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 LT |
45 | #include "xfs_rw.h" |
46 | #include "xfs_refcache.h" | |
47 | #include "xfs_buf_item.h" | |
a844f451 | 48 | #include "xfs_log_priv.h" |
1da177e4 | 49 | #include "xfs_dir2_trace.h" |
a844f451 | 50 | #include "xfs_extfree_item.h" |
1da177e4 LT |
51 | #include "xfs_acl.h" |
52 | #include "xfs_attr.h" | |
53 | #include "xfs_clnt.h" | |
2a82b8be DC |
54 | #include "xfs_mru_cache.h" |
55 | #include "xfs_filestream.h" | |
e13a73f0 | 56 | #include "xfs_fsops.h" |
739bfb2a | 57 | #include "xfs_vnodeops.h" |
48c872a9 | 58 | #include "xfs_vfsops.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 | */ | |
72 | xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t), | |
73 | "xfs_bmap_free_item"); | |
74 | xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t), | |
75 | "xfs_btree_cur"); | |
1da177e4 LT |
76 | xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans"); |
77 | xfs_da_state_zone = | |
78 | kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state"); | |
79 | xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf"); | |
8758280f NS |
80 | xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork"); |
81 | xfs_acl_zone_init(xfs_acl_zone, "xfs_acl"); | |
2a82b8be DC |
82 | xfs_mru_cache_init(); |
83 | xfs_filestream_init(); | |
1da177e4 LT |
84 | |
85 | /* | |
86 | * The size of the zone allocated buf log item is the maximum | |
87 | * size possible under XFS. This wastes a little bit of memory, | |
88 | * but it is much faster. | |
89 | */ | |
90 | xfs_buf_item_zone = | |
91 | kmem_zone_init((sizeof(xfs_buf_log_item_t) + | |
92 | (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) / | |
93 | NBWORD) * sizeof(int))), | |
94 | "xfs_buf_item"); | |
8758280f NS |
95 | xfs_efd_zone = |
96 | kmem_zone_init((sizeof(xfs_efd_log_item_t) + | |
97 | ((XFS_EFD_MAX_FAST_EXTENTS - 1) * | |
98 | sizeof(xfs_extent_t))), | |
1da177e4 | 99 | "xfs_efd_item"); |
8758280f NS |
100 | xfs_efi_zone = |
101 | kmem_zone_init((sizeof(xfs_efi_log_item_t) + | |
102 | ((XFS_EFI_MAX_FAST_EXTENTS - 1) * | |
103 | sizeof(xfs_extent_t))), | |
1da177e4 | 104 | "xfs_efi_item"); |
8758280f NS |
105 | |
106 | /* | |
107 | * These zones warrant special memory allocator hints | |
108 | */ | |
109 | xfs_inode_zone = | |
110 | kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode", | |
111 | KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | | |
112 | KM_ZONE_SPREAD, NULL); | |
113 | xfs_ili_zone = | |
114 | kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili", | |
115 | KM_ZONE_SPREAD, NULL); | |
da353b0d DC |
116 | xfs_icluster_zone = |
117 | kmem_zone_init_flags(sizeof(xfs_icluster_t), "xfs_icluster", | |
8758280f | 118 | KM_ZONE_SPREAD, NULL); |
1da177e4 LT |
119 | |
120 | /* | |
121 | * Allocate global trace buffers. | |
122 | */ | |
123 | #ifdef XFS_ALLOC_TRACE | |
124 | xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP); | |
125 | #endif | |
126 | #ifdef XFS_BMAP_TRACE | |
127 | xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP); | |
128 | #endif | |
129 | #ifdef XFS_BMBT_TRACE | |
130 | xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP); | |
131 | #endif | |
1da177e4 LT |
132 | #ifdef XFS_ATTR_TRACE |
133 | xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP); | |
134 | #endif | |
135 | #ifdef XFS_DIR2_TRACE | |
136 | xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP); | |
137 | #endif | |
138 | ||
139 | xfs_dir_startup(); | |
140 | ||
141 | #if (defined(DEBUG) || defined(INDUCE_IO_ERROR)) | |
142 | xfs_error_test_init(); | |
143 | #endif /* DEBUG || INDUCE_IO_ERROR */ | |
144 | ||
145 | xfs_init_procfs(); | |
146 | xfs_sysctl_register(); | |
147 | return 0; | |
148 | } | |
149 | ||
de2eeea6 | 150 | void __exit |
1da177e4 LT |
151 | xfs_cleanup(void) |
152 | { | |
1da177e4 | 153 | extern kmem_zone_t *xfs_inode_zone; |
1da177e4 LT |
154 | extern kmem_zone_t *xfs_efd_zone; |
155 | extern kmem_zone_t *xfs_efi_zone; | |
da353b0d | 156 | extern kmem_zone_t *xfs_icluster_zone; |
1da177e4 LT |
157 | |
158 | xfs_cleanup_procfs(); | |
159 | xfs_sysctl_unregister(); | |
160 | xfs_refcache_destroy(); | |
2a82b8be DC |
161 | xfs_filestream_uninit(); |
162 | xfs_mru_cache_uninit(); | |
1da177e4 LT |
163 | xfs_acl_zone_destroy(xfs_acl_zone); |
164 | ||
165 | #ifdef XFS_DIR2_TRACE | |
166 | ktrace_free(xfs_dir2_trace_buf); | |
167 | #endif | |
168 | #ifdef XFS_ATTR_TRACE | |
169 | ktrace_free(xfs_attr_trace_buf); | |
170 | #endif | |
1da177e4 LT |
171 | #ifdef XFS_BMBT_TRACE |
172 | ktrace_free(xfs_bmbt_trace_buf); | |
173 | #endif | |
174 | #ifdef XFS_BMAP_TRACE | |
175 | ktrace_free(xfs_bmap_trace_buf); | |
176 | #endif | |
177 | #ifdef XFS_ALLOC_TRACE | |
178 | ktrace_free(xfs_alloc_trace_buf); | |
179 | #endif | |
180 | ||
3758dee9 NS |
181 | kmem_zone_destroy(xfs_bmap_free_item_zone); |
182 | kmem_zone_destroy(xfs_btree_cur_zone); | |
183 | kmem_zone_destroy(xfs_inode_zone); | |
184 | kmem_zone_destroy(xfs_trans_zone); | |
185 | kmem_zone_destroy(xfs_da_state_zone); | |
186 | kmem_zone_destroy(xfs_dabuf_zone); | |
187 | kmem_zone_destroy(xfs_buf_item_zone); | |
188 | kmem_zone_destroy(xfs_efd_zone); | |
189 | kmem_zone_destroy(xfs_efi_zone); | |
190 | kmem_zone_destroy(xfs_ifork_zone); | |
191 | kmem_zone_destroy(xfs_ili_zone); | |
da353b0d | 192 | kmem_zone_destroy(xfs_icluster_zone); |
1da177e4 LT |
193 | } |
194 | ||
195 | /* | |
196 | * xfs_start_flags | |
197 | * | |
198 | * This function fills in xfs_mount_t fields based on mount args. | |
199 | * Note: the superblock has _not_ yet been read in. | |
200 | */ | |
201 | STATIC int | |
202 | xfs_start_flags( | |
1da177e4 LT |
203 | struct xfs_mount_args *ap, |
204 | struct xfs_mount *mp) | |
205 | { | |
206 | /* Values are in BBs */ | |
207 | if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) { | |
208 | /* | |
209 | * At this point the superblock has not been read | |
210 | * in, therefore we do not know the block size. | |
211 | * Before the mount call ends we will convert | |
212 | * these to FSBs. | |
213 | */ | |
214 | mp->m_dalign = ap->sunit; | |
215 | mp->m_swidth = ap->swidth; | |
216 | } | |
217 | ||
218 | if (ap->logbufs != -1 && | |
1da177e4 | 219 | ap->logbufs != 0 && |
1da177e4 LT |
220 | (ap->logbufs < XLOG_MIN_ICLOGS || |
221 | ap->logbufs > XLOG_MAX_ICLOGS)) { | |
222 | cmn_err(CE_WARN, | |
223 | "XFS: invalid logbufs value: %d [not %d-%d]", | |
224 | ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS); | |
225 | return XFS_ERROR(EINVAL); | |
226 | } | |
227 | mp->m_logbufs = ap->logbufs; | |
228 | if (ap->logbufsize != -1 && | |
cfcbbbd0 | 229 | ap->logbufsize != 0 && |
dcb3b83f ES |
230 | (ap->logbufsize < XLOG_MIN_RECORD_BSIZE || |
231 | ap->logbufsize > XLOG_MAX_RECORD_BSIZE || | |
232 | !is_power_of_2(ap->logbufsize))) { | |
1da177e4 LT |
233 | cmn_err(CE_WARN, |
234 | "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]", | |
235 | ap->logbufsize); | |
236 | return XFS_ERROR(EINVAL); | |
237 | } | |
1da177e4 LT |
238 | mp->m_logbsize = ap->logbufsize; |
239 | mp->m_fsname_len = strlen(ap->fsname) + 1; | |
240 | mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP); | |
241 | strcpy(mp->m_fsname, ap->fsname); | |
fc1f8c1c NS |
242 | if (ap->rtname[0]) { |
243 | mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP); | |
244 | strcpy(mp->m_rtname, ap->rtname); | |
245 | } | |
246 | if (ap->logname[0]) { | |
247 | mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP); | |
248 | strcpy(mp->m_logname, ap->logname); | |
249 | } | |
1da177e4 LT |
250 | |
251 | if (ap->flags & XFSMNT_WSYNC) | |
252 | mp->m_flags |= XFS_MOUNT_WSYNC; | |
253 | #if XFS_BIG_INUMS | |
254 | if (ap->flags & XFSMNT_INO64) { | |
255 | mp->m_flags |= XFS_MOUNT_INO64; | |
256 | mp->m_inoadd = XFS_INO64_OFFSET; | |
257 | } | |
258 | #endif | |
1da177e4 LT |
259 | if (ap->flags & XFSMNT_RETERR) |
260 | mp->m_flags |= XFS_MOUNT_RETERR; | |
1da177e4 LT |
261 | if (ap->flags & XFSMNT_NOALIGN) |
262 | mp->m_flags |= XFS_MOUNT_NOALIGN; | |
1da177e4 LT |
263 | if (ap->flags & XFSMNT_SWALLOC) |
264 | mp->m_flags |= XFS_MOUNT_SWALLOC; | |
1da177e4 LT |
265 | if (ap->flags & XFSMNT_OSYNCISOSYNC) |
266 | mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC; | |
1da177e4 | 267 | if (ap->flags & XFSMNT_32BITINODES) |
c11e2c36 | 268 | mp->m_flags |= XFS_MOUNT_32BITINODES; |
1da177e4 LT |
269 | |
270 | if (ap->flags & XFSMNT_IOSIZE) { | |
271 | if (ap->iosizelog > XFS_MAX_IO_LOG || | |
272 | ap->iosizelog < XFS_MIN_IO_LOG) { | |
273 | cmn_err(CE_WARN, | |
274 | "XFS: invalid log iosize: %d [not %d-%d]", | |
275 | ap->iosizelog, XFS_MIN_IO_LOG, | |
276 | XFS_MAX_IO_LOG); | |
277 | return XFS_ERROR(EINVAL); | |
278 | } | |
279 | ||
280 | mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE; | |
281 | mp->m_readio_log = mp->m_writeio_log = ap->iosizelog; | |
282 | } | |
283 | ||
1da177e4 LT |
284 | if (ap->flags & XFSMNT_IDELETE) |
285 | mp->m_flags |= XFS_MOUNT_IDELETE; | |
1da177e4 LT |
286 | if (ap->flags & XFSMNT_DIRSYNC) |
287 | mp->m_flags |= XFS_MOUNT_DIRSYNC; | |
13059ff0 NS |
288 | if (ap->flags & XFSMNT_ATTR2) |
289 | mp->m_flags |= XFS_MOUNT_ATTR2; | |
e8c8b3a7 | 290 | |
e718eeb4 NS |
291 | if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE) |
292 | mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE; | |
293 | ||
1da177e4 LT |
294 | /* |
295 | * no recovery flag requires a read-only mount | |
296 | */ | |
297 | if (ap->flags & XFSMNT_NORECOVERY) { | |
bd186aa9 | 298 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { |
1da177e4 LT |
299 | cmn_err(CE_WARN, |
300 | "XFS: tried to mount a FS read-write without recovery!"); | |
301 | return XFS_ERROR(EINVAL); | |
302 | } | |
303 | mp->m_flags |= XFS_MOUNT_NORECOVERY; | |
304 | } | |
305 | ||
306 | if (ap->flags & XFSMNT_NOUUID) | |
307 | mp->m_flags |= XFS_MOUNT_NOUUID; | |
f538d4da CH |
308 | if (ap->flags & XFSMNT_BARRIER) |
309 | mp->m_flags |= XFS_MOUNT_BARRIER; | |
4ef19ddd CH |
310 | else |
311 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
1da177e4 | 312 | |
2a82b8be DC |
313 | if (ap->flags2 & XFSMNT2_FILESTREAMS) |
314 | mp->m_flags |= XFS_MOUNT_FILESTREAMS; | |
315 | ||
293688ec | 316 | if (ap->flags & XFSMNT_DMAPI) |
bd186aa9 | 317 | mp->m_flags |= XFS_MOUNT_DMAPI; |
1da177e4 LT |
318 | return 0; |
319 | } | |
320 | ||
321 | /* | |
322 | * This function fills in xfs_mount_t fields based on mount args. | |
323 | * Note: the superblock _has_ now been read in. | |
324 | */ | |
325 | STATIC int | |
326 | xfs_finish_flags( | |
1da177e4 LT |
327 | struct xfs_mount_args *ap, |
328 | struct xfs_mount *mp) | |
329 | { | |
bd186aa9 | 330 | int ronly = (mp->m_flags & XFS_MOUNT_RDONLY); |
1da177e4 LT |
331 | |
332 | /* Fail a mount where the logbuf is smaller then the log stripe */ | |
333 | if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) { | |
c2cd2550 | 334 | if ((ap->logbufsize <= 0) && |
1da177e4 LT |
335 | (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) { |
336 | mp->m_logbsize = mp->m_sb.sb_logsunit; | |
c2cd2550 NS |
337 | } else if (ap->logbufsize > 0 && |
338 | ap->logbufsize < mp->m_sb.sb_logsunit) { | |
1da177e4 LT |
339 | cmn_err(CE_WARN, |
340 | "XFS: logbuf size must be greater than or equal to log stripe size"); | |
341 | return XFS_ERROR(EINVAL); | |
342 | } | |
343 | } else { | |
344 | /* Fail a mount if the logbuf is larger than 32K */ | |
345 | if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) { | |
346 | cmn_err(CE_WARN, | |
347 | "XFS: logbuf size for version 1 logs must be 16K or 32K"); | |
348 | return XFS_ERROR(EINVAL); | |
349 | } | |
350 | } | |
351 | ||
13059ff0 NS |
352 | if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) { |
353 | mp->m_flags |= XFS_MOUNT_ATTR2; | |
354 | } | |
355 | ||
1da177e4 LT |
356 | /* |
357 | * prohibit r/w mounts of read-only filesystems | |
358 | */ | |
359 | if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) { | |
360 | cmn_err(CE_WARN, | |
361 | "XFS: cannot mount a read-only filesystem as read-write"); | |
362 | return XFS_ERROR(EROFS); | |
363 | } | |
364 | ||
1da177e4 LT |
365 | /* |
366 | * check for shared mount. | |
367 | */ | |
368 | if (ap->flags & XFSMNT_SHARED) { | |
369 | if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb)) | |
370 | return XFS_ERROR(EINVAL); | |
371 | ||
372 | /* | |
373 | * For IRIX 6.5, shared mounts must have the shared | |
374 | * version bit set, have the persistent readonly | |
375 | * field set, must be version 0 and can only be mounted | |
376 | * read-only. | |
377 | */ | |
378 | if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) || | |
379 | (mp->m_sb.sb_shared_vn != 0)) | |
380 | return XFS_ERROR(EINVAL); | |
381 | ||
382 | mp->m_flags |= XFS_MOUNT_SHARED; | |
383 | ||
384 | /* | |
385 | * Shared XFS V0 can't deal with DMI. Return EINVAL. | |
386 | */ | |
387 | if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI)) | |
388 | return XFS_ERROR(EINVAL); | |
389 | } | |
390 | ||
b09cc771 CH |
391 | if (ap->flags & XFSMNT_UQUOTA) { |
392 | mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE); | |
393 | if (ap->flags & XFSMNT_UQUOTAENF) | |
394 | mp->m_qflags |= XFS_UQUOTA_ENFD; | |
395 | } | |
396 | ||
397 | if (ap->flags & XFSMNT_GQUOTA) { | |
398 | mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE); | |
399 | if (ap->flags & XFSMNT_GQUOTAENF) | |
400 | mp->m_qflags |= XFS_OQUOTA_ENFD; | |
401 | } else if (ap->flags & XFSMNT_PQUOTA) { | |
402 | mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE); | |
403 | if (ap->flags & XFSMNT_PQUOTAENF) | |
404 | mp->m_qflags |= XFS_OQUOTA_ENFD; | |
405 | } | |
406 | ||
1da177e4 LT |
407 | return 0; |
408 | } | |
409 | ||
410 | /* | |
411 | * xfs_mount | |
412 | * | |
413 | * The file system configurations are: | |
414 | * (1) device (partition) with data and internal log | |
415 | * (2) logical volume with data and log subvolumes. | |
416 | * (3) logical volume with data, log, and realtime subvolumes. | |
417 | * | |
418 | * We only have to handle opening the log and realtime volumes here if | |
419 | * they are present. The data subvolume has already been opened by | |
420 | * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev. | |
421 | */ | |
48c872a9 | 422 | int |
1da177e4 | 423 | xfs_mount( |
48c872a9 | 424 | struct xfs_mount *mp, |
1da177e4 LT |
425 | struct xfs_mount_args *args, |
426 | cred_t *credp) | |
427 | { | |
1da177e4 LT |
428 | struct block_device *ddev, *logdev, *rtdev; |
429 | int flags = 0, error; | |
430 | ||
b267ce99 | 431 | ddev = mp->m_super->s_bdev; |
1da177e4 LT |
432 | logdev = rtdev = NULL; |
433 | ||
293688ec | 434 | error = xfs_dmops_get(mp, args); |
b09cc771 CH |
435 | if (error) |
436 | return error; | |
437 | error = xfs_qmops_get(mp, args); | |
293688ec CH |
438 | if (error) |
439 | return error; | |
440 | ||
764d1f89 NS |
441 | if (args->flags & XFSMNT_QUIET) |
442 | flags |= XFS_MFSI_QUIET; | |
443 | ||
1da177e4 LT |
444 | /* |
445 | * Open real time and log devices - order is important. | |
446 | */ | |
447 | if (args->logname[0]) { | |
448 | error = xfs_blkdev_get(mp, args->logname, &logdev); | |
449 | if (error) | |
450 | return error; | |
451 | } | |
452 | if (args->rtname[0]) { | |
453 | error = xfs_blkdev_get(mp, args->rtname, &rtdev); | |
454 | if (error) { | |
455 | xfs_blkdev_put(logdev); | |
456 | return error; | |
457 | } | |
458 | ||
459 | if (rtdev == ddev || rtdev == logdev) { | |
460 | cmn_err(CE_WARN, | |
461 | "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev."); | |
462 | xfs_blkdev_put(logdev); | |
463 | xfs_blkdev_put(rtdev); | |
464 | return EINVAL; | |
465 | } | |
466 | } | |
467 | ||
468 | /* | |
469 | * Setup xfs_mount buffer target pointers | |
470 | */ | |
471 | error = ENOMEM; | |
472 | mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0); | |
473 | if (!mp->m_ddev_targp) { | |
474 | xfs_blkdev_put(logdev); | |
475 | xfs_blkdev_put(rtdev); | |
476 | return error; | |
477 | } | |
478 | if (rtdev) { | |
479 | mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1); | |
49ee6c91 JJ |
480 | if (!mp->m_rtdev_targp) { |
481 | xfs_blkdev_put(logdev); | |
482 | xfs_blkdev_put(rtdev); | |
1da177e4 | 483 | goto error0; |
49ee6c91 | 484 | } |
1da177e4 LT |
485 | } |
486 | mp->m_logdev_targp = (logdev && logdev != ddev) ? | |
487 | xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp; | |
49ee6c91 JJ |
488 | if (!mp->m_logdev_targp) { |
489 | xfs_blkdev_put(logdev); | |
490 | xfs_blkdev_put(rtdev); | |
1da177e4 | 491 | goto error0; |
49ee6c91 | 492 | } |
1da177e4 LT |
493 | |
494 | /* | |
495 | * Setup flags based on mount(2) options and then the superblock | |
496 | */ | |
b267ce99 | 497 | error = xfs_start_flags(args, mp); |
1da177e4 LT |
498 | if (error) |
499 | goto error1; | |
764d1f89 | 500 | error = xfs_readsb(mp, flags); |
1da177e4 LT |
501 | if (error) |
502 | goto error1; | |
b267ce99 | 503 | error = xfs_finish_flags(args, mp); |
1da177e4 LT |
504 | if (error) |
505 | goto error2; | |
506 | ||
507 | /* | |
508 | * Setup xfs_mount buffer target pointers based on superblock | |
509 | */ | |
510 | error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize, | |
511 | mp->m_sb.sb_sectsize); | |
512 | if (!error && logdev && logdev != ddev) { | |
513 | unsigned int log_sector_size = BBSIZE; | |
514 | ||
515 | if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb)) | |
516 | log_sector_size = mp->m_sb.sb_logsectsize; | |
517 | error = xfs_setsize_buftarg(mp->m_logdev_targp, | |
518 | mp->m_sb.sb_blocksize, | |
519 | log_sector_size); | |
520 | } | |
521 | if (!error && rtdev) | |
522 | error = xfs_setsize_buftarg(mp->m_rtdev_targp, | |
523 | mp->m_sb.sb_blocksize, | |
524 | mp->m_sb.sb_sectsize); | |
525 | if (error) | |
526 | goto error2; | |
527 | ||
b2ea401b | 528 | if (mp->m_flags & XFS_MOUNT_BARRIER) |
c7d437da CH |
529 | xfs_mountfs_check_barriers(mp); |
530 | ||
2a82b8be DC |
531 | if ((error = xfs_filestream_mount(mp))) |
532 | goto error2; | |
533 | ||
541d7d3c | 534 | error = xfs_mountfs(mp, flags); |
f538d4da CH |
535 | if (error) |
536 | goto error2; | |
537 | ||
293688ec CH |
538 | XFS_SEND_MOUNT(mp, DM_RIGHT_NULL, args->mtpt, args->fsname); |
539 | ||
f538d4da CH |
540 | return 0; |
541 | ||
1da177e4 LT |
542 | error2: |
543 | if (mp->m_sb_bp) | |
544 | xfs_freesb(mp); | |
545 | error1: | |
546 | xfs_binval(mp->m_ddev_targp); | |
547 | if (logdev && logdev != ddev) | |
548 | xfs_binval(mp->m_logdev_targp); | |
549 | if (rtdev) | |
550 | xfs_binval(mp->m_rtdev_targp); | |
551 | error0: | |
552 | xfs_unmountfs_close(mp, credp); | |
b09cc771 | 553 | xfs_qmops_put(mp); |
293688ec | 554 | xfs_dmops_put(mp); |
1da177e4 LT |
555 | return error; |
556 | } | |
557 | ||
48c872a9 | 558 | int |
1da177e4 | 559 | xfs_unmount( |
48c872a9 | 560 | xfs_mount_t *mp, |
1da177e4 LT |
561 | int flags, |
562 | cred_t *credp) | |
563 | { | |
1da177e4 | 564 | xfs_inode_t *rip; |
67fcaa73 | 565 | bhv_vnode_t *rvp; |
1da177e4 LT |
566 | int unmount_event_wanted = 0; |
567 | int unmount_event_flags = 0; | |
568 | int xfs_unmountfs_needed = 0; | |
569 | int error; | |
570 | ||
571 | rip = mp->m_rootip; | |
572 | rvp = XFS_ITOV(rip); | |
573 | ||
948c6d4f | 574 | #ifdef HAVE_DMAPI |
bd186aa9 | 575 | if (mp->m_flags & XFS_MOUNT_DMAPI) { |
b267ce99 | 576 | error = XFS_SEND_PREUNMOUNT(mp, |
1da177e4 LT |
577 | rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL, |
578 | NULL, NULL, 0, 0, | |
579 | (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))? | |
580 | 0:DM_FLAGS_UNWANTED); | |
581 | if (error) | |
582 | return XFS_ERROR(error); | |
583 | unmount_event_wanted = 1; | |
584 | unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))? | |
585 | 0 : DM_FLAGS_UNWANTED; | |
586 | } | |
948c6d4f | 587 | #endif |
1da177e4 LT |
588 | /* |
589 | * First blow any referenced inode from this file system | |
590 | * out of the reference cache, and delete the timer. | |
591 | */ | |
592 | xfs_refcache_purge_mp(mp); | |
593 | ||
2a82b8be DC |
594 | /* |
595 | * Blow away any referenced inode in the filestreams cache. | |
596 | * This can and will cause log traffic as inodes go inactive | |
597 | * here. | |
598 | */ | |
599 | xfs_filestream_unmount(mp); | |
600 | ||
1da177e4 LT |
601 | XFS_bflush(mp->m_ddev_targp); |
602 | error = xfs_unmount_flush(mp, 0); | |
603 | if (error) | |
604 | goto out; | |
605 | ||
606 | ASSERT(vn_count(rvp) == 1); | |
607 | ||
608 | /* | |
609 | * Drop the reference count | |
610 | */ | |
611 | VN_RELE(rvp); | |
612 | ||
613 | /* | |
614 | * If we're forcing a shutdown, typically because of a media error, | |
615 | * we want to make sure we invalidate dirty pages that belong to | |
616 | * referenced vnodes as well. | |
617 | */ | |
618 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
48c872a9 | 619 | error = xfs_sync(mp, SYNC_WAIT | SYNC_CLOSE); |
1da177e4 LT |
620 | ASSERT(error != EFSCORRUPTED); |
621 | } | |
622 | xfs_unmountfs_needed = 1; | |
623 | ||
624 | out: | |
625 | /* Send DMAPI event, if required. | |
626 | * Then do xfs_unmountfs() if needed. | |
627 | * Then return error (or zero). | |
628 | */ | |
629 | if (unmount_event_wanted) { | |
630 | /* Note: mp structure must still exist for | |
631 | * XFS_SEND_UNMOUNT() call. | |
632 | */ | |
b267ce99 | 633 | XFS_SEND_UNMOUNT(mp, error == 0 ? rvp : NULL, |
1da177e4 LT |
634 | DM_RIGHT_NULL, 0, error, unmount_event_flags); |
635 | } | |
636 | if (xfs_unmountfs_needed) { | |
637 | /* | |
638 | * Call common unmount function to flush to disk | |
639 | * and free the super block buffer & mount structures. | |
640 | */ | |
641 | xfs_unmountfs(mp, credp); | |
b09cc771 | 642 | xfs_qmops_put(mp); |
293688ec | 643 | xfs_dmops_put(mp); |
f541d270 | 644 | kmem_free(mp, sizeof(xfs_mount_t)); |
1da177e4 LT |
645 | } |
646 | ||
647 | return XFS_ERROR(error); | |
648 | } | |
649 | ||
f898d6c0 CH |
650 | STATIC int |
651 | xfs_quiesce_fs( | |
652 | xfs_mount_t *mp) | |
653 | { | |
654 | int count = 0, pincount; | |
3758dee9 | 655 | |
f898d6c0 CH |
656 | xfs_refcache_purge_mp(mp); |
657 | xfs_flush_buftarg(mp->m_ddev_targp, 0); | |
658 | xfs_finish_reclaim_all(mp, 0); | |
659 | ||
660 | /* This loop must run at least twice. | |
661 | * The first instance of the loop will flush | |
662 | * most meta data but that will generate more | |
663 | * meta data (typically directory updates). | |
664 | * Which then must be flushed and logged before | |
665 | * we can write the unmount record. | |
3758dee9 | 666 | */ |
f898d6c0 | 667 | do { |
516b2e7c | 668 | xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL); |
f898d6c0 CH |
669 | pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1); |
670 | if (!pincount) { | |
671 | delay(50); | |
672 | count++; | |
673 | } | |
674 | } while (count < 2); | |
675 | ||
676 | return 0; | |
677 | } | |
1da177e4 | 678 | |
516b2e7c DC |
679 | /* |
680 | * Second stage of a quiesce. The data is already synced, now we have to take | |
681 | * care of the metadata. New transactions are already blocked, so we need to | |
682 | * wait for any remaining transactions to drain out before proceding. | |
683 | */ | |
9909c4aa | 684 | void |
516b2e7c DC |
685 | xfs_attr_quiesce( |
686 | xfs_mount_t *mp) | |
687 | { | |
688 | /* wait for all modifications to complete */ | |
689 | while (atomic_read(&mp->m_active_trans) > 0) | |
690 | delay(100); | |
691 | ||
692 | /* flush inodes and push all remaining buffers out to disk */ | |
693 | xfs_quiesce_fs(mp); | |
694 | ||
695 | ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0); | |
696 | ||
697 | /* Push the superblock and write an unmount record */ | |
698 | xfs_log_sbcount(mp, 1); | |
699 | xfs_log_unmount_write(mp); | |
700 | xfs_unmountfs_writesb(mp); | |
701 | } | |
702 | ||
48c872a9 | 703 | int |
1da177e4 | 704 | xfs_mntupdate( |
48c872a9 | 705 | struct xfs_mount *mp, |
1da177e4 LT |
706 | int *flags, |
707 | struct xfs_mount_args *args) | |
708 | { | |
d08d389d | 709 | if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */ |
bd186aa9 CH |
710 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
711 | mp->m_flags &= ~XFS_MOUNT_RDONLY; | |
d08d389d NS |
712 | if (args->flags & XFSMNT_BARRIER) { |
713 | mp->m_flags |= XFS_MOUNT_BARRIER; | |
f538d4da | 714 | xfs_mountfs_check_barriers(mp); |
d08d389d NS |
715 | } else { |
716 | mp->m_flags &= ~XFS_MOUNT_BARRIER; | |
717 | } | |
bd186aa9 | 718 | } else if (!(mp->m_flags & XFS_MOUNT_RDONLY)) { /* rw -> ro */ |
2a82b8be | 719 | xfs_filestream_flush(mp); |
745f6919 | 720 | xfs_sync(mp, SYNC_DATA_QUIESCE); |
516b2e7c | 721 | xfs_attr_quiesce(mp); |
bd186aa9 | 722 | mp->m_flags |= XFS_MOUNT_RDONLY; |
1da177e4 | 723 | } |
1da177e4 LT |
724 | return 0; |
725 | } | |
726 | ||
727 | /* | |
728 | * xfs_unmount_flush implements a set of flush operation on special | |
729 | * inodes, which are needed as a separate set of operations so that | |
730 | * they can be called as part of relocation process. | |
731 | */ | |
732 | int | |
733 | xfs_unmount_flush( | |
734 | xfs_mount_t *mp, /* Mount structure we are getting | |
735 | rid of. */ | |
736 | int relocation) /* Called from vfs relocation. */ | |
737 | { | |
738 | xfs_inode_t *rip = mp->m_rootip; | |
739 | xfs_inode_t *rbmip; | |
740 | xfs_inode_t *rsumip = NULL; | |
67fcaa73 | 741 | bhv_vnode_t *rvp = XFS_ITOV(rip); |
1da177e4 LT |
742 | int error; |
743 | ||
f7c66ce3 | 744 | xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); |
1da177e4 LT |
745 | xfs_iflock(rip); |
746 | ||
747 | /* | |
748 | * Flush out the real time inodes. | |
749 | */ | |
750 | if ((rbmip = mp->m_rbmip) != NULL) { | |
751 | xfs_ilock(rbmip, XFS_ILOCK_EXCL); | |
752 | xfs_iflock(rbmip); | |
753 | error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC); | |
754 | xfs_iunlock(rbmip, XFS_ILOCK_EXCL); | |
755 | ||
756 | if (error == EFSCORRUPTED) | |
757 | goto fscorrupt_out; | |
758 | ||
759 | ASSERT(vn_count(XFS_ITOV(rbmip)) == 1); | |
760 | ||
761 | rsumip = mp->m_rsumip; | |
762 | xfs_ilock(rsumip, XFS_ILOCK_EXCL); | |
763 | xfs_iflock(rsumip); | |
764 | error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC); | |
765 | xfs_iunlock(rsumip, XFS_ILOCK_EXCL); | |
766 | ||
767 | if (error == EFSCORRUPTED) | |
768 | goto fscorrupt_out; | |
769 | ||
770 | ASSERT(vn_count(XFS_ITOV(rsumip)) == 1); | |
771 | } | |
772 | ||
773 | /* | |
774 | * Synchronously flush root inode to disk | |
775 | */ | |
776 | error = xfs_iflush(rip, XFS_IFLUSH_SYNC); | |
777 | if (error == EFSCORRUPTED) | |
778 | goto fscorrupt_out2; | |
779 | ||
780 | if (vn_count(rvp) != 1 && !relocation) { | |
781 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
782 | return XFS_ERROR(EBUSY); | |
783 | } | |
784 | ||
785 | /* | |
786 | * Release dquot that rootinode, rbmino and rsumino might be holding, | |
787 | * flush and purge the quota inodes. | |
788 | */ | |
789 | error = XFS_QM_UNMOUNT(mp); | |
790 | if (error == EFSCORRUPTED) | |
791 | goto fscorrupt_out2; | |
792 | ||
793 | if (rbmip) { | |
794 | VN_RELE(XFS_ITOV(rbmip)); | |
795 | VN_RELE(XFS_ITOV(rsumip)); | |
796 | } | |
797 | ||
798 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
799 | return 0; | |
800 | ||
801 | fscorrupt_out: | |
802 | xfs_ifunlock(rip); | |
803 | ||
804 | fscorrupt_out2: | |
805 | xfs_iunlock(rip, XFS_ILOCK_EXCL); | |
806 | ||
807 | return XFS_ERROR(EFSCORRUPTED); | |
808 | } | |
809 | ||
1da177e4 LT |
810 | /* |
811 | * xfs_sync flushes any pending I/O to file system vfsp. | |
812 | * | |
813 | * This routine is called by vfs_sync() to make sure that things make it | |
814 | * out to disk eventually, on sync() system calls to flush out everything, | |
815 | * and when the file system is unmounted. For the vfs_sync() case, all | |
816 | * we really need to do is sync out the log to make all of our meta-data | |
817 | * updates permanent (except for timestamps). For calls from pflushd(), | |
818 | * dirty pages are kept moving by calling pdflush() on the inodes | |
819 | * containing them. We also flush the inodes that we can lock without | |
820 | * sleeping and the superblock if we can lock it without sleeping from | |
821 | * vfs_sync() so that items at the tail of the log are always moving out. | |
822 | * | |
823 | * Flags: | |
824 | * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want | |
825 | * to sleep if we can help it. All we really need | |
826 | * to do is ensure that the log is synced at least | |
827 | * periodically. We also push the inodes and | |
828 | * superblock if we can lock them without sleeping | |
829 | * and they are not pinned. | |
830 | * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not | |
831 | * set, then we really want to lock each inode and flush | |
832 | * it. | |
833 | * SYNC_WAIT - All the flushes that take place in this call should | |
834 | * be synchronous. | |
835 | * SYNC_DELWRI - This tells us to push dirty pages associated with | |
836 | * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to | |
837 | * determine if they should be flushed sync, async, or | |
838 | * delwri. | |
839 | * SYNC_CLOSE - This flag is passed when the system is being | |
c41564b5 | 840 | * unmounted. We should sync and invalidate everything. |
1da177e4 LT |
841 | * SYNC_FSDATA - This indicates that the caller would like to make |
842 | * sure the superblock is safe on disk. We can ensure | |
c41564b5 | 843 | * this by simply making sure the log gets flushed |
1da177e4 LT |
844 | * if SYNC_BDFLUSH is set, and by actually writing it |
845 | * out otherwise. | |
3c0dc77b DC |
846 | * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete |
847 | * before we return (including direct I/O). Forms the drain | |
848 | * side of the write barrier needed to safely quiesce the | |
849 | * filesystem. | |
1da177e4 LT |
850 | * |
851 | */ | |
48c872a9 | 852 | int |
1da177e4 | 853 | xfs_sync( |
48c872a9 CH |
854 | xfs_mount_t *mp, |
855 | int flags) | |
1da177e4 | 856 | { |
b09cc771 CH |
857 | int error; |
858 | ||
859 | /* | |
860 | * Get the Quota Manager to flush the dquots. | |
861 | * | |
862 | * If XFS quota support is not enabled or this filesystem | |
863 | * instance does not use quotas XFS_QM_DQSYNC will always | |
864 | * return zero. | |
865 | */ | |
866 | error = XFS_QM_DQSYNC(mp, flags); | |
867 | if (error) { | |
868 | /* | |
869 | * If we got an IO error, we will be shutting down. | |
870 | * So, there's nothing more for us to do here. | |
871 | */ | |
872 | ASSERT(error != EIO || XFS_FORCED_SHUTDOWN(mp)); | |
873 | if (XFS_FORCED_SHUTDOWN(mp)) | |
874 | return XFS_ERROR(error); | |
875 | } | |
1da177e4 | 876 | |
2a82b8be DC |
877 | if (flags & SYNC_IOWAIT) |
878 | xfs_filestream_flush(mp); | |
879 | ||
2823945f | 880 | return xfs_syncsub(mp, flags, NULL); |
1da177e4 LT |
881 | } |
882 | ||
883 | /* | |
884 | * xfs sync routine for internal use | |
885 | * | |
b83bd138 | 886 | * This routine supports all of the flags defined for the generic vfs_sync |
1f9b3b64 | 887 | * interface as explained above under xfs_sync. |
1da177e4 LT |
888 | * |
889 | */ | |
ee34807a | 890 | int |
1da177e4 LT |
891 | xfs_sync_inodes( |
892 | xfs_mount_t *mp, | |
893 | int flags, | |
1da177e4 LT |
894 | int *bypassed) |
895 | { | |
896 | xfs_inode_t *ip = NULL; | |
67fcaa73 | 897 | bhv_vnode_t *vp = NULL; |
1da177e4 LT |
898 | int error; |
899 | int last_error; | |
900 | uint64_t fflag; | |
901 | uint lock_flags; | |
902 | uint base_lock_flags; | |
903 | boolean_t mount_locked; | |
904 | boolean_t vnode_refed; | |
905 | int preempt; | |
1da177e4 LT |
906 | xfs_iptr_t *ipointer; |
907 | #ifdef DEBUG | |
908 | boolean_t ipointer_in = B_FALSE; | |
909 | ||
910 | #define IPOINTER_SET ipointer_in = B_TRUE | |
911 | #define IPOINTER_CLR ipointer_in = B_FALSE | |
912 | #else | |
913 | #define IPOINTER_SET | |
914 | #define IPOINTER_CLR | |
915 | #endif | |
916 | ||
917 | ||
918 | /* Insert a marker record into the inode list after inode ip. The list | |
919 | * must be locked when this is called. After the call the list will no | |
920 | * longer be locked. | |
921 | */ | |
922 | #define IPOINTER_INSERT(ip, mp) { \ | |
923 | ASSERT(ipointer_in == B_FALSE); \ | |
924 | ipointer->ip_mnext = ip->i_mnext; \ | |
925 | ipointer->ip_mprev = ip; \ | |
926 | ip->i_mnext = (xfs_inode_t *)ipointer; \ | |
927 | ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \ | |
928 | preempt = 0; \ | |
929 | XFS_MOUNT_IUNLOCK(mp); \ | |
930 | mount_locked = B_FALSE; \ | |
931 | IPOINTER_SET; \ | |
932 | } | |
933 | ||
934 | /* Remove the marker from the inode list. If the marker was the only item | |
935 | * in the list then there are no remaining inodes and we should zero out | |
936 | * the whole list. If we are the current head of the list then move the head | |
937 | * past us. | |
938 | */ | |
939 | #define IPOINTER_REMOVE(ip, mp) { \ | |
940 | ASSERT(ipointer_in == B_TRUE); \ | |
941 | if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \ | |
942 | ip = ipointer->ip_mnext; \ | |
943 | ip->i_mprev = ipointer->ip_mprev; \ | |
944 | ipointer->ip_mprev->i_mnext = ip; \ | |
945 | if (mp->m_inodes == (xfs_inode_t *)ipointer) { \ | |
946 | mp->m_inodes = ip; \ | |
947 | } \ | |
948 | } else { \ | |
949 | ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \ | |
950 | mp->m_inodes = NULL; \ | |
951 | ip = NULL; \ | |
952 | } \ | |
953 | IPOINTER_CLR; \ | |
954 | } | |
955 | ||
956 | #define XFS_PREEMPT_MASK 0x7f | |
957 | ||
44866d39 LM |
958 | ASSERT(!(flags & SYNC_BDFLUSH)); |
959 | ||
1da177e4 LT |
960 | if (bypassed) |
961 | *bypassed = 0; | |
bd186aa9 | 962 | if (mp->m_flags & XFS_MOUNT_RDONLY) |
1da177e4 LT |
963 | return 0; |
964 | error = 0; | |
965 | last_error = 0; | |
966 | preempt = 0; | |
967 | ||
968 | /* Allocate a reference marker */ | |
969 | ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP); | |
970 | ||
971 | fflag = XFS_B_ASYNC; /* default is don't wait */ | |
44866d39 | 972 | if (flags & SYNC_DELWRI) |
1da177e4 LT |
973 | fflag = XFS_B_DELWRI; |
974 | if (flags & SYNC_WAIT) | |
975 | fflag = 0; /* synchronous overrides all */ | |
976 | ||
977 | base_lock_flags = XFS_ILOCK_SHARED; | |
978 | if (flags & (SYNC_DELWRI | SYNC_CLOSE)) { | |
979 | /* | |
980 | * We need the I/O lock if we're going to call any of | |
981 | * the flush/inval routines. | |
982 | */ | |
983 | base_lock_flags |= XFS_IOLOCK_SHARED; | |
984 | } | |
985 | ||
986 | XFS_MOUNT_ILOCK(mp); | |
987 | ||
988 | ip = mp->m_inodes; | |
989 | ||
990 | mount_locked = B_TRUE; | |
991 | vnode_refed = B_FALSE; | |
992 | ||
993 | IPOINTER_CLR; | |
994 | ||
995 | do { | |
996 | ASSERT(ipointer_in == B_FALSE); | |
997 | ASSERT(vnode_refed == B_FALSE); | |
998 | ||
999 | lock_flags = base_lock_flags; | |
1000 | ||
1001 | /* | |
1002 | * There were no inodes in the list, just break out | |
1003 | * of the loop. | |
1004 | */ | |
1005 | if (ip == NULL) { | |
1006 | break; | |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * We found another sync thread marker - skip it | |
1011 | */ | |
1012 | if (ip->i_mount == NULL) { | |
1013 | ip = ip->i_mnext; | |
1014 | continue; | |
1015 | } | |
1016 | ||
1017 | vp = XFS_ITOV_NULL(ip); | |
1018 | ||
1019 | /* | |
1020 | * If the vnode is gone then this is being torn down, | |
1021 | * call reclaim if it is flushed, else let regular flush | |
1022 | * code deal with it later in the loop. | |
1023 | */ | |
1024 | ||
1025 | if (vp == NULL) { | |
1026 | /* Skip ones already in reclaim */ | |
1027 | if (ip->i_flags & XFS_IRECLAIM) { | |
1028 | ip = ip->i_mnext; | |
1029 | continue; | |
1030 | } | |
1031 | if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) { | |
1032 | ip = ip->i_mnext; | |
1033 | } else if ((xfs_ipincount(ip) == 0) && | |
1034 | xfs_iflock_nowait(ip)) { | |
1035 | IPOINTER_INSERT(ip, mp); | |
1036 | ||
1037 | xfs_finish_reclaim(ip, 1, | |
1038 | XFS_IFLUSH_DELWRI_ELSE_ASYNC); | |
1039 | ||
1040 | XFS_MOUNT_ILOCK(mp); | |
1041 | mount_locked = B_TRUE; | |
1042 | IPOINTER_REMOVE(ip, mp); | |
1043 | } else { | |
1044 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1045 | ip = ip->i_mnext; | |
1046 | } | |
1047 | continue; | |
1048 | } | |
1049 | ||
1050 | if (VN_BAD(vp)) { | |
1051 | ip = ip->i_mnext; | |
1052 | continue; | |
1053 | } | |
1054 | ||
1055 | if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) { | |
1056 | XFS_MOUNT_IUNLOCK(mp); | |
1057 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1058 | return 0; | |
1059 | } | |
1060 | ||
1da177e4 LT |
1061 | /* |
1062 | * Try to lock without sleeping. We're out of order with | |
1063 | * the inode list lock here, so if we fail we need to drop | |
1064 | * the mount lock and try again. If we're called from | |
1065 | * bdflush() here, then don't bother. | |
1066 | * | |
1067 | * The inode lock here actually coordinates with the | |
1068 | * almost spurious inode lock in xfs_ireclaim() to prevent | |
1069 | * the vnode we handle here without a reference from | |
1070 | * being freed while we reference it. If we lock the inode | |
1071 | * while it's on the mount list here, then the spurious inode | |
1072 | * lock in xfs_ireclaim() after the inode is pulled from | |
1073 | * the mount list will sleep until we release it here. | |
1074 | * This keeps the vnode from being freed while we reference | |
cdb62687 | 1075 | * it. |
1da177e4 LT |
1076 | */ |
1077 | if (xfs_ilock_nowait(ip, lock_flags) == 0) { | |
44866d39 | 1078 | if (vp == NULL) { |
1da177e4 LT |
1079 | ip = ip->i_mnext; |
1080 | continue; | |
1081 | } | |
1082 | ||
cdb62687 | 1083 | vp = vn_grab(vp); |
1da177e4 | 1084 | if (vp == NULL) { |
cdb62687 | 1085 | ip = ip->i_mnext; |
1da177e4 LT |
1086 | continue; |
1087 | } | |
1088 | ||
cdb62687 | 1089 | IPOINTER_INSERT(ip, mp); |
1da177e4 LT |
1090 | xfs_ilock(ip, lock_flags); |
1091 | ||
1092 | ASSERT(vp == XFS_ITOV(ip)); | |
1093 | ASSERT(ip->i_mount == mp); | |
1094 | ||
1095 | vnode_refed = B_TRUE; | |
1096 | } | |
1097 | ||
1098 | /* From here on in the loop we may have a marker record | |
1099 | * in the inode list. | |
1100 | */ | |
1101 | ||
40095b64 DC |
1102 | /* |
1103 | * If we have to flush data or wait for I/O completion | |
1104 | * we need to drop the ilock that we currently hold. | |
1105 | * If we need to drop the lock, insert a marker if we | |
1106 | * have not already done so. | |
1107 | */ | |
1108 | if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) || | |
1109 | ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) { | |
1110 | if (mount_locked) { | |
1111 | IPOINTER_INSERT(ip, mp); | |
1da177e4 | 1112 | } |
40095b64 | 1113 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
1da177e4 | 1114 | |
40095b64 DC |
1115 | if (flags & SYNC_CLOSE) { |
1116 | /* Shutdown case. Flush and invalidate. */ | |
1117 | if (XFS_FORCED_SHUTDOWN(mp)) | |
739bfb2a CH |
1118 | xfs_tosspages(ip, 0, -1, |
1119 | FI_REMAPF); | |
40095b64 | 1120 | else |
739bfb2a CH |
1121 | error = xfs_flushinval_pages(ip, |
1122 | 0, -1, FI_REMAPF); | |
40095b64 | 1123 | } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) { |
739bfb2a | 1124 | error = xfs_flush_pages(ip, 0, |
67fcaa73 | 1125 | -1, fflag, FI_NONE); |
1da177e4 LT |
1126 | } |
1127 | ||
40095b64 DC |
1128 | /* |
1129 | * When freezing, we need to wait ensure all I/O (including direct | |
1130 | * I/O) is complete to ensure no further data modification can take | |
1131 | * place after this point | |
1132 | */ | |
1133 | if (flags & SYNC_IOWAIT) | |
b677c210 | 1134 | vn_iowait(ip); |
40095b64 DC |
1135 | |
1136 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
1da177e4 LT |
1137 | } |
1138 | ||
44866d39 LM |
1139 | if ((flags & SYNC_ATTR) && |
1140 | (ip->i_update_core || | |
1141 | (ip->i_itemp && ip->i_itemp->ili_format.ilf_fields))) { | |
1142 | if (mount_locked) | |
1143 | IPOINTER_INSERT(ip, mp); | |
1da177e4 | 1144 | |
44866d39 LM |
1145 | if (flags & SYNC_WAIT) { |
1146 | xfs_iflock(ip); | |
1147 | error = xfs_iflush(ip, XFS_IFLUSH_SYNC); | |
1da177e4 | 1148 | |
44866d39 LM |
1149 | /* |
1150 | * If we can't acquire the flush lock, then the inode | |
1151 | * is already being flushed so don't bother waiting. | |
1152 | * | |
1153 | * If we can lock it then do a delwri flush so we can | |
1154 | * combine multiple inode flushes in each disk write. | |
1155 | */ | |
1156 | } else if (xfs_iflock_nowait(ip)) { | |
1157 | error = xfs_iflush(ip, XFS_IFLUSH_DELWRI); | |
1158 | } else if (bypassed) { | |
1159 | (*bypassed)++; | |
1da177e4 LT |
1160 | } |
1161 | } | |
1162 | ||
1163 | if (lock_flags != 0) { | |
1164 | xfs_iunlock(ip, lock_flags); | |
1165 | } | |
1166 | ||
1167 | if (vnode_refed) { | |
1168 | /* | |
1169 | * If we had to take a reference on the vnode | |
1170 | * above, then wait until after we've unlocked | |
1171 | * the inode to release the reference. This is | |
1172 | * because we can be already holding the inode | |
1173 | * lock when VN_RELE() calls xfs_inactive(). | |
1174 | * | |
1175 | * Make sure to drop the mount lock before calling | |
1176 | * VN_RELE() so that we don't trip over ourselves if | |
1177 | * we have to go for the mount lock again in the | |
1178 | * inactive code. | |
1179 | */ | |
1180 | if (mount_locked) { | |
1181 | IPOINTER_INSERT(ip, mp); | |
1182 | } | |
1183 | ||
1184 | VN_RELE(vp); | |
1185 | ||
1186 | vnode_refed = B_FALSE; | |
1187 | } | |
1188 | ||
1189 | if (error) { | |
1190 | last_error = error; | |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * bail out if the filesystem is corrupted. | |
1195 | */ | |
1196 | if (error == EFSCORRUPTED) { | |
1197 | if (!mount_locked) { | |
1198 | XFS_MOUNT_ILOCK(mp); | |
1199 | IPOINTER_REMOVE(ip, mp); | |
1200 | } | |
1201 | XFS_MOUNT_IUNLOCK(mp); | |
1202 | ASSERT(ipointer_in == B_FALSE); | |
1203 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1204 | return XFS_ERROR(error); | |
1205 | } | |
1206 | ||
1207 | /* Let other threads have a chance at the mount lock | |
1208 | * if we have looped many times without dropping the | |
1209 | * lock. | |
1210 | */ | |
1211 | if ((++preempt & XFS_PREEMPT_MASK) == 0) { | |
1212 | if (mount_locked) { | |
1213 | IPOINTER_INSERT(ip, mp); | |
1214 | } | |
1215 | } | |
1216 | ||
1217 | if (mount_locked == B_FALSE) { | |
1218 | XFS_MOUNT_ILOCK(mp); | |
1219 | mount_locked = B_TRUE; | |
1220 | IPOINTER_REMOVE(ip, mp); | |
1221 | continue; | |
1222 | } | |
1223 | ||
1224 | ASSERT(ipointer_in == B_FALSE); | |
1225 | ip = ip->i_mnext; | |
1226 | ||
1227 | } while (ip != mp->m_inodes); | |
1228 | ||
1229 | XFS_MOUNT_IUNLOCK(mp); | |
1230 | ||
1231 | ASSERT(ipointer_in == B_FALSE); | |
1232 | ||
1233 | kmem_free(ipointer, sizeof(xfs_iptr_t)); | |
1234 | return XFS_ERROR(last_error); | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * xfs sync routine for internal use | |
1239 | * | |
b83bd138 | 1240 | * This routine supports all of the flags defined for the generic vfs_sync |
1f9b3b64 | 1241 | * interface as explained above under xfs_sync. |
1da177e4 LT |
1242 | * |
1243 | */ | |
1244 | int | |
1245 | xfs_syncsub( | |
1246 | xfs_mount_t *mp, | |
1247 | int flags, | |
1da177e4 LT |
1248 | int *bypassed) |
1249 | { | |
1250 | int error = 0; | |
1251 | int last_error = 0; | |
1252 | uint log_flags = XFS_LOG_FORCE; | |
1253 | xfs_buf_t *bp; | |
1254 | xfs_buf_log_item_t *bip; | |
1255 | ||
1256 | /* | |
1257 | * Sync out the log. This ensures that the log is periodically | |
1258 | * flushed even if there is not enough activity to fill it up. | |
1259 | */ | |
1260 | if (flags & SYNC_WAIT) | |
1261 | log_flags |= XFS_LOG_SYNC; | |
1262 | ||
1263 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1264 | ||
1265 | if (flags & (SYNC_ATTR|SYNC_DELWRI)) { | |
1266 | if (flags & SYNC_BDFLUSH) | |
1267 | xfs_finish_reclaim_all(mp, 1); | |
1268 | else | |
1f9b3b64 | 1269 | error = xfs_sync_inodes(mp, flags, bypassed); |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * Flushing out dirty data above probably generated more | |
1274 | * log activity, so if this isn't vfs_sync() then flush | |
1275 | * the log again. | |
1276 | */ | |
1277 | if (flags & SYNC_DELWRI) { | |
1278 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1279 | } | |
1280 | ||
1281 | if (flags & SYNC_FSDATA) { | |
1282 | /* | |
1283 | * If this is vfs_sync() then only sync the superblock | |
1284 | * if we can lock it without sleeping and it is not pinned. | |
1285 | */ | |
1286 | if (flags & SYNC_BDFLUSH) { | |
1287 | bp = xfs_getsb(mp, XFS_BUF_TRYLOCK); | |
1288 | if (bp != NULL) { | |
1289 | bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*); | |
1290 | if ((bip != NULL) && | |
1291 | xfs_buf_item_dirty(bip)) { | |
1292 | if (!(XFS_BUF_ISPINNED(bp))) { | |
1293 | XFS_BUF_ASYNC(bp); | |
1294 | error = xfs_bwrite(mp, bp); | |
1295 | } else { | |
1296 | xfs_buf_relse(bp); | |
1297 | } | |
1298 | } else { | |
1299 | xfs_buf_relse(bp); | |
1300 | } | |
1301 | } | |
1302 | } else { | |
1303 | bp = xfs_getsb(mp, 0); | |
1304 | /* | |
1305 | * If the buffer is pinned then push on the log so | |
1306 | * we won't get stuck waiting in the write for | |
1307 | * someone, maybe ourselves, to flush the log. | |
1308 | * Even though we just pushed the log above, we | |
1309 | * did not have the superblock buffer locked at | |
1310 | * that point so it can become pinned in between | |
1311 | * there and here. | |
1312 | */ | |
1313 | if (XFS_BUF_ISPINNED(bp)) | |
1314 | xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE); | |
1315 | if (flags & SYNC_WAIT) | |
1316 | XFS_BUF_UNASYNC(bp); | |
1317 | else | |
1318 | XFS_BUF_ASYNC(bp); | |
1319 | error = xfs_bwrite(mp, bp); | |
1320 | } | |
1321 | if (error) { | |
1322 | last_error = error; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * If this is the periodic sync, then kick some entries out of | |
1328 | * the reference cache. This ensures that idle entries are | |
1329 | * eventually kicked out of the cache. | |
1330 | */ | |
1331 | if (flags & SYNC_REFCACHE) { | |
cde410a9 NS |
1332 | if (flags & SYNC_WAIT) |
1333 | xfs_refcache_purge_mp(mp); | |
1334 | else | |
1335 | xfs_refcache_purge_some(mp); | |
1da177e4 LT |
1336 | } |
1337 | ||
92821e2b DC |
1338 | /* |
1339 | * If asked, update the disk superblock with incore counter values if we | |
1340 | * are using non-persistent counters so that they don't get too far out | |
1341 | * of sync if we crash or get a forced shutdown. We don't want to force | |
1342 | * this to disk, just get a transaction into the iclogs.... | |
1343 | */ | |
1344 | if (flags & SYNC_SUPER) | |
1345 | xfs_log_sbcount(mp, 0); | |
1346 | ||
1da177e4 LT |
1347 | /* |
1348 | * Now check to see if the log needs a "dummy" transaction. | |
1349 | */ | |
1350 | ||
1351 | if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) { | |
1352 | xfs_trans_t *tp; | |
1353 | xfs_inode_t *ip; | |
1354 | ||
1355 | /* | |
1356 | * Put a dummy transaction in the log to tell | |
1357 | * recovery that all others are OK. | |
1358 | */ | |
1359 | tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1); | |
1360 | if ((error = xfs_trans_reserve(tp, 0, | |
1361 | XFS_ICHANGE_LOG_RES(mp), | |
1362 | 0, 0, 0))) { | |
1363 | xfs_trans_cancel(tp, 0); | |
1364 | return error; | |
1365 | } | |
1366 | ||
1367 | ip = mp->m_rootip; | |
1368 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1369 | ||
1370 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
1371 | xfs_trans_ihold(tp, ip); | |
1372 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
1c72bf90 | 1373 | error = xfs_trans_commit(tp, 0); |
1da177e4 LT |
1374 | xfs_iunlock(ip, XFS_ILOCK_EXCL); |
1375 | xfs_log_force(mp, (xfs_lsn_t)0, log_flags); | |
1376 | } | |
1377 | ||
1378 | /* | |
1379 | * When shutting down, we need to insure that the AIL is pushed | |
1380 | * to disk or the filesystem can appear corrupt from the PROM. | |
1381 | */ | |
1382 | if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) { | |
1383 | XFS_bflush(mp->m_ddev_targp); | |
1384 | if (mp->m_rtdev_targp) { | |
1385 | XFS_bflush(mp->m_rtdev_targp); | |
1386 | } | |
1387 | } | |
1388 | ||
1389 | return XFS_ERROR(last_error); | |
1390 | } |