1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2010 Red Hat, Inc.
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_btree.h"
14 #include "xfs_alloc_btree.h"
15 #include "xfs_alloc.h"
16 #include "xfs_error.h"
17 #include "xfs_extent_busy.h"
18 #include "xfs_trace.h"
28 uint64_t *blocks_trimmed
)
30 struct block_device
*bdev
= mp
->m_ddev_targp
->bt_bdev
;
31 struct xfs_btree_cur
*cur
;
33 struct xfs_perag
*pag
;
37 pag
= xfs_perag_get(mp
, agno
);
40 * Force out the log. This means any transactions that might have freed
41 * space before we take the AGF buffer lock are now on disk, and the
42 * volatile disk cache is flushed.
44 xfs_log_force(mp
, XFS_LOG_SYNC
);
46 error
= xfs_alloc_read_agf(mp
, NULL
, agno
, 0, &agbp
);
50 cur
= xfs_allocbt_init_cursor(mp
, NULL
, agbp
, agno
, XFS_BTNUM_CNT
);
53 * Look up the longest btree in the AGF and start with it.
55 error
= xfs_alloc_lookup_ge(cur
, 0,
56 be32_to_cpu(XFS_BUF_TO_AGF(agbp
)->agf_longest
), &i
);
61 * Loop until we are done with all extents that are large
62 * enough to be worth discarding.
70 error
= xfs_alloc_get_rec(cur
, &fbno
, &flen
, &i
);
73 XFS_WANT_CORRUPTED_GOTO(mp
, i
== 1, out_del_cursor
);
74 ASSERT(flen
<= be32_to_cpu(XFS_BUF_TO_AGF(agbp
)->agf_longest
));
77 * use daddr format for all range/len calculations as that is
78 * the format the range/len variables are supplied in by
81 dbno
= XFS_AGB_TO_DADDR(mp
, agno
, fbno
);
82 dlen
= XFS_FSB_TO_BB(mp
, flen
);
88 trace_xfs_discard_toosmall(mp
, agno
, fbno
, flen
);
93 * If the extent is entirely outside of the range we are
94 * supposed to discard skip it. Do not bother to trim
95 * down partially overlapping ranges for now.
97 if (dbno
+ dlen
< start
|| dbno
> end
) {
98 trace_xfs_discard_exclude(mp
, agno
, fbno
, flen
);
103 * If any blocks in the range are still busy, skip the
104 * discard and try again the next time.
106 if (xfs_extent_busy_search(mp
, agno
, fbno
, flen
)) {
107 trace_xfs_discard_busy(mp
, agno
, fbno
, flen
);
111 trace_xfs_discard_extent(mp
, agno
, fbno
, flen
);
112 error
= blkdev_issue_discard(bdev
, dbno
, dlen
, GFP_NOFS
, 0);
115 *blocks_trimmed
+= flen
;
118 error
= xfs_btree_decrement(cur
, 0, &i
);
122 if (fatal_signal_pending(current
)) {
123 error
= -ERESTARTSYS
;
129 xfs_btree_del_cursor(cur
, error
);
137 * trim a range of the filesystem.
139 * Note: the parameters passed from userspace are byte ranges into the
140 * filesystem which does not match to the format we use for filesystem block
141 * addressing. FSB addressing is sparse (AGNO|AGBNO), while the incoming format
142 * is a linear address range. Hence we need to use DADDR based conversions and
143 * comparisons for determining the correct offset and regions to trim.
147 struct xfs_mount
*mp
,
148 struct fstrim_range __user
*urange
)
150 struct request_queue
*q
= bdev_get_queue(mp
->m_ddev_targp
->bt_bdev
);
151 unsigned int granularity
= q
->limits
.discard_granularity
;
152 struct fstrim_range range
;
153 xfs_daddr_t start
, end
, minlen
;
154 xfs_agnumber_t start_agno
, end_agno
, agno
;
155 uint64_t blocks_trimmed
= 0;
156 int error
, last_error
= 0;
158 if (!capable(CAP_SYS_ADMIN
))
160 if (!blk_queue_discard(q
))
164 * We haven't recovered the log, so we cannot use our bnobt-guided
165 * storage zapping commands.
167 if (mp
->m_flags
& XFS_MOUNT_NORECOVERY
)
170 if (copy_from_user(&range
, urange
, sizeof(range
)))
173 range
.minlen
= max_t(u64
, granularity
, range
.minlen
);
174 minlen
= BTOBB(range
.minlen
);
176 * Truncating down the len isn't actually quite correct, but using
177 * BBTOB would mean we trivially get overflows for values
178 * of ULLONG_MAX or slightly lower. And ULLONG_MAX is the default
179 * used by the fstrim application. In the end it really doesn't
180 * matter as trimming blocks is an advisory interface.
182 if (range
.start
>= XFS_FSB_TO_B(mp
, mp
->m_sb
.sb_dblocks
) ||
183 range
.minlen
> XFS_FSB_TO_B(mp
, mp
->m_ag_max_usable
) ||
184 range
.len
< mp
->m_sb
.sb_blocksize
)
187 start
= BTOBB(range
.start
);
188 end
= start
+ BTOBBT(range
.len
) - 1;
190 if (end
> XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
) - 1)
191 end
= XFS_FSB_TO_BB(mp
, mp
->m_sb
.sb_dblocks
)- 1;
193 start_agno
= xfs_daddr_to_agno(mp
, start
);
194 end_agno
= xfs_daddr_to_agno(mp
, end
);
196 for (agno
= start_agno
; agno
<= end_agno
; agno
++) {
197 error
= xfs_trim_extents(mp
, agno
, start
, end
, minlen
,
201 if (error
== -ERESTARTSYS
)
209 range
.len
= XFS_FSB_TO_B(mp
, blocks_trimmed
);
210 if (copy_to_user(urange
, &range
, sizeof(range
)))