xref: /openbmc/linux/fs/xfs/xfs_super.c (revision 7051924f771722c6dd235e693742cda6488ac700)
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_inum.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_inode.h"
30 #include "xfs_btree.h"
31 #include "xfs_bmap.h"
32 #include "xfs_alloc.h"
33 #include "xfs_error.h"
34 #include "xfs_fsops.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_log.h"
38 #include "xfs_log_priv.h"
39 #include "xfs_da_btree.h"
40 #include "xfs_dir2.h"
41 #include "xfs_extfree_item.h"
42 #include "xfs_mru_cache.h"
43 #include "xfs_inode_item.h"
44 #include "xfs_icache.h"
45 #include "xfs_trace.h"
46 #include "xfs_icreate_item.h"
47 #include "xfs_dinode.h"
48 #include "xfs_filestream.h"
49 #include "xfs_quota.h"
50 
51 #include <linux/namei.h>
52 #include <linux/init.h>
53 #include <linux/slab.h>
54 #include <linux/mount.h>
55 #include <linux/mempool.h>
56 #include <linux/writeback.h>
57 #include <linux/kthread.h>
58 #include <linux/freezer.h>
59 #include <linux/parser.h>
60 
61 static const struct super_operations xfs_super_operations;
62 static kmem_zone_t *xfs_ioend_zone;
63 mempool_t *xfs_ioend_pool;
64 struct kset *xfs_kset;
65 
66 #define MNTOPT_LOGBUFS	"logbufs"	/* number of XFS log buffers */
67 #define MNTOPT_LOGBSIZE	"logbsize"	/* size of XFS log buffers */
68 #define MNTOPT_LOGDEV	"logdev"	/* log device */
69 #define MNTOPT_RTDEV	"rtdev"		/* realtime I/O device */
70 #define MNTOPT_BIOSIZE	"biosize"	/* log2 of preferred buffered io size */
71 #define MNTOPT_WSYNC	"wsync"		/* safe-mode nfs compatible mount */
72 #define MNTOPT_NOALIGN	"noalign"	/* turn off stripe alignment */
73 #define MNTOPT_SWALLOC	"swalloc"	/* turn on stripe width allocation */
74 #define MNTOPT_SUNIT	"sunit"		/* data volume stripe unit */
75 #define MNTOPT_SWIDTH	"swidth"	/* data volume stripe width */
76 #define MNTOPT_NOUUID	"nouuid"	/* ignore filesystem UUID */
77 #define MNTOPT_MTPT	"mtpt"		/* filesystem mount point */
78 #define MNTOPT_GRPID	"grpid"		/* group-ID from parent directory */
79 #define MNTOPT_NOGRPID	"nogrpid"	/* group-ID from current process */
80 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
81 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
82 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
83 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
84 #define MNTOPT_BARRIER	"barrier"	/* use writer barriers for log write and
85 					 * unwritten extent conversion */
86 #define MNTOPT_NOBARRIER "nobarrier"	/* .. disable */
87 #define MNTOPT_64BITINODE   "inode64"	/* inodes can be allocated anywhere */
88 #define MNTOPT_32BITINODE   "inode32"	/* inode allocation limited to
89 					 * XFS_MAXINUMBER_32 */
90 #define MNTOPT_IKEEP	"ikeep"		/* do not free empty inode clusters */
91 #define MNTOPT_NOIKEEP	"noikeep"	/* free empty inode clusters */
92 #define MNTOPT_LARGEIO	   "largeio"	/* report large I/O sizes in stat() */
93 #define MNTOPT_NOLARGEIO   "nolargeio"	/* do not report large I/O sizes
94 					 * in stat(). */
95 #define MNTOPT_ATTR2	"attr2"		/* do use attr2 attribute format */
96 #define MNTOPT_NOATTR2	"noattr2"	/* do not use attr2 attribute format */
97 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
98 #define MNTOPT_QUOTA	"quota"		/* disk quotas (user) */
99 #define MNTOPT_NOQUOTA	"noquota"	/* no quotas */
100 #define MNTOPT_USRQUOTA	"usrquota"	/* user quota enabled */
101 #define MNTOPT_GRPQUOTA	"grpquota"	/* group quota enabled */
102 #define MNTOPT_PRJQUOTA	"prjquota"	/* project quota enabled */
103 #define MNTOPT_UQUOTA	"uquota"	/* user quota (IRIX variant) */
104 #define MNTOPT_GQUOTA	"gquota"	/* group quota (IRIX variant) */
105 #define MNTOPT_PQUOTA	"pquota"	/* project quota (IRIX variant) */
106 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
107 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
108 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
109 #define MNTOPT_QUOTANOENF  "qnoenforce"	/* same as uqnoenforce */
110 #define MNTOPT_DELAYLOG    "delaylog"	/* Delayed logging enabled */
111 #define MNTOPT_NODELAYLOG  "nodelaylog"	/* Delayed logging disabled */
112 #define MNTOPT_DISCARD	   "discard"	/* Discard unused blocks */
113 #define MNTOPT_NODISCARD   "nodiscard"	/* Do not discard unused blocks */
114 
115 /*
116  * Table driven mount option parser.
117  *
118  * Currently only used for remount, but it will be used for mount
119  * in the future, too.
120  */
121 enum {
122 	Opt_barrier,
123 	Opt_nobarrier,
124 	Opt_inode64,
125 	Opt_inode32,
126 	Opt_err
127 };
128 
129 static const match_table_t tokens = {
130 	{Opt_barrier, "barrier"},
131 	{Opt_nobarrier, "nobarrier"},
132 	{Opt_inode64, "inode64"},
133 	{Opt_inode32, "inode32"},
134 	{Opt_err, NULL}
135 };
136 
137 
138 STATIC unsigned long
139 suffix_kstrtoint(char *s, unsigned int base, int *res)
140 {
141 	int	last, shift_left_factor = 0, _res;
142 	char	*value = s;
143 
144 	last = strlen(value) - 1;
145 	if (value[last] == 'K' || value[last] == 'k') {
146 		shift_left_factor = 10;
147 		value[last] = '\0';
148 	}
149 	if (value[last] == 'M' || value[last] == 'm') {
150 		shift_left_factor = 20;
151 		value[last] = '\0';
152 	}
153 	if (value[last] == 'G' || value[last] == 'g') {
154 		shift_left_factor = 30;
155 		value[last] = '\0';
156 	}
157 
158 	if (kstrtoint(s, base, &_res))
159 		return -EINVAL;
160 	*res = _res << shift_left_factor;
161 	return 0;
162 }
163 
164 /*
165  * This function fills in xfs_mount_t fields based on mount args.
166  * Note: the superblock has _not_ yet been read in.
167  *
168  * Note that this function leaks the various device name allocations on
169  * failure.  The caller takes care of them.
170  */
171 STATIC int
172 xfs_parseargs(
173 	struct xfs_mount	*mp,
174 	char			*options)
175 {
176 	struct super_block	*sb = mp->m_super;
177 	char			*this_char, *value;
178 	int			dsunit = 0;
179 	int			dswidth = 0;
180 	int			iosize = 0;
181 	__uint8_t		iosizelog = 0;
182 
183 	/*
184 	 * set up the mount name first so all the errors will refer to the
185 	 * correct device.
186 	 */
187 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
188 	if (!mp->m_fsname)
189 		return -ENOMEM;
190 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
191 
192 	/*
193 	 * Copy binary VFS mount flags we are interested in.
194 	 */
195 	if (sb->s_flags & MS_RDONLY)
196 		mp->m_flags |= XFS_MOUNT_RDONLY;
197 	if (sb->s_flags & MS_DIRSYNC)
198 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
199 	if (sb->s_flags & MS_SYNCHRONOUS)
200 		mp->m_flags |= XFS_MOUNT_WSYNC;
201 
202 	/*
203 	 * Set some default flags that could be cleared by the mount option
204 	 * parsing.
205 	 */
206 	mp->m_flags |= XFS_MOUNT_BARRIER;
207 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
208 
209 	/*
210 	 * These can be overridden by the mount option parsing.
211 	 */
212 	mp->m_logbufs = -1;
213 	mp->m_logbsize = -1;
214 
215 	if (!options)
216 		goto done;
217 
218 	while ((this_char = strsep(&options, ",")) != NULL) {
219 		if (!*this_char)
220 			continue;
221 		if ((value = strchr(this_char, '=')) != NULL)
222 			*value++ = 0;
223 
224 		if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
225 			if (!value || !*value) {
226 				xfs_warn(mp, "%s option requires an argument",
227 					this_char);
228 				return -EINVAL;
229 			}
230 			if (kstrtoint(value, 10, &mp->m_logbufs))
231 				return -EINVAL;
232 		} else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
233 			if (!value || !*value) {
234 				xfs_warn(mp, "%s option requires an argument",
235 					this_char);
236 				return -EINVAL;
237 			}
238 			if (suffix_kstrtoint(value, 10, &mp->m_logbsize))
239 				return -EINVAL;
240 		} else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
241 			if (!value || !*value) {
242 				xfs_warn(mp, "%s option requires an argument",
243 					this_char);
244 				return -EINVAL;
245 			}
246 			mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
247 			if (!mp->m_logname)
248 				return -ENOMEM;
249 		} else if (!strcmp(this_char, MNTOPT_MTPT)) {
250 			xfs_warn(mp, "%s option not allowed on this system",
251 				this_char);
252 			return -EINVAL;
253 		} else if (!strcmp(this_char, MNTOPT_RTDEV)) {
254 			if (!value || !*value) {
255 				xfs_warn(mp, "%s option requires an argument",
256 					this_char);
257 				return -EINVAL;
258 			}
259 			mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
260 			if (!mp->m_rtname)
261 				return -ENOMEM;
262 		} else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
263 			if (!value || !*value) {
264 				xfs_warn(mp, "%s option requires an argument",
265 					this_char);
266 				return -EINVAL;
267 			}
268 			if (kstrtoint(value, 10, &iosize))
269 				return -EINVAL;
270 			iosizelog = ffs(iosize) - 1;
271 		} else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
272 			if (!value || !*value) {
273 				xfs_warn(mp, "%s option requires an argument",
274 					this_char);
275 				return -EINVAL;
276 			}
277 			if (suffix_kstrtoint(value, 10, &iosize))
278 				return -EINVAL;
279 			iosizelog = ffs(iosize) - 1;
280 		} else if (!strcmp(this_char, MNTOPT_GRPID) ||
281 			   !strcmp(this_char, MNTOPT_BSDGROUPS)) {
282 			mp->m_flags |= XFS_MOUNT_GRPID;
283 		} else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
284 			   !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
285 			mp->m_flags &= ~XFS_MOUNT_GRPID;
286 		} else if (!strcmp(this_char, MNTOPT_WSYNC)) {
287 			mp->m_flags |= XFS_MOUNT_WSYNC;
288 		} else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
289 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
290 		} else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
291 			mp->m_flags |= XFS_MOUNT_NOALIGN;
292 		} else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
293 			mp->m_flags |= XFS_MOUNT_SWALLOC;
294 		} else if (!strcmp(this_char, MNTOPT_SUNIT)) {
295 			if (!value || !*value) {
296 				xfs_warn(mp, "%s option requires an argument",
297 					this_char);
298 				return -EINVAL;
299 			}
300 			if (kstrtoint(value, 10, &dsunit))
301 				return -EINVAL;
302 		} else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
303 			if (!value || !*value) {
304 				xfs_warn(mp, "%s option requires an argument",
305 					this_char);
306 				return -EINVAL;
307 			}
308 			if (kstrtoint(value, 10, &dswidth))
309 				return -EINVAL;
310 		} else if (!strcmp(this_char, MNTOPT_32BITINODE)) {
311 			mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
312 		} else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
313 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
314 		} else if (!strcmp(this_char, MNTOPT_NOUUID)) {
315 			mp->m_flags |= XFS_MOUNT_NOUUID;
316 		} else if (!strcmp(this_char, MNTOPT_BARRIER)) {
317 			mp->m_flags |= XFS_MOUNT_BARRIER;
318 		} else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
319 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
320 		} else if (!strcmp(this_char, MNTOPT_IKEEP)) {
321 			mp->m_flags |= XFS_MOUNT_IKEEP;
322 		} else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
323 			mp->m_flags &= ~XFS_MOUNT_IKEEP;
324 		} else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
325 			mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
326 		} else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
327 			mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
328 		} else if (!strcmp(this_char, MNTOPT_ATTR2)) {
329 			mp->m_flags |= XFS_MOUNT_ATTR2;
330 		} else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
331 			mp->m_flags &= ~XFS_MOUNT_ATTR2;
332 			mp->m_flags |= XFS_MOUNT_NOATTR2;
333 		} else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
334 			mp->m_flags |= XFS_MOUNT_FILESTREAMS;
335 		} else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
336 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
337 			mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
338 			mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
339 		} else if (!strcmp(this_char, MNTOPT_QUOTA) ||
340 			   !strcmp(this_char, MNTOPT_UQUOTA) ||
341 			   !strcmp(this_char, MNTOPT_USRQUOTA)) {
342 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
343 					 XFS_UQUOTA_ENFD);
344 		} else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
345 			   !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
346 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
347 			mp->m_qflags &= ~XFS_UQUOTA_ENFD;
348 		} else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
349 			   !strcmp(this_char, MNTOPT_PRJQUOTA)) {
350 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
351 					 XFS_PQUOTA_ENFD);
352 		} else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
353 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
354 			mp->m_qflags &= ~XFS_PQUOTA_ENFD;
355 		} else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
356 			   !strcmp(this_char, MNTOPT_GRPQUOTA)) {
357 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
358 					 XFS_GQUOTA_ENFD);
359 		} else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
360 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
361 			mp->m_qflags &= ~XFS_GQUOTA_ENFD;
362 		} else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
363 			xfs_warn(mp,
364 	"delaylog is the default now, option is deprecated.");
365 		} else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
366 			xfs_warn(mp,
367 	"nodelaylog support has been removed, option is deprecated.");
368 		} else if (!strcmp(this_char, MNTOPT_DISCARD)) {
369 			mp->m_flags |= XFS_MOUNT_DISCARD;
370 		} else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
371 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
372 		} else if (!strcmp(this_char, "ihashsize")) {
373 			xfs_warn(mp,
374 	"ihashsize no longer used, option is deprecated.");
375 		} else if (!strcmp(this_char, "osyncisdsync")) {
376 			xfs_warn(mp,
377 	"osyncisdsync has no effect, option is deprecated.");
378 		} else if (!strcmp(this_char, "osyncisosync")) {
379 			xfs_warn(mp,
380 	"osyncisosync has no effect, option is deprecated.");
381 		} else if (!strcmp(this_char, "irixsgid")) {
382 			xfs_warn(mp,
383 	"irixsgid is now a sysctl(2) variable, option is deprecated.");
384 		} else {
385 			xfs_warn(mp, "unknown mount option [%s].", this_char);
386 			return -EINVAL;
387 		}
388 	}
389 
390 	/*
391 	 * no recovery flag requires a read-only mount
392 	 */
393 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
394 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
395 		xfs_warn(mp, "no-recovery mounts must be read-only.");
396 		return -EINVAL;
397 	}
398 
399 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
400 		xfs_warn(mp,
401 	"sunit and swidth options incompatible with the noalign option");
402 		return -EINVAL;
403 	}
404 
405 #ifndef CONFIG_XFS_QUOTA
406 	if (XFS_IS_QUOTA_RUNNING(mp)) {
407 		xfs_warn(mp, "quota support not available in this kernel.");
408 		return -EINVAL;
409 	}
410 #endif
411 
412 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
413 		xfs_warn(mp, "sunit and swidth must be specified together");
414 		return -EINVAL;
415 	}
416 
417 	if (dsunit && (dswidth % dsunit != 0)) {
418 		xfs_warn(mp,
419 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
420 			dswidth, dsunit);
421 		return -EINVAL;
422 	}
423 
424 done:
425 	if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
426 		/*
427 		 * At this point the superblock has not been read
428 		 * in, therefore we do not know the block size.
429 		 * Before the mount call ends we will convert
430 		 * these to FSBs.
431 		 */
432 		mp->m_dalign = dsunit;
433 		mp->m_swidth = dswidth;
434 	}
435 
436 	if (mp->m_logbufs != -1 &&
437 	    mp->m_logbufs != 0 &&
438 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
439 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
440 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
441 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
442 		return -EINVAL;
443 	}
444 	if (mp->m_logbsize != -1 &&
445 	    mp->m_logbsize !=  0 &&
446 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
447 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
448 	     !is_power_of_2(mp->m_logbsize))) {
449 		xfs_warn(mp,
450 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
451 			mp->m_logbsize);
452 		return -EINVAL;
453 	}
454 
455 	if (iosizelog) {
456 		if (iosizelog > XFS_MAX_IO_LOG ||
457 		    iosizelog < XFS_MIN_IO_LOG) {
458 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
459 				iosizelog, XFS_MIN_IO_LOG,
460 				XFS_MAX_IO_LOG);
461 			return -EINVAL;
462 		}
463 
464 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
465 		mp->m_readio_log = iosizelog;
466 		mp->m_writeio_log = iosizelog;
467 	}
468 
469 	return 0;
470 }
471 
472 struct proc_xfs_info {
473 	int	flag;
474 	char	*str;
475 };
476 
477 STATIC int
478 xfs_showargs(
479 	struct xfs_mount	*mp,
480 	struct seq_file		*m)
481 {
482 	static struct proc_xfs_info xfs_info_set[] = {
483 		/* the few simple ones we can get from the mount struct */
484 		{ XFS_MOUNT_IKEEP,		"," MNTOPT_IKEEP },
485 		{ XFS_MOUNT_WSYNC,		"," MNTOPT_WSYNC },
486 		{ XFS_MOUNT_NOALIGN,		"," MNTOPT_NOALIGN },
487 		{ XFS_MOUNT_SWALLOC,		"," MNTOPT_SWALLOC },
488 		{ XFS_MOUNT_NOUUID,		"," MNTOPT_NOUUID },
489 		{ XFS_MOUNT_NORECOVERY,		"," MNTOPT_NORECOVERY },
490 		{ XFS_MOUNT_ATTR2,		"," MNTOPT_ATTR2 },
491 		{ XFS_MOUNT_FILESTREAMS,	"," MNTOPT_FILESTREAM },
492 		{ XFS_MOUNT_GRPID,		"," MNTOPT_GRPID },
493 		{ XFS_MOUNT_DISCARD,		"," MNTOPT_DISCARD },
494 		{ XFS_MOUNT_SMALL_INUMS,	"," MNTOPT_32BITINODE },
495 		{ 0, NULL }
496 	};
497 	static struct proc_xfs_info xfs_info_unset[] = {
498 		/* the few simple ones we can get from the mount struct */
499 		{ XFS_MOUNT_COMPAT_IOSIZE,	"," MNTOPT_LARGEIO },
500 		{ XFS_MOUNT_BARRIER,		"," MNTOPT_NOBARRIER },
501 		{ XFS_MOUNT_SMALL_INUMS,	"," MNTOPT_64BITINODE },
502 		{ 0, NULL }
503 	};
504 	struct proc_xfs_info	*xfs_infop;
505 
506 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
507 		if (mp->m_flags & xfs_infop->flag)
508 			seq_puts(m, xfs_infop->str);
509 	}
510 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
511 		if (!(mp->m_flags & xfs_infop->flag))
512 			seq_puts(m, xfs_infop->str);
513 	}
514 
515 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
516 		seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
517 				(int)(1 << mp->m_writeio_log) >> 10);
518 
519 	if (mp->m_logbufs > 0)
520 		seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
521 	if (mp->m_logbsize > 0)
522 		seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
523 
524 	if (mp->m_logname)
525 		seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
526 	if (mp->m_rtname)
527 		seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
528 
529 	if (mp->m_dalign > 0)
530 		seq_printf(m, "," MNTOPT_SUNIT "=%d",
531 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
532 	if (mp->m_swidth > 0)
533 		seq_printf(m, "," MNTOPT_SWIDTH "=%d",
534 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
535 
536 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
537 		seq_puts(m, "," MNTOPT_USRQUOTA);
538 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
539 		seq_puts(m, "," MNTOPT_UQUOTANOENF);
540 
541 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
542 		if (mp->m_qflags & XFS_PQUOTA_ENFD)
543 			seq_puts(m, "," MNTOPT_PRJQUOTA);
544 		else
545 			seq_puts(m, "," MNTOPT_PQUOTANOENF);
546 	}
547 	if (mp->m_qflags & XFS_GQUOTA_ACCT) {
548 		if (mp->m_qflags & XFS_GQUOTA_ENFD)
549 			seq_puts(m, "," MNTOPT_GRPQUOTA);
550 		else
551 			seq_puts(m, "," MNTOPT_GQUOTANOENF);
552 	}
553 
554 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
555 		seq_puts(m, "," MNTOPT_NOQUOTA);
556 
557 	return 0;
558 }
559 __uint64_t
560 xfs_max_file_offset(
561 	unsigned int		blockshift)
562 {
563 	unsigned int		pagefactor = 1;
564 	unsigned int		bitshift = BITS_PER_LONG - 1;
565 
566 	/* Figure out maximum filesize, on Linux this can depend on
567 	 * the filesystem blocksize (on 32 bit platforms).
568 	 * __block_write_begin does this in an [unsigned] long...
569 	 *      page->index << (PAGE_CACHE_SHIFT - bbits)
570 	 * So, for page sized blocks (4K on 32 bit platforms),
571 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
572 	 *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
573 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
574 	 * Note1: get_block_t takes a long (implicit cast from above)
575 	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
576 	 * can optionally convert the [unsigned] long from above into
577 	 * an [unsigned] long long.
578 	 */
579 
580 #if BITS_PER_LONG == 32
581 # if defined(CONFIG_LBDAF)
582 	ASSERT(sizeof(sector_t) == 8);
583 	pagefactor = PAGE_CACHE_SIZE;
584 	bitshift = BITS_PER_LONG;
585 # else
586 	pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
587 # endif
588 #endif
589 
590 	return (((__uint64_t)pagefactor) << bitshift) - 1;
591 }
592 
593 /*
594  * xfs_set_inode32() and xfs_set_inode64() are passed an agcount
595  * because in the growfs case, mp->m_sb.sb_agcount is not updated
596  * yet to the potentially higher ag count.
597  */
598 xfs_agnumber_t
599 xfs_set_inode32(struct xfs_mount *mp, xfs_agnumber_t agcount)
600 {
601 	xfs_agnumber_t	index = 0;
602 	xfs_agnumber_t	maxagi = 0;
603 	xfs_sb_t	*sbp = &mp->m_sb;
604 	xfs_agnumber_t	max_metadata;
605 	xfs_agino_t	agino;
606 	xfs_ino_t	ino;
607 	xfs_perag_t	*pag;
608 
609 	/* Calculate how much should be reserved for inodes to meet
610 	 * the max inode percentage.
611 	 */
612 	if (mp->m_maxicount) {
613 		__uint64_t	icount;
614 
615 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
616 		do_div(icount, 100);
617 		icount += sbp->sb_agblocks - 1;
618 		do_div(icount, sbp->sb_agblocks);
619 		max_metadata = icount;
620 	} else {
621 		max_metadata = agcount;
622 	}
623 
624 	agino =	XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
625 
626 	for (index = 0; index < agcount; index++) {
627 		ino = XFS_AGINO_TO_INO(mp, index, agino);
628 
629 		if (ino > XFS_MAXINUMBER_32) {
630 			pag = xfs_perag_get(mp, index);
631 			pag->pagi_inodeok = 0;
632 			pag->pagf_metadata = 0;
633 			xfs_perag_put(pag);
634 			continue;
635 		}
636 
637 		pag = xfs_perag_get(mp, index);
638 		pag->pagi_inodeok = 1;
639 		maxagi++;
640 		if (index < max_metadata)
641 			pag->pagf_metadata = 1;
642 		xfs_perag_put(pag);
643 	}
644 	mp->m_flags |= (XFS_MOUNT_32BITINODES |
645 			XFS_MOUNT_SMALL_INUMS);
646 
647 	return maxagi;
648 }
649 
650 xfs_agnumber_t
651 xfs_set_inode64(struct xfs_mount *mp, xfs_agnumber_t agcount)
652 {
653 	xfs_agnumber_t index = 0;
654 
655 	for (index = 0; index < agcount; index++) {
656 		struct xfs_perag	*pag;
657 
658 		pag = xfs_perag_get(mp, index);
659 		pag->pagi_inodeok = 1;
660 		pag->pagf_metadata = 0;
661 		xfs_perag_put(pag);
662 	}
663 
664 	/* There is no need for lock protection on m_flags,
665 	 * the rw_semaphore of the VFS superblock is locked
666 	 * during mount/umount/remount operations, so this is
667 	 * enough to avoid concurency on the m_flags field
668 	 */
669 	mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
670 			 XFS_MOUNT_SMALL_INUMS);
671 	return index;
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\n", 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 	if (rtdev)
795 		xfs_blkdev_put(rtdev);
796  out_close_logdev:
797 	if (logdev && logdev != ddev)
798 		xfs_blkdev_put(logdev);
799  out:
800 	return error;
801 }
802 
803 /*
804  * Setup xfs_mount buffer target pointers based on superblock
805  */
806 STATIC int
807 xfs_setup_devices(
808 	struct xfs_mount	*mp)
809 {
810 	int			error;
811 
812 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
813 	if (error)
814 		return error;
815 
816 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
817 		unsigned int	log_sector_size = BBSIZE;
818 
819 		if (xfs_sb_version_hassector(&mp->m_sb))
820 			log_sector_size = mp->m_sb.sb_logsectsize;
821 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
822 					    log_sector_size);
823 		if (error)
824 			return error;
825 	}
826 	if (mp->m_rtdev_targp) {
827 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
828 					    mp->m_sb.sb_sectsize);
829 		if (error)
830 			return error;
831 	}
832 
833 	return 0;
834 }
835 
836 STATIC int
837 xfs_init_mount_workqueues(
838 	struct xfs_mount	*mp)
839 {
840 	mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
841 			WQ_MEM_RECLAIM, 0, mp->m_fsname);
842 	if (!mp->m_data_workqueue)
843 		goto out;
844 
845 	mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
846 			WQ_MEM_RECLAIM, 0, mp->m_fsname);
847 	if (!mp->m_unwritten_workqueue)
848 		goto out_destroy_data_iodone_queue;
849 
850 	mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
851 			WQ_MEM_RECLAIM, 0, mp->m_fsname);
852 	if (!mp->m_cil_workqueue)
853 		goto out_destroy_unwritten;
854 
855 	mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
856 			0, 0, mp->m_fsname);
857 	if (!mp->m_reclaim_workqueue)
858 		goto out_destroy_cil;
859 
860 	mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
861 			0, 0, mp->m_fsname);
862 	if (!mp->m_log_workqueue)
863 		goto out_destroy_reclaim;
864 
865 	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
866 			0, 0, mp->m_fsname);
867 	if (!mp->m_eofblocks_workqueue)
868 		goto out_destroy_log;
869 
870 	return 0;
871 
872 out_destroy_log:
873 	destroy_workqueue(mp->m_log_workqueue);
874 out_destroy_reclaim:
875 	destroy_workqueue(mp->m_reclaim_workqueue);
876 out_destroy_cil:
877 	destroy_workqueue(mp->m_cil_workqueue);
878 out_destroy_unwritten:
879 	destroy_workqueue(mp->m_unwritten_workqueue);
880 out_destroy_data_iodone_queue:
881 	destroy_workqueue(mp->m_data_workqueue);
882 out:
883 	return -ENOMEM;
884 }
885 
886 STATIC void
887 xfs_destroy_mount_workqueues(
888 	struct xfs_mount	*mp)
889 {
890 	destroy_workqueue(mp->m_eofblocks_workqueue);
891 	destroy_workqueue(mp->m_log_workqueue);
892 	destroy_workqueue(mp->m_reclaim_workqueue);
893 	destroy_workqueue(mp->m_cil_workqueue);
894 	destroy_workqueue(mp->m_data_workqueue);
895 	destroy_workqueue(mp->m_unwritten_workqueue);
896 }
897 
898 /*
899  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
900  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
901  * for IO to complete so that we effectively throttle multiple callers to the
902  * rate at which IO is completing.
903  */
904 void
905 xfs_flush_inodes(
906 	struct xfs_mount	*mp)
907 {
908 	struct super_block	*sb = mp->m_super;
909 
910 	if (down_read_trylock(&sb->s_umount)) {
911 		sync_inodes_sb(sb);
912 		up_read(&sb->s_umount);
913 	}
914 }
915 
916 /* Catch misguided souls that try to use this interface on XFS */
917 STATIC struct inode *
918 xfs_fs_alloc_inode(
919 	struct super_block	*sb)
920 {
921 	BUG();
922 	return NULL;
923 }
924 
925 /*
926  * Now that the generic code is guaranteed not to be accessing
927  * the linux inode, we can reclaim the inode.
928  */
929 STATIC void
930 xfs_fs_destroy_inode(
931 	struct inode		*inode)
932 {
933 	struct xfs_inode	*ip = XFS_I(inode);
934 
935 	trace_xfs_destroy_inode(ip);
936 
937 	XFS_STATS_INC(vn_reclaim);
938 
939 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
940 
941 	/*
942 	 * We should never get here with one of the reclaim flags already set.
943 	 */
944 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
945 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
946 
947 	/*
948 	 * We always use background reclaim here because even if the
949 	 * inode is clean, it still may be under IO and hence we have
950 	 * to take the flush lock. The background reclaim path handles
951 	 * this more efficiently than we can here, so simply let background
952 	 * reclaim tear down all inodes.
953 	 */
954 	xfs_inode_set_reclaim_tag(ip);
955 }
956 
957 /*
958  * Slab object creation initialisation for the XFS inode.
959  * This covers only the idempotent fields in the XFS inode;
960  * all other fields need to be initialised on allocation
961  * from the slab. This avoids the need to repeatedly initialise
962  * fields in the xfs inode that left in the initialise state
963  * when freeing the inode.
964  */
965 STATIC void
966 xfs_fs_inode_init_once(
967 	void			*inode)
968 {
969 	struct xfs_inode	*ip = inode;
970 
971 	memset(ip, 0, sizeof(struct xfs_inode));
972 
973 	/* vfs inode */
974 	inode_init_once(VFS_I(ip));
975 
976 	/* xfs inode */
977 	atomic_set(&ip->i_pincount, 0);
978 	spin_lock_init(&ip->i_flags_lock);
979 
980 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
981 		     "xfsino", ip->i_ino);
982 }
983 
984 STATIC void
985 xfs_fs_evict_inode(
986 	struct inode		*inode)
987 {
988 	xfs_inode_t		*ip = XFS_I(inode);
989 
990 	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
991 
992 	trace_xfs_evict_inode(ip);
993 
994 	truncate_inode_pages_final(&inode->i_data);
995 	clear_inode(inode);
996 	XFS_STATS_INC(vn_rele);
997 	XFS_STATS_INC(vn_remove);
998 	XFS_STATS_DEC(vn_active);
999 
1000 	xfs_inactive(ip);
1001 }
1002 
1003 /*
1004  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1005  * serialised against cache hits here via the inode->i_lock and igrab() in
1006  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1007  * racing with us, and it avoids needing to grab a spinlock here for every inode
1008  * we drop the final reference on.
1009  */
1010 STATIC int
1011 xfs_fs_drop_inode(
1012 	struct inode		*inode)
1013 {
1014 	struct xfs_inode	*ip = XFS_I(inode);
1015 
1016 	return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1017 }
1018 
1019 STATIC void
1020 xfs_free_fsname(
1021 	struct xfs_mount	*mp)
1022 {
1023 	kfree(mp->m_fsname);
1024 	kfree(mp->m_rtname);
1025 	kfree(mp->m_logname);
1026 }
1027 
1028 STATIC void
1029 xfs_fs_put_super(
1030 	struct super_block	*sb)
1031 {
1032 	struct xfs_mount	*mp = XFS_M(sb);
1033 
1034 	xfs_filestream_unmount(mp);
1035 	xfs_unmountfs(mp);
1036 
1037 	xfs_freesb(mp);
1038 	xfs_icsb_destroy_counters(mp);
1039 	xfs_destroy_mount_workqueues(mp);
1040 	xfs_close_devices(mp);
1041 	xfs_free_fsname(mp);
1042 	kfree(mp);
1043 }
1044 
1045 STATIC int
1046 xfs_fs_sync_fs(
1047 	struct super_block	*sb,
1048 	int			wait)
1049 {
1050 	struct xfs_mount	*mp = XFS_M(sb);
1051 
1052 	/*
1053 	 * Doing anything during the async pass would be counterproductive.
1054 	 */
1055 	if (!wait)
1056 		return 0;
1057 
1058 	xfs_log_force(mp, XFS_LOG_SYNC);
1059 	if (laptop_mode) {
1060 		/*
1061 		 * The disk must be active because we're syncing.
1062 		 * We schedule log work now (now that the disk is
1063 		 * active) instead of later (when it might not be).
1064 		 */
1065 		flush_delayed_work(&mp->m_log->l_work);
1066 	}
1067 
1068 	return 0;
1069 }
1070 
1071 STATIC int
1072 xfs_fs_statfs(
1073 	struct dentry		*dentry,
1074 	struct kstatfs		*statp)
1075 {
1076 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1077 	xfs_sb_t		*sbp = &mp->m_sb;
1078 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
1079 	__uint64_t		fakeinos, id;
1080 	xfs_extlen_t		lsize;
1081 	__int64_t		ffree;
1082 
1083 	statp->f_type = XFS_SB_MAGIC;
1084 	statp->f_namelen = MAXNAMELEN - 1;
1085 
1086 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1087 	statp->f_fsid.val[0] = (u32)id;
1088 	statp->f_fsid.val[1] = (u32)(id >> 32);
1089 
1090 	xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1091 
1092 	spin_lock(&mp->m_sb_lock);
1093 	statp->f_bsize = sbp->sb_blocksize;
1094 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1095 	statp->f_blocks = sbp->sb_dblocks - lsize;
1096 	statp->f_bfree = statp->f_bavail =
1097 				sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1098 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
1099 	statp->f_files =
1100 	    MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1101 	if (mp->m_maxicount)
1102 		statp->f_files = min_t(typeof(statp->f_files),
1103 					statp->f_files,
1104 					mp->m_maxicount);
1105 
1106 	/* make sure statp->f_ffree does not underflow */
1107 	ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1108 	statp->f_ffree = max_t(__int64_t, ffree, 0);
1109 
1110 	spin_unlock(&mp->m_sb_lock);
1111 
1112 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1113 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1114 			      (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1115 		xfs_qm_statvfs(ip, statp);
1116 	return 0;
1117 }
1118 
1119 STATIC void
1120 xfs_save_resvblks(struct xfs_mount *mp)
1121 {
1122 	__uint64_t resblks = 0;
1123 
1124 	mp->m_resblks_save = mp->m_resblks;
1125 	xfs_reserve_blocks(mp, &resblks, NULL);
1126 }
1127 
1128 STATIC void
1129 xfs_restore_resvblks(struct xfs_mount *mp)
1130 {
1131 	__uint64_t resblks;
1132 
1133 	if (mp->m_resblks_save) {
1134 		resblks = mp->m_resblks_save;
1135 		mp->m_resblks_save = 0;
1136 	} else
1137 		resblks = xfs_default_resblks(mp);
1138 
1139 	xfs_reserve_blocks(mp, &resblks, NULL);
1140 }
1141 
1142 /*
1143  * Trigger writeback of all the dirty metadata in the file system.
1144  *
1145  * This ensures that the metadata is written to their location on disk rather
1146  * than just existing in transactions in the log. This means after a quiesce
1147  * there is no log replay required to write the inodes to disk - this is the
1148  * primary difference between a sync and a quiesce.
1149  *
1150  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1151  * it is started again when appropriate.
1152  */
1153 static void
1154 xfs_quiesce_attr(
1155 	struct xfs_mount	*mp)
1156 {
1157 	int	error = 0;
1158 
1159 	/* wait for all modifications to complete */
1160 	while (atomic_read(&mp->m_active_trans) > 0)
1161 		delay(100);
1162 
1163 	/* force the log to unpin objects from the now complete transactions */
1164 	xfs_log_force(mp, XFS_LOG_SYNC);
1165 
1166 	/* reclaim inodes to do any IO before the freeze completes */
1167 	xfs_reclaim_inodes(mp, 0);
1168 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1169 
1170 	/* Push the superblock and write an unmount record */
1171 	error = xfs_log_sbcount(mp);
1172 	if (error)
1173 		xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1174 				"Frozen image may not be consistent.");
1175 	/*
1176 	 * Just warn here till VFS can correctly support
1177 	 * read-only remount without racing.
1178 	 */
1179 	WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1180 
1181 	xfs_log_quiesce(mp);
1182 }
1183 
1184 STATIC int
1185 xfs_fs_remount(
1186 	struct super_block	*sb,
1187 	int			*flags,
1188 	char			*options)
1189 {
1190 	struct xfs_mount	*mp = XFS_M(sb);
1191 	xfs_sb_t		*sbp = &mp->m_sb;
1192 	substring_t		args[MAX_OPT_ARGS];
1193 	char			*p;
1194 	int			error;
1195 
1196 	sync_filesystem(sb);
1197 	while ((p = strsep(&options, ",")) != NULL) {
1198 		int token;
1199 
1200 		if (!*p)
1201 			continue;
1202 
1203 		token = match_token(p, tokens, args);
1204 		switch (token) {
1205 		case Opt_barrier:
1206 			mp->m_flags |= XFS_MOUNT_BARRIER;
1207 			break;
1208 		case Opt_nobarrier:
1209 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
1210 			break;
1211 		case Opt_inode64:
1212 			mp->m_maxagi = xfs_set_inode64(mp, sbp->sb_agcount);
1213 			break;
1214 		case Opt_inode32:
1215 			mp->m_maxagi = xfs_set_inode32(mp, sbp->sb_agcount);
1216 			break;
1217 		default:
1218 			/*
1219 			 * Logically we would return an error here to prevent
1220 			 * users from believing they might have changed
1221 			 * mount options using remount which can't be changed.
1222 			 *
1223 			 * But unfortunately mount(8) adds all options from
1224 			 * mtab and fstab to the mount arguments in some cases
1225 			 * so we can't blindly reject options, but have to
1226 			 * check for each specified option if it actually
1227 			 * differs from the currently set option and only
1228 			 * reject it if that's the case.
1229 			 *
1230 			 * Until that is implemented we return success for
1231 			 * every remount request, and silently ignore all
1232 			 * options that we can't actually change.
1233 			 */
1234 #if 0
1235 			xfs_info(mp,
1236 		"mount option \"%s\" not supported for remount", p);
1237 			return -EINVAL;
1238 #else
1239 			break;
1240 #endif
1241 		}
1242 	}
1243 
1244 	/* ro -> rw */
1245 	if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1246 		mp->m_flags &= ~XFS_MOUNT_RDONLY;
1247 
1248 		/*
1249 		 * If this is the first remount to writeable state we
1250 		 * might have some superblock changes to update.
1251 		 */
1252 		if (mp->m_update_flags) {
1253 			error = xfs_mount_log_sb(mp, mp->m_update_flags);
1254 			if (error) {
1255 				xfs_warn(mp, "failed to write sb changes");
1256 				return error;
1257 			}
1258 			mp->m_update_flags = 0;
1259 		}
1260 
1261 		/*
1262 		 * Fill out the reserve pool if it is empty. Use the stashed
1263 		 * value if it is non-zero, otherwise go with the default.
1264 		 */
1265 		xfs_restore_resvblks(mp);
1266 		xfs_log_work_queue(mp);
1267 	}
1268 
1269 	/* rw -> ro */
1270 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1271 		/*
1272 		 * Before we sync the metadata, we need to free up the reserve
1273 		 * block pool so that the used block count in the superblock on
1274 		 * disk is correct at the end of the remount. Stash the current
1275 		 * reserve pool size so that if we get remounted rw, we can
1276 		 * return it to the same size.
1277 		 */
1278 		xfs_save_resvblks(mp);
1279 		xfs_quiesce_attr(mp);
1280 		mp->m_flags |= XFS_MOUNT_RDONLY;
1281 	}
1282 
1283 	return 0;
1284 }
1285 
1286 /*
1287  * Second stage of a freeze. The data is already frozen so we only
1288  * need to take care of the metadata. Once that's done write a dummy
1289  * record to dirty the log in case of a crash while frozen.
1290  */
1291 STATIC int
1292 xfs_fs_freeze(
1293 	struct super_block	*sb)
1294 {
1295 	struct xfs_mount	*mp = XFS_M(sb);
1296 
1297 	xfs_save_resvblks(mp);
1298 	xfs_quiesce_attr(mp);
1299 	return xfs_fs_log_dummy(mp);
1300 }
1301 
1302 STATIC int
1303 xfs_fs_unfreeze(
1304 	struct super_block	*sb)
1305 {
1306 	struct xfs_mount	*mp = XFS_M(sb);
1307 
1308 	xfs_restore_resvblks(mp);
1309 	xfs_log_work_queue(mp);
1310 	return 0;
1311 }
1312 
1313 STATIC int
1314 xfs_fs_show_options(
1315 	struct seq_file		*m,
1316 	struct dentry		*root)
1317 {
1318 	return xfs_showargs(XFS_M(root->d_sb), m);
1319 }
1320 
1321 /*
1322  * This function fills in xfs_mount_t fields based on mount args.
1323  * Note: the superblock _has_ now been read in.
1324  */
1325 STATIC int
1326 xfs_finish_flags(
1327 	struct xfs_mount	*mp)
1328 {
1329 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1330 
1331 	/* Fail a mount where the logbuf is smaller than the log stripe */
1332 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1333 		if (mp->m_logbsize <= 0 &&
1334 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1335 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1336 		} else if (mp->m_logbsize > 0 &&
1337 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1338 			xfs_warn(mp,
1339 		"logbuf size must be greater than or equal to log stripe size");
1340 			return -EINVAL;
1341 		}
1342 	} else {
1343 		/* Fail a mount if the logbuf is larger than 32K */
1344 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1345 			xfs_warn(mp,
1346 		"logbuf size for version 1 logs must be 16K or 32K");
1347 			return -EINVAL;
1348 		}
1349 	}
1350 
1351 	/*
1352 	 * V5 filesystems always use attr2 format for attributes.
1353 	 */
1354 	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1355 	    (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1356 		xfs_warn(mp,
1357 "Cannot mount a V5 filesystem as %s. %s is always enabled for V5 filesystems.",
1358 			MNTOPT_NOATTR2, MNTOPT_ATTR2);
1359 		return -EINVAL;
1360 	}
1361 
1362 	/*
1363 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1364 	 * told by noattr2 to turn it off
1365 	 */
1366 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1367 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1368 		mp->m_flags |= XFS_MOUNT_ATTR2;
1369 
1370 	/*
1371 	 * prohibit r/w mounts of read-only filesystems
1372 	 */
1373 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1374 		xfs_warn(mp,
1375 			"cannot mount a read-only filesystem as read-write");
1376 		return -EROFS;
1377 	}
1378 
1379 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1380 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1381 	    !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1382 		xfs_warn(mp,
1383 		  "Super block does not support project and group quota together");
1384 		return -EINVAL;
1385 	}
1386 
1387 	return 0;
1388 }
1389 
1390 STATIC int
1391 xfs_fs_fill_super(
1392 	struct super_block	*sb,
1393 	void			*data,
1394 	int			silent)
1395 {
1396 	struct inode		*root;
1397 	struct xfs_mount	*mp = NULL;
1398 	int			flags = 0, error = -ENOMEM;
1399 
1400 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1401 	if (!mp)
1402 		goto out;
1403 
1404 	spin_lock_init(&mp->m_sb_lock);
1405 	mutex_init(&mp->m_growlock);
1406 	atomic_set(&mp->m_active_trans, 0);
1407 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1408 	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1409 
1410 	mp->m_super = sb;
1411 	sb->s_fs_info = mp;
1412 
1413 	error = xfs_parseargs(mp, (char *)data);
1414 	if (error)
1415 		goto out_free_fsname;
1416 
1417 	sb_min_blocksize(sb, BBSIZE);
1418 	sb->s_xattr = xfs_xattr_handlers;
1419 	sb->s_export_op = &xfs_export_operations;
1420 #ifdef CONFIG_XFS_QUOTA
1421 	sb->s_qcop = &xfs_quotactl_operations;
1422 #endif
1423 	sb->s_op = &xfs_super_operations;
1424 
1425 	if (silent)
1426 		flags |= XFS_MFSI_QUIET;
1427 
1428 	error = xfs_open_devices(mp);
1429 	if (error)
1430 		goto out_free_fsname;
1431 
1432 	error = xfs_init_mount_workqueues(mp);
1433 	if (error)
1434 		goto out_close_devices;
1435 
1436 	error = xfs_icsb_init_counters(mp);
1437 	if (error)
1438 		goto out_destroy_workqueues;
1439 
1440 	error = xfs_readsb(mp, flags);
1441 	if (error)
1442 		goto out_destroy_counters;
1443 
1444 	error = xfs_finish_flags(mp);
1445 	if (error)
1446 		goto out_free_sb;
1447 
1448 	error = xfs_setup_devices(mp);
1449 	if (error)
1450 		goto out_free_sb;
1451 
1452 	error = xfs_filestream_mount(mp);
1453 	if (error)
1454 		goto out_free_sb;
1455 
1456 	/*
1457 	 * we must configure the block size in the superblock before we run the
1458 	 * full mount process as the mount process can lookup and cache inodes.
1459 	 */
1460 	sb->s_magic = XFS_SB_MAGIC;
1461 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1462 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1463 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1464 	sb->s_max_links = XFS_MAXLINK;
1465 	sb->s_time_gran = 1;
1466 	set_posix_acl_flag(sb);
1467 
1468 	/* version 5 superblocks support inode version counters. */
1469 	if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1470 		sb->s_flags |= MS_I_VERSION;
1471 
1472 	error = xfs_mountfs(mp);
1473 	if (error)
1474 		goto out_filestream_unmount;
1475 
1476 	root = igrab(VFS_I(mp->m_rootip));
1477 	if (!root) {
1478 		error = -ENOENT;
1479 		goto out_unmount;
1480 	}
1481 	sb->s_root = d_make_root(root);
1482 	if (!sb->s_root) {
1483 		error = -ENOMEM;
1484 		goto out_unmount;
1485 	}
1486 
1487 	return 0;
1488 
1489  out_filestream_unmount:
1490 	xfs_filestream_unmount(mp);
1491  out_free_sb:
1492 	xfs_freesb(mp);
1493  out_destroy_counters:
1494 	xfs_icsb_destroy_counters(mp);
1495 out_destroy_workqueues:
1496 	xfs_destroy_mount_workqueues(mp);
1497  out_close_devices:
1498 	xfs_close_devices(mp);
1499  out_free_fsname:
1500 	xfs_free_fsname(mp);
1501 	kfree(mp);
1502  out:
1503 	return error;
1504 
1505  out_unmount:
1506 	xfs_filestream_unmount(mp);
1507 	xfs_unmountfs(mp);
1508 	goto out_free_sb;
1509 }
1510 
1511 STATIC struct dentry *
1512 xfs_fs_mount(
1513 	struct file_system_type	*fs_type,
1514 	int			flags,
1515 	const char		*dev_name,
1516 	void			*data)
1517 {
1518 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1519 }
1520 
1521 static long
1522 xfs_fs_nr_cached_objects(
1523 	struct super_block	*sb,
1524 	int			nid)
1525 {
1526 	return xfs_reclaim_inodes_count(XFS_M(sb));
1527 }
1528 
1529 static long
1530 xfs_fs_free_cached_objects(
1531 	struct super_block	*sb,
1532 	long			nr_to_scan,
1533 	int			nid)
1534 {
1535 	return xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1536 }
1537 
1538 static const struct super_operations xfs_super_operations = {
1539 	.alloc_inode		= xfs_fs_alloc_inode,
1540 	.destroy_inode		= xfs_fs_destroy_inode,
1541 	.evict_inode		= xfs_fs_evict_inode,
1542 	.drop_inode		= xfs_fs_drop_inode,
1543 	.put_super		= xfs_fs_put_super,
1544 	.sync_fs		= xfs_fs_sync_fs,
1545 	.freeze_fs		= xfs_fs_freeze,
1546 	.unfreeze_fs		= xfs_fs_unfreeze,
1547 	.statfs			= xfs_fs_statfs,
1548 	.remount_fs		= xfs_fs_remount,
1549 	.show_options		= xfs_fs_show_options,
1550 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1551 	.free_cached_objects	= xfs_fs_free_cached_objects,
1552 };
1553 
1554 static struct file_system_type xfs_fs_type = {
1555 	.owner			= THIS_MODULE,
1556 	.name			= "xfs",
1557 	.mount			= xfs_fs_mount,
1558 	.kill_sb		= kill_block_super,
1559 	.fs_flags		= FS_REQUIRES_DEV,
1560 };
1561 MODULE_ALIAS_FS("xfs");
1562 
1563 STATIC int __init
1564 xfs_init_zones(void)
1565 {
1566 
1567 	xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1568 	if (!xfs_ioend_zone)
1569 		goto out;
1570 
1571 	xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1572 						  xfs_ioend_zone);
1573 	if (!xfs_ioend_pool)
1574 		goto out_destroy_ioend_zone;
1575 
1576 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1577 						"xfs_log_ticket");
1578 	if (!xfs_log_ticket_zone)
1579 		goto out_destroy_ioend_pool;
1580 
1581 	xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1582 						"xfs_bmap_free_item");
1583 	if (!xfs_bmap_free_item_zone)
1584 		goto out_destroy_log_ticket_zone;
1585 
1586 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1587 						"xfs_btree_cur");
1588 	if (!xfs_btree_cur_zone)
1589 		goto out_destroy_bmap_free_item_zone;
1590 
1591 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1592 						"xfs_da_state");
1593 	if (!xfs_da_state_zone)
1594 		goto out_destroy_btree_cur_zone;
1595 
1596 	xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1597 	if (!xfs_ifork_zone)
1598 		goto out_destroy_da_state_zone;
1599 
1600 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1601 	if (!xfs_trans_zone)
1602 		goto out_destroy_ifork_zone;
1603 
1604 	xfs_log_item_desc_zone =
1605 		kmem_zone_init(sizeof(struct xfs_log_item_desc),
1606 			       "xfs_log_item_desc");
1607 	if (!xfs_log_item_desc_zone)
1608 		goto out_destroy_trans_zone;
1609 
1610 	/*
1611 	 * The size of the zone allocated buf log item is the maximum
1612 	 * size possible under XFS.  This wastes a little bit of memory,
1613 	 * but it is much faster.
1614 	 */
1615 	xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1616 					   "xfs_buf_item");
1617 	if (!xfs_buf_item_zone)
1618 		goto out_destroy_log_item_desc_zone;
1619 
1620 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1621 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1622 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1623 	if (!xfs_efd_zone)
1624 		goto out_destroy_buf_item_zone;
1625 
1626 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1627 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1628 				sizeof(xfs_extent_t))), "xfs_efi_item");
1629 	if (!xfs_efi_zone)
1630 		goto out_destroy_efd_zone;
1631 
1632 	xfs_inode_zone =
1633 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1634 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1635 			xfs_fs_inode_init_once);
1636 	if (!xfs_inode_zone)
1637 		goto out_destroy_efi_zone;
1638 
1639 	xfs_ili_zone =
1640 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1641 					KM_ZONE_SPREAD, NULL);
1642 	if (!xfs_ili_zone)
1643 		goto out_destroy_inode_zone;
1644 	xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1645 					"xfs_icr");
1646 	if (!xfs_icreate_zone)
1647 		goto out_destroy_ili_zone;
1648 
1649 	return 0;
1650 
1651  out_destroy_ili_zone:
1652 	kmem_zone_destroy(xfs_ili_zone);
1653  out_destroy_inode_zone:
1654 	kmem_zone_destroy(xfs_inode_zone);
1655  out_destroy_efi_zone:
1656 	kmem_zone_destroy(xfs_efi_zone);
1657  out_destroy_efd_zone:
1658 	kmem_zone_destroy(xfs_efd_zone);
1659  out_destroy_buf_item_zone:
1660 	kmem_zone_destroy(xfs_buf_item_zone);
1661  out_destroy_log_item_desc_zone:
1662 	kmem_zone_destroy(xfs_log_item_desc_zone);
1663  out_destroy_trans_zone:
1664 	kmem_zone_destroy(xfs_trans_zone);
1665  out_destroy_ifork_zone:
1666 	kmem_zone_destroy(xfs_ifork_zone);
1667  out_destroy_da_state_zone:
1668 	kmem_zone_destroy(xfs_da_state_zone);
1669  out_destroy_btree_cur_zone:
1670 	kmem_zone_destroy(xfs_btree_cur_zone);
1671  out_destroy_bmap_free_item_zone:
1672 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1673  out_destroy_log_ticket_zone:
1674 	kmem_zone_destroy(xfs_log_ticket_zone);
1675  out_destroy_ioend_pool:
1676 	mempool_destroy(xfs_ioend_pool);
1677  out_destroy_ioend_zone:
1678 	kmem_zone_destroy(xfs_ioend_zone);
1679  out:
1680 	return -ENOMEM;
1681 }
1682 
1683 STATIC void
1684 xfs_destroy_zones(void)
1685 {
1686 	/*
1687 	 * Make sure all delayed rcu free are flushed before we
1688 	 * destroy caches.
1689 	 */
1690 	rcu_barrier();
1691 	kmem_zone_destroy(xfs_icreate_zone);
1692 	kmem_zone_destroy(xfs_ili_zone);
1693 	kmem_zone_destroy(xfs_inode_zone);
1694 	kmem_zone_destroy(xfs_efi_zone);
1695 	kmem_zone_destroy(xfs_efd_zone);
1696 	kmem_zone_destroy(xfs_buf_item_zone);
1697 	kmem_zone_destroy(xfs_log_item_desc_zone);
1698 	kmem_zone_destroy(xfs_trans_zone);
1699 	kmem_zone_destroy(xfs_ifork_zone);
1700 	kmem_zone_destroy(xfs_da_state_zone);
1701 	kmem_zone_destroy(xfs_btree_cur_zone);
1702 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1703 	kmem_zone_destroy(xfs_log_ticket_zone);
1704 	mempool_destroy(xfs_ioend_pool);
1705 	kmem_zone_destroy(xfs_ioend_zone);
1706 
1707 }
1708 
1709 STATIC int __init
1710 xfs_init_workqueues(void)
1711 {
1712 	/*
1713 	 * The allocation workqueue can be used in memory reclaim situations
1714 	 * (writepage path), and parallelism is only limited by the number of
1715 	 * AGs in all the filesystems mounted. Hence use the default large
1716 	 * max_active value for this workqueue.
1717 	 */
1718 	xfs_alloc_wq = alloc_workqueue("xfsalloc", WQ_MEM_RECLAIM, 0);
1719 	if (!xfs_alloc_wq)
1720 		return -ENOMEM;
1721 
1722 	return 0;
1723 }
1724 
1725 STATIC void
1726 xfs_destroy_workqueues(void)
1727 {
1728 	destroy_workqueue(xfs_alloc_wq);
1729 }
1730 
1731 STATIC int __init
1732 init_xfs_fs(void)
1733 {
1734 	int			error;
1735 
1736 	printk(KERN_INFO XFS_VERSION_STRING " with "
1737 			 XFS_BUILD_OPTIONS " enabled\n");
1738 
1739 	xfs_dir_startup();
1740 
1741 	error = xfs_init_zones();
1742 	if (error)
1743 		goto out;
1744 
1745 	error = xfs_init_workqueues();
1746 	if (error)
1747 		goto out_destroy_zones;
1748 
1749 	error = xfs_mru_cache_init();
1750 	if (error)
1751 		goto out_destroy_wq;
1752 
1753 	error = xfs_buf_init();
1754 	if (error)
1755 		goto out_mru_cache_uninit;
1756 
1757 	error = xfs_init_procfs();
1758 	if (error)
1759 		goto out_buf_terminate;
1760 
1761 	error = xfs_sysctl_register();
1762 	if (error)
1763 		goto out_cleanup_procfs;
1764 
1765 	xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
1766 	if (!xfs_kset) {
1767 		error = -ENOMEM;
1768 		goto out_sysctl_unregister;;
1769 	}
1770 
1771 	error = xfs_qm_init();
1772 	if (error)
1773 		goto out_kset_unregister;
1774 
1775 	error = register_filesystem(&xfs_fs_type);
1776 	if (error)
1777 		goto out_qm_exit;
1778 	return 0;
1779 
1780  out_qm_exit:
1781 	xfs_qm_exit();
1782  out_kset_unregister:
1783 	kset_unregister(xfs_kset);
1784  out_sysctl_unregister:
1785 	xfs_sysctl_unregister();
1786  out_cleanup_procfs:
1787 	xfs_cleanup_procfs();
1788  out_buf_terminate:
1789 	xfs_buf_terminate();
1790  out_mru_cache_uninit:
1791 	xfs_mru_cache_uninit();
1792  out_destroy_wq:
1793 	xfs_destroy_workqueues();
1794  out_destroy_zones:
1795 	xfs_destroy_zones();
1796  out:
1797 	return error;
1798 }
1799 
1800 STATIC void __exit
1801 exit_xfs_fs(void)
1802 {
1803 	xfs_qm_exit();
1804 	unregister_filesystem(&xfs_fs_type);
1805 	kset_unregister(xfs_kset);
1806 	xfs_sysctl_unregister();
1807 	xfs_cleanup_procfs();
1808 	xfs_buf_terminate();
1809 	xfs_mru_cache_uninit();
1810 	xfs_destroy_workqueues();
1811 	xfs_destroy_zones();
1812 }
1813 
1814 module_init(init_xfs_fs);
1815 module_exit(exit_xfs_fs);
1816 
1817 MODULE_AUTHOR("Silicon Graphics, Inc.");
1818 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1819 MODULE_LICENSE("GPL");
1820