Minki/linux
1
0
Fork 1
mirror of https://github.com/torvalds/linux.git synced 2024-11-24 21:21:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

xfs: use rtgroup busy extent list for FITRIM

Browse Source 
For filesystems that have rtgroups and hence use the busy extent list
for freed rt space, use that busy extent list so that FITRIM can issue
discard commands asynchronously without worrying about other callers
accidentally allocating and using space that is being discarded.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
This commit is contained in:
Darrick J. Wong 2024-11-03 20:19:36 -08:00
parent 7e85fc2394
commit a3315d1130
1 changed files with 144 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

146
fs/xfs/xfs_discard.c
View File

@ -73,6 +73,8 @@
* extent search so that it overlaps in flight discard IO.
*/
#define XFS_DISCARD_MAX_EXAMINE (100)
struct workqueue_struct *xfs_discard_wq;
static void
@ -185,7 +187,7 @@ xfs_trim_gather_extents(
struct xfs_buf *agbp;
int error;
int i;
int batch = 100;
int batch = XFS_DISCARD_MAX_EXAMINE;
/*
* Force out the log. This means any transactions that might have freed
@ -565,6 +567,7 @@ xfs_trim_gather_rtextent(
return 0;
}
/* Trim extents on an !rtgroups realtime device */
static int
xfs_trim_rtextents(
struct xfs_rtgroup *rtg,
@ -619,6 +622,140 @@ xfs_trim_rtextents(
return error;
}
struct xfs_trim_rtgroup {
/* list of rtgroup extents to free */
struct xfs_busy_extents *extents;
/* minimum length that caller allows us to trim */
xfs_rtblock_t minlen_fsb;
/* restart point for the rtbitmap walk */
xfs_rtxnum_t restart_rtx;
/* number of extents to examine before stopping to issue discard ios */
int batch;
/* number of extents queued for discard */
int queued;
};
static int
xfs_trim_gather_rtgroup_extent(
struct xfs_rtgroup *rtg,
struct xfs_trans *tp,
const struct xfs_rtalloc_rec *rec,
void *priv)
{
struct xfs_trim_rtgroup *tr = priv;
xfs_rgblock_t rgbno;
xfs_extlen_t len;
if (--tr->batch <= 0) {
/*
* If we've checked a large number of extents, update the
* cursor to point at this extent so we restart the next batch
* from this extent.
*/
tr->restart_rtx = rec->ar_startext;
return -ECANCELED;
}
rgbno = xfs_rtx_to_rgbno(rtg, rec->ar_startext);
len = xfs_rtxlen_to_extlen(rtg_mount(rtg), rec->ar_extcount);
/* Ignore too small. */
if (len < tr->minlen_fsb) {
trace_xfs_discard_toosmall(rtg_group(rtg), rgbno, len);
return 0;
}
/*
* If any blocks in the range are still busy, skip the discard and try
* again the next time.
*/
if (xfs_extent_busy_search(rtg_group(rtg), rgbno, len)) {
trace_xfs_discard_busy(rtg_group(rtg), rgbno, len);
return 0;
}
xfs_extent_busy_insert_discard(rtg_group(rtg), rgbno, len,
&tr->extents->extent_list);
tr->queued++;
tr->restart_rtx = rec->ar_startext + rec->ar_extcount;
return 0;
}
/* Trim extents in this rtgroup using the busy extent machinery. */
static int
xfs_trim_rtgroup_extents(
struct xfs_rtgroup *rtg,
xfs_rtxnum_t low,
xfs_rtxnum_t high,
xfs_daddr_t minlen)
{
struct xfs_mount *mp = rtg_mount(rtg);
struct xfs_trim_rtgroup tr = {
.minlen_fsb = XFS_BB_TO_FSB(mp, minlen),
};
struct xfs_trans *tp;
int error;
error = xfs_trans_alloc_empty(mp, &tp);
if (error)
return error;
/*
* Walk the free ranges between low and high. The query_range function
* trims the extents returned.
*/
do {
tr.extents = kzalloc(sizeof(*tr.extents), GFP_KERNEL);
if (!tr.extents) {
error = -ENOMEM;
break;
}
tr.queued = 0;
tr.batch = XFS_DISCARD_MAX_EXAMINE;
tr.extents->owner = tr.extents;
INIT_LIST_HEAD(&tr.extents->extent_list);
xfs_rtgroup_lock(rtg, XFS_RTGLOCK_BITMAP_SHARED);
error = xfs_rtalloc_query_range(rtg, tp, low, high,
xfs_trim_gather_rtgroup_extent, &tr);
xfs_rtgroup_unlock(rtg, XFS_RTGLOCK_BITMAP_SHARED);
if (error == -ECANCELED)
error = 0;
if (error) {
kfree(tr.extents);
break;
}
if (!tr.queued)
break;
/*
* We hand the extent list to the discard function here so the
* discarded extents can be removed from the busy extent list.
* This allows the discards to run asynchronously with
* gathering the next round of extents to discard.
*
* However, we must ensure that we do not reference the extent
* list after this function call, as it may have been freed by
* the time control returns to us.
*/
error = xfs_discard_extents(rtg_mount(rtg), tr.extents);
if (error)
break;
low = tr.restart_rtx;
} while (!xfs_trim_should_stop() && low <= high);
xfs_trans_cancel(tp);
return error;
}
static int
xfs_trim_rtdev_extents(
struct xfs_mount *mp,
@ -657,7 +794,12 @@ xfs_trim_rtdev_extents(
if (rtg_rgno(rtg) == end_rgno)
rtg_end = min(rtg_end, end_rtx);
error = xfs_trim_rtextents(rtg, start_rtx, rtg_end, minlen);
if (xfs_has_rtgroups(mp))
error = xfs_trim_rtgroup_extents(rtg, start_rtx,
rtg_end, minlen);
else
error = xfs_trim_rtextents(rtg, start_rtx, rtg_end,
minlen);
if (error)
last_error = error;