xref: /linux/fs/xfs/xfs_super.c (revision b43ab901d671e3e3cad425ea5e9a3c74e266dcdd)
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_bit.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_dir2.h"
27 #include "xfs_alloc.h"
28 #include "xfs_quota.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_btree.h"
36 #include "xfs_ialloc.h"
37 #include "xfs_bmap.h"
38 #include "xfs_rtalloc.h"
39 #include "xfs_error.h"
40 #include "xfs_itable.h"
41 #include "xfs_fsops.h"
42 #include "xfs_attr.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_utils.h"
45 #include "xfs_vnodeops.h"
46 #include "xfs_log_priv.h"
47 #include "xfs_trans_priv.h"
48 #include "xfs_filestream.h"
49 #include "xfs_da_btree.h"
50 #include "xfs_extfree_item.h"
51 #include "xfs_mru_cache.h"
52 #include "xfs_inode_item.h"
53 #include "xfs_sync.h"
54 #include "xfs_trace.h"
55 
56 #include <linux/namei.h>
57 #include <linux/init.h>
58 #include <linux/slab.h>
59 #include <linux/mount.h>
60 #include <linux/mempool.h>
61 #include <linux/writeback.h>
62 #include <linux/kthread.h>
63 #include <linux/freezer.h>
64 #include <linux/parser.h>
65 
66 static const struct super_operations xfs_super_operations;
67 static kmem_zone_t *xfs_ioend_zone;
68 mempool_t *xfs_ioend_pool;
69 
70 #define MNTOPT_LOGBUFS	"logbufs"	/* number of XFS log buffers */
71 #define MNTOPT_LOGBSIZE	"logbsize"	/* size of XFS log buffers */
72 #define MNTOPT_LOGDEV	"logdev"	/* log device */
73 #define MNTOPT_RTDEV	"rtdev"		/* realtime I/O device */
74 #define MNTOPT_BIOSIZE	"biosize"	/* log2 of preferred buffered io size */
75 #define MNTOPT_WSYNC	"wsync"		/* safe-mode nfs compatible mount */
76 #define MNTOPT_NOALIGN	"noalign"	/* turn off stripe alignment */
77 #define MNTOPT_SWALLOC	"swalloc"	/* turn on stripe width allocation */
78 #define MNTOPT_SUNIT	"sunit"		/* data volume stripe unit */
79 #define MNTOPT_SWIDTH	"swidth"	/* data volume stripe width */
80 #define MNTOPT_NOUUID	"nouuid"	/* ignore filesystem UUID */
81 #define MNTOPT_MTPT	"mtpt"		/* filesystem mount point */
82 #define MNTOPT_GRPID	"grpid"		/* group-ID from parent directory */
83 #define MNTOPT_NOGRPID	"nogrpid"	/* group-ID from current process */
84 #define MNTOPT_BSDGROUPS    "bsdgroups"    /* group-ID from parent directory */
85 #define MNTOPT_SYSVGROUPS   "sysvgroups"   /* group-ID from current process */
86 #define MNTOPT_ALLOCSIZE    "allocsize"    /* preferred allocation size */
87 #define MNTOPT_NORECOVERY   "norecovery"   /* don't run XFS recovery */
88 #define MNTOPT_BARRIER	"barrier"	/* use writer barriers for log write and
89 					 * unwritten extent conversion */
90 #define MNTOPT_NOBARRIER "nobarrier"	/* .. disable */
91 #define MNTOPT_64BITINODE   "inode64"	/* inodes can be allocated anywhere */
92 #define MNTOPT_IKEEP	"ikeep"		/* do not free empty inode clusters */
93 #define MNTOPT_NOIKEEP	"noikeep"	/* free empty inode clusters */
94 #define MNTOPT_LARGEIO	   "largeio"	/* report large I/O sizes in stat() */
95 #define MNTOPT_NOLARGEIO   "nolargeio"	/* do not report large I/O sizes
96 					 * in stat(). */
97 #define MNTOPT_ATTR2	"attr2"		/* do use attr2 attribute format */
98 #define MNTOPT_NOATTR2	"noattr2"	/* do not use attr2 attribute format */
99 #define MNTOPT_FILESTREAM  "filestreams" /* use filestreams allocator */
100 #define MNTOPT_QUOTA	"quota"		/* disk quotas (user) */
101 #define MNTOPT_NOQUOTA	"noquota"	/* no quotas */
102 #define MNTOPT_USRQUOTA	"usrquota"	/* user quota enabled */
103 #define MNTOPT_GRPQUOTA	"grpquota"	/* group quota enabled */
104 #define MNTOPT_PRJQUOTA	"prjquota"	/* project quota enabled */
105 #define MNTOPT_UQUOTA	"uquota"	/* user quota (IRIX variant) */
106 #define MNTOPT_GQUOTA	"gquota"	/* group quota (IRIX variant) */
107 #define MNTOPT_PQUOTA	"pquota"	/* project quota (IRIX variant) */
108 #define MNTOPT_UQUOTANOENF "uqnoenforce"/* user quota limit enforcement */
109 #define MNTOPT_GQUOTANOENF "gqnoenforce"/* group quota limit enforcement */
110 #define MNTOPT_PQUOTANOENF "pqnoenforce"/* project quota limit enforcement */
111 #define MNTOPT_QUOTANOENF  "qnoenforce"	/* same as uqnoenforce */
112 #define MNTOPT_DELAYLOG    "delaylog"	/* Delayed logging enabled */
113 #define MNTOPT_NODELAYLOG  "nodelaylog"	/* Delayed logging disabled */
114 #define MNTOPT_DISCARD	   "discard"	/* Discard unused blocks */
115 #define MNTOPT_NODISCARD   "nodiscard"	/* Do not discard unused blocks */
116 
117 /*
118  * Table driven mount option parser.
119  *
120  * Currently only used for remount, but it will be used for mount
121  * in the future, too.
122  */
123 enum {
124 	Opt_barrier, Opt_nobarrier, Opt_err
125 };
126 
127 static const match_table_t tokens = {
128 	{Opt_barrier, "barrier"},
129 	{Opt_nobarrier, "nobarrier"},
130 	{Opt_err, NULL}
131 };
132 
133 
134 STATIC unsigned long
135 suffix_strtoul(char *s, char **endp, unsigned int base)
136 {
137 	int	last, shift_left_factor = 0;
138 	char	*value = s;
139 
140 	last = strlen(value) - 1;
141 	if (value[last] == 'K' || value[last] == 'k') {
142 		shift_left_factor = 10;
143 		value[last] = '\0';
144 	}
145 	if (value[last] == 'M' || value[last] == 'm') {
146 		shift_left_factor = 20;
147 		value[last] = '\0';
148 	}
149 	if (value[last] == 'G' || value[last] == 'g') {
150 		shift_left_factor = 30;
151 		value[last] = '\0';
152 	}
153 
154 	return simple_strtoul((const char *)s, endp, base) << shift_left_factor;
155 }
156 
157 /*
158  * This function fills in xfs_mount_t fields based on mount args.
159  * Note: the superblock has _not_ yet been read in.
160  *
161  * Note that this function leaks the various device name allocations on
162  * failure.  The caller takes care of them.
163  */
164 STATIC int
165 xfs_parseargs(
166 	struct xfs_mount	*mp,
167 	char			*options)
168 {
169 	struct super_block	*sb = mp->m_super;
170 	char			*this_char, *value, *eov;
171 	int			dsunit = 0;
172 	int			dswidth = 0;
173 	int			iosize = 0;
174 	__uint8_t		iosizelog = 0;
175 
176 	/*
177 	 * set up the mount name first so all the errors will refer to the
178 	 * correct device.
179 	 */
180 	mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
181 	if (!mp->m_fsname)
182 		return ENOMEM;
183 	mp->m_fsname_len = strlen(mp->m_fsname) + 1;
184 
185 	/*
186 	 * Copy binary VFS mount flags we are interested in.
187 	 */
188 	if (sb->s_flags & MS_RDONLY)
189 		mp->m_flags |= XFS_MOUNT_RDONLY;
190 	if (sb->s_flags & MS_DIRSYNC)
191 		mp->m_flags |= XFS_MOUNT_DIRSYNC;
192 	if (sb->s_flags & MS_SYNCHRONOUS)
193 		mp->m_flags |= XFS_MOUNT_WSYNC;
194 
195 	/*
196 	 * Set some default flags that could be cleared by the mount option
197 	 * parsing.
198 	 */
199 	mp->m_flags |= XFS_MOUNT_BARRIER;
200 	mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
201 	mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
202 
203 	/*
204 	 * These can be overridden by the mount option parsing.
205 	 */
206 	mp->m_logbufs = -1;
207 	mp->m_logbsize = -1;
208 
209 	if (!options)
210 		goto done;
211 
212 	while ((this_char = strsep(&options, ",")) != NULL) {
213 		if (!*this_char)
214 			continue;
215 		if ((value = strchr(this_char, '=')) != NULL)
216 			*value++ = 0;
217 
218 		if (!strcmp(this_char, MNTOPT_LOGBUFS)) {
219 			if (!value || !*value) {
220 				xfs_warn(mp, "%s option requires an argument",
221 					this_char);
222 				return EINVAL;
223 			}
224 			mp->m_logbufs = simple_strtoul(value, &eov, 10);
225 		} else if (!strcmp(this_char, MNTOPT_LOGBSIZE)) {
226 			if (!value || !*value) {
227 				xfs_warn(mp, "%s option requires an argument",
228 					this_char);
229 				return EINVAL;
230 			}
231 			mp->m_logbsize = suffix_strtoul(value, &eov, 10);
232 		} else if (!strcmp(this_char, MNTOPT_LOGDEV)) {
233 			if (!value || !*value) {
234 				xfs_warn(mp, "%s option requires an argument",
235 					this_char);
236 				return EINVAL;
237 			}
238 			mp->m_logname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
239 			if (!mp->m_logname)
240 				return ENOMEM;
241 		} else if (!strcmp(this_char, MNTOPT_MTPT)) {
242 			xfs_warn(mp, "%s option not allowed on this system",
243 				this_char);
244 			return EINVAL;
245 		} else if (!strcmp(this_char, MNTOPT_RTDEV)) {
246 			if (!value || !*value) {
247 				xfs_warn(mp, "%s option requires an argument",
248 					this_char);
249 				return EINVAL;
250 			}
251 			mp->m_rtname = kstrndup(value, MAXNAMELEN, GFP_KERNEL);
252 			if (!mp->m_rtname)
253 				return ENOMEM;
254 		} else if (!strcmp(this_char, MNTOPT_BIOSIZE)) {
255 			if (!value || !*value) {
256 				xfs_warn(mp, "%s option requires an argument",
257 					this_char);
258 				return EINVAL;
259 			}
260 			iosize = simple_strtoul(value, &eov, 10);
261 			iosizelog = ffs(iosize) - 1;
262 		} else if (!strcmp(this_char, MNTOPT_ALLOCSIZE)) {
263 			if (!value || !*value) {
264 				xfs_warn(mp, "%s option requires an argument",
265 					this_char);
266 				return EINVAL;
267 			}
268 			iosize = suffix_strtoul(value, &eov, 10);
269 			iosizelog = ffs(iosize) - 1;
270 		} else if (!strcmp(this_char, MNTOPT_GRPID) ||
271 			   !strcmp(this_char, MNTOPT_BSDGROUPS)) {
272 			mp->m_flags |= XFS_MOUNT_GRPID;
273 		} else if (!strcmp(this_char, MNTOPT_NOGRPID) ||
274 			   !strcmp(this_char, MNTOPT_SYSVGROUPS)) {
275 			mp->m_flags &= ~XFS_MOUNT_GRPID;
276 		} else if (!strcmp(this_char, MNTOPT_WSYNC)) {
277 			mp->m_flags |= XFS_MOUNT_WSYNC;
278 		} else if (!strcmp(this_char, MNTOPT_NORECOVERY)) {
279 			mp->m_flags |= XFS_MOUNT_NORECOVERY;
280 		} else if (!strcmp(this_char, MNTOPT_NOALIGN)) {
281 			mp->m_flags |= XFS_MOUNT_NOALIGN;
282 		} else if (!strcmp(this_char, MNTOPT_SWALLOC)) {
283 			mp->m_flags |= XFS_MOUNT_SWALLOC;
284 		} else if (!strcmp(this_char, MNTOPT_SUNIT)) {
285 			if (!value || !*value) {
286 				xfs_warn(mp, "%s option requires an argument",
287 					this_char);
288 				return EINVAL;
289 			}
290 			dsunit = simple_strtoul(value, &eov, 10);
291 		} else if (!strcmp(this_char, MNTOPT_SWIDTH)) {
292 			if (!value || !*value) {
293 				xfs_warn(mp, "%s option requires an argument",
294 					this_char);
295 				return EINVAL;
296 			}
297 			dswidth = simple_strtoul(value, &eov, 10);
298 		} else if (!strcmp(this_char, MNTOPT_64BITINODE)) {
299 			mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
300 #if !XFS_BIG_INUMS
301 			xfs_warn(mp, "%s option not allowed on this system",
302 				this_char);
303 			return EINVAL;
304 #endif
305 		} else if (!strcmp(this_char, MNTOPT_NOUUID)) {
306 			mp->m_flags |= XFS_MOUNT_NOUUID;
307 		} else if (!strcmp(this_char, MNTOPT_BARRIER)) {
308 			mp->m_flags |= XFS_MOUNT_BARRIER;
309 		} else if (!strcmp(this_char, MNTOPT_NOBARRIER)) {
310 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
311 		} else if (!strcmp(this_char, MNTOPT_IKEEP)) {
312 			mp->m_flags |= XFS_MOUNT_IKEEP;
313 		} else if (!strcmp(this_char, MNTOPT_NOIKEEP)) {
314 			mp->m_flags &= ~XFS_MOUNT_IKEEP;
315 		} else if (!strcmp(this_char, MNTOPT_LARGEIO)) {
316 			mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
317 		} else if (!strcmp(this_char, MNTOPT_NOLARGEIO)) {
318 			mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
319 		} else if (!strcmp(this_char, MNTOPT_ATTR2)) {
320 			mp->m_flags |= XFS_MOUNT_ATTR2;
321 		} else if (!strcmp(this_char, MNTOPT_NOATTR2)) {
322 			mp->m_flags &= ~XFS_MOUNT_ATTR2;
323 			mp->m_flags |= XFS_MOUNT_NOATTR2;
324 		} else if (!strcmp(this_char, MNTOPT_FILESTREAM)) {
325 			mp->m_flags |= XFS_MOUNT_FILESTREAMS;
326 		} else if (!strcmp(this_char, MNTOPT_NOQUOTA)) {
327 			mp->m_qflags &= ~(XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
328 					  XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
329 					  XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
330 					  XFS_UQUOTA_ENFD | XFS_OQUOTA_ENFD);
331 		} else if (!strcmp(this_char, MNTOPT_QUOTA) ||
332 			   !strcmp(this_char, MNTOPT_UQUOTA) ||
333 			   !strcmp(this_char, MNTOPT_USRQUOTA)) {
334 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
335 					 XFS_UQUOTA_ENFD);
336 		} else if (!strcmp(this_char, MNTOPT_QUOTANOENF) ||
337 			   !strcmp(this_char, MNTOPT_UQUOTANOENF)) {
338 			mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
339 			mp->m_qflags &= ~XFS_UQUOTA_ENFD;
340 		} else if (!strcmp(this_char, MNTOPT_PQUOTA) ||
341 			   !strcmp(this_char, MNTOPT_PRJQUOTA)) {
342 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
343 					 XFS_OQUOTA_ENFD);
344 		} else if (!strcmp(this_char, MNTOPT_PQUOTANOENF)) {
345 			mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
346 			mp->m_qflags &= ~XFS_OQUOTA_ENFD;
347 		} else if (!strcmp(this_char, MNTOPT_GQUOTA) ||
348 			   !strcmp(this_char, MNTOPT_GRPQUOTA)) {
349 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
350 					 XFS_OQUOTA_ENFD);
351 		} else if (!strcmp(this_char, MNTOPT_GQUOTANOENF)) {
352 			mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
353 			mp->m_qflags &= ~XFS_OQUOTA_ENFD;
354 		} else if (!strcmp(this_char, MNTOPT_DELAYLOG)) {
355 			xfs_warn(mp,
356 	"delaylog is the default now, option is deprecated.");
357 		} else if (!strcmp(this_char, MNTOPT_NODELAYLOG)) {
358 			xfs_warn(mp,
359 	"nodelaylog support has been removed, option is deprecated.");
360 		} else if (!strcmp(this_char, MNTOPT_DISCARD)) {
361 			mp->m_flags |= XFS_MOUNT_DISCARD;
362 		} else if (!strcmp(this_char, MNTOPT_NODISCARD)) {
363 			mp->m_flags &= ~XFS_MOUNT_DISCARD;
364 		} else if (!strcmp(this_char, "ihashsize")) {
365 			xfs_warn(mp,
366 	"ihashsize no longer used, option is deprecated.");
367 		} else if (!strcmp(this_char, "osyncisdsync")) {
368 			xfs_warn(mp,
369 	"osyncisdsync has no effect, option is deprecated.");
370 		} else if (!strcmp(this_char, "osyncisosync")) {
371 			xfs_warn(mp,
372 	"osyncisosync has no effect, option is deprecated.");
373 		} else if (!strcmp(this_char, "irixsgid")) {
374 			xfs_warn(mp,
375 	"irixsgid is now a sysctl(2) variable, option is deprecated.");
376 		} else {
377 			xfs_warn(mp, "unknown mount option [%s].", this_char);
378 			return EINVAL;
379 		}
380 	}
381 
382 	/*
383 	 * no recovery flag requires a read-only mount
384 	 */
385 	if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
386 	    !(mp->m_flags & XFS_MOUNT_RDONLY)) {
387 		xfs_warn(mp, "no-recovery mounts must be read-only.");
388 		return EINVAL;
389 	}
390 
391 	if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
392 		xfs_warn(mp,
393 	"sunit and swidth options incompatible with the noalign option");
394 		return EINVAL;
395 	}
396 
397 #ifndef CONFIG_XFS_QUOTA
398 	if (XFS_IS_QUOTA_RUNNING(mp)) {
399 		xfs_warn(mp, "quota support not available in this kernel.");
400 		return EINVAL;
401 	}
402 #endif
403 
404 	if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
405 	    (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE))) {
406 		xfs_warn(mp, "cannot mount with both project and group quota");
407 		return EINVAL;
408 	}
409 
410 	if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
411 		xfs_warn(mp, "sunit and swidth must be specified together");
412 		return EINVAL;
413 	}
414 
415 	if (dsunit && (dswidth % dsunit != 0)) {
416 		xfs_warn(mp,
417 	"stripe width (%d) must be a multiple of the stripe unit (%d)",
418 			dswidth, dsunit);
419 		return EINVAL;
420 	}
421 
422 done:
423 	if (!(mp->m_flags & XFS_MOUNT_NOALIGN)) {
424 		/*
425 		 * At this point the superblock has not been read
426 		 * in, therefore we do not know the block size.
427 		 * Before the mount call ends we will convert
428 		 * these to FSBs.
429 		 */
430 		if (dsunit) {
431 			mp->m_dalign = dsunit;
432 			mp->m_flags |= XFS_MOUNT_RETERR;
433 		}
434 
435 		if (dswidth)
436 			mp->m_swidth = dswidth;
437 	}
438 
439 	if (mp->m_logbufs != -1 &&
440 	    mp->m_logbufs != 0 &&
441 	    (mp->m_logbufs < XLOG_MIN_ICLOGS ||
442 	     mp->m_logbufs > XLOG_MAX_ICLOGS)) {
443 		xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
444 			mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
445 		return XFS_ERROR(EINVAL);
446 	}
447 	if (mp->m_logbsize != -1 &&
448 	    mp->m_logbsize !=  0 &&
449 	    (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
450 	     mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
451 	     !is_power_of_2(mp->m_logbsize))) {
452 		xfs_warn(mp,
453 			"invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
454 			mp->m_logbsize);
455 		return XFS_ERROR(EINVAL);
456 	}
457 
458 	if (iosizelog) {
459 		if (iosizelog > XFS_MAX_IO_LOG ||
460 		    iosizelog < XFS_MIN_IO_LOG) {
461 			xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
462 				iosizelog, XFS_MIN_IO_LOG,
463 				XFS_MAX_IO_LOG);
464 			return XFS_ERROR(EINVAL);
465 		}
466 
467 		mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
468 		mp->m_readio_log = iosizelog;
469 		mp->m_writeio_log = iosizelog;
470 	}
471 
472 	return 0;
473 }
474 
475 struct proc_xfs_info {
476 	int	flag;
477 	char	*str;
478 };
479 
480 STATIC int
481 xfs_showargs(
482 	struct xfs_mount	*mp,
483 	struct seq_file		*m)
484 {
485 	static struct proc_xfs_info xfs_info_set[] = {
486 		/* the few simple ones we can get from the mount struct */
487 		{ XFS_MOUNT_IKEEP,		"," MNTOPT_IKEEP },
488 		{ XFS_MOUNT_WSYNC,		"," MNTOPT_WSYNC },
489 		{ XFS_MOUNT_NOALIGN,		"," MNTOPT_NOALIGN },
490 		{ XFS_MOUNT_SWALLOC,		"," MNTOPT_SWALLOC },
491 		{ XFS_MOUNT_NOUUID,		"," MNTOPT_NOUUID },
492 		{ XFS_MOUNT_NORECOVERY,		"," MNTOPT_NORECOVERY },
493 		{ XFS_MOUNT_ATTR2,		"," MNTOPT_ATTR2 },
494 		{ XFS_MOUNT_FILESTREAMS,	"," MNTOPT_FILESTREAM },
495 		{ XFS_MOUNT_GRPID,		"," MNTOPT_GRPID },
496 		{ XFS_MOUNT_DISCARD,		"," MNTOPT_DISCARD },
497 		{ 0, NULL }
498 	};
499 	static struct proc_xfs_info xfs_info_unset[] = {
500 		/* the few simple ones we can get from the mount struct */
501 		{ XFS_MOUNT_COMPAT_IOSIZE,	"," MNTOPT_LARGEIO },
502 		{ XFS_MOUNT_BARRIER,		"," MNTOPT_NOBARRIER },
503 		{ XFS_MOUNT_SMALL_INUMS,	"," MNTOPT_64BITINODE },
504 		{ 0, NULL }
505 	};
506 	struct proc_xfs_info	*xfs_infop;
507 
508 	for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
509 		if (mp->m_flags & xfs_infop->flag)
510 			seq_puts(m, xfs_infop->str);
511 	}
512 	for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
513 		if (!(mp->m_flags & xfs_infop->flag))
514 			seq_puts(m, xfs_infop->str);
515 	}
516 
517 	if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
518 		seq_printf(m, "," MNTOPT_ALLOCSIZE "=%dk",
519 				(int)(1 << mp->m_writeio_log) >> 10);
520 
521 	if (mp->m_logbufs > 0)
522 		seq_printf(m, "," MNTOPT_LOGBUFS "=%d", mp->m_logbufs);
523 	if (mp->m_logbsize > 0)
524 		seq_printf(m, "," MNTOPT_LOGBSIZE "=%dk", mp->m_logbsize >> 10);
525 
526 	if (mp->m_logname)
527 		seq_printf(m, "," MNTOPT_LOGDEV "=%s", mp->m_logname);
528 	if (mp->m_rtname)
529 		seq_printf(m, "," MNTOPT_RTDEV "=%s", mp->m_rtname);
530 
531 	if (mp->m_dalign > 0)
532 		seq_printf(m, "," MNTOPT_SUNIT "=%d",
533 				(int)XFS_FSB_TO_BB(mp, mp->m_dalign));
534 	if (mp->m_swidth > 0)
535 		seq_printf(m, "," MNTOPT_SWIDTH "=%d",
536 				(int)XFS_FSB_TO_BB(mp, mp->m_swidth));
537 
538 	if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
539 		seq_puts(m, "," MNTOPT_USRQUOTA);
540 	else if (mp->m_qflags & XFS_UQUOTA_ACCT)
541 		seq_puts(m, "," MNTOPT_UQUOTANOENF);
542 
543 	/* Either project or group quotas can be active, not both */
544 
545 	if (mp->m_qflags & XFS_PQUOTA_ACCT) {
546 		if (mp->m_qflags & XFS_OQUOTA_ENFD)
547 			seq_puts(m, "," MNTOPT_PRJQUOTA);
548 		else
549 			seq_puts(m, "," MNTOPT_PQUOTANOENF);
550 	} else if (mp->m_qflags & XFS_GQUOTA_ACCT) {
551 		if (mp->m_qflags & XFS_OQUOTA_ENFD)
552 			seq_puts(m, "," MNTOPT_GRPQUOTA);
553 		else
554 			seq_puts(m, "," MNTOPT_GQUOTANOENF);
555 	}
556 
557 	if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
558 		seq_puts(m, "," MNTOPT_NOQUOTA);
559 
560 	return 0;
561 }
562 __uint64_t
563 xfs_max_file_offset(
564 	unsigned int		blockshift)
565 {
566 	unsigned int		pagefactor = 1;
567 	unsigned int		bitshift = BITS_PER_LONG - 1;
568 
569 	/* Figure out maximum filesize, on Linux this can depend on
570 	 * the filesystem blocksize (on 32 bit platforms).
571 	 * __block_write_begin does this in an [unsigned] long...
572 	 *      page->index << (PAGE_CACHE_SHIFT - bbits)
573 	 * So, for page sized blocks (4K on 32 bit platforms),
574 	 * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
575 	 *      (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1)
576 	 * but for smaller blocksizes it is less (bbits = log2 bsize).
577 	 * Note1: get_block_t takes a long (implicit cast from above)
578 	 * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
579 	 * can optionally convert the [unsigned] long from above into
580 	 * an [unsigned] long long.
581 	 */
582 
583 #if BITS_PER_LONG == 32
584 # if defined(CONFIG_LBDAF)
585 	ASSERT(sizeof(sector_t) == 8);
586 	pagefactor = PAGE_CACHE_SIZE;
587 	bitshift = BITS_PER_LONG;
588 # else
589 	pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift);
590 # endif
591 #endif
592 
593 	return (((__uint64_t)pagefactor) << bitshift) - 1;
594 }
595 
596 STATIC int
597 xfs_blkdev_get(
598 	xfs_mount_t		*mp,
599 	const char		*name,
600 	struct block_device	**bdevp)
601 {
602 	int			error = 0;
603 
604 	*bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
605 				    mp);
606 	if (IS_ERR(*bdevp)) {
607 		error = PTR_ERR(*bdevp);
608 		xfs_warn(mp, "Invalid device [%s], error=%d\n", name, error);
609 	}
610 
611 	return -error;
612 }
613 
614 STATIC void
615 xfs_blkdev_put(
616 	struct block_device	*bdev)
617 {
618 	if (bdev)
619 		blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
620 }
621 
622 void
623 xfs_blkdev_issue_flush(
624 	xfs_buftarg_t		*buftarg)
625 {
626 	blkdev_issue_flush(buftarg->bt_bdev, GFP_KERNEL, NULL);
627 }
628 
629 STATIC void
630 xfs_close_devices(
631 	struct xfs_mount	*mp)
632 {
633 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
634 		struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
635 		xfs_free_buftarg(mp, mp->m_logdev_targp);
636 		xfs_blkdev_put(logdev);
637 	}
638 	if (mp->m_rtdev_targp) {
639 		struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
640 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
641 		xfs_blkdev_put(rtdev);
642 	}
643 	xfs_free_buftarg(mp, mp->m_ddev_targp);
644 }
645 
646 /*
647  * The file system configurations are:
648  *	(1) device (partition) with data and internal log
649  *	(2) logical volume with data and log subvolumes.
650  *	(3) logical volume with data, log, and realtime subvolumes.
651  *
652  * We only have to handle opening the log and realtime volumes here if
653  * they are present.  The data subvolume has already been opened by
654  * get_sb_bdev() and is stored in sb->s_bdev.
655  */
656 STATIC int
657 xfs_open_devices(
658 	struct xfs_mount	*mp)
659 {
660 	struct block_device	*ddev = mp->m_super->s_bdev;
661 	struct block_device	*logdev = NULL, *rtdev = NULL;
662 	int			error;
663 
664 	/*
665 	 * Open real time and log devices - order is important.
666 	 */
667 	if (mp->m_logname) {
668 		error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
669 		if (error)
670 			goto out;
671 	}
672 
673 	if (mp->m_rtname) {
674 		error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
675 		if (error)
676 			goto out_close_logdev;
677 
678 		if (rtdev == ddev || rtdev == logdev) {
679 			xfs_warn(mp,
680 	"Cannot mount filesystem with identical rtdev and ddev/logdev.");
681 			error = EINVAL;
682 			goto out_close_rtdev;
683 		}
684 	}
685 
686 	/*
687 	 * Setup xfs_mount buffer target pointers
688 	 */
689 	error = ENOMEM;
690 	mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, 0, mp->m_fsname);
691 	if (!mp->m_ddev_targp)
692 		goto out_close_rtdev;
693 
694 	if (rtdev) {
695 		mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, 1,
696 							mp->m_fsname);
697 		if (!mp->m_rtdev_targp)
698 			goto out_free_ddev_targ;
699 	}
700 
701 	if (logdev && logdev != ddev) {
702 		mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, 1,
703 							mp->m_fsname);
704 		if (!mp->m_logdev_targp)
705 			goto out_free_rtdev_targ;
706 	} else {
707 		mp->m_logdev_targp = mp->m_ddev_targp;
708 	}
709 
710 	return 0;
711 
712  out_free_rtdev_targ:
713 	if (mp->m_rtdev_targp)
714 		xfs_free_buftarg(mp, mp->m_rtdev_targp);
715  out_free_ddev_targ:
716 	xfs_free_buftarg(mp, mp->m_ddev_targp);
717  out_close_rtdev:
718 	if (rtdev)
719 		xfs_blkdev_put(rtdev);
720  out_close_logdev:
721 	if (logdev && logdev != ddev)
722 		xfs_blkdev_put(logdev);
723  out:
724 	return error;
725 }
726 
727 /*
728  * Setup xfs_mount buffer target pointers based on superblock
729  */
730 STATIC int
731 xfs_setup_devices(
732 	struct xfs_mount	*mp)
733 {
734 	int			error;
735 
736 	error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_blocksize,
737 				    mp->m_sb.sb_sectsize);
738 	if (error)
739 		return error;
740 
741 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
742 		unsigned int	log_sector_size = BBSIZE;
743 
744 		if (xfs_sb_version_hassector(&mp->m_sb))
745 			log_sector_size = mp->m_sb.sb_logsectsize;
746 		error = xfs_setsize_buftarg(mp->m_logdev_targp,
747 					    mp->m_sb.sb_blocksize,
748 					    log_sector_size);
749 		if (error)
750 			return error;
751 	}
752 	if (mp->m_rtdev_targp) {
753 		error = xfs_setsize_buftarg(mp->m_rtdev_targp,
754 					    mp->m_sb.sb_blocksize,
755 					    mp->m_sb.sb_sectsize);
756 		if (error)
757 			return error;
758 	}
759 
760 	return 0;
761 }
762 
763 /* Catch misguided souls that try to use this interface on XFS */
764 STATIC struct inode *
765 xfs_fs_alloc_inode(
766 	struct super_block	*sb)
767 {
768 	BUG();
769 	return NULL;
770 }
771 
772 /*
773  * Now that the generic code is guaranteed not to be accessing
774  * the linux inode, we can reclaim the inode.
775  */
776 STATIC void
777 xfs_fs_destroy_inode(
778 	struct inode		*inode)
779 {
780 	struct xfs_inode	*ip = XFS_I(inode);
781 
782 	trace_xfs_destroy_inode(ip);
783 
784 	XFS_STATS_INC(vn_reclaim);
785 
786 	/* bad inode, get out here ASAP */
787 	if (is_bad_inode(inode))
788 		goto out_reclaim;
789 
790 	ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
791 
792 	/*
793 	 * We should never get here with one of the reclaim flags already set.
794 	 */
795 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
796 	ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
797 
798 	/*
799 	 * We always use background reclaim here because even if the
800 	 * inode is clean, it still may be under IO and hence we have
801 	 * to take the flush lock. The background reclaim path handles
802 	 * this more efficiently than we can here, so simply let background
803 	 * reclaim tear down all inodes.
804 	 */
805 out_reclaim:
806 	xfs_inode_set_reclaim_tag(ip);
807 }
808 
809 /*
810  * Slab object creation initialisation for the XFS inode.
811  * This covers only the idempotent fields in the XFS inode;
812  * all other fields need to be initialised on allocation
813  * from the slab. This avoids the need to repeatedly initialise
814  * fields in the xfs inode that left in the initialise state
815  * when freeing the inode.
816  */
817 STATIC void
818 xfs_fs_inode_init_once(
819 	void			*inode)
820 {
821 	struct xfs_inode	*ip = inode;
822 
823 	memset(ip, 0, sizeof(struct xfs_inode));
824 
825 	/* vfs inode */
826 	inode_init_once(VFS_I(ip));
827 
828 	/* xfs inode */
829 	atomic_set(&ip->i_pincount, 0);
830 	spin_lock_init(&ip->i_flags_lock);
831 
832 	mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
833 		     "xfsino", ip->i_ino);
834 }
835 
836 /*
837  * Dirty the XFS inode when mark_inode_dirty_sync() is called so that
838  * we catch unlogged VFS level updates to the inode.
839  *
840  * We need the barrier() to maintain correct ordering between unlogged
841  * updates and the transaction commit code that clears the i_update_core
842  * field. This requires all updates to be completed before marking the
843  * inode dirty.
844  */
845 STATIC void
846 xfs_fs_dirty_inode(
847 	struct inode	*inode,
848 	int		flags)
849 {
850 	barrier();
851 	XFS_I(inode)->i_update_core = 1;
852 }
853 
854 STATIC int
855 xfs_fs_write_inode(
856 	struct inode		*inode,
857 	struct writeback_control *wbc)
858 {
859 	struct xfs_inode	*ip = XFS_I(inode);
860 	struct xfs_mount	*mp = ip->i_mount;
861 	int			error = EAGAIN;
862 
863 	trace_xfs_write_inode(ip);
864 
865 	if (XFS_FORCED_SHUTDOWN(mp))
866 		return -XFS_ERROR(EIO);
867 
868 	if (wbc->sync_mode == WB_SYNC_ALL || wbc->for_kupdate) {
869 		/*
870 		 * Make sure the inode has made it it into the log.  Instead
871 		 * of forcing it all the way to stable storage using a
872 		 * synchronous transaction we let the log force inside the
873 		 * ->sync_fs call do that for thus, which reduces the number
874 		 * of synchronous log forces dramatically.
875 		 */
876 		error = xfs_log_dirty_inode(ip, NULL, 0);
877 		if (error)
878 			goto out;
879 		return 0;
880 	} else {
881 		if (!ip->i_update_core)
882 			return 0;
883 
884 		/*
885 		 * We make this non-blocking if the inode is contended, return
886 		 * EAGAIN to indicate to the caller that they did not succeed.
887 		 * This prevents the flush path from blocking on inodes inside
888 		 * another operation right now, they get caught later by
889 		 * xfs_sync.
890 		 */
891 		if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED))
892 			goto out;
893 
894 		if (xfs_ipincount(ip) || !xfs_iflock_nowait(ip))
895 			goto out_unlock;
896 
897 		/*
898 		 * Now we have the flush lock and the inode is not pinned, we
899 		 * can check if the inode is really clean as we know that
900 		 * there are no pending transaction completions, it is not
901 		 * waiting on the delayed write queue and there is no IO in
902 		 * progress.
903 		 */
904 		if (xfs_inode_clean(ip)) {
905 			xfs_ifunlock(ip);
906 			error = 0;
907 			goto out_unlock;
908 		}
909 		error = xfs_iflush(ip, SYNC_TRYLOCK);
910 	}
911 
912  out_unlock:
913 	xfs_iunlock(ip, XFS_ILOCK_SHARED);
914  out:
915 	/*
916 	 * if we failed to write out the inode then mark
917 	 * it dirty again so we'll try again later.
918 	 */
919 	if (error)
920 		xfs_mark_inode_dirty_sync(ip);
921 	return -error;
922 }
923 
924 STATIC void
925 xfs_fs_evict_inode(
926 	struct inode		*inode)
927 {
928 	xfs_inode_t		*ip = XFS_I(inode);
929 
930 	trace_xfs_evict_inode(ip);
931 
932 	truncate_inode_pages(&inode->i_data, 0);
933 	end_writeback(inode);
934 	XFS_STATS_INC(vn_rele);
935 	XFS_STATS_INC(vn_remove);
936 	XFS_STATS_DEC(vn_active);
937 
938 	/*
939 	 * The iolock is used by the file system to coordinate reads,
940 	 * writes, and block truncates.  Up to this point the lock
941 	 * protected concurrent accesses by users of the inode.  But
942 	 * from here forward we're doing some final processing of the
943 	 * inode because we're done with it, and although we reuse the
944 	 * iolock for protection it is really a distinct lock class
945 	 * (in the lockdep sense) from before.  To keep lockdep happy
946 	 * (and basically indicate what we are doing), we explicitly
947 	 * re-init the iolock here.
948 	 */
949 	ASSERT(!rwsem_is_locked(&ip->i_iolock.mr_lock));
950 	mrlock_init(&ip->i_iolock, MRLOCK_BARRIER, "xfsio", ip->i_ino);
951 	lockdep_set_class_and_name(&ip->i_iolock.mr_lock,
952 			&xfs_iolock_reclaimable, "xfs_iolock_reclaimable");
953 
954 	xfs_inactive(ip);
955 }
956 
957 STATIC void
958 xfs_free_fsname(
959 	struct xfs_mount	*mp)
960 {
961 	kfree(mp->m_fsname);
962 	kfree(mp->m_rtname);
963 	kfree(mp->m_logname);
964 }
965 
966 STATIC void
967 xfs_fs_put_super(
968 	struct super_block	*sb)
969 {
970 	struct xfs_mount	*mp = XFS_M(sb);
971 
972 	xfs_syncd_stop(mp);
973 
974 	/*
975 	 * Blow away any referenced inode in the filestreams cache.
976 	 * This can and will cause log traffic as inodes go inactive
977 	 * here.
978 	 */
979 	xfs_filestream_unmount(mp);
980 
981 	xfs_flush_buftarg(mp->m_ddev_targp, 1);
982 
983 	xfs_unmountfs(mp);
984 	xfs_freesb(mp);
985 	xfs_icsb_destroy_counters(mp);
986 	xfs_close_devices(mp);
987 	xfs_free_fsname(mp);
988 	kfree(mp);
989 }
990 
991 STATIC int
992 xfs_fs_sync_fs(
993 	struct super_block	*sb,
994 	int			wait)
995 {
996 	struct xfs_mount	*mp = XFS_M(sb);
997 	int			error;
998 
999 	/*
1000 	 * Doing anything during the async pass would be counterproductive.
1001 	 */
1002 	if (!wait)
1003 		return 0;
1004 
1005 	error = xfs_quiesce_data(mp);
1006 	if (error)
1007 		return -error;
1008 
1009 	if (laptop_mode) {
1010 		/*
1011 		 * The disk must be active because we're syncing.
1012 		 * We schedule xfssyncd now (now that the disk is
1013 		 * active) instead of later (when it might not be).
1014 		 */
1015 		flush_delayed_work_sync(&mp->m_sync_work);
1016 	}
1017 
1018 	return 0;
1019 }
1020 
1021 STATIC int
1022 xfs_fs_statfs(
1023 	struct dentry		*dentry,
1024 	struct kstatfs		*statp)
1025 {
1026 	struct xfs_mount	*mp = XFS_M(dentry->d_sb);
1027 	xfs_sb_t		*sbp = &mp->m_sb;
1028 	struct xfs_inode	*ip = XFS_I(dentry->d_inode);
1029 	__uint64_t		fakeinos, id;
1030 	xfs_extlen_t		lsize;
1031 	__int64_t		ffree;
1032 
1033 	statp->f_type = XFS_SB_MAGIC;
1034 	statp->f_namelen = MAXNAMELEN - 1;
1035 
1036 	id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1037 	statp->f_fsid.val[0] = (u32)id;
1038 	statp->f_fsid.val[1] = (u32)(id >> 32);
1039 
1040 	xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
1041 
1042 	spin_lock(&mp->m_sb_lock);
1043 	statp->f_bsize = sbp->sb_blocksize;
1044 	lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1045 	statp->f_blocks = sbp->sb_dblocks - lsize;
1046 	statp->f_bfree = statp->f_bavail =
1047 				sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1048 	fakeinos = statp->f_bfree << sbp->sb_inopblog;
1049 	statp->f_files =
1050 	    MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
1051 	if (mp->m_maxicount)
1052 		statp->f_files = min_t(typeof(statp->f_files),
1053 					statp->f_files,
1054 					mp->m_maxicount);
1055 
1056 	/* make sure statp->f_ffree does not underflow */
1057 	ffree = statp->f_files - (sbp->sb_icount - sbp->sb_ifree);
1058 	statp->f_ffree = max_t(__int64_t, ffree, 0);
1059 
1060 	spin_unlock(&mp->m_sb_lock);
1061 
1062 	if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) ||
1063 	    ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))) ==
1064 			      (XFS_PQUOTA_ACCT|XFS_OQUOTA_ENFD))
1065 		xfs_qm_statvfs(ip, statp);
1066 	return 0;
1067 }
1068 
1069 STATIC void
1070 xfs_save_resvblks(struct xfs_mount *mp)
1071 {
1072 	__uint64_t resblks = 0;
1073 
1074 	mp->m_resblks_save = mp->m_resblks;
1075 	xfs_reserve_blocks(mp, &resblks, NULL);
1076 }
1077 
1078 STATIC void
1079 xfs_restore_resvblks(struct xfs_mount *mp)
1080 {
1081 	__uint64_t resblks;
1082 
1083 	if (mp->m_resblks_save) {
1084 		resblks = mp->m_resblks_save;
1085 		mp->m_resblks_save = 0;
1086 	} else
1087 		resblks = xfs_default_resblks(mp);
1088 
1089 	xfs_reserve_blocks(mp, &resblks, NULL);
1090 }
1091 
1092 STATIC int
1093 xfs_fs_remount(
1094 	struct super_block	*sb,
1095 	int			*flags,
1096 	char			*options)
1097 {
1098 	struct xfs_mount	*mp = XFS_M(sb);
1099 	substring_t		args[MAX_OPT_ARGS];
1100 	char			*p;
1101 	int			error;
1102 
1103 	while ((p = strsep(&options, ",")) != NULL) {
1104 		int token;
1105 
1106 		if (!*p)
1107 			continue;
1108 
1109 		token = match_token(p, tokens, args);
1110 		switch (token) {
1111 		case Opt_barrier:
1112 			mp->m_flags |= XFS_MOUNT_BARRIER;
1113 			break;
1114 		case Opt_nobarrier:
1115 			mp->m_flags &= ~XFS_MOUNT_BARRIER;
1116 			break;
1117 		default:
1118 			/*
1119 			 * Logically we would return an error here to prevent
1120 			 * users from believing they might have changed
1121 			 * mount options using remount which can't be changed.
1122 			 *
1123 			 * But unfortunately mount(8) adds all options from
1124 			 * mtab and fstab to the mount arguments in some cases
1125 			 * so we can't blindly reject options, but have to
1126 			 * check for each specified option if it actually
1127 			 * differs from the currently set option and only
1128 			 * reject it if that's the case.
1129 			 *
1130 			 * Until that is implemented we return success for
1131 			 * every remount request, and silently ignore all
1132 			 * options that we can't actually change.
1133 			 */
1134 #if 0
1135 			xfs_info(mp,
1136 		"mount option \"%s\" not supported for remount\n", p);
1137 			return -EINVAL;
1138 #else
1139 			break;
1140 #endif
1141 		}
1142 	}
1143 
1144 	/* ro -> rw */
1145 	if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & MS_RDONLY)) {
1146 		mp->m_flags &= ~XFS_MOUNT_RDONLY;
1147 
1148 		/*
1149 		 * If this is the first remount to writeable state we
1150 		 * might have some superblock changes to update.
1151 		 */
1152 		if (mp->m_update_flags) {
1153 			error = xfs_mount_log_sb(mp, mp->m_update_flags);
1154 			if (error) {
1155 				xfs_warn(mp, "failed to write sb changes");
1156 				return error;
1157 			}
1158 			mp->m_update_flags = 0;
1159 		}
1160 
1161 		/*
1162 		 * Fill out the reserve pool if it is empty. Use the stashed
1163 		 * value if it is non-zero, otherwise go with the default.
1164 		 */
1165 		xfs_restore_resvblks(mp);
1166 	}
1167 
1168 	/* rw -> ro */
1169 	if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
1170 		/*
1171 		 * After we have synced the data but before we sync the
1172 		 * metadata, we need to free up the reserve block pool so that
1173 		 * the used block count in the superblock on disk is correct at
1174 		 * the end of the remount. Stash the current reserve pool size
1175 		 * so that if we get remounted rw, we can return it to the same
1176 		 * size.
1177 		 */
1178 
1179 		xfs_quiesce_data(mp);
1180 		xfs_save_resvblks(mp);
1181 		xfs_quiesce_attr(mp);
1182 		mp->m_flags |= XFS_MOUNT_RDONLY;
1183 	}
1184 
1185 	return 0;
1186 }
1187 
1188 /*
1189  * Second stage of a freeze. The data is already frozen so we only
1190  * need to take care of the metadata. Once that's done write a dummy
1191  * record to dirty the log in case of a crash while frozen.
1192  */
1193 STATIC int
1194 xfs_fs_freeze(
1195 	struct super_block	*sb)
1196 {
1197 	struct xfs_mount	*mp = XFS_M(sb);
1198 
1199 	xfs_save_resvblks(mp);
1200 	xfs_quiesce_attr(mp);
1201 	return -xfs_fs_log_dummy(mp);
1202 }
1203 
1204 STATIC int
1205 xfs_fs_unfreeze(
1206 	struct super_block	*sb)
1207 {
1208 	struct xfs_mount	*mp = XFS_M(sb);
1209 
1210 	xfs_restore_resvblks(mp);
1211 	return 0;
1212 }
1213 
1214 STATIC int
1215 xfs_fs_show_options(
1216 	struct seq_file		*m,
1217 	struct dentry		*root)
1218 {
1219 	return -xfs_showargs(XFS_M(root->d_sb), m);
1220 }
1221 
1222 /*
1223  * This function fills in xfs_mount_t fields based on mount args.
1224  * Note: the superblock _has_ now been read in.
1225  */
1226 STATIC int
1227 xfs_finish_flags(
1228 	struct xfs_mount	*mp)
1229 {
1230 	int			ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1231 
1232 	/* Fail a mount where the logbuf is smaller than the log stripe */
1233 	if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1234 		if (mp->m_logbsize <= 0 &&
1235 		    mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1236 			mp->m_logbsize = mp->m_sb.sb_logsunit;
1237 		} else if (mp->m_logbsize > 0 &&
1238 			   mp->m_logbsize < mp->m_sb.sb_logsunit) {
1239 			xfs_warn(mp,
1240 		"logbuf size must be greater than or equal to log stripe size");
1241 			return XFS_ERROR(EINVAL);
1242 		}
1243 	} else {
1244 		/* Fail a mount if the logbuf is larger than 32K */
1245 		if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1246 			xfs_warn(mp,
1247 		"logbuf size for version 1 logs must be 16K or 32K");
1248 			return XFS_ERROR(EINVAL);
1249 		}
1250 	}
1251 
1252 	/*
1253 	 * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1254 	 * told by noattr2 to turn it off
1255 	 */
1256 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1257 	    !(mp->m_flags & XFS_MOUNT_NOATTR2))
1258 		mp->m_flags |= XFS_MOUNT_ATTR2;
1259 
1260 	/*
1261 	 * prohibit r/w mounts of read-only filesystems
1262 	 */
1263 	if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1264 		xfs_warn(mp,
1265 			"cannot mount a read-only filesystem as read-write");
1266 		return XFS_ERROR(EROFS);
1267 	}
1268 
1269 	return 0;
1270 }
1271 
1272 STATIC int
1273 xfs_fs_fill_super(
1274 	struct super_block	*sb,
1275 	void			*data,
1276 	int			silent)
1277 {
1278 	struct inode		*root;
1279 	struct xfs_mount	*mp = NULL;
1280 	int			flags = 0, error = ENOMEM;
1281 
1282 	mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1283 	if (!mp)
1284 		goto out;
1285 
1286 	spin_lock_init(&mp->m_sb_lock);
1287 	mutex_init(&mp->m_growlock);
1288 	atomic_set(&mp->m_active_trans, 0);
1289 
1290 	mp->m_super = sb;
1291 	sb->s_fs_info = mp;
1292 
1293 	error = xfs_parseargs(mp, (char *)data);
1294 	if (error)
1295 		goto out_free_fsname;
1296 
1297 	sb_min_blocksize(sb, BBSIZE);
1298 	sb->s_xattr = xfs_xattr_handlers;
1299 	sb->s_export_op = &xfs_export_operations;
1300 #ifdef CONFIG_XFS_QUOTA
1301 	sb->s_qcop = &xfs_quotactl_operations;
1302 #endif
1303 	sb->s_op = &xfs_super_operations;
1304 
1305 	if (silent)
1306 		flags |= XFS_MFSI_QUIET;
1307 
1308 	error = xfs_open_devices(mp);
1309 	if (error)
1310 		goto out_free_fsname;
1311 
1312 	error = xfs_icsb_init_counters(mp);
1313 	if (error)
1314 		goto out_close_devices;
1315 
1316 	error = xfs_readsb(mp, flags);
1317 	if (error)
1318 		goto out_destroy_counters;
1319 
1320 	error = xfs_finish_flags(mp);
1321 	if (error)
1322 		goto out_free_sb;
1323 
1324 	error = xfs_setup_devices(mp);
1325 	if (error)
1326 		goto out_free_sb;
1327 
1328 	error = xfs_filestream_mount(mp);
1329 	if (error)
1330 		goto out_free_sb;
1331 
1332 	/*
1333 	 * we must configure the block size in the superblock before we run the
1334 	 * full mount process as the mount process can lookup and cache inodes.
1335 	 * For the same reason we must also initialise the syncd and register
1336 	 * the inode cache shrinker so that inodes can be reclaimed during
1337 	 * operations like a quotacheck that iterate all inodes in the
1338 	 * filesystem.
1339 	 */
1340 	sb->s_magic = XFS_SB_MAGIC;
1341 	sb->s_blocksize = mp->m_sb.sb_blocksize;
1342 	sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1343 	sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1344 	sb->s_time_gran = 1;
1345 	set_posix_acl_flag(sb);
1346 
1347 	error = xfs_mountfs(mp);
1348 	if (error)
1349 		goto out_filestream_unmount;
1350 
1351 	error = xfs_syncd_init(mp);
1352 	if (error)
1353 		goto out_unmount;
1354 
1355 	root = igrab(VFS_I(mp->m_rootip));
1356 	if (!root) {
1357 		error = ENOENT;
1358 		goto out_syncd_stop;
1359 	}
1360 	if (is_bad_inode(root)) {
1361 		error = EINVAL;
1362 		goto out_syncd_stop;
1363 	}
1364 	sb->s_root = d_alloc_root(root);
1365 	if (!sb->s_root) {
1366 		error = ENOMEM;
1367 		goto out_iput;
1368 	}
1369 
1370 	return 0;
1371 
1372  out_filestream_unmount:
1373 	xfs_filestream_unmount(mp);
1374  out_free_sb:
1375 	xfs_freesb(mp);
1376  out_destroy_counters:
1377 	xfs_icsb_destroy_counters(mp);
1378  out_close_devices:
1379 	xfs_close_devices(mp);
1380  out_free_fsname:
1381 	xfs_free_fsname(mp);
1382 	kfree(mp);
1383  out:
1384 	return -error;
1385 
1386  out_iput:
1387 	iput(root);
1388  out_syncd_stop:
1389 	xfs_syncd_stop(mp);
1390  out_unmount:
1391 	/*
1392 	 * Blow away any referenced inode in the filestreams cache.
1393 	 * This can and will cause log traffic as inodes go inactive
1394 	 * here.
1395 	 */
1396 	xfs_filestream_unmount(mp);
1397 
1398 	xfs_flush_buftarg(mp->m_ddev_targp, 1);
1399 
1400 	xfs_unmountfs(mp);
1401 	goto out_free_sb;
1402 }
1403 
1404 STATIC struct dentry *
1405 xfs_fs_mount(
1406 	struct file_system_type	*fs_type,
1407 	int			flags,
1408 	const char		*dev_name,
1409 	void			*data)
1410 {
1411 	return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1412 }
1413 
1414 static int
1415 xfs_fs_nr_cached_objects(
1416 	struct super_block	*sb)
1417 {
1418 	return xfs_reclaim_inodes_count(XFS_M(sb));
1419 }
1420 
1421 static void
1422 xfs_fs_free_cached_objects(
1423 	struct super_block	*sb,
1424 	int			nr_to_scan)
1425 {
1426 	xfs_reclaim_inodes_nr(XFS_M(sb), nr_to_scan);
1427 }
1428 
1429 static const struct super_operations xfs_super_operations = {
1430 	.alloc_inode		= xfs_fs_alloc_inode,
1431 	.destroy_inode		= xfs_fs_destroy_inode,
1432 	.dirty_inode		= xfs_fs_dirty_inode,
1433 	.write_inode		= xfs_fs_write_inode,
1434 	.evict_inode		= xfs_fs_evict_inode,
1435 	.put_super		= xfs_fs_put_super,
1436 	.sync_fs		= xfs_fs_sync_fs,
1437 	.freeze_fs		= xfs_fs_freeze,
1438 	.unfreeze_fs		= xfs_fs_unfreeze,
1439 	.statfs			= xfs_fs_statfs,
1440 	.remount_fs		= xfs_fs_remount,
1441 	.show_options		= xfs_fs_show_options,
1442 	.nr_cached_objects	= xfs_fs_nr_cached_objects,
1443 	.free_cached_objects	= xfs_fs_free_cached_objects,
1444 };
1445 
1446 static struct file_system_type xfs_fs_type = {
1447 	.owner			= THIS_MODULE,
1448 	.name			= "xfs",
1449 	.mount			= xfs_fs_mount,
1450 	.kill_sb		= kill_block_super,
1451 	.fs_flags		= FS_REQUIRES_DEV,
1452 };
1453 
1454 STATIC int __init
1455 xfs_init_zones(void)
1456 {
1457 
1458 	xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend");
1459 	if (!xfs_ioend_zone)
1460 		goto out;
1461 
1462 	xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE,
1463 						  xfs_ioend_zone);
1464 	if (!xfs_ioend_pool)
1465 		goto out_destroy_ioend_zone;
1466 
1467 	xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1468 						"xfs_log_ticket");
1469 	if (!xfs_log_ticket_zone)
1470 		goto out_destroy_ioend_pool;
1471 
1472 	xfs_bmap_free_item_zone = kmem_zone_init(sizeof(xfs_bmap_free_item_t),
1473 						"xfs_bmap_free_item");
1474 	if (!xfs_bmap_free_item_zone)
1475 		goto out_destroy_log_ticket_zone;
1476 
1477 	xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1478 						"xfs_btree_cur");
1479 	if (!xfs_btree_cur_zone)
1480 		goto out_destroy_bmap_free_item_zone;
1481 
1482 	xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1483 						"xfs_da_state");
1484 	if (!xfs_da_state_zone)
1485 		goto out_destroy_btree_cur_zone;
1486 
1487 	xfs_dabuf_zone = kmem_zone_init(sizeof(xfs_dabuf_t), "xfs_dabuf");
1488 	if (!xfs_dabuf_zone)
1489 		goto out_destroy_da_state_zone;
1490 
1491 	xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1492 	if (!xfs_ifork_zone)
1493 		goto out_destroy_dabuf_zone;
1494 
1495 	xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1496 	if (!xfs_trans_zone)
1497 		goto out_destroy_ifork_zone;
1498 
1499 	xfs_log_item_desc_zone =
1500 		kmem_zone_init(sizeof(struct xfs_log_item_desc),
1501 			       "xfs_log_item_desc");
1502 	if (!xfs_log_item_desc_zone)
1503 		goto out_destroy_trans_zone;
1504 
1505 	/*
1506 	 * The size of the zone allocated buf log item is the maximum
1507 	 * size possible under XFS.  This wastes a little bit of memory,
1508 	 * but it is much faster.
1509 	 */
1510 	xfs_buf_item_zone = kmem_zone_init((sizeof(xfs_buf_log_item_t) +
1511 				(((XFS_MAX_BLOCKSIZE / XFS_BLF_CHUNK) /
1512 				  NBWORD) * sizeof(int))), "xfs_buf_item");
1513 	if (!xfs_buf_item_zone)
1514 		goto out_destroy_log_item_desc_zone;
1515 
1516 	xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1517 			((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1518 				 sizeof(xfs_extent_t))), "xfs_efd_item");
1519 	if (!xfs_efd_zone)
1520 		goto out_destroy_buf_item_zone;
1521 
1522 	xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1523 			((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1524 				sizeof(xfs_extent_t))), "xfs_efi_item");
1525 	if (!xfs_efi_zone)
1526 		goto out_destroy_efd_zone;
1527 
1528 	xfs_inode_zone =
1529 		kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1530 			KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD,
1531 			xfs_fs_inode_init_once);
1532 	if (!xfs_inode_zone)
1533 		goto out_destroy_efi_zone;
1534 
1535 	xfs_ili_zone =
1536 		kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1537 					KM_ZONE_SPREAD, NULL);
1538 	if (!xfs_ili_zone)
1539 		goto out_destroy_inode_zone;
1540 
1541 	return 0;
1542 
1543  out_destroy_inode_zone:
1544 	kmem_zone_destroy(xfs_inode_zone);
1545  out_destroy_efi_zone:
1546 	kmem_zone_destroy(xfs_efi_zone);
1547  out_destroy_efd_zone:
1548 	kmem_zone_destroy(xfs_efd_zone);
1549  out_destroy_buf_item_zone:
1550 	kmem_zone_destroy(xfs_buf_item_zone);
1551  out_destroy_log_item_desc_zone:
1552 	kmem_zone_destroy(xfs_log_item_desc_zone);
1553  out_destroy_trans_zone:
1554 	kmem_zone_destroy(xfs_trans_zone);
1555  out_destroy_ifork_zone:
1556 	kmem_zone_destroy(xfs_ifork_zone);
1557  out_destroy_dabuf_zone:
1558 	kmem_zone_destroy(xfs_dabuf_zone);
1559  out_destroy_da_state_zone:
1560 	kmem_zone_destroy(xfs_da_state_zone);
1561  out_destroy_btree_cur_zone:
1562 	kmem_zone_destroy(xfs_btree_cur_zone);
1563  out_destroy_bmap_free_item_zone:
1564 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1565  out_destroy_log_ticket_zone:
1566 	kmem_zone_destroy(xfs_log_ticket_zone);
1567  out_destroy_ioend_pool:
1568 	mempool_destroy(xfs_ioend_pool);
1569  out_destroy_ioend_zone:
1570 	kmem_zone_destroy(xfs_ioend_zone);
1571  out:
1572 	return -ENOMEM;
1573 }
1574 
1575 STATIC void
1576 xfs_destroy_zones(void)
1577 {
1578 	kmem_zone_destroy(xfs_ili_zone);
1579 	kmem_zone_destroy(xfs_inode_zone);
1580 	kmem_zone_destroy(xfs_efi_zone);
1581 	kmem_zone_destroy(xfs_efd_zone);
1582 	kmem_zone_destroy(xfs_buf_item_zone);
1583 	kmem_zone_destroy(xfs_log_item_desc_zone);
1584 	kmem_zone_destroy(xfs_trans_zone);
1585 	kmem_zone_destroy(xfs_ifork_zone);
1586 	kmem_zone_destroy(xfs_dabuf_zone);
1587 	kmem_zone_destroy(xfs_da_state_zone);
1588 	kmem_zone_destroy(xfs_btree_cur_zone);
1589 	kmem_zone_destroy(xfs_bmap_free_item_zone);
1590 	kmem_zone_destroy(xfs_log_ticket_zone);
1591 	mempool_destroy(xfs_ioend_pool);
1592 	kmem_zone_destroy(xfs_ioend_zone);
1593 
1594 }
1595 
1596 STATIC int __init
1597 xfs_init_workqueues(void)
1598 {
1599 	/*
1600 	 * We never want to the same work item to run twice, reclaiming inodes
1601 	 * or idling the log is not going to get any faster by multiple CPUs
1602 	 * competing for ressources.  Use the default large max_active value
1603 	 * so that even lots of filesystems can perform these task in parallel.
1604 	 */
1605 	xfs_syncd_wq = alloc_workqueue("xfssyncd", WQ_NON_REENTRANT, 0);
1606 	if (!xfs_syncd_wq)
1607 		return -ENOMEM;
1608 	return 0;
1609 }
1610 
1611 STATIC void
1612 xfs_destroy_workqueues(void)
1613 {
1614 	destroy_workqueue(xfs_syncd_wq);
1615 }
1616 
1617 STATIC int __init
1618 init_xfs_fs(void)
1619 {
1620 	int			error;
1621 
1622 	printk(KERN_INFO XFS_VERSION_STRING " with "
1623 			 XFS_BUILD_OPTIONS " enabled\n");
1624 
1625 	xfs_dir_startup();
1626 
1627 	error = xfs_init_zones();
1628 	if (error)
1629 		goto out;
1630 
1631 	error = xfs_init_workqueues();
1632 	if (error)
1633 		goto out_destroy_zones;
1634 
1635 	error = xfs_mru_cache_init();
1636 	if (error)
1637 		goto out_destroy_wq;
1638 
1639 	error = xfs_filestream_init();
1640 	if (error)
1641 		goto out_mru_cache_uninit;
1642 
1643 	error = xfs_buf_init();
1644 	if (error)
1645 		goto out_filestream_uninit;
1646 
1647 	error = xfs_init_procfs();
1648 	if (error)
1649 		goto out_buf_terminate;
1650 
1651 	error = xfs_sysctl_register();
1652 	if (error)
1653 		goto out_cleanup_procfs;
1654 
1655 	vfs_initquota();
1656 
1657 	error = register_filesystem(&xfs_fs_type);
1658 	if (error)
1659 		goto out_sysctl_unregister;
1660 	return 0;
1661 
1662  out_sysctl_unregister:
1663 	xfs_sysctl_unregister();
1664  out_cleanup_procfs:
1665 	xfs_cleanup_procfs();
1666  out_buf_terminate:
1667 	xfs_buf_terminate();
1668  out_filestream_uninit:
1669 	xfs_filestream_uninit();
1670  out_mru_cache_uninit:
1671 	xfs_mru_cache_uninit();
1672  out_destroy_wq:
1673 	xfs_destroy_workqueues();
1674  out_destroy_zones:
1675 	xfs_destroy_zones();
1676  out:
1677 	return error;
1678 }
1679 
1680 STATIC void __exit
1681 exit_xfs_fs(void)
1682 {
1683 	vfs_exitquota();
1684 	unregister_filesystem(&xfs_fs_type);
1685 	xfs_sysctl_unregister();
1686 	xfs_cleanup_procfs();
1687 	xfs_buf_terminate();
1688 	xfs_filestream_uninit();
1689 	xfs_mru_cache_uninit();
1690 	xfs_destroy_workqueues();
1691 	xfs_destroy_zones();
1692 }
1693 
1694 module_init(init_xfs_fs);
1695 module_exit(exit_xfs_fs);
1696 
1697 MODULE_AUTHOR("Silicon Graphics, Inc.");
1698 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
1699 MODULE_LICENSE("GPL");
1700