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 LT |
274 | /* |
275 | * Allocate a (locked) buffer to hold the superblock. | |
276 | * This will be kept around at all times to optimize | |
277 | * access to the superblock. | |
278 | */ | |
26af6552 | 279 | reread: |
ba372674 DC |
280 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
281 | BTOBB(sector_size), 0, &bp, buf_ops); | |
282 | if (error) { | |
eab4e633 | 283 | if (loud) |
e721f504 | 284 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 285 | /* bad CRC means corrupted metadata */ |
2451337d DC |
286 | if (error == -EFSBADCRC) |
287 | error = -EFSCORRUPTED; | |
ba372674 | 288 | return error; |
eab4e633 | 289 | } |
1da177e4 LT |
290 | |
291 | /* | |
292 | * Initialize the mount structure from the superblock. | |
1da177e4 | 293 | */ |
556b8883 | 294 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
295 | |
296 | /* | |
297 | * If we haven't validated the superblock, do so now before we try | |
298 | * to check the sector size and reread the superblock appropriately. | |
299 | */ | |
300 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
301 | if (loud) | |
302 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 303 | error = -EINVAL; |
556b8883 DC |
304 | goto release_buf; |
305 | } | |
ff55068c | 306 | |
1da177e4 LT |
307 | /* |
308 | * We must be able to do sector-sized and sector-aligned IO. | |
309 | */ | |
04a1e6c5 | 310 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
311 | if (loud) |
312 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 313 | sector_size, sbp->sb_sectsize); |
2451337d | 314 | error = -ENOSYS; |
26af6552 | 315 | goto release_buf; |
1da177e4 LT |
316 | } |
317 | ||
daba5427 | 318 | if (buf_ops == NULL) { |
556b8883 DC |
319 | /* |
320 | * Re-read the superblock so the buffer is correctly sized, | |
321 | * and properly verified. | |
322 | */ | |
1da177e4 | 323 | xfs_buf_relse(bp); |
04a1e6c5 | 324 | sector_size = sbp->sb_sectsize; |
daba5427 | 325 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 326 | goto reread; |
1da177e4 LT |
327 | } |
328 | ||
5681ca40 | 329 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 330 | |
04a1e6c5 DC |
331 | /* no need to be quiet anymore, so reset the buf ops */ |
332 | bp->b_ops = &xfs_sb_buf_ops; | |
333 | ||
1da177e4 | 334 | mp->m_sb_bp = bp; |
26af6552 | 335 | xfs_buf_unlock(bp); |
1da177e4 LT |
336 | return 0; |
337 | ||
26af6552 DC |
338 | release_buf: |
339 | xfs_buf_relse(bp); | |
1da177e4 LT |
340 | return error; |
341 | } | |
342 | ||
1da177e4 | 343 | /* |
0771fb45 | 344 | * Update alignment values based on mount options and sb values |
1da177e4 | 345 | */ |
0771fb45 | 346 | STATIC int |
7884bc86 | 347 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 348 | { |
1da177e4 | 349 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 350 | |
4249023a | 351 | if (mp->m_dalign) { |
1da177e4 LT |
352 | /* |
353 | * If stripe unit and stripe width are not multiples | |
354 | * of the fs blocksize turn off alignment. | |
355 | */ | |
356 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
357 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
358 | xfs_warn(mp, |
359 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
360 | sbp->sb_blocksize); | |
2451337d | 361 | return -EINVAL; |
1da177e4 LT |
362 | } else { |
363 | /* | |
364 | * Convert the stripe unit and width to FSBs. | |
365 | */ | |
366 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
367 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 368 | xfs_warn(mp, |
39a45d84 JL |
369 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
370 | sbp->sb_agblocks); | |
2451337d | 371 | return -EINVAL; |
1da177e4 LT |
372 | } else if (mp->m_dalign) { |
373 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
374 | } else { | |
39a45d84 JL |
375 | xfs_warn(mp, |
376 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
377 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 378 | return -EINVAL; |
1da177e4 LT |
379 | } |
380 | } | |
381 | ||
382 | /* | |
383 | * Update superblock with new values | |
384 | * and log changes | |
385 | */ | |
62118709 | 386 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
387 | if (sbp->sb_unit != mp->m_dalign) { |
388 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 389 | mp->m_update_sb = true; |
1da177e4 LT |
390 | } |
391 | if (sbp->sb_width != mp->m_swidth) { | |
392 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 393 | mp->m_update_sb = true; |
1da177e4 | 394 | } |
34d7f603 JL |
395 | } else { |
396 | xfs_warn(mp, | |
397 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 398 | return -EINVAL; |
1da177e4 LT |
399 | } |
400 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 401 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
402 | mp->m_dalign = sbp->sb_unit; |
403 | mp->m_swidth = sbp->sb_width; | |
404 | } | |
405 | ||
0771fb45 ES |
406 | return 0; |
407 | } | |
1da177e4 | 408 | |
0771fb45 ES |
409 | /* |
410 | * Set the maximum inode count for this filesystem | |
411 | */ | |
412 | STATIC void | |
413 | xfs_set_maxicount(xfs_mount_t *mp) | |
414 | { | |
415 | xfs_sb_t *sbp = &(mp->m_sb); | |
416 | __uint64_t icount; | |
1da177e4 | 417 | |
0771fb45 ES |
418 | if (sbp->sb_imax_pct) { |
419 | /* | |
420 | * Make sure the maximum inode count is a multiple | |
421 | * of the units we allocate inodes in. | |
1da177e4 | 422 | */ |
1da177e4 LT |
423 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
424 | do_div(icount, 100); | |
425 | do_div(icount, mp->m_ialloc_blks); | |
426 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
427 | sbp->sb_inopblog; | |
0771fb45 | 428 | } else { |
1da177e4 | 429 | mp->m_maxicount = 0; |
1da177e4 | 430 | } |
0771fb45 ES |
431 | } |
432 | ||
433 | /* | |
434 | * Set the default minimum read and write sizes unless | |
435 | * already specified in a mount option. | |
436 | * We use smaller I/O sizes when the file system | |
437 | * is being used for NFS service (wsync mount option). | |
438 | */ | |
439 | STATIC void | |
440 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
441 | { | |
442 | xfs_sb_t *sbp = &(mp->m_sb); | |
443 | int readio_log, writeio_log; | |
1da177e4 | 444 | |
1da177e4 LT |
445 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
446 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
447 | readio_log = XFS_WSYNC_READIO_LOG; | |
448 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
449 | } else { | |
450 | readio_log = XFS_READIO_LOG_LARGE; | |
451 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
452 | } | |
453 | } else { | |
454 | readio_log = mp->m_readio_log; | |
455 | writeio_log = mp->m_writeio_log; | |
456 | } | |
457 | ||
1da177e4 LT |
458 | if (sbp->sb_blocklog > readio_log) { |
459 | mp->m_readio_log = sbp->sb_blocklog; | |
460 | } else { | |
461 | mp->m_readio_log = readio_log; | |
462 | } | |
463 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
464 | if (sbp->sb_blocklog > writeio_log) { | |
465 | mp->m_writeio_log = sbp->sb_blocklog; | |
466 | } else { | |
467 | mp->m_writeio_log = writeio_log; | |
468 | } | |
469 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 470 | } |
1da177e4 | 471 | |
055388a3 DC |
472 | /* |
473 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
474 | */ | |
475 | void | |
476 | xfs_set_low_space_thresholds( | |
477 | struct xfs_mount *mp) | |
478 | { | |
479 | int i; | |
480 | ||
481 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
482 | __uint64_t space = mp->m_sb.sb_dblocks; | |
483 | ||
484 | do_div(space, 100); | |
485 | mp->m_low_space[i] = space * (i + 1); | |
486 | } | |
487 | } | |
488 | ||
489 | ||
0771fb45 ES |
490 | /* |
491 | * Set whether we're using inode alignment. | |
492 | */ | |
493 | STATIC void | |
494 | xfs_set_inoalignment(xfs_mount_t *mp) | |
495 | { | |
62118709 | 496 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
497 | mp->m_sb.sb_inoalignmt >= |
498 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
499 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
500 | else | |
501 | mp->m_inoalign_mask = 0; | |
502 | /* | |
503 | * If we are using stripe alignment, check whether | |
504 | * the stripe unit is a multiple of the inode alignment | |
505 | */ | |
506 | if (mp->m_dalign && mp->m_inoalign_mask && | |
507 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
508 | mp->m_sinoalign = mp->m_dalign; | |
509 | else | |
510 | mp->m_sinoalign = 0; | |
0771fb45 ES |
511 | } |
512 | ||
513 | /* | |
0471f62e | 514 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
515 | */ |
516 | STATIC int | |
ba372674 DC |
517 | xfs_check_sizes( |
518 | struct xfs_mount *mp) | |
0771fb45 | 519 | { |
ba372674 | 520 | struct xfs_buf *bp; |
0771fb45 | 521 | xfs_daddr_t d; |
ba372674 | 522 | int error; |
0771fb45 | 523 | |
1da177e4 LT |
524 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
525 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 526 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 527 | return -EFBIG; |
1da177e4 | 528 | } |
ba372674 | 529 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 530 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
531 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
532 | if (error) { | |
0b932ccc | 533 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 534 | return error; |
1da177e4 | 535 | } |
1922c949 | 536 | xfs_buf_relse(bp); |
1da177e4 | 537 | |
ba372674 DC |
538 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
539 | return 0; | |
540 | ||
541 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
542 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
543 | xfs_warn(mp, "log size mismatch detected"); | |
544 | return -EFBIG; | |
545 | } | |
546 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 547 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
548 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
549 | if (error) { | |
550 | xfs_warn(mp, "log device read failed"); | |
551 | return error; | |
0771fb45 | 552 | } |
ba372674 | 553 | xfs_buf_relse(bp); |
0771fb45 ES |
554 | return 0; |
555 | } | |
556 | ||
7d095257 CH |
557 | /* |
558 | * Clear the quotaflags in memory and in the superblock. | |
559 | */ | |
560 | int | |
561 | xfs_mount_reset_sbqflags( | |
562 | struct xfs_mount *mp) | |
563 | { | |
7d095257 CH |
564 | mp->m_qflags = 0; |
565 | ||
61e63ecb | 566 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
567 | if (mp->m_sb.sb_qflags == 0) |
568 | return 0; | |
569 | spin_lock(&mp->m_sb_lock); | |
570 | mp->m_sb.sb_qflags = 0; | |
571 | spin_unlock(&mp->m_sb_lock); | |
572 | ||
61e63ecb | 573 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
574 | return 0; |
575 | ||
61e63ecb | 576 | return xfs_sync_sb(mp, false); |
7d095257 CH |
577 | } |
578 | ||
d5db0f97 ES |
579 | __uint64_t |
580 | xfs_default_resblks(xfs_mount_t *mp) | |
581 | { | |
582 | __uint64_t resblks; | |
583 | ||
584 | /* | |
8babd8a2 DC |
585 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
586 | * smaller. This is intended to cover concurrent allocation | |
587 | * transactions when we initially hit enospc. These each require a 4 | |
588 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
589 | * allocation reservations. | |
d5db0f97 ES |
590 | */ |
591 | resblks = mp->m_sb.sb_dblocks; | |
592 | do_div(resblks, 20); | |
8babd8a2 | 593 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
594 | return resblks; |
595 | } | |
596 | ||
0771fb45 | 597 | /* |
0771fb45 ES |
598 | * This function does the following on an initial mount of a file system: |
599 | * - reads the superblock from disk and init the mount struct | |
600 | * - if we're a 32-bit kernel, do a size check on the superblock | |
601 | * so we don't mount terabyte filesystems | |
602 | * - init mount struct realtime fields | |
603 | * - allocate inode hash table for fs | |
604 | * - init directory manager | |
605 | * - perform recovery and init the log manager | |
606 | */ | |
607 | int | |
608 | xfs_mountfs( | |
f0b2efad | 609 | struct xfs_mount *mp) |
0771fb45 | 610 | { |
f0b2efad BF |
611 | struct xfs_sb *sbp = &(mp->m_sb); |
612 | struct xfs_inode *rip; | |
613 | __uint64_t resblks; | |
614 | uint quotamount = 0; | |
615 | uint quotaflags = 0; | |
616 | int error = 0; | |
0771fb45 | 617 | |
ff55068c | 618 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 619 | |
ee1c0908 | 620 | /* |
074e427b DC |
621 | * Check for a mismatched features2 values. Older kernels read & wrote |
622 | * into the wrong sb offset for sb_features2 on some platforms due to | |
623 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
624 | * which made older superblock reading/writing routines swap it as a | |
625 | * 64-bit value. | |
ee1c0908 | 626 | * |
e6957ea4 ES |
627 | * For backwards compatibility, we make both slots equal. |
628 | * | |
074e427b DC |
629 | * If we detect a mismatched field, we OR the set bits into the existing |
630 | * features2 field in case it has already been modified; we don't want | |
631 | * to lose any features. We then update the bad location with the ORed | |
632 | * value so that older kernels will see any features2 flags. The | |
633 | * superblock writeback code ensures the new sb_features2 is copied to | |
634 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 635 | */ |
e6957ea4 | 636 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 637 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 638 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 639 | mp->m_update_sb = true; |
e6957ea4 ES |
640 | |
641 | /* | |
642 | * Re-check for ATTR2 in case it was found in bad_features2 | |
643 | * slot. | |
644 | */ | |
7c12f296 TS |
645 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
646 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 647 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
648 | } |
649 | ||
650 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
651 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
652 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 653 | mp->m_update_sb = true; |
e6957ea4 | 654 | |
7c12f296 TS |
655 | /* update sb_versionnum for the clearing of the morebits */ |
656 | if (!sbp->sb_features2) | |
61e63ecb | 657 | mp->m_update_sb = true; |
ee1c0908 DC |
658 | } |
659 | ||
263997a6 DC |
660 | /* always use v2 inodes by default now */ |
661 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
662 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 663 | mp->m_update_sb = true; |
263997a6 DC |
664 | } |
665 | ||
0771fb45 ES |
666 | /* |
667 | * Check if sb_agblocks is aligned at stripe boundary | |
668 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
669 | * allocator alignment is within an ag, therefore ag has | |
670 | * to be aligned at stripe boundary. | |
671 | */ | |
7884bc86 | 672 | error = xfs_update_alignment(mp); |
0771fb45 | 673 | if (error) |
f9057e3d | 674 | goto out; |
0771fb45 ES |
675 | |
676 | xfs_alloc_compute_maxlevels(mp); | |
677 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
678 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
679 | xfs_ialloc_compute_maxlevels(mp); | |
680 | ||
681 | xfs_set_maxicount(mp); | |
682 | ||
e6b3bb78 CM |
683 | /* enable fail_at_unmount as default */ |
684 | mp->m_fail_unmount = 1; | |
685 | ||
a31b1d3d | 686 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
687 | if (error) |
688 | goto out; | |
1da177e4 | 689 | |
225e4635 BD |
690 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
691 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
692 | if (error) |
693 | goto out_remove_sysfs; | |
694 | ||
192852be | 695 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
696 | if (error) |
697 | goto out_del_stats; | |
698 | ||
192852be CM |
699 | |
700 | error = xfs_uuid_mount(mp); | |
701 | if (error) | |
702 | goto out_remove_error_sysfs; | |
703 | ||
0771fb45 ES |
704 | /* |
705 | * Set the minimum read and write sizes | |
706 | */ | |
707 | xfs_set_rw_sizes(mp); | |
708 | ||
055388a3 DC |
709 | /* set the low space thresholds for dynamic preallocation */ |
710 | xfs_set_low_space_thresholds(mp); | |
711 | ||
0771fb45 ES |
712 | /* |
713 | * Set the inode cluster size. | |
714 | * This may still be overridden by the file system | |
715 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
716 | * |
717 | * For v5 filesystems, scale the cluster size with the inode size to | |
718 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
719 | * has set the inode alignment value appropriately for larger cluster | |
720 | * sizes. | |
0771fb45 ES |
721 | */ |
722 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
723 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
724 | int new_size = mp->m_inode_cluster_size; | |
725 | ||
726 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
727 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
728 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 729 | } |
0771fb45 | 730 | |
e5376fc1 BF |
731 | /* |
732 | * If enabled, sparse inode chunk alignment is expected to match the | |
733 | * cluster size. Full inode chunk alignment must match the chunk size, | |
734 | * but that is checked on sb read verification... | |
735 | */ | |
736 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
737 | mp->m_sb.sb_spino_align != | |
738 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { | |
739 | xfs_warn(mp, | |
740 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
741 | mp->m_sb.sb_spino_align, | |
742 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); | |
743 | error = -EINVAL; | |
744 | goto out_remove_uuid; | |
745 | } | |
746 | ||
0771fb45 ES |
747 | /* |
748 | * Set inode alignment fields | |
749 | */ | |
750 | xfs_set_inoalignment(mp); | |
751 | ||
752 | /* | |
c2bfbc9b | 753 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 754 | */ |
4249023a | 755 | error = xfs_check_sizes(mp); |
0771fb45 | 756 | if (error) |
f9057e3d | 757 | goto out_remove_uuid; |
0771fb45 | 758 | |
1da177e4 LT |
759 | /* |
760 | * Initialize realtime fields in the mount structure | |
761 | */ | |
0771fb45 ES |
762 | error = xfs_rtmount_init(mp); |
763 | if (error) { | |
0b932ccc | 764 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 765 | goto out_remove_uuid; |
1da177e4 LT |
766 | } |
767 | ||
1da177e4 LT |
768 | /* |
769 | * Copies the low order bits of the timestamp and the randomly | |
770 | * set "sequence" number out of a UUID. | |
771 | */ | |
772 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
773 | ||
1da177e4 LT |
774 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
775 | ||
0650b554 DC |
776 | error = xfs_da_mount(mp); |
777 | if (error) { | |
778 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
779 | goto out_remove_uuid; | |
780 | } | |
1da177e4 LT |
781 | |
782 | /* | |
783 | * Initialize the precomputed transaction reservations values. | |
784 | */ | |
785 | xfs_trans_init(mp); | |
786 | ||
1da177e4 LT |
787 | /* |
788 | * Allocate and initialize the per-ag data. | |
789 | */ | |
1c1c6ebc | 790 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 791 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
792 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
793 | if (error) { | |
0b932ccc | 794 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 795 | goto out_free_dir; |
1c1c6ebc | 796 | } |
1da177e4 | 797 | |
f9057e3d | 798 | if (!sbp->sb_logblocks) { |
0b932ccc | 799 | xfs_warn(mp, "no log defined"); |
f9057e3d | 800 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 801 | error = -EFSCORRUPTED; |
f9057e3d CH |
802 | goto out_free_perag; |
803 | } | |
804 | ||
1da177e4 | 805 | /* |
f0b2efad BF |
806 | * Log's mount-time initialization. The first part of recovery can place |
807 | * some items on the AIL, to be handled when recovery is finished or | |
808 | * cancelled. | |
1da177e4 | 809 | */ |
f9057e3d CH |
810 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
811 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
812 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
813 | if (error) { | |
0b932ccc | 814 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 815 | goto out_fail_wait; |
1da177e4 LT |
816 | } |
817 | ||
92821e2b DC |
818 | /* |
819 | * Now the log is mounted, we know if it was an unclean shutdown or | |
820 | * not. If it was, with the first phase of recovery has completed, we | |
821 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
822 | * but they are recovered transactionally in the second recovery phase | |
823 | * later. | |
824 | * | |
825 | * Hence we can safely re-initialise incore superblock counters from | |
826 | * the per-ag data. These may not be correct if the filesystem was not | |
827 | * cleanly unmounted, so we need to wait for recovery to finish before | |
828 | * doing this. | |
829 | * | |
830 | * If the filesystem was cleanly unmounted, then we can trust the | |
831 | * values in the superblock to be correct and we don't need to do | |
832 | * anything here. | |
833 | * | |
834 | * If we are currently making the filesystem, the initialisation will | |
835 | * fail as the perag data is in an undefined state. | |
836 | */ | |
92821e2b DC |
837 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
838 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
839 | !mp->m_sb.sb_inprogress) { | |
840 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 841 | if (error) |
6eee8972 | 842 | goto out_log_dealloc; |
92821e2b | 843 | } |
f9057e3d | 844 | |
1da177e4 LT |
845 | /* |
846 | * Get and sanity-check the root inode. | |
847 | * Save the pointer to it in the mount structure. | |
848 | */ | |
7b6259e7 | 849 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 850 | if (error) { |
0b932ccc | 851 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 852 | goto out_log_dealloc; |
1da177e4 LT |
853 | } |
854 | ||
855 | ASSERT(rip != NULL); | |
1da177e4 | 856 | |
c19b3b05 | 857 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 858 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 859 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
860 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
861 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
862 | mp); | |
2451337d | 863 | error = -EFSCORRUPTED; |
f9057e3d | 864 | goto out_rele_rip; |
1da177e4 LT |
865 | } |
866 | mp->m_rootip = rip; /* save it */ | |
867 | ||
868 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
869 | ||
870 | /* | |
871 | * Initialize realtime inode pointers in the mount structure | |
872 | */ | |
0771fb45 ES |
873 | error = xfs_rtmount_inodes(mp); |
874 | if (error) { | |
1da177e4 LT |
875 | /* |
876 | * Free up the root inode. | |
877 | */ | |
0b932ccc | 878 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 879 | goto out_rele_rip; |
1da177e4 LT |
880 | } |
881 | ||
882 | /* | |
7884bc86 CH |
883 | * If this is a read-only mount defer the superblock updates until |
884 | * the next remount into writeable mode. Otherwise we would never | |
885 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 886 | */ |
61e63ecb DC |
887 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
888 | error = xfs_sync_sb(mp, false); | |
e5720eec | 889 | if (error) { |
0b932ccc | 890 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 891 | goto out_rtunmount; |
e5720eec DC |
892 | } |
893 | } | |
1da177e4 LT |
894 | |
895 | /* | |
896 | * Initialise the XFS quota management subsystem for this mount | |
897 | */ | |
7d095257 CH |
898 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
899 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
900 | if (error) | |
901 | goto out_rtunmount; | |
902 | } else { | |
903 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
904 | ||
905 | /* | |
906 | * If a file system had quotas running earlier, but decided to | |
907 | * mount without -o uquota/pquota/gquota options, revoke the | |
908 | * quotachecked license. | |
909 | */ | |
910 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 911 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
912 | error = xfs_mount_reset_sbqflags(mp); |
913 | if (error) | |
a70a4fa5 | 914 | goto out_rtunmount; |
7d095257 CH |
915 | } |
916 | } | |
1da177e4 LT |
917 | |
918 | /* | |
f0b2efad BF |
919 | * Finish recovering the file system. This part needed to be delayed |
920 | * until after the root and real-time bitmap inodes were consistently | |
921 | * read in. | |
1da177e4 | 922 | */ |
4249023a | 923 | error = xfs_log_mount_finish(mp); |
1da177e4 | 924 | if (error) { |
0b932ccc | 925 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 926 | goto out_rtunmount; |
1da177e4 LT |
927 | } |
928 | ||
929 | /* | |
930 | * Complete the quota initialisation, post-log-replay component. | |
931 | */ | |
7d095257 CH |
932 | if (quotamount) { |
933 | ASSERT(mp->m_qflags == 0); | |
934 | mp->m_qflags = quotaflags; | |
935 | ||
936 | xfs_qm_mount_quotas(mp); | |
937 | } | |
938 | ||
84e1e99f DC |
939 | /* |
940 | * Now we are mounted, reserve a small amount of unused space for | |
941 | * privileged transactions. This is needed so that transaction | |
942 | * space required for critical operations can dip into this pool | |
943 | * when at ENOSPC. This is needed for operations like create with | |
944 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
945 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
946 | * |
947 | * This may drive us straight to ENOSPC on mount, but that implies | |
948 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 949 | */ |
d5db0f97 ES |
950 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
951 | resblks = xfs_default_resblks(mp); | |
952 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
953 | if (error) | |
0b932ccc DC |
954 | xfs_warn(mp, |
955 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 956 | } |
84e1e99f | 957 | |
1da177e4 LT |
958 | return 0; |
959 | ||
b93b6e43 CH |
960 | out_rtunmount: |
961 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 962 | out_rele_rip: |
43355099 | 963 | IRELE(rip); |
0ae120f8 BF |
964 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
965 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
f9057e3d | 966 | out_log_dealloc: |
e6b3bb78 | 967 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 968 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
969 | out_fail_wait: |
970 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
971 | xfs_wait_buftarg(mp->m_logdev_targp); | |
972 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 973 | out_free_perag: |
ff4f038c | 974 | xfs_free_perag(mp); |
0650b554 DC |
975 | out_free_dir: |
976 | xfs_da_unmount(mp); | |
f9057e3d | 977 | out_remove_uuid: |
27174203 | 978 | xfs_uuid_unmount(mp); |
192852be CM |
979 | out_remove_error_sysfs: |
980 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
981 | out_del_stats: |
982 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
983 | out_remove_sysfs: |
984 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 985 | out: |
1da177e4 LT |
986 | return error; |
987 | } | |
988 | ||
989 | /* | |
1da177e4 LT |
990 | * This flushes out the inodes,dquots and the superblock, unmounts the |
991 | * log and makes sure that incore structures are freed. | |
992 | */ | |
41b5c2e7 CH |
993 | void |
994 | xfs_unmountfs( | |
995 | struct xfs_mount *mp) | |
1da177e4 | 996 | { |
41b5c2e7 CH |
997 | __uint64_t resblks; |
998 | int error; | |
1da177e4 | 999 | |
579b62fa BF |
1000 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1001 | ||
7d095257 | 1002 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1003 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1004 | IRELE(mp->m_rootip); |
1005 | ||
641c56fb DC |
1006 | /* |
1007 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1008 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1009 | * the transaction is still sitting in a iclog. The stale inodes |
1010 | * on that buffer will have their flush locks held until the | |
1011 | * transaction hits the disk and the callbacks run. the inode | |
1012 | * flush takes the flush lock unconditionally and with nothing to | |
1013 | * push out the iclog we will never get that unlocked. hence we | |
1014 | * need to force the log first. | |
1015 | */ | |
a14a348b | 1016 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1017 | |
e6b3bb78 CM |
1018 | /* |
1019 | * We now need to tell the world we are unmounting. This will allow | |
1020 | * us to detect that the filesystem is going away and we should error | |
1021 | * out anything that we have been retrying in the background. This will | |
1022 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1023 | */ | |
1024 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1025 | ||
c854363e | 1026 | /* |
211e4d43 CH |
1027 | * Flush all pending changes from the AIL. |
1028 | */ | |
1029 | xfs_ail_push_all_sync(mp->m_ail); | |
1030 | ||
1031 | /* | |
1032 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1033 | * inodes and none should be pinned or locked, but use synchronous |
1034 | * reclaim just to be sure. We can stop background inode reclaim | |
1035 | * here as well if it is still running. | |
c854363e | 1036 | */ |
7e18530b | 1037 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1038 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1039 | |
7d095257 | 1040 | xfs_qm_unmount(mp); |
a357a121 | 1041 | |
84e1e99f DC |
1042 | /* |
1043 | * Unreserve any blocks we have so that when we unmount we don't account | |
1044 | * the reserved free space as used. This is really only necessary for | |
1045 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1046 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1047 | * |
1048 | * We don't bother correcting this elsewhere for lazy superblock | |
1049 | * counting because on mount of an unclean filesystem we reconstruct the | |
1050 | * correct counter value and this is irrelevant. | |
1051 | * | |
1052 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1053 | * we only every apply deltas to the superblock and hence the incore | |
1054 | * value does not matter.... | |
1055 | */ | |
1056 | resblks = 0; | |
714082bc DC |
1057 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1058 | if (error) | |
0b932ccc | 1059 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1060 | "Freespace may not be correct on next mount."); |
1061 | ||
adab0f67 | 1062 | error = xfs_log_sbcount(mp); |
e5720eec | 1063 | if (error) |
0b932ccc | 1064 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1065 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1066 | |
225e4635 | 1067 | |
21b699c8 | 1068 | xfs_log_unmount(mp); |
0650b554 | 1069 | xfs_da_unmount(mp); |
27174203 | 1070 | xfs_uuid_unmount(mp); |
1da177e4 | 1071 | |
1550d0b0 | 1072 | #if defined(DEBUG) |
0ce4cfd4 | 1073 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1074 | #endif |
ff4f038c | 1075 | xfs_free_perag(mp); |
a31b1d3d | 1076 | |
192852be | 1077 | xfs_error_sysfs_del(mp); |
225e4635 | 1078 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1079 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1080 | } |
1081 | ||
91ee575f BF |
1082 | /* |
1083 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1084 | * freeze level for which modifications should not be allowed. This allows | |
1085 | * certain operations to proceed while the freeze sequence is in progress, if | |
1086 | * necessary. | |
1087 | */ | |
1088 | bool | |
1089 | xfs_fs_writable( | |
1090 | struct xfs_mount *mp, | |
1091 | int level) | |
92821e2b | 1092 | { |
91ee575f BF |
1093 | ASSERT(level > SB_UNFROZEN); |
1094 | if ((mp->m_super->s_writers.frozen >= level) || | |
1095 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1096 | return false; | |
1097 | ||
1098 | return true; | |
92821e2b DC |
1099 | } |
1100 | ||
1101 | /* | |
b2ce3974 AE |
1102 | * xfs_log_sbcount |
1103 | * | |
adab0f67 | 1104 | * Sync the superblock counters to disk. |
b2ce3974 | 1105 | * |
91ee575f BF |
1106 | * Note this code can be called during the process of freezing, so we use the |
1107 | * transaction allocator that does not block when the transaction subsystem is | |
1108 | * in its frozen state. | |
92821e2b DC |
1109 | */ |
1110 | int | |
adab0f67 | 1111 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1112 | { |
91ee575f BF |
1113 | /* allow this to proceed during the freeze sequence... */ |
1114 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1115 | return 0; |
1116 | ||
92821e2b DC |
1117 | /* |
1118 | * we don't need to do this if we are updating the superblock | |
1119 | * counters on every modification. | |
1120 | */ | |
1121 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1122 | return 0; | |
1123 | ||
61e63ecb | 1124 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1125 | } |
1126 | ||
8c1903d3 DC |
1127 | /* |
1128 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1129 | * of 128 so we don't need to take the counter lock on every update. | |
1130 | */ | |
1131 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1132 | int |
1133 | xfs_mod_icount( | |
1134 | struct xfs_mount *mp, | |
1135 | int64_t delta) | |
1136 | { | |
8c1903d3 DC |
1137 | __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
1138 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { | |
501ab323 DC |
1139 | ASSERT(0); |
1140 | percpu_counter_add(&mp->m_icount, -delta); | |
1141 | return -EINVAL; | |
1142 | } | |
1143 | return 0; | |
1144 | } | |
1145 | ||
e88b64ea DC |
1146 | int |
1147 | xfs_mod_ifree( | |
1148 | struct xfs_mount *mp, | |
1149 | int64_t delta) | |
1150 | { | |
1151 | percpu_counter_add(&mp->m_ifree, delta); | |
1152 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1153 | ASSERT(0); | |
1154 | percpu_counter_add(&mp->m_ifree, -delta); | |
1155 | return -EINVAL; | |
1156 | } | |
1157 | return 0; | |
1158 | } | |
0d485ada | 1159 | |
8c1903d3 DC |
1160 | /* |
1161 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1162 | * only occurs on frequent small block count updates such as in the delayed | |
1163 | * allocation path for buffered writes (page a time updates). Hence we set | |
1164 | * a large batch count (1024) to minimise global counter updates except when | |
1165 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1166 | */ | |
1167 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1168 | int |
1169 | xfs_mod_fdblocks( | |
1170 | struct xfs_mount *mp, | |
1171 | int64_t delta, | |
1172 | bool rsvd) | |
1173 | { | |
1174 | int64_t lcounter; | |
1175 | long long res_used; | |
1176 | s32 batch; | |
1177 | ||
1178 | if (delta > 0) { | |
1179 | /* | |
1180 | * If the reserve pool is depleted, put blocks back into it | |
1181 | * first. Most of the time the pool is full. | |
1182 | */ | |
1183 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1184 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1185 | return 0; | |
1186 | } | |
1187 | ||
1188 | spin_lock(&mp->m_sb_lock); | |
1189 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1190 | ||
1191 | if (res_used > delta) { | |
1192 | mp->m_resblks_avail += delta; | |
1193 | } else { | |
1194 | delta -= res_used; | |
1195 | mp->m_resblks_avail = mp->m_resblks; | |
1196 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1197 | } | |
1198 | spin_unlock(&mp->m_sb_lock); | |
1199 | return 0; | |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * Taking blocks away, need to be more accurate the closer we | |
1204 | * are to zero. | |
1205 | * | |
0d485ada DC |
1206 | * If the counter has a value of less than 2 * max batch size, |
1207 | * then make everything serialise as we are real close to | |
1208 | * ENOSPC. | |
1209 | */ | |
8c1903d3 DC |
1210 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1211 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1212 | batch = 1; |
1213 | else | |
8c1903d3 | 1214 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada DC |
1215 | |
1216 | __percpu_counter_add(&mp->m_fdblocks, delta, batch); | |
8c1903d3 DC |
1217 | if (__percpu_counter_compare(&mp->m_fdblocks, XFS_ALLOC_SET_ASIDE(mp), |
1218 | XFS_FDBLOCKS_BATCH) >= 0) { | |
0d485ada DC |
1219 | /* we had space! */ |
1220 | return 0; | |
1221 | } | |
1222 | ||
1223 | /* | |
1224 | * lock up the sb for dipping into reserves before releasing the space | |
1225 | * that took us to ENOSPC. | |
1226 | */ | |
1227 | spin_lock(&mp->m_sb_lock); | |
1228 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1229 | if (!rsvd) | |
1230 | goto fdblocks_enospc; | |
1231 | ||
1232 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1233 | if (lcounter >= 0) { | |
1234 | mp->m_resblks_avail = lcounter; | |
1235 | spin_unlock(&mp->m_sb_lock); | |
1236 | return 0; | |
1237 | } | |
1238 | printk_once(KERN_WARNING | |
1239 | "Filesystem \"%s\": reserve blocks depleted! " | |
1240 | "Consider increasing reserve pool size.", | |
1241 | mp->m_fsname); | |
1242 | fdblocks_enospc: | |
1243 | spin_unlock(&mp->m_sb_lock); | |
1244 | return -ENOSPC; | |
1245 | } | |
1246 | ||
bab98bbe DC |
1247 | int |
1248 | xfs_mod_frextents( | |
1249 | struct xfs_mount *mp, | |
1250 | int64_t delta) | |
1251 | { | |
1252 | int64_t lcounter; | |
1253 | int ret = 0; | |
1254 | ||
1255 | spin_lock(&mp->m_sb_lock); | |
1256 | lcounter = mp->m_sb.sb_frextents + delta; | |
1257 | if (lcounter < 0) | |
1258 | ret = -ENOSPC; | |
1259 | else | |
1260 | mp->m_sb.sb_frextents = lcounter; | |
1261 | spin_unlock(&mp->m_sb_lock); | |
1262 | return ret; | |
1263 | } | |
1264 | ||
1da177e4 LT |
1265 | /* |
1266 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1267 | * The buffer is returned locked and read in from disk. | |
1268 | * The buffer should be released with a call to xfs_brelse(). | |
1269 | * | |
1270 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1271 | * the superblock buffer if it can be locked without sleeping. | |
1272 | * If it can't then we'll return NULL. | |
1273 | */ | |
0c842ad4 | 1274 | struct xfs_buf * |
1da177e4 | 1275 | xfs_getsb( |
0c842ad4 CH |
1276 | struct xfs_mount *mp, |
1277 | int flags) | |
1da177e4 | 1278 | { |
0c842ad4 | 1279 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1280 | |
0c842ad4 CH |
1281 | if (!xfs_buf_trylock(bp)) { |
1282 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1283 | return NULL; |
0c842ad4 | 1284 | xfs_buf_lock(bp); |
1da177e4 | 1285 | } |
0c842ad4 | 1286 | |
72790aa1 | 1287 | xfs_buf_hold(bp); |
b0388bf1 | 1288 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1289 | return bp; |
1da177e4 LT |
1290 | } |
1291 | ||
1292 | /* | |
1293 | * Used to free the superblock along various error paths. | |
1294 | */ | |
1295 | void | |
1296 | xfs_freesb( | |
26af6552 | 1297 | struct xfs_mount *mp) |
1da177e4 | 1298 | { |
26af6552 | 1299 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1300 | |
26af6552 | 1301 | xfs_buf_lock(bp); |
1da177e4 | 1302 | mp->m_sb_bp = NULL; |
26af6552 | 1303 | xfs_buf_relse(bp); |
1da177e4 LT |
1304 | } |
1305 | ||
dda35b8f CH |
1306 | /* |
1307 | * If the underlying (data/log/rt) device is readonly, there are some | |
1308 | * operations that cannot proceed. | |
1309 | */ | |
1310 | int | |
1311 | xfs_dev_is_read_only( | |
1312 | struct xfs_mount *mp, | |
1313 | char *message) | |
1314 | { | |
1315 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1316 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1317 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1318 | xfs_notice(mp, "%s required on read-only device.", message); |
1319 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1320 | return -EROFS; |
dda35b8f CH |
1321 | } |
1322 | return 0; | |
1323 | } |