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