linux/fs/xfs/xfs_rtalloc.c

1480 lines
39 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/*
* Copyright (c) 2000-2005 Silicon Graphics, Inc.
* All Rights Reserved.
*/
#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_bit.h"
#include "xfs_mount.h"
#include "xfs_inode.h"
#include "xfs_bmap.h"
#include "xfs_bmap_btree.h"
#include "xfs_trans.h"
#include "xfs_trans_space.h"
#include "xfs_icache.h"
#include "xfs_rtalloc.h"
#include "xfs_sb.h"
#include "xfs_rtbitmap.h"
/*
* Read and return the summary information for a given extent size,
* bitmap block combination.
* Keeps track of a current summary block, so we don't keep reading
* it from the buffer cache.
*/
static int
xfs_rtget_summary(
xfs_mount_t *mp, /* file system mount structure */
xfs_trans_t *tp, /* transaction pointer */
int log, /* log2 of extent size */
xfs_fileoff_t bbno, /* bitmap block number */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
xfs_suminfo_t *sum) /* out: summary info for this block */
{
return xfs_rtmodify_summary_int(mp, tp, log, bbno, 0, rbpp, rsb, sum);
}
/*
* Return whether there are any free extents in the size range given
* by low and high, for the bitmap block bbno.
*/
STATIC int /* error */
xfs_rtany_summary(
xfs_mount_t *mp, /* file system mount structure */
xfs_trans_t *tp, /* transaction pointer */
int low, /* low log2 extent size */
int high, /* high log2 extent size */
xfs_fileoff_t bbno, /* bitmap block number */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
int *stat) /* out: any good extents here? */
{
int error; /* error value */
int log; /* loop counter, log2 of ext. size */
xfs_suminfo_t sum; /* summary data */
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
/* There are no extents at levels < m_rsum_cache[bbno]. */
if (mp->m_rsum_cache && low < mp->m_rsum_cache[bbno])
low = mp->m_rsum_cache[bbno];
/*
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
* Loop over logs of extent sizes.
*/
for (log = low; log <= high; log++) {
/*
* Get one summary datum.
*/
error = xfs_rtget_summary(mp, tp, log, bbno, rbpp, rsb, &sum);
if (error) {
return error;
}
/*
* If there are any, return success.
*/
if (sum) {
*stat = 1;
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
goto out;
}
}
/*
* Found nothing, return failure.
*/
*stat = 0;
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
out:
/* There were no extents at levels < log. */
if (mp->m_rsum_cache && log > mp->m_rsum_cache[bbno])
mp->m_rsum_cache[bbno] = log;
return 0;
}
/*
* Copy and transform the summary file, given the old and new
* parameters in the mount structures.
*/
STATIC int /* error */
xfs_rtcopy_summary(
xfs_mount_t *omp, /* old file system mount point */
xfs_mount_t *nmp, /* new file system mount point */
xfs_trans_t *tp) /* transaction pointer */
{
xfs_fileoff_t bbno; /* bitmap block number */
struct xfs_buf *bp; /* summary buffer */
int error; /* error return value */
int log; /* summary level number (log length) */
xfs_suminfo_t sum; /* summary data */
xfs_fileoff_t sumbno; /* summary block number */
bp = NULL;
for (log = omp->m_rsumlevels - 1; log >= 0; log--) {
for (bbno = omp->m_sb.sb_rbmblocks - 1;
(xfs_srtblock_t)bbno >= 0;
bbno--) {
error = xfs_rtget_summary(omp, tp, log, bbno, &bp,
&sumbno, &sum);
if (error)
return error;
if (sum == 0)
continue;
error = xfs_rtmodify_summary(omp, tp, log, bbno, -sum,
&bp, &sumbno);
if (error)
return error;
error = xfs_rtmodify_summary(nmp, tp, log, bbno, sum,
&bp, &sumbno);
if (error)
return error;
ASSERT(sum > 0);
}
}
return 0;
}
/*
* Mark an extent specified by start and len allocated.
* Updates all the summary information as well as the bitmap.
*/
STATIC int /* error */
xfs_rtallocate_range(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxnum_t start, /* start rtext to allocate */
xfs_rtxlen_t len, /* length to allocate */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb) /* in/out: summary block number */
{
xfs_rtxnum_t end; /* end of the allocated rtext */
int error; /* error value */
xfs_rtxnum_t postblock = 0; /* first rtext allocated > end */
xfs_rtxnum_t preblock = 0; /* first rtext allocated < start */
end = start + len - 1;
/*
* Assume we're allocating out of the middle of a free extent.
* We need to find the beginning and end of the extent so we can
* properly update the summary.
*/
error = xfs_rtfind_back(mp, tp, start, 0, &preblock);
if (error) {
return error;
}
/*
* Find the next allocated block (end of free extent).
*/
error = xfs_rtfind_forw(mp, tp, end, mp->m_sb.sb_rextents - 1,
&postblock);
if (error) {
return error;
}
/*
* Decrement the summary information corresponding to the entire
* (old) free extent.
*/
error = xfs_rtmodify_summary(mp, tp,
XFS_RTBLOCKLOG(postblock + 1 - preblock),
xfs_rtx_to_rbmblock(mp, preblock), -1, rbpp, rsb);
if (error) {
return error;
}
/*
* If there are blocks not being allocated at the front of the
* old extent, add summary data for them to be free.
*/
if (preblock < start) {
error = xfs_rtmodify_summary(mp, tp,
XFS_RTBLOCKLOG(start - preblock),
xfs_rtx_to_rbmblock(mp, preblock), 1, rbpp, rsb);
if (error) {
return error;
}
}
/*
* If there are blocks not being allocated at the end of the
* old extent, add summary data for them to be free.
*/
if (postblock > end) {
error = xfs_rtmodify_summary(mp, tp,
XFS_RTBLOCKLOG(postblock - end),
xfs_rtx_to_rbmblock(mp, end + 1), 1, rbpp, rsb);
if (error) {
return error;
}
}
/*
* Modify the bitmap to mark this extent allocated.
*/
error = xfs_rtmodify_range(mp, tp, start, len, 0);
return error;
}
/*
* Make sure we don't run off the end of the rt volume. Be careful that
* adjusting maxlen downwards doesn't cause us to fail the alignment checks.
*/
static inline xfs_rtxlen_t
xfs_rtallocate_clamp_len(
struct xfs_mount *mp,
xfs_rtxnum_t startrtx,
xfs_rtxlen_t rtxlen,
xfs_rtxlen_t prod)
{
xfs_rtxlen_t ret;
ret = min(mp->m_sb.sb_rextents, startrtx + rtxlen) - startrtx;
return rounddown(ret, prod);
}
/*
* Attempt to allocate an extent minlen<=len<=maxlen starting from
* bitmap block bbno. If we don't get maxlen then use prod to trim
* the length, if given. Returns error; returns starting block in *rtx.
* The lengths are all in rtextents.
*/
STATIC int /* error */
xfs_rtallocate_extent_block(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_fileoff_t bbno, /* bitmap block number */
xfs_rtxlen_t minlen, /* minimum length to allocate */
xfs_rtxlen_t maxlen, /* maximum length to allocate */
xfs_rtxlen_t *len, /* out: actual length allocated */
xfs_rtxnum_t *nextp, /* out: next rtext to try */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
xfs_rtxlen_t prod, /* extent product factor */
xfs_rtxnum_t *rtx) /* out: start rtext allocated */
{
xfs_rtxnum_t besti; /* best rtext found so far */
xfs_rtxnum_t bestlen; /* best length found so far */
xfs_rtxnum_t end; /* last rtext in chunk */
int error; /* error value */
xfs_rtxnum_t i; /* current rtext trying */
xfs_rtxnum_t next; /* next rtext to try */
int stat; /* status from internal calls */
/*
* Loop over all the extents starting in this bitmap block,
* looking for one that's long enough.
*/
for (i = xfs_rbmblock_to_rtx(mp, bbno), besti = -1, bestlen = 0,
end = xfs_rbmblock_to_rtx(mp, bbno + 1) - 1;
i <= end;
i++) {
/* Make sure we don't scan off the end of the rt volume. */
maxlen = xfs_rtallocate_clamp_len(mp, i, maxlen, prod);
/*
* See if there's a free extent of maxlen starting at i.
* If it's not so then next will contain the first non-free.
*/
error = xfs_rtcheck_range(mp, tp, i, maxlen, 1, &next, &stat);
if (error) {
return error;
}
if (stat) {
/*
* i for maxlen is all free, allocate and return that.
*/
error = xfs_rtallocate_range(mp, tp, i, maxlen, rbpp,
rsb);
if (error) {
return error;
}
*len = maxlen;
*rtx = i;
return 0;
}
/*
* In the case where we have a variable-sized allocation
* request, figure out how big this free piece is,
* and if it's big enough for the minimum, and the best
* so far, remember it.
*/
if (minlen < maxlen) {
xfs_rtxnum_t thislen; /* this extent size */
thislen = next - i;
if (thislen >= minlen && thislen > bestlen) {
besti = i;
bestlen = thislen;
}
}
/*
* If not done yet, find the start of the next free space.
*/
if (next < end) {
error = xfs_rtfind_forw(mp, tp, next, end, &i);
if (error) {
return error;
}
} else
break;
}
/*
* Searched the whole thing & didn't find a maxlen free extent.
*/
if (minlen < maxlen && besti != -1) {
xfs_rtxlen_t p; /* amount to trim length by */
/*
* If size should be a multiple of prod, make that so.
*/
if (prod > 1) {
div_u64_rem(bestlen, prod, &p);
if (p)
bestlen -= p;
}
/*
* Allocate besti for bestlen & return that.
*/
error = xfs_rtallocate_range(mp, tp, besti, bestlen, rbpp, rsb);
if (error) {
return error;
}
*len = bestlen;
*rtx = besti;
return 0;
}
/*
* Allocation failed. Set *nextp to the next block to try.
*/
*nextp = next;
*rtx = NULLRTEXTNO;
return 0;
}
/*
* Allocate an extent of length minlen<=len<=maxlen, starting at block
* bno. If we don't get maxlen then use prod to trim the length, if given.
* Returns error; returns starting block in *rtx.
* The lengths are all in rtextents.
*/
STATIC int /* error */
xfs_rtallocate_extent_exact(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxnum_t start, /* starting rtext number to allocate */
xfs_rtxlen_t minlen, /* minimum length to allocate */
xfs_rtxlen_t maxlen, /* maximum length to allocate */
xfs_rtxlen_t *len, /* out: actual length allocated */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
xfs_rtxlen_t prod, /* extent product factor */
xfs_rtxnum_t *rtx) /* out: start rtext allocated */
{
int error; /* error value */
xfs_rtxlen_t i; /* extent length trimmed due to prod */
int isfree; /* extent is free */
xfs_rtxnum_t next; /* next rtext to try (dummy) */
ASSERT(minlen % prod == 0);
ASSERT(maxlen % prod == 0);
/*
* Check if the range in question (for maxlen) is free.
*/
error = xfs_rtcheck_range(mp, tp, start, maxlen, 1, &next, &isfree);
if (error) {
return error;
}
if (isfree) {
/*
* If it is, allocate it and return success.
*/
error = xfs_rtallocate_range(mp, tp, start, maxlen, rbpp, rsb);
if (error) {
return error;
}
*len = maxlen;
*rtx = start;
return 0;
}
/*
* If not, allocate what there is, if it's at least minlen.
*/
maxlen = next - start;
if (maxlen < minlen) {
/*
* Failed, return failure status.
*/
*rtx = NULLRTEXTNO;
return 0;
}
/*
* Trim off tail of extent, if prod is specified.
*/
if (prod > 1 && (i = maxlen % prod)) {
maxlen -= i;
if (maxlen < minlen) {
/*
* Now we can't do it, return failure status.
*/
*rtx = NULLRTEXTNO;
return 0;
}
}
/*
* Allocate what we can and return it.
*/
error = xfs_rtallocate_range(mp, tp, start, maxlen, rbpp, rsb);
if (error) {
return error;
}
*len = maxlen;
*rtx = start;
return 0;
}
/*
* Allocate an extent of length minlen<=len<=maxlen, starting as near
* to start as possible. If we don't get maxlen then use prod to trim
* the length, if given. The lengths are all in rtextents.
*/
STATIC int /* error */
xfs_rtallocate_extent_near(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxnum_t start, /* starting rtext number to allocate */
xfs_rtxlen_t minlen, /* minimum length to allocate */
xfs_rtxlen_t maxlen, /* maximum length to allocate */
xfs_rtxlen_t *len, /* out: actual length allocated */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
xfs_rtxlen_t prod, /* extent product factor */
xfs_rtxnum_t *rtx) /* out: start rtext allocated */
{
int any; /* any useful extents from summary */
xfs_fileoff_t bbno; /* bitmap block number */
int error; /* error value */
int i; /* bitmap block offset (loop control) */
int j; /* secondary loop control */
int log2len; /* log2 of minlen */
xfs_rtxnum_t n; /* next rtext to try */
xfs_rtxnum_t r; /* result rtext */
ASSERT(minlen % prod == 0);
ASSERT(maxlen % prod == 0);
/*
* If the block number given is off the end, silently set it to
* the last block.
*/
if (start >= mp->m_sb.sb_rextents)
start = mp->m_sb.sb_rextents - 1;
/* Make sure we don't run off the end of the rt volume. */
maxlen = xfs_rtallocate_clamp_len(mp, start, maxlen, prod);
if (maxlen < minlen) {
*rtx = NULLRTEXTNO;
return 0;
}
/*
* Try the exact allocation first.
*/
error = xfs_rtallocate_extent_exact(mp, tp, start, minlen, maxlen, len,
rbpp, rsb, prod, &r);
if (error) {
return error;
}
/*
* If the exact allocation worked, return that.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
bbno = xfs_rtx_to_rbmblock(mp, start);
i = 0;
ASSERT(minlen != 0);
log2len = xfs_highbit32(minlen);
/*
* Loop over all bitmap blocks (bbno + i is current block).
*/
for (;;) {
/*
* Get summary information of extents of all useful levels
* starting in this bitmap block.
*/
error = xfs_rtany_summary(mp, tp, log2len, mp->m_rsumlevels - 1,
bbno + i, rbpp, rsb, &any);
if (error) {
return error;
}
/*
* If there are any useful extents starting here, try
* allocating one.
*/
if (any) {
/*
* On the positive side of the starting location.
*/
if (i >= 0) {
/*
* Try to allocate an extent starting in
* this block.
*/
error = xfs_rtallocate_extent_block(mp, tp,
bbno + i, minlen, maxlen, len, &n, rbpp,
rsb, prod, &r);
if (error) {
return error;
}
/*
* If it worked, return it.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
}
/*
* On the negative side of the starting location.
*/
else { /* i < 0 */
/*
* Loop backwards through the bitmap blocks from
* the starting point-1 up to where we are now.
* There should be an extent which ends in this
* bitmap block and is long enough.
*/
for (j = -1; j > i; j--) {
/*
* Grab the summary information for
* this bitmap block.
*/
error = xfs_rtany_summary(mp, tp,
log2len, mp->m_rsumlevels - 1,
bbno + j, rbpp, rsb, &any);
if (error) {
return error;
}
/*
* If there's no extent given in the
* summary that means the extent we
* found must carry over from an
* earlier block. If there is an
* extent given, we've already tried
* that allocation, don't do it again.
*/
if (any)
continue;
error = xfs_rtallocate_extent_block(mp,
tp, bbno + j, minlen, maxlen,
len, &n, rbpp, rsb, prod, &r);
if (error) {
return error;
}
/*
* If it works, return the extent.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
}
/*
* There weren't intervening bitmap blocks
* with a long enough extent, or the
* allocation didn't work for some reason
* (i.e. it's a little * too short).
* Try to allocate from the summary block
* that we found.
*/
error = xfs_rtallocate_extent_block(mp, tp,
bbno + i, minlen, maxlen, len, &n, rbpp,
rsb, prod, &r);
if (error) {
return error;
}
/*
* If it works, return the extent.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
}
}
/*
* Loop control. If we were on the positive side, and there's
* still more blocks on the negative side, go there.
*/
if (i > 0 && (int)bbno - i >= 0)
i = -i;
/*
* If positive, and no more negative, but there are more
* positive, go there.
*/
else if (i > 0 && (int)bbno + i < mp->m_sb.sb_rbmblocks - 1)
i++;
/*
* If negative or 0 (just started), and there are positive
* blocks to go, go there. The 0 case moves to block 1.
*/
else if (i <= 0 && (int)bbno - i < mp->m_sb.sb_rbmblocks - 1)
i = 1 - i;
/*
* If negative or 0 and there are more negative blocks,
* go there.
*/
else if (i <= 0 && (int)bbno + i > 0)
i--;
/*
* Must be done. Return failure.
*/
else
break;
}
*rtx = NULLRTEXTNO;
return 0;
}
/*
* Allocate an extent of length minlen<=len<=maxlen, with no position
* specified. If we don't get maxlen then use prod to trim
* the length, if given. The lengths are all in rtextents.
*/
STATIC int /* error */
xfs_rtallocate_extent_size(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxlen_t minlen, /* minimum length to allocate */
xfs_rtxlen_t maxlen, /* maximum length to allocate */
xfs_rtxlen_t *len, /* out: actual length allocated */
struct xfs_buf **rbpp, /* in/out: summary block buffer */
xfs_fileoff_t *rsb, /* in/out: summary block number */
xfs_rtxlen_t prod, /* extent product factor */
xfs_rtxnum_t *rtx) /* out: start rtext allocated */
{
int error; /* error value */
xfs_fileoff_t i; /* bitmap block number */
int l; /* level number (loop control) */
xfs_rtxnum_t n; /* next rtext to be tried */
xfs_rtxnum_t r; /* result rtext number */
xfs_suminfo_t sum; /* summary information for extents */
ASSERT(minlen % prod == 0);
ASSERT(maxlen % prod == 0);
ASSERT(maxlen != 0);
/*
* Loop over all the levels starting with maxlen.
* At each level, look at all the bitmap blocks, to see if there
* are extents starting there that are long enough (>= maxlen).
* Note, only on the initial level can the allocation fail if
* the summary says there's an extent.
*/
for (l = xfs_highbit32(maxlen); l < mp->m_rsumlevels; l++) {
/*
* Loop over all the bitmap blocks.
*/
for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
/*
* Get the summary for this level/block.
*/
error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
&sum);
if (error) {
return error;
}
/*
* Nothing there, on to the next block.
*/
if (!sum)
continue;
/*
* Try allocating the extent.
*/
error = xfs_rtallocate_extent_block(mp, tp, i, maxlen,
maxlen, len, &n, rbpp, rsb, prod, &r);
if (error) {
return error;
}
/*
* If it worked, return that.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
/*
* If the "next block to try" returned from the
* allocator is beyond the next bitmap block,
* skip to that bitmap block.
*/
if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
i = xfs_rtx_to_rbmblock(mp, n) - 1;
}
}
/*
* Didn't find any maxlen blocks. Try smaller ones, unless
* we're asking for a fixed size extent.
*/
if (minlen > --maxlen) {
*rtx = NULLRTEXTNO;
return 0;
}
ASSERT(minlen != 0);
ASSERT(maxlen != 0);
/*
* Loop over sizes, from maxlen down to minlen.
* This time, when we do the allocations, allow smaller ones
* to succeed.
*/
for (l = xfs_highbit32(maxlen); l >= xfs_highbit32(minlen); l--) {
/*
* Loop over all the bitmap blocks, try an allocation
* starting in that block.
*/
for (i = 0; i < mp->m_sb.sb_rbmblocks; i++) {
/*
* Get the summary information for this level/block.
*/
error = xfs_rtget_summary(mp, tp, l, i, rbpp, rsb,
&sum);
if (error) {
return error;
}
/*
* If nothing there, go on to next.
*/
if (!sum)
continue;
/*
* Try the allocation. Make sure the specified
* minlen/maxlen are in the possible range for
* this summary level.
*/
error = xfs_rtallocate_extent_block(mp, tp, i,
XFS_RTMAX(minlen, 1 << l),
XFS_RTMIN(maxlen, (1 << (l + 1)) - 1),
len, &n, rbpp, rsb, prod, &r);
if (error) {
return error;
}
/*
* If it worked, return that extent.
*/
if (r != NULLRTEXTNO) {
*rtx = r;
return 0;
}
/*
* If the "next block to try" returned from the
* allocator is beyond the next bitmap block,
* skip to that bitmap block.
*/
if (xfs_rtx_to_rbmblock(mp, n) > i + 1)
i = xfs_rtx_to_rbmblock(mp, n) - 1;
}
}
/*
* Got nothing, return failure.
*/
*rtx = NULLRTEXTNO;
return 0;
}
/*
* Allocate space to the bitmap or summary file, and zero it, for growfs.
*/
STATIC int
xfs_growfs_rt_alloc(
struct xfs_mount *mp, /* file system mount point */
xfs_extlen_t oblocks, /* old count of blocks */
xfs_extlen_t nblocks, /* new count of blocks */
struct xfs_inode *ip) /* inode (bitmap/summary) */
{
xfs_fileoff_t bno; /* block number in file */
struct xfs_buf *bp; /* temporary buffer for zeroing */
xfs_daddr_t d; /* disk block address */
int error; /* error return value */
xfs_fsblock_t fsbno; /* filesystem block for bno */
struct xfs_bmbt_irec map; /* block map output */
int nmap; /* number of block maps */
int resblks; /* space reservation */
xfs: Set xfs_buf type flag when growing summary/bitmap files The following sequence of commands, mkfs.xfs -f -m reflink=0 -r rtdev=/dev/loop1,size=10M /dev/loop0 mount -o rtdev=/dev/loop1 /dev/loop0 /mnt xfs_growfs /mnt ... causes the following call trace to be printed on the console, XFS: Assertion failed: (bip->bli_flags & XFS_BLI_STALE) || (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF), file: fs/xfs/xfs_buf_item.c, line: 331 Call Trace: xfs_buf_item_format+0x632/0x680 ? kmem_alloc_large+0x29/0x90 ? kmem_alloc+0x70/0x120 ? xfs_log_commit_cil+0x132/0x940 xfs_log_commit_cil+0x26f/0x940 ? xfs_buf_item_init+0x1ad/0x240 ? xfs_growfs_rt_alloc+0x1fc/0x280 __xfs_trans_commit+0xac/0x370 xfs_growfs_rt_alloc+0x1fc/0x280 xfs_growfs_rt+0x1a0/0x5e0 xfs_file_ioctl+0x3fd/0xc70 ? selinux_file_ioctl+0x174/0x220 ksys_ioctl+0x87/0xc0 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x3e/0x70 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This occurs because the buffer being formatted has the value of XFS_BLFT_UNKNOWN_BUF assigned to the 'type' subfield of bip->bli_formats->blf_flags. This commit fixes the issue by assigning one of XFS_BLFT_RTSUMMARY_BUF and XFS_BLFT_RTBITMAP_BUF to the 'type' subfield of bip->bli_formats->blf_flags before committing the corresponding transaction. Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-09-16 03:50:42 +00:00
enum xfs_blft buf_type;
struct xfs_trans *tp;
xfs: Set xfs_buf type flag when growing summary/bitmap files The following sequence of commands, mkfs.xfs -f -m reflink=0 -r rtdev=/dev/loop1,size=10M /dev/loop0 mount -o rtdev=/dev/loop1 /dev/loop0 /mnt xfs_growfs /mnt ... causes the following call trace to be printed on the console, XFS: Assertion failed: (bip->bli_flags & XFS_BLI_STALE) || (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF), file: fs/xfs/xfs_buf_item.c, line: 331 Call Trace: xfs_buf_item_format+0x632/0x680 ? kmem_alloc_large+0x29/0x90 ? kmem_alloc+0x70/0x120 ? xfs_log_commit_cil+0x132/0x940 xfs_log_commit_cil+0x26f/0x940 ? xfs_buf_item_init+0x1ad/0x240 ? xfs_growfs_rt_alloc+0x1fc/0x280 __xfs_trans_commit+0xac/0x370 xfs_growfs_rt_alloc+0x1fc/0x280 xfs_growfs_rt+0x1a0/0x5e0 xfs_file_ioctl+0x3fd/0xc70 ? selinux_file_ioctl+0x174/0x220 ksys_ioctl+0x87/0xc0 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x3e/0x70 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This occurs because the buffer being formatted has the value of XFS_BLFT_UNKNOWN_BUF assigned to the 'type' subfield of bip->bli_formats->blf_flags. This commit fixes the issue by assigning one of XFS_BLFT_RTSUMMARY_BUF and XFS_BLFT_RTBITMAP_BUF to the 'type' subfield of bip->bli_formats->blf_flags before committing the corresponding transaction. Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-09-16 03:50:42 +00:00
if (ip == mp->m_rsumip)
buf_type = XFS_BLFT_RTSUMMARY_BUF;
else
buf_type = XFS_BLFT_RTBITMAP_BUF;
/*
* Allocate space to the file, as necessary.
*/
while (oblocks < nblocks) {
resblks = XFS_GROWFSRT_SPACE_RES(mp, nblocks - oblocks);
/*
* Reserve space & log for one extent added to the file.
*/
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtalloc, resblks,
0, 0, &tp);
if (error)
return error;
/*
* Lock the inode.
*/
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
XFS_IEXT_ADD_NOSPLIT_CNT);
if (error == -EFBIG)
error = xfs_iext_count_upgrade(tp, ip,
XFS_IEXT_ADD_NOSPLIT_CNT);
if (error)
goto out_trans_cancel;
/*
* Allocate blocks to the bitmap file.
*/
nmap = 1;
error = xfs_bmapi_write(tp, ip, oblocks, nblocks - oblocks,
xfs: don't set bmapi total block req where minleft is xfs_bmapi_write() takes a total block requirement parameter that is passed down to the block allocation code and is used to specify the total block requirement of the associated transaction. This is used to try and select an AG that can not only satisfy the requested extent allocation, but can also accommodate subsequent allocations that might be required to complete the transaction. For example, additional bmbt block allocations may be required on insertion of the resulting extent to an inode data fork. While it's important for callers to calculate and reserve such extra blocks in the transaction, it is not necessary to pass the total value to xfs_bmapi_write() in all cases. The latter automatically sets minleft to ensure that sufficient free blocks remain after the allocation attempt to expand the format of the associated inode (i.e., such as extent to btree conversion, btree splits, etc). Therefore, any callers that pass a total block requirement of the bmap mapping length plus worst case bmbt expansion essentially specify the additional reservation requirement twice. These callers can pass a total of zero to rely on the bmapi minleft policy. Beyond being superfluous, the primary motivation for this change is that the total reservation logic in the bmbt code is dubious in scenarios where minlen < maxlen and a maxlen extent cannot be allocated (which is more common for data extent allocations where contiguity is not required). The total value is based on maxlen in the xfs_bmapi_write() caller. If the bmbt code falls back to an allocation between minlen and maxlen, that allocation will not succeed until total is reset to minlen, which essentially throws away any additional reservation included in total by the caller. In addition, the total value is not reset until after alignment is dropped, which means that such callers drop alignment far too aggressively than necessary. Update all callers of xfs_bmapi_write() that pass a total block value of the mapping length plus bmbt reservation to instead pass zero and rely on xfs_bmapi_minleft() to enforce the bmbt reservation requirement. This trades off slightly less conservative AG selection for the ability to preserve alignment in more scenarios. xfs_bmapi_write() callers that incorporate unrelated or additional reservations in total beyond what is already included in minleft must continue to use the former. Signed-off-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>
2019-10-21 16:26:48 +00:00
XFS_BMAPI_METADATA, 0, &map, &nmap);
if (!error && nmap < 1)
error = -ENOSPC;
if (error)
goto out_trans_cancel;
/*
* Free any blocks freed up in the transaction, then commit.
*/
error = xfs_trans_commit(tp);
if (error)
return error;
/*
* Now we need to clear the allocated blocks.
* Do this one block per transaction, to keep it simple.
*/
for (bno = map.br_startoff, fsbno = map.br_startblock;
bno < map.br_startoff + map.br_blockcount;
bno++, fsbno++) {
/*
* Reserve log for one block zeroing.
*/
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtzero,
0, 0, 0, &tp);
if (error)
return error;
/*
* Lock the bitmap inode.
*/
xfs_ilock(ip, XFS_ILOCK_EXCL);
xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
/*
* Get a buffer for the block.
*/
d = XFS_FSB_TO_DADDR(mp, fsbno);
error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
mp->m_bsize, 0, &bp);
if (error)
goto out_trans_cancel;
xfs: Set xfs_buf type flag when growing summary/bitmap files The following sequence of commands, mkfs.xfs -f -m reflink=0 -r rtdev=/dev/loop1,size=10M /dev/loop0 mount -o rtdev=/dev/loop1 /dev/loop0 /mnt xfs_growfs /mnt ... causes the following call trace to be printed on the console, XFS: Assertion failed: (bip->bli_flags & XFS_BLI_STALE) || (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF), file: fs/xfs/xfs_buf_item.c, line: 331 Call Trace: xfs_buf_item_format+0x632/0x680 ? kmem_alloc_large+0x29/0x90 ? kmem_alloc+0x70/0x120 ? xfs_log_commit_cil+0x132/0x940 xfs_log_commit_cil+0x26f/0x940 ? xfs_buf_item_init+0x1ad/0x240 ? xfs_growfs_rt_alloc+0x1fc/0x280 __xfs_trans_commit+0xac/0x370 xfs_growfs_rt_alloc+0x1fc/0x280 xfs_growfs_rt+0x1a0/0x5e0 xfs_file_ioctl+0x3fd/0xc70 ? selinux_file_ioctl+0x174/0x220 ksys_ioctl+0x87/0xc0 __x64_sys_ioctl+0x16/0x20 do_syscall_64+0x3e/0x70 entry_SYSCALL_64_after_hwframe+0x44/0xa9 This occurs because the buffer being formatted has the value of XFS_BLFT_UNKNOWN_BUF assigned to the 'type' subfield of bip->bli_formats->blf_flags. This commit fixes the issue by assigning one of XFS_BLFT_RTSUMMARY_BUF and XFS_BLFT_RTBITMAP_BUF to the 'type' subfield of bip->bli_formats->blf_flags before committing the corresponding transaction. Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Chandan Babu R <chandanrlinux@gmail.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2020-09-16 03:50:42 +00:00
xfs_trans_buf_set_type(tp, bp, buf_type);
xfs: Set xfs_buf's b_ops member when zeroing bitmap/summary files In xfs_growfs_rt(), we enlarge bitmap and summary files by allocating new blocks for both files. For each of the new blocks allocated, we allocate an xfs_buf, zero the payload, log the contents and commit the transaction. Hence these buffers will eventually find themselves appended to list at xfs_ail->ail_buf_list. Later, xfs_growfs_rt() loops across all of the new blocks belonging to the bitmap inode to set the bitmap values to 1. In doing so, it allocates a new transaction and invokes the following sequence of functions, - xfs_rtfree_range() - xfs_rtmodify_range() - xfs_rtbuf_get() We pass '&xfs_rtbuf_ops' as the ops pointer to xfs_trans_read_buf(). - xfs_trans_read_buf() We find the xfs_buf of interest in per-ag hash table, invoke xfs_buf_reverify() which ends up assigning '&xfs_rtbuf_ops' to xfs_buf->b_ops. On the other hand, if xfs_growfs_rt_alloc() had allocated a few blocks for the bitmap inode and returned with an error, all the xfs_bufs corresponding to the new bitmap blocks that have been allocated would continue to be on xfs_ail->ail_buf_list list without ever having a non-NULL value assigned to their b_ops members. An AIL flush operation would then trigger the following warning message to be printed on the console, XFS (loop0): _xfs_buf_ioapply: no buf ops on daddr 0x58 len 8 00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000040: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000050: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 00000070: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ CPU: 3 PID: 449 Comm: xfsaild/loop0 Not tainted 5.8.0-rc4-chandan-00038-g4d8c2b9de9ab-dirty #37 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014 Call Trace: dump_stack+0x57/0x70 _xfs_buf_ioapply+0x37c/0x3b0 ? xfs_rw_bdev+0x1e0/0x1e0 ? xfs_buf_delwri_submit_buffers+0xd4/0x210 __xfs_buf_submit+0x6d/0x1f0 xfs_buf_delwri_submit_buffers+0xd4/0x210 xfsaild+0x2c8/0x9e0 ? __switch_to_asm+0x42/0x70 ? xfs_trans_ail_cursor_first+0x80/0x80 kthread+0xfe/0x140 ? kthread_park+0x90/0x90 ret_from_fork+0x22/0x30 This message indicates that the xfs_buf had its b_ops member set to NULL. This commit fixes the issue by assigning "&xfs_rtbuf_ops" to b_ops member of each of the xfs_bufs logged by xfs_growfs_rt_alloc(). 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>
2020-09-17 18:12:08 +00:00
bp->b_ops = &xfs_rtbuf_ops;
memset(bp->b_addr, 0, mp->m_sb.sb_blocksize);
xfs_trans_log_buf(tp, bp, 0, mp->m_sb.sb_blocksize - 1);
/*
* Commit the transaction.
*/
error = xfs_trans_commit(tp);
if (error)
return error;
}
/*
* Go on to the next extent, if any.
*/
oblocks = map.br_startoff + map.br_blockcount;
}
return 0;
out_trans_cancel:
xfs_trans_cancel(tp);
return error;
}
static void
xfs_alloc_rsum_cache(
xfs_mount_t *mp, /* file system mount structure */
xfs_extlen_t rbmblocks) /* number of rt bitmap blocks */
{
/*
* The rsum cache is initialized to all zeroes, which is trivially a
* lower bound on the minimum level with any free extents. We can
* continue without the cache if it couldn't be allocated.
*/
mp->m_rsum_cache = kvzalloc(rbmblocks, GFP_KERNEL);
if (!mp->m_rsum_cache)
xfs_warn(mp, "could not allocate realtime summary cache");
}
/*
* Visible (exported) functions.
*/
/*
* Grow the realtime area of the filesystem.
*/
int
xfs_growfs_rt(
xfs_mount_t *mp, /* mount point for filesystem */
xfs_growfs_rt_t *in) /* growfs rt input struct */
{
xfs_fileoff_t bmbno; /* bitmap block number */
struct xfs_buf *bp; /* temporary buffer */
int error; /* error return value */
xfs_mount_t *nmp; /* new (fake) mount structure */
xfs_rfsblock_t nrblocks; /* new number of realtime blocks */
xfs_extlen_t nrbmblocks; /* new number of rt bitmap blocks */
xfs_rtxnum_t nrextents; /* new number of realtime extents */
uint8_t nrextslog; /* new log2 of sb_rextents */
xfs_extlen_t nrsumblocks; /* new number of summary blocks */
uint nrsumlevels; /* new rt summary levels */
uint nrsumsize; /* new size of rt summary, bytes */
xfs_sb_t *nsbp; /* new superblock */
xfs_extlen_t rbmblocks; /* current number of rt bitmap blocks */
xfs_extlen_t rsumblocks; /* current number of rt summary blks */
xfs_sb_t *sbp; /* old superblock */
xfs_fileoff_t sumbno; /* summary block number */
uint8_t *rsum_cache; /* old summary cache */
sbp = &mp->m_sb;
if (!capable(CAP_SYS_ADMIN))
return -EPERM;
/* Needs to have been mounted with an rt device. */
if (!XFS_IS_REALTIME_MOUNT(mp))
return -EINVAL;
/*
* Mount should fail if the rt bitmap/summary files don't load, but
* we'll check anyway.
*/
if (!mp->m_rbmip || !mp->m_rsumip)
return -EINVAL;
/* Shrink not supported. */
if (in->newblocks <= sbp->sb_rblocks)
return -EINVAL;
/* Can only change rt extent size when adding rt volume. */
if (sbp->sb_rblocks > 0 && in->extsize != sbp->sb_rextsize)
return -EINVAL;
/* Range check the extent size. */
if (XFS_FSB_TO_B(mp, in->extsize) > XFS_MAX_RTEXTSIZE ||
XFS_FSB_TO_B(mp, in->extsize) < XFS_MIN_RTEXTSIZE)
return -EINVAL;
/* Unsupported realtime features. */
if (xfs_has_rmapbt(mp) || xfs_has_reflink(mp) || xfs_has_quota(mp))
return -EOPNOTSUPP;
nrblocks = in->newblocks;
error = xfs_sb_validate_fsb_count(sbp, nrblocks);
if (error)
return error;
/*
* Read in the last block of the device, make sure it exists.
*/
error = xfs_buf_read_uncached(mp->m_rtdev_targp,
XFS_FSB_TO_BB(mp, nrblocks - 1),
XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
if (error)
return error;
xfs_buf_relse(bp);
/*
* Calculate new parameters. These are the final values to be reached.
*/
nrextents = nrblocks;
do_div(nrextents, in->extsize);
nrbmblocks = howmany_64(nrextents, NBBY * sbp->sb_blocksize);
nrextslog = xfs_highbit32(nrextents);
nrsumlevels = nrextslog + 1;
nrsumsize = (uint)sizeof(xfs_suminfo_t) * nrsumlevels * nrbmblocks;
nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
/*
* New summary size can't be more than half the size of
* the log. This prevents us from getting a log overflow,
* since we'll log basically the whole summary file at once.
*/
if (nrsumblocks > (mp->m_sb.sb_logblocks >> 1))
return -EINVAL;
/*
* Get the old block counts for bitmap and summary inodes.
* These can't change since other growfs callers are locked out.
*/
rbmblocks = XFS_B_TO_FSB(mp, mp->m_rbmip->i_disk_size);
rsumblocks = XFS_B_TO_FSB(mp, mp->m_rsumip->i_disk_size);
/*
* Allocate space to the bitmap and summary files, as necessary.
*/
error = xfs_growfs_rt_alloc(mp, rbmblocks, nrbmblocks, mp->m_rbmip);
if (error)
return error;
error = xfs_growfs_rt_alloc(mp, rsumblocks, nrsumblocks, mp->m_rsumip);
if (error)
return error;
rsum_cache = mp->m_rsum_cache;
if (nrbmblocks != sbp->sb_rbmblocks)
xfs_alloc_rsum_cache(mp, nrbmblocks);
/*
* Allocate a new (fake) mount/sb.
*/
nmp = kmem_alloc(sizeof(*nmp), 0);
/*
* Loop over the bitmap blocks.
* We will do everything one bitmap block at a time.
* Skip the current block if it is exactly full.
* This also deals with the case where there were no rtextents before.
*/
for (bmbno = sbp->sb_rbmblocks -
((sbp->sb_rextents & ((1 << mp->m_blkbit_log) - 1)) != 0);
bmbno < nrbmblocks;
bmbno++) {
struct xfs_trans *tp;
xfs_rfsblock_t nrblocks_step;
*nmp = *mp;
nsbp = &nmp->m_sb;
/*
* Calculate new sb and mount fields for this round.
*/
nsbp->sb_rextsize = in->extsize;
nmp->m_rtxblklog = -1; /* don't use shift or masking */
nsbp->sb_rbmblocks = bmbno + 1;
nrblocks_step = (bmbno + 1) * NBBY * nsbp->sb_blocksize *
nsbp->sb_rextsize;
nsbp->sb_rblocks = min(nrblocks, nrblocks_step);
nsbp->sb_rextents = xfs_rtb_to_rtx(nmp, nsbp->sb_rblocks);
ASSERT(nsbp->sb_rextents != 0);
nsbp->sb_rextslog = xfs_highbit32(nsbp->sb_rextents);
nrsumlevels = nmp->m_rsumlevels = nsbp->sb_rextslog + 1;
nrsumsize =
(uint)sizeof(xfs_suminfo_t) * nrsumlevels *
nsbp->sb_rbmblocks;
nrsumblocks = XFS_B_TO_FSB(mp, nrsumsize);
nmp->m_rsumsize = nrsumsize = XFS_FSB_TO_B(mp, nrsumblocks);
/*
* Start a transaction, get the log reservation.
*/
error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growrtfree, 0, 0, 0,
&tp);
if (error)
break;
/*
* Lock out other callers by grabbing the bitmap inode lock.
*/
xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL | XFS_ILOCK_RTBITMAP);
xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
/*
* Update the bitmap inode's size ondisk and incore. We need
* to update the incore size so that inode inactivation won't
* punch what it thinks are "posteof" blocks.
*/
mp->m_rbmip->i_disk_size =
nsbp->sb_rbmblocks * nsbp->sb_blocksize;
i_size_write(VFS_I(mp->m_rbmip), mp->m_rbmip->i_disk_size);
xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
/*
* Get the summary inode into the transaction.
*/
xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL | XFS_ILOCK_RTSUM);
xfs_trans_ijoin(tp, mp->m_rsumip, XFS_ILOCK_EXCL);
/*
* Update the summary inode's size. We need to update the
* incore size so that inode inactivation won't punch what it
* thinks are "posteof" blocks.
*/
mp->m_rsumip->i_disk_size = nmp->m_rsumsize;
i_size_write(VFS_I(mp->m_rsumip), mp->m_rsumip->i_disk_size);
xfs_trans_log_inode(tp, mp->m_rsumip, XFS_ILOG_CORE);
/*
* Copy summary data from old to new sizes.
* Do this when the real size (not block-aligned) changes.
*/
if (sbp->sb_rbmblocks != nsbp->sb_rbmblocks ||
mp->m_rsumlevels != nmp->m_rsumlevels) {
error = xfs_rtcopy_summary(mp, nmp, tp);
if (error)
goto error_cancel;
}
/*
* Update superblock fields.
*/
if (nsbp->sb_rextsize != sbp->sb_rextsize)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSIZE,
nsbp->sb_rextsize - sbp->sb_rextsize);
if (nsbp->sb_rbmblocks != sbp->sb_rbmblocks)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBMBLOCKS,
nsbp->sb_rbmblocks - sbp->sb_rbmblocks);
if (nsbp->sb_rblocks != sbp->sb_rblocks)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_RBLOCKS,
nsbp->sb_rblocks - sbp->sb_rblocks);
if (nsbp->sb_rextents != sbp->sb_rextents)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTENTS,
nsbp->sb_rextents - sbp->sb_rextents);
if (nsbp->sb_rextslog != sbp->sb_rextslog)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_REXTSLOG,
nsbp->sb_rextslog - sbp->sb_rextslog);
/*
* Free new extent.
*/
bp = NULL;
error = xfs_rtfree_range(nmp, tp, sbp->sb_rextents,
nsbp->sb_rextents - sbp->sb_rextents, &bp, &sumbno);
if (error) {
error_cancel:
xfs_trans_cancel(tp);
break;
}
/*
* Mark more blocks free in the superblock.
*/
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS,
nsbp->sb_rextents - sbp->sb_rextents);
/*
* Update mp values into the real mp structure.
*/
mp->m_rsumlevels = nrsumlevels;
mp->m_rsumsize = nrsumsize;
error = xfs_trans_commit(tp);
if (error)
break;
/* Ensure the mount RT feature flag is now set. */
mp->m_features |= XFS_FEAT_REALTIME;
}
if (error)
goto out_free;
/* Update secondary superblocks now the physical grow has completed */
error = xfs_update_secondary_sbs(mp);
out_free:
/*
* Free the fake mp structure.
*/
kmem_free(nmp);
/*
* If we had to allocate a new rsum_cache, we either need to free the
* old one (if we succeeded) or free the new one and restore the old one
* (if there was an error).
*/
if (rsum_cache != mp->m_rsum_cache) {
if (error) {
kmem_free(mp->m_rsum_cache);
mp->m_rsum_cache = rsum_cache;
} else {
kmem_free(rsum_cache);
}
}
return error;
}
/*
* Allocate an extent in the realtime subvolume, with the usual allocation
* parameters. The length units are all in realtime extents, as is the
* result block number.
*/
int /* error */
xfs_rtallocate_extent(
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxnum_t start, /* starting rtext number to allocate */
xfs_rtxlen_t minlen, /* minimum length to allocate */
xfs_rtxlen_t maxlen, /* maximum length to allocate */
xfs_rtxlen_t *len, /* out: actual length allocated */
int wasdel, /* was a delayed allocation extent */
xfs_rtxlen_t prod, /* extent product factor */
xfs_rtxnum_t *rtblock) /* out: start rtext allocated */
{
xfs_mount_t *mp = tp->t_mountp;
int error; /* error value */
xfs_rtxnum_t r; /* result allocated rtext */
xfs_fileoff_t sb; /* summary file block number */
struct xfs_buf *sumbp; /* summary file block buffer */
ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
ASSERT(minlen > 0 && minlen <= maxlen);
/*
* If prod is set then figure out what to do to minlen and maxlen.
*/
if (prod > 1) {
xfs_rtxlen_t i;
if ((i = maxlen % prod))
maxlen -= i;
if ((i = minlen % prod))
minlen += prod - i;
if (maxlen < minlen) {
*rtblock = NULLRTEXTNO;
return 0;
}
}
retry:
sumbp = NULL;
if (start == 0) {
error = xfs_rtallocate_extent_size(mp, tp, minlen, maxlen, len,
&sumbp, &sb, prod, &r);
} else {
error = xfs_rtallocate_extent_near(mp, tp, start, minlen, maxlen,
len, &sumbp, &sb, prod, &r);
}
if (error)
return error;
/*
* If it worked, update the superblock.
*/
if (r != NULLRTEXTNO) {
long slen = (long)*len;
ASSERT(*len >= minlen && *len <= maxlen);
if (wasdel)
xfs_trans_mod_sb(tp, XFS_TRANS_SB_RES_FREXTENTS, -slen);
else
xfs_trans_mod_sb(tp, XFS_TRANS_SB_FREXTENTS, -slen);
} else if (prod > 1) {
prod = 1;
goto retry;
}
*rtblock = r;
return 0;
}
/*
* Initialize realtime fields in the mount structure.
*/
int /* error */
xfs_rtmount_init(
struct xfs_mount *mp) /* file system mount structure */
{
struct xfs_buf *bp; /* buffer for last block of subvolume */
struct xfs_sb *sbp; /* filesystem superblock copy in mount */
xfs_daddr_t d; /* address of last block of subvolume */
int error;
sbp = &mp->m_sb;
if (sbp->sb_rblocks == 0)
return 0;
if (mp->m_rtdev_targp == NULL) {
xfs_warn(mp,
"Filesystem has a realtime volume, use rtdev=device option");
return -ENODEV;
}
mp->m_rsumlevels = sbp->sb_rextslog + 1;
mp->m_rsumsize =
(uint)sizeof(xfs_suminfo_t) * mp->m_rsumlevels *
sbp->sb_rbmblocks;
mp->m_rsumsize = roundup(mp->m_rsumsize, sbp->sb_blocksize);
mp->m_rbmip = mp->m_rsumip = NULL;
/*
* Check that the realtime section is an ok size.
*/
d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_rblocks);
if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_rblocks) {
xfs_warn(mp, "realtime mount -- %llu != %llu",
(unsigned long long) XFS_BB_TO_FSB(mp, d),
(unsigned long long) mp->m_sb.sb_rblocks);
return -EFBIG;
}
error = xfs_buf_read_uncached(mp->m_rtdev_targp,
d - XFS_FSB_TO_BB(mp, 1),
XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
if (error) {
xfs_warn(mp, "realtime device size check failed");
return error;
}
xfs_buf_relse(bp);
return 0;
}
xfs: recalculate free rt extents after log recovery I've been observing periodic corruption reports from xfs_scrub involving the free rt extent counter (frextents) while running xfs/141. That test uses an error injection knob to induce a torn write to the log, and an arbitrary number of recovery mounts, frextents will count fewer free rt extents than can be found the rtbitmap. The root cause of the problem is a combination of the misuse of sb_frextents in the incore mount to reflect both incore reservations made by running transactions as well as the actual count of free rt extents on disk. The following sequence can reproduce the undercount: Thread 1 Thread 2 xfs_trans_alloc(rtextents=3) xfs_mod_frextents(-3) <blocks> xfs_attr_set() xfs_bmap_attr_addfork() xfs_add_attr2() xfs_log_sb() xfs_sb_to_disk() xfs_trans_commit() <log flushed to disk> <log goes down> Note that thread 1 subtracts 3 from sb_frextents even though it never commits to using that space. Thread 2 writes the undercounted value to the ondisk superblock and logs it to the xattr transaction, which is then flushed to disk. At next mount, log recovery will find the logged superblock and write that back into the filesystem. At the end of log recovery, we reread the superblock and install the recovered undercounted frextents value into the incore superblock. From that point on, we've effectively leaked thread 1's transaction reservation. The correct fix for this is to separate the incore reservation from the ondisk usage, but that's a matter for the next patch. Because the kernel has been logging superblocks with undercounted frextents for a very long time and we don't demand that sysadmins run xfs_repair after a crash, fix the undercount by recomputing frextents after log recovery. Gating this on log recovery is a reasonable balance (I think) between correcting the problem and slowing down every mount attempt. Note that xfs_repair will fix undercounted frextents. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-11 20:49:42 +00:00
static int
xfs_rtalloc_count_frextent(
struct xfs_mount *mp,
struct xfs_trans *tp,
const struct xfs_rtalloc_rec *rec,
void *priv)
{
uint64_t *valp = priv;
*valp += rec->ar_extcount;
return 0;
}
/*
* Reinitialize the number of free realtime extents from the realtime bitmap.
* Callers must ensure that there is no other activity in the filesystem.
*/
int
xfs_rtalloc_reinit_frextents(
struct xfs_mount *mp)
{
uint64_t val = 0;
int error;
xfs_ilock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
xfs: recalculate free rt extents after log recovery I've been observing periodic corruption reports from xfs_scrub involving the free rt extent counter (frextents) while running xfs/141. That test uses an error injection knob to induce a torn write to the log, and an arbitrary number of recovery mounts, frextents will count fewer free rt extents than can be found the rtbitmap. The root cause of the problem is a combination of the misuse of sb_frextents in the incore mount to reflect both incore reservations made by running transactions as well as the actual count of free rt extents on disk. The following sequence can reproduce the undercount: Thread 1 Thread 2 xfs_trans_alloc(rtextents=3) xfs_mod_frextents(-3) <blocks> xfs_attr_set() xfs_bmap_attr_addfork() xfs_add_attr2() xfs_log_sb() xfs_sb_to_disk() xfs_trans_commit() <log flushed to disk> <log goes down> Note that thread 1 subtracts 3 from sb_frextents even though it never commits to using that space. Thread 2 writes the undercounted value to the ondisk superblock and logs it to the xattr transaction, which is then flushed to disk. At next mount, log recovery will find the logged superblock and write that back into the filesystem. At the end of log recovery, we reread the superblock and install the recovered undercounted frextents value into the incore superblock. From that point on, we've effectively leaked thread 1's transaction reservation. The correct fix for this is to separate the incore reservation from the ondisk usage, but that's a matter for the next patch. Because the kernel has been logging superblocks with undercounted frextents for a very long time and we don't demand that sysadmins run xfs_repair after a crash, fix the undercount by recomputing frextents after log recovery. Gating this on log recovery is a reasonable balance (I think) between correcting the problem and slowing down every mount attempt. Note that xfs_repair will fix undercounted frextents. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-11 20:49:42 +00:00
error = xfs_rtalloc_query_all(mp, NULL, xfs_rtalloc_count_frextent,
&val);
xfs_iunlock(mp->m_rbmip, XFS_ILOCK_SHARED | XFS_ILOCK_RTBITMAP);
xfs: recalculate free rt extents after log recovery I've been observing periodic corruption reports from xfs_scrub involving the free rt extent counter (frextents) while running xfs/141. That test uses an error injection knob to induce a torn write to the log, and an arbitrary number of recovery mounts, frextents will count fewer free rt extents than can be found the rtbitmap. The root cause of the problem is a combination of the misuse of sb_frextents in the incore mount to reflect both incore reservations made by running transactions as well as the actual count of free rt extents on disk. The following sequence can reproduce the undercount: Thread 1 Thread 2 xfs_trans_alloc(rtextents=3) xfs_mod_frextents(-3) <blocks> xfs_attr_set() xfs_bmap_attr_addfork() xfs_add_attr2() xfs_log_sb() xfs_sb_to_disk() xfs_trans_commit() <log flushed to disk> <log goes down> Note that thread 1 subtracts 3 from sb_frextents even though it never commits to using that space. Thread 2 writes the undercounted value to the ondisk superblock and logs it to the xattr transaction, which is then flushed to disk. At next mount, log recovery will find the logged superblock and write that back into the filesystem. At the end of log recovery, we reread the superblock and install the recovered undercounted frextents value into the incore superblock. From that point on, we've effectively leaked thread 1's transaction reservation. The correct fix for this is to separate the incore reservation from the ondisk usage, but that's a matter for the next patch. Because the kernel has been logging superblocks with undercounted frextents for a very long time and we don't demand that sysadmins run xfs_repair after a crash, fix the undercount by recomputing frextents after log recovery. Gating this on log recovery is a reasonable balance (I think) between correcting the problem and slowing down every mount attempt. Note that xfs_repair will fix undercounted frextents. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-11 20:49:42 +00:00
if (error)
return error;
spin_lock(&mp->m_sb_lock);
mp->m_sb.sb_frextents = val;
spin_unlock(&mp->m_sb_lock);
percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
xfs: recalculate free rt extents after log recovery I've been observing periodic corruption reports from xfs_scrub involving the free rt extent counter (frextents) while running xfs/141. That test uses an error injection knob to induce a torn write to the log, and an arbitrary number of recovery mounts, frextents will count fewer free rt extents than can be found the rtbitmap. The root cause of the problem is a combination of the misuse of sb_frextents in the incore mount to reflect both incore reservations made by running transactions as well as the actual count of free rt extents on disk. The following sequence can reproduce the undercount: Thread 1 Thread 2 xfs_trans_alloc(rtextents=3) xfs_mod_frextents(-3) <blocks> xfs_attr_set() xfs_bmap_attr_addfork() xfs_add_attr2() xfs_log_sb() xfs_sb_to_disk() xfs_trans_commit() <log flushed to disk> <log goes down> Note that thread 1 subtracts 3 from sb_frextents even though it never commits to using that space. Thread 2 writes the undercounted value to the ondisk superblock and logs it to the xattr transaction, which is then flushed to disk. At next mount, log recovery will find the logged superblock and write that back into the filesystem. At the end of log recovery, we reread the superblock and install the recovered undercounted frextents value into the incore superblock. From that point on, we've effectively leaked thread 1's transaction reservation. The correct fix for this is to separate the incore reservation from the ondisk usage, but that's a matter for the next patch. Because the kernel has been logging superblocks with undercounted frextents for a very long time and we don't demand that sysadmins run xfs_repair after a crash, fix the undercount by recomputing frextents after log recovery. Gating this on log recovery is a reasonable balance (I think) between correcting the problem and slowing down every mount attempt. Note that xfs_repair will fix undercounted frextents. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
2022-04-11 20:49:42 +00:00
return 0;
}
/*
* Read in the bmbt of an rt metadata inode so that we never have to load them
* at runtime. This enables the use of shared ILOCKs for rtbitmap scans. Use
* an empty transaction to avoid deadlocking on loops in the bmbt.
*/
static inline int
xfs_rtmount_iread_extents(
struct xfs_inode *ip,
unsigned int lock_class)
{
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc_empty(ip->i_mount, &tp);
if (error)
return error;
xfs_ilock(ip, XFS_ILOCK_EXCL | lock_class);
error = xfs_iread_extents(tp, ip, XFS_DATA_FORK);
if (error)
goto out_unlock;
if (xfs_inode_has_attr_fork(ip)) {
error = xfs_iread_extents(tp, ip, XFS_ATTR_FORK);
if (error)
goto out_unlock;
}
out_unlock:
xfs_iunlock(ip, XFS_ILOCK_EXCL | lock_class);
xfs_trans_cancel(tp);
return error;
}
/*
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
* Get the bitmap and summary inodes and the summary cache into the mount
* structure at mount time.
*/
int /* error */
xfs_rtmount_inodes(
xfs_mount_t *mp) /* file system mount structure */
{
int error; /* error return value */
xfs_sb_t *sbp;
sbp = &mp->m_sb;
error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip);
if (error)
return error;
ASSERT(mp->m_rbmip != NULL);
error = xfs_rtmount_iread_extents(mp->m_rbmip, XFS_ILOCK_RTBITMAP);
if (error)
goto out_rele_bitmap;
error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip);
if (error)
goto out_rele_bitmap;
ASSERT(mp->m_rsumip != NULL);
error = xfs_rtmount_iread_extents(mp->m_rsumip, XFS_ILOCK_RTSUM);
if (error)
goto out_rele_summary;
xfs_alloc_rsum_cache(mp, sbp->sb_rbmblocks);
return 0;
out_rele_summary:
xfs_irele(mp->m_rsumip);
out_rele_bitmap:
xfs_irele(mp->m_rbmip);
return error;
}
void
xfs_rtunmount_inodes(
struct xfs_mount *mp)
{
xfs: cache minimum realtime summary level The realtime summary is a two-dimensional array on disk, effectively: u32 rsum[log2(number of realtime extents) + 1][number of blocks in the bitmap] rsum[log][bbno] is the number of extents of size 2**log which start in bitmap block bbno. xfs_rtallocate_extent_near() uses xfs_rtany_summary() to check whether rsum[log][bbno] != 0 for any log level. However, the summary array is stored in row-major order (i.e., like an array in C), so all of these entries are not adjacent, but rather spread across the entire summary file. In the worst case (a full bitmap block), xfs_rtany_summary() has to check every level. This means that on a moderately-used realtime device, an allocation will waste a lot of time finding, reading, and releasing buffers for the realtime summary. In particular, one of our storage services (which runs on servers with 8 very slow CPUs and 15 8 TB XFS realtime filesystems) spends almost 5% of its CPU cycles in xfs_rtbuf_get() and xfs_trans_brelse() called from xfs_rtany_summary(). One solution would be to also store the summary with the dimensions swapped. However, this would require a disk format change to a very old component of XFS. Instead, we can cache the minimum size which contains any extents. We do so lazily; rather than guaranteeing that the cache contains the precise minimum, it always contains a loose lower bound which we tighten when we read or update a summary block. This only uses a few kilobytes of memory and is already serialized via the realtime bitmap and summary inode locks, so the cost is minimal. With this change, the same workload only spends 0.2% of its CPU cycles in the realtime allocator. Signed-off-by: Omar Sandoval <osandov@fb.com> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
2018-12-12 16:46:32 +00:00
kmem_free(mp->m_rsum_cache);
if (mp->m_rbmip)
xfs_irele(mp->m_rbmip);
if (mp->m_rsumip)
xfs_irele(mp->m_rsumip);
}
/*
* Pick an extent for allocation at the start of a new realtime file.
* Use the sequence number stored in the atime field of the bitmap inode.
* Translate this to a fraction of the rtextents, and return the product
* of rtextents and the fraction.
* The fraction sequence is 0, 1/2, 1/4, 3/4, 1/8, ..., 7/8, 1/16, ...
*/
int /* error */
xfs_rtpick_extent(
xfs_mount_t *mp, /* file system mount point */
xfs_trans_t *tp, /* transaction pointer */
xfs_rtxlen_t len, /* allocation length (rtextents) */
xfs_rtxnum_t *pick) /* result rt extent */
{
xfs_rtxnum_t b; /* result rtext */
int log2; /* log of sequence number */
uint64_t resid; /* residual after log removed */
uint64_t seq; /* sequence number of file creation */
uint64_t *seqp; /* pointer to seqno in inode */
ASSERT(xfs_isilocked(mp->m_rbmip, XFS_ILOCK_EXCL));
seqp = (uint64_t *)&VFS_I(mp->m_rbmip)->i_atime;
if (!(mp->m_rbmip->i_diflags & XFS_DIFLAG_NEWRTBM)) {
mp->m_rbmip->i_diflags |= XFS_DIFLAG_NEWRTBM;
*seqp = 0;
}
seq = *seqp;
if ((log2 = xfs_highbit64(seq)) == -1)
b = 0;
else {
resid = seq - (1ULL << log2);
b = (mp->m_sb.sb_rextents * ((resid << 1) + 1ULL)) >>
(log2 + 1);
if (b >= mp->m_sb.sb_rextents)
div64_u64_rem(b, mp->m_sb.sb_rextents, &b);
if (b + len > mp->m_sb.sb_rextents)
b = mp->m_sb.sb_rextents - len;
}
*seqp = seq + 1;
xfs_trans_log_inode(tp, mp->m_rbmip, XFS_ILOG_CORE);
*pick = b;
return 0;
}