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