1 /*
2  * XFS filesystem operations.
3  *
4  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
5  *
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.
9  *
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.
13  *
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.
20  *
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.
24  *
25  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
26  * Mountain View, CA  94043, or:
27  *
28  * http://www.sgi.com
29  *
30  * For further information regarding this notice, see:
31  *
32  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
33  */
34 
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"
72 
73 STATIC int xfs_sync(bhv_desc_t *, int, cred_t *);
74 
75 int
xfs_init(void)76 xfs_init(void)
77 {
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
87 
88 	/*
89 	 * Initialize all of the zone allocators we use.
90 	 */
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");
100 
101 	/*
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.
105 	 */
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");
122 
123 	/*
124 	 * Allocate global trace buffers.
125 	 */
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
144 
145 	xfs_dir_startup();
146 
147 #if (defined(DEBUG) || defined(INDUCE_IO_ERROR))
148 	xfs_error_test_init();
149 #endif /* DEBUG || INDUCE_IO_ERROR */
150 
151 	xfs_init_procfs();
152 	xfs_sysctl_register();
153 	return 0;
154 }
155 
156 void
xfs_cleanup(void)157 xfs_cleanup(void)
158 {
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;
169 
170 	xfs_cleanup_procfs();
171 	xfs_sysctl_unregister();
172 	xfs_refcache_destroy();
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_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
193 
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);
206 }
207 
208 /*
209  * xfs_start_flags
210  *
211  * This function fills in xfs_mount_t fields based on mount args.
212  * Note: the superblock has _not_ yet been read in.
213  */
214 STATIC int
xfs_start_flags(struct vfs * vfs,struct xfs_mount_args * ap,struct xfs_mount * mp)215 xfs_start_flags(
216 	struct vfs		*vfs,
217 	struct xfs_mount_args	*ap,
218 	struct xfs_mount	*mp)
219 {
220 	/* Values are in BBs */
221 	if ((ap->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
222 		/*
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.
227 		 */
228 		mp->m_dalign = ap->sunit;
229 		mp->m_swidth = ap->swidth;
230 	}
231 
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);
242 	}
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);
254 	}
255 	mp->m_logbsize = ap->logbufsize;
256 	mp->m_fsname_len = strlen(ap->fsname) + 1;
257 	mp->m_fsname = kmem_alloc(mp->m_fsname_len, KM_SLEEP);
258 	strcpy(mp->m_fsname, ap->fsname);
259 
260 	/*
261 	 * Pull in the 'wsync' and 'ino64' mount options before we do the real
262 	 * work of mounting and recovery.  The arg pointer will
263 	 * be NULL when we are being called from the root mount code.
264 	 */
265 	if (ap->flags & XFSMNT_WSYNC)
266 		mp->m_flags |= XFS_MOUNT_WSYNC;
267 #if XFS_BIG_INUMS
268 	if (ap->flags & XFSMNT_INO64) {
269 		mp->m_flags |= XFS_MOUNT_INO64;
270 		mp->m_inoadd = XFS_INO64_OFFSET;
271 	}
272 #endif
273 	if (ap->flags & XFSMNT_NOATIME)
274 		mp->m_flags |= XFS_MOUNT_NOATIME;
275 
276 	if (ap->flags & XFSMNT_RETERR)
277 		mp->m_flags |= XFS_MOUNT_RETERR;
278 
279 	if (ap->flags & XFSMNT_NOALIGN)
280 		mp->m_flags |= XFS_MOUNT_NOALIGN;
281 
282 	if (ap->flags & XFSMNT_SWALLOC)
283 		mp->m_flags |= XFS_MOUNT_SWALLOC;
284 
285 	if (ap->flags & XFSMNT_OSYNCISOSYNC)
286 		mp->m_flags |= XFS_MOUNT_OSYNCISOSYNC;
287 
288 	if (ap->flags & XFSMNT_32BITINODES)
289 		mp->m_flags |= (XFS_MOUNT_32BITINODES | XFS_MOUNT_32BITINOOPT);
290 
291 	if (ap->flags & XFSMNT_IOSIZE) {
292 		if (ap->iosizelog > XFS_MAX_IO_LOG ||
293 		    ap->iosizelog < XFS_MIN_IO_LOG) {
294 			cmn_err(CE_WARN,
295 		"XFS: invalid log iosize: %d [not %d-%d]",
296 				ap->iosizelog, XFS_MIN_IO_LOG,
297 				XFS_MAX_IO_LOG);
298 			return XFS_ERROR(EINVAL);
299 		}
300 
301 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
302 		mp->m_readio_log = mp->m_writeio_log = ap->iosizelog;
303 	}
304 	if (ap->flags & XFSMNT_IDELETE)
305 		mp->m_flags |= XFS_MOUNT_IDELETE;
306 
307 	/*
308 	 * no recovery flag requires a read-only mount
309 	 */
310 	if (ap->flags & XFSMNT_NORECOVERY) {
311 		if (!(vfs->vfs_flag & VFS_RDONLY)) {
312 			cmn_err(CE_WARN,
313 	"XFS: tried to mount a FS read-write without recovery!");
314 			return XFS_ERROR(EINVAL);
315 		}
316 		mp->m_flags |= XFS_MOUNT_NORECOVERY;
317 	}
318 
319 	if (ap->flags & XFSMNT_NOUUID)
320 		mp->m_flags |= XFS_MOUNT_NOUUID;
321 	if (ap->flags & XFSMNT_NOLOGFLUSH)
322 		mp->m_flags |= XFS_MOUNT_NOLOGFLUSH;
323 
324 	return 0;
325 }
326 
327 /*
328  * This function fills in xfs_mount_t fields based on mount args.
329  * Note: the superblock _has_ now been read in.
330  */
331 STATIC int
xfs_finish_flags(struct vfs * vfs,struct xfs_mount_args * ap,struct xfs_mount * mp)332 xfs_finish_flags(
333 	struct vfs		*vfs,
334 	struct xfs_mount_args	*ap,
335 	struct xfs_mount	*mp)
336 {
337 	int			ronly = (vfs->vfs_flag & VFS_RDONLY);
338 
339 	/* Fail a mount where the logbuf is smaller then the log stripe */
340 	if (XFS_SB_VERSION_HASLOGV2(&mp->m_sb)) {
341 		if ((ap->logbufsize == -1) &&
342 		    (mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE)) {
343 			mp->m_logbsize = mp->m_sb.sb_logsunit;
344 		} else if (ap->logbufsize < mp->m_sb.sb_logsunit) {
345 			cmn_err(CE_WARN,
346 	"XFS: logbuf size must be greater than or equal to log stripe size");
347 			return XFS_ERROR(EINVAL);
348 		}
349 	} else {
350 		/* Fail a mount if the logbuf is larger than 32K */
351 		if (ap->logbufsize > XLOG_BIG_RECORD_BSIZE) {
352 			cmn_err(CE_WARN,
353 	"XFS: logbuf size for version 1 logs must be 16K or 32K");
354 			return XFS_ERROR(EINVAL);
355 		}
356 	}
357 
358 	/*
359 	 * prohibit r/w mounts of read-only filesystems
360 	 */
361 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
362 		cmn_err(CE_WARN,
363 	"XFS: cannot mount a read-only filesystem as read-write");
364 		return XFS_ERROR(EROFS);
365 	}
366 
367 	/*
368 	 * disallow mount attempts with (IRIX) project quota enabled
369 	 */
370 	if (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
371 	    (mp->m_sb.sb_qflags & XFS_PQUOTA_ACCT)) {
372 		cmn_err(CE_WARN,
373 	"XFS: cannot mount a filesystem with IRIX project quota enabled");
374 		return XFS_ERROR(ENOSYS);
375 	}
376 
377 	/*
378 	 * check for shared mount.
379 	 */
380 	if (ap->flags & XFSMNT_SHARED) {
381 		if (!XFS_SB_VERSION_HASSHARED(&mp->m_sb))
382 			return XFS_ERROR(EINVAL);
383 
384 		/*
385 		 * For IRIX 6.5, shared mounts must have the shared
386 		 * version bit set, have the persistent readonly
387 		 * field set, must be version 0 and can only be mounted
388 		 * read-only.
389 		 */
390 		if (!ronly || !(mp->m_sb.sb_flags & XFS_SBF_READONLY) ||
391 		     (mp->m_sb.sb_shared_vn != 0))
392 			return XFS_ERROR(EINVAL);
393 
394 		mp->m_flags |= XFS_MOUNT_SHARED;
395 
396 		/*
397 		 * Shared XFS V0 can't deal with DMI.  Return EINVAL.
398 		 */
399 		if (mp->m_sb.sb_shared_vn == 0 && (ap->flags & XFSMNT_DMAPI))
400 			return XFS_ERROR(EINVAL);
401 	}
402 
403 	return 0;
404 }
405 
406 /*
407  * xfs_mount
408  *
409  * The file system configurations are:
410  *	(1) device (partition) with data and internal log
411  *	(2) logical volume with data and log subvolumes.
412  *	(3) logical volume with data, log, and realtime subvolumes.
413  *
414  * We only have to handle opening the log and realtime volumes here if
415  * they are present.  The data subvolume has already been opened by
416  * get_sb_bdev() and is stored in vfsp->vfs_super->s_bdev.
417  */
418 STATIC int
xfs_mount(struct bhv_desc * bhvp,struct xfs_mount_args * args,cred_t * credp)419 xfs_mount(
420 	struct bhv_desc		*bhvp,
421 	struct xfs_mount_args	*args,
422 	cred_t			*credp)
423 {
424 	struct vfs		*vfsp = bhvtovfs(bhvp);
425 	struct bhv_desc		*p;
426 	struct xfs_mount	*mp = XFS_BHVTOM(bhvp);
427 	struct block_device	*ddev, *logdev, *rtdev;
428 	int			flags = 0, error;
429 
430 	ddev = vfsp->vfs_super->s_bdev;
431 	logdev = rtdev = NULL;
432 
433 	/*
434 	 * Setup xfs_mount function vectors from available behaviors
435 	 */
436 	p = vfs_bhv_lookup(vfsp, VFS_POSITION_DM);
437 	mp->m_dm_ops = p ? *(xfs_dmops_t *) vfs_bhv_custom(p) : xfs_dmcore_stub;
438 	p = vfs_bhv_lookup(vfsp, VFS_POSITION_QM);
439 	mp->m_qm_ops = p ? *(xfs_qmops_t *) vfs_bhv_custom(p) : xfs_qmcore_stub;
440 	p = vfs_bhv_lookup(vfsp, VFS_POSITION_IO);
441 	mp->m_io_ops = p ? *(xfs_ioops_t *) vfs_bhv_custom(p) : xfs_iocore_xfs;
442 
443 	/*
444 	 * Open real time and log devices - order is important.
445 	 */
446 	if (args->logname[0]) {
447 		error = xfs_blkdev_get(mp, args->logname, &logdev);
448 		if (error)
449 			return error;
450 	}
451 	if (args->rtname[0]) {
452 		error = xfs_blkdev_get(mp, args->rtname, &rtdev);
453 		if (error) {
454 			xfs_blkdev_put(logdev);
455 			return error;
456 		}
457 
458 		if (rtdev == ddev || rtdev == logdev) {
459 			cmn_err(CE_WARN,
460 	"XFS: Cannot mount filesystem with identical rtdev and ddev/logdev.");
461 			xfs_blkdev_put(logdev);
462 			xfs_blkdev_put(rtdev);
463 			return EINVAL;
464 		}
465 	}
466 
467 	/*
468 	 * Setup xfs_mount buffer target pointers
469 	 */
470 	error = ENOMEM;
471 	mp->m_ddev_targp = xfs_alloc_buftarg(ddev);
472 	if (!mp->m_ddev_targp) {
473 		xfs_blkdev_put(logdev);
474 		xfs_blkdev_put(rtdev);
475 		return error;
476 	}
477 	if (rtdev) {
478 		mp->m_rtdev_targp = xfs_alloc_buftarg(rtdev);
479 		if (!mp->m_rtdev_targp)
480 			goto error0;
481 	}
482 	mp->m_logdev_targp = (logdev && logdev != ddev) ?
483 				xfs_alloc_buftarg(logdev) : mp->m_ddev_targp;
484 	if (!mp->m_logdev_targp)
485 		goto error0;
486 
487 	/*
488 	 * Setup flags based on mount(2) options and then the superblock
489 	 */
490 	error = xfs_start_flags(vfsp, args, mp);
491 	if (error)
492 		goto error1;
493 	error = xfs_readsb(mp);
494 	if (error)
495 		goto error1;
496 	error = xfs_finish_flags(vfsp, args, mp);
497 	if (error)
498 		goto error2;
499 
500 	/*
501 	 * Setup xfs_mount buffer target pointers based on superblock
502 	 */
503 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
504 				    mp->m_sb.sb_sectsize);
505 	if (!error && logdev && logdev != ddev) {
506 		unsigned int	log_sector_size = BBSIZE;
507 
508 		if (XFS_SB_VERSION_HASSECTOR(&mp->m_sb))
509 			log_sector_size = mp->m_sb.sb_logsectsize;
510 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
511 					    mp->m_sb.sb_blocksize,
512 					    log_sector_size);
513 	}
514 	if (!error && rtdev)
515 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
516 					    mp->m_sb.sb_blocksize,
517 					    mp->m_sb.sb_sectsize);
518 	if (error)
519 		goto error2;
520 
521 	error = XFS_IOINIT(vfsp, args, flags);
522 	if (!error)
523 		return 0;
524 error2:
525 	if (mp->m_sb_bp)
526 		xfs_freesb(mp);
527 error1:
528 	xfs_binval(mp->m_ddev_targp);
529 	if (logdev && logdev != ddev)
530 		xfs_binval(mp->m_logdev_targp);
531 	if (rtdev)
532 		xfs_binval(mp->m_rtdev_targp);
533 error0:
534 	xfs_unmountfs_close(mp, credp);
535 	return error;
536 }
537 
538 STATIC int
xfs_unmount(bhv_desc_t * bdp,int flags,cred_t * credp)539 xfs_unmount(
540 	bhv_desc_t	*bdp,
541 	int		flags,
542 	cred_t		*credp)
543 {
544 	struct vfs	*vfsp = bhvtovfs(bdp);
545 	xfs_mount_t	*mp = XFS_BHVTOM(bdp);
546 	xfs_inode_t	*rip;
547 	vnode_t		*rvp;
548 	int		unmount_event_wanted = 0;
549 	int		unmount_event_flags = 0;
550 	int		xfs_unmountfs_needed = 0;
551 	int		error;
552 
553 	rip = mp->m_rootip;
554 	rvp = XFS_ITOV(rip);
555 
556 	if (vfsp->vfs_flag & VFS_DMI) {
557 		error = XFS_SEND_PREUNMOUNT(mp, vfsp,
558 				rvp, DM_RIGHT_NULL, rvp, DM_RIGHT_NULL,
559 				NULL, NULL, 0, 0,
560 				(mp->m_dmevmask & (1<<DM_EVENT_PREUNMOUNT))?
561 					0:DM_FLAGS_UNWANTED);
562 			if (error)
563 				return XFS_ERROR(error);
564 		unmount_event_wanted = 1;
565 		unmount_event_flags = (mp->m_dmevmask & (1<<DM_EVENT_UNMOUNT))?
566 					0 : DM_FLAGS_UNWANTED;
567 	}
568 
569 	/*
570 	 * First blow any referenced inode from this file system
571 	 * out of the reference cache, and delete the timer.
572 	 */
573 	xfs_refcache_purge_mp(mp);
574 
575 	XFS_bflush(mp->m_ddev_targp);
576 	error = xfs_unmount_flush(mp, 0);
577 	if (error)
578 		goto out;
579 
580 	ASSERT(vn_count(rvp) == 1);
581 
582 	/*
583 	 * Drop the reference count
584 	 */
585 	VN_RELE(rvp);
586 
587 	/*
588 	 * If we're forcing a shutdown, typically because of a media error,
589 	 * we want to make sure we invalidate dirty pages that belong to
590 	 * referenced vnodes as well.
591 	 */
592 	if (XFS_FORCED_SHUTDOWN(mp)) {
593 		error = xfs_sync(&mp->m_bhv,
594 			 (SYNC_WAIT | SYNC_CLOSE), credp);
595 		ASSERT(error != EFSCORRUPTED);
596 	}
597 	xfs_unmountfs_needed = 1;
598 
599 out:
600 	/*	Send DMAPI event, if required.
601 	 *	Then do xfs_unmountfs() if needed.
602 	 *	Then return error (or zero).
603 	 */
604 	if (unmount_event_wanted) {
605 		/* Note: mp structure must still exist for
606 		 * XFS_SEND_UNMOUNT() call.
607 		 */
608 		XFS_SEND_UNMOUNT(mp, vfsp, error == 0 ? rvp : NULL,
609 			DM_RIGHT_NULL, 0, error, unmount_event_flags);
610 	}
611 	if (xfs_unmountfs_needed) {
612 		/*
613 		 * Call common unmount function to flush to disk
614 		 * and free the super block buffer & mount structures.
615 		 */
616 		xfs_unmountfs(mp, credp);
617 	}
618 
619 	return XFS_ERROR(error);
620 }
621 
622 #define REMOUNT_READONLY_FLAGS	(SYNC_REMOUNT|SYNC_ATTR|SYNC_WAIT)
623 
624 STATIC int
xfs_mntupdate(bhv_desc_t * bdp,int * flags,struct xfs_mount_args * args)625 xfs_mntupdate(
626 	bhv_desc_t			*bdp,
627 	int				*flags,
628 	struct xfs_mount_args		*args)
629 {
630 	struct vfs	*vfsp = bhvtovfs(bdp);
631 	xfs_mount_t	*mp = XFS_BHVTOM(bdp);
632 	int		pincount, error;
633 	int		count = 0;
634 
635 	if (args->flags & XFSMNT_NOATIME)
636 		mp->m_flags |= XFS_MOUNT_NOATIME;
637 	else
638 		mp->m_flags &= ~XFS_MOUNT_NOATIME;
639 
640 	if (!(vfsp->vfs_flag & VFS_RDONLY)) {
641 		VFS_SYNC(vfsp, SYNC_FSDATA|SYNC_BDFLUSH|SYNC_ATTR, NULL, error);
642 	}
643 
644 	if (*flags & MS_RDONLY) {
645 		xfs_refcache_purge_mp(mp);
646 		xfs_flush_buftarg(mp->m_ddev_targp, 0);
647 		xfs_finish_reclaim_all(mp, 0);
648 
649 		/* This loop must run at least twice.
650 		 * The first instance of the loop will flush
651 		 * most meta data but that will generate more
652 		 * meta data (typically directory updates).
653 		 * Which then must be flushed and logged before
654 		 * we can write the unmount record.
655 		 */
656 		do {
657 			VFS_SYNC(vfsp, REMOUNT_READONLY_FLAGS, NULL, error);
658 			pincount = xfs_flush_buftarg(mp->m_ddev_targp, 1);
659 			if (!pincount) {
660 				delay(50);
661 				count++;
662 			}
663 		} while (count < 2);
664 
665 		/* Ok now write out an unmount record */
666 		xfs_log_unmount_write(mp);
667 		xfs_unmountfs_writesb(mp);
668 		vfsp->vfs_flag |= VFS_RDONLY;
669 	} else {
670 		vfsp->vfs_flag &= ~VFS_RDONLY;
671 	}
672 
673 	return 0;
674 }
675 
676 /*
677  * xfs_unmount_flush implements a set of flush operation on special
678  * inodes, which are needed as a separate set of operations so that
679  * they can be called as part of relocation process.
680  */
681 int
xfs_unmount_flush(xfs_mount_t * mp,int relocation)682 xfs_unmount_flush(
683 	xfs_mount_t	*mp,		/* Mount structure we are getting
684 					   rid of. */
685 	int             relocation)	/* Called from vfs relocation. */
686 {
687 	xfs_inode_t	*rip = mp->m_rootip;
688 	xfs_inode_t	*rbmip;
689 	xfs_inode_t	*rsumip = NULL;
690 	vnode_t		*rvp = XFS_ITOV(rip);
691 	int		error;
692 
693 	xfs_ilock(rip, XFS_ILOCK_EXCL);
694 	xfs_iflock(rip);
695 
696 	/*
697 	 * Flush out the real time inodes.
698 	 */
699 	if ((rbmip = mp->m_rbmip) != NULL) {
700 		xfs_ilock(rbmip, XFS_ILOCK_EXCL);
701 		xfs_iflock(rbmip);
702 		error = xfs_iflush(rbmip, XFS_IFLUSH_SYNC);
703 		xfs_iunlock(rbmip, XFS_ILOCK_EXCL);
704 
705 		if (error == EFSCORRUPTED)
706 			goto fscorrupt_out;
707 
708 		ASSERT(vn_count(XFS_ITOV(rbmip)) == 1);
709 
710 		rsumip = mp->m_rsumip;
711 		xfs_ilock(rsumip, XFS_ILOCK_EXCL);
712 		xfs_iflock(rsumip);
713 		error = xfs_iflush(rsumip, XFS_IFLUSH_SYNC);
714 		xfs_iunlock(rsumip, XFS_ILOCK_EXCL);
715 
716 		if (error == EFSCORRUPTED)
717 			goto fscorrupt_out;
718 
719 		ASSERT(vn_count(XFS_ITOV(rsumip)) == 1);
720 	}
721 
722 	/*
723 	 * Synchronously flush root inode to disk
724 	 */
725 	error = xfs_iflush(rip, XFS_IFLUSH_SYNC);
726 	if (error == EFSCORRUPTED)
727 		goto fscorrupt_out2;
728 
729 	if (vn_count(rvp) != 1 && !relocation) {
730 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
731 		return XFS_ERROR(EBUSY);
732 	}
733 
734 	/*
735 	 * Release dquot that rootinode, rbmino and rsumino might be holding,
736 	 * flush and purge the quota inodes.
737 	 */
738 	error = XFS_QM_UNMOUNT(mp);
739 	if (error == EFSCORRUPTED)
740 		goto fscorrupt_out2;
741 
742 	if (rbmip) {
743 		VN_RELE(XFS_ITOV(rbmip));
744 		VN_RELE(XFS_ITOV(rsumip));
745 	}
746 
747 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
748 	return 0;
749 
750 fscorrupt_out:
751 	xfs_ifunlock(rip);
752 
753 fscorrupt_out2:
754 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
755 
756 	return XFS_ERROR(EFSCORRUPTED);
757 }
758 
759 /*
760  * xfs_root extracts the root vnode from a vfs.
761  *
762  * vfsp -- the vfs struct for the desired file system
763  * vpp  -- address of the caller's vnode pointer which should be
764  *         set to the desired fs root vnode
765  */
766 STATIC int
xfs_root(bhv_desc_t * bdp,vnode_t ** vpp)767 xfs_root(
768 	bhv_desc_t	*bdp,
769 	vnode_t		**vpp)
770 {
771 	vnode_t		*vp;
772 
773 	vp = XFS_ITOV((XFS_BHVTOM(bdp))->m_rootip);
774 	VN_HOLD(vp);
775 	*vpp = vp;
776 	return 0;
777 }
778 
779 /*
780  * xfs_statvfs
781  *
782  * Fill in the statvfs structure for the given file system.  We use
783  * the superblock lock in the mount structure to ensure a consistent
784  * snapshot of the counters returned.
785  */
786 STATIC int
xfs_statvfs(bhv_desc_t * bdp,xfs_statfs_t * statp,vnode_t * vp)787 xfs_statvfs(
788 	bhv_desc_t	*bdp,
789 	xfs_statfs_t	*statp,
790 	vnode_t		*vp)
791 {
792 	__uint64_t	fakeinos;
793 	xfs_extlen_t	lsize;
794 	xfs_mount_t	*mp;
795 	xfs_sb_t	*sbp;
796 	unsigned long	s;
797 
798 	mp = XFS_BHVTOM(bdp);
799 	sbp = &(mp->m_sb);
800 
801 	statp->f_type = XFS_SB_MAGIC;
802 
803 	s = XFS_SB_LOCK(mp);
804 	statp->f_bsize = sbp->sb_blocksize;
805 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
806 	statp->f_blocks = sbp->sb_dblocks - lsize;
807 	statp->f_bfree = statp->f_bavail = sbp->sb_fdblocks;
808 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
809 #if XFS_BIG_INUMS
810 	fakeinos += mp->m_inoadd;
811 #endif
812 	statp->f_files =
813 	    MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
814 	if (mp->m_maxicount)
815 #if XFS_BIG_INUMS
816 		if (!mp->m_inoadd)
817 #endif
818 			statp->f_files = min_t(typeof(statp->f_files),
819 						statp->f_files,
820 						mp->m_maxicount);
821 	statp->f_ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
822 	XFS_SB_UNLOCK(mp, s);
823 
824 	statp->f_fsid.val[0] = mp->m_dev;
825 	statp->f_fsid.val[1] = 0;
826 	statp->f_namelen = MAXNAMELEN - 1;
827 
828 	return 0;
829 }
830 
831 
832 /*
833  * xfs_sync flushes any pending I/O to file system vfsp.
834  *
835  * This routine is called by vfs_sync() to make sure that things make it
836  * out to disk eventually, on sync() system calls to flush out everything,
837  * and when the file system is unmounted.  For the vfs_sync() case, all
838  * we really need to do is sync out the log to make all of our meta-data
839  * updates permanent (except for timestamps).  For calls from pflushd(),
840  * dirty pages are kept moving by calling pdflush() on the inodes
841  * containing them.  We also flush the inodes that we can lock without
842  * sleeping and the superblock if we can lock it without sleeping from
843  * vfs_sync() so that items at the tail of the log are always moving out.
844  *
845  * Flags:
846  *      SYNC_BDFLUSH - We're being called from vfs_sync() so we don't want
847  *		       to sleep if we can help it.  All we really need
848  *		       to do is ensure that the log is synced at least
849  *		       periodically.  We also push the inodes and
850  *		       superblock if we can lock them without sleeping
851  *			and they are not pinned.
852  *      SYNC_ATTR    - We need to flush the inodes.  If SYNC_BDFLUSH is not
853  *		       set, then we really want to lock each inode and flush
854  *		       it.
855  *      SYNC_WAIT    - All the flushes that take place in this call should
856  *		       be synchronous.
857  *      SYNC_DELWRI  - This tells us to push dirty pages associated with
858  *		       inodes.  SYNC_WAIT and SYNC_BDFLUSH are used to
859  *		       determine if they should be flushed sync, async, or
860  *		       delwri.
861  *      SYNC_CLOSE   - This flag is passed when the system is being
862  *		       unmounted.  We should sync and invalidate everthing.
863  *      SYNC_FSDATA  - This indicates that the caller would like to make
864  *		       sure the superblock is safe on disk.  We can ensure
865  *		       this by simply makeing sure the log gets flushed
866  *		       if SYNC_BDFLUSH is set, and by actually writing it
867  *		       out otherwise.
868  *
869  */
870 /*ARGSUSED*/
871 STATIC int
xfs_sync(bhv_desc_t * bdp,int flags,cred_t * credp)872 xfs_sync(
873 	bhv_desc_t	*bdp,
874 	int		flags,
875 	cred_t		*credp)
876 {
877 	xfs_mount_t	*mp;
878 
879 	mp = XFS_BHVTOM(bdp);
880 	return (xfs_syncsub(mp, flags, 0, NULL));
881 }
882 
883 /*
884  * xfs sync routine for internal use
885  *
886  * This routine supports all of the flags defined for the generic VFS_SYNC
887  * interface as explained above under xfs_sync.  In the interests of not
888  * changing interfaces within the 6.5 family, additional internallly-
889  * required functions are specified within a separate xflags parameter,
890  * only available by calling this routine.
891  *
892  */
893 STATIC int
xfs_sync_inodes(xfs_mount_t * mp,int flags,int xflags,int * bypassed)894 xfs_sync_inodes(
895 	xfs_mount_t	*mp,
896 	int		flags,
897 	int             xflags,
898 	int             *bypassed)
899 {
900 	xfs_inode_t	*ip = NULL;
901 	xfs_inode_t	*ip_next;
902 	xfs_buf_t	*bp;
903 	vnode_t		*vp = NULL;
904 	vmap_t		vmap;
905 	int		error;
906 	int		last_error;
907 	uint64_t	fflag;
908 	uint		lock_flags;
909 	uint		base_lock_flags;
910 	boolean_t	mount_locked;
911 	boolean_t	vnode_refed;
912 	int		preempt;
913 	xfs_dinode_t	*dip;
914 	xfs_iptr_t	*ipointer;
915 #ifdef DEBUG
916 	boolean_t	ipointer_in = B_FALSE;
917 
918 #define IPOINTER_SET	ipointer_in = B_TRUE
919 #define IPOINTER_CLR	ipointer_in = B_FALSE
920 #else
921 #define IPOINTER_SET
922 #define IPOINTER_CLR
923 #endif
924 
925 
926 /* Insert a marker record into the inode list after inode ip. The list
927  * must be locked when this is called. After the call the list will no
928  * longer be locked.
929  */
930 #define IPOINTER_INSERT(ip, mp)	{ \
931 		ASSERT(ipointer_in == B_FALSE); \
932 		ipointer->ip_mnext = ip->i_mnext; \
933 		ipointer->ip_mprev = ip; \
934 		ip->i_mnext = (xfs_inode_t *)ipointer; \
935 		ipointer->ip_mnext->i_mprev = (xfs_inode_t *)ipointer; \
936 		preempt = 0; \
937 		XFS_MOUNT_IUNLOCK(mp); \
938 		mount_locked = B_FALSE; \
939 		IPOINTER_SET; \
940 	}
941 
942 /* Remove the marker from the inode list. If the marker was the only item
943  * in the list then there are no remaining inodes and we should zero out
944  * the whole list. If we are the current head of the list then move the head
945  * past us.
946  */
947 #define IPOINTER_REMOVE(ip, mp)	{ \
948 		ASSERT(ipointer_in == B_TRUE); \
949 		if (ipointer->ip_mnext != (xfs_inode_t *)ipointer) { \
950 			ip = ipointer->ip_mnext; \
951 			ip->i_mprev = ipointer->ip_mprev; \
952 			ipointer->ip_mprev->i_mnext = ip; \
953 			if (mp->m_inodes == (xfs_inode_t *)ipointer) { \
954 				mp->m_inodes = ip; \
955 			} \
956 		} else { \
957 			ASSERT(mp->m_inodes == (xfs_inode_t *)ipointer); \
958 			mp->m_inodes = NULL; \
959 			ip = NULL; \
960 		} \
961 		IPOINTER_CLR; \
962 	}
963 
964 #define XFS_PREEMPT_MASK	0x7f
965 
966 	if (bypassed)
967 		*bypassed = 0;
968 	if (XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY)
969 		return 0;
970 	error = 0;
971 	last_error = 0;
972 	preempt = 0;
973 
974 	/* Allocate a reference marker */
975 	ipointer = (xfs_iptr_t *)kmem_zalloc(sizeof(xfs_iptr_t), KM_SLEEP);
976 
977 	fflag = XFS_B_ASYNC;		/* default is don't wait */
978 	if (flags & SYNC_BDFLUSH)
979 		fflag = XFS_B_DELWRI;
980 	if (flags & SYNC_WAIT)
981 		fflag = 0;		/* synchronous overrides all */
982 
983 	base_lock_flags = XFS_ILOCK_SHARED;
984 	if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
985 		/*
986 		 * We need the I/O lock if we're going to call any of
987 		 * the flush/inval routines.
988 		 */
989 		base_lock_flags |= XFS_IOLOCK_SHARED;
990 	}
991 
992 	XFS_MOUNT_ILOCK(mp);
993 
994 	ip = mp->m_inodes;
995 
996 	mount_locked = B_TRUE;
997 	vnode_refed  = B_FALSE;
998 
999 	IPOINTER_CLR;
1000 
1001 	do {
1002 		ASSERT(ipointer_in == B_FALSE);
1003 		ASSERT(vnode_refed == B_FALSE);
1004 
1005 		lock_flags = base_lock_flags;
1006 
1007 		/*
1008 		 * There were no inodes in the list, just break out
1009 		 * of the loop.
1010 		 */
1011 		if (ip == NULL) {
1012 			break;
1013 		}
1014 
1015 		/*
1016 		 * We found another sync thread marker - skip it
1017 		 */
1018 		if (ip->i_mount == NULL) {
1019 			ip = ip->i_mnext;
1020 			continue;
1021 		}
1022 
1023 		vp = XFS_ITOV_NULL(ip);
1024 
1025 		/*
1026 		 * If the vnode is gone then this is being torn down,
1027 		 * call reclaim if it is flushed, else let regular flush
1028 		 * code deal with it later in the loop.
1029 		 */
1030 
1031 		if (vp == NULL) {
1032 			/* Skip ones already in reclaim */
1033 			if (ip->i_flags & XFS_IRECLAIM) {
1034 				ip = ip->i_mnext;
1035 				continue;
1036 			}
1037 			if (xfs_ilock_nowait(ip, XFS_ILOCK_EXCL) == 0) {
1038 				ip = ip->i_mnext;
1039 			} else if ((xfs_ipincount(ip) == 0) &&
1040 				    xfs_iflock_nowait(ip)) {
1041 				IPOINTER_INSERT(ip, mp);
1042 
1043 				xfs_finish_reclaim(ip, 1,
1044 						XFS_IFLUSH_DELWRI_ELSE_ASYNC);
1045 
1046 				XFS_MOUNT_ILOCK(mp);
1047 				mount_locked = B_TRUE;
1048 				IPOINTER_REMOVE(ip, mp);
1049 			} else {
1050 				xfs_iunlock(ip, XFS_ILOCK_EXCL);
1051 				ip = ip->i_mnext;
1052 			}
1053 			continue;
1054 		}
1055 
1056 		if (VN_BAD(vp)) {
1057 			ip = ip->i_mnext;
1058 			continue;
1059 		}
1060 
1061 		if (XFS_FORCED_SHUTDOWN(mp) && !(flags & SYNC_CLOSE)) {
1062 			XFS_MOUNT_IUNLOCK(mp);
1063 			kmem_free(ipointer, sizeof(xfs_iptr_t));
1064 			return 0;
1065 		}
1066 
1067 		/*
1068 		 * If this is just vfs_sync() or pflushd() calling
1069 		 * then we can skip inodes for which it looks like
1070 		 * there is nothing to do.  Since we don't have the
1071 		 * inode locked this is racey, but these are periodic
1072 		 * calls so it doesn't matter.  For the others we want
1073 		 * to know for sure, so we at least try to lock them.
1074 		 */
1075 		if (flags & SYNC_BDFLUSH) {
1076 			if (((ip->i_itemp == NULL) ||
1077 			     !(ip->i_itemp->ili_format.ilf_fields &
1078 			       XFS_ILOG_ALL)) &&
1079 			    (ip->i_update_core == 0)) {
1080 				ip = ip->i_mnext;
1081 				continue;
1082 			}
1083 		}
1084 
1085 		/*
1086 		 * Try to lock without sleeping.  We're out of order with
1087 		 * the inode list lock here, so if we fail we need to drop
1088 		 * the mount lock and try again.  If we're called from
1089 		 * bdflush() here, then don't bother.
1090 		 *
1091 		 * The inode lock here actually coordinates with the
1092 		 * almost spurious inode lock in xfs_ireclaim() to prevent
1093 		 * the vnode we handle here without a reference from
1094 		 * being freed while we reference it.  If we lock the inode
1095 		 * while it's on the mount list here, then the spurious inode
1096 		 * lock in xfs_ireclaim() after the inode is pulled from
1097 		 * the mount list will sleep until we release it here.
1098 		 * This keeps the vnode from being freed while we reference
1099 		 * it.  It is also cheaper and simpler than actually doing
1100 		 * a vn_get() for every inode we touch here.
1101 		 */
1102 		if (xfs_ilock_nowait(ip, lock_flags) == 0) {
1103 
1104 			if ((flags & SYNC_BDFLUSH) || (vp == NULL)) {
1105 				ip = ip->i_mnext;
1106 				continue;
1107 			}
1108 
1109 			/*
1110 			 * We need to unlock the inode list lock in order
1111 			 * to lock the inode. Insert a marker record into
1112 			 * the inode list to remember our position, dropping
1113 			 * the lock is now done inside the IPOINTER_INSERT
1114 			 * macro.
1115 			 *
1116 			 * We also use the inode list lock to protect us
1117 			 * in taking a snapshot of the vnode version number
1118 			 * for use in calling vn_get().
1119 			 */
1120 			VMAP(vp, vmap);
1121 			IPOINTER_INSERT(ip, mp);
1122 
1123 			vp = vn_get(vp, &vmap);
1124 			if (vp == NULL) {
1125 				/*
1126 				 * The vnode was reclaimed once we let go
1127 				 * of the inode list lock.  Skip to the
1128 				 * next list entry. Remove the marker.
1129 				 */
1130 
1131 				XFS_MOUNT_ILOCK(mp);
1132 
1133 				mount_locked = B_TRUE;
1134 				vnode_refed  = B_FALSE;
1135 
1136 				IPOINTER_REMOVE(ip, mp);
1137 
1138 				continue;
1139 			}
1140 
1141 			xfs_ilock(ip, lock_flags);
1142 
1143 			ASSERT(vp == XFS_ITOV(ip));
1144 			ASSERT(ip->i_mount == mp);
1145 
1146 			vnode_refed = B_TRUE;
1147 		}
1148 
1149 		/* From here on in the loop we may have a marker record
1150 		 * in the inode list.
1151 		 */
1152 
1153 		if ((flags & SYNC_CLOSE)  && (vp != NULL)) {
1154 			/*
1155 			 * This is the shutdown case.  We just need to
1156 			 * flush and invalidate all the pages associated
1157 			 * with the inode.  Drop the inode lock since
1158 			 * we can't hold it across calls to the buffer
1159 			 * cache.
1160 			 *
1161 			 * We don't set the VREMAPPING bit in the vnode
1162 			 * here, because we don't hold the vnode lock
1163 			 * exclusively.  It doesn't really matter, though,
1164 			 * because we only come here when we're shutting
1165 			 * down anyway.
1166 			 */
1167 			xfs_iunlock(ip, XFS_ILOCK_SHARED);
1168 
1169 			if (XFS_FORCED_SHUTDOWN(mp)) {
1170 				VOP_TOSS_PAGES(vp, 0, -1, FI_REMAPF);
1171 			} else {
1172 				VOP_FLUSHINVAL_PAGES(vp, 0, -1, FI_REMAPF);
1173 			}
1174 
1175 			xfs_ilock(ip, XFS_ILOCK_SHARED);
1176 
1177 		} else if ((flags & SYNC_DELWRI) && (vp != NULL)) {
1178 			if (VN_DIRTY(vp)) {
1179 				/* We need to have dropped the lock here,
1180 				 * so insert a marker if we have not already
1181 				 * done so.
1182 				 */
1183 				if (mount_locked) {
1184 					IPOINTER_INSERT(ip, mp);
1185 				}
1186 
1187 				/*
1188 				 * Drop the inode lock since we can't hold it
1189 				 * across calls to the buffer cache.
1190 				 */
1191 				xfs_iunlock(ip, XFS_ILOCK_SHARED);
1192 				VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
1193 							fflag, FI_NONE, error);
1194 				xfs_ilock(ip, XFS_ILOCK_SHARED);
1195 			}
1196 
1197 		}
1198 
1199 		if (flags & SYNC_BDFLUSH) {
1200 			if ((flags & SYNC_ATTR) &&
1201 			    ((ip->i_update_core) ||
1202 			     ((ip->i_itemp != NULL) &&
1203 			      (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1204 
1205 				/* Insert marker and drop lock if not already
1206 				 * done.
1207 				 */
1208 				if (mount_locked) {
1209 					IPOINTER_INSERT(ip, mp);
1210 				}
1211 
1212 				/*
1213 				 * We don't want the periodic flushing of the
1214 				 * inodes by vfs_sync() to interfere with
1215 				 * I/O to the file, especially read I/O
1216 				 * where it is only the access time stamp
1217 				 * that is being flushed out.  To prevent
1218 				 * long periods where we have both inode
1219 				 * locks held shared here while reading the
1220 				 * inode's buffer in from disk, we drop the
1221 				 * inode lock while reading in the inode
1222 				 * buffer.  We have to release the buffer
1223 				 * and reacquire the inode lock so that they
1224 				 * are acquired in the proper order (inode
1225 				 * locks first).  The buffer will go at the
1226 				 * end of the lru chain, though, so we can
1227 				 * expect it to still be there when we go
1228 				 * for it again in xfs_iflush().
1229 				 */
1230 				if ((xfs_ipincount(ip) == 0) &&
1231 				    xfs_iflock_nowait(ip)) {
1232 
1233 					xfs_ifunlock(ip);
1234 					xfs_iunlock(ip, XFS_ILOCK_SHARED);
1235 
1236 					error = xfs_itobp(mp, NULL, ip,
1237 							  &dip, &bp, 0);
1238 					if (!error) {
1239 						xfs_buf_relse(bp);
1240 					} else {
1241 						/* Bailing out, remove the
1242 						 * marker and free it.
1243 						 */
1244 						XFS_MOUNT_ILOCK(mp);
1245 
1246 						IPOINTER_REMOVE(ip, mp);
1247 
1248 						XFS_MOUNT_IUNLOCK(mp);
1249 
1250 						ASSERT(!(lock_flags &
1251 							XFS_IOLOCK_SHARED));
1252 
1253 						kmem_free(ipointer,
1254 							sizeof(xfs_iptr_t));
1255 						return (0);
1256 					}
1257 
1258 					/*
1259 					 * Since we dropped the inode lock,
1260 					 * the inode may have been reclaimed.
1261 					 * Therefore, we reacquire the mount
1262 					 * lock and check to see if we were the
1263 					 * inode reclaimed. If this happened
1264 					 * then the ipointer marker will no
1265 					 * longer point back at us. In this
1266 					 * case, move ip along to the inode
1267 					 * after the marker, remove the marker
1268 					 * and continue.
1269 					 */
1270 					XFS_MOUNT_ILOCK(mp);
1271 					mount_locked = B_TRUE;
1272 
1273 					if (ip != ipointer->ip_mprev) {
1274 						IPOINTER_REMOVE(ip, mp);
1275 
1276 						ASSERT(!vnode_refed);
1277 						ASSERT(!(lock_flags &
1278 							XFS_IOLOCK_SHARED));
1279 						continue;
1280 					}
1281 
1282 					ASSERT(ip->i_mount == mp);
1283 
1284 					if (xfs_ilock_nowait(ip,
1285 						    XFS_ILOCK_SHARED) == 0) {
1286 						ASSERT(ip->i_mount == mp);
1287 						/*
1288 						 * We failed to reacquire
1289 						 * the inode lock without
1290 						 * sleeping, so just skip
1291 						 * the inode for now.  We
1292 						 * clear the ILOCK bit from
1293 						 * the lock_flags so that we
1294 						 * won't try to drop a lock
1295 						 * we don't hold below.
1296 						 */
1297 						lock_flags &= ~XFS_ILOCK_SHARED;
1298 						IPOINTER_REMOVE(ip_next, mp);
1299 					} else if ((xfs_ipincount(ip) == 0) &&
1300 						   xfs_iflock_nowait(ip)) {
1301 						ASSERT(ip->i_mount == mp);
1302 						/*
1303 						 * Since this is vfs_sync()
1304 						 * calling we only flush the
1305 						 * inode out if we can lock
1306 						 * it without sleeping and
1307 						 * it is not pinned.  Drop
1308 						 * the mount lock here so
1309 						 * that we don't hold it for
1310 						 * too long. We already have
1311 						 * a marker in the list here.
1312 						 */
1313 						XFS_MOUNT_IUNLOCK(mp);
1314 						mount_locked = B_FALSE;
1315 						error = xfs_iflush(ip,
1316 							   XFS_IFLUSH_DELWRI);
1317 					} else {
1318 						ASSERT(ip->i_mount == mp);
1319 						IPOINTER_REMOVE(ip_next, mp);
1320 					}
1321 				}
1322 
1323 			}
1324 
1325 		} else {
1326 			if ((flags & SYNC_ATTR) &&
1327 			    ((ip->i_update_core) ||
1328 			     ((ip->i_itemp != NULL) &&
1329 			      (ip->i_itemp->ili_format.ilf_fields != 0)))) {
1330 				if (mount_locked) {
1331 					IPOINTER_INSERT(ip, mp);
1332 				}
1333 
1334 				if (flags & SYNC_WAIT) {
1335 					xfs_iflock(ip);
1336 					error = xfs_iflush(ip,
1337 							   XFS_IFLUSH_SYNC);
1338 				} else {
1339 					/*
1340 					 * If we can't acquire the flush
1341 					 * lock, then the inode is already
1342 					 * being flushed so don't bother
1343 					 * waiting.  If we can lock it then
1344 					 * do a delwri flush so we can
1345 					 * combine multiple inode flushes
1346 					 * in each disk write.
1347 					 */
1348 					if (xfs_iflock_nowait(ip)) {
1349 						error = xfs_iflush(ip,
1350 							   XFS_IFLUSH_DELWRI);
1351 					}
1352 					else if (bypassed)
1353 						(*bypassed)++;
1354 				}
1355 			}
1356 		}
1357 
1358 		if (lock_flags != 0) {
1359 			xfs_iunlock(ip, lock_flags);
1360 		}
1361 
1362 		if (vnode_refed) {
1363 			/*
1364 			 * If we had to take a reference on the vnode
1365 			 * above, then wait until after we've unlocked
1366 			 * the inode to release the reference.  This is
1367 			 * because we can be already holding the inode
1368 			 * lock when VN_RELE() calls xfs_inactive().
1369 			 *
1370 			 * Make sure to drop the mount lock before calling
1371 			 * VN_RELE() so that we don't trip over ourselves if
1372 			 * we have to go for the mount lock again in the
1373 			 * inactive code.
1374 			 */
1375 			if (mount_locked) {
1376 				IPOINTER_INSERT(ip, mp);
1377 			}
1378 
1379 			VN_RELE(vp);
1380 
1381 			vnode_refed = B_FALSE;
1382 		}
1383 
1384 		if (error) {
1385 			last_error = error;
1386 		}
1387 
1388 		/*
1389 		 * bail out if the filesystem is corrupted.
1390 		 */
1391 		if (error == EFSCORRUPTED)  {
1392 			if (!mount_locked) {
1393 				XFS_MOUNT_ILOCK(mp);
1394 				IPOINTER_REMOVE(ip, mp);
1395 			}
1396 			XFS_MOUNT_IUNLOCK(mp);
1397 			ASSERT(ipointer_in == B_FALSE);
1398 			kmem_free(ipointer, sizeof(xfs_iptr_t));
1399 			return XFS_ERROR(error);
1400 		}
1401 
1402 		/* Let other threads have a chance at the mount lock
1403 		 * if we have looped many times without dropping the
1404 		 * lock.
1405 		 */
1406 		if ((++preempt & XFS_PREEMPT_MASK) == 0) {
1407 			if (mount_locked) {
1408 				IPOINTER_INSERT(ip, mp);
1409 			}
1410 		}
1411 
1412 		if (mount_locked == B_FALSE) {
1413 			XFS_MOUNT_ILOCK(mp);
1414 			mount_locked = B_TRUE;
1415 			IPOINTER_REMOVE(ip, mp);
1416 			continue;
1417 		}
1418 
1419 		ASSERT(ipointer_in == B_FALSE);
1420 		ip = ip->i_mnext;
1421 
1422 	} while (ip != mp->m_inodes);
1423 
1424 	XFS_MOUNT_IUNLOCK(mp);
1425 
1426 	ASSERT(ipointer_in == B_FALSE);
1427 
1428 	kmem_free(ipointer, sizeof(xfs_iptr_t));
1429 	return XFS_ERROR(last_error);
1430 }
1431 
1432 /*
1433  * xfs sync routine for internal use
1434  *
1435  * This routine supports all of the flags defined for the generic VFS_SYNC
1436  * interface as explained above under xfs_sync.  In the interests of not
1437  * changing interfaces within the 6.5 family, additional internallly-
1438  * required functions are specified within a separate xflags parameter,
1439  * only available by calling this routine.
1440  *
1441  */
1442 int
xfs_syncsub(xfs_mount_t * mp,int flags,int xflags,int * bypassed)1443 xfs_syncsub(
1444 	xfs_mount_t	*mp,
1445 	int		flags,
1446 	int             xflags,
1447 	int             *bypassed)
1448 {
1449 	int		error = 0;
1450 	int		last_error = 0;
1451 	uint		log_flags = XFS_LOG_FORCE;
1452 	xfs_buf_t	*bp;
1453 	xfs_buf_log_item_t	*bip;
1454 
1455 	/*
1456 	 * Sync out the log.  This ensures that the log is periodically
1457 	 * flushed even if there is not enough activity to fill it up.
1458 	 */
1459 	if (flags & SYNC_WAIT)
1460 		log_flags |= XFS_LOG_SYNC;
1461 
1462 	xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1463 
1464 	if (flags & (SYNC_ATTR|SYNC_DELWRI)) {
1465 		if (flags & SYNC_BDFLUSH)
1466 			xfs_finish_reclaim_all(mp, 1);
1467 		else
1468 			error = xfs_sync_inodes(mp, flags, xflags, bypassed);
1469 	}
1470 
1471 	/*
1472 	 * Flushing out dirty data above probably generated more
1473 	 * log activity, so if this isn't vfs_sync() then flush
1474 	 * the log again.
1475 	 */
1476 	if (flags & SYNC_DELWRI) {
1477 		xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1478 	}
1479 
1480 	if (flags & SYNC_FSDATA) {
1481 		/*
1482 		 * If this is vfs_sync() then only sync the superblock
1483 		 * if we can lock it without sleeping and it is not pinned.
1484 		 */
1485 		if (flags & SYNC_BDFLUSH) {
1486 			bp = xfs_getsb(mp, XFS_BUF_TRYLOCK);
1487 			if (bp != NULL) {
1488 				bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
1489 				if ((bip != NULL) &&
1490 				    xfs_buf_item_dirty(bip)) {
1491 					if (!(XFS_BUF_ISPINNED(bp))) {
1492 						XFS_BUF_ASYNC(bp);
1493 						error = xfs_bwrite(mp, bp);
1494 					} else {
1495 						xfs_buf_relse(bp);
1496 					}
1497 				} else {
1498 					xfs_buf_relse(bp);
1499 				}
1500 			}
1501 		} else {
1502 			bp = xfs_getsb(mp, 0);
1503 			/*
1504 			 * If the buffer is pinned then push on the log so
1505 			 * we won't get stuck waiting in the write for
1506 			 * someone, maybe ourselves, to flush the log.
1507 			 * Even though we just pushed the log above, we
1508 			 * did not have the superblock buffer locked at
1509 			 * that point so it can become pinned in between
1510 			 * there and here.
1511 			 */
1512 			if (XFS_BUF_ISPINNED(bp))
1513 				xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE);
1514 			if (flags & SYNC_WAIT)
1515 				XFS_BUF_UNASYNC(bp);
1516 			else
1517 				XFS_BUF_ASYNC(bp);
1518 			error = xfs_bwrite(mp, bp);
1519 		}
1520 		if (error) {
1521 			last_error = error;
1522 		}
1523 	}
1524 
1525 	/*
1526 	 * If this is the periodic sync, then kick some entries out of
1527 	 * the reference cache.  This ensures that idle entries are
1528 	 * eventually kicked out of the cache.
1529 	 */
1530 	if (flags & SYNC_REFCACHE) {
1531 		if (flags & SYNC_WAIT)
1532 			xfs_refcache_purge_mp(mp);
1533 		else
1534 			xfs_refcache_purge_some(mp);
1535 	}
1536 
1537 	/*
1538 	 * Now check to see if the log needs a "dummy" transaction.
1539 	 */
1540 
1541 	if (!(flags & SYNC_REMOUNT) && xfs_log_need_covered(mp)) {
1542 		xfs_trans_t *tp;
1543 		xfs_inode_t *ip;
1544 
1545 		/*
1546 		 * Put a dummy transaction in the log to tell
1547 		 * recovery that all others are OK.
1548 		 */
1549 		tp = xfs_trans_alloc(mp, XFS_TRANS_DUMMY1);
1550 		if ((error = xfs_trans_reserve(tp, 0,
1551 				XFS_ICHANGE_LOG_RES(mp),
1552 				0, 0, 0)))  {
1553 			xfs_trans_cancel(tp, 0);
1554 			return error;
1555 		}
1556 
1557 		ip = mp->m_rootip;
1558 		xfs_ilock(ip, XFS_ILOCK_EXCL);
1559 
1560 		xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1561 		xfs_trans_ihold(tp, ip);
1562 		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1563 		error = xfs_trans_commit(tp, 0, NULL);
1564 		xfs_iunlock(ip, XFS_ILOCK_EXCL);
1565 		xfs_log_force(mp, (xfs_lsn_t)0, log_flags);
1566 	}
1567 
1568 	/*
1569 	 * When shutting down, we need to insure that the AIL is pushed
1570 	 * to disk or the filesystem can appear corrupt from the PROM.
1571 	 */
1572 	if ((flags & (SYNC_CLOSE|SYNC_WAIT)) == (SYNC_CLOSE|SYNC_WAIT)) {
1573 		XFS_bflush(mp->m_ddev_targp);
1574 		if (mp->m_rtdev_targp) {
1575 			XFS_bflush(mp->m_rtdev_targp);
1576 		}
1577 	}
1578 
1579 	return XFS_ERROR(last_error);
1580 }
1581 
1582 /*
1583  * xfs_vget - called by DMAPI and NFSD to get vnode from file handle
1584  */
1585 STATIC int
xfs_vget(bhv_desc_t * bdp,vnode_t ** vpp,fid_t * fidp)1586 xfs_vget(
1587 	bhv_desc_t	*bdp,
1588 	vnode_t		**vpp,
1589 	fid_t		*fidp)
1590 {
1591 	xfs_mount_t	*mp = XFS_BHVTOM(bdp);
1592 	xfs_fid_t	*xfid = (struct xfs_fid *)fidp;
1593 	xfs_inode_t	*ip;
1594 	int		error;
1595 	xfs_ino_t	ino;
1596 	unsigned int	igen;
1597 
1598 	/*
1599 	 * Invalid.  Since handles can be created in user space and passed in
1600 	 * via gethandle(), this is not cause for a panic.
1601 	 */
1602 	if (xfid->xfs_fid_len != sizeof(*xfid) - sizeof(xfid->xfs_fid_len))
1603 		return XFS_ERROR(EINVAL);
1604 
1605 	ino  = xfid->xfs_fid_ino;
1606 	igen = xfid->xfs_fid_gen;
1607 
1608 	/*
1609 	 * NFS can sometimes send requests for ino 0.  Fail them gracefully.
1610 	 */
1611 	if (ino == 0)
1612 		return XFS_ERROR(ESTALE);
1613 
1614 	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
1615 	if (error) {
1616 		*vpp = NULL;
1617 		return error;
1618 	}
1619 
1620 	if (ip == NULL) {
1621 		*vpp = NULL;
1622 		return XFS_ERROR(EIO);
1623 	}
1624 
1625 	if (ip->i_d.di_mode == 0 || ip->i_d.di_gen != igen) {
1626 		xfs_iput_new(ip, XFS_ILOCK_SHARED);
1627 		*vpp = NULL;
1628 		return XFS_ERROR(ENOENT);
1629 	}
1630 
1631 	*vpp = XFS_ITOV(ip);
1632 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
1633 	return 0;
1634 }
1635 
1636 
1637 #define MNTOPT_LOGBUFS	"logbufs"	/* number of XFS log buffers */
1638 #define MNTOPT_LOGBSIZE	"logbsize"	/* size of XFS log buffers */
1639 #define MNTOPT_LOGDEV	"logdev"	/* log device */
1640 #define MNTOPT_RTDEV	"rtdev"		/* realtime I/O device */
1641 #define MNTOPT_BIOSIZE	"biosize"	/* log2 of preferred buffered io size */
1642 #define MNTOPT_WSYNC	"wsync"		/* safe-mode nfs compatible mount */
1643 #define MNTOPT_INO64	"ino64"		/* force inodes into 64-bit range */
1644 #define MNTOPT_NOALIGN	"noalign"	/* turn off stripe alignment */
1645 #define MNTOPT_SWALLOC	"swalloc"	/* turn on stripe width allocation */
1646 #define MNTOPT_SUNIT	"sunit"		/* data volume stripe unit */
1647 #define MNTOPT_SWIDTH	"swidth"	/* data volume stripe width */
1648 #define MNTOPT_NOUUID	"nouuid"	/* ignore filesystem UUID */
1649 #define MNTOPT_MTPT	"mtpt"		/* filesystem mount point */
1650 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
1651 #define MNTOPT_NOLOGFLUSH   "nologflush"   /* don't hard flush on log writes */
1652 #define MNTOPT_OSYNCISOSYNC "osyncisosync" /* o_sync is REALLY o_sync */
1653 #define MNTOPT_64BITINODE   "inode64"	/* inodes can be allocated anywhere */
1654 #define MNTOPT_IKEEP	"ikeep"		/* do not free empty inode clusters */
1655 #define MNTOPT_NOIKEEP	"noikeep"	/* free empty inode clusters */
1656 
1657 
1658 int
xfs_parseargs(struct bhv_desc * bhv,char * options,struct xfs_mount_args * args,int update)1659 xfs_parseargs(
1660 	struct bhv_desc		*bhv,
1661 	char			*options,
1662 	struct xfs_mount_args	*args,
1663 	int			update)
1664 {
1665 	struct vfs		*vfsp = bhvtovfs(bhv);
1666 	char			*this_char, *value, *eov;
1667 	int			dsunit, dswidth, vol_dsunit, vol_dswidth;
1668 	int			iosize;
1669 
1670 #if 0	/* XXX: off by default, until some remaining issues ironed out */
1671 	args->flags |= XFSMNT_IDELETE; /* default to on */
1672 #endif
1673 
1674 	if (!options)
1675 		return 0;
1676 
1677 	iosize = dsunit = dswidth = vol_dsunit = vol_dswidth = 0;
1678 
1679 	while ((this_char = strsep(&options, ",")) != NULL) {
1680 		if (!*this_char)
1681 			continue;
1682 		if ((value = strchr(this_char, '=')) != NULL)
1683 			*value++ = 0;
1684 
1685 		if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
1686 			if (!value || !*value) {
1687 				printk("XFS: %s option requires an argument\n",
1688 					MNTOPT_LOGBUFS);
1689 				return EINVAL;
1690 			}
1691 			args->logbufs = simple_strtoul(value, &eov, 10);
1692 		} else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
1693 			int	last, in_kilobytes = 0;
1694 
1695 			if (!value || !*value) {
1696 				printk("XFS: %s option requires an argument\n",
1697 					MNTOPT_LOGBSIZE);
1698 				return EINVAL;
1699 			}
1700 			last = strlen(value) - 1;
1701 			if (value[last] == 'K' || value[last] == 'k') {
1702 				in_kilobytes = 1;
1703 				value[last] = '\0';
1704 			}
1705 			args->logbufsize = simple_strtoul(value, &eov, 10);
1706 			if (in_kilobytes)
1707 				args->logbufsize <<= 10;
1708 		} else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
1709 			if (!value || !*value) {
1710 				printk("XFS: %s option requires an argument\n",
1711 					MNTOPT_LOGDEV);
1712 				return EINVAL;
1713 			}
1714 			strncpy(args->logname, value, MAXNAMELEN);
1715 		} else if (!strcmp(this_char, MNTOPT_MTPT)) {
1716 			if (!value || !*value) {
1717 				printk("XFS: %s option requires an argument\n",
1718 					MNTOPT_MTPT);
1719 				return EINVAL;
1720 			}
1721 			strncpy(args->mtpt, value, MAXNAMELEN);
1722 		} else if (!strcmp(this_char, MNTOPT_RTDEV)) {
1723 			if (!value || !*value) {
1724 				printk("XFS: %s option requires an argument\n",
1725 					MNTOPT_RTDEV);
1726 				return EINVAL;
1727 			}
1728 			strncpy(args->rtname, value, MAXNAMELEN);
1729 		} else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
1730 			if (!value || !*value) {
1731 				printk("XFS: %s option requires an argument\n",
1732 					MNTOPT_BIOSIZE);
1733 				return EINVAL;
1734 			}
1735 			iosize = simple_strtoul(value, &eov, 10);
1736 			args->flags |= XFSMNT_IOSIZE;
1737 			args->iosizelog = (uint8_t) iosize;
1738 		} else if (!strcmp(this_char, MNTOPT_WSYNC)) {
1739 			args->flags |= XFSMNT_WSYNC;
1740 		} else if (!strcmp(this_char, MNTOPT_OSYNCISOSYNC)) {
1741 			args->flags |= XFSMNT_OSYNCISOSYNC;
1742 		} else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
1743 			args->flags |= XFSMNT_NORECOVERY;
1744 		} else if (!strcmp(this_char, MNTOPT_INO64)) {
1745 			args->flags |= XFSMNT_INO64;
1746 #if !XFS_BIG_INUMS
1747 			printk("XFS: %s option not allowed on this system\n",
1748 				MNTOPT_INO64);
1749 			return EINVAL;
1750 #endif
1751 		} else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
1752 			args->flags |= XFSMNT_NOALIGN;
1753 		} else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
1754 			args->flags |= XFSMNT_SWALLOC;
1755 		} else if (!strcmp(this_char, MNTOPT_SUNIT)) {
1756 			if (!value || !*value) {
1757 				printk("XFS: %s option requires an argument\n",
1758 					MNTOPT_SUNIT);
1759 				return EINVAL;
1760 			}
1761 			dsunit = simple_strtoul(value, &eov, 10);
1762 		} else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
1763 			if (!value || !*value) {
1764 				printk("XFS: %s option requires an argument\n",
1765 					MNTOPT_SWIDTH);
1766 				return EINVAL;
1767 			}
1768 			dswidth = simple_strtoul(value, &eov, 10);
1769 		} else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
1770 			args->flags &= ~XFSMNT_32BITINODES;
1771 #if !XFS_BIG_INUMS
1772 			printk("XFS: %s option not allowed on this system\n",
1773 				MNTOPT_64BITINODE);
1774 			return EINVAL;
1775 #endif
1776 		} else if (!strcmp(this_char, MNTOPT_NOUUID)) {
1777 			args->flags |= XFSMNT_NOUUID;
1778 		} else if (!strcmp(this_char, MNTOPT_NOLOGFLUSH)) {
1779 			args->flags |= XFSMNT_NOLOGFLUSH;
1780 		} else if (!strcmp(this_char, MNTOPT_IKEEP)) {
1781 			args->flags &= ~XFSMNT_IDELETE;
1782 		} else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
1783 			args->flags |= XFSMNT_IDELETE;
1784 		} else if (!strcmp(this_char, "osyncisdsync")) {
1785 			/* no-op, this is now the default */
1786 printk("XFS: osyncisdsync is now the default, option is deprecated.\n");
1787 		} else if (!strcmp(this_char, "irixsgid")) {
1788 printk("XFS: irixsgid is now a sysctl(2) variable, option is deprecated.\n");
1789 		} else {
1790 			printk("XFS: unknown mount option [%s].\n", this_char);
1791 			return EINVAL;
1792 		}
1793 	}
1794 
1795 	if (args->flags & XFSMNT_NORECOVERY) {
1796 		if ((vfsp->vfs_flag & VFS_RDONLY) == 0) {
1797 			printk("XFS: no-recovery mounts must be read-only.\n");
1798 			return EINVAL;
1799 		}
1800 	}
1801 
1802 	if ((args->flags & XFSMNT_NOALIGN) && (dsunit || dswidth)) {
1803 		printk(
1804 	"XFS: sunit and swidth options incompatible with the noalign option\n");
1805 		return EINVAL;
1806 	}
1807 
1808 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
1809 		printk("XFS: sunit and swidth must be specified together\n");
1810 		return EINVAL;
1811 	}
1812 
1813 	if (dsunit && (dswidth % dsunit != 0)) {
1814 		printk(
1815 	"XFS: stripe width (%d) must be a multiple of the stripe unit (%d)\n",
1816 			dswidth, dsunit);
1817 		return EINVAL;
1818 	}
1819 
1820 	if ((args->flags & XFSMNT_NOALIGN) != XFSMNT_NOALIGN) {
1821 		if (dsunit) {
1822 			args->sunit = dsunit;
1823 			args->flags |= XFSMNT_RETERR;
1824 		} else {
1825 			args->sunit = vol_dsunit;
1826 		}
1827 		dswidth ? (args->swidth = dswidth) :
1828 			  (args->swidth = vol_dswidth);
1829 	} else {
1830 		args->sunit = args->swidth = 0;
1831 	}
1832 
1833 	return 0;
1834 }
1835 
1836 int
xfs_showargs(struct bhv_desc * bhv,struct seq_file * m)1837 xfs_showargs(
1838 	struct bhv_desc		*bhv,
1839 	struct seq_file		*m)
1840 {
1841 	static struct proc_xfs_info {
1842 		int	flag;
1843 		char	*str;
1844 	} xfs_info[] = {
1845 		/* the few simple ones we can get from the mount struct */
1846 		{ XFS_MOUNT_WSYNC,		"," MNTOPT_WSYNC },
1847 		{ XFS_MOUNT_INO64,		"," MNTOPT_INO64 },
1848 		{ XFS_MOUNT_NOALIGN,		"," MNTOPT_NOALIGN },
1849 		{ XFS_MOUNT_SWALLOC,		"," MNTOPT_SWALLOC },
1850 		{ XFS_MOUNT_NOUUID,		"," MNTOPT_NOUUID },
1851 		{ XFS_MOUNT_NORECOVERY,		"," MNTOPT_NORECOVERY },
1852 		{ XFS_MOUNT_OSYNCISOSYNC,	"," MNTOPT_OSYNCISOSYNC },
1853 		{ XFS_MOUNT_NOLOGFLUSH,		"," MNTOPT_NOLOGFLUSH },
1854 		{ XFS_MOUNT_IDELETE,		"," MNTOPT_NOIKEEP },
1855 		{ 0, NULL }
1856 	};
1857 	struct proc_xfs_info	*xfs_infop;
1858 	struct xfs_mount	*mp = XFS_BHVTOM(bhv);
1859 
1860 	for (xfs_infop = xfs_info; xfs_infop->flag; xfs_infop++) {
1861 		if (mp->m_flags & xfs_infop->flag)
1862 			seq_puts(m, xfs_infop->str);
1863 	}
1864 
1865 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
1866 		seq_printf(m, "," MNTOPT_BIOSIZE "=%d", mp->m_writeio_log);
1867 
1868 	if (mp->m_logbufs > 0)
1869 		seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
1870 
1871 	if (mp->m_logbsize > 0)
1872 		seq_printf(m, "," MNTOPT_LOGBSIZE "=%d", mp->m_logbsize);
1873 
1874 	if (mp->m_ddev_targp != mp->m_logdev_targp)
1875 		seq_printf(m, "," MNTOPT_LOGDEV "=%s",
1876 				XFS_BUFTARG_NAME(mp->m_logdev_targp));
1877 
1878 	if (mp->m_rtdev_targp && mp->m_ddev_targp != mp->m_rtdev_targp)
1879 		seq_printf(m, "," MNTOPT_RTDEV "=%s",
1880 				XFS_BUFTARG_NAME(mp->m_rtdev_targp));
1881 
1882 	if (mp->m_dalign > 0)
1883 		seq_printf(m, "," MNTOPT_SUNIT "=%d",
1884 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
1885 
1886 	if (mp->m_swidth > 0)
1887 		seq_printf(m, "," MNTOPT_SWIDTH "=%d",
1888 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
1889 
1890 	if (!(mp->m_flags & XFS_MOUNT_32BITINOOPT))
1891 		seq_printf(m, "," MNTOPT_64BITINODE);
1892 
1893 	return 0;
1894 }
1895 
1896 STATIC void
xfs_freeze(bhv_desc_t * bdp)1897 xfs_freeze(
1898 	bhv_desc_t	*bdp)
1899 {
1900 	xfs_mount_t	*mp = XFS_BHVTOM(bdp);
1901 
1902 	while (atomic_read(&mp->m_active_trans) > 0)
1903 		delay(100);
1904 
1905 	/* Push the superblock and write an unmount record */
1906 	xfs_log_unmount_write(mp);
1907 	xfs_unmountfs_writesb(mp);
1908 }
1909 
1910 
1911 vfsops_t xfs_vfsops = {
1912 	BHV_IDENTITY_INIT(VFS_BHV_XFS,VFS_POSITION_XFS),
1913 	.vfs_parseargs		= xfs_parseargs,
1914 	.vfs_showargs		= xfs_showargs,
1915 	.vfs_mount		= xfs_mount,
1916 	.vfs_unmount		= xfs_unmount,
1917 	.vfs_mntupdate		= xfs_mntupdate,
1918 	.vfs_root		= xfs_root,
1919 	.vfs_statvfs		= xfs_statvfs,
1920 	.vfs_sync		= xfs_sync,
1921 	.vfs_vget		= xfs_vget,
1922 	.vfs_dmapiops		= (vfs_dmapiops_t)fs_nosys,
1923 	.vfs_quotactl		= (vfs_quotactl_t)fs_nosys,
1924 	.vfs_get_inode		= xfs_get_inode,
1925 	.vfs_init_vnode		= xfs_initialize_vnode,
1926 	.vfs_force_shutdown	= xfs_do_force_shutdown,
1927 	.vfs_freeze		= xfs_freeze,
1928 };
1929