1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/blkdev.h>
18 #include <linux/kthread.h>
19 #include <linux/export.h>
20 #include <linux/namei.h>
21 #include <linux/mount.h>
22 #include <linux/gfs2_ondisk.h>
23 #include <linux/quotaops.h>
24 #include <linux/lockdep.h>
25 #include <linux/module.h>
26
27 #include "gfs2.h"
28 #include "incore.h"
29 #include "bmap.h"
30 #include "glock.h"
31 #include "glops.h"
32 #include "inode.h"
33 #include "recovery.h"
34 #include "rgrp.h"
35 #include "super.h"
36 #include "sys.h"
37 #include "util.h"
38 #include "log.h"
39 #include "quota.h"
40 #include "dir.h"
41 #include "meta_io.h"
42 #include "trace_gfs2.h"
43
44 #define DO 0
45 #define UNDO 1
46
47 /**
48 * gfs2_tune_init - Fill a gfs2_tune structure with default values
49 * @gt: tune
50 *
51 */
52
53 static void gfs2_tune_init(struct gfs2_tune *gt)
54 {
55 spin_lock_init(>->gt_spin);
56
57 gt->gt_quota_warn_period = 10;
58 gt->gt_quota_scale_num = 1;
59 gt->gt_quota_scale_den = 1;
60 gt->gt_new_files_jdata = 0;
61 gt->gt_max_readahead = BIT(18);
62 gt->gt_complain_secs = 10;
63 }
64
65 static struct gfs2_sbd *init_sbd(struct super_block *sb)
66 {
67 struct gfs2_sbd *sdp;
68 struct address_space *mapping;
69
70 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
71 if (!sdp)
72 return NULL;
73
74 sdp->sd_vfs = sb;
75 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
76 if (!sdp->sd_lkstats) {
77 kfree(sdp);
78 return NULL;
79 }
80 sb->s_fs_info = sdp;
81
82 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
83 gfs2_tune_init(&sdp->sd_tune);
84
85 init_waitqueue_head(&sdp->sd_glock_wait);
86 atomic_set(&sdp->sd_glock_disposal, 0);
87 init_completion(&sdp->sd_locking_init);
88 init_completion(&sdp->sd_wdack);
89 spin_lock_init(&sdp->sd_statfs_spin);
90
91 spin_lock_init(&sdp->sd_rindex_spin);
92 sdp->sd_rindex_tree.rb_node = NULL;
93
94 INIT_LIST_HEAD(&sdp->sd_jindex_list);
95 spin_lock_init(&sdp->sd_jindex_spin);
96 mutex_init(&sdp->sd_jindex_mutex);
97 init_completion(&sdp->sd_journal_ready);
98
99 INIT_LIST_HEAD(&sdp->sd_quota_list);
100 mutex_init(&sdp->sd_quota_mutex);
101 mutex_init(&sdp->sd_quota_sync_mutex);
102 init_waitqueue_head(&sdp->sd_quota_wait);
103 INIT_LIST_HEAD(&sdp->sd_trunc_list);
104 spin_lock_init(&sdp->sd_trunc_lock);
105 spin_lock_init(&sdp->sd_bitmap_lock);
106
107 mapping = &sdp->sd_aspace;
108
109 address_space_init_once(mapping);
110 mapping->a_ops = &gfs2_rgrp_aops;
111 mapping->host = sb->s_bdev->bd_inode;
112 mapping->flags = 0;
113 mapping_set_gfp_mask(mapping, GFP_NOFS);
114 mapping->private_data = NULL;
115 mapping->writeback_index = 0;
116
117 spin_lock_init(&sdp->sd_log_lock);
118 atomic_set(&sdp->sd_log_pinned, 0);
119 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
120 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
121 spin_lock_init(&sdp->sd_ordered_lock);
122
123 init_waitqueue_head(&sdp->sd_log_waitq);
124 init_waitqueue_head(&sdp->sd_logd_waitq);
125 spin_lock_init(&sdp->sd_ail_lock);
126 INIT_LIST_HEAD(&sdp->sd_ail1_list);
127 INIT_LIST_HEAD(&sdp->sd_ail2_list);
128
129 init_rwsem(&sdp->sd_log_flush_lock);
130 atomic_set(&sdp->sd_log_in_flight, 0);
131 atomic_set(&sdp->sd_reserving_log, 0);
132 init_waitqueue_head(&sdp->sd_reserving_log_wait);
133 init_waitqueue_head(&sdp->sd_log_flush_wait);
134 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
135 mutex_init(&sdp->sd_freeze_mutex);
136
137 return sdp;
138 }
139
140
141 /**
142 * gfs2_check_sb - Check superblock
143 * @sdp: the filesystem
144 * @sb: The superblock
145 * @silent: Don't print a message if the check fails
146 *
147 * Checks the version code of the FS is one that we understand how to
148 * read and that the sizes of the various on-disk structures have not
149 * changed.
150 */
151
152 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
153 {
154 struct gfs2_sb_host *sb = &sdp->sd_sb;
155
156 if (sb->sb_magic != GFS2_MAGIC ||
157 sb->sb_type != GFS2_METATYPE_SB) {
158 if (!silent)
159 pr_warn("not a GFS2 filesystem\n");
160 return -EINVAL;
161 }
162
163 if (sb->sb_fs_format != GFS2_FORMAT_FS ||
164 sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
165 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
166 return -EINVAL;
167 }
168
169 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
170 (sb->sb_bsize & (sb->sb_bsize - 1))) {
171 pr_warn("Invalid superblock size\n");
172 return -EINVAL;
173 }
174 if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
175 pr_warn("Invalid block size shift\n");
176 return -EINVAL;
177 }
178 return 0;
179 }
180
181 static void end_bio_io_page(struct bio *bio)
182 {
183 struct page *page = bio->bi_private;
184
185 if (!bio->bi_error)
186 SetPageUptodate(page);
187 else
188 pr_warn("error %d reading superblock\n", bio->bi_error);
189 unlock_page(page);
190 }
191
192 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
193 {
194 struct gfs2_sb_host *sb = &sdp->sd_sb;
195 struct super_block *s = sdp->sd_vfs;
196 const struct gfs2_sb *str = buf;
197
198 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
199 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
200 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
201 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
202 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
203 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
204 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
205 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
206 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
207 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
208 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
209
210 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
211 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
212 memcpy(s->s_uuid, str->sb_uuid, 16);
213 }
214
215 /**
216 * gfs2_read_super - Read the gfs2 super block from disk
217 * @sdp: The GFS2 super block
218 * @sector: The location of the super block
219 * @error: The error code to return
220 *
221 * This uses the bio functions to read the super block from disk
222 * because we want to be 100% sure that we never read cached data.
223 * A super block is read twice only during each GFS2 mount and is
224 * never written to by the filesystem. The first time its read no
225 * locks are held, and the only details which are looked at are those
226 * relating to the locking protocol. Once locking is up and working,
227 * the sb is read again under the lock to establish the location of
228 * the master directory (contains pointers to journals etc) and the
229 * root directory.
230 *
231 * Returns: 0 on success or error
232 */
233
234 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
235 {
236 struct super_block *sb = sdp->sd_vfs;
237 struct gfs2_sb *p;
238 struct page *page;
239 struct bio *bio;
240
241 page = alloc_page(GFP_NOFS);
242 if (unlikely(!page))
243 return -ENOMEM;
244
245 ClearPageUptodate(page);
246 ClearPageDirty(page);
247 lock_page(page);
248
249 bio = bio_alloc(GFP_NOFS, 1);
250 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
251 bio->bi_bdev = sb->s_bdev;
252 bio_add_page(bio, page, PAGE_SIZE, 0);
253
254 bio->bi_end_io = end_bio_io_page;
255 bio->bi_private = page;
256 bio_set_op_attrs(bio, REQ_OP_READ, READ_SYNC | REQ_META);
257 submit_bio(bio);
258 wait_on_page_locked(page);
259 bio_put(bio);
260 if (!PageUptodate(page)) {
261 __free_page(page);
262 return -EIO;
263 }
264 p = kmap(page);
265 gfs2_sb_in(sdp, p);
266 kunmap(page);
267 __free_page(page);
268 return gfs2_check_sb(sdp, silent);
269 }
270
271 /**
272 * gfs2_read_sb - Read super block
273 * @sdp: The GFS2 superblock
274 * @silent: Don't print message if mount fails
275 *
276 */
277
278 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
279 {
280 u32 hash_blocks, ind_blocks, leaf_blocks;
281 u32 tmp_blocks;
282 unsigned int x;
283 int error;
284
285 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
286 if (error) {
287 if (!silent)
288 fs_err(sdp, "can't read superblock\n");
289 return error;
290 }
291
292 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
293 GFS2_BASIC_BLOCK_SHIFT;
294 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
295 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
296 sizeof(struct gfs2_dinode)) / sizeof(u64);
297 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
298 sizeof(struct gfs2_meta_header)) / sizeof(u64);
299 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
300 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
301 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
302 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
303 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
304 sizeof(struct gfs2_meta_header)) /
305 sizeof(struct gfs2_quota_change);
306 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
307 sizeof(struct gfs2_meta_header))
308 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
309
310 /* Compute maximum reservation required to add a entry to a directory */
311
312 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
313 sdp->sd_jbsize);
314
315 ind_blocks = 0;
316 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
317 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
318 ind_blocks += tmp_blocks;
319 }
320
321 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
322
323 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
324
325 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
326 sizeof(struct gfs2_dinode);
327 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
328 for (x = 2;; x++) {
329 u64 space, d;
330 u32 m;
331
332 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
333 d = space;
334 m = do_div(d, sdp->sd_inptrs);
335
336 if (d != sdp->sd_heightsize[x - 1] || m)
337 break;
338 sdp->sd_heightsize[x] = space;
339 }
340 sdp->sd_max_height = x;
341 sdp->sd_heightsize[x] = ~0;
342 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
343
344 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
345 sizeof(struct gfs2_dinode);
346 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
347 for (x = 2;; x++) {
348 u64 space, d;
349 u32 m;
350
351 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
352 d = space;
353 m = do_div(d, sdp->sd_inptrs);
354
355 if (d != sdp->sd_jheightsize[x - 1] || m)
356 break;
357 sdp->sd_jheightsize[x] = space;
358 }
359 sdp->sd_max_jheight = x;
360 sdp->sd_jheightsize[x] = ~0;
361 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
362
363 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
364 sizeof(struct gfs2_leaf)) /
365 GFS2_MIN_DIRENT_SIZE;
366 return 0;
367 }
368
369 static int init_names(struct gfs2_sbd *sdp, int silent)
370 {
371 char *proto, *table;
372 int error = 0;
373
374 proto = sdp->sd_args.ar_lockproto;
375 table = sdp->sd_args.ar_locktable;
376
377 /* Try to autodetect */
378
379 if (!proto[0] || !table[0]) {
380 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
381 if (error)
382 return error;
383
384 if (!proto[0])
385 proto = sdp->sd_sb.sb_lockproto;
386 if (!table[0])
387 table = sdp->sd_sb.sb_locktable;
388 }
389
390 if (!table[0])
391 table = sdp->sd_vfs->s_id;
392
393 BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
394
395 strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
396 strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
397
398 table = sdp->sd_table_name;
399 while ((table = strchr(table, '/')))
400 *table = '_';
401
402 return error;
403 }
404
405 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
406 int undo)
407 {
408 int error = 0;
409
410 if (undo)
411 goto fail_trans;
412
413 error = gfs2_glock_nq_num(sdp,
414 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
415 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
416 mount_gh);
417 if (error) {
418 fs_err(sdp, "can't acquire mount glock: %d\n", error);
419 goto fail;
420 }
421
422 error = gfs2_glock_nq_num(sdp,
423 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
424 LM_ST_SHARED,
425 LM_FLAG_NOEXP | GL_EXACT,
426 &sdp->sd_live_gh);
427 if (error) {
428 fs_err(sdp, "can't acquire live glock: %d\n", error);
429 goto fail_mount;
430 }
431
432 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
433 CREATE, &sdp->sd_rename_gl);
434 if (error) {
435 fs_err(sdp, "can't create rename glock: %d\n", error);
436 goto fail_live;
437 }
438
439 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
440 CREATE, &sdp->sd_freeze_gl);
441 if (error) {
442 fs_err(sdp, "can't create transaction glock: %d\n", error);
443 goto fail_rename;
444 }
445
446 return 0;
447
448 fail_trans:
449 gfs2_glock_put(sdp->sd_freeze_gl);
450 fail_rename:
451 gfs2_glock_put(sdp->sd_rename_gl);
452 fail_live:
453 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
454 fail_mount:
455 gfs2_glock_dq_uninit(mount_gh);
456 fail:
457 return error;
458 }
459
460 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
461 u64 no_addr, const char *name)
462 {
463 struct gfs2_sbd *sdp = sb->s_fs_info;
464 struct dentry *dentry;
465 struct inode *inode;
466
467 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
468 GFS2_BLKST_FREE /* ignore */);
469 if (IS_ERR(inode)) {
470 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
471 return PTR_ERR(inode);
472 }
473 dentry = d_make_root(inode);
474 if (!dentry) {
475 fs_err(sdp, "can't alloc %s dentry\n", name);
476 return -ENOMEM;
477 }
478 *dptr = dentry;
479 return 0;
480 }
481
482 static int init_sb(struct gfs2_sbd *sdp, int silent)
483 {
484 struct super_block *sb = sdp->sd_vfs;
485 struct gfs2_holder sb_gh;
486 u64 no_addr;
487 int ret;
488
489 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
490 LM_ST_SHARED, 0, &sb_gh);
491 if (ret) {
492 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
493 return ret;
494 }
495
496 ret = gfs2_read_sb(sdp, silent);
497 if (ret) {
498 fs_err(sdp, "can't read superblock: %d\n", ret);
499 goto out;
500 }
501
502 /* Set up the buffer cache and SB for real */
503 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
504 ret = -EINVAL;
505 fs_err(sdp, "FS block size (%u) is too small for device "
506 "block size (%u)\n",
507 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
508 goto out;
509 }
510 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
511 ret = -EINVAL;
512 fs_err(sdp, "FS block size (%u) is too big for machine "
513 "page size (%u)\n",
514 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
515 goto out;
516 }
517 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
518
519 /* Get the root inode */
520 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
521 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
522 if (ret)
523 goto out;
524
525 /* Get the master inode */
526 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
527 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
528 if (ret) {
529 dput(sdp->sd_root_dir);
530 goto out;
531 }
532 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
533 out:
534 gfs2_glock_dq_uninit(&sb_gh);
535 return ret;
536 }
537
538 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
539 {
540 char *message = "FIRSTMOUNT=Done";
541 char *envp[] = { message, NULL };
542
543 fs_info(sdp, "first mount done, others may mount\n");
544
545 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
546 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
547
548 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
549 }
550
551 /**
552 * gfs2_jindex_hold - Grab a lock on the jindex
553 * @sdp: The GFS2 superblock
554 * @ji_gh: the holder for the jindex glock
555 *
556 * Returns: errno
557 */
558
559 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
560 {
561 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
562 struct qstr name;
563 char buf[20];
564 struct gfs2_jdesc *jd;
565 int error;
566
567 name.name = buf;
568
569 mutex_lock(&sdp->sd_jindex_mutex);
570
571 for (;;) {
572 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
573 if (error)
574 break;
575
576 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
577 name.hash = gfs2_disk_hash(name.name, name.len);
578
579 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
580 if (error == -ENOENT) {
581 error = 0;
582 break;
583 }
584
585 gfs2_glock_dq_uninit(ji_gh);
586
587 if (error)
588 break;
589
590 error = -ENOMEM;
591 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
592 if (!jd)
593 break;
594
595 INIT_LIST_HEAD(&jd->extent_list);
596 INIT_LIST_HEAD(&jd->jd_revoke_list);
597
598 INIT_WORK(&jd->jd_work, gfs2_recover_func);
599 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
600 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
601 if (!jd->jd_inode)
602 error = -ENOENT;
603 else
604 error = PTR_ERR(jd->jd_inode);
605 kfree(jd);
606 break;
607 }
608
609 spin_lock(&sdp->sd_jindex_spin);
610 jd->jd_jid = sdp->sd_journals++;
611 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
612 spin_unlock(&sdp->sd_jindex_spin);
613 }
614
615 mutex_unlock(&sdp->sd_jindex_mutex);
616
617 return error;
618 }
619
620 /**
621 * check_journal_clean - Make sure a journal is clean for a spectator mount
622 * @sdp: The GFS2 superblock
623 * @jd: The journal descriptor
624 *
625 * Returns: 0 if the journal is clean or locked, else an error
626 */
627 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
628 {
629 int error;
630 struct gfs2_holder j_gh;
631 struct gfs2_log_header_host head;
632 struct gfs2_inode *ip;
633
634 ip = GFS2_I(jd->jd_inode);
635 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
636 GL_EXACT | GL_NOCACHE, &j_gh);
637 if (error) {
638 fs_err(sdp, "Error locking journal for spectator mount.\n");
639 return -EPERM;
640 }
641 error = gfs2_jdesc_check(jd);
642 if (error) {
643 fs_err(sdp, "Error checking journal for spectator mount.\n");
644 goto out_unlock;
645 }
646 error = gfs2_find_jhead(jd, &head);
647 if (error) {
648 fs_err(sdp, "Error parsing journal for spectator mount.\n");
649 goto out_unlock;
650 }
651 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
652 error = -EPERM;
653 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
654 "must not be a spectator.\n", jd->jd_jid);
655 }
656
657 out_unlock:
658 gfs2_glock_dq_uninit(&j_gh);
659 return error;
660 }
661
662 static int init_journal(struct gfs2_sbd *sdp, int undo)
663 {
664 struct inode *master = d_inode(sdp->sd_master_dir);
665 struct gfs2_holder ji_gh;
666 struct gfs2_inode *ip;
667 int jindex = 1;
668 int error = 0;
669
670 if (undo) {
671 jindex = 0;
672 goto fail_jinode_gh;
673 }
674
675 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
676 if (IS_ERR(sdp->sd_jindex)) {
677 fs_err(sdp, "can't lookup journal index: %d\n", error);
678 return PTR_ERR(sdp->sd_jindex);
679 }
680
681 /* Load in the journal index special file */
682
683 error = gfs2_jindex_hold(sdp, &ji_gh);
684 if (error) {
685 fs_err(sdp, "can't read journal index: %d\n", error);
686 goto fail;
687 }
688
689 error = -EUSERS;
690 if (!gfs2_jindex_size(sdp)) {
691 fs_err(sdp, "no journals!\n");
692 goto fail_jindex;
693 }
694
695 if (sdp->sd_args.ar_spectator) {
696 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
697 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
698 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
699 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
700 } else {
701 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
702 fs_err(sdp, "can't mount journal #%u\n",
703 sdp->sd_lockstruct.ls_jid);
704 fs_err(sdp, "there are only %u journals (0 - %u)\n",
705 gfs2_jindex_size(sdp),
706 gfs2_jindex_size(sdp) - 1);
707 goto fail_jindex;
708 }
709 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
710
711 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
712 &gfs2_journal_glops,
713 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
714 &sdp->sd_journal_gh);
715 if (error) {
716 fs_err(sdp, "can't acquire journal glock: %d\n", error);
717 goto fail_jindex;
718 }
719
720 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
721 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
722 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
723 &sdp->sd_jinode_gh);
724 if (error) {
725 fs_err(sdp, "can't acquire journal inode glock: %d\n",
726 error);
727 goto fail_journal_gh;
728 }
729
730 error = gfs2_jdesc_check(sdp->sd_jdesc);
731 if (error) {
732 fs_err(sdp, "my journal (%u) is bad: %d\n",
733 sdp->sd_jdesc->jd_jid, error);
734 goto fail_jinode_gh;
735 }
736 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
737 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
738 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
739
740 /* Map the extents for this journal's blocks */
741 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
742 }
743 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
744
745 if (sdp->sd_lockstruct.ls_first) {
746 unsigned int x;
747 for (x = 0; x < sdp->sd_journals; x++) {
748 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
749
750 if (sdp->sd_args.ar_spectator) {
751 error = check_journal_clean(sdp, jd);
752 if (error)
753 goto fail_jinode_gh;
754 continue;
755 }
756 error = gfs2_recover_journal(jd, true);
757 if (error) {
758 fs_err(sdp, "error recovering journal %u: %d\n",
759 x, error);
760 goto fail_jinode_gh;
761 }
762 }
763
764 gfs2_others_may_mount(sdp);
765 } else if (!sdp->sd_args.ar_spectator) {
766 error = gfs2_recover_journal(sdp->sd_jdesc, true);
767 if (error) {
768 fs_err(sdp, "error recovering my journal: %d\n", error);
769 goto fail_jinode_gh;
770 }
771 }
772
773 sdp->sd_log_idle = 1;
774 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
775 gfs2_glock_dq_uninit(&ji_gh);
776 jindex = 0;
777 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
778 return 0;
779
780 fail_jinode_gh:
781 if (!sdp->sd_args.ar_spectator)
782 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
783 fail_journal_gh:
784 if (!sdp->sd_args.ar_spectator)
785 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
786 fail_jindex:
787 gfs2_jindex_free(sdp);
788 if (jindex)
789 gfs2_glock_dq_uninit(&ji_gh);
790 fail:
791 iput(sdp->sd_jindex);
792 return error;
793 }
794
795 static struct lock_class_key gfs2_quota_imutex_key;
796
797 static int init_inodes(struct gfs2_sbd *sdp, int undo)
798 {
799 int error = 0;
800 struct inode *master = d_inode(sdp->sd_master_dir);
801
802 if (undo)
803 goto fail_qinode;
804
805 error = init_journal(sdp, undo);
806 complete_all(&sdp->sd_journal_ready);
807 if (error)
808 goto fail;
809
810 /* Read in the master statfs inode */
811 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
812 if (IS_ERR(sdp->sd_statfs_inode)) {
813 error = PTR_ERR(sdp->sd_statfs_inode);
814 fs_err(sdp, "can't read in statfs inode: %d\n", error);
815 goto fail_journal;
816 }
817
818 /* Read in the resource index inode */
819 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
820 if (IS_ERR(sdp->sd_rindex)) {
821 error = PTR_ERR(sdp->sd_rindex);
822 fs_err(sdp, "can't get resource index inode: %d\n", error);
823 goto fail_statfs;
824 }
825 sdp->sd_rindex_uptodate = 0;
826
827 /* Read in the quota inode */
828 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
829 if (IS_ERR(sdp->sd_quota_inode)) {
830 error = PTR_ERR(sdp->sd_quota_inode);
831 fs_err(sdp, "can't get quota file inode: %d\n", error);
832 goto fail_rindex;
833 }
834 /*
835 * i_mutex on quota files is special. Since this inode is hidden system
836 * file, we are safe to define locking ourselves.
837 */
838 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
839 &gfs2_quota_imutex_key);
840
841 error = gfs2_rindex_update(sdp);
842 if (error)
843 goto fail_qinode;
844
845 return 0;
846
847 fail_qinode:
848 iput(sdp->sd_quota_inode);
849 fail_rindex:
850 gfs2_clear_rgrpd(sdp);
851 iput(sdp->sd_rindex);
852 fail_statfs:
853 iput(sdp->sd_statfs_inode);
854 fail_journal:
855 init_journal(sdp, UNDO);
856 fail:
857 return error;
858 }
859
860 static int init_per_node(struct gfs2_sbd *sdp, int undo)
861 {
862 struct inode *pn = NULL;
863 char buf[30];
864 int error = 0;
865 struct gfs2_inode *ip;
866 struct inode *master = d_inode(sdp->sd_master_dir);
867
868 if (sdp->sd_args.ar_spectator)
869 return 0;
870
871 if (undo)
872 goto fail_qc_gh;
873
874 pn = gfs2_lookup_simple(master, "per_node");
875 if (IS_ERR(pn)) {
876 error = PTR_ERR(pn);
877 fs_err(sdp, "can't find per_node directory: %d\n", error);
878 return error;
879 }
880
881 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
882 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
883 if (IS_ERR(sdp->sd_sc_inode)) {
884 error = PTR_ERR(sdp->sd_sc_inode);
885 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
886 goto fail;
887 }
888
889 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
890 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
891 if (IS_ERR(sdp->sd_qc_inode)) {
892 error = PTR_ERR(sdp->sd_qc_inode);
893 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
894 goto fail_ut_i;
895 }
896
897 iput(pn);
898 pn = NULL;
899
900 ip = GFS2_I(sdp->sd_sc_inode);
901 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
902 &sdp->sd_sc_gh);
903 if (error) {
904 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
905 goto fail_qc_i;
906 }
907
908 ip = GFS2_I(sdp->sd_qc_inode);
909 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
910 &sdp->sd_qc_gh);
911 if (error) {
912 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
913 goto fail_ut_gh;
914 }
915
916 return 0;
917
918 fail_qc_gh:
919 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
920 fail_ut_gh:
921 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
922 fail_qc_i:
923 iput(sdp->sd_qc_inode);
924 fail_ut_i:
925 iput(sdp->sd_sc_inode);
926 fail:
927 iput(pn);
928 return error;
929 }
930
931 static const match_table_t nolock_tokens = {
932 { Opt_jid, "jid=%d", },
933 { Opt_err, NULL },
934 };
935
936 static const struct lm_lockops nolock_ops = {
937 .lm_proto_name = "lock_nolock",
938 .lm_put_lock = gfs2_glock_free,
939 .lm_tokens = &nolock_tokens,
940 };
941
942 /**
943 * gfs2_lm_mount - mount a locking protocol
944 * @sdp: the filesystem
945 * @args: mount arguments
946 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
947 *
948 * Returns: errno
949 */
950
951 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
952 {
953 const struct lm_lockops *lm;
954 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
955 struct gfs2_args *args = &sdp->sd_args;
956 const char *proto = sdp->sd_proto_name;
957 const char *table = sdp->sd_table_name;
958 char *o, *options;
959 int ret;
960
961 if (!strcmp("lock_nolock", proto)) {
962 lm = &nolock_ops;
963 sdp->sd_args.ar_localflocks = 1;
964 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
965 } else if (!strcmp("lock_dlm", proto)) {
966 lm = &gfs2_dlm_ops;
967 #endif
968 } else {
969 pr_info("can't find protocol %s\n", proto);
970 return -ENOENT;
971 }
972
973 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
974
975 ls->ls_ops = lm;
976 ls->ls_first = 1;
977
978 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
979 substring_t tmp[MAX_OPT_ARGS];
980 int token, option;
981
982 if (!o || !*o)
983 continue;
984
985 token = match_token(o, *lm->lm_tokens, tmp);
986 switch (token) {
987 case Opt_jid:
988 ret = match_int(&tmp[0], &option);
989 if (ret || option < 0)
990 goto hostdata_error;
991 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
992 ls->ls_jid = option;
993 break;
994 case Opt_id:
995 case Opt_nodir:
996 /* Obsolete, but left for backward compat purposes */
997 break;
998 case Opt_first:
999 ret = match_int(&tmp[0], &option);
1000 if (ret || (option != 0 && option != 1))
1001 goto hostdata_error;
1002 ls->ls_first = option;
1003 break;
1004 case Opt_err:
1005 default:
1006 hostdata_error:
1007 fs_info(sdp, "unknown hostdata (%s)\n", o);
1008 return -EINVAL;
1009 }
1010 }
1011
1012 if (lm->lm_mount == NULL) {
1013 fs_info(sdp, "Now mounting FS...\n");
1014 complete_all(&sdp->sd_locking_init);
1015 return 0;
1016 }
1017 ret = lm->lm_mount(sdp, table);
1018 if (ret == 0)
1019 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1020 complete_all(&sdp->sd_locking_init);
1021 return ret;
1022 }
1023
1024 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1025 {
1026 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1027 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1028 lm->lm_unmount)
1029 lm->lm_unmount(sdp);
1030 }
1031
1032 static int wait_on_journal(struct gfs2_sbd *sdp)
1033 {
1034 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1035 return 0;
1036
1037 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1038 ? -EINTR : 0;
1039 }
1040
1041 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1042 {
1043 struct super_block *sb = sdp->sd_vfs;
1044 char ro[20];
1045 char spectator[20];
1046 char *envp[] = { ro, spectator, NULL };
1047 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1048 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1049 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1050 }
1051
1052 /**
1053 * fill_super - Read in superblock
1054 * @sb: The VFS superblock
1055 * @data: Mount options
1056 * @silent: Don't complain if it's not a GFS2 filesystem
1057 *
1058 * Returns: errno
1059 */
1060
1061 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1062 {
1063 struct gfs2_sbd *sdp;
1064 struct gfs2_holder mount_gh;
1065 int error;
1066
1067 sdp = init_sbd(sb);
1068 if (!sdp) {
1069 pr_warn("can't alloc struct gfs2_sbd\n");
1070 return -ENOMEM;
1071 }
1072 sdp->sd_args = *args;
1073
1074 if (sdp->sd_args.ar_spectator) {
1075 sb->s_flags |= MS_RDONLY;
1076 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1077 }
1078 if (sdp->sd_args.ar_posix_acl)
1079 sb->s_flags |= MS_POSIXACL;
1080 if (sdp->sd_args.ar_nobarrier)
1081 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1082
1083 sb->s_flags |= MS_NOSEC;
1084 sb->s_magic = GFS2_MAGIC;
1085 sb->s_op = &gfs2_super_ops;
1086 sb->s_d_op = &gfs2_dops;
1087 sb->s_export_op = &gfs2_export_ops;
1088 sb->s_xattr = gfs2_xattr_handlers;
1089 sb->s_qcop = &gfs2_quotactl_ops;
1090 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1091 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1092 sb->s_time_gran = 1;
1093 sb->s_maxbytes = MAX_LFS_FILESIZE;
1094
1095 /* Set up the buffer cache and fill in some fake block size values
1096 to allow us to read-in the on-disk superblock. */
1097 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1098 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1099 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1100 GFS2_BASIC_BLOCK_SHIFT;
1101 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1102
1103 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1104 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1105 if (sdp->sd_args.ar_statfs_quantum) {
1106 sdp->sd_tune.gt_statfs_slow = 0;
1107 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1108 } else {
1109 sdp->sd_tune.gt_statfs_slow = 1;
1110 sdp->sd_tune.gt_statfs_quantum = 30;
1111 }
1112
1113 error = init_names(sdp, silent);
1114 if (error) {
1115 /* In this case, we haven't initialized sysfs, so we have to
1116 manually free the sdp. */
1117 free_percpu(sdp->sd_lkstats);
1118 kfree(sdp);
1119 sb->s_fs_info = NULL;
1120 return error;
1121 }
1122
1123 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1124
1125 error = gfs2_sys_fs_add(sdp);
1126 /*
1127 * If we hit an error here, gfs2_sys_fs_add will have called function
1128 * kobject_put which causes the sysfs usage count to go to zero, which
1129 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1130 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1131 * kobject_put to free sdp.
1132 */
1133 if (error)
1134 return error;
1135
1136 gfs2_create_debugfs_file(sdp);
1137
1138 error = gfs2_lm_mount(sdp, silent);
1139 if (error)
1140 goto fail_debug;
1141
1142 error = init_locking(sdp, &mount_gh, DO);
1143 if (error)
1144 goto fail_lm;
1145
1146 error = init_sb(sdp, silent);
1147 if (error)
1148 goto fail_locking;
1149
1150 error = wait_on_journal(sdp);
1151 if (error)
1152 goto fail_sb;
1153
1154 /*
1155 * If user space has failed to join the cluster or some similar
1156 * failure has occurred, then the journal id will contain a
1157 * negative (error) number. This will then be returned to the
1158 * caller (of the mount syscall). We do this even for spectator
1159 * mounts (which just write a jid of 0 to indicate "ok" even though
1160 * the jid is unused in the spectator case)
1161 */
1162 if (sdp->sd_lockstruct.ls_jid < 0) {
1163 error = sdp->sd_lockstruct.ls_jid;
1164 sdp->sd_lockstruct.ls_jid = 0;
1165 goto fail_sb;
1166 }
1167
1168 if (sdp->sd_args.ar_spectator)
1169 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1170 sdp->sd_table_name);
1171 else
1172 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1173 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1174
1175 error = init_inodes(sdp, DO);
1176 if (error)
1177 goto fail_sb;
1178
1179 error = init_per_node(sdp, DO);
1180 if (error)
1181 goto fail_inodes;
1182
1183 error = gfs2_statfs_init(sdp);
1184 if (error) {
1185 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1186 goto fail_per_node;
1187 }
1188
1189 if (!(sb->s_flags & MS_RDONLY)) {
1190 error = gfs2_make_fs_rw(sdp);
1191 if (error) {
1192 fs_err(sdp, "can't make FS RW: %d\n", error);
1193 goto fail_per_node;
1194 }
1195 }
1196
1197 gfs2_glock_dq_uninit(&mount_gh);
1198 gfs2_online_uevent(sdp);
1199 return 0;
1200
1201 fail_per_node:
1202 init_per_node(sdp, UNDO);
1203 fail_inodes:
1204 init_inodes(sdp, UNDO);
1205 fail_sb:
1206 if (sdp->sd_root_dir)
1207 dput(sdp->sd_root_dir);
1208 if (sdp->sd_master_dir)
1209 dput(sdp->sd_master_dir);
1210 if (sb->s_root)
1211 dput(sb->s_root);
1212 sb->s_root = NULL;
1213 fail_locking:
1214 init_locking(sdp, &mount_gh, UNDO);
1215 fail_lm:
1216 complete_all(&sdp->sd_journal_ready);
1217 gfs2_gl_hash_clear(sdp);
1218 gfs2_lm_unmount(sdp);
1219 fail_debug:
1220 gfs2_delete_debugfs_file(sdp);
1221 free_percpu(sdp->sd_lkstats);
1222 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1223 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1224 gfs2_sys_fs_del(sdp);
1225 sb->s_fs_info = NULL;
1226 return error;
1227 }
1228
1229 static int set_gfs2_super(struct super_block *s, void *data)
1230 {
1231 s->s_bdev = data;
1232 s->s_dev = s->s_bdev->bd_dev;
1233
1234 /*
1235 * We set the bdi here to the queue backing, file systems can
1236 * overwrite this in ->fill_super()
1237 */
1238 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1239 return 0;
1240 }
1241
1242 static int test_gfs2_super(struct super_block *s, void *ptr)
1243 {
1244 struct block_device *bdev = ptr;
1245 return (bdev == s->s_bdev);
1246 }
1247
1248 /**
1249 * gfs2_mount - Get the GFS2 superblock
1250 * @fs_type: The GFS2 filesystem type
1251 * @flags: Mount flags
1252 * @dev_name: The name of the device
1253 * @data: The mount arguments
1254 *
1255 * Q. Why not use get_sb_bdev() ?
1256 * A. We need to select one of two root directories to mount, independent
1257 * of whether this is the initial, or subsequent, mount of this sb
1258 *
1259 * Returns: 0 or -ve on error
1260 */
1261
1262 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1263 const char *dev_name, void *data)
1264 {
1265 struct block_device *bdev;
1266 struct super_block *s;
1267 fmode_t mode = FMODE_READ | FMODE_EXCL;
1268 int error;
1269 struct gfs2_args args;
1270 struct gfs2_sbd *sdp;
1271
1272 if (!(flags & MS_RDONLY))
1273 mode |= FMODE_WRITE;
1274
1275 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1276 if (IS_ERR(bdev))
1277 return ERR_CAST(bdev);
1278
1279 /*
1280 * once the super is inserted into the list by sget, s_umount
1281 * will protect the lockfs code from trying to start a snapshot
1282 * while we are mounting
1283 */
1284 mutex_lock(&bdev->bd_fsfreeze_mutex);
1285 if (bdev->bd_fsfreeze_count > 0) {
1286 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1287 error = -EBUSY;
1288 goto error_bdev;
1289 }
1290 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1292 error = PTR_ERR(s);
1293 if (IS_ERR(s))
1294 goto error_bdev;
1295
1296 if (s->s_root) {
1297 /*
1298 * s_umount nests inside bd_mutex during
1299 * __invalidate_device(). blkdev_put() acquires
1300 * bd_mutex and can't be called under s_umount. Drop
1301 * s_umount temporarily. This is safe as we're
1302 * holding an active reference.
1303 */
1304 up_write(&s->s_umount);
1305 blkdev_put(bdev, mode);
1306 down_write(&s->s_umount);
1307 } else {
1308 /* s_mode must be set before deactivate_locked_super calls */
1309 s->s_mode = mode;
1310 }
1311
1312 memset(&args, 0, sizeof(args));
1313 args.ar_quota = GFS2_QUOTA_DEFAULT;
1314 args.ar_data = GFS2_DATA_DEFAULT;
1315 args.ar_commit = 30;
1316 args.ar_statfs_quantum = 30;
1317 args.ar_quota_quantum = 60;
1318 args.ar_errors = GFS2_ERRORS_DEFAULT;
1319
1320 error = gfs2_mount_args(&args, data);
1321 if (error) {
1322 pr_warn("can't parse mount arguments\n");
1323 goto error_super;
1324 }
1325
1326 if (s->s_root) {
1327 error = -EBUSY;
1328 if ((flags ^ s->s_flags) & MS_RDONLY)
1329 goto error_super;
1330 } else {
1331 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1332 sb_set_blocksize(s, block_size(bdev));
1333 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1334 if (error)
1335 goto error_super;
1336 s->s_flags |= MS_ACTIVE;
1337 bdev->bd_super = s;
1338 }
1339
1340 sdp = s->s_fs_info;
1341 if (args.ar_meta)
1342 return dget(sdp->sd_master_dir);
1343 else
1344 return dget(sdp->sd_root_dir);
1345
1346 error_super:
1347 deactivate_locked_super(s);
1348 return ERR_PTR(error);
1349 error_bdev:
1350 blkdev_put(bdev, mode);
1351 return ERR_PTR(error);
1352 }
1353
1354 static int set_meta_super(struct super_block *s, void *ptr)
1355 {
1356 return -EINVAL;
1357 }
1358
1359 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1360 int flags, const char *dev_name, void *data)
1361 {
1362 struct super_block *s;
1363 struct gfs2_sbd *sdp;
1364 struct path path;
1365 int error;
1366
1367 if (!dev_name || !*dev_name)
1368 return ERR_PTR(-EINVAL);
1369
1370 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1371 if (error) {
1372 pr_warn("path_lookup on %s returned error %d\n",
1373 dev_name, error);
1374 return ERR_PTR(error);
1375 }
1376 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1377 path.dentry->d_sb->s_bdev);
1378 path_put(&path);
1379 if (IS_ERR(s)) {
1380 pr_warn("gfs2 mount does not exist\n");
1381 return ERR_CAST(s);
1382 }
1383 if ((flags ^ s->s_flags) & MS_RDONLY) {
1384 deactivate_locked_super(s);
1385 return ERR_PTR(-EBUSY);
1386 }
1387 sdp = s->s_fs_info;
1388 return dget(sdp->sd_master_dir);
1389 }
1390
1391 static void gfs2_kill_sb(struct super_block *sb)
1392 {
1393 struct gfs2_sbd *sdp = sb->s_fs_info;
1394
1395 if (sdp == NULL) {
1396 kill_block_super(sb);
1397 return;
1398 }
1399
1400 gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1401 dput(sdp->sd_root_dir);
1402 dput(sdp->sd_master_dir);
1403 sdp->sd_root_dir = NULL;
1404 sdp->sd_master_dir = NULL;
1405 shrink_dcache_sb(sb);
1406 gfs2_delete_debugfs_file(sdp);
1407 free_percpu(sdp->sd_lkstats);
1408 kill_block_super(sb);
1409 }
1410
1411 struct file_system_type gfs2_fs_type = {
1412 .name = "gfs2",
1413 .fs_flags = FS_REQUIRES_DEV,
1414 .mount = gfs2_mount,
1415 .kill_sb = gfs2_kill_sb,
1416 .owner = THIS_MODULE,
1417 };
1418 MODULE_ALIAS_FS("gfs2");
1419
1420 struct file_system_type gfs2meta_fs_type = {
1421 .name = "gfs2meta",
1422 .fs_flags = FS_REQUIRES_DEV,
1423 .mount = gfs2_mount_meta,
1424 .owner = THIS_MODULE,
1425 };
1426 MODULE_ALIAS_FS("gfs2meta");
1427
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.