linux/fs/bcachefs/fs-io.c

1082 lines
27 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
#ifndef NO_BCACHEFS_FS
#include "bcachefs.h"
#include "alloc_foreground.h"
#include "bkey_buf.h"
#include "btree_update.h"
#include "buckets.h"
#include "clock.h"
#include "error.h"
#include "extents.h"
#include "extent_update.h"
#include "fs.h"
#include "fs-io.h"
#include "fs-io-buffered.h"
#include "fs-io-pagecache.h"
#include "fsck.h"
#include "inode.h"
#include "journal.h"
#include "io_misc.h"
#include "keylist.h"
#include "quota.h"
#include "reflink.h"
#include "trace.h"
#include <linux/aio.h>
#include <linux/backing-dev.h>
#include <linux/falloc.h>
#include <linux/migrate.h>
#include <linux/mmu_context.h>
#include <linux/pagevec.h>
#include <linux/rmap.h>
#include <linux/sched/signal.h>
#include <linux/task_io_accounting_ops.h>
#include <linux/uio.h>
#include <trace/events/writeback.h>
struct nocow_flush {
struct closure *cl;
struct bch_dev *ca;
struct bio bio;
};
static void nocow_flush_endio(struct bio *_bio)
{
struct nocow_flush *bio = container_of(_bio, struct nocow_flush, bio);
closure_put(bio->cl);
percpu_ref_put(&bio->ca->io_ref);
bio_put(&bio->bio);
}
void bch2_inode_flush_nocow_writes_async(struct bch_fs *c,
struct bch_inode_info *inode,
struct closure *cl)
{
struct nocow_flush *bio;
struct bch_dev *ca;
struct bch_devs_mask devs;
unsigned dev;
dev = find_first_bit(inode->ei_devs_need_flush.d, BCH_SB_MEMBERS_MAX);
if (dev == BCH_SB_MEMBERS_MAX)
return;
devs = inode->ei_devs_need_flush;
memset(&inode->ei_devs_need_flush, 0, sizeof(inode->ei_devs_need_flush));
for_each_set_bit(dev, devs.d, BCH_SB_MEMBERS_MAX) {
rcu_read_lock();
ca = rcu_dereference(c->devs[dev]);
if (ca && !percpu_ref_tryget(&ca->io_ref))
ca = NULL;
rcu_read_unlock();
if (!ca)
continue;
bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev, 0,
REQ_OP_WRITE|REQ_PREFLUSH,
GFP_KERNEL,
&c->nocow_flush_bioset),
struct nocow_flush, bio);
bio->cl = cl;
bio->ca = ca;
bio->bio.bi_end_io = nocow_flush_endio;
closure_bio_submit(&bio->bio, cl);
}
}
static int bch2_inode_flush_nocow_writes(struct bch_fs *c,
struct bch_inode_info *inode)
{
struct closure cl;
closure_init_stack(&cl);
bch2_inode_flush_nocow_writes_async(c, inode, &cl);
closure_sync(&cl);
return 0;
}
/* i_size updates: */
struct inode_new_size {
loff_t new_size;
u64 now;
unsigned fields;
};
static int inode_set_size(struct btree_trans *trans,
struct bch_inode_info *inode,
struct bch_inode_unpacked *bi,
void *p)
{
struct inode_new_size *s = p;
bi->bi_size = s->new_size;
if (s->fields & ATTR_ATIME)
bi->bi_atime = s->now;
if (s->fields & ATTR_MTIME)
bi->bi_mtime = s->now;
if (s->fields & ATTR_CTIME)
bi->bi_ctime = s->now;
return 0;
}
int __must_check bch2_write_inode_size(struct bch_fs *c,
struct bch_inode_info *inode,
loff_t new_size, unsigned fields)
{
struct inode_new_size s = {
.new_size = new_size,
.now = bch2_current_time(c),
.fields = fields,
};
return bch2_write_inode(c, inode, inode_set_size, &s, fields);
}
void __bch2_i_sectors_acct(struct bch_fs *c, struct bch_inode_info *inode,
struct quota_res *quota_res, s64 sectors)
{
bch2_fs_inconsistent_on((s64) inode->v.i_blocks + sectors < 0, c,
"inode %lu i_blocks underflow: %llu + %lli < 0 (ondisk %lli)",
inode->v.i_ino, (u64) inode->v.i_blocks, sectors,
inode->ei_inode.bi_sectors);
inode->v.i_blocks += sectors;
#ifdef CONFIG_BCACHEFS_QUOTA
if (quota_res &&
!test_bit(EI_INODE_SNAPSHOT, &inode->ei_flags) &&
sectors > 0) {
BUG_ON(sectors > quota_res->sectors);
BUG_ON(sectors > inode->ei_quota_reserved);
quota_res->sectors -= sectors;
inode->ei_quota_reserved -= sectors;
} else {
bch2_quota_acct(c, inode->ei_qid, Q_SPC, sectors, KEY_TYPE_QUOTA_WARN);
}
#endif
}
/* fsync: */
/*
* inode->ei_inode.bi_journal_seq won't be up to date since it's set in an
* insert trigger: look up the btree inode instead
*/
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
static int bch2_flush_inode(struct bch_fs *c,
struct bch_inode_info *inode)
{
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
struct bch_inode_unpacked u;
int ret;
if (c->opts.journal_flush_disabled)
return 0;
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
ret = bch2_inode_find_by_inum(c, inode_inum(inode), &u);
if (ret)
return ret;
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
return bch2_journal_flush_seq(&c->journal, u.bi_journal_seq) ?:
bch2_inode_flush_nocow_writes(c, inode);
}
int bch2_fsync(struct file *file, loff_t start, loff_t end, int datasync)
{
struct bch_inode_info *inode = file_bch_inode(file);
struct bch_fs *c = inode->v.i_sb->s_fs_info;
int ret;
ret = file_write_and_wait_range(file, start, end);
if (ret)
goto out;
ret = sync_inode_metadata(&inode->v, 1);
if (ret)
goto out;
ret = bch2_flush_inode(c, inode);
out:
return bch2_err_class(ret);
}
/* truncate: */
static inline int range_has_data(struct bch_fs *c, u32 subvol,
struct bpos start,
struct bpos end)
{
struct btree_trans *trans = bch2_trans_get(c);
struct btree_iter iter;
struct bkey_s_c k;
int ret = 0;
retry:
bch2_trans_begin(trans);
ret = bch2_subvolume_get_snapshot(trans, subvol, &start.snapshot);
if (ret)
goto err;
for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents, start, end, 0, k, ret)
bcachefs: fix truncate overflow if folio is beyond EOF generic/083 occasionally reproduces a panic caused by an overflow when accessing the bch_folio_sector array of the folio being processed by __bch2_truncate_folio(). The immediate cause of the overflow is that the folio offset is beyond i_size, and therefore the sector index calculation underflows on subtraction of the folio offset. One cause of this is mainly observed on nocow mounts. When nocow is enabled, fallocate performs physical block allocation (as opposed to block reservation in cow mode), which range_has_data() then interprets as valid data that requires partial zeroing on truncate. Therefore, if a post-eof zero range request lands across post-eof preallocated blocks, __bch2_truncate_folio() may actually create a post-eof folio in order to perform zeroing. To avoid this problem, update range_has_data() to filter out unwritten blocks from folio creation and partial zeroing. Even though we should never create folios beyond EOF like this, the mere existence of such folios is not necessarily a fatal error. Fix up the truncate code to warn about this condition and not overflow the sector array and possibly crash the system. The addition of this warning without the corresponding unwritten extent fix has shown that various other fstests are able to reproduce this problem fairly frequently, but often in ways that doesn't necessarily result in a kernel panic or a change in user observable behavior, and therefore the problem goes undetected. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-29 13:49:04 +00:00
if (bkey_extent_is_data(k.k) && !bkey_extent_is_unwritten(k)) {
ret = 1;
break;
}
start = iter.pos;
bch2_trans_iter_exit(trans, &iter);
err:
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
goto retry;
bch2_trans_put(trans);
return ret;
}
static int __bch2_truncate_folio(struct bch_inode_info *inode,
pgoff_t index, loff_t start, loff_t end)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct address_space *mapping = inode->v.i_mapping;
struct bch_folio *s;
unsigned start_offset;
unsigned end_offset;
unsigned i;
struct folio *folio;
s64 i_sectors_delta = 0;
int ret = 0;
u64 end_pos;
folio = filemap_lock_folio(mapping, index);
if (IS_ERR_OR_NULL(folio)) {
/*
* XXX: we're doing two index lookups when we end up reading the
* folio
*/
ret = range_has_data(c, inode->ei_subvol,
POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT)),
POS(inode->v.i_ino, (index << PAGE_SECTORS_SHIFT) + PAGE_SECTORS));
if (ret <= 0)
return ret;
folio = __filemap_get_folio(mapping, index,
FGP_LOCK|FGP_CREAT, GFP_KERNEL);
if (IS_ERR_OR_NULL(folio)) {
ret = -ENOMEM;
goto out;
}
}
BUG_ON(start >= folio_end_pos(folio));
BUG_ON(end <= folio_pos(folio));
start_offset = max(start, folio_pos(folio)) - folio_pos(folio);
end_offset = min_t(u64, end, folio_end_pos(folio)) - folio_pos(folio);
/* Folio boundary? Nothing to do */
if (start_offset == 0 &&
end_offset == folio_size(folio)) {
ret = 0;
goto unlock;
}
s = bch2_folio_create(folio, 0);
if (!s) {
ret = -ENOMEM;
goto unlock;
}
if (!folio_test_uptodate(folio)) {
ret = bch2_read_single_folio(folio, mapping);
if (ret)
goto unlock;
}
ret = bch2_folio_set(c, inode_inum(inode), &folio, 1);
if (ret)
goto unlock;
for (i = round_up(start_offset, block_bytes(c)) >> 9;
i < round_down(end_offset, block_bytes(c)) >> 9;
i++) {
s->s[i].nr_replicas = 0;
i_sectors_delta -= s->s[i].state == SECTOR_dirty;
bch2_folio_sector_set(folio, s, i, SECTOR_unallocated);
}
bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
/*
* Caller needs to know whether this folio will be written out by
* writeback - doing an i_size update if necessary - or whether it will
bcachefs: fix truncate overflow if folio is beyond EOF generic/083 occasionally reproduces a panic caused by an overflow when accessing the bch_folio_sector array of the folio being processed by __bch2_truncate_folio(). The immediate cause of the overflow is that the folio offset is beyond i_size, and therefore the sector index calculation underflows on subtraction of the folio offset. One cause of this is mainly observed on nocow mounts. When nocow is enabled, fallocate performs physical block allocation (as opposed to block reservation in cow mode), which range_has_data() then interprets as valid data that requires partial zeroing on truncate. Therefore, if a post-eof zero range request lands across post-eof preallocated blocks, __bch2_truncate_folio() may actually create a post-eof folio in order to perform zeroing. To avoid this problem, update range_has_data() to filter out unwritten blocks from folio creation and partial zeroing. Even though we should never create folios beyond EOF like this, the mere existence of such folios is not necessarily a fatal error. Fix up the truncate code to warn about this condition and not overflow the sector array and possibly crash the system. The addition of this warning without the corresponding unwritten extent fix has shown that various other fstests are able to reproduce this problem fairly frequently, but often in ways that doesn't necessarily result in a kernel panic or a change in user observable behavior, and therefore the problem goes undetected. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-29 13:49:04 +00:00
* be responsible for the i_size update.
*
* Note that we shouldn't ever see a folio beyond EOF, but check and
* warn if so. This has been observed by failure to clean up folios
* after a short write and there's still a chance reclaim will fix
* things up.
*/
bcachefs: fix truncate overflow if folio is beyond EOF generic/083 occasionally reproduces a panic caused by an overflow when accessing the bch_folio_sector array of the folio being processed by __bch2_truncate_folio(). The immediate cause of the overflow is that the folio offset is beyond i_size, and therefore the sector index calculation underflows on subtraction of the folio offset. One cause of this is mainly observed on nocow mounts. When nocow is enabled, fallocate performs physical block allocation (as opposed to block reservation in cow mode), which range_has_data() then interprets as valid data that requires partial zeroing on truncate. Therefore, if a post-eof zero range request lands across post-eof preallocated blocks, __bch2_truncate_folio() may actually create a post-eof folio in order to perform zeroing. To avoid this problem, update range_has_data() to filter out unwritten blocks from folio creation and partial zeroing. Even though we should never create folios beyond EOF like this, the mere existence of such folios is not necessarily a fatal error. Fix up the truncate code to warn about this condition and not overflow the sector array and possibly crash the system. The addition of this warning without the corresponding unwritten extent fix has shown that various other fstests are able to reproduce this problem fairly frequently, but often in ways that doesn't necessarily result in a kernel panic or a change in user observable behavior, and therefore the problem goes undetected. Signed-off-by: Brian Foster <bfoster@redhat.com> Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2023-03-29 13:49:04 +00:00
WARN_ON_ONCE(folio_pos(folio) >= inode->v.i_size);
end_pos = folio_end_pos(folio);
if (inode->v.i_size > folio_pos(folio))
end_pos = min_t(u64, inode->v.i_size, end_pos);
ret = s->s[folio_pos_to_s(folio, end_pos - 1)].state >= SECTOR_dirty;
folio_zero_segment(folio, start_offset, end_offset);
/*
* Bit of a hack - we don't want truncate to fail due to -ENOSPC.
*
* XXX: because we aren't currently tracking whether the folio has actual
* data in it (vs. just 0s, or only partially written) this wrong. ick.
*/
BUG_ON(bch2_get_folio_disk_reservation(c, inode, folio, false));
/*
* This removes any writeable userspace mappings; we need to force
* .page_mkwrite to be called again before any mmapped writes, to
* redirty the full page:
*/
folio_mkclean(folio);
filemap_dirty_folio(mapping, folio);
unlock:
folio_unlock(folio);
folio_put(folio);
out:
return ret;
}
static int bch2_truncate_folio(struct bch_inode_info *inode, loff_t from)
{
return __bch2_truncate_folio(inode, from >> PAGE_SHIFT,
from, ANYSINT_MAX(loff_t));
}
static int bch2_truncate_folios(struct bch_inode_info *inode,
loff_t start, loff_t end)
{
int ret = __bch2_truncate_folio(inode, start >> PAGE_SHIFT,
start, end);
if (ret >= 0 &&
start >> PAGE_SHIFT != end >> PAGE_SHIFT)
ret = __bch2_truncate_folio(inode,
(end - 1) >> PAGE_SHIFT,
start, end);
return ret;
}
static int bch2_extend(struct mnt_idmap *idmap,
struct bch_inode_info *inode,
struct bch_inode_unpacked *inode_u,
struct iattr *iattr)
{
struct address_space *mapping = inode->v.i_mapping;
int ret;
/*
* sync appends:
*
* this has to be done _before_ extending i_size:
*/
ret = filemap_write_and_wait_range(mapping, inode_u->bi_size, S64_MAX);
if (ret)
return ret;
truncate_setsize(&inode->v, iattr->ia_size);
return bch2_setattr_nonsize(idmap, inode, iattr);
}
int bchfs_truncate(struct mnt_idmap *idmap,
struct bch_inode_info *inode, struct iattr *iattr)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct address_space *mapping = inode->v.i_mapping;
struct bch_inode_unpacked inode_u;
s64 i_sectors_delta = 0;
int ret = 0;
/*
* If the truncate call with change the size of the file, the
* cmtimes should be updated. If the size will not change, we
* do not need to update the cmtimes.
*/
if (iattr->ia_size != inode->v.i_size) {
if (!(iattr->ia_valid & ATTR_MTIME))
ktime_get_coarse_real_ts64(&iattr->ia_mtime);
if (!(iattr->ia_valid & ATTR_CTIME))
ktime_get_coarse_real_ts64(&iattr->ia_ctime);
iattr->ia_valid |= ATTR_MTIME|ATTR_CTIME;
}
inode_dio_wait(&inode->v);
bch2_pagecache_block_get(inode);
ret = bch2_inode_find_by_inum(c, inode_inum(inode), &inode_u);
if (ret)
goto err;
/*
* check this before next assertion; on filesystem error our normal
* invariants are a bit broken (truncate has to truncate the page cache
* before the inode).
*/
ret = bch2_journal_error(&c->journal);
if (ret)
goto err;
WARN_ONCE(!test_bit(EI_INODE_ERROR, &inode->ei_flags) &&
inode->v.i_size < inode_u.bi_size,
"truncate spotted in mem i_size < btree i_size: %llu < %llu\n",
(u64) inode->v.i_size, inode_u.bi_size);
if (iattr->ia_size > inode->v.i_size) {
ret = bch2_extend(idmap, inode, &inode_u, iattr);
goto err;
}
iattr->ia_valid &= ~ATTR_SIZE;
ret = bch2_truncate_folio(inode, iattr->ia_size);
if (unlikely(ret < 0))
goto err;
truncate_setsize(&inode->v, iattr->ia_size);
/*
* When extending, we're going to write the new i_size to disk
* immediately so we need to flush anything above the current on disk
* i_size first:
*
* Also, when extending we need to flush the page that i_size currently
* straddles - if it's mapped to userspace, we need to ensure that
* userspace has to redirty it and call .mkwrite -> set_page_dirty
* again to allocate the part of the page that was extended.
*/
if (iattr->ia_size > inode_u.bi_size)
ret = filemap_write_and_wait_range(mapping,
inode_u.bi_size,
iattr->ia_size - 1);
else if (iattr->ia_size & (PAGE_SIZE - 1))
ret = filemap_write_and_wait_range(mapping,
round_down(iattr->ia_size, PAGE_SIZE),
iattr->ia_size - 1);
if (ret)
goto err;
ret = bch2_truncate(c, inode_inum(inode), iattr->ia_size, &i_sectors_delta);
bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
if (unlikely(ret)) {
/*
* If we error here, VFS caches are now inconsistent with btree
*/
set_bit(EI_INODE_ERROR, &inode->ei_flags);
goto err;
}
bch2_fs_inconsistent_on(!inode->v.i_size && inode->v.i_blocks &&
!bch2_journal_error(&c->journal), c,
"inode %lu truncated to 0 but i_blocks %llu (ondisk %lli)",
inode->v.i_ino, (u64) inode->v.i_blocks,
inode->ei_inode.bi_sectors);
ret = bch2_setattr_nonsize(idmap, inode, iattr);
err:
bch2_pagecache_block_put(inode);
return bch2_err_class(ret);
}
/* fallocate: */
static int inode_update_times_fn(struct btree_trans *trans,
struct bch_inode_info *inode,
struct bch_inode_unpacked *bi, void *p)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
bi->bi_mtime = bi->bi_ctime = bch2_current_time(c);
return 0;
}
static long bchfs_fpunch(struct bch_inode_info *inode, loff_t offset, loff_t len)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
u64 end = offset + len;
u64 block_start = round_up(offset, block_bytes(c));
u64 block_end = round_down(end, block_bytes(c));
bool truncated_last_page;
int ret = 0;
ret = bch2_truncate_folios(inode, offset, end);
if (unlikely(ret < 0))
goto err;
truncated_last_page = ret;
truncate_pagecache_range(&inode->v, offset, end - 1);
if (block_start < block_end) {
s64 i_sectors_delta = 0;
ret = bch2_fpunch(c, inode_inum(inode),
block_start >> 9, block_end >> 9,
&i_sectors_delta);
bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
}
mutex_lock(&inode->ei_update_lock);
if (end >= inode->v.i_size && !truncated_last_page) {
ret = bch2_write_inode_size(c, inode, inode->v.i_size,
ATTR_MTIME|ATTR_CTIME);
} else {
ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
ATTR_MTIME|ATTR_CTIME);
}
mutex_unlock(&inode->ei_update_lock);
err:
return ret;
}
static long bchfs_fcollapse_finsert(struct bch_inode_info *inode,
loff_t offset, loff_t len,
bool insert)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct address_space *mapping = inode->v.i_mapping;
s64 i_sectors_delta = 0;
int ret = 0;
if ((offset | len) & (block_bytes(c) - 1))
return -EINVAL;
if (insert) {
if (offset >= inode->v.i_size)
return -EINVAL;
} else {
if (offset + len >= inode->v.i_size)
return -EINVAL;
}
ret = bch2_write_invalidate_inode_pages_range(mapping, offset, LLONG_MAX);
if (ret)
return ret;
if (insert)
i_size_write(&inode->v, inode->v.i_size + len);
ret = bch2_fcollapse_finsert(c, inode_inum(inode), offset >> 9, len >> 9,
insert, &i_sectors_delta);
if (!ret && !insert)
i_size_write(&inode->v, inode->v.i_size - len);
bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta);
return ret;
}
static int __bchfs_fallocate(struct bch_inode_info *inode, int mode,
u64 start_sector, u64 end_sector)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct btree_trans *trans = bch2_trans_get(c);
struct btree_iter iter;
struct bpos end_pos = POS(inode->v.i_ino, end_sector);
struct bch_io_opts opts;
int ret = 0;
bch2_inode_opts_get(&opts, c, &inode->ei_inode);
bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
POS(inode->v.i_ino, start_sector),
BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
while (!ret && bkey_lt(iter.pos, end_pos)) {
s64 i_sectors_delta = 0;
struct quota_res quota_res = { 0 };
struct bkey_s_c k;
unsigned sectors;
bool is_allocation;
u64 hole_start, hole_end;
u32 snapshot;
bch2_trans_begin(trans);
ret = bch2_subvolume_get_snapshot(trans,
inode->ei_subvol, &snapshot);
if (ret)
goto bkey_err;
bch2_btree_iter_set_snapshot(&iter, snapshot);
k = bch2_btree_iter_peek_slot(&iter);
if ((ret = bkey_err(k)))
goto bkey_err;
hole_start = iter.pos.offset;
hole_end = bpos_min(k.k->p, end_pos).offset;
is_allocation = bkey_extent_is_allocation(k.k);
/* already reserved */
if (bkey_extent_is_reservation(k) &&
bch2_bkey_nr_ptrs_fully_allocated(k) >= opts.data_replicas) {
bch2_btree_iter_advance(&iter);
continue;
}
if (bkey_extent_is_data(k.k) &&
!(mode & FALLOC_FL_ZERO_RANGE)) {
bch2_btree_iter_advance(&iter);
continue;
}
if (!(mode & FALLOC_FL_ZERO_RANGE)) {
/*
* Lock ordering - can't be holding btree locks while
* blocking on a folio lock:
*/
if (bch2_clamp_data_hole(&inode->v,
&hole_start,
&hole_end,
opts.data_replicas, true))
ret = drop_locks_do(trans,
(bch2_clamp_data_hole(&inode->v,
&hole_start,
&hole_end,
opts.data_replicas, false), 0));
bch2_btree_iter_set_pos(&iter, POS(iter.pos.inode, hole_start));
if (ret)
goto bkey_err;
if (hole_start == hole_end)
continue;
}
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
sectors = hole_end - hole_start;
if (!is_allocation) {
ret = bch2_quota_reservation_add(c, inode,
&quota_res, sectors, true);
if (unlikely(ret))
goto bkey_err;
}
ret = bch2_extent_fallocate(trans, inode_inum(inode), &iter,
sectors, opts, &i_sectors_delta,
writepoint_hashed((unsigned long) current));
if (ret)
goto bkey_err;
bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
if (bch2_mark_pagecache_reserved(inode, &hole_start,
iter.pos.offset, true))
drop_locks_do(trans,
bch2_mark_pagecache_reserved(inode, &hole_start,
iter.pos.offset, false));
bkey_err:
bch2_quota_reservation_put(c, inode, &quota_res);
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
ret = 0;
}
if (bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)) {
struct quota_res quota_res = { 0 };
s64 i_sectors_delta = 0;
bch2_fpunch_at(trans, &iter, inode_inum(inode),
end_sector, &i_sectors_delta);
bch2_i_sectors_acct(c, inode, &quota_res, i_sectors_delta);
bch2_quota_reservation_put(c, inode, &quota_res);
}
bch2_trans_iter_exit(trans, &iter);
bch2_trans_put(trans);
return ret;
}
static long bchfs_fallocate(struct bch_inode_info *inode, int mode,
loff_t offset, loff_t len)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
u64 end = offset + len;
u64 block_start = round_down(offset, block_bytes(c));
u64 block_end = round_up(end, block_bytes(c));
bool truncated_last_page = false;
int ret, ret2 = 0;
if (!(mode & FALLOC_FL_KEEP_SIZE) && end > inode->v.i_size) {
ret = inode_newsize_ok(&inode->v, end);
if (ret)
return ret;
}
if (mode & FALLOC_FL_ZERO_RANGE) {
ret = bch2_truncate_folios(inode, offset, end);
if (unlikely(ret < 0))
return ret;
truncated_last_page = ret;
truncate_pagecache_range(&inode->v, offset, end - 1);
block_start = round_up(offset, block_bytes(c));
block_end = round_down(end, block_bytes(c));
}
ret = __bchfs_fallocate(inode, mode, block_start >> 9, block_end >> 9);
/*
* On -ENOSPC in ZERO_RANGE mode, we still want to do the inode update,
* so that the VFS cache i_size is consistent with the btree i_size:
*/
if (ret &&
!(bch2_err_matches(ret, ENOSPC) && (mode & FALLOC_FL_ZERO_RANGE)))
return ret;
if (mode & FALLOC_FL_KEEP_SIZE && end > inode->v.i_size)
end = inode->v.i_size;
if (end >= inode->v.i_size &&
(((mode & FALLOC_FL_ZERO_RANGE) && !truncated_last_page) ||
!(mode & FALLOC_FL_KEEP_SIZE))) {
spin_lock(&inode->v.i_lock);
i_size_write(&inode->v, end);
spin_unlock(&inode->v.i_lock);
mutex_lock(&inode->ei_update_lock);
ret2 = bch2_write_inode_size(c, inode, end, 0);
mutex_unlock(&inode->ei_update_lock);
}
return ret ?: ret2;
}
long bch2_fallocate_dispatch(struct file *file, int mode,
loff_t offset, loff_t len)
{
struct bch_inode_info *inode = file_bch_inode(file);
struct bch_fs *c = inode->v.i_sb->s_fs_info;
long ret;
if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_fallocate))
return -EROFS;
inode_lock(&inode->v);
inode_dio_wait(&inode->v);
bch2_pagecache_block_get(inode);
ret = file_modified(file);
if (ret)
goto err;
if (!(mode & ~(FALLOC_FL_KEEP_SIZE|FALLOC_FL_ZERO_RANGE)))
ret = bchfs_fallocate(inode, mode, offset, len);
else if (mode == (FALLOC_FL_PUNCH_HOLE|FALLOC_FL_KEEP_SIZE))
ret = bchfs_fpunch(inode, offset, len);
else if (mode == FALLOC_FL_INSERT_RANGE)
ret = bchfs_fcollapse_finsert(inode, offset, len, true);
else if (mode == FALLOC_FL_COLLAPSE_RANGE)
ret = bchfs_fcollapse_finsert(inode, offset, len, false);
else
ret = -EOPNOTSUPP;
err:
bch2_pagecache_block_put(inode);
inode_unlock(&inode->v);
bch2_write_ref_put(c, BCH_WRITE_REF_fallocate);
return bch2_err_class(ret);
}
/*
* Take a quota reservation for unallocated blocks in a given file range
* Does not check pagecache
*/
static int quota_reserve_range(struct bch_inode_info *inode,
struct quota_res *res,
u64 start, u64 end)
{
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct btree_trans *trans = bch2_trans_get(c);
struct btree_iter iter;
struct bkey_s_c k;
u32 snapshot;
u64 sectors = end - start;
u64 pos = start;
int ret;
retry:
bch2_trans_begin(trans);
ret = bch2_subvolume_get_snapshot(trans, inode->ei_subvol, &snapshot);
if (ret)
goto err;
bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
SPOS(inode->v.i_ino, pos, snapshot), 0);
while (!(ret = btree_trans_too_many_iters(trans)) &&
(k = bch2_btree_iter_peek_upto(&iter, POS(inode->v.i_ino, end - 1))).k &&
!(ret = bkey_err(k))) {
if (bkey_extent_is_allocation(k.k)) {
u64 s = min(end, k.k->p.offset) -
max(start, bkey_start_offset(k.k));
BUG_ON(s > sectors);
sectors -= s;
}
bch2_btree_iter_advance(&iter);
}
pos = iter.pos.offset;
bch2_trans_iter_exit(trans, &iter);
err:
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
goto retry;
bch2_trans_put(trans);
return ret ?: bch2_quota_reservation_add(c, inode, res, sectors, true);
}
loff_t bch2_remap_file_range(struct file *file_src, loff_t pos_src,
struct file *file_dst, loff_t pos_dst,
loff_t len, unsigned remap_flags)
{
struct bch_inode_info *src = file_bch_inode(file_src);
struct bch_inode_info *dst = file_bch_inode(file_dst);
struct bch_fs *c = src->v.i_sb->s_fs_info;
struct quota_res quota_res = { 0 };
s64 i_sectors_delta = 0;
u64 aligned_len;
loff_t ret = 0;
if (remap_flags & ~(REMAP_FILE_DEDUP|REMAP_FILE_ADVISORY))
return -EINVAL;
if (remap_flags & REMAP_FILE_DEDUP)
return -EOPNOTSUPP;
if ((pos_src & (block_bytes(c) - 1)) ||
(pos_dst & (block_bytes(c) - 1)))
return -EINVAL;
if (src == dst &&
abs(pos_src - pos_dst) < len)
return -EINVAL;
lock_two_nondirectories(&src->v, &dst->v);
bch2_lock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
inode_dio_wait(&src->v);
inode_dio_wait(&dst->v);
ret = generic_remap_file_range_prep(file_src, pos_src,
file_dst, pos_dst,
&len, remap_flags);
if (ret < 0 || len == 0)
goto err;
aligned_len = round_up((u64) len, block_bytes(c));
ret = bch2_write_invalidate_inode_pages_range(dst->v.i_mapping,
pos_dst, pos_dst + len - 1);
if (ret)
goto err;
ret = quota_reserve_range(dst, &quota_res, pos_dst >> 9,
(pos_dst + aligned_len) >> 9);
if (ret)
goto err;
file_update_time(file_dst);
bch2_mark_pagecache_unallocated(src, pos_src >> 9,
(pos_src + aligned_len) >> 9);
ret = bch2_remap_range(c,
inode_inum(dst), pos_dst >> 9,
inode_inum(src), pos_src >> 9,
aligned_len >> 9,
pos_dst + len, &i_sectors_delta);
if (ret < 0)
goto err;
/*
* due to alignment, we might have remapped slightly more than requsted
*/
ret = min((u64) ret << 9, (u64) len);
bch2_i_sectors_acct(c, dst, &quota_res, i_sectors_delta);
spin_lock(&dst->v.i_lock);
if (pos_dst + ret > dst->v.i_size)
i_size_write(&dst->v, pos_dst + ret);
spin_unlock(&dst->v.i_lock);
if ((file_dst->f_flags & (__O_SYNC | O_DSYNC)) ||
IS_SYNC(file_inode(file_dst)))
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
ret = bch2_flush_inode(c, dst);
err:
bch2_quota_reservation_put(c, dst, &quota_res);
bch2_unlock_inodes(INODE_PAGECACHE_BLOCK, src, dst);
unlock_two_nondirectories(&src->v, &dst->v);
return bch2_err_class(ret);
}
/* fseek: */
static loff_t bch2_seek_data(struct file *file, u64 offset)
{
struct bch_inode_info *inode = file_bch_inode(file);
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct btree_trans *trans;
struct btree_iter iter;
struct bkey_s_c k;
subvol_inum inum = inode_inum(inode);
u64 isize, next_data = MAX_LFS_FILESIZE;
u32 snapshot;
int ret;
isize = i_size_read(&inode->v);
if (offset >= isize)
return -ENXIO;
trans = bch2_trans_get(c);
retry:
bch2_trans_begin(trans);
ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
if (ret)
goto err;
for_each_btree_key_upto_norestart(trans, iter, BTREE_ID_extents,
SPOS(inode->v.i_ino, offset >> 9, snapshot),
POS(inode->v.i_ino, U64_MAX),
0, k, ret) {
if (bkey_extent_is_data(k.k)) {
next_data = max(offset, bkey_start_offset(k.k) << 9);
break;
} else if (k.k->p.offset >> 9 > isize)
break;
}
bch2_trans_iter_exit(trans, &iter);
err:
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
goto retry;
bch2_trans_put(trans);
if (ret)
return ret;
if (next_data > offset)
next_data = bch2_seek_pagecache_data(&inode->v,
offset, next_data, 0, false);
if (next_data >= isize)
return -ENXIO;
return vfs_setpos(file, next_data, MAX_LFS_FILESIZE);
}
static loff_t bch2_seek_hole(struct file *file, u64 offset)
{
struct bch_inode_info *inode = file_bch_inode(file);
struct bch_fs *c = inode->v.i_sb->s_fs_info;
struct btree_trans *trans;
struct btree_iter iter;
struct bkey_s_c k;
subvol_inum inum = inode_inum(inode);
u64 isize, next_hole = MAX_LFS_FILESIZE;
u32 snapshot;
int ret;
isize = i_size_read(&inode->v);
if (offset >= isize)
return -ENXIO;
trans = bch2_trans_get(c);
retry:
bch2_trans_begin(trans);
ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
if (ret)
goto err;
for_each_btree_key_norestart(trans, iter, BTREE_ID_extents,
SPOS(inode->v.i_ino, offset >> 9, snapshot),
BTREE_ITER_SLOTS, k, ret) {
if (k.k->p.inode != inode->v.i_ino) {
next_hole = bch2_seek_pagecache_hole(&inode->v,
offset, MAX_LFS_FILESIZE, 0, false);
break;
} else if (!bkey_extent_is_data(k.k)) {
next_hole = bch2_seek_pagecache_hole(&inode->v,
max(offset, bkey_start_offset(k.k) << 9),
k.k->p.offset << 9, 0, false);
if (next_hole < k.k->p.offset << 9)
break;
} else {
offset = max(offset, bkey_start_offset(k.k) << 9);
}
}
bch2_trans_iter_exit(trans, &iter);
err:
if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
goto retry;
bch2_trans_put(trans);
if (ret)
return ret;
if (next_hole > isize)
next_hole = isize;
return vfs_setpos(file, next_hole, MAX_LFS_FILESIZE);
}
loff_t bch2_llseek(struct file *file, loff_t offset, int whence)
{
loff_t ret;
switch (whence) {
case SEEK_SET:
case SEEK_CUR:
case SEEK_END:
ret = generic_file_llseek(file, offset, whence);
break;
case SEEK_DATA:
ret = bch2_seek_data(file, offset);
break;
case SEEK_HOLE:
ret = bch2_seek_hole(file, offset);
break;
default:
ret = -EINVAL;
break;
}
return bch2_err_class(ret);
}
void bch2_fs_fsio_exit(struct bch_fs *c)
{
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
bioset_exit(&c->nocow_flush_bioset);
}
int bch2_fs_fsio_init(struct bch_fs *c)
{
if (bioset_init(&c->nocow_flush_bioset,
bcachefs: Nocow support This adds support for nocow mode, where we do writes in-place when possible. Patch components: - New boolean filesystem and inode option, nocow: note that when nocow is enabled, data checksumming and compression are implicitly disabled - To prevent in-place writes from racing with data moves (data_update.c) or bucket reuse (i.e. a bucket being reused and re-allocated while a nocow write is in flight, we have a new locking mechanism. Buckets can be locked for either data update or data move, using a fixed size hash table of two_state_shared locks. We don't have any chaining, meaning updates and moves to different buckets that hash to the same lock will wait unnecessarily - we'll want to watch for this becoming an issue. - The allocator path also needs to check for in-place writes in flight to a given bucket before giving it out: thus we add another counter to bucket_alloc_state so we can track this. - Fsync now may need to issue cache flushes to block devices instead of flushing the journal. We add a device bitmask to bch_inode_info, ei_devs_need_flush, which tracks devices that need to have flushes issued - note that this will lead to unnecessary flushes when other codepaths have already issued flushes, we may want to replace this with a sequence number. - New nocow write path: look up extents, and if they're writable write to them - otherwise fall back to the normal COW write path. XXX: switch to sequence numbers instead of bitmask for devs needing journal flush XXX: ei_quota_lock being a mutex means bch2_nocow_write_done() needs to run in process context - see if we can improve this Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
2022-11-02 21:12:00 +00:00
1, offsetof(struct nocow_flush, bio), 0))
return -BCH_ERR_ENOMEM_nocow_flush_bioset_init;
return 0;
}
#endif /* NO_BCACHEFS_FS */