gfs2: don't withdraw if init_threads() got interrupted
[sfrench/cifs-2.6.git] / fs / gfs2 / ops_fstype.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
5  */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
24 #include <linux/fs_parser.h>
25
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "recovery.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "sys.h"
36 #include "util.h"
37 #include "log.h"
38 #include "quota.h"
39 #include "dir.h"
40 #include "meta_io.h"
41 #include "trace_gfs2.h"
42 #include "lops.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 void free_sbd(struct gfs2_sbd *sdp)
66 {
67         if (sdp->sd_lkstats)
68                 free_percpu(sdp->sd_lkstats);
69         kfree(sdp);
70 }
71
72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
73 {
74         struct gfs2_sbd *sdp;
75         struct address_space *mapping;
76
77         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
78         if (!sdp)
79                 return NULL;
80
81         sdp->sd_vfs = sb;
82         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
83         if (!sdp->sd_lkstats)
84                 goto fail;
85         sb->s_fs_info = sdp;
86
87         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
88         gfs2_tune_init(&sdp->sd_tune);
89
90         init_waitqueue_head(&sdp->sd_kill_wait);
91         init_waitqueue_head(&sdp->sd_async_glock_wait);
92         atomic_set(&sdp->sd_glock_disposal, 0);
93         init_completion(&sdp->sd_locking_init);
94         init_completion(&sdp->sd_wdack);
95         spin_lock_init(&sdp->sd_statfs_spin);
96
97         spin_lock_init(&sdp->sd_rindex_spin);
98         sdp->sd_rindex_tree.rb_node = NULL;
99
100         INIT_LIST_HEAD(&sdp->sd_jindex_list);
101         spin_lock_init(&sdp->sd_jindex_spin);
102         mutex_init(&sdp->sd_jindex_mutex);
103         init_completion(&sdp->sd_journal_ready);
104
105         INIT_LIST_HEAD(&sdp->sd_quota_list);
106         mutex_init(&sdp->sd_quota_mutex);
107         mutex_init(&sdp->sd_quota_sync_mutex);
108         init_waitqueue_head(&sdp->sd_quota_wait);
109         spin_lock_init(&sdp->sd_bitmap_lock);
110
111         INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
112
113         mapping = &sdp->sd_aspace;
114
115         address_space_init_once(mapping);
116         mapping->a_ops = &gfs2_rgrp_aops;
117         mapping->host = sb->s_bdev->bd_inode;
118         mapping->flags = 0;
119         mapping_set_gfp_mask(mapping, GFP_NOFS);
120         mapping->private_data = NULL;
121         mapping->writeback_index = 0;
122
123         spin_lock_init(&sdp->sd_log_lock);
124         atomic_set(&sdp->sd_log_pinned, 0);
125         INIT_LIST_HEAD(&sdp->sd_log_revokes);
126         INIT_LIST_HEAD(&sdp->sd_log_ordered);
127         spin_lock_init(&sdp->sd_ordered_lock);
128
129         init_waitqueue_head(&sdp->sd_log_waitq);
130         init_waitqueue_head(&sdp->sd_logd_waitq);
131         spin_lock_init(&sdp->sd_ail_lock);
132         INIT_LIST_HEAD(&sdp->sd_ail1_list);
133         INIT_LIST_HEAD(&sdp->sd_ail2_list);
134
135         init_rwsem(&sdp->sd_log_flush_lock);
136         atomic_set(&sdp->sd_log_in_flight, 0);
137         init_waitqueue_head(&sdp->sd_log_flush_wait);
138         mutex_init(&sdp->sd_freeze_mutex);
139
140         return sdp;
141
142 fail:
143         free_sbd(sdp);
144         return NULL;
145 }
146
147 /**
148  * gfs2_check_sb - Check superblock
149  * @sdp: the filesystem
150  * @silent: Don't print a message if the check fails
151  *
152  * Checks the version code of the FS is one that we understand how to
153  * read and that the sizes of the various on-disk structures have not
154  * changed.
155  */
156
157 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
158 {
159         struct gfs2_sb_host *sb = &sdp->sd_sb;
160
161         if (sb->sb_magic != GFS2_MAGIC ||
162             sb->sb_type != GFS2_METATYPE_SB) {
163                 if (!silent)
164                         pr_warn("not a GFS2 filesystem\n");
165                 return -EINVAL;
166         }
167
168         if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
169             sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
170             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
171                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
172                 return -EINVAL;
173         }
174
175         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
176             (sb->sb_bsize & (sb->sb_bsize - 1))) {
177                 pr_warn("Invalid block size\n");
178                 return -EINVAL;
179         }
180         if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
181                 pr_warn("Invalid block size shift\n");
182                 return -EINVAL;
183         }
184         return 0;
185 }
186
187 static void end_bio_io_page(struct bio *bio)
188 {
189         struct page *page = bio->bi_private;
190
191         if (!bio->bi_status)
192                 SetPageUptodate(page);
193         else
194                 pr_warn("error %d reading superblock\n", bio->bi_status);
195         unlock_page(page);
196 }
197
198 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
199 {
200         struct gfs2_sb_host *sb = &sdp->sd_sb;
201         struct super_block *s = sdp->sd_vfs;
202         const struct gfs2_sb *str = buf;
203
204         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
205         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
206         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
207         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
208         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
209         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
210         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
211         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
212         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
213         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
214
215         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
216         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
217         memcpy(&s->s_uuid, str->sb_uuid, 16);
218 }
219
220 /**
221  * gfs2_read_super - Read the gfs2 super block from disk
222  * @sdp: The GFS2 super block
223  * @sector: The location of the super block
224  * @silent: Don't print a message if the check fails
225  *
226  * This uses the bio functions to read the super block from disk
227  * because we want to be 100% sure that we never read cached data.
228  * A super block is read twice only during each GFS2 mount and is
229  * never written to by the filesystem. The first time its read no
230  * locks are held, and the only details which are looked at are those
231  * relating to the locking protocol. Once locking is up and working,
232  * the sb is read again under the lock to establish the location of
233  * the master directory (contains pointers to journals etc) and the
234  * root directory.
235  *
236  * Returns: 0 on success or error
237  */
238
239 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
240 {
241         struct super_block *sb = sdp->sd_vfs;
242         struct gfs2_sb *p;
243         struct page *page;
244         struct bio *bio;
245
246         page = alloc_page(GFP_NOFS);
247         if (unlikely(!page))
248                 return -ENOMEM;
249
250         ClearPageUptodate(page);
251         ClearPageDirty(page);
252         lock_page(page);
253
254         bio = bio_alloc(sb->s_bdev, 1, REQ_OP_READ | REQ_META, GFP_NOFS);
255         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
256         __bio_add_page(bio, page, PAGE_SIZE, 0);
257
258         bio->bi_end_io = end_bio_io_page;
259         bio->bi_private = page;
260         submit_bio(bio);
261         wait_on_page_locked(page);
262         bio_put(bio);
263         if (!PageUptodate(page)) {
264                 __free_page(page);
265                 return -EIO;
266         }
267         p = kmap(page);
268         gfs2_sb_in(sdp, p);
269         kunmap(page);
270         __free_page(page);
271         return gfs2_check_sb(sdp, silent);
272 }
273
274 /**
275  * gfs2_read_sb - Read super block
276  * @sdp: The GFS2 superblock
277  * @silent: Don't print message if mount fails
278  *
279  */
280
281 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
282 {
283         u32 hash_blocks, ind_blocks, leaf_blocks;
284         u32 tmp_blocks;
285         unsigned int x;
286         int error;
287
288         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
289         if (error) {
290                 if (!silent)
291                         fs_err(sdp, "can't read superblock\n");
292                 return error;
293         }
294
295         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
296         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
297         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
298                           sizeof(struct gfs2_dinode)) / sizeof(u64);
299         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
300                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
301         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
302                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
303         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
304         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
305         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
306         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
307         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
308                                 sizeof(struct gfs2_meta_header)) /
309                                 sizeof(struct gfs2_quota_change);
310         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
311                                      sizeof(struct gfs2_meta_header))
312                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
313
314         /*
315          * We always keep at least one block reserved for revokes in
316          * transactions.  This greatly simplifies allocating additional
317          * revoke blocks.
318          */
319         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
320
321         /* Compute maximum reservation required to add a entry to a directory */
322
323         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
324                              sdp->sd_jbsize);
325
326         ind_blocks = 0;
327         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
328                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
329                 ind_blocks += tmp_blocks;
330         }
331
332         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
333
334         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
335
336         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
337                                 sizeof(struct gfs2_dinode);
338         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
339         for (x = 2;; x++) {
340                 u64 space, d;
341                 u32 m;
342
343                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
344                 d = space;
345                 m = do_div(d, sdp->sd_inptrs);
346
347                 if (d != sdp->sd_heightsize[x - 1] || m)
348                         break;
349                 sdp->sd_heightsize[x] = space;
350         }
351         sdp->sd_max_height = x;
352         sdp->sd_heightsize[x] = ~0;
353         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
354
355         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
356                                       sizeof(struct gfs2_leaf)) /
357                                      GFS2_MIN_DIRENT_SIZE;
358         return 0;
359 }
360
361 static int init_names(struct gfs2_sbd *sdp, int silent)
362 {
363         char *proto, *table;
364         int error = 0;
365
366         proto = sdp->sd_args.ar_lockproto;
367         table = sdp->sd_args.ar_locktable;
368
369         /*  Try to autodetect  */
370
371         if (!proto[0] || !table[0]) {
372                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
373                 if (error)
374                         return error;
375
376                 if (!proto[0])
377                         proto = sdp->sd_sb.sb_lockproto;
378                 if (!table[0])
379                         table = sdp->sd_sb.sb_locktable;
380         }
381
382         if (!table[0])
383                 table = sdp->sd_vfs->s_id;
384
385         BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
386
387         strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
388         strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
389
390         table = sdp->sd_table_name;
391         while ((table = strchr(table, '/')))
392                 *table = '_';
393
394         return error;
395 }
396
397 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
398                         int undo)
399 {
400         int error = 0;
401
402         if (undo)
403                 goto fail_trans;
404
405         error = gfs2_glock_nq_num(sdp,
406                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
407                                   LM_ST_EXCLUSIVE,
408                                   LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
409                                   mount_gh);
410         if (error) {
411                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
412                 goto fail;
413         }
414
415         error = gfs2_glock_nq_num(sdp,
416                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
417                                   LM_ST_SHARED,
418                                   LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
419                                   &sdp->sd_live_gh);
420         if (error) {
421                 fs_err(sdp, "can't acquire live glock: %d\n", error);
422                 goto fail_mount;
423         }
424
425         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
426                                CREATE, &sdp->sd_rename_gl);
427         if (error) {
428                 fs_err(sdp, "can't create rename glock: %d\n", error);
429                 goto fail_live;
430         }
431
432         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
433                                CREATE, &sdp->sd_freeze_gl);
434         if (error) {
435                 fs_err(sdp, "can't create freeze glock: %d\n", error);
436                 goto fail_rename;
437         }
438
439         return 0;
440
441 fail_trans:
442         gfs2_glock_put(sdp->sd_freeze_gl);
443 fail_rename:
444         gfs2_glock_put(sdp->sd_rename_gl);
445 fail_live:
446         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
447 fail_mount:
448         gfs2_glock_dq_uninit(mount_gh);
449 fail:
450         return error;
451 }
452
453 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
454                             u64 no_addr, const char *name)
455 {
456         struct gfs2_sbd *sdp = sb->s_fs_info;
457         struct dentry *dentry;
458         struct inode *inode;
459
460         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
461                                   GFS2_BLKST_FREE /* ignore */);
462         if (IS_ERR(inode)) {
463                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
464                 return PTR_ERR(inode);
465         }
466         dentry = d_make_root(inode);
467         if (!dentry) {
468                 fs_err(sdp, "can't alloc %s dentry\n", name);
469                 return -ENOMEM;
470         }
471         *dptr = dentry;
472         return 0;
473 }
474
475 static int init_sb(struct gfs2_sbd *sdp, int silent)
476 {
477         struct super_block *sb = sdp->sd_vfs;
478         struct gfs2_holder sb_gh;
479         u64 no_addr;
480         int ret;
481
482         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
483                                 LM_ST_SHARED, 0, &sb_gh);
484         if (ret) {
485                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
486                 return ret;
487         }
488
489         ret = gfs2_read_sb(sdp, silent);
490         if (ret) {
491                 fs_err(sdp, "can't read superblock: %d\n", ret);
492                 goto out;
493         }
494
495         switch(sdp->sd_sb.sb_fs_format) {
496         case GFS2_FS_FORMAT_MAX:
497                 sb->s_xattr = gfs2_xattr_handlers_max;
498                 break;
499
500         case GFS2_FS_FORMAT_MIN:
501                 sb->s_xattr = gfs2_xattr_handlers_min;
502                 break;
503
504         default:
505                 BUG();
506         }
507
508         /* Set up the buffer cache and SB for real */
509         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
510                 ret = -EINVAL;
511                 fs_err(sdp, "FS block size (%u) is too small for device "
512                        "block size (%u)\n",
513                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
514                 goto out;
515         }
516         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
517                 ret = -EINVAL;
518                 fs_err(sdp, "FS block size (%u) is too big for machine "
519                        "page size (%u)\n",
520                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
521                 goto out;
522         }
523         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
524
525         /* Get the root inode */
526         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
527         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
528         if (ret)
529                 goto out;
530
531         /* Get the master inode */
532         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
533         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
534         if (ret) {
535                 dput(sdp->sd_root_dir);
536                 goto out;
537         }
538         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
539 out:
540         gfs2_glock_dq_uninit(&sb_gh);
541         return ret;
542 }
543
544 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
545 {
546         char *message = "FIRSTMOUNT=Done";
547         char *envp[] = { message, NULL };
548
549         fs_info(sdp, "first mount done, others may mount\n");
550
551         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
552                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
553
554         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
555 }
556
557 /**
558  * gfs2_jindex_hold - Grab a lock on the jindex
559  * @sdp: The GFS2 superblock
560  * @ji_gh: the holder for the jindex glock
561  *
562  * Returns: errno
563  */
564
565 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
566 {
567         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
568         struct qstr name;
569         char buf[20];
570         struct gfs2_jdesc *jd;
571         int error;
572
573         name.name = buf;
574
575         mutex_lock(&sdp->sd_jindex_mutex);
576
577         for (;;) {
578                 struct gfs2_inode *jip;
579
580                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
581                 if (error)
582                         break;
583
584                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
585                 name.hash = gfs2_disk_hash(name.name, name.len);
586
587                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
588                 if (error == -ENOENT) {
589                         error = 0;
590                         break;
591                 }
592
593                 gfs2_glock_dq_uninit(ji_gh);
594
595                 if (error)
596                         break;
597
598                 error = -ENOMEM;
599                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
600                 if (!jd)
601                         break;
602
603                 INIT_LIST_HEAD(&jd->extent_list);
604                 INIT_LIST_HEAD(&jd->jd_revoke_list);
605
606                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
607                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
608                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
609                         if (!jd->jd_inode)
610                                 error = -ENOENT;
611                         else
612                                 error = PTR_ERR(jd->jd_inode);
613                         kfree(jd);
614                         break;
615                 }
616
617                 d_mark_dontcache(jd->jd_inode);
618                 spin_lock(&sdp->sd_jindex_spin);
619                 jd->jd_jid = sdp->sd_journals++;
620                 jip = GFS2_I(jd->jd_inode);
621                 jd->jd_no_addr = jip->i_no_addr;
622                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
623                 spin_unlock(&sdp->sd_jindex_spin);
624         }
625
626         mutex_unlock(&sdp->sd_jindex_mutex);
627
628         return error;
629 }
630
631 /**
632  * init_statfs - look up and initialize master and local (per node) statfs inodes
633  * @sdp: The GFS2 superblock
634  *
635  * This should be called after the jindex is initialized in init_journal() and
636  * before gfs2_journal_recovery() is called because we need to be able to write
637  * to these inodes during recovery.
638  *
639  * Returns: errno
640  */
641 static int init_statfs(struct gfs2_sbd *sdp)
642 {
643         int error = 0;
644         struct inode *master = d_inode(sdp->sd_master_dir);
645         struct inode *pn = NULL;
646         char buf[30];
647         struct gfs2_jdesc *jd;
648         struct gfs2_inode *ip;
649
650         sdp->sd_statfs_inode = gfs2_lookup_meta(master, "statfs");
651         if (IS_ERR(sdp->sd_statfs_inode)) {
652                 error = PTR_ERR(sdp->sd_statfs_inode);
653                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
654                 goto out;
655         }
656         if (sdp->sd_args.ar_spectator)
657                 goto out;
658
659         pn = gfs2_lookup_meta(master, "per_node");
660         if (IS_ERR(pn)) {
661                 error = PTR_ERR(pn);
662                 fs_err(sdp, "can't find per_node directory: %d\n", error);
663                 goto put_statfs;
664         }
665
666         /* For each jid, lookup the corresponding local statfs inode in the
667          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
668         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
669                 struct local_statfs_inode *lsi =
670                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
671                 if (!lsi) {
672                         error = -ENOMEM;
673                         goto free_local;
674                 }
675                 sprintf(buf, "statfs_change%u", jd->jd_jid);
676                 lsi->si_sc_inode = gfs2_lookup_meta(pn, buf);
677                 if (IS_ERR(lsi->si_sc_inode)) {
678                         error = PTR_ERR(lsi->si_sc_inode);
679                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
680                                jd->jd_jid, error);
681                         kfree(lsi);
682                         goto free_local;
683                 }
684                 lsi->si_jid = jd->jd_jid;
685                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
686                         sdp->sd_sc_inode = lsi->si_sc_inode;
687
688                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
689         }
690
691         iput(pn);
692         pn = NULL;
693         ip = GFS2_I(sdp->sd_sc_inode);
694         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
695                                    &sdp->sd_sc_gh);
696         if (error) {
697                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
698                 goto free_local;
699         }
700         /* read in the local statfs buffer - other nodes don't change it. */
701         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
702         if (error) {
703                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
704                 goto unlock_sd_gh;
705         }
706         return 0;
707
708 unlock_sd_gh:
709         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
710 free_local:
711         free_local_statfs_inodes(sdp);
712         iput(pn);
713 put_statfs:
714         iput(sdp->sd_statfs_inode);
715 out:
716         return error;
717 }
718
719 /* Uninitialize and free up memory used by the list of statfs inodes */
720 static void uninit_statfs(struct gfs2_sbd *sdp)
721 {
722         if (!sdp->sd_args.ar_spectator) {
723                 brelse(sdp->sd_sc_bh);
724                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
725                 free_local_statfs_inodes(sdp);
726         }
727         iput(sdp->sd_statfs_inode);
728 }
729
730 static int init_journal(struct gfs2_sbd *sdp, int undo)
731 {
732         struct inode *master = d_inode(sdp->sd_master_dir);
733         struct gfs2_holder ji_gh;
734         struct gfs2_inode *ip;
735         int error = 0;
736
737         gfs2_holder_mark_uninitialized(&ji_gh);
738         if (undo)
739                 goto fail_statfs;
740
741         sdp->sd_jindex = gfs2_lookup_meta(master, "jindex");
742         if (IS_ERR(sdp->sd_jindex)) {
743                 fs_err(sdp, "can't lookup journal index: %d\n", error);
744                 return PTR_ERR(sdp->sd_jindex);
745         }
746
747         /* Load in the journal index special file */
748
749         error = gfs2_jindex_hold(sdp, &ji_gh);
750         if (error) {
751                 fs_err(sdp, "can't read journal index: %d\n", error);
752                 goto fail;
753         }
754
755         error = -EUSERS;
756         if (!gfs2_jindex_size(sdp)) {
757                 fs_err(sdp, "no journals!\n");
758                 goto fail_jindex;
759         }
760
761         atomic_set(&sdp->sd_log_blks_needed, 0);
762         if (sdp->sd_args.ar_spectator) {
763                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
764                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
765                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
766                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
767         } else {
768                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
769                         fs_err(sdp, "can't mount journal #%u\n",
770                                sdp->sd_lockstruct.ls_jid);
771                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
772                                gfs2_jindex_size(sdp),
773                                gfs2_jindex_size(sdp) - 1);
774                         goto fail_jindex;
775                 }
776                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
777
778                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
779                                           &gfs2_journal_glops,
780                                           LM_ST_EXCLUSIVE,
781                                           LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
782                                           &sdp->sd_journal_gh);
783                 if (error) {
784                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
785                         goto fail_jindex;
786                 }
787
788                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
789                 sdp->sd_jinode_gl = ip->i_gl;
790                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
791                                            LM_FLAG_NOEXP | GL_EXACT |
792                                            GL_NOCACHE | GL_NOPID,
793                                            &sdp->sd_jinode_gh);
794                 if (error) {
795                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
796                                error);
797                         goto fail_journal_gh;
798                 }
799
800                 error = gfs2_jdesc_check(sdp->sd_jdesc);
801                 if (error) {
802                         fs_err(sdp, "my journal (%u) is bad: %d\n",
803                                sdp->sd_jdesc->jd_jid, error);
804                         goto fail_jinode_gh;
805                 }
806                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
807                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
808                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
809
810                 /* Map the extents for this journal's blocks */
811                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
812         }
813         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
814
815         /* Lookup statfs inodes here so journal recovery can use them. */
816         error = init_statfs(sdp);
817         if (error)
818                 goto fail_jinode_gh;
819
820         if (sdp->sd_lockstruct.ls_first) {
821                 unsigned int x;
822                 for (x = 0; x < sdp->sd_journals; x++) {
823                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
824
825                         if (sdp->sd_args.ar_spectator) {
826                                 error = check_journal_clean(sdp, jd, true);
827                                 if (error)
828                                         goto fail_statfs;
829                                 continue;
830                         }
831                         error = gfs2_recover_journal(jd, true);
832                         if (error) {
833                                 fs_err(sdp, "error recovering journal %u: %d\n",
834                                        x, error);
835                                 goto fail_statfs;
836                         }
837                 }
838
839                 gfs2_others_may_mount(sdp);
840         } else if (!sdp->sd_args.ar_spectator) {
841                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
842                 if (error) {
843                         fs_err(sdp, "error recovering my journal: %d\n", error);
844                         goto fail_statfs;
845                 }
846         }
847
848         sdp->sd_log_idle = 1;
849         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
850         gfs2_glock_dq_uninit(&ji_gh);
851         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
852         return 0;
853
854 fail_statfs:
855         uninit_statfs(sdp);
856 fail_jinode_gh:
857         /* A withdraw may have done dq/uninit so now we need to check it */
858         if (!sdp->sd_args.ar_spectator &&
859             gfs2_holder_initialized(&sdp->sd_jinode_gh))
860                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
861 fail_journal_gh:
862         if (!sdp->sd_args.ar_spectator &&
863             gfs2_holder_initialized(&sdp->sd_journal_gh))
864                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
865 fail_jindex:
866         gfs2_jindex_free(sdp);
867         if (gfs2_holder_initialized(&ji_gh))
868                 gfs2_glock_dq_uninit(&ji_gh);
869 fail:
870         iput(sdp->sd_jindex);
871         return error;
872 }
873
874 static struct lock_class_key gfs2_quota_imutex_key;
875
876 static int init_inodes(struct gfs2_sbd *sdp, int undo)
877 {
878         int error = 0;
879         struct inode *master = d_inode(sdp->sd_master_dir);
880
881         if (undo)
882                 goto fail_qinode;
883
884         error = init_journal(sdp, undo);
885         complete_all(&sdp->sd_journal_ready);
886         if (error)
887                 goto fail;
888
889         /* Read in the resource index inode */
890         sdp->sd_rindex = gfs2_lookup_meta(master, "rindex");
891         if (IS_ERR(sdp->sd_rindex)) {
892                 error = PTR_ERR(sdp->sd_rindex);
893                 fs_err(sdp, "can't get resource index inode: %d\n", error);
894                 goto fail_journal;
895         }
896         sdp->sd_rindex_uptodate = 0;
897
898         /* Read in the quota inode */
899         sdp->sd_quota_inode = gfs2_lookup_meta(master, "quota");
900         if (IS_ERR(sdp->sd_quota_inode)) {
901                 error = PTR_ERR(sdp->sd_quota_inode);
902                 fs_err(sdp, "can't get quota file inode: %d\n", error);
903                 goto fail_rindex;
904         }
905         /*
906          * i_rwsem on quota files is special. Since this inode is hidden system
907          * file, we are safe to define locking ourselves.
908          */
909         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
910                           &gfs2_quota_imutex_key);
911
912         error = gfs2_rindex_update(sdp);
913         if (error)
914                 goto fail_qinode;
915
916         return 0;
917
918 fail_qinode:
919         iput(sdp->sd_quota_inode);
920 fail_rindex:
921         gfs2_clear_rgrpd(sdp);
922         iput(sdp->sd_rindex);
923 fail_journal:
924         init_journal(sdp, UNDO);
925 fail:
926         return error;
927 }
928
929 static int init_per_node(struct gfs2_sbd *sdp, int undo)
930 {
931         struct inode *pn = NULL;
932         char buf[30];
933         int error = 0;
934         struct gfs2_inode *ip;
935         struct inode *master = d_inode(sdp->sd_master_dir);
936
937         if (sdp->sd_args.ar_spectator)
938                 return 0;
939
940         if (undo)
941                 goto fail_qc_gh;
942
943         pn = gfs2_lookup_meta(master, "per_node");
944         if (IS_ERR(pn)) {
945                 error = PTR_ERR(pn);
946                 fs_err(sdp, "can't find per_node directory: %d\n", error);
947                 return error;
948         }
949
950         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
951         sdp->sd_qc_inode = gfs2_lookup_meta(pn, buf);
952         if (IS_ERR(sdp->sd_qc_inode)) {
953                 error = PTR_ERR(sdp->sd_qc_inode);
954                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
955                 goto fail_ut_i;
956         }
957
958         iput(pn);
959         pn = NULL;
960
961         ip = GFS2_I(sdp->sd_qc_inode);
962         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
963                                    &sdp->sd_qc_gh);
964         if (error) {
965                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
966                 goto fail_qc_i;
967         }
968
969         return 0;
970
971 fail_qc_gh:
972         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
973 fail_qc_i:
974         iput(sdp->sd_qc_inode);
975 fail_ut_i:
976         iput(pn);
977         return error;
978 }
979
980 static const match_table_t nolock_tokens = {
981         { Opt_jid, "jid=%d", },
982         { Opt_err, NULL },
983 };
984
985 static const struct lm_lockops nolock_ops = {
986         .lm_proto_name = "lock_nolock",
987         .lm_put_lock = gfs2_glock_free,
988         .lm_tokens = &nolock_tokens,
989 };
990
991 /**
992  * gfs2_lm_mount - mount a locking protocol
993  * @sdp: the filesystem
994  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
995  *
996  * Returns: errno
997  */
998
999 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1000 {
1001         const struct lm_lockops *lm;
1002         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1003         struct gfs2_args *args = &sdp->sd_args;
1004         const char *proto = sdp->sd_proto_name;
1005         const char *table = sdp->sd_table_name;
1006         char *o, *options;
1007         int ret;
1008
1009         if (!strcmp("lock_nolock", proto)) {
1010                 lm = &nolock_ops;
1011                 sdp->sd_args.ar_localflocks = 1;
1012 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1013         } else if (!strcmp("lock_dlm", proto)) {
1014                 lm = &gfs2_dlm_ops;
1015 #endif
1016         } else {
1017                 pr_info("can't find protocol %s\n", proto);
1018                 return -ENOENT;
1019         }
1020
1021         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1022
1023         ls->ls_ops = lm;
1024         ls->ls_first = 1;
1025
1026         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1027                 substring_t tmp[MAX_OPT_ARGS];
1028                 int token, option;
1029
1030                 if (!o || !*o)
1031                         continue;
1032
1033                 token = match_token(o, *lm->lm_tokens, tmp);
1034                 switch (token) {
1035                 case Opt_jid:
1036                         ret = match_int(&tmp[0], &option);
1037                         if (ret || option < 0) 
1038                                 goto hostdata_error;
1039                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1040                                 ls->ls_jid = option;
1041                         break;
1042                 case Opt_id:
1043                 case Opt_nodir:
1044                         /* Obsolete, but left for backward compat purposes */
1045                         break;
1046                 case Opt_first:
1047                         ret = match_int(&tmp[0], &option);
1048                         if (ret || (option != 0 && option != 1))
1049                                 goto hostdata_error;
1050                         ls->ls_first = option;
1051                         break;
1052                 case Opt_err:
1053                 default:
1054 hostdata_error:
1055                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1056                         return -EINVAL;
1057                 }
1058         }
1059
1060         if (lm->lm_mount == NULL) {
1061                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1062                 complete_all(&sdp->sd_locking_init);
1063                 return 0;
1064         }
1065         ret = lm->lm_mount(sdp, table);
1066         if (ret == 0)
1067                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1068                         sdp->sd_sb.sb_fs_format);
1069         complete_all(&sdp->sd_locking_init);
1070         return ret;
1071 }
1072
1073 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1074 {
1075         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1076         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1077                 lm->lm_unmount(sdp);
1078 }
1079
1080 static int wait_on_journal(struct gfs2_sbd *sdp)
1081 {
1082         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1083                 return 0;
1084
1085         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1086                 ? -EINTR : 0;
1087 }
1088
1089 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1090 {
1091         struct super_block *sb = sdp->sd_vfs;
1092         char ro[20];
1093         char spectator[20];
1094         char *envp[] = { ro, spectator, NULL };
1095         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1096         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1097         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1098 }
1099
1100 static int init_threads(struct gfs2_sbd *sdp)
1101 {
1102         struct task_struct *p;
1103         int error = 0;
1104
1105         p = kthread_create(gfs2_logd, sdp, "gfs2_logd/%s", sdp->sd_fsname);
1106         if (IS_ERR(p)) {
1107                 error = PTR_ERR(p);
1108                 fs_err(sdp, "can't create logd thread: %d\n", error);
1109                 return error;
1110         }
1111         get_task_struct(p);
1112         sdp->sd_logd_process = p;
1113
1114         p = kthread_create(gfs2_quotad, sdp, "gfs2_quotad/%s", sdp->sd_fsname);
1115         if (IS_ERR(p)) {
1116                 error = PTR_ERR(p);
1117                 fs_err(sdp, "can't create quotad thread: %d\n", error);
1118                 goto fail;
1119         }
1120         get_task_struct(p);
1121         sdp->sd_quotad_process = p;
1122
1123         wake_up_process(sdp->sd_logd_process);
1124         wake_up_process(sdp->sd_quotad_process);
1125         return 0;
1126
1127 fail:
1128         kthread_stop(sdp->sd_logd_process);
1129         put_task_struct(sdp->sd_logd_process);
1130         sdp->sd_logd_process = NULL;
1131         return error;
1132 }
1133
1134 void gfs2_destroy_threads(struct gfs2_sbd *sdp)
1135 {
1136         if (sdp->sd_logd_process) {
1137                 kthread_stop(sdp->sd_logd_process);
1138                 put_task_struct(sdp->sd_logd_process);
1139                 sdp->sd_logd_process = NULL;
1140         }
1141         if (sdp->sd_quotad_process) {
1142                 kthread_stop(sdp->sd_quotad_process);
1143                 put_task_struct(sdp->sd_quotad_process);
1144                 sdp->sd_quotad_process = NULL;
1145         }
1146 }
1147
1148 /**
1149  * gfs2_fill_super - Read in superblock
1150  * @sb: The VFS superblock
1151  * @fc: Mount options and flags
1152  *
1153  * Returns: -errno
1154  */
1155 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1156 {
1157         struct gfs2_args *args = fc->fs_private;
1158         int silent = fc->sb_flags & SB_SILENT;
1159         struct gfs2_sbd *sdp;
1160         struct gfs2_holder mount_gh;
1161         int error;
1162
1163         sdp = init_sbd(sb);
1164         if (!sdp) {
1165                 pr_warn("can't alloc struct gfs2_sbd\n");
1166                 return -ENOMEM;
1167         }
1168         sdp->sd_args = *args;
1169
1170         if (sdp->sd_args.ar_spectator) {
1171                 sb->s_flags |= SB_RDONLY;
1172                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1173         }
1174         if (sdp->sd_args.ar_posix_acl)
1175                 sb->s_flags |= SB_POSIXACL;
1176         if (sdp->sd_args.ar_nobarrier)
1177                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1178
1179         sb->s_flags |= SB_NOSEC;
1180         sb->s_magic = GFS2_MAGIC;
1181         sb->s_op = &gfs2_super_ops;
1182         sb->s_d_op = &gfs2_dops;
1183         sb->s_export_op = &gfs2_export_ops;
1184         sb->s_qcop = &gfs2_quotactl_ops;
1185         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1186         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1187         sb->s_time_gran = 1;
1188         sb->s_maxbytes = MAX_LFS_FILESIZE;
1189
1190         /* Set up the buffer cache and fill in some fake block size values
1191            to allow us to read-in the on-disk superblock. */
1192         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, 512);
1193         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1194         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
1195         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1196
1197         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1198         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1199         if (sdp->sd_args.ar_statfs_quantum) {
1200                 sdp->sd_tune.gt_statfs_slow = 0;
1201                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1202         } else {
1203                 sdp->sd_tune.gt_statfs_slow = 1;
1204                 sdp->sd_tune.gt_statfs_quantum = 30;
1205         }
1206
1207         error = init_names(sdp, silent);
1208         if (error)
1209                 goto fail_free;
1210
1211         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1212
1213         sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s",
1214                         WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname);
1215         error = -ENOMEM;
1216         if (!sdp->sd_delete_wq)
1217                 goto fail_free;
1218
1219         error = gfs2_sys_fs_add(sdp);
1220         if (error)
1221                 goto fail_delete_wq;
1222
1223         gfs2_create_debugfs_file(sdp);
1224
1225         error = gfs2_lm_mount(sdp, silent);
1226         if (error)
1227                 goto fail_debug;
1228
1229         error = init_locking(sdp, &mount_gh, DO);
1230         if (error)
1231                 goto fail_lm;
1232
1233         error = init_sb(sdp, silent);
1234         if (error)
1235                 goto fail_locking;
1236
1237         /* Turn rgrplvb on by default if fs format is recent enough */
1238         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1239                 sdp->sd_args.ar_rgrplvb = 1;
1240
1241         error = wait_on_journal(sdp);
1242         if (error)
1243                 goto fail_sb;
1244
1245         /*
1246          * If user space has failed to join the cluster or some similar
1247          * failure has occurred, then the journal id will contain a
1248          * negative (error) number. This will then be returned to the
1249          * caller (of the mount syscall). We do this even for spectator
1250          * mounts (which just write a jid of 0 to indicate "ok" even though
1251          * the jid is unused in the spectator case)
1252          */
1253         if (sdp->sd_lockstruct.ls_jid < 0) {
1254                 error = sdp->sd_lockstruct.ls_jid;
1255                 sdp->sd_lockstruct.ls_jid = 0;
1256                 goto fail_sb;
1257         }
1258
1259         if (sdp->sd_args.ar_spectator)
1260                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1261                          sdp->sd_table_name);
1262         else
1263                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1264                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1265
1266         error = init_inodes(sdp, DO);
1267         if (error)
1268                 goto fail_sb;
1269
1270         error = init_per_node(sdp, DO);
1271         if (error)
1272                 goto fail_inodes;
1273
1274         error = gfs2_statfs_init(sdp);
1275         if (error) {
1276                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1277                 goto fail_per_node;
1278         }
1279
1280         if (!sb_rdonly(sb)) {
1281                 error = init_threads(sdp);
1282                 if (error)
1283                         goto fail_per_node;
1284         }
1285
1286         error = gfs2_freeze_lock_shared(sdp);
1287         if (error)
1288                 goto fail_per_node;
1289
1290         if (!sb_rdonly(sb))
1291                 error = gfs2_make_fs_rw(sdp);
1292
1293         if (error) {
1294                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
1295                 gfs2_destroy_threads(sdp);
1296                 fs_err(sdp, "can't make FS RW: %d\n", error);
1297                 goto fail_per_node;
1298         }
1299         gfs2_glock_dq_uninit(&mount_gh);
1300         gfs2_online_uevent(sdp);
1301         return 0;
1302
1303 fail_per_node:
1304         init_per_node(sdp, UNDO);
1305 fail_inodes:
1306         init_inodes(sdp, UNDO);
1307 fail_sb:
1308         if (sdp->sd_root_dir)
1309                 dput(sdp->sd_root_dir);
1310         if (sdp->sd_master_dir)
1311                 dput(sdp->sd_master_dir);
1312         if (sb->s_root)
1313                 dput(sb->s_root);
1314         sb->s_root = NULL;
1315 fail_locking:
1316         init_locking(sdp, &mount_gh, UNDO);
1317 fail_lm:
1318         complete_all(&sdp->sd_journal_ready);
1319         gfs2_gl_hash_clear(sdp);
1320         gfs2_lm_unmount(sdp);
1321 fail_debug:
1322         gfs2_delete_debugfs_file(sdp);
1323         gfs2_sys_fs_del(sdp);
1324 fail_delete_wq:
1325         destroy_workqueue(sdp->sd_delete_wq);
1326 fail_free:
1327         free_sbd(sdp);
1328         sb->s_fs_info = NULL;
1329         return error;
1330 }
1331
1332 /**
1333  * gfs2_get_tree - Get the GFS2 superblock and root directory
1334  * @fc: The filesystem context
1335  *
1336  * Returns: 0 or -errno on error
1337  */
1338 static int gfs2_get_tree(struct fs_context *fc)
1339 {
1340         struct gfs2_args *args = fc->fs_private;
1341         struct gfs2_sbd *sdp;
1342         int error;
1343
1344         error = get_tree_bdev(fc, gfs2_fill_super);
1345         if (error)
1346                 return error;
1347
1348         sdp = fc->root->d_sb->s_fs_info;
1349         dput(fc->root);
1350         if (args->ar_meta)
1351                 fc->root = dget(sdp->sd_master_dir);
1352         else
1353                 fc->root = dget(sdp->sd_root_dir);
1354         return 0;
1355 }
1356
1357 static void gfs2_fc_free(struct fs_context *fc)
1358 {
1359         struct gfs2_args *args = fc->fs_private;
1360
1361         kfree(args);
1362 }
1363
1364 enum gfs2_param {
1365         Opt_lockproto,
1366         Opt_locktable,
1367         Opt_hostdata,
1368         Opt_spectator,
1369         Opt_ignore_local_fs,
1370         Opt_localflocks,
1371         Opt_localcaching,
1372         Opt_debug,
1373         Opt_upgrade,
1374         Opt_acl,
1375         Opt_quota,
1376         Opt_quota_flag,
1377         Opt_suiddir,
1378         Opt_data,
1379         Opt_meta,
1380         Opt_discard,
1381         Opt_commit,
1382         Opt_errors,
1383         Opt_statfs_quantum,
1384         Opt_statfs_percent,
1385         Opt_quota_quantum,
1386         Opt_barrier,
1387         Opt_rgrplvb,
1388         Opt_loccookie,
1389 };
1390
1391 static const struct constant_table gfs2_param_quota[] = {
1392         {"off",        GFS2_QUOTA_OFF},
1393         {"account",    GFS2_QUOTA_ACCOUNT},
1394         {"on",         GFS2_QUOTA_ON},
1395         {"quiet",      GFS2_QUOTA_QUIET},
1396         {}
1397 };
1398
1399 enum opt_data {
1400         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1401         Opt_data_ordered   = GFS2_DATA_ORDERED,
1402 };
1403
1404 static const struct constant_table gfs2_param_data[] = {
1405         {"writeback",  Opt_data_writeback },
1406         {"ordered",    Opt_data_ordered },
1407         {}
1408 };
1409
1410 enum opt_errors {
1411         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1412         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1413 };
1414
1415 static const struct constant_table gfs2_param_errors[] = {
1416         {"withdraw",   Opt_errors_withdraw },
1417         {"panic",      Opt_errors_panic },
1418         {}
1419 };
1420
1421 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1422         fsparam_string ("lockproto",          Opt_lockproto),
1423         fsparam_string ("locktable",          Opt_locktable),
1424         fsparam_string ("hostdata",           Opt_hostdata),
1425         fsparam_flag   ("spectator",          Opt_spectator),
1426         fsparam_flag   ("norecovery",         Opt_spectator),
1427         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1428         fsparam_flag   ("localflocks",        Opt_localflocks),
1429         fsparam_flag   ("localcaching",       Opt_localcaching),
1430         fsparam_flag_no("debug",              Opt_debug),
1431         fsparam_flag   ("upgrade",            Opt_upgrade),
1432         fsparam_flag_no("acl",                Opt_acl),
1433         fsparam_flag_no("suiddir",            Opt_suiddir),
1434         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1435         fsparam_flag   ("meta",               Opt_meta),
1436         fsparam_flag_no("discard",            Opt_discard),
1437         fsparam_s32    ("commit",             Opt_commit),
1438         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1439         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1440         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1441         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1442         fsparam_flag_no("barrier",            Opt_barrier),
1443         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1444         fsparam_flag_no("loccookie",          Opt_loccookie),
1445         /* quota can be a flag or an enum so it gets special treatment */
1446         fsparam_flag_no("quota",              Opt_quota_flag),
1447         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1448         {}
1449 };
1450
1451 /* Parse a single mount parameter */
1452 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1453 {
1454         struct gfs2_args *args = fc->fs_private;
1455         struct fs_parse_result result;
1456         int o;
1457
1458         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1459         if (o < 0)
1460                 return o;
1461
1462         switch (o) {
1463         case Opt_lockproto:
1464                 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1465                 break;
1466         case Opt_locktable:
1467                 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1468                 break;
1469         case Opt_hostdata:
1470                 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1471                 break;
1472         case Opt_spectator:
1473                 args->ar_spectator = 1;
1474                 break;
1475         case Opt_ignore_local_fs:
1476                 /* Retained for backwards compat only */
1477                 break;
1478         case Opt_localflocks:
1479                 args->ar_localflocks = 1;
1480                 break;
1481         case Opt_localcaching:
1482                 /* Retained for backwards compat only */
1483                 break;
1484         case Opt_debug:
1485                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1486                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1487                 args->ar_debug = result.boolean;
1488                 break;
1489         case Opt_upgrade:
1490                 /* Retained for backwards compat only */
1491                 break;
1492         case Opt_acl:
1493                 args->ar_posix_acl = result.boolean;
1494                 break;
1495         case Opt_quota_flag:
1496                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1497                 break;
1498         case Opt_quota:
1499                 args->ar_quota = result.int_32;
1500                 break;
1501         case Opt_suiddir:
1502                 args->ar_suiddir = result.boolean;
1503                 break;
1504         case Opt_data:
1505                 /* The uint_32 result maps directly to GFS2_DATA_* */
1506                 args->ar_data = result.uint_32;
1507                 break;
1508         case Opt_meta:
1509                 args->ar_meta = 1;
1510                 break;
1511         case Opt_discard:
1512                 args->ar_discard = result.boolean;
1513                 break;
1514         case Opt_commit:
1515                 if (result.int_32 <= 0)
1516                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1517                 args->ar_commit = result.int_32;
1518                 break;
1519         case Opt_statfs_quantum:
1520                 if (result.int_32 < 0)
1521                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1522                 args->ar_statfs_quantum = result.int_32;
1523                 break;
1524         case Opt_quota_quantum:
1525                 if (result.int_32 <= 0)
1526                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1527                 args->ar_quota_quantum = result.int_32;
1528                 break;
1529         case Opt_statfs_percent:
1530                 if (result.int_32 < 0 || result.int_32 > 100)
1531                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1532                 args->ar_statfs_percent = result.int_32;
1533                 break;
1534         case Opt_errors:
1535                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1536                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1537                 args->ar_errors = result.uint_32;
1538                 break;
1539         case Opt_barrier:
1540                 args->ar_nobarrier = result.boolean;
1541                 break;
1542         case Opt_rgrplvb:
1543                 args->ar_rgrplvb = result.boolean;
1544                 args->ar_got_rgrplvb = 1;
1545                 break;
1546         case Opt_loccookie:
1547                 args->ar_loccookie = result.boolean;
1548                 break;
1549         default:
1550                 return invalfc(fc, "invalid mount option: %s", param->key);
1551         }
1552         return 0;
1553 }
1554
1555 static int gfs2_reconfigure(struct fs_context *fc)
1556 {
1557         struct super_block *sb = fc->root->d_sb;
1558         struct gfs2_sbd *sdp = sb->s_fs_info;
1559         struct gfs2_args *oldargs = &sdp->sd_args;
1560         struct gfs2_args *newargs = fc->fs_private;
1561         struct gfs2_tune *gt = &sdp->sd_tune;
1562         int error = 0;
1563
1564         sync_filesystem(sb);
1565
1566         spin_lock(&gt->gt_spin);
1567         oldargs->ar_commit = gt->gt_logd_secs;
1568         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1569         if (gt->gt_statfs_slow)
1570                 oldargs->ar_statfs_quantum = 0;
1571         else
1572                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1573         spin_unlock(&gt->gt_spin);
1574
1575         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1576                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1577                 return -EINVAL;
1578         }
1579         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1580                 errorfc(fc, "reconfiguration of lock table not allowed");
1581                 return -EINVAL;
1582         }
1583         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1584                 errorfc(fc, "reconfiguration of host data not allowed");
1585                 return -EINVAL;
1586         }
1587         if (newargs->ar_spectator != oldargs->ar_spectator) {
1588                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1589                 return -EINVAL;
1590         }
1591         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1592                 errorfc(fc, "reconfiguration of localflocks not allowed");
1593                 return -EINVAL;
1594         }
1595         if (newargs->ar_meta != oldargs->ar_meta) {
1596                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1597                 return -EINVAL;
1598         }
1599         if (oldargs->ar_spectator)
1600                 fc->sb_flags |= SB_RDONLY;
1601
1602         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1603                 if (fc->sb_flags & SB_RDONLY) {
1604                         gfs2_make_fs_ro(sdp);
1605                 } else {
1606                         error = gfs2_make_fs_rw(sdp);
1607                         if (error)
1608                                 errorfc(fc, "unable to remount read-write");
1609                 }
1610         }
1611         sdp->sd_args = *newargs;
1612
1613         if (sdp->sd_args.ar_posix_acl)
1614                 sb->s_flags |= SB_POSIXACL;
1615         else
1616                 sb->s_flags &= ~SB_POSIXACL;
1617         if (sdp->sd_args.ar_nobarrier)
1618                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1619         else
1620                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1621         spin_lock(&gt->gt_spin);
1622         gt->gt_logd_secs = newargs->ar_commit;
1623         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1624         if (newargs->ar_statfs_quantum) {
1625                 gt->gt_statfs_slow = 0;
1626                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1627         }
1628         else {
1629                 gt->gt_statfs_slow = 1;
1630                 gt->gt_statfs_quantum = 30;
1631         }
1632         spin_unlock(&gt->gt_spin);
1633
1634         gfs2_online_uevent(sdp);
1635         return error;
1636 }
1637
1638 static const struct fs_context_operations gfs2_context_ops = {
1639         .free        = gfs2_fc_free,
1640         .parse_param = gfs2_parse_param,
1641         .get_tree    = gfs2_get_tree,
1642         .reconfigure = gfs2_reconfigure,
1643 };
1644
1645 /* Set up the filesystem mount context */
1646 static int gfs2_init_fs_context(struct fs_context *fc)
1647 {
1648         struct gfs2_args *args;
1649
1650         args = kmalloc(sizeof(*args), GFP_KERNEL);
1651         if (args == NULL)
1652                 return -ENOMEM;
1653
1654         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1655                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1656
1657                 *args = sdp->sd_args;
1658         } else {
1659                 memset(args, 0, sizeof(*args));
1660                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1661                 args->ar_data = GFS2_DATA_DEFAULT;
1662                 args->ar_commit = 30;
1663                 args->ar_statfs_quantum = 30;
1664                 args->ar_quota_quantum = 60;
1665                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1666         }
1667         fc->fs_private = args;
1668         fc->ops = &gfs2_context_ops;
1669         return 0;
1670 }
1671
1672 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1673 {
1674         return -EINVAL;
1675 }
1676
1677 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1678 {
1679         return (fc->sget_key == s->s_bdev);
1680 }
1681
1682 static int gfs2_meta_get_tree(struct fs_context *fc)
1683 {
1684         struct super_block *s;
1685         struct gfs2_sbd *sdp;
1686         struct path path;
1687         int error;
1688
1689         if (!fc->source || !*fc->source)
1690                 return -EINVAL;
1691
1692         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1693         if (error) {
1694                 pr_warn("path_lookup on %s returned error %d\n",
1695                         fc->source, error);
1696                 return error;
1697         }
1698         fc->fs_type = &gfs2_fs_type;
1699         fc->sget_key = path.dentry->d_sb->s_bdev;
1700         s = sget_fc(fc, test_meta_super, set_meta_super);
1701         path_put(&path);
1702         if (IS_ERR(s)) {
1703                 pr_warn("gfs2 mount does not exist\n");
1704                 return PTR_ERR(s);
1705         }
1706         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1707                 deactivate_locked_super(s);
1708                 return -EBUSY;
1709         }
1710         sdp = s->s_fs_info;
1711         fc->root = dget(sdp->sd_master_dir);
1712         return 0;
1713 }
1714
1715 static const struct fs_context_operations gfs2_meta_context_ops = {
1716         .free        = gfs2_fc_free,
1717         .get_tree    = gfs2_meta_get_tree,
1718 };
1719
1720 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1721 {
1722         int ret = gfs2_init_fs_context(fc);
1723
1724         if (ret)
1725                 return ret;
1726
1727         fc->ops = &gfs2_meta_context_ops;
1728         return 0;
1729 }
1730
1731 /**
1732  * gfs2_evict_inodes - evict inodes cooperatively
1733  * @sb: the superblock
1734  *
1735  * When evicting an inode with a zero link count, we are trying to upgrade the
1736  * inode's iopen glock from SH to EX mode in order to determine if we can
1737  * delete the inode.  The other nodes are supposed to evict the inode from
1738  * their caches if they can, and to poke the inode's inode glock if they cannot
1739  * do so.  Either behavior allows gfs2_upgrade_iopen_glock() to proceed
1740  * quickly, but if the other nodes are not cooperating, the lock upgrading
1741  * attempt will time out.  Since inodes are evicted sequentially, this can add
1742  * up quickly.
1743  *
1744  * Function evict_inodes() tries to keep the s_inode_list_lock list locked over
1745  * a long time, which prevents other inodes from being evicted concurrently.
1746  * This precludes the cooperative behavior we are looking for.  This special
1747  * version of evict_inodes() avoids that.
1748  *
1749  * Modeled after drop_pagecache_sb().
1750  */
1751 static void gfs2_evict_inodes(struct super_block *sb)
1752 {
1753         struct inode *inode, *toput_inode = NULL;
1754         struct gfs2_sbd *sdp = sb->s_fs_info;
1755
1756         set_bit(SDF_EVICTING, &sdp->sd_flags);
1757
1758         spin_lock(&sb->s_inode_list_lock);
1759         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1760                 spin_lock(&inode->i_lock);
1761                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) &&
1762                     !need_resched()) {
1763                         spin_unlock(&inode->i_lock);
1764                         continue;
1765                 }
1766                 atomic_inc(&inode->i_count);
1767                 spin_unlock(&inode->i_lock);
1768                 spin_unlock(&sb->s_inode_list_lock);
1769
1770                 iput(toput_inode);
1771                 toput_inode = inode;
1772
1773                 cond_resched();
1774                 spin_lock(&sb->s_inode_list_lock);
1775         }
1776         spin_unlock(&sb->s_inode_list_lock);
1777         iput(toput_inode);
1778 }
1779
1780 static void gfs2_kill_sb(struct super_block *sb)
1781 {
1782         struct gfs2_sbd *sdp = sb->s_fs_info;
1783
1784         if (sdp == NULL) {
1785                 kill_block_super(sb);
1786                 return;
1787         }
1788
1789         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1790         dput(sdp->sd_root_dir);
1791         dput(sdp->sd_master_dir);
1792         sdp->sd_root_dir = NULL;
1793         sdp->sd_master_dir = NULL;
1794         shrink_dcache_sb(sb);
1795
1796         gfs2_evict_inodes(sb);
1797
1798         /*
1799          * Flush and then drain the delete workqueue here (via
1800          * destroy_workqueue()) to ensure that any delete work that
1801          * may be running will also see the SDF_KILL flag.
1802          */
1803         set_bit(SDF_KILL, &sdp->sd_flags);
1804         gfs2_flush_delete_work(sdp);
1805         destroy_workqueue(sdp->sd_delete_wq);
1806
1807         kill_block_super(sb);
1808 }
1809
1810 struct file_system_type gfs2_fs_type = {
1811         .name = "gfs2",
1812         .fs_flags = FS_REQUIRES_DEV,
1813         .init_fs_context = gfs2_init_fs_context,
1814         .parameters = gfs2_fs_parameters,
1815         .kill_sb = gfs2_kill_sb,
1816         .owner = THIS_MODULE,
1817 };
1818 MODULE_ALIAS_FS("gfs2");
1819
1820 struct file_system_type gfs2meta_fs_type = {
1821         .name = "gfs2meta",
1822         .fs_flags = FS_REQUIRES_DEV,
1823         .init_fs_context = gfs2_meta_init_fs_context,
1824         .owner = THIS_MODULE,
1825 };
1826 MODULE_ALIAS_FS("gfs2meta");