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)) { | |
0b932ccc | 136 | xfs_warn(mp, "Filesystem has nil UUID - can't mount"); |
27174203 CH |
137 | return XFS_ERROR(EINVAL); |
138 | } | |
139 | ||
140 | mutex_lock(&xfs_uuid_table_mutex); | |
141 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
142 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
143 | hole = i; | |
144 | continue; | |
145 | } | |
146 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
147 | goto out_duplicate; | |
148 | } | |
149 | ||
150 | if (hole < 0) { | |
151 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
152 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
153 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
154 | KM_SLEEP); | |
155 | hole = xfs_uuid_table_size++; | |
156 | } | |
157 | xfs_uuid_table[hole] = *uuid; | |
158 | mutex_unlock(&xfs_uuid_table_mutex); | |
159 | ||
160 | return 0; | |
161 | ||
162 | out_duplicate: | |
163 | mutex_unlock(&xfs_uuid_table_mutex); | |
0b932ccc | 164 | xfs_warn(mp, "Filesystem has duplicate UUID - can't mount"); |
27174203 CH |
165 | return XFS_ERROR(EINVAL); |
166 | } | |
167 | ||
168 | STATIC void | |
169 | xfs_uuid_unmount( | |
170 | struct xfs_mount *mp) | |
171 | { | |
172 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
173 | int i; | |
174 | ||
175 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
176 | return; | |
177 | ||
178 | mutex_lock(&xfs_uuid_table_mutex); | |
179 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
180 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
181 | continue; | |
182 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
183 | continue; | |
184 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
185 | break; | |
186 | } | |
187 | ASSERT(i < xfs_uuid_table_size); | |
188 | mutex_unlock(&xfs_uuid_table_mutex); | |
189 | } | |
190 | ||
191 | ||
0fa800fb DC |
192 | /* |
193 | * Reference counting access wrappers to the perag structures. | |
e176579e DC |
194 | * Because we never free per-ag structures, the only thing we |
195 | * have to protect against changes is the tree structure itself. | |
0fa800fb DC |
196 | */ |
197 | struct xfs_perag * | |
198 | xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno) | |
199 | { | |
200 | struct xfs_perag *pag; | |
201 | int ref = 0; | |
202 | ||
e176579e | 203 | rcu_read_lock(); |
0fa800fb DC |
204 | pag = radix_tree_lookup(&mp->m_perag_tree, agno); |
205 | if (pag) { | |
206 | ASSERT(atomic_read(&pag->pag_ref) >= 0); | |
0fa800fb DC |
207 | ref = atomic_inc_return(&pag->pag_ref); |
208 | } | |
e176579e | 209 | rcu_read_unlock(); |
0fa800fb DC |
210 | trace_xfs_perag_get(mp, agno, ref, _RET_IP_); |
211 | return pag; | |
212 | } | |
213 | ||
65d0f205 DC |
214 | /* |
215 | * search from @first to find the next perag with the given tag set. | |
216 | */ | |
217 | struct xfs_perag * | |
218 | xfs_perag_get_tag( | |
219 | struct xfs_mount *mp, | |
220 | xfs_agnumber_t first, | |
221 | int tag) | |
222 | { | |
223 | struct xfs_perag *pag; | |
224 | int found; | |
225 | int ref; | |
226 | ||
227 | rcu_read_lock(); | |
228 | found = radix_tree_gang_lookup_tag(&mp->m_perag_tree, | |
229 | (void **)&pag, first, 1, tag); | |
230 | if (found <= 0) { | |
231 | rcu_read_unlock(); | |
232 | return NULL; | |
233 | } | |
234 | ref = atomic_inc_return(&pag->pag_ref); | |
235 | rcu_read_unlock(); | |
236 | trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_); | |
237 | return pag; | |
238 | } | |
239 | ||
0fa800fb DC |
240 | void |
241 | xfs_perag_put(struct xfs_perag *pag) | |
242 | { | |
243 | int ref; | |
244 | ||
245 | ASSERT(atomic_read(&pag->pag_ref) > 0); | |
246 | ref = atomic_dec_return(&pag->pag_ref); | |
247 | trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_); | |
248 | } | |
249 | ||
e176579e DC |
250 | STATIC void |
251 | __xfs_free_perag( | |
252 | struct rcu_head *head) | |
253 | { | |
254 | struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head); | |
255 | ||
256 | ASSERT(atomic_read(&pag->pag_ref) == 0); | |
257 | kmem_free(pag); | |
258 | } | |
259 | ||
1da177e4 | 260 | /* |
e176579e | 261 | * Free up the per-ag resources associated with the mount structure. |
1da177e4 | 262 | */ |
c962fb79 | 263 | STATIC void |
ff4f038c | 264 | xfs_free_perag( |
745f6919 | 265 | xfs_mount_t *mp) |
1da177e4 | 266 | { |
1c1c6ebc DC |
267 | xfs_agnumber_t agno; |
268 | struct xfs_perag *pag; | |
269 | ||
270 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
271 | spin_lock(&mp->m_perag_lock); | |
272 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
273 | spin_unlock(&mp->m_perag_lock); | |
e176579e | 274 | ASSERT(pag); |
f83282a8 | 275 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
e176579e | 276 | call_rcu(&pag->rcu_head, __xfs_free_perag); |
1da177e4 | 277 | } |
1da177e4 LT |
278 | } |
279 | ||
4cc929ee NS |
280 | /* |
281 | * Check size of device based on the (data/realtime) block count. | |
282 | * Note: this check is used by the growfs code as well as mount. | |
283 | */ | |
284 | int | |
285 | xfs_sb_validate_fsb_count( | |
286 | xfs_sb_t *sbp, | |
287 | __uint64_t nblocks) | |
288 | { | |
289 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
290 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
291 | ||
292 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
293 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
657a4cff | 294 | return EFBIG; |
4cc929ee NS |
295 | #else /* Limited by UINT_MAX of sectors */ |
296 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
657a4cff | 297 | return EFBIG; |
4cc929ee NS |
298 | #endif |
299 | return 0; | |
300 | } | |
1da177e4 LT |
301 | |
302 | /* | |
303 | * Check the validity of the SB found. | |
304 | */ | |
305 | STATIC int | |
306 | xfs_mount_validate_sb( | |
307 | xfs_mount_t *mp, | |
764d1f89 NS |
308 | xfs_sb_t *sbp, |
309 | int flags) | |
1da177e4 | 310 | { |
af34e09d DC |
311 | int loud = !(flags & XFS_MFSI_QUIET); |
312 | ||
1da177e4 LT |
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) { | |
af34e09d DC |
321 | if (loud) |
322 | xfs_warn(mp, "bad magic number"); | |
1da177e4 LT |
323 | return XFS_ERROR(EWRONGFS); |
324 | } | |
325 | ||
62118709 | 326 | if (!xfs_sb_good_version(sbp)) { |
af34e09d DC |
327 | if (loud) |
328 | xfs_warn(mp, "bad version"); | |
1da177e4 LT |
329 | return XFS_ERROR(EWRONGFS); |
330 | } | |
331 | ||
332 | if (unlikely( | |
333 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
af34e09d DC |
334 | if (loud) |
335 | xfs_warn(mp, | |
336 | "filesystem is marked as having an external log; " | |
337 | "specify logdev on the mount command line."); | |
764d1f89 | 338 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
339 | } |
340 | ||
341 | if (unlikely( | |
342 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
af34e09d DC |
343 | if (loud) |
344 | xfs_warn(mp, | |
345 | "filesystem is marked as having an internal log; " | |
346 | "do not specify logdev on the mount command line."); | |
764d1f89 | 347 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
348 | } |
349 | ||
350 | /* | |
351 | * More sanity checking. These were stolen directly from | |
352 | * xfs_repair. | |
353 | */ | |
354 | if (unlikely( | |
355 | sbp->sb_agcount <= 0 || | |
356 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
357 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
358 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
359 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 360 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
361 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
362 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
363 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
364 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 365 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
366 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
367 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
368 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
369 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 370 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 371 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
372 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
373 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 374 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
af34e09d DC |
375 | if (loud) |
376 | xfs_warn(mp, "SB sanity check 1 failed"); | |
1da177e4 LT |
377 | return XFS_ERROR(EFSCORRUPTED); |
378 | } | |
379 | ||
380 | /* | |
381 | * Sanity check AG count, size fields against data size field | |
382 | */ | |
383 | if (unlikely( | |
384 | sbp->sb_dblocks == 0 || | |
385 | sbp->sb_dblocks > | |
386 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
387 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
388 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
af34e09d DC |
389 | if (loud) |
390 | xfs_warn(mp, "SB sanity check 2 failed"); | |
1da177e4 LT |
391 | return XFS_ERROR(EFSCORRUPTED); |
392 | } | |
393 | ||
2edbddd5 LM |
394 | /* |
395 | * Until this is fixed only page-sized or smaller data blocks work. | |
396 | */ | |
397 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
af34e09d DC |
398 | if (loud) { |
399 | xfs_warn(mp, | |
400 | "File system with blocksize %d bytes. " | |
401 | "Only pagesize (%ld) or less will currently work.", | |
402 | sbp->sb_blocksize, PAGE_SIZE); | |
403 | } | |
2edbddd5 LM |
404 | return XFS_ERROR(ENOSYS); |
405 | } | |
406 | ||
1a5902c5 CH |
407 | /* |
408 | * Currently only very few inode sizes are supported. | |
409 | */ | |
410 | switch (sbp->sb_inodesize) { | |
411 | case 256: | |
412 | case 512: | |
413 | case 1024: | |
414 | case 2048: | |
415 | break; | |
416 | default: | |
af34e09d DC |
417 | if (loud) |
418 | xfs_warn(mp, "inode size of %d bytes not supported", | |
419 | sbp->sb_inodesize); | |
1a5902c5 CH |
420 | return XFS_ERROR(ENOSYS); |
421 | } | |
422 | ||
4cc929ee NS |
423 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
424 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
af34e09d DC |
425 | if (loud) |
426 | xfs_warn(mp, | |
427 | "file system too large to be mounted on this system."); | |
657a4cff | 428 | return XFS_ERROR(EFBIG); |
1da177e4 LT |
429 | } |
430 | ||
431 | if (unlikely(sbp->sb_inprogress)) { | |
af34e09d DC |
432 | if (loud) |
433 | xfs_warn(mp, "file system busy"); | |
1da177e4 LT |
434 | return XFS_ERROR(EFSCORRUPTED); |
435 | } | |
436 | ||
de20614b NS |
437 | /* |
438 | * Version 1 directory format has never worked on Linux. | |
439 | */ | |
62118709 | 440 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
af34e09d DC |
441 | if (loud) |
442 | xfs_warn(mp, | |
443 | "file system using version 1 directory format"); | |
de20614b NS |
444 | return XFS_ERROR(ENOSYS); |
445 | } | |
446 | ||
1da177e4 LT |
447 | return 0; |
448 | } | |
449 | ||
1c1c6ebc | 450 | int |
c11e2c36 | 451 | xfs_initialize_perag( |
c11e2c36 | 452 | xfs_mount_t *mp, |
1c1c6ebc DC |
453 | xfs_agnumber_t agcount, |
454 | xfs_agnumber_t *maxagi) | |
1da177e4 LT |
455 | { |
456 | xfs_agnumber_t index, max_metadata; | |
8b26c582 | 457 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
458 | xfs_perag_t *pag; |
459 | xfs_agino_t agino; | |
460 | xfs_ino_t ino; | |
461 | xfs_sb_t *sbp = &mp->m_sb; | |
8b26c582 | 462 | int error = -ENOMEM; |
1da177e4 | 463 | |
1c1c6ebc DC |
464 | /* |
465 | * Walk the current per-ag tree so we don't try to initialise AGs | |
466 | * that already exist (growfs case). Allocate and insert all the | |
467 | * AGs we don't find ready for initialisation. | |
468 | */ | |
469 | for (index = 0; index < agcount; index++) { | |
470 | pag = xfs_perag_get(mp, index); | |
471 | if (pag) { | |
472 | xfs_perag_put(pag); | |
473 | continue; | |
474 | } | |
8b26c582 DC |
475 | if (!first_initialised) |
476 | first_initialised = index; | |
fb3b504a | 477 | |
1c1c6ebc DC |
478 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
479 | if (!pag) | |
8b26c582 | 480 | goto out_unwind; |
fb3b504a CH |
481 | pag->pag_agno = index; |
482 | pag->pag_mount = mp; | |
1a427ab0 | 483 | spin_lock_init(&pag->pag_ici_lock); |
69b491c2 | 484 | mutex_init(&pag->pag_ici_reclaim_lock); |
fb3b504a | 485 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); |
74f75a0c DC |
486 | spin_lock_init(&pag->pag_buf_lock); |
487 | pag->pag_buf_tree = RB_ROOT; | |
fb3b504a | 488 | |
1c1c6ebc | 489 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 490 | goto out_unwind; |
fb3b504a | 491 | |
1c1c6ebc DC |
492 | spin_lock(&mp->m_perag_lock); |
493 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
494 | BUG(); | |
495 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
496 | radix_tree_preload_end(); |
497 | error = -EEXIST; | |
498 | goto out_unwind; | |
1c1c6ebc DC |
499 | } |
500 | spin_unlock(&mp->m_perag_lock); | |
501 | radix_tree_preload_end(); | |
502 | } | |
503 | ||
fb3b504a CH |
504 | /* |
505 | * If we mount with the inode64 option, or no inode overflows | |
506 | * the legacy 32-bit address space clear the inode32 option. | |
1da177e4 | 507 | */ |
fb3b504a CH |
508 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); |
509 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
510 | ||
511 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32) | |
1da177e4 | 512 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
fb3b504a | 513 | else |
1da177e4 | 514 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; |
1da177e4 | 515 | |
1da177e4 | 516 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { |
fb3b504a CH |
517 | /* |
518 | * Calculate how much should be reserved for inodes to meet | |
519 | * the max inode percentage. | |
1da177e4 LT |
520 | */ |
521 | if (mp->m_maxicount) { | |
522 | __uint64_t icount; | |
523 | ||
524 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
525 | do_div(icount, 100); | |
526 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 527 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
528 | max_metadata = icount; |
529 | } else { | |
530 | max_metadata = agcount; | |
531 | } | |
fb3b504a | 532 | |
1da177e4 LT |
533 | for (index = 0; index < agcount; index++) { |
534 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
fb3b504a | 535 | if (ino > XFS_MAXINUMBER_32) { |
1da177e4 LT |
536 | index++; |
537 | break; | |
538 | } | |
539 | ||
44b56e0a | 540 | pag = xfs_perag_get(mp, index); |
1da177e4 LT |
541 | pag->pagi_inodeok = 1; |
542 | if (index < max_metadata) | |
543 | pag->pagf_metadata = 1; | |
44b56e0a | 544 | xfs_perag_put(pag); |
1da177e4 LT |
545 | } |
546 | } else { | |
1da177e4 | 547 | for (index = 0; index < agcount; index++) { |
44b56e0a | 548 | pag = xfs_perag_get(mp, index); |
1da177e4 | 549 | pag->pagi_inodeok = 1; |
44b56e0a | 550 | xfs_perag_put(pag); |
1da177e4 LT |
551 | } |
552 | } | |
fb3b504a | 553 | |
1c1c6ebc DC |
554 | if (maxagi) |
555 | *maxagi = index; | |
556 | return 0; | |
8b26c582 DC |
557 | |
558 | out_unwind: | |
559 | kmem_free(pag); | |
560 | for (; index > first_initialised; index--) { | |
561 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
562 | kmem_free(pag); | |
563 | } | |
564 | return error; | |
1da177e4 LT |
565 | } |
566 | ||
2bdf7cd0 CH |
567 | void |
568 | xfs_sb_from_disk( | |
569 | xfs_sb_t *to, | |
570 | xfs_dsb_t *from) | |
571 | { | |
572 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
573 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
574 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
575 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
576 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
577 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
578 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
579 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
580 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
581 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
582 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
583 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
584 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
585 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
586 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
587 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
588 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
589 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
590 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
591 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
592 | to->sb_blocklog = from->sb_blocklog; | |
593 | to->sb_sectlog = from->sb_sectlog; | |
594 | to->sb_inodelog = from->sb_inodelog; | |
595 | to->sb_inopblog = from->sb_inopblog; | |
596 | to->sb_agblklog = from->sb_agblklog; | |
597 | to->sb_rextslog = from->sb_rextslog; | |
598 | to->sb_inprogress = from->sb_inprogress; | |
599 | to->sb_imax_pct = from->sb_imax_pct; | |
600 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
601 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
602 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
603 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
604 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
605 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
606 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
607 | to->sb_flags = from->sb_flags; | |
608 | to->sb_shared_vn = from->sb_shared_vn; | |
609 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
610 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
611 | to->sb_width = be32_to_cpu(from->sb_width); | |
612 | to->sb_dirblklog = from->sb_dirblklog; | |
613 | to->sb_logsectlog = from->sb_logsectlog; | |
614 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
615 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
616 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 617 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
618 | } |
619 | ||
1da177e4 | 620 | /* |
2bdf7cd0 | 621 | * Copy in core superblock to ondisk one. |
1da177e4 | 622 | * |
2bdf7cd0 | 623 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
624 | */ |
625 | void | |
2bdf7cd0 CH |
626 | xfs_sb_to_disk( |
627 | xfs_dsb_t *to, | |
628 | xfs_sb_t *from, | |
1da177e4 LT |
629 | __int64_t fields) |
630 | { | |
2bdf7cd0 CH |
631 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
632 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
633 | xfs_sb_field_t f; |
634 | int first; | |
635 | int size; | |
636 | ||
1da177e4 | 637 | ASSERT(fields); |
1da177e4 LT |
638 | if (!fields) |
639 | return; | |
640 | ||
1da177e4 LT |
641 | while (fields) { |
642 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
643 | first = xfs_sb_info[f].offset; | |
644 | size = xfs_sb_info[f + 1].offset - first; | |
645 | ||
646 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
647 | ||
648 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 649 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
650 | } else { |
651 | switch (size) { | |
652 | case 2: | |
2bdf7cd0 CH |
653 | *(__be16 *)(to_ptr + first) = |
654 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
655 | break; |
656 | case 4: | |
2bdf7cd0 CH |
657 | *(__be32 *)(to_ptr + first) = |
658 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
659 | break; |
660 | case 8: | |
2bdf7cd0 CH |
661 | *(__be64 *)(to_ptr + first) = |
662 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
663 | break; |
664 | default: | |
665 | ASSERT(0); | |
666 | } | |
667 | } | |
668 | ||
669 | fields &= ~(1LL << f); | |
670 | } | |
671 | } | |
672 | ||
673 | /* | |
674 | * xfs_readsb | |
675 | * | |
676 | * Does the initial read of the superblock. | |
677 | */ | |
678 | int | |
764d1f89 | 679 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
680 | { |
681 | unsigned int sector_size; | |
1da177e4 | 682 | xfs_buf_t *bp; |
1da177e4 | 683 | int error; |
af34e09d | 684 | int loud = !(flags & XFS_MFSI_QUIET); |
1da177e4 LT |
685 | |
686 | ASSERT(mp->m_sb_bp == NULL); | |
687 | ASSERT(mp->m_ddev_targp != NULL); | |
688 | ||
689 | /* | |
690 | * Allocate a (locked) buffer to hold the superblock. | |
691 | * This will be kept around at all times to optimize | |
692 | * access to the superblock. | |
693 | */ | |
694 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
26af6552 DC |
695 | |
696 | reread: | |
697 | bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp, | |
698 | XFS_SB_DADDR, sector_size, 0); | |
699 | if (!bp) { | |
af34e09d DC |
700 | if (loud) |
701 | xfs_warn(mp, "SB buffer read failed"); | |
26af6552 | 702 | return EIO; |
1da177e4 | 703 | } |
1da177e4 LT |
704 | |
705 | /* | |
706 | * Initialize the mount structure from the superblock. | |
707 | * But first do some basic consistency checking. | |
708 | */ | |
2bdf7cd0 | 709 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
764d1f89 | 710 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 711 | if (error) { |
af34e09d DC |
712 | if (loud) |
713 | xfs_warn(mp, "SB validate failed"); | |
26af6552 | 714 | goto release_buf; |
1da177e4 LT |
715 | } |
716 | ||
717 | /* | |
718 | * We must be able to do sector-sized and sector-aligned IO. | |
719 | */ | |
720 | if (sector_size > mp->m_sb.sb_sectsize) { | |
af34e09d DC |
721 | if (loud) |
722 | xfs_warn(mp, "device supports %u byte sectors (not %u)", | |
723 | sector_size, mp->m_sb.sb_sectsize); | |
1da177e4 | 724 | error = ENOSYS; |
26af6552 | 725 | goto release_buf; |
1da177e4 LT |
726 | } |
727 | ||
728 | /* | |
729 | * If device sector size is smaller than the superblock size, | |
730 | * re-read the superblock so the buffer is correctly sized. | |
731 | */ | |
732 | if (sector_size < mp->m_sb.sb_sectsize) { | |
1da177e4 LT |
733 | xfs_buf_relse(bp); |
734 | sector_size = mp->m_sb.sb_sectsize; | |
26af6552 | 735 | goto reread; |
1da177e4 LT |
736 | } |
737 | ||
5478eead LM |
738 | /* Initialize per-cpu counters */ |
739 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 740 | |
1da177e4 | 741 | mp->m_sb_bp = bp; |
26af6552 | 742 | xfs_buf_unlock(bp); |
1da177e4 LT |
743 | return 0; |
744 | ||
26af6552 DC |
745 | release_buf: |
746 | xfs_buf_relse(bp); | |
1da177e4 LT |
747 | return error; |
748 | } | |
749 | ||
750 | ||
751 | /* | |
752 | * xfs_mount_common | |
753 | * | |
754 | * Mount initialization code establishing various mount | |
755 | * fields from the superblock associated with the given | |
756 | * mount structure | |
757 | */ | |
ba0f32d4 | 758 | STATIC void |
1da177e4 LT |
759 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
760 | { | |
1da177e4 | 761 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 762 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
763 | mp->m_maxagi = mp->m_sb.sb_agcount; |
764 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
765 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
766 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
767 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
768 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
769 | mp->m_blockmask = sbp->sb_blocksize - 1; |
770 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
771 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 772 | |
60197e8d CH |
773 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
774 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
775 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
776 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
777 | ||
778 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
779 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
780 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
781 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
782 | ||
783 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
784 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
785 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
786 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
787 | |
788 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
789 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
790 | sbp->sb_inopblock); | |
791 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
792 | } | |
92821e2b DC |
793 | |
794 | /* | |
795 | * xfs_initialize_perag_data | |
796 | * | |
797 | * Read in each per-ag structure so we can count up the number of | |
798 | * allocated inodes, free inodes and used filesystem blocks as this | |
799 | * information is no longer persistent in the superblock. Once we have | |
800 | * this information, write it into the in-core superblock structure. | |
801 | */ | |
802 | STATIC int | |
803 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
804 | { | |
805 | xfs_agnumber_t index; | |
806 | xfs_perag_t *pag; | |
807 | xfs_sb_t *sbp = &mp->m_sb; | |
808 | uint64_t ifree = 0; | |
809 | uint64_t ialloc = 0; | |
810 | uint64_t bfree = 0; | |
811 | uint64_t bfreelst = 0; | |
812 | uint64_t btree = 0; | |
813 | int error; | |
92821e2b DC |
814 | |
815 | for (index = 0; index < agcount; index++) { | |
816 | /* | |
817 | * read the agf, then the agi. This gets us | |
9da096fd | 818 | * all the information we need and populates the |
92821e2b DC |
819 | * per-ag structures for us. |
820 | */ | |
821 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
822 | if (error) | |
823 | return error; | |
824 | ||
825 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
826 | if (error) | |
827 | return error; | |
44b56e0a | 828 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
829 | ifree += pag->pagi_freecount; |
830 | ialloc += pag->pagi_count; | |
831 | bfree += pag->pagf_freeblks; | |
832 | bfreelst += pag->pagf_flcount; | |
833 | btree += pag->pagf_btreeblks; | |
44b56e0a | 834 | xfs_perag_put(pag); |
92821e2b DC |
835 | } |
836 | /* | |
837 | * Overwrite incore superblock counters with just-read data | |
838 | */ | |
3685c2a1 | 839 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
840 | sbp->sb_ifree = ifree; |
841 | sbp->sb_icount = ialloc; | |
842 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 843 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
844 | |
845 | /* Fixup the per-cpu counters as well. */ | |
846 | xfs_icsb_reinit_counters(mp); | |
847 | ||
848 | return 0; | |
849 | } | |
850 | ||
1da177e4 | 851 | /* |
0771fb45 | 852 | * Update alignment values based on mount options and sb values |
1da177e4 | 853 | */ |
0771fb45 | 854 | STATIC int |
7884bc86 | 855 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 856 | { |
1da177e4 | 857 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 858 | |
4249023a | 859 | if (mp->m_dalign) { |
1da177e4 LT |
860 | /* |
861 | * If stripe unit and stripe width are not multiples | |
862 | * of the fs blocksize turn off alignment. | |
863 | */ | |
864 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
865 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
866 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0b932ccc | 867 | xfs_warn(mp, "alignment check 1 failed"); |
0771fb45 | 868 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
869 | } |
870 | mp->m_dalign = mp->m_swidth = 0; | |
871 | } else { | |
872 | /* | |
873 | * Convert the stripe unit and width to FSBs. | |
874 | */ | |
875 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
876 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
877 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 878 | return XFS_ERROR(EINVAL); |
1da177e4 | 879 | } |
53487786 DC |
880 | xfs_warn(mp, |
881 | "stripe alignment turned off: sunit(%d)/swidth(%d) " | |
882 | "incompatible with agsize(%d)", | |
1da177e4 LT |
883 | mp->m_dalign, mp->m_swidth, |
884 | sbp->sb_agblocks); | |
885 | ||
886 | mp->m_dalign = 0; | |
887 | mp->m_swidth = 0; | |
888 | } else if (mp->m_dalign) { | |
889 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
890 | } else { | |
891 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
53487786 DC |
892 | xfs_warn(mp, |
893 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
894 | mp->m_dalign, | |
1da177e4 | 895 | mp->m_blockmask +1); |
0771fb45 | 896 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
897 | } |
898 | mp->m_swidth = 0; | |
899 | } | |
900 | } | |
901 | ||
902 | /* | |
903 | * Update superblock with new values | |
904 | * and log changes | |
905 | */ | |
62118709 | 906 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
907 | if (sbp->sb_unit != mp->m_dalign) { |
908 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 909 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
910 | } |
911 | if (sbp->sb_width != mp->m_swidth) { | |
912 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 913 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
914 | } |
915 | } | |
916 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 917 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
918 | mp->m_dalign = sbp->sb_unit; |
919 | mp->m_swidth = sbp->sb_width; | |
920 | } | |
921 | ||
0771fb45 ES |
922 | return 0; |
923 | } | |
1da177e4 | 924 | |
0771fb45 ES |
925 | /* |
926 | * Set the maximum inode count for this filesystem | |
927 | */ | |
928 | STATIC void | |
929 | xfs_set_maxicount(xfs_mount_t *mp) | |
930 | { | |
931 | xfs_sb_t *sbp = &(mp->m_sb); | |
932 | __uint64_t icount; | |
1da177e4 | 933 | |
0771fb45 ES |
934 | if (sbp->sb_imax_pct) { |
935 | /* | |
936 | * Make sure the maximum inode count is a multiple | |
937 | * of the units we allocate inodes in. | |
1da177e4 | 938 | */ |
1da177e4 LT |
939 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
940 | do_div(icount, 100); | |
941 | do_div(icount, mp->m_ialloc_blks); | |
942 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
943 | sbp->sb_inopblog; | |
0771fb45 | 944 | } else { |
1da177e4 | 945 | mp->m_maxicount = 0; |
1da177e4 | 946 | } |
0771fb45 ES |
947 | } |
948 | ||
949 | /* | |
950 | * Set the default minimum read and write sizes unless | |
951 | * already specified in a mount option. | |
952 | * We use smaller I/O sizes when the file system | |
953 | * is being used for NFS service (wsync mount option). | |
954 | */ | |
955 | STATIC void | |
956 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
957 | { | |
958 | xfs_sb_t *sbp = &(mp->m_sb); | |
959 | int readio_log, writeio_log; | |
1da177e4 | 960 | |
1da177e4 LT |
961 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
962 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
963 | readio_log = XFS_WSYNC_READIO_LOG; | |
964 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
965 | } else { | |
966 | readio_log = XFS_READIO_LOG_LARGE; | |
967 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
968 | } | |
969 | } else { | |
970 | readio_log = mp->m_readio_log; | |
971 | writeio_log = mp->m_writeio_log; | |
972 | } | |
973 | ||
1da177e4 LT |
974 | if (sbp->sb_blocklog > readio_log) { |
975 | mp->m_readio_log = sbp->sb_blocklog; | |
976 | } else { | |
977 | mp->m_readio_log = readio_log; | |
978 | } | |
979 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
980 | if (sbp->sb_blocklog > writeio_log) { | |
981 | mp->m_writeio_log = sbp->sb_blocklog; | |
982 | } else { | |
983 | mp->m_writeio_log = writeio_log; | |
984 | } | |
985 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 986 | } |
1da177e4 | 987 | |
055388a3 DC |
988 | /* |
989 | * precalculate the low space thresholds for dynamic speculative preallocation. | |
990 | */ | |
991 | void | |
992 | xfs_set_low_space_thresholds( | |
993 | struct xfs_mount *mp) | |
994 | { | |
995 | int i; | |
996 | ||
997 | for (i = 0; i < XFS_LOWSP_MAX; i++) { | |
998 | __uint64_t space = mp->m_sb.sb_dblocks; | |
999 | ||
1000 | do_div(space, 100); | |
1001 | mp->m_low_space[i] = space * (i + 1); | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | ||
0771fb45 ES |
1006 | /* |
1007 | * Set whether we're using inode alignment. | |
1008 | */ | |
1009 | STATIC void | |
1010 | xfs_set_inoalignment(xfs_mount_t *mp) | |
1011 | { | |
62118709 | 1012 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
1013 | mp->m_sb.sb_inoalignmt >= |
1014 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
1015 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
1016 | else | |
1017 | mp->m_inoalign_mask = 0; | |
1018 | /* | |
1019 | * If we are using stripe alignment, check whether | |
1020 | * the stripe unit is a multiple of the inode alignment | |
1021 | */ | |
1022 | if (mp->m_dalign && mp->m_inoalign_mask && | |
1023 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
1024 | mp->m_sinoalign = mp->m_dalign; | |
1025 | else | |
1026 | mp->m_sinoalign = 0; | |
0771fb45 ES |
1027 | } |
1028 | ||
1029 | /* | |
1030 | * Check that the data (and log if separate) are an ok size. | |
1031 | */ | |
1032 | STATIC int | |
4249023a | 1033 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
1034 | { |
1035 | xfs_buf_t *bp; | |
1036 | xfs_daddr_t d; | |
0771fb45 | 1037 | |
1da177e4 LT |
1038 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
1039 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
0b932ccc | 1040 | xfs_warn(mp, "filesystem size mismatch detected"); |
657a4cff | 1041 | return XFS_ERROR(EFBIG); |
1da177e4 | 1042 | } |
1922c949 DC |
1043 | bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp, |
1044 | d - XFS_FSS_TO_BB(mp, 1), | |
1045 | BBTOB(XFS_FSS_TO_BB(mp, 1)), 0); | |
1046 | if (!bp) { | |
0b932ccc | 1047 | xfs_warn(mp, "last sector read failed"); |
1922c949 | 1048 | return EIO; |
1da177e4 | 1049 | } |
1922c949 | 1050 | xfs_buf_relse(bp); |
1da177e4 | 1051 | |
4249023a | 1052 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
1053 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
1054 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
0b932ccc | 1055 | xfs_warn(mp, "log size mismatch detected"); |
657a4cff | 1056 | return XFS_ERROR(EFBIG); |
1da177e4 | 1057 | } |
1922c949 DC |
1058 | bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp, |
1059 | d - XFS_FSB_TO_BB(mp, 1), | |
1060 | XFS_FSB_TO_B(mp, 1), 0); | |
1061 | if (!bp) { | |
0b932ccc | 1062 | xfs_warn(mp, "log device read failed"); |
1922c949 | 1063 | return EIO; |
0771fb45 | 1064 | } |
1922c949 | 1065 | xfs_buf_relse(bp); |
0771fb45 ES |
1066 | } |
1067 | return 0; | |
1068 | } | |
1069 | ||
7d095257 CH |
1070 | /* |
1071 | * Clear the quotaflags in memory and in the superblock. | |
1072 | */ | |
1073 | int | |
1074 | xfs_mount_reset_sbqflags( | |
1075 | struct xfs_mount *mp) | |
1076 | { | |
1077 | int error; | |
1078 | struct xfs_trans *tp; | |
1079 | ||
1080 | mp->m_qflags = 0; | |
1081 | ||
1082 | /* | |
1083 | * It is OK to look at sb_qflags here in mount path, | |
1084 | * without m_sb_lock. | |
1085 | */ | |
1086 | if (mp->m_sb.sb_qflags == 0) | |
1087 | return 0; | |
1088 | spin_lock(&mp->m_sb_lock); | |
1089 | mp->m_sb.sb_qflags = 0; | |
1090 | spin_unlock(&mp->m_sb_lock); | |
1091 | ||
1092 | /* | |
1093 | * If the fs is readonly, let the incore superblock run | |
1094 | * with quotas off but don't flush the update out to disk | |
1095 | */ | |
1096 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1097 | return 0; | |
1098 | ||
1099 | #ifdef QUOTADEBUG | |
53487786 | 1100 | xfs_notice(mp, "Writing superblock quota changes"); |
7d095257 CH |
1101 | #endif |
1102 | ||
1103 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); | |
1104 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1105 | XFS_DEFAULT_LOG_COUNT); | |
1106 | if (error) { | |
1107 | xfs_trans_cancel(tp, 0); | |
53487786 | 1108 | xfs_alert(mp, "%s: Superblock update failed!", __func__); |
7d095257 CH |
1109 | return error; |
1110 | } | |
1111 | ||
1112 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1113 | return xfs_trans_commit(tp, 0); | |
1114 | } | |
1115 | ||
d5db0f97 ES |
1116 | __uint64_t |
1117 | xfs_default_resblks(xfs_mount_t *mp) | |
1118 | { | |
1119 | __uint64_t resblks; | |
1120 | ||
1121 | /* | |
8babd8a2 DC |
1122 | * We default to 5% or 8192 fsbs of space reserved, whichever is |
1123 | * smaller. This is intended to cover concurrent allocation | |
1124 | * transactions when we initially hit enospc. These each require a 4 | |
1125 | * block reservation. Hence by default we cover roughly 2000 concurrent | |
1126 | * allocation reservations. | |
d5db0f97 ES |
1127 | */ |
1128 | resblks = mp->m_sb.sb_dblocks; | |
1129 | do_div(resblks, 20); | |
8babd8a2 | 1130 | resblks = min_t(__uint64_t, resblks, 8192); |
d5db0f97 ES |
1131 | return resblks; |
1132 | } | |
1133 | ||
0771fb45 | 1134 | /* |
0771fb45 ES |
1135 | * This function does the following on an initial mount of a file system: |
1136 | * - reads the superblock from disk and init the mount struct | |
1137 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1138 | * so we don't mount terabyte filesystems | |
1139 | * - init mount struct realtime fields | |
1140 | * - allocate inode hash table for fs | |
1141 | * - init directory manager | |
1142 | * - perform recovery and init the log manager | |
1143 | */ | |
1144 | int | |
1145 | xfs_mountfs( | |
4249023a | 1146 | xfs_mount_t *mp) |
0771fb45 ES |
1147 | { |
1148 | xfs_sb_t *sbp = &(mp->m_sb); | |
1149 | xfs_inode_t *rip; | |
0771fb45 | 1150 | __uint64_t resblks; |
7d095257 CH |
1151 | uint quotamount = 0; |
1152 | uint quotaflags = 0; | |
0771fb45 ES |
1153 | int error = 0; |
1154 | ||
0771fb45 ES |
1155 | xfs_mount_common(mp, sbp); |
1156 | ||
ee1c0908 | 1157 | /* |
e6957ea4 ES |
1158 | * Check for a mismatched features2 values. Older kernels |
1159 | * read & wrote into the wrong sb offset for sb_features2 | |
1160 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1161 | * when sb_features2 was added, which made older superblock | |
1162 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1163 | * |
e6957ea4 ES |
1164 | * For backwards compatibility, we make both slots equal. |
1165 | * | |
1166 | * If we detect a mismatched field, we OR the set bits into the | |
1167 | * existing features2 field in case it has already been modified; we | |
1168 | * don't want to lose any features. We then update the bad location | |
1169 | * with the ORed value so that older kernels will see any features2 | |
1170 | * flags, and mark the two fields as needing updates once the | |
1171 | * transaction subsystem is online. | |
ee1c0908 | 1172 | */ |
e6957ea4 | 1173 | if (xfs_sb_has_mismatched_features2(sbp)) { |
0b932ccc | 1174 | xfs_warn(mp, "correcting sb_features alignment problem"); |
ee1c0908 | 1175 | sbp->sb_features2 |= sbp->sb_bad_features2; |
e6957ea4 | 1176 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1177 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1178 | |
1179 | /* | |
1180 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1181 | * slot. | |
1182 | */ | |
7c12f296 TS |
1183 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1184 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1185 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1186 | } |
1187 | ||
1188 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1189 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1190 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1191 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1192 | |
7c12f296 TS |
1193 | /* update sb_versionnum for the clearing of the morebits */ |
1194 | if (!sbp->sb_features2) | |
7884bc86 | 1195 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1196 | } |
1197 | ||
0771fb45 ES |
1198 | /* |
1199 | * Check if sb_agblocks is aligned at stripe boundary | |
1200 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1201 | * allocator alignment is within an ag, therefore ag has | |
1202 | * to be aligned at stripe boundary. | |
1203 | */ | |
7884bc86 | 1204 | error = xfs_update_alignment(mp); |
0771fb45 | 1205 | if (error) |
f9057e3d | 1206 | goto out; |
0771fb45 ES |
1207 | |
1208 | xfs_alloc_compute_maxlevels(mp); | |
1209 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1210 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1211 | xfs_ialloc_compute_maxlevels(mp); | |
1212 | ||
1213 | xfs_set_maxicount(mp); | |
1214 | ||
1215 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
1216 | ||
27174203 CH |
1217 | error = xfs_uuid_mount(mp); |
1218 | if (error) | |
1219 | goto out; | |
1da177e4 | 1220 | |
0771fb45 ES |
1221 | /* |
1222 | * Set the minimum read and write sizes | |
1223 | */ | |
1224 | xfs_set_rw_sizes(mp); | |
1225 | ||
055388a3 DC |
1226 | /* set the low space thresholds for dynamic preallocation */ |
1227 | xfs_set_low_space_thresholds(mp); | |
1228 | ||
0771fb45 ES |
1229 | /* |
1230 | * Set the inode cluster size. | |
1231 | * This may still be overridden by the file system | |
1232 | * block size if it is larger than the chosen cluster size. | |
1233 | */ | |
1234 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1235 | ||
1236 | /* | |
1237 | * Set inode alignment fields | |
1238 | */ | |
1239 | xfs_set_inoalignment(mp); | |
1240 | ||
1241 | /* | |
1242 | * Check that the data (and log if separate) are an ok size. | |
1243 | */ | |
4249023a | 1244 | error = xfs_check_sizes(mp); |
0771fb45 | 1245 | if (error) |
f9057e3d | 1246 | goto out_remove_uuid; |
0771fb45 | 1247 | |
1da177e4 LT |
1248 | /* |
1249 | * Initialize realtime fields in the mount structure | |
1250 | */ | |
0771fb45 ES |
1251 | error = xfs_rtmount_init(mp); |
1252 | if (error) { | |
0b932ccc | 1253 | xfs_warn(mp, "RT mount failed"); |
f9057e3d | 1254 | goto out_remove_uuid; |
1da177e4 LT |
1255 | } |
1256 | ||
1da177e4 LT |
1257 | /* |
1258 | * Copies the low order bits of the timestamp and the randomly | |
1259 | * set "sequence" number out of a UUID. | |
1260 | */ | |
1261 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1262 | ||
1da177e4 LT |
1263 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1264 | ||
f6c2d1fa | 1265 | xfs_dir_mount(mp); |
1da177e4 LT |
1266 | |
1267 | /* | |
1268 | * Initialize the attribute manager's entries. | |
1269 | */ | |
1270 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1271 | ||
1272 | /* | |
1273 | * Initialize the precomputed transaction reservations values. | |
1274 | */ | |
1275 | xfs_trans_init(mp); | |
1276 | ||
1da177e4 LT |
1277 | /* |
1278 | * Allocate and initialize the per-ag data. | |
1279 | */ | |
1c1c6ebc | 1280 | spin_lock_init(&mp->m_perag_lock); |
9b98b6f3 | 1281 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC); |
1c1c6ebc DC |
1282 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); |
1283 | if (error) { | |
0b932ccc | 1284 | xfs_warn(mp, "Failed per-ag init: %d", error); |
f9057e3d | 1285 | goto out_remove_uuid; |
1c1c6ebc | 1286 | } |
1da177e4 | 1287 | |
f9057e3d | 1288 | if (!sbp->sb_logblocks) { |
0b932ccc | 1289 | xfs_warn(mp, "no log defined"); |
f9057e3d CH |
1290 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); |
1291 | error = XFS_ERROR(EFSCORRUPTED); | |
1292 | goto out_free_perag; | |
1293 | } | |
1294 | ||
1da177e4 LT |
1295 | /* |
1296 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1297 | */ | |
f9057e3d CH |
1298 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1299 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1300 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1301 | if (error) { | |
0b932ccc | 1302 | xfs_warn(mp, "log mount failed"); |
f9057e3d | 1303 | goto out_free_perag; |
1da177e4 LT |
1304 | } |
1305 | ||
92821e2b DC |
1306 | /* |
1307 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1308 | * not. If it was, with the first phase of recovery has completed, we | |
1309 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1310 | * but they are recovered transactionally in the second recovery phase | |
1311 | * later. | |
1312 | * | |
1313 | * Hence we can safely re-initialise incore superblock counters from | |
1314 | * the per-ag data. These may not be correct if the filesystem was not | |
1315 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1316 | * doing this. | |
1317 | * | |
1318 | * If the filesystem was cleanly unmounted, then we can trust the | |
1319 | * values in the superblock to be correct and we don't need to do | |
1320 | * anything here. | |
1321 | * | |
1322 | * If we are currently making the filesystem, the initialisation will | |
1323 | * fail as the perag data is in an undefined state. | |
1324 | */ | |
92821e2b DC |
1325 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1326 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1327 | !mp->m_sb.sb_inprogress) { | |
1328 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d CH |
1329 | if (error) |
1330 | goto out_free_perag; | |
92821e2b | 1331 | } |
f9057e3d | 1332 | |
1da177e4 LT |
1333 | /* |
1334 | * Get and sanity-check the root inode. | |
1335 | * Save the pointer to it in the mount structure. | |
1336 | */ | |
7b6259e7 | 1337 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip); |
1da177e4 | 1338 | if (error) { |
0b932ccc | 1339 | xfs_warn(mp, "failed to read root inode"); |
f9057e3d | 1340 | goto out_log_dealloc; |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | ASSERT(rip != NULL); | |
1da177e4 LT |
1344 | |
1345 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
0b932ccc | 1346 | xfs_warn(mp, "corrupted root inode %llu: not a directory", |
b6574520 | 1347 | (unsigned long long)rip->i_ino); |
1da177e4 LT |
1348 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1349 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1350 | mp); | |
1351 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1352 | goto out_rele_rip; |
1da177e4 LT |
1353 | } |
1354 | mp->m_rootip = rip; /* save it */ | |
1355 | ||
1356 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1357 | ||
1358 | /* | |
1359 | * Initialize realtime inode pointers in the mount structure | |
1360 | */ | |
0771fb45 ES |
1361 | error = xfs_rtmount_inodes(mp); |
1362 | if (error) { | |
1da177e4 LT |
1363 | /* |
1364 | * Free up the root inode. | |
1365 | */ | |
0b932ccc | 1366 | xfs_warn(mp, "failed to read RT inodes"); |
f9057e3d | 1367 | goto out_rele_rip; |
1da177e4 LT |
1368 | } |
1369 | ||
1370 | /* | |
7884bc86 CH |
1371 | * If this is a read-only mount defer the superblock updates until |
1372 | * the next remount into writeable mode. Otherwise we would never | |
1373 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1374 | */ |
7884bc86 CH |
1375 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1376 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec | 1377 | if (error) { |
0b932ccc | 1378 | xfs_warn(mp, "failed to write sb changes"); |
b93b6e43 | 1379 | goto out_rtunmount; |
e5720eec DC |
1380 | } |
1381 | } | |
1da177e4 LT |
1382 | |
1383 | /* | |
1384 | * Initialise the XFS quota management subsystem for this mount | |
1385 | */ | |
7d095257 CH |
1386 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1387 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1388 | if (error) | |
1389 | goto out_rtunmount; | |
1390 | } else { | |
1391 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1392 | ||
1393 | /* | |
1394 | * If a file system had quotas running earlier, but decided to | |
1395 | * mount without -o uquota/pquota/gquota options, revoke the | |
1396 | * quotachecked license. | |
1397 | */ | |
1398 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
0b932ccc | 1399 | xfs_notice(mp, "resetting quota flags"); |
7d095257 CH |
1400 | error = xfs_mount_reset_sbqflags(mp); |
1401 | if (error) | |
1402 | return error; | |
1403 | } | |
1404 | } | |
1da177e4 LT |
1405 | |
1406 | /* | |
1407 | * Finish recovering the file system. This part needed to be | |
1408 | * delayed until after the root and real-time bitmap inodes | |
1409 | * were consistently read in. | |
1410 | */ | |
4249023a | 1411 | error = xfs_log_mount_finish(mp); |
1da177e4 | 1412 | if (error) { |
0b932ccc | 1413 | xfs_warn(mp, "log mount finish failed"); |
b93b6e43 | 1414 | goto out_rtunmount; |
1da177e4 LT |
1415 | } |
1416 | ||
1417 | /* | |
1418 | * Complete the quota initialisation, post-log-replay component. | |
1419 | */ | |
7d095257 CH |
1420 | if (quotamount) { |
1421 | ASSERT(mp->m_qflags == 0); | |
1422 | mp->m_qflags = quotaflags; | |
1423 | ||
1424 | xfs_qm_mount_quotas(mp); | |
1425 | } | |
1426 | ||
84e1e99f DC |
1427 | /* |
1428 | * Now we are mounted, reserve a small amount of unused space for | |
1429 | * privileged transactions. This is needed so that transaction | |
1430 | * space required for critical operations can dip into this pool | |
1431 | * when at ENOSPC. This is needed for operations like create with | |
1432 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1433 | * are not allowed to use this reserved space. | |
8babd8a2 DC |
1434 | * |
1435 | * This may drive us straight to ENOSPC on mount, but that implies | |
1436 | * we were already there on the last unmount. Warn if this occurs. | |
84e1e99f | 1437 | */ |
d5db0f97 ES |
1438 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1439 | resblks = xfs_default_resblks(mp); | |
1440 | error = xfs_reserve_blocks(mp, &resblks, NULL); | |
1441 | if (error) | |
0b932ccc DC |
1442 | xfs_warn(mp, |
1443 | "Unable to allocate reserve blocks. Continuing without reserve pool."); | |
d5db0f97 | 1444 | } |
84e1e99f | 1445 | |
1da177e4 LT |
1446 | return 0; |
1447 | ||
b93b6e43 CH |
1448 | out_rtunmount: |
1449 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1450 | out_rele_rip: |
43355099 | 1451 | IRELE(rip); |
f9057e3d | 1452 | out_log_dealloc: |
21b699c8 | 1453 | xfs_log_unmount(mp); |
f9057e3d | 1454 | out_free_perag: |
ff4f038c | 1455 | xfs_free_perag(mp); |
f9057e3d | 1456 | out_remove_uuid: |
27174203 | 1457 | xfs_uuid_unmount(mp); |
f9057e3d | 1458 | out: |
1da177e4 LT |
1459 | return error; |
1460 | } | |
1461 | ||
1462 | /* | |
1da177e4 LT |
1463 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1464 | * log and makes sure that incore structures are freed. | |
1465 | */ | |
41b5c2e7 CH |
1466 | void |
1467 | xfs_unmountfs( | |
1468 | struct xfs_mount *mp) | |
1da177e4 | 1469 | { |
41b5c2e7 CH |
1470 | __uint64_t resblks; |
1471 | int error; | |
1da177e4 | 1472 | |
7d095257 | 1473 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1474 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1475 | IRELE(mp->m_rootip); |
1476 | ||
641c56fb DC |
1477 | /* |
1478 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1479 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1480 | * the transaction is still sitting in a iclog. The stale inodes |
1481 | * on that buffer will have their flush locks held until the | |
1482 | * transaction hits the disk and the callbacks run. the inode | |
1483 | * flush takes the flush lock unconditionally and with nothing to | |
1484 | * push out the iclog we will never get that unlocked. hence we | |
1485 | * need to force the log first. | |
1486 | */ | |
a14a348b | 1487 | xfs_log_force(mp, XFS_LOG_SYNC); |
c854363e DC |
1488 | |
1489 | /* | |
1490 | * Do a delwri reclaim pass first so that as many dirty inodes are | |
1491 | * queued up for IO as possible. Then flush the buffers before making | |
1492 | * a synchronous path to catch all the remaining inodes are reclaimed. | |
1493 | * This makes the reclaim process as quick as possible by avoiding | |
1494 | * synchronous writeout and blocking on inodes already in the delwri | |
1495 | * state as much as possible. | |
1496 | */ | |
1497 | xfs_reclaim_inodes(mp, 0); | |
1498 | XFS_bflush(mp->m_ddev_targp); | |
1499 | xfs_reclaim_inodes(mp, SYNC_WAIT); | |
1da177e4 | 1500 | |
7d095257 | 1501 | xfs_qm_unmount(mp); |
a357a121 | 1502 | |
1da177e4 LT |
1503 | /* |
1504 | * Flush out the log synchronously so that we know for sure | |
1505 | * that nothing is pinned. This is important because bflush() | |
1506 | * will skip pinned buffers. | |
1507 | */ | |
a14a348b | 1508 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 LT |
1509 | |
1510 | xfs_binval(mp->m_ddev_targp); | |
1511 | if (mp->m_rtdev_targp) { | |
1512 | xfs_binval(mp->m_rtdev_targp); | |
1513 | } | |
1514 | ||
84e1e99f DC |
1515 | /* |
1516 | * Unreserve any blocks we have so that when we unmount we don't account | |
1517 | * the reserved free space as used. This is really only necessary for | |
1518 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1519 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1520 | * |
1521 | * We don't bother correcting this elsewhere for lazy superblock | |
1522 | * counting because on mount of an unclean filesystem we reconstruct the | |
1523 | * correct counter value and this is irrelevant. | |
1524 | * | |
1525 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1526 | * we only every apply deltas to the superblock and hence the incore | |
1527 | * value does not matter.... | |
1528 | */ | |
1529 | resblks = 0; | |
714082bc DC |
1530 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1531 | if (error) | |
0b932ccc | 1532 | xfs_warn(mp, "Unable to free reserved block pool. " |
714082bc DC |
1533 | "Freespace may not be correct on next mount."); |
1534 | ||
e5720eec DC |
1535 | error = xfs_log_sbcount(mp, 1); |
1536 | if (error) | |
0b932ccc | 1537 | xfs_warn(mp, "Unable to update superblock counters. " |
e5720eec | 1538 | "Freespace may not be correct on next mount."); |
1da177e4 | 1539 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1540 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
21b699c8 CH |
1541 | xfs_log_unmount_write(mp); |
1542 | xfs_log_unmount(mp); | |
27174203 | 1543 | xfs_uuid_unmount(mp); |
1da177e4 | 1544 | |
1550d0b0 | 1545 | #if defined(DEBUG) |
0ce4cfd4 | 1546 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1547 | #endif |
ff4f038c | 1548 | xfs_free_perag(mp); |
1da177e4 LT |
1549 | } |
1550 | ||
ba0f32d4 | 1551 | STATIC void |
1da177e4 LT |
1552 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1553 | { | |
1554 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1555 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1556 | if (mp->m_rtdev_targp) | |
1557 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1558 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1559 | } | |
1560 | ||
92821e2b DC |
1561 | int |
1562 | xfs_fs_writable(xfs_mount_t *mp) | |
1563 | { | |
b267ce99 | 1564 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1565 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1566 | } |
1567 | ||
1568 | /* | |
92821e2b DC |
1569 | * Called either periodically to keep the on disk superblock values |
1570 | * roughly up to date or from unmount to make sure the values are | |
1571 | * correct on a clean unmount. | |
92821e2b DC |
1572 | */ |
1573 | int | |
1574 | xfs_log_sbcount( | |
1575 | xfs_mount_t *mp, | |
1576 | uint sync) | |
1577 | { | |
1578 | xfs_trans_t *tp; | |
1579 | int error; | |
1580 | ||
1581 | if (!xfs_fs_writable(mp)) | |
1582 | return 0; | |
1583 | ||
d4d90b57 | 1584 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1585 | |
1586 | /* | |
1587 | * we don't need to do this if we are updating the superblock | |
1588 | * counters on every modification. | |
1589 | */ | |
1590 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1591 | return 0; | |
1592 | ||
7a249cf8 CH |
1593 | /* |
1594 | * We can be called during the process of freezing, so make sure | |
1595 | * we go ahead even if the frozen for new transactions. We will | |
1596 | * always use a sync transaction in the freeze path to make sure | |
1597 | * the transaction has completed by the time we return. | |
1598 | */ | |
1599 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP, false); | |
92821e2b DC |
1600 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1601 | XFS_DEFAULT_LOG_COUNT); | |
1602 | if (error) { | |
1603 | xfs_trans_cancel(tp, 0); | |
1604 | return error; | |
1605 | } | |
1606 | ||
1607 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1608 | if (sync) | |
1609 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1610 | error = xfs_trans_commit(tp, 0); |
1611 | return error; | |
92821e2b DC |
1612 | } |
1613 | ||
1da177e4 LT |
1614 | int |
1615 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1616 | { | |
1617 | xfs_buf_t *sbp; | |
1da177e4 LT |
1618 | int error = 0; |
1619 | ||
1620 | /* | |
1621 | * skip superblock write if fs is read-only, or | |
1622 | * if we are doing a forced umount. | |
1623 | */ | |
bd186aa9 | 1624 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1625 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1626 | |
92821e2b | 1627 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1628 | |
1da177e4 LT |
1629 | XFS_BUF_UNDONE(sbp); |
1630 | XFS_BUF_UNREAD(sbp); | |
1631 | XFS_BUF_UNDELAYWRITE(sbp); | |
1632 | XFS_BUF_WRITE(sbp); | |
1633 | XFS_BUF_UNASYNC(sbp); | |
1634 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1635 | xfsbdstrat(mp, sbp); | |
1a1a3e97 | 1636 | error = xfs_buf_iowait(sbp); |
1da177e4 LT |
1637 | if (error) |
1638 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1639 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
92821e2b | 1640 | xfs_buf_relse(sbp); |
1da177e4 | 1641 | } |
014c2544 | 1642 | return error; |
1da177e4 LT |
1643 | } |
1644 | ||
1645 | /* | |
1646 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1647 | * in-core superblock into the superblock buffer to be logged. | |
1648 | * It does not provide the higher level of locking that is | |
1649 | * needed to protect the in-core superblock from concurrent | |
1650 | * access. | |
1651 | */ | |
1652 | void | |
1653 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1654 | { | |
1655 | xfs_buf_t *bp; | |
1656 | int first; | |
1657 | int last; | |
1658 | xfs_mount_t *mp; | |
1da177e4 LT |
1659 | xfs_sb_field_t f; |
1660 | ||
1661 | ASSERT(fields); | |
1662 | if (!fields) | |
1663 | return; | |
1664 | mp = tp->t_mountp; | |
1665 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1666 | first = sizeof(xfs_sb_t); |
1667 | last = 0; | |
1668 | ||
1669 | /* translate/copy */ | |
1670 | ||
2bdf7cd0 | 1671 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1672 | |
1673 | /* find modified range */ | |
587aa0fe DC |
1674 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); |
1675 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1676 | last = xfs_sb_info[f + 1].offset - 1; | |
1da177e4 LT |
1677 | |
1678 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1679 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1680 | first = xfs_sb_info[f].offset; | |
1681 | ||
1da177e4 LT |
1682 | xfs_trans_log_buf(tp, bp, first, last); |
1683 | } | |
d210a28c | 1684 | |
d210a28c | 1685 | |
1da177e4 LT |
1686 | /* |
1687 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1688 | * a delta to a specified field in the in-core superblock. Simply | |
1689 | * switch on the field indicated and apply the delta to that field. | |
1690 | * Fields are not allowed to dip below zero, so if the delta would | |
1691 | * do this do not apply it and return EINVAL. | |
1692 | * | |
3685c2a1 | 1693 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1694 | */ |
d96f8f89 | 1695 | STATIC int |
20f4ebf2 DC |
1696 | xfs_mod_incore_sb_unlocked( |
1697 | xfs_mount_t *mp, | |
1698 | xfs_sb_field_t field, | |
1699 | int64_t delta, | |
1700 | int rsvd) | |
1da177e4 LT |
1701 | { |
1702 | int scounter; /* short counter for 32 bit fields */ | |
1703 | long long lcounter; /* long counter for 64 bit fields */ | |
1704 | long long res_used, rem; | |
1705 | ||
1706 | /* | |
1707 | * With the in-core superblock spin lock held, switch | |
1708 | * on the indicated field. Apply the delta to the | |
1709 | * proper field. If the fields value would dip below | |
1710 | * 0, then do not apply the delta and return EINVAL. | |
1711 | */ | |
1712 | switch (field) { | |
1713 | case XFS_SBS_ICOUNT: | |
1714 | lcounter = (long long)mp->m_sb.sb_icount; | |
1715 | lcounter += delta; | |
1716 | if (lcounter < 0) { | |
1717 | ASSERT(0); | |
014c2544 | 1718 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1719 | } |
1720 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1721 | return 0; |
1da177e4 LT |
1722 | case XFS_SBS_IFREE: |
1723 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1724 | lcounter += delta; | |
1725 | if (lcounter < 0) { | |
1726 | ASSERT(0); | |
014c2544 | 1727 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1728 | } |
1729 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1730 | return 0; |
1da177e4 | 1731 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1732 | lcounter = (long long) |
1733 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1734 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1735 | ||
1736 | if (delta > 0) { /* Putting blocks back */ | |
1737 | if (res_used > delta) { | |
1738 | mp->m_resblks_avail += delta; | |
1739 | } else { | |
1740 | rem = delta - res_used; | |
1741 | mp->m_resblks_avail = mp->m_resblks; | |
1742 | lcounter += rem; | |
1743 | } | |
1744 | } else { /* Taking blocks away */ | |
1da177e4 | 1745 | lcounter += delta; |
8babd8a2 DC |
1746 | if (lcounter >= 0) { |
1747 | mp->m_sb.sb_fdblocks = lcounter + | |
1748 | XFS_ALLOC_SET_ASIDE(mp); | |
1749 | return 0; | |
1750 | } | |
1da177e4 | 1751 | |
8babd8a2 DC |
1752 | /* |
1753 | * We are out of blocks, use any available reserved | |
1754 | * blocks if were allowed to. | |
1755 | */ | |
1756 | if (!rsvd) | |
1757 | return XFS_ERROR(ENOSPC); | |
1da177e4 | 1758 | |
8babd8a2 DC |
1759 | lcounter = (long long)mp->m_resblks_avail + delta; |
1760 | if (lcounter >= 0) { | |
1761 | mp->m_resblks_avail = lcounter; | |
1762 | return 0; | |
1da177e4 | 1763 | } |
8babd8a2 DC |
1764 | printk_once(KERN_WARNING |
1765 | "Filesystem \"%s\": reserve blocks depleted! " | |
1766 | "Consider increasing reserve pool size.", | |
1767 | mp->m_fsname); | |
1768 | return XFS_ERROR(ENOSPC); | |
1da177e4 LT |
1769 | } |
1770 | ||
4be536de | 1771 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1772 | return 0; |
1da177e4 LT |
1773 | case XFS_SBS_FREXTENTS: |
1774 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1775 | lcounter += delta; | |
1776 | if (lcounter < 0) { | |
014c2544 | 1777 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1778 | } |
1779 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1780 | return 0; |
1da177e4 LT |
1781 | case XFS_SBS_DBLOCKS: |
1782 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1783 | lcounter += delta; | |
1784 | if (lcounter < 0) { | |
1785 | ASSERT(0); | |
014c2544 | 1786 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1787 | } |
1788 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1789 | return 0; |
1da177e4 LT |
1790 | case XFS_SBS_AGCOUNT: |
1791 | scounter = mp->m_sb.sb_agcount; | |
1792 | scounter += delta; | |
1793 | if (scounter < 0) { | |
1794 | ASSERT(0); | |
014c2544 | 1795 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1796 | } |
1797 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1798 | return 0; |
1da177e4 LT |
1799 | case XFS_SBS_IMAX_PCT: |
1800 | scounter = mp->m_sb.sb_imax_pct; | |
1801 | scounter += delta; | |
1802 | if (scounter < 0) { | |
1803 | ASSERT(0); | |
014c2544 | 1804 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1805 | } |
1806 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1807 | return 0; |
1da177e4 LT |
1808 | case XFS_SBS_REXTSIZE: |
1809 | scounter = mp->m_sb.sb_rextsize; | |
1810 | scounter += delta; | |
1811 | if (scounter < 0) { | |
1812 | ASSERT(0); | |
014c2544 | 1813 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1814 | } |
1815 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1816 | return 0; |
1da177e4 LT |
1817 | case XFS_SBS_RBMBLOCKS: |
1818 | scounter = mp->m_sb.sb_rbmblocks; | |
1819 | scounter += delta; | |
1820 | if (scounter < 0) { | |
1821 | ASSERT(0); | |
014c2544 | 1822 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1823 | } |
1824 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1825 | return 0; |
1da177e4 LT |
1826 | case XFS_SBS_RBLOCKS: |
1827 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1828 | lcounter += delta; | |
1829 | if (lcounter < 0) { | |
1830 | ASSERT(0); | |
014c2544 | 1831 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1832 | } |
1833 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1834 | return 0; |
1da177e4 LT |
1835 | case XFS_SBS_REXTENTS: |
1836 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1837 | lcounter += delta; | |
1838 | if (lcounter < 0) { | |
1839 | ASSERT(0); | |
014c2544 | 1840 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1841 | } |
1842 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1843 | return 0; |
1da177e4 LT |
1844 | case XFS_SBS_REXTSLOG: |
1845 | scounter = mp->m_sb.sb_rextslog; | |
1846 | scounter += delta; | |
1847 | if (scounter < 0) { | |
1848 | ASSERT(0); | |
014c2544 | 1849 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1850 | } |
1851 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1852 | return 0; |
1da177e4 LT |
1853 | default: |
1854 | ASSERT(0); | |
014c2544 | 1855 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1856 | } |
1857 | } | |
1858 | ||
1859 | /* | |
1860 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1861 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1862 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1863 | * routine to do the work. |
1864 | */ | |
1865 | int | |
20f4ebf2 | 1866 | xfs_mod_incore_sb( |
96540c78 CH |
1867 | struct xfs_mount *mp, |
1868 | xfs_sb_field_t field, | |
1869 | int64_t delta, | |
1870 | int rsvd) | |
1da177e4 | 1871 | { |
96540c78 | 1872 | int status; |
1da177e4 | 1873 | |
8d280b98 | 1874 | #ifdef HAVE_PERCPU_SB |
96540c78 | 1875 | ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS); |
8d280b98 | 1876 | #endif |
96540c78 CH |
1877 | spin_lock(&mp->m_sb_lock); |
1878 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); | |
1879 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 | 1880 | |
014c2544 | 1881 | return status; |
1da177e4 LT |
1882 | } |
1883 | ||
1884 | /* | |
1b040712 | 1885 | * Change more than one field in the in-core superblock structure at a time. |
1da177e4 | 1886 | * |
1b040712 CH |
1887 | * The fields and changes to those fields are specified in the array of |
1888 | * xfs_mod_sb structures passed in. Either all of the specified deltas | |
1889 | * will be applied or none of them will. If any modified field dips below 0, | |
1890 | * then all modifications will be backed out and EINVAL will be returned. | |
1891 | * | |
1892 | * Note that this function may not be used for the superblock values that | |
1893 | * are tracked with the in-memory per-cpu counters - a direct call to | |
1894 | * xfs_icsb_modify_counters is required for these. | |
1da177e4 LT |
1895 | */ |
1896 | int | |
1b040712 CH |
1897 | xfs_mod_incore_sb_batch( |
1898 | struct xfs_mount *mp, | |
1899 | xfs_mod_sb_t *msb, | |
1900 | uint nmsb, | |
1901 | int rsvd) | |
1da177e4 | 1902 | { |
45c51b99 | 1903 | xfs_mod_sb_t *msbp; |
1b040712 | 1904 | int error = 0; |
1da177e4 LT |
1905 | |
1906 | /* | |
1b040712 CH |
1907 | * Loop through the array of mod structures and apply each individually. |
1908 | * If any fail, then back out all those which have already been applied. | |
1909 | * Do all of this within the scope of the m_sb_lock so that all of the | |
1910 | * changes will be atomic. | |
1da177e4 | 1911 | */ |
3685c2a1 | 1912 | spin_lock(&mp->m_sb_lock); |
45c51b99 | 1913 | for (msbp = msb; msbp < (msb + nmsb); msbp++) { |
1b040712 CH |
1914 | ASSERT(msbp->msb_field < XFS_SBS_ICOUNT || |
1915 | msbp->msb_field > XFS_SBS_FDBLOCKS); | |
8d280b98 | 1916 | |
1b040712 CH |
1917 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, |
1918 | msbp->msb_delta, rsvd); | |
1919 | if (error) | |
1920 | goto unwind; | |
1da177e4 | 1921 | } |
1b040712 CH |
1922 | spin_unlock(&mp->m_sb_lock); |
1923 | return 0; | |
1da177e4 | 1924 | |
1b040712 CH |
1925 | unwind: |
1926 | while (--msbp >= msb) { | |
1927 | error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field, | |
1928 | -msbp->msb_delta, rsvd); | |
1929 | ASSERT(error == 0); | |
1da177e4 | 1930 | } |
3685c2a1 | 1931 | spin_unlock(&mp->m_sb_lock); |
1b040712 | 1932 | return error; |
1da177e4 LT |
1933 | } |
1934 | ||
1935 | /* | |
1936 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1937 | * The buffer is returned locked and read in from disk. | |
1938 | * The buffer should be released with a call to xfs_brelse(). | |
1939 | * | |
1940 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1941 | * the superblock buffer if it can be locked without sleeping. | |
1942 | * If it can't then we'll return NULL. | |
1943 | */ | |
0c842ad4 | 1944 | struct xfs_buf * |
1da177e4 | 1945 | xfs_getsb( |
0c842ad4 CH |
1946 | struct xfs_mount *mp, |
1947 | int flags) | |
1da177e4 | 1948 | { |
0c842ad4 | 1949 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1950 | |
0c842ad4 CH |
1951 | if (!xfs_buf_trylock(bp)) { |
1952 | if (flags & XBF_TRYLOCK) | |
1da177e4 | 1953 | return NULL; |
0c842ad4 | 1954 | xfs_buf_lock(bp); |
1da177e4 | 1955 | } |
0c842ad4 | 1956 | |
1da177e4 LT |
1957 | XFS_BUF_HOLD(bp); |
1958 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1959 | return bp; |
1da177e4 LT |
1960 | } |
1961 | ||
1962 | /* | |
1963 | * Used to free the superblock along various error paths. | |
1964 | */ | |
1965 | void | |
1966 | xfs_freesb( | |
26af6552 | 1967 | struct xfs_mount *mp) |
1da177e4 | 1968 | { |
26af6552 | 1969 | struct xfs_buf *bp = mp->m_sb_bp; |
1da177e4 | 1970 | |
26af6552 | 1971 | xfs_buf_lock(bp); |
1da177e4 | 1972 | mp->m_sb_bp = NULL; |
26af6552 | 1973 | xfs_buf_relse(bp); |
1da177e4 LT |
1974 | } |
1975 | ||
1da177e4 LT |
1976 | /* |
1977 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1978 | * be altered by the mount options, as well as any potential sb_features2 |
1979 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1980 | */ |
7884bc86 | 1981 | int |
ee1c0908 | 1982 | xfs_mount_log_sb( |
1da177e4 LT |
1983 | xfs_mount_t *mp, |
1984 | __int64_t fields) | |
1985 | { | |
1986 | xfs_trans_t *tp; | |
e5720eec | 1987 | int error; |
1da177e4 | 1988 | |
ee1c0908 | 1989 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1990 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1991 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1992 | |
1993 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
1994 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1995 | XFS_DEFAULT_LOG_COUNT); | |
1996 | if (error) { | |
1da177e4 | 1997 | xfs_trans_cancel(tp, 0); |
e5720eec | 1998 | return error; |
1da177e4 LT |
1999 | } |
2000 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
2001 | error = xfs_trans_commit(tp, 0); |
2002 | return error; | |
1da177e4 | 2003 | } |
8d280b98 | 2004 | |
dda35b8f CH |
2005 | /* |
2006 | * If the underlying (data/log/rt) device is readonly, there are some | |
2007 | * operations that cannot proceed. | |
2008 | */ | |
2009 | int | |
2010 | xfs_dev_is_read_only( | |
2011 | struct xfs_mount *mp, | |
2012 | char *message) | |
2013 | { | |
2014 | if (xfs_readonly_buftarg(mp->m_ddev_targp) || | |
2015 | xfs_readonly_buftarg(mp->m_logdev_targp) || | |
2016 | (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) { | |
0b932ccc DC |
2017 | xfs_notice(mp, "%s required on read-only device.", message); |
2018 | xfs_notice(mp, "write access unavailable, cannot proceed."); | |
dda35b8f CH |
2019 | return EROFS; |
2020 | } | |
2021 | return 0; | |
2022 | } | |
8d280b98 DC |
2023 | |
2024 | #ifdef HAVE_PERCPU_SB | |
2025 | /* | |
2026 | * Per-cpu incore superblock counters | |
2027 | * | |
2028 | * Simple concept, difficult implementation | |
2029 | * | |
2030 | * Basically, replace the incore superblock counters with a distributed per cpu | |
2031 | * counter for contended fields (e.g. free block count). | |
2032 | * | |
2033 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
2034 | * hence needs to be accurately read when we are running low on space. Hence | |
2035 | * there is a method to enable and disable the per-cpu counters based on how | |
2036 | * much "stuff" is available in them. | |
2037 | * | |
2038 | * Basically, a counter is enabled if there is enough free resource to justify | |
2039 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2040 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2041 | * re-distribute the available resources. | |
2042 | * | |
2043 | * If, once we redistributed the available resources, we still get a failure, | |
2044 | * we disable the per-cpu counter and go through the slow path. | |
2045 | * | |
2046 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2047 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2048 | * the global superblock. We do this after disabling the counter to prevent |
2049 | * more threads from queueing up on the counter. | |
2050 | * | |
2051 | * Essentially, this means that we still need a lock in the fast path to enable | |
2052 | * synchronisation between the global counters and the per-cpu counters. This | |
2053 | * is not a problem because the lock will be local to a CPU almost all the time | |
2054 | * and have little contention except when we get to ENOSPC conditions. | |
2055 | * | |
2056 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2057 | * path while we do things like enabling and disabling counters and | |
2058 | * synchronising the counters. | |
2059 | * | |
2060 | * Locking rules: | |
2061 | * | |
3685c2a1 | 2062 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2063 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2064 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2065 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2066 | * 5. modifying global counters requires holding m_sb_lock |
2067 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2068 | * and _none_ of the per-cpu locks. |
2069 | * | |
2070 | * Disabled counters are only ever re-enabled by a balance operation | |
2071 | * that results in more free resources per CPU than a given threshold. | |
2072 | * To ensure counters don't remain disabled, they are rebalanced when | |
2073 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2074 | * is present to prevent thrashing). | |
e8234a68 DC |
2075 | */ |
2076 | ||
5a67e4c5 | 2077 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2078 | /* |
2079 | * hot-plug CPU notifier support. | |
8d280b98 | 2080 | * |
5a67e4c5 CS |
2081 | * We need a notifier per filesystem as we need to be able to identify |
2082 | * the filesystem to balance the counters out. This is achieved by | |
2083 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2084 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2085 | */ |
e8234a68 DC |
2086 | STATIC int |
2087 | xfs_icsb_cpu_notify( | |
2088 | struct notifier_block *nfb, | |
2089 | unsigned long action, | |
2090 | void *hcpu) | |
2091 | { | |
2092 | xfs_icsb_cnts_t *cntp; | |
2093 | xfs_mount_t *mp; | |
e8234a68 DC |
2094 | |
2095 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2096 | cntp = (xfs_icsb_cnts_t *) | |
2097 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2098 | switch (action) { | |
2099 | case CPU_UP_PREPARE: | |
8bb78442 | 2100 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2101 | /* Easy Case - initialize the area and locks, and |
2102 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2103 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2104 | break; |
2105 | case CPU_ONLINE: | |
8bb78442 | 2106 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2107 | xfs_icsb_lock(mp); |
45af6c6d CH |
2108 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2109 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2110 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2111 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2112 | break; |
2113 | case CPU_DEAD: | |
8bb78442 | 2114 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2115 | /* Disable all the counters, then fold the dead cpu's |
2116 | * count into the total on the global superblock and | |
2117 | * re-enable the counters. */ | |
03135cf7 | 2118 | xfs_icsb_lock(mp); |
3685c2a1 | 2119 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2120 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2121 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2122 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2123 | ||
2124 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2125 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2126 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2127 | ||
01e1b69c | 2128 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2129 | |
45af6c6d CH |
2130 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2131 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2132 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2133 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2134 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2135 | break; |
2136 | } | |
2137 | ||
2138 | return NOTIFY_OK; | |
2139 | } | |
5a67e4c5 | 2140 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2141 | |
8d280b98 DC |
2142 | int |
2143 | xfs_icsb_init_counters( | |
2144 | xfs_mount_t *mp) | |
2145 | { | |
2146 | xfs_icsb_cnts_t *cntp; | |
2147 | int i; | |
2148 | ||
2149 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2150 | if (mp->m_sb_cnts == NULL) | |
2151 | return -ENOMEM; | |
2152 | ||
5a67e4c5 | 2153 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2154 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2155 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2156 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2157 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2158 | |
8d280b98 DC |
2159 | for_each_online_cpu(i) { |
2160 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2161 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2162 | } |
20b64285 DC |
2163 | |
2164 | mutex_init(&mp->m_icsb_mutex); | |
2165 | ||
8d280b98 DC |
2166 | /* |
2167 | * start with all counters disabled so that the | |
2168 | * initial balance kicks us off correctly | |
2169 | */ | |
2170 | mp->m_icsb_counters = -1; | |
2171 | return 0; | |
2172 | } | |
2173 | ||
5478eead LM |
2174 | void |
2175 | xfs_icsb_reinit_counters( | |
2176 | xfs_mount_t *mp) | |
2177 | { | |
2178 | xfs_icsb_lock(mp); | |
2179 | /* | |
2180 | * start with all counters disabled so that the | |
2181 | * initial balance kicks us off correctly | |
2182 | */ | |
2183 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2184 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2185 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2186 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2187 | xfs_icsb_unlock(mp); |
2188 | } | |
2189 | ||
c962fb79 | 2190 | void |
8d280b98 DC |
2191 | xfs_icsb_destroy_counters( |
2192 | xfs_mount_t *mp) | |
2193 | { | |
e8234a68 | 2194 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2195 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2196 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2197 | } |
03135cf7 | 2198 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2199 | } |
2200 | ||
b8f82a4a | 2201 | STATIC void |
01e1b69c DC |
2202 | xfs_icsb_lock_cntr( |
2203 | xfs_icsb_cnts_t *icsbp) | |
2204 | { | |
2205 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2206 | ndelay(1000); | |
2207 | } | |
2208 | } | |
2209 | ||
b8f82a4a | 2210 | STATIC void |
01e1b69c DC |
2211 | xfs_icsb_unlock_cntr( |
2212 | xfs_icsb_cnts_t *icsbp) | |
2213 | { | |
2214 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2215 | } | |
2216 | ||
8d280b98 | 2217 | |
b8f82a4a | 2218 | STATIC void |
8d280b98 DC |
2219 | xfs_icsb_lock_all_counters( |
2220 | xfs_mount_t *mp) | |
2221 | { | |
2222 | xfs_icsb_cnts_t *cntp; | |
2223 | int i; | |
2224 | ||
2225 | for_each_online_cpu(i) { | |
2226 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2227 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2228 | } |
2229 | } | |
2230 | ||
b8f82a4a | 2231 | STATIC void |
8d280b98 DC |
2232 | xfs_icsb_unlock_all_counters( |
2233 | xfs_mount_t *mp) | |
2234 | { | |
2235 | xfs_icsb_cnts_t *cntp; | |
2236 | int i; | |
2237 | ||
2238 | for_each_online_cpu(i) { | |
2239 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2240 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2241 | } |
2242 | } | |
2243 | ||
2244 | STATIC void | |
2245 | xfs_icsb_count( | |
2246 | xfs_mount_t *mp, | |
2247 | xfs_icsb_cnts_t *cnt, | |
2248 | int flags) | |
2249 | { | |
2250 | xfs_icsb_cnts_t *cntp; | |
2251 | int i; | |
2252 | ||
2253 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2254 | ||
2255 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2256 | xfs_icsb_lock_all_counters(mp); | |
2257 | ||
2258 | for_each_online_cpu(i) { | |
2259 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2260 | cnt->icsb_icount += cntp->icsb_icount; | |
2261 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2262 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2263 | } | |
2264 | ||
2265 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2266 | xfs_icsb_unlock_all_counters(mp); | |
2267 | } | |
2268 | ||
2269 | STATIC int | |
2270 | xfs_icsb_counter_disabled( | |
2271 | xfs_mount_t *mp, | |
2272 | xfs_sb_field_t field) | |
2273 | { | |
2274 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2275 | return test_bit(field, &mp->m_icsb_counters); | |
2276 | } | |
2277 | ||
36fbe6e6 | 2278 | STATIC void |
8d280b98 DC |
2279 | xfs_icsb_disable_counter( |
2280 | xfs_mount_t *mp, | |
2281 | xfs_sb_field_t field) | |
2282 | { | |
2283 | xfs_icsb_cnts_t cnt; | |
2284 | ||
2285 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2286 | ||
20b64285 DC |
2287 | /* |
2288 | * If we are already disabled, then there is nothing to do | |
2289 | * here. We check before locking all the counters to avoid | |
2290 | * the expensive lock operation when being called in the | |
2291 | * slow path and the counter is already disabled. This is | |
2292 | * safe because the only time we set or clear this state is under | |
2293 | * the m_icsb_mutex. | |
2294 | */ | |
2295 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2296 | return; |
20b64285 | 2297 | |
8d280b98 DC |
2298 | xfs_icsb_lock_all_counters(mp); |
2299 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2300 | /* drain back to superblock */ | |
2301 | ||
ce46193b | 2302 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2303 | switch(field) { |
2304 | case XFS_SBS_ICOUNT: | |
2305 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2306 | break; | |
2307 | case XFS_SBS_IFREE: | |
2308 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2309 | break; | |
2310 | case XFS_SBS_FDBLOCKS: | |
2311 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2312 | break; | |
2313 | default: | |
2314 | BUG(); | |
2315 | } | |
2316 | } | |
2317 | ||
2318 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2319 | } |
2320 | ||
2321 | STATIC void | |
2322 | xfs_icsb_enable_counter( | |
2323 | xfs_mount_t *mp, | |
2324 | xfs_sb_field_t field, | |
2325 | uint64_t count, | |
2326 | uint64_t resid) | |
2327 | { | |
2328 | xfs_icsb_cnts_t *cntp; | |
2329 | int i; | |
2330 | ||
2331 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2332 | ||
2333 | xfs_icsb_lock_all_counters(mp); | |
2334 | for_each_online_cpu(i) { | |
2335 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2336 | switch (field) { | |
2337 | case XFS_SBS_ICOUNT: | |
2338 | cntp->icsb_icount = count + resid; | |
2339 | break; | |
2340 | case XFS_SBS_IFREE: | |
2341 | cntp->icsb_ifree = count + resid; | |
2342 | break; | |
2343 | case XFS_SBS_FDBLOCKS: | |
2344 | cntp->icsb_fdblocks = count + resid; | |
2345 | break; | |
2346 | default: | |
2347 | BUG(); | |
2348 | break; | |
2349 | } | |
2350 | resid = 0; | |
2351 | } | |
2352 | clear_bit(field, &mp->m_icsb_counters); | |
2353 | xfs_icsb_unlock_all_counters(mp); | |
2354 | } | |
2355 | ||
dbcabad1 | 2356 | void |
d4d90b57 | 2357 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2358 | xfs_mount_t *mp, |
2359 | int flags) | |
2360 | { | |
2361 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2362 | |
8d280b98 DC |
2363 | xfs_icsb_count(mp, &cnt, flags); |
2364 | ||
8d280b98 DC |
2365 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2366 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2367 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2368 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2369 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2370 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2371 | } |
2372 | ||
2373 | /* | |
2374 | * Accurate update of per-cpu counters to incore superblock | |
2375 | */ | |
d4d90b57 | 2376 | void |
8d280b98 | 2377 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2378 | xfs_mount_t *mp, |
2379 | int flags) | |
8d280b98 | 2380 | { |
d4d90b57 CH |
2381 | spin_lock(&mp->m_sb_lock); |
2382 | xfs_icsb_sync_counters_locked(mp, flags); | |
2383 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2384 | } |
2385 | ||
2386 | /* | |
2387 | * Balance and enable/disable counters as necessary. | |
2388 | * | |
20b64285 DC |
2389 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2390 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2391 | * free blocks is something far enough zero that we aren't going thrash when we | |
2392 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2393 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2394 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2395 | * reenabled. | |
2396 | * | |
2397 | * Note that we can be called when counters are already disabled. | |
2398 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2399 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2400 | */ |
20b64285 DC |
2401 | |
2402 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2403 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2404 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2405 | STATIC void |
45af6c6d | 2406 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2407 | xfs_mount_t *mp, |
2408 | xfs_sb_field_t field, | |
20b64285 | 2409 | int min_per_cpu) |
8d280b98 | 2410 | { |
6fdf8ccc | 2411 | uint64_t count, resid; |
8d280b98 | 2412 | int weight = num_online_cpus(); |
20b64285 | 2413 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2414 | |
8d280b98 DC |
2415 | /* disable counter and sync counter */ |
2416 | xfs_icsb_disable_counter(mp, field); | |
2417 | ||
2418 | /* update counters - first CPU gets residual*/ | |
2419 | switch (field) { | |
2420 | case XFS_SBS_ICOUNT: | |
2421 | count = mp->m_sb.sb_icount; | |
2422 | resid = do_div(count, weight); | |
20b64285 | 2423 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2424 | return; |
8d280b98 DC |
2425 | break; |
2426 | case XFS_SBS_IFREE: | |
2427 | count = mp->m_sb.sb_ifree; | |
2428 | resid = do_div(count, weight); | |
20b64285 | 2429 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2430 | return; |
8d280b98 DC |
2431 | break; |
2432 | case XFS_SBS_FDBLOCKS: | |
2433 | count = mp->m_sb.sb_fdblocks; | |
2434 | resid = do_div(count, weight); | |
20b64285 | 2435 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2436 | return; |
8d280b98 DC |
2437 | break; |
2438 | default: | |
2439 | BUG(); | |
6fdf8ccc | 2440 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2441 | break; |
2442 | } | |
2443 | ||
2444 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2445 | } |
2446 | ||
2447 | STATIC void | |
2448 | xfs_icsb_balance_counter( | |
2449 | xfs_mount_t *mp, | |
2450 | xfs_sb_field_t fields, | |
2451 | int min_per_cpu) | |
2452 | { | |
2453 | spin_lock(&mp->m_sb_lock); | |
2454 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2455 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2456 | } |
2457 | ||
1b040712 | 2458 | int |
20b64285 | 2459 | xfs_icsb_modify_counters( |
8d280b98 DC |
2460 | xfs_mount_t *mp, |
2461 | xfs_sb_field_t field, | |
20f4ebf2 | 2462 | int64_t delta, |
20b64285 | 2463 | int rsvd) |
8d280b98 DC |
2464 | { |
2465 | xfs_icsb_cnts_t *icsbp; | |
2466 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2467 | int ret = 0; |
8d280b98 | 2468 | |
20b64285 | 2469 | might_sleep(); |
8d280b98 | 2470 | again: |
7a9e02d6 CL |
2471 | preempt_disable(); |
2472 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2473 | |
2474 | /* | |
2475 | * if the counter is disabled, go to slow path | |
2476 | */ | |
8d280b98 DC |
2477 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2478 | goto slow_path; | |
20b64285 DC |
2479 | xfs_icsb_lock_cntr(icsbp); |
2480 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2481 | xfs_icsb_unlock_cntr(icsbp); | |
2482 | goto slow_path; | |
2483 | } | |
8d280b98 DC |
2484 | |
2485 | switch (field) { | |
2486 | case XFS_SBS_ICOUNT: | |
2487 | lcounter = icsbp->icsb_icount; | |
2488 | lcounter += delta; | |
2489 | if (unlikely(lcounter < 0)) | |
20b64285 | 2490 | goto balance_counter; |
8d280b98 DC |
2491 | icsbp->icsb_icount = lcounter; |
2492 | break; | |
2493 | ||
2494 | case XFS_SBS_IFREE: | |
2495 | lcounter = icsbp->icsb_ifree; | |
2496 | lcounter += delta; | |
2497 | if (unlikely(lcounter < 0)) | |
20b64285 | 2498 | goto balance_counter; |
8d280b98 DC |
2499 | icsbp->icsb_ifree = lcounter; |
2500 | break; | |
2501 | ||
2502 | case XFS_SBS_FDBLOCKS: | |
2503 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2504 | ||
4be536de | 2505 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2506 | lcounter += delta; |
2507 | if (unlikely(lcounter < 0)) | |
20b64285 | 2508 | goto balance_counter; |
4be536de | 2509 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2510 | break; |
2511 | default: | |
2512 | BUG(); | |
2513 | break; | |
2514 | } | |
01e1b69c | 2515 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2516 | preempt_enable(); |
8d280b98 DC |
2517 | return 0; |
2518 | ||
8d280b98 | 2519 | slow_path: |
7a9e02d6 | 2520 | preempt_enable(); |
8d280b98 | 2521 | |
20b64285 DC |
2522 | /* |
2523 | * serialise with a mutex so we don't burn lots of cpu on | |
2524 | * the superblock lock. We still need to hold the superblock | |
2525 | * lock, however, when we modify the global structures. | |
2526 | */ | |
03135cf7 | 2527 | xfs_icsb_lock(mp); |
20b64285 DC |
2528 | |
2529 | /* | |
2530 | * Now running atomically. | |
2531 | * | |
2532 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2533 | * Drop the lock and try again in the fast path.... | |
2534 | */ | |
2535 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2536 | xfs_icsb_unlock(mp); |
8d280b98 | 2537 | goto again; |
8d280b98 DC |
2538 | } |
2539 | ||
20b64285 DC |
2540 | /* |
2541 | * The counter is currently disabled. Because we are | |
2542 | * running atomically here, we know a rebalance cannot | |
2543 | * be in progress. Hence we can go straight to operating | |
2544 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2545 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2546 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2547 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2548 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2549 | * directly on the global counters. | |
2550 | */ | |
3685c2a1 | 2551 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2552 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2553 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2554 | |
20b64285 DC |
2555 | /* |
2556 | * Now that we've modified the global superblock, we | |
2557 | * may be able to re-enable the distributed counters | |
2558 | * (e.g. lots of space just got freed). After that | |
2559 | * we are done. | |
2560 | */ | |
2561 | if (ret != ENOSPC) | |
45af6c6d | 2562 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2563 | xfs_icsb_unlock(mp); |
8d280b98 | 2564 | return ret; |
8d280b98 | 2565 | |
20b64285 DC |
2566 | balance_counter: |
2567 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2568 | preempt_enable(); |
8d280b98 | 2569 | |
20b64285 DC |
2570 | /* |
2571 | * We may have multiple threads here if multiple per-cpu | |
2572 | * counters run dry at the same time. This will mean we can | |
2573 | * do more balances than strictly necessary but it is not | |
2574 | * the common slowpath case. | |
2575 | */ | |
03135cf7 | 2576 | xfs_icsb_lock(mp); |
20b64285 DC |
2577 | |
2578 | /* | |
2579 | * running atomically. | |
2580 | * | |
2581 | * This will leave the counter in the correct state for future | |
2582 | * accesses. After the rebalance, we simply try again and our retry | |
2583 | * will either succeed through the fast path or slow path without | |
2584 | * another balance operation being required. | |
2585 | */ | |
45af6c6d | 2586 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2587 | xfs_icsb_unlock(mp); |
20b64285 | 2588 | goto again; |
8d280b98 | 2589 | } |
20b64285 | 2590 | |
8d280b98 | 2591 | #endif |