]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
70a9883c | 8 | #include "xfs_shared.h" |
239880ef DC |
9 | #include "xfs_format.h" |
10 | #include "xfs_log_format.h" | |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_sb.h" |
1da177e4 | 14 | #include "xfs_mount.h" |
1da177e4 | 15 | #include "xfs_inode.h" |
a4fbe6ab | 16 | #include "xfs_dir2.h" |
a844f451 | 17 | #include "xfs_ialloc.h" |
1da177e4 LT |
18 | #include "xfs_alloc.h" |
19 | #include "xfs_rtalloc.h" | |
20 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
21 | #include "xfs_trans.h" |
22 | #include "xfs_trans_priv.h" | |
23 | #include "xfs_log.h" | |
1da177e4 | 24 | #include "xfs_error.h" |
1da177e4 LT |
25 | #include "xfs_quota.h" |
26 | #include "xfs_fsops.h" | |
6d8b79cf | 27 | #include "xfs_icache.h" |
a31b1d3d | 28 | #include "xfs_sysfs.h" |
035e00ac | 29 | #include "xfs_rmap_btree.h" |
1946b91c | 30 | #include "xfs_refcount_btree.h" |
174edb0e | 31 | #include "xfs_reflink.h" |
ebf55872 | 32 | #include "xfs_extent_busy.h" |
39353ff6 | 33 | #include "xfs_health.h" |
13eaec4b | 34 | #include "xfs_trace.h" |
9bbafc71 | 35 | #include "xfs_ag.h" |
1da177e4 | 36 | |
27174203 CH |
37 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
38 | static int xfs_uuid_table_size; | |
39 | static uuid_t *xfs_uuid_table; | |
40 | ||
af3b6382 DW |
41 | void |
42 | xfs_uuid_table_free(void) | |
43 | { | |
44 | if (xfs_uuid_table_size == 0) | |
45 | return; | |
46 | kmem_free(xfs_uuid_table); | |
47 | xfs_uuid_table = NULL; | |
48 | xfs_uuid_table_size = 0; | |
49 | } | |
50 | ||
27174203 CH |
51 | /* |
52 | * See if the UUID is unique among mounted XFS filesystems. | |
53 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
54 | */ | |
55 | STATIC int | |
56 | xfs_uuid_mount( | |
57 | struct xfs_mount *mp) | |
58 | { | |
59 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
60 | int hole, i; | |
61 | ||
8f720d9f | 62 | /* Publish UUID in struct super_block */ |
85787090 | 63 | uuid_copy(&mp->m_super->s_uuid, uuid); |
8f720d9f | 64 | |
27174203 CH |
65 | if (mp->m_flags & XFS_MOUNT_NOUUID) |
66 | return 0; | |
67 | ||
d905fdaa AG |
68 | if (uuid_is_null(uuid)) { |
69 | xfs_warn(mp, "Filesystem has null UUID - can't mount"); | |
2451337d | 70 | return -EINVAL; |
27174203 CH |
71 | } |
72 | ||
73 | mutex_lock(&xfs_uuid_table_mutex); | |
74 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 75 | if (uuid_is_null(&xfs_uuid_table[i])) { |
27174203 CH |
76 | hole = i; |
77 | continue; | |
78 | } | |
79 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
80 | goto out_duplicate; | |
81 | } | |
82 | ||
83 | if (hole < 0) { | |
771915c4 | 84 | xfs_uuid_table = krealloc(xfs_uuid_table, |
27174203 | 85 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), |
771915c4 | 86 | GFP_KERNEL | __GFP_NOFAIL); |
27174203 CH |
87 | hole = xfs_uuid_table_size++; |
88 | } | |
89 | xfs_uuid_table[hole] = *uuid; | |
90 | mutex_unlock(&xfs_uuid_table_mutex); | |
91 | ||
92 | return 0; | |
93 | ||
94 | out_duplicate: | |
95 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 96 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 97 | return -EINVAL; |
27174203 CH |
98 | } |
99 | ||
100 | STATIC void | |
101 | xfs_uuid_unmount( | |
102 | struct xfs_mount *mp) | |
103 | { | |
104 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
105 | int i; | |
106 | ||
107 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
108 | return; | |
109 | ||
110 | mutex_lock(&xfs_uuid_table_mutex); | |
111 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 112 | if (uuid_is_null(&xfs_uuid_table[i])) |
27174203 CH |
113 | continue; |
114 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
115 | continue; | |
116 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
117 | break; | |
118 | } | |
119 | ASSERT(i < xfs_uuid_table_size); | |
120 | mutex_unlock(&xfs_uuid_table_mutex); | |
121 | } | |
122 | ||
4cc929ee NS |
123 | /* |
124 | * Check size of device based on the (data/realtime) block count. | |
125 | * Note: this check is used by the growfs code as well as mount. | |
126 | */ | |
127 | int | |
128 | xfs_sb_validate_fsb_count( | |
129 | xfs_sb_t *sbp, | |
c8ce540d | 130 | uint64_t nblocks) |
4cc929ee NS |
131 | { |
132 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
133 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
134 | ||
d5cf09ba | 135 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 136 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 137 | return -EFBIG; |
4cc929ee NS |
138 | return 0; |
139 | } | |
1da177e4 | 140 | |
1da177e4 LT |
141 | /* |
142 | * xfs_readsb | |
143 | * | |
144 | * Does the initial read of the superblock. | |
145 | */ | |
146 | int | |
ff55068c DC |
147 | xfs_readsb( |
148 | struct xfs_mount *mp, | |
149 | int flags) | |
1da177e4 LT |
150 | { |
151 | unsigned int sector_size; | |
04a1e6c5 DC |
152 | struct xfs_buf *bp; |
153 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 154 | int error; |
af34e09d | 155 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 156 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
157 | |
158 | ASSERT(mp->m_sb_bp == NULL); | |
159 | ASSERT(mp->m_ddev_targp != NULL); | |
160 | ||
daba5427 ES |
161 | /* |
162 | * For the initial read, we must guess at the sector | |
163 | * size based on the block device. It's enough to | |
164 | * get the sb_sectsize out of the superblock and | |
165 | * then reread with the proper length. | |
166 | * We don't verify it yet, because it may not be complete. | |
167 | */ | |
168 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
169 | buf_ops = NULL; | |
170 | ||
1da177e4 | 171 | /* |
c891c30a BF |
172 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
173 | * around at all times to optimize access to the superblock. Therefore, | |
174 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
175 | * elevated. | |
1da177e4 | 176 | */ |
26af6552 | 177 | reread: |
ba372674 | 178 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
179 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
180 | buf_ops); | |
ba372674 | 181 | if (error) { |
eab4e633 | 182 | if (loud) |
e721f504 | 183 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 184 | /* bad CRC means corrupted metadata */ |
2451337d DC |
185 | if (error == -EFSBADCRC) |
186 | error = -EFSCORRUPTED; | |
ba372674 | 187 | return error; |
eab4e633 | 188 | } |
1da177e4 LT |
189 | |
190 | /* | |
191 | * Initialize the mount structure from the superblock. | |
1da177e4 | 192 | */ |
3e6e8afd | 193 | xfs_sb_from_disk(sbp, bp->b_addr); |
556b8883 DC |
194 | |
195 | /* | |
196 | * If we haven't validated the superblock, do so now before we try | |
197 | * to check the sector size and reread the superblock appropriately. | |
198 | */ | |
199 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
200 | if (loud) | |
201 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 202 | error = -EINVAL; |
556b8883 DC |
203 | goto release_buf; |
204 | } | |
ff55068c | 205 | |
1da177e4 LT |
206 | /* |
207 | * We must be able to do sector-sized and sector-aligned IO. | |
208 | */ | |
04a1e6c5 | 209 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
210 | if (loud) |
211 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 212 | sector_size, sbp->sb_sectsize); |
2451337d | 213 | error = -ENOSYS; |
26af6552 | 214 | goto release_buf; |
1da177e4 LT |
215 | } |
216 | ||
daba5427 | 217 | if (buf_ops == NULL) { |
556b8883 DC |
218 | /* |
219 | * Re-read the superblock so the buffer is correctly sized, | |
220 | * and properly verified. | |
221 | */ | |
1da177e4 | 222 | xfs_buf_relse(bp); |
04a1e6c5 | 223 | sector_size = sbp->sb_sectsize; |
daba5427 | 224 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 225 | goto reread; |
1da177e4 LT |
226 | } |
227 | ||
5681ca40 | 228 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 229 | |
04a1e6c5 DC |
230 | /* no need to be quiet anymore, so reset the buf ops */ |
231 | bp->b_ops = &xfs_sb_buf_ops; | |
232 | ||
1da177e4 | 233 | mp->m_sb_bp = bp; |
26af6552 | 234 | xfs_buf_unlock(bp); |
1da177e4 LT |
235 | return 0; |
236 | ||
26af6552 DC |
237 | release_buf: |
238 | xfs_buf_relse(bp); | |
1da177e4 LT |
239 | return error; |
240 | } | |
241 | ||
13eaec4b DW |
242 | /* |
243 | * If the sunit/swidth change would move the precomputed root inode value, we | |
244 | * must reject the ondisk change because repair will stumble over that. | |
245 | * However, we allow the mount to proceed because we never rejected this | |
246 | * combination before. Returns true to update the sb, false otherwise. | |
247 | */ | |
248 | static inline int | |
249 | xfs_check_new_dalign( | |
250 | struct xfs_mount *mp, | |
251 | int new_dalign, | |
252 | bool *update_sb) | |
253 | { | |
254 | struct xfs_sb *sbp = &mp->m_sb; | |
255 | xfs_ino_t calc_ino; | |
256 | ||
257 | calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign); | |
258 | trace_xfs_check_new_dalign(mp, new_dalign, calc_ino); | |
259 | ||
260 | if (sbp->sb_rootino == calc_ino) { | |
261 | *update_sb = true; | |
262 | return 0; | |
263 | } | |
264 | ||
265 | xfs_warn(mp, | |
266 | "Cannot change stripe alignment; would require moving root inode."); | |
267 | ||
268 | /* | |
269 | * XXX: Next time we add a new incompat feature, this should start | |
270 | * returning -EINVAL to fail the mount. Until then, spit out a warning | |
271 | * that we're ignoring the administrator's instructions. | |
272 | */ | |
273 | xfs_warn(mp, "Skipping superblock stripe alignment update."); | |
274 | *update_sb = false; | |
275 | return 0; | |
276 | } | |
277 | ||
1da177e4 | 278 | /* |
4f5b1b3a DW |
279 | * If we were provided with new sunit/swidth values as mount options, make sure |
280 | * that they pass basic alignment and superblock feature checks, and convert | |
281 | * them into the same units (FSB) that everything else expects. This step | |
282 | * /must/ be done before computing the inode geometry. | |
1da177e4 | 283 | */ |
0771fb45 | 284 | STATIC int |
4f5b1b3a DW |
285 | xfs_validate_new_dalign( |
286 | struct xfs_mount *mp) | |
1da177e4 | 287 | { |
4f5b1b3a DW |
288 | if (mp->m_dalign == 0) |
289 | return 0; | |
1da177e4 | 290 | |
4f5b1b3a DW |
291 | /* |
292 | * If stripe unit and stripe width are not multiples | |
293 | * of the fs blocksize turn off alignment. | |
294 | */ | |
295 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
296 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
297 | xfs_warn(mp, | |
298 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
299 | mp->m_sb.sb_blocksize); | |
300 | return -EINVAL; | |
301 | } else { | |
1da177e4 | 302 | /* |
4f5b1b3a | 303 | * Convert the stripe unit and width to FSBs. |
1da177e4 | 304 | */ |
4f5b1b3a DW |
305 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); |
306 | if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) { | |
39a45d84 | 307 | xfs_warn(mp, |
4f5b1b3a DW |
308 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
309 | mp->m_sb.sb_agblocks); | |
2451337d | 310 | return -EINVAL; |
4f5b1b3a DW |
311 | } else if (mp->m_dalign) { |
312 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
34d7f603 JL |
313 | } else { |
314 | xfs_warn(mp, | |
4f5b1b3a DW |
315 | "alignment check failed: sunit(%d) less than bsize(%d)", |
316 | mp->m_dalign, mp->m_sb.sb_blocksize); | |
2451337d | 317 | return -EINVAL; |
1da177e4 | 318 | } |
4f5b1b3a DW |
319 | } |
320 | ||
321 | if (!xfs_sb_version_hasdalign(&mp->m_sb)) { | |
322 | xfs_warn(mp, | |
323 | "cannot change alignment: superblock does not support data alignment"); | |
324 | return -EINVAL; | |
325 | } | |
326 | ||
327 | return 0; | |
328 | } | |
329 | ||
330 | /* Update alignment values based on mount options and sb values. */ | |
331 | STATIC int | |
332 | xfs_update_alignment( | |
333 | struct xfs_mount *mp) | |
334 | { | |
335 | struct xfs_sb *sbp = &mp->m_sb; | |
336 | ||
337 | if (mp->m_dalign) { | |
13eaec4b DW |
338 | bool update_sb; |
339 | int error; | |
340 | ||
4f5b1b3a DW |
341 | if (sbp->sb_unit == mp->m_dalign && |
342 | sbp->sb_width == mp->m_swidth) | |
343 | return 0; | |
344 | ||
13eaec4b DW |
345 | error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb); |
346 | if (error || !update_sb) | |
347 | return error; | |
348 | ||
4f5b1b3a DW |
349 | sbp->sb_unit = mp->m_dalign; |
350 | sbp->sb_width = mp->m_swidth; | |
351 | mp->m_update_sb = true; | |
1da177e4 | 352 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && |
62118709 | 353 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
4f5b1b3a DW |
354 | mp->m_dalign = sbp->sb_unit; |
355 | mp->m_swidth = sbp->sb_width; | |
1da177e4 LT |
356 | } |
357 | ||
0771fb45 ES |
358 | return 0; |
359 | } | |
1da177e4 | 360 | |
055388a3 DC |
361 | /* |
362 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
363 | */ | |
364 | void | |
365 | xfs_set_low_space_thresholds( | |
366 | struct xfs_mount *mp) | |
367 | { | |
65f03d86 DW |
368 | uint64_t dblocks = mp->m_sb.sb_dblocks; |
369 | uint64_t rtexts = mp->m_sb.sb_rextents; | |
370 | int i; | |
055388a3 | 371 | |
65f03d86 DW |
372 | do_div(dblocks, 100); |
373 | do_div(rtexts, 100); | |
055388a3 | 374 | |
65f03d86 DW |
375 | for (i = 0; i < XFS_LOWSP_MAX; i++) { |
376 | mp->m_low_space[i] = dblocks * (i + 1); | |
377 | mp->m_low_rtexts[i] = rtexts * (i + 1); | |
055388a3 DC |
378 | } |
379 | } | |
380 | ||
0771fb45 | 381 | /* |
0471f62e | 382 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
383 | */ |
384 | STATIC int | |
ba372674 DC |
385 | xfs_check_sizes( |
386 | struct xfs_mount *mp) | |
0771fb45 | 387 | { |
ba372674 | 388 | struct xfs_buf *bp; |
0771fb45 | 389 | xfs_daddr_t d; |
ba372674 | 390 | int error; |
0771fb45 | 391 | |
1da177e4 LT |
392 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
393 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 394 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 395 | return -EFBIG; |
1da177e4 | 396 | } |
ba372674 | 397 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 398 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
399 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
400 | if (error) { | |
0b932ccc | 401 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 402 | return error; |
1da177e4 | 403 | } |
1922c949 | 404 | xfs_buf_relse(bp); |
1da177e4 | 405 | |
ba372674 DC |
406 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
407 | return 0; | |
408 | ||
409 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
410 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
411 | xfs_warn(mp, "log size mismatch detected"); | |
412 | return -EFBIG; | |
413 | } | |
414 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 415 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
416 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
417 | if (error) { | |
418 | xfs_warn(mp, "log device read failed"); | |
419 | return error; | |
0771fb45 | 420 | } |
ba372674 | 421 | xfs_buf_relse(bp); |
0771fb45 ES |
422 | return 0; |
423 | } | |
424 | ||
7d095257 CH |
425 | /* |
426 | * Clear the quotaflags in memory and in the superblock. | |
427 | */ | |
428 | int | |
429 | xfs_mount_reset_sbqflags( | |
430 | struct xfs_mount *mp) | |
431 | { | |
7d095257 CH |
432 | mp->m_qflags = 0; |
433 | ||
61e63ecb | 434 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
435 | if (mp->m_sb.sb_qflags == 0) |
436 | return 0; | |
437 | spin_lock(&mp->m_sb_lock); | |
438 | mp->m_sb.sb_qflags = 0; | |
439 | spin_unlock(&mp->m_sb_lock); | |
440 | ||
61e63ecb | 441 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
442 | return 0; |
443 | ||
61e63ecb | 444 | return xfs_sync_sb(mp, false); |
7d095257 CH |
445 | } |
446 | ||
c8ce540d | 447 | uint64_t |
d5db0f97 ES |
448 | xfs_default_resblks(xfs_mount_t *mp) |
449 | { | |
c8ce540d | 450 | uint64_t resblks; |
d5db0f97 ES |
451 | |
452 | /* | |
8babd8a2 DC |
453 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
454 | * smaller. This is intended to cover concurrent allocation | |
455 | * transactions when we initially hit enospc. These each require a 4 | |
456 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
457 | * allocation reservations. | |
d5db0f97 ES |
458 | */ |
459 | resblks = mp->m_sb.sb_dblocks; | |
460 | do_div(resblks, 20); | |
c8ce540d | 461 | resblks = min_t(uint64_t, resblks, 8192); |
d5db0f97 ES |
462 | return resblks; |
463 | } | |
464 | ||
2e9e6481 DW |
465 | /* Ensure the summary counts are correct. */ |
466 | STATIC int | |
467 | xfs_check_summary_counts( | |
468 | struct xfs_mount *mp) | |
469 | { | |
470 | /* | |
471 | * The AG0 superblock verifier rejects in-progress filesystems, | |
472 | * so we should never see the flag set this far into mounting. | |
473 | */ | |
474 | if (mp->m_sb.sb_inprogress) { | |
475 | xfs_err(mp, "sb_inprogress set after log recovery??"); | |
476 | WARN_ON(1); | |
477 | return -EFSCORRUPTED; | |
478 | } | |
479 | ||
480 | /* | |
481 | * Now the log is mounted, we know if it was an unclean shutdown or | |
482 | * not. If it was, with the first phase of recovery has completed, we | |
483 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
484 | * but they are recovered transactionally in the second recovery phase | |
485 | * later. | |
486 | * | |
487 | * If the log was clean when we mounted, we can check the summary | |
488 | * counters. If any of them are obviously incorrect, we can recompute | |
489 | * them from the AGF headers in the next step. | |
490 | */ | |
491 | if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
492 | (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks || | |
00d22a1c | 493 | !xfs_verify_icount(mp, mp->m_sb.sb_icount) || |
2e9e6481 | 494 | mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) |
39353ff6 | 495 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
2e9e6481 DW |
496 | |
497 | /* | |
498 | * We can safely re-initialise incore superblock counters from the | |
499 | * per-ag data. These may not be correct if the filesystem was not | |
500 | * cleanly unmounted, so we waited for recovery to finish before doing | |
501 | * this. | |
502 | * | |
503 | * If the filesystem was cleanly unmounted or the previous check did | |
504 | * not flag anything weird, then we can trust the values in the | |
505 | * superblock to be correct and we don't need to do anything here. | |
506 | * Otherwise, recalculate the summary counters. | |
507 | */ | |
508 | if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
509 | XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) && | |
39353ff6 | 510 | !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) |
2e9e6481 DW |
511 | return 0; |
512 | ||
513 | return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount); | |
514 | } | |
515 | ||
d336f7eb DW |
516 | /* |
517 | * Flush and reclaim dirty inodes in preparation for unmount. Inodes and | |
518 | * internal inode structures can be sitting in the CIL and AIL at this point, | |
519 | * so we need to unpin them, write them back and/or reclaim them before unmount | |
ab23a776 DC |
520 | * can proceed. In other words, callers are required to have inactivated all |
521 | * inodes. | |
d336f7eb DW |
522 | * |
523 | * An inode cluster that has been freed can have its buffer still pinned in | |
524 | * memory because the transaction is still sitting in a iclog. The stale inodes | |
525 | * on that buffer will be pinned to the buffer until the transaction hits the | |
526 | * disk and the callbacks run. Pushing the AIL will skip the stale inodes and | |
527 | * may never see the pinned buffer, so nothing will push out the iclog and | |
528 | * unpin the buffer. | |
529 | * | |
530 | * Hence we need to force the log to unpin everything first. However, log | |
531 | * forces don't wait for the discards they issue to complete, so we have to | |
532 | * explicitly wait for them to complete here as well. | |
533 | * | |
534 | * Then we can tell the world we are unmounting so that error handling knows | |
535 | * that the filesystem is going away and we should error out anything that we | |
536 | * have been retrying in the background. This will prevent never-ending | |
537 | * retries in AIL pushing from hanging the unmount. | |
538 | * | |
539 | * Finally, we can push the AIL to clean all the remaining dirty objects, then | |
540 | * reclaim the remaining inodes that are still in memory at this point in time. | |
541 | */ | |
542 | static void | |
543 | xfs_unmount_flush_inodes( | |
544 | struct xfs_mount *mp) | |
545 | { | |
546 | xfs_log_force(mp, XFS_LOG_SYNC); | |
547 | xfs_extent_busy_wait_all(mp); | |
548 | flush_workqueue(xfs_discard_wq); | |
549 | ||
550 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
551 | ||
552 | xfs_ail_push_all_sync(mp->m_ail); | |
ab23a776 | 553 | xfs_inodegc_stop(mp); |
d336f7eb DW |
554 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
555 | xfs_reclaim_inodes(mp); | |
556 | xfs_health_unmount(mp); | |
557 | } | |
558 | ||
b2941046 DC |
559 | static void |
560 | xfs_mount_setup_inode_geom( | |
561 | struct xfs_mount *mp) | |
562 | { | |
563 | struct xfs_ino_geometry *igeo = M_IGEO(mp); | |
564 | ||
565 | igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp); | |
566 | ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp)); | |
567 | ||
568 | xfs_ialloc_setup_geometry(mp); | |
569 | } | |
570 | ||
0771fb45 | 571 | /* |
0771fb45 ES |
572 | * This function does the following on an initial mount of a file system: |
573 | * - reads the superblock from disk and init the mount struct | |
574 | * - if we're a 32-bit kernel, do a size check on the superblock | |
575 | * so we don't mount terabyte filesystems | |
576 | * - init mount struct realtime fields | |
577 | * - allocate inode hash table for fs | |
578 | * - init directory manager | |
579 | * - perform recovery and init the log manager | |
580 | */ | |
581 | int | |
582 | xfs_mountfs( | |
f0b2efad | 583 | struct xfs_mount *mp) |
0771fb45 | 584 | { |
f0b2efad BF |
585 | struct xfs_sb *sbp = &(mp->m_sb); |
586 | struct xfs_inode *rip; | |
ef325959 | 587 | struct xfs_ino_geometry *igeo = M_IGEO(mp); |
c8ce540d | 588 | uint64_t resblks; |
f0b2efad BF |
589 | uint quotamount = 0; |
590 | uint quotaflags = 0; | |
591 | int error = 0; | |
0771fb45 | 592 | |
ff55068c | 593 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 594 | |
ee1c0908 | 595 | /* |
074e427b DC |
596 | * Check for a mismatched features2 values. Older kernels read & wrote |
597 | * into the wrong sb offset for sb_features2 on some platforms due to | |
598 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
599 | * which made older superblock reading/writing routines swap it as a | |
600 | * 64-bit value. | |
ee1c0908 | 601 | * |
e6957ea4 ES |
602 | * For backwards compatibility, we make both slots equal. |
603 | * | |
074e427b DC |
604 | * If we detect a mismatched field, we OR the set bits into the existing |
605 | * features2 field in case it has already been modified; we don't want | |
606 | * to lose any features. We then update the bad location with the ORed | |
607 | * value so that older kernels will see any features2 flags. The | |
608 | * superblock writeback code ensures the new sb_features2 is copied to | |
609 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 610 | */ |
e6957ea4 | 611 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 612 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 613 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 614 | mp->m_update_sb = true; |
e6957ea4 ES |
615 | |
616 | /* | |
617 | * Re-check for ATTR2 in case it was found in bad_features2 | |
618 | * slot. | |
619 | */ | |
7c12f296 TS |
620 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
621 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 622 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
623 | } |
624 | ||
625 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
626 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
627 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 628 | mp->m_update_sb = true; |
e6957ea4 | 629 | |
7c12f296 TS |
630 | /* update sb_versionnum for the clearing of the morebits */ |
631 | if (!sbp->sb_features2) | |
61e63ecb | 632 | mp->m_update_sb = true; |
ee1c0908 DC |
633 | } |
634 | ||
263997a6 DC |
635 | /* always use v2 inodes by default now */ |
636 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
637 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 638 | mp->m_update_sb = true; |
263997a6 DC |
639 | } |
640 | ||
0771fb45 | 641 | /* |
4f5b1b3a DW |
642 | * If we were given new sunit/swidth options, do some basic validation |
643 | * checks and convert the incore dalign and swidth values to the | |
644 | * same units (FSB) that everything else uses. This /must/ happen | |
645 | * before computing the inode geometry. | |
0771fb45 | 646 | */ |
4f5b1b3a | 647 | error = xfs_validate_new_dalign(mp); |
0771fb45 | 648 | if (error) |
f9057e3d | 649 | goto out; |
0771fb45 ES |
650 | |
651 | xfs_alloc_compute_maxlevels(mp); | |
652 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
653 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
b2941046 | 654 | xfs_mount_setup_inode_geom(mp); |
035e00ac | 655 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 656 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 | 657 | |
4f5b1b3a DW |
658 | /* |
659 | * Check if sb_agblocks is aligned at stripe boundary. If sb_agblocks | |
660 | * is NOT aligned turn off m_dalign since allocator alignment is within | |
661 | * an ag, therefore ag has to be aligned at stripe boundary. Note that | |
662 | * we must compute the free space and rmap btree geometry before doing | |
663 | * this. | |
664 | */ | |
665 | error = xfs_update_alignment(mp); | |
666 | if (error) | |
667 | goto out; | |
668 | ||
e6b3bb78 | 669 | /* enable fail_at_unmount as default */ |
749f24f3 | 670 | mp->m_fail_unmount = true; |
e6b3bb78 | 671 | |
e1d3d218 IK |
672 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, |
673 | NULL, mp->m_super->s_id); | |
27174203 CH |
674 | if (error) |
675 | goto out; | |
1da177e4 | 676 | |
225e4635 BD |
677 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
678 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
679 | if (error) |
680 | goto out_remove_sysfs; | |
681 | ||
192852be | 682 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
683 | if (error) |
684 | goto out_del_stats; | |
685 | ||
31965ef3 DW |
686 | error = xfs_errortag_init(mp); |
687 | if (error) | |
688 | goto out_remove_error_sysfs; | |
192852be CM |
689 | |
690 | error = xfs_uuid_mount(mp); | |
691 | if (error) | |
31965ef3 | 692 | goto out_remove_errortag; |
192852be | 693 | |
0771fb45 | 694 | /* |
2fcddee8 CH |
695 | * Update the preferred write size based on the information from the |
696 | * on-disk superblock. | |
0771fb45 | 697 | */ |
2fcddee8 CH |
698 | mp->m_allocsize_log = |
699 | max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log); | |
700 | mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog); | |
0771fb45 | 701 | |
055388a3 DC |
702 | /* set the low space thresholds for dynamic preallocation */ |
703 | xfs_set_low_space_thresholds(mp); | |
704 | ||
e5376fc1 BF |
705 | /* |
706 | * If enabled, sparse inode chunk alignment is expected to match the | |
707 | * cluster size. Full inode chunk alignment must match the chunk size, | |
708 | * but that is checked on sb read verification... | |
709 | */ | |
710 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
711 | mp->m_sb.sb_spino_align != | |
490d451f | 712 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) { |
e5376fc1 BF |
713 | xfs_warn(mp, |
714 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
715 | mp->m_sb.sb_spino_align, | |
490d451f | 716 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)); |
e5376fc1 BF |
717 | error = -EINVAL; |
718 | goto out_remove_uuid; | |
719 | } | |
720 | ||
0771fb45 | 721 | /* |
c2bfbc9b | 722 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 723 | */ |
4249023a | 724 | error = xfs_check_sizes(mp); |
0771fb45 | 725 | if (error) |
f9057e3d | 726 | goto out_remove_uuid; |
0771fb45 | 727 | |
1da177e4 LT |
728 | /* |
729 | * Initialize realtime fields in the mount structure | |
730 | */ | |
0771fb45 ES |
731 | error = xfs_rtmount_init(mp); |
732 | if (error) { | |
0b932ccc | 733 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 734 | goto out_remove_uuid; |
1da177e4 LT |
735 | } |
736 | ||
1da177e4 LT |
737 | /* |
738 | * Copies the low order bits of the timestamp and the randomly | |
739 | * set "sequence" number out of a UUID. | |
740 | */ | |
cb0ba6cc CH |
741 | mp->m_fixedfsid[0] = |
742 | (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) | | |
743 | get_unaligned_be16(&sbp->sb_uuid.b[4]); | |
744 | mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]); | |
1da177e4 | 745 | |
0650b554 DC |
746 | error = xfs_da_mount(mp); |
747 | if (error) { | |
748 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
749 | goto out_remove_uuid; | |
750 | } | |
1da177e4 LT |
751 | |
752 | /* | |
753 | * Initialize the precomputed transaction reservations values. | |
754 | */ | |
755 | xfs_trans_init(mp); | |
756 | ||
1da177e4 LT |
757 | /* |
758 | * Allocate and initialize the per-ag data. | |
759 | */ | |
1c1c6ebc DC |
760 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
761 | if (error) { | |
0b932ccc | 762 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 763 | goto out_free_dir; |
1c1c6ebc | 764 | } |
1da177e4 | 765 | |
a71895c5 | 766 | if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) { |
0b932ccc | 767 | xfs_warn(mp, "no log defined"); |
2451337d | 768 | error = -EFSCORRUPTED; |
f9057e3d CH |
769 | goto out_free_perag; |
770 | } | |
771 | ||
40b1de00 DW |
772 | error = xfs_inodegc_register_shrinker(mp); |
773 | if (error) | |
774 | goto out_fail_wait; | |
775 | ||
1da177e4 | 776 | /* |
f0b2efad BF |
777 | * Log's mount-time initialization. The first part of recovery can place |
778 | * some items on the AIL, to be handled when recovery is finished or | |
779 | * cancelled. | |
1da177e4 | 780 | */ |
f9057e3d CH |
781 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
782 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
783 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
784 | if (error) { | |
0b932ccc | 785 | xfs_warn(mp, "log mount failed"); |
40b1de00 | 786 | goto out_inodegc_shrinker; |
1da177e4 LT |
787 | } |
788 | ||
2e9e6481 DW |
789 | /* Make sure the summary counts are ok. */ |
790 | error = xfs_check_summary_counts(mp); | |
791 | if (error) | |
792 | goto out_log_dealloc; | |
f9057e3d | 793 | |
ab23a776 DC |
794 | /* Enable background inode inactivation workers. */ |
795 | xfs_inodegc_start(mp); | |
6f649091 | 796 | xfs_blockgc_start(mp); |
ab23a776 | 797 | |
1da177e4 LT |
798 | /* |
799 | * Get and sanity-check the root inode. | |
800 | * Save the pointer to it in the mount structure. | |
801 | */ | |
541b5acc DC |
802 | error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED, |
803 | XFS_ILOCK_EXCL, &rip); | |
1da177e4 | 804 | if (error) { |
541b5acc DC |
805 | xfs_warn(mp, |
806 | "Failed to read root inode 0x%llx, error %d", | |
807 | sbp->sb_rootino, -error); | |
f9057e3d | 808 | goto out_log_dealloc; |
1da177e4 LT |
809 | } |
810 | ||
811 | ASSERT(rip != NULL); | |
1da177e4 | 812 | |
a71895c5 | 813 | if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 814 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 815 | (unsigned long long)rip->i_ino); |
1da177e4 | 816 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
2451337d | 817 | error = -EFSCORRUPTED; |
f9057e3d | 818 | goto out_rele_rip; |
1da177e4 LT |
819 | } |
820 | mp->m_rootip = rip; /* save it */ | |
821 | ||
822 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
823 | ||
824 | /* | |
825 | * Initialize realtime inode pointers in the mount structure | |
826 | */ | |
0771fb45 ES |
827 | error = xfs_rtmount_inodes(mp); |
828 | if (error) { | |
1da177e4 LT |
829 | /* |
830 | * Free up the root inode. | |
831 | */ | |
0b932ccc | 832 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 833 | goto out_rele_rip; |
1da177e4 LT |
834 | } |
835 | ||
836 | /* | |
7884bc86 CH |
837 | * If this is a read-only mount defer the superblock updates until |
838 | * the next remount into writeable mode. Otherwise we would never | |
839 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 840 | */ |
61e63ecb DC |
841 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
842 | error = xfs_sync_sb(mp, false); | |
e5720eec | 843 | if (error) { |
0b932ccc | 844 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 845 | goto out_rtunmount; |
e5720eec DC |
846 | } |
847 | } | |
1da177e4 LT |
848 | |
849 | /* | |
850 | * Initialise the XFS quota management subsystem for this mount | |
851 | */ | |
149e53af | 852 | if (XFS_IS_QUOTA_ON(mp)) { |
7d095257 CH |
853 | error = xfs_qm_newmount(mp, "amount, "aflags); |
854 | if (error) | |
855 | goto out_rtunmount; | |
856 | } else { | |
7d095257 CH |
857 | /* |
858 | * If a file system had quotas running earlier, but decided to | |
859 | * mount without -o uquota/pquota/gquota options, revoke the | |
860 | * quotachecked license. | |
861 | */ | |
862 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 863 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
864 | error = xfs_mount_reset_sbqflags(mp); |
865 | if (error) | |
a70a4fa5 | 866 | goto out_rtunmount; |
7d095257 CH |
867 | } |
868 | } | |
1da177e4 LT |
869 | |
870 | /* | |
f0b2efad BF |
871 | * Finish recovering the file system. This part needed to be delayed |
872 | * until after the root and real-time bitmap inodes were consistently | |
81ed9475 DW |
873 | * read in. Temporarily create per-AG space reservations for metadata |
874 | * btree shape changes because space freeing transactions (for inode | |
875 | * inactivation) require the per-AG reservation in lieu of reserving | |
876 | * blocks. | |
1da177e4 | 877 | */ |
81ed9475 DW |
878 | error = xfs_fs_reserve_ag_blocks(mp); |
879 | if (error && error == -ENOSPC) | |
880 | xfs_warn(mp, | |
881 | "ENOSPC reserving per-AG metadata pool, log recovery may fail."); | |
4249023a | 882 | error = xfs_log_mount_finish(mp); |
81ed9475 | 883 | xfs_fs_unreserve_ag_blocks(mp); |
1da177e4 | 884 | if (error) { |
0b932ccc | 885 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 886 | goto out_rtunmount; |
1da177e4 LT |
887 | } |
888 | ||
ddeb14f4 DC |
889 | /* |
890 | * Now the log is fully replayed, we can transition to full read-only | |
891 | * mode for read-only mounts. This will sync all the metadata and clean | |
892 | * the log so that the recovery we just performed does not have to be | |
893 | * replayed again on the next mount. | |
894 | * | |
895 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
896 | * semantically identical operations. | |
897 | */ | |
898 | if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) == | |
899 | XFS_MOUNT_RDONLY) { | |
ea2064da | 900 | xfs_log_clean(mp); |
ddeb14f4 DC |
901 | } |
902 | ||
1da177e4 LT |
903 | /* |
904 | * Complete the quota initialisation, post-log-replay component. | |
905 | */ | |
7d095257 CH |
906 | if (quotamount) { |
907 | ASSERT(mp->m_qflags == 0); | |
908 | mp->m_qflags = quotaflags; | |
909 | ||
910 | xfs_qm_mount_quotas(mp); | |
911 | } | |
912 | ||
84e1e99f DC |
913 | /* |
914 | * Now we are mounted, reserve a small amount of unused space for | |
915 | * privileged transactions. This is needed so that transaction | |
916 | * space required for critical operations can dip into this pool | |
917 | * when at ENOSPC. This is needed for operations like create with | |
918 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
919 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
920 | * |
921 | * This may drive us straight to ENOSPC on mount, but that implies | |
922 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 923 | */ |
d5db0f97 ES |
924 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
925 | resblks = xfs_default_resblks(mp); | |
926 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
927 | if (error) | |
0b932ccc DC |
928 | xfs_warn(mp, |
929 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
174edb0e DW |
930 | |
931 | /* Recover any CoW blocks that never got remapped. */ | |
932 | error = xfs_reflink_recover_cow(mp); | |
933 | if (error) { | |
934 | xfs_err(mp, | |
935 | "Error %d recovering leftover CoW allocations.", error); | |
936 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
937 | goto out_quota; | |
938 | } | |
84d69619 DW |
939 | |
940 | /* Reserve AG blocks for future btree expansion. */ | |
941 | error = xfs_fs_reserve_ag_blocks(mp); | |
942 | if (error && error != -ENOSPC) | |
943 | goto out_agresv; | |
d5db0f97 | 944 | } |
84e1e99f | 945 | |
1da177e4 LT |
946 | return 0; |
947 | ||
84d69619 DW |
948 | out_agresv: |
949 | xfs_fs_unreserve_ag_blocks(mp); | |
174edb0e DW |
950 | out_quota: |
951 | xfs_qm_unmount_quotas(mp); | |
b93b6e43 CH |
952 | out_rtunmount: |
953 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 954 | out_rele_rip: |
44a8736b | 955 | xfs_irele(rip); |
77aff8c7 DW |
956 | /* Clean out dquots that might be in memory after quotacheck. */ |
957 | xfs_qm_unmount(mp); | |
ab23a776 DC |
958 | |
959 | /* | |
960 | * Inactivate all inodes that might still be in memory after a log | |
961 | * intent recovery failure so that reclaim can free them. Metadata | |
962 | * inodes and the root directory shouldn't need inactivation, but the | |
963 | * mount failed for some reason, so pull down all the state and flee. | |
964 | */ | |
965 | xfs_inodegc_flush(mp); | |
966 | ||
2d1d1da3 | 967 | /* |
d336f7eb | 968 | * Flush all inode reclamation work and flush the log. |
2d1d1da3 DW |
969 | * We have to do this /after/ rtunmount and qm_unmount because those |
970 | * two will have scheduled delayed reclaim for the rt/quota inodes. | |
971 | * | |
972 | * This is slightly different from the unmountfs call sequence | |
973 | * because we could be tearing down a partially set up mount. In | |
974 | * particular, if log_mount_finish fails we bail out without calling | |
975 | * qm_unmount_quotas and therefore rely on qm_unmount to release the | |
976 | * quota inodes. | |
977 | */ | |
d336f7eb | 978 | xfs_unmount_flush_inodes(mp); |
f9057e3d | 979 | out_log_dealloc: |
f0b2efad | 980 | xfs_log_mount_cancel(mp); |
40b1de00 DW |
981 | out_inodegc_shrinker: |
982 | unregister_shrinker(&mp->m_inodegc_shrinker); | |
d4f3512b DC |
983 | out_fail_wait: |
984 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
10fb9ac1 BF |
985 | xfs_buftarg_drain(mp->m_logdev_targp); |
986 | xfs_buftarg_drain(mp->m_ddev_targp); | |
f9057e3d | 987 | out_free_perag: |
ff4f038c | 988 | xfs_free_perag(mp); |
0650b554 DC |
989 | out_free_dir: |
990 | xfs_da_unmount(mp); | |
f9057e3d | 991 | out_remove_uuid: |
27174203 | 992 | xfs_uuid_unmount(mp); |
31965ef3 DW |
993 | out_remove_errortag: |
994 | xfs_errortag_del(mp); | |
192852be CM |
995 | out_remove_error_sysfs: |
996 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
997 | out_del_stats: |
998 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
999 | out_remove_sysfs: |
1000 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1001 | out: |
1da177e4 LT |
1002 | return error; |
1003 | } | |
1004 | ||
1005 | /* | |
1da177e4 LT |
1006 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1007 | * log and makes sure that incore structures are freed. | |
1008 | */ | |
41b5c2e7 CH |
1009 | void |
1010 | xfs_unmountfs( | |
1011 | struct xfs_mount *mp) | |
1da177e4 | 1012 | { |
c8ce540d | 1013 | uint64_t resblks; |
41b5c2e7 | 1014 | int error; |
1da177e4 | 1015 | |
ab23a776 DC |
1016 | /* |
1017 | * Perform all on-disk metadata updates required to inactivate inodes | |
1018 | * that the VFS evicted earlier in the unmount process. Freeing inodes | |
1019 | * and discarding CoW fork preallocations can cause shape changes to | |
1020 | * the free inode and refcount btrees, respectively, so we must finish | |
1021 | * this before we discard the metadata space reservations. Metadata | |
1022 | * inodes and the root directory do not require inactivation. | |
1023 | */ | |
1024 | xfs_inodegc_flush(mp); | |
1025 | ||
c9a6526f | 1026 | xfs_blockgc_stop(mp); |
84d69619 | 1027 | xfs_fs_unreserve_ag_blocks(mp); |
7d095257 | 1028 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1029 | xfs_rtunmount_inodes(mp); |
44a8736b | 1030 | xfs_irele(mp->m_rootip); |
77508ec8 | 1031 | |
d336f7eb | 1032 | xfs_unmount_flush_inodes(mp); |
1da177e4 | 1033 | |
7d095257 | 1034 | xfs_qm_unmount(mp); |
a357a121 | 1035 | |
84e1e99f DC |
1036 | /* |
1037 | * Unreserve any blocks we have so that when we unmount we don't account | |
1038 | * the reserved free space as used. This is really only necessary for | |
1039 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1040 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1041 | * |
1042 | * We don't bother correcting this elsewhere for lazy superblock | |
1043 | * counting because on mount of an unclean filesystem we reconstruct the | |
1044 | * correct counter value and this is irrelevant. | |
1045 | * | |
1046 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1047 | * we only every apply deltas to the superblock and hence the incore | |
1048 | * value does not matter.... | |
1049 | */ | |
1050 | resblks = 0; | |
714082bc DC |
1051 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1052 | if (error) | |
0b932ccc | 1053 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1054 | "Freespace may not be correct on next mount."); |
1055 | ||
21b699c8 | 1056 | xfs_log_unmount(mp); |
0650b554 | 1057 | xfs_da_unmount(mp); |
27174203 | 1058 | xfs_uuid_unmount(mp); |
1da177e4 | 1059 | |
1550d0b0 | 1060 | #if defined(DEBUG) |
31965ef3 | 1061 | xfs_errortag_clearall(mp); |
1da177e4 | 1062 | #endif |
40b1de00 | 1063 | unregister_shrinker(&mp->m_inodegc_shrinker); |
ff4f038c | 1064 | xfs_free_perag(mp); |
a31b1d3d | 1065 | |
31965ef3 | 1066 | xfs_errortag_del(mp); |
192852be | 1067 | xfs_error_sysfs_del(mp); |
225e4635 | 1068 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1069 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1070 | } |
1071 | ||
91ee575f BF |
1072 | /* |
1073 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1074 | * freeze level for which modifications should not be allowed. This allows | |
1075 | * certain operations to proceed while the freeze sequence is in progress, if | |
1076 | * necessary. | |
1077 | */ | |
1078 | bool | |
1079 | xfs_fs_writable( | |
1080 | struct xfs_mount *mp, | |
1081 | int level) | |
92821e2b | 1082 | { |
91ee575f BF |
1083 | ASSERT(level > SB_UNFROZEN); |
1084 | if ((mp->m_super->s_writers.frozen >= level) || | |
1085 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1086 | return false; | |
1087 | ||
1088 | return true; | |
92821e2b DC |
1089 | } |
1090 | ||
0d485ada DC |
1091 | int |
1092 | xfs_mod_fdblocks( | |
1093 | struct xfs_mount *mp, | |
1094 | int64_t delta, | |
1095 | bool rsvd) | |
1096 | { | |
1097 | int64_t lcounter; | |
1098 | long long res_used; | |
1099 | s32 batch; | |
fd43cf60 | 1100 | uint64_t set_aside; |
0d485ada DC |
1101 | |
1102 | if (delta > 0) { | |
1103 | /* | |
1104 | * If the reserve pool is depleted, put blocks back into it | |
1105 | * first. Most of the time the pool is full. | |
1106 | */ | |
1107 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1108 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1109 | return 0; | |
1110 | } | |
1111 | ||
1112 | spin_lock(&mp->m_sb_lock); | |
1113 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1114 | ||
1115 | if (res_used > delta) { | |
1116 | mp->m_resblks_avail += delta; | |
1117 | } else { | |
1118 | delta -= res_used; | |
1119 | mp->m_resblks_avail = mp->m_resblks; | |
1120 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1121 | } | |
1122 | spin_unlock(&mp->m_sb_lock); | |
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * Taking blocks away, need to be more accurate the closer we | |
1128 | * are to zero. | |
1129 | * | |
0d485ada DC |
1130 | * If the counter has a value of less than 2 * max batch size, |
1131 | * then make everything serialise as we are real close to | |
1132 | * ENOSPC. | |
1133 | */ | |
8c1903d3 DC |
1134 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1135 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1136 | batch = 1; |
1137 | else | |
8c1903d3 | 1138 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada | 1139 | |
fd43cf60 BF |
1140 | /* |
1141 | * Set aside allocbt blocks because these blocks are tracked as free | |
1142 | * space but not available for allocation. Technically this means that a | |
1143 | * single reservation cannot consume all remaining free space, but the | |
1144 | * ratio of allocbt blocks to usable free blocks should be rather small. | |
1145 | * The tradeoff without this is that filesystems that maintain high | |
1146 | * perag block reservations can over reserve physical block availability | |
1147 | * and fail physical allocation, which leads to much more serious | |
1148 | * problems (i.e. transaction abort, pagecache discards, etc.) than | |
1149 | * slightly premature -ENOSPC. | |
1150 | */ | |
1151 | set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks); | |
104b4e51 | 1152 | percpu_counter_add_batch(&mp->m_fdblocks, delta, batch); |
fd43cf60 | 1153 | if (__percpu_counter_compare(&mp->m_fdblocks, set_aside, |
8c1903d3 | 1154 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1155 | /* we had space! */ |
1156 | return 0; | |
1157 | } | |
1158 | ||
1159 | /* | |
1160 | * lock up the sb for dipping into reserves before releasing the space | |
1161 | * that took us to ENOSPC. | |
1162 | */ | |
1163 | spin_lock(&mp->m_sb_lock); | |
1164 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1165 | if (!rsvd) | |
1166 | goto fdblocks_enospc; | |
1167 | ||
1168 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1169 | if (lcounter >= 0) { | |
1170 | mp->m_resblks_avail = lcounter; | |
1171 | spin_unlock(&mp->m_sb_lock); | |
1172 | return 0; | |
1173 | } | |
ec43f6da ES |
1174 | xfs_warn_once(mp, |
1175 | "Reserve blocks depleted! Consider increasing reserve pool size."); | |
1176 | ||
0d485ada DC |
1177 | fdblocks_enospc: |
1178 | spin_unlock(&mp->m_sb_lock); | |
1179 | return -ENOSPC; | |
1180 | } | |
1181 | ||
bab98bbe DC |
1182 | int |
1183 | xfs_mod_frextents( | |
1184 | struct xfs_mount *mp, | |
1185 | int64_t delta) | |
1186 | { | |
1187 | int64_t lcounter; | |
1188 | int ret = 0; | |
1189 | ||
1190 | spin_lock(&mp->m_sb_lock); | |
1191 | lcounter = mp->m_sb.sb_frextents + delta; | |
1192 | if (lcounter < 0) | |
1193 | ret = -ENOSPC; | |
1194 | else | |
1195 | mp->m_sb.sb_frextents = lcounter; | |
1196 | spin_unlock(&mp->m_sb_lock); | |
1197 | return ret; | |
1198 | } | |
1199 | ||
1da177e4 LT |
1200 | /* |
1201 | * Used to free the superblock along various error paths. | |
1202 | */ | |
1203 | void | |
1204 | xfs_freesb( | |
26af6552 | 1205 | struct xfs_mount *mp) |
1da177e4 | 1206 | { |
26af6552 | 1207 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1208 | |
26af6552 | 1209 | xfs_buf_lock(bp); |
1da177e4 | 1210 | mp->m_sb_bp = NULL; |
26af6552 | 1211 | xfs_buf_relse(bp); |
1da177e4 LT |
1212 | } |
1213 | ||
dda35b8f CH |
1214 | /* |
1215 | * If the underlying (data/log/rt) device is readonly, there are some | |
1216 | * operations that cannot proceed. | |
1217 | */ | |
1218 | int | |
1219 | xfs_dev_is_read_only( | |
1220 | struct xfs_mount *mp, | |
1221 | char *message) | |
1222 | { | |
1223 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1224 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1225 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1226 | xfs_notice(mp, "%s required on read-only device.", message); |
1227 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1228 | return -EROFS; |
dda35b8f CH |
1229 | } |
1230 | return 0; | |
1231 | } | |
f467cad9 DW |
1232 | |
1233 | /* Force the summary counters to be recalculated at next mount. */ | |
1234 | void | |
1235 | xfs_force_summary_recalc( | |
1236 | struct xfs_mount *mp) | |
1237 | { | |
1238 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1239 | return; | |
1240 | ||
39353ff6 | 1241 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
f467cad9 | 1242 | } |
9fe82b8c | 1243 | |
908ce71e DW |
1244 | /* |
1245 | * Enable a log incompat feature flag in the primary superblock. The caller | |
1246 | * cannot have any other transactions in progress. | |
1247 | */ | |
1248 | int | |
1249 | xfs_add_incompat_log_feature( | |
1250 | struct xfs_mount *mp, | |
1251 | uint32_t feature) | |
1252 | { | |
1253 | struct xfs_dsb *dsb; | |
1254 | int error; | |
1255 | ||
1256 | ASSERT(hweight32(feature) == 1); | |
1257 | ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN)); | |
1258 | ||
1259 | /* | |
1260 | * Force the log to disk and kick the background AIL thread to reduce | |
1261 | * the chances that the bwrite will stall waiting for the AIL to unpin | |
1262 | * the primary superblock buffer. This isn't a data integrity | |
1263 | * operation, so we don't need a synchronous push. | |
1264 | */ | |
1265 | error = xfs_log_force(mp, XFS_LOG_SYNC); | |
1266 | if (error) | |
1267 | return error; | |
1268 | xfs_ail_push_all(mp->m_ail); | |
1269 | ||
1270 | /* | |
1271 | * Lock the primary superblock buffer to serialize all callers that | |
1272 | * are trying to set feature bits. | |
1273 | */ | |
1274 | xfs_buf_lock(mp->m_sb_bp); | |
1275 | xfs_buf_hold(mp->m_sb_bp); | |
1276 | ||
1277 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
1278 | error = -EIO; | |
1279 | goto rele; | |
1280 | } | |
1281 | ||
1282 | if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature)) | |
1283 | goto rele; | |
1284 | ||
1285 | /* | |
1286 | * Write the primary superblock to disk immediately, because we need | |
1287 | * the log_incompat bit to be set in the primary super now to protect | |
1288 | * the log items that we're going to commit later. | |
1289 | */ | |
1290 | dsb = mp->m_sb_bp->b_addr; | |
1291 | xfs_sb_to_disk(dsb, &mp->m_sb); | |
1292 | dsb->sb_features_log_incompat |= cpu_to_be32(feature); | |
1293 | error = xfs_bwrite(mp->m_sb_bp); | |
1294 | if (error) | |
1295 | goto shutdown; | |
1296 | ||
1297 | /* | |
1298 | * Add the feature bits to the incore superblock before we unlock the | |
1299 | * buffer. | |
1300 | */ | |
1301 | xfs_sb_add_incompat_log_features(&mp->m_sb, feature); | |
1302 | xfs_buf_relse(mp->m_sb_bp); | |
1303 | ||
1304 | /* Log the superblock to disk. */ | |
1305 | return xfs_sync_sb(mp, false); | |
1306 | shutdown: | |
1307 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1308 | rele: | |
1309 | xfs_buf_relse(mp->m_sb_bp); | |
1310 | return error; | |
1311 | } | |
1312 | ||
1313 | /* | |
1314 | * Clear all the log incompat flags from the superblock. | |
1315 | * | |
1316 | * The caller cannot be in a transaction, must ensure that the log does not | |
1317 | * contain any log items protected by any log incompat bit, and must ensure | |
1318 | * that there are no other threads that depend on the state of the log incompat | |
1319 | * feature flags in the primary super. | |
1320 | * | |
1321 | * Returns true if the superblock is dirty. | |
1322 | */ | |
1323 | bool | |
1324 | xfs_clear_incompat_log_features( | |
1325 | struct xfs_mount *mp) | |
1326 | { | |
1327 | bool ret = false; | |
1328 | ||
1329 | if (!xfs_sb_version_hascrc(&mp->m_sb) || | |
1330 | !xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1331 | XFS_SB_FEAT_INCOMPAT_LOG_ALL) || | |
1332 | XFS_FORCED_SHUTDOWN(mp)) | |
1333 | return false; | |
1334 | ||
1335 | /* | |
1336 | * Update the incore superblock. We synchronize on the primary super | |
1337 | * buffer lock to be consistent with the add function, though at least | |
1338 | * in theory this shouldn't be necessary. | |
1339 | */ | |
1340 | xfs_buf_lock(mp->m_sb_bp); | |
1341 | xfs_buf_hold(mp->m_sb_bp); | |
1342 | ||
1343 | if (xfs_sb_has_incompat_log_feature(&mp->m_sb, | |
1344 | XFS_SB_FEAT_INCOMPAT_LOG_ALL)) { | |
1345 | xfs_info(mp, "Clearing log incompat feature flags."); | |
1346 | xfs_sb_remove_incompat_log_features(&mp->m_sb); | |
1347 | ret = true; | |
1348 | } | |
1349 | ||
1350 | xfs_buf_relse(mp->m_sb_bp); | |
1351 | return ret; | |
1352 | } | |
1353 | ||
9fe82b8c DW |
1354 | /* |
1355 | * Update the in-core delayed block counter. | |
1356 | * | |
1357 | * We prefer to update the counter without having to take a spinlock for every | |
1358 | * counter update (i.e. batching). Each change to delayed allocation | |
1359 | * reservations can change can easily exceed the default percpu counter | |
1360 | * batching, so we use a larger batch factor here. | |
1361 | * | |
1362 | * Note that we don't currently have any callers requiring fast summation | |
1363 | * (e.g. percpu_counter_read) so we can use a big batch value here. | |
1364 | */ | |
1365 | #define XFS_DELALLOC_BATCH (4096) | |
1366 | void | |
1367 | xfs_mod_delalloc( | |
1368 | struct xfs_mount *mp, | |
1369 | int64_t delta) | |
1370 | { | |
1371 | percpu_counter_add_batch(&mp->m_delalloc_blks, delta, | |
1372 | XFS_DELALLOC_BATCH); | |
1373 | } |