]>
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" |
a844f451 | 25 | #include "xfs_inum.h" |
1da177e4 LT |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
57062787 | 29 | #include "xfs_da_format.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" |
8f80587b | 44 | #include "xfs_dinode.h" |
a31b1d3d | 45 | #include "xfs_sysfs.h" |
0b1b213f | 46 | |
1da177e4 | 47 | |
8d280b98 | 48 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 49 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
50 | int); |
51 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
52 | int); | |
36fbe6e6 | 53 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
54 | #else |
55 | ||
45af6c6d CH |
56 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
57 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
58 | #endif |
59 | ||
27174203 CH |
60 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
61 | static int xfs_uuid_table_size; | |
62 | static uuid_t *xfs_uuid_table; | |
63 | ||
a31b1d3d BF |
64 | extern struct kset *xfs_kset; |
65 | ||
27174203 CH |
66 | /* |
67 | * See if the UUID is unique among mounted XFS filesystems. | |
68 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
69 | */ | |
70 | STATIC int | |
71 | xfs_uuid_mount( | |
72 | struct xfs_mount *mp) | |
73 | { | |
74 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
75 | int hole, i; | |
76 | ||
77 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
78 | return 0; | |
79 | ||
80 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 81 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 82 | return -EINVAL; |
27174203 CH |
83 | } |
84 | ||
85 | mutex_lock(&xfs_uuid_table_mutex); | |
86 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
87 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
88 | hole = i; | |
89 | continue; | |
90 | } | |
91 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
92 | goto out_duplicate; | |
93 | } | |
94 | ||
95 | if (hole < 0) { | |
96 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
97 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
98 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
99 | KM_SLEEP); | |
100 | hole = xfs_uuid_table_size++; | |
101 | } | |
102 | xfs_uuid_table[hole] = *uuid; | |
103 | mutex_unlock(&xfs_uuid_table_mutex); | |
104 | ||
105 | return 0; | |
106 | ||
107 | out_duplicate: | |
108 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 109 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
2451337d | 110 | return -EINVAL; |
27174203 CH |
111 | } |
112 | ||
113 | STATIC void | |
114 | xfs_uuid_unmount( | |
115 | struct xfs_mount *mp) | |
116 | { | |
117 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
118 | int i; | |
119 | ||
120 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
121 | return; | |
122 | ||
123 | mutex_lock(&xfs_uuid_table_mutex); | |
124 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
125 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
126 | continue; | |
127 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
128 | continue; | |
129 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
130 | break; | |
131 | } | |
132 | ASSERT(i < xfs_uuid_table_size); | |
133 | mutex_unlock(&xfs_uuid_table_mutex); | |
134 | } | |
135 | ||
136 | ||
e176579e DC |
137 | STATIC void |
138 | __xfs_free_perag( | |
139 | struct rcu_head *head) | |
140 | { | |
141 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
142 | ||
143 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
144 | kmem_free(pag); | |
145 | } | |
146 | ||
1da177e4 | 147 | /* |
e176579e | 148 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 149 | */ |
c962fb79 | 150 | STATIC void |
ff4f038c | 151 | xfs_free_perag( |
745f6919 | 152 | xfs_mount_t *mp) |
1da177e4 | 153 | { |
1c1c6ebc DC |
154 | xfs_agnumber_t agno; |
155 | struct xfs_perag *pag; | |
156 | ||
157 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
158 | spin_lock(&mp->m_perag_lock); | |
159 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
160 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 161 | ASSERT(pag); |
f83282a8 | 162 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 163 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 164 | } |
1da177e4 LT |
165 | } |
166 | ||
4cc929ee NS |
167 | /* |
168 | * Check size of device based on the (data/realtime) block count. | |
169 | * Note: this check is used by the growfs code as well as mount. | |
170 | */ | |
171 | int | |
172 | xfs_sb_validate_fsb_count( | |
173 | xfs_sb_t *sbp, | |
174 | __uint64_t nblocks) | |
175 | { | |
176 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
177 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
178 | ||
179 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
180 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
2451337d | 181 | return -EFBIG; |
4cc929ee NS |
182 | #else /* Limited by UINT_MAX of sectors */ |
183 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
2451337d | 184 | return -EFBIG; |
4cc929ee NS |
185 | #endif |
186 | return 0; | |
187 | } | |
1da177e4 | 188 | |
1c1c6ebc | 189 | int |
c11e2c36 | 190 | xfs_initialize_perag( |
c11e2c36 | 191 | xfs_mount_t *mp, |
1c1c6ebc DC |
192 | xfs_agnumber_t agcount, |
193 | xfs_agnumber_t *maxagi) | |
1da177e4 | 194 | { |
2d2194f6 | 195 | xfs_agnumber_t index; |
8b26c582 | 196 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
197 | xfs_perag_t *pag; |
198 | xfs_agino_t agino; | |
199 | xfs_ino_t ino; | |
200 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 201 | int error = -ENOMEM; |
1da177e4 | 202 | |
1c1c6ebc DC |
203 | /* |
204 | * Walk the current per-ag tree so we don't try to initialise AGs | |
205 | * that already exist (growfs case). Allocate and insert all the | |
206 | * AGs we don't find ready for initialisation. | |
207 | */ | |
208 | for (index = 0; index < agcount; index++) { | |
209 | pag = xfs_perag_get(mp, index); | |
210 | if (pag) { | |
211 | xfs_perag_put(pag); | |
212 | continue; | |
213 | } | |
8b26c582 DC |
214 | if (!first_initialised) |
215 | first_initialised = index; | |
fb3b504a | 216 | |
1c1c6ebc DC |
217 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
218 | if (!pag) | |
8b26c582 | 219 | goto out_unwind; |
fb3b504a CH |
220 | pag->pag_agno = index; |
221 | pag->pag_mount = mp; | |
1a427ab0 | 222 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 223 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 224 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
225 | spin_lock_init(&pag->pag_buf_lock); |
226 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 227 | |
1c1c6ebc | 228 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 229 | goto out_unwind; |
fb3b504a | 230 | |
1c1c6ebc DC |
231 | spin_lock(&mp->m_perag_lock); |
232 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
233 | BUG(); | |
234 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
235 | radix_tree_preload_end(); |
236 | error = -EEXIST; | |
237 | goto out_unwind; | |
1c1c6ebc DC |
238 | } |
239 | spin_unlock(&mp->m_perag_lock); | |
240 | radix_tree_preload_end(); | |
241 | } | |
242 | ||
fb3b504a CH |
243 | /* |
244 | * If we mount with the inode64 option, or no inode overflows | |
245 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 246 | */ |
fb3b504a CH |
247 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
248 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
249 | ||
250 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 251 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 252 | else |
1da177e4 | 253 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 254 | |
2d2194f6 CM |
255 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
256 | index = xfs_set_inode32(mp); | |
257 | else | |
258 | index = xfs_set_inode64(mp); | |
fb3b504a | 259 | |
1c1c6ebc DC |
260 | if (maxagi) |
261 | *maxagi = index; | |
262 | return 0; | |
8b26c582 DC |
263 | |
264 | out_unwind: | |
265 | kmem_free(pag); | |
266 | for (; index > first_initialised; index--) { | |
267 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
268 | kmem_free(pag); | |
269 | } | |
270 | return error; | |
1da177e4 LT |
271 | } |
272 | ||
1da177e4 LT |
273 | /* |
274 | * xfs_readsb | |
275 | * | |
276 | * Does the initial read of the superblock. | |
277 | */ | |
278 | int | |
ff55068c DC |
279 | xfs_readsb( |
280 | struct xfs_mount *mp, | |
281 | int flags) | |
1da177e4 LT |
282 | { |
283 | unsigned int sector_size; | |
04a1e6c5 DC |
284 | struct xfs_buf *bp; |
285 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 286 | int error; |
af34e09d | 287 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 288 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
289 | |
290 | ASSERT(mp->m_sb_bp == NULL); | |
291 | ASSERT(mp->m_ddev_targp != NULL); | |
292 | ||
daba5427 ES |
293 | /* |
294 | * For the initial read, we must guess at the sector | |
295 | * size based on the block device. It's enough to | |
296 | * get the sb_sectsize out of the superblock and | |
297 | * then reread with the proper length. | |
298 | * We don't verify it yet, because it may not be complete. | |
299 | */ | |
300 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
301 | buf_ops = NULL; | |
302 | ||
1da177e4 LT |
303 | /* |
304 | * Allocate a (locked) buffer to hold the superblock. | |
305 | * This will be kept around at all times to optimize | |
306 | * access to the superblock. | |
307 | */ | |
26af6552 | 308 | reread: |
e70b73f8 | 309 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
daba5427 | 310 | BTOBB(sector_size), 0, buf_ops); |
26af6552 | 311 | if (!bp) { |
af34e09d DC |
312 | if (loud) |
313 | xfs_warn(mp, "SB buffer read failed"); | |
2451337d | 314 | return -EIO; |
1da177e4 | 315 | } |
eab4e633 DC |
316 | if (bp->b_error) { |
317 | error = bp->b_error; | |
318 | if (loud) | |
e721f504 | 319 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 320 | /* bad CRC means corrupted metadata */ |
2451337d DC |
321 | if (error == -EFSBADCRC) |
322 | error = -EFSCORRUPTED; | |
eab4e633 DC |
323 | goto release_buf; |
324 | } | |
1da177e4 LT |
325 | |
326 | /* | |
327 | * Initialize the mount structure from the superblock. | |
1da177e4 | 328 | */ |
556b8883 DC |
329 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
330 | xfs_sb_quota_from_disk(sbp); | |
331 | ||
332 | /* | |
333 | * If we haven't validated the superblock, do so now before we try | |
334 | * to check the sector size and reread the superblock appropriately. | |
335 | */ | |
336 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
337 | if (loud) | |
338 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 339 | error = -EINVAL; |
556b8883 DC |
340 | goto release_buf; |
341 | } | |
ff55068c | 342 | |
1da177e4 LT |
343 | /* |
344 | * We must be able to do sector-sized and sector-aligned IO. | |
345 | */ | |
04a1e6c5 | 346 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
347 | if (loud) |
348 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 349 | sector_size, sbp->sb_sectsize); |
2451337d | 350 | error = -ENOSYS; |
26af6552 | 351 | goto release_buf; |
1da177e4 LT |
352 | } |
353 | ||
daba5427 | 354 | if (buf_ops == NULL) { |
556b8883 DC |
355 | /* |
356 | * Re-read the superblock so the buffer is correctly sized, | |
357 | * and properly verified. | |
358 | */ | |
1da177e4 | 359 | xfs_buf_relse(bp); |
04a1e6c5 | 360 | sector_size = sbp->sb_sectsize; |
daba5427 | 361 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 362 | goto reread; |
1da177e4 LT |
363 | } |
364 | ||
5478eead LM |
365 | /* Initialize per-cpu counters */ |
366 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 367 | |
04a1e6c5 DC |
368 | /* no need to be quiet anymore, so reset the buf ops */ |
369 | bp->b_ops = &xfs_sb_buf_ops; | |
370 | ||
1da177e4 | 371 | mp->m_sb_bp = bp; |
26af6552 | 372 | xfs_buf_unlock(bp); |
1da177e4 LT |
373 | return 0; |
374 | ||
26af6552 DC |
375 | release_buf: |
376 | xfs_buf_relse(bp); | |
1da177e4 LT |
377 | return error; |
378 | } | |
379 | ||
1da177e4 | 380 | /* |
0771fb45 | 381 | * Update alignment values based on mount options and sb values |
1da177e4 | 382 | */ |
0771fb45 | 383 | STATIC int |
7884bc86 | 384 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 385 | { |
1da177e4 | 386 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 387 | |
4249023a | 388 | if (mp->m_dalign) { |
1da177e4 LT |
389 | /* |
390 | * If stripe unit and stripe width are not multiples | |
391 | * of the fs blocksize turn off alignment. | |
392 | */ | |
393 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
394 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
395 | xfs_warn(mp, |
396 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
397 | sbp->sb_blocksize); | |
2451337d | 398 | return -EINVAL; |
1da177e4 LT |
399 | } else { |
400 | /* | |
401 | * Convert the stripe unit and width to FSBs. | |
402 | */ | |
403 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
404 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 405 | xfs_warn(mp, |
39a45d84 JL |
406 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
407 | sbp->sb_agblocks); | |
2451337d | 408 | return -EINVAL; |
1da177e4 LT |
409 | } else if (mp->m_dalign) { |
410 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
411 | } else { | |
39a45d84 JL |
412 | xfs_warn(mp, |
413 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
414 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 415 | return -EINVAL; |
1da177e4 LT |
416 | } |
417 | } | |
418 | ||
419 | /* | |
420 | * Update superblock with new values | |
421 | * and log changes | |
422 | */ | |
62118709 | 423 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
424 | if (sbp->sb_unit != mp->m_dalign) { |
425 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 426 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
427 | } |
428 | if (sbp->sb_width != mp->m_swidth) { | |
429 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 430 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 | 431 | } |
34d7f603 JL |
432 | } else { |
433 | xfs_warn(mp, | |
434 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 435 | return -EINVAL; |
1da177e4 LT |
436 | } |
437 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 438 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
439 | mp->m_dalign = sbp->sb_unit; |
440 | mp->m_swidth = sbp->sb_width; | |
441 | } | |
442 | ||
0771fb45 ES |
443 | return 0; |
444 | } | |
1da177e4 | 445 | |
0771fb45 ES |
446 | /* |
447 | * Set the maximum inode count for this filesystem | |
448 | */ | |
449 | STATIC void | |
450 | xfs_set_maxicount(xfs_mount_t *mp) | |
451 | { | |
452 | xfs_sb_t *sbp = &(mp->m_sb); | |
453 | __uint64_t icount; | |
1da177e4 | 454 | |
0771fb45 ES |
455 | if (sbp->sb_imax_pct) { |
456 | /* | |
457 | * Make sure the maximum inode count is a multiple | |
458 | * of the units we allocate inodes in. | |
1da177e4 | 459 | */ |
1da177e4 LT |
460 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
461 | do_div(icount, 100); | |
462 | do_div(icount, mp->m_ialloc_blks); | |
463 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
464 | sbp->sb_inopblog; | |
0771fb45 | 465 | } else { |
1da177e4 | 466 | mp->m_maxicount = 0; |
1da177e4 | 467 | } |
0771fb45 ES |
468 | } |
469 | ||
470 | /* | |
471 | * Set the default minimum read and write sizes unless | |
472 | * already specified in a mount option. | |
473 | * We use smaller I/O sizes when the file system | |
474 | * is being used for NFS service (wsync mount option). | |
475 | */ | |
476 | STATIC void | |
477 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
478 | { | |
479 | xfs_sb_t *sbp = &(mp->m_sb); | |
480 | int readio_log, writeio_log; | |
1da177e4 | 481 | |
1da177e4 LT |
482 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
483 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
484 | readio_log = XFS_WSYNC_READIO_LOG; | |
485 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
486 | } else { | |
487 | readio_log = XFS_READIO_LOG_LARGE; | |
488 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
489 | } | |
490 | } else { | |
491 | readio_log = mp->m_readio_log; | |
492 | writeio_log = mp->m_writeio_log; | |
493 | } | |
494 | ||
1da177e4 LT |
495 | if (sbp->sb_blocklog > readio_log) { |
496 | mp->m_readio_log = sbp->sb_blocklog; | |
497 | } else { | |
498 | mp->m_readio_log = readio_log; | |
499 | } | |
500 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
501 | if (sbp->sb_blocklog > writeio_log) { | |
502 | mp->m_writeio_log = sbp->sb_blocklog; | |
503 | } else { | |
504 | mp->m_writeio_log = writeio_log; | |
505 | } | |
506 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 507 | } |
1da177e4 | 508 | |
055388a3 DC |
509 | /* |
510 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
511 | */ | |
512 | void | |
513 | xfs_set_low_space_thresholds( | |
514 | struct xfs_mount *mp) | |
515 | { | |
516 | int i; | |
517 | ||
518 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
519 | __uint64_t space = mp->m_sb.sb_dblocks; | |
520 | ||
521 | do_div(space, 100); | |
522 | mp->m_low_space[i] = space * (i + 1); | |
523 | } | |
524 | } | |
525 | ||
526 | ||
0771fb45 ES |
527 | /* |
528 | * Set whether we're using inode alignment. | |
529 | */ | |
530 | STATIC void | |
531 | xfs_set_inoalignment(xfs_mount_t *mp) | |
532 | { | |
62118709 | 533 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
534 | mp->m_sb.sb_inoalignmt >= |
535 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
536 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
537 | else | |
538 | mp->m_inoalign_mask = 0; | |
539 | /* | |
540 | * If we are using stripe alignment, check whether | |
541 | * the stripe unit is a multiple of the inode alignment | |
542 | */ | |
543 | if (mp->m_dalign && mp->m_inoalign_mask && | |
544 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
545 | mp->m_sinoalign = mp->m_dalign; | |
546 | else | |
547 | mp->m_sinoalign = 0; | |
0771fb45 ES |
548 | } |
549 | ||
550 | /* | |
0471f62e | 551 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
552 | */ |
553 | STATIC int | |
4249023a | 554 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
555 | { |
556 | xfs_buf_t *bp; | |
557 | xfs_daddr_t d; | |
0771fb45 | 558 | |
1da177e4 LT |
559 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
560 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 561 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 562 | return -EFBIG; |
1da177e4 | 563 | } |
e70b73f8 | 564 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 565 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 566 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 567 | if (!bp) { |
0b932ccc | 568 | xfs_warn(mp, "last sector read failed"); |
2451337d | 569 | return -EIO; |
1da177e4 | 570 | } |
1922c949 | 571 | xfs_buf_relse(bp); |
1da177e4 | 572 | |
4249023a | 573 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
574 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
575 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 576 | xfs_warn(mp, "log size mismatch detected"); |
2451337d | 577 | return -EFBIG; |
1da177e4 | 578 | } |
e70b73f8 | 579 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 580 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 581 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 582 | if (!bp) { |
0b932ccc | 583 | xfs_warn(mp, "log device read failed"); |
2451337d | 584 | return -EIO; |
0771fb45 | 585 | } |
1922c949 | 586 | xfs_buf_relse(bp); |
0771fb45 ES |
587 | } |
588 | return 0; | |
589 | } | |
590 | ||
7d095257 CH |
591 | /* |
592 | * Clear the quotaflags in memory and in the superblock. | |
593 | */ | |
594 | int | |
595 | xfs_mount_reset_sbqflags( | |
596 | struct xfs_mount *mp) | |
597 | { | |
598 | int error; | |
599 | struct xfs_trans *tp; | |
600 | ||
601 | mp->m_qflags = 0; | |
602 | ||
603 | /* | |
604 | * It is OK to look at sb_qflags here in mount path, | |
605 | * without m_sb_lock. | |
606 | */ | |
607 | if (mp->m_sb.sb_qflags == 0) | |
608 | return 0; | |
609 | spin_lock(&mp->m_sb_lock); | |
610 | mp->m_sb.sb_qflags = 0; | |
611 | spin_unlock(&mp->m_sb_lock); | |
612 | ||
613 | /* | |
614 | * If the fs is readonly, let the incore superblock run | |
615 | * with quotas off but don't flush the update out to disk | |
616 | */ | |
617 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
618 | return 0; | |
619 | ||
7d095257 | 620 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
3d3c8b52 | 621 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0); |
7d095257 CH |
622 | if (error) { |
623 | xfs_trans_cancel(tp, 0); | |
53487786 | 624 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
625 | return error; |
626 | } | |
627 | ||
628 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
629 | return xfs_trans_commit(tp, 0); | |
630 | } | |
631 | ||
d5db0f97 ES |
632 | __uint64_t |
633 | xfs_default_resblks(xfs_mount_t *mp) | |
634 | { | |
635 | __uint64_t resblks; | |
636 | ||
637 | /* | |
8babd8a2 DC |
638 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
639 | * smaller. This is intended to cover concurrent allocation | |
640 | * transactions when we initially hit enospc. These each require a 4 | |
641 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
642 | * allocation reservations. | |
d5db0f97 ES |
643 | */ |
644 | resblks = mp->m_sb.sb_dblocks; | |
645 | do_div(resblks, 20); | |
8babd8a2 | 646 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
647 | return resblks; |
648 | } | |
649 | ||
0771fb45 | 650 | /* |
0771fb45 ES |
651 | * This function does the following on an initial mount of a file system: |
652 | * - reads the superblock from disk and init the mount struct | |
653 | * - if we're a 32-bit kernel, do a size check on the superblock | |
654 | * so we don't mount terabyte filesystems | |
655 | * - init mount struct realtime fields | |
656 | * - allocate inode hash table for fs | |
657 | * - init directory manager | |
658 | * - perform recovery and init the log manager | |
659 | */ | |
660 | int | |
661 | xfs_mountfs( | |
4249023a | 662 | xfs_mount_t *mp) |
0771fb45 ES |
663 | { |
664 | xfs_sb_t *sbp = &(mp->m_sb); | |
665 | xfs_inode_t *rip; | |
0771fb45 | 666 | __uint64_t resblks; |
7d095257 CH |
667 | uint quotamount = 0; |
668 | uint quotaflags = 0; | |
0771fb45 ES |
669 | int error = 0; |
670 | ||
ff55068c | 671 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 672 | |
ee1c0908 | 673 | /* |
e6957ea4 ES |
674 | * Check for a mismatched features2 values. Older kernels |
675 | * read & wrote into the wrong sb offset for sb_features2 | |
676 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
677 | * when sb_features2 was added, which made older superblock | |
678 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 679 | * |
e6957ea4 ES |
680 | * For backwards compatibility, we make both slots equal. |
681 | * | |
682 | * If we detect a mismatched field, we OR the set bits into the | |
683 | * existing features2 field in case it has already been modified; we | |
684 | * don't want to lose any features. We then update the bad location | |
685 | * with the ORed value so that older kernels will see any features2 | |
686 | * flags, and mark the two fields as needing updates once the | |
687 | * transaction subsystem is online. | |
ee1c0908 | 688 | */ |
e6957ea4 | 689 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 690 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 691 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 692 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 693 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
694 | |
695 | /* | |
696 | * Re-check for ATTR2 in case it was found in bad_features2 | |
697 | * slot. | |
698 | */ | |
7c12f296 TS |
699 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
700 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 701 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
702 | } |
703 | ||
704 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
705 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
706 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 707 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 708 | |
7c12f296 TS |
709 | /* update sb_versionnum for the clearing of the morebits */ |
710 | if (!sbp->sb_features2) | |
7884bc86 | 711 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
712 | } |
713 | ||
263997a6 DC |
714 | /* always use v2 inodes by default now */ |
715 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
716 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
717 | mp->m_update_flags |= XFS_SB_VERSIONNUM; | |
718 | } | |
719 | ||
0771fb45 ES |
720 | /* |
721 | * Check if sb_agblocks is aligned at stripe boundary | |
722 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
723 | * allocator alignment is within an ag, therefore ag has | |
724 | * to be aligned at stripe boundary. | |
725 | */ | |
7884bc86 | 726 | error = xfs_update_alignment(mp); |
0771fb45 | 727 | if (error) |
f9057e3d | 728 | goto out; |
0771fb45 ES |
729 | |
730 | xfs_alloc_compute_maxlevels(mp); | |
731 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
732 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
733 | xfs_ialloc_compute_maxlevels(mp); | |
734 | ||
735 | xfs_set_maxicount(mp); | |
736 | ||
a31b1d3d BF |
737 | mp->m_kobj.kobject.kset = xfs_kset; |
738 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); | |
27174203 CH |
739 | if (error) |
740 | goto out; | |
1da177e4 | 741 | |
a31b1d3d BF |
742 | error = xfs_uuid_mount(mp); |
743 | if (error) | |
744 | goto out_remove_sysfs; | |
745 | ||
0771fb45 ES |
746 | /* |
747 | * Set the minimum read and write sizes | |
748 | */ | |
749 | xfs_set_rw_sizes(mp); | |
750 | ||
055388a3 DC |
751 | /* set the low space thresholds for dynamic preallocation */ |
752 | xfs_set_low_space_thresholds(mp); | |
753 | ||
0771fb45 ES |
754 | /* |
755 | * Set the inode cluster size. | |
756 | * This may still be overridden by the file system | |
757 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
758 | * |
759 | * For v5 filesystems, scale the cluster size with the inode size to | |
760 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
761 | * has set the inode alignment value appropriately for larger cluster | |
762 | * sizes. | |
0771fb45 ES |
763 | */ |
764 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
765 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
766 | int new_size = mp->m_inode_cluster_size; | |
767 | ||
768 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
769 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
770 | mp->m_inode_cluster_size = new_size; | |
8f80587b | 771 | } |
0771fb45 ES |
772 | |
773 | /* | |
774 | * Set inode alignment fields | |
775 | */ | |
776 | xfs_set_inoalignment(mp); | |
777 | ||
778 | /* | |
c2bfbc9b | 779 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 780 | */ |
4249023a | 781 | error = xfs_check_sizes(mp); |
0771fb45 | 782 | if (error) |
f9057e3d | 783 | goto out_remove_uuid; |
0771fb45 | 784 | |
1da177e4 LT |
785 | /* |
786 | * Initialize realtime fields in the mount structure | |
787 | */ | |
0771fb45 ES |
788 | error = xfs_rtmount_init(mp); |
789 | if (error) { | |
0b932ccc | 790 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 791 | goto out_remove_uuid; |
1da177e4 LT |
792 | } |
793 | ||
1da177e4 LT |
794 | /* |
795 | * Copies the low order bits of the timestamp and the randomly | |
796 | * set "sequence" number out of a UUID. | |
797 | */ | |
798 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
799 | ||
1da177e4 LT |
800 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
801 | ||
0650b554 DC |
802 | error = xfs_da_mount(mp); |
803 | if (error) { | |
804 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
805 | goto out_remove_uuid; | |
806 | } | |
1da177e4 LT |
807 | |
808 | /* | |
809 | * Initialize the precomputed transaction reservations values. | |
810 | */ | |
811 | xfs_trans_init(mp); | |
812 | ||
1da177e4 LT |
813 | /* |
814 | * Allocate and initialize the per-ag data. | |
815 | */ | |
1c1c6ebc | 816 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 817 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
818 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
819 | if (error) { | |
0b932ccc | 820 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 821 | goto out_free_dir; |
1c1c6ebc | 822 | } |
1da177e4 | 823 | |
f9057e3d | 824 | if (!sbp->sb_logblocks) { |
0b932ccc | 825 | xfs_warn(mp, "no log defined"); |
f9057e3d | 826 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 827 | error = -EFSCORRUPTED; |
f9057e3d CH |
828 | goto out_free_perag; |
829 | } | |
830 | ||
1da177e4 LT |
831 | /* |
832 | * log's mount-time initialization. Perform 1st part recovery if needed | |
833 | */ | |
f9057e3d CH |
834 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
835 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
836 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
837 | if (error) { | |
0b932ccc | 838 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 839 | goto out_fail_wait; |
1da177e4 LT |
840 | } |
841 | ||
92821e2b DC |
842 | /* |
843 | * Now the log is mounted, we know if it was an unclean shutdown or | |
844 | * not. If it was, with the first phase of recovery has completed, we | |
845 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
846 | * but they are recovered transactionally in the second recovery phase | |
847 | * later. | |
848 | * | |
849 | * Hence we can safely re-initialise incore superblock counters from | |
850 | * the per-ag data. These may not be correct if the filesystem was not | |
851 | * cleanly unmounted, so we need to wait for recovery to finish before | |
852 | * doing this. | |
853 | * | |
854 | * If the filesystem was cleanly unmounted, then we can trust the | |
855 | * values in the superblock to be correct and we don't need to do | |
856 | * anything here. | |
857 | * | |
858 | * If we are currently making the filesystem, the initialisation will | |
859 | * fail as the perag data is in an undefined state. | |
860 | */ | |
92821e2b DC |
861 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
862 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
863 | !mp->m_sb.sb_inprogress) { | |
864 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 865 | if (error) |
a70a4fa5 | 866 | goto out_log_dealloc;; |
92821e2b | 867 | } |
f9057e3d | 868 | |
1da177e4 LT |
869 | /* |
870 | * Get and sanity-check the root inode. | |
871 | * Save the pointer to it in the mount structure. | |
872 | */ | |
7b6259e7 | 873 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 874 | if (error) { |
0b932ccc | 875 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 876 | goto out_log_dealloc; |
1da177e4 LT |
877 | } |
878 | ||
879 | ASSERT(rip != NULL); | |
1da177e4 | 880 | |
abbede1b | 881 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 882 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 883 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
884 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
885 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
886 | mp); | |
2451337d | 887 | error = -EFSCORRUPTED; |
f9057e3d | 888 | goto out_rele_rip; |
1da177e4 LT |
889 | } |
890 | mp->m_rootip = rip; /* save it */ | |
891 | ||
892 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
893 | ||
894 | /* | |
895 | * Initialize realtime inode pointers in the mount structure | |
896 | */ | |
0771fb45 ES |
897 | error = xfs_rtmount_inodes(mp); |
898 | if (error) { | |
1da177e4 LT |
899 | /* |
900 | * Free up the root inode. | |
901 | */ | |
0b932ccc | 902 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 903 | goto out_rele_rip; |
1da177e4 LT |
904 | } |
905 | ||
906 | /* | |
7884bc86 CH |
907 | * If this is a read-only mount defer the superblock updates until |
908 | * the next remount into writeable mode. Otherwise we would never | |
909 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 910 | */ |
7884bc86 CH |
911 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
912 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 913 | if (error) { |
0b932ccc | 914 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 915 | goto out_rtunmount; |
e5720eec DC |
916 | } |
917 | } | |
1da177e4 LT |
918 | |
919 | /* | |
920 | * Initialise the XFS quota management subsystem for this mount | |
921 | */ | |
7d095257 CH |
922 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
923 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
924 | if (error) | |
925 | goto out_rtunmount; | |
926 | } else { | |
927 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
928 | ||
929 | /* | |
930 | * If a file system had quotas running earlier, but decided to | |
931 | * mount without -o uquota/pquota/gquota options, revoke the | |
932 | * quotachecked license. | |
933 | */ | |
934 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 935 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
936 | error = xfs_mount_reset_sbqflags(mp); |
937 | if (error) | |
a70a4fa5 | 938 | goto out_rtunmount; |
7d095257 CH |
939 | } |
940 | } | |
1da177e4 LT |
941 | |
942 | /* | |
943 | * Finish recovering the file system. This part needed to be | |
944 | * delayed until after the root and real-time bitmap inodes | |
945 | * were consistently read in. | |
946 | */ | |
4249023a | 947 | error = xfs_log_mount_finish(mp); |
1da177e4 | 948 | if (error) { |
0b932ccc | 949 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 950 | goto out_rtunmount; |
1da177e4 LT |
951 | } |
952 | ||
953 | /* | |
954 | * Complete the quota initialisation, post-log-replay component. | |
955 | */ | |
7d095257 CH |
956 | if (quotamount) { |
957 | ASSERT(mp->m_qflags == 0); | |
958 | mp->m_qflags = quotaflags; | |
959 | ||
960 | xfs_qm_mount_quotas(mp); | |
961 | } | |
962 | ||
84e1e99f DC |
963 | /* |
964 | * Now we are mounted, reserve a small amount of unused space for | |
965 | * privileged transactions. This is needed so that transaction | |
966 | * space required for critical operations can dip into this pool | |
967 | * when at ENOSPC. This is needed for operations like create with | |
968 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
969 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
970 | * |
971 | * This may drive us straight to ENOSPC on mount, but that implies | |
972 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 973 | */ |
d5db0f97 ES |
974 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
975 | resblks = xfs_default_resblks(mp); | |
976 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
977 | if (error) | |
0b932ccc DC |
978 | xfs_warn(mp, |
979 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 980 | } |
84e1e99f | 981 | |
1da177e4 LT |
982 | return 0; |
983 | ||
b93b6e43 CH |
984 | out_rtunmount: |
985 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 986 | out_rele_rip: |
43355099 | 987 | IRELE(rip); |
f9057e3d | 988 | out_log_dealloc: |
21b699c8 | 989 | xfs_log_unmount(mp); |
d4f3512b DC |
990 | out_fail_wait: |
991 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
992 | xfs_wait_buftarg(mp->m_logdev_targp); | |
993 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 994 | out_free_perag: |
ff4f038c | 995 | xfs_free_perag(mp); |
0650b554 DC |
996 | out_free_dir: |
997 | xfs_da_unmount(mp); | |
f9057e3d | 998 | out_remove_uuid: |
27174203 | 999 | xfs_uuid_unmount(mp); |
a31b1d3d BF |
1000 | out_remove_sysfs: |
1001 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 1002 | out: |
1da177e4 LT |
1003 | return error; |
1004 | } | |
1005 | ||
1006 | /* | |
1da177e4 LT |
1007 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1008 | * log and makes sure that incore structures are freed. | |
1009 | */ | |
41b5c2e7 CH |
1010 | void |
1011 | xfs_unmountfs( | |
1012 | struct xfs_mount *mp) | |
1da177e4 | 1013 | { |
41b5c2e7 CH |
1014 | __uint64_t resblks; |
1015 | int error; | |
1da177e4 | 1016 | |
579b62fa BF |
1017 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1018 | ||
7d095257 | 1019 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1020 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1021 | IRELE(mp->m_rootip); |
1022 | ||
641c56fb DC |
1023 | /* |
1024 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1025 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1026 | * the transaction is still sitting in a iclog. The stale inodes |
1027 | * on that buffer will have their flush locks held until the | |
1028 | * transaction hits the disk and the callbacks run. the inode | |
1029 | * flush takes the flush lock unconditionally and with nothing to | |
1030 | * push out the iclog we will never get that unlocked. hence we | |
1031 | * need to force the log first. | |
1032 | */ | |
a14a348b | 1033 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1034 | |
1035 | /* | |
211e4d43 CH |
1036 | * Flush all pending changes from the AIL. |
1037 | */ | |
1038 | xfs_ail_push_all_sync(mp->m_ail); | |
1039 | ||
1040 | /* | |
1041 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1042 | * inodes and none should be pinned or locked, but use synchronous |
1043 | * reclaim just to be sure. We can stop background inode reclaim | |
1044 | * here as well if it is still running. | |
c854363e | 1045 | */ |
7e18530b | 1046 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1047 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1048 | |
7d095257 | 1049 | xfs_qm_unmount(mp); |
a357a121 | 1050 | |
84e1e99f DC |
1051 | /* |
1052 | * Unreserve any blocks we have so that when we unmount we don't account | |
1053 | * the reserved free space as used. This is really only necessary for | |
1054 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1055 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1056 | * |
1057 | * We don't bother correcting this elsewhere for lazy superblock | |
1058 | * counting because on mount of an unclean filesystem we reconstruct the | |
1059 | * correct counter value and this is irrelevant. | |
1060 | * | |
1061 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1062 | * we only every apply deltas to the superblock and hence the incore | |
1063 | * value does not matter.... | |
1064 | */ | |
1065 | resblks = 0; | |
714082bc DC |
1066 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1067 | if (error) | |
0b932ccc | 1068 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1069 | "Freespace may not be correct on next mount."); |
1070 | ||
adab0f67 | 1071 | error = xfs_log_sbcount(mp); |
e5720eec | 1072 | if (error) |
0b932ccc | 1073 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1074 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1075 | |
21b699c8 | 1076 | xfs_log_unmount(mp); |
0650b554 | 1077 | xfs_da_unmount(mp); |
27174203 | 1078 | xfs_uuid_unmount(mp); |
1da177e4 | 1079 | |
1550d0b0 | 1080 | #if defined(DEBUG) |
0ce4cfd4 | 1081 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1082 | #endif |
ff4f038c | 1083 | xfs_free_perag(mp); |
a31b1d3d BF |
1084 | |
1085 | xfs_sysfs_del(&mp->m_kobj); | |
1da177e4 LT |
1086 | } |
1087 | ||
92821e2b DC |
1088 | int |
1089 | xfs_fs_writable(xfs_mount_t *mp) | |
1090 | { | |
d9457dc0 | 1091 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1092 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1093 | } |
1094 | ||
1095 | /* | |
b2ce3974 AE |
1096 | * xfs_log_sbcount |
1097 | * | |
adab0f67 | 1098 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1099 | * |
1100 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1101 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1102 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1103 | */ |
1104 | int | |
adab0f67 | 1105 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1106 | { |
1107 | xfs_trans_t *tp; | |
1108 | int error; | |
1109 | ||
1110 | if (!xfs_fs_writable(mp)) | |
1111 | return 0; | |
1112 | ||
d4d90b57 | 1113 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1114 | |
1115 | /* | |
1116 | * we don't need to do this if we are updating the superblock | |
1117 | * counters on every modification. | |
1118 | */ | |
1119 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1120 | return 0; | |
1121 | ||
b2ce3974 | 1122 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
3d3c8b52 | 1123 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
92821e2b DC |
1124 | if (error) { |
1125 | xfs_trans_cancel(tp, 0); | |
1126 | return error; | |
1127 | } | |
1128 | ||
1129 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1130 | xfs_trans_set_sync(tp); |
e5720eec DC |
1131 | error = xfs_trans_commit(tp, 0); |
1132 | return error; | |
92821e2b DC |
1133 | } |
1134 | ||
1da177e4 | 1135 | /* |
99e738b7 | 1136 | * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply |
1da177e4 LT |
1137 | * a delta to a specified field in the in-core superblock. Simply |
1138 | * switch on the field indicated and apply the delta to that field. | |
1139 | * Fields are not allowed to dip below zero, so if the delta would | |
1140 | * do this do not apply it and return EINVAL. | |
1141 | * | |
3685c2a1 | 1142 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1143 | */ |
d96f8f89 | 1144 | STATIC int |
20f4ebf2 DC |
1145 | xfs_mod_incore_sb_unlocked( |
1146 | xfs_mount_t *mp, | |
1147 | xfs_sb_field_t field, | |
1148 | int64_t delta, | |
1149 | int rsvd) | |
1da177e4 LT |
1150 | { |
1151 | int scounter; /* short counter for 32 bit fields */ | |
1152 | long long lcounter; /* long counter for 64 bit fields */ | |
1153 | long long res_used, rem; | |
1154 | ||
1155 | /* | |
1156 | * With the in-core superblock spin lock held, switch | |
1157 | * on the indicated field. Apply the delta to the | |
1158 | * proper field. If the fields value would dip below | |
1159 | * 0, then do not apply the delta and return EINVAL. | |
1160 | */ | |
1161 | switch (field) { | |
1162 | case XFS_SBS_ICOUNT: | |
1163 | lcounter = (long long)mp->m_sb.sb_icount; | |
1164 | lcounter += delta; | |
1165 | if (lcounter < 0) { | |
1166 | ASSERT(0); | |
2451337d | 1167 | return -EINVAL; |
1da177e4 LT |
1168 | } |
1169 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1170 | return 0; |
1da177e4 LT |
1171 | case XFS_SBS_IFREE: |
1172 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1173 | lcounter += delta; | |
1174 | if (lcounter < 0) { | |
1175 | ASSERT(0); | |
2451337d | 1176 | return -EINVAL; |
1da177e4 LT |
1177 | } |
1178 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1179 | return 0; |
1da177e4 | 1180 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1181 | lcounter = (long long) |
1182 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1183 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1184 | ||
1185 | if (delta > 0) { /* Putting blocks back */ | |
1186 | if (res_used > delta) { | |
1187 | mp->m_resblks_avail += delta; | |
1188 | } else { | |
1189 | rem = delta - res_used; | |
1190 | mp->m_resblks_avail = mp->m_resblks; | |
1191 | lcounter += rem; | |
1192 | } | |
1193 | } else { /* Taking blocks away */ | |
1da177e4 | 1194 | lcounter += delta; |
8babd8a2 DC |
1195 | if (lcounter >= 0) { |
1196 | mp->m_sb.sb_fdblocks = lcounter + | |
1197 | XFS_ALLOC_SET_ASIDE(mp); | |
1198 | return 0; | |
1199 | } | |
1da177e4 | 1200 | |
8babd8a2 DC |
1201 | /* |
1202 | * We are out of blocks, use any available reserved | |
1203 | * blocks if were allowed to. | |
1204 | */ | |
1205 | if (!rsvd) | |
2451337d | 1206 | return -ENOSPC; |
1da177e4 | 1207 | |
8babd8a2 DC |
1208 | lcounter = (long long)mp->m_resblks_avail + delta; |
1209 | if (lcounter >= 0) { | |
1210 | mp->m_resblks_avail = lcounter; | |
1211 | return 0; | |
1da177e4 | 1212 | } |
8babd8a2 DC |
1213 | printk_once(KERN_WARNING |
1214 | "Filesystem \"%s\": reserve blocks depleted! " | |
1215 | "Consider increasing reserve pool size.", | |
1216 | mp->m_fsname); | |
2451337d | 1217 | return -ENOSPC; |
1da177e4 LT |
1218 | } |
1219 | ||
4be536de | 1220 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1221 | return 0; |
1da177e4 LT |
1222 | case XFS_SBS_FREXTENTS: |
1223 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1224 | lcounter += delta; | |
1225 | if (lcounter < 0) { | |
2451337d | 1226 | return -ENOSPC; |
1da177e4 LT |
1227 | } |
1228 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1229 | return 0; |
1da177e4 LT |
1230 | case XFS_SBS_DBLOCKS: |
1231 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1232 | lcounter += delta; | |
1233 | if (lcounter < 0) { | |
1234 | ASSERT(0); | |
2451337d | 1235 | return -EINVAL; |
1da177e4 LT |
1236 | } |
1237 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1238 | return 0; |
1da177e4 LT |
1239 | case XFS_SBS_AGCOUNT: |
1240 | scounter = mp->m_sb.sb_agcount; | |
1241 | scounter += delta; | |
1242 | if (scounter < 0) { | |
1243 | ASSERT(0); | |
2451337d | 1244 | return -EINVAL; |
1da177e4 LT |
1245 | } |
1246 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1247 | return 0; |
1da177e4 LT |
1248 | case XFS_SBS_IMAX_PCT: |
1249 | scounter = mp->m_sb.sb_imax_pct; | |
1250 | scounter += delta; | |
1251 | if (scounter < 0) { | |
1252 | ASSERT(0); | |
2451337d | 1253 | return -EINVAL; |
1da177e4 LT |
1254 | } |
1255 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1256 | return 0; |
1da177e4 LT |
1257 | case XFS_SBS_REXTSIZE: |
1258 | scounter = mp->m_sb.sb_rextsize; | |
1259 | scounter += delta; | |
1260 | if (scounter < 0) { | |
1261 | ASSERT(0); | |
2451337d | 1262 | return -EINVAL; |
1da177e4 LT |
1263 | } |
1264 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1265 | return 0; |
1da177e4 LT |
1266 | case XFS_SBS_RBMBLOCKS: |
1267 | scounter = mp->m_sb.sb_rbmblocks; | |
1268 | scounter += delta; | |
1269 | if (scounter < 0) { | |
1270 | ASSERT(0); | |
2451337d | 1271 | return -EINVAL; |
1da177e4 LT |
1272 | } |
1273 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1274 | return 0; |
1da177e4 LT |
1275 | case XFS_SBS_RBLOCKS: |
1276 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1277 | lcounter += delta; | |
1278 | if (lcounter < 0) { | |
1279 | ASSERT(0); | |
2451337d | 1280 | return -EINVAL; |
1da177e4 LT |
1281 | } |
1282 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1283 | return 0; |
1da177e4 LT |
1284 | case XFS_SBS_REXTENTS: |
1285 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1286 | lcounter += delta; | |
1287 | if (lcounter < 0) { | |
1288 | ASSERT(0); | |
2451337d | 1289 | return -EINVAL; |
1da177e4 LT |
1290 | } |
1291 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1292 | return 0; |
1da177e4 LT |
1293 | case XFS_SBS_REXTSLOG: |
1294 | scounter = mp->m_sb.sb_rextslog; | |
1295 | scounter += delta; | |
1296 | if (scounter < 0) { | |
1297 | ASSERT(0); | |
2451337d | 1298 | return -EINVAL; |
1da177e4 LT |
1299 | } |
1300 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1301 | return 0; |
1da177e4 LT |
1302 | default: |
1303 | ASSERT(0); | |
2451337d | 1304 | return -EINVAL; |
1da177e4 LT |
1305 | } |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1310 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1311 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1312 | * routine to do the work. |
1313 | */ | |
1314 | int | |
20f4ebf2 | 1315 | xfs_mod_incore_sb( |
96540c78 CH |
1316 | struct xfs_mount *mp, |
1317 | xfs_sb_field_t field, | |
1318 | int64_t delta, | |
1319 | int rsvd) | |
1da177e4 | 1320 | { |
96540c78 | 1321 | int status; |
1da177e4 | 1322 | |
8d280b98 | 1323 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1324 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1325 | #endif |
96540c78 CH |
1326 | spin_lock(&mp->m_sb_lock); |
1327 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1328 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1329 | |
014c2544 | 1330 | return status; |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | /* | |
1b040712 | 1334 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1335 | * |
1b040712 CH |
1336 | * The fields and changes to those fields are specified in the array of |
1337 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1338 | * will be applied or none of them will. If any modified field dips below 0, | |
1339 | * then all modifications will be backed out and EINVAL will be returned. | |
1340 | * | |
1341 | * Note that this function may not be used for the superblock values that | |
1342 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1343 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1344 | */ |
1345 | int | |
1b040712 CH |
1346 | xfs_mod_incore_sb_batch( |
1347 | struct xfs_mount *mp, | |
1348 | xfs_mod_sb_t *msb, | |
1349 | uint nmsb, | |
1350 | int rsvd) | |
1da177e4 | 1351 | { |
45c51b99 | 1352 | xfs_mod_sb_t *msbp; |
1b040712 | 1353 | int error = 0; |
1da177e4 LT |
1354 | |
1355 | /* | |
1b040712 CH |
1356 | * Loop through the array of mod structures and apply each individually. |
1357 | * If any fail, then back out all those which have already been applied. | |
1358 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1359 | * changes will be atomic. | |
1da177e4 | 1360 | */ |
3685c2a1 | 1361 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1362 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1363 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1364 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1365 | |
1b040712 CH |
1366 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1367 | msbp->msb_delta, rsvd); | |
1368 | if (error) | |
1369 | goto unwind; | |
1da177e4 | 1370 | } |
1b040712 CH |
1371 | spin_unlock(&mp->m_sb_lock); |
1372 | return 0; | |
1da177e4 | 1373 | |
1b040712 CH |
1374 | unwind: |
1375 | while (--msbp >= msb) { | |
1376 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1377 | -msbp->msb_delta, rsvd); | |
1378 | ASSERT(error == 0); | |
1da177e4 | 1379 | } |
3685c2a1 | 1380 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1381 | return error; |
1da177e4 LT |
1382 | } |
1383 | ||
1384 | /* | |
1385 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1386 | * The buffer is returned locked and read in from disk. | |
1387 | * The buffer should be released with a call to xfs_brelse(). | |
1388 | * | |
1389 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1390 | * the superblock buffer if it can be locked without sleeping. | |
1391 | * If it can't then we'll return NULL. | |
1392 | */ | |
0c842ad4 | 1393 | struct xfs_buf * |
1da177e4 | 1394 | xfs_getsb( |
0c842ad4 CH |
1395 | struct xfs_mount *mp, |
1396 | int flags) | |
1da177e4 | 1397 | { |
0c842ad4 | 1398 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1399 | |
0c842ad4 CH |
1400 | if (!xfs_buf_trylock(bp)) { |
1401 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1402 | return NULL; |
0c842ad4 | 1403 | xfs_buf_lock(bp); |
1da177e4 | 1404 | } |
0c842ad4 | 1405 | |
72790aa1 | 1406 | xfs_buf_hold(bp); |
1da177e4 | 1407 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 1408 | return bp; |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | /* | |
1412 | * Used to free the superblock along various error paths. | |
1413 | */ | |
1414 | void | |
1415 | xfs_freesb( | |
26af6552 | 1416 | struct xfs_mount *mp) |
1da177e4 | 1417 | { |
26af6552 | 1418 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1419 | |
26af6552 | 1420 | xfs_buf_lock(bp); |
1da177e4 | 1421 | mp->m_sb_bp = NULL; |
26af6552 | 1422 | xfs_buf_relse(bp); |
1da177e4 LT |
1423 | } |
1424 | ||
1da177e4 LT |
1425 | /* |
1426 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1427 | * be altered by the mount options, as well as any potential sb_features2 |
1428 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1429 | */ |
7884bc86 | 1430 | int |
ee1c0908 | 1431 | xfs_mount_log_sb( |
1da177e4 LT |
1432 | xfs_mount_t *mp, |
1433 | __int64_t fields) | |
1434 | { | |
1435 | xfs_trans_t *tp; | |
e5720eec | 1436 | int error; |
1da177e4 | 1437 | |
ee1c0908 | 1438 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1439 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1440 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1441 | |
1442 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
3d3c8b52 | 1443 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
e5720eec | 1444 | if (error) { |
1da177e4 | 1445 | xfs_trans_cancel(tp, 0); |
e5720eec | 1446 | return error; |
1da177e4 LT |
1447 | } |
1448 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1449 | error = xfs_trans_commit(tp, 0); |
1450 | return error; | |
1da177e4 | 1451 | } |
8d280b98 | 1452 | |
dda35b8f CH |
1453 | /* |
1454 | * If the underlying (data/log/rt) device is readonly, there are some | |
1455 | * operations that cannot proceed. | |
1456 | */ | |
1457 | int | |
1458 | xfs_dev_is_read_only( | |
1459 | struct xfs_mount *mp, | |
1460 | char *message) | |
1461 | { | |
1462 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1463 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1464 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1465 | xfs_notice(mp, "%s required on read-only device.", message); |
1466 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1467 | return -EROFS; |
dda35b8f CH |
1468 | } |
1469 | return 0; | |
1470 | } | |
8d280b98 DC |
1471 | |
1472 | #ifdef HAVE_PERCPU_SB | |
1473 | /* | |
1474 | * Per-cpu incore superblock counters | |
1475 | * | |
1476 | * Simple concept, difficult implementation | |
1477 | * | |
1478 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1479 | * counter for contended fields (e.g. free block count). | |
1480 | * | |
1481 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1482 | * hence needs to be accurately read when we are running low on space. Hence | |
1483 | * there is a method to enable and disable the per-cpu counters based on how | |
1484 | * much "stuff" is available in them. | |
1485 | * | |
1486 | * Basically, a counter is enabled if there is enough free resource to justify | |
1487 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1488 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1489 | * re-distribute the available resources. | |
1490 | * | |
1491 | * If, once we redistributed the available resources, we still get a failure, | |
1492 | * we disable the per-cpu counter and go through the slow path. | |
1493 | * | |
1494 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 1495 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
1496 | * the global superblock. We do this after disabling the counter to prevent |
1497 | * more threads from queueing up on the counter. | |
1498 | * | |
1499 | * Essentially, this means that we still need a lock in the fast path to enable | |
1500 | * synchronisation between the global counters and the per-cpu counters. This | |
1501 | * is not a problem because the lock will be local to a CPU almost all the time | |
1502 | * and have little contention except when we get to ENOSPC conditions. | |
1503 | * | |
1504 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1505 | * path while we do things like enabling and disabling counters and | |
1506 | * synchronising the counters. | |
1507 | * | |
1508 | * Locking rules: | |
1509 | * | |
3685c2a1 | 1510 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1511 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1512 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1513 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1514 | * 5. modifying global counters requires holding m_sb_lock |
1515 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1516 | * and _none_ of the per-cpu locks. |
1517 | * | |
1518 | * Disabled counters are only ever re-enabled by a balance operation | |
1519 | * that results in more free resources per CPU than a given threshold. | |
1520 | * To ensure counters don't remain disabled, they are rebalanced when | |
1521 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1522 | * is present to prevent thrashing). | |
e8234a68 DC |
1523 | */ |
1524 | ||
5a67e4c5 | 1525 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1526 | /* |
1527 | * hot-plug CPU notifier support. | |
8d280b98 | 1528 | * |
5a67e4c5 CS |
1529 | * We need a notifier per filesystem as we need to be able to identify |
1530 | * the filesystem to balance the counters out. This is achieved by | |
1531 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1532 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1533 | */ |
e8234a68 DC |
1534 | STATIC int |
1535 | xfs_icsb_cpu_notify( | |
1536 | struct notifier_block *nfb, | |
1537 | unsigned long action, | |
1538 | void *hcpu) | |
1539 | { | |
1540 | xfs_icsb_cnts_t *cntp; | |
1541 | xfs_mount_t *mp; | |
e8234a68 DC |
1542 | |
1543 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1544 | cntp = (xfs_icsb_cnts_t *) | |
1545 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1546 | switch (action) { | |
1547 | case CPU_UP_PREPARE: | |
8bb78442 | 1548 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1549 | /* Easy Case - initialize the area and locks, and |
1550 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1551 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1552 | break; |
1553 | case CPU_ONLINE: | |
8bb78442 | 1554 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1555 | xfs_icsb_lock(mp); |
45af6c6d CH |
1556 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1557 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1558 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1559 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1560 | break; |
1561 | case CPU_DEAD: | |
8bb78442 | 1562 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1563 | /* Disable all the counters, then fold the dead cpu's |
1564 | * count into the total on the global superblock and | |
1565 | * re-enable the counters. */ | |
03135cf7 | 1566 | xfs_icsb_lock(mp); |
3685c2a1 | 1567 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1568 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
1569 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1570 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1571 | ||
1572 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1573 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1574 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1575 | ||
01e1b69c | 1576 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1577 | |
45af6c6d CH |
1578 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
1579 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
1580 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1581 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1582 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1583 | break; |
1584 | } | |
1585 | ||
1586 | return NOTIFY_OK; | |
1587 | } | |
5a67e4c5 | 1588 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1589 | |
8d280b98 DC |
1590 | int |
1591 | xfs_icsb_init_counters( | |
1592 | xfs_mount_t *mp) | |
1593 | { | |
1594 | xfs_icsb_cnts_t *cntp; | |
1595 | int i; | |
1596 | ||
1597 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1598 | if (mp->m_sb_cnts == NULL) | |
1599 | return -ENOMEM; | |
1600 | ||
1601 | for_each_online_cpu(i) { | |
1602 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1603 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 1604 | } |
20b64285 DC |
1605 | |
1606 | mutex_init(&mp->m_icsb_mutex); | |
1607 | ||
8d280b98 DC |
1608 | /* |
1609 | * start with all counters disabled so that the | |
1610 | * initial balance kicks us off correctly | |
1611 | */ | |
1612 | mp->m_icsb_counters = -1; | |
46677e67 RW |
1613 | |
1614 | #ifdef CONFIG_HOTPLUG_CPU | |
1615 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; | |
1616 | mp->m_icsb_notifier.priority = 0; | |
1617 | register_hotcpu_notifier(&mp->m_icsb_notifier); | |
1618 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1619 | ||
8d280b98 DC |
1620 | return 0; |
1621 | } | |
1622 | ||
5478eead LM |
1623 | void |
1624 | xfs_icsb_reinit_counters( | |
1625 | xfs_mount_t *mp) | |
1626 | { | |
1627 | xfs_icsb_lock(mp); | |
1628 | /* | |
1629 | * start with all counters disabled so that the | |
1630 | * initial balance kicks us off correctly | |
1631 | */ | |
1632 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
1633 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1634 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1635 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
1636 | xfs_icsb_unlock(mp); |
1637 | } | |
1638 | ||
c962fb79 | 1639 | void |
8d280b98 DC |
1640 | xfs_icsb_destroy_counters( |
1641 | xfs_mount_t *mp) | |
1642 | { | |
e8234a68 | 1643 | if (mp->m_sb_cnts) { |
5a67e4c5 | 1644 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 1645 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 1646 | } |
03135cf7 | 1647 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
1648 | } |
1649 | ||
b8f82a4a | 1650 | STATIC void |
01e1b69c DC |
1651 | xfs_icsb_lock_cntr( |
1652 | xfs_icsb_cnts_t *icsbp) | |
1653 | { | |
1654 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1655 | ndelay(1000); | |
1656 | } | |
1657 | } | |
1658 | ||
b8f82a4a | 1659 | STATIC void |
01e1b69c DC |
1660 | xfs_icsb_unlock_cntr( |
1661 | xfs_icsb_cnts_t *icsbp) | |
1662 | { | |
1663 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1664 | } | |
1665 | ||
8d280b98 | 1666 | |
b8f82a4a | 1667 | STATIC void |
8d280b98 DC |
1668 | xfs_icsb_lock_all_counters( |
1669 | xfs_mount_t *mp) | |
1670 | { | |
1671 | xfs_icsb_cnts_t *cntp; | |
1672 | int i; | |
1673 | ||
1674 | for_each_online_cpu(i) { | |
1675 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1676 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1677 | } |
1678 | } | |
1679 | ||
b8f82a4a | 1680 | STATIC void |
8d280b98 DC |
1681 | xfs_icsb_unlock_all_counters( |
1682 | xfs_mount_t *mp) | |
1683 | { | |
1684 | xfs_icsb_cnts_t *cntp; | |
1685 | int i; | |
1686 | ||
1687 | for_each_online_cpu(i) { | |
1688 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1689 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1690 | } |
1691 | } | |
1692 | ||
1693 | STATIC void | |
1694 | xfs_icsb_count( | |
1695 | xfs_mount_t *mp, | |
1696 | xfs_icsb_cnts_t *cnt, | |
1697 | int flags) | |
1698 | { | |
1699 | xfs_icsb_cnts_t *cntp; | |
1700 | int i; | |
1701 | ||
1702 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1703 | ||
1704 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1705 | xfs_icsb_lock_all_counters(mp); | |
1706 | ||
1707 | for_each_online_cpu(i) { | |
1708 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
1709 | cnt->icsb_icount += cntp->icsb_icount; | |
1710 | cnt->icsb_ifree += cntp->icsb_ifree; | |
1711 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1712 | } | |
1713 | ||
1714 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1715 | xfs_icsb_unlock_all_counters(mp); | |
1716 | } | |
1717 | ||
1718 | STATIC int | |
1719 | xfs_icsb_counter_disabled( | |
1720 | xfs_mount_t *mp, | |
1721 | xfs_sb_field_t field) | |
1722 | { | |
1723 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1724 | return test_bit(field, &mp->m_icsb_counters); | |
1725 | } | |
1726 | ||
36fbe6e6 | 1727 | STATIC void |
8d280b98 DC |
1728 | xfs_icsb_disable_counter( |
1729 | xfs_mount_t *mp, | |
1730 | xfs_sb_field_t field) | |
1731 | { | |
1732 | xfs_icsb_cnts_t cnt; | |
1733 | ||
1734 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1735 | ||
20b64285 DC |
1736 | /* |
1737 | * If we are already disabled, then there is nothing to do | |
1738 | * here. We check before locking all the counters to avoid | |
1739 | * the expensive lock operation when being called in the | |
1740 | * slow path and the counter is already disabled. This is | |
1741 | * safe because the only time we set or clear this state is under | |
1742 | * the m_icsb_mutex. | |
1743 | */ | |
1744 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 1745 | return; |
20b64285 | 1746 | |
8d280b98 DC |
1747 | xfs_icsb_lock_all_counters(mp); |
1748 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1749 | /* drain back to superblock */ | |
1750 | ||
ce46193b | 1751 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
1752 | switch(field) { |
1753 | case XFS_SBS_ICOUNT: | |
1754 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1755 | break; | |
1756 | case XFS_SBS_IFREE: | |
1757 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1758 | break; | |
1759 | case XFS_SBS_FDBLOCKS: | |
1760 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1761 | break; | |
1762 | default: | |
1763 | BUG(); | |
1764 | } | |
1765 | } | |
1766 | ||
1767 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
1768 | } |
1769 | ||
1770 | STATIC void | |
1771 | xfs_icsb_enable_counter( | |
1772 | xfs_mount_t *mp, | |
1773 | xfs_sb_field_t field, | |
1774 | uint64_t count, | |
1775 | uint64_t resid) | |
1776 | { | |
1777 | xfs_icsb_cnts_t *cntp; | |
1778 | int i; | |
1779 | ||
1780 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1781 | ||
1782 | xfs_icsb_lock_all_counters(mp); | |
1783 | for_each_online_cpu(i) { | |
1784 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1785 | switch (field) { | |
1786 | case XFS_SBS_ICOUNT: | |
1787 | cntp->icsb_icount = count + resid; | |
1788 | break; | |
1789 | case XFS_SBS_IFREE: | |
1790 | cntp->icsb_ifree = count + resid; | |
1791 | break; | |
1792 | case XFS_SBS_FDBLOCKS: | |
1793 | cntp->icsb_fdblocks = count + resid; | |
1794 | break; | |
1795 | default: | |
1796 | BUG(); | |
1797 | break; | |
1798 | } | |
1799 | resid = 0; | |
1800 | } | |
1801 | clear_bit(field, &mp->m_icsb_counters); | |
1802 | xfs_icsb_unlock_all_counters(mp); | |
1803 | } | |
1804 | ||
dbcabad1 | 1805 | void |
d4d90b57 | 1806 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
1807 | xfs_mount_t *mp, |
1808 | int flags) | |
1809 | { | |
1810 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 1811 | |
8d280b98 DC |
1812 | xfs_icsb_count(mp, &cnt, flags); |
1813 | ||
8d280b98 DC |
1814 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
1815 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1816 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
1817 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1818 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1819 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
1820 | } |
1821 | ||
1822 | /* | |
1823 | * Accurate update of per-cpu counters to incore superblock | |
1824 | */ | |
d4d90b57 | 1825 | void |
8d280b98 | 1826 | xfs_icsb_sync_counters( |
d4d90b57 CH |
1827 | xfs_mount_t *mp, |
1828 | int flags) | |
8d280b98 | 1829 | { |
d4d90b57 CH |
1830 | spin_lock(&mp->m_sb_lock); |
1831 | xfs_icsb_sync_counters_locked(mp, flags); | |
1832 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1833 | } |
1834 | ||
1835 | /* | |
1836 | * Balance and enable/disable counters as necessary. | |
1837 | * | |
20b64285 DC |
1838 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
1839 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
1840 | * free blocks is something far enough zero that we aren't going thrash when we | |
1841 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
1842 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
1843 | * be distributed to a single CPU but each CPU has enough blocks to be | |
1844 | * reenabled. | |
1845 | * | |
1846 | * Note that we can be called when counters are already disabled. | |
1847 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
1848 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 1849 | */ |
20b64285 DC |
1850 | |
1851 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 1852 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 1853 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 1854 | STATIC void |
45af6c6d | 1855 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
1856 | xfs_mount_t *mp, |
1857 | xfs_sb_field_t field, | |
20b64285 | 1858 | int min_per_cpu) |
8d280b98 | 1859 | { |
6fdf8ccc | 1860 | uint64_t count, resid; |
8d280b98 | 1861 | int weight = num_online_cpus(); |
20b64285 | 1862 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 1863 | |
8d280b98 DC |
1864 | /* disable counter and sync counter */ |
1865 | xfs_icsb_disable_counter(mp, field); | |
1866 | ||
1867 | /* update counters - first CPU gets residual*/ | |
1868 | switch (field) { | |
1869 | case XFS_SBS_ICOUNT: | |
1870 | count = mp->m_sb.sb_icount; | |
1871 | resid = do_div(count, weight); | |
20b64285 | 1872 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1873 | return; |
8d280b98 DC |
1874 | break; |
1875 | case XFS_SBS_IFREE: | |
1876 | count = mp->m_sb.sb_ifree; | |
1877 | resid = do_div(count, weight); | |
20b64285 | 1878 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1879 | return; |
8d280b98 DC |
1880 | break; |
1881 | case XFS_SBS_FDBLOCKS: | |
1882 | count = mp->m_sb.sb_fdblocks; | |
1883 | resid = do_div(count, weight); | |
20b64285 | 1884 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 1885 | return; |
8d280b98 DC |
1886 | break; |
1887 | default: | |
1888 | BUG(); | |
6fdf8ccc | 1889 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
1890 | break; |
1891 | } | |
1892 | ||
1893 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
1894 | } |
1895 | ||
1896 | STATIC void | |
1897 | xfs_icsb_balance_counter( | |
1898 | xfs_mount_t *mp, | |
1899 | xfs_sb_field_t fields, | |
1900 | int min_per_cpu) | |
1901 | { | |
1902 | spin_lock(&mp->m_sb_lock); | |
1903 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
1904 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1905 | } |
1906 | ||
1b040712 | 1907 | int |
20b64285 | 1908 | xfs_icsb_modify_counters( |
8d280b98 DC |
1909 | xfs_mount_t *mp, |
1910 | xfs_sb_field_t field, | |
20f4ebf2 | 1911 | int64_t delta, |
20b64285 | 1912 | int rsvd) |
8d280b98 DC |
1913 | { |
1914 | xfs_icsb_cnts_t *icsbp; | |
1915 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 1916 | int ret = 0; |
8d280b98 | 1917 | |
20b64285 | 1918 | might_sleep(); |
8d280b98 | 1919 | again: |
7a9e02d6 CL |
1920 | preempt_disable(); |
1921 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
1922 | |
1923 | /* | |
1924 | * if the counter is disabled, go to slow path | |
1925 | */ | |
8d280b98 DC |
1926 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
1927 | goto slow_path; | |
20b64285 DC |
1928 | xfs_icsb_lock_cntr(icsbp); |
1929 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
1930 | xfs_icsb_unlock_cntr(icsbp); | |
1931 | goto slow_path; | |
1932 | } | |
8d280b98 DC |
1933 | |
1934 | switch (field) { | |
1935 | case XFS_SBS_ICOUNT: | |
1936 | lcounter = icsbp->icsb_icount; | |
1937 | lcounter += delta; | |
1938 | if (unlikely(lcounter < 0)) | |
20b64285 | 1939 | goto balance_counter; |
8d280b98 DC |
1940 | icsbp->icsb_icount = lcounter; |
1941 | break; | |
1942 | ||
1943 | case XFS_SBS_IFREE: | |
1944 | lcounter = icsbp->icsb_ifree; | |
1945 | lcounter += delta; | |
1946 | if (unlikely(lcounter < 0)) | |
20b64285 | 1947 | goto balance_counter; |
8d280b98 DC |
1948 | icsbp->icsb_ifree = lcounter; |
1949 | break; | |
1950 | ||
1951 | case XFS_SBS_FDBLOCKS: | |
1952 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
1953 | ||
4be536de | 1954 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1955 | lcounter += delta; |
1956 | if (unlikely(lcounter < 0)) | |
20b64285 | 1957 | goto balance_counter; |
4be536de | 1958 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1959 | break; |
1960 | default: | |
1961 | BUG(); | |
1962 | break; | |
1963 | } | |
01e1b69c | 1964 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 1965 | preempt_enable(); |
8d280b98 DC |
1966 | return 0; |
1967 | ||
8d280b98 | 1968 | slow_path: |
7a9e02d6 | 1969 | preempt_enable(); |
8d280b98 | 1970 | |
20b64285 DC |
1971 | /* |
1972 | * serialise with a mutex so we don't burn lots of cpu on | |
1973 | * the superblock lock. We still need to hold the superblock | |
1974 | * lock, however, when we modify the global structures. | |
1975 | */ | |
03135cf7 | 1976 | xfs_icsb_lock(mp); |
20b64285 DC |
1977 | |
1978 | /* | |
1979 | * Now running atomically. | |
1980 | * | |
1981 | * If the counter is enabled, someone has beaten us to rebalancing. | |
1982 | * Drop the lock and try again in the fast path.... | |
1983 | */ | |
1984 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 1985 | xfs_icsb_unlock(mp); |
8d280b98 | 1986 | goto again; |
8d280b98 DC |
1987 | } |
1988 | ||
20b64285 DC |
1989 | /* |
1990 | * The counter is currently disabled. Because we are | |
1991 | * running atomically here, we know a rebalance cannot | |
1992 | * be in progress. Hence we can go straight to operating | |
1993 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 1994 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 1995 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 1996 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
1997 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
1998 | * directly on the global counters. | |
1999 | */ | |
3685c2a1 | 2000 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2001 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2002 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2003 | |
20b64285 DC |
2004 | /* |
2005 | * Now that we've modified the global superblock, we | |
2006 | * may be able to re-enable the distributed counters | |
2007 | * (e.g. lots of space just got freed). After that | |
2008 | * we are done. | |
2009 | */ | |
2451337d | 2010 | if (ret != -ENOSPC) |
45af6c6d | 2011 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2012 | xfs_icsb_unlock(mp); |
8d280b98 | 2013 | return ret; |
8d280b98 | 2014 | |
20b64285 DC |
2015 | balance_counter: |
2016 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2017 | preempt_enable(); |
8d280b98 | 2018 | |
20b64285 DC |
2019 | /* |
2020 | * We may have multiple threads here if multiple per-cpu | |
2021 | * counters run dry at the same time. This will mean we can | |
2022 | * do more balances than strictly necessary but it is not | |
2023 | * the common slowpath case. | |
2024 | */ | |
03135cf7 | 2025 | xfs_icsb_lock(mp); |
20b64285 DC |
2026 | |
2027 | /* | |
2028 | * running atomically. | |
2029 | * | |
2030 | * This will leave the counter in the correct state for future | |
2031 | * accesses. After the rebalance, we simply try again and our retry | |
2032 | * will either succeed through the fast path or slow path without | |
2033 | * another balance operation being required. | |
2034 | */ | |
45af6c6d | 2035 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2036 | xfs_icsb_unlock(mp); |
20b64285 | 2037 | goto again; |
8d280b98 | 2038 | } |
20b64285 | 2039 | |
8d280b98 | 2040 | #endif |