Merge tag 'rproc-v6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/remoteproc...
[platform/kernel/linux-rpi.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_glock_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         atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
139         mutex_init(&sdp->sd_freeze_mutex);
140
141         return sdp;
142
143 fail:
144         free_sbd(sdp);
145         return NULL;
146 }
147
148 /**
149  * gfs2_check_sb - Check superblock
150  * @sdp: the filesystem
151  * @silent: Don't print a message if the check fails
152  *
153  * Checks the version code of the FS is one that we understand how to
154  * read and that the sizes of the various on-disk structures have not
155  * changed.
156  */
157
158 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
159 {
160         struct gfs2_sb_host *sb = &sdp->sd_sb;
161
162         if (sb->sb_magic != GFS2_MAGIC ||
163             sb->sb_type != GFS2_METATYPE_SB) {
164                 if (!silent)
165                         pr_warn("not a GFS2 filesystem\n");
166                 return -EINVAL;
167         }
168
169         if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
170             sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
171             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
172                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
173                 return -EINVAL;
174         }
175
176         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
177             (sb->sb_bsize & (sb->sb_bsize - 1))) {
178                 pr_warn("Invalid block size\n");
179                 return -EINVAL;
180         }
181         if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
182                 pr_warn("Invalid block size shift\n");
183                 return -EINVAL;
184         }
185         return 0;
186 }
187
188 static void end_bio_io_page(struct bio *bio)
189 {
190         struct page *page = bio->bi_private;
191
192         if (!bio->bi_status)
193                 SetPageUptodate(page);
194         else
195                 pr_warn("error %d reading superblock\n", bio->bi_status);
196         unlock_page(page);
197 }
198
199 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
200 {
201         struct gfs2_sb_host *sb = &sdp->sd_sb;
202         struct super_block *s = sdp->sd_vfs;
203         const struct gfs2_sb *str = buf;
204
205         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
206         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
207         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
208         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
209         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
210         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
211         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
212         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
213         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
214         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
215
216         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
217         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
218         memcpy(&s->s_uuid, str->sb_uuid, 16);
219 }
220
221 /**
222  * gfs2_read_super - Read the gfs2 super block from disk
223  * @sdp: The GFS2 super block
224  * @sector: The location of the super block
225  * @silent: Don't print a message if the check fails
226  *
227  * This uses the bio functions to read the super block from disk
228  * because we want to be 100% sure that we never read cached data.
229  * A super block is read twice only during each GFS2 mount and is
230  * never written to by the filesystem. The first time its read no
231  * locks are held, and the only details which are looked at are those
232  * relating to the locking protocol. Once locking is up and working,
233  * the sb is read again under the lock to establish the location of
234  * the master directory (contains pointers to journals etc) and the
235  * root directory.
236  *
237  * Returns: 0 on success or error
238  */
239
240 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
241 {
242         struct super_block *sb = sdp->sd_vfs;
243         struct gfs2_sb *p;
244         struct page *page;
245         struct bio *bio;
246
247         page = alloc_page(GFP_NOFS);
248         if (unlikely(!page))
249                 return -ENOMEM;
250
251         ClearPageUptodate(page);
252         ClearPageDirty(page);
253         lock_page(page);
254
255         bio = bio_alloc(sb->s_bdev, 1, REQ_OP_READ | REQ_META, GFP_NOFS);
256         bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
257         __bio_add_page(bio, page, PAGE_SIZE, 0);
258
259         bio->bi_end_io = end_bio_io_page;
260         bio->bi_private = page;
261         submit_bio(bio);
262         wait_on_page_locked(page);
263         bio_put(bio);
264         if (!PageUptodate(page)) {
265                 __free_page(page);
266                 return -EIO;
267         }
268         p = kmap(page);
269         gfs2_sb_in(sdp, p);
270         kunmap(page);
271         __free_page(page);
272         return gfs2_check_sb(sdp, silent);
273 }
274
275 /**
276  * gfs2_read_sb - Read super block
277  * @sdp: The GFS2 superblock
278  * @silent: Don't print message if mount fails
279  *
280  */
281
282 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
283 {
284         u32 hash_blocks, ind_blocks, leaf_blocks;
285         u32 tmp_blocks;
286         unsigned int x;
287         int error;
288
289         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
290         if (error) {
291                 if (!silent)
292                         fs_err(sdp, "can't read superblock\n");
293                 return error;
294         }
295
296         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
297                                GFS2_BASIC_BLOCK_SHIFT;
298         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
299         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
300                           sizeof(struct gfs2_dinode)) / sizeof(u64);
301         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
302                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
303         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
304                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
305         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
306         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
307         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
308         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
309         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
310                                 sizeof(struct gfs2_meta_header)) /
311                                 sizeof(struct gfs2_quota_change);
312         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
313                                      sizeof(struct gfs2_meta_header))
314                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
315
316         /*
317          * We always keep at least one block reserved for revokes in
318          * transactions.  This greatly simplifies allocating additional
319          * revoke blocks.
320          */
321         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
322
323         /* Compute maximum reservation required to add a entry to a directory */
324
325         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
326                              sdp->sd_jbsize);
327
328         ind_blocks = 0;
329         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
330                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
331                 ind_blocks += tmp_blocks;
332         }
333
334         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
335
336         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
337
338         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
339                                 sizeof(struct gfs2_dinode);
340         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
341         for (x = 2;; x++) {
342                 u64 space, d;
343                 u32 m;
344
345                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
346                 d = space;
347                 m = do_div(d, sdp->sd_inptrs);
348
349                 if (d != sdp->sd_heightsize[x - 1] || m)
350                         break;
351                 sdp->sd_heightsize[x] = space;
352         }
353         sdp->sd_max_height = x;
354         sdp->sd_heightsize[x] = ~0;
355         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
356
357         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
358                                       sizeof(struct gfs2_leaf)) /
359                                      GFS2_MIN_DIRENT_SIZE;
360         return 0;
361 }
362
363 static int init_names(struct gfs2_sbd *sdp, int silent)
364 {
365         char *proto, *table;
366         int error = 0;
367
368         proto = sdp->sd_args.ar_lockproto;
369         table = sdp->sd_args.ar_locktable;
370
371         /*  Try to autodetect  */
372
373         if (!proto[0] || !table[0]) {
374                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
375                 if (error)
376                         return error;
377
378                 if (!proto[0])
379                         proto = sdp->sd_sb.sb_lockproto;
380                 if (!table[0])
381                         table = sdp->sd_sb.sb_locktable;
382         }
383
384         if (!table[0])
385                 table = sdp->sd_vfs->s_id;
386
387         BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
388
389         strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
390         strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
391
392         table = sdp->sd_table_name;
393         while ((table = strchr(table, '/')))
394                 *table = '_';
395
396         return error;
397 }
398
399 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
400                         int undo)
401 {
402         int error = 0;
403
404         if (undo)
405                 goto fail_trans;
406
407         error = gfs2_glock_nq_num(sdp,
408                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
409                                   LM_ST_EXCLUSIVE,
410                                   LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
411                                   mount_gh);
412         if (error) {
413                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
414                 goto fail;
415         }
416
417         error = gfs2_glock_nq_num(sdp,
418                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
419                                   LM_ST_SHARED,
420                                   LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
421                                   &sdp->sd_live_gh);
422         if (error) {
423                 fs_err(sdp, "can't acquire live glock: %d\n", error);
424                 goto fail_mount;
425         }
426
427         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
428                                CREATE, &sdp->sd_rename_gl);
429         if (error) {
430                 fs_err(sdp, "can't create rename glock: %d\n", error);
431                 goto fail_live;
432         }
433
434         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
435                                CREATE, &sdp->sd_freeze_gl);
436         if (error) {
437                 fs_err(sdp, "can't create transaction glock: %d\n", error);
438                 goto fail_rename;
439         }
440
441         return 0;
442
443 fail_trans:
444         gfs2_glock_put(sdp->sd_freeze_gl);
445 fail_rename:
446         gfs2_glock_put(sdp->sd_rename_gl);
447 fail_live:
448         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
449 fail_mount:
450         gfs2_glock_dq_uninit(mount_gh);
451 fail:
452         return error;
453 }
454
455 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
456                             u64 no_addr, const char *name)
457 {
458         struct gfs2_sbd *sdp = sb->s_fs_info;
459         struct dentry *dentry;
460         struct inode *inode;
461
462         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
463                                   GFS2_BLKST_FREE /* ignore */);
464         if (IS_ERR(inode)) {
465                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
466                 return PTR_ERR(inode);
467         }
468         dentry = d_make_root(inode);
469         if (!dentry) {
470                 fs_err(sdp, "can't alloc %s dentry\n", name);
471                 return -ENOMEM;
472         }
473         *dptr = dentry;
474         return 0;
475 }
476
477 static int init_sb(struct gfs2_sbd *sdp, int silent)
478 {
479         struct super_block *sb = sdp->sd_vfs;
480         struct gfs2_holder sb_gh;
481         u64 no_addr;
482         int ret;
483
484         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
485                                 LM_ST_SHARED, 0, &sb_gh);
486         if (ret) {
487                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
488                 return ret;
489         }
490
491         ret = gfs2_read_sb(sdp, silent);
492         if (ret) {
493                 fs_err(sdp, "can't read superblock: %d\n", ret);
494                 goto out;
495         }
496
497         switch(sdp->sd_sb.sb_fs_format) {
498         case GFS2_FS_FORMAT_MAX:
499                 sb->s_xattr = gfs2_xattr_handlers_max;
500                 break;
501
502         case GFS2_FS_FORMAT_MIN:
503                 sb->s_xattr = gfs2_xattr_handlers_min;
504                 break;
505
506         default:
507                 BUG();
508         }
509
510         /* Set up the buffer cache and SB for real */
511         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
512                 ret = -EINVAL;
513                 fs_err(sdp, "FS block size (%u) is too small for device "
514                        "block size (%u)\n",
515                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
516                 goto out;
517         }
518         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
519                 ret = -EINVAL;
520                 fs_err(sdp, "FS block size (%u) is too big for machine "
521                        "page size (%u)\n",
522                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
523                 goto out;
524         }
525         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
526
527         /* Get the root inode */
528         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
529         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
530         if (ret)
531                 goto out;
532
533         /* Get the master inode */
534         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
535         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
536         if (ret) {
537                 dput(sdp->sd_root_dir);
538                 goto out;
539         }
540         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
541 out:
542         gfs2_glock_dq_uninit(&sb_gh);
543         return ret;
544 }
545
546 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
547 {
548         char *message = "FIRSTMOUNT=Done";
549         char *envp[] = { message, NULL };
550
551         fs_info(sdp, "first mount done, others may mount\n");
552
553         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
554                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
555
556         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
557 }
558
559 /**
560  * gfs2_jindex_hold - Grab a lock on the jindex
561  * @sdp: The GFS2 superblock
562  * @ji_gh: the holder for the jindex glock
563  *
564  * Returns: errno
565  */
566
567 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
568 {
569         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
570         struct qstr name;
571         char buf[20];
572         struct gfs2_jdesc *jd;
573         int error;
574
575         name.name = buf;
576
577         mutex_lock(&sdp->sd_jindex_mutex);
578
579         for (;;) {
580                 struct gfs2_inode *jip;
581
582                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
583                 if (error)
584                         break;
585
586                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
587                 name.hash = gfs2_disk_hash(name.name, name.len);
588
589                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
590                 if (error == -ENOENT) {
591                         error = 0;
592                         break;
593                 }
594
595                 gfs2_glock_dq_uninit(ji_gh);
596
597                 if (error)
598                         break;
599
600                 error = -ENOMEM;
601                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
602                 if (!jd)
603                         break;
604
605                 INIT_LIST_HEAD(&jd->extent_list);
606                 INIT_LIST_HEAD(&jd->jd_revoke_list);
607
608                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
609                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
610                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
611                         if (!jd->jd_inode)
612                                 error = -ENOENT;
613                         else
614                                 error = PTR_ERR(jd->jd_inode);
615                         kfree(jd);
616                         break;
617                 }
618
619                 d_mark_dontcache(jd->jd_inode);
620                 spin_lock(&sdp->sd_jindex_spin);
621                 jd->jd_jid = sdp->sd_journals++;
622                 jip = GFS2_I(jd->jd_inode);
623                 jd->jd_no_addr = jip->i_no_addr;
624                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
625                 spin_unlock(&sdp->sd_jindex_spin);
626         }
627
628         mutex_unlock(&sdp->sd_jindex_mutex);
629
630         return error;
631 }
632
633 /**
634  * init_statfs - look up and initialize master and local (per node) statfs inodes
635  * @sdp: The GFS2 superblock
636  *
637  * This should be called after the jindex is initialized in init_journal() and
638  * before gfs2_journal_recovery() is called because we need to be able to write
639  * to these inodes during recovery.
640  *
641  * Returns: errno
642  */
643 static int init_statfs(struct gfs2_sbd *sdp)
644 {
645         int error = 0;
646         struct inode *master = d_inode(sdp->sd_master_dir);
647         struct inode *pn = NULL;
648         char buf[30];
649         struct gfs2_jdesc *jd;
650         struct gfs2_inode *ip;
651
652         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
653         if (IS_ERR(sdp->sd_statfs_inode)) {
654                 error = PTR_ERR(sdp->sd_statfs_inode);
655                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
656                 goto out;
657         }
658         if (sdp->sd_args.ar_spectator)
659                 goto out;
660
661         pn = gfs2_lookup_simple(master, "per_node");
662         if (IS_ERR(pn)) {
663                 error = PTR_ERR(pn);
664                 fs_err(sdp, "can't find per_node directory: %d\n", error);
665                 goto put_statfs;
666         }
667
668         /* For each jid, lookup the corresponding local statfs inode in the
669          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
670         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
671                 struct local_statfs_inode *lsi =
672                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
673                 if (!lsi) {
674                         error = -ENOMEM;
675                         goto free_local;
676                 }
677                 sprintf(buf, "statfs_change%u", jd->jd_jid);
678                 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
679                 if (IS_ERR(lsi->si_sc_inode)) {
680                         error = PTR_ERR(lsi->si_sc_inode);
681                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
682                                jd->jd_jid, error);
683                         kfree(lsi);
684                         goto free_local;
685                 }
686                 lsi->si_jid = jd->jd_jid;
687                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
688                         sdp->sd_sc_inode = lsi->si_sc_inode;
689
690                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
691         }
692
693         iput(pn);
694         pn = NULL;
695         ip = GFS2_I(sdp->sd_sc_inode);
696         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
697                                    &sdp->sd_sc_gh);
698         if (error) {
699                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
700                 goto free_local;
701         }
702         /* read in the local statfs buffer - other nodes don't change it. */
703         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
704         if (error) {
705                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
706                 goto unlock_sd_gh;
707         }
708         return 0;
709
710 unlock_sd_gh:
711         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
712 free_local:
713         free_local_statfs_inodes(sdp);
714         iput(pn);
715 put_statfs:
716         iput(sdp->sd_statfs_inode);
717 out:
718         return error;
719 }
720
721 /* Uninitialize and free up memory used by the list of statfs inodes */
722 static void uninit_statfs(struct gfs2_sbd *sdp)
723 {
724         if (!sdp->sd_args.ar_spectator) {
725                 brelse(sdp->sd_sc_bh);
726                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
727                 free_local_statfs_inodes(sdp);
728         }
729         iput(sdp->sd_statfs_inode);
730 }
731
732 static int init_journal(struct gfs2_sbd *sdp, int undo)
733 {
734         struct inode *master = d_inode(sdp->sd_master_dir);
735         struct gfs2_holder ji_gh;
736         struct gfs2_inode *ip;
737         int error = 0;
738
739         gfs2_holder_mark_uninitialized(&ji_gh);
740         if (undo)
741                 goto fail_statfs;
742
743         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
744         if (IS_ERR(sdp->sd_jindex)) {
745                 fs_err(sdp, "can't lookup journal index: %d\n", error);
746                 return PTR_ERR(sdp->sd_jindex);
747         }
748
749         /* Load in the journal index special file */
750
751         error = gfs2_jindex_hold(sdp, &ji_gh);
752         if (error) {
753                 fs_err(sdp, "can't read journal index: %d\n", error);
754                 goto fail;
755         }
756
757         error = -EUSERS;
758         if (!gfs2_jindex_size(sdp)) {
759                 fs_err(sdp, "no journals!\n");
760                 goto fail_jindex;
761         }
762
763         atomic_set(&sdp->sd_log_blks_needed, 0);
764         if (sdp->sd_args.ar_spectator) {
765                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
766                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
767                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
768                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
769         } else {
770                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
771                         fs_err(sdp, "can't mount journal #%u\n",
772                                sdp->sd_lockstruct.ls_jid);
773                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
774                                gfs2_jindex_size(sdp),
775                                gfs2_jindex_size(sdp) - 1);
776                         goto fail_jindex;
777                 }
778                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
779
780                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
781                                           &gfs2_journal_glops,
782                                           LM_ST_EXCLUSIVE,
783                                           LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
784                                           &sdp->sd_journal_gh);
785                 if (error) {
786                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
787                         goto fail_jindex;
788                 }
789
790                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
791                 sdp->sd_jinode_gl = ip->i_gl;
792                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
793                                            LM_FLAG_NOEXP | GL_EXACT |
794                                            GL_NOCACHE | GL_NOPID,
795                                            &sdp->sd_jinode_gh);
796                 if (error) {
797                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
798                                error);
799                         goto fail_journal_gh;
800                 }
801
802                 error = gfs2_jdesc_check(sdp->sd_jdesc);
803                 if (error) {
804                         fs_err(sdp, "my journal (%u) is bad: %d\n",
805                                sdp->sd_jdesc->jd_jid, error);
806                         goto fail_jinode_gh;
807                 }
808                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
809                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
810                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
811
812                 /* Map the extents for this journal's blocks */
813                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
814         }
815         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
816
817         /* Lookup statfs inodes here so journal recovery can use them. */
818         error = init_statfs(sdp);
819         if (error)
820                 goto fail_jinode_gh;
821
822         if (sdp->sd_lockstruct.ls_first) {
823                 unsigned int x;
824                 for (x = 0; x < sdp->sd_journals; x++) {
825                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
826
827                         if (sdp->sd_args.ar_spectator) {
828                                 error = check_journal_clean(sdp, jd, true);
829                                 if (error)
830                                         goto fail_statfs;
831                                 continue;
832                         }
833                         error = gfs2_recover_journal(jd, true);
834                         if (error) {
835                                 fs_err(sdp, "error recovering journal %u: %d\n",
836                                        x, error);
837                                 goto fail_statfs;
838                         }
839                 }
840
841                 gfs2_others_may_mount(sdp);
842         } else if (!sdp->sd_args.ar_spectator) {
843                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
844                 if (error) {
845                         fs_err(sdp, "error recovering my journal: %d\n", error);
846                         goto fail_statfs;
847                 }
848         }
849
850         sdp->sd_log_idle = 1;
851         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
852         gfs2_glock_dq_uninit(&ji_gh);
853         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
854         return 0;
855
856 fail_statfs:
857         uninit_statfs(sdp);
858 fail_jinode_gh:
859         /* A withdraw may have done dq/uninit so now we need to check it */
860         if (!sdp->sd_args.ar_spectator &&
861             gfs2_holder_initialized(&sdp->sd_jinode_gh))
862                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
863 fail_journal_gh:
864         if (!sdp->sd_args.ar_spectator &&
865             gfs2_holder_initialized(&sdp->sd_journal_gh))
866                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
867 fail_jindex:
868         gfs2_jindex_free(sdp);
869         if (gfs2_holder_initialized(&ji_gh))
870                 gfs2_glock_dq_uninit(&ji_gh);
871 fail:
872         iput(sdp->sd_jindex);
873         return error;
874 }
875
876 static struct lock_class_key gfs2_quota_imutex_key;
877
878 static int init_inodes(struct gfs2_sbd *sdp, int undo)
879 {
880         int error = 0;
881         struct inode *master = d_inode(sdp->sd_master_dir);
882
883         if (undo)
884                 goto fail_qinode;
885
886         error = init_journal(sdp, undo);
887         complete_all(&sdp->sd_journal_ready);
888         if (error)
889                 goto fail;
890
891         /* Read in the resource index inode */
892         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
893         if (IS_ERR(sdp->sd_rindex)) {
894                 error = PTR_ERR(sdp->sd_rindex);
895                 fs_err(sdp, "can't get resource index inode: %d\n", error);
896                 goto fail_journal;
897         }
898         sdp->sd_rindex_uptodate = 0;
899
900         /* Read in the quota inode */
901         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
902         if (IS_ERR(sdp->sd_quota_inode)) {
903                 error = PTR_ERR(sdp->sd_quota_inode);
904                 fs_err(sdp, "can't get quota file inode: %d\n", error);
905                 goto fail_rindex;
906         }
907         /*
908          * i_rwsem on quota files is special. Since this inode is hidden system
909          * file, we are safe to define locking ourselves.
910          */
911         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
912                           &gfs2_quota_imutex_key);
913
914         error = gfs2_rindex_update(sdp);
915         if (error)
916                 goto fail_qinode;
917
918         return 0;
919
920 fail_qinode:
921         iput(sdp->sd_quota_inode);
922 fail_rindex:
923         gfs2_clear_rgrpd(sdp);
924         iput(sdp->sd_rindex);
925 fail_journal:
926         init_journal(sdp, UNDO);
927 fail:
928         return error;
929 }
930
931 static int init_per_node(struct gfs2_sbd *sdp, int undo)
932 {
933         struct inode *pn = NULL;
934         char buf[30];
935         int error = 0;
936         struct gfs2_inode *ip;
937         struct inode *master = d_inode(sdp->sd_master_dir);
938
939         if (sdp->sd_args.ar_spectator)
940                 return 0;
941
942         if (undo)
943                 goto fail_qc_gh;
944
945         pn = gfs2_lookup_simple(master, "per_node");
946         if (IS_ERR(pn)) {
947                 error = PTR_ERR(pn);
948                 fs_err(sdp, "can't find per_node directory: %d\n", error);
949                 return error;
950         }
951
952         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
953         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
954         if (IS_ERR(sdp->sd_qc_inode)) {
955                 error = PTR_ERR(sdp->sd_qc_inode);
956                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
957                 goto fail_ut_i;
958         }
959
960         iput(pn);
961         pn = NULL;
962
963         ip = GFS2_I(sdp->sd_qc_inode);
964         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
965                                    &sdp->sd_qc_gh);
966         if (error) {
967                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
968                 goto fail_qc_i;
969         }
970
971         return 0;
972
973 fail_qc_gh:
974         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
975 fail_qc_i:
976         iput(sdp->sd_qc_inode);
977 fail_ut_i:
978         iput(pn);
979         return error;
980 }
981
982 static const match_table_t nolock_tokens = {
983         { Opt_jid, "jid=%d", },
984         { Opt_err, NULL },
985 };
986
987 static const struct lm_lockops nolock_ops = {
988         .lm_proto_name = "lock_nolock",
989         .lm_put_lock = gfs2_glock_free,
990         .lm_tokens = &nolock_tokens,
991 };
992
993 /**
994  * gfs2_lm_mount - mount a locking protocol
995  * @sdp: the filesystem
996  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
997  *
998  * Returns: errno
999  */
1000
1001 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1002 {
1003         const struct lm_lockops *lm;
1004         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1005         struct gfs2_args *args = &sdp->sd_args;
1006         const char *proto = sdp->sd_proto_name;
1007         const char *table = sdp->sd_table_name;
1008         char *o, *options;
1009         int ret;
1010
1011         if (!strcmp("lock_nolock", proto)) {
1012                 lm = &nolock_ops;
1013                 sdp->sd_args.ar_localflocks = 1;
1014 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1015         } else if (!strcmp("lock_dlm", proto)) {
1016                 lm = &gfs2_dlm_ops;
1017 #endif
1018         } else {
1019                 pr_info("can't find protocol %s\n", proto);
1020                 return -ENOENT;
1021         }
1022
1023         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1024
1025         ls->ls_ops = lm;
1026         ls->ls_first = 1;
1027
1028         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1029                 substring_t tmp[MAX_OPT_ARGS];
1030                 int token, option;
1031
1032                 if (!o || !*o)
1033                         continue;
1034
1035                 token = match_token(o, *lm->lm_tokens, tmp);
1036                 switch (token) {
1037                 case Opt_jid:
1038                         ret = match_int(&tmp[0], &option);
1039                         if (ret || option < 0) 
1040                                 goto hostdata_error;
1041                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1042                                 ls->ls_jid = option;
1043                         break;
1044                 case Opt_id:
1045                 case Opt_nodir:
1046                         /* Obsolete, but left for backward compat purposes */
1047                         break;
1048                 case Opt_first:
1049                         ret = match_int(&tmp[0], &option);
1050                         if (ret || (option != 0 && option != 1))
1051                                 goto hostdata_error;
1052                         ls->ls_first = option;
1053                         break;
1054                 case Opt_err:
1055                 default:
1056 hostdata_error:
1057                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1058                         return -EINVAL;
1059                 }
1060         }
1061
1062         if (lm->lm_mount == NULL) {
1063                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1064                 complete_all(&sdp->sd_locking_init);
1065                 return 0;
1066         }
1067         ret = lm->lm_mount(sdp, table);
1068         if (ret == 0)
1069                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1070                         sdp->sd_sb.sb_fs_format);
1071         complete_all(&sdp->sd_locking_init);
1072         return ret;
1073 }
1074
1075 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1076 {
1077         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1078         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1079                 lm->lm_unmount(sdp);
1080 }
1081
1082 static int wait_on_journal(struct gfs2_sbd *sdp)
1083 {
1084         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1085                 return 0;
1086
1087         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1088                 ? -EINTR : 0;
1089 }
1090
1091 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1092 {
1093         struct super_block *sb = sdp->sd_vfs;
1094         char ro[20];
1095         char spectator[20];
1096         char *envp[] = { ro, spectator, NULL };
1097         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1098         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1099         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1100 }
1101
1102 static int init_threads(struct gfs2_sbd *sdp)
1103 {
1104         struct task_struct *p;
1105         int error = 0;
1106
1107         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1108         if (IS_ERR(p)) {
1109                 error = PTR_ERR(p);
1110                 fs_err(sdp, "can't start logd thread: %d\n", error);
1111                 return error;
1112         }
1113         sdp->sd_logd_process = p;
1114
1115         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1116         if (IS_ERR(p)) {
1117                 error = PTR_ERR(p);
1118                 fs_err(sdp, "can't start quotad thread: %d\n", error);
1119                 goto fail;
1120         }
1121         sdp->sd_quotad_process = p;
1122         return 0;
1123
1124 fail:
1125         kthread_stop(sdp->sd_logd_process);
1126         sdp->sd_logd_process = NULL;
1127         return error;
1128 }
1129
1130 /**
1131  * gfs2_fill_super - Read in superblock
1132  * @sb: The VFS superblock
1133  * @fc: Mount options and flags
1134  *
1135  * Returns: -errno
1136  */
1137 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1138 {
1139         struct gfs2_args *args = fc->fs_private;
1140         int silent = fc->sb_flags & SB_SILENT;
1141         struct gfs2_sbd *sdp;
1142         struct gfs2_holder mount_gh;
1143         struct gfs2_holder freeze_gh;
1144         int error;
1145
1146         sdp = init_sbd(sb);
1147         if (!sdp) {
1148                 pr_warn("can't alloc struct gfs2_sbd\n");
1149                 return -ENOMEM;
1150         }
1151         sdp->sd_args = *args;
1152
1153         if (sdp->sd_args.ar_spectator) {
1154                 sb->s_flags |= SB_RDONLY;
1155                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1156         }
1157         if (sdp->sd_args.ar_posix_acl)
1158                 sb->s_flags |= SB_POSIXACL;
1159         if (sdp->sd_args.ar_nobarrier)
1160                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1161
1162         sb->s_flags |= SB_NOSEC;
1163         sb->s_magic = GFS2_MAGIC;
1164         sb->s_op = &gfs2_super_ops;
1165         sb->s_d_op = &gfs2_dops;
1166         sb->s_export_op = &gfs2_export_ops;
1167         sb->s_qcop = &gfs2_quotactl_ops;
1168         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1169         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1170         sb->s_time_gran = 1;
1171         sb->s_maxbytes = MAX_LFS_FILESIZE;
1172
1173         /* Set up the buffer cache and fill in some fake block size values
1174            to allow us to read-in the on-disk superblock. */
1175         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1176         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1177         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1178                                GFS2_BASIC_BLOCK_SHIFT;
1179         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1180
1181         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1182         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1183         if (sdp->sd_args.ar_statfs_quantum) {
1184                 sdp->sd_tune.gt_statfs_slow = 0;
1185                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1186         } else {
1187                 sdp->sd_tune.gt_statfs_slow = 1;
1188                 sdp->sd_tune.gt_statfs_quantum = 30;
1189         }
1190
1191         error = init_names(sdp, silent);
1192         if (error)
1193                 goto fail_free;
1194
1195         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1196
1197         sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s",
1198                         WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname);
1199         error = -ENOMEM;
1200         if (!sdp->sd_delete_wq)
1201                 goto fail_free;
1202
1203         error = gfs2_sys_fs_add(sdp);
1204         if (error)
1205                 goto fail_delete_wq;
1206
1207         gfs2_create_debugfs_file(sdp);
1208
1209         error = gfs2_lm_mount(sdp, silent);
1210         if (error)
1211                 goto fail_debug;
1212
1213         error = init_locking(sdp, &mount_gh, DO);
1214         if (error)
1215                 goto fail_lm;
1216
1217         error = init_sb(sdp, silent);
1218         if (error)
1219                 goto fail_locking;
1220
1221         /* Turn rgrplvb on by default if fs format is recent enough */
1222         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1223                 sdp->sd_args.ar_rgrplvb = 1;
1224
1225         error = wait_on_journal(sdp);
1226         if (error)
1227                 goto fail_sb;
1228
1229         /*
1230          * If user space has failed to join the cluster or some similar
1231          * failure has occurred, then the journal id will contain a
1232          * negative (error) number. This will then be returned to the
1233          * caller (of the mount syscall). We do this even for spectator
1234          * mounts (which just write a jid of 0 to indicate "ok" even though
1235          * the jid is unused in the spectator case)
1236          */
1237         if (sdp->sd_lockstruct.ls_jid < 0) {
1238                 error = sdp->sd_lockstruct.ls_jid;
1239                 sdp->sd_lockstruct.ls_jid = 0;
1240                 goto fail_sb;
1241         }
1242
1243         if (sdp->sd_args.ar_spectator)
1244                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1245                          sdp->sd_table_name);
1246         else
1247                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1248                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1249
1250         error = init_inodes(sdp, DO);
1251         if (error)
1252                 goto fail_sb;
1253
1254         error = init_per_node(sdp, DO);
1255         if (error)
1256                 goto fail_inodes;
1257
1258         error = gfs2_statfs_init(sdp);
1259         if (error) {
1260                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1261                 goto fail_per_node;
1262         }
1263
1264         if (!sb_rdonly(sb)) {
1265                 error = init_threads(sdp);
1266                 if (error) {
1267                         gfs2_withdraw_delayed(sdp);
1268                         goto fail_per_node;
1269                 }
1270         }
1271
1272         error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1273         if (error)
1274                 goto fail_per_node;
1275
1276         if (!sb_rdonly(sb))
1277                 error = gfs2_make_fs_rw(sdp);
1278
1279         gfs2_freeze_unlock(&freeze_gh);
1280         if (error) {
1281                 if (sdp->sd_quotad_process)
1282                         kthread_stop(sdp->sd_quotad_process);
1283                 sdp->sd_quotad_process = NULL;
1284                 if (sdp->sd_logd_process)
1285                         kthread_stop(sdp->sd_logd_process);
1286                 sdp->sd_logd_process = NULL;
1287                 fs_err(sdp, "can't make FS RW: %d\n", error);
1288                 goto fail_per_node;
1289         }
1290         gfs2_glock_dq_uninit(&mount_gh);
1291         gfs2_online_uevent(sdp);
1292         return 0;
1293
1294 fail_per_node:
1295         init_per_node(sdp, UNDO);
1296 fail_inodes:
1297         init_inodes(sdp, UNDO);
1298 fail_sb:
1299         if (sdp->sd_root_dir)
1300                 dput(sdp->sd_root_dir);
1301         if (sdp->sd_master_dir)
1302                 dput(sdp->sd_master_dir);
1303         if (sb->s_root)
1304                 dput(sb->s_root);
1305         sb->s_root = NULL;
1306 fail_locking:
1307         init_locking(sdp, &mount_gh, UNDO);
1308 fail_lm:
1309         complete_all(&sdp->sd_journal_ready);
1310         gfs2_gl_hash_clear(sdp);
1311         gfs2_lm_unmount(sdp);
1312 fail_debug:
1313         gfs2_delete_debugfs_file(sdp);
1314         gfs2_sys_fs_del(sdp);
1315 fail_delete_wq:
1316         destroy_workqueue(sdp->sd_delete_wq);
1317 fail_free:
1318         free_sbd(sdp);
1319         sb->s_fs_info = NULL;
1320         return error;
1321 }
1322
1323 /**
1324  * gfs2_get_tree - Get the GFS2 superblock and root directory
1325  * @fc: The filesystem context
1326  *
1327  * Returns: 0 or -errno on error
1328  */
1329 static int gfs2_get_tree(struct fs_context *fc)
1330 {
1331         struct gfs2_args *args = fc->fs_private;
1332         struct gfs2_sbd *sdp;
1333         int error;
1334
1335         error = get_tree_bdev(fc, gfs2_fill_super);
1336         if (error)
1337                 return error;
1338
1339         sdp = fc->root->d_sb->s_fs_info;
1340         dput(fc->root);
1341         if (args->ar_meta)
1342                 fc->root = dget(sdp->sd_master_dir);
1343         else
1344                 fc->root = dget(sdp->sd_root_dir);
1345         return 0;
1346 }
1347
1348 static void gfs2_fc_free(struct fs_context *fc)
1349 {
1350         struct gfs2_args *args = fc->fs_private;
1351
1352         kfree(args);
1353 }
1354
1355 enum gfs2_param {
1356         Opt_lockproto,
1357         Opt_locktable,
1358         Opt_hostdata,
1359         Opt_spectator,
1360         Opt_ignore_local_fs,
1361         Opt_localflocks,
1362         Opt_localcaching,
1363         Opt_debug,
1364         Opt_upgrade,
1365         Opt_acl,
1366         Opt_quota,
1367         Opt_quota_flag,
1368         Opt_suiddir,
1369         Opt_data,
1370         Opt_meta,
1371         Opt_discard,
1372         Opt_commit,
1373         Opt_errors,
1374         Opt_statfs_quantum,
1375         Opt_statfs_percent,
1376         Opt_quota_quantum,
1377         Opt_barrier,
1378         Opt_rgrplvb,
1379         Opt_loccookie,
1380 };
1381
1382 static const struct constant_table gfs2_param_quota[] = {
1383         {"off",        GFS2_QUOTA_OFF},
1384         {"account",    GFS2_QUOTA_ACCOUNT},
1385         {"on",         GFS2_QUOTA_ON},
1386         {}
1387 };
1388
1389 enum opt_data {
1390         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1391         Opt_data_ordered   = GFS2_DATA_ORDERED,
1392 };
1393
1394 static const struct constant_table gfs2_param_data[] = {
1395         {"writeback",  Opt_data_writeback },
1396         {"ordered",    Opt_data_ordered },
1397         {}
1398 };
1399
1400 enum opt_errors {
1401         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1402         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1403 };
1404
1405 static const struct constant_table gfs2_param_errors[] = {
1406         {"withdraw",   Opt_errors_withdraw },
1407         {"panic",      Opt_errors_panic },
1408         {}
1409 };
1410
1411 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1412         fsparam_string ("lockproto",          Opt_lockproto),
1413         fsparam_string ("locktable",          Opt_locktable),
1414         fsparam_string ("hostdata",           Opt_hostdata),
1415         fsparam_flag   ("spectator",          Opt_spectator),
1416         fsparam_flag   ("norecovery",         Opt_spectator),
1417         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1418         fsparam_flag   ("localflocks",        Opt_localflocks),
1419         fsparam_flag   ("localcaching",       Opt_localcaching),
1420         fsparam_flag_no("debug",              Opt_debug),
1421         fsparam_flag   ("upgrade",            Opt_upgrade),
1422         fsparam_flag_no("acl",                Opt_acl),
1423         fsparam_flag_no("suiddir",            Opt_suiddir),
1424         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1425         fsparam_flag   ("meta",               Opt_meta),
1426         fsparam_flag_no("discard",            Opt_discard),
1427         fsparam_s32    ("commit",             Opt_commit),
1428         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1429         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1430         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1431         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1432         fsparam_flag_no("barrier",            Opt_barrier),
1433         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1434         fsparam_flag_no("loccookie",          Opt_loccookie),
1435         /* quota can be a flag or an enum so it gets special treatment */
1436         fsparam_flag_no("quota",              Opt_quota_flag),
1437         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1438         {}
1439 };
1440
1441 /* Parse a single mount parameter */
1442 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1443 {
1444         struct gfs2_args *args = fc->fs_private;
1445         struct fs_parse_result result;
1446         int o;
1447
1448         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1449         if (o < 0)
1450                 return o;
1451
1452         switch (o) {
1453         case Opt_lockproto:
1454                 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1455                 break;
1456         case Opt_locktable:
1457                 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1458                 break;
1459         case Opt_hostdata:
1460                 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1461                 break;
1462         case Opt_spectator:
1463                 args->ar_spectator = 1;
1464                 break;
1465         case Opt_ignore_local_fs:
1466                 /* Retained for backwards compat only */
1467                 break;
1468         case Opt_localflocks:
1469                 args->ar_localflocks = 1;
1470                 break;
1471         case Opt_localcaching:
1472                 /* Retained for backwards compat only */
1473                 break;
1474         case Opt_debug:
1475                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1476                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1477                 args->ar_debug = result.boolean;
1478                 break;
1479         case Opt_upgrade:
1480                 /* Retained for backwards compat only */
1481                 break;
1482         case Opt_acl:
1483                 args->ar_posix_acl = result.boolean;
1484                 break;
1485         case Opt_quota_flag:
1486                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1487                 break;
1488         case Opt_quota:
1489                 args->ar_quota = result.int_32;
1490                 break;
1491         case Opt_suiddir:
1492                 args->ar_suiddir = result.boolean;
1493                 break;
1494         case Opt_data:
1495                 /* The uint_32 result maps directly to GFS2_DATA_* */
1496                 args->ar_data = result.uint_32;
1497                 break;
1498         case Opt_meta:
1499                 args->ar_meta = 1;
1500                 break;
1501         case Opt_discard:
1502                 args->ar_discard = result.boolean;
1503                 break;
1504         case Opt_commit:
1505                 if (result.int_32 <= 0)
1506                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1507                 args->ar_commit = result.int_32;
1508                 break;
1509         case Opt_statfs_quantum:
1510                 if (result.int_32 < 0)
1511                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1512                 args->ar_statfs_quantum = result.int_32;
1513                 break;
1514         case Opt_quota_quantum:
1515                 if (result.int_32 <= 0)
1516                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1517                 args->ar_quota_quantum = result.int_32;
1518                 break;
1519         case Opt_statfs_percent:
1520                 if (result.int_32 < 0 || result.int_32 > 100)
1521                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1522                 args->ar_statfs_percent = result.int_32;
1523                 break;
1524         case Opt_errors:
1525                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1526                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1527                 args->ar_errors = result.uint_32;
1528                 break;
1529         case Opt_barrier:
1530                 args->ar_nobarrier = result.boolean;
1531                 break;
1532         case Opt_rgrplvb:
1533                 args->ar_rgrplvb = result.boolean;
1534                 args->ar_got_rgrplvb = 1;
1535                 break;
1536         case Opt_loccookie:
1537                 args->ar_loccookie = result.boolean;
1538                 break;
1539         default:
1540                 return invalfc(fc, "invalid mount option: %s", param->key);
1541         }
1542         return 0;
1543 }
1544
1545 static int gfs2_reconfigure(struct fs_context *fc)
1546 {
1547         struct super_block *sb = fc->root->d_sb;
1548         struct gfs2_sbd *sdp = sb->s_fs_info;
1549         struct gfs2_args *oldargs = &sdp->sd_args;
1550         struct gfs2_args *newargs = fc->fs_private;
1551         struct gfs2_tune *gt = &sdp->sd_tune;
1552         int error = 0;
1553
1554         sync_filesystem(sb);
1555
1556         spin_lock(&gt->gt_spin);
1557         oldargs->ar_commit = gt->gt_logd_secs;
1558         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1559         if (gt->gt_statfs_slow)
1560                 oldargs->ar_statfs_quantum = 0;
1561         else
1562                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1563         spin_unlock(&gt->gt_spin);
1564
1565         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1566                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1567                 return -EINVAL;
1568         }
1569         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1570                 errorfc(fc, "reconfiguration of lock table not allowed");
1571                 return -EINVAL;
1572         }
1573         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1574                 errorfc(fc, "reconfiguration of host data not allowed");
1575                 return -EINVAL;
1576         }
1577         if (newargs->ar_spectator != oldargs->ar_spectator) {
1578                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1579                 return -EINVAL;
1580         }
1581         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1582                 errorfc(fc, "reconfiguration of localflocks not allowed");
1583                 return -EINVAL;
1584         }
1585         if (newargs->ar_meta != oldargs->ar_meta) {
1586                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1587                 return -EINVAL;
1588         }
1589         if (oldargs->ar_spectator)
1590                 fc->sb_flags |= SB_RDONLY;
1591
1592         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1593                 struct gfs2_holder freeze_gh;
1594
1595                 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1596                 if (error)
1597                         return -EINVAL;
1598
1599                 if (fc->sb_flags & SB_RDONLY) {
1600                         gfs2_make_fs_ro(sdp);
1601                 } else {
1602                         error = gfs2_make_fs_rw(sdp);
1603                         if (error)
1604                                 errorfc(fc, "unable to remount read-write");
1605                 }
1606                 gfs2_freeze_unlock(&freeze_gh);
1607         }
1608         sdp->sd_args = *newargs;
1609
1610         if (sdp->sd_args.ar_posix_acl)
1611                 sb->s_flags |= SB_POSIXACL;
1612         else
1613                 sb->s_flags &= ~SB_POSIXACL;
1614         if (sdp->sd_args.ar_nobarrier)
1615                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1616         else
1617                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1618         spin_lock(&gt->gt_spin);
1619         gt->gt_logd_secs = newargs->ar_commit;
1620         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1621         if (newargs->ar_statfs_quantum) {
1622                 gt->gt_statfs_slow = 0;
1623                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1624         }
1625         else {
1626                 gt->gt_statfs_slow = 1;
1627                 gt->gt_statfs_quantum = 30;
1628         }
1629         spin_unlock(&gt->gt_spin);
1630
1631         gfs2_online_uevent(sdp);
1632         return error;
1633 }
1634
1635 static const struct fs_context_operations gfs2_context_ops = {
1636         .free        = gfs2_fc_free,
1637         .parse_param = gfs2_parse_param,
1638         .get_tree    = gfs2_get_tree,
1639         .reconfigure = gfs2_reconfigure,
1640 };
1641
1642 /* Set up the filesystem mount context */
1643 static int gfs2_init_fs_context(struct fs_context *fc)
1644 {
1645         struct gfs2_args *args;
1646
1647         args = kmalloc(sizeof(*args), GFP_KERNEL);
1648         if (args == NULL)
1649                 return -ENOMEM;
1650
1651         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1652                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1653
1654                 *args = sdp->sd_args;
1655         } else {
1656                 memset(args, 0, sizeof(*args));
1657                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1658                 args->ar_data = GFS2_DATA_DEFAULT;
1659                 args->ar_commit = 30;
1660                 args->ar_statfs_quantum = 30;
1661                 args->ar_quota_quantum = 60;
1662                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1663         }
1664         fc->fs_private = args;
1665         fc->ops = &gfs2_context_ops;
1666         return 0;
1667 }
1668
1669 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1670 {
1671         return -EINVAL;
1672 }
1673
1674 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1675 {
1676         return (fc->sget_key == s->s_bdev);
1677 }
1678
1679 static int gfs2_meta_get_tree(struct fs_context *fc)
1680 {
1681         struct super_block *s;
1682         struct gfs2_sbd *sdp;
1683         struct path path;
1684         int error;
1685
1686         if (!fc->source || !*fc->source)
1687                 return -EINVAL;
1688
1689         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1690         if (error) {
1691                 pr_warn("path_lookup on %s returned error %d\n",
1692                         fc->source, error);
1693                 return error;
1694         }
1695         fc->fs_type = &gfs2_fs_type;
1696         fc->sget_key = path.dentry->d_sb->s_bdev;
1697         s = sget_fc(fc, test_meta_super, set_meta_super);
1698         path_put(&path);
1699         if (IS_ERR(s)) {
1700                 pr_warn("gfs2 mount does not exist\n");
1701                 return PTR_ERR(s);
1702         }
1703         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1704                 deactivate_locked_super(s);
1705                 return -EBUSY;
1706         }
1707         sdp = s->s_fs_info;
1708         fc->root = dget(sdp->sd_master_dir);
1709         return 0;
1710 }
1711
1712 static const struct fs_context_operations gfs2_meta_context_ops = {
1713         .free        = gfs2_fc_free,
1714         .get_tree    = gfs2_meta_get_tree,
1715 };
1716
1717 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1718 {
1719         int ret = gfs2_init_fs_context(fc);
1720
1721         if (ret)
1722                 return ret;
1723
1724         fc->ops = &gfs2_meta_context_ops;
1725         return 0;
1726 }
1727
1728 /**
1729  * gfs2_evict_inodes - evict inodes cooperatively
1730  * @sb: the superblock
1731  *
1732  * When evicting an inode with a zero link count, we are trying to upgrade the
1733  * inode's iopen glock from SH to EX mode in order to determine if we can
1734  * delete the inode.  The other nodes are supposed to evict the inode from
1735  * their caches if they can, and to poke the inode's inode glock if they cannot
1736  * do so.  Either behavior allows gfs2_upgrade_iopen_glock() to proceed
1737  * quickly, but if the other nodes are not cooperating, the lock upgrading
1738  * attempt will time out.  Since inodes are evicted sequentially, this can add
1739  * up quickly.
1740  *
1741  * Function evict_inodes() tries to keep the s_inode_list_lock list locked over
1742  * a long time, which prevents other inodes from being evicted concurrently.
1743  * This precludes the cooperative behavior we are looking for.  This special
1744  * version of evict_inodes() avoids that.
1745  *
1746  * Modeled after drop_pagecache_sb().
1747  */
1748 static void gfs2_evict_inodes(struct super_block *sb)
1749 {
1750         struct inode *inode, *toput_inode = NULL;
1751         struct gfs2_sbd *sdp = sb->s_fs_info;
1752
1753         set_bit(SDF_EVICTING, &sdp->sd_flags);
1754
1755         spin_lock(&sb->s_inode_list_lock);
1756         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1757                 spin_lock(&inode->i_lock);
1758                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) &&
1759                     !need_resched()) {
1760                         spin_unlock(&inode->i_lock);
1761                         continue;
1762                 }
1763                 atomic_inc(&inode->i_count);
1764                 spin_unlock(&inode->i_lock);
1765                 spin_unlock(&sb->s_inode_list_lock);
1766
1767                 iput(toput_inode);
1768                 toput_inode = inode;
1769
1770                 cond_resched();
1771                 spin_lock(&sb->s_inode_list_lock);
1772         }
1773         spin_unlock(&sb->s_inode_list_lock);
1774         iput(toput_inode);
1775 }
1776
1777 static void gfs2_kill_sb(struct super_block *sb)
1778 {
1779         struct gfs2_sbd *sdp = sb->s_fs_info;
1780
1781         if (sdp == NULL) {
1782                 kill_block_super(sb);
1783                 return;
1784         }
1785
1786         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1787         dput(sdp->sd_root_dir);
1788         dput(sdp->sd_master_dir);
1789         sdp->sd_root_dir = NULL;
1790         sdp->sd_master_dir = NULL;
1791         shrink_dcache_sb(sb);
1792
1793         gfs2_evict_inodes(sb);
1794
1795         /*
1796          * Flush and then drain the delete workqueue here (via
1797          * destroy_workqueue()) to ensure that any delete work that
1798          * may be running will also see the SDF_DEACTIVATING flag.
1799          */
1800         set_bit(SDF_DEACTIVATING, &sdp->sd_flags);
1801         gfs2_flush_delete_work(sdp);
1802         destroy_workqueue(sdp->sd_delete_wq);
1803
1804         kill_block_super(sb);
1805 }
1806
1807 struct file_system_type gfs2_fs_type = {
1808         .name = "gfs2",
1809         .fs_flags = FS_REQUIRES_DEV,
1810         .init_fs_context = gfs2_init_fs_context,
1811         .parameters = gfs2_fs_parameters,
1812         .kill_sb = gfs2_kill_sb,
1813         .owner = THIS_MODULE,
1814 };
1815 MODULE_ALIAS_FS("gfs2");
1816
1817 struct file_system_type gfs2meta_fs_type = {
1818         .name = "gfs2meta",
1819         .fs_flags = FS_REQUIRES_DEV,
1820         .init_fs_context = gfs2_meta_init_fs_context,
1821         .owner = THIS_MODULE,
1822 };
1823 MODULE_ALIAS_FS("gfs2meta");