2 * Copyright (C) 2010 Red Hat, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "xfs_mount.h"
23 #include "xfs_quota.h"
24 #include "xfs_trans.h"
25 #include "xfs_alloc_btree.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_ialloc_btree.h"
28 #include "xfs_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
32 #include "xfs_extent_busy.h"
33 #include "xfs_discard.h"
34 #include "xfs_trace.h"
43 __uint64_t *blocks_trimmed)
45 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
46 struct xfs_btree_cur *cur;
48 struct xfs_perag *pag;
52 pag = xfs_perag_get(mp, agno);
54 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
58 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_CNT);
61 * Force out the log. This means any transactions that might have freed
62 * space before we took the AGF buffer lock are now on disk, and the
63 * volatile disk cache is flushed.
65 xfs_log_force(mp, XFS_LOG_SYNC);
68 * Look up the longest btree in the AGF and start with it.
70 error = xfs_alloc_lookup_ge(cur, 0,
71 be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest), &i);
76 * Loop until we are done with all extents that are large
77 * enough to be worth discarding.
85 error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
88 XFS_WANT_CORRUPTED_GOTO(i == 1, out_del_cursor);
89 ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
92 * use daddr format for all range/len calculations as that is
93 * the format the range/len variables are supplied in by
96 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
97 dlen = XFS_FSB_TO_BB(mp, flen);
100 * Too small? Give up.
103 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
108 * If the extent is entirely outside of the range we are
109 * supposed to discard skip it. Do not bother to trim
110 * down partially overlapping ranges for now.
112 if (dbno + dlen < start || dbno > end) {
113 trace_xfs_discard_exclude(mp, agno, fbno, flen);
118 * If any blocks in the range are still busy, skip the
119 * discard and try again the next time.
121 if (xfs_extent_busy_search(mp, agno, fbno, flen)) {
122 trace_xfs_discard_busy(mp, agno, fbno, flen);
126 trace_xfs_discard_extent(mp, agno, fbno, flen);
127 error = -blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS, 0);
130 *blocks_trimmed += flen;
133 error = xfs_btree_decrement(cur, 0, &i);
139 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
147 * trim a range of the filesystem.
149 * Note: the parameters passed from userspace are byte ranges into the
150 * filesystem which does not match to the format we use for filesystem block
151 * addressing. FSB addressing is sparse (AGNO|AGBNO), while the incoming format
152 * is a linear address range. Hence we need to use DADDR based conversions and
153 * comparisons for determining the correct offset and regions to trim.
157 struct xfs_mount *mp,
158 struct fstrim_range __user *urange)
160 struct request_queue *q = mp->m_ddev_targp->bt_bdev->bd_disk->queue;
161 unsigned int granularity = q->limits.discard_granularity;
162 struct fstrim_range range;
163 xfs_daddr_t start, end, minlen;
164 xfs_agnumber_t start_agno, end_agno, agno;
165 __uint64_t blocks_trimmed = 0;
166 int error, last_error = 0;
168 if (!capable(CAP_SYS_ADMIN))
169 return -XFS_ERROR(EPERM);
170 if (!blk_queue_discard(q))
171 return -XFS_ERROR(EOPNOTSUPP);
172 if (copy_from_user(&range, urange, sizeof(range)))
173 return -XFS_ERROR(EFAULT);
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, XFS_ALLOC_AG_MAX_USABLE(mp)))
184 return -XFS_ERROR(EINVAL);
186 start = BTOBB(range.start);
187 end = start + BTOBBT(range.len) - 1;
188 minlen = BTOBB(max_t(u64, granularity, range.minlen));
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,
206 range.len = XFS_FSB_TO_B(mp, blocks_trimmed);
207 if (copy_to_user(urange, &range, sizeof(range)))
208 return -XFS_ERROR(EFAULT);
214 struct xfs_mount *mp,
215 struct list_head *list)
217 struct xfs_extent_busy *busyp;
220 list_for_each_entry(busyp, list, list) {
221 trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
224 error = -blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
225 XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
226 XFS_FSB_TO_BB(mp, busyp->length),
228 if (error && error != EOPNOTSUPP) {
230 "discard failed for extent [0x%llu,%u], error %d",
231 (unsigned long long)busyp->bno,