]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
6ca1c906 | 20 | #include "xfs_format.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 | 22 | #include "xfs_log.h" |
a844f451 | 23 | #include "xfs_inum.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
211e4d43 | 25 | #include "xfs_trans_priv.h" |
1da177e4 LT |
26 | #include "xfs_sb.h" |
27 | #include "xfs_ag.h" | |
1da177e4 | 28 | #include "xfs_mount.h" |
2b9ab5ab DC |
29 | #include "xfs_da_btree.h" |
30 | #include "xfs_dir2_format.h" | |
31 | #include "xfs_dir2.h" | |
1da177e4 | 32 | #include "xfs_bmap_btree.h" |
a844f451 | 33 | #include "xfs_alloc_btree.h" |
1da177e4 | 34 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_alloc.h" |
40 | #include "xfs_rtalloc.h" | |
41 | #include "xfs_bmap.h" | |
42 | #include "xfs_error.h" | |
1da177e4 LT |
43 | #include "xfs_quota.h" |
44 | #include "xfs_fsops.h" | |
43355099 | 45 | #include "xfs_utils.h" |
0b1b213f | 46 | #include "xfs_trace.h" |
6d8b79cf | 47 | #include "xfs_icache.h" |
04a1e6c5 DC |
48 | #include "xfs_cksum.h" |
49 | #include "xfs_buf_item.h" | |
0b1b213f | 50 | |
1da177e4 | 51 | |
8d280b98 | 52 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 53 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
54 | int); |
55 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
56 | int); | |
36fbe6e6 | 57 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
58 | #else |
59 | ||
45af6c6d CH |
60 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
61 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
62 | #endif |
63 | ||
1df84c93 | 64 | static const struct { |
8d280b98 DC |
65 | short offset; |
66 | short type; /* 0 = integer | |
67 | * 1 = binary / string (no translation) | |
68 | */ | |
1da177e4 LT |
69 | } xfs_sb_info[] = { |
70 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
71 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
72 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
73 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
74 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
75 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
76 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
77 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
78 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
79 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
80 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
81 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
82 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
84 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
87 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
88 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
90 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
95 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
96 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
98 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
101 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
111 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
113 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
114 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
ee1c0908 | 115 | { offsetof(xfs_sb_t, sb_bad_features2), 0 }, |
04a1e6c5 DC |
116 | { offsetof(xfs_sb_t, sb_features_compat), 0 }, |
117 | { offsetof(xfs_sb_t, sb_features_ro_compat), 0 }, | |
118 | { offsetof(xfs_sb_t, sb_features_incompat), 0 }, | |
e721f504 | 119 | { offsetof(xfs_sb_t, sb_features_log_incompat), 0 }, |
04a1e6c5 | 120 | { offsetof(xfs_sb_t, sb_crc), 0 }, |
e721f504 | 121 | { offsetof(xfs_sb_t, sb_pad), 0 }, |
04a1e6c5 DC |
122 | { offsetof(xfs_sb_t, sb_pquotino), 0 }, |
123 | { offsetof(xfs_sb_t, sb_lsn), 0 }, | |
1da177e4 LT |
124 | { sizeof(xfs_sb_t), 0 } |
125 | }; | |
126 | ||
27174203 CH |
127 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
128 | static int xfs_uuid_table_size; | |
129 | static uuid_t *xfs_uuid_table; | |
130 | ||
131 | /* | |
132 | * See if the UUID is unique among mounted XFS filesystems. | |
133 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
134 | */ | |
135 | STATIC int | |
136 | xfs_uuid_mount( | |
137 | struct xfs_mount *mp) | |
138 | { | |
139 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
140 | int hole, i; | |
141 | ||
142 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
143 | return 0; | |
144 | ||
145 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 146 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
147 | return XFS_ERROR(EINVAL); |
148 | } | |
149 | ||
150 | mutex_lock(&xfs_uuid_table_mutex); | |
151 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
152 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
153 | hole = i; | |
154 | continue; | |
155 | } | |
156 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
157 | goto out_duplicate; | |
158 | } | |
159 | ||
160 | if (hole < 0) { | |
161 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
162 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
163 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
164 | KM_SLEEP); | |
165 | hole = xfs_uuid_table_size++; | |
166 | } | |
167 | xfs_uuid_table[hole] = *uuid; | |
168 | mutex_unlock(&xfs_uuid_table_mutex); | |
169 | ||
170 | return 0; | |
171 | ||
172 | out_duplicate: | |
173 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 174 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
27174203 CH |
175 | return XFS_ERROR(EINVAL); |
176 | } | |
177 | ||
178 | STATIC void | |
179 | xfs_uuid_unmount( | |
180 | struct xfs_mount *mp) | |
181 | { | |
182 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
183 | int i; | |
184 | ||
185 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
186 | return; | |
187 | ||
188 | mutex_lock(&xfs_uuid_table_mutex); | |
189 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
190 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
191 | continue; | |
192 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
193 | continue; | |
194 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
195 | break; | |
196 | } | |
197 | ASSERT(i < xfs_uuid_table_size); | |
198 | mutex_unlock(&xfs_uuid_table_mutex); | |
199 | } | |
200 | ||
201 | ||
0fa800fb DC |
202 | /* |
203 | * Reference counting access wrappers to the perag structures. | |
e176579e DC |
204 | * Because we never free per-ag structures, the only thing we |
205 | * have to protect against changes is the tree structure itself. | |
0fa800fb DC |
206 | */ |
207 | struct xfs_perag * | |
208 | xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno) | |
209 | { | |
210 | struct xfs_perag *pag; | |
211 | int ref = 0; | |
212 | ||
e176579e | 213 | rcu_read_lock(); |
0fa800fb DC |
214 | pag = radix_tree_lookup(&mp->m_perag_tree, agno); |
215 | if (pag) { | |
216 | ASSERT(atomic_read(&pag->pag_ref) >= 0); | |
0fa800fb DC |
217 | ref = atomic_inc_return(&pag->pag_ref); |
218 | } | |
e176579e | 219 | rcu_read_unlock(); |
0fa800fb DC |
220 | trace_xfs_perag_get(mp, agno, ref, _RET_IP_); |
221 | return pag; | |
222 | } | |
223 | ||
65d0f205 DC |
224 | /* |
225 | * search from @first to find the next perag with the given tag set. | |
226 | */ | |
227 | struct xfs_perag * | |
228 | xfs_perag_get_tag( | |
229 | struct xfs_mount *mp, | |
230 | xfs_agnumber_t first, | |
231 | int tag) | |
232 | { | |
233 | struct xfs_perag *pag; | |
234 | int found; | |
235 | int ref; | |
236 | ||
237 | rcu_read_lock(); | |
238 | found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, | |
239 | (void **)&pag, first, 1, tag); | |
240 | if (found <= 0) { | |
241 | rcu_read_unlock(); | |
242 | return NULL; | |
243 | } | |
244 | ref = atomic_inc_return(&pag->pag_ref); | |
245 | rcu_read_unlock(); | |
246 | trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_); | |
247 | return pag; | |
248 | } | |
249 | ||
0fa800fb DC |
250 | void |
251 | xfs_perag_put(struct xfs_perag *pag) | |
252 | { | |
253 | int ref; | |
254 | ||
255 | ASSERT(atomic_read(&pag->pag_ref) > 0); | |
256 | ref = atomic_dec_return(&pag->pag_ref); | |
257 | trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_); | |
258 | } | |
259 | ||
e176579e DC |
260 | STATIC void |
261 | __xfs_free_perag( | |
262 | struct rcu_head *head) | |
263 | { | |
264 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
265 | ||
266 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
267 | kmem_free(pag); | |
268 | } | |
269 | ||
1da177e4 | 270 | /* |
e176579e | 271 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 272 | */ |
c962fb79 | 273 | STATIC void |
ff4f038c | 274 | xfs_free_perag( |
745f6919 | 275 | xfs_mount_t *mp) |
1da177e4 | 276 | { |
1c1c6ebc DC |
277 | xfs_agnumber_t agno; |
278 | struct xfs_perag *pag; | |
279 | ||
280 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
281 | spin_lock(&mp->m_perag_lock); | |
282 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
283 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 284 | ASSERT(pag); |
f83282a8 | 285 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 286 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 287 | } |
1da177e4 LT |
288 | } |
289 | ||
4cc929ee NS |
290 | /* |
291 | * Check size of device based on the (data/realtime) block count. | |
292 | * Note: this check is used by the growfs code as well as mount. | |
293 | */ | |
294 | int | |
295 | xfs_sb_validate_fsb_count( | |
296 | xfs_sb_t *sbp, | |
297 | __uint64_t nblocks) | |
298 | { | |
299 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
300 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
301 | ||
302 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
303 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 304 | return EFBIG; |
4cc929ee NS |
305 | #else /* Limited by UINT_MAX of sectors */ |
306 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 307 | return EFBIG; |
4cc929ee NS |
308 | #endif |
309 | return 0; | |
310 | } | |
1da177e4 LT |
311 | |
312 | /* | |
313 | * Check the validity of the SB found. | |
314 | */ | |
315 | STATIC int | |
316 | xfs_mount_validate_sb( | |
317 | xfs_mount_t *mp, | |
764d1f89 | 318 | xfs_sb_t *sbp, |
34510185 DC |
319 | bool check_inprogress, |
320 | bool check_version) | |
1da177e4 | 321 | { |
af34e09d | 322 | |
1da177e4 LT |
323 | /* |
324 | * If the log device and data device have the | |
325 | * same device number, the log is internal. | |
326 | * Consequently, the sb_logstart should be non-zero. If | |
327 | * we have a zero sb_logstart in this case, we may be trying to mount | |
328 | * a volume filesystem in a non-volume manner. | |
329 | */ | |
330 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
98021821 | 331 | xfs_warn(mp, "bad magic number"); |
1da177e4 LT |
332 | return XFS_ERROR(EWRONGFS); |
333 | } | |
334 | ||
04a1e6c5 | 335 | |
62118709 | 336 | if (!xfs_sb_good_version(sbp)) { |
98021821 | 337 | xfs_warn(mp, "bad version"); |
1da177e4 LT |
338 | return XFS_ERROR(EWRONGFS); |
339 | } | |
340 | ||
04a1e6c5 | 341 | /* |
e721f504 | 342 | * Version 5 superblock feature mask validation. Reject combinations the |
34510185 DC |
343 | * kernel cannot support up front before checking anything else. For |
344 | * write validation, we don't need to check feature masks. | |
04a1e6c5 | 345 | */ |
34510185 | 346 | if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) { |
04a1e6c5 | 347 | xfs_alert(mp, |
e721f504 DC |
348 | "Version 5 superblock detected. This kernel has EXPERIMENTAL support enabled!\n" |
349 | "Use of these features in this kernel is at your own risk!"); | |
350 | ||
351 | if (xfs_sb_has_compat_feature(sbp, | |
352 | XFS_SB_FEAT_COMPAT_UNKNOWN)) { | |
353 | xfs_warn(mp, | |
354 | "Superblock has unknown compatible features (0x%x) enabled.\n" | |
355 | "Using a more recent kernel is recommended.", | |
356 | (sbp->sb_features_compat & | |
357 | XFS_SB_FEAT_COMPAT_UNKNOWN)); | |
358 | } | |
359 | ||
360 | if (xfs_sb_has_ro_compat_feature(sbp, | |
361 | XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { | |
362 | xfs_alert(mp, | |
363 | "Superblock has unknown read-only compatible features (0x%x) enabled.", | |
364 | (sbp->sb_features_ro_compat & | |
365 | XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); | |
366 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { | |
367 | xfs_warn(mp, | |
368 | "Attempted to mount read-only compatible filesystem read-write.\n" | |
369 | "Filesystem can only be safely mounted read only."); | |
370 | return XFS_ERROR(EINVAL); | |
371 | } | |
372 | } | |
373 | if (xfs_sb_has_incompat_feature(sbp, | |
374 | XFS_SB_FEAT_INCOMPAT_UNKNOWN)) { | |
375 | xfs_warn(mp, | |
376 | "Superblock has unknown incompatible features (0x%x) enabled.\n" | |
377 | "Filesystem can not be safely mounted by this kernel.", | |
378 | (sbp->sb_features_incompat & | |
379 | XFS_SB_FEAT_INCOMPAT_UNKNOWN)); | |
380 | return XFS_ERROR(EINVAL); | |
381 | } | |
04a1e6c5 DC |
382 | } |
383 | ||
d892d586 CS |
384 | if (xfs_sb_version_has_pquotino(sbp)) { |
385 | if (sbp->sb_qflags & (XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD)) { | |
386 | xfs_notice(mp, | |
387 | "Version 5 of Super block has XFS_OQUOTA bits.\n"); | |
388 | return XFS_ERROR(EFSCORRUPTED); | |
389 | } | |
390 | } else if (sbp->sb_qflags & (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD | | |
391 | XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) { | |
392 | xfs_notice(mp, | |
393 | "Superblock earlier than Version 5 has XFS_[PQ]UOTA_{ENFD|CHKD} bits.\n"); | |
394 | return XFS_ERROR(EFSCORRUPTED); | |
395 | } | |
396 | ||
1da177e4 LT |
397 | if (unlikely( |
398 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
98021821 | 399 | xfs_warn(mp, |
af34e09d DC |
400 | "filesystem is marked as having an external log; " |
401 | "specify logdev on the mount command line."); | |
764d1f89 | 402 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
403 | } |
404 | ||
405 | if (unlikely( | |
406 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
98021821 | 407 | xfs_warn(mp, |
af34e09d DC |
408 | "filesystem is marked as having an internal log; " |
409 | "do not specify logdev on the mount command line."); | |
764d1f89 | 410 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
411 | } |
412 | ||
413 | /* | |
c0e090ce | 414 | * More sanity checking. Most of these were stolen directly from |
1da177e4 LT |
415 | * xfs_repair. |
416 | */ | |
417 | if (unlikely( | |
418 | sbp->sb_agcount <= 0 || | |
419 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
420 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
421 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
422 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 423 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
424 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
425 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
426 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
427 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 428 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
429 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
430 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
431 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
432 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 433 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 434 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
435 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
436 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
c0e090ce ES |
437 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) || |
438 | sbp->sb_dblocks == 0 || | |
439 | sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) || | |
440 | sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) { | |
98021821 | 441 | XFS_CORRUPTION_ERROR("SB sanity check failed", |
c0e090ce | 442 | XFS_ERRLEVEL_LOW, mp, sbp); |
1da177e4 LT |
443 | return XFS_ERROR(EFSCORRUPTED); |
444 | } | |
445 | ||
2edbddd5 LM |
446 | /* |
447 | * Until this is fixed only page-sized or smaller data blocks work. | |
448 | */ | |
449 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
98021821 | 450 | xfs_warn(mp, |
af34e09d DC |
451 | "File system with blocksize %d bytes. " |
452 | "Only pagesize (%ld) or less will currently work.", | |
453 | sbp->sb_blocksize, PAGE_SIZE); | |
2edbddd5 LM |
454 | return XFS_ERROR(ENOSYS); |
455 | } | |
456 | ||
1a5902c5 CH |
457 | /* |
458 | * Currently only very few inode sizes are supported. | |
459 | */ | |
460 | switch (sbp->sb_inodesize) { | |
461 | case 256: | |
462 | case 512: | |
463 | case 1024: | |
464 | case 2048: | |
465 | break; | |
466 | default: | |
98021821 | 467 | xfs_warn(mp, "inode size of %d bytes not supported", |
af34e09d | 468 | sbp->sb_inodesize); |
1a5902c5 CH |
469 | return XFS_ERROR(ENOSYS); |
470 | } | |
471 | ||
4cc929ee NS |
472 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
473 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
98021821 | 474 | xfs_warn(mp, |
af34e09d | 475 | "file system too large to be mounted on this system."); |
657a4cff | 476 | return XFS_ERROR(EFBIG); |
1da177e4 LT |
477 | } |
478 | ||
98021821 DC |
479 | if (check_inprogress && sbp->sb_inprogress) { |
480 | xfs_warn(mp, "Offline file system operation in progress!"); | |
1da177e4 LT |
481 | return XFS_ERROR(EFSCORRUPTED); |
482 | } | |
483 | ||
de20614b NS |
484 | /* |
485 | * Version 1 directory format has never worked on Linux. | |
486 | */ | |
62118709 | 487 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
98021821 | 488 | xfs_warn(mp, "file system using version 1 directory format"); |
de20614b NS |
489 | return XFS_ERROR(ENOSYS); |
490 | } | |
491 | ||
1da177e4 LT |
492 | return 0; |
493 | } | |
494 | ||
1c1c6ebc | 495 | int |
c11e2c36 | 496 | xfs_initialize_perag( |
c11e2c36 | 497 | xfs_mount_t *mp, |
1c1c6ebc DC |
498 | xfs_agnumber_t agcount, |
499 | xfs_agnumber_t *maxagi) | |
1da177e4 | 500 | { |
2d2194f6 | 501 | xfs_agnumber_t index; |
8b26c582 | 502 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
503 | xfs_perag_t *pag; |
504 | xfs_agino_t agino; | |
505 | xfs_ino_t ino; | |
506 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 507 | int error = -ENOMEM; |
1da177e4 | 508 | |
1c1c6ebc DC |
509 | /* |
510 | * Walk the current per-ag tree so we don't try to initialise AGs | |
511 | * that already exist (growfs case). Allocate and insert all the | |
512 | * AGs we don't find ready for initialisation. | |
513 | */ | |
514 | for (index = 0; index < agcount; index++) { | |
515 | pag = xfs_perag_get(mp, index); | |
516 | if (pag) { | |
517 | xfs_perag_put(pag); | |
518 | continue; | |
519 | } | |
8b26c582 DC |
520 | if (!first_initialised) |
521 | first_initialised = index; | |
fb3b504a | 522 | |
1c1c6ebc DC |
523 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
524 | if (!pag) | |
8b26c582 | 525 | goto out_unwind; |
fb3b504a CH |
526 | pag->pag_agno = index; |
527 | pag->pag_mount = mp; | |
1a427ab0 | 528 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 529 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 530 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
531 | spin_lock_init(&pag->pag_buf_lock); |
532 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 533 | |
1c1c6ebc | 534 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 535 | goto out_unwind; |
fb3b504a | 536 | |
1c1c6ebc DC |
537 | spin_lock(&mp->m_perag_lock); |
538 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
539 | BUG(); | |
540 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
541 | radix_tree_preload_end(); |
542 | error = -EEXIST; | |
543 | goto out_unwind; | |
1c1c6ebc DC |
544 | } |
545 | spin_unlock(&mp->m_perag_lock); | |
546 | radix_tree_preload_end(); | |
547 | } | |
548 | ||
fb3b504a CH |
549 | /* |
550 | * If we mount with the inode64 option, or no inode overflows | |
551 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 552 | */ |
fb3b504a CH |
553 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
554 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
555 | ||
556 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 557 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 558 | else |
1da177e4 | 559 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 560 | |
2d2194f6 CM |
561 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
562 | index = xfs_set_inode32(mp); | |
563 | else | |
564 | index = xfs_set_inode64(mp); | |
fb3b504a | 565 | |
1c1c6ebc DC |
566 | if (maxagi) |
567 | *maxagi = index; | |
568 | return 0; | |
8b26c582 DC |
569 | |
570 | out_unwind: | |
571 | kmem_free(pag); | |
572 | for (; index > first_initialised; index--) { | |
573 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
574 | kmem_free(pag); | |
575 | } | |
576 | return error; | |
1da177e4 LT |
577 | } |
578 | ||
83e782e1 CS |
579 | static void |
580 | xfs_sb_quota_from_disk(struct xfs_sb *sbp) | |
581 | { | |
01026297 CS |
582 | /* |
583 | * older mkfs doesn't initialize quota inodes to NULLFSINO. This | |
584 | * leads to in-core values having two different values for a quota | |
585 | * inode to be invalid: 0 and NULLFSINO. Change it to a single value | |
586 | * NULLFSINO. | |
587 | * | |
588 | * Note that this change affect only the in-core values. These | |
589 | * values are not written back to disk unless any quota information | |
590 | * is written to the disk. Even in that case, sb_pquotino field is | |
591 | * not written to disk unless the superblock supports pquotino. | |
592 | */ | |
593 | if (sbp->sb_uquotino == 0) | |
594 | sbp->sb_uquotino = NULLFSINO; | |
595 | if (sbp->sb_gquotino == 0) | |
596 | sbp->sb_gquotino = NULLFSINO; | |
597 | if (sbp->sb_pquotino == 0) | |
598 | sbp->sb_pquotino = NULLFSINO; | |
599 | ||
d892d586 CS |
600 | /* |
601 | * We need to do these manipilations only if we are working | |
602 | * with an older version of on-disk superblock. | |
603 | */ | |
604 | if (xfs_sb_version_has_pquotino(sbp)) | |
605 | return; | |
606 | ||
83e782e1 CS |
607 | if (sbp->sb_qflags & XFS_OQUOTA_ENFD) |
608 | sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ? | |
609 | XFS_PQUOTA_ENFD : XFS_GQUOTA_ENFD; | |
610 | if (sbp->sb_qflags & XFS_OQUOTA_CHKD) | |
611 | sbp->sb_qflags |= (sbp->sb_qflags & XFS_PQUOTA_ACCT) ? | |
612 | XFS_PQUOTA_CHKD : XFS_GQUOTA_CHKD; | |
613 | sbp->sb_qflags &= ~(XFS_OQUOTA_ENFD | XFS_OQUOTA_CHKD); | |
d892d586 CS |
614 | |
615 | if (sbp->sb_qflags & XFS_PQUOTA_ACCT) { | |
616 | /* | |
617 | * In older version of superblock, on-disk superblock only | |
618 | * has sb_gquotino, and in-core superblock has both sb_gquotino | |
619 | * and sb_pquotino. But, only one of them is supported at any | |
620 | * point of time. So, if PQUOTA is set in disk superblock, | |
621 | * copy over sb_gquotino to sb_pquotino. | |
622 | */ | |
623 | sbp->sb_pquotino = sbp->sb_gquotino; | |
624 | sbp->sb_gquotino = NULLFSINO; | |
625 | } | |
83e782e1 CS |
626 | } |
627 | ||
2bdf7cd0 CH |
628 | void |
629 | xfs_sb_from_disk( | |
98021821 | 630 | struct xfs_sb *to, |
2bdf7cd0 CH |
631 | xfs_dsb_t *from) |
632 | { | |
633 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
634 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
635 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
636 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
637 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
638 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
639 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
640 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
641 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
642 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
643 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
644 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
645 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
646 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
647 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
648 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
649 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
650 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
651 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
652 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
653 | to->sb_blocklog = from->sb_blocklog; | |
654 | to->sb_sectlog = from->sb_sectlog; | |
655 | to->sb_inodelog = from->sb_inodelog; | |
656 | to->sb_inopblog = from->sb_inopblog; | |
657 | to->sb_agblklog = from->sb_agblklog; | |
658 | to->sb_rextslog = from->sb_rextslog; | |
659 | to->sb_inprogress = from->sb_inprogress; | |
660 | to->sb_imax_pct = from->sb_imax_pct; | |
661 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
662 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
663 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
664 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
665 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
666 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
667 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
668 | to->sb_flags = from->sb_flags; | |
669 | to->sb_shared_vn = from->sb_shared_vn; | |
670 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
671 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
672 | to->sb_width = be32_to_cpu(from->sb_width); | |
673 | to->sb_dirblklog = from->sb_dirblklog; | |
674 | to->sb_logsectlog = from->sb_logsectlog; | |
675 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
676 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
677 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 678 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
04a1e6c5 DC |
679 | to->sb_features_compat = be32_to_cpu(from->sb_features_compat); |
680 | to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat); | |
681 | to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat); | |
e721f504 DC |
682 | to->sb_features_log_incompat = |
683 | be32_to_cpu(from->sb_features_log_incompat); | |
684 | to->sb_pad = 0; | |
04a1e6c5 DC |
685 | to->sb_pquotino = be64_to_cpu(from->sb_pquotino); |
686 | to->sb_lsn = be64_to_cpu(from->sb_lsn); | |
2bdf7cd0 CH |
687 | } |
688 | ||
83e782e1 CS |
689 | static inline void |
690 | xfs_sb_quota_to_disk( | |
691 | xfs_dsb_t *to, | |
692 | xfs_sb_t *from, | |
693 | __int64_t *fields) | |
694 | { | |
695 | __uint16_t qflags = from->sb_qflags; | |
696 | ||
d892d586 CS |
697 | /* |
698 | * We need to do these manipilations only if we are working | |
699 | * with an older version of on-disk superblock. | |
700 | */ | |
701 | if (xfs_sb_version_has_pquotino(from)) | |
702 | return; | |
703 | ||
83e782e1 CS |
704 | if (*fields & XFS_SB_QFLAGS) { |
705 | /* | |
706 | * The in-core version of sb_qflags do not have | |
707 | * XFS_OQUOTA_* flags, whereas the on-disk version | |
708 | * does. So, convert incore XFS_{PG}QUOTA_* flags | |
709 | * to on-disk XFS_OQUOTA_* flags. | |
710 | */ | |
711 | qflags &= ~(XFS_PQUOTA_ENFD | XFS_PQUOTA_CHKD | | |
712 | XFS_GQUOTA_ENFD | XFS_GQUOTA_CHKD); | |
713 | ||
714 | if (from->sb_qflags & | |
715 | (XFS_PQUOTA_ENFD | XFS_GQUOTA_ENFD)) | |
716 | qflags |= XFS_OQUOTA_ENFD; | |
717 | if (from->sb_qflags & | |
718 | (XFS_PQUOTA_CHKD | XFS_GQUOTA_CHKD)) | |
719 | qflags |= XFS_OQUOTA_CHKD; | |
720 | to->sb_qflags = cpu_to_be16(qflags); | |
721 | *fields &= ~XFS_SB_QFLAGS; | |
722 | } | |
d892d586 CS |
723 | |
724 | /* | |
725 | * GQUOTINO and PQUOTINO cannot be used together in versions | |
726 | * of superblock that do not have pquotino. from->sb_flags | |
727 | * tells us which quota is active and should be copied to | |
728 | * disk. | |
729 | */ | |
730 | if ((*fields & XFS_SB_GQUOTINO) && | |
731 | (from->sb_qflags & XFS_GQUOTA_ACCT)) | |
732 | to->sb_gquotino = cpu_to_be64(from->sb_gquotino); | |
733 | else if ((*fields & XFS_SB_PQUOTINO) && | |
734 | (from->sb_qflags & XFS_PQUOTA_ACCT)) | |
735 | to->sb_gquotino = cpu_to_be64(from->sb_pquotino); | |
736 | ||
737 | *fields &= ~(XFS_SB_PQUOTINO | XFS_SB_GQUOTINO); | |
83e782e1 CS |
738 | } |
739 | ||
1da177e4 | 740 | /* |
2bdf7cd0 | 741 | * Copy in core superblock to ondisk one. |
1da177e4 | 742 | * |
2bdf7cd0 | 743 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
744 | */ |
745 | void | |
2bdf7cd0 CH |
746 | xfs_sb_to_disk( |
747 | xfs_dsb_t *to, | |
748 | xfs_sb_t *from, | |
1da177e4 LT |
749 | __int64_t fields) |
750 | { | |
2bdf7cd0 CH |
751 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
752 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
753 | xfs_sb_field_t f; |
754 | int first; | |
755 | int size; | |
756 | ||
1da177e4 | 757 | ASSERT(fields); |
1da177e4 LT |
758 | if (!fields) |
759 | return; | |
760 | ||
83e782e1 | 761 | xfs_sb_quota_to_disk(to, from, &fields); |
1da177e4 LT |
762 | while (fields) { |
763 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
764 | first = xfs_sb_info[f].offset; | |
765 | size = xfs_sb_info[f + 1].offset - first; | |
766 | ||
767 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
768 | ||
769 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 770 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
771 | } else { |
772 | switch (size) { | |
773 | case 2: | |
2bdf7cd0 CH |
774 | *(__be16 *)(to_ptr + first) = |
775 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
776 | break; |
777 | case 4: | |
2bdf7cd0 CH |
778 | *(__be32 *)(to_ptr + first) = |
779 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
780 | break; |
781 | case 8: | |
2bdf7cd0 CH |
782 | *(__be64 *)(to_ptr + first) = |
783 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
784 | break; |
785 | default: | |
786 | ASSERT(0); | |
787 | } | |
788 | } | |
789 | ||
790 | fields &= ~(1LL << f); | |
791 | } | |
792 | } | |
793 | ||
04a1e6c5 | 794 | static int |
612cfbfe | 795 | xfs_sb_verify( |
34510185 DC |
796 | struct xfs_buf *bp, |
797 | bool check_version) | |
98021821 DC |
798 | { |
799 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
800 | struct xfs_sb sb; | |
98021821 DC |
801 | |
802 | xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp)); | |
803 | ||
804 | /* | |
805 | * Only check the in progress field for the primary superblock as | |
806 | * mkfs.xfs doesn't clear it from secondary superblocks. | |
807 | */ | |
34510185 DC |
808 | return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR, |
809 | check_version); | |
612cfbfe DC |
810 | } |
811 | ||
04a1e6c5 DC |
812 | /* |
813 | * If the superblock has the CRC feature bit set or the CRC field is non-null, | |
814 | * check that the CRC is valid. We check the CRC field is non-null because a | |
815 | * single bit error could clear the feature bit and unused parts of the | |
816 | * superblock are supposed to be zero. Hence a non-null crc field indicates that | |
817 | * we've potentially lost a feature bit and we should check it anyway. | |
818 | */ | |
1813dd64 | 819 | static void |
612cfbfe DC |
820 | xfs_sb_read_verify( |
821 | struct xfs_buf *bp) | |
822 | { | |
04a1e6c5 DC |
823 | struct xfs_mount *mp = bp->b_target->bt_mount; |
824 | struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp); | |
825 | int error; | |
826 | ||
827 | /* | |
828 | * open code the version check to avoid needing to convert the entire | |
829 | * superblock from disk order just to check the version number | |
830 | */ | |
831 | if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) && | |
832 | (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) == | |
833 | XFS_SB_VERSION_5) || | |
834 | dsb->sb_crc != 0)) { | |
835 | ||
836 | if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize), | |
837 | offsetof(struct xfs_sb, sb_crc))) { | |
838 | error = EFSCORRUPTED; | |
839 | goto out_error; | |
840 | } | |
841 | } | |
34510185 | 842 | error = xfs_sb_verify(bp, true); |
04a1e6c5 DC |
843 | |
844 | out_error: | |
845 | if (error) { | |
846 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
847 | xfs_buf_ioerror(bp, error); | |
848 | } | |
98021821 DC |
849 | } |
850 | ||
851 | /* | |
852 | * We may be probed for a filesystem match, so we may not want to emit | |
853 | * messages when the superblock buffer is not actually an XFS superblock. | |
854 | * If we find an XFS superblock, the run a normal, noisy mount because we are | |
855 | * really going to mount it and want to know about errors. | |
856 | */ | |
1813dd64 | 857 | static void |
98021821 DC |
858 | xfs_sb_quiet_read_verify( |
859 | struct xfs_buf *bp) | |
860 | { | |
04a1e6c5 | 861 | struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp); |
98021821 | 862 | |
98021821 | 863 | |
04a1e6c5 | 864 | if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) { |
98021821 DC |
865 | /* XFS filesystem, verify noisily! */ |
866 | xfs_sb_read_verify(bp); | |
867 | return; | |
868 | } | |
869 | /* quietly fail */ | |
aeb4f20a | 870 | xfs_buf_ioerror(bp, EWRONGFS); |
98021821 DC |
871 | } |
872 | ||
1813dd64 DC |
873 | static void |
874 | xfs_sb_write_verify( | |
04a1e6c5 | 875 | struct xfs_buf *bp) |
1813dd64 | 876 | { |
04a1e6c5 DC |
877 | struct xfs_mount *mp = bp->b_target->bt_mount; |
878 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
879 | int error; | |
880 | ||
34510185 | 881 | error = xfs_sb_verify(bp, false); |
04a1e6c5 DC |
882 | if (error) { |
883 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
884 | xfs_buf_ioerror(bp, error); | |
885 | return; | |
886 | } | |
887 | ||
888 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
889 | return; | |
890 | ||
891 | if (bip) | |
892 | XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
893 | ||
894 | xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), | |
895 | offsetof(struct xfs_sb, sb_crc)); | |
1813dd64 DC |
896 | } |
897 | ||
898 | const struct xfs_buf_ops xfs_sb_buf_ops = { | |
899 | .verify_read = xfs_sb_read_verify, | |
900 | .verify_write = xfs_sb_write_verify, | |
901 | }; | |
902 | ||
903 | static const struct xfs_buf_ops xfs_sb_quiet_buf_ops = { | |
904 | .verify_read = xfs_sb_quiet_read_verify, | |
905 | .verify_write = xfs_sb_write_verify, | |
906 | }; | |
907 | ||
1da177e4 LT |
908 | /* |
909 | * xfs_readsb | |
910 | * | |
911 | * Does the initial read of the superblock. | |
912 | */ | |
913 | int | |
764d1f89 | 914 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
915 | { |
916 | unsigned int sector_size; | |
04a1e6c5 DC |
917 | struct xfs_buf *bp; |
918 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 919 | int error; |
af34e09d | 920 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
921 | |
922 | ASSERT(mp->m_sb_bp == NULL); | |
923 | ASSERT(mp->m_ddev_targp != NULL); | |
924 | ||
925 | /* | |
926 | * Allocate a (locked) buffer to hold the superblock. | |
927 | * This will be kept around at all times to optimize | |
928 | * access to the superblock. | |
929 | */ | |
930 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
931 | |
932 | reread: | |
e70b73f8 | 933 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
98021821 | 934 | BTOBB(sector_size), 0, |
1813dd64 DC |
935 | loud ? &xfs_sb_buf_ops |
936 | : &xfs_sb_quiet_buf_ops); | |
26af6552 | 937 | if (!bp) { |
af34e09d DC |
938 | if (loud) |
939 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 940 | return EIO; |
1da177e4 | 941 | } |
eab4e633 DC |
942 | if (bp->b_error) { |
943 | error = bp->b_error; | |
944 | if (loud) | |
e721f504 | 945 | xfs_warn(mp, "SB validate failed with error %d.", error); |
eab4e633 DC |
946 | goto release_buf; |
947 | } | |
1da177e4 LT |
948 | |
949 | /* | |
950 | * Initialize the mount structure from the superblock. | |
1da177e4 | 951 | */ |
98021821 | 952 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 953 | |
83e782e1 | 954 | xfs_sb_quota_from_disk(&mp->m_sb); |
1da177e4 LT |
955 | /* |
956 | * We must be able to do sector-sized and sector-aligned IO. | |
957 | */ | |
04a1e6c5 | 958 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
959 | if (loud) |
960 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 961 | sector_size, sbp->sb_sectsize); |
1da177e4 | 962 | error = ENOSYS; |
26af6552 | 963 | goto release_buf; |
1da177e4 LT |
964 | } |
965 | ||
966 | /* | |
967 | * If device sector size is smaller than the superblock size, | |
968 | * re-read the superblock so the buffer is correctly sized. | |
969 | */ | |
04a1e6c5 | 970 | if (sector_size < sbp->sb_sectsize) { |
1da177e4 | 971 | xfs_buf_relse(bp); |
04a1e6c5 | 972 | sector_size = sbp->sb_sectsize; |
26af6552 | 973 | goto reread; |
1da177e4 LT |
974 | } |
975 | ||
5478eead LM |
976 | /* Initialize per-cpu counters */ |
977 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 978 | |
04a1e6c5 DC |
979 | /* no need to be quiet anymore, so reset the buf ops */ |
980 | bp->b_ops = &xfs_sb_buf_ops; | |
981 | ||
1da177e4 | 982 | mp->m_sb_bp = bp; |
26af6552 | 983 | xfs_buf_unlock(bp); |
1da177e4 LT |
984 | return 0; |
985 | ||
26af6552 DC |
986 | release_buf: |
987 | xfs_buf_relse(bp); | |
1da177e4 LT |
988 | return error; |
989 | } | |
990 | ||
991 | ||
992 | /* | |
993 | * xfs_mount_common | |
994 | * | |
995 | * Mount initialization code establishing various mount | |
996 | * fields from the superblock associated with the given | |
997 | * mount structure | |
998 | */ | |
ba0f32d4 | 999 | STATIC void |
1da177e4 LT |
1000 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
1001 | { | |
1da177e4 | 1002 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 1003 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
1004 | mp->m_maxagi = mp->m_sb.sb_agcount; |
1005 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
1006 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
1007 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
1008 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
1009 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
1010 | mp->m_blockmask = sbp->sb_blocksize - 1; |
1011 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
1012 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 1013 | |
60197e8d CH |
1014 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
1015 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
1016 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
1017 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
1018 | ||
1019 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
1020 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
1021 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
1022 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
1023 | ||
1024 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
1025 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
1026 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
1027 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
1028 | |
1029 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
1030 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
1031 | sbp->sb_inopblock); | |
1032 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
1033 | } | |
92821e2b DC |
1034 | |
1035 | /* | |
1036 | * xfs_initialize_perag_data | |
1037 | * | |
1038 | * Read in each per-ag structure so we can count up the number of | |
1039 | * allocated inodes, free inodes and used filesystem blocks as this | |
1040 | * information is no longer persistent in the superblock. Once we have | |
1041 | * this information, write it into the in-core superblock structure. | |
1042 | */ | |
1043 | STATIC int | |
1044 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
1045 | { | |
1046 | xfs_agnumber_t index; | |
1047 | xfs_perag_t *pag; | |
1048 | xfs_sb_t *sbp = &mp->m_sb; | |
1049 | uint64_t ifree = 0; | |
1050 | uint64_t ialloc = 0; | |
1051 | uint64_t bfree = 0; | |
1052 | uint64_t bfreelst = 0; | |
1053 | uint64_t btree = 0; | |
1054 | int error; | |
92821e2b DC |
1055 | |
1056 | for (index = 0; index < agcount; index++) { | |
1057 | /* | |
1058 | * read the agf, then the agi. This gets us | |
9da096fd | 1059 | * all the information we need and populates the |
92821e2b DC |
1060 | * per-ag structures for us. |
1061 | */ | |
1062 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
1063 | if (error) | |
1064 | return error; | |
1065 | ||
1066 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
1067 | if (error) | |
1068 | return error; | |
44b56e0a | 1069 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
1070 | ifree += pag->pagi_freecount; |
1071 | ialloc += pag->pagi_count; | |
1072 | bfree += pag->pagf_freeblks; | |
1073 | bfreelst += pag->pagf_flcount; | |
1074 | btree += pag->pagf_btreeblks; | |
44b56e0a | 1075 | xfs_perag_put(pag); |
92821e2b DC |
1076 | } |
1077 | /* | |
1078 | * Overwrite incore superblock counters with just-read data | |
1079 | */ | |
3685c2a1 | 1080 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
1081 | sbp->sb_ifree = ifree; |
1082 | sbp->sb_icount = ialloc; | |
1083 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 1084 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
1085 | |
1086 | /* Fixup the per-cpu counters as well. */ | |
1087 | xfs_icsb_reinit_counters(mp); | |
1088 | ||
1089 | return 0; | |
1090 | } | |
1091 | ||
1da177e4 | 1092 | /* |
0771fb45 | 1093 | * Update alignment values based on mount options and sb values |
1da177e4 | 1094 | */ |
0771fb45 | 1095 | STATIC int |
7884bc86 | 1096 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 1097 | { |
1da177e4 | 1098 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 1099 | |
4249023a | 1100 | if (mp->m_dalign) { |
1da177e4 LT |
1101 | /* |
1102 | * If stripe unit and stripe width are not multiples | |
1103 | * of the fs blocksize turn off alignment. | |
1104 | */ | |
1105 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
1106 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
39a45d84 JL |
1107 | xfs_warn(mp, |
1108 | "alignment check failed: sunit/swidth vs. blocksize(%d)", | |
1109 | sbp->sb_blocksize); | |
1110 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
1111 | } else { |
1112 | /* | |
1113 | * Convert the stripe unit and width to FSBs. | |
1114 | */ | |
1115 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
1116 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
53487786 | 1117 | xfs_warn(mp, |
39a45d84 JL |
1118 | "alignment check failed: sunit/swidth vs. agsize(%d)", |
1119 | sbp->sb_agblocks); | |
1120 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
1121 | } else if (mp->m_dalign) { |
1122 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
1123 | } else { | |
39a45d84 JL |
1124 | xfs_warn(mp, |
1125 | "alignment check failed: sunit(%d) less than bsize(%d)", | |
1126 | mp->m_dalign, sbp->sb_blocksize); | |
1127 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
1128 | } |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * Update superblock with new values | |
1133 | * and log changes | |
1134 | */ | |
62118709 | 1135 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
1136 | if (sbp->sb_unit != mp->m_dalign) { |
1137 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 1138 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
1139 | } |
1140 | if (sbp->sb_width != mp->m_swidth) { | |
1141 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 1142 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 | 1143 | } |
34d7f603 JL |
1144 | } else { |
1145 | xfs_warn(mp, | |
1146 | "cannot change alignment: superblock does not support data alignment"); | |
1147 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
1148 | } |
1149 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 1150 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
1151 | mp->m_dalign = sbp->sb_unit; |
1152 | mp->m_swidth = sbp->sb_width; | |
1153 | } | |
1154 | ||
0771fb45 ES |
1155 | return 0; |
1156 | } | |
1da177e4 | 1157 | |
0771fb45 ES |
1158 | /* |
1159 | * Set the maximum inode count for this filesystem | |
1160 | */ | |
1161 | STATIC void | |
1162 | xfs_set_maxicount(xfs_mount_t *mp) | |
1163 | { | |
1164 | xfs_sb_t *sbp = &(mp->m_sb); | |
1165 | __uint64_t icount; | |
1da177e4 | 1166 | |
0771fb45 ES |
1167 | if (sbp->sb_imax_pct) { |
1168 | /* | |
1169 | * Make sure the maximum inode count is a multiple | |
1170 | * of the units we allocate inodes in. | |
1da177e4 | 1171 | */ |
1da177e4 LT |
1172 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
1173 | do_div(icount, 100); | |
1174 | do_div(icount, mp->m_ialloc_blks); | |
1175 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
1176 | sbp->sb_inopblog; | |
0771fb45 | 1177 | } else { |
1da177e4 | 1178 | mp->m_maxicount = 0; |
1da177e4 | 1179 | } |
0771fb45 ES |
1180 | } |
1181 | ||
1182 | /* | |
1183 | * Set the default minimum read and write sizes unless | |
1184 | * already specified in a mount option. | |
1185 | * We use smaller I/O sizes when the file system | |
1186 | * is being used for NFS service (wsync mount option). | |
1187 | */ | |
1188 | STATIC void | |
1189 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
1190 | { | |
1191 | xfs_sb_t *sbp = &(mp->m_sb); | |
1192 | int readio_log, writeio_log; | |
1da177e4 | 1193 | |
1da177e4 LT |
1194 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
1195 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
1196 | readio_log = XFS_WSYNC_READIO_LOG; | |
1197 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
1198 | } else { | |
1199 | readio_log = XFS_READIO_LOG_LARGE; | |
1200 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
1201 | } | |
1202 | } else { | |
1203 | readio_log = mp->m_readio_log; | |
1204 | writeio_log = mp->m_writeio_log; | |
1205 | } | |
1206 | ||
1da177e4 LT |
1207 | if (sbp->sb_blocklog > readio_log) { |
1208 | mp->m_readio_log = sbp->sb_blocklog; | |
1209 | } else { | |
1210 | mp->m_readio_log = readio_log; | |
1211 | } | |
1212 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
1213 | if (sbp->sb_blocklog > writeio_log) { | |
1214 | mp->m_writeio_log = sbp->sb_blocklog; | |
1215 | } else { | |
1216 | mp->m_writeio_log = writeio_log; | |
1217 | } | |
1218 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 1219 | } |
1da177e4 | 1220 | |
055388a3 DC |
1221 | /* |
1222 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
1223 | */ | |
1224 | void | |
1225 | xfs_set_low_space_thresholds( | |
1226 | struct xfs_mount *mp) | |
1227 | { | |
1228 | int i; | |
1229 | ||
1230 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
1231 | __uint64_t space = mp->m_sb.sb_dblocks; | |
1232 | ||
1233 | do_div(space, 100); | |
1234 | mp->m_low_space[i] = space * (i + 1); | |
1235 | } | |
1236 | } | |
1237 | ||
1238 | ||
0771fb45 ES |
1239 | /* |
1240 | * Set whether we're using inode alignment. | |
1241 | */ | |
1242 | STATIC void | |
1243 | xfs_set_inoalignment(xfs_mount_t *mp) | |
1244 | { | |
62118709 | 1245 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
1246 | mp->m_sb.sb_inoalignmt >= |
1247 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
1248 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
1249 | else | |
1250 | mp->m_inoalign_mask = 0; | |
1251 | /* | |
1252 | * If we are using stripe alignment, check whether | |
1253 | * the stripe unit is a multiple of the inode alignment | |
1254 | */ | |
1255 | if (mp->m_dalign && mp->m_inoalign_mask && | |
1256 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
1257 | mp->m_sinoalign = mp->m_dalign; | |
1258 | else | |
1259 | mp->m_sinoalign = 0; | |
0771fb45 ES |
1260 | } |
1261 | ||
1262 | /* | |
1263 | * Check that the data (and log if separate) are an ok size. | |
1264 | */ | |
1265 | STATIC int | |
4249023a | 1266 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
1267 | { |
1268 | xfs_buf_t *bp; | |
1269 | xfs_daddr_t d; | |
0771fb45 | 1270 | |
1da177e4 LT |
1271 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
1272 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 1273 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 1274 | return XFS_ERROR(EFBIG); |
1da177e4 | 1275 | } |
e70b73f8 | 1276 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 1277 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 1278 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 1279 | if (!bp) { |
0b932ccc | 1280 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 1281 | return EIO; |
1da177e4 | 1282 | } |
1922c949 | 1283 | xfs_buf_relse(bp); |
1da177e4 | 1284 | |
4249023a | 1285 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
1286 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
1287 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 1288 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 1289 | return XFS_ERROR(EFBIG); |
1da177e4 | 1290 | } |
e70b73f8 | 1291 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 1292 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 1293 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 1294 | if (!bp) { |
0b932ccc | 1295 | xfs_warn(mp, "log device read failed"); |
1922c949 | 1296 | return EIO; |
0771fb45 | 1297 | } |
1922c949 | 1298 | xfs_buf_relse(bp); |
0771fb45 ES |
1299 | } |
1300 | return 0; | |
1301 | } | |
1302 | ||
7d095257 CH |
1303 | /* |
1304 | * Clear the quotaflags in memory and in the superblock. | |
1305 | */ | |
1306 | int | |
1307 | xfs_mount_reset_sbqflags( | |
1308 | struct xfs_mount *mp) | |
1309 | { | |
1310 | int error; | |
1311 | struct xfs_trans *tp; | |
1312 | ||
1313 | mp->m_qflags = 0; | |
1314 | ||
1315 | /* | |
1316 | * It is OK to look at sb_qflags here in mount path, | |
1317 | * without m_sb_lock. | |
1318 | */ | |
1319 | if (mp->m_sb.sb_qflags == 0) | |
1320 | return 0; | |
1321 | spin_lock(&mp->m_sb_lock); | |
1322 | mp->m_sb.sb_qflags = 0; | |
1323 | spin_unlock(&mp->m_sb_lock); | |
1324 | ||
1325 | /* | |
1326 | * If the fs is readonly, let the incore superblock run | |
1327 | * with quotas off but don't flush the update out to disk | |
1328 | */ | |
1329 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1330 | return 0; | |
1331 | ||
7d095257 | 1332 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
b0c10b98 JL |
1333 | error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp), |
1334 | 0, 0, XFS_DEFAULT_LOG_COUNT); | |
7d095257 CH |
1335 | if (error) { |
1336 | xfs_trans_cancel(tp, 0); | |
53487786 | 1337 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
1338 | return error; |
1339 | } | |
1340 | ||
1341 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1342 | return xfs_trans_commit(tp, 0); | |
1343 | } | |
1344 | ||
d5db0f97 ES |
1345 | __uint64_t |
1346 | xfs_default_resblks(xfs_mount_t *mp) | |
1347 | { | |
1348 | __uint64_t resblks; | |
1349 | ||
1350 | /* | |
8babd8a2 DC |
1351 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
1352 | * smaller. This is intended to cover concurrent allocation | |
1353 | * transactions when we initially hit enospc. These each require a 4 | |
1354 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
1355 | * allocation reservations. | |
d5db0f97 ES |
1356 | */ |
1357 | resblks = mp->m_sb.sb_dblocks; | |
1358 | do_div(resblks, 20); | |
8babd8a2 | 1359 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
1360 | return resblks; |
1361 | } | |
1362 | ||
0771fb45 | 1363 | /* |
0771fb45 ES |
1364 | * This function does the following on an initial mount of a file system: |
1365 | * - reads the superblock from disk and init the mount struct | |
1366 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1367 | * so we don't mount terabyte filesystems | |
1368 | * - init mount struct realtime fields | |
1369 | * - allocate inode hash table for fs | |
1370 | * - init directory manager | |
1371 | * - perform recovery and init the log manager | |
1372 | */ | |
1373 | int | |
1374 | xfs_mountfs( | |
4249023a | 1375 | xfs_mount_t *mp) |
0771fb45 ES |
1376 | { |
1377 | xfs_sb_t *sbp = &(mp->m_sb); | |
1378 | xfs_inode_t *rip; | |
0771fb45 | 1379 | __uint64_t resblks; |
7d095257 CH |
1380 | uint quotamount = 0; |
1381 | uint quotaflags = 0; | |
0771fb45 ES |
1382 | int error = 0; |
1383 | ||
0771fb45 ES |
1384 | xfs_mount_common(mp, sbp); |
1385 | ||
ee1c0908 | 1386 | /* |
e6957ea4 ES |
1387 | * Check for a mismatched features2 values. Older kernels |
1388 | * read & wrote into the wrong sb offset for sb_features2 | |
1389 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1390 | * when sb_features2 was added, which made older superblock | |
1391 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1392 | * |
e6957ea4 ES |
1393 | * For backwards compatibility, we make both slots equal. |
1394 | * | |
1395 | * If we detect a mismatched field, we OR the set bits into the | |
1396 | * existing features2 field in case it has already been modified; we | |
1397 | * don't want to lose any features. We then update the bad location | |
1398 | * with the ORed value so that older kernels will see any features2 | |
1399 | * flags, and mark the two fields as needing updates once the | |
1400 | * transaction subsystem is online. | |
ee1c0908 | 1401 | */ |
e6957ea4 | 1402 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 1403 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 1404 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 1405 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1406 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1407 | |
1408 | /* | |
1409 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1410 | * slot. | |
1411 | */ | |
7c12f296 TS |
1412 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1413 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1414 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1415 | } |
1416 | ||
1417 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1418 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1419 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1420 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1421 | |
7c12f296 TS |
1422 | /* update sb_versionnum for the clearing of the morebits */ |
1423 | if (!sbp->sb_features2) | |
7884bc86 | 1424 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1425 | } |
1426 | ||
0771fb45 ES |
1427 | /* |
1428 | * Check if sb_agblocks is aligned at stripe boundary | |
1429 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1430 | * allocator alignment is within an ag, therefore ag has | |
1431 | * to be aligned at stripe boundary. | |
1432 | */ | |
7884bc86 | 1433 | error = xfs_update_alignment(mp); |
0771fb45 | 1434 | if (error) |
f9057e3d | 1435 | goto out; |
0771fb45 ES |
1436 | |
1437 | xfs_alloc_compute_maxlevels(mp); | |
1438 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1439 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1440 | xfs_ialloc_compute_maxlevels(mp); | |
1441 | ||
1442 | xfs_set_maxicount(mp); | |
1443 | ||
27174203 CH |
1444 | error = xfs_uuid_mount(mp); |
1445 | if (error) | |
1446 | goto out; | |
1da177e4 | 1447 | |
0771fb45 ES |
1448 | /* |
1449 | * Set the minimum read and write sizes | |
1450 | */ | |
1451 | xfs_set_rw_sizes(mp); | |
1452 | ||
055388a3 DC |
1453 | /* set the low space thresholds for dynamic preallocation */ |
1454 | xfs_set_low_space_thresholds(mp); | |
1455 | ||
0771fb45 ES |
1456 | /* |
1457 | * Set the inode cluster size. | |
1458 | * This may still be overridden by the file system | |
1459 | * block size if it is larger than the chosen cluster size. | |
1460 | */ | |
1461 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1462 | ||
1463 | /* | |
1464 | * Set inode alignment fields | |
1465 | */ | |
1466 | xfs_set_inoalignment(mp); | |
1467 | ||
1468 | /* | |
1469 | * Check that the data (and log if separate) are an ok size. | |
1470 | */ | |
4249023a | 1471 | error = xfs_check_sizes(mp); |
0771fb45 | 1472 | if (error) |
f9057e3d | 1473 | goto out_remove_uuid; |
0771fb45 | 1474 | |
1da177e4 LT |
1475 | /* |
1476 | * Initialize realtime fields in the mount structure | |
1477 | */ | |
0771fb45 ES |
1478 | error = xfs_rtmount_init(mp); |
1479 | if (error) { | |
0b932ccc | 1480 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 1481 | goto out_remove_uuid; |
1da177e4 LT |
1482 | } |
1483 | ||
1da177e4 LT |
1484 | /* |
1485 | * Copies the low order bits of the timestamp and the randomly | |
1486 | * set "sequence" number out of a UUID. | |
1487 | */ | |
1488 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1489 | ||
1da177e4 LT |
1490 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1491 | ||
f6c2d1fa | 1492 | xfs_dir_mount(mp); |
1da177e4 LT |
1493 | |
1494 | /* | |
1495 | * Initialize the attribute manager's entries. | |
1496 | */ | |
1497 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1498 | ||
1499 | /* | |
1500 | * Initialize the precomputed transaction reservations values. | |
1501 | */ | |
1502 | xfs_trans_init(mp); | |
1503 | ||
1da177e4 LT |
1504 | /* |
1505 | * Allocate and initialize the per-ag data. | |
1506 | */ | |
1c1c6ebc | 1507 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 1508 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
1509 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
1510 | if (error) { | |
0b932ccc | 1511 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 1512 | goto out_remove_uuid; |
1c1c6ebc | 1513 | } |
1da177e4 | 1514 | |
f9057e3d | 1515 | if (!sbp->sb_logblocks) { |
0b932ccc | 1516 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
1517 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
1518 | error = XFS_ERROR(EFSCORRUPTED); | |
1519 | goto out_free_perag; | |
1520 | } | |
1521 | ||
1da177e4 LT |
1522 | /* |
1523 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1524 | */ | |
f9057e3d CH |
1525 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1526 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1527 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1528 | if (error) { | |
0b932ccc | 1529 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 1530 | goto out_fail_wait; |
1da177e4 LT |
1531 | } |
1532 | ||
92821e2b DC |
1533 | /* |
1534 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1535 | * not. If it was, with the first phase of recovery has completed, we | |
1536 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1537 | * but they are recovered transactionally in the second recovery phase | |
1538 | * later. | |
1539 | * | |
1540 | * Hence we can safely re-initialise incore superblock counters from | |
1541 | * the per-ag data. These may not be correct if the filesystem was not | |
1542 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1543 | * doing this. | |
1544 | * | |
1545 | * If the filesystem was cleanly unmounted, then we can trust the | |
1546 | * values in the superblock to be correct and we don't need to do | |
1547 | * anything here. | |
1548 | * | |
1549 | * If we are currently making the filesystem, the initialisation will | |
1550 | * fail as the perag data is in an undefined state. | |
1551 | */ | |
92821e2b DC |
1552 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1553 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1554 | !mp->m_sb.sb_inprogress) { | |
1555 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 1556 | if (error) |
d4f3512b | 1557 | goto out_fail_wait; |
92821e2b | 1558 | } |
f9057e3d | 1559 | |
1da177e4 LT |
1560 | /* |
1561 | * Get and sanity-check the root inode. | |
1562 | * Save the pointer to it in the mount structure. | |
1563 | */ | |
7b6259e7 | 1564 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 1565 | if (error) { |
0b932ccc | 1566 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 1567 | goto out_log_dealloc; |
1da177e4 LT |
1568 | } |
1569 | ||
1570 | ASSERT(rip != NULL); | |
1da177e4 | 1571 | |
abbede1b | 1572 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 1573 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 1574 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
1575 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1576 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1577 | mp); | |
1578 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1579 | goto out_rele_rip; |
1da177e4 LT |
1580 | } |
1581 | mp->m_rootip = rip; /* save it */ | |
1582 | ||
1583 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1584 | ||
1585 | /* | |
1586 | * Initialize realtime inode pointers in the mount structure | |
1587 | */ | |
0771fb45 ES |
1588 | error = xfs_rtmount_inodes(mp); |
1589 | if (error) { | |
1da177e4 LT |
1590 | /* |
1591 | * Free up the root inode. | |
1592 | */ | |
0b932ccc | 1593 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 1594 | goto out_rele_rip; |
1da177e4 LT |
1595 | } |
1596 | ||
1597 | /* | |
7884bc86 CH |
1598 | * If this is a read-only mount defer the superblock updates until |
1599 | * the next remount into writeable mode. Otherwise we would never | |
1600 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1601 | */ |
7884bc86 CH |
1602 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1603 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 1604 | if (error) { |
0b932ccc | 1605 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 1606 | goto out_rtunmount; |
e5720eec DC |
1607 | } |
1608 | } | |
1da177e4 LT |
1609 | |
1610 | /* | |
1611 | * Initialise the XFS quota management subsystem for this mount | |
1612 | */ | |
7d095257 CH |
1613 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1614 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1615 | if (error) | |
1616 | goto out_rtunmount; | |
1617 | } else { | |
1618 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1619 | ||
1620 | /* | |
1621 | * If a file system had quotas running earlier, but decided to | |
1622 | * mount without -o uquota/pquota/gquota options, revoke the | |
1623 | * quotachecked license. | |
1624 | */ | |
1625 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 1626 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
1627 | error = xfs_mount_reset_sbqflags(mp); |
1628 | if (error) | |
1629 | return error; | |
1630 | } | |
1631 | } | |
1da177e4 LT |
1632 | |
1633 | /* | |
1634 | * Finish recovering the file system. This part needed to be | |
1635 | * delayed until after the root and real-time bitmap inodes | |
1636 | * were consistently read in. | |
1637 | */ | |
4249023a | 1638 | error = xfs_log_mount_finish(mp); |
1da177e4 | 1639 | if (error) { |
0b932ccc | 1640 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 1641 | goto out_rtunmount; |
1da177e4 LT |
1642 | } |
1643 | ||
1644 | /* | |
1645 | * Complete the quota initialisation, post-log-replay component. | |
1646 | */ | |
7d095257 CH |
1647 | if (quotamount) { |
1648 | ASSERT(mp->m_qflags == 0); | |
1649 | mp->m_qflags = quotaflags; | |
1650 | ||
1651 | xfs_qm_mount_quotas(mp); | |
1652 | } | |
1653 | ||
84e1e99f DC |
1654 | /* |
1655 | * Now we are mounted, reserve a small amount of unused space for | |
1656 | * privileged transactions. This is needed so that transaction | |
1657 | * space required for critical operations can dip into this pool | |
1658 | * when at ENOSPC. This is needed for operations like create with | |
1659 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1660 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
1661 | * |
1662 | * This may drive us straight to ENOSPC on mount, but that implies | |
1663 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 1664 | */ |
d5db0f97 ES |
1665 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1666 | resblks = xfs_default_resblks(mp); | |
1667 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
1668 | if (error) | |
0b932ccc DC |
1669 | xfs_warn(mp, |
1670 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 1671 | } |
84e1e99f | 1672 | |
1da177e4 LT |
1673 | return 0; |
1674 | ||
b93b6e43 CH |
1675 | out_rtunmount: |
1676 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1677 | out_rele_rip: |
43355099 | 1678 | IRELE(rip); |
f9057e3d | 1679 | out_log_dealloc: |
21b699c8 | 1680 | xfs_log_unmount(mp); |
d4f3512b DC |
1681 | out_fail_wait: |
1682 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
1683 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1684 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1685 | out_free_perag: |
ff4f038c | 1686 | xfs_free_perag(mp); |
f9057e3d | 1687 | out_remove_uuid: |
27174203 | 1688 | xfs_uuid_unmount(mp); |
f9057e3d | 1689 | out: |
1da177e4 LT |
1690 | return error; |
1691 | } | |
1692 | ||
1693 | /* | |
1da177e4 LT |
1694 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1695 | * log and makes sure that incore structures are freed. | |
1696 | */ | |
41b5c2e7 CH |
1697 | void |
1698 | xfs_unmountfs( | |
1699 | struct xfs_mount *mp) | |
1da177e4 | 1700 | { |
41b5c2e7 CH |
1701 | __uint64_t resblks; |
1702 | int error; | |
1da177e4 | 1703 | |
579b62fa BF |
1704 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1705 | ||
7d095257 | 1706 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1707 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1708 | IRELE(mp->m_rootip); |
1709 | ||
641c56fb DC |
1710 | /* |
1711 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1712 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1713 | * the transaction is still sitting in a iclog. The stale inodes |
1714 | * on that buffer will have their flush locks held until the | |
1715 | * transaction hits the disk and the callbacks run. the inode | |
1716 | * flush takes the flush lock unconditionally and with nothing to | |
1717 | * push out the iclog we will never get that unlocked. hence we | |
1718 | * need to force the log first. | |
1719 | */ | |
a14a348b | 1720 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1721 | |
1722 | /* | |
211e4d43 CH |
1723 | * Flush all pending changes from the AIL. |
1724 | */ | |
1725 | xfs_ail_push_all_sync(mp->m_ail); | |
1726 | ||
1727 | /* | |
1728 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1729 | * inodes and none should be pinned or locked, but use synchronous |
1730 | * reclaim just to be sure. We can stop background inode reclaim | |
1731 | * here as well if it is still running. | |
c854363e | 1732 | */ |
7e18530b | 1733 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1734 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1735 | |
7d095257 | 1736 | xfs_qm_unmount(mp); |
a357a121 | 1737 | |
84e1e99f DC |
1738 | /* |
1739 | * Unreserve any blocks we have so that when we unmount we don't account | |
1740 | * the reserved free space as used. This is really only necessary for | |
1741 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1742 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1743 | * |
1744 | * We don't bother correcting this elsewhere for lazy superblock | |
1745 | * counting because on mount of an unclean filesystem we reconstruct the | |
1746 | * correct counter value and this is irrelevant. | |
1747 | * | |
1748 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1749 | * we only every apply deltas to the superblock and hence the incore | |
1750 | * value does not matter.... | |
1751 | */ | |
1752 | resblks = 0; | |
714082bc DC |
1753 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1754 | if (error) | |
0b932ccc | 1755 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1756 | "Freespace may not be correct on next mount."); |
1757 | ||
adab0f67 | 1758 | error = xfs_log_sbcount(mp); |
e5720eec | 1759 | if (error) |
0b932ccc | 1760 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1761 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1762 | |
21b699c8 | 1763 | xfs_log_unmount(mp); |
27174203 | 1764 | xfs_uuid_unmount(mp); |
1da177e4 | 1765 | |
1550d0b0 | 1766 | #if defined(DEBUG) |
0ce4cfd4 | 1767 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1768 | #endif |
ff4f038c | 1769 | xfs_free_perag(mp); |
1da177e4 LT |
1770 | } |
1771 | ||
92821e2b DC |
1772 | int |
1773 | xfs_fs_writable(xfs_mount_t *mp) | |
1774 | { | |
d9457dc0 | 1775 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1776 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1777 | } |
1778 | ||
1779 | /* | |
b2ce3974 AE |
1780 | * xfs_log_sbcount |
1781 | * | |
adab0f67 | 1782 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1783 | * |
1784 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1785 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1786 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1787 | */ |
1788 | int | |
adab0f67 | 1789 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1790 | { |
1791 | xfs_trans_t *tp; | |
1792 | int error; | |
1793 | ||
1794 | if (!xfs_fs_writable(mp)) | |
1795 | return 0; | |
1796 | ||
d4d90b57 | 1797 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1798 | |
1799 | /* | |
1800 | * we don't need to do this if we are updating the superblock | |
1801 | * counters on every modification. | |
1802 | */ | |
1803 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1804 | return 0; | |
1805 | ||
b2ce3974 | 1806 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
e457274b JL |
1807 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
1808 | XFS_DEFAULT_LOG_COUNT); | |
92821e2b DC |
1809 | if (error) { |
1810 | xfs_trans_cancel(tp, 0); | |
1811 | return error; | |
1812 | } | |
1813 | ||
1814 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1815 | xfs_trans_set_sync(tp); |
e5720eec DC |
1816 | error = xfs_trans_commit(tp, 0); |
1817 | return error; | |
92821e2b DC |
1818 | } |
1819 | ||
1da177e4 LT |
1820 | /* |
1821 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1822 | * in-core superblock into the superblock buffer to be logged. | |
1823 | * It does not provide the higher level of locking that is | |
1824 | * needed to protect the in-core superblock from concurrent | |
1825 | * access. | |
1826 | */ | |
1827 | void | |
1828 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1829 | { | |
1830 | xfs_buf_t *bp; | |
1831 | int first; | |
1832 | int last; | |
1833 | xfs_mount_t *mp; | |
1da177e4 LT |
1834 | xfs_sb_field_t f; |
1835 | ||
1836 | ASSERT(fields); | |
1837 | if (!fields) | |
1838 | return; | |
1839 | mp = tp->t_mountp; | |
1840 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1841 | first = sizeof(xfs_sb_t); |
1842 | last = 0; | |
1843 | ||
1844 | /* translate/copy */ | |
1845 | ||
2bdf7cd0 | 1846 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1847 | |
1848 | /* find modified range */ | |
587aa0fe DC |
1849 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); |
1850 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1851 | last = xfs_sb_info[f + 1].offset - 1; | |
1da177e4 LT |
1852 | |
1853 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1854 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1855 | first = xfs_sb_info[f].offset; | |
1856 | ||
04a1e6c5 | 1857 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF); |
1da177e4 LT |
1858 | xfs_trans_log_buf(tp, bp, first, last); |
1859 | } | |
d210a28c | 1860 | |
d210a28c | 1861 | |
1da177e4 LT |
1862 | /* |
1863 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1864 | * a delta to a specified field in the in-core superblock. Simply | |
1865 | * switch on the field indicated and apply the delta to that field. | |
1866 | * Fields are not allowed to dip below zero, so if the delta would | |
1867 | * do this do not apply it and return EINVAL. | |
1868 | * | |
3685c2a1 | 1869 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1870 | */ |
d96f8f89 | 1871 | STATIC int |
20f4ebf2 DC |
1872 | xfs_mod_incore_sb_unlocked( |
1873 | xfs_mount_t *mp, | |
1874 | xfs_sb_field_t field, | |
1875 | int64_t delta, | |
1876 | int rsvd) | |
1da177e4 LT |
1877 | { |
1878 | int scounter; /* short counter for 32 bit fields */ | |
1879 | long long lcounter; /* long counter for 64 bit fields */ | |
1880 | long long res_used, rem; | |
1881 | ||
1882 | /* | |
1883 | * With the in-core superblock spin lock held, switch | |
1884 | * on the indicated field. Apply the delta to the | |
1885 | * proper field. If the fields value would dip below | |
1886 | * 0, then do not apply the delta and return EINVAL. | |
1887 | */ | |
1888 | switch (field) { | |
1889 | case XFS_SBS_ICOUNT: | |
1890 | lcounter = (long long)mp->m_sb.sb_icount; | |
1891 | lcounter += delta; | |
1892 | if (lcounter < 0) { | |
1893 | ASSERT(0); | |
014c2544 | 1894 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1895 | } |
1896 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1897 | return 0; |
1da177e4 LT |
1898 | case XFS_SBS_IFREE: |
1899 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1900 | lcounter += delta; | |
1901 | if (lcounter < 0) { | |
1902 | ASSERT(0); | |
014c2544 | 1903 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1904 | } |
1905 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1906 | return 0; |
1da177e4 | 1907 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1908 | lcounter = (long long) |
1909 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1910 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1911 | ||
1912 | if (delta > 0) { /* Putting blocks back */ | |
1913 | if (res_used > delta) { | |
1914 | mp->m_resblks_avail += delta; | |
1915 | } else { | |
1916 | rem = delta - res_used; | |
1917 | mp->m_resblks_avail = mp->m_resblks; | |
1918 | lcounter += rem; | |
1919 | } | |
1920 | } else { /* Taking blocks away */ | |
1da177e4 | 1921 | lcounter += delta; |
8babd8a2 DC |
1922 | if (lcounter >= 0) { |
1923 | mp->m_sb.sb_fdblocks = lcounter + | |
1924 | XFS_ALLOC_SET_ASIDE(mp); | |
1925 | return 0; | |
1926 | } | |
1da177e4 | 1927 | |
8babd8a2 DC |
1928 | /* |
1929 | * We are out of blocks, use any available reserved | |
1930 | * blocks if were allowed to. | |
1931 | */ | |
1932 | if (!rsvd) | |
1933 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1934 | |
8babd8a2 DC |
1935 | lcounter = (long long)mp->m_resblks_avail + delta; |
1936 | if (lcounter >= 0) { | |
1937 | mp->m_resblks_avail = lcounter; | |
1938 | return 0; | |
1da177e4 | 1939 | } |
8babd8a2 DC |
1940 | printk_once(KERN_WARNING |
1941 | "Filesystem \"%s\": reserve blocks depleted! " | |
1942 | "Consider increasing reserve pool size.", | |
1943 | mp->m_fsname); | |
1944 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1945 | } |
1946 | ||
4be536de | 1947 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1948 | return 0; |
1da177e4 LT |
1949 | case XFS_SBS_FREXTENTS: |
1950 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1951 | lcounter += delta; | |
1952 | if (lcounter < 0) { | |
014c2544 | 1953 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1954 | } |
1955 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1956 | return 0; |
1da177e4 LT |
1957 | case XFS_SBS_DBLOCKS: |
1958 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1959 | lcounter += delta; | |
1960 | if (lcounter < 0) { | |
1961 | ASSERT(0); | |
014c2544 | 1962 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1963 | } |
1964 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1965 | return 0; |
1da177e4 LT |
1966 | case XFS_SBS_AGCOUNT: |
1967 | scounter = mp->m_sb.sb_agcount; | |
1968 | scounter += delta; | |
1969 | if (scounter < 0) { | |
1970 | ASSERT(0); | |
014c2544 | 1971 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1972 | } |
1973 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1974 | return 0; |
1da177e4 LT |
1975 | case XFS_SBS_IMAX_PCT: |
1976 | scounter = mp->m_sb.sb_imax_pct; | |
1977 | scounter += delta; | |
1978 | if (scounter < 0) { | |
1979 | ASSERT(0); | |
014c2544 | 1980 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1981 | } |
1982 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1983 | return 0; |
1da177e4 LT |
1984 | case XFS_SBS_REXTSIZE: |
1985 | scounter = mp->m_sb.sb_rextsize; | |
1986 | scounter += delta; | |
1987 | if (scounter < 0) { | |
1988 | ASSERT(0); | |
014c2544 | 1989 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1990 | } |
1991 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1992 | return 0; |
1da177e4 LT |
1993 | case XFS_SBS_RBMBLOCKS: |
1994 | scounter = mp->m_sb.sb_rbmblocks; | |
1995 | scounter += delta; | |
1996 | if (scounter < 0) { | |
1997 | ASSERT(0); | |
014c2544 | 1998 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1999 | } |
2000 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 2001 | return 0; |
1da177e4 LT |
2002 | case XFS_SBS_RBLOCKS: |
2003 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
2004 | lcounter += delta; | |
2005 | if (lcounter < 0) { | |
2006 | ASSERT(0); | |
014c2544 | 2007 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
2008 | } |
2009 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 2010 | return 0; |
1da177e4 LT |
2011 | case XFS_SBS_REXTENTS: |
2012 | lcounter = (long long)mp->m_sb.sb_rextents; | |
2013 | lcounter += delta; | |
2014 | if (lcounter < 0) { | |
2015 | ASSERT(0); | |
014c2544 | 2016 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
2017 | } |
2018 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 2019 | return 0; |
1da177e4 LT |
2020 | case XFS_SBS_REXTSLOG: |
2021 | scounter = mp->m_sb.sb_rextslog; | |
2022 | scounter += delta; | |
2023 | if (scounter < 0) { | |
2024 | ASSERT(0); | |
014c2544 | 2025 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
2026 | } |
2027 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 2028 | return 0; |
1da177e4 LT |
2029 | default: |
2030 | ASSERT(0); | |
014c2544 | 2031 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
2032 | } |
2033 | } | |
2034 | ||
2035 | /* | |
2036 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
2037 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 2038 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
2039 | * routine to do the work. |
2040 | */ | |
2041 | int | |
20f4ebf2 | 2042 | xfs_mod_incore_sb( |
96540c78 CH |
2043 | struct xfs_mount *mp, |
2044 | xfs_sb_field_t field, | |
2045 | int64_t delta, | |
2046 | int rsvd) | |
1da177e4 | 2047 | { |
96540c78 | 2048 | int status; |
1da177e4 | 2049 | |
8d280b98 | 2050 | #ifdef HAVE_PERCPU_SB |
96540c78 | 2051 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 2052 | #endif |
96540c78 CH |
2053 | spin_lock(&mp->m_sb_lock); |
2054 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
2055 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 2056 | |
014c2544 | 2057 | return status; |
1da177e4 LT |
2058 | } |
2059 | ||
2060 | /* | |
1b040712 | 2061 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 2062 | * |
1b040712 CH |
2063 | * The fields and changes to those fields are specified in the array of |
2064 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
2065 | * will be applied or none of them will. If any modified field dips below 0, | |
2066 | * then all modifications will be backed out and EINVAL will be returned. | |
2067 | * | |
2068 | * Note that this function may not be used for the superblock values that | |
2069 | * are tracked with the in-memory per-cpu counters - a direct call to | |
2070 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
2071 | */ |
2072 | int | |
1b040712 CH |
2073 | xfs_mod_incore_sb_batch( |
2074 | struct xfs_mount *mp, | |
2075 | xfs_mod_sb_t *msb, | |
2076 | uint nmsb, | |
2077 | int rsvd) | |
1da177e4 | 2078 | { |
45c51b99 | 2079 | xfs_mod_sb_t *msbp; |
1b040712 | 2080 | int error = 0; |
1da177e4 LT |
2081 | |
2082 | /* | |
1b040712 CH |
2083 | * Loop through the array of mod structures and apply each individually. |
2084 | * If any fail, then back out all those which have already been applied. | |
2085 | * Do all of this within the scope of the m_sb_lock so that all of the | |
2086 | * changes will be atomic. | |
1da177e4 | 2087 | */ |
3685c2a1 | 2088 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 2089 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
2090 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
2091 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 2092 | |
1b040712 CH |
2093 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
2094 | msbp->msb_delta, rsvd); | |
2095 | if (error) | |
2096 | goto unwind; | |
1da177e4 | 2097 | } |
1b040712 CH |
2098 | spin_unlock(&mp->m_sb_lock); |
2099 | return 0; | |
1da177e4 | 2100 | |
1b040712 CH |
2101 | unwind: |
2102 | while (--msbp >= msb) { | |
2103 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
2104 | -msbp->msb_delta, rsvd); | |
2105 | ASSERT(error == 0); | |
1da177e4 | 2106 | } |
3685c2a1 | 2107 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 2108 | return error; |
1da177e4 LT |
2109 | } |
2110 | ||
2111 | /* | |
2112 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
2113 | * The buffer is returned locked and read in from disk. | |
2114 | * The buffer should be released with a call to xfs_brelse(). | |
2115 | * | |
2116 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
2117 | * the superblock buffer if it can be locked without sleeping. | |
2118 | * If it can't then we'll return NULL. | |
2119 | */ | |
0c842ad4 | 2120 | struct xfs_buf * |
1da177e4 | 2121 | xfs_getsb( |
0c842ad4 CH |
2122 | struct xfs_mount *mp, |
2123 | int flags) | |
1da177e4 | 2124 | { |
0c842ad4 | 2125 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 2126 | |
0c842ad4 CH |
2127 | if (!xfs_buf_trylock(bp)) { |
2128 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 2129 | return NULL; |
0c842ad4 | 2130 | xfs_buf_lock(bp); |
1da177e4 | 2131 | } |
0c842ad4 | 2132 | |
72790aa1 | 2133 | xfs_buf_hold(bp); |
1da177e4 | 2134 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 2135 | return bp; |
1da177e4 LT |
2136 | } |
2137 | ||
2138 | /* | |
2139 | * Used to free the superblock along various error paths. | |
2140 | */ | |
2141 | void | |
2142 | xfs_freesb( | |
26af6552 | 2143 | struct xfs_mount *mp) |
1da177e4 | 2144 | { |
26af6552 | 2145 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 2146 | |
26af6552 | 2147 | xfs_buf_lock(bp); |
1da177e4 | 2148 | mp->m_sb_bp = NULL; |
26af6552 | 2149 | xfs_buf_relse(bp); |
1da177e4 LT |
2150 | } |
2151 | ||
1da177e4 LT |
2152 | /* |
2153 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
2154 | * be altered by the mount options, as well as any potential sb_features2 |
2155 | * fixup. Only the first superblock is updated. | |
1da177e4 | 2156 | */ |
7884bc86 | 2157 | int |
ee1c0908 | 2158 | xfs_mount_log_sb( |
1da177e4 LT |
2159 | xfs_mount_t *mp, |
2160 | __int64_t fields) | |
2161 | { | |
2162 | xfs_trans_t *tp; | |
e5720eec | 2163 | int error; |
1da177e4 | 2164 | |
ee1c0908 | 2165 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
2166 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
2167 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
2168 | |
2169 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
5166ab06 JL |
2170 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
2171 | XFS_DEFAULT_LOG_COUNT); | |
e5720eec | 2172 | if (error) { |
1da177e4 | 2173 | xfs_trans_cancel(tp, 0); |
e5720eec | 2174 | return error; |
1da177e4 LT |
2175 | } |
2176 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
2177 | error = xfs_trans_commit(tp, 0); |
2178 | return error; | |
1da177e4 | 2179 | } |
8d280b98 | 2180 | |
dda35b8f CH |
2181 | /* |
2182 | * If the underlying (data/log/rt) device is readonly, there are some | |
2183 | * operations that cannot proceed. | |
2184 | */ | |
2185 | int | |
2186 | xfs_dev_is_read_only( | |
2187 | struct xfs_mount *mp, | |
2188 | char *message) | |
2189 | { | |
2190 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
2191 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
2192 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
2193 | xfs_notice(mp, "%s required on read-only device.", message); |
2194 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
2195 | return EROFS; |
2196 | } | |
2197 | return 0; | |
2198 | } | |
8d280b98 DC |
2199 | |
2200 | #ifdef HAVE_PERCPU_SB | |
2201 | /* | |
2202 | * Per-cpu incore superblock counters | |
2203 | * | |
2204 | * Simple concept, difficult implementation | |
2205 | * | |
2206 | * Basically, replace the incore superblock counters with a distributed per cpu | |
2207 | * counter for contended fields (e.g. free block count). | |
2208 | * | |
2209 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
2210 | * hence needs to be accurately read when we are running low on space. Hence | |
2211 | * there is a method to enable and disable the per-cpu counters based on how | |
2212 | * much "stuff" is available in them. | |
2213 | * | |
2214 | * Basically, a counter is enabled if there is enough free resource to justify | |
2215 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2216 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2217 | * re-distribute the available resources. | |
2218 | * | |
2219 | * If, once we redistributed the available resources, we still get a failure, | |
2220 | * we disable the per-cpu counter and go through the slow path. | |
2221 | * | |
2222 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2223 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2224 | * the global superblock. We do this after disabling the counter to prevent |
2225 | * more threads from queueing up on the counter. | |
2226 | * | |
2227 | * Essentially, this means that we still need a lock in the fast path to enable | |
2228 | * synchronisation between the global counters and the per-cpu counters. This | |
2229 | * is not a problem because the lock will be local to a CPU almost all the time | |
2230 | * and have little contention except when we get to ENOSPC conditions. | |
2231 | * | |
2232 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2233 | * path while we do things like enabling and disabling counters and | |
2234 | * synchronising the counters. | |
2235 | * | |
2236 | * Locking rules: | |
2237 | * | |
3685c2a1 | 2238 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2239 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2240 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2241 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2242 | * 5. modifying global counters requires holding m_sb_lock |
2243 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2244 | * and _none_ of the per-cpu locks. |
2245 | * | |
2246 | * Disabled counters are only ever re-enabled by a balance operation | |
2247 | * that results in more free resources per CPU than a given threshold. | |
2248 | * To ensure counters don't remain disabled, they are rebalanced when | |
2249 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2250 | * is present to prevent thrashing). | |
e8234a68 DC |
2251 | */ |
2252 | ||
5a67e4c5 | 2253 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2254 | /* |
2255 | * hot-plug CPU notifier support. | |
8d280b98 | 2256 | * |
5a67e4c5 CS |
2257 | * We need a notifier per filesystem as we need to be able to identify |
2258 | * the filesystem to balance the counters out. This is achieved by | |
2259 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2260 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2261 | */ |
e8234a68 DC |
2262 | STATIC int |
2263 | xfs_icsb_cpu_notify( | |
2264 | struct notifier_block *nfb, | |
2265 | unsigned long action, | |
2266 | void *hcpu) | |
2267 | { | |
2268 | xfs_icsb_cnts_t *cntp; | |
2269 | xfs_mount_t *mp; | |
e8234a68 DC |
2270 | |
2271 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2272 | cntp = (xfs_icsb_cnts_t *) | |
2273 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2274 | switch (action) { | |
2275 | case CPU_UP_PREPARE: | |
8bb78442 | 2276 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2277 | /* Easy Case - initialize the area and locks, and |
2278 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2279 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2280 | break; |
2281 | case CPU_ONLINE: | |
8bb78442 | 2282 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2283 | xfs_icsb_lock(mp); |
45af6c6d CH |
2284 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2285 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2286 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2287 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2288 | break; |
2289 | case CPU_DEAD: | |
8bb78442 | 2290 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2291 | /* Disable all the counters, then fold the dead cpu's |
2292 | * count into the total on the global superblock and | |
2293 | * re-enable the counters. */ | |
03135cf7 | 2294 | xfs_icsb_lock(mp); |
3685c2a1 | 2295 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2296 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2297 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2298 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2299 | ||
2300 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2301 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2302 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2303 | ||
01e1b69c | 2304 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2305 | |
45af6c6d CH |
2306 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2307 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2308 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2309 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2310 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2311 | break; |
2312 | } | |
2313 | ||
2314 | return NOTIFY_OK; | |
2315 | } | |
5a67e4c5 | 2316 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2317 | |
8d280b98 DC |
2318 | int |
2319 | xfs_icsb_init_counters( | |
2320 | xfs_mount_t *mp) | |
2321 | { | |
2322 | xfs_icsb_cnts_t *cntp; | |
2323 | int i; | |
2324 | ||
2325 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2326 | if (mp->m_sb_cnts == NULL) | |
2327 | return -ENOMEM; | |
2328 | ||
5a67e4c5 | 2329 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2330 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2331 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2332 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2333 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2334 | |
8d280b98 DC |
2335 | for_each_online_cpu(i) { |
2336 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2337 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2338 | } |
20b64285 DC |
2339 | |
2340 | mutex_init(&mp->m_icsb_mutex); | |
2341 | ||
8d280b98 DC |
2342 | /* |
2343 | * start with all counters disabled so that the | |
2344 | * initial balance kicks us off correctly | |
2345 | */ | |
2346 | mp->m_icsb_counters = -1; | |
2347 | return 0; | |
2348 | } | |
2349 | ||
5478eead LM |
2350 | void |
2351 | xfs_icsb_reinit_counters( | |
2352 | xfs_mount_t *mp) | |
2353 | { | |
2354 | xfs_icsb_lock(mp); | |
2355 | /* | |
2356 | * start with all counters disabled so that the | |
2357 | * initial balance kicks us off correctly | |
2358 | */ | |
2359 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2360 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2361 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2362 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2363 | xfs_icsb_unlock(mp); |
2364 | } | |
2365 | ||
c962fb79 | 2366 | void |
8d280b98 DC |
2367 | xfs_icsb_destroy_counters( |
2368 | xfs_mount_t *mp) | |
2369 | { | |
e8234a68 | 2370 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2371 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2372 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2373 | } |
03135cf7 | 2374 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2375 | } |
2376 | ||
b8f82a4a | 2377 | STATIC void |
01e1b69c DC |
2378 | xfs_icsb_lock_cntr( |
2379 | xfs_icsb_cnts_t *icsbp) | |
2380 | { | |
2381 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2382 | ndelay(1000); | |
2383 | } | |
2384 | } | |
2385 | ||
b8f82a4a | 2386 | STATIC void |
01e1b69c DC |
2387 | xfs_icsb_unlock_cntr( |
2388 | xfs_icsb_cnts_t *icsbp) | |
2389 | { | |
2390 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2391 | } | |
2392 | ||
8d280b98 | 2393 | |
b8f82a4a | 2394 | STATIC void |
8d280b98 DC |
2395 | xfs_icsb_lock_all_counters( |
2396 | xfs_mount_t *mp) | |
2397 | { | |
2398 | xfs_icsb_cnts_t *cntp; | |
2399 | int i; | |
2400 | ||
2401 | for_each_online_cpu(i) { | |
2402 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2403 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2404 | } |
2405 | } | |
2406 | ||
b8f82a4a | 2407 | STATIC void |
8d280b98 DC |
2408 | xfs_icsb_unlock_all_counters( |
2409 | xfs_mount_t *mp) | |
2410 | { | |
2411 | xfs_icsb_cnts_t *cntp; | |
2412 | int i; | |
2413 | ||
2414 | for_each_online_cpu(i) { | |
2415 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2416 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2417 | } |
2418 | } | |
2419 | ||
2420 | STATIC void | |
2421 | xfs_icsb_count( | |
2422 | xfs_mount_t *mp, | |
2423 | xfs_icsb_cnts_t *cnt, | |
2424 | int flags) | |
2425 | { | |
2426 | xfs_icsb_cnts_t *cntp; | |
2427 | int i; | |
2428 | ||
2429 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2430 | ||
2431 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2432 | xfs_icsb_lock_all_counters(mp); | |
2433 | ||
2434 | for_each_online_cpu(i) { | |
2435 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2436 | cnt->icsb_icount += cntp->icsb_icount; | |
2437 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2438 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2439 | } | |
2440 | ||
2441 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2442 | xfs_icsb_unlock_all_counters(mp); | |
2443 | } | |
2444 | ||
2445 | STATIC int | |
2446 | xfs_icsb_counter_disabled( | |
2447 | xfs_mount_t *mp, | |
2448 | xfs_sb_field_t field) | |
2449 | { | |
2450 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2451 | return test_bit(field, &mp->m_icsb_counters); | |
2452 | } | |
2453 | ||
36fbe6e6 | 2454 | STATIC void |
8d280b98 DC |
2455 | xfs_icsb_disable_counter( |
2456 | xfs_mount_t *mp, | |
2457 | xfs_sb_field_t field) | |
2458 | { | |
2459 | xfs_icsb_cnts_t cnt; | |
2460 | ||
2461 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2462 | ||
20b64285 DC |
2463 | /* |
2464 | * If we are already disabled, then there is nothing to do | |
2465 | * here. We check before locking all the counters to avoid | |
2466 | * the expensive lock operation when being called in the | |
2467 | * slow path and the counter is already disabled. This is | |
2468 | * safe because the only time we set or clear this state is under | |
2469 | * the m_icsb_mutex. | |
2470 | */ | |
2471 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2472 | return; |
20b64285 | 2473 | |
8d280b98 DC |
2474 | xfs_icsb_lock_all_counters(mp); |
2475 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2476 | /* drain back to superblock */ | |
2477 | ||
ce46193b | 2478 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2479 | switch(field) { |
2480 | case XFS_SBS_ICOUNT: | |
2481 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2482 | break; | |
2483 | case XFS_SBS_IFREE: | |
2484 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2485 | break; | |
2486 | case XFS_SBS_FDBLOCKS: | |
2487 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2488 | break; | |
2489 | default: | |
2490 | BUG(); | |
2491 | } | |
2492 | } | |
2493 | ||
2494 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2495 | } |
2496 | ||
2497 | STATIC void | |
2498 | xfs_icsb_enable_counter( | |
2499 | xfs_mount_t *mp, | |
2500 | xfs_sb_field_t field, | |
2501 | uint64_t count, | |
2502 | uint64_t resid) | |
2503 | { | |
2504 | xfs_icsb_cnts_t *cntp; | |
2505 | int i; | |
2506 | ||
2507 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2508 | ||
2509 | xfs_icsb_lock_all_counters(mp); | |
2510 | for_each_online_cpu(i) { | |
2511 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2512 | switch (field) { | |
2513 | case XFS_SBS_ICOUNT: | |
2514 | cntp->icsb_icount = count + resid; | |
2515 | break; | |
2516 | case XFS_SBS_IFREE: | |
2517 | cntp->icsb_ifree = count + resid; | |
2518 | break; | |
2519 | case XFS_SBS_FDBLOCKS: | |
2520 | cntp->icsb_fdblocks = count + resid; | |
2521 | break; | |
2522 | default: | |
2523 | BUG(); | |
2524 | break; | |
2525 | } | |
2526 | resid = 0; | |
2527 | } | |
2528 | clear_bit(field, &mp->m_icsb_counters); | |
2529 | xfs_icsb_unlock_all_counters(mp); | |
2530 | } | |
2531 | ||
dbcabad1 | 2532 | void |
d4d90b57 | 2533 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2534 | xfs_mount_t *mp, |
2535 | int flags) | |
2536 | { | |
2537 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2538 | |
8d280b98 DC |
2539 | xfs_icsb_count(mp, &cnt, flags); |
2540 | ||
8d280b98 DC |
2541 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2542 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2543 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2544 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2545 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2546 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2547 | } |
2548 | ||
2549 | /* | |
2550 | * Accurate update of per-cpu counters to incore superblock | |
2551 | */ | |
d4d90b57 | 2552 | void |
8d280b98 | 2553 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2554 | xfs_mount_t *mp, |
2555 | int flags) | |
8d280b98 | 2556 | { |
d4d90b57 CH |
2557 | spin_lock(&mp->m_sb_lock); |
2558 | xfs_icsb_sync_counters_locked(mp, flags); | |
2559 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2560 | } |
2561 | ||
2562 | /* | |
2563 | * Balance and enable/disable counters as necessary. | |
2564 | * | |
20b64285 DC |
2565 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2566 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2567 | * free blocks is something far enough zero that we aren't going thrash when we | |
2568 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2569 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2570 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2571 | * reenabled. | |
2572 | * | |
2573 | * Note that we can be called when counters are already disabled. | |
2574 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2575 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2576 | */ |
20b64285 DC |
2577 | |
2578 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2579 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2580 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2581 | STATIC void |
45af6c6d | 2582 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2583 | xfs_mount_t *mp, |
2584 | xfs_sb_field_t field, | |
20b64285 | 2585 | int min_per_cpu) |
8d280b98 | 2586 | { |
6fdf8ccc | 2587 | uint64_t count, resid; |
8d280b98 | 2588 | int weight = num_online_cpus(); |
20b64285 | 2589 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2590 | |
8d280b98 DC |
2591 | /* disable counter and sync counter */ |
2592 | xfs_icsb_disable_counter(mp, field); | |
2593 | ||
2594 | /* update counters - first CPU gets residual*/ | |
2595 | switch (field) { | |
2596 | case XFS_SBS_ICOUNT: | |
2597 | count = mp->m_sb.sb_icount; | |
2598 | resid = do_div(count, weight); | |
20b64285 | 2599 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2600 | return; |
8d280b98 DC |
2601 | break; |
2602 | case XFS_SBS_IFREE: | |
2603 | count = mp->m_sb.sb_ifree; | |
2604 | resid = do_div(count, weight); | |
20b64285 | 2605 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2606 | return; |
8d280b98 DC |
2607 | break; |
2608 | case XFS_SBS_FDBLOCKS: | |
2609 | count = mp->m_sb.sb_fdblocks; | |
2610 | resid = do_div(count, weight); | |
20b64285 | 2611 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2612 | return; |
8d280b98 DC |
2613 | break; |
2614 | default: | |
2615 | BUG(); | |
6fdf8ccc | 2616 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2617 | break; |
2618 | } | |
2619 | ||
2620 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2621 | } |
2622 | ||
2623 | STATIC void | |
2624 | xfs_icsb_balance_counter( | |
2625 | xfs_mount_t *mp, | |
2626 | xfs_sb_field_t fields, | |
2627 | int min_per_cpu) | |
2628 | { | |
2629 | spin_lock(&mp->m_sb_lock); | |
2630 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2631 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2632 | } |
2633 | ||
1b040712 | 2634 | int |
20b64285 | 2635 | xfs_icsb_modify_counters( |
8d280b98 DC |
2636 | xfs_mount_t *mp, |
2637 | xfs_sb_field_t field, | |
20f4ebf2 | 2638 | int64_t delta, |
20b64285 | 2639 | int rsvd) |
8d280b98 DC |
2640 | { |
2641 | xfs_icsb_cnts_t *icsbp; | |
2642 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2643 | int ret = 0; |
8d280b98 | 2644 | |
20b64285 | 2645 | might_sleep(); |
8d280b98 | 2646 | again: |
7a9e02d6 CL |
2647 | preempt_disable(); |
2648 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2649 | |
2650 | /* | |
2651 | * if the counter is disabled, go to slow path | |
2652 | */ | |
8d280b98 DC |
2653 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2654 | goto slow_path; | |
20b64285 DC |
2655 | xfs_icsb_lock_cntr(icsbp); |
2656 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2657 | xfs_icsb_unlock_cntr(icsbp); | |
2658 | goto slow_path; | |
2659 | } | |
8d280b98 DC |
2660 | |
2661 | switch (field) { | |
2662 | case XFS_SBS_ICOUNT: | |
2663 | lcounter = icsbp->icsb_icount; | |
2664 | lcounter += delta; | |
2665 | if (unlikely(lcounter < 0)) | |
20b64285 | 2666 | goto balance_counter; |
8d280b98 DC |
2667 | icsbp->icsb_icount = lcounter; |
2668 | break; | |
2669 | ||
2670 | case XFS_SBS_IFREE: | |
2671 | lcounter = icsbp->icsb_ifree; | |
2672 | lcounter += delta; | |
2673 | if (unlikely(lcounter < 0)) | |
20b64285 | 2674 | goto balance_counter; |
8d280b98 DC |
2675 | icsbp->icsb_ifree = lcounter; |
2676 | break; | |
2677 | ||
2678 | case XFS_SBS_FDBLOCKS: | |
2679 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2680 | ||
4be536de | 2681 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2682 | lcounter += delta; |
2683 | if (unlikely(lcounter < 0)) | |
20b64285 | 2684 | goto balance_counter; |
4be536de | 2685 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2686 | break; |
2687 | default: | |
2688 | BUG(); | |
2689 | break; | |
2690 | } | |
01e1b69c | 2691 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2692 | preempt_enable(); |
8d280b98 DC |
2693 | return 0; |
2694 | ||
8d280b98 | 2695 | slow_path: |
7a9e02d6 | 2696 | preempt_enable(); |
8d280b98 | 2697 | |
20b64285 DC |
2698 | /* |
2699 | * serialise with a mutex so we don't burn lots of cpu on | |
2700 | * the superblock lock. We still need to hold the superblock | |
2701 | * lock, however, when we modify the global structures. | |
2702 | */ | |
03135cf7 | 2703 | xfs_icsb_lock(mp); |
20b64285 DC |
2704 | |
2705 | /* | |
2706 | * Now running atomically. | |
2707 | * | |
2708 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2709 | * Drop the lock and try again in the fast path.... | |
2710 | */ | |
2711 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2712 | xfs_icsb_unlock(mp); |
8d280b98 | 2713 | goto again; |
8d280b98 DC |
2714 | } |
2715 | ||
20b64285 DC |
2716 | /* |
2717 | * The counter is currently disabled. Because we are | |
2718 | * running atomically here, we know a rebalance cannot | |
2719 | * be in progress. Hence we can go straight to operating | |
2720 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2721 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2722 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2723 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2724 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2725 | * directly on the global counters. | |
2726 | */ | |
3685c2a1 | 2727 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2728 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2729 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2730 | |
20b64285 DC |
2731 | /* |
2732 | * Now that we've modified the global superblock, we | |
2733 | * may be able to re-enable the distributed counters | |
2734 | * (e.g. lots of space just got freed). After that | |
2735 | * we are done. | |
2736 | */ | |
2737 | if (ret != ENOSPC) | |
45af6c6d | 2738 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2739 | xfs_icsb_unlock(mp); |
8d280b98 | 2740 | return ret; |
8d280b98 | 2741 | |
20b64285 DC |
2742 | balance_counter: |
2743 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2744 | preempt_enable(); |
8d280b98 | 2745 | |
20b64285 DC |
2746 | /* |
2747 | * We may have multiple threads here if multiple per-cpu | |
2748 | * counters run dry at the same time. This will mean we can | |
2749 | * do more balances than strictly necessary but it is not | |
2750 | * the common slowpath case. | |
2751 | */ | |
03135cf7 | 2752 | xfs_icsb_lock(mp); |
20b64285 DC |
2753 | |
2754 | /* | |
2755 | * running atomically. | |
2756 | * | |
2757 | * This will leave the counter in the correct state for future | |
2758 | * accesses. After the rebalance, we simply try again and our retry | |
2759 | * will either succeed through the fast path or slow path without | |
2760 | * another balance operation being required. | |
2761 | */ | |
45af6c6d | 2762 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2763 | xfs_icsb_unlock(mp); |
20b64285 | 2764 | goto again; |
8d280b98 | 2765 | } |
20b64285 | 2766 | |
8d280b98 | 2767 | #endif |