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 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/blkdev.h>
16 #include <linux/kthread.h>
17 #include <linux/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
22
23 #include "gfs2.h"
24 #include "incore.h"
25 #include "bmap.h"
26 #include "glock.h"
27 #include "glops.h"
28 #include "inode.h"
29 #include "recovery.h"
30 #include "rgrp.h"
31 #include "super.h"
32 #include "sys.h"
33 #include "util.h"
34 #include "log.h"
35 #include "quota.h"
36 #include "dir.h"
37 #include "trace_gfs2.h"
38
39 #define DO 0
40 #define UNDO 1
41
42 /**
43 * gfs2_tune_init - Fill a gfs2_tune structure with default values
44 * @gt: tune
45 *
46 */
47
gfs2_tune_init(struct gfs2_tune * gt)48 static void gfs2_tune_init(struct gfs2_tune *gt)
49 {
50 spin_lock_init(>->gt_spin);
51
52 gt->gt_quota_simul_sync = 64;
53 gt->gt_quota_warn_period = 10;
54 gt->gt_quota_scale_num = 1;
55 gt->gt_quota_scale_den = 1;
56 gt->gt_new_files_jdata = 0;
57 gt->gt_max_readahead = 1 << 18;
58 gt->gt_complain_secs = 10;
59 }
60
init_sbd(struct super_block * sb)61 static struct gfs2_sbd *init_sbd(struct super_block *sb)
62 {
63 struct gfs2_sbd *sdp;
64
65 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
66 if (!sdp)
67 return NULL;
68
69 sb->s_fs_info = sdp;
70 sdp->sd_vfs = sb;
71 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
72 gfs2_tune_init(&sdp->sd_tune);
73
74 init_waitqueue_head(&sdp->sd_glock_wait);
75 atomic_set(&sdp->sd_glock_disposal, 0);
76 init_completion(&sdp->sd_locking_init);
77 spin_lock_init(&sdp->sd_statfs_spin);
78
79 spin_lock_init(&sdp->sd_rindex_spin);
80 mutex_init(&sdp->sd_rindex_mutex);
81 sdp->sd_rindex_tree.rb_node = NULL;
82
83 INIT_LIST_HEAD(&sdp->sd_jindex_list);
84 spin_lock_init(&sdp->sd_jindex_spin);
85 mutex_init(&sdp->sd_jindex_mutex);
86
87 INIT_LIST_HEAD(&sdp->sd_quota_list);
88 mutex_init(&sdp->sd_quota_mutex);
89 init_waitqueue_head(&sdp->sd_quota_wait);
90 INIT_LIST_HEAD(&sdp->sd_trunc_list);
91 spin_lock_init(&sdp->sd_trunc_lock);
92
93 spin_lock_init(&sdp->sd_log_lock);
94 atomic_set(&sdp->sd_log_pinned, 0);
95 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
96 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
97 INIT_LIST_HEAD(&sdp->sd_log_le_rg);
98 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
99 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
100
101 init_waitqueue_head(&sdp->sd_log_waitq);
102 init_waitqueue_head(&sdp->sd_logd_waitq);
103 spin_lock_init(&sdp->sd_ail_lock);
104 INIT_LIST_HEAD(&sdp->sd_ail1_list);
105 INIT_LIST_HEAD(&sdp->sd_ail2_list);
106
107 init_rwsem(&sdp->sd_log_flush_lock);
108 atomic_set(&sdp->sd_log_in_flight, 0);
109 init_waitqueue_head(&sdp->sd_log_flush_wait);
110
111 INIT_LIST_HEAD(&sdp->sd_revoke_list);
112
113 mutex_init(&sdp->sd_freeze_lock);
114
115 return sdp;
116 }
117
118
119 /**
120 * gfs2_check_sb - Check superblock
121 * @sdp: the filesystem
122 * @sb: The superblock
123 * @silent: Don't print a message if the check fails
124 *
125 * Checks the version code of the FS is one that we understand how to
126 * read and that the sizes of the various on-disk structures have not
127 * changed.
128 */
129
gfs2_check_sb(struct gfs2_sbd * sdp,int silent)130 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
131 {
132 struct gfs2_sb_host *sb = &sdp->sd_sb;
133
134 if (sb->sb_magic != GFS2_MAGIC ||
135 sb->sb_type != GFS2_METATYPE_SB) {
136 if (!silent)
137 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
138 return -EINVAL;
139 }
140
141 /* If format numbers match exactly, we're done. */
142
143 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
144 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
145 return 0;
146
147 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
148
149 return -EINVAL;
150 }
151
end_bio_io_page(struct bio * bio,int error)152 static void end_bio_io_page(struct bio *bio, int error)
153 {
154 struct page *page = bio->bi_private;
155
156 if (!error)
157 SetPageUptodate(page);
158 else
159 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
160 unlock_page(page);
161 }
162
gfs2_sb_in(struct gfs2_sbd * sdp,const void * buf)163 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
164 {
165 struct gfs2_sb_host *sb = &sdp->sd_sb;
166 struct super_block *s = sdp->sd_vfs;
167 const struct gfs2_sb *str = buf;
168
169 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
170 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
171 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
172 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
173 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
174 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
175 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
176 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
177 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
178 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
179 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
180
181 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
182 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
183 memcpy(s->s_uuid, str->sb_uuid, 16);
184 }
185
186 /**
187 * gfs2_read_super - Read the gfs2 super block from disk
188 * @sdp: The GFS2 super block
189 * @sector: The location of the super block
190 * @error: The error code to return
191 *
192 * This uses the bio functions to read the super block from disk
193 * because we want to be 100% sure that we never read cached data.
194 * A super block is read twice only during each GFS2 mount and is
195 * never written to by the filesystem. The first time its read no
196 * locks are held, and the only details which are looked at are those
197 * relating to the locking protocol. Once locking is up and working,
198 * the sb is read again under the lock to establish the location of
199 * the master directory (contains pointers to journals etc) and the
200 * root directory.
201 *
202 * Returns: 0 on success or error
203 */
204
gfs2_read_super(struct gfs2_sbd * sdp,sector_t sector,int silent)205 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
206 {
207 struct super_block *sb = sdp->sd_vfs;
208 struct gfs2_sb *p;
209 struct page *page;
210 struct bio *bio;
211
212 page = alloc_page(GFP_NOFS);
213 if (unlikely(!page))
214 return -ENOBUFS;
215
216 ClearPageUptodate(page);
217 ClearPageDirty(page);
218 lock_page(page);
219
220 bio = bio_alloc(GFP_NOFS, 1);
221 bio->bi_sector = sector * (sb->s_blocksize >> 9);
222 bio->bi_bdev = sb->s_bdev;
223 bio_add_page(bio, page, PAGE_SIZE, 0);
224
225 bio->bi_end_io = end_bio_io_page;
226 bio->bi_private = page;
227 submit_bio(READ_SYNC | REQ_META, bio);
228 wait_on_page_locked(page);
229 bio_put(bio);
230 if (!PageUptodate(page)) {
231 __free_page(page);
232 return -EIO;
233 }
234 p = kmap(page);
235 gfs2_sb_in(sdp, p);
236 kunmap(page);
237 __free_page(page);
238 return gfs2_check_sb(sdp, silent);
239 }
240
241 /**
242 * gfs2_read_sb - Read super block
243 * @sdp: The GFS2 superblock
244 * @silent: Don't print message if mount fails
245 *
246 */
247
gfs2_read_sb(struct gfs2_sbd * sdp,int silent)248 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
249 {
250 u32 hash_blocks, ind_blocks, leaf_blocks;
251 u32 tmp_blocks;
252 unsigned int x;
253 int error;
254
255 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
256 if (error) {
257 if (!silent)
258 fs_err(sdp, "can't read superblock\n");
259 return error;
260 }
261
262 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
263 GFS2_BASIC_BLOCK_SHIFT;
264 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
265 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
266 sizeof(struct gfs2_dinode)) / sizeof(u64);
267 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
268 sizeof(struct gfs2_meta_header)) / sizeof(u64);
269 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
270 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
271 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
272 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
273 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
274 sizeof(struct gfs2_meta_header)) /
275 sizeof(struct gfs2_quota_change);
276
277 /* Compute maximum reservation required to add a entry to a directory */
278
279 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
280 sdp->sd_jbsize);
281
282 ind_blocks = 0;
283 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
284 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
285 ind_blocks += tmp_blocks;
286 }
287
288 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
289
290 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
291
292 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
293 sizeof(struct gfs2_dinode);
294 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
295 for (x = 2;; x++) {
296 u64 space, d;
297 u32 m;
298
299 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
300 d = space;
301 m = do_div(d, sdp->sd_inptrs);
302
303 if (d != sdp->sd_heightsize[x - 1] || m)
304 break;
305 sdp->sd_heightsize[x] = space;
306 }
307 sdp->sd_max_height = x;
308 sdp->sd_heightsize[x] = ~0;
309 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
310
311 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
312 sizeof(struct gfs2_dinode);
313 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
314 for (x = 2;; x++) {
315 u64 space, d;
316 u32 m;
317
318 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
319 d = space;
320 m = do_div(d, sdp->sd_inptrs);
321
322 if (d != sdp->sd_jheightsize[x - 1] || m)
323 break;
324 sdp->sd_jheightsize[x] = space;
325 }
326 sdp->sd_max_jheight = x;
327 sdp->sd_jheightsize[x] = ~0;
328 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
329
330 return 0;
331 }
332
init_names(struct gfs2_sbd * sdp,int silent)333 static int init_names(struct gfs2_sbd *sdp, int silent)
334 {
335 char *proto, *table;
336 int error = 0;
337
338 proto = sdp->sd_args.ar_lockproto;
339 table = sdp->sd_args.ar_locktable;
340
341 /* Try to autodetect */
342
343 if (!proto[0] || !table[0]) {
344 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
345 if (error)
346 return error;
347
348 if (!proto[0])
349 proto = sdp->sd_sb.sb_lockproto;
350 if (!table[0])
351 table = sdp->sd_sb.sb_locktable;
352 }
353
354 if (!table[0])
355 table = sdp->sd_vfs->s_id;
356
357 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
358 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
359
360 table = sdp->sd_table_name;
361 while ((table = strchr(table, '/')))
362 *table = '_';
363
364 return error;
365 }
366
init_locking(struct gfs2_sbd * sdp,struct gfs2_holder * mount_gh,int undo)367 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
368 int undo)
369 {
370 int error = 0;
371
372 if (undo)
373 goto fail_trans;
374
375 error = gfs2_glock_nq_num(sdp,
376 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
377 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
378 mount_gh);
379 if (error) {
380 fs_err(sdp, "can't acquire mount glock: %d\n", error);
381 goto fail;
382 }
383
384 error = gfs2_glock_nq_num(sdp,
385 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
386 LM_ST_SHARED,
387 LM_FLAG_NOEXP | GL_EXACT,
388 &sdp->sd_live_gh);
389 if (error) {
390 fs_err(sdp, "can't acquire live glock: %d\n", error);
391 goto fail_mount;
392 }
393
394 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
395 CREATE, &sdp->sd_rename_gl);
396 if (error) {
397 fs_err(sdp, "can't create rename glock: %d\n", error);
398 goto fail_live;
399 }
400
401 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
402 CREATE, &sdp->sd_trans_gl);
403 if (error) {
404 fs_err(sdp, "can't create transaction glock: %d\n", error);
405 goto fail_rename;
406 }
407
408 return 0;
409
410 fail_trans:
411 gfs2_glock_put(sdp->sd_trans_gl);
412 fail_rename:
413 gfs2_glock_put(sdp->sd_rename_gl);
414 fail_live:
415 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
416 fail_mount:
417 gfs2_glock_dq_uninit(mount_gh);
418 fail:
419 return error;
420 }
421
gfs2_lookup_root(struct super_block * sb,struct dentry ** dptr,u64 no_addr,const char * name)422 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
423 u64 no_addr, const char *name)
424 {
425 struct gfs2_sbd *sdp = sb->s_fs_info;
426 struct dentry *dentry;
427 struct inode *inode;
428
429 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
430 if (IS_ERR(inode)) {
431 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
432 return PTR_ERR(inode);
433 }
434 dentry = d_alloc_root(inode);
435 if (!dentry) {
436 fs_err(sdp, "can't alloc %s dentry\n", name);
437 iput(inode);
438 return -ENOMEM;
439 }
440 *dptr = dentry;
441 return 0;
442 }
443
init_sb(struct gfs2_sbd * sdp,int silent)444 static int init_sb(struct gfs2_sbd *sdp, int silent)
445 {
446 struct super_block *sb = sdp->sd_vfs;
447 struct gfs2_holder sb_gh;
448 u64 no_addr;
449 int ret;
450
451 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
452 LM_ST_SHARED, 0, &sb_gh);
453 if (ret) {
454 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
455 return ret;
456 }
457
458 ret = gfs2_read_sb(sdp, silent);
459 if (ret) {
460 fs_err(sdp, "can't read superblock: %d\n", ret);
461 goto out;
462 }
463
464 /* Set up the buffer cache and SB for real */
465 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
466 ret = -EINVAL;
467 fs_err(sdp, "FS block size (%u) is too small for device "
468 "block size (%u)\n",
469 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
470 goto out;
471 }
472 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
473 ret = -EINVAL;
474 fs_err(sdp, "FS block size (%u) is too big for machine "
475 "page size (%u)\n",
476 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
477 goto out;
478 }
479 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
480
481 /* Get the root inode */
482 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
483 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
484 if (ret)
485 goto out;
486
487 /* Get the master inode */
488 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
489 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
490 if (ret) {
491 dput(sdp->sd_root_dir);
492 goto out;
493 }
494 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
495 out:
496 gfs2_glock_dq_uninit(&sb_gh);
497 return ret;
498 }
499
500 /**
501 * map_journal_extents - create a reusable "extent" mapping from all logical
502 * blocks to all physical blocks for the given journal. This will save
503 * us time when writing journal blocks. Most journals will have only one
504 * extent that maps all their logical blocks. That's because gfs2.mkfs
505 * arranges the journal blocks sequentially to maximize performance.
506 * So the extent would map the first block for the entire file length.
507 * However, gfs2_jadd can happen while file activity is happening, so
508 * those journals may not be sequential. Less likely is the case where
509 * the users created their own journals by mounting the metafs and
510 * laying it out. But it's still possible. These journals might have
511 * several extents.
512 *
513 * TODO: This should be done in bigger chunks rather than one block at a time,
514 * but since it's only done at mount time, I'm not worried about the
515 * time it takes.
516 */
map_journal_extents(struct gfs2_sbd * sdp)517 static int map_journal_extents(struct gfs2_sbd *sdp)
518 {
519 struct gfs2_jdesc *jd = sdp->sd_jdesc;
520 unsigned int lb;
521 u64 db, prev_db; /* logical block, disk block, prev disk block */
522 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
523 struct gfs2_journal_extent *jext = NULL;
524 struct buffer_head bh;
525 int rc = 0;
526
527 prev_db = 0;
528
529 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
530 bh.b_state = 0;
531 bh.b_blocknr = 0;
532 bh.b_size = 1 << ip->i_inode.i_blkbits;
533 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
534 db = bh.b_blocknr;
535 if (rc || !db) {
536 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
537 "%u db=%llu\n", rc, lb, (unsigned long long)db);
538 break;
539 }
540 if (!prev_db || db != prev_db + 1) {
541 jext = kzalloc(sizeof(struct gfs2_journal_extent),
542 GFP_KERNEL);
543 if (!jext) {
544 printk(KERN_INFO "GFS2 error: out of memory "
545 "mapping journal extents.\n");
546 rc = -ENOMEM;
547 break;
548 }
549 jext->dblock = db;
550 jext->lblock = lb;
551 jext->blocks = 1;
552 list_add_tail(&jext->extent_list, &jd->extent_list);
553 } else {
554 jext->blocks++;
555 }
556 prev_db = db;
557 }
558 return rc;
559 }
560
gfs2_others_may_mount(struct gfs2_sbd * sdp)561 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
562 {
563 char *message = "FIRSTMOUNT=Done";
564 char *envp[] = { message, NULL };
565
566 fs_info(sdp, "first mount done, others may mount\n");
567
568 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
569 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
570
571 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
572 }
573
574 /**
575 * gfs2_jindex_hold - Grab a lock on the jindex
576 * @sdp: The GFS2 superblock
577 * @ji_gh: the holder for the jindex glock
578 *
579 * Returns: errno
580 */
581
gfs2_jindex_hold(struct gfs2_sbd * sdp,struct gfs2_holder * ji_gh)582 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
583 {
584 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
585 struct qstr name;
586 char buf[20];
587 struct gfs2_jdesc *jd;
588 int error;
589
590 name.name = buf;
591
592 mutex_lock(&sdp->sd_jindex_mutex);
593
594 for (;;) {
595 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
596 if (error)
597 break;
598
599 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
600 name.hash = gfs2_disk_hash(name.name, name.len);
601
602 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
603 if (error == -ENOENT) {
604 error = 0;
605 break;
606 }
607
608 gfs2_glock_dq_uninit(ji_gh);
609
610 if (error)
611 break;
612
613 error = -ENOMEM;
614 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
615 if (!jd)
616 break;
617
618 INIT_LIST_HEAD(&jd->extent_list);
619 INIT_WORK(&jd->jd_work, gfs2_recover_func);
620 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
621 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
622 if (!jd->jd_inode)
623 error = -ENOENT;
624 else
625 error = PTR_ERR(jd->jd_inode);
626 kfree(jd);
627 break;
628 }
629
630 spin_lock(&sdp->sd_jindex_spin);
631 jd->jd_jid = sdp->sd_journals++;
632 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
633 spin_unlock(&sdp->sd_jindex_spin);
634 }
635
636 mutex_unlock(&sdp->sd_jindex_mutex);
637
638 return error;
639 }
640
init_journal(struct gfs2_sbd * sdp,int undo)641 static int init_journal(struct gfs2_sbd *sdp, int undo)
642 {
643 struct inode *master = sdp->sd_master_dir->d_inode;
644 struct gfs2_holder ji_gh;
645 struct gfs2_inode *ip;
646 int jindex = 1;
647 int error = 0;
648
649 if (undo) {
650 jindex = 0;
651 goto fail_jinode_gh;
652 }
653
654 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
655 if (IS_ERR(sdp->sd_jindex)) {
656 fs_err(sdp, "can't lookup journal index: %d\n", error);
657 return PTR_ERR(sdp->sd_jindex);
658 }
659
660 /* Load in the journal index special file */
661
662 error = gfs2_jindex_hold(sdp, &ji_gh);
663 if (error) {
664 fs_err(sdp, "can't read journal index: %d\n", error);
665 goto fail;
666 }
667
668 error = -EUSERS;
669 if (!gfs2_jindex_size(sdp)) {
670 fs_err(sdp, "no journals!\n");
671 goto fail_jindex;
672 }
673
674 if (sdp->sd_args.ar_spectator) {
675 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
676 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
677 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
678 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
679 } else {
680 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
681 fs_err(sdp, "can't mount journal #%u\n",
682 sdp->sd_lockstruct.ls_jid);
683 fs_err(sdp, "there are only %u journals (0 - %u)\n",
684 gfs2_jindex_size(sdp),
685 gfs2_jindex_size(sdp) - 1);
686 goto fail_jindex;
687 }
688 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
689
690 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
691 &gfs2_journal_glops,
692 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
693 &sdp->sd_journal_gh);
694 if (error) {
695 fs_err(sdp, "can't acquire journal glock: %d\n", error);
696 goto fail_jindex;
697 }
698
699 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
700 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
701 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
702 &sdp->sd_jinode_gh);
703 if (error) {
704 fs_err(sdp, "can't acquire journal inode glock: %d\n",
705 error);
706 goto fail_journal_gh;
707 }
708
709 error = gfs2_jdesc_check(sdp->sd_jdesc);
710 if (error) {
711 fs_err(sdp, "my journal (%u) is bad: %d\n",
712 sdp->sd_jdesc->jd_jid, error);
713 goto fail_jinode_gh;
714 }
715 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
716 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
717 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
718
719 /* Map the extents for this journal's blocks */
720 map_journal_extents(sdp);
721 }
722 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
723
724 if (sdp->sd_lockstruct.ls_first) {
725 unsigned int x;
726 for (x = 0; x < sdp->sd_journals; x++) {
727 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
728 true);
729 if (error) {
730 fs_err(sdp, "error recovering journal %u: %d\n",
731 x, error);
732 goto fail_jinode_gh;
733 }
734 }
735
736 gfs2_others_may_mount(sdp);
737 } else if (!sdp->sd_args.ar_spectator) {
738 error = gfs2_recover_journal(sdp->sd_jdesc, true);
739 if (error) {
740 fs_err(sdp, "error recovering my journal: %d\n", error);
741 goto fail_jinode_gh;
742 }
743 }
744
745 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
746 gfs2_glock_dq_uninit(&ji_gh);
747 jindex = 0;
748
749 return 0;
750
751 fail_jinode_gh:
752 if (!sdp->sd_args.ar_spectator)
753 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
754 fail_journal_gh:
755 if (!sdp->sd_args.ar_spectator)
756 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
757 fail_jindex:
758 gfs2_jindex_free(sdp);
759 if (jindex)
760 gfs2_glock_dq_uninit(&ji_gh);
761 fail:
762 iput(sdp->sd_jindex);
763 return error;
764 }
765
766
init_inodes(struct gfs2_sbd * sdp,int undo)767 static int init_inodes(struct gfs2_sbd *sdp, int undo)
768 {
769 int error = 0;
770 struct inode *master = sdp->sd_master_dir->d_inode;
771
772 if (undo)
773 goto fail_qinode;
774
775 error = init_journal(sdp, undo);
776 if (error)
777 goto fail;
778
779 /* Read in the master statfs inode */
780 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
781 if (IS_ERR(sdp->sd_statfs_inode)) {
782 error = PTR_ERR(sdp->sd_statfs_inode);
783 fs_err(sdp, "can't read in statfs inode: %d\n", error);
784 goto fail_journal;
785 }
786
787 /* Read in the resource index inode */
788 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
789 if (IS_ERR(sdp->sd_rindex)) {
790 error = PTR_ERR(sdp->sd_rindex);
791 fs_err(sdp, "can't get resource index inode: %d\n", error);
792 goto fail_statfs;
793 }
794 sdp->sd_rindex_uptodate = 0;
795
796 /* Read in the quota inode */
797 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
798 if (IS_ERR(sdp->sd_quota_inode)) {
799 error = PTR_ERR(sdp->sd_quota_inode);
800 fs_err(sdp, "can't get quota file inode: %d\n", error);
801 goto fail_rindex;
802 }
803
804 error = gfs2_rindex_update(sdp);
805 if (error)
806 goto fail_qinode;
807
808 return 0;
809
810 fail_qinode:
811 iput(sdp->sd_quota_inode);
812 fail_rindex:
813 gfs2_clear_rgrpd(sdp);
814 iput(sdp->sd_rindex);
815 fail_statfs:
816 iput(sdp->sd_statfs_inode);
817 fail_journal:
818 init_journal(sdp, UNDO);
819 fail:
820 return error;
821 }
822
init_per_node(struct gfs2_sbd * sdp,int undo)823 static int init_per_node(struct gfs2_sbd *sdp, int undo)
824 {
825 struct inode *pn = NULL;
826 char buf[30];
827 int error = 0;
828 struct gfs2_inode *ip;
829 struct inode *master = sdp->sd_master_dir->d_inode;
830
831 if (sdp->sd_args.ar_spectator)
832 return 0;
833
834 if (undo)
835 goto fail_qc_gh;
836
837 pn = gfs2_lookup_simple(master, "per_node");
838 if (IS_ERR(pn)) {
839 error = PTR_ERR(pn);
840 fs_err(sdp, "can't find per_node directory: %d\n", error);
841 return error;
842 }
843
844 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
845 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
846 if (IS_ERR(sdp->sd_sc_inode)) {
847 error = PTR_ERR(sdp->sd_sc_inode);
848 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
849 goto fail;
850 }
851
852 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
853 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
854 if (IS_ERR(sdp->sd_qc_inode)) {
855 error = PTR_ERR(sdp->sd_qc_inode);
856 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
857 goto fail_ut_i;
858 }
859
860 iput(pn);
861 pn = NULL;
862
863 ip = GFS2_I(sdp->sd_sc_inode);
864 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
865 &sdp->sd_sc_gh);
866 if (error) {
867 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
868 goto fail_qc_i;
869 }
870
871 ip = GFS2_I(sdp->sd_qc_inode);
872 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
873 &sdp->sd_qc_gh);
874 if (error) {
875 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
876 goto fail_ut_gh;
877 }
878
879 return 0;
880
881 fail_qc_gh:
882 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
883 fail_ut_gh:
884 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
885 fail_qc_i:
886 iput(sdp->sd_qc_inode);
887 fail_ut_i:
888 iput(sdp->sd_sc_inode);
889 fail:
890 if (pn)
891 iput(pn);
892 return error;
893 }
894
init_threads(struct gfs2_sbd * sdp,int undo)895 static int init_threads(struct gfs2_sbd *sdp, int undo)
896 {
897 struct task_struct *p;
898 int error = 0;
899
900 if (undo)
901 goto fail_quotad;
902
903 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
904 error = IS_ERR(p);
905 if (error) {
906 fs_err(sdp, "can't start logd thread: %d\n", error);
907 return error;
908 }
909 sdp->sd_logd_process = p;
910
911 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
912 error = IS_ERR(p);
913 if (error) {
914 fs_err(sdp, "can't start quotad thread: %d\n", error);
915 goto fail;
916 }
917 sdp->sd_quotad_process = p;
918
919 return 0;
920
921
922 fail_quotad:
923 kthread_stop(sdp->sd_quotad_process);
924 fail:
925 kthread_stop(sdp->sd_logd_process);
926 return error;
927 }
928
929 static const match_table_t nolock_tokens = {
930 { Opt_jid, "jid=%d\n", },
931 { Opt_err, NULL },
932 };
933
934 static const struct lm_lockops nolock_ops = {
935 .lm_proto_name = "lock_nolock",
936 .lm_put_lock = gfs2_glock_free,
937 .lm_tokens = &nolock_tokens,
938 };
939
940 /**
941 * gfs2_lm_mount - mount a locking protocol
942 * @sdp: the filesystem
943 * @args: mount arguments
944 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
945 *
946 * Returns: errno
947 */
948
gfs2_lm_mount(struct gfs2_sbd * sdp,int silent)949 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
950 {
951 const struct lm_lockops *lm;
952 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
953 struct gfs2_args *args = &sdp->sd_args;
954 const char *proto = sdp->sd_proto_name;
955 const char *table = sdp->sd_table_name;
956 char *o, *options;
957 int ret;
958
959 if (!strcmp("lock_nolock", proto)) {
960 lm = &nolock_ops;
961 sdp->sd_args.ar_localflocks = 1;
962 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
963 } else if (!strcmp("lock_dlm", proto)) {
964 lm = &gfs2_dlm_ops;
965 #endif
966 } else {
967 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
968 return -ENOENT;
969 }
970
971 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
972
973 ls->ls_ops = lm;
974 ls->ls_first = 1;
975
976 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
977 substring_t tmp[MAX_OPT_ARGS];
978 int token, option;
979
980 if (!o || !*o)
981 continue;
982
983 token = match_token(o, *lm->lm_tokens, tmp);
984 switch (token) {
985 case Opt_jid:
986 ret = match_int(&tmp[0], &option);
987 if (ret || option < 0)
988 goto hostdata_error;
989 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
990 ls->ls_jid = option;
991 break;
992 case Opt_id:
993 /* Obsolete, but left for backward compat purposes */
994 break;
995 case Opt_first:
996 ret = match_int(&tmp[0], &option);
997 if (ret || (option != 0 && option != 1))
998 goto hostdata_error;
999 ls->ls_first = option;
1000 break;
1001 case Opt_nodir:
1002 ret = match_int(&tmp[0], &option);
1003 if (ret || (option != 0 && option != 1))
1004 goto hostdata_error;
1005 ls->ls_nodir = option;
1006 break;
1007 case Opt_err:
1008 default:
1009 hostdata_error:
1010 fs_info(sdp, "unknown hostdata (%s)\n", o);
1011 return -EINVAL;
1012 }
1013 }
1014
1015 if (lm->lm_mount == NULL) {
1016 fs_info(sdp, "Now mounting FS...\n");
1017 complete_all(&sdp->sd_locking_init);
1018 return 0;
1019 }
1020 ret = lm->lm_mount(sdp, table);
1021 if (ret == 0)
1022 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1023 complete_all(&sdp->sd_locking_init);
1024 return ret;
1025 }
1026
gfs2_lm_unmount(struct gfs2_sbd * sdp)1027 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1028 {
1029 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1030 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1031 lm->lm_unmount)
1032 lm->lm_unmount(sdp);
1033 }
1034
gfs2_journalid_wait(void * word)1035 static int gfs2_journalid_wait(void *word)
1036 {
1037 if (signal_pending(current))
1038 return -EINTR;
1039 schedule();
1040 return 0;
1041 }
1042
wait_on_journal(struct gfs2_sbd * sdp)1043 static int wait_on_journal(struct gfs2_sbd *sdp)
1044 {
1045 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1046 return 0;
1047
1048 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1049 }
1050
gfs2_online_uevent(struct gfs2_sbd * sdp)1051 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1052 {
1053 struct super_block *sb = sdp->sd_vfs;
1054 char ro[20];
1055 char spectator[20];
1056 char *envp[] = { ro, spectator, NULL };
1057 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1058 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1059 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1060 }
1061
1062 /**
1063 * fill_super - Read in superblock
1064 * @sb: The VFS superblock
1065 * @data: Mount options
1066 * @silent: Don't complain if it's not a GFS2 filesystem
1067 *
1068 * Returns: errno
1069 */
1070
fill_super(struct super_block * sb,struct gfs2_args * args,int silent)1071 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1072 {
1073 struct gfs2_sbd *sdp;
1074 struct gfs2_holder mount_gh;
1075 int error;
1076
1077 sdp = init_sbd(sb);
1078 if (!sdp) {
1079 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1080 return -ENOMEM;
1081 }
1082 sdp->sd_args = *args;
1083
1084 if (sdp->sd_args.ar_spectator) {
1085 sb->s_flags |= MS_RDONLY;
1086 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1087 }
1088 if (sdp->sd_args.ar_posix_acl)
1089 sb->s_flags |= MS_POSIXACL;
1090 if (sdp->sd_args.ar_nobarrier)
1091 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1092
1093 sb->s_flags |= MS_NOSEC;
1094 sb->s_magic = GFS2_MAGIC;
1095 sb->s_op = &gfs2_super_ops;
1096 sb->s_d_op = &gfs2_dops;
1097 sb->s_export_op = &gfs2_export_ops;
1098 sb->s_xattr = gfs2_xattr_handlers;
1099 sb->s_qcop = &gfs2_quotactl_ops;
1100 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1101 sb->s_time_gran = 1;
1102 sb->s_maxbytes = MAX_LFS_FILESIZE;
1103
1104 /* Set up the buffer cache and fill in some fake block size values
1105 to allow us to read-in the on-disk superblock. */
1106 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1107 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1108 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1109 GFS2_BASIC_BLOCK_SHIFT;
1110 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1111
1112 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1113 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1114 if (sdp->sd_args.ar_statfs_quantum) {
1115 sdp->sd_tune.gt_statfs_slow = 0;
1116 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1117 } else {
1118 sdp->sd_tune.gt_statfs_slow = 1;
1119 sdp->sd_tune.gt_statfs_quantum = 30;
1120 }
1121
1122 error = init_names(sdp, silent);
1123 if (error)
1124 goto fail;
1125
1126 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1127
1128 gfs2_create_debugfs_file(sdp);
1129
1130 error = gfs2_sys_fs_add(sdp);
1131 if (error)
1132 goto fail;
1133
1134 error = gfs2_lm_mount(sdp, silent);
1135 if (error)
1136 goto fail_sys;
1137
1138 error = init_locking(sdp, &mount_gh, DO);
1139 if (error)
1140 goto fail_lm;
1141
1142 error = init_sb(sdp, silent);
1143 if (error)
1144 goto fail_locking;
1145
1146 error = wait_on_journal(sdp);
1147 if (error)
1148 goto fail_sb;
1149
1150 /*
1151 * If user space has failed to join the cluster or some similar
1152 * failure has occurred, then the journal id will contain a
1153 * negative (error) number. This will then be returned to the
1154 * caller (of the mount syscall). We do this even for spectator
1155 * mounts (which just write a jid of 0 to indicate "ok" even though
1156 * the jid is unused in the spectator case)
1157 */
1158 if (sdp->sd_lockstruct.ls_jid < 0) {
1159 error = sdp->sd_lockstruct.ls_jid;
1160 sdp->sd_lockstruct.ls_jid = 0;
1161 goto fail_sb;
1162 }
1163
1164 if (sdp->sd_args.ar_spectator)
1165 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1166 sdp->sd_table_name);
1167 else
1168 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1169 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1170
1171 error = init_inodes(sdp, DO);
1172 if (error)
1173 goto fail_sb;
1174
1175 error = init_per_node(sdp, DO);
1176 if (error)
1177 goto fail_inodes;
1178
1179 error = gfs2_statfs_init(sdp);
1180 if (error) {
1181 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1182 goto fail_per_node;
1183 }
1184
1185 error = init_threads(sdp, DO);
1186 if (error)
1187 goto fail_per_node;
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_threads;
1194 }
1195 }
1196
1197 gfs2_glock_dq_uninit(&mount_gh);
1198 gfs2_online_uevent(sdp);
1199 return 0;
1200
1201 fail_threads:
1202 init_threads(sdp, UNDO);
1203 fail_per_node:
1204 init_per_node(sdp, UNDO);
1205 fail_inodes:
1206 init_inodes(sdp, UNDO);
1207 fail_sb:
1208 if (sdp->sd_root_dir)
1209 dput(sdp->sd_root_dir);
1210 if (sdp->sd_master_dir)
1211 dput(sdp->sd_master_dir);
1212 if (sb->s_root)
1213 dput(sb->s_root);
1214 sb->s_root = NULL;
1215 fail_locking:
1216 init_locking(sdp, &mount_gh, UNDO);
1217 fail_lm:
1218 gfs2_gl_hash_clear(sdp);
1219 gfs2_lm_unmount(sdp);
1220 fail_sys:
1221 gfs2_sys_fs_del(sdp);
1222 fail:
1223 gfs2_delete_debugfs_file(sdp);
1224 kfree(sdp);
1225 sb->s_fs_info = NULL;
1226 return error;
1227 }
1228
set_gfs2_super(struct super_block * s,void * data)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
test_gfs2_super(struct super_block * s,void * ptr)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
gfs2_mount(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)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, 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 blkdev_put(bdev, mode);
1298
1299 memset(&args, 0, sizeof(args));
1300 args.ar_quota = GFS2_QUOTA_DEFAULT;
1301 args.ar_data = GFS2_DATA_DEFAULT;
1302 args.ar_commit = 30;
1303 args.ar_statfs_quantum = 30;
1304 args.ar_quota_quantum = 60;
1305 args.ar_errors = GFS2_ERRORS_DEFAULT;
1306
1307 error = gfs2_mount_args(&args, data);
1308 if (error) {
1309 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1310 goto error_super;
1311 }
1312
1313 if (s->s_root) {
1314 error = -EBUSY;
1315 if ((flags ^ s->s_flags) & MS_RDONLY)
1316 goto error_super;
1317 } else {
1318 char b[BDEVNAME_SIZE];
1319
1320 s->s_flags = flags;
1321 s->s_mode = mode;
1322 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1323 sb_set_blocksize(s, block_size(bdev));
1324 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1325 if (error)
1326 goto error_super;
1327 s->s_flags |= MS_ACTIVE;
1328 bdev->bd_super = s;
1329 }
1330
1331 sdp = s->s_fs_info;
1332 if (args.ar_meta)
1333 return dget(sdp->sd_master_dir);
1334 else
1335 return dget(sdp->sd_root_dir);
1336
1337 error_super:
1338 deactivate_locked_super(s);
1339 return ERR_PTR(error);
1340 error_bdev:
1341 blkdev_put(bdev, mode);
1342 return ERR_PTR(error);
1343 }
1344
set_meta_super(struct super_block * s,void * ptr)1345 static int set_meta_super(struct super_block *s, void *ptr)
1346 {
1347 return -EINVAL;
1348 }
1349
gfs2_mount_meta(struct file_system_type * fs_type,int flags,const char * dev_name,void * data)1350 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1351 int flags, const char *dev_name, void *data)
1352 {
1353 struct super_block *s;
1354 struct gfs2_sbd *sdp;
1355 struct path path;
1356 int error;
1357
1358 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1359 if (error) {
1360 printk(KERN_WARNING "GFS2: 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,
1365 path.dentry->d_inode->i_sb->s_bdev);
1366 path_put(&path);
1367 if (IS_ERR(s)) {
1368 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1369 return ERR_CAST(s);
1370 }
1371 if ((flags ^ s->s_flags) & MS_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_meta_syncfs(sdp);
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 kill_block_super(sb);
1395 gfs2_delete_debugfs_file(sdp);
1396 kfree(sdp);
1397 }
1398
1399 struct file_system_type gfs2_fs_type = {
1400 .name = "gfs2",
1401 .fs_flags = FS_REQUIRES_DEV,
1402 .mount = gfs2_mount,
1403 .kill_sb = gfs2_kill_sb,
1404 .owner = THIS_MODULE,
1405 };
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
1414