b108e7ba81af73568b71ec0b163bf68fbbde8948
[sfrench/cifs-2.6.git] / fs / gfs2 / ops_fstype.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
4  *
5  * This copyrighted material is made available to anyone wishing to use,
6  * modify, copy, or redistribute it subject to the terms and conditions
7  * of the GNU General Public License version 2.
8  */
9
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/blkdev.h>
18 #include <linux/kthread.h>
19 #include <linux/export.h>
20 #include <linux/namei.h>
21 #include <linux/mount.h>
22 #include <linux/gfs2_ondisk.h>
23 #include <linux/quotaops.h>
24 #include <linux/lockdep.h>
25 #include <linux/module.h>
26
27 #include "gfs2.h"
28 #include "incore.h"
29 #include "bmap.h"
30 #include "glock.h"
31 #include "glops.h"
32 #include "inode.h"
33 #include "recovery.h"
34 #include "rgrp.h"
35 #include "super.h"
36 #include "sys.h"
37 #include "util.h"
38 #include "log.h"
39 #include "quota.h"
40 #include "dir.h"
41 #include "meta_io.h"
42 #include "trace_gfs2.h"
43
44 #define DO 0
45 #define UNDO 1
46
47 /**
48  * gfs2_tune_init - Fill a gfs2_tune structure with default values
49  * @gt: tune
50  *
51  */
52
53 static void gfs2_tune_init(struct gfs2_tune *gt)
54 {
55         spin_lock_init(&gt->gt_spin);
56
57         gt->gt_quota_warn_period = 10;
58         gt->gt_quota_scale_num = 1;
59         gt->gt_quota_scale_den = 1;
60         gt->gt_new_files_jdata = 0;
61         gt->gt_max_readahead = BIT(18);
62         gt->gt_complain_secs = 10;
63 }
64
65 static struct gfs2_sbd *init_sbd(struct super_block *sb)
66 {
67         struct gfs2_sbd *sdp;
68         struct address_space *mapping;
69
70         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
71         if (!sdp)
72                 return NULL;
73
74         sb->s_fs_info = sdp;
75         sdp->sd_vfs = sb;
76         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
77         if (!sdp->sd_lkstats) {
78                 kfree(sdp);
79                 return NULL;
80         }
81
82         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
83         gfs2_tune_init(&sdp->sd_tune);
84
85         init_waitqueue_head(&sdp->sd_glock_wait);
86         atomic_set(&sdp->sd_glock_disposal, 0);
87         init_completion(&sdp->sd_locking_init);
88         init_completion(&sdp->sd_wdack);
89         spin_lock_init(&sdp->sd_statfs_spin);
90
91         spin_lock_init(&sdp->sd_rindex_spin);
92         sdp->sd_rindex_tree.rb_node = NULL;
93
94         INIT_LIST_HEAD(&sdp->sd_jindex_list);
95         spin_lock_init(&sdp->sd_jindex_spin);
96         mutex_init(&sdp->sd_jindex_mutex);
97         init_completion(&sdp->sd_journal_ready);
98
99         INIT_LIST_HEAD(&sdp->sd_quota_list);
100         mutex_init(&sdp->sd_quota_mutex);
101         mutex_init(&sdp->sd_quota_sync_mutex);
102         init_waitqueue_head(&sdp->sd_quota_wait);
103         INIT_LIST_HEAD(&sdp->sd_trunc_list);
104         spin_lock_init(&sdp->sd_trunc_lock);
105         spin_lock_init(&sdp->sd_bitmap_lock);
106
107         mapping = &sdp->sd_aspace;
108
109         address_space_init_once(mapping);
110         mapping->a_ops = &gfs2_rgrp_aops;
111         mapping->host = sb->s_bdev->bd_inode;
112         mapping->flags = 0;
113         mapping_set_gfp_mask(mapping, GFP_NOFS);
114         mapping->private_data = NULL;
115         mapping->writeback_index = 0;
116
117         spin_lock_init(&sdp->sd_log_lock);
118         atomic_set(&sdp->sd_log_pinned, 0);
119         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
120         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
121         spin_lock_init(&sdp->sd_ordered_lock);
122
123         init_waitqueue_head(&sdp->sd_log_waitq);
124         init_waitqueue_head(&sdp->sd_logd_waitq);
125         spin_lock_init(&sdp->sd_ail_lock);
126         INIT_LIST_HEAD(&sdp->sd_ail1_list);
127         INIT_LIST_HEAD(&sdp->sd_ail2_list);
128
129         init_rwsem(&sdp->sd_log_flush_lock);
130         atomic_set(&sdp->sd_log_in_flight, 0);
131         atomic_set(&sdp->sd_reserving_log, 0);
132         init_waitqueue_head(&sdp->sd_reserving_log_wait);
133         init_waitqueue_head(&sdp->sd_log_flush_wait);
134         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
135         mutex_init(&sdp->sd_freeze_mutex);
136
137         return sdp;
138 }
139
140
141 /**
142  * gfs2_check_sb - Check superblock
143  * @sdp: the filesystem
144  * @sb: The superblock
145  * @silent: Don't print a message if the check fails
146  *
147  * Checks the version code of the FS is one that we understand how to
148  * read and that the sizes of the various on-disk structures have not
149  * changed.
150  */
151
152 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
153 {
154         struct gfs2_sb_host *sb = &sdp->sd_sb;
155
156         if (sb->sb_magic != GFS2_MAGIC ||
157             sb->sb_type != GFS2_METATYPE_SB) {
158                 if (!silent)
159                         pr_warn("not a GFS2 filesystem\n");
160                 return -EINVAL;
161         }
162
163         /*  If format numbers match exactly, we're done.  */
164
165         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
166             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
167                 return 0;
168
169         fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
170
171         return -EINVAL;
172 }
173
174 static void end_bio_io_page(struct bio *bio)
175 {
176         struct page *page = bio->bi_private;
177
178         if (!bio->bi_error)
179                 SetPageUptodate(page);
180         else
181                 pr_warn("error %d reading superblock\n", bio->bi_error);
182         unlock_page(page);
183 }
184
185 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
186 {
187         struct gfs2_sb_host *sb = &sdp->sd_sb;
188         struct super_block *s = sdp->sd_vfs;
189         const struct gfs2_sb *str = buf;
190
191         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
192         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
193         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
194         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
195         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
196         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
197         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
198         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
199         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
200         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
201         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
202
203         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
204         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
205         memcpy(s->s_uuid, str->sb_uuid, 16);
206 }
207
208 /**
209  * gfs2_read_super - Read the gfs2 super block from disk
210  * @sdp: The GFS2 super block
211  * @sector: The location of the super block
212  * @error: The error code to return
213  *
214  * This uses the bio functions to read the super block from disk
215  * because we want to be 100% sure that we never read cached data.
216  * A super block is read twice only during each GFS2 mount and is
217  * never written to by the filesystem. The first time its read no
218  * locks are held, and the only details which are looked at are those
219  * relating to the locking protocol. Once locking is up and working,
220  * the sb is read again under the lock to establish the location of
221  * the master directory (contains pointers to journals etc) and the
222  * root directory.
223  *
224  * Returns: 0 on success or error
225  */
226
227 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
228 {
229         struct super_block *sb = sdp->sd_vfs;
230         struct gfs2_sb *p;
231         struct page *page;
232         struct bio *bio;
233
234         page = alloc_page(GFP_NOFS);
235         if (unlikely(!page))
236                 return -ENOMEM;
237
238         ClearPageUptodate(page);
239         ClearPageDirty(page);
240         lock_page(page);
241
242         bio = bio_alloc(GFP_NOFS, 1);
243         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
244         bio->bi_bdev = sb->s_bdev;
245         bio_add_page(bio, page, PAGE_SIZE, 0);
246
247         bio->bi_end_io = end_bio_io_page;
248         bio->bi_private = page;
249         bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
250         submit_bio(bio);
251         wait_on_page_locked(page);
252         bio_put(bio);
253         if (!PageUptodate(page)) {
254                 __free_page(page);
255                 return -EIO;
256         }
257         p = kmap(page);
258         gfs2_sb_in(sdp, p);
259         kunmap(page);
260         __free_page(page);
261         return gfs2_check_sb(sdp, silent);
262 }
263
264 /**
265  * gfs2_read_sb - Read super block
266  * @sdp: The GFS2 superblock
267  * @silent: Don't print message if mount fails
268  *
269  */
270
271 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
272 {
273         u32 hash_blocks, ind_blocks, leaf_blocks;
274         u32 tmp_blocks;
275         unsigned int x;
276         int error;
277
278         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
279         if (error) {
280                 if (!silent)
281                         fs_err(sdp, "can't read superblock\n");
282                 return error;
283         }
284
285         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
286                                GFS2_BASIC_BLOCK_SHIFT;
287         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
288         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
289                           sizeof(struct gfs2_dinode)) / sizeof(u64);
290         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
291                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
292         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
293         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
294         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
295         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
296         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
297                                 sizeof(struct gfs2_meta_header)) /
298                                 sizeof(struct gfs2_quota_change);
299         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
300                                      sizeof(struct gfs2_meta_header))
301                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
302
303         /* Compute maximum reservation required to add a entry to a directory */
304
305         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
306                              sdp->sd_jbsize);
307
308         ind_blocks = 0;
309         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
310                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
311                 ind_blocks += tmp_blocks;
312         }
313
314         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
315
316         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
317
318         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
319                                 sizeof(struct gfs2_dinode);
320         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
321         for (x = 2;; x++) {
322                 u64 space, d;
323                 u32 m;
324
325                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
326                 d = space;
327                 m = do_div(d, sdp->sd_inptrs);
328
329                 if (d != sdp->sd_heightsize[x - 1] || m)
330                         break;
331                 sdp->sd_heightsize[x] = space;
332         }
333         sdp->sd_max_height = x;
334         sdp->sd_heightsize[x] = ~0;
335         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
336
337         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
338                                  sizeof(struct gfs2_dinode);
339         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
340         for (x = 2;; x++) {
341                 u64 space, d;
342                 u32 m;
343
344                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
345                 d = space;
346                 m = do_div(d, sdp->sd_inptrs);
347
348                 if (d != sdp->sd_jheightsize[x - 1] || m)
349                         break;
350                 sdp->sd_jheightsize[x] = space;
351         }
352         sdp->sd_max_jheight = x;
353         sdp->sd_jheightsize[x] = ~0;
354         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
355
356         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
357                                       sizeof(struct gfs2_leaf)) /
358                                      GFS2_MIN_DIRENT_SIZE;
359         return 0;
360 }
361
362 static int init_names(struct gfs2_sbd *sdp, int silent)
363 {
364         char *proto, *table;
365         int error = 0;
366
367         proto = sdp->sd_args.ar_lockproto;
368         table = sdp->sd_args.ar_locktable;
369
370         /*  Try to autodetect  */
371
372         if (!proto[0] || !table[0]) {
373                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
374                 if (error)
375                         return error;
376
377                 if (!proto[0])
378                         proto = sdp->sd_sb.sb_lockproto;
379                 if (!table[0])
380                         table = sdp->sd_sb.sb_locktable;
381         }
382
383         if (!table[0])
384                 table = sdp->sd_vfs->s_id;
385
386         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
387         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
388
389         table = sdp->sd_table_name;
390         while ((table = strchr(table, '/')))
391                 *table = '_';
392
393         return error;
394 }
395
396 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
397                         int undo)
398 {
399         int error = 0;
400
401         if (undo)
402                 goto fail_trans;
403
404         error = gfs2_glock_nq_num(sdp,
405                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
406                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
407                                   mount_gh);
408         if (error) {
409                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
410                 goto fail;
411         }
412
413         error = gfs2_glock_nq_num(sdp,
414                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
415                                   LM_ST_SHARED,
416                                   LM_FLAG_NOEXP | GL_EXACT,
417                                   &sdp->sd_live_gh);
418         if (error) {
419                 fs_err(sdp, "can't acquire live glock: %d\n", error);
420                 goto fail_mount;
421         }
422
423         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
424                                CREATE, &sdp->sd_rename_gl);
425         if (error) {
426                 fs_err(sdp, "can't create rename glock: %d\n", error);
427                 goto fail_live;
428         }
429
430         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
431                                CREATE, &sdp->sd_freeze_gl);
432         if (error) {
433                 fs_err(sdp, "can't create transaction glock: %d\n", error);
434                 goto fail_rename;
435         }
436
437         return 0;
438
439 fail_trans:
440         gfs2_glock_put(sdp->sd_freeze_gl);
441 fail_rename:
442         gfs2_glock_put(sdp->sd_rename_gl);
443 fail_live:
444         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
445 fail_mount:
446         gfs2_glock_dq_uninit(mount_gh);
447 fail:
448         return error;
449 }
450
451 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
452                             u64 no_addr, const char *name)
453 {
454         struct gfs2_sbd *sdp = sb->s_fs_info;
455         struct dentry *dentry;
456         struct inode *inode;
457
458         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
459                                   GFS2_BLKST_FREE /* ignore */);
460         if (IS_ERR(inode)) {
461                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
462                 return PTR_ERR(inode);
463         }
464         dentry = d_make_root(inode);
465         if (!dentry) {
466                 fs_err(sdp, "can't alloc %s dentry\n", name);
467                 return -ENOMEM;
468         }
469         *dptr = dentry;
470         return 0;
471 }
472
473 static int init_sb(struct gfs2_sbd *sdp, int silent)
474 {
475         struct super_block *sb = sdp->sd_vfs;
476         struct gfs2_holder sb_gh;
477         u64 no_addr;
478         int ret;
479
480         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
481                                 LM_ST_SHARED, 0, &sb_gh);
482         if (ret) {
483                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
484                 return ret;
485         }
486
487         ret = gfs2_read_sb(sdp, silent);
488         if (ret) {
489                 fs_err(sdp, "can't read superblock: %d\n", ret);
490                 goto out;
491         }
492
493         /* Set up the buffer cache and SB for real */
494         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
495                 ret = -EINVAL;
496                 fs_err(sdp, "FS block size (%u) is too small for device "
497                        "block size (%u)\n",
498                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
499                 goto out;
500         }
501         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
502                 ret = -EINVAL;
503                 fs_err(sdp, "FS block size (%u) is too big for machine "
504                        "page size (%u)\n",
505                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
506                 goto out;
507         }
508         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
509
510         /* Get the root inode */
511         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
512         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
513         if (ret)
514                 goto out;
515
516         /* Get the master inode */
517         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
518         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
519         if (ret) {
520                 dput(sdp->sd_root_dir);
521                 goto out;
522         }
523         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
524 out:
525         gfs2_glock_dq_uninit(&sb_gh);
526         return ret;
527 }
528
529 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
530 {
531         char *message = "FIRSTMOUNT=Done";
532         char *envp[] = { message, NULL };
533
534         fs_info(sdp, "first mount done, others may mount\n");
535
536         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
537                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
538
539         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
540 }
541
542 /**
543  * gfs2_jindex_hold - Grab a lock on the jindex
544  * @sdp: The GFS2 superblock
545  * @ji_gh: the holder for the jindex glock
546  *
547  * Returns: errno
548  */
549
550 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
551 {
552         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
553         struct qstr name;
554         char buf[20];
555         struct gfs2_jdesc *jd;
556         int error;
557
558         name.name = buf;
559
560         mutex_lock(&sdp->sd_jindex_mutex);
561
562         for (;;) {
563                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
564                 if (error)
565                         break;
566
567                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
568                 name.hash = gfs2_disk_hash(name.name, name.len);
569
570                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
571                 if (error == -ENOENT) {
572                         error = 0;
573                         break;
574                 }
575
576                 gfs2_glock_dq_uninit(ji_gh);
577
578                 if (error)
579                         break;
580
581                 error = -ENOMEM;
582                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
583                 if (!jd)
584                         break;
585
586                 INIT_LIST_HEAD(&jd->extent_list);
587                 INIT_LIST_HEAD(&jd->jd_revoke_list);
588
589                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
590                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
591                 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
592                         if (!jd->jd_inode)
593                                 error = -ENOENT;
594                         else
595                                 error = PTR_ERR(jd->jd_inode);
596                         kfree(jd);
597                         break;
598                 }
599
600                 spin_lock(&sdp->sd_jindex_spin);
601                 jd->jd_jid = sdp->sd_journals++;
602                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
603                 spin_unlock(&sdp->sd_jindex_spin);
604         }
605
606         mutex_unlock(&sdp->sd_jindex_mutex);
607
608         return error;
609 }
610
611 /**
612  * check_journal_clean - Make sure a journal is clean for a spectator mount
613  * @sdp: The GFS2 superblock
614  * @jd: The journal descriptor
615  *
616  * Returns: 0 if the journal is clean or locked, else an error
617  */
618 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
619 {
620         int error;
621         struct gfs2_holder j_gh;
622         struct gfs2_log_header_host head;
623         struct gfs2_inode *ip;
624
625         ip = GFS2_I(jd->jd_inode);
626         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
627                                    GL_EXACT | GL_NOCACHE, &j_gh);
628         if (error) {
629                 fs_err(sdp, "Error locking journal for spectator mount.\n");
630                 return -EPERM;
631         }
632         error = gfs2_jdesc_check(jd);
633         if (error) {
634                 fs_err(sdp, "Error checking journal for spectator mount.\n");
635                 goto out_unlock;
636         }
637         error = gfs2_find_jhead(jd, &head);
638         if (error) {
639                 fs_err(sdp, "Error parsing journal for spectator mount.\n");
640                 goto out_unlock;
641         }
642         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
643                 error = -EPERM;
644                 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
645                        "must not be a spectator.\n", jd->jd_jid);
646         }
647
648 out_unlock:
649         gfs2_glock_dq_uninit(&j_gh);
650         return error;
651 }
652
653 static int init_journal(struct gfs2_sbd *sdp, int undo)
654 {
655         struct inode *master = d_inode(sdp->sd_master_dir);
656         struct gfs2_holder ji_gh;
657         struct gfs2_inode *ip;
658         int jindex = 1;
659         int error = 0;
660
661         if (undo) {
662                 jindex = 0;
663                 goto fail_jinode_gh;
664         }
665
666         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
667         if (IS_ERR(sdp->sd_jindex)) {
668                 fs_err(sdp, "can't lookup journal index: %d\n", error);
669                 return PTR_ERR(sdp->sd_jindex);
670         }
671
672         /* Load in the journal index special file */
673
674         error = gfs2_jindex_hold(sdp, &ji_gh);
675         if (error) {
676                 fs_err(sdp, "can't read journal index: %d\n", error);
677                 goto fail;
678         }
679
680         error = -EUSERS;
681         if (!gfs2_jindex_size(sdp)) {
682                 fs_err(sdp, "no journals!\n");
683                 goto fail_jindex;
684         }
685
686         atomic_set(&sdp->sd_log_blks_needed, 0);
687         if (sdp->sd_args.ar_spectator) {
688                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
689                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
690                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
691                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
692         } else {
693                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
694                         fs_err(sdp, "can't mount journal #%u\n",
695                                sdp->sd_lockstruct.ls_jid);
696                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
697                                gfs2_jindex_size(sdp),
698                                gfs2_jindex_size(sdp) - 1);
699                         goto fail_jindex;
700                 }
701                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
702
703                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
704                                           &gfs2_journal_glops,
705                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
706                                           &sdp->sd_journal_gh);
707                 if (error) {
708                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
709                         goto fail_jindex;
710                 }
711
712                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
713                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
714                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
715                                            &sdp->sd_jinode_gh);
716                 if (error) {
717                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
718                                error);
719                         goto fail_journal_gh;
720                 }
721
722                 error = gfs2_jdesc_check(sdp->sd_jdesc);
723                 if (error) {
724                         fs_err(sdp, "my journal (%u) is bad: %d\n",
725                                sdp->sd_jdesc->jd_jid, error);
726                         goto fail_jinode_gh;
727                 }
728                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
729                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
730                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
731
732                 /* Map the extents for this journal's blocks */
733                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
734         }
735         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
736
737         if (sdp->sd_lockstruct.ls_first) {
738                 unsigned int x;
739                 for (x = 0; x < sdp->sd_journals; x++) {
740                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
741
742                         if (sdp->sd_args.ar_spectator) {
743                                 error = check_journal_clean(sdp, jd);
744                                 if (error)
745                                         goto fail_jinode_gh;
746                                 continue;
747                         }
748                         error = gfs2_recover_journal(jd, true);
749                         if (error) {
750                                 fs_err(sdp, "error recovering journal %u: %d\n",
751                                        x, error);
752                                 goto fail_jinode_gh;
753                         }
754                 }
755
756                 gfs2_others_may_mount(sdp);
757         } else if (!sdp->sd_args.ar_spectator) {
758                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
759                 if (error) {
760                         fs_err(sdp, "error recovering my journal: %d\n", error);
761                         goto fail_jinode_gh;
762                 }
763         }
764
765         sdp->sd_log_idle = 1;
766         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
767         gfs2_glock_dq_uninit(&ji_gh);
768         jindex = 0;
769         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
770         return 0;
771
772 fail_jinode_gh:
773         if (!sdp->sd_args.ar_spectator)
774                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
775 fail_journal_gh:
776         if (!sdp->sd_args.ar_spectator)
777                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
778 fail_jindex:
779         gfs2_jindex_free(sdp);
780         if (jindex)
781                 gfs2_glock_dq_uninit(&ji_gh);
782 fail:
783         iput(sdp->sd_jindex);
784         return error;
785 }
786
787 static struct lock_class_key gfs2_quota_imutex_key;
788
789 static int init_inodes(struct gfs2_sbd *sdp, int undo)
790 {
791         int error = 0;
792         struct inode *master = d_inode(sdp->sd_master_dir);
793
794         if (undo)
795                 goto fail_qinode;
796
797         error = init_journal(sdp, undo);
798         complete_all(&sdp->sd_journal_ready);
799         if (error)
800                 goto fail;
801
802         /* Read in the master statfs inode */
803         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
804         if (IS_ERR(sdp->sd_statfs_inode)) {
805                 error = PTR_ERR(sdp->sd_statfs_inode);
806                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
807                 goto fail_journal;
808         }
809
810         /* Read in the resource index inode */
811         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
812         if (IS_ERR(sdp->sd_rindex)) {
813                 error = PTR_ERR(sdp->sd_rindex);
814                 fs_err(sdp, "can't get resource index inode: %d\n", error);
815                 goto fail_statfs;
816         }
817         sdp->sd_rindex_uptodate = 0;
818
819         /* Read in the quota inode */
820         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
821         if (IS_ERR(sdp->sd_quota_inode)) {
822                 error = PTR_ERR(sdp->sd_quota_inode);
823                 fs_err(sdp, "can't get quota file inode: %d\n", error);
824                 goto fail_rindex;
825         }
826         /*
827          * i_mutex on quota files is special. Since this inode is hidden system
828          * file, we are safe to define locking ourselves.
829          */
830         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
831                           &gfs2_quota_imutex_key);
832
833         error = gfs2_rindex_update(sdp);
834         if (error)
835                 goto fail_qinode;
836
837         return 0;
838
839 fail_qinode:
840         iput(sdp->sd_quota_inode);
841 fail_rindex:
842         gfs2_clear_rgrpd(sdp);
843         iput(sdp->sd_rindex);
844 fail_statfs:
845         iput(sdp->sd_statfs_inode);
846 fail_journal:
847         init_journal(sdp, UNDO);
848 fail:
849         return error;
850 }
851
852 static int init_per_node(struct gfs2_sbd *sdp, int undo)
853 {
854         struct inode *pn = NULL;
855         char buf[30];
856         int error = 0;
857         struct gfs2_inode *ip;
858         struct inode *master = d_inode(sdp->sd_master_dir);
859
860         if (sdp->sd_args.ar_spectator)
861                 return 0;
862
863         if (undo)
864                 goto fail_qc_gh;
865
866         pn = gfs2_lookup_simple(master, "per_node");
867         if (IS_ERR(pn)) {
868                 error = PTR_ERR(pn);
869                 fs_err(sdp, "can't find per_node directory: %d\n", error);
870                 return error;
871         }
872
873         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
874         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
875         if (IS_ERR(sdp->sd_sc_inode)) {
876                 error = PTR_ERR(sdp->sd_sc_inode);
877                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
878                 goto fail;
879         }
880
881         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
882         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
883         if (IS_ERR(sdp->sd_qc_inode)) {
884                 error = PTR_ERR(sdp->sd_qc_inode);
885                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
886                 goto fail_ut_i;
887         }
888
889         iput(pn);
890         pn = NULL;
891
892         ip = GFS2_I(sdp->sd_sc_inode);
893         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
894                                    &sdp->sd_sc_gh);
895         if (error) {
896                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
897                 goto fail_qc_i;
898         }
899
900         ip = GFS2_I(sdp->sd_qc_inode);
901         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
902                                    &sdp->sd_qc_gh);
903         if (error) {
904                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
905                 goto fail_ut_gh;
906         }
907
908         return 0;
909
910 fail_qc_gh:
911         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
912 fail_ut_gh:
913         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
914 fail_qc_i:
915         iput(sdp->sd_qc_inode);
916 fail_ut_i:
917         iput(sdp->sd_sc_inode);
918 fail:
919         iput(pn);
920         return error;
921 }
922
923 static const match_table_t nolock_tokens = {
924         { Opt_jid, "jid=%d\n", },
925         { Opt_err, NULL },
926 };
927
928 static const struct lm_lockops nolock_ops = {
929         .lm_proto_name = "lock_nolock",
930         .lm_put_lock = gfs2_glock_free,
931         .lm_tokens = &nolock_tokens,
932 };
933
934 /**
935  * gfs2_lm_mount - mount a locking protocol
936  * @sdp: the filesystem
937  * @args: mount arguments
938  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
939  *
940  * Returns: errno
941  */
942
943 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
944 {
945         const struct lm_lockops *lm;
946         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
947         struct gfs2_args *args = &sdp->sd_args;
948         const char *proto = sdp->sd_proto_name;
949         const char *table = sdp->sd_table_name;
950         char *o, *options;
951         int ret;
952
953         if (!strcmp("lock_nolock", proto)) {
954                 lm = &nolock_ops;
955                 sdp->sd_args.ar_localflocks = 1;
956 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
957         } else if (!strcmp("lock_dlm", proto)) {
958                 lm = &gfs2_dlm_ops;
959 #endif
960         } else {
961                 pr_info("can't find protocol %s\n", proto);
962                 return -ENOENT;
963         }
964
965         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
966
967         ls->ls_ops = lm;
968         ls->ls_first = 1;
969
970         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
971                 substring_t tmp[MAX_OPT_ARGS];
972                 int token, option;
973
974                 if (!o || !*o)
975                         continue;
976
977                 token = match_token(o, *lm->lm_tokens, tmp);
978                 switch (token) {
979                 case Opt_jid:
980                         ret = match_int(&tmp[0], &option);
981                         if (ret || option < 0) 
982                                 goto hostdata_error;
983                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
984                                 ls->ls_jid = option;
985                         break;
986                 case Opt_id:
987                 case Opt_nodir:
988                         /* Obsolete, but left for backward compat purposes */
989                         break;
990                 case Opt_first:
991                         ret = match_int(&tmp[0], &option);
992                         if (ret || (option != 0 && option != 1))
993                                 goto hostdata_error;
994                         ls->ls_first = option;
995                         break;
996                 case Opt_err:
997                 default:
998 hostdata_error:
999                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1000                         return -EINVAL;
1001                 }
1002         }
1003
1004         if (lm->lm_mount == NULL) {
1005                 fs_info(sdp, "Now mounting FS...\n");
1006                 complete_all(&sdp->sd_locking_init);
1007                 return 0;
1008         }
1009         ret = lm->lm_mount(sdp, table);
1010         if (ret == 0)
1011                 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1012         complete_all(&sdp->sd_locking_init);
1013         return ret;
1014 }
1015
1016 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1017 {
1018         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1019         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1020             lm->lm_unmount)
1021                 lm->lm_unmount(sdp);
1022 }
1023
1024 static int wait_on_journal(struct gfs2_sbd *sdp)
1025 {
1026         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1027                 return 0;
1028
1029         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1030                 ? -EINTR : 0;
1031 }
1032
1033 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1034 {
1035         struct super_block *sb = sdp->sd_vfs;
1036         char ro[20];
1037         char spectator[20];
1038         char *envp[] = { ro, spectator, NULL };
1039         sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1040         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1041         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1042 }
1043
1044 /**
1045  * fill_super - Read in superblock
1046  * @sb: The VFS superblock
1047  * @data: Mount options
1048  * @silent: Don't complain if it's not a GFS2 filesystem
1049  *
1050  * Returns: errno
1051  */
1052
1053 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1054 {
1055         struct gfs2_sbd *sdp;
1056         struct gfs2_holder mount_gh;
1057         int error;
1058
1059         sdp = init_sbd(sb);
1060         if (!sdp) {
1061                 pr_warn("can't alloc struct gfs2_sbd\n");
1062                 return -ENOMEM;
1063         }
1064         sdp->sd_args = *args;
1065
1066         if (sdp->sd_args.ar_spectator) {
1067                 sb->s_flags |= MS_RDONLY;
1068                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1069         }
1070         if (sdp->sd_args.ar_posix_acl)
1071                 sb->s_flags |= MS_POSIXACL;
1072         if (sdp->sd_args.ar_nobarrier)
1073                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1074
1075         sb->s_flags |= MS_NOSEC;
1076         sb->s_magic = GFS2_MAGIC;
1077         sb->s_op = &gfs2_super_ops;
1078         sb->s_d_op = &gfs2_dops;
1079         sb->s_export_op = &gfs2_export_ops;
1080         sb->s_xattr = gfs2_xattr_handlers;
1081         sb->s_qcop = &gfs2_quotactl_ops;
1082         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1083         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1084         sb->s_time_gran = 1;
1085         sb->s_maxbytes = MAX_LFS_FILESIZE;
1086
1087         /* Set up the buffer cache and fill in some fake block size values
1088            to allow us to read-in the on-disk superblock. */
1089         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1090         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1091         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1092                                GFS2_BASIC_BLOCK_SHIFT;
1093         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1094
1095         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1096         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1097         if (sdp->sd_args.ar_statfs_quantum) {
1098                 sdp->sd_tune.gt_statfs_slow = 0;
1099                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1100         } else {
1101                 sdp->sd_tune.gt_statfs_slow = 1;
1102                 sdp->sd_tune.gt_statfs_quantum = 30;
1103         }
1104
1105         error = init_names(sdp, silent);
1106         if (error) {
1107                 /* In this case, we haven't initialized sysfs, so we have to
1108                    manually free the sdp. */
1109                 free_percpu(sdp->sd_lkstats);
1110                 kfree(sdp);
1111                 sb->s_fs_info = NULL;
1112                 return error;
1113         }
1114
1115         snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1116
1117         error = gfs2_sys_fs_add(sdp);
1118         /*
1119          * If we hit an error here, gfs2_sys_fs_add will have called function
1120          * kobject_put which causes the sysfs usage count to go to zero, which
1121          * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1122          * Subsequent error paths here will call gfs2_sys_fs_del, which also
1123          * kobject_put to free sdp.
1124          */
1125         if (error)
1126                 return error;
1127
1128         gfs2_create_debugfs_file(sdp);
1129
1130         error = gfs2_lm_mount(sdp, silent);
1131         if (error)
1132                 goto fail_debug;
1133
1134         error = init_locking(sdp, &mount_gh, DO);
1135         if (error)
1136                 goto fail_lm;
1137
1138         error = init_sb(sdp, silent);
1139         if (error)
1140                 goto fail_locking;
1141
1142         error = wait_on_journal(sdp);
1143         if (error)
1144                 goto fail_sb;
1145
1146         /*
1147          * If user space has failed to join the cluster or some similar
1148          * failure has occurred, then the journal id will contain a
1149          * negative (error) number. This will then be returned to the
1150          * caller (of the mount syscall). We do this even for spectator
1151          * mounts (which just write a jid of 0 to indicate "ok" even though
1152          * the jid is unused in the spectator case)
1153          */
1154         if (sdp->sd_lockstruct.ls_jid < 0) {
1155                 error = sdp->sd_lockstruct.ls_jid;
1156                 sdp->sd_lockstruct.ls_jid = 0;
1157                 goto fail_sb;
1158         }
1159
1160         if (sdp->sd_args.ar_spectator)
1161                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1162                          sdp->sd_table_name);
1163         else
1164                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1165                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1166
1167         error = init_inodes(sdp, DO);
1168         if (error)
1169                 goto fail_sb;
1170
1171         error = init_per_node(sdp, DO);
1172         if (error)
1173                 goto fail_inodes;
1174
1175         error = gfs2_statfs_init(sdp);
1176         if (error) {
1177                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1178                 goto fail_per_node;
1179         }
1180
1181         if (!(sb->s_flags & MS_RDONLY)) {
1182                 error = gfs2_make_fs_rw(sdp);
1183                 if (error) {
1184                         fs_err(sdp, "can't make FS RW: %d\n", error);
1185                         goto fail_per_node;
1186                 }
1187         }
1188
1189         gfs2_glock_dq_uninit(&mount_gh);
1190         gfs2_online_uevent(sdp);
1191         return 0;
1192
1193 fail_per_node:
1194         init_per_node(sdp, UNDO);
1195 fail_inodes:
1196         init_inodes(sdp, UNDO);
1197 fail_sb:
1198         if (sdp->sd_root_dir)
1199                 dput(sdp->sd_root_dir);
1200         if (sdp->sd_master_dir)
1201                 dput(sdp->sd_master_dir);
1202         if (sb->s_root)
1203                 dput(sb->s_root);
1204         sb->s_root = NULL;
1205 fail_locking:
1206         init_locking(sdp, &mount_gh, UNDO);
1207 fail_lm:
1208         complete_all(&sdp->sd_journal_ready);
1209         gfs2_gl_hash_clear(sdp);
1210         gfs2_lm_unmount(sdp);
1211 fail_debug:
1212         gfs2_delete_debugfs_file(sdp);
1213         free_percpu(sdp->sd_lkstats);
1214         /* gfs2_sys_fs_del must be the last thing we do, since it causes
1215          * sysfs to call function gfs2_sbd_release, which frees sdp. */
1216         gfs2_sys_fs_del(sdp);
1217         sb->s_fs_info = NULL;
1218         return error;
1219 }
1220
1221 static int set_gfs2_super(struct super_block *s, void *data)
1222 {
1223         s->s_bdev = data;
1224         s->s_dev = s->s_bdev->bd_dev;
1225
1226         /*
1227          * We set the bdi here to the queue backing, file systems can
1228          * overwrite this in ->fill_super()
1229          */
1230         s->s_bdi = bdev_get_queue(s->s_bdev)->backing_dev_info;
1231         return 0;
1232 }
1233
1234 static int test_gfs2_super(struct super_block *s, void *ptr)
1235 {
1236         struct block_device *bdev = ptr;
1237         return (bdev == s->s_bdev);
1238 }
1239
1240 /**
1241  * gfs2_mount - Get the GFS2 superblock
1242  * @fs_type: The GFS2 filesystem type
1243  * @flags: Mount flags
1244  * @dev_name: The name of the device
1245  * @data: The mount arguments
1246  *
1247  * Q. Why not use get_sb_bdev() ?
1248  * A. We need to select one of two root directories to mount, independent
1249  *    of whether this is the initial, or subsequent, mount of this sb
1250  *
1251  * Returns: 0 or -ve on error
1252  */
1253
1254 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1255                        const char *dev_name, void *data)
1256 {
1257         struct block_device *bdev;
1258         struct super_block *s;
1259         fmode_t mode = FMODE_READ | FMODE_EXCL;
1260         int error;
1261         struct gfs2_args args;
1262         struct gfs2_sbd *sdp;
1263
1264         if (!(flags & MS_RDONLY))
1265                 mode |= FMODE_WRITE;
1266
1267         bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1268         if (IS_ERR(bdev))
1269                 return ERR_CAST(bdev);
1270
1271         /*
1272          * once the super is inserted into the list by sget, s_umount
1273          * will protect the lockfs code from trying to start a snapshot
1274          * while we are mounting
1275          */
1276         mutex_lock(&bdev->bd_fsfreeze_mutex);
1277         if (bdev->bd_fsfreeze_count > 0) {
1278                 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1279                 error = -EBUSY;
1280                 goto error_bdev;
1281         }
1282         s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1283         mutex_unlock(&bdev->bd_fsfreeze_mutex);
1284         error = PTR_ERR(s);
1285         if (IS_ERR(s))
1286                 goto error_bdev;
1287
1288         if (s->s_root) {
1289                 /*
1290                  * s_umount nests inside bd_mutex during
1291                  * __invalidate_device().  blkdev_put() acquires
1292                  * bd_mutex and can't be called under s_umount.  Drop
1293                  * s_umount temporarily.  This is safe as we're
1294                  * holding an active reference.
1295                  */
1296                 up_write(&s->s_umount);
1297                 blkdev_put(bdev, mode);
1298                 down_write(&s->s_umount);
1299         } else {
1300                 /* s_mode must be set before deactivate_locked_super calls */
1301                 s->s_mode = mode;
1302         }
1303
1304         memset(&args, 0, sizeof(args));
1305         args.ar_quota = GFS2_QUOTA_DEFAULT;
1306         args.ar_data = GFS2_DATA_DEFAULT;
1307         args.ar_commit = 30;
1308         args.ar_statfs_quantum = 30;
1309         args.ar_quota_quantum = 60;
1310         args.ar_errors = GFS2_ERRORS_DEFAULT;
1311
1312         error = gfs2_mount_args(&args, data);
1313         if (error) {
1314                 pr_warn("can't parse mount arguments\n");
1315                 goto error_super;
1316         }
1317
1318         if (s->s_root) {
1319                 error = -EBUSY;
1320                 if ((flags ^ s->s_flags) & MS_RDONLY)
1321                         goto error_super;
1322         } else {
1323                 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1324                 sb_set_blocksize(s, block_size(bdev));
1325                 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1326                 if (error)
1327                         goto error_super;
1328                 s->s_flags |= MS_ACTIVE;
1329                 bdev->bd_super = s;
1330         }
1331
1332         sdp = s->s_fs_info;
1333         if (args.ar_meta)
1334                 return dget(sdp->sd_master_dir);
1335         else
1336                 return dget(sdp->sd_root_dir);
1337
1338 error_super:
1339         deactivate_locked_super(s);
1340         return ERR_PTR(error);
1341 error_bdev:
1342         blkdev_put(bdev, mode);
1343         return ERR_PTR(error);
1344 }
1345
1346 static int set_meta_super(struct super_block *s, void *ptr)
1347 {
1348         return -EINVAL;
1349 }
1350
1351 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1352                         int flags, const char *dev_name, void *data)
1353 {
1354         struct super_block *s;
1355         struct gfs2_sbd *sdp;
1356         struct path path;
1357         int error;
1358
1359         error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1360         if (error) {
1361                 pr_warn("path_lookup on %s returned error %d\n",
1362                         dev_name, error);
1363                 return ERR_PTR(error);
1364         }
1365         s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1366                  path.dentry->d_sb->s_bdev);
1367         path_put(&path);
1368         if (IS_ERR(s)) {
1369                 pr_warn("gfs2 mount does not exist\n");
1370                 return ERR_CAST(s);
1371         }
1372         if ((flags ^ s->s_flags) & MS_RDONLY) {
1373                 deactivate_locked_super(s);
1374                 return ERR_PTR(-EBUSY);
1375         }
1376         sdp = s->s_fs_info;
1377         return dget(sdp->sd_master_dir);
1378 }
1379
1380 static void gfs2_kill_sb(struct super_block *sb)
1381 {
1382         struct gfs2_sbd *sdp = sb->s_fs_info;
1383
1384         if (sdp == NULL) {
1385                 kill_block_super(sb);
1386                 return;
1387         }
1388
1389         gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1390         dput(sdp->sd_root_dir);
1391         dput(sdp->sd_master_dir);
1392         sdp->sd_root_dir = NULL;
1393         sdp->sd_master_dir = NULL;
1394         shrink_dcache_sb(sb);
1395         gfs2_delete_debugfs_file(sdp);
1396         free_percpu(sdp->sd_lkstats);
1397         kill_block_super(sb);
1398 }
1399
1400 struct file_system_type gfs2_fs_type = {
1401         .name = "gfs2",
1402         .fs_flags = FS_REQUIRES_DEV,
1403         .mount = gfs2_mount,
1404         .kill_sb = gfs2_kill_sb,
1405         .owner = THIS_MODULE,
1406 };
1407 MODULE_ALIAS_FS("gfs2");
1408
1409 struct file_system_type gfs2meta_fs_type = {
1410         .name = "gfs2meta",
1411         .fs_flags = FS_REQUIRES_DEV,
1412         .mount = gfs2_mount_meta,
1413         .owner = THIS_MODULE,
1414 };
1415 MODULE_ALIAS_FS("gfs2meta");