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