xref: /openbmc/linux/fs/xfs/xfs_super.c (revision 6dfcd296)
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 
19 #include "xfs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_inode.h"
28 #include "xfs_btree.h"
29 #include "xfs_bmap.h"
30 #include "xfs_alloc.h"
31 #include "xfs_error.h"
32 #include "xfs_fsops.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_log.h"
36 #include "xfs_log_priv.h"
37 #include "xfs_da_btree.h"
38 #include "xfs_dir2.h"
39 #include "xfs_extfree_item.h"
40 #include "xfs_mru_cache.h"
41 #include "xfs_inode_item.h"
42 #include "xfs_icache.h"
43 #include "xfs_trace.h"
44 #include "xfs_icreate_item.h"
45 #include "xfs_filestream.h"
46 #include "xfs_quota.h"
47 #include "xfs_sysfs.h"
48 #include "xfs_ondisk.h"
49 #include "xfs_rmap_item.h"
50 #include "xfs_refcount_item.h"
51 #include "xfs_bmap_item.h"
52 #include "xfs_reflink.h"
53 
54 #include <linux/namei.h>
55 #include <linux/init.h>
56 #include <linux/slab.h>
57 #include <linux/mount.h>
58 #include <linux/mempool.h>
59 #include <linux/writeback.h>
60 #include <linux/kthread.h>
61 #include <linux/freezer.h>
62 #include <linux/parser.h>
63 
64 static const struct super_operations xfs_super_operations;
65 struct bio_set *xfs_ioend_bioset;
66 
67 static struct kset *xfs_kset;		/* top-level xfs sysfs dir */
68 #ifdef DEBUG
69 static struct xfs_kobj xfs_dbg_kobj;	/* global debug sysfs attrs */
70 #endif
71 
72 /*
73  * Table driven mount option parser.
74  */
75 enum {
76 	Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
77 	Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
78 	Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
79 	Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
80 	Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
81 	Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
82 	Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
83 	Opt_uquota, Opt_gquota, Opt_pquota,
84 	Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
85 	Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
86 };
87 
88 static const match_table_t tokens = {
89 	{Opt_logbufs,	"logbufs=%u"},	/* number of XFS log buffers */
90 	{Opt_logbsize,	"logbsize=%s"},	/* size of XFS log buffers */
91 	{Opt_logdev,	"logdev=%s"},	/* log device */
92 	{Opt_rtdev,	"rtdev=%s"},	/* realtime I/O device */
93 	{Opt_biosize,	"biosize=%u"},	/* log2 of preferred buffered io size */
94 	{Opt_wsync,	"wsync"},	/* safe-mode nfs compatible mount */
95 	{Opt_noalign,	"noalign"},	/* turn off stripe alignment */
96 	{Opt_swalloc,	"swalloc"},	/* turn on stripe width allocation */
97 	{Opt_sunit,	"sunit=%u"},	/* data volume stripe unit */
98 	{Opt_swidth,	"swidth=%u"},	/* data volume stripe width */
99 	{Opt_nouuid,	"nouuid"},	/* ignore filesystem UUID */
100 	{Opt_mtpt,	"mtpt"},	/* filesystem mount point */
101 	{Opt_grpid,	"grpid"},	/* group-ID from parent directory */
102 	{Opt_nogrpid,	"nogrpid"},	/* group-ID from current process */
103 	{Opt_bsdgroups,	"bsdgroups"},	/* group-ID from parent directory */
104 	{Opt_sysvgroups,"sysvgroups"},	/* group-ID from current process */
105 	{Opt_allocsize,	"allocsize=%s"},/* preferred allocation size */
106 	{Opt_norecovery,"norecovery"},	/* don't run XFS recovery */
107 	{Opt_barrier,	"barrier"},	/* use writer barriers for log write and
108 					 * unwritten extent conversion */
109 	{Opt_nobarrier,	"nobarrier"},	/* .. disable */
110 	{Opt_inode64,	"inode64"},	/* inodes can be allocated anywhere */
111 	{Opt_inode32,   "inode32"},	/* inode allocation limited to
112 					 * XFS_MAXINUMBER_32 */
113 	{Opt_ikeep,	"ikeep"},	/* do not free empty inode clusters */
114 	{Opt_noikeep,	"noikeep"},	/* free empty inode clusters */
115 	{Opt_largeio,	"largeio"},	/* report large I/O sizes in stat() */
116 	{Opt_nolargeio,	"nolargeio"},	/* do not report large I/O sizes
117 					 * in stat(). */
118 	{Opt_attr2,	"attr2"},	/* do use attr2 attribute format */
119 	{Opt_noattr2,	"noattr2"},	/* do not use attr2 attribute format */
120 	{Opt_filestreams,"filestreams"},/* use filestreams allocator */
121 	{Opt_quota,	"quota"},	/* disk quotas (user) */
122 	{Opt_noquota,	"noquota"},	/* no quotas */
123 	{Opt_usrquota,	"usrquota"},	/* user quota enabled */
124 	{Opt_grpquota,	"grpquota"},	/* group quota enabled */
125 	{Opt_prjquota,	"prjquota"},	/* project quota enabled */
126 	{Opt_uquota,	"uquota"},	/* user quota (IRIX variant) */
127 	{Opt_gquota,	"gquota"},	/* group quota (IRIX variant) */
128 	{Opt_pquota,	"pquota"},	/* project quota (IRIX variant) */
129 	{Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
130 	{Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
131 	{Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
132 	{Opt_qnoenforce, "qnoenforce"},	/* same as uqnoenforce */
133 	{Opt_discard,	"discard"},	/* Discard unused blocks */
134 	{Opt_nodiscard,	"nodiscard"},	/* Do not discard unused blocks */
135 
136 	{Opt_dax,	"dax"},		/* Enable direct access to bdev pages */
137 	{Opt_err,	NULL},
138 };
139 
140 
141 STATIC int
142 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
143 {
144 	int	last, shift_left_factor = 0, _res;
145 	char	*value;
146 	int	ret = 0;
147 
148 	value = match_strdup(s);
149 	if (!value)
150 		return -ENOMEM;
151 
152 	last = strlen(value) - 1;
153 	if (value[last] == 'K' || value[last] == 'k') {
154 		shift_left_factor = 10;
155 		value[last] = '\0';
156 	}
157 	if (value[last] == 'M' || value[last] == 'm') {
158 		shift_left_factor = 20;
159 		value[last] = '\0';
160 	}
161 	if (value[last] == 'G' || value[last] == 'g') {
162 		shift_left_factor = 30;
163 		value[last] = '\0';
164 	}
165 
166 	if (kstrtoint(value, base, &_res))
167 		ret = -EINVAL;
168 	kfree(value);
169 	*res = _res << shift_left_factor;
170 	return ret;
171 }
172 
173 /*
174  * This function fills in xfs_mount_t fields based on mount args.
175  * Note: the superblock has _not_ yet been read in.
176  *
177  * Note that this function leaks the various device name allocations on
178  * failure.  The caller takes care of them.
179  *
180  * *sb is const because this is also used to test options on the remount
181  * path, and we don't want this to have any side effects at remount time.
182  * Today this function does not change *sb, but just to future-proof...
183  */
184 STATIC int
185 xfs_parseargs(
186 	struct xfs_mount	*mp,
187 	char			*options)
188 {
189 	const struct super_block *sb = mp->m_super;
190 	char			*p;
191 	substring_t		args[MAX_OPT_ARGS];
192 	int			dsunit = 0;
193 	int			dswidth = 0;
194 	int			iosize = 0;
195 	__uint8_t		iosizelog = 0;
196 
197 	/*
198 	 * set up the mount name first so all the errors will refer to the
199 	 * correct device.
200 	 */
201 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
202 	if (!mp->m_fsname)
203 		return -ENOMEM;
204 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
205 
206 	/*
207 	 * Copy binary VFS mount flags we are interested in.
208 	 */
209 	if (sb->s_flags & MS_RDONLY)
210 		mp->m_flags |= XFS_MOUNT_RDONLY;
211 	if (sb->s_flags & MS_DIRSYNC)
212 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
213 	if (sb->s_flags & MS_SYNCHRONOUS)
214 		mp->m_flags |= XFS_MOUNT_WSYNC;
215 
216 	/*
217 	 * Set some default flags that could be cleared by the mount option
218 	 * parsing.
219 	 */
220 	mp->m_flags |= XFS_MOUNT_BARRIER;
221 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
222 
223 	/*
224 	 * These can be overridden by the mount option parsing.
225 	 */
226 	mp->m_logbufs = -1;
227 	mp->m_logbsize = -1;
228 
229 	if (!options)
230 		goto done;
231 
232 	while ((p = strsep(&options, ",")) != NULL) {
233 		int		token;
234 
235 		if (!*p)
236 			continue;
237 
238 		token = match_token(p, tokens, args);
239 		switch (token) {
240 		case Opt_logbufs:
241 			if (match_int(args, &mp->m_logbufs))
242 				return -EINVAL;
243 			break;
244 		case Opt_logbsize:
245 			if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
246 				return -EINVAL;
247 			break;
248 		case Opt_logdev:
249 			mp->m_logname = match_strdup(args);
250 			if (!mp->m_logname)
251 				return -ENOMEM;
252 			break;
253 		case Opt_mtpt:
254 			xfs_warn(mp, "%s option not allowed on this system", p);
255 			return -EINVAL;
256 		case Opt_rtdev:
257 			mp->m_rtname = match_strdup(args);
258 			if (!mp->m_rtname)
259 				return -ENOMEM;
260 			break;
261 		case Opt_allocsize:
262 		case Opt_biosize:
263 			if (suffix_kstrtoint(args, 10, &iosize))
264 				return -EINVAL;
265 			iosizelog = ffs(iosize) - 1;
266 			break;
267 		case Opt_grpid:
268 		case Opt_bsdgroups:
269 			mp->m_flags |= XFS_MOUNT_GRPID;
270 			break;
271 		case Opt_nogrpid:
272 		case Opt_sysvgroups:
273 			mp->m_flags &= ~XFS_MOUNT_GRPID;
274 			break;
275 		case Opt_wsync:
276 			mp->m_flags |= XFS_MOUNT_WSYNC;
277 			break;
278 		case Opt_norecovery:
279 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
280 			break;
281 		case Opt_noalign:
282 			mp->m_flags |= XFS_MOUNT_NOALIGN;
283 			break;
284 		case Opt_swalloc:
285 			mp->m_flags |= XFS_MOUNT_SWALLOC;
286 			break;
287 		case Opt_sunit:
288 			if (match_int(args, &dsunit))
289 				return -EINVAL;
290 			break;
291 		case Opt_swidth:
292 			if (match_int(args, &dswidth))
293 				return -EINVAL;
294 			break;
295 		case Opt_inode32:
296 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
297 			break;
298 		case Opt_inode64:
299 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
300 			break;
301 		case Opt_nouuid:
302 			mp->m_flags |= XFS_MOUNT_NOUUID;
303 			break;
304 		case Opt_barrier:
305 			mp->m_flags |= XFS_MOUNT_BARRIER;
306 			break;
307 		case Opt_nobarrier:
308 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
309 			break;
310 		case Opt_ikeep:
311 			mp->m_flags |= XFS_MOUNT_IKEEP;
312 			break;
313 		case Opt_noikeep:
314 			mp->m_flags &= ~XFS_MOUNT_IKEEP;
315 			break;
316 		case Opt_largeio:
317 			mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
318 			break;
319 		case Opt_nolargeio:
320 			mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
321 			break;
322 		case Opt_attr2:
323 			mp->m_flags |= XFS_MOUNT_ATTR2;
324 			break;
325 		case Opt_noattr2:
326 			mp->m_flags &= ~XFS_MOUNT_ATTR2;
327 			mp->m_flags |= XFS_MOUNT_NOATTR2;
328 			break;
329 		case Opt_filestreams:
330 			mp->m_flags |= XFS_MOUNT_FILESTREAMS;
331 			break;
332 		case Opt_noquota:
333 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
334 			mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
335 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
336 			break;
337 		case Opt_quota:
338 		case Opt_uquota:
339 		case Opt_usrquota:
340 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
341 					 XFS_UQUOTA_ENFD);
342 			break;
343 		case Opt_qnoenforce:
344 		case Opt_uqnoenforce:
345 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
346 			mp->m_qflags &= ~XFS_UQUOTA_ENFD;
347 			break;
348 		case Opt_pquota:
349 		case Opt_prjquota:
350 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
351 					 XFS_PQUOTA_ENFD);
352 			break;
353 		case Opt_pqnoenforce:
354 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
355 			mp->m_qflags &= ~XFS_PQUOTA_ENFD;
356 			break;
357 		case Opt_gquota:
358 		case Opt_grpquota:
359 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
360 					 XFS_GQUOTA_ENFD);
361 			break;
362 		case Opt_gqnoenforce:
363 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
364 			mp->m_qflags &= ~XFS_GQUOTA_ENFD;
365 			break;
366 		case Opt_discard:
367 			mp->m_flags |= XFS_MOUNT_DISCARD;
368 			break;
369 		case Opt_nodiscard:
370 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
371 			break;
372 #ifdef CONFIG_FS_DAX
373 		case Opt_dax:
374 			mp->m_flags |= XFS_MOUNT_DAX;
375 			break;
376 #endif
377 		default:
378 			xfs_warn(mp, "unknown mount option [%s].", p);
379 			return -EINVAL;
380 		}
381 	}
382 
383 	/*
384 	 * no recovery flag requires a read-only mount
385 	 */
386 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
387 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
388 		xfs_warn(mp, "no-recovery mounts must be read-only.");
389 		return -EINVAL;
390 	}
391 
392 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
393 		xfs_warn(mp,
394 	"sunit and swidth options incompatible with the noalign option");
395 		return -EINVAL;
396 	}
397 
398 #ifndef CONFIG_XFS_QUOTA
399 	if (XFS_IS_QUOTA_RUNNING(mp)) {
400 		xfs_warn(mp, "quota support not available in this kernel.");
401 		return -EINVAL;
402 	}
403 #endif
404 
405 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
406 		xfs_warn(mp, "sunit and swidth must be specified together");
407 		return -EINVAL;
408 	}
409 
410 	if (dsunit && (dswidth % dsunit != 0)) {
411 		xfs_warn(mp,
412 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
413 			dswidth, dsunit);
414 		return -EINVAL;
415 	}
416 
417 done:
418 	if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
419 		/*
420 		 * At this point the superblock has not been read
421 		 * in, therefore we do not know the block size.
422 		 * Before the mount call ends we will convert
423 		 * these to FSBs.
424 		 */
425 		mp->m_dalign = dsunit;
426 		mp->m_swidth = dswidth;
427 	}
428 
429 	if (mp->m_logbufs != -1 &&
430 	    mp->m_logbufs != 0 &&
431 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
432 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
433 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
434 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
435 		return -EINVAL;
436 	}
437 	if (mp->m_logbsize != -1 &&
438 	    mp->m_logbsize !=  0 &&
439 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
440 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
441 	     !is_power_of_2(mp->m_logbsize))) {
442 		xfs_warn(mp,
443 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
444 			mp->m_logbsize);
445 		return -EINVAL;
446 	}
447 
448 	if (iosizelog) {
449 		if (iosizelog > XFS_MAX_IO_LOG ||
450 		    iosizelog < XFS_MIN_IO_LOG) {
451 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
452 				iosizelog, XFS_MIN_IO_LOG,
453 				XFS_MAX_IO_LOG);
454 			return -EINVAL;
455 		}
456 
457 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
458 		mp->m_readio_log = iosizelog;
459 		mp->m_writeio_log = iosizelog;
460 	}
461 
462 	return 0;
463 }
464 
465 struct proc_xfs_info {
466 	uint64_t	flag;
467 	char		*str;
468 };
469 
470 STATIC int
471 xfs_showargs(
472 	struct xfs_mount	*mp,
473 	struct seq_file		*m)
474 {
475 	static struct proc_xfs_info xfs_info_set[] = {
476 		/* the few simple ones we can get from the mount struct */
477 		{ XFS_MOUNT_IKEEP,		",ikeep" },
478 		{ XFS_MOUNT_WSYNC,		",wsync" },
479 		{ XFS_MOUNT_NOALIGN,		",noalign" },
480 		{ XFS_MOUNT_SWALLOC,		",swalloc" },
481 		{ XFS_MOUNT_NOUUID,		",nouuid" },
482 		{ XFS_MOUNT_NORECOVERY,		",norecovery" },
483 		{ XFS_MOUNT_ATTR2,		",attr2" },
484 		{ XFS_MOUNT_FILESTREAMS,	",filestreams" },
485 		{ XFS_MOUNT_GRPID,		",grpid" },
486 		{ XFS_MOUNT_DISCARD,		",discard" },
487 		{ XFS_MOUNT_SMALL_INUMS,	",inode32" },
488 		{ XFS_MOUNT_DAX,		",dax" },
489 		{ 0, NULL }
490 	};
491 	static struct proc_xfs_info xfs_info_unset[] = {
492 		/* the few simple ones we can get from the mount struct */
493 		{ XFS_MOUNT_COMPAT_IOSIZE,	",largeio" },
494 		{ XFS_MOUNT_BARRIER,		",nobarrier" },
495 		{ XFS_MOUNT_SMALL_INUMS,	",inode64" },
496 		{ 0, NULL }
497 	};
498 	struct proc_xfs_info	*xfs_infop;
499 
500 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
501 		if (mp->m_flags & xfs_infop->flag)
502 			seq_puts(m, xfs_infop->str);
503 	}
504 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
505 		if (!(mp->m_flags & xfs_infop->flag))
506 			seq_puts(m, xfs_infop->str);
507 	}
508 
509 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
510 		seq_printf(m, ",allocsize=%dk",
511 				(int)(1 << mp->m_writeio_log) >> 10);
512 
513 	if (mp->m_logbufs > 0)
514 		seq_printf(m, ",logbufs=%d", mp->m_logbufs);
515 	if (mp->m_logbsize > 0)
516 		seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
517 
518 	if (mp->m_logname)
519 		seq_show_option(m, "logdev", mp->m_logname);
520 	if (mp->m_rtname)
521 		seq_show_option(m, "rtdev", mp->m_rtname);
522 
523 	if (mp->m_dalign > 0)
524 		seq_printf(m, ",sunit=%d",
525 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
526 	if (mp->m_swidth > 0)
527 		seq_printf(m, ",swidth=%d",
528 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
529 
530 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
531 		seq_puts(m, ",usrquota");
532 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
533 		seq_puts(m, ",uqnoenforce");
534 
535 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
536 		if (mp->m_qflags & XFS_PQUOTA_ENFD)
537 			seq_puts(m, ",prjquota");
538 		else
539 			seq_puts(m, ",pqnoenforce");
540 	}
541 	if (mp->m_qflags & XFS_GQUOTA_ACCT) {
542 		if (mp->m_qflags & XFS_GQUOTA_ENFD)
543 			seq_puts(m, ",grpquota");
544 		else
545 			seq_puts(m, ",gqnoenforce");
546 	}
547 
548 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
549 		seq_puts(m, ",noquota");
550 
551 	return 0;
552 }
553 static __uint64_t
554 xfs_max_file_offset(
555 	unsigned int		blockshift)
556 {
557 	unsigned int		pagefactor = 1;
558 	unsigned int		bitshift = BITS_PER_LONG - 1;
559 
560 	/* Figure out maximum filesize, on Linux this can depend on
561 	 * the filesystem blocksize (on 32 bit platforms).
562 	 * __block_write_begin does this in an [unsigned] long...
563 	 *      page->index << (PAGE_SHIFT - bbits)
564 	 * So, for page sized blocks (4K on 32 bit platforms),
565 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
566 	 *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
567 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
568 	 * Note1: get_block_t takes a long (implicit cast from above)
569 	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
570 	 * can optionally convert the [unsigned] long from above into
571 	 * an [unsigned] long long.
572 	 */
573 
574 #if BITS_PER_LONG == 32
575 # if defined(CONFIG_LBDAF)
576 	ASSERT(sizeof(sector_t) == 8);
577 	pagefactor = PAGE_SIZE;
578 	bitshift = BITS_PER_LONG;
579 # else
580 	pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
581 # endif
582 #endif
583 
584 	return (((__uint64_t)pagefactor) << bitshift) - 1;
585 }
586 
587 /*
588  * Set parameters for inode allocation heuristics, taking into account
589  * filesystem size and inode32/inode64 mount options; i.e. specifically
590  * whether or not XFS_MOUNT_SMALL_INUMS is set.
591  *
592  * Inode allocation patterns are altered only if inode32 is requested
593  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
594  * If altered, XFS_MOUNT_32BITINODES is set as well.
595  *
596  * An agcount independent of that in the mount structure is provided
597  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
598  * to the potentially higher ag count.
599  *
600  * Returns the maximum AG index which may contain inodes.
601  */
602 xfs_agnumber_t
603 xfs_set_inode_alloc(
604 	struct xfs_mount *mp,
605 	xfs_agnumber_t	agcount)
606 {
607 	xfs_agnumber_t	index;
608 	xfs_agnumber_t	maxagi = 0;
609 	xfs_sb_t	*sbp = &mp->m_sb;
610 	xfs_agnumber_t	max_metadata;
611 	xfs_agino_t	agino;
612 	xfs_ino_t	ino;
613 
614 	/*
615 	 * Calculate how much should be reserved for inodes to meet
616 	 * the max inode percentage.  Used only for inode32.
617 	 */
618 	if (mp->m_maxicount) {
619 		__uint64_t	icount;
620 
621 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
622 		do_div(icount, 100);
623 		icount += sbp->sb_agblocks - 1;
624 		do_div(icount, sbp->sb_agblocks);
625 		max_metadata = icount;
626 	} else {
627 		max_metadata = agcount;
628 	}
629 
630 	/* Get the last possible inode in the filesystem */
631 	agino =	XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
632 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
633 
634 	/*
635 	 * If user asked for no more than 32-bit inodes, and the fs is
636 	 * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
637 	 * the allocator to accommodate the request.
638 	 */
639 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
640 		mp->m_flags |= XFS_MOUNT_32BITINODES;
641 	else
642 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
643 
644 	for (index = 0; index < agcount; index++) {
645 		struct xfs_perag	*pag;
646 
647 		ino = XFS_AGINO_TO_INO(mp, index, agino);
648 
649 		pag = xfs_perag_get(mp, index);
650 
651 		if (mp->m_flags & XFS_MOUNT_32BITINODES) {
652 			if (ino > XFS_MAXINUMBER_32) {
653 				pag->pagi_inodeok = 0;
654 				pag->pagf_metadata = 0;
655 			} else {
656 				pag->pagi_inodeok = 1;
657 				maxagi++;
658 				if (index < max_metadata)
659 					pag->pagf_metadata = 1;
660 				else
661 					pag->pagf_metadata = 0;
662 			}
663 		} else {
664 			pag->pagi_inodeok = 1;
665 			pag->pagf_metadata = 0;
666 		}
667 
668 		xfs_perag_put(pag);
669 	}
670 
671 	return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
672 }
673 
674 STATIC int
675 xfs_blkdev_get(
676 	xfs_mount_t		*mp,
677 	const char		*name,
678 	struct block_device	**bdevp)
679 {
680 	int			error = 0;
681 
682 	*bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
683 				    mp);
684 	if (IS_ERR(*bdevp)) {
685 		error = PTR_ERR(*bdevp);
686 		xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
687 	}
688 
689 	return error;
690 }
691 
692 STATIC void
693 xfs_blkdev_put(
694 	struct block_device	*bdev)
695 {
696 	if (bdev)
697 		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
698 }
699 
700 void
701 xfs_blkdev_issue_flush(
702 	xfs_buftarg_t		*buftarg)
703 {
704 	blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
705 }
706 
707 STATIC void
708 xfs_close_devices(
709 	struct xfs_mount	*mp)
710 {
711 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
712 		struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
713 		xfs_free_buftarg(mp, mp->m_logdev_targp);
714 		xfs_blkdev_put(logdev);
715 	}
716 	if (mp->m_rtdev_targp) {
717 		struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
718 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
719 		xfs_blkdev_put(rtdev);
720 	}
721 	xfs_free_buftarg(mp, mp->m_ddev_targp);
722 }
723 
724 /*
725  * The file system configurations are:
726  *	(1) device (partition) with data and internal log
727  *	(2) logical volume with data and log subvolumes.
728  *	(3) logical volume with data, log, and realtime subvolumes.
729  *
730  * We only have to handle opening the log and realtime volumes here if
731  * they are present.  The data subvolume has already been opened by
732  * get_sb_bdev() and is stored in sb->s_bdev.
733  */
734 STATIC int
735 xfs_open_devices(
736 	struct xfs_mount	*mp)
737 {
738 	struct block_device	*ddev = mp->m_super->s_bdev;
739 	struct block_device	*logdev = NULL, *rtdev = NULL;
740 	int			error;
741 
742 	/*
743 	 * Open real time and log devices - order is important.
744 	 */
745 	if (mp->m_logname) {
746 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
747 		if (error)
748 			goto out;
749 	}
750 
751 	if (mp->m_rtname) {
752 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
753 		if (error)
754 			goto out_close_logdev;
755 
756 		if (rtdev == ddev || rtdev == logdev) {
757 			xfs_warn(mp,
758 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
759 			error = -EINVAL;
760 			goto out_close_rtdev;
761 		}
762 	}
763 
764 	/*
765 	 * Setup xfs_mount buffer target pointers
766 	 */
767 	error = -ENOMEM;
768 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
769 	if (!mp->m_ddev_targp)
770 		goto out_close_rtdev;
771 
772 	if (rtdev) {
773 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
774 		if (!mp->m_rtdev_targp)
775 			goto out_free_ddev_targ;
776 	}
777 
778 	if (logdev && logdev != ddev) {
779 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
780 		if (!mp->m_logdev_targp)
781 			goto out_free_rtdev_targ;
782 	} else {
783 		mp->m_logdev_targp = mp->m_ddev_targp;
784 	}
785 
786 	return 0;
787 
788  out_free_rtdev_targ:
789 	if (mp->m_rtdev_targp)
790 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
791  out_free_ddev_targ:
792 	xfs_free_buftarg(mp, mp->m_ddev_targp);
793  out_close_rtdev:
794 	xfs_blkdev_put(rtdev);
795  out_close_logdev:
796 	if (logdev && logdev != ddev)
797 		xfs_blkdev_put(logdev);
798  out:
799 	return error;
800 }
801 
802 /*
803  * Setup xfs_mount buffer target pointers based on superblock
804  */
805 STATIC int
806 xfs_setup_devices(
807 	struct xfs_mount	*mp)
808 {
809 	int			error;
810 
811 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
812 	if (error)
813 		return error;
814 
815 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
816 		unsigned int	log_sector_size = BBSIZE;
817 
818 		if (xfs_sb_version_hassector(&mp->m_sb))
819 			log_sector_size = mp->m_sb.sb_logsectsize;
820 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
821 					    log_sector_size);
822 		if (error)
823 			return error;
824 	}
825 	if (mp->m_rtdev_targp) {
826 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
827 					    mp->m_sb.sb_sectsize);
828 		if (error)
829 			return error;
830 	}
831 
832 	return 0;
833 }
834 
835 STATIC int
836 xfs_init_mount_workqueues(
837 	struct xfs_mount	*mp)
838 {
839 	mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
840 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
841 	if (!mp->m_buf_workqueue)
842 		goto out;
843 
844 	mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
845 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
846 	if (!mp->m_data_workqueue)
847 		goto out_destroy_buf;
848 
849 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
850 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
851 	if (!mp->m_unwritten_workqueue)
852 		goto out_destroy_data_iodone_queue;
853 
854 	mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
855 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
856 	if (!mp->m_cil_workqueue)
857 		goto out_destroy_unwritten;
858 
859 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
860 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
861 	if (!mp->m_reclaim_workqueue)
862 		goto out_destroy_cil;
863 
864 	mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
865 			WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
866 			mp->m_fsname);
867 	if (!mp->m_log_workqueue)
868 		goto out_destroy_reclaim;
869 
870 	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
871 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
872 	if (!mp->m_eofblocks_workqueue)
873 		goto out_destroy_log;
874 
875 	return 0;
876 
877 out_destroy_log:
878 	destroy_workqueue(mp->m_log_workqueue);
879 out_destroy_reclaim:
880 	destroy_workqueue(mp->m_reclaim_workqueue);
881 out_destroy_cil:
882 	destroy_workqueue(mp->m_cil_workqueue);
883 out_destroy_unwritten:
884 	destroy_workqueue(mp->m_unwritten_workqueue);
885 out_destroy_data_iodone_queue:
886 	destroy_workqueue(mp->m_data_workqueue);
887 out_destroy_buf:
888 	destroy_workqueue(mp->m_buf_workqueue);
889 out:
890 	return -ENOMEM;
891 }
892 
893 STATIC void
894 xfs_destroy_mount_workqueues(
895 	struct xfs_mount	*mp)
896 {
897 	destroy_workqueue(mp->m_eofblocks_workqueue);
898 	destroy_workqueue(mp->m_log_workqueue);
899 	destroy_workqueue(mp->m_reclaim_workqueue);
900 	destroy_workqueue(mp->m_cil_workqueue);
901 	destroy_workqueue(mp->m_data_workqueue);
902 	destroy_workqueue(mp->m_unwritten_workqueue);
903 	destroy_workqueue(mp->m_buf_workqueue);
904 }
905 
906 /*
907  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
908  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
909  * for IO to complete so that we effectively throttle multiple callers to the
910  * rate at which IO is completing.
911  */
912 void
913 xfs_flush_inodes(
914 	struct xfs_mount	*mp)
915 {
916 	struct super_block	*sb = mp->m_super;
917 
918 	if (down_read_trylock(&sb->s_umount)) {
919 		sync_inodes_sb(sb);
920 		up_read(&sb->s_umount);
921 	}
922 }
923 
924 /* Catch misguided souls that try to use this interface on XFS */
925 STATIC struct inode *
926 xfs_fs_alloc_inode(
927 	struct super_block	*sb)
928 {
929 	BUG();
930 	return NULL;
931 }
932 
933 /*
934  * Now that the generic code is guaranteed not to be accessing
935  * the linux inode, we can inactivate and reclaim the inode.
936  */
937 STATIC void
938 xfs_fs_destroy_inode(
939 	struct inode		*inode)
940 {
941 	struct xfs_inode	*ip = XFS_I(inode);
942 	int			error;
943 
944 	trace_xfs_destroy_inode(ip);
945 
946 	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
947 	XFS_STATS_INC(ip->i_mount, vn_rele);
948 	XFS_STATS_INC(ip->i_mount, vn_remove);
949 
950 	if (xfs_is_reflink_inode(ip)) {
951 		error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF);
952 		if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
953 			xfs_warn(ip->i_mount,
954 "Error %d while evicting CoW blocks for inode %llu.",
955 					error, ip->i_ino);
956 	}
957 
958 	xfs_inactive(ip);
959 
960 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
961 	XFS_STATS_INC(ip->i_mount, vn_reclaim);
962 
963 	/*
964 	 * We should never get here with one of the reclaim flags already set.
965 	 */
966 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
967 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
968 
969 	/*
970 	 * We always use background reclaim here because even if the
971 	 * inode is clean, it still may be under IO and hence we have
972 	 * to take the flush lock. The background reclaim path handles
973 	 * this more efficiently than we can here, so simply let background
974 	 * reclaim tear down all inodes.
975 	 */
976 	xfs_inode_set_reclaim_tag(ip);
977 }
978 
979 /*
980  * Slab object creation initialisation for the XFS inode.
981  * This covers only the idempotent fields in the XFS inode;
982  * all other fields need to be initialised on allocation
983  * from the slab. This avoids the need to repeatedly initialise
984  * fields in the xfs inode that left in the initialise state
985  * when freeing the inode.
986  */
987 STATIC void
988 xfs_fs_inode_init_once(
989 	void			*inode)
990 {
991 	struct xfs_inode	*ip = inode;
992 
993 	memset(ip, 0, sizeof(struct xfs_inode));
994 
995 	/* vfs inode */
996 	inode_init_once(VFS_I(ip));
997 
998 	/* xfs inode */
999 	atomic_set(&ip->i_pincount, 0);
1000 	spin_lock_init(&ip->i_flags_lock);
1001 
1002 	mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1003 		     "xfsino", ip->i_ino);
1004 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1005 		     "xfsino", ip->i_ino);
1006 }
1007 
1008 /*
1009  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1010  * serialised against cache hits here via the inode->i_lock and igrab() in
1011  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1012  * racing with us, and it avoids needing to grab a spinlock here for every inode
1013  * we drop the final reference on.
1014  */
1015 STATIC int
1016 xfs_fs_drop_inode(
1017 	struct inode		*inode)
1018 {
1019 	struct xfs_inode	*ip = XFS_I(inode);
1020 
1021 	/*
1022 	 * If this unlinked inode is in the middle of recovery, don't
1023 	 * drop the inode just yet; log recovery will take care of
1024 	 * that.  See the comment for this inode flag.
1025 	 */
1026 	if (ip->i_flags & XFS_IRECOVERY) {
1027 		ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1028 		return 0;
1029 	}
1030 
1031 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1032 }
1033 
1034 STATIC void
1035 xfs_free_fsname(
1036 	struct xfs_mount	*mp)
1037 {
1038 	kfree(mp->m_fsname);
1039 	kfree(mp->m_rtname);
1040 	kfree(mp->m_logname);
1041 }
1042 
1043 STATIC int
1044 xfs_fs_sync_fs(
1045 	struct super_block	*sb,
1046 	int			wait)
1047 {
1048 	struct xfs_mount	*mp = XFS_M(sb);
1049 
1050 	/*
1051 	 * Doing anything during the async pass would be counterproductive.
1052 	 */
1053 	if (!wait)
1054 		return 0;
1055 
1056 	xfs_log_force(mp, XFS_LOG_SYNC);
1057 	if (laptop_mode) {
1058 		/*
1059 		 * The disk must be active because we're syncing.
1060 		 * We schedule log work now (now that the disk is
1061 		 * active) instead of later (when it might not be).
1062 		 */
1063 		flush_delayed_work(&mp->m_log->l_work);
1064 	}
1065 
1066 	return 0;
1067 }
1068 
1069 STATIC int
1070 xfs_fs_statfs(
1071 	struct dentry		*dentry,
1072 	struct kstatfs		*statp)
1073 {
1074 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1075 	xfs_sb_t		*sbp = &mp->m_sb;
1076 	struct xfs_inode	*ip = XFS_I(d_inode(dentry));
1077 	__uint64_t		fakeinos, id;
1078 	__uint64_t		icount;
1079 	__uint64_t		ifree;
1080 	__uint64_t		fdblocks;
1081 	xfs_extlen_t		lsize;
1082 	__int64_t		ffree;
1083 
1084 	statp->f_type = XFS_SB_MAGIC;
1085 	statp->f_namelen = MAXNAMELEN - 1;
1086 
1087 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1088 	statp->f_fsid.val[0] = (u32)id;
1089 	statp->f_fsid.val[1] = (u32)(id >> 32);
1090 
1091 	icount = percpu_counter_sum(&mp->m_icount);
1092 	ifree = percpu_counter_sum(&mp->m_ifree);
1093 	fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1094 
1095 	spin_lock(&mp->m_sb_lock);
1096 	statp->f_bsize = sbp->sb_blocksize;
1097 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1098 	statp->f_blocks = sbp->sb_dblocks - lsize;
1099 	spin_unlock(&mp->m_sb_lock);
1100 
1101 	statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1102 	statp->f_bavail = statp->f_bfree;
1103 
1104 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
1105 	statp->f_files = MIN(icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1106 	if (mp->m_maxicount)
1107 		statp->f_files = min_t(typeof(statp->f_files),
1108 					statp->f_files,
1109 					mp->m_maxicount);
1110 
1111 	/* If sb_icount overshot maxicount, report actual allocation */
1112 	statp->f_files = max_t(typeof(statp->f_files),
1113 					statp->f_files,
1114 					sbp->sb_icount);
1115 
1116 	/* make sure statp->f_ffree does not underflow */
1117 	ffree = statp->f_files - (icount - ifree);
1118 	statp->f_ffree = max_t(__int64_t, ffree, 0);
1119 
1120 
1121 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1122 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1123 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1124 		xfs_qm_statvfs(ip, statp);
1125 	return 0;
1126 }
1127 
1128 STATIC void
1129 xfs_save_resvblks(struct xfs_mount *mp)
1130 {
1131 	__uint64_t resblks = 0;
1132 
1133 	mp->m_resblks_save = mp->m_resblks;
1134 	xfs_reserve_blocks(mp, &resblks, NULL);
1135 }
1136 
1137 STATIC void
1138 xfs_restore_resvblks(struct xfs_mount *mp)
1139 {
1140 	__uint64_t resblks;
1141 
1142 	if (mp->m_resblks_save) {
1143 		resblks = mp->m_resblks_save;
1144 		mp->m_resblks_save = 0;
1145 	} else
1146 		resblks = xfs_default_resblks(mp);
1147 
1148 	xfs_reserve_blocks(mp, &resblks, NULL);
1149 }
1150 
1151 /*
1152  * Trigger writeback of all the dirty metadata in the file system.
1153  *
1154  * This ensures that the metadata is written to their location on disk rather
1155  * than just existing in transactions in the log. This means after a quiesce
1156  * there is no log replay required to write the inodes to disk - this is the
1157  * primary difference between a sync and a quiesce.
1158  *
1159  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1160  * it is started again when appropriate.
1161  */
1162 void
1163 xfs_quiesce_attr(
1164 	struct xfs_mount	*mp)
1165 {
1166 	int	error = 0;
1167 
1168 	/* wait for all modifications to complete */
1169 	while (atomic_read(&mp->m_active_trans) > 0)
1170 		delay(100);
1171 
1172 	/* force the log to unpin objects from the now complete transactions */
1173 	xfs_log_force(mp, XFS_LOG_SYNC);
1174 
1175 	/* reclaim inodes to do any IO before the freeze completes */
1176 	xfs_reclaim_inodes(mp, 0);
1177 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1178 
1179 	/* Push the superblock and write an unmount record */
1180 	error = xfs_log_sbcount(mp);
1181 	if (error)
1182 		xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1183 				"Frozen image may not be consistent.");
1184 	/*
1185 	 * Just warn here till VFS can correctly support
1186 	 * read-only remount without racing.
1187 	 */
1188 	WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1189 
1190 	xfs_log_quiesce(mp);
1191 }
1192 
1193 STATIC int
1194 xfs_test_remount_options(
1195 	struct super_block	*sb,
1196 	struct xfs_mount	*mp,
1197 	char			*options)
1198 {
1199 	int			error = 0;
1200 	struct xfs_mount	*tmp_mp;
1201 
1202 	tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1203 	if (!tmp_mp)
1204 		return -ENOMEM;
1205 
1206 	tmp_mp->m_super = sb;
1207 	error = xfs_parseargs(tmp_mp, options);
1208 	xfs_free_fsname(tmp_mp);
1209 	kfree(tmp_mp);
1210 
1211 	return error;
1212 }
1213 
1214 STATIC int
1215 xfs_fs_remount(
1216 	struct super_block	*sb,
1217 	int			*flags,
1218 	char			*options)
1219 {
1220 	struct xfs_mount	*mp = XFS_M(sb);
1221 	xfs_sb_t		*sbp = &mp->m_sb;
1222 	substring_t		args[MAX_OPT_ARGS];
1223 	char			*p;
1224 	int			error;
1225 
1226 	/* First, check for complete junk; i.e. invalid options */
1227 	error = xfs_test_remount_options(sb, mp, options);
1228 	if (error)
1229 		return error;
1230 
1231 	sync_filesystem(sb);
1232 	while ((p = strsep(&options, ",")) != NULL) {
1233 		int token;
1234 
1235 		if (!*p)
1236 			continue;
1237 
1238 		token = match_token(p, tokens, args);
1239 		switch (token) {
1240 		case Opt_barrier:
1241 			mp->m_flags |= XFS_MOUNT_BARRIER;
1242 			break;
1243 		case Opt_nobarrier:
1244 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
1245 			break;
1246 		case Opt_inode64:
1247 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1248 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1249 			break;
1250 		case Opt_inode32:
1251 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1252 			mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1253 			break;
1254 		default:
1255 			/*
1256 			 * Logically we would return an error here to prevent
1257 			 * users from believing they might have changed
1258 			 * mount options using remount which can't be changed.
1259 			 *
1260 			 * But unfortunately mount(8) adds all options from
1261 			 * mtab and fstab to the mount arguments in some cases
1262 			 * so we can't blindly reject options, but have to
1263 			 * check for each specified option if it actually
1264 			 * differs from the currently set option and only
1265 			 * reject it if that's the case.
1266 			 *
1267 			 * Until that is implemented we return success for
1268 			 * every remount request, and silently ignore all
1269 			 * options that we can't actually change.
1270 			 */
1271 #if 0
1272 			xfs_info(mp,
1273 		"mount option \"%s\" not supported for remount", p);
1274 			return -EINVAL;
1275 #else
1276 			break;
1277 #endif
1278 		}
1279 	}
1280 
1281 	/* ro -> rw */
1282 	if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1283 		if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1284 			xfs_warn(mp,
1285 		"ro->rw transition prohibited on norecovery mount");
1286 			return -EINVAL;
1287 		}
1288 
1289 		if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1290 		    xfs_sb_has_ro_compat_feature(sbp,
1291 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1292 			xfs_warn(mp,
1293 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1294 				(sbp->sb_features_ro_compat &
1295 					XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1296 			return -EINVAL;
1297 		}
1298 
1299 		mp->m_flags &= ~XFS_MOUNT_RDONLY;
1300 
1301 		/*
1302 		 * If this is the first remount to writeable state we
1303 		 * might have some superblock changes to update.
1304 		 */
1305 		if (mp->m_update_sb) {
1306 			error = xfs_sync_sb(mp, false);
1307 			if (error) {
1308 				xfs_warn(mp, "failed to write sb changes");
1309 				return error;
1310 			}
1311 			mp->m_update_sb = false;
1312 		}
1313 
1314 		/*
1315 		 * Fill out the reserve pool if it is empty. Use the stashed
1316 		 * value if it is non-zero, otherwise go with the default.
1317 		 */
1318 		xfs_restore_resvblks(mp);
1319 		xfs_log_work_queue(mp);
1320 		xfs_queue_eofblocks(mp);
1321 
1322 		/* Recover any CoW blocks that never got remapped. */
1323 		error = xfs_reflink_recover_cow(mp);
1324 		if (error) {
1325 			xfs_err(mp,
1326 	"Error %d recovering leftover CoW allocations.", error);
1327 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1328 			return error;
1329 		}
1330 
1331 		/* Create the per-AG metadata reservation pool .*/
1332 		error = xfs_fs_reserve_ag_blocks(mp);
1333 		if (error && error != -ENOSPC)
1334 			return error;
1335 	}
1336 
1337 	/* rw -> ro */
1338 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1339 		/* Free the per-AG metadata reservation pool. */
1340 		error = xfs_fs_unreserve_ag_blocks(mp);
1341 		if (error) {
1342 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1343 			return error;
1344 		}
1345 
1346 		/*
1347 		 * Before we sync the metadata, we need to free up the reserve
1348 		 * block pool so that the used block count in the superblock on
1349 		 * disk is correct at the end of the remount. Stash the current
1350 		 * reserve pool size so that if we get remounted rw, we can
1351 		 * return it to the same size.
1352 		 */
1353 		xfs_save_resvblks(mp);
1354 
1355 		/*
1356 		 * Cancel background eofb scanning so it cannot race with the
1357 		 * final log force+buftarg wait and deadlock the remount.
1358 		 */
1359 		cancel_delayed_work_sync(&mp->m_eofblocks_work);
1360 
1361 		xfs_quiesce_attr(mp);
1362 		mp->m_flags |= XFS_MOUNT_RDONLY;
1363 	}
1364 
1365 	return 0;
1366 }
1367 
1368 /*
1369  * Second stage of a freeze. The data is already frozen so we only
1370  * need to take care of the metadata. Once that's done sync the superblock
1371  * to the log to dirty it in case of a crash while frozen. This ensures that we
1372  * will recover the unlinked inode lists on the next mount.
1373  */
1374 STATIC int
1375 xfs_fs_freeze(
1376 	struct super_block	*sb)
1377 {
1378 	struct xfs_mount	*mp = XFS_M(sb);
1379 
1380 	xfs_save_resvblks(mp);
1381 	xfs_quiesce_attr(mp);
1382 	return xfs_sync_sb(mp, true);
1383 }
1384 
1385 STATIC int
1386 xfs_fs_unfreeze(
1387 	struct super_block	*sb)
1388 {
1389 	struct xfs_mount	*mp = XFS_M(sb);
1390 
1391 	xfs_restore_resvblks(mp);
1392 	xfs_log_work_queue(mp);
1393 	return 0;
1394 }
1395 
1396 STATIC int
1397 xfs_fs_show_options(
1398 	struct seq_file		*m,
1399 	struct dentry		*root)
1400 {
1401 	return xfs_showargs(XFS_M(root->d_sb), m);
1402 }
1403 
1404 /*
1405  * This function fills in xfs_mount_t fields based on mount args.
1406  * Note: the superblock _has_ now been read in.
1407  */
1408 STATIC int
1409 xfs_finish_flags(
1410 	struct xfs_mount	*mp)
1411 {
1412 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1413 
1414 	/* Fail a mount where the logbuf is smaller than the log stripe */
1415 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1416 		if (mp->m_logbsize <= 0 &&
1417 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1418 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1419 		} else if (mp->m_logbsize > 0 &&
1420 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1421 			xfs_warn(mp,
1422 		"logbuf size must be greater than or equal to log stripe size");
1423 			return -EINVAL;
1424 		}
1425 	} else {
1426 		/* Fail a mount if the logbuf is larger than 32K */
1427 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1428 			xfs_warn(mp,
1429 		"logbuf size for version 1 logs must be 16K or 32K");
1430 			return -EINVAL;
1431 		}
1432 	}
1433 
1434 	/*
1435 	 * V5 filesystems always use attr2 format for attributes.
1436 	 */
1437 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1438 	    (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1439 		xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1440 			     "attr2 is always enabled for V5 filesystems.");
1441 		return -EINVAL;
1442 	}
1443 
1444 	/*
1445 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1446 	 * told by noattr2 to turn it off
1447 	 */
1448 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1449 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1450 		mp->m_flags |= XFS_MOUNT_ATTR2;
1451 
1452 	/*
1453 	 * prohibit r/w mounts of read-only filesystems
1454 	 */
1455 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1456 		xfs_warn(mp,
1457 			"cannot mount a read-only filesystem as read-write");
1458 		return -EROFS;
1459 	}
1460 
1461 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1462 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1463 	    !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1464 		xfs_warn(mp,
1465 		  "Super block does not support project and group quota together");
1466 		return -EINVAL;
1467 	}
1468 
1469 	return 0;
1470 }
1471 
1472 static int
1473 xfs_init_percpu_counters(
1474 	struct xfs_mount	*mp)
1475 {
1476 	int		error;
1477 
1478 	error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1479 	if (error)
1480 		return -ENOMEM;
1481 
1482 	error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1483 	if (error)
1484 		goto free_icount;
1485 
1486 	error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1487 	if (error)
1488 		goto free_ifree;
1489 
1490 	return 0;
1491 
1492 free_ifree:
1493 	percpu_counter_destroy(&mp->m_ifree);
1494 free_icount:
1495 	percpu_counter_destroy(&mp->m_icount);
1496 	return -ENOMEM;
1497 }
1498 
1499 void
1500 xfs_reinit_percpu_counters(
1501 	struct xfs_mount	*mp)
1502 {
1503 	percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1504 	percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1505 	percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1506 }
1507 
1508 static void
1509 xfs_destroy_percpu_counters(
1510 	struct xfs_mount	*mp)
1511 {
1512 	percpu_counter_destroy(&mp->m_icount);
1513 	percpu_counter_destroy(&mp->m_ifree);
1514 	percpu_counter_destroy(&mp->m_fdblocks);
1515 }
1516 
1517 STATIC int
1518 xfs_fs_fill_super(
1519 	struct super_block	*sb,
1520 	void			*data,
1521 	int			silent)
1522 {
1523 	struct inode		*root;
1524 	struct xfs_mount	*mp = NULL;
1525 	int			flags = 0, error = -ENOMEM;
1526 
1527 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1528 	if (!mp)
1529 		goto out;
1530 
1531 	spin_lock_init(&mp->m_sb_lock);
1532 	mutex_init(&mp->m_growlock);
1533 	atomic_set(&mp->m_active_trans, 0);
1534 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1535 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1536 	INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1537 	mp->m_kobj.kobject.kset = xfs_kset;
1538 
1539 	mp->m_super = sb;
1540 	sb->s_fs_info = mp;
1541 
1542 	error = xfs_parseargs(mp, (char *)data);
1543 	if (error)
1544 		goto out_free_fsname;
1545 
1546 	sb_min_blocksize(sb, BBSIZE);
1547 	sb->s_xattr = xfs_xattr_handlers;
1548 	sb->s_export_op = &xfs_export_operations;
1549 #ifdef CONFIG_XFS_QUOTA
1550 	sb->s_qcop = &xfs_quotactl_operations;
1551 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1552 #endif
1553 	sb->s_op = &xfs_super_operations;
1554 
1555 	if (silent)
1556 		flags |= XFS_MFSI_QUIET;
1557 
1558 	error = xfs_open_devices(mp);
1559 	if (error)
1560 		goto out_free_fsname;
1561 
1562 	error = xfs_init_mount_workqueues(mp);
1563 	if (error)
1564 		goto out_close_devices;
1565 
1566 	error = xfs_init_percpu_counters(mp);
1567 	if (error)
1568 		goto out_destroy_workqueues;
1569 
1570 	/* Allocate stats memory before we do operations that might use it */
1571 	mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1572 	if (!mp->m_stats.xs_stats) {
1573 		error = -ENOMEM;
1574 		goto out_destroy_counters;
1575 	}
1576 
1577 	error = xfs_readsb(mp, flags);
1578 	if (error)
1579 		goto out_free_stats;
1580 
1581 	error = xfs_finish_flags(mp);
1582 	if (error)
1583 		goto out_free_sb;
1584 
1585 	error = xfs_setup_devices(mp);
1586 	if (error)
1587 		goto out_free_sb;
1588 
1589 	error = xfs_filestream_mount(mp);
1590 	if (error)
1591 		goto out_free_sb;
1592 
1593 	/*
1594 	 * we must configure the block size in the superblock before we run the
1595 	 * full mount process as the mount process can lookup and cache inodes.
1596 	 */
1597 	sb->s_magic = XFS_SB_MAGIC;
1598 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1599 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1600 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1601 	sb->s_max_links = XFS_MAXLINK;
1602 	sb->s_time_gran = 1;
1603 	set_posix_acl_flag(sb);
1604 
1605 	/* version 5 superblocks support inode version counters. */
1606 	if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1607 		sb->s_flags |= MS_I_VERSION;
1608 
1609 	if (mp->m_flags & XFS_MOUNT_DAX) {
1610 		xfs_warn(mp,
1611 		"DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1612 
1613 		error = bdev_dax_supported(sb, sb->s_blocksize);
1614 		if (error) {
1615 			xfs_alert(mp,
1616 			"DAX unsupported by block device. Turning off DAX.");
1617 			mp->m_flags &= ~XFS_MOUNT_DAX;
1618 		}
1619 		if (xfs_sb_version_hasreflink(&mp->m_sb))
1620 			xfs_alert(mp,
1621 		"DAX and reflink have not been tested together!");
1622 	}
1623 
1624 	if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1625 		if (mp->m_sb.sb_rblocks) {
1626 			xfs_alert(mp,
1627 	"EXPERIMENTAL reverse mapping btree not compatible with realtime device!");
1628 			error = -EINVAL;
1629 			goto out_filestream_unmount;
1630 		}
1631 		xfs_alert(mp,
1632 	"EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
1633 	}
1634 
1635 	if (xfs_sb_version_hasreflink(&mp->m_sb))
1636 		xfs_alert(mp,
1637 	"EXPERIMENTAL reflink feature enabled. Use at your own risk!");
1638 
1639 	error = xfs_mountfs(mp);
1640 	if (error)
1641 		goto out_filestream_unmount;
1642 
1643 	root = igrab(VFS_I(mp->m_rootip));
1644 	if (!root) {
1645 		error = -ENOENT;
1646 		goto out_unmount;
1647 	}
1648 	sb->s_root = d_make_root(root);
1649 	if (!sb->s_root) {
1650 		error = -ENOMEM;
1651 		goto out_unmount;
1652 	}
1653 
1654 	return 0;
1655 
1656  out_filestream_unmount:
1657 	xfs_filestream_unmount(mp);
1658  out_free_sb:
1659 	xfs_freesb(mp);
1660  out_free_stats:
1661 	free_percpu(mp->m_stats.xs_stats);
1662  out_destroy_counters:
1663 	xfs_destroy_percpu_counters(mp);
1664  out_destroy_workqueues:
1665 	xfs_destroy_mount_workqueues(mp);
1666  out_close_devices:
1667 	xfs_close_devices(mp);
1668  out_free_fsname:
1669 	xfs_free_fsname(mp);
1670 	kfree(mp);
1671  out:
1672 	return error;
1673 
1674  out_unmount:
1675 	xfs_filestream_unmount(mp);
1676 	xfs_unmountfs(mp);
1677 	goto out_free_sb;
1678 }
1679 
1680 STATIC void
1681 xfs_fs_put_super(
1682 	struct super_block	*sb)
1683 {
1684 	struct xfs_mount	*mp = XFS_M(sb);
1685 
1686 	xfs_notice(mp, "Unmounting Filesystem");
1687 	xfs_filestream_unmount(mp);
1688 	xfs_unmountfs(mp);
1689 
1690 	xfs_freesb(mp);
1691 	free_percpu(mp->m_stats.xs_stats);
1692 	xfs_destroy_percpu_counters(mp);
1693 	xfs_destroy_mount_workqueues(mp);
1694 	xfs_close_devices(mp);
1695 	xfs_free_fsname(mp);
1696 	kfree(mp);
1697 }
1698 
1699 STATIC struct dentry *
1700 xfs_fs_mount(
1701 	struct file_system_type	*fs_type,
1702 	int			flags,
1703 	const char		*dev_name,
1704 	void			*data)
1705 {
1706 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1707 }
1708 
1709 static long
1710 xfs_fs_nr_cached_objects(
1711 	struct super_block	*sb,
1712 	struct shrink_control	*sc)
1713 {
1714 	return xfs_reclaim_inodes_count(XFS_M(sb));
1715 }
1716 
1717 static long
1718 xfs_fs_free_cached_objects(
1719 	struct super_block	*sb,
1720 	struct shrink_control	*sc)
1721 {
1722 	return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1723 }
1724 
1725 static const struct super_operations xfs_super_operations = {
1726 	.alloc_inode		= xfs_fs_alloc_inode,
1727 	.destroy_inode		= xfs_fs_destroy_inode,
1728 	.drop_inode		= xfs_fs_drop_inode,
1729 	.put_super		= xfs_fs_put_super,
1730 	.sync_fs		= xfs_fs_sync_fs,
1731 	.freeze_fs		= xfs_fs_freeze,
1732 	.unfreeze_fs		= xfs_fs_unfreeze,
1733 	.statfs			= xfs_fs_statfs,
1734 	.remount_fs		= xfs_fs_remount,
1735 	.show_options		= xfs_fs_show_options,
1736 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1737 	.free_cached_objects	= xfs_fs_free_cached_objects,
1738 };
1739 
1740 static struct file_system_type xfs_fs_type = {
1741 	.owner			= THIS_MODULE,
1742 	.name			= "xfs",
1743 	.mount			= xfs_fs_mount,
1744 	.kill_sb		= kill_block_super,
1745 	.fs_flags		= FS_REQUIRES_DEV,
1746 };
1747 MODULE_ALIAS_FS("xfs");
1748 
1749 STATIC int __init
1750 xfs_init_zones(void)
1751 {
1752 	xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1753 			offsetof(struct xfs_ioend, io_inline_bio));
1754 	if (!xfs_ioend_bioset)
1755 		goto out;
1756 
1757 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1758 						"xfs_log_ticket");
1759 	if (!xfs_log_ticket_zone)
1760 		goto out_free_ioend_bioset;
1761 
1762 	xfs_bmap_free_item_zone = kmem_zone_init(
1763 			sizeof(struct xfs_extent_free_item),
1764 			"xfs_bmap_free_item");
1765 	if (!xfs_bmap_free_item_zone)
1766 		goto out_destroy_log_ticket_zone;
1767 
1768 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1769 						"xfs_btree_cur");
1770 	if (!xfs_btree_cur_zone)
1771 		goto out_destroy_bmap_free_item_zone;
1772 
1773 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1774 						"xfs_da_state");
1775 	if (!xfs_da_state_zone)
1776 		goto out_destroy_btree_cur_zone;
1777 
1778 	xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1779 	if (!xfs_ifork_zone)
1780 		goto out_destroy_da_state_zone;
1781 
1782 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1783 	if (!xfs_trans_zone)
1784 		goto out_destroy_ifork_zone;
1785 
1786 	xfs_log_item_desc_zone =
1787 		kmem_zone_init(sizeof(struct xfs_log_item_desc),
1788 			       "xfs_log_item_desc");
1789 	if (!xfs_log_item_desc_zone)
1790 		goto out_destroy_trans_zone;
1791 
1792 	/*
1793 	 * The size of the zone allocated buf log item is the maximum
1794 	 * size possible under XFS.  This wastes a little bit of memory,
1795 	 * but it is much faster.
1796 	 */
1797 	xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1798 					   "xfs_buf_item");
1799 	if (!xfs_buf_item_zone)
1800 		goto out_destroy_log_item_desc_zone;
1801 
1802 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1803 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1804 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1805 	if (!xfs_efd_zone)
1806 		goto out_destroy_buf_item_zone;
1807 
1808 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1809 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1810 				sizeof(xfs_extent_t))), "xfs_efi_item");
1811 	if (!xfs_efi_zone)
1812 		goto out_destroy_efd_zone;
1813 
1814 	xfs_inode_zone =
1815 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1816 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1817 			KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1818 	if (!xfs_inode_zone)
1819 		goto out_destroy_efi_zone;
1820 
1821 	xfs_ili_zone =
1822 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1823 					KM_ZONE_SPREAD, NULL);
1824 	if (!xfs_ili_zone)
1825 		goto out_destroy_inode_zone;
1826 	xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1827 					"xfs_icr");
1828 	if (!xfs_icreate_zone)
1829 		goto out_destroy_ili_zone;
1830 
1831 	xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1832 			"xfs_rud_item");
1833 	if (!xfs_rud_zone)
1834 		goto out_destroy_icreate_zone;
1835 
1836 	xfs_rui_zone = kmem_zone_init(
1837 			xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1838 			"xfs_rui_item");
1839 	if (!xfs_rui_zone)
1840 		goto out_destroy_rud_zone;
1841 
1842 	xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1843 			"xfs_cud_item");
1844 	if (!xfs_cud_zone)
1845 		goto out_destroy_rui_zone;
1846 
1847 	xfs_cui_zone = kmem_zone_init(
1848 			xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1849 			"xfs_cui_item");
1850 	if (!xfs_cui_zone)
1851 		goto out_destroy_cud_zone;
1852 
1853 	xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1854 			"xfs_bud_item");
1855 	if (!xfs_bud_zone)
1856 		goto out_destroy_cui_zone;
1857 
1858 	xfs_bui_zone = kmem_zone_init(
1859 			xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1860 			"xfs_bui_item");
1861 	if (!xfs_bui_zone)
1862 		goto out_destroy_bud_zone;
1863 
1864 	return 0;
1865 
1866  out_destroy_bud_zone:
1867 	kmem_zone_destroy(xfs_bud_zone);
1868  out_destroy_cui_zone:
1869 	kmem_zone_destroy(xfs_cui_zone);
1870  out_destroy_cud_zone:
1871 	kmem_zone_destroy(xfs_cud_zone);
1872  out_destroy_rui_zone:
1873 	kmem_zone_destroy(xfs_rui_zone);
1874  out_destroy_rud_zone:
1875 	kmem_zone_destroy(xfs_rud_zone);
1876  out_destroy_icreate_zone:
1877 	kmem_zone_destroy(xfs_icreate_zone);
1878  out_destroy_ili_zone:
1879 	kmem_zone_destroy(xfs_ili_zone);
1880  out_destroy_inode_zone:
1881 	kmem_zone_destroy(xfs_inode_zone);
1882  out_destroy_efi_zone:
1883 	kmem_zone_destroy(xfs_efi_zone);
1884  out_destroy_efd_zone:
1885 	kmem_zone_destroy(xfs_efd_zone);
1886  out_destroy_buf_item_zone:
1887 	kmem_zone_destroy(xfs_buf_item_zone);
1888  out_destroy_log_item_desc_zone:
1889 	kmem_zone_destroy(xfs_log_item_desc_zone);
1890  out_destroy_trans_zone:
1891 	kmem_zone_destroy(xfs_trans_zone);
1892  out_destroy_ifork_zone:
1893 	kmem_zone_destroy(xfs_ifork_zone);
1894  out_destroy_da_state_zone:
1895 	kmem_zone_destroy(xfs_da_state_zone);
1896  out_destroy_btree_cur_zone:
1897 	kmem_zone_destroy(xfs_btree_cur_zone);
1898  out_destroy_bmap_free_item_zone:
1899 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1900  out_destroy_log_ticket_zone:
1901 	kmem_zone_destroy(xfs_log_ticket_zone);
1902  out_free_ioend_bioset:
1903 	bioset_free(xfs_ioend_bioset);
1904  out:
1905 	return -ENOMEM;
1906 }
1907 
1908 STATIC void
1909 xfs_destroy_zones(void)
1910 {
1911 	/*
1912 	 * Make sure all delayed rcu free are flushed before we
1913 	 * destroy caches.
1914 	 */
1915 	rcu_barrier();
1916 	kmem_zone_destroy(xfs_bui_zone);
1917 	kmem_zone_destroy(xfs_bud_zone);
1918 	kmem_zone_destroy(xfs_cui_zone);
1919 	kmem_zone_destroy(xfs_cud_zone);
1920 	kmem_zone_destroy(xfs_rui_zone);
1921 	kmem_zone_destroy(xfs_rud_zone);
1922 	kmem_zone_destroy(xfs_icreate_zone);
1923 	kmem_zone_destroy(xfs_ili_zone);
1924 	kmem_zone_destroy(xfs_inode_zone);
1925 	kmem_zone_destroy(xfs_efi_zone);
1926 	kmem_zone_destroy(xfs_efd_zone);
1927 	kmem_zone_destroy(xfs_buf_item_zone);
1928 	kmem_zone_destroy(xfs_log_item_desc_zone);
1929 	kmem_zone_destroy(xfs_trans_zone);
1930 	kmem_zone_destroy(xfs_ifork_zone);
1931 	kmem_zone_destroy(xfs_da_state_zone);
1932 	kmem_zone_destroy(xfs_btree_cur_zone);
1933 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1934 	kmem_zone_destroy(xfs_log_ticket_zone);
1935 	bioset_free(xfs_ioend_bioset);
1936 }
1937 
1938 STATIC int __init
1939 xfs_init_workqueues(void)
1940 {
1941 	/*
1942 	 * The allocation workqueue can be used in memory reclaim situations
1943 	 * (writepage path), and parallelism is only limited by the number of
1944 	 * AGs in all the filesystems mounted. Hence use the default large
1945 	 * max_active value for this workqueue.
1946 	 */
1947 	xfs_alloc_wq = alloc_workqueue("xfsalloc",
1948 			WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1949 	if (!xfs_alloc_wq)
1950 		return -ENOMEM;
1951 
1952 	return 0;
1953 }
1954 
1955 STATIC void
1956 xfs_destroy_workqueues(void)
1957 {
1958 	destroy_workqueue(xfs_alloc_wq);
1959 }
1960 
1961 STATIC int __init
1962 init_xfs_fs(void)
1963 {
1964 	int			error;
1965 
1966 	xfs_check_ondisk_structs();
1967 
1968 	printk(KERN_INFO XFS_VERSION_STRING " with "
1969 			 XFS_BUILD_OPTIONS " enabled\n");
1970 
1971 	xfs_extent_free_init_defer_op();
1972 	xfs_rmap_update_init_defer_op();
1973 	xfs_refcount_update_init_defer_op();
1974 	xfs_bmap_update_init_defer_op();
1975 
1976 	xfs_dir_startup();
1977 
1978 	error = xfs_init_zones();
1979 	if (error)
1980 		goto out;
1981 
1982 	error = xfs_init_workqueues();
1983 	if (error)
1984 		goto out_destroy_zones;
1985 
1986 	error = xfs_mru_cache_init();
1987 	if (error)
1988 		goto out_destroy_wq;
1989 
1990 	error = xfs_buf_init();
1991 	if (error)
1992 		goto out_mru_cache_uninit;
1993 
1994 	error = xfs_init_procfs();
1995 	if (error)
1996 		goto out_buf_terminate;
1997 
1998 	error = xfs_sysctl_register();
1999 	if (error)
2000 		goto out_cleanup_procfs;
2001 
2002 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2003 	if (!xfs_kset) {
2004 		error = -ENOMEM;
2005 		goto out_sysctl_unregister;
2006 	}
2007 
2008 	xfsstats.xs_kobj.kobject.kset = xfs_kset;
2009 
2010 	xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2011 	if (!xfsstats.xs_stats) {
2012 		error = -ENOMEM;
2013 		goto out_kset_unregister;
2014 	}
2015 
2016 	error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2017 			       "stats");
2018 	if (error)
2019 		goto out_free_stats;
2020 
2021 #ifdef DEBUG
2022 	xfs_dbg_kobj.kobject.kset = xfs_kset;
2023 	error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2024 	if (error)
2025 		goto out_remove_stats_kobj;
2026 #endif
2027 
2028 	error = xfs_qm_init();
2029 	if (error)
2030 		goto out_remove_dbg_kobj;
2031 
2032 	error = register_filesystem(&xfs_fs_type);
2033 	if (error)
2034 		goto out_qm_exit;
2035 	return 0;
2036 
2037  out_qm_exit:
2038 	xfs_qm_exit();
2039  out_remove_dbg_kobj:
2040 #ifdef DEBUG
2041 	xfs_sysfs_del(&xfs_dbg_kobj);
2042  out_remove_stats_kobj:
2043 #endif
2044 	xfs_sysfs_del(&xfsstats.xs_kobj);
2045  out_free_stats:
2046 	free_percpu(xfsstats.xs_stats);
2047  out_kset_unregister:
2048 	kset_unregister(xfs_kset);
2049  out_sysctl_unregister:
2050 	xfs_sysctl_unregister();
2051  out_cleanup_procfs:
2052 	xfs_cleanup_procfs();
2053  out_buf_terminate:
2054 	xfs_buf_terminate();
2055  out_mru_cache_uninit:
2056 	xfs_mru_cache_uninit();
2057  out_destroy_wq:
2058 	xfs_destroy_workqueues();
2059  out_destroy_zones:
2060 	xfs_destroy_zones();
2061  out:
2062 	return error;
2063 }
2064 
2065 STATIC void __exit
2066 exit_xfs_fs(void)
2067 {
2068 	xfs_qm_exit();
2069 	unregister_filesystem(&xfs_fs_type);
2070 #ifdef DEBUG
2071 	xfs_sysfs_del(&xfs_dbg_kobj);
2072 #endif
2073 	xfs_sysfs_del(&xfsstats.xs_kobj);
2074 	free_percpu(xfsstats.xs_stats);
2075 	kset_unregister(xfs_kset);
2076 	xfs_sysctl_unregister();
2077 	xfs_cleanup_procfs();
2078 	xfs_buf_terminate();
2079 	xfs_mru_cache_uninit();
2080 	xfs_destroy_workqueues();
2081 	xfs_destroy_zones();
2082 	xfs_uuid_table_free();
2083 }
2084 
2085 module_init(init_xfs_fs);
2086 module_exit(exit_xfs_fs);
2087 
2088 MODULE_AUTHOR("Silicon Graphics, Inc.");
2089 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2090 MODULE_LICENSE("GPL");
2091