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" |
0b1b213f | 34 | |
1da177e4 | 35 | |
27174203 CH |
36 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
37 | static int xfs_uuid_table_size; | |
38 | static uuid_t *xfs_uuid_table; | |
39 | ||
af3b6382 DW |
40 | void |
41 | xfs_uuid_table_free(void) | |
42 | { | |
43 | if (xfs_uuid_table_size == 0) | |
44 | return; | |
45 | kmem_free(xfs_uuid_table); | |
46 | xfs_uuid_table = NULL; | |
47 | xfs_uuid_table_size = 0; | |
48 | } | |
49 | ||
27174203 CH |
50 | /* |
51 | * See if the UUID is unique among mounted XFS filesystems. | |
52 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
53 | */ | |
54 | STATIC int | |
55 | xfs_uuid_mount( | |
56 | struct xfs_mount *mp) | |
57 | { | |
58 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
59 | int hole, i; | |
60 | ||
8f720d9f | 61 | /* Publish UUID in struct super_block */ |
85787090 | 62 | uuid_copy(&mp->m_super->s_uuid, uuid); |
8f720d9f | 63 | |
27174203 CH |
64 | if (mp->m_flags & XFS_MOUNT_NOUUID) |
65 | return 0; | |
66 | ||
d905fdaa AG |
67 | if (uuid_is_null(uuid)) { |
68 | xfs_warn(mp, "Filesystem has null UUID - can't mount"); | |
2451337d | 69 | return -EINVAL; |
27174203 CH |
70 | } |
71 | ||
72 | mutex_lock(&xfs_uuid_table_mutex); | |
73 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 74 | if (uuid_is_null(&xfs_uuid_table[i])) { |
27174203 CH |
75 | hole = i; |
76 | continue; | |
77 | } | |
78 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
79 | goto out_duplicate; | |
80 | } | |
81 | ||
82 | if (hole < 0) { | |
83 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
84 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
707e0dda | 85 | 0); |
27174203 CH |
86 | hole = xfs_uuid_table_size++; |
87 | } | |
88 | xfs_uuid_table[hole] = *uuid; | |
89 | mutex_unlock(&xfs_uuid_table_mutex); | |
90 | ||
91 | return 0; | |
92 | ||
93 | out_duplicate: | |
94 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 95 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 96 | return -EINVAL; |
27174203 CH |
97 | } |
98 | ||
99 | STATIC void | |
100 | xfs_uuid_unmount( | |
101 | struct xfs_mount *mp) | |
102 | { | |
103 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
104 | int i; | |
105 | ||
106 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
107 | return; | |
108 | ||
109 | mutex_lock(&xfs_uuid_table_mutex); | |
110 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
d905fdaa | 111 | if (uuid_is_null(&xfs_uuid_table[i])) |
27174203 CH |
112 | continue; |
113 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
114 | continue; | |
115 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
116 | break; | |
117 | } | |
118 | ASSERT(i < xfs_uuid_table_size); | |
119 | mutex_unlock(&xfs_uuid_table_mutex); | |
120 | } | |
121 | ||
122 | ||
e176579e DC |
123 | STATIC void |
124 | __xfs_free_perag( | |
125 | struct rcu_head *head) | |
126 | { | |
127 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
128 | ||
129 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
130 | kmem_free(pag); | |
131 | } | |
132 | ||
1da177e4 | 133 | /* |
e176579e | 134 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 135 | */ |
c962fb79 | 136 | STATIC void |
ff4f038c | 137 | xfs_free_perag( |
745f6919 | 138 | xfs_mount_t *mp) |
1da177e4 | 139 | { |
1c1c6ebc DC |
140 | xfs_agnumber_t agno; |
141 | struct xfs_perag *pag; | |
142 | ||
143 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
144 | spin_lock(&mp->m_perag_lock); | |
145 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
146 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 147 | ASSERT(pag); |
f83282a8 | 148 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
9b247179 | 149 | xfs_iunlink_destroy(pag); |
6031e73a | 150 | xfs_buf_hash_destroy(pag); |
1da06189 | 151 | mutex_destroy(&pag->pag_ici_reclaim_lock); |
e176579e | 152 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 153 | } |
1da177e4 LT |
154 | } |
155 | ||
4cc929ee NS |
156 | /* |
157 | * Check size of device based on the (data/realtime) block count. | |
158 | * Note: this check is used by the growfs code as well as mount. | |
159 | */ | |
160 | int | |
161 | xfs_sb_validate_fsb_count( | |
162 | xfs_sb_t *sbp, | |
c8ce540d | 163 | uint64_t nblocks) |
4cc929ee NS |
164 | { |
165 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
166 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
167 | ||
d5cf09ba | 168 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 169 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 170 | return -EFBIG; |
4cc929ee NS |
171 | return 0; |
172 | } | |
1da177e4 | 173 | |
1c1c6ebc | 174 | int |
c11e2c36 | 175 | xfs_initialize_perag( |
c11e2c36 | 176 | xfs_mount_t *mp, |
1c1c6ebc DC |
177 | xfs_agnumber_t agcount, |
178 | xfs_agnumber_t *maxagi) | |
1da177e4 | 179 | { |
2d2194f6 | 180 | xfs_agnumber_t index; |
b20fe473 | 181 | xfs_agnumber_t first_initialised = NULLAGNUMBER; |
1da177e4 | 182 | xfs_perag_t *pag; |
8b26c582 | 183 | int error = -ENOMEM; |
1da177e4 | 184 | |
1c1c6ebc DC |
185 | /* |
186 | * Walk the current per-ag tree so we don't try to initialise AGs | |
187 | * that already exist (growfs case). Allocate and insert all the | |
188 | * AGs we don't find ready for initialisation. | |
189 | */ | |
190 | for (index = 0; index < agcount; index++) { | |
191 | pag = xfs_perag_get(mp, index); | |
192 | if (pag) { | |
193 | xfs_perag_put(pag); | |
194 | continue; | |
195 | } | |
fb3b504a | 196 | |
1c1c6ebc DC |
197 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
198 | if (!pag) | |
b20fe473 | 199 | goto out_unwind_new_pags; |
fb3b504a CH |
200 | pag->pag_agno = index; |
201 | pag->pag_mount = mp; | |
1a427ab0 | 202 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 203 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 204 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
6031e73a | 205 | if (xfs_buf_hash_init(pag)) |
b20fe473 | 206 | goto out_free_pag; |
ebf55872 | 207 | init_waitqueue_head(&pag->pagb_wait); |
ff23f4af DW |
208 | spin_lock_init(&pag->pagb_lock); |
209 | pag->pagb_count = 0; | |
210 | pag->pagb_tree = RB_ROOT; | |
fb3b504a | 211 | |
1c1c6ebc | 212 | if (radix_tree_preload(GFP_NOFS)) |
b20fe473 | 213 | goto out_hash_destroy; |
fb3b504a | 214 | |
1c1c6ebc DC |
215 | spin_lock(&mp->m_perag_lock); |
216 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
eb2e9994 | 217 | WARN_ON_ONCE(1); |
1c1c6ebc | 218 | spin_unlock(&mp->m_perag_lock); |
8b26c582 DC |
219 | radix_tree_preload_end(); |
220 | error = -EEXIST; | |
b20fe473 | 221 | goto out_hash_destroy; |
1c1c6ebc DC |
222 | } |
223 | spin_unlock(&mp->m_perag_lock); | |
224 | radix_tree_preload_end(); | |
b20fe473 BD |
225 | /* first new pag is fully initialized */ |
226 | if (first_initialised == NULLAGNUMBER) | |
227 | first_initialised = index; | |
9b247179 DW |
228 | error = xfs_iunlink_init(pag); |
229 | if (error) | |
230 | goto out_hash_destroy; | |
6772c1f1 | 231 | spin_lock_init(&pag->pag_state_lock); |
1c1c6ebc DC |
232 | } |
233 | ||
12c3f05c | 234 | index = xfs_set_inode_alloc(mp, agcount); |
fb3b504a | 235 | |
1c1c6ebc DC |
236 | if (maxagi) |
237 | *maxagi = index; | |
8018026e DW |
238 | |
239 | mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp); | |
1c1c6ebc | 240 | return 0; |
8b26c582 | 241 | |
b20fe473 | 242 | out_hash_destroy: |
6031e73a | 243 | xfs_buf_hash_destroy(pag); |
b20fe473 | 244 | out_free_pag: |
1da06189 | 245 | mutex_destroy(&pag->pag_ici_reclaim_lock); |
8b26c582 | 246 | kmem_free(pag); |
b20fe473 BD |
247 | out_unwind_new_pags: |
248 | /* unwind any prior newly initialized pags */ | |
249 | for (index = first_initialised; index < agcount; index++) { | |
8b26c582 | 250 | pag = radix_tree_delete(&mp->m_perag_tree, index); |
b20fe473 BD |
251 | if (!pag) |
252 | break; | |
6031e73a | 253 | xfs_buf_hash_destroy(pag); |
9b247179 | 254 | xfs_iunlink_destroy(pag); |
1da06189 | 255 | mutex_destroy(&pag->pag_ici_reclaim_lock); |
8b26c582 DC |
256 | kmem_free(pag); |
257 | } | |
258 | return error; | |
1da177e4 LT |
259 | } |
260 | ||
1da177e4 LT |
261 | /* |
262 | * xfs_readsb | |
263 | * | |
264 | * Does the initial read of the superblock. | |
265 | */ | |
266 | int | |
ff55068c DC |
267 | xfs_readsb( |
268 | struct xfs_mount *mp, | |
269 | int flags) | |
1da177e4 LT |
270 | { |
271 | unsigned int sector_size; | |
04a1e6c5 DC |
272 | struct xfs_buf *bp; |
273 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 274 | int error; |
af34e09d | 275 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 276 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
277 | |
278 | ASSERT(mp->m_sb_bp == NULL); | |
279 | ASSERT(mp->m_ddev_targp != NULL); | |
280 | ||
daba5427 ES |
281 | /* |
282 | * For the initial read, we must guess at the sector | |
283 | * size based on the block device. It's enough to | |
284 | * get the sb_sectsize out of the superblock and | |
285 | * then reread with the proper length. | |
286 | * We don't verify it yet, because it may not be complete. | |
287 | */ | |
288 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
289 | buf_ops = NULL; | |
290 | ||
1da177e4 | 291 | /* |
c891c30a BF |
292 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
293 | * around at all times to optimize access to the superblock. Therefore, | |
294 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
295 | * elevated. | |
1da177e4 | 296 | */ |
26af6552 | 297 | reread: |
ba372674 | 298 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
299 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
300 | buf_ops); | |
ba372674 | 301 | if (error) { |
eab4e633 | 302 | if (loud) |
e721f504 | 303 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 304 | /* bad CRC means corrupted metadata */ |
2451337d DC |
305 | if (error == -EFSBADCRC) |
306 | error = -EFSCORRUPTED; | |
ba372674 | 307 | return error; |
eab4e633 | 308 | } |
1da177e4 LT |
309 | |
310 | /* | |
311 | * Initialize the mount structure from the superblock. | |
1da177e4 | 312 | */ |
556b8883 | 313 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
314 | |
315 | /* | |
316 | * If we haven't validated the superblock, do so now before we try | |
317 | * to check the sector size and reread the superblock appropriately. | |
318 | */ | |
319 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
320 | if (loud) | |
321 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 322 | error = -EINVAL; |
556b8883 DC |
323 | goto release_buf; |
324 | } | |
ff55068c | 325 | |
1da177e4 LT |
326 | /* |
327 | * We must be able to do sector-sized and sector-aligned IO. | |
328 | */ | |
04a1e6c5 | 329 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
330 | if (loud) |
331 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 332 | sector_size, sbp->sb_sectsize); |
2451337d | 333 | error = -ENOSYS; |
26af6552 | 334 | goto release_buf; |
1da177e4 LT |
335 | } |
336 | ||
daba5427 | 337 | if (buf_ops == NULL) { |
556b8883 DC |
338 | /* |
339 | * Re-read the superblock so the buffer is correctly sized, | |
340 | * and properly verified. | |
341 | */ | |
1da177e4 | 342 | xfs_buf_relse(bp); |
04a1e6c5 | 343 | sector_size = sbp->sb_sectsize; |
daba5427 | 344 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 345 | goto reread; |
1da177e4 LT |
346 | } |
347 | ||
5681ca40 | 348 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 349 | |
04a1e6c5 DC |
350 | /* no need to be quiet anymore, so reset the buf ops */ |
351 | bp->b_ops = &xfs_sb_buf_ops; | |
352 | ||
1da177e4 | 353 | mp->m_sb_bp = bp; |
26af6552 | 354 | xfs_buf_unlock(bp); |
1da177e4 LT |
355 | return 0; |
356 | ||
26af6552 DC |
357 | release_buf: |
358 | xfs_buf_relse(bp); | |
1da177e4 LT |
359 | return error; |
360 | } | |
361 | ||
1da177e4 | 362 | /* |
0771fb45 | 363 | * Update alignment values based on mount options and sb values |
1da177e4 | 364 | */ |
0771fb45 | 365 | STATIC int |
7884bc86 | 366 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 367 | { |
1da177e4 | 368 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 369 | |
4249023a | 370 | if (mp->m_dalign) { |
1da177e4 LT |
371 | /* |
372 | * If stripe unit and stripe width are not multiples | |
373 | * of the fs blocksize turn off alignment. | |
374 | */ | |
375 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
376 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
377 | xfs_warn(mp, |
378 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
379 | sbp->sb_blocksize); | |
2451337d | 380 | return -EINVAL; |
1da177e4 LT |
381 | } else { |
382 | /* | |
383 | * Convert the stripe unit and width to FSBs. | |
384 | */ | |
385 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
386 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 387 | xfs_warn(mp, |
39a45d84 JL |
388 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
389 | sbp->sb_agblocks); | |
2451337d | 390 | return -EINVAL; |
1da177e4 LT |
391 | } else if (mp->m_dalign) { |
392 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
393 | } else { | |
39a45d84 JL |
394 | xfs_warn(mp, |
395 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
396 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 397 | return -EINVAL; |
1da177e4 LT |
398 | } |
399 | } | |
400 | ||
401 | /* | |
402 | * Update superblock with new values | |
403 | * and log changes | |
404 | */ | |
62118709 | 405 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
406 | if (sbp->sb_unit != mp->m_dalign) { |
407 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 408 | mp->m_update_sb = true; |
1da177e4 LT |
409 | } |
410 | if (sbp->sb_width != mp->m_swidth) { | |
411 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 412 | mp->m_update_sb = true; |
1da177e4 | 413 | } |
34d7f603 JL |
414 | } else { |
415 | xfs_warn(mp, | |
416 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 417 | return -EINVAL; |
1da177e4 LT |
418 | } |
419 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 420 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
421 | mp->m_dalign = sbp->sb_unit; |
422 | mp->m_swidth = sbp->sb_width; | |
423 | } | |
424 | ||
0771fb45 ES |
425 | return 0; |
426 | } | |
1da177e4 | 427 | |
0771fb45 ES |
428 | /* |
429 | * Set the default minimum read and write sizes unless | |
430 | * already specified in a mount option. | |
431 | * We use smaller I/O sizes when the file system | |
432 | * is being used for NFS service (wsync mount option). | |
433 | */ | |
434 | STATIC void | |
435 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
436 | { | |
437 | xfs_sb_t *sbp = &(mp->m_sb); | |
438 | int readio_log, writeio_log; | |
1da177e4 | 439 | |
1da177e4 LT |
440 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
441 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
442 | readio_log = XFS_WSYNC_READIO_LOG; | |
443 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
444 | } else { | |
445 | readio_log = XFS_READIO_LOG_LARGE; | |
446 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
447 | } | |
448 | } else { | |
449 | readio_log = mp->m_readio_log; | |
450 | writeio_log = mp->m_writeio_log; | |
451 | } | |
452 | ||
1da177e4 LT |
453 | if (sbp->sb_blocklog > readio_log) { |
454 | mp->m_readio_log = sbp->sb_blocklog; | |
455 | } else { | |
456 | mp->m_readio_log = readio_log; | |
457 | } | |
458 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
459 | if (sbp->sb_blocklog > writeio_log) { | |
460 | mp->m_writeio_log = sbp->sb_blocklog; | |
461 | } else { | |
462 | mp->m_writeio_log = writeio_log; | |
463 | } | |
464 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 465 | } |
1da177e4 | 466 | |
055388a3 DC |
467 | /* |
468 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
469 | */ | |
470 | void | |
471 | xfs_set_low_space_thresholds( | |
472 | struct xfs_mount *mp) | |
473 | { | |
474 | int i; | |
475 | ||
476 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
c8ce540d | 477 | uint64_t space = mp->m_sb.sb_dblocks; |
055388a3 DC |
478 | |
479 | do_div(space, 100); | |
480 | mp->m_low_space[i] = space * (i + 1); | |
481 | } | |
482 | } | |
483 | ||
0771fb45 | 484 | /* |
0471f62e | 485 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
486 | */ |
487 | STATIC int | |
ba372674 DC |
488 | xfs_check_sizes( |
489 | struct xfs_mount *mp) | |
0771fb45 | 490 | { |
ba372674 | 491 | struct xfs_buf *bp; |
0771fb45 | 492 | xfs_daddr_t d; |
ba372674 | 493 | int error; |
0771fb45 | 494 | |
1da177e4 LT |
495 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
496 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 497 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 498 | return -EFBIG; |
1da177e4 | 499 | } |
ba372674 | 500 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 501 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
502 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
503 | if (error) { | |
0b932ccc | 504 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 505 | return error; |
1da177e4 | 506 | } |
1922c949 | 507 | xfs_buf_relse(bp); |
1da177e4 | 508 | |
ba372674 DC |
509 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
510 | return 0; | |
511 | ||
512 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
513 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
514 | xfs_warn(mp, "log size mismatch detected"); | |
515 | return -EFBIG; | |
516 | } | |
517 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 518 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
519 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
520 | if (error) { | |
521 | xfs_warn(mp, "log device read failed"); | |
522 | return error; | |
0771fb45 | 523 | } |
ba372674 | 524 | xfs_buf_relse(bp); |
0771fb45 ES |
525 | return 0; |
526 | } | |
527 | ||
7d095257 CH |
528 | /* |
529 | * Clear the quotaflags in memory and in the superblock. | |
530 | */ | |
531 | int | |
532 | xfs_mount_reset_sbqflags( | |
533 | struct xfs_mount *mp) | |
534 | { | |
7d095257 CH |
535 | mp->m_qflags = 0; |
536 | ||
61e63ecb | 537 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
538 | if (mp->m_sb.sb_qflags == 0) |
539 | return 0; | |
540 | spin_lock(&mp->m_sb_lock); | |
541 | mp->m_sb.sb_qflags = 0; | |
542 | spin_unlock(&mp->m_sb_lock); | |
543 | ||
61e63ecb | 544 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
545 | return 0; |
546 | ||
61e63ecb | 547 | return xfs_sync_sb(mp, false); |
7d095257 CH |
548 | } |
549 | ||
c8ce540d | 550 | uint64_t |
d5db0f97 ES |
551 | xfs_default_resblks(xfs_mount_t *mp) |
552 | { | |
c8ce540d | 553 | uint64_t resblks; |
d5db0f97 ES |
554 | |
555 | /* | |
8babd8a2 DC |
556 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
557 | * smaller. This is intended to cover concurrent allocation | |
558 | * transactions when we initially hit enospc. These each require a 4 | |
559 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
560 | * allocation reservations. | |
d5db0f97 ES |
561 | */ |
562 | resblks = mp->m_sb.sb_dblocks; | |
563 | do_div(resblks, 20); | |
c8ce540d | 564 | resblks = min_t(uint64_t, resblks, 8192); |
d5db0f97 ES |
565 | return resblks; |
566 | } | |
567 | ||
2e9e6481 DW |
568 | /* Ensure the summary counts are correct. */ |
569 | STATIC int | |
570 | xfs_check_summary_counts( | |
571 | struct xfs_mount *mp) | |
572 | { | |
573 | /* | |
574 | * The AG0 superblock verifier rejects in-progress filesystems, | |
575 | * so we should never see the flag set this far into mounting. | |
576 | */ | |
577 | if (mp->m_sb.sb_inprogress) { | |
578 | xfs_err(mp, "sb_inprogress set after log recovery??"); | |
579 | WARN_ON(1); | |
580 | return -EFSCORRUPTED; | |
581 | } | |
582 | ||
583 | /* | |
584 | * Now the log is mounted, we know if it was an unclean shutdown or | |
585 | * not. If it was, with the first phase of recovery has completed, we | |
586 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
587 | * but they are recovered transactionally in the second recovery phase | |
588 | * later. | |
589 | * | |
590 | * If the log was clean when we mounted, we can check the summary | |
591 | * counters. If any of them are obviously incorrect, we can recompute | |
592 | * them from the AGF headers in the next step. | |
593 | */ | |
594 | if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
595 | (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks || | |
00d22a1c | 596 | !xfs_verify_icount(mp, mp->m_sb.sb_icount) || |
2e9e6481 | 597 | mp->m_sb.sb_ifree > mp->m_sb.sb_icount)) |
39353ff6 | 598 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
2e9e6481 DW |
599 | |
600 | /* | |
601 | * We can safely re-initialise incore superblock counters from the | |
602 | * per-ag data. These may not be correct if the filesystem was not | |
603 | * cleanly unmounted, so we waited for recovery to finish before doing | |
604 | * this. | |
605 | * | |
606 | * If the filesystem was cleanly unmounted or the previous check did | |
607 | * not flag anything weird, then we can trust the values in the | |
608 | * superblock to be correct and we don't need to do anything here. | |
609 | * Otherwise, recalculate the summary counters. | |
610 | */ | |
611 | if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) || | |
612 | XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) && | |
39353ff6 | 613 | !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS)) |
2e9e6481 DW |
614 | return 0; |
615 | ||
616 | return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount); | |
617 | } | |
618 | ||
0771fb45 | 619 | /* |
0771fb45 ES |
620 | * This function does the following on an initial mount of a file system: |
621 | * - reads the superblock from disk and init the mount struct | |
622 | * - if we're a 32-bit kernel, do a size check on the superblock | |
623 | * so we don't mount terabyte filesystems | |
624 | * - init mount struct realtime fields | |
625 | * - allocate inode hash table for fs | |
626 | * - init directory manager | |
627 | * - perform recovery and init the log manager | |
628 | */ | |
629 | int | |
630 | xfs_mountfs( | |
f0b2efad | 631 | struct xfs_mount *mp) |
0771fb45 | 632 | { |
f0b2efad BF |
633 | struct xfs_sb *sbp = &(mp->m_sb); |
634 | struct xfs_inode *rip; | |
ef325959 | 635 | struct xfs_ino_geometry *igeo = M_IGEO(mp); |
c8ce540d | 636 | uint64_t resblks; |
f0b2efad BF |
637 | uint quotamount = 0; |
638 | uint quotaflags = 0; | |
639 | int error = 0; | |
0771fb45 | 640 | |
ff55068c | 641 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 642 | |
ee1c0908 | 643 | /* |
074e427b DC |
644 | * Check for a mismatched features2 values. Older kernels read & wrote |
645 | * into the wrong sb offset for sb_features2 on some platforms due to | |
646 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
647 | * which made older superblock reading/writing routines swap it as a | |
648 | * 64-bit value. | |
ee1c0908 | 649 | * |
e6957ea4 ES |
650 | * For backwards compatibility, we make both slots equal. |
651 | * | |
074e427b DC |
652 | * If we detect a mismatched field, we OR the set bits into the existing |
653 | * features2 field in case it has already been modified; we don't want | |
654 | * to lose any features. We then update the bad location with the ORed | |
655 | * value so that older kernels will see any features2 flags. The | |
656 | * superblock writeback code ensures the new sb_features2 is copied to | |
657 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 658 | */ |
e6957ea4 | 659 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 660 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 661 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 662 | mp->m_update_sb = true; |
e6957ea4 ES |
663 | |
664 | /* | |
665 | * Re-check for ATTR2 in case it was found in bad_features2 | |
666 | * slot. | |
667 | */ | |
7c12f296 TS |
668 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
669 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 670 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
671 | } |
672 | ||
673 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
674 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
675 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 676 | mp->m_update_sb = true; |
e6957ea4 | 677 | |
7c12f296 TS |
678 | /* update sb_versionnum for the clearing of the morebits */ |
679 | if (!sbp->sb_features2) | |
61e63ecb | 680 | mp->m_update_sb = true; |
ee1c0908 DC |
681 | } |
682 | ||
263997a6 DC |
683 | /* always use v2 inodes by default now */ |
684 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
685 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 686 | mp->m_update_sb = true; |
263997a6 DC |
687 | } |
688 | ||
0771fb45 ES |
689 | /* |
690 | * Check if sb_agblocks is aligned at stripe boundary | |
691 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
692 | * allocator alignment is within an ag, therefore ag has | |
693 | * to be aligned at stripe boundary. | |
694 | */ | |
7884bc86 | 695 | error = xfs_update_alignment(mp); |
0771fb45 | 696 | if (error) |
f9057e3d | 697 | goto out; |
0771fb45 ES |
698 | |
699 | xfs_alloc_compute_maxlevels(mp); | |
700 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
701 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
494dba7b | 702 | xfs_ialloc_setup_geometry(mp); |
035e00ac | 703 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 704 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 | 705 | |
e6b3bb78 | 706 | /* enable fail_at_unmount as default */ |
749f24f3 | 707 | mp->m_fail_unmount = true; |
e6b3bb78 | 708 | |
a31b1d3d | 709 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
710 | if (error) |
711 | goto out; | |
1da177e4 | 712 | |
225e4635 BD |
713 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
714 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
715 | if (error) |
716 | goto out_remove_sysfs; | |
717 | ||
192852be | 718 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
719 | if (error) |
720 | goto out_del_stats; | |
721 | ||
31965ef3 DW |
722 | error = xfs_errortag_init(mp); |
723 | if (error) | |
724 | goto out_remove_error_sysfs; | |
192852be CM |
725 | |
726 | error = xfs_uuid_mount(mp); | |
727 | if (error) | |
31965ef3 | 728 | goto out_remove_errortag; |
192852be | 729 | |
0771fb45 ES |
730 | /* |
731 | * Set the minimum read and write sizes | |
732 | */ | |
733 | xfs_set_rw_sizes(mp); | |
734 | ||
055388a3 DC |
735 | /* set the low space thresholds for dynamic preallocation */ |
736 | xfs_set_low_space_thresholds(mp); | |
737 | ||
e5376fc1 BF |
738 | /* |
739 | * If enabled, sparse inode chunk alignment is expected to match the | |
740 | * cluster size. Full inode chunk alignment must match the chunk size, | |
741 | * but that is checked on sb read verification... | |
742 | */ | |
743 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
744 | mp->m_sb.sb_spino_align != | |
490d451f | 745 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) { |
e5376fc1 BF |
746 | xfs_warn(mp, |
747 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
748 | mp->m_sb.sb_spino_align, | |
490d451f | 749 | XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)); |
e5376fc1 BF |
750 | error = -EINVAL; |
751 | goto out_remove_uuid; | |
752 | } | |
753 | ||
0771fb45 | 754 | /* |
c2bfbc9b | 755 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 756 | */ |
4249023a | 757 | error = xfs_check_sizes(mp); |
0771fb45 | 758 | if (error) |
f9057e3d | 759 | goto out_remove_uuid; |
0771fb45 | 760 | |
1da177e4 LT |
761 | /* |
762 | * Initialize realtime fields in the mount structure | |
763 | */ | |
0771fb45 ES |
764 | error = xfs_rtmount_init(mp); |
765 | if (error) { | |
0b932ccc | 766 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 767 | goto out_remove_uuid; |
1da177e4 LT |
768 | } |
769 | ||
1da177e4 LT |
770 | /* |
771 | * Copies the low order bits of the timestamp and the randomly | |
772 | * set "sequence" number out of a UUID. | |
773 | */ | |
cb0ba6cc CH |
774 | mp->m_fixedfsid[0] = |
775 | (get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) | | |
776 | get_unaligned_be16(&sbp->sb_uuid.b[4]); | |
777 | mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]); | |
1da177e4 | 778 | |
0650b554 DC |
779 | error = xfs_da_mount(mp); |
780 | if (error) { | |
781 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
782 | goto out_remove_uuid; | |
783 | } | |
1da177e4 LT |
784 | |
785 | /* | |
786 | * Initialize the precomputed transaction reservations values. | |
787 | */ | |
788 | xfs_trans_init(mp); | |
789 | ||
1da177e4 LT |
790 | /* |
791 | * Allocate and initialize the per-ag data. | |
792 | */ | |
1c1c6ebc DC |
793 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
794 | if (error) { | |
0b932ccc | 795 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 796 | goto out_free_dir; |
1c1c6ebc | 797 | } |
1da177e4 | 798 | |
f9057e3d | 799 | if (!sbp->sb_logblocks) { |
0b932ccc | 800 | xfs_warn(mp, "no log defined"); |
f9057e3d | 801 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 802 | error = -EFSCORRUPTED; |
f9057e3d CH |
803 | goto out_free_perag; |
804 | } | |
805 | ||
1da177e4 | 806 | /* |
f0b2efad BF |
807 | * Log's mount-time initialization. The first part of recovery can place |
808 | * some items on the AIL, to be handled when recovery is finished or | |
809 | * cancelled. | |
1da177e4 | 810 | */ |
f9057e3d CH |
811 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
812 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
813 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
814 | if (error) { | |
0b932ccc | 815 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 816 | goto out_fail_wait; |
1da177e4 LT |
817 | } |
818 | ||
2e9e6481 DW |
819 | /* Make sure the summary counts are ok. */ |
820 | error = xfs_check_summary_counts(mp); | |
821 | if (error) | |
822 | goto out_log_dealloc; | |
f9057e3d | 823 | |
1da177e4 LT |
824 | /* |
825 | * Get and sanity-check the root inode. | |
826 | * Save the pointer to it in the mount structure. | |
827 | */ | |
541b5acc DC |
828 | error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED, |
829 | XFS_ILOCK_EXCL, &rip); | |
1da177e4 | 830 | if (error) { |
541b5acc DC |
831 | xfs_warn(mp, |
832 | "Failed to read root inode 0x%llx, error %d", | |
833 | sbp->sb_rootino, -error); | |
f9057e3d | 834 | goto out_log_dealloc; |
1da177e4 LT |
835 | } |
836 | ||
837 | ASSERT(rip != NULL); | |
1da177e4 | 838 | |
c19b3b05 | 839 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 840 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 841 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
842 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
843 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
844 | mp); | |
2451337d | 845 | error = -EFSCORRUPTED; |
f9057e3d | 846 | goto out_rele_rip; |
1da177e4 LT |
847 | } |
848 | mp->m_rootip = rip; /* save it */ | |
849 | ||
850 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
851 | ||
852 | /* | |
853 | * Initialize realtime inode pointers in the mount structure | |
854 | */ | |
0771fb45 ES |
855 | error = xfs_rtmount_inodes(mp); |
856 | if (error) { | |
1da177e4 LT |
857 | /* |
858 | * Free up the root inode. | |
859 | */ | |
0b932ccc | 860 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 861 | goto out_rele_rip; |
1da177e4 LT |
862 | } |
863 | ||
864 | /* | |
7884bc86 CH |
865 | * If this is a read-only mount defer the superblock updates until |
866 | * the next remount into writeable mode. Otherwise we would never | |
867 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 868 | */ |
61e63ecb DC |
869 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
870 | error = xfs_sync_sb(mp, false); | |
e5720eec | 871 | if (error) { |
0b932ccc | 872 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 873 | goto out_rtunmount; |
e5720eec DC |
874 | } |
875 | } | |
1da177e4 LT |
876 | |
877 | /* | |
878 | * Initialise the XFS quota management subsystem for this mount | |
879 | */ | |
7d095257 CH |
880 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
881 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
882 | if (error) | |
883 | goto out_rtunmount; | |
884 | } else { | |
885 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
886 | ||
887 | /* | |
888 | * If a file system had quotas running earlier, but decided to | |
889 | * mount without -o uquota/pquota/gquota options, revoke the | |
890 | * quotachecked license. | |
891 | */ | |
892 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 893 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
894 | error = xfs_mount_reset_sbqflags(mp); |
895 | if (error) | |
a70a4fa5 | 896 | goto out_rtunmount; |
7d095257 CH |
897 | } |
898 | } | |
1da177e4 LT |
899 | |
900 | /* | |
f0b2efad BF |
901 | * Finish recovering the file system. This part needed to be delayed |
902 | * until after the root and real-time bitmap inodes were consistently | |
903 | * read in. | |
1da177e4 | 904 | */ |
4249023a | 905 | error = xfs_log_mount_finish(mp); |
1da177e4 | 906 | if (error) { |
0b932ccc | 907 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 908 | goto out_rtunmount; |
1da177e4 LT |
909 | } |
910 | ||
ddeb14f4 DC |
911 | /* |
912 | * Now the log is fully replayed, we can transition to full read-only | |
913 | * mode for read-only mounts. This will sync all the metadata and clean | |
914 | * the log so that the recovery we just performed does not have to be | |
915 | * replayed again on the next mount. | |
916 | * | |
917 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
918 | * semantically identical operations. | |
919 | */ | |
920 | if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) == | |
921 | XFS_MOUNT_RDONLY) { | |
922 | xfs_quiesce_attr(mp); | |
923 | } | |
924 | ||
1da177e4 LT |
925 | /* |
926 | * Complete the quota initialisation, post-log-replay component. | |
927 | */ | |
7d095257 CH |
928 | if (quotamount) { |
929 | ASSERT(mp->m_qflags == 0); | |
930 | mp->m_qflags = quotaflags; | |
931 | ||
932 | xfs_qm_mount_quotas(mp); | |
933 | } | |
934 | ||
84e1e99f DC |
935 | /* |
936 | * Now we are mounted, reserve a small amount of unused space for | |
937 | * privileged transactions. This is needed so that transaction | |
938 | * space required for critical operations can dip into this pool | |
939 | * when at ENOSPC. This is needed for operations like create with | |
940 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
941 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
942 | * |
943 | * This may drive us straight to ENOSPC on mount, but that implies | |
944 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 945 | */ |
d5db0f97 ES |
946 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
947 | resblks = xfs_default_resblks(mp); | |
948 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
949 | if (error) | |
0b932ccc DC |
950 | xfs_warn(mp, |
951 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
174edb0e DW |
952 | |
953 | /* Recover any CoW blocks that never got remapped. */ | |
954 | error = xfs_reflink_recover_cow(mp); | |
955 | if (error) { | |
956 | xfs_err(mp, | |
957 | "Error %d recovering leftover CoW allocations.", error); | |
958 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
959 | goto out_quota; | |
960 | } | |
84d69619 DW |
961 | |
962 | /* Reserve AG blocks for future btree expansion. */ | |
963 | error = xfs_fs_reserve_ag_blocks(mp); | |
964 | if (error && error != -ENOSPC) | |
965 | goto out_agresv; | |
d5db0f97 | 966 | } |
84e1e99f | 967 | |
1da177e4 LT |
968 | return 0; |
969 | ||
84d69619 DW |
970 | out_agresv: |
971 | xfs_fs_unreserve_ag_blocks(mp); | |
174edb0e DW |
972 | out_quota: |
973 | xfs_qm_unmount_quotas(mp); | |
b93b6e43 CH |
974 | out_rtunmount: |
975 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 976 | out_rele_rip: |
44a8736b | 977 | xfs_irele(rip); |
77aff8c7 DW |
978 | /* Clean out dquots that might be in memory after quotacheck. */ |
979 | xfs_qm_unmount(mp); | |
2d1d1da3 DW |
980 | /* |
981 | * Cancel all delayed reclaim work and reclaim the inodes directly. | |
982 | * We have to do this /after/ rtunmount and qm_unmount because those | |
983 | * two will have scheduled delayed reclaim for the rt/quota inodes. | |
984 | * | |
985 | * This is slightly different from the unmountfs call sequence | |
986 | * because we could be tearing down a partially set up mount. In | |
987 | * particular, if log_mount_finish fails we bail out without calling | |
988 | * qm_unmount_quotas and therefore rely on qm_unmount to release the | |
989 | * quota inodes. | |
990 | */ | |
991 | cancel_delayed_work_sync(&mp->m_reclaim_work); | |
992 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
519841c2 | 993 | xfs_health_unmount(mp); |
f9057e3d | 994 | out_log_dealloc: |
e6b3bb78 | 995 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 996 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
997 | out_fail_wait: |
998 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
999 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1000 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1001 | out_free_perag: |
ff4f038c | 1002 | xfs_free_perag(mp); |
0650b554 DC |
1003 | out_free_dir: |
1004 | xfs_da_unmount(mp); | |
f9057e3d | 1005 | out_remove_uuid: |
27174203 | 1006 | xfs_uuid_unmount(mp); |
31965ef3 DW |
1007 | out_remove_errortag: |
1008 | xfs_errortag_del(mp); | |
192852be CM |
1009 | out_remove_error_sysfs: |
1010 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
1011 | out_del_stats: |
1012 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
1013 | out_remove_sysfs: |
1014 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1015 | out: |
1da177e4 LT |
1016 | return error; |
1017 | } | |
1018 | ||
1019 | /* | |
1da177e4 LT |
1020 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1021 | * log and makes sure that incore structures are freed. | |
1022 | */ | |
41b5c2e7 CH |
1023 | void |
1024 | xfs_unmountfs( | |
1025 | struct xfs_mount *mp) | |
1da177e4 | 1026 | { |
c8ce540d | 1027 | uint64_t resblks; |
41b5c2e7 | 1028 | int error; |
1da177e4 | 1029 | |
ed30dcbd | 1030 | xfs_stop_block_reaping(mp); |
84d69619 | 1031 | xfs_fs_unreserve_ag_blocks(mp); |
7d095257 | 1032 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1033 | xfs_rtunmount_inodes(mp); |
44a8736b | 1034 | xfs_irele(mp->m_rootip); |
77508ec8 | 1035 | |
641c56fb DC |
1036 | /* |
1037 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1038 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1039 | * the transaction is still sitting in a iclog. The stale inodes |
1040 | * on that buffer will have their flush locks held until the | |
1041 | * transaction hits the disk and the callbacks run. the inode | |
1042 | * flush takes the flush lock unconditionally and with nothing to | |
1043 | * push out the iclog we will never get that unlocked. hence we | |
1044 | * need to force the log first. | |
1045 | */ | |
a14a348b | 1046 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1047 | |
ebf55872 CH |
1048 | /* |
1049 | * Wait for all busy extents to be freed, including completion of | |
1050 | * any discard operation. | |
1051 | */ | |
1052 | xfs_extent_busy_wait_all(mp); | |
4560e78f | 1053 | flush_workqueue(xfs_discard_wq); |
ebf55872 | 1054 | |
e6b3bb78 CM |
1055 | /* |
1056 | * We now need to tell the world we are unmounting. This will allow | |
1057 | * us to detect that the filesystem is going away and we should error | |
1058 | * out anything that we have been retrying in the background. This will | |
1059 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1060 | */ | |
1061 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1062 | ||
c854363e | 1063 | /* |
211e4d43 CH |
1064 | * Flush all pending changes from the AIL. |
1065 | */ | |
1066 | xfs_ail_push_all_sync(mp->m_ail); | |
1067 | ||
1068 | /* | |
1069 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1070 | * inodes and none should be pinned or locked, but use synchronous |
1071 | * reclaim just to be sure. We can stop background inode reclaim | |
1072 | * here as well if it is still running. | |
c854363e | 1073 | */ |
7e18530b | 1074 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1075 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
519841c2 | 1076 | xfs_health_unmount(mp); |
1da177e4 | 1077 | |
7d095257 | 1078 | xfs_qm_unmount(mp); |
a357a121 | 1079 | |
84e1e99f DC |
1080 | /* |
1081 | * Unreserve any blocks we have so that when we unmount we don't account | |
1082 | * the reserved free space as used. This is really only necessary for | |
1083 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1084 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1085 | * |
1086 | * We don't bother correcting this elsewhere for lazy superblock | |
1087 | * counting because on mount of an unclean filesystem we reconstruct the | |
1088 | * correct counter value and this is irrelevant. | |
1089 | * | |
1090 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1091 | * we only every apply deltas to the superblock and hence the incore | |
1092 | * value does not matter.... | |
1093 | */ | |
1094 | resblks = 0; | |
714082bc DC |
1095 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1096 | if (error) | |
0b932ccc | 1097 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1098 | "Freespace may not be correct on next mount."); |
1099 | ||
adab0f67 | 1100 | error = xfs_log_sbcount(mp); |
e5720eec | 1101 | if (error) |
0b932ccc | 1102 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1103 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1104 | |
225e4635 | 1105 | |
21b699c8 | 1106 | xfs_log_unmount(mp); |
0650b554 | 1107 | xfs_da_unmount(mp); |
27174203 | 1108 | xfs_uuid_unmount(mp); |
1da177e4 | 1109 | |
1550d0b0 | 1110 | #if defined(DEBUG) |
31965ef3 | 1111 | xfs_errortag_clearall(mp); |
1da177e4 | 1112 | #endif |
ff4f038c | 1113 | xfs_free_perag(mp); |
a31b1d3d | 1114 | |
31965ef3 | 1115 | xfs_errortag_del(mp); |
192852be | 1116 | xfs_error_sysfs_del(mp); |
225e4635 | 1117 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1118 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1119 | } |
1120 | ||
91ee575f BF |
1121 | /* |
1122 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1123 | * freeze level for which modifications should not be allowed. This allows | |
1124 | * certain operations to proceed while the freeze sequence is in progress, if | |
1125 | * necessary. | |
1126 | */ | |
1127 | bool | |
1128 | xfs_fs_writable( | |
1129 | struct xfs_mount *mp, | |
1130 | int level) | |
92821e2b | 1131 | { |
91ee575f BF |
1132 | ASSERT(level > SB_UNFROZEN); |
1133 | if ((mp->m_super->s_writers.frozen >= level) || | |
1134 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1135 | return false; | |
1136 | ||
1137 | return true; | |
92821e2b DC |
1138 | } |
1139 | ||
1140 | /* | |
b2ce3974 AE |
1141 | * xfs_log_sbcount |
1142 | * | |
adab0f67 | 1143 | * Sync the superblock counters to disk. |
b2ce3974 | 1144 | * |
91ee575f BF |
1145 | * Note this code can be called during the process of freezing, so we use the |
1146 | * transaction allocator that does not block when the transaction subsystem is | |
1147 | * in its frozen state. | |
92821e2b DC |
1148 | */ |
1149 | int | |
adab0f67 | 1150 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1151 | { |
91ee575f BF |
1152 | /* allow this to proceed during the freeze sequence... */ |
1153 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1154 | return 0; |
1155 | ||
92821e2b DC |
1156 | /* |
1157 | * we don't need to do this if we are updating the superblock | |
1158 | * counters on every modification. | |
1159 | */ | |
1160 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1161 | return 0; | |
1162 | ||
61e63ecb | 1163 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1164 | } |
1165 | ||
8c1903d3 DC |
1166 | /* |
1167 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1168 | * of 128 so we don't need to take the counter lock on every update. | |
1169 | */ | |
1170 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1171 | int |
1172 | xfs_mod_icount( | |
1173 | struct xfs_mount *mp, | |
1174 | int64_t delta) | |
1175 | { | |
104b4e51 | 1176 | percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
8c1903d3 | 1177 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { |
501ab323 DC |
1178 | ASSERT(0); |
1179 | percpu_counter_add(&mp->m_icount, -delta); | |
1180 | return -EINVAL; | |
1181 | } | |
1182 | return 0; | |
1183 | } | |
1184 | ||
e88b64ea DC |
1185 | int |
1186 | xfs_mod_ifree( | |
1187 | struct xfs_mount *mp, | |
1188 | int64_t delta) | |
1189 | { | |
1190 | percpu_counter_add(&mp->m_ifree, delta); | |
1191 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1192 | ASSERT(0); | |
1193 | percpu_counter_add(&mp->m_ifree, -delta); | |
1194 | return -EINVAL; | |
1195 | } | |
1196 | return 0; | |
1197 | } | |
0d485ada | 1198 | |
8c1903d3 DC |
1199 | /* |
1200 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1201 | * only occurs on frequent small block count updates such as in the delayed | |
1202 | * allocation path for buffered writes (page a time updates). Hence we set | |
1203 | * a large batch count (1024) to minimise global counter updates except when | |
1204 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1205 | */ | |
1206 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1207 | int |
1208 | xfs_mod_fdblocks( | |
1209 | struct xfs_mount *mp, | |
1210 | int64_t delta, | |
1211 | bool rsvd) | |
1212 | { | |
1213 | int64_t lcounter; | |
1214 | long long res_used; | |
1215 | s32 batch; | |
1216 | ||
1217 | if (delta > 0) { | |
1218 | /* | |
1219 | * If the reserve pool is depleted, put blocks back into it | |
1220 | * first. Most of the time the pool is full. | |
1221 | */ | |
1222 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1223 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1224 | return 0; | |
1225 | } | |
1226 | ||
1227 | spin_lock(&mp->m_sb_lock); | |
1228 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1229 | ||
1230 | if (res_used > delta) { | |
1231 | mp->m_resblks_avail += delta; | |
1232 | } else { | |
1233 | delta -= res_used; | |
1234 | mp->m_resblks_avail = mp->m_resblks; | |
1235 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1236 | } | |
1237 | spin_unlock(&mp->m_sb_lock); | |
1238 | return 0; | |
1239 | } | |
1240 | ||
1241 | /* | |
1242 | * Taking blocks away, need to be more accurate the closer we | |
1243 | * are to zero. | |
1244 | * | |
0d485ada DC |
1245 | * If the counter has a value of less than 2 * max batch size, |
1246 | * then make everything serialise as we are real close to | |
1247 | * ENOSPC. | |
1248 | */ | |
8c1903d3 DC |
1249 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1250 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1251 | batch = 1; |
1252 | else | |
8c1903d3 | 1253 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada | 1254 | |
104b4e51 | 1255 | percpu_counter_add_batch(&mp->m_fdblocks, delta, batch); |
52548852 | 1256 | if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside, |
8c1903d3 | 1257 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1258 | /* we had space! */ |
1259 | return 0; | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * lock up the sb for dipping into reserves before releasing the space | |
1264 | * that took us to ENOSPC. | |
1265 | */ | |
1266 | spin_lock(&mp->m_sb_lock); | |
1267 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1268 | if (!rsvd) | |
1269 | goto fdblocks_enospc; | |
1270 | ||
1271 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1272 | if (lcounter >= 0) { | |
1273 | mp->m_resblks_avail = lcounter; | |
1274 | spin_unlock(&mp->m_sb_lock); | |
1275 | return 0; | |
1276 | } | |
1277 | printk_once(KERN_WARNING | |
1278 | "Filesystem \"%s\": reserve blocks depleted! " | |
1279 | "Consider increasing reserve pool size.", | |
1280 | mp->m_fsname); | |
1281 | fdblocks_enospc: | |
1282 | spin_unlock(&mp->m_sb_lock); | |
1283 | return -ENOSPC; | |
1284 | } | |
1285 | ||
bab98bbe DC |
1286 | int |
1287 | xfs_mod_frextents( | |
1288 | struct xfs_mount *mp, | |
1289 | int64_t delta) | |
1290 | { | |
1291 | int64_t lcounter; | |
1292 | int ret = 0; | |
1293 | ||
1294 | spin_lock(&mp->m_sb_lock); | |
1295 | lcounter = mp->m_sb.sb_frextents + delta; | |
1296 | if (lcounter < 0) | |
1297 | ret = -ENOSPC; | |
1298 | else | |
1299 | mp->m_sb.sb_frextents = lcounter; | |
1300 | spin_unlock(&mp->m_sb_lock); | |
1301 | return ret; | |
1302 | } | |
1303 | ||
1da177e4 LT |
1304 | /* |
1305 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1306 | * The buffer is returned locked and read in from disk. | |
1307 | * The buffer should be released with a call to xfs_brelse(). | |
1da177e4 | 1308 | */ |
0c842ad4 | 1309 | struct xfs_buf * |
1da177e4 | 1310 | xfs_getsb( |
8c9ce2f7 | 1311 | struct xfs_mount *mp) |
1da177e4 | 1312 | { |
0c842ad4 | 1313 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1314 | |
8c9ce2f7 | 1315 | xfs_buf_lock(bp); |
72790aa1 | 1316 | xfs_buf_hold(bp); |
b0388bf1 | 1317 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1318 | return bp; |
1da177e4 LT |
1319 | } |
1320 | ||
1321 | /* | |
1322 | * Used to free the superblock along various error paths. | |
1323 | */ | |
1324 | void | |
1325 | xfs_freesb( | |
26af6552 | 1326 | struct xfs_mount *mp) |
1da177e4 | 1327 | { |
26af6552 | 1328 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1329 | |
26af6552 | 1330 | xfs_buf_lock(bp); |
1da177e4 | 1331 | mp->m_sb_bp = NULL; |
26af6552 | 1332 | xfs_buf_relse(bp); |
1da177e4 LT |
1333 | } |
1334 | ||
dda35b8f CH |
1335 | /* |
1336 | * If the underlying (data/log/rt) device is readonly, there are some | |
1337 | * operations that cannot proceed. | |
1338 | */ | |
1339 | int | |
1340 | xfs_dev_is_read_only( | |
1341 | struct xfs_mount *mp, | |
1342 | char *message) | |
1343 | { | |
1344 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1345 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1346 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1347 | xfs_notice(mp, "%s required on read-only device.", message); |
1348 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1349 | return -EROFS; |
dda35b8f CH |
1350 | } |
1351 | return 0; | |
1352 | } | |
f467cad9 DW |
1353 | |
1354 | /* Force the summary counters to be recalculated at next mount. */ | |
1355 | void | |
1356 | xfs_force_summary_recalc( | |
1357 | struct xfs_mount *mp) | |
1358 | { | |
1359 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1360 | return; | |
1361 | ||
39353ff6 | 1362 | xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS); |
f467cad9 | 1363 | } |
9fe82b8c DW |
1364 | |
1365 | /* | |
1366 | * Update the in-core delayed block counter. | |
1367 | * | |
1368 | * We prefer to update the counter without having to take a spinlock for every | |
1369 | * counter update (i.e. batching). Each change to delayed allocation | |
1370 | * reservations can change can easily exceed the default percpu counter | |
1371 | * batching, so we use a larger batch factor here. | |
1372 | * | |
1373 | * Note that we don't currently have any callers requiring fast summation | |
1374 | * (e.g. percpu_counter_read) so we can use a big batch value here. | |
1375 | */ | |
1376 | #define XFS_DELALLOC_BATCH (4096) | |
1377 | void | |
1378 | xfs_mod_delalloc( | |
1379 | struct xfs_mount *mp, | |
1380 | int64_t delta) | |
1381 | { | |
1382 | percpu_counter_add_batch(&mp->m_delalloc_blks, delta, | |
1383 | XFS_DELALLOC_BATCH); | |
1384 | } |