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" |
1da177e4 | 20 | #include "xfs_types.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_dir2.h" |
1da177e4 | 29 | #include "xfs_mount.h" |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_btree.h" |
36 | #include "xfs_ialloc.h" | |
1da177e4 LT |
37 | #include "xfs_alloc.h" |
38 | #include "xfs_rtalloc.h" | |
39 | #include "xfs_bmap.h" | |
40 | #include "xfs_error.h" | |
1da177e4 LT |
41 | #include "xfs_quota.h" |
42 | #include "xfs_fsops.h" | |
43355099 | 43 | #include "xfs_utils.h" |
0b1b213f | 44 | #include "xfs_trace.h" |
6d8b79cf | 45 | #include "xfs_icache.h" |
04a1e6c5 DC |
46 | #include "xfs_cksum.h" |
47 | #include "xfs_buf_item.h" | |
0b1b213f | 48 | |
1da177e4 | 49 | |
8d280b98 | 50 | #ifdef HAVE_PERCPU_SB |
20f4ebf2 | 51 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
52 | int); |
53 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
54 | int); | |
36fbe6e6 | 55 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
56 | #else |
57 | ||
45af6c6d CH |
58 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
59 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 DC |
60 | #endif |
61 | ||
1df84c93 | 62 | static const struct { |
8d280b98 DC |
63 | short offset; |
64 | short type; /* 0 = integer | |
65 | * 1 = binary / string (no translation) | |
66 | */ | |
1da177e4 LT |
67 | } xfs_sb_info[] = { |
68 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
69 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
70 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
71 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
72 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
73 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
74 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
75 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
76 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
77 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
78 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
79 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
80 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
81 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
82 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
84 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
87 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
88 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
90 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
95 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
96 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
98 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
101 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
111 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
ee1c0908 | 113 | { offsetof(xfs_sb_t, sb_bad_features2), 0 }, |
04a1e6c5 DC |
114 | { offsetof(xfs_sb_t, sb_features_compat), 0 }, |
115 | { offsetof(xfs_sb_t, sb_features_ro_compat), 0 }, | |
116 | { offsetof(xfs_sb_t, sb_features_incompat), 0 }, | |
e721f504 | 117 | { offsetof(xfs_sb_t, sb_features_log_incompat), 0 }, |
04a1e6c5 | 118 | { offsetof(xfs_sb_t, sb_crc), 0 }, |
e721f504 | 119 | { offsetof(xfs_sb_t, sb_pad), 0 }, |
04a1e6c5 DC |
120 | { offsetof(xfs_sb_t, sb_pquotino), 0 }, |
121 | { offsetof(xfs_sb_t, sb_lsn), 0 }, | |
1da177e4 LT |
122 | { sizeof(xfs_sb_t), 0 } |
123 | }; | |
124 | ||
27174203 CH |
125 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
126 | static int xfs_uuid_table_size; | |
127 | static uuid_t *xfs_uuid_table; | |
128 | ||
129 | /* | |
130 | * See if the UUID is unique among mounted XFS filesystems. | |
131 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
132 | */ | |
133 | STATIC int | |
134 | xfs_uuid_mount( | |
135 | struct xfs_mount *mp) | |
136 | { | |
137 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
138 | int hole, i; | |
139 | ||
140 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
141 | return 0; | |
142 | ||
143 | if (uuid_is_nil(uuid)) { | |
0b932ccc | 144 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
145 | return XFS_ERROR(EINVAL); |
146 | } | |
147 | ||
148 | mutex_lock(&xfs_uuid_table_mutex); | |
149 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
150 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
151 | hole = i; | |
152 | continue; | |
153 | } | |
154 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
155 | goto out_duplicate; | |
156 | } | |
157 | ||
158 | if (hole < 0) { | |
159 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
160 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
161 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
162 | KM_SLEEP); | |
163 | hole = xfs_uuid_table_size++; | |
164 | } | |
165 | xfs_uuid_table[hole] = *uuid; | |
166 | mutex_unlock(&xfs_uuid_table_mutex); | |
167 | ||
168 | return 0; | |
169 | ||
170 | out_duplicate: | |
171 | mutex_unlock(&xfs_uuid_table_mutex); | |
021000e5 | 172 | xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid); |
27174203 CH |
173 | return XFS_ERROR(EINVAL); |
174 | } | |
175 | ||
176 | STATIC void | |
177 | xfs_uuid_unmount( | |
178 | struct xfs_mount *mp) | |
179 | { | |
180 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
181 | int i; | |
182 | ||
183 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
184 | return; | |
185 | ||
186 | mutex_lock(&xfs_uuid_table_mutex); | |
187 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
188 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
189 | continue; | |
190 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
191 | continue; | |
192 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
193 | break; | |
194 | } | |
195 | ASSERT(i < xfs_uuid_table_size); | |
196 | mutex_unlock(&xfs_uuid_table_mutex); | |
197 | } | |
198 | ||
199 | ||
0fa800fb DC |
200 | /* |
201 | * Reference counting access wrappers to the perag structures. | |
e176579e DC |
202 | * Because we never free per-ag structures, the only thing we |
203 | * have to protect against changes is the tree structure itself. | |
0fa800fb DC |
204 | */ |
205 | struct xfs_perag * | |
206 | xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno) | |
207 | { | |
208 | struct xfs_perag *pag; | |
209 | int ref = 0; | |
210 | ||
e176579e | 211 | rcu_read_lock(); |
0fa800fb DC |
212 | pag = radix_tree_lookup(&mp->m_perag_tree, agno); |
213 | if (pag) { | |
214 | ASSERT(atomic_read(&pag->pag_ref) >= 0); | |
0fa800fb DC |
215 | ref = atomic_inc_return(&pag->pag_ref); |
216 | } | |
e176579e | 217 | rcu_read_unlock(); |
0fa800fb DC |
218 | trace_xfs_perag_get(mp, agno, ref, _RET_IP_); |
219 | return pag; | |
220 | } | |
221 | ||
65d0f205 DC |
222 | /* |
223 | * search from @first to find the next perag with the given tag set. | |
224 | */ | |
225 | struct xfs_perag * | |
226 | xfs_perag_get_tag( | |
227 | struct xfs_mount *mp, | |
228 | xfs_agnumber_t first, | |
229 | int tag) | |
230 | { | |
231 | struct xfs_perag *pag; | |
232 | int found; | |
233 | int ref; | |
234 | ||
235 | rcu_read_lock(); | |
236 | found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, | |
237 | (void **)&pag, first, 1, tag); | |
238 | if (found <= 0) { | |
239 | rcu_read_unlock(); | |
240 | return NULL; | |
241 | } | |
242 | ref = atomic_inc_return(&pag->pag_ref); | |
243 | rcu_read_unlock(); | |
244 | trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_); | |
245 | return pag; | |
246 | } | |
247 | ||
0fa800fb DC |
248 | void |
249 | xfs_perag_put(struct xfs_perag *pag) | |
250 | { | |
251 | int ref; | |
252 | ||
253 | ASSERT(atomic_read(&pag->pag_ref) > 0); | |
254 | ref = atomic_dec_return(&pag->pag_ref); | |
255 | trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_); | |
256 | } | |
257 | ||
e176579e DC |
258 | STATIC void |
259 | __xfs_free_perag( | |
260 | struct rcu_head *head) | |
261 | { | |
262 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
263 | ||
264 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
265 | kmem_free(pag); | |
266 | } | |
267 | ||
1da177e4 | 268 | /* |
e176579e | 269 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 270 | */ |
c962fb79 | 271 | STATIC void |
ff4f038c | 272 | xfs_free_perag( |
745f6919 | 273 | xfs_mount_t *mp) |
1da177e4 | 274 | { |
1c1c6ebc DC |
275 | xfs_agnumber_t agno; |
276 | struct xfs_perag *pag; | |
277 | ||
278 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
279 | spin_lock(&mp->m_perag_lock); | |
280 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
281 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 282 | ASSERT(pag); |
f83282a8 | 283 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 284 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 285 | } |
1da177e4 LT |
286 | } |
287 | ||
4cc929ee NS |
288 | /* |
289 | * Check size of device based on the (data/realtime) block count. | |
290 | * Note: this check is used by the growfs code as well as mount. | |
291 | */ | |
292 | int | |
293 | xfs_sb_validate_fsb_count( | |
294 | xfs_sb_t *sbp, | |
295 | __uint64_t nblocks) | |
296 | { | |
297 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
298 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
299 | ||
300 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
301 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 302 | return EFBIG; |
4cc929ee NS |
303 | #else /* Limited by UINT_MAX of sectors */ |
304 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 305 | return EFBIG; |
4cc929ee NS |
306 | #endif |
307 | return 0; | |
308 | } | |
1da177e4 LT |
309 | |
310 | /* | |
311 | * Check the validity of the SB found. | |
312 | */ | |
313 | STATIC int | |
314 | xfs_mount_validate_sb( | |
315 | xfs_mount_t *mp, | |
764d1f89 | 316 | xfs_sb_t *sbp, |
47ad2fcb DC |
317 | bool check_inprogress, |
318 | bool check_version) | |
1da177e4 | 319 | { |
af34e09d | 320 | |
1da177e4 LT |
321 | /* |
322 | * If the log device and data device have the | |
323 | * same device number, the log is internal. | |
324 | * Consequently, the sb_logstart should be non-zero. If | |
325 | * we have a zero sb_logstart in this case, we may be trying to mount | |
326 | * a volume filesystem in a non-volume manner. | |
327 | */ | |
328 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
98021821 | 329 | xfs_warn(mp, "bad magic number"); |
1da177e4 LT |
330 | return XFS_ERROR(EWRONGFS); |
331 | } | |
332 | ||
04a1e6c5 | 333 | |
62118709 | 334 | if (!xfs_sb_good_version(sbp)) { |
98021821 | 335 | xfs_warn(mp, "bad version"); |
1da177e4 LT |
336 | return XFS_ERROR(EWRONGFS); |
337 | } | |
338 | ||
04a1e6c5 | 339 | /* |
e721f504 | 340 | * Version 5 superblock feature mask validation. Reject combinations the |
47ad2fcb DC |
341 | * kernel cannot support up front before checking anything else. For |
342 | * write validation, we don't need to check feature masks. | |
04a1e6c5 | 343 | */ |
47ad2fcb | 344 | if (check_version && XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5) { |
04a1e6c5 | 345 | xfs_alert(mp, |
e721f504 DC |
346 | "Version 5 superblock detected. This kernel has EXPERIMENTAL support enabled!\n" |
347 | "Use of these features in this kernel is at your own risk!"); | |
348 | ||
349 | if (xfs_sb_has_compat_feature(sbp, | |
350 | XFS_SB_FEAT_COMPAT_UNKNOWN)) { | |
351 | xfs_warn(mp, | |
352 | "Superblock has unknown compatible features (0x%x) enabled.\n" | |
353 | "Using a more recent kernel is recommended.", | |
354 | (sbp->sb_features_compat & | |
355 | XFS_SB_FEAT_COMPAT_UNKNOWN)); | |
356 | } | |
357 | ||
358 | if (xfs_sb_has_ro_compat_feature(sbp, | |
359 | XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) { | |
360 | xfs_alert(mp, | |
361 | "Superblock has unknown read-only compatible features (0x%x) enabled.", | |
362 | (sbp->sb_features_ro_compat & | |
363 | XFS_SB_FEAT_RO_COMPAT_UNKNOWN)); | |
364 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { | |
365 | xfs_warn(mp, | |
366 | "Attempted to mount read-only compatible filesystem read-write.\n" | |
367 | "Filesystem can only be safely mounted read only."); | |
368 | return XFS_ERROR(EINVAL); | |
369 | } | |
370 | } | |
371 | if (xfs_sb_has_incompat_feature(sbp, | |
372 | XFS_SB_FEAT_INCOMPAT_UNKNOWN)) { | |
373 | xfs_warn(mp, | |
374 | "Superblock has unknown incompatible features (0x%x) enabled.\n" | |
375 | "Filesystem can not be safely mounted by this kernel.", | |
376 | (sbp->sb_features_incompat & | |
377 | XFS_SB_FEAT_INCOMPAT_UNKNOWN)); | |
378 | return XFS_ERROR(EINVAL); | |
379 | } | |
04a1e6c5 DC |
380 | } |
381 | ||
1da177e4 LT |
382 | if (unlikely( |
383 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
98021821 | 384 | xfs_warn(mp, |
af34e09d DC |
385 | "filesystem is marked as having an external log; " |
386 | "specify logdev on the mount command line."); | |
764d1f89 | 387 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
388 | } |
389 | ||
390 | if (unlikely( | |
391 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
98021821 | 392 | xfs_warn(mp, |
af34e09d DC |
393 | "filesystem is marked as having an internal log; " |
394 | "do not specify logdev on the mount command line."); | |
764d1f89 | 395 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
396 | } |
397 | ||
398 | /* | |
c0e090ce | 399 | * More sanity checking. Most of these were stolen directly from |
1da177e4 LT |
400 | * xfs_repair. |
401 | */ | |
402 | if (unlikely( | |
403 | sbp->sb_agcount <= 0 || | |
404 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
405 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
406 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
407 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 408 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
409 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
410 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
411 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
412 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 413 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
414 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
415 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
416 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
417 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 418 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 419 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
420 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
421 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
c0e090ce ES |
422 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */) || |
423 | sbp->sb_dblocks == 0 || | |
424 | sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp) || | |
425 | sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) { | |
98021821 | 426 | XFS_CORRUPTION_ERROR("SB sanity check failed", |
c0e090ce | 427 | XFS_ERRLEVEL_LOW, mp, sbp); |
1da177e4 LT |
428 | return XFS_ERROR(EFSCORRUPTED); |
429 | } | |
430 | ||
2edbddd5 LM |
431 | /* |
432 | * Until this is fixed only page-sized or smaller data blocks work. | |
433 | */ | |
434 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
98021821 | 435 | xfs_warn(mp, |
af34e09d DC |
436 | "File system with blocksize %d bytes. " |
437 | "Only pagesize (%ld) or less will currently work.", | |
438 | sbp->sb_blocksize, PAGE_SIZE); | |
2edbddd5 LM |
439 | return XFS_ERROR(ENOSYS); |
440 | } | |
441 | ||
1a5902c5 CH |
442 | /* |
443 | * Currently only very few inode sizes are supported. | |
444 | */ | |
445 | switch (sbp->sb_inodesize) { | |
446 | case 256: | |
447 | case 512: | |
448 | case 1024: | |
449 | case 2048: | |
450 | break; | |
451 | default: | |
98021821 | 452 | xfs_warn(mp, "inode size of %d bytes not supported", |
af34e09d | 453 | sbp->sb_inodesize); |
1a5902c5 CH |
454 | return XFS_ERROR(ENOSYS); |
455 | } | |
456 | ||
4cc929ee NS |
457 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
458 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
98021821 | 459 | xfs_warn(mp, |
af34e09d | 460 | "file system too large to be mounted on this system."); |
657a4cff | 461 | return XFS_ERROR(EFBIG); |
1da177e4 LT |
462 | } |
463 | ||
98021821 DC |
464 | if (check_inprogress && sbp->sb_inprogress) { |
465 | xfs_warn(mp, "Offline file system operation in progress!"); | |
1da177e4 LT |
466 | return XFS_ERROR(EFSCORRUPTED); |
467 | } | |
468 | ||
de20614b NS |
469 | /* |
470 | * Version 1 directory format has never worked on Linux. | |
471 | */ | |
62118709 | 472 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
98021821 | 473 | xfs_warn(mp, "file system using version 1 directory format"); |
de20614b NS |
474 | return XFS_ERROR(ENOSYS); |
475 | } | |
476 | ||
1da177e4 LT |
477 | return 0; |
478 | } | |
479 | ||
1c1c6ebc | 480 | int |
c11e2c36 | 481 | xfs_initialize_perag( |
c11e2c36 | 482 | xfs_mount_t *mp, |
1c1c6ebc DC |
483 | xfs_agnumber_t agcount, |
484 | xfs_agnumber_t *maxagi) | |
1da177e4 | 485 | { |
2d2194f6 | 486 | xfs_agnumber_t index; |
8b26c582 | 487 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
488 | xfs_perag_t *pag; |
489 | xfs_agino_t agino; | |
490 | xfs_ino_t ino; | |
491 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 492 | int error = -ENOMEM; |
1da177e4 | 493 | |
1c1c6ebc DC |
494 | /* |
495 | * Walk the current per-ag tree so we don't try to initialise AGs | |
496 | * that already exist (growfs case). Allocate and insert all the | |
497 | * AGs we don't find ready for initialisation. | |
498 | */ | |
499 | for (index = 0; index < agcount; index++) { | |
500 | pag = xfs_perag_get(mp, index); | |
501 | if (pag) { | |
502 | xfs_perag_put(pag); | |
503 | continue; | |
504 | } | |
8b26c582 DC |
505 | if (!first_initialised) |
506 | first_initialised = index; | |
fb3b504a | 507 | |
1c1c6ebc DC |
508 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
509 | if (!pag) | |
8b26c582 | 510 | goto out_unwind; |
fb3b504a CH |
511 | pag->pag_agno = index; |
512 | pag->pag_mount = mp; | |
1a427ab0 | 513 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 514 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 515 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
516 | spin_lock_init(&pag->pag_buf_lock); |
517 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 518 | |
1c1c6ebc | 519 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 520 | goto out_unwind; |
fb3b504a | 521 | |
1c1c6ebc DC |
522 | spin_lock(&mp->m_perag_lock); |
523 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
524 | BUG(); | |
525 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
526 | radix_tree_preload_end(); |
527 | error = -EEXIST; | |
528 | goto out_unwind; | |
1c1c6ebc DC |
529 | } |
530 | spin_unlock(&mp->m_perag_lock); | |
531 | radix_tree_preload_end(); | |
532 | } | |
533 | ||
fb3b504a CH |
534 | /* |
535 | * If we mount with the inode64 option, or no inode overflows | |
536 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 537 | */ |
fb3b504a CH |
538 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
539 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
540 | ||
541 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 542 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 543 | else |
1da177e4 | 544 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 545 | |
2d2194f6 CM |
546 | if (mp->m_flags & XFS_MOUNT_32BITINODES) |
547 | index = xfs_set_inode32(mp); | |
548 | else | |
549 | index = xfs_set_inode64(mp); | |
fb3b504a | 550 | |
1c1c6ebc DC |
551 | if (maxagi) |
552 | *maxagi = index; | |
553 | return 0; | |
8b26c582 DC |
554 | |
555 | out_unwind: | |
556 | kmem_free(pag); | |
557 | for (; index > first_initialised; index--) { | |
558 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
559 | kmem_free(pag); | |
560 | } | |
561 | return error; | |
1da177e4 LT |
562 | } |
563 | ||
2bdf7cd0 CH |
564 | void |
565 | xfs_sb_from_disk( | |
98021821 | 566 | struct xfs_sb *to, |
2bdf7cd0 CH |
567 | xfs_dsb_t *from) |
568 | { | |
569 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
570 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
571 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
572 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
573 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
574 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
575 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
576 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
577 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
578 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
579 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
580 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
581 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
582 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
583 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
584 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
585 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
586 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
587 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
588 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
589 | to->sb_blocklog = from->sb_blocklog; | |
590 | to->sb_sectlog = from->sb_sectlog; | |
591 | to->sb_inodelog = from->sb_inodelog; | |
592 | to->sb_inopblog = from->sb_inopblog; | |
593 | to->sb_agblklog = from->sb_agblklog; | |
594 | to->sb_rextslog = from->sb_rextslog; | |
595 | to->sb_inprogress = from->sb_inprogress; | |
596 | to->sb_imax_pct = from->sb_imax_pct; | |
597 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
598 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
599 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
600 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
601 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
602 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
603 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
604 | to->sb_flags = from->sb_flags; | |
605 | to->sb_shared_vn = from->sb_shared_vn; | |
606 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
607 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
608 | to->sb_width = be32_to_cpu(from->sb_width); | |
609 | to->sb_dirblklog = from->sb_dirblklog; | |
610 | to->sb_logsectlog = from->sb_logsectlog; | |
611 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
612 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
613 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 614 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
04a1e6c5 DC |
615 | to->sb_features_compat = be32_to_cpu(from->sb_features_compat); |
616 | to->sb_features_ro_compat = be32_to_cpu(from->sb_features_ro_compat); | |
617 | to->sb_features_incompat = be32_to_cpu(from->sb_features_incompat); | |
e721f504 DC |
618 | to->sb_features_log_incompat = |
619 | be32_to_cpu(from->sb_features_log_incompat); | |
620 | to->sb_pad = 0; | |
04a1e6c5 DC |
621 | to->sb_pquotino = be64_to_cpu(from->sb_pquotino); |
622 | to->sb_lsn = be64_to_cpu(from->sb_lsn); | |
2bdf7cd0 CH |
623 | } |
624 | ||
1da177e4 | 625 | /* |
2bdf7cd0 | 626 | * Copy in core superblock to ondisk one. |
1da177e4 | 627 | * |
2bdf7cd0 | 628 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
629 | */ |
630 | void | |
2bdf7cd0 CH |
631 | xfs_sb_to_disk( |
632 | xfs_dsb_t *to, | |
633 | xfs_sb_t *from, | |
1da177e4 LT |
634 | __int64_t fields) |
635 | { | |
2bdf7cd0 CH |
636 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
637 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
638 | xfs_sb_field_t f; |
639 | int first; | |
640 | int size; | |
641 | ||
1da177e4 | 642 | ASSERT(fields); |
1da177e4 LT |
643 | if (!fields) |
644 | return; | |
645 | ||
1da177e4 LT |
646 | while (fields) { |
647 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
648 | first = xfs_sb_info[f].offset; | |
649 | size = xfs_sb_info[f + 1].offset - first; | |
650 | ||
651 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
652 | ||
653 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 654 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
655 | } else { |
656 | switch (size) { | |
657 | case 2: | |
2bdf7cd0 CH |
658 | *(__be16 *)(to_ptr + first) = |
659 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
660 | break; |
661 | case 4: | |
2bdf7cd0 CH |
662 | *(__be32 *)(to_ptr + first) = |
663 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
664 | break; |
665 | case 8: | |
2bdf7cd0 CH |
666 | *(__be64 *)(to_ptr + first) = |
667 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
668 | break; |
669 | default: | |
670 | ASSERT(0); | |
671 | } | |
672 | } | |
673 | ||
674 | fields &= ~(1LL << f); | |
675 | } | |
676 | } | |
677 | ||
04a1e6c5 | 678 | static int |
612cfbfe | 679 | xfs_sb_verify( |
47ad2fcb DC |
680 | struct xfs_buf *bp, |
681 | bool check_version) | |
98021821 DC |
682 | { |
683 | struct xfs_mount *mp = bp->b_target->bt_mount; | |
684 | struct xfs_sb sb; | |
98021821 DC |
685 | |
686 | xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp)); | |
687 | ||
688 | /* | |
689 | * Only check the in progress field for the primary superblock as | |
690 | * mkfs.xfs doesn't clear it from secondary superblocks. | |
691 | */ | |
47ad2fcb DC |
692 | return xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR, |
693 | check_version); | |
612cfbfe DC |
694 | } |
695 | ||
04a1e6c5 DC |
696 | /* |
697 | * If the superblock has the CRC feature bit set or the CRC field is non-null, | |
698 | * check that the CRC is valid. We check the CRC field is non-null because a | |
699 | * single bit error could clear the feature bit and unused parts of the | |
700 | * superblock are supposed to be zero. Hence a non-null crc field indicates that | |
701 | * we've potentially lost a feature bit and we should check it anyway. | |
702 | */ | |
1813dd64 | 703 | static void |
612cfbfe DC |
704 | xfs_sb_read_verify( |
705 | struct xfs_buf *bp) | |
706 | { | |
04a1e6c5 DC |
707 | struct xfs_mount *mp = bp->b_target->bt_mount; |
708 | struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp); | |
709 | int error; | |
710 | ||
711 | /* | |
712 | * open code the version check to avoid needing to convert the entire | |
713 | * superblock from disk order just to check the version number | |
714 | */ | |
715 | if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC) && | |
716 | (((be16_to_cpu(dsb->sb_versionnum) & XFS_SB_VERSION_NUMBITS) == | |
717 | XFS_SB_VERSION_5) || | |
718 | dsb->sb_crc != 0)) { | |
719 | ||
720 | if (!xfs_verify_cksum(bp->b_addr, be16_to_cpu(dsb->sb_sectsize), | |
721 | offsetof(struct xfs_sb, sb_crc))) { | |
722 | error = EFSCORRUPTED; | |
723 | goto out_error; | |
724 | } | |
725 | } | |
47ad2fcb | 726 | error = xfs_sb_verify(bp, true); |
04a1e6c5 DC |
727 | |
728 | out_error: | |
729 | if (error) { | |
730 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
731 | xfs_buf_ioerror(bp, error); | |
732 | } | |
98021821 DC |
733 | } |
734 | ||
735 | /* | |
736 | * We may be probed for a filesystem match, so we may not want to emit | |
737 | * messages when the superblock buffer is not actually an XFS superblock. | |
738 | * If we find an XFS superblock, the run a normal, noisy mount because we are | |
739 | * really going to mount it and want to know about errors. | |
740 | */ | |
1813dd64 | 741 | static void |
98021821 DC |
742 | xfs_sb_quiet_read_verify( |
743 | struct xfs_buf *bp) | |
744 | { | |
04a1e6c5 | 745 | struct xfs_dsb *dsb = XFS_BUF_TO_SBP(bp); |
98021821 | 746 | |
98021821 | 747 | |
04a1e6c5 | 748 | if (dsb->sb_magicnum == cpu_to_be32(XFS_SB_MAGIC)) { |
98021821 DC |
749 | /* XFS filesystem, verify noisily! */ |
750 | xfs_sb_read_verify(bp); | |
751 | return; | |
752 | } | |
753 | /* quietly fail */ | |
aeb4f20a | 754 | xfs_buf_ioerror(bp, EWRONGFS); |
98021821 DC |
755 | } |
756 | ||
1813dd64 DC |
757 | static void |
758 | xfs_sb_write_verify( | |
04a1e6c5 | 759 | struct xfs_buf *bp) |
1813dd64 | 760 | { |
04a1e6c5 DC |
761 | struct xfs_mount *mp = bp->b_target->bt_mount; |
762 | struct xfs_buf_log_item *bip = bp->b_fspriv; | |
763 | int error; | |
764 | ||
47ad2fcb | 765 | error = xfs_sb_verify(bp, false); |
04a1e6c5 DC |
766 | if (error) { |
767 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr); | |
768 | xfs_buf_ioerror(bp, error); | |
769 | return; | |
770 | } | |
771 | ||
772 | if (!xfs_sb_version_hascrc(&mp->m_sb)) | |
773 | return; | |
774 | ||
775 | if (bip) | |
776 | XFS_BUF_TO_SBP(bp)->sb_lsn = cpu_to_be64(bip->bli_item.li_lsn); | |
777 | ||
778 | xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), | |
779 | offsetof(struct xfs_sb, sb_crc)); | |
1813dd64 DC |
780 | } |
781 | ||
782 | const struct xfs_buf_ops xfs_sb_buf_ops = { | |
783 | .verify_read = xfs_sb_read_verify, | |
784 | .verify_write = xfs_sb_write_verify, | |
785 | }; | |
786 | ||
787 | static const struct xfs_buf_ops xfs_sb_quiet_buf_ops = { | |
788 | .verify_read = xfs_sb_quiet_read_verify, | |
789 | .verify_write = xfs_sb_write_verify, | |
790 | }; | |
791 | ||
1da177e4 LT |
792 | /* |
793 | * xfs_readsb | |
794 | * | |
795 | * Does the initial read of the superblock. | |
796 | */ | |
797 | int | |
764d1f89 | 798 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
799 | { |
800 | unsigned int sector_size; | |
04a1e6c5 DC |
801 | struct xfs_buf *bp; |
802 | struct xfs_sb *sbp = &mp->m_sb; | |
1da177e4 | 803 | int error; |
af34e09d | 804 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
805 | |
806 | ASSERT(mp->m_sb_bp == NULL); | |
807 | ASSERT(mp->m_ddev_targp != NULL); | |
808 | ||
809 | /* | |
810 | * Allocate a (locked) buffer to hold the superblock. | |
811 | * This will be kept around at all times to optimize | |
812 | * access to the superblock. | |
813 | */ | |
814 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
815 | |
816 | reread: | |
e70b73f8 | 817 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR, |
98021821 | 818 | BTOBB(sector_size), 0, |
1813dd64 DC |
819 | loud ? &xfs_sb_buf_ops |
820 | : &xfs_sb_quiet_buf_ops); | |
26af6552 | 821 | if (!bp) { |
af34e09d DC |
822 | if (loud) |
823 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 824 | return EIO; |
1da177e4 | 825 | } |
eab4e633 DC |
826 | if (bp->b_error) { |
827 | error = bp->b_error; | |
828 | if (loud) | |
e721f504 | 829 | xfs_warn(mp, "SB validate failed with error %d.", error); |
eab4e633 DC |
830 | goto release_buf; |
831 | } | |
1da177e4 LT |
832 | |
833 | /* | |
834 | * Initialize the mount structure from the superblock. | |
1da177e4 | 835 | */ |
98021821 | 836 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 LT |
837 | |
838 | /* | |
839 | * We must be able to do sector-sized and sector-aligned IO. | |
840 | */ | |
04a1e6c5 | 841 | if (sector_size > sbp->sb_sectsize) { |
af34e09d DC |
842 | if (loud) |
843 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
04a1e6c5 | 844 | sector_size, sbp->sb_sectsize); |
1da177e4 | 845 | error = ENOSYS; |
26af6552 | 846 | goto release_buf; |
1da177e4 LT |
847 | } |
848 | ||
849 | /* | |
850 | * If device sector size is smaller than the superblock size, | |
851 | * re-read the superblock so the buffer is correctly sized. | |
852 | */ | |
04a1e6c5 | 853 | if (sector_size < sbp->sb_sectsize) { |
1da177e4 | 854 | xfs_buf_relse(bp); |
04a1e6c5 | 855 | sector_size = sbp->sb_sectsize; |
26af6552 | 856 | goto reread; |
1da177e4 LT |
857 | } |
858 | ||
5478eead LM |
859 | /* Initialize per-cpu counters */ |
860 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 861 | |
04a1e6c5 DC |
862 | /* no need to be quiet anymore, so reset the buf ops */ |
863 | bp->b_ops = &xfs_sb_buf_ops; | |
864 | ||
1da177e4 | 865 | mp->m_sb_bp = bp; |
26af6552 | 866 | xfs_buf_unlock(bp); |
1da177e4 LT |
867 | return 0; |
868 | ||
26af6552 DC |
869 | release_buf: |
870 | xfs_buf_relse(bp); | |
1da177e4 LT |
871 | return error; |
872 | } | |
873 | ||
874 | ||
875 | /* | |
876 | * xfs_mount_common | |
877 | * | |
878 | * Mount initialization code establishing various mount | |
879 | * fields from the superblock associated with the given | |
880 | * mount structure | |
881 | */ | |
ba0f32d4 | 882 | STATIC void |
1da177e4 LT |
883 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
884 | { | |
1da177e4 | 885 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 886 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
887 | mp->m_maxagi = mp->m_sb.sb_agcount; |
888 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
889 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
890 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
891 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
892 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
893 | mp->m_blockmask = sbp->sb_blocksize - 1; |
894 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
895 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 896 | |
60197e8d CH |
897 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
898 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
899 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
900 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
901 | ||
902 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
903 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
904 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
905 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
906 | ||
907 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
908 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
909 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
910 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
911 | |
912 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
913 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
914 | sbp->sb_inopblock); | |
915 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
916 | } | |
92821e2b DC |
917 | |
918 | /* | |
919 | * xfs_initialize_perag_data | |
920 | * | |
921 | * Read in each per-ag structure so we can count up the number of | |
922 | * allocated inodes, free inodes and used filesystem blocks as this | |
923 | * information is no longer persistent in the superblock. Once we have | |
924 | * this information, write it into the in-core superblock structure. | |
925 | */ | |
926 | STATIC int | |
927 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
928 | { | |
929 | xfs_agnumber_t index; | |
930 | xfs_perag_t *pag; | |
931 | xfs_sb_t *sbp = &mp->m_sb; | |
932 | uint64_t ifree = 0; | |
933 | uint64_t ialloc = 0; | |
934 | uint64_t bfree = 0; | |
935 | uint64_t bfreelst = 0; | |
936 | uint64_t btree = 0; | |
937 | int error; | |
92821e2b DC |
938 | |
939 | for (index = 0; index < agcount; index++) { | |
940 | /* | |
941 | * read the agf, then the agi. This gets us | |
9da096fd | 942 | * all the information we need and populates the |
92821e2b DC |
943 | * per-ag structures for us. |
944 | */ | |
945 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
946 | if (error) | |
947 | return error; | |
948 | ||
949 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
950 | if (error) | |
951 | return error; | |
44b56e0a | 952 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
953 | ifree += pag->pagi_freecount; |
954 | ialloc += pag->pagi_count; | |
955 | bfree += pag->pagf_freeblks; | |
956 | bfreelst += pag->pagf_flcount; | |
957 | btree += pag->pagf_btreeblks; | |
44b56e0a | 958 | xfs_perag_put(pag); |
92821e2b DC |
959 | } |
960 | /* | |
961 | * Overwrite incore superblock counters with just-read data | |
962 | */ | |
3685c2a1 | 963 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
964 | sbp->sb_ifree = ifree; |
965 | sbp->sb_icount = ialloc; | |
966 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 967 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
968 | |
969 | /* Fixup the per-cpu counters as well. */ | |
970 | xfs_icsb_reinit_counters(mp); | |
971 | ||
972 | return 0; | |
973 | } | |
974 | ||
1da177e4 | 975 | /* |
0771fb45 | 976 | * Update alignment values based on mount options and sb values |
1da177e4 | 977 | */ |
0771fb45 | 978 | STATIC int |
7884bc86 | 979 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 980 | { |
1da177e4 | 981 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 982 | |
4249023a | 983 | if (mp->m_dalign) { |
1da177e4 LT |
984 | /* |
985 | * If stripe unit and stripe width are not multiples | |
986 | * of the fs blocksize turn off alignment. | |
987 | */ | |
988 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
989 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
990 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
c0e090ce ES |
991 | xfs_warn(mp, "alignment check failed: " |
992 | "(sunit/swidth vs. blocksize)"); | |
0771fb45 | 993 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
994 | } |
995 | mp->m_dalign = mp->m_swidth = 0; | |
996 | } else { | |
997 | /* | |
998 | * Convert the stripe unit and width to FSBs. | |
999 | */ | |
1000 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
1001 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
1002 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
c0e090ce ES |
1003 | xfs_warn(mp, "alignment check failed: " |
1004 | "(sunit/swidth vs. ag size)"); | |
0771fb45 | 1005 | return XFS_ERROR(EINVAL); |
1da177e4 | 1006 | } |
53487786 DC |
1007 | xfs_warn(mp, |
1008 | "stripe alignment turned off: sunit(%d)/swidth(%d) " | |
1009 | "incompatible with agsize(%d)", | |
1da177e4 LT |
1010 | mp->m_dalign, mp->m_swidth, |
1011 | sbp->sb_agblocks); | |
1012 | ||
1013 | mp->m_dalign = 0; | |
1014 | mp->m_swidth = 0; | |
1015 | } else if (mp->m_dalign) { | |
1016 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
1017 | } else { | |
1018 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
c0e090ce ES |
1019 | xfs_warn(mp, "alignment check failed: " |
1020 | "sunit(%d) less than bsize(%d)", | |
53487786 | 1021 | mp->m_dalign, |
1da177e4 | 1022 | mp->m_blockmask +1); |
0771fb45 | 1023 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1024 | } |
1025 | mp->m_swidth = 0; | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Update superblock with new values | |
1031 | * and log changes | |
1032 | */ | |
62118709 | 1033 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
1034 | if (sbp->sb_unit != mp->m_dalign) { |
1035 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 1036 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
1037 | } |
1038 | if (sbp->sb_width != mp->m_swidth) { | |
1039 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 1040 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
1041 | } |
1042 | } | |
1043 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 1044 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
1045 | mp->m_dalign = sbp->sb_unit; |
1046 | mp->m_swidth = sbp->sb_width; | |
1047 | } | |
1048 | ||
0771fb45 ES |
1049 | return 0; |
1050 | } | |
1da177e4 | 1051 | |
0771fb45 ES |
1052 | /* |
1053 | * Set the maximum inode count for this filesystem | |
1054 | */ | |
1055 | STATIC void | |
1056 | xfs_set_maxicount(xfs_mount_t *mp) | |
1057 | { | |
1058 | xfs_sb_t *sbp = &(mp->m_sb); | |
1059 | __uint64_t icount; | |
1da177e4 | 1060 | |
0771fb45 ES |
1061 | if (sbp->sb_imax_pct) { |
1062 | /* | |
1063 | * Make sure the maximum inode count is a multiple | |
1064 | * of the units we allocate inodes in. | |
1da177e4 | 1065 | */ |
1da177e4 LT |
1066 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
1067 | do_div(icount, 100); | |
1068 | do_div(icount, mp->m_ialloc_blks); | |
1069 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
1070 | sbp->sb_inopblog; | |
0771fb45 | 1071 | } else { |
1da177e4 | 1072 | mp->m_maxicount = 0; |
1da177e4 | 1073 | } |
0771fb45 ES |
1074 | } |
1075 | ||
1076 | /* | |
1077 | * Set the default minimum read and write sizes unless | |
1078 | * already specified in a mount option. | |
1079 | * We use smaller I/O sizes when the file system | |
1080 | * is being used for NFS service (wsync mount option). | |
1081 | */ | |
1082 | STATIC void | |
1083 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
1084 | { | |
1085 | xfs_sb_t *sbp = &(mp->m_sb); | |
1086 | int readio_log, writeio_log; | |
1da177e4 | 1087 | |
1da177e4 LT |
1088 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
1089 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
1090 | readio_log = XFS_WSYNC_READIO_LOG; | |
1091 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
1092 | } else { | |
1093 | readio_log = XFS_READIO_LOG_LARGE; | |
1094 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
1095 | } | |
1096 | } else { | |
1097 | readio_log = mp->m_readio_log; | |
1098 | writeio_log = mp->m_writeio_log; | |
1099 | } | |
1100 | ||
1da177e4 LT |
1101 | if (sbp->sb_blocklog > readio_log) { |
1102 | mp->m_readio_log = sbp->sb_blocklog; | |
1103 | } else { | |
1104 | mp->m_readio_log = readio_log; | |
1105 | } | |
1106 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
1107 | if (sbp->sb_blocklog > writeio_log) { | |
1108 | mp->m_writeio_log = sbp->sb_blocklog; | |
1109 | } else { | |
1110 | mp->m_writeio_log = writeio_log; | |
1111 | } | |
1112 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 1113 | } |
1da177e4 | 1114 | |
055388a3 DC |
1115 | /* |
1116 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
1117 | */ | |
1118 | void | |
1119 | xfs_set_low_space_thresholds( | |
1120 | struct xfs_mount *mp) | |
1121 | { | |
1122 | int i; | |
1123 | ||
1124 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
1125 | __uint64_t space = mp->m_sb.sb_dblocks; | |
1126 | ||
1127 | do_div(space, 100); | |
1128 | mp->m_low_space[i] = space * (i + 1); | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | ||
0771fb45 ES |
1133 | /* |
1134 | * Set whether we're using inode alignment. | |
1135 | */ | |
1136 | STATIC void | |
1137 | xfs_set_inoalignment(xfs_mount_t *mp) | |
1138 | { | |
62118709 | 1139 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
1140 | mp->m_sb.sb_inoalignmt >= |
1141 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
1142 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
1143 | else | |
1144 | mp->m_inoalign_mask = 0; | |
1145 | /* | |
1146 | * If we are using stripe alignment, check whether | |
1147 | * the stripe unit is a multiple of the inode alignment | |
1148 | */ | |
1149 | if (mp->m_dalign && mp->m_inoalign_mask && | |
1150 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
1151 | mp->m_sinoalign = mp->m_dalign; | |
1152 | else | |
1153 | mp->m_sinoalign = 0; | |
0771fb45 ES |
1154 | } |
1155 | ||
1156 | /* | |
1157 | * Check that the data (and log if separate) are an ok size. | |
1158 | */ | |
1159 | STATIC int | |
4249023a | 1160 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
1161 | { |
1162 | xfs_buf_t *bp; | |
1163 | xfs_daddr_t d; | |
0771fb45 | 1164 | |
1da177e4 LT |
1165 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
1166 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 1167 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 1168 | return XFS_ERROR(EFBIG); |
1da177e4 | 1169 | } |
e70b73f8 | 1170 | bp = xfs_buf_read_uncached(mp->m_ddev_targp, |
1922c949 | 1171 | d - XFS_FSS_TO_BB(mp, 1), |
c3f8fc73 | 1172 | XFS_FSS_TO_BB(mp, 1), 0, NULL); |
1922c949 | 1173 | if (!bp) { |
0b932ccc | 1174 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 1175 | return EIO; |
1da177e4 | 1176 | } |
1922c949 | 1177 | xfs_buf_relse(bp); |
1da177e4 | 1178 | |
4249023a | 1179 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
1180 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
1181 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 1182 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 1183 | return XFS_ERROR(EFBIG); |
1da177e4 | 1184 | } |
e70b73f8 | 1185 | bp = xfs_buf_read_uncached(mp->m_logdev_targp, |
1922c949 | 1186 | d - XFS_FSB_TO_BB(mp, 1), |
c3f8fc73 | 1187 | XFS_FSB_TO_BB(mp, 1), 0, NULL); |
1922c949 | 1188 | if (!bp) { |
0b932ccc | 1189 | xfs_warn(mp, "log device read failed"); |
1922c949 | 1190 | return EIO; |
0771fb45 | 1191 | } |
1922c949 | 1192 | xfs_buf_relse(bp); |
0771fb45 ES |
1193 | } |
1194 | return 0; | |
1195 | } | |
1196 | ||
7d095257 CH |
1197 | /* |
1198 | * Clear the quotaflags in memory and in the superblock. | |
1199 | */ | |
1200 | int | |
1201 | xfs_mount_reset_sbqflags( | |
1202 | struct xfs_mount *mp) | |
1203 | { | |
1204 | int error; | |
1205 | struct xfs_trans *tp; | |
1206 | ||
1207 | mp->m_qflags = 0; | |
1208 | ||
1209 | /* | |
1210 | * It is OK to look at sb_qflags here in mount path, | |
1211 | * without m_sb_lock. | |
1212 | */ | |
1213 | if (mp->m_sb.sb_qflags == 0) | |
1214 | return 0; | |
1215 | spin_lock(&mp->m_sb_lock); | |
1216 | mp->m_sb.sb_qflags = 0; | |
1217 | spin_unlock(&mp->m_sb_lock); | |
1218 | ||
1219 | /* | |
1220 | * If the fs is readonly, let the incore superblock run | |
1221 | * with quotas off but don't flush the update out to disk | |
1222 | */ | |
1223 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1224 | return 0; | |
1225 | ||
7d095257 | 1226 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); |
b0c10b98 JL |
1227 | error = xfs_trans_reserve(tp, 0, XFS_QM_SBCHANGE_LOG_RES(mp), |
1228 | 0, 0, XFS_DEFAULT_LOG_COUNT); | |
7d095257 CH |
1229 | if (error) { |
1230 | xfs_trans_cancel(tp, 0); | |
53487786 | 1231 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
1232 | return error; |
1233 | } | |
1234 | ||
1235 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1236 | return xfs_trans_commit(tp, 0); | |
1237 | } | |
1238 | ||
d5db0f97 ES |
1239 | __uint64_t |
1240 | xfs_default_resblks(xfs_mount_t *mp) | |
1241 | { | |
1242 | __uint64_t resblks; | |
1243 | ||
1244 | /* | |
8babd8a2 DC |
1245 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
1246 | * smaller. This is intended to cover concurrent allocation | |
1247 | * transactions when we initially hit enospc. These each require a 4 | |
1248 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
1249 | * allocation reservations. | |
d5db0f97 ES |
1250 | */ |
1251 | resblks = mp->m_sb.sb_dblocks; | |
1252 | do_div(resblks, 20); | |
8babd8a2 | 1253 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
1254 | return resblks; |
1255 | } | |
1256 | ||
0771fb45 | 1257 | /* |
0771fb45 ES |
1258 | * This function does the following on an initial mount of a file system: |
1259 | * - reads the superblock from disk and init the mount struct | |
1260 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1261 | * so we don't mount terabyte filesystems | |
1262 | * - init mount struct realtime fields | |
1263 | * - allocate inode hash table for fs | |
1264 | * - init directory manager | |
1265 | * - perform recovery and init the log manager | |
1266 | */ | |
1267 | int | |
1268 | xfs_mountfs( | |
4249023a | 1269 | xfs_mount_t *mp) |
0771fb45 ES |
1270 | { |
1271 | xfs_sb_t *sbp = &(mp->m_sb); | |
1272 | xfs_inode_t *rip; | |
0771fb45 | 1273 | __uint64_t resblks; |
7d095257 CH |
1274 | uint quotamount = 0; |
1275 | uint quotaflags = 0; | |
0771fb45 ES |
1276 | int error = 0; |
1277 | ||
0771fb45 ES |
1278 | xfs_mount_common(mp, sbp); |
1279 | ||
ee1c0908 | 1280 | /* |
e6957ea4 ES |
1281 | * Check for a mismatched features2 values. Older kernels |
1282 | * read & wrote into the wrong sb offset for sb_features2 | |
1283 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1284 | * when sb_features2 was added, which made older superblock | |
1285 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1286 | * |
e6957ea4 ES |
1287 | * For backwards compatibility, we make both slots equal. |
1288 | * | |
1289 | * If we detect a mismatched field, we OR the set bits into the | |
1290 | * existing features2 field in case it has already been modified; we | |
1291 | * don't want to lose any features. We then update the bad location | |
1292 | * with the ORed value so that older kernels will see any features2 | |
1293 | * flags, and mark the two fields as needing updates once the | |
1294 | * transaction subsystem is online. | |
ee1c0908 | 1295 | */ |
e6957ea4 | 1296 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 1297 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 1298 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 1299 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1300 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1301 | |
1302 | /* | |
1303 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1304 | * slot. | |
1305 | */ | |
7c12f296 TS |
1306 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1307 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1308 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1309 | } |
1310 | ||
1311 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1312 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1313 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1314 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1315 | |
7c12f296 TS |
1316 | /* update sb_versionnum for the clearing of the morebits */ |
1317 | if (!sbp->sb_features2) | |
7884bc86 | 1318 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1319 | } |
1320 | ||
0771fb45 ES |
1321 | /* |
1322 | * Check if sb_agblocks is aligned at stripe boundary | |
1323 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1324 | * allocator alignment is within an ag, therefore ag has | |
1325 | * to be aligned at stripe boundary. | |
1326 | */ | |
7884bc86 | 1327 | error = xfs_update_alignment(mp); |
0771fb45 | 1328 | if (error) |
f9057e3d | 1329 | goto out; |
0771fb45 ES |
1330 | |
1331 | xfs_alloc_compute_maxlevels(mp); | |
1332 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1333 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1334 | xfs_ialloc_compute_maxlevels(mp); | |
1335 | ||
1336 | xfs_set_maxicount(mp); | |
1337 | ||
27174203 CH |
1338 | error = xfs_uuid_mount(mp); |
1339 | if (error) | |
1340 | goto out; | |
1da177e4 | 1341 | |
0771fb45 ES |
1342 | /* |
1343 | * Set the minimum read and write sizes | |
1344 | */ | |
1345 | xfs_set_rw_sizes(mp); | |
1346 | ||
055388a3 DC |
1347 | /* set the low space thresholds for dynamic preallocation */ |
1348 | xfs_set_low_space_thresholds(mp); | |
1349 | ||
0771fb45 ES |
1350 | /* |
1351 | * Set the inode cluster size. | |
1352 | * This may still be overridden by the file system | |
1353 | * block size if it is larger than the chosen cluster size. | |
1354 | */ | |
1355 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1356 | ||
1357 | /* | |
1358 | * Set inode alignment fields | |
1359 | */ | |
1360 | xfs_set_inoalignment(mp); | |
1361 | ||
1362 | /* | |
1363 | * Check that the data (and log if separate) are an ok size. | |
1364 | */ | |
4249023a | 1365 | error = xfs_check_sizes(mp); |
0771fb45 | 1366 | if (error) |
f9057e3d | 1367 | goto out_remove_uuid; |
0771fb45 | 1368 | |
1da177e4 LT |
1369 | /* |
1370 | * Initialize realtime fields in the mount structure | |
1371 | */ | |
0771fb45 ES |
1372 | error = xfs_rtmount_init(mp); |
1373 | if (error) { | |
0b932ccc | 1374 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 1375 | goto out_remove_uuid; |
1da177e4 LT |
1376 | } |
1377 | ||
1da177e4 LT |
1378 | /* |
1379 | * Copies the low order bits of the timestamp and the randomly | |
1380 | * set "sequence" number out of a UUID. | |
1381 | */ | |
1382 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1383 | ||
1da177e4 LT |
1384 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1385 | ||
f6c2d1fa | 1386 | xfs_dir_mount(mp); |
1da177e4 LT |
1387 | |
1388 | /* | |
1389 | * Initialize the attribute manager's entries. | |
1390 | */ | |
1391 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1392 | ||
1393 | /* | |
1394 | * Initialize the precomputed transaction reservations values. | |
1395 | */ | |
1396 | xfs_trans_init(mp); | |
1397 | ||
1da177e4 LT |
1398 | /* |
1399 | * Allocate and initialize the per-ag data. | |
1400 | */ | |
1c1c6ebc | 1401 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 1402 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
1403 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
1404 | if (error) { | |
0b932ccc | 1405 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 1406 | goto out_remove_uuid; |
1c1c6ebc | 1407 | } |
1da177e4 | 1408 | |
f9057e3d | 1409 | if (!sbp->sb_logblocks) { |
0b932ccc | 1410 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
1411 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
1412 | error = XFS_ERROR(EFSCORRUPTED); | |
1413 | goto out_free_perag; | |
1414 | } | |
1415 | ||
1da177e4 LT |
1416 | /* |
1417 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1418 | */ | |
f9057e3d CH |
1419 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1420 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1421 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1422 | if (error) { | |
0b932ccc | 1423 | xfs_warn(mp, "log mount failed"); |
d4f3512b | 1424 | goto out_fail_wait; |
1da177e4 LT |
1425 | } |
1426 | ||
92821e2b DC |
1427 | /* |
1428 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1429 | * not. If it was, with the first phase of recovery has completed, we | |
1430 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1431 | * but they are recovered transactionally in the second recovery phase | |
1432 | * later. | |
1433 | * | |
1434 | * Hence we can safely re-initialise incore superblock counters from | |
1435 | * the per-ag data. These may not be correct if the filesystem was not | |
1436 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1437 | * doing this. | |
1438 | * | |
1439 | * If the filesystem was cleanly unmounted, then we can trust the | |
1440 | * values in the superblock to be correct and we don't need to do | |
1441 | * anything here. | |
1442 | * | |
1443 | * If we are currently making the filesystem, the initialisation will | |
1444 | * fail as the perag data is in an undefined state. | |
1445 | */ | |
92821e2b DC |
1446 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1447 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1448 | !mp->m_sb.sb_inprogress) { | |
1449 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d | 1450 | if (error) |
d4f3512b | 1451 | goto out_fail_wait; |
92821e2b | 1452 | } |
f9057e3d | 1453 | |
1da177e4 LT |
1454 | /* |
1455 | * Get and sanity-check the root inode. | |
1456 | * Save the pointer to it in the mount structure. | |
1457 | */ | |
7b6259e7 | 1458 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 1459 | if (error) { |
0b932ccc | 1460 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 1461 | goto out_log_dealloc; |
1da177e4 LT |
1462 | } |
1463 | ||
1464 | ASSERT(rip != NULL); | |
1da177e4 | 1465 | |
abbede1b | 1466 | if (unlikely(!S_ISDIR(rip->i_d.di_mode))) { |
0b932ccc | 1467 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 1468 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
1469 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1470 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1471 | mp); | |
1472 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1473 | goto out_rele_rip; |
1da177e4 LT |
1474 | } |
1475 | mp->m_rootip = rip; /* save it */ | |
1476 | ||
1477 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1478 | ||
1479 | /* | |
1480 | * Initialize realtime inode pointers in the mount structure | |
1481 | */ | |
0771fb45 ES |
1482 | error = xfs_rtmount_inodes(mp); |
1483 | if (error) { | |
1da177e4 LT |
1484 | /* |
1485 | * Free up the root inode. | |
1486 | */ | |
0b932ccc | 1487 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 1488 | goto out_rele_rip; |
1da177e4 LT |
1489 | } |
1490 | ||
1491 | /* | |
7884bc86 CH |
1492 | * If this is a read-only mount defer the superblock updates until |
1493 | * the next remount into writeable mode. Otherwise we would never | |
1494 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1495 | */ |
7884bc86 CH |
1496 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1497 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 1498 | if (error) { |
0b932ccc | 1499 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 1500 | goto out_rtunmount; |
e5720eec DC |
1501 | } |
1502 | } | |
1da177e4 LT |
1503 | |
1504 | /* | |
1505 | * Initialise the XFS quota management subsystem for this mount | |
1506 | */ | |
7d095257 CH |
1507 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1508 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1509 | if (error) | |
1510 | goto out_rtunmount; | |
1511 | } else { | |
1512 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1513 | ||
1514 | /* | |
1515 | * If a file system had quotas running earlier, but decided to | |
1516 | * mount without -o uquota/pquota/gquota options, revoke the | |
1517 | * quotachecked license. | |
1518 | */ | |
1519 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 1520 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
1521 | error = xfs_mount_reset_sbqflags(mp); |
1522 | if (error) | |
1523 | return error; | |
1524 | } | |
1525 | } | |
1da177e4 LT |
1526 | |
1527 | /* | |
1528 | * Finish recovering the file system. This part needed to be | |
1529 | * delayed until after the root and real-time bitmap inodes | |
1530 | * were consistently read in. | |
1531 | */ | |
4249023a | 1532 | error = xfs_log_mount_finish(mp); |
1da177e4 | 1533 | if (error) { |
0b932ccc | 1534 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 1535 | goto out_rtunmount; |
1da177e4 LT |
1536 | } |
1537 | ||
1538 | /* | |
1539 | * Complete the quota initialisation, post-log-replay component. | |
1540 | */ | |
7d095257 CH |
1541 | if (quotamount) { |
1542 | ASSERT(mp->m_qflags == 0); | |
1543 | mp->m_qflags = quotaflags; | |
1544 | ||
1545 | xfs_qm_mount_quotas(mp); | |
1546 | } | |
1547 | ||
84e1e99f DC |
1548 | /* |
1549 | * Now we are mounted, reserve a small amount of unused space for | |
1550 | * privileged transactions. This is needed so that transaction | |
1551 | * space required for critical operations can dip into this pool | |
1552 | * when at ENOSPC. This is needed for operations like create with | |
1553 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1554 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
1555 | * |
1556 | * This may drive us straight to ENOSPC on mount, but that implies | |
1557 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 1558 | */ |
d5db0f97 ES |
1559 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1560 | resblks = xfs_default_resblks(mp); | |
1561 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
1562 | if (error) | |
0b932ccc DC |
1563 | xfs_warn(mp, |
1564 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 1565 | } |
84e1e99f | 1566 | |
1da177e4 LT |
1567 | return 0; |
1568 | ||
b93b6e43 CH |
1569 | out_rtunmount: |
1570 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1571 | out_rele_rip: |
43355099 | 1572 | IRELE(rip); |
f9057e3d | 1573 | out_log_dealloc: |
21b699c8 | 1574 | xfs_log_unmount(mp); |
d4f3512b DC |
1575 | out_fail_wait: |
1576 | if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) | |
1577 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1578 | xfs_wait_buftarg(mp->m_ddev_targp); | |
f9057e3d | 1579 | out_free_perag: |
ff4f038c | 1580 | xfs_free_perag(mp); |
f9057e3d | 1581 | out_remove_uuid: |
27174203 | 1582 | xfs_uuid_unmount(mp); |
f9057e3d | 1583 | out: |
1da177e4 LT |
1584 | return error; |
1585 | } | |
1586 | ||
1587 | /* | |
1da177e4 LT |
1588 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1589 | * log and makes sure that incore structures are freed. | |
1590 | */ | |
41b5c2e7 CH |
1591 | void |
1592 | xfs_unmountfs( | |
1593 | struct xfs_mount *mp) | |
1da177e4 | 1594 | { |
41b5c2e7 CH |
1595 | __uint64_t resblks; |
1596 | int error; | |
1da177e4 | 1597 | |
579b62fa BF |
1598 | cancel_delayed_work_sync(&mp->m_eofblocks_work); |
1599 | ||
7d095257 | 1600 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1601 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1602 | IRELE(mp->m_rootip); |
1603 | ||
641c56fb DC |
1604 | /* |
1605 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1606 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1607 | * the transaction is still sitting in a iclog. The stale inodes |
1608 | * on that buffer will have their flush locks held until the | |
1609 | * transaction hits the disk and the callbacks run. the inode | |
1610 | * flush takes the flush lock unconditionally and with nothing to | |
1611 | * push out the iclog we will never get that unlocked. hence we | |
1612 | * need to force the log first. | |
1613 | */ | |
a14a348b | 1614 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1615 | |
1616 | /* | |
211e4d43 CH |
1617 | * Flush all pending changes from the AIL. |
1618 | */ | |
1619 | xfs_ail_push_all_sync(mp->m_ail); | |
1620 | ||
1621 | /* | |
1622 | * And reclaim all inodes. At this point there should be no dirty | |
7e18530b DC |
1623 | * inodes and none should be pinned or locked, but use synchronous |
1624 | * reclaim just to be sure. We can stop background inode reclaim | |
1625 | * here as well if it is still running. | |
c854363e | 1626 | */ |
7e18530b | 1627 | cancel_delayed_work_sync(&mp->m_reclaim_work); |
c854363e | 1628 | xfs_reclaim_inodes(mp, SYNC_WAIT); |
1da177e4 | 1629 | |
7d095257 | 1630 | xfs_qm_unmount(mp); |
a357a121 | 1631 | |
84e1e99f DC |
1632 | /* |
1633 | * Unreserve any blocks we have so that when we unmount we don't account | |
1634 | * the reserved free space as used. This is really only necessary for | |
1635 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1636 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1637 | * |
1638 | * We don't bother correcting this elsewhere for lazy superblock | |
1639 | * counting because on mount of an unclean filesystem we reconstruct the | |
1640 | * correct counter value and this is irrelevant. | |
1641 | * | |
1642 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1643 | * we only every apply deltas to the superblock and hence the incore | |
1644 | * value does not matter.... | |
1645 | */ | |
1646 | resblks = 0; | |
714082bc DC |
1647 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1648 | if (error) | |
0b932ccc | 1649 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1650 | "Freespace may not be correct on next mount."); |
1651 | ||
adab0f67 | 1652 | error = xfs_log_sbcount(mp); |
e5720eec | 1653 | if (error) |
0b932ccc | 1654 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1655 | "Freespace may not be correct on next mount."); |
87c7bec7 | 1656 | |
21b699c8 | 1657 | xfs_log_unmount(mp); |
27174203 | 1658 | xfs_uuid_unmount(mp); |
1da177e4 | 1659 | |
1550d0b0 | 1660 | #if defined(DEBUG) |
0ce4cfd4 | 1661 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1662 | #endif |
ff4f038c | 1663 | xfs_free_perag(mp); |
1da177e4 LT |
1664 | } |
1665 | ||
92821e2b DC |
1666 | int |
1667 | xfs_fs_writable(xfs_mount_t *mp) | |
1668 | { | |
d9457dc0 | 1669 | return !(mp->m_super->s_writers.frozen || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1670 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1671 | } |
1672 | ||
1673 | /* | |
b2ce3974 AE |
1674 | * xfs_log_sbcount |
1675 | * | |
adab0f67 | 1676 | * Sync the superblock counters to disk. |
b2ce3974 AE |
1677 | * |
1678 | * Note this code can be called during the process of freezing, so | |
adab0f67 | 1679 | * we may need to use the transaction allocator which does not |
b2ce3974 | 1680 | * block when the transaction subsystem is in its frozen state. |
92821e2b DC |
1681 | */ |
1682 | int | |
adab0f67 | 1683 | xfs_log_sbcount(xfs_mount_t *mp) |
92821e2b DC |
1684 | { |
1685 | xfs_trans_t *tp; | |
1686 | int error; | |
1687 | ||
1688 | if (!xfs_fs_writable(mp)) | |
1689 | return 0; | |
1690 | ||
d4d90b57 | 1691 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1692 | |
1693 | /* | |
1694 | * we don't need to do this if we are updating the superblock | |
1695 | * counters on every modification. | |
1696 | */ | |
1697 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1698 | return 0; | |
1699 | ||
b2ce3974 | 1700 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
e457274b JL |
1701 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
1702 | XFS_DEFAULT_LOG_COUNT); | |
92821e2b DC |
1703 | if (error) { |
1704 | xfs_trans_cancel(tp, 0); | |
1705 | return error; | |
1706 | } | |
1707 | ||
1708 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
adab0f67 | 1709 | xfs_trans_set_sync(tp); |
e5720eec DC |
1710 | error = xfs_trans_commit(tp, 0); |
1711 | return error; | |
92821e2b DC |
1712 | } |
1713 | ||
1da177e4 LT |
1714 | /* |
1715 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1716 | * in-core superblock into the superblock buffer to be logged. | |
1717 | * It does not provide the higher level of locking that is | |
1718 | * needed to protect the in-core superblock from concurrent | |
1719 | * access. | |
1720 | */ | |
1721 | void | |
1722 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1723 | { | |
1724 | xfs_buf_t *bp; | |
1725 | int first; | |
1726 | int last; | |
1727 | xfs_mount_t *mp; | |
1da177e4 LT |
1728 | xfs_sb_field_t f; |
1729 | ||
1730 | ASSERT(fields); | |
1731 | if (!fields) | |
1732 | return; | |
1733 | mp = tp->t_mountp; | |
1734 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1735 | first = sizeof(xfs_sb_t); |
1736 | last = 0; | |
1737 | ||
1738 | /* translate/copy */ | |
1739 | ||
2bdf7cd0 | 1740 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1741 | |
1742 | /* find modified range */ | |
587aa0fe DC |
1743 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); |
1744 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1745 | last = xfs_sb_info[f + 1].offset - 1; | |
1da177e4 LT |
1746 | |
1747 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1748 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1749 | first = xfs_sb_info[f].offset; | |
1750 | ||
04a1e6c5 | 1751 | xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SB_BUF); |
1da177e4 LT |
1752 | xfs_trans_log_buf(tp, bp, first, last); |
1753 | } | |
d210a28c | 1754 | |
d210a28c | 1755 | |
1da177e4 LT |
1756 | /* |
1757 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1758 | * a delta to a specified field in the in-core superblock. Simply | |
1759 | * switch on the field indicated and apply the delta to that field. | |
1760 | * Fields are not allowed to dip below zero, so if the delta would | |
1761 | * do this do not apply it and return EINVAL. | |
1762 | * | |
3685c2a1 | 1763 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1764 | */ |
d96f8f89 | 1765 | STATIC int |
20f4ebf2 DC |
1766 | xfs_mod_incore_sb_unlocked( |
1767 | xfs_mount_t *mp, | |
1768 | xfs_sb_field_t field, | |
1769 | int64_t delta, | |
1770 | int rsvd) | |
1da177e4 LT |
1771 | { |
1772 | int scounter; /* short counter for 32 bit fields */ | |
1773 | long long lcounter; /* long counter for 64 bit fields */ | |
1774 | long long res_used, rem; | |
1775 | ||
1776 | /* | |
1777 | * With the in-core superblock spin lock held, switch | |
1778 | * on the indicated field. Apply the delta to the | |
1779 | * proper field. If the fields value would dip below | |
1780 | * 0, then do not apply the delta and return EINVAL. | |
1781 | */ | |
1782 | switch (field) { | |
1783 | case XFS_SBS_ICOUNT: | |
1784 | lcounter = (long long)mp->m_sb.sb_icount; | |
1785 | lcounter += delta; | |
1786 | if (lcounter < 0) { | |
1787 | ASSERT(0); | |
014c2544 | 1788 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1789 | } |
1790 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1791 | return 0; |
1da177e4 LT |
1792 | case XFS_SBS_IFREE: |
1793 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1794 | lcounter += delta; | |
1795 | if (lcounter < 0) { | |
1796 | ASSERT(0); | |
014c2544 | 1797 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1798 | } |
1799 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1800 | return 0; |
1da177e4 | 1801 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1802 | lcounter = (long long) |
1803 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1804 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1805 | ||
1806 | if (delta > 0) { /* Putting blocks back */ | |
1807 | if (res_used > delta) { | |
1808 | mp->m_resblks_avail += delta; | |
1809 | } else { | |
1810 | rem = delta - res_used; | |
1811 | mp->m_resblks_avail = mp->m_resblks; | |
1812 | lcounter += rem; | |
1813 | } | |
1814 | } else { /* Taking blocks away */ | |
1da177e4 | 1815 | lcounter += delta; |
8babd8a2 DC |
1816 | if (lcounter >= 0) { |
1817 | mp->m_sb.sb_fdblocks = lcounter + | |
1818 | XFS_ALLOC_SET_ASIDE(mp); | |
1819 | return 0; | |
1820 | } | |
1da177e4 | 1821 | |
8babd8a2 DC |
1822 | /* |
1823 | * We are out of blocks, use any available reserved | |
1824 | * blocks if were allowed to. | |
1825 | */ | |
1826 | if (!rsvd) | |
1827 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1828 | |
8babd8a2 DC |
1829 | lcounter = (long long)mp->m_resblks_avail + delta; |
1830 | if (lcounter >= 0) { | |
1831 | mp->m_resblks_avail = lcounter; | |
1832 | return 0; | |
1da177e4 | 1833 | } |
8babd8a2 DC |
1834 | printk_once(KERN_WARNING |
1835 | "Filesystem \"%s\": reserve blocks depleted! " | |
1836 | "Consider increasing reserve pool size.", | |
1837 | mp->m_fsname); | |
1838 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1839 | } |
1840 | ||
4be536de | 1841 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1842 | return 0; |
1da177e4 LT |
1843 | case XFS_SBS_FREXTENTS: |
1844 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1845 | lcounter += delta; | |
1846 | if (lcounter < 0) { | |
014c2544 | 1847 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1848 | } |
1849 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1850 | return 0; |
1da177e4 LT |
1851 | case XFS_SBS_DBLOCKS: |
1852 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1853 | lcounter += delta; | |
1854 | if (lcounter < 0) { | |
1855 | ASSERT(0); | |
014c2544 | 1856 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1857 | } |
1858 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1859 | return 0; |
1da177e4 LT |
1860 | case XFS_SBS_AGCOUNT: |
1861 | scounter = mp->m_sb.sb_agcount; | |
1862 | scounter += delta; | |
1863 | if (scounter < 0) { | |
1864 | ASSERT(0); | |
014c2544 | 1865 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1866 | } |
1867 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1868 | return 0; |
1da177e4 LT |
1869 | case XFS_SBS_IMAX_PCT: |
1870 | scounter = mp->m_sb.sb_imax_pct; | |
1871 | scounter += delta; | |
1872 | if (scounter < 0) { | |
1873 | ASSERT(0); | |
014c2544 | 1874 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1875 | } |
1876 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1877 | return 0; |
1da177e4 LT |
1878 | case XFS_SBS_REXTSIZE: |
1879 | scounter = mp->m_sb.sb_rextsize; | |
1880 | scounter += delta; | |
1881 | if (scounter < 0) { | |
1882 | ASSERT(0); | |
014c2544 | 1883 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1884 | } |
1885 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1886 | return 0; |
1da177e4 LT |
1887 | case XFS_SBS_RBMBLOCKS: |
1888 | scounter = mp->m_sb.sb_rbmblocks; | |
1889 | scounter += delta; | |
1890 | if (scounter < 0) { | |
1891 | ASSERT(0); | |
014c2544 | 1892 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1893 | } |
1894 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1895 | return 0; |
1da177e4 LT |
1896 | case XFS_SBS_RBLOCKS: |
1897 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1898 | lcounter += delta; | |
1899 | if (lcounter < 0) { | |
1900 | ASSERT(0); | |
014c2544 | 1901 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1902 | } |
1903 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1904 | return 0; |
1da177e4 LT |
1905 | case XFS_SBS_REXTENTS: |
1906 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1907 | lcounter += delta; | |
1908 | if (lcounter < 0) { | |
1909 | ASSERT(0); | |
014c2544 | 1910 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1911 | } |
1912 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1913 | return 0; |
1da177e4 LT |
1914 | case XFS_SBS_REXTSLOG: |
1915 | scounter = mp->m_sb.sb_rextslog; | |
1916 | scounter += delta; | |
1917 | if (scounter < 0) { | |
1918 | ASSERT(0); | |
014c2544 | 1919 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1920 | } |
1921 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1922 | return 0; |
1da177e4 LT |
1923 | default: |
1924 | ASSERT(0); | |
014c2544 | 1925 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1926 | } |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1931 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1932 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1933 | * routine to do the work. |
1934 | */ | |
1935 | int | |
20f4ebf2 | 1936 | xfs_mod_incore_sb( |
96540c78 CH |
1937 | struct xfs_mount *mp, |
1938 | xfs_sb_field_t field, | |
1939 | int64_t delta, | |
1940 | int rsvd) | |
1da177e4 | 1941 | { |
96540c78 | 1942 | int status; |
1da177e4 | 1943 | |
8d280b98 | 1944 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1945 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1946 | #endif |
96540c78 CH |
1947 | spin_lock(&mp->m_sb_lock); |
1948 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1949 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1950 | |
014c2544 | 1951 | return status; |
1da177e4 LT |
1952 | } |
1953 | ||
1954 | /* | |
1b040712 | 1955 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1956 | * |
1b040712 CH |
1957 | * The fields and changes to those fields are specified in the array of |
1958 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1959 | * will be applied or none of them will. If any modified field dips below 0, | |
1960 | * then all modifications will be backed out and EINVAL will be returned. | |
1961 | * | |
1962 | * Note that this function may not be used for the superblock values that | |
1963 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1964 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1965 | */ |
1966 | int | |
1b040712 CH |
1967 | xfs_mod_incore_sb_batch( |
1968 | struct xfs_mount *mp, | |
1969 | xfs_mod_sb_t *msb, | |
1970 | uint nmsb, | |
1971 | int rsvd) | |
1da177e4 | 1972 | { |
45c51b99 | 1973 | xfs_mod_sb_t *msbp; |
1b040712 | 1974 | int error = 0; |
1da177e4 LT |
1975 | |
1976 | /* | |
1b040712 CH |
1977 | * Loop through the array of mod structures and apply each individually. |
1978 | * If any fail, then back out all those which have already been applied. | |
1979 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1980 | * changes will be atomic. | |
1da177e4 | 1981 | */ |
3685c2a1 | 1982 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1983 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1984 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1985 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1986 | |
1b040712 CH |
1987 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1988 | msbp->msb_delta, rsvd); | |
1989 | if (error) | |
1990 | goto unwind; | |
1da177e4 | 1991 | } |
1b040712 CH |
1992 | spin_unlock(&mp->m_sb_lock); |
1993 | return 0; | |
1da177e4 | 1994 | |
1b040712 CH |
1995 | unwind: |
1996 | while (--msbp >= msb) { | |
1997 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1998 | -msbp->msb_delta, rsvd); | |
1999 | ASSERT(error == 0); | |
1da177e4 | 2000 | } |
3685c2a1 | 2001 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 2002 | return error; |
1da177e4 LT |
2003 | } |
2004 | ||
2005 | /* | |
2006 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
2007 | * The buffer is returned locked and read in from disk. | |
2008 | * The buffer should be released with a call to xfs_brelse(). | |
2009 | * | |
2010 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
2011 | * the superblock buffer if it can be locked without sleeping. | |
2012 | * If it can't then we'll return NULL. | |
2013 | */ | |
0c842ad4 | 2014 | struct xfs_buf * |
1da177e4 | 2015 | xfs_getsb( |
0c842ad4 CH |
2016 | struct xfs_mount *mp, |
2017 | int flags) | |
1da177e4 | 2018 | { |
0c842ad4 | 2019 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 2020 | |
0c842ad4 CH |
2021 | if (!xfs_buf_trylock(bp)) { |
2022 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 2023 | return NULL; |
0c842ad4 | 2024 | xfs_buf_lock(bp); |
1da177e4 | 2025 | } |
0c842ad4 | 2026 | |
72790aa1 | 2027 | xfs_buf_hold(bp); |
1da177e4 | 2028 | ASSERT(XFS_BUF_ISDONE(bp)); |
014c2544 | 2029 | return bp; |
1da177e4 LT |
2030 | } |
2031 | ||
2032 | /* | |
2033 | * Used to free the superblock along various error paths. | |
2034 | */ | |
2035 | void | |
2036 | xfs_freesb( | |
26af6552 | 2037 | struct xfs_mount *mp) |
1da177e4 | 2038 | { |
26af6552 | 2039 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 2040 | |
26af6552 | 2041 | xfs_buf_lock(bp); |
1da177e4 | 2042 | mp->m_sb_bp = NULL; |
26af6552 | 2043 | xfs_buf_relse(bp); |
1da177e4 LT |
2044 | } |
2045 | ||
1da177e4 LT |
2046 | /* |
2047 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
2048 | * be altered by the mount options, as well as any potential sb_features2 |
2049 | * fixup. Only the first superblock is updated. | |
1da177e4 | 2050 | */ |
7884bc86 | 2051 | int |
ee1c0908 | 2052 | xfs_mount_log_sb( |
1da177e4 LT |
2053 | xfs_mount_t *mp, |
2054 | __int64_t fields) | |
2055 | { | |
2056 | xfs_trans_t *tp; | |
e5720eec | 2057 | int error; |
1da177e4 | 2058 | |
ee1c0908 | 2059 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
2060 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
2061 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
2062 | |
2063 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
5166ab06 JL |
2064 | error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0, |
2065 | XFS_DEFAULT_LOG_COUNT); | |
e5720eec | 2066 | if (error) { |
1da177e4 | 2067 | xfs_trans_cancel(tp, 0); |
e5720eec | 2068 | return error; |
1da177e4 LT |
2069 | } |
2070 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
2071 | error = xfs_trans_commit(tp, 0); |
2072 | return error; | |
1da177e4 | 2073 | } |
8d280b98 | 2074 | |
dda35b8f CH |
2075 | /* |
2076 | * If the underlying (data/log/rt) device is readonly, there are some | |
2077 | * operations that cannot proceed. | |
2078 | */ | |
2079 | int | |
2080 | xfs_dev_is_read_only( | |
2081 | struct xfs_mount *mp, | |
2082 | char *message) | |
2083 | { | |
2084 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
2085 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
2086 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
2087 | xfs_notice(mp, "%s required on read-only device.", message); |
2088 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
2089 | return EROFS; |
2090 | } | |
2091 | return 0; | |
2092 | } | |
8d280b98 DC |
2093 | |
2094 | #ifdef HAVE_PERCPU_SB | |
2095 | /* | |
2096 | * Per-cpu incore superblock counters | |
2097 | * | |
2098 | * Simple concept, difficult implementation | |
2099 | * | |
2100 | * Basically, replace the incore superblock counters with a distributed per cpu | |
2101 | * counter for contended fields (e.g. free block count). | |
2102 | * | |
2103 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
2104 | * hence needs to be accurately read when we are running low on space. Hence | |
2105 | * there is a method to enable and disable the per-cpu counters based on how | |
2106 | * much "stuff" is available in them. | |
2107 | * | |
2108 | * Basically, a counter is enabled if there is enough free resource to justify | |
2109 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2110 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2111 | * re-distribute the available resources. | |
2112 | * | |
2113 | * If, once we redistributed the available resources, we still get a failure, | |
2114 | * we disable the per-cpu counter and go through the slow path. | |
2115 | * | |
2116 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2117 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2118 | * the global superblock. We do this after disabling the counter to prevent |
2119 | * more threads from queueing up on the counter. | |
2120 | * | |
2121 | * Essentially, this means that we still need a lock in the fast path to enable | |
2122 | * synchronisation between the global counters and the per-cpu counters. This | |
2123 | * is not a problem because the lock will be local to a CPU almost all the time | |
2124 | * and have little contention except when we get to ENOSPC conditions. | |
2125 | * | |
2126 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2127 | * path while we do things like enabling and disabling counters and | |
2128 | * synchronising the counters. | |
2129 | * | |
2130 | * Locking rules: | |
2131 | * | |
3685c2a1 | 2132 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2133 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2134 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2135 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2136 | * 5. modifying global counters requires holding m_sb_lock |
2137 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2138 | * and _none_ of the per-cpu locks. |
2139 | * | |
2140 | * Disabled counters are only ever re-enabled by a balance operation | |
2141 | * that results in more free resources per CPU than a given threshold. | |
2142 | * To ensure counters don't remain disabled, they are rebalanced when | |
2143 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2144 | * is present to prevent thrashing). | |
e8234a68 DC |
2145 | */ |
2146 | ||
5a67e4c5 | 2147 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2148 | /* |
2149 | * hot-plug CPU notifier support. | |
8d280b98 | 2150 | * |
5a67e4c5 CS |
2151 | * We need a notifier per filesystem as we need to be able to identify |
2152 | * the filesystem to balance the counters out. This is achieved by | |
2153 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2154 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2155 | */ |
e8234a68 DC |
2156 | STATIC int |
2157 | xfs_icsb_cpu_notify( | |
2158 | struct notifier_block *nfb, | |
2159 | unsigned long action, | |
2160 | void *hcpu) | |
2161 | { | |
2162 | xfs_icsb_cnts_t *cntp; | |
2163 | xfs_mount_t *mp; | |
e8234a68 DC |
2164 | |
2165 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2166 | cntp = (xfs_icsb_cnts_t *) | |
2167 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2168 | switch (action) { | |
2169 | case CPU_UP_PREPARE: | |
8bb78442 | 2170 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2171 | /* Easy Case - initialize the area and locks, and |
2172 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2173 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2174 | break; |
2175 | case CPU_ONLINE: | |
8bb78442 | 2176 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2177 | xfs_icsb_lock(mp); |
45af6c6d CH |
2178 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2179 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2180 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2181 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2182 | break; |
2183 | case CPU_DEAD: | |
8bb78442 | 2184 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2185 | /* Disable all the counters, then fold the dead cpu's |
2186 | * count into the total on the global superblock and | |
2187 | * re-enable the counters. */ | |
03135cf7 | 2188 | xfs_icsb_lock(mp); |
3685c2a1 | 2189 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2190 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2191 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2192 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2193 | ||
2194 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2195 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2196 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2197 | ||
01e1b69c | 2198 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2199 | |
45af6c6d CH |
2200 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2201 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2202 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2203 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2204 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2205 | break; |
2206 | } | |
2207 | ||
2208 | return NOTIFY_OK; | |
2209 | } | |
5a67e4c5 | 2210 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2211 | |
8d280b98 DC |
2212 | int |
2213 | xfs_icsb_init_counters( | |
2214 | xfs_mount_t *mp) | |
2215 | { | |
2216 | xfs_icsb_cnts_t *cntp; | |
2217 | int i; | |
2218 | ||
2219 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2220 | if (mp->m_sb_cnts == NULL) | |
2221 | return -ENOMEM; | |
2222 | ||
5a67e4c5 | 2223 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2224 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2225 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2226 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2227 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2228 | |
8d280b98 DC |
2229 | for_each_online_cpu(i) { |
2230 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2231 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2232 | } |
20b64285 DC |
2233 | |
2234 | mutex_init(&mp->m_icsb_mutex); | |
2235 | ||
8d280b98 DC |
2236 | /* |
2237 | * start with all counters disabled so that the | |
2238 | * initial balance kicks us off correctly | |
2239 | */ | |
2240 | mp->m_icsb_counters = -1; | |
2241 | return 0; | |
2242 | } | |
2243 | ||
5478eead LM |
2244 | void |
2245 | xfs_icsb_reinit_counters( | |
2246 | xfs_mount_t *mp) | |
2247 | { | |
2248 | xfs_icsb_lock(mp); | |
2249 | /* | |
2250 | * start with all counters disabled so that the | |
2251 | * initial balance kicks us off correctly | |
2252 | */ | |
2253 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2254 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2255 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2256 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2257 | xfs_icsb_unlock(mp); |
2258 | } | |
2259 | ||
c962fb79 | 2260 | void |
8d280b98 DC |
2261 | xfs_icsb_destroy_counters( |
2262 | xfs_mount_t *mp) | |
2263 | { | |
e8234a68 | 2264 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2265 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2266 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2267 | } |
03135cf7 | 2268 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2269 | } |
2270 | ||
b8f82a4a | 2271 | STATIC void |
01e1b69c DC |
2272 | xfs_icsb_lock_cntr( |
2273 | xfs_icsb_cnts_t *icsbp) | |
2274 | { | |
2275 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2276 | ndelay(1000); | |
2277 | } | |
2278 | } | |
2279 | ||
b8f82a4a | 2280 | STATIC void |
01e1b69c DC |
2281 | xfs_icsb_unlock_cntr( |
2282 | xfs_icsb_cnts_t *icsbp) | |
2283 | { | |
2284 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2285 | } | |
2286 | ||
8d280b98 | 2287 | |
b8f82a4a | 2288 | STATIC void |
8d280b98 DC |
2289 | xfs_icsb_lock_all_counters( |
2290 | xfs_mount_t *mp) | |
2291 | { | |
2292 | xfs_icsb_cnts_t *cntp; | |
2293 | int i; | |
2294 | ||
2295 | for_each_online_cpu(i) { | |
2296 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2297 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2298 | } |
2299 | } | |
2300 | ||
b8f82a4a | 2301 | STATIC void |
8d280b98 DC |
2302 | xfs_icsb_unlock_all_counters( |
2303 | xfs_mount_t *mp) | |
2304 | { | |
2305 | xfs_icsb_cnts_t *cntp; | |
2306 | int i; | |
2307 | ||
2308 | for_each_online_cpu(i) { | |
2309 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2310 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2311 | } |
2312 | } | |
2313 | ||
2314 | STATIC void | |
2315 | xfs_icsb_count( | |
2316 | xfs_mount_t *mp, | |
2317 | xfs_icsb_cnts_t *cnt, | |
2318 | int flags) | |
2319 | { | |
2320 | xfs_icsb_cnts_t *cntp; | |
2321 | int i; | |
2322 | ||
2323 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2324 | ||
2325 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2326 | xfs_icsb_lock_all_counters(mp); | |
2327 | ||
2328 | for_each_online_cpu(i) { | |
2329 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2330 | cnt->icsb_icount += cntp->icsb_icount; | |
2331 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2332 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2333 | } | |
2334 | ||
2335 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2336 | xfs_icsb_unlock_all_counters(mp); | |
2337 | } | |
2338 | ||
2339 | STATIC int | |
2340 | xfs_icsb_counter_disabled( | |
2341 | xfs_mount_t *mp, | |
2342 | xfs_sb_field_t field) | |
2343 | { | |
2344 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2345 | return test_bit(field, &mp->m_icsb_counters); | |
2346 | } | |
2347 | ||
36fbe6e6 | 2348 | STATIC void |
8d280b98 DC |
2349 | xfs_icsb_disable_counter( |
2350 | xfs_mount_t *mp, | |
2351 | xfs_sb_field_t field) | |
2352 | { | |
2353 | xfs_icsb_cnts_t cnt; | |
2354 | ||
2355 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2356 | ||
20b64285 DC |
2357 | /* |
2358 | * If we are already disabled, then there is nothing to do | |
2359 | * here. We check before locking all the counters to avoid | |
2360 | * the expensive lock operation when being called in the | |
2361 | * slow path and the counter is already disabled. This is | |
2362 | * safe because the only time we set or clear this state is under | |
2363 | * the m_icsb_mutex. | |
2364 | */ | |
2365 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2366 | return; |
20b64285 | 2367 | |
8d280b98 DC |
2368 | xfs_icsb_lock_all_counters(mp); |
2369 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2370 | /* drain back to superblock */ | |
2371 | ||
ce46193b | 2372 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2373 | switch(field) { |
2374 | case XFS_SBS_ICOUNT: | |
2375 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2376 | break; | |
2377 | case XFS_SBS_IFREE: | |
2378 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2379 | break; | |
2380 | case XFS_SBS_FDBLOCKS: | |
2381 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2382 | break; | |
2383 | default: | |
2384 | BUG(); | |
2385 | } | |
2386 | } | |
2387 | ||
2388 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2389 | } |
2390 | ||
2391 | STATIC void | |
2392 | xfs_icsb_enable_counter( | |
2393 | xfs_mount_t *mp, | |
2394 | xfs_sb_field_t field, | |
2395 | uint64_t count, | |
2396 | uint64_t resid) | |
2397 | { | |
2398 | xfs_icsb_cnts_t *cntp; | |
2399 | int i; | |
2400 | ||
2401 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2402 | ||
2403 | xfs_icsb_lock_all_counters(mp); | |
2404 | for_each_online_cpu(i) { | |
2405 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2406 | switch (field) { | |
2407 | case XFS_SBS_ICOUNT: | |
2408 | cntp->icsb_icount = count + resid; | |
2409 | break; | |
2410 | case XFS_SBS_IFREE: | |
2411 | cntp->icsb_ifree = count + resid; | |
2412 | break; | |
2413 | case XFS_SBS_FDBLOCKS: | |
2414 | cntp->icsb_fdblocks = count + resid; | |
2415 | break; | |
2416 | default: | |
2417 | BUG(); | |
2418 | break; | |
2419 | } | |
2420 | resid = 0; | |
2421 | } | |
2422 | clear_bit(field, &mp->m_icsb_counters); | |
2423 | xfs_icsb_unlock_all_counters(mp); | |
2424 | } | |
2425 | ||
dbcabad1 | 2426 | void |
d4d90b57 | 2427 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2428 | xfs_mount_t *mp, |
2429 | int flags) | |
2430 | { | |
2431 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2432 | |
8d280b98 DC |
2433 | xfs_icsb_count(mp, &cnt, flags); |
2434 | ||
8d280b98 DC |
2435 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2436 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2437 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2438 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2439 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2440 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2441 | } |
2442 | ||
2443 | /* | |
2444 | * Accurate update of per-cpu counters to incore superblock | |
2445 | */ | |
d4d90b57 | 2446 | void |
8d280b98 | 2447 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2448 | xfs_mount_t *mp, |
2449 | int flags) | |
8d280b98 | 2450 | { |
d4d90b57 CH |
2451 | spin_lock(&mp->m_sb_lock); |
2452 | xfs_icsb_sync_counters_locked(mp, flags); | |
2453 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2454 | } |
2455 | ||
2456 | /* | |
2457 | * Balance and enable/disable counters as necessary. | |
2458 | * | |
20b64285 DC |
2459 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2460 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2461 | * free blocks is something far enough zero that we aren't going thrash when we | |
2462 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2463 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2464 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2465 | * reenabled. | |
2466 | * | |
2467 | * Note that we can be called when counters are already disabled. | |
2468 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2469 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2470 | */ |
20b64285 DC |
2471 | |
2472 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2473 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2474 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2475 | STATIC void |
45af6c6d | 2476 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2477 | xfs_mount_t *mp, |
2478 | xfs_sb_field_t field, | |
20b64285 | 2479 | int min_per_cpu) |
8d280b98 | 2480 | { |
6fdf8ccc | 2481 | uint64_t count, resid; |
8d280b98 | 2482 | int weight = num_online_cpus(); |
20b64285 | 2483 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2484 | |
8d280b98 DC |
2485 | /* disable counter and sync counter */ |
2486 | xfs_icsb_disable_counter(mp, field); | |
2487 | ||
2488 | /* update counters - first CPU gets residual*/ | |
2489 | switch (field) { | |
2490 | case XFS_SBS_ICOUNT: | |
2491 | count = mp->m_sb.sb_icount; | |
2492 | resid = do_div(count, weight); | |
20b64285 | 2493 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2494 | return; |
8d280b98 DC |
2495 | break; |
2496 | case XFS_SBS_IFREE: | |
2497 | count = mp->m_sb.sb_ifree; | |
2498 | resid = do_div(count, weight); | |
20b64285 | 2499 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2500 | return; |
8d280b98 DC |
2501 | break; |
2502 | case XFS_SBS_FDBLOCKS: | |
2503 | count = mp->m_sb.sb_fdblocks; | |
2504 | resid = do_div(count, weight); | |
20b64285 | 2505 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2506 | return; |
8d280b98 DC |
2507 | break; |
2508 | default: | |
2509 | BUG(); | |
6fdf8ccc | 2510 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2511 | break; |
2512 | } | |
2513 | ||
2514 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2515 | } |
2516 | ||
2517 | STATIC void | |
2518 | xfs_icsb_balance_counter( | |
2519 | xfs_mount_t *mp, | |
2520 | xfs_sb_field_t fields, | |
2521 | int min_per_cpu) | |
2522 | { | |
2523 | spin_lock(&mp->m_sb_lock); | |
2524 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2525 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2526 | } |
2527 | ||
1b040712 | 2528 | int |
20b64285 | 2529 | xfs_icsb_modify_counters( |
8d280b98 DC |
2530 | xfs_mount_t *mp, |
2531 | xfs_sb_field_t field, | |
20f4ebf2 | 2532 | int64_t delta, |
20b64285 | 2533 | int rsvd) |
8d280b98 DC |
2534 | { |
2535 | xfs_icsb_cnts_t *icsbp; | |
2536 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2537 | int ret = 0; |
8d280b98 | 2538 | |
20b64285 | 2539 | might_sleep(); |
8d280b98 | 2540 | again: |
7a9e02d6 CL |
2541 | preempt_disable(); |
2542 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2543 | |
2544 | /* | |
2545 | * if the counter is disabled, go to slow path | |
2546 | */ | |
8d280b98 DC |
2547 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2548 | goto slow_path; | |
20b64285 DC |
2549 | xfs_icsb_lock_cntr(icsbp); |
2550 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2551 | xfs_icsb_unlock_cntr(icsbp); | |
2552 | goto slow_path; | |
2553 | } | |
8d280b98 DC |
2554 | |
2555 | switch (field) { | |
2556 | case XFS_SBS_ICOUNT: | |
2557 | lcounter = icsbp->icsb_icount; | |
2558 | lcounter += delta; | |
2559 | if (unlikely(lcounter < 0)) | |
20b64285 | 2560 | goto balance_counter; |
8d280b98 DC |
2561 | icsbp->icsb_icount = lcounter; |
2562 | break; | |
2563 | ||
2564 | case XFS_SBS_IFREE: | |
2565 | lcounter = icsbp->icsb_ifree; | |
2566 | lcounter += delta; | |
2567 | if (unlikely(lcounter < 0)) | |
20b64285 | 2568 | goto balance_counter; |
8d280b98 DC |
2569 | icsbp->icsb_ifree = lcounter; |
2570 | break; | |
2571 | ||
2572 | case XFS_SBS_FDBLOCKS: | |
2573 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2574 | ||
4be536de | 2575 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2576 | lcounter += delta; |
2577 | if (unlikely(lcounter < 0)) | |
20b64285 | 2578 | goto balance_counter; |
4be536de | 2579 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2580 | break; |
2581 | default: | |
2582 | BUG(); | |
2583 | break; | |
2584 | } | |
01e1b69c | 2585 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2586 | preempt_enable(); |
8d280b98 DC |
2587 | return 0; |
2588 | ||
8d280b98 | 2589 | slow_path: |
7a9e02d6 | 2590 | preempt_enable(); |
8d280b98 | 2591 | |
20b64285 DC |
2592 | /* |
2593 | * serialise with a mutex so we don't burn lots of cpu on | |
2594 | * the superblock lock. We still need to hold the superblock | |
2595 | * lock, however, when we modify the global structures. | |
2596 | */ | |
03135cf7 | 2597 | xfs_icsb_lock(mp); |
20b64285 DC |
2598 | |
2599 | /* | |
2600 | * Now running atomically. | |
2601 | * | |
2602 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2603 | * Drop the lock and try again in the fast path.... | |
2604 | */ | |
2605 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2606 | xfs_icsb_unlock(mp); |
8d280b98 | 2607 | goto again; |
8d280b98 DC |
2608 | } |
2609 | ||
20b64285 DC |
2610 | /* |
2611 | * The counter is currently disabled. Because we are | |
2612 | * running atomically here, we know a rebalance cannot | |
2613 | * be in progress. Hence we can go straight to operating | |
2614 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2615 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2616 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2617 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2618 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2619 | * directly on the global counters. | |
2620 | */ | |
3685c2a1 | 2621 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2622 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2623 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2624 | |
20b64285 DC |
2625 | /* |
2626 | * Now that we've modified the global superblock, we | |
2627 | * may be able to re-enable the distributed counters | |
2628 | * (e.g. lots of space just got freed). After that | |
2629 | * we are done. | |
2630 | */ | |
2631 | if (ret != ENOSPC) | |
45af6c6d | 2632 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2633 | xfs_icsb_unlock(mp); |
8d280b98 | 2634 | return ret; |
8d280b98 | 2635 | |
20b64285 DC |
2636 | balance_counter: |
2637 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2638 | preempt_enable(); |
8d280b98 | 2639 | |
20b64285 DC |
2640 | /* |
2641 | * We may have multiple threads here if multiple per-cpu | |
2642 | * counters run dry at the same time. This will mean we can | |
2643 | * do more balances than strictly necessary but it is not | |
2644 | * the common slowpath case. | |
2645 | */ | |
03135cf7 | 2646 | xfs_icsb_lock(mp); |
20b64285 DC |
2647 | |
2648 | /* | |
2649 | * running atomically. | |
2650 | * | |
2651 | * This will leave the counter in the correct state for future | |
2652 | * accesses. After the rebalance, we simply try again and our retry | |
2653 | * will either succeed through the fast path or slow path without | |
2654 | * another balance operation being required. | |
2655 | */ | |
45af6c6d | 2656 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2657 | xfs_icsb_unlock(mp); |
20b64285 | 2658 | goto again; |
8d280b98 | 2659 | } |
20b64285 | 2660 | |
8d280b98 | 2661 | #endif |