?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[XFS] optimize dmapi event tests w/o dmapi config
[linux-2.6/mini2440.git] / fs / xfs / xfs_vfsops.c
blob4a27648b544628b11bd3fb26000097090604d622
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_dir2_sf.h"
35 #include "xfs_attr_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
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"
43 #include "xfs_error.h"
44 #include "xfs_bmap.h"
45 #include "xfs_rw.h"
46 #include "xfs_refcache.h"
47 #include "xfs_buf_item.h"
48 #include "xfs_log_priv.h"
49 #include "xfs_dir2_trace.h"
50 #include "xfs_extfree_item.h"
51 #include "xfs_acl.h"
52 #include "xfs_attr.h"
53 #include "xfs_clnt.h"
54 #include "xfs_mru_cache.h"
55 #include "xfs_filestream.h"
56 #include "xfs_fsops.h"
57
58 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
59
60 int
61 xfs_init(void)
62 {
63 extern kmem_zone_t *xfs_bmap_free_item_zone;
64 extern kmem_zone_t *xfs_btree_cur_zone;
65 extern kmem_zone_t *xfs_trans_zone;
66 extern kmem_zone_t *xfs_buf_item_zone;
67 extern kmem_zone_t *xfs_dabuf_zone;
68 #ifdef XFS_DABUF_DEBUG
69 extern lock_t xfs_dabuf_global_lock;
70 spinlock_init(&xfs_dabuf_global_lock, "xfsda");
71 #endif
72
73 /*
74 * Initialize all of the zone allocators we use.
75 */
76 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
77 "xfs_bmap_free_item");
78 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
79 "xfs_btree_cur");
80 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
81 xfs_da_state_zone =
82 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
83 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
84 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
85 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
86 xfs_mru_cache_init();
87 xfs_filestream_init();
88
89 /*
90 * The size of the zone allocated buf log item is the maximum
91 * size possible under XFS. This wastes a little bit of memory,
92 * but it is much faster.
93 */
94 xfs_buf_item_zone =
95 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
96 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
97 NBWORD) * sizeof(int))),
98 "xfs_buf_item");
99 xfs_efd_zone =
100 kmem_zone_init((sizeof(xfs_efd_log_item_t) +
101 ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
102 sizeof(xfs_extent_t))),
103 "xfs_efd_item");
104 xfs_efi_zone =
105 kmem_zone_init((sizeof(xfs_efi_log_item_t) +
106 ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
107 sizeof(xfs_extent_t))),
108 "xfs_efi_item");
109
110 /*
111 * These zones warrant special memory allocator hints
112 */
113 xfs_inode_zone =
114 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
115 KM_ZONE_HWALIGN | KM_ZONE_RECLAIM |
116 KM_ZONE_SPREAD, NULL);
117 xfs_ili_zone =
118 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
119 KM_ZONE_SPREAD, NULL);
120 xfs_chashlist_zone =
121 kmem_zone_init_flags(sizeof(xfs_chashlist_t), "xfs_chashlist",
122 KM_ZONE_SPREAD, NULL);
123
124 /*
125 * Allocate global trace buffers.
126 */
127 #ifdef XFS_ALLOC_TRACE
128 xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
129 #endif
130 #ifdef XFS_BMAP_TRACE
131 xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
132 #endif
133 #ifdef XFS_BMBT_TRACE
134 xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
135 #endif
136 #ifdef XFS_ATTR_TRACE
137 xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
138 #endif
139 #ifdef XFS_DIR2_TRACE
140 xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
141 #endif
142
143 xfs_dir_startup();
144
145 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
146 xfs_error_test_init();
147 #endif /* DEBUG || INDUCE_IO_ERROR */
148
149 xfs_init_procfs();
150 xfs_sysctl_register();
151 return 0;
152 }
153
154 void
155 xfs_cleanup(void)
156 {
157 extern kmem_zone_t *xfs_bmap_free_item_zone;
158 extern kmem_zone_t *xfs_btree_cur_zone;
159 extern kmem_zone_t *xfs_inode_zone;
160 extern kmem_zone_t *xfs_trans_zone;
161 extern kmem_zone_t *xfs_da_state_zone;
162 extern kmem_zone_t *xfs_dabuf_zone;
163 extern kmem_zone_t *xfs_efd_zone;
164 extern kmem_zone_t *xfs_efi_zone;
165 extern kmem_zone_t *xfs_buf_item_zone;
166 extern kmem_zone_t *xfs_chashlist_zone;
167
168 xfs_cleanup_procfs();
169 xfs_sysctl_unregister();
170 xfs_refcache_destroy();
171 xfs_filestream_uninit();
172 xfs_mru_cache_uninit();
173 xfs_acl_zone_destroy(xfs_acl_zone);
174
175 #ifdef XFS_DIR2_TRACE
176 ktrace_free(xfs_dir2_trace_buf);
177 #endif
178 #ifdef XFS_ATTR_TRACE
179 ktrace_free(xfs_attr_trace_buf);
180 #endif
181 #ifdef XFS_BMBT_TRACE
182 ktrace_free(xfs_bmbt_trace_buf);
183 #endif
184 #ifdef XFS_BMAP_TRACE
185 ktrace_free(xfs_bmap_trace_buf);
186 #endif
187 #ifdef XFS_ALLOC_TRACE
188 ktrace_free(xfs_alloc_trace_buf);
189 #endif
190
191 kmem_zone_destroy(xfs_bmap_free_item_zone);
192 kmem_zone_destroy(xfs_btree_cur_zone);
193 kmem_zone_destroy(xfs_inode_zone);
194 kmem_zone_destroy(xfs_trans_zone);
195 kmem_zone_destroy(xfs_da_state_zone);
196 kmem_zone_destroy(xfs_dabuf_zone);
197 kmem_zone_destroy(xfs_buf_item_zone);
198 kmem_zone_destroy(xfs_efd_zone);
199 kmem_zone_destroy(xfs_efi_zone);
200 kmem_zone_destroy(xfs_ifork_zone);
201 kmem_zone_destroy(xfs_ili_zone);
202 kmem_zone_destroy(xfs_chashlist_zone);
203 }
204
205 /*
206 * xfs_start_flags
207 *
208 * This function fills in xfs_mount_t fields based on mount args.
209 * Note: the superblock has _not_ yet been read in.
210 */
211 STATIC int
212 xfs_start_flags(
213 struct bhv_vfs *vfs,
214 struct xfs_mount_args *ap,
215 struct xfs_mount *mp)
216 {
217 /* Values are in BBs */
218 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
219 /*
220 * At this point the superblock has not been read
221 * in, therefore we do not know the block size.
222 * Before the mount call ends we will convert
223 * these to FSBs.
224 */
225 mp->m_dalign = ap->sunit;
226 mp->m_swidth = ap->swidth;
227 }
228
229 if (ap->logbufs != -1 &&
230 ap->logbufs != 0 &&
231 (ap->logbufs < XLOG_MIN_ICLOGS ||
232 ap->logbufs > XLOG_MAX_ICLOGS)) {
233 cmn_err(CE_WARN,
234 "XFS: invalid logbufs value: %d [not %d-%d]",
235 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
236 return XFS_ERROR(EINVAL);
237 }
238 mp->m_logbufs = ap->logbufs;
239 if (ap->logbufsize != -1 &&
240 ap->logbufsize != 0 &&
241 (ap->logbufsize < XLOG_MIN_RECORD_BSIZE ||
242 ap->logbufsize > XLOG_MAX_RECORD_BSIZE ||
243 !is_power_of_2(ap->logbufsize))) {
244 cmn_err(CE_WARN,
245 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
246 ap->logbufsize);
247 return XFS_ERROR(EINVAL);
248 }
249 mp->m_ihsize = ap->ihashsize;
250 mp->m_logbsize = ap->logbufsize;
251 mp->m_fsname_len = strlen(ap->fsname) + 1;
252 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
253 strcpy(mp->m_fsname, ap->fsname);
254 if (ap->rtname[0]) {
255 mp->m_rtname = kmem_alloc(strlen(ap->rtname) + 1, KM_SLEEP);
256 strcpy(mp->m_rtname, ap->rtname);
257 }
258 if (ap->logname[0]) {
259 mp->m_logname = kmem_alloc(strlen(ap->logname) + 1, KM_SLEEP);
260 strcpy(mp->m_logname, ap->logname);
261 }
262
263 if (ap->flags & XFSMNT_WSYNC)
264 mp->m_flags |= XFS_MOUNT_WSYNC;
265 #if XFS_BIG_INUMS
266 if (ap->flags & XFSMNT_INO64) {
267 mp->m_flags |= XFS_MOUNT_INO64;
268 mp->m_inoadd = XFS_INO64_OFFSET;
269 }
270 #endif
271 if (ap->flags & XFSMNT_RETERR)
272 mp->m_flags |= XFS_MOUNT_RETERR;
273 if (ap->flags & XFSMNT_NOALIGN)
274 mp->m_flags |= XFS_MOUNT_NOALIGN;
275 if (ap->flags & XFSMNT_SWALLOC)
276 mp->m_flags |= XFS_MOUNT_SWALLOC;
277 if (ap->flags & XFSMNT_OSYNCISOSYNC)
278 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
279 if (ap->flags & XFSMNT_32BITINODES)
280 mp->m_flags |= XFS_MOUNT_32BITINODES;
281
282 if (ap->flags & XFSMNT_IOSIZE) {
283 if (ap->iosizelog > XFS_MAX_IO_LOG ||
284 ap->iosizelog < XFS_MIN_IO_LOG) {
285 cmn_err(CE_WARN,
286 "XFS: invalid log iosize: %d [not %d-%d]",
287 ap->iosizelog, XFS_MIN_IO_LOG,
288 XFS_MAX_IO_LOG);
289 return XFS_ERROR(EINVAL);
290 }
291
292 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
293 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
294 }
295
296 if (ap->flags & XFSMNT_IHASHSIZE)
297 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
298 if (ap->flags & XFSMNT_IDELETE)
299 mp->m_flags |= XFS_MOUNT_IDELETE;
300 if (ap->flags & XFSMNT_DIRSYNC)
301 mp->m_flags |= XFS_MOUNT_DIRSYNC;
302 if (ap->flags & XFSMNT_ATTR2)
303 mp->m_flags |= XFS_MOUNT_ATTR2;
304
305 if (ap->flags2 & XFSMNT2_COMPAT_IOSIZE)
306 mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
307
308 /*
309 * no recovery flag requires a read-only mount
310 */
311 if (ap->flags & XFSMNT_NORECOVERY) {
312 if (!(vfs->vfs_flag & VFS_RDONLY)) {
313 cmn_err(CE_WARN,
314 "XFS: tried to mount a FS read-write without recovery!");
315 return XFS_ERROR(EINVAL);
316 }
317 mp->m_flags |= XFS_MOUNT_NORECOVERY;
318 }
319
320 if (ap->flags & XFSMNT_NOUUID)
321 mp->m_flags |= XFS_MOUNT_NOUUID;
322 if (ap->flags & XFSMNT_BARRIER)
323 mp->m_flags |= XFS_MOUNT_BARRIER;
324 else
325 mp->m_flags &= ~XFS_MOUNT_BARRIER;
326
327 if (ap->flags2 & XFSMNT2_FILESTREAMS)
328 mp->m_flags |= XFS_MOUNT_FILESTREAMS;
329
330 return 0;
331 }
332
333 /*
334 * This function fills in xfs_mount_t fields based on mount args.
335 * Note: the superblock _has_ now been read in.
336 */
337 STATIC int
338 xfs_finish_flags(
339 struct bhv_vfs *vfs,
340 struct xfs_mount_args *ap,
341 struct xfs_mount *mp)
342 {
343 int ronly = (vfs->vfs_flag & VFS_RDONLY);
344
345 /* Fail a mount where the logbuf is smaller then the log stripe */
346 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
347 if ((ap->logbufsize <= 0) &&
348 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
349 mp->m_logbsize = mp->m_sb.sb_logsunit;
350 } else if (ap->logbufsize > 0 &&
351 ap->logbufsize < mp->m_sb.sb_logsunit) {
352 cmn_err(CE_WARN,
353 "XFS: logbuf size must be greater than or equal to log stripe size");
354 return XFS_ERROR(EINVAL);
355 }
356 } else {
357 /* Fail a mount if the logbuf is larger than 32K */
358 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
359 cmn_err(CE_WARN,
360 "XFS: logbuf size for version 1 logs must be 16K or 32K");
361 return XFS_ERROR(EINVAL);
362 }
363 }
364
365 if (XFS_SB_VERSION_HASATTR2(&mp->m_sb)) {
366 mp->m_flags |= XFS_MOUNT_ATTR2;
367 }
368
369 /*
370 * prohibit r/w mounts of read-only filesystems
371 */
372 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
373 cmn_err(CE_WARN,
374 "XFS: cannot mount a read-only filesystem as read-write");
375 return XFS_ERROR(EROFS);
376 }
377
378 /*
379 * check for shared mount.
380 */
381 if (ap->flags & XFSMNT_SHARED) {
382 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
383 return XFS_ERROR(EINVAL);
384
385 /*
386 * For IRIX 6.5, shared mounts must have the shared
387 * version bit set, have the persistent readonly
388 * field set, must be version 0 and can only be mounted
389 * read-only.
390 */
391 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
392 (mp->m_sb.sb_shared_vn != 0))
393 return XFS_ERROR(EINVAL);
394
395 mp->m_flags |= XFS_MOUNT_SHARED;
396
397 /*
398 * Shared XFS V0 can't deal with DMI. Return EINVAL.
399 */
400 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
401 return XFS_ERROR(EINVAL);
402 }
403
404 return 0;
405 }
406
407 /*
408 * xfs_mount
409 *
410 * The file system configurations are:
411 * (1) device (partition) with data and internal log
412 * (2) logical volume with data and log subvolumes.
413 * (3) logical volume with data, log, and realtime subvolumes.
414 *
415 * We only have to handle opening the log and realtime volumes here if
416 * they are present. The data subvolume has already been opened by
417 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
418 */
419 STATIC int
420 xfs_mount(
421 struct bhv_desc *bhvp,
422 struct xfs_mount_args *args,
423 cred_t *credp)
424 {
425 struct bhv_vfs *vfsp = bhvtovfs(bhvp);
426 struct bhv_desc *p;
427 struct xfs_mount *mp = XFS_BHVTOM(bhvp);
428 struct block_device *ddev, *logdev, *rtdev;
429 int flags = 0, error;
430
431 ddev = vfsp->vfs_super->s_bdev;
432 logdev = rtdev = NULL;
433
434 /*
435 * Setup xfs_mount function vectors from available behaviors
436 */
437 p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
438 mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
439 p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
440 mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
441 p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
442 mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
443
444 if (args->flags & XFSMNT_QUIET)
445 flags |= XFS_MFSI_QUIET;
446
447 /*
448 * Open real time and log devices - order is important.
449 */
450 if (args->logname[0]) {
451 error = xfs_blkdev_get(mp, args->logname, &logdev);
452 if (error)
453 return error;
454 }
455 if (args->rtname[0]) {
456 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
457 if (error) {
458 xfs_blkdev_put(logdev);
459 return error;
460 }
461
462 if (rtdev == ddev || rtdev == logdev) {
463 cmn_err(CE_WARN,
464 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
465 xfs_blkdev_put(logdev);
466 xfs_blkdev_put(rtdev);
467 return EINVAL;
468 }
469 }
470
471 /*
472 * Setup xfs_mount buffer target pointers
473 */
474 error = ENOMEM;
475 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
476 if (!mp->m_ddev_targp) {
477 xfs_blkdev_put(logdev);
478 xfs_blkdev_put(rtdev);
479 return error;
480 }
481 if (rtdev) {
482 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
483 if (!mp->m_rtdev_targp) {
484 xfs_blkdev_put(logdev);
485 xfs_blkdev_put(rtdev);
486 goto error0;
487 }
488 }
489 mp->m_logdev_targp = (logdev && logdev != ddev) ?
490 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
491 if (!mp->m_logdev_targp) {
492 xfs_blkdev_put(logdev);
493 xfs_blkdev_put(rtdev);
494 goto error0;
495 }
496
497 /*
498 * Setup flags based on mount(2) options and then the superblock
499 */
500 error = xfs_start_flags(vfsp, args, mp);
501 if (error)
502 goto error1;
503 error = xfs_readsb(mp, flags);
504 if (error)
505 goto error1;
506 error = xfs_finish_flags(vfsp, args, mp);
507 if (error)
508 goto error2;
509
510 /*
511 * Setup xfs_mount buffer target pointers based on superblock
512 */
513 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
514 mp->m_sb.sb_sectsize);
515 if (!error && logdev && logdev != ddev) {
516 unsigned int log_sector_size = BBSIZE;
517
518 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
519 log_sector_size = mp->m_sb.sb_logsectsize;
520 error = xfs_setsize_buftarg(mp->m_logdev_targp,
521 mp->m_sb.sb_blocksize,
522 log_sector_size);
523 }
524 if (!error && rtdev)
525 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
526 mp->m_sb.sb_blocksize,
527 mp->m_sb.sb_sectsize);
528 if (error)
529 goto error2;
530
531 if (mp->m_flags & XFS_MOUNT_BARRIER)
532 xfs_mountfs_check_barriers(mp);
533
534 if ((error = xfs_filestream_mount(mp)))
535 goto error2;
536
537 error = XFS_IOINIT(vfsp, args, flags);
538 if (error)
539 goto error2;
540
541 return 0;
542
543 error2:
544 if (mp->m_sb_bp)
545 xfs_freesb(mp);
546 error1:
547 xfs_binval(mp->m_ddev_targp);
548 if (logdev && logdev != ddev)
549 xfs_binval(mp->m_logdev_targp);
550 if (rtdev)
551 xfs_binval(mp->m_rtdev_targp);
552 error0:
553 xfs_unmountfs_close(mp, credp);
554 return error;
555 }
556
557 STATIC int
558 xfs_unmount(
559 bhv_desc_t *bdp,
560 int flags,
561 cred_t *credp)
562 {
563 bhv_vfs_t *vfsp = bhvtovfs(bdp);
564 xfs_mount_t *mp = XFS_BHVTOM(bdp);
565 xfs_inode_t *rip;
566 bhv_vnode_t *rvp;
567 int unmount_event_wanted = 0;
568 int unmount_event_flags = 0;
569 int xfs_unmountfs_needed = 0;
570 int error;
571
572 rip = mp->m_rootip;
573 rvp = XFS_ITOV(rip);
574
575 #ifdef HAVE_DMAPI
576 if (vfsp->vfs_flag & VFS_DMI) {
577 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
578 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
579 NULL, NULL, 0, 0,
580 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
581 0:DM_FLAGS_UNWANTED);
582 if (error)
583 return XFS_ERROR(error);
584 unmount_event_wanted = 1;
585 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
586 0 : DM_FLAGS_UNWANTED;
587 }
588 #endif
589 /*
590 * First blow any referenced inode from this file system
591 * out of the reference cache, and delete the timer.
592 */
593 xfs_refcache_purge_mp(mp);
594
595 /*
596 * Blow away any referenced inode in the filestreams cache.
597 * This can and will cause log traffic as inodes go inactive
598 * here.
599 */
600 xfs_filestream_unmount(mp);
601
602 XFS_bflush(mp->m_ddev_targp);
603 error = xfs_unmount_flush(mp, 0);
604 if (error)
605 goto out;
606
607 ASSERT(vn_count(rvp) == 1);
608
609 /*
610 * Drop the reference count
611 */
612 VN_RELE(rvp);
613
614 /*
615 * If we're forcing a shutdown, typically because of a media error,
616 * we want to make sure we invalidate dirty pages that belong to
617 * referenced vnodes as well.
618 */
619 if (XFS_FORCED_SHUTDOWN(mp)) {
620 error = xfs_sync(&mp->m_bhv,
621 (SYNC_WAIT | SYNC_CLOSE), credp);
622 ASSERT(error != EFSCORRUPTED);
623 }
624 xfs_unmountfs_needed = 1;
625
626 out:
627 /* Send DMAPI event, if required.
628 * Then do xfs_unmountfs() if needed.
629 * Then return error (or zero).
630 */
631 if (unmount_event_wanted) {
632 /* Note: mp structure must still exist for
633 * XFS_SEND_UNMOUNT() call.
634 */
635 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
636 DM_RIGHT_NULL, 0, error, unmount_event_flags);
637 }
638 if (xfs_unmountfs_needed) {
639 /*
640 * Call common unmount function to flush to disk
641 * and free the super block buffer & mount structures.
642 */
643 xfs_unmountfs(mp, credp);
644 }
645
646 return XFS_ERROR(error);
647 }
648
649 STATIC int
650 xfs_quiesce_fs(
651 xfs_mount_t *mp)
652 {
653 int count = 0, pincount;
654
655 xfs_refcache_purge_mp(mp);
656 xfs_flush_buftarg(mp->m_ddev_targp, 0);
657 xfs_finish_reclaim_all(mp, 0);
658
659 /* This loop must run at least twice.
660 * The first instance of the loop will flush
661 * most meta data but that will generate more
662 * meta data (typically directory updates).
663 * Which then must be flushed and logged before
664 * we can write the unmount record.
665 */
666 do {
667 xfs_syncsub(mp, SYNC_INODE_QUIESCE, NULL);
668 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
669 if (!pincount) {
670 delay(50);
671 count++;
672 }
673 } while (count < 2);
674
675 return 0;
676 }
677
678 /*
679 * Second stage of a quiesce. The data is already synced, now we have to take
680 * care of the metadata. New transactions are already blocked, so we need to
681 * wait for any remaining transactions to drain out before proceding.
682 */
683 STATIC void
684 xfs_attr_quiesce(
685 xfs_mount_t *mp)
686 {
687 /* wait for all modifications to complete */
688 while (atomic_read(&mp->m_active_trans) > 0)
689 delay(100);
690
691 /* flush inodes and push all remaining buffers out to disk */
692 xfs_quiesce_fs(mp);
693
694 ASSERT_ALWAYS(atomic_read(&mp->m_active_trans) == 0);
695
696 /* Push the superblock and write an unmount record */
697 xfs_log_sbcount(mp, 1);
698 xfs_log_unmount_write(mp);
699 xfs_unmountfs_writesb(mp);
700 }
701
702 STATIC int
703 xfs_mntupdate(
704 bhv_desc_t *bdp,
705 int *flags,
706 struct xfs_mount_args *args)
707 {
708 bhv_vfs_t *vfsp = bhvtovfs(bdp);
709 xfs_mount_t *mp = XFS_BHVTOM(bdp);
710
711 if (!(*flags & MS_RDONLY)) { /* rw/ro -> rw */
712 if (vfsp->vfs_flag & VFS_RDONLY)
713 vfsp->vfs_flag &= ~VFS_RDONLY;
714 if (args->flags & XFSMNT_BARRIER) {
715 mp->m_flags |= XFS_MOUNT_BARRIER;
716 xfs_mountfs_check_barriers(mp);
717 } else {
718 mp->m_flags &= ~XFS_MOUNT_BARRIER;
719 }
720 } else if (!(vfsp->vfs_flag & VFS_RDONLY)) { /* rw -> ro */
721 xfs_filestream_flush(mp);
722 bhv_vfs_sync(vfsp, SYNC_DATA_QUIESCE, NULL);
723 xfs_attr_quiesce(mp);
724 vfsp->vfs_flag |= VFS_RDONLY;
725 }
726 return 0;
727 }
728
729 /*
730 * xfs_unmount_flush implements a set of flush operation on special
731 * inodes, which are needed as a separate set of operations so that
732 * they can be called as part of relocation process.
733 */
734 int
735 xfs_unmount_flush(
736 xfs_mount_t *mp, /* Mount structure we are getting
737 rid of. */
738 int relocation) /* Called from vfs relocation. */
739 {
740 xfs_inode_t *rip = mp->m_rootip;
741 xfs_inode_t *rbmip;
742 xfs_inode_t *rsumip = NULL;
743 bhv_vnode_t *rvp = XFS_ITOV(rip);
744 int error;
745
746 xfs_ilock(rip, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
747 xfs_iflock(rip);
748
749 /*
750 * Flush out the real time inodes.
751 */
752 if ((rbmip = mp->m_rbmip) != NULL) {
753 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
754 xfs_iflock(rbmip);
755 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
756 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
757
758 if (error == EFSCORRUPTED)
759 goto fscorrupt_out;
760
761 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
762
763 rsumip = mp->m_rsumip;
764 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
765 xfs_iflock(rsumip);
766 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
767 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
768
769 if (error == EFSCORRUPTED)
770 goto fscorrupt_out;
771
772 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
773 }
774
775 /*
776 * Synchronously flush root inode to disk
777 */
778 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
779 if (error == EFSCORRUPTED)
780 goto fscorrupt_out2;
781
782 if (vn_count(rvp) != 1 && !relocation) {
783 xfs_iunlock(rip, XFS_ILOCK_EXCL);
784 return XFS_ERROR(EBUSY);
785 }
786
787 /*
788 * Release dquot that rootinode, rbmino and rsumino might be holding,
789 * flush and purge the quota inodes.
790 */
791 error = XFS_QM_UNMOUNT(mp);
792 if (error == EFSCORRUPTED)
793 goto fscorrupt_out2;
794
795 if (rbmip) {
796 VN_RELE(XFS_ITOV(rbmip));
797 VN_RELE(XFS_ITOV(rsumip));
798 }
799
800 xfs_iunlock(rip, XFS_ILOCK_EXCL);
801 return 0;
802
803 fscorrupt_out:
804 xfs_ifunlock(rip);
805
806 fscorrupt_out2:
807 xfs_iunlock(rip, XFS_ILOCK_EXCL);
808
809 return XFS_ERROR(EFSCORRUPTED);
810 }
811
812 /*
813 * xfs_root extracts the root vnode from a vfs.
814 *
815 * vfsp -- the vfs struct for the desired file system
816 * vpp -- address of the caller's vnode pointer which should be
817 * set to the desired fs root vnode
818 */
819 STATIC int
820 xfs_root(
821 bhv_desc_t *bdp,
822 bhv_vnode_t **vpp)
823 {
824 bhv_vnode_t *vp;
825
826 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
827 VN_HOLD(vp);
828 *vpp = vp;
829 return 0;
830 }
831
832 /*
833 * xfs_statvfs
834 *
835 * Fill in the statvfs structure for the given file system. We use
836 * the superblock lock in the mount structure to ensure a consistent
837 * snapshot of the counters returned.
838 */
839 STATIC int
840 xfs_statvfs(
841 bhv_desc_t *bdp,
842 bhv_statvfs_t *statp,
843 bhv_vnode_t *vp)
844 {
845 __uint64_t fakeinos;
846 xfs_extlen_t lsize;
847 xfs_mount_t *mp;
848 xfs_sb_t *sbp;
849 unsigned long s;
850
851 mp = XFS_BHVTOM(bdp);
852 sbp = &(mp->m_sb);
853
854 statp->f_type = XFS_SB_MAGIC;
855
856 xfs_icsb_sync_counters_flags(mp, XFS_ICSB_LAZY_COUNT);
857 s = XFS_SB_LOCK(mp);
858 statp->f_bsize = sbp->sb_blocksize;
859 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
860 statp->f_blocks = sbp->sb_dblocks - lsize;
861 statp->f_bfree = statp->f_bavail =
862 sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
863 fakeinos = statp->f_bfree << sbp->sb_inopblog;
864 #if XFS_BIG_INUMS
865 fakeinos += mp->m_inoadd;
866 #endif
867 statp->f_files =
868 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
869 if (mp->m_maxicount)
870 #if XFS_BIG_INUMS
871 if (!mp->m_inoadd)
872 #endif
873 statp->f_files = min_t(typeof(statp->f_files),
874 statp->f_files,
875 mp->m_maxicount);
876 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
877 XFS_SB_UNLOCK(mp, s);
878
879 xfs_statvfs_fsid(statp, mp);
880 statp->f_namelen = MAXNAMELEN - 1;
881
882 return 0;
883 }
884
885
886 /*
887 * xfs_sync flushes any pending I/O to file system vfsp.
888 *
889 * This routine is called by vfs_sync() to make sure that things make it
890 * out to disk eventually, on sync() system calls to flush out everything,
891 * and when the file system is unmounted. For the vfs_sync() case, all
892 * we really need to do is sync out the log to make all of our meta-data
893 * updates permanent (except for timestamps). For calls from pflushd(),
894 * dirty pages are kept moving by calling pdflush() on the inodes
895 * containing them. We also flush the inodes that we can lock without
896 * sleeping and the superblock if we can lock it without sleeping from
897 * vfs_sync() so that items at the tail of the log are always moving out.
898 *
899 * Flags:
900 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
901 * to sleep if we can help it. All we really need
902 * to do is ensure that the log is synced at least
903 * periodically. We also push the inodes and
904 * superblock if we can lock them without sleeping
905 * and they are not pinned.
906 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
907 * set, then we really want to lock each inode and flush
908 * it.
909 * SYNC_WAIT - All the flushes that take place in this call should
910 * be synchronous.
911 * SYNC_DELWRI - This tells us to push dirty pages associated with
912 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
913 * determine if they should be flushed sync, async, or
914 * delwri.
915 * SYNC_CLOSE - This flag is passed when the system is being
916 * unmounted. We should sync and invalidate everything.
917 * SYNC_FSDATA - This indicates that the caller would like to make
918 * sure the superblock is safe on disk. We can ensure
919 * this by simply making sure the log gets flushed
920 * if SYNC_BDFLUSH is set, and by actually writing it
921 * out otherwise.
922 * SYNC_IOWAIT - The caller wants us to wait for all data I/O to complete
923 * before we return (including direct I/O). Forms the drain
924 * side of the write barrier needed to safely quiesce the
925 * filesystem.
926 *
927 */
928 /*ARGSUSED*/
929 STATIC int
930 xfs_sync(
931 bhv_desc_t *bdp,
932 int flags,
933 cred_t *credp)
934 {
935 xfs_mount_t *mp = XFS_BHVTOM(bdp);
936
937 if (flags & SYNC_IOWAIT)
938 xfs_filestream_flush(mp);
939
940 return xfs_syncsub(mp, flags, NULL);
941 }
942
943 /*
944 * xfs sync routine for internal use
945 *
946 * This routine supports all of the flags defined for the generic vfs_sync
947 * interface as explained above under xfs_sync.
948 *
949 */
950 int
951 xfs_sync_inodes(
952 xfs_mount_t *mp,
953 int flags,
954 int *bypassed)
955 {
956 xfs_inode_t *ip = NULL;
957 xfs_inode_t *ip_next;
958 xfs_buf_t *bp;
959 bhv_vnode_t *vp = NULL;
960 int error;
961 int last_error;
962 uint64_t fflag;
963 uint lock_flags;
964 uint base_lock_flags;
965 boolean_t mount_locked;
966 boolean_t vnode_refed;
967 int preempt;
968 xfs_dinode_t *dip;
969 xfs_iptr_t *ipointer;
970 #ifdef DEBUG
971 boolean_t ipointer_in = B_FALSE;
972
973 #define IPOINTER_SET ipointer_in = B_TRUE
974 #define IPOINTER_CLR ipointer_in = B_FALSE
975 #else
976 #define IPOINTER_SET
977 #define IPOINTER_CLR
978 #endif
979
980
981 /* Insert a marker record into the inode list after inode ip. The list
982 * must be locked when this is called. After the call the list will no
983 * longer be locked.
984 */
985 #define IPOINTER_INSERT(ip, mp) { \
986 ASSERT(ipointer_in == B_FALSE); \
987 ipointer->ip_mnext = ip->i_mnext; \
988 ipointer->ip_mprev = ip; \
989 ip->i_mnext = (xfs_inode_t *)ipointer; \
990 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
991 preempt = 0; \
992 XFS_MOUNT_IUNLOCK(mp); \
993 mount_locked = B_FALSE; \
994 IPOINTER_SET; \
995 }
996
997 /* Remove the marker from the inode list. If the marker was the only item
998 * in the list then there are no remaining inodes and we should zero out
999 * the whole list. If we are the current head of the list then move the head
1000 * past us.
1001 */
1002 #define IPOINTER_REMOVE(ip, mp) { \
1003 ASSERT(ipointer_in == B_TRUE); \
1004 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
1005 ip = ipointer->ip_mnext; \
1006 ip->i_mprev = ipointer->ip_mprev; \
1007 ipointer->ip_mprev->i_mnext = ip; \
1008 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
1009 mp->m_inodes = ip; \
1010 } \
1011 } else { \
1012 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
1013 mp->m_inodes = NULL; \
1014 ip = NULL; \
1015 } \
1016 IPOINTER_CLR; \
1017 }
1018
1019 #define XFS_PREEMPT_MASK 0x7f
1020
1021 if (bypassed)
1022 *bypassed = 0;
1023 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
1024 return 0;
1025 error = 0;
1026 last_error = 0;
1027 preempt = 0;
1028
1029 /* Allocate a reference marker */
1030 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
1031
1032 fflag = XFS_B_ASYNC; /* default is don't wait */
1033 if (flags & (SYNC_BDFLUSH | SYNC_DELWRI))
1034 fflag = XFS_B_DELWRI;
1035 if (flags & SYNC_WAIT)
1036 fflag = 0; /* synchronous overrides all */
1037
1038 base_lock_flags = XFS_ILOCK_SHARED;
1039 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
1040 /*
1041 * We need the I/O lock if we're going to call any of
1042 * the flush/inval routines.
1043 */
1044 base_lock_flags |= XFS_IOLOCK_SHARED;
1045 }
1046
1047 XFS_MOUNT_ILOCK(mp);
1048
1049 ip = mp->m_inodes;
1050
1051 mount_locked = B_TRUE;
1052 vnode_refed = B_FALSE;
1053
1054 IPOINTER_CLR;
1055
1056 do {
1057 ASSERT(ipointer_in == B_FALSE);
1058 ASSERT(vnode_refed == B_FALSE);
1059
1060 lock_flags = base_lock_flags;
1061
1062 /*
1063 * There were no inodes in the list, just break out
1064 * of the loop.
1065 */
1066 if (ip == NULL) {
1067 break;
1068 }
1069
1070 /*
1071 * We found another sync thread marker - skip it
1072 */
1073 if (ip->i_mount == NULL) {
1074 ip = ip->i_mnext;
1075 continue;
1076 }
1077
1078 vp = XFS_ITOV_NULL(ip);
1079
1080 /*
1081 * If the vnode is gone then this is being torn down,
1082 * call reclaim if it is flushed, else let regular flush
1083 * code deal with it later in the loop.
1084 */
1085
1086 if (vp == NULL) {
1087 /* Skip ones already in reclaim */
1088 if (ip->i_flags & XFS_IRECLAIM) {
1089 ip = ip->i_mnext;
1090 continue;
1091 }
1092 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1093 ip = ip->i_mnext;
1094 } else if ((xfs_ipincount(ip) == 0) &&
1095 xfs_iflock_nowait(ip)) {
1096 IPOINTER_INSERT(ip, mp);
1097
1098 xfs_finish_reclaim(ip, 1,
1099 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1100
1101 XFS_MOUNT_ILOCK(mp);
1102 mount_locked = B_TRUE;
1103 IPOINTER_REMOVE(ip, mp);
1104 } else {
1105 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1106 ip = ip->i_mnext;
1107 }
1108 continue;
1109 }
1110
1111 if (VN_BAD(vp)) {
1112 ip = ip->i_mnext;
1113 continue;
1114 }
1115
1116 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1117 XFS_MOUNT_IUNLOCK(mp);
1118 kmem_free(ipointer, sizeof(xfs_iptr_t));
1119 return 0;
1120 }
1121
1122 /*
1123 * If this is just vfs_sync() or pflushd() calling
1124 * then we can skip inodes for which it looks like
1125 * there is nothing to do. Since we don't have the
1126 * inode locked this is racy, but these are periodic
1127 * calls so it doesn't matter. For the others we want
1128 * to know for sure, so we at least try to lock them.
1129 */
1130 if (flags & SYNC_BDFLUSH) {
1131 if (((ip->i_itemp == NULL) ||
1132 !(ip->i_itemp->ili_format.ilf_fields &
1133 XFS_ILOG_ALL)) &&
1134 (ip->i_update_core == 0)) {
1135 ip = ip->i_mnext;
1136 continue;
1137 }
1138 }
1139
1140 /*
1141 * Try to lock without sleeping. We're out of order with
1142 * the inode list lock here, so if we fail we need to drop
1143 * the mount lock and try again. If we're called from
1144 * bdflush() here, then don't bother.
1145 *
1146 * The inode lock here actually coordinates with the
1147 * almost spurious inode lock in xfs_ireclaim() to prevent
1148 * the vnode we handle here without a reference from
1149 * being freed while we reference it. If we lock the inode
1150 * while it's on the mount list here, then the spurious inode
1151 * lock in xfs_ireclaim() after the inode is pulled from
1152 * the mount list will sleep until we release it here.
1153 * This keeps the vnode from being freed while we reference
1154 * it.
1155 */
1156 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1157 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1158 ip = ip->i_mnext;
1159 continue;
1160 }
1161
1162 vp = vn_grab(vp);
1163 if (vp == NULL) {
1164 ip = ip->i_mnext;
1165 continue;
1166 }
1167
1168 IPOINTER_INSERT(ip, mp);
1169 xfs_ilock(ip, lock_flags);
1170
1171 ASSERT(vp == XFS_ITOV(ip));
1172 ASSERT(ip->i_mount == mp);
1173
1174 vnode_refed = B_TRUE;
1175 }
1176
1177 /* From here on in the loop we may have a marker record
1178 * in the inode list.
1179 */
1180
1181 /*
1182 * If we have to flush data or wait for I/O completion
1183 * we need to drop the ilock that we currently hold.
1184 * If we need to drop the lock, insert a marker if we
1185 * have not already done so.
1186 */
1187 if ((flags & (SYNC_CLOSE|SYNC_IOWAIT)) ||
1188 ((flags & SYNC_DELWRI) && VN_DIRTY(vp))) {
1189 if (mount_locked) {
1190 IPOINTER_INSERT(ip, mp);
1191 }
1192 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1193
1194 if (flags & SYNC_CLOSE) {
1195 /* Shutdown case. Flush and invalidate. */
1196 if (XFS_FORCED_SHUTDOWN(mp))
1197 bhv_vop_toss_pages(vp, 0, -1, FI_REMAPF);
1198 else
1199 error = bhv_vop_flushinval_pages(vp, 0,
1200 -1, FI_REMAPF);
1201 } else if ((flags & SYNC_DELWRI) && VN_DIRTY(vp)) {
1202 error = bhv_vop_flush_pages(vp, (xfs_off_t)0,
1203 -1, fflag, FI_NONE);
1204 }
1205
1206 /*
1207 * When freezing, we need to wait ensure all I/O (including direct
1208 * I/O) is complete to ensure no further data modification can take
1209 * place after this point
1210 */
1211 if (flags & SYNC_IOWAIT)
1212 vn_iowait(vp);
1213
1214 xfs_ilock(ip, XFS_ILOCK_SHARED);
1215 }
1216
1217 if (flags & SYNC_BDFLUSH) {
1218 if ((flags & SYNC_ATTR) &&
1219 ((ip->i_update_core) ||
1220 ((ip->i_itemp != NULL) &&
1221 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1222
1223 /* Insert marker and drop lock if not already
1224 * done.
1225 */
1226 if (mount_locked) {
1227 IPOINTER_INSERT(ip, mp);
1228 }
1229
1230 /*
1231 * We don't want the periodic flushing of the
1232 * inodes by vfs_sync() to interfere with
1233 * I/O to the file, especially read I/O
1234 * where it is only the access time stamp
1235 * that is being flushed out. To prevent
1236 * long periods where we have both inode
1237 * locks held shared here while reading the
1238 * inode's buffer in from disk, we drop the
1239 * inode lock while reading in the inode
1240 * buffer. We have to release the buffer
1241 * and reacquire the inode lock so that they
1242 * are acquired in the proper order (inode
1243 * locks first). The buffer will go at the
1244 * end of the lru chain, though, so we can
1245 * expect it to still be there when we go
1246 * for it again in xfs_iflush().
1247 */
1248 if ((xfs_ipincount(ip) == 0) &&
1249 xfs_iflock_nowait(ip)) {
1250
1251 xfs_ifunlock(ip);
1252 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1253
1254 error = xfs_itobp(mp, NULL, ip,
1255 &dip, &bp, 0, 0);
1256 if (!error) {
1257 xfs_buf_relse(bp);
1258 } else {
1259 /* Bailing out, remove the
1260 * marker and free it.
1261 */
1262 XFS_MOUNT_ILOCK(mp);
1263 IPOINTER_REMOVE(ip, mp);
1264 XFS_MOUNT_IUNLOCK(mp);
1265
1266 ASSERT(!(lock_flags &
1267 XFS_IOLOCK_SHARED));
1268
1269 kmem_free(ipointer,
1270 sizeof(xfs_iptr_t));
1271 return (0);
1272 }
1273
1274 /*
1275 * Since we dropped the inode lock,
1276 * the inode may have been reclaimed.
1277 * Therefore, we reacquire the mount
1278 * lock and check to see if we were the
1279 * inode reclaimed. If this happened
1280 * then the ipointer marker will no
1281 * longer point back at us. In this
1282 * case, move ip along to the inode
1283 * after the marker, remove the marker
1284 * and continue.
1285 */
1286 XFS_MOUNT_ILOCK(mp);
1287 mount_locked = B_TRUE;
1288
1289 if (ip != ipointer->ip_mprev) {
1290 IPOINTER_REMOVE(ip, mp);
1291
1292 ASSERT(!vnode_refed);
1293 ASSERT(!(lock_flags &
1294 XFS_IOLOCK_SHARED));
1295 continue;
1296 }
1297
1298 ASSERT(ip->i_mount == mp);
1299
1300 if (xfs_ilock_nowait(ip,
1301 XFS_ILOCK_SHARED) == 0) {
1302 ASSERT(ip->i_mount == mp);
1303 /*
1304 * We failed to reacquire
1305 * the inode lock without
1306 * sleeping, so just skip
1307 * the inode for now. We
1308 * clear the ILOCK bit from
1309 * the lock_flags so that we
1310 * won't try to drop a lock
1311 * we don't hold below.
1312 */
1313 lock_flags &= ~XFS_ILOCK_SHARED;
1314 IPOINTER_REMOVE(ip_next, mp);
1315 } else if ((xfs_ipincount(ip) == 0) &&
1316 xfs_iflock_nowait(ip)) {
1317 ASSERT(ip->i_mount == mp);
1318 /*
1319 * Since this is vfs_sync()
1320 * calling we only flush the
1321 * inode out if we can lock
1322 * it without sleeping and
1323 * it is not pinned. Drop
1324 * the mount lock here so
1325 * that we don't hold it for
1326 * too long. We already have
1327 * a marker in the list here.
1328 */
1329 XFS_MOUNT_IUNLOCK(mp);
1330 mount_locked = B_FALSE;
1331 error = xfs_iflush(ip,
1332 XFS_IFLUSH_DELWRI);
1333 } else {
1334 ASSERT(ip->i_mount == mp);
1335 IPOINTER_REMOVE(ip_next, mp);
1336 }
1337 }
1338
1339 }
1340
1341 } else {
1342 if ((flags & SYNC_ATTR) &&
1343 ((ip->i_update_core) ||
1344 ((ip->i_itemp != NULL) &&
1345 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1346 if (mount_locked) {
1347 IPOINTER_INSERT(ip, mp);
1348 }
1349
1350 if (flags & SYNC_WAIT) {
1351 xfs_iflock(ip);
1352 error = xfs_iflush(ip,
1353 XFS_IFLUSH_SYNC);
1354 } else {
1355 /*
1356 * If we can't acquire the flush
1357 * lock, then the inode is already
1358 * being flushed so don't bother
1359 * waiting. If we can lock it then
1360 * do a delwri flush so we can
1361 * combine multiple inode flushes
1362 * in each disk write.
1363 */
1364 if (xfs_iflock_nowait(ip)) {
1365 error = xfs_iflush(ip,
1366 XFS_IFLUSH_DELWRI);
1367 }
1368 else if (bypassed)
1369 (*bypassed)++;
1370 }
1371 }
1372 }
1373
1374 if (lock_flags != 0) {
1375 xfs_iunlock(ip, lock_flags);
1376 }
1377
1378 if (vnode_refed) {
1379 /*
1380 * If we had to take a reference on the vnode
1381 * above, then wait until after we've unlocked
1382 * the inode to release the reference. This is
1383 * because we can be already holding the inode
1384 * lock when VN_RELE() calls xfs_inactive().
1385 *
1386 * Make sure to drop the mount lock before calling
1387 * VN_RELE() so that we don't trip over ourselves if
1388 * we have to go for the mount lock again in the
1389 * inactive code.
1390 */
1391 if (mount_locked) {
1392 IPOINTER_INSERT(ip, mp);
1393 }
1394
1395 VN_RELE(vp);
1396
1397 vnode_refed = B_FALSE;
1398 }
1399
1400 if (error) {
1401 last_error = error;
1402 }
1403
1404 /*
1405 * bail out if the filesystem is corrupted.
1406 */
1407 if (error == EFSCORRUPTED) {
1408 if (!mount_locked) {
1409 XFS_MOUNT_ILOCK(mp);
1410 IPOINTER_REMOVE(ip, mp);
1411 }
1412 XFS_MOUNT_IUNLOCK(mp);
1413 ASSERT(ipointer_in == B_FALSE);
1414 kmem_free(ipointer, sizeof(xfs_iptr_t));
1415 return XFS_ERROR(error);
1416 }
1417
1418 /* Let other threads have a chance at the mount lock
1419 * if we have looped many times without dropping the
1420 * lock.
1421 */
1422 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1423 if (mount_locked) {
1424 IPOINTER_INSERT(ip, mp);
1425 }
1426 }
1427
1428 if (mount_locked == B_FALSE) {
1429 XFS_MOUNT_ILOCK(mp);
1430 mount_locked = B_TRUE;
1431 IPOINTER_REMOVE(ip, mp);
1432 continue;
1433 }
1434
1435 ASSERT(ipointer_in == B_FALSE);
1436 ip = ip->i_mnext;
1437
1438 } while (ip != mp->m_inodes);
1439
1440 XFS_MOUNT_IUNLOCK(mp);
1441
1442 ASSERT(ipointer_in == B_FALSE);
1443
1444 kmem_free(ipointer, sizeof(xfs_iptr_t));
1445 return XFS_ERROR(last_error);
1446 }
1447
1448 /*
1449 * xfs sync routine for internal use
1450 *
1451 * This routine supports all of the flags defined for the generic vfs_sync
1452 * interface as explained above under xfs_sync.
1453 *
1454 */
1455 int
1456 xfs_syncsub(
1457 xfs_mount_t *mp,
1458 int flags,
1459 int *bypassed)
1460 {
1461 int error = 0;
1462 int last_error = 0;
1463 uint log_flags = XFS_LOG_FORCE;
1464 xfs_buf_t *bp;
1465 xfs_buf_log_item_t *bip;
1466
1467 /*
1468 * Sync out the log. This ensures that the log is periodically
1469 * flushed even if there is not enough activity to fill it up.
1470 */
1471 if (flags & SYNC_WAIT)
1472 log_flags |= XFS_LOG_SYNC;
1473
1474 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1475
1476 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1477 if (flags & SYNC_BDFLUSH)
1478 xfs_finish_reclaim_all(mp, 1);
1479 else
1480 error = xfs_sync_inodes(mp, flags, bypassed);
1481 }
1482
1483 /*
1484 * Flushing out dirty data above probably generated more
1485 * log activity, so if this isn't vfs_sync() then flush
1486 * the log again.
1487 */
1488 if (flags & SYNC_DELWRI) {
1489 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1490 }
1491
1492 if (flags & SYNC_FSDATA) {
1493 /*
1494 * If this is vfs_sync() then only sync the superblock
1495 * if we can lock it without sleeping and it is not pinned.
1496 */
1497 if (flags & SYNC_BDFLUSH) {
1498 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1499 if (bp != NULL) {
1500 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1501 if ((bip != NULL) &&
1502 xfs_buf_item_dirty(bip)) {
1503 if (!(XFS_BUF_ISPINNED(bp))) {
1504 XFS_BUF_ASYNC(bp);
1505 error = xfs_bwrite(mp, bp);
1506 } else {
1507 xfs_buf_relse(bp);
1508 }
1509 } else {
1510 xfs_buf_relse(bp);
1511 }
1512 }
1513 } else {
1514 bp = xfs_getsb(mp, 0);
1515 /*
1516 * If the buffer is pinned then push on the log so
1517 * we won't get stuck waiting in the write for
1518 * someone, maybe ourselves, to flush the log.
1519 * Even though we just pushed the log above, we
1520 * did not have the superblock buffer locked at
1521 * that point so it can become pinned in between
1522 * there and here.
1523 */
1524 if (XFS_BUF_ISPINNED(bp))
1525 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1526 if (flags & SYNC_WAIT)
1527 XFS_BUF_UNASYNC(bp);
1528 else
1529 XFS_BUF_ASYNC(bp);
1530 error = xfs_bwrite(mp, bp);
1531 }
1532 if (error) {
1533 last_error = error;
1534 }
1535 }
1536
1537 /*
1538 * If this is the periodic sync, then kick some entries out of
1539 * the reference cache. This ensures that idle entries are
1540 * eventually kicked out of the cache.
1541 */
1542 if (flags & SYNC_REFCACHE) {
1543 if (flags & SYNC_WAIT)
1544 xfs_refcache_purge_mp(mp);
1545 else
1546 xfs_refcache_purge_some(mp);
1547 }
1548
1549 /*
1550 * If asked, update the disk superblock with incore counter values if we
1551 * are using non-persistent counters so that they don't get too far out
1552 * of sync if we crash or get a forced shutdown. We don't want to force
1553 * this to disk, just get a transaction into the iclogs....
1554 */
1555 if (flags & SYNC_SUPER)
1556 xfs_log_sbcount(mp, 0);
1557
1558 /*
1559 * Now check to see if the log needs a "dummy" transaction.
1560 */
1561
1562 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1563 xfs_trans_t *tp;
1564 xfs_inode_t *ip;
1565
1566 /*
1567 * Put a dummy transaction in the log to tell
1568 * recovery that all others are OK.
1569 */
1570 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1571 if ((error = xfs_trans_reserve(tp, 0,
1572 XFS_ICHANGE_LOG_RES(mp),
1573 0, 0, 0))) {
1574 xfs_trans_cancel(tp, 0);
1575 return error;
1576 }
1577
1578 ip = mp->m_rootip;
1579 xfs_ilock(ip, XFS_ILOCK_EXCL);
1580
1581 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1582 xfs_trans_ihold(tp, ip);
1583 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1584 error = xfs_trans_commit(tp, 0);
1585 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1586 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1587 }
1588
1589 /*
1590 * When shutting down, we need to insure that the AIL is pushed
1591 * to disk or the filesystem can appear corrupt from the PROM.
1592 */
1593 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1594 XFS_bflush(mp->m_ddev_targp);
1595 if (mp->m_rtdev_targp) {
1596 XFS_bflush(mp->m_rtdev_targp);
1597 }
1598 }
1599
1600 return XFS_ERROR(last_error);
1601 }
1602
1603 /*
1604 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1605 */
1606 STATIC int
1607 xfs_vget(
1608 bhv_desc_t *bdp,
1609 bhv_vnode_t **vpp,
1610 fid_t *fidp)
1611 {
1612 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1613 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1614 xfs_inode_t *ip;
1615 int error;
1616 xfs_ino_t ino;
1617 unsigned int igen;
1618
1619 /*
1620 * Invalid. Since handles can be created in user space and passed in
1621 * via gethandle(), this is not cause for a panic.
1622 */
1623 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1624 return XFS_ERROR(EINVAL);
1625
1626 ino = xfid->xfs_fid_ino;
1627 igen = xfid->xfs_fid_gen;
1628
1629 /*
1630 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1631 */
1632 if (ino == 0)
1633 return XFS_ERROR(ESTALE);
1634
1635 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1636 if (error) {
1637 *vpp = NULL;
1638 return error;
1639 }
1640
1641 if (ip == NULL) {
1642 *vpp = NULL;
1643 return XFS_ERROR(EIO);
1644 }
1645
1646 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1647 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1648 *vpp = NULL;
1649 return XFS_ERROR(ENOENT);
1650 }
1651
1652 *vpp = XFS_ITOV(ip);
1653 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1654 return 0;
1655 }
1656
1657
1658 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1659 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1660 #define MNTOPT_LOGDEV "logdev" /* log device */
1661 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1662 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1663 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1664 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1665 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1666 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1667 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1668 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1669 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1670 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1671 #define MNTOPT_GRPID "grpid" /* group-ID from parent directory */
1672 #define MNTOPT_NOGRPID "nogrpid" /* group-ID from current process */
1673 #define MNTOPT_BSDGROUPS "bsdgroups" /* group-ID from parent directory */
1674 #define MNTOPT_SYSVGROUPS "sysvgroups" /* group-ID from current process */
1675 #define MNTOPT_ALLOCSIZE "allocsize" /* preferred allocation size */
1676 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1677 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1678 #define MNTOPT_BARRIER "barrier" /* use writer barriers for log write and
1679 * unwritten extent conversion */
1680 #define MNTOPT_NOBARRIER "nobarrier" /* .. disable */
1681 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1682 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1683 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1684 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1685 #define MNTOPT_LARGEIO "largeio" /* report large I/O sizes in stat() */
1686 #define MNTOPT_NOLARGEIO "nolargeio" /* do not report large I/O sizes
1687 * in stat(). */
1688 #define MNTOPT_ATTR2 "attr2" /* do use attr2 attribute format */
1689 #define MNTOPT_NOATTR2 "noattr2" /* do not use attr2 attribute format */
1690 #define MNTOPT_FILESTREAM "filestreams" /* use filestreams allocator */
1691
1692 STATIC unsigned long
1693 suffix_strtoul(char *s, char **endp, unsigned int base)
1694 {
1695 int last, shift_left_factor = 0;
1696 char *value = s;
1697
1698 last = strlen(value) - 1;
1699 if (value[last] == 'K' || value[last] == 'k') {
1700 shift_left_factor = 10;
1701 value[last] = '\0';
1702 }
1703 if (value[last] == 'M' || value[last] == 'm') {
1704 shift_left_factor = 20;
1705 value[last] = '\0';
1706 }
1707 if (value[last] == 'G' || value[last] == 'g') {
1708 shift_left_factor = 30;
1709 value[last] = '\0';
1710 }
1711
1712 return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
1713 }
1714
1715 STATIC int
1716 xfs_parseargs(
1717 struct bhv_desc *bhv,
1718 char *options,
1719 struct xfs_mount_args *args,
1720 int update)
1721 {
1722 bhv_vfs_t *vfsp = bhvtovfs(bhv);
1723 char *this_char, *value, *eov;
1724 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1725 int iosize;
1726
1727 args->flags |= XFSMNT_IDELETE;
1728 args->flags |= XFSMNT_BARRIER;
1729 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1730
1731 if (!options)
1732 goto done;
1733
1734 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1735
1736 while ((this_char = strsep(&options, ",")) != NULL) {
1737 if (!*this_char)
1738 continue;
1739 if ((value = strchr(this_char, '=')) != NULL)
1740 *value++ = 0;
1741
1742 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1743 if (!value || !*value) {
1744 cmn_err(CE_WARN,
1745 "XFS: %s option requires an argument",
1746 this_char);
1747 return EINVAL;
1748 }
1749 args->logbufs = simple_strtoul(value, &eov, 10);
1750 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1751 if (!value || !*value) {
1752 cmn_err(CE_WARN,
1753 "XFS: %s option requires an argument",
1754 this_char);
1755 return EINVAL;
1756 }
1757 args->logbufsize = suffix_strtoul(value, &eov, 10);
1758 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1759 if (!value || !*value) {
1760 cmn_err(CE_WARN,
1761 "XFS: %s option requires an argument",
1762 this_char);
1763 return EINVAL;
1764 }
1765 strncpy(args->logname, value, MAXNAMELEN);
1766 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1767 if (!value || !*value) {
1768 cmn_err(CE_WARN,
1769 "XFS: %s option requires an argument",
1770 this_char);
1771 return EINVAL;
1772 }
1773 strncpy(args->mtpt, value, MAXNAMELEN);
1774 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1775 if (!value || !*value) {
1776 cmn_err(CE_WARN,
1777 "XFS: %s option requires an argument",
1778 this_char);
1779 return EINVAL;
1780 }
1781 strncpy(args->rtname, value, MAXNAMELEN);
1782 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1783 if (!value || !*value) {
1784 cmn_err(CE_WARN,
1785 "XFS: %s option requires an argument",
1786 this_char);
1787 return EINVAL;
1788 }
1789 iosize = simple_strtoul(value, &eov, 10);
1790 args->flags |= XFSMNT_IOSIZE;
1791 args->iosizelog = (uint8_t) iosize;
1792 } else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
1793 if (!value || !*value) {
1794 cmn_err(CE_WARN,
1795 "XFS: %s option requires an argument",
1796 this_char);
1797 return EINVAL;
1798 }
1799 iosize = suffix_strtoul(value, &eov, 10);
1800 args->flags |= XFSMNT_IOSIZE;
1801 args->iosizelog = ffs(iosize) - 1;
1802 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1803 if (!value || !*value) {
1804 cmn_err(CE_WARN,
1805 "XFS: %s option requires an argument",
1806 this_char);
1807 return EINVAL;
1808 }
1809 args->flags |= XFSMNT_IHASHSIZE;
1810 args->ihashsize = simple_strtoul(value, &eov, 10);
1811 } else if (!strcmp(this_char, MNTOPT_GRPID) ||
1812 !strcmp(this_char, MNTOPT_BSDGROUPS)) {
1813 vfsp->vfs_flag |= VFS_GRPID;
1814 } else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
1815 !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
1816 vfsp->vfs_flag &= ~VFS_GRPID;
1817 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1818 args->flags |= XFSMNT_WSYNC;
1819 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1820 args->flags |= XFSMNT_OSYNCISOSYNC;
1821 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1822 args->flags |= XFSMNT_NORECOVERY;
1823 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1824 args->flags |= XFSMNT_INO64;
1825 #if !XFS_BIG_INUMS
1826 cmn_err(CE_WARN,
1827 "XFS: %s option not allowed on this system",
1828 this_char);
1829 return EINVAL;
1830 #endif
1831 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1832 args->flags |= XFSMNT_NOALIGN;
1833 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1834 args->flags |= XFSMNT_SWALLOC;
1835 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1836 if (!value || !*value) {
1837 cmn_err(CE_WARN,
1838 "XFS: %s option requires an argument",
1839 this_char);
1840 return EINVAL;
1841 }
1842 dsunit = simple_strtoul(value, &eov, 10);
1843 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1844 if (!value || !*value) {
1845 cmn_err(CE_WARN,
1846 "XFS: %s option requires an argument",
1847 this_char);
1848 return EINVAL;
1849 }
1850 dswidth = simple_strtoul(value, &eov, 10);
1851 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1852 args->flags &= ~XFSMNT_32BITINODES;
1853 #if !XFS_BIG_INUMS
1854 cmn_err(CE_WARN,
1855 "XFS: %s option not allowed on this system",
1856 this_char);
1857 return EINVAL;
1858 #endif
1859 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1860 args->flags |= XFSMNT_NOUUID;
1861 } else if (!strcmp(this_char, MNTOPT_BARRIER)) {
1862 args->flags |= XFSMNT_BARRIER;
1863 } else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
1864 args->flags &= ~XFSMNT_BARRIER;
1865 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1866 args->flags &= ~XFSMNT_IDELETE;
1867 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1868 args->flags |= XFSMNT_IDELETE;
1869 } else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
1870 args->flags2 &= ~XFSMNT2_COMPAT_IOSIZE;
1871 } else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
1872 args->flags2 |= XFSMNT2_COMPAT_IOSIZE;
1873 } else if (!strcmp(this_char, MNTOPT_ATTR2)) {
1874 args->flags |= XFSMNT_ATTR2;
1875 } else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
1876 args->flags &= ~XFSMNT_ATTR2;
1877 } else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
1878 args->flags2 |= XFSMNT2_FILESTREAMS;
1879 } else if (!strcmp(this_char, "osyncisdsync")) {
1880 /* no-op, this is now the default */
1881 cmn_err(CE_WARN,
1882 "XFS: osyncisdsync is now the default, option is deprecated.");
1883 } else if (!strcmp(this_char, "irixsgid")) {
1884 cmn_err(CE_WARN,
1885 "XFS: irixsgid is now a sysctl(2) variable, option is deprecated.");
1886 } else {
1887 cmn_err(CE_WARN,
1888 "XFS: unknown mount option [%s].", this_char);
1889 return EINVAL;
1890 }
1891 }
1892
1893 if (args->flags & XFSMNT_NORECOVERY) {
1894 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1895 cmn_err(CE_WARN,
1896 "XFS: no-recovery mounts must be read-only.");
1897 return EINVAL;
1898 }
1899 }
1900
1901 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1902 cmn_err(CE_WARN,
1903 "XFS: sunit and swidth options incompatible with the noalign option");
1904 return EINVAL;
1905 }
1906
1907 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1908 cmn_err(CE_WARN,
1909 "XFS: sunit and swidth must be specified together");
1910 return EINVAL;
1911 }
1912
1913 if (dsunit && (dswidth % dsunit != 0)) {
1914 cmn_err(CE_WARN,
1915 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)",
1916 dswidth, dsunit);
1917 return EINVAL;
1918 }
1919
1920 if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1921 if (dsunit) {
1922 args->sunit = dsunit;
1923 args->flags |= XFSMNT_RETERR;
1924 } else {
1925 args->sunit = vol_dsunit;
1926 }
1927 dswidth ? (args->swidth = dswidth) :
1928 (args->swidth = vol_dswidth);
1929 } else {
1930 args->sunit = args->swidth = 0;
1931 }
1932
1933 done:
1934 if (args->flags & XFSMNT_32BITINODES)
1935 vfsp->vfs_flag |= VFS_32BITINODES;
1936 if (args->flags2)
1937 args->flags |= XFSMNT_FLAGS2;
1938 return 0;
1939 }
1940
1941 STATIC int
1942 xfs_showargs(
1943 struct bhv_desc *bhv,
1944 struct seq_file *m)
1945 {
1946 static struct proc_xfs_info {
1947 int flag;
1948 char *str;
1949 } xfs_info[] = {
1950 /* the few simple ones we can get from the mount struct */
1951 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
1952 { XFS_MOUNT_INO64, "," MNTOPT_INO64 },
1953 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
1954 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
1955 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
1956 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
1957 { XFS_MOUNT_OSYNCISOSYNC, "," MNTOPT_OSYNCISOSYNC },
1958 { 0, NULL }
1959 };
1960 struct proc_xfs_info *xfs_infop;
1961 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1962 struct bhv_vfs *vfsp = XFS_MTOVFS(mp);
1963
1964 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1965 if (mp->m_flags & xfs_infop->flag)
1966 seq_puts(m, xfs_infop->str);
1967 }
1968
1969 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1970 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", (int)mp->m_ihsize);
1971
1972 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1973 seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
1974 (int)(1 << mp->m_writeio_log) >> 10);
1975
1976 if (mp->m_logbufs > 0)
1977 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1978 if (mp->m_logbsize > 0)
1979 seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
1980
1981 if (mp->m_logname)
1982 seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
1983 if (mp->m_rtname)
1984 seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
1985
1986 if (mp->m_dalign > 0)
1987 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1988 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1989 if (mp->m_swidth > 0)
1990 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1991 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1992
1993 if (!(mp->m_flags & XFS_MOUNT_IDELETE))
1994 seq_printf(m, "," MNTOPT_IKEEP);
1995 if (!(mp->m_flags & XFS_MOUNT_COMPAT_IOSIZE))
1996 seq_printf(m, "," MNTOPT_LARGEIO);
1997
1998 if (!(vfsp->vfs_flag & VFS_32BITINODES))
1999 seq_printf(m, "," MNTOPT_64BITINODE);
2000 if (vfsp->vfs_flag & VFS_GRPID)
2001 seq_printf(m, "," MNTOPT_GRPID);
2002
2003 return 0;
2004 }
2005
2006 /*
2007 * Second stage of a freeze. The data is already frozen so we only
2008 * need to take care of themetadata. Once that's done write a dummy
2009 * record to dirty the log in case of a crash while frozen.
2010 */
2011 STATIC void
2012 xfs_freeze(
2013 bhv_desc_t *bdp)
2014 {
2015 xfs_mount_t *mp = XFS_BHVTOM(bdp);
2016
2017 xfs_attr_quiesce(mp);
2018 xfs_fs_log_dummy(mp);
2019 }
2020
2021
2022 bhv_vfsops_t xfs_vfsops = {
2023 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
2024 .vfs_parseargs = xfs_parseargs,
2025 .vfs_showargs = xfs_showargs,
2026 .vfs_mount = xfs_mount,
2027 .vfs_unmount = xfs_unmount,
2028 .vfs_mntupdate = xfs_mntupdate,
2029 .vfs_root = xfs_root,
2030 .vfs_statvfs = xfs_statvfs,
2031 .vfs_sync = xfs_sync,
2032 .vfs_vget = xfs_vget,
2033 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
2034 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
2035 .vfs_init_vnode = xfs_initialize_vnode,
2036 .vfs_force_shutdown = xfs_do_force_shutdown,
2037 .vfs_freeze = xfs_freeze,
2038 };
Support for the Chinese Samsung S3C2440 based development boards