linux/fs/xfs/xfs_inode.c

3820 lines
106 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2006 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#include <linux/iversion.h>
#include "xfs.h"
#include "xfs_fs.h"
#include "xfs_shared.h"
#include "xfs_format.h"
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
#include "xfs_mount.h"
#include "xfs_defer.h"
#include "xfs_inode.h"
#include "xfs_dir2.h"
#include "xfs_attr.h"
#include "xfs_trans_space.h"
#include "xfs_trans.h"
#include "xfs_buf_item.h"
#include "xfs_inode_item.h"
#include "xfs_ialloc.h"
#include "xfs_bmap.h"
#include "xfs_bmap_util.h"
#include "xfs_errortag.h"
#include "xfs_error.h"
#include "xfs_quota.h"
[XFS] Concurrent Multi-File Data Streams In media spaces, video is often stored in a frame-per-file format. When dealing with uncompressed realtime HD video streams in this format, it is crucial that files do not get fragmented and that multiple files a placed contiguously on disk. When multiple streams are being ingested and played out at the same time, it is critical that the filesystem does not cross the streams and interleave them together as this creates seek and readahead cache miss latency and prevents both ingest and playout from meeting frame rate targets. This patch set creates a "stream of files" concept into the allocator to place all the data from a single stream contiguously on disk so that RAID array readahead can be used effectively. Each additional stream gets placed in different allocation groups within the filesystem, thereby ensuring that we don't cross any streams. When an AG fills up, we select a new AG for the stream that is not in use. The core of the functionality is the stream tracking - each inode that we create in a directory needs to be associated with the directories' stream. Hence every time we create a file, we look up the directories' stream object and associate the new file with that object. Once we have a stream object for a file, we use the AG that the stream object point to for allocations. If we can't allocate in that AG (e.g. it is full) we move the entire stream to another AG. Other inodes in the same stream are moved to the new AG on their next allocation (i.e. lazy update). Stream objects are kept in a cache and hold a reference on the inode. Hence the inode cannot be reclaimed while there is an outstanding stream reference. This means that on unlink we need to remove the stream association and we also need to flush all the associations on certain events that want to reclaim all unreferenced inodes (e.g. filesystem freeze). SGI-PV: 964469 SGI-Modid: xfs-linux-melb:xfs-kern:29096a Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Barry Naujok <bnaujok@sgi.com> Signed-off-by: Donald Douwsma <donaldd@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Tim Shimmin <tes@sgi.com> Signed-off-by: Vlad Apostolov <vapo@sgi.com>
2007-07-11 01:09:12 +00:00
#include "xfs_filestream.h"
xfs: event tracing support Convert the old xfs tracing support that could only be used with the out of tree kdb and xfsidbg patches to use the generic event tracer. To use it make sure CONFIG_EVENT_TRACING is enabled and then enable all xfs trace channels by: echo 1 > /sys/kernel/debug/tracing/events/xfs/enable or alternatively enable single events by just doing the same in one event subdirectory, e.g. echo 1 > /sys/kernel/debug/tracing/events/xfs/xfs_ihold/enable or set more complex filters, etc. In Documentation/trace/events.txt all this is desctribed in more detail. To reads the events do a cat /sys/kernel/debug/tracing/trace Compared to the last posting this patch converts the tracing mostly to the one tracepoint per callsite model that other users of the new tracing facility also employ. This allows a very fine-grained control of the tracing, a cleaner output of the traces and also enables the perf tool to use each tracepoint as a virtual performance counter, allowing us to e.g. count how often certain workloads git various spots in XFS. Take a look at http://lwn.net/Articles/346470/ for some examples. Also the btree tracing isn't included at all yet, as it will require additional core tracing features not in mainline yet, I plan to deliver it later. And the really nice thing about this patch is that it actually removes many lines of code while adding this nice functionality: fs/xfs/Makefile | 8 fs/xfs/linux-2.6/xfs_acl.c | 1 fs/xfs/linux-2.6/xfs_aops.c | 52 - fs/xfs/linux-2.6/xfs_aops.h | 2 fs/xfs/linux-2.6/xfs_buf.c | 117 +-- fs/xfs/linux-2.6/xfs_buf.h | 33 fs/xfs/linux-2.6/xfs_fs_subr.c | 3 fs/xfs/linux-2.6/xfs_ioctl.c | 1 fs/xfs/linux-2.6/xfs_ioctl32.c | 1 fs/xfs/linux-2.6/xfs_iops.c | 1 fs/xfs/linux-2.6/xfs_linux.h | 1 fs/xfs/linux-2.6/xfs_lrw.c | 87 -- fs/xfs/linux-2.6/xfs_lrw.h | 45 - fs/xfs/linux-2.6/xfs_super.c | 104 --- fs/xfs/linux-2.6/xfs_super.h | 7 fs/xfs/linux-2.6/xfs_sync.c | 1 fs/xfs/linux-2.6/xfs_trace.c | 75 ++ fs/xfs/linux-2.6/xfs_trace.h | 1369 +++++++++++++++++++++++++++++++++++++++++ fs/xfs/linux-2.6/xfs_vnode.h | 4 fs/xfs/quota/xfs_dquot.c | 110 --- fs/xfs/quota/xfs_dquot.h | 21 fs/xfs/quota/xfs_qm.c | 40 - fs/xfs/quota/xfs_qm_syscalls.c | 4 fs/xfs/support/ktrace.c | 323 --------- fs/xfs/support/ktrace.h | 85 -- fs/xfs/xfs.h | 16 fs/xfs/xfs_ag.h | 14 fs/xfs/xfs_alloc.c | 230 +----- fs/xfs/xfs_alloc.h | 27 fs/xfs/xfs_alloc_btree.c | 1 fs/xfs/xfs_attr.c | 107 --- fs/xfs/xfs_attr.h | 10 fs/xfs/xfs_attr_leaf.c | 14 fs/xfs/xfs_attr_sf.h | 40 - fs/xfs/xfs_bmap.c | 507 +++------------ fs/xfs/xfs_bmap.h | 49 - fs/xfs/xfs_bmap_btree.c | 6 fs/xfs/xfs_btree.c | 5 fs/xfs/xfs_btree_trace.h | 17 fs/xfs/xfs_buf_item.c | 87 -- fs/xfs/xfs_buf_item.h | 20 fs/xfs/xfs_da_btree.c | 3 fs/xfs/xfs_da_btree.h | 7 fs/xfs/xfs_dfrag.c | 2 fs/xfs/xfs_dir2.c | 8 fs/xfs/xfs_dir2_block.c | 20 fs/xfs/xfs_dir2_leaf.c | 21 fs/xfs/xfs_dir2_node.c | 27 fs/xfs/xfs_dir2_sf.c | 26 fs/xfs/xfs_dir2_trace.c | 216 ------ fs/xfs/xfs_dir2_trace.h | 72 -- fs/xfs/xfs_filestream.c | 8 fs/xfs/xfs_fsops.c | 2 fs/xfs/xfs_iget.c | 111 --- fs/xfs/xfs_inode.c | 67 -- fs/xfs/xfs_inode.h | 76 -- fs/xfs/xfs_inode_item.c | 5 fs/xfs/xfs_iomap.c | 85 -- fs/xfs/xfs_iomap.h | 8 fs/xfs/xfs_log.c | 181 +---- fs/xfs/xfs_log_priv.h | 20 fs/xfs/xfs_log_recover.c | 1 fs/xfs/xfs_mount.c | 2 fs/xfs/xfs_quota.h | 8 fs/xfs/xfs_rename.c | 1 fs/xfs/xfs_rtalloc.c | 1 fs/xfs/xfs_rw.c | 3 fs/xfs/xfs_trans.h | 47 + fs/xfs/xfs_trans_buf.c | 62 - fs/xfs/xfs_vnodeops.c | 8 70 files changed, 2151 insertions(+), 2592 deletions(-) Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Alex Elder <aelder@sgi.com>
2009-12-14 23:14:59 +00:00
#include "xfs_trace.h"
#include "xfs_icache.h"
#include "xfs_symlink.h"
#include "xfs_trans_priv.h"
#include "xfs_log.h"
#include "xfs_bmap_btree.h"
#include "xfs_reflink.h"
#include "xfs_ag.h"
kmem_zone_t *xfs_inode_zone;
/*
* Used in xfs_itruncate_extents(). This is the maximum number of extents
* freed from a file in a single transaction.
*/
#define XFS_ITRUNC_MAX_EXTENTS 2
STATIC int xfs_iunlink(struct xfs_trans *, struct xfs_inode *);
STATIC int xfs_iunlink_remove(struct xfs_trans *tp, struct xfs_perag *pag,
struct xfs_inode *);
/*
* helper function to extract extent size hint from inode
*/
xfs_extlen_t
xfs_get_extsz_hint(
struct xfs_inode *ip)
{
/*
* No point in aligning allocations if we need to COW to actually
* write to them.
*/
if (xfs_is_always_cow_inode(ip))
return 0;
if ((ip->i_diflags & XFS_DIFLAG_EXTSIZE) && ip->i_extsize)
return ip->i_extsize;
if (XFS_IS_REALTIME_INODE(ip))
return ip->i_mount->m_sb.sb_rextsize;
return 0;
}
/*
* Helper function to extract CoW extent size hint from inode.
* Between the extent size hint and the CoW extent size hint, we
* return the greater of the two. If the value is zero (automatic),
* use the default size.
*/
xfs_extlen_t
xfs_get_cowextsz_hint(
struct xfs_inode *ip)
{
xfs_extlen_t a, b;
a = 0;
if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
a = ip->i_cowextsize;
b = xfs_get_extsz_hint(ip);
a = max(a, b);
if (a == 0)
return XFS_DEFAULT_COWEXTSZ_HINT;
return a;
}
/*
* These two are wrapper routines around the xfs_ilock() routine used to
* centralize some grungy code. They are used in places that wish to lock the
* inode solely for reading the extents. The reason these places can't just
* call xfs_ilock(ip, XFS_ILOCK_SHARED) is that the inode lock also guards to
* bringing in of the extents from disk for a file in b-tree format. If the
* inode is in b-tree format, then we need to lock the inode exclusively until
* the extents are read in. Locking it exclusively all the time would limit
* our parallelism unnecessarily, though. What we do instead is check to see
* if the extents have been read in yet, and only lock the inode exclusively
* if they have not.
*
* The functions return a value which should be given to the corresponding
* xfs_iunlock() call.
*/
uint
xfs_ilock_data_map_shared(
struct xfs_inode *ip)
{
uint lock_mode = XFS_ILOCK_SHARED;
if (xfs_need_iread_extents(&ip->i_df))
lock_mode = XFS_ILOCK_EXCL;
xfs_ilock(ip, lock_mode);
return lock_mode;
}
uint
xfs_ilock_attr_map_shared(
struct xfs_inode *ip)
{
uint lock_mode = XFS_ILOCK_SHARED;
if (ip->i_afp && xfs_need_iread_extents(ip->i_afp))
lock_mode = XFS_ILOCK_EXCL;
xfs_ilock(ip, lock_mode);
return lock_mode;
}
/*
* In addition to i_rwsem in the VFS inode, the xfs inode contains 2
* multi-reader locks: i_mmap_lock and the i_lock. This routine allows
* various combinations of the locks to be obtained.
*
* The 3 locks should always be ordered so that the IO lock is obtained first,
* the mmap lock second and the ilock last in order to prevent deadlock.
*
* Basic locking order:
*
* i_rwsem -> i_mmap_lock -> page_lock -> i_ilock
*
* mmap_lock locking order:
*
* i_rwsem -> page lock -> mmap_lock
* mmap_lock -> i_mmap_lock -> page_lock
*
* The difference in mmap_lock locking order mean that we cannot hold the
* i_mmap_lock over syscall based read(2)/write(2) based IO. These IO paths can
* fault in pages during copy in/out (for buffered IO) or require the mmap_lock
* in get_user_pages() to map the user pages into the kernel address space for
* direct IO. Similarly the i_rwsem cannot be taken inside a page fault because
* page faults already hold the mmap_lock.
*
* Hence to serialise fully against both syscall and mmap based IO, we need to
* take both the i_rwsem and the i_mmap_lock. These locks should *only* be both
* taken in places where we need to invalidate the page cache in a race
* free manner (e.g. truncate, hole punch and other extent manipulation
* functions).
*/
void
xfs_ilock(
xfs_inode_t *ip,
uint lock_flags)
{
trace_xfs_ilock(ip, lock_flags, _RET_IP_);
/*
* You can't set both SHARED and EXCL for the same lock,
* and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
* and XFS_ILOCK_EXCL are valid values to set in lock_flags.
*/
ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
if (lock_flags & XFS_IOLOCK_EXCL) {
down_write_nested(&VFS_I(ip)->i_rwsem,
XFS_IOLOCK_DEP(lock_flags));
} else if (lock_flags & XFS_IOLOCK_SHARED) {
down_read_nested(&VFS_I(ip)->i_rwsem,
XFS_IOLOCK_DEP(lock_flags));
}
if (lock_flags & XFS_MMAPLOCK_EXCL)
mrupdate_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
else if (lock_flags & XFS_MMAPLOCK_SHARED)
mraccess_nested(&ip->i_mmaplock, XFS_MMAPLOCK_DEP(lock_flags));
if (lock_flags & XFS_ILOCK_EXCL)
mrupdate_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
else if (lock_flags & XFS_ILOCK_SHARED)
mraccess_nested(&ip->i_lock, XFS_ILOCK_DEP(lock_flags));
}
/*
* This is just like xfs_ilock(), except that the caller
* is guaranteed not to sleep. It returns 1 if it gets
* the requested locks and 0 otherwise. If the IO lock is
* obtained but the inode lock cannot be, then the IO lock
* is dropped before returning.
*
* ip -- the inode being locked
* lock_flags -- this parameter indicates the inode's locks to be
* to be locked. See the comment for xfs_ilock() for a list
* of valid values.
*/
int
xfs_ilock_nowait(
xfs_inode_t *ip,
uint lock_flags)
{
trace_xfs_ilock_nowait(ip, lock_flags, _RET_IP_);
/*
* You can't set both SHARED and EXCL for the same lock,
* and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
* and XFS_ILOCK_EXCL are valid values to set in lock_flags.
*/
ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
if (lock_flags & XFS_IOLOCK_EXCL) {
if (!down_write_trylock(&VFS_I(ip)->i_rwsem))
goto out;
} else if (lock_flags & XFS_IOLOCK_SHARED) {
if (!down_read_trylock(&VFS_I(ip)->i_rwsem))
goto out;
}
if (lock_flags & XFS_MMAPLOCK_EXCL) {
if (!mrtryupdate(&ip->i_mmaplock))
goto out_undo_iolock;
} else if (lock_flags & XFS_MMAPLOCK_SHARED) {
if (!mrtryaccess(&ip->i_mmaplock))
goto out_undo_iolock;
}
if (lock_flags & XFS_ILOCK_EXCL) {
if (!mrtryupdate(&ip->i_lock))
goto out_undo_mmaplock;
} else if (lock_flags & XFS_ILOCK_SHARED) {
if (!mrtryaccess(&ip->i_lock))
goto out_undo_mmaplock;
}
return 1;
out_undo_mmaplock:
if (lock_flags & XFS_MMAPLOCK_EXCL)
mrunlock_excl(&ip->i_mmaplock);
else if (lock_flags & XFS_MMAPLOCK_SHARED)
mrunlock_shared(&ip->i_mmaplock);
out_undo_iolock:
if (lock_flags & XFS_IOLOCK_EXCL)
up_write(&VFS_I(ip)->i_rwsem);
else if (lock_flags & XFS_IOLOCK_SHARED)
up_read(&VFS_I(ip)->i_rwsem);
out:
return 0;
}
/*
* xfs_iunlock() is used to drop the inode locks acquired with
* xfs_ilock() and xfs_ilock_nowait(). The caller must pass
* in the flags given to xfs_ilock() or xfs_ilock_nowait() so
* that we know which locks to drop.
*
* ip -- the inode being unlocked
* lock_flags -- this parameter indicates the inode's locks to be
* to be unlocked. See the comment for xfs_ilock() for a list
* of valid values for this parameter.
*
*/
void
xfs_iunlock(
xfs_inode_t *ip,
uint lock_flags)
{
/*
* You can't set both SHARED and EXCL for the same lock,
* and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
* and XFS_ILOCK_EXCL are valid values to set in lock_flags.
*/
ASSERT((lock_flags & (XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL)) !=
(XFS_IOLOCK_SHARED | XFS_IOLOCK_EXCL));
ASSERT((lock_flags & (XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL)) !=
(XFS_MMAPLOCK_SHARED | XFS_MMAPLOCK_EXCL));
ASSERT((lock_flags & (XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)) !=
(XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT((lock_flags & ~(XFS_LOCK_MASK | XFS_LOCK_SUBCLASS_MASK)) == 0);
ASSERT(lock_flags != 0);
if (lock_flags & XFS_IOLOCK_EXCL)
up_write(&VFS_I(ip)->i_rwsem);
else if (lock_flags & XFS_IOLOCK_SHARED)
up_read(&VFS_I(ip)->i_rwsem);
if (lock_flags & XFS_MMAPLOCK_EXCL)
mrunlock_excl(&ip->i_mmaplock);
else if (lock_flags & XFS_MMAPLOCK_SHARED)
mrunlock_shared(&ip->i_mmaplock);
if (lock_flags & XFS_ILOCK_EXCL)
mrunlock_excl(&ip->i_lock);
else if (lock_flags & XFS_ILOCK_SHARED)
mrunlock_shared(&ip->i_lock);
trace_xfs_iunlock(ip, lock_flags, _RET_IP_);
}
/*
* give up write locks. the i/o lock cannot be held nested
* if it is being demoted.
*/
void
xfs_ilock_demote(
xfs_inode_t *ip,
uint lock_flags)
{
ASSERT(lock_flags & (XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL));
ASSERT((lock_flags &
~(XFS_IOLOCK_EXCL|XFS_MMAPLOCK_EXCL|XFS_ILOCK_EXCL)) == 0);
if (lock_flags & XFS_ILOCK_EXCL)
mrdemote(&ip->i_lock);
if (lock_flags & XFS_MMAPLOCK_EXCL)
mrdemote(&ip->i_mmaplock);
if (lock_flags & XFS_IOLOCK_EXCL)
downgrade_write(&VFS_I(ip)->i_rwsem);
trace_xfs_ilock_demote(ip, lock_flags, _RET_IP_);
}
#if defined(DEBUG) || defined(XFS_WARN)
int
xfs_isilocked(
xfs_inode_t *ip,
uint lock_flags)
{
if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
if (!(lock_flags & XFS_ILOCK_SHARED))
return !!ip->i_lock.mr_writer;
return rwsem_is_locked(&ip->i_lock.mr_lock);
}
if (lock_flags & (XFS_MMAPLOCK_EXCL|XFS_MMAPLOCK_SHARED)) {
if (!(lock_flags & XFS_MMAPLOCK_SHARED))
return !!ip->i_mmaplock.mr_writer;
return rwsem_is_locked(&ip->i_mmaplock.mr_lock);
}
if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
if (!(lock_flags & XFS_IOLOCK_SHARED))
return !debug_locks ||
lockdep_is_held_type(&VFS_I(ip)->i_rwsem, 0);
return rwsem_is_locked(&VFS_I(ip)->i_rwsem);
}
ASSERT(0);
return 0;
}
#endif
/*
* xfs_lockdep_subclass_ok() is only used in an ASSERT, so is only called when
* DEBUG or XFS_WARN is set. And MAX_LOCKDEP_SUBCLASSES is then only defined
* when CONFIG_LOCKDEP is set. Hence the complex define below to avoid build
* errors and warnings.
*/
#if (defined(DEBUG) || defined(XFS_WARN)) && defined(CONFIG_LOCKDEP)
static bool
xfs_lockdep_subclass_ok(
int subclass)
{
return subclass < MAX_LOCKDEP_SUBCLASSES;
}
#else
#define xfs_lockdep_subclass_ok(subclass) (true)
#endif
/*
* Bump the subclass so xfs_lock_inodes() acquires each lock with a different
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
* value. This can be called for any type of inode lock combination, including
* parent locking. Care must be taken to ensure we don't overrun the subclass
* storage fields in the class mask we build.
*/
static inline int
xfs_lock_inumorder(int lock_mode, int subclass)
{
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
int class = 0;
ASSERT(!(lock_mode & (XFS_ILOCK_PARENT | XFS_ILOCK_RTBITMAP |
XFS_ILOCK_RTSUM)));
ASSERT(xfs_lockdep_subclass_ok(subclass));
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
if (lock_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)) {
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT(subclass <= XFS_IOLOCK_MAX_SUBCLASS);
class += subclass << XFS_IOLOCK_SHIFT;
}
if (lock_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) {
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT(subclass <= XFS_MMAPLOCK_MAX_SUBCLASS);
class += subclass << XFS_MMAPLOCK_SHIFT;
}
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
if (lock_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)) {
ASSERT(subclass <= XFS_ILOCK_MAX_SUBCLASS);
class += subclass << XFS_ILOCK_SHIFT;
}
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
return (lock_mode & ~XFS_LOCK_SUBCLASS_MASK) | class;
}
/*
* The following routine will lock n inodes in exclusive mode. We assume the
* caller calls us with the inodes in i_ino order.
*
* We need to detect deadlock where an inode that we lock is in the AIL and we
* start waiting for another inode that is locked by a thread in a long running
* transaction (such as truncate). This can result in deadlock since the long
* running trans might need to wait for the inode we just locked in order to
* push the tail and free space in the log.
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
*
* xfs_lock_inodes() can only be used to lock one type of lock at a time -
* the iolock, the mmaplock or the ilock, but not more than one at a time. If we
* lock more than one at a time, lockdep will report false positives saying we
* have violated locking orders.
*/
static void
xfs_lock_inodes(
struct xfs_inode **ips,
int inodes,
uint lock_mode)
{
int attempts = 0, i, j, try_lock;
struct xfs_log_item *lp;
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
/*
* Currently supports between 2 and 5 inodes with exclusive locking. We
* support an arbitrary depth of locking here, but absolute limits on
* inodes depend on the type of locking and the limits placed by
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
* lockdep annotations in xfs_lock_inumorder. These are all checked by
* the asserts.
*/
ASSERT(ips && inodes >= 2 && inodes <= 5);
xfs: clean up inode lockdep annotations Lockdep annotations are a maintenance nightmare. Locking has to be modified to suit the limitations of the annotations, and we're always having to fix the annotations because they are unable to express the complexity of locking heirarchies correctly. So, next up, we've got more issues with lockdep annotations for inode locking w.r.t. XFS_LOCK_PARENT: - lockdep classes are exclusive and can't be ORed together to form new classes. - IOLOCK needs multiple PARENT subclasses to express the changes needed for the readdir locking rework needed to stop the endless flow of lockdep false positives involving readdir calling filldir under the ILOCK. - there are only 8 unique lockdep subclasses available, so we can't create a generic solution. IOWs we need to treat the 3-bit space available to each lock type differently: - IOLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 IOLOCK subclasses - at least 2 IOLOCK_PARENT subclasses - MMAPLOCK uses xfs_lock_two_inodes(), so needs: - at least 2 MMAPLOCK subclasses - ILOCK uses xfs_lock_inodes with up to 5 inodes, so needs: - at least 5 ILOCK subclasses - one ILOCK_PARENT subclass - one RTBITMAP subclass - one RTSUM subclass For the IOLOCK, split the space into two sets of subclasses. For the MMAPLOCK, just use half the space for the one subclass to match the non-parent lock classes of the IOLOCK. For the ILOCK, use 0-4 as the ILOCK subclasses, 5-7 for the remaining individual subclasses. Because they are now all different, modify xfs_lock_inumorder() to handle the nested subclasses, and to assert fail if passed an invalid subclass. Further, annotate xfs_lock_inodes() to assert fail if an invalid combination of lock primitives and inode counts are passed that would result in a lockdep subclass annotation overflow. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:32:49 +00:00
ASSERT(lock_mode & (XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL |
XFS_ILOCK_EXCL));
ASSERT(!(lock_mode & (XFS_IOLOCK_SHARED | XFS_MMAPLOCK_SHARED |
XFS_ILOCK_SHARED)));
ASSERT(!(lock_mode & XFS_MMAPLOCK_EXCL) ||
inodes <= XFS_MMAPLOCK_MAX_SUBCLASS + 1);
ASSERT(!(lock_mode & XFS_ILOCK_EXCL) ||
inodes <= XFS_ILOCK_MAX_SUBCLASS + 1);
if (lock_mode & XFS_IOLOCK_EXCL) {
ASSERT(!(lock_mode & (XFS_MMAPLOCK_EXCL | XFS_ILOCK_EXCL)));
} else if (lock_mode & XFS_MMAPLOCK_EXCL)
ASSERT(!(lock_mode & XFS_ILOCK_EXCL));
try_lock = 0;
i = 0;
again:
for (; i < inodes; i++) {
ASSERT(ips[i]);
if (i && (ips[i] == ips[i - 1])) /* Already locked */
continue;
/*
* If try_lock is not set yet, make sure all locked inodes are
* not in the AIL. If any are, set try_lock to be used later.
*/
if (!try_lock) {
for (j = (i - 1); j >= 0 && !try_lock; j--) {
lp = &ips[j]->i_itemp->ili_item;
if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags))
try_lock++;
}
}
/*
* If any of the previous locks we have locked is in the AIL,
* we must TRY to get the second and subsequent locks. If
* we can't get any, we must release all we have
* and try again.
*/
if (!try_lock) {
xfs_ilock(ips[i], xfs_lock_inumorder(lock_mode, i));
continue;
}
/* try_lock means we have an inode locked that is in the AIL. */
ASSERT(i != 0);
if (xfs_ilock_nowait(ips[i], xfs_lock_inumorder(lock_mode, i)))
continue;
/*
* Unlock all previous guys and try again. xfs_iunlock will try
* to push the tail if the inode is in the AIL.
*/
attempts++;
for (j = i - 1; j >= 0; j--) {
/*
* Check to see if we've already unlocked this one. Not
* the first one going back, and the inode ptr is the
* same.
*/
if (j != (i - 1) && ips[j] == ips[j + 1])
continue;
xfs_iunlock(ips[j], lock_mode);
}
if ((attempts % 5) == 0) {
delay(1); /* Don't just spin the CPU */
}
i = 0;
try_lock = 0;
goto again;
}
}
/*
* xfs_lock_two_inodes() can only be used to lock one type of lock at a time -
* the mmaplock or the ilock, but not more than one type at a time. If we lock
* more than one at a time, lockdep will report false positives saying we have
* violated locking orders. The iolock must be double-locked separately since
* we use i_rwsem for that. We now support taking one lock EXCL and the other
* SHARED.
*/
void
xfs_lock_two_inodes(
struct xfs_inode *ip0,
uint ip0_mode,
struct xfs_inode *ip1,
uint ip1_mode)
{
struct xfs_inode *temp;
uint mode_temp;
int attempts = 0;
struct xfs_log_item *lp;
ASSERT(hweight32(ip0_mode) == 1);
ASSERT(hweight32(ip1_mode) == 1);
ASSERT(!(ip0_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
ASSERT(!(ip1_mode & (XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL)));
ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
!(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
!(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
ASSERT(!(ip1_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
!(ip0_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
ASSERT(!(ip0_mode & (XFS_MMAPLOCK_SHARED|XFS_MMAPLOCK_EXCL)) ||
!(ip1_mode & (XFS_ILOCK_SHARED|XFS_ILOCK_EXCL)));
ASSERT(ip0->i_ino != ip1->i_ino);
if (ip0->i_ino > ip1->i_ino) {
temp = ip0;
ip0 = ip1;
ip1 = temp;
mode_temp = ip0_mode;
ip0_mode = ip1_mode;
ip1_mode = mode_temp;
}
again:
xfs_ilock(ip0, xfs_lock_inumorder(ip0_mode, 0));
/*
* If the first lock we have locked is in the AIL, we must TRY to get
* the second lock. If we can't get it, we must release the first one
* and try again.
*/
lp = &ip0->i_itemp->ili_item;
if (lp && test_bit(XFS_LI_IN_AIL, &lp->li_flags)) {
if (!xfs_ilock_nowait(ip1, xfs_lock_inumorder(ip1_mode, 1))) {
xfs_iunlock(ip0, ip0_mode);
if ((++attempts % 5) == 0)
delay(1); /* Don't just spin the CPU */
goto again;
}
} else {
xfs_ilock(ip1, xfs_lock_inumorder(ip1_mode, 1));
}
}
uint
xfs_ip2xflags(
struct xfs_inode *ip)
{
uint flags = 0;
if (ip->i_diflags & XFS_DIFLAG_ANY) {
if (ip->i_diflags & XFS_DIFLAG_REALTIME)
flags |= FS_XFLAG_REALTIME;
if (ip->i_diflags & XFS_DIFLAG_PREALLOC)
flags |= FS_XFLAG_PREALLOC;
if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE)
flags |= FS_XFLAG_IMMUTABLE;
if (ip->i_diflags & XFS_DIFLAG_APPEND)
flags |= FS_XFLAG_APPEND;
if (ip->i_diflags & XFS_DIFLAG_SYNC)
flags |= FS_XFLAG_SYNC;
if (ip->i_diflags & XFS_DIFLAG_NOATIME)
flags |= FS_XFLAG_NOATIME;
if (ip->i_diflags & XFS_DIFLAG_NODUMP)
flags |= FS_XFLAG_NODUMP;
if (ip->i_diflags & XFS_DIFLAG_RTINHERIT)
flags |= FS_XFLAG_RTINHERIT;
if (ip->i_diflags & XFS_DIFLAG_PROJINHERIT)
flags |= FS_XFLAG_PROJINHERIT;
if (ip->i_diflags & XFS_DIFLAG_NOSYMLINKS)
flags |= FS_XFLAG_NOSYMLINKS;
if (ip->i_diflags & XFS_DIFLAG_EXTSIZE)
flags |= FS_XFLAG_EXTSIZE;
if (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT)
flags |= FS_XFLAG_EXTSZINHERIT;
if (ip->i_diflags & XFS_DIFLAG_NODEFRAG)
flags |= FS_XFLAG_NODEFRAG;
if (ip->i_diflags & XFS_DIFLAG_FILESTREAM)
flags |= FS_XFLAG_FILESTREAM;
}
if (ip->i_diflags2 & XFS_DIFLAG2_ANY) {
if (ip->i_diflags2 & XFS_DIFLAG2_DAX)
flags |= FS_XFLAG_DAX;
if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE)
flags |= FS_XFLAG_COWEXTSIZE;
}
if (XFS_IFORK_Q(ip))
flags |= FS_XFLAG_HASATTR;
return flags;
}
/*
* Lookups up an inode from "name". If ci_name is not NULL, then a CI match
* is allowed, otherwise it has to be an exact match. If a CI match is found,
* ci_name->name will point to a the actual name (caller must free) or
* will be set to NULL if an exact match is found.
*/
int
xfs_lookup(
xfs_inode_t *dp,
struct xfs_name *name,
xfs_inode_t **ipp,
struct xfs_name *ci_name)
{
xfs_ino_t inum;
int error;
trace_xfs_lookup(dp, name);
if (XFS_FORCED_SHUTDOWN(dp->i_mount))
return -EIO;
error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
if (error)
xfs: stop holding ILOCK over filldir callbacks The recent change to the readdir locking made in 40194ec ("xfs: reinstate the ilock in xfs_readdir") for CXFS directory sanity was probably the wrong thing to do. Deep in the readdir code we can take page faults in the filldir callback, and so taking a page fault while holding an inode ilock creates a new set of locking issues that lockdep warns all over the place about. The locking order for regular inodes w.r.t. page faults is io_lock -> pagefault -> mmap_sem -> ilock. The directory readdir code now triggers ilock -> page fault -> mmap_sem. While we cannot deadlock at this point, it inverts all the locking patterns that lockdep normally sees on XFS inodes, and so triggers lockdep. We worked around this with commit 93a8614 ("xfs: fix directory inode iolock lockdep false positive"), but that then just moved the lockdep warning to deeper in the page fault path and triggered on security inode locks. Fixing the shmem issue there just moved the lockdep reports somewhere else, and now we are getting false positives from filesystem freezing annotations getting confused. Further, if we enter memory reclaim in a readdir path, we now get lockdep warning about potential deadlocks because the ilock is held when we enter reclaim. This, again, is different to a regular file in that we never allow memory reclaim to run while holding the ilock for regular files. Hence lockdep now throws ilock->kmalloc->reclaim->ilock warnings. Basically, the problem is that the ilock is being used to protect the directory data and the inode metadata, whereas for a regular file the iolock protects the data and the ilock protects the metadata. From the VFS perspective, the i_mutex serialises all accesses to the directory data, and so not holding the ilock for readdir doesn't matter. The issue is that CXFS doesn't access directory data via the VFS, so it has no "data serialisaton" mechanism. Hence we need to hold the IOLOCK in the correct places to provide this low level directory data access serialisation. The ilock can then be used just when the extent list needs to be read, just like we do for regular files. The directory modification code can take the iolock exclusive when the ilock is also taken, and this then ensures that readdir is correct excluded while modifications are in progress. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:33:00 +00:00
goto out_unlock;
error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp);
if (error)
goto out_free_name;
return 0;
out_free_name:
if (ci_name)
kmem_free(ci_name->name);
xfs: stop holding ILOCK over filldir callbacks The recent change to the readdir locking made in 40194ec ("xfs: reinstate the ilock in xfs_readdir") for CXFS directory sanity was probably the wrong thing to do. Deep in the readdir code we can take page faults in the filldir callback, and so taking a page fault while holding an inode ilock creates a new set of locking issues that lockdep warns all over the place about. The locking order for regular inodes w.r.t. page faults is io_lock -> pagefault -> mmap_sem -> ilock. The directory readdir code now triggers ilock -> page fault -> mmap_sem. While we cannot deadlock at this point, it inverts all the locking patterns that lockdep normally sees on XFS inodes, and so triggers lockdep. We worked around this with commit 93a8614 ("xfs: fix directory inode iolock lockdep false positive"), but that then just moved the lockdep warning to deeper in the page fault path and triggered on security inode locks. Fixing the shmem issue there just moved the lockdep reports somewhere else, and now we are getting false positives from filesystem freezing annotations getting confused. Further, if we enter memory reclaim in a readdir path, we now get lockdep warning about potential deadlocks because the ilock is held when we enter reclaim. This, again, is different to a regular file in that we never allow memory reclaim to run while holding the ilock for regular files. Hence lockdep now throws ilock->kmalloc->reclaim->ilock warnings. Basically, the problem is that the ilock is being used to protect the directory data and the inode metadata, whereas for a regular file the iolock protects the data and the ilock protects the metadata. From the VFS perspective, the i_mutex serialises all accesses to the directory data, and so not holding the ilock for readdir doesn't matter. The issue is that CXFS doesn't access directory data via the VFS, so it has no "data serialisaton" mechanism. Hence we need to hold the IOLOCK in the correct places to provide this low level directory data access serialisation. The ilock can then be used just when the extent list needs to be read, just like we do for regular files. The directory modification code can take the iolock exclusive when the ilock is also taken, and this then ensures that readdir is correct excluded while modifications are in progress. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-08-19 00:33:00 +00:00
out_unlock:
*ipp = NULL;
return error;
}
/* Propagate di_flags from a parent inode to a child inode. */
static void
xfs_inode_inherit_flags(
struct xfs_inode *ip,
const struct xfs_inode *pip)
{
unsigned int di_flags = 0;
xfs: validate extsz hints against rt extent size when rtinherit is set The RTINHERIT bit can be set on a directory so that newly created regular files will have the REALTIME bit set to store their data on the realtime volume. If an extent size hint (and EXTSZINHERIT) are set on the directory, the hint will also be copied into the new file. As pointed out in previous patches, for realtime files we require the extent size hint be an integer multiple of the realtime extent, but we don't perform the same validation on a directory with both RTINHERIT and EXTSZINHERIT set, even though the only use-case of that combination is to propagate extent size hints into new realtime files. This leads to inode corruption errors when the bad values are propagated. Because there may be existing filesystems with such a configuration, we cannot simply amend the inode verifier to trip on these directories and call it a day because that will cause previously "working" filesystems to start throwing errors abruptly. Note that it's valid to have directories with rtinherit set even if there is no realtime volume, in which case the problem does not manifest because rtinherit is ignored if there's no realtime device; and it's possible that someone set the flag, crashed, repaired the filesystem (which clears the hint on the realtime file) and continued. Therefore, mitigate this issue in several ways: First, if we try to write out an inode with both rtinherit/extszinherit set and an unaligned extent size hint, turn off the hint to correct the error. Second, if someone tries to misconfigure a directory via the fssetxattr ioctl, fail the ioctl. Third, reverify both extent size hint values when we propagate heritable inode attributes from parent to child, to prevent misconfigurations from spreading. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2021-05-12 19:51:26 +00:00
xfs_failaddr_t failaddr;
umode_t mode = VFS_I(ip)->i_mode;
if (S_ISDIR(mode)) {
if (pip->i_diflags & XFS_DIFLAG_RTINHERIT)
di_flags |= XFS_DIFLAG_RTINHERIT;
if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
di_flags |= XFS_DIFLAG_EXTSZINHERIT;
ip->i_extsize = pip->i_extsize;
}
if (pip->i_diflags & XFS_DIFLAG_PROJINHERIT)
di_flags |= XFS_DIFLAG_PROJINHERIT;
} else if (S_ISREG(mode)) {
if ((pip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
xfs_sb_version_hasrealtime(&ip->i_mount->m_sb))
di_flags |= XFS_DIFLAG_REALTIME;
if (pip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) {
di_flags |= XFS_DIFLAG_EXTSIZE;
ip->i_extsize = pip->i_extsize;
}
}
if ((pip->i_diflags & XFS_DIFLAG_NOATIME) &&
xfs_inherit_noatime)
di_flags |= XFS_DIFLAG_NOATIME;
if ((pip->i_diflags & XFS_DIFLAG_NODUMP) &&
xfs_inherit_nodump)
di_flags |= XFS_DIFLAG_NODUMP;
if ((pip->i_diflags & XFS_DIFLAG_SYNC) &&
xfs_inherit_sync)
di_flags |= XFS_DIFLAG_SYNC;
if ((pip->i_diflags & XFS_DIFLAG_NOSYMLINKS) &&
xfs_inherit_nosymlinks)
di_flags |= XFS_DIFLAG_NOSYMLINKS;
if ((pip->i_diflags & XFS_DIFLAG_NODEFRAG) &&
xfs_inherit_nodefrag)
di_flags |= XFS_DIFLAG_NODEFRAG;
if (pip->i_diflags & XFS_DIFLAG_FILESTREAM)
di_flags |= XFS_DIFLAG_FILESTREAM;
ip->i_diflags |= di_flags;
xfs: validate extsz hints against rt extent size when rtinherit is set The RTINHERIT bit can be set on a directory so that newly created regular files will have the REALTIME bit set to store their data on the realtime volume. If an extent size hint (and EXTSZINHERIT) are set on the directory, the hint will also be copied into the new file. As pointed out in previous patches, for realtime files we require the extent size hint be an integer multiple of the realtime extent, but we don't perform the same validation on a directory with both RTINHERIT and EXTSZINHERIT set, even though the only use-case of that combination is to propagate extent size hints into new realtime files. This leads to inode corruption errors when the bad values are propagated. Because there may be existing filesystems with such a configuration, we cannot simply amend the inode verifier to trip on these directories and call it a day because that will cause previously "working" filesystems to start throwing errors abruptly. Note that it's valid to have directories with rtinherit set even if there is no realtime volume, in which case the problem does not manifest because rtinherit is ignored if there's no realtime device; and it's possible that someone set the flag, crashed, repaired the filesystem (which clears the hint on the realtime file) and continued. Therefore, mitigate this issue in several ways: First, if we try to write out an inode with both rtinherit/extszinherit set and an unaligned extent size hint, turn off the hint to correct the error. Second, if someone tries to misconfigure a directory via the fssetxattr ioctl, fail the ioctl. Third, reverify both extent size hint values when we propagate heritable inode attributes from parent to child, to prevent misconfigurations from spreading. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2021-05-12 19:51:26 +00:00
/*
* Inode verifiers on older kernels only check that the extent size
* hint is an integer multiple of the rt extent size on realtime files.
* They did not check the hint alignment on a directory with both
* rtinherit and extszinherit flags set. If the misaligned hint is
* propagated from a directory into a new realtime file, new file
* allocations will fail due to math errors in the rt allocator and/or
* trip the verifiers. Validate the hint settings in the new file so
* that we don't let broken hints propagate.
*/
failaddr = xfs_inode_validate_extsize(ip->i_mount, ip->i_extsize,
VFS_I(ip)->i_mode, ip->i_diflags);
if (failaddr) {
ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
XFS_DIFLAG_EXTSZINHERIT);
ip->i_extsize = 0;
}
}
/* Propagate di_flags2 from a parent inode to a child inode. */
static void
xfs_inode_inherit_flags2(
struct xfs_inode *ip,
const struct xfs_inode *pip)
{
xfs: validate extsz hints against rt extent size when rtinherit is set The RTINHERIT bit can be set on a directory so that newly created regular files will have the REALTIME bit set to store their data on the realtime volume. If an extent size hint (and EXTSZINHERIT) are set on the directory, the hint will also be copied into the new file. As pointed out in previous patches, for realtime files we require the extent size hint be an integer multiple of the realtime extent, but we don't perform the same validation on a directory with both RTINHERIT and EXTSZINHERIT set, even though the only use-case of that combination is to propagate extent size hints into new realtime files. This leads to inode corruption errors when the bad values are propagated. Because there may be existing filesystems with such a configuration, we cannot simply amend the inode verifier to trip on these directories and call it a day because that will cause previously "working" filesystems to start throwing errors abruptly. Note that it's valid to have directories with rtinherit set even if there is no realtime volume, in which case the problem does not manifest because rtinherit is ignored if there's no realtime device; and it's possible that someone set the flag, crashed, repaired the filesystem (which clears the hint on the realtime file) and continued. Therefore, mitigate this issue in several ways: First, if we try to write out an inode with both rtinherit/extszinherit set and an unaligned extent size hint, turn off the hint to correct the error. Second, if someone tries to misconfigure a directory via the fssetxattr ioctl, fail the ioctl. Third, reverify both extent size hint values when we propagate heritable inode attributes from parent to child, to prevent misconfigurations from spreading. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2021-05-12 19:51:26 +00:00
xfs_failaddr_t failaddr;
if (pip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) {
ip->i_diflags2 |= XFS_DIFLAG2_COWEXTSIZE;
ip->i_cowextsize = pip->i_cowextsize;
}
if (pip->i_diflags2 & XFS_DIFLAG2_DAX)
ip->i_diflags2 |= XFS_DIFLAG2_DAX;
xfs: validate extsz hints against rt extent size when rtinherit is set The RTINHERIT bit can be set on a directory so that newly created regular files will have the REALTIME bit set to store their data on the realtime volume. If an extent size hint (and EXTSZINHERIT) are set on the directory, the hint will also be copied into the new file. As pointed out in previous patches, for realtime files we require the extent size hint be an integer multiple of the realtime extent, but we don't perform the same validation on a directory with both RTINHERIT and EXTSZINHERIT set, even though the only use-case of that combination is to propagate extent size hints into new realtime files. This leads to inode corruption errors when the bad values are propagated. Because there may be existing filesystems with such a configuration, we cannot simply amend the inode verifier to trip on these directories and call it a day because that will cause previously "working" filesystems to start throwing errors abruptly. Note that it's valid to have directories with rtinherit set even if there is no realtime volume, in which case the problem does not manifest because rtinherit is ignored if there's no realtime device; and it's possible that someone set the flag, crashed, repaired the filesystem (which clears the hint on the realtime file) and continued. Therefore, mitigate this issue in several ways: First, if we try to write out an inode with both rtinherit/extszinherit set and an unaligned extent size hint, turn off the hint to correct the error. Second, if someone tries to misconfigure a directory via the fssetxattr ioctl, fail the ioctl. Third, reverify both extent size hint values when we propagate heritable inode attributes from parent to child, to prevent misconfigurations from spreading. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com>
2021-05-12 19:51:26 +00:00
/* Don't let invalid cowextsize hints propagate. */
failaddr = xfs_inode_validate_cowextsize(ip->i_mount, ip->i_cowextsize,
VFS_I(ip)->i_mode, ip->i_diflags, ip->i_diflags2);
if (failaddr) {
ip->i_diflags2 &= ~XFS_DIFLAG2_COWEXTSIZE;
ip->i_cowextsize = 0;
}
}
/*
* Initialise a newly allocated inode and return the in-core inode to the
* caller locked exclusively.
*/
int
xfs_init_new_inode(
struct user_namespace *mnt_userns,
struct xfs_trans *tp,
struct xfs_inode *pip,
xfs_ino_t ino,
umode_t mode,
xfs_nlink_t nlink,
dev_t rdev,
prid_t prid,
xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com>
2021-03-22 16:52:03 +00:00
bool init_xattrs,
struct xfs_inode **ipp)
{
struct inode *dir = pip ? VFS_I(pip) : NULL;
struct xfs_mount *mp = tp->t_mountp;
struct xfs_inode *ip;
unsigned int flags;
int error;
struct timespec64 tv;
struct inode *inode;
/*
* Protect against obviously corrupt allocation btree records. Later
* xfs_iget checks will catch re-allocation of other active in-memory
* and on-disk inodes. If we don't catch reallocating the parent inode
* here we will deadlock in xfs_iget() so we have to do these checks
* first.
*/
if ((pip && ino == pip->i_ino) || !xfs_verify_dir_ino(mp, ino)) {
xfs_alert(mp, "Allocated a known in-use inode 0x%llx!", ino);
return -EFSCORRUPTED;
}
/*
* Get the in-core inode with the lock held exclusively to prevent
* others from looking at until we're done.
*/
error = xfs_iget(mp, tp, ino, XFS_IGET_CREATE, XFS_ILOCK_EXCL, &ip);
if (error)
return error;
ASSERT(ip != NULL);
inode = VFS_I(ip);
set_nlink(inode, nlink);
inode->i_rdev = rdev;
ip->i_projid = prid;
if (dir && !(dir->i_mode & S_ISGID) &&
(mp->m_flags & XFS_MOUNT_GRPID)) {
inode_fsuid_set(inode, mnt_userns);
inode->i_gid = dir->i_gid;
inode->i_mode = mode;
} else {
idmapped-mounts-v5.12 -----BEGIN PGP SIGNATURE----- iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCYCegywAKCRCRxhvAZXjc ouJ6AQDlf+7jCQlQdeKKoN9QDFfMzG1ooemat36EpRRTONaGuAD8D9A4sUsG4+5f 4IU5Lj9oY4DEmF8HenbWK2ZHsesL2Qg= =yPaw -----END PGP SIGNATURE----- Merge tag 'idmapped-mounts-v5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux Pull idmapped mounts from Christian Brauner: "This introduces idmapped mounts which has been in the making for some time. Simply put, different mounts can expose the same file or directory with different ownership. This initial implementation comes with ports for fat, ext4 and with Christoph's port for xfs with more filesystems being actively worked on by independent people and maintainers. Idmapping mounts handle a wide range of long standing use-cases. Here are just a few: - Idmapped mounts make it possible to easily share files between multiple users or multiple machines especially in complex scenarios. For example, idmapped mounts will be used in the implementation of portable home directories in systemd-homed.service(8) where they allow users to move their home directory to an external storage device and use it on multiple computers where they are assigned different uids and gids. This effectively makes it possible to assign random uids and gids at login time. - It is possible to share files from the host with unprivileged containers without having to change ownership permanently through chown(2). - It is possible to idmap a container's rootfs and without having to mangle every file. For example, Chromebooks use it to share the user's Download folder with their unprivileged containers in their Linux subsystem. - It is possible to share files between containers with non-overlapping idmappings. - Filesystem that lack a proper concept of ownership such as fat can use idmapped mounts to implement discretionary access (DAC) permission checking. - They allow users to efficiently changing ownership on a per-mount basis without having to (recursively) chown(2) all files. In contrast to chown (2) changing ownership of large sets of files is instantenous with idmapped mounts. This is especially useful when ownership of a whole root filesystem of a virtual machine or container is changed. With idmapped mounts a single syscall mount_setattr syscall will be sufficient to change the ownership of all files. - Idmapped mounts always take the current ownership into account as idmappings specify what a given uid or gid is supposed to be mapped to. This contrasts with the chown(2) syscall which cannot by itself take the current ownership of the files it changes into account. It simply changes the ownership to the specified uid and gid. This is especially problematic when recursively chown(2)ing a large set of files which is commong with the aforementioned portable home directory and container and vm scenario. - Idmapped mounts allow to change ownership locally, restricting it to specific mounts, and temporarily as the ownership changes only apply as long as the mount exists. Several userspace projects have either already put up patches and pull-requests for this feature or will do so should you decide to pull this: - systemd: In a wide variety of scenarios but especially right away in their implementation of portable home directories. https://systemd.io/HOME_DIRECTORY/ - container runtimes: containerd, runC, LXD:To share data between host and unprivileged containers, unprivileged and privileged containers, etc. The pull request for idmapped mounts support in containerd, the default Kubernetes runtime is already up for quite a while now: https://github.com/containerd/containerd/pull/4734 - The virtio-fs developers and several users have expressed interest in using this feature with virtual machines once virtio-fs is ported. - ChromeOS: Sharing host-directories with unprivileged containers. I've tightly synced with all those projects and all of those listed here have also expressed their need/desire for this feature on the mailing list. For more info on how people use this there's a bunch of talks about this too. Here's just two recent ones: https://www.cncf.io/wp-content/uploads/2020/12/Rootless-Containers-in-Gitpod.pdf https://fosdem.org/2021/schedule/event/containers_idmap/ This comes with an extensive xfstests suite covering both ext4 and xfs: https://git.kernel.org/brauner/xfstests-dev/h/idmapped_mounts It covers truncation, creation, opening, xattrs, vfscaps, setid execution, setgid inheritance and more both with idmapped and non-idmapped mounts. It already helped to discover an unrelated xfs setgid inheritance bug which has since been fixed in mainline. It will be sent for inclusion with the xfstests project should you decide to merge this. In order to support per-mount idmappings vfsmounts are marked with user namespaces. The idmapping of the user namespace will be used to map the ids of vfs objects when they are accessed through that mount. By default all vfsmounts are marked with the initial user namespace. The initial user namespace is used to indicate that a mount is not idmapped. All operations behave as before and this is verified in the testsuite. Based on prior discussions we want to attach the whole user namespace and not just a dedicated idmapping struct. This allows us to reuse all the helpers that already exist for dealing with idmappings instead of introducing a whole new range of helpers. In addition, if we decide in the future that we are confident enough to enable unprivileged users to setup idmapped mounts the permission checking can take into account whether the caller is privileged in the user namespace the mount is currently marked with. The user namespace the mount will be marked with can be specified by passing a file descriptor refering to the user namespace as an argument to the new mount_setattr() syscall together with the new MOUNT_ATTR_IDMAP flag. The system call follows the openat2() pattern of extensibility. The following conditions must be met in order to create an idmapped mount: - The caller must currently have the CAP_SYS_ADMIN capability in the user namespace the underlying filesystem has been mounted in. - The underlying filesystem must support idmapped mounts. - The mount must not already be idmapped. This also implies that the idmapping of a mount cannot be altered once it has been idmapped. - The mount must be a detached/anonymous mount, i.e. it must have been created by calling open_tree() with the OPEN_TREE_CLONE flag and it must not already have been visible in the filesystem. The last two points guarantee easier semantics for userspace and the kernel and make the implementation significantly simpler. By default vfsmounts are marked with the initial user namespace and no behavioral or performance changes are observed. The manpage with a detailed description can be found here: https://git.kernel.org/brauner/man-pages/c/1d7b902e2875a1ff342e036a9f866a995640aea8 In order to support idmapped mounts, filesystems need to be changed and mark themselves with the FS_ALLOW_IDMAP flag in fs_flags. The patches to convert individual filesystem are not very large or complicated overall as can be seen from the included fat, ext4, and xfs ports. Patches for other filesystems are actively worked on and will be sent out separately. The xfstestsuite can be used to verify that port has been done correctly. The mount_setattr() syscall is motivated independent of the idmapped mounts patches and it's been around since July 2019. One of the most valuable features of the new mount api is the ability to perform mounts based on file descriptors only. Together with the lookup restrictions available in the openat2() RESOLVE_* flag namespace which we added in v5.6 this is the first time we are close to hardened and race-free (e.g. symlinks) mounting and path resolution. While userspace has started porting to the new mount api to mount proper filesystems and create new bind-mounts it is currently not possible to change mount options of an already existing bind mount in the new mount api since the mount_setattr() syscall is missing. With the addition of the mount_setattr() syscall we remove this last restriction and userspace can now fully port to the new mount api, covering every use-case the old mount api could. We also add the crucial ability to recursively change mount options for a whole mount tree, both removing and adding mount options at the same time. This syscall has been requested multiple times by various people and projects. There is a simple tool available at https://github.com/brauner/mount-idmapped that allows to create idmapped mounts so people can play with this patch series. I'll add support for the regular mount binary should you decide to pull this in the following weeks: Here's an example to a simple idmapped mount of another user's home directory: u1001@f2-vm:/$ sudo ./mount --idmap both:1000:1001:1 /home/ubuntu/ /mnt u1001@f2-vm:/$ ls -al /home/ubuntu/ total 28 drwxr-xr-x 2 ubuntu ubuntu 4096 Oct 28 22:07 . drwxr-xr-x 4 root root 4096 Oct 28 04:00 .. -rw------- 1 ubuntu ubuntu 3154 Oct 28 22:12 .bash_history -rw-r--r-- 1 ubuntu ubuntu 220 Feb 25 2020 .bash_logout -rw-r--r-- 1 ubuntu ubuntu 3771 Feb 25 2020 .bashrc -rw-r--r-- 1 ubuntu ubuntu 807 Feb 25 2020 .profile -rw-r--r-- 1 ubuntu ubuntu 0 Oct 16 16:11 .sudo_as_admin_successful -rw------- 1 ubuntu ubuntu 1144 Oct 28 00:43 .viminfo u1001@f2-vm:/$ ls -al /mnt/ total 28 drwxr-xr-x 2 u1001 u1001 4096 Oct 28 22:07 . drwxr-xr-x 29 root root 4096 Oct 28 22:01 .. -rw------- 1 u1001 u1001 3154 Oct 28 22:12 .bash_history -rw-r--r-- 1 u1001 u1001 220 Feb 25 2020 .bash_logout -rw-r--r-- 1 u1001 u1001 3771 Feb 25 2020 .bashrc -rw-r--r-- 1 u1001 u1001 807 Feb 25 2020 .profile -rw-r--r-- 1 u1001 u1001 0 Oct 16 16:11 .sudo_as_admin_successful -rw------- 1 u1001 u1001 1144 Oct 28 00:43 .viminfo u1001@f2-vm:/$ touch /mnt/my-file u1001@f2-vm:/$ setfacl -m u:1001:rwx /mnt/my-file u1001@f2-vm:/$ sudo setcap -n 1001 cap_net_raw+ep /mnt/my-file u1001@f2-vm:/$ ls -al /mnt/my-file -rw-rwxr--+ 1 u1001 u1001 0 Oct 28 22:14 /mnt/my-file u1001@f2-vm:/$ ls -al /home/ubuntu/my-file -rw-rwxr--+ 1 ubuntu ubuntu 0 Oct 28 22:14 /home/ubuntu/my-file u1001@f2-vm:/$ getfacl /mnt/my-file getfacl: Removing leading '/' from absolute path names # file: mnt/my-file # owner: u1001 # group: u1001 user::rw- user:u1001:rwx group::rw- mask::rwx other::r-- u1001@f2-vm:/$ getfacl /home/ubuntu/my-file getfacl: Removing leading '/' from absolute path names # file: home/ubuntu/my-file # owner: ubuntu # group: ubuntu user::rw- user:ubuntu:rwx group::rw- mask::rwx other::r--" * tag 'idmapped-mounts-v5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux: (41 commits) xfs: remove the possibly unused mp variable in xfs_file_compat_ioctl xfs: support idmapped mounts ext4: support idmapped mounts fat: handle idmapped mounts tests: add mount_setattr() selftests fs: introduce MOUNT_ATTR_IDMAP fs: add mount_setattr() fs: add attr_flags_to_mnt_flags helper fs: split out functions to hold writers namespace: only take read lock in do_reconfigure_mnt() mount: make {lock,unlock}_mount_hash() static namespace: take lock_mount_hash() directly when changing flags nfs: do not export idmapped mounts overlayfs: do not mount on top of idmapped mounts ecryptfs: do not mount on top of idmapped mounts ima: handle idmapped mounts apparmor: handle idmapped mounts fs: make helpers idmap mount aware exec: handle idmapped mounts would_dump: handle idmapped mounts ...
2021-02-23 21:39:45 +00:00
inode_init_owner(mnt_userns, inode, dir, mode);
}
/*
* If the group ID of the new file does not match the effective group
* ID or one of the supplementary group IDs, the S_ISGID bit is cleared
* (and only if the irix_sgid_inherit compatibility variable is set).
*/
if (irix_sgid_inherit &&
(inode->i_mode & S_ISGID) &&
!in_group_p(i_gid_into_mnt(mnt_userns, inode)))
inode->i_mode &= ~S_ISGID;
ip->i_disk_size = 0;
ip->i_df.if_nextents = 0;
ASSERT(ip->i_nblocks == 0);
tv = current_time(inode);
inode->i_mtime = tv;
inode->i_atime = tv;
inode->i_ctime = tv;
ip->i_extsize = 0;
ip->i_diflags = 0;
if (xfs_sb_version_has_v3inode(&mp->m_sb)) {
inode_set_iversion(inode, 1);
ip->i_cowextsize = 0;
ip->i_crtime = tv;
}
flags = XFS_ILOG_CORE;
switch (mode & S_IFMT) {
case S_IFIFO:
case S_IFCHR:
case S_IFBLK:
case S_IFSOCK:
ip->i_df.if_format = XFS_DINODE_FMT_DEV;
flags |= XFS_ILOG_DEV;
break;
case S_IFREG:
case S_IFDIR:
if (pip && (pip->i_diflags & XFS_DIFLAG_ANY))
xfs_inode_inherit_flags(ip, pip);
if (pip && (pip->i_diflags2 & XFS_DIFLAG2_ANY))
xfs_inode_inherit_flags2(ip, pip);
xfs: Fix fall-through warnings for Clang In preparation to enable -Wimplicit-fallthrough for Clang, fix the following warnings by replacing /* fall through */ comments, and its variants, with the new pseudo-keyword macro fallthrough: fs/xfs/libxfs/xfs_alloc.c:3167:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/libxfs/xfs_da_btree.c:286:3: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/libxfs/xfs_ag_resv.c:346:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/libxfs/xfs_ag_resv.c:388:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_bmap_util.c:246:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_export.c:88:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_export.c:96:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_file.c:867:3: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_ioctl.c:562:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_ioctl.c:1548:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_iomap.c:1040:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_inode.c:852:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_log.c:2627:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/xfs_trans_buf.c:298:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/bmap.c:275:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/btree.c:48:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/common.c:85:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/common.c:138:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/common.c:698:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/dabtree.c:51:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/repair.c:951:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] fs/xfs/scrub/agheader.c:89:2: warning: unannotated fall-through between switch labels [-Wimplicit-fallthrough] Notice that Clang doesn't recognize /* fall through */ comments as implicit fall-through markings, so in order to globally enable -Wimplicit-fallthrough for Clang, these comments need to be replaced with fallthrough; in the whole codebase. Link: https://github.com/KSPP/linux/issues/115 Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
2021-04-20 22:54:36 +00:00
fallthrough;
case S_IFLNK:
ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
ip->i_df.if_bytes = 0;
xfs: use a b+tree for the in-core extent list Replace the current linear list and the indirection array for the in-core extent list with a b+tree to avoid the need for larger memory allocations for the indirection array when lots of extents are present. The current extent list implementations leads to heavy pressure on the memory allocator when modifying files with a high extent count, and can lead to high latencies because of that. The replacement is a b+tree with a few quirks. The leaf nodes directly store the extent record in two u64 values. The encoding is a little bit different from the existing in-core extent records so that the start offset and length which are required for lookups can be retreived with simple mask operations. The inner nodes store a 64-bit key containing the start offset in the first half of the node, and the pointers to the next lower level in the second half. In either case we walk the node from the beginninig to the end and do a linear search, as that is more efficient for the low number of cache lines touched during a search (2 for the inner nodes, 4 for the leaf nodes) than a binary search. We store termination markers (zero length for the leaf nodes, an otherwise impossible high bit for the inner nodes) to terminate the key list / records instead of storing a count to use the available cache lines as efficiently as possible. One quirk of the algorithm is that while we normally split a node half and half like usual btree implementations we just spill over entries added at the very end of the list to a new node on its own. This means we get a 100% fill grade for the common cases of bulk insertion when reading an inode into memory, and when only sequentially appending to a file. The downside is a slightly higher chance of splits on the first random insertions. Both insert and removal manually recurse into the lower levels, but the bulk deletion of the whole tree is still implemented as a recursive function call, although one limited by the overall depth and with very little stack usage in every iteration. For the first few extents we dynamically grow the list from a single extent to the next powers of two until we have a first full leaf block and that building the actual tree. The code started out based on the generic lib/btree.c code from Joern Engel based on earlier work from Peter Zijlstra, but has since been rewritten beyond recognition. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2017-11-03 17:34:46 +00:00
ip->i_df.if_u1.if_root = NULL;
break;
default:
ASSERT(0);
}
xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com>
2021-03-22 16:52:03 +00:00
/*
* If we need to create attributes immediately after allocating the
* inode, initialise an empty attribute fork right now. We use the
* default fork offset for attributes here as we don't know exactly what
* size or how many attributes we might be adding. We can do this
* safely here because we know the data fork is completely empty and
* this saves us from needing to run a separate transaction to set the
* fork offset in the immediate future.
*/
if (init_xattrs && xfs_sb_version_hasattr(&mp->m_sb)) {
ip->i_forkoff = xfs_default_attroffset(ip) >> 3;
xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com>
2021-03-22 16:52:03 +00:00
ip->i_afp = xfs_ifork_alloc(XFS_DINODE_FMT_EXTENTS, 0);
}
/*
* Log the new values stuffed into the inode.
*/
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
xfs_trans_log_inode(tp, ip, flags);
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
/* now that we have an i_mode we can setup the inode structure */
xfs_setup_inode(ip);
*ipp = ip;
return 0;
}
/*
* Decrement the link count on an inode & log the change. If this causes the
* link count to go to zero, move the inode to AGI unlinked list so that it can
* be freed when the last active reference goes away via xfs_inactive().
*/
static int /* error */
xfs_droplink(
xfs_trans_t *tp,
xfs_inode_t *ip)
{
xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
drop_nlink(VFS_I(ip));
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
if (VFS_I(ip)->i_nlink)
return 0;
return xfs_iunlink(tp, ip);
}
/*
* Increment the link count on an inode & log the change.
*/
static void
xfs_bumplink(
xfs_trans_t *tp,
xfs_inode_t *ip)
{
xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_CHG);
inc_nlink(VFS_I(ip));
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
}
int
xfs_create(
struct user_namespace *mnt_userns,
xfs_inode_t *dp,
struct xfs_name *name,
umode_t mode,
dev_t rdev,
xfs: initialise attr fork on inode create When we allocate a new inode, we often need to add an attribute to the inode as part of the create. This can happen as a result of needing to add default ACLs or security labels before the inode is made visible to userspace. This is highly inefficient right now. We do the create transaction to allocate the inode, then we do an "add attr fork" transaction to modify the just created empty inode to set the inode fork offset to allow attributes to be stored, then we go and do the attribute creation. This means 3 transactions instead of 1 to allocate an inode, and this greatly increases the load on the CIL commit code, resulting in excessive contention on the CIL spin locks and performance degradation: 18.99% [kernel] [k] __pv_queued_spin_lock_slowpath 3.57% [kernel] [k] do_raw_spin_lock 2.51% [kernel] [k] __raw_callee_save___pv_queued_spin_unlock 2.48% [kernel] [k] memcpy 2.34% [kernel] [k] xfs_log_commit_cil The typical profile resulting from running fsmark on a selinux enabled filesytem is adds this overhead to the create path: - 15.30% xfs_init_security - 15.23% security_inode_init_security - 13.05% xfs_initxattrs - 12.94% xfs_attr_set - 6.75% xfs_bmap_add_attrfork - 5.51% xfs_trans_commit - 5.48% __xfs_trans_commit - 5.35% xfs_log_commit_cil - 3.86% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.70% xfs_trans_alloc 0.52% xfs_trans_reserve - 5.41% xfs_attr_set_args - 5.39% xfs_attr_set_shortform.constprop.0 - 4.46% xfs_trans_commit - 4.46% __xfs_trans_commit - 4.33% xfs_log_commit_cil - 2.74% _raw_spin_lock - do_raw_spin_lock __pv_queued_spin_lock_slowpath 0.60% xfs_inode_item_format 0.90% xfs_attr_try_sf_addname - 1.99% selinux_inode_init_security - 1.02% security_sid_to_context_force - 1.00% security_sid_to_context_core - 0.92% sidtab_entry_to_string - 0.90% sidtab_sid2str_get 0.59% sidtab_sid2str_put.part.0 - 0.82% selinux_determine_inode_label - 0.77% security_transition_sid 0.70% security_compute_sid.part.0 And fsmark creation rate performance drops by ~25%. The key point to note here is that half the additional overhead comes from adding the attribute fork to the newly created inode. That's crazy, considering we can do this same thing at inode create time with a couple of lines of code and no extra overhead. So, if we know we are going to add an attribute immediately after creating the inode, let's just initialise the attribute fork inside the create transaction and chop that whole chunk of code out of the create fast path. This completely removes the performance drop caused by enabling SELinux, and the profile looks like: - 8.99% xfs_init_security - 9.00% security_inode_init_security - 6.43% xfs_initxattrs - 6.37% xfs_attr_set - 5.45% xfs_attr_set_args - 5.42% xfs_attr_set_shortform.constprop.0 - 4.51% xfs_trans_commit - 4.54% __xfs_trans_commit - 4.59% xfs_log_commit_cil - 2.67% _raw_spin_lock - 3.28% do_raw_spin_lock 3.08% __pv_queued_spin_lock_slowpath 0.66% xfs_inode_item_format - 0.90% xfs_attr_try_sf_addname - 0.60% xfs_trans_alloc - 2.35% selinux_inode_init_security - 1.25% security_sid_to_context_force - 1.21% security_sid_to_context_core - 1.19% sidtab_entry_to_string - 1.20% sidtab_sid2str_get - 0.86% sidtab_sid2str_put.part.0 - 0.62% _raw_spin_lock_irqsave - 0.77% do_raw_spin_lock __pv_queued_spin_lock_slowpath - 0.84% selinux_determine_inode_label - 0.83% security_transition_sid 0.86% security_compute_sid.part.0 Which indicates the XFS overhead of creating the selinux xattr has been halved. This doesn't fix the CIL lock contention problem, just means it's not a limiting factor for this workload. Lock contention in the security subsystems is going to be an issue soon, though... Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> [djwong: fix compilation error when CONFIG_SECURITY=n] Reviewed-by: Darrick J. Wong <djwong@kernel.org> Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Gao Xiang <hsiangkao@redhat.com>
2021-03-22 16:52:03 +00:00
bool init_xattrs,
xfs_inode_t **ipp)
{
int is_dir = S_ISDIR(mode);
struct xfs_mount *mp = dp->i_mount;
struct xfs_inode *ip = NULL;
struct xfs_trans *tp = NULL;
int error;
bool unlock_dp_on_error = false;
prid_t prid;
struct xfs_dquot *udqp = NULL;
struct xfs_dquot *gdqp = NULL;
struct xfs_dquot *pdqp = NULL;
struct xfs_trans_res *tres;
uint resblks;
xfs_ino_t ino;
trace_xfs_create(dp, name);
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
prid = xfs_get_initial_prid(dp);
/*
* Make sure that we have allocated dquot(s) on disk.
*/
error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
mapped_fsgid(mnt_userns), prid,
XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
&udqp, &gdqp, &pdqp);
if (error)
return error;
if (is_dir) {
resblks = XFS_MKDIR_SPACE_RES(mp, name->len);
tres = &M_RES(mp)->tr_mkdir;
} else {
resblks = XFS_CREATE_SPACE_RES(mp, name->len);
tres = &M_RES(mp)->tr_create;
}
/*
* Initially assume that the file does not exist and
* reserve the resources for that case. If that is not
* the case we'll drop the one we have and get a more
* appropriate transaction later.
*/
error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
&tp);
if (error == -ENOSPC) {
/* flush outstanding delalloc blocks and retry */
xfs_flush_inodes(mp);
error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp,
resblks, &tp);
}
if (error)
goto out_release_dquots;
xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
unlock_dp_on_error = true;
error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK,
XFS_IEXT_DIR_MANIP_CNT(mp));
if (error)
goto out_trans_cancel;
/*
* A newly created regular or special file just has one directory
* entry pointing to them, but a directory also the "." entry
* pointing to itself.
*/
error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
if (!error)
error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
is_dir ? 2 : 1, rdev, prid, init_xattrs, &ip);
if (error)
goto out_trans_cancel;
/*
* Now we join the directory inode to the transaction. We do not do it
* earlier because xfs_dialloc might commit the previous transaction
* (and release all the locks). An error from here on will result in
* the transaction cancel unlocking dp so don't do it explicitly in the
* error path.
*/
xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
unlock_dp_on_error = false;
error = xfs_dir_createname(tp, dp, name, ip->i_ino,
resblks - XFS_IALLOC_SPACE_RES(mp));
if (error) {
ASSERT(error != -ENOSPC);
goto out_trans_cancel;
}
xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
if (is_dir) {
error = xfs_dir_init(tp, ip, dp);
if (error)
goto out_trans_cancel;
xfs_bumplink(tp, dp);
}
/*
* If this is a synchronous mount, make sure that the
* create transaction goes to disk before returning to
* the user.
*/
if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
xfs_trans_set_sync(tp);
/*
* Attach the dquot(s) to the inodes and modify them incore.
* These ids of the inode couldn't have changed since the new
* inode has been locked ever since it was created.
*/
xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
error = xfs_trans_commit(tp);
if (error)
goto out_release_inode;
xfs_qm_dqrele(udqp);
xfs_qm_dqrele(gdqp);
xfs_qm_dqrele(pdqp);
*ipp = ip;
return 0;
out_trans_cancel:
xfs_trans_cancel(tp);
out_release_inode:
/*
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
* Wait until after the current transaction is aborted to finish the
* setup of the inode and release the inode. This prevents recursive
* transactions and deadlocks from xfs_inactive.
*/
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
if (ip) {
xfs_finish_inode_setup(ip);
xfs_irele(ip);
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
}
out_release_dquots:
xfs_qm_dqrele(udqp);
xfs_qm_dqrele(gdqp);
xfs_qm_dqrele(pdqp);
if (unlock_dp_on_error)
xfs_iunlock(dp, XFS_ILOCK_EXCL);
return error;
}
int
xfs_create_tmpfile(
struct user_namespace *mnt_userns,
struct xfs_inode *dp,
xfs: fix tmpfile/selinux deadlock and initialize security xfstests generic/004 reproduces an ilock deadlock using the tmpfile interface when selinux is enabled. This occurs because xfs_create_tmpfile() takes the ilock and then calls d_tmpfile(). The latter eventually calls into xfs_xattr_get() which attempts to get the lock again. E.g.: xfs_io D ffffffff81c134c0 4096 3561 3560 0x00000080 ffff8801176a1a68 0000000000000046 ffff8800b401b540 ffff8801176a1fd8 00000000001d5800 00000000001d5800 ffff8800b401b540 ffff8800b401b540 ffff8800b73a6bd0 fffffffeffffffff ffff8800b73a6bd8 ffff8800b5ddb480 Call Trace: [<ffffffff8177f969>] schedule+0x29/0x70 [<ffffffff81783a65>] rwsem_down_read_failed+0xc5/0x120 [<ffffffffa05aa97f>] ? xfs_ilock_attr_map_shared+0x1f/0x50 [xfs] [<ffffffff813b3434>] call_rwsem_down_read_failed+0x14/0x30 [<ffffffff810ed179>] ? down_read_nested+0x89/0xa0 [<ffffffffa05aa7f2>] ? xfs_ilock+0x122/0x250 [xfs] [<ffffffffa05aa7f2>] xfs_ilock+0x122/0x250 [xfs] [<ffffffffa05aa97f>] xfs_ilock_attr_map_shared+0x1f/0x50 [xfs] [<ffffffffa05701d0>] xfs_attr_get+0x90/0xe0 [xfs] [<ffffffffa0565e07>] xfs_xattr_get+0x37/0x50 [xfs] [<ffffffff8124842f>] generic_getxattr+0x4f/0x70 [<ffffffff8133fd9e>] inode_doinit_with_dentry+0x1ae/0x650 [<ffffffff81340e0c>] selinux_d_instantiate+0x1c/0x20 [<ffffffff813351bb>] security_d_instantiate+0x1b/0x30 [<ffffffff81237db0>] d_instantiate+0x50/0x70 [<ffffffff81237e85>] d_tmpfile+0xb5/0xc0 [<ffffffffa05add02>] xfs_create_tmpfile+0x362/0x410 [xfs] [<ffffffffa0559ac8>] xfs_vn_tmpfile+0x18/0x20 [xfs] [<ffffffff81230388>] path_openat+0x228/0x6a0 [<ffffffff810230f9>] ? sched_clock+0x9/0x10 [<ffffffff8105a427>] ? kvm_clock_read+0x27/0x40 [<ffffffff8124054f>] ? __alloc_fd+0xaf/0x1f0 [<ffffffff8123101a>] do_filp_open+0x3a/0x90 [<ffffffff817845e7>] ? _raw_spin_unlock+0x27/0x40 [<ffffffff8124054f>] ? __alloc_fd+0xaf/0x1f0 [<ffffffff8121e3ce>] do_sys_open+0x12e/0x210 [<ffffffff8121e4ce>] SyS_open+0x1e/0x20 [<ffffffff8178eda9>] system_call_fastpath+0x16/0x1b xfs_vn_tmpfile() also fails to initialize security on the newly created inode. Pull the d_tmpfile() call up into xfs_vn_tmpfile() after the transaction has been committed and the inode unlocked. Also, initialize security on the inode based on the parent directory provided via the tmpfile call. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-16 22:15:30 +00:00
umode_t mode,
struct xfs_inode **ipp)
{
struct xfs_mount *mp = dp->i_mount;
struct xfs_inode *ip = NULL;
struct xfs_trans *tp = NULL;
int error;
prid_t prid;
struct xfs_dquot *udqp = NULL;
struct xfs_dquot *gdqp = NULL;
struct xfs_dquot *pdqp = NULL;
struct xfs_trans_res *tres;
uint resblks;
xfs_ino_t ino;
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
prid = xfs_get_initial_prid(dp);
/*
* Make sure that we have allocated dquot(s) on disk.
*/
error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns),
mapped_fsgid(mnt_userns), prid,
XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT,
&udqp, &gdqp, &pdqp);
if (error)
return error;
resblks = XFS_IALLOC_SPACE_RES(mp);
tres = &M_RES(mp)->tr_create_tmpfile;
error = xfs_trans_alloc_icreate(mp, tres, udqp, gdqp, pdqp, resblks,
&tp);
if (error)
goto out_release_dquots;
error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
if (!error)
error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
0, 0, prid, false, &ip);
if (error)
goto out_trans_cancel;
if (mp->m_flags & XFS_MOUNT_WSYNC)
xfs_trans_set_sync(tp);
/*
* Attach the dquot(s) to the inodes and modify them incore.
* These ids of the inode couldn't have changed since the new
* inode has been locked ever since it was created.
*/
xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
error = xfs_iunlink(tp, ip);
if (error)
goto out_trans_cancel;
error = xfs_trans_commit(tp);
if (error)
goto out_release_inode;
xfs_qm_dqrele(udqp);
xfs_qm_dqrele(gdqp);
xfs_qm_dqrele(pdqp);
xfs: fix tmpfile/selinux deadlock and initialize security xfstests generic/004 reproduces an ilock deadlock using the tmpfile interface when selinux is enabled. This occurs because xfs_create_tmpfile() takes the ilock and then calls d_tmpfile(). The latter eventually calls into xfs_xattr_get() which attempts to get the lock again. E.g.: xfs_io D ffffffff81c134c0 4096 3561 3560 0x00000080 ffff8801176a1a68 0000000000000046 ffff8800b401b540 ffff8801176a1fd8 00000000001d5800 00000000001d5800 ffff8800b401b540 ffff8800b401b540 ffff8800b73a6bd0 fffffffeffffffff ffff8800b73a6bd8 ffff8800b5ddb480 Call Trace: [<ffffffff8177f969>] schedule+0x29/0x70 [<ffffffff81783a65>] rwsem_down_read_failed+0xc5/0x120 [<ffffffffa05aa97f>] ? xfs_ilock_attr_map_shared+0x1f/0x50 [xfs] [<ffffffff813b3434>] call_rwsem_down_read_failed+0x14/0x30 [<ffffffff810ed179>] ? down_read_nested+0x89/0xa0 [<ffffffffa05aa7f2>] ? xfs_ilock+0x122/0x250 [xfs] [<ffffffffa05aa7f2>] xfs_ilock+0x122/0x250 [xfs] [<ffffffffa05aa97f>] xfs_ilock_attr_map_shared+0x1f/0x50 [xfs] [<ffffffffa05701d0>] xfs_attr_get+0x90/0xe0 [xfs] [<ffffffffa0565e07>] xfs_xattr_get+0x37/0x50 [xfs] [<ffffffff8124842f>] generic_getxattr+0x4f/0x70 [<ffffffff8133fd9e>] inode_doinit_with_dentry+0x1ae/0x650 [<ffffffff81340e0c>] selinux_d_instantiate+0x1c/0x20 [<ffffffff813351bb>] security_d_instantiate+0x1b/0x30 [<ffffffff81237db0>] d_instantiate+0x50/0x70 [<ffffffff81237e85>] d_tmpfile+0xb5/0xc0 [<ffffffffa05add02>] xfs_create_tmpfile+0x362/0x410 [xfs] [<ffffffffa0559ac8>] xfs_vn_tmpfile+0x18/0x20 [xfs] [<ffffffff81230388>] path_openat+0x228/0x6a0 [<ffffffff810230f9>] ? sched_clock+0x9/0x10 [<ffffffff8105a427>] ? kvm_clock_read+0x27/0x40 [<ffffffff8124054f>] ? __alloc_fd+0xaf/0x1f0 [<ffffffff8123101a>] do_filp_open+0x3a/0x90 [<ffffffff817845e7>] ? _raw_spin_unlock+0x27/0x40 [<ffffffff8124054f>] ? __alloc_fd+0xaf/0x1f0 [<ffffffff8121e3ce>] do_sys_open+0x12e/0x210 [<ffffffff8121e4ce>] SyS_open+0x1e/0x20 [<ffffffff8178eda9>] system_call_fastpath+0x16/0x1b xfs_vn_tmpfile() also fails to initialize security on the newly created inode. Pull the d_tmpfile() call up into xfs_vn_tmpfile() after the transaction has been committed and the inode unlocked. Also, initialize security on the inode based on the parent directory provided via the tmpfile call. Signed-off-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Dave Chinner <david@fromorbit.com>
2014-04-16 22:15:30 +00:00
*ipp = ip;
return 0;
out_trans_cancel:
xfs_trans_cancel(tp);
out_release_inode:
/*
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
* Wait until after the current transaction is aborted to finish the
* setup of the inode and release the inode. This prevents recursive
* transactions and deadlocks from xfs_inactive.
*/
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
if (ip) {
xfs_finish_inode_setup(ip);
xfs_irele(ip);
xfs: inodes are new until the dentry cache is set up Al Viro noticed a generic set of issues to do with filehandle lookup racing with dentry cache setup. They involve a filehandle lookup occurring while an inode is being created and the filehandle lookup racing with the dentry creation for the real file. This can lead to multiple dentries for the one path being instantiated. There are a host of other issues around this same set of paths. The underlying cause is that file handle lookup only waits on inode cache instantiation rather than full dentry cache instantiation. XFS is mostly immune to the problems discovered due to it's own internal inode cache, but there are a couple of corner cases where races can happen. We currently clear the XFS_INEW flag when the inode is fully set up after insertion into the cache. Newly allocated inodes are inserted locked and so aren't usable until the allocation transaction commits. This, however, occurs before the dentry and security information is fully initialised and hence the inode is unlocked and available for lookups to find too early. To solve the problem, only clear the XFS_INEW flag for newly created inodes once the dentry is fully instantiated. This means lookups will retry until the XFS_INEW flag is removed from the inode and hence avoids the race conditions in questions. THis also means that xfs_create(), xfs_create_tmpfile() and xfs_symlink() need to finish the setup of the inode in their error paths if we had allocated the inode but failed later in the creation process. xfs_symlink(), in particular, needed a lot of help to make it's error handling match that of xfs_create(). Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-02-23 11:38:08 +00:00
}
out_release_dquots:
xfs_qm_dqrele(udqp);
xfs_qm_dqrele(gdqp);
xfs_qm_dqrele(pdqp);
return error;
}
int
xfs_link(
xfs_inode_t *tdp,
xfs_inode_t *sip,
struct xfs_name *target_name)
{
xfs_mount_t *mp = tdp->i_mount;
xfs_trans_t *tp;
int error;
int resblks;
trace_xfs_link(tdp, target_name);
ASSERT(!S_ISDIR(VFS_I(sip)->i_mode));
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
error = xfs_qm_dqattach(sip);
if (error)
goto std_return;
error = xfs_qm_dqattach(tdp);
if (error)
goto std_return;
resblks = XFS_LINK_SPACE_RES(mp, target_name->len);
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, resblks, 0, 0, &tp);
if (error == -ENOSPC) {
resblks = 0;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_link, 0, 0, 0, &tp);
}
if (error)
goto std_return;
xfs_lock_two_inodes(sip, XFS_ILOCK_EXCL, tdp, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, sip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, tdp, XFS_ILOCK_EXCL);
error = xfs_iext_count_may_overflow(tdp, XFS_DATA_FORK,
XFS_IEXT_DIR_MANIP_CNT(mp));
if (error)
goto error_return;
/*
* If we are using project inheritance, we only allow hard link
* creation in our tree when the project IDs are the same; else
* the tree quota mechanism could be circumvented.
*/
if (unlikely((tdp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
tdp->i_projid != sip->i_projid)) {
error = -EXDEV;
goto error_return;
}
if (!resblks) {
error = xfs_dir_canenter(tp, tdp, target_name);
if (error)
goto error_return;
}
/*
* Handle initial link state of O_TMPFILE inode
*/
if (VFS_I(sip)->i_nlink == 0) {
struct xfs_perag *pag;
pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, sip->i_ino));
error = xfs_iunlink_remove(tp, pag, sip);
xfs_perag_put(pag);
if (error)
goto error_return;
}
error = xfs_dir_createname(tp, tdp, target_name, sip->i_ino,
resblks);
if (error)
goto error_return;
xfs_trans_ichgtime(tp, tdp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs_trans_log_inode(tp, tdp, XFS_ILOG_CORE);
xfs_bumplink(tp, sip);
/*
* If this is a synchronous mount, make sure that the
* link transaction goes to disk before returning to
* the user.
*/
if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
xfs_trans_set_sync(tp);
return xfs_trans_commit(tp);
error_return:
xfs_trans_cancel(tp);
std_return:
return error;
}
/* Clear the reflink flag and the cowblocks tag if possible. */
static void
xfs_itruncate_clear_reflink_flags(
struct xfs_inode *ip)
{
struct xfs_ifork *dfork;
struct xfs_ifork *cfork;
if (!xfs_is_reflink_inode(ip))
return;
dfork = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
cfork = XFS_IFORK_PTR(ip, XFS_COW_FORK);
if (dfork->if_bytes == 0 && cfork->if_bytes == 0)
ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
if (cfork->if_bytes == 0)
xfs_inode_clear_cowblocks_tag(ip);
}
/*
* Free up the underlying blocks past new_size. The new size must be smaller
* than the current size. This routine can be used both for the attribute and
* data fork, and does not modify the inode size, which is left to the caller.
*
* The transaction passed to this routine must have made a permanent log
* reservation of at least XFS_ITRUNCATE_LOG_RES. This routine may commit the
* given transaction and start new ones, so make sure everything involved in
* the transaction is tidy before calling here. Some transaction will be
* returned to the caller to be committed. The incoming transaction must
* already include the inode, and both inode locks must be held exclusively.
* The inode must also be "held" within the transaction. On return the inode
* will be "held" within the returned transaction. This routine does NOT
* require any disk space to be reserved for it within the transaction.
*
* If we get an error, we must return with the inode locked and linked into the
* current transaction. This keeps things simple for the higher level code,
* because it always knows that the inode is locked and held in the transaction
* that returns to it whether errors occur or not. We don't mark the inode
* dirty on error so that transactions can be easily aborted if possible.
*/
int
xfs_itruncate_extents_flags(
struct xfs_trans **tpp,
struct xfs_inode *ip,
int whichfork,
xfs_fsize_t new_size,
int flags)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_trans *tp = *tpp;
xfs_fileoff_t first_unmap_block;
xfs_filblks_t unmap_len;
int error = 0;
ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
ASSERT(!atomic_read(&VFS_I(ip)->i_count) ||
xfs_isilocked(ip, XFS_IOLOCK_EXCL));
ASSERT(new_size <= XFS_ISIZE(ip));
ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES);
ASSERT(ip->i_itemp != NULL);
ASSERT(ip->i_itemp->ili_lock_flags == 0);
ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
trace_xfs_itruncate_extents_start(ip, new_size);
flags |= xfs_bmapi_aflag(whichfork);
/*
* Since it is possible for space to become allocated beyond
* the end of the file (in a crash where the space is allocated
* but the inode size is not yet updated), simply remove any
* blocks which show up between the new EOF and the maximum
* possible file size.
*
* We have to free all the blocks to the bmbt maximum offset, even if
* the page cache can't scale that far.
*/
first_unmap_block = XFS_B_TO_FSB(mp, (xfs_ufsize_t)new_size);
if (!xfs_verify_fileoff(mp, first_unmap_block)) {
WARN_ON_ONCE(first_unmap_block > XFS_MAX_FILEOFF);
return 0;
}
unmap_len = XFS_MAX_FILEOFF - first_unmap_block + 1;
while (unmap_len > 0) {
ASSERT(tp->t_firstblock == NULLFSBLOCK);
error = __xfs_bunmapi(tp, ip, first_unmap_block, &unmap_len,
flags, XFS_ITRUNC_MAX_EXTENTS);
if (error)
goto out;
/* free the just unmapped extents */
error = xfs_defer_finish(&tp);
if (error)
goto out;
}
if (whichfork == XFS_DATA_FORK) {
/* Remove all pending CoW reservations. */
error = xfs_reflink_cancel_cow_blocks(ip, &tp,
first_unmap_block, XFS_MAX_FILEOFF, true);
if (error)
goto out;
xfs_itruncate_clear_reflink_flags(ip);
}
/*
* Always re-log the inode so that our permanent transaction can keep
* on rolling it forward in the log.
*/
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
trace_xfs_itruncate_extents_end(ip, new_size);
out:
*tpp = tp;
return error;
}
int
xfs_release(
xfs_inode_t *ip)
{
xfs_mount_t *mp = ip->i_mount;
int error = 0;
if (!S_ISREG(VFS_I(ip)->i_mode) || (VFS_I(ip)->i_mode == 0))
return 0;
/* If this is a read-only mount, don't do this (would generate I/O) */
if (mp->m_flags & XFS_MOUNT_RDONLY)
return 0;
if (!XFS_FORCED_SHUTDOWN(mp)) {
int truncated;
/*
* If we previously truncated this file and removed old data
* in the process, we want to initiate "early" writeout on
* the last close. This is an attempt to combat the notorious
* NULL files problem which is particularly noticeable from a
* truncate down, buffered (re-)write (delalloc), followed by
* a crash. What we are effectively doing here is
* significantly reducing the time window where we'd otherwise
* be exposed to that problem.
*/
truncated = xfs_iflags_test_and_clear(ip, XFS_ITRUNCATED);
if (truncated) {
xfs_iflags_clear(ip, XFS_IDIRTY_RELEASE);
if (ip->i_delayed_blks > 0) {
error = filemap_flush(VFS_I(ip)->i_mapping);
if (error)
return error;
}
}
}
if (VFS_I(ip)->i_nlink == 0)
return 0;
/*
* If we can't get the iolock just skip truncating the blocks past EOF
* because we could deadlock with the mmap_lock otherwise. We'll get
* another chance to drop them once the last reference to the inode is
* dropped, so we'll never leak blocks permanently.
*/
if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL))
return 0;
if (xfs_can_free_eofblocks(ip, false)) {
/*
* Check if the inode is being opened, written and closed
* frequently and we have delayed allocation blocks outstanding
* (e.g. streaming writes from the NFS server), truncating the
* blocks past EOF will cause fragmentation to occur.
*
* In this case don't do the truncation, but we have to be
* careful how we detect this case. Blocks beyond EOF show up as
* i_delayed_blks even when the inode is clean, so we need to
* truncate them away first before checking for a dirty release.
* Hence on the first dirty close we will still remove the
* speculative allocation, but after that we will leave it in
* place.
*/
if (xfs_iflags_test(ip, XFS_IDIRTY_RELEASE))
goto out_unlock;
error = xfs_free_eofblocks(ip);
if (error)
goto out_unlock;
/* delalloc blocks after truncation means it really is dirty */
if (ip->i_delayed_blks)
xfs_iflags_set(ip, XFS_IDIRTY_RELEASE);
}
out_unlock:
xfs_iunlock(ip, XFS_IOLOCK_EXCL);
return error;
}
/*
* xfs_inactive_truncate
*
* Called to perform a truncate when an inode becomes unlinked.
*/
STATIC int
xfs_inactive_truncate(
struct xfs_inode *ip)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
if (error) {
ASSERT(XFS_FORCED_SHUTDOWN(mp));
return error;
}
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, 0);
/*
* Log the inode size first to prevent stale data exposure in the event
* of a system crash before the truncate completes. See the related
* comment in xfs_vn_setattr_size() for details.
*/
ip->i_disk_size = 0;
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, 0);
if (error)
goto error_trans_cancel;
ASSERT(ip->i_df.if_nextents == 0);
error = xfs_trans_commit(tp);
if (error)
goto error_unlock;
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return 0;
error_trans_cancel:
xfs_trans_cancel(tp);
error_unlock:
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return error;
}
/*
* xfs_inactive_ifree()
*
* Perform the inode free when an inode is unlinked.
*/
STATIC int
xfs_inactive_ifree(
struct xfs_inode *ip)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_trans *tp;
int error;
/*
* We try to use a per-AG reservation for any block needed by the finobt
* tree, but as the finobt feature predates the per-AG reservation
* support a degraded file system might not have enough space for the
* reservation at mount time. In that case try to dip into the reserved
* pool and pray.
*
* Send a warning if the reservation does happen to fail, as the inode
* now remains allocated and sits on the unlinked list until the fs is
* repaired.
*/
if (unlikely(mp->m_finobt_nores)) {
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree,
XFS_IFREE_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
&tp);
} else {
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ifree, 0, 0, 0, &tp);
}
if (error) {
if (error == -ENOSPC) {
xfs_warn_ratelimited(mp,
"Failed to remove inode(s) from unlinked list. "
"Please free space, unmount and run xfs_repair.");
} else {
ASSERT(XFS_FORCED_SHUTDOWN(mp));
}
return error;
}
xfs: Don't allow logging of XFS_ISTALE inodes In tracking down a problem in this patchset, I discovered we are reclaiming dirty stale inodes. This wasn't discovered until inodes were always attached to the cluster buffer and then the rcu callback that freed inodes was assert failing because the inode still had an active pointer to the cluster buffer after it had been reclaimed. Debugging the issue indicated that this was a pre-existing issue resulting from the way the inodes are handled in xfs_inactive_ifree. When we free a cluster buffer from xfs_ifree_cluster, all the inodes in cache are marked XFS_ISTALE. Those that are clean have nothing else done to them and so eventually get cleaned up by background reclaim. i.e. it is assumed we'll never dirty/relog an inode marked XFS_ISTALE. On journal commit dirty stale inodes as are handled by both buffer and inode log items to run though xfs_istale_done() and removed from the AIL (buffer log item commit) or the log item will simply unpin it because the buffer log item will clean it. What happens to any specific inode is entirely dependent on which log item wins the commit race, but the result is the same - stale inodes are clean, not attached to the cluster buffer, and not in the AIL. Hence inode reclaim can just free these inodes without further care. However, if the stale inode is relogged, it gets dirtied again and relogged into the CIL. Most of the time this isn't an issue, because relogging simply changes the inode's location in the current checkpoint. Problems arise, however, when the CIL checkpoints between two transactions in the xfs_inactive_ifree() deferops processing. This results in the XFS_ISTALE inode being redirtied and inserted into the CIL without any of the other stale cluster buffer infrastructure being in place. Hence on journal commit, it simply gets unpinned, so it remains dirty in memory. Everything in inode writeback avoids XFS_ISTALE inodes so it can't be written back, and it is not tracked in the AIL so there's not even a trigger to attempt to clean the inode. Hence the inode just sits dirty in memory until inode reclaim comes along, sees that it is XFS_ISTALE, and goes to reclaim it. This reclaiming of a dirty inode caused use after free, list corruptions and other nasty issues later in this patchset. Hence this patch addresses a violation of the "never log XFS_ISTALE inodes" caused by the deferops processing rolling a transaction and relogging a stale inode in xfs_inactive_free. It also adds a bunch of asserts to catch this problem in debug kernels so that we don't reintroduce this problem in future. Reproducer for this issue was generic/558 on a v4 filesystem. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:45 +00:00
/*
* We do not hold the inode locked across the entire rolling transaction
* here. We only need to hold it for the first transaction that
* xfs_ifree() builds, which may mark the inode XFS_ISTALE if the
* underlying cluster buffer is freed. Relogging an XFS_ISTALE inode
* here breaks the relationship between cluster buffer invalidation and
* stale inode invalidation on cluster buffer item journal commit
* completion, and can result in leaving dirty stale inodes hanging
* around in memory.
*
* We have no need for serialising this inode operation against other
* operations - we freed the inode and hence reallocation is required
* and that will serialise on reallocating the space the deferops need
* to free. Hence we can unlock the inode on the first commit of
* the transaction rather than roll it right through the deferops. This
* avoids relogging the XFS_ISTALE inode.
*
* We check that xfs_ifree() hasn't grown an internal transaction roll
* by asserting that the inode is still locked when it returns.
*/
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs: Don't allow logging of XFS_ISTALE inodes In tracking down a problem in this patchset, I discovered we are reclaiming dirty stale inodes. This wasn't discovered until inodes were always attached to the cluster buffer and then the rcu callback that freed inodes was assert failing because the inode still had an active pointer to the cluster buffer after it had been reclaimed. Debugging the issue indicated that this was a pre-existing issue resulting from the way the inodes are handled in xfs_inactive_ifree. When we free a cluster buffer from xfs_ifree_cluster, all the inodes in cache are marked XFS_ISTALE. Those that are clean have nothing else done to them and so eventually get cleaned up by background reclaim. i.e. it is assumed we'll never dirty/relog an inode marked XFS_ISTALE. On journal commit dirty stale inodes as are handled by both buffer and inode log items to run though xfs_istale_done() and removed from the AIL (buffer log item commit) or the log item will simply unpin it because the buffer log item will clean it. What happens to any specific inode is entirely dependent on which log item wins the commit race, but the result is the same - stale inodes are clean, not attached to the cluster buffer, and not in the AIL. Hence inode reclaim can just free these inodes without further care. However, if the stale inode is relogged, it gets dirtied again and relogged into the CIL. Most of the time this isn't an issue, because relogging simply changes the inode's location in the current checkpoint. Problems arise, however, when the CIL checkpoints between two transactions in the xfs_inactive_ifree() deferops processing. This results in the XFS_ISTALE inode being redirtied and inserted into the CIL without any of the other stale cluster buffer infrastructure being in place. Hence on journal commit, it simply gets unpinned, so it remains dirty in memory. Everything in inode writeback avoids XFS_ISTALE inodes so it can't be written back, and it is not tracked in the AIL so there's not even a trigger to attempt to clean the inode. Hence the inode just sits dirty in memory until inode reclaim comes along, sees that it is XFS_ISTALE, and goes to reclaim it. This reclaiming of a dirty inode caused use after free, list corruptions and other nasty issues later in this patchset. Hence this patch addresses a violation of the "never log XFS_ISTALE inodes" caused by the deferops processing rolling a transaction and relogging a stale inode in xfs_inactive_free. It also adds a bunch of asserts to catch this problem in debug kernels so that we don't reintroduce this problem in future. Reproducer for this issue was generic/558 on a v4 filesystem. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:45 +00:00
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
error = xfs_ifree(tp, ip);
xfs: Don't allow logging of XFS_ISTALE inodes In tracking down a problem in this patchset, I discovered we are reclaiming dirty stale inodes. This wasn't discovered until inodes were always attached to the cluster buffer and then the rcu callback that freed inodes was assert failing because the inode still had an active pointer to the cluster buffer after it had been reclaimed. Debugging the issue indicated that this was a pre-existing issue resulting from the way the inodes are handled in xfs_inactive_ifree. When we free a cluster buffer from xfs_ifree_cluster, all the inodes in cache are marked XFS_ISTALE. Those that are clean have nothing else done to them and so eventually get cleaned up by background reclaim. i.e. it is assumed we'll never dirty/relog an inode marked XFS_ISTALE. On journal commit dirty stale inodes as are handled by both buffer and inode log items to run though xfs_istale_done() and removed from the AIL (buffer log item commit) or the log item will simply unpin it because the buffer log item will clean it. What happens to any specific inode is entirely dependent on which log item wins the commit race, but the result is the same - stale inodes are clean, not attached to the cluster buffer, and not in the AIL. Hence inode reclaim can just free these inodes without further care. However, if the stale inode is relogged, it gets dirtied again and relogged into the CIL. Most of the time this isn't an issue, because relogging simply changes the inode's location in the current checkpoint. Problems arise, however, when the CIL checkpoints between two transactions in the xfs_inactive_ifree() deferops processing. This results in the XFS_ISTALE inode being redirtied and inserted into the CIL without any of the other stale cluster buffer infrastructure being in place. Hence on journal commit, it simply gets unpinned, so it remains dirty in memory. Everything in inode writeback avoids XFS_ISTALE inodes so it can't be written back, and it is not tracked in the AIL so there's not even a trigger to attempt to clean the inode. Hence the inode just sits dirty in memory until inode reclaim comes along, sees that it is XFS_ISTALE, and goes to reclaim it. This reclaiming of a dirty inode caused use after free, list corruptions and other nasty issues later in this patchset. Hence this patch addresses a violation of the "never log XFS_ISTALE inodes" caused by the deferops processing rolling a transaction and relogging a stale inode in xfs_inactive_free. It also adds a bunch of asserts to catch this problem in debug kernels so that we don't reintroduce this problem in future. Reproducer for this issue was generic/558 on a v4 filesystem. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:45 +00:00
ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
if (error) {
/*
* If we fail to free the inode, shut down. The cancel
* might do that, we need to make sure. Otherwise the
* inode might be lost for a long time or forever.
*/
if (!XFS_FORCED_SHUTDOWN(mp)) {
xfs_notice(mp, "%s: xfs_ifree returned error %d",
__func__, error);
xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
}
xfs_trans_cancel(tp);
return error;
}
/*
* Credit the quota account(s). The inode is gone.
*/
xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
/*
* Just ignore errors at this point. There is nothing we can do except
* to try to keep going. Make sure it's not a silent error.
*/
error = xfs_trans_commit(tp);
if (error)
xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
__func__, error);
return 0;
}
/*
* Returns true if we need to update the on-disk metadata before we can free
* the memory used by this inode. Updates include freeing post-eof
* preallocations; freeing COW staging extents; and marking the inode free in
* the inobt if it is on the unlinked list.
*/
bool
xfs_inode_needs_inactive(
struct xfs_inode *ip)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_ifork *cow_ifp = XFS_IFORK_PTR(ip, XFS_COW_FORK);
/*
* If the inode is already free, then there can be nothing
* to clean up here.
*/
if (VFS_I(ip)->i_mode == 0)
return false;
/* If this is a read-only mount, don't do this (would generate I/O) */
if (mp->m_flags & XFS_MOUNT_RDONLY)
return false;
/* If the log isn't running, push inodes straight to reclaim. */
if (XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_NORECOVERY))
return false;
/* Metadata inodes require explicit resource cleanup. */
if (xfs_is_metadata_inode(ip))
return false;
/* Want to clean out the cow blocks if there are any. */
if (cow_ifp && cow_ifp->if_bytes > 0)
return true;
/* Unlinked files must be freed. */
if (VFS_I(ip)->i_nlink == 0)
return true;
/*
* This file isn't being freed, so check if there are post-eof blocks
* to free. @force is true because we are evicting an inode from the
* cache. Post-eof blocks must be freed, lest we end up with broken
* free space accounting.
*
* Note: don't bother with iolock here since lockdep complains about
* acquiring it in reclaim context. We have the only reference to the
* inode at this point anyways.
*/
return xfs_can_free_eofblocks(ip, true);
}
/*
* xfs_inactive
*
* This is called when the vnode reference count for the vnode
* goes to zero. If the file has been unlinked, then it must
* now be truncated. Also, we clear all of the read-ahead state
* kept for the inode here since the file is now closed.
*/
void
xfs_inactive(
xfs_inode_t *ip)
{
struct xfs_mount *mp;
int error;
int truncate = 0;
/*
* If the inode is already free, then there can be nothing
* to clean up here.
*/
if (VFS_I(ip)->i_mode == 0) {
ASSERT(ip->i_df.if_broot_bytes == 0);
goto out;
}
mp = ip->i_mount;
ASSERT(!xfs_iflags_test(ip, XFS_IRECOVERY));
/* If this is a read-only mount, don't do this (would generate I/O) */
if (mp->m_flags & XFS_MOUNT_RDONLY)
goto out;
/* Metadata inodes require explicit resource cleanup. */
if (xfs_is_metadata_inode(ip))
goto out;
/* Try to clean out the cow blocks if there are any. */
if (xfs_inode_has_cow_data(ip))
xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
if (VFS_I(ip)->i_nlink != 0) {
/*
* force is true because we are evicting an inode from the
* cache. Post-eof blocks must be freed, lest we end up with
* broken free space accounting.
*
* Note: don't bother with iolock here since lockdep complains
* about acquiring it in reclaim context. We have the only
* reference to the inode at this point anyways.
*/
if (xfs_can_free_eofblocks(ip, true))
xfs_free_eofblocks(ip);
goto out;
}
if (S_ISREG(VFS_I(ip)->i_mode) &&
(ip->i_disk_size != 0 || XFS_ISIZE(ip) != 0 ||
ip->i_df.if_nextents > 0 || ip->i_delayed_blks > 0))
truncate = 1;
error = xfs_qm_dqattach(ip);
if (error)
goto out;
if (S_ISLNK(VFS_I(ip)->i_mode))
error = xfs_inactive_symlink(ip);
else if (truncate)
error = xfs_inactive_truncate(ip);
if (error)
goto out;
/*
* If there are attributes associated with the file then blow them away
* now. The code calls a routine that recursively deconstructs the
* attribute fork. If also blows away the in-core attribute fork.
*/
if (XFS_IFORK_Q(ip)) {
error = xfs_attr_inactive(ip);
if (error)
goto out;
}
ASSERT(!ip->i_afp);
ASSERT(ip->i_forkoff == 0);
/*
* Free the inode.
*/
xfs_inactive_ifree(ip);
out:
/*
* We're done making metadata updates for this inode, so we can release
* the attached dquots.
*/
xfs_qm_dqdetach(ip);
}
/*
* In-Core Unlinked List Lookups
* =============================
*
* Every inode is supposed to be reachable from some other piece of metadata
* with the exception of the root directory. Inodes with a connection to a
* file descriptor but not linked from anywhere in the on-disk directory tree
* are collectively known as unlinked inodes, though the filesystem itself
* maintains links to these inodes so that on-disk metadata are consistent.
*
* XFS implements a per-AG on-disk hash table of unlinked inodes. The AGI
* header contains a number of buckets that point to an inode, and each inode
* record has a pointer to the next inode in the hash chain. This
* singly-linked list causes scaling problems in the iunlink remove function
* because we must walk that list to find the inode that points to the inode
* being removed from the unlinked hash bucket list.
*
* What if we modelled the unlinked list as a collection of records capturing
* "X.next_unlinked = Y" relations? If we indexed those records on Y, we'd
* have a fast way to look up unlinked list predecessors, which avoids the
* slow list walk. That's exactly what we do here (in-core) with a per-AG
* rhashtable.
*
* Because this is a backref cache, we ignore operational failures since the
* iunlink code can fall back to the slow bucket walk. The only errors that
* should bubble out are for obviously incorrect situations.
*
* All users of the backref cache MUST hold the AGI buffer lock to serialize
* access or have otherwise provided for concurrency control.
*/
/* Capture a "X.next_unlinked = Y" relationship. */
struct xfs_iunlink {
struct rhash_head iu_rhash_head;
xfs_agino_t iu_agino; /* X */
xfs_agino_t iu_next_unlinked; /* Y */
};
/* Unlinked list predecessor lookup hashtable construction */
static int
xfs_iunlink_obj_cmpfn(
struct rhashtable_compare_arg *arg,
const void *obj)
{
const xfs_agino_t *key = arg->key;
const struct xfs_iunlink *iu = obj;
if (iu->iu_next_unlinked != *key)
return 1;
return 0;
}
static const struct rhashtable_params xfs_iunlink_hash_params = {
.min_size = XFS_AGI_UNLINKED_BUCKETS,
.key_len = sizeof(xfs_agino_t),
.key_offset = offsetof(struct xfs_iunlink,
iu_next_unlinked),
.head_offset = offsetof(struct xfs_iunlink, iu_rhash_head),
.automatic_shrinking = true,
.obj_cmpfn = xfs_iunlink_obj_cmpfn,
};
/*
* Return X, where X.next_unlinked == @agino. Returns NULLAGINO if no such
* relation is found.
*/
static xfs_agino_t
xfs_iunlink_lookup_backref(
struct xfs_perag *pag,
xfs_agino_t agino)
{
struct xfs_iunlink *iu;
iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
xfs_iunlink_hash_params);
return iu ? iu->iu_agino : NULLAGINO;
}
/*
* Take ownership of an iunlink cache entry and insert it into the hash table.
* If successful, the entry will be owned by the cache; if not, it is freed.
* Either way, the caller does not own @iu after this call.
*/
static int
xfs_iunlink_insert_backref(
struct xfs_perag *pag,
struct xfs_iunlink *iu)
{
int error;
error = rhashtable_insert_fast(&pag->pagi_unlinked_hash,
&iu->iu_rhash_head, xfs_iunlink_hash_params);
/*
* Fail loudly if there already was an entry because that's a sign of
* corruption of in-memory data. Also fail loudly if we see an error
* code we didn't anticipate from the rhashtable code. Currently we
* only anticipate ENOMEM.
*/
if (error) {
WARN(error != -ENOMEM, "iunlink cache insert error %d", error);
kmem_free(iu);
}
/*
* Absorb any runtime errors that aren't a result of corruption because
* this is a cache and we can always fall back to bucket list scanning.
*/
if (error != 0 && error != -EEXIST)
error = 0;
return error;
}
/* Remember that @prev_agino.next_unlinked = @this_agino. */
static int
xfs_iunlink_add_backref(
struct xfs_perag *pag,
xfs_agino_t prev_agino,
xfs_agino_t this_agino)
{
struct xfs_iunlink *iu;
if (XFS_TEST_ERROR(false, pag->pag_mount, XFS_ERRTAG_IUNLINK_FALLBACK))
return 0;
iu = kmem_zalloc(sizeof(*iu), KM_NOFS);
iu->iu_agino = prev_agino;
iu->iu_next_unlinked = this_agino;
return xfs_iunlink_insert_backref(pag, iu);
}
/*
* Replace X.next_unlinked = @agino with X.next_unlinked = @next_unlinked.
* If @next_unlinked is NULLAGINO, we drop the backref and exit. If there
* wasn't any such entry then we don't bother.
*/
static int
xfs_iunlink_change_backref(
struct xfs_perag *pag,
xfs_agino_t agino,
xfs_agino_t next_unlinked)
{
struct xfs_iunlink *iu;
int error;
/* Look up the old entry; if there wasn't one then exit. */
iu = rhashtable_lookup_fast(&pag->pagi_unlinked_hash, &agino,
xfs_iunlink_hash_params);
if (!iu)
return 0;
/*
* Remove the entry. This shouldn't ever return an error, but if we
* couldn't remove the old entry we don't want to add it again to the
* hash table, and if the entry disappeared on us then someone's
* violated the locking rules and we need to fail loudly. Either way
* we cannot remove the inode because internal state is or would have
* been corrupt.
*/
error = rhashtable_remove_fast(&pag->pagi_unlinked_hash,
&iu->iu_rhash_head, xfs_iunlink_hash_params);
if (error)
return error;
/* If there is no new next entry just free our item and return. */
if (next_unlinked == NULLAGINO) {
kmem_free(iu);
return 0;
}
/* Update the entry and re-add it to the hash table. */
iu->iu_next_unlinked = next_unlinked;
return xfs_iunlink_insert_backref(pag, iu);
}
/* Set up the in-core predecessor structures. */
int
xfs_iunlink_init(
struct xfs_perag *pag)
{
return rhashtable_init(&pag->pagi_unlinked_hash,
&xfs_iunlink_hash_params);
}
/* Free the in-core predecessor structures. */
static void
xfs_iunlink_free_item(
void *ptr,
void *arg)
{
struct xfs_iunlink *iu = ptr;
bool *freed_anything = arg;
*freed_anything = true;
kmem_free(iu);
}
void
xfs_iunlink_destroy(
struct xfs_perag *pag)
{
bool freed_anything = false;
rhashtable_free_and_destroy(&pag->pagi_unlinked_hash,
xfs_iunlink_free_item, &freed_anything);
ASSERT(freed_anything == false || XFS_FORCED_SHUTDOWN(pag->pag_mount));
}
/*
* Point the AGI unlinked bucket at an inode and log the results. The caller
* is responsible for validating the old value.
*/
STATIC int
xfs_iunlink_update_bucket(
struct xfs_trans *tp,
struct xfs_perag *pag,
struct xfs_buf *agibp,
unsigned int bucket_index,
xfs_agino_t new_agino)
{
struct xfs_agi *agi = agibp->b_addr;
xfs_agino_t old_value;
int offset;
ASSERT(xfs_verify_agino_or_null(tp->t_mountp, pag->pag_agno, new_agino));
old_value = be32_to_cpu(agi->agi_unlinked[bucket_index]);
trace_xfs_iunlink_update_bucket(tp->t_mountp, pag->pag_agno, bucket_index,
old_value, new_agino);
/*
* We should never find the head of the list already set to the value
* passed in because either we're adding or removing ourselves from the
* head of the list.
*/
if (old_value == new_agino) {
xfs_buf_mark_corrupt(agibp);
return -EFSCORRUPTED;
}
agi->agi_unlinked[bucket_index] = cpu_to_be32(new_agino);
offset = offsetof(struct xfs_agi, agi_unlinked) +
(sizeof(xfs_agino_t) * bucket_index);
xfs_trans_log_buf(tp, agibp, offset, offset + sizeof(xfs_agino_t) - 1);
return 0;
}
/* Set an on-disk inode's next_unlinked pointer. */
STATIC void
xfs_iunlink_update_dinode(
struct xfs_trans *tp,
struct xfs_perag *pag,
xfs_agino_t agino,
struct xfs_buf *ibp,
struct xfs_dinode *dip,
struct xfs_imap *imap,
xfs_agino_t next_agino)
{
struct xfs_mount *mp = tp->t_mountp;
int offset;
ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
trace_xfs_iunlink_update_dinode(mp, pag->pag_agno, agino,
be32_to_cpu(dip->di_next_unlinked), next_agino);
dip->di_next_unlinked = cpu_to_be32(next_agino);
offset = imap->im_boffset +
offsetof(struct xfs_dinode, di_next_unlinked);
/* need to recalc the inode CRC if appropriate */
xfs_dinode_calc_crc(mp, dip);
xfs_trans_inode_buf(tp, ibp);
xfs_trans_log_buf(tp, ibp, offset, offset + sizeof(xfs_agino_t) - 1);
}
/* Set an in-core inode's unlinked pointer and return the old value. */
STATIC int
xfs_iunlink_update_inode(
struct xfs_trans *tp,
struct xfs_inode *ip,
struct xfs_perag *pag,
xfs_agino_t next_agino,
xfs_agino_t *old_next_agino)
{
struct xfs_mount *mp = tp->t_mountp;
struct xfs_dinode *dip;
struct xfs_buf *ibp;
xfs_agino_t old_value;
int error;
ASSERT(xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino));
error = xfs_imap_to_bp(mp, tp, &ip->i_imap, &ibp);
if (error)
return error;
dip = xfs_buf_offset(ibp, ip->i_imap.im_boffset);
/* Make sure the old pointer isn't garbage. */
old_value = be32_to_cpu(dip->di_next_unlinked);
if (!xfs_verify_agino_or_null(mp, pag->pag_agno, old_value)) {
xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,
sizeof(*dip), __this_address);
error = -EFSCORRUPTED;
goto out;
}
/*
* Since we're updating a linked list, we should never find that the
* current pointer is the same as the new value, unless we're
* terminating the list.
*/
*old_next_agino = old_value;
if (old_value == next_agino) {
if (next_agino != NULLAGINO) {
xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
dip, sizeof(*dip), __this_address);
error = -EFSCORRUPTED;
}
goto out;
}
/* Ok, update the new pointer. */
xfs_iunlink_update_dinode(tp, pag, XFS_INO_TO_AGINO(mp, ip->i_ino),
ibp, dip, &ip->i_imap, next_agino);
return 0;
out:
xfs_trans_brelse(tp, ibp);
return error;
}
/*
* This is called when the inode's link count has gone to 0 or we are creating
* a tmpfile via O_TMPFILE. The inode @ip must have nlink == 0.
*
* We place the on-disk inode on a list in the AGI. It will be pulled from this
* list when the inode is freed.
*/
STATIC int
xfs_iunlink(
struct xfs_trans *tp,
struct xfs_inode *ip)
{
struct xfs_mount *mp = tp->t_mountp;
struct xfs_perag *pag;
struct xfs_agi *agi;
struct xfs_buf *agibp;
xfs_agino_t next_agino;
xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
int error;
ASSERT(VFS_I(ip)->i_nlink == 0);
ASSERT(VFS_I(ip)->i_mode != 0);
trace_xfs_iunlink(ip);
pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
/* Get the agi buffer first. It ensures lock ordering on the list. */
error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
[XFS] get_bulkall() could return incorrect inode state In the following scenario xfs_bulkstat() returns incorrect stale inode state: 1. File_A is created and its inode synced to disk. 2. File_A is unlinked and doesn't exist anymore. 3. Filesystem sync is invoked. 4. File_B is created. File_B happens to reclaim File_A's inode. 5. xfs_bulkstat() is called and detects File_B but reports the incorrect File_A inode state. Explanation for the incorrect inode state is that inodes are not immediately synced on file create for performance reasons. This leaves the on-disk inode buffer uninitialized (or with old state from a previous generation inode) and this is what xfs_bulkstat() would report. The patch marks the on-disk inode buffer "dirty" on unlink. When the inode is reclaimed (by a new file create), xfs_bulkstat() would filter this inode by the "dirty" mark. Once the inode is flushed to disk, the on-disk buffer "dirty" mark is automatically removed and a following xfs_bulkstat() would return the correct inode state. Marking the on-disk inode buffer "dirty" on unlink is achieved by setting the on-disk di_nlink field to 0. Note that the in-core di_nlink has already been set to 0 and a corresponding transaction logged by xfs_droplink(). This is an exception from the rule that any on-disk inode buffer changes has to be followed by a disk write (inode flush). Synchronizing the in-core to on-disk di_nlink values in advance (before the actual inode flush to disk) should be fine in this case because the inode is already unlinked and it would never change its di_nlink again for this inode generation. SGI-PV: 970842 SGI-Modid: xfs-linux-melb:xfs-kern:29757a Signed-off-by: Vlad Apostolov <vapo@sgi.com> Signed-off-by: Alex Elder <aelder@sgi.com> Signed-off-by: David Chinner <dgc@sgi.com> Signed-off-by: Christoph Hellwig <hch@infradead.org> Signed-off-by: Mark Goodwin <markgw@sgi.com> Signed-off-by: Tim Shimmin <tes@sgi.com>
2007-10-11 07:44:18 +00:00
if (error)
goto out;
agi = agibp->b_addr;
/*
* Get the index into the agi hash table for the list this inode will
* go on. Make sure the pointer isn't garbage and that this inode
* isn't already on the list.
*/
next_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
if (next_agino == agino ||
!xfs_verify_agino_or_null(mp, pag->pag_agno, next_agino)) {
xfs_buf_mark_corrupt(agibp);
error = -EFSCORRUPTED;
goto out;
}
if (next_agino != NULLAGINO) {
xfs_agino_t old_agino;
/*
* There is already another inode in the bucket, so point this
* inode to the current head of the list.
*/
error = xfs_iunlink_update_inode(tp, ip, pag, next_agino,
&old_agino);
if (error)
goto out;
ASSERT(old_agino == NULLAGINO);
/*
* agino has been unlinked, add a backref from the next inode
* back to agino.
*/
error = xfs_iunlink_add_backref(pag, agino, next_agino);
if (error)
goto out;
}
/* Point the head of the list to point to this inode. */
error = xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index, agino);
out:
xfs_perag_put(pag);
return error;
}
/* Return the imap, dinode pointer, and buffer for an inode. */
STATIC int
xfs_iunlink_map_ino(
struct xfs_trans *tp,
xfs_agnumber_t agno,
xfs_agino_t agino,
struct xfs_imap *imap,
struct xfs_dinode **dipp,
struct xfs_buf **bpp)
{
struct xfs_mount *mp = tp->t_mountp;
int error;
imap->im_blkno = 0;
error = xfs_imap(mp, tp, XFS_AGINO_TO_INO(mp, agno, agino), imap, 0);
if (error) {
xfs_warn(mp, "%s: xfs_imap returned error %d.",
__func__, error);
return error;
}
error = xfs_imap_to_bp(mp, tp, imap, bpp);
if (error) {
xfs_warn(mp, "%s: xfs_imap_to_bp returned error %d.",
__func__, error);
return error;
}
*dipp = xfs_buf_offset(*bpp, imap->im_boffset);
return 0;
}
/*
* Walk the unlinked chain from @head_agino until we find the inode that
* points to @target_agino. Return the inode number, map, dinode pointer,
* and inode cluster buffer of that inode as @agino, @imap, @dipp, and @bpp.
*
* @tp, @pag, @head_agino, and @target_agino are input parameters.
* @agino, @imap, @dipp, and @bpp are all output parameters.
*
* Do not call this function if @target_agino is the head of the list.
*/
STATIC int
xfs_iunlink_map_prev(
struct xfs_trans *tp,
struct xfs_perag *pag,
xfs_agino_t head_agino,
xfs_agino_t target_agino,
xfs_agino_t *agino,
struct xfs_imap *imap,
struct xfs_dinode **dipp,
struct xfs_buf **bpp)
{
struct xfs_mount *mp = tp->t_mountp;
xfs_agino_t next_agino;
int error;
ASSERT(head_agino != target_agino);
*bpp = NULL;
/* See if our backref cache can find it faster. */
*agino = xfs_iunlink_lookup_backref(pag, target_agino);
if (*agino != NULLAGINO) {
error = xfs_iunlink_map_ino(tp, pag->pag_agno, *agino, imap,
dipp, bpp);
if (error)
return error;
if (be32_to_cpu((*dipp)->di_next_unlinked) == target_agino)
return 0;
/*
* If we get here the cache contents were corrupt, so drop the
* buffer and fall back to walking the bucket list.
*/
xfs_trans_brelse(tp, *bpp);
*bpp = NULL;
WARN_ON_ONCE(1);
}
trace_xfs_iunlink_map_prev_fallback(mp, pag->pag_agno);
/* Otherwise, walk the entire bucket until we find it. */
next_agino = head_agino;
while (next_agino != target_agino) {
xfs_agino_t unlinked_agino;
if (*bpp)
xfs_trans_brelse(tp, *bpp);
*agino = next_agino;
error = xfs_iunlink_map_ino(tp, pag->pag_agno, next_agino, imap,
dipp, bpp);
if (error)
return error;
unlinked_agino = be32_to_cpu((*dipp)->di_next_unlinked);
/*
* Make sure this pointer is valid and isn't an obvious
* infinite loop.
*/
if (!xfs_verify_agino(mp, pag->pag_agno, unlinked_agino) ||
next_agino == unlinked_agino) {
XFS_CORRUPTION_ERROR(__func__,
XFS_ERRLEVEL_LOW, mp,
*dipp, sizeof(**dipp));
error = -EFSCORRUPTED;
return error;
}
next_agino = unlinked_agino;
}
return 0;
}
/*
* Pull the on-disk inode from the AGI unlinked list.
*/
STATIC int
xfs_iunlink_remove(
struct xfs_trans *tp,
struct xfs_perag *pag,
struct xfs_inode *ip)
{
struct xfs_mount *mp = tp->t_mountp;
struct xfs_agi *agi;
struct xfs_buf *agibp;
struct xfs_buf *last_ibp;
struct xfs_dinode *last_dip = NULL;
xfs_agino_t agino = XFS_INO_TO_AGINO(mp, ip->i_ino);
xfs_agino_t next_agino;
xfs_agino_t head_agino;
short bucket_index = agino % XFS_AGI_UNLINKED_BUCKETS;
int error;
trace_xfs_iunlink_remove(ip);
/* Get the agi buffer first. It ensures lock ordering on the list. */
error = xfs_read_agi(mp, tp, pag->pag_agno, &agibp);
if (error)
return error;
agi = agibp->b_addr;
/*
* Get the index into the agi hash table for the list this inode will
* go on. Make sure the head pointer isn't garbage.
*/
head_agino = be32_to_cpu(agi->agi_unlinked[bucket_index]);
if (!xfs_verify_agino(mp, pag->pag_agno, head_agino)) {
XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
agi, sizeof(*agi));
return -EFSCORRUPTED;
}
/*
* Set our inode's next_unlinked pointer to NULL and then return
* the old pointer value so that we can update whatever was previous
* to us in the list to point to whatever was next in the list.
*/
error = xfs_iunlink_update_inode(tp, ip, pag, NULLAGINO, &next_agino);
if (error)
return error;
/*
* If there was a backref pointing from the next inode back to this
* one, remove it because we've removed this inode from the list.
*
* Later, if this inode was in the middle of the list we'll update
* this inode's backref to point from the next inode.
*/
if (next_agino != NULLAGINO) {
error = xfs_iunlink_change_backref(pag, next_agino, NULLAGINO);
if (error)
return error;
}
if (head_agino != agino) {
struct xfs_imap imap;
xfs_agino_t prev_agino;
/* We need to search the list for the inode being freed. */
error = xfs_iunlink_map_prev(tp, pag, head_agino, agino,
&prev_agino, &imap, &last_dip, &last_ibp);
if (error)
return error;
/* Point the previous inode on the list to the next inode. */
xfs_iunlink_update_dinode(tp, pag, prev_agino, last_ibp,
last_dip, &imap, next_agino);
/*
* Now we deal with the backref for this inode. If this inode
* pointed at a real inode, change the backref that pointed to
* us to point to our old next. If this inode was the end of
* the list, delete the backref that pointed to us. Note that
* change_backref takes care of deleting the backref if
* next_agino is NULLAGINO.
*/
return xfs_iunlink_change_backref(agibp->b_pag, agino,
next_agino);
}
/* Point the head of the list to the next unlinked inode. */
return xfs_iunlink_update_bucket(tp, pag, agibp, bucket_index,
next_agino);
}
/*
* Look up the inode number specified and if it is not already marked XFS_ISTALE
* mark it stale. We should only find clean inodes in this lookup that aren't
* already stale.
*/
static void
xfs_ifree_mark_inode_stale(
struct xfs_perag *pag,
struct xfs_inode *free_ip,
xfs_ino_t inum)
{
struct xfs_mount *mp = pag->pag_mount;
struct xfs_inode_log_item *iip;
struct xfs_inode *ip;
retry:
rcu_read_lock();
ip = radix_tree_lookup(&pag->pag_ici_root, XFS_INO_TO_AGINO(mp, inum));
/* Inode not in memory, nothing to do */
if (!ip) {
rcu_read_unlock();
return;
}
/*
* because this is an RCU protected lookup, we could find a recently
* freed or even reallocated inode during the lookup. We need to check
* under the i_flags_lock for a valid inode here. Skip it if it is not
* valid, the wrong inode or stale.
*/
spin_lock(&ip->i_flags_lock);
if (ip->i_ino != inum || __xfs_iflags_test(ip, XFS_ISTALE))
goto out_iflags_unlock;
/*
* Don't try to lock/unlock the current inode, but we _cannot_ skip the
* other inodes that we did not find in the list attached to the buffer
* and are not already marked stale. If we can't lock it, back off and
* retry.
*/
if (ip != free_ip) {
if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) {
spin_unlock(&ip->i_flags_lock);
rcu_read_unlock();
delay(1);
goto retry;
}
}
ip->i_flags |= XFS_ISTALE;
/*
* If the inode is flushing, it is already attached to the buffer. All
* we needed to do here is mark the inode stale so buffer IO completion
* will remove it from the AIL.
*/
iip = ip->i_itemp;
if (__xfs_iflags_test(ip, XFS_IFLUSHING)) {
ASSERT(!list_empty(&iip->ili_item.li_bio_list));
ASSERT(iip->ili_last_fields);
goto out_iunlock;
}
/*
* Inodes not attached to the buffer can be released immediately.
* Everything else has to go through xfs_iflush_abort() on journal
* commit as the flock synchronises removal of the inode from the
* cluster buffer against inode reclaim.
*/
if (!iip || list_empty(&iip->ili_item.li_bio_list))
goto out_iunlock;
__xfs_iflags_set(ip, XFS_IFLUSHING);
spin_unlock(&ip->i_flags_lock);
rcu_read_unlock();
/* we have a dirty inode in memory that has not yet been flushed. */
spin_lock(&iip->ili_lock);
iip->ili_last_fields = iip->ili_fields;
iip->ili_fields = 0;
iip->ili_fsync_fields = 0;
spin_unlock(&iip->ili_lock);
ASSERT(iip->ili_last_fields);
if (ip != free_ip)
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return;
out_iunlock:
if (ip != free_ip)
xfs_iunlock(ip, XFS_ILOCK_EXCL);
out_iflags_unlock:
spin_unlock(&ip->i_flags_lock);
rcu_read_unlock();
}
/*
* A big issue when freeing the inode cluster is that we _cannot_ skip any
* inodes that are in memory - they all must be marked stale and attached to
* the cluster buffer.
*/
static int
xfs_ifree_cluster(
struct xfs_trans *tp,
struct xfs_perag *pag,
struct xfs_inode *free_ip,
struct xfs_icluster *xic)
{
struct xfs_mount *mp = free_ip->i_mount;
struct xfs_ino_geometry *igeo = M_IGEO(mp);
struct xfs_buf *bp;
xfs_daddr_t blkno;
xfs_ino_t inum = xic->first_ino;
int nbufs;
xfs: fix race in inode cluster freeing failing to stale inodes When an inode cluster is freed, it needs to mark all inodes in memory as XFS_ISTALE before marking the buffer as stale. This is eeded because the inodes have a different life cycle to the buffer, and once the buffer is torn down during transaction completion, we must ensure none of the inodes get written back (which is what XFS_ISTALE does). Unfortunately, xfs_ifree_cluster() has some bugs that lead to inodes not being marked with XFS_ISTALE. This shows up when xfs_iflush() is called on these inodes either during inode reclaim or tail pushing on the AIL. The buffer is read back, but no longer contains inodes and so triggers assert failures and shutdowns. This was reproducable with at run.dbench10 invocation from xfstests. There are two main causes of xfs_ifree_cluster() failing. The first is simple - it checks in-memory inodes it finds in the per-ag icache to see if they are clean without holding the flush lock. if they are clean it skips them completely. However, If an inode is flushed delwri, it will appear clean, but is not guaranteed to be written back until the flush lock has been dropped. Hence we may have raced on the clean check and the inode may actually be dirty. Hence always mark inodes found in memory stale before we check properly if they are clean. The second is more complex, and makes the first problem easier to hit. Basically the in-memory inode scan is done with full knowledge it can be racing with inode flushing and AIl tail pushing, which means that inodes that it can't get the flush lock on might not be attached to the buffer after then in-memory inode scan due to IO completion occurring. This is actually documented in the code as "needs better interlocking". i.e. this is a zero-day bug. Effectively, the in-memory scan must be done while the inode buffer is locked and Io cannot be issued on it while we do the in-memory inode scan. This ensures that inodes we couldn't get the flush lock on are guaranteed to be attached to the cluster buffer, so we can then catch all in-memory inodes and mark them stale. Now that the inode cluster buffer is locked before the in-memory scan is done, there is no need for the two-phase update of the in-memory inodes, so simplify the code into two loops and remove the allocation of the temporary buffer used to hold locked inodes across the phases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-06-03 06:22:29 +00:00
int i, j;
int ioffset;
int error;
nbufs = igeo->ialloc_blks / igeo->blocks_per_cluster;
for (j = 0; j < nbufs; j++, inum += igeo->inodes_per_cluster) {
/*
* The allocation bitmap tells us which inodes of the chunk were
* physically allocated. Skip the cluster if an inode falls into
* a sparse region.
*/
ioffset = inum - xic->first_ino;
if ((xic->alloc & XFS_INOBT_MASK(ioffset)) == 0) {
ASSERT(ioffset % igeo->inodes_per_cluster == 0);
continue;
}
blkno = XFS_AGB_TO_DADDR(mp, XFS_INO_TO_AGNO(mp, inum),
XFS_INO_TO_AGBNO(mp, inum));
xfs: fix race in inode cluster freeing failing to stale inodes When an inode cluster is freed, it needs to mark all inodes in memory as XFS_ISTALE before marking the buffer as stale. This is eeded because the inodes have a different life cycle to the buffer, and once the buffer is torn down during transaction completion, we must ensure none of the inodes get written back (which is what XFS_ISTALE does). Unfortunately, xfs_ifree_cluster() has some bugs that lead to inodes not being marked with XFS_ISTALE. This shows up when xfs_iflush() is called on these inodes either during inode reclaim or tail pushing on the AIL. The buffer is read back, but no longer contains inodes and so triggers assert failures and shutdowns. This was reproducable with at run.dbench10 invocation from xfstests. There are two main causes of xfs_ifree_cluster() failing. The first is simple - it checks in-memory inodes it finds in the per-ag icache to see if they are clean without holding the flush lock. if they are clean it skips them completely. However, If an inode is flushed delwri, it will appear clean, but is not guaranteed to be written back until the flush lock has been dropped. Hence we may have raced on the clean check and the inode may actually be dirty. Hence always mark inodes found in memory stale before we check properly if they are clean. The second is more complex, and makes the first problem easier to hit. Basically the in-memory inode scan is done with full knowledge it can be racing with inode flushing and AIl tail pushing, which means that inodes that it can't get the flush lock on might not be attached to the buffer after then in-memory inode scan due to IO completion occurring. This is actually documented in the code as "needs better interlocking". i.e. this is a zero-day bug. Effectively, the in-memory scan must be done while the inode buffer is locked and Io cannot be issued on it while we do the in-memory inode scan. This ensures that inodes we couldn't get the flush lock on are guaranteed to be attached to the cluster buffer, so we can then catch all in-memory inodes and mark them stale. Now that the inode cluster buffer is locked before the in-memory scan is done, there is no need for the two-phase update of the in-memory inodes, so simplify the code into two loops and remove the allocation of the temporary buffer used to hold locked inodes across the phases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-06-03 06:22:29 +00:00
/*
* We obtain and lock the backing buffer first in the process
* here to ensure dirty inodes attached to the buffer remain in
* the flushing state while we mark them stale.
*
xfs: fix race in inode cluster freeing failing to stale inodes When an inode cluster is freed, it needs to mark all inodes in memory as XFS_ISTALE before marking the buffer as stale. This is eeded because the inodes have a different life cycle to the buffer, and once the buffer is torn down during transaction completion, we must ensure none of the inodes get written back (which is what XFS_ISTALE does). Unfortunately, xfs_ifree_cluster() has some bugs that lead to inodes not being marked with XFS_ISTALE. This shows up when xfs_iflush() is called on these inodes either during inode reclaim or tail pushing on the AIL. The buffer is read back, but no longer contains inodes and so triggers assert failures and shutdowns. This was reproducable with at run.dbench10 invocation from xfstests. There are two main causes of xfs_ifree_cluster() failing. The first is simple - it checks in-memory inodes it finds in the per-ag icache to see if they are clean without holding the flush lock. if they are clean it skips them completely. However, If an inode is flushed delwri, it will appear clean, but is not guaranteed to be written back until the flush lock has been dropped. Hence we may have raced on the clean check and the inode may actually be dirty. Hence always mark inodes found in memory stale before we check properly if they are clean. The second is more complex, and makes the first problem easier to hit. Basically the in-memory inode scan is done with full knowledge it can be racing with inode flushing and AIl tail pushing, which means that inodes that it can't get the flush lock on might not be attached to the buffer after then in-memory inode scan due to IO completion occurring. This is actually documented in the code as "needs better interlocking". i.e. this is a zero-day bug. Effectively, the in-memory scan must be done while the inode buffer is locked and Io cannot be issued on it while we do the in-memory inode scan. This ensures that inodes we couldn't get the flush lock on are guaranteed to be attached to the cluster buffer, so we can then catch all in-memory inodes and mark them stale. Now that the inode cluster buffer is locked before the in-memory scan is done, there is no need for the two-phase update of the in-memory inodes, so simplify the code into two loops and remove the allocation of the temporary buffer used to hold locked inodes across the phases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-06-03 06:22:29 +00:00
* If we scan the in-memory inodes first, then buffer IO can
* complete before we get a lock on it, and hence we may fail
* to mark all the active inodes on the buffer stale.
*/
error = xfs_trans_get_buf(tp, mp->m_ddev_targp, blkno,
mp->m_bsize * igeo->blocks_per_cluster,
XBF_UNMAPPED, &bp);
if (error)
return error;
/*
* This buffer may not have been correctly initialised as we
* didn't read it from disk. That's not important because we are
* only using to mark the buffer as stale in the log, and to
* attach stale cached inodes on it. That means it will never be
* dispatched for IO. If it is, we want to know about it, and we
* want it to fail. We can acheive this by adding a write
* verifier to the buffer.
*/
bp->b_ops = &xfs_inode_buf_ops;
xfs: fix race in inode cluster freeing failing to stale inodes When an inode cluster is freed, it needs to mark all inodes in memory as XFS_ISTALE before marking the buffer as stale. This is eeded because the inodes have a different life cycle to the buffer, and once the buffer is torn down during transaction completion, we must ensure none of the inodes get written back (which is what XFS_ISTALE does). Unfortunately, xfs_ifree_cluster() has some bugs that lead to inodes not being marked with XFS_ISTALE. This shows up when xfs_iflush() is called on these inodes either during inode reclaim or tail pushing on the AIL. The buffer is read back, but no longer contains inodes and so triggers assert failures and shutdowns. This was reproducable with at run.dbench10 invocation from xfstests. There are two main causes of xfs_ifree_cluster() failing. The first is simple - it checks in-memory inodes it finds in the per-ag icache to see if they are clean without holding the flush lock. if they are clean it skips them completely. However, If an inode is flushed delwri, it will appear clean, but is not guaranteed to be written back until the flush lock has been dropped. Hence we may have raced on the clean check and the inode may actually be dirty. Hence always mark inodes found in memory stale before we check properly if they are clean. The second is more complex, and makes the first problem easier to hit. Basically the in-memory inode scan is done with full knowledge it can be racing with inode flushing and AIl tail pushing, which means that inodes that it can't get the flush lock on might not be attached to the buffer after then in-memory inode scan due to IO completion occurring. This is actually documented in the code as "needs better interlocking". i.e. this is a zero-day bug. Effectively, the in-memory scan must be done while the inode buffer is locked and Io cannot be issued on it while we do the in-memory inode scan. This ensures that inodes we couldn't get the flush lock on are guaranteed to be attached to the cluster buffer, so we can then catch all in-memory inodes and mark them stale. Now that the inode cluster buffer is locked before the in-memory scan is done, there is no need for the two-phase update of the in-memory inodes, so simplify the code into two loops and remove the allocation of the temporary buffer used to hold locked inodes across the phases. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
2010-06-03 06:22:29 +00:00
/*
* Now we need to set all the cached clean inodes as XFS_ISTALE,
* too. This requires lookups, and will skip inodes that we've
* already marked XFS_ISTALE.
*/
for (i = 0; i < igeo->inodes_per_cluster; i++)
xfs_ifree_mark_inode_stale(pag, free_ip, inum + i);
xfs_trans_stale_inode_buf(tp, bp);
xfs_trans_binval(tp, bp);
}
return 0;
}
/*
* This is called to return an inode to the inode free list.
* The inode should already be truncated to 0 length and have
* no pages associated with it. This routine also assumes that
* the inode is already a part of the transaction.
*
* The on-disk copy of the inode will have been added to the list
* of unlinked inodes in the AGI. We need to remove the inode from
* that list atomically with respect to freeing it here.
*/
int
xfs_ifree(
struct xfs_trans *tp,
struct xfs_inode *ip)
{
struct xfs_mount *mp = ip->i_mount;
struct xfs_perag *pag;
struct xfs_icluster xic = { 0 };
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
struct xfs_inode_log_item *iip = ip->i_itemp;
int error;
ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
ASSERT(VFS_I(ip)->i_nlink == 0);
ASSERT(ip->i_df.if_nextents == 0);
ASSERT(ip->i_disk_size == 0 || !S_ISREG(VFS_I(ip)->i_mode));
ASSERT(ip->i_nblocks == 0);
pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
/*
* Pull the on-disk inode from the AGI unlinked list.
*/
error = xfs_iunlink_remove(tp, pag, ip);
if (error)
goto out;
error = xfs_difree(tp, pag, ip->i_ino, &xic);
if (error)
goto out;
/*
* Free any local-format data sitting around before we reset the
* data fork to extents format. Note that the attr fork data has
* already been freed by xfs_attr_inactive.
*/
if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
kmem_free(ip->i_df.if_u1.if_data);
ip->i_df.if_u1.if_data = NULL;
ip->i_df.if_bytes = 0;
}
VFS_I(ip)->i_mode = 0; /* mark incore inode as free */
ip->i_diflags = 0;
ip->i_diflags2 = mp->m_ino_geo.new_diflags2;
ip->i_forkoff = 0; /* mark the attr fork not in use */
ip->i_df.if_format = XFS_DINODE_FMT_EXTENTS;
if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS))
xfs_iflags_clear(ip, XFS_IPRESERVE_DM_FIELDS);
/* Don't attempt to replay owner changes for a deleted inode */
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
spin_lock(&iip->ili_lock);
iip->ili_fields &= ~(XFS_ILOG_AOWNER | XFS_ILOG_DOWNER);
spin_unlock(&iip->ili_lock);
/*
* Bump the generation count so no one will be confused
* by reincarnations of this inode.
*/
VFS_I(ip)->i_generation++;
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
if (xic.deleted)
error = xfs_ifree_cluster(tp, pag, ip, &xic);
out:
xfs_perag_put(pag);
return error;
}
/*
* This is called to unpin an inode. The caller must have the inode locked
* in at least shared mode so that the buffer cannot be subsequently pinned
* once someone is waiting for it to be unpinned.
*/
static void
xfs_iunpin(
struct xfs_inode *ip)
{
ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
trace_xfs_inode_unpin_nowait(ip, _RET_IP_);
/* Give the log a push to start the unpinning I/O */
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xfs_log_force_seq(ip->i_mount, ip->i_itemp->ili_commit_seq, 0, NULL);
}
static void
__xfs_iunpin_wait(
struct xfs_inode *ip)
{
wait_queue_head_t *wq = bit_waitqueue(&ip->i_flags, __XFS_IPINNED_BIT);
DEFINE_WAIT_BIT(wait, &ip->i_flags, __XFS_IPINNED_BIT);
xfs_iunpin(ip);
do {
prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
if (xfs_ipincount(ip))
io_schedule();
} while (xfs_ipincount(ip));
finish_wait(wq, &wait.wq_entry);
}
void
xfs_iunpin_wait(
struct xfs_inode *ip)
{
if (xfs_ipincount(ip))
__xfs_iunpin_wait(ip);
}
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
/*
* Removing an inode from the namespace involves removing the directory entry
* and dropping the link count on the inode. Removing the directory entry can
* result in locking an AGF (directory blocks were freed) and removing a link
* count can result in placing the inode on an unlinked list which results in
* locking an AGI.
*
* The big problem here is that we have an ordering constraint on AGF and AGI
* locking - inode allocation locks the AGI, then can allocate a new extent for
* new inodes, locking the AGF after the AGI. Similarly, freeing the inode
* removes the inode from the unlinked list, requiring that we lock the AGI
* first, and then freeing the inode can result in an inode chunk being freed
* and hence freeing disk space requiring that we lock an AGF.
*
* Hence the ordering that is imposed by other parts of the code is AGI before
* AGF. This means we cannot remove the directory entry before we drop the inode
* reference count and put it on the unlinked list as this results in a lock
* order of AGF then AGI, and this can deadlock against inode allocation and
* freeing. Therefore we must drop the link counts before we remove the
* directory entry.
*
* This is still safe from a transactional point of view - it is not until we
* get to xfs_defer_finish() that we have the possibility of multiple
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
* transactions in this operation. Hence as long as we remove the directory
* entry and drop the link count in the first transaction of the remove
* operation, there are no transactional constraints on the ordering here.
*/
int
xfs_remove(
xfs_inode_t *dp,
struct xfs_name *name,
xfs_inode_t *ip)
{
xfs_mount_t *mp = dp->i_mount;
xfs_trans_t *tp = NULL;
int is_dir = S_ISDIR(VFS_I(ip)->i_mode);
int error = 0;
uint resblks;
trace_xfs_remove(dp, name);
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
error = xfs_qm_dqattach(dp);
if (error)
goto std_return;
error = xfs_qm_dqattach(ip);
if (error)
goto std_return;
/*
* We try to get the real space reservation first,
* allowing for directory btree deletion(s) implying
* possible bmap insert(s). If we can't get the space
* reservation then we use 0 instead, and avoid the bmap
* btree insert(s) in the directory code by, if the bmap
* insert tries to happen, instead trimming the LAST
* block from the directory.
*/
resblks = XFS_REMOVE_SPACE_RES(mp);
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, resblks, 0, 0, &tp);
if (error == -ENOSPC) {
resblks = 0;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_remove, 0, 0, 0,
&tp);
}
if (error) {
ASSERT(error != -ENOSPC);
goto std_return;
}
xfs_lock_two_inodes(dp, XFS_ILOCK_EXCL, ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
/*
* If we're removing a directory perform some additional validation.
*/
if (is_dir) {
ASSERT(VFS_I(ip)->i_nlink >= 2);
if (VFS_I(ip)->i_nlink != 2) {
error = -ENOTEMPTY;
goto out_trans_cancel;
}
if (!xfs_dir_isempty(ip)) {
error = -ENOTEMPTY;
goto out_trans_cancel;
}
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
/* Drop the link from ip's "..". */
error = xfs_droplink(tp, dp);
if (error)
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
goto out_trans_cancel;
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
/* Drop the "." link from ip to self. */
error = xfs_droplink(tp, ip);
if (error)
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
goto out_trans_cancel;
/*
* Point the unlinked child directory's ".." entry to the root
* directory to eliminate back-references to inodes that may
* get freed before the child directory is closed. If the fs
* gets shrunk, this can lead to dirent inode validation errors.
*/
if (dp->i_ino != tp->t_mountp->m_sb.sb_rootino) {
error = xfs_dir_replace(tp, ip, &xfs_name_dotdot,
tp->t_mountp->m_sb.sb_rootino, 0);
if (error)
return error;
}
} else {
/*
* When removing a non-directory we need to log the parent
* inode here. For a directory this is done implicitly
* by the xfs_droplink call for the ".." entry.
*/
xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
}
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
/* Drop the link from dp to ip. */
error = xfs_droplink(tp, ip);
if (error)
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
goto out_trans_cancel;
error = xfs_dir_removename(tp, dp, name, ip->i_ino, resblks);
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
if (error) {
ASSERT(error != -ENOENT);
goto out_trans_cancel;
xfs: xfs_remove deadlocks due to inverted AGF vs AGI lock ordering Removing an inode from the namespace involves removing the directory entry and dropping the link count on the inode. Removing the directory entry can result in locking an AGF (directory blocks were freed) and removing a link count can result in placing the inode on an unlinked list which results in locking an AGI. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Similarly, freeing the inode removes the inode from the unlinked list, requiring that we lock the AGI first, and then freeing the inode can result in an inode chunk being freed and hence freeing disk space requiring that we lock an AGF. Hence the ordering that is imposed by other parts of the code is AGI before AGF. This means we cannot remove the directory entry before we drop the inode reference count and put it on the unlinked list as this results in a lock order of AGF then AGI, and this can deadlock against inode allocation and freeing. Therefore we must drop the link counts before we remove the directory entry. This is still safe from a transactional point of view - it is not until we get to xfs_bmap_finish() that we have the possibility of multiple transactions in this operation. Hence as long as we remove the directory entry and drop the link count in the first transaction of the remove operation, there are no transactional constraints on the ordering here. Change the ordering of the operations in the xfs_remove() function to align the ordering of AGI and AGF locking to match that of the rest of the code. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Ben Myers <bpm@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-10-29 11:11:44 +00:00
}
/*
* If this is a synchronous mount, make sure that the
* remove transaction goes to disk before returning to
* the user.
*/
if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
xfs_trans_set_sync(tp);
error = xfs_trans_commit(tp);
if (error)
goto std_return;
if (is_dir && xfs_inode_is_filestream(ip))
xfs_filestream_deassociate(ip);
return 0;
out_trans_cancel:
xfs_trans_cancel(tp);
std_return:
return error;
}
/*
* Enter all inodes for a rename transaction into a sorted array.
*/
#define __XFS_SORT_INODES 5
STATIC void
xfs_sort_for_rename(
struct xfs_inode *dp1, /* in: old (source) directory inode */
struct xfs_inode *dp2, /* in: new (target) directory inode */
struct xfs_inode *ip1, /* in: inode of old entry */
struct xfs_inode *ip2, /* in: inode of new entry */
struct xfs_inode *wip, /* in: whiteout inode */
struct xfs_inode **i_tab,/* out: sorted array of inodes */
int *num_inodes) /* in/out: inodes in array */
{
int i, j;
ASSERT(*num_inodes == __XFS_SORT_INODES);
memset(i_tab, 0, *num_inodes * sizeof(struct xfs_inode *));
/*
* i_tab contains a list of pointers to inodes. We initialize
* the table here & we'll sort it. We will then use it to
* order the acquisition of the inode locks.
*
* Note that the table may contain duplicates. e.g., dp1 == dp2.
*/
i = 0;
i_tab[i++] = dp1;
i_tab[i++] = dp2;
i_tab[i++] = ip1;
if (ip2)
i_tab[i++] = ip2;
if (wip)
i_tab[i++] = wip;
*num_inodes = i;
/*
* Sort the elements via bubble sort. (Remember, there are at
* most 5 elements to sort, so this is adequate.)
*/
for (i = 0; i < *num_inodes; i++) {
for (j = 1; j < *num_inodes; j++) {
if (i_tab[j]->i_ino < i_tab[j-1]->i_ino) {
struct xfs_inode *temp = i_tab[j];
i_tab[j] = i_tab[j-1];
i_tab[j-1] = temp;
}
}
}
}
static int
xfs_finish_rename(
struct xfs_trans *tp)
{
/*
* If this is a synchronous mount, make sure that the rename transaction
* goes to disk before returning to the user.
*/
if (tp->t_mountp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC))
xfs_trans_set_sync(tp);
return xfs_trans_commit(tp);
}
/*
* xfs_cross_rename()
*
* responsible for handling RENAME_EXCHANGE flag in renameat2() syscall
*/
STATIC int
xfs_cross_rename(
struct xfs_trans *tp,
struct xfs_inode *dp1,
struct xfs_name *name1,
struct xfs_inode *ip1,
struct xfs_inode *dp2,
struct xfs_name *name2,
struct xfs_inode *ip2,
int spaceres)
{
int error = 0;
int ip1_flags = 0;
int ip2_flags = 0;
int dp2_flags = 0;
/* Swap inode number for dirent in first parent */
error = xfs_dir_replace(tp, dp1, name1, ip2->i_ino, spaceres);
if (error)
goto out_trans_abort;
/* Swap inode number for dirent in second parent */
error = xfs_dir_replace(tp, dp2, name2, ip1->i_ino, spaceres);
if (error)
goto out_trans_abort;
/*
* If we're renaming one or more directories across different parents,
* update the respective ".." entries (and link counts) to match the new
* parents.
*/
if (dp1 != dp2) {
dp2_flags = XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
if (S_ISDIR(VFS_I(ip2)->i_mode)) {
error = xfs_dir_replace(tp, ip2, &xfs_name_dotdot,
dp1->i_ino, spaceres);
if (error)
goto out_trans_abort;
/* transfer ip2 ".." reference to dp1 */
if (!S_ISDIR(VFS_I(ip1)->i_mode)) {
error = xfs_droplink(tp, dp2);
if (error)
goto out_trans_abort;
xfs_bumplink(tp, dp1);
}
/*
* Although ip1 isn't changed here, userspace needs
* to be warned about the change, so that applications
* relying on it (like backup ones), will properly
* notify the change
*/
ip1_flags |= XFS_ICHGTIME_CHG;
ip2_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
}
if (S_ISDIR(VFS_I(ip1)->i_mode)) {
error = xfs_dir_replace(tp, ip1, &xfs_name_dotdot,
dp2->i_ino, spaceres);
if (error)
goto out_trans_abort;
/* transfer ip1 ".." reference to dp2 */
if (!S_ISDIR(VFS_I(ip2)->i_mode)) {
error = xfs_droplink(tp, dp1);
if (error)
goto out_trans_abort;
xfs_bumplink(tp, dp2);
}
/*
* Although ip2 isn't changed here, userspace needs
* to be warned about the change, so that applications
* relying on it (like backup ones), will properly
* notify the change
*/
ip1_flags |= XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG;
ip2_flags |= XFS_ICHGTIME_CHG;
}
}
if (ip1_flags) {
xfs_trans_ichgtime(tp, ip1, ip1_flags);
xfs_trans_log_inode(tp, ip1, XFS_ILOG_CORE);
}
if (ip2_flags) {
xfs_trans_ichgtime(tp, ip2, ip2_flags);
xfs_trans_log_inode(tp, ip2, XFS_ILOG_CORE);
}
if (dp2_flags) {
xfs_trans_ichgtime(tp, dp2, dp2_flags);
xfs_trans_log_inode(tp, dp2, XFS_ILOG_CORE);
}
xfs_trans_ichgtime(tp, dp1, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs_trans_log_inode(tp, dp1, XFS_ILOG_CORE);
return xfs_finish_rename(tp);
out_trans_abort:
xfs_trans_cancel(tp);
return error;
}
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
/*
* xfs_rename_alloc_whiteout()
*
* Return a referenced, unlinked, unlocked inode that can be used as a
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
* whiteout in a rename transaction. We use a tmpfile inode here so that if we
* crash between allocating the inode and linking it into the rename transaction
* recovery will free the inode and we won't leak it.
*/
static int
xfs_rename_alloc_whiteout(
struct user_namespace *mnt_userns,
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
struct xfs_inode *dp,
struct xfs_inode **wip)
{
struct xfs_inode *tmpfile;
int error;
error = xfs_create_tmpfile(mnt_userns, dp, S_IFCHR | WHITEOUT_MODE,
&tmpfile);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
if (error)
return error;
/*
* Prepare the tmpfile inode as if it were created through the VFS.
* Complete the inode setup and flag it as linkable. nlink is already
* zero, so we can skip the drop_nlink.
*/
xfs_setup_iops(tmpfile);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
xfs_finish_inode_setup(tmpfile);
VFS_I(tmpfile)->i_state |= I_LINKABLE;
*wip = tmpfile;
return 0;
}
/*
* xfs_rename
*/
int
xfs_rename(
struct user_namespace *mnt_userns,
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
struct xfs_inode *src_dp,
struct xfs_name *src_name,
struct xfs_inode *src_ip,
struct xfs_inode *target_dp,
struct xfs_name *target_name,
struct xfs_inode *target_ip,
unsigned int flags)
{
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
struct xfs_mount *mp = src_dp->i_mount;
struct xfs_trans *tp;
struct xfs_inode *wip = NULL; /* whiteout inode */
struct xfs_inode *inodes[__XFS_SORT_INODES];
int i;
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
int num_inodes = __XFS_SORT_INODES;
bool new_parent = (src_dp != target_dp);
bool src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
int spaceres;
int error;
trace_xfs_rename(src_dp, target_dp, src_name, target_name);
if ((flags & RENAME_EXCHANGE) && !target_ip)
return -EINVAL;
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
/*
* If we are doing a whiteout operation, allocate the whiteout inode
* we will be placing at the target and ensure the type is set
* appropriately.
*/
if (flags & RENAME_WHITEOUT) {
ASSERT(!(flags & (RENAME_NOREPLACE | RENAME_EXCHANGE)));
error = xfs_rename_alloc_whiteout(mnt_userns, target_dp, &wip);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
if (error)
return error;
/* setup target dirent info as whiteout */
src_name->type = XFS_DIR3_FT_CHRDEV;
}
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
xfs_sort_for_rename(src_dp, target_dp, src_ip, target_ip, wip,
inodes, &num_inodes);
spaceres = XFS_RENAME_SPACE_RES(mp, target_name->len);
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, spaceres, 0, 0, &tp);
if (error == -ENOSPC) {
spaceres = 0;
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_rename, 0, 0, 0,
&tp);
}
if (error)
goto out_release_wip;
/*
* Attach the dquots to the inodes
*/
error = xfs_qm_vop_rename_dqattach(inodes);
if (error)
goto out_trans_cancel;
/*
* Lock all the participating inodes. Depending upon whether
* the target_name exists in the target directory, and
* whether the target directory is the same as the source
* directory, we can lock from 2 to 4 inodes.
*/
xfs_lock_inodes(inodes, num_inodes, XFS_ILOCK_EXCL);
/*
* Join all the inodes to the transaction. From this point on,
* we can rely on either trans_commit or trans_cancel to unlock
* them.
*/
xfs_trans_ijoin(tp, src_dp, XFS_ILOCK_EXCL);
if (new_parent)
xfs_trans_ijoin(tp, target_dp, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, src_ip, XFS_ILOCK_EXCL);
if (target_ip)
xfs_trans_ijoin(tp, target_ip, XFS_ILOCK_EXCL);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
if (wip)
xfs_trans_ijoin(tp, wip, XFS_ILOCK_EXCL);
/*
* If we are using project inheritance, we only allow renames
* into our tree when the project IDs are the same; else the
* tree quota mechanism would be circumvented.
*/
if (unlikely((target_dp->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
target_dp->i_projid != src_ip->i_projid)) {
error = -EXDEV;
goto out_trans_cancel;
}
/* RENAME_EXCHANGE is unique from here on. */
if (flags & RENAME_EXCHANGE)
return xfs_cross_rename(tp, src_dp, src_name, src_ip,
target_dp, target_name, target_ip,
spaceres);
/*
xfs: Fix deadlock between AGI and AGF with RENAME_WHITEOUT When performing rename operation with RENAME_WHITEOUT flag, we will hold AGF lock to allocate or free extents in manipulating the dirents firstly, and then doing the xfs_iunlink_remove() call last to hold AGI lock to modify the tmpfile info, so we the lock order AGI->AGF. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Hence the ordering that is imposed by other parts of the code is AGI before AGF. So we get an ABBA deadlock between the AGI and AGF here. Process A: Call trace: ? __schedule+0x2bd/0x620 schedule+0x33/0x90 schedule_timeout+0x17d/0x290 __down_common+0xef/0x125 ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agf+0xa6/0x180 [xfs] ? schedule_timeout+0x17d/0x290 xfs_alloc_read_agf+0x52/0x1f0 [xfs] xfs_alloc_fix_freelist+0x432/0x590 [xfs] ? down+0x3b/0x50 ? xfs_buf_lock+0x34/0xf0 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] xfs_alloc_vextent+0x301/0x6c0 [xfs] xfs_ialloc_ag_alloc+0x182/0x700 [xfs] ? _xfs_trans_bjoin+0x72/0xf0 [xfs] xfs_dialloc+0x116/0x290 [xfs] xfs_ialloc+0x6d/0x5e0 [xfs] ? xfs_log_reserve+0x165/0x280 [xfs] xfs_dir_ialloc+0x8c/0x240 [xfs] xfs_create+0x35a/0x610 [xfs] xfs_generic_create+0x1f1/0x2f0 [xfs] ... Process B: Call trace: ? __schedule+0x2bd/0x620 ? xfs_bmapi_allocate+0x245/0x380 [xfs] schedule+0x33/0x90 schedule_timeout+0x17d/0x290 ? xfs_buf_find+0x1fd/0x6c0 [xfs] __down_common+0xef/0x125 ? xfs_buf_get_map+0x37/0x230 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agi+0xa8/0x160 [xfs] xfs_iunlink_remove+0x6f/0x2a0 [xfs] ? current_time+0x46/0x80 ? xfs_trans_ichgtime+0x39/0xb0 [xfs] xfs_rename+0x57a/0xae0 [xfs] xfs_vn_rename+0xe4/0x150 [xfs] ... In this patch we move the xfs_iunlink_remove() call to before acquiring the AGF lock to preserve correct AGI/AGF locking order. Signed-off-by: kaixuxia <kaixuxia@tencent.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-04 04:06:50 +00:00
* Check for expected errors before we dirty the transaction
* so we can return an error without a transaction abort.
xfs: Check for extent overflow when renaming dir entries A rename operation is essentially a directory entry remove operation from the perspective of parent directory (i.e. src_dp) of rename's source. Hence the only place where we check for extent count overflow for src_dp is in xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns -ENOSPC when it detects a possible extent count overflow and in response, the higher layers of directory handling code do the following: 1. Data/Free blocks: XFS lets these blocks linger until a future remove operation removes them. 2. Dabtree blocks: XFS swaps the blocks with the last block in the Leaf space and unmaps the last block. For target_dp, there are two cases depending on whether the destination directory entry exists or not. When destination directory entry does not exist (i.e. target_ip == NULL), extent count overflow check is performed only when transaction has a non-zero sized space reservation associated with it. With a zero-sized space reservation, XFS allows a rename operation to continue only when the directory has sufficient free space in its data/leaf/free space blocks to hold the new entry. When destination directory entry exists (i.e. target_ip != NULL), all we need to do is change the inode number associated with the already existing entry. Hence there is no need to perform an extent count overflow check. Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2021-01-23 00:48:13 +00:00
*
* Extent count overflow check:
*
* From the perspective of src_dp, a rename operation is essentially a
* directory entry remove operation. Hence the only place where we check
* for extent count overflow for src_dp is in
* xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns
* -ENOSPC when it detects a possible extent count overflow and in
* response, the higher layers of directory handling code do the
* following:
* 1. Data/Free blocks: XFS lets these blocks linger until a
* future remove operation removes them.
* 2. Dabtree blocks: XFS swaps the blocks with the last block in the
* Leaf space and unmaps the last block.
*
* For target_dp, there are two cases depending on whether the
* destination directory entry exists or not.
*
* When destination directory entry does not exist (i.e. target_ip ==
* NULL), extent count overflow check is performed only when transaction
* has a non-zero sized space reservation associated with it. With a
* zero-sized space reservation, XFS allows a rename operation to
* continue only when the directory has sufficient free space in its
* data/leaf/free space blocks to hold the new entry.
*
* When destination directory entry exists (i.e. target_ip != NULL), all
* we need to do is change the inode number associated with the already
* existing entry. Hence there is no need to perform an extent count
* overflow check.
*/
if (target_ip == NULL) {
/*
* If there's no space reservation, check the entry will
* fit before actually inserting it.
*/
if (!spaceres) {
error = xfs_dir_canenter(tp, target_dp, target_name);
if (error)
goto out_trans_cancel;
xfs: Check for extent overflow when renaming dir entries A rename operation is essentially a directory entry remove operation from the perspective of parent directory (i.e. src_dp) of rename's source. Hence the only place where we check for extent count overflow for src_dp is in xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns -ENOSPC when it detects a possible extent count overflow and in response, the higher layers of directory handling code do the following: 1. Data/Free blocks: XFS lets these blocks linger until a future remove operation removes them. 2. Dabtree blocks: XFS swaps the blocks with the last block in the Leaf space and unmaps the last block. For target_dp, there are two cases depending on whether the destination directory entry exists or not. When destination directory entry does not exist (i.e. target_ip == NULL), extent count overflow check is performed only when transaction has a non-zero sized space reservation associated with it. With a zero-sized space reservation, XFS allows a rename operation to continue only when the directory has sufficient free space in its data/leaf/free space blocks to hold the new entry. When destination directory entry exists (i.e. target_ip != NULL), all we need to do is change the inode number associated with the already existing entry. Hence there is no need to perform an extent count overflow check. Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2021-01-23 00:48:13 +00:00
} else {
error = xfs_iext_count_may_overflow(target_dp,
XFS_DATA_FORK,
XFS_IEXT_DIR_MANIP_CNT(mp));
if (error)
goto out_trans_cancel;
}
xfs: Fix deadlock between AGI and AGF with RENAME_WHITEOUT When performing rename operation with RENAME_WHITEOUT flag, we will hold AGF lock to allocate or free extents in manipulating the dirents firstly, and then doing the xfs_iunlink_remove() call last to hold AGI lock to modify the tmpfile info, so we the lock order AGI->AGF. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Hence the ordering that is imposed by other parts of the code is AGI before AGF. So we get an ABBA deadlock between the AGI and AGF here. Process A: Call trace: ? __schedule+0x2bd/0x620 schedule+0x33/0x90 schedule_timeout+0x17d/0x290 __down_common+0xef/0x125 ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agf+0xa6/0x180 [xfs] ? schedule_timeout+0x17d/0x290 xfs_alloc_read_agf+0x52/0x1f0 [xfs] xfs_alloc_fix_freelist+0x432/0x590 [xfs] ? down+0x3b/0x50 ? xfs_buf_lock+0x34/0xf0 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] xfs_alloc_vextent+0x301/0x6c0 [xfs] xfs_ialloc_ag_alloc+0x182/0x700 [xfs] ? _xfs_trans_bjoin+0x72/0xf0 [xfs] xfs_dialloc+0x116/0x290 [xfs] xfs_ialloc+0x6d/0x5e0 [xfs] ? xfs_log_reserve+0x165/0x280 [xfs] xfs_dir_ialloc+0x8c/0x240 [xfs] xfs_create+0x35a/0x610 [xfs] xfs_generic_create+0x1f1/0x2f0 [xfs] ... Process B: Call trace: ? __schedule+0x2bd/0x620 ? xfs_bmapi_allocate+0x245/0x380 [xfs] schedule+0x33/0x90 schedule_timeout+0x17d/0x290 ? xfs_buf_find+0x1fd/0x6c0 [xfs] __down_common+0xef/0x125 ? xfs_buf_get_map+0x37/0x230 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agi+0xa8/0x160 [xfs] xfs_iunlink_remove+0x6f/0x2a0 [xfs] ? current_time+0x46/0x80 ? xfs_trans_ichgtime+0x39/0xb0 [xfs] xfs_rename+0x57a/0xae0 [xfs] xfs_vn_rename+0xe4/0x150 [xfs] ... In this patch we move the xfs_iunlink_remove() call to before acquiring the AGF lock to preserve correct AGI/AGF locking order. Signed-off-by: kaixuxia <kaixuxia@tencent.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-04 04:06:50 +00:00
} else {
/*
* If target exists and it's a directory, check that whether
* it can be destroyed.
*/
if (S_ISDIR(VFS_I(target_ip)->i_mode) &&
(!xfs_dir_isempty(target_ip) ||
(VFS_I(target_ip)->i_nlink > 2))) {
error = -EEXIST;
goto out_trans_cancel;
}
}
/*
* Lock the AGI buffers we need to handle bumping the nlink of the
* whiteout inode off the unlinked list and to handle dropping the
* nlink of the target inode. Per locking order rules, do this in
* increasing AG order and before directory block allocation tries to
* grab AGFs because we grab AGIs before AGFs.
*
* The (vfs) caller must ensure that if src is a directory then
* target_ip is either null or an empty directory.
*/
for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
if (inodes[i] == wip ||
(inodes[i] == target_ip &&
(VFS_I(target_ip)->i_nlink == 1 || src_is_directory))) {
struct xfs_buf *bp;
xfs_agnumber_t agno;
agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
error = xfs_read_agi(mp, tp, agno, &bp);
if (error)
goto out_trans_cancel;
}
}
xfs: Fix deadlock between AGI and AGF with RENAME_WHITEOUT When performing rename operation with RENAME_WHITEOUT flag, we will hold AGF lock to allocate or free extents in manipulating the dirents firstly, and then doing the xfs_iunlink_remove() call last to hold AGI lock to modify the tmpfile info, so we the lock order AGI->AGF. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Hence the ordering that is imposed by other parts of the code is AGI before AGF. So we get an ABBA deadlock between the AGI and AGF here. Process A: Call trace: ? __schedule+0x2bd/0x620 schedule+0x33/0x90 schedule_timeout+0x17d/0x290 __down_common+0xef/0x125 ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agf+0xa6/0x180 [xfs] ? schedule_timeout+0x17d/0x290 xfs_alloc_read_agf+0x52/0x1f0 [xfs] xfs_alloc_fix_freelist+0x432/0x590 [xfs] ? down+0x3b/0x50 ? xfs_buf_lock+0x34/0xf0 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] xfs_alloc_vextent+0x301/0x6c0 [xfs] xfs_ialloc_ag_alloc+0x182/0x700 [xfs] ? _xfs_trans_bjoin+0x72/0xf0 [xfs] xfs_dialloc+0x116/0x290 [xfs] xfs_ialloc+0x6d/0x5e0 [xfs] ? xfs_log_reserve+0x165/0x280 [xfs] xfs_dir_ialloc+0x8c/0x240 [xfs] xfs_create+0x35a/0x610 [xfs] xfs_generic_create+0x1f1/0x2f0 [xfs] ... Process B: Call trace: ? __schedule+0x2bd/0x620 ? xfs_bmapi_allocate+0x245/0x380 [xfs] schedule+0x33/0x90 schedule_timeout+0x17d/0x290 ? xfs_buf_find+0x1fd/0x6c0 [xfs] __down_common+0xef/0x125 ? xfs_buf_get_map+0x37/0x230 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agi+0xa8/0x160 [xfs] xfs_iunlink_remove+0x6f/0x2a0 [xfs] ? current_time+0x46/0x80 ? xfs_trans_ichgtime+0x39/0xb0 [xfs] xfs_rename+0x57a/0xae0 [xfs] xfs_vn_rename+0xe4/0x150 [xfs] ... In this patch we move the xfs_iunlink_remove() call to before acquiring the AGF lock to preserve correct AGI/AGF locking order. Signed-off-by: kaixuxia <kaixuxia@tencent.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-04 04:06:50 +00:00
/*
* Directory entry creation below may acquire the AGF. Remove
* the whiteout from the unlinked list first to preserve correct
* AGI/AGF locking order. This dirties the transaction so failures
* after this point will abort and log recovery will clean up the
* mess.
*
* For whiteouts, we need to bump the link count on the whiteout
* inode. After this point, we have a real link, clear the tmpfile
* state flag from the inode so it doesn't accidentally get misused
* in future.
*/
if (wip) {
struct xfs_perag *pag;
xfs: Fix deadlock between AGI and AGF with RENAME_WHITEOUT When performing rename operation with RENAME_WHITEOUT flag, we will hold AGF lock to allocate or free extents in manipulating the dirents firstly, and then doing the xfs_iunlink_remove() call last to hold AGI lock to modify the tmpfile info, so we the lock order AGI->AGF. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Hence the ordering that is imposed by other parts of the code is AGI before AGF. So we get an ABBA deadlock between the AGI and AGF here. Process A: Call trace: ? __schedule+0x2bd/0x620 schedule+0x33/0x90 schedule_timeout+0x17d/0x290 __down_common+0xef/0x125 ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agf+0xa6/0x180 [xfs] ? schedule_timeout+0x17d/0x290 xfs_alloc_read_agf+0x52/0x1f0 [xfs] xfs_alloc_fix_freelist+0x432/0x590 [xfs] ? down+0x3b/0x50 ? xfs_buf_lock+0x34/0xf0 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] xfs_alloc_vextent+0x301/0x6c0 [xfs] xfs_ialloc_ag_alloc+0x182/0x700 [xfs] ? _xfs_trans_bjoin+0x72/0xf0 [xfs] xfs_dialloc+0x116/0x290 [xfs] xfs_ialloc+0x6d/0x5e0 [xfs] ? xfs_log_reserve+0x165/0x280 [xfs] xfs_dir_ialloc+0x8c/0x240 [xfs] xfs_create+0x35a/0x610 [xfs] xfs_generic_create+0x1f1/0x2f0 [xfs] ... Process B: Call trace: ? __schedule+0x2bd/0x620 ? xfs_bmapi_allocate+0x245/0x380 [xfs] schedule+0x33/0x90 schedule_timeout+0x17d/0x290 ? xfs_buf_find+0x1fd/0x6c0 [xfs] __down_common+0xef/0x125 ? xfs_buf_get_map+0x37/0x230 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agi+0xa8/0x160 [xfs] xfs_iunlink_remove+0x6f/0x2a0 [xfs] ? current_time+0x46/0x80 ? xfs_trans_ichgtime+0x39/0xb0 [xfs] xfs_rename+0x57a/0xae0 [xfs] xfs_vn_rename+0xe4/0x150 [xfs] ... In this patch we move the xfs_iunlink_remove() call to before acquiring the AGF lock to preserve correct AGI/AGF locking order. Signed-off-by: kaixuxia <kaixuxia@tencent.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-04 04:06:50 +00:00
ASSERT(VFS_I(wip)->i_nlink == 0);
pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, wip->i_ino));
error = xfs_iunlink_remove(tp, pag, wip);
xfs_perag_put(pag);
xfs: Fix deadlock between AGI and AGF with RENAME_WHITEOUT When performing rename operation with RENAME_WHITEOUT flag, we will hold AGF lock to allocate or free extents in manipulating the dirents firstly, and then doing the xfs_iunlink_remove() call last to hold AGI lock to modify the tmpfile info, so we the lock order AGI->AGF. The big problem here is that we have an ordering constraint on AGF and AGI locking - inode allocation locks the AGI, then can allocate a new extent for new inodes, locking the AGF after the AGI. Hence the ordering that is imposed by other parts of the code is AGI before AGF. So we get an ABBA deadlock between the AGI and AGF here. Process A: Call trace: ? __schedule+0x2bd/0x620 schedule+0x33/0x90 schedule_timeout+0x17d/0x290 __down_common+0xef/0x125 ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agf+0xa6/0x180 [xfs] ? schedule_timeout+0x17d/0x290 xfs_alloc_read_agf+0x52/0x1f0 [xfs] xfs_alloc_fix_freelist+0x432/0x590 [xfs] ? down+0x3b/0x50 ? xfs_buf_lock+0x34/0xf0 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] xfs_alloc_vextent+0x301/0x6c0 [xfs] xfs_ialloc_ag_alloc+0x182/0x700 [xfs] ? _xfs_trans_bjoin+0x72/0xf0 [xfs] xfs_dialloc+0x116/0x290 [xfs] xfs_ialloc+0x6d/0x5e0 [xfs] ? xfs_log_reserve+0x165/0x280 [xfs] xfs_dir_ialloc+0x8c/0x240 [xfs] xfs_create+0x35a/0x610 [xfs] xfs_generic_create+0x1f1/0x2f0 [xfs] ... Process B: Call trace: ? __schedule+0x2bd/0x620 ? xfs_bmapi_allocate+0x245/0x380 [xfs] schedule+0x33/0x90 schedule_timeout+0x17d/0x290 ? xfs_buf_find+0x1fd/0x6c0 [xfs] __down_common+0xef/0x125 ? xfs_buf_get_map+0x37/0x230 [xfs] ? xfs_buf_find+0x215/0x6c0 [xfs] down+0x3b/0x50 xfs_buf_lock+0x34/0xf0 [xfs] xfs_buf_find+0x215/0x6c0 [xfs] xfs_buf_get_map+0x37/0x230 [xfs] xfs_buf_read_map+0x29/0x190 [xfs] xfs_trans_read_buf_map+0x13d/0x520 [xfs] xfs_read_agi+0xa8/0x160 [xfs] xfs_iunlink_remove+0x6f/0x2a0 [xfs] ? current_time+0x46/0x80 ? xfs_trans_ichgtime+0x39/0xb0 [xfs] xfs_rename+0x57a/0xae0 [xfs] xfs_vn_rename+0xe4/0x150 [xfs] ... In this patch we move the xfs_iunlink_remove() call to before acquiring the AGF lock to preserve correct AGI/AGF locking order. Signed-off-by: kaixuxia <kaixuxia@tencent.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2019-09-04 04:06:50 +00:00
if (error)
goto out_trans_cancel;
xfs_bumplink(tp, wip);
VFS_I(wip)->i_state &= ~I_LINKABLE;
}
/*
* Set up the target.
*/
if (target_ip == NULL) {
/*
* If target does not exist and the rename crosses
* directories, adjust the target directory link count
* to account for the ".." reference from the new entry.
*/
error = xfs_dir_createname(tp, target_dp, target_name,
src_ip->i_ino, spaceres);
if (error)
goto out_trans_cancel;
xfs_trans_ichgtime(tp, target_dp,
XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
if (new_parent && src_is_directory) {
xfs_bumplink(tp, target_dp);
}
} else { /* target_ip != NULL */
/*
* Link the source inode under the target name.
* If the source inode is a directory and we are moving
* it across directories, its ".." entry will be
* inconsistent until we replace that down below.
*
* In case there is already an entry with the same
* name at the destination directory, remove it first.
*/
error = xfs_dir_replace(tp, target_dp, target_name,
src_ip->i_ino, spaceres);
if (error)
goto out_trans_cancel;
xfs_trans_ichgtime(tp, target_dp,
XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
/*
* Decrement the link count on the target since the target
* dir no longer points to it.
*/
error = xfs_droplink(tp, target_ip);
if (error)
goto out_trans_cancel;
if (src_is_directory) {
/*
* Drop the link from the old "." entry.
*/
error = xfs_droplink(tp, target_ip);
if (error)
goto out_trans_cancel;
}
} /* target_ip != NULL */
/*
* Remove the source.
*/
if (new_parent && src_is_directory) {
/*
* Rewrite the ".." entry to point to the new
* directory.
*/
error = xfs_dir_replace(tp, src_ip, &xfs_name_dotdot,
target_dp->i_ino, spaceres);
ASSERT(error != -EEXIST);
if (error)
goto out_trans_cancel;
}
/*
* We always want to hit the ctime on the source inode.
*
* This isn't strictly required by the standards since the source
* inode isn't really being changed, but old unix file systems did
* it and some incremental backup programs won't work without it.
*/
xfs_trans_ichgtime(tp, src_ip, XFS_ICHGTIME_CHG);
xfs_trans_log_inode(tp, src_ip, XFS_ILOG_CORE);
/*
* Adjust the link count on src_dp. This is necessary when
* renaming a directory, either within one parent when
* the target existed, or across two parent directories.
*/
if (src_is_directory && (new_parent || target_ip != NULL)) {
/*
* Decrement link count on src_directory since the
* entry that's moved no longer points to it.
*/
error = xfs_droplink(tp, src_dp);
if (error)
goto out_trans_cancel;
}
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
/*
* For whiteouts, we only need to update the source dirent with the
* inode number of the whiteout inode rather than removing it
* altogether.
*/
if (wip) {
error = xfs_dir_replace(tp, src_dp, src_name, wip->i_ino,
spaceres);
xfs: Check for extent overflow when renaming dir entries A rename operation is essentially a directory entry remove operation from the perspective of parent directory (i.e. src_dp) of rename's source. Hence the only place where we check for extent count overflow for src_dp is in xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns -ENOSPC when it detects a possible extent count overflow and in response, the higher layers of directory handling code do the following: 1. Data/Free blocks: XFS lets these blocks linger until a future remove operation removes them. 2. Dabtree blocks: XFS swaps the blocks with the last block in the Leaf space and unmaps the last block. For target_dp, there are two cases depending on whether the destination directory entry exists or not. When destination directory entry does not exist (i.e. target_ip == NULL), extent count overflow check is performed only when transaction has a non-zero sized space reservation associated with it. With a zero-sized space reservation, XFS allows a rename operation to continue only when the directory has sufficient free space in its data/leaf/free space blocks to hold the new entry. When destination directory entry exists (i.e. target_ip != NULL), all we need to do is change the inode number associated with the already existing entry. Hence there is no need to perform an extent count overflow check. Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2021-01-23 00:48:13 +00:00
} else {
/*
* NOTE: We don't need to check for extent count overflow here
* because the dir remove name code will leave the dir block in
* place if the extent count would overflow.
*/
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
error = xfs_dir_removename(tp, src_dp, src_name, src_ip->i_ino,
spaceres);
xfs: Check for extent overflow when renaming dir entries A rename operation is essentially a directory entry remove operation from the perspective of parent directory (i.e. src_dp) of rename's source. Hence the only place where we check for extent count overflow for src_dp is in xfs_bmap_del_extent_real(). xfs_bmap_del_extent_real() returns -ENOSPC when it detects a possible extent count overflow and in response, the higher layers of directory handling code do the following: 1. Data/Free blocks: XFS lets these blocks linger until a future remove operation removes them. 2. Dabtree blocks: XFS swaps the blocks with the last block in the Leaf space and unmaps the last block. For target_dp, there are two cases depending on whether the destination directory entry exists or not. When destination directory entry does not exist (i.e. target_ip == NULL), extent count overflow check is performed only when transaction has a non-zero sized space reservation associated with it. With a zero-sized space reservation, XFS allows a rename operation to continue only when the directory has sufficient free space in its data/leaf/free space blocks to hold the new entry. When destination directory entry exists (i.e. target_ip != NULL), all we need to do is change the inode number associated with the already existing entry. Hence there is no need to perform an extent count overflow check. Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2021-01-23 00:48:13 +00:00
}
if (error)
goto out_trans_cancel;
xfs_trans_ichgtime(tp, src_dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
xfs_trans_log_inode(tp, src_dp, XFS_ILOG_CORE);
if (new_parent)
xfs_trans_log_inode(tp, target_dp, XFS_ILOG_CORE);
error = xfs_finish_rename(tp);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
if (wip)
xfs_irele(wip);
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
return error;
out_trans_cancel:
xfs_trans_cancel(tp);
out_release_wip:
xfs: add RENAME_WHITEOUT support Whiteouts are used by overlayfs - it has a crazy convention that a whiteout is a character device inode with a major:minor of 0:0. Because it's not documented anywhere, here's an example of what RENAME_WHITEOUT does on ext4: # echo foo > /mnt/scratch/foo # echo bar > /mnt/scratch/bar # ls -l /mnt/scratch total 24 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar -rw-r--r-- 1 root root 4 Feb 11 20:22 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # src/renameat2 -w /mnt/scratch/foo /mnt/scratch/bar # ls -l /mnt/scratch total 20 -rw-r--r-- 1 root root 4 Feb 11 20:22 bar c--------- 1 root root 0, 0 Feb 11 20:23 foo drwx------ 2 root root 16384 Feb 11 20:18 lost+found # cat /mnt/scratch/bar foo # In XFS rename terms, the operation that has been done is that source (foo) has been moved to the target (bar), which is like a nomal rename operation, but rather than the source being removed, it have been replaced with a whiteout. We can't allocate whiteout inodes within the rename transaction due to allocation being a multi-commit transaction: rename needs to be a single, atomic commit. Hence we have several options here, form most efficient to least efficient: - use DT_WHT in the target dirent and do no whiteout inode allocation. The main issue with this approach is that we need hooks in lookup to create a virtual chardev inode to present to userspace and in places where we might need to modify the dirent e.g. unlink. Overlayfs also needs to be taught about DT_WHT. Most invasive change, lowest overhead. - create a special whiteout inode in the root directory (e.g. a ".wino" dirent) and then hardlink every new whiteout to it. This means we only need to create a single whiteout inode, and rename simply creates a hardlink to it. We can use DT_WHT for these, though using DT_CHR means we won't have to modify overlayfs, nor anything in userspace. Downside is we have to look up the whiteout inode on every operation and create it if it doesn't exist. - copy ext4: create a special whiteout chardev inode for every whiteout. This is more complex than the above options because of the lack of atomicity between inode creation and the rename operation, requiring us to create a tmpfile inode and then linking it into the directory structure during the rename. At least with a tmpfile inode crashes between the create and rename doesn't leave unreferenced inodes or directory pollution around. By far the simplest thing to do in the short term is to copy ext4. While it is the most inefficient way of supporting whiteouts, but as an initial implementation we can simply reuse existing functions and add a small amount of extra code the the rename operation. When we get full whiteout support in the VFS (via the dentry cache) we can then look to supporting DT_WHT method outlined as the first method of supporting whiteouts. But until then, we'll stick with what overlayfs expects us to be: dumb and stupid. Signed-off-by: Dave Chinner <dchinner@redhat.com>
2015-03-25 03:08:08 +00:00
if (wip)
xfs_irele(wip);
return error;
}
static int
xfs_iflush(
struct xfs_inode *ip,
struct xfs_buf *bp)
{
struct xfs_inode_log_item *iip = ip->i_itemp;
struct xfs_dinode *dip;
struct xfs_mount *mp = ip->i_mount;
int error;
ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
ASSERT(xfs_iflags_test(ip, XFS_IFLUSHING));
ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_BTREE ||
ip->i_df.if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK));
ASSERT(iip->ili_item.li_buf == bp);
dip = xfs_buf_offset(bp, ip->i_imap.im_boffset);
/*
* We don't flush the inode if any of the following checks fail, but we
* do still update the log item and attach to the backing buffer as if
* the flush happened. This is a formality to facilitate predictable
* error handling as the caller will shutdown and fail the buffer.
*/
error = -EFSCORRUPTED;
if (XFS_TEST_ERROR(dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC),
mp, XFS_ERRTAG_IFLUSH_1)) {
xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
"%s: Bad inode %Lu magic number 0x%x, ptr "PTR_FMT,
__func__, ip->i_ino, be16_to_cpu(dip->di_magic), dip);
goto flush_out;
}
if (S_ISREG(VFS_I(ip)->i_mode)) {
if (XFS_TEST_ERROR(
ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
ip->i_df.if_format != XFS_DINODE_FMT_BTREE,
mp, XFS_ERRTAG_IFLUSH_3)) {
xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
"%s: Bad regular inode %Lu, ptr "PTR_FMT,
__func__, ip->i_ino, ip);
goto flush_out;
}
} else if (S_ISDIR(VFS_I(ip)->i_mode)) {
if (XFS_TEST_ERROR(
ip->i_df.if_format != XFS_DINODE_FMT_EXTENTS &&
ip->i_df.if_format != XFS_DINODE_FMT_BTREE &&
ip->i_df.if_format != XFS_DINODE_FMT_LOCAL,
mp, XFS_ERRTAG_IFLUSH_4)) {
xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
"%s: Bad directory inode %Lu, ptr "PTR_FMT,
__func__, ip->i_ino, ip);
goto flush_out;
}
}
if (XFS_TEST_ERROR(ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp) >
ip->i_nblocks, mp, XFS_ERRTAG_IFLUSH_5)) {
xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
"%s: detected corrupt incore inode %Lu, "
"total extents = %d, nblocks = %Ld, ptr "PTR_FMT,
__func__, ip->i_ino,
ip->i_df.if_nextents + xfs_ifork_nextents(ip->i_afp),
ip->i_nblocks, ip);
goto flush_out;
}
if (XFS_TEST_ERROR(ip->i_forkoff > mp->m_sb.sb_inodesize,
mp, XFS_ERRTAG_IFLUSH_6)) {
xfs_alert_tag(mp, XFS_PTAG_IFLUSH,
"%s: bad inode %Lu, forkoff 0x%x, ptr "PTR_FMT,
__func__, ip->i_ino, ip->i_forkoff, ip);
goto flush_out;
}
xfs: di_flushiter considered harmful When we made all inode updates transactional, we no longer needed the log recovery detection for inodes being newer on disk than the transaction being replayed - it was redundant as replay of the log would always result in the latest version of the inode would be on disk. It was redundant, but left in place because it wasn't considered to be a problem. However, with the new "don't read inodes on create" optimisation, flushiter has come back to bite us. Essentially, the optimisation made always initialises flushiter to zero in the create transaction, and so if we then crash and run recovery and the inode already on disk has a non-zero flushiter it will skip recovery of that inode. As a result, log recovery does the wrong thing and we end up with a corrupt filesystem. Because we have to support old kernel to new kernel upgrades, we can't just get rid of the flushiter support in log recovery as we might be upgrading from a kernel that doesn't have fully transactional inode updates. Unfortunately, for v4 superblocks there is no way to guarantee that log recovery knows about this fact. We cannot add a new inode format flag to say it's a "special inode create" because it won't be understood by older kernels and so recovery could do the wrong thing on downgrade. We cannot specially detect the combination of zero mode/non-zero flushiter on disk to non-zero mode, zero flushiter in the log item during recovery because wrapping of the flushiter can result in false detection. Hence that makes this "don't use flushiter" optimisation limited to a disk format that guarantees that we don't need it. And that means the only fix here is to limit the "no read IO on create" optimisation to version 5 superblocks.... Reported-by: Markus Trippelsdorf <markus@trippelsdorf.de> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Mark Tinguely <tinguely@sgi.com> Signed-off-by: Ben Myers <bpm@sgi.com>
2013-07-24 05:47:30 +00:00
/*
* Inode item log recovery for v2 inodes are dependent on the flushiter
* count for correct sequencing. We bump the flush iteration count so
* we can detect flushes which postdate a log record during recovery.
* This is redundant as we now log every change and hence this can't
* happen but we need to still do it to ensure backwards compatibility
* with old kernels that predate logging all inode changes.
*/
if (!xfs_sb_version_has_v3inode(&mp->m_sb))
ip->i_flushiter++;
/*
* If there are inline format data / attr forks attached to this inode,
* make sure they are not corrupt.
*/
if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL &&
xfs_ifork_verify_local_data(ip))
goto flush_out;
if (ip->i_afp && ip->i_afp->if_format == XFS_DINODE_FMT_LOCAL &&
xfs_ifork_verify_local_attr(ip))
goto flush_out;
/*
* Copy the dirty parts of the inode into the on-disk inode. We always
* copy out the core of the inode, because if the inode is dirty at all
* the core must be.
*/
xfs_inode_to_disk(ip, dip, iip->ili_item.li_lsn);
/* Wrap, we never let the log put out DI_MAX_FLUSH */
if (!xfs_sb_version_has_v3inode(&mp->m_sb)) {
if (ip->i_flushiter == DI_MAX_FLUSH)
ip->i_flushiter = 0;
}
xfs_iflush_fork(ip, dip, iip, XFS_DATA_FORK);
if (XFS_IFORK_Q(ip))
xfs_iflush_fork(ip, dip, iip, XFS_ATTR_FORK);
/*
* We've recorded everything logged in the inode, so we'd like to clear
* the ili_fields bits so we don't log and flush things unnecessarily.
* However, we can't stop logging all this information until the data
* we've copied into the disk buffer is written to disk. If we did we
* might overwrite the copy of the inode in the log with all the data
* after re-logging only part of it, and in the face of a crash we
* wouldn't have all the data we need to recover.
*
* What we do is move the bits to the ili_last_fields field. When
* logging the inode, these bits are moved back to the ili_fields field.
* In the xfs_buf_inode_iodone() routine we clear ili_last_fields, since
* we know that the information those bits represent is permanently on
* disk. As long as the flush completes before the inode is logged
* again, then both ili_fields and ili_last_fields will be cleared.
*/
error = 0;
flush_out:
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
spin_lock(&iip->ili_lock);
iip->ili_last_fields = iip->ili_fields;
iip->ili_fields = 0;
xfs: optimise away log forces on timestamp updates for fdatasync xfs: timestamp updates cause excessive fdatasync log traffic Sage Weil reported that a ceph test workload was writing to the log on every fdatasync during an overwrite workload. Event tracing showed that the only metadata modification being made was the timestamp updates during the write(2) syscall, but fdatasync(2) is supposed to ignore them. The key observation was that the transactions in the log all looked like this: INODE: #regs: 4 ino: 0x8b flags: 0x45 dsize: 32 And contained a flags field of 0x45 or 0x85, and had data and attribute forks following the inode core. This means that the timestamp updates were triggering dirty relogging of previously logged parts of the inode that hadn't yet been flushed back to disk. There are two parts to this problem. The first is that XFS relogs dirty regions in subsequent transactions, so it carries around the fields that have been dirtied since the last time the inode was written back to disk, not since the last time the inode was forced into the log. The second part is that on v5 filesystems, the inode change count update during inode dirtying also sets the XFS_ILOG_CORE flag, so on v5 filesystems this makes a timestamp update dirty the entire inode. As a result when fdatasync is run, it looks at the dirty fields in the inode, and sees more than just the timestamp flag, even though the only metadata change since the last fdatasync was just the timestamps. Hence we force the log on every subsequent fdatasync even though it is not needed. To fix this, add a new field to the inode log item that tracks changes since the last time fsync/fdatasync forced the log to flush the changes to the journal. This flag is updated when we dirty the inode, but we do it before updating the change count so it does not carry the "core dirty" flag from timestamp updates. The fields are zeroed when the inode is marked clean (due to writeback/freeing) or when an fsync/datasync forces the log. Hence if we only dirty the timestamps on the inode between fsync/fdatasync calls, the fdatasync will not trigger another log force. Over 100 runs of the test program: Ext4 baseline: runtime: 1.63s +/- 0.24s avg lat: 1.59ms +/- 0.24ms iops: ~2000 XFS, vanilla kernel: runtime: 2.45s +/- 0.18s avg lat: 2.39ms +/- 0.18ms log forces: ~400/s iops: ~1000 XFS, patched kernel: runtime: 1.49s +/- 0.26s avg lat: 1.46ms +/- 0.25ms log forces: ~30/s iops: ~1500 Reported-by: Sage Weil <sage@redhat.com> Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2015-11-03 02:14:59 +00:00
iip->ili_fsync_fields = 0;
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
spin_unlock(&iip->ili_lock);
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
/*
* Store the current LSN of the inode so that we can tell whether the
* item has moved in the AIL from xfs_buf_inode_iodone().
xfs: add an inode item lock The inode log item is kind of special in that it can be aggregating new changes in memory at the same time time existing changes are being written back to disk. This means there are fields in the log item that are accessed concurrently from contexts that don't share any locking at all. e.g. updating ili_last_fields occurs at flush time under the ILOCK_EXCL and flush lock at flush time, under the flush lock at IO completion time, and is read under the ILOCK_EXCL when the inode is logged. Hence there is no actual serialisation between reading the field during logging of the inode in transactions vs clearing the field in IO completion. We currently get away with this by the fact that we are only clearing fields in IO completion, and nothing bad happens if we accidentally log more of the inode than we actually modify. Worst case is we consume a tiny bit more memory and log bandwidth. However, if we want to do more complex state manipulations on the log item that requires updates at all three of these potential locations, we need to have some mechanism of serialising those operations. To do this, introduce a spinlock into the log item to serialise internal state. This could be done via the xfs_inode i_flags_lock, but this then leads to potential lock inversion issues where inode flag updates need to occur inside locks that best nest inside the inode log item locks (e.g. marking inodes stale during inode cluster freeing). Using a separate spinlock avoids these sorts of problems and simplifies future code. This does not touch the use of ili_fields in the item formatting code - that is entirely protected by the ILOCK_EXCL at this point in time, so it remains untouched. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-06-29 21:48:46 +00:00
*/
xfs_trans_ail_copy_lsn(mp->m_ail, &iip->ili_flush_lsn,
&iip->ili_item.li_lsn);
/* generate the checksum. */
xfs_dinode_calc_crc(mp, dip);
return error;
}
/*
* Non-blocking flush of dirty inode metadata into the backing buffer.
*
* The caller must have a reference to the inode and hold the cluster buffer
* locked. The function will walk across all the inodes on the cluster buffer it
* can find and lock without blocking, and flush them to the cluster buffer.
*
* On successful flushing of at least one inode, the caller must write out the
* buffer and release it. If no inodes are flushed, -EAGAIN will be returned and
* the caller needs to release the buffer. On failure, the filesystem will be
* shut down, the buffer will have been unlocked and released, and EFSCORRUPTED
* will be returned.
*/
int
xfs_iflush_cluster(
struct xfs_buf *bp)
{
struct xfs_mount *mp = bp->b_mount;
struct xfs_log_item *lip, *n;
struct xfs_inode *ip;
struct xfs_inode_log_item *iip;
int clcount = 0;
int error = 0;
/*
* We must use the safe variant here as on shutdown xfs_iflush_abort()
* can remove itself from the list.
*/
list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {
iip = (struct xfs_inode_log_item *)lip;
ip = iip->ili_inode;
/*
* Quick and dirty check to avoid locks if possible.
*/
if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING))
continue;
if (xfs_ipincount(ip))
continue;
/*
* The inode is still attached to the buffer, which means it is
* dirty but reclaim might try to grab it. Check carefully for
* that, and grab the ilock while still holding the i_flags_lock
* to guarantee reclaim will not be able to reclaim this inode
* once we drop the i_flags_lock.
*/
spin_lock(&ip->i_flags_lock);
ASSERT(!__xfs_iflags_test(ip, XFS_ISTALE));
if (__xfs_iflags_test(ip, XFS_IRECLAIM | XFS_IFLUSHING)) {
spin_unlock(&ip->i_flags_lock);
continue;
}
/*
* ILOCK will pin the inode against reclaim and prevent
* concurrent transactions modifying the inode while we are
* flushing the inode. If we get the lock, set the flushing
* state before we drop the i_flags_lock.
*/
if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
spin_unlock(&ip->i_flags_lock);
continue;
}
__xfs_iflags_set(ip, XFS_IFLUSHING);
spin_unlock(&ip->i_flags_lock);
/*
* Abort flushing this inode if we are shut down because the
* inode may not currently be in the AIL. This can occur when
* log I/O failure unpins the inode without inserting into the
* AIL, leaving a dirty/unpinned inode attached to the buffer
* that otherwise looks like it should be flushed.
*/
if (XFS_FORCED_SHUTDOWN(mp)) {
xfs_iunpin_wait(ip);
xfs_iflush_abort(ip);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
error = -EIO;
continue;
}
/* don't block waiting on a log force to unpin dirty inodes */
if (xfs_ipincount(ip)) {
xfs_iflags_clear(ip, XFS_IFLUSHING);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
continue;
}
if (!xfs_inode_clean(ip))
error = xfs_iflush(ip, bp);
else
xfs_iflags_clear(ip, XFS_IFLUSHING);
xfs_iunlock(ip, XFS_ILOCK_SHARED);
if (error)
break;
clcount++;
}
if (error) {
bp->b_flags |= XBF_ASYNC;
xfs_buf_ioend_fail(bp);
xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
return error;
}
if (!clcount)
return -EAGAIN;
XFS_STATS_INC(mp, xs_icluster_flushcnt);
XFS_STATS_ADD(mp, xs_icluster_flushinode, clcount);
return 0;
}
/* Release an inode. */
void
xfs_irele(
struct xfs_inode *ip)
{
trace_xfs_irele(ip, _RET_IP_);
iput(VFS_I(ip));
}
/*
* Ensure all commited transactions touching the inode are written to the log.
*/
int
xfs_log_force_inode(
struct xfs_inode *ip)
{
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
xfs_csn_t seq = 0;
xfs_ilock(ip, XFS_ILOCK_SHARED);
if (xfs_ipincount(ip))
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
seq = ip->i_itemp->ili_commit_seq;
xfs_iunlock(ip, XFS_ILOCK_SHARED);
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
if (!seq)
return 0;
xfs: xfs_log_force_lsn isn't passed a LSN In doing an investigation into AIL push stalls, I was looking at the log force code to see if an async CIL push could be done instead. This lead me to xfs_log_force_lsn() and looking at how it works. xfs_log_force_lsn() is only called from inode synchronisation contexts such as fsync(), and it takes the ip->i_itemp->ili_last_lsn value as the LSN to sync the log to. This gets passed to xlog_cil_force_lsn() via xfs_log_force_lsn() to flush the CIL to the journal, and then used by xfs_log_force_lsn() to flush the iclogs to the journal. The problem is that ip->i_itemp->ili_last_lsn does not store a log sequence number. What it stores is passed to it from the ->iop_committing method, which is called by xfs_log_commit_cil(). The value this passes to the iop_committing method is the CIL context sequence number that the item was committed to. As it turns out, xlog_cil_force_lsn() converts the sequence to an actual commit LSN for the related context and returns that to xfs_log_force_lsn(). xfs_log_force_lsn() overwrites it's "lsn" variable that contained a sequence with an actual LSN and then uses that to sync the iclogs. This caused me some confusion for a while, even though I originally wrote all this code a decade ago. ->iop_committing is only used by a couple of log item types, and only inode items use the sequence number it is passed. Let's clean up the API, CIL structures and inode log item to call it a sequence number, and make it clear that the high level code is using CIL sequence numbers and not on-disk LSNs for integrity synchronisation purposes. Signed-off-by: Dave Chinner <dchinner@redhat.com> Reviewed-by: Brian Foster <bfoster@redhat.com> Reviewed-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Allison Henderson <allison.henderson@oracle.com> Signed-off-by: Darrick J. Wong <djwong@kernel.org>
2021-06-18 15:21:52 +00:00
return xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC, NULL);
}
/*
* Grab the exclusive iolock for a data copy from src to dest, making sure to
* abide vfs locking order (lowest pointer value goes first) and breaking the
* layout leases before proceeding. The loop is needed because we cannot call
* the blocking break_layout() with the iolocks held, and therefore have to
* back out both locks.
*/
static int
xfs_iolock_two_inodes_and_break_layout(
struct inode *src,
struct inode *dest)
{
int error;
if (src > dest)
swap(src, dest);
retry:
/* Wait to break both inodes' layouts before we start locking. */
error = break_layout(src, true);
if (error)
return error;
if (src != dest) {
error = break_layout(dest, true);
if (error)
return error;
}
/* Lock one inode and make sure nobody got in and leased it. */
inode_lock(src);
error = break_layout(src, false);
if (error) {
inode_unlock(src);
if (error == -EWOULDBLOCK)
goto retry;
return error;
}
if (src == dest)
return 0;
/* Lock the other inode and make sure nobody got in and leased it. */
inode_lock_nested(dest, I_MUTEX_NONDIR2);
error = break_layout(dest, false);
if (error) {
inode_unlock(src);
inode_unlock(dest);
if (error == -EWOULDBLOCK)
goto retry;
return error;
}
return 0;
}
/*
* Lock two inodes so that userspace cannot initiate I/O via file syscalls or
* mmap activity.
*/
int
xfs_ilock2_io_mmap(
struct xfs_inode *ip1,
struct xfs_inode *ip2)
{
int ret;
ret = xfs_iolock_two_inodes_and_break_layout(VFS_I(ip1), VFS_I(ip2));
if (ret)
return ret;
if (ip1 == ip2)
xfs_ilock(ip1, XFS_MMAPLOCK_EXCL);
else
xfs_lock_two_inodes(ip1, XFS_MMAPLOCK_EXCL,
ip2, XFS_MMAPLOCK_EXCL);
return 0;
}
/* Unlock both inodes to allow IO and mmap activity. */
void
xfs_iunlock2_io_mmap(
struct xfs_inode *ip1,
struct xfs_inode *ip2)
{
bool same_inode = (ip1 == ip2);
xfs_iunlock(ip2, XFS_MMAPLOCK_EXCL);
if (!same_inode)
xfs_iunlock(ip1, XFS_MMAPLOCK_EXCL);
inode_unlock(VFS_I(ip2));
if (!same_inode)
inode_unlock(VFS_I(ip1));
}