]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
70a9883c | 20 | #include "xfs_shared.h" |
239880ef DC |
21 | #include "xfs_format.h" |
22 | #include "xfs_log_format.h" | |
23 | #include "xfs_trans_resv.h" | |
a844f451 | 24 | #include "xfs_bit.h" |
a844f451 | 25 | #include "xfs_inum.h" |
1da177e4 LT |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
57062787 | 29 | #include "xfs_da_format.h" |
1da177e4 | 30 | #include "xfs_inode.h" |
a4fbe6ab | 31 | #include "xfs_dir2.h" |
a844f451 | 32 | #include "xfs_ialloc.h" |
1da177e4 LT |
33 | #include "xfs_alloc.h" |
34 | #include "xfs_rtalloc.h" | |
35 | #include "xfs_bmap.h" | |
a4fbe6ab DC |
36 | #include "xfs_trans.h" |
37 | #include "xfs_trans_priv.h" | |
38 | #include "xfs_log.h" | |
1da177e4 | 39 | #include "xfs_error.h" |
1da177e4 LT |
40 | #include "xfs_quota.h" |
41 | #include "xfs_fsops.h" | |
0b1b213f | 42 | #include "xfs_trace.h" |
6d8b79cf | 43 | #include "xfs_icache.h" |
8f80587b | 44 | #include "xfs_dinode.h" |
0b1b213f | 45 | |
1da177e4 | 46 | |
8d280b98 | 47 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 48 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
49 | int); |
50 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
51 | int); | |
36fbe6e6 | 52 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
53 | #else |
54 | ||
45af6c6d CH |
55 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
56 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
57 | #endif |
58 | ||
27174203 CH |
59 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
60 | static int xfs_uuid_table_size; | |
61 | static uuid_t *xfs_uuid_table; | |
62 | ||
63 | /* | |
64 | * See if the UUID is unique among mounted XFS filesystems. | |
65 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
66 | */ | |
67 | STATIC int | |
68 | xfs_uuid_mount( | |
69 | struct xfs_mount *mp) | |
70 | { | |
71 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
72 | int hole, i; | |
73 | ||
74 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
75 | return 0; | |
76 | ||
77 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 78 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
79 | return XFS_ERROR(EINVAL); |
80 | } | |
81 | ||
82 | mutex_lock(&xfs_uuid_table_mutex); | |
83 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
84 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
85 | hole = i; | |
86 | continue; | |
87 | } | |
88 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
89 | goto out_duplicate; | |
90 | } | |
91 | ||
92 | if (hole < 0) { | |
93 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
94 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
95 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
96 | KM_SLEEP); | |
97 | hole = xfs_uuid_table_size++; | |
98 | } | |
99 | xfs_uuid_table[hole] = *uuid; | |
100 | mutex_unlock(&xfs_uuid_table_mutex); | |
101 | ||
102 | return 0; | |
103 | ||
104 | out_duplicate: | |
105 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 106 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
27174203 CH |
107 | return XFS_ERROR(EINVAL); |
108 | } | |
109 | ||
110 | STATIC void | |
111 | xfs_uuid_unmount( | |
112 | struct xfs_mount *mp) | |
113 | { | |
114 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
115 | int i; | |
116 | ||
117 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
118 | return; | |
119 | ||
120 | mutex_lock(&xfs_uuid_table_mutex); | |
121 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
122 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
123 | continue; | |
124 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
125 | continue; | |
126 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
127 | break; | |
128 | } | |
129 | ASSERT(i < xfs_uuid_table_size); | |
130 | mutex_unlock(&xfs_uuid_table_mutex); | |
131 | } | |
132 | ||
133 | ||
e176579e DC |
134 | STATIC void |
135 | __xfs_free_perag( | |
136 | struct rcu_head *head) | |
137 | { | |
138 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
139 | ||
140 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
141 | kmem_free(pag); | |
142 | } | |
143 | ||
1da177e4 | 144 | /* |
e176579e | 145 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 146 | */ |
c962fb79 | 147 | STATIC void |
ff4f038c | 148 | xfs_free_perag( |
745f6919 | 149 | xfs_mount_t *mp) |
1da177e4 | 150 | { |
1c1c6ebc DC |
151 | xfs_agnumber_t agno; |
152 | struct xfs_perag *pag; | |
153 | ||
154 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
155 | spin_lock(&mp->m_perag_lock); | |
156 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
157 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 158 | ASSERT(pag); |
f83282a8 | 159 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 160 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 161 | } |
1da177e4 LT |
162 | } |
163 | ||
4cc929ee NS |
164 | /* |
165 | * Check size of device based on the (data/realtime) block count. | |
166 | * Note: this check is used by the growfs code as well as mount. | |
167 | */ | |
168 | int | |
169 | xfs_sb_validate_fsb_count( | |
170 | xfs_sb_t *sbp, | |
171 | __uint64_t nblocks) | |
172 | { | |
173 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
174 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
175 | ||
176 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
177 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 178 | return EFBIG; |
4cc929ee NS |
179 | #else /* Limited by UINT_MAX of sectors */ |
180 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 181 | return EFBIG; |
4cc929ee NS |
182 | #endif |
183 | return 0; | |
184 | } | |
1da177e4 | 185 | |
1c1c6ebc | 186 | int |
c11e2c36 | 187 | xfs_initialize_perag( |
c11e2c36 | 188 | xfs_mount_t *mp, |
1c1c6ebc DC |
189 | xfs_agnumber_t agcount, |
190 | xfs_agnumber_t *maxagi) | |
1da177e4 | 191 | { |
2d2194f6 | 192 | xfs_agnumber_t index; |
8b26c582 | 193 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
194 | xfs_perag_t *pag; |
195 | xfs_agino_t agino; | |
196 | xfs_ino_t ino; | |
197 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 198 | int error = -ENOMEM; |
1da177e4 | 199 | |
1c1c6ebc DC |
200 | /* |
201 | * Walk the current per-ag tree so we don't try to initialise AGs | |
202 | * that already exist (growfs case). Allocate and insert all the | |
203 | * AGs we don't find ready for initialisation. | |
204 | */ | |
205 | for (index = 0; index < agcount; index++) { | |
206 | pag = xfs_perag_get(mp, index); | |
207 | if (pag) { | |
208 | xfs_perag_put(pag); | |
209 | continue; | |
210 | } | |
8b26c582 DC |
211 | if (!first_initialised) |
212 | first_initialised = index; | |
fb3b504a | 213 | |
1c1c6ebc DC |
214 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
215 | if (!pag) | |
8b26c582 | 216 | goto out_unwind; |
fb3b504a CH |
217 | pag->pag_agno = index; |
218 | pag->pag_mount = mp; | |
1a427ab0 | 219 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 220 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 221 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
222 | spin_lock_init(&pag->pag_buf_lock); |
223 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 224 | |
1c1c6ebc | 225 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 226 | goto out_unwind; |
fb3b504a | 227 | |
1c1c6ebc DC |
228 | spin_lock(&mp->m_perag_lock); |
229 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
230 | BUG(); | |
231 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
232 | radix_tree_preload_end(); |
233 | error = -EEXIST; | |
234 | goto out_unwind; | |
1c1c6ebc DC |
235 | } |
236 | spin_unlock(&mp->m_perag_lock); | |
237 | radix_tree_preload_end(); | |
238 | } | |
239 | ||
fb3b504a CH |
240 | /* |
241 | * If we mount with the inode64 option, or no inode overflows | |
242 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 243 | */ |
fb3b504a CH |
244 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
245 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
246 | ||
247 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 248 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 249 | else |
1da177e4 | 250 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 251 | |
2d2194f6 CM |
252 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
253 | index = xfs_set_inode32(mp); | |
254 | else | |
255 | index = xfs_set_inode64(mp); | |
fb3b504a | 256 | |
1c1c6ebc DC |
257 | if (maxagi) |
258 | *maxagi = index; | |
259 | return 0; | |
8b26c582 DC |
260 | |
261 | out_unwind: | |
262 | kmem_free(pag); | |
263 | for (; index > first_initialised; index--) { | |
264 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
265 | kmem_free(pag); | |
266 | } | |
267 | return error; | |
1da177e4 LT |
268 | } |
269 | ||
1da177e4 LT |
270 | /* |
271 | * xfs_readsb | |
272 | * | |
273 | * Does the initial read of the superblock. | |
274 | */ | |
275 | int | |
ff55068c DC |
276 | xfs_readsb( |
277 | struct xfs_mount *mp, | |
278 | int flags) | |
1da177e4 LT |
279 | { |
280 | unsigned int sector_size; | |
04a1e6c5 DC |
281 | struct xfs_buf *bp; |
282 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 283 | int error; |
af34e09d | 284 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
285 | |
286 | ASSERT(mp->m_sb_bp == NULL); | |
287 | ASSERT(mp->m_ddev_targp != NULL); | |
288 | ||
289 | /* | |
290 | * Allocate a (locked) buffer to hold the superblock. | |
291 | * This will be kept around at all times to optimize | |
292 | * access to the superblock. | |
293 | */ | |
294 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
295 | |
296 | reread: | |
e70b73f8 | 297 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
98021821 | 298 | BTOBB(sector_size), 0, |
1813dd64 DC |
299 | loud ? &xfs_sb_buf_ops |
300 | : &xfs_sb_quiet_buf_ops); | |
26af6552 | 301 | if (!bp) { |
af34e09d DC |
302 | if (loud) |
303 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 304 | return EIO; |
1da177e4 | 305 | } |
eab4e633 DC |
306 | if (bp->b_error) { |
307 | error = bp->b_error; | |
308 | if (loud) | |
e721f504 | 309 | xfs_warn(mp, "SB validate failed with error %d.", error); |
eab4e633 DC |
310 | goto release_buf; |
311 | } | |
1da177e4 LT |
312 | |
313 | /* | |
314 | * Initialize the mount structure from the superblock. | |
1da177e4 | 315 | */ |
98021821 | 316 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
83e782e1 | 317 | xfs_sb_quota_from_disk(&mp->m_sb); |
ff55068c | 318 | |
1da177e4 LT |
319 | /* |
320 | * We must be able to do sector-sized and sector-aligned IO. | |
321 | */ | |
04a1e6c5 | 322 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
323 | if (loud) |
324 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 325 | sector_size, sbp->sb_sectsize); |
1da177e4 | 326 | error = ENOSYS; |
26af6552 | 327 | goto release_buf; |
1da177e4 LT |
328 | } |
329 | ||
330 | /* | |
331 | * If device sector size is smaller than the superblock size, | |
332 | * re-read the superblock so the buffer is correctly sized. | |
333 | */ | |
04a1e6c5 | 334 | if (sector_size < sbp->sb_sectsize) { |
1da177e4 | 335 | xfs_buf_relse(bp); |
04a1e6c5 | 336 | sector_size = sbp->sb_sectsize; |
26af6552 | 337 | goto reread; |
1da177e4 LT |
338 | } |
339 | ||
5478eead LM |
340 | /* Initialize per-cpu counters */ |
341 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 342 | |
04a1e6c5 DC |
343 | /* no need to be quiet anymore, so reset the buf ops */ |
344 | bp->b_ops = &xfs_sb_buf_ops; | |
345 | ||
1da177e4 | 346 | mp->m_sb_bp = bp; |
26af6552 | 347 | xfs_buf_unlock(bp); |
1da177e4 LT |
348 | return 0; |
349 | ||
26af6552 DC |
350 | release_buf: |
351 | xfs_buf_relse(bp); | |
1da177e4 LT |
352 | return error; |
353 | } | |
354 | ||
1da177e4 | 355 | /* |
0771fb45 | 356 | * Update alignment values based on mount options and sb values |
1da177e4 | 357 | */ |
0771fb45 | 358 | STATIC int |
7884bc86 | 359 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 360 | { |
1da177e4 | 361 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 362 | |
4249023a | 363 | if (mp->m_dalign) { |
1da177e4 LT |
364 | /* |
365 | * If stripe unit and stripe width are not multiples | |
366 | * of the fs blocksize turn off alignment. | |
367 | */ | |
368 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
369 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
370 | xfs_warn(mp, |
371 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
372 | sbp->sb_blocksize); | |
373 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
374 | } else { |
375 | /* | |
376 | * Convert the stripe unit and width to FSBs. | |
377 | */ | |
378 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
379 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 380 | xfs_warn(mp, |
39a45d84 JL |
381 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
382 | sbp->sb_agblocks); | |
383 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
384 | } else if (mp->m_dalign) { |
385 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
386 | } else { | |
39a45d84 JL |
387 | xfs_warn(mp, |
388 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
389 | mp->m_dalign, sbp->sb_blocksize); | |
390 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
391 | } |
392 | } | |
393 | ||
394 | /* | |
395 | * Update superblock with new values | |
396 | * and log changes | |
397 | */ | |
62118709 | 398 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
399 | if (sbp->sb_unit != mp->m_dalign) { |
400 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 401 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
402 | } |
403 | if (sbp->sb_width != mp->m_swidth) { | |
404 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 405 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 | 406 | } |
34d7f603 JL |
407 | } else { |
408 | xfs_warn(mp, | |
409 | "cannot change alignment: superblock does not support data alignment"); | |
410 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
411 | } |
412 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 413 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
414 | mp->m_dalign = sbp->sb_unit; |
415 | mp->m_swidth = sbp->sb_width; | |
416 | } | |
417 | ||
0771fb45 ES |
418 | return 0; |
419 | } | |
1da177e4 | 420 | |
0771fb45 ES |
421 | /* |
422 | * Set the maximum inode count for this filesystem | |
423 | */ | |
424 | STATIC void | |
425 | xfs_set_maxicount(xfs_mount_t *mp) | |
426 | { | |
427 | xfs_sb_t *sbp = &(mp->m_sb); | |
428 | __uint64_t icount; | |
1da177e4 | 429 | |
0771fb45 ES |
430 | if (sbp->sb_imax_pct) { |
431 | /* | |
432 | * Make sure the maximum inode count is a multiple | |
433 | * of the units we allocate inodes in. | |
1da177e4 | 434 | */ |
1da177e4 LT |
435 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
436 | do_div(icount, 100); | |
437 | do_div(icount, mp->m_ialloc_blks); | |
438 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
439 | sbp->sb_inopblog; | |
0771fb45 | 440 | } else { |
1da177e4 | 441 | mp->m_maxicount = 0; |
1da177e4 | 442 | } |
0771fb45 ES |
443 | } |
444 | ||
445 | /* | |
446 | * Set the default minimum read and write sizes unless | |
447 | * already specified in a mount option. | |
448 | * We use smaller I/O sizes when the file system | |
449 | * is being used for NFS service (wsync mount option). | |
450 | */ | |
451 | STATIC void | |
452 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
453 | { | |
454 | xfs_sb_t *sbp = &(mp->m_sb); | |
455 | int readio_log, writeio_log; | |
1da177e4 | 456 | |
1da177e4 LT |
457 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
458 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
459 | readio_log = XFS_WSYNC_READIO_LOG; | |
460 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
461 | } else { | |
462 | readio_log = XFS_READIO_LOG_LARGE; | |
463 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
464 | } | |
465 | } else { | |
466 | readio_log = mp->m_readio_log; | |
467 | writeio_log = mp->m_writeio_log; | |
468 | } | |
469 | ||
1da177e4 LT |
470 | if (sbp->sb_blocklog > readio_log) { |
471 | mp->m_readio_log = sbp->sb_blocklog; | |
472 | } else { | |
473 | mp->m_readio_log = readio_log; | |
474 | } | |
475 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
476 | if (sbp->sb_blocklog > writeio_log) { | |
477 | mp->m_writeio_log = sbp->sb_blocklog; | |
478 | } else { | |
479 | mp->m_writeio_log = writeio_log; | |
480 | } | |
481 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 482 | } |
1da177e4 | 483 | |
055388a3 DC |
484 | /* |
485 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
486 | */ | |
487 | void | |
488 | xfs_set_low_space_thresholds( | |
489 | struct xfs_mount *mp) | |
490 | { | |
491 | int i; | |
492 | ||
493 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
494 | __uint64_t space = mp->m_sb.sb_dblocks; | |
495 | ||
496 | do_div(space, 100); | |
497 | mp->m_low_space[i] = space * (i + 1); | |
498 | } | |
499 | } | |
500 | ||
501 | ||
0771fb45 ES |
502 | /* |
503 | * Set whether we're using inode alignment. | |
504 | */ | |
505 | STATIC void | |
506 | xfs_set_inoalignment(xfs_mount_t *mp) | |
507 | { | |
62118709 | 508 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
509 | mp->m_sb.sb_inoalignmt >= |
510 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
511 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
512 | else | |
513 | mp->m_inoalign_mask = 0; | |
514 | /* | |
515 | * If we are using stripe alignment, check whether | |
516 | * the stripe unit is a multiple of the inode alignment | |
517 | */ | |
518 | if (mp->m_dalign && mp->m_inoalign_mask && | |
519 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
520 | mp->m_sinoalign = mp->m_dalign; | |
521 | else | |
522 | mp->m_sinoalign = 0; | |
0771fb45 ES |
523 | } |
524 | ||
525 | /* | |
0471f62e | 526 | * Check that the data (and log if separate) is an ok size. |
0771fb45 ES |
527 | */ |
528 | STATIC int | |
4249023a | 529 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
530 | { |
531 | xfs_buf_t *bp; | |
532 | xfs_daddr_t d; | |
0771fb45 | 533 | |
1da177e4 LT |
534 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
535 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 536 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 537 | return XFS_ERROR(EFBIG); |
1da177e4 | 538 | } |
e70b73f8 | 539 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 540 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 541 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 542 | if (!bp) { |
0b932ccc | 543 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 544 | return EIO; |
1da177e4 | 545 | } |
1922c949 | 546 | xfs_buf_relse(bp); |
1da177e4 | 547 | |
4249023a | 548 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
549 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
550 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 551 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 552 | return XFS_ERROR(EFBIG); |
1da177e4 | 553 | } |
e70b73f8 | 554 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 555 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 556 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 557 | if (!bp) { |
0b932ccc | 558 | xfs_warn(mp, "log device read failed"); |
1922c949 | 559 | return EIO; |
0771fb45 | 560 | } |
1922c949 | 561 | xfs_buf_relse(bp); |
0771fb45 ES |
562 | } |
563 | return 0; | |
564 | } | |
565 | ||
7d095257 CH |
566 | /* |
567 | * Clear the quotaflags in memory and in the superblock. | |
568 | */ | |
569 | int | |
570 | xfs_mount_reset_sbqflags( | |
571 | struct xfs_mount *mp) | |
572 | { | |
573 | int error; | |
574 | struct xfs_trans *tp; | |
575 | ||
576 | mp->m_qflags = 0; | |
577 | ||
578 | /* | |
579 | * It is OK to look at sb_qflags here in mount path, | |
580 | * without m_sb_lock. | |
581 | */ | |
582 | if (mp->m_sb.sb_qflags == 0) | |
583 | return 0; | |
584 | spin_lock(&mp->m_sb_lock); | |
585 | mp->m_sb.sb_qflags = 0; | |
586 | spin_unlock(&mp->m_sb_lock); | |
587 | ||
588 | /* | |
589 | * If the fs is readonly, let the incore superblock run | |
590 | * with quotas off but don't flush the update out to disk | |
591 | */ | |
592 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
593 | return 0; | |
594 | ||
7d095257 | 595 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
3d3c8b52 | 596 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_qm_sbchange, 0, 0); |
7d095257 CH |
597 | if (error) { |
598 | xfs_trans_cancel(tp, 0); | |
53487786 | 599 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
600 | return error; |
601 | } | |
602 | ||
603 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
604 | return xfs_trans_commit(tp, 0); | |
605 | } | |
606 | ||
d5db0f97 ES |
607 | __uint64_t |
608 | xfs_default_resblks(xfs_mount_t *mp) | |
609 | { | |
610 | __uint64_t resblks; | |
611 | ||
612 | /* | |
8babd8a2 DC |
613 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
614 | * smaller. This is intended to cover concurrent allocation | |
615 | * transactions when we initially hit enospc. These each require a 4 | |
616 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
617 | * allocation reservations. | |
d5db0f97 ES |
618 | */ |
619 | resblks = mp->m_sb.sb_dblocks; | |
620 | do_div(resblks, 20); | |
8babd8a2 | 621 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
622 | return resblks; |
623 | } | |
624 | ||
0771fb45 | 625 | /* |
0771fb45 ES |
626 | * This function does the following on an initial mount of a file system: |
627 | * - reads the superblock from disk and init the mount struct | |
628 | * - if we're a 32-bit kernel, do a size check on the superblock | |
629 | * so we don't mount terabyte filesystems | |
630 | * - init mount struct realtime fields | |
631 | * - allocate inode hash table for fs | |
632 | * - init directory manager | |
633 | * - perform recovery and init the log manager | |
634 | */ | |
635 | int | |
636 | xfs_mountfs( | |
4249023a | 637 | xfs_mount_t *mp) |
0771fb45 ES |
638 | { |
639 | xfs_sb_t *sbp = &(mp->m_sb); | |
640 | xfs_inode_t *rip; | |
0771fb45 | 641 | __uint64_t resblks; |
7d095257 CH |
642 | uint quotamount = 0; |
643 | uint quotaflags = 0; | |
0771fb45 ES |
644 | int error = 0; |
645 | ||
ff55068c | 646 | xfs_sb_mount_common(mp, sbp); |
0771fb45 | 647 | |
ee1c0908 | 648 | /* |
e6957ea4 ES |
649 | * Check for a mismatched features2 values. Older kernels |
650 | * read & wrote into the wrong sb offset for sb_features2 | |
651 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
652 | * when sb_features2 was added, which made older superblock | |
653 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 654 | * |
e6957ea4 ES |
655 | * For backwards compatibility, we make both slots equal. |
656 | * | |
657 | * If we detect a mismatched field, we OR the set bits into the | |
658 | * existing features2 field in case it has already been modified; we | |
659 | * don't want to lose any features. We then update the bad location | |
660 | * with the ORed value so that older kernels will see any features2 | |
661 | * flags, and mark the two fields as needing updates once the | |
662 | * transaction subsystem is online. | |
ee1c0908 | 663 | */ |
e6957ea4 | 664 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 665 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 666 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 667 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 668 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
669 | |
670 | /* | |
671 | * Re-check for ATTR2 in case it was found in bad_features2 | |
672 | * slot. | |
673 | */ | |
7c12f296 TS |
674 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
675 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 676 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
677 | } |
678 | ||
679 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
680 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
681 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 682 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 683 | |
7c12f296 TS |
684 | /* update sb_versionnum for the clearing of the morebits */ |
685 | if (!sbp->sb_features2) | |
7884bc86 | 686 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
687 | } |
688 | ||
0771fb45 ES |
689 | /* |
690 | * Check if sb_agblocks is aligned at stripe boundary | |
691 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
692 | * allocator alignment is within an ag, therefore ag has | |
693 | * to be aligned at stripe boundary. | |
694 | */ | |
7884bc86 | 695 | error = xfs_update_alignment(mp); |
0771fb45 | 696 | if (error) |
f9057e3d | 697 | goto out; |
0771fb45 ES |
698 | |
699 | xfs_alloc_compute_maxlevels(mp); | |
700 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
701 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
702 | xfs_ialloc_compute_maxlevels(mp); | |
703 | ||
704 | xfs_set_maxicount(mp); | |
705 | ||
27174203 CH |
706 | error = xfs_uuid_mount(mp); |
707 | if (error) | |
708 | goto out; | |
1da177e4 | 709 | |
0771fb45 ES |
710 | /* |
711 | * Set the minimum read and write sizes | |
712 | */ | |
713 | xfs_set_rw_sizes(mp); | |
714 | ||
055388a3 DC |
715 | /* set the low space thresholds for dynamic preallocation */ |
716 | xfs_set_low_space_thresholds(mp); | |
717 | ||
0771fb45 ES |
718 | /* |
719 | * Set the inode cluster size. | |
720 | * This may still be overridden by the file system | |
721 | * block size if it is larger than the chosen cluster size. | |
8f80587b DC |
722 | * |
723 | * For v5 filesystems, scale the cluster size with the inode size to | |
724 | * keep a constant ratio of inode per cluster buffer, but only if mkfs | |
725 | * has set the inode alignment value appropriately for larger cluster | |
726 | * sizes. | |
0771fb45 ES |
727 | */ |
728 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
8f80587b DC |
729 | if (xfs_sb_version_hascrc(&mp->m_sb)) { |
730 | int new_size = mp->m_inode_cluster_size; | |
731 | ||
732 | new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE; | |
733 | if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size)) | |
734 | mp->m_inode_cluster_size = new_size; | |
735 | xfs_info(mp, "Using inode cluster size of %d bytes", | |
736 | mp->m_inode_cluster_size); | |
737 | } | |
0771fb45 ES |
738 | |
739 | /* | |
740 | * Set inode alignment fields | |
741 | */ | |
742 | xfs_set_inoalignment(mp); | |
743 | ||
744 | /* | |
c2bfbc9b | 745 | * Check that the data (and log if separate) is an ok size. |
0771fb45 | 746 | */ |
4249023a | 747 | error = xfs_check_sizes(mp); |
0771fb45 | 748 | if (error) |
f9057e3d | 749 | goto out_remove_uuid; |
0771fb45 | 750 | |
1da177e4 LT |
751 | /* |
752 | * Initialize realtime fields in the mount structure | |
753 | */ | |
0771fb45 ES |
754 | error = xfs_rtmount_init(mp); |
755 | if (error) { | |
0b932ccc | 756 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 757 | goto out_remove_uuid; |
1da177e4 LT |
758 | } |
759 | ||
1da177e4 LT |
760 | /* |
761 | * Copies the low order bits of the timestamp and the randomly | |
762 | * set "sequence" number out of a UUID. | |
763 | */ | |
764 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
765 | ||
1da177e4 LT |
766 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
767 | ||
f6c2d1fa | 768 | xfs_dir_mount(mp); |
1da177e4 LT |
769 | |
770 | /* | |
771 | * Initialize the attribute manager's entries. | |
772 | */ | |
773 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
774 | ||
775 | /* | |
776 | * Initialize the precomputed transaction reservations values. | |
777 | */ | |
778 | xfs_trans_init(mp); | |
779 | ||
1da177e4 LT |
780 | /* |
781 | * Allocate and initialize the per-ag data. | |
782 | */ | |
1c1c6ebc | 783 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 784 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
785 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
786 | if (error) { | |
0b932ccc | 787 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 788 | goto out_remove_uuid; |
1c1c6ebc | 789 | } |
1da177e4 | 790 | |
f9057e3d | 791 | if (!sbp->sb_logblocks) { |
0b932ccc | 792 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
793 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
794 | error = XFS_ERROR(EFSCORRUPTED); | |
795 | goto out_free_perag; | |
796 | } | |
797 | ||
1da177e4 LT |
798 | /* |
799 | * log's mount-time initialization. Perform 1st part recovery if needed | |
800 | */ | |
f9057e3d CH |
801 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
802 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
803 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
804 | if (error) { | |
0b932ccc | 805 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 806 | goto out_fail_wait; |
1da177e4 LT |
807 | } |
808 | ||
92821e2b DC |
809 | /* |
810 | * Now the log is mounted, we know if it was an unclean shutdown or | |
811 | * not. If it was, with the first phase of recovery has completed, we | |
812 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
813 | * but they are recovered transactionally in the second recovery phase | |
814 | * later. | |
815 | * | |
816 | * Hence we can safely re-initialise incore superblock counters from | |
817 | * the per-ag data. These may not be correct if the filesystem was not | |
818 | * cleanly unmounted, so we need to wait for recovery to finish before | |
819 | * doing this. | |
820 | * | |
821 | * If the filesystem was cleanly unmounted, then we can trust the | |
822 | * values in the superblock to be correct and we don't need to do | |
823 | * anything here. | |
824 | * | |
825 | * If we are currently making the filesystem, the initialisation will | |
826 | * fail as the perag data is in an undefined state. | |
827 | */ | |
92821e2b DC |
828 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
829 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
830 | !mp->m_sb.sb_inprogress) { | |
831 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 832 | if (error) |
d4f3512b | 833 | goto out_fail_wait; |
92821e2b | 834 | } |
f9057e3d | 835 | |
1da177e4 LT |
836 | /* |
837 | * Get and sanity-check the root inode. | |
838 | * Save the pointer to it in the mount structure. | |
839 | */ | |
7b6259e7 | 840 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 841 | if (error) { |
0b932ccc | 842 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 843 | goto out_log_dealloc; |
1da177e4 LT |
844 | } |
845 | ||
846 | ASSERT(rip != NULL); | |
1da177e4 | 847 | |
abbede1b | 848 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 849 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 850 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
851 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
852 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
853 | mp); | |
854 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 855 | goto out_rele_rip; |
1da177e4 LT |
856 | } |
857 | mp->m_rootip = rip; /* save it */ | |
858 | ||
859 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
860 | ||
861 | /* | |
862 | * Initialize realtime inode pointers in the mount structure | |
863 | */ | |
0771fb45 ES |
864 | error = xfs_rtmount_inodes(mp); |
865 | if (error) { | |
1da177e4 LT |
866 | /* |
867 | * Free up the root inode. | |
868 | */ | |
0b932ccc | 869 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 870 | goto out_rele_rip; |
1da177e4 LT |
871 | } |
872 | ||
873 | /* | |
7884bc86 CH |
874 | * If this is a read-only mount defer the superblock updates until |
875 | * the next remount into writeable mode. Otherwise we would never | |
876 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 877 | */ |
7884bc86 CH |
878 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
879 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 880 | if (error) { |
0b932ccc | 881 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 882 | goto out_rtunmount; |
e5720eec DC |
883 | } |
884 | } | |
1da177e4 LT |
885 | |
886 | /* | |
887 | * Initialise the XFS quota management subsystem for this mount | |
888 | */ | |
7d095257 CH |
889 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
890 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
891 | if (error) | |
892 | goto out_rtunmount; | |
893 | } else { | |
894 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
895 | ||
896 | /* | |
897 | * If a file system had quotas running earlier, but decided to | |
898 | * mount without -o uquota/pquota/gquota options, revoke the | |
899 | * quotachecked license. | |
900 | */ | |
901 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 902 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
903 | error = xfs_mount_reset_sbqflags(mp); |
904 | if (error) | |
905 | return error; | |
906 | } | |
907 | } | |
1da177e4 LT |
908 | |
909 | /* | |
910 | * Finish recovering the file system. This part needed to be | |
911 | * delayed until after the root and real-time bitmap inodes | |
912 | * were consistently read in. | |
913 | */ | |
4249023a | 914 | error = xfs_log_mount_finish(mp); |
1da177e4 | 915 | if (error) { |
0b932ccc | 916 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 917 | goto out_rtunmount; |
1da177e4 LT |
918 | } |
919 | ||
920 | /* | |
921 | * Complete the quota initialisation, post-log-replay component. | |
922 | */ | |
7d095257 CH |
923 | if (quotamount) { |
924 | ASSERT(mp->m_qflags == 0); | |
925 | mp->m_qflags = quotaflags; | |
926 | ||
927 | xfs_qm_mount_quotas(mp); | |
928 | } | |
929 | ||
84e1e99f DC |
930 | /* |
931 | * Now we are mounted, reserve a small amount of unused space for | |
932 | * privileged transactions. This is needed so that transaction | |
933 | * space required for critical operations can dip into this pool | |
934 | * when at ENOSPC. This is needed for operations like create with | |
935 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
936 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
937 | * |
938 | * This may drive us straight to ENOSPC on mount, but that implies | |
939 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 940 | */ |
d5db0f97 ES |
941 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
942 | resblks = xfs_default_resblks(mp); | |
943 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
944 | if (error) | |
0b932ccc DC |
945 | xfs_warn(mp, |
946 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 947 | } |
84e1e99f | 948 | |
1da177e4 LT |
949 | return 0; |
950 | ||
b93b6e43 CH |
951 | out_rtunmount: |
952 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 953 | out_rele_rip: |
43355099 | 954 | IRELE(rip); |
f9057e3d | 955 | out_log_dealloc: |
21b699c8 | 956 | xfs_log_unmount(mp); |
d4f3512b DC |
957 | out_fail_wait: |
958 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
959 | xfs_wait_buftarg(mp->m_logdev_targp); | |
960 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 961 | out_free_perag: |
ff4f038c | 962 | xfs_free_perag(mp); |
f9057e3d | 963 | out_remove_uuid: |
27174203 | 964 | xfs_uuid_unmount(mp); |
f9057e3d | 965 | out: |
1da177e4 LT |
966 | return error; |
967 | } | |
968 | ||
969 | /* | |
1da177e4 LT |
970 | * This flushes out the inodes,dquots and the superblock, unmounts the |
971 | * log and makes sure that incore structures are freed. | |
972 | */ | |
41b5c2e7 CH |
973 | void |
974 | xfs_unmountfs( | |
975 | struct xfs_mount *mp) | |
1da177e4 | 976 | { |
41b5c2e7 CH |
977 | __uint64_t resblks; |
978 | int error; | |
1da177e4 | 979 | |
579b62fa BF |
980 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
981 | ||
7d095257 | 982 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 983 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
984 | IRELE(mp->m_rootip); |
985 | ||
641c56fb DC |
986 | /* |
987 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 988 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
989 | * the transaction is still sitting in a iclog. The stale inodes |
990 | * on that buffer will have their flush locks held until the | |
991 | * transaction hits the disk and the callbacks run. the inode | |
992 | * flush takes the flush lock unconditionally and with nothing to | |
993 | * push out the iclog we will never get that unlocked. hence we | |
994 | * need to force the log first. | |
995 | */ | |
a14a348b | 996 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
997 | |
998 | /* | |
211e4d43 CH |
999 | * Flush all pending changes from the AIL. |
1000 | */ | |
1001 | xfs_ail_push_all_sync(mp->m_ail); | |
1002 | ||
1003 | /* | |
1004 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1005 | * inodes and none should be pinned or locked, but use synchronous |
1006 | * reclaim just to be sure. We can stop background inode reclaim | |
1007 | * here as well if it is still running. | |
c854363e | 1008 | */ |
7e18530b | 1009 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1010 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1011 | |
7d095257 | 1012 | xfs_qm_unmount(mp); |
a357a121 | 1013 | |
84e1e99f DC |
1014 | /* |
1015 | * Unreserve any blocks we have so that when we unmount we don't account | |
1016 | * the reserved free space as used. This is really only necessary for | |
1017 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1018 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1019 | * |
1020 | * We don't bother correcting this elsewhere for lazy superblock | |
1021 | * counting because on mount of an unclean filesystem we reconstruct the | |
1022 | * correct counter value and this is irrelevant. | |
1023 | * | |
1024 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1025 | * we only every apply deltas to the superblock and hence the incore | |
1026 | * value does not matter.... | |
1027 | */ | |
1028 | resblks = 0; | |
714082bc DC |
1029 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1030 | if (error) | |
0b932ccc | 1031 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1032 | "Freespace may not be correct on next mount."); |
1033 | ||
adab0f67 | 1034 | error = xfs_log_sbcount(mp); |
e5720eec | 1035 | if (error) |
0b932ccc | 1036 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1037 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1038 | |
21b699c8 | 1039 | xfs_log_unmount(mp); |
27174203 | 1040 | xfs_uuid_unmount(mp); |
1da177e4 | 1041 | |
1550d0b0 | 1042 | #if defined(DEBUG) |
0ce4cfd4 | 1043 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1044 | #endif |
ff4f038c | 1045 | xfs_free_perag(mp); |
1da177e4 LT |
1046 | } |
1047 | ||
92821e2b DC |
1048 | int |
1049 | xfs_fs_writable(xfs_mount_t *mp) | |
1050 | { | |
d9457dc0 | 1051 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1052 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1053 | } |
1054 | ||
1055 | /* | |
b2ce3974 AE |
1056 | * xfs_log_sbcount |
1057 | * | |
adab0f67 | 1058 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1059 | * |
1060 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1061 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1062 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1063 | */ |
1064 | int | |
adab0f67 | 1065 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1066 | { |
1067 | xfs_trans_t *tp; | |
1068 | int error; | |
1069 | ||
1070 | if (!xfs_fs_writable(mp)) | |
1071 | return 0; | |
1072 | ||
d4d90b57 | 1073 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1074 | |
1075 | /* | |
1076 | * we don't need to do this if we are updating the superblock | |
1077 | * counters on every modification. | |
1078 | */ | |
1079 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1080 | return 0; | |
1081 | ||
b2ce3974 | 1082 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
3d3c8b52 | 1083 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
92821e2b DC |
1084 | if (error) { |
1085 | xfs_trans_cancel(tp, 0); | |
1086 | return error; | |
1087 | } | |
1088 | ||
1089 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1090 | xfs_trans_set_sync(tp); |
e5720eec DC |
1091 | error = xfs_trans_commit(tp, 0); |
1092 | return error; | |
92821e2b DC |
1093 | } |
1094 | ||
1da177e4 | 1095 | /* |
99e738b7 | 1096 | * xfs_mod_incore_sb_unlocked() is a utility routine commonly used to apply |
1da177e4 LT |
1097 | * a delta to a specified field in the in-core superblock. Simply |
1098 | * switch on the field indicated and apply the delta to that field. | |
1099 | * Fields are not allowed to dip below zero, so if the delta would | |
1100 | * do this do not apply it and return EINVAL. | |
1101 | * | |
3685c2a1 | 1102 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1103 | */ |
d96f8f89 | 1104 | STATIC int |
20f4ebf2 DC |
1105 | xfs_mod_incore_sb_unlocked( |
1106 | xfs_mount_t *mp, | |
1107 | xfs_sb_field_t field, | |
1108 | int64_t delta, | |
1109 | int rsvd) | |
1da177e4 LT |
1110 | { |
1111 | int scounter; /* short counter for 32 bit fields */ | |
1112 | long long lcounter; /* long counter for 64 bit fields */ | |
1113 | long long res_used, rem; | |
1114 | ||
1115 | /* | |
1116 | * With the in-core superblock spin lock held, switch | |
1117 | * on the indicated field. Apply the delta to the | |
1118 | * proper field. If the fields value would dip below | |
1119 | * 0, then do not apply the delta and return EINVAL. | |
1120 | */ | |
1121 | switch (field) { | |
1122 | case XFS_SBS_ICOUNT: | |
1123 | lcounter = (long long)mp->m_sb.sb_icount; | |
1124 | lcounter += delta; | |
1125 | if (lcounter < 0) { | |
1126 | ASSERT(0); | |
014c2544 | 1127 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1128 | } |
1129 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1130 | return 0; |
1da177e4 LT |
1131 | case XFS_SBS_IFREE: |
1132 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1133 | lcounter += delta; | |
1134 | if (lcounter < 0) { | |
1135 | ASSERT(0); | |
014c2544 | 1136 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1137 | } |
1138 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1139 | return 0; |
1da177e4 | 1140 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1141 | lcounter = (long long) |
1142 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1143 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1144 | ||
1145 | if (delta > 0) { /* Putting blocks back */ | |
1146 | if (res_used > delta) { | |
1147 | mp->m_resblks_avail += delta; | |
1148 | } else { | |
1149 | rem = delta - res_used; | |
1150 | mp->m_resblks_avail = mp->m_resblks; | |
1151 | lcounter += rem; | |
1152 | } | |
1153 | } else { /* Taking blocks away */ | |
1da177e4 | 1154 | lcounter += delta; |
8babd8a2 DC |
1155 | if (lcounter >= 0) { |
1156 | mp->m_sb.sb_fdblocks = lcounter + | |
1157 | XFS_ALLOC_SET_ASIDE(mp); | |
1158 | return 0; | |
1159 | } | |
1da177e4 | 1160 | |
8babd8a2 DC |
1161 | /* |
1162 | * We are out of blocks, use any available reserved | |
1163 | * blocks if were allowed to. | |
1164 | */ | |
1165 | if (!rsvd) | |
1166 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1167 | |
8babd8a2 DC |
1168 | lcounter = (long long)mp->m_resblks_avail + delta; |
1169 | if (lcounter >= 0) { | |
1170 | mp->m_resblks_avail = lcounter; | |
1171 | return 0; | |
1da177e4 | 1172 | } |
8babd8a2 DC |
1173 | printk_once(KERN_WARNING |
1174 | "Filesystem \"%s\": reserve blocks depleted! " | |
1175 | "Consider increasing reserve pool size.", | |
1176 | mp->m_fsname); | |
1177 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1178 | } |
1179 | ||
4be536de | 1180 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1181 | return 0; |
1da177e4 LT |
1182 | case XFS_SBS_FREXTENTS: |
1183 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1184 | lcounter += delta; | |
1185 | if (lcounter < 0) { | |
014c2544 | 1186 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1187 | } |
1188 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1189 | return 0; |
1da177e4 LT |
1190 | case XFS_SBS_DBLOCKS: |
1191 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1192 | lcounter += delta; | |
1193 | if (lcounter < 0) { | |
1194 | ASSERT(0); | |
014c2544 | 1195 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1196 | } |
1197 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1198 | return 0; |
1da177e4 LT |
1199 | case XFS_SBS_AGCOUNT: |
1200 | scounter = mp->m_sb.sb_agcount; | |
1201 | scounter += delta; | |
1202 | if (scounter < 0) { | |
1203 | ASSERT(0); | |
014c2544 | 1204 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1205 | } |
1206 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1207 | return 0; |
1da177e4 LT |
1208 | case XFS_SBS_IMAX_PCT: |
1209 | scounter = mp->m_sb.sb_imax_pct; | |
1210 | scounter += delta; | |
1211 | if (scounter < 0) { | |
1212 | ASSERT(0); | |
014c2544 | 1213 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1214 | } |
1215 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1216 | return 0; |
1da177e4 LT |
1217 | case XFS_SBS_REXTSIZE: |
1218 | scounter = mp->m_sb.sb_rextsize; | |
1219 | scounter += delta; | |
1220 | if (scounter < 0) { | |
1221 | ASSERT(0); | |
014c2544 | 1222 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1223 | } |
1224 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1225 | return 0; |
1da177e4 LT |
1226 | case XFS_SBS_RBMBLOCKS: |
1227 | scounter = mp->m_sb.sb_rbmblocks; | |
1228 | scounter += delta; | |
1229 | if (scounter < 0) { | |
1230 | ASSERT(0); | |
014c2544 | 1231 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1232 | } |
1233 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1234 | return 0; |
1da177e4 LT |
1235 | case XFS_SBS_RBLOCKS: |
1236 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1237 | lcounter += delta; | |
1238 | if (lcounter < 0) { | |
1239 | ASSERT(0); | |
014c2544 | 1240 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1241 | } |
1242 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1243 | return 0; |
1da177e4 LT |
1244 | case XFS_SBS_REXTENTS: |
1245 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1246 | lcounter += delta; | |
1247 | if (lcounter < 0) { | |
1248 | ASSERT(0); | |
014c2544 | 1249 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1250 | } |
1251 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1252 | return 0; |
1da177e4 LT |
1253 | case XFS_SBS_REXTSLOG: |
1254 | scounter = mp->m_sb.sb_rextslog; | |
1255 | scounter += delta; | |
1256 | if (scounter < 0) { | |
1257 | ASSERT(0); | |
014c2544 | 1258 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1259 | } |
1260 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1261 | return 0; |
1da177e4 LT |
1262 | default: |
1263 | ASSERT(0); | |
014c2544 | 1264 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1265 | } |
1266 | } | |
1267 | ||
1268 | /* | |
1269 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1270 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1271 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1272 | * routine to do the work. |
1273 | */ | |
1274 | int | |
20f4ebf2 | 1275 | xfs_mod_incore_sb( |
96540c78 CH |
1276 | struct xfs_mount *mp, |
1277 | xfs_sb_field_t field, | |
1278 | int64_t delta, | |
1279 | int rsvd) | |
1da177e4 | 1280 | { |
96540c78 | 1281 | int status; |
1da177e4 | 1282 | |
8d280b98 | 1283 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1284 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1285 | #endif |
96540c78 CH |
1286 | spin_lock(&mp->m_sb_lock); |
1287 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1288 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1289 | |
014c2544 | 1290 | return status; |
1da177e4 LT |
1291 | } |
1292 | ||
1293 | /* | |
1b040712 | 1294 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1295 | * |
1b040712 CH |
1296 | * The fields and changes to those fields are specified in the array of |
1297 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1298 | * will be applied or none of them will. If any modified field dips below 0, | |
1299 | * then all modifications will be backed out and EINVAL will be returned. | |
1300 | * | |
1301 | * Note that this function may not be used for the superblock values that | |
1302 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1303 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1304 | */ |
1305 | int | |
1b040712 CH |
1306 | xfs_mod_incore_sb_batch( |
1307 | struct xfs_mount *mp, | |
1308 | xfs_mod_sb_t *msb, | |
1309 | uint nmsb, | |
1310 | int rsvd) | |
1da177e4 | 1311 | { |
45c51b99 | 1312 | xfs_mod_sb_t *msbp; |
1b040712 | 1313 | int error = 0; |
1da177e4 LT |
1314 | |
1315 | /* | |
1b040712 CH |
1316 | * Loop through the array of mod structures and apply each individually. |
1317 | * If any fail, then back out all those which have already been applied. | |
1318 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1319 | * changes will be atomic. | |
1da177e4 | 1320 | */ |
3685c2a1 | 1321 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1322 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1323 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1324 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1325 | |
1b040712 CH |
1326 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1327 | msbp->msb_delta, rsvd); | |
1328 | if (error) | |
1329 | goto unwind; | |
1da177e4 | 1330 | } |
1b040712 CH |
1331 | spin_unlock(&mp->m_sb_lock); |
1332 | return 0; | |
1da177e4 | 1333 | |
1b040712 CH |
1334 | unwind: |
1335 | while (--msbp >= msb) { | |
1336 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1337 | -msbp->msb_delta, rsvd); | |
1338 | ASSERT(error == 0); | |
1da177e4 | 1339 | } |
3685c2a1 | 1340 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1341 | return error; |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | /* | |
1345 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1346 | * The buffer is returned locked and read in from disk. | |
1347 | * The buffer should be released with a call to xfs_brelse(). | |
1348 | * | |
1349 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1350 | * the superblock buffer if it can be locked without sleeping. | |
1351 | * If it can't then we'll return NULL. | |
1352 | */ | |
0c842ad4 | 1353 | struct xfs_buf * |
1da177e4 | 1354 | xfs_getsb( |
0c842ad4 CH |
1355 | struct xfs_mount *mp, |
1356 | int flags) | |
1da177e4 | 1357 | { |
0c842ad4 | 1358 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1359 | |
0c842ad4 CH |
1360 | if (!xfs_buf_trylock(bp)) { |
1361 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1362 | return NULL; |
0c842ad4 | 1363 | xfs_buf_lock(bp); |
1da177e4 | 1364 | } |
0c842ad4 | 1365 | |
72790aa1 | 1366 | xfs_buf_hold(bp); |
1da177e4 | 1367 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 1368 | return bp; |
1da177e4 LT |
1369 | } |
1370 | ||
1371 | /* | |
1372 | * Used to free the superblock along various error paths. | |
1373 | */ | |
1374 | void | |
1375 | xfs_freesb( | |
26af6552 | 1376 | struct xfs_mount *mp) |
1da177e4 | 1377 | { |
26af6552 | 1378 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1379 | |
26af6552 | 1380 | xfs_buf_lock(bp); |
1da177e4 | 1381 | mp->m_sb_bp = NULL; |
26af6552 | 1382 | xfs_buf_relse(bp); |
1da177e4 LT |
1383 | } |
1384 | ||
1da177e4 LT |
1385 | /* |
1386 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1387 | * be altered by the mount options, as well as any potential sb_features2 |
1388 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1389 | */ |
7884bc86 | 1390 | int |
ee1c0908 | 1391 | xfs_mount_log_sb( |
1da177e4 LT |
1392 | xfs_mount_t *mp, |
1393 | __int64_t fields) | |
1394 | { | |
1395 | xfs_trans_t *tp; | |
e5720eec | 1396 | int error; |
1da177e4 | 1397 | |
ee1c0908 | 1398 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1399 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1400 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1401 | |
1402 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
3d3c8b52 | 1403 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0); |
e5720eec | 1404 | if (error) { |
1da177e4 | 1405 | xfs_trans_cancel(tp, 0); |
e5720eec | 1406 | return error; |
1da177e4 LT |
1407 | } |
1408 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1409 | error = xfs_trans_commit(tp, 0); |
1410 | return error; | |
1da177e4 | 1411 | } |
8d280b98 | 1412 | |
dda35b8f CH |
1413 | /* |
1414 | * If the underlying (data/log/rt) device is readonly, there are some | |
1415 | * operations that cannot proceed. | |
1416 | */ | |
1417 | int | |
1418 | xfs_dev_is_read_only( | |
1419 | struct xfs_mount *mp, | |
1420 | char *message) | |
1421 | { | |
1422 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
1423 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
1424 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
1425 | xfs_notice(mp, "%s required on read-only device.", message); |
1426 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
1427 | return EROFS; |
1428 | } | |
1429 | return 0; | |
1430 | } | |
8d280b98 DC |
1431 | |
1432 | #ifdef HAVE_PERCPU_SB | |
1433 | /* | |
1434 | * Per-cpu incore superblock counters | |
1435 | * | |
1436 | * Simple concept, difficult implementation | |
1437 | * | |
1438 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1439 | * counter for contended fields (e.g. free block count). | |
1440 | * | |
1441 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1442 | * hence needs to be accurately read when we are running low on space. Hence | |
1443 | * there is a method to enable and disable the per-cpu counters based on how | |
1444 | * much "stuff" is available in them. | |
1445 | * | |
1446 | * Basically, a counter is enabled if there is enough free resource to justify | |
1447 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
1448 | * ENOSPC), then we disable the counters to synchronise all callers and | |
1449 | * re-distribute the available resources. | |
1450 | * | |
1451 | * If, once we redistributed the available resources, we still get a failure, | |
1452 | * we disable the per-cpu counter and go through the slow path. | |
1453 | * | |
1454 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 1455 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
1456 | * the global superblock. We do this after disabling the counter to prevent |
1457 | * more threads from queueing up on the counter. | |
1458 | * | |
1459 | * Essentially, this means that we still need a lock in the fast path to enable | |
1460 | * synchronisation between the global counters and the per-cpu counters. This | |
1461 | * is not a problem because the lock will be local to a CPU almost all the time | |
1462 | * and have little contention except when we get to ENOSPC conditions. | |
1463 | * | |
1464 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
1465 | * path while we do things like enabling and disabling counters and | |
1466 | * synchronising the counters. | |
1467 | * | |
1468 | * Locking rules: | |
1469 | * | |
3685c2a1 | 1470 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 1471 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 1472 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 1473 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
1474 | * 5. modifying global counters requires holding m_sb_lock |
1475 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
1476 | * and _none_ of the per-cpu locks. |
1477 | * | |
1478 | * Disabled counters are only ever re-enabled by a balance operation | |
1479 | * that results in more free resources per CPU than a given threshold. | |
1480 | * To ensure counters don't remain disabled, they are rebalanced when | |
1481 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
1482 | * is present to prevent thrashing). | |
e8234a68 DC |
1483 | */ |
1484 | ||
5a67e4c5 | 1485 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
1486 | /* |
1487 | * hot-plug CPU notifier support. | |
8d280b98 | 1488 | * |
5a67e4c5 CS |
1489 | * We need a notifier per filesystem as we need to be able to identify |
1490 | * the filesystem to balance the counters out. This is achieved by | |
1491 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
1492 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 1493 | */ |
e8234a68 DC |
1494 | STATIC int |
1495 | xfs_icsb_cpu_notify( | |
1496 | struct notifier_block *nfb, | |
1497 | unsigned long action, | |
1498 | void *hcpu) | |
1499 | { | |
1500 | xfs_icsb_cnts_t *cntp; | |
1501 | xfs_mount_t *mp; | |
e8234a68 DC |
1502 | |
1503 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
1504 | cntp = (xfs_icsb_cnts_t *) | |
1505 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
1506 | switch (action) { | |
1507 | case CPU_UP_PREPARE: | |
8bb78442 | 1508 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
1509 | /* Easy Case - initialize the area and locks, and |
1510 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 1511 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
1512 | break; |
1513 | case CPU_ONLINE: | |
8bb78442 | 1514 | case CPU_ONLINE_FROZEN: |
03135cf7 | 1515 | xfs_icsb_lock(mp); |
45af6c6d CH |
1516 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1517 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1518 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 1519 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1520 | break; |
1521 | case CPU_DEAD: | |
8bb78442 | 1522 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
1523 | /* Disable all the counters, then fold the dead cpu's |
1524 | * count into the total on the global superblock and | |
1525 | * re-enable the counters. */ | |
03135cf7 | 1526 | xfs_icsb_lock(mp); |
3685c2a1 | 1527 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
1528 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
1529 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
1530 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
1531 | ||
1532 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
1533 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
1534 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
1535 | ||
01e1b69c | 1536 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 1537 | |
45af6c6d CH |
1538 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
1539 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
1540 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 1541 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 1542 | xfs_icsb_unlock(mp); |
e8234a68 DC |
1543 | break; |
1544 | } | |
1545 | ||
1546 | return NOTIFY_OK; | |
1547 | } | |
5a67e4c5 | 1548 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 1549 | |
8d280b98 DC |
1550 | int |
1551 | xfs_icsb_init_counters( | |
1552 | xfs_mount_t *mp) | |
1553 | { | |
1554 | xfs_icsb_cnts_t *cntp; | |
1555 | int i; | |
1556 | ||
1557 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
1558 | if (mp->m_sb_cnts == NULL) | |
1559 | return -ENOMEM; | |
1560 | ||
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; | |
46677e67 RW |
1573 | |
1574 | #ifdef CONFIG_HOTPLUG_CPU | |
1575 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; | |
1576 | mp->m_icsb_notifier.priority = 0; | |
1577 | register_hotcpu_notifier(&mp->m_icsb_notifier); | |
1578 | #endif /* CONFIG_HOTPLUG_CPU */ | |
1579 | ||
8d280b98 DC |
1580 | return 0; |
1581 | } | |
1582 | ||
5478eead LM |
1583 | void |
1584 | xfs_icsb_reinit_counters( | |
1585 | xfs_mount_t *mp) | |
1586 | { | |
1587 | xfs_icsb_lock(mp); | |
1588 | /* | |
1589 | * start with all counters disabled so that the | |
1590 | * initial balance kicks us off correctly | |
1591 | */ | |
1592 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
1593 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
1594 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
1595 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
1596 | xfs_icsb_unlock(mp); |
1597 | } | |
1598 | ||
c962fb79 | 1599 | void |
8d280b98 DC |
1600 | xfs_icsb_destroy_counters( |
1601 | xfs_mount_t *mp) | |
1602 | { | |
e8234a68 | 1603 | if (mp->m_sb_cnts) { |
5a67e4c5 | 1604 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 1605 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 1606 | } |
03135cf7 | 1607 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
1608 | } |
1609 | ||
b8f82a4a | 1610 | STATIC void |
01e1b69c DC |
1611 | xfs_icsb_lock_cntr( |
1612 | xfs_icsb_cnts_t *icsbp) | |
1613 | { | |
1614 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
1615 | ndelay(1000); | |
1616 | } | |
1617 | } | |
1618 | ||
b8f82a4a | 1619 | STATIC void |
01e1b69c DC |
1620 | xfs_icsb_unlock_cntr( |
1621 | xfs_icsb_cnts_t *icsbp) | |
1622 | { | |
1623 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
1624 | } | |
1625 | ||
8d280b98 | 1626 | |
b8f82a4a | 1627 | STATIC void |
8d280b98 DC |
1628 | xfs_icsb_lock_all_counters( |
1629 | xfs_mount_t *mp) | |
1630 | { | |
1631 | xfs_icsb_cnts_t *cntp; | |
1632 | int i; | |
1633 | ||
1634 | for_each_online_cpu(i) { | |
1635 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1636 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
1637 | } |
1638 | } | |
1639 | ||
b8f82a4a | 1640 | STATIC void |
8d280b98 DC |
1641 | xfs_icsb_unlock_all_counters( |
1642 | xfs_mount_t *mp) | |
1643 | { | |
1644 | xfs_icsb_cnts_t *cntp; | |
1645 | int i; | |
1646 | ||
1647 | for_each_online_cpu(i) { | |
1648 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 1649 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
1650 | } |
1651 | } | |
1652 | ||
1653 | STATIC void | |
1654 | xfs_icsb_count( | |
1655 | xfs_mount_t *mp, | |
1656 | xfs_icsb_cnts_t *cnt, | |
1657 | int flags) | |
1658 | { | |
1659 | xfs_icsb_cnts_t *cntp; | |
1660 | int i; | |
1661 | ||
1662 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
1663 | ||
1664 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1665 | xfs_icsb_lock_all_counters(mp); | |
1666 | ||
1667 | for_each_online_cpu(i) { | |
1668 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
1669 | cnt->icsb_icount += cntp->icsb_icount; | |
1670 | cnt->icsb_ifree += cntp->icsb_ifree; | |
1671 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
1672 | } | |
1673 | ||
1674 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
1675 | xfs_icsb_unlock_all_counters(mp); | |
1676 | } | |
1677 | ||
1678 | STATIC int | |
1679 | xfs_icsb_counter_disabled( | |
1680 | xfs_mount_t *mp, | |
1681 | xfs_sb_field_t field) | |
1682 | { | |
1683 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1684 | return test_bit(field, &mp->m_icsb_counters); | |
1685 | } | |
1686 | ||
36fbe6e6 | 1687 | STATIC void |
8d280b98 DC |
1688 | xfs_icsb_disable_counter( |
1689 | xfs_mount_t *mp, | |
1690 | xfs_sb_field_t field) | |
1691 | { | |
1692 | xfs_icsb_cnts_t cnt; | |
1693 | ||
1694 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1695 | ||
20b64285 DC |
1696 | /* |
1697 | * If we are already disabled, then there is nothing to do | |
1698 | * here. We check before locking all the counters to avoid | |
1699 | * the expensive lock operation when being called in the | |
1700 | * slow path and the counter is already disabled. This is | |
1701 | * safe because the only time we set or clear this state is under | |
1702 | * the m_icsb_mutex. | |
1703 | */ | |
1704 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 1705 | return; |
20b64285 | 1706 | |
8d280b98 DC |
1707 | xfs_icsb_lock_all_counters(mp); |
1708 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
1709 | /* drain back to superblock */ | |
1710 | ||
ce46193b | 1711 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
1712 | switch(field) { |
1713 | case XFS_SBS_ICOUNT: | |
1714 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1715 | break; | |
1716 | case XFS_SBS_IFREE: | |
1717 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1718 | break; | |
1719 | case XFS_SBS_FDBLOCKS: | |
1720 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
1721 | break; | |
1722 | default: | |
1723 | BUG(); | |
1724 | } | |
1725 | } | |
1726 | ||
1727 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
1728 | } |
1729 | ||
1730 | STATIC void | |
1731 | xfs_icsb_enable_counter( | |
1732 | xfs_mount_t *mp, | |
1733 | xfs_sb_field_t field, | |
1734 | uint64_t count, | |
1735 | uint64_t resid) | |
1736 | { | |
1737 | xfs_icsb_cnts_t *cntp; | |
1738 | int i; | |
1739 | ||
1740 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
1741 | ||
1742 | xfs_icsb_lock_all_counters(mp); | |
1743 | for_each_online_cpu(i) { | |
1744 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
1745 | switch (field) { | |
1746 | case XFS_SBS_ICOUNT: | |
1747 | cntp->icsb_icount = count + resid; | |
1748 | break; | |
1749 | case XFS_SBS_IFREE: | |
1750 | cntp->icsb_ifree = count + resid; | |
1751 | break; | |
1752 | case XFS_SBS_FDBLOCKS: | |
1753 | cntp->icsb_fdblocks = count + resid; | |
1754 | break; | |
1755 | default: | |
1756 | BUG(); | |
1757 | break; | |
1758 | } | |
1759 | resid = 0; | |
1760 | } | |
1761 | clear_bit(field, &mp->m_icsb_counters); | |
1762 | xfs_icsb_unlock_all_counters(mp); | |
1763 | } | |
1764 | ||
dbcabad1 | 1765 | void |
d4d90b57 | 1766 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
1767 | xfs_mount_t *mp, |
1768 | int flags) | |
1769 | { | |
1770 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 1771 | |
8d280b98 DC |
1772 | xfs_icsb_count(mp, &cnt, flags); |
1773 | ||
8d280b98 DC |
1774 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
1775 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
1776 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
1777 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
1778 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
1779 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
1780 | } |
1781 | ||
1782 | /* | |
1783 | * Accurate update of per-cpu counters to incore superblock | |
1784 | */ | |
d4d90b57 | 1785 | void |
8d280b98 | 1786 | xfs_icsb_sync_counters( |
d4d90b57 CH |
1787 | xfs_mount_t *mp, |
1788 | int flags) | |
8d280b98 | 1789 | { |
d4d90b57 CH |
1790 | spin_lock(&mp->m_sb_lock); |
1791 | xfs_icsb_sync_counters_locked(mp, flags); | |
1792 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1793 | } |
1794 | ||
1795 | /* | |
1796 | * Balance and enable/disable counters as necessary. | |
1797 | * | |
20b64285 DC |
1798 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
1799 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
1800 | * free blocks is something far enough zero that we aren't going thrash when we | |
1801 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
1802 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
1803 | * be distributed to a single CPU but each CPU has enough blocks to be | |
1804 | * reenabled. | |
1805 | * | |
1806 | * Note that we can be called when counters are already disabled. | |
1807 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
1808 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 1809 | */ |
20b64285 DC |
1810 | |
1811 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 1812 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 1813 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 1814 | STATIC void |
45af6c6d | 1815 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
1816 | xfs_mount_t *mp, |
1817 | xfs_sb_field_t field, | |
20b64285 | 1818 | int min_per_cpu) |
8d280b98 | 1819 | { |
6fdf8ccc | 1820 | uint64_t count, resid; |
8d280b98 | 1821 | int weight = num_online_cpus(); |
20b64285 | 1822 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 1823 | |
8d280b98 DC |
1824 | /* disable counter and sync counter */ |
1825 | xfs_icsb_disable_counter(mp, field); | |
1826 | ||
1827 | /* update counters - first CPU gets residual*/ | |
1828 | switch (field) { | |
1829 | case XFS_SBS_ICOUNT: | |
1830 | count = mp->m_sb.sb_icount; | |
1831 | resid = do_div(count, weight); | |
20b64285 | 1832 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1833 | return; |
8d280b98 DC |
1834 | break; |
1835 | case XFS_SBS_IFREE: | |
1836 | count = mp->m_sb.sb_ifree; | |
1837 | resid = do_div(count, weight); | |
20b64285 | 1838 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 1839 | return; |
8d280b98 DC |
1840 | break; |
1841 | case XFS_SBS_FDBLOCKS: | |
1842 | count = mp->m_sb.sb_fdblocks; | |
1843 | resid = do_div(count, weight); | |
20b64285 | 1844 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 1845 | return; |
8d280b98 DC |
1846 | break; |
1847 | default: | |
1848 | BUG(); | |
6fdf8ccc | 1849 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
1850 | break; |
1851 | } | |
1852 | ||
1853 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
1854 | } |
1855 | ||
1856 | STATIC void | |
1857 | xfs_icsb_balance_counter( | |
1858 | xfs_mount_t *mp, | |
1859 | xfs_sb_field_t fields, | |
1860 | int min_per_cpu) | |
1861 | { | |
1862 | spin_lock(&mp->m_sb_lock); | |
1863 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
1864 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
1865 | } |
1866 | ||
1b040712 | 1867 | int |
20b64285 | 1868 | xfs_icsb_modify_counters( |
8d280b98 DC |
1869 | xfs_mount_t *mp, |
1870 | xfs_sb_field_t field, | |
20f4ebf2 | 1871 | int64_t delta, |
20b64285 | 1872 | int rsvd) |
8d280b98 DC |
1873 | { |
1874 | xfs_icsb_cnts_t *icsbp; | |
1875 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 1876 | int ret = 0; |
8d280b98 | 1877 | |
20b64285 | 1878 | might_sleep(); |
8d280b98 | 1879 | again: |
7a9e02d6 CL |
1880 | preempt_disable(); |
1881 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
1882 | |
1883 | /* | |
1884 | * if the counter is disabled, go to slow path | |
1885 | */ | |
8d280b98 DC |
1886 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
1887 | goto slow_path; | |
20b64285 DC |
1888 | xfs_icsb_lock_cntr(icsbp); |
1889 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
1890 | xfs_icsb_unlock_cntr(icsbp); | |
1891 | goto slow_path; | |
1892 | } | |
8d280b98 DC |
1893 | |
1894 | switch (field) { | |
1895 | case XFS_SBS_ICOUNT: | |
1896 | lcounter = icsbp->icsb_icount; | |
1897 | lcounter += delta; | |
1898 | if (unlikely(lcounter < 0)) | |
20b64285 | 1899 | goto balance_counter; |
8d280b98 DC |
1900 | icsbp->icsb_icount = lcounter; |
1901 | break; | |
1902 | ||
1903 | case XFS_SBS_IFREE: | |
1904 | lcounter = icsbp->icsb_ifree; | |
1905 | lcounter += delta; | |
1906 | if (unlikely(lcounter < 0)) | |
20b64285 | 1907 | goto balance_counter; |
8d280b98 DC |
1908 | icsbp->icsb_ifree = lcounter; |
1909 | break; | |
1910 | ||
1911 | case XFS_SBS_FDBLOCKS: | |
1912 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
1913 | ||
4be536de | 1914 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1915 | lcounter += delta; |
1916 | if (unlikely(lcounter < 0)) | |
20b64285 | 1917 | goto balance_counter; |
4be536de | 1918 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
1919 | break; |
1920 | default: | |
1921 | BUG(); | |
1922 | break; | |
1923 | } | |
01e1b69c | 1924 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 1925 | preempt_enable(); |
8d280b98 DC |
1926 | return 0; |
1927 | ||
8d280b98 | 1928 | slow_path: |
7a9e02d6 | 1929 | preempt_enable(); |
8d280b98 | 1930 | |
20b64285 DC |
1931 | /* |
1932 | * serialise with a mutex so we don't burn lots of cpu on | |
1933 | * the superblock lock. We still need to hold the superblock | |
1934 | * lock, however, when we modify the global structures. | |
1935 | */ | |
03135cf7 | 1936 | xfs_icsb_lock(mp); |
20b64285 DC |
1937 | |
1938 | /* | |
1939 | * Now running atomically. | |
1940 | * | |
1941 | * If the counter is enabled, someone has beaten us to rebalancing. | |
1942 | * Drop the lock and try again in the fast path.... | |
1943 | */ | |
1944 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 1945 | xfs_icsb_unlock(mp); |
8d280b98 | 1946 | goto again; |
8d280b98 DC |
1947 | } |
1948 | ||
20b64285 DC |
1949 | /* |
1950 | * The counter is currently disabled. Because we are | |
1951 | * running atomically here, we know a rebalance cannot | |
1952 | * be in progress. Hence we can go straight to operating | |
1953 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 1954 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 1955 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 1956 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
1957 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
1958 | * directly on the global counters. | |
1959 | */ | |
3685c2a1 | 1960 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1961 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1962 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 1963 | |
20b64285 DC |
1964 | /* |
1965 | * Now that we've modified the global superblock, we | |
1966 | * may be able to re-enable the distributed counters | |
1967 | * (e.g. lots of space just got freed). After that | |
1968 | * we are done. | |
1969 | */ | |
1970 | if (ret != ENOSPC) | |
45af6c6d | 1971 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 1972 | xfs_icsb_unlock(mp); |
8d280b98 | 1973 | return ret; |
8d280b98 | 1974 | |
20b64285 DC |
1975 | balance_counter: |
1976 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 1977 | preempt_enable(); |
8d280b98 | 1978 | |
20b64285 DC |
1979 | /* |
1980 | * We may have multiple threads here if multiple per-cpu | |
1981 | * counters run dry at the same time. This will mean we can | |
1982 | * do more balances than strictly necessary but it is not | |
1983 | * the common slowpath case. | |
1984 | */ | |
03135cf7 | 1985 | xfs_icsb_lock(mp); |
20b64285 DC |
1986 | |
1987 | /* | |
1988 | * running atomically. | |
1989 | * | |
1990 | * This will leave the counter in the correct state for future | |
1991 | * accesses. After the rebalance, we simply try again and our retry | |
1992 | * will either succeed through the fast path or slow path without | |
1993 | * another balance operation being required. | |
1994 | */ | |
45af6c6d | 1995 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 1996 | xfs_icsb_unlock(mp); |
20b64285 | 1997 | goto again; |
8d280b98 | 1998 | } |
20b64285 | 1999 | |
8d280b98 | 2000 | #endif |