]>
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 | ||
0fa800fb DC |
203 | /* |
204 | * Reference counting access wrappers to the perag structures. | |
205 | */ | |
206 | struct xfs_perag * | |
207 | xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno) | |
208 | { | |
209 | struct xfs_perag *pag; | |
210 | int ref = 0; | |
211 | ||
212 | spin_lock(&mp->m_perag_lock); | |
213 | pag = radix_tree_lookup(&mp->m_perag_tree, agno); | |
214 | if (pag) { | |
215 | ASSERT(atomic_read(&pag->pag_ref) >= 0); | |
216 | /* catch leaks in the positive direction during testing */ | |
217 | ASSERT(atomic_read(&pag->pag_ref) < 1000); | |
218 | ref = atomic_inc_return(&pag->pag_ref); | |
219 | } | |
220 | spin_unlock(&mp->m_perag_lock); | |
221 | trace_xfs_perag_get(mp, agno, ref, _RET_IP_); | |
222 | return pag; | |
223 | } | |
224 | ||
225 | void | |
226 | xfs_perag_put(struct xfs_perag *pag) | |
227 | { | |
228 | int ref; | |
229 | ||
230 | ASSERT(atomic_read(&pag->pag_ref) > 0); | |
231 | ref = atomic_dec_return(&pag->pag_ref); | |
232 | trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_); | |
233 | } | |
234 | ||
1da177e4 LT |
235 | /* |
236 | * Free up the resources associated with a mount structure. Assume that | |
237 | * the structure was initially zeroed, so we can tell which fields got | |
238 | * initialized. | |
239 | */ | |
c962fb79 | 240 | STATIC void |
ff4f038c | 241 | xfs_free_perag( |
745f6919 | 242 | xfs_mount_t *mp) |
1da177e4 | 243 | { |
1c1c6ebc DC |
244 | xfs_agnumber_t agno; |
245 | struct xfs_perag *pag; | |
246 | ||
247 | for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) { | |
248 | spin_lock(&mp->m_perag_lock); | |
249 | pag = radix_tree_delete(&mp->m_perag_tree, agno); | |
e57336ff | 250 | ASSERT(pag); |
aed3bb90 | 251 | ASSERT(atomic_read(&pag->pag_ref) == 0); |
1c1c6ebc | 252 | spin_unlock(&mp->m_perag_lock); |
1c1c6ebc | 253 | kmem_free(pag); |
1da177e4 | 254 | } |
1da177e4 LT |
255 | } |
256 | ||
4cc929ee NS |
257 | /* |
258 | * Check size of device based on the (data/realtime) block count. | |
259 | * Note: this check is used by the growfs code as well as mount. | |
260 | */ | |
261 | int | |
262 | xfs_sb_validate_fsb_count( | |
263 | xfs_sb_t *sbp, | |
264 | __uint64_t nblocks) | |
265 | { | |
266 | ASSERT(PAGE_SHIFT >= sbp->sb_blocklog); | |
267 | ASSERT(sbp->sb_blocklog >= BBSHIFT); | |
268 | ||
269 | #if XFS_BIG_BLKNOS /* Limited by ULONG_MAX of page cache index */ | |
270 | if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX) | |
271 | return E2BIG; | |
272 | #else /* Limited by UINT_MAX of sectors */ | |
273 | if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX) | |
274 | return E2BIG; | |
275 | #endif | |
276 | return 0; | |
277 | } | |
1da177e4 LT |
278 | |
279 | /* | |
280 | * Check the validity of the SB found. | |
281 | */ | |
282 | STATIC int | |
283 | xfs_mount_validate_sb( | |
284 | xfs_mount_t *mp, | |
764d1f89 NS |
285 | xfs_sb_t *sbp, |
286 | int flags) | |
1da177e4 LT |
287 | { |
288 | /* | |
289 | * If the log device and data device have the | |
290 | * same device number, the log is internal. | |
291 | * Consequently, the sb_logstart should be non-zero. If | |
292 | * we have a zero sb_logstart in this case, we may be trying to mount | |
293 | * a volume filesystem in a non-volume manner. | |
294 | */ | |
295 | if (sbp->sb_magicnum != XFS_SB_MAGIC) { | |
764d1f89 | 296 | xfs_fs_mount_cmn_err(flags, "bad magic number"); |
1da177e4 LT |
297 | return XFS_ERROR(EWRONGFS); |
298 | } | |
299 | ||
62118709 | 300 | if (!xfs_sb_good_version(sbp)) { |
764d1f89 | 301 | xfs_fs_mount_cmn_err(flags, "bad version"); |
1da177e4 LT |
302 | return XFS_ERROR(EWRONGFS); |
303 | } | |
304 | ||
305 | if (unlikely( | |
306 | sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) { | |
764d1f89 NS |
307 | xfs_fs_mount_cmn_err(flags, |
308 | "filesystem is marked as having an external log; " | |
309 | "specify logdev on the\nmount command line."); | |
310 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
311 | } |
312 | ||
313 | if (unlikely( | |
314 | sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) { | |
764d1f89 NS |
315 | xfs_fs_mount_cmn_err(flags, |
316 | "filesystem is marked as having an internal log; " | |
317 | "do not specify logdev on\nthe mount command line."); | |
318 | return XFS_ERROR(EINVAL); | |
1da177e4 LT |
319 | } |
320 | ||
321 | /* | |
322 | * More sanity checking. These were stolen directly from | |
323 | * xfs_repair. | |
324 | */ | |
325 | if (unlikely( | |
326 | sbp->sb_agcount <= 0 || | |
327 | sbp->sb_sectsize < XFS_MIN_SECTORSIZE || | |
328 | sbp->sb_sectsize > XFS_MAX_SECTORSIZE || | |
329 | sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG || | |
330 | sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG || | |
2ac00af7 | 331 | sbp->sb_sectsize != (1 << sbp->sb_sectlog) || |
1da177e4 LT |
332 | sbp->sb_blocksize < XFS_MIN_BLOCKSIZE || |
333 | sbp->sb_blocksize > XFS_MAX_BLOCKSIZE || | |
334 | sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG || | |
335 | sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG || | |
2ac00af7 | 336 | sbp->sb_blocksize != (1 << sbp->sb_blocklog) || |
1da177e4 LT |
337 | sbp->sb_inodesize < XFS_DINODE_MIN_SIZE || |
338 | sbp->sb_inodesize > XFS_DINODE_MAX_SIZE || | |
9f989c94 NS |
339 | sbp->sb_inodelog < XFS_DINODE_MIN_LOG || |
340 | sbp->sb_inodelog > XFS_DINODE_MAX_LOG || | |
2ac00af7 | 341 | sbp->sb_inodesize != (1 << sbp->sb_inodelog) || |
9f989c94 | 342 | (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog) || |
1da177e4 LT |
343 | (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE) || |
344 | (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE) || | |
e50bd16f | 345 | (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) { |
764d1f89 | 346 | xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed"); |
1da177e4 LT |
347 | return XFS_ERROR(EFSCORRUPTED); |
348 | } | |
349 | ||
350 | /* | |
351 | * Sanity check AG count, size fields against data size field | |
352 | */ | |
353 | if (unlikely( | |
354 | sbp->sb_dblocks == 0 || | |
355 | sbp->sb_dblocks > | |
356 | (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks || | |
357 | sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) * | |
358 | sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) { | |
764d1f89 | 359 | xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed"); |
1da177e4 LT |
360 | return XFS_ERROR(EFSCORRUPTED); |
361 | } | |
362 | ||
2edbddd5 LM |
363 | /* |
364 | * Until this is fixed only page-sized or smaller data blocks work. | |
365 | */ | |
366 | if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) { | |
367 | xfs_fs_mount_cmn_err(flags, | |
368 | "file system with blocksize %d bytes", | |
369 | sbp->sb_blocksize); | |
370 | xfs_fs_mount_cmn_err(flags, | |
371 | "only pagesize (%ld) or less will currently work.", | |
372 | PAGE_SIZE); | |
373 | return XFS_ERROR(ENOSYS); | |
374 | } | |
375 | ||
1a5902c5 CH |
376 | /* |
377 | * Currently only very few inode sizes are supported. | |
378 | */ | |
379 | switch (sbp->sb_inodesize) { | |
380 | case 256: | |
381 | case 512: | |
382 | case 1024: | |
383 | case 2048: | |
384 | break; | |
385 | default: | |
386 | xfs_fs_mount_cmn_err(flags, | |
387 | "inode size of %d bytes not supported", | |
388 | sbp->sb_inodesize); | |
389 | return XFS_ERROR(ENOSYS); | |
390 | } | |
391 | ||
4cc929ee NS |
392 | if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) || |
393 | xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) { | |
764d1f89 NS |
394 | xfs_fs_mount_cmn_err(flags, |
395 | "file system too large to be mounted on this system."); | |
1da177e4 LT |
396 | return XFS_ERROR(E2BIG); |
397 | } | |
398 | ||
399 | if (unlikely(sbp->sb_inprogress)) { | |
764d1f89 | 400 | xfs_fs_mount_cmn_err(flags, "file system busy"); |
1da177e4 LT |
401 | return XFS_ERROR(EFSCORRUPTED); |
402 | } | |
403 | ||
de20614b NS |
404 | /* |
405 | * Version 1 directory format has never worked on Linux. | |
406 | */ | |
62118709 | 407 | if (unlikely(!xfs_sb_version_hasdirv2(sbp))) { |
764d1f89 NS |
408 | xfs_fs_mount_cmn_err(flags, |
409 | "file system using version 1 directory format"); | |
de20614b NS |
410 | return XFS_ERROR(ENOSYS); |
411 | } | |
412 | ||
1da177e4 LT |
413 | return 0; |
414 | } | |
415 | ||
da353b0d DC |
416 | STATIC void |
417 | xfs_initialize_perag_icache( | |
418 | xfs_perag_t *pag) | |
419 | { | |
420 | if (!pag->pag_ici_init) { | |
421 | rwlock_init(&pag->pag_ici_lock); | |
422 | INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC); | |
423 | pag->pag_ici_init = 1; | |
424 | } | |
425 | } | |
426 | ||
1c1c6ebc | 427 | int |
c11e2c36 | 428 | xfs_initialize_perag( |
c11e2c36 | 429 | xfs_mount_t *mp, |
1c1c6ebc DC |
430 | xfs_agnumber_t agcount, |
431 | xfs_agnumber_t *maxagi) | |
1da177e4 LT |
432 | { |
433 | xfs_agnumber_t index, max_metadata; | |
8b26c582 | 434 | xfs_agnumber_t first_initialised = 0; |
1da177e4 LT |
435 | xfs_perag_t *pag; |
436 | xfs_agino_t agino; | |
437 | xfs_ino_t ino; | |
438 | xfs_sb_t *sbp = &mp->m_sb; | |
439 | xfs_ino_t max_inum = XFS_MAXINUMBER_32; | |
8b26c582 | 440 | int error = -ENOMEM; |
1da177e4 LT |
441 | |
442 | /* Check to see if the filesystem can overflow 32 bit inodes */ | |
443 | agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0); | |
444 | ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino); | |
445 | ||
1c1c6ebc DC |
446 | /* |
447 | * Walk the current per-ag tree so we don't try to initialise AGs | |
448 | * that already exist (growfs case). Allocate and insert all the | |
449 | * AGs we don't find ready for initialisation. | |
450 | */ | |
451 | for (index = 0; index < agcount; index++) { | |
452 | pag = xfs_perag_get(mp, index); | |
453 | if (pag) { | |
454 | xfs_perag_put(pag); | |
455 | continue; | |
456 | } | |
8b26c582 DC |
457 | if (!first_initialised) |
458 | first_initialised = index; | |
1c1c6ebc DC |
459 | pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL); |
460 | if (!pag) | |
8b26c582 | 461 | goto out_unwind; |
1c1c6ebc | 462 | if (radix_tree_preload(GFP_NOFS)) |
8b26c582 | 463 | goto out_unwind; |
1c1c6ebc DC |
464 | spin_lock(&mp->m_perag_lock); |
465 | if (radix_tree_insert(&mp->m_perag_tree, index, pag)) { | |
466 | BUG(); | |
467 | spin_unlock(&mp->m_perag_lock); | |
8b26c582 DC |
468 | radix_tree_preload_end(); |
469 | error = -EEXIST; | |
470 | goto out_unwind; | |
1c1c6ebc | 471 | } |
0fa800fb DC |
472 | pag->pag_agno = index; |
473 | pag->pag_mount = mp; | |
1c1c6ebc DC |
474 | spin_unlock(&mp->m_perag_lock); |
475 | radix_tree_preload_end(); | |
476 | } | |
477 | ||
1da177e4 LT |
478 | /* Clear the mount flag if no inode can overflow 32 bits |
479 | * on this filesystem, or if specifically requested.. | |
480 | */ | |
bd186aa9 | 481 | if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) { |
1da177e4 LT |
482 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
483 | } else { | |
484 | mp->m_flags &= ~XFS_MOUNT_32BITINODES; | |
485 | } | |
486 | ||
487 | /* If we can overflow then setup the ag headers accordingly */ | |
488 | if (mp->m_flags & XFS_MOUNT_32BITINODES) { | |
489 | /* Calculate how much should be reserved for inodes to | |
490 | * meet the max inode percentage. | |
491 | */ | |
492 | if (mp->m_maxicount) { | |
493 | __uint64_t icount; | |
494 | ||
495 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; | |
496 | do_div(icount, 100); | |
497 | icount += sbp->sb_agblocks - 1; | |
a749ee86 | 498 | do_div(icount, sbp->sb_agblocks); |
1da177e4 LT |
499 | max_metadata = icount; |
500 | } else { | |
501 | max_metadata = agcount; | |
502 | } | |
503 | for (index = 0; index < agcount; index++) { | |
504 | ino = XFS_AGINO_TO_INO(mp, index, agino); | |
505 | if (ino > max_inum) { | |
506 | index++; | |
507 | break; | |
508 | } | |
509 | ||
c41564b5 | 510 | /* This ag is preferred for inodes */ |
44b56e0a | 511 | pag = xfs_perag_get(mp, index); |
1da177e4 LT |
512 | pag->pagi_inodeok = 1; |
513 | if (index < max_metadata) | |
514 | pag->pagf_metadata = 1; | |
da353b0d | 515 | xfs_initialize_perag_icache(pag); |
44b56e0a | 516 | xfs_perag_put(pag); |
1da177e4 LT |
517 | } |
518 | } else { | |
519 | /* Setup default behavior for smaller filesystems */ | |
520 | for (index = 0; index < agcount; index++) { | |
44b56e0a | 521 | pag = xfs_perag_get(mp, index); |
1da177e4 | 522 | pag->pagi_inodeok = 1; |
da353b0d | 523 | xfs_initialize_perag_icache(pag); |
44b56e0a | 524 | xfs_perag_put(pag); |
1da177e4 LT |
525 | } |
526 | } | |
1c1c6ebc DC |
527 | if (maxagi) |
528 | *maxagi = index; | |
529 | return 0; | |
8b26c582 DC |
530 | |
531 | out_unwind: | |
532 | kmem_free(pag); | |
533 | for (; index > first_initialised; index--) { | |
534 | pag = radix_tree_delete(&mp->m_perag_tree, index); | |
535 | kmem_free(pag); | |
536 | } | |
537 | return error; | |
1da177e4 LT |
538 | } |
539 | ||
2bdf7cd0 CH |
540 | void |
541 | xfs_sb_from_disk( | |
542 | xfs_sb_t *to, | |
543 | xfs_dsb_t *from) | |
544 | { | |
545 | to->sb_magicnum = be32_to_cpu(from->sb_magicnum); | |
546 | to->sb_blocksize = be32_to_cpu(from->sb_blocksize); | |
547 | to->sb_dblocks = be64_to_cpu(from->sb_dblocks); | |
548 | to->sb_rblocks = be64_to_cpu(from->sb_rblocks); | |
549 | to->sb_rextents = be64_to_cpu(from->sb_rextents); | |
550 | memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid)); | |
551 | to->sb_logstart = be64_to_cpu(from->sb_logstart); | |
552 | to->sb_rootino = be64_to_cpu(from->sb_rootino); | |
553 | to->sb_rbmino = be64_to_cpu(from->sb_rbmino); | |
554 | to->sb_rsumino = be64_to_cpu(from->sb_rsumino); | |
555 | to->sb_rextsize = be32_to_cpu(from->sb_rextsize); | |
556 | to->sb_agblocks = be32_to_cpu(from->sb_agblocks); | |
557 | to->sb_agcount = be32_to_cpu(from->sb_agcount); | |
558 | to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks); | |
559 | to->sb_logblocks = be32_to_cpu(from->sb_logblocks); | |
560 | to->sb_versionnum = be16_to_cpu(from->sb_versionnum); | |
561 | to->sb_sectsize = be16_to_cpu(from->sb_sectsize); | |
562 | to->sb_inodesize = be16_to_cpu(from->sb_inodesize); | |
563 | to->sb_inopblock = be16_to_cpu(from->sb_inopblock); | |
564 | memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname)); | |
565 | to->sb_blocklog = from->sb_blocklog; | |
566 | to->sb_sectlog = from->sb_sectlog; | |
567 | to->sb_inodelog = from->sb_inodelog; | |
568 | to->sb_inopblog = from->sb_inopblog; | |
569 | to->sb_agblklog = from->sb_agblklog; | |
570 | to->sb_rextslog = from->sb_rextslog; | |
571 | to->sb_inprogress = from->sb_inprogress; | |
572 | to->sb_imax_pct = from->sb_imax_pct; | |
573 | to->sb_icount = be64_to_cpu(from->sb_icount); | |
574 | to->sb_ifree = be64_to_cpu(from->sb_ifree); | |
575 | to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks); | |
576 | to->sb_frextents = be64_to_cpu(from->sb_frextents); | |
577 | to->sb_uquotino = be64_to_cpu(from->sb_uquotino); | |
578 | to->sb_gquotino = be64_to_cpu(from->sb_gquotino); | |
579 | to->sb_qflags = be16_to_cpu(from->sb_qflags); | |
580 | to->sb_flags = from->sb_flags; | |
581 | to->sb_shared_vn = from->sb_shared_vn; | |
582 | to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt); | |
583 | to->sb_unit = be32_to_cpu(from->sb_unit); | |
584 | to->sb_width = be32_to_cpu(from->sb_width); | |
585 | to->sb_dirblklog = from->sb_dirblklog; | |
586 | to->sb_logsectlog = from->sb_logsectlog; | |
587 | to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize); | |
588 | to->sb_logsunit = be32_to_cpu(from->sb_logsunit); | |
589 | to->sb_features2 = be32_to_cpu(from->sb_features2); | |
ee1c0908 | 590 | to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2); |
2bdf7cd0 CH |
591 | } |
592 | ||
1da177e4 | 593 | /* |
2bdf7cd0 | 594 | * Copy in core superblock to ondisk one. |
1da177e4 | 595 | * |
2bdf7cd0 | 596 | * The fields argument is mask of superblock fields to copy. |
1da177e4 LT |
597 | */ |
598 | void | |
2bdf7cd0 CH |
599 | xfs_sb_to_disk( |
600 | xfs_dsb_t *to, | |
601 | xfs_sb_t *from, | |
1da177e4 LT |
602 | __int64_t fields) |
603 | { | |
2bdf7cd0 CH |
604 | xfs_caddr_t to_ptr = (xfs_caddr_t)to; |
605 | xfs_caddr_t from_ptr = (xfs_caddr_t)from; | |
1da177e4 LT |
606 | xfs_sb_field_t f; |
607 | int first; | |
608 | int size; | |
609 | ||
1da177e4 | 610 | ASSERT(fields); |
1da177e4 LT |
611 | if (!fields) |
612 | return; | |
613 | ||
1da177e4 LT |
614 | while (fields) { |
615 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
616 | first = xfs_sb_info[f].offset; | |
617 | size = xfs_sb_info[f + 1].offset - first; | |
618 | ||
619 | ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1); | |
620 | ||
621 | if (size == 1 || xfs_sb_info[f].type == 1) { | |
2bdf7cd0 | 622 | memcpy(to_ptr + first, from_ptr + first, size); |
1da177e4 LT |
623 | } else { |
624 | switch (size) { | |
625 | case 2: | |
2bdf7cd0 CH |
626 | *(__be16 *)(to_ptr + first) = |
627 | cpu_to_be16(*(__u16 *)(from_ptr + first)); | |
1da177e4 LT |
628 | break; |
629 | case 4: | |
2bdf7cd0 CH |
630 | *(__be32 *)(to_ptr + first) = |
631 | cpu_to_be32(*(__u32 *)(from_ptr + first)); | |
1da177e4 LT |
632 | break; |
633 | case 8: | |
2bdf7cd0 CH |
634 | *(__be64 *)(to_ptr + first) = |
635 | cpu_to_be64(*(__u64 *)(from_ptr + first)); | |
1da177e4 LT |
636 | break; |
637 | default: | |
638 | ASSERT(0); | |
639 | } | |
640 | } | |
641 | ||
642 | fields &= ~(1LL << f); | |
643 | } | |
644 | } | |
645 | ||
646 | /* | |
647 | * xfs_readsb | |
648 | * | |
649 | * Does the initial read of the superblock. | |
650 | */ | |
651 | int | |
764d1f89 | 652 | xfs_readsb(xfs_mount_t *mp, int flags) |
1da177e4 LT |
653 | { |
654 | unsigned int sector_size; | |
655 | unsigned int extra_flags; | |
656 | xfs_buf_t *bp; | |
1da177e4 LT |
657 | int error; |
658 | ||
659 | ASSERT(mp->m_sb_bp == NULL); | |
660 | ASSERT(mp->m_ddev_targp != NULL); | |
661 | ||
662 | /* | |
663 | * Allocate a (locked) buffer to hold the superblock. | |
664 | * This will be kept around at all times to optimize | |
665 | * access to the superblock. | |
666 | */ | |
667 | sector_size = xfs_getsize_buftarg(mp->m_ddev_targp); | |
0cadda1c | 668 | extra_flags = XBF_LOCK | XBF_FS_MANAGED | XBF_MAPPED; |
1da177e4 | 669 | |
6ad112bf CH |
670 | bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, BTOBB(sector_size), |
671 | extra_flags); | |
1da177e4 | 672 | if (!bp || XFS_BUF_ISERROR(bp)) { |
764d1f89 | 673 | xfs_fs_mount_cmn_err(flags, "SB read failed"); |
1da177e4 LT |
674 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
675 | goto fail; | |
676 | } | |
677 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
678 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
679 | ||
680 | /* | |
681 | * Initialize the mount structure from the superblock. | |
682 | * But first do some basic consistency checking. | |
683 | */ | |
2bdf7cd0 | 684 | xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp)); |
1da177e4 | 685 | |
764d1f89 | 686 | error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags); |
1da177e4 | 687 | if (error) { |
764d1f89 | 688 | xfs_fs_mount_cmn_err(flags, "SB validate failed"); |
1da177e4 LT |
689 | goto fail; |
690 | } | |
691 | ||
692 | /* | |
693 | * We must be able to do sector-sized and sector-aligned IO. | |
694 | */ | |
695 | if (sector_size > mp->m_sb.sb_sectsize) { | |
764d1f89 NS |
696 | xfs_fs_mount_cmn_err(flags, |
697 | "device supports only %u byte sectors (not %u)", | |
1da177e4 LT |
698 | sector_size, mp->m_sb.sb_sectsize); |
699 | error = ENOSYS; | |
700 | goto fail; | |
701 | } | |
702 | ||
703 | /* | |
704 | * If device sector size is smaller than the superblock size, | |
705 | * re-read the superblock so the buffer is correctly sized. | |
706 | */ | |
707 | if (sector_size < mp->m_sb.sb_sectsize) { | |
708 | XFS_BUF_UNMANAGE(bp); | |
709 | xfs_buf_relse(bp); | |
710 | sector_size = mp->m_sb.sb_sectsize; | |
6ad112bf CH |
711 | bp = xfs_buf_read(mp->m_ddev_targp, XFS_SB_DADDR, |
712 | BTOBB(sector_size), extra_flags); | |
1da177e4 | 713 | if (!bp || XFS_BUF_ISERROR(bp)) { |
764d1f89 | 714 | xfs_fs_mount_cmn_err(flags, "SB re-read failed"); |
1da177e4 LT |
715 | error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM; |
716 | goto fail; | |
717 | } | |
718 | ASSERT(XFS_BUF_ISBUSY(bp)); | |
719 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); | |
720 | } | |
721 | ||
5478eead LM |
722 | /* Initialize per-cpu counters */ |
723 | xfs_icsb_reinit_counters(mp); | |
8d280b98 | 724 | |
1da177e4 LT |
725 | mp->m_sb_bp = bp; |
726 | xfs_buf_relse(bp); | |
727 | ASSERT(XFS_BUF_VALUSEMA(bp) > 0); | |
728 | return 0; | |
729 | ||
730 | fail: | |
731 | if (bp) { | |
732 | XFS_BUF_UNMANAGE(bp); | |
733 | xfs_buf_relse(bp); | |
734 | } | |
735 | return error; | |
736 | } | |
737 | ||
738 | ||
739 | /* | |
740 | * xfs_mount_common | |
741 | * | |
742 | * Mount initialization code establishing various mount | |
743 | * fields from the superblock associated with the given | |
744 | * mount structure | |
745 | */ | |
ba0f32d4 | 746 | STATIC void |
1da177e4 LT |
747 | xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp) |
748 | { | |
1da177e4 | 749 | mp->m_agfrotor = mp->m_agirotor = 0; |
007c61c6 | 750 | spin_lock_init(&mp->m_agirotor_lock); |
1da177e4 LT |
751 | mp->m_maxagi = mp->m_sb.sb_agcount; |
752 | mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG; | |
753 | mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT; | |
754 | mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT; | |
755 | mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1; | |
756 | mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog; | |
1da177e4 LT |
757 | mp->m_blockmask = sbp->sb_blocksize - 1; |
758 | mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG; | |
759 | mp->m_blockwmask = mp->m_blockwsize - 1; | |
1da177e4 | 760 | |
60197e8d CH |
761 | mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1); |
762 | mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
763 | mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2; | |
764 | mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2; | |
765 | ||
766 | mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1); | |
767 | mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0); | |
768 | mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2; | |
769 | mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2; | |
770 | ||
771 | mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1); | |
772 | mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0); | |
773 | mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2; | |
774 | mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2; | |
1da177e4 LT |
775 | |
776 | mp->m_bsize = XFS_FSB_TO_BB(mp, 1); | |
777 | mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK, | |
778 | sbp->sb_inopblock); | |
779 | mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog; | |
780 | } | |
92821e2b DC |
781 | |
782 | /* | |
783 | * xfs_initialize_perag_data | |
784 | * | |
785 | * Read in each per-ag structure so we can count up the number of | |
786 | * allocated inodes, free inodes and used filesystem blocks as this | |
787 | * information is no longer persistent in the superblock. Once we have | |
788 | * this information, write it into the in-core superblock structure. | |
789 | */ | |
790 | STATIC int | |
791 | xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount) | |
792 | { | |
793 | xfs_agnumber_t index; | |
794 | xfs_perag_t *pag; | |
795 | xfs_sb_t *sbp = &mp->m_sb; | |
796 | uint64_t ifree = 0; | |
797 | uint64_t ialloc = 0; | |
798 | uint64_t bfree = 0; | |
799 | uint64_t bfreelst = 0; | |
800 | uint64_t btree = 0; | |
801 | int error; | |
92821e2b DC |
802 | |
803 | for (index = 0; index < agcount; index++) { | |
804 | /* | |
805 | * read the agf, then the agi. This gets us | |
9da096fd | 806 | * all the information we need and populates the |
92821e2b DC |
807 | * per-ag structures for us. |
808 | */ | |
809 | error = xfs_alloc_pagf_init(mp, NULL, index, 0); | |
810 | if (error) | |
811 | return error; | |
812 | ||
813 | error = xfs_ialloc_pagi_init(mp, NULL, index); | |
814 | if (error) | |
815 | return error; | |
44b56e0a | 816 | pag = xfs_perag_get(mp, index); |
92821e2b DC |
817 | ifree += pag->pagi_freecount; |
818 | ialloc += pag->pagi_count; | |
819 | bfree += pag->pagf_freeblks; | |
820 | bfreelst += pag->pagf_flcount; | |
821 | btree += pag->pagf_btreeblks; | |
44b56e0a | 822 | xfs_perag_put(pag); |
92821e2b DC |
823 | } |
824 | /* | |
825 | * Overwrite incore superblock counters with just-read data | |
826 | */ | |
3685c2a1 | 827 | spin_lock(&mp->m_sb_lock); |
92821e2b DC |
828 | sbp->sb_ifree = ifree; |
829 | sbp->sb_icount = ialloc; | |
830 | sbp->sb_fdblocks = bfree + bfreelst + btree; | |
3685c2a1 | 831 | spin_unlock(&mp->m_sb_lock); |
92821e2b DC |
832 | |
833 | /* Fixup the per-cpu counters as well. */ | |
834 | xfs_icsb_reinit_counters(mp); | |
835 | ||
836 | return 0; | |
837 | } | |
838 | ||
1da177e4 | 839 | /* |
0771fb45 | 840 | * Update alignment values based on mount options and sb values |
1da177e4 | 841 | */ |
0771fb45 | 842 | STATIC int |
7884bc86 | 843 | xfs_update_alignment(xfs_mount_t *mp) |
1da177e4 | 844 | { |
1da177e4 | 845 | xfs_sb_t *sbp = &(mp->m_sb); |
1da177e4 | 846 | |
4249023a | 847 | if (mp->m_dalign) { |
1da177e4 LT |
848 | /* |
849 | * If stripe unit and stripe width are not multiples | |
850 | * of the fs blocksize turn off alignment. | |
851 | */ | |
852 | if ((BBTOB(mp->m_dalign) & mp->m_blockmask) || | |
853 | (BBTOB(mp->m_swidth) & mp->m_blockmask)) { | |
854 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
855 | cmn_err(CE_WARN, | |
856 | "XFS: alignment check 1 failed"); | |
0771fb45 | 857 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
858 | } |
859 | mp->m_dalign = mp->m_swidth = 0; | |
860 | } else { | |
861 | /* | |
862 | * Convert the stripe unit and width to FSBs. | |
863 | */ | |
864 | mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign); | |
865 | if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) { | |
866 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
0771fb45 | 867 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
868 | } |
869 | xfs_fs_cmn_err(CE_WARN, mp, | |
870 | "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)", | |
871 | mp->m_dalign, mp->m_swidth, | |
872 | sbp->sb_agblocks); | |
873 | ||
874 | mp->m_dalign = 0; | |
875 | mp->m_swidth = 0; | |
876 | } else if (mp->m_dalign) { | |
877 | mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth); | |
878 | } else { | |
879 | if (mp->m_flags & XFS_MOUNT_RETERR) { | |
880 | xfs_fs_cmn_err(CE_WARN, mp, | |
881 | "stripe alignment turned off: sunit(%d) less than bsize(%d)", | |
882 | mp->m_dalign, | |
883 | mp->m_blockmask +1); | |
0771fb45 | 884 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
885 | } |
886 | mp->m_swidth = 0; | |
887 | } | |
888 | } | |
889 | ||
890 | /* | |
891 | * Update superblock with new values | |
892 | * and log changes | |
893 | */ | |
62118709 | 894 | if (xfs_sb_version_hasdalign(sbp)) { |
1da177e4 LT |
895 | if (sbp->sb_unit != mp->m_dalign) { |
896 | sbp->sb_unit = mp->m_dalign; | |
7884bc86 | 897 | mp->m_update_flags |= XFS_SB_UNIT; |
1da177e4 LT |
898 | } |
899 | if (sbp->sb_width != mp->m_swidth) { | |
900 | sbp->sb_width = mp->m_swidth; | |
7884bc86 | 901 | mp->m_update_flags |= XFS_SB_WIDTH; |
1da177e4 LT |
902 | } |
903 | } | |
904 | } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN && | |
62118709 | 905 | xfs_sb_version_hasdalign(&mp->m_sb)) { |
1da177e4 LT |
906 | mp->m_dalign = sbp->sb_unit; |
907 | mp->m_swidth = sbp->sb_width; | |
908 | } | |
909 | ||
0771fb45 ES |
910 | return 0; |
911 | } | |
1da177e4 | 912 | |
0771fb45 ES |
913 | /* |
914 | * Set the maximum inode count for this filesystem | |
915 | */ | |
916 | STATIC void | |
917 | xfs_set_maxicount(xfs_mount_t *mp) | |
918 | { | |
919 | xfs_sb_t *sbp = &(mp->m_sb); | |
920 | __uint64_t icount; | |
1da177e4 | 921 | |
0771fb45 ES |
922 | if (sbp->sb_imax_pct) { |
923 | /* | |
924 | * Make sure the maximum inode count is a multiple | |
925 | * of the units we allocate inodes in. | |
1da177e4 | 926 | */ |
1da177e4 LT |
927 | icount = sbp->sb_dblocks * sbp->sb_imax_pct; |
928 | do_div(icount, 100); | |
929 | do_div(icount, mp->m_ialloc_blks); | |
930 | mp->m_maxicount = (icount * mp->m_ialloc_blks) << | |
931 | sbp->sb_inopblog; | |
0771fb45 | 932 | } else { |
1da177e4 | 933 | mp->m_maxicount = 0; |
1da177e4 | 934 | } |
0771fb45 ES |
935 | } |
936 | ||
937 | /* | |
938 | * Set the default minimum read and write sizes unless | |
939 | * already specified in a mount option. | |
940 | * We use smaller I/O sizes when the file system | |
941 | * is being used for NFS service (wsync mount option). | |
942 | */ | |
943 | STATIC void | |
944 | xfs_set_rw_sizes(xfs_mount_t *mp) | |
945 | { | |
946 | xfs_sb_t *sbp = &(mp->m_sb); | |
947 | int readio_log, writeio_log; | |
1da177e4 | 948 | |
1da177e4 LT |
949 | if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) { |
950 | if (mp->m_flags & XFS_MOUNT_WSYNC) { | |
951 | readio_log = XFS_WSYNC_READIO_LOG; | |
952 | writeio_log = XFS_WSYNC_WRITEIO_LOG; | |
953 | } else { | |
954 | readio_log = XFS_READIO_LOG_LARGE; | |
955 | writeio_log = XFS_WRITEIO_LOG_LARGE; | |
956 | } | |
957 | } else { | |
958 | readio_log = mp->m_readio_log; | |
959 | writeio_log = mp->m_writeio_log; | |
960 | } | |
961 | ||
1da177e4 LT |
962 | if (sbp->sb_blocklog > readio_log) { |
963 | mp->m_readio_log = sbp->sb_blocklog; | |
964 | } else { | |
965 | mp->m_readio_log = readio_log; | |
966 | } | |
967 | mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog); | |
968 | if (sbp->sb_blocklog > writeio_log) { | |
969 | mp->m_writeio_log = sbp->sb_blocklog; | |
970 | } else { | |
971 | mp->m_writeio_log = writeio_log; | |
972 | } | |
973 | mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog); | |
0771fb45 | 974 | } |
1da177e4 | 975 | |
0771fb45 ES |
976 | /* |
977 | * Set whether we're using inode alignment. | |
978 | */ | |
979 | STATIC void | |
980 | xfs_set_inoalignment(xfs_mount_t *mp) | |
981 | { | |
62118709 | 982 | if (xfs_sb_version_hasalign(&mp->m_sb) && |
1da177e4 LT |
983 | mp->m_sb.sb_inoalignmt >= |
984 | XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) | |
985 | mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1; | |
986 | else | |
987 | mp->m_inoalign_mask = 0; | |
988 | /* | |
989 | * If we are using stripe alignment, check whether | |
990 | * the stripe unit is a multiple of the inode alignment | |
991 | */ | |
992 | if (mp->m_dalign && mp->m_inoalign_mask && | |
993 | !(mp->m_dalign & mp->m_inoalign_mask)) | |
994 | mp->m_sinoalign = mp->m_dalign; | |
995 | else | |
996 | mp->m_sinoalign = 0; | |
0771fb45 ES |
997 | } |
998 | ||
999 | /* | |
1000 | * Check that the data (and log if separate) are an ok size. | |
1001 | */ | |
1002 | STATIC int | |
4249023a | 1003 | xfs_check_sizes(xfs_mount_t *mp) |
0771fb45 ES |
1004 | { |
1005 | xfs_buf_t *bp; | |
1006 | xfs_daddr_t d; | |
1007 | int error; | |
1008 | ||
1da177e4 LT |
1009 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks); |
1010 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) { | |
1011 | cmn_err(CE_WARN, "XFS: size check 1 failed"); | |
0771fb45 | 1012 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
1013 | } |
1014 | error = xfs_read_buf(mp, mp->m_ddev_targp, | |
1015 | d - XFS_FSS_TO_BB(mp, 1), | |
1016 | XFS_FSS_TO_BB(mp, 1), 0, &bp); | |
1017 | if (!error) { | |
1018 | xfs_buf_relse(bp); | |
1019 | } else { | |
1020 | cmn_err(CE_WARN, "XFS: size check 2 failed"); | |
0771fb45 | 1021 | if (error == ENOSPC) |
1da177e4 | 1022 | error = XFS_ERROR(E2BIG); |
0771fb45 | 1023 | return error; |
1da177e4 LT |
1024 | } |
1025 | ||
4249023a | 1026 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
1da177e4 LT |
1027 | d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks); |
1028 | if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) { | |
1029 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 1030 | return XFS_ERROR(E2BIG); |
1da177e4 LT |
1031 | } |
1032 | error = xfs_read_buf(mp, mp->m_logdev_targp, | |
1033 | d - XFS_FSB_TO_BB(mp, 1), | |
1034 | XFS_FSB_TO_BB(mp, 1), 0, &bp); | |
1035 | if (!error) { | |
1036 | xfs_buf_relse(bp); | |
1037 | } else { | |
1038 | cmn_err(CE_WARN, "XFS: size check 3 failed"); | |
0771fb45 | 1039 | if (error == ENOSPC) |
1da177e4 | 1040 | error = XFS_ERROR(E2BIG); |
0771fb45 ES |
1041 | return error; |
1042 | } | |
1043 | } | |
1044 | return 0; | |
1045 | } | |
1046 | ||
7d095257 CH |
1047 | /* |
1048 | * Clear the quotaflags in memory and in the superblock. | |
1049 | */ | |
1050 | int | |
1051 | xfs_mount_reset_sbqflags( | |
1052 | struct xfs_mount *mp) | |
1053 | { | |
1054 | int error; | |
1055 | struct xfs_trans *tp; | |
1056 | ||
1057 | mp->m_qflags = 0; | |
1058 | ||
1059 | /* | |
1060 | * It is OK to look at sb_qflags here in mount path, | |
1061 | * without m_sb_lock. | |
1062 | */ | |
1063 | if (mp->m_sb.sb_qflags == 0) | |
1064 | return 0; | |
1065 | spin_lock(&mp->m_sb_lock); | |
1066 | mp->m_sb.sb_qflags = 0; | |
1067 | spin_unlock(&mp->m_sb_lock); | |
1068 | ||
1069 | /* | |
1070 | * If the fs is readonly, let the incore superblock run | |
1071 | * with quotas off but don't flush the update out to disk | |
1072 | */ | |
1073 | if (mp->m_flags & XFS_MOUNT_RDONLY) | |
1074 | return 0; | |
1075 | ||
1076 | #ifdef QUOTADEBUG | |
1077 | xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes"); | |
1078 | #endif | |
1079 | ||
1080 | tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE); | |
1081 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, | |
1082 | XFS_DEFAULT_LOG_COUNT); | |
1083 | if (error) { | |
1084 | xfs_trans_cancel(tp, 0); | |
1085 | xfs_fs_cmn_err(CE_ALERT, mp, | |
1086 | "xfs_mount_reset_sbqflags: Superblock update failed!"); | |
1087 | return error; | |
1088 | } | |
1089 | ||
1090 | xfs_mod_sb(tp, XFS_SB_QFLAGS); | |
1091 | return xfs_trans_commit(tp, 0); | |
1092 | } | |
1093 | ||
0771fb45 | 1094 | /* |
0771fb45 ES |
1095 | * This function does the following on an initial mount of a file system: |
1096 | * - reads the superblock from disk and init the mount struct | |
1097 | * - if we're a 32-bit kernel, do a size check on the superblock | |
1098 | * so we don't mount terabyte filesystems | |
1099 | * - init mount struct realtime fields | |
1100 | * - allocate inode hash table for fs | |
1101 | * - init directory manager | |
1102 | * - perform recovery and init the log manager | |
1103 | */ | |
1104 | int | |
1105 | xfs_mountfs( | |
4249023a | 1106 | xfs_mount_t *mp) |
0771fb45 ES |
1107 | { |
1108 | xfs_sb_t *sbp = &(mp->m_sb); | |
1109 | xfs_inode_t *rip; | |
0771fb45 | 1110 | __uint64_t resblks; |
7d095257 CH |
1111 | uint quotamount = 0; |
1112 | uint quotaflags = 0; | |
0771fb45 ES |
1113 | int error = 0; |
1114 | ||
0771fb45 ES |
1115 | xfs_mount_common(mp, sbp); |
1116 | ||
ee1c0908 | 1117 | /* |
e6957ea4 ES |
1118 | * Check for a mismatched features2 values. Older kernels |
1119 | * read & wrote into the wrong sb offset for sb_features2 | |
1120 | * on some platforms due to xfs_sb_t not being 64bit size aligned | |
1121 | * when sb_features2 was added, which made older superblock | |
1122 | * reading/writing routines swap it as a 64-bit value. | |
ee1c0908 | 1123 | * |
e6957ea4 ES |
1124 | * For backwards compatibility, we make both slots equal. |
1125 | * | |
1126 | * If we detect a mismatched field, we OR the set bits into the | |
1127 | * existing features2 field in case it has already been modified; we | |
1128 | * don't want to lose any features. We then update the bad location | |
1129 | * with the ORed value so that older kernels will see any features2 | |
1130 | * flags, and mark the two fields as needing updates once the | |
1131 | * transaction subsystem is online. | |
ee1c0908 | 1132 | */ |
e6957ea4 | 1133 | if (xfs_sb_has_mismatched_features2(sbp)) { |
ee1c0908 DC |
1134 | cmn_err(CE_WARN, |
1135 | "XFS: correcting sb_features alignment problem"); | |
1136 | sbp->sb_features2 |= sbp->sb_bad_features2; | |
e6957ea4 | 1137 | sbp->sb_bad_features2 = sbp->sb_features2; |
7884bc86 | 1138 | mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2; |
e6957ea4 ES |
1139 | |
1140 | /* | |
1141 | * Re-check for ATTR2 in case it was found in bad_features2 | |
1142 | * slot. | |
1143 | */ | |
7c12f296 TS |
1144 | if (xfs_sb_version_hasattr2(&mp->m_sb) && |
1145 | !(mp->m_flags & XFS_MOUNT_NOATTR2)) | |
e6957ea4 | 1146 | mp->m_flags |= XFS_MOUNT_ATTR2; |
7c12f296 TS |
1147 | } |
1148 | ||
1149 | if (xfs_sb_version_hasattr2(&mp->m_sb) && | |
1150 | (mp->m_flags & XFS_MOUNT_NOATTR2)) { | |
1151 | xfs_sb_version_removeattr2(&mp->m_sb); | |
7884bc86 | 1152 | mp->m_update_flags |= XFS_SB_FEATURES2; |
e6957ea4 | 1153 | |
7c12f296 TS |
1154 | /* update sb_versionnum for the clearing of the morebits */ |
1155 | if (!sbp->sb_features2) | |
7884bc86 | 1156 | mp->m_update_flags |= XFS_SB_VERSIONNUM; |
ee1c0908 DC |
1157 | } |
1158 | ||
0771fb45 ES |
1159 | /* |
1160 | * Check if sb_agblocks is aligned at stripe boundary | |
1161 | * If sb_agblocks is NOT aligned turn off m_dalign since | |
1162 | * allocator alignment is within an ag, therefore ag has | |
1163 | * to be aligned at stripe boundary. | |
1164 | */ | |
7884bc86 | 1165 | error = xfs_update_alignment(mp); |
0771fb45 | 1166 | if (error) |
f9057e3d | 1167 | goto out; |
0771fb45 ES |
1168 | |
1169 | xfs_alloc_compute_maxlevels(mp); | |
1170 | xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK); | |
1171 | xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK); | |
1172 | xfs_ialloc_compute_maxlevels(mp); | |
1173 | ||
1174 | xfs_set_maxicount(mp); | |
1175 | ||
1176 | mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog); | |
1177 | ||
27174203 CH |
1178 | error = xfs_uuid_mount(mp); |
1179 | if (error) | |
1180 | goto out; | |
1da177e4 | 1181 | |
0771fb45 ES |
1182 | /* |
1183 | * Set the minimum read and write sizes | |
1184 | */ | |
1185 | xfs_set_rw_sizes(mp); | |
1186 | ||
1187 | /* | |
1188 | * Set the inode cluster size. | |
1189 | * This may still be overridden by the file system | |
1190 | * block size if it is larger than the chosen cluster size. | |
1191 | */ | |
1192 | mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE; | |
1193 | ||
1194 | /* | |
1195 | * Set inode alignment fields | |
1196 | */ | |
1197 | xfs_set_inoalignment(mp); | |
1198 | ||
1199 | /* | |
1200 | * Check that the data (and log if separate) are an ok size. | |
1201 | */ | |
4249023a | 1202 | error = xfs_check_sizes(mp); |
0771fb45 | 1203 | if (error) |
f9057e3d | 1204 | goto out_remove_uuid; |
0771fb45 | 1205 | |
1da177e4 LT |
1206 | /* |
1207 | * Initialize realtime fields in the mount structure | |
1208 | */ | |
0771fb45 ES |
1209 | error = xfs_rtmount_init(mp); |
1210 | if (error) { | |
1da177e4 | 1211 | cmn_err(CE_WARN, "XFS: RT mount failed"); |
f9057e3d | 1212 | goto out_remove_uuid; |
1da177e4 LT |
1213 | } |
1214 | ||
1da177e4 LT |
1215 | /* |
1216 | * Copies the low order bits of the timestamp and the randomly | |
1217 | * set "sequence" number out of a UUID. | |
1218 | */ | |
1219 | uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid); | |
1220 | ||
1da177e4 LT |
1221 | mp->m_dmevmask = 0; /* not persistent; set after each mount */ |
1222 | ||
f6c2d1fa | 1223 | xfs_dir_mount(mp); |
1da177e4 LT |
1224 | |
1225 | /* | |
1226 | * Initialize the attribute manager's entries. | |
1227 | */ | |
1228 | mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100; | |
1229 | ||
1230 | /* | |
1231 | * Initialize the precomputed transaction reservations values. | |
1232 | */ | |
1233 | xfs_trans_init(mp); | |
1234 | ||
1da177e4 LT |
1235 | /* |
1236 | * Allocate and initialize the per-ag data. | |
1237 | */ | |
1c1c6ebc DC |
1238 | spin_lock_init(&mp->m_perag_lock); |
1239 | INIT_RADIX_TREE(&mp->m_perag_tree, GFP_NOFS); | |
1240 | error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi); | |
1241 | if (error) { | |
1242 | cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error); | |
f9057e3d | 1243 | goto out_remove_uuid; |
1c1c6ebc | 1244 | } |
1da177e4 | 1245 | |
f9057e3d CH |
1246 | if (!sbp->sb_logblocks) { |
1247 | cmn_err(CE_WARN, "XFS: no log defined"); | |
1248 | XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp); | |
1249 | error = XFS_ERROR(EFSCORRUPTED); | |
1250 | goto out_free_perag; | |
1251 | } | |
1252 | ||
1da177e4 LT |
1253 | /* |
1254 | * log's mount-time initialization. Perform 1st part recovery if needed | |
1255 | */ | |
f9057e3d CH |
1256 | error = xfs_log_mount(mp, mp->m_logdev_targp, |
1257 | XFS_FSB_TO_DADDR(mp, sbp->sb_logstart), | |
1258 | XFS_FSB_TO_BB(mp, sbp->sb_logblocks)); | |
1259 | if (error) { | |
1260 | cmn_err(CE_WARN, "XFS: log mount failed"); | |
1261 | goto out_free_perag; | |
1da177e4 LT |
1262 | } |
1263 | ||
92821e2b DC |
1264 | /* |
1265 | * Now the log is mounted, we know if it was an unclean shutdown or | |
1266 | * not. If it was, with the first phase of recovery has completed, we | |
1267 | * have consistent AG blocks on disk. We have not recovered EFIs yet, | |
1268 | * but they are recovered transactionally in the second recovery phase | |
1269 | * later. | |
1270 | * | |
1271 | * Hence we can safely re-initialise incore superblock counters from | |
1272 | * the per-ag data. These may not be correct if the filesystem was not | |
1273 | * cleanly unmounted, so we need to wait for recovery to finish before | |
1274 | * doing this. | |
1275 | * | |
1276 | * If the filesystem was cleanly unmounted, then we can trust the | |
1277 | * values in the superblock to be correct and we don't need to do | |
1278 | * anything here. | |
1279 | * | |
1280 | * If we are currently making the filesystem, the initialisation will | |
1281 | * fail as the perag data is in an undefined state. | |
1282 | */ | |
92821e2b DC |
1283 | if (xfs_sb_version_haslazysbcount(&mp->m_sb) && |
1284 | !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) && | |
1285 | !mp->m_sb.sb_inprogress) { | |
1286 | error = xfs_initialize_perag_data(mp, sbp->sb_agcount); | |
f9057e3d CH |
1287 | if (error) |
1288 | goto out_free_perag; | |
92821e2b | 1289 | } |
f9057e3d | 1290 | |
1da177e4 LT |
1291 | /* |
1292 | * Get and sanity-check the root inode. | |
1293 | * Save the pointer to it in the mount structure. | |
1294 | */ | |
1295 | error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0); | |
1296 | if (error) { | |
1297 | cmn_err(CE_WARN, "XFS: failed to read root inode"); | |
f9057e3d | 1298 | goto out_log_dealloc; |
1da177e4 LT |
1299 | } |
1300 | ||
1301 | ASSERT(rip != NULL); | |
1da177e4 LT |
1302 | |
1303 | if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) { | |
1304 | cmn_err(CE_WARN, "XFS: corrupted root inode"); | |
b6574520 NS |
1305 | cmn_err(CE_WARN, "Device %s - root %llu is not a directory", |
1306 | XFS_BUFTARG_NAME(mp->m_ddev_targp), | |
1307 | (unsigned long long)rip->i_ino); | |
1da177e4 LT |
1308 | xfs_iunlock(rip, XFS_ILOCK_EXCL); |
1309 | XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW, | |
1310 | mp); | |
1311 | error = XFS_ERROR(EFSCORRUPTED); | |
f9057e3d | 1312 | goto out_rele_rip; |
1da177e4 LT |
1313 | } |
1314 | mp->m_rootip = rip; /* save it */ | |
1315 | ||
1316 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
1317 | ||
1318 | /* | |
1319 | * Initialize realtime inode pointers in the mount structure | |
1320 | */ | |
0771fb45 ES |
1321 | error = xfs_rtmount_inodes(mp); |
1322 | if (error) { | |
1da177e4 LT |
1323 | /* |
1324 | * Free up the root inode. | |
1325 | */ | |
1326 | cmn_err(CE_WARN, "XFS: failed to read RT inodes"); | |
f9057e3d | 1327 | goto out_rele_rip; |
1da177e4 LT |
1328 | } |
1329 | ||
1330 | /* | |
7884bc86 CH |
1331 | * If this is a read-only mount defer the superblock updates until |
1332 | * the next remount into writeable mode. Otherwise we would never | |
1333 | * perform the update e.g. for the root filesystem. | |
1da177e4 | 1334 | */ |
7884bc86 CH |
1335 | if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1336 | error = xfs_mount_log_sb(mp, mp->m_update_flags); | |
e5720eec DC |
1337 | if (error) { |
1338 | cmn_err(CE_WARN, "XFS: failed to write sb changes"); | |
b93b6e43 | 1339 | goto out_rtunmount; |
e5720eec DC |
1340 | } |
1341 | } | |
1da177e4 LT |
1342 | |
1343 | /* | |
1344 | * Initialise the XFS quota management subsystem for this mount | |
1345 | */ | |
7d095257 CH |
1346 | if (XFS_IS_QUOTA_RUNNING(mp)) { |
1347 | error = xfs_qm_newmount(mp, "amount, "aflags); | |
1348 | if (error) | |
1349 | goto out_rtunmount; | |
1350 | } else { | |
1351 | ASSERT(!XFS_IS_QUOTA_ON(mp)); | |
1352 | ||
1353 | /* | |
1354 | * If a file system had quotas running earlier, but decided to | |
1355 | * mount without -o uquota/pquota/gquota options, revoke the | |
1356 | * quotachecked license. | |
1357 | */ | |
1358 | if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) { | |
1359 | cmn_err(CE_NOTE, | |
1360 | "XFS: resetting qflags for filesystem %s", | |
1361 | mp->m_fsname); | |
1362 | ||
1363 | error = xfs_mount_reset_sbqflags(mp); | |
1364 | if (error) | |
1365 | return error; | |
1366 | } | |
1367 | } | |
1da177e4 LT |
1368 | |
1369 | /* | |
1370 | * Finish recovering the file system. This part needed to be | |
1371 | * delayed until after the root and real-time bitmap inodes | |
1372 | * were consistently read in. | |
1373 | */ | |
4249023a | 1374 | error = xfs_log_mount_finish(mp); |
1da177e4 LT |
1375 | if (error) { |
1376 | cmn_err(CE_WARN, "XFS: log mount finish failed"); | |
b93b6e43 | 1377 | goto out_rtunmount; |
1da177e4 LT |
1378 | } |
1379 | ||
1380 | /* | |
1381 | * Complete the quota initialisation, post-log-replay component. | |
1382 | */ | |
7d095257 CH |
1383 | if (quotamount) { |
1384 | ASSERT(mp->m_qflags == 0); | |
1385 | mp->m_qflags = quotaflags; | |
1386 | ||
1387 | xfs_qm_mount_quotas(mp); | |
1388 | } | |
1389 | ||
1390 | #if defined(DEBUG) && defined(XFS_LOUD_RECOVERY) | |
1391 | if (XFS_IS_QUOTA_ON(mp)) | |
1392 | xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas turned on"); | |
1393 | else | |
1394 | xfs_fs_cmn_err(CE_NOTE, mp, "Disk quotas not turned on"); | |
1395 | #endif | |
1da177e4 | 1396 | |
84e1e99f DC |
1397 | /* |
1398 | * Now we are mounted, reserve a small amount of unused space for | |
1399 | * privileged transactions. This is needed so that transaction | |
1400 | * space required for critical operations can dip into this pool | |
1401 | * when at ENOSPC. This is needed for operations like create with | |
1402 | * attr, unwritten extent conversion at ENOSPC, etc. Data allocations | |
1403 | * are not allowed to use this reserved space. | |
1404 | * | |
1405 | * We default to 5% or 1024 fsbs of space reserved, whichever is smaller. | |
1406 | * This may drive us straight to ENOSPC on mount, but that implies | |
714082bc | 1407 | * we were already there on the last unmount. Warn if this occurs. |
84e1e99f | 1408 | */ |
39726be2 CH |
1409 | resblks = mp->m_sb.sb_dblocks; |
1410 | do_div(resblks, 20); | |
1411 | resblks = min_t(__uint64_t, resblks, 1024); | |
714082bc DC |
1412 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1413 | if (error) | |
1414 | cmn_err(CE_WARN, "XFS: Unable to allocate reserve blocks. " | |
1415 | "Continuing without a reserve pool."); | |
84e1e99f | 1416 | |
1da177e4 LT |
1417 | return 0; |
1418 | ||
b93b6e43 CH |
1419 | out_rtunmount: |
1420 | xfs_rtunmount_inodes(mp); | |
f9057e3d | 1421 | out_rele_rip: |
43355099 | 1422 | IRELE(rip); |
f9057e3d | 1423 | out_log_dealloc: |
21b699c8 | 1424 | xfs_log_unmount(mp); |
f9057e3d | 1425 | out_free_perag: |
ff4f038c | 1426 | xfs_free_perag(mp); |
f9057e3d | 1427 | out_remove_uuid: |
27174203 | 1428 | xfs_uuid_unmount(mp); |
f9057e3d | 1429 | out: |
1da177e4 LT |
1430 | return error; |
1431 | } | |
1432 | ||
1433 | /* | |
1da177e4 LT |
1434 | * This flushes out the inodes,dquots and the superblock, unmounts the |
1435 | * log and makes sure that incore structures are freed. | |
1436 | */ | |
41b5c2e7 CH |
1437 | void |
1438 | xfs_unmountfs( | |
1439 | struct xfs_mount *mp) | |
1da177e4 | 1440 | { |
41b5c2e7 CH |
1441 | __uint64_t resblks; |
1442 | int error; | |
1da177e4 | 1443 | |
7d095257 | 1444 | xfs_qm_unmount_quotas(mp); |
b93b6e43 | 1445 | xfs_rtunmount_inodes(mp); |
77508ec8 CH |
1446 | IRELE(mp->m_rootip); |
1447 | ||
641c56fb DC |
1448 | /* |
1449 | * We can potentially deadlock here if we have an inode cluster | |
9da096fd | 1450 | * that has been freed has its buffer still pinned in memory because |
641c56fb DC |
1451 | * the transaction is still sitting in a iclog. The stale inodes |
1452 | * on that buffer will have their flush locks held until the | |
1453 | * transaction hits the disk and the callbacks run. the inode | |
1454 | * flush takes the flush lock unconditionally and with nothing to | |
1455 | * push out the iclog we will never get that unlocked. hence we | |
1456 | * need to force the log first. | |
1457 | */ | |
a14a348b | 1458 | xfs_log_force(mp, XFS_LOG_SYNC); |
abc10647 | 1459 | xfs_reclaim_inodes(mp, XFS_IFLUSH_ASYNC); |
1da177e4 | 1460 | |
7d095257 | 1461 | xfs_qm_unmount(mp); |
a357a121 | 1462 | |
1da177e4 LT |
1463 | /* |
1464 | * Flush out the log synchronously so that we know for sure | |
1465 | * that nothing is pinned. This is important because bflush() | |
1466 | * will skip pinned buffers. | |
1467 | */ | |
a14a348b | 1468 | xfs_log_force(mp, XFS_LOG_SYNC); |
1da177e4 LT |
1469 | |
1470 | xfs_binval(mp->m_ddev_targp); | |
1471 | if (mp->m_rtdev_targp) { | |
1472 | xfs_binval(mp->m_rtdev_targp); | |
1473 | } | |
1474 | ||
84e1e99f DC |
1475 | /* |
1476 | * Unreserve any blocks we have so that when we unmount we don't account | |
1477 | * the reserved free space as used. This is really only necessary for | |
1478 | * lazy superblock counting because it trusts the incore superblock | |
9da096fd | 1479 | * counters to be absolutely correct on clean unmount. |
84e1e99f DC |
1480 | * |
1481 | * We don't bother correcting this elsewhere for lazy superblock | |
1482 | * counting because on mount of an unclean filesystem we reconstruct the | |
1483 | * correct counter value and this is irrelevant. | |
1484 | * | |
1485 | * For non-lazy counter filesystems, this doesn't matter at all because | |
1486 | * we only every apply deltas to the superblock and hence the incore | |
1487 | * value does not matter.... | |
1488 | */ | |
1489 | resblks = 0; | |
714082bc DC |
1490 | error = xfs_reserve_blocks(mp, &resblks, NULL); |
1491 | if (error) | |
1492 | cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. " | |
1493 | "Freespace may not be correct on next mount."); | |
1494 | ||
e5720eec DC |
1495 | error = xfs_log_sbcount(mp, 1); |
1496 | if (error) | |
1497 | cmn_err(CE_WARN, "XFS: Unable to update superblock counters. " | |
1498 | "Freespace may not be correct on next mount."); | |
1da177e4 | 1499 | xfs_unmountfs_writesb(mp); |
1da177e4 | 1500 | xfs_unmountfs_wait(mp); /* wait for async bufs */ |
21b699c8 CH |
1501 | xfs_log_unmount_write(mp); |
1502 | xfs_log_unmount(mp); | |
27174203 | 1503 | xfs_uuid_unmount(mp); |
1da177e4 | 1504 | |
1550d0b0 | 1505 | #if defined(DEBUG) |
0ce4cfd4 | 1506 | xfs_errortag_clearall(mp, 0); |
1da177e4 | 1507 | #endif |
ff4f038c | 1508 | xfs_free_perag(mp); |
1da177e4 LT |
1509 | } |
1510 | ||
ba0f32d4 | 1511 | STATIC void |
1da177e4 LT |
1512 | xfs_unmountfs_wait(xfs_mount_t *mp) |
1513 | { | |
1514 | if (mp->m_logdev_targp != mp->m_ddev_targp) | |
1515 | xfs_wait_buftarg(mp->m_logdev_targp); | |
1516 | if (mp->m_rtdev_targp) | |
1517 | xfs_wait_buftarg(mp->m_rtdev_targp); | |
1518 | xfs_wait_buftarg(mp->m_ddev_targp); | |
1519 | } | |
1520 | ||
92821e2b DC |
1521 | int |
1522 | xfs_fs_writable(xfs_mount_t *mp) | |
1523 | { | |
b267ce99 | 1524 | return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) || |
bd186aa9 | 1525 | (mp->m_flags & XFS_MOUNT_RDONLY)); |
92821e2b DC |
1526 | } |
1527 | ||
1528 | /* | |
1529 | * xfs_log_sbcount | |
1530 | * | |
1531 | * Called either periodically to keep the on disk superblock values | |
1532 | * roughly up to date or from unmount to make sure the values are | |
1533 | * correct on a clean unmount. | |
1534 | * | |
1535 | * Note this code can be called during the process of freezing, so | |
1536 | * we may need to use the transaction allocator which does not not | |
1537 | * block when the transaction subsystem is in its frozen state. | |
1538 | */ | |
1539 | int | |
1540 | xfs_log_sbcount( | |
1541 | xfs_mount_t *mp, | |
1542 | uint sync) | |
1543 | { | |
1544 | xfs_trans_t *tp; | |
1545 | int error; | |
1546 | ||
1547 | if (!xfs_fs_writable(mp)) | |
1548 | return 0; | |
1549 | ||
d4d90b57 | 1550 | xfs_icsb_sync_counters(mp, 0); |
92821e2b DC |
1551 | |
1552 | /* | |
1553 | * we don't need to do this if we are updating the superblock | |
1554 | * counters on every modification. | |
1555 | */ | |
1556 | if (!xfs_sb_version_haslazysbcount(&mp->m_sb)) | |
1557 | return 0; | |
1558 | ||
80641dc6 | 1559 | tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP); |
92821e2b DC |
1560 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
1561 | XFS_DEFAULT_LOG_COUNT); | |
1562 | if (error) { | |
1563 | xfs_trans_cancel(tp, 0); | |
1564 | return error; | |
1565 | } | |
1566 | ||
1567 | xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS); | |
1568 | if (sync) | |
1569 | xfs_trans_set_sync(tp); | |
e5720eec DC |
1570 | error = xfs_trans_commit(tp, 0); |
1571 | return error; | |
92821e2b DC |
1572 | } |
1573 | ||
1da177e4 LT |
1574 | int |
1575 | xfs_unmountfs_writesb(xfs_mount_t *mp) | |
1576 | { | |
1577 | xfs_buf_t *sbp; | |
1da177e4 LT |
1578 | int error = 0; |
1579 | ||
1580 | /* | |
1581 | * skip superblock write if fs is read-only, or | |
1582 | * if we are doing a forced umount. | |
1583 | */ | |
bd186aa9 | 1584 | if (!((mp->m_flags & XFS_MOUNT_RDONLY) || |
1da177e4 | 1585 | XFS_FORCED_SHUTDOWN(mp))) { |
8d280b98 | 1586 | |
92821e2b | 1587 | sbp = xfs_getsb(mp, 0); |
8d280b98 | 1588 | |
1da177e4 LT |
1589 | XFS_BUF_UNDONE(sbp); |
1590 | XFS_BUF_UNREAD(sbp); | |
1591 | XFS_BUF_UNDELAYWRITE(sbp); | |
1592 | XFS_BUF_WRITE(sbp); | |
1593 | XFS_BUF_UNASYNC(sbp); | |
1594 | ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp); | |
1595 | xfsbdstrat(mp, sbp); | |
1da177e4 LT |
1596 | error = xfs_iowait(sbp); |
1597 | if (error) | |
1598 | xfs_ioerror_alert("xfs_unmountfs_writesb", | |
1599 | mp, sbp, XFS_BUF_ADDR(sbp)); | |
92821e2b | 1600 | xfs_buf_relse(sbp); |
1da177e4 | 1601 | } |
014c2544 | 1602 | return error; |
1da177e4 LT |
1603 | } |
1604 | ||
1605 | /* | |
1606 | * xfs_mod_sb() can be used to copy arbitrary changes to the | |
1607 | * in-core superblock into the superblock buffer to be logged. | |
1608 | * It does not provide the higher level of locking that is | |
1609 | * needed to protect the in-core superblock from concurrent | |
1610 | * access. | |
1611 | */ | |
1612 | void | |
1613 | xfs_mod_sb(xfs_trans_t *tp, __int64_t fields) | |
1614 | { | |
1615 | xfs_buf_t *bp; | |
1616 | int first; | |
1617 | int last; | |
1618 | xfs_mount_t *mp; | |
1da177e4 LT |
1619 | xfs_sb_field_t f; |
1620 | ||
1621 | ASSERT(fields); | |
1622 | if (!fields) | |
1623 | return; | |
1624 | mp = tp->t_mountp; | |
1625 | bp = xfs_trans_getsb(tp, mp, 0); | |
1da177e4 LT |
1626 | first = sizeof(xfs_sb_t); |
1627 | last = 0; | |
1628 | ||
1629 | /* translate/copy */ | |
1630 | ||
2bdf7cd0 | 1631 | xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields); |
1da177e4 LT |
1632 | |
1633 | /* find modified range */ | |
587aa0fe DC |
1634 | f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields); |
1635 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1636 | last = xfs_sb_info[f + 1].offset - 1; | |
1da177e4 LT |
1637 | |
1638 | f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields); | |
1639 | ASSERT((1LL << f) & XFS_SB_MOD_BITS); | |
1640 | first = xfs_sb_info[f].offset; | |
1641 | ||
1da177e4 LT |
1642 | xfs_trans_log_buf(tp, bp, first, last); |
1643 | } | |
d210a28c | 1644 | |
d210a28c | 1645 | |
1da177e4 LT |
1646 | /* |
1647 | * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply | |
1648 | * a delta to a specified field in the in-core superblock. Simply | |
1649 | * switch on the field indicated and apply the delta to that field. | |
1650 | * Fields are not allowed to dip below zero, so if the delta would | |
1651 | * do this do not apply it and return EINVAL. | |
1652 | * | |
3685c2a1 | 1653 | * The m_sb_lock must be held when this routine is called. |
1da177e4 | 1654 | */ |
d96f8f89 | 1655 | STATIC int |
20f4ebf2 DC |
1656 | xfs_mod_incore_sb_unlocked( |
1657 | xfs_mount_t *mp, | |
1658 | xfs_sb_field_t field, | |
1659 | int64_t delta, | |
1660 | int rsvd) | |
1da177e4 LT |
1661 | { |
1662 | int scounter; /* short counter for 32 bit fields */ | |
1663 | long long lcounter; /* long counter for 64 bit fields */ | |
1664 | long long res_used, rem; | |
1665 | ||
1666 | /* | |
1667 | * With the in-core superblock spin lock held, switch | |
1668 | * on the indicated field. Apply the delta to the | |
1669 | * proper field. If the fields value would dip below | |
1670 | * 0, then do not apply the delta and return EINVAL. | |
1671 | */ | |
1672 | switch (field) { | |
1673 | case XFS_SBS_ICOUNT: | |
1674 | lcounter = (long long)mp->m_sb.sb_icount; | |
1675 | lcounter += delta; | |
1676 | if (lcounter < 0) { | |
1677 | ASSERT(0); | |
014c2544 | 1678 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1679 | } |
1680 | mp->m_sb.sb_icount = lcounter; | |
014c2544 | 1681 | return 0; |
1da177e4 LT |
1682 | case XFS_SBS_IFREE: |
1683 | lcounter = (long long)mp->m_sb.sb_ifree; | |
1684 | lcounter += delta; | |
1685 | if (lcounter < 0) { | |
1686 | ASSERT(0); | |
014c2544 | 1687 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1688 | } |
1689 | mp->m_sb.sb_ifree = lcounter; | |
014c2544 | 1690 | return 0; |
1da177e4 | 1691 | case XFS_SBS_FDBLOCKS: |
4be536de DC |
1692 | lcounter = (long long) |
1693 | mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); | |
1da177e4 LT |
1694 | res_used = (long long)(mp->m_resblks - mp->m_resblks_avail); |
1695 | ||
1696 | if (delta > 0) { /* Putting blocks back */ | |
1697 | if (res_used > delta) { | |
1698 | mp->m_resblks_avail += delta; | |
1699 | } else { | |
1700 | rem = delta - res_used; | |
1701 | mp->m_resblks_avail = mp->m_resblks; | |
1702 | lcounter += rem; | |
1703 | } | |
1704 | } else { /* Taking blocks away */ | |
1705 | ||
1706 | lcounter += delta; | |
1707 | ||
1708 | /* | |
1709 | * If were out of blocks, use any available reserved blocks if | |
1710 | * were allowed to. | |
1711 | */ | |
1712 | ||
1713 | if (lcounter < 0) { | |
1714 | if (rsvd) { | |
1715 | lcounter = (long long)mp->m_resblks_avail + delta; | |
1716 | if (lcounter < 0) { | |
014c2544 | 1717 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1718 | } |
1719 | mp->m_resblks_avail = lcounter; | |
014c2544 | 1720 | return 0; |
1da177e4 | 1721 | } else { /* not reserved */ |
014c2544 | 1722 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1723 | } |
1724 | } | |
1725 | } | |
1726 | ||
4be536de | 1727 | mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
014c2544 | 1728 | return 0; |
1da177e4 LT |
1729 | case XFS_SBS_FREXTENTS: |
1730 | lcounter = (long long)mp->m_sb.sb_frextents; | |
1731 | lcounter += delta; | |
1732 | if (lcounter < 0) { | |
014c2544 | 1733 | return XFS_ERROR(ENOSPC); |
1da177e4 LT |
1734 | } |
1735 | mp->m_sb.sb_frextents = lcounter; | |
014c2544 | 1736 | return 0; |
1da177e4 LT |
1737 | case XFS_SBS_DBLOCKS: |
1738 | lcounter = (long long)mp->m_sb.sb_dblocks; | |
1739 | lcounter += delta; | |
1740 | if (lcounter < 0) { | |
1741 | ASSERT(0); | |
014c2544 | 1742 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1743 | } |
1744 | mp->m_sb.sb_dblocks = lcounter; | |
014c2544 | 1745 | return 0; |
1da177e4 LT |
1746 | case XFS_SBS_AGCOUNT: |
1747 | scounter = mp->m_sb.sb_agcount; | |
1748 | scounter += delta; | |
1749 | if (scounter < 0) { | |
1750 | ASSERT(0); | |
014c2544 | 1751 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1752 | } |
1753 | mp->m_sb.sb_agcount = scounter; | |
014c2544 | 1754 | return 0; |
1da177e4 LT |
1755 | case XFS_SBS_IMAX_PCT: |
1756 | scounter = mp->m_sb.sb_imax_pct; | |
1757 | scounter += delta; | |
1758 | if (scounter < 0) { | |
1759 | ASSERT(0); | |
014c2544 | 1760 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1761 | } |
1762 | mp->m_sb.sb_imax_pct = scounter; | |
014c2544 | 1763 | return 0; |
1da177e4 LT |
1764 | case XFS_SBS_REXTSIZE: |
1765 | scounter = mp->m_sb.sb_rextsize; | |
1766 | scounter += delta; | |
1767 | if (scounter < 0) { | |
1768 | ASSERT(0); | |
014c2544 | 1769 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1770 | } |
1771 | mp->m_sb.sb_rextsize = scounter; | |
014c2544 | 1772 | return 0; |
1da177e4 LT |
1773 | case XFS_SBS_RBMBLOCKS: |
1774 | scounter = mp->m_sb.sb_rbmblocks; | |
1775 | scounter += delta; | |
1776 | if (scounter < 0) { | |
1777 | ASSERT(0); | |
014c2544 | 1778 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1779 | } |
1780 | mp->m_sb.sb_rbmblocks = scounter; | |
014c2544 | 1781 | return 0; |
1da177e4 LT |
1782 | case XFS_SBS_RBLOCKS: |
1783 | lcounter = (long long)mp->m_sb.sb_rblocks; | |
1784 | lcounter += delta; | |
1785 | if (lcounter < 0) { | |
1786 | ASSERT(0); | |
014c2544 | 1787 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1788 | } |
1789 | mp->m_sb.sb_rblocks = lcounter; | |
014c2544 | 1790 | return 0; |
1da177e4 LT |
1791 | case XFS_SBS_REXTENTS: |
1792 | lcounter = (long long)mp->m_sb.sb_rextents; | |
1793 | lcounter += delta; | |
1794 | if (lcounter < 0) { | |
1795 | ASSERT(0); | |
014c2544 | 1796 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1797 | } |
1798 | mp->m_sb.sb_rextents = lcounter; | |
014c2544 | 1799 | return 0; |
1da177e4 LT |
1800 | case XFS_SBS_REXTSLOG: |
1801 | scounter = mp->m_sb.sb_rextslog; | |
1802 | scounter += delta; | |
1803 | if (scounter < 0) { | |
1804 | ASSERT(0); | |
014c2544 | 1805 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1806 | } |
1807 | mp->m_sb.sb_rextslog = scounter; | |
014c2544 | 1808 | return 0; |
1da177e4 LT |
1809 | default: |
1810 | ASSERT(0); | |
014c2544 | 1811 | return XFS_ERROR(EINVAL); |
1da177e4 LT |
1812 | } |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * xfs_mod_incore_sb() is used to change a field in the in-core | |
1817 | * superblock structure by the specified delta. This modification | |
3685c2a1 | 1818 | * is protected by the m_sb_lock. Just use the xfs_mod_incore_sb_unlocked() |
1da177e4 LT |
1819 | * routine to do the work. |
1820 | */ | |
1821 | int | |
20f4ebf2 DC |
1822 | xfs_mod_incore_sb( |
1823 | xfs_mount_t *mp, | |
1824 | xfs_sb_field_t field, | |
1825 | int64_t delta, | |
1826 | int rsvd) | |
1da177e4 | 1827 | { |
1da177e4 LT |
1828 | int status; |
1829 | ||
8d280b98 DC |
1830 | /* check for per-cpu counters */ |
1831 | switch (field) { | |
1832 | #ifdef HAVE_PERCPU_SB | |
1833 | case XFS_SBS_ICOUNT: | |
1834 | case XFS_SBS_IFREE: | |
1835 | case XFS_SBS_FDBLOCKS: | |
1836 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
1837 | status = xfs_icsb_modify_counters(mp, field, | |
1838 | delta, rsvd); | |
1839 | break; | |
1840 | } | |
1841 | /* FALLTHROUGH */ | |
1842 | #endif | |
1843 | default: | |
3685c2a1 | 1844 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 1845 | status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 1846 | spin_unlock(&mp->m_sb_lock); |
8d280b98 DC |
1847 | break; |
1848 | } | |
1849 | ||
014c2544 | 1850 | return status; |
1da177e4 LT |
1851 | } |
1852 | ||
1853 | /* | |
1854 | * xfs_mod_incore_sb_batch() is used to change more than one field | |
1855 | * in the in-core superblock structure at a time. This modification | |
1856 | * is protected by a lock internal to this module. The fields and | |
1857 | * changes to those fields are specified in the array of xfs_mod_sb | |
1858 | * structures passed in. | |
1859 | * | |
1860 | * Either all of the specified deltas will be applied or none of | |
1861 | * them will. If any modified field dips below 0, then all modifications | |
1862 | * will be backed out and EINVAL will be returned. | |
1863 | */ | |
1864 | int | |
1865 | xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd) | |
1866 | { | |
1da177e4 LT |
1867 | int status=0; |
1868 | xfs_mod_sb_t *msbp; | |
1869 | ||
1870 | /* | |
1871 | * Loop through the array of mod structures and apply each | |
1872 | * individually. If any fail, then back out all those | |
1873 | * which have already been applied. Do all of this within | |
3685c2a1 | 1874 | * the scope of the m_sb_lock so that all of the changes will |
1da177e4 LT |
1875 | * be atomic. |
1876 | */ | |
3685c2a1 | 1877 | spin_lock(&mp->m_sb_lock); |
1da177e4 LT |
1878 | msbp = &msb[0]; |
1879 | for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) { | |
1880 | /* | |
1881 | * Apply the delta at index n. If it fails, break | |
1882 | * from the loop so we'll fall into the undo loop | |
1883 | * below. | |
1884 | */ | |
8d280b98 DC |
1885 | switch (msbp->msb_field) { |
1886 | #ifdef HAVE_PERCPU_SB | |
1887 | case XFS_SBS_ICOUNT: | |
1888 | case XFS_SBS_IFREE: | |
1889 | case XFS_SBS_FDBLOCKS: | |
1890 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1891 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1892 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1893 | msbp->msb_field, |
1894 | msbp->msb_delta, rsvd); | |
3685c2a1 | 1895 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1896 | break; |
1897 | } | |
1898 | /* FALLTHROUGH */ | |
1899 | #endif | |
1900 | default: | |
1901 | status = xfs_mod_incore_sb_unlocked(mp, | |
1902 | msbp->msb_field, | |
1903 | msbp->msb_delta, rsvd); | |
1904 | break; | |
1905 | } | |
1906 | ||
1da177e4 LT |
1907 | if (status != 0) { |
1908 | break; | |
1909 | } | |
1910 | } | |
1911 | ||
1912 | /* | |
1913 | * If we didn't complete the loop above, then back out | |
1914 | * any changes made to the superblock. If you add code | |
1915 | * between the loop above and here, make sure that you | |
1916 | * preserve the value of status. Loop back until | |
1917 | * we step below the beginning of the array. Make sure | |
1918 | * we don't touch anything back there. | |
1919 | */ | |
1920 | if (status != 0) { | |
1921 | msbp--; | |
1922 | while (msbp >= msb) { | |
8d280b98 DC |
1923 | switch (msbp->msb_field) { |
1924 | #ifdef HAVE_PERCPU_SB | |
1925 | case XFS_SBS_ICOUNT: | |
1926 | case XFS_SBS_IFREE: | |
1927 | case XFS_SBS_FDBLOCKS: | |
1928 | if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) { | |
3685c2a1 | 1929 | spin_unlock(&mp->m_sb_lock); |
20b64285 | 1930 | status = xfs_icsb_modify_counters(mp, |
8d280b98 DC |
1931 | msbp->msb_field, |
1932 | -(msbp->msb_delta), | |
1933 | rsvd); | |
3685c2a1 | 1934 | spin_lock(&mp->m_sb_lock); |
8d280b98 DC |
1935 | break; |
1936 | } | |
1937 | /* FALLTHROUGH */ | |
1938 | #endif | |
1939 | default: | |
1940 | status = xfs_mod_incore_sb_unlocked(mp, | |
1941 | msbp->msb_field, | |
1942 | -(msbp->msb_delta), | |
1943 | rsvd); | |
1944 | break; | |
1945 | } | |
1da177e4 LT |
1946 | ASSERT(status == 0); |
1947 | msbp--; | |
1948 | } | |
1949 | } | |
3685c2a1 | 1950 | spin_unlock(&mp->m_sb_lock); |
014c2544 | 1951 | return status; |
1da177e4 LT |
1952 | } |
1953 | ||
1954 | /* | |
1955 | * xfs_getsb() is called to obtain the buffer for the superblock. | |
1956 | * The buffer is returned locked and read in from disk. | |
1957 | * The buffer should be released with a call to xfs_brelse(). | |
1958 | * | |
1959 | * If the flags parameter is BUF_TRYLOCK, then we'll only return | |
1960 | * the superblock buffer if it can be locked without sleeping. | |
1961 | * If it can't then we'll return NULL. | |
1962 | */ | |
1963 | xfs_buf_t * | |
1964 | xfs_getsb( | |
1965 | xfs_mount_t *mp, | |
1966 | int flags) | |
1967 | { | |
1968 | xfs_buf_t *bp; | |
1969 | ||
1970 | ASSERT(mp->m_sb_bp != NULL); | |
1971 | bp = mp->m_sb_bp; | |
0cadda1c | 1972 | if (flags & XBF_TRYLOCK) { |
1da177e4 LT |
1973 | if (!XFS_BUF_CPSEMA(bp)) { |
1974 | return NULL; | |
1975 | } | |
1976 | } else { | |
1977 | XFS_BUF_PSEMA(bp, PRIBIO); | |
1978 | } | |
1979 | XFS_BUF_HOLD(bp); | |
1980 | ASSERT(XFS_BUF_ISDONE(bp)); | |
014c2544 | 1981 | return bp; |
1da177e4 LT |
1982 | } |
1983 | ||
1984 | /* | |
1985 | * Used to free the superblock along various error paths. | |
1986 | */ | |
1987 | void | |
1988 | xfs_freesb( | |
1989 | xfs_mount_t *mp) | |
1990 | { | |
1991 | xfs_buf_t *bp; | |
1992 | ||
1993 | /* | |
1994 | * Use xfs_getsb() so that the buffer will be locked | |
1995 | * when we call xfs_buf_relse(). | |
1996 | */ | |
1997 | bp = xfs_getsb(mp, 0); | |
1998 | XFS_BUF_UNMANAGE(bp); | |
1999 | xfs_buf_relse(bp); | |
2000 | mp->m_sb_bp = NULL; | |
2001 | } | |
2002 | ||
1da177e4 LT |
2003 | /* |
2004 | * Used to log changes to the superblock unit and width fields which could | |
e6957ea4 ES |
2005 | * be altered by the mount options, as well as any potential sb_features2 |
2006 | * fixup. Only the first superblock is updated. | |
1da177e4 | 2007 | */ |
7884bc86 | 2008 | int |
ee1c0908 | 2009 | xfs_mount_log_sb( |
1da177e4 LT |
2010 | xfs_mount_t *mp, |
2011 | __int64_t fields) | |
2012 | { | |
2013 | xfs_trans_t *tp; | |
e5720eec | 2014 | int error; |
1da177e4 | 2015 | |
ee1c0908 | 2016 | ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID | |
4b166de0 DC |
2017 | XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 | |
2018 | XFS_SB_VERSIONNUM)); | |
1da177e4 LT |
2019 | |
2020 | tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT); | |
e5720eec DC |
2021 | error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0, |
2022 | XFS_DEFAULT_LOG_COUNT); | |
2023 | if (error) { | |
1da177e4 | 2024 | xfs_trans_cancel(tp, 0); |
e5720eec | 2025 | return error; |
1da177e4 LT |
2026 | } |
2027 | xfs_mod_sb(tp, fields); | |
e5720eec DC |
2028 | error = xfs_trans_commit(tp, 0); |
2029 | return error; | |
1da177e4 | 2030 | } |
8d280b98 DC |
2031 | |
2032 | ||
2033 | #ifdef HAVE_PERCPU_SB | |
2034 | /* | |
2035 | * Per-cpu incore superblock counters | |
2036 | * | |
2037 | * Simple concept, difficult implementation | |
2038 | * | |
2039 | * Basically, replace the incore superblock counters with a distributed per cpu | |
2040 | * counter for contended fields (e.g. free block count). | |
2041 | * | |
2042 | * Difficulties arise in that the incore sb is used for ENOSPC checking, and | |
2043 | * hence needs to be accurately read when we are running low on space. Hence | |
2044 | * there is a method to enable and disable the per-cpu counters based on how | |
2045 | * much "stuff" is available in them. | |
2046 | * | |
2047 | * Basically, a counter is enabled if there is enough free resource to justify | |
2048 | * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local | |
2049 | * ENOSPC), then we disable the counters to synchronise all callers and | |
2050 | * re-distribute the available resources. | |
2051 | * | |
2052 | * If, once we redistributed the available resources, we still get a failure, | |
2053 | * we disable the per-cpu counter and go through the slow path. | |
2054 | * | |
2055 | * The slow path is the current xfs_mod_incore_sb() function. This means that | |
9da096fd | 2056 | * when we disable a per-cpu counter, we need to drain its resources back to |
8d280b98 DC |
2057 | * the global superblock. We do this after disabling the counter to prevent |
2058 | * more threads from queueing up on the counter. | |
2059 | * | |
2060 | * Essentially, this means that we still need a lock in the fast path to enable | |
2061 | * synchronisation between the global counters and the per-cpu counters. This | |
2062 | * is not a problem because the lock will be local to a CPU almost all the time | |
2063 | * and have little contention except when we get to ENOSPC conditions. | |
2064 | * | |
2065 | * Basically, this lock becomes a barrier that enables us to lock out the fast | |
2066 | * path while we do things like enabling and disabling counters and | |
2067 | * synchronising the counters. | |
2068 | * | |
2069 | * Locking rules: | |
2070 | * | |
3685c2a1 | 2071 | * 1. m_sb_lock before picking up per-cpu locks |
8d280b98 | 2072 | * 2. per-cpu locks always picked up via for_each_online_cpu() order |
3685c2a1 | 2073 | * 3. accurate counter sync requires m_sb_lock + per cpu locks |
8d280b98 | 2074 | * 4. modifying per-cpu counters requires holding per-cpu lock |
3685c2a1 ES |
2075 | * 5. modifying global counters requires holding m_sb_lock |
2076 | * 6. enabling or disabling a counter requires holding the m_sb_lock | |
8d280b98 DC |
2077 | * and _none_ of the per-cpu locks. |
2078 | * | |
2079 | * Disabled counters are only ever re-enabled by a balance operation | |
2080 | * that results in more free resources per CPU than a given threshold. | |
2081 | * To ensure counters don't remain disabled, they are rebalanced when | |
2082 | * the global resource goes above a higher threshold (i.e. some hysteresis | |
2083 | * is present to prevent thrashing). | |
e8234a68 DC |
2084 | */ |
2085 | ||
5a67e4c5 | 2086 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2087 | /* |
2088 | * hot-plug CPU notifier support. | |
8d280b98 | 2089 | * |
5a67e4c5 CS |
2090 | * We need a notifier per filesystem as we need to be able to identify |
2091 | * the filesystem to balance the counters out. This is achieved by | |
2092 | * having a notifier block embedded in the xfs_mount_t and doing pointer | |
2093 | * magic to get the mount pointer from the notifier block address. | |
8d280b98 | 2094 | */ |
e8234a68 DC |
2095 | STATIC int |
2096 | xfs_icsb_cpu_notify( | |
2097 | struct notifier_block *nfb, | |
2098 | unsigned long action, | |
2099 | void *hcpu) | |
2100 | { | |
2101 | xfs_icsb_cnts_t *cntp; | |
2102 | xfs_mount_t *mp; | |
e8234a68 DC |
2103 | |
2104 | mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier); | |
2105 | cntp = (xfs_icsb_cnts_t *) | |
2106 | per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu); | |
2107 | switch (action) { | |
2108 | case CPU_UP_PREPARE: | |
8bb78442 | 2109 | case CPU_UP_PREPARE_FROZEN: |
e8234a68 DC |
2110 | /* Easy Case - initialize the area and locks, and |
2111 | * then rebalance when online does everything else for us. */ | |
01e1b69c | 2112 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 DC |
2113 | break; |
2114 | case CPU_ONLINE: | |
8bb78442 | 2115 | case CPU_ONLINE_FROZEN: |
03135cf7 | 2116 | xfs_icsb_lock(mp); |
45af6c6d CH |
2117 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2118 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2119 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
03135cf7 | 2120 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2121 | break; |
2122 | case CPU_DEAD: | |
8bb78442 | 2123 | case CPU_DEAD_FROZEN: |
e8234a68 DC |
2124 | /* Disable all the counters, then fold the dead cpu's |
2125 | * count into the total on the global superblock and | |
2126 | * re-enable the counters. */ | |
03135cf7 | 2127 | xfs_icsb_lock(mp); |
3685c2a1 | 2128 | spin_lock(&mp->m_sb_lock); |
e8234a68 DC |
2129 | xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT); |
2130 | xfs_icsb_disable_counter(mp, XFS_SBS_IFREE); | |
2131 | xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS); | |
2132 | ||
2133 | mp->m_sb.sb_icount += cntp->icsb_icount; | |
2134 | mp->m_sb.sb_ifree += cntp->icsb_ifree; | |
2135 | mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks; | |
2136 | ||
01e1b69c | 2137 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
e8234a68 | 2138 | |
45af6c6d CH |
2139 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0); |
2140 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0); | |
2141 | xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0); | |
3685c2a1 | 2142 | spin_unlock(&mp->m_sb_lock); |
03135cf7 | 2143 | xfs_icsb_unlock(mp); |
e8234a68 DC |
2144 | break; |
2145 | } | |
2146 | ||
2147 | return NOTIFY_OK; | |
2148 | } | |
5a67e4c5 | 2149 | #endif /* CONFIG_HOTPLUG_CPU */ |
e8234a68 | 2150 | |
8d280b98 DC |
2151 | int |
2152 | xfs_icsb_init_counters( | |
2153 | xfs_mount_t *mp) | |
2154 | { | |
2155 | xfs_icsb_cnts_t *cntp; | |
2156 | int i; | |
2157 | ||
2158 | mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t); | |
2159 | if (mp->m_sb_cnts == NULL) | |
2160 | return -ENOMEM; | |
2161 | ||
5a67e4c5 | 2162 | #ifdef CONFIG_HOTPLUG_CPU |
e8234a68 DC |
2163 | mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify; |
2164 | mp->m_icsb_notifier.priority = 0; | |
5a67e4c5 CS |
2165 | register_hotcpu_notifier(&mp->m_icsb_notifier); |
2166 | #endif /* CONFIG_HOTPLUG_CPU */ | |
e8234a68 | 2167 | |
8d280b98 DC |
2168 | for_each_online_cpu(i) { |
2169 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2170 | memset(cntp, 0, sizeof(xfs_icsb_cnts_t)); |
8d280b98 | 2171 | } |
20b64285 DC |
2172 | |
2173 | mutex_init(&mp->m_icsb_mutex); | |
2174 | ||
8d280b98 DC |
2175 | /* |
2176 | * start with all counters disabled so that the | |
2177 | * initial balance kicks us off correctly | |
2178 | */ | |
2179 | mp->m_icsb_counters = -1; | |
2180 | return 0; | |
2181 | } | |
2182 | ||
5478eead LM |
2183 | void |
2184 | xfs_icsb_reinit_counters( | |
2185 | xfs_mount_t *mp) | |
2186 | { | |
2187 | xfs_icsb_lock(mp); | |
2188 | /* | |
2189 | * start with all counters disabled so that the | |
2190 | * initial balance kicks us off correctly | |
2191 | */ | |
2192 | mp->m_icsb_counters = -1; | |
45af6c6d CH |
2193 | xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0); |
2194 | xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0); | |
2195 | xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0); | |
5478eead LM |
2196 | xfs_icsb_unlock(mp); |
2197 | } | |
2198 | ||
c962fb79 | 2199 | void |
8d280b98 DC |
2200 | xfs_icsb_destroy_counters( |
2201 | xfs_mount_t *mp) | |
2202 | { | |
e8234a68 | 2203 | if (mp->m_sb_cnts) { |
5a67e4c5 | 2204 | unregister_hotcpu_notifier(&mp->m_icsb_notifier); |
8d280b98 | 2205 | free_percpu(mp->m_sb_cnts); |
e8234a68 | 2206 | } |
03135cf7 | 2207 | mutex_destroy(&mp->m_icsb_mutex); |
8d280b98 DC |
2208 | } |
2209 | ||
b8f82a4a | 2210 | STATIC void |
01e1b69c DC |
2211 | xfs_icsb_lock_cntr( |
2212 | xfs_icsb_cnts_t *icsbp) | |
2213 | { | |
2214 | while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) { | |
2215 | ndelay(1000); | |
2216 | } | |
2217 | } | |
2218 | ||
b8f82a4a | 2219 | STATIC void |
01e1b69c DC |
2220 | xfs_icsb_unlock_cntr( |
2221 | xfs_icsb_cnts_t *icsbp) | |
2222 | { | |
2223 | clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags); | |
2224 | } | |
2225 | ||
8d280b98 | 2226 | |
b8f82a4a | 2227 | STATIC void |
8d280b98 DC |
2228 | xfs_icsb_lock_all_counters( |
2229 | xfs_mount_t *mp) | |
2230 | { | |
2231 | xfs_icsb_cnts_t *cntp; | |
2232 | int i; | |
2233 | ||
2234 | for_each_online_cpu(i) { | |
2235 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2236 | xfs_icsb_lock_cntr(cntp); |
8d280b98 DC |
2237 | } |
2238 | } | |
2239 | ||
b8f82a4a | 2240 | STATIC void |
8d280b98 DC |
2241 | xfs_icsb_unlock_all_counters( |
2242 | xfs_mount_t *mp) | |
2243 | { | |
2244 | xfs_icsb_cnts_t *cntp; | |
2245 | int i; | |
2246 | ||
2247 | for_each_online_cpu(i) { | |
2248 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
01e1b69c | 2249 | xfs_icsb_unlock_cntr(cntp); |
8d280b98 DC |
2250 | } |
2251 | } | |
2252 | ||
2253 | STATIC void | |
2254 | xfs_icsb_count( | |
2255 | xfs_mount_t *mp, | |
2256 | xfs_icsb_cnts_t *cnt, | |
2257 | int flags) | |
2258 | { | |
2259 | xfs_icsb_cnts_t *cntp; | |
2260 | int i; | |
2261 | ||
2262 | memset(cnt, 0, sizeof(xfs_icsb_cnts_t)); | |
2263 | ||
2264 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2265 | xfs_icsb_lock_all_counters(mp); | |
2266 | ||
2267 | for_each_online_cpu(i) { | |
2268 | cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i); | |
2269 | cnt->icsb_icount += cntp->icsb_icount; | |
2270 | cnt->icsb_ifree += cntp->icsb_ifree; | |
2271 | cnt->icsb_fdblocks += cntp->icsb_fdblocks; | |
2272 | } | |
2273 | ||
2274 | if (!(flags & XFS_ICSB_LAZY_COUNT)) | |
2275 | xfs_icsb_unlock_all_counters(mp); | |
2276 | } | |
2277 | ||
2278 | STATIC int | |
2279 | xfs_icsb_counter_disabled( | |
2280 | xfs_mount_t *mp, | |
2281 | xfs_sb_field_t field) | |
2282 | { | |
2283 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2284 | return test_bit(field, &mp->m_icsb_counters); | |
2285 | } | |
2286 | ||
36fbe6e6 | 2287 | STATIC void |
8d280b98 DC |
2288 | xfs_icsb_disable_counter( |
2289 | xfs_mount_t *mp, | |
2290 | xfs_sb_field_t field) | |
2291 | { | |
2292 | xfs_icsb_cnts_t cnt; | |
2293 | ||
2294 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2295 | ||
20b64285 DC |
2296 | /* |
2297 | * If we are already disabled, then there is nothing to do | |
2298 | * here. We check before locking all the counters to avoid | |
2299 | * the expensive lock operation when being called in the | |
2300 | * slow path and the counter is already disabled. This is | |
2301 | * safe because the only time we set or clear this state is under | |
2302 | * the m_icsb_mutex. | |
2303 | */ | |
2304 | if (xfs_icsb_counter_disabled(mp, field)) | |
36fbe6e6 | 2305 | return; |
20b64285 | 2306 | |
8d280b98 DC |
2307 | xfs_icsb_lock_all_counters(mp); |
2308 | if (!test_and_set_bit(field, &mp->m_icsb_counters)) { | |
2309 | /* drain back to superblock */ | |
2310 | ||
ce46193b | 2311 | xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT); |
8d280b98 DC |
2312 | switch(field) { |
2313 | case XFS_SBS_ICOUNT: | |
2314 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2315 | break; | |
2316 | case XFS_SBS_IFREE: | |
2317 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2318 | break; | |
2319 | case XFS_SBS_FDBLOCKS: | |
2320 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
2321 | break; | |
2322 | default: | |
2323 | BUG(); | |
2324 | } | |
2325 | } | |
2326 | ||
2327 | xfs_icsb_unlock_all_counters(mp); | |
8d280b98 DC |
2328 | } |
2329 | ||
2330 | STATIC void | |
2331 | xfs_icsb_enable_counter( | |
2332 | xfs_mount_t *mp, | |
2333 | xfs_sb_field_t field, | |
2334 | uint64_t count, | |
2335 | uint64_t resid) | |
2336 | { | |
2337 | xfs_icsb_cnts_t *cntp; | |
2338 | int i; | |
2339 | ||
2340 | ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS)); | |
2341 | ||
2342 | xfs_icsb_lock_all_counters(mp); | |
2343 | for_each_online_cpu(i) { | |
2344 | cntp = per_cpu_ptr(mp->m_sb_cnts, i); | |
2345 | switch (field) { | |
2346 | case XFS_SBS_ICOUNT: | |
2347 | cntp->icsb_icount = count + resid; | |
2348 | break; | |
2349 | case XFS_SBS_IFREE: | |
2350 | cntp->icsb_ifree = count + resid; | |
2351 | break; | |
2352 | case XFS_SBS_FDBLOCKS: | |
2353 | cntp->icsb_fdblocks = count + resid; | |
2354 | break; | |
2355 | default: | |
2356 | BUG(); | |
2357 | break; | |
2358 | } | |
2359 | resid = 0; | |
2360 | } | |
2361 | clear_bit(field, &mp->m_icsb_counters); | |
2362 | xfs_icsb_unlock_all_counters(mp); | |
2363 | } | |
2364 | ||
dbcabad1 | 2365 | void |
d4d90b57 | 2366 | xfs_icsb_sync_counters_locked( |
8d280b98 DC |
2367 | xfs_mount_t *mp, |
2368 | int flags) | |
2369 | { | |
2370 | xfs_icsb_cnts_t cnt; | |
8d280b98 | 2371 | |
8d280b98 DC |
2372 | xfs_icsb_count(mp, &cnt, flags); |
2373 | ||
8d280b98 DC |
2374 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT)) |
2375 | mp->m_sb.sb_icount = cnt.icsb_icount; | |
2376 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE)) | |
2377 | mp->m_sb.sb_ifree = cnt.icsb_ifree; | |
2378 | if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS)) | |
2379 | mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks; | |
8d280b98 DC |
2380 | } |
2381 | ||
2382 | /* | |
2383 | * Accurate update of per-cpu counters to incore superblock | |
2384 | */ | |
d4d90b57 | 2385 | void |
8d280b98 | 2386 | xfs_icsb_sync_counters( |
d4d90b57 CH |
2387 | xfs_mount_t *mp, |
2388 | int flags) | |
8d280b98 | 2389 | { |
d4d90b57 CH |
2390 | spin_lock(&mp->m_sb_lock); |
2391 | xfs_icsb_sync_counters_locked(mp, flags); | |
2392 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * Balance and enable/disable counters as necessary. | |
2397 | * | |
20b64285 DC |
2398 | * Thresholds for re-enabling counters are somewhat magic. inode counts are |
2399 | * chosen to be the same number as single on disk allocation chunk per CPU, and | |
2400 | * free blocks is something far enough zero that we aren't going thrash when we | |
2401 | * get near ENOSPC. We also need to supply a minimum we require per cpu to | |
2402 | * prevent looping endlessly when xfs_alloc_space asks for more than will | |
2403 | * be distributed to a single CPU but each CPU has enough blocks to be | |
2404 | * reenabled. | |
2405 | * | |
2406 | * Note that we can be called when counters are already disabled. | |
2407 | * xfs_icsb_disable_counter() optimises the counter locking in this case to | |
2408 | * prevent locking every per-cpu counter needlessly. | |
8d280b98 | 2409 | */ |
20b64285 DC |
2410 | |
2411 | #define XFS_ICSB_INO_CNTR_REENABLE (uint64_t)64 | |
4be536de | 2412 | #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \ |
20b64285 | 2413 | (uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp)) |
8d280b98 | 2414 | STATIC void |
45af6c6d | 2415 | xfs_icsb_balance_counter_locked( |
8d280b98 DC |
2416 | xfs_mount_t *mp, |
2417 | xfs_sb_field_t field, | |
20b64285 | 2418 | int min_per_cpu) |
8d280b98 | 2419 | { |
6fdf8ccc | 2420 | uint64_t count, resid; |
8d280b98 | 2421 | int weight = num_online_cpus(); |
20b64285 | 2422 | uint64_t min = (uint64_t)min_per_cpu; |
8d280b98 | 2423 | |
8d280b98 DC |
2424 | /* disable counter and sync counter */ |
2425 | xfs_icsb_disable_counter(mp, field); | |
2426 | ||
2427 | /* update counters - first CPU gets residual*/ | |
2428 | switch (field) { | |
2429 | case XFS_SBS_ICOUNT: | |
2430 | count = mp->m_sb.sb_icount; | |
2431 | resid = do_div(count, weight); | |
20b64285 | 2432 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2433 | return; |
8d280b98 DC |
2434 | break; |
2435 | case XFS_SBS_IFREE: | |
2436 | count = mp->m_sb.sb_ifree; | |
2437 | resid = do_div(count, weight); | |
20b64285 | 2438 | if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE)) |
45af6c6d | 2439 | return; |
8d280b98 DC |
2440 | break; |
2441 | case XFS_SBS_FDBLOCKS: | |
2442 | count = mp->m_sb.sb_fdblocks; | |
2443 | resid = do_div(count, weight); | |
20b64285 | 2444 | if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp))) |
45af6c6d | 2445 | return; |
8d280b98 DC |
2446 | break; |
2447 | default: | |
2448 | BUG(); | |
6fdf8ccc | 2449 | count = resid = 0; /* quiet, gcc */ |
8d280b98 DC |
2450 | break; |
2451 | } | |
2452 | ||
2453 | xfs_icsb_enable_counter(mp, field, count, resid); | |
45af6c6d CH |
2454 | } |
2455 | ||
2456 | STATIC void | |
2457 | xfs_icsb_balance_counter( | |
2458 | xfs_mount_t *mp, | |
2459 | xfs_sb_field_t fields, | |
2460 | int min_per_cpu) | |
2461 | { | |
2462 | spin_lock(&mp->m_sb_lock); | |
2463 | xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu); | |
2464 | spin_unlock(&mp->m_sb_lock); | |
8d280b98 DC |
2465 | } |
2466 | ||
a8272ce0 | 2467 | STATIC int |
20b64285 | 2468 | xfs_icsb_modify_counters( |
8d280b98 DC |
2469 | xfs_mount_t *mp, |
2470 | xfs_sb_field_t field, | |
20f4ebf2 | 2471 | int64_t delta, |
20b64285 | 2472 | int rsvd) |
8d280b98 DC |
2473 | { |
2474 | xfs_icsb_cnts_t *icsbp; | |
2475 | long long lcounter; /* long counter for 64 bit fields */ | |
7a9e02d6 | 2476 | int ret = 0; |
8d280b98 | 2477 | |
20b64285 | 2478 | might_sleep(); |
8d280b98 | 2479 | again: |
7a9e02d6 CL |
2480 | preempt_disable(); |
2481 | icsbp = this_cpu_ptr(mp->m_sb_cnts); | |
20b64285 DC |
2482 | |
2483 | /* | |
2484 | * if the counter is disabled, go to slow path | |
2485 | */ | |
8d280b98 DC |
2486 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) |
2487 | goto slow_path; | |
20b64285 DC |
2488 | xfs_icsb_lock_cntr(icsbp); |
2489 | if (unlikely(xfs_icsb_counter_disabled(mp, field))) { | |
2490 | xfs_icsb_unlock_cntr(icsbp); | |
2491 | goto slow_path; | |
2492 | } | |
8d280b98 DC |
2493 | |
2494 | switch (field) { | |
2495 | case XFS_SBS_ICOUNT: | |
2496 | lcounter = icsbp->icsb_icount; | |
2497 | lcounter += delta; | |
2498 | if (unlikely(lcounter < 0)) | |
20b64285 | 2499 | goto balance_counter; |
8d280b98 DC |
2500 | icsbp->icsb_icount = lcounter; |
2501 | break; | |
2502 | ||
2503 | case XFS_SBS_IFREE: | |
2504 | lcounter = icsbp->icsb_ifree; | |
2505 | lcounter += delta; | |
2506 | if (unlikely(lcounter < 0)) | |
20b64285 | 2507 | goto balance_counter; |
8d280b98 DC |
2508 | icsbp->icsb_ifree = lcounter; |
2509 | break; | |
2510 | ||
2511 | case XFS_SBS_FDBLOCKS: | |
2512 | BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0); | |
2513 | ||
4be536de | 2514 | lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2515 | lcounter += delta; |
2516 | if (unlikely(lcounter < 0)) | |
20b64285 | 2517 | goto balance_counter; |
4be536de | 2518 | icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp); |
8d280b98 DC |
2519 | break; |
2520 | default: | |
2521 | BUG(); | |
2522 | break; | |
2523 | } | |
01e1b69c | 2524 | xfs_icsb_unlock_cntr(icsbp); |
7a9e02d6 | 2525 | preempt_enable(); |
8d280b98 DC |
2526 | return 0; |
2527 | ||
8d280b98 | 2528 | slow_path: |
7a9e02d6 | 2529 | preempt_enable(); |
8d280b98 | 2530 | |
20b64285 DC |
2531 | /* |
2532 | * serialise with a mutex so we don't burn lots of cpu on | |
2533 | * the superblock lock. We still need to hold the superblock | |
2534 | * lock, however, when we modify the global structures. | |
2535 | */ | |
03135cf7 | 2536 | xfs_icsb_lock(mp); |
20b64285 DC |
2537 | |
2538 | /* | |
2539 | * Now running atomically. | |
2540 | * | |
2541 | * If the counter is enabled, someone has beaten us to rebalancing. | |
2542 | * Drop the lock and try again in the fast path.... | |
2543 | */ | |
2544 | if (!(xfs_icsb_counter_disabled(mp, field))) { | |
03135cf7 | 2545 | xfs_icsb_unlock(mp); |
8d280b98 | 2546 | goto again; |
8d280b98 DC |
2547 | } |
2548 | ||
20b64285 DC |
2549 | /* |
2550 | * The counter is currently disabled. Because we are | |
2551 | * running atomically here, we know a rebalance cannot | |
2552 | * be in progress. Hence we can go straight to operating | |
2553 | * on the global superblock. We do not call xfs_mod_incore_sb() | |
3685c2a1 | 2554 | * here even though we need to get the m_sb_lock. Doing so |
20b64285 | 2555 | * will cause us to re-enter this function and deadlock. |
3685c2a1 | 2556 | * Hence we get the m_sb_lock ourselves and then call |
20b64285 DC |
2557 | * xfs_mod_incore_sb_unlocked() as the unlocked path operates |
2558 | * directly on the global counters. | |
2559 | */ | |
3685c2a1 | 2560 | spin_lock(&mp->m_sb_lock); |
8d280b98 | 2561 | ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd); |
3685c2a1 | 2562 | spin_unlock(&mp->m_sb_lock); |
8d280b98 | 2563 | |
20b64285 DC |
2564 | /* |
2565 | * Now that we've modified the global superblock, we | |
2566 | * may be able to re-enable the distributed counters | |
2567 | * (e.g. lots of space just got freed). After that | |
2568 | * we are done. | |
2569 | */ | |
2570 | if (ret != ENOSPC) | |
45af6c6d | 2571 | xfs_icsb_balance_counter(mp, field, 0); |
03135cf7 | 2572 | xfs_icsb_unlock(mp); |
8d280b98 | 2573 | return ret; |
8d280b98 | 2574 | |
20b64285 DC |
2575 | balance_counter: |
2576 | xfs_icsb_unlock_cntr(icsbp); | |
7a9e02d6 | 2577 | preempt_enable(); |
8d280b98 | 2578 | |
20b64285 DC |
2579 | /* |
2580 | * We may have multiple threads here if multiple per-cpu | |
2581 | * counters run dry at the same time. This will mean we can | |
2582 | * do more balances than strictly necessary but it is not | |
2583 | * the common slowpath case. | |
2584 | */ | |
03135cf7 | 2585 | xfs_icsb_lock(mp); |
20b64285 DC |
2586 | |
2587 | /* | |
2588 | * running atomically. | |
2589 | * | |
2590 | * This will leave the counter in the correct state for future | |
2591 | * accesses. After the rebalance, we simply try again and our retry | |
2592 | * will either succeed through the fast path or slow path without | |
2593 | * another balance operation being required. | |
2594 | */ | |
45af6c6d | 2595 | xfs_icsb_balance_counter(mp, field, delta); |
03135cf7 | 2596 | xfs_icsb_unlock(mp); |
20b64285 | 2597 | goto again; |
8d280b98 | 2598 | } |
20b64285 | 2599 | |
8d280b98 | 2600 | #endif |