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" |
0b1b213f | 47 | |
1da177e4 | 48 | |
27174203 CH |
49 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
50 | static int xfs_uuid_table_size; | |
51 | static uuid_t *xfs_uuid_table; | |
52 | ||
af3b6382 DW |
53 | void |
54 | xfs_uuid_table_free(void) | |
55 | { | |
56 | if (xfs_uuid_table_size == 0) | |
57 | return; | |
58 | kmem_free(xfs_uuid_table); | |
59 | xfs_uuid_table = NULL; | |
60 | xfs_uuid_table_size = 0; | |
61 | } | |
62 | ||
27174203 CH |
63 | /* |
64 | * See if the UUID is unique among mounted XFS filesystems. | |
65 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
66 | */ | |
67 | STATIC int | |
68 | xfs_uuid_mount( | |
69 | struct xfs_mount *mp) | |
70 | { | |
71 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
72 | int hole, i; | |
73 | ||
74 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
75 | return 0; | |
76 | ||
77 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 78 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 79 | return -EINVAL; |
27174203 CH |
80 | } |
81 | ||
82 | mutex_lock(&xfs_uuid_table_mutex); | |
83 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
84 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
85 | hole = i; | |
86 | continue; | |
87 | } | |
88 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
89 | goto out_duplicate; | |
90 | } | |
91 | ||
92 | if (hole < 0) { | |
93 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
94 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
27174203 CH |
95 | KM_SLEEP); |
96 | hole = xfs_uuid_table_size++; | |
97 | } | |
98 | xfs_uuid_table[hole] = *uuid; | |
99 | mutex_unlock(&xfs_uuid_table_mutex); | |
100 | ||
101 | return 0; | |
102 | ||
103 | out_duplicate: | |
104 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 105 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 106 | return -EINVAL; |
27174203 CH |
107 | } |
108 | ||
109 | STATIC void | |
110 | xfs_uuid_unmount( | |
111 | struct xfs_mount *mp) | |
112 | { | |
113 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
114 | int i; | |
115 | ||
116 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
117 | return; | |
118 | ||
119 | mutex_lock(&xfs_uuid_table_mutex); | |
120 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
121 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
122 | continue; | |
123 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
124 | continue; | |
125 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
126 | break; | |
127 | } | |
128 | ASSERT(i < xfs_uuid_table_size); | |
129 | mutex_unlock(&xfs_uuid_table_mutex); | |
130 | } | |
131 | ||
132 | ||
e176579e DC |
133 | STATIC void |
134 | __xfs_free_perag( | |
135 | struct rcu_head *head) | |
136 | { | |
137 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
138 | ||
139 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
140 | kmem_free(pag); | |
141 | } | |
142 | ||
1da177e4 | 143 | /* |
e176579e | 144 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 145 | */ |
c962fb79 | 146 | STATIC void |
ff4f038c | 147 | xfs_free_perag( |
745f6919 | 148 | xfs_mount_t *mp) |
1da177e4 | 149 | { |
1c1c6ebc DC |
150 | xfs_agnumber_t agno; |
151 | struct xfs_perag *pag; | |
152 | ||
153 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
154 | spin_lock(&mp->m_perag_lock); | |
155 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
156 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 157 | ASSERT(pag); |
f83282a8 | 158 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 159 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 160 | } |
1da177e4 LT |
161 | } |
162 | ||
4cc929ee NS |
163 | /* |
164 | * Check size of device based on the (data/realtime) block count. | |
165 | * Note: this check is used by the growfs code as well as mount. | |
166 | */ | |
167 | int | |
168 | xfs_sb_validate_fsb_count( | |
169 | xfs_sb_t *sbp, | |
170 | __uint64_t nblocks) | |
171 | { | |
172 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
173 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
174 | ||
d5cf09ba | 175 | /* Limited by ULONG_MAX of page cache index */ |
09cbfeaf | 176 | if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) |
2451337d | 177 | return -EFBIG; |
4cc929ee NS |
178 | return 0; |
179 | } | |
1da177e4 | 180 | |
1c1c6ebc | 181 | int |
c11e2c36 | 182 | xfs_initialize_perag( |
c11e2c36 | 183 | xfs_mount_t *mp, |
1c1c6ebc DC |
184 | xfs_agnumber_t agcount, |
185 | xfs_agnumber_t *maxagi) | |
1da177e4 | 186 | { |
2d2194f6 | 187 | xfs_agnumber_t index; |
8b26c582 | 188 | xfs_agnumber_t first_initialised = 0; |
1da177e4 | 189 | xfs_perag_t *pag; |
8b26c582 | 190 | int error = -ENOMEM; |
1da177e4 | 191 | |
1c1c6ebc DC |
192 | /* |
193 | * Walk the current per-ag tree so we don't try to initialise AGs | |
194 | * that already exist (growfs case). Allocate and insert all the | |
195 | * AGs we don't find ready for initialisation. | |
196 | */ | |
197 | for (index = 0; index < agcount; index++) { | |
198 | pag = xfs_perag_get(mp, index); | |
199 | if (pag) { | |
200 | xfs_perag_put(pag); | |
201 | continue; | |
202 | } | |
8b26c582 DC |
203 | if (!first_initialised) |
204 | first_initialised = index; | |
fb3b504a | 205 | |
1c1c6ebc DC |
206 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
207 | if (!pag) | |
8b26c582 | 208 | goto out_unwind; |
fb3b504a CH |
209 | pag->pag_agno = index; |
210 | pag->pag_mount = mp; | |
1a427ab0 | 211 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 212 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 213 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
214 | spin_lock_init(&pag->pag_buf_lock); |
215 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 216 | |
1c1c6ebc | 217 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 218 | goto out_unwind; |
fb3b504a | 219 | |
1c1c6ebc DC |
220 | spin_lock(&mp->m_perag_lock); |
221 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
222 | BUG(); | |
223 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
224 | radix_tree_preload_end(); |
225 | error = -EEXIST; | |
226 | goto out_unwind; | |
1c1c6ebc DC |
227 | } |
228 | spin_unlock(&mp->m_perag_lock); | |
229 | radix_tree_preload_end(); | |
230 | } | |
231 | ||
12c3f05c | 232 | index = xfs_set_inode_alloc(mp, agcount); |
fb3b504a | 233 | |
1c1c6ebc DC |
234 | if (maxagi) |
235 | *maxagi = index; | |
8018026e DW |
236 | |
237 | mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp); | |
1c1c6ebc | 238 | return 0; |
8b26c582 DC |
239 | |
240 | out_unwind: | |
241 | kmem_free(pag); | |
242 | for (; index > first_initialised; index--) { | |
243 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
244 | kmem_free(pag); | |
245 | } | |
246 | return error; | |
1da177e4 LT |
247 | } |
248 | ||
1da177e4 LT |
249 | /* |
250 | * xfs_readsb | |
251 | * | |
252 | * Does the initial read of the superblock. | |
253 | */ | |
254 | int | |
ff55068c DC |
255 | xfs_readsb( |
256 | struct xfs_mount *mp, | |
257 | int flags) | |
1da177e4 LT |
258 | { |
259 | unsigned int sector_size; | |
04a1e6c5 DC |
260 | struct xfs_buf *bp; |
261 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 262 | int error; |
af34e09d | 263 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 264 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
265 | |
266 | ASSERT(mp->m_sb_bp == NULL); | |
267 | ASSERT(mp->m_ddev_targp != NULL); | |
268 | ||
daba5427 ES |
269 | /* |
270 | * For the initial read, we must guess at the sector | |
271 | * size based on the block device. It's enough to | |
272 | * get the sb_sectsize out of the superblock and | |
273 | * then reread with the proper length. | |
274 | * We don't verify it yet, because it may not be complete. | |
275 | */ | |
276 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
277 | buf_ops = NULL; | |
278 | ||
1da177e4 | 279 | /* |
c891c30a BF |
280 | * Allocate a (locked) buffer to hold the superblock. This will be kept |
281 | * around at all times to optimize access to the superblock. Therefore, | |
282 | * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count | |
283 | * elevated. | |
1da177e4 | 284 | */ |
26af6552 | 285 | reread: |
ba372674 | 286 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
c891c30a BF |
287 | BTOBB(sector_size), XBF_NO_IOACCT, &bp, |
288 | buf_ops); | |
ba372674 | 289 | if (error) { |
eab4e633 | 290 | if (loud) |
e721f504 | 291 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 292 | /* bad CRC means corrupted metadata */ |
2451337d DC |
293 | if (error == -EFSBADCRC) |
294 | error = -EFSCORRUPTED; | |
ba372674 | 295 | return error; |
eab4e633 | 296 | } |
1da177e4 LT |
297 | |
298 | /* | |
299 | * Initialize the mount structure from the superblock. | |
1da177e4 | 300 | */ |
556b8883 | 301 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
302 | |
303 | /* | |
304 | * If we haven't validated the superblock, do so now before we try | |
305 | * to check the sector size and reread the superblock appropriately. | |
306 | */ | |
307 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
308 | if (loud) | |
309 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 310 | error = -EINVAL; |
556b8883 DC |
311 | goto release_buf; |
312 | } | |
ff55068c | 313 | |
1da177e4 LT |
314 | /* |
315 | * We must be able to do sector-sized and sector-aligned IO. | |
316 | */ | |
04a1e6c5 | 317 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
318 | if (loud) |
319 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 320 | sector_size, sbp->sb_sectsize); |
2451337d | 321 | error = -ENOSYS; |
26af6552 | 322 | goto release_buf; |
1da177e4 LT |
323 | } |
324 | ||
daba5427 | 325 | if (buf_ops == NULL) { |
556b8883 DC |
326 | /* |
327 | * Re-read the superblock so the buffer is correctly sized, | |
328 | * and properly verified. | |
329 | */ | |
1da177e4 | 330 | xfs_buf_relse(bp); |
04a1e6c5 | 331 | sector_size = sbp->sb_sectsize; |
daba5427 | 332 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 333 | goto reread; |
1da177e4 LT |
334 | } |
335 | ||
5681ca40 | 336 | xfs_reinit_percpu_counters(mp); |
8d280b98 | 337 | |
04a1e6c5 DC |
338 | /* no need to be quiet anymore, so reset the buf ops */ |
339 | bp->b_ops = &xfs_sb_buf_ops; | |
340 | ||
1da177e4 | 341 | mp->m_sb_bp = bp; |
26af6552 | 342 | xfs_buf_unlock(bp); |
1da177e4 LT |
343 | return 0; |
344 | ||
26af6552 DC |
345 | release_buf: |
346 | xfs_buf_relse(bp); | |
1da177e4 LT |
347 | return error; |
348 | } | |
349 | ||
1da177e4 | 350 | /* |
0771fb45 | 351 | * Update alignment values based on mount options and sb values |
1da177e4 | 352 | */ |
0771fb45 | 353 | STATIC int |
7884bc86 | 354 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 355 | { |
1da177e4 | 356 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 357 | |
4249023a | 358 | if (mp->m_dalign) { |
1da177e4 LT |
359 | /* |
360 | * If stripe unit and stripe width are not multiples | |
361 | * of the fs blocksize turn off alignment. | |
362 | */ | |
363 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
364 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
365 | xfs_warn(mp, |
366 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
367 | sbp->sb_blocksize); | |
2451337d | 368 | return -EINVAL; |
1da177e4 LT |
369 | } else { |
370 | /* | |
371 | * Convert the stripe unit and width to FSBs. | |
372 | */ | |
373 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
374 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 375 | xfs_warn(mp, |
39a45d84 JL |
376 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
377 | sbp->sb_agblocks); | |
2451337d | 378 | return -EINVAL; |
1da177e4 LT |
379 | } else if (mp->m_dalign) { |
380 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
381 | } else { | |
39a45d84 JL |
382 | xfs_warn(mp, |
383 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
384 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 385 | return -EINVAL; |
1da177e4 LT |
386 | } |
387 | } | |
388 | ||
389 | /* | |
390 | * Update superblock with new values | |
391 | * and log changes | |
392 | */ | |
62118709 | 393 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
394 | if (sbp->sb_unit != mp->m_dalign) { |
395 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 396 | mp->m_update_sb = true; |
1da177e4 LT |
397 | } |
398 | if (sbp->sb_width != mp->m_swidth) { | |
399 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 400 | mp->m_update_sb = true; |
1da177e4 | 401 | } |
34d7f603 JL |
402 | } else { |
403 | xfs_warn(mp, | |
404 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 405 | return -EINVAL; |
1da177e4 LT |
406 | } |
407 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 408 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
409 | mp->m_dalign = sbp->sb_unit; |
410 | mp->m_swidth = sbp->sb_width; | |
411 | } | |
412 | ||
0771fb45 ES |
413 | return 0; |
414 | } | |
1da177e4 | 415 | |
0771fb45 ES |
416 | /* |
417 | * Set the maximum inode count for this filesystem | |
418 | */ | |
419 | STATIC void | |
420 | xfs_set_maxicount(xfs_mount_t *mp) | |
421 | { | |
422 | xfs_sb_t *sbp = &(mp->m_sb); | |
423 | __uint64_t icount; | |
1da177e4 | 424 | |
0771fb45 ES |
425 | if (sbp->sb_imax_pct) { |
426 | /* | |
427 | * Make sure the maximum inode count is a multiple | |
428 | * of the units we allocate inodes in. | |
1da177e4 | 429 | */ |
1da177e4 LT |
430 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
431 | do_div(icount, 100); | |
432 | do_div(icount, mp->m_ialloc_blks); | |
433 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
434 | sbp->sb_inopblog; | |
0771fb45 | 435 | } else { |
1da177e4 | 436 | mp->m_maxicount = 0; |
1da177e4 | 437 | } |
0771fb45 ES |
438 | } |
439 | ||
440 | /* | |
441 | * Set the default minimum read and write sizes unless | |
442 | * already specified in a mount option. | |
443 | * We use smaller I/O sizes when the file system | |
444 | * is being used for NFS service (wsync mount option). | |
445 | */ | |
446 | STATIC void | |
447 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
448 | { | |
449 | xfs_sb_t *sbp = &(mp->m_sb); | |
450 | int readio_log, writeio_log; | |
1da177e4 | 451 | |
1da177e4 LT |
452 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
453 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
454 | readio_log = XFS_WSYNC_READIO_LOG; | |
455 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
456 | } else { | |
457 | readio_log = XFS_READIO_LOG_LARGE; | |
458 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
459 | } | |
460 | } else { | |
461 | readio_log = mp->m_readio_log; | |
462 | writeio_log = mp->m_writeio_log; | |
463 | } | |
464 | ||
1da177e4 LT |
465 | if (sbp->sb_blocklog > readio_log) { |
466 | mp->m_readio_log = sbp->sb_blocklog; | |
467 | } else { | |
468 | mp->m_readio_log = readio_log; | |
469 | } | |
470 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
471 | if (sbp->sb_blocklog > writeio_log) { | |
472 | mp->m_writeio_log = sbp->sb_blocklog; | |
473 | } else { | |
474 | mp->m_writeio_log = writeio_log; | |
475 | } | |
476 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 477 | } |
1da177e4 | 478 | |
055388a3 DC |
479 | /* |
480 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
481 | */ | |
482 | void | |
483 | xfs_set_low_space_thresholds( | |
484 | struct xfs_mount *mp) | |
485 | { | |
486 | int i; | |
487 | ||
488 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
489 | __uint64_t space = mp->m_sb.sb_dblocks; | |
490 | ||
491 | do_div(space, 100); | |
492 | mp->m_low_space[i] = space * (i + 1); | |
493 | } | |
494 | } | |
495 | ||
496 | ||
0771fb45 ES |
497 | /* |
498 | * Set whether we're using inode alignment. | |
499 | */ | |
500 | STATIC void | |
501 | xfs_set_inoalignment(xfs_mount_t *mp) | |
502 | { | |
62118709 | 503 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
504 | mp->m_sb.sb_inoalignmt >= |
505 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
506 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
507 | else | |
508 | mp->m_inoalign_mask = 0; | |
509 | /* | |
510 | * If we are using stripe alignment, check whether | |
511 | * the stripe unit is a multiple of the inode alignment | |
512 | */ | |
513 | if (mp->m_dalign && mp->m_inoalign_mask && | |
514 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
515 | mp->m_sinoalign = mp->m_dalign; | |
516 | else | |
517 | mp->m_sinoalign = 0; | |
0771fb45 ES |
518 | } |
519 | ||
520 | /* | |
0471f62e | 521 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
522 | */ |
523 | STATIC int | |
ba372674 DC |
524 | xfs_check_sizes( |
525 | struct xfs_mount *mp) | |
0771fb45 | 526 | { |
ba372674 | 527 | struct xfs_buf *bp; |
0771fb45 | 528 | xfs_daddr_t d; |
ba372674 | 529 | int error; |
0771fb45 | 530 | |
1da177e4 LT |
531 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
532 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 533 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 534 | return -EFBIG; |
1da177e4 | 535 | } |
ba372674 | 536 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 537 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
538 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
539 | if (error) { | |
0b932ccc | 540 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 541 | return error; |
1da177e4 | 542 | } |
1922c949 | 543 | xfs_buf_relse(bp); |
1da177e4 | 544 | |
ba372674 DC |
545 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
546 | return 0; | |
547 | ||
548 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
549 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
550 | xfs_warn(mp, "log size mismatch detected"); | |
551 | return -EFBIG; | |
552 | } | |
553 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 554 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
555 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
556 | if (error) { | |
557 | xfs_warn(mp, "log device read failed"); | |
558 | return error; | |
0771fb45 | 559 | } |
ba372674 | 560 | xfs_buf_relse(bp); |
0771fb45 ES |
561 | return 0; |
562 | } | |
563 | ||
7d095257 CH |
564 | /* |
565 | * Clear the quotaflags in memory and in the superblock. | |
566 | */ | |
567 | int | |
568 | xfs_mount_reset_sbqflags( | |
569 | struct xfs_mount *mp) | |
570 | { | |
7d095257 CH |
571 | mp->m_qflags = 0; |
572 | ||
61e63ecb | 573 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
574 | if (mp->m_sb.sb_qflags == 0) |
575 | return 0; | |
576 | spin_lock(&mp->m_sb_lock); | |
577 | mp->m_sb.sb_qflags = 0; | |
578 | spin_unlock(&mp->m_sb_lock); | |
579 | ||
61e63ecb | 580 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
581 | return 0; |
582 | ||
61e63ecb | 583 | return xfs_sync_sb(mp, false); |
7d095257 CH |
584 | } |
585 | ||
d5db0f97 ES |
586 | __uint64_t |
587 | xfs_default_resblks(xfs_mount_t *mp) | |
588 | { | |
589 | __uint64_t resblks; | |
590 | ||
591 | /* | |
8babd8a2 DC |
592 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
593 | * smaller. This is intended to cover concurrent allocation | |
594 | * transactions when we initially hit enospc. These each require a 4 | |
595 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
596 | * allocation reservations. | |
d5db0f97 ES |
597 | */ |
598 | resblks = mp->m_sb.sb_dblocks; | |
599 | do_div(resblks, 20); | |
8babd8a2 | 600 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
601 | return resblks; |
602 | } | |
603 | ||
0771fb45 | 604 | /* |
0771fb45 ES |
605 | * This function does the following on an initial mount of a file system: |
606 | * - reads the superblock from disk and init the mount struct | |
607 | * - if we're a 32-bit kernel, do a size check on the superblock | |
608 | * so we don't mount terabyte filesystems | |
609 | * - init mount struct realtime fields | |
610 | * - allocate inode hash table for fs | |
611 | * - init directory manager | |
612 | * - perform recovery and init the log manager | |
613 | */ | |
614 | int | |
615 | xfs_mountfs( | |
f0b2efad | 616 | struct xfs_mount *mp) |
0771fb45 | 617 | { |
f0b2efad BF |
618 | struct xfs_sb *sbp = &(mp->m_sb); |
619 | struct xfs_inode *rip; | |
620 | __uint64_t resblks; | |
621 | uint quotamount = 0; | |
622 | uint quotaflags = 0; | |
623 | int error = 0; | |
0771fb45 | 624 | |
ff55068c | 625 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 626 | |
ee1c0908 | 627 | /* |
074e427b DC |
628 | * Check for a mismatched features2 values. Older kernels read & wrote |
629 | * into the wrong sb offset for sb_features2 on some platforms due to | |
630 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
631 | * which made older superblock reading/writing routines swap it as a | |
632 | * 64-bit value. | |
ee1c0908 | 633 | * |
e6957ea4 ES |
634 | * For backwards compatibility, we make both slots equal. |
635 | * | |
074e427b DC |
636 | * If we detect a mismatched field, we OR the set bits into the existing |
637 | * features2 field in case it has already been modified; we don't want | |
638 | * to lose any features. We then update the bad location with the ORed | |
639 | * value so that older kernels will see any features2 flags. The | |
640 | * superblock writeback code ensures the new sb_features2 is copied to | |
641 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 642 | */ |
e6957ea4 | 643 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 644 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 645 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 646 | mp->m_update_sb = true; |
e6957ea4 ES |
647 | |
648 | /* | |
649 | * Re-check for ATTR2 in case it was found in bad_features2 | |
650 | * slot. | |
651 | */ | |
7c12f296 TS |
652 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
653 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 654 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
655 | } |
656 | ||
657 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
658 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
659 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 660 | mp->m_update_sb = true; |
e6957ea4 | 661 | |
7c12f296 TS |
662 | /* update sb_versionnum for the clearing of the morebits */ |
663 | if (!sbp->sb_features2) | |
61e63ecb | 664 | mp->m_update_sb = true; |
ee1c0908 DC |
665 | } |
666 | ||
263997a6 DC |
667 | /* always use v2 inodes by default now */ |
668 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
669 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 670 | mp->m_update_sb = true; |
263997a6 DC |
671 | } |
672 | ||
0771fb45 ES |
673 | /* |
674 | * Check if sb_agblocks is aligned at stripe boundary | |
675 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
676 | * allocator alignment is within an ag, therefore ag has | |
677 | * to be aligned at stripe boundary. | |
678 | */ | |
7884bc86 | 679 | error = xfs_update_alignment(mp); |
0771fb45 | 680 | if (error) |
f9057e3d | 681 | goto out; |
0771fb45 ES |
682 | |
683 | xfs_alloc_compute_maxlevels(mp); | |
684 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
685 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
686 | xfs_ialloc_compute_maxlevels(mp); | |
035e00ac | 687 | xfs_rmapbt_compute_maxlevels(mp); |
1946b91c | 688 | xfs_refcountbt_compute_maxlevels(mp); |
0771fb45 ES |
689 | |
690 | xfs_set_maxicount(mp); | |
691 | ||
e6b3bb78 CM |
692 | /* enable fail_at_unmount as default */ |
693 | mp->m_fail_unmount = 1; | |
694 | ||
a31b1d3d | 695 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
696 | if (error) |
697 | goto out; | |
1da177e4 | 698 | |
225e4635 BD |
699 | error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype, |
700 | &mp->m_kobj, "stats"); | |
a31b1d3d BF |
701 | if (error) |
702 | goto out_remove_sysfs; | |
703 | ||
192852be | 704 | error = xfs_error_sysfs_init(mp); |
225e4635 BD |
705 | if (error) |
706 | goto out_del_stats; | |
707 | ||
192852be CM |
708 | |
709 | error = xfs_uuid_mount(mp); | |
710 | if (error) | |
711 | goto out_remove_error_sysfs; | |
712 | ||
0771fb45 ES |
713 | /* |
714 | * Set the minimum read and write sizes | |
715 | */ | |
716 | xfs_set_rw_sizes(mp); | |
717 | ||
055388a3 DC |
718 | /* set the low space thresholds for dynamic preallocation */ |
719 | xfs_set_low_space_thresholds(mp); | |
720 | ||
0771fb45 ES |
721 | /* |
722 | * Set the inode cluster size. | |
723 | * This may still be overridden by the file system | |
724 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
725 | * |
726 | * For v5 filesystems, scale the cluster size with the inode size to | |
727 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
728 | * has set the inode alignment value appropriately for larger cluster | |
729 | * sizes. | |
0771fb45 ES |
730 | */ |
731 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
732 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
733 | int new_size = mp->m_inode_cluster_size; | |
734 | ||
735 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
736 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
737 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 738 | } |
0771fb45 | 739 | |
e5376fc1 BF |
740 | /* |
741 | * If enabled, sparse inode chunk alignment is expected to match the | |
742 | * cluster size. Full inode chunk alignment must match the chunk size, | |
743 | * but that is checked on sb read verification... | |
744 | */ | |
745 | if (xfs_sb_version_hassparseinodes(&mp->m_sb) && | |
746 | mp->m_sb.sb_spino_align != | |
747 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) { | |
748 | xfs_warn(mp, | |
749 | "Sparse inode block alignment (%u) must match cluster size (%llu).", | |
750 | mp->m_sb.sb_spino_align, | |
751 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)); | |
752 | error = -EINVAL; | |
753 | goto out_remove_uuid; | |
754 | } | |
755 | ||
0771fb45 ES |
756 | /* |
757 | * Set inode alignment fields | |
758 | */ | |
759 | xfs_set_inoalignment(mp); | |
760 | ||
761 | /* | |
c2bfbc9b | 762 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 763 | */ |
4249023a | 764 | error = xfs_check_sizes(mp); |
0771fb45 | 765 | if (error) |
f9057e3d | 766 | goto out_remove_uuid; |
0771fb45 | 767 | |
1da177e4 LT |
768 | /* |
769 | * Initialize realtime fields in the mount structure | |
770 | */ | |
0771fb45 ES |
771 | error = xfs_rtmount_init(mp); |
772 | if (error) { | |
0b932ccc | 773 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 774 | goto out_remove_uuid; |
1da177e4 LT |
775 | } |
776 | ||
1da177e4 LT |
777 | /* |
778 | * Copies the low order bits of the timestamp and the randomly | |
779 | * set "sequence" number out of a UUID. | |
780 | */ | |
781 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
782 | ||
1da177e4 LT |
783 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
784 | ||
0650b554 DC |
785 | error = xfs_da_mount(mp); |
786 | if (error) { | |
787 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
788 | goto out_remove_uuid; | |
789 | } | |
1da177e4 LT |
790 | |
791 | /* | |
792 | * Initialize the precomputed transaction reservations values. | |
793 | */ | |
794 | xfs_trans_init(mp); | |
795 | ||
1da177e4 LT |
796 | /* |
797 | * Allocate and initialize the per-ag data. | |
798 | */ | |
1c1c6ebc | 799 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 800 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
801 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
802 | if (error) { | |
0b932ccc | 803 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 804 | goto out_free_dir; |
1c1c6ebc | 805 | } |
1da177e4 | 806 | |
f9057e3d | 807 | if (!sbp->sb_logblocks) { |
0b932ccc | 808 | xfs_warn(mp, "no log defined"); |
f9057e3d | 809 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 810 | error = -EFSCORRUPTED; |
f9057e3d CH |
811 | goto out_free_perag; |
812 | } | |
813 | ||
1da177e4 | 814 | /* |
f0b2efad BF |
815 | * Log's mount-time initialization. The first part of recovery can place |
816 | * some items on the AIL, to be handled when recovery is finished or | |
817 | * cancelled. | |
1da177e4 | 818 | */ |
f9057e3d CH |
819 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
820 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
821 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
822 | if (error) { | |
0b932ccc | 823 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 824 | goto out_fail_wait; |
1da177e4 LT |
825 | } |
826 | ||
92821e2b DC |
827 | /* |
828 | * Now the log is mounted, we know if it was an unclean shutdown or | |
829 | * not. If it was, with the first phase of recovery has completed, we | |
830 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
831 | * but they are recovered transactionally in the second recovery phase | |
832 | * later. | |
833 | * | |
834 | * Hence we can safely re-initialise incore superblock counters from | |
835 | * the per-ag data. These may not be correct if the filesystem was not | |
836 | * cleanly unmounted, so we need to wait for recovery to finish before | |
837 | * doing this. | |
838 | * | |
839 | * If the filesystem was cleanly unmounted, then we can trust the | |
840 | * values in the superblock to be correct and we don't need to do | |
841 | * anything here. | |
842 | * | |
843 | * If we are currently making the filesystem, the initialisation will | |
844 | * fail as the perag data is in an undefined state. | |
845 | */ | |
92821e2b DC |
846 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
847 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
848 | !mp->m_sb.sb_inprogress) { | |
849 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 850 | if (error) |
6eee8972 | 851 | goto out_log_dealloc; |
92821e2b | 852 | } |
f9057e3d | 853 | |
1da177e4 LT |
854 | /* |
855 | * Get and sanity-check the root inode. | |
856 | * Save the pointer to it in the mount structure. | |
857 | */ | |
7b6259e7 | 858 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 859 | if (error) { |
0b932ccc | 860 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 861 | goto out_log_dealloc; |
1da177e4 LT |
862 | } |
863 | ||
864 | ASSERT(rip != NULL); | |
1da177e4 | 865 | |
c19b3b05 | 866 | if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) { |
0b932ccc | 867 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 868 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
869 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
870 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
871 | mp); | |
2451337d | 872 | error = -EFSCORRUPTED; |
f9057e3d | 873 | goto out_rele_rip; |
1da177e4 LT |
874 | } |
875 | mp->m_rootip = rip; /* save it */ | |
876 | ||
877 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
878 | ||
879 | /* | |
880 | * Initialize realtime inode pointers in the mount structure | |
881 | */ | |
0771fb45 ES |
882 | error = xfs_rtmount_inodes(mp); |
883 | if (error) { | |
1da177e4 LT |
884 | /* |
885 | * Free up the root inode. | |
886 | */ | |
0b932ccc | 887 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 888 | goto out_rele_rip; |
1da177e4 LT |
889 | } |
890 | ||
891 | /* | |
7884bc86 CH |
892 | * If this is a read-only mount defer the superblock updates until |
893 | * the next remount into writeable mode. Otherwise we would never | |
894 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 895 | */ |
61e63ecb DC |
896 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
897 | error = xfs_sync_sb(mp, false); | |
e5720eec | 898 | if (error) { |
0b932ccc | 899 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 900 | goto out_rtunmount; |
e5720eec DC |
901 | } |
902 | } | |
1da177e4 LT |
903 | |
904 | /* | |
905 | * Initialise the XFS quota management subsystem for this mount | |
906 | */ | |
7d095257 CH |
907 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
908 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
909 | if (error) | |
910 | goto out_rtunmount; | |
911 | } else { | |
912 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
913 | ||
914 | /* | |
915 | * If a file system had quotas running earlier, but decided to | |
916 | * mount without -o uquota/pquota/gquota options, revoke the | |
917 | * quotachecked license. | |
918 | */ | |
919 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 920 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
921 | error = xfs_mount_reset_sbqflags(mp); |
922 | if (error) | |
a70a4fa5 | 923 | goto out_rtunmount; |
7d095257 CH |
924 | } |
925 | } | |
1da177e4 | 926 | |
17c12bcd DW |
927 | /* |
928 | * During the second phase of log recovery, we need iget and | |
929 | * iput to behave like they do for an active filesystem. | |
930 | * xfs_fs_drop_inode needs to be able to prevent the deletion | |
931 | * of inodes before we're done replaying log items on those | |
932 | * inodes. | |
933 | */ | |
934 | mp->m_super->s_flags |= MS_ACTIVE; | |
935 | ||
1da177e4 | 936 | /* |
f0b2efad BF |
937 | * Finish recovering the file system. This part needed to be delayed |
938 | * until after the root and real-time bitmap inodes were consistently | |
939 | * read in. | |
1da177e4 | 940 | */ |
4249023a | 941 | error = xfs_log_mount_finish(mp); |
1da177e4 | 942 | if (error) { |
0b932ccc | 943 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 944 | goto out_rtunmount; |
1da177e4 LT |
945 | } |
946 | ||
ddeb14f4 DC |
947 | /* |
948 | * Now the log is fully replayed, we can transition to full read-only | |
949 | * mode for read-only mounts. This will sync all the metadata and clean | |
950 | * the log so that the recovery we just performed does not have to be | |
951 | * replayed again on the next mount. | |
952 | * | |
953 | * We use the same quiesce mechanism as the rw->ro remount, as they are | |
954 | * semantically identical operations. | |
955 | */ | |
956 | if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) == | |
957 | XFS_MOUNT_RDONLY) { | |
958 | xfs_quiesce_attr(mp); | |
959 | } | |
960 | ||
1da177e4 LT |
961 | /* |
962 | * Complete the quota initialisation, post-log-replay component. | |
963 | */ | |
7d095257 CH |
964 | if (quotamount) { |
965 | ASSERT(mp->m_qflags == 0); | |
966 | mp->m_qflags = quotaflags; | |
967 | ||
968 | xfs_qm_mount_quotas(mp); | |
969 | } | |
970 | ||
84e1e99f DC |
971 | /* |
972 | * Now we are mounted, reserve a small amount of unused space for | |
973 | * privileged transactions. This is needed so that transaction | |
974 | * space required for critical operations can dip into this pool | |
975 | * when at ENOSPC. This is needed for operations like create with | |
976 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
977 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
978 | * |
979 | * This may drive us straight to ENOSPC on mount, but that implies | |
980 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 981 | */ |
d5db0f97 ES |
982 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
983 | resblks = xfs_default_resblks(mp); | |
984 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
985 | if (error) | |
0b932ccc DC |
986 | xfs_warn(mp, |
987 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 988 | } |
84e1e99f | 989 | |
1da177e4 LT |
990 | return 0; |
991 | ||
b93b6e43 CH |
992 | out_rtunmount: |
993 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 994 | out_rele_rip: |
43355099 | 995 | IRELE(rip); |
0ae120f8 BF |
996 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
997 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
f9057e3d | 998 | out_log_dealloc: |
e6b3bb78 | 999 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; |
f0b2efad | 1000 | xfs_log_mount_cancel(mp); |
d4f3512b DC |
1001 | out_fail_wait: |
1002 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
1003 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1004 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1005 | out_free_perag: |
ff4f038c | 1006 | xfs_free_perag(mp); |
0650b554 DC |
1007 | out_free_dir: |
1008 | xfs_da_unmount(mp); | |
f9057e3d | 1009 | out_remove_uuid: |
27174203 | 1010 | xfs_uuid_unmount(mp); |
192852be CM |
1011 | out_remove_error_sysfs: |
1012 | xfs_error_sysfs_del(mp); | |
225e4635 BD |
1013 | out_del_stats: |
1014 | xfs_sysfs_del(&mp->m_stats.xs_kobj); | |
a31b1d3d BF |
1015 | out_remove_sysfs: |
1016 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1017 | out: |
1da177e4 LT |
1018 | return error; |
1019 | } | |
1020 | ||
1021 | /* | |
1da177e4 LT |
1022 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1023 | * log and makes sure that incore structures are freed. | |
1024 | */ | |
41b5c2e7 CH |
1025 | void |
1026 | xfs_unmountfs( | |
1027 | struct xfs_mount *mp) | |
1da177e4 | 1028 | { |
41b5c2e7 CH |
1029 | __uint64_t resblks; |
1030 | int error; | |
1da177e4 | 1031 | |
579b62fa BF |
1032 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1033 | ||
7d095257 | 1034 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1035 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1036 | IRELE(mp->m_rootip); |
1037 | ||
641c56fb DC |
1038 | /* |
1039 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1040 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1041 | * the transaction is still sitting in a iclog. The stale inodes |
1042 | * on that buffer will have their flush locks held until the | |
1043 | * transaction hits the disk and the callbacks run. the inode | |
1044 | * flush takes the flush lock unconditionally and with nothing to | |
1045 | * push out the iclog we will never get that unlocked. hence we | |
1046 | * need to force the log first. | |
1047 | */ | |
a14a348b | 1048 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e | 1049 | |
e6b3bb78 CM |
1050 | /* |
1051 | * We now need to tell the world we are unmounting. This will allow | |
1052 | * us to detect that the filesystem is going away and we should error | |
1053 | * out anything that we have been retrying in the background. This will | |
1054 | * prevent neverending retries in AIL pushing from hanging the unmount. | |
1055 | */ | |
1056 | mp->m_flags |= XFS_MOUNT_UNMOUNTING; | |
1057 | ||
c854363e | 1058 | /* |
211e4d43 CH |
1059 | * Flush all pending changes from the AIL. |
1060 | */ | |
1061 | xfs_ail_push_all_sync(mp->m_ail); | |
1062 | ||
1063 | /* | |
1064 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1065 | * inodes and none should be pinned or locked, but use synchronous |
1066 | * reclaim just to be sure. We can stop background inode reclaim | |
1067 | * here as well if it is still running. | |
c854363e | 1068 | */ |
7e18530b | 1069 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1070 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1071 | |
7d095257 | 1072 | xfs_qm_unmount(mp); |
a357a121 | 1073 | |
84e1e99f DC |
1074 | /* |
1075 | * Unreserve any blocks we have so that when we unmount we don't account | |
1076 | * the reserved free space as used. This is really only necessary for | |
1077 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1078 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1079 | * |
1080 | * We don't bother correcting this elsewhere for lazy superblock | |
1081 | * counting because on mount of an unclean filesystem we reconstruct the | |
1082 | * correct counter value and this is irrelevant. | |
1083 | * | |
1084 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1085 | * we only every apply deltas to the superblock and hence the incore | |
1086 | * value does not matter.... | |
1087 | */ | |
1088 | resblks = 0; | |
714082bc DC |
1089 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1090 | if (error) | |
0b932ccc | 1091 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1092 | "Freespace may not be correct on next mount."); |
1093 | ||
adab0f67 | 1094 | error = xfs_log_sbcount(mp); |
e5720eec | 1095 | if (error) |
0b932ccc | 1096 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1097 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1098 | |
225e4635 | 1099 | |
21b699c8 | 1100 | xfs_log_unmount(mp); |
0650b554 | 1101 | xfs_da_unmount(mp); |
27174203 | 1102 | xfs_uuid_unmount(mp); |
1da177e4 | 1103 | |
1550d0b0 | 1104 | #if defined(DEBUG) |
0ce4cfd4 | 1105 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1106 | #endif |
ff4f038c | 1107 | xfs_free_perag(mp); |
a31b1d3d | 1108 | |
192852be | 1109 | xfs_error_sysfs_del(mp); |
225e4635 | 1110 | xfs_sysfs_del(&mp->m_stats.xs_kobj); |
a31b1d3d | 1111 | xfs_sysfs_del(&mp->m_kobj); |
1da177e4 LT |
1112 | } |
1113 | ||
91ee575f BF |
1114 | /* |
1115 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1116 | * freeze level for which modifications should not be allowed. This allows | |
1117 | * certain operations to proceed while the freeze sequence is in progress, if | |
1118 | * necessary. | |
1119 | */ | |
1120 | bool | |
1121 | xfs_fs_writable( | |
1122 | struct xfs_mount *mp, | |
1123 | int level) | |
92821e2b | 1124 | { |
91ee575f BF |
1125 | ASSERT(level > SB_UNFROZEN); |
1126 | if ((mp->m_super->s_writers.frozen >= level) || | |
1127 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1128 | return false; | |
1129 | ||
1130 | return true; | |
92821e2b DC |
1131 | } |
1132 | ||
1133 | /* | |
b2ce3974 AE |
1134 | * xfs_log_sbcount |
1135 | * | |
adab0f67 | 1136 | * Sync the superblock counters to disk. |
b2ce3974 | 1137 | * |
91ee575f BF |
1138 | * Note this code can be called during the process of freezing, so we use the |
1139 | * transaction allocator that does not block when the transaction subsystem is | |
1140 | * in its frozen state. | |
92821e2b DC |
1141 | */ |
1142 | int | |
adab0f67 | 1143 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1144 | { |
91ee575f BF |
1145 | /* allow this to proceed during the freeze sequence... */ |
1146 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1147 | return 0; |
1148 | ||
92821e2b DC |
1149 | /* |
1150 | * we don't need to do this if we are updating the superblock | |
1151 | * counters on every modification. | |
1152 | */ | |
1153 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1154 | return 0; | |
1155 | ||
61e63ecb | 1156 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1157 | } |
1158 | ||
8c1903d3 DC |
1159 | /* |
1160 | * Deltas for the inode count are +/-64, hence we use a large batch size | |
1161 | * of 128 so we don't need to take the counter lock on every update. | |
1162 | */ | |
1163 | #define XFS_ICOUNT_BATCH 128 | |
501ab323 DC |
1164 | int |
1165 | xfs_mod_icount( | |
1166 | struct xfs_mount *mp, | |
1167 | int64_t delta) | |
1168 | { | |
8c1903d3 DC |
1169 | __percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH); |
1170 | if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) { | |
501ab323 DC |
1171 | ASSERT(0); |
1172 | percpu_counter_add(&mp->m_icount, -delta); | |
1173 | return -EINVAL; | |
1174 | } | |
1175 | return 0; | |
1176 | } | |
1177 | ||
e88b64ea DC |
1178 | int |
1179 | xfs_mod_ifree( | |
1180 | struct xfs_mount *mp, | |
1181 | int64_t delta) | |
1182 | { | |
1183 | percpu_counter_add(&mp->m_ifree, delta); | |
1184 | if (percpu_counter_compare(&mp->m_ifree, 0) < 0) { | |
1185 | ASSERT(0); | |
1186 | percpu_counter_add(&mp->m_ifree, -delta); | |
1187 | return -EINVAL; | |
1188 | } | |
1189 | return 0; | |
1190 | } | |
0d485ada | 1191 | |
8c1903d3 DC |
1192 | /* |
1193 | * Deltas for the block count can vary from 1 to very large, but lock contention | |
1194 | * only occurs on frequent small block count updates such as in the delayed | |
1195 | * allocation path for buffered writes (page a time updates). Hence we set | |
1196 | * a large batch count (1024) to minimise global counter updates except when | |
1197 | * we get near to ENOSPC and we have to be very accurate with our updates. | |
1198 | */ | |
1199 | #define XFS_FDBLOCKS_BATCH 1024 | |
0d485ada DC |
1200 | int |
1201 | xfs_mod_fdblocks( | |
1202 | struct xfs_mount *mp, | |
1203 | int64_t delta, | |
1204 | bool rsvd) | |
1205 | { | |
1206 | int64_t lcounter; | |
1207 | long long res_used; | |
1208 | s32 batch; | |
1209 | ||
1210 | if (delta > 0) { | |
1211 | /* | |
1212 | * If the reserve pool is depleted, put blocks back into it | |
1213 | * first. Most of the time the pool is full. | |
1214 | */ | |
1215 | if (likely(mp->m_resblks == mp->m_resblks_avail)) { | |
1216 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1217 | return 0; | |
1218 | } | |
1219 | ||
1220 | spin_lock(&mp->m_sb_lock); | |
1221 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); | |
1222 | ||
1223 | if (res_used > delta) { | |
1224 | mp->m_resblks_avail += delta; | |
1225 | } else { | |
1226 | delta -= res_used; | |
1227 | mp->m_resblks_avail = mp->m_resblks; | |
1228 | percpu_counter_add(&mp->m_fdblocks, delta); | |
1229 | } | |
1230 | spin_unlock(&mp->m_sb_lock); | |
1231 | return 0; | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * Taking blocks away, need to be more accurate the closer we | |
1236 | * are to zero. | |
1237 | * | |
0d485ada DC |
1238 | * If the counter has a value of less than 2 * max batch size, |
1239 | * then make everything serialise as we are real close to | |
1240 | * ENOSPC. | |
1241 | */ | |
8c1903d3 DC |
1242 | if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH, |
1243 | XFS_FDBLOCKS_BATCH) < 0) | |
0d485ada DC |
1244 | batch = 1; |
1245 | else | |
8c1903d3 | 1246 | batch = XFS_FDBLOCKS_BATCH; |
0d485ada DC |
1247 | |
1248 | __percpu_counter_add(&mp->m_fdblocks, delta, batch); | |
52548852 | 1249 | if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside, |
8c1903d3 | 1250 | XFS_FDBLOCKS_BATCH) >= 0) { |
0d485ada DC |
1251 | /* we had space! */ |
1252 | return 0; | |
1253 | } | |
1254 | ||
1255 | /* | |
1256 | * lock up the sb for dipping into reserves before releasing the space | |
1257 | * that took us to ENOSPC. | |
1258 | */ | |
1259 | spin_lock(&mp->m_sb_lock); | |
1260 | percpu_counter_add(&mp->m_fdblocks, -delta); | |
1261 | if (!rsvd) | |
1262 | goto fdblocks_enospc; | |
1263 | ||
1264 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1265 | if (lcounter >= 0) { | |
1266 | mp->m_resblks_avail = lcounter; | |
1267 | spin_unlock(&mp->m_sb_lock); | |
1268 | return 0; | |
1269 | } | |
1270 | printk_once(KERN_WARNING | |
1271 | "Filesystem \"%s\": reserve blocks depleted! " | |
1272 | "Consider increasing reserve pool size.", | |
1273 | mp->m_fsname); | |
1274 | fdblocks_enospc: | |
1275 | spin_unlock(&mp->m_sb_lock); | |
1276 | return -ENOSPC; | |
1277 | } | |
1278 | ||
bab98bbe DC |
1279 | int |
1280 | xfs_mod_frextents( | |
1281 | struct xfs_mount *mp, | |
1282 | int64_t delta) | |
1283 | { | |
1284 | int64_t lcounter; | |
1285 | int ret = 0; | |
1286 | ||
1287 | spin_lock(&mp->m_sb_lock); | |
1288 | lcounter = mp->m_sb.sb_frextents + delta; | |
1289 | if (lcounter < 0) | |
1290 | ret = -ENOSPC; | |
1291 | else | |
1292 | mp->m_sb.sb_frextents = lcounter; | |
1293 | spin_unlock(&mp->m_sb_lock); | |
1294 | return ret; | |
1295 | } | |
1296 | ||
1da177e4 LT |
1297 | /* |
1298 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1299 | * The buffer is returned locked and read in from disk. | |
1300 | * The buffer should be released with a call to xfs_brelse(). | |
1301 | * | |
1302 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1303 | * the superblock buffer if it can be locked without sleeping. | |
1304 | * If it can't then we'll return NULL. | |
1305 | */ | |
0c842ad4 | 1306 | struct xfs_buf * |
1da177e4 | 1307 | xfs_getsb( |
0c842ad4 CH |
1308 | struct xfs_mount *mp, |
1309 | int flags) | |
1da177e4 | 1310 | { |
0c842ad4 | 1311 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1312 | |
0c842ad4 CH |
1313 | if (!xfs_buf_trylock(bp)) { |
1314 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1315 | return NULL; |
0c842ad4 | 1316 | xfs_buf_lock(bp); |
1da177e4 | 1317 | } |
0c842ad4 | 1318 | |
72790aa1 | 1319 | xfs_buf_hold(bp); |
b0388bf1 | 1320 | ASSERT(bp->b_flags & XBF_DONE); |
014c2544 | 1321 | return bp; |
1da177e4 LT |
1322 | } |
1323 | ||
1324 | /* | |
1325 | * Used to free the superblock along various error paths. | |
1326 | */ | |
1327 | void | |
1328 | xfs_freesb( | |
26af6552 | 1329 | struct xfs_mount *mp) |
1da177e4 | 1330 | { |
26af6552 | 1331 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1332 | |
26af6552 | 1333 | xfs_buf_lock(bp); |
1da177e4 | 1334 | mp->m_sb_bp = NULL; |
26af6552 | 1335 | xfs_buf_relse(bp); |
1da177e4 LT |
1336 | } |
1337 | ||
dda35b8f CH |
1338 | /* |
1339 | * If the underlying (data/log/rt) device is readonly, there are some | |
1340 | * operations that cannot proceed. | |
1341 | */ | |
1342 | int | |
1343 | xfs_dev_is_read_only( | |
1344 | struct xfs_mount *mp, | |
1345 | char *message) | |
1346 | { | |
1347 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1348 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1349 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1350 | xfs_notice(mp, "%s required on read-only device.", message); |
1351 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1352 | return -EROFS; |
dda35b8f CH |
1353 | } |
1354 | return 0; | |
1355 | } |