Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[muen/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 #include "xfs_reflink.h"
53
54 #include <linux/namei.h>
55 #include <linux/dax.h>
56 #include <linux/init.h>
57 #include <linux/slab.h>
58 #include <linux/mount.h>
59 #include <linux/mempool.h>
60 #include <linux/writeback.h>
61 #include <linux/kthread.h>
62 #include <linux/freezer.h>
63 #include <linux/parser.h>
64
65 static const struct super_operations xfs_super_operations;
66 struct bio_set *xfs_ioend_bioset;
67
68 static struct kset *xfs_kset;           /* top-level xfs sysfs dir */
69 #ifdef DEBUG
70 static struct xfs_kobj xfs_dbg_kobj;    /* global debug sysfs attrs */
71 #endif
72
73 /*
74  * Table driven mount option parser.
75  */
76 enum {
77         Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev, Opt_biosize,
78         Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
79         Opt_mtpt, Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
80         Opt_allocsize, Opt_norecovery, Opt_barrier, Opt_nobarrier,
81         Opt_inode64, Opt_inode32, Opt_ikeep, Opt_noikeep,
82         Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2, Opt_filestreams,
83         Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota, Opt_prjquota,
84         Opt_uquota, Opt_gquota, Opt_pquota,
85         Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
86         Opt_discard, Opt_nodiscard, Opt_dax, Opt_err,
87 };
88
89 static const match_table_t tokens = {
90         {Opt_logbufs,   "logbufs=%u"},  /* number of XFS log buffers */
91         {Opt_logbsize,  "logbsize=%s"}, /* size of XFS log buffers */
92         {Opt_logdev,    "logdev=%s"},   /* log device */
93         {Opt_rtdev,     "rtdev=%s"},    /* realtime I/O device */
94         {Opt_biosize,   "biosize=%u"},  /* log2 of preferred buffered io size */
95         {Opt_wsync,     "wsync"},       /* safe-mode nfs compatible mount */
96         {Opt_noalign,   "noalign"},     /* turn off stripe alignment */
97         {Opt_swalloc,   "swalloc"},     /* turn on stripe width allocation */
98         {Opt_sunit,     "sunit=%u"},    /* data volume stripe unit */
99         {Opt_swidth,    "swidth=%u"},   /* data volume stripe width */
100         {Opt_nouuid,    "nouuid"},      /* ignore filesystem UUID */
101         {Opt_mtpt,      "mtpt"},        /* filesystem mount point */
102         {Opt_grpid,     "grpid"},       /* group-ID from parent directory */
103         {Opt_nogrpid,   "nogrpid"},     /* group-ID from current process */
104         {Opt_bsdgroups, "bsdgroups"},   /* group-ID from parent directory */
105         {Opt_sysvgroups,"sysvgroups"},  /* group-ID from current process */
106         {Opt_allocsize, "allocsize=%s"},/* preferred allocation size */
107         {Opt_norecovery,"norecovery"},  /* don't run XFS recovery */
108         {Opt_inode64,   "inode64"},     /* inodes can be allocated anywhere */
109         {Opt_inode32,   "inode32"},     /* inode allocation limited to
110                                          * XFS_MAXINUMBER_32 */
111         {Opt_ikeep,     "ikeep"},       /* do not free empty inode clusters */
112         {Opt_noikeep,   "noikeep"},     /* free empty inode clusters */
113         {Opt_largeio,   "largeio"},     /* report large I/O sizes in stat() */
114         {Opt_nolargeio, "nolargeio"},   /* do not report large I/O sizes
115                                          * in stat(). */
116         {Opt_attr2,     "attr2"},       /* do use attr2 attribute format */
117         {Opt_noattr2,   "noattr2"},     /* do not use attr2 attribute format */
118         {Opt_filestreams,"filestreams"},/* use filestreams allocator */
119         {Opt_quota,     "quota"},       /* disk quotas (user) */
120         {Opt_noquota,   "noquota"},     /* no quotas */
121         {Opt_usrquota,  "usrquota"},    /* user quota enabled */
122         {Opt_grpquota,  "grpquota"},    /* group quota enabled */
123         {Opt_prjquota,  "prjquota"},    /* project quota enabled */
124         {Opt_uquota,    "uquota"},      /* user quota (IRIX variant) */
125         {Opt_gquota,    "gquota"},      /* group quota (IRIX variant) */
126         {Opt_pquota,    "pquota"},      /* project quota (IRIX variant) */
127         {Opt_uqnoenforce,"uqnoenforce"},/* user quota limit enforcement */
128         {Opt_gqnoenforce,"gqnoenforce"},/* group quota limit enforcement */
129         {Opt_pqnoenforce,"pqnoenforce"},/* project quota limit enforcement */
130         {Opt_qnoenforce, "qnoenforce"}, /* same as uqnoenforce */
131         {Opt_discard,   "discard"},     /* Discard unused blocks */
132         {Opt_nodiscard, "nodiscard"},   /* Do not discard unused blocks */
133
134         {Opt_dax,       "dax"},         /* Enable direct access to bdev pages */
135
136         /* Deprecated mount options scheduled for removal */
137         {Opt_barrier,   "barrier"},     /* use writer barriers for log write and
138                                          * unwritten extent conversion */
139         {Opt_nobarrier, "nobarrier"},   /* .. disable */
140
141         {Opt_err,       NULL},
142 };
143
144
145 STATIC int
146 suffix_kstrtoint(const substring_t *s, unsigned int base, int *res)
147 {
148         int     last, shift_left_factor = 0, _res;
149         char    *value;
150         int     ret = 0;
151
152         value = match_strdup(s);
153         if (!value)
154                 return -ENOMEM;
155
156         last = strlen(value) - 1;
157         if (value[last] == 'K' || value[last] == 'k') {
158                 shift_left_factor = 10;
159                 value[last] = '\0';
160         }
161         if (value[last] == 'M' || value[last] == 'm') {
162                 shift_left_factor = 20;
163                 value[last] = '\0';
164         }
165         if (value[last] == 'G' || value[last] == 'g') {
166                 shift_left_factor = 30;
167                 value[last] = '\0';
168         }
169
170         if (kstrtoint(value, base, &_res))
171                 ret = -EINVAL;
172         kfree(value);
173         *res = _res << shift_left_factor;
174         return ret;
175 }
176
177 /*
178  * This function fills in xfs_mount_t fields based on mount args.
179  * Note: the superblock has _not_ yet been read in.
180  *
181  * Note that this function leaks the various device name allocations on
182  * failure.  The caller takes care of them.
183  *
184  * *sb is const because this is also used to test options on the remount
185  * path, and we don't want this to have any side effects at remount time.
186  * Today this function does not change *sb, but just to future-proof...
187  */
188 STATIC int
189 xfs_parseargs(
190         struct xfs_mount        *mp,
191         char                    *options)
192 {
193         const struct super_block *sb = mp->m_super;
194         char                    *p;
195         substring_t             args[MAX_OPT_ARGS];
196         int                     dsunit = 0;
197         int                     dswidth = 0;
198         int                     iosize = 0;
199         uint8_t                 iosizelog = 0;
200
201         /*
202          * set up the mount name first so all the errors will refer to the
203          * correct device.
204          */
205         mp->m_fsname = kstrndup(sb->s_id, MAXNAMELEN, GFP_KERNEL);
206         if (!mp->m_fsname)
207                 return -ENOMEM;
208         mp->m_fsname_len = strlen(mp->m_fsname) + 1;
209
210         /*
211          * Copy binary VFS mount flags we are interested in.
212          */
213         if (sb_rdonly(sb))
214                 mp->m_flags |= XFS_MOUNT_RDONLY;
215         if (sb->s_flags & SB_DIRSYNC)
216                 mp->m_flags |= XFS_MOUNT_DIRSYNC;
217         if (sb->s_flags & SB_SYNCHRONOUS)
218                 mp->m_flags |= XFS_MOUNT_WSYNC;
219
220         /*
221          * Set some default flags that could be cleared by the mount option
222          * parsing.
223          */
224         mp->m_flags |= XFS_MOUNT_BARRIER;
225         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
226
227         /*
228          * These can be overridden by the mount option parsing.
229          */
230         mp->m_logbufs = -1;
231         mp->m_logbsize = -1;
232
233         if (!options)
234                 goto done;
235
236         while ((p = strsep(&options, ",")) != NULL) {
237                 int             token;
238
239                 if (!*p)
240                         continue;
241
242                 token = match_token(p, tokens, args);
243                 switch (token) {
244                 case Opt_logbufs:
245                         if (match_int(args, &mp->m_logbufs))
246                                 return -EINVAL;
247                         break;
248                 case Opt_logbsize:
249                         if (suffix_kstrtoint(args, 10, &mp->m_logbsize))
250                                 return -EINVAL;
251                         break;
252                 case Opt_logdev:
253                         mp->m_logname = match_strdup(args);
254                         if (!mp->m_logname)
255                                 return -ENOMEM;
256                         break;
257                 case Opt_mtpt:
258                         xfs_warn(mp, "%s option not allowed on this system", p);
259                         return -EINVAL;
260                 case Opt_rtdev:
261                         mp->m_rtname = match_strdup(args);
262                         if (!mp->m_rtname)
263                                 return -ENOMEM;
264                         break;
265                 case Opt_allocsize:
266                 case Opt_biosize:
267                         if (suffix_kstrtoint(args, 10, &iosize))
268                                 return -EINVAL;
269                         iosizelog = ffs(iosize) - 1;
270                         break;
271                 case Opt_grpid:
272                 case Opt_bsdgroups:
273                         mp->m_flags |= XFS_MOUNT_GRPID;
274                         break;
275                 case Opt_nogrpid:
276                 case Opt_sysvgroups:
277                         mp->m_flags &= ~XFS_MOUNT_GRPID;
278                         break;
279                 case Opt_wsync:
280                         mp->m_flags |= XFS_MOUNT_WSYNC;
281                         break;
282                 case Opt_norecovery:
283                         mp->m_flags |= XFS_MOUNT_NORECOVERY;
284                         break;
285                 case Opt_noalign:
286                         mp->m_flags |= XFS_MOUNT_NOALIGN;
287                         break;
288                 case Opt_swalloc:
289                         mp->m_flags |= XFS_MOUNT_SWALLOC;
290                         break;
291                 case Opt_sunit:
292                         if (match_int(args, &dsunit))
293                                 return -EINVAL;
294                         break;
295                 case Opt_swidth:
296                         if (match_int(args, &dswidth))
297                                 return -EINVAL;
298                         break;
299                 case Opt_inode32:
300                         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
301                         break;
302                 case Opt_inode64:
303                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
304                         break;
305                 case Opt_nouuid:
306                         mp->m_flags |= XFS_MOUNT_NOUUID;
307                         break;
308                 case Opt_ikeep:
309                         mp->m_flags |= XFS_MOUNT_IKEEP;
310                         break;
311                 case Opt_noikeep:
312                         mp->m_flags &= ~XFS_MOUNT_IKEEP;
313                         break;
314                 case Opt_largeio:
315                         mp->m_flags &= ~XFS_MOUNT_COMPAT_IOSIZE;
316                         break;
317                 case Opt_nolargeio:
318                         mp->m_flags |= XFS_MOUNT_COMPAT_IOSIZE;
319                         break;
320                 case Opt_attr2:
321                         mp->m_flags |= XFS_MOUNT_ATTR2;
322                         break;
323                 case Opt_noattr2:
324                         mp->m_flags &= ~XFS_MOUNT_ATTR2;
325                         mp->m_flags |= XFS_MOUNT_NOATTR2;
326                         break;
327                 case Opt_filestreams:
328                         mp->m_flags |= XFS_MOUNT_FILESTREAMS;
329                         break;
330                 case Opt_noquota:
331                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
332                         mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
333                         mp->m_qflags &= ~XFS_ALL_QUOTA_ACTIVE;
334                         break;
335                 case Opt_quota:
336                 case Opt_uquota:
337                 case Opt_usrquota:
338                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE |
339                                          XFS_UQUOTA_ENFD);
340                         break;
341                 case Opt_qnoenforce:
342                 case Opt_uqnoenforce:
343                         mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ACTIVE);
344                         mp->m_qflags &= ~XFS_UQUOTA_ENFD;
345                         break;
346                 case Opt_pquota:
347                 case Opt_prjquota:
348                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE |
349                                          XFS_PQUOTA_ENFD);
350                         break;
351                 case Opt_pqnoenforce:
352                         mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE);
353                         mp->m_qflags &= ~XFS_PQUOTA_ENFD;
354                         break;
355                 case Opt_gquota:
356                 case Opt_grpquota:
357                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE |
358                                          XFS_GQUOTA_ENFD);
359                         break;
360                 case Opt_gqnoenforce:
361                         mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE);
362                         mp->m_qflags &= ~XFS_GQUOTA_ENFD;
363                         break;
364                 case Opt_discard:
365                         mp->m_flags |= XFS_MOUNT_DISCARD;
366                         break;
367                 case Opt_nodiscard:
368                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
369                         break;
370 #ifdef CONFIG_FS_DAX
371                 case Opt_dax:
372                         mp->m_flags |= XFS_MOUNT_DAX;
373                         break;
374 #endif
375                 case Opt_barrier:
376                         xfs_warn(mp, "%s option is deprecated, ignoring.", p);
377                         mp->m_flags |= XFS_MOUNT_BARRIER;
378                         break;
379                 case Opt_nobarrier:
380                         xfs_warn(mp, "%s option is deprecated, ignoring.", p);
381                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
382                         break;
383                 default:
384                         xfs_warn(mp, "unknown mount option [%s].", p);
385                         return -EINVAL;
386                 }
387         }
388
389         /*
390          * no recovery flag requires a read-only mount
391          */
392         if ((mp->m_flags & XFS_MOUNT_NORECOVERY) &&
393             !(mp->m_flags & XFS_MOUNT_RDONLY)) {
394                 xfs_warn(mp, "no-recovery mounts must be read-only.");
395                 return -EINVAL;
396         }
397
398         if ((mp->m_flags & XFS_MOUNT_NOALIGN) && (dsunit || dswidth)) {
399                 xfs_warn(mp,
400         "sunit and swidth options incompatible with the noalign option");
401                 return -EINVAL;
402         }
403
404 #ifndef CONFIG_XFS_QUOTA
405         if (XFS_IS_QUOTA_RUNNING(mp)) {
406                 xfs_warn(mp, "quota support not available in this kernel.");
407                 return -EINVAL;
408         }
409 #endif
410
411         if ((dsunit && !dswidth) || (!dsunit && dswidth)) {
412                 xfs_warn(mp, "sunit and swidth must be specified together");
413                 return -EINVAL;
414         }
415
416         if (dsunit && (dswidth % dsunit != 0)) {
417                 xfs_warn(mp,
418         "stripe width (%d) must be a multiple of the stripe unit (%d)",
419                         dswidth, dsunit);
420                 return -EINVAL;
421         }
422
423 done:
424         if (dsunit && !(mp->m_flags & XFS_MOUNT_NOALIGN)) {
425                 /*
426                  * At this point the superblock has not been read
427                  * in, therefore we do not know the block size.
428                  * Before the mount call ends we will convert
429                  * these to FSBs.
430                  */
431                 mp->m_dalign = dsunit;
432                 mp->m_swidth = dswidth;
433         }
434
435         if (mp->m_logbufs != -1 &&
436             mp->m_logbufs != 0 &&
437             (mp->m_logbufs < XLOG_MIN_ICLOGS ||
438              mp->m_logbufs > XLOG_MAX_ICLOGS)) {
439                 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
440                         mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
441                 return -EINVAL;
442         }
443         if (mp->m_logbsize != -1 &&
444             mp->m_logbsize !=  0 &&
445             (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
446              mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
447              !is_power_of_2(mp->m_logbsize))) {
448                 xfs_warn(mp,
449                         "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
450                         mp->m_logbsize);
451                 return -EINVAL;
452         }
453
454         if (iosizelog) {
455                 if (iosizelog > XFS_MAX_IO_LOG ||
456                     iosizelog < XFS_MIN_IO_LOG) {
457                         xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
458                                 iosizelog, XFS_MIN_IO_LOG,
459                                 XFS_MAX_IO_LOG);
460                         return -EINVAL;
461                 }
462
463                 mp->m_flags |= XFS_MOUNT_DFLT_IOSIZE;
464                 mp->m_readio_log = iosizelog;
465                 mp->m_writeio_log = iosizelog;
466         }
467
468         return 0;
469 }
470
471 struct proc_xfs_info {
472         uint64_t        flag;
473         char            *str;
474 };
475
476 STATIC int
477 xfs_showargs(
478         struct xfs_mount        *mp,
479         struct seq_file         *m)
480 {
481         static struct proc_xfs_info xfs_info_set[] = {
482                 /* the few simple ones we can get from the mount struct */
483                 { XFS_MOUNT_IKEEP,              ",ikeep" },
484                 { XFS_MOUNT_WSYNC,              ",wsync" },
485                 { XFS_MOUNT_NOALIGN,            ",noalign" },
486                 { XFS_MOUNT_SWALLOC,            ",swalloc" },
487                 { XFS_MOUNT_NOUUID,             ",nouuid" },
488                 { XFS_MOUNT_NORECOVERY,         ",norecovery" },
489                 { XFS_MOUNT_ATTR2,              ",attr2" },
490                 { XFS_MOUNT_FILESTREAMS,        ",filestreams" },
491                 { XFS_MOUNT_GRPID,              ",grpid" },
492                 { XFS_MOUNT_DISCARD,            ",discard" },
493                 { XFS_MOUNT_SMALL_INUMS,        ",inode32" },
494                 { XFS_MOUNT_DAX,                ",dax" },
495                 { 0, NULL }
496         };
497         static struct proc_xfs_info xfs_info_unset[] = {
498                 /* the few simple ones we can get from the mount struct */
499                 { XFS_MOUNT_COMPAT_IOSIZE,      ",largeio" },
500                 { XFS_MOUNT_BARRIER,            ",nobarrier" },
501                 { XFS_MOUNT_SMALL_INUMS,        ",inode64" },
502                 { 0, NULL }
503         };
504         struct proc_xfs_info    *xfs_infop;
505
506         for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
507                 if (mp->m_flags & xfs_infop->flag)
508                         seq_puts(m, xfs_infop->str);
509         }
510         for (xfs_infop = xfs_info_unset; xfs_infop->flag; xfs_infop++) {
511                 if (!(mp->m_flags & xfs_infop->flag))
512                         seq_puts(m, xfs_infop->str);
513         }
514
515         if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
516                 seq_printf(m, ",allocsize=%dk",
517                                 (int)(1 << mp->m_writeio_log) >> 10);
518
519         if (mp->m_logbufs > 0)
520                 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
521         if (mp->m_logbsize > 0)
522                 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
523
524         if (mp->m_logname)
525                 seq_show_option(m, "logdev", mp->m_logname);
526         if (mp->m_rtname)
527                 seq_show_option(m, "rtdev", mp->m_rtname);
528
529         if (mp->m_dalign > 0)
530                 seq_printf(m, ",sunit=%d",
531                                 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
532         if (mp->m_swidth > 0)
533                 seq_printf(m, ",swidth=%d",
534                                 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
535
536         if (mp->m_qflags & (XFS_UQUOTA_ACCT|XFS_UQUOTA_ENFD))
537                 seq_puts(m, ",usrquota");
538         else if (mp->m_qflags & XFS_UQUOTA_ACCT)
539                 seq_puts(m, ",uqnoenforce");
540
541         if (mp->m_qflags & XFS_PQUOTA_ACCT) {
542                 if (mp->m_qflags & XFS_PQUOTA_ENFD)
543                         seq_puts(m, ",prjquota");
544                 else
545                         seq_puts(m, ",pqnoenforce");
546         }
547         if (mp->m_qflags & XFS_GQUOTA_ACCT) {
548                 if (mp->m_qflags & XFS_GQUOTA_ENFD)
549                         seq_puts(m, ",grpquota");
550                 else
551                         seq_puts(m, ",gqnoenforce");
552         }
553
554         if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
555                 seq_puts(m, ",noquota");
556
557         return 0;
558 }
559 static uint64_t
560 xfs_max_file_offset(
561         unsigned int            blockshift)
562 {
563         unsigned int            pagefactor = 1;
564         unsigned int            bitshift = BITS_PER_LONG - 1;
565
566         /* Figure out maximum filesize, on Linux this can depend on
567          * the filesystem blocksize (on 32 bit platforms).
568          * __block_write_begin does this in an [unsigned] long...
569          *      page->index << (PAGE_SHIFT - bbits)
570          * So, for page sized blocks (4K on 32 bit platforms),
571          * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is
572          *      (((u64)PAGE_SIZE << (BITS_PER_LONG-1))-1)
573          * but for smaller blocksizes it is less (bbits = log2 bsize).
574          * Note1: get_block_t takes a long (implicit cast from above)
575          * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch
576          * can optionally convert the [unsigned] long from above into
577          * an [unsigned] long long.
578          */
579
580 #if BITS_PER_LONG == 32
581 # if defined(CONFIG_LBDAF)
582         ASSERT(sizeof(sector_t) == 8);
583         pagefactor = PAGE_SIZE;
584         bitshift = BITS_PER_LONG;
585 # else
586         pagefactor = PAGE_SIZE >> (PAGE_SHIFT - blockshift);
587 # endif
588 #endif
589
590         return (((uint64_t)pagefactor) << bitshift) - 1;
591 }
592
593 /*
594  * Set parameters for inode allocation heuristics, taking into account
595  * filesystem size and inode32/inode64 mount options; i.e. specifically
596  * whether or not XFS_MOUNT_SMALL_INUMS is set.
597  *
598  * Inode allocation patterns are altered only if inode32 is requested
599  * (XFS_MOUNT_SMALL_INUMS), and the filesystem is sufficiently large.
600  * If altered, XFS_MOUNT_32BITINODES is set as well.
601  *
602  * An agcount independent of that in the mount structure is provided
603  * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
604  * to the potentially higher ag count.
605  *
606  * Returns the maximum AG index which may contain inodes.
607  */
608 xfs_agnumber_t
609 xfs_set_inode_alloc(
610         struct xfs_mount *mp,
611         xfs_agnumber_t  agcount)
612 {
613         xfs_agnumber_t  index;
614         xfs_agnumber_t  maxagi = 0;
615         xfs_sb_t        *sbp = &mp->m_sb;
616         xfs_agnumber_t  max_metadata;
617         xfs_agino_t     agino;
618         xfs_ino_t       ino;
619
620         /*
621          * Calculate how much should be reserved for inodes to meet
622          * the max inode percentage.  Used only for inode32.
623          */
624         if (mp->m_maxicount) {
625                 uint64_t        icount;
626
627                 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
628                 do_div(icount, 100);
629                 icount += sbp->sb_agblocks - 1;
630                 do_div(icount, sbp->sb_agblocks);
631                 max_metadata = icount;
632         } else {
633                 max_metadata = agcount;
634         }
635
636         /* Get the last possible inode in the filesystem */
637         agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
638         ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
639
640         /*
641          * If user asked for no more than 32-bit inodes, and the fs is
642          * sufficiently large, set XFS_MOUNT_32BITINODES if we must alter
643          * the allocator to accommodate the request.
644          */
645         if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
646                 mp->m_flags |= XFS_MOUNT_32BITINODES;
647         else
648                 mp->m_flags &= ~XFS_MOUNT_32BITINODES;
649
650         for (index = 0; index < agcount; index++) {
651                 struct xfs_perag        *pag;
652
653                 ino = XFS_AGINO_TO_INO(mp, index, agino);
654
655                 pag = xfs_perag_get(mp, index);
656
657                 if (mp->m_flags & XFS_MOUNT_32BITINODES) {
658                         if (ino > XFS_MAXINUMBER_32) {
659                                 pag->pagi_inodeok = 0;
660                                 pag->pagf_metadata = 0;
661                         } else {
662                                 pag->pagi_inodeok = 1;
663                                 maxagi++;
664                                 if (index < max_metadata)
665                                         pag->pagf_metadata = 1;
666                                 else
667                                         pag->pagf_metadata = 0;
668                         }
669                 } else {
670                         pag->pagi_inodeok = 1;
671                         pag->pagf_metadata = 0;
672                 }
673
674                 xfs_perag_put(pag);
675         }
676
677         return (mp->m_flags & XFS_MOUNT_32BITINODES) ? maxagi : agcount;
678 }
679
680 STATIC int
681 xfs_blkdev_get(
682         xfs_mount_t             *mp,
683         const char              *name,
684         struct block_device     **bdevp)
685 {
686         int                     error = 0;
687
688         *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
689                                     mp);
690         if (IS_ERR(*bdevp)) {
691                 error = PTR_ERR(*bdevp);
692                 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
693         }
694
695         return error;
696 }
697
698 STATIC void
699 xfs_blkdev_put(
700         struct block_device     *bdev)
701 {
702         if (bdev)
703                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
704 }
705
706 void
707 xfs_blkdev_issue_flush(
708         xfs_buftarg_t           *buftarg)
709 {
710         blkdev_issue_flush(buftarg->bt_bdev, GFP_NOFS, NULL);
711 }
712
713 STATIC void
714 xfs_close_devices(
715         struct xfs_mount        *mp)
716 {
717         struct dax_device *dax_ddev = mp->m_ddev_targp->bt_daxdev;
718
719         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
720                 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
721                 struct dax_device *dax_logdev = mp->m_logdev_targp->bt_daxdev;
722
723                 xfs_free_buftarg(mp, mp->m_logdev_targp);
724                 xfs_blkdev_put(logdev);
725                 fs_put_dax(dax_logdev);
726         }
727         if (mp->m_rtdev_targp) {
728                 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
729                 struct dax_device *dax_rtdev = mp->m_rtdev_targp->bt_daxdev;
730
731                 xfs_free_buftarg(mp, mp->m_rtdev_targp);
732                 xfs_blkdev_put(rtdev);
733                 fs_put_dax(dax_rtdev);
734         }
735         xfs_free_buftarg(mp, mp->m_ddev_targp);
736         fs_put_dax(dax_ddev);
737 }
738
739 /*
740  * The file system configurations are:
741  *      (1) device (partition) with data and internal log
742  *      (2) logical volume with data and log subvolumes.
743  *      (3) logical volume with data, log, and realtime subvolumes.
744  *
745  * We only have to handle opening the log and realtime volumes here if
746  * they are present.  The data subvolume has already been opened by
747  * get_sb_bdev() and is stored in sb->s_bdev.
748  */
749 STATIC int
750 xfs_open_devices(
751         struct xfs_mount        *mp)
752 {
753         struct block_device     *ddev = mp->m_super->s_bdev;
754         struct dax_device       *dax_ddev = fs_dax_get_by_bdev(ddev);
755         struct dax_device       *dax_logdev = NULL, *dax_rtdev = NULL;
756         struct block_device     *logdev = NULL, *rtdev = NULL;
757         int                     error;
758
759         /*
760          * Open real time and log devices - order is important.
761          */
762         if (mp->m_logname) {
763                 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
764                 if (error)
765                         goto out;
766                 dax_logdev = fs_dax_get_by_bdev(logdev);
767         }
768
769         if (mp->m_rtname) {
770                 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
771                 if (error)
772                         goto out_close_logdev;
773
774                 if (rtdev == ddev || rtdev == logdev) {
775                         xfs_warn(mp,
776         "Cannot mount filesystem with identical rtdev and ddev/logdev.");
777                         error = -EINVAL;
778                         goto out_close_rtdev;
779                 }
780                 dax_rtdev = fs_dax_get_by_bdev(rtdev);
781         }
782
783         /*
784          * Setup xfs_mount buffer target pointers
785          */
786         error = -ENOMEM;
787         mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev, dax_ddev);
788         if (!mp->m_ddev_targp)
789                 goto out_close_rtdev;
790
791         if (rtdev) {
792                 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev, dax_rtdev);
793                 if (!mp->m_rtdev_targp)
794                         goto out_free_ddev_targ;
795         }
796
797         if (logdev && logdev != ddev) {
798                 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev, dax_logdev);
799                 if (!mp->m_logdev_targp)
800                         goto out_free_rtdev_targ;
801         } else {
802                 mp->m_logdev_targp = mp->m_ddev_targp;
803         }
804
805         return 0;
806
807  out_free_rtdev_targ:
808         if (mp->m_rtdev_targp)
809                 xfs_free_buftarg(mp, mp->m_rtdev_targp);
810  out_free_ddev_targ:
811         xfs_free_buftarg(mp, mp->m_ddev_targp);
812  out_close_rtdev:
813         xfs_blkdev_put(rtdev);
814         fs_put_dax(dax_rtdev);
815  out_close_logdev:
816         if (logdev && logdev != ddev) {
817                 xfs_blkdev_put(logdev);
818                 fs_put_dax(dax_logdev);
819         }
820  out:
821         fs_put_dax(dax_ddev);
822         return error;
823 }
824
825 /*
826  * Setup xfs_mount buffer target pointers based on superblock
827  */
828 STATIC int
829 xfs_setup_devices(
830         struct xfs_mount        *mp)
831 {
832         int                     error;
833
834         error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
835         if (error)
836                 return error;
837
838         if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
839                 unsigned int    log_sector_size = BBSIZE;
840
841                 if (xfs_sb_version_hassector(&mp->m_sb))
842                         log_sector_size = mp->m_sb.sb_logsectsize;
843                 error = xfs_setsize_buftarg(mp->m_logdev_targp,
844                                             log_sector_size);
845                 if (error)
846                         return error;
847         }
848         if (mp->m_rtdev_targp) {
849                 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
850                                             mp->m_sb.sb_sectsize);
851                 if (error)
852                         return error;
853         }
854
855         return 0;
856 }
857
858 STATIC int
859 xfs_init_mount_workqueues(
860         struct xfs_mount        *mp)
861 {
862         mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
863                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 1, mp->m_fsname);
864         if (!mp->m_buf_workqueue)
865                 goto out;
866
867         mp->m_data_workqueue = alloc_workqueue("xfs-data/%s",
868                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
869         if (!mp->m_data_workqueue)
870                 goto out_destroy_buf;
871
872         mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
873                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
874         if (!mp->m_unwritten_workqueue)
875                 goto out_destroy_data_iodone_queue;
876
877         mp->m_cil_workqueue = alloc_workqueue("xfs-cil/%s",
878                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
879         if (!mp->m_cil_workqueue)
880                 goto out_destroy_unwritten;
881
882         mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
883                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
884         if (!mp->m_reclaim_workqueue)
885                 goto out_destroy_cil;
886
887         mp->m_log_workqueue = alloc_workqueue("xfs-log/%s",
888                         WQ_MEM_RECLAIM|WQ_FREEZABLE|WQ_HIGHPRI, 0,
889                         mp->m_fsname);
890         if (!mp->m_log_workqueue)
891                 goto out_destroy_reclaim;
892
893         mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
894                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0, mp->m_fsname);
895         if (!mp->m_eofblocks_workqueue)
896                 goto out_destroy_log;
897
898         mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s", WQ_FREEZABLE, 0,
899                                                mp->m_fsname);
900         if (!mp->m_sync_workqueue)
901                 goto out_destroy_eofb;
902
903         return 0;
904
905 out_destroy_eofb:
906         destroy_workqueue(mp->m_eofblocks_workqueue);
907 out_destroy_log:
908         destroy_workqueue(mp->m_log_workqueue);
909 out_destroy_reclaim:
910         destroy_workqueue(mp->m_reclaim_workqueue);
911 out_destroy_cil:
912         destroy_workqueue(mp->m_cil_workqueue);
913 out_destroy_unwritten:
914         destroy_workqueue(mp->m_unwritten_workqueue);
915 out_destroy_data_iodone_queue:
916         destroy_workqueue(mp->m_data_workqueue);
917 out_destroy_buf:
918         destroy_workqueue(mp->m_buf_workqueue);
919 out:
920         return -ENOMEM;
921 }
922
923 STATIC void
924 xfs_destroy_mount_workqueues(
925         struct xfs_mount        *mp)
926 {
927         destroy_workqueue(mp->m_sync_workqueue);
928         destroy_workqueue(mp->m_eofblocks_workqueue);
929         destroy_workqueue(mp->m_log_workqueue);
930         destroy_workqueue(mp->m_reclaim_workqueue);
931         destroy_workqueue(mp->m_cil_workqueue);
932         destroy_workqueue(mp->m_data_workqueue);
933         destroy_workqueue(mp->m_unwritten_workqueue);
934         destroy_workqueue(mp->m_buf_workqueue);
935 }
936
937 /*
938  * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
939  * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
940  * for IO to complete so that we effectively throttle multiple callers to the
941  * rate at which IO is completing.
942  */
943 void
944 xfs_flush_inodes(
945         struct xfs_mount        *mp)
946 {
947         struct super_block      *sb = mp->m_super;
948
949         if (down_read_trylock(&sb->s_umount)) {
950                 sync_inodes_sb(sb);
951                 up_read(&sb->s_umount);
952         }
953 }
954
955 /* Catch misguided souls that try to use this interface on XFS */
956 STATIC struct inode *
957 xfs_fs_alloc_inode(
958         struct super_block      *sb)
959 {
960         BUG();
961         return NULL;
962 }
963
964 /*
965  * Now that the generic code is guaranteed not to be accessing
966  * the linux inode, we can inactivate and reclaim the inode.
967  */
968 STATIC void
969 xfs_fs_destroy_inode(
970         struct inode            *inode)
971 {
972         struct xfs_inode        *ip = XFS_I(inode);
973         int                     error;
974
975         trace_xfs_destroy_inode(ip);
976
977         ASSERT(!rwsem_is_locked(&inode->i_rwsem));
978         XFS_STATS_INC(ip->i_mount, vn_rele);
979         XFS_STATS_INC(ip->i_mount, vn_remove);
980
981         if (xfs_is_reflink_inode(ip)) {
982                 error = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, true);
983                 if (error && !XFS_FORCED_SHUTDOWN(ip->i_mount))
984                         xfs_warn(ip->i_mount,
985 "Error %d while evicting CoW blocks for inode %llu.",
986                                         error, ip->i_ino);
987         }
988
989         xfs_inactive(ip);
990
991         ASSERT(XFS_FORCED_SHUTDOWN(ip->i_mount) || ip->i_delayed_blks == 0);
992         XFS_STATS_INC(ip->i_mount, vn_reclaim);
993
994         /*
995          * We should never get here with one of the reclaim flags already set.
996          */
997         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIMABLE));
998         ASSERT_ALWAYS(!xfs_iflags_test(ip, XFS_IRECLAIM));
999
1000         /*
1001          * We always use background reclaim here because even if the
1002          * inode is clean, it still may be under IO and hence we have
1003          * to take the flush lock. The background reclaim path handles
1004          * this more efficiently than we can here, so simply let background
1005          * reclaim tear down all inodes.
1006          */
1007         xfs_inode_set_reclaim_tag(ip);
1008 }
1009
1010 /*
1011  * Slab object creation initialisation for the XFS inode.
1012  * This covers only the idempotent fields in the XFS inode;
1013  * all other fields need to be initialised on allocation
1014  * from the slab. This avoids the need to repeatedly initialise
1015  * fields in the xfs inode that left in the initialise state
1016  * when freeing the inode.
1017  */
1018 STATIC void
1019 xfs_fs_inode_init_once(
1020         void                    *inode)
1021 {
1022         struct xfs_inode        *ip = inode;
1023
1024         memset(ip, 0, sizeof(struct xfs_inode));
1025
1026         /* vfs inode */
1027         inode_init_once(VFS_I(ip));
1028
1029         /* xfs inode */
1030         atomic_set(&ip->i_pincount, 0);
1031         spin_lock_init(&ip->i_flags_lock);
1032
1033         mrlock_init(&ip->i_mmaplock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1034                      "xfsino", ip->i_ino);
1035         mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
1036                      "xfsino", ip->i_ino);
1037 }
1038
1039 /*
1040  * We do an unlocked check for XFS_IDONTCACHE here because we are already
1041  * serialised against cache hits here via the inode->i_lock and igrab() in
1042  * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
1043  * racing with us, and it avoids needing to grab a spinlock here for every inode
1044  * we drop the final reference on.
1045  */
1046 STATIC int
1047 xfs_fs_drop_inode(
1048         struct inode            *inode)
1049 {
1050         struct xfs_inode        *ip = XFS_I(inode);
1051
1052         /*
1053          * If this unlinked inode is in the middle of recovery, don't
1054          * drop the inode just yet; log recovery will take care of
1055          * that.  See the comment for this inode flag.
1056          */
1057         if (ip->i_flags & XFS_IRECOVERY) {
1058                 ASSERT(ip->i_mount->m_log->l_flags & XLOG_RECOVERY_NEEDED);
1059                 return 0;
1060         }
1061
1062         return generic_drop_inode(inode) || (ip->i_flags & XFS_IDONTCACHE);
1063 }
1064
1065 STATIC void
1066 xfs_free_fsname(
1067         struct xfs_mount        *mp)
1068 {
1069         kfree(mp->m_fsname);
1070         kfree(mp->m_rtname);
1071         kfree(mp->m_logname);
1072 }
1073
1074 STATIC int
1075 xfs_fs_sync_fs(
1076         struct super_block      *sb,
1077         int                     wait)
1078 {
1079         struct xfs_mount        *mp = XFS_M(sb);
1080
1081         /*
1082          * Doing anything during the async pass would be counterproductive.
1083          */
1084         if (!wait)
1085                 return 0;
1086
1087         xfs_log_force(mp, XFS_LOG_SYNC);
1088         if (laptop_mode) {
1089                 /*
1090                  * The disk must be active because we're syncing.
1091                  * We schedule log work now (now that the disk is
1092                  * active) instead of later (when it might not be).
1093                  */
1094                 flush_delayed_work(&mp->m_log->l_work);
1095         }
1096
1097         return 0;
1098 }
1099
1100 STATIC int
1101 xfs_fs_statfs(
1102         struct dentry           *dentry,
1103         struct kstatfs          *statp)
1104 {
1105         struct xfs_mount        *mp = XFS_M(dentry->d_sb);
1106         xfs_sb_t                *sbp = &mp->m_sb;
1107         struct xfs_inode        *ip = XFS_I(d_inode(dentry));
1108         uint64_t                fakeinos, id;
1109         uint64_t                icount;
1110         uint64_t                ifree;
1111         uint64_t                fdblocks;
1112         xfs_extlen_t            lsize;
1113         int64_t                 ffree;
1114
1115         statp->f_type = XFS_SB_MAGIC;
1116         statp->f_namelen = MAXNAMELEN - 1;
1117
1118         id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
1119         statp->f_fsid.val[0] = (u32)id;
1120         statp->f_fsid.val[1] = (u32)(id >> 32);
1121
1122         icount = percpu_counter_sum(&mp->m_icount);
1123         ifree = percpu_counter_sum(&mp->m_ifree);
1124         fdblocks = percpu_counter_sum(&mp->m_fdblocks);
1125
1126         spin_lock(&mp->m_sb_lock);
1127         statp->f_bsize = sbp->sb_blocksize;
1128         lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
1129         statp->f_blocks = sbp->sb_dblocks - lsize;
1130         spin_unlock(&mp->m_sb_lock);
1131
1132         statp->f_bfree = fdblocks - mp->m_alloc_set_aside;
1133         statp->f_bavail = statp->f_bfree;
1134
1135         fakeinos = statp->f_bfree << sbp->sb_inopblog;
1136         statp->f_files = MIN(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
1137         if (mp->m_maxicount)
1138                 statp->f_files = min_t(typeof(statp->f_files),
1139                                         statp->f_files,
1140                                         mp->m_maxicount);
1141
1142         /* If sb_icount overshot maxicount, report actual allocation */
1143         statp->f_files = max_t(typeof(statp->f_files),
1144                                         statp->f_files,
1145                                         sbp->sb_icount);
1146
1147         /* make sure statp->f_ffree does not underflow */
1148         ffree = statp->f_files - (icount - ifree);
1149         statp->f_ffree = max_t(int64_t, ffree, 0);
1150
1151
1152         if ((ip->i_d.di_flags & XFS_DIFLAG_PROJINHERIT) &&
1153             ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
1154                               (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
1155                 xfs_qm_statvfs(ip, statp);
1156         return 0;
1157 }
1158
1159 STATIC void
1160 xfs_save_resvblks(struct xfs_mount *mp)
1161 {
1162         uint64_t resblks = 0;
1163
1164         mp->m_resblks_save = mp->m_resblks;
1165         xfs_reserve_blocks(mp, &resblks, NULL);
1166 }
1167
1168 STATIC void
1169 xfs_restore_resvblks(struct xfs_mount *mp)
1170 {
1171         uint64_t resblks;
1172
1173         if (mp->m_resblks_save) {
1174                 resblks = mp->m_resblks_save;
1175                 mp->m_resblks_save = 0;
1176         } else
1177                 resblks = xfs_default_resblks(mp);
1178
1179         xfs_reserve_blocks(mp, &resblks, NULL);
1180 }
1181
1182 /*
1183  * Trigger writeback of all the dirty metadata in the file system.
1184  *
1185  * This ensures that the metadata is written to their location on disk rather
1186  * than just existing in transactions in the log. This means after a quiesce
1187  * there is no log replay required to write the inodes to disk - this is the
1188  * primary difference between a sync and a quiesce.
1189  *
1190  * Note: xfs_log_quiesce() stops background log work - the callers must ensure
1191  * it is started again when appropriate.
1192  */
1193 void
1194 xfs_quiesce_attr(
1195         struct xfs_mount        *mp)
1196 {
1197         int     error = 0;
1198
1199         /* wait for all modifications to complete */
1200         while (atomic_read(&mp->m_active_trans) > 0)
1201                 delay(100);
1202
1203         /* force the log to unpin objects from the now complete transactions */
1204         xfs_log_force(mp, XFS_LOG_SYNC);
1205
1206         /* reclaim inodes to do any IO before the freeze completes */
1207         xfs_reclaim_inodes(mp, 0);
1208         xfs_reclaim_inodes(mp, SYNC_WAIT);
1209
1210         /* Push the superblock and write an unmount record */
1211         error = xfs_log_sbcount(mp);
1212         if (error)
1213                 xfs_warn(mp, "xfs_attr_quiesce: failed to log sb changes. "
1214                                 "Frozen image may not be consistent.");
1215         /*
1216          * Just warn here till VFS can correctly support
1217          * read-only remount without racing.
1218          */
1219         WARN_ON(atomic_read(&mp->m_active_trans) != 0);
1220
1221         xfs_log_quiesce(mp);
1222 }
1223
1224 STATIC int
1225 xfs_test_remount_options(
1226         struct super_block      *sb,
1227         struct xfs_mount        *mp,
1228         char                    *options)
1229 {
1230         int                     error = 0;
1231         struct xfs_mount        *tmp_mp;
1232
1233         tmp_mp = kmem_zalloc(sizeof(*tmp_mp), KM_MAYFAIL);
1234         if (!tmp_mp)
1235                 return -ENOMEM;
1236
1237         tmp_mp->m_super = sb;
1238         error = xfs_parseargs(tmp_mp, options);
1239         xfs_free_fsname(tmp_mp);
1240         kmem_free(tmp_mp);
1241
1242         return error;
1243 }
1244
1245 STATIC int
1246 xfs_fs_remount(
1247         struct super_block      *sb,
1248         int                     *flags,
1249         char                    *options)
1250 {
1251         struct xfs_mount        *mp = XFS_M(sb);
1252         xfs_sb_t                *sbp = &mp->m_sb;
1253         substring_t             args[MAX_OPT_ARGS];
1254         char                    *p;
1255         int                     error;
1256
1257         /* First, check for complete junk; i.e. invalid options */
1258         error = xfs_test_remount_options(sb, mp, options);
1259         if (error)
1260                 return error;
1261
1262         sync_filesystem(sb);
1263         while ((p = strsep(&options, ",")) != NULL) {
1264                 int token;
1265
1266                 if (!*p)
1267                         continue;
1268
1269                 token = match_token(p, tokens, args);
1270                 switch (token) {
1271                 case Opt_barrier:
1272                         xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1273                         mp->m_flags |= XFS_MOUNT_BARRIER;
1274                         break;
1275                 case Opt_nobarrier:
1276                         xfs_warn(mp, "%s option is deprecated, ignoring.", p);
1277                         mp->m_flags &= ~XFS_MOUNT_BARRIER;
1278                         break;
1279                 case Opt_inode64:
1280                         mp->m_flags &= ~XFS_MOUNT_SMALL_INUMS;
1281                         mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1282                         break;
1283                 case Opt_inode32:
1284                         mp->m_flags |= XFS_MOUNT_SMALL_INUMS;
1285                         mp->m_maxagi = xfs_set_inode_alloc(mp, sbp->sb_agcount);
1286                         break;
1287                 default:
1288                         /*
1289                          * Logically we would return an error here to prevent
1290                          * users from believing they might have changed
1291                          * mount options using remount which can't be changed.
1292                          *
1293                          * But unfortunately mount(8) adds all options from
1294                          * mtab and fstab to the mount arguments in some cases
1295                          * so we can't blindly reject options, but have to
1296                          * check for each specified option if it actually
1297                          * differs from the currently set option and only
1298                          * reject it if that's the case.
1299                          *
1300                          * Until that is implemented we return success for
1301                          * every remount request, and silently ignore all
1302                          * options that we can't actually change.
1303                          */
1304 #if 0
1305                         xfs_info(mp,
1306                 "mount option \"%s\" not supported for remount", p);
1307                         return -EINVAL;
1308 #else
1309                         break;
1310 #endif
1311                 }
1312         }
1313
1314         /* ro -> rw */
1315         if ((mp->m_flags & XFS_MOUNT_RDONLY) && !(*flags & SB_RDONLY)) {
1316                 if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
1317                         xfs_warn(mp,
1318                 "ro->rw transition prohibited on norecovery mount");
1319                         return -EINVAL;
1320                 }
1321
1322                 if (XFS_SB_VERSION_NUM(sbp) == XFS_SB_VERSION_5 &&
1323                     xfs_sb_has_ro_compat_feature(sbp,
1324                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1325                         xfs_warn(mp,
1326 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1327                                 (sbp->sb_features_ro_compat &
1328                                         XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1329                         return -EINVAL;
1330                 }
1331
1332                 mp->m_flags &= ~XFS_MOUNT_RDONLY;
1333
1334                 /*
1335                  * If this is the first remount to writeable state we
1336                  * might have some superblock changes to update.
1337                  */
1338                 if (mp->m_update_sb) {
1339                         error = xfs_sync_sb(mp, false);
1340                         if (error) {
1341                                 xfs_warn(mp, "failed to write sb changes");
1342                                 return error;
1343                         }
1344                         mp->m_update_sb = false;
1345                 }
1346
1347                 /*
1348                  * Fill out the reserve pool if it is empty. Use the stashed
1349                  * value if it is non-zero, otherwise go with the default.
1350                  */
1351                 xfs_restore_resvblks(mp);
1352                 xfs_log_work_queue(mp);
1353                 xfs_queue_eofblocks(mp);
1354
1355                 /* Recover any CoW blocks that never got remapped. */
1356                 error = xfs_reflink_recover_cow(mp);
1357                 if (error) {
1358                         xfs_err(mp,
1359         "Error %d recovering leftover CoW allocations.", error);
1360                         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1361                         return error;
1362                 }
1363
1364                 /* Create the per-AG metadata reservation pool .*/
1365                 error = xfs_fs_reserve_ag_blocks(mp);
1366                 if (error && error != -ENOSPC)
1367                         return error;
1368         }
1369
1370         /* rw -> ro */
1371         if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & SB_RDONLY)) {
1372                 /* Free the per-AG metadata reservation pool. */
1373                 error = xfs_fs_unreserve_ag_blocks(mp);
1374                 if (error) {
1375                         xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1376                         return error;
1377                 }
1378
1379                 /*
1380                  * Before we sync the metadata, we need to free up the reserve
1381                  * block pool so that the used block count in the superblock on
1382                  * disk is correct at the end of the remount. Stash the current
1383                  * reserve pool size so that if we get remounted rw, we can
1384                  * return it to the same size.
1385                  */
1386                 xfs_save_resvblks(mp);
1387
1388                 /*
1389                  * Cancel background eofb scanning so it cannot race with the
1390                  * final log force+buftarg wait and deadlock the remount.
1391                  */
1392                 cancel_delayed_work_sync(&mp->m_eofblocks_work);
1393
1394                 xfs_quiesce_attr(mp);
1395                 mp->m_flags |= XFS_MOUNT_RDONLY;
1396         }
1397
1398         return 0;
1399 }
1400
1401 /*
1402  * Second stage of a freeze. The data is already frozen so we only
1403  * need to take care of the metadata. Once that's done sync the superblock
1404  * to the log to dirty it in case of a crash while frozen. This ensures that we
1405  * will recover the unlinked inode lists on the next mount.
1406  */
1407 STATIC int
1408 xfs_fs_freeze(
1409         struct super_block      *sb)
1410 {
1411         struct xfs_mount        *mp = XFS_M(sb);
1412
1413         xfs_save_resvblks(mp);
1414         xfs_quiesce_attr(mp);
1415         return xfs_sync_sb(mp, true);
1416 }
1417
1418 STATIC int
1419 xfs_fs_unfreeze(
1420         struct super_block      *sb)
1421 {
1422         struct xfs_mount        *mp = XFS_M(sb);
1423
1424         xfs_restore_resvblks(mp);
1425         xfs_log_work_queue(mp);
1426         return 0;
1427 }
1428
1429 STATIC int
1430 xfs_fs_show_options(
1431         struct seq_file         *m,
1432         struct dentry           *root)
1433 {
1434         return xfs_showargs(XFS_M(root->d_sb), m);
1435 }
1436
1437 /*
1438  * This function fills in xfs_mount_t fields based on mount args.
1439  * Note: the superblock _has_ now been read in.
1440  */
1441 STATIC int
1442 xfs_finish_flags(
1443         struct xfs_mount        *mp)
1444 {
1445         int                     ronly = (mp->m_flags & XFS_MOUNT_RDONLY);
1446
1447         /* Fail a mount where the logbuf is smaller than the log stripe */
1448         if (xfs_sb_version_haslogv2(&mp->m_sb)) {
1449                 if (mp->m_logbsize <= 0 &&
1450                     mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
1451                         mp->m_logbsize = mp->m_sb.sb_logsunit;
1452                 } else if (mp->m_logbsize > 0 &&
1453                            mp->m_logbsize < mp->m_sb.sb_logsunit) {
1454                         xfs_warn(mp,
1455                 "logbuf size must be greater than or equal to log stripe size");
1456                         return -EINVAL;
1457                 }
1458         } else {
1459                 /* Fail a mount if the logbuf is larger than 32K */
1460                 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1461                         xfs_warn(mp,
1462                 "logbuf size for version 1 logs must be 16K or 32K");
1463                         return -EINVAL;
1464                 }
1465         }
1466
1467         /*
1468          * V5 filesystems always use attr2 format for attributes.
1469          */
1470         if (xfs_sb_version_hascrc(&mp->m_sb) &&
1471             (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1472                 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1473                              "attr2 is always enabled for V5 filesystems.");
1474                 return -EINVAL;
1475         }
1476
1477         /*
1478          * mkfs'ed attr2 will turn on attr2 mount unless explicitly
1479          * told by noattr2 to turn it off
1480          */
1481         if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1482             !(mp->m_flags & XFS_MOUNT_NOATTR2))
1483                 mp->m_flags |= XFS_MOUNT_ATTR2;
1484
1485         /*
1486          * prohibit r/w mounts of read-only filesystems
1487          */
1488         if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !ronly) {
1489                 xfs_warn(mp,
1490                         "cannot mount a read-only filesystem as read-write");
1491                 return -EROFS;
1492         }
1493
1494         if ((mp->m_qflags & (XFS_GQUOTA_ACCT | XFS_GQUOTA_ACTIVE)) &&
1495             (mp->m_qflags & (XFS_PQUOTA_ACCT | XFS_PQUOTA_ACTIVE)) &&
1496             !xfs_sb_version_has_pquotino(&mp->m_sb)) {
1497                 xfs_warn(mp,
1498                   "Super block does not support project and group quota together");
1499                 return -EINVAL;
1500         }
1501
1502         return 0;
1503 }
1504
1505 static int
1506 xfs_init_percpu_counters(
1507         struct xfs_mount        *mp)
1508 {
1509         int             error;
1510
1511         error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1512         if (error)
1513                 return -ENOMEM;
1514
1515         error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1516         if (error)
1517                 goto free_icount;
1518
1519         error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1520         if (error)
1521                 goto free_ifree;
1522
1523         return 0;
1524
1525 free_ifree:
1526         percpu_counter_destroy(&mp->m_ifree);
1527 free_icount:
1528         percpu_counter_destroy(&mp->m_icount);
1529         return -ENOMEM;
1530 }
1531
1532 void
1533 xfs_reinit_percpu_counters(
1534         struct xfs_mount        *mp)
1535 {
1536         percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1537         percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1538         percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1539 }
1540
1541 static void
1542 xfs_destroy_percpu_counters(
1543         struct xfs_mount        *mp)
1544 {
1545         percpu_counter_destroy(&mp->m_icount);
1546         percpu_counter_destroy(&mp->m_ifree);
1547         percpu_counter_destroy(&mp->m_fdblocks);
1548 }
1549
1550 STATIC int
1551 xfs_fs_fill_super(
1552         struct super_block      *sb,
1553         void                    *data,
1554         int                     silent)
1555 {
1556         struct inode            *root;
1557         struct xfs_mount        *mp = NULL;
1558         int                     flags = 0, error = -ENOMEM;
1559
1560         mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL);
1561         if (!mp)
1562                 goto out;
1563
1564         spin_lock_init(&mp->m_sb_lock);
1565         mutex_init(&mp->m_growlock);
1566         atomic_set(&mp->m_active_trans, 0);
1567         INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1568         INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
1569         INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
1570         mp->m_kobj.kobject.kset = xfs_kset;
1571
1572         mp->m_super = sb;
1573         sb->s_fs_info = mp;
1574
1575         error = xfs_parseargs(mp, (char *)data);
1576         if (error)
1577                 goto out_free_fsname;
1578
1579         sb_min_blocksize(sb, BBSIZE);
1580         sb->s_xattr = xfs_xattr_handlers;
1581         sb->s_export_op = &xfs_export_operations;
1582 #ifdef CONFIG_XFS_QUOTA
1583         sb->s_qcop = &xfs_quotactl_operations;
1584         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1585 #endif
1586         sb->s_op = &xfs_super_operations;
1587
1588         if (silent)
1589                 flags |= XFS_MFSI_QUIET;
1590
1591         error = xfs_open_devices(mp);
1592         if (error)
1593                 goto out_free_fsname;
1594
1595         error = xfs_init_mount_workqueues(mp);
1596         if (error)
1597                 goto out_close_devices;
1598
1599         error = xfs_init_percpu_counters(mp);
1600         if (error)
1601                 goto out_destroy_workqueues;
1602
1603         /* Allocate stats memory before we do operations that might use it */
1604         mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1605         if (!mp->m_stats.xs_stats) {
1606                 error = -ENOMEM;
1607                 goto out_destroy_counters;
1608         }
1609
1610         error = xfs_readsb(mp, flags);
1611         if (error)
1612                 goto out_free_stats;
1613
1614         error = xfs_finish_flags(mp);
1615         if (error)
1616                 goto out_free_sb;
1617
1618         error = xfs_setup_devices(mp);
1619         if (error)
1620                 goto out_free_sb;
1621
1622         error = xfs_filestream_mount(mp);
1623         if (error)
1624                 goto out_free_sb;
1625
1626         /*
1627          * we must configure the block size in the superblock before we run the
1628          * full mount process as the mount process can lookup and cache inodes.
1629          */
1630         sb->s_magic = XFS_SB_MAGIC;
1631         sb->s_blocksize = mp->m_sb.sb_blocksize;
1632         sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1633         sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits);
1634         sb->s_max_links = XFS_MAXLINK;
1635         sb->s_time_gran = 1;
1636         set_posix_acl_flag(sb);
1637
1638         /* version 5 superblocks support inode version counters. */
1639         if (XFS_SB_VERSION_NUM(&mp->m_sb) == XFS_SB_VERSION_5)
1640                 sb->s_flags |= SB_I_VERSION;
1641
1642         if (mp->m_flags & XFS_MOUNT_DAX) {
1643                 xfs_warn(mp,
1644                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1645
1646                 error = bdev_dax_supported(sb, sb->s_blocksize);
1647                 if (error) {
1648                         xfs_alert(mp,
1649                         "DAX unsupported by block device. Turning off DAX.");
1650                         mp->m_flags &= ~XFS_MOUNT_DAX;
1651                 }
1652                 if (xfs_sb_version_hasreflink(&mp->m_sb))
1653                         xfs_alert(mp,
1654                 "DAX and reflink have not been tested together!");
1655         }
1656
1657         if (mp->m_flags & XFS_MOUNT_DISCARD) {
1658                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1659
1660                 if (!blk_queue_discard(q)) {
1661                         xfs_warn(mp, "mounting with \"discard\" option, but "
1662                                         "the device does not support discard");
1663                         mp->m_flags &= ~XFS_MOUNT_DISCARD;
1664                 }
1665         }
1666
1667         if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1668                 if (mp->m_sb.sb_rblocks) {
1669                         xfs_alert(mp,
1670         "EXPERIMENTAL reverse mapping btree not compatible with realtime device!");
1671                         error = -EINVAL;
1672                         goto out_filestream_unmount;
1673                 }
1674                 xfs_alert(mp,
1675         "EXPERIMENTAL reverse mapping btree feature enabled. Use at your own risk!");
1676         }
1677
1678         if (xfs_sb_version_hasreflink(&mp->m_sb))
1679                 xfs_alert(mp,
1680         "EXPERIMENTAL reflink feature enabled. Use at your own risk!");
1681
1682         error = xfs_mountfs(mp);
1683         if (error)
1684                 goto out_filestream_unmount;
1685
1686         root = igrab(VFS_I(mp->m_rootip));
1687         if (!root) {
1688                 error = -ENOENT;
1689                 goto out_unmount;
1690         }
1691         sb->s_root = d_make_root(root);
1692         if (!sb->s_root) {
1693                 error = -ENOMEM;
1694                 goto out_unmount;
1695         }
1696
1697         return 0;
1698
1699  out_filestream_unmount:
1700         xfs_filestream_unmount(mp);
1701  out_free_sb:
1702         xfs_freesb(mp);
1703  out_free_stats:
1704         free_percpu(mp->m_stats.xs_stats);
1705  out_destroy_counters:
1706         xfs_destroy_percpu_counters(mp);
1707  out_destroy_workqueues:
1708         xfs_destroy_mount_workqueues(mp);
1709  out_close_devices:
1710         xfs_close_devices(mp);
1711  out_free_fsname:
1712         xfs_free_fsname(mp);
1713         kfree(mp);
1714  out:
1715         return error;
1716
1717  out_unmount:
1718         xfs_filestream_unmount(mp);
1719         xfs_unmountfs(mp);
1720         goto out_free_sb;
1721 }
1722
1723 STATIC void
1724 xfs_fs_put_super(
1725         struct super_block      *sb)
1726 {
1727         struct xfs_mount        *mp = XFS_M(sb);
1728
1729         xfs_notice(mp, "Unmounting Filesystem");
1730         xfs_filestream_unmount(mp);
1731         xfs_unmountfs(mp);
1732
1733         xfs_freesb(mp);
1734         free_percpu(mp->m_stats.xs_stats);
1735         xfs_destroy_percpu_counters(mp);
1736         xfs_destroy_mount_workqueues(mp);
1737         xfs_close_devices(mp);
1738         xfs_free_fsname(mp);
1739         kfree(mp);
1740 }
1741
1742 STATIC struct dentry *
1743 xfs_fs_mount(
1744         struct file_system_type *fs_type,
1745         int                     flags,
1746         const char              *dev_name,
1747         void                    *data)
1748 {
1749         return mount_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super);
1750 }
1751
1752 static long
1753 xfs_fs_nr_cached_objects(
1754         struct super_block      *sb,
1755         struct shrink_control   *sc)
1756 {
1757         return xfs_reclaim_inodes_count(XFS_M(sb));
1758 }
1759
1760 static long
1761 xfs_fs_free_cached_objects(
1762         struct super_block      *sb,
1763         struct shrink_control   *sc)
1764 {
1765         return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1766 }
1767
1768 static const struct super_operations xfs_super_operations = {
1769         .alloc_inode            = xfs_fs_alloc_inode,
1770         .destroy_inode          = xfs_fs_destroy_inode,
1771         .drop_inode             = xfs_fs_drop_inode,
1772         .put_super              = xfs_fs_put_super,
1773         .sync_fs                = xfs_fs_sync_fs,
1774         .freeze_fs              = xfs_fs_freeze,
1775         .unfreeze_fs            = xfs_fs_unfreeze,
1776         .statfs                 = xfs_fs_statfs,
1777         .remount_fs             = xfs_fs_remount,
1778         .show_options           = xfs_fs_show_options,
1779         .nr_cached_objects      = xfs_fs_nr_cached_objects,
1780         .free_cached_objects    = xfs_fs_free_cached_objects,
1781 };
1782
1783 static struct file_system_type xfs_fs_type = {
1784         .owner                  = THIS_MODULE,
1785         .name                   = "xfs",
1786         .mount                  = xfs_fs_mount,
1787         .kill_sb                = kill_block_super,
1788         .fs_flags               = FS_REQUIRES_DEV,
1789 };
1790 MODULE_ALIAS_FS("xfs");
1791
1792 STATIC int __init
1793 xfs_init_zones(void)
1794 {
1795         xfs_ioend_bioset = bioset_create(4 * MAX_BUF_PER_PAGE,
1796                         offsetof(struct xfs_ioend, io_inline_bio),
1797                         BIOSET_NEED_BVECS);
1798         if (!xfs_ioend_bioset)
1799                 goto out;
1800
1801         xfs_log_ticket_zone = kmem_zone_init(sizeof(xlog_ticket_t),
1802                                                 "xfs_log_ticket");
1803         if (!xfs_log_ticket_zone)
1804                 goto out_free_ioend_bioset;
1805
1806         xfs_bmap_free_item_zone = kmem_zone_init(
1807                         sizeof(struct xfs_extent_free_item),
1808                         "xfs_bmap_free_item");
1809         if (!xfs_bmap_free_item_zone)
1810                 goto out_destroy_log_ticket_zone;
1811
1812         xfs_btree_cur_zone = kmem_zone_init(sizeof(xfs_btree_cur_t),
1813                                                 "xfs_btree_cur");
1814         if (!xfs_btree_cur_zone)
1815                 goto out_destroy_bmap_free_item_zone;
1816
1817         xfs_da_state_zone = kmem_zone_init(sizeof(xfs_da_state_t),
1818                                                 "xfs_da_state");
1819         if (!xfs_da_state_zone)
1820                 goto out_destroy_btree_cur_zone;
1821
1822         xfs_ifork_zone = kmem_zone_init(sizeof(xfs_ifork_t), "xfs_ifork");
1823         if (!xfs_ifork_zone)
1824                 goto out_destroy_da_state_zone;
1825
1826         xfs_trans_zone = kmem_zone_init(sizeof(xfs_trans_t), "xfs_trans");
1827         if (!xfs_trans_zone)
1828                 goto out_destroy_ifork_zone;
1829
1830         xfs_log_item_desc_zone =
1831                 kmem_zone_init(sizeof(struct xfs_log_item_desc),
1832                                "xfs_log_item_desc");
1833         if (!xfs_log_item_desc_zone)
1834                 goto out_destroy_trans_zone;
1835
1836         /*
1837          * The size of the zone allocated buf log item is the maximum
1838          * size possible under XFS.  This wastes a little bit of memory,
1839          * but it is much faster.
1840          */
1841         xfs_buf_item_zone = kmem_zone_init(sizeof(struct xfs_buf_log_item),
1842                                            "xfs_buf_item");
1843         if (!xfs_buf_item_zone)
1844                 goto out_destroy_log_item_desc_zone;
1845
1846         xfs_efd_zone = kmem_zone_init((sizeof(xfs_efd_log_item_t) +
1847                         ((XFS_EFD_MAX_FAST_EXTENTS - 1) *
1848                                  sizeof(xfs_extent_t))), "xfs_efd_item");
1849         if (!xfs_efd_zone)
1850                 goto out_destroy_buf_item_zone;
1851
1852         xfs_efi_zone = kmem_zone_init((sizeof(xfs_efi_log_item_t) +
1853                         ((XFS_EFI_MAX_FAST_EXTENTS - 1) *
1854                                 sizeof(xfs_extent_t))), "xfs_efi_item");
1855         if (!xfs_efi_zone)
1856                 goto out_destroy_efd_zone;
1857
1858         xfs_inode_zone =
1859                 kmem_zone_init_flags(sizeof(xfs_inode_t), "xfs_inode",
1860                         KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | KM_ZONE_SPREAD |
1861                         KM_ZONE_ACCOUNT, xfs_fs_inode_init_once);
1862         if (!xfs_inode_zone)
1863                 goto out_destroy_efi_zone;
1864
1865         xfs_ili_zone =
1866                 kmem_zone_init_flags(sizeof(xfs_inode_log_item_t), "xfs_ili",
1867                                         KM_ZONE_SPREAD, NULL);
1868         if (!xfs_ili_zone)
1869                 goto out_destroy_inode_zone;
1870         xfs_icreate_zone = kmem_zone_init(sizeof(struct xfs_icreate_item),
1871                                         "xfs_icr");
1872         if (!xfs_icreate_zone)
1873                 goto out_destroy_ili_zone;
1874
1875         xfs_rud_zone = kmem_zone_init(sizeof(struct xfs_rud_log_item),
1876                         "xfs_rud_item");
1877         if (!xfs_rud_zone)
1878                 goto out_destroy_icreate_zone;
1879
1880         xfs_rui_zone = kmem_zone_init(
1881                         xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
1882                         "xfs_rui_item");
1883         if (!xfs_rui_zone)
1884                 goto out_destroy_rud_zone;
1885
1886         xfs_cud_zone = kmem_zone_init(sizeof(struct xfs_cud_log_item),
1887                         "xfs_cud_item");
1888         if (!xfs_cud_zone)
1889                 goto out_destroy_rui_zone;
1890
1891         xfs_cui_zone = kmem_zone_init(
1892                         xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
1893                         "xfs_cui_item");
1894         if (!xfs_cui_zone)
1895                 goto out_destroy_cud_zone;
1896
1897         xfs_bud_zone = kmem_zone_init(sizeof(struct xfs_bud_log_item),
1898                         "xfs_bud_item");
1899         if (!xfs_bud_zone)
1900                 goto out_destroy_cui_zone;
1901
1902         xfs_bui_zone = kmem_zone_init(
1903                         xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
1904                         "xfs_bui_item");
1905         if (!xfs_bui_zone)
1906                 goto out_destroy_bud_zone;
1907
1908         return 0;
1909
1910  out_destroy_bud_zone:
1911         kmem_zone_destroy(xfs_bud_zone);
1912  out_destroy_cui_zone:
1913         kmem_zone_destroy(xfs_cui_zone);
1914  out_destroy_cud_zone:
1915         kmem_zone_destroy(xfs_cud_zone);
1916  out_destroy_rui_zone:
1917         kmem_zone_destroy(xfs_rui_zone);
1918  out_destroy_rud_zone:
1919         kmem_zone_destroy(xfs_rud_zone);
1920  out_destroy_icreate_zone:
1921         kmem_zone_destroy(xfs_icreate_zone);
1922  out_destroy_ili_zone:
1923         kmem_zone_destroy(xfs_ili_zone);
1924  out_destroy_inode_zone:
1925         kmem_zone_destroy(xfs_inode_zone);
1926  out_destroy_efi_zone:
1927         kmem_zone_destroy(xfs_efi_zone);
1928  out_destroy_efd_zone:
1929         kmem_zone_destroy(xfs_efd_zone);
1930  out_destroy_buf_item_zone:
1931         kmem_zone_destroy(xfs_buf_item_zone);
1932  out_destroy_log_item_desc_zone:
1933         kmem_zone_destroy(xfs_log_item_desc_zone);
1934  out_destroy_trans_zone:
1935         kmem_zone_destroy(xfs_trans_zone);
1936  out_destroy_ifork_zone:
1937         kmem_zone_destroy(xfs_ifork_zone);
1938  out_destroy_da_state_zone:
1939         kmem_zone_destroy(xfs_da_state_zone);
1940  out_destroy_btree_cur_zone:
1941         kmem_zone_destroy(xfs_btree_cur_zone);
1942  out_destroy_bmap_free_item_zone:
1943         kmem_zone_destroy(xfs_bmap_free_item_zone);
1944  out_destroy_log_ticket_zone:
1945         kmem_zone_destroy(xfs_log_ticket_zone);
1946  out_free_ioend_bioset:
1947         bioset_free(xfs_ioend_bioset);
1948  out:
1949         return -ENOMEM;
1950 }
1951
1952 STATIC void
1953 xfs_destroy_zones(void)
1954 {
1955         /*
1956          * Make sure all delayed rcu free are flushed before we
1957          * destroy caches.
1958          */
1959         rcu_barrier();
1960         kmem_zone_destroy(xfs_bui_zone);
1961         kmem_zone_destroy(xfs_bud_zone);
1962         kmem_zone_destroy(xfs_cui_zone);
1963         kmem_zone_destroy(xfs_cud_zone);
1964         kmem_zone_destroy(xfs_rui_zone);
1965         kmem_zone_destroy(xfs_rud_zone);
1966         kmem_zone_destroy(xfs_icreate_zone);
1967         kmem_zone_destroy(xfs_ili_zone);
1968         kmem_zone_destroy(xfs_inode_zone);
1969         kmem_zone_destroy(xfs_efi_zone);
1970         kmem_zone_destroy(xfs_efd_zone);
1971         kmem_zone_destroy(xfs_buf_item_zone);
1972         kmem_zone_destroy(xfs_log_item_desc_zone);
1973         kmem_zone_destroy(xfs_trans_zone);
1974         kmem_zone_destroy(xfs_ifork_zone);
1975         kmem_zone_destroy(xfs_da_state_zone);
1976         kmem_zone_destroy(xfs_btree_cur_zone);
1977         kmem_zone_destroy(xfs_bmap_free_item_zone);
1978         kmem_zone_destroy(xfs_log_ticket_zone);
1979         bioset_free(xfs_ioend_bioset);
1980 }
1981
1982 STATIC int __init
1983 xfs_init_workqueues(void)
1984 {
1985         /*
1986          * The allocation workqueue can be used in memory reclaim situations
1987          * (writepage path), and parallelism is only limited by the number of
1988          * AGs in all the filesystems mounted. Hence use the default large
1989          * max_active value for this workqueue.
1990          */
1991         xfs_alloc_wq = alloc_workqueue("xfsalloc",
1992                         WQ_MEM_RECLAIM|WQ_FREEZABLE, 0);
1993         if (!xfs_alloc_wq)
1994                 return -ENOMEM;
1995
1996         xfs_discard_wq = alloc_workqueue("xfsdiscard", WQ_UNBOUND, 0);
1997         if (!xfs_discard_wq)
1998                 goto out_free_alloc_wq;
1999
2000         return 0;
2001 out_free_alloc_wq:
2002         destroy_workqueue(xfs_alloc_wq);
2003         return -ENOMEM;
2004 }
2005
2006 STATIC void
2007 xfs_destroy_workqueues(void)
2008 {
2009         destroy_workqueue(xfs_discard_wq);
2010         destroy_workqueue(xfs_alloc_wq);
2011 }
2012
2013 STATIC int __init
2014 init_xfs_fs(void)
2015 {
2016         int                     error;
2017
2018         xfs_check_ondisk_structs();
2019
2020         printk(KERN_INFO XFS_VERSION_STRING " with "
2021                          XFS_BUILD_OPTIONS " enabled\n");
2022
2023         xfs_extent_free_init_defer_op();
2024         xfs_rmap_update_init_defer_op();
2025         xfs_refcount_update_init_defer_op();
2026         xfs_bmap_update_init_defer_op();
2027
2028         xfs_dir_startup();
2029
2030         error = xfs_init_zones();
2031         if (error)
2032                 goto out;
2033
2034         error = xfs_init_workqueues();
2035         if (error)
2036                 goto out_destroy_zones;
2037
2038         error = xfs_mru_cache_init();
2039         if (error)
2040                 goto out_destroy_wq;
2041
2042         error = xfs_buf_init();
2043         if (error)
2044                 goto out_mru_cache_uninit;
2045
2046         error = xfs_init_procfs();
2047         if (error)
2048                 goto out_buf_terminate;
2049
2050         error = xfs_sysctl_register();
2051         if (error)
2052                 goto out_cleanup_procfs;
2053
2054         xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2055         if (!xfs_kset) {
2056                 error = -ENOMEM;
2057                 goto out_sysctl_unregister;
2058         }
2059
2060         xfsstats.xs_kobj.kobject.kset = xfs_kset;
2061
2062         xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2063         if (!xfsstats.xs_stats) {
2064                 error = -ENOMEM;
2065                 goto out_kset_unregister;
2066         }
2067
2068         error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2069                                "stats");
2070         if (error)
2071                 goto out_free_stats;
2072
2073 #ifdef DEBUG
2074         xfs_dbg_kobj.kobject.kset = xfs_kset;
2075         error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2076         if (error)
2077                 goto out_remove_stats_kobj;
2078 #endif
2079
2080         error = xfs_qm_init();
2081         if (error)
2082                 goto out_remove_dbg_kobj;
2083
2084         error = register_filesystem(&xfs_fs_type);
2085         if (error)
2086                 goto out_qm_exit;
2087         return 0;
2088
2089  out_qm_exit:
2090         xfs_qm_exit();
2091  out_remove_dbg_kobj:
2092 #ifdef DEBUG
2093         xfs_sysfs_del(&xfs_dbg_kobj);
2094  out_remove_stats_kobj:
2095 #endif
2096         xfs_sysfs_del(&xfsstats.xs_kobj);
2097  out_free_stats:
2098         free_percpu(xfsstats.xs_stats);
2099  out_kset_unregister:
2100         kset_unregister(xfs_kset);
2101  out_sysctl_unregister:
2102         xfs_sysctl_unregister();
2103  out_cleanup_procfs:
2104         xfs_cleanup_procfs();
2105  out_buf_terminate:
2106         xfs_buf_terminate();
2107  out_mru_cache_uninit:
2108         xfs_mru_cache_uninit();
2109  out_destroy_wq:
2110         xfs_destroy_workqueues();
2111  out_destroy_zones:
2112         xfs_destroy_zones();
2113  out:
2114         return error;
2115 }
2116
2117 STATIC void __exit
2118 exit_xfs_fs(void)
2119 {
2120         xfs_qm_exit();
2121         unregister_filesystem(&xfs_fs_type);
2122 #ifdef DEBUG
2123         xfs_sysfs_del(&xfs_dbg_kobj);
2124 #endif
2125         xfs_sysfs_del(&xfsstats.xs_kobj);
2126         free_percpu(xfsstats.xs_stats);
2127         kset_unregister(xfs_kset);
2128         xfs_sysctl_unregister();
2129         xfs_cleanup_procfs();
2130         xfs_buf_terminate();
2131         xfs_mru_cache_uninit();
2132         xfs_destroy_workqueues();
2133         xfs_destroy_zones();
2134         xfs_uuid_table_free();
2135 }
2136
2137 module_init(init_xfs_fs);
2138 module_exit(exit_xfs_fs);
2139
2140 MODULE_AUTHOR("Silicon Graphics, Inc.");
2141 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2142 MODULE_LICENSE("GPL");