]>
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 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
1da177e4 | 30 | #include "xfs_bmap_btree.h" |
a844f451 | 31 | #include "xfs_alloc_btree.h" |
1da177e4 | 32 | #include "xfs_ialloc_btree.h" |
1da177e4 | 33 | #include "xfs_dir2_sf.h" |
a844f451 | 34 | #include "xfs_attr_sf.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_btree.h" |
38 | #include "xfs_ialloc.h" | |
1da177e4 LT |
39 | #include "xfs_alloc.h" |
40 | #include "xfs_rtalloc.h" | |
41 | #include "xfs_bmap.h" | |
42 | #include "xfs_error.h" | |
1da177e4 LT |
43 | #include "xfs_rw.h" |
44 | #include "xfs_quota.h" | |
45 | #include "xfs_fsops.h" | |
43355099 | 46 | #include "xfs_utils.h" |
0b1b213f CH |
47 | #include "xfs_trace.h" |
48 | ||
1da177e4 | 49 | |
ba0f32d4 | 50 | STATIC void xfs_unmountfs_wait(xfs_mount_t *); |
1da177e4 | 51 | |
8d280b98 DC |
52 | |
53 | #ifdef HAVE_PERCPU_SB | |
20f4ebf2 | 54 | STATIC void xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t, |
45af6c6d CH |
55 | int); |
56 | STATIC void xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t, | |
57 | int); | |
8d280b98 | 58 | STATIC int xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t, |
20f4ebf2 | 59 | int64_t, int); |
36fbe6e6 | 60 | STATIC void xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t); |
8d280b98 DC |
61 | |
62 | #else | |
63 | ||
45af6c6d CH |
64 | #define xfs_icsb_balance_counter(mp, a, b) do { } while (0) |
65 | #define xfs_icsb_balance_counter_locked(mp, a, b) do { } while (0) | |
8d280b98 | 66 | #define xfs_icsb_modify_counters(mp, a, b, c) do { } while (0) |
8d280b98 DC |
67 | |
68 | #endif | |
69 | ||
1df84c93 | 70 | static const struct { |
8d280b98 DC |
71 | short offset; |
72 | short type; /* 0 = integer | |
73 | * 1 = binary / string (no translation) | |
74 | */ | |
1da177e4 LT |
75 | } xfs_sb_info[] = { |
76 | { offsetof(xfs_sb_t, sb_magicnum), 0 }, | |
77 | { offsetof(xfs_sb_t, sb_blocksize), 0 }, | |
78 | { offsetof(xfs_sb_t, sb_dblocks), 0 }, | |
79 | { offsetof(xfs_sb_t, sb_rblocks), 0 }, | |
80 | { offsetof(xfs_sb_t, sb_rextents), 0 }, | |
81 | { offsetof(xfs_sb_t, sb_uuid), 1 }, | |
82 | { offsetof(xfs_sb_t, sb_logstart), 0 }, | |
83 | { offsetof(xfs_sb_t, sb_rootino), 0 }, | |
84 | { offsetof(xfs_sb_t, sb_rbmino), 0 }, | |
85 | { offsetof(xfs_sb_t, sb_rsumino), 0 }, | |
86 | { offsetof(xfs_sb_t, sb_rextsize), 0 }, | |
87 | { offsetof(xfs_sb_t, sb_agblocks), 0 }, | |
88 | { offsetof(xfs_sb_t, sb_agcount), 0 }, | |
89 | { offsetof(xfs_sb_t, sb_rbmblocks), 0 }, | |
90 | { offsetof(xfs_sb_t, sb_logblocks), 0 }, | |
91 | { offsetof(xfs_sb_t, sb_versionnum), 0 }, | |
92 | { offsetof(xfs_sb_t, sb_sectsize), 0 }, | |
93 | { offsetof(xfs_sb_t, sb_inodesize), 0 }, | |
94 | { offsetof(xfs_sb_t, sb_inopblock), 0 }, | |
95 | { offsetof(xfs_sb_t, sb_fname[0]), 1 }, | |
96 | { offsetof(xfs_sb_t, sb_blocklog), 0 }, | |
97 | { offsetof(xfs_sb_t, sb_sectlog), 0 }, | |
98 | { offsetof(xfs_sb_t, sb_inodelog), 0 }, | |
99 | { offsetof(xfs_sb_t, sb_inopblog), 0 }, | |
100 | { offsetof(xfs_sb_t, sb_agblklog), 0 }, | |
101 | { offsetof(xfs_sb_t, sb_rextslog), 0 }, | |
102 | { offsetof(xfs_sb_t, sb_inprogress), 0 }, | |
103 | { offsetof(xfs_sb_t, sb_imax_pct), 0 }, | |
104 | { offsetof(xfs_sb_t, sb_icount), 0 }, | |
105 | { offsetof(xfs_sb_t, sb_ifree), 0 }, | |
106 | { offsetof(xfs_sb_t, sb_fdblocks), 0 }, | |
107 | { offsetof(xfs_sb_t, sb_frextents), 0 }, | |
108 | { offsetof(xfs_sb_t, sb_uquotino), 0 }, | |
109 | { offsetof(xfs_sb_t, sb_gquotino), 0 }, | |
110 | { offsetof(xfs_sb_t, sb_qflags), 0 }, | |
111 | { offsetof(xfs_sb_t, sb_flags), 0 }, | |
112 | { offsetof(xfs_sb_t, sb_shared_vn), 0 }, | |
113 | { offsetof(xfs_sb_t, sb_inoalignmt), 0 }, | |
114 | { offsetof(xfs_sb_t, sb_unit), 0 }, | |
115 | { offsetof(xfs_sb_t, sb_width), 0 }, | |
116 | { offsetof(xfs_sb_t, sb_dirblklog), 0 }, | |
117 | { offsetof(xfs_sb_t, sb_logsectlog), 0 }, | |
118 | { offsetof(xfs_sb_t, sb_logsectsize),0 }, | |
119 | { offsetof(xfs_sb_t, sb_logsunit), 0 }, | |
120 | { offsetof(xfs_sb_t, sb_features2), 0 }, | |
ee1c0908 | 121 | { offsetof(xfs_sb_t, sb_bad_features2), 0 }, |
1da177e4 LT |
122 | { sizeof(xfs_sb_t), 0 } |
123 | }; | |
124 | ||
27174203 CH |
125 | static DEFINE_MUTEX(xfs_uuid_table_mutex); |
126 | static int xfs_uuid_table_size; | |
127 | static uuid_t *xfs_uuid_table; | |
128 | ||
129 | /* | |
130 | * See if the UUID is unique among mounted XFS filesystems. | |
131 | * Mount fails if UUID is nil or a FS with the same UUID is already mounted. | |
132 | */ | |
133 | STATIC int | |
134 | xfs_uuid_mount( | |
135 | struct xfs_mount *mp) | |
136 | { | |
137 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
138 | int hole, i; | |
139 | ||
140 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
141 | return 0; | |
142 | ||
143 | if (uuid_is_nil(uuid)) { | |
144 | cmn_err(CE_WARN, | |
145 | "XFS: Filesystem %s has nil UUID - can't mount", | |
146 | mp->m_fsname); | |
147 | return XFS_ERROR(EINVAL); | |
148 | } | |
149 | ||
150 | mutex_lock(&xfs_uuid_table_mutex); | |
151 | for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) { | |
152 | if (uuid_is_nil(&xfs_uuid_table[i])) { | |
153 | hole = i; | |
154 | continue; | |
155 | } | |
156 | if (uuid_equal(uuid, &xfs_uuid_table[i])) | |
157 | goto out_duplicate; | |
158 | } | |
159 | ||
160 | if (hole < 0) { | |
161 | xfs_uuid_table = kmem_realloc(xfs_uuid_table, | |
162 | (xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table), | |
163 | xfs_uuid_table_size * sizeof(*xfs_uuid_table), | |
164 | KM_SLEEP); | |
165 | hole = xfs_uuid_table_size++; | |
166 | } | |
167 | xfs_uuid_table[hole] = *uuid; | |
168 | mutex_unlock(&xfs_uuid_table_mutex); | |
169 | ||
170 | return 0; | |
171 | ||
172 | out_duplicate: | |
173 | mutex_unlock(&xfs_uuid_table_mutex); | |
174 | cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount", | |
175 | mp->m_fsname); | |
176 | return XFS_ERROR(EINVAL); | |
177 | } | |
178 | ||
179 | STATIC void | |
180 | xfs_uuid_unmount( | |
181 | struct xfs_mount *mp) | |
182 | { | |
183 | uuid_t *uuid = &mp->m_sb.sb_uuid; | |
184 | int i; | |
185 | ||
186 | if (mp->m_flags & XFS_MOUNT_NOUUID) | |
187 | return; | |
188 | ||
189 | mutex_lock(&xfs_uuid_table_mutex); | |
190 | for (i = 0; i < xfs_uuid_table_size; i++) { | |
191 | if (uuid_is_nil(&xfs_uuid_table[i])) | |
192 | continue; | |
193 | if (!uuid_equal(uuid, &xfs_uuid_table[i])) | |
194 | continue; | |
195 | memset(&xfs_uuid_table[i], 0, sizeof(uuid_t)); | |
196 | break; | |
197 | } | |
198 | ASSERT(i < xfs_uuid_table_size); | |
199 | mutex_unlock(&xfs_uuid_table_mutex); | |
200 | } | |
201 | ||
202 | ||
1da177e4 LT |
203 | /* |
204 | * Free up the resources associated with a mount structure. Assume that | |
205 | * the structure was initially zeroed, so we can tell which fields got | |
206 | * initialized. | |
207 | */ | |
c962fb79 | 208 | STATIC void |
ff4f038c | 209 | xfs_free_perag( |
745f6919 | 210 | xfs_mount_t *mp) |
1da177e4 | 211 | { |
1c1c6ebc DC |
212 | xfs_agnumber_t agno; |
213 | struct xfs_perag *pag; | |
214 | ||
215 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
216 | spin_lock(&mp->m_perag_lock); | |
217 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
218 | spin_unlock(&mp->m_perag_lock); | |
219 | ASSERT(pag); | |
220 | kmem_free(pag->pagb_list); | |
221 | kmem_free(pag); | |
1da177e4 | 222 | } |
1da177e4 LT |
223 | } |
224 | ||
4cc929ee NS |
225 | /* |
226 | * Check size of device based on the (data/realtime) block count. | |
227 | * Note: this check is used by the growfs code as well as mount. | |
228 | */ | |
229 | int | |
230 | xfs_sb_validate_fsb_count( | |
231 | xfs_sb_t *sbp, | |
232 | __uint64_t nblocks) | |
233 | { | |
234 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
235 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
236 | ||
237 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
238 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
239 | return E2BIG; | |
240 | #else /* Limited by UINT_MAX of sectors */ | |
241 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
242 | return E2BIG; | |
243 | #endif | |
244 | return 0; | |
245 | } | |
1da177e4 LT |
246 | |
247 | /* | |
248 | * Check the validity of the SB found. | |
249 | */ | |
250 | STATIC int | |
251 | xfs_mount_validate_sb( | |
252 | xfs_mount_t *mp, | |
764d1f89 NS |
253 | xfs_sb_t *sbp, |
254 | int flags) | |
1da177e4 LT |
255 | { |
256 | /* | |
257 | * If the log device and data device have the | |
258 | * same device number, the log is internal. | |
259 | * Consequently, the sb_logstart should be non-zero. If | |
260 | * we have a zero sb_logstart in this case, we may be trying to mount | |
261 | * a volume filesystem in a non-volume manner. | |
262 | */ | |
263 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 264 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
265 | return XFS_ERROR(EWRONGFS); |
266 | } | |
267 | ||
62118709 | 268 | if (!xfs_sb_good_version(sbp)) { |
764d1f89 | 269 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
270 | return XFS_ERROR(EWRONGFS); |
271 | } | |
272 | ||
273 | if (unlikely( | |
274 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
275 | xfs_fs_mount_cmn_err(flags, |
276 | "filesystem is marked as having an external log; " | |
277 | "specify logdev on the\nmount command line."); | |
278 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
279 | } |
280 | ||
281 | if (unlikely( | |
282 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
283 | xfs_fs_mount_cmn_err(flags, |
284 | "filesystem is marked as having an internal log; " | |
285 | "do not specify logdev on\nthe mount command line."); | |
286 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
287 | } |
288 | ||
289 | /* | |
290 | * More sanity checking. These were stolen directly from | |
291 | * xfs_repair. | |
292 | */ | |
293 | if (unlikely( | |
294 | sbp->sb_agcount <= 0 || | |
295 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
296 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
297 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
298 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 299 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
300 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
301 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
302 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
303 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 304 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
305 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
306 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
307 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
308 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 309 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 310 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
311 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
312 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 313 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 314 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
315 | return XFS_ERROR(EFSCORRUPTED); |
316 | } | |
317 | ||
318 | /* | |
319 | * Sanity check AG count, size fields against data size field | |
320 | */ | |
321 | if (unlikely( | |
322 | sbp->sb_dblocks == 0 || | |
323 | sbp->sb_dblocks > | |
324 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
325 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
326 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 327 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
328 | return XFS_ERROR(EFSCORRUPTED); |
329 | } | |
330 | ||
2edbddd5 LM |
331 | /* |
332 | * Until this is fixed only page-sized or smaller data blocks work. | |
333 | */ | |
334 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
335 | xfs_fs_mount_cmn_err(flags, | |
336 | "file system with blocksize %d bytes", | |
337 | sbp->sb_blocksize); | |
338 | xfs_fs_mount_cmn_err(flags, | |
339 | "only pagesize (%ld) or less will currently work.", | |
340 | PAGE_SIZE); | |
341 | return XFS_ERROR(ENOSYS); | |
342 | } | |
343 | ||
1a5902c5 CH |
344 | /* |
345 | * Currently only very few inode sizes are supported. | |
346 | */ | |
347 | switch (sbp->sb_inodesize) { | |
348 | case 256: | |
349 | case 512: | |
350 | case 1024: | |
351 | case 2048: | |
352 | break; | |
353 | default: | |
354 | xfs_fs_mount_cmn_err(flags, | |
355 | "inode size of %d bytes not supported", | |
356 | sbp->sb_inodesize); | |
357 | return XFS_ERROR(ENOSYS); | |
358 | } | |
359 | ||
4cc929ee NS |
360 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
361 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
764d1f89 NS |
362 | xfs_fs_mount_cmn_err(flags, |
363 | "file system too large to be mounted on this system."); | |
1da177e4 LT |
364 | return XFS_ERROR(E2BIG); |
365 | } | |
366 | ||
367 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 368 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
369 | return XFS_ERROR(EFSCORRUPTED); |
370 | } | |
371 | ||
de20614b NS |
372 | /* |
373 | * Version 1 directory format has never worked on Linux. | |
374 | */ | |
62118709 | 375 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
764d1f89 NS |
376 | xfs_fs_mount_cmn_err(flags, |
377 | "file system using version 1 directory format"); | |
de20614b NS |
378 | return XFS_ERROR(ENOSYS); |
379 | } | |
380 | ||
1da177e4 LT |
381 | return 0; |
382 | } | |
383 | ||
da353b0d DC |
384 | STATIC void |
385 | xfs_initialize_perag_icache( | |
386 | xfs_perag_t *pag) | |
387 | { | |
388 | if (!pag->pag_ici_init) { | |
389 | rwlock_init(&pag->pag_ici_lock); | |
390 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); | |
391 | pag->pag_ici_init = 1; | |
392 | } | |
393 | } | |
394 | ||
1c1c6ebc | 395 | int |
c11e2c36 | 396 | xfs_initialize_perag( |
c11e2c36 | 397 | xfs_mount_t *mp, |
1c1c6ebc DC |
398 | xfs_agnumber_t agcount, |
399 | xfs_agnumber_t *maxagi) | |
1da177e4 LT |
400 | { |
401 | xfs_agnumber_t index, max_metadata; | |
402 | xfs_perag_t *pag; | |
403 | xfs_agino_t agino; | |
404 | xfs_ino_t ino; | |
405 | xfs_sb_t *sbp = &mp->m_sb; | |
406 | xfs_ino_t max_inum = XFS_MAXINUMBER_32; | |
407 | ||
408 | /* Check to see if the filesystem can overflow 32 bit inodes */ | |
409 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); | |
410 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
411 | ||
1c1c6ebc DC |
412 | /* |
413 | * Walk the current per-ag tree so we don't try to initialise AGs | |
414 | * that already exist (growfs case). Allocate and insert all the | |
415 | * AGs we don't find ready for initialisation. | |
416 | */ | |
417 | for (index = 0; index < agcount; index++) { | |
418 | pag = xfs_perag_get(mp, index); | |
419 | if (pag) { | |
420 | xfs_perag_put(pag); | |
421 | continue; | |
422 | } | |
423 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); | |
424 | if (!pag) | |
425 | return -ENOMEM; | |
426 | if (radix_tree_preload(GFP_NOFS)) | |
427 | return -ENOMEM; | |
428 | spin_lock(&mp->m_perag_lock); | |
429 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
430 | BUG(); | |
431 | spin_unlock(&mp->m_perag_lock); | |
432 | kmem_free(pag); | |
433 | return -EEXIST; | |
434 | } | |
435 | spin_unlock(&mp->m_perag_lock); | |
436 | radix_tree_preload_end(); | |
437 | } | |
438 | ||
1da177e4 LT |
439 | /* Clear the mount flag if no inode can overflow 32 bits |
440 | * on this filesystem, or if specifically requested.. | |
441 | */ | |
bd186aa9 | 442 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) { |
1da177e4 LT |
443 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
444 | } else { | |
445 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; | |
446 | } | |
447 | ||
448 | /* If we can overflow then setup the ag headers accordingly */ | |
449 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { | |
450 | /* Calculate how much should be reserved for inodes to | |
451 | * meet the max inode percentage. | |
452 | */ | |
453 | if (mp->m_maxicount) { | |
454 | __uint64_t icount; | |
455 | ||
456 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
457 | do_div(icount, 100); | |
458 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 459 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
460 | max_metadata = icount; |
461 | } else { | |
462 | max_metadata = agcount; | |
463 | } | |
464 | for (index = 0; index < agcount; index++) { | |
465 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
466 | if (ino > max_inum) { | |
467 | index++; | |
468 | break; | |
469 | } | |
470 | ||
c41564b5 | 471 | /* This ag is preferred for inodes */ |
44b56e0a | 472 | pag = xfs_perag_get(mp, index); |
1da177e4 LT |
473 | pag->pagi_inodeok = 1; |
474 | if (index < max_metadata) | |
475 | pag->pagf_metadata = 1; | |
da353b0d | 476 | xfs_initialize_perag_icache(pag); |
44b56e0a | 477 | xfs_perag_put(pag); |
1da177e4 LT |
478 | } |
479 | } else { | |
480 | /* Setup default behavior for smaller filesystems */ | |
481 | for (index = 0; index < agcount; index++) { | |
44b56e0a | 482 | pag = xfs_perag_get(mp, index); |
1da177e4 | 483 | pag->pagi_inodeok = 1; |
da353b0d | 484 | xfs_initialize_perag_icache(pag); |
44b56e0a | 485 | xfs_perag_put(pag); |
1da177e4 LT |
486 | } |
487 | } | |
1c1c6ebc DC |
488 | if (maxagi) |
489 | *maxagi = index; | |
490 | return 0; | |
1da177e4 LT |
491 | } |
492 | ||
2bdf7cd0 CH |
493 | void |
494 | xfs_sb_from_disk( | |
495 | xfs_sb_t *to, | |
496 | xfs_dsb_t *from) | |
497 | { | |
498 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
499 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
500 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
501 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
502 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
503 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
504 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
505 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
506 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
507 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
508 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
509 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
510 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
511 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
512 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
513 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
514 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
515 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
516 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
517 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
518 | to->sb_blocklog = from->sb_blocklog; | |
519 | to->sb_sectlog = from->sb_sectlog; | |
520 | to->sb_inodelog = from->sb_inodelog; | |
521 | to->sb_inopblog = from->sb_inopblog; | |
522 | to->sb_agblklog = from->sb_agblklog; | |
523 | to->sb_rextslog = from->sb_rextslog; | |
524 | to->sb_inprogress = from->sb_inprogress; | |
525 | to->sb_imax_pct = from->sb_imax_pct; | |
526 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
527 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
528 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
529 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
530 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
531 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
532 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
533 | to->sb_flags = from->sb_flags; | |
534 | to->sb_shared_vn = from->sb_shared_vn; | |
535 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
536 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
537 | to->sb_width = be32_to_cpu(from->sb_width); | |
538 | to->sb_dirblklog = from->sb_dirblklog; | |
539 | to->sb_logsectlog = from->sb_logsectlog; | |
540 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
541 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
542 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 543 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
544 | } |
545 | ||
1da177e4 | 546 | /* |
2bdf7cd0 | 547 | * Copy in core superblock to ondisk one. |
1da177e4 | 548 | * |
2bdf7cd0 | 549 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
550 | */ |
551 | void | |
2bdf7cd0 CH |
552 | xfs_sb_to_disk( |
553 | xfs_dsb_t *to, | |
554 | xfs_sb_t *from, | |
1da177e4 LT |
555 | __int64_t fields) |
556 | { | |
2bdf7cd0 CH |
557 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
558 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
559 | xfs_sb_field_t f; |
560 | int first; | |
561 | int size; | |
562 | ||
1da177e4 | 563 | ASSERT(fields); |
1da177e4 LT |
564 | if (!fields) |
565 | return; | |
566 | ||
1da177e4 LT |
567 | while (fields) { |
568 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
569 | first = xfs_sb_info[f].offset; | |
570 | size = xfs_sb_info[f + 1].offset - first; | |
571 | ||
572 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
573 | ||
574 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 575 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
576 | } else { |
577 | switch (size) { | |
578 | case 2: | |
2bdf7cd0 CH |
579 | *(__be16 *)(to_ptr + first) = |
580 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
581 | break; |
582 | case 4: | |
2bdf7cd0 CH |
583 | *(__be32 *)(to_ptr + first) = |
584 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
585 | break; |
586 | case 8: | |
2bdf7cd0 CH |
587 | *(__be64 *)(to_ptr + first) = |
588 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
589 | break; |
590 | default: | |
591 | ASSERT(0); | |
592 | } | |
593 | } | |
594 | ||
595 | fields &= ~(1LL << f); | |
596 | } | |
597 | } | |
598 | ||
599 | /* | |
600 | * xfs_readsb | |
601 | * | |
602 | * Does the initial read of the superblock. | |
603 | */ | |
604 | int | |
764d1f89 | 605 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
606 | { |
607 | unsigned int sector_size; | |
608 | unsigned int extra_flags; | |
609 | xfs_buf_t *bp; | |
1da177e4 LT |
610 | int error; |
611 | ||
612 | ASSERT(mp->m_sb_bp == NULL); | |
613 | ASSERT(mp->m_ddev_targp != NULL); | |
614 | ||
615 | /* | |
616 | * Allocate a (locked) buffer to hold the superblock. | |
617 | * This will be kept around at all times to optimize | |
618 | * access to the superblock. | |
619 | */ | |
620 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
621 | extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED; | |
622 | ||
6ad112bf CH |
623 | bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size), |
624 | extra_flags); | |
1da177e4 | 625 | if (!bp || XFS_BUF_ISERROR(bp)) { |
764d1f89 | 626 | xfs_fs_mount_cmn_err(flags, "SB read failed"); |
1da177e4 LT |
627 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
628 | goto fail; | |
629 | } | |
630 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
631 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
632 | ||
633 | /* | |
634 | * Initialize the mount structure from the superblock. | |
635 | * But first do some basic consistency checking. | |
636 | */ | |
2bdf7cd0 | 637 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 638 | |
764d1f89 | 639 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 640 | if (error) { |
764d1f89 | 641 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
1da177e4 LT |
642 | goto fail; |
643 | } | |
644 | ||
645 | /* | |
646 | * We must be able to do sector-sized and sector-aligned IO. | |
647 | */ | |
648 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
649 | xfs_fs_mount_cmn_err(flags, |
650 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
651 | sector_size, mp->m_sb.sb_sectsize); |
652 | error = ENOSYS; | |
653 | goto fail; | |
654 | } | |
655 | ||
656 | /* | |
657 | * If device sector size is smaller than the superblock size, | |
658 | * re-read the superblock so the buffer is correctly sized. | |
659 | */ | |
660 | if (sector_size < mp->m_sb.sb_sectsize) { | |
661 | XFS_BUF_UNMANAGE(bp); | |
662 | xfs_buf_relse(bp); | |
663 | sector_size = mp->m_sb.sb_sectsize; | |
6ad112bf CH |
664 | bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, |
665 | BTOBB(sector_size), extra_flags); | |
1da177e4 | 666 | if (!bp || XFS_BUF_ISERROR(bp)) { |
764d1f89 | 667 | xfs_fs_mount_cmn_err(flags, "SB re-read failed"); |
1da177e4 LT |
668 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
669 | goto fail; | |
670 | } | |
671 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
672 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
673 | } | |
674 | ||
5478eead LM |
675 | /* Initialize per-cpu counters */ |
676 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 677 | |
1da177e4 LT |
678 | mp->m_sb_bp = bp; |
679 | xfs_buf_relse(bp); | |
680 | ASSERT(XFS_BUF_VALUSEMA(bp) > 0); | |
681 | return 0; | |
682 | ||
683 | fail: | |
684 | if (bp) { | |
685 | XFS_BUF_UNMANAGE(bp); | |
686 | xfs_buf_relse(bp); | |
687 | } | |
688 | return error; | |
689 | } | |
690 | ||
691 | ||
692 | /* | |
693 | * xfs_mount_common | |
694 | * | |
695 | * Mount initialization code establishing various mount | |
696 | * fields from the superblock associated with the given | |
697 | * mount structure | |
698 | */ | |
ba0f32d4 | 699 | STATIC void |
1da177e4 LT |
700 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
701 | { | |
1da177e4 | 702 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 703 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
704 | mp->m_maxagi = mp->m_sb.sb_agcount; |
705 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
706 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
707 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
708 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
709 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
710 | mp->m_blockmask = sbp->sb_blocksize - 1; |
711 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
712 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 713 | |
60197e8d CH |
714 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
715 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
716 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
717 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
718 | ||
719 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
720 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
721 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
722 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
723 | ||
724 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
725 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
726 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
727 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
728 | |
729 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
730 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
731 | sbp->sb_inopblock); | |
732 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
733 | } | |
92821e2b DC |
734 | |
735 | /* | |
736 | * xfs_initialize_perag_data | |
737 | * | |
738 | * Read in each per-ag structure so we can count up the number of | |
739 | * allocated inodes, free inodes and used filesystem blocks as this | |
740 | * information is no longer persistent in the superblock. Once we have | |
741 | * this information, write it into the in-core superblock structure. | |
742 | */ | |
743 | STATIC int | |
744 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
745 | { | |
746 | xfs_agnumber_t index; | |
747 | xfs_perag_t *pag; | |
748 | xfs_sb_t *sbp = &mp->m_sb; | |
749 | uint64_t ifree = 0; | |
750 | uint64_t ialloc = 0; | |
751 | uint64_t bfree = 0; | |
752 | uint64_t bfreelst = 0; | |
753 | uint64_t btree = 0; | |
754 | int error; | |
92821e2b DC |
755 | |
756 | for (index = 0; index < agcount; index++) { | |
757 | /* | |
758 | * read the agf, then the agi. This gets us | |
9da096fd | 759 | * all the information we need and populates the |
92821e2b DC |
760 | * per-ag structures for us. |
761 | */ | |
762 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
763 | if (error) | |
764 | return error; | |
765 | ||
766 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
767 | if (error) | |
768 | return error; | |
44b56e0a | 769 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
770 | ifree += pag->pagi_freecount; |
771 | ialloc += pag->pagi_count; | |
772 | bfree += pag->pagf_freeblks; | |
773 | bfreelst += pag->pagf_flcount; | |
774 | btree += pag->pagf_btreeblks; | |
44b56e0a | 775 | xfs_perag_put(pag); |
92821e2b DC |
776 | } |
777 | /* | |
778 | * Overwrite incore superblock counters with just-read data | |
779 | */ | |
3685c2a1 | 780 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
781 | sbp->sb_ifree = ifree; |
782 | sbp->sb_icount = ialloc; | |
783 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 784 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
785 | |
786 | /* Fixup the per-cpu counters as well. */ | |
787 | xfs_icsb_reinit_counters(mp); | |
788 | ||
789 | return 0; | |
790 | } | |
791 | ||
1da177e4 | 792 | /* |
0771fb45 | 793 | * Update alignment values based on mount options and sb values |
1da177e4 | 794 | */ |
0771fb45 | 795 | STATIC int |
7884bc86 | 796 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 797 | { |
1da177e4 | 798 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 799 | |
4249023a | 800 | if (mp->m_dalign) { |
1da177e4 LT |
801 | /* |
802 | * If stripe unit and stripe width are not multiples | |
803 | * of the fs blocksize turn off alignment. | |
804 | */ | |
805 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
806 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
807 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
808 | cmn_err(CE_WARN, | |
809 | "XFS: alignment check 1 failed"); | |
0771fb45 | 810 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
811 | } |
812 | mp->m_dalign = mp->m_swidth = 0; | |
813 | } else { | |
814 | /* | |
815 | * Convert the stripe unit and width to FSBs. | |
816 | */ | |
817 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
818 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
819 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 820 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
821 | } |
822 | xfs_fs_cmn_err(CE_WARN, mp, | |
823 | "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)", | |
824 | mp->m_dalign, mp->m_swidth, | |
825 | sbp->sb_agblocks); | |
826 | ||
827 | mp->m_dalign = 0; | |
828 | mp->m_swidth = 0; | |
829 | } else if (mp->m_dalign) { | |
830 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
831 | } else { | |
832 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
833 | xfs_fs_cmn_err(CE_WARN, mp, | |
834 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
835 | mp->m_dalign, | |
836 | mp->m_blockmask +1); | |
0771fb45 | 837 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
838 | } |
839 | mp->m_swidth = 0; | |
840 | } | |
841 | } | |
842 | ||
843 | /* | |
844 | * Update superblock with new values | |
845 | * and log changes | |
846 | */ | |
62118709 | 847 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
848 | if (sbp->sb_unit != mp->m_dalign) { |
849 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 850 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
851 | } |
852 | if (sbp->sb_width != mp->m_swidth) { | |
853 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 854 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
855 | } |
856 | } | |
857 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 858 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
859 | mp->m_dalign = sbp->sb_unit; |
860 | mp->m_swidth = sbp->sb_width; | |
861 | } | |
862 | ||
0771fb45 ES |
863 | return 0; |
864 | } | |
1da177e4 | 865 | |
0771fb45 ES |
866 | /* |
867 | * Set the maximum inode count for this filesystem | |
868 | */ | |
869 | STATIC void | |
870 | xfs_set_maxicount(xfs_mount_t *mp) | |
871 | { | |
872 | xfs_sb_t *sbp = &(mp->m_sb); | |
873 | __uint64_t icount; | |
1da177e4 | 874 | |
0771fb45 ES |
875 | if (sbp->sb_imax_pct) { |
876 | /* | |
877 | * Make sure the maximum inode count is a multiple | |
878 | * of the units we allocate inodes in. | |
1da177e4 | 879 | */ |
1da177e4 LT |
880 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
881 | do_div(icount, 100); | |
882 | do_div(icount, mp->m_ialloc_blks); | |
883 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
884 | sbp->sb_inopblog; | |
0771fb45 | 885 | } else { |
1da177e4 | 886 | mp->m_maxicount = 0; |
1da177e4 | 887 | } |
0771fb45 ES |
888 | } |
889 | ||
890 | /* | |
891 | * Set the default minimum read and write sizes unless | |
892 | * already specified in a mount option. | |
893 | * We use smaller I/O sizes when the file system | |
894 | * is being used for NFS service (wsync mount option). | |
895 | */ | |
896 | STATIC void | |
897 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
898 | { | |
899 | xfs_sb_t *sbp = &(mp->m_sb); | |
900 | int readio_log, writeio_log; | |
1da177e4 | 901 | |
1da177e4 LT |
902 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
903 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
904 | readio_log = XFS_WSYNC_READIO_LOG; | |
905 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
906 | } else { | |
907 | readio_log = XFS_READIO_LOG_LARGE; | |
908 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
909 | } | |
910 | } else { | |
911 | readio_log = mp->m_readio_log; | |
912 | writeio_log = mp->m_writeio_log; | |
913 | } | |
914 | ||
1da177e4 LT |
915 | if (sbp->sb_blocklog > readio_log) { |
916 | mp->m_readio_log = sbp->sb_blocklog; | |
917 | } else { | |
918 | mp->m_readio_log = readio_log; | |
919 | } | |
920 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
921 | if (sbp->sb_blocklog > writeio_log) { | |
922 | mp->m_writeio_log = sbp->sb_blocklog; | |
923 | } else { | |
924 | mp->m_writeio_log = writeio_log; | |
925 | } | |
926 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 927 | } |
1da177e4 | 928 | |
0771fb45 ES |
929 | /* |
930 | * Set whether we're using inode alignment. | |
931 | */ | |
932 | STATIC void | |
933 | xfs_set_inoalignment(xfs_mount_t *mp) | |
934 | { | |
62118709 | 935 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
936 | mp->m_sb.sb_inoalignmt >= |
937 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
938 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
939 | else | |
940 | mp->m_inoalign_mask = 0; | |
941 | /* | |
942 | * If we are using stripe alignment, check whether | |
943 | * the stripe unit is a multiple of the inode alignment | |
944 | */ | |
945 | if (mp->m_dalign && mp->m_inoalign_mask && | |
946 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
947 | mp->m_sinoalign = mp->m_dalign; | |
948 | else | |
949 | mp->m_sinoalign = 0; | |
0771fb45 ES |
950 | } |
951 | ||
952 | /* | |
953 | * Check that the data (and log if separate) are an ok size. | |
954 | */ | |
955 | STATIC int | |
4249023a | 956 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
957 | { |
958 | xfs_buf_t *bp; | |
959 | xfs_daddr_t d; | |
960 | int error; | |
961 | ||
1da177e4 LT |
962 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
963 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
964 | cmn_err(CE_WARN, "XFS: size check 1 failed"); | |
0771fb45 | 965 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
966 | } |
967 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
968 | d - XFS_FSS_TO_BB(mp, 1), | |
969 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
970 | if (!error) { | |
971 | xfs_buf_relse(bp); | |
972 | } else { | |
973 | cmn_err(CE_WARN, "XFS: size check 2 failed"); | |
0771fb45 | 974 | if (error == ENOSPC) |
1da177e4 | 975 | error = XFS_ERROR(E2BIG); |
0771fb45 | 976 | return error; |
1da177e4 LT |
977 | } |
978 | ||
4249023a | 979 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
980 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
981 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
982 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 983 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
984 | } |
985 | error = xfs_read_buf(mp, mp->m_logdev_targp, | |
986 | d - XFS_FSB_TO_BB(mp, 1), | |
987 | XFS_FSB_TO_BB(mp, 1), 0, &bp); | |
988 | if (!error) { | |
989 | xfs_buf_relse(bp); | |
990 | } else { | |
991 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 992 | if (error == ENOSPC) |
1da177e4 | 993 | error = XFS_ERROR(E2BIG); |
0771fb45 ES |
994 | return error; |
995 | } | |
996 | } | |
997 | return 0; | |
998 | } | |
999 | ||
7d095257 CH |
1000 | /* |
1001 | * Clear the quotaflags in memory and in the superblock. | |
1002 | */ | |
1003 | int | |
1004 | xfs_mount_reset_sbqflags( | |
1005 | struct xfs_mount *mp) | |
1006 | { | |
1007 | int error; | |
1008 | struct xfs_trans *tp; | |
1009 | ||
1010 | mp->m_qflags = 0; | |
1011 | ||
1012 | /* | |
1013 | * It is OK to look at sb_qflags here in mount path, | |
1014 | * without m_sb_lock. | |
1015 | */ | |
1016 | if (mp->m_sb.sb_qflags == 0) | |
1017 | return 0; | |
1018 | spin_lock(&mp->m_sb_lock); | |
1019 | mp->m_sb.sb_qflags = 0; | |
1020 | spin_unlock(&mp->m_sb_lock); | |
1021 | ||
1022 | /* | |
1023 | * If the fs is readonly, let the incore superblock run | |
1024 | * with quotas off but don't flush the update out to disk | |
1025 | */ | |
1026 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1027 | return 0; | |
1028 | ||
1029 | #ifdef QUOTADEBUG | |
1030 | xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes"); | |
1031 | #endif | |
1032 | ||
1033 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); | |
1034 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1035 | XFS_DEFAULT_LOG_COUNT); | |
1036 | if (error) { | |
1037 | xfs_trans_cancel(tp, 0); | |
1038 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1039 | "xfs_mount_reset_sbqflags: Superblock update failed!"); | |
1040 | return error; | |
1041 | } | |
1042 | ||
1043 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1044 | return xfs_trans_commit(tp, 0); | |
1045 | } | |
1046 | ||
0771fb45 | 1047 | /* |
0771fb45 ES |
1048 | * This function does the following on an initial mount of a file system: |
1049 | * - reads the superblock from disk and init the mount struct | |
1050 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1051 | * so we don't mount terabyte filesystems | |
1052 | * - init mount struct realtime fields | |
1053 | * - allocate inode hash table for fs | |
1054 | * - init directory manager | |
1055 | * - perform recovery and init the log manager | |
1056 | */ | |
1057 | int | |
1058 | xfs_mountfs( | |
4249023a | 1059 | xfs_mount_t *mp) |
0771fb45 ES |
1060 | { |
1061 | xfs_sb_t *sbp = &(mp->m_sb); | |
1062 | xfs_inode_t *rip; | |
0771fb45 | 1063 | __uint64_t resblks; |
7d095257 CH |
1064 | uint quotamount = 0; |
1065 | uint quotaflags = 0; | |
0771fb45 ES |
1066 | int error = 0; |
1067 | ||
0771fb45 ES |
1068 | xfs_mount_common(mp, sbp); |
1069 | ||
ee1c0908 | 1070 | /* |
e6957ea4 ES |
1071 | * Check for a mismatched features2 values. Older kernels |
1072 | * read & wrote into the wrong sb offset for sb_features2 | |
1073 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1074 | * when sb_features2 was added, which made older superblock | |
1075 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1076 | * |
e6957ea4 ES |
1077 | * For backwards compatibility, we make both slots equal. |
1078 | * | |
1079 | * If we detect a mismatched field, we OR the set bits into the | |
1080 | * existing features2 field in case it has already been modified; we | |
1081 | * don't want to lose any features. We then update the bad location | |
1082 | * with the ORed value so that older kernels will see any features2 | |
1083 | * flags, and mark the two fields as needing updates once the | |
1084 | * transaction subsystem is online. | |
ee1c0908 | 1085 | */ |
e6957ea4 | 1086 | if (xfs_sb_has_mismatched_features2(sbp)) { |
ee1c0908 DC |
1087 | cmn_err(CE_WARN, |
1088 | "XFS: correcting sb_features alignment problem"); | |
1089 | sbp->sb_features2 |= sbp->sb_bad_features2; | |
e6957ea4 | 1090 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1091 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1092 | |
1093 | /* | |
1094 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1095 | * slot. | |
1096 | */ | |
7c12f296 TS |
1097 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1098 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1099 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1100 | } |
1101 | ||
1102 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1103 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1104 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1105 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1106 | |
7c12f296 TS |
1107 | /* update sb_versionnum for the clearing of the morebits */ |
1108 | if (!sbp->sb_features2) | |
7884bc86 | 1109 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1110 | } |
1111 | ||
0771fb45 ES |
1112 | /* |
1113 | * Check if sb_agblocks is aligned at stripe boundary | |
1114 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1115 | * allocator alignment is within an ag, therefore ag has | |
1116 | * to be aligned at stripe boundary. | |
1117 | */ | |
7884bc86 | 1118 | error = xfs_update_alignment(mp); |
0771fb45 | 1119 | if (error) |
f9057e3d | 1120 | goto out; |
0771fb45 ES |
1121 | |
1122 | xfs_alloc_compute_maxlevels(mp); | |
1123 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1124 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1125 | xfs_ialloc_compute_maxlevels(mp); | |
1126 | ||
1127 | xfs_set_maxicount(mp); | |
1128 | ||
1129 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
1130 | ||
27174203 CH |
1131 | error = xfs_uuid_mount(mp); |
1132 | if (error) | |
1133 | goto out; | |
1da177e4 | 1134 | |
0771fb45 ES |
1135 | /* |
1136 | * Set the minimum read and write sizes | |
1137 | */ | |
1138 | xfs_set_rw_sizes(mp); | |
1139 | ||
1140 | /* | |
1141 | * Set the inode cluster size. | |
1142 | * This may still be overridden by the file system | |
1143 | * block size if it is larger than the chosen cluster size. | |
1144 | */ | |
1145 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1146 | ||
1147 | /* | |
1148 | * Set inode alignment fields | |
1149 | */ | |
1150 | xfs_set_inoalignment(mp); | |
1151 | ||
1152 | /* | |
1153 | * Check that the data (and log if separate) are an ok size. | |
1154 | */ | |
4249023a | 1155 | error = xfs_check_sizes(mp); |
0771fb45 | 1156 | if (error) |
f9057e3d | 1157 | goto out_remove_uuid; |
0771fb45 | 1158 | |
1da177e4 LT |
1159 | /* |
1160 | * Initialize realtime fields in the mount structure | |
1161 | */ | |
0771fb45 ES |
1162 | error = xfs_rtmount_init(mp); |
1163 | if (error) { | |
1da177e4 | 1164 | cmn_err(CE_WARN, "XFS: RT mount failed"); |
f9057e3d | 1165 | goto out_remove_uuid; |
1da177e4 LT |
1166 | } |
1167 | ||
1da177e4 LT |
1168 | /* |
1169 | * Copies the low order bits of the timestamp and the randomly | |
1170 | * set "sequence" number out of a UUID. | |
1171 | */ | |
1172 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1173 | ||
1da177e4 LT |
1174 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1175 | ||
f6c2d1fa | 1176 | xfs_dir_mount(mp); |
1da177e4 LT |
1177 | |
1178 | /* | |
1179 | * Initialize the attribute manager's entries. | |
1180 | */ | |
1181 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1182 | ||
1183 | /* | |
1184 | * Initialize the precomputed transaction reservations values. | |
1185 | */ | |
1186 | xfs_trans_init(mp); | |
1187 | ||
1da177e4 LT |
1188 | /* |
1189 | * Allocate and initialize the per-ag data. | |
1190 | */ | |
1c1c6ebc DC |
1191 | spin_lock_init(&mp->m_perag_lock); |
1192 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_NOFS); | |
1193 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); | |
1194 | if (error) { | |
1195 | cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error); | |
f9057e3d | 1196 | goto out_remove_uuid; |
1c1c6ebc | 1197 | } |
1da177e4 | 1198 | |
f9057e3d CH |
1199 | if (!sbp->sb_logblocks) { |
1200 | cmn_err(CE_WARN, "XFS: no log defined"); | |
1201 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); | |
1202 | error = XFS_ERROR(EFSCORRUPTED); | |
1203 | goto out_free_perag; | |
1204 | } | |
1205 | ||
1da177e4 LT |
1206 | /* |
1207 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1208 | */ | |
f9057e3d CH |
1209 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1210 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1211 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1212 | if (error) { | |
1213 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
1214 | goto out_free_perag; | |
1da177e4 LT |
1215 | } |
1216 | ||
92821e2b DC |
1217 | /* |
1218 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1219 | * not. If it was, with the first phase of recovery has completed, we | |
1220 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1221 | * but they are recovered transactionally in the second recovery phase | |
1222 | * later. | |
1223 | * | |
1224 | * Hence we can safely re-initialise incore superblock counters from | |
1225 | * the per-ag data. These may not be correct if the filesystem was not | |
1226 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1227 | * doing this. | |
1228 | * | |
1229 | * If the filesystem was cleanly unmounted, then we can trust the | |
1230 | * values in the superblock to be correct and we don't need to do | |
1231 | * anything here. | |
1232 | * | |
1233 | * If we are currently making the filesystem, the initialisation will | |
1234 | * fail as the perag data is in an undefined state. | |
1235 | */ | |
92821e2b DC |
1236 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1237 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1238 | !mp->m_sb.sb_inprogress) { | |
1239 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d CH |
1240 | if (error) |
1241 | goto out_free_perag; | |
92821e2b | 1242 | } |
f9057e3d | 1243 | |
1da177e4 LT |
1244 | /* |
1245 | * Get and sanity-check the root inode. | |
1246 | * Save the pointer to it in the mount structure. | |
1247 | */ | |
1248 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0); | |
1249 | if (error) { | |
1250 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
f9057e3d | 1251 | goto out_log_dealloc; |
1da177e4 LT |
1252 | } |
1253 | ||
1254 | ASSERT(rip != NULL); | |
1da177e4 LT |
1255 | |
1256 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1257 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1258 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1259 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1260 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1261 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1262 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1263 | mp); | |
1264 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1265 | goto out_rele_rip; |
1da177e4 LT |
1266 | } |
1267 | mp->m_rootip = rip; /* save it */ | |
1268 | ||
1269 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1270 | ||
1271 | /* | |
1272 | * Initialize realtime inode pointers in the mount structure | |
1273 | */ | |
0771fb45 ES |
1274 | error = xfs_rtmount_inodes(mp); |
1275 | if (error) { | |
1da177e4 LT |
1276 | /* |
1277 | * Free up the root inode. | |
1278 | */ | |
1279 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
f9057e3d | 1280 | goto out_rele_rip; |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | /* | |
7884bc86 CH |
1284 | * If this is a read-only mount defer the superblock updates until |
1285 | * the next remount into writeable mode. Otherwise we would never | |
1286 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1287 | */ |
7884bc86 CH |
1288 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1289 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec DC |
1290 | if (error) { |
1291 | cmn_err(CE_WARN, "XFS: failed to write sb changes"); | |
b93b6e43 | 1292 | goto out_rtunmount; |
e5720eec DC |
1293 | } |
1294 | } | |
1da177e4 LT |
1295 | |
1296 | /* | |
1297 | * Initialise the XFS quota management subsystem for this mount | |
1298 | */ | |
7d095257 CH |
1299 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1300 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1301 | if (error) | |
1302 | goto out_rtunmount; | |
1303 | } else { | |
1304 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1305 | ||
1306 | /* | |
1307 | * If a file system had quotas running earlier, but decided to | |
1308 | * mount without -o uquota/pquota/gquota options, revoke the | |
1309 | * quotachecked license. | |
1310 | */ | |
1311 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
1312 | cmn_err(CE_NOTE, | |
1313 | "XFS: resetting qflags for filesystem %s", | |
1314 | mp->m_fsname); | |
1315 | ||
1316 | error = xfs_mount_reset_sbqflags(mp); | |
1317 | if (error) | |
1318 | return error; | |
1319 | } | |
1320 | } | |
1da177e4 LT |
1321 | |
1322 | /* | |
1323 | * Finish recovering the file system. This part needed to be | |
1324 | * delayed until after the root and real-time bitmap inodes | |
1325 | * were consistently read in. | |
1326 | */ | |
4249023a | 1327 | error = xfs_log_mount_finish(mp); |
1da177e4 LT |
1328 | if (error) { |
1329 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
b93b6e43 | 1330 | goto out_rtunmount; |
1da177e4 LT |
1331 | } |
1332 | ||
1333 | /* | |
1334 | * Complete the quota initialisation, post-log-replay component. | |
1335 | */ | |
7d095257 CH |
1336 | if (quotamount) { |
1337 | ASSERT(mp->m_qflags == 0); | |
1338 | mp->m_qflags = quotaflags; | |
1339 | ||
1340 | xfs_qm_mount_quotas(mp); | |
1341 | } | |
1342 | ||
1343 | #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY) | |
1344 | if (XFS_IS_QUOTA_ON(mp)) | |
1345 | xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas turned on"); | |
1346 | else | |
1347 | xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas not turned on"); | |
1348 | #endif | |
1da177e4 | 1349 | |
84e1e99f DC |
1350 | /* |
1351 | * Now we are mounted, reserve a small amount of unused space for | |
1352 | * privileged transactions. This is needed so that transaction | |
1353 | * space required for critical operations can dip into this pool | |
1354 | * when at ENOSPC. This is needed for operations like create with | |
1355 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1356 | * are not allowed to use this reserved space. | |
1357 | * | |
1358 | * We default to 5% or 1024 fsbs of space reserved, whichever is smaller. | |
1359 | * This may drive us straight to ENOSPC on mount, but that implies | |
714082bc | 1360 | * we were already there on the last unmount. Warn if this occurs. |
84e1e99f | 1361 | */ |
39726be2 CH |
1362 | resblks = mp->m_sb.sb_dblocks; |
1363 | do_div(resblks, 20); | |
1364 | resblks = min_t(__uint64_t, resblks, 1024); | |
714082bc DC |
1365 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1366 | if (error) | |
1367 | cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. " | |
1368 | "Continuing without a reserve pool."); | |
84e1e99f | 1369 | |
1da177e4 LT |
1370 | return 0; |
1371 | ||
b93b6e43 CH |
1372 | out_rtunmount: |
1373 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1374 | out_rele_rip: |
43355099 | 1375 | IRELE(rip); |
f9057e3d | 1376 | out_log_dealloc: |
21b699c8 | 1377 | xfs_log_unmount(mp); |
f9057e3d | 1378 | out_free_perag: |
ff4f038c | 1379 | xfs_free_perag(mp); |
f9057e3d | 1380 | out_remove_uuid: |
27174203 | 1381 | xfs_uuid_unmount(mp); |
f9057e3d | 1382 | out: |
1da177e4 LT |
1383 | return error; |
1384 | } | |
1385 | ||
1386 | /* | |
1da177e4 LT |
1387 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1388 | * log and makes sure that incore structures are freed. | |
1389 | */ | |
41b5c2e7 CH |
1390 | void |
1391 | xfs_unmountfs( | |
1392 | struct xfs_mount *mp) | |
1da177e4 | 1393 | { |
41b5c2e7 CH |
1394 | __uint64_t resblks; |
1395 | int error; | |
1da177e4 | 1396 | |
7d095257 | 1397 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1398 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1399 | IRELE(mp->m_rootip); |
1400 | ||
641c56fb DC |
1401 | /* |
1402 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1403 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1404 | * the transaction is still sitting in a iclog. The stale inodes |
1405 | * on that buffer will have their flush locks held until the | |
1406 | * transaction hits the disk and the callbacks run. the inode | |
1407 | * flush takes the flush lock unconditionally and with nothing to | |
1408 | * push out the iclog we will never get that unlocked. hence we | |
1409 | * need to force the log first. | |
1410 | */ | |
1411 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
abc10647 | 1412 | xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC); |
1da177e4 | 1413 | |
7d095257 | 1414 | xfs_qm_unmount(mp); |
a357a121 | 1415 | |
1da177e4 LT |
1416 | /* |
1417 | * Flush out the log synchronously so that we know for sure | |
1418 | * that nothing is pinned. This is important because bflush() | |
1419 | * will skip pinned buffers. | |
1420 | */ | |
1421 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC); | |
1422 | ||
1423 | xfs_binval(mp->m_ddev_targp); | |
1424 | if (mp->m_rtdev_targp) { | |
1425 | xfs_binval(mp->m_rtdev_targp); | |
1426 | } | |
1427 | ||
84e1e99f DC |
1428 | /* |
1429 | * Unreserve any blocks we have so that when we unmount we don't account | |
1430 | * the reserved free space as used. This is really only necessary for | |
1431 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1432 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1433 | * |
1434 | * We don't bother correcting this elsewhere for lazy superblock | |
1435 | * counting because on mount of an unclean filesystem we reconstruct the | |
1436 | * correct counter value and this is irrelevant. | |
1437 | * | |
1438 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1439 | * we only every apply deltas to the superblock and hence the incore | |
1440 | * value does not matter.... | |
1441 | */ | |
1442 | resblks = 0; | |
714082bc DC |
1443 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1444 | if (error) | |
1445 | cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. " | |
1446 | "Freespace may not be correct on next mount."); | |
1447 | ||
e5720eec DC |
1448 | error = xfs_log_sbcount(mp, 1); |
1449 | if (error) | |
1450 | cmn_err(CE_WARN, "XFS: Unable to update superblock counters. " | |
1451 | "Freespace may not be correct on next mount."); | |
1da177e4 | 1452 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1453 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
21b699c8 CH |
1454 | xfs_log_unmount_write(mp); |
1455 | xfs_log_unmount(mp); | |
27174203 | 1456 | xfs_uuid_unmount(mp); |
1da177e4 | 1457 | |
1550d0b0 | 1458 | #if defined(DEBUG) |
0ce4cfd4 | 1459 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1460 | #endif |
ff4f038c | 1461 | xfs_free_perag(mp); |
1da177e4 LT |
1462 | } |
1463 | ||
ba0f32d4 | 1464 | STATIC void |
1da177e4 LT |
1465 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1466 | { | |
1467 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1468 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1469 | if (mp->m_rtdev_targp) | |
1470 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1471 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1472 | } | |
1473 | ||
92821e2b DC |
1474 | int |
1475 | xfs_fs_writable(xfs_mount_t *mp) | |
1476 | { | |
b267ce99 | 1477 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1478 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1479 | } |
1480 | ||
1481 | /* | |
1482 | * xfs_log_sbcount | |
1483 | * | |
1484 | * Called either periodically to keep the on disk superblock values | |
1485 | * roughly up to date or from unmount to make sure the values are | |
1486 | * correct on a clean unmount. | |
1487 | * | |
1488 | * Note this code can be called during the process of freezing, so | |
1489 | * we may need to use the transaction allocator which does not not | |
1490 | * block when the transaction subsystem is in its frozen state. | |
1491 | */ | |
1492 | int | |
1493 | xfs_log_sbcount( | |
1494 | xfs_mount_t *mp, | |
1495 | uint sync) | |
1496 | { | |
1497 | xfs_trans_t *tp; | |
1498 | int error; | |
1499 | ||
1500 | if (!xfs_fs_writable(mp)) | |
1501 | return 0; | |
1502 | ||
d4d90b57 | 1503 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1504 | |
1505 | /* | |
1506 | * we don't need to do this if we are updating the superblock | |
1507 | * counters on every modification. | |
1508 | */ | |
1509 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1510 | return 0; | |
1511 | ||
80641dc6 | 1512 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
92821e2b DC |
1513 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1514 | XFS_DEFAULT_LOG_COUNT); | |
1515 | if (error) { | |
1516 | xfs_trans_cancel(tp, 0); | |
1517 | return error; | |
1518 | } | |
1519 | ||
1520 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1521 | if (sync) | |
1522 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1523 | error = xfs_trans_commit(tp, 0); |
1524 | return error; | |
92821e2b DC |
1525 | } |
1526 | ||
1da177e4 LT |
1527 | int |
1528 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1529 | { | |
1530 | xfs_buf_t *sbp; | |
1da177e4 LT |
1531 | int error = 0; |
1532 | ||
1533 | /* | |
1534 | * skip superblock write if fs is read-only, or | |
1535 | * if we are doing a forced umount. | |
1536 | */ | |
bd186aa9 | 1537 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1538 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1539 | |
92821e2b | 1540 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1541 | |
1da177e4 LT |
1542 | XFS_BUF_UNDONE(sbp); |
1543 | XFS_BUF_UNREAD(sbp); | |
1544 | XFS_BUF_UNDELAYWRITE(sbp); | |
1545 | XFS_BUF_WRITE(sbp); | |
1546 | XFS_BUF_UNASYNC(sbp); | |
1547 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1548 | xfsbdstrat(mp, sbp); | |
1da177e4 LT |
1549 | error = xfs_iowait(sbp); |
1550 | if (error) | |
1551 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1552 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
92821e2b | 1553 | xfs_buf_relse(sbp); |
1da177e4 | 1554 | } |
014c2544 | 1555 | return error; |
1da177e4 LT |
1556 | } |
1557 | ||
1558 | /* | |
1559 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1560 | * in-core superblock into the superblock buffer to be logged. | |
1561 | * It does not provide the higher level of locking that is | |
1562 | * needed to protect the in-core superblock from concurrent | |
1563 | * access. | |
1564 | */ | |
1565 | void | |
1566 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1567 | { | |
1568 | xfs_buf_t *bp; | |
1569 | int first; | |
1570 | int last; | |
1571 | xfs_mount_t *mp; | |
1da177e4 LT |
1572 | xfs_sb_field_t f; |
1573 | ||
1574 | ASSERT(fields); | |
1575 | if (!fields) | |
1576 | return; | |
1577 | mp = tp->t_mountp; | |
1578 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1579 | first = sizeof(xfs_sb_t); |
1580 | last = 0; | |
1581 | ||
1582 | /* translate/copy */ | |
1583 | ||
2bdf7cd0 | 1584 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1585 | |
1586 | /* find modified range */ | |
1587 | ||
1588 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1589 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1590 | first = xfs_sb_info[f].offset; | |
1591 | ||
1592 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); | |
1593 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1594 | last = xfs_sb_info[f + 1].offset - 1; | |
1595 | ||
1596 | xfs_trans_log_buf(tp, bp, first, last); | |
1597 | } | |
d210a28c | 1598 | |
d210a28c | 1599 | |
1da177e4 LT |
1600 | /* |
1601 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1602 | * a delta to a specified field in the in-core superblock. Simply | |
1603 | * switch on the field indicated and apply the delta to that field. | |
1604 | * Fields are not allowed to dip below zero, so if the delta would | |
1605 | * do this do not apply it and return EINVAL. | |
1606 | * | |
3685c2a1 | 1607 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1608 | */ |
d96f8f89 | 1609 | STATIC int |
20f4ebf2 DC |
1610 | xfs_mod_incore_sb_unlocked( |
1611 | xfs_mount_t *mp, | |
1612 | xfs_sb_field_t field, | |
1613 | int64_t delta, | |
1614 | int rsvd) | |
1da177e4 LT |
1615 | { |
1616 | int scounter; /* short counter for 32 bit fields */ | |
1617 | long long lcounter; /* long counter for 64 bit fields */ | |
1618 | long long res_used, rem; | |
1619 | ||
1620 | /* | |
1621 | * With the in-core superblock spin lock held, switch | |
1622 | * on the indicated field. Apply the delta to the | |
1623 | * proper field. If the fields value would dip below | |
1624 | * 0, then do not apply the delta and return EINVAL. | |
1625 | */ | |
1626 | switch (field) { | |
1627 | case XFS_SBS_ICOUNT: | |
1628 | lcounter = (long long)mp->m_sb.sb_icount; | |
1629 | lcounter += delta; | |
1630 | if (lcounter < 0) { | |
1631 | ASSERT(0); | |
014c2544 | 1632 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1633 | } |
1634 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1635 | return 0; |
1da177e4 LT |
1636 | case XFS_SBS_IFREE: |
1637 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1638 | lcounter += delta; | |
1639 | if (lcounter < 0) { | |
1640 | ASSERT(0); | |
014c2544 | 1641 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1642 | } |
1643 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1644 | return 0; |
1da177e4 | 1645 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1646 | lcounter = (long long) |
1647 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1648 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1649 | ||
1650 | if (delta > 0) { /* Putting blocks back */ | |
1651 | if (res_used > delta) { | |
1652 | mp->m_resblks_avail += delta; | |
1653 | } else { | |
1654 | rem = delta - res_used; | |
1655 | mp->m_resblks_avail = mp->m_resblks; | |
1656 | lcounter += rem; | |
1657 | } | |
1658 | } else { /* Taking blocks away */ | |
1659 | ||
1660 | lcounter += delta; | |
1661 | ||
1662 | /* | |
1663 | * If were out of blocks, use any available reserved blocks if | |
1664 | * were allowed to. | |
1665 | */ | |
1666 | ||
1667 | if (lcounter < 0) { | |
1668 | if (rsvd) { | |
1669 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1670 | if (lcounter < 0) { | |
014c2544 | 1671 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1672 | } |
1673 | mp->m_resblks_avail = lcounter; | |
014c2544 | 1674 | return 0; |
1da177e4 | 1675 | } else { /* not reserved */ |
014c2544 | 1676 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1677 | } |
1678 | } | |
1679 | } | |
1680 | ||
4be536de | 1681 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1682 | return 0; |
1da177e4 LT |
1683 | case XFS_SBS_FREXTENTS: |
1684 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1685 | lcounter += delta; | |
1686 | if (lcounter < 0) { | |
014c2544 | 1687 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1688 | } |
1689 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1690 | return 0; |
1da177e4 LT |
1691 | case XFS_SBS_DBLOCKS: |
1692 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1693 | lcounter += delta; | |
1694 | if (lcounter < 0) { | |
1695 | ASSERT(0); | |
014c2544 | 1696 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1697 | } |
1698 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1699 | return 0; |
1da177e4 LT |
1700 | case XFS_SBS_AGCOUNT: |
1701 | scounter = mp->m_sb.sb_agcount; | |
1702 | scounter += delta; | |
1703 | if (scounter < 0) { | |
1704 | ASSERT(0); | |
014c2544 | 1705 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1706 | } |
1707 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1708 | return 0; |
1da177e4 LT |
1709 | case XFS_SBS_IMAX_PCT: |
1710 | scounter = mp->m_sb.sb_imax_pct; | |
1711 | scounter += delta; | |
1712 | if (scounter < 0) { | |
1713 | ASSERT(0); | |
014c2544 | 1714 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1715 | } |
1716 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1717 | return 0; |
1da177e4 LT |
1718 | case XFS_SBS_REXTSIZE: |
1719 | scounter = mp->m_sb.sb_rextsize; | |
1720 | scounter += delta; | |
1721 | if (scounter < 0) { | |
1722 | ASSERT(0); | |
014c2544 | 1723 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1724 | } |
1725 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1726 | return 0; |
1da177e4 LT |
1727 | case XFS_SBS_RBMBLOCKS: |
1728 | scounter = mp->m_sb.sb_rbmblocks; | |
1729 | scounter += delta; | |
1730 | if (scounter < 0) { | |
1731 | ASSERT(0); | |
014c2544 | 1732 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1733 | } |
1734 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1735 | return 0; |
1da177e4 LT |
1736 | case XFS_SBS_RBLOCKS: |
1737 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1738 | lcounter += delta; | |
1739 | if (lcounter < 0) { | |
1740 | ASSERT(0); | |
014c2544 | 1741 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1742 | } |
1743 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1744 | return 0; |
1da177e4 LT |
1745 | case XFS_SBS_REXTENTS: |
1746 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1747 | lcounter += delta; | |
1748 | if (lcounter < 0) { | |
1749 | ASSERT(0); | |
014c2544 | 1750 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1751 | } |
1752 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1753 | return 0; |
1da177e4 LT |
1754 | case XFS_SBS_REXTSLOG: |
1755 | scounter = mp->m_sb.sb_rextslog; | |
1756 | scounter += delta; | |
1757 | if (scounter < 0) { | |
1758 | ASSERT(0); | |
014c2544 | 1759 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1760 | } |
1761 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1762 | return 0; |
1da177e4 LT |
1763 | default: |
1764 | ASSERT(0); | |
014c2544 | 1765 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1766 | } |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1771 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1772 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1773 | * routine to do the work. |
1774 | */ | |
1775 | int | |
20f4ebf2 DC |
1776 | xfs_mod_incore_sb( |
1777 | xfs_mount_t *mp, | |
1778 | xfs_sb_field_t field, | |
1779 | int64_t delta, | |
1780 | int rsvd) | |
1da177e4 | 1781 | { |
1da177e4 LT |
1782 | int status; |
1783 | ||
8d280b98 DC |
1784 | /* check for per-cpu counters */ |
1785 | switch (field) { | |
1786 | #ifdef HAVE_PERCPU_SB | |
1787 | case XFS_SBS_ICOUNT: | |
1788 | case XFS_SBS_IFREE: | |
1789 | case XFS_SBS_FDBLOCKS: | |
1790 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1791 | status = xfs_icsb_modify_counters(mp, field, | |
1792 | delta, rsvd); | |
1793 | break; | |
1794 | } | |
1795 | /* FALLTHROUGH */ | |
1796 | #endif | |
1797 | default: | |
3685c2a1 | 1798 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1799 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1800 | spin_unlock(&mp->m_sb_lock); |
8d280b98 DC |
1801 | break; |
1802 | } | |
1803 | ||
014c2544 | 1804 | return status; |
1da177e4 LT |
1805 | } |
1806 | ||
1807 | /* | |
1808 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1809 | * in the in-core superblock structure at a time. This modification | |
1810 | * is protected by a lock internal to this module. The fields and | |
1811 | * changes to those fields are specified in the array of xfs_mod_sb | |
1812 | * structures passed in. | |
1813 | * | |
1814 | * Either all of the specified deltas will be applied or none of | |
1815 | * them will. If any modified field dips below 0, then all modifications | |
1816 | * will be backed out and EINVAL will be returned. | |
1817 | */ | |
1818 | int | |
1819 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1820 | { | |
1da177e4 LT |
1821 | int status=0; |
1822 | xfs_mod_sb_t *msbp; | |
1823 | ||
1824 | /* | |
1825 | * Loop through the array of mod structures and apply each | |
1826 | * individually. If any fail, then back out all those | |
1827 | * which have already been applied. Do all of this within | |
3685c2a1 | 1828 | * the scope of the m_sb_lock so that all of the changes will |
1da177e4 LT |
1829 | * be atomic. |
1830 | */ | |
3685c2a1 | 1831 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
1832 | msbp = &msb[0]; |
1833 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1834 | /* | |
1835 | * Apply the delta at index n. If it fails, break | |
1836 | * from the loop so we'll fall into the undo loop | |
1837 | * below. | |
1838 | */ | |
8d280b98 DC |
1839 | switch (msbp->msb_field) { |
1840 | #ifdef HAVE_PERCPU_SB | |
1841 | case XFS_SBS_ICOUNT: | |
1842 | case XFS_SBS_IFREE: | |
1843 | case XFS_SBS_FDBLOCKS: | |
1844 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1845 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1846 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1847 | msbp->msb_field, |
1848 | msbp->msb_delta, rsvd); | |
3685c2a1 | 1849 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1850 | break; |
1851 | } | |
1852 | /* FALLTHROUGH */ | |
1853 | #endif | |
1854 | default: | |
1855 | status = xfs_mod_incore_sb_unlocked(mp, | |
1856 | msbp->msb_field, | |
1857 | msbp->msb_delta, rsvd); | |
1858 | break; | |
1859 | } | |
1860 | ||
1da177e4 LT |
1861 | if (status != 0) { |
1862 | break; | |
1863 | } | |
1864 | } | |
1865 | ||
1866 | /* | |
1867 | * If we didn't complete the loop above, then back out | |
1868 | * any changes made to the superblock. If you add code | |
1869 | * between the loop above and here, make sure that you | |
1870 | * preserve the value of status. Loop back until | |
1871 | * we step below the beginning of the array. Make sure | |
1872 | * we don't touch anything back there. | |
1873 | */ | |
1874 | if (status != 0) { | |
1875 | msbp--; | |
1876 | while (msbp >= msb) { | |
8d280b98 DC |
1877 | switch (msbp->msb_field) { |
1878 | #ifdef HAVE_PERCPU_SB | |
1879 | case XFS_SBS_ICOUNT: | |
1880 | case XFS_SBS_IFREE: | |
1881 | case XFS_SBS_FDBLOCKS: | |
1882 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1883 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1884 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1885 | msbp->msb_field, |
1886 | -(msbp->msb_delta), | |
1887 | rsvd); | |
3685c2a1 | 1888 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1889 | break; |
1890 | } | |
1891 | /* FALLTHROUGH */ | |
1892 | #endif | |
1893 | default: | |
1894 | status = xfs_mod_incore_sb_unlocked(mp, | |
1895 | msbp->msb_field, | |
1896 | -(msbp->msb_delta), | |
1897 | rsvd); | |
1898 | break; | |
1899 | } | |
1da177e4 LT |
1900 | ASSERT(status == 0); |
1901 | msbp--; | |
1902 | } | |
1903 | } | |
3685c2a1 | 1904 | spin_unlock(&mp->m_sb_lock); |
014c2544 | 1905 | return status; |
1da177e4 LT |
1906 | } |
1907 | ||
1908 | /* | |
1909 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1910 | * The buffer is returned locked and read in from disk. | |
1911 | * The buffer should be released with a call to xfs_brelse(). | |
1912 | * | |
1913 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1914 | * the superblock buffer if it can be locked without sleeping. | |
1915 | * If it can't then we'll return NULL. | |
1916 | */ | |
1917 | xfs_buf_t * | |
1918 | xfs_getsb( | |
1919 | xfs_mount_t *mp, | |
1920 | int flags) | |
1921 | { | |
1922 | xfs_buf_t *bp; | |
1923 | ||
1924 | ASSERT(mp->m_sb_bp != NULL); | |
1925 | bp = mp->m_sb_bp; | |
1926 | if (flags & XFS_BUF_TRYLOCK) { | |
1927 | if (!XFS_BUF_CPSEMA(bp)) { | |
1928 | return NULL; | |
1929 | } | |
1930 | } else { | |
1931 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1932 | } | |
1933 | XFS_BUF_HOLD(bp); | |
1934 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1935 | return bp; |
1da177e4 LT |
1936 | } |
1937 | ||
1938 | /* | |
1939 | * Used to free the superblock along various error paths. | |
1940 | */ | |
1941 | void | |
1942 | xfs_freesb( | |
1943 | xfs_mount_t *mp) | |
1944 | { | |
1945 | xfs_buf_t *bp; | |
1946 | ||
1947 | /* | |
1948 | * Use xfs_getsb() so that the buffer will be locked | |
1949 | * when we call xfs_buf_relse(). | |
1950 | */ | |
1951 | bp = xfs_getsb(mp, 0); | |
1952 | XFS_BUF_UNMANAGE(bp); | |
1953 | xfs_buf_relse(bp); | |
1954 | mp->m_sb_bp = NULL; | |
1955 | } | |
1956 | ||
1da177e4 LT |
1957 | /* |
1958 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
1959 | * be altered by the mount options, as well as any potential sb_features2 |
1960 | * fixup. Only the first superblock is updated. | |
1da177e4 | 1961 | */ |
7884bc86 | 1962 | int |
ee1c0908 | 1963 | xfs_mount_log_sb( |
1da177e4 LT |
1964 | xfs_mount_t *mp, |
1965 | __int64_t fields) | |
1966 | { | |
1967 | xfs_trans_t *tp; | |
e5720eec | 1968 | int error; |
1da177e4 | 1969 | |
ee1c0908 | 1970 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
1971 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
1972 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
1973 | |
1974 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
1975 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1976 | XFS_DEFAULT_LOG_COUNT); | |
1977 | if (error) { | |
1da177e4 | 1978 | xfs_trans_cancel(tp, 0); |
e5720eec | 1979 | return error; |
1da177e4 LT |
1980 | } |
1981 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
1982 | error = xfs_trans_commit(tp, 0); |
1983 | return error; | |
1da177e4 | 1984 | } |
8d280b98 DC |
1985 | |
1986 | ||
1987 | #ifdef HAVE_PERCPU_SB | |
1988 | /* | |
1989 | * Per-cpu incore superblock counters | |
1990 | * | |
1991 | * Simple concept, difficult implementation | |
1992 | * | |
1993 | * Basically, replace the incore superblock counters with a distributed per cpu | |
1994 | * counter for contended fields (e.g. free block count). | |
1995 | * | |
1996 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
1997 | * hence needs to be accurately read when we are running low on space. Hence | |
1998 | * there is a method to enable and disable the per-cpu counters based on how | |
1999 | * much "stuff" is available in them. | |
2000 | * | |
2001 | * Basically, a counter is enabled if there is enough free resource to justify | |
2002 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2003 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2004 | * re-distribute the available resources. | |
2005 | * | |
2006 | * If, once we redistributed the available resources, we still get a failure, | |
2007 | * we disable the per-cpu counter and go through the slow path. | |
2008 | * | |
2009 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2010 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2011 | * the global superblock. We do this after disabling the counter to prevent |
2012 | * more threads from queueing up on the counter. | |
2013 | * | |
2014 | * Essentially, this means that we still need a lock in the fast path to enable | |
2015 | * synchronisation between the global counters and the per-cpu counters. This | |
2016 | * is not a problem because the lock will be local to a CPU almost all the time | |
2017 | * and have little contention except when we get to ENOSPC conditions. | |
2018 | * | |
2019 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2020 | * path while we do things like enabling and disabling counters and | |
2021 | * synchronising the counters. | |
2022 | * | |
2023 | * Locking rules: | |
2024 | * | |
3685c2a1 | 2025 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2026 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2027 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2028 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2029 | * 5. modifying global counters requires holding m_sb_lock |
2030 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2031 | * and _none_ of the per-cpu locks. |
2032 | * | |
2033 | * Disabled counters are only ever re-enabled by a balance operation | |
2034 | * that results in more free resources per CPU than a given threshold. | |
2035 | * To ensure counters don't remain disabled, they are rebalanced when | |
2036 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2037 | * is present to prevent thrashing). | |
e8234a68 DC |
2038 | */ |
2039 | ||
5a67e4c5 | 2040 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2041 | /* |
2042 | * hot-plug CPU notifier support. | |
8d280b98 | 2043 | * |
5a67e4c5 CS |
2044 | * We need a notifier per filesystem as we need to be able to identify |
2045 | * the filesystem to balance the counters out. This is achieved by | |
2046 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2047 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2048 | */ |
e8234a68 DC |
2049 | STATIC int |
2050 | xfs_icsb_cpu_notify( | |
2051 | struct notifier_block *nfb, | |
2052 | unsigned long action, | |
2053 | void *hcpu) | |
2054 | { | |
2055 | xfs_icsb_cnts_t *cntp; | |
2056 | xfs_mount_t *mp; | |
e8234a68 DC |
2057 | |
2058 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2059 | cntp = (xfs_icsb_cnts_t *) | |
2060 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2061 | switch (action) { | |
2062 | case CPU_UP_PREPARE: | |
8bb78442 | 2063 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2064 | /* Easy Case - initialize the area and locks, and |
2065 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2066 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2067 | break; |
2068 | case CPU_ONLINE: | |
8bb78442 | 2069 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2070 | xfs_icsb_lock(mp); |
45af6c6d CH |
2071 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2072 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2073 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2074 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2075 | break; |
2076 | case CPU_DEAD: | |
8bb78442 | 2077 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2078 | /* Disable all the counters, then fold the dead cpu's |
2079 | * count into the total on the global superblock and | |
2080 | * re-enable the counters. */ | |
03135cf7 | 2081 | xfs_icsb_lock(mp); |
3685c2a1 | 2082 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2083 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2084 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2085 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2086 | ||
2087 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2088 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2089 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2090 | ||
01e1b69c | 2091 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2092 | |
45af6c6d CH |
2093 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2094 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2095 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2096 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2097 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2098 | break; |
2099 | } | |
2100 | ||
2101 | return NOTIFY_OK; | |
2102 | } | |
5a67e4c5 | 2103 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2104 | |
8d280b98 DC |
2105 | int |
2106 | xfs_icsb_init_counters( | |
2107 | xfs_mount_t *mp) | |
2108 | { | |
2109 | xfs_icsb_cnts_t *cntp; | |
2110 | int i; | |
2111 | ||
2112 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2113 | if (mp->m_sb_cnts == NULL) | |
2114 | return -ENOMEM; | |
2115 | ||
5a67e4c5 | 2116 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2117 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2118 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2119 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2120 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2121 | |
8d280b98 DC |
2122 | for_each_online_cpu(i) { |
2123 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2124 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2125 | } |
20b64285 DC |
2126 | |
2127 | mutex_init(&mp->m_icsb_mutex); | |
2128 | ||
8d280b98 DC |
2129 | /* |
2130 | * start with all counters disabled so that the | |
2131 | * initial balance kicks us off correctly | |
2132 | */ | |
2133 | mp->m_icsb_counters = -1; | |
2134 | return 0; | |
2135 | } | |
2136 | ||
5478eead LM |
2137 | void |
2138 | xfs_icsb_reinit_counters( | |
2139 | xfs_mount_t *mp) | |
2140 | { | |
2141 | xfs_icsb_lock(mp); | |
2142 | /* | |
2143 | * start with all counters disabled so that the | |
2144 | * initial balance kicks us off correctly | |
2145 | */ | |
2146 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2147 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2148 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2149 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2150 | xfs_icsb_unlock(mp); |
2151 | } | |
2152 | ||
c962fb79 | 2153 | void |
8d280b98 DC |
2154 | xfs_icsb_destroy_counters( |
2155 | xfs_mount_t *mp) | |
2156 | { | |
e8234a68 | 2157 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2158 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2159 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2160 | } |
03135cf7 | 2161 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2162 | } |
2163 | ||
b8f82a4a | 2164 | STATIC void |
01e1b69c DC |
2165 | xfs_icsb_lock_cntr( |
2166 | xfs_icsb_cnts_t *icsbp) | |
2167 | { | |
2168 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2169 | ndelay(1000); | |
2170 | } | |
2171 | } | |
2172 | ||
b8f82a4a | 2173 | STATIC void |
01e1b69c DC |
2174 | xfs_icsb_unlock_cntr( |
2175 | xfs_icsb_cnts_t *icsbp) | |
2176 | { | |
2177 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2178 | } | |
2179 | ||
8d280b98 | 2180 | |
b8f82a4a | 2181 | STATIC void |
8d280b98 DC |
2182 | xfs_icsb_lock_all_counters( |
2183 | xfs_mount_t *mp) | |
2184 | { | |
2185 | xfs_icsb_cnts_t *cntp; | |
2186 | int i; | |
2187 | ||
2188 | for_each_online_cpu(i) { | |
2189 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2190 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2191 | } |
2192 | } | |
2193 | ||
b8f82a4a | 2194 | STATIC void |
8d280b98 DC |
2195 | xfs_icsb_unlock_all_counters( |
2196 | xfs_mount_t *mp) | |
2197 | { | |
2198 | xfs_icsb_cnts_t *cntp; | |
2199 | int i; | |
2200 | ||
2201 | for_each_online_cpu(i) { | |
2202 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2203 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2204 | } |
2205 | } | |
2206 | ||
2207 | STATIC void | |
2208 | xfs_icsb_count( | |
2209 | xfs_mount_t *mp, | |
2210 | xfs_icsb_cnts_t *cnt, | |
2211 | int flags) | |
2212 | { | |
2213 | xfs_icsb_cnts_t *cntp; | |
2214 | int i; | |
2215 | ||
2216 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2217 | ||
2218 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2219 | xfs_icsb_lock_all_counters(mp); | |
2220 | ||
2221 | for_each_online_cpu(i) { | |
2222 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2223 | cnt->icsb_icount += cntp->icsb_icount; | |
2224 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2225 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2226 | } | |
2227 | ||
2228 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2229 | xfs_icsb_unlock_all_counters(mp); | |
2230 | } | |
2231 | ||
2232 | STATIC int | |
2233 | xfs_icsb_counter_disabled( | |
2234 | xfs_mount_t *mp, | |
2235 | xfs_sb_field_t field) | |
2236 | { | |
2237 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2238 | return test_bit(field, &mp->m_icsb_counters); | |
2239 | } | |
2240 | ||
36fbe6e6 | 2241 | STATIC void |
8d280b98 DC |
2242 | xfs_icsb_disable_counter( |
2243 | xfs_mount_t *mp, | |
2244 | xfs_sb_field_t field) | |
2245 | { | |
2246 | xfs_icsb_cnts_t cnt; | |
2247 | ||
2248 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2249 | ||
20b64285 DC |
2250 | /* |
2251 | * If we are already disabled, then there is nothing to do | |
2252 | * here. We check before locking all the counters to avoid | |
2253 | * the expensive lock operation when being called in the | |
2254 | * slow path and the counter is already disabled. This is | |
2255 | * safe because the only time we set or clear this state is under | |
2256 | * the m_icsb_mutex. | |
2257 | */ | |
2258 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2259 | return; |
20b64285 | 2260 | |
8d280b98 DC |
2261 | xfs_icsb_lock_all_counters(mp); |
2262 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2263 | /* drain back to superblock */ | |
2264 | ||
ce46193b | 2265 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2266 | switch(field) { |
2267 | case XFS_SBS_ICOUNT: | |
2268 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2269 | break; | |
2270 | case XFS_SBS_IFREE: | |
2271 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2272 | break; | |
2273 | case XFS_SBS_FDBLOCKS: | |
2274 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2275 | break; | |
2276 | default: | |
2277 | BUG(); | |
2278 | } | |
2279 | } | |
2280 | ||
2281 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2282 | } |
2283 | ||
2284 | STATIC void | |
2285 | xfs_icsb_enable_counter( | |
2286 | xfs_mount_t *mp, | |
2287 | xfs_sb_field_t field, | |
2288 | uint64_t count, | |
2289 | uint64_t resid) | |
2290 | { | |
2291 | xfs_icsb_cnts_t *cntp; | |
2292 | int i; | |
2293 | ||
2294 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2295 | ||
2296 | xfs_icsb_lock_all_counters(mp); | |
2297 | for_each_online_cpu(i) { | |
2298 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2299 | switch (field) { | |
2300 | case XFS_SBS_ICOUNT: | |
2301 | cntp->icsb_icount = count + resid; | |
2302 | break; | |
2303 | case XFS_SBS_IFREE: | |
2304 | cntp->icsb_ifree = count + resid; | |
2305 | break; | |
2306 | case XFS_SBS_FDBLOCKS: | |
2307 | cntp->icsb_fdblocks = count + resid; | |
2308 | break; | |
2309 | default: | |
2310 | BUG(); | |
2311 | break; | |
2312 | } | |
2313 | resid = 0; | |
2314 | } | |
2315 | clear_bit(field, &mp->m_icsb_counters); | |
2316 | xfs_icsb_unlock_all_counters(mp); | |
2317 | } | |
2318 | ||
dbcabad1 | 2319 | void |
d4d90b57 | 2320 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2321 | xfs_mount_t *mp, |
2322 | int flags) | |
2323 | { | |
2324 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2325 | |
8d280b98 DC |
2326 | xfs_icsb_count(mp, &cnt, flags); |
2327 | ||
8d280b98 DC |
2328 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2329 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2330 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2331 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2332 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2333 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2334 | } |
2335 | ||
2336 | /* | |
2337 | * Accurate update of per-cpu counters to incore superblock | |
2338 | */ | |
d4d90b57 | 2339 | void |
8d280b98 | 2340 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2341 | xfs_mount_t *mp, |
2342 | int flags) | |
8d280b98 | 2343 | { |
d4d90b57 CH |
2344 | spin_lock(&mp->m_sb_lock); |
2345 | xfs_icsb_sync_counters_locked(mp, flags); | |
2346 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2347 | } |
2348 | ||
2349 | /* | |
2350 | * Balance and enable/disable counters as necessary. | |
2351 | * | |
20b64285 DC |
2352 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2353 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2354 | * free blocks is something far enough zero that we aren't going thrash when we | |
2355 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2356 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2357 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2358 | * reenabled. | |
2359 | * | |
2360 | * Note that we can be called when counters are already disabled. | |
2361 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2362 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2363 | */ |
20b64285 DC |
2364 | |
2365 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2366 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2367 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2368 | STATIC void |
45af6c6d | 2369 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2370 | xfs_mount_t *mp, |
2371 | xfs_sb_field_t field, | |
20b64285 | 2372 | int min_per_cpu) |
8d280b98 | 2373 | { |
6fdf8ccc | 2374 | uint64_t count, resid; |
8d280b98 | 2375 | int weight = num_online_cpus(); |
20b64285 | 2376 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2377 | |
8d280b98 DC |
2378 | /* disable counter and sync counter */ |
2379 | xfs_icsb_disable_counter(mp, field); | |
2380 | ||
2381 | /* update counters - first CPU gets residual*/ | |
2382 | switch (field) { | |
2383 | case XFS_SBS_ICOUNT: | |
2384 | count = mp->m_sb.sb_icount; | |
2385 | resid = do_div(count, weight); | |
20b64285 | 2386 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2387 | return; |
8d280b98 DC |
2388 | break; |
2389 | case XFS_SBS_IFREE: | |
2390 | count = mp->m_sb.sb_ifree; | |
2391 | resid = do_div(count, weight); | |
20b64285 | 2392 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2393 | return; |
8d280b98 DC |
2394 | break; |
2395 | case XFS_SBS_FDBLOCKS: | |
2396 | count = mp->m_sb.sb_fdblocks; | |
2397 | resid = do_div(count, weight); | |
20b64285 | 2398 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2399 | return; |
8d280b98 DC |
2400 | break; |
2401 | default: | |
2402 | BUG(); | |
6fdf8ccc | 2403 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2404 | break; |
2405 | } | |
2406 | ||
2407 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2408 | } |
2409 | ||
2410 | STATIC void | |
2411 | xfs_icsb_balance_counter( | |
2412 | xfs_mount_t *mp, | |
2413 | xfs_sb_field_t fields, | |
2414 | int min_per_cpu) | |
2415 | { | |
2416 | spin_lock(&mp->m_sb_lock); | |
2417 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2418 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2419 | } |
2420 | ||
a8272ce0 | 2421 | STATIC int |
20b64285 | 2422 | xfs_icsb_modify_counters( |
8d280b98 DC |
2423 | xfs_mount_t *mp, |
2424 | xfs_sb_field_t field, | |
20f4ebf2 | 2425 | int64_t delta, |
20b64285 | 2426 | int rsvd) |
8d280b98 DC |
2427 | { |
2428 | xfs_icsb_cnts_t *icsbp; | |
2429 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2430 | int ret = 0; |
8d280b98 | 2431 | |
20b64285 | 2432 | might_sleep(); |
8d280b98 | 2433 | again: |
7a9e02d6 CL |
2434 | preempt_disable(); |
2435 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2436 | |
2437 | /* | |
2438 | * if the counter is disabled, go to slow path | |
2439 | */ | |
8d280b98 DC |
2440 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2441 | goto slow_path; | |
20b64285 DC |
2442 | xfs_icsb_lock_cntr(icsbp); |
2443 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2444 | xfs_icsb_unlock_cntr(icsbp); | |
2445 | goto slow_path; | |
2446 | } | |
8d280b98 DC |
2447 | |
2448 | switch (field) { | |
2449 | case XFS_SBS_ICOUNT: | |
2450 | lcounter = icsbp->icsb_icount; | |
2451 | lcounter += delta; | |
2452 | if (unlikely(lcounter < 0)) | |
20b64285 | 2453 | goto balance_counter; |
8d280b98 DC |
2454 | icsbp->icsb_icount = lcounter; |
2455 | break; | |
2456 | ||
2457 | case XFS_SBS_IFREE: | |
2458 | lcounter = icsbp->icsb_ifree; | |
2459 | lcounter += delta; | |
2460 | if (unlikely(lcounter < 0)) | |
20b64285 | 2461 | goto balance_counter; |
8d280b98 DC |
2462 | icsbp->icsb_ifree = lcounter; |
2463 | break; | |
2464 | ||
2465 | case XFS_SBS_FDBLOCKS: | |
2466 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2467 | ||
4be536de | 2468 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2469 | lcounter += delta; |
2470 | if (unlikely(lcounter < 0)) | |
20b64285 | 2471 | goto balance_counter; |
4be536de | 2472 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2473 | break; |
2474 | default: | |
2475 | BUG(); | |
2476 | break; | |
2477 | } | |
01e1b69c | 2478 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2479 | preempt_enable(); |
8d280b98 DC |
2480 | return 0; |
2481 | ||
8d280b98 | 2482 | slow_path: |
7a9e02d6 | 2483 | preempt_enable(); |
8d280b98 | 2484 | |
20b64285 DC |
2485 | /* |
2486 | * serialise with a mutex so we don't burn lots of cpu on | |
2487 | * the superblock lock. We still need to hold the superblock | |
2488 | * lock, however, when we modify the global structures. | |
2489 | */ | |
03135cf7 | 2490 | xfs_icsb_lock(mp); |
20b64285 DC |
2491 | |
2492 | /* | |
2493 | * Now running atomically. | |
2494 | * | |
2495 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2496 | * Drop the lock and try again in the fast path.... | |
2497 | */ | |
2498 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2499 | xfs_icsb_unlock(mp); |
8d280b98 | 2500 | goto again; |
8d280b98 DC |
2501 | } |
2502 | ||
20b64285 DC |
2503 | /* |
2504 | * The counter is currently disabled. Because we are | |
2505 | * running atomically here, we know a rebalance cannot | |
2506 | * be in progress. Hence we can go straight to operating | |
2507 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2508 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2509 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2510 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2511 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2512 | * directly on the global counters. | |
2513 | */ | |
3685c2a1 | 2514 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2515 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2516 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2517 | |
20b64285 DC |
2518 | /* |
2519 | * Now that we've modified the global superblock, we | |
2520 | * may be able to re-enable the distributed counters | |
2521 | * (e.g. lots of space just got freed). After that | |
2522 | * we are done. | |
2523 | */ | |
2524 | if (ret != ENOSPC) | |
45af6c6d | 2525 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2526 | xfs_icsb_unlock(mp); |
8d280b98 | 2527 | return ret; |
8d280b98 | 2528 | |
20b64285 DC |
2529 | balance_counter: |
2530 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2531 | preempt_enable(); |
8d280b98 | 2532 | |
20b64285 DC |
2533 | /* |
2534 | * We may have multiple threads here if multiple per-cpu | |
2535 | * counters run dry at the same time. This will mean we can | |
2536 | * do more balances than strictly necessary but it is not | |
2537 | * the common slowpath case. | |
2538 | */ | |
03135cf7 | 2539 | xfs_icsb_lock(mp); |
20b64285 DC |
2540 | |
2541 | /* | |
2542 | * running atomically. | |
2543 | * | |
2544 | * This will leave the counter in the correct state for future | |
2545 | * accesses. After the rebalance, we simply try again and our retry | |
2546 | * will either succeed through the fast path or slow path without | |
2547 | * another balance operation being required. | |
2548 | */ | |
45af6c6d | 2549 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2550 | xfs_icsb_unlock(mp); |
20b64285 | 2551 | goto again; |
8d280b98 | 2552 | } |
20b64285 | 2553 | |
8d280b98 | 2554 | #endif |