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 #include <linux/backing-dev.h>
27
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "glock.h"
32 #include "glops.h"
33 #include "inode.h"
34 #include "recovery.h"
35 #include "rgrp.h"
36 #include "super.h"
37 #include "sys.h"
38 #include "util.h"
39 #include "log.h"
40 #include "quota.h"
41 #include "dir.h"
42 #include "meta_io.h"
43 #include "trace_gfs2.h"
44
45 #define DO 0
46 #define UNDO 1
47
48 /**
49 * gfs2_tune_init - Fill a gfs2_tune structure with default values
50 * @gt: tune
51 *
52 */
53
gfs2_tune_init(struct gfs2_tune * gt)54 static void gfs2_tune_init(struct gfs2_tune *gt)
55 {
56 spin_lock_init(>->gt_spin);
57
58 gt->gt_quota_warn_period = 10;
59 gt->gt_quota_scale_num = 1;
60 gt->gt_quota_scale_den = 1;
61 gt->gt_new_files_jdata = 0;
62 gt->gt_max_readahead = BIT(18);
63 gt->gt_complain_secs = 10;
64 }
65
init_sbd(struct super_block * sb)66 static struct gfs2_sbd *init_sbd(struct super_block *sb)
67 {
68 struct gfs2_sbd *sdp;
69 struct address_space *mapping;
70
71 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
72 if (!sdp)
73 return NULL;
74
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 sb->s_fs_info = sdp;
82
83 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
84 gfs2_tune_init(&sdp->sd_tune);
85
86 init_waitqueue_head(&sdp->sd_glock_wait);
87 atomic_set(&sdp->sd_glock_disposal, 0);
88 init_completion(&sdp->sd_locking_init);
89 init_completion(&sdp->sd_wdack);
90 spin_lock_init(&sdp->sd_statfs_spin);
91
92 spin_lock_init(&sdp->sd_rindex_spin);
93 sdp->sd_rindex_tree.rb_node = NULL;
94
95 INIT_LIST_HEAD(&sdp->sd_jindex_list);
96 spin_lock_init(&sdp->sd_jindex_spin);
97 mutex_init(&sdp->sd_jindex_mutex);
98 init_completion(&sdp->sd_journal_ready);
99
100 INIT_LIST_HEAD(&sdp->sd_quota_list);
101 mutex_init(&sdp->sd_quota_mutex);
102 mutex_init(&sdp->sd_quota_sync_mutex);
103 init_waitqueue_head(&sdp->sd_quota_wait);
104 INIT_LIST_HEAD(&sdp->sd_trunc_list);
105 spin_lock_init(&sdp->sd_trunc_lock);
106 spin_lock_init(&sdp->sd_bitmap_lock);
107
108 mapping = &sdp->sd_aspace;
109
110 address_space_init_once(mapping);
111 mapping->a_ops = &gfs2_rgrp_aops;
112 mapping->host = sb->s_bdev->bd_inode;
113 mapping->flags = 0;
114 mapping_set_gfp_mask(mapping, GFP_NOFS);
115 mapping->private_data = NULL;
116 mapping->writeback_index = 0;
117
118 spin_lock_init(&sdp->sd_log_lock);
119 atomic_set(&sdp->sd_log_pinned, 0);
120 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
121 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
122 spin_lock_init(&sdp->sd_ordered_lock);
123
124 init_waitqueue_head(&sdp->sd_log_waitq);
125 init_waitqueue_head(&sdp->sd_logd_waitq);
126 spin_lock_init(&sdp->sd_ail_lock);
127 INIT_LIST_HEAD(&sdp->sd_ail1_list);
128 INIT_LIST_HEAD(&sdp->sd_ail2_list);
129
130 init_rwsem(&sdp->sd_log_flush_lock);
131 atomic_set(&sdp->sd_log_in_flight, 0);
132 atomic_set(&sdp->sd_reserving_log, 0);
133 init_waitqueue_head(&sdp->sd_reserving_log_wait);
134 init_waitqueue_head(&sdp->sd_log_flush_wait);
135 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
136 mutex_init(&sdp->sd_freeze_mutex);
137
138 return sdp;
139 }
140
141
142 /**
143 * gfs2_check_sb - Check superblock
144 * @sdp: the filesystem
145 * @sb: The superblock
146 * @silent: Don't print a message if the check fails
147 *
148 * Checks the version code of the FS is one that we understand how to
149 * read and that the sizes of the various on-disk structures have not
150 * changed.
151 */
152
gfs2_check_sb(struct gfs2_sbd * sdp,int silent)153 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
154 {
155 struct gfs2_sb_host *sb = &sdp->sd_sb;
156
157 if (sb->sb_magic != GFS2_MAGIC ||
158 sb->sb_type != GFS2_METATYPE_SB) {
159 if (!silent)
160 pr_warn("not a GFS2 filesystem\n");
161 return -EINVAL;
162 }
163
164 if (sb->sb_fs_format != GFS2_FORMAT_FS ||
165 sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
166 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
167 return -EINVAL;
168 }
169
170 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
171 (sb->sb_bsize & (sb->sb_bsize - 1))) {
172 pr_warn("Invalid superblock size\n");
173 return -EINVAL;
174 }
175 if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
176 pr_warn("Invalid block size shift\n");
177 return -EINVAL;
178 }
179 return 0;
180 }
181
end_bio_io_page(struct bio * bio)182 static void end_bio_io_page(struct bio *bio)
183 {
184 struct page *page = bio->bi_private;
185
186 if (!bio->bi_status)
187 SetPageUptodate(page);
188 else
189 pr_warn("error %d reading superblock\n", bio->bi_status);
190 unlock_page(page);
191 }
192
gfs2_sb_in(struct gfs2_sbd * sdp,const void * buf)193 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
194 {
195 struct gfs2_sb_host *sb = &sdp->sd_sb;
196 struct super_block *s = sdp->sd_vfs;
197 const struct gfs2_sb *str = buf;
198
199 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
200 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
201 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
202 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
203 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
204 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
205 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
206 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
207 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
208 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
209 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
210
211 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
212 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
213 memcpy(&s->s_uuid, str->sb_uuid, 16);
214 }
215
216 /**
217 * gfs2_read_super - Read the gfs2 super block from disk
218 * @sdp: The GFS2 super block
219 * @sector: The location of the super block
220 * @error: The error code to return
221 *
222 * This uses the bio functions to read the super block from disk
223 * because we want to be 100% sure that we never read cached data.
224 * A super block is read twice only during each GFS2 mount and is
225 * never written to by the filesystem. The first time its read no
226 * locks are held, and the only details which are looked at are those
227 * relating to the locking protocol. Once locking is up and working,
228 * the sb is read again under the lock to establish the location of
229 * the master directory (contains pointers to journals etc) and the
230 * root directory.
231 *
232 * Returns: 0 on success or error
233 */
234
gfs2_read_super(struct gfs2_sbd * sdp,sector_t sector,int silent)235 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
236 {
237 struct super_block *sb = sdp->sd_vfs;
238 struct gfs2_sb *p;
239 struct page *page;
240 struct bio *bio;
241
242 page = alloc_page(GFP_NOFS);
243 if (unlikely(!page))
244 return -ENOMEM;
245
246 ClearPageUptodate(page);
247 ClearPageDirty(page);
248 lock_page(page);
249
250 bio = bio_alloc(GFP_NOFS, 1);
251 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
252 bio_set_dev(bio, sb->s_bdev);
253 bio_add_page(bio, page, PAGE_SIZE, 0);
254
255 bio->bi_end_io = end_bio_io_page;
256 bio->bi_private = page;
257 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
258 submit_bio(bio);
259 wait_on_page_locked(page);
260 bio_put(bio);
261 if (!PageUptodate(page)) {
262 __free_page(page);
263 return -EIO;
264 }
265 p = kmap(page);
266 gfs2_sb_in(sdp, p);
267 kunmap(page);
268 __free_page(page);
269 return gfs2_check_sb(sdp, silent);
270 }
271
272 /**
273 * gfs2_read_sb - Read super block
274 * @sdp: The GFS2 superblock
275 * @silent: Don't print message if mount fails
276 *
277 */
278
gfs2_read_sb(struct gfs2_sbd * sdp,int silent)279 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
280 {
281 u32 hash_blocks, ind_blocks, leaf_blocks;
282 u32 tmp_blocks;
283 unsigned int x;
284 int error;
285
286 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
287 if (error) {
288 if (!silent)
289 fs_err(sdp, "can't read superblock\n");
290 return error;
291 }
292
293 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
294 GFS2_BASIC_BLOCK_SHIFT;
295 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
296 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
297 sizeof(struct gfs2_dinode)) / sizeof(u64);
298 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
299 sizeof(struct gfs2_meta_header)) / sizeof(u64);
300 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
301 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
302 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
303 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
304 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
305 sizeof(struct gfs2_meta_header)) /
306 sizeof(struct gfs2_quota_change);
307 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
308 sizeof(struct gfs2_meta_header))
309 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
310
311 /* Compute maximum reservation required to add a entry to a directory */
312
313 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
314 sdp->sd_jbsize);
315
316 ind_blocks = 0;
317 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
318 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
319 ind_blocks += tmp_blocks;
320 }
321
322 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
323
324 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
325
326 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
327 sizeof(struct gfs2_dinode);
328 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
329 for (x = 2;; x++) {
330 u64 space, d;
331 u32 m;
332
333 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
334 d = space;
335 m = do_div(d, sdp->sd_inptrs);
336
337 if (d != sdp->sd_heightsize[x - 1] || m)
338 break;
339 sdp->sd_heightsize[x] = space;
340 }
341 sdp->sd_max_height = x;
342 sdp->sd_heightsize[x] = ~0;
343 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
344
345 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
346 sizeof(struct gfs2_leaf)) /
347 GFS2_MIN_DIRENT_SIZE;
348 return 0;
349 }
350
init_names(struct gfs2_sbd * sdp,int silent)351 static int init_names(struct gfs2_sbd *sdp, int silent)
352 {
353 char *proto, *table;
354 int error = 0;
355
356 proto = sdp->sd_args.ar_lockproto;
357 table = sdp->sd_args.ar_locktable;
358
359 /* Try to autodetect */
360
361 if (!proto[0] || !table[0]) {
362 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
363 if (error)
364 return error;
365
366 if (!proto[0])
367 proto = sdp->sd_sb.sb_lockproto;
368 if (!table[0])
369 table = sdp->sd_sb.sb_locktable;
370 }
371
372 if (!table[0])
373 table = sdp->sd_vfs->s_id;
374
375 BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
376
377 strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
378 strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
379
380 table = sdp->sd_table_name;
381 while ((table = strchr(table, '/')))
382 *table = '_';
383
384 return error;
385 }
386
init_locking(struct gfs2_sbd * sdp,struct gfs2_holder * mount_gh,int undo)387 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
388 int undo)
389 {
390 int error = 0;
391
392 if (undo)
393 goto fail_trans;
394
395 error = gfs2_glock_nq_num(sdp,
396 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
397 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
398 mount_gh);
399 if (error) {
400 fs_err(sdp, "can't acquire mount glock: %d\n", error);
401 goto fail;
402 }
403
404 error = gfs2_glock_nq_num(sdp,
405 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
406 LM_ST_SHARED,
407 LM_FLAG_NOEXP | GL_EXACT,
408 &sdp->sd_live_gh);
409 if (error) {
410 fs_err(sdp, "can't acquire live glock: %d\n", error);
411 goto fail_mount;
412 }
413
414 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
415 CREATE, &sdp->sd_rename_gl);
416 if (error) {
417 fs_err(sdp, "can't create rename glock: %d\n", error);
418 goto fail_live;
419 }
420
421 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
422 CREATE, &sdp->sd_freeze_gl);
423 if (error) {
424 fs_err(sdp, "can't create transaction glock: %d\n", error);
425 goto fail_rename;
426 }
427
428 return 0;
429
430 fail_trans:
431 gfs2_glock_put(sdp->sd_freeze_gl);
432 fail_rename:
433 gfs2_glock_put(sdp->sd_rename_gl);
434 fail_live:
435 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
436 fail_mount:
437 gfs2_glock_dq_uninit(mount_gh);
438 fail:
439 return error;
440 }
441
gfs2_lookup_root(struct super_block * sb,struct dentry ** dptr,u64 no_addr,const char * name)442 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
443 u64 no_addr, const char *name)
444 {
445 struct gfs2_sbd *sdp = sb->s_fs_info;
446 struct dentry *dentry;
447 struct inode *inode;
448
449 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
450 GFS2_BLKST_FREE /* ignore */);
451 if (IS_ERR(inode)) {
452 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
453 return PTR_ERR(inode);
454 }
455 dentry = d_make_root(inode);
456 if (!dentry) {
457 fs_err(sdp, "can't alloc %s dentry\n", name);
458 return -ENOMEM;
459 }
460 *dptr = dentry;
461 return 0;
462 }
463
init_sb(struct gfs2_sbd * sdp,int silent)464 static int init_sb(struct gfs2_sbd *sdp, int silent)
465 {
466 struct super_block *sb = sdp->sd_vfs;
467 struct gfs2_holder sb_gh;
468 u64 no_addr;
469 int ret;
470
471 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
472 LM_ST_SHARED, 0, &sb_gh);
473 if (ret) {
474 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
475 return ret;
476 }
477
478 ret = gfs2_read_sb(sdp, silent);
479 if (ret) {
480 fs_err(sdp, "can't read superblock: %d\n", ret);
481 goto out;
482 }
483
484 /* Set up the buffer cache and SB for real */
485 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
486 ret = -EINVAL;
487 fs_err(sdp, "FS block size (%u) is too small for device "
488 "block size (%u)\n",
489 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
490 goto out;
491 }
492 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
493 ret = -EINVAL;
494 fs_err(sdp, "FS block size (%u) is too big for machine "
495 "page size (%u)\n",
496 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
497 goto out;
498 }
499 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
500
501 /* Get the root inode */
502 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
503 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
504 if (ret)
505 goto out;
506
507 /* Get the master inode */
508 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
509 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
510 if (ret) {
511 dput(sdp->sd_root_dir);
512 goto out;
513 }
514 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
515 out:
516 gfs2_glock_dq_uninit(&sb_gh);
517 return ret;
518 }
519
gfs2_others_may_mount(struct gfs2_sbd * sdp)520 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
521 {
522 char *message = "FIRSTMOUNT=Done";
523 char *envp[] = { message, NULL };
524
525 fs_info(sdp, "first mount done, others may mount\n");
526
527 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
528 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
529
530 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
531 }
532
533 /**
534 * gfs2_jindex_hold - Grab a lock on the jindex
535 * @sdp: The GFS2 superblock
536 * @ji_gh: the holder for the jindex glock
537 *
538 * Returns: errno
539 */
540
gfs2_jindex_hold(struct gfs2_sbd * sdp,struct gfs2_holder * ji_gh)541 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
542 {
543 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
544 struct qstr name;
545 char buf[20];
546 struct gfs2_jdesc *jd;
547 int error;
548
549 name.name = buf;
550
551 mutex_lock(&sdp->sd_jindex_mutex);
552
553 for (;;) {
554 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
555 if (error)
556 break;
557
558 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
559 name.hash = gfs2_disk_hash(name.name, name.len);
560
561 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
562 if (error == -ENOENT) {
563 error = 0;
564 break;
565 }
566
567 gfs2_glock_dq_uninit(ji_gh);
568
569 if (error)
570 break;
571
572 error = -ENOMEM;
573 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
574 if (!jd)
575 break;
576
577 INIT_LIST_HEAD(&jd->extent_list);
578 INIT_LIST_HEAD(&jd->jd_revoke_list);
579
580 INIT_WORK(&jd->jd_work, gfs2_recover_func);
581 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
582 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
583 if (!jd->jd_inode)
584 error = -ENOENT;
585 else
586 error = PTR_ERR(jd->jd_inode);
587 kfree(jd);
588 break;
589 }
590
591 spin_lock(&sdp->sd_jindex_spin);
592 jd->jd_jid = sdp->sd_journals++;
593 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
594 spin_unlock(&sdp->sd_jindex_spin);
595 }
596
597 mutex_unlock(&sdp->sd_jindex_mutex);
598
599 return error;
600 }
601
602 /**
603 * check_journal_clean - Make sure a journal is clean for a spectator mount
604 * @sdp: The GFS2 superblock
605 * @jd: The journal descriptor
606 *
607 * Returns: 0 if the journal is clean or locked, else an error
608 */
check_journal_clean(struct gfs2_sbd * sdp,struct gfs2_jdesc * jd)609 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
610 {
611 int error;
612 struct gfs2_holder j_gh;
613 struct gfs2_log_header_host head;
614 struct gfs2_inode *ip;
615
616 ip = GFS2_I(jd->jd_inode);
617 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
618 GL_EXACT | GL_NOCACHE, &j_gh);
619 if (error) {
620 fs_err(sdp, "Error locking journal for spectator mount.\n");
621 return -EPERM;
622 }
623 error = gfs2_jdesc_check(jd);
624 if (error) {
625 fs_err(sdp, "Error checking journal for spectator mount.\n");
626 goto out_unlock;
627 }
628 error = gfs2_find_jhead(jd, &head);
629 if (error) {
630 fs_err(sdp, "Error parsing journal for spectator mount.\n");
631 goto out_unlock;
632 }
633 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
634 error = -EPERM;
635 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
636 "must not be a spectator.\n", jd->jd_jid);
637 }
638
639 out_unlock:
640 gfs2_glock_dq_uninit(&j_gh);
641 return error;
642 }
643
init_journal(struct gfs2_sbd * sdp,int undo)644 static int init_journal(struct gfs2_sbd *sdp, int undo)
645 {
646 struct inode *master = d_inode(sdp->sd_master_dir);
647 struct gfs2_holder ji_gh;
648 struct gfs2_inode *ip;
649 int jindex = 1;
650 int error = 0;
651
652 if (undo) {
653 jindex = 0;
654 goto fail_jinode_gh;
655 }
656
657 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
658 if (IS_ERR(sdp->sd_jindex)) {
659 fs_err(sdp, "can't lookup journal index: %d\n", error);
660 return PTR_ERR(sdp->sd_jindex);
661 }
662
663 /* Load in the journal index special file */
664
665 error = gfs2_jindex_hold(sdp, &ji_gh);
666 if (error) {
667 fs_err(sdp, "can't read journal index: %d\n", error);
668 goto fail;
669 }
670
671 error = -EUSERS;
672 if (!gfs2_jindex_size(sdp)) {
673 fs_err(sdp, "no journals!\n");
674 goto fail_jindex;
675 }
676
677 atomic_set(&sdp->sd_log_blks_needed, 0);
678 if (sdp->sd_args.ar_spectator) {
679 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
680 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
681 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
682 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
683 } else {
684 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
685 fs_err(sdp, "can't mount journal #%u\n",
686 sdp->sd_lockstruct.ls_jid);
687 fs_err(sdp, "there are only %u journals (0 - %u)\n",
688 gfs2_jindex_size(sdp),
689 gfs2_jindex_size(sdp) - 1);
690 goto fail_jindex;
691 }
692 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
693
694 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
695 &gfs2_journal_glops,
696 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
697 &sdp->sd_journal_gh);
698 if (error) {
699 fs_err(sdp, "can't acquire journal glock: %d\n", error);
700 goto fail_jindex;
701 }
702
703 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
704 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
705 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
706 &sdp->sd_jinode_gh);
707 if (error) {
708 fs_err(sdp, "can't acquire journal inode glock: %d\n",
709 error);
710 goto fail_journal_gh;
711 }
712
713 error = gfs2_jdesc_check(sdp->sd_jdesc);
714 if (error) {
715 fs_err(sdp, "my journal (%u) is bad: %d\n",
716 sdp->sd_jdesc->jd_jid, error);
717 goto fail_jinode_gh;
718 }
719 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
720 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
721 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
722
723 /* Map the extents for this journal's blocks */
724 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
725 }
726 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
727
728 if (sdp->sd_lockstruct.ls_first) {
729 unsigned int x;
730 for (x = 0; x < sdp->sd_journals; x++) {
731 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
732
733 if (sdp->sd_args.ar_spectator) {
734 error = check_journal_clean(sdp, jd);
735 if (error)
736 goto fail_jinode_gh;
737 continue;
738 }
739 error = gfs2_recover_journal(jd, true);
740 if (error) {
741 fs_err(sdp, "error recovering journal %u: %d\n",
742 x, error);
743 goto fail_jinode_gh;
744 }
745 }
746
747 gfs2_others_may_mount(sdp);
748 } else if (!sdp->sd_args.ar_spectator) {
749 error = gfs2_recover_journal(sdp->sd_jdesc, true);
750 if (error) {
751 fs_err(sdp, "error recovering my journal: %d\n", error);
752 goto fail_jinode_gh;
753 }
754 }
755
756 sdp->sd_log_idle = 1;
757 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
758 gfs2_glock_dq_uninit(&ji_gh);
759 jindex = 0;
760 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
761 return 0;
762
763 fail_jinode_gh:
764 if (!sdp->sd_args.ar_spectator)
765 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
766 fail_journal_gh:
767 if (!sdp->sd_args.ar_spectator)
768 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
769 fail_jindex:
770 gfs2_jindex_free(sdp);
771 if (jindex)
772 gfs2_glock_dq_uninit(&ji_gh);
773 fail:
774 iput(sdp->sd_jindex);
775 return error;
776 }
777
778 static struct lock_class_key gfs2_quota_imutex_key;
779
init_inodes(struct gfs2_sbd * sdp,int undo)780 static int init_inodes(struct gfs2_sbd *sdp, int undo)
781 {
782 int error = 0;
783 struct inode *master = d_inode(sdp->sd_master_dir);
784
785 if (undo)
786 goto fail_qinode;
787
788 error = init_journal(sdp, undo);
789 complete_all(&sdp->sd_journal_ready);
790 if (error)
791 goto fail;
792
793 /* Read in the master statfs inode */
794 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
795 if (IS_ERR(sdp->sd_statfs_inode)) {
796 error = PTR_ERR(sdp->sd_statfs_inode);
797 fs_err(sdp, "can't read in statfs inode: %d\n", error);
798 goto fail_journal;
799 }
800
801 /* Read in the resource index inode */
802 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
803 if (IS_ERR(sdp->sd_rindex)) {
804 error = PTR_ERR(sdp->sd_rindex);
805 fs_err(sdp, "can't get resource index inode: %d\n", error);
806 goto fail_statfs;
807 }
808 sdp->sd_rindex_uptodate = 0;
809
810 /* Read in the quota inode */
811 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
812 if (IS_ERR(sdp->sd_quota_inode)) {
813 error = PTR_ERR(sdp->sd_quota_inode);
814 fs_err(sdp, "can't get quota file inode: %d\n", error);
815 goto fail_rindex;
816 }
817 /*
818 * i_rwsem on quota files is special. Since this inode is hidden system
819 * file, we are safe to define locking ourselves.
820 */
821 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
822 &gfs2_quota_imutex_key);
823
824 error = gfs2_rindex_update(sdp);
825 if (error)
826 goto fail_qinode;
827
828 return 0;
829
830 fail_qinode:
831 iput(sdp->sd_quota_inode);
832 fail_rindex:
833 gfs2_clear_rgrpd(sdp);
834 iput(sdp->sd_rindex);
835 fail_statfs:
836 iput(sdp->sd_statfs_inode);
837 fail_journal:
838 init_journal(sdp, UNDO);
839 fail:
840 return error;
841 }
842
init_per_node(struct gfs2_sbd * sdp,int undo)843 static int init_per_node(struct gfs2_sbd *sdp, int undo)
844 {
845 struct inode *pn = NULL;
846 char buf[30];
847 int error = 0;
848 struct gfs2_inode *ip;
849 struct inode *master = d_inode(sdp->sd_master_dir);
850
851 if (sdp->sd_args.ar_spectator)
852 return 0;
853
854 if (undo)
855 goto fail_qc_gh;
856
857 pn = gfs2_lookup_simple(master, "per_node");
858 if (IS_ERR(pn)) {
859 error = PTR_ERR(pn);
860 fs_err(sdp, "can't find per_node directory: %d\n", error);
861 return error;
862 }
863
864 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
865 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
866 if (IS_ERR(sdp->sd_sc_inode)) {
867 error = PTR_ERR(sdp->sd_sc_inode);
868 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
869 goto fail;
870 }
871
872 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
873 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
874 if (IS_ERR(sdp->sd_qc_inode)) {
875 error = PTR_ERR(sdp->sd_qc_inode);
876 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
877 goto fail_ut_i;
878 }
879
880 iput(pn);
881 pn = NULL;
882
883 ip = GFS2_I(sdp->sd_sc_inode);
884 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
885 &sdp->sd_sc_gh);
886 if (error) {
887 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
888 goto fail_qc_i;
889 }
890
891 ip = GFS2_I(sdp->sd_qc_inode);
892 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
893 &sdp->sd_qc_gh);
894 if (error) {
895 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
896 goto fail_ut_gh;
897 }
898
899 return 0;
900
901 fail_qc_gh:
902 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
903 fail_ut_gh:
904 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
905 fail_qc_i:
906 iput(sdp->sd_qc_inode);
907 fail_ut_i:
908 iput(sdp->sd_sc_inode);
909 fail:
910 iput(pn);
911 return error;
912 }
913
914 static const match_table_t nolock_tokens = {
915 { Opt_jid, "jid=%d", },
916 { Opt_err, NULL },
917 };
918
919 static const struct lm_lockops nolock_ops = {
920 .lm_proto_name = "lock_nolock",
921 .lm_put_lock = gfs2_glock_free,
922 .lm_tokens = &nolock_tokens,
923 };
924
925 /**
926 * gfs2_lm_mount - mount a locking protocol
927 * @sdp: the filesystem
928 * @args: mount arguments
929 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
930 *
931 * Returns: errno
932 */
933
gfs2_lm_mount(struct gfs2_sbd * sdp,int silent)934 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
935 {
936 const struct lm_lockops *lm;
937 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
938 struct gfs2_args *args = &sdp->sd_args;
939 const char *proto = sdp->sd_proto_name;
940 const char *table = sdp->sd_table_name;
941 char *o, *options;
942 int ret;
943
944 if (!strcmp("lock_nolock", proto)) {
945 lm = &nolock_ops;
946 sdp->sd_args.ar_localflocks = 1;
947 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
948 } else if (!strcmp("lock_dlm", proto)) {
949 lm = &gfs2_dlm_ops;
950 #endif
951 } else {
952 pr_info("can't find protocol %s\n", proto);
953 return -ENOENT;
954 }
955
956 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
957
958 ls->ls_ops = lm;
959 ls->ls_first = 1;
960
961 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
962 substring_t tmp[MAX_OPT_ARGS];
963 int token, option;
964
965 if (!o || !*o)
966 continue;
967
968 token = match_token(o, *lm->lm_tokens, tmp);
969 switch (token) {
970 case Opt_jid:
971 ret = match_int(&tmp[0], &option);
972 if (ret || option < 0)
973 goto hostdata_error;
974 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
975 ls->ls_jid = option;
976 break;
977 case Opt_id:
978 case Opt_nodir:
979 /* Obsolete, but left for backward compat purposes */
980 break;
981 case Opt_first:
982 ret = match_int(&tmp[0], &option);
983 if (ret || (option != 0 && option != 1))
984 goto hostdata_error;
985 ls->ls_first = option;
986 break;
987 case Opt_err:
988 default:
989 hostdata_error:
990 fs_info(sdp, "unknown hostdata (%s)\n", o);
991 return -EINVAL;
992 }
993 }
994
995 if (lm->lm_mount == NULL) {
996 fs_info(sdp, "Now mounting FS...\n");
997 complete_all(&sdp->sd_locking_init);
998 return 0;
999 }
1000 ret = lm->lm_mount(sdp, table);
1001 if (ret == 0)
1002 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1003 complete_all(&sdp->sd_locking_init);
1004 return ret;
1005 }
1006
gfs2_lm_unmount(struct gfs2_sbd * sdp)1007 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1008 {
1009 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1010 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1011 lm->lm_unmount)
1012 lm->lm_unmount(sdp);
1013 }
1014
wait_on_journal(struct gfs2_sbd * sdp)1015 static int wait_on_journal(struct gfs2_sbd *sdp)
1016 {
1017 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1018 return 0;
1019
1020 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1021 ? -EINTR : 0;
1022 }
1023
gfs2_online_uevent(struct gfs2_sbd * sdp)1024 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1025 {
1026 struct super_block *sb = sdp->sd_vfs;
1027 char ro[20];
1028 char spectator[20];
1029 char *envp[] = { ro, spectator, NULL };
1030 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1031 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1032 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1033 }
1034
1035 /**
1036 * fill_super - Read in superblock
1037 * @sb: The VFS superblock
1038 * @data: Mount options
1039 * @silent: Don't complain if it's not a GFS2 filesystem
1040 *
1041 * Returns: errno
1042 */
1043
fill_super(struct super_block * sb,struct gfs2_args * args,int silent)1044 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1045 {
1046 struct gfs2_sbd *sdp;
1047 struct gfs2_holder mount_gh;
1048 int error;
1049
1050 sdp = init_sbd(sb);
1051 if (!sdp) {
1052 pr_warn("can't alloc struct gfs2_sbd\n");
1053 return -ENOMEM;
1054 }
1055 sdp->sd_args = *args;
1056
1057 if (sdp->sd_args.ar_spectator) {
1058 sb->s_flags |= SB_RDONLY;
1059 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1060 }
1061 if (sdp->sd_args.ar_posix_acl)
1062 sb->s_flags |= SB_POSIXACL;
1063 if (sdp->sd_args.ar_nobarrier)
1064 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1065
1066 sb->s_flags |= SB_NOSEC;
1067 sb->s_magic = GFS2_MAGIC;
1068 sb->s_op = &gfs2_super_ops;
1069 sb->s_d_op = &gfs2_dops;
1070 sb->s_export_op = &gfs2_export_ops;
1071 sb->s_xattr = gfs2_xattr_handlers;
1072 sb->s_qcop = &gfs2_quotactl_ops;
1073 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1074 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1075 sb->s_time_gran = 1;
1076 sb->s_maxbytes = MAX_LFS_FILESIZE;
1077
1078 /* Set up the buffer cache and fill in some fake block size values
1079 to allow us to read-in the on-disk superblock. */
1080 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1081 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1082 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1083 GFS2_BASIC_BLOCK_SHIFT;
1084 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1085
1086 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1087 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1088 if (sdp->sd_args.ar_statfs_quantum) {
1089 sdp->sd_tune.gt_statfs_slow = 0;
1090 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1091 } else {
1092 sdp->sd_tune.gt_statfs_slow = 1;
1093 sdp->sd_tune.gt_statfs_quantum = 30;
1094 }
1095
1096 error = init_names(sdp, silent);
1097 if (error) {
1098 /* In this case, we haven't initialized sysfs, so we have to
1099 manually free the sdp. */
1100 free_percpu(sdp->sd_lkstats);
1101 kfree(sdp);
1102 sb->s_fs_info = NULL;
1103 return error;
1104 }
1105
1106 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1107
1108 error = gfs2_sys_fs_add(sdp);
1109 /*
1110 * If we hit an error here, gfs2_sys_fs_add will have called function
1111 * kobject_put which causes the sysfs usage count to go to zero, which
1112 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1113 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1114 * kobject_put to free sdp.
1115 */
1116 if (error)
1117 return error;
1118
1119 gfs2_create_debugfs_file(sdp);
1120
1121 error = gfs2_lm_mount(sdp, silent);
1122 if (error)
1123 goto fail_debug;
1124
1125 error = init_locking(sdp, &mount_gh, DO);
1126 if (error)
1127 goto fail_lm;
1128
1129 error = init_sb(sdp, silent);
1130 if (error)
1131 goto fail_locking;
1132
1133 error = wait_on_journal(sdp);
1134 if (error)
1135 goto fail_sb;
1136
1137 /*
1138 * If user space has failed to join the cluster or some similar
1139 * failure has occurred, then the journal id will contain a
1140 * negative (error) number. This will then be returned to the
1141 * caller (of the mount syscall). We do this even for spectator
1142 * mounts (which just write a jid of 0 to indicate "ok" even though
1143 * the jid is unused in the spectator case)
1144 */
1145 if (sdp->sd_lockstruct.ls_jid < 0) {
1146 error = sdp->sd_lockstruct.ls_jid;
1147 sdp->sd_lockstruct.ls_jid = 0;
1148 goto fail_sb;
1149 }
1150
1151 if (sdp->sd_args.ar_spectator)
1152 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1153 sdp->sd_table_name);
1154 else
1155 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1156 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1157
1158 error = init_inodes(sdp, DO);
1159 if (error)
1160 goto fail_sb;
1161
1162 error = init_per_node(sdp, DO);
1163 if (error)
1164 goto fail_inodes;
1165
1166 error = gfs2_statfs_init(sdp);
1167 if (error) {
1168 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1169 goto fail_per_node;
1170 }
1171
1172 if (sb_rdonly(sb)) {
1173 struct gfs2_holder freeze_gh;
1174
1175 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1176 GL_EXACT, &freeze_gh);
1177 if (error) {
1178 fs_err(sdp, "can't make FS RO: %d\n", error);
1179 goto fail_per_node;
1180 }
1181 gfs2_glock_dq_uninit(&freeze_gh);
1182 } else {
1183 error = gfs2_make_fs_rw(sdp);
1184 if (error) {
1185 fs_err(sdp, "can't make FS RW: %d\n", error);
1186 goto fail_per_node;
1187 }
1188 }
1189
1190 gfs2_glock_dq_uninit(&mount_gh);
1191 gfs2_online_uevent(sdp);
1192 return 0;
1193
1194 fail_per_node:
1195 init_per_node(sdp, UNDO);
1196 fail_inodes:
1197 init_inodes(sdp, UNDO);
1198 fail_sb:
1199 if (sdp->sd_root_dir)
1200 dput(sdp->sd_root_dir);
1201 if (sdp->sd_master_dir)
1202 dput(sdp->sd_master_dir);
1203 if (sb->s_root)
1204 dput(sb->s_root);
1205 sb->s_root = NULL;
1206 fail_locking:
1207 init_locking(sdp, &mount_gh, UNDO);
1208 fail_lm:
1209 complete_all(&sdp->sd_journal_ready);
1210 gfs2_gl_hash_clear(sdp);
1211 gfs2_lm_unmount(sdp);
1212 fail_debug:
1213 gfs2_delete_debugfs_file(sdp);
1214 free_percpu(sdp->sd_lkstats);
1215 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1216 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1217 gfs2_sys_fs_del(sdp);
1218 sb->s_fs_info = NULL;
1219 return error;
1220 }
1221
set_gfs2_super(struct super_block * s,void * data)1222 static int set_gfs2_super(struct super_block *s, void *data)
1223 {
1224 s->s_bdev = data;
1225 s->s_dev = s->s_bdev->bd_dev;
1226 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1227 return 0;
1228 }
1229
test_gfs2_super(struct super_block * s,void * ptr)1230 static int test_gfs2_super(struct super_block *s, void *ptr)
1231 {
1232 struct block_device *bdev = ptr;
1233 return (bdev == s->s_bdev);
1234 }
1235
1236 /**
1237 * gfs2_mount - Get the GFS2 superblock
1238 * @fs_type: The GFS2 filesystem type
1239 * @flags: Mount flags
1240 * @dev_name: The name of the device
1241 * @data: The mount arguments
1242 *
1243 * Q. Why not use get_sb_bdev() ?
1244 * A. We need to select one of two root directories to mount, independent
1245 * of whether this is the initial, or subsequent, mount of this sb
1246 *
1247 * Returns: 0 or -ve on error
1248 */
1249
gfs2_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1250 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1251 const char *dev_name, void *data)
1252 {
1253 struct block_device *bdev;
1254 struct super_block *s;
1255 fmode_t mode = FMODE_READ | FMODE_EXCL;
1256 int error;
1257 struct gfs2_args args;
1258 struct gfs2_sbd *sdp;
1259
1260 if (!(flags & SB_RDONLY))
1261 mode |= FMODE_WRITE;
1262
1263 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1264 if (IS_ERR(bdev))
1265 return ERR_CAST(bdev);
1266
1267 /*
1268 * once the super is inserted into the list by sget, s_umount
1269 * will protect the lockfs code from trying to start a snapshot
1270 * while we are mounting
1271 */
1272 mutex_lock(&bdev->bd_fsfreeze_mutex);
1273 if (bdev->bd_fsfreeze_count > 0) {
1274 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1275 error = -EBUSY;
1276 goto error_bdev;
1277 }
1278 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1280 error = PTR_ERR(s);
1281 if (IS_ERR(s))
1282 goto error_bdev;
1283
1284 if (s->s_root) {
1285 /*
1286 * s_umount nests inside bd_mutex during
1287 * __invalidate_device(). blkdev_put() acquires
1288 * bd_mutex and can't be called under s_umount. Drop
1289 * s_umount temporarily. This is safe as we're
1290 * holding an active reference.
1291 */
1292 up_write(&s->s_umount);
1293 blkdev_put(bdev, mode);
1294 down_write(&s->s_umount);
1295 } else {
1296 /* s_mode must be set before deactivate_locked_super calls */
1297 s->s_mode = mode;
1298 }
1299
1300 memset(&args, 0, sizeof(args));
1301 args.ar_quota = GFS2_QUOTA_DEFAULT;
1302 args.ar_data = GFS2_DATA_DEFAULT;
1303 args.ar_commit = 30;
1304 args.ar_statfs_quantum = 30;
1305 args.ar_quota_quantum = 60;
1306 args.ar_errors = GFS2_ERRORS_DEFAULT;
1307
1308 error = gfs2_mount_args(&args, data);
1309 if (error) {
1310 pr_warn("can't parse mount arguments\n");
1311 goto error_super;
1312 }
1313
1314 if (s->s_root) {
1315 error = -EBUSY;
1316 if ((flags ^ s->s_flags) & SB_RDONLY)
1317 goto error_super;
1318 } else {
1319 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1320 sb_set_blocksize(s, block_size(bdev));
1321 error = fill_super(s, &args, flags & SB_SILENT ? 1 : 0);
1322 if (error)
1323 goto error_super;
1324 s->s_flags |= SB_ACTIVE;
1325 bdev->bd_super = s;
1326 }
1327
1328 sdp = s->s_fs_info;
1329 if (args.ar_meta)
1330 return dget(sdp->sd_master_dir);
1331 else
1332 return dget(sdp->sd_root_dir);
1333
1334 error_super:
1335 deactivate_locked_super(s);
1336 return ERR_PTR(error);
1337 error_bdev:
1338 blkdev_put(bdev, mode);
1339 return ERR_PTR(error);
1340 }
1341
set_meta_super(struct super_block * s,void * ptr)1342 static int set_meta_super(struct super_block *s, void *ptr)
1343 {
1344 return -EINVAL;
1345 }
1346
gfs2_mount_meta(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1347 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1348 int flags, const char *dev_name, void *data)
1349 {
1350 struct super_block *s;
1351 struct gfs2_sbd *sdp;
1352 struct path path;
1353 int error;
1354
1355 if (!dev_name || !*dev_name)
1356 return ERR_PTR(-EINVAL);
1357
1358 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1359 if (error) {
1360 pr_warn("path_lookup on %s returned error %d\n",
1361 dev_name, error);
1362 return ERR_PTR(error);
1363 }
1364 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1365 path.dentry->d_sb->s_bdev);
1366 path_put(&path);
1367 if (IS_ERR(s)) {
1368 pr_warn("gfs2 mount does not exist\n");
1369 return ERR_CAST(s);
1370 }
1371 if ((flags ^ s->s_flags) & SB_RDONLY) {
1372 deactivate_locked_super(s);
1373 return ERR_PTR(-EBUSY);
1374 }
1375 sdp = s->s_fs_info;
1376 return dget(sdp->sd_master_dir);
1377 }
1378
gfs2_kill_sb(struct super_block * sb)1379 static void gfs2_kill_sb(struct super_block *sb)
1380 {
1381 struct gfs2_sbd *sdp = sb->s_fs_info;
1382
1383 if (sdp == NULL) {
1384 kill_block_super(sb);
1385 return;
1386 }
1387
1388 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1389 dput(sdp->sd_root_dir);
1390 dput(sdp->sd_master_dir);
1391 sdp->sd_root_dir = NULL;
1392 sdp->sd_master_dir = NULL;
1393 shrink_dcache_sb(sb);
1394 free_percpu(sdp->sd_lkstats);
1395 kill_block_super(sb);
1396 }
1397
1398 struct file_system_type gfs2_fs_type = {
1399 .name = "gfs2",
1400 .fs_flags = FS_REQUIRES_DEV,
1401 .mount = gfs2_mount,
1402 .kill_sb = gfs2_kill_sb,
1403 .owner = THIS_MODULE,
1404 };
1405 MODULE_ALIAS_FS("gfs2");
1406
1407 struct file_system_type gfs2meta_fs_type = {
1408 .name = "gfs2meta",
1409 .fs_flags = FS_REQUIRES_DEV,
1410 .mount = gfs2_mount_meta,
1411 .owner = THIS_MODULE,
1412 };
1413 MODULE_ALIAS_FS("gfs2meta");
1414