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_format.h"
8 #include "xfs_log_format.h"
9 #include "xfs_trans_resv.h"
10 #include "xfs_sb.h"
11 #include "xfs_mount.h"
12 #include "xfs_quota.h"
13 #include "xfs_inode.h"
14 #include "xfs_btree.h"
15 #include "xfs_alloc_btree.h"
16 #include "xfs_alloc.h"
17 #include "xfs_error.h"
18 #include "xfs_extent_busy.h"
19 #include "xfs_discard.h"
20 #include "xfs_trace.h"
21 #include "xfs_log.h"
22
23 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)24 xfs_trim_extents(
25 struct xfs_mount *mp,
26 xfs_agnumber_t agno,
27 xfs_daddr_t start,
28 xfs_daddr_t end,
29 xfs_daddr_t minlen,
30 uint64_t *blocks_trimmed)
31 {
32 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
33 struct xfs_btree_cur *cur;
34 struct xfs_buf *agbp;
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(mp, NULL, agno, 0, &agbp);
49 if (error || !agbp)
50 goto out_put_perag;
51
52 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_CNT);
53
54 /*
55 * Look up the longest btree in the AGF and start with it.
56 */
57 error = xfs_alloc_lookup_ge(cur, 0,
58 be32_to_cpu(XFS_BUF_TO_AGF(agbp)->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 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_del_cursor);
76 ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
77
78 /*
79 * use daddr format for all range/len calculations as that is
80 * the format the range/len variables are supplied in by
81 * userspace.
82 */
83 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
84 dlen = XFS_FSB_TO_BB(mp, flen);
85
86 /*
87 * Too small? Give up.
88 */
89 if (dlen < minlen) {
90 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
91 goto out_del_cursor;
92 }
93
94 /*
95 * If the extent is entirely outside of the range we are
96 * supposed to discard skip it. Do not bother to trim
97 * down partially overlapping ranges for now.
98 */
99 if (dbno + dlen < start || dbno > end) {
100 trace_xfs_discard_exclude(mp, agno, fbno, flen);
101 goto next_extent;
102 }
103
104 /*
105 * If any blocks in the range are still busy, skip the
106 * discard and try again the next time.
107 */
108 if (xfs_extent_busy_search(mp, agno, fbno, flen)) {
109 trace_xfs_discard_busy(mp, agno, fbno, flen);
110 goto next_extent;
111 }
112
113 trace_xfs_discard_extent(mp, agno, fbno, flen);
114 error = blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS, 0);
115 if (error)
116 goto out_del_cursor;
117 *blocks_trimmed += flen;
118
119 next_extent:
120 error = xfs_btree_decrement(cur, 0, &i);
121 if (error)
122 goto out_del_cursor;
123
124 if (fatal_signal_pending(current)) {
125 error = -ERESTARTSYS;
126 goto out_del_cursor;
127 }
128 }
129
130 out_del_cursor:
131 xfs_btree_del_cursor(cur, error);
132 xfs_buf_relse(agbp);
133 out_put_perag:
134 xfs_perag_put(pag);
135 return error;
136 }
137
138 /*
139 * trim a range of the filesystem.
140 *
141 * Note: the parameters passed from userspace are byte ranges into the
142 * filesystem which does not match to the format we use for filesystem block
143 * addressing. FSB addressing is sparse (AGNO|AGBNO), while the incoming format
144 * is a linear address range. Hence we need to use DADDR based conversions and
145 * comparisons for determining the correct offset and regions to trim.
146 */
147 int
xfs_ioc_trim(struct xfs_mount * mp,struct fstrim_range __user * urange)148 xfs_ioc_trim(
149 struct xfs_mount *mp,
150 struct fstrim_range __user *urange)
151 {
152 struct request_queue *q = bdev_get_queue(mp->m_ddev_targp->bt_bdev);
153 unsigned int granularity = q->limits.discard_granularity;
154 struct fstrim_range range;
155 xfs_daddr_t start, end, minlen;
156 xfs_agnumber_t start_agno, end_agno, agno;
157 uint64_t blocks_trimmed = 0;
158 int error, last_error = 0;
159
160 if (!capable(CAP_SYS_ADMIN))
161 return -EPERM;
162 if (!blk_queue_discard(q))
163 return -EOPNOTSUPP;
164 if (copy_from_user(&range, urange, sizeof(range)))
165 return -EFAULT;
166
167 /*
168 * Truncating down the len isn't actually quite correct, but using
169 * BBTOB would mean we trivially get overflows for values
170 * of ULLONG_MAX or slightly lower. And ULLONG_MAX is the default
171 * used by the fstrim application. In the end it really doesn't
172 * matter as trimming blocks is an advisory interface.
173 */
174 if (range.start >= XFS_FSB_TO_B(mp, mp->m_sb.sb_dblocks) ||
175 range.minlen > XFS_FSB_TO_B(mp, mp->m_ag_max_usable) ||
176 range.len < mp->m_sb.sb_blocksize)
177 return -EINVAL;
178
179 start = BTOBB(range.start);
180 end = start + BTOBBT(range.len) - 1;
181 minlen = BTOBB(max_t(u64, granularity, range.minlen));
182
183 if (end > XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) - 1)
184 end = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)- 1;
185
186 start_agno = xfs_daddr_to_agno(mp, start);
187 end_agno = xfs_daddr_to_agno(mp, end);
188
189 for (agno = start_agno; agno <= end_agno; agno++) {
190 error = xfs_trim_extents(mp, agno, start, end, minlen,
191 &blocks_trimmed);
192 if (error) {
193 last_error = error;
194 if (error == -ERESTARTSYS)
195 break;
196 }
197 }
198
199 if (last_error)
200 return last_error;
201
202 range.len = XFS_FSB_TO_B(mp, blocks_trimmed);
203 if (copy_to_user(urange, &range, sizeof(range)))
204 return -EFAULT;
205 return 0;
206 }
207