Linux-2.6.12-rc2
[linux-2.6/next.git] / fs / xfs / xfs_vfsops.c
blob00aae9c6a904e09865cd727aad754c8f83ba43ef
1 /*
2 * XFS filesystem operations.
4 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of version 2 of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
14 * Further, this software is distributed without any warranty that it is
15 * free of the rightful claim of any third person regarding infringement
16 * or the like. Any license provided herein, whether implied or
17 * otherwise, applies only to this software file. Patent licenses, if
18 * any, provided herein do not apply to combinations of this program with
19 * other software, or any other product whatsoever.
21 * You should have received a copy of the GNU General Public License along
22 * with this program; if not, write the Free Software Foundation, Inc., 59
23 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
25 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
26 * Mountain View, CA 94043, or:
28 * http://www.sgi.com
30 * For further information regarding this notice, see:
32 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
35 #include "xfs.h"
36 #include "xfs_macros.h"
37 #include "xfs_types.h"
38 #include "xfs_inum.h"
39 #include "xfs_log.h"
40 #include "xfs_trans.h"
41 #include "xfs_sb.h"
42 #include "xfs_dir.h"
43 #include "xfs_dir2.h"
44 #include "xfs_dmapi.h"
45 #include "xfs_mount.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_alloc_btree.h"
49 #include "xfs_btree.h"
50 #include "xfs_alloc.h"
51 #include "xfs_ialloc.h"
52 #include "xfs_attr_sf.h"
53 #include "xfs_dir_sf.h"
54 #include "xfs_dir2_sf.h"
55 #include "xfs_dinode.h"
56 #include "xfs_inode_item.h"
57 #include "xfs_inode.h"
58 #include "xfs_ag.h"
59 #include "xfs_error.h"
60 #include "xfs_bmap.h"
61 #include "xfs_da_btree.h"
62 #include "xfs_rw.h"
63 #include "xfs_refcache.h"
64 #include "xfs_buf_item.h"
65 #include "xfs_extfree_item.h"
66 #include "xfs_quota.h"
67 #include "xfs_dir2_trace.h"
68 #include "xfs_acl.h"
69 #include "xfs_attr.h"
70 #include "xfs_clnt.h"
71 #include "xfs_log_priv.h"
73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
75 int
76 xfs_init(void)
78 extern kmem_zone_t *xfs_bmap_free_item_zone;
79 extern kmem_zone_t *xfs_btree_cur_zone;
80 extern kmem_zone_t *xfs_trans_zone;
81 extern kmem_zone_t *xfs_buf_item_zone;
82 extern kmem_zone_t *xfs_dabuf_zone;
83 #ifdef XFS_DABUF_DEBUG
84 extern lock_t xfs_dabuf_global_lock;
85 spinlock_init(&xfs_dabuf_global_lock, "xfsda");
86 #endif
89 * Initialize all of the zone allocators we use.
91 xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
92 "xfs_bmap_free_item");
93 xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
94 "xfs_btree_cur");
95 xfs_inode_zone = kmem_zone_init(sizeof(xfs_inode_t), "xfs_inode");
96 xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
97 xfs_da_state_zone =
98 kmem_zone_init(sizeof(xfs_da_state_t), "xfs_da_state");
99 xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
102 * The size of the zone allocated buf log item is the maximum
103 * size possible under XFS. This wastes a little bit of memory,
104 * but it is much faster.
106 xfs_buf_item_zone =
107 kmem_zone_init((sizeof(xfs_buf_log_item_t) +
108 (((XFS_MAX_BLOCKSIZE / XFS_BLI_CHUNK) /
109 NBWORD) * sizeof(int))),
110 "xfs_buf_item");
111 xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
112 ((XFS_EFD_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
113 "xfs_efd_item");
114 xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
115 ((XFS_EFI_MAX_FAST_EXTENTS - 1) * sizeof(xfs_extent_t))),
116 "xfs_efi_item");
117 xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
118 xfs_ili_zone = kmem_zone_init(sizeof(xfs_inode_log_item_t), "xfs_ili");
119 xfs_chashlist_zone = kmem_zone_init(sizeof(xfs_chashlist_t),
120 "xfs_chashlist");
121 xfs_acl_zone_init(xfs_acl_zone, "xfs_acl");
124 * Allocate global trace buffers.
126 #ifdef XFS_ALLOC_TRACE
127 xfs_alloc_trace_buf = ktrace_alloc(XFS_ALLOC_TRACE_SIZE, KM_SLEEP);
128 #endif
129 #ifdef XFS_BMAP_TRACE
130 xfs_bmap_trace_buf = ktrace_alloc(XFS_BMAP_TRACE_SIZE, KM_SLEEP);
131 #endif
132 #ifdef XFS_BMBT_TRACE
133 xfs_bmbt_trace_buf = ktrace_alloc(XFS_BMBT_TRACE_SIZE, KM_SLEEP);
134 #endif
135 #ifdef XFS_DIR_TRACE
136 xfs_dir_trace_buf = ktrace_alloc(XFS_DIR_TRACE_SIZE, KM_SLEEP);
137 #endif
138 #ifdef XFS_ATTR_TRACE
139 xfs_attr_trace_buf = ktrace_alloc(XFS_ATTR_TRACE_SIZE, KM_SLEEP);
140 #endif
141 #ifdef XFS_DIR2_TRACE
142 xfs_dir2_trace_buf = ktrace_alloc(XFS_DIR2_GTRACE_SIZE, KM_SLEEP);
143 #endif
145 xfs_dir_startup();
147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
148 xfs_error_test_init();
149 #endif /* DEBUG || INDUCE_IO_ERROR */
151 xfs_init_procfs();
152 xfs_sysctl_register();
153 return 0;
156 void
157 xfs_cleanup(void)
159 extern kmem_zone_t *xfs_bmap_free_item_zone;
160 extern kmem_zone_t *xfs_btree_cur_zone;
161 extern kmem_zone_t *xfs_inode_zone;
162 extern kmem_zone_t *xfs_trans_zone;
163 extern kmem_zone_t *xfs_da_state_zone;
164 extern kmem_zone_t *xfs_dabuf_zone;
165 extern kmem_zone_t *xfs_efd_zone;
166 extern kmem_zone_t *xfs_efi_zone;
167 extern kmem_zone_t *xfs_buf_item_zone;
168 extern kmem_zone_t *xfs_chashlist_zone;
170 xfs_cleanup_procfs();
171 xfs_sysctl_unregister();
172 xfs_refcache_destroy();
173 xfs_acl_zone_destroy(xfs_acl_zone);
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_DIR_TRACE
182 ktrace_free(xfs_dir_trace_buf);
183 #endif
184 #ifdef XFS_BMBT_TRACE
185 ktrace_free(xfs_bmbt_trace_buf);
186 #endif
187 #ifdef XFS_BMAP_TRACE
188 ktrace_free(xfs_bmap_trace_buf);
189 #endif
190 #ifdef XFS_ALLOC_TRACE
191 ktrace_free(xfs_alloc_trace_buf);
192 #endif
194 kmem_cache_destroy(xfs_bmap_free_item_zone);
195 kmem_cache_destroy(xfs_btree_cur_zone);
196 kmem_cache_destroy(xfs_inode_zone);
197 kmem_cache_destroy(xfs_trans_zone);
198 kmem_cache_destroy(xfs_da_state_zone);
199 kmem_cache_destroy(xfs_dabuf_zone);
200 kmem_cache_destroy(xfs_buf_item_zone);
201 kmem_cache_destroy(xfs_efd_zone);
202 kmem_cache_destroy(xfs_efi_zone);
203 kmem_cache_destroy(xfs_ifork_zone);
204 kmem_cache_destroy(xfs_ili_zone);
205 kmem_cache_destroy(xfs_chashlist_zone);
209 * xfs_start_flags
211 * This function fills in xfs_mount_t fields based on mount args.
212 * Note: the superblock has _not_ yet been read in.
214 STATIC int
215 xfs_start_flags(
216 struct vfs *vfs,
217 struct xfs_mount_args *ap,
218 struct xfs_mount *mp)
220 /* Values are in BBs */
221 if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
223 * At this point the superblock has not been read
224 * in, therefore we do not know the block size.
225 * Before the mount call ends we will convert
226 * these to FSBs.
228 mp->m_dalign = ap->sunit;
229 mp->m_swidth = ap->swidth;
232 if (ap->logbufs != -1 &&
233 #if defined(DEBUG) || defined(XLOG_NOLOG)
234 ap->logbufs != 0 &&
235 #endif
236 (ap->logbufs < XLOG_MIN_ICLOGS ||
237 ap->logbufs > XLOG_MAX_ICLOGS)) {
238 cmn_err(CE_WARN,
239 "XFS: invalid logbufs value: %d [not %d-%d]",
240 ap->logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
241 return XFS_ERROR(EINVAL);
243 mp->m_logbufs = ap->logbufs;
244 if (ap->logbufsize != -1 &&
245 ap->logbufsize != 16 * 1024 &&
246 ap->logbufsize != 32 * 1024 &&
247 ap->logbufsize != 64 * 1024 &&
248 ap->logbufsize != 128 * 1024 &&
249 ap->logbufsize != 256 * 1024) {
250 cmn_err(CE_WARN,
251 "XFS: invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
252 ap->logbufsize);
253 return XFS_ERROR(EINVAL);
255 mp->m_ihsize = ap->ihashsize;
256 mp->m_logbsize = ap->logbufsize;
257 mp->m_fsname_len = strlen(ap->fsname) + 1;
258 mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
259 strcpy(mp->m_fsname, ap->fsname);
261 if (ap->flags & XFSMNT_WSYNC)
262 mp->m_flags |= XFS_MOUNT_WSYNC;
263 #if XFS_BIG_INUMS
264 if (ap->flags & XFSMNT_INO64) {
265 mp->m_flags |= XFS_MOUNT_INO64;
266 mp->m_inoadd = XFS_INO64_OFFSET;
268 #endif
269 if (ap->flags & XFSMNT_NOATIME)
270 mp->m_flags |= XFS_MOUNT_NOATIME;
272 if (ap->flags & XFSMNT_RETERR)
273 mp->m_flags |= XFS_MOUNT_RETERR;
275 if (ap->flags & XFSMNT_NOALIGN)
276 mp->m_flags |= XFS_MOUNT_NOALIGN;
278 if (ap->flags & XFSMNT_SWALLOC)
279 mp->m_flags |= XFS_MOUNT_SWALLOC;
281 if (ap->flags & XFSMNT_OSYNCISOSYNC)
282 mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
284 if (ap->flags & XFSMNT_32BITINODES)
285 mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
287 if (ap->flags & XFSMNT_IOSIZE) {
288 if (ap->iosizelog > XFS_MAX_IO_LOG ||
289 ap->iosizelog < XFS_MIN_IO_LOG) {
290 cmn_err(CE_WARN,
291 "XFS: invalid log iosize: %d [not %d-%d]",
292 ap->iosizelog, XFS_MIN_IO_LOG,
293 XFS_MAX_IO_LOG);
294 return XFS_ERROR(EINVAL);
297 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
298 mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
301 if (ap->flags & XFSMNT_IHASHSIZE)
302 mp->m_flags |= XFS_MOUNT_IHASHSIZE;
304 if (ap->flags & XFSMNT_IDELETE)
305 mp->m_flags |= XFS_MOUNT_IDELETE;
307 if (ap->flags & XFSMNT_DIRSYNC)
308 mp->m_flags |= XFS_MOUNT_DIRSYNC;
311 * no recovery flag requires a read-only mount
313 if (ap->flags & XFSMNT_NORECOVERY) {
314 if (!(vfs->vfs_flag & VFS_RDONLY)) {
315 cmn_err(CE_WARN,
316 "XFS: tried to mount a FS read-write without recovery!");
317 return XFS_ERROR(EINVAL);
319 mp->m_flags |= XFS_MOUNT_NORECOVERY;
322 if (ap->flags & XFSMNT_NOUUID)
323 mp->m_flags |= XFS_MOUNT_NOUUID;
324 if (ap->flags & XFSMNT_NOLOGFLUSH)
325 mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
327 return 0;
331 * This function fills in xfs_mount_t fields based on mount args.
332 * Note: the superblock _has_ now been read in.
334 STATIC int
335 xfs_finish_flags(
336 struct vfs *vfs,
337 struct xfs_mount_args *ap,
338 struct xfs_mount *mp)
340 int ronly = (vfs->vfs_flag & VFS_RDONLY);
342 /* Fail a mount where the logbuf is smaller then the log stripe */
343 if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
344 if ((ap->logbufsize == -1) &&
345 (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
346 mp->m_logbsize = mp->m_sb.sb_logsunit;
347 } else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
348 cmn_err(CE_WARN,
349 "XFS: logbuf size must be greater than or equal to log stripe size");
350 return XFS_ERROR(EINVAL);
352 } else {
353 /* Fail a mount if the logbuf is larger than 32K */
354 if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
355 cmn_err(CE_WARN,
356 "XFS: logbuf size for version 1 logs must be 16K or 32K");
357 return XFS_ERROR(EINVAL);
362 * prohibit r/w mounts of read-only filesystems
364 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
365 cmn_err(CE_WARN,
366 "XFS: cannot mount a read-only filesystem as read-write");
367 return XFS_ERROR(EROFS);
371 * disallow mount attempts with (IRIX) project quota enabled
373 if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
374 (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
375 cmn_err(CE_WARN,
376 "XFS: cannot mount a filesystem with IRIX project quota enabled");
377 return XFS_ERROR(ENOSYS);
381 * check for shared mount.
383 if (ap->flags & XFSMNT_SHARED) {
384 if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
385 return XFS_ERROR(EINVAL);
388 * For IRIX 6.5, shared mounts must have the shared
389 * version bit set, have the persistent readonly
390 * field set, must be version 0 and can only be mounted
391 * read-only.
393 if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
394 (mp->m_sb.sb_shared_vn != 0))
395 return XFS_ERROR(EINVAL);
397 mp->m_flags |= XFS_MOUNT_SHARED;
400 * Shared XFS V0 can't deal with DMI. Return EINVAL.
402 if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
403 return XFS_ERROR(EINVAL);
406 return 0;
410 * xfs_mount
412 * The file system configurations are:
413 * (1) device (partition) with data and internal log
414 * (2) logical volume with data and log subvolumes.
415 * (3) logical volume with data, log, and realtime subvolumes.
417 * We only have to handle opening the log and realtime volumes here if
418 * they are present. The data subvolume has already been opened by
419 * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
421 STATIC int
422 xfs_mount(
423 struct bhv_desc *bhvp,
424 struct xfs_mount_args *args,
425 cred_t *credp)
427 struct vfs *vfsp = bhvtovfs(bhvp);
428 struct bhv_desc *p;
429 struct xfs_mount *mp = XFS_BHVTOM(bhvp);
430 struct block_device *ddev, *logdev, *rtdev;
431 int flags = 0, error;
433 ddev = vfsp->vfs_super->s_bdev;
434 logdev = rtdev = NULL;
437 * Setup xfs_mount function vectors from available behaviors
439 p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
440 mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
441 p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
442 mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
443 p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
444 mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
447 * Open real time and log devices - order is important.
449 if (args->logname[0]) {
450 error = xfs_blkdev_get(mp, args->logname, &logdev);
451 if (error)
452 return error;
454 if (args->rtname[0]) {
455 error = xfs_blkdev_get(mp, args->rtname, &rtdev);
456 if (error) {
457 xfs_blkdev_put(logdev);
458 return error;
461 if (rtdev == ddev || rtdev == logdev) {
462 cmn_err(CE_WARN,
463 "XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
464 xfs_blkdev_put(logdev);
465 xfs_blkdev_put(rtdev);
466 return EINVAL;
471 * Setup xfs_mount buffer target pointers
473 error = ENOMEM;
474 mp->m_ddev_targp = xfs_alloc_buftarg(ddev, 0);
475 if (!mp->m_ddev_targp) {
476 xfs_blkdev_put(logdev);
477 xfs_blkdev_put(rtdev);
478 return error;
480 if (rtdev) {
481 mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev, 1);
482 if (!mp->m_rtdev_targp)
483 goto error0;
485 mp->m_logdev_targp = (logdev && logdev != ddev) ?
486 xfs_alloc_buftarg(logdev, 1) : mp->m_ddev_targp;
487 if (!mp->m_logdev_targp)
488 goto error0;
491 * Setup flags based on mount(2) options and then the superblock
493 error = xfs_start_flags(vfsp, args, mp);
494 if (error)
495 goto error1;
496 error = xfs_readsb(mp);
497 if (error)
498 goto error1;
499 error = xfs_finish_flags(vfsp, args, mp);
500 if (error)
501 goto error2;
504 * Setup xfs_mount buffer target pointers based on superblock
506 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
507 mp->m_sb.sb_sectsize);
508 if (!error && logdev && logdev != ddev) {
509 unsigned int log_sector_size = BBSIZE;
511 if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
512 log_sector_size = mp->m_sb.sb_logsectsize;
513 error = xfs_setsize_buftarg(mp->m_logdev_targp,
514 mp->m_sb.sb_blocksize,
515 log_sector_size);
517 if (!error && rtdev)
518 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
519 mp->m_sb.sb_blocksize,
520 mp->m_sb.sb_sectsize);
521 if (error)
522 goto error2;
524 error = XFS_IOINIT(vfsp, args, flags);
525 if (!error)
526 return 0;
527 error2:
528 if (mp->m_sb_bp)
529 xfs_freesb(mp);
530 error1:
531 xfs_binval(mp->m_ddev_targp);
532 if (logdev && logdev != ddev)
533 xfs_binval(mp->m_logdev_targp);
534 if (rtdev)
535 xfs_binval(mp->m_rtdev_targp);
536 error0:
537 xfs_unmountfs_close(mp, credp);
538 return error;
541 STATIC int
542 xfs_unmount(
543 bhv_desc_t *bdp,
544 int flags,
545 cred_t *credp)
547 struct vfs *vfsp = bhvtovfs(bdp);
548 xfs_mount_t *mp = XFS_BHVTOM(bdp);
549 xfs_inode_t *rip;
550 vnode_t *rvp;
551 int unmount_event_wanted = 0;
552 int unmount_event_flags = 0;
553 int xfs_unmountfs_needed = 0;
554 int error;
556 rip = mp->m_rootip;
557 rvp = XFS_ITOV(rip);
559 if (vfsp->vfs_flag & VFS_DMI) {
560 error = XFS_SEND_PREUNMOUNT(mp, vfsp,
561 rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
562 NULL, NULL, 0, 0,
563 (mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
564 0:DM_FLAGS_UNWANTED);
565 if (error)
566 return XFS_ERROR(error);
567 unmount_event_wanted = 1;
568 unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
569 0 : DM_FLAGS_UNWANTED;
573 * First blow any referenced inode from this file system
574 * out of the reference cache, and delete the timer.
576 xfs_refcache_purge_mp(mp);
578 XFS_bflush(mp->m_ddev_targp);
579 error = xfs_unmount_flush(mp, 0);
580 if (error)
581 goto out;
583 ASSERT(vn_count(rvp) == 1);
586 * Drop the reference count
588 VN_RELE(rvp);
591 * If we're forcing a shutdown, typically because of a media error,
592 * we want to make sure we invalidate dirty pages that belong to
593 * referenced vnodes as well.
595 if (XFS_FORCED_SHUTDOWN(mp)) {
596 error = xfs_sync(&mp->m_bhv,
597 (SYNC_WAIT | SYNC_CLOSE), credp);
598 ASSERT(error != EFSCORRUPTED);
600 xfs_unmountfs_needed = 1;
602 out:
603 /* Send DMAPI event, if required.
604 * Then do xfs_unmountfs() if needed.
605 * Then return error (or zero).
607 if (unmount_event_wanted) {
608 /* Note: mp structure must still exist for
609 * XFS_SEND_UNMOUNT() call.
611 XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
612 DM_RIGHT_NULL, 0, error, unmount_event_flags);
614 if (xfs_unmountfs_needed) {
616 * Call common unmount function to flush to disk
617 * and free the super block buffer & mount structures.
619 xfs_unmountfs(mp, credp);
622 return XFS_ERROR(error);
625 #define REMOUNT_READONLY_FLAGS (SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
627 STATIC int
628 xfs_mntupdate(
629 bhv_desc_t *bdp,
630 int *flags,
631 struct xfs_mount_args *args)
633 struct vfs *vfsp = bhvtovfs(bdp);
634 xfs_mount_t *mp = XFS_BHVTOM(bdp);
635 int pincount, error;
636 int count = 0;
638 if (args->flags & XFSMNT_NOATIME)
639 mp->m_flags |= XFS_MOUNT_NOATIME;
640 else
641 mp->m_flags &= ~XFS_MOUNT_NOATIME;
643 if (!(vfsp->vfs_flag & VFS_RDONLY)) {
644 VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
647 if (*flags & MS_RDONLY) {
648 xfs_refcache_purge_mp(mp);
649 xfs_flush_buftarg(mp->m_ddev_targp, 0);
650 xfs_finish_reclaim_all(mp, 0);
652 /* This loop must run at least twice.
653 * The first instance of the loop will flush
654 * most meta data but that will generate more
655 * meta data (typically directory updates).
656 * Which then must be flushed and logged before
657 * we can write the unmount record.
659 do {
660 VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
661 pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
662 if (!pincount) {
663 delay(50);
664 count++;
666 } while (count < 2);
668 /* Ok now write out an unmount record */
669 xfs_log_unmount_write(mp);
670 xfs_unmountfs_writesb(mp);
671 vfsp->vfs_flag |= VFS_RDONLY;
672 } else {
673 vfsp->vfs_flag &= ~VFS_RDONLY;
676 return 0;
680 * xfs_unmount_flush implements a set of flush operation on special
681 * inodes, which are needed as a separate set of operations so that
682 * they can be called as part of relocation process.
685 xfs_unmount_flush(
686 xfs_mount_t *mp, /* Mount structure we are getting
687 rid of. */
688 int relocation) /* Called from vfs relocation. */
690 xfs_inode_t *rip = mp->m_rootip;
691 xfs_inode_t *rbmip;
692 xfs_inode_t *rsumip = NULL;
693 vnode_t *rvp = XFS_ITOV(rip);
694 int error;
696 xfs_ilock(rip, XFS_ILOCK_EXCL);
697 xfs_iflock(rip);
700 * Flush out the real time inodes.
702 if ((rbmip = mp->m_rbmip) != NULL) {
703 xfs_ilock(rbmip, XFS_ILOCK_EXCL);
704 xfs_iflock(rbmip);
705 error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
706 xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
708 if (error == EFSCORRUPTED)
709 goto fscorrupt_out;
711 ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
713 rsumip = mp->m_rsumip;
714 xfs_ilock(rsumip, XFS_ILOCK_EXCL);
715 xfs_iflock(rsumip);
716 error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
717 xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
719 if (error == EFSCORRUPTED)
720 goto fscorrupt_out;
722 ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
726 * Synchronously flush root inode to disk
728 error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
729 if (error == EFSCORRUPTED)
730 goto fscorrupt_out2;
732 if (vn_count(rvp) != 1 && !relocation) {
733 xfs_iunlock(rip, XFS_ILOCK_EXCL);
734 return XFS_ERROR(EBUSY);
738 * Release dquot that rootinode, rbmino and rsumino might be holding,
739 * flush and purge the quota inodes.
741 error = XFS_QM_UNMOUNT(mp);
742 if (error == EFSCORRUPTED)
743 goto fscorrupt_out2;
745 if (rbmip) {
746 VN_RELE(XFS_ITOV(rbmip));
747 VN_RELE(XFS_ITOV(rsumip));
750 xfs_iunlock(rip, XFS_ILOCK_EXCL);
751 return 0;
753 fscorrupt_out:
754 xfs_ifunlock(rip);
756 fscorrupt_out2:
757 xfs_iunlock(rip, XFS_ILOCK_EXCL);
759 return XFS_ERROR(EFSCORRUPTED);
763 * xfs_root extracts the root vnode from a vfs.
765 * vfsp -- the vfs struct for the desired file system
766 * vpp -- address of the caller's vnode pointer which should be
767 * set to the desired fs root vnode
769 STATIC int
770 xfs_root(
771 bhv_desc_t *bdp,
772 vnode_t **vpp)
774 vnode_t *vp;
776 vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
777 VN_HOLD(vp);
778 *vpp = vp;
779 return 0;
783 * xfs_statvfs
785 * Fill in the statvfs structure for the given file system. We use
786 * the superblock lock in the mount structure to ensure a consistent
787 * snapshot of the counters returned.
789 STATIC int
790 xfs_statvfs(
791 bhv_desc_t *bdp,
792 xfs_statfs_t *statp,
793 vnode_t *vp)
795 __uint64_t fakeinos;
796 xfs_extlen_t lsize;
797 xfs_mount_t *mp;
798 xfs_sb_t *sbp;
799 unsigned long s;
800 u64 id;
802 mp = XFS_BHVTOM(bdp);
803 sbp = &(mp->m_sb);
805 statp->f_type = XFS_SB_MAGIC;
807 s = XFS_SB_LOCK(mp);
808 statp->f_bsize = sbp->sb_blocksize;
809 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
810 statp->f_blocks = sbp->sb_dblocks - lsize;
811 statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
812 fakeinos = statp->f_bfree << sbp->sb_inopblog;
813 #if XFS_BIG_INUMS
814 fakeinos += mp->m_inoadd;
815 #endif
816 statp->f_files =
817 MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
818 if (mp->m_maxicount)
819 #if XFS_BIG_INUMS
820 if (!mp->m_inoadd)
821 #endif
822 statp->f_files = min_t(typeof(statp->f_files),
823 statp->f_files,
824 mp->m_maxicount);
825 statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
826 XFS_SB_UNLOCK(mp, s);
828 id = huge_encode_dev(mp->m_dev);
829 statp->f_fsid.val[0] = (u32)id;
830 statp->f_fsid.val[1] = (u32)(id >> 32);
831 statp->f_namelen = MAXNAMELEN - 1;
833 return 0;
838 * xfs_sync flushes any pending I/O to file system vfsp.
840 * This routine is called by vfs_sync() to make sure that things make it
841 * out to disk eventually, on sync() system calls to flush out everything,
842 * and when the file system is unmounted. For the vfs_sync() case, all
843 * we really need to do is sync out the log to make all of our meta-data
844 * updates permanent (except for timestamps). For calls from pflushd(),
845 * dirty pages are kept moving by calling pdflush() on the inodes
846 * containing them. We also flush the inodes that we can lock without
847 * sleeping and the superblock if we can lock it without sleeping from
848 * vfs_sync() so that items at the tail of the log are always moving out.
850 * Flags:
851 * SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
852 * to sleep if we can help it. All we really need
853 * to do is ensure that the log is synced at least
854 * periodically. We also push the inodes and
855 * superblock if we can lock them without sleeping
856 * and they are not pinned.
857 * SYNC_ATTR - We need to flush the inodes. If SYNC_BDFLUSH is not
858 * set, then we really want to lock each inode and flush
859 * it.
860 * SYNC_WAIT - All the flushes that take place in this call should
861 * be synchronous.
862 * SYNC_DELWRI - This tells us to push dirty pages associated with
863 * inodes. SYNC_WAIT and SYNC_BDFLUSH are used to
864 * determine if they should be flushed sync, async, or
865 * delwri.
866 * SYNC_CLOSE - This flag is passed when the system is being
867 * unmounted. We should sync and invalidate everthing.
868 * SYNC_FSDATA - This indicates that the caller would like to make
869 * sure the superblock is safe on disk. We can ensure
870 * this by simply makeing sure the log gets flushed
871 * if SYNC_BDFLUSH is set, and by actually writing it
872 * out otherwise.
875 /*ARGSUSED*/
876 STATIC int
877 xfs_sync(
878 bhv_desc_t *bdp,
879 int flags,
880 cred_t *credp)
882 xfs_mount_t *mp;
884 mp = XFS_BHVTOM(bdp);
885 return (xfs_syncsub(mp, flags, 0, NULL));
889 * xfs sync routine for internal use
891 * This routine supports all of the flags defined for the generic VFS_SYNC
892 * interface as explained above under xfs_sync. In the interests of not
893 * changing interfaces within the 6.5 family, additional internallly-
894 * required functions are specified within a separate xflags parameter,
895 * only available by calling this routine.
898 STATIC int
899 xfs_sync_inodes(
900 xfs_mount_t *mp,
901 int flags,
902 int xflags,
903 int *bypassed)
905 xfs_inode_t *ip = NULL;
906 xfs_inode_t *ip_next;
907 xfs_buf_t *bp;
908 vnode_t *vp = NULL;
909 vmap_t vmap;
910 int error;
911 int last_error;
912 uint64_t fflag;
913 uint lock_flags;
914 uint base_lock_flags;
915 boolean_t mount_locked;
916 boolean_t vnode_refed;
917 int preempt;
918 xfs_dinode_t *dip;
919 xfs_iptr_t *ipointer;
920 #ifdef DEBUG
921 boolean_t ipointer_in = B_FALSE;
923 #define IPOINTER_SET ipointer_in = B_TRUE
924 #define IPOINTER_CLR ipointer_in = B_FALSE
925 #else
926 #define IPOINTER_SET
927 #define IPOINTER_CLR
928 #endif
931 /* Insert a marker record into the inode list after inode ip. The list
932 * must be locked when this is called. After the call the list will no
933 * longer be locked.
935 #define IPOINTER_INSERT(ip, mp) { \
936 ASSERT(ipointer_in == B_FALSE); \
937 ipointer->ip_mnext = ip->i_mnext; \
938 ipointer->ip_mprev = ip; \
939 ip->i_mnext = (xfs_inode_t *)ipointer; \
940 ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
941 preempt = 0; \
942 XFS_MOUNT_IUNLOCK(mp); \
943 mount_locked = B_FALSE; \
944 IPOINTER_SET; \
947 /* Remove the marker from the inode list. If the marker was the only item
948 * in the list then there are no remaining inodes and we should zero out
949 * the whole list. If we are the current head of the list then move the head
950 * past us.
952 #define IPOINTER_REMOVE(ip, mp) { \
953 ASSERT(ipointer_in == B_TRUE); \
954 if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
955 ip = ipointer->ip_mnext; \
956 ip->i_mprev = ipointer->ip_mprev; \
957 ipointer->ip_mprev->i_mnext = ip; \
958 if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
959 mp->m_inodes = ip; \
961 } else { \
962 ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
963 mp->m_inodes = NULL; \
964 ip = NULL; \
966 IPOINTER_CLR; \
969 #define XFS_PREEMPT_MASK 0x7f
971 if (bypassed)
972 *bypassed = 0;
973 if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
974 return 0;
975 error = 0;
976 last_error = 0;
977 preempt = 0;
979 /* Allocate a reference marker */
980 ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
982 fflag = XFS_B_ASYNC; /* default is don't wait */
983 if (flags & SYNC_BDFLUSH)
984 fflag = XFS_B_DELWRI;
985 if (flags & SYNC_WAIT)
986 fflag = 0; /* synchronous overrides all */
988 base_lock_flags = XFS_ILOCK_SHARED;
989 if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
991 * We need the I/O lock if we're going to call any of
992 * the flush/inval routines.
994 base_lock_flags |= XFS_IOLOCK_SHARED;
997 XFS_MOUNT_ILOCK(mp);
999 ip = mp->m_inodes;
1001 mount_locked = B_TRUE;
1002 vnode_refed = B_FALSE;
1004 IPOINTER_CLR;
1006 do {
1007 ASSERT(ipointer_in == B_FALSE);
1008 ASSERT(vnode_refed == B_FALSE);
1010 lock_flags = base_lock_flags;
1013 * There were no inodes in the list, just break out
1014 * of the loop.
1016 if (ip == NULL) {
1017 break;
1021 * We found another sync thread marker - skip it
1023 if (ip->i_mount == NULL) {
1024 ip = ip->i_mnext;
1025 continue;
1028 vp = XFS_ITOV_NULL(ip);
1031 * If the vnode is gone then this is being torn down,
1032 * call reclaim if it is flushed, else let regular flush
1033 * code deal with it later in the loop.
1036 if (vp == NULL) {
1037 /* Skip ones already in reclaim */
1038 if (ip->i_flags & XFS_IRECLAIM) {
1039 ip = ip->i_mnext;
1040 continue;
1042 if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1043 ip = ip->i_mnext;
1044 } else if ((xfs_ipincount(ip) == 0) &&
1045 xfs_iflock_nowait(ip)) {
1046 IPOINTER_INSERT(ip, mp);
1048 xfs_finish_reclaim(ip, 1,
1049 XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1051 XFS_MOUNT_ILOCK(mp);
1052 mount_locked = B_TRUE;
1053 IPOINTER_REMOVE(ip, mp);
1054 } else {
1055 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1056 ip = ip->i_mnext;
1058 continue;
1061 if (VN_BAD(vp)) {
1062 ip = ip->i_mnext;
1063 continue;
1066 if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1067 XFS_MOUNT_IUNLOCK(mp);
1068 kmem_free(ipointer, sizeof(xfs_iptr_t));
1069 return 0;
1073 * If this is just vfs_sync() or pflushd() calling
1074 * then we can skip inodes for which it looks like
1075 * there is nothing to do. Since we don't have the
1076 * inode locked this is racey, but these are periodic
1077 * calls so it doesn't matter. For the others we want
1078 * to know for sure, so we at least try to lock them.
1080 if (flags & SYNC_BDFLUSH) {
1081 if (((ip->i_itemp == NULL) ||
1082 !(ip->i_itemp->ili_format.ilf_fields &
1083 XFS_ILOG_ALL)) &&
1084 (ip->i_update_core == 0)) {
1085 ip = ip->i_mnext;
1086 continue;
1091 * Try to lock without sleeping. We're out of order with
1092 * the inode list lock here, so if we fail we need to drop
1093 * the mount lock and try again. If we're called from
1094 * bdflush() here, then don't bother.
1096 * The inode lock here actually coordinates with the
1097 * almost spurious inode lock in xfs_ireclaim() to prevent
1098 * the vnode we handle here without a reference from
1099 * being freed while we reference it. If we lock the inode
1100 * while it's on the mount list here, then the spurious inode
1101 * lock in xfs_ireclaim() after the inode is pulled from
1102 * the mount list will sleep until we release it here.
1103 * This keeps the vnode from being freed while we reference
1104 * it. It is also cheaper and simpler than actually doing
1105 * a vn_get() for every inode we touch here.
1107 if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1109 if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1110 ip = ip->i_mnext;
1111 continue;
1115 * We need to unlock the inode list lock in order
1116 * to lock the inode. Insert a marker record into
1117 * the inode list to remember our position, dropping
1118 * the lock is now done inside the IPOINTER_INSERT
1119 * macro.
1121 * We also use the inode list lock to protect us
1122 * in taking a snapshot of the vnode version number
1123 * for use in calling vn_get().
1125 VMAP(vp, vmap);
1126 IPOINTER_INSERT(ip, mp);
1128 vp = vn_get(vp, &vmap);
1129 if (vp == NULL) {
1131 * The vnode was reclaimed once we let go
1132 * of the inode list lock. Skip to the
1133 * next list entry. Remove the marker.
1136 XFS_MOUNT_ILOCK(mp);
1138 mount_locked = B_TRUE;
1139 vnode_refed = B_FALSE;
1141 IPOINTER_REMOVE(ip, mp);
1143 continue;
1146 xfs_ilock(ip, lock_flags);
1148 ASSERT(vp == XFS_ITOV(ip));
1149 ASSERT(ip->i_mount == mp);
1151 vnode_refed = B_TRUE;
1154 /* From here on in the loop we may have a marker record
1155 * in the inode list.
1158 if ((flags & SYNC_CLOSE) && (vp != NULL)) {
1160 * This is the shutdown case. We just need to
1161 * flush and invalidate all the pages associated
1162 * with the inode. Drop the inode lock since
1163 * we can't hold it across calls to the buffer
1164 * cache.
1166 * We don't set the VREMAPPING bit in the vnode
1167 * here, because we don't hold the vnode lock
1168 * exclusively. It doesn't really matter, though,
1169 * because we only come here when we're shutting
1170 * down anyway.
1172 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1174 if (XFS_FORCED_SHUTDOWN(mp)) {
1175 VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1176 } else {
1177 VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1180 xfs_ilock(ip, XFS_ILOCK_SHARED);
1182 } else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1183 if (VN_DIRTY(vp)) {
1184 /* We need to have dropped the lock here,
1185 * so insert a marker if we have not already
1186 * done so.
1188 if (mount_locked) {
1189 IPOINTER_INSERT(ip, mp);
1193 * Drop the inode lock since we can't hold it
1194 * across calls to the buffer cache.
1196 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1197 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1198 fflag, FI_NONE, error);
1199 xfs_ilock(ip, XFS_ILOCK_SHARED);
1204 if (flags & SYNC_BDFLUSH) {
1205 if ((flags & SYNC_ATTR) &&
1206 ((ip->i_update_core) ||
1207 ((ip->i_itemp != NULL) &&
1208 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1210 /* Insert marker and drop lock if not already
1211 * done.
1213 if (mount_locked) {
1214 IPOINTER_INSERT(ip, mp);
1218 * We don't want the periodic flushing of the
1219 * inodes by vfs_sync() to interfere with
1220 * I/O to the file, especially read I/O
1221 * where it is only the access time stamp
1222 * that is being flushed out. To prevent
1223 * long periods where we have both inode
1224 * locks held shared here while reading the
1225 * inode's buffer in from disk, we drop the
1226 * inode lock while reading in the inode
1227 * buffer. We have to release the buffer
1228 * and reacquire the inode lock so that they
1229 * are acquired in the proper order (inode
1230 * locks first). The buffer will go at the
1231 * end of the lru chain, though, so we can
1232 * expect it to still be there when we go
1233 * for it again in xfs_iflush().
1235 if ((xfs_ipincount(ip) == 0) &&
1236 xfs_iflock_nowait(ip)) {
1238 xfs_ifunlock(ip);
1239 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1241 error = xfs_itobp(mp, NULL, ip,
1242 &dip, &bp, 0);
1243 if (!error) {
1244 xfs_buf_relse(bp);
1245 } else {
1246 /* Bailing out, remove the
1247 * marker and free it.
1249 XFS_MOUNT_ILOCK(mp);
1251 IPOINTER_REMOVE(ip, mp);
1253 XFS_MOUNT_IUNLOCK(mp);
1255 ASSERT(!(lock_flags &
1256 XFS_IOLOCK_SHARED));
1258 kmem_free(ipointer,
1259 sizeof(xfs_iptr_t));
1260 return (0);
1264 * Since we dropped the inode lock,
1265 * the inode may have been reclaimed.
1266 * Therefore, we reacquire the mount
1267 * lock and check to see if we were the
1268 * inode reclaimed. If this happened
1269 * then the ipointer marker will no
1270 * longer point back at us. In this
1271 * case, move ip along to the inode
1272 * after the marker, remove the marker
1273 * and continue.
1275 XFS_MOUNT_ILOCK(mp);
1276 mount_locked = B_TRUE;
1278 if (ip != ipointer->ip_mprev) {
1279 IPOINTER_REMOVE(ip, mp);
1281 ASSERT(!vnode_refed);
1282 ASSERT(!(lock_flags &
1283 XFS_IOLOCK_SHARED));
1284 continue;
1287 ASSERT(ip->i_mount == mp);
1289 if (xfs_ilock_nowait(ip,
1290 XFS_ILOCK_SHARED) == 0) {
1291 ASSERT(ip->i_mount == mp);
1293 * We failed to reacquire
1294 * the inode lock without
1295 * sleeping, so just skip
1296 * the inode for now. We
1297 * clear the ILOCK bit from
1298 * the lock_flags so that we
1299 * won't try to drop a lock
1300 * we don't hold below.
1302 lock_flags &= ~XFS_ILOCK_SHARED;
1303 IPOINTER_REMOVE(ip_next, mp);
1304 } else if ((xfs_ipincount(ip) == 0) &&
1305 xfs_iflock_nowait(ip)) {
1306 ASSERT(ip->i_mount == mp);
1308 * Since this is vfs_sync()
1309 * calling we only flush the
1310 * inode out if we can lock
1311 * it without sleeping and
1312 * it is not pinned. Drop
1313 * the mount lock here so
1314 * that we don't hold it for
1315 * too long. We already have
1316 * a marker in the list here.
1318 XFS_MOUNT_IUNLOCK(mp);
1319 mount_locked = B_FALSE;
1320 error = xfs_iflush(ip,
1321 XFS_IFLUSH_DELWRI);
1322 } else {
1323 ASSERT(ip->i_mount == mp);
1324 IPOINTER_REMOVE(ip_next, mp);
1330 } else {
1331 if ((flags & SYNC_ATTR) &&
1332 ((ip->i_update_core) ||
1333 ((ip->i_itemp != NULL) &&
1334 (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1335 if (mount_locked) {
1336 IPOINTER_INSERT(ip, mp);
1339 if (flags & SYNC_WAIT) {
1340 xfs_iflock(ip);
1341 error = xfs_iflush(ip,
1342 XFS_IFLUSH_SYNC);
1343 } else {
1345 * If we can't acquire the flush
1346 * lock, then the inode is already
1347 * being flushed so don't bother
1348 * waiting. If we can lock it then
1349 * do a delwri flush so we can
1350 * combine multiple inode flushes
1351 * in each disk write.
1353 if (xfs_iflock_nowait(ip)) {
1354 error = xfs_iflush(ip,
1355 XFS_IFLUSH_DELWRI);
1357 else if (bypassed)
1358 (*bypassed)++;
1363 if (lock_flags != 0) {
1364 xfs_iunlock(ip, lock_flags);
1367 if (vnode_refed) {
1369 * If we had to take a reference on the vnode
1370 * above, then wait until after we've unlocked
1371 * the inode to release the reference. This is
1372 * because we can be already holding the inode
1373 * lock when VN_RELE() calls xfs_inactive().
1375 * Make sure to drop the mount lock before calling
1376 * VN_RELE() so that we don't trip over ourselves if
1377 * we have to go for the mount lock again in the
1378 * inactive code.
1380 if (mount_locked) {
1381 IPOINTER_INSERT(ip, mp);
1384 VN_RELE(vp);
1386 vnode_refed = B_FALSE;
1389 if (error) {
1390 last_error = error;
1394 * bail out if the filesystem is corrupted.
1396 if (error == EFSCORRUPTED) {
1397 if (!mount_locked) {
1398 XFS_MOUNT_ILOCK(mp);
1399 IPOINTER_REMOVE(ip, mp);
1401 XFS_MOUNT_IUNLOCK(mp);
1402 ASSERT(ipointer_in == B_FALSE);
1403 kmem_free(ipointer, sizeof(xfs_iptr_t));
1404 return XFS_ERROR(error);
1407 /* Let other threads have a chance at the mount lock
1408 * if we have looped many times without dropping the
1409 * lock.
1411 if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1412 if (mount_locked) {
1413 IPOINTER_INSERT(ip, mp);
1417 if (mount_locked == B_FALSE) {
1418 XFS_MOUNT_ILOCK(mp);
1419 mount_locked = B_TRUE;
1420 IPOINTER_REMOVE(ip, mp);
1421 continue;
1424 ASSERT(ipointer_in == B_FALSE);
1425 ip = ip->i_mnext;
1427 } while (ip != mp->m_inodes);
1429 XFS_MOUNT_IUNLOCK(mp);
1431 ASSERT(ipointer_in == B_FALSE);
1433 kmem_free(ipointer, sizeof(xfs_iptr_t));
1434 return XFS_ERROR(last_error);
1438 * xfs sync routine for internal use
1440 * This routine supports all of the flags defined for the generic VFS_SYNC
1441 * interface as explained above under xfs_sync. In the interests of not
1442 * changing interfaces within the 6.5 family, additional internallly-
1443 * required functions are specified within a separate xflags parameter,
1444 * only available by calling this routine.
1448 xfs_syncsub(
1449 xfs_mount_t *mp,
1450 int flags,
1451 int xflags,
1452 int *bypassed)
1454 int error = 0;
1455 int last_error = 0;
1456 uint log_flags = XFS_LOG_FORCE;
1457 xfs_buf_t *bp;
1458 xfs_buf_log_item_t *bip;
1461 * Sync out the log. This ensures that the log is periodically
1462 * flushed even if there is not enough activity to fill it up.
1464 if (flags & SYNC_WAIT)
1465 log_flags |= XFS_LOG_SYNC;
1467 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1469 if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1470 if (flags & SYNC_BDFLUSH)
1471 xfs_finish_reclaim_all(mp, 1);
1472 else
1473 error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1477 * Flushing out dirty data above probably generated more
1478 * log activity, so if this isn't vfs_sync() then flush
1479 * the log again.
1481 if (flags & SYNC_DELWRI) {
1482 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1485 if (flags & SYNC_FSDATA) {
1487 * If this is vfs_sync() then only sync the superblock
1488 * if we can lock it without sleeping and it is not pinned.
1490 if (flags & SYNC_BDFLUSH) {
1491 bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1492 if (bp != NULL) {
1493 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1494 if ((bip != NULL) &&
1495 xfs_buf_item_dirty(bip)) {
1496 if (!(XFS_BUF_ISPINNED(bp))) {
1497 XFS_BUF_ASYNC(bp);
1498 error = xfs_bwrite(mp, bp);
1499 } else {
1500 xfs_buf_relse(bp);
1502 } else {
1503 xfs_buf_relse(bp);
1506 } else {
1507 bp = xfs_getsb(mp, 0);
1509 * If the buffer is pinned then push on the log so
1510 * we won't get stuck waiting in the write for
1511 * someone, maybe ourselves, to flush the log.
1512 * Even though we just pushed the log above, we
1513 * did not have the superblock buffer locked at
1514 * that point so it can become pinned in between
1515 * there and here.
1517 if (XFS_BUF_ISPINNED(bp))
1518 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1519 if (flags & SYNC_WAIT)
1520 XFS_BUF_UNASYNC(bp);
1521 else
1522 XFS_BUF_ASYNC(bp);
1523 error = xfs_bwrite(mp, bp);
1525 if (error) {
1526 last_error = error;
1531 * If this is the periodic sync, then kick some entries out of
1532 * the reference cache. This ensures that idle entries are
1533 * eventually kicked out of the cache.
1535 if (flags & SYNC_REFCACHE) {
1536 xfs_refcache_purge_some(mp);
1540 * Now check to see if the log needs a "dummy" transaction.
1543 if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1544 xfs_trans_t *tp;
1545 xfs_inode_t *ip;
1548 * Put a dummy transaction in the log to tell
1549 * recovery that all others are OK.
1551 tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1552 if ((error = xfs_trans_reserve(tp, 0,
1553 XFS_ICHANGE_LOG_RES(mp),
1554 0, 0, 0))) {
1555 xfs_trans_cancel(tp, 0);
1556 return error;
1559 ip = mp->m_rootip;
1560 xfs_ilock(ip, XFS_ILOCK_EXCL);
1562 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1563 xfs_trans_ihold(tp, ip);
1564 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1565 error = xfs_trans_commit(tp, 0, NULL);
1566 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1567 xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1571 * When shutting down, we need to insure that the AIL is pushed
1572 * to disk or the filesystem can appear corrupt from the PROM.
1574 if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1575 XFS_bflush(mp->m_ddev_targp);
1576 if (mp->m_rtdev_targp) {
1577 XFS_bflush(mp->m_rtdev_targp);
1581 return XFS_ERROR(last_error);
1585 * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1587 STATIC int
1588 xfs_vget(
1589 bhv_desc_t *bdp,
1590 vnode_t **vpp,
1591 fid_t *fidp)
1593 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1594 xfs_fid_t *xfid = (struct xfs_fid *)fidp;
1595 xfs_inode_t *ip;
1596 int error;
1597 xfs_ino_t ino;
1598 unsigned int igen;
1601 * Invalid. Since handles can be created in user space and passed in
1602 * via gethandle(), this is not cause for a panic.
1604 if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1605 return XFS_ERROR(EINVAL);
1607 ino = xfid->xfs_fid_ino;
1608 igen = xfid->xfs_fid_gen;
1611 * NFS can sometimes send requests for ino 0. Fail them gracefully.
1613 if (ino == 0)
1614 return XFS_ERROR(ESTALE);
1616 error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1617 if (error) {
1618 *vpp = NULL;
1619 return error;
1622 if (ip == NULL) {
1623 *vpp = NULL;
1624 return XFS_ERROR(EIO);
1627 if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1628 xfs_iput_new(ip, XFS_ILOCK_SHARED);
1629 *vpp = NULL;
1630 return XFS_ERROR(ENOENT);
1633 *vpp = XFS_ITOV(ip);
1634 xfs_iunlock(ip, XFS_ILOCK_SHARED);
1635 return 0;
1639 #define MNTOPT_LOGBUFS "logbufs" /* number of XFS log buffers */
1640 #define MNTOPT_LOGBSIZE "logbsize" /* size of XFS log buffers */
1641 #define MNTOPT_LOGDEV "logdev" /* log device */
1642 #define MNTOPT_RTDEV "rtdev" /* realtime I/O device */
1643 #define MNTOPT_BIOSIZE "biosize" /* log2 of preferred buffered io size */
1644 #define MNTOPT_WSYNC "wsync" /* safe-mode nfs compatible mount */
1645 #define MNTOPT_INO64 "ino64" /* force inodes into 64-bit range */
1646 #define MNTOPT_NOALIGN "noalign" /* turn off stripe alignment */
1647 #define MNTOPT_SWALLOC "swalloc" /* turn on stripe width allocation */
1648 #define MNTOPT_SUNIT "sunit" /* data volume stripe unit */
1649 #define MNTOPT_SWIDTH "swidth" /* data volume stripe width */
1650 #define MNTOPT_NOUUID "nouuid" /* ignore filesystem UUID */
1651 #define MNTOPT_MTPT "mtpt" /* filesystem mount point */
1652 #define MNTOPT_IHASHSIZE "ihashsize" /* size of inode hash table */
1653 #define MNTOPT_NORECOVERY "norecovery" /* don't run XFS recovery */
1654 #define MNTOPT_NOLOGFLUSH "nologflush" /* don't hard flush on log writes */
1655 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1656 #define MNTOPT_64BITINODE "inode64" /* inodes can be allocated anywhere */
1657 #define MNTOPT_IKEEP "ikeep" /* do not free empty inode clusters */
1658 #define MNTOPT_NOIKEEP "noikeep" /* free empty inode clusters */
1662 xfs_parseargs(
1663 struct bhv_desc *bhv,
1664 char *options,
1665 struct xfs_mount_args *args,
1666 int update)
1668 struct vfs *vfsp = bhvtovfs(bhv);
1669 char *this_char, *value, *eov;
1670 int dsunit, dswidth, vol_dsunit, vol_dswidth;
1671 int iosize;
1673 #if 0 /* XXX: off by default, until some remaining issues ironed out */
1674 args->flags |= XFSMNT_IDELETE; /* default to on */
1675 #endif
1677 if (!options)
1678 return 0;
1680 iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1682 while ((this_char = strsep(&options, ",")) != NULL) {
1683 if (!*this_char)
1684 continue;
1685 if ((value = strchr(this_char, '=')) != NULL)
1686 *value++ = 0;
1688 if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1689 if (!value || !*value) {
1690 printk("XFS: %s option requires an argument\n",
1691 MNTOPT_LOGBUFS);
1692 return EINVAL;
1694 args->logbufs = simple_strtoul(value, &eov, 10);
1695 } else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1696 int last, in_kilobytes = 0;
1698 if (!value || !*value) {
1699 printk("XFS: %s option requires an argument\n",
1700 MNTOPT_LOGBSIZE);
1701 return EINVAL;
1703 last = strlen(value) - 1;
1704 if (value[last] == 'K' || value[last] == 'k') {
1705 in_kilobytes = 1;
1706 value[last] = '\0';
1708 args->logbufsize = simple_strtoul(value, &eov, 10);
1709 if (in_kilobytes)
1710 args->logbufsize <<= 10;
1711 } else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1712 if (!value || !*value) {
1713 printk("XFS: %s option requires an argument\n",
1714 MNTOPT_LOGDEV);
1715 return EINVAL;
1717 strncpy(args->logname, value, MAXNAMELEN);
1718 } else if (!strcmp(this_char, MNTOPT_MTPT)) {
1719 if (!value || !*value) {
1720 printk("XFS: %s option requires an argument\n",
1721 MNTOPT_MTPT);
1722 return EINVAL;
1724 strncpy(args->mtpt, value, MAXNAMELEN);
1725 } else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1726 if (!value || !*value) {
1727 printk("XFS: %s option requires an argument\n",
1728 MNTOPT_RTDEV);
1729 return EINVAL;
1731 strncpy(args->rtname, value, MAXNAMELEN);
1732 } else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1733 if (!value || !*value) {
1734 printk("XFS: %s option requires an argument\n",
1735 MNTOPT_BIOSIZE);
1736 return EINVAL;
1738 iosize = simple_strtoul(value, &eov, 10);
1739 args->flags |= XFSMNT_IOSIZE;
1740 args->iosizelog = (uint8_t) iosize;
1741 } else if (!strcmp(this_char, MNTOPT_IHASHSIZE)) {
1742 if (!value || !*value) {
1743 printk("XFS: %s option requires an argument\n",
1744 this_char);
1745 return EINVAL;
1747 args->flags |= XFSMNT_IHASHSIZE;
1748 args->ihashsize = simple_strtoul(value, &eov, 10);
1749 } else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1750 args->flags |= XFSMNT_WSYNC;
1751 } else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1752 args->flags |= XFSMNT_OSYNCISOSYNC;
1753 } else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1754 args->flags |= XFSMNT_NORECOVERY;
1755 } else if (!strcmp(this_char, MNTOPT_INO64)) {
1756 args->flags |= XFSMNT_INO64;
1757 #if !XFS_BIG_INUMS
1758 printk("XFS: %s option not allowed on this system\n",
1759 MNTOPT_INO64);
1760 return EINVAL;
1761 #endif
1762 } else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1763 args->flags |= XFSMNT_NOALIGN;
1764 } else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1765 args->flags |= XFSMNT_SWALLOC;
1766 } else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1767 if (!value || !*value) {
1768 printk("XFS: %s option requires an argument\n",
1769 MNTOPT_SUNIT);
1770 return EINVAL;
1772 dsunit = simple_strtoul(value, &eov, 10);
1773 } else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1774 if (!value || !*value) {
1775 printk("XFS: %s option requires an argument\n",
1776 MNTOPT_SWIDTH);
1777 return EINVAL;
1779 dswidth = simple_strtoul(value, &eov, 10);
1780 } else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1781 args->flags &= ~XFSMNT_32BITINODES;
1782 #if !XFS_BIG_INUMS
1783 printk("XFS: %s option not allowed on this system\n",
1784 MNTOPT_64BITINODE);
1785 return EINVAL;
1786 #endif
1787 } else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1788 args->flags |= XFSMNT_NOUUID;
1789 } else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1790 args->flags |= XFSMNT_NOLOGFLUSH;
1791 } else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1792 args->flags &= ~XFSMNT_IDELETE;
1793 } else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1794 args->flags |= XFSMNT_IDELETE;
1795 } else if (!strcmp(this_char, "osyncisdsync")) {
1796 /* no-op, this is now the default */
1797 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1798 } else if (!strcmp(this_char, "irixsgid")) {
1799 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1800 } else {
1801 printk("XFS: unknown mount option [%s].\n", this_char);
1802 return EINVAL;
1806 if (args->flags & XFSMNT_NORECOVERY) {
1807 if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1808 printk("XFS: no-recovery mounts must be read-only.\n");
1809 return EINVAL;
1813 if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1814 printk(
1815 "XFS: sunit and swidth options incompatible with the noalign option\n");
1816 return EINVAL;
1819 if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1820 printk("XFS: sunit and swidth must be specified together\n");
1821 return EINVAL;
1824 if (dsunit && (dswidth % dsunit != 0)) {
1825 printk(
1826 "XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1827 dswidth, dsunit);
1828 return EINVAL;
1831 if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1832 if (dsunit) {
1833 args->sunit = dsunit;
1834 args->flags |= XFSMNT_RETERR;
1835 } else {
1836 args->sunit = vol_dsunit;
1838 dswidth ? (args->swidth = dswidth) :
1839 (args->swidth = vol_dswidth);
1840 } else {
1841 args->sunit = args->swidth = 0;
1844 return 0;
1848 xfs_showargs(
1849 struct bhv_desc *bhv,
1850 struct seq_file *m)
1852 static struct proc_xfs_info {
1853 int flag;
1854 char *str;
1855 } xfs_info[] = {
1856 /* the few simple ones we can get from the mount struct */
1857 { XFS_MOUNT_WSYNC, "," MNTOPT_WSYNC },
1858 { XFS_MOUNT_INO64, "," MNTOPT_INO64 },
1859 { XFS_MOUNT_NOALIGN, "," MNTOPT_NOALIGN },
1860 { XFS_MOUNT_SWALLOC, "," MNTOPT_SWALLOC },
1861 { XFS_MOUNT_NOUUID, "," MNTOPT_NOUUID },
1862 { XFS_MOUNT_NORECOVERY, "," MNTOPT_NORECOVERY },
1863 { XFS_MOUNT_OSYNCISOSYNC, "," MNTOPT_OSYNCISOSYNC },
1864 { XFS_MOUNT_NOLOGFLUSH, "," MNTOPT_NOLOGFLUSH },
1865 { XFS_MOUNT_IDELETE, "," MNTOPT_NOIKEEP },
1866 { 0, NULL }
1868 struct proc_xfs_info *xfs_infop;
1869 struct xfs_mount *mp = XFS_BHVTOM(bhv);
1871 for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1872 if (mp->m_flags & xfs_infop->flag)
1873 seq_puts(m, xfs_infop->str);
1876 if (mp->m_flags & XFS_MOUNT_IHASHSIZE)
1877 seq_printf(m, "," MNTOPT_IHASHSIZE "=%d", mp->m_ihsize);
1879 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1880 seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1882 if (mp->m_logbufs > 0)
1883 seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1885 if (mp->m_logbsize > 0)
1886 seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1888 if (mp->m_ddev_targp != mp->m_logdev_targp)
1889 seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1890 XFS_BUFTARG_NAME(mp->m_logdev_targp));
1892 if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1893 seq_printf(m, "," MNTOPT_RTDEV "=%s",
1894 XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1896 if (mp->m_dalign > 0)
1897 seq_printf(m, "," MNTOPT_SUNIT "=%d",
1898 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1900 if (mp->m_swidth > 0)
1901 seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1902 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1904 if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1905 seq_printf(m, "," MNTOPT_64BITINODE);
1907 return 0;
1910 STATIC void
1911 xfs_freeze(
1912 bhv_desc_t *bdp)
1914 xfs_mount_t *mp = XFS_BHVTOM(bdp);
1916 while (atomic_read(&mp->m_active_trans) > 0)
1917 delay(100);
1919 /* Push the superblock and write an unmount record */
1920 xfs_log_unmount_write(mp);
1921 xfs_unmountfs_writesb(mp);
1925 vfsops_t xfs_vfsops = {
1926 BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1927 .vfs_parseargs = xfs_parseargs,
1928 .vfs_showargs = xfs_showargs,
1929 .vfs_mount = xfs_mount,
1930 .vfs_unmount = xfs_unmount,
1931 .vfs_mntupdate = xfs_mntupdate,
1932 .vfs_root = xfs_root,
1933 .vfs_statvfs = xfs_statvfs,
1934 .vfs_sync = xfs_sync,
1935 .vfs_vget = xfs_vget,
1936 .vfs_dmapiops = (vfs_dmapiops_t)fs_nosys,
1937 .vfs_quotactl = (vfs_quotactl_t)fs_nosys,
1938 .vfs_init_vnode = xfs_initialize_vnode,
1939 .vfs_force_shutdown = xfs_do_force_shutdown,
1940 .vfs_freeze = xfs_freeze,