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" | |
0b1b213f | 45 | #include "xfs_trace.h" |
6d8b79cf | 46 | #include "xfs_icache.h" |
04a1e6c5 DC |
47 | #include "xfs_cksum.h" |
48 | #include "xfs_buf_item.h" | |
0b1b213f | 49 | |
1da177e4 | 50 | |
8d280b98 | 51 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 52 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
53 | int); |
54 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
55 | int); | |
36fbe6e6 | 56 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
57 | #else |
58 | ||
45af6c6d CH |
59 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
60 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
61 | #endif |
62 | ||
27174203 CH |
63 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
64 | static int xfs_uuid_table_size; | |
65 | static uuid_t *xfs_uuid_table; | |
66 | ||
67 | /* | |
68 | * See if the UUID is unique among mounted XFS filesystems. | |
69 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
70 | */ | |
71 | STATIC int | |
72 | xfs_uuid_mount( | |
73 | struct xfs_mount *mp) | |
74 | { | |
75 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
76 | int hole, i; | |
77 | ||
78 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
79 | return 0; | |
80 | ||
81 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 82 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
83 | return XFS_ERROR(EINVAL); |
84 | } | |
85 | ||
86 | mutex_lock(&xfs_uuid_table_mutex); | |
87 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
88 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
89 | hole = i; | |
90 | continue; | |
91 | } | |
92 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
93 | goto out_duplicate; | |
94 | } | |
95 | ||
96 | if (hole < 0) { | |
97 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
98 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
99 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
100 | KM_SLEEP); | |
101 | hole = xfs_uuid_table_size++; | |
102 | } | |
103 | xfs_uuid_table[hole] = *uuid; | |
104 | mutex_unlock(&xfs_uuid_table_mutex); | |
105 | ||
106 | return 0; | |
107 | ||
108 | out_duplicate: | |
109 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 110 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
27174203 CH |
111 | return XFS_ERROR(EINVAL); |
112 | } | |
113 | ||
114 | STATIC void | |
115 | xfs_uuid_unmount( | |
116 | struct xfs_mount *mp) | |
117 | { | |
118 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
119 | int i; | |
120 | ||
121 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
122 | return; | |
123 | ||
124 | mutex_lock(&xfs_uuid_table_mutex); | |
125 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
126 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
127 | continue; | |
128 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
129 | continue; | |
130 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
131 | break; | |
132 | } | |
133 | ASSERT(i < xfs_uuid_table_size); | |
134 | mutex_unlock(&xfs_uuid_table_mutex); | |
135 | } | |
136 | ||
137 | ||
e176579e DC |
138 | STATIC void |
139 | __xfs_free_perag( | |
140 | struct rcu_head *head) | |
141 | { | |
142 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
143 | ||
144 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
145 | kmem_free(pag); | |
146 | } | |
147 | ||
1da177e4 | 148 | /* |
e176579e | 149 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 150 | */ |
c962fb79 | 151 | STATIC void |
ff4f038c | 152 | xfs_free_perag( |
745f6919 | 153 | xfs_mount_t *mp) |
1da177e4 | 154 | { |
1c1c6ebc DC |
155 | xfs_agnumber_t agno; |
156 | struct xfs_perag *pag; | |
157 | ||
158 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
159 | spin_lock(&mp->m_perag_lock); | |
160 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
161 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 162 | ASSERT(pag); |
f83282a8 | 163 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 164 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 165 | } |
1da177e4 LT |
166 | } |
167 | ||
4cc929ee NS |
168 | /* |
169 | * Check size of device based on the (data/realtime) block count. | |
170 | * Note: this check is used by the growfs code as well as mount. | |
171 | */ | |
172 | int | |
173 | xfs_sb_validate_fsb_count( | |
174 | xfs_sb_t *sbp, | |
175 | __uint64_t nblocks) | |
176 | { | |
177 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
178 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
179 | ||
180 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
181 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 182 | return EFBIG; |
4cc929ee NS |
183 | #else /* Limited by UINT_MAX of sectors */ |
184 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 185 | return EFBIG; |
4cc929ee NS |
186 | #endif |
187 | return 0; | |
188 | } | |
1da177e4 | 189 | |
1c1c6ebc | 190 | int |
c11e2c36 | 191 | xfs_initialize_perag( |
c11e2c36 | 192 | xfs_mount_t *mp, |
1c1c6ebc DC |
193 | xfs_agnumber_t agcount, |
194 | xfs_agnumber_t *maxagi) | |
1da177e4 | 195 | { |
2d2194f6 | 196 | xfs_agnumber_t index; |
8b26c582 | 197 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
198 | xfs_perag_t *pag; |
199 | xfs_agino_t agino; | |
200 | xfs_ino_t ino; | |
201 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 202 | int error = -ENOMEM; |
1da177e4 | 203 | |
1c1c6ebc DC |
204 | /* |
205 | * Walk the current per-ag tree so we don't try to initialise AGs | |
206 | * that already exist (growfs case). Allocate and insert all the | |
207 | * AGs we don't find ready for initialisation. | |
208 | */ | |
209 | for (index = 0; index < agcount; index++) { | |
210 | pag = xfs_perag_get(mp, index); | |
211 | if (pag) { | |
212 | xfs_perag_put(pag); | |
213 | continue; | |
214 | } | |
8b26c582 DC |
215 | if (!first_initialised) |
216 | first_initialised = index; | |
fb3b504a | 217 | |
1c1c6ebc DC |
218 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
219 | if (!pag) | |
8b26c582 | 220 | goto out_unwind; |
fb3b504a CH |
221 | pag->pag_agno = index; |
222 | pag->pag_mount = mp; | |
1a427ab0 | 223 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 224 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 225 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
226 | spin_lock_init(&pag->pag_buf_lock); |
227 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 228 | |
1c1c6ebc | 229 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 230 | goto out_unwind; |
fb3b504a | 231 | |
1c1c6ebc DC |
232 | spin_lock(&mp->m_perag_lock); |
233 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
234 | BUG(); | |
235 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
236 | radix_tree_preload_end(); |
237 | error = -EEXIST; | |
238 | goto out_unwind; | |
1c1c6ebc DC |
239 | } |
240 | spin_unlock(&mp->m_perag_lock); | |
241 | radix_tree_preload_end(); | |
242 | } | |
243 | ||
fb3b504a CH |
244 | /* |
245 | * If we mount with the inode64 option, or no inode overflows | |
246 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 247 | */ |
fb3b504a CH |
248 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
249 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
250 | ||
251 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 252 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 253 | else |
1da177e4 | 254 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 255 | |
2d2194f6 CM |
256 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
257 | index = xfs_set_inode32(mp); | |
258 | else | |
259 | index = xfs_set_inode64(mp); | |
fb3b504a | 260 | |
1c1c6ebc DC |
261 | if (maxagi) |
262 | *maxagi = index; | |
263 | return 0; | |
8b26c582 DC |
264 | |
265 | out_unwind: | |
266 | kmem_free(pag); | |
267 | for (; index > first_initialised; index--) { | |
268 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
269 | kmem_free(pag); | |
270 | } | |
271 | return error; | |
1da177e4 LT |
272 | } |
273 | ||
1da177e4 LT |
274 | /* |
275 | * xfs_readsb | |
276 | * | |
277 | * Does the initial read of the superblock. | |
278 | */ | |
279 | int | |
ff55068c DC |
280 | xfs_readsb( |
281 | struct xfs_mount *mp, | |
282 | int flags) | |
1da177e4 LT |
283 | { |
284 | unsigned int sector_size; | |
04a1e6c5 DC |
285 | struct xfs_buf *bp; |
286 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 287 | int error; |
af34e09d | 288 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
289 | |
290 | ASSERT(mp->m_sb_bp == NULL); | |
291 | ASSERT(mp->m_ddev_targp != NULL); | |
292 | ||
293 | /* | |
294 | * Allocate a (locked) buffer to hold the superblock. | |
295 | * This will be kept around at all times to optimize | |
296 | * access to the superblock. | |
297 | */ | |
298 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
299 | |
300 | reread: | |
e70b73f8 | 301 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
98021821 | 302 | BTOBB(sector_size), 0, |
1813dd64 DC |
303 | loud ? &xfs_sb_buf_ops |
304 | : &xfs_sb_quiet_buf_ops); | |
26af6552 | 305 | if (!bp) { |
af34e09d DC |
306 | if (loud) |
307 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 308 | return EIO; |
1da177e4 | 309 | } |
eab4e633 DC |
310 | if (bp->b_error) { |
311 | error = bp->b_error; | |
312 | if (loud) | |
e721f504 | 313 | xfs_warn(mp, "SB validate failed with error %d.", error); |
eab4e633 DC |
314 | goto release_buf; |
315 | } | |
1da177e4 LT |
316 | |
317 | /* | |
318 | * Initialize the mount structure from the superblock. | |
1da177e4 | 319 | */ |
98021821 | 320 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
83e782e1 | 321 | xfs_sb_quota_from_disk(&mp->m_sb); |
ff55068c | 322 | |
1da177e4 LT |
323 | /* |
324 | * We must be able to do sector-sized and sector-aligned IO. | |
325 | */ | |
04a1e6c5 | 326 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
327 | if (loud) |
328 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 329 | sector_size, sbp->sb_sectsize); |
1da177e4 | 330 | error = ENOSYS; |
26af6552 | 331 | goto release_buf; |
1da177e4 LT |
332 | } |
333 | ||
334 | /* | |
335 | * If device sector size is smaller than the superblock size, | |
336 | * re-read the superblock so the buffer is correctly sized. | |
337 | */ | |
04a1e6c5 | 338 | if (sector_size < sbp->sb_sectsize) { |
1da177e4 | 339 | xfs_buf_relse(bp); |
04a1e6c5 | 340 | sector_size = sbp->sb_sectsize; |
26af6552 | 341 | goto reread; |
1da177e4 LT |
342 | } |
343 | ||
5478eead LM |
344 | /* Initialize per-cpu counters */ |
345 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 346 | |
04a1e6c5 DC |
347 | /* no need to be quiet anymore, so reset the buf ops */ |
348 | bp->b_ops = &xfs_sb_buf_ops; | |
349 | ||
1da177e4 | 350 | mp->m_sb_bp = bp; |
26af6552 | 351 | xfs_buf_unlock(bp); |
1da177e4 LT |
352 | return 0; |
353 | ||
26af6552 DC |
354 | release_buf: |
355 | xfs_buf_relse(bp); | |
1da177e4 LT |
356 | return error; |
357 | } | |
358 | ||
1da177e4 | 359 | /* |
0771fb45 | 360 | * Update alignment values based on mount options and sb values |
1da177e4 | 361 | */ |
0771fb45 | 362 | STATIC int |
7884bc86 | 363 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 364 | { |
1da177e4 | 365 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 366 | |
4249023a | 367 | if (mp->m_dalign) { |
1da177e4 LT |
368 | /* |
369 | * If stripe unit and stripe width are not multiples | |
370 | * of the fs blocksize turn off alignment. | |
371 | */ | |
372 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
373 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
374 | xfs_warn(mp, |
375 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
376 | sbp->sb_blocksize); | |
377 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
378 | } else { |
379 | /* | |
380 | * Convert the stripe unit and width to FSBs. | |
381 | */ | |
382 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
383 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 384 | xfs_warn(mp, |
39a45d84 JL |
385 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
386 | sbp->sb_agblocks); | |
387 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
388 | } else if (mp->m_dalign) { |
389 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
390 | } else { | |
39a45d84 JL |
391 | xfs_warn(mp, |
392 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
393 | mp->m_dalign, sbp->sb_blocksize); | |
394 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
395 | } |
396 | } | |
397 | ||
398 | /* | |
399 | * Update superblock with new values | |
400 | * and log changes | |
401 | */ | |
62118709 | 402 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
403 | if (sbp->sb_unit != mp->m_dalign) { |
404 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 405 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
406 | } |
407 | if (sbp->sb_width != mp->m_swidth) { | |
408 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 409 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 | 410 | } |
34d7f603 JL |
411 | } else { |
412 | xfs_warn(mp, | |
413 | "cannot change alignment: superblock does not support data alignment"); | |
414 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
415 | } |
416 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 417 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
418 | mp->m_dalign = sbp->sb_unit; |
419 | mp->m_swidth = sbp->sb_width; | |
420 | } | |
421 | ||
0771fb45 ES |
422 | return 0; |
423 | } | |
1da177e4 | 424 | |
0771fb45 ES |
425 | /* |
426 | * Set the maximum inode count for this filesystem | |
427 | */ | |
428 | STATIC void | |
429 | xfs_set_maxicount(xfs_mount_t *mp) | |
430 | { | |
431 | xfs_sb_t *sbp = &(mp->m_sb); | |
432 | __uint64_t icount; | |
1da177e4 | 433 | |
0771fb45 ES |
434 | if (sbp->sb_imax_pct) { |
435 | /* | |
436 | * Make sure the maximum inode count is a multiple | |
437 | * of the units we allocate inodes in. | |
1da177e4 | 438 | */ |
1da177e4 LT |
439 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
440 | do_div(icount, 100); | |
441 | do_div(icount, mp->m_ialloc_blks); | |
442 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
443 | sbp->sb_inopblog; | |
0771fb45 | 444 | } else { |
1da177e4 | 445 | mp->m_maxicount = 0; |
1da177e4 | 446 | } |
0771fb45 ES |
447 | } |
448 | ||
449 | /* | |
450 | * Set the default minimum read and write sizes unless | |
451 | * already specified in a mount option. | |
452 | * We use smaller I/O sizes when the file system | |
453 | * is being used for NFS service (wsync mount option). | |
454 | */ | |
455 | STATIC void | |
456 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
457 | { | |
458 | xfs_sb_t *sbp = &(mp->m_sb); | |
459 | int readio_log, writeio_log; | |
1da177e4 | 460 | |
1da177e4 LT |
461 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
462 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
463 | readio_log = XFS_WSYNC_READIO_LOG; | |
464 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
465 | } else { | |
466 | readio_log = XFS_READIO_LOG_LARGE; | |
467 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
468 | } | |
469 | } else { | |
470 | readio_log = mp->m_readio_log; | |
471 | writeio_log = mp->m_writeio_log; | |
472 | } | |
473 | ||
1da177e4 LT |
474 | if (sbp->sb_blocklog > readio_log) { |
475 | mp->m_readio_log = sbp->sb_blocklog; | |
476 | } else { | |
477 | mp->m_readio_log = readio_log; | |
478 | } | |
479 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
480 | if (sbp->sb_blocklog > writeio_log) { | |
481 | mp->m_writeio_log = sbp->sb_blocklog; | |
482 | } else { | |
483 | mp->m_writeio_log = writeio_log; | |
484 | } | |
485 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 486 | } |
1da177e4 | 487 | |
055388a3 DC |
488 | /* |
489 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
490 | */ | |
491 | void | |
492 | xfs_set_low_space_thresholds( | |
493 | struct xfs_mount *mp) | |
494 | { | |
495 | int i; | |
496 | ||
497 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
498 | __uint64_t space = mp->m_sb.sb_dblocks; | |
499 | ||
500 | do_div(space, 100); | |
501 | mp->m_low_space[i] = space * (i + 1); | |
502 | } | |
503 | } | |
504 | ||
505 | ||
0771fb45 ES |
506 | /* |
507 | * Set whether we're using inode alignment. | |
508 | */ | |
509 | STATIC void | |
510 | xfs_set_inoalignment(xfs_mount_t *mp) | |
511 | { | |
62118709 | 512 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
513 | mp->m_sb.sb_inoalignmt >= |
514 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
515 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
516 | else | |
517 | mp->m_inoalign_mask = 0; | |
518 | /* | |
519 | * If we are using stripe alignment, check whether | |
520 | * the stripe unit is a multiple of the inode alignment | |
521 | */ | |
522 | if (mp->m_dalign && mp->m_inoalign_mask && | |
523 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
524 | mp->m_sinoalign = mp->m_dalign; | |
525 | else | |
526 | mp->m_sinoalign = 0; | |
0771fb45 ES |
527 | } |
528 | ||
529 | /* | |
0471f62e | 530 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
531 | */ |
532 | STATIC int | |
4249023a | 533 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
534 | { |
535 | xfs_buf_t *bp; | |
536 | xfs_daddr_t d; | |
0771fb45 | 537 | |
1da177e4 LT |
538 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
539 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 540 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 541 | return XFS_ERROR(EFBIG); |
1da177e4 | 542 | } |
e70b73f8 | 543 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 544 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 545 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 546 | if (!bp) { |
0b932ccc | 547 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 548 | return EIO; |
1da177e4 | 549 | } |
1922c949 | 550 | xfs_buf_relse(bp); |
1da177e4 | 551 | |
4249023a | 552 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
553 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
554 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 555 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 556 | return XFS_ERROR(EFBIG); |
1da177e4 | 557 | } |
e70b73f8 | 558 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 559 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 560 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 561 | if (!bp) { |
0b932ccc | 562 | xfs_warn(mp, "log device read failed"); |
1922c949 | 563 | return EIO; |
0771fb45 | 564 | } |
1922c949 | 565 | xfs_buf_relse(bp); |
0771fb45 ES |
566 | } |
567 | return 0; | |
568 | } | |
569 | ||
7d095257 CH |
570 | /* |
571 | * Clear the quotaflags in memory and in the superblock. | |
572 | */ | |
573 | int | |
574 | xfs_mount_reset_sbqflags( | |
575 | struct xfs_mount *mp) | |
576 | { | |
577 | int error; | |
578 | struct xfs_trans *tp; | |
579 | ||
580 | mp->m_qflags = 0; | |
581 | ||
582 | /* | |
583 | * It is OK to look at sb_qflags here in mount path, | |
584 | * without m_sb_lock. | |
585 | */ | |
586 | if (mp->m_sb.sb_qflags == 0) | |
587 | return 0; | |
588 | spin_lock(&mp->m_sb_lock); | |
589 | mp->m_sb.sb_qflags = 0; | |
590 | spin_unlock(&mp->m_sb_lock); | |
591 | ||
592 | /* | |
593 | * If the fs is readonly, let the incore superblock run | |
594 | * with quotas off but don't flush the update out to disk | |
595 | */ | |
596 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
597 | return 0; | |
598 | ||
7d095257 | 599 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
3d3c8b52 | 600 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0); |
7d095257 CH |
601 | if (error) { |
602 | xfs_trans_cancel(tp, 0); | |
53487786 | 603 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
604 | return error; |
605 | } | |
606 | ||
607 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
608 | return xfs_trans_commit(tp, 0); | |
609 | } | |
610 | ||
d5db0f97 ES |
611 | __uint64_t |
612 | xfs_default_resblks(xfs_mount_t *mp) | |
613 | { | |
614 | __uint64_t resblks; | |
615 | ||
616 | /* | |
8babd8a2 DC |
617 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
618 | * smaller. This is intended to cover concurrent allocation | |
619 | * transactions when we initially hit enospc. These each require a 4 | |
620 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
621 | * allocation reservations. | |
d5db0f97 ES |
622 | */ |
623 | resblks = mp->m_sb.sb_dblocks; | |
624 | do_div(resblks, 20); | |
8babd8a2 | 625 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
626 | return resblks; |
627 | } | |
628 | ||
0771fb45 | 629 | /* |
0771fb45 ES |
630 | * This function does the following on an initial mount of a file system: |
631 | * - reads the superblock from disk and init the mount struct | |
632 | * - if we're a 32-bit kernel, do a size check on the superblock | |
633 | * so we don't mount terabyte filesystems | |
634 | * - init mount struct realtime fields | |
635 | * - allocate inode hash table for fs | |
636 | * - init directory manager | |
637 | * - perform recovery and init the log manager | |
638 | */ | |
639 | int | |
640 | xfs_mountfs( | |
4249023a | 641 | xfs_mount_t *mp) |
0771fb45 ES |
642 | { |
643 | xfs_sb_t *sbp = &(mp->m_sb); | |
644 | xfs_inode_t *rip; | |
0771fb45 | 645 | __uint64_t resblks; |
7d095257 CH |
646 | uint quotamount = 0; |
647 | uint quotaflags = 0; | |
0771fb45 ES |
648 | int error = 0; |
649 | ||
ff55068c | 650 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 651 | |
ee1c0908 | 652 | /* |
e6957ea4 ES |
653 | * Check for a mismatched features2 values. Older kernels |
654 | * read & wrote into the wrong sb offset for sb_features2 | |
655 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
656 | * when sb_features2 was added, which made older superblock | |
657 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 658 | * |
e6957ea4 ES |
659 | * For backwards compatibility, we make both slots equal. |
660 | * | |
661 | * If we detect a mismatched field, we OR the set bits into the | |
662 | * existing features2 field in case it has already been modified; we | |
663 | * don't want to lose any features. We then update the bad location | |
664 | * with the ORed value so that older kernels will see any features2 | |
665 | * flags, and mark the two fields as needing updates once the | |
666 | * transaction subsystem is online. | |
ee1c0908 | 667 | */ |
e6957ea4 | 668 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 669 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 670 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 671 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 672 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
673 | |
674 | /* | |
675 | * Re-check for ATTR2 in case it was found in bad_features2 | |
676 | * slot. | |
677 | */ | |
7c12f296 TS |
678 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
679 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 680 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
681 | } |
682 | ||
683 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
684 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
685 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 686 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 687 | |
7c12f296 TS |
688 | /* update sb_versionnum for the clearing of the morebits */ |
689 | if (!sbp->sb_features2) | |
7884bc86 | 690 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
691 | } |
692 | ||
0771fb45 ES |
693 | /* |
694 | * Check if sb_agblocks is aligned at stripe boundary | |
695 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
696 | * allocator alignment is within an ag, therefore ag has | |
697 | * to be aligned at stripe boundary. | |
698 | */ | |
7884bc86 | 699 | error = xfs_update_alignment(mp); |
0771fb45 | 700 | if (error) |
f9057e3d | 701 | goto out; |
0771fb45 ES |
702 | |
703 | xfs_alloc_compute_maxlevels(mp); | |
704 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
705 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
706 | xfs_ialloc_compute_maxlevels(mp); | |
707 | ||
708 | xfs_set_maxicount(mp); | |
709 | ||
27174203 CH |
710 | error = xfs_uuid_mount(mp); |
711 | if (error) | |
712 | goto out; | |
1da177e4 | 713 | |
0771fb45 ES |
714 | /* |
715 | * Set the minimum read and write sizes | |
716 | */ | |
717 | xfs_set_rw_sizes(mp); | |
718 | ||
055388a3 DC |
719 | /* set the low space thresholds for dynamic preallocation */ |
720 | xfs_set_low_space_thresholds(mp); | |
721 | ||
0771fb45 ES |
722 | /* |
723 | * Set the inode cluster size. | |
724 | * This may still be overridden by the file system | |
725 | * block size if it is larger than the chosen cluster size. | |
726 | */ | |
727 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
728 | ||
729 | /* | |
730 | * Set inode alignment fields | |
731 | */ | |
732 | xfs_set_inoalignment(mp); | |
733 | ||
734 | /* | |
735 | * Check that the data (and log if separate) are an ok size. | |
736 | */ | |
4249023a | 737 | error = xfs_check_sizes(mp); |
0771fb45 | 738 | if (error) |
f9057e3d | 739 | goto out_remove_uuid; |
0771fb45 | 740 | |
1da177e4 LT |
741 | /* |
742 | * Initialize realtime fields in the mount structure | |
743 | */ | |
0771fb45 ES |
744 | error = xfs_rtmount_init(mp); |
745 | if (error) { | |
0b932ccc | 746 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 747 | goto out_remove_uuid; |
1da177e4 LT |
748 | } |
749 | ||
1da177e4 LT |
750 | /* |
751 | * Copies the low order bits of the timestamp and the randomly | |
752 | * set "sequence" number out of a UUID. | |
753 | */ | |
754 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
755 | ||
1da177e4 LT |
756 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
757 | ||
f6c2d1fa | 758 | xfs_dir_mount(mp); |
1da177e4 LT |
759 | |
760 | /* | |
761 | * Initialize the attribute manager's entries. | |
762 | */ | |
763 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
764 | ||
765 | /* | |
766 | * Initialize the precomputed transaction reservations values. | |
767 | */ | |
768 | xfs_trans_init(mp); | |
769 | ||
1da177e4 LT |
770 | /* |
771 | * Allocate and initialize the per-ag data. | |
772 | */ | |
1c1c6ebc | 773 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 774 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
775 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
776 | if (error) { | |
0b932ccc | 777 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 778 | goto out_remove_uuid; |
1c1c6ebc | 779 | } |
1da177e4 | 780 | |
f9057e3d | 781 | if (!sbp->sb_logblocks) { |
0b932ccc | 782 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
783 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
784 | error = XFS_ERROR(EFSCORRUPTED); | |
785 | goto out_free_perag; | |
786 | } | |
787 | ||
1da177e4 LT |
788 | /* |
789 | * log's mount-time initialization. Perform 1st part recovery if needed | |
790 | */ | |
f9057e3d CH |
791 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
792 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
793 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
794 | if (error) { | |
0b932ccc | 795 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 796 | goto out_fail_wait; |
1da177e4 LT |
797 | } |
798 | ||
92821e2b DC |
799 | /* |
800 | * Now the log is mounted, we know if it was an unclean shutdown or | |
801 | * not. If it was, with the first phase of recovery has completed, we | |
802 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
803 | * but they are recovered transactionally in the second recovery phase | |
804 | * later. | |
805 | * | |
806 | * Hence we can safely re-initialise incore superblock counters from | |
807 | * the per-ag data. These may not be correct if the filesystem was not | |
808 | * cleanly unmounted, so we need to wait for recovery to finish before | |
809 | * doing this. | |
810 | * | |
811 | * If the filesystem was cleanly unmounted, then we can trust the | |
812 | * values in the superblock to be correct and we don't need to do | |
813 | * anything here. | |
814 | * | |
815 | * If we are currently making the filesystem, the initialisation will | |
816 | * fail as the perag data is in an undefined state. | |
817 | */ | |
92821e2b DC |
818 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
819 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
820 | !mp->m_sb.sb_inprogress) { | |
821 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 822 | if (error) |
d4f3512b | 823 | goto out_fail_wait; |
92821e2b | 824 | } |
f9057e3d | 825 | |
1da177e4 LT |
826 | /* |
827 | * Get and sanity-check the root inode. | |
828 | * Save the pointer to it in the mount structure. | |
829 | */ | |
7b6259e7 | 830 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 831 | if (error) { |
0b932ccc | 832 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 833 | goto out_log_dealloc; |
1da177e4 LT |
834 | } |
835 | ||
836 | ASSERT(rip != NULL); | |
1da177e4 | 837 | |
abbede1b | 838 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 839 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 840 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
841 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
842 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
843 | mp); | |
844 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 845 | goto out_rele_rip; |
1da177e4 LT |
846 | } |
847 | mp->m_rootip = rip; /* save it */ | |
848 | ||
849 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
850 | ||
851 | /* | |
852 | * Initialize realtime inode pointers in the mount structure | |
853 | */ | |
0771fb45 ES |
854 | error = xfs_rtmount_inodes(mp); |
855 | if (error) { | |
1da177e4 LT |
856 | /* |
857 | * Free up the root inode. | |
858 | */ | |
0b932ccc | 859 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 860 | goto out_rele_rip; |
1da177e4 LT |
861 | } |
862 | ||
863 | /* | |
7884bc86 CH |
864 | * If this is a read-only mount defer the superblock updates until |
865 | * the next remount into writeable mode. Otherwise we would never | |
866 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 867 | */ |
7884bc86 CH |
868 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
869 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 870 | if (error) { |
0b932ccc | 871 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 872 | goto out_rtunmount; |
e5720eec DC |
873 | } |
874 | } | |
1da177e4 LT |
875 | |
876 | /* | |
877 | * Initialise the XFS quota management subsystem for this mount | |
878 | */ | |
7d095257 CH |
879 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
880 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
881 | if (error) | |
882 | goto out_rtunmount; | |
883 | } else { | |
884 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
885 | ||
886 | /* | |
887 | * If a file system had quotas running earlier, but decided to | |
888 | * mount without -o uquota/pquota/gquota options, revoke the | |
889 | * quotachecked license. | |
890 | */ | |
891 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 892 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
893 | error = xfs_mount_reset_sbqflags(mp); |
894 | if (error) | |
895 | return error; | |
896 | } | |
897 | } | |
1da177e4 LT |
898 | |
899 | /* | |
900 | * Finish recovering the file system. This part needed to be | |
901 | * delayed until after the root and real-time bitmap inodes | |
902 | * were consistently read in. | |
903 | */ | |
4249023a | 904 | error = xfs_log_mount_finish(mp); |
1da177e4 | 905 | if (error) { |
0b932ccc | 906 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 907 | goto out_rtunmount; |
1da177e4 LT |
908 | } |
909 | ||
910 | /* | |
911 | * Complete the quota initialisation, post-log-replay component. | |
912 | */ | |
7d095257 CH |
913 | if (quotamount) { |
914 | ASSERT(mp->m_qflags == 0); | |
915 | mp->m_qflags = quotaflags; | |
916 | ||
917 | xfs_qm_mount_quotas(mp); | |
918 | } | |
919 | ||
84e1e99f DC |
920 | /* |
921 | * Now we are mounted, reserve a small amount of unused space for | |
922 | * privileged transactions. This is needed so that transaction | |
923 | * space required for critical operations can dip into this pool | |
924 | * when at ENOSPC. This is needed for operations like create with | |
925 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
926 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
927 | * |
928 | * This may drive us straight to ENOSPC on mount, but that implies | |
929 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 930 | */ |
d5db0f97 ES |
931 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
932 | resblks = xfs_default_resblks(mp); | |
933 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
934 | if (error) | |
0b932ccc DC |
935 | xfs_warn(mp, |
936 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 937 | } |
84e1e99f | 938 | |
1da177e4 LT |
939 | return 0; |
940 | ||
b93b6e43 CH |
941 | out_rtunmount: |
942 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 943 | out_rele_rip: |
43355099 | 944 | IRELE(rip); |
f9057e3d | 945 | out_log_dealloc: |
21b699c8 | 946 | xfs_log_unmount(mp); |
d4f3512b DC |
947 | out_fail_wait: |
948 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
949 | xfs_wait_buftarg(mp->m_logdev_targp); | |
950 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 951 | out_free_perag: |
ff4f038c | 952 | xfs_free_perag(mp); |
f9057e3d | 953 | out_remove_uuid: |
27174203 | 954 | xfs_uuid_unmount(mp); |
f9057e3d | 955 | out: |
1da177e4 LT |
956 | return error; |
957 | } | |
958 | ||
959 | /* | |
1da177e4 LT |
960 | * This flushes out the inodes,dquots and the superblock, unmounts the |
961 | * log and makes sure that incore structures are freed. | |
962 | */ | |
41b5c2e7 CH |
963 | void |
964 | xfs_unmountfs( | |
965 | struct xfs_mount *mp) | |
1da177e4 | 966 | { |
41b5c2e7 CH |
967 | __uint64_t resblks; |
968 | int error; | |
1da177e4 | 969 | |
579b62fa BF |
970 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
971 | ||
7d095257 | 972 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 973 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
974 | IRELE(mp->m_rootip); |
975 | ||
641c56fb DC |
976 | /* |
977 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 978 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
979 | * the transaction is still sitting in a iclog. The stale inodes |
980 | * on that buffer will have their flush locks held until the | |
981 | * transaction hits the disk and the callbacks run. the inode | |
982 | * flush takes the flush lock unconditionally and with nothing to | |
983 | * push out the iclog we will never get that unlocked. hence we | |
984 | * need to force the log first. | |
985 | */ | |
a14a348b | 986 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
987 | |
988 | /* | |
211e4d43 CH |
989 | * Flush all pending changes from the AIL. |
990 | */ | |
991 | xfs_ail_push_all_sync(mp->m_ail); | |
992 | ||
993 | /* | |
994 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
995 | * inodes and none should be pinned or locked, but use synchronous |
996 | * reclaim just to be sure. We can stop background inode reclaim | |
997 | * here as well if it is still running. | |
c854363e | 998 | */ |
7e18530b | 999 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1000 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1001 | |
7d095257 | 1002 | xfs_qm_unmount(mp); |
a357a121 | 1003 | |
84e1e99f DC |
1004 | /* |
1005 | * Unreserve any blocks we have so that when we unmount we don't account | |
1006 | * the reserved free space as used. This is really only necessary for | |
1007 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1008 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1009 | * |
1010 | * We don't bother correcting this elsewhere for lazy superblock | |
1011 | * counting because on mount of an unclean filesystem we reconstruct the | |
1012 | * correct counter value and this is irrelevant. | |
1013 | * | |
1014 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1015 | * we only every apply deltas to the superblock and hence the incore | |
1016 | * value does not matter.... | |
1017 | */ | |
1018 | resblks = 0; | |
714082bc DC |
1019 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1020 | if (error) | |
0b932ccc | 1021 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1022 | "Freespace may not be correct on next mount."); |
1023 | ||
adab0f67 | 1024 | error = xfs_log_sbcount(mp); |
e5720eec | 1025 | if (error) |
0b932ccc | 1026 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1027 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1028 | |
21b699c8 | 1029 | xfs_log_unmount(mp); |
27174203 | 1030 | xfs_uuid_unmount(mp); |
1da177e4 | 1031 | |
1550d0b0 | 1032 | #if defined(DEBUG) |
0ce4cfd4 | 1033 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1034 | #endif |
ff4f038c | 1035 | xfs_free_perag(mp); |
1da177e4 LT |
1036 | } |
1037 | ||
92821e2b DC |
1038 | int |
1039 | xfs_fs_writable(xfs_mount_t *mp) | |
1040 | { | |
d9457dc0 | 1041 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1042 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1043 | } |
1044 | ||
1045 | /* | |
b2ce3974 AE |
1046 | * xfs_log_sbcount |
1047 | * | |
adab0f67 | 1048 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1049 | * |
1050 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1051 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1052 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1053 | */ |
1054 | int | |
adab0f67 | 1055 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1056 | { |
1057 | xfs_trans_t *tp; | |
1058 | int error; | |
1059 | ||
1060 | if (!xfs_fs_writable(mp)) | |
1061 | return 0; | |
1062 | ||
d4d90b57 | 1063 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1064 | |
1065 | /* | |
1066 | * we don't need to do this if we are updating the superblock | |
1067 | * counters on every modification. | |
1068 | */ | |
1069 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1070 | return 0; | |
1071 | ||
b2ce3974 | 1072 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
3d3c8b52 | 1073 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
92821e2b DC |
1074 | if (error) { |
1075 | xfs_trans_cancel(tp, 0); | |
1076 | return error; | |
1077 | } | |
1078 | ||
1079 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1080 | xfs_trans_set_sync(tp); |
e5720eec DC |
1081 | error = xfs_trans_commit(tp, 0); |
1082 | return error; | |
92821e2b DC |
1083 | } |
1084 | ||
1da177e4 | 1085 | /* |
99e738b7 | 1086 | * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply |
1da177e4 LT |
1087 | * a delta to a specified field in the in-core superblock. Simply |
1088 | * switch on the field indicated and apply the delta to that field. | |
1089 | * Fields are not allowed to dip below zero, so if the delta would | |
1090 | * do this do not apply it and return EINVAL. | |
1091 | * | |
3685c2a1 | 1092 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1093 | */ |
d96f8f89 | 1094 | STATIC int |
20f4ebf2 DC |
1095 | xfs_mod_incore_sb_unlocked( |
1096 | xfs_mount_t *mp, | |
1097 | xfs_sb_field_t field, | |
1098 | int64_t delta, | |
1099 | int rsvd) | |
1da177e4 LT |
1100 | { |
1101 | int scounter; /* short counter for 32 bit fields */ | |
1102 | long long lcounter; /* long counter for 64 bit fields */ | |
1103 | long long res_used, rem; | |
1104 | ||
1105 | /* | |
1106 | * With the in-core superblock spin lock held, switch | |
1107 | * on the indicated field. Apply the delta to the | |
1108 | * proper field. If the fields value would dip below | |
1109 | * 0, then do not apply the delta and return EINVAL. | |
1110 | */ | |
1111 | switch (field) { | |
1112 | case XFS_SBS_ICOUNT: | |
1113 | lcounter = (long long)mp->m_sb.sb_icount; | |
1114 | lcounter += delta; | |
1115 | if (lcounter < 0) { | |
1116 | ASSERT(0); | |
014c2544 | 1117 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1118 | } |
1119 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1120 | return 0; |
1da177e4 LT |
1121 | case XFS_SBS_IFREE: |
1122 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1123 | lcounter += delta; | |
1124 | if (lcounter < 0) { | |
1125 | ASSERT(0); | |
014c2544 | 1126 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1127 | } |
1128 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1129 | return 0; |
1da177e4 | 1130 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1131 | lcounter = (long long) |
1132 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1133 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1134 | ||
1135 | if (delta > 0) { /* Putting blocks back */ | |
1136 | if (res_used > delta) { | |
1137 | mp->m_resblks_avail += delta; | |
1138 | } else { | |
1139 | rem = delta - res_used; | |
1140 | mp->m_resblks_avail = mp->m_resblks; | |
1141 | lcounter += rem; | |
1142 | } | |
1143 | } else { /* Taking blocks away */ | |
1da177e4 | 1144 | lcounter += delta; |
8babd8a2 DC |
1145 | if (lcounter >= 0) { |
1146 | mp->m_sb.sb_fdblocks = lcounter + | |
1147 | XFS_ALLOC_SET_ASIDE(mp); | |
1148 | return 0; | |
1149 | } | |
1da177e4 | 1150 | |
8babd8a2 DC |
1151 | /* |
1152 | * We are out of blocks, use any available reserved | |
1153 | * blocks if were allowed to. | |
1154 | */ | |
1155 | if (!rsvd) | |
1156 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1157 | |
8babd8a2 DC |
1158 | lcounter = (long long)mp->m_resblks_avail + delta; |
1159 | if (lcounter >= 0) { | |
1160 | mp->m_resblks_avail = lcounter; | |
1161 | return 0; | |
1da177e4 | 1162 | } |
8babd8a2 DC |
1163 | printk_once(KERN_WARNING |
1164 | "Filesystem \"%s\": reserve blocks depleted! " | |
1165 | "Consider increasing reserve pool size.", | |
1166 | mp->m_fsname); | |
1167 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1168 | } |
1169 | ||
4be536de | 1170 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1171 | return 0; |
1da177e4 LT |
1172 | case XFS_SBS_FREXTENTS: |
1173 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1174 | lcounter += delta; | |
1175 | if (lcounter < 0) { | |
014c2544 | 1176 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1177 | } |
1178 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1179 | return 0; |
1da177e4 LT |
1180 | case XFS_SBS_DBLOCKS: |
1181 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1182 | lcounter += delta; | |
1183 | if (lcounter < 0) { | |
1184 | ASSERT(0); | |
014c2544 | 1185 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1186 | } |
1187 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1188 | return 0; |
1da177e4 LT |
1189 | case XFS_SBS_AGCOUNT: |
1190 | scounter = mp->m_sb.sb_agcount; | |
1191 | scounter += delta; | |
1192 | if (scounter < 0) { | |
1193 | ASSERT(0); | |
014c2544 | 1194 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1195 | } |
1196 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1197 | return 0; |
1da177e4 LT |
1198 | case XFS_SBS_IMAX_PCT: |
1199 | scounter = mp->m_sb.sb_imax_pct; | |
1200 | scounter += delta; | |
1201 | if (scounter < 0) { | |
1202 | ASSERT(0); | |
014c2544 | 1203 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1204 | } |
1205 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1206 | return 0; |
1da177e4 LT |
1207 | case XFS_SBS_REXTSIZE: |
1208 | scounter = mp->m_sb.sb_rextsize; | |
1209 | scounter += delta; | |
1210 | if (scounter < 0) { | |
1211 | ASSERT(0); | |
014c2544 | 1212 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1213 | } |
1214 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1215 | return 0; |
1da177e4 LT |
1216 | case XFS_SBS_RBMBLOCKS: |
1217 | scounter = mp->m_sb.sb_rbmblocks; | |
1218 | scounter += delta; | |
1219 | if (scounter < 0) { | |
1220 | ASSERT(0); | |
014c2544 | 1221 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1222 | } |
1223 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1224 | return 0; |
1da177e4 LT |
1225 | case XFS_SBS_RBLOCKS: |
1226 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1227 | lcounter += delta; | |
1228 | if (lcounter < 0) { | |
1229 | ASSERT(0); | |
014c2544 | 1230 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1231 | } |
1232 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1233 | return 0; |
1da177e4 LT |
1234 | case XFS_SBS_REXTENTS: |
1235 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1236 | lcounter += delta; | |
1237 | if (lcounter < 0) { | |
1238 | ASSERT(0); | |
014c2544 | 1239 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1240 | } |
1241 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1242 | return 0; |
1da177e4 LT |
1243 | case XFS_SBS_REXTSLOG: |
1244 | scounter = mp->m_sb.sb_rextslog; | |
1245 | scounter += delta; | |
1246 | if (scounter < 0) { | |
1247 | ASSERT(0); | |
014c2544 | 1248 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1249 | } |
1250 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1251 | return 0; |
1da177e4 LT |
1252 | default: |
1253 | ASSERT(0); | |
014c2544 | 1254 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1255 | } |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1260 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1261 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1262 | * routine to do the work. |
1263 | */ | |
1264 | int | |
20f4ebf2 | 1265 | xfs_mod_incore_sb( |
96540c78 CH |
1266 | struct xfs_mount *mp, |
1267 | xfs_sb_field_t field, | |
1268 | int64_t delta, | |
1269 | int rsvd) | |
1da177e4 | 1270 | { |
96540c78 | 1271 | int status; |
1da177e4 | 1272 | |
8d280b98 | 1273 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1274 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1275 | #endif |
96540c78 CH |
1276 | spin_lock(&mp->m_sb_lock); |
1277 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1278 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1279 | |
014c2544 | 1280 | return status; |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | /* | |
1b040712 | 1284 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1285 | * |
1b040712 CH |
1286 | * The fields and changes to those fields are specified in the array of |
1287 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1288 | * will be applied or none of them will. If any modified field dips below 0, | |
1289 | * then all modifications will be backed out and EINVAL will be returned. | |
1290 | * | |
1291 | * Note that this function may not be used for the superblock values that | |
1292 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1293 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1294 | */ |
1295 | int | |
1b040712 CH |
1296 | xfs_mod_incore_sb_batch( |
1297 | struct xfs_mount *mp, | |
1298 | xfs_mod_sb_t *msb, | |
1299 | uint nmsb, | |
1300 | int rsvd) | |
1da177e4 | 1301 | { |
45c51b99 | 1302 | xfs_mod_sb_t *msbp; |
1b040712 | 1303 | int error = 0; |
1da177e4 LT |
1304 | |
1305 | /* | |
1b040712 CH |
1306 | * Loop through the array of mod structures and apply each individually. |
1307 | * If any fail, then back out all those which have already been applied. | |
1308 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1309 | * changes will be atomic. | |
1da177e4 | 1310 | */ |
3685c2a1 | 1311 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1312 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1313 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1314 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1315 | |
1b040712 CH |
1316 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1317 | msbp->msb_delta, rsvd); | |
1318 | if (error) | |
1319 | goto unwind; | |
1da177e4 | 1320 | } |
1b040712 CH |
1321 | spin_unlock(&mp->m_sb_lock); |
1322 | return 0; | |
1da177e4 | 1323 | |
1b040712 CH |
1324 | unwind: |
1325 | while (--msbp >= msb) { | |
1326 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1327 | -msbp->msb_delta, rsvd); | |
1328 | ASSERT(error == 0); | |
1da177e4 | 1329 | } |
3685c2a1 | 1330 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1331 | return error; |
1da177e4 LT |
1332 | } |
1333 | ||
1334 | /* | |
1335 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1336 | * The buffer is returned locked and read in from disk. | |
1337 | * The buffer should be released with a call to xfs_brelse(). | |
1338 | * | |
1339 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1340 | * the superblock buffer if it can be locked without sleeping. | |
1341 | * If it can't then we'll return NULL. | |
1342 | */ | |
0c842ad4 | 1343 | struct xfs_buf * |
1da177e4 | 1344 | xfs_getsb( |
0c842ad4 CH |
1345 | struct xfs_mount *mp, |
1346 | int flags) | |
1da177e4 | 1347 | { |
0c842ad4 | 1348 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1349 | |
0c842ad4 CH |
1350 | if (!xfs_buf_trylock(bp)) { |
1351 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1352 | return NULL; |
0c842ad4 | 1353 | xfs_buf_lock(bp); |
1da177e4 | 1354 | } |
0c842ad4 | 1355 | |
72790aa1 | 1356 | xfs_buf_hold(bp); |
1da177e4 | 1357 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 1358 | return bp; |
1da177e4 LT |
1359 | } |
1360 | ||
1361 | /* | |
1362 | * Used to free the superblock along various error paths. | |
1363 | */ | |
1364 | void | |
1365 | xfs_freesb( | |
26af6552 | 1366 | struct xfs_mount *mp) |
1da177e4 | 1367 | { |
26af6552 | 1368 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1369 | |
26af6552 | 1370 | xfs_buf_lock(bp); |
1da177e4 | 1371 | mp->m_sb_bp = NULL; |
26af6552 | 1372 | xfs_buf_relse(bp); |
1da177e4 LT |
1373 | } |
1374 | ||
1da177e4 LT |
1375 | /* |
1376 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1377 | * be altered by the mount options, as well as any potential sb_features2 |
1378 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1379 | */ |
7884bc86 | 1380 | int |
ee1c0908 | 1381 | xfs_mount_log_sb( |
1da177e4 LT |
1382 | xfs_mount_t *mp, |
1383 | __int64_t fields) | |
1384 | { | |
1385 | xfs_trans_t *tp; | |
e5720eec | 1386 | int error; |
1da177e4 | 1387 | |
ee1c0908 | 1388 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1389 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1390 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1391 | |
1392 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
3d3c8b52 | 1393 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
e5720eec | 1394 | if (error) { |
1da177e4 | 1395 | xfs_trans_cancel(tp, 0); |
e5720eec | 1396 | return error; |
1da177e4 LT |
1397 | } |
1398 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1399 | error = xfs_trans_commit(tp, 0); |
1400 | return error; | |
1da177e4 | 1401 | } |
8d280b98 | 1402 | |
dda35b8f CH |
1403 | /* |
1404 | * If the underlying (data/log/rt) device is readonly, there are some | |
1405 | * operations that cannot proceed. | |
1406 | */ | |
1407 | int | |
1408 | xfs_dev_is_read_only( | |
1409 | struct xfs_mount *mp, | |
1410 | char *message) | |
1411 | { | |
1412 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1413 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1414 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1415 | xfs_notice(mp, "%s required on read-only device.", message); |
1416 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
1417 | return EROFS; |
1418 | } | |
1419 | return 0; | |
1420 | } | |
8d280b98 DC |
1421 | |
1422 | #ifdef HAVE_PERCPU_SB | |
1423 | /* | |
1424 | * Per-cpu incore superblock counters | |
1425 | * | |
1426 | * Simple concept, difficult implementation | |
1427 | * | |
1428 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1429 | * counter for contended fields (e.g. free block count). | |
1430 | * | |
1431 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1432 | * hence needs to be accurately read when we are running low on space. Hence | |
1433 | * there is a method to enable and disable the per-cpu counters based on how | |
1434 | * much "stuff" is available in them. | |
1435 | * | |
1436 | * Basically, a counter is enabled if there is enough free resource to justify | |
1437 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1438 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1439 | * re-distribute the available resources. | |
1440 | * | |
1441 | * If, once we redistributed the available resources, we still get a failure, | |
1442 | * we disable the per-cpu counter and go through the slow path. | |
1443 | * | |
1444 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 1445 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
1446 | * the global superblock. We do this after disabling the counter to prevent |
1447 | * more threads from queueing up on the counter. | |
1448 | * | |
1449 | * Essentially, this means that we still need a lock in the fast path to enable | |
1450 | * synchronisation between the global counters and the per-cpu counters. This | |
1451 | * is not a problem because the lock will be local to a CPU almost all the time | |
1452 | * and have little contention except when we get to ENOSPC conditions. | |
1453 | * | |
1454 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1455 | * path while we do things like enabling and disabling counters and | |
1456 | * synchronising the counters. | |
1457 | * | |
1458 | * Locking rules: | |
1459 | * | |
3685c2a1 | 1460 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1461 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1462 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1463 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1464 | * 5. modifying global counters requires holding m_sb_lock |
1465 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1466 | * and _none_ of the per-cpu locks. |
1467 | * | |
1468 | * Disabled counters are only ever re-enabled by a balance operation | |
1469 | * that results in more free resources per CPU than a given threshold. | |
1470 | * To ensure counters don't remain disabled, they are rebalanced when | |
1471 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1472 | * is present to prevent thrashing). | |
e8234a68 DC |
1473 | */ |
1474 | ||
5a67e4c5 | 1475 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1476 | /* |
1477 | * hot-plug CPU notifier support. | |
8d280b98 | 1478 | * |
5a67e4c5 CS |
1479 | * We need a notifier per filesystem as we need to be able to identify |
1480 | * the filesystem to balance the counters out. This is achieved by | |
1481 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1482 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1483 | */ |
e8234a68 DC |
1484 | STATIC int |
1485 | xfs_icsb_cpu_notify( | |
1486 | struct notifier_block *nfb, | |
1487 | unsigned long action, | |
1488 | void *hcpu) | |
1489 | { | |
1490 | xfs_icsb_cnts_t *cntp; | |
1491 | xfs_mount_t *mp; | |
e8234a68 DC |
1492 | |
1493 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1494 | cntp = (xfs_icsb_cnts_t *) | |
1495 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1496 | switch (action) { | |
1497 | case CPU_UP_PREPARE: | |
8bb78442 | 1498 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1499 | /* Easy Case - initialize the area and locks, and |
1500 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1501 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1502 | break; |
1503 | case CPU_ONLINE: | |
8bb78442 | 1504 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1505 | xfs_icsb_lock(mp); |
45af6c6d CH |
1506 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1507 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1508 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1509 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1510 | break; |
1511 | case CPU_DEAD: | |
8bb78442 | 1512 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1513 | /* Disable all the counters, then fold the dead cpu's |
1514 | * count into the total on the global superblock and | |
1515 | * re-enable the counters. */ | |
03135cf7 | 1516 | xfs_icsb_lock(mp); |
3685c2a1 | 1517 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1518 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
1519 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1520 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1521 | ||
1522 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1523 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1524 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1525 | ||
01e1b69c | 1526 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1527 | |
45af6c6d CH |
1528 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
1529 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
1530 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1531 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1532 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1533 | break; |
1534 | } | |
1535 | ||
1536 | return NOTIFY_OK; | |
1537 | } | |
5a67e4c5 | 1538 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1539 | |
8d280b98 DC |
1540 | int |
1541 | xfs_icsb_init_counters( | |
1542 | xfs_mount_t *mp) | |
1543 | { | |
1544 | xfs_icsb_cnts_t *cntp; | |
1545 | int i; | |
1546 | ||
1547 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1548 | if (mp->m_sb_cnts == NULL) | |
1549 | return -ENOMEM; | |
1550 | ||
5a67e4c5 | 1551 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1552 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
1553 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
1554 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
1555 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 1556 | |
8d280b98 DC |
1557 | for_each_online_cpu(i) { |
1558 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1559 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 1560 | } |
20b64285 DC |
1561 | |
1562 | mutex_init(&mp->m_icsb_mutex); | |
1563 | ||
8d280b98 DC |
1564 | /* |
1565 | * start with all counters disabled so that the | |
1566 | * initial balance kicks us off correctly | |
1567 | */ | |
1568 | mp->m_icsb_counters = -1; | |
1569 | return 0; | |
1570 | } | |
1571 | ||
5478eead LM |
1572 | void |
1573 | xfs_icsb_reinit_counters( | |
1574 | xfs_mount_t *mp) | |
1575 | { | |
1576 | xfs_icsb_lock(mp); | |
1577 | /* | |
1578 | * start with all counters disabled so that the | |
1579 | * initial balance kicks us off correctly | |
1580 | */ | |
1581 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
1582 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1583 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1584 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
1585 | xfs_icsb_unlock(mp); |
1586 | } | |
1587 | ||
c962fb79 | 1588 | void |
8d280b98 DC |
1589 | xfs_icsb_destroy_counters( |
1590 | xfs_mount_t *mp) | |
1591 | { | |
e8234a68 | 1592 | if (mp->m_sb_cnts) { |
5a67e4c5 | 1593 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 1594 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 1595 | } |
03135cf7 | 1596 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
1597 | } |
1598 | ||
b8f82a4a | 1599 | STATIC void |
01e1b69c DC |
1600 | xfs_icsb_lock_cntr( |
1601 | xfs_icsb_cnts_t *icsbp) | |
1602 | { | |
1603 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1604 | ndelay(1000); | |
1605 | } | |
1606 | } | |
1607 | ||
b8f82a4a | 1608 | STATIC void |
01e1b69c DC |
1609 | xfs_icsb_unlock_cntr( |
1610 | xfs_icsb_cnts_t *icsbp) | |
1611 | { | |
1612 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1613 | } | |
1614 | ||
8d280b98 | 1615 | |
b8f82a4a | 1616 | STATIC void |
8d280b98 DC |
1617 | xfs_icsb_lock_all_counters( |
1618 | xfs_mount_t *mp) | |
1619 | { | |
1620 | xfs_icsb_cnts_t *cntp; | |
1621 | int i; | |
1622 | ||
1623 | for_each_online_cpu(i) { | |
1624 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1625 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1626 | } |
1627 | } | |
1628 | ||
b8f82a4a | 1629 | STATIC void |
8d280b98 DC |
1630 | xfs_icsb_unlock_all_counters( |
1631 | xfs_mount_t *mp) | |
1632 | { | |
1633 | xfs_icsb_cnts_t *cntp; | |
1634 | int i; | |
1635 | ||
1636 | for_each_online_cpu(i) { | |
1637 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1638 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1639 | } |
1640 | } | |
1641 | ||
1642 | STATIC void | |
1643 | xfs_icsb_count( | |
1644 | xfs_mount_t *mp, | |
1645 | xfs_icsb_cnts_t *cnt, | |
1646 | int flags) | |
1647 | { | |
1648 | xfs_icsb_cnts_t *cntp; | |
1649 | int i; | |
1650 | ||
1651 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1652 | ||
1653 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1654 | xfs_icsb_lock_all_counters(mp); | |
1655 | ||
1656 | for_each_online_cpu(i) { | |
1657 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
1658 | cnt->icsb_icount += cntp->icsb_icount; | |
1659 | cnt->icsb_ifree += cntp->icsb_ifree; | |
1660 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1661 | } | |
1662 | ||
1663 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1664 | xfs_icsb_unlock_all_counters(mp); | |
1665 | } | |
1666 | ||
1667 | STATIC int | |
1668 | xfs_icsb_counter_disabled( | |
1669 | xfs_mount_t *mp, | |
1670 | xfs_sb_field_t field) | |
1671 | { | |
1672 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1673 | return test_bit(field, &mp->m_icsb_counters); | |
1674 | } | |
1675 | ||
36fbe6e6 | 1676 | STATIC void |
8d280b98 DC |
1677 | xfs_icsb_disable_counter( |
1678 | xfs_mount_t *mp, | |
1679 | xfs_sb_field_t field) | |
1680 | { | |
1681 | xfs_icsb_cnts_t cnt; | |
1682 | ||
1683 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1684 | ||
20b64285 DC |
1685 | /* |
1686 | * If we are already disabled, then there is nothing to do | |
1687 | * here. We check before locking all the counters to avoid | |
1688 | * the expensive lock operation when being called in the | |
1689 | * slow path and the counter is already disabled. This is | |
1690 | * safe because the only time we set or clear this state is under | |
1691 | * the m_icsb_mutex. | |
1692 | */ | |
1693 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 1694 | return; |
20b64285 | 1695 | |
8d280b98 DC |
1696 | xfs_icsb_lock_all_counters(mp); |
1697 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1698 | /* drain back to superblock */ | |
1699 | ||
ce46193b | 1700 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
1701 | switch(field) { |
1702 | case XFS_SBS_ICOUNT: | |
1703 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1704 | break; | |
1705 | case XFS_SBS_IFREE: | |
1706 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1707 | break; | |
1708 | case XFS_SBS_FDBLOCKS: | |
1709 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1710 | break; | |
1711 | default: | |
1712 | BUG(); | |
1713 | } | |
1714 | } | |
1715 | ||
1716 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
1717 | } |
1718 | ||
1719 | STATIC void | |
1720 | xfs_icsb_enable_counter( | |
1721 | xfs_mount_t *mp, | |
1722 | xfs_sb_field_t field, | |
1723 | uint64_t count, | |
1724 | uint64_t resid) | |
1725 | { | |
1726 | xfs_icsb_cnts_t *cntp; | |
1727 | int i; | |
1728 | ||
1729 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1730 | ||
1731 | xfs_icsb_lock_all_counters(mp); | |
1732 | for_each_online_cpu(i) { | |
1733 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1734 | switch (field) { | |
1735 | case XFS_SBS_ICOUNT: | |
1736 | cntp->icsb_icount = count + resid; | |
1737 | break; | |
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_ICOUNT)) |
1764 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1765 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
1766 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1767 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1768 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
1769 | } |
1770 | ||
1771 | /* | |
1772 | * Accurate update of per-cpu counters to incore superblock | |
1773 | */ | |
d4d90b57 | 1774 | void |
8d280b98 | 1775 | xfs_icsb_sync_counters( |
d4d90b57 CH |
1776 | xfs_mount_t *mp, |
1777 | int flags) | |
8d280b98 | 1778 | { |
d4d90b57 CH |
1779 | spin_lock(&mp->m_sb_lock); |
1780 | xfs_icsb_sync_counters_locked(mp, flags); | |
1781 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1782 | } |
1783 | ||
1784 | /* | |
1785 | * Balance and enable/disable counters as necessary. | |
1786 | * | |
20b64285 DC |
1787 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
1788 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
1789 | * free blocks is something far enough zero that we aren't going thrash when we | |
1790 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
1791 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
1792 | * be distributed to a single CPU but each CPU has enough blocks to be | |
1793 | * reenabled. | |
1794 | * | |
1795 | * Note that we can be called when counters are already disabled. | |
1796 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
1797 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 1798 | */ |
20b64285 DC |
1799 | |
1800 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 1801 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 1802 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 1803 | STATIC void |
45af6c6d | 1804 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
1805 | xfs_mount_t *mp, |
1806 | xfs_sb_field_t field, | |
20b64285 | 1807 | int min_per_cpu) |
8d280b98 | 1808 | { |
6fdf8ccc | 1809 | uint64_t count, resid; |
8d280b98 | 1810 | int weight = num_online_cpus(); |
20b64285 | 1811 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 1812 | |
8d280b98 DC |
1813 | /* disable counter and sync counter */ |
1814 | xfs_icsb_disable_counter(mp, field); | |
1815 | ||
1816 | /* update counters - first CPU gets residual*/ | |
1817 | switch (field) { | |
1818 | case XFS_SBS_ICOUNT: | |
1819 | count = mp->m_sb.sb_icount; | |
1820 | resid = do_div(count, weight); | |
20b64285 | 1821 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1822 | return; |
8d280b98 DC |
1823 | break; |
1824 | case XFS_SBS_IFREE: | |
1825 | count = mp->m_sb.sb_ifree; | |
1826 | resid = do_div(count, weight); | |
20b64285 | 1827 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1828 | return; |
8d280b98 DC |
1829 | break; |
1830 | case XFS_SBS_FDBLOCKS: | |
1831 | count = mp->m_sb.sb_fdblocks; | |
1832 | resid = do_div(count, weight); | |
20b64285 | 1833 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 1834 | return; |
8d280b98 DC |
1835 | break; |
1836 | default: | |
1837 | BUG(); | |
6fdf8ccc | 1838 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
1839 | break; |
1840 | } | |
1841 | ||
1842 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
1843 | } |
1844 | ||
1845 | STATIC void | |
1846 | xfs_icsb_balance_counter( | |
1847 | xfs_mount_t *mp, | |
1848 | xfs_sb_field_t fields, | |
1849 | int min_per_cpu) | |
1850 | { | |
1851 | spin_lock(&mp->m_sb_lock); | |
1852 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
1853 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1854 | } |
1855 | ||
1b040712 | 1856 | int |
20b64285 | 1857 | xfs_icsb_modify_counters( |
8d280b98 DC |
1858 | xfs_mount_t *mp, |
1859 | xfs_sb_field_t field, | |
20f4ebf2 | 1860 | int64_t delta, |
20b64285 | 1861 | int rsvd) |
8d280b98 DC |
1862 | { |
1863 | xfs_icsb_cnts_t *icsbp; | |
1864 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 1865 | int ret = 0; |
8d280b98 | 1866 | |
20b64285 | 1867 | might_sleep(); |
8d280b98 | 1868 | again: |
7a9e02d6 CL |
1869 | preempt_disable(); |
1870 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
1871 | |
1872 | /* | |
1873 | * if the counter is disabled, go to slow path | |
1874 | */ | |
8d280b98 DC |
1875 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
1876 | goto slow_path; | |
20b64285 DC |
1877 | xfs_icsb_lock_cntr(icsbp); |
1878 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
1879 | xfs_icsb_unlock_cntr(icsbp); | |
1880 | goto slow_path; | |
1881 | } | |
8d280b98 DC |
1882 | |
1883 | switch (field) { | |
1884 | case XFS_SBS_ICOUNT: | |
1885 | lcounter = icsbp->icsb_icount; | |
1886 | lcounter += delta; | |
1887 | if (unlikely(lcounter < 0)) | |
20b64285 | 1888 | goto balance_counter; |
8d280b98 DC |
1889 | icsbp->icsb_icount = lcounter; |
1890 | break; | |
1891 | ||
1892 | case XFS_SBS_IFREE: | |
1893 | lcounter = icsbp->icsb_ifree; | |
1894 | lcounter += delta; | |
1895 | if (unlikely(lcounter < 0)) | |
20b64285 | 1896 | goto balance_counter; |
8d280b98 DC |
1897 | icsbp->icsb_ifree = lcounter; |
1898 | break; | |
1899 | ||
1900 | case XFS_SBS_FDBLOCKS: | |
1901 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
1902 | ||
4be536de | 1903 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1904 | lcounter += delta; |
1905 | if (unlikely(lcounter < 0)) | |
20b64285 | 1906 | goto balance_counter; |
4be536de | 1907 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1908 | break; |
1909 | default: | |
1910 | BUG(); | |
1911 | break; | |
1912 | } | |
01e1b69c | 1913 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 1914 | preempt_enable(); |
8d280b98 DC |
1915 | return 0; |
1916 | ||
8d280b98 | 1917 | slow_path: |
7a9e02d6 | 1918 | preempt_enable(); |
8d280b98 | 1919 | |
20b64285 DC |
1920 | /* |
1921 | * serialise with a mutex so we don't burn lots of cpu on | |
1922 | * the superblock lock. We still need to hold the superblock | |
1923 | * lock, however, when we modify the global structures. | |
1924 | */ | |
03135cf7 | 1925 | xfs_icsb_lock(mp); |
20b64285 DC |
1926 | |
1927 | /* | |
1928 | * Now running atomically. | |
1929 | * | |
1930 | * If the counter is enabled, someone has beaten us to rebalancing. | |
1931 | * Drop the lock and try again in the fast path.... | |
1932 | */ | |
1933 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 1934 | xfs_icsb_unlock(mp); |
8d280b98 | 1935 | goto again; |
8d280b98 DC |
1936 | } |
1937 | ||
20b64285 DC |
1938 | /* |
1939 | * The counter is currently disabled. Because we are | |
1940 | * running atomically here, we know a rebalance cannot | |
1941 | * be in progress. Hence we can go straight to operating | |
1942 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 1943 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 1944 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 1945 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
1946 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
1947 | * directly on the global counters. | |
1948 | */ | |
3685c2a1 | 1949 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1950 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1951 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 1952 | |
20b64285 DC |
1953 | /* |
1954 | * Now that we've modified the global superblock, we | |
1955 | * may be able to re-enable the distributed counters | |
1956 | * (e.g. lots of space just got freed). After that | |
1957 | * we are done. | |
1958 | */ | |
1959 | if (ret != ENOSPC) | |
45af6c6d | 1960 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 1961 | xfs_icsb_unlock(mp); |
8d280b98 | 1962 | return ret; |
8d280b98 | 1963 | |
20b64285 DC |
1964 | balance_counter: |
1965 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 1966 | preempt_enable(); |
8d280b98 | 1967 | |
20b64285 DC |
1968 | /* |
1969 | * We may have multiple threads here if multiple per-cpu | |
1970 | * counters run dry at the same time. This will mean we can | |
1971 | * do more balances than strictly necessary but it is not | |
1972 | * the common slowpath case. | |
1973 | */ | |
03135cf7 | 1974 | xfs_icsb_lock(mp); |
20b64285 DC |
1975 | |
1976 | /* | |
1977 | * running atomically. | |
1978 | * | |
1979 | * This will leave the counter in the correct state for future | |
1980 | * accesses. After the rebalance, we simply try again and our retry | |
1981 | * will either succeed through the fast path or slow path without | |
1982 | * another balance operation being required. | |
1983 | */ | |
45af6c6d | 1984 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 1985 | xfs_icsb_unlock(mp); |
20b64285 | 1986 | goto again; |
8d280b98 | 1987 | } |
20b64285 | 1988 | |
8d280b98 | 1989 | #endif |