OSDN Git Service

perf/x86/uncore: Correct the number of CHAs on EMR
[tomoyo/tomoyo-test1.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         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 -
296                                GFS2_BASIC_BLOCK_SHIFT;
297         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
298         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
299                           sizeof(struct gfs2_dinode)) / sizeof(u64);
300         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
301                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
302         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
303                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
304         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
305         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
306         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
307         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
308         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
309                                 sizeof(struct gfs2_meta_header)) /
310                                 sizeof(struct gfs2_quota_change);
311         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
312                                      sizeof(struct gfs2_meta_header))
313                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
314
315         /*
316          * We always keep at least one block reserved for revokes in
317          * transactions.  This greatly simplifies allocating additional
318          * revoke blocks.
319          */
320         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
321
322         /* Compute maximum reservation required to add a entry to a directory */
323
324         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
325                              sdp->sd_jbsize);
326
327         ind_blocks = 0;
328         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
329                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
330                 ind_blocks += tmp_blocks;
331         }
332
333         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
334
335         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
336
337         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
338                                 sizeof(struct gfs2_dinode);
339         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
340         for (x = 2;; x++) {
341                 u64 space, d;
342                 u32 m;
343
344                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
345                 d = space;
346                 m = do_div(d, sdp->sd_inptrs);
347
348                 if (d != sdp->sd_heightsize[x - 1] || m)
349                         break;
350                 sdp->sd_heightsize[x] = space;
351         }
352         sdp->sd_max_height = x;
353         sdp->sd_heightsize[x] = ~0;
354         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
355
356         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
357                                       sizeof(struct gfs2_leaf)) /
358                                      GFS2_MIN_DIRENT_SIZE;
359         return 0;
360 }
361
362 static int init_names(struct gfs2_sbd *sdp, int silent)
363 {
364         char *proto, *table;
365         int error = 0;
366
367         proto = sdp->sd_args.ar_lockproto;
368         table = sdp->sd_args.ar_locktable;
369
370         /*  Try to autodetect  */
371
372         if (!proto[0] || !table[0]) {
373                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
374                 if (error)
375                         return error;
376
377                 if (!proto[0])
378                         proto = sdp->sd_sb.sb_lockproto;
379                 if (!table[0])
380                         table = sdp->sd_sb.sb_locktable;
381         }
382
383         if (!table[0])
384                 table = sdp->sd_vfs->s_id;
385
386         BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
387
388         strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
389         strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
390
391         table = sdp->sd_table_name;
392         while ((table = strchr(table, '/')))
393                 *table = '_';
394
395         return error;
396 }
397
398 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
399                         int undo)
400 {
401         int error = 0;
402
403         if (undo)
404                 goto fail_trans;
405
406         error = gfs2_glock_nq_num(sdp,
407                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
408                                   LM_ST_EXCLUSIVE,
409                                   LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
410                                   mount_gh);
411         if (error) {
412                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
413                 goto fail;
414         }
415
416         error = gfs2_glock_nq_num(sdp,
417                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
418                                   LM_ST_SHARED,
419                                   LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
420                                   &sdp->sd_live_gh);
421         if (error) {
422                 fs_err(sdp, "can't acquire live glock: %d\n", error);
423                 goto fail_mount;
424         }
425
426         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
427                                CREATE, &sdp->sd_rename_gl);
428         if (error) {
429                 fs_err(sdp, "can't create rename glock: %d\n", error);
430                 goto fail_live;
431         }
432
433         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
434                                CREATE, &sdp->sd_freeze_gl);
435         if (error) {
436                 fs_err(sdp, "can't create freeze glock: %d\n", error);
437                 goto fail_rename;
438         }
439
440         return 0;
441
442 fail_trans:
443         gfs2_glock_put(sdp->sd_freeze_gl);
444 fail_rename:
445         gfs2_glock_put(sdp->sd_rename_gl);
446 fail_live:
447         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
448 fail_mount:
449         gfs2_glock_dq_uninit(mount_gh);
450 fail:
451         return error;
452 }
453
454 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
455                             u64 no_addr, const char *name)
456 {
457         struct gfs2_sbd *sdp = sb->s_fs_info;
458         struct dentry *dentry;
459         struct inode *inode;
460
461         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
462                                   GFS2_BLKST_FREE /* ignore */);
463         if (IS_ERR(inode)) {
464                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
465                 return PTR_ERR(inode);
466         }
467         dentry = d_make_root(inode);
468         if (!dentry) {
469                 fs_err(sdp, "can't alloc %s dentry\n", name);
470                 return -ENOMEM;
471         }
472         *dptr = dentry;
473         return 0;
474 }
475
476 static int init_sb(struct gfs2_sbd *sdp, int silent)
477 {
478         struct super_block *sb = sdp->sd_vfs;
479         struct gfs2_holder sb_gh;
480         u64 no_addr;
481         int ret;
482
483         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
484                                 LM_ST_SHARED, 0, &sb_gh);
485         if (ret) {
486                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
487                 return ret;
488         }
489
490         ret = gfs2_read_sb(sdp, silent);
491         if (ret) {
492                 fs_err(sdp, "can't read superblock: %d\n", ret);
493                 goto out;
494         }
495
496         switch(sdp->sd_sb.sb_fs_format) {
497         case GFS2_FS_FORMAT_MAX:
498                 sb->s_xattr = gfs2_xattr_handlers_max;
499                 break;
500
501         case GFS2_FS_FORMAT_MIN:
502                 sb->s_xattr = gfs2_xattr_handlers_min;
503                 break;
504
505         default:
506                 BUG();
507         }
508
509         /* Set up the buffer cache and SB for real */
510         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
511                 ret = -EINVAL;
512                 fs_err(sdp, "FS block size (%u) is too small for device "
513                        "block size (%u)\n",
514                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
515                 goto out;
516         }
517         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
518                 ret = -EINVAL;
519                 fs_err(sdp, "FS block size (%u) is too big for machine "
520                        "page size (%u)\n",
521                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
522                 goto out;
523         }
524         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
525
526         /* Get the root inode */
527         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
528         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
529         if (ret)
530                 goto out;
531
532         /* Get the master inode */
533         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
534         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
535         if (ret) {
536                 dput(sdp->sd_root_dir);
537                 goto out;
538         }
539         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
540 out:
541         gfs2_glock_dq_uninit(&sb_gh);
542         return ret;
543 }
544
545 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
546 {
547         char *message = "FIRSTMOUNT=Done";
548         char *envp[] = { message, NULL };
549
550         fs_info(sdp, "first mount done, others may mount\n");
551
552         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
553                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
554
555         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
556 }
557
558 /**
559  * gfs2_jindex_hold - Grab a lock on the jindex
560  * @sdp: The GFS2 superblock
561  * @ji_gh: the holder for the jindex glock
562  *
563  * Returns: errno
564  */
565
566 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
567 {
568         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
569         struct qstr name;
570         char buf[20];
571         struct gfs2_jdesc *jd;
572         int error;
573
574         name.name = buf;
575
576         mutex_lock(&sdp->sd_jindex_mutex);
577
578         for (;;) {
579                 struct gfs2_inode *jip;
580
581                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
582                 if (error)
583                         break;
584
585                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
586                 name.hash = gfs2_disk_hash(name.name, name.len);
587
588                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
589                 if (error == -ENOENT) {
590                         error = 0;
591                         break;
592                 }
593
594                 gfs2_glock_dq_uninit(ji_gh);
595
596                 if (error)
597                         break;
598
599                 error = -ENOMEM;
600                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
601                 if (!jd)
602                         break;
603
604                 INIT_LIST_HEAD(&jd->extent_list);
605                 INIT_LIST_HEAD(&jd->jd_revoke_list);
606
607                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
608                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
609                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
610                         if (!jd->jd_inode)
611                                 error = -ENOENT;
612                         else
613                                 error = PTR_ERR(jd->jd_inode);
614                         kfree(jd);
615                         break;
616                 }
617
618                 d_mark_dontcache(jd->jd_inode);
619                 spin_lock(&sdp->sd_jindex_spin);
620                 jd->jd_jid = sdp->sd_journals++;
621                 jip = GFS2_I(jd->jd_inode);
622                 jd->jd_no_addr = jip->i_no_addr;
623                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
624                 spin_unlock(&sdp->sd_jindex_spin);
625         }
626
627         mutex_unlock(&sdp->sd_jindex_mutex);
628
629         return error;
630 }
631
632 /**
633  * init_statfs - look up and initialize master and local (per node) statfs inodes
634  * @sdp: The GFS2 superblock
635  *
636  * This should be called after the jindex is initialized in init_journal() and
637  * before gfs2_journal_recovery() is called because we need to be able to write
638  * to these inodes during recovery.
639  *
640  * Returns: errno
641  */
642 static int init_statfs(struct gfs2_sbd *sdp)
643 {
644         int error = 0;
645         struct inode *master = d_inode(sdp->sd_master_dir);
646         struct inode *pn = NULL;
647         char buf[30];
648         struct gfs2_jdesc *jd;
649         struct gfs2_inode *ip;
650
651         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
652         if (IS_ERR(sdp->sd_statfs_inode)) {
653                 error = PTR_ERR(sdp->sd_statfs_inode);
654                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
655                 goto out;
656         }
657         if (sdp->sd_args.ar_spectator)
658                 goto out;
659
660         pn = gfs2_lookup_simple(master, "per_node");
661         if (IS_ERR(pn)) {
662                 error = PTR_ERR(pn);
663                 fs_err(sdp, "can't find per_node directory: %d\n", error);
664                 goto put_statfs;
665         }
666
667         /* For each jid, lookup the corresponding local statfs inode in the
668          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
669         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
670                 struct local_statfs_inode *lsi =
671                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
672                 if (!lsi) {
673                         error = -ENOMEM;
674                         goto free_local;
675                 }
676                 sprintf(buf, "statfs_change%u", jd->jd_jid);
677                 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
678                 if (IS_ERR(lsi->si_sc_inode)) {
679                         error = PTR_ERR(lsi->si_sc_inode);
680                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
681                                jd->jd_jid, error);
682                         kfree(lsi);
683                         goto free_local;
684                 }
685                 lsi->si_jid = jd->jd_jid;
686                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
687                         sdp->sd_sc_inode = lsi->si_sc_inode;
688
689                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
690         }
691
692         iput(pn);
693         pn = NULL;
694         ip = GFS2_I(sdp->sd_sc_inode);
695         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
696                                    &sdp->sd_sc_gh);
697         if (error) {
698                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
699                 goto free_local;
700         }
701         /* read in the local statfs buffer - other nodes don't change it. */
702         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
703         if (error) {
704                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
705                 goto unlock_sd_gh;
706         }
707         return 0;
708
709 unlock_sd_gh:
710         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
711 free_local:
712         free_local_statfs_inodes(sdp);
713         iput(pn);
714 put_statfs:
715         iput(sdp->sd_statfs_inode);
716 out:
717         return error;
718 }
719
720 /* Uninitialize and free up memory used by the list of statfs inodes */
721 static void uninit_statfs(struct gfs2_sbd *sdp)
722 {
723         if (!sdp->sd_args.ar_spectator) {
724                 brelse(sdp->sd_sc_bh);
725                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
726                 free_local_statfs_inodes(sdp);
727         }
728         iput(sdp->sd_statfs_inode);
729 }
730
731 static int init_journal(struct gfs2_sbd *sdp, int undo)
732 {
733         struct inode *master = d_inode(sdp->sd_master_dir);
734         struct gfs2_holder ji_gh;
735         struct gfs2_inode *ip;
736         int error = 0;
737
738         gfs2_holder_mark_uninitialized(&ji_gh);
739         if (undo)
740                 goto fail_statfs;
741
742         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
743         if (IS_ERR(sdp->sd_jindex)) {
744                 fs_err(sdp, "can't lookup journal index: %d\n", error);
745                 return PTR_ERR(sdp->sd_jindex);
746         }
747
748         /* Load in the journal index special file */
749
750         error = gfs2_jindex_hold(sdp, &ji_gh);
751         if (error) {
752                 fs_err(sdp, "can't read journal index: %d\n", error);
753                 goto fail;
754         }
755
756         error = -EUSERS;
757         if (!gfs2_jindex_size(sdp)) {
758                 fs_err(sdp, "no journals!\n");
759                 goto fail_jindex;
760         }
761
762         atomic_set(&sdp->sd_log_blks_needed, 0);
763         if (sdp->sd_args.ar_spectator) {
764                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
765                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
766                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
767                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
768         } else {
769                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
770                         fs_err(sdp, "can't mount journal #%u\n",
771                                sdp->sd_lockstruct.ls_jid);
772                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
773                                gfs2_jindex_size(sdp),
774                                gfs2_jindex_size(sdp) - 1);
775                         goto fail_jindex;
776                 }
777                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
778
779                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
780                                           &gfs2_journal_glops,
781                                           LM_ST_EXCLUSIVE,
782                                           LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
783                                           &sdp->sd_journal_gh);
784                 if (error) {
785                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
786                         goto fail_jindex;
787                 }
788
789                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
790                 sdp->sd_jinode_gl = ip->i_gl;
791                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
792                                            LM_FLAG_NOEXP | GL_EXACT |
793                                            GL_NOCACHE | GL_NOPID,
794                                            &sdp->sd_jinode_gh);
795                 if (error) {
796                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
797                                error);
798                         goto fail_journal_gh;
799                 }
800
801                 error = gfs2_jdesc_check(sdp->sd_jdesc);
802                 if (error) {
803                         fs_err(sdp, "my journal (%u) is bad: %d\n",
804                                sdp->sd_jdesc->jd_jid, error);
805                         goto fail_jinode_gh;
806                 }
807                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
808                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
809                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
810
811                 /* Map the extents for this journal's blocks */
812                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
813         }
814         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
815
816         /* Lookup statfs inodes here so journal recovery can use them. */
817         error = init_statfs(sdp);
818         if (error)
819                 goto fail_jinode_gh;
820
821         if (sdp->sd_lockstruct.ls_first) {
822                 unsigned int x;
823                 for (x = 0; x < sdp->sd_journals; x++) {
824                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
825
826                         if (sdp->sd_args.ar_spectator) {
827                                 error = check_journal_clean(sdp, jd, true);
828                                 if (error)
829                                         goto fail_statfs;
830                                 continue;
831                         }
832                         error = gfs2_recover_journal(jd, true);
833                         if (error) {
834                                 fs_err(sdp, "error recovering journal %u: %d\n",
835                                        x, error);
836                                 goto fail_statfs;
837                         }
838                 }
839
840                 gfs2_others_may_mount(sdp);
841         } else if (!sdp->sd_args.ar_spectator) {
842                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
843                 if (error) {
844                         fs_err(sdp, "error recovering my journal: %d\n", error);
845                         goto fail_statfs;
846                 }
847         }
848
849         sdp->sd_log_idle = 1;
850         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
851         gfs2_glock_dq_uninit(&ji_gh);
852         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
853         return 0;
854
855 fail_statfs:
856         uninit_statfs(sdp);
857 fail_jinode_gh:
858         /* A withdraw may have done dq/uninit so now we need to check it */
859         if (!sdp->sd_args.ar_spectator &&
860             gfs2_holder_initialized(&sdp->sd_jinode_gh))
861                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
862 fail_journal_gh:
863         if (!sdp->sd_args.ar_spectator &&
864             gfs2_holder_initialized(&sdp->sd_journal_gh))
865                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
866 fail_jindex:
867         gfs2_jindex_free(sdp);
868         if (gfs2_holder_initialized(&ji_gh))
869                 gfs2_glock_dq_uninit(&ji_gh);
870 fail:
871         iput(sdp->sd_jindex);
872         return error;
873 }
874
875 static struct lock_class_key gfs2_quota_imutex_key;
876
877 static int init_inodes(struct gfs2_sbd *sdp, int undo)
878 {
879         int error = 0;
880         struct inode *master = d_inode(sdp->sd_master_dir);
881
882         if (undo)
883                 goto fail_qinode;
884
885         error = init_journal(sdp, undo);
886         complete_all(&sdp->sd_journal_ready);
887         if (error)
888                 goto fail;
889
890         /* Read in the resource index inode */
891         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
892         if (IS_ERR(sdp->sd_rindex)) {
893                 error = PTR_ERR(sdp->sd_rindex);
894                 fs_err(sdp, "can't get resource index inode: %d\n", error);
895                 goto fail_journal;
896         }
897         sdp->sd_rindex_uptodate = 0;
898
899         /* Read in the quota inode */
900         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
901         if (IS_ERR(sdp->sd_quota_inode)) {
902                 error = PTR_ERR(sdp->sd_quota_inode);
903                 fs_err(sdp, "can't get quota file inode: %d\n", error);
904                 goto fail_rindex;
905         }
906         /*
907          * i_rwsem on quota files is special. Since this inode is hidden system
908          * file, we are safe to define locking ourselves.
909          */
910         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
911                           &gfs2_quota_imutex_key);
912
913         error = gfs2_rindex_update(sdp);
914         if (error)
915                 goto fail_qinode;
916
917         return 0;
918
919 fail_qinode:
920         iput(sdp->sd_quota_inode);
921 fail_rindex:
922         gfs2_clear_rgrpd(sdp);
923         iput(sdp->sd_rindex);
924 fail_journal:
925         init_journal(sdp, UNDO);
926 fail:
927         return error;
928 }
929
930 static int init_per_node(struct gfs2_sbd *sdp, int undo)
931 {
932         struct inode *pn = NULL;
933         char buf[30];
934         int error = 0;
935         struct gfs2_inode *ip;
936         struct inode *master = d_inode(sdp->sd_master_dir);
937
938         if (sdp->sd_args.ar_spectator)
939                 return 0;
940
941         if (undo)
942                 goto fail_qc_gh;
943
944         pn = gfs2_lookup_simple(master, "per_node");
945         if (IS_ERR(pn)) {
946                 error = PTR_ERR(pn);
947                 fs_err(sdp, "can't find per_node directory: %d\n", error);
948                 return error;
949         }
950
951         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
952         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
953         if (IS_ERR(sdp->sd_qc_inode)) {
954                 error = PTR_ERR(sdp->sd_qc_inode);
955                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
956                 goto fail_ut_i;
957         }
958
959         iput(pn);
960         pn = NULL;
961
962         ip = GFS2_I(sdp->sd_qc_inode);
963         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
964                                    &sdp->sd_qc_gh);
965         if (error) {
966                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
967                 goto fail_qc_i;
968         }
969
970         return 0;
971
972 fail_qc_gh:
973         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
974 fail_qc_i:
975         iput(sdp->sd_qc_inode);
976 fail_ut_i:
977         iput(pn);
978         return error;
979 }
980
981 static const match_table_t nolock_tokens = {
982         { Opt_jid, "jid=%d", },
983         { Opt_err, NULL },
984 };
985
986 static const struct lm_lockops nolock_ops = {
987         .lm_proto_name = "lock_nolock",
988         .lm_put_lock = gfs2_glock_free,
989         .lm_tokens = &nolock_tokens,
990 };
991
992 /**
993  * gfs2_lm_mount - mount a locking protocol
994  * @sdp: the filesystem
995  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
996  *
997  * Returns: errno
998  */
999
1000 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1001 {
1002         const struct lm_lockops *lm;
1003         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1004         struct gfs2_args *args = &sdp->sd_args;
1005         const char *proto = sdp->sd_proto_name;
1006         const char *table = sdp->sd_table_name;
1007         char *o, *options;
1008         int ret;
1009
1010         if (!strcmp("lock_nolock", proto)) {
1011                 lm = &nolock_ops;
1012                 sdp->sd_args.ar_localflocks = 1;
1013 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1014         } else if (!strcmp("lock_dlm", proto)) {
1015                 lm = &gfs2_dlm_ops;
1016 #endif
1017         } else {
1018                 pr_info("can't find protocol %s\n", proto);
1019                 return -ENOENT;
1020         }
1021
1022         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1023
1024         ls->ls_ops = lm;
1025         ls->ls_first = 1;
1026
1027         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1028                 substring_t tmp[MAX_OPT_ARGS];
1029                 int token, option;
1030
1031                 if (!o || !*o)
1032                         continue;
1033
1034                 token = match_token(o, *lm->lm_tokens, tmp);
1035                 switch (token) {
1036                 case Opt_jid:
1037                         ret = match_int(&tmp[0], &option);
1038                         if (ret || option < 0) 
1039                                 goto hostdata_error;
1040                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1041                                 ls->ls_jid = option;
1042                         break;
1043                 case Opt_id:
1044                 case Opt_nodir:
1045                         /* Obsolete, but left for backward compat purposes */
1046                         break;
1047                 case Opt_first:
1048                         ret = match_int(&tmp[0], &option);
1049                         if (ret || (option != 0 && option != 1))
1050                                 goto hostdata_error;
1051                         ls->ls_first = option;
1052                         break;
1053                 case Opt_err:
1054                 default:
1055 hostdata_error:
1056                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1057                         return -EINVAL;
1058                 }
1059         }
1060
1061         if (lm->lm_mount == NULL) {
1062                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1063                 complete_all(&sdp->sd_locking_init);
1064                 return 0;
1065         }
1066         ret = lm->lm_mount(sdp, table);
1067         if (ret == 0)
1068                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1069                         sdp->sd_sb.sb_fs_format);
1070         complete_all(&sdp->sd_locking_init);
1071         return ret;
1072 }
1073
1074 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1075 {
1076         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1077         if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1078                 lm->lm_unmount(sdp);
1079 }
1080
1081 static int wait_on_journal(struct gfs2_sbd *sdp)
1082 {
1083         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1084                 return 0;
1085
1086         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1087                 ? -EINTR : 0;
1088 }
1089
1090 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1091 {
1092         struct super_block *sb = sdp->sd_vfs;
1093         char ro[20];
1094         char spectator[20];
1095         char *envp[] = { ro, spectator, NULL };
1096         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1097         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1098         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1099 }
1100
1101 static int init_threads(struct gfs2_sbd *sdp)
1102 {
1103         struct task_struct *p;
1104         int error = 0;
1105
1106         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1107         if (IS_ERR(p)) {
1108                 error = PTR_ERR(p);
1109                 fs_err(sdp, "can't start logd thread: %d\n", error);
1110                 return error;
1111         }
1112         sdp->sd_logd_process = p;
1113
1114         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1115         if (IS_ERR(p)) {
1116                 error = PTR_ERR(p);
1117                 fs_err(sdp, "can't start quotad thread: %d\n", error);
1118                 goto fail;
1119         }
1120         sdp->sd_quotad_process = p;
1121         return 0;
1122
1123 fail:
1124         kthread_stop(sdp->sd_logd_process);
1125         sdp->sd_logd_process = NULL;
1126         return error;
1127 }
1128
1129 /**
1130  * gfs2_fill_super - Read in superblock
1131  * @sb: The VFS superblock
1132  * @fc: Mount options and flags
1133  *
1134  * Returns: -errno
1135  */
1136 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1137 {
1138         struct gfs2_args *args = fc->fs_private;
1139         int silent = fc->sb_flags & SB_SILENT;
1140         struct gfs2_sbd *sdp;
1141         struct gfs2_holder mount_gh;
1142         int error;
1143
1144         sdp = init_sbd(sb);
1145         if (!sdp) {
1146                 pr_warn("can't alloc struct gfs2_sbd\n");
1147                 return -ENOMEM;
1148         }
1149         sdp->sd_args = *args;
1150
1151         if (sdp->sd_args.ar_spectator) {
1152                 sb->s_flags |= SB_RDONLY;
1153                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1154         }
1155         if (sdp->sd_args.ar_posix_acl)
1156                 sb->s_flags |= SB_POSIXACL;
1157         if (sdp->sd_args.ar_nobarrier)
1158                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1159
1160         sb->s_flags |= SB_NOSEC;
1161         sb->s_magic = GFS2_MAGIC;
1162         sb->s_op = &gfs2_super_ops;
1163         sb->s_d_op = &gfs2_dops;
1164         sb->s_export_op = &gfs2_export_ops;
1165         sb->s_qcop = &gfs2_quotactl_ops;
1166         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1167         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1168         sb->s_time_gran = 1;
1169         sb->s_maxbytes = MAX_LFS_FILESIZE;
1170
1171         /* Set up the buffer cache and fill in some fake block size values
1172            to allow us to read-in the on-disk superblock. */
1173         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1174         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1175         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1176                                GFS2_BASIC_BLOCK_SHIFT;
1177         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1178
1179         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1180         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1181         if (sdp->sd_args.ar_statfs_quantum) {
1182                 sdp->sd_tune.gt_statfs_slow = 0;
1183                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1184         } else {
1185                 sdp->sd_tune.gt_statfs_slow = 1;
1186                 sdp->sd_tune.gt_statfs_quantum = 30;
1187         }
1188
1189         error = init_names(sdp, silent);
1190         if (error)
1191                 goto fail_free;
1192
1193         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1194
1195         sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s",
1196                         WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname);
1197         error = -ENOMEM;
1198         if (!sdp->sd_delete_wq)
1199                 goto fail_free;
1200
1201         error = gfs2_sys_fs_add(sdp);
1202         if (error)
1203                 goto fail_delete_wq;
1204
1205         gfs2_create_debugfs_file(sdp);
1206
1207         error = gfs2_lm_mount(sdp, silent);
1208         if (error)
1209                 goto fail_debug;
1210
1211         error = init_locking(sdp, &mount_gh, DO);
1212         if (error)
1213                 goto fail_lm;
1214
1215         error = init_sb(sdp, silent);
1216         if (error)
1217                 goto fail_locking;
1218
1219         /* Turn rgrplvb on by default if fs format is recent enough */
1220         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1221                 sdp->sd_args.ar_rgrplvb = 1;
1222
1223         error = wait_on_journal(sdp);
1224         if (error)
1225                 goto fail_sb;
1226
1227         /*
1228          * If user space has failed to join the cluster or some similar
1229          * failure has occurred, then the journal id will contain a
1230          * negative (error) number. This will then be returned to the
1231          * caller (of the mount syscall). We do this even for spectator
1232          * mounts (which just write a jid of 0 to indicate "ok" even though
1233          * the jid is unused in the spectator case)
1234          */
1235         if (sdp->sd_lockstruct.ls_jid < 0) {
1236                 error = sdp->sd_lockstruct.ls_jid;
1237                 sdp->sd_lockstruct.ls_jid = 0;
1238                 goto fail_sb;
1239         }
1240
1241         if (sdp->sd_args.ar_spectator)
1242                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1243                          sdp->sd_table_name);
1244         else
1245                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1246                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1247
1248         error = init_inodes(sdp, DO);
1249         if (error)
1250                 goto fail_sb;
1251
1252         error = init_per_node(sdp, DO);
1253         if (error)
1254                 goto fail_inodes;
1255
1256         error = gfs2_statfs_init(sdp);
1257         if (error) {
1258                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1259                 goto fail_per_node;
1260         }
1261
1262         if (!sb_rdonly(sb)) {
1263                 error = init_threads(sdp);
1264                 if (error) {
1265                         gfs2_withdraw_delayed(sdp);
1266                         goto fail_per_node;
1267                 }
1268         }
1269
1270         error = gfs2_freeze_lock_shared(sdp);
1271         if (error)
1272                 goto fail_per_node;
1273
1274         if (!sb_rdonly(sb))
1275                 error = gfs2_make_fs_rw(sdp);
1276
1277         if (error) {
1278                 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
1279                 if (sdp->sd_quotad_process)
1280                         kthread_stop(sdp->sd_quotad_process);
1281                 sdp->sd_quotad_process = NULL;
1282                 if (sdp->sd_logd_process)
1283                         kthread_stop(sdp->sd_logd_process);
1284                 sdp->sd_logd_process = NULL;
1285                 fs_err(sdp, "can't make FS RW: %d\n", error);
1286                 goto fail_per_node;
1287         }
1288         gfs2_glock_dq_uninit(&mount_gh);
1289         gfs2_online_uevent(sdp);
1290         return 0;
1291
1292 fail_per_node:
1293         init_per_node(sdp, UNDO);
1294 fail_inodes:
1295         init_inodes(sdp, UNDO);
1296 fail_sb:
1297         if (sdp->sd_root_dir)
1298                 dput(sdp->sd_root_dir);
1299         if (sdp->sd_master_dir)
1300                 dput(sdp->sd_master_dir);
1301         if (sb->s_root)
1302                 dput(sb->s_root);
1303         sb->s_root = NULL;
1304 fail_locking:
1305         init_locking(sdp, &mount_gh, UNDO);
1306 fail_lm:
1307         complete_all(&sdp->sd_journal_ready);
1308         gfs2_gl_hash_clear(sdp);
1309         gfs2_lm_unmount(sdp);
1310 fail_debug:
1311         gfs2_delete_debugfs_file(sdp);
1312         gfs2_sys_fs_del(sdp);
1313 fail_delete_wq:
1314         destroy_workqueue(sdp->sd_delete_wq);
1315 fail_free:
1316         free_sbd(sdp);
1317         sb->s_fs_info = NULL;
1318         return error;
1319 }
1320
1321 /**
1322  * gfs2_get_tree - Get the GFS2 superblock and root directory
1323  * @fc: The filesystem context
1324  *
1325  * Returns: 0 or -errno on error
1326  */
1327 static int gfs2_get_tree(struct fs_context *fc)
1328 {
1329         struct gfs2_args *args = fc->fs_private;
1330         struct gfs2_sbd *sdp;
1331         int error;
1332
1333         error = get_tree_bdev(fc, gfs2_fill_super);
1334         if (error)
1335                 return error;
1336
1337         sdp = fc->root->d_sb->s_fs_info;
1338         dput(fc->root);
1339         if (args->ar_meta)
1340                 fc->root = dget(sdp->sd_master_dir);
1341         else
1342                 fc->root = dget(sdp->sd_root_dir);
1343         return 0;
1344 }
1345
1346 static void gfs2_fc_free(struct fs_context *fc)
1347 {
1348         struct gfs2_args *args = fc->fs_private;
1349
1350         kfree(args);
1351 }
1352
1353 enum gfs2_param {
1354         Opt_lockproto,
1355         Opt_locktable,
1356         Opt_hostdata,
1357         Opt_spectator,
1358         Opt_ignore_local_fs,
1359         Opt_localflocks,
1360         Opt_localcaching,
1361         Opt_debug,
1362         Opt_upgrade,
1363         Opt_acl,
1364         Opt_quota,
1365         Opt_quota_flag,
1366         Opt_suiddir,
1367         Opt_data,
1368         Opt_meta,
1369         Opt_discard,
1370         Opt_commit,
1371         Opt_errors,
1372         Opt_statfs_quantum,
1373         Opt_statfs_percent,
1374         Opt_quota_quantum,
1375         Opt_barrier,
1376         Opt_rgrplvb,
1377         Opt_loccookie,
1378 };
1379
1380 static const struct constant_table gfs2_param_quota[] = {
1381         {"off",        GFS2_QUOTA_OFF},
1382         {"account",    GFS2_QUOTA_ACCOUNT},
1383         {"on",         GFS2_QUOTA_ON},
1384         {}
1385 };
1386
1387 enum opt_data {
1388         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1389         Opt_data_ordered   = GFS2_DATA_ORDERED,
1390 };
1391
1392 static const struct constant_table gfs2_param_data[] = {
1393         {"writeback",  Opt_data_writeback },
1394         {"ordered",    Opt_data_ordered },
1395         {}
1396 };
1397
1398 enum opt_errors {
1399         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1400         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1401 };
1402
1403 static const struct constant_table gfs2_param_errors[] = {
1404         {"withdraw",   Opt_errors_withdraw },
1405         {"panic",      Opt_errors_panic },
1406         {}
1407 };
1408
1409 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1410         fsparam_string ("lockproto",          Opt_lockproto),
1411         fsparam_string ("locktable",          Opt_locktable),
1412         fsparam_string ("hostdata",           Opt_hostdata),
1413         fsparam_flag   ("spectator",          Opt_spectator),
1414         fsparam_flag   ("norecovery",         Opt_spectator),
1415         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1416         fsparam_flag   ("localflocks",        Opt_localflocks),
1417         fsparam_flag   ("localcaching",       Opt_localcaching),
1418         fsparam_flag_no("debug",              Opt_debug),
1419         fsparam_flag   ("upgrade",            Opt_upgrade),
1420         fsparam_flag_no("acl",                Opt_acl),
1421         fsparam_flag_no("suiddir",            Opt_suiddir),
1422         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1423         fsparam_flag   ("meta",               Opt_meta),
1424         fsparam_flag_no("discard",            Opt_discard),
1425         fsparam_s32    ("commit",             Opt_commit),
1426         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1427         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1428         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1429         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1430         fsparam_flag_no("barrier",            Opt_barrier),
1431         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1432         fsparam_flag_no("loccookie",          Opt_loccookie),
1433         /* quota can be a flag or an enum so it gets special treatment */
1434         fsparam_flag_no("quota",              Opt_quota_flag),
1435         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1436         {}
1437 };
1438
1439 /* Parse a single mount parameter */
1440 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1441 {
1442         struct gfs2_args *args = fc->fs_private;
1443         struct fs_parse_result result;
1444         int o;
1445
1446         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1447         if (o < 0)
1448                 return o;
1449
1450         switch (o) {
1451         case Opt_lockproto:
1452                 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1453                 break;
1454         case Opt_locktable:
1455                 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1456                 break;
1457         case Opt_hostdata:
1458                 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1459                 break;
1460         case Opt_spectator:
1461                 args->ar_spectator = 1;
1462                 break;
1463         case Opt_ignore_local_fs:
1464                 /* Retained for backwards compat only */
1465                 break;
1466         case Opt_localflocks:
1467                 args->ar_localflocks = 1;
1468                 break;
1469         case Opt_localcaching:
1470                 /* Retained for backwards compat only */
1471                 break;
1472         case Opt_debug:
1473                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1474                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1475                 args->ar_debug = result.boolean;
1476                 break;
1477         case Opt_upgrade:
1478                 /* Retained for backwards compat only */
1479                 break;
1480         case Opt_acl:
1481                 args->ar_posix_acl = result.boolean;
1482                 break;
1483         case Opt_quota_flag:
1484                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1485                 break;
1486         case Opt_quota:
1487                 args->ar_quota = result.int_32;
1488                 break;
1489         case Opt_suiddir:
1490                 args->ar_suiddir = result.boolean;
1491                 break;
1492         case Opt_data:
1493                 /* The uint_32 result maps directly to GFS2_DATA_* */
1494                 args->ar_data = result.uint_32;
1495                 break;
1496         case Opt_meta:
1497                 args->ar_meta = 1;
1498                 break;
1499         case Opt_discard:
1500                 args->ar_discard = result.boolean;
1501                 break;
1502         case Opt_commit:
1503                 if (result.int_32 <= 0)
1504                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1505                 args->ar_commit = result.int_32;
1506                 break;
1507         case Opt_statfs_quantum:
1508                 if (result.int_32 < 0)
1509                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1510                 args->ar_statfs_quantum = result.int_32;
1511                 break;
1512         case Opt_quota_quantum:
1513                 if (result.int_32 <= 0)
1514                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1515                 args->ar_quota_quantum = result.int_32;
1516                 break;
1517         case Opt_statfs_percent:
1518                 if (result.int_32 < 0 || result.int_32 > 100)
1519                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1520                 args->ar_statfs_percent = result.int_32;
1521                 break;
1522         case Opt_errors:
1523                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1524                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1525                 args->ar_errors = result.uint_32;
1526                 break;
1527         case Opt_barrier:
1528                 args->ar_nobarrier = result.boolean;
1529                 break;
1530         case Opt_rgrplvb:
1531                 args->ar_rgrplvb = result.boolean;
1532                 args->ar_got_rgrplvb = 1;
1533                 break;
1534         case Opt_loccookie:
1535                 args->ar_loccookie = result.boolean;
1536                 break;
1537         default:
1538                 return invalfc(fc, "invalid mount option: %s", param->key);
1539         }
1540         return 0;
1541 }
1542
1543 static int gfs2_reconfigure(struct fs_context *fc)
1544 {
1545         struct super_block *sb = fc->root->d_sb;
1546         struct gfs2_sbd *sdp = sb->s_fs_info;
1547         struct gfs2_args *oldargs = &sdp->sd_args;
1548         struct gfs2_args *newargs = fc->fs_private;
1549         struct gfs2_tune *gt = &sdp->sd_tune;
1550         int error = 0;
1551
1552         sync_filesystem(sb);
1553
1554         spin_lock(&gt->gt_spin);
1555         oldargs->ar_commit = gt->gt_logd_secs;
1556         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1557         if (gt->gt_statfs_slow)
1558                 oldargs->ar_statfs_quantum = 0;
1559         else
1560                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1561         spin_unlock(&gt->gt_spin);
1562
1563         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1564                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1565                 return -EINVAL;
1566         }
1567         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1568                 errorfc(fc, "reconfiguration of lock table not allowed");
1569                 return -EINVAL;
1570         }
1571         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1572                 errorfc(fc, "reconfiguration of host data not allowed");
1573                 return -EINVAL;
1574         }
1575         if (newargs->ar_spectator != oldargs->ar_spectator) {
1576                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1577                 return -EINVAL;
1578         }
1579         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1580                 errorfc(fc, "reconfiguration of localflocks not allowed");
1581                 return -EINVAL;
1582         }
1583         if (newargs->ar_meta != oldargs->ar_meta) {
1584                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1585                 return -EINVAL;
1586         }
1587         if (oldargs->ar_spectator)
1588                 fc->sb_flags |= SB_RDONLY;
1589
1590         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1591                 if (fc->sb_flags & SB_RDONLY) {
1592                         gfs2_make_fs_ro(sdp);
1593                 } else {
1594                         error = gfs2_make_fs_rw(sdp);
1595                         if (error)
1596                                 errorfc(fc, "unable to remount read-write");
1597                 }
1598         }
1599         sdp->sd_args = *newargs;
1600
1601         if (sdp->sd_args.ar_posix_acl)
1602                 sb->s_flags |= SB_POSIXACL;
1603         else
1604                 sb->s_flags &= ~SB_POSIXACL;
1605         if (sdp->sd_args.ar_nobarrier)
1606                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1607         else
1608                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1609         spin_lock(&gt->gt_spin);
1610         gt->gt_logd_secs = newargs->ar_commit;
1611         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1612         if (newargs->ar_statfs_quantum) {
1613                 gt->gt_statfs_slow = 0;
1614                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1615         }
1616         else {
1617                 gt->gt_statfs_slow = 1;
1618                 gt->gt_statfs_quantum = 30;
1619         }
1620         spin_unlock(&gt->gt_spin);
1621
1622         gfs2_online_uevent(sdp);
1623         return error;
1624 }
1625
1626 static const struct fs_context_operations gfs2_context_ops = {
1627         .free        = gfs2_fc_free,
1628         .parse_param = gfs2_parse_param,
1629         .get_tree    = gfs2_get_tree,
1630         .reconfigure = gfs2_reconfigure,
1631 };
1632
1633 /* Set up the filesystem mount context */
1634 static int gfs2_init_fs_context(struct fs_context *fc)
1635 {
1636         struct gfs2_args *args;
1637
1638         args = kmalloc(sizeof(*args), GFP_KERNEL);
1639         if (args == NULL)
1640                 return -ENOMEM;
1641
1642         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1643                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1644
1645                 *args = sdp->sd_args;
1646         } else {
1647                 memset(args, 0, sizeof(*args));
1648                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1649                 args->ar_data = GFS2_DATA_DEFAULT;
1650                 args->ar_commit = 30;
1651                 args->ar_statfs_quantum = 30;
1652                 args->ar_quota_quantum = 60;
1653                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1654         }
1655         fc->fs_private = args;
1656         fc->ops = &gfs2_context_ops;
1657         return 0;
1658 }
1659
1660 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1661 {
1662         return -EINVAL;
1663 }
1664
1665 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1666 {
1667         return (fc->sget_key == s->s_bdev);
1668 }
1669
1670 static int gfs2_meta_get_tree(struct fs_context *fc)
1671 {
1672         struct super_block *s;
1673         struct gfs2_sbd *sdp;
1674         struct path path;
1675         int error;
1676
1677         if (!fc->source || !*fc->source)
1678                 return -EINVAL;
1679
1680         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1681         if (error) {
1682                 pr_warn("path_lookup on %s returned error %d\n",
1683                         fc->source, error);
1684                 return error;
1685         }
1686         fc->fs_type = &gfs2_fs_type;
1687         fc->sget_key = path.dentry->d_sb->s_bdev;
1688         s = sget_fc(fc, test_meta_super, set_meta_super);
1689         path_put(&path);
1690         if (IS_ERR(s)) {
1691                 pr_warn("gfs2 mount does not exist\n");
1692                 return PTR_ERR(s);
1693         }
1694         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1695                 deactivate_locked_super(s);
1696                 return -EBUSY;
1697         }
1698         sdp = s->s_fs_info;
1699         fc->root = dget(sdp->sd_master_dir);
1700         return 0;
1701 }
1702
1703 static const struct fs_context_operations gfs2_meta_context_ops = {
1704         .free        = gfs2_fc_free,
1705         .get_tree    = gfs2_meta_get_tree,
1706 };
1707
1708 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1709 {
1710         int ret = gfs2_init_fs_context(fc);
1711
1712         if (ret)
1713                 return ret;
1714
1715         fc->ops = &gfs2_meta_context_ops;
1716         return 0;
1717 }
1718
1719 /**
1720  * gfs2_evict_inodes - evict inodes cooperatively
1721  * @sb: the superblock
1722  *
1723  * When evicting an inode with a zero link count, we are trying to upgrade the
1724  * inode's iopen glock from SH to EX mode in order to determine if we can
1725  * delete the inode.  The other nodes are supposed to evict the inode from
1726  * their caches if they can, and to poke the inode's inode glock if they cannot
1727  * do so.  Either behavior allows gfs2_upgrade_iopen_glock() to proceed
1728  * quickly, but if the other nodes are not cooperating, the lock upgrading
1729  * attempt will time out.  Since inodes are evicted sequentially, this can add
1730  * up quickly.
1731  *
1732  * Function evict_inodes() tries to keep the s_inode_list_lock list locked over
1733  * a long time, which prevents other inodes from being evicted concurrently.
1734  * This precludes the cooperative behavior we are looking for.  This special
1735  * version of evict_inodes() avoids that.
1736  *
1737  * Modeled after drop_pagecache_sb().
1738  */
1739 static void gfs2_evict_inodes(struct super_block *sb)
1740 {
1741         struct inode *inode, *toput_inode = NULL;
1742         struct gfs2_sbd *sdp = sb->s_fs_info;
1743
1744         set_bit(SDF_EVICTING, &sdp->sd_flags);
1745
1746         spin_lock(&sb->s_inode_list_lock);
1747         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1748                 spin_lock(&inode->i_lock);
1749                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) &&
1750                     !need_resched()) {
1751                         spin_unlock(&inode->i_lock);
1752                         continue;
1753                 }
1754                 atomic_inc(&inode->i_count);
1755                 spin_unlock(&inode->i_lock);
1756                 spin_unlock(&sb->s_inode_list_lock);
1757
1758                 iput(toput_inode);
1759                 toput_inode = inode;
1760
1761                 cond_resched();
1762                 spin_lock(&sb->s_inode_list_lock);
1763         }
1764         spin_unlock(&sb->s_inode_list_lock);
1765         iput(toput_inode);
1766 }
1767
1768 static void gfs2_kill_sb(struct super_block *sb)
1769 {
1770         struct gfs2_sbd *sdp = sb->s_fs_info;
1771
1772         if (sdp == NULL) {
1773                 kill_block_super(sb);
1774                 return;
1775         }
1776
1777         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1778         dput(sdp->sd_root_dir);
1779         dput(sdp->sd_master_dir);
1780         sdp->sd_root_dir = NULL;
1781         sdp->sd_master_dir = NULL;
1782         shrink_dcache_sb(sb);
1783
1784         gfs2_evict_inodes(sb);
1785
1786         /*
1787          * Flush and then drain the delete workqueue here (via
1788          * destroy_workqueue()) to ensure that any delete work that
1789          * may be running will also see the SDF_DEACTIVATING flag.
1790          */
1791         set_bit(SDF_DEACTIVATING, &sdp->sd_flags);
1792         gfs2_flush_delete_work(sdp);
1793         destroy_workqueue(sdp->sd_delete_wq);
1794
1795         kill_block_super(sb);
1796 }
1797
1798 struct file_system_type gfs2_fs_type = {
1799         .name = "gfs2",
1800         .fs_flags = FS_REQUIRES_DEV,
1801         .init_fs_context = gfs2_init_fs_context,
1802         .parameters = gfs2_fs_parameters,
1803         .kill_sb = gfs2_kill_sb,
1804         .owner = THIS_MODULE,
1805 };
1806 MODULE_ALIAS_FS("gfs2");
1807
1808 struct file_system_type gfs2meta_fs_type = {
1809         .name = "gfs2meta",
1810         .fs_flags = FS_REQUIRES_DEV,
1811         .init_fs_context = gfs2_meta_init_fs_context,
1812         .owner = THIS_MODULE,
1813 };
1814 MODULE_ALIAS_FS("gfs2meta");