]>
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 LT |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" | |
26 | #include "xfs_ag.h" | |
1da177e4 | 27 | #include "xfs_dir2.h" |
1da177e4 | 28 | #include "xfs_mount.h" |
1da177e4 | 29 | #include "xfs_bmap_btree.h" |
a844f451 | 30 | #include "xfs_alloc_btree.h" |
1da177e4 | 31 | #include "xfs_ialloc_btree.h" |
1da177e4 LT |
32 | #include "xfs_dinode.h" |
33 | #include "xfs_inode.h" | |
a844f451 NS |
34 | #include "xfs_btree.h" |
35 | #include "xfs_ialloc.h" | |
1da177e4 LT |
36 | #include "xfs_alloc.h" |
37 | #include "xfs_rtalloc.h" | |
38 | #include "xfs_bmap.h" | |
39 | #include "xfs_error.h" | |
1da177e4 LT |
40 | #include "xfs_rw.h" |
41 | #include "xfs_quota.h" | |
42 | #include "xfs_fsops.h" | |
43355099 | 43 | #include "xfs_utils.h" |
0b1b213f CH |
44 | #include "xfs_trace.h" |
45 | ||
1da177e4 | 46 | |
ba0f32d4 | 47 | STATIC void xfs_unmountfs_wait(xfs_mount_t *); |
1da177e4 | 48 | |
8d280b98 DC |
49 | |
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 }, |
1da177e4 LT |
114 | { sizeof(xfs_sb_t), 0 } |
115 | }; | |
116 | ||
27174203 CH |
117 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
118 | static int xfs_uuid_table_size; | |
119 | static uuid_t *xfs_uuid_table; | |
120 | ||
121 | /* | |
122 | * See if the UUID is unique among mounted XFS filesystems. | |
123 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
124 | */ | |
125 | STATIC int | |
126 | xfs_uuid_mount( | |
127 | struct xfs_mount *mp) | |
128 | { | |
129 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
130 | int hole, i; | |
131 | ||
132 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
133 | return 0; | |
134 | ||
135 | if (uuid_is_nil(uuid)) { | |
136 | cmn_err(CE_WARN, | |
137 | "XFS: Filesystem %s has nil UUID - can't mount", | |
138 | mp->m_fsname); | |
139 | return XFS_ERROR(EINVAL); | |
140 | } | |
141 | ||
142 | mutex_lock(&xfs_uuid_table_mutex); | |
143 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
144 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
145 | hole = i; | |
146 | continue; | |
147 | } | |
148 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
149 | goto out_duplicate; | |
150 | } | |
151 | ||
152 | if (hole < 0) { | |
153 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
154 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
155 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
156 | KM_SLEEP); | |
157 | hole = xfs_uuid_table_size++; | |
158 | } | |
159 | xfs_uuid_table[hole] = *uuid; | |
160 | mutex_unlock(&xfs_uuid_table_mutex); | |
161 | ||
162 | return 0; | |
163 | ||
164 | out_duplicate: | |
165 | mutex_unlock(&xfs_uuid_table_mutex); | |
166 | cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount", | |
167 | mp->m_fsname); | |
168 | return XFS_ERROR(EINVAL); | |
169 | } | |
170 | ||
171 | STATIC void | |
172 | xfs_uuid_unmount( | |
173 | struct xfs_mount *mp) | |
174 | { | |
175 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
176 | int i; | |
177 | ||
178 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
179 | return; | |
180 | ||
181 | mutex_lock(&xfs_uuid_table_mutex); | |
182 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
183 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
184 | continue; | |
185 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
186 | continue; | |
187 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
188 | break; | |
189 | } | |
190 | ASSERT(i < xfs_uuid_table_size); | |
191 | mutex_unlock(&xfs_uuid_table_mutex); | |
192 | } | |
193 | ||
194 | ||
0fa800fb DC |
195 | /* |
196 | * Reference counting access wrappers to the perag structures. | |
e176579e DC |
197 | * Because we never free per-ag structures, the only thing we |
198 | * have to protect against changes is the tree structure itself. | |
0fa800fb DC |
199 | */ |
200 | struct xfs_perag * | |
201 | xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno) | |
202 | { | |
203 | struct xfs_perag *pag; | |
204 | int ref = 0; | |
205 | ||
e176579e | 206 | rcu_read_lock(); |
0fa800fb DC |
207 | pag = radix_tree_lookup(&mp->m_perag_tree, agno); |
208 | if (pag) { | |
209 | ASSERT(atomic_read(&pag->pag_ref) >= 0); | |
0fa800fb DC |
210 | ref = atomic_inc_return(&pag->pag_ref); |
211 | } | |
e176579e | 212 | rcu_read_unlock(); |
0fa800fb DC |
213 | trace_xfs_perag_get(mp, agno, ref, _RET_IP_); |
214 | return pag; | |
215 | } | |
216 | ||
65d0f205 DC |
217 | /* |
218 | * search from @first to find the next perag with the given tag set. | |
219 | */ | |
220 | struct xfs_perag * | |
221 | xfs_perag_get_tag( | |
222 | struct xfs_mount *mp, | |
223 | xfs_agnumber_t first, | |
224 | int tag) | |
225 | { | |
226 | struct xfs_perag *pag; | |
227 | int found; | |
228 | int ref; | |
229 | ||
230 | rcu_read_lock(); | |
231 | found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, | |
232 | (void **)&pag, first, 1, tag); | |
233 | if (found <= 0) { | |
234 | rcu_read_unlock(); | |
235 | return NULL; | |
236 | } | |
237 | ref = atomic_inc_return(&pag->pag_ref); | |
238 | rcu_read_unlock(); | |
239 | trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_); | |
240 | return pag; | |
241 | } | |
242 | ||
0fa800fb DC |
243 | void |
244 | xfs_perag_put(struct xfs_perag *pag) | |
245 | { | |
246 | int ref; | |
247 | ||
248 | ASSERT(atomic_read(&pag->pag_ref) > 0); | |
249 | ref = atomic_dec_return(&pag->pag_ref); | |
250 | trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_); | |
251 | } | |
252 | ||
e176579e DC |
253 | STATIC void |
254 | __xfs_free_perag( | |
255 | struct rcu_head *head) | |
256 | { | |
257 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
258 | ||
259 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
260 | kmem_free(pag); | |
261 | } | |
262 | ||
1da177e4 | 263 | /* |
e176579e | 264 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 265 | */ |
c962fb79 | 266 | STATIC void |
ff4f038c | 267 | xfs_free_perag( |
745f6919 | 268 | xfs_mount_t *mp) |
1da177e4 | 269 | { |
1c1c6ebc DC |
270 | xfs_agnumber_t agno; |
271 | struct xfs_perag *pag; | |
272 | ||
273 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
274 | spin_lock(&mp->m_perag_lock); | |
275 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
276 | spin_unlock(&mp->m_perag_lock); | |
e176579e DC |
277 | ASSERT(pag); |
278 | call_rcu(&pag->rcu_head, __xfs_free_perag); | |
1da177e4 | 279 | } |
1da177e4 LT |
280 | } |
281 | ||
4cc929ee NS |
282 | /* |
283 | * Check size of device based on the (data/realtime) block count. | |
284 | * Note: this check is used by the growfs code as well as mount. | |
285 | */ | |
286 | int | |
287 | xfs_sb_validate_fsb_count( | |
288 | xfs_sb_t *sbp, | |
289 | __uint64_t nblocks) | |
290 | { | |
291 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
292 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
293 | ||
294 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
295 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 296 | return EFBIG; |
4cc929ee NS |
297 | #else /* Limited by UINT_MAX of sectors */ |
298 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 299 | return EFBIG; |
4cc929ee NS |
300 | #endif |
301 | return 0; | |
302 | } | |
1da177e4 LT |
303 | |
304 | /* | |
305 | * Check the validity of the SB found. | |
306 | */ | |
307 | STATIC int | |
308 | xfs_mount_validate_sb( | |
309 | xfs_mount_t *mp, | |
764d1f89 NS |
310 | xfs_sb_t *sbp, |
311 | int flags) | |
1da177e4 LT |
312 | { |
313 | /* | |
314 | * If the log device and data device have the | |
315 | * same device number, the log is internal. | |
316 | * Consequently, the sb_logstart should be non-zero. If | |
317 | * we have a zero sb_logstart in this case, we may be trying to mount | |
318 | * a volume filesystem in a non-volume manner. | |
319 | */ | |
320 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 321 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
322 | return XFS_ERROR(EWRONGFS); |
323 | } | |
324 | ||
62118709 | 325 | if (!xfs_sb_good_version(sbp)) { |
764d1f89 | 326 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
327 | return XFS_ERROR(EWRONGFS); |
328 | } | |
329 | ||
330 | if (unlikely( | |
331 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
332 | xfs_fs_mount_cmn_err(flags, |
333 | "filesystem is marked as having an external log; " | |
334 | "specify logdev on the\nmount command line."); | |
335 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
336 | } |
337 | ||
338 | if (unlikely( | |
339 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
340 | xfs_fs_mount_cmn_err(flags, |
341 | "filesystem is marked as having an internal log; " | |
342 | "do not specify logdev on\nthe mount command line."); | |
343 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
344 | } |
345 | ||
346 | /* | |
347 | * More sanity checking. These were stolen directly from | |
348 | * xfs_repair. | |
349 | */ | |
350 | if (unlikely( | |
351 | sbp->sb_agcount <= 0 || | |
352 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
353 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
354 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
355 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 356 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
357 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
358 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
359 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
360 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 361 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
362 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
363 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
364 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
365 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 366 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 367 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
368 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
369 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 370 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 371 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
372 | return XFS_ERROR(EFSCORRUPTED); |
373 | } | |
374 | ||
375 | /* | |
376 | * Sanity check AG count, size fields against data size field | |
377 | */ | |
378 | if (unlikely( | |
379 | sbp->sb_dblocks == 0 || | |
380 | sbp->sb_dblocks > | |
381 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
382 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
383 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 384 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
385 | return XFS_ERROR(EFSCORRUPTED); |
386 | } | |
387 | ||
2edbddd5 LM |
388 | /* |
389 | * Until this is fixed only page-sized or smaller data blocks work. | |
390 | */ | |
391 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
392 | xfs_fs_mount_cmn_err(flags, | |
393 | "file system with blocksize %d bytes", | |
394 | sbp->sb_blocksize); | |
395 | xfs_fs_mount_cmn_err(flags, | |
396 | "only pagesize (%ld) or less will currently work.", | |
397 | PAGE_SIZE); | |
398 | return XFS_ERROR(ENOSYS); | |
399 | } | |
400 | ||
1a5902c5 CH |
401 | /* |
402 | * Currently only very few inode sizes are supported. | |
403 | */ | |
404 | switch (sbp->sb_inodesize) { | |
405 | case 256: | |
406 | case 512: | |
407 | case 1024: | |
408 | case 2048: | |
409 | break; | |
410 | default: | |
411 | xfs_fs_mount_cmn_err(flags, | |
412 | "inode size of %d bytes not supported", | |
413 | sbp->sb_inodesize); | |
414 | return XFS_ERROR(ENOSYS); | |
415 | } | |
416 | ||
4cc929ee NS |
417 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
418 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
764d1f89 NS |
419 | xfs_fs_mount_cmn_err(flags, |
420 | "file system too large to be mounted on this system."); | |
657a4cff | 421 | return XFS_ERROR(EFBIG); |
1da177e4 LT |
422 | } |
423 | ||
424 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 425 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
426 | return XFS_ERROR(EFSCORRUPTED); |
427 | } | |
428 | ||
de20614b NS |
429 | /* |
430 | * Version 1 directory format has never worked on Linux. | |
431 | */ | |
62118709 | 432 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
764d1f89 NS |
433 | xfs_fs_mount_cmn_err(flags, |
434 | "file system using version 1 directory format"); | |
de20614b NS |
435 | return XFS_ERROR(ENOSYS); |
436 | } | |
437 | ||
1da177e4 LT |
438 | return 0; |
439 | } | |
440 | ||
1c1c6ebc | 441 | int |
c11e2c36 | 442 | xfs_initialize_perag( |
c11e2c36 | 443 | xfs_mount_t *mp, |
1c1c6ebc DC |
444 | xfs_agnumber_t agcount, |
445 | xfs_agnumber_t *maxagi) | |
1da177e4 LT |
446 | { |
447 | xfs_agnumber_t index, max_metadata; | |
8b26c582 | 448 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
449 | xfs_perag_t *pag; |
450 | xfs_agino_t agino; | |
451 | xfs_ino_t ino; | |
452 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 453 | int error = -ENOMEM; |
1da177e4 | 454 | |
1c1c6ebc DC |
455 | /* |
456 | * Walk the current per-ag tree so we don't try to initialise AGs | |
457 | * that already exist (growfs case). Allocate and insert all the | |
458 | * AGs we don't find ready for initialisation. | |
459 | */ | |
460 | for (index = 0; index < agcount; index++) { | |
461 | pag = xfs_perag_get(mp, index); | |
462 | if (pag) { | |
463 | xfs_perag_put(pag); | |
464 | continue; | |
465 | } | |
8b26c582 DC |
466 | if (!first_initialised) |
467 | first_initialised = index; | |
fb3b504a | 468 | |
1c1c6ebc DC |
469 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
470 | if (!pag) | |
8b26c582 | 471 | goto out_unwind; |
fb3b504a CH |
472 | pag->pag_agno = index; |
473 | pag->pag_mount = mp; | |
474 | rwlock_init(&pag->pag_ici_lock); | |
69b491c2 | 475 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 476 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
477 | spin_lock_init(&pag->pag_buf_lock); |
478 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 479 | |
1c1c6ebc | 480 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 481 | goto out_unwind; |
fb3b504a | 482 | |
1c1c6ebc DC |
483 | spin_lock(&mp->m_perag_lock); |
484 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
485 | BUG(); | |
486 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
487 | radix_tree_preload_end(); |
488 | error = -EEXIST; | |
489 | goto out_unwind; | |
1c1c6ebc DC |
490 | } |
491 | spin_unlock(&mp->m_perag_lock); | |
492 | radix_tree_preload_end(); | |
493 | } | |
494 | ||
fb3b504a CH |
495 | /* |
496 | * If we mount with the inode64 option, or no inode overflows | |
497 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 498 | */ |
fb3b504a CH |
499 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
500 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
501 | ||
502 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 503 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 504 | else |
1da177e4 | 505 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 506 | |
1da177e4 | 507 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { |
fb3b504a CH |
508 | /* |
509 | * Calculate how much should be reserved for inodes to meet | |
510 | * the max inode percentage. | |
1da177e4 LT |
511 | */ |
512 | if (mp->m_maxicount) { | |
513 | __uint64_t icount; | |
514 | ||
515 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
516 | do_div(icount, 100); | |
517 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 518 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
519 | max_metadata = icount; |
520 | } else { | |
521 | max_metadata = agcount; | |
522 | } | |
fb3b504a | 523 | |
1da177e4 LT |
524 | for (index = 0; index < agcount; index++) { |
525 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
fb3b504a | 526 | if (ino > XFS_MAXINUMBER_32) { |
1da177e4 LT |
527 | index++; |
528 | break; | |
529 | } | |
530 | ||
44b56e0a | 531 | pag = xfs_perag_get(mp, index); |
1da177e4 LT |
532 | pag->pagi_inodeok = 1; |
533 | if (index < max_metadata) | |
534 | pag->pagf_metadata = 1; | |
44b56e0a | 535 | xfs_perag_put(pag); |
1da177e4 LT |
536 | } |
537 | } else { | |
1da177e4 | 538 | for (index = 0; index < agcount; index++) { |
44b56e0a | 539 | pag = xfs_perag_get(mp, index); |
1da177e4 | 540 | pag->pagi_inodeok = 1; |
44b56e0a | 541 | xfs_perag_put(pag); |
1da177e4 LT |
542 | } |
543 | } | |
fb3b504a | 544 | |
1c1c6ebc DC |
545 | if (maxagi) |
546 | *maxagi = index; | |
547 | return 0; | |
8b26c582 DC |
548 | |
549 | out_unwind: | |
550 | kmem_free(pag); | |
551 | for (; index > first_initialised; index--) { | |
552 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
553 | kmem_free(pag); | |
554 | } | |
555 | return error; | |
1da177e4 LT |
556 | } |
557 | ||
2bdf7cd0 CH |
558 | void |
559 | xfs_sb_from_disk( | |
560 | xfs_sb_t *to, | |
561 | xfs_dsb_t *from) | |
562 | { | |
563 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
564 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
565 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
566 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
567 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
568 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
569 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
570 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
571 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
572 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
573 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
574 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
575 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
576 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
577 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
578 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
579 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
580 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
581 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
582 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
583 | to->sb_blocklog = from->sb_blocklog; | |
584 | to->sb_sectlog = from->sb_sectlog; | |
585 | to->sb_inodelog = from->sb_inodelog; | |
586 | to->sb_inopblog = from->sb_inopblog; | |
587 | to->sb_agblklog = from->sb_agblklog; | |
588 | to->sb_rextslog = from->sb_rextslog; | |
589 | to->sb_inprogress = from->sb_inprogress; | |
590 | to->sb_imax_pct = from->sb_imax_pct; | |
591 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
592 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
593 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
594 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
595 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
596 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
597 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
598 | to->sb_flags = from->sb_flags; | |
599 | to->sb_shared_vn = from->sb_shared_vn; | |
600 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
601 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
602 | to->sb_width = be32_to_cpu(from->sb_width); | |
603 | to->sb_dirblklog = from->sb_dirblklog; | |
604 | to->sb_logsectlog = from->sb_logsectlog; | |
605 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
606 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
607 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 608 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
609 | } |
610 | ||
1da177e4 | 611 | /* |
2bdf7cd0 | 612 | * Copy in core superblock to ondisk one. |
1da177e4 | 613 | * |
2bdf7cd0 | 614 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
615 | */ |
616 | void | |
2bdf7cd0 CH |
617 | xfs_sb_to_disk( |
618 | xfs_dsb_t *to, | |
619 | xfs_sb_t *from, | |
1da177e4 LT |
620 | __int64_t fields) |
621 | { | |
2bdf7cd0 CH |
622 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
623 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
624 | xfs_sb_field_t f; |
625 | int first; | |
626 | int size; | |
627 | ||
1da177e4 | 628 | ASSERT(fields); |
1da177e4 LT |
629 | if (!fields) |
630 | return; | |
631 | ||
1da177e4 LT |
632 | while (fields) { |
633 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
634 | first = xfs_sb_info[f].offset; | |
635 | size = xfs_sb_info[f + 1].offset - first; | |
636 | ||
637 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
638 | ||
639 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 640 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
641 | } else { |
642 | switch (size) { | |
643 | case 2: | |
2bdf7cd0 CH |
644 | *(__be16 *)(to_ptr + first) = |
645 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
646 | break; |
647 | case 4: | |
2bdf7cd0 CH |
648 | *(__be32 *)(to_ptr + first) = |
649 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
650 | break; |
651 | case 8: | |
2bdf7cd0 CH |
652 | *(__be64 *)(to_ptr + first) = |
653 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
654 | break; |
655 | default: | |
656 | ASSERT(0); | |
657 | } | |
658 | } | |
659 | ||
660 | fields &= ~(1LL << f); | |
661 | } | |
662 | } | |
663 | ||
664 | /* | |
665 | * xfs_readsb | |
666 | * | |
667 | * Does the initial read of the superblock. | |
668 | */ | |
669 | int | |
764d1f89 | 670 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
671 | { |
672 | unsigned int sector_size; | |
1da177e4 | 673 | xfs_buf_t *bp; |
1da177e4 LT |
674 | int error; |
675 | ||
676 | ASSERT(mp->m_sb_bp == NULL); | |
677 | ASSERT(mp->m_ddev_targp != NULL); | |
678 | ||
679 | /* | |
680 | * Allocate a (locked) buffer to hold the superblock. | |
681 | * This will be kept around at all times to optimize | |
682 | * access to the superblock. | |
683 | */ | |
684 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
685 | |
686 | reread: | |
687 | bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp, | |
688 | XFS_SB_DADDR, sector_size, 0); | |
689 | if (!bp) { | |
690 | xfs_fs_mount_cmn_err(flags, "SB buffer read failed"); | |
691 | return EIO; | |
1da177e4 | 692 | } |
1da177e4 LT |
693 | |
694 | /* | |
695 | * Initialize the mount structure from the superblock. | |
696 | * But first do some basic consistency checking. | |
697 | */ | |
2bdf7cd0 | 698 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
764d1f89 | 699 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 700 | if (error) { |
764d1f89 | 701 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
26af6552 | 702 | goto release_buf; |
1da177e4 LT |
703 | } |
704 | ||
705 | /* | |
706 | * We must be able to do sector-sized and sector-aligned IO. | |
707 | */ | |
708 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
709 | xfs_fs_mount_cmn_err(flags, |
710 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
711 | sector_size, mp->m_sb.sb_sectsize); |
712 | error = ENOSYS; | |
26af6552 | 713 | goto release_buf; |
1da177e4 LT |
714 | } |
715 | ||
716 | /* | |
717 | * If device sector size is smaller than the superblock size, | |
718 | * re-read the superblock so the buffer is correctly sized. | |
719 | */ | |
720 | if (sector_size < mp->m_sb.sb_sectsize) { | |
1da177e4 LT |
721 | xfs_buf_relse(bp); |
722 | sector_size = mp->m_sb.sb_sectsize; | |
26af6552 | 723 | goto reread; |
1da177e4 LT |
724 | } |
725 | ||
5478eead LM |
726 | /* Initialize per-cpu counters */ |
727 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 728 | |
1da177e4 | 729 | mp->m_sb_bp = bp; |
26af6552 | 730 | xfs_buf_unlock(bp); |
1da177e4 LT |
731 | return 0; |
732 | ||
26af6552 DC |
733 | release_buf: |
734 | xfs_buf_relse(bp); | |
1da177e4 LT |
735 | return error; |
736 | } | |
737 | ||
738 | ||
739 | /* | |
740 | * xfs_mount_common | |
741 | * | |
742 | * Mount initialization code establishing various mount | |
743 | * fields from the superblock associated with the given | |
744 | * mount structure | |
745 | */ | |
ba0f32d4 | 746 | STATIC void |
1da177e4 LT |
747 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
748 | { | |
1da177e4 | 749 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 750 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
751 | mp->m_maxagi = mp->m_sb.sb_agcount; |
752 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
753 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
754 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
755 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
756 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
757 | mp->m_blockmask = sbp->sb_blocksize - 1; |
758 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
759 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 760 | |
60197e8d CH |
761 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
762 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
763 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
764 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
765 | ||
766 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
767 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
768 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
769 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
770 | ||
771 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
772 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
773 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
774 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
775 | |
776 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
777 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
778 | sbp->sb_inopblock); | |
779 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
780 | } | |
92821e2b DC |
781 | |
782 | /* | |
783 | * xfs_initialize_perag_data | |
784 | * | |
785 | * Read in each per-ag structure so we can count up the number of | |
786 | * allocated inodes, free inodes and used filesystem blocks as this | |
787 | * information is no longer persistent in the superblock. Once we have | |
788 | * this information, write it into the in-core superblock structure. | |
789 | */ | |
790 | STATIC int | |
791 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
792 | { | |
793 | xfs_agnumber_t index; | |
794 | xfs_perag_t *pag; | |
795 | xfs_sb_t *sbp = &mp->m_sb; | |
796 | uint64_t ifree = 0; | |
797 | uint64_t ialloc = 0; | |
798 | uint64_t bfree = 0; | |
799 | uint64_t bfreelst = 0; | |
800 | uint64_t btree = 0; | |
801 | int error; | |
92821e2b DC |
802 | |
803 | for (index = 0; index < agcount; index++) { | |
804 | /* | |
805 | * read the agf, then the agi. This gets us | |
9da096fd | 806 | * all the information we need and populates the |
92821e2b DC |
807 | * per-ag structures for us. |
808 | */ | |
809 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
810 | if (error) | |
811 | return error; | |
812 | ||
813 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
814 | if (error) | |
815 | return error; | |
44b56e0a | 816 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
817 | ifree += pag->pagi_freecount; |
818 | ialloc += pag->pagi_count; | |
819 | bfree += pag->pagf_freeblks; | |
820 | bfreelst += pag->pagf_flcount; | |
821 | btree += pag->pagf_btreeblks; | |
44b56e0a | 822 | xfs_perag_put(pag); |
92821e2b DC |
823 | } |
824 | /* | |
825 | * Overwrite incore superblock counters with just-read data | |
826 | */ | |
3685c2a1 | 827 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
828 | sbp->sb_ifree = ifree; |
829 | sbp->sb_icount = ialloc; | |
830 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 831 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
832 | |
833 | /* Fixup the per-cpu counters as well. */ | |
834 | xfs_icsb_reinit_counters(mp); | |
835 | ||
836 | return 0; | |
837 | } | |
838 | ||
1da177e4 | 839 | /* |
0771fb45 | 840 | * Update alignment values based on mount options and sb values |
1da177e4 | 841 | */ |
0771fb45 | 842 | STATIC int |
7884bc86 | 843 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 844 | { |
1da177e4 | 845 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 846 | |
4249023a | 847 | if (mp->m_dalign) { |
1da177e4 LT |
848 | /* |
849 | * If stripe unit and stripe width are not multiples | |
850 | * of the fs blocksize turn off alignment. | |
851 | */ | |
852 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
853 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
854 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
855 | cmn_err(CE_WARN, | |
856 | "XFS: alignment check 1 failed"); | |
0771fb45 | 857 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
858 | } |
859 | mp->m_dalign = mp->m_swidth = 0; | |
860 | } else { | |
861 | /* | |
862 | * Convert the stripe unit and width to FSBs. | |
863 | */ | |
864 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
865 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
866 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 867 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
868 | } |
869 | xfs_fs_cmn_err(CE_WARN, mp, | |
870 | "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)", | |
871 | mp->m_dalign, mp->m_swidth, | |
872 | sbp->sb_agblocks); | |
873 | ||
874 | mp->m_dalign = 0; | |
875 | mp->m_swidth = 0; | |
876 | } else if (mp->m_dalign) { | |
877 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
878 | } else { | |
879 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
880 | xfs_fs_cmn_err(CE_WARN, mp, | |
881 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
882 | mp->m_dalign, | |
883 | mp->m_blockmask +1); | |
0771fb45 | 884 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
885 | } |
886 | mp->m_swidth = 0; | |
887 | } | |
888 | } | |
889 | ||
890 | /* | |
891 | * Update superblock with new values | |
892 | * and log changes | |
893 | */ | |
62118709 | 894 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
895 | if (sbp->sb_unit != mp->m_dalign) { |
896 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 897 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
898 | } |
899 | if (sbp->sb_width != mp->m_swidth) { | |
900 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 901 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
902 | } |
903 | } | |
904 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 905 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
906 | mp->m_dalign = sbp->sb_unit; |
907 | mp->m_swidth = sbp->sb_width; | |
908 | } | |
909 | ||
0771fb45 ES |
910 | return 0; |
911 | } | |
1da177e4 | 912 | |
0771fb45 ES |
913 | /* |
914 | * Set the maximum inode count for this filesystem | |
915 | */ | |
916 | STATIC void | |
917 | xfs_set_maxicount(xfs_mount_t *mp) | |
918 | { | |
919 | xfs_sb_t *sbp = &(mp->m_sb); | |
920 | __uint64_t icount; | |
1da177e4 | 921 | |
0771fb45 ES |
922 | if (sbp->sb_imax_pct) { |
923 | /* | |
924 | * Make sure the maximum inode count is a multiple | |
925 | * of the units we allocate inodes in. | |
1da177e4 | 926 | */ |
1da177e4 LT |
927 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
928 | do_div(icount, 100); | |
929 | do_div(icount, mp->m_ialloc_blks); | |
930 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
931 | sbp->sb_inopblog; | |
0771fb45 | 932 | } else { |
1da177e4 | 933 | mp->m_maxicount = 0; |
1da177e4 | 934 | } |
0771fb45 ES |
935 | } |
936 | ||
937 | /* | |
938 | * Set the default minimum read and write sizes unless | |
939 | * already specified in a mount option. | |
940 | * We use smaller I/O sizes when the file system | |
941 | * is being used for NFS service (wsync mount option). | |
942 | */ | |
943 | STATIC void | |
944 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
945 | { | |
946 | xfs_sb_t *sbp = &(mp->m_sb); | |
947 | int readio_log, writeio_log; | |
1da177e4 | 948 | |
1da177e4 LT |
949 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
950 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
951 | readio_log = XFS_WSYNC_READIO_LOG; | |
952 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
953 | } else { | |
954 | readio_log = XFS_READIO_LOG_LARGE; | |
955 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
956 | } | |
957 | } else { | |
958 | readio_log = mp->m_readio_log; | |
959 | writeio_log = mp->m_writeio_log; | |
960 | } | |
961 | ||
1da177e4 LT |
962 | if (sbp->sb_blocklog > readio_log) { |
963 | mp->m_readio_log = sbp->sb_blocklog; | |
964 | } else { | |
965 | mp->m_readio_log = readio_log; | |
966 | } | |
967 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
968 | if (sbp->sb_blocklog > writeio_log) { | |
969 | mp->m_writeio_log = sbp->sb_blocklog; | |
970 | } else { | |
971 | mp->m_writeio_log = writeio_log; | |
972 | } | |
973 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 974 | } |
1da177e4 | 975 | |
0771fb45 ES |
976 | /* |
977 | * Set whether we're using inode alignment. | |
978 | */ | |
979 | STATIC void | |
980 | xfs_set_inoalignment(xfs_mount_t *mp) | |
981 | { | |
62118709 | 982 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
983 | mp->m_sb.sb_inoalignmt >= |
984 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
985 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
986 | else | |
987 | mp->m_inoalign_mask = 0; | |
988 | /* | |
989 | * If we are using stripe alignment, check whether | |
990 | * the stripe unit is a multiple of the inode alignment | |
991 | */ | |
992 | if (mp->m_dalign && mp->m_inoalign_mask && | |
993 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
994 | mp->m_sinoalign = mp->m_dalign; | |
995 | else | |
996 | mp->m_sinoalign = 0; | |
0771fb45 ES |
997 | } |
998 | ||
999 | /* | |
1000 | * Check that the data (and log if separate) are an ok size. | |
1001 | */ | |
1002 | STATIC int | |
4249023a | 1003 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
1004 | { |
1005 | xfs_buf_t *bp; | |
1006 | xfs_daddr_t d; | |
0771fb45 | 1007 | |
1da177e4 LT |
1008 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
1009 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
1922c949 | 1010 | cmn_err(CE_WARN, "XFS: filesystem size mismatch detected"); |
657a4cff | 1011 | return XFS_ERROR(EFBIG); |
1da177e4 | 1012 | } |
1922c949 DC |
1013 | bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp, |
1014 | d - XFS_FSS_TO_BB(mp, 1), | |
1015 | BBTOB(XFS_FSS_TO_BB(mp, 1)), 0); | |
1016 | if (!bp) { | |
1017 | cmn_err(CE_WARN, "XFS: last sector read failed"); | |
1018 | return EIO; | |
1da177e4 | 1019 | } |
1922c949 | 1020 | xfs_buf_relse(bp); |
1da177e4 | 1021 | |
4249023a | 1022 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
1023 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
1024 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
1922c949 | 1025 | cmn_err(CE_WARN, "XFS: log size mismatch detected"); |
657a4cff | 1026 | return XFS_ERROR(EFBIG); |
1da177e4 | 1027 | } |
1922c949 DC |
1028 | bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp, |
1029 | d - XFS_FSB_TO_BB(mp, 1), | |
1030 | XFS_FSB_TO_B(mp, 1), 0); | |
1031 | if (!bp) { | |
1032 | cmn_err(CE_WARN, "XFS: log device read failed"); | |
1033 | return EIO; | |
0771fb45 | 1034 | } |
1922c949 | 1035 | xfs_buf_relse(bp); |
0771fb45 ES |
1036 | } |
1037 | return 0; | |
1038 | } | |
1039 | ||
7d095257 CH |
1040 | /* |
1041 | * Clear the quotaflags in memory and in the superblock. | |
1042 | */ | |
1043 | int | |
1044 | xfs_mount_reset_sbqflags( | |
1045 | struct xfs_mount *mp) | |
1046 | { | |
1047 | int error; | |
1048 | struct xfs_trans *tp; | |
1049 | ||
1050 | mp->m_qflags = 0; | |
1051 | ||
1052 | /* | |
1053 | * It is OK to look at sb_qflags here in mount path, | |
1054 | * without m_sb_lock. | |
1055 | */ | |
1056 | if (mp->m_sb.sb_qflags == 0) | |
1057 | return 0; | |
1058 | spin_lock(&mp->m_sb_lock); | |
1059 | mp->m_sb.sb_qflags = 0; | |
1060 | spin_unlock(&mp->m_sb_lock); | |
1061 | ||
1062 | /* | |
1063 | * If the fs is readonly, let the incore superblock run | |
1064 | * with quotas off but don't flush the update out to disk | |
1065 | */ | |
1066 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1067 | return 0; | |
1068 | ||
1069 | #ifdef QUOTADEBUG | |
1070 | xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes"); | |
1071 | #endif | |
1072 | ||
1073 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); | |
1074 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1075 | XFS_DEFAULT_LOG_COUNT); | |
1076 | if (error) { | |
1077 | xfs_trans_cancel(tp, 0); | |
1078 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1079 | "xfs_mount_reset_sbqflags: Superblock update failed!"); | |
1080 | return error; | |
1081 | } | |
1082 | ||
1083 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1084 | return xfs_trans_commit(tp, 0); | |
1085 | } | |
1086 | ||
d5db0f97 ES |
1087 | __uint64_t |
1088 | xfs_default_resblks(xfs_mount_t *mp) | |
1089 | { | |
1090 | __uint64_t resblks; | |
1091 | ||
1092 | /* | |
8babd8a2 DC |
1093 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
1094 | * smaller. This is intended to cover concurrent allocation | |
1095 | * transactions when we initially hit enospc. These each require a 4 | |
1096 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
1097 | * allocation reservations. | |
d5db0f97 ES |
1098 | */ |
1099 | resblks = mp->m_sb.sb_dblocks; | |
1100 | do_div(resblks, 20); | |
8babd8a2 | 1101 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
1102 | return resblks; |
1103 | } | |
1104 | ||
0771fb45 | 1105 | /* |
0771fb45 ES |
1106 | * This function does the following on an initial mount of a file system: |
1107 | * - reads the superblock from disk and init the mount struct | |
1108 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1109 | * so we don't mount terabyte filesystems | |
1110 | * - init mount struct realtime fields | |
1111 | * - allocate inode hash table for fs | |
1112 | * - init directory manager | |
1113 | * - perform recovery and init the log manager | |
1114 | */ | |
1115 | int | |
1116 | xfs_mountfs( | |
4249023a | 1117 | xfs_mount_t *mp) |
0771fb45 ES |
1118 | { |
1119 | xfs_sb_t *sbp = &(mp->m_sb); | |
1120 | xfs_inode_t *rip; | |
0771fb45 | 1121 | __uint64_t resblks; |
7d095257 CH |
1122 | uint quotamount = 0; |
1123 | uint quotaflags = 0; | |
0771fb45 ES |
1124 | int error = 0; |
1125 | ||
0771fb45 ES |
1126 | xfs_mount_common(mp, sbp); |
1127 | ||
ee1c0908 | 1128 | /* |
e6957ea4 ES |
1129 | * Check for a mismatched features2 values. Older kernels |
1130 | * read & wrote into the wrong sb offset for sb_features2 | |
1131 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1132 | * when sb_features2 was added, which made older superblock | |
1133 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1134 | * |
e6957ea4 ES |
1135 | * For backwards compatibility, we make both slots equal. |
1136 | * | |
1137 | * If we detect a mismatched field, we OR the set bits into the | |
1138 | * existing features2 field in case it has already been modified; we | |
1139 | * don't want to lose any features. We then update the bad location | |
1140 | * with the ORed value so that older kernels will see any features2 | |
1141 | * flags, and mark the two fields as needing updates once the | |
1142 | * transaction subsystem is online. | |
ee1c0908 | 1143 | */ |
e6957ea4 | 1144 | if (xfs_sb_has_mismatched_features2(sbp)) { |
ee1c0908 DC |
1145 | cmn_err(CE_WARN, |
1146 | "XFS: correcting sb_features alignment problem"); | |
1147 | sbp->sb_features2 |= sbp->sb_bad_features2; | |
e6957ea4 | 1148 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1149 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1150 | |
1151 | /* | |
1152 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1153 | * slot. | |
1154 | */ | |
7c12f296 TS |
1155 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1156 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1157 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1158 | } |
1159 | ||
1160 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1161 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1162 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1163 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1164 | |
7c12f296 TS |
1165 | /* update sb_versionnum for the clearing of the morebits */ |
1166 | if (!sbp->sb_features2) | |
7884bc86 | 1167 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1168 | } |
1169 | ||
0771fb45 ES |
1170 | /* |
1171 | * Check if sb_agblocks is aligned at stripe boundary | |
1172 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1173 | * allocator alignment is within an ag, therefore ag has | |
1174 | * to be aligned at stripe boundary. | |
1175 | */ | |
7884bc86 | 1176 | error = xfs_update_alignment(mp); |
0771fb45 | 1177 | if (error) |
f9057e3d | 1178 | goto out; |
0771fb45 ES |
1179 | |
1180 | xfs_alloc_compute_maxlevels(mp); | |
1181 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1182 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1183 | xfs_ialloc_compute_maxlevels(mp); | |
1184 | ||
1185 | xfs_set_maxicount(mp); | |
1186 | ||
1187 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
1188 | ||
27174203 CH |
1189 | error = xfs_uuid_mount(mp); |
1190 | if (error) | |
1191 | goto out; | |
1da177e4 | 1192 | |
0771fb45 ES |
1193 | /* |
1194 | * Set the minimum read and write sizes | |
1195 | */ | |
1196 | xfs_set_rw_sizes(mp); | |
1197 | ||
1198 | /* | |
1199 | * Set the inode cluster size. | |
1200 | * This may still be overridden by the file system | |
1201 | * block size if it is larger than the chosen cluster size. | |
1202 | */ | |
1203 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1204 | ||
1205 | /* | |
1206 | * Set inode alignment fields | |
1207 | */ | |
1208 | xfs_set_inoalignment(mp); | |
1209 | ||
1210 | /* | |
1211 | * Check that the data (and log if separate) are an ok size. | |
1212 | */ | |
4249023a | 1213 | error = xfs_check_sizes(mp); |
0771fb45 | 1214 | if (error) |
f9057e3d | 1215 | goto out_remove_uuid; |
0771fb45 | 1216 | |
1da177e4 LT |
1217 | /* |
1218 | * Initialize realtime fields in the mount structure | |
1219 | */ | |
0771fb45 ES |
1220 | error = xfs_rtmount_init(mp); |
1221 | if (error) { | |
1da177e4 | 1222 | cmn_err(CE_WARN, "XFS: RT mount failed"); |
f9057e3d | 1223 | goto out_remove_uuid; |
1da177e4 LT |
1224 | } |
1225 | ||
1da177e4 LT |
1226 | /* |
1227 | * Copies the low order bits of the timestamp and the randomly | |
1228 | * set "sequence" number out of a UUID. | |
1229 | */ | |
1230 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1231 | ||
1da177e4 LT |
1232 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1233 | ||
f6c2d1fa | 1234 | xfs_dir_mount(mp); |
1da177e4 LT |
1235 | |
1236 | /* | |
1237 | * Initialize the attribute manager's entries. | |
1238 | */ | |
1239 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1240 | ||
1241 | /* | |
1242 | * Initialize the precomputed transaction reservations values. | |
1243 | */ | |
1244 | xfs_trans_init(mp); | |
1245 | ||
1da177e4 LT |
1246 | /* |
1247 | * Allocate and initialize the per-ag data. | |
1248 | */ | |
1c1c6ebc | 1249 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 1250 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
1251 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
1252 | if (error) { | |
1253 | cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error); | |
f9057e3d | 1254 | goto out_remove_uuid; |
1c1c6ebc | 1255 | } |
1da177e4 | 1256 | |
f9057e3d CH |
1257 | if (!sbp->sb_logblocks) { |
1258 | cmn_err(CE_WARN, "XFS: no log defined"); | |
1259 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); | |
1260 | error = XFS_ERROR(EFSCORRUPTED); | |
1261 | goto out_free_perag; | |
1262 | } | |
1263 | ||
1da177e4 LT |
1264 | /* |
1265 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1266 | */ | |
f9057e3d CH |
1267 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1268 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1269 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1270 | if (error) { | |
1271 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
1272 | goto out_free_perag; | |
1da177e4 LT |
1273 | } |
1274 | ||
92821e2b DC |
1275 | /* |
1276 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1277 | * not. If it was, with the first phase of recovery has completed, we | |
1278 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1279 | * but they are recovered transactionally in the second recovery phase | |
1280 | * later. | |
1281 | * | |
1282 | * Hence we can safely re-initialise incore superblock counters from | |
1283 | * the per-ag data. These may not be correct if the filesystem was not | |
1284 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1285 | * doing this. | |
1286 | * | |
1287 | * If the filesystem was cleanly unmounted, then we can trust the | |
1288 | * values in the superblock to be correct and we don't need to do | |
1289 | * anything here. | |
1290 | * | |
1291 | * If we are currently making the filesystem, the initialisation will | |
1292 | * fail as the perag data is in an undefined state. | |
1293 | */ | |
92821e2b DC |
1294 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1295 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1296 | !mp->m_sb.sb_inprogress) { | |
1297 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d CH |
1298 | if (error) |
1299 | goto out_free_perag; | |
92821e2b | 1300 | } |
f9057e3d | 1301 | |
1da177e4 LT |
1302 | /* |
1303 | * Get and sanity-check the root inode. | |
1304 | * Save the pointer to it in the mount structure. | |
1305 | */ | |
7b6259e7 | 1306 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 LT |
1307 | if (error) { |
1308 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
f9057e3d | 1309 | goto out_log_dealloc; |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | ASSERT(rip != NULL); | |
1da177e4 LT |
1313 | |
1314 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1315 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1316 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1317 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1318 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1319 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1320 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1321 | mp); | |
1322 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1323 | goto out_rele_rip; |
1da177e4 LT |
1324 | } |
1325 | mp->m_rootip = rip; /* save it */ | |
1326 | ||
1327 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1328 | ||
1329 | /* | |
1330 | * Initialize realtime inode pointers in the mount structure | |
1331 | */ | |
0771fb45 ES |
1332 | error = xfs_rtmount_inodes(mp); |
1333 | if (error) { | |
1da177e4 LT |
1334 | /* |
1335 | * Free up the root inode. | |
1336 | */ | |
1337 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
f9057e3d | 1338 | goto out_rele_rip; |
1da177e4 LT |
1339 | } |
1340 | ||
1341 | /* | |
7884bc86 CH |
1342 | * If this is a read-only mount defer the superblock updates until |
1343 | * the next remount into writeable mode. Otherwise we would never | |
1344 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1345 | */ |
7884bc86 CH |
1346 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1347 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec DC |
1348 | if (error) { |
1349 | cmn_err(CE_WARN, "XFS: failed to write sb changes"); | |
b93b6e43 | 1350 | goto out_rtunmount; |
e5720eec DC |
1351 | } |
1352 | } | |
1da177e4 LT |
1353 | |
1354 | /* | |
1355 | * Initialise the XFS quota management subsystem for this mount | |
1356 | */ | |
7d095257 CH |
1357 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1358 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1359 | if (error) | |
1360 | goto out_rtunmount; | |
1361 | } else { | |
1362 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1363 | ||
1364 | /* | |
1365 | * If a file system had quotas running earlier, but decided to | |
1366 | * mount without -o uquota/pquota/gquota options, revoke the | |
1367 | * quotachecked license. | |
1368 | */ | |
1369 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
1370 | cmn_err(CE_NOTE, | |
1371 | "XFS: resetting qflags for filesystem %s", | |
1372 | mp->m_fsname); | |
1373 | ||
1374 | error = xfs_mount_reset_sbqflags(mp); | |
1375 | if (error) | |
1376 | return error; | |
1377 | } | |
1378 | } | |
1da177e4 LT |
1379 | |
1380 | /* | |
1381 | * Finish recovering the file system. This part needed to be | |
1382 | * delayed until after the root and real-time bitmap inodes | |
1383 | * were consistently read in. | |
1384 | */ | |
4249023a | 1385 | error = xfs_log_mount_finish(mp); |
1da177e4 LT |
1386 | if (error) { |
1387 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
b93b6e43 | 1388 | goto out_rtunmount; |
1da177e4 LT |
1389 | } |
1390 | ||
1391 | /* | |
1392 | * Complete the quota initialisation, post-log-replay component. | |
1393 | */ | |
7d095257 CH |
1394 | if (quotamount) { |
1395 | ASSERT(mp->m_qflags == 0); | |
1396 | mp->m_qflags = quotaflags; | |
1397 | ||
1398 | xfs_qm_mount_quotas(mp); | |
1399 | } | |
1400 | ||
84e1e99f DC |
1401 | /* |
1402 | * Now we are mounted, reserve a small amount of unused space for | |
1403 | * privileged transactions. This is needed so that transaction | |
1404 | * space required for critical operations can dip into this pool | |
1405 | * when at ENOSPC. This is needed for operations like create with | |
1406 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1407 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
1408 | * |
1409 | * This may drive us straight to ENOSPC on mount, but that implies | |
1410 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 1411 | */ |
d5db0f97 ES |
1412 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1413 | resblks = xfs_default_resblks(mp); | |
1414 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
1415 | if (error) | |
1416 | cmn_err(CE_WARN, "XFS: Unable to allocate reserve " | |
1417 | "blocks. Continuing without a reserve pool."); | |
1418 | } | |
84e1e99f | 1419 | |
1da177e4 LT |
1420 | return 0; |
1421 | ||
b93b6e43 CH |
1422 | out_rtunmount: |
1423 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1424 | out_rele_rip: |
43355099 | 1425 | IRELE(rip); |
f9057e3d | 1426 | out_log_dealloc: |
21b699c8 | 1427 | xfs_log_unmount(mp); |
f9057e3d | 1428 | out_free_perag: |
ff4f038c | 1429 | xfs_free_perag(mp); |
f9057e3d | 1430 | out_remove_uuid: |
27174203 | 1431 | xfs_uuid_unmount(mp); |
f9057e3d | 1432 | out: |
1da177e4 LT |
1433 | return error; |
1434 | } | |
1435 | ||
1436 | /* | |
1da177e4 LT |
1437 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1438 | * log and makes sure that incore structures are freed. | |
1439 | */ | |
41b5c2e7 CH |
1440 | void |
1441 | xfs_unmountfs( | |
1442 | struct xfs_mount *mp) | |
1da177e4 | 1443 | { |
41b5c2e7 CH |
1444 | __uint64_t resblks; |
1445 | int error; | |
1da177e4 | 1446 | |
7d095257 | 1447 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1448 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1449 | IRELE(mp->m_rootip); |
1450 | ||
641c56fb DC |
1451 | /* |
1452 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1453 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1454 | * the transaction is still sitting in a iclog. The stale inodes |
1455 | * on that buffer will have their flush locks held until the | |
1456 | * transaction hits the disk and the callbacks run. the inode | |
1457 | * flush takes the flush lock unconditionally and with nothing to | |
1458 | * push out the iclog we will never get that unlocked. hence we | |
1459 | * need to force the log first. | |
1460 | */ | |
a14a348b | 1461 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1462 | |
1463 | /* | |
1464 | * Do a delwri reclaim pass first so that as many dirty inodes are | |
1465 | * queued up for IO as possible. Then flush the buffers before making | |
1466 | * a synchronous path to catch all the remaining inodes are reclaimed. | |
1467 | * This makes the reclaim process as quick as possible by avoiding | |
1468 | * synchronous writeout and blocking on inodes already in the delwri | |
1469 | * state as much as possible. | |
1470 | */ | |
1471 | xfs_reclaim_inodes(mp, 0); | |
1472 | XFS_bflush(mp->m_ddev_targp); | |
1473 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
1da177e4 | 1474 | |
7d095257 | 1475 | xfs_qm_unmount(mp); |
a357a121 | 1476 | |
1da177e4 LT |
1477 | /* |
1478 | * Flush out the log synchronously so that we know for sure | |
1479 | * that nothing is pinned. This is important because bflush() | |
1480 | * will skip pinned buffers. | |
1481 | */ | |
a14a348b | 1482 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 LT |
1483 | |
1484 | xfs_binval(mp->m_ddev_targp); | |
1485 | if (mp->m_rtdev_targp) { | |
1486 | xfs_binval(mp->m_rtdev_targp); | |
1487 | } | |
1488 | ||
84e1e99f DC |
1489 | /* |
1490 | * Unreserve any blocks we have so that when we unmount we don't account | |
1491 | * the reserved free space as used. This is really only necessary for | |
1492 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1493 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1494 | * |
1495 | * We don't bother correcting this elsewhere for lazy superblock | |
1496 | * counting because on mount of an unclean filesystem we reconstruct the | |
1497 | * correct counter value and this is irrelevant. | |
1498 | * | |
1499 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1500 | * we only every apply deltas to the superblock and hence the incore | |
1501 | * value does not matter.... | |
1502 | */ | |
1503 | resblks = 0; | |
714082bc DC |
1504 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1505 | if (error) | |
1506 | cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. " | |
1507 | "Freespace may not be correct on next mount."); | |
1508 | ||
e5720eec DC |
1509 | error = xfs_log_sbcount(mp, 1); |
1510 | if (error) | |
1511 | cmn_err(CE_WARN, "XFS: Unable to update superblock counters. " | |
1512 | "Freespace may not be correct on next mount."); | |
1da177e4 | 1513 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1514 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
21b699c8 CH |
1515 | xfs_log_unmount_write(mp); |
1516 | xfs_log_unmount(mp); | |
27174203 | 1517 | xfs_uuid_unmount(mp); |
1da177e4 | 1518 | |
1550d0b0 | 1519 | #if defined(DEBUG) |
0ce4cfd4 | 1520 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1521 | #endif |
ff4f038c | 1522 | xfs_free_perag(mp); |
1da177e4 LT |
1523 | } |
1524 | ||
ba0f32d4 | 1525 | STATIC void |
1da177e4 LT |
1526 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1527 | { | |
1528 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1529 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1530 | if (mp->m_rtdev_targp) | |
1531 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1532 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1533 | } | |
1534 | ||
92821e2b DC |
1535 | int |
1536 | xfs_fs_writable(xfs_mount_t *mp) | |
1537 | { | |
b267ce99 | 1538 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1539 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1540 | } |
1541 | ||
1542 | /* | |
1543 | * xfs_log_sbcount | |
1544 | * | |
1545 | * Called either periodically to keep the on disk superblock values | |
1546 | * roughly up to date or from unmount to make sure the values are | |
1547 | * correct on a clean unmount. | |
1548 | * | |
1549 | * Note this code can be called during the process of freezing, so | |
1550 | * we may need to use the transaction allocator which does not not | |
1551 | * block when the transaction subsystem is in its frozen state. | |
1552 | */ | |
1553 | int | |
1554 | xfs_log_sbcount( | |
1555 | xfs_mount_t *mp, | |
1556 | uint sync) | |
1557 | { | |
1558 | xfs_trans_t *tp; | |
1559 | int error; | |
1560 | ||
1561 | if (!xfs_fs_writable(mp)) | |
1562 | return 0; | |
1563 | ||
d4d90b57 | 1564 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1565 | |
1566 | /* | |
1567 | * we don't need to do this if we are updating the superblock | |
1568 | * counters on every modification. | |
1569 | */ | |
1570 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1571 | return 0; | |
1572 | ||
80641dc6 | 1573 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
92821e2b DC |
1574 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1575 | XFS_DEFAULT_LOG_COUNT); | |
1576 | if (error) { | |
1577 | xfs_trans_cancel(tp, 0); | |
1578 | return error; | |
1579 | } | |
1580 | ||
1581 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1582 | if (sync) | |
1583 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1584 | error = xfs_trans_commit(tp, 0); |
1585 | return error; | |
92821e2b DC |
1586 | } |
1587 | ||
1da177e4 LT |
1588 | int |
1589 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1590 | { | |
1591 | xfs_buf_t *sbp; | |
1da177e4 LT |
1592 | int error = 0; |
1593 | ||
1594 | /* | |
1595 | * skip superblock write if fs is read-only, or | |
1596 | * if we are doing a forced umount. | |
1597 | */ | |
bd186aa9 | 1598 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1599 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1600 | |
92821e2b | 1601 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1602 | |
1da177e4 LT |
1603 | XFS_BUF_UNDONE(sbp); |
1604 | XFS_BUF_UNREAD(sbp); | |
1605 | XFS_BUF_UNDELAYWRITE(sbp); | |
1606 | XFS_BUF_WRITE(sbp); | |
1607 | XFS_BUF_UNASYNC(sbp); | |
1608 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1609 | xfsbdstrat(mp, sbp); | |
1da177e4 LT |
1610 | error = xfs_iowait(sbp); |
1611 | if (error) | |
1612 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1613 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
92821e2b | 1614 | xfs_buf_relse(sbp); |
1da177e4 | 1615 | } |
014c2544 | 1616 | return error; |
1da177e4 LT |
1617 | } |
1618 | ||
1619 | /* | |
1620 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1621 | * in-core superblock into the superblock buffer to be logged. | |
1622 | * It does not provide the higher level of locking that is | |
1623 | * needed to protect the in-core superblock from concurrent | |
1624 | * access. | |
1625 | */ | |
1626 | void | |
1627 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1628 | { | |
1629 | xfs_buf_t *bp; | |
1630 | int first; | |
1631 | int last; | |
1632 | xfs_mount_t *mp; | |
1da177e4 LT |
1633 | xfs_sb_field_t f; |
1634 | ||
1635 | ASSERT(fields); | |
1636 | if (!fields) | |
1637 | return; | |
1638 | mp = tp->t_mountp; | |
1639 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1640 | first = sizeof(xfs_sb_t); |
1641 | last = 0; | |
1642 | ||
1643 | /* translate/copy */ | |
1644 | ||
2bdf7cd0 | 1645 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1646 | |
1647 | /* find modified range */ | |
587aa0fe DC |
1648 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); |
1649 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1650 | last = xfs_sb_info[f + 1].offset - 1; | |
1da177e4 LT |
1651 | |
1652 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1653 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1654 | first = xfs_sb_info[f].offset; | |
1655 | ||
1da177e4 LT |
1656 | xfs_trans_log_buf(tp, bp, first, last); |
1657 | } | |
d210a28c | 1658 | |
d210a28c | 1659 | |
1da177e4 LT |
1660 | /* |
1661 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1662 | * a delta to a specified field in the in-core superblock. Simply | |
1663 | * switch on the field indicated and apply the delta to that field. | |
1664 | * Fields are not allowed to dip below zero, so if the delta would | |
1665 | * do this do not apply it and return EINVAL. | |
1666 | * | |
3685c2a1 | 1667 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1668 | */ |
d96f8f89 | 1669 | STATIC int |
20f4ebf2 DC |
1670 | xfs_mod_incore_sb_unlocked( |
1671 | xfs_mount_t *mp, | |
1672 | xfs_sb_field_t field, | |
1673 | int64_t delta, | |
1674 | int rsvd) | |
1da177e4 LT |
1675 | { |
1676 | int scounter; /* short counter for 32 bit fields */ | |
1677 | long long lcounter; /* long counter for 64 bit fields */ | |
1678 | long long res_used, rem; | |
1679 | ||
1680 | /* | |
1681 | * With the in-core superblock spin lock held, switch | |
1682 | * on the indicated field. Apply the delta to the | |
1683 | * proper field. If the fields value would dip below | |
1684 | * 0, then do not apply the delta and return EINVAL. | |
1685 | */ | |
1686 | switch (field) { | |
1687 | case XFS_SBS_ICOUNT: | |
1688 | lcounter = (long long)mp->m_sb.sb_icount; | |
1689 | lcounter += delta; | |
1690 | if (lcounter < 0) { | |
1691 | ASSERT(0); | |
014c2544 | 1692 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1693 | } |
1694 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1695 | return 0; |
1da177e4 LT |
1696 | case XFS_SBS_IFREE: |
1697 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1698 | lcounter += delta; | |
1699 | if (lcounter < 0) { | |
1700 | ASSERT(0); | |
014c2544 | 1701 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1702 | } |
1703 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1704 | return 0; |
1da177e4 | 1705 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1706 | lcounter = (long long) |
1707 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1708 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1709 | ||
1710 | if (delta > 0) { /* Putting blocks back */ | |
1711 | if (res_used > delta) { | |
1712 | mp->m_resblks_avail += delta; | |
1713 | } else { | |
1714 | rem = delta - res_used; | |
1715 | mp->m_resblks_avail = mp->m_resblks; | |
1716 | lcounter += rem; | |
1717 | } | |
1718 | } else { /* Taking blocks away */ | |
1da177e4 | 1719 | lcounter += delta; |
8babd8a2 DC |
1720 | if (lcounter >= 0) { |
1721 | mp->m_sb.sb_fdblocks = lcounter + | |
1722 | XFS_ALLOC_SET_ASIDE(mp); | |
1723 | return 0; | |
1724 | } | |
1da177e4 | 1725 | |
8babd8a2 DC |
1726 | /* |
1727 | * We are out of blocks, use any available reserved | |
1728 | * blocks if were allowed to. | |
1729 | */ | |
1730 | if (!rsvd) | |
1731 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1732 | |
8babd8a2 DC |
1733 | lcounter = (long long)mp->m_resblks_avail + delta; |
1734 | if (lcounter >= 0) { | |
1735 | mp->m_resblks_avail = lcounter; | |
1736 | return 0; | |
1da177e4 | 1737 | } |
8babd8a2 DC |
1738 | printk_once(KERN_WARNING |
1739 | "Filesystem \"%s\": reserve blocks depleted! " | |
1740 | "Consider increasing reserve pool size.", | |
1741 | mp->m_fsname); | |
1742 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1743 | } |
1744 | ||
4be536de | 1745 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1746 | return 0; |
1da177e4 LT |
1747 | case XFS_SBS_FREXTENTS: |
1748 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1749 | lcounter += delta; | |
1750 | if (lcounter < 0) { | |
014c2544 | 1751 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1752 | } |
1753 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1754 | return 0; |
1da177e4 LT |
1755 | case XFS_SBS_DBLOCKS: |
1756 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1757 | lcounter += delta; | |
1758 | if (lcounter < 0) { | |
1759 | ASSERT(0); | |
014c2544 | 1760 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1761 | } |
1762 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1763 | return 0; |
1da177e4 LT |
1764 | case XFS_SBS_AGCOUNT: |
1765 | scounter = mp->m_sb.sb_agcount; | |
1766 | scounter += delta; | |
1767 | if (scounter < 0) { | |
1768 | ASSERT(0); | |
014c2544 | 1769 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1770 | } |
1771 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1772 | return 0; |
1da177e4 LT |
1773 | case XFS_SBS_IMAX_PCT: |
1774 | scounter = mp->m_sb.sb_imax_pct; | |
1775 | scounter += delta; | |
1776 | if (scounter < 0) { | |
1777 | ASSERT(0); | |
014c2544 | 1778 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1779 | } |
1780 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1781 | return 0; |
1da177e4 LT |
1782 | case XFS_SBS_REXTSIZE: |
1783 | scounter = mp->m_sb.sb_rextsize; | |
1784 | scounter += delta; | |
1785 | if (scounter < 0) { | |
1786 | ASSERT(0); | |
014c2544 | 1787 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1788 | } |
1789 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1790 | return 0; |
1da177e4 LT |
1791 | case XFS_SBS_RBMBLOCKS: |
1792 | scounter = mp->m_sb.sb_rbmblocks; | |
1793 | scounter += delta; | |
1794 | if (scounter < 0) { | |
1795 | ASSERT(0); | |
014c2544 | 1796 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1797 | } |
1798 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1799 | return 0; |
1da177e4 LT |
1800 | case XFS_SBS_RBLOCKS: |
1801 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1802 | lcounter += delta; | |
1803 | if (lcounter < 0) { | |
1804 | ASSERT(0); | |
014c2544 | 1805 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1806 | } |
1807 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1808 | return 0; |
1da177e4 LT |
1809 | case XFS_SBS_REXTENTS: |
1810 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1811 | lcounter += delta; | |
1812 | if (lcounter < 0) { | |
1813 | ASSERT(0); | |
014c2544 | 1814 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1815 | } |
1816 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1817 | return 0; |
1da177e4 LT |
1818 | case XFS_SBS_REXTSLOG: |
1819 | scounter = mp->m_sb.sb_rextslog; | |
1820 | scounter += delta; | |
1821 | if (scounter < 0) { | |
1822 | ASSERT(0); | |
014c2544 | 1823 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1824 | } |
1825 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1826 | return 0; |
1da177e4 LT |
1827 | default: |
1828 | ASSERT(0); | |
014c2544 | 1829 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1830 | } |
1831 | } | |
1832 | ||
1833 | /* | |
1834 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1835 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1836 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1837 | * routine to do the work. |
1838 | */ | |
1839 | int | |
20f4ebf2 | 1840 | xfs_mod_incore_sb( |
96540c78 CH |
1841 | struct xfs_mount *mp, |
1842 | xfs_sb_field_t field, | |
1843 | int64_t delta, | |
1844 | int rsvd) | |
1da177e4 | 1845 | { |
96540c78 | 1846 | int status; |
1da177e4 | 1847 | |
8d280b98 | 1848 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1849 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1850 | #endif |
96540c78 CH |
1851 | spin_lock(&mp->m_sb_lock); |
1852 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1853 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1854 | |
014c2544 | 1855 | return status; |
1da177e4 LT |
1856 | } |
1857 | ||
1858 | /* | |
1859 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1860 | * in the in-core superblock structure at a time. This modification | |
1861 | * is protected by a lock internal to this module. The fields and | |
1862 | * changes to those fields are specified in the array of xfs_mod_sb | |
1863 | * structures passed in. | |
1864 | * | |
1865 | * Either all of the specified deltas will be applied or none of | |
1866 | * them will. If any modified field dips below 0, then all modifications | |
1867 | * will be backed out and EINVAL will be returned. | |
1868 | */ | |
1869 | int | |
1870 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1871 | { | |
1da177e4 LT |
1872 | int status=0; |
1873 | xfs_mod_sb_t *msbp; | |
1874 | ||
1875 | /* | |
1876 | * Loop through the array of mod structures and apply each | |
1877 | * individually. If any fail, then back out all those | |
1878 | * which have already been applied. Do all of this within | |
3685c2a1 | 1879 | * the scope of the m_sb_lock so that all of the changes will |
1da177e4 LT |
1880 | * be atomic. |
1881 | */ | |
3685c2a1 | 1882 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
1883 | msbp = &msb[0]; |
1884 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1885 | /* | |
1886 | * Apply the delta at index n. If it fails, break | |
1887 | * from the loop so we'll fall into the undo loop | |
1888 | * below. | |
1889 | */ | |
8d280b98 DC |
1890 | switch (msbp->msb_field) { |
1891 | #ifdef HAVE_PERCPU_SB | |
1892 | case XFS_SBS_ICOUNT: | |
1893 | case XFS_SBS_IFREE: | |
1894 | case XFS_SBS_FDBLOCKS: | |
61ba35de CH |
1895 | spin_unlock(&mp->m_sb_lock); |
1896 | status = xfs_icsb_modify_counters(mp, | |
1897 | msbp->msb_field, | |
1898 | msbp->msb_delta, rsvd); | |
1899 | spin_lock(&mp->m_sb_lock); | |
1900 | break; | |
8d280b98 DC |
1901 | #endif |
1902 | default: | |
1903 | status = xfs_mod_incore_sb_unlocked(mp, | |
1904 | msbp->msb_field, | |
1905 | msbp->msb_delta, rsvd); | |
1906 | break; | |
1907 | } | |
1908 | ||
1da177e4 LT |
1909 | if (status != 0) { |
1910 | break; | |
1911 | } | |
1912 | } | |
1913 | ||
1914 | /* | |
1915 | * If we didn't complete the loop above, then back out | |
1916 | * any changes made to the superblock. If you add code | |
1917 | * between the loop above and here, make sure that you | |
1918 | * preserve the value of status. Loop back until | |
1919 | * we step below the beginning of the array. Make sure | |
1920 | * we don't touch anything back there. | |
1921 | */ | |
1922 | if (status != 0) { | |
1923 | msbp--; | |
1924 | while (msbp >= msb) { | |
8d280b98 DC |
1925 | switch (msbp->msb_field) { |
1926 | #ifdef HAVE_PERCPU_SB | |
1927 | case XFS_SBS_ICOUNT: | |
1928 | case XFS_SBS_IFREE: | |
1929 | case XFS_SBS_FDBLOCKS: | |
61ba35de CH |
1930 | spin_unlock(&mp->m_sb_lock); |
1931 | status = xfs_icsb_modify_counters(mp, | |
1932 | msbp->msb_field, | |
1933 | -(msbp->msb_delta), | |
1934 | rsvd); | |
1935 | spin_lock(&mp->m_sb_lock); | |
1936 | break; | |
8d280b98 DC |
1937 | #endif |
1938 | default: | |
1939 | status = xfs_mod_incore_sb_unlocked(mp, | |
1940 | msbp->msb_field, | |
1941 | -(msbp->msb_delta), | |
1942 | rsvd); | |
1943 | break; | |
1944 | } | |
1da177e4 LT |
1945 | ASSERT(status == 0); |
1946 | msbp--; | |
1947 | } | |
1948 | } | |
3685c2a1 | 1949 | spin_unlock(&mp->m_sb_lock); |
014c2544 | 1950 | return status; |
1da177e4 LT |
1951 | } |
1952 | ||
1953 | /* | |
1954 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1955 | * The buffer is returned locked and read in from disk. | |
1956 | * The buffer should be released with a call to xfs_brelse(). | |
1957 | * | |
1958 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1959 | * the superblock buffer if it can be locked without sleeping. | |
1960 | * If it can't then we'll return NULL. | |
1961 | */ | |
1962 | xfs_buf_t * | |
1963 | xfs_getsb( | |
1964 | xfs_mount_t *mp, | |
1965 | int flags) | |
1966 | { | |
1967 | xfs_buf_t *bp; | |
1968 | ||
1969 | ASSERT(mp->m_sb_bp != NULL); | |
1970 | bp = mp->m_sb_bp; | |
0cadda1c | 1971 | if (flags & XBF_TRYLOCK) { |
1da177e4 LT |
1972 | if (!XFS_BUF_CPSEMA(bp)) { |
1973 | return NULL; | |
1974 | } | |
1975 | } else { | |
1976 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1977 | } | |
1978 | XFS_BUF_HOLD(bp); | |
1979 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1980 | return bp; |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | /* | |
1984 | * Used to free the superblock along various error paths. | |
1985 | */ | |
1986 | void | |
1987 | xfs_freesb( | |
26af6552 | 1988 | struct xfs_mount *mp) |
1da177e4 | 1989 | { |
26af6552 | 1990 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1991 | |
26af6552 | 1992 | xfs_buf_lock(bp); |
1da177e4 | 1993 | mp->m_sb_bp = NULL; |
26af6552 | 1994 | xfs_buf_relse(bp); |
1da177e4 LT |
1995 | } |
1996 | ||
1da177e4 LT |
1997 | /* |
1998 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1999 | * be altered by the mount options, as well as any potential sb_features2 |
2000 | * fixup. Only the first superblock is updated. | |
1da177e4 | 2001 | */ |
7884bc86 | 2002 | int |
ee1c0908 | 2003 | xfs_mount_log_sb( |
1da177e4 LT |
2004 | xfs_mount_t *mp, |
2005 | __int64_t fields) | |
2006 | { | |
2007 | xfs_trans_t *tp; | |
e5720eec | 2008 | int error; |
1da177e4 | 2009 | |
ee1c0908 | 2010 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
2011 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
2012 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
2013 | |
2014 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
2015 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
2016 | XFS_DEFAULT_LOG_COUNT); | |
2017 | if (error) { | |
1da177e4 | 2018 | xfs_trans_cancel(tp, 0); |
e5720eec | 2019 | return error; |
1da177e4 LT |
2020 | } |
2021 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
2022 | error = xfs_trans_commit(tp, 0); |
2023 | return error; | |
1da177e4 | 2024 | } |
8d280b98 | 2025 | |
dda35b8f CH |
2026 | /* |
2027 | * If the underlying (data/log/rt) device is readonly, there are some | |
2028 | * operations that cannot proceed. | |
2029 | */ | |
2030 | int | |
2031 | xfs_dev_is_read_only( | |
2032 | struct xfs_mount *mp, | |
2033 | char *message) | |
2034 | { | |
2035 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
2036 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
2037 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
2038 | cmn_err(CE_NOTE, | |
2039 | "XFS: %s required on read-only device.", message); | |
2040 | cmn_err(CE_NOTE, | |
2041 | "XFS: write access unavailable, cannot proceed."); | |
2042 | return EROFS; | |
2043 | } | |
2044 | return 0; | |
2045 | } | |
8d280b98 DC |
2046 | |
2047 | #ifdef HAVE_PERCPU_SB | |
2048 | /* | |
2049 | * Per-cpu incore superblock counters | |
2050 | * | |
2051 | * Simple concept, difficult implementation | |
2052 | * | |
2053 | * Basically, replace the incore superblock counters with a distributed per cpu | |
2054 | * counter for contended fields (e.g. free block count). | |
2055 | * | |
2056 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
2057 | * hence needs to be accurately read when we are running low on space. Hence | |
2058 | * there is a method to enable and disable the per-cpu counters based on how | |
2059 | * much "stuff" is available in them. | |
2060 | * | |
2061 | * Basically, a counter is enabled if there is enough free resource to justify | |
2062 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2063 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2064 | * re-distribute the available resources. | |
2065 | * | |
2066 | * If, once we redistributed the available resources, we still get a failure, | |
2067 | * we disable the per-cpu counter and go through the slow path. | |
2068 | * | |
2069 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2070 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2071 | * the global superblock. We do this after disabling the counter to prevent |
2072 | * more threads from queueing up on the counter. | |
2073 | * | |
2074 | * Essentially, this means that we still need a lock in the fast path to enable | |
2075 | * synchronisation between the global counters and the per-cpu counters. This | |
2076 | * is not a problem because the lock will be local to a CPU almost all the time | |
2077 | * and have little contention except when we get to ENOSPC conditions. | |
2078 | * | |
2079 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2080 | * path while we do things like enabling and disabling counters and | |
2081 | * synchronising the counters. | |
2082 | * | |
2083 | * Locking rules: | |
2084 | * | |
3685c2a1 | 2085 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2086 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2087 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2088 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2089 | * 5. modifying global counters requires holding m_sb_lock |
2090 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2091 | * and _none_ of the per-cpu locks. |
2092 | * | |
2093 | * Disabled counters are only ever re-enabled by a balance operation | |
2094 | * that results in more free resources per CPU than a given threshold. | |
2095 | * To ensure counters don't remain disabled, they are rebalanced when | |
2096 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2097 | * is present to prevent thrashing). | |
e8234a68 DC |
2098 | */ |
2099 | ||
5a67e4c5 | 2100 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2101 | /* |
2102 | * hot-plug CPU notifier support. | |
8d280b98 | 2103 | * |
5a67e4c5 CS |
2104 | * We need a notifier per filesystem as we need to be able to identify |
2105 | * the filesystem to balance the counters out. This is achieved by | |
2106 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2107 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2108 | */ |
e8234a68 DC |
2109 | STATIC int |
2110 | xfs_icsb_cpu_notify( | |
2111 | struct notifier_block *nfb, | |
2112 | unsigned long action, | |
2113 | void *hcpu) | |
2114 | { | |
2115 | xfs_icsb_cnts_t *cntp; | |
2116 | xfs_mount_t *mp; | |
e8234a68 DC |
2117 | |
2118 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2119 | cntp = (xfs_icsb_cnts_t *) | |
2120 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2121 | switch (action) { | |
2122 | case CPU_UP_PREPARE: | |
8bb78442 | 2123 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2124 | /* Easy Case - initialize the area and locks, and |
2125 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2126 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2127 | break; |
2128 | case CPU_ONLINE: | |
8bb78442 | 2129 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2130 | xfs_icsb_lock(mp); |
45af6c6d CH |
2131 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2132 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2133 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2134 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2135 | break; |
2136 | case CPU_DEAD: | |
8bb78442 | 2137 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2138 | /* Disable all the counters, then fold the dead cpu's |
2139 | * count into the total on the global superblock and | |
2140 | * re-enable the counters. */ | |
03135cf7 | 2141 | xfs_icsb_lock(mp); |
3685c2a1 | 2142 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2143 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2144 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2145 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2146 | ||
2147 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2148 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2149 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2150 | ||
01e1b69c | 2151 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2152 | |
45af6c6d CH |
2153 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2154 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2155 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2156 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2157 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2158 | break; |
2159 | } | |
2160 | ||
2161 | return NOTIFY_OK; | |
2162 | } | |
5a67e4c5 | 2163 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2164 | |
8d280b98 DC |
2165 | int |
2166 | xfs_icsb_init_counters( | |
2167 | xfs_mount_t *mp) | |
2168 | { | |
2169 | xfs_icsb_cnts_t *cntp; | |
2170 | int i; | |
2171 | ||
2172 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2173 | if (mp->m_sb_cnts == NULL) | |
2174 | return -ENOMEM; | |
2175 | ||
5a67e4c5 | 2176 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2177 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2178 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2179 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2180 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2181 | |
8d280b98 DC |
2182 | for_each_online_cpu(i) { |
2183 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2184 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2185 | } |
20b64285 DC |
2186 | |
2187 | mutex_init(&mp->m_icsb_mutex); | |
2188 | ||
8d280b98 DC |
2189 | /* |
2190 | * start with all counters disabled so that the | |
2191 | * initial balance kicks us off correctly | |
2192 | */ | |
2193 | mp->m_icsb_counters = -1; | |
2194 | return 0; | |
2195 | } | |
2196 | ||
5478eead LM |
2197 | void |
2198 | xfs_icsb_reinit_counters( | |
2199 | xfs_mount_t *mp) | |
2200 | { | |
2201 | xfs_icsb_lock(mp); | |
2202 | /* | |
2203 | * start with all counters disabled so that the | |
2204 | * initial balance kicks us off correctly | |
2205 | */ | |
2206 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2207 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2208 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2209 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2210 | xfs_icsb_unlock(mp); |
2211 | } | |
2212 | ||
c962fb79 | 2213 | void |
8d280b98 DC |
2214 | xfs_icsb_destroy_counters( |
2215 | xfs_mount_t *mp) | |
2216 | { | |
e8234a68 | 2217 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2218 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2219 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2220 | } |
03135cf7 | 2221 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2222 | } |
2223 | ||
b8f82a4a | 2224 | STATIC void |
01e1b69c DC |
2225 | xfs_icsb_lock_cntr( |
2226 | xfs_icsb_cnts_t *icsbp) | |
2227 | { | |
2228 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2229 | ndelay(1000); | |
2230 | } | |
2231 | } | |
2232 | ||
b8f82a4a | 2233 | STATIC void |
01e1b69c DC |
2234 | xfs_icsb_unlock_cntr( |
2235 | xfs_icsb_cnts_t *icsbp) | |
2236 | { | |
2237 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2238 | } | |
2239 | ||
8d280b98 | 2240 | |
b8f82a4a | 2241 | STATIC void |
8d280b98 DC |
2242 | xfs_icsb_lock_all_counters( |
2243 | xfs_mount_t *mp) | |
2244 | { | |
2245 | xfs_icsb_cnts_t *cntp; | |
2246 | int i; | |
2247 | ||
2248 | for_each_online_cpu(i) { | |
2249 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2250 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2251 | } |
2252 | } | |
2253 | ||
b8f82a4a | 2254 | STATIC void |
8d280b98 DC |
2255 | xfs_icsb_unlock_all_counters( |
2256 | xfs_mount_t *mp) | |
2257 | { | |
2258 | xfs_icsb_cnts_t *cntp; | |
2259 | int i; | |
2260 | ||
2261 | for_each_online_cpu(i) { | |
2262 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2263 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2264 | } |
2265 | } | |
2266 | ||
2267 | STATIC void | |
2268 | xfs_icsb_count( | |
2269 | xfs_mount_t *mp, | |
2270 | xfs_icsb_cnts_t *cnt, | |
2271 | int flags) | |
2272 | { | |
2273 | xfs_icsb_cnts_t *cntp; | |
2274 | int i; | |
2275 | ||
2276 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2277 | ||
2278 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2279 | xfs_icsb_lock_all_counters(mp); | |
2280 | ||
2281 | for_each_online_cpu(i) { | |
2282 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2283 | cnt->icsb_icount += cntp->icsb_icount; | |
2284 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2285 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2286 | } | |
2287 | ||
2288 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2289 | xfs_icsb_unlock_all_counters(mp); | |
2290 | } | |
2291 | ||
2292 | STATIC int | |
2293 | xfs_icsb_counter_disabled( | |
2294 | xfs_mount_t *mp, | |
2295 | xfs_sb_field_t field) | |
2296 | { | |
2297 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2298 | return test_bit(field, &mp->m_icsb_counters); | |
2299 | } | |
2300 | ||
36fbe6e6 | 2301 | STATIC void |
8d280b98 DC |
2302 | xfs_icsb_disable_counter( |
2303 | xfs_mount_t *mp, | |
2304 | xfs_sb_field_t field) | |
2305 | { | |
2306 | xfs_icsb_cnts_t cnt; | |
2307 | ||
2308 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2309 | ||
20b64285 DC |
2310 | /* |
2311 | * If we are already disabled, then there is nothing to do | |
2312 | * here. We check before locking all the counters to avoid | |
2313 | * the expensive lock operation when being called in the | |
2314 | * slow path and the counter is already disabled. This is | |
2315 | * safe because the only time we set or clear this state is under | |
2316 | * the m_icsb_mutex. | |
2317 | */ | |
2318 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2319 | return; |
20b64285 | 2320 | |
8d280b98 DC |
2321 | xfs_icsb_lock_all_counters(mp); |
2322 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2323 | /* drain back to superblock */ | |
2324 | ||
ce46193b | 2325 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2326 | switch(field) { |
2327 | case XFS_SBS_ICOUNT: | |
2328 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2329 | break; | |
2330 | case XFS_SBS_IFREE: | |
2331 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2332 | break; | |
2333 | case XFS_SBS_FDBLOCKS: | |
2334 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2335 | break; | |
2336 | default: | |
2337 | BUG(); | |
2338 | } | |
2339 | } | |
2340 | ||
2341 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2342 | } |
2343 | ||
2344 | STATIC void | |
2345 | xfs_icsb_enable_counter( | |
2346 | xfs_mount_t *mp, | |
2347 | xfs_sb_field_t field, | |
2348 | uint64_t count, | |
2349 | uint64_t resid) | |
2350 | { | |
2351 | xfs_icsb_cnts_t *cntp; | |
2352 | int i; | |
2353 | ||
2354 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2355 | ||
2356 | xfs_icsb_lock_all_counters(mp); | |
2357 | for_each_online_cpu(i) { | |
2358 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2359 | switch (field) { | |
2360 | case XFS_SBS_ICOUNT: | |
2361 | cntp->icsb_icount = count + resid; | |
2362 | break; | |
2363 | case XFS_SBS_IFREE: | |
2364 | cntp->icsb_ifree = count + resid; | |
2365 | break; | |
2366 | case XFS_SBS_FDBLOCKS: | |
2367 | cntp->icsb_fdblocks = count + resid; | |
2368 | break; | |
2369 | default: | |
2370 | BUG(); | |
2371 | break; | |
2372 | } | |
2373 | resid = 0; | |
2374 | } | |
2375 | clear_bit(field, &mp->m_icsb_counters); | |
2376 | xfs_icsb_unlock_all_counters(mp); | |
2377 | } | |
2378 | ||
dbcabad1 | 2379 | void |
d4d90b57 | 2380 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2381 | xfs_mount_t *mp, |
2382 | int flags) | |
2383 | { | |
2384 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2385 | |
8d280b98 DC |
2386 | xfs_icsb_count(mp, &cnt, flags); |
2387 | ||
8d280b98 DC |
2388 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2389 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2390 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2391 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2392 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2393 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2394 | } |
2395 | ||
2396 | /* | |
2397 | * Accurate update of per-cpu counters to incore superblock | |
2398 | */ | |
d4d90b57 | 2399 | void |
8d280b98 | 2400 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2401 | xfs_mount_t *mp, |
2402 | int flags) | |
8d280b98 | 2403 | { |
d4d90b57 CH |
2404 | spin_lock(&mp->m_sb_lock); |
2405 | xfs_icsb_sync_counters_locked(mp, flags); | |
2406 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2407 | } |
2408 | ||
2409 | /* | |
2410 | * Balance and enable/disable counters as necessary. | |
2411 | * | |
20b64285 DC |
2412 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2413 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2414 | * free blocks is something far enough zero that we aren't going thrash when we | |
2415 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2416 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2417 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2418 | * reenabled. | |
2419 | * | |
2420 | * Note that we can be called when counters are already disabled. | |
2421 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2422 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2423 | */ |
20b64285 DC |
2424 | |
2425 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2426 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2427 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2428 | STATIC void |
45af6c6d | 2429 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2430 | xfs_mount_t *mp, |
2431 | xfs_sb_field_t field, | |
20b64285 | 2432 | int min_per_cpu) |
8d280b98 | 2433 | { |
6fdf8ccc | 2434 | uint64_t count, resid; |
8d280b98 | 2435 | int weight = num_online_cpus(); |
20b64285 | 2436 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2437 | |
8d280b98 DC |
2438 | /* disable counter and sync counter */ |
2439 | xfs_icsb_disable_counter(mp, field); | |
2440 | ||
2441 | /* update counters - first CPU gets residual*/ | |
2442 | switch (field) { | |
2443 | case XFS_SBS_ICOUNT: | |
2444 | count = mp->m_sb.sb_icount; | |
2445 | resid = do_div(count, weight); | |
20b64285 | 2446 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2447 | return; |
8d280b98 DC |
2448 | break; |
2449 | case XFS_SBS_IFREE: | |
2450 | count = mp->m_sb.sb_ifree; | |
2451 | resid = do_div(count, weight); | |
20b64285 | 2452 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2453 | return; |
8d280b98 DC |
2454 | break; |
2455 | case XFS_SBS_FDBLOCKS: | |
2456 | count = mp->m_sb.sb_fdblocks; | |
2457 | resid = do_div(count, weight); | |
20b64285 | 2458 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2459 | return; |
8d280b98 DC |
2460 | break; |
2461 | default: | |
2462 | BUG(); | |
6fdf8ccc | 2463 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2464 | break; |
2465 | } | |
2466 | ||
2467 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2468 | } |
2469 | ||
2470 | STATIC void | |
2471 | xfs_icsb_balance_counter( | |
2472 | xfs_mount_t *mp, | |
2473 | xfs_sb_field_t fields, | |
2474 | int min_per_cpu) | |
2475 | { | |
2476 | spin_lock(&mp->m_sb_lock); | |
2477 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2478 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2479 | } |
2480 | ||
a8272ce0 | 2481 | STATIC int |
20b64285 | 2482 | xfs_icsb_modify_counters( |
8d280b98 DC |
2483 | xfs_mount_t *mp, |
2484 | xfs_sb_field_t field, | |
20f4ebf2 | 2485 | int64_t delta, |
20b64285 | 2486 | int rsvd) |
8d280b98 DC |
2487 | { |
2488 | xfs_icsb_cnts_t *icsbp; | |
2489 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2490 | int ret = 0; |
8d280b98 | 2491 | |
20b64285 | 2492 | might_sleep(); |
8d280b98 | 2493 | again: |
7a9e02d6 CL |
2494 | preempt_disable(); |
2495 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2496 | |
2497 | /* | |
2498 | * if the counter is disabled, go to slow path | |
2499 | */ | |
8d280b98 DC |
2500 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2501 | goto slow_path; | |
20b64285 DC |
2502 | xfs_icsb_lock_cntr(icsbp); |
2503 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2504 | xfs_icsb_unlock_cntr(icsbp); | |
2505 | goto slow_path; | |
2506 | } | |
8d280b98 DC |
2507 | |
2508 | switch (field) { | |
2509 | case XFS_SBS_ICOUNT: | |
2510 | lcounter = icsbp->icsb_icount; | |
2511 | lcounter += delta; | |
2512 | if (unlikely(lcounter < 0)) | |
20b64285 | 2513 | goto balance_counter; |
8d280b98 DC |
2514 | icsbp->icsb_icount = lcounter; |
2515 | break; | |
2516 | ||
2517 | case XFS_SBS_IFREE: | |
2518 | lcounter = icsbp->icsb_ifree; | |
2519 | lcounter += delta; | |
2520 | if (unlikely(lcounter < 0)) | |
20b64285 | 2521 | goto balance_counter; |
8d280b98 DC |
2522 | icsbp->icsb_ifree = lcounter; |
2523 | break; | |
2524 | ||
2525 | case XFS_SBS_FDBLOCKS: | |
2526 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2527 | ||
4be536de | 2528 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2529 | lcounter += delta; |
2530 | if (unlikely(lcounter < 0)) | |
20b64285 | 2531 | goto balance_counter; |
4be536de | 2532 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2533 | break; |
2534 | default: | |
2535 | BUG(); | |
2536 | break; | |
2537 | } | |
01e1b69c | 2538 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2539 | preempt_enable(); |
8d280b98 DC |
2540 | return 0; |
2541 | ||
8d280b98 | 2542 | slow_path: |
7a9e02d6 | 2543 | preempt_enable(); |
8d280b98 | 2544 | |
20b64285 DC |
2545 | /* |
2546 | * serialise with a mutex so we don't burn lots of cpu on | |
2547 | * the superblock lock. We still need to hold the superblock | |
2548 | * lock, however, when we modify the global structures. | |
2549 | */ | |
03135cf7 | 2550 | xfs_icsb_lock(mp); |
20b64285 DC |
2551 | |
2552 | /* | |
2553 | * Now running atomically. | |
2554 | * | |
2555 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2556 | * Drop the lock and try again in the fast path.... | |
2557 | */ | |
2558 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2559 | xfs_icsb_unlock(mp); |
8d280b98 | 2560 | goto again; |
8d280b98 DC |
2561 | } |
2562 | ||
20b64285 DC |
2563 | /* |
2564 | * The counter is currently disabled. Because we are | |
2565 | * running atomically here, we know a rebalance cannot | |
2566 | * be in progress. Hence we can go straight to operating | |
2567 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2568 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2569 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2570 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2571 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2572 | * directly on the global counters. | |
2573 | */ | |
3685c2a1 | 2574 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2575 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2576 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2577 | |
20b64285 DC |
2578 | /* |
2579 | * Now that we've modified the global superblock, we | |
2580 | * may be able to re-enable the distributed counters | |
2581 | * (e.g. lots of space just got freed). After that | |
2582 | * we are done. | |
2583 | */ | |
2584 | if (ret != ENOSPC) | |
45af6c6d | 2585 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2586 | xfs_icsb_unlock(mp); |
8d280b98 | 2587 | return ret; |
8d280b98 | 2588 | |
20b64285 DC |
2589 | balance_counter: |
2590 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2591 | preempt_enable(); |
8d280b98 | 2592 | |
20b64285 DC |
2593 | /* |
2594 | * We may have multiple threads here if multiple per-cpu | |
2595 | * counters run dry at the same time. This will mean we can | |
2596 | * do more balances than strictly necessary but it is not | |
2597 | * the common slowpath case. | |
2598 | */ | |
03135cf7 | 2599 | xfs_icsb_lock(mp); |
20b64285 DC |
2600 | |
2601 | /* | |
2602 | * running atomically. | |
2603 | * | |
2604 | * This will leave the counter in the correct state for future | |
2605 | * accesses. After the rebalance, we simply try again and our retry | |
2606 | * will either succeed through the fast path or slow path without | |
2607 | * another balance operation being required. | |
2608 | */ | |
45af6c6d | 2609 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2610 | xfs_icsb_unlock(mp); |
20b64285 | 2611 | goto again; |
8d280b98 | 2612 | } |
20b64285 | 2613 | |
8d280b98 | 2614 | #endif |