Loading...
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
53static void gfs2_tune_init(struct gfs2_tune *gt)
54{
55 spin_lock_init(>->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
65void free_sbd(struct gfs2_sbd *sdp)
66{
67 if (sdp->sd_lkstats)
68 free_percpu(sdp->sd_lkstats);
69 kfree(sdp);
70}
71
72static 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
143fail:
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
158static 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
188static 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
199static 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
240static 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
282static 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
363static 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
399static 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
443fail_trans:
444 gfs2_glock_put(sdp->sd_freeze_gl);
445fail_rename:
446 gfs2_glock_put(sdp->sd_rename_gl);
447fail_live:
448 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
449fail_mount:
450 gfs2_glock_dq_uninit(mount_gh);
451fail:
452 return error;
453}
454
455static 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
477static 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);
541out:
542 gfs2_glock_dq_uninit(&sb_gh);
543 return ret;
544}
545
546static 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
567static 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 */
643static 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
710unlock_sd_gh:
711 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
712free_local:
713 free_local_statfs_inodes(sdp);
714 iput(pn);
715put_statfs:
716 iput(sdp->sd_statfs_inode);
717out:
718 return error;
719}
720
721/* Uninitialize and free up memory used by the list of statfs inodes */
722static 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
732static 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 jindex = 1;
738 int error = 0;
739
740 if (undo) {
741 jindex = 0;
742 goto fail_statfs;
743 }
744
745 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
746 if (IS_ERR(sdp->sd_jindex)) {
747 fs_err(sdp, "can't lookup journal index: %d\n", error);
748 return PTR_ERR(sdp->sd_jindex);
749 }
750
751 /* Load in the journal index special file */
752
753 error = gfs2_jindex_hold(sdp, &ji_gh);
754 if (error) {
755 fs_err(sdp, "can't read journal index: %d\n", error);
756 goto fail;
757 }
758
759 error = -EUSERS;
760 if (!gfs2_jindex_size(sdp)) {
761 fs_err(sdp, "no journals!\n");
762 goto fail_jindex;
763 }
764
765 atomic_set(&sdp->sd_log_blks_needed, 0);
766 if (sdp->sd_args.ar_spectator) {
767 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
768 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
769 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
770 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
771 } else {
772 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
773 fs_err(sdp, "can't mount journal #%u\n",
774 sdp->sd_lockstruct.ls_jid);
775 fs_err(sdp, "there are only %u journals (0 - %u)\n",
776 gfs2_jindex_size(sdp),
777 gfs2_jindex_size(sdp) - 1);
778 goto fail_jindex;
779 }
780 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
781
782 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
783 &gfs2_journal_glops,
784 LM_ST_EXCLUSIVE,
785 LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
786 &sdp->sd_journal_gh);
787 if (error) {
788 fs_err(sdp, "can't acquire journal glock: %d\n", error);
789 goto fail_jindex;
790 }
791
792 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
793 sdp->sd_jinode_gl = ip->i_gl;
794 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
795 LM_FLAG_NOEXP | GL_EXACT |
796 GL_NOCACHE | GL_NOPID,
797 &sdp->sd_jinode_gh);
798 if (error) {
799 fs_err(sdp, "can't acquire journal inode glock: %d\n",
800 error);
801 goto fail_journal_gh;
802 }
803
804 error = gfs2_jdesc_check(sdp->sd_jdesc);
805 if (error) {
806 fs_err(sdp, "my journal (%u) is bad: %d\n",
807 sdp->sd_jdesc->jd_jid, error);
808 goto fail_jinode_gh;
809 }
810 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
811 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
812 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
813
814 /* Map the extents for this journal's blocks */
815 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
816 }
817 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
818
819 /* Lookup statfs inodes here so journal recovery can use them. */
820 error = init_statfs(sdp);
821 if (error)
822 goto fail_jinode_gh;
823
824 if (sdp->sd_lockstruct.ls_first) {
825 unsigned int x;
826 for (x = 0; x < sdp->sd_journals; x++) {
827 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
828
829 if (sdp->sd_args.ar_spectator) {
830 error = check_journal_clean(sdp, jd, true);
831 if (error)
832 goto fail_statfs;
833 continue;
834 }
835 error = gfs2_recover_journal(jd, true);
836 if (error) {
837 fs_err(sdp, "error recovering journal %u: %d\n",
838 x, error);
839 goto fail_statfs;
840 }
841 }
842
843 gfs2_others_may_mount(sdp);
844 } else if (!sdp->sd_args.ar_spectator) {
845 error = gfs2_recover_journal(sdp->sd_jdesc, true);
846 if (error) {
847 fs_err(sdp, "error recovering my journal: %d\n", error);
848 goto fail_statfs;
849 }
850 }
851
852 sdp->sd_log_idle = 1;
853 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
854 gfs2_glock_dq_uninit(&ji_gh);
855 jindex = 0;
856 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
857 return 0;
858
859fail_statfs:
860 uninit_statfs(sdp);
861fail_jinode_gh:
862 /* A withdraw may have done dq/uninit so now we need to check it */
863 if (!sdp->sd_args.ar_spectator &&
864 gfs2_holder_initialized(&sdp->sd_jinode_gh))
865 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
866fail_journal_gh:
867 if (!sdp->sd_args.ar_spectator &&
868 gfs2_holder_initialized(&sdp->sd_journal_gh))
869 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
870fail_jindex:
871 gfs2_jindex_free(sdp);
872 if (jindex)
873 gfs2_glock_dq_uninit(&ji_gh);
874fail:
875 iput(sdp->sd_jindex);
876 return error;
877}
878
879static struct lock_class_key gfs2_quota_imutex_key;
880
881static int init_inodes(struct gfs2_sbd *sdp, int undo)
882{
883 int error = 0;
884 struct inode *master = d_inode(sdp->sd_master_dir);
885
886 if (undo)
887 goto fail_qinode;
888
889 error = init_journal(sdp, undo);
890 complete_all(&sdp->sd_journal_ready);
891 if (error)
892 goto fail;
893
894 /* Read in the resource index inode */
895 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
896 if (IS_ERR(sdp->sd_rindex)) {
897 error = PTR_ERR(sdp->sd_rindex);
898 fs_err(sdp, "can't get resource index inode: %d\n", error);
899 goto fail_journal;
900 }
901 sdp->sd_rindex_uptodate = 0;
902
903 /* Read in the quota inode */
904 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
905 if (IS_ERR(sdp->sd_quota_inode)) {
906 error = PTR_ERR(sdp->sd_quota_inode);
907 fs_err(sdp, "can't get quota file inode: %d\n", error);
908 goto fail_rindex;
909 }
910 /*
911 * i_rwsem on quota files is special. Since this inode is hidden system
912 * file, we are safe to define locking ourselves.
913 */
914 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
915 &gfs2_quota_imutex_key);
916
917 error = gfs2_rindex_update(sdp);
918 if (error)
919 goto fail_qinode;
920
921 return 0;
922
923fail_qinode:
924 iput(sdp->sd_quota_inode);
925fail_rindex:
926 gfs2_clear_rgrpd(sdp);
927 iput(sdp->sd_rindex);
928fail_journal:
929 init_journal(sdp, UNDO);
930fail:
931 return error;
932}
933
934static int init_per_node(struct gfs2_sbd *sdp, int undo)
935{
936 struct inode *pn = NULL;
937 char buf[30];
938 int error = 0;
939 struct gfs2_inode *ip;
940 struct inode *master = d_inode(sdp->sd_master_dir);
941
942 if (sdp->sd_args.ar_spectator)
943 return 0;
944
945 if (undo)
946 goto fail_qc_gh;
947
948 pn = gfs2_lookup_simple(master, "per_node");
949 if (IS_ERR(pn)) {
950 error = PTR_ERR(pn);
951 fs_err(sdp, "can't find per_node directory: %d\n", error);
952 return error;
953 }
954
955 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
956 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
957 if (IS_ERR(sdp->sd_qc_inode)) {
958 error = PTR_ERR(sdp->sd_qc_inode);
959 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
960 goto fail_ut_i;
961 }
962
963 iput(pn);
964 pn = NULL;
965
966 ip = GFS2_I(sdp->sd_qc_inode);
967 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
968 &sdp->sd_qc_gh);
969 if (error) {
970 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
971 goto fail_qc_i;
972 }
973
974 return 0;
975
976fail_qc_gh:
977 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
978fail_qc_i:
979 iput(sdp->sd_qc_inode);
980fail_ut_i:
981 iput(pn);
982 return error;
983}
984
985static const match_table_t nolock_tokens = {
986 { Opt_jid, "jid=%d", },
987 { Opt_err, NULL },
988};
989
990static const struct lm_lockops nolock_ops = {
991 .lm_proto_name = "lock_nolock",
992 .lm_put_lock = gfs2_glock_free,
993 .lm_tokens = &nolock_tokens,
994};
995
996/**
997 * gfs2_lm_mount - mount a locking protocol
998 * @sdp: the filesystem
999 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
1000 *
1001 * Returns: errno
1002 */
1003
1004static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1005{
1006 const struct lm_lockops *lm;
1007 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1008 struct gfs2_args *args = &sdp->sd_args;
1009 const char *proto = sdp->sd_proto_name;
1010 const char *table = sdp->sd_table_name;
1011 char *o, *options;
1012 int ret;
1013
1014 if (!strcmp("lock_nolock", proto)) {
1015 lm = &nolock_ops;
1016 sdp->sd_args.ar_localflocks = 1;
1017#ifdef CONFIG_GFS2_FS_LOCKING_DLM
1018 } else if (!strcmp("lock_dlm", proto)) {
1019 lm = &gfs2_dlm_ops;
1020#endif
1021 } else {
1022 pr_info("can't find protocol %s\n", proto);
1023 return -ENOENT;
1024 }
1025
1026 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1027
1028 ls->ls_ops = lm;
1029 ls->ls_first = 1;
1030
1031 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1032 substring_t tmp[MAX_OPT_ARGS];
1033 int token, option;
1034
1035 if (!o || !*o)
1036 continue;
1037
1038 token = match_token(o, *lm->lm_tokens, tmp);
1039 switch (token) {
1040 case Opt_jid:
1041 ret = match_int(&tmp[0], &option);
1042 if (ret || option < 0)
1043 goto hostdata_error;
1044 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1045 ls->ls_jid = option;
1046 break;
1047 case Opt_id:
1048 case Opt_nodir:
1049 /* Obsolete, but left for backward compat purposes */
1050 break;
1051 case Opt_first:
1052 ret = match_int(&tmp[0], &option);
1053 if (ret || (option != 0 && option != 1))
1054 goto hostdata_error;
1055 ls->ls_first = option;
1056 break;
1057 case Opt_err:
1058 default:
1059hostdata_error:
1060 fs_info(sdp, "unknown hostdata (%s)\n", o);
1061 return -EINVAL;
1062 }
1063 }
1064
1065 if (lm->lm_mount == NULL) {
1066 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1067 complete_all(&sdp->sd_locking_init);
1068 return 0;
1069 }
1070 ret = lm->lm_mount(sdp, table);
1071 if (ret == 0)
1072 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1073 sdp->sd_sb.sb_fs_format);
1074 complete_all(&sdp->sd_locking_init);
1075 return ret;
1076}
1077
1078void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1079{
1080 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1081 if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1082 lm->lm_unmount(sdp);
1083}
1084
1085static int wait_on_journal(struct gfs2_sbd *sdp)
1086{
1087 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1088 return 0;
1089
1090 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1091 ? -EINTR : 0;
1092}
1093
1094void gfs2_online_uevent(struct gfs2_sbd *sdp)
1095{
1096 struct super_block *sb = sdp->sd_vfs;
1097 char ro[20];
1098 char spectator[20];
1099 char *envp[] = { ro, spectator, NULL };
1100 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1101 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1102 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1103}
1104
1105static int init_threads(struct gfs2_sbd *sdp)
1106{
1107 struct task_struct *p;
1108 int error = 0;
1109
1110 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1111 if (IS_ERR(p)) {
1112 error = PTR_ERR(p);
1113 fs_err(sdp, "can't start logd thread: %d\n", error);
1114 return error;
1115 }
1116 sdp->sd_logd_process = p;
1117
1118 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1119 if (IS_ERR(p)) {
1120 error = PTR_ERR(p);
1121 fs_err(sdp, "can't start quotad thread: %d\n", error);
1122 goto fail;
1123 }
1124 sdp->sd_quotad_process = p;
1125 return 0;
1126
1127fail:
1128 kthread_stop(sdp->sd_logd_process);
1129 sdp->sd_logd_process = NULL;
1130 return error;
1131}
1132
1133/**
1134 * gfs2_fill_super - Read in superblock
1135 * @sb: The VFS superblock
1136 * @fc: Mount options and flags
1137 *
1138 * Returns: -errno
1139 */
1140static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1141{
1142 struct gfs2_args *args = fc->fs_private;
1143 int silent = fc->sb_flags & SB_SILENT;
1144 struct gfs2_sbd *sdp;
1145 struct gfs2_holder mount_gh;
1146 struct gfs2_holder freeze_gh;
1147 int error;
1148
1149 sdp = init_sbd(sb);
1150 if (!sdp) {
1151 pr_warn("can't alloc struct gfs2_sbd\n");
1152 return -ENOMEM;
1153 }
1154 sdp->sd_args = *args;
1155
1156 if (sdp->sd_args.ar_spectator) {
1157 sb->s_flags |= SB_RDONLY;
1158 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1159 }
1160 if (sdp->sd_args.ar_posix_acl)
1161 sb->s_flags |= SB_POSIXACL;
1162 if (sdp->sd_args.ar_nobarrier)
1163 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1164
1165 sb->s_flags |= SB_NOSEC;
1166 sb->s_magic = GFS2_MAGIC;
1167 sb->s_op = &gfs2_super_ops;
1168 sb->s_d_op = &gfs2_dops;
1169 sb->s_export_op = &gfs2_export_ops;
1170 sb->s_qcop = &gfs2_quotactl_ops;
1171 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1172 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1173 sb->s_time_gran = 1;
1174 sb->s_maxbytes = MAX_LFS_FILESIZE;
1175
1176 /* Set up the buffer cache and fill in some fake block size values
1177 to allow us to read-in the on-disk superblock. */
1178 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1179 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1180 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1181 GFS2_BASIC_BLOCK_SHIFT;
1182 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1183
1184 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1185 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1186 if (sdp->sd_args.ar_statfs_quantum) {
1187 sdp->sd_tune.gt_statfs_slow = 0;
1188 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1189 } else {
1190 sdp->sd_tune.gt_statfs_slow = 1;
1191 sdp->sd_tune.gt_statfs_quantum = 30;
1192 }
1193
1194 error = init_names(sdp, silent);
1195 if (error)
1196 goto fail_free;
1197
1198 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1199
1200 error = gfs2_sys_fs_add(sdp);
1201 if (error)
1202 goto fail_free;
1203
1204 gfs2_create_debugfs_file(sdp);
1205
1206 error = gfs2_lm_mount(sdp, silent);
1207 if (error)
1208 goto fail_debug;
1209
1210 error = init_locking(sdp, &mount_gh, DO);
1211 if (error)
1212 goto fail_lm;
1213
1214 error = init_sb(sdp, silent);
1215 if (error)
1216 goto fail_locking;
1217
1218 /* Turn rgrplvb on by default if fs format is recent enough */
1219 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1220 sdp->sd_args.ar_rgrplvb = 1;
1221
1222 error = wait_on_journal(sdp);
1223 if (error)
1224 goto fail_sb;
1225
1226 /*
1227 * If user space has failed to join the cluster or some similar
1228 * failure has occurred, then the journal id will contain a
1229 * negative (error) number. This will then be returned to the
1230 * caller (of the mount syscall). We do this even for spectator
1231 * mounts (which just write a jid of 0 to indicate "ok" even though
1232 * the jid is unused in the spectator case)
1233 */
1234 if (sdp->sd_lockstruct.ls_jid < 0) {
1235 error = sdp->sd_lockstruct.ls_jid;
1236 sdp->sd_lockstruct.ls_jid = 0;
1237 goto fail_sb;
1238 }
1239
1240 if (sdp->sd_args.ar_spectator)
1241 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1242 sdp->sd_table_name);
1243 else
1244 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1245 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1246
1247 error = init_inodes(sdp, DO);
1248 if (error)
1249 goto fail_sb;
1250
1251 error = init_per_node(sdp, DO);
1252 if (error)
1253 goto fail_inodes;
1254
1255 error = gfs2_statfs_init(sdp);
1256 if (error) {
1257 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1258 goto fail_per_node;
1259 }
1260
1261 if (!sb_rdonly(sb)) {
1262 error = init_threads(sdp);
1263 if (error) {
1264 gfs2_withdraw_delayed(sdp);
1265 goto fail_per_node;
1266 }
1267 }
1268
1269 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1270 if (error)
1271 goto fail_per_node;
1272
1273 if (!sb_rdonly(sb))
1274 error = gfs2_make_fs_rw(sdp);
1275
1276 gfs2_freeze_unlock(&freeze_gh);
1277 if (error) {
1278 if (sdp->sd_quotad_process)
1279 kthread_stop(sdp->sd_quotad_process);
1280 sdp->sd_quotad_process = NULL;
1281 if (sdp->sd_logd_process)
1282 kthread_stop(sdp->sd_logd_process);
1283 sdp->sd_logd_process = NULL;
1284 fs_err(sdp, "can't make FS RW: %d\n", error);
1285 goto fail_per_node;
1286 }
1287 gfs2_glock_dq_uninit(&mount_gh);
1288 gfs2_online_uevent(sdp);
1289 return 0;
1290
1291fail_per_node:
1292 init_per_node(sdp, UNDO);
1293fail_inodes:
1294 init_inodes(sdp, UNDO);
1295fail_sb:
1296 if (sdp->sd_root_dir)
1297 dput(sdp->sd_root_dir);
1298 if (sdp->sd_master_dir)
1299 dput(sdp->sd_master_dir);
1300 if (sb->s_root)
1301 dput(sb->s_root);
1302 sb->s_root = NULL;
1303fail_locking:
1304 init_locking(sdp, &mount_gh, UNDO);
1305fail_lm:
1306 complete_all(&sdp->sd_journal_ready);
1307 gfs2_gl_hash_clear(sdp);
1308 gfs2_lm_unmount(sdp);
1309fail_debug:
1310 gfs2_delete_debugfs_file(sdp);
1311 gfs2_sys_fs_del(sdp);
1312fail_free:
1313 free_sbd(sdp);
1314 sb->s_fs_info = NULL;
1315 return error;
1316}
1317
1318/**
1319 * gfs2_get_tree - Get the GFS2 superblock and root directory
1320 * @fc: The filesystem context
1321 *
1322 * Returns: 0 or -errno on error
1323 */
1324static int gfs2_get_tree(struct fs_context *fc)
1325{
1326 struct gfs2_args *args = fc->fs_private;
1327 struct gfs2_sbd *sdp;
1328 int error;
1329
1330 error = get_tree_bdev(fc, gfs2_fill_super);
1331 if (error)
1332 return error;
1333
1334 sdp = fc->root->d_sb->s_fs_info;
1335 dput(fc->root);
1336 if (args->ar_meta)
1337 fc->root = dget(sdp->sd_master_dir);
1338 else
1339 fc->root = dget(sdp->sd_root_dir);
1340 return 0;
1341}
1342
1343static void gfs2_fc_free(struct fs_context *fc)
1344{
1345 struct gfs2_args *args = fc->fs_private;
1346
1347 kfree(args);
1348}
1349
1350enum gfs2_param {
1351 Opt_lockproto,
1352 Opt_locktable,
1353 Opt_hostdata,
1354 Opt_spectator,
1355 Opt_ignore_local_fs,
1356 Opt_localflocks,
1357 Opt_localcaching,
1358 Opt_debug,
1359 Opt_upgrade,
1360 Opt_acl,
1361 Opt_quota,
1362 Opt_quota_flag,
1363 Opt_suiddir,
1364 Opt_data,
1365 Opt_meta,
1366 Opt_discard,
1367 Opt_commit,
1368 Opt_errors,
1369 Opt_statfs_quantum,
1370 Opt_statfs_percent,
1371 Opt_quota_quantum,
1372 Opt_barrier,
1373 Opt_rgrplvb,
1374 Opt_loccookie,
1375};
1376
1377static const struct constant_table gfs2_param_quota[] = {
1378 {"off", GFS2_QUOTA_OFF},
1379 {"account", GFS2_QUOTA_ACCOUNT},
1380 {"on", GFS2_QUOTA_ON},
1381 {}
1382};
1383
1384enum opt_data {
1385 Opt_data_writeback = GFS2_DATA_WRITEBACK,
1386 Opt_data_ordered = GFS2_DATA_ORDERED,
1387};
1388
1389static const struct constant_table gfs2_param_data[] = {
1390 {"writeback", Opt_data_writeback },
1391 {"ordered", Opt_data_ordered },
1392 {}
1393};
1394
1395enum opt_errors {
1396 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1397 Opt_errors_panic = GFS2_ERRORS_PANIC,
1398};
1399
1400static const struct constant_table gfs2_param_errors[] = {
1401 {"withdraw", Opt_errors_withdraw },
1402 {"panic", Opt_errors_panic },
1403 {}
1404};
1405
1406static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1407 fsparam_string ("lockproto", Opt_lockproto),
1408 fsparam_string ("locktable", Opt_locktable),
1409 fsparam_string ("hostdata", Opt_hostdata),
1410 fsparam_flag ("spectator", Opt_spectator),
1411 fsparam_flag ("norecovery", Opt_spectator),
1412 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs),
1413 fsparam_flag ("localflocks", Opt_localflocks),
1414 fsparam_flag ("localcaching", Opt_localcaching),
1415 fsparam_flag_no("debug", Opt_debug),
1416 fsparam_flag ("upgrade", Opt_upgrade),
1417 fsparam_flag_no("acl", Opt_acl),
1418 fsparam_flag_no("suiddir", Opt_suiddir),
1419 fsparam_enum ("data", Opt_data, gfs2_param_data),
1420 fsparam_flag ("meta", Opt_meta),
1421 fsparam_flag_no("discard", Opt_discard),
1422 fsparam_s32 ("commit", Opt_commit),
1423 fsparam_enum ("errors", Opt_errors, gfs2_param_errors),
1424 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum),
1425 fsparam_s32 ("statfs_percent", Opt_statfs_percent),
1426 fsparam_s32 ("quota_quantum", Opt_quota_quantum),
1427 fsparam_flag_no("barrier", Opt_barrier),
1428 fsparam_flag_no("rgrplvb", Opt_rgrplvb),
1429 fsparam_flag_no("loccookie", Opt_loccookie),
1430 /* quota can be a flag or an enum so it gets special treatment */
1431 fsparam_flag_no("quota", Opt_quota_flag),
1432 fsparam_enum("quota", Opt_quota, gfs2_param_quota),
1433 {}
1434};
1435
1436/* Parse a single mount parameter */
1437static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1438{
1439 struct gfs2_args *args = fc->fs_private;
1440 struct fs_parse_result result;
1441 int o;
1442
1443 o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1444 if (o < 0)
1445 return o;
1446
1447 switch (o) {
1448 case Opt_lockproto:
1449 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1450 break;
1451 case Opt_locktable:
1452 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1453 break;
1454 case Opt_hostdata:
1455 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1456 break;
1457 case Opt_spectator:
1458 args->ar_spectator = 1;
1459 break;
1460 case Opt_ignore_local_fs:
1461 /* Retained for backwards compat only */
1462 break;
1463 case Opt_localflocks:
1464 args->ar_localflocks = 1;
1465 break;
1466 case Opt_localcaching:
1467 /* Retained for backwards compat only */
1468 break;
1469 case Opt_debug:
1470 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1471 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1472 args->ar_debug = result.boolean;
1473 break;
1474 case Opt_upgrade:
1475 /* Retained for backwards compat only */
1476 break;
1477 case Opt_acl:
1478 args->ar_posix_acl = result.boolean;
1479 break;
1480 case Opt_quota_flag:
1481 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1482 break;
1483 case Opt_quota:
1484 args->ar_quota = result.int_32;
1485 break;
1486 case Opt_suiddir:
1487 args->ar_suiddir = result.boolean;
1488 break;
1489 case Opt_data:
1490 /* The uint_32 result maps directly to GFS2_DATA_* */
1491 args->ar_data = result.uint_32;
1492 break;
1493 case Opt_meta:
1494 args->ar_meta = 1;
1495 break;
1496 case Opt_discard:
1497 args->ar_discard = result.boolean;
1498 break;
1499 case Opt_commit:
1500 if (result.int_32 <= 0)
1501 return invalfc(fc, "commit mount option requires a positive numeric argument");
1502 args->ar_commit = result.int_32;
1503 break;
1504 case Opt_statfs_quantum:
1505 if (result.int_32 < 0)
1506 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1507 args->ar_statfs_quantum = result.int_32;
1508 break;
1509 case Opt_quota_quantum:
1510 if (result.int_32 <= 0)
1511 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1512 args->ar_quota_quantum = result.int_32;
1513 break;
1514 case Opt_statfs_percent:
1515 if (result.int_32 < 0 || result.int_32 > 100)
1516 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1517 args->ar_statfs_percent = result.int_32;
1518 break;
1519 case Opt_errors:
1520 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1521 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1522 args->ar_errors = result.uint_32;
1523 break;
1524 case Opt_barrier:
1525 args->ar_nobarrier = result.boolean;
1526 break;
1527 case Opt_rgrplvb:
1528 args->ar_rgrplvb = result.boolean;
1529 args->ar_got_rgrplvb = 1;
1530 break;
1531 case Opt_loccookie:
1532 args->ar_loccookie = result.boolean;
1533 break;
1534 default:
1535 return invalfc(fc, "invalid mount option: %s", param->key);
1536 }
1537 return 0;
1538}
1539
1540static int gfs2_reconfigure(struct fs_context *fc)
1541{
1542 struct super_block *sb = fc->root->d_sb;
1543 struct gfs2_sbd *sdp = sb->s_fs_info;
1544 struct gfs2_args *oldargs = &sdp->sd_args;
1545 struct gfs2_args *newargs = fc->fs_private;
1546 struct gfs2_tune *gt = &sdp->sd_tune;
1547 int error = 0;
1548
1549 sync_filesystem(sb);
1550
1551 spin_lock(>->gt_spin);
1552 oldargs->ar_commit = gt->gt_logd_secs;
1553 oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1554 if (gt->gt_statfs_slow)
1555 oldargs->ar_statfs_quantum = 0;
1556 else
1557 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1558 spin_unlock(>->gt_spin);
1559
1560 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1561 errorfc(fc, "reconfiguration of locking protocol not allowed");
1562 return -EINVAL;
1563 }
1564 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1565 errorfc(fc, "reconfiguration of lock table not allowed");
1566 return -EINVAL;
1567 }
1568 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1569 errorfc(fc, "reconfiguration of host data not allowed");
1570 return -EINVAL;
1571 }
1572 if (newargs->ar_spectator != oldargs->ar_spectator) {
1573 errorfc(fc, "reconfiguration of spectator mode not allowed");
1574 return -EINVAL;
1575 }
1576 if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1577 errorfc(fc, "reconfiguration of localflocks not allowed");
1578 return -EINVAL;
1579 }
1580 if (newargs->ar_meta != oldargs->ar_meta) {
1581 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1582 return -EINVAL;
1583 }
1584 if (oldargs->ar_spectator)
1585 fc->sb_flags |= SB_RDONLY;
1586
1587 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1588 struct gfs2_holder freeze_gh;
1589
1590 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1591 if (error)
1592 return -EINVAL;
1593
1594 if (fc->sb_flags & SB_RDONLY) {
1595 gfs2_make_fs_ro(sdp);
1596 } else {
1597 error = gfs2_make_fs_rw(sdp);
1598 if (error)
1599 errorfc(fc, "unable to remount read-write");
1600 }
1601 gfs2_freeze_unlock(&freeze_gh);
1602 }
1603 sdp->sd_args = *newargs;
1604
1605 if (sdp->sd_args.ar_posix_acl)
1606 sb->s_flags |= SB_POSIXACL;
1607 else
1608 sb->s_flags &= ~SB_POSIXACL;
1609 if (sdp->sd_args.ar_nobarrier)
1610 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1611 else
1612 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1613 spin_lock(>->gt_spin);
1614 gt->gt_logd_secs = newargs->ar_commit;
1615 gt->gt_quota_quantum = newargs->ar_quota_quantum;
1616 if (newargs->ar_statfs_quantum) {
1617 gt->gt_statfs_slow = 0;
1618 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1619 }
1620 else {
1621 gt->gt_statfs_slow = 1;
1622 gt->gt_statfs_quantum = 30;
1623 }
1624 spin_unlock(>->gt_spin);
1625
1626 gfs2_online_uevent(sdp);
1627 return error;
1628}
1629
1630static const struct fs_context_operations gfs2_context_ops = {
1631 .free = gfs2_fc_free,
1632 .parse_param = gfs2_parse_param,
1633 .get_tree = gfs2_get_tree,
1634 .reconfigure = gfs2_reconfigure,
1635};
1636
1637/* Set up the filesystem mount context */
1638static int gfs2_init_fs_context(struct fs_context *fc)
1639{
1640 struct gfs2_args *args;
1641
1642 args = kmalloc(sizeof(*args), GFP_KERNEL);
1643 if (args == NULL)
1644 return -ENOMEM;
1645
1646 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1647 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1648
1649 *args = sdp->sd_args;
1650 } else {
1651 memset(args, 0, sizeof(*args));
1652 args->ar_quota = GFS2_QUOTA_DEFAULT;
1653 args->ar_data = GFS2_DATA_DEFAULT;
1654 args->ar_commit = 30;
1655 args->ar_statfs_quantum = 30;
1656 args->ar_quota_quantum = 60;
1657 args->ar_errors = GFS2_ERRORS_DEFAULT;
1658 }
1659 fc->fs_private = args;
1660 fc->ops = &gfs2_context_ops;
1661 return 0;
1662}
1663
1664static int set_meta_super(struct super_block *s, struct fs_context *fc)
1665{
1666 return -EINVAL;
1667}
1668
1669static int test_meta_super(struct super_block *s, struct fs_context *fc)
1670{
1671 return (fc->sget_key == s->s_bdev);
1672}
1673
1674static int gfs2_meta_get_tree(struct fs_context *fc)
1675{
1676 struct super_block *s;
1677 struct gfs2_sbd *sdp;
1678 struct path path;
1679 int error;
1680
1681 if (!fc->source || !*fc->source)
1682 return -EINVAL;
1683
1684 error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1685 if (error) {
1686 pr_warn("path_lookup on %s returned error %d\n",
1687 fc->source, error);
1688 return error;
1689 }
1690 fc->fs_type = &gfs2_fs_type;
1691 fc->sget_key = path.dentry->d_sb->s_bdev;
1692 s = sget_fc(fc, test_meta_super, set_meta_super);
1693 path_put(&path);
1694 if (IS_ERR(s)) {
1695 pr_warn("gfs2 mount does not exist\n");
1696 return PTR_ERR(s);
1697 }
1698 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1699 deactivate_locked_super(s);
1700 return -EBUSY;
1701 }
1702 sdp = s->s_fs_info;
1703 fc->root = dget(sdp->sd_master_dir);
1704 return 0;
1705}
1706
1707static const struct fs_context_operations gfs2_meta_context_ops = {
1708 .free = gfs2_fc_free,
1709 .get_tree = gfs2_meta_get_tree,
1710};
1711
1712static int gfs2_meta_init_fs_context(struct fs_context *fc)
1713{
1714 int ret = gfs2_init_fs_context(fc);
1715
1716 if (ret)
1717 return ret;
1718
1719 fc->ops = &gfs2_meta_context_ops;
1720 return 0;
1721}
1722
1723static void gfs2_kill_sb(struct super_block *sb)
1724{
1725 struct gfs2_sbd *sdp = sb->s_fs_info;
1726
1727 if (sdp == NULL) {
1728 kill_block_super(sb);
1729 return;
1730 }
1731
1732 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1733 dput(sdp->sd_root_dir);
1734 dput(sdp->sd_master_dir);
1735 sdp->sd_root_dir = NULL;
1736 sdp->sd_master_dir = NULL;
1737 shrink_dcache_sb(sb);
1738 kill_block_super(sb);
1739}
1740
1741struct file_system_type gfs2_fs_type = {
1742 .name = "gfs2",
1743 .fs_flags = FS_REQUIRES_DEV,
1744 .init_fs_context = gfs2_init_fs_context,
1745 .parameters = gfs2_fs_parameters,
1746 .kill_sb = gfs2_kill_sb,
1747 .owner = THIS_MODULE,
1748};
1749MODULE_ALIAS_FS("gfs2");
1750
1751struct file_system_type gfs2meta_fs_type = {
1752 .name = "gfs2meta",
1753 .fs_flags = FS_REQUIRES_DEV,
1754 .init_fs_context = gfs2_meta_init_fs_context,
1755 .owner = THIS_MODULE,
1756};
1757MODULE_ALIAS_FS("gfs2meta");
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12#include <linux/sched.h>
13#include <linux/slab.h>
14#include <linux/spinlock.h>
15#include <linux/completion.h>
16#include <linux/buffer_head.h>
17#include <linux/blkdev.h>
18#include <linux/kthread.h>
19#include <linux/export.h>
20#include <linux/namei.h>
21#include <linux/mount.h>
22#include <linux/gfs2_ondisk.h>
23#include <linux/quotaops.h>
24#include <linux/lockdep.h>
25#include <linux/module.h>
26
27#include "gfs2.h"
28#include "incore.h"
29#include "bmap.h"
30#include "glock.h"
31#include "glops.h"
32#include "inode.h"
33#include "recovery.h"
34#include "rgrp.h"
35#include "super.h"
36#include "sys.h"
37#include "util.h"
38#include "log.h"
39#include "quota.h"
40#include "dir.h"
41#include "meta_io.h"
42#include "trace_gfs2.h"
43
44#define DO 0
45#define UNDO 1
46
47/**
48 * gfs2_tune_init - Fill a gfs2_tune structure with default values
49 * @gt: tune
50 *
51 */
52
53static void gfs2_tune_init(struct gfs2_tune *gt)
54{
55 spin_lock_init(>->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 = 1 << 18;
62 gt->gt_complain_secs = 10;
63}
64
65static struct gfs2_sbd *init_sbd(struct super_block *sb)
66{
67 struct gfs2_sbd *sdp;
68 struct address_space *mapping;
69
70 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
71 if (!sdp)
72 return NULL;
73
74 sb->s_fs_info = sdp;
75 sdp->sd_vfs = sb;
76 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
77 if (!sdp->sd_lkstats) {
78 kfree(sdp);
79 return NULL;
80 }
81
82 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
83 gfs2_tune_init(&sdp->sd_tune);
84
85 init_waitqueue_head(&sdp->sd_glock_wait);
86 atomic_set(&sdp->sd_glock_disposal, 0);
87 init_completion(&sdp->sd_locking_init);
88 init_completion(&sdp->sd_wdack);
89 spin_lock_init(&sdp->sd_statfs_spin);
90
91 spin_lock_init(&sdp->sd_rindex_spin);
92 sdp->sd_rindex_tree.rb_node = NULL;
93
94 INIT_LIST_HEAD(&sdp->sd_jindex_list);
95 spin_lock_init(&sdp->sd_jindex_spin);
96 mutex_init(&sdp->sd_jindex_mutex);
97 init_completion(&sdp->sd_journal_ready);
98
99 INIT_LIST_HEAD(&sdp->sd_quota_list);
100 mutex_init(&sdp->sd_quota_mutex);
101 mutex_init(&sdp->sd_quota_sync_mutex);
102 init_waitqueue_head(&sdp->sd_quota_wait);
103 INIT_LIST_HEAD(&sdp->sd_trunc_list);
104 spin_lock_init(&sdp->sd_trunc_lock);
105 spin_lock_init(&sdp->sd_bitmap_lock);
106
107 mapping = &sdp->sd_aspace;
108
109 address_space_init_once(mapping);
110 mapping->a_ops = &gfs2_rgrp_aops;
111 mapping->host = sb->s_bdev->bd_inode;
112 mapping->flags = 0;
113 mapping_set_gfp_mask(mapping, GFP_NOFS);
114 mapping->private_data = NULL;
115 mapping->writeback_index = 0;
116
117 spin_lock_init(&sdp->sd_log_lock);
118 atomic_set(&sdp->sd_log_pinned, 0);
119 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
120 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
121 spin_lock_init(&sdp->sd_ordered_lock);
122
123 init_waitqueue_head(&sdp->sd_log_waitq);
124 init_waitqueue_head(&sdp->sd_logd_waitq);
125 spin_lock_init(&sdp->sd_ail_lock);
126 INIT_LIST_HEAD(&sdp->sd_ail1_list);
127 INIT_LIST_HEAD(&sdp->sd_ail2_list);
128
129 init_rwsem(&sdp->sd_log_flush_lock);
130 atomic_set(&sdp->sd_log_in_flight, 0);
131 atomic_set(&sdp->sd_reserving_log, 0);
132 init_waitqueue_head(&sdp->sd_reserving_log_wait);
133 init_waitqueue_head(&sdp->sd_log_flush_wait);
134 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
135 mutex_init(&sdp->sd_freeze_mutex);
136
137 return sdp;
138}
139
140
141/**
142 * gfs2_check_sb - Check superblock
143 * @sdp: the filesystem
144 * @sb: The superblock
145 * @silent: Don't print a message if the check fails
146 *
147 * Checks the version code of the FS is one that we understand how to
148 * read and that the sizes of the various on-disk structures have not
149 * changed.
150 */
151
152static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
153{
154 struct gfs2_sb_host *sb = &sdp->sd_sb;
155
156 if (sb->sb_magic != GFS2_MAGIC ||
157 sb->sb_type != GFS2_METATYPE_SB) {
158 if (!silent)
159 pr_warn("not a GFS2 filesystem\n");
160 return -EINVAL;
161 }
162
163 /* If format numbers match exactly, we're done. */
164
165 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
166 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
167 return 0;
168
169 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
170
171 return -EINVAL;
172}
173
174static void end_bio_io_page(struct bio *bio)
175{
176 struct page *page = bio->bi_private;
177
178 if (!bio->bi_error)
179 SetPageUptodate(page);
180 else
181 pr_warn("error %d reading superblock\n", bio->bi_error);
182 unlock_page(page);
183}
184
185static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
186{
187 struct gfs2_sb_host *sb = &sdp->sd_sb;
188 struct super_block *s = sdp->sd_vfs;
189 const struct gfs2_sb *str = buf;
190
191 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
192 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
193 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
194 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
195 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
196 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
197 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
198 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
199 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
200 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
201 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
202
203 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
204 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
205 memcpy(s->s_uuid, str->sb_uuid, 16);
206}
207
208/**
209 * gfs2_read_super - Read the gfs2 super block from disk
210 * @sdp: The GFS2 super block
211 * @sector: The location of the super block
212 * @error: The error code to return
213 *
214 * This uses the bio functions to read the super block from disk
215 * because we want to be 100% sure that we never read cached data.
216 * A super block is read twice only during each GFS2 mount and is
217 * never written to by the filesystem. The first time its read no
218 * locks are held, and the only details which are looked at are those
219 * relating to the locking protocol. Once locking is up and working,
220 * the sb is read again under the lock to establish the location of
221 * the master directory (contains pointers to journals etc) and the
222 * root directory.
223 *
224 * Returns: 0 on success or error
225 */
226
227static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
228{
229 struct super_block *sb = sdp->sd_vfs;
230 struct gfs2_sb *p;
231 struct page *page;
232 struct bio *bio;
233
234 page = alloc_page(GFP_NOFS);
235 if (unlikely(!page))
236 return -ENOMEM;
237
238 ClearPageUptodate(page);
239 ClearPageDirty(page);
240 lock_page(page);
241
242 bio = bio_alloc(GFP_NOFS, 1);
243 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
244 bio->bi_bdev = sb->s_bdev;
245 bio_add_page(bio, page, PAGE_SIZE, 0);
246
247 bio->bi_end_io = end_bio_io_page;
248 bio->bi_private = page;
249 submit_bio(READ_SYNC | REQ_META, bio);
250 wait_on_page_locked(page);
251 bio_put(bio);
252 if (!PageUptodate(page)) {
253 __free_page(page);
254 return -EIO;
255 }
256 p = kmap(page);
257 gfs2_sb_in(sdp, p);
258 kunmap(page);
259 __free_page(page);
260 return gfs2_check_sb(sdp, silent);
261}
262
263/**
264 * gfs2_read_sb - Read super block
265 * @sdp: The GFS2 superblock
266 * @silent: Don't print message if mount fails
267 *
268 */
269
270static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
271{
272 u32 hash_blocks, ind_blocks, leaf_blocks;
273 u32 tmp_blocks;
274 unsigned int x;
275 int error;
276
277 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
278 if (error) {
279 if (!silent)
280 fs_err(sdp, "can't read superblock\n");
281 return error;
282 }
283
284 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
285 GFS2_BASIC_BLOCK_SHIFT;
286 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
287 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
288 sizeof(struct gfs2_dinode)) / sizeof(u64);
289 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
290 sizeof(struct gfs2_meta_header)) / sizeof(u64);
291 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
292 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
293 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
294 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
295 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
296 sizeof(struct gfs2_meta_header)) /
297 sizeof(struct gfs2_quota_change);
298 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
299 sizeof(struct gfs2_meta_header))
300 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
301
302 /* Compute maximum reservation required to add a entry to a directory */
303
304 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
305 sdp->sd_jbsize);
306
307 ind_blocks = 0;
308 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
309 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
310 ind_blocks += tmp_blocks;
311 }
312
313 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
314
315 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
316
317 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
318 sizeof(struct gfs2_dinode);
319 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
320 for (x = 2;; x++) {
321 u64 space, d;
322 u32 m;
323
324 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
325 d = space;
326 m = do_div(d, sdp->sd_inptrs);
327
328 if (d != sdp->sd_heightsize[x - 1] || m)
329 break;
330 sdp->sd_heightsize[x] = space;
331 }
332 sdp->sd_max_height = x;
333 sdp->sd_heightsize[x] = ~0;
334 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
335
336 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
337 sizeof(struct gfs2_dinode);
338 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
339 for (x = 2;; x++) {
340 u64 space, d;
341 u32 m;
342
343 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
344 d = space;
345 m = do_div(d, sdp->sd_inptrs);
346
347 if (d != sdp->sd_jheightsize[x - 1] || m)
348 break;
349 sdp->sd_jheightsize[x] = space;
350 }
351 sdp->sd_max_jheight = x;
352 sdp->sd_jheightsize[x] = ~0;
353 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
354
355 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
356 sizeof(struct gfs2_leaf)) /
357 GFS2_MIN_DIRENT_SIZE;
358 return 0;
359}
360
361static int init_names(struct gfs2_sbd *sdp, int silent)
362{
363 char *proto, *table;
364 int error = 0;
365
366 proto = sdp->sd_args.ar_lockproto;
367 table = sdp->sd_args.ar_locktable;
368
369 /* Try to autodetect */
370
371 if (!proto[0] || !table[0]) {
372 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
373 if (error)
374 return error;
375
376 if (!proto[0])
377 proto = sdp->sd_sb.sb_lockproto;
378 if (!table[0])
379 table = sdp->sd_sb.sb_locktable;
380 }
381
382 if (!table[0])
383 table = sdp->sd_vfs->s_id;
384
385 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
386 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
387
388 table = sdp->sd_table_name;
389 while ((table = strchr(table, '/')))
390 *table = '_';
391
392 return error;
393}
394
395static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
396 int undo)
397{
398 int error = 0;
399
400 if (undo)
401 goto fail_trans;
402
403 error = gfs2_glock_nq_num(sdp,
404 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
405 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
406 mount_gh);
407 if (error) {
408 fs_err(sdp, "can't acquire mount glock: %d\n", error);
409 goto fail;
410 }
411
412 error = gfs2_glock_nq_num(sdp,
413 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
414 LM_ST_SHARED,
415 LM_FLAG_NOEXP | GL_EXACT,
416 &sdp->sd_live_gh);
417 if (error) {
418 fs_err(sdp, "can't acquire live glock: %d\n", error);
419 goto fail_mount;
420 }
421
422 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
423 CREATE, &sdp->sd_rename_gl);
424 if (error) {
425 fs_err(sdp, "can't create rename glock: %d\n", error);
426 goto fail_live;
427 }
428
429 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
430 CREATE, &sdp->sd_freeze_gl);
431 if (error) {
432 fs_err(sdp, "can't create transaction glock: %d\n", error);
433 goto fail_rename;
434 }
435
436 return 0;
437
438fail_trans:
439 gfs2_glock_put(sdp->sd_freeze_gl);
440fail_rename:
441 gfs2_glock_put(sdp->sd_rename_gl);
442fail_live:
443 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
444fail_mount:
445 gfs2_glock_dq_uninit(mount_gh);
446fail:
447 return error;
448}
449
450static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
451 u64 no_addr, const char *name)
452{
453 struct gfs2_sbd *sdp = sb->s_fs_info;
454 struct dentry *dentry;
455 struct inode *inode;
456
457 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0);
458 if (IS_ERR(inode)) {
459 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
460 return PTR_ERR(inode);
461 }
462 dentry = d_make_root(inode);
463 if (!dentry) {
464 fs_err(sdp, "can't alloc %s dentry\n", name);
465 return -ENOMEM;
466 }
467 *dptr = dentry;
468 return 0;
469}
470
471static int init_sb(struct gfs2_sbd *sdp, int silent)
472{
473 struct super_block *sb = sdp->sd_vfs;
474 struct gfs2_holder sb_gh;
475 u64 no_addr;
476 int ret;
477
478 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
479 LM_ST_SHARED, 0, &sb_gh);
480 if (ret) {
481 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
482 return ret;
483 }
484
485 ret = gfs2_read_sb(sdp, silent);
486 if (ret) {
487 fs_err(sdp, "can't read superblock: %d\n", ret);
488 goto out;
489 }
490
491 /* Set up the buffer cache and SB for real */
492 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
493 ret = -EINVAL;
494 fs_err(sdp, "FS block size (%u) is too small for device "
495 "block size (%u)\n",
496 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
497 goto out;
498 }
499 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
500 ret = -EINVAL;
501 fs_err(sdp, "FS block size (%u) is too big for machine "
502 "page size (%u)\n",
503 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
504 goto out;
505 }
506 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
507
508 /* Get the root inode */
509 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
510 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
511 if (ret)
512 goto out;
513
514 /* Get the master inode */
515 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
516 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
517 if (ret) {
518 dput(sdp->sd_root_dir);
519 goto out;
520 }
521 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
522out:
523 gfs2_glock_dq_uninit(&sb_gh);
524 return ret;
525}
526
527static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
528{
529 char *message = "FIRSTMOUNT=Done";
530 char *envp[] = { message, NULL };
531
532 fs_info(sdp, "first mount done, others may mount\n");
533
534 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
535 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
536
537 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
538}
539
540/**
541 * gfs2_jindex_hold - Grab a lock on the jindex
542 * @sdp: The GFS2 superblock
543 * @ji_gh: the holder for the jindex glock
544 *
545 * Returns: errno
546 */
547
548static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
549{
550 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
551 struct qstr name;
552 char buf[20];
553 struct gfs2_jdesc *jd;
554 int error;
555
556 name.name = buf;
557
558 mutex_lock(&sdp->sd_jindex_mutex);
559
560 for (;;) {
561 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
562 if (error)
563 break;
564
565 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
566 name.hash = gfs2_disk_hash(name.name, name.len);
567
568 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
569 if (error == -ENOENT) {
570 error = 0;
571 break;
572 }
573
574 gfs2_glock_dq_uninit(ji_gh);
575
576 if (error)
577 break;
578
579 error = -ENOMEM;
580 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
581 if (!jd)
582 break;
583
584 INIT_LIST_HEAD(&jd->extent_list);
585 INIT_LIST_HEAD(&jd->jd_revoke_list);
586
587 INIT_WORK(&jd->jd_work, gfs2_recover_func);
588 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
589 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
590 if (!jd->jd_inode)
591 error = -ENOENT;
592 else
593 error = PTR_ERR(jd->jd_inode);
594 kfree(jd);
595 break;
596 }
597
598 spin_lock(&sdp->sd_jindex_spin);
599 jd->jd_jid = sdp->sd_journals++;
600 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
601 spin_unlock(&sdp->sd_jindex_spin);
602 }
603
604 mutex_unlock(&sdp->sd_jindex_mutex);
605
606 return error;
607}
608
609/**
610 * check_journal_clean - Make sure a journal is clean for a spectator mount
611 * @sdp: The GFS2 superblock
612 * @jd: The journal descriptor
613 *
614 * Returns: 0 if the journal is clean or locked, else an error
615 */
616static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
617{
618 int error;
619 struct gfs2_holder j_gh;
620 struct gfs2_log_header_host head;
621 struct gfs2_inode *ip;
622
623 ip = GFS2_I(jd->jd_inode);
624 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
625 GL_EXACT | GL_NOCACHE, &j_gh);
626 if (error) {
627 fs_err(sdp, "Error locking journal for spectator mount.\n");
628 return -EPERM;
629 }
630 error = gfs2_jdesc_check(jd);
631 if (error) {
632 fs_err(sdp, "Error checking journal for spectator mount.\n");
633 goto out_unlock;
634 }
635 error = gfs2_find_jhead(jd, &head);
636 if (error) {
637 fs_err(sdp, "Error parsing journal for spectator mount.\n");
638 goto out_unlock;
639 }
640 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
641 error = -EPERM;
642 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
643 "must not be a spectator.\n", jd->jd_jid);
644 }
645
646out_unlock:
647 gfs2_glock_dq_uninit(&j_gh);
648 return error;
649}
650
651static int init_journal(struct gfs2_sbd *sdp, int undo)
652{
653 struct inode *master = d_inode(sdp->sd_master_dir);
654 struct gfs2_holder ji_gh;
655 struct gfs2_inode *ip;
656 int jindex = 1;
657 int error = 0;
658
659 if (undo) {
660 jindex = 0;
661 goto fail_jinode_gh;
662 }
663
664 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
665 if (IS_ERR(sdp->sd_jindex)) {
666 fs_err(sdp, "can't lookup journal index: %d\n", error);
667 return PTR_ERR(sdp->sd_jindex);
668 }
669
670 /* Load in the journal index special file */
671
672 error = gfs2_jindex_hold(sdp, &ji_gh);
673 if (error) {
674 fs_err(sdp, "can't read journal index: %d\n", error);
675 goto fail;
676 }
677
678 error = -EUSERS;
679 if (!gfs2_jindex_size(sdp)) {
680 fs_err(sdp, "no journals!\n");
681 goto fail_jindex;
682 }
683
684 if (sdp->sd_args.ar_spectator) {
685 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
686 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
687 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
688 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
689 } else {
690 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
691 fs_err(sdp, "can't mount journal #%u\n",
692 sdp->sd_lockstruct.ls_jid);
693 fs_err(sdp, "there are only %u journals (0 - %u)\n",
694 gfs2_jindex_size(sdp),
695 gfs2_jindex_size(sdp) - 1);
696 goto fail_jindex;
697 }
698 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
699
700 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
701 &gfs2_journal_glops,
702 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
703 &sdp->sd_journal_gh);
704 if (error) {
705 fs_err(sdp, "can't acquire journal glock: %d\n", error);
706 goto fail_jindex;
707 }
708
709 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
710 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
711 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
712 &sdp->sd_jinode_gh);
713 if (error) {
714 fs_err(sdp, "can't acquire journal inode glock: %d\n",
715 error);
716 goto fail_journal_gh;
717 }
718
719 error = gfs2_jdesc_check(sdp->sd_jdesc);
720 if (error) {
721 fs_err(sdp, "my journal (%u) is bad: %d\n",
722 sdp->sd_jdesc->jd_jid, error);
723 goto fail_jinode_gh;
724 }
725 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
726 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
727 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
728
729 /* Map the extents for this journal's blocks */
730 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
731 }
732 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
733
734 if (sdp->sd_lockstruct.ls_first) {
735 unsigned int x;
736 for (x = 0; x < sdp->sd_journals; x++) {
737 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
738
739 if (sdp->sd_args.ar_spectator) {
740 error = check_journal_clean(sdp, jd);
741 if (error)
742 goto fail_jinode_gh;
743 continue;
744 }
745 error = gfs2_recover_journal(jd, true);
746 if (error) {
747 fs_err(sdp, "error recovering journal %u: %d\n",
748 x, error);
749 goto fail_jinode_gh;
750 }
751 }
752
753 gfs2_others_may_mount(sdp);
754 } else if (!sdp->sd_args.ar_spectator) {
755 error = gfs2_recover_journal(sdp->sd_jdesc, true);
756 if (error) {
757 fs_err(sdp, "error recovering my journal: %d\n", error);
758 goto fail_jinode_gh;
759 }
760 }
761
762 sdp->sd_log_idle = 1;
763 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
764 gfs2_glock_dq_uninit(&ji_gh);
765 jindex = 0;
766 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
767 return 0;
768
769fail_jinode_gh:
770 if (!sdp->sd_args.ar_spectator)
771 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
772fail_journal_gh:
773 if (!sdp->sd_args.ar_spectator)
774 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
775fail_jindex:
776 gfs2_jindex_free(sdp);
777 if (jindex)
778 gfs2_glock_dq_uninit(&ji_gh);
779fail:
780 iput(sdp->sd_jindex);
781 return error;
782}
783
784static struct lock_class_key gfs2_quota_imutex_key;
785
786static int init_inodes(struct gfs2_sbd *sdp, int undo)
787{
788 int error = 0;
789 struct inode *master = d_inode(sdp->sd_master_dir);
790
791 if (undo)
792 goto fail_qinode;
793
794 error = init_journal(sdp, undo);
795 complete_all(&sdp->sd_journal_ready);
796 if (error)
797 goto fail;
798
799 /* Read in the master statfs inode */
800 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
801 if (IS_ERR(sdp->sd_statfs_inode)) {
802 error = PTR_ERR(sdp->sd_statfs_inode);
803 fs_err(sdp, "can't read in statfs inode: %d\n", error);
804 goto fail_journal;
805 }
806
807 /* Read in the resource index inode */
808 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
809 if (IS_ERR(sdp->sd_rindex)) {
810 error = PTR_ERR(sdp->sd_rindex);
811 fs_err(sdp, "can't get resource index inode: %d\n", error);
812 goto fail_statfs;
813 }
814 sdp->sd_rindex_uptodate = 0;
815
816 /* Read in the quota inode */
817 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
818 if (IS_ERR(sdp->sd_quota_inode)) {
819 error = PTR_ERR(sdp->sd_quota_inode);
820 fs_err(sdp, "can't get quota file inode: %d\n", error);
821 goto fail_rindex;
822 }
823 /*
824 * i_mutex on quota files is special. Since this inode is hidden system
825 * file, we are safe to define locking ourselves.
826 */
827 lockdep_set_class(&sdp->sd_quota_inode->i_mutex,
828 &gfs2_quota_imutex_key);
829
830 error = gfs2_rindex_update(sdp);
831 if (error)
832 goto fail_qinode;
833
834 return 0;
835
836fail_qinode:
837 iput(sdp->sd_quota_inode);
838fail_rindex:
839 gfs2_clear_rgrpd(sdp);
840 iput(sdp->sd_rindex);
841fail_statfs:
842 iput(sdp->sd_statfs_inode);
843fail_journal:
844 init_journal(sdp, UNDO);
845fail:
846 return error;
847}
848
849static int init_per_node(struct gfs2_sbd *sdp, int undo)
850{
851 struct inode *pn = NULL;
852 char buf[30];
853 int error = 0;
854 struct gfs2_inode *ip;
855 struct inode *master = d_inode(sdp->sd_master_dir);
856
857 if (sdp->sd_args.ar_spectator)
858 return 0;
859
860 if (undo)
861 goto fail_qc_gh;
862
863 pn = gfs2_lookup_simple(master, "per_node");
864 if (IS_ERR(pn)) {
865 error = PTR_ERR(pn);
866 fs_err(sdp, "can't find per_node directory: %d\n", error);
867 return error;
868 }
869
870 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
871 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
872 if (IS_ERR(sdp->sd_sc_inode)) {
873 error = PTR_ERR(sdp->sd_sc_inode);
874 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
875 goto fail;
876 }
877
878 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
879 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
880 if (IS_ERR(sdp->sd_qc_inode)) {
881 error = PTR_ERR(sdp->sd_qc_inode);
882 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
883 goto fail_ut_i;
884 }
885
886 iput(pn);
887 pn = NULL;
888
889 ip = GFS2_I(sdp->sd_sc_inode);
890 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
891 &sdp->sd_sc_gh);
892 if (error) {
893 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
894 goto fail_qc_i;
895 }
896
897 ip = GFS2_I(sdp->sd_qc_inode);
898 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
899 &sdp->sd_qc_gh);
900 if (error) {
901 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
902 goto fail_ut_gh;
903 }
904
905 return 0;
906
907fail_qc_gh:
908 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
909fail_ut_gh:
910 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
911fail_qc_i:
912 iput(sdp->sd_qc_inode);
913fail_ut_i:
914 iput(sdp->sd_sc_inode);
915fail:
916 iput(pn);
917 return error;
918}
919
920static const match_table_t nolock_tokens = {
921 { Opt_jid, "jid=%d\n", },
922 { Opt_err, NULL },
923};
924
925static const struct lm_lockops nolock_ops = {
926 .lm_proto_name = "lock_nolock",
927 .lm_put_lock = gfs2_glock_free,
928 .lm_tokens = &nolock_tokens,
929};
930
931/**
932 * gfs2_lm_mount - mount a locking protocol
933 * @sdp: the filesystem
934 * @args: mount arguments
935 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
936 *
937 * Returns: errno
938 */
939
940static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
941{
942 const struct lm_lockops *lm;
943 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
944 struct gfs2_args *args = &sdp->sd_args;
945 const char *proto = sdp->sd_proto_name;
946 const char *table = sdp->sd_table_name;
947 char *o, *options;
948 int ret;
949
950 if (!strcmp("lock_nolock", proto)) {
951 lm = &nolock_ops;
952 sdp->sd_args.ar_localflocks = 1;
953#ifdef CONFIG_GFS2_FS_LOCKING_DLM
954 } else if (!strcmp("lock_dlm", proto)) {
955 lm = &gfs2_dlm_ops;
956#endif
957 } else {
958 pr_info("can't find protocol %s\n", proto);
959 return -ENOENT;
960 }
961
962 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
963
964 ls->ls_ops = lm;
965 ls->ls_first = 1;
966
967 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
968 substring_t tmp[MAX_OPT_ARGS];
969 int token, option;
970
971 if (!o || !*o)
972 continue;
973
974 token = match_token(o, *lm->lm_tokens, tmp);
975 switch (token) {
976 case Opt_jid:
977 ret = match_int(&tmp[0], &option);
978 if (ret || option < 0)
979 goto hostdata_error;
980 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
981 ls->ls_jid = option;
982 break;
983 case Opt_id:
984 case Opt_nodir:
985 /* Obsolete, but left for backward compat purposes */
986 break;
987 case Opt_first:
988 ret = match_int(&tmp[0], &option);
989 if (ret || (option != 0 && option != 1))
990 goto hostdata_error;
991 ls->ls_first = option;
992 break;
993 case Opt_err:
994 default:
995hostdata_error:
996 fs_info(sdp, "unknown hostdata (%s)\n", o);
997 return -EINVAL;
998 }
999 }
1000
1001 if (lm->lm_mount == NULL) {
1002 fs_info(sdp, "Now mounting FS...\n");
1003 complete_all(&sdp->sd_locking_init);
1004 return 0;
1005 }
1006 ret = lm->lm_mount(sdp, table);
1007 if (ret == 0)
1008 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1009 complete_all(&sdp->sd_locking_init);
1010 return ret;
1011}
1012
1013void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1014{
1015 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1016 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1017 lm->lm_unmount)
1018 lm->lm_unmount(sdp);
1019}
1020
1021static int wait_on_journal(struct gfs2_sbd *sdp)
1022{
1023 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1024 return 0;
1025
1026 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1027 ? -EINTR : 0;
1028}
1029
1030void gfs2_online_uevent(struct gfs2_sbd *sdp)
1031{
1032 struct super_block *sb = sdp->sd_vfs;
1033 char ro[20];
1034 char spectator[20];
1035 char *envp[] = { ro, spectator, NULL };
1036 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1037 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1038 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1039}
1040
1041/**
1042 * fill_super - Read in superblock
1043 * @sb: The VFS superblock
1044 * @data: Mount options
1045 * @silent: Don't complain if it's not a GFS2 filesystem
1046 *
1047 * Returns: errno
1048 */
1049
1050static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1051{
1052 struct gfs2_sbd *sdp;
1053 struct gfs2_holder mount_gh;
1054 int error;
1055
1056 sdp = init_sbd(sb);
1057 if (!sdp) {
1058 pr_warn("can't alloc struct gfs2_sbd\n");
1059 return -ENOMEM;
1060 }
1061 sdp->sd_args = *args;
1062
1063 if (sdp->sd_args.ar_spectator) {
1064 sb->s_flags |= MS_RDONLY;
1065 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1066 }
1067 if (sdp->sd_args.ar_posix_acl)
1068 sb->s_flags |= MS_POSIXACL;
1069 if (sdp->sd_args.ar_nobarrier)
1070 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1071
1072 sb->s_flags |= MS_NOSEC;
1073 sb->s_magic = GFS2_MAGIC;
1074 sb->s_op = &gfs2_super_ops;
1075 sb->s_d_op = &gfs2_dops;
1076 sb->s_export_op = &gfs2_export_ops;
1077 sb->s_xattr = gfs2_xattr_handlers;
1078 sb->s_qcop = &gfs2_quotactl_ops;
1079 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1080 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1081 sb->s_time_gran = 1;
1082 sb->s_maxbytes = MAX_LFS_FILESIZE;
1083
1084 /* Set up the buffer cache and fill in some fake block size values
1085 to allow us to read-in the on-disk superblock. */
1086 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1087 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1088 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1089 GFS2_BASIC_BLOCK_SHIFT;
1090 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1091
1092 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1093 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1094 if (sdp->sd_args.ar_statfs_quantum) {
1095 sdp->sd_tune.gt_statfs_slow = 0;
1096 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1097 } else {
1098 sdp->sd_tune.gt_statfs_slow = 1;
1099 sdp->sd_tune.gt_statfs_quantum = 30;
1100 }
1101
1102 error = init_names(sdp, silent);
1103 if (error) {
1104 /* In this case, we haven't initialized sysfs, so we have to
1105 manually free the sdp. */
1106 free_percpu(sdp->sd_lkstats);
1107 kfree(sdp);
1108 sb->s_fs_info = NULL;
1109 return error;
1110 }
1111
1112 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1113
1114 error = gfs2_sys_fs_add(sdp);
1115 /*
1116 * If we hit an error here, gfs2_sys_fs_add will have called function
1117 * kobject_put which causes the sysfs usage count to go to zero, which
1118 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1119 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1120 * kobject_put to free sdp.
1121 */
1122 if (error)
1123 return error;
1124
1125 gfs2_create_debugfs_file(sdp);
1126
1127 error = gfs2_lm_mount(sdp, silent);
1128 if (error)
1129 goto fail_debug;
1130
1131 error = init_locking(sdp, &mount_gh, DO);
1132 if (error)
1133 goto fail_lm;
1134
1135 error = init_sb(sdp, silent);
1136 if (error)
1137 goto fail_locking;
1138
1139 error = wait_on_journal(sdp);
1140 if (error)
1141 goto fail_sb;
1142
1143 /*
1144 * If user space has failed to join the cluster or some similar
1145 * failure has occurred, then the journal id will contain a
1146 * negative (error) number. This will then be returned to the
1147 * caller (of the mount syscall). We do this even for spectator
1148 * mounts (which just write a jid of 0 to indicate "ok" even though
1149 * the jid is unused in the spectator case)
1150 */
1151 if (sdp->sd_lockstruct.ls_jid < 0) {
1152 error = sdp->sd_lockstruct.ls_jid;
1153 sdp->sd_lockstruct.ls_jid = 0;
1154 goto fail_sb;
1155 }
1156
1157 if (sdp->sd_args.ar_spectator)
1158 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1159 sdp->sd_table_name);
1160 else
1161 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1162 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1163
1164 error = init_inodes(sdp, DO);
1165 if (error)
1166 goto fail_sb;
1167
1168 error = init_per_node(sdp, DO);
1169 if (error)
1170 goto fail_inodes;
1171
1172 error = gfs2_statfs_init(sdp);
1173 if (error) {
1174 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1175 goto fail_per_node;
1176 }
1177
1178 if (!(sb->s_flags & MS_RDONLY)) {
1179 error = gfs2_make_fs_rw(sdp);
1180 if (error) {
1181 fs_err(sdp, "can't make FS RW: %d\n", error);
1182 goto fail_per_node;
1183 }
1184 }
1185
1186 gfs2_glock_dq_uninit(&mount_gh);
1187 gfs2_online_uevent(sdp);
1188 return 0;
1189
1190fail_per_node:
1191 init_per_node(sdp, UNDO);
1192fail_inodes:
1193 init_inodes(sdp, UNDO);
1194fail_sb:
1195 if (sdp->sd_root_dir)
1196 dput(sdp->sd_root_dir);
1197 if (sdp->sd_master_dir)
1198 dput(sdp->sd_master_dir);
1199 if (sb->s_root)
1200 dput(sb->s_root);
1201 sb->s_root = NULL;
1202fail_locking:
1203 init_locking(sdp, &mount_gh, UNDO);
1204fail_lm:
1205 complete_all(&sdp->sd_journal_ready);
1206 gfs2_gl_hash_clear(sdp);
1207 gfs2_lm_unmount(sdp);
1208fail_debug:
1209 gfs2_delete_debugfs_file(sdp);
1210 free_percpu(sdp->sd_lkstats);
1211 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1212 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1213 gfs2_sys_fs_del(sdp);
1214 sb->s_fs_info = NULL;
1215 return error;
1216}
1217
1218static int set_gfs2_super(struct super_block *s, void *data)
1219{
1220 s->s_bdev = data;
1221 s->s_dev = s->s_bdev->bd_dev;
1222
1223 /*
1224 * We set the bdi here to the queue backing, file systems can
1225 * overwrite this in ->fill_super()
1226 */
1227 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1228 return 0;
1229}
1230
1231static int test_gfs2_super(struct super_block *s, void *ptr)
1232{
1233 struct block_device *bdev = ptr;
1234 return (bdev == s->s_bdev);
1235}
1236
1237/**
1238 * gfs2_mount - Get the GFS2 superblock
1239 * @fs_type: The GFS2 filesystem type
1240 * @flags: Mount flags
1241 * @dev_name: The name of the device
1242 * @data: The mount arguments
1243 *
1244 * Q. Why not use get_sb_bdev() ?
1245 * A. We need to select one of two root directories to mount, independent
1246 * of whether this is the initial, or subsequent, mount of this sb
1247 *
1248 * Returns: 0 or -ve on error
1249 */
1250
1251static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1252 const char *dev_name, void *data)
1253{
1254 struct block_device *bdev;
1255 struct super_block *s;
1256 fmode_t mode = FMODE_READ | FMODE_EXCL;
1257 int error;
1258 struct gfs2_args args;
1259 struct gfs2_sbd *sdp;
1260
1261 if (!(flags & MS_RDONLY))
1262 mode |= FMODE_WRITE;
1263
1264 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1265 if (IS_ERR(bdev))
1266 return ERR_CAST(bdev);
1267
1268 /*
1269 * once the super is inserted into the list by sget, s_umount
1270 * will protect the lockfs code from trying to start a snapshot
1271 * while we are mounting
1272 */
1273 mutex_lock(&bdev->bd_fsfreeze_mutex);
1274 if (bdev->bd_fsfreeze_count > 0) {
1275 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1276 error = -EBUSY;
1277 goto error_bdev;
1278 }
1279 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1281 error = PTR_ERR(s);
1282 if (IS_ERR(s))
1283 goto error_bdev;
1284
1285 if (s->s_root) {
1286 /*
1287 * s_umount nests inside bd_mutex during
1288 * __invalidate_device(). blkdev_put() acquires
1289 * bd_mutex and can't be called under s_umount. Drop
1290 * s_umount temporarily. This is safe as we're
1291 * holding an active reference.
1292 */
1293 up_write(&s->s_umount);
1294 blkdev_put(bdev, mode);
1295 down_write(&s->s_umount);
1296 } else {
1297 /* s_mode must be set before deactivate_locked_super calls */
1298 s->s_mode = mode;
1299 }
1300
1301 memset(&args, 0, sizeof(args));
1302 args.ar_quota = GFS2_QUOTA_DEFAULT;
1303 args.ar_data = GFS2_DATA_DEFAULT;
1304 args.ar_commit = 30;
1305 args.ar_statfs_quantum = 30;
1306 args.ar_quota_quantum = 60;
1307 args.ar_errors = GFS2_ERRORS_DEFAULT;
1308
1309 error = gfs2_mount_args(&args, data);
1310 if (error) {
1311 pr_warn("can't parse mount arguments\n");
1312 goto error_super;
1313 }
1314
1315 if (s->s_root) {
1316 error = -EBUSY;
1317 if ((flags ^ s->s_flags) & MS_RDONLY)
1318 goto error_super;
1319 } else {
1320 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1321 sb_set_blocksize(s, block_size(bdev));
1322 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1323 if (error)
1324 goto error_super;
1325 s->s_flags |= MS_ACTIVE;
1326 bdev->bd_super = s;
1327 }
1328
1329 sdp = s->s_fs_info;
1330 if (args.ar_meta)
1331 return dget(sdp->sd_master_dir);
1332 else
1333 return dget(sdp->sd_root_dir);
1334
1335error_super:
1336 deactivate_locked_super(s);
1337 return ERR_PTR(error);
1338error_bdev:
1339 blkdev_put(bdev, mode);
1340 return ERR_PTR(error);
1341}
1342
1343static int set_meta_super(struct super_block *s, void *ptr)
1344{
1345 return -EINVAL;
1346}
1347
1348static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1349 int flags, const char *dev_name, void *data)
1350{
1351 struct super_block *s;
1352 struct gfs2_sbd *sdp;
1353 struct path path;
1354 int error;
1355
1356 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1357 if (error) {
1358 pr_warn("path_lookup on %s returned error %d\n",
1359 dev_name, error);
1360 return ERR_PTR(error);
1361 }
1362 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1363 d_inode(path.dentry)->i_sb->s_bdev);
1364 path_put(&path);
1365 if (IS_ERR(s)) {
1366 pr_warn("gfs2 mount does not exist\n");
1367 return ERR_CAST(s);
1368 }
1369 if ((flags ^ s->s_flags) & MS_RDONLY) {
1370 deactivate_locked_super(s);
1371 return ERR_PTR(-EBUSY);
1372 }
1373 sdp = s->s_fs_info;
1374 return dget(sdp->sd_master_dir);
1375}
1376
1377static void gfs2_kill_sb(struct super_block *sb)
1378{
1379 struct gfs2_sbd *sdp = sb->s_fs_info;
1380
1381 if (sdp == NULL) {
1382 kill_block_super(sb);
1383 return;
1384 }
1385
1386 gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1387 dput(sdp->sd_root_dir);
1388 dput(sdp->sd_master_dir);
1389 sdp->sd_root_dir = NULL;
1390 sdp->sd_master_dir = NULL;
1391 shrink_dcache_sb(sb);
1392 gfs2_delete_debugfs_file(sdp);
1393 free_percpu(sdp->sd_lkstats);
1394 kill_block_super(sb);
1395}
1396
1397struct file_system_type gfs2_fs_type = {
1398 .name = "gfs2",
1399 .fs_flags = FS_REQUIRES_DEV,
1400 .mount = gfs2_mount,
1401 .kill_sb = gfs2_kill_sb,
1402 .owner = THIS_MODULE,
1403};
1404MODULE_ALIAS_FS("gfs2");
1405
1406struct file_system_type gfs2meta_fs_type = {
1407 .name = "gfs2meta",
1408 .fs_flags = FS_REQUIRES_DEV,
1409 .mount = gfs2_mount_meta,
1410 .owner = THIS_MODULE,
1411};
1412MODULE_ALIAS_FS("gfs2meta");