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_mount.h"
12 #include "xfs_btree.h"
13 #include "xfs_alloc_btree.h"
14 #include "xfs_alloc.h"
15 #include "xfs_discard.h"
16 #include "xfs_error.h"
17 #include "xfs_extent_busy.h"
18 #include "xfs_trace.h"
19 #include "xfs_log.h"
20 #include "xfs_ag.h"
21
22 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)23 xfs_trim_extents(
24 struct xfs_mount *mp,
25 xfs_agnumber_t agno,
26 xfs_daddr_t start,
27 xfs_daddr_t end,
28 xfs_daddr_t minlen,
29 uint64_t *blocks_trimmed)
30 {
31 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
32 struct xfs_btree_cur *cur;
33 struct xfs_buf *agbp;
34 struct xfs_agf *agf;
35 struct xfs_perag *pag;
36 int error;
37 int i;
38
39 pag = xfs_perag_get(mp, agno);
40
41 /*
42 * Force out the log. This means any transactions that might have freed
43 * space before we take the AGF buffer lock are now on disk, and the
44 * volatile disk cache is flushed.
45 */
46 xfs_log_force(mp, XFS_LOG_SYNC);
47
48 error = xfs_alloc_read_agf(pag, NULL, 0, &agbp);
49 if (error)
50 goto out_put_perag;
51 agf = agbp->b_addr;
52
53 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, pag, XFS_BTNUM_CNT);
54
55 /*
56 * Look up the longest btree in the AGF and start with it.
57 */
58 error = xfs_alloc_lookup_ge(cur, 0, be32_to_cpu(agf->agf_longest), &i);
59 if (error)
60 goto out_del_cursor;
61
62 /*
63 * Loop until we are done with all extents that are large
64 * enough to be worth discarding.
65 */
66 while (i) {
67 xfs_agblock_t fbno;
68 xfs_extlen_t flen;
69 xfs_daddr_t dbno;
70 xfs_extlen_t dlen;
71
72 error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
73 if (error)
74 goto out_del_cursor;
75 if (XFS_IS_CORRUPT(mp, i != 1)) {
76 error = -EFSCORRUPTED;
77 goto out_del_cursor;
78 }
79 ASSERT(flen <= be32_to_cpu(agf->agf_longest));
80
81 /*
82 * use daddr format for all range/len calculations as that is
83 * the format the range/len variables are supplied in by
84 * userspace.
85 */
86 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
87 dlen = XFS_FSB_TO_BB(mp, flen);
88
89 /*
90 * Too small? Give up.
91 */
92 if (dlen < minlen) {
93 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
94 goto out_del_cursor;
95 }
96
97 /*
98 * If the extent is entirely outside of the range we are
99 * supposed to discard skip it. Do not bother to trim
100 * down partially overlapping ranges for now.
101 */
102 if (dbno + dlen < start || dbno > end) {
103 trace_xfs_discard_exclude(mp, agno, fbno, flen);
104 goto next_extent;
105 }
106
107 /*
108 * If any blocks in the range are still busy, skip the
109 * discard and try again the next time.
110 */
111 if (xfs_extent_busy_search(mp, pag, fbno, flen)) {
112 trace_xfs_discard_busy(mp, agno, fbno, flen);
113 goto next_extent;
114 }
115
116 trace_xfs_discard_extent(mp, agno, fbno, flen);
117 error = blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS);
118 if (error)
119 goto out_del_cursor;
120 *blocks_trimmed += flen;
121
122 next_extent:
123 error = xfs_btree_decrement(cur, 0, &i);
124 if (error)
125 goto out_del_cursor;
126
127 if (fatal_signal_pending(current)) {
128 error = -ERESTARTSYS;
129 goto out_del_cursor;
130 }
131 }
132
133 out_del_cursor:
134 xfs_btree_del_cursor(cur, error);
135 xfs_buf_relse(agbp);
136 out_put_perag:
137 xfs_perag_put(pag);
138 return error;
139 }
140
141 /*
142 * trim a range of the filesystem.
143 *
144 * Note: the parameters passed from userspace are byte ranges into the
145 * filesystem which does not match to the format we use for filesystem block
146 * addressing. FSB addressing is sparse (AGNO|AGBNO), while the incoming format
147 * is a linear address range. Hence we need to use DADDR based conversions and
148 * comparisons for determining the correct offset and regions to trim.
149 */
150 int
xfs_ioc_trim(struct xfs_mount * mp,struct fstrim_range __user * urange)151 xfs_ioc_trim(
152 struct xfs_mount *mp,
153 struct fstrim_range __user *urange)
154 {
155 unsigned int granularity =
156 bdev_discard_granularity(mp->m_ddev_targp->bt_bdev);
157 struct fstrim_range range;
158 xfs_daddr_t start, end, minlen;
159 xfs_agnumber_t start_agno, end_agno, agno;
160 uint64_t blocks_trimmed = 0;
161 int error, last_error = 0;
162
163 if (!capable(CAP_SYS_ADMIN))
164 return -EPERM;
165 if (!bdev_max_discard_sectors(mp->m_ddev_targp->bt_bdev))
166 return -EOPNOTSUPP;
167
168 /*
169 * We haven't recovered the log, so we cannot use our bnobt-guided
170 * storage zapping commands.
171 */
172 if (xfs_has_norecovery(mp))
173 return -EROFS;
174
175 if (copy_from_user(&range, urange, sizeof(range)))
176 return -EFAULT;
177
178 range.minlen = max_t(u64, granularity, range.minlen);
179 minlen = BTOBB(range.minlen);
180 /*
181 * Truncating down the len isn't actually quite correct, but using
182 * BBTOB would mean we trivially get overflows for values
183 * of ULLONG_MAX or slightly lower. And ULLONG_MAX is the default
184 * used by the fstrim application. In the end it really doesn't
185 * matter as trimming blocks is an advisory interface.
186 */
187 if (range.start >= XFS_FSB_TO_B(mp, mp->m_sb.sb_dblocks) ||
188 range.minlen > XFS_FSB_TO_B(mp, mp->m_ag_max_usable) ||
189 range.len < mp->m_sb.sb_blocksize)
190 return -EINVAL;
191
192 start = BTOBB(range.start);
193 end = start + BTOBBT(range.len) - 1;
194
195 if (end > XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) - 1)
196 end = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)- 1;
197
198 start_agno = xfs_daddr_to_agno(mp, start);
199 end_agno = xfs_daddr_to_agno(mp, end);
200
201 for (agno = start_agno; agno <= end_agno; agno++) {
202 error = xfs_trim_extents(mp, agno, start, end, minlen,
203 &blocks_trimmed);
204 if (error) {
205 last_error = error;
206 if (error == -ERESTARTSYS)
207 break;
208 }
209 }
210
211 if (last_error)
212 return last_error;
213
214 range.len = XFS_FSB_TO_B(mp, blocks_trimmed);
215 if (copy_to_user(urange, &range, sizeof(range)))
216 return -EFAULT;
217 return 0;
218 }
219