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" | |
a844f451 | 26 | #include "xfs_ag.h" |
1da177e4 LT |
27 | #include "xfs_dir2.h" |
28 | #include "xfs_dmapi.h" | |
29 | #include "xfs_mount.h" | |
a844f451 | 30 | #include "xfs_da_btree.h" |
1da177e4 LT |
31 | #include "xfs_bmap_btree.h" |
32 | #include "xfs_ialloc_btree.h" | |
33 | #include "xfs_alloc_btree.h" | |
1da177e4 | 34 | #include "xfs_dir2_sf.h" |
a844f451 | 35 | #include "xfs_attr_sf.h" |
1da177e4 | 36 | #include "xfs_dinode.h" |
1da177e4 | 37 | #include "xfs_inode.h" |
a844f451 NS |
38 | #include "xfs_inode_item.h" |
39 | #include "xfs_btree.h" | |
40 | #include "xfs_alloc.h" | |
41 | #include "xfs_ialloc.h" | |
42 | #include "xfs_quota.h" | |
1da177e4 LT |
43 | #include "xfs_error.h" |
44 | #include "xfs_bmap.h" | |
1da177e4 | 45 | #include "xfs_rw.h" |
1da177e4 | 46 | #include "xfs_buf_item.h" |
a844f451 | 47 | #include "xfs_log_priv.h" |
1da177e4 | 48 | #include "xfs_dir2_trace.h" |
a844f451 | 49 | #include "xfs_extfree_item.h" |
1da177e4 LT |
50 | #include "xfs_acl.h" |
51 | #include "xfs_attr.h" | |
52 | #include "xfs_clnt.h" | |
2a82b8be DC |
53 | #include "xfs_mru_cache.h" |
54 | #include "xfs_filestream.h" | |
e13a73f0 | 55 | #include "xfs_fsops.h" |
739bfb2a | 56 | #include "xfs_vnodeops.h" |
48c872a9 | 57 | #include "xfs_vfsops.h" |
43355099 | 58 | #include "xfs_utils.h" |
739bfb2a | 59 | |
1da177e4 | 60 | |
de2eeea6 | 61 | int __init |
1da177e4 LT |
62 | xfs_init(void) |
63 | { | |
1da177e4 | 64 | #ifdef XFS_DABUF_DEBUG |
703e1f0f | 65 | extern spinlock_t xfs_dabuf_global_lock; |
007c61c6 | 66 | spin_lock_init(&xfs_dabuf_global_lock); |
1da177e4 LT |
67 | #endif |
68 | ||
69 | /* | |
70 | * Initialize all of the zone allocators we use. | |
71 | */ | |
eb01c9cd DC |
72 | xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t), |
73 | "xfs_log_ticket"); | |
1da177e4 | 74 | xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t), |
eb01c9cd | 75 | "xfs_bmap_free_item"); |
1da177e4 | 76 | xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t), |
eb01c9cd DC |
77 | "xfs_btree_cur"); |
78 | xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t), | |
79 | "xfs_da_state"); | |
1da177e4 | 80 | xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf"); |
8758280f | 81 | xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork"); |
eb01c9cd | 82 | xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans"); |
8758280f | 83 | xfs_acl_zone_init(xfs_acl_zone, "xfs_acl"); |
2a82b8be DC |
84 | xfs_mru_cache_init(); |
85 | xfs_filestream_init(); | |
1da177e4 LT |
86 | |
87 | /* | |
88 | * The size of the zone allocated buf log item is the maximum | |
89 | * size possible under XFS. This wastes a little bit of memory, | |
90 | * but it is much faster. | |
91 | */ | |
92 | xfs_buf_item_zone = | |
93 | kmem_zone_init((sizeof(xfs_buf_log_item_t) + | |
94 | (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) / | |
95 | NBWORD) * sizeof(int))), | |
96 | "xfs_buf_item"); | |
8758280f NS |
97 | xfs_efd_zone = |
98 | kmem_zone_init((sizeof(xfs_efd_log_item_t) + | |
99 | ((XFS_EFD_MAX_FAST_EXTENTS - 1) * | |
100 | sizeof(xfs_extent_t))), | |
1da177e4 | 101 | "xfs_efd_item"); |
8758280f NS |
102 | xfs_efi_zone = |
103 | kmem_zone_init((sizeof(xfs_efi_log_item_t) + | |
104 | ((XFS_EFI_MAX_FAST_EXTENTS - 1) * | |
105 | sizeof(xfs_extent_t))), | |
1da177e4 | 106 | "xfs_efi_item"); |
8758280f NS |
107 | |
108 | /* | |
109 | * These zones warrant special memory allocator hints | |
110 | */ | |
111 | xfs_inode_zone = | |
112 | kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode", | |
113 | KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | | |
114 | KM_ZONE_SPREAD, NULL); | |
115 | xfs_ili_zone = | |
116 | kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili", | |
117 | KM_ZONE_SPREAD, NULL); | |
1da177e4 LT |
118 | |
119 | /* | |
120 | * Allocate global trace buffers. | |
121 | */ | |
122 | #ifdef XFS_ALLOC_TRACE | |
123 | xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP); | |
124 | #endif | |
125 | #ifdef XFS_BMAP_TRACE | |
126 | xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP); | |
127 | #endif | |
128 | #ifdef XFS_BMBT_TRACE | |
129 | xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP); | |
130 | #endif | |
1da177e4 LT |
131 | #ifdef XFS_ATTR_TRACE |
132 | xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP); | |
133 | #endif | |
134 | #ifdef XFS_DIR2_TRACE | |
135 | xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP); | |
136 | #endif | |
137 | ||
138 | xfs_dir_startup(); | |
139 | ||
140 | #if (defined(DEBUG) || defined(INDUCE_IO_ERROR)) | |
141 | xfs_error_test_init(); | |
142 | #endif /* DEBUG || INDUCE_IO_ERROR */ | |
143 | ||
144 | xfs_init_procfs(); | |
145 | xfs_sysctl_register(); | |
146 | return 0; | |
147 | } | |
148 | ||
de2eeea6 | 149 | void __exit |
1da177e4 LT |
150 | xfs_cleanup(void) |
151 | { | |
1da177e4 | 152 | extern kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
153 | extern kmem_zone_t *xfs_efd_zone; |
154 | extern kmem_zone_t *xfs_efi_zone; | |
1da177e4 LT |
155 | |
156 | xfs_cleanup_procfs(); | |
157 | xfs_sysctl_unregister(); | |
2a82b8be DC |
158 | xfs_filestream_uninit(); |
159 | xfs_mru_cache_uninit(); | |
1da177e4 LT |
160 | xfs_acl_zone_destroy(xfs_acl_zone); |
161 | ||
162 | #ifdef XFS_DIR2_TRACE | |
163 | ktrace_free(xfs_dir2_trace_buf); | |
164 | #endif | |
165 | #ifdef XFS_ATTR_TRACE | |
166 | ktrace_free(xfs_attr_trace_buf); | |
167 | #endif | |
1da177e4 LT |
168 | #ifdef XFS_BMBT_TRACE |
169 | ktrace_free(xfs_bmbt_trace_buf); | |
170 | #endif | |
171 | #ifdef XFS_BMAP_TRACE | |
172 | ktrace_free(xfs_bmap_trace_buf); | |
173 | #endif | |
174 | #ifdef XFS_ALLOC_TRACE | |
175 | ktrace_free(xfs_alloc_trace_buf); | |
176 | #endif | |
177 | ||
3758dee9 NS |
178 | kmem_zone_destroy(xfs_bmap_free_item_zone); |
179 | kmem_zone_destroy(xfs_btree_cur_zone); | |
180 | kmem_zone_destroy(xfs_inode_zone); | |
181 | kmem_zone_destroy(xfs_trans_zone); | |
182 | kmem_zone_destroy(xfs_da_state_zone); | |
183 | kmem_zone_destroy(xfs_dabuf_zone); | |
184 | kmem_zone_destroy(xfs_buf_item_zone); | |
185 | kmem_zone_destroy(xfs_efd_zone); | |
186 | kmem_zone_destroy(xfs_efi_zone); | |
187 | kmem_zone_destroy(xfs_ifork_zone); | |
188 | kmem_zone_destroy(xfs_ili_zone); | |
fe0754f0 | 189 | kmem_zone_destroy(xfs_log_ticket_zone); |
1da177e4 LT |
190 | } |
191 | ||
192 | /* | |
193 | * xfs_start_flags | |
194 | * | |
195 | * This function fills in xfs_mount_t fields based on mount args. | |
196 | * Note: the superblock has _not_ yet been read in. | |
197 | */ | |
198 | STATIC int | |
199 | xfs_start_flags( | |
1da177e4 LT |
200 | struct xfs_mount_args *ap, |
201 | struct xfs_mount *mp) | |
202 | { | |
203 | /* Values are in BBs */ | |
204 | if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) { | |
205 | /* | |
206 | * At this point the superblock has not been read | |
207 | * in, therefore we do not know the block size. | |
208 | * Before the mount call ends we will convert | |
209 | * these to FSBs. | |
210 | */ | |
211 | mp->m_dalign = ap->sunit; | |
212 | mp->m_swidth = ap->swidth; | |
213 | } | |
214 | ||
215 | if (ap->logbufs != -1 && | |
1da177e4 | 216 | ap->logbufs != 0 && |
1da177e4 LT |
217 | (ap->logbufs < XLOG_MIN_ICLOGS || |
218 | ap->logbufs > XLOG_MAX_ICLOGS)) { | |
219 | cmn_err(CE_WARN, | |
220 | "XFS: invalid logbufs value: %d [not %d-%d]", | |
221 | ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS); | |
222 | return XFS_ERROR(EINVAL); | |
223 | } | |
224 | mp->m_logbufs = ap->logbufs; | |
225 | if (ap->logbufsize != -1 && | |
cfcbbbd0 | 226 | ap->logbufsize != 0 && |
dcb3b83f ES |
227 | (ap->logbufsize < XLOG_MIN_RECORD_BSIZE || |
228 | ap->logbufsize > XLOG_MAX_RECORD_BSIZE || | |
229 | !is_power_of_2(ap->logbufsize))) { | |
1da177e4 LT |
230 | cmn_err(CE_WARN, |
231 | "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]", | |
232 | ap->logbufsize); | |
233 | return XFS_ERROR(EINVAL); | |
234 | } | |
1da177e4 LT |
235 | mp->m_logbsize = ap->logbufsize; |
236 | mp->m_fsname_len = strlen(ap->fsname) + 1; | |
237 | mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP); | |
238 | strcpy(mp->m_fsname, ap->fsname); | |
fc1f8c1c NS |
239 | if (ap->rtname[0]) { |
240 | mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP); | |
241 | strcpy(mp->m_rtname, ap->rtname); | |
242 | } | |
243 | if (ap->logname[0]) { | |
244 | mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP); | |
245 | strcpy(mp->m_logname, ap->logname); | |
246 | } | |
1da177e4 LT |
247 | |
248 | if (ap->flags & XFSMNT_WSYNC) | |
249 | mp->m_flags |= XFS_MOUNT_WSYNC; | |
250 | #if XFS_BIG_INUMS | |
251 | if (ap->flags & XFSMNT_INO64) { | |
252 | mp->m_flags |= XFS_MOUNT_INO64; | |
253 | mp->m_inoadd = XFS_INO64_OFFSET; | |
254 | } | |
255 | #endif | |
1da177e4 LT |
256 | if (ap->flags & XFSMNT_RETERR) |
257 | mp->m_flags |= XFS_MOUNT_RETERR; | |
1da177e4 LT |
258 | if (ap->flags & XFSMNT_NOALIGN) |
259 | mp->m_flags |= XFS_MOUNT_NOALIGN; | |
1da177e4 LT |
260 | if (ap->flags & XFSMNT_SWALLOC) |
261 | mp->m_flags |= XFS_MOUNT_SWALLOC; | |
1da177e4 LT |
262 | if (ap->flags & XFSMNT_OSYNCISOSYNC) |
263 | mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC; | |
1da177e4 | 264 | if (ap->flags & XFSMNT_32BITINODES) |
c11e2c36 | 265 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
1da177e4 LT |
266 | |
267 | if (ap->flags & XFSMNT_IOSIZE) { | |
268 | if (ap->iosizelog > XFS_MAX_IO_LOG || | |
269 | ap->iosizelog < XFS_MIN_IO_LOG) { | |
270 | cmn_err(CE_WARN, | |
271 | "XFS: invalid log iosize: %d [not %d-%d]", | |
272 | ap->iosizelog, XFS_MIN_IO_LOG, | |
273 | XFS_MAX_IO_LOG); | |
274 | return XFS_ERROR(EINVAL); | |
275 | } | |
276 | ||
277 | mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE; | |
278 | mp->m_readio_log = mp->m_writeio_log = ap->iosizelog; | |
279 | } | |
280 | ||
1bd960ee JJS |
281 | if (ap->flags & XFSMNT_IKEEP) |
282 | mp->m_flags |= XFS_MOUNT_IKEEP; | |
1da177e4 LT |
283 | if (ap->flags & XFSMNT_DIRSYNC) |
284 | mp->m_flags |= XFS_MOUNT_DIRSYNC; | |
13059ff0 NS |
285 | if (ap->flags & XFSMNT_ATTR2) |
286 | mp->m_flags |= XFS_MOUNT_ATTR2; | |
e8c8b3a7 | 287 | |
e718eeb4 NS |
288 | if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE) |
289 | mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; | |
290 | ||
1da177e4 LT |
291 | /* |
292 | * no recovery flag requires a read-only mount | |
293 | */ | |
294 | if (ap->flags & XFSMNT_NORECOVERY) { | |
bd186aa9 | 295 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1da177e4 LT |
296 | cmn_err(CE_WARN, |
297 | "XFS: tried to mount a FS read-write without recovery!"); | |
298 | return XFS_ERROR(EINVAL); | |
299 | } | |
300 | mp->m_flags |= XFS_MOUNT_NORECOVERY; | |
301 | } | |
302 | ||
303 | if (ap->flags & XFSMNT_NOUUID) | |
304 | mp->m_flags |= XFS_MOUNT_NOUUID; | |
f538d4da CH |
305 | if (ap->flags & XFSMNT_BARRIER) |
306 | mp->m_flags |= XFS_MOUNT_BARRIER; | |
4ef19ddd CH |
307 | else |
308 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
1da177e4 | 309 | |
2a82b8be DC |
310 | if (ap->flags2 & XFSMNT2_FILESTREAMS) |
311 | mp->m_flags |= XFS_MOUNT_FILESTREAMS; | |
312 | ||
293688ec | 313 | if (ap->flags & XFSMNT_DMAPI) |
bd186aa9 | 314 | mp->m_flags |= XFS_MOUNT_DMAPI; |
1da177e4 LT |
315 | return 0; |
316 | } | |
317 | ||
318 | /* | |
319 | * This function fills in xfs_mount_t fields based on mount args. | |
320 | * Note: the superblock _has_ now been read in. | |
321 | */ | |
322 | STATIC int | |
323 | xfs_finish_flags( | |
1da177e4 LT |
324 | struct xfs_mount_args *ap, |
325 | struct xfs_mount *mp) | |
326 | { | |
bd186aa9 | 327 | int ronly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
328 | |
329 | /* Fail a mount where the logbuf is smaller then the log stripe */ | |
62118709 | 330 | if (xfs_sb_version_haslogv2(&mp->m_sb)) { |
c2cd2550 | 331 | if ((ap->logbufsize <= 0) && |
1da177e4 LT |
332 | (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) { |
333 | mp->m_logbsize = mp->m_sb.sb_logsunit; | |
c2cd2550 NS |
334 | } else if (ap->logbufsize > 0 && |
335 | ap->logbufsize < mp->m_sb.sb_logsunit) { | |
1da177e4 LT |
336 | cmn_err(CE_WARN, |
337 | "XFS: logbuf size must be greater than or equal to log stripe size"); | |
338 | return XFS_ERROR(EINVAL); | |
339 | } | |
340 | } else { | |
341 | /* Fail a mount if the logbuf is larger than 32K */ | |
342 | if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) { | |
343 | cmn_err(CE_WARN, | |
344 | "XFS: logbuf size for version 1 logs must be 16K or 32K"); | |
345 | return XFS_ERROR(EINVAL); | |
346 | } | |
347 | } | |
348 | ||
62118709 | 349 | if (xfs_sb_version_hasattr2(&mp->m_sb)) |
13059ff0 | 350 | mp->m_flags |= XFS_MOUNT_ATTR2; |
13059ff0 | 351 | |
1da177e4 LT |
352 | /* |
353 | * prohibit r/w mounts of read-only filesystems | |
354 | */ | |
355 | if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) { | |
356 | cmn_err(CE_WARN, | |
357 | "XFS: cannot mount a read-only filesystem as read-write"); | |
358 | return XFS_ERROR(EROFS); | |
359 | } | |
360 | ||
1da177e4 LT |
361 | /* |
362 | * check for shared mount. | |
363 | */ | |
364 | if (ap->flags & XFSMNT_SHARED) { | |
62118709 | 365 | if (!xfs_sb_version_hasshared(&mp->m_sb)) |
1da177e4 LT |
366 | return XFS_ERROR(EINVAL); |
367 | ||
368 | /* | |
369 | * For IRIX 6.5, shared mounts must have the shared | |
370 | * version bit set, have the persistent readonly | |
371 | * field set, must be version 0 and can only be mounted | |
372 | * read-only. | |
373 | */ | |
374 | if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) || | |
375 | (mp->m_sb.sb_shared_vn != 0)) | |
376 | return XFS_ERROR(EINVAL); | |
377 | ||
378 | mp->m_flags |= XFS_MOUNT_SHARED; | |
379 | ||
380 | /* | |
381 | * Shared XFS V0 can't deal with DMI. Return EINVAL. | |
382 | */ | |
383 | if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI)) | |
384 | return XFS_ERROR(EINVAL); | |
385 | } | |
386 | ||
b09cc771 CH |
387 | if (ap->flags & XFSMNT_UQUOTA) { |
388 | mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE); | |
389 | if (ap->flags & XFSMNT_UQUOTAENF) | |
390 | mp->m_qflags |= XFS_UQUOTA_ENFD; | |
391 | } | |
392 | ||
393 | if (ap->flags & XFSMNT_GQUOTA) { | |
394 | mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE); | |
395 | if (ap->flags & XFSMNT_GQUOTAENF) | |
396 | mp->m_qflags |= XFS_OQUOTA_ENFD; | |
397 | } else if (ap->flags & XFSMNT_PQUOTA) { | |
398 | mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE); | |
399 | if (ap->flags & XFSMNT_PQUOTAENF) | |
400 | mp->m_qflags |= XFS_OQUOTA_ENFD; | |
401 | } | |
402 | ||
1da177e4 LT |
403 | return 0; |
404 | } | |
405 | ||
406 | /* | |
407 | * xfs_mount | |
408 | * | |
409 | * The file system configurations are: | |
410 | * (1) device (partition) with data and internal log | |
411 | * (2) logical volume with data and log subvolumes. | |
412 | * (3) logical volume with data, log, and realtime subvolumes. | |
413 | * | |
414 | * We only have to handle opening the log and realtime volumes here if | |
415 | * they are present. The data subvolume has already been opened by | |
416 | * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev. | |
417 | */ | |
48c872a9 | 418 | int |
1da177e4 | 419 | xfs_mount( |
48c872a9 | 420 | struct xfs_mount *mp, |
1da177e4 LT |
421 | struct xfs_mount_args *args, |
422 | cred_t *credp) | |
423 | { | |
1da177e4 LT |
424 | struct block_device *ddev, *logdev, *rtdev; |
425 | int flags = 0, error; | |
426 | ||
b267ce99 | 427 | ddev = mp->m_super->s_bdev; |
1da177e4 LT |
428 | logdev = rtdev = NULL; |
429 | ||
293688ec | 430 | error = xfs_dmops_get(mp, args); |
b09cc771 CH |
431 | if (error) |
432 | return error; | |
433 | error = xfs_qmops_get(mp, args); | |
293688ec CH |
434 | if (error) |
435 | return error; | |
436 | ||
764d1f89 NS |
437 | if (args->flags & XFSMNT_QUIET) |
438 | flags |= XFS_MFSI_QUIET; | |
439 | ||
1da177e4 LT |
440 | /* |
441 | * Open real time and log devices - order is important. | |
442 | */ | |
443 | if (args->logname[0]) { | |
444 | error = xfs_blkdev_get(mp, args->logname, &logdev); | |
445 | if (error) | |
446 | return error; | |
447 | } | |
448 | if (args->rtname[0]) { | |
449 | error = xfs_blkdev_get(mp, args->rtname, &rtdev); | |
450 | if (error) { | |
451 | xfs_blkdev_put(logdev); | |
452 | return error; | |
453 | } | |
454 | ||
455 | if (rtdev == ddev || rtdev == logdev) { | |
456 | cmn_err(CE_WARN, | |
457 | "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev."); | |
458 | xfs_blkdev_put(logdev); | |
459 | xfs_blkdev_put(rtdev); | |
460 | return EINVAL; | |
461 | } | |
462 | } | |
463 | ||
464 | /* | |
465 | * Setup xfs_mount buffer target pointers | |
466 | */ | |
467 | error = ENOMEM; | |
468 | mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0); | |
469 | if (!mp->m_ddev_targp) { | |
470 | xfs_blkdev_put(logdev); | |
471 | xfs_blkdev_put(rtdev); | |
472 | return error; | |
473 | } | |
474 | if (rtdev) { | |
475 | mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1); | |
49ee6c91 JJ |
476 | if (!mp->m_rtdev_targp) { |
477 | xfs_blkdev_put(logdev); | |
478 | xfs_blkdev_put(rtdev); | |
1da177e4 | 479 | goto error0; |
49ee6c91 | 480 | } |
1da177e4 LT |
481 | } |
482 | mp->m_logdev_targp = (logdev && logdev != ddev) ? | |
483 | xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp; | |
49ee6c91 JJ |
484 | if (!mp->m_logdev_targp) { |
485 | xfs_blkdev_put(logdev); | |
486 | xfs_blkdev_put(rtdev); | |
1da177e4 | 487 | goto error0; |
49ee6c91 | 488 | } |
1da177e4 LT |
489 | |
490 | /* | |
491 | * Setup flags based on mount(2) options and then the superblock | |
492 | */ | |
b267ce99 | 493 | error = xfs_start_flags(args, mp); |
1da177e4 LT |
494 | if (error) |
495 | goto error1; | |
764d1f89 | 496 | error = xfs_readsb(mp, flags); |
1da177e4 LT |
497 | if (error) |
498 | goto error1; | |
b267ce99 | 499 | error = xfs_finish_flags(args, mp); |
1da177e4 LT |
500 | if (error) |
501 | goto error2; | |
502 | ||
503 | /* | |
504 | * Setup xfs_mount buffer target pointers based on superblock | |
505 | */ | |
506 | error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize, | |
507 | mp->m_sb.sb_sectsize); | |
508 | if (!error && logdev && logdev != ddev) { | |
509 | unsigned int log_sector_size = BBSIZE; | |
510 | ||
62118709 | 511 | if (xfs_sb_version_hassector(&mp->m_sb)) |
1da177e4 LT |
512 | log_sector_size = mp->m_sb.sb_logsectsize; |
513 | error = xfs_setsize_buftarg(mp->m_logdev_targp, | |
514 | mp->m_sb.sb_blocksize, | |
515 | log_sector_size); | |
516 | } | |
517 | if (!error && rtdev) | |
518 | error = xfs_setsize_buftarg(mp->m_rtdev_targp, | |
519 | mp->m_sb.sb_blocksize, | |
520 | mp->m_sb.sb_sectsize); | |
521 | if (error) | |
522 | goto error2; | |
523 | ||
b2ea401b | 524 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
c7d437da CH |
525 | xfs_mountfs_check_barriers(mp); |
526 | ||
2a82b8be DC |
527 | if ((error = xfs_filestream_mount(mp))) |
528 | goto error2; | |
529 | ||
541d7d3c | 530 | error = xfs_mountfs(mp, flags); |
f538d4da CH |
531 | if (error) |
532 | goto error2; | |
533 | ||
293688ec CH |
534 | XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname); |
535 | ||
f538d4da CH |
536 | return 0; |
537 | ||
1da177e4 LT |
538 | error2: |
539 | if (mp->m_sb_bp) | |
540 | xfs_freesb(mp); | |
541 | error1: | |
542 | xfs_binval(mp->m_ddev_targp); | |
543 | if (logdev && logdev != ddev) | |
544 | xfs_binval(mp->m_logdev_targp); | |
545 | if (rtdev) | |
546 | xfs_binval(mp->m_rtdev_targp); | |
547 | error0: | |
548 | xfs_unmountfs_close(mp, credp); | |
b09cc771 | 549 | xfs_qmops_put(mp); |
293688ec | 550 | xfs_dmops_put(mp); |
1da177e4 LT |
551 | return error; |
552 | } | |
553 | ||
48c872a9 | 554 | int |
1da177e4 | 555 | xfs_unmount( |
48c872a9 | 556 | xfs_mount_t *mp, |
1da177e4 LT |
557 | int flags, |
558 | cred_t *credp) | |
559 | { | |
1da177e4 | 560 | xfs_inode_t *rip; |
67fcaa73 | 561 | bhv_vnode_t *rvp; |
1da177e4 LT |
562 | int unmount_event_wanted = 0; |
563 | int unmount_event_flags = 0; | |
564 | int xfs_unmountfs_needed = 0; | |
565 | int error; | |
566 | ||
567 | rip = mp->m_rootip; | |
568 | rvp = XFS_ITOV(rip); | |
569 | ||
948c6d4f | 570 | #ifdef HAVE_DMAPI |
bd186aa9 | 571 | if (mp->m_flags & XFS_MOUNT_DMAPI) { |
b267ce99 | 572 | error = XFS_SEND_PREUNMOUNT(mp, |
bc4ac74a | 573 | rip, DM_RIGHT_NULL, rip, DM_RIGHT_NULL, |
1da177e4 LT |
574 | NULL, NULL, 0, 0, |
575 | (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))? | |
576 | 0:DM_FLAGS_UNWANTED); | |
577 | if (error) | |
578 | return XFS_ERROR(error); | |
579 | unmount_event_wanted = 1; | |
580 | unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))? | |
581 | 0 : DM_FLAGS_UNWANTED; | |
582 | } | |
948c6d4f | 583 | #endif |
1da177e4 | 584 | |
2a82b8be DC |
585 | /* |
586 | * Blow away any referenced inode in the filestreams cache. | |
587 | * This can and will cause log traffic as inodes go inactive | |
588 | * here. | |
589 | */ | |
590 | xfs_filestream_unmount(mp); | |
591 | ||
1da177e4 LT |
592 | XFS_bflush(mp->m_ddev_targp); |
593 | error = xfs_unmount_flush(mp, 0); | |
594 | if (error) | |
595 | goto out; | |
596 | ||
597 | ASSERT(vn_count(rvp) == 1); | |
598 | ||
599 | /* | |
600 | * Drop the reference count | |
601 | */ | |
43355099 | 602 | IRELE(rip); |
1da177e4 LT |
603 | |
604 | /* | |
605 | * If we're forcing a shutdown, typically because of a media error, | |
606 | * we want to make sure we invalidate dirty pages that belong to | |
607 | * referenced vnodes as well. | |
608 | */ | |
609 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
48c872a9 | 610 | error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE); |
1da177e4 LT |
611 | ASSERT(error != EFSCORRUPTED); |
612 | } | |
613 | xfs_unmountfs_needed = 1; | |
614 | ||
615 | out: | |
616 | /* Send DMAPI event, if required. | |
617 | * Then do xfs_unmountfs() if needed. | |
618 | * Then return error (or zero). | |
619 | */ | |
620 | if (unmount_event_wanted) { | |
621 | /* Note: mp structure must still exist for | |
622 | * XFS_SEND_UNMOUNT() call. | |
623 | */ | |
bc4ac74a | 624 | XFS_SEND_UNMOUNT(mp, error == 0 ? rip : NULL, |
1da177e4 LT |
625 | DM_RIGHT_NULL, 0, error, unmount_event_flags); |
626 | } | |
627 | if (xfs_unmountfs_needed) { | |
628 | /* | |
629 | * Call common unmount function to flush to disk | |
630 | * and free the super block buffer & mount structures. | |
631 | */ | |
632 | xfs_unmountfs(mp, credp); | |
b09cc771 | 633 | xfs_qmops_put(mp); |
293688ec | 634 | xfs_dmops_put(mp); |
f541d270 | 635 | kmem_free(mp, sizeof(xfs_mount_t)); |
1da177e4 LT |
636 | } |
637 | ||
638 | return XFS_ERROR(error); | |
639 | } | |
640 | ||
3c85c36c | 641 | STATIC void |
f898d6c0 CH |
642 | xfs_quiesce_fs( |
643 | xfs_mount_t *mp) | |
644 | { | |
645 | int count = 0, pincount; | |
3758dee9 | 646 | |
f898d6c0 CH |
647 | xfs_flush_buftarg(mp->m_ddev_targp, 0); |
648 | xfs_finish_reclaim_all(mp, 0); | |
649 | ||
650 | /* This loop must run at least twice. | |
651 | * The first instance of the loop will flush | |
652 | * most meta data but that will generate more | |
653 | * meta data (typically directory updates). | |
654 | * Which then must be flushed and logged before | |
655 | * we can write the unmount record. | |
3758dee9 | 656 | */ |
f898d6c0 | 657 | do { |
516b2e7c | 658 | xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL); |
f898d6c0 CH |
659 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
660 | if (!pincount) { | |
661 | delay(50); | |
662 | count++; | |
663 | } | |
664 | } while (count < 2); | |
f898d6c0 | 665 | } |
1da177e4 | 666 | |
516b2e7c DC |
667 | /* |
668 | * Second stage of a quiesce. The data is already synced, now we have to take | |
669 | * care of the metadata. New transactions are already blocked, so we need to | |
670 | * wait for any remaining transactions to drain out before proceding. | |
671 | */ | |
9909c4aa | 672 | void |
516b2e7c DC |
673 | xfs_attr_quiesce( |
674 | xfs_mount_t *mp) | |
675 | { | |
e5720eec DC |
676 | int error = 0; |
677 | ||
516b2e7c DC |
678 | /* wait for all modifications to complete */ |
679 | while (atomic_read(&mp->m_active_trans) > 0) | |
680 | delay(100); | |
681 | ||
682 | /* flush inodes and push all remaining buffers out to disk */ | |
683 | xfs_quiesce_fs(mp); | |
684 | ||
685 | ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0); | |
686 | ||
687 | /* Push the superblock and write an unmount record */ | |
e5720eec DC |
688 | error = xfs_log_sbcount(mp, 1); |
689 | if (error) | |
690 | xfs_fs_cmn_err(CE_WARN, mp, | |
691 | "xfs_attr_quiesce: failed to log sb changes. " | |
692 | "Frozen image may not be consistent."); | |
516b2e7c DC |
693 | xfs_log_unmount_write(mp); |
694 | xfs_unmountfs_writesb(mp); | |
695 | } | |
696 | ||
48c872a9 | 697 | int |
1da177e4 | 698 | xfs_mntupdate( |
48c872a9 | 699 | struct xfs_mount *mp, |
1da177e4 LT |
700 | int *flags, |
701 | struct xfs_mount_args *args) | |
702 | { | |
d08d389d | 703 | if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */ |
bd186aa9 CH |
704 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
705 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
d08d389d NS |
706 | if (args->flags & XFSMNT_BARRIER) { |
707 | mp->m_flags |= XFS_MOUNT_BARRIER; | |
f538d4da | 708 | xfs_mountfs_check_barriers(mp); |
d08d389d NS |
709 | } else { |
710 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
711 | } | |
bd186aa9 | 712 | } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */ |
2a82b8be | 713 | xfs_filestream_flush(mp); |
745f6919 | 714 | xfs_sync(mp, SYNC_DATA_QUIESCE); |
516b2e7c | 715 | xfs_attr_quiesce(mp); |
bd186aa9 | 716 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 717 | } |
1da177e4 LT |
718 | return 0; |
719 | } | |
720 | ||
721 | /* | |
722 | * xfs_unmount_flush implements a set of flush operation on special | |
723 | * inodes, which are needed as a separate set of operations so that | |
724 | * they can be called as part of relocation process. | |
725 | */ | |
726 | int | |
727 | xfs_unmount_flush( | |
728 | xfs_mount_t *mp, /* Mount structure we are getting | |
729 | rid of. */ | |
730 | int relocation) /* Called from vfs relocation. */ | |
731 | { | |
732 | xfs_inode_t *rip = mp->m_rootip; | |
733 | xfs_inode_t *rbmip; | |
734 | xfs_inode_t *rsumip = NULL; | |
67fcaa73 | 735 | bhv_vnode_t *rvp = XFS_ITOV(rip); |
1da177e4 LT |
736 | int error; |
737 | ||
f7c66ce3 | 738 | xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
739 | xfs_iflock(rip); |
740 | ||
741 | /* | |
742 | * Flush out the real time inodes. | |
743 | */ | |
744 | if ((rbmip = mp->m_rbmip) != NULL) { | |
745 | xfs_ilock(rbmip, XFS_ILOCK_EXCL); | |
746 | xfs_iflock(rbmip); | |
747 | error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC); | |
748 | xfs_iunlock(rbmip, XFS_ILOCK_EXCL); | |
749 | ||
750 | if (error == EFSCORRUPTED) | |
751 | goto fscorrupt_out; | |
752 | ||
753 | ASSERT(vn_count(XFS_ITOV(rbmip)) == 1); | |
754 | ||
755 | rsumip = mp->m_rsumip; | |
756 | xfs_ilock(rsumip, XFS_ILOCK_EXCL); | |
757 | xfs_iflock(rsumip); | |
758 | error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC); | |
759 | xfs_iunlock(rsumip, XFS_ILOCK_EXCL); | |
760 | ||
761 | if (error == EFSCORRUPTED) | |
762 | goto fscorrupt_out; | |
763 | ||
764 | ASSERT(vn_count(XFS_ITOV(rsumip)) == 1); | |
765 | } | |
766 | ||
767 | /* | |
768 | * Synchronously flush root inode to disk | |
769 | */ | |
770 | error = xfs_iflush(rip, XFS_IFLUSH_SYNC); | |
771 | if (error == EFSCORRUPTED) | |
772 | goto fscorrupt_out2; | |
773 | ||
774 | if (vn_count(rvp) != 1 && !relocation) { | |
775 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
776 | return XFS_ERROR(EBUSY); | |
777 | } | |
778 | ||
779 | /* | |
780 | * Release dquot that rootinode, rbmino and rsumino might be holding, | |
781 | * flush and purge the quota inodes. | |
782 | */ | |
783 | error = XFS_QM_UNMOUNT(mp); | |
784 | if (error == EFSCORRUPTED) | |
785 | goto fscorrupt_out2; | |
786 | ||
787 | if (rbmip) { | |
43355099 CH |
788 | IRELE(rbmip); |
789 | IRELE(rsumip); | |
1da177e4 LT |
790 | } |
791 | ||
792 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
793 | return 0; | |
794 | ||
795 | fscorrupt_out: | |
796 | xfs_ifunlock(rip); | |
797 | ||
798 | fscorrupt_out2: | |
799 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
800 | ||
801 | return XFS_ERROR(EFSCORRUPTED); | |
802 | } | |
803 | ||
1da177e4 LT |
804 | /* |
805 | * xfs_sync flushes any pending I/O to file system vfsp. | |
806 | * | |
807 | * This routine is called by vfs_sync() to make sure that things make it | |
808 | * out to disk eventually, on sync() system calls to flush out everything, | |
809 | * and when the file system is unmounted. For the vfs_sync() case, all | |
810 | * we really need to do is sync out the log to make all of our meta-data | |
811 | * updates permanent (except for timestamps). For calls from pflushd(), | |
812 | * dirty pages are kept moving by calling pdflush() on the inodes | |
813 | * containing them. We also flush the inodes that we can lock without | |
814 | * sleeping and the superblock if we can lock it without sleeping from | |
815 | * vfs_sync() so that items at the tail of the log are always moving out. | |
816 | * | |
817 | * Flags: | |
818 | * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want | |
819 | * to sleep if we can help it. All we really need | |
820 | * to do is ensure that the log is synced at least | |
821 | * periodically. We also push the inodes and | |
822 | * superblock if we can lock them without sleeping | |
823 | * and they are not pinned. | |
824 | * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not | |
825 | * set, then we really want to lock each inode and flush | |
826 | * it. | |
827 | * SYNC_WAIT - All the flushes that take place in this call should | |
828 | * be synchronous. | |
829 | * SYNC_DELWRI - This tells us to push dirty pages associated with | |
830 | * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to | |
831 | * determine if they should be flushed sync, async, or | |
832 | * delwri. | |
833 | * SYNC_CLOSE - This flag is passed when the system is being | |
c41564b5 | 834 | * unmounted. We should sync and invalidate everything. |
1da177e4 LT |
835 | * SYNC_FSDATA - This indicates that the caller would like to make |
836 | * sure the superblock is safe on disk. We can ensure | |
c41564b5 | 837 | * this by simply making sure the log gets flushed |
1da177e4 LT |
838 | * if SYNC_BDFLUSH is set, and by actually writing it |
839 | * out otherwise. | |
3c0dc77b DC |
840 | * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete |
841 | * before we return (including direct I/O). Forms the drain | |
842 | * side of the write barrier needed to safely quiesce the | |
843 | * filesystem. | |
1da177e4 LT |
844 | * |
845 | */ | |
48c872a9 | 846 | int |
1da177e4 | 847 | xfs_sync( |
48c872a9 CH |
848 | xfs_mount_t *mp, |
849 | int flags) | |
1da177e4 | 850 | { |
b09cc771 CH |
851 | int error; |
852 | ||
853 | /* | |
854 | * Get the Quota Manager to flush the dquots. | |
855 | * | |
856 | * If XFS quota support is not enabled or this filesystem | |
857 | * instance does not use quotas XFS_QM_DQSYNC will always | |
858 | * return zero. | |
859 | */ | |
860 | error = XFS_QM_DQSYNC(mp, flags); | |
861 | if (error) { | |
862 | /* | |
863 | * If we got an IO error, we will be shutting down. | |
864 | * So, there's nothing more for us to do here. | |
865 | */ | |
866 | ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp)); | |
867 | if (XFS_FORCED_SHUTDOWN(mp)) | |
868 | return XFS_ERROR(error); | |
869 | } | |
1da177e4 | 870 | |
2a82b8be DC |
871 | if (flags & SYNC_IOWAIT) |
872 | xfs_filestream_flush(mp); | |
873 | ||
2823945f | 874 | return xfs_syncsub(mp, flags, NULL); |
1da177e4 LT |
875 | } |
876 | ||
877 | /* | |
878 | * xfs sync routine for internal use | |
879 | * | |
b83bd138 | 880 | * This routine supports all of the flags defined for the generic vfs_sync |
1f9b3b64 | 881 | * interface as explained above under xfs_sync. |
1da177e4 LT |
882 | * |
883 | */ | |
ee34807a | 884 | int |
1da177e4 LT |
885 | xfs_sync_inodes( |
886 | xfs_mount_t *mp, | |
887 | int flags, | |
1da177e4 LT |
888 | int *bypassed) |
889 | { | |
890 | xfs_inode_t *ip = NULL; | |
67fcaa73 | 891 | bhv_vnode_t *vp = NULL; |
1da177e4 LT |
892 | int error; |
893 | int last_error; | |
894 | uint64_t fflag; | |
895 | uint lock_flags; | |
896 | uint base_lock_flags; | |
897 | boolean_t mount_locked; | |
898 | boolean_t vnode_refed; | |
899 | int preempt; | |
1da177e4 LT |
900 | xfs_iptr_t *ipointer; |
901 | #ifdef DEBUG | |
902 | boolean_t ipointer_in = B_FALSE; | |
903 | ||
904 | #define IPOINTER_SET ipointer_in = B_TRUE | |
905 | #define IPOINTER_CLR ipointer_in = B_FALSE | |
906 | #else | |
907 | #define IPOINTER_SET | |
908 | #define IPOINTER_CLR | |
909 | #endif | |
910 | ||
911 | ||
912 | /* Insert a marker record into the inode list after inode ip. The list | |
913 | * must be locked when this is called. After the call the list will no | |
914 | * longer be locked. | |
915 | */ | |
916 | #define IPOINTER_INSERT(ip, mp) { \ | |
917 | ASSERT(ipointer_in == B_FALSE); \ | |
918 | ipointer->ip_mnext = ip->i_mnext; \ | |
919 | ipointer->ip_mprev = ip; \ | |
920 | ip->i_mnext = (xfs_inode_t *)ipointer; \ | |
921 | ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \ | |
922 | preempt = 0; \ | |
923 | XFS_MOUNT_IUNLOCK(mp); \ | |
924 | mount_locked = B_FALSE; \ | |
925 | IPOINTER_SET; \ | |
926 | } | |
927 | ||
928 | /* Remove the marker from the inode list. If the marker was the only item | |
929 | * in the list then there are no remaining inodes and we should zero out | |
930 | * the whole list. If we are the current head of the list then move the head | |
931 | * past us. | |
932 | */ | |
933 | #define IPOINTER_REMOVE(ip, mp) { \ | |
934 | ASSERT(ipointer_in == B_TRUE); \ | |
935 | if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \ | |
936 | ip = ipointer->ip_mnext; \ | |
937 | ip->i_mprev = ipointer->ip_mprev; \ | |
938 | ipointer->ip_mprev->i_mnext = ip; \ | |
939 | if (mp->m_inodes == (xfs_inode_t *)ipointer) { \ | |
940 | mp->m_inodes = ip; \ | |
941 | } \ | |
942 | } else { \ | |
943 | ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \ | |
944 | mp->m_inodes = NULL; \ | |
945 | ip = NULL; \ | |
946 | } \ | |
947 | IPOINTER_CLR; \ | |
948 | } | |
949 | ||
950 | #define XFS_PREEMPT_MASK 0x7f | |
951 | ||
44866d39 LM |
952 | ASSERT(!(flags & SYNC_BDFLUSH)); |
953 | ||
1da177e4 LT |
954 | if (bypassed) |
955 | *bypassed = 0; | |
bd186aa9 | 956 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
957 | return 0; |
958 | error = 0; | |
959 | last_error = 0; | |
960 | preempt = 0; | |
961 | ||
962 | /* Allocate a reference marker */ | |
963 | ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP); | |
964 | ||
965 | fflag = XFS_B_ASYNC; /* default is don't wait */ | |
44866d39 | 966 | if (flags & SYNC_DELWRI) |
1da177e4 LT |
967 | fflag = XFS_B_DELWRI; |
968 | if (flags & SYNC_WAIT) | |
969 | fflag = 0; /* synchronous overrides all */ | |
970 | ||
971 | base_lock_flags = XFS_ILOCK_SHARED; | |
972 | if (flags & (SYNC_DELWRI | SYNC_CLOSE)) { | |
973 | /* | |
974 | * We need the I/O lock if we're going to call any of | |
975 | * the flush/inval routines. | |
976 | */ | |
977 | base_lock_flags |= XFS_IOLOCK_SHARED; | |
978 | } | |
979 | ||
980 | XFS_MOUNT_ILOCK(mp); | |
981 | ||
982 | ip = mp->m_inodes; | |
983 | ||
984 | mount_locked = B_TRUE; | |
985 | vnode_refed = B_FALSE; | |
986 | ||
987 | IPOINTER_CLR; | |
988 | ||
989 | do { | |
990 | ASSERT(ipointer_in == B_FALSE); | |
991 | ASSERT(vnode_refed == B_FALSE); | |
992 | ||
993 | lock_flags = base_lock_flags; | |
994 | ||
995 | /* | |
996 | * There were no inodes in the list, just break out | |
997 | * of the loop. | |
998 | */ | |
999 | if (ip == NULL) { | |
1000 | break; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * We found another sync thread marker - skip it | |
1005 | */ | |
1006 | if (ip->i_mount == NULL) { | |
1007 | ip = ip->i_mnext; | |
1008 | continue; | |
1009 | } | |
1010 | ||
1011 | vp = XFS_ITOV_NULL(ip); | |
1012 | ||
1013 | /* | |
1014 | * If the vnode is gone then this is being torn down, | |
1015 | * call reclaim if it is flushed, else let regular flush | |
1016 | * code deal with it later in the loop. | |
1017 | */ | |
1018 | ||
1019 | if (vp == NULL) { | |
1020 | /* Skip ones already in reclaim */ | |
1021 | if (ip->i_flags & XFS_IRECLAIM) { | |
1022 | ip = ip->i_mnext; | |
1023 | continue; | |
1024 | } | |
1025 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) { | |
1026 | ip = ip->i_mnext; | |
1027 | } else if ((xfs_ipincount(ip) == 0) && | |
1028 | xfs_iflock_nowait(ip)) { | |
1029 | IPOINTER_INSERT(ip, mp); | |
1030 | ||
1031 | xfs_finish_reclaim(ip, 1, | |
1032 | XFS_IFLUSH_DELWRI_ELSE_ASYNC); | |
1033 | ||
1034 | XFS_MOUNT_ILOCK(mp); | |
1035 | mount_locked = B_TRUE; | |
1036 | IPOINTER_REMOVE(ip, mp); | |
1037 | } else { | |
1038 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1039 | ip = ip->i_mnext; | |
1040 | } | |
1041 | continue; | |
1042 | } | |
1043 | ||
1044 | if (VN_BAD(vp)) { | |
1045 | ip = ip->i_mnext; | |
1046 | continue; | |
1047 | } | |
1048 | ||
1049 | if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) { | |
1050 | XFS_MOUNT_IUNLOCK(mp); | |
1051 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1052 | return 0; | |
1053 | } | |
1054 | ||
1da177e4 LT |
1055 | /* |
1056 | * Try to lock without sleeping. We're out of order with | |
1057 | * the inode list lock here, so if we fail we need to drop | |
1058 | * the mount lock and try again. If we're called from | |
1059 | * bdflush() here, then don't bother. | |
1060 | * | |
1061 | * The inode lock here actually coordinates with the | |
1062 | * almost spurious inode lock in xfs_ireclaim() to prevent | |
1063 | * the vnode we handle here without a reference from | |
1064 | * being freed while we reference it. If we lock the inode | |
1065 | * while it's on the mount list here, then the spurious inode | |
1066 | * lock in xfs_ireclaim() after the inode is pulled from | |
1067 | * the mount list will sleep until we release it here. | |
1068 | * This keeps the vnode from being freed while we reference | |
cdb62687 | 1069 | * it. |
1da177e4 LT |
1070 | */ |
1071 | if (xfs_ilock_nowait(ip, lock_flags) == 0) { | |
44866d39 | 1072 | if (vp == NULL) { |
1da177e4 LT |
1073 | ip = ip->i_mnext; |
1074 | continue; | |
1075 | } | |
1076 | ||
cdb62687 | 1077 | vp = vn_grab(vp); |
1da177e4 | 1078 | if (vp == NULL) { |
cdb62687 | 1079 | ip = ip->i_mnext; |
1da177e4 LT |
1080 | continue; |
1081 | } | |
1082 | ||
cdb62687 | 1083 | IPOINTER_INSERT(ip, mp); |
1da177e4 LT |
1084 | xfs_ilock(ip, lock_flags); |
1085 | ||
1086 | ASSERT(vp == XFS_ITOV(ip)); | |
1087 | ASSERT(ip->i_mount == mp); | |
1088 | ||
1089 | vnode_refed = B_TRUE; | |
1090 | } | |
1091 | ||
1092 | /* From here on in the loop we may have a marker record | |
1093 | * in the inode list. | |
1094 | */ | |
1095 | ||
40095b64 DC |
1096 | /* |
1097 | * If we have to flush data or wait for I/O completion | |
1098 | * we need to drop the ilock that we currently hold. | |
1099 | * If we need to drop the lock, insert a marker if we | |
1100 | * have not already done so. | |
1101 | */ | |
1102 | if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) || | |
1103 | ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) { | |
1104 | if (mount_locked) { | |
1105 | IPOINTER_INSERT(ip, mp); | |
1da177e4 | 1106 | } |
40095b64 | 1107 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 | 1108 | |
40095b64 DC |
1109 | if (flags & SYNC_CLOSE) { |
1110 | /* Shutdown case. Flush and invalidate. */ | |
1111 | if (XFS_FORCED_SHUTDOWN(mp)) | |
739bfb2a CH |
1112 | xfs_tosspages(ip, 0, -1, |
1113 | FI_REMAPF); | |
40095b64 | 1114 | else |
739bfb2a CH |
1115 | error = xfs_flushinval_pages(ip, |
1116 | 0, -1, FI_REMAPF); | |
40095b64 | 1117 | } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) { |
739bfb2a | 1118 | error = xfs_flush_pages(ip, 0, |
67fcaa73 | 1119 | -1, fflag, FI_NONE); |
1da177e4 LT |
1120 | } |
1121 | ||
40095b64 DC |
1122 | /* |
1123 | * When freezing, we need to wait ensure all I/O (including direct | |
1124 | * I/O) is complete to ensure no further data modification can take | |
1125 | * place after this point | |
1126 | */ | |
1127 | if (flags & SYNC_IOWAIT) | |
b677c210 | 1128 | vn_iowait(ip); |
40095b64 DC |
1129 | |
1130 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1da177e4 LT |
1131 | } |
1132 | ||
44866d39 LM |
1133 | if ((flags & SYNC_ATTR) && |
1134 | (ip->i_update_core || | |
1135 | (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) { | |
1136 | if (mount_locked) | |
1137 | IPOINTER_INSERT(ip, mp); | |
1da177e4 | 1138 | |
44866d39 LM |
1139 | if (flags & SYNC_WAIT) { |
1140 | xfs_iflock(ip); | |
1141 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
1da177e4 | 1142 | |
44866d39 LM |
1143 | /* |
1144 | * If we can't acquire the flush lock, then the inode | |
1145 | * is already being flushed so don't bother waiting. | |
1146 | * | |
1147 | * If we can lock it then do a delwri flush so we can | |
1148 | * combine multiple inode flushes in each disk write. | |
1149 | */ | |
1150 | } else if (xfs_iflock_nowait(ip)) { | |
1151 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
1152 | } else if (bypassed) { | |
1153 | (*bypassed)++; | |
1da177e4 LT |
1154 | } |
1155 | } | |
1156 | ||
1157 | if (lock_flags != 0) { | |
1158 | xfs_iunlock(ip, lock_flags); | |
1159 | } | |
1160 | ||
1161 | if (vnode_refed) { | |
1162 | /* | |
1163 | * If we had to take a reference on the vnode | |
1164 | * above, then wait until after we've unlocked | |
1165 | * the inode to release the reference. This is | |
1166 | * because we can be already holding the inode | |
43355099 | 1167 | * lock when IRELE() calls xfs_inactive(). |
1da177e4 LT |
1168 | * |
1169 | * Make sure to drop the mount lock before calling | |
43355099 | 1170 | * IRELE() so that we don't trip over ourselves if |
1da177e4 LT |
1171 | * we have to go for the mount lock again in the |
1172 | * inactive code. | |
1173 | */ | |
1174 | if (mount_locked) { | |
1175 | IPOINTER_INSERT(ip, mp); | |
1176 | } | |
1177 | ||
43355099 | 1178 | IRELE(ip); |
1da177e4 LT |
1179 | |
1180 | vnode_refed = B_FALSE; | |
1181 | } | |
1182 | ||
1183 | if (error) { | |
1184 | last_error = error; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * bail out if the filesystem is corrupted. | |
1189 | */ | |
1190 | if (error == EFSCORRUPTED) { | |
1191 | if (!mount_locked) { | |
1192 | XFS_MOUNT_ILOCK(mp); | |
1193 | IPOINTER_REMOVE(ip, mp); | |
1194 | } | |
1195 | XFS_MOUNT_IUNLOCK(mp); | |
1196 | ASSERT(ipointer_in == B_FALSE); | |
1197 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1198 | return XFS_ERROR(error); | |
1199 | } | |
1200 | ||
1201 | /* Let other threads have a chance at the mount lock | |
1202 | * if we have looped many times without dropping the | |
1203 | * lock. | |
1204 | */ | |
1205 | if ((++preempt & XFS_PREEMPT_MASK) == 0) { | |
1206 | if (mount_locked) { | |
1207 | IPOINTER_INSERT(ip, mp); | |
1208 | } | |
1209 | } | |
1210 | ||
1211 | if (mount_locked == B_FALSE) { | |
1212 | XFS_MOUNT_ILOCK(mp); | |
1213 | mount_locked = B_TRUE; | |
1214 | IPOINTER_REMOVE(ip, mp); | |
1215 | continue; | |
1216 | } | |
1217 | ||
1218 | ASSERT(ipointer_in == B_FALSE); | |
1219 | ip = ip->i_mnext; | |
1220 | ||
1221 | } while (ip != mp->m_inodes); | |
1222 | ||
1223 | XFS_MOUNT_IUNLOCK(mp); | |
1224 | ||
1225 | ASSERT(ipointer_in == B_FALSE); | |
1226 | ||
1227 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1228 | return XFS_ERROR(last_error); | |
1229 | } | |
1230 | ||
1231 | /* | |
1232 | * xfs sync routine for internal use | |
1233 | * | |
b83bd138 | 1234 | * This routine supports all of the flags defined for the generic vfs_sync |
1f9b3b64 | 1235 | * interface as explained above under xfs_sync. |
1da177e4 LT |
1236 | * |
1237 | */ | |
1238 | int | |
1239 | xfs_syncsub( | |
1240 | xfs_mount_t *mp, | |
1241 | int flags, | |
1da177e4 LT |
1242 | int *bypassed) |
1243 | { | |
1244 | int error = 0; | |
1245 | int last_error = 0; | |
1246 | uint log_flags = XFS_LOG_FORCE; | |
1247 | xfs_buf_t *bp; | |
1248 | xfs_buf_log_item_t *bip; | |
1249 | ||
1250 | /* | |
1251 | * Sync out the log. This ensures that the log is periodically | |
1252 | * flushed even if there is not enough activity to fill it up. | |
1253 | */ | |
1254 | if (flags & SYNC_WAIT) | |
1255 | log_flags |= XFS_LOG_SYNC; | |
1256 | ||
1257 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1258 | ||
1259 | if (flags & (SYNC_ATTR|SYNC_DELWRI)) { | |
1260 | if (flags & SYNC_BDFLUSH) | |
1261 | xfs_finish_reclaim_all(mp, 1); | |
1262 | else | |
1f9b3b64 | 1263 | error = xfs_sync_inodes(mp, flags, bypassed); |
1da177e4 LT |
1264 | } |
1265 | ||
1266 | /* | |
1267 | * Flushing out dirty data above probably generated more | |
1268 | * log activity, so if this isn't vfs_sync() then flush | |
1269 | * the log again. | |
1270 | */ | |
1271 | if (flags & SYNC_DELWRI) { | |
1272 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1273 | } | |
1274 | ||
1275 | if (flags & SYNC_FSDATA) { | |
1276 | /* | |
1277 | * If this is vfs_sync() then only sync the superblock | |
1278 | * if we can lock it without sleeping and it is not pinned. | |
1279 | */ | |
1280 | if (flags & SYNC_BDFLUSH) { | |
1281 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
1282 | if (bp != NULL) { | |
1283 | bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*); | |
1284 | if ((bip != NULL) && | |
1285 | xfs_buf_item_dirty(bip)) { | |
1286 | if (!(XFS_BUF_ISPINNED(bp))) { | |
1287 | XFS_BUF_ASYNC(bp); | |
1288 | error = xfs_bwrite(mp, bp); | |
1289 | } else { | |
1290 | xfs_buf_relse(bp); | |
1291 | } | |
1292 | } else { | |
1293 | xfs_buf_relse(bp); | |
1294 | } | |
1295 | } | |
1296 | } else { | |
1297 | bp = xfs_getsb(mp, 0); | |
1298 | /* | |
1299 | * If the buffer is pinned then push on the log so | |
1300 | * we won't get stuck waiting in the write for | |
1301 | * someone, maybe ourselves, to flush the log. | |
1302 | * Even though we just pushed the log above, we | |
1303 | * did not have the superblock buffer locked at | |
1304 | * that point so it can become pinned in between | |
1305 | * there and here. | |
1306 | */ | |
1307 | if (XFS_BUF_ISPINNED(bp)) | |
1308 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1309 | if (flags & SYNC_WAIT) | |
1310 | XFS_BUF_UNASYNC(bp); | |
1311 | else | |
1312 | XFS_BUF_ASYNC(bp); | |
1313 | error = xfs_bwrite(mp, bp); | |
1314 | } | |
1315 | if (error) { | |
1316 | last_error = error; | |
1317 | } | |
1318 | } | |
1319 | ||
1da177e4 LT |
1320 | /* |
1321 | * Now check to see if the log needs a "dummy" transaction. | |
1322 | */ | |
1da177e4 LT |
1323 | if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) { |
1324 | xfs_trans_t *tp; | |
1325 | xfs_inode_t *ip; | |
1326 | ||
1327 | /* | |
1328 | * Put a dummy transaction in the log to tell | |
1329 | * recovery that all others are OK. | |
1330 | */ | |
1331 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
1332 | if ((error = xfs_trans_reserve(tp, 0, | |
1333 | XFS_ICHANGE_LOG_RES(mp), | |
1334 | 0, 0, 0))) { | |
1335 | xfs_trans_cancel(tp, 0); | |
1336 | return error; | |
1337 | } | |
1338 | ||
1339 | ip = mp->m_rootip; | |
1340 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1341 | ||
1342 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1343 | xfs_trans_ihold(tp, ip); | |
1344 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1c72bf90 | 1345 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1346 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1347 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * When shutting down, we need to insure that the AIL is pushed | |
1352 | * to disk or the filesystem can appear corrupt from the PROM. | |
1353 | */ | |
1354 | if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) { | |
1355 | XFS_bflush(mp->m_ddev_targp); | |
1356 | if (mp->m_rtdev_targp) { | |
1357 | XFS_bflush(mp->m_rtdev_targp); | |
1358 | } | |
1359 | } | |
1360 | ||
1361 | return XFS_ERROR(last_error); | |
1362 | } |