Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 | 27 | #include "xfs_da_format.h" |
9a2cc41c | 28 | #include "xfs_da_btree.h" |
1da177e4 | 29 | #include "xfs_inode.h" |
a4fbe6ab | 30 | #include "xfs_dir2.h" |
a844f451 | 31 | #include "xfs_ialloc.h" |
1da177e4 LT |
32 | #include "xfs_alloc.h" |
33 | #include "xfs_rtalloc.h" | |
34 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
35 | #include "xfs_trans.h" |
36 | #include "xfs_trans_priv.h" | |
37 | #include "xfs_log.h" | |
1da177e4 | 38 | #include "xfs_error.h" |
1da177e4 LT |
39 | #include "xfs_quota.h" |
40 | #include "xfs_fsops.h" | |
0b1b213f | 41 | #include "xfs_trace.h" |
6d8b79cf | 42 | #include "xfs_icache.h" |
a31b1d3d | 43 | #include "xfs_sysfs.h" |
0b1b213f | 44 | |
1da177e4 | 45 | |
27174203 CH |
46 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
47 | static int xfs_uuid_table_size; | |
48 | static uuid_t *xfs_uuid_table; | |
49 | ||
af3b6382 DW |
50 | void |
51 | xfs_uuid_table_free(void) | |
52 | { | |
53 | if (xfs_uuid_table_size == 0) | |
54 | return; | |
55 | kmem_free(xfs_uuid_table); | |
56 | xfs_uuid_table = NULL; | |
57 | xfs_uuid_table_size = 0; | |
58 | } | |
59 | ||
27174203 CH |
60 | /* |
61 | * See if the UUID is unique among mounted XFS filesystems. | |
62 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
63 | */ | |
64 | STATIC int | |
65 | xfs_uuid_mount( | |
66 | struct xfs_mount *mp) | |
67 | { | |
68 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
69 | int hole, i; | |
70 | ||
71 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
72 | return 0; | |
73 | ||
74 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 75 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 76 | return -EINVAL; |
27174203 CH |
77 | } |
78 | ||
79 | mutex_lock(&xfs_uuid_table_mutex); | |
80 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
81 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
82 | hole = i; | |
83 | continue; | |
84 | } | |
85 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
86 | goto out_duplicate; | |
87 | } | |
88 | ||
89 | if (hole < 0) { | |
90 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
91 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
27174203 CH |
92 | KM_SLEEP); |
93 | hole = xfs_uuid_table_size++; | |
94 | } | |
95 | xfs_uuid_table[hole] = *uuid; | |
96 | mutex_unlock(&xfs_uuid_table_mutex); | |
97 | ||
98 | return 0; | |
99 | ||
100 | out_duplicate: | |
101 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 102 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 103 | return -EINVAL; |
27174203 CH |
104 | } |
105 | ||
106 | STATIC void | |
107 | xfs_uuid_unmount( | |
108 | struct xfs_mount *mp) | |
109 | { | |
110 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
111 | int i; | |
112 | ||
113 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
114 | return; | |
115 | ||
116 | mutex_lock(&xfs_uuid_table_mutex); | |
117 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
118 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
119 | continue; | |
120 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
121 | continue; | |
122 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
123 | break; | |
124 | } | |
125 | ASSERT(i < xfs_uuid_table_size); | |
126 | mutex_unlock(&xfs_uuid_table_mutex); | |
127 | } | |
128 | ||
129 | ||
e176579e DC |
130 | STATIC void |
131 | __xfs_free_perag( | |
132 | struct rcu_head *head) | |
133 | { | |
134 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
135 | ||
136 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
137 | kmem_free(pag); | |
138 | } | |
139 | ||
1da177e4 | 140 | /* |
e176579e | 141 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 142 | */ |
c962fb79 | 143 | STATIC void |
ff4f038c | 144 | xfs_free_perag( |
745f6919 | 145 | xfs_mount_t *mp) |
1da177e4 | 146 | { |
1c1c6ebc DC |
147 | xfs_agnumber_t agno; |
148 | struct xfs_perag *pag; | |
149 | ||
150 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
151 | spin_lock(&mp->m_perag_lock); | |
152 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
153 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 154 | ASSERT(pag); |
f83282a8 | 155 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 156 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 157 | } |
1da177e4 LT |
158 | } |
159 | ||
4cc929ee NS |
160 | /* |
161 | * Check size of device based on the (data/realtime) block count. | |
162 | * Note: this check is used by the growfs code as well as mount. | |
163 | */ | |
164 | int | |
165 | xfs_sb_validate_fsb_count( | |
166 | xfs_sb_t *sbp, | |
167 | __uint64_t nblocks) | |
168 | { | |
169 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
170 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
171 | ||
d5cf09ba | 172 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 173 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 174 | return -EFBIG; |
4cc929ee NS |
175 | return 0; |
176 | } | |
1da177e4 | 177 | |
1c1c6ebc | 178 | int |
c11e2c36 | 179 | xfs_initialize_perag( |
c11e2c36 | 180 | xfs_mount_t *mp, |
1c1c6ebc DC |
181 | xfs_agnumber_t agcount, |
182 | xfs_agnumber_t *maxagi) | |
1da177e4 | 183 | { |
2d2194f6 | 184 | xfs_agnumber_t index; |
8b26c582 | 185 | xfs_agnumber_t first_initialised = 0; |
1da177e4 | 186 | xfs_perag_t *pag; |
8b26c582 | 187 | int error = -ENOMEM; |
1da177e4 | 188 | |
1c1c6ebc DC |
189 | /* |
190 | * Walk the current per-ag tree so we don't try to initialise AGs | |
191 | * that already exist (growfs case). Allocate and insert all the | |
192 | * AGs we don't find ready for initialisation. | |
193 | */ | |
194 | for (index = 0; index < agcount; index++) { | |
195 | pag = xfs_perag_get(mp, index); | |
196 | if (pag) { | |
197 | xfs_perag_put(pag); | |
198 | continue; | |
199 | } | |
8b26c582 DC |
200 | if (!first_initialised) |
201 | first_initialised = index; | |
fb3b504a | 202 | |
1c1c6ebc DC |
203 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
204 | if (!pag) | |
8b26c582 | 205 | goto out_unwind; |
fb3b504a CH |
206 | pag->pag_agno = index; |
207 | pag->pag_mount = mp; | |
1a427ab0 | 208 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 209 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 210 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
211 | spin_lock_init(&pag->pag_buf_lock); |
212 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 213 | |
1c1c6ebc | 214 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 215 | goto out_unwind; |
fb3b504a | 216 | |
1c1c6ebc DC |
217 | spin_lock(&mp->m_perag_lock); |
218 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
219 | BUG(); | |
220 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
221 | radix_tree_preload_end(); |
222 | error = -EEXIST; | |
223 | goto out_unwind; | |
1c1c6ebc DC |
224 | } |
225 | spin_unlock(&mp->m_perag_lock); | |
226 | radix_tree_preload_end(); | |
227 | } | |
228 | ||
12c3f05c | 229 | index = xfs_set_inode_alloc(mp, agcount); |
fb3b504a | 230 | |
1c1c6ebc DC |
231 | if (maxagi) |
232 | *maxagi = index; | |
233 | return 0; | |
8b26c582 DC |
234 | |
235 | out_unwind: | |
236 | kmem_free(pag); | |
237 | for (; index > first_initialised; index--) { | |
238 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
239 | kmem_free(pag); | |
240 | } | |
241 | return error; | |
1da177e4 LT |
242 | } |
243 | ||
1da177e4 LT |
244 | /* |
245 | * xfs_readsb | |
246 | * | |
247 | * Does the initial read of the superblock. | |
248 | */ | |
249 | int | |
ff55068c DC |
250 | xfs_readsb( |
251 | struct xfs_mount *mp, | |
252 | int flags) | |
1da177e4 LT |
253 | { |
254 | unsigned int sector_size; | |
04a1e6c5 DC |
255 | struct xfs_buf *bp; |
256 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 257 | int error; |
af34e09d | 258 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 259 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
260 | |
261 | ASSERT(mp->m_sb_bp == NULL); | |
262 | ASSERT(mp->m_ddev_targp != NULL); | |
263 | ||
daba5427 ES |
264 | /* |
265 | * For the initial read, we must guess at the sector | |
266 | * size based on the block device. It's enough to | |
267 | * get the sb_sectsize out of the superblock and | |
268 | * then reread with the proper length. | |
269 | * We don't verify it yet, because it may not be complete. | |
270 | */ | |
271 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
272 | buf_ops = NULL; | |
273 | ||
1da177e4 | 274 | /* |
c891c30a BF |
275 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
276 | * around at all times to optimize access to the superblock. Therefore, | |
277 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
278 | * elevated. | |
1da177e4 | 279 | */ |
26af6552 | 280 | reread: |
ba372674 | 281 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
282 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
283 | buf_ops); | |
ba372674 | 284 | if (error) { |
eab4e633 | 285 | if (loud) |
e721f504 | 286 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 287 | /* bad CRC means corrupted metadata */ |
2451337d DC |
288 | if (error == -EFSBADCRC) |
289 | error = -EFSCORRUPTED; | |
ba372674 | 290 | return error; |
eab4e633 | 291 | } |
1da177e4 LT |
292 | |
293 | /* | |
294 | * Initialize the mount structure from the superblock. | |
1da177e4 | 295 | */ |
556b8883 | 296 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
297 | |
298 | /* | |
299 | * If we haven't validated the superblock, do so now before we try | |
300 | * to check the sector size and reread the superblock appropriately. | |
301 | */ | |
302 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
303 | if (loud) | |
304 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 305 | error = -EINVAL; |
556b8883 DC |
306 | goto release_buf; |
307 | } | |
ff55068c | 308 | |
1da177e4 LT |
309 | /* |
310 | * We must be able to do sector-sized and sector-aligned IO. | |
311 | */ | |
04a1e6c5 | 312 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
313 | if (loud) |
314 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 315 | sector_size, sbp->sb_sectsize); |
2451337d | 316 | error = -ENOSYS; |
26af6552 | 317 | goto release_buf; |
1da177e4 LT |
318 | } |
319 | ||
daba5427 | 320 | if (buf_ops == NULL) { |
556b8883 DC |
321 | /* |
322 | * Re-read the superblock so the buffer is correctly sized, | |
323 | * and properly verified. | |
324 | */ | |
1da177e4 | 325 | xfs_buf_relse(bp); |
04a1e6c5 | 326 | sector_size = sbp->sb_sectsize; |
daba5427 | 327 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 328 | goto reread; |
1da177e4 LT |
329 | } |
330 | ||
5681ca40 | 331 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 332 | |
04a1e6c5 DC |
333 | /* no need to be quiet anymore, so reset the buf ops */ |
334 | bp->b_ops = &xfs_sb_buf_ops; | |
335 | ||
1da177e4 | 336 | mp->m_sb_bp = bp; |
26af6552 | 337 | xfs_buf_unlock(bp); |
1da177e4 LT |
338 | return 0; |
339 | ||
26af6552 DC |
340 | release_buf: |
341 | xfs_buf_relse(bp); | |
1da177e4 LT |
342 | return error; |
343 | } | |
344 | ||
1da177e4 | 345 | /* |
0771fb45 | 346 | * Update alignment values based on mount options and sb values |
1da177e4 | 347 | */ |
0771fb45 | 348 | STATIC int |
7884bc86 | 349 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 350 | { |
1da177e4 | 351 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 352 | |
4249023a | 353 | if (mp->m_dalign) { |
1da177e4 LT |
354 | /* |
355 | * If stripe unit and stripe width are not multiples | |
356 | * of the fs blocksize turn off alignment. | |
357 | */ | |
358 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
359 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
360 | xfs_warn(mp, |
361 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
362 | sbp->sb_blocksize); | |
2451337d | 363 | return -EINVAL; |
1da177e4 LT |
364 | } else { |
365 | /* | |
366 | * Convert the stripe unit and width to FSBs. | |
367 | */ | |
368 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
369 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 370 | xfs_warn(mp, |
39a45d84 JL |
371 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
372 | sbp->sb_agblocks); | |
2451337d | 373 | return -EINVAL; |
1da177e4 LT |
374 | } else if (mp->m_dalign) { |
375 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
376 | } else { | |
39a45d84 JL |
377 | xfs_warn(mp, |
378 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
379 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 380 | return -EINVAL; |
1da177e4 LT |
381 | } |
382 | } | |
383 | ||
384 | /* | |
385 | * Update superblock with new values | |
386 | * and log changes | |
387 | */ | |
62118709 | 388 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
389 | if (sbp->sb_unit != mp->m_dalign) { |
390 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 391 | mp->m_update_sb = true; |
1da177e4 LT |
392 | } |
393 | if (sbp->sb_width != mp->m_swidth) { | |
394 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 395 | mp->m_update_sb = true; |
1da177e4 | 396 | } |
34d7f603 JL |
397 | } else { |
398 | xfs_warn(mp, | |
399 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 400 | return -EINVAL; |
1da177e4 LT |
401 | } |
402 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 403 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
404 | mp->m_dalign = sbp->sb_unit; |
405 | mp->m_swidth = sbp->sb_width; | |
406 | } | |
407 | ||
0771fb45 ES |
408 | return 0; |
409 | } | |
1da177e4 | 410 | |
0771fb45 ES |
411 | /* |
412 | * Set the maximum inode count for this filesystem | |
413 | */ | |
414 | STATIC void | |
415 | xfs_set_maxicount(xfs_mount_t *mp) | |
416 | { | |
417 | xfs_sb_t *sbp = &(mp->m_sb); | |
418 | __uint64_t icount; | |
1da177e4 | 419 | |
0771fb45 ES |
420 | if (sbp->sb_imax_pct) { |
421 | /* | |
422 | * Make sure the maximum inode count is a multiple | |
423 | * of the units we allocate inodes in. | |
1da177e4 | 424 | */ |
1da177e4 LT |
425 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
426 | do_div(icount, 100); | |
427 | do_div(icount, mp->m_ialloc_blks); | |
428 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
429 | sbp->sb_inopblog; | |
0771fb45 | 430 | } else { |
1da177e4 | 431 | mp->m_maxicount = 0; |
1da177e4 | 432 | } |
0771fb45 ES |
433 | } |
434 | ||
435 | /* | |
436 | * Set the default minimum read and write sizes unless | |
437 | * already specified in a mount option. | |
438 | * We use smaller I/O sizes when the file system | |
439 | * is being used for NFS service (wsync mount option). | |
440 | */ | |
441 | STATIC void | |
442 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
443 | { | |
444 | xfs_sb_t *sbp = &(mp->m_sb); | |
445 | int readio_log, writeio_log; | |
1da177e4 | 446 | |
1da177e4 LT |
447 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
448 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
449 | readio_log = XFS_WSYNC_READIO_LOG; | |
450 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
451 | } else { | |
452 | readio_log = XFS_READIO_LOG_LARGE; | |
453 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
454 | } | |
455 | } else { | |
456 | readio_log = mp->m_readio_log; | |
457 | writeio_log = mp->m_writeio_log; | |
458 | } | |
459 | ||
1da177e4 LT |
460 | if (sbp->sb_blocklog > readio_log) { |
461 | mp->m_readio_log = sbp->sb_blocklog; | |
462 | } else { | |
463 | mp->m_readio_log = readio_log; | |
464 | } | |
465 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
466 | if (sbp->sb_blocklog > writeio_log) { | |
467 | mp->m_writeio_log = sbp->sb_blocklog; | |
468 | } else { | |
469 | mp->m_writeio_log = writeio_log; | |
470 | } | |
471 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 472 | } |
1da177e4 | 473 | |
055388a3 DC |
474 | /* |
475 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
476 | */ | |
477 | void | |
478 | xfs_set_low_space_thresholds( | |
479 | struct xfs_mount *mp) | |
480 | { | |
481 | int i; | |
482 | ||
483 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
484 | __uint64_t space = mp->m_sb.sb_dblocks; | |
485 | ||
486 | do_div(space, 100); | |
487 | mp->m_low_space[i] = space * (i + 1); | |
488 | } | |
489 | } | |
490 | ||
491 | ||
0771fb45 ES |
492 | /* |
493 | * Set whether we're using inode alignment. | |
494 | */ | |
495 | STATIC void | |
496 | xfs_set_inoalignment(xfs_mount_t *mp) | |
497 | { | |
62118709 | 498 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
499 | mp->m_sb.sb_inoalignmt >= |
500 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
501 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
502 | else | |
503 | mp->m_inoalign_mask = 0; | |
504 | /* | |
505 | * If we are using stripe alignment, check whether | |
506 | * the stripe unit is a multiple of the inode alignment | |
507 | */ | |
508 | if (mp->m_dalign && mp->m_inoalign_mask && | |
509 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
510 | mp->m_sinoalign = mp->m_dalign; | |
511 | else | |
512 | mp->m_sinoalign = 0; | |
0771fb45 ES |
513 | } |
514 | ||
515 | /* | |
0471f62e | 516 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
517 | */ |
518 | STATIC int | |
ba372674 DC |
519 | xfs_check_sizes( |
520 | struct xfs_mount *mp) | |
0771fb45 | 521 | { |
ba372674 | 522 | struct xfs_buf *bp; |
0771fb45 | 523 | xfs_daddr_t d; |
ba372674 | 524 | int error; |
0771fb45 | 525 | |
1da177e4 LT |
526 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
527 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 528 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 529 | return -EFBIG; |
1da177e4 | 530 | } |
ba372674 | 531 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 532 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
533 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
534 | if (error) { | |
0b932ccc | 535 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 536 | return error; |
1da177e4 | 537 | } |
1922c949 | 538 | xfs_buf_relse(bp); |
1da177e4 | 539 | |
ba372674 DC |
540 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
541 | return 0; | |
542 | ||
543 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
544 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
545 | xfs_warn(mp, "log size mismatch detected"); | |
546 | return -EFBIG; | |
547 | } | |
548 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 549 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
550 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
551 | if (error) { | |
552 | xfs_warn(mp, "log device read failed"); | |
553 | return error; | |
0771fb45 | 554 | } |
ba372674 | 555 | xfs_buf_relse(bp); |
0771fb45 ES |
556 | return 0; |
557 | } | |
558 | ||
7d095257 CH |
559 | /* |
560 | * Clear the quotaflags in memory and in the superblock. | |
561 | */ | |
562 | int | |
563 | xfs_mount_reset_sbqflags( | |
564 | struct xfs_mount *mp) | |
565 | { | |
7d095257 CH |
566 | mp->m_qflags = 0; |
567 | ||
61e63ecb | 568 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
569 | if (mp->m_sb.sb_qflags == 0) |
570 | return 0; | |
571 | spin_lock(&mp->m_sb_lock); | |
572 | mp->m_sb.sb_qflags = 0; | |
573 | spin_unlock(&mp->m_sb_lock); | |
574 | ||
61e63ecb | 575 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
576 | return 0; |
577 | ||
61e63ecb | 578 | return xfs_sync_sb(mp, false); |
7d095257 CH |
579 | } |
580 | ||
d5db0f97 ES |
581 | __uint64_t |
582 | xfs_default_resblks(xfs_mount_t *mp) | |
583 | { | |
584 | __uint64_t resblks; | |
585 | ||
586 | /* | |
8babd8a2 DC |
587 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
588 | * smaller. This is intended to cover concurrent allocation | |
589 | * transactions when we initially hit enospc. These each require a 4 | |
590 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
591 | * allocation reservations. | |
d5db0f97 ES |
592 | */ |
593 | resblks = mp->m_sb.sb_dblocks; | |
594 | do_div(resblks, 20); | |
8babd8a2 | 595 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
596 | return resblks; |
597 | } | |
598 | ||
0771fb45 | 599 | /* |
0771fb45 ES |
600 | * This function does the following on an initial mount of a file system: |
601 | * - reads the superblock from disk and init the mount struct | |
602 | * - if we're a 32-bit kernel, do a size check on the superblock | |
603 | * so we don't mount terabyte filesystems | |
604 | * - init mount struct realtime fields | |
605 | * - allocate inode hash table for fs | |
606 | * - init directory manager | |
607 | * - perform recovery and init the log manager | |
608 | */ | |
609 | int | |
610 | xfs_mountfs( | |
f0b2efad | 611 | struct xfs_mount *mp) |
0771fb45 | 612 | { |
f0b2efad BF |
613 | struct xfs_sb *sbp = &(mp->m_sb); |
614 | struct xfs_inode *rip; | |
615 | __uint64_t resblks; | |
616 | uint quotamount = 0; | |
617 | uint quotaflags = 0; | |
618 | int error = 0; | |
0771fb45 | 619 | |
ff55068c | 620 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 621 | |
ee1c0908 | 622 | /* |
074e427b DC |
623 | * Check for a mismatched features2 values. Older kernels read & wrote |
624 | * into the wrong sb offset for sb_features2 on some platforms due to | |
625 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
626 | * which made older superblock reading/writing routines swap it as a | |
627 | * 64-bit value. | |
ee1c0908 | 628 | * |
e6957ea4 ES |
629 | * For backwards compatibility, we make both slots equal. |
630 | * | |
074e427b DC |
631 | * If we detect a mismatched field, we OR the set bits into the existing |
632 | * features2 field in case it has already been modified; we don't want | |
633 | * to lose any features. We then update the bad location with the ORed | |
634 | * value so that older kernels will see any features2 flags. The | |
635 | * superblock writeback code ensures the new sb_features2 is copied to | |
636 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 637 | */ |
e6957ea4 | 638 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 639 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 640 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 641 | mp->m_update_sb = true; |
e6957ea4 ES |
642 | |
643 | /* | |
644 | * Re-check for ATTR2 in case it was found in bad_features2 | |
645 | * slot. | |
646 | */ | |
7c12f296 TS |
647 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
648 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 649 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
650 | } |
651 | ||
652 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
653 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
654 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 655 | mp->m_update_sb = true; |
e6957ea4 | 656 | |
7c12f296 TS |
657 | /* update sb_versionnum for the clearing of the morebits */ |
658 | if (!sbp->sb_features2) | |
61e63ecb | 659 | mp->m_update_sb = true; |
ee1c0908 DC |
660 | } |
661 | ||
263997a6 DC |
662 | /* always use v2 inodes by default now */ |
663 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
664 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 665 | mp->m_update_sb = true; |
263997a6 DC |
666 | } |
667 | ||
0771fb45 ES |
668 | /* |
669 | * Check if sb_agblocks is aligned at stripe boundary | |
670 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
671 | * allocator alignment is within an ag, therefore ag has | |
672 | * to be aligned at stripe boundary. | |
673 | */ | |
7884bc86 | 674 | error = xfs_update_alignment(mp); |
0771fb45 | 675 | if (error) |
f9057e3d | 676 | goto out; |
0771fb45 ES |
677 | |
678 | xfs_alloc_compute_maxlevels(mp); | |
679 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
680 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
681 | xfs_ialloc_compute_maxlevels(mp); | |
682 | ||
683 | xfs_set_maxicount(mp); | |
684 | ||
e6b3bb78 CM |
685 | /* enable fail_at_unmount as default */ |
686 | mp->m_fail_unmount = 1; | |
687 | ||
a31b1d3d | 688 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
689 | if (error) |
690 | goto out; | |
1da177e4 | 691 | |
225e4635 BD |
692 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
693 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
694 | if (error) |
695 | goto out_remove_sysfs; | |
696 | ||
192852be | 697 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
698 | if (error) |
699 | goto out_del_stats; | |
700 | ||
192852be CM |
701 | |
702 | error = xfs_uuid_mount(mp); | |
703 | if (error) | |
704 | goto out_remove_error_sysfs; | |
705 | ||
0771fb45 ES |
706 | /* |
707 | * Set the minimum read and write sizes | |
708 | */ | |
709 | xfs_set_rw_sizes(mp); | |
710 | ||
055388a3 DC |
711 | /* set the low space thresholds for dynamic preallocation */ |
712 | xfs_set_low_space_thresholds(mp); | |
713 | ||
0771fb45 ES |
714 | /* |
715 | * Set the inode cluster size. | |
716 | * This may still be overridden by the file system | |
717 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
718 | * |
719 | * For v5 filesystems, scale the cluster size with the inode size to | |
720 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
721 | * has set the inode alignment value appropriately for larger cluster | |
722 | * sizes. | |
0771fb45 ES |
723 | */ |
724 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
725 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
726 | int new_size = mp->m_inode_cluster_size; | |
727 | ||
728 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
729 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
730 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 731 | } |
0771fb45 | 732 | |
e5376fc1 BF |
733 | /* |
734 | * If enabled, sparse inode chunk alignment is expected to match the | |
735 | * cluster size. Full inode chunk alignment must match the chunk size, | |
736 | * but that is checked on sb read verification... | |
737 | */ | |
738 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
739 | mp->m_sb.sb_spino_align != | |
740 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { | |
741 | xfs_warn(mp, | |
742 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
743 | mp->m_sb.sb_spino_align, | |
744 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); | |
745 | error = -EINVAL; | |
746 | goto out_remove_uuid; | |
747 | } | |
748 | ||
0771fb45 ES |
749 | /* |
750 | * Set inode alignment fields | |
751 | */ | |
752 | xfs_set_inoalignment(mp); | |
753 | ||
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 | */ | |
774 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
775 | ||
1da177e4 LT |
776 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
777 | ||
0650b554 DC |
778 | error = xfs_da_mount(mp); |
779 | if (error) { | |
780 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
781 | goto out_remove_uuid; | |
782 | } | |
1da177e4 LT |
783 | |
784 | /* | |
785 | * Initialize the precomputed transaction reservations values. | |
786 | */ | |
787 | xfs_trans_init(mp); | |
788 | ||
1da177e4 LT |
789 | /* |
790 | * Allocate and initialize the per-ag data. | |
791 | */ | |
1c1c6ebc | 792 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 793 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
794 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
795 | if (error) { | |
0b932ccc | 796 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 797 | goto out_free_dir; |
1c1c6ebc | 798 | } |
1da177e4 | 799 | |
f9057e3d | 800 | if (!sbp->sb_logblocks) { |
0b932ccc | 801 | xfs_warn(mp, "no log defined"); |
f9057e3d | 802 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 803 | error = -EFSCORRUPTED; |
f9057e3d CH |
804 | goto out_free_perag; |
805 | } | |
806 | ||
1da177e4 | 807 | /* |
f0b2efad BF |
808 | * Log's mount-time initialization. The first part of recovery can place |
809 | * some items on the AIL, to be handled when recovery is finished or | |
810 | * cancelled. | |
1da177e4 | 811 | */ |
f9057e3d CH |
812 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
813 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
814 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
815 | if (error) { | |
0b932ccc | 816 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 817 | goto out_fail_wait; |
1da177e4 LT |
818 | } |
819 | ||
92821e2b DC |
820 | /* |
821 | * Now the log is mounted, we know if it was an unclean shutdown or | |
822 | * not. If it was, with the first phase of recovery has completed, we | |
823 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
824 | * but they are recovered transactionally in the second recovery phase | |
825 | * later. | |
826 | * | |
827 | * Hence we can safely re-initialise incore superblock counters from | |
828 | * the per-ag data. These may not be correct if the filesystem was not | |
829 | * cleanly unmounted, so we need to wait for recovery to finish before | |
830 | * doing this. | |
831 | * | |
832 | * If the filesystem was cleanly unmounted, then we can trust the | |
833 | * values in the superblock to be correct and we don't need to do | |
834 | * anything here. | |
835 | * | |
836 | * If we are currently making the filesystem, the initialisation will | |
837 | * fail as the perag data is in an undefined state. | |
838 | */ | |
92821e2b DC |
839 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
840 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
841 | !mp->m_sb.sb_inprogress) { | |
842 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 843 | if (error) |
6eee8972 | 844 | goto out_log_dealloc; |
92821e2b | 845 | } |
f9057e3d | 846 | |
1da177e4 LT |
847 | /* |
848 | * Get and sanity-check the root inode. | |
849 | * Save the pointer to it in the mount structure. | |
850 | */ | |
7b6259e7 | 851 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 852 | if (error) { |
0b932ccc | 853 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 854 | goto out_log_dealloc; |
1da177e4 LT |
855 | } |
856 | ||
857 | ASSERT(rip != NULL); | |
1da177e4 | 858 | |
c19b3b05 | 859 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 860 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 861 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
862 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
863 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
864 | mp); | |
2451337d | 865 | error = -EFSCORRUPTED; |
f9057e3d | 866 | goto out_rele_rip; |
1da177e4 LT |
867 | } |
868 | mp->m_rootip = rip; /* save it */ | |
869 | ||
870 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
871 | ||
872 | /* | |
873 | * Initialize realtime inode pointers in the mount structure | |
874 | */ | |
0771fb45 ES |
875 | error = xfs_rtmount_inodes(mp); |
876 | if (error) { | |
1da177e4 LT |
877 | /* |
878 | * Free up the root inode. | |
879 | */ | |
0b932ccc | 880 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 881 | goto out_rele_rip; |
1da177e4 LT |
882 | } |
883 | ||
884 | /* | |
7884bc86 CH |
885 | * If this is a read-only mount defer the superblock updates until |
886 | * the next remount into writeable mode. Otherwise we would never | |
887 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 888 | */ |
61e63ecb DC |
889 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
890 | error = xfs_sync_sb(mp, false); | |
e5720eec | 891 | if (error) { |
0b932ccc | 892 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 893 | goto out_rtunmount; |
e5720eec DC |
894 | } |
895 | } | |
1da177e4 LT |
896 | |
897 | /* | |
898 | * Initialise the XFS quota management subsystem for this mount | |
899 | */ | |
7d095257 CH |
900 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
901 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
902 | if (error) | |
903 | goto out_rtunmount; | |
904 | } else { | |
905 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
906 | ||
907 | /* | |
908 | * If a file system had quotas running earlier, but decided to | |
909 | * mount without -o uquota/pquota/gquota options, revoke the | |
910 | * quotachecked license. | |
911 | */ | |
912 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 913 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
914 | error = xfs_mount_reset_sbqflags(mp); |
915 | if (error) | |
a70a4fa5 | 916 | goto out_rtunmount; |
7d095257 CH |
917 | } |
918 | } | |
1da177e4 LT |
919 | |
920 | /* | |
f0b2efad BF |
921 | * Finish recovering the file system. This part needed to be delayed |
922 | * until after the root and real-time bitmap inodes were consistently | |
923 | * read in. | |
1da177e4 | 924 | */ |
4249023a | 925 | error = xfs_log_mount_finish(mp); |
1da177e4 | 926 | if (error) { |
0b932ccc | 927 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 928 | goto out_rtunmount; |
1da177e4 LT |
929 | } |
930 | ||
931 | /* | |
932 | * Complete the quota initialisation, post-log-replay component. | |
933 | */ | |
7d095257 CH |
934 | if (quotamount) { |
935 | ASSERT(mp->m_qflags == 0); | |
936 | mp->m_qflags = quotaflags; | |
937 | ||
938 | xfs_qm_mount_quotas(mp); | |
939 | } | |
940 | ||
84e1e99f DC |
941 | /* |
942 | * Now we are mounted, reserve a small amount of unused space for | |
943 | * privileged transactions. This is needed so that transaction | |
944 | * space required for critical operations can dip into this pool | |
945 | * when at ENOSPC. This is needed for operations like create with | |
946 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
947 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
948 | * |
949 | * This may drive us straight to ENOSPC on mount, but that implies | |
950 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 951 | */ |
d5db0f97 ES |
952 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
953 | resblks = xfs_default_resblks(mp); | |
954 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
955 | if (error) | |
0b932ccc DC |
956 | xfs_warn(mp, |
957 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 958 | } |
84e1e99f | 959 | |
1da177e4 LT |
960 | return 0; |
961 | ||
b93b6e43 CH |
962 | out_rtunmount: |
963 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 964 | out_rele_rip: |
43355099 | 965 | IRELE(rip); |
0ae120f8 BF |
966 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
967 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
f9057e3d | 968 | out_log_dealloc: |
e6b3bb78 | 969 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 970 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
971 | out_fail_wait: |
972 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
973 | xfs_wait_buftarg(mp->m_logdev_targp); | |
974 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 975 | out_free_perag: |
ff4f038c | 976 | xfs_free_perag(mp); |
0650b554 DC |
977 | out_free_dir: |
978 | xfs_da_unmount(mp); | |
f9057e3d | 979 | out_remove_uuid: |
27174203 | 980 | xfs_uuid_unmount(mp); |
192852be CM |
981 | out_remove_error_sysfs: |
982 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
983 | out_del_stats: |
984 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
985 | out_remove_sysfs: |
986 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 987 | out: |
1da177e4 LT |
988 | return error; |
989 | } | |
990 | ||
991 | /* | |
1da177e4 LT |
992 | * This flushes out the inodes,dquots and the superblock, unmounts the |
993 | * log and makes sure that incore structures are freed. | |
994 | */ | |
41b5c2e7 CH |
995 | void |
996 | xfs_unmountfs( | |
997 | struct xfs_mount *mp) | |
1da177e4 | 998 | { |
41b5c2e7 CH |
999 | __uint64_t resblks; |
1000 | int error; | |
1da177e4 | 1001 | |
579b62fa BF |
1002 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1003 | ||
7d095257 | 1004 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1005 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1006 | IRELE(mp->m_rootip); |
1007 | ||
641c56fb DC |
1008 | /* |
1009 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1010 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1011 | * the transaction is still sitting in a iclog. The stale inodes |
1012 | * on that buffer will have their flush locks held until the | |
1013 | * transaction hits the disk and the callbacks run. the inode | |
1014 | * flush takes the flush lock unconditionally and with nothing to | |
1015 | * push out the iclog we will never get that unlocked. hence we | |
1016 | * need to force the log first. | |
1017 | */ | |
a14a348b | 1018 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1019 | |
e6b3bb78 CM |
1020 | /* |
1021 | * We now need to tell the world we are unmounting. This will allow | |
1022 | * us to detect that the filesystem is going away and we should error | |
1023 | * out anything that we have been retrying in the background. This will | |
1024 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1025 | */ | |
1026 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1027 | ||
c854363e | 1028 | /* |
211e4d43 CH |
1029 | * Flush all pending changes from the AIL. |
1030 | */ | |
1031 | xfs_ail_push_all_sync(mp->m_ail); | |
1032 | ||
1033 | /* | |
1034 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1035 | * inodes and none should be pinned or locked, but use synchronous |
1036 | * reclaim just to be sure. We can stop background inode reclaim | |
1037 | * here as well if it is still running. | |
c854363e | 1038 | */ |
7e18530b | 1039 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1040 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1041 | |
7d095257 | 1042 | xfs_qm_unmount(mp); |
a357a121 | 1043 | |
84e1e99f DC |
1044 | /* |
1045 | * Unreserve any blocks we have so that when we unmount we don't account | |
1046 | * the reserved free space as used. This is really only necessary for | |
1047 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1048 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1049 | * |
1050 | * We don't bother correcting this elsewhere for lazy superblock | |
1051 | * counting because on mount of an unclean filesystem we reconstruct the | |
1052 | * correct counter value and this is irrelevant. | |
1053 | * | |
1054 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1055 | * we only every apply deltas to the superblock and hence the incore | |
1056 | * value does not matter.... | |
1057 | */ | |
1058 | resblks = 0; | |
714082bc DC |
1059 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1060 | if (error) | |
0b932ccc | 1061 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1062 | "Freespace may not be correct on next mount."); |
1063 | ||
adab0f67 | 1064 | error = xfs_log_sbcount(mp); |
e5720eec | 1065 | if (error) |
0b932ccc | 1066 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1067 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1068 | |
225e4635 | 1069 | |
21b699c8 | 1070 | xfs_log_unmount(mp); |
0650b554 | 1071 | xfs_da_unmount(mp); |
27174203 | 1072 | xfs_uuid_unmount(mp); |
1da177e4 | 1073 | |
1550d0b0 | 1074 | #if defined(DEBUG) |
0ce4cfd4 | 1075 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1076 | #endif |
ff4f038c | 1077 | xfs_free_perag(mp); |
a31b1d3d | 1078 | |
192852be | 1079 | xfs_error_sysfs_del(mp); |
225e4635 | 1080 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1081 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1082 | } |
1083 | ||
91ee575f BF |
1084 | /* |
1085 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1086 | * freeze level for which modifications should not be allowed. This allows | |
1087 | * certain operations to proceed while the freeze sequence is in progress, if | |
1088 | * necessary. | |
1089 | */ | |
1090 | bool | |
1091 | xfs_fs_writable( | |
1092 | struct xfs_mount *mp, | |
1093 | int level) | |
92821e2b | 1094 | { |
91ee575f BF |
1095 | ASSERT(level > SB_UNFROZEN); |
1096 | if ((mp->m_super->s_writers.frozen >= level) || | |
1097 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1098 | return false; | |
1099 | ||
1100 | return true; | |
92821e2b DC |
1101 | } |
1102 | ||
1103 | /* | |
b2ce3974 AE |
1104 | * xfs_log_sbcount |
1105 | * | |
adab0f67 | 1106 | * Sync the superblock counters to disk. |
b2ce3974 | 1107 | * |
91ee575f BF |
1108 | * Note this code can be called during the process of freezing, so we use the |
1109 | * transaction allocator that does not block when the transaction subsystem is | |
1110 | * in its frozen state. | |
92821e2b DC |
1111 | */ |
1112 | int | |
adab0f67 | 1113 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1114 | { |
91ee575f BF |
1115 | /* allow this to proceed during the freeze sequence... */ |
1116 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1117 | return 0; |
1118 | ||
92821e2b DC |
1119 | /* |
1120 | * we don't need to do this if we are updating the superblock | |
1121 | * counters on every modification. | |
1122 | */ | |
1123 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1124 | return 0; | |
1125 | ||
61e63ecb | 1126 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1127 | } |
1128 | ||
8c1903d3 DC |
1129 | /* |
1130 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1131 | * of 128 so we don't need to take the counter lock on every update. | |
1132 | */ | |
1133 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1134 | int |
1135 | xfs_mod_icount( | |
1136 | struct xfs_mount *mp, | |
1137 | int64_t delta) | |
1138 | { | |
8c1903d3 DC |
1139 | __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
1140 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { | |
501ab323 DC |
1141 | ASSERT(0); |
1142 | percpu_counter_add(&mp->m_icount, -delta); | |
1143 | return -EINVAL; | |
1144 | } | |
1145 | return 0; | |
1146 | } | |
1147 | ||
e88b64ea DC |
1148 | int |
1149 | xfs_mod_ifree( | |
1150 | struct xfs_mount *mp, | |
1151 | int64_t delta) | |
1152 | { | |
1153 | percpu_counter_add(&mp->m_ifree, delta); | |
1154 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1155 | ASSERT(0); | |
1156 | percpu_counter_add(&mp->m_ifree, -delta); | |
1157 | return -EINVAL; | |
1158 | } | |
1159 | return 0; | |
1160 | } | |
0d485ada | 1161 | |
8c1903d3 DC |
1162 | /* |
1163 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1164 | * only occurs on frequent small block count updates such as in the delayed | |
1165 | * allocation path for buffered writes (page a time updates). Hence we set | |
1166 | * a large batch count (1024) to minimise global counter updates except when | |
1167 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1168 | */ | |
1169 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1170 | int |
1171 | xfs_mod_fdblocks( | |
1172 | struct xfs_mount *mp, | |
1173 | int64_t delta, | |
1174 | bool rsvd) | |
1175 | { | |
1176 | int64_t lcounter; | |
1177 | long long res_used; | |
1178 | s32 batch; | |
1179 | ||
1180 | if (delta > 0) { | |
1181 | /* | |
1182 | * If the reserve pool is depleted, put blocks back into it | |
1183 | * first. Most of the time the pool is full. | |
1184 | */ | |
1185 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1186 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | spin_lock(&mp->m_sb_lock); | |
1191 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1192 | ||
1193 | if (res_used > delta) { | |
1194 | mp->m_resblks_avail += delta; | |
1195 | } else { | |
1196 | delta -= res_used; | |
1197 | mp->m_resblks_avail = mp->m_resblks; | |
1198 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1199 | } | |
1200 | spin_unlock(&mp->m_sb_lock); | |
1201 | return 0; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Taking blocks away, need to be more accurate the closer we | |
1206 | * are to zero. | |
1207 | * | |
0d485ada DC |
1208 | * If the counter has a value of less than 2 * max batch size, |
1209 | * then make everything serialise as we are real close to | |
1210 | * ENOSPC. | |
1211 | */ | |
8c1903d3 DC |
1212 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1213 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1214 | batch = 1; |
1215 | else | |
8c1903d3 | 1216 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada DC |
1217 | |
1218 | __percpu_counter_add(&mp->m_fdblocks, delta, batch); | |
8c1903d3 DC |
1219 | if (__percpu_counter_compare(&mp->m_fdblocks, XFS_ALLOC_SET_ASIDE(mp), |
1220 | XFS_FDBLOCKS_BATCH) >= 0) { | |
0d485ada DC |
1221 | /* we had space! */ |
1222 | return 0; | |
1223 | } | |
1224 | ||
1225 | /* | |
1226 | * lock up the sb for dipping into reserves before releasing the space | |
1227 | * that took us to ENOSPC. | |
1228 | */ | |
1229 | spin_lock(&mp->m_sb_lock); | |
1230 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1231 | if (!rsvd) | |
1232 | goto fdblocks_enospc; | |
1233 | ||
1234 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1235 | if (lcounter >= 0) { | |
1236 | mp->m_resblks_avail = lcounter; | |
1237 | spin_unlock(&mp->m_sb_lock); | |
1238 | return 0; | |
1239 | } | |
1240 | printk_once(KERN_WARNING | |
1241 | "Filesystem \"%s\": reserve blocks depleted! " | |
1242 | "Consider increasing reserve pool size.", | |
1243 | mp->m_fsname); | |
1244 | fdblocks_enospc: | |
1245 | spin_unlock(&mp->m_sb_lock); | |
1246 | return -ENOSPC; | |
1247 | } | |
1248 | ||
bab98bbe DC |
1249 | int |
1250 | xfs_mod_frextents( | |
1251 | struct xfs_mount *mp, | |
1252 | int64_t delta) | |
1253 | { | |
1254 | int64_t lcounter; | |
1255 | int ret = 0; | |
1256 | ||
1257 | spin_lock(&mp->m_sb_lock); | |
1258 | lcounter = mp->m_sb.sb_frextents + delta; | |
1259 | if (lcounter < 0) | |
1260 | ret = -ENOSPC; | |
1261 | else | |
1262 | mp->m_sb.sb_frextents = lcounter; | |
1263 | spin_unlock(&mp->m_sb_lock); | |
1264 | return ret; | |
1265 | } | |
1266 | ||
1da177e4 LT |
1267 | /* |
1268 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1269 | * The buffer is returned locked and read in from disk. | |
1270 | * The buffer should be released with a call to xfs_brelse(). | |
1271 | * | |
1272 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1273 | * the superblock buffer if it can be locked without sleeping. | |
1274 | * If it can't then we'll return NULL. | |
1275 | */ | |
0c842ad4 | 1276 | struct xfs_buf * |
1da177e4 | 1277 | xfs_getsb( |
0c842ad4 CH |
1278 | struct xfs_mount *mp, |
1279 | int flags) | |
1da177e4 | 1280 | { |
0c842ad4 | 1281 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1282 | |
0c842ad4 CH |
1283 | if (!xfs_buf_trylock(bp)) { |
1284 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1285 | return NULL; |
0c842ad4 | 1286 | xfs_buf_lock(bp); |
1da177e4 | 1287 | } |
0c842ad4 | 1288 | |
72790aa1 | 1289 | xfs_buf_hold(bp); |
b0388bf1 | 1290 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1291 | return bp; |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | /* | |
1295 | * Used to free the superblock along various error paths. | |
1296 | */ | |
1297 | void | |
1298 | xfs_freesb( | |
26af6552 | 1299 | struct xfs_mount *mp) |
1da177e4 | 1300 | { |
26af6552 | 1301 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1302 | |
26af6552 | 1303 | xfs_buf_lock(bp); |
1da177e4 | 1304 | mp->m_sb_bp = NULL; |
26af6552 | 1305 | xfs_buf_relse(bp); |
1da177e4 LT |
1306 | } |
1307 | ||
dda35b8f CH |
1308 | /* |
1309 | * If the underlying (data/log/rt) device is readonly, there are some | |
1310 | * operations that cannot proceed. | |
1311 | */ | |
1312 | int | |
1313 | xfs_dev_is_read_only( | |
1314 | struct xfs_mount *mp, | |
1315 | char *message) | |
1316 | { | |
1317 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1318 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1319 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1320 | xfs_notice(mp, "%s required on read-only device.", message); |
1321 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1322 | return -EROFS; |
dda35b8f CH |
1323 | } |
1324 | return 0; | |
1325 | } |