]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
1da177e4 | 25 | #include "xfs_sb.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
57062787 | 27 | #include "xfs_da_format.h" |
9a2cc41c | 28 | #include "xfs_da_btree.h" |
1da177e4 | 29 | #include "xfs_inode.h" |
a4fbe6ab | 30 | #include "xfs_dir2.h" |
a844f451 | 31 | #include "xfs_ialloc.h" |
1da177e4 LT |
32 | #include "xfs_alloc.h" |
33 | #include "xfs_rtalloc.h" | |
34 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
35 | #include "xfs_trans.h" |
36 | #include "xfs_trans_priv.h" | |
37 | #include "xfs_log.h" | |
1da177e4 | 38 | #include "xfs_error.h" |
1da177e4 LT |
39 | #include "xfs_quota.h" |
40 | #include "xfs_fsops.h" | |
0b1b213f | 41 | #include "xfs_trace.h" |
6d8b79cf | 42 | #include "xfs_icache.h" |
a31b1d3d | 43 | #include "xfs_sysfs.h" |
0b1b213f | 44 | |
1da177e4 | 45 | |
8d280b98 | 46 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 47 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
48 | int); |
49 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
50 | int); | |
36fbe6e6 | 51 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
52 | #else |
53 | ||
45af6c6d CH |
54 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
55 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
56 | #endif |
57 | ||
27174203 CH |
58 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
59 | static int xfs_uuid_table_size; | |
60 | static uuid_t *xfs_uuid_table; | |
61 | ||
62 | /* | |
63 | * See if the UUID is unique among mounted XFS filesystems. | |
64 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
65 | */ | |
66 | STATIC int | |
67 | xfs_uuid_mount( | |
68 | struct xfs_mount *mp) | |
69 | { | |
70 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
71 | int hole, i; | |
72 | ||
73 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
74 | return 0; | |
75 | ||
76 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 77 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
2451337d | 78 | return -EINVAL; |
27174203 CH |
79 | } |
80 | ||
81 | mutex_lock(&xfs_uuid_table_mutex); | |
82 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
83 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
84 | hole = i; | |
85 | continue; | |
86 | } | |
87 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
88 | goto out_duplicate; | |
89 | } | |
90 | ||
91 | if (hole < 0) { | |
92 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
93 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
94 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
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 */ |
4cc929ee | 176 | if (nblocks >> (PAGE_CACHE_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 LT |
189 | xfs_perag_t *pag; |
190 | xfs_agino_t agino; | |
191 | xfs_ino_t ino; | |
192 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 193 | int error = -ENOMEM; |
1da177e4 | 194 | |
1c1c6ebc DC |
195 | /* |
196 | * Walk the current per-ag tree so we don't try to initialise AGs | |
197 | * that already exist (growfs case). Allocate and insert all the | |
198 | * AGs we don't find ready for initialisation. | |
199 | */ | |
200 | for (index = 0; index < agcount; index++) { | |
201 | pag = xfs_perag_get(mp, index); | |
202 | if (pag) { | |
203 | xfs_perag_put(pag); | |
204 | continue; | |
205 | } | |
8b26c582 DC |
206 | if (!first_initialised) |
207 | first_initialised = index; | |
fb3b504a | 208 | |
1c1c6ebc DC |
209 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
210 | if (!pag) | |
8b26c582 | 211 | goto out_unwind; |
fb3b504a CH |
212 | pag->pag_agno = index; |
213 | pag->pag_mount = mp; | |
1a427ab0 | 214 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 215 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 216 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
217 | spin_lock_init(&pag->pag_buf_lock); |
218 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 219 | |
1c1c6ebc | 220 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 221 | goto out_unwind; |
fb3b504a | 222 | |
1c1c6ebc DC |
223 | spin_lock(&mp->m_perag_lock); |
224 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
225 | BUG(); | |
226 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
227 | radix_tree_preload_end(); |
228 | error = -EEXIST; | |
229 | goto out_unwind; | |
1c1c6ebc DC |
230 | } |
231 | spin_unlock(&mp->m_perag_lock); | |
232 | radix_tree_preload_end(); | |
233 | } | |
234 | ||
fb3b504a CH |
235 | /* |
236 | * If we mount with the inode64 option, or no inode overflows | |
237 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 238 | */ |
fb3b504a CH |
239 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
240 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
241 | ||
242 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 243 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 244 | else |
1da177e4 | 245 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 246 | |
2d2194f6 | 247 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
9de67c3b | 248 | index = xfs_set_inode32(mp, agcount); |
2d2194f6 | 249 | else |
9de67c3b | 250 | index = xfs_set_inode64(mp, agcount); |
fb3b504a | 251 | |
1c1c6ebc DC |
252 | if (maxagi) |
253 | *maxagi = index; | |
254 | return 0; | |
8b26c582 DC |
255 | |
256 | out_unwind: | |
257 | kmem_free(pag); | |
258 | for (; index > first_initialised; index--) { | |
259 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
260 | kmem_free(pag); | |
261 | } | |
262 | return error; | |
1da177e4 LT |
263 | } |
264 | ||
1da177e4 LT |
265 | /* |
266 | * xfs_readsb | |
267 | * | |
268 | * Does the initial read of the superblock. | |
269 | */ | |
270 | int | |
ff55068c DC |
271 | xfs_readsb( |
272 | struct xfs_mount *mp, | |
273 | int flags) | |
1da177e4 LT |
274 | { |
275 | unsigned int sector_size; | |
04a1e6c5 DC |
276 | struct xfs_buf *bp; |
277 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 278 | int error; |
af34e09d | 279 | int loud = !(flags & XFS_MFSI_QUIET); |
daba5427 | 280 | const struct xfs_buf_ops *buf_ops; |
1da177e4 LT |
281 | |
282 | ASSERT(mp->m_sb_bp == NULL); | |
283 | ASSERT(mp->m_ddev_targp != NULL); | |
284 | ||
daba5427 ES |
285 | /* |
286 | * For the initial read, we must guess at the sector | |
287 | * size based on the block device. It's enough to | |
288 | * get the sb_sectsize out of the superblock and | |
289 | * then reread with the proper length. | |
290 | * We don't verify it yet, because it may not be complete. | |
291 | */ | |
292 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
293 | buf_ops = NULL; | |
294 | ||
1da177e4 LT |
295 | /* |
296 | * Allocate a (locked) buffer to hold the superblock. | |
297 | * This will be kept around at all times to optimize | |
298 | * access to the superblock. | |
299 | */ | |
26af6552 | 300 | reread: |
ba372674 DC |
301 | error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
302 | BTOBB(sector_size), 0, &bp, buf_ops); | |
303 | if (error) { | |
eab4e633 | 304 | if (loud) |
e721f504 | 305 | xfs_warn(mp, "SB validate failed with error %d.", error); |
ac75a1f7 | 306 | /* bad CRC means corrupted metadata */ |
2451337d DC |
307 | if (error == -EFSBADCRC) |
308 | error = -EFSCORRUPTED; | |
ba372674 | 309 | return error; |
eab4e633 | 310 | } |
1da177e4 LT |
311 | |
312 | /* | |
313 | * Initialize the mount structure from the superblock. | |
1da177e4 | 314 | */ |
556b8883 | 315 | xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp)); |
556b8883 DC |
316 | |
317 | /* | |
318 | * If we haven't validated the superblock, do so now before we try | |
319 | * to check the sector size and reread the superblock appropriately. | |
320 | */ | |
321 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
322 | if (loud) | |
323 | xfs_warn(mp, "Invalid superblock magic number"); | |
2451337d | 324 | error = -EINVAL; |
556b8883 DC |
325 | goto release_buf; |
326 | } | |
ff55068c | 327 | |
1da177e4 LT |
328 | /* |
329 | * We must be able to do sector-sized and sector-aligned IO. | |
330 | */ | |
04a1e6c5 | 331 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
332 | if (loud) |
333 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 334 | sector_size, sbp->sb_sectsize); |
2451337d | 335 | error = -ENOSYS; |
26af6552 | 336 | goto release_buf; |
1da177e4 LT |
337 | } |
338 | ||
daba5427 | 339 | if (buf_ops == NULL) { |
556b8883 DC |
340 | /* |
341 | * Re-read the superblock so the buffer is correctly sized, | |
342 | * and properly verified. | |
343 | */ | |
1da177e4 | 344 | xfs_buf_relse(bp); |
04a1e6c5 | 345 | sector_size = sbp->sb_sectsize; |
daba5427 | 346 | buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops; |
26af6552 | 347 | goto reread; |
1da177e4 LT |
348 | } |
349 | ||
5478eead LM |
350 | /* Initialize per-cpu counters */ |
351 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 352 | |
04a1e6c5 DC |
353 | /* no need to be quiet anymore, so reset the buf ops */ |
354 | bp->b_ops = &xfs_sb_buf_ops; | |
355 | ||
1da177e4 | 356 | mp->m_sb_bp = bp; |
26af6552 | 357 | xfs_buf_unlock(bp); |
1da177e4 LT |
358 | return 0; |
359 | ||
26af6552 DC |
360 | release_buf: |
361 | xfs_buf_relse(bp); | |
1da177e4 LT |
362 | return error; |
363 | } | |
364 | ||
1da177e4 | 365 | /* |
0771fb45 | 366 | * Update alignment values based on mount options and sb values |
1da177e4 | 367 | */ |
0771fb45 | 368 | STATIC int |
7884bc86 | 369 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 370 | { |
1da177e4 | 371 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 372 | |
4249023a | 373 | if (mp->m_dalign) { |
1da177e4 LT |
374 | /* |
375 | * If stripe unit and stripe width are not multiples | |
376 | * of the fs blocksize turn off alignment. | |
377 | */ | |
378 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
379 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
380 | xfs_warn(mp, |
381 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
382 | sbp->sb_blocksize); | |
2451337d | 383 | return -EINVAL; |
1da177e4 LT |
384 | } else { |
385 | /* | |
386 | * Convert the stripe unit and width to FSBs. | |
387 | */ | |
388 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
389 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 390 | xfs_warn(mp, |
39a45d84 JL |
391 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
392 | sbp->sb_agblocks); | |
2451337d | 393 | return -EINVAL; |
1da177e4 LT |
394 | } else if (mp->m_dalign) { |
395 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
396 | } else { | |
39a45d84 JL |
397 | xfs_warn(mp, |
398 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
399 | mp->m_dalign, sbp->sb_blocksize); | |
2451337d | 400 | return -EINVAL; |
1da177e4 LT |
401 | } |
402 | } | |
403 | ||
404 | /* | |
405 | * Update superblock with new values | |
406 | * and log changes | |
407 | */ | |
62118709 | 408 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
409 | if (sbp->sb_unit != mp->m_dalign) { |
410 | sbp->sb_unit = mp->m_dalign; | |
61e63ecb | 411 | mp->m_update_sb = true; |
1da177e4 LT |
412 | } |
413 | if (sbp->sb_width != mp->m_swidth) { | |
414 | sbp->sb_width = mp->m_swidth; | |
61e63ecb | 415 | mp->m_update_sb = true; |
1da177e4 | 416 | } |
34d7f603 JL |
417 | } else { |
418 | xfs_warn(mp, | |
419 | "cannot change alignment: superblock does not support data alignment"); | |
2451337d | 420 | return -EINVAL; |
1da177e4 LT |
421 | } |
422 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 423 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
424 | mp->m_dalign = sbp->sb_unit; |
425 | mp->m_swidth = sbp->sb_width; | |
426 | } | |
427 | ||
0771fb45 ES |
428 | return 0; |
429 | } | |
1da177e4 | 430 | |
0771fb45 ES |
431 | /* |
432 | * Set the maximum inode count for this filesystem | |
433 | */ | |
434 | STATIC void | |
435 | xfs_set_maxicount(xfs_mount_t *mp) | |
436 | { | |
437 | xfs_sb_t *sbp = &(mp->m_sb); | |
438 | __uint64_t icount; | |
1da177e4 | 439 | |
0771fb45 ES |
440 | if (sbp->sb_imax_pct) { |
441 | /* | |
442 | * Make sure the maximum inode count is a multiple | |
443 | * of the units we allocate inodes in. | |
1da177e4 | 444 | */ |
1da177e4 LT |
445 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
446 | do_div(icount, 100); | |
447 | do_div(icount, mp->m_ialloc_blks); | |
448 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
449 | sbp->sb_inopblog; | |
0771fb45 | 450 | } else { |
1da177e4 | 451 | mp->m_maxicount = 0; |
1da177e4 | 452 | } |
0771fb45 ES |
453 | } |
454 | ||
455 | /* | |
456 | * Set the default minimum read and write sizes unless | |
457 | * already specified in a mount option. | |
458 | * We use smaller I/O sizes when the file system | |
459 | * is being used for NFS service (wsync mount option). | |
460 | */ | |
461 | STATIC void | |
462 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
463 | { | |
464 | xfs_sb_t *sbp = &(mp->m_sb); | |
465 | int readio_log, writeio_log; | |
1da177e4 | 466 | |
1da177e4 LT |
467 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
468 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
469 | readio_log = XFS_WSYNC_READIO_LOG; | |
470 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
471 | } else { | |
472 | readio_log = XFS_READIO_LOG_LARGE; | |
473 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
474 | } | |
475 | } else { | |
476 | readio_log = mp->m_readio_log; | |
477 | writeio_log = mp->m_writeio_log; | |
478 | } | |
479 | ||
1da177e4 LT |
480 | if (sbp->sb_blocklog > readio_log) { |
481 | mp->m_readio_log = sbp->sb_blocklog; | |
482 | } else { | |
483 | mp->m_readio_log = readio_log; | |
484 | } | |
485 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
486 | if (sbp->sb_blocklog > writeio_log) { | |
487 | mp->m_writeio_log = sbp->sb_blocklog; | |
488 | } else { | |
489 | mp->m_writeio_log = writeio_log; | |
490 | } | |
491 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 492 | } |
1da177e4 | 493 | |
055388a3 DC |
494 | /* |
495 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
496 | */ | |
497 | void | |
498 | xfs_set_low_space_thresholds( | |
499 | struct xfs_mount *mp) | |
500 | { | |
501 | int i; | |
502 | ||
503 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
504 | __uint64_t space = mp->m_sb.sb_dblocks; | |
505 | ||
506 | do_div(space, 100); | |
507 | mp->m_low_space[i] = space * (i + 1); | |
508 | } | |
509 | } | |
510 | ||
511 | ||
0771fb45 ES |
512 | /* |
513 | * Set whether we're using inode alignment. | |
514 | */ | |
515 | STATIC void | |
516 | xfs_set_inoalignment(xfs_mount_t *mp) | |
517 | { | |
62118709 | 518 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
519 | mp->m_sb.sb_inoalignmt >= |
520 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
521 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
522 | else | |
523 | mp->m_inoalign_mask = 0; | |
524 | /* | |
525 | * If we are using stripe alignment, check whether | |
526 | * the stripe unit is a multiple of the inode alignment | |
527 | */ | |
528 | if (mp->m_dalign && mp->m_inoalign_mask && | |
529 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
530 | mp->m_sinoalign = mp->m_dalign; | |
531 | else | |
532 | mp->m_sinoalign = 0; | |
0771fb45 ES |
533 | } |
534 | ||
535 | /* | |
0471f62e | 536 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
537 | */ |
538 | STATIC int | |
ba372674 DC |
539 | xfs_check_sizes( |
540 | struct xfs_mount *mp) | |
0771fb45 | 541 | { |
ba372674 | 542 | struct xfs_buf *bp; |
0771fb45 | 543 | xfs_daddr_t d; |
ba372674 | 544 | int error; |
0771fb45 | 545 | |
1da177e4 LT |
546 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
547 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 548 | xfs_warn(mp, "filesystem size mismatch detected"); |
2451337d | 549 | return -EFBIG; |
1da177e4 | 550 | } |
ba372674 | 551 | error = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 552 | d - XFS_FSS_TO_BB(mp, 1), |
ba372674 DC |
553 | XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL); |
554 | if (error) { | |
0b932ccc | 555 | xfs_warn(mp, "last sector read failed"); |
ba372674 | 556 | return error; |
1da177e4 | 557 | } |
1922c949 | 558 | xfs_buf_relse(bp); |
1da177e4 | 559 | |
ba372674 DC |
560 | if (mp->m_logdev_targp == mp->m_ddev_targp) |
561 | return 0; | |
562 | ||
563 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); | |
564 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
565 | xfs_warn(mp, "log size mismatch detected"); | |
566 | return -EFBIG; | |
567 | } | |
568 | error = xfs_buf_read_uncached(mp->m_logdev_targp, | |
1922c949 | 569 | d - XFS_FSB_TO_BB(mp, 1), |
ba372674 DC |
570 | XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL); |
571 | if (error) { | |
572 | xfs_warn(mp, "log device read failed"); | |
573 | return error; | |
0771fb45 | 574 | } |
ba372674 | 575 | xfs_buf_relse(bp); |
0771fb45 ES |
576 | return 0; |
577 | } | |
578 | ||
7d095257 CH |
579 | /* |
580 | * Clear the quotaflags in memory and in the superblock. | |
581 | */ | |
582 | int | |
583 | xfs_mount_reset_sbqflags( | |
584 | struct xfs_mount *mp) | |
585 | { | |
7d095257 CH |
586 | mp->m_qflags = 0; |
587 | ||
61e63ecb | 588 | /* It is OK to look at sb_qflags in the mount path without m_sb_lock. */ |
7d095257 CH |
589 | if (mp->m_sb.sb_qflags == 0) |
590 | return 0; | |
591 | spin_lock(&mp->m_sb_lock); | |
592 | mp->m_sb.sb_qflags = 0; | |
593 | spin_unlock(&mp->m_sb_lock); | |
594 | ||
61e63ecb | 595 | if (!xfs_fs_writable(mp, SB_FREEZE_WRITE)) |
7d095257 CH |
596 | return 0; |
597 | ||
61e63ecb | 598 | return xfs_sync_sb(mp, false); |
7d095257 CH |
599 | } |
600 | ||
d5db0f97 ES |
601 | __uint64_t |
602 | xfs_default_resblks(xfs_mount_t *mp) | |
603 | { | |
604 | __uint64_t resblks; | |
605 | ||
606 | /* | |
8babd8a2 DC |
607 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
608 | * smaller. This is intended to cover concurrent allocation | |
609 | * transactions when we initially hit enospc. These each require a 4 | |
610 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
611 | * allocation reservations. | |
d5db0f97 ES |
612 | */ |
613 | resblks = mp->m_sb.sb_dblocks; | |
614 | do_div(resblks, 20); | |
8babd8a2 | 615 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
616 | return resblks; |
617 | } | |
618 | ||
0771fb45 | 619 | /* |
0771fb45 ES |
620 | * This function does the following on an initial mount of a file system: |
621 | * - reads the superblock from disk and init the mount struct | |
622 | * - if we're a 32-bit kernel, do a size check on the superblock | |
623 | * so we don't mount terabyte filesystems | |
624 | * - init mount struct realtime fields | |
625 | * - allocate inode hash table for fs | |
626 | * - init directory manager | |
627 | * - perform recovery and init the log manager | |
628 | */ | |
629 | int | |
630 | xfs_mountfs( | |
4249023a | 631 | xfs_mount_t *mp) |
0771fb45 ES |
632 | { |
633 | xfs_sb_t *sbp = &(mp->m_sb); | |
634 | xfs_inode_t *rip; | |
0771fb45 | 635 | __uint64_t resblks; |
7d095257 CH |
636 | uint quotamount = 0; |
637 | uint quotaflags = 0; | |
0771fb45 ES |
638 | int error = 0; |
639 | ||
ff55068c | 640 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 641 | |
ee1c0908 | 642 | /* |
074e427b DC |
643 | * Check for a mismatched features2 values. Older kernels read & wrote |
644 | * into the wrong sb offset for sb_features2 on some platforms due to | |
645 | * xfs_sb_t not being 64bit size aligned when sb_features2 was added, | |
646 | * which made older superblock reading/writing routines swap it as a | |
647 | * 64-bit value. | |
ee1c0908 | 648 | * |
e6957ea4 ES |
649 | * For backwards compatibility, we make both slots equal. |
650 | * | |
074e427b DC |
651 | * If we detect a mismatched field, we OR the set bits into the existing |
652 | * features2 field in case it has already been modified; we don't want | |
653 | * to lose any features. We then update the bad location with the ORed | |
654 | * value so that older kernels will see any features2 flags. The | |
655 | * superblock writeback code ensures the new sb_features2 is copied to | |
656 | * sb_bad_features2 before it is logged or written to disk. | |
ee1c0908 | 657 | */ |
e6957ea4 | 658 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 659 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 660 | sbp->sb_features2 |= sbp->sb_bad_features2; |
61e63ecb | 661 | mp->m_update_sb = true; |
e6957ea4 ES |
662 | |
663 | /* | |
664 | * Re-check for ATTR2 in case it was found in bad_features2 | |
665 | * slot. | |
666 | */ | |
7c12f296 TS |
667 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
668 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 669 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
670 | } |
671 | ||
672 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
673 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
674 | xfs_sb_version_removeattr2(&mp->m_sb); | |
61e63ecb | 675 | mp->m_update_sb = true; |
e6957ea4 | 676 | |
7c12f296 TS |
677 | /* update sb_versionnum for the clearing of the morebits */ |
678 | if (!sbp->sb_features2) | |
61e63ecb | 679 | mp->m_update_sb = true; |
ee1c0908 DC |
680 | } |
681 | ||
263997a6 DC |
682 | /* always use v2 inodes by default now */ |
683 | if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) { | |
684 | mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT; | |
61e63ecb | 685 | mp->m_update_sb = true; |
263997a6 DC |
686 | } |
687 | ||
0771fb45 ES |
688 | /* |
689 | * Check if sb_agblocks is aligned at stripe boundary | |
690 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
691 | * allocator alignment is within an ag, therefore ag has | |
692 | * to be aligned at stripe boundary. | |
693 | */ | |
7884bc86 | 694 | error = xfs_update_alignment(mp); |
0771fb45 | 695 | if (error) |
f9057e3d | 696 | goto out; |
0771fb45 ES |
697 | |
698 | xfs_alloc_compute_maxlevels(mp); | |
699 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
700 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
701 | xfs_ialloc_compute_maxlevels(mp); | |
702 | ||
703 | xfs_set_maxicount(mp); | |
704 | ||
a31b1d3d | 705 | error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname); |
27174203 CH |
706 | if (error) |
707 | goto out; | |
1da177e4 | 708 | |
a31b1d3d BF |
709 | error = xfs_uuid_mount(mp); |
710 | if (error) | |
711 | goto out_remove_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 ES |
739 | |
740 | /* | |
741 | * Set inode alignment fields | |
742 | */ | |
743 | xfs_set_inoalignment(mp); | |
744 | ||
745 | /* | |
c2bfbc9b | 746 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 747 | */ |
4249023a | 748 | error = xfs_check_sizes(mp); |
0771fb45 | 749 | if (error) |
f9057e3d | 750 | goto out_remove_uuid; |
0771fb45 | 751 | |
1da177e4 LT |
752 | /* |
753 | * Initialize realtime fields in the mount structure | |
754 | */ | |
0771fb45 ES |
755 | error = xfs_rtmount_init(mp); |
756 | if (error) { | |
0b932ccc | 757 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 758 | goto out_remove_uuid; |
1da177e4 LT |
759 | } |
760 | ||
1da177e4 LT |
761 | /* |
762 | * Copies the low order bits of the timestamp and the randomly | |
763 | * set "sequence" number out of a UUID. | |
764 | */ | |
765 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
766 | ||
1da177e4 LT |
767 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
768 | ||
0650b554 DC |
769 | error = xfs_da_mount(mp); |
770 | if (error) { | |
771 | xfs_warn(mp, "Failed dir/attr init: %d", error); | |
772 | goto out_remove_uuid; | |
773 | } | |
1da177e4 LT |
774 | |
775 | /* | |
776 | * Initialize the precomputed transaction reservations values. | |
777 | */ | |
778 | xfs_trans_init(mp); | |
779 | ||
1da177e4 LT |
780 | /* |
781 | * Allocate and initialize the per-ag data. | |
782 | */ | |
1c1c6ebc | 783 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 784 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
785 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
786 | if (error) { | |
0b932ccc | 787 | xfs_warn(mp, "Failed per-ag init: %d", error); |
0650b554 | 788 | goto out_free_dir; |
1c1c6ebc | 789 | } |
1da177e4 | 790 | |
f9057e3d | 791 | if (!sbp->sb_logblocks) { |
0b932ccc | 792 | xfs_warn(mp, "no log defined"); |
f9057e3d | 793 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
2451337d | 794 | error = -EFSCORRUPTED; |
f9057e3d CH |
795 | goto out_free_perag; |
796 | } | |
797 | ||
1da177e4 LT |
798 | /* |
799 | * log's mount-time initialization. Perform 1st part recovery if needed | |
800 | */ | |
f9057e3d CH |
801 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
802 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
803 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
804 | if (error) { | |
0b932ccc | 805 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 806 | goto out_fail_wait; |
1da177e4 LT |
807 | } |
808 | ||
92821e2b DC |
809 | /* |
810 | * Now the log is mounted, we know if it was an unclean shutdown or | |
811 | * not. If it was, with the first phase of recovery has completed, we | |
812 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
813 | * but they are recovered transactionally in the second recovery phase | |
814 | * later. | |
815 | * | |
816 | * Hence we can safely re-initialise incore superblock counters from | |
817 | * the per-ag data. These may not be correct if the filesystem was not | |
818 | * cleanly unmounted, so we need to wait for recovery to finish before | |
819 | * doing this. | |
820 | * | |
821 | * If the filesystem was cleanly unmounted, then we can trust the | |
822 | * values in the superblock to be correct and we don't need to do | |
823 | * anything here. | |
824 | * | |
825 | * If we are currently making the filesystem, the initialisation will | |
826 | * fail as the perag data is in an undefined state. | |
827 | */ | |
92821e2b DC |
828 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
829 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
830 | !mp->m_sb.sb_inprogress) { | |
831 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 832 | if (error) |
6eee8972 | 833 | goto out_log_dealloc; |
92821e2b | 834 | } |
f9057e3d | 835 | |
1da177e4 LT |
836 | /* |
837 | * Get and sanity-check the root inode. | |
838 | * Save the pointer to it in the mount structure. | |
839 | */ | |
7b6259e7 | 840 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 841 | if (error) { |
0b932ccc | 842 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 843 | goto out_log_dealloc; |
1da177e4 LT |
844 | } |
845 | ||
846 | ASSERT(rip != NULL); | |
1da177e4 | 847 | |
abbede1b | 848 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 849 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 850 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
851 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
852 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
853 | mp); | |
2451337d | 854 | error = -EFSCORRUPTED; |
f9057e3d | 855 | goto out_rele_rip; |
1da177e4 LT |
856 | } |
857 | mp->m_rootip = rip; /* save it */ | |
858 | ||
859 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
860 | ||
861 | /* | |
862 | * Initialize realtime inode pointers in the mount structure | |
863 | */ | |
0771fb45 ES |
864 | error = xfs_rtmount_inodes(mp); |
865 | if (error) { | |
1da177e4 LT |
866 | /* |
867 | * Free up the root inode. | |
868 | */ | |
0b932ccc | 869 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 870 | goto out_rele_rip; |
1da177e4 LT |
871 | } |
872 | ||
873 | /* | |
7884bc86 CH |
874 | * If this is a read-only mount defer the superblock updates until |
875 | * the next remount into writeable mode. Otherwise we would never | |
876 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 877 | */ |
61e63ecb DC |
878 | if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
879 | error = xfs_sync_sb(mp, false); | |
e5720eec | 880 | if (error) { |
0b932ccc | 881 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 882 | goto out_rtunmount; |
e5720eec DC |
883 | } |
884 | } | |
1da177e4 LT |
885 | |
886 | /* | |
887 | * Initialise the XFS quota management subsystem for this mount | |
888 | */ | |
7d095257 CH |
889 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
890 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
891 | if (error) | |
892 | goto out_rtunmount; | |
893 | } else { | |
894 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
895 | ||
896 | /* | |
897 | * If a file system had quotas running earlier, but decided to | |
898 | * mount without -o uquota/pquota/gquota options, revoke the | |
899 | * quotachecked license. | |
900 | */ | |
901 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 902 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
903 | error = xfs_mount_reset_sbqflags(mp); |
904 | if (error) | |
a70a4fa5 | 905 | goto out_rtunmount; |
7d095257 CH |
906 | } |
907 | } | |
1da177e4 LT |
908 | |
909 | /* | |
910 | * Finish recovering the file system. This part needed to be | |
911 | * delayed until after the root and real-time bitmap inodes | |
912 | * were consistently read in. | |
913 | */ | |
4249023a | 914 | error = xfs_log_mount_finish(mp); |
1da177e4 | 915 | if (error) { |
0b932ccc | 916 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 917 | goto out_rtunmount; |
1da177e4 LT |
918 | } |
919 | ||
920 | /* | |
921 | * Complete the quota initialisation, post-log-replay component. | |
922 | */ | |
7d095257 CH |
923 | if (quotamount) { |
924 | ASSERT(mp->m_qflags == 0); | |
925 | mp->m_qflags = quotaflags; | |
926 | ||
927 | xfs_qm_mount_quotas(mp); | |
928 | } | |
929 | ||
84e1e99f DC |
930 | /* |
931 | * Now we are mounted, reserve a small amount of unused space for | |
932 | * privileged transactions. This is needed so that transaction | |
933 | * space required for critical operations can dip into this pool | |
934 | * when at ENOSPC. This is needed for operations like create with | |
935 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
936 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
937 | * |
938 | * This may drive us straight to ENOSPC on mount, but that implies | |
939 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 940 | */ |
d5db0f97 ES |
941 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
942 | resblks = xfs_default_resblks(mp); | |
943 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
944 | if (error) | |
0b932ccc DC |
945 | xfs_warn(mp, |
946 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 947 | } |
84e1e99f | 948 | |
1da177e4 LT |
949 | return 0; |
950 | ||
b93b6e43 CH |
951 | out_rtunmount: |
952 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 953 | out_rele_rip: |
43355099 | 954 | IRELE(rip); |
f9057e3d | 955 | out_log_dealloc: |
21b699c8 | 956 | xfs_log_unmount(mp); |
d4f3512b DC |
957 | out_fail_wait: |
958 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
959 | xfs_wait_buftarg(mp->m_logdev_targp); | |
960 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 961 | out_free_perag: |
ff4f038c | 962 | xfs_free_perag(mp); |
0650b554 DC |
963 | out_free_dir: |
964 | xfs_da_unmount(mp); | |
f9057e3d | 965 | out_remove_uuid: |
27174203 | 966 | xfs_uuid_unmount(mp); |
a31b1d3d BF |
967 | out_remove_sysfs: |
968 | xfs_sysfs_del(&mp->m_kobj); | |
f9057e3d | 969 | out: |
1da177e4 LT |
970 | return error; |
971 | } | |
972 | ||
973 | /* | |
1da177e4 LT |
974 | * This flushes out the inodes,dquots and the superblock, unmounts the |
975 | * log and makes sure that incore structures are freed. | |
976 | */ | |
41b5c2e7 CH |
977 | void |
978 | xfs_unmountfs( | |
979 | struct xfs_mount *mp) | |
1da177e4 | 980 | { |
41b5c2e7 CH |
981 | __uint64_t resblks; |
982 | int error; | |
1da177e4 | 983 | |
579b62fa BF |
984 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
985 | ||
7d095257 | 986 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 987 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
988 | IRELE(mp->m_rootip); |
989 | ||
641c56fb DC |
990 | /* |
991 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 992 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
993 | * the transaction is still sitting in a iclog. The stale inodes |
994 | * on that buffer will have their flush locks held until the | |
995 | * transaction hits the disk and the callbacks run. the inode | |
996 | * flush takes the flush lock unconditionally and with nothing to | |
997 | * push out the iclog we will never get that unlocked. hence we | |
998 | * need to force the log first. | |
999 | */ | |
a14a348b | 1000 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1001 | |
1002 | /* | |
211e4d43 CH |
1003 | * Flush all pending changes from the AIL. |
1004 | */ | |
1005 | xfs_ail_push_all_sync(mp->m_ail); | |
1006 | ||
1007 | /* | |
1008 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1009 | * inodes and none should be pinned or locked, but use synchronous |
1010 | * reclaim just to be sure. We can stop background inode reclaim | |
1011 | * here as well if it is still running. | |
c854363e | 1012 | */ |
7e18530b | 1013 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1014 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1015 | |
7d095257 | 1016 | xfs_qm_unmount(mp); |
a357a121 | 1017 | |
84e1e99f DC |
1018 | /* |
1019 | * Unreserve any blocks we have so that when we unmount we don't account | |
1020 | * the reserved free space as used. This is really only necessary for | |
1021 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1022 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1023 | * |
1024 | * We don't bother correcting this elsewhere for lazy superblock | |
1025 | * counting because on mount of an unclean filesystem we reconstruct the | |
1026 | * correct counter value and this is irrelevant. | |
1027 | * | |
1028 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1029 | * we only every apply deltas to the superblock and hence the incore | |
1030 | * value does not matter.... | |
1031 | */ | |
1032 | resblks = 0; | |
714082bc DC |
1033 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1034 | if (error) | |
0b932ccc | 1035 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1036 | "Freespace may not be correct on next mount."); |
1037 | ||
adab0f67 | 1038 | error = xfs_log_sbcount(mp); |
e5720eec | 1039 | if (error) |
0b932ccc | 1040 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1041 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1042 | |
21b699c8 | 1043 | xfs_log_unmount(mp); |
0650b554 | 1044 | xfs_da_unmount(mp); |
27174203 | 1045 | xfs_uuid_unmount(mp); |
1da177e4 | 1046 | |
1550d0b0 | 1047 | #if defined(DEBUG) |
0ce4cfd4 | 1048 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1049 | #endif |
ff4f038c | 1050 | xfs_free_perag(mp); |
a31b1d3d BF |
1051 | |
1052 | xfs_sysfs_del(&mp->m_kobj); | |
1da177e4 LT |
1053 | } |
1054 | ||
91ee575f BF |
1055 | /* |
1056 | * Determine whether modifications can proceed. The caller specifies the minimum | |
1057 | * freeze level for which modifications should not be allowed. This allows | |
1058 | * certain operations to proceed while the freeze sequence is in progress, if | |
1059 | * necessary. | |
1060 | */ | |
1061 | bool | |
1062 | xfs_fs_writable( | |
1063 | struct xfs_mount *mp, | |
1064 | int level) | |
92821e2b | 1065 | { |
91ee575f BF |
1066 | ASSERT(level > SB_UNFROZEN); |
1067 | if ((mp->m_super->s_writers.frozen >= level) || | |
1068 | XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY)) | |
1069 | return false; | |
1070 | ||
1071 | return true; | |
92821e2b DC |
1072 | } |
1073 | ||
1074 | /* | |
b2ce3974 AE |
1075 | * xfs_log_sbcount |
1076 | * | |
adab0f67 | 1077 | * Sync the superblock counters to disk. |
b2ce3974 | 1078 | * |
91ee575f BF |
1079 | * Note this code can be called during the process of freezing, so we use the |
1080 | * transaction allocator that does not block when the transaction subsystem is | |
1081 | * in its frozen state. | |
92821e2b DC |
1082 | */ |
1083 | int | |
adab0f67 | 1084 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b | 1085 | { |
91ee575f BF |
1086 | /* allow this to proceed during the freeze sequence... */ |
1087 | if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) | |
92821e2b DC |
1088 | return 0; |
1089 | ||
d4d90b57 | 1090 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1091 | |
1092 | /* | |
1093 | * we don't need to do this if we are updating the superblock | |
1094 | * counters on every modification. | |
1095 | */ | |
1096 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1097 | return 0; | |
1098 | ||
61e63ecb | 1099 | return xfs_sync_sb(mp, true); |
92821e2b DC |
1100 | } |
1101 | ||
501ab323 DC |
1102 | int |
1103 | xfs_mod_icount( | |
1104 | struct xfs_mount *mp, | |
1105 | int64_t delta) | |
1106 | { | |
1107 | /* deltas are +/-64, hence the large batch size of 128. */ | |
1108 | __percpu_counter_add(&mp->m_icount, delta, 128); | |
1109 | if (percpu_counter_compare(&mp->m_icount, 0) < 0) { | |
1110 | ASSERT(0); | |
1111 | percpu_counter_add(&mp->m_icount, -delta); | |
1112 | return -EINVAL; | |
1113 | } | |
1114 | return 0; | |
1115 | } | |
1116 | ||
1da177e4 | 1117 | /* |
99e738b7 | 1118 | * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply |
1da177e4 LT |
1119 | * a delta to a specified field in the in-core superblock. Simply |
1120 | * switch on the field indicated and apply the delta to that field. | |
1121 | * Fields are not allowed to dip below zero, so if the delta would | |
1122 | * do this do not apply it and return EINVAL. | |
1123 | * | |
3685c2a1 | 1124 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1125 | */ |
d96f8f89 | 1126 | STATIC int |
20f4ebf2 DC |
1127 | xfs_mod_incore_sb_unlocked( |
1128 | xfs_mount_t *mp, | |
1129 | xfs_sb_field_t field, | |
1130 | int64_t delta, | |
1131 | int rsvd) | |
1da177e4 LT |
1132 | { |
1133 | int scounter; /* short counter for 32 bit fields */ | |
1134 | long long lcounter; /* long counter for 64 bit fields */ | |
1135 | long long res_used, rem; | |
1136 | ||
1137 | /* | |
1138 | * With the in-core superblock spin lock held, switch | |
1139 | * on the indicated field. Apply the delta to the | |
1140 | * proper field. If the fields value would dip below | |
1141 | * 0, then do not apply the delta and return EINVAL. | |
1142 | */ | |
1143 | switch (field) { | |
1144 | case XFS_SBS_ICOUNT: | |
501ab323 DC |
1145 | ASSERT(0); |
1146 | return -ENOSPC; | |
1da177e4 LT |
1147 | case XFS_SBS_IFREE: |
1148 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1149 | lcounter += delta; | |
1150 | if (lcounter < 0) { | |
1151 | ASSERT(0); | |
2451337d | 1152 | return -EINVAL; |
1da177e4 LT |
1153 | } |
1154 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1155 | return 0; |
1da177e4 | 1156 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1157 | lcounter = (long long) |
1158 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1159 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1160 | ||
1161 | if (delta > 0) { /* Putting blocks back */ | |
1162 | if (res_used > delta) { | |
1163 | mp->m_resblks_avail += delta; | |
1164 | } else { | |
1165 | rem = delta - res_used; | |
1166 | mp->m_resblks_avail = mp->m_resblks; | |
1167 | lcounter += rem; | |
1168 | } | |
1169 | } else { /* Taking blocks away */ | |
1da177e4 | 1170 | lcounter += delta; |
8babd8a2 DC |
1171 | if (lcounter >= 0) { |
1172 | mp->m_sb.sb_fdblocks = lcounter + | |
1173 | XFS_ALLOC_SET_ASIDE(mp); | |
1174 | return 0; | |
1175 | } | |
1da177e4 | 1176 | |
8babd8a2 DC |
1177 | /* |
1178 | * We are out of blocks, use any available reserved | |
1179 | * blocks if were allowed to. | |
1180 | */ | |
1181 | if (!rsvd) | |
2451337d | 1182 | return -ENOSPC; |
1da177e4 | 1183 | |
8babd8a2 DC |
1184 | lcounter = (long long)mp->m_resblks_avail + delta; |
1185 | if (lcounter >= 0) { | |
1186 | mp->m_resblks_avail = lcounter; | |
1187 | return 0; | |
1da177e4 | 1188 | } |
8babd8a2 DC |
1189 | printk_once(KERN_WARNING |
1190 | "Filesystem \"%s\": reserve blocks depleted! " | |
1191 | "Consider increasing reserve pool size.", | |
1192 | mp->m_fsname); | |
2451337d | 1193 | return -ENOSPC; |
1da177e4 LT |
1194 | } |
1195 | ||
4be536de | 1196 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1197 | return 0; |
1da177e4 LT |
1198 | case XFS_SBS_FREXTENTS: |
1199 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1200 | lcounter += delta; | |
1201 | if (lcounter < 0) { | |
2451337d | 1202 | return -ENOSPC; |
1da177e4 LT |
1203 | } |
1204 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1205 | return 0; |
1da177e4 LT |
1206 | case XFS_SBS_DBLOCKS: |
1207 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1208 | lcounter += delta; | |
1209 | if (lcounter < 0) { | |
1210 | ASSERT(0); | |
2451337d | 1211 | return -EINVAL; |
1da177e4 LT |
1212 | } |
1213 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1214 | return 0; |
1da177e4 LT |
1215 | case XFS_SBS_AGCOUNT: |
1216 | scounter = mp->m_sb.sb_agcount; | |
1217 | scounter += delta; | |
1218 | if (scounter < 0) { | |
1219 | ASSERT(0); | |
2451337d | 1220 | return -EINVAL; |
1da177e4 LT |
1221 | } |
1222 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1223 | return 0; |
1da177e4 LT |
1224 | case XFS_SBS_IMAX_PCT: |
1225 | scounter = mp->m_sb.sb_imax_pct; | |
1226 | scounter += delta; | |
1227 | if (scounter < 0) { | |
1228 | ASSERT(0); | |
2451337d | 1229 | return -EINVAL; |
1da177e4 LT |
1230 | } |
1231 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1232 | return 0; |
1da177e4 LT |
1233 | case XFS_SBS_REXTSIZE: |
1234 | scounter = mp->m_sb.sb_rextsize; | |
1235 | scounter += delta; | |
1236 | if (scounter < 0) { | |
1237 | ASSERT(0); | |
2451337d | 1238 | return -EINVAL; |
1da177e4 LT |
1239 | } |
1240 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1241 | return 0; |
1da177e4 LT |
1242 | case XFS_SBS_RBMBLOCKS: |
1243 | scounter = mp->m_sb.sb_rbmblocks; | |
1244 | scounter += delta; | |
1245 | if (scounter < 0) { | |
1246 | ASSERT(0); | |
2451337d | 1247 | return -EINVAL; |
1da177e4 LT |
1248 | } |
1249 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1250 | return 0; |
1da177e4 LT |
1251 | case XFS_SBS_RBLOCKS: |
1252 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1253 | lcounter += delta; | |
1254 | if (lcounter < 0) { | |
1255 | ASSERT(0); | |
2451337d | 1256 | return -EINVAL; |
1da177e4 LT |
1257 | } |
1258 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1259 | return 0; |
1da177e4 LT |
1260 | case XFS_SBS_REXTENTS: |
1261 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1262 | lcounter += delta; | |
1263 | if (lcounter < 0) { | |
1264 | ASSERT(0); | |
2451337d | 1265 | return -EINVAL; |
1da177e4 LT |
1266 | } |
1267 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1268 | return 0; |
1da177e4 LT |
1269 | case XFS_SBS_REXTSLOG: |
1270 | scounter = mp->m_sb.sb_rextslog; | |
1271 | scounter += delta; | |
1272 | if (scounter < 0) { | |
1273 | ASSERT(0); | |
2451337d | 1274 | return -EINVAL; |
1da177e4 LT |
1275 | } |
1276 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1277 | return 0; |
1da177e4 LT |
1278 | default: |
1279 | ASSERT(0); | |
2451337d | 1280 | return -EINVAL; |
1da177e4 LT |
1281 | } |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1286 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1287 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1288 | * routine to do the work. |
1289 | */ | |
1290 | int | |
20f4ebf2 | 1291 | xfs_mod_incore_sb( |
96540c78 CH |
1292 | struct xfs_mount *mp, |
1293 | xfs_sb_field_t field, | |
1294 | int64_t delta, | |
1295 | int rsvd) | |
1da177e4 | 1296 | { |
96540c78 | 1297 | int status; |
1da177e4 | 1298 | |
8d280b98 | 1299 | #ifdef HAVE_PERCPU_SB |
501ab323 | 1300 | ASSERT(field < XFS_SBS_IFREE || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1301 | #endif |
501ab323 | 1302 | |
96540c78 CH |
1303 | spin_lock(&mp->m_sb_lock); |
1304 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1305 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1306 | |
014c2544 | 1307 | return status; |
1da177e4 LT |
1308 | } |
1309 | ||
1310 | /* | |
1b040712 | 1311 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1312 | * |
1b040712 CH |
1313 | * The fields and changes to those fields are specified in the array of |
1314 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1315 | * will be applied or none of them will. If any modified field dips below 0, | |
1316 | * then all modifications will be backed out and EINVAL will be returned. | |
1317 | * | |
1318 | * Note that this function may not be used for the superblock values that | |
1319 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1320 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1321 | */ |
1322 | int | |
1b040712 CH |
1323 | xfs_mod_incore_sb_batch( |
1324 | struct xfs_mount *mp, | |
1325 | xfs_mod_sb_t *msb, | |
1326 | uint nmsb, | |
1327 | int rsvd) | |
1da177e4 | 1328 | { |
45c51b99 | 1329 | xfs_mod_sb_t *msbp; |
1b040712 | 1330 | int error = 0; |
1da177e4 LT |
1331 | |
1332 | /* | |
1b040712 CH |
1333 | * Loop through the array of mod structures and apply each individually. |
1334 | * If any fail, then back out all those which have already been applied. | |
1335 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1336 | * changes will be atomic. | |
1da177e4 | 1337 | */ |
3685c2a1 | 1338 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1339 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1340 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1341 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1342 | |
1b040712 CH |
1343 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1344 | msbp->msb_delta, rsvd); | |
1345 | if (error) | |
1346 | goto unwind; | |
1da177e4 | 1347 | } |
1b040712 CH |
1348 | spin_unlock(&mp->m_sb_lock); |
1349 | return 0; | |
1da177e4 | 1350 | |
1b040712 CH |
1351 | unwind: |
1352 | while (--msbp >= msb) { | |
1353 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1354 | -msbp->msb_delta, rsvd); | |
1355 | ASSERT(error == 0); | |
1da177e4 | 1356 | } |
3685c2a1 | 1357 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1358 | return error; |
1da177e4 LT |
1359 | } |
1360 | ||
1361 | /* | |
1362 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1363 | * The buffer is returned locked and read in from disk. | |
1364 | * The buffer should be released with a call to xfs_brelse(). | |
1365 | * | |
1366 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1367 | * the superblock buffer if it can be locked without sleeping. | |
1368 | * If it can't then we'll return NULL. | |
1369 | */ | |
0c842ad4 | 1370 | struct xfs_buf * |
1da177e4 | 1371 | xfs_getsb( |
0c842ad4 CH |
1372 | struct xfs_mount *mp, |
1373 | int flags) | |
1da177e4 | 1374 | { |
0c842ad4 | 1375 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1376 | |
0c842ad4 CH |
1377 | if (!xfs_buf_trylock(bp)) { |
1378 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1379 | return NULL; |
0c842ad4 | 1380 | xfs_buf_lock(bp); |
1da177e4 | 1381 | } |
0c842ad4 | 1382 | |
72790aa1 | 1383 | xfs_buf_hold(bp); |
1da177e4 | 1384 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 1385 | return bp; |
1da177e4 LT |
1386 | } |
1387 | ||
1388 | /* | |
1389 | * Used to free the superblock along various error paths. | |
1390 | */ | |
1391 | void | |
1392 | xfs_freesb( | |
26af6552 | 1393 | struct xfs_mount *mp) |
1da177e4 | 1394 | { |
26af6552 | 1395 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1396 | |
26af6552 | 1397 | xfs_buf_lock(bp); |
1da177e4 | 1398 | mp->m_sb_bp = NULL; |
26af6552 | 1399 | xfs_buf_relse(bp); |
1da177e4 LT |
1400 | } |
1401 | ||
dda35b8f CH |
1402 | /* |
1403 | * If the underlying (data/log/rt) device is readonly, there are some | |
1404 | * operations that cannot proceed. | |
1405 | */ | |
1406 | int | |
1407 | xfs_dev_is_read_only( | |
1408 | struct xfs_mount *mp, | |
1409 | char *message) | |
1410 | { | |
1411 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1412 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1413 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1414 | xfs_notice(mp, "%s required on read-only device.", message); |
1415 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
2451337d | 1416 | return -EROFS; |
dda35b8f CH |
1417 | } |
1418 | return 0; | |
1419 | } | |
8d280b98 DC |
1420 | |
1421 | #ifdef HAVE_PERCPU_SB | |
1422 | /* | |
1423 | * Per-cpu incore superblock counters | |
1424 | * | |
1425 | * Simple concept, difficult implementation | |
1426 | * | |
1427 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1428 | * counter for contended fields (e.g. free block count). | |
1429 | * | |
1430 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1431 | * hence needs to be accurately read when we are running low on space. Hence | |
1432 | * there is a method to enable and disable the per-cpu counters based on how | |
1433 | * much "stuff" is available in them. | |
1434 | * | |
1435 | * Basically, a counter is enabled if there is enough free resource to justify | |
1436 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1437 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1438 | * re-distribute the available resources. | |
1439 | * | |
1440 | * If, once we redistributed the available resources, we still get a failure, | |
1441 | * we disable the per-cpu counter and go through the slow path. | |
1442 | * | |
1443 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 1444 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
1445 | * the global superblock. We do this after disabling the counter to prevent |
1446 | * more threads from queueing up on the counter. | |
1447 | * | |
1448 | * Essentially, this means that we still need a lock in the fast path to enable | |
1449 | * synchronisation between the global counters and the per-cpu counters. This | |
1450 | * is not a problem because the lock will be local to a CPU almost all the time | |
1451 | * and have little contention except when we get to ENOSPC conditions. | |
1452 | * | |
1453 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1454 | * path while we do things like enabling and disabling counters and | |
1455 | * synchronising the counters. | |
1456 | * | |
1457 | * Locking rules: | |
1458 | * | |
3685c2a1 | 1459 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1460 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1461 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1462 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1463 | * 5. modifying global counters requires holding m_sb_lock |
1464 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1465 | * and _none_ of the per-cpu locks. |
1466 | * | |
1467 | * Disabled counters are only ever re-enabled by a balance operation | |
1468 | * that results in more free resources per CPU than a given threshold. | |
1469 | * To ensure counters don't remain disabled, they are rebalanced when | |
1470 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1471 | * is present to prevent thrashing). | |
e8234a68 DC |
1472 | */ |
1473 | ||
5a67e4c5 | 1474 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1475 | /* |
1476 | * hot-plug CPU notifier support. | |
8d280b98 | 1477 | * |
5a67e4c5 CS |
1478 | * We need a notifier per filesystem as we need to be able to identify |
1479 | * the filesystem to balance the counters out. This is achieved by | |
1480 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1481 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1482 | */ |
e8234a68 DC |
1483 | STATIC int |
1484 | xfs_icsb_cpu_notify( | |
1485 | struct notifier_block *nfb, | |
1486 | unsigned long action, | |
1487 | void *hcpu) | |
1488 | { | |
1489 | xfs_icsb_cnts_t *cntp; | |
1490 | xfs_mount_t *mp; | |
e8234a68 DC |
1491 | |
1492 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1493 | cntp = (xfs_icsb_cnts_t *) | |
1494 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1495 | switch (action) { | |
1496 | case CPU_UP_PREPARE: | |
8bb78442 | 1497 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1498 | /* Easy Case - initialize the area and locks, and |
1499 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1500 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1501 | break; |
1502 | case CPU_ONLINE: | |
8bb78442 | 1503 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1504 | xfs_icsb_lock(mp); |
45af6c6d CH |
1505 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); |
1506 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1507 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1508 | break; |
1509 | case CPU_DEAD: | |
8bb78442 | 1510 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1511 | /* Disable all the counters, then fold the dead cpu's |
1512 | * count into the total on the global superblock and | |
1513 | * re-enable the counters. */ | |
03135cf7 | 1514 | xfs_icsb_lock(mp); |
3685c2a1 | 1515 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1516 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); |
1517 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1518 | ||
e8234a68 DC |
1519 | mp->m_sb.sb_ifree += cntp->icsb_ifree; |
1520 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1521 | ||
01e1b69c | 1522 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1523 | |
45af6c6d CH |
1524 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); |
1525 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1526 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1527 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1528 | break; |
1529 | } | |
1530 | ||
1531 | return NOTIFY_OK; | |
1532 | } | |
5a67e4c5 | 1533 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1534 | |
8d280b98 DC |
1535 | int |
1536 | xfs_icsb_init_counters( | |
1537 | xfs_mount_t *mp) | |
1538 | { | |
1539 | xfs_icsb_cnts_t *cntp; | |
501ab323 | 1540 | int error; |
8d280b98 DC |
1541 | int i; |
1542 | ||
501ab323 DC |
1543 | error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL); |
1544 | if (error) | |
1545 | return error; | |
1546 | ||
8d280b98 | 1547 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); |
501ab323 DC |
1548 | if (!mp->m_sb_cnts) { |
1549 | percpu_counter_destroy(&mp->m_icount); | |
8d280b98 | 1550 | return -ENOMEM; |
501ab323 | 1551 | } |
8d280b98 DC |
1552 | |
1553 | for_each_online_cpu(i) { | |
1554 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1555 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 1556 | } |
20b64285 DC |
1557 | |
1558 | mutex_init(&mp->m_icsb_mutex); | |
1559 | ||
8d280b98 DC |
1560 | /* |
1561 | * start with all counters disabled so that the | |
1562 | * initial balance kicks us off correctly | |
1563 | */ | |
1564 | mp->m_icsb_counters = -1; | |
46677e67 RW |
1565 | |
1566 | #ifdef CONFIG_HOTPLUG_CPU | |
1567 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; | |
1568 | mp->m_icsb_notifier.priority = 0; | |
1569 | register_hotcpu_notifier(&mp->m_icsb_notifier); | |
1570 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1571 | ||
8d280b98 DC |
1572 | return 0; |
1573 | } | |
1574 | ||
5478eead LM |
1575 | void |
1576 | xfs_icsb_reinit_counters( | |
1577 | xfs_mount_t *mp) | |
1578 | { | |
501ab323 DC |
1579 | percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount); |
1580 | ||
5478eead LM |
1581 | xfs_icsb_lock(mp); |
1582 | /* | |
1583 | * start with all counters disabled so that the | |
1584 | * initial balance kicks us off correctly | |
1585 | */ | |
1586 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
1587 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); |
1588 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
1589 | xfs_icsb_unlock(mp); |
1590 | } | |
1591 | ||
c962fb79 | 1592 | void |
8d280b98 DC |
1593 | xfs_icsb_destroy_counters( |
1594 | xfs_mount_t *mp) | |
1595 | { | |
e8234a68 | 1596 | if (mp->m_sb_cnts) { |
5a67e4c5 | 1597 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 1598 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 1599 | } |
501ab323 DC |
1600 | |
1601 | percpu_counter_destroy(&mp->m_icount); | |
1602 | ||
03135cf7 | 1603 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
1604 | } |
1605 | ||
b8f82a4a | 1606 | STATIC void |
01e1b69c DC |
1607 | xfs_icsb_lock_cntr( |
1608 | xfs_icsb_cnts_t *icsbp) | |
1609 | { | |
1610 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1611 | ndelay(1000); | |
1612 | } | |
1613 | } | |
1614 | ||
b8f82a4a | 1615 | STATIC void |
01e1b69c DC |
1616 | xfs_icsb_unlock_cntr( |
1617 | xfs_icsb_cnts_t *icsbp) | |
1618 | { | |
1619 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1620 | } | |
1621 | ||
8d280b98 | 1622 | |
b8f82a4a | 1623 | STATIC void |
8d280b98 DC |
1624 | xfs_icsb_lock_all_counters( |
1625 | xfs_mount_t *mp) | |
1626 | { | |
1627 | xfs_icsb_cnts_t *cntp; | |
1628 | int i; | |
1629 | ||
1630 | for_each_online_cpu(i) { | |
1631 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1632 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1633 | } |
1634 | } | |
1635 | ||
b8f82a4a | 1636 | STATIC void |
8d280b98 DC |
1637 | xfs_icsb_unlock_all_counters( |
1638 | xfs_mount_t *mp) | |
1639 | { | |
1640 | xfs_icsb_cnts_t *cntp; | |
1641 | int i; | |
1642 | ||
1643 | for_each_online_cpu(i) { | |
1644 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1645 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1646 | } |
1647 | } | |
1648 | ||
1649 | STATIC void | |
1650 | xfs_icsb_count( | |
1651 | xfs_mount_t *mp, | |
1652 | xfs_icsb_cnts_t *cnt, | |
1653 | int flags) | |
1654 | { | |
1655 | xfs_icsb_cnts_t *cntp; | |
1656 | int i; | |
1657 | ||
1658 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1659 | ||
1660 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1661 | xfs_icsb_lock_all_counters(mp); | |
1662 | ||
1663 | for_each_online_cpu(i) { | |
1664 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
8d280b98 DC |
1665 | cnt->icsb_ifree += cntp->icsb_ifree; |
1666 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1667 | } | |
1668 | ||
1669 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1670 | xfs_icsb_unlock_all_counters(mp); | |
1671 | } | |
1672 | ||
1673 | STATIC int | |
1674 | xfs_icsb_counter_disabled( | |
1675 | xfs_mount_t *mp, | |
1676 | xfs_sb_field_t field) | |
1677 | { | |
501ab323 | 1678 | ASSERT((field >= XFS_SBS_IFREE) && (field <= XFS_SBS_FDBLOCKS)); |
8d280b98 DC |
1679 | return test_bit(field, &mp->m_icsb_counters); |
1680 | } | |
1681 | ||
36fbe6e6 | 1682 | STATIC void |
8d280b98 DC |
1683 | xfs_icsb_disable_counter( |
1684 | xfs_mount_t *mp, | |
1685 | xfs_sb_field_t field) | |
1686 | { | |
1687 | xfs_icsb_cnts_t cnt; | |
1688 | ||
501ab323 | 1689 | ASSERT((field >= XFS_SBS_IFREE) && (field <= XFS_SBS_FDBLOCKS)); |
8d280b98 | 1690 | |
20b64285 DC |
1691 | /* |
1692 | * If we are already disabled, then there is nothing to do | |
1693 | * here. We check before locking all the counters to avoid | |
1694 | * the expensive lock operation when being called in the | |
1695 | * slow path and the counter is already disabled. This is | |
1696 | * safe because the only time we set or clear this state is under | |
1697 | * the m_icsb_mutex. | |
1698 | */ | |
1699 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 1700 | return; |
20b64285 | 1701 | |
8d280b98 DC |
1702 | xfs_icsb_lock_all_counters(mp); |
1703 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1704 | /* drain back to superblock */ | |
1705 | ||
ce46193b | 1706 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 | 1707 | switch(field) { |
8d280b98 DC |
1708 | case XFS_SBS_IFREE: |
1709 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1710 | break; | |
1711 | case XFS_SBS_FDBLOCKS: | |
1712 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1713 | break; | |
1714 | default: | |
1715 | BUG(); | |
1716 | } | |
1717 | } | |
1718 | ||
1719 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
1720 | } |
1721 | ||
1722 | STATIC void | |
1723 | xfs_icsb_enable_counter( | |
1724 | xfs_mount_t *mp, | |
1725 | xfs_sb_field_t field, | |
1726 | uint64_t count, | |
1727 | uint64_t resid) | |
1728 | { | |
1729 | xfs_icsb_cnts_t *cntp; | |
1730 | int i; | |
1731 | ||
501ab323 | 1732 | ASSERT((field >= XFS_SBS_IFREE) && (field <= XFS_SBS_FDBLOCKS)); |
8d280b98 DC |
1733 | |
1734 | xfs_icsb_lock_all_counters(mp); | |
1735 | for_each_online_cpu(i) { | |
1736 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1737 | switch (field) { | |
8d280b98 DC |
1738 | case XFS_SBS_IFREE: |
1739 | cntp->icsb_ifree = count + resid; | |
1740 | break; | |
1741 | case XFS_SBS_FDBLOCKS: | |
1742 | cntp->icsb_fdblocks = count + resid; | |
1743 | break; | |
1744 | default: | |
1745 | BUG(); | |
1746 | break; | |
1747 | } | |
1748 | resid = 0; | |
1749 | } | |
1750 | clear_bit(field, &mp->m_icsb_counters); | |
1751 | xfs_icsb_unlock_all_counters(mp); | |
1752 | } | |
1753 | ||
dbcabad1 | 1754 | void |
d4d90b57 | 1755 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
1756 | xfs_mount_t *mp, |
1757 | int flags) | |
1758 | { | |
1759 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 1760 | |
8d280b98 DC |
1761 | xfs_icsb_count(mp, &cnt, flags); |
1762 | ||
8d280b98 DC |
1763 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) |
1764 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1765 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1766 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
1767 | } |
1768 | ||
1769 | /* | |
1770 | * Accurate update of per-cpu counters to incore superblock | |
1771 | */ | |
d4d90b57 | 1772 | void |
8d280b98 | 1773 | xfs_icsb_sync_counters( |
d4d90b57 CH |
1774 | xfs_mount_t *mp, |
1775 | int flags) | |
8d280b98 | 1776 | { |
d4d90b57 CH |
1777 | spin_lock(&mp->m_sb_lock); |
1778 | xfs_icsb_sync_counters_locked(mp, flags); | |
1779 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1780 | } |
1781 | ||
1782 | /* | |
1783 | * Balance and enable/disable counters as necessary. | |
1784 | * | |
20b64285 DC |
1785 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
1786 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
1787 | * free blocks is something far enough zero that we aren't going thrash when we | |
1788 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
1789 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
1790 | * be distributed to a single CPU but each CPU has enough blocks to be | |
1791 | * reenabled. | |
1792 | * | |
1793 | * Note that we can be called when counters are already disabled. | |
1794 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
1795 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 1796 | */ |
20b64285 DC |
1797 | |
1798 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 1799 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 1800 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 1801 | STATIC void |
45af6c6d | 1802 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
1803 | xfs_mount_t *mp, |
1804 | xfs_sb_field_t field, | |
20b64285 | 1805 | int min_per_cpu) |
8d280b98 | 1806 | { |
6fdf8ccc | 1807 | uint64_t count, resid; |
8d280b98 | 1808 | int weight = num_online_cpus(); |
20b64285 | 1809 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 1810 | |
8d280b98 DC |
1811 | /* disable counter and sync counter */ |
1812 | xfs_icsb_disable_counter(mp, field); | |
1813 | ||
1814 | /* update counters - first CPU gets residual*/ | |
1815 | switch (field) { | |
8d280b98 DC |
1816 | case XFS_SBS_IFREE: |
1817 | count = mp->m_sb.sb_ifree; | |
1818 | resid = do_div(count, weight); | |
20b64285 | 1819 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1820 | return; |
8d280b98 DC |
1821 | break; |
1822 | case XFS_SBS_FDBLOCKS: | |
1823 | count = mp->m_sb.sb_fdblocks; | |
1824 | resid = do_div(count, weight); | |
20b64285 | 1825 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 1826 | return; |
8d280b98 DC |
1827 | break; |
1828 | default: | |
1829 | BUG(); | |
6fdf8ccc | 1830 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
1831 | break; |
1832 | } | |
1833 | ||
1834 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
1835 | } |
1836 | ||
1837 | STATIC void | |
1838 | xfs_icsb_balance_counter( | |
1839 | xfs_mount_t *mp, | |
1840 | xfs_sb_field_t fields, | |
1841 | int min_per_cpu) | |
1842 | { | |
1843 | spin_lock(&mp->m_sb_lock); | |
1844 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
1845 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1846 | } |
1847 | ||
1b040712 | 1848 | int |
20b64285 | 1849 | xfs_icsb_modify_counters( |
8d280b98 DC |
1850 | xfs_mount_t *mp, |
1851 | xfs_sb_field_t field, | |
20f4ebf2 | 1852 | int64_t delta, |
20b64285 | 1853 | int rsvd) |
8d280b98 DC |
1854 | { |
1855 | xfs_icsb_cnts_t *icsbp; | |
1856 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 1857 | int ret = 0; |
8d280b98 | 1858 | |
20b64285 | 1859 | might_sleep(); |
8d280b98 | 1860 | again: |
7a9e02d6 CL |
1861 | preempt_disable(); |
1862 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
1863 | |
1864 | /* | |
1865 | * if the counter is disabled, go to slow path | |
1866 | */ | |
8d280b98 DC |
1867 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
1868 | goto slow_path; | |
20b64285 DC |
1869 | xfs_icsb_lock_cntr(icsbp); |
1870 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
1871 | xfs_icsb_unlock_cntr(icsbp); | |
1872 | goto slow_path; | |
1873 | } | |
8d280b98 DC |
1874 | |
1875 | switch (field) { | |
8d280b98 DC |
1876 | case XFS_SBS_IFREE: |
1877 | lcounter = icsbp->icsb_ifree; | |
1878 | lcounter += delta; | |
1879 | if (unlikely(lcounter < 0)) | |
20b64285 | 1880 | goto balance_counter; |
8d280b98 DC |
1881 | icsbp->icsb_ifree = lcounter; |
1882 | break; | |
1883 | ||
1884 | case XFS_SBS_FDBLOCKS: | |
1885 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
1886 | ||
4be536de | 1887 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1888 | lcounter += delta; |
1889 | if (unlikely(lcounter < 0)) | |
20b64285 | 1890 | goto balance_counter; |
4be536de | 1891 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1892 | break; |
1893 | default: | |
1894 | BUG(); | |
1895 | break; | |
1896 | } | |
01e1b69c | 1897 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 1898 | preempt_enable(); |
8d280b98 DC |
1899 | return 0; |
1900 | ||
8d280b98 | 1901 | slow_path: |
7a9e02d6 | 1902 | preempt_enable(); |
8d280b98 | 1903 | |
20b64285 DC |
1904 | /* |
1905 | * serialise with a mutex so we don't burn lots of cpu on | |
1906 | * the superblock lock. We still need to hold the superblock | |
1907 | * lock, however, when we modify the global structures. | |
1908 | */ | |
03135cf7 | 1909 | xfs_icsb_lock(mp); |
20b64285 DC |
1910 | |
1911 | /* | |
1912 | * Now running atomically. | |
1913 | * | |
1914 | * If the counter is enabled, someone has beaten us to rebalancing. | |
1915 | * Drop the lock and try again in the fast path.... | |
1916 | */ | |
1917 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 1918 | xfs_icsb_unlock(mp); |
8d280b98 | 1919 | goto again; |
8d280b98 DC |
1920 | } |
1921 | ||
20b64285 DC |
1922 | /* |
1923 | * The counter is currently disabled. Because we are | |
1924 | * running atomically here, we know a rebalance cannot | |
1925 | * be in progress. Hence we can go straight to operating | |
1926 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 1927 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 1928 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 1929 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
1930 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
1931 | * directly on the global counters. | |
1932 | */ | |
3685c2a1 | 1933 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1934 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1935 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 1936 | |
20b64285 DC |
1937 | /* |
1938 | * Now that we've modified the global superblock, we | |
1939 | * may be able to re-enable the distributed counters | |
1940 | * (e.g. lots of space just got freed). After that | |
1941 | * we are done. | |
1942 | */ | |
2451337d | 1943 | if (ret != -ENOSPC) |
45af6c6d | 1944 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 1945 | xfs_icsb_unlock(mp); |
8d280b98 | 1946 | return ret; |
8d280b98 | 1947 | |
20b64285 DC |
1948 | balance_counter: |
1949 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 1950 | preempt_enable(); |
8d280b98 | 1951 | |
20b64285 DC |
1952 | /* |
1953 | * We may have multiple threads here if multiple per-cpu | |
1954 | * counters run dry at the same time. This will mean we can | |
1955 | * do more balances than strictly necessary but it is not | |
1956 | * the common slowpath case. | |
1957 | */ | |
03135cf7 | 1958 | xfs_icsb_lock(mp); |
20b64285 DC |
1959 | |
1960 | /* | |
1961 | * running atomically. | |
1962 | * | |
1963 | * This will leave the counter in the correct state for future | |
1964 | * accesses. After the rebalance, we simply try again and our retry | |
1965 | * will either succeed through the fast path or slow path without | |
1966 | * another balance operation being required. | |
1967 | */ | |
45af6c6d | 1968 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 1969 | xfs_icsb_unlock(mp); |
20b64285 | 1970 | goto again; |
8d280b98 | 1971 | } |
20b64285 | 1972 | |
8d280b98 | 1973 | #endif |