]>
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" |
3ab78df2 | 27 | #include "xfs_defer.h" |
57062787 | 28 | #include "xfs_da_format.h" |
9a2cc41c | 29 | #include "xfs_da_btree.h" |
1da177e4 | 30 | #include "xfs_inode.h" |
a4fbe6ab | 31 | #include "xfs_dir2.h" |
a844f451 | 32 | #include "xfs_ialloc.h" |
1da177e4 LT |
33 | #include "xfs_alloc.h" |
34 | #include "xfs_rtalloc.h" | |
35 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
36 | #include "xfs_trans.h" |
37 | #include "xfs_trans_priv.h" | |
38 | #include "xfs_log.h" | |
1da177e4 | 39 | #include "xfs_error.h" |
1da177e4 LT |
40 | #include "xfs_quota.h" |
41 | #include "xfs_fsops.h" | |
0b1b213f | 42 | #include "xfs_trace.h" |
6d8b79cf | 43 | #include "xfs_icache.h" |
a31b1d3d | 44 | #include "xfs_sysfs.h" |
035e00ac | 45 | #include "xfs_rmap_btree.h" |
1946b91c | 46 | #include "xfs_refcount_btree.h" |
174edb0e | 47 | #include "xfs_reflink.h" |
0b1b213f | 48 | |
1da177e4 | 49 | |
27174203 CH |
50 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
51 | static int xfs_uuid_table_size; | |
52 | static uuid_t *xfs_uuid_table; | |
53 | ||
af3b6382 DW |
54 | void |
55 | xfs_uuid_table_free(void) | |
56 | { | |
57 | if (xfs_uuid_table_size == 0) | |
58 | return; | |
59 | kmem_free(xfs_uuid_table); | |
60 | xfs_uuid_table = NULL; | |
61 | xfs_uuid_table_size = 0; | |
62 | } | |
63 | ||
27174203 CH |
64 | /* |
65 | * See if the UUID is unique among mounted XFS filesystems. | |
66 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
67 | */ | |
68 | STATIC int | |
69 | xfs_uuid_mount( | |
70 | struct xfs_mount *mp) | |
71 | { | |
72 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
73 | int hole, i; | |
74 | ||
75 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
76 | return 0; | |
77 | ||
78 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 79 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 80 | return -EINVAL; |
27174203 CH |
81 | } |
82 | ||
83 | mutex_lock(&xfs_uuid_table_mutex); | |
84 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
85 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
86 | hole = i; | |
87 | continue; | |
88 | } | |
89 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
90 | goto out_duplicate; | |
91 | } | |
92 | ||
93 | if (hole < 0) { | |
94 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
95 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
27174203 CH |
96 | KM_SLEEP); |
97 | hole = xfs_uuid_table_size++; | |
98 | } | |
99 | xfs_uuid_table[hole] = *uuid; | |
100 | mutex_unlock(&xfs_uuid_table_mutex); | |
101 | ||
102 | return 0; | |
103 | ||
104 | out_duplicate: | |
105 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 106 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 107 | return -EINVAL; |
27174203 CH |
108 | } |
109 | ||
110 | STATIC void | |
111 | xfs_uuid_unmount( | |
112 | struct xfs_mount *mp) | |
113 | { | |
114 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
115 | int i; | |
116 | ||
117 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
118 | return; | |
119 | ||
120 | mutex_lock(&xfs_uuid_table_mutex); | |
121 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
122 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
123 | continue; | |
124 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
125 | continue; | |
126 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
127 | break; | |
128 | } | |
129 | ASSERT(i < xfs_uuid_table_size); | |
130 | mutex_unlock(&xfs_uuid_table_mutex); | |
131 | } | |
132 | ||
133 | ||
e176579e DC |
134 | STATIC void |
135 | __xfs_free_perag( | |
136 | struct rcu_head *head) | |
137 | { | |
138 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
139 | ||
140 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
141 | kmem_free(pag); | |
142 | } | |
143 | ||
1da177e4 | 144 | /* |
e176579e | 145 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 146 | */ |
c962fb79 | 147 | STATIC void |
ff4f038c | 148 | xfs_free_perag( |
745f6919 | 149 | xfs_mount_t *mp) |
1da177e4 | 150 | { |
1c1c6ebc DC |
151 | xfs_agnumber_t agno; |
152 | struct xfs_perag *pag; | |
153 | ||
154 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
155 | spin_lock(&mp->m_perag_lock); | |
156 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
157 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 158 | ASSERT(pag); |
f83282a8 | 159 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
6031e73a | 160 | xfs_buf_hash_destroy(pag); |
e176579e | 161 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 162 | } |
1da177e4 LT |
163 | } |
164 | ||
4cc929ee NS |
165 | /* |
166 | * Check size of device based on the (data/realtime) block count. | |
167 | * Note: this check is used by the growfs code as well as mount. | |
168 | */ | |
169 | int | |
170 | xfs_sb_validate_fsb_count( | |
171 | xfs_sb_t *sbp, | |
172 | __uint64_t nblocks) | |
173 | { | |
174 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
175 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
176 | ||
d5cf09ba | 177 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 178 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 179 | return -EFBIG; |
4cc929ee NS |
180 | return 0; |
181 | } | |
1da177e4 | 182 | |
1c1c6ebc | 183 | int |
c11e2c36 | 184 | xfs_initialize_perag( |
c11e2c36 | 185 | xfs_mount_t *mp, |
1c1c6ebc DC |
186 | xfs_agnumber_t agcount, |
187 | xfs_agnumber_t *maxagi) | |
1da177e4 | 188 | { |
2d2194f6 | 189 | xfs_agnumber_t index; |
8b26c582 | 190 | xfs_agnumber_t first_initialised = 0; |
1da177e4 | 191 | xfs_perag_t *pag; |
8b26c582 | 192 | int error = -ENOMEM; |
1da177e4 | 193 | |
1c1c6ebc DC |
194 | /* |
195 | * Walk the current per-ag tree so we don't try to initialise AGs | |
196 | * that already exist (growfs case). Allocate and insert all the | |
197 | * AGs we don't find ready for initialisation. | |
198 | */ | |
199 | for (index = 0; index < agcount; index++) { | |
200 | pag = xfs_perag_get(mp, index); | |
201 | if (pag) { | |
202 | xfs_perag_put(pag); | |
203 | continue; | |
204 | } | |
8b26c582 DC |
205 | if (!first_initialised) |
206 | first_initialised = index; | |
fb3b504a | 207 | |
1c1c6ebc DC |
208 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
209 | if (!pag) | |
8b26c582 | 210 | goto out_unwind; |
fb3b504a CH |
211 | pag->pag_agno = index; |
212 | pag->pag_mount = mp; | |
1a427ab0 | 213 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 214 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 215 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
6031e73a LS |
216 | if (xfs_buf_hash_init(pag)) |
217 | goto out_unwind; | |
fb3b504a | 218 | |
1c1c6ebc | 219 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 220 | goto out_unwind; |
fb3b504a | 221 | |
1c1c6ebc DC |
222 | spin_lock(&mp->m_perag_lock); |
223 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
224 | BUG(); | |
225 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
226 | radix_tree_preload_end(); |
227 | error = -EEXIST; | |
228 | goto out_unwind; | |
1c1c6ebc DC |
229 | } |
230 | spin_unlock(&mp->m_perag_lock); | |
231 | radix_tree_preload_end(); | |
232 | } | |
233 | ||
12c3f05c | 234 | index = xfs_set_inode_alloc(mp, agcount); |
fb3b504a | 235 | |
1c1c6ebc DC |
236 | if (maxagi) |
237 | *maxagi = index; | |
8018026e DW |
238 | |
239 | mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp); | |
1c1c6ebc | 240 | return 0; |
8b26c582 DC |
241 | |
242 | out_unwind: | |
6031e73a | 243 | xfs_buf_hash_destroy(pag); |
8b26c582 DC |
244 | kmem_free(pag); |
245 | for (; index > first_initialised; index--) { | |
246 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
6031e73a | 247 | xfs_buf_hash_destroy(pag); |
8b26c582 DC |
248 | kmem_free(pag); |
249 | } | |
250 | return error; | |
1da177e4 LT |
251 | } |
252 | ||
1da177e4 LT |
253 | /* |
254 | * xfs_readsb | |
255 | * | |
256 | * Does the initial read of the superblock. | |
257 | */ | |
258 | int | |
ff55068c DC |
259 | xfs_readsb( |
260 | struct xfs_mount *mp, | |
261 | int flags) | |
1da177e4 LT |
262 | { |
263 | unsigned int sector_size; | |
04a1e6c5 DC |
264 | struct xfs_buf *bp; |
265 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 266 | int error; |
af34e09d | 267 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 268 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
269 | |
270 | ASSERT(mp->m_sb_bp == NULL); | |
271 | ASSERT(mp->m_ddev_targp != NULL); | |
272 | ||
daba5427 ES |
273 | /* |
274 | * For the initial read, we must guess at the sector | |
275 | * size based on the block device. It's enough to | |
276 | * get the sb_sectsize out of the superblock and | |
277 | * then reread with the proper length. | |
278 | * We don't verify it yet, because it may not be complete. | |
279 | */ | |
280 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
281 | buf_ops = NULL; | |
282 | ||
1da177e4 | 283 | /* |
c891c30a BF |
284 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
285 | * around at all times to optimize access to the superblock. Therefore, | |
286 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
287 | * elevated. | |
1da177e4 | 288 | */ |
26af6552 | 289 | reread: |
ba372674 | 290 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
291 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
292 | buf_ops); | |
ba372674 | 293 | if (error) { |
eab4e633 | 294 | if (loud) |
e721f504 | 295 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 296 | /* bad CRC means corrupted metadata */ |
2451337d DC |
297 | if (error == -EFSBADCRC) |
298 | error = -EFSCORRUPTED; | |
ba372674 | 299 | return error; |
eab4e633 | 300 | } |
1da177e4 LT |
301 | |
302 | /* | |
303 | * Initialize the mount structure from the superblock. | |
1da177e4 | 304 | */ |
556b8883 | 305 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
306 | |
307 | /* | |
308 | * If we haven't validated the superblock, do so now before we try | |
309 | * to check the sector size and reread the superblock appropriately. | |
310 | */ | |
311 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
312 | if (loud) | |
313 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 314 | error = -EINVAL; |
556b8883 DC |
315 | goto release_buf; |
316 | } | |
ff55068c | 317 | |
1da177e4 LT |
318 | /* |
319 | * We must be able to do sector-sized and sector-aligned IO. | |
320 | */ | |
04a1e6c5 | 321 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
322 | if (loud) |
323 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 324 | sector_size, sbp->sb_sectsize); |
2451337d | 325 | error = -ENOSYS; |
26af6552 | 326 | goto release_buf; |
1da177e4 LT |
327 | } |
328 | ||
daba5427 | 329 | if (buf_ops == NULL) { |
556b8883 DC |
330 | /* |
331 | * Re-read the superblock so the buffer is correctly sized, | |
332 | * and properly verified. | |
333 | */ | |
1da177e4 | 334 | xfs_buf_relse(bp); |
04a1e6c5 | 335 | sector_size = sbp->sb_sectsize; |
daba5427 | 336 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 337 | goto reread; |
1da177e4 LT |
338 | } |
339 | ||
5681ca40 | 340 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 341 | |
04a1e6c5 DC |
342 | /* no need to be quiet anymore, so reset the buf ops */ |
343 | bp->b_ops = &xfs_sb_buf_ops; | |
344 | ||
1da177e4 | 345 | mp->m_sb_bp = bp; |
26af6552 | 346 | xfs_buf_unlock(bp); |
1da177e4 LT |
347 | return 0; |
348 | ||
26af6552 DC |
349 | release_buf: |
350 | xfs_buf_relse(bp); | |
1da177e4 LT |
351 | return error; |
352 | } | |
353 | ||
1da177e4 | 354 | /* |
0771fb45 | 355 | * Update alignment values based on mount options and sb values |
1da177e4 | 356 | */ |
0771fb45 | 357 | STATIC int |
7884bc86 | 358 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 359 | { |
1da177e4 | 360 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 361 | |
4249023a | 362 | if (mp->m_dalign) { |
1da177e4 LT |
363 | /* |
364 | * If stripe unit and stripe width are not multiples | |
365 | * of the fs blocksize turn off alignment. | |
366 | */ | |
367 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
368 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
369 | xfs_warn(mp, |
370 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
371 | sbp->sb_blocksize); | |
2451337d | 372 | return -EINVAL; |
1da177e4 LT |
373 | } else { |
374 | /* | |
375 | * Convert the stripe unit and width to FSBs. | |
376 | */ | |
377 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
378 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 379 | xfs_warn(mp, |
39a45d84 JL |
380 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
381 | sbp->sb_agblocks); | |
2451337d | 382 | return -EINVAL; |
1da177e4 LT |
383 | } else if (mp->m_dalign) { |
384 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
385 | } else { | |
39a45d84 JL |
386 | xfs_warn(mp, |
387 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
388 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 389 | return -EINVAL; |
1da177e4 LT |
390 | } |
391 | } | |
392 | ||
393 | /* | |
394 | * Update superblock with new values | |
395 | * and log changes | |
396 | */ | |
62118709 | 397 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
398 | if (sbp->sb_unit != mp->m_dalign) { |
399 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 400 | mp->m_update_sb = true; |
1da177e4 LT |
401 | } |
402 | if (sbp->sb_width != mp->m_swidth) { | |
403 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 404 | mp->m_update_sb = true; |
1da177e4 | 405 | } |
34d7f603 JL |
406 | } else { |
407 | xfs_warn(mp, | |
408 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 409 | return -EINVAL; |
1da177e4 LT |
410 | } |
411 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 412 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
413 | mp->m_dalign = sbp->sb_unit; |
414 | mp->m_swidth = sbp->sb_width; | |
415 | } | |
416 | ||
0771fb45 ES |
417 | return 0; |
418 | } | |
1da177e4 | 419 | |
0771fb45 ES |
420 | /* |
421 | * Set the maximum inode count for this filesystem | |
422 | */ | |
423 | STATIC void | |
424 | xfs_set_maxicount(xfs_mount_t *mp) | |
425 | { | |
426 | xfs_sb_t *sbp = &(mp->m_sb); | |
427 | __uint64_t icount; | |
1da177e4 | 428 | |
0771fb45 ES |
429 | if (sbp->sb_imax_pct) { |
430 | /* | |
431 | * Make sure the maximum inode count is a multiple | |
432 | * of the units we allocate inodes in. | |
1da177e4 | 433 | */ |
1da177e4 LT |
434 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
435 | do_div(icount, 100); | |
436 | do_div(icount, mp->m_ialloc_blks); | |
437 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
438 | sbp->sb_inopblog; | |
0771fb45 | 439 | } else { |
1da177e4 | 440 | mp->m_maxicount = 0; |
1da177e4 | 441 | } |
0771fb45 ES |
442 | } |
443 | ||
444 | /* | |
445 | * Set the default minimum read and write sizes unless | |
446 | * already specified in a mount option. | |
447 | * We use smaller I/O sizes when the file system | |
448 | * is being used for NFS service (wsync mount option). | |
449 | */ | |
450 | STATIC void | |
451 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
452 | { | |
453 | xfs_sb_t *sbp = &(mp->m_sb); | |
454 | int readio_log, writeio_log; | |
1da177e4 | 455 | |
1da177e4 LT |
456 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
457 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
458 | readio_log = XFS_WSYNC_READIO_LOG; | |
459 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
460 | } else { | |
461 | readio_log = XFS_READIO_LOG_LARGE; | |
462 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
463 | } | |
464 | } else { | |
465 | readio_log = mp->m_readio_log; | |
466 | writeio_log = mp->m_writeio_log; | |
467 | } | |
468 | ||
1da177e4 LT |
469 | if (sbp->sb_blocklog > readio_log) { |
470 | mp->m_readio_log = sbp->sb_blocklog; | |
471 | } else { | |
472 | mp->m_readio_log = readio_log; | |
473 | } | |
474 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
475 | if (sbp->sb_blocklog > writeio_log) { | |
476 | mp->m_writeio_log = sbp->sb_blocklog; | |
477 | } else { | |
478 | mp->m_writeio_log = writeio_log; | |
479 | } | |
480 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 481 | } |
1da177e4 | 482 | |
055388a3 DC |
483 | /* |
484 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
485 | */ | |
486 | void | |
487 | xfs_set_low_space_thresholds( | |
488 | struct xfs_mount *mp) | |
489 | { | |
490 | int i; | |
491 | ||
492 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
493 | __uint64_t space = mp->m_sb.sb_dblocks; | |
494 | ||
495 | do_div(space, 100); | |
496 | mp->m_low_space[i] = space * (i + 1); | |
497 | } | |
498 | } | |
499 | ||
500 | ||
0771fb45 ES |
501 | /* |
502 | * Set whether we're using inode alignment. | |
503 | */ | |
504 | STATIC void | |
505 | xfs_set_inoalignment(xfs_mount_t *mp) | |
506 | { | |
62118709 | 507 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
508 | mp->m_sb.sb_inoalignmt >= |
509 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
510 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
511 | else | |
512 | mp->m_inoalign_mask = 0; | |
513 | /* | |
514 | * If we are using stripe alignment, check whether | |
515 | * the stripe unit is a multiple of the inode alignment | |
516 | */ | |
517 | if (mp->m_dalign && mp->m_inoalign_mask && | |
518 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
519 | mp->m_sinoalign = mp->m_dalign; | |
520 | else | |
521 | mp->m_sinoalign = 0; | |
0771fb45 ES |
522 | } |
523 | ||
524 | /* | |
0471f62e | 525 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
526 | */ |
527 | STATIC int | |
ba372674 DC |
528 | xfs_check_sizes( |
529 | struct xfs_mount *mp) | |
0771fb45 | 530 | { |
ba372674 | 531 | struct xfs_buf *bp; |
0771fb45 | 532 | xfs_daddr_t d; |
ba372674 | 533 | int error; |
0771fb45 | 534 | |
1da177e4 LT |
535 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
536 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 537 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 538 | return -EFBIG; |
1da177e4 | 539 | } |
ba372674 | 540 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 541 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
542 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
543 | if (error) { | |
0b932ccc | 544 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 545 | return error; |
1da177e4 | 546 | } |
1922c949 | 547 | xfs_buf_relse(bp); |
1da177e4 | 548 | |
ba372674 DC |
549 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
550 | return 0; | |
551 | ||
552 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
553 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
554 | xfs_warn(mp, "log size mismatch detected"); | |
555 | return -EFBIG; | |
556 | } | |
557 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 558 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
559 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
560 | if (error) { | |
561 | xfs_warn(mp, "log device read failed"); | |
562 | return error; | |
0771fb45 | 563 | } |
ba372674 | 564 | xfs_buf_relse(bp); |
0771fb45 ES |
565 | return 0; |
566 | } | |
567 | ||
7d095257 CH |
568 | /* |
569 | * Clear the quotaflags in memory and in the superblock. | |
570 | */ | |
571 | int | |
572 | xfs_mount_reset_sbqflags( | |
573 | struct xfs_mount *mp) | |
574 | { | |
7d095257 CH |
575 | mp->m_qflags = 0; |
576 | ||
61e63ecb | 577 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
578 | if (mp->m_sb.sb_qflags == 0) |
579 | return 0; | |
580 | spin_lock(&mp->m_sb_lock); | |
581 | mp->m_sb.sb_qflags = 0; | |
582 | spin_unlock(&mp->m_sb_lock); | |
583 | ||
61e63ecb | 584 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
585 | return 0; |
586 | ||
61e63ecb | 587 | return xfs_sync_sb(mp, false); |
7d095257 CH |
588 | } |
589 | ||
d5db0f97 ES |
590 | __uint64_t |
591 | xfs_default_resblks(xfs_mount_t *mp) | |
592 | { | |
593 | __uint64_t resblks; | |
594 | ||
595 | /* | |
8babd8a2 DC |
596 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
597 | * smaller. This is intended to cover concurrent allocation | |
598 | * transactions when we initially hit enospc. These each require a 4 | |
599 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
600 | * allocation reservations. | |
d5db0f97 ES |
601 | */ |
602 | resblks = mp->m_sb.sb_dblocks; | |
603 | do_div(resblks, 20); | |
8babd8a2 | 604 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
605 | return resblks; |
606 | } | |
607 | ||
0771fb45 | 608 | /* |
0771fb45 ES |
609 | * This function does the following on an initial mount of a file system: |
610 | * - reads the superblock from disk and init the mount struct | |
611 | * - if we're a 32-bit kernel, do a size check on the superblock | |
612 | * so we don't mount terabyte filesystems | |
613 | * - init mount struct realtime fields | |
614 | * - allocate inode hash table for fs | |
615 | * - init directory manager | |
616 | * - perform recovery and init the log manager | |
617 | */ | |
618 | int | |
619 | xfs_mountfs( | |
f0b2efad | 620 | struct xfs_mount *mp) |
0771fb45 | 621 | { |
f0b2efad BF |
622 | struct xfs_sb *sbp = &(mp->m_sb); |
623 | struct xfs_inode *rip; | |
624 | __uint64_t resblks; | |
625 | uint quotamount = 0; | |
626 | uint quotaflags = 0; | |
627 | int error = 0; | |
0771fb45 | 628 | |
ff55068c | 629 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 630 | |
ee1c0908 | 631 | /* |
074e427b DC |
632 | * Check for a mismatched features2 values. Older kernels read & wrote |
633 | * into the wrong sb offset for sb_features2 on some platforms due to | |
634 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
635 | * which made older superblock reading/writing routines swap it as a | |
636 | * 64-bit value. | |
ee1c0908 | 637 | * |
e6957ea4 ES |
638 | * For backwards compatibility, we make both slots equal. |
639 | * | |
074e427b DC |
640 | * If we detect a mismatched field, we OR the set bits into the existing |
641 | * features2 field in case it has already been modified; we don't want | |
642 | * to lose any features. We then update the bad location with the ORed | |
643 | * value so that older kernels will see any features2 flags. The | |
644 | * superblock writeback code ensures the new sb_features2 is copied to | |
645 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 646 | */ |
e6957ea4 | 647 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 648 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 649 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 650 | mp->m_update_sb = true; |
e6957ea4 ES |
651 | |
652 | /* | |
653 | * Re-check for ATTR2 in case it was found in bad_features2 | |
654 | * slot. | |
655 | */ | |
7c12f296 TS |
656 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
657 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 658 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
659 | } |
660 | ||
661 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
662 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
663 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 664 | mp->m_update_sb = true; |
e6957ea4 | 665 | |
7c12f296 TS |
666 | /* update sb_versionnum for the clearing of the morebits */ |
667 | if (!sbp->sb_features2) | |
61e63ecb | 668 | mp->m_update_sb = true; |
ee1c0908 DC |
669 | } |
670 | ||
263997a6 DC |
671 | /* always use v2 inodes by default now */ |
672 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
673 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 674 | mp->m_update_sb = true; |
263997a6 DC |
675 | } |
676 | ||
0771fb45 ES |
677 | /* |
678 | * Check if sb_agblocks is aligned at stripe boundary | |
679 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
680 | * allocator alignment is within an ag, therefore ag has | |
681 | * to be aligned at stripe boundary. | |
682 | */ | |
7884bc86 | 683 | error = xfs_update_alignment(mp); |
0771fb45 | 684 | if (error) |
f9057e3d | 685 | goto out; |
0771fb45 ES |
686 | |
687 | xfs_alloc_compute_maxlevels(mp); | |
688 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
689 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
690 | xfs_ialloc_compute_maxlevels(mp); | |
035e00ac | 691 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 692 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 ES |
693 | |
694 | xfs_set_maxicount(mp); | |
695 | ||
e6b3bb78 CM |
696 | /* enable fail_at_unmount as default */ |
697 | mp->m_fail_unmount = 1; | |
698 | ||
a31b1d3d | 699 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
700 | if (error) |
701 | goto out; | |
1da177e4 | 702 | |
225e4635 BD |
703 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
704 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
705 | if (error) |
706 | goto out_remove_sysfs; | |
707 | ||
192852be | 708 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
709 | if (error) |
710 | goto out_del_stats; | |
711 | ||
192852be CM |
712 | |
713 | error = xfs_uuid_mount(mp); | |
714 | if (error) | |
715 | goto out_remove_error_sysfs; | |
716 | ||
0771fb45 ES |
717 | /* |
718 | * Set the minimum read and write sizes | |
719 | */ | |
720 | xfs_set_rw_sizes(mp); | |
721 | ||
055388a3 DC |
722 | /* set the low space thresholds for dynamic preallocation */ |
723 | xfs_set_low_space_thresholds(mp); | |
724 | ||
0771fb45 ES |
725 | /* |
726 | * Set the inode cluster size. | |
727 | * This may still be overridden by the file system | |
728 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
729 | * |
730 | * For v5 filesystems, scale the cluster size with the inode size to | |
731 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
732 | * has set the inode alignment value appropriately for larger cluster | |
733 | * sizes. | |
0771fb45 ES |
734 | */ |
735 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
736 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
737 | int new_size = mp->m_inode_cluster_size; | |
738 | ||
739 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
740 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
741 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 742 | } |
0771fb45 | 743 | |
e5376fc1 BF |
744 | /* |
745 | * If enabled, sparse inode chunk alignment is expected to match the | |
746 | * cluster size. Full inode chunk alignment must match the chunk size, | |
747 | * but that is checked on sb read verification... | |
748 | */ | |
749 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
750 | mp->m_sb.sb_spino_align != | |
751 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { | |
752 | xfs_warn(mp, | |
753 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
754 | mp->m_sb.sb_spino_align, | |
755 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); | |
756 | error = -EINVAL; | |
757 | goto out_remove_uuid; | |
758 | } | |
759 | ||
0771fb45 ES |
760 | /* |
761 | * Set inode alignment fields | |
762 | */ | |
763 | xfs_set_inoalignment(mp); | |
764 | ||
765 | /* | |
c2bfbc9b | 766 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 767 | */ |
4249023a | 768 | error = xfs_check_sizes(mp); |
0771fb45 | 769 | if (error) |
f9057e3d | 770 | goto out_remove_uuid; |
0771fb45 | 771 | |
1da177e4 LT |
772 | /* |
773 | * Initialize realtime fields in the mount structure | |
774 | */ | |
0771fb45 ES |
775 | error = xfs_rtmount_init(mp); |
776 | if (error) { | |
0b932ccc | 777 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 778 | goto out_remove_uuid; |
1da177e4 LT |
779 | } |
780 | ||
1da177e4 LT |
781 | /* |
782 | * Copies the low order bits of the timestamp and the randomly | |
783 | * set "sequence" number out of a UUID. | |
784 | */ | |
785 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
786 | ||
1da177e4 LT |
787 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
788 | ||
0650b554 DC |
789 | error = xfs_da_mount(mp); |
790 | if (error) { | |
791 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
792 | goto out_remove_uuid; | |
793 | } | |
1da177e4 LT |
794 | |
795 | /* | |
796 | * Initialize the precomputed transaction reservations values. | |
797 | */ | |
798 | xfs_trans_init(mp); | |
799 | ||
1da177e4 LT |
800 | /* |
801 | * Allocate and initialize the per-ag data. | |
802 | */ | |
1c1c6ebc | 803 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 804 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
805 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
806 | if (error) { | |
0b932ccc | 807 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 808 | goto out_free_dir; |
1c1c6ebc | 809 | } |
1da177e4 | 810 | |
f9057e3d | 811 | if (!sbp->sb_logblocks) { |
0b932ccc | 812 | xfs_warn(mp, "no log defined"); |
f9057e3d | 813 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 814 | error = -EFSCORRUPTED; |
f9057e3d CH |
815 | goto out_free_perag; |
816 | } | |
817 | ||
1da177e4 | 818 | /* |
f0b2efad BF |
819 | * Log's mount-time initialization. The first part of recovery can place |
820 | * some items on the AIL, to be handled when recovery is finished or | |
821 | * cancelled. | |
1da177e4 | 822 | */ |
f9057e3d CH |
823 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
824 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
825 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
826 | if (error) { | |
0b932ccc | 827 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 828 | goto out_fail_wait; |
1da177e4 LT |
829 | } |
830 | ||
92821e2b DC |
831 | /* |
832 | * Now the log is mounted, we know if it was an unclean shutdown or | |
833 | * not. If it was, with the first phase of recovery has completed, we | |
834 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
835 | * but they are recovered transactionally in the second recovery phase | |
836 | * later. | |
837 | * | |
838 | * Hence we can safely re-initialise incore superblock counters from | |
839 | * the per-ag data. These may not be correct if the filesystem was not | |
840 | * cleanly unmounted, so we need to wait for recovery to finish before | |
841 | * doing this. | |
842 | * | |
843 | * If the filesystem was cleanly unmounted, then we can trust the | |
844 | * values in the superblock to be correct and we don't need to do | |
845 | * anything here. | |
846 | * | |
847 | * If we are currently making the filesystem, the initialisation will | |
848 | * fail as the perag data is in an undefined state. | |
849 | */ | |
92821e2b DC |
850 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
851 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
852 | !mp->m_sb.sb_inprogress) { | |
853 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 854 | if (error) |
6eee8972 | 855 | goto out_log_dealloc; |
92821e2b | 856 | } |
f9057e3d | 857 | |
1da177e4 LT |
858 | /* |
859 | * Get and sanity-check the root inode. | |
860 | * Save the pointer to it in the mount structure. | |
861 | */ | |
7b6259e7 | 862 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 863 | if (error) { |
0b932ccc | 864 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 865 | goto out_log_dealloc; |
1da177e4 LT |
866 | } |
867 | ||
868 | ASSERT(rip != NULL); | |
1da177e4 | 869 | |
c19b3b05 | 870 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 871 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 872 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
873 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
874 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
875 | mp); | |
2451337d | 876 | error = -EFSCORRUPTED; |
f9057e3d | 877 | goto out_rele_rip; |
1da177e4 LT |
878 | } |
879 | mp->m_rootip = rip; /* save it */ | |
880 | ||
881 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
882 | ||
883 | /* | |
884 | * Initialize realtime inode pointers in the mount structure | |
885 | */ | |
0771fb45 ES |
886 | error = xfs_rtmount_inodes(mp); |
887 | if (error) { | |
1da177e4 LT |
888 | /* |
889 | * Free up the root inode. | |
890 | */ | |
0b932ccc | 891 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 892 | goto out_rele_rip; |
1da177e4 LT |
893 | } |
894 | ||
895 | /* | |
7884bc86 CH |
896 | * If this is a read-only mount defer the superblock updates until |
897 | * the next remount into writeable mode. Otherwise we would never | |
898 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 899 | */ |
61e63ecb DC |
900 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
901 | error = xfs_sync_sb(mp, false); | |
e5720eec | 902 | if (error) { |
0b932ccc | 903 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 904 | goto out_rtunmount; |
e5720eec DC |
905 | } |
906 | } | |
1da177e4 LT |
907 | |
908 | /* | |
909 | * Initialise the XFS quota management subsystem for this mount | |
910 | */ | |
7d095257 CH |
911 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
912 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
913 | if (error) | |
914 | goto out_rtunmount; | |
915 | } else { | |
916 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
917 | ||
918 | /* | |
919 | * If a file system had quotas running earlier, but decided to | |
920 | * mount without -o uquota/pquota/gquota options, revoke the | |
921 | * quotachecked license. | |
922 | */ | |
923 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 924 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
925 | error = xfs_mount_reset_sbqflags(mp); |
926 | if (error) | |
a70a4fa5 | 927 | goto out_rtunmount; |
7d095257 CH |
928 | } |
929 | } | |
1da177e4 | 930 | |
17c12bcd DW |
931 | /* |
932 | * During the second phase of log recovery, we need iget and | |
933 | * iput to behave like they do for an active filesystem. | |
934 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
935 | * of inodes before we're done replaying log items on those | |
936 | * inodes. | |
937 | */ | |
938 | mp->m_super->s_flags |= MS_ACTIVE; | |
939 | ||
1da177e4 | 940 | /* |
f0b2efad BF |
941 | * Finish recovering the file system. This part needed to be delayed |
942 | * until after the root and real-time bitmap inodes were consistently | |
943 | * read in. | |
1da177e4 | 944 | */ |
4249023a | 945 | error = xfs_log_mount_finish(mp); |
1da177e4 | 946 | if (error) { |
0b932ccc | 947 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 948 | goto out_rtunmount; |
1da177e4 LT |
949 | } |
950 | ||
ddeb14f4 DC |
951 | /* |
952 | * Now the log is fully replayed, we can transition to full read-only | |
953 | * mode for read-only mounts. This will sync all the metadata and clean | |
954 | * the log so that the recovery we just performed does not have to be | |
955 | * replayed again on the next mount. | |
956 | * | |
957 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
958 | * semantically identical operations. | |
959 | */ | |
960 | if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) == | |
961 | XFS_MOUNT_RDONLY) { | |
962 | xfs_quiesce_attr(mp); | |
963 | } | |
964 | ||
1da177e4 LT |
965 | /* |
966 | * Complete the quota initialisation, post-log-replay component. | |
967 | */ | |
7d095257 CH |
968 | if (quotamount) { |
969 | ASSERT(mp->m_qflags == 0); | |
970 | mp->m_qflags = quotaflags; | |
971 | ||
972 | xfs_qm_mount_quotas(mp); | |
973 | } | |
974 | ||
84e1e99f DC |
975 | /* |
976 | * Now we are mounted, reserve a small amount of unused space for | |
977 | * privileged transactions. This is needed so that transaction | |
978 | * space required for critical operations can dip into this pool | |
979 | * when at ENOSPC. This is needed for operations like create with | |
980 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
981 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
982 | * |
983 | * This may drive us straight to ENOSPC on mount, but that implies | |
984 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 985 | */ |
d5db0f97 ES |
986 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
987 | resblks = xfs_default_resblks(mp); | |
988 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
989 | if (error) | |
0b932ccc DC |
990 | xfs_warn(mp, |
991 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
174edb0e DW |
992 | |
993 | /* Recover any CoW blocks that never got remapped. */ | |
994 | error = xfs_reflink_recover_cow(mp); | |
995 | if (error) { | |
996 | xfs_err(mp, | |
997 | "Error %d recovering leftover CoW allocations.", error); | |
998 | xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE); | |
999 | goto out_quota; | |
1000 | } | |
84d69619 DW |
1001 | |
1002 | /* Reserve AG blocks for future btree expansion. */ | |
1003 | error = xfs_fs_reserve_ag_blocks(mp); | |
1004 | if (error && error != -ENOSPC) | |
1005 | goto out_agresv; | |
d5db0f97 | 1006 | } |
84e1e99f | 1007 | |
1da177e4 LT |
1008 | return 0; |
1009 | ||
84d69619 DW |
1010 | out_agresv: |
1011 | xfs_fs_unreserve_ag_blocks(mp); | |
174edb0e DW |
1012 | out_quota: |
1013 | xfs_qm_unmount_quotas(mp); | |
b93b6e43 | 1014 | out_rtunmount: |
d0992452 | 1015 | mp->m_super->s_flags &= ~MS_ACTIVE; |
b93b6e43 | 1016 | xfs_rtunmount_inodes(mp); |
f9057e3d | 1017 | out_rele_rip: |
43355099 | 1018 | IRELE(rip); |
0ae120f8 BF |
1019 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
1020 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
f9057e3d | 1021 | out_log_dealloc: |
e6b3bb78 | 1022 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 1023 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
1024 | out_fail_wait: |
1025 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
1026 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1027 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1028 | out_free_perag: |
ff4f038c | 1029 | xfs_free_perag(mp); |
0650b554 DC |
1030 | out_free_dir: |
1031 | xfs_da_unmount(mp); | |
f9057e3d | 1032 | out_remove_uuid: |
27174203 | 1033 | xfs_uuid_unmount(mp); |
192852be CM |
1034 | out_remove_error_sysfs: |
1035 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
1036 | out_del_stats: |
1037 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
1038 | out_remove_sysfs: |
1039 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1040 | out: |
1da177e4 LT |
1041 | return error; |
1042 | } | |
1043 | ||
1044 | /* | |
1da177e4 LT |
1045 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1046 | * log and makes sure that incore structures are freed. | |
1047 | */ | |
41b5c2e7 CH |
1048 | void |
1049 | xfs_unmountfs( | |
1050 | struct xfs_mount *mp) | |
1da177e4 | 1051 | { |
41b5c2e7 CH |
1052 | __uint64_t resblks; |
1053 | int error; | |
1da177e4 | 1054 | |
579b62fa | 1055 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
83104d44 | 1056 | cancel_delayed_work_sync(&mp->m_cowblocks_work); |
579b62fa | 1057 | |
84d69619 | 1058 | xfs_fs_unreserve_ag_blocks(mp); |
7d095257 | 1059 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1060 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1061 | IRELE(mp->m_rootip); |
1062 | ||
641c56fb DC |
1063 | /* |
1064 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1065 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1066 | * the transaction is still sitting in a iclog. The stale inodes |
1067 | * on that buffer will have their flush locks held until the | |
1068 | * transaction hits the disk and the callbacks run. the inode | |
1069 | * flush takes the flush lock unconditionally and with nothing to | |
1070 | * push out the iclog we will never get that unlocked. hence we | |
1071 | * need to force the log first. | |
1072 | */ | |
a14a348b | 1073 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1074 | |
e6b3bb78 CM |
1075 | /* |
1076 | * We now need to tell the world we are unmounting. This will allow | |
1077 | * us to detect that the filesystem is going away and we should error | |
1078 | * out anything that we have been retrying in the background. This will | |
1079 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1080 | */ | |
1081 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1082 | ||
c854363e | 1083 | /* |
211e4d43 CH |
1084 | * Flush all pending changes from the AIL. |
1085 | */ | |
1086 | xfs_ail_push_all_sync(mp->m_ail); | |
1087 | ||
1088 | /* | |
1089 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1090 | * inodes and none should be pinned or locked, but use synchronous |
1091 | * reclaim just to be sure. We can stop background inode reclaim | |
1092 | * here as well if it is still running. | |
c854363e | 1093 | */ |
7e18530b | 1094 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1095 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1096 | |
7d095257 | 1097 | xfs_qm_unmount(mp); |
a357a121 | 1098 | |
84e1e99f DC |
1099 | /* |
1100 | * Unreserve any blocks we have so that when we unmount we don't account | |
1101 | * the reserved free space as used. This is really only necessary for | |
1102 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1103 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1104 | * |
1105 | * We don't bother correcting this elsewhere for lazy superblock | |
1106 | * counting because on mount of an unclean filesystem we reconstruct the | |
1107 | * correct counter value and this is irrelevant. | |
1108 | * | |
1109 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1110 | * we only every apply deltas to the superblock and hence the incore | |
1111 | * value does not matter.... | |
1112 | */ | |
1113 | resblks = 0; | |
714082bc DC |
1114 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1115 | if (error) | |
0b932ccc | 1116 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1117 | "Freespace may not be correct on next mount."); |
1118 | ||
adab0f67 | 1119 | error = xfs_log_sbcount(mp); |
e5720eec | 1120 | if (error) |
0b932ccc | 1121 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1122 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1123 | |
225e4635 | 1124 | |
21b699c8 | 1125 | xfs_log_unmount(mp); |
0650b554 | 1126 | xfs_da_unmount(mp); |
27174203 | 1127 | xfs_uuid_unmount(mp); |
1da177e4 | 1128 | |
1550d0b0 | 1129 | #if defined(DEBUG) |
0ce4cfd4 | 1130 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1131 | #endif |
ff4f038c | 1132 | xfs_free_perag(mp); |
a31b1d3d | 1133 | |
192852be | 1134 | xfs_error_sysfs_del(mp); |
225e4635 | 1135 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1136 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1137 | } |
1138 | ||
91ee575f BF |
1139 | /* |
1140 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1141 | * freeze level for which modifications should not be allowed. This allows | |
1142 | * certain operations to proceed while the freeze sequence is in progress, if | |
1143 | * necessary. | |
1144 | */ | |
1145 | bool | |
1146 | xfs_fs_writable( | |
1147 | struct xfs_mount *mp, | |
1148 | int level) | |
92821e2b | 1149 | { |
91ee575f BF |
1150 | ASSERT(level > SB_UNFROZEN); |
1151 | if ((mp->m_super->s_writers.frozen >= level) || | |
1152 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1153 | return false; | |
1154 | ||
1155 | return true; | |
92821e2b DC |
1156 | } |
1157 | ||
1158 | /* | |
b2ce3974 AE |
1159 | * xfs_log_sbcount |
1160 | * | |
adab0f67 | 1161 | * Sync the superblock counters to disk. |
b2ce3974 | 1162 | * |
91ee575f BF |
1163 | * Note this code can be called during the process of freezing, so we use the |
1164 | * transaction allocator that does not block when the transaction subsystem is | |
1165 | * in its frozen state. | |
92821e2b DC |
1166 | */ |
1167 | int | |
adab0f67 | 1168 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1169 | { |
91ee575f BF |
1170 | /* allow this to proceed during the freeze sequence... */ |
1171 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1172 | return 0; |
1173 | ||
92821e2b DC |
1174 | /* |
1175 | * we don't need to do this if we are updating the superblock | |
1176 | * counters on every modification. | |
1177 | */ | |
1178 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1179 | return 0; | |
1180 | ||
61e63ecb | 1181 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1182 | } |
1183 | ||
8c1903d3 DC |
1184 | /* |
1185 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1186 | * of 128 so we don't need to take the counter lock on every update. | |
1187 | */ | |
1188 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1189 | int |
1190 | xfs_mod_icount( | |
1191 | struct xfs_mount *mp, | |
1192 | int64_t delta) | |
1193 | { | |
8c1903d3 DC |
1194 | __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
1195 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { | |
501ab323 DC |
1196 | ASSERT(0); |
1197 | percpu_counter_add(&mp->m_icount, -delta); | |
1198 | return -EINVAL; | |
1199 | } | |
1200 | return 0; | |
1201 | } | |
1202 | ||
e88b64ea DC |
1203 | int |
1204 | xfs_mod_ifree( | |
1205 | struct xfs_mount *mp, | |
1206 | int64_t delta) | |
1207 | { | |
1208 | percpu_counter_add(&mp->m_ifree, delta); | |
1209 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1210 | ASSERT(0); | |
1211 | percpu_counter_add(&mp->m_ifree, -delta); | |
1212 | return -EINVAL; | |
1213 | } | |
1214 | return 0; | |
1215 | } | |
0d485ada | 1216 | |
8c1903d3 DC |
1217 | /* |
1218 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1219 | * only occurs on frequent small block count updates such as in the delayed | |
1220 | * allocation path for buffered writes (page a time updates). Hence we set | |
1221 | * a large batch count (1024) to minimise global counter updates except when | |
1222 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1223 | */ | |
1224 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1225 | int |
1226 | xfs_mod_fdblocks( | |
1227 | struct xfs_mount *mp, | |
1228 | int64_t delta, | |
1229 | bool rsvd) | |
1230 | { | |
1231 | int64_t lcounter; | |
1232 | long long res_used; | |
1233 | s32 batch; | |
1234 | ||
1235 | if (delta > 0) { | |
1236 | /* | |
1237 | * If the reserve pool is depleted, put blocks back into it | |
1238 | * first. Most of the time the pool is full. | |
1239 | */ | |
1240 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1241 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1242 | return 0; | |
1243 | } | |
1244 | ||
1245 | spin_lock(&mp->m_sb_lock); | |
1246 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1247 | ||
1248 | if (res_used > delta) { | |
1249 | mp->m_resblks_avail += delta; | |
1250 | } else { | |
1251 | delta -= res_used; | |
1252 | mp->m_resblks_avail = mp->m_resblks; | |
1253 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1254 | } | |
1255 | spin_unlock(&mp->m_sb_lock); | |
1256 | return 0; | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Taking blocks away, need to be more accurate the closer we | |
1261 | * are to zero. | |
1262 | * | |
0d485ada DC |
1263 | * If the counter has a value of less than 2 * max batch size, |
1264 | * then make everything serialise as we are real close to | |
1265 | * ENOSPC. | |
1266 | */ | |
8c1903d3 DC |
1267 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1268 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1269 | batch = 1; |
1270 | else | |
8c1903d3 | 1271 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada DC |
1272 | |
1273 | __percpu_counter_add(&mp->m_fdblocks, delta, batch); | |
52548852 | 1274 | if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside, |
8c1903d3 | 1275 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1276 | /* we had space! */ |
1277 | return 0; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * lock up the sb for dipping into reserves before releasing the space | |
1282 | * that took us to ENOSPC. | |
1283 | */ | |
1284 | spin_lock(&mp->m_sb_lock); | |
1285 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1286 | if (!rsvd) | |
1287 | goto fdblocks_enospc; | |
1288 | ||
1289 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1290 | if (lcounter >= 0) { | |
1291 | mp->m_resblks_avail = lcounter; | |
1292 | spin_unlock(&mp->m_sb_lock); | |
1293 | return 0; | |
1294 | } | |
1295 | printk_once(KERN_WARNING | |
1296 | "Filesystem \"%s\": reserve blocks depleted! " | |
1297 | "Consider increasing reserve pool size.", | |
1298 | mp->m_fsname); | |
1299 | fdblocks_enospc: | |
1300 | spin_unlock(&mp->m_sb_lock); | |
1301 | return -ENOSPC; | |
1302 | } | |
1303 | ||
bab98bbe DC |
1304 | int |
1305 | xfs_mod_frextents( | |
1306 | struct xfs_mount *mp, | |
1307 | int64_t delta) | |
1308 | { | |
1309 | int64_t lcounter; | |
1310 | int ret = 0; | |
1311 | ||
1312 | spin_lock(&mp->m_sb_lock); | |
1313 | lcounter = mp->m_sb.sb_frextents + delta; | |
1314 | if (lcounter < 0) | |
1315 | ret = -ENOSPC; | |
1316 | else | |
1317 | mp->m_sb.sb_frextents = lcounter; | |
1318 | spin_unlock(&mp->m_sb_lock); | |
1319 | return ret; | |
1320 | } | |
1321 | ||
1da177e4 LT |
1322 | /* |
1323 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1324 | * The buffer is returned locked and read in from disk. | |
1325 | * The buffer should be released with a call to xfs_brelse(). | |
1326 | * | |
1327 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1328 | * the superblock buffer if it can be locked without sleeping. | |
1329 | * If it can't then we'll return NULL. | |
1330 | */ | |
0c842ad4 | 1331 | struct xfs_buf * |
1da177e4 | 1332 | xfs_getsb( |
0c842ad4 CH |
1333 | struct xfs_mount *mp, |
1334 | int flags) | |
1da177e4 | 1335 | { |
0c842ad4 | 1336 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1337 | |
0c842ad4 CH |
1338 | if (!xfs_buf_trylock(bp)) { |
1339 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1340 | return NULL; |
0c842ad4 | 1341 | xfs_buf_lock(bp); |
1da177e4 | 1342 | } |
0c842ad4 | 1343 | |
72790aa1 | 1344 | xfs_buf_hold(bp); |
b0388bf1 | 1345 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1346 | return bp; |
1da177e4 LT |
1347 | } |
1348 | ||
1349 | /* | |
1350 | * Used to free the superblock along various error paths. | |
1351 | */ | |
1352 | void | |
1353 | xfs_freesb( | |
26af6552 | 1354 | struct xfs_mount *mp) |
1da177e4 | 1355 | { |
26af6552 | 1356 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1357 | |
26af6552 | 1358 | xfs_buf_lock(bp); |
1da177e4 | 1359 | mp->m_sb_bp = NULL; |
26af6552 | 1360 | xfs_buf_relse(bp); |
1da177e4 LT |
1361 | } |
1362 | ||
dda35b8f CH |
1363 | /* |
1364 | * If the underlying (data/log/rt) device is readonly, there are some | |
1365 | * operations that cannot proceed. | |
1366 | */ | |
1367 | int | |
1368 | xfs_dev_is_read_only( | |
1369 | struct xfs_mount *mp, | |
1370 | char *message) | |
1371 | { | |
1372 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1373 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1374 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1375 | xfs_notice(mp, "%s required on read-only device.", message); |
1376 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1377 | return -EROFS; |
dda35b8f CH |
1378 | } |
1379 | return 0; | |
1380 | } |