1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2010 Red Hat, Inc.
4 * All Rights Reserved.
5 */
6 #include "xfs.h"
7 #include "xfs_shared.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_sb.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"
19 #include "xfs_log.h"
20
21 STATIC int
xfs_trim_extents(struct xfs_mount * mp,xfs_agnumber_t agno,xfs_daddr_t start,xfs_daddr_t end,xfs_daddr_t minlen,uint64_t * blocks_trimmed)22 xfs_trim_extents(
23 struct xfs_mount *mp,
24 xfs_agnumber_t agno,
25 xfs_daddr_t start,
26 xfs_daddr_t end,
27 xfs_daddr_t minlen,
28 uint64_t *blocks_trimmed)
29 {
30 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
31 struct xfs_btree_cur *cur;
32 struct xfs_buf *agbp;
33 struct xfs_perag *pag;
34 int error;
35 int i;
36
37 pag = xfs_perag_get(mp, agno);
38
39 /*
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.
43 */
44 xfs_log_force(mp, XFS_LOG_SYNC);
45
46 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
47 if (error || !agbp)
48 goto out_put_perag;
49
50 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_CNT);
51
52 /*
53 * Look up the longest btree in the AGF and start with it.
54 */
55 error = xfs_alloc_lookup_ge(cur, 0,
56 be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest), &i);
57 if (error)
58 goto out_del_cursor;
59
60 /*
61 * Loop until we are done with all extents that are large
62 * enough to be worth discarding.
63 */
64 while (i) {
65 xfs_agblock_t fbno;
66 xfs_extlen_t flen;
67 xfs_daddr_t dbno;
68 xfs_extlen_t dlen;
69
70 error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
71 if (error)
72 goto out_del_cursor;
73 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_del_cursor);
74 ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
75
76 /*
77 * use daddr format for all range/len calculations as that is
78 * the format the range/len variables are supplied in by
79 * userspace.
80 */
81 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
82 dlen = XFS_FSB_TO_BB(mp, flen);
83
84 /*
85 * Too small? Give up.
86 */
87 if (dlen < minlen) {
88 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
89 goto out_del_cursor;
90 }
91
92 /*
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.
96 */
97 if (dbno + dlen < start || dbno > end) {
98 trace_xfs_discard_exclude(mp, agno, fbno, flen);
99 goto next_extent;
100 }
101
102 /*
103 * If any blocks in the range are still busy, skip the
104 * discard and try again the next time.
105 */
106 if (xfs_extent_busy_search(mp, agno, fbno, flen)) {
107 trace_xfs_discard_busy(mp, agno, fbno, flen);
108 goto next_extent;
109 }
110
111 trace_xfs_discard_extent(mp, agno, fbno, flen);
112 error = blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS, 0);
113 if (error)
114 goto out_del_cursor;
115 *blocks_trimmed += flen;
116
117 next_extent:
118 error = xfs_btree_decrement(cur, 0, &i);
119 if (error)
120 goto out_del_cursor;
121
122 if (fatal_signal_pending(current)) {
123 error = -ERESTARTSYS;
124 goto out_del_cursor;
125 }
126 }
127
128 out_del_cursor:
129 xfs_btree_del_cursor(cur, error);
130 xfs_buf_relse(agbp);
131 out_put_perag:
132 xfs_perag_put(pag);
133 return error;
134 }
135
136 /*
137 * trim a range of the filesystem.
138 *
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.
144 */
145 int
xfs_ioc_trim(struct xfs_mount * mp,struct fstrim_range __user * urange)146 xfs_ioc_trim(
147 struct xfs_mount *mp,
148 struct fstrim_range __user *urange)
149 {
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;
157
158 if (!capable(CAP_SYS_ADMIN))
159 return -EPERM;
160 if (!blk_queue_discard(q))
161 return -EOPNOTSUPP;
162
163 /*
164 * We haven't recovered the log, so we cannot use our bnobt-guided
165 * storage zapping commands.
166 */
167 if (mp->m_flags & XFS_MOUNT_NORECOVERY)
168 return -EROFS;
169
170 if (copy_from_user(&range, urange, sizeof(range)))
171 return -EFAULT;
172
173 range.minlen = max_t(u64, granularity, range.minlen);
174 minlen = BTOBB(range.minlen);
175 /*
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.
181 */
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)
185 return -EINVAL;
186
187 start = BTOBB(range.start);
188 end = start + BTOBBT(range.len) - 1;
189
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;
192
193 start_agno = xfs_daddr_to_agno(mp, start);
194 end_agno = xfs_daddr_to_agno(mp, end);
195
196 for (agno = start_agno; agno <= end_agno; agno++) {
197 error = xfs_trim_extents(mp, agno, start, end, minlen,
198 &blocks_trimmed);
199 if (error) {
200 last_error = error;
201 if (error == -ERESTARTSYS)
202 break;
203 }
204 }
205
206 if (last_error)
207 return last_error;
208
209 range.len = XFS_FSB_TO_B(mp, blocks_trimmed);
210 if (copy_to_user(urange, &range, sizeof(range)))
211 return -EFAULT;
212 return 0;
213 }
214