Loading...
1/*
2 * Copyright (C) 2010 Red Hat, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18#include "xfs.h"
19#include "xfs_format.h"
20#include "xfs_log_format.h"
21#include "xfs_trans_resv.h"
22#include "xfs_sb.h"
23#include "xfs_mount.h"
24#include "xfs_quota.h"
25#include "xfs_inode.h"
26#include "xfs_btree.h"
27#include "xfs_alloc_btree.h"
28#include "xfs_alloc.h"
29#include "xfs_error.h"
30#include "xfs_extent_busy.h"
31#include "xfs_discard.h"
32#include "xfs_trace.h"
33#include "xfs_log.h"
34
35STATIC int
36xfs_trim_extents(
37 struct xfs_mount *mp,
38 xfs_agnumber_t agno,
39 xfs_daddr_t start,
40 xfs_daddr_t end,
41 xfs_daddr_t minlen,
42 uint64_t *blocks_trimmed)
43{
44 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
45 struct xfs_btree_cur *cur;
46 struct xfs_buf *agbp;
47 struct xfs_perag *pag;
48 int error;
49 int i;
50
51 pag = xfs_perag_get(mp, agno);
52
53 /*
54 * Force out the log. This means any transactions that might have freed
55 * space before we take the AGF buffer lock are now on disk, and the
56 * volatile disk cache is flushed.
57 */
58 xfs_log_force(mp, XFS_LOG_SYNC);
59
60 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
61 if (error || !agbp)
62 goto out_put_perag;
63
64 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_CNT);
65
66 /*
67 * Look up the longest btree in the AGF and start with it.
68 */
69 error = xfs_alloc_lookup_ge(cur, 0,
70 be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest), &i);
71 if (error)
72 goto out_del_cursor;
73
74 /*
75 * Loop until we are done with all extents that are large
76 * enough to be worth discarding.
77 */
78 while (i) {
79 xfs_agblock_t fbno;
80 xfs_extlen_t flen;
81 xfs_daddr_t dbno;
82 xfs_extlen_t dlen;
83
84 error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
85 if (error)
86 goto out_del_cursor;
87 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, out_del_cursor);
88 ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
89
90 /*
91 * use daddr format for all range/len calculations as that is
92 * the format the range/len variables are supplied in by
93 * userspace.
94 */
95 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
96 dlen = XFS_FSB_TO_BB(mp, flen);
97
98 /*
99 * Too small? Give up.
100 */
101 if (dlen < minlen) {
102 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
103 goto out_del_cursor;
104 }
105
106 /*
107 * If the extent is entirely outside of the range we are
108 * supposed to discard skip it. Do not bother to trim
109 * down partially overlapping ranges for now.
110 */
111 if (dbno + dlen < start || dbno > end) {
112 trace_xfs_discard_exclude(mp, agno, fbno, flen);
113 goto next_extent;
114 }
115
116 /*
117 * If any blocks in the range are still busy, skip the
118 * discard and try again the next time.
119 */
120 if (xfs_extent_busy_search(mp, agno, fbno, flen)) {
121 trace_xfs_discard_busy(mp, agno, fbno, flen);
122 goto next_extent;
123 }
124
125 trace_xfs_discard_extent(mp, agno, fbno, flen);
126 error = blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS, 0);
127 if (error)
128 goto out_del_cursor;
129 *blocks_trimmed += flen;
130
131next_extent:
132 error = xfs_btree_decrement(cur, 0, &i);
133 if (error)
134 goto out_del_cursor;
135
136 if (fatal_signal_pending(current)) {
137 error = -ERESTARTSYS;
138 goto out_del_cursor;
139 }
140 }
141
142out_del_cursor:
143 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
144 xfs_buf_relse(agbp);
145out_put_perag:
146 xfs_perag_put(pag);
147 return error;
148}
149
150/*
151 * trim a range of the filesystem.
152 *
153 * Note: the parameters passed from userspace are byte ranges into the
154 * filesystem which does not match to the format we use for filesystem block
155 * addressing. FSB addressing is sparse (AGNO|AGBNO), while the incoming format
156 * is a linear address range. Hence we need to use DADDR based conversions and
157 * comparisons for determining the correct offset and regions to trim.
158 */
159int
160xfs_ioc_trim(
161 struct xfs_mount *mp,
162 struct fstrim_range __user *urange)
163{
164 struct request_queue *q = bdev_get_queue(mp->m_ddev_targp->bt_bdev);
165 unsigned int granularity = q->limits.discard_granularity;
166 struct fstrim_range range;
167 xfs_daddr_t start, end, minlen;
168 xfs_agnumber_t start_agno, end_agno, agno;
169 uint64_t blocks_trimmed = 0;
170 int error, last_error = 0;
171
172 if (!capable(CAP_SYS_ADMIN))
173 return -EPERM;
174 if (!blk_queue_discard(q))
175 return -EOPNOTSUPP;
176 if (copy_from_user(&range, urange, sizeof(range)))
177 return -EFAULT;
178
179 /*
180 * Truncating down the len isn't actually quite correct, but using
181 * BBTOB would mean we trivially get overflows for values
182 * of ULLONG_MAX or slightly lower. And ULLONG_MAX is the default
183 * used by the fstrim application. In the end it really doesn't
184 * matter as trimming blocks is an advisory interface.
185 */
186 if (range.start >= XFS_FSB_TO_B(mp, mp->m_sb.sb_dblocks) ||
187 range.minlen > XFS_FSB_TO_B(mp, mp->m_ag_max_usable) ||
188 range.len < mp->m_sb.sb_blocksize)
189 return -EINVAL;
190
191 start = BTOBB(range.start);
192 end = start + BTOBBT(range.len) - 1;
193 minlen = BTOBB(max_t(u64, granularity, range.minlen));
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}
1/*
2 * Copyright (C) 2010 Red Hat, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18#include "xfs.h"
19#include "xfs_format.h"
20#include "xfs_log_format.h"
21#include "xfs_trans_resv.h"
22#include "xfs_sb.h"
23#include "xfs_ag.h"
24#include "xfs_mount.h"
25#include "xfs_quota.h"
26#include "xfs_inode.h"
27#include "xfs_btree.h"
28#include "xfs_alloc_btree.h"
29#include "xfs_alloc.h"
30#include "xfs_error.h"
31#include "xfs_extent_busy.h"
32#include "xfs_discard.h"
33#include "xfs_trace.h"
34#include "xfs_log.h"
35
36STATIC int
37xfs_trim_extents(
38 struct xfs_mount *mp,
39 xfs_agnumber_t agno,
40 xfs_daddr_t start,
41 xfs_daddr_t end,
42 xfs_daddr_t minlen,
43 __uint64_t *blocks_trimmed)
44{
45 struct block_device *bdev = mp->m_ddev_targp->bt_bdev;
46 struct xfs_btree_cur *cur;
47 struct xfs_buf *agbp;
48 struct xfs_perag *pag;
49 int error;
50 int i;
51
52 pag = xfs_perag_get(mp, agno);
53
54 error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agbp);
55 if (error || !agbp)
56 goto out_put_perag;
57
58 cur = xfs_allocbt_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_CNT);
59
60 /*
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.
64 */
65 xfs_log_force(mp, XFS_LOG_SYNC);
66
67 /*
68 * Look up the longest btree in the AGF and start with it.
69 */
70 error = xfs_alloc_lookup_ge(cur, 0,
71 be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest), &i);
72 if (error)
73 goto out_del_cursor;
74
75 /*
76 * Loop until we are done with all extents that are large
77 * enough to be worth discarding.
78 */
79 while (i) {
80 xfs_agblock_t fbno;
81 xfs_extlen_t flen;
82 xfs_daddr_t dbno;
83 xfs_extlen_t dlen;
84
85 error = xfs_alloc_get_rec(cur, &fbno, &flen, &i);
86 if (error)
87 goto out_del_cursor;
88 XFS_WANT_CORRUPTED_GOTO(i == 1, out_del_cursor);
89 ASSERT(flen <= be32_to_cpu(XFS_BUF_TO_AGF(agbp)->agf_longest));
90
91 /*
92 * use daddr format for all range/len calculations as that is
93 * the format the range/len variables are supplied in by
94 * userspace.
95 */
96 dbno = XFS_AGB_TO_DADDR(mp, agno, fbno);
97 dlen = XFS_FSB_TO_BB(mp, flen);
98
99 /*
100 * Too small? Give up.
101 */
102 if (dlen < minlen) {
103 trace_xfs_discard_toosmall(mp, agno, fbno, flen);
104 goto out_del_cursor;
105 }
106
107 /*
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.
111 */
112 if (dbno + dlen < start || dbno > end) {
113 trace_xfs_discard_exclude(mp, agno, fbno, flen);
114 goto next_extent;
115 }
116
117 /*
118 * If any blocks in the range are still busy, skip the
119 * discard and try again the next time.
120 */
121 if (xfs_extent_busy_search(mp, agno, fbno, flen)) {
122 trace_xfs_discard_busy(mp, agno, fbno, flen);
123 goto next_extent;
124 }
125
126 trace_xfs_discard_extent(mp, agno, fbno, flen);
127 error = -blkdev_issue_discard(bdev, dbno, dlen, GFP_NOFS, 0);
128 if (error)
129 goto out_del_cursor;
130 *blocks_trimmed += flen;
131
132next_extent:
133 error = xfs_btree_decrement(cur, 0, &i);
134 if (error)
135 goto out_del_cursor;
136 }
137
138out_del_cursor:
139 xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
140 xfs_buf_relse(agbp);
141out_put_perag:
142 xfs_perag_put(pag);
143 return error;
144}
145
146/*
147 * trim a range of the filesystem.
148 *
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.
154 */
155int
156xfs_ioc_trim(
157 struct xfs_mount *mp,
158 struct fstrim_range __user *urange)
159{
160 struct request_queue *q = bdev_get_queue(mp->m_ddev_targp->bt_bdev);
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;
167
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);
174
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, XFS_ALLOC_AG_MAX_USABLE(mp)) ||
184 range.len < mp->m_sb.sb_blocksize)
185 return -XFS_ERROR(EINVAL);
186
187 start = BTOBB(range.start);
188 end = start + BTOBBT(range.len) - 1;
189 minlen = BTOBB(max_t(u64, granularity, range.minlen));
190
191 if (end > XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks) - 1)
192 end = XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)- 1;
193
194 start_agno = xfs_daddr_to_agno(mp, start);
195 end_agno = xfs_daddr_to_agno(mp, end);
196
197 for (agno = start_agno; agno <= end_agno; agno++) {
198 error = -xfs_trim_extents(mp, agno, start, end, minlen,
199 &blocks_trimmed);
200 if (error)
201 last_error = error;
202 }
203
204 if (last_error)
205 return last_error;
206
207 range.len = XFS_FSB_TO_B(mp, blocks_trimmed);
208 if (copy_to_user(urange, &range, sizeof(range)))
209 return -XFS_ERROR(EFAULT);
210 return 0;
211}
212
213int
214xfs_discard_extents(
215 struct xfs_mount *mp,
216 struct list_head *list)
217{
218 struct xfs_extent_busy *busyp;
219 int error = 0;
220
221 list_for_each_entry(busyp, list, list) {
222 trace_xfs_discard_extent(mp, busyp->agno, busyp->bno,
223 busyp->length);
224
225 error = -blkdev_issue_discard(mp->m_ddev_targp->bt_bdev,
226 XFS_AGB_TO_DADDR(mp, busyp->agno, busyp->bno),
227 XFS_FSB_TO_BB(mp, busyp->length),
228 GFP_NOFS, 0);
229 if (error && error != EOPNOTSUPP) {
230 xfs_info(mp,
231 "discard failed for extent [0x%llu,%u], error %d",
232 (unsigned long long)busyp->bno,
233 busyp->length,
234 error);
235 return error;
236 }
237 }
238
239 return 0;
240}