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" |
6ca1c906 | 20 | #include "xfs_format.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
211e4d43 | 25 | #include "xfs_trans_priv.h" |
1da177e4 LT |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
2b9ab5ab DC |
29 | #include "xfs_da_btree.h" |
30 | #include "xfs_dir2_format.h" | |
31 | #include "xfs_dir2.h" | |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_alloc.h" |
40 | #include "xfs_rtalloc.h" | |
41 | #include "xfs_bmap.h" | |
42 | #include "xfs_error.h" | |
1da177e4 LT |
43 | #include "xfs_quota.h" |
44 | #include "xfs_fsops.h" | |
43355099 | 45 | #include "xfs_utils.h" |
0b1b213f | 46 | #include "xfs_trace.h" |
6d8b79cf | 47 | #include "xfs_icache.h" |
04a1e6c5 DC |
48 | #include "xfs_cksum.h" |
49 | #include "xfs_buf_item.h" | |
0b1b213f | 50 | |
1da177e4 | 51 | |
8d280b98 | 52 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 53 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
54 | int); |
55 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
56 | int); | |
36fbe6e6 | 57 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
58 | #else |
59 | ||
45af6c6d CH |
60 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
61 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
62 | #endif |
63 | ||
27174203 CH |
64 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
65 | static int xfs_uuid_table_size; | |
66 | static uuid_t *xfs_uuid_table; | |
67 | ||
68 | /* | |
69 | * See if the UUID is unique among mounted XFS filesystems. | |
70 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
71 | */ | |
72 | STATIC int | |
73 | xfs_uuid_mount( | |
74 | struct xfs_mount *mp) | |
75 | { | |
76 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
77 | int hole, i; | |
78 | ||
79 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
80 | return 0; | |
81 | ||
82 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 83 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
84 | return XFS_ERROR(EINVAL); |
85 | } | |
86 | ||
87 | mutex_lock(&xfs_uuid_table_mutex); | |
88 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
89 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
90 | hole = i; | |
91 | continue; | |
92 | } | |
93 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
94 | goto out_duplicate; | |
95 | } | |
96 | ||
97 | if (hole < 0) { | |
98 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
99 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
100 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
101 | KM_SLEEP); | |
102 | hole = xfs_uuid_table_size++; | |
103 | } | |
104 | xfs_uuid_table[hole] = *uuid; | |
105 | mutex_unlock(&xfs_uuid_table_mutex); | |
106 | ||
107 | return 0; | |
108 | ||
109 | out_duplicate: | |
110 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 111 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
27174203 CH |
112 | return XFS_ERROR(EINVAL); |
113 | } | |
114 | ||
115 | STATIC void | |
116 | xfs_uuid_unmount( | |
117 | struct xfs_mount *mp) | |
118 | { | |
119 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
120 | int i; | |
121 | ||
122 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
123 | return; | |
124 | ||
125 | mutex_lock(&xfs_uuid_table_mutex); | |
126 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
127 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
128 | continue; | |
129 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
130 | continue; | |
131 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
132 | break; | |
133 | } | |
134 | ASSERT(i < xfs_uuid_table_size); | |
135 | mutex_unlock(&xfs_uuid_table_mutex); | |
136 | } | |
137 | ||
138 | ||
e176579e DC |
139 | STATIC void |
140 | __xfs_free_perag( | |
141 | struct rcu_head *head) | |
142 | { | |
143 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
144 | ||
145 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
146 | kmem_free(pag); | |
147 | } | |
148 | ||
1da177e4 | 149 | /* |
e176579e | 150 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 151 | */ |
c962fb79 | 152 | STATIC void |
ff4f038c | 153 | xfs_free_perag( |
745f6919 | 154 | xfs_mount_t *mp) |
1da177e4 | 155 | { |
1c1c6ebc DC |
156 | xfs_agnumber_t agno; |
157 | struct xfs_perag *pag; | |
158 | ||
159 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
160 | spin_lock(&mp->m_perag_lock); | |
161 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
162 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 163 | ASSERT(pag); |
f83282a8 | 164 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 165 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 166 | } |
1da177e4 LT |
167 | } |
168 | ||
4cc929ee NS |
169 | /* |
170 | * Check size of device based on the (data/realtime) block count. | |
171 | * Note: this check is used by the growfs code as well as mount. | |
172 | */ | |
173 | int | |
174 | xfs_sb_validate_fsb_count( | |
175 | xfs_sb_t *sbp, | |
176 | __uint64_t nblocks) | |
177 | { | |
178 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
179 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
180 | ||
181 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
182 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 183 | return EFBIG; |
4cc929ee NS |
184 | #else /* Limited by UINT_MAX of sectors */ |
185 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 186 | return EFBIG; |
4cc929ee NS |
187 | #endif |
188 | return 0; | |
189 | } | |
1da177e4 | 190 | |
1c1c6ebc | 191 | int |
c11e2c36 | 192 | xfs_initialize_perag( |
c11e2c36 | 193 | xfs_mount_t *mp, |
1c1c6ebc DC |
194 | xfs_agnumber_t agcount, |
195 | xfs_agnumber_t *maxagi) | |
1da177e4 | 196 | { |
2d2194f6 | 197 | xfs_agnumber_t index; |
8b26c582 | 198 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
199 | xfs_perag_t *pag; |
200 | xfs_agino_t agino; | |
201 | xfs_ino_t ino; | |
202 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 203 | int error = -ENOMEM; |
1da177e4 | 204 | |
1c1c6ebc DC |
205 | /* |
206 | * Walk the current per-ag tree so we don't try to initialise AGs | |
207 | * that already exist (growfs case). Allocate and insert all the | |
208 | * AGs we don't find ready for initialisation. | |
209 | */ | |
210 | for (index = 0; index < agcount; index++) { | |
211 | pag = xfs_perag_get(mp, index); | |
212 | if (pag) { | |
213 | xfs_perag_put(pag); | |
214 | continue; | |
215 | } | |
8b26c582 DC |
216 | if (!first_initialised) |
217 | first_initialised = index; | |
fb3b504a | 218 | |
1c1c6ebc DC |
219 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
220 | if (!pag) | |
8b26c582 | 221 | goto out_unwind; |
fb3b504a CH |
222 | pag->pag_agno = index; |
223 | pag->pag_mount = mp; | |
1a427ab0 | 224 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 225 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 226 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
227 | spin_lock_init(&pag->pag_buf_lock); |
228 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 229 | |
1c1c6ebc | 230 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 231 | goto out_unwind; |
fb3b504a | 232 | |
1c1c6ebc DC |
233 | spin_lock(&mp->m_perag_lock); |
234 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
235 | BUG(); | |
236 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
237 | radix_tree_preload_end(); |
238 | error = -EEXIST; | |
239 | goto out_unwind; | |
1c1c6ebc DC |
240 | } |
241 | spin_unlock(&mp->m_perag_lock); | |
242 | radix_tree_preload_end(); | |
243 | } | |
244 | ||
fb3b504a CH |
245 | /* |
246 | * If we mount with the inode64 option, or no inode overflows | |
247 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 248 | */ |
fb3b504a CH |
249 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
250 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
251 | ||
252 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 253 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 254 | else |
1da177e4 | 255 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 256 | |
2d2194f6 CM |
257 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
258 | index = xfs_set_inode32(mp); | |
259 | else | |
260 | index = xfs_set_inode64(mp); | |
fb3b504a | 261 | |
1c1c6ebc DC |
262 | if (maxagi) |
263 | *maxagi = index; | |
264 | return 0; | |
8b26c582 DC |
265 | |
266 | out_unwind: | |
267 | kmem_free(pag); | |
268 | for (; index > first_initialised; index--) { | |
269 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
270 | kmem_free(pag); | |
271 | } | |
272 | return error; | |
1da177e4 LT |
273 | } |
274 | ||
1da177e4 LT |
275 | /* |
276 | * xfs_readsb | |
277 | * | |
278 | * Does the initial read of the superblock. | |
279 | */ | |
280 | int | |
ff55068c DC |
281 | xfs_readsb( |
282 | struct xfs_mount *mp, | |
283 | int flags) | |
1da177e4 LT |
284 | { |
285 | unsigned int sector_size; | |
04a1e6c5 DC |
286 | struct xfs_buf *bp; |
287 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 288 | int error; |
af34e09d | 289 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
290 | |
291 | ASSERT(mp->m_sb_bp == NULL); | |
292 | ASSERT(mp->m_ddev_targp != NULL); | |
293 | ||
294 | /* | |
295 | * Allocate a (locked) buffer to hold the superblock. | |
296 | * This will be kept around at all times to optimize | |
297 | * access to the superblock. | |
298 | */ | |
299 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
300 | |
301 | reread: | |
e70b73f8 | 302 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
98021821 | 303 | BTOBB(sector_size), 0, |
1813dd64 DC |
304 | loud ? &xfs_sb_buf_ops |
305 | : &xfs_sb_quiet_buf_ops); | |
26af6552 | 306 | if (!bp) { |
af34e09d DC |
307 | if (loud) |
308 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 309 | return EIO; |
1da177e4 | 310 | } |
eab4e633 DC |
311 | if (bp->b_error) { |
312 | error = bp->b_error; | |
313 | if (loud) | |
e721f504 | 314 | xfs_warn(mp, "SB validate failed with error %d.", error); |
eab4e633 DC |
315 | goto release_buf; |
316 | } | |
1da177e4 LT |
317 | |
318 | /* | |
319 | * Initialize the mount structure from the superblock. | |
1da177e4 | 320 | */ |
98021821 | 321 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
83e782e1 | 322 | xfs_sb_quota_from_disk(&mp->m_sb); |
ff55068c | 323 | |
1da177e4 LT |
324 | /* |
325 | * We must be able to do sector-sized and sector-aligned IO. | |
326 | */ | |
04a1e6c5 | 327 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
328 | if (loud) |
329 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 330 | sector_size, sbp->sb_sectsize); |
1da177e4 | 331 | error = ENOSYS; |
26af6552 | 332 | goto release_buf; |
1da177e4 LT |
333 | } |
334 | ||
335 | /* | |
336 | * If device sector size is smaller than the superblock size, | |
337 | * re-read the superblock so the buffer is correctly sized. | |
338 | */ | |
04a1e6c5 | 339 | if (sector_size < sbp->sb_sectsize) { |
1da177e4 | 340 | xfs_buf_relse(bp); |
04a1e6c5 | 341 | sector_size = sbp->sb_sectsize; |
26af6552 | 342 | goto reread; |
1da177e4 LT |
343 | } |
344 | ||
5478eead LM |
345 | /* Initialize per-cpu counters */ |
346 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 347 | |
04a1e6c5 DC |
348 | /* no need to be quiet anymore, so reset the buf ops */ |
349 | bp->b_ops = &xfs_sb_buf_ops; | |
350 | ||
1da177e4 | 351 | mp->m_sb_bp = bp; |
26af6552 | 352 | xfs_buf_unlock(bp); |
1da177e4 LT |
353 | return 0; |
354 | ||
26af6552 DC |
355 | release_buf: |
356 | xfs_buf_relse(bp); | |
1da177e4 LT |
357 | return error; |
358 | } | |
359 | ||
1da177e4 | 360 | /* |
0771fb45 | 361 | * Update alignment values based on mount options and sb values |
1da177e4 | 362 | */ |
0771fb45 | 363 | STATIC int |
7884bc86 | 364 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 365 | { |
1da177e4 | 366 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 367 | |
4249023a | 368 | if (mp->m_dalign) { |
1da177e4 LT |
369 | /* |
370 | * If stripe unit and stripe width are not multiples | |
371 | * of the fs blocksize turn off alignment. | |
372 | */ | |
373 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
374 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
375 | xfs_warn(mp, |
376 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
377 | sbp->sb_blocksize); | |
378 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
379 | } else { |
380 | /* | |
381 | * Convert the stripe unit and width to FSBs. | |
382 | */ | |
383 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
384 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 385 | xfs_warn(mp, |
39a45d84 JL |
386 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
387 | sbp->sb_agblocks); | |
388 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
389 | } else if (mp->m_dalign) { |
390 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
391 | } else { | |
39a45d84 JL |
392 | xfs_warn(mp, |
393 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
394 | mp->m_dalign, sbp->sb_blocksize); | |
395 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
396 | } |
397 | } | |
398 | ||
399 | /* | |
400 | * Update superblock with new values | |
401 | * and log changes | |
402 | */ | |
62118709 | 403 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
404 | if (sbp->sb_unit != mp->m_dalign) { |
405 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 406 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
407 | } |
408 | if (sbp->sb_width != mp->m_swidth) { | |
409 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 410 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 | 411 | } |
34d7f603 JL |
412 | } else { |
413 | xfs_warn(mp, | |
414 | "cannot change alignment: superblock does not support data alignment"); | |
415 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
416 | } |
417 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 418 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
419 | mp->m_dalign = sbp->sb_unit; |
420 | mp->m_swidth = sbp->sb_width; | |
421 | } | |
422 | ||
0771fb45 ES |
423 | return 0; |
424 | } | |
1da177e4 | 425 | |
0771fb45 ES |
426 | /* |
427 | * Set the maximum inode count for this filesystem | |
428 | */ | |
429 | STATIC void | |
430 | xfs_set_maxicount(xfs_mount_t *mp) | |
431 | { | |
432 | xfs_sb_t *sbp = &(mp->m_sb); | |
433 | __uint64_t icount; | |
1da177e4 | 434 | |
0771fb45 ES |
435 | if (sbp->sb_imax_pct) { |
436 | /* | |
437 | * Make sure the maximum inode count is a multiple | |
438 | * of the units we allocate inodes in. | |
1da177e4 | 439 | */ |
1da177e4 LT |
440 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
441 | do_div(icount, 100); | |
442 | do_div(icount, mp->m_ialloc_blks); | |
443 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
444 | sbp->sb_inopblog; | |
0771fb45 | 445 | } else { |
1da177e4 | 446 | mp->m_maxicount = 0; |
1da177e4 | 447 | } |
0771fb45 ES |
448 | } |
449 | ||
450 | /* | |
451 | * Set the default minimum read and write sizes unless | |
452 | * already specified in a mount option. | |
453 | * We use smaller I/O sizes when the file system | |
454 | * is being used for NFS service (wsync mount option). | |
455 | */ | |
456 | STATIC void | |
457 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
458 | { | |
459 | xfs_sb_t *sbp = &(mp->m_sb); | |
460 | int readio_log, writeio_log; | |
1da177e4 | 461 | |
1da177e4 LT |
462 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
463 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
464 | readio_log = XFS_WSYNC_READIO_LOG; | |
465 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
466 | } else { | |
467 | readio_log = XFS_READIO_LOG_LARGE; | |
468 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
469 | } | |
470 | } else { | |
471 | readio_log = mp->m_readio_log; | |
472 | writeio_log = mp->m_writeio_log; | |
473 | } | |
474 | ||
1da177e4 LT |
475 | if (sbp->sb_blocklog > readio_log) { |
476 | mp->m_readio_log = sbp->sb_blocklog; | |
477 | } else { | |
478 | mp->m_readio_log = readio_log; | |
479 | } | |
480 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
481 | if (sbp->sb_blocklog > writeio_log) { | |
482 | mp->m_writeio_log = sbp->sb_blocklog; | |
483 | } else { | |
484 | mp->m_writeio_log = writeio_log; | |
485 | } | |
486 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 487 | } |
1da177e4 | 488 | |
055388a3 DC |
489 | /* |
490 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
491 | */ | |
492 | void | |
493 | xfs_set_low_space_thresholds( | |
494 | struct xfs_mount *mp) | |
495 | { | |
496 | int i; | |
497 | ||
498 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
499 | __uint64_t space = mp->m_sb.sb_dblocks; | |
500 | ||
501 | do_div(space, 100); | |
502 | mp->m_low_space[i] = space * (i + 1); | |
503 | } | |
504 | } | |
505 | ||
506 | ||
0771fb45 ES |
507 | /* |
508 | * Set whether we're using inode alignment. | |
509 | */ | |
510 | STATIC void | |
511 | xfs_set_inoalignment(xfs_mount_t *mp) | |
512 | { | |
62118709 | 513 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
514 | mp->m_sb.sb_inoalignmt >= |
515 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
516 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
517 | else | |
518 | mp->m_inoalign_mask = 0; | |
519 | /* | |
520 | * If we are using stripe alignment, check whether | |
521 | * the stripe unit is a multiple of the inode alignment | |
522 | */ | |
523 | if (mp->m_dalign && mp->m_inoalign_mask && | |
524 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
525 | mp->m_sinoalign = mp->m_dalign; | |
526 | else | |
527 | mp->m_sinoalign = 0; | |
0771fb45 ES |
528 | } |
529 | ||
530 | /* | |
531 | * Check that the data (and log if separate) are an ok size. | |
532 | */ | |
533 | STATIC int | |
4249023a | 534 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
535 | { |
536 | xfs_buf_t *bp; | |
537 | xfs_daddr_t d; | |
0771fb45 | 538 | |
1da177e4 LT |
539 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
540 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 541 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 542 | return XFS_ERROR(EFBIG); |
1da177e4 | 543 | } |
e70b73f8 | 544 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 545 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 546 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 547 | if (!bp) { |
0b932ccc | 548 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 549 | return EIO; |
1da177e4 | 550 | } |
1922c949 | 551 | xfs_buf_relse(bp); |
1da177e4 | 552 | |
4249023a | 553 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
554 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
555 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 556 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 557 | return XFS_ERROR(EFBIG); |
1da177e4 | 558 | } |
e70b73f8 | 559 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 560 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 561 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 562 | if (!bp) { |
0b932ccc | 563 | xfs_warn(mp, "log device read failed"); |
1922c949 | 564 | return EIO; |
0771fb45 | 565 | } |
1922c949 | 566 | xfs_buf_relse(bp); |
0771fb45 ES |
567 | } |
568 | return 0; | |
569 | } | |
570 | ||
7d095257 CH |
571 | /* |
572 | * Clear the quotaflags in memory and in the superblock. | |
573 | */ | |
574 | int | |
575 | xfs_mount_reset_sbqflags( | |
576 | struct xfs_mount *mp) | |
577 | { | |
578 | int error; | |
579 | struct xfs_trans *tp; | |
580 | ||
581 | mp->m_qflags = 0; | |
582 | ||
583 | /* | |
584 | * It is OK to look at sb_qflags here in mount path, | |
585 | * without m_sb_lock. | |
586 | */ | |
587 | if (mp->m_sb.sb_qflags == 0) | |
588 | return 0; | |
589 | spin_lock(&mp->m_sb_lock); | |
590 | mp->m_sb.sb_qflags = 0; | |
591 | spin_unlock(&mp->m_sb_lock); | |
592 | ||
593 | /* | |
594 | * If the fs is readonly, let the incore superblock run | |
595 | * with quotas off but don't flush the update out to disk | |
596 | */ | |
597 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
598 | return 0; | |
599 | ||
7d095257 | 600 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
b0c10b98 JL |
601 | error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp), |
602 | 0, 0, XFS_DEFAULT_LOG_COUNT); | |
7d095257 CH |
603 | if (error) { |
604 | xfs_trans_cancel(tp, 0); | |
53487786 | 605 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
606 | return error; |
607 | } | |
608 | ||
609 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
610 | return xfs_trans_commit(tp, 0); | |
611 | } | |
612 | ||
d5db0f97 ES |
613 | __uint64_t |
614 | xfs_default_resblks(xfs_mount_t *mp) | |
615 | { | |
616 | __uint64_t resblks; | |
617 | ||
618 | /* | |
8babd8a2 DC |
619 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
620 | * smaller. This is intended to cover concurrent allocation | |
621 | * transactions when we initially hit enospc. These each require a 4 | |
622 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
623 | * allocation reservations. | |
d5db0f97 ES |
624 | */ |
625 | resblks = mp->m_sb.sb_dblocks; | |
626 | do_div(resblks, 20); | |
8babd8a2 | 627 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
628 | return resblks; |
629 | } | |
630 | ||
0771fb45 | 631 | /* |
0771fb45 ES |
632 | * This function does the following on an initial mount of a file system: |
633 | * - reads the superblock from disk and init the mount struct | |
634 | * - if we're a 32-bit kernel, do a size check on the superblock | |
635 | * so we don't mount terabyte filesystems | |
636 | * - init mount struct realtime fields | |
637 | * - allocate inode hash table for fs | |
638 | * - init directory manager | |
639 | * - perform recovery and init the log manager | |
640 | */ | |
641 | int | |
642 | xfs_mountfs( | |
4249023a | 643 | xfs_mount_t *mp) |
0771fb45 ES |
644 | { |
645 | xfs_sb_t *sbp = &(mp->m_sb); | |
646 | xfs_inode_t *rip; | |
0771fb45 | 647 | __uint64_t resblks; |
7d095257 CH |
648 | uint quotamount = 0; |
649 | uint quotaflags = 0; | |
0771fb45 ES |
650 | int error = 0; |
651 | ||
ff55068c | 652 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 653 | |
ee1c0908 | 654 | /* |
e6957ea4 ES |
655 | * Check for a mismatched features2 values. Older kernels |
656 | * read & wrote into the wrong sb offset for sb_features2 | |
657 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
658 | * when sb_features2 was added, which made older superblock | |
659 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 660 | * |
e6957ea4 ES |
661 | * For backwards compatibility, we make both slots equal. |
662 | * | |
663 | * If we detect a mismatched field, we OR the set bits into the | |
664 | * existing features2 field in case it has already been modified; we | |
665 | * don't want to lose any features. We then update the bad location | |
666 | * with the ORed value so that older kernels will see any features2 | |
667 | * flags, and mark the two fields as needing updates once the | |
668 | * transaction subsystem is online. | |
ee1c0908 | 669 | */ |
e6957ea4 | 670 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 671 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 672 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 673 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 674 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
675 | |
676 | /* | |
677 | * Re-check for ATTR2 in case it was found in bad_features2 | |
678 | * slot. | |
679 | */ | |
7c12f296 TS |
680 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
681 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 682 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
683 | } |
684 | ||
685 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
686 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
687 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 688 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 689 | |
7c12f296 TS |
690 | /* update sb_versionnum for the clearing of the morebits */ |
691 | if (!sbp->sb_features2) | |
7884bc86 | 692 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
693 | } |
694 | ||
0771fb45 ES |
695 | /* |
696 | * Check if sb_agblocks is aligned at stripe boundary | |
697 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
698 | * allocator alignment is within an ag, therefore ag has | |
699 | * to be aligned at stripe boundary. | |
700 | */ | |
7884bc86 | 701 | error = xfs_update_alignment(mp); |
0771fb45 | 702 | if (error) |
f9057e3d | 703 | goto out; |
0771fb45 ES |
704 | |
705 | xfs_alloc_compute_maxlevels(mp); | |
706 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
707 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
708 | xfs_ialloc_compute_maxlevels(mp); | |
709 | ||
710 | xfs_set_maxicount(mp); | |
711 | ||
27174203 CH |
712 | error = xfs_uuid_mount(mp); |
713 | if (error) | |
714 | goto out; | |
1da177e4 | 715 | |
0771fb45 ES |
716 | /* |
717 | * Set the minimum read and write sizes | |
718 | */ | |
719 | xfs_set_rw_sizes(mp); | |
720 | ||
055388a3 DC |
721 | /* set the low space thresholds for dynamic preallocation */ |
722 | xfs_set_low_space_thresholds(mp); | |
723 | ||
0771fb45 ES |
724 | /* |
725 | * Set the inode cluster size. | |
726 | * This may still be overridden by the file system | |
727 | * block size if it is larger than the chosen cluster size. | |
728 | */ | |
729 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
730 | ||
731 | /* | |
732 | * Set inode alignment fields | |
733 | */ | |
734 | xfs_set_inoalignment(mp); | |
735 | ||
736 | /* | |
737 | * Check that the data (and log if separate) are an ok size. | |
738 | */ | |
4249023a | 739 | error = xfs_check_sizes(mp); |
0771fb45 | 740 | if (error) |
f9057e3d | 741 | goto out_remove_uuid; |
0771fb45 | 742 | |
1da177e4 LT |
743 | /* |
744 | * Initialize realtime fields in the mount structure | |
745 | */ | |
0771fb45 ES |
746 | error = xfs_rtmount_init(mp); |
747 | if (error) { | |
0b932ccc | 748 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 749 | goto out_remove_uuid; |
1da177e4 LT |
750 | } |
751 | ||
1da177e4 LT |
752 | /* |
753 | * Copies the low order bits of the timestamp and the randomly | |
754 | * set "sequence" number out of a UUID. | |
755 | */ | |
756 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
757 | ||
1da177e4 LT |
758 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
759 | ||
f6c2d1fa | 760 | xfs_dir_mount(mp); |
1da177e4 LT |
761 | |
762 | /* | |
763 | * Initialize the attribute manager's entries. | |
764 | */ | |
765 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
766 | ||
767 | /* | |
768 | * Initialize the precomputed transaction reservations values. | |
769 | */ | |
770 | xfs_trans_init(mp); | |
771 | ||
1da177e4 LT |
772 | /* |
773 | * Allocate and initialize the per-ag data. | |
774 | */ | |
1c1c6ebc | 775 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 776 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
777 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
778 | if (error) { | |
0b932ccc | 779 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 780 | goto out_remove_uuid; |
1c1c6ebc | 781 | } |
1da177e4 | 782 | |
f9057e3d | 783 | if (!sbp->sb_logblocks) { |
0b932ccc | 784 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
785 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
786 | error = XFS_ERROR(EFSCORRUPTED); | |
787 | goto out_free_perag; | |
788 | } | |
789 | ||
1da177e4 LT |
790 | /* |
791 | * log's mount-time initialization. Perform 1st part recovery if needed | |
792 | */ | |
f9057e3d CH |
793 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
794 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
795 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
796 | if (error) { | |
0b932ccc | 797 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 798 | goto out_fail_wait; |
1da177e4 LT |
799 | } |
800 | ||
92821e2b DC |
801 | /* |
802 | * Now the log is mounted, we know if it was an unclean shutdown or | |
803 | * not. If it was, with the first phase of recovery has completed, we | |
804 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
805 | * but they are recovered transactionally in the second recovery phase | |
806 | * later. | |
807 | * | |
808 | * Hence we can safely re-initialise incore superblock counters from | |
809 | * the per-ag data. These may not be correct if the filesystem was not | |
810 | * cleanly unmounted, so we need to wait for recovery to finish before | |
811 | * doing this. | |
812 | * | |
813 | * If the filesystem was cleanly unmounted, then we can trust the | |
814 | * values in the superblock to be correct and we don't need to do | |
815 | * anything here. | |
816 | * | |
817 | * If we are currently making the filesystem, the initialisation will | |
818 | * fail as the perag data is in an undefined state. | |
819 | */ | |
92821e2b DC |
820 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
821 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
822 | !mp->m_sb.sb_inprogress) { | |
823 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 824 | if (error) |
d4f3512b | 825 | goto out_fail_wait; |
92821e2b | 826 | } |
f9057e3d | 827 | |
1da177e4 LT |
828 | /* |
829 | * Get and sanity-check the root inode. | |
830 | * Save the pointer to it in the mount structure. | |
831 | */ | |
7b6259e7 | 832 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 833 | if (error) { |
0b932ccc | 834 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 835 | goto out_log_dealloc; |
1da177e4 LT |
836 | } |
837 | ||
838 | ASSERT(rip != NULL); | |
1da177e4 | 839 | |
abbede1b | 840 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 841 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 842 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
843 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
844 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
845 | mp); | |
846 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 847 | goto out_rele_rip; |
1da177e4 LT |
848 | } |
849 | mp->m_rootip = rip; /* save it */ | |
850 | ||
851 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
852 | ||
853 | /* | |
854 | * Initialize realtime inode pointers in the mount structure | |
855 | */ | |
0771fb45 ES |
856 | error = xfs_rtmount_inodes(mp); |
857 | if (error) { | |
1da177e4 LT |
858 | /* |
859 | * Free up the root inode. | |
860 | */ | |
0b932ccc | 861 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 862 | goto out_rele_rip; |
1da177e4 LT |
863 | } |
864 | ||
865 | /* | |
7884bc86 CH |
866 | * If this is a read-only mount defer the superblock updates until |
867 | * the next remount into writeable mode. Otherwise we would never | |
868 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 869 | */ |
7884bc86 CH |
870 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
871 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 872 | if (error) { |
0b932ccc | 873 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 874 | goto out_rtunmount; |
e5720eec DC |
875 | } |
876 | } | |
1da177e4 LT |
877 | |
878 | /* | |
879 | * Initialise the XFS quota management subsystem for this mount | |
880 | */ | |
7d095257 CH |
881 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
882 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
883 | if (error) | |
884 | goto out_rtunmount; | |
885 | } else { | |
886 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
887 | ||
888 | /* | |
889 | * If a file system had quotas running earlier, but decided to | |
890 | * mount without -o uquota/pquota/gquota options, revoke the | |
891 | * quotachecked license. | |
892 | */ | |
893 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 894 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
895 | error = xfs_mount_reset_sbqflags(mp); |
896 | if (error) | |
897 | return error; | |
898 | } | |
899 | } | |
1da177e4 LT |
900 | |
901 | /* | |
902 | * Finish recovering the file system. This part needed to be | |
903 | * delayed until after the root and real-time bitmap inodes | |
904 | * were consistently read in. | |
905 | */ | |
4249023a | 906 | error = xfs_log_mount_finish(mp); |
1da177e4 | 907 | if (error) { |
0b932ccc | 908 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 909 | goto out_rtunmount; |
1da177e4 LT |
910 | } |
911 | ||
912 | /* | |
913 | * Complete the quota initialisation, post-log-replay component. | |
914 | */ | |
7d095257 CH |
915 | if (quotamount) { |
916 | ASSERT(mp->m_qflags == 0); | |
917 | mp->m_qflags = quotaflags; | |
918 | ||
919 | xfs_qm_mount_quotas(mp); | |
920 | } | |
921 | ||
84e1e99f DC |
922 | /* |
923 | * Now we are mounted, reserve a small amount of unused space for | |
924 | * privileged transactions. This is needed so that transaction | |
925 | * space required for critical operations can dip into this pool | |
926 | * when at ENOSPC. This is needed for operations like create with | |
927 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
928 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
929 | * |
930 | * This may drive us straight to ENOSPC on mount, but that implies | |
931 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 932 | */ |
d5db0f97 ES |
933 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
934 | resblks = xfs_default_resblks(mp); | |
935 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
936 | if (error) | |
0b932ccc DC |
937 | xfs_warn(mp, |
938 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 939 | } |
84e1e99f | 940 | |
1da177e4 LT |
941 | return 0; |
942 | ||
b93b6e43 CH |
943 | out_rtunmount: |
944 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 945 | out_rele_rip: |
43355099 | 946 | IRELE(rip); |
f9057e3d | 947 | out_log_dealloc: |
21b699c8 | 948 | xfs_log_unmount(mp); |
d4f3512b DC |
949 | out_fail_wait: |
950 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
951 | xfs_wait_buftarg(mp->m_logdev_targp); | |
952 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 953 | out_free_perag: |
ff4f038c | 954 | xfs_free_perag(mp); |
f9057e3d | 955 | out_remove_uuid: |
27174203 | 956 | xfs_uuid_unmount(mp); |
f9057e3d | 957 | out: |
1da177e4 LT |
958 | return error; |
959 | } | |
960 | ||
961 | /* | |
1da177e4 LT |
962 | * This flushes out the inodes,dquots and the superblock, unmounts the |
963 | * log and makes sure that incore structures are freed. | |
964 | */ | |
41b5c2e7 CH |
965 | void |
966 | xfs_unmountfs( | |
967 | struct xfs_mount *mp) | |
1da177e4 | 968 | { |
41b5c2e7 CH |
969 | __uint64_t resblks; |
970 | int error; | |
1da177e4 | 971 | |
579b62fa BF |
972 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
973 | ||
7d095257 | 974 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 975 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
976 | IRELE(mp->m_rootip); |
977 | ||
641c56fb DC |
978 | /* |
979 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 980 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
981 | * the transaction is still sitting in a iclog. The stale inodes |
982 | * on that buffer will have their flush locks held until the | |
983 | * transaction hits the disk and the callbacks run. the inode | |
984 | * flush takes the flush lock unconditionally and with nothing to | |
985 | * push out the iclog we will never get that unlocked. hence we | |
986 | * need to force the log first. | |
987 | */ | |
a14a348b | 988 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
989 | |
990 | /* | |
211e4d43 CH |
991 | * Flush all pending changes from the AIL. |
992 | */ | |
993 | xfs_ail_push_all_sync(mp->m_ail); | |
994 | ||
995 | /* | |
996 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
997 | * inodes and none should be pinned or locked, but use synchronous |
998 | * reclaim just to be sure. We can stop background inode reclaim | |
999 | * here as well if it is still running. | |
c854363e | 1000 | */ |
7e18530b | 1001 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1002 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1003 | |
7d095257 | 1004 | xfs_qm_unmount(mp); |
a357a121 | 1005 | |
84e1e99f DC |
1006 | /* |
1007 | * Unreserve any blocks we have so that when we unmount we don't account | |
1008 | * the reserved free space as used. This is really only necessary for | |
1009 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1010 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1011 | * |
1012 | * We don't bother correcting this elsewhere for lazy superblock | |
1013 | * counting because on mount of an unclean filesystem we reconstruct the | |
1014 | * correct counter value and this is irrelevant. | |
1015 | * | |
1016 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1017 | * we only every apply deltas to the superblock and hence the incore | |
1018 | * value does not matter.... | |
1019 | */ | |
1020 | resblks = 0; | |
714082bc DC |
1021 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1022 | if (error) | |
0b932ccc | 1023 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1024 | "Freespace may not be correct on next mount."); |
1025 | ||
adab0f67 | 1026 | error = xfs_log_sbcount(mp); |
e5720eec | 1027 | if (error) |
0b932ccc | 1028 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1029 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1030 | |
21b699c8 | 1031 | xfs_log_unmount(mp); |
27174203 | 1032 | xfs_uuid_unmount(mp); |
1da177e4 | 1033 | |
1550d0b0 | 1034 | #if defined(DEBUG) |
0ce4cfd4 | 1035 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1036 | #endif |
ff4f038c | 1037 | xfs_free_perag(mp); |
1da177e4 LT |
1038 | } |
1039 | ||
92821e2b DC |
1040 | int |
1041 | xfs_fs_writable(xfs_mount_t *mp) | |
1042 | { | |
d9457dc0 | 1043 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1044 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1045 | } |
1046 | ||
1047 | /* | |
b2ce3974 AE |
1048 | * xfs_log_sbcount |
1049 | * | |
adab0f67 | 1050 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1051 | * |
1052 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1053 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1054 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1055 | */ |
1056 | int | |
adab0f67 | 1057 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1058 | { |
1059 | xfs_trans_t *tp; | |
1060 | int error; | |
1061 | ||
1062 | if (!xfs_fs_writable(mp)) | |
1063 | return 0; | |
1064 | ||
d4d90b57 | 1065 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1066 | |
1067 | /* | |
1068 | * we don't need to do this if we are updating the superblock | |
1069 | * counters on every modification. | |
1070 | */ | |
1071 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1072 | return 0; | |
1073 | ||
b2ce3974 | 1074 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
e457274b JL |
1075 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
1076 | XFS_DEFAULT_LOG_COUNT); | |
92821e2b DC |
1077 | if (error) { |
1078 | xfs_trans_cancel(tp, 0); | |
1079 | return error; | |
1080 | } | |
1081 | ||
1082 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1083 | xfs_trans_set_sync(tp); |
e5720eec DC |
1084 | error = xfs_trans_commit(tp, 0); |
1085 | return error; | |
92821e2b DC |
1086 | } |
1087 | ||
1da177e4 LT |
1088 | /* |
1089 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1090 | * a delta to a specified field in the in-core superblock. Simply | |
1091 | * switch on the field indicated and apply the delta to that field. | |
1092 | * Fields are not allowed to dip below zero, so if the delta would | |
1093 | * do this do not apply it and return EINVAL. | |
1094 | * | |
3685c2a1 | 1095 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1096 | */ |
d96f8f89 | 1097 | STATIC int |
20f4ebf2 DC |
1098 | xfs_mod_incore_sb_unlocked( |
1099 | xfs_mount_t *mp, | |
1100 | xfs_sb_field_t field, | |
1101 | int64_t delta, | |
1102 | int rsvd) | |
1da177e4 LT |
1103 | { |
1104 | int scounter; /* short counter for 32 bit fields */ | |
1105 | long long lcounter; /* long counter for 64 bit fields */ | |
1106 | long long res_used, rem; | |
1107 | ||
1108 | /* | |
1109 | * With the in-core superblock spin lock held, switch | |
1110 | * on the indicated field. Apply the delta to the | |
1111 | * proper field. If the fields value would dip below | |
1112 | * 0, then do not apply the delta and return EINVAL. | |
1113 | */ | |
1114 | switch (field) { | |
1115 | case XFS_SBS_ICOUNT: | |
1116 | lcounter = (long long)mp->m_sb.sb_icount; | |
1117 | lcounter += delta; | |
1118 | if (lcounter < 0) { | |
1119 | ASSERT(0); | |
014c2544 | 1120 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1121 | } |
1122 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1123 | return 0; |
1da177e4 LT |
1124 | case XFS_SBS_IFREE: |
1125 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1126 | lcounter += delta; | |
1127 | if (lcounter < 0) { | |
1128 | ASSERT(0); | |
014c2544 | 1129 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1130 | } |
1131 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1132 | return 0; |
1da177e4 | 1133 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1134 | lcounter = (long long) |
1135 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1136 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1137 | ||
1138 | if (delta > 0) { /* Putting blocks back */ | |
1139 | if (res_used > delta) { | |
1140 | mp->m_resblks_avail += delta; | |
1141 | } else { | |
1142 | rem = delta - res_used; | |
1143 | mp->m_resblks_avail = mp->m_resblks; | |
1144 | lcounter += rem; | |
1145 | } | |
1146 | } else { /* Taking blocks away */ | |
1da177e4 | 1147 | lcounter += delta; |
8babd8a2 DC |
1148 | if (lcounter >= 0) { |
1149 | mp->m_sb.sb_fdblocks = lcounter + | |
1150 | XFS_ALLOC_SET_ASIDE(mp); | |
1151 | return 0; | |
1152 | } | |
1da177e4 | 1153 | |
8babd8a2 DC |
1154 | /* |
1155 | * We are out of blocks, use any available reserved | |
1156 | * blocks if were allowed to. | |
1157 | */ | |
1158 | if (!rsvd) | |
1159 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1160 | |
8babd8a2 DC |
1161 | lcounter = (long long)mp->m_resblks_avail + delta; |
1162 | if (lcounter >= 0) { | |
1163 | mp->m_resblks_avail = lcounter; | |
1164 | return 0; | |
1da177e4 | 1165 | } |
8babd8a2 DC |
1166 | printk_once(KERN_WARNING |
1167 | "Filesystem \"%s\": reserve blocks depleted! " | |
1168 | "Consider increasing reserve pool size.", | |
1169 | mp->m_fsname); | |
1170 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1171 | } |
1172 | ||
4be536de | 1173 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1174 | return 0; |
1da177e4 LT |
1175 | case XFS_SBS_FREXTENTS: |
1176 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1177 | lcounter += delta; | |
1178 | if (lcounter < 0) { | |
014c2544 | 1179 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1180 | } |
1181 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1182 | return 0; |
1da177e4 LT |
1183 | case XFS_SBS_DBLOCKS: |
1184 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1185 | lcounter += delta; | |
1186 | if (lcounter < 0) { | |
1187 | ASSERT(0); | |
014c2544 | 1188 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1189 | } |
1190 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1191 | return 0; |
1da177e4 LT |
1192 | case XFS_SBS_AGCOUNT: |
1193 | scounter = mp->m_sb.sb_agcount; | |
1194 | scounter += delta; | |
1195 | if (scounter < 0) { | |
1196 | ASSERT(0); | |
014c2544 | 1197 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1198 | } |
1199 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1200 | return 0; |
1da177e4 LT |
1201 | case XFS_SBS_IMAX_PCT: |
1202 | scounter = mp->m_sb.sb_imax_pct; | |
1203 | scounter += delta; | |
1204 | if (scounter < 0) { | |
1205 | ASSERT(0); | |
014c2544 | 1206 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1207 | } |
1208 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1209 | return 0; |
1da177e4 LT |
1210 | case XFS_SBS_REXTSIZE: |
1211 | scounter = mp->m_sb.sb_rextsize; | |
1212 | scounter += delta; | |
1213 | if (scounter < 0) { | |
1214 | ASSERT(0); | |
014c2544 | 1215 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1216 | } |
1217 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1218 | return 0; |
1da177e4 LT |
1219 | case XFS_SBS_RBMBLOCKS: |
1220 | scounter = mp->m_sb.sb_rbmblocks; | |
1221 | scounter += delta; | |
1222 | if (scounter < 0) { | |
1223 | ASSERT(0); | |
014c2544 | 1224 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1225 | } |
1226 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1227 | return 0; |
1da177e4 LT |
1228 | case XFS_SBS_RBLOCKS: |
1229 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1230 | lcounter += delta; | |
1231 | if (lcounter < 0) { | |
1232 | ASSERT(0); | |
014c2544 | 1233 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1234 | } |
1235 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1236 | return 0; |
1da177e4 LT |
1237 | case XFS_SBS_REXTENTS: |
1238 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1239 | lcounter += delta; | |
1240 | if (lcounter < 0) { | |
1241 | ASSERT(0); | |
014c2544 | 1242 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1243 | } |
1244 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1245 | return 0; |
1da177e4 LT |
1246 | case XFS_SBS_REXTSLOG: |
1247 | scounter = mp->m_sb.sb_rextslog; | |
1248 | scounter += delta; | |
1249 | if (scounter < 0) { | |
1250 | ASSERT(0); | |
014c2544 | 1251 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1252 | } |
1253 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1254 | return 0; |
1da177e4 LT |
1255 | default: |
1256 | ASSERT(0); | |
014c2544 | 1257 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1258 | } |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1263 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1264 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1265 | * routine to do the work. |
1266 | */ | |
1267 | int | |
20f4ebf2 | 1268 | xfs_mod_incore_sb( |
96540c78 CH |
1269 | struct xfs_mount *mp, |
1270 | xfs_sb_field_t field, | |
1271 | int64_t delta, | |
1272 | int rsvd) | |
1da177e4 | 1273 | { |
96540c78 | 1274 | int status; |
1da177e4 | 1275 | |
8d280b98 | 1276 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1277 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1278 | #endif |
96540c78 CH |
1279 | spin_lock(&mp->m_sb_lock); |
1280 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1281 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1282 | |
014c2544 | 1283 | return status; |
1da177e4 LT |
1284 | } |
1285 | ||
1286 | /* | |
1b040712 | 1287 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1288 | * |
1b040712 CH |
1289 | * The fields and changes to those fields are specified in the array of |
1290 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1291 | * will be applied or none of them will. If any modified field dips below 0, | |
1292 | * then all modifications will be backed out and EINVAL will be returned. | |
1293 | * | |
1294 | * Note that this function may not be used for the superblock values that | |
1295 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1296 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1297 | */ |
1298 | int | |
1b040712 CH |
1299 | xfs_mod_incore_sb_batch( |
1300 | struct xfs_mount *mp, | |
1301 | xfs_mod_sb_t *msb, | |
1302 | uint nmsb, | |
1303 | int rsvd) | |
1da177e4 | 1304 | { |
45c51b99 | 1305 | xfs_mod_sb_t *msbp; |
1b040712 | 1306 | int error = 0; |
1da177e4 LT |
1307 | |
1308 | /* | |
1b040712 CH |
1309 | * Loop through the array of mod structures and apply each individually. |
1310 | * If any fail, then back out all those which have already been applied. | |
1311 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1312 | * changes will be atomic. | |
1da177e4 | 1313 | */ |
3685c2a1 | 1314 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1315 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1316 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1317 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1318 | |
1b040712 CH |
1319 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1320 | msbp->msb_delta, rsvd); | |
1321 | if (error) | |
1322 | goto unwind; | |
1da177e4 | 1323 | } |
1b040712 CH |
1324 | spin_unlock(&mp->m_sb_lock); |
1325 | return 0; | |
1da177e4 | 1326 | |
1b040712 CH |
1327 | unwind: |
1328 | while (--msbp >= msb) { | |
1329 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1330 | -msbp->msb_delta, rsvd); | |
1331 | ASSERT(error == 0); | |
1da177e4 | 1332 | } |
3685c2a1 | 1333 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1334 | return error; |
1da177e4 LT |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1339 | * The buffer is returned locked and read in from disk. | |
1340 | * The buffer should be released with a call to xfs_brelse(). | |
1341 | * | |
1342 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1343 | * the superblock buffer if it can be locked without sleeping. | |
1344 | * If it can't then we'll return NULL. | |
1345 | */ | |
0c842ad4 | 1346 | struct xfs_buf * |
1da177e4 | 1347 | xfs_getsb( |
0c842ad4 CH |
1348 | struct xfs_mount *mp, |
1349 | int flags) | |
1da177e4 | 1350 | { |
0c842ad4 | 1351 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1352 | |
0c842ad4 CH |
1353 | if (!xfs_buf_trylock(bp)) { |
1354 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1355 | return NULL; |
0c842ad4 | 1356 | xfs_buf_lock(bp); |
1da177e4 | 1357 | } |
0c842ad4 | 1358 | |
72790aa1 | 1359 | xfs_buf_hold(bp); |
1da177e4 | 1360 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 1361 | return bp; |
1da177e4 LT |
1362 | } |
1363 | ||
1364 | /* | |
1365 | * Used to free the superblock along various error paths. | |
1366 | */ | |
1367 | void | |
1368 | xfs_freesb( | |
26af6552 | 1369 | struct xfs_mount *mp) |
1da177e4 | 1370 | { |
26af6552 | 1371 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1372 | |
26af6552 | 1373 | xfs_buf_lock(bp); |
1da177e4 | 1374 | mp->m_sb_bp = NULL; |
26af6552 | 1375 | xfs_buf_relse(bp); |
1da177e4 LT |
1376 | } |
1377 | ||
1da177e4 LT |
1378 | /* |
1379 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1380 | * be altered by the mount options, as well as any potential sb_features2 |
1381 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1382 | */ |
7884bc86 | 1383 | int |
ee1c0908 | 1384 | xfs_mount_log_sb( |
1da177e4 LT |
1385 | xfs_mount_t *mp, |
1386 | __int64_t fields) | |
1387 | { | |
1388 | xfs_trans_t *tp; | |
e5720eec | 1389 | int error; |
1da177e4 | 1390 | |
ee1c0908 | 1391 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1392 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1393 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1394 | |
1395 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
5166ab06 JL |
1396 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
1397 | XFS_DEFAULT_LOG_COUNT); | |
e5720eec | 1398 | if (error) { |
1da177e4 | 1399 | xfs_trans_cancel(tp, 0); |
e5720eec | 1400 | return error; |
1da177e4 LT |
1401 | } |
1402 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1403 | error = xfs_trans_commit(tp, 0); |
1404 | return error; | |
1da177e4 | 1405 | } |
8d280b98 | 1406 | |
dda35b8f CH |
1407 | /* |
1408 | * If the underlying (data/log/rt) device is readonly, there are some | |
1409 | * operations that cannot proceed. | |
1410 | */ | |
1411 | int | |
1412 | xfs_dev_is_read_only( | |
1413 | struct xfs_mount *mp, | |
1414 | char *message) | |
1415 | { | |
1416 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1417 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1418 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1419 | xfs_notice(mp, "%s required on read-only device.", message); |
1420 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
1421 | return EROFS; |
1422 | } | |
1423 | return 0; | |
1424 | } | |
8d280b98 DC |
1425 | |
1426 | #ifdef HAVE_PERCPU_SB | |
1427 | /* | |
1428 | * Per-cpu incore superblock counters | |
1429 | * | |
1430 | * Simple concept, difficult implementation | |
1431 | * | |
1432 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1433 | * counter for contended fields (e.g. free block count). | |
1434 | * | |
1435 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1436 | * hence needs to be accurately read when we are running low on space. Hence | |
1437 | * there is a method to enable and disable the per-cpu counters based on how | |
1438 | * much "stuff" is available in them. | |
1439 | * | |
1440 | * Basically, a counter is enabled if there is enough free resource to justify | |
1441 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1442 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1443 | * re-distribute the available resources. | |
1444 | * | |
1445 | * If, once we redistributed the available resources, we still get a failure, | |
1446 | * we disable the per-cpu counter and go through the slow path. | |
1447 | * | |
1448 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 1449 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
1450 | * the global superblock. We do this after disabling the counter to prevent |
1451 | * more threads from queueing up on the counter. | |
1452 | * | |
1453 | * Essentially, this means that we still need a lock in the fast path to enable | |
1454 | * synchronisation between the global counters and the per-cpu counters. This | |
1455 | * is not a problem because the lock will be local to a CPU almost all the time | |
1456 | * and have little contention except when we get to ENOSPC conditions. | |
1457 | * | |
1458 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1459 | * path while we do things like enabling and disabling counters and | |
1460 | * synchronising the counters. | |
1461 | * | |
1462 | * Locking rules: | |
1463 | * | |
3685c2a1 | 1464 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1465 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1466 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1467 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1468 | * 5. modifying global counters requires holding m_sb_lock |
1469 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1470 | * and _none_ of the per-cpu locks. |
1471 | * | |
1472 | * Disabled counters are only ever re-enabled by a balance operation | |
1473 | * that results in more free resources per CPU than a given threshold. | |
1474 | * To ensure counters don't remain disabled, they are rebalanced when | |
1475 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1476 | * is present to prevent thrashing). | |
e8234a68 DC |
1477 | */ |
1478 | ||
5a67e4c5 | 1479 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1480 | /* |
1481 | * hot-plug CPU notifier support. | |
8d280b98 | 1482 | * |
5a67e4c5 CS |
1483 | * We need a notifier per filesystem as we need to be able to identify |
1484 | * the filesystem to balance the counters out. This is achieved by | |
1485 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1486 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1487 | */ |
e8234a68 DC |
1488 | STATIC int |
1489 | xfs_icsb_cpu_notify( | |
1490 | struct notifier_block *nfb, | |
1491 | unsigned long action, | |
1492 | void *hcpu) | |
1493 | { | |
1494 | xfs_icsb_cnts_t *cntp; | |
1495 | xfs_mount_t *mp; | |
e8234a68 DC |
1496 | |
1497 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1498 | cntp = (xfs_icsb_cnts_t *) | |
1499 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1500 | switch (action) { | |
1501 | case CPU_UP_PREPARE: | |
8bb78442 | 1502 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1503 | /* Easy Case - initialize the area and locks, and |
1504 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1505 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1506 | break; |
1507 | case CPU_ONLINE: | |
8bb78442 | 1508 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1509 | xfs_icsb_lock(mp); |
45af6c6d CH |
1510 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1511 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1512 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1513 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1514 | break; |
1515 | case CPU_DEAD: | |
8bb78442 | 1516 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1517 | /* Disable all the counters, then fold the dead cpu's |
1518 | * count into the total on the global superblock and | |
1519 | * re-enable the counters. */ | |
03135cf7 | 1520 | xfs_icsb_lock(mp); |
3685c2a1 | 1521 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1522 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
1523 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1524 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1525 | ||
1526 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1527 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1528 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1529 | ||
01e1b69c | 1530 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1531 | |
45af6c6d CH |
1532 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
1533 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
1534 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1535 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1536 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1537 | break; |
1538 | } | |
1539 | ||
1540 | return NOTIFY_OK; | |
1541 | } | |
5a67e4c5 | 1542 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1543 | |
8d280b98 DC |
1544 | int |
1545 | xfs_icsb_init_counters( | |
1546 | xfs_mount_t *mp) | |
1547 | { | |
1548 | xfs_icsb_cnts_t *cntp; | |
1549 | int i; | |
1550 | ||
1551 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1552 | if (mp->m_sb_cnts == NULL) | |
1553 | return -ENOMEM; | |
1554 | ||
5a67e4c5 | 1555 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1556 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
1557 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
1558 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
1559 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 1560 | |
8d280b98 DC |
1561 | for_each_online_cpu(i) { |
1562 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1563 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 1564 | } |
20b64285 DC |
1565 | |
1566 | mutex_init(&mp->m_icsb_mutex); | |
1567 | ||
8d280b98 DC |
1568 | /* |
1569 | * start with all counters disabled so that the | |
1570 | * initial balance kicks us off correctly | |
1571 | */ | |
1572 | mp->m_icsb_counters = -1; | |
1573 | return 0; | |
1574 | } | |
1575 | ||
5478eead LM |
1576 | void |
1577 | xfs_icsb_reinit_counters( | |
1578 | xfs_mount_t *mp) | |
1579 | { | |
1580 | xfs_icsb_lock(mp); | |
1581 | /* | |
1582 | * start with all counters disabled so that the | |
1583 | * initial balance kicks us off correctly | |
1584 | */ | |
1585 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
1586 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
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 | } |
03135cf7 | 1600 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
1601 | } |
1602 | ||
b8f82a4a | 1603 | STATIC void |
01e1b69c DC |
1604 | xfs_icsb_lock_cntr( |
1605 | xfs_icsb_cnts_t *icsbp) | |
1606 | { | |
1607 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1608 | ndelay(1000); | |
1609 | } | |
1610 | } | |
1611 | ||
b8f82a4a | 1612 | STATIC void |
01e1b69c DC |
1613 | xfs_icsb_unlock_cntr( |
1614 | xfs_icsb_cnts_t *icsbp) | |
1615 | { | |
1616 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1617 | } | |
1618 | ||
8d280b98 | 1619 | |
b8f82a4a | 1620 | STATIC void |
8d280b98 DC |
1621 | xfs_icsb_lock_all_counters( |
1622 | xfs_mount_t *mp) | |
1623 | { | |
1624 | xfs_icsb_cnts_t *cntp; | |
1625 | int i; | |
1626 | ||
1627 | for_each_online_cpu(i) { | |
1628 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1629 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1630 | } |
1631 | } | |
1632 | ||
b8f82a4a | 1633 | STATIC void |
8d280b98 DC |
1634 | xfs_icsb_unlock_all_counters( |
1635 | xfs_mount_t *mp) | |
1636 | { | |
1637 | xfs_icsb_cnts_t *cntp; | |
1638 | int i; | |
1639 | ||
1640 | for_each_online_cpu(i) { | |
1641 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1642 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1643 | } |
1644 | } | |
1645 | ||
1646 | STATIC void | |
1647 | xfs_icsb_count( | |
1648 | xfs_mount_t *mp, | |
1649 | xfs_icsb_cnts_t *cnt, | |
1650 | int flags) | |
1651 | { | |
1652 | xfs_icsb_cnts_t *cntp; | |
1653 | int i; | |
1654 | ||
1655 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1656 | ||
1657 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1658 | xfs_icsb_lock_all_counters(mp); | |
1659 | ||
1660 | for_each_online_cpu(i) { | |
1661 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
1662 | cnt->icsb_icount += cntp->icsb_icount; | |
1663 | cnt->icsb_ifree += cntp->icsb_ifree; | |
1664 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1665 | } | |
1666 | ||
1667 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1668 | xfs_icsb_unlock_all_counters(mp); | |
1669 | } | |
1670 | ||
1671 | STATIC int | |
1672 | xfs_icsb_counter_disabled( | |
1673 | xfs_mount_t *mp, | |
1674 | xfs_sb_field_t field) | |
1675 | { | |
1676 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1677 | return test_bit(field, &mp->m_icsb_counters); | |
1678 | } | |
1679 | ||
36fbe6e6 | 1680 | STATIC void |
8d280b98 DC |
1681 | xfs_icsb_disable_counter( |
1682 | xfs_mount_t *mp, | |
1683 | xfs_sb_field_t field) | |
1684 | { | |
1685 | xfs_icsb_cnts_t cnt; | |
1686 | ||
1687 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1688 | ||
20b64285 DC |
1689 | /* |
1690 | * If we are already disabled, then there is nothing to do | |
1691 | * here. We check before locking all the counters to avoid | |
1692 | * the expensive lock operation when being called in the | |
1693 | * slow path and the counter is already disabled. This is | |
1694 | * safe because the only time we set or clear this state is under | |
1695 | * the m_icsb_mutex. | |
1696 | */ | |
1697 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 1698 | return; |
20b64285 | 1699 | |
8d280b98 DC |
1700 | xfs_icsb_lock_all_counters(mp); |
1701 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1702 | /* drain back to superblock */ | |
1703 | ||
ce46193b | 1704 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
1705 | switch(field) { |
1706 | case XFS_SBS_ICOUNT: | |
1707 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1708 | break; | |
1709 | case XFS_SBS_IFREE: | |
1710 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1711 | break; | |
1712 | case XFS_SBS_FDBLOCKS: | |
1713 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1714 | break; | |
1715 | default: | |
1716 | BUG(); | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
1721 | } |
1722 | ||
1723 | STATIC void | |
1724 | xfs_icsb_enable_counter( | |
1725 | xfs_mount_t *mp, | |
1726 | xfs_sb_field_t field, | |
1727 | uint64_t count, | |
1728 | uint64_t resid) | |
1729 | { | |
1730 | xfs_icsb_cnts_t *cntp; | |
1731 | int i; | |
1732 | ||
1733 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1734 | ||
1735 | xfs_icsb_lock_all_counters(mp); | |
1736 | for_each_online_cpu(i) { | |
1737 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1738 | switch (field) { | |
1739 | case XFS_SBS_ICOUNT: | |
1740 | cntp->icsb_icount = count + resid; | |
1741 | break; | |
1742 | case XFS_SBS_IFREE: | |
1743 | cntp->icsb_ifree = count + resid; | |
1744 | break; | |
1745 | case XFS_SBS_FDBLOCKS: | |
1746 | cntp->icsb_fdblocks = count + resid; | |
1747 | break; | |
1748 | default: | |
1749 | BUG(); | |
1750 | break; | |
1751 | } | |
1752 | resid = 0; | |
1753 | } | |
1754 | clear_bit(field, &mp->m_icsb_counters); | |
1755 | xfs_icsb_unlock_all_counters(mp); | |
1756 | } | |
1757 | ||
dbcabad1 | 1758 | void |
d4d90b57 | 1759 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
1760 | xfs_mount_t *mp, |
1761 | int flags) | |
1762 | { | |
1763 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 1764 | |
8d280b98 DC |
1765 | xfs_icsb_count(mp, &cnt, flags); |
1766 | ||
8d280b98 DC |
1767 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
1768 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1769 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
1770 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1771 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1772 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * Accurate update of per-cpu counters to incore superblock | |
1777 | */ | |
d4d90b57 | 1778 | void |
8d280b98 | 1779 | xfs_icsb_sync_counters( |
d4d90b57 CH |
1780 | xfs_mount_t *mp, |
1781 | int flags) | |
8d280b98 | 1782 | { |
d4d90b57 CH |
1783 | spin_lock(&mp->m_sb_lock); |
1784 | xfs_icsb_sync_counters_locked(mp, flags); | |
1785 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1786 | } |
1787 | ||
1788 | /* | |
1789 | * Balance and enable/disable counters as necessary. | |
1790 | * | |
20b64285 DC |
1791 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
1792 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
1793 | * free blocks is something far enough zero that we aren't going thrash when we | |
1794 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
1795 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
1796 | * be distributed to a single CPU but each CPU has enough blocks to be | |
1797 | * reenabled. | |
1798 | * | |
1799 | * Note that we can be called when counters are already disabled. | |
1800 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
1801 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 1802 | */ |
20b64285 DC |
1803 | |
1804 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 1805 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 1806 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 1807 | STATIC void |
45af6c6d | 1808 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
1809 | xfs_mount_t *mp, |
1810 | xfs_sb_field_t field, | |
20b64285 | 1811 | int min_per_cpu) |
8d280b98 | 1812 | { |
6fdf8ccc | 1813 | uint64_t count, resid; |
8d280b98 | 1814 | int weight = num_online_cpus(); |
20b64285 | 1815 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 1816 | |
8d280b98 DC |
1817 | /* disable counter and sync counter */ |
1818 | xfs_icsb_disable_counter(mp, field); | |
1819 | ||
1820 | /* update counters - first CPU gets residual*/ | |
1821 | switch (field) { | |
1822 | case XFS_SBS_ICOUNT: | |
1823 | count = mp->m_sb.sb_icount; | |
1824 | resid = do_div(count, weight); | |
20b64285 | 1825 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1826 | return; |
8d280b98 DC |
1827 | break; |
1828 | case XFS_SBS_IFREE: | |
1829 | count = mp->m_sb.sb_ifree; | |
1830 | resid = do_div(count, weight); | |
20b64285 | 1831 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1832 | return; |
8d280b98 DC |
1833 | break; |
1834 | case XFS_SBS_FDBLOCKS: | |
1835 | count = mp->m_sb.sb_fdblocks; | |
1836 | resid = do_div(count, weight); | |
20b64285 | 1837 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 1838 | return; |
8d280b98 DC |
1839 | break; |
1840 | default: | |
1841 | BUG(); | |
6fdf8ccc | 1842 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
1843 | break; |
1844 | } | |
1845 | ||
1846 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
1847 | } |
1848 | ||
1849 | STATIC void | |
1850 | xfs_icsb_balance_counter( | |
1851 | xfs_mount_t *mp, | |
1852 | xfs_sb_field_t fields, | |
1853 | int min_per_cpu) | |
1854 | { | |
1855 | spin_lock(&mp->m_sb_lock); | |
1856 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
1857 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1858 | } |
1859 | ||
1b040712 | 1860 | int |
20b64285 | 1861 | xfs_icsb_modify_counters( |
8d280b98 DC |
1862 | xfs_mount_t *mp, |
1863 | xfs_sb_field_t field, | |
20f4ebf2 | 1864 | int64_t delta, |
20b64285 | 1865 | int rsvd) |
8d280b98 DC |
1866 | { |
1867 | xfs_icsb_cnts_t *icsbp; | |
1868 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 1869 | int ret = 0; |
8d280b98 | 1870 | |
20b64285 | 1871 | might_sleep(); |
8d280b98 | 1872 | again: |
7a9e02d6 CL |
1873 | preempt_disable(); |
1874 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
1875 | |
1876 | /* | |
1877 | * if the counter is disabled, go to slow path | |
1878 | */ | |
8d280b98 DC |
1879 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
1880 | goto slow_path; | |
20b64285 DC |
1881 | xfs_icsb_lock_cntr(icsbp); |
1882 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
1883 | xfs_icsb_unlock_cntr(icsbp); | |
1884 | goto slow_path; | |
1885 | } | |
8d280b98 DC |
1886 | |
1887 | switch (field) { | |
1888 | case XFS_SBS_ICOUNT: | |
1889 | lcounter = icsbp->icsb_icount; | |
1890 | lcounter += delta; | |
1891 | if (unlikely(lcounter < 0)) | |
20b64285 | 1892 | goto balance_counter; |
8d280b98 DC |
1893 | icsbp->icsb_icount = lcounter; |
1894 | break; | |
1895 | ||
1896 | case XFS_SBS_IFREE: | |
1897 | lcounter = icsbp->icsb_ifree; | |
1898 | lcounter += delta; | |
1899 | if (unlikely(lcounter < 0)) | |
20b64285 | 1900 | goto balance_counter; |
8d280b98 DC |
1901 | icsbp->icsb_ifree = lcounter; |
1902 | break; | |
1903 | ||
1904 | case XFS_SBS_FDBLOCKS: | |
1905 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
1906 | ||
4be536de | 1907 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1908 | lcounter += delta; |
1909 | if (unlikely(lcounter < 0)) | |
20b64285 | 1910 | goto balance_counter; |
4be536de | 1911 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1912 | break; |
1913 | default: | |
1914 | BUG(); | |
1915 | break; | |
1916 | } | |
01e1b69c | 1917 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 1918 | preempt_enable(); |
8d280b98 DC |
1919 | return 0; |
1920 | ||
8d280b98 | 1921 | slow_path: |
7a9e02d6 | 1922 | preempt_enable(); |
8d280b98 | 1923 | |
20b64285 DC |
1924 | /* |
1925 | * serialise with a mutex so we don't burn lots of cpu on | |
1926 | * the superblock lock. We still need to hold the superblock | |
1927 | * lock, however, when we modify the global structures. | |
1928 | */ | |
03135cf7 | 1929 | xfs_icsb_lock(mp); |
20b64285 DC |
1930 | |
1931 | /* | |
1932 | * Now running atomically. | |
1933 | * | |
1934 | * If the counter is enabled, someone has beaten us to rebalancing. | |
1935 | * Drop the lock and try again in the fast path.... | |
1936 | */ | |
1937 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 1938 | xfs_icsb_unlock(mp); |
8d280b98 | 1939 | goto again; |
8d280b98 DC |
1940 | } |
1941 | ||
20b64285 DC |
1942 | /* |
1943 | * The counter is currently disabled. Because we are | |
1944 | * running atomically here, we know a rebalance cannot | |
1945 | * be in progress. Hence we can go straight to operating | |
1946 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 1947 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 1948 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 1949 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
1950 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
1951 | * directly on the global counters. | |
1952 | */ | |
3685c2a1 | 1953 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1954 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1955 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 1956 | |
20b64285 DC |
1957 | /* |
1958 | * Now that we've modified the global superblock, we | |
1959 | * may be able to re-enable the distributed counters | |
1960 | * (e.g. lots of space just got freed). After that | |
1961 | * we are done. | |
1962 | */ | |
1963 | if (ret != ENOSPC) | |
45af6c6d | 1964 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 1965 | xfs_icsb_unlock(mp); |
8d280b98 | 1966 | return ret; |
8d280b98 | 1967 | |
20b64285 DC |
1968 | balance_counter: |
1969 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 1970 | preempt_enable(); |
8d280b98 | 1971 | |
20b64285 DC |
1972 | /* |
1973 | * We may have multiple threads here if multiple per-cpu | |
1974 | * counters run dry at the same time. This will mean we can | |
1975 | * do more balances than strictly necessary but it is not | |
1976 | * the common slowpath case. | |
1977 | */ | |
03135cf7 | 1978 | xfs_icsb_lock(mp); |
20b64285 DC |
1979 | |
1980 | /* | |
1981 | * running atomically. | |
1982 | * | |
1983 | * This will leave the counter in the correct state for future | |
1984 | * accesses. After the rebalance, we simply try again and our retry | |
1985 | * will either succeed through the fast path or slow path without | |
1986 | * another balance operation being required. | |
1987 | */ | |
45af6c6d | 1988 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 1989 | xfs_icsb_unlock(mp); |
20b64285 | 1990 | goto again; |
8d280b98 | 1991 | } |
20b64285 | 1992 | |
8d280b98 | 1993 | #endif |