Loading...
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
48static 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
61static 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 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
72 if (!sdp->sd_lkstats) {
73 kfree(sdp);
74 return NULL;
75 }
76
77 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
78 gfs2_tune_init(&sdp->sd_tune);
79
80 init_waitqueue_head(&sdp->sd_glock_wait);
81 atomic_set(&sdp->sd_glock_disposal, 0);
82 init_completion(&sdp->sd_locking_init);
83 spin_lock_init(&sdp->sd_statfs_spin);
84
85 spin_lock_init(&sdp->sd_rindex_spin);
86 sdp->sd_rindex_tree.rb_node = NULL;
87
88 INIT_LIST_HEAD(&sdp->sd_jindex_list);
89 spin_lock_init(&sdp->sd_jindex_spin);
90 mutex_init(&sdp->sd_jindex_mutex);
91
92 INIT_LIST_HEAD(&sdp->sd_quota_list);
93 mutex_init(&sdp->sd_quota_mutex);
94 init_waitqueue_head(&sdp->sd_quota_wait);
95 INIT_LIST_HEAD(&sdp->sd_trunc_list);
96 spin_lock_init(&sdp->sd_trunc_lock);
97
98 spin_lock_init(&sdp->sd_log_lock);
99 atomic_set(&sdp->sd_log_pinned, 0);
100 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
101 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
102 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
103 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
104
105 init_waitqueue_head(&sdp->sd_log_waitq);
106 init_waitqueue_head(&sdp->sd_logd_waitq);
107 spin_lock_init(&sdp->sd_ail_lock);
108 INIT_LIST_HEAD(&sdp->sd_ail1_list);
109 INIT_LIST_HEAD(&sdp->sd_ail2_list);
110
111 init_rwsem(&sdp->sd_log_flush_lock);
112 atomic_set(&sdp->sd_log_in_flight, 0);
113 init_waitqueue_head(&sdp->sd_log_flush_wait);
114
115 INIT_LIST_HEAD(&sdp->sd_revoke_list);
116
117 mutex_init(&sdp->sd_freeze_lock);
118
119 return sdp;
120}
121
122
123/**
124 * gfs2_check_sb - Check superblock
125 * @sdp: the filesystem
126 * @sb: The superblock
127 * @silent: Don't print a message if the check fails
128 *
129 * Checks the version code of the FS is one that we understand how to
130 * read and that the sizes of the various on-disk structures have not
131 * changed.
132 */
133
134static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
135{
136 struct gfs2_sb_host *sb = &sdp->sd_sb;
137
138 if (sb->sb_magic != GFS2_MAGIC ||
139 sb->sb_type != GFS2_METATYPE_SB) {
140 if (!silent)
141 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
142 return -EINVAL;
143 }
144
145 /* If format numbers match exactly, we're done. */
146
147 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
148 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
149 return 0;
150
151 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
152
153 return -EINVAL;
154}
155
156static void end_bio_io_page(struct bio *bio, int error)
157{
158 struct page *page = bio->bi_private;
159
160 if (!error)
161 SetPageUptodate(page);
162 else
163 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
164 unlock_page(page);
165}
166
167static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
168{
169 struct gfs2_sb_host *sb = &sdp->sd_sb;
170 struct super_block *s = sdp->sd_vfs;
171 const struct gfs2_sb *str = buf;
172
173 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
174 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
175 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
176 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
177 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
178 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
179 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
180 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
181 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
182 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
183 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
184
185 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
186 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
187 memcpy(s->s_uuid, str->sb_uuid, 16);
188}
189
190/**
191 * gfs2_read_super - Read the gfs2 super block from disk
192 * @sdp: The GFS2 super block
193 * @sector: The location of the super block
194 * @error: The error code to return
195 *
196 * This uses the bio functions to read the super block from disk
197 * because we want to be 100% sure that we never read cached data.
198 * A super block is read twice only during each GFS2 mount and is
199 * never written to by the filesystem. The first time its read no
200 * locks are held, and the only details which are looked at are those
201 * relating to the locking protocol. Once locking is up and working,
202 * the sb is read again under the lock to establish the location of
203 * the master directory (contains pointers to journals etc) and the
204 * root directory.
205 *
206 * Returns: 0 on success or error
207 */
208
209static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
210{
211 struct super_block *sb = sdp->sd_vfs;
212 struct gfs2_sb *p;
213 struct page *page;
214 struct bio *bio;
215
216 page = alloc_page(GFP_NOFS);
217 if (unlikely(!page))
218 return -ENOBUFS;
219
220 ClearPageUptodate(page);
221 ClearPageDirty(page);
222 lock_page(page);
223
224 bio = bio_alloc(GFP_NOFS, 1);
225 bio->bi_sector = sector * (sb->s_blocksize >> 9);
226 bio->bi_bdev = sb->s_bdev;
227 bio_add_page(bio, page, PAGE_SIZE, 0);
228
229 bio->bi_end_io = end_bio_io_page;
230 bio->bi_private = page;
231 submit_bio(READ_SYNC | REQ_META, bio);
232 wait_on_page_locked(page);
233 bio_put(bio);
234 if (!PageUptodate(page)) {
235 __free_page(page);
236 return -EIO;
237 }
238 p = kmap(page);
239 gfs2_sb_in(sdp, p);
240 kunmap(page);
241 __free_page(page);
242 return gfs2_check_sb(sdp, silent);
243}
244
245/**
246 * gfs2_read_sb - Read super block
247 * @sdp: The GFS2 superblock
248 * @silent: Don't print message if mount fails
249 *
250 */
251
252static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
253{
254 u32 hash_blocks, ind_blocks, leaf_blocks;
255 u32 tmp_blocks;
256 unsigned int x;
257 int error;
258
259 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
260 if (error) {
261 if (!silent)
262 fs_err(sdp, "can't read superblock\n");
263 return error;
264 }
265
266 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
267 GFS2_BASIC_BLOCK_SHIFT;
268 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
269 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
270 sizeof(struct gfs2_dinode)) / sizeof(u64);
271 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
272 sizeof(struct gfs2_meta_header)) / sizeof(u64);
273 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
274 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
275 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
276 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
277 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
278 sizeof(struct gfs2_meta_header)) /
279 sizeof(struct gfs2_quota_change);
280
281 /* Compute maximum reservation required to add a entry to a directory */
282
283 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
284 sdp->sd_jbsize);
285
286 ind_blocks = 0;
287 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
288 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
289 ind_blocks += tmp_blocks;
290 }
291
292 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
293
294 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
295
296 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
297 sizeof(struct gfs2_dinode);
298 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
299 for (x = 2;; x++) {
300 u64 space, d;
301 u32 m;
302
303 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
304 d = space;
305 m = do_div(d, sdp->sd_inptrs);
306
307 if (d != sdp->sd_heightsize[x - 1] || m)
308 break;
309 sdp->sd_heightsize[x] = space;
310 }
311 sdp->sd_max_height = x;
312 sdp->sd_heightsize[x] = ~0;
313 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
314
315 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
316 sizeof(struct gfs2_dinode);
317 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
318 for (x = 2;; x++) {
319 u64 space, d;
320 u32 m;
321
322 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
323 d = space;
324 m = do_div(d, sdp->sd_inptrs);
325
326 if (d != sdp->sd_jheightsize[x - 1] || m)
327 break;
328 sdp->sd_jheightsize[x] = space;
329 }
330 sdp->sd_max_jheight = x;
331 sdp->sd_jheightsize[x] = ~0;
332 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
333
334 return 0;
335}
336
337static int init_names(struct gfs2_sbd *sdp, int silent)
338{
339 char *proto, *table;
340 int error = 0;
341
342 proto = sdp->sd_args.ar_lockproto;
343 table = sdp->sd_args.ar_locktable;
344
345 /* Try to autodetect */
346
347 if (!proto[0] || !table[0]) {
348 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
349 if (error)
350 return error;
351
352 if (!proto[0])
353 proto = sdp->sd_sb.sb_lockproto;
354 if (!table[0])
355 table = sdp->sd_sb.sb_locktable;
356 }
357
358 if (!table[0])
359 table = sdp->sd_vfs->s_id;
360
361 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
362 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
363
364 table = sdp->sd_table_name;
365 while ((table = strchr(table, '/')))
366 *table = '_';
367
368 return error;
369}
370
371static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
372 int undo)
373{
374 int error = 0;
375
376 if (undo)
377 goto fail_trans;
378
379 error = gfs2_glock_nq_num(sdp,
380 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
381 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
382 mount_gh);
383 if (error) {
384 fs_err(sdp, "can't acquire mount glock: %d\n", error);
385 goto fail;
386 }
387
388 error = gfs2_glock_nq_num(sdp,
389 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
390 LM_ST_SHARED,
391 LM_FLAG_NOEXP | GL_EXACT,
392 &sdp->sd_live_gh);
393 if (error) {
394 fs_err(sdp, "can't acquire live glock: %d\n", error);
395 goto fail_mount;
396 }
397
398 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
399 CREATE, &sdp->sd_rename_gl);
400 if (error) {
401 fs_err(sdp, "can't create rename glock: %d\n", error);
402 goto fail_live;
403 }
404
405 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
406 CREATE, &sdp->sd_trans_gl);
407 if (error) {
408 fs_err(sdp, "can't create transaction glock: %d\n", error);
409 goto fail_rename;
410 }
411
412 return 0;
413
414fail_trans:
415 gfs2_glock_put(sdp->sd_trans_gl);
416fail_rename:
417 gfs2_glock_put(sdp->sd_rename_gl);
418fail_live:
419 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
420fail_mount:
421 gfs2_glock_dq_uninit(mount_gh);
422fail:
423 return error;
424}
425
426static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
427 u64 no_addr, const char *name)
428{
429 struct gfs2_sbd *sdp = sb->s_fs_info;
430 struct dentry *dentry;
431 struct inode *inode;
432
433 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
434 if (IS_ERR(inode)) {
435 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
436 return PTR_ERR(inode);
437 }
438 dentry = d_make_root(inode);
439 if (!dentry) {
440 fs_err(sdp, "can't alloc %s dentry\n", name);
441 return -ENOMEM;
442 }
443 *dptr = dentry;
444 return 0;
445}
446
447static int init_sb(struct gfs2_sbd *sdp, int silent)
448{
449 struct super_block *sb = sdp->sd_vfs;
450 struct gfs2_holder sb_gh;
451 u64 no_addr;
452 int ret;
453
454 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
455 LM_ST_SHARED, 0, &sb_gh);
456 if (ret) {
457 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
458 return ret;
459 }
460
461 ret = gfs2_read_sb(sdp, silent);
462 if (ret) {
463 fs_err(sdp, "can't read superblock: %d\n", ret);
464 goto out;
465 }
466
467 /* Set up the buffer cache and SB for real */
468 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
469 ret = -EINVAL;
470 fs_err(sdp, "FS block size (%u) is too small for device "
471 "block size (%u)\n",
472 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
473 goto out;
474 }
475 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
476 ret = -EINVAL;
477 fs_err(sdp, "FS block size (%u) is too big for machine "
478 "page size (%u)\n",
479 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
480 goto out;
481 }
482 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
483
484 /* Get the root inode */
485 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
486 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
487 if (ret)
488 goto out;
489
490 /* Get the master inode */
491 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
492 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
493 if (ret) {
494 dput(sdp->sd_root_dir);
495 goto out;
496 }
497 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
498out:
499 gfs2_glock_dq_uninit(&sb_gh);
500 return ret;
501}
502
503/**
504 * map_journal_extents - create a reusable "extent" mapping from all logical
505 * blocks to all physical blocks for the given journal. This will save
506 * us time when writing journal blocks. Most journals will have only one
507 * extent that maps all their logical blocks. That's because gfs2.mkfs
508 * arranges the journal blocks sequentially to maximize performance.
509 * So the extent would map the first block for the entire file length.
510 * However, gfs2_jadd can happen while file activity is happening, so
511 * those journals may not be sequential. Less likely is the case where
512 * the users created their own journals by mounting the metafs and
513 * laying it out. But it's still possible. These journals might have
514 * several extents.
515 *
516 * TODO: This should be done in bigger chunks rather than one block at a time,
517 * but since it's only done at mount time, I'm not worried about the
518 * time it takes.
519 */
520static int map_journal_extents(struct gfs2_sbd *sdp)
521{
522 struct gfs2_jdesc *jd = sdp->sd_jdesc;
523 unsigned int lb;
524 u64 db, prev_db; /* logical block, disk block, prev disk block */
525 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
526 struct gfs2_journal_extent *jext = NULL;
527 struct buffer_head bh;
528 int rc = 0;
529
530 prev_db = 0;
531
532 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
533 bh.b_state = 0;
534 bh.b_blocknr = 0;
535 bh.b_size = 1 << ip->i_inode.i_blkbits;
536 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
537 db = bh.b_blocknr;
538 if (rc || !db) {
539 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
540 "%u db=%llu\n", rc, lb, (unsigned long long)db);
541 break;
542 }
543 if (!prev_db || db != prev_db + 1) {
544 jext = kzalloc(sizeof(struct gfs2_journal_extent),
545 GFP_KERNEL);
546 if (!jext) {
547 printk(KERN_INFO "GFS2 error: out of memory "
548 "mapping journal extents.\n");
549 rc = -ENOMEM;
550 break;
551 }
552 jext->dblock = db;
553 jext->lblock = lb;
554 jext->blocks = 1;
555 list_add_tail(&jext->extent_list, &jd->extent_list);
556 } else {
557 jext->blocks++;
558 }
559 prev_db = db;
560 }
561 return rc;
562}
563
564static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
565{
566 char *message = "FIRSTMOUNT=Done";
567 char *envp[] = { message, NULL };
568
569 fs_info(sdp, "first mount done, others may mount\n");
570
571 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
572 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
573
574 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
575}
576
577/**
578 * gfs2_jindex_hold - Grab a lock on the jindex
579 * @sdp: The GFS2 superblock
580 * @ji_gh: the holder for the jindex glock
581 *
582 * Returns: errno
583 */
584
585static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
586{
587 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
588 struct qstr name;
589 char buf[20];
590 struct gfs2_jdesc *jd;
591 int error;
592
593 name.name = buf;
594
595 mutex_lock(&sdp->sd_jindex_mutex);
596
597 for (;;) {
598 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
599 if (error)
600 break;
601
602 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
603 name.hash = gfs2_disk_hash(name.name, name.len);
604
605 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
606 if (error == -ENOENT) {
607 error = 0;
608 break;
609 }
610
611 gfs2_glock_dq_uninit(ji_gh);
612
613 if (error)
614 break;
615
616 error = -ENOMEM;
617 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
618 if (!jd)
619 break;
620
621 INIT_LIST_HEAD(&jd->extent_list);
622 INIT_WORK(&jd->jd_work, gfs2_recover_func);
623 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
624 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
625 if (!jd->jd_inode)
626 error = -ENOENT;
627 else
628 error = PTR_ERR(jd->jd_inode);
629 kfree(jd);
630 break;
631 }
632
633 spin_lock(&sdp->sd_jindex_spin);
634 jd->jd_jid = sdp->sd_journals++;
635 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
636 spin_unlock(&sdp->sd_jindex_spin);
637 }
638
639 mutex_unlock(&sdp->sd_jindex_mutex);
640
641 return error;
642}
643
644static int init_journal(struct gfs2_sbd *sdp, int undo)
645{
646 struct inode *master = sdp->sd_master_dir->d_inode;
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 if (sdp->sd_args.ar_spectator) {
678 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
679 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
680 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
681 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
682 } else {
683 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
684 fs_err(sdp, "can't mount journal #%u\n",
685 sdp->sd_lockstruct.ls_jid);
686 fs_err(sdp, "there are only %u journals (0 - %u)\n",
687 gfs2_jindex_size(sdp),
688 gfs2_jindex_size(sdp) - 1);
689 goto fail_jindex;
690 }
691 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
692
693 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
694 &gfs2_journal_glops,
695 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
696 &sdp->sd_journal_gh);
697 if (error) {
698 fs_err(sdp, "can't acquire journal glock: %d\n", error);
699 goto fail_jindex;
700 }
701
702 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
703 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
704 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
705 &sdp->sd_jinode_gh);
706 if (error) {
707 fs_err(sdp, "can't acquire journal inode glock: %d\n",
708 error);
709 goto fail_journal_gh;
710 }
711
712 error = gfs2_jdesc_check(sdp->sd_jdesc);
713 if (error) {
714 fs_err(sdp, "my journal (%u) is bad: %d\n",
715 sdp->sd_jdesc->jd_jid, error);
716 goto fail_jinode_gh;
717 }
718 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
719 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
720 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
721
722 /* Map the extents for this journal's blocks */
723 map_journal_extents(sdp);
724 }
725 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
726
727 if (sdp->sd_lockstruct.ls_first) {
728 unsigned int x;
729 for (x = 0; x < sdp->sd_journals; x++) {
730 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
731 true);
732 if (error) {
733 fs_err(sdp, "error recovering journal %u: %d\n",
734 x, error);
735 goto fail_jinode_gh;
736 }
737 }
738
739 gfs2_others_may_mount(sdp);
740 } else if (!sdp->sd_args.ar_spectator) {
741 error = gfs2_recover_journal(sdp->sd_jdesc, true);
742 if (error) {
743 fs_err(sdp, "error recovering my journal: %d\n", error);
744 goto fail_jinode_gh;
745 }
746 }
747
748 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
749 gfs2_glock_dq_uninit(&ji_gh);
750 jindex = 0;
751
752 return 0;
753
754fail_jinode_gh:
755 if (!sdp->sd_args.ar_spectator)
756 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
757fail_journal_gh:
758 if (!sdp->sd_args.ar_spectator)
759 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
760fail_jindex:
761 gfs2_jindex_free(sdp);
762 if (jindex)
763 gfs2_glock_dq_uninit(&ji_gh);
764fail:
765 iput(sdp->sd_jindex);
766 return error;
767}
768
769
770static int init_inodes(struct gfs2_sbd *sdp, int undo)
771{
772 int error = 0;
773 struct inode *master = sdp->sd_master_dir->d_inode;
774
775 if (undo)
776 goto fail_qinode;
777
778 error = init_journal(sdp, undo);
779 if (error)
780 goto fail;
781
782 /* Read in the master statfs inode */
783 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
784 if (IS_ERR(sdp->sd_statfs_inode)) {
785 error = PTR_ERR(sdp->sd_statfs_inode);
786 fs_err(sdp, "can't read in statfs inode: %d\n", error);
787 goto fail_journal;
788 }
789
790 /* Read in the resource index inode */
791 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
792 if (IS_ERR(sdp->sd_rindex)) {
793 error = PTR_ERR(sdp->sd_rindex);
794 fs_err(sdp, "can't get resource index inode: %d\n", error);
795 goto fail_statfs;
796 }
797 sdp->sd_rindex_uptodate = 0;
798
799 /* Read in the quota inode */
800 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
801 if (IS_ERR(sdp->sd_quota_inode)) {
802 error = PTR_ERR(sdp->sd_quota_inode);
803 fs_err(sdp, "can't get quota file inode: %d\n", error);
804 goto fail_rindex;
805 }
806
807 error = gfs2_rindex_update(sdp);
808 if (error)
809 goto fail_qinode;
810
811 return 0;
812
813fail_qinode:
814 iput(sdp->sd_quota_inode);
815fail_rindex:
816 gfs2_clear_rgrpd(sdp);
817 iput(sdp->sd_rindex);
818fail_statfs:
819 iput(sdp->sd_statfs_inode);
820fail_journal:
821 init_journal(sdp, UNDO);
822fail:
823 return error;
824}
825
826static int init_per_node(struct gfs2_sbd *sdp, int undo)
827{
828 struct inode *pn = NULL;
829 char buf[30];
830 int error = 0;
831 struct gfs2_inode *ip;
832 struct inode *master = sdp->sd_master_dir->d_inode;
833
834 if (sdp->sd_args.ar_spectator)
835 return 0;
836
837 if (undo)
838 goto fail_qc_gh;
839
840 pn = gfs2_lookup_simple(master, "per_node");
841 if (IS_ERR(pn)) {
842 error = PTR_ERR(pn);
843 fs_err(sdp, "can't find per_node directory: %d\n", error);
844 return error;
845 }
846
847 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
848 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
849 if (IS_ERR(sdp->sd_sc_inode)) {
850 error = PTR_ERR(sdp->sd_sc_inode);
851 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
852 goto fail;
853 }
854
855 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
856 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
857 if (IS_ERR(sdp->sd_qc_inode)) {
858 error = PTR_ERR(sdp->sd_qc_inode);
859 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
860 goto fail_ut_i;
861 }
862
863 iput(pn);
864 pn = NULL;
865
866 ip = GFS2_I(sdp->sd_sc_inode);
867 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
868 &sdp->sd_sc_gh);
869 if (error) {
870 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
871 goto fail_qc_i;
872 }
873
874 ip = GFS2_I(sdp->sd_qc_inode);
875 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
876 &sdp->sd_qc_gh);
877 if (error) {
878 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
879 goto fail_ut_gh;
880 }
881
882 return 0;
883
884fail_qc_gh:
885 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
886fail_ut_gh:
887 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
888fail_qc_i:
889 iput(sdp->sd_qc_inode);
890fail_ut_i:
891 iput(sdp->sd_sc_inode);
892fail:
893 if (pn)
894 iput(pn);
895 return error;
896}
897
898static int init_threads(struct gfs2_sbd *sdp, int undo)
899{
900 struct task_struct *p;
901 int error = 0;
902
903 if (undo)
904 goto fail_quotad;
905
906 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
907 error = IS_ERR(p);
908 if (error) {
909 fs_err(sdp, "can't start logd thread: %d\n", error);
910 return error;
911 }
912 sdp->sd_logd_process = p;
913
914 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
915 error = IS_ERR(p);
916 if (error) {
917 fs_err(sdp, "can't start quotad thread: %d\n", error);
918 goto fail;
919 }
920 sdp->sd_quotad_process = p;
921
922 return 0;
923
924
925fail_quotad:
926 kthread_stop(sdp->sd_quotad_process);
927fail:
928 kthread_stop(sdp->sd_logd_process);
929 return error;
930}
931
932static const match_table_t nolock_tokens = {
933 { Opt_jid, "jid=%d\n", },
934 { Opt_err, NULL },
935};
936
937static const struct lm_lockops nolock_ops = {
938 .lm_proto_name = "lock_nolock",
939 .lm_put_lock = gfs2_glock_free,
940 .lm_tokens = &nolock_tokens,
941};
942
943/**
944 * gfs2_lm_mount - mount a locking protocol
945 * @sdp: the filesystem
946 * @args: mount arguments
947 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
948 *
949 * Returns: errno
950 */
951
952static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
953{
954 const struct lm_lockops *lm;
955 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
956 struct gfs2_args *args = &sdp->sd_args;
957 const char *proto = sdp->sd_proto_name;
958 const char *table = sdp->sd_table_name;
959 char *o, *options;
960 int ret;
961
962 if (!strcmp("lock_nolock", proto)) {
963 lm = &nolock_ops;
964 sdp->sd_args.ar_localflocks = 1;
965#ifdef CONFIG_GFS2_FS_LOCKING_DLM
966 } else if (!strcmp("lock_dlm", proto)) {
967 lm = &gfs2_dlm_ops;
968#endif
969 } else {
970 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
971 return -ENOENT;
972 }
973
974 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
975
976 ls->ls_ops = lm;
977 ls->ls_first = 1;
978
979 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
980 substring_t tmp[MAX_OPT_ARGS];
981 int token, option;
982
983 if (!o || !*o)
984 continue;
985
986 token = match_token(o, *lm->lm_tokens, tmp);
987 switch (token) {
988 case Opt_jid:
989 ret = match_int(&tmp[0], &option);
990 if (ret || option < 0)
991 goto hostdata_error;
992 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
993 ls->ls_jid = option;
994 break;
995 case Opt_id:
996 case Opt_nodir:
997 /* Obsolete, but left for backward compat purposes */
998 break;
999 case Opt_first:
1000 ret = match_int(&tmp[0], &option);
1001 if (ret || (option != 0 && option != 1))
1002 goto hostdata_error;
1003 ls->ls_first = option;
1004 break;
1005 case Opt_err:
1006 default:
1007hostdata_error:
1008 fs_info(sdp, "unknown hostdata (%s)\n", o);
1009 return -EINVAL;
1010 }
1011 }
1012
1013 if (lm->lm_mount == NULL) {
1014 fs_info(sdp, "Now mounting FS...\n");
1015 complete_all(&sdp->sd_locking_init);
1016 return 0;
1017 }
1018 ret = lm->lm_mount(sdp, table);
1019 if (ret == 0)
1020 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1021 complete_all(&sdp->sd_locking_init);
1022 return ret;
1023}
1024
1025void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1026{
1027 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1028 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1029 lm->lm_unmount)
1030 lm->lm_unmount(sdp);
1031}
1032
1033static int gfs2_journalid_wait(void *word)
1034{
1035 if (signal_pending(current))
1036 return -EINTR;
1037 schedule();
1038 return 0;
1039}
1040
1041static int wait_on_journal(struct gfs2_sbd *sdp)
1042{
1043 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1044 return 0;
1045
1046 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1047}
1048
1049void gfs2_online_uevent(struct gfs2_sbd *sdp)
1050{
1051 struct super_block *sb = sdp->sd_vfs;
1052 char ro[20];
1053 char spectator[20];
1054 char *envp[] = { ro, spectator, NULL };
1055 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1056 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1057 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1058}
1059
1060/**
1061 * fill_super - Read in superblock
1062 * @sb: The VFS superblock
1063 * @data: Mount options
1064 * @silent: Don't complain if it's not a GFS2 filesystem
1065 *
1066 * Returns: errno
1067 */
1068
1069static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1070{
1071 struct gfs2_sbd *sdp;
1072 struct gfs2_holder mount_gh;
1073 int error;
1074
1075 sdp = init_sbd(sb);
1076 if (!sdp) {
1077 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1078 return -ENOMEM;
1079 }
1080 sdp->sd_args = *args;
1081
1082 if (sdp->sd_args.ar_spectator) {
1083 sb->s_flags |= MS_RDONLY;
1084 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1085 }
1086 if (sdp->sd_args.ar_posix_acl)
1087 sb->s_flags |= MS_POSIXACL;
1088 if (sdp->sd_args.ar_nobarrier)
1089 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1090
1091 sb->s_flags |= MS_NOSEC;
1092 sb->s_magic = GFS2_MAGIC;
1093 sb->s_op = &gfs2_super_ops;
1094 sb->s_d_op = &gfs2_dops;
1095 sb->s_export_op = &gfs2_export_ops;
1096 sb->s_xattr = gfs2_xattr_handlers;
1097 sb->s_qcop = &gfs2_quotactl_ops;
1098 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1099 sb->s_time_gran = 1;
1100 sb->s_maxbytes = MAX_LFS_FILESIZE;
1101
1102 /* Set up the buffer cache and fill in some fake block size values
1103 to allow us to read-in the on-disk superblock. */
1104 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1105 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1106 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1107 GFS2_BASIC_BLOCK_SHIFT;
1108 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1109
1110 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1111 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1112 if (sdp->sd_args.ar_statfs_quantum) {
1113 sdp->sd_tune.gt_statfs_slow = 0;
1114 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1115 } else {
1116 sdp->sd_tune.gt_statfs_slow = 1;
1117 sdp->sd_tune.gt_statfs_quantum = 30;
1118 }
1119
1120 error = init_names(sdp, silent);
1121 if (error)
1122 goto fail;
1123
1124 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1125
1126 gfs2_create_debugfs_file(sdp);
1127
1128 error = gfs2_sys_fs_add(sdp);
1129 if (error)
1130 goto fail;
1131
1132 error = gfs2_lm_mount(sdp, silent);
1133 if (error)
1134 goto fail_sys;
1135
1136 error = init_locking(sdp, &mount_gh, DO);
1137 if (error)
1138 goto fail_lm;
1139
1140 error = init_sb(sdp, silent);
1141 if (error)
1142 goto fail_locking;
1143
1144 error = wait_on_journal(sdp);
1145 if (error)
1146 goto fail_sb;
1147
1148 /*
1149 * If user space has failed to join the cluster or some similar
1150 * failure has occurred, then the journal id will contain a
1151 * negative (error) number. This will then be returned to the
1152 * caller (of the mount syscall). We do this even for spectator
1153 * mounts (which just write a jid of 0 to indicate "ok" even though
1154 * the jid is unused in the spectator case)
1155 */
1156 if (sdp->sd_lockstruct.ls_jid < 0) {
1157 error = sdp->sd_lockstruct.ls_jid;
1158 sdp->sd_lockstruct.ls_jid = 0;
1159 goto fail_sb;
1160 }
1161
1162 if (sdp->sd_args.ar_spectator)
1163 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1164 sdp->sd_table_name);
1165 else
1166 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1167 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1168
1169 error = init_inodes(sdp, DO);
1170 if (error)
1171 goto fail_sb;
1172
1173 error = init_per_node(sdp, DO);
1174 if (error)
1175 goto fail_inodes;
1176
1177 error = gfs2_statfs_init(sdp);
1178 if (error) {
1179 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1180 goto fail_per_node;
1181 }
1182
1183 error = init_threads(sdp, DO);
1184 if (error)
1185 goto fail_per_node;
1186
1187 if (!(sb->s_flags & MS_RDONLY)) {
1188 error = gfs2_make_fs_rw(sdp);
1189 if (error) {
1190 fs_err(sdp, "can't make FS RW: %d\n", error);
1191 goto fail_threads;
1192 }
1193 }
1194
1195 gfs2_glock_dq_uninit(&mount_gh);
1196 gfs2_online_uevent(sdp);
1197 return 0;
1198
1199fail_threads:
1200 init_threads(sdp, UNDO);
1201fail_per_node:
1202 init_per_node(sdp, UNDO);
1203fail_inodes:
1204 init_inodes(sdp, UNDO);
1205fail_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;
1213fail_locking:
1214 init_locking(sdp, &mount_gh, UNDO);
1215fail_lm:
1216 gfs2_gl_hash_clear(sdp);
1217 gfs2_lm_unmount(sdp);
1218fail_sys:
1219 gfs2_sys_fs_del(sdp);
1220fail:
1221 gfs2_delete_debugfs_file(sdp);
1222 free_percpu(sdp->sd_lkstats);
1223 kfree(sdp);
1224 sb->s_fs_info = NULL;
1225 return error;
1226}
1227
1228static int set_gfs2_super(struct super_block *s, void *data)
1229{
1230 s->s_bdev = data;
1231 s->s_dev = s->s_bdev->bd_dev;
1232
1233 /*
1234 * We set the bdi here to the queue backing, file systems can
1235 * overwrite this in ->fill_super()
1236 */
1237 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1238 return 0;
1239}
1240
1241static int test_gfs2_super(struct super_block *s, void *ptr)
1242{
1243 struct block_device *bdev = ptr;
1244 return (bdev == s->s_bdev);
1245}
1246
1247/**
1248 * gfs2_mount - Get the GFS2 superblock
1249 * @fs_type: The GFS2 filesystem type
1250 * @flags: Mount flags
1251 * @dev_name: The name of the device
1252 * @data: The mount arguments
1253 *
1254 * Q. Why not use get_sb_bdev() ?
1255 * A. We need to select one of two root directories to mount, independent
1256 * of whether this is the initial, or subsequent, mount of this sb
1257 *
1258 * Returns: 0 or -ve on error
1259 */
1260
1261static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1262 const char *dev_name, void *data)
1263{
1264 struct block_device *bdev;
1265 struct super_block *s;
1266 fmode_t mode = FMODE_READ | FMODE_EXCL;
1267 int error;
1268 struct gfs2_args args;
1269 struct gfs2_sbd *sdp;
1270
1271 if (!(flags & MS_RDONLY))
1272 mode |= FMODE_WRITE;
1273
1274 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1275 if (IS_ERR(bdev))
1276 return ERR_CAST(bdev);
1277
1278 /*
1279 * once the super is inserted into the list by sget, s_umount
1280 * will protect the lockfs code from trying to start a snapshot
1281 * while we are mounting
1282 */
1283 mutex_lock(&bdev->bd_fsfreeze_mutex);
1284 if (bdev->bd_fsfreeze_count > 0) {
1285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1286 error = -EBUSY;
1287 goto error_bdev;
1288 }
1289 s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1290 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1291 error = PTR_ERR(s);
1292 if (IS_ERR(s))
1293 goto error_bdev;
1294
1295 if (s->s_root)
1296 blkdev_put(bdev, mode);
1297
1298 memset(&args, 0, sizeof(args));
1299 args.ar_quota = GFS2_QUOTA_DEFAULT;
1300 args.ar_data = GFS2_DATA_DEFAULT;
1301 args.ar_commit = 30;
1302 args.ar_statfs_quantum = 30;
1303 args.ar_quota_quantum = 60;
1304 args.ar_errors = GFS2_ERRORS_DEFAULT;
1305
1306 error = gfs2_mount_args(&args, data);
1307 if (error) {
1308 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1309 goto error_super;
1310 }
1311
1312 if (s->s_root) {
1313 error = -EBUSY;
1314 if ((flags ^ s->s_flags) & MS_RDONLY)
1315 goto error_super;
1316 } else {
1317 char b[BDEVNAME_SIZE];
1318
1319 s->s_flags = flags;
1320 s->s_mode = mode;
1321 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1322 sb_set_blocksize(s, block_size(bdev));
1323 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1324 if (error)
1325 goto error_super;
1326 s->s_flags |= MS_ACTIVE;
1327 bdev->bd_super = s;
1328 }
1329
1330 sdp = s->s_fs_info;
1331 if (args.ar_meta)
1332 return dget(sdp->sd_master_dir);
1333 else
1334 return dget(sdp->sd_root_dir);
1335
1336error_super:
1337 deactivate_locked_super(s);
1338 return ERR_PTR(error);
1339error_bdev:
1340 blkdev_put(bdev, mode);
1341 return ERR_PTR(error);
1342}
1343
1344static int set_meta_super(struct super_block *s, void *ptr)
1345{
1346 return -EINVAL;
1347}
1348
1349static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1350 int flags, const char *dev_name, void *data)
1351{
1352 struct super_block *s;
1353 struct gfs2_sbd *sdp;
1354 struct path path;
1355 int error;
1356
1357 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1358 if (error) {
1359 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1360 dev_name, error);
1361 return ERR_PTR(error);
1362 }
1363 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1364 path.dentry->d_inode->i_sb->s_bdev);
1365 path_put(&path);
1366 if (IS_ERR(s)) {
1367 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1368 return ERR_CAST(s);
1369 }
1370 if ((flags ^ s->s_flags) & MS_RDONLY) {
1371 deactivate_locked_super(s);
1372 return ERR_PTR(-EBUSY);
1373 }
1374 sdp = s->s_fs_info;
1375 return dget(sdp->sd_master_dir);
1376}
1377
1378static void gfs2_kill_sb(struct super_block *sb)
1379{
1380 struct gfs2_sbd *sdp = sb->s_fs_info;
1381
1382 if (sdp == NULL) {
1383 kill_block_super(sb);
1384 return;
1385 }
1386
1387 gfs2_meta_syncfs(sdp);
1388 dput(sdp->sd_root_dir);
1389 dput(sdp->sd_master_dir);
1390 sdp->sd_root_dir = NULL;
1391 sdp->sd_master_dir = NULL;
1392 shrink_dcache_sb(sb);
1393 kill_block_super(sb);
1394 gfs2_delete_debugfs_file(sdp);
1395 free_percpu(sdp->sd_lkstats);
1396 kfree(sdp);
1397}
1398
1399struct 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
1407struct 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
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
54static 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
66static 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 sb->s_fs_info = sdp;
76 sdp->sd_vfs = sb;
77 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
78 if (!sdp->sd_lkstats) {
79 kfree(sdp);
80 return NULL;
81 }
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
153static 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 format numbers match exactly, we're done. */
165
166 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
167 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
168 return 0;
169
170 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
171
172 return -EINVAL;
173}
174
175static void end_bio_io_page(struct bio *bio)
176{
177 struct page *page = bio->bi_private;
178
179 if (!bio->bi_status)
180 SetPageUptodate(page);
181 else
182 pr_warn("error %d reading superblock\n", bio->bi_status);
183 unlock_page(page);
184}
185
186static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
187{
188 struct gfs2_sb_host *sb = &sdp->sd_sb;
189 struct super_block *s = sdp->sd_vfs;
190 const struct gfs2_sb *str = buf;
191
192 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
193 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
194 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
195 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
196 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
197 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
198 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
199 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
200 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
201 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
202 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
203
204 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
205 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
206 memcpy(&s->s_uuid, str->sb_uuid, 16);
207}
208
209/**
210 * gfs2_read_super - Read the gfs2 super block from disk
211 * @sdp: The GFS2 super block
212 * @sector: The location of the super block
213 * @error: The error code to return
214 *
215 * This uses the bio functions to read the super block from disk
216 * because we want to be 100% sure that we never read cached data.
217 * A super block is read twice only during each GFS2 mount and is
218 * never written to by the filesystem. The first time its read no
219 * locks are held, and the only details which are looked at are those
220 * relating to the locking protocol. Once locking is up and working,
221 * the sb is read again under the lock to establish the location of
222 * the master directory (contains pointers to journals etc) and the
223 * root directory.
224 *
225 * Returns: 0 on success or error
226 */
227
228static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
229{
230 struct super_block *sb = sdp->sd_vfs;
231 struct gfs2_sb *p;
232 struct page *page;
233 struct bio *bio;
234
235 page = alloc_page(GFP_NOFS);
236 if (unlikely(!page))
237 return -ENOMEM;
238
239 ClearPageUptodate(page);
240 ClearPageDirty(page);
241 lock_page(page);
242
243 bio = bio_alloc(GFP_NOFS, 1);
244 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
245 bio_set_dev(bio, sb->s_bdev);
246 bio_add_page(bio, page, PAGE_SIZE, 0);
247
248 bio->bi_end_io = end_bio_io_page;
249 bio->bi_private = page;
250 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
251 submit_bio(bio);
252 wait_on_page_locked(page);
253 bio_put(bio);
254 if (!PageUptodate(page)) {
255 __free_page(page);
256 return -EIO;
257 }
258 p = kmap(page);
259 gfs2_sb_in(sdp, p);
260 kunmap(page);
261 __free_page(page);
262 return gfs2_check_sb(sdp, silent);
263}
264
265/**
266 * gfs2_read_sb - Read super block
267 * @sdp: The GFS2 superblock
268 * @silent: Don't print message if mount fails
269 *
270 */
271
272static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
273{
274 u32 hash_blocks, ind_blocks, leaf_blocks;
275 u32 tmp_blocks;
276 unsigned int x;
277 int error;
278
279 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
280 if (error) {
281 if (!silent)
282 fs_err(sdp, "can't read superblock\n");
283 return error;
284 }
285
286 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
287 GFS2_BASIC_BLOCK_SHIFT;
288 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
289 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
290 sizeof(struct gfs2_dinode)) / sizeof(u64);
291 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
292 sizeof(struct gfs2_meta_header)) / sizeof(u64);
293 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
294 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
295 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
296 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
297 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
298 sizeof(struct gfs2_meta_header)) /
299 sizeof(struct gfs2_quota_change);
300 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
301 sizeof(struct gfs2_meta_header))
302 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
303
304 /* Compute maximum reservation required to add a entry to a directory */
305
306 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
307 sdp->sd_jbsize);
308
309 ind_blocks = 0;
310 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
311 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
312 ind_blocks += tmp_blocks;
313 }
314
315 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
316
317 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
318
319 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
320 sizeof(struct gfs2_dinode);
321 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
322 for (x = 2;; x++) {
323 u64 space, d;
324 u32 m;
325
326 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
327 d = space;
328 m = do_div(d, sdp->sd_inptrs);
329
330 if (d != sdp->sd_heightsize[x - 1] || m)
331 break;
332 sdp->sd_heightsize[x] = space;
333 }
334 sdp->sd_max_height = x;
335 sdp->sd_heightsize[x] = ~0;
336 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
337
338 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
339 sizeof(struct gfs2_dinode);
340 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
341 for (x = 2;; x++) {
342 u64 space, d;
343 u32 m;
344
345 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
346 d = space;
347 m = do_div(d, sdp->sd_inptrs);
348
349 if (d != sdp->sd_jheightsize[x - 1] || m)
350 break;
351 sdp->sd_jheightsize[x] = space;
352 }
353 sdp->sd_max_jheight = x;
354 sdp->sd_jheightsize[x] = ~0;
355 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
356
357 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
358 sizeof(struct gfs2_leaf)) /
359 GFS2_MIN_DIRENT_SIZE;
360 return 0;
361}
362
363static int init_names(struct gfs2_sbd *sdp, int silent)
364{
365 char *proto, *table;
366 int error = 0;
367
368 proto = sdp->sd_args.ar_lockproto;
369 table = sdp->sd_args.ar_locktable;
370
371 /* Try to autodetect */
372
373 if (!proto[0] || !table[0]) {
374 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
375 if (error)
376 return error;
377
378 if (!proto[0])
379 proto = sdp->sd_sb.sb_lockproto;
380 if (!table[0])
381 table = sdp->sd_sb.sb_locktable;
382 }
383
384 if (!table[0])
385 table = sdp->sd_vfs->s_id;
386
387 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
388 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
389
390 table = sdp->sd_table_name;
391 while ((table = strchr(table, '/')))
392 *table = '_';
393
394 return error;
395}
396
397static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
398 int undo)
399{
400 int error = 0;
401
402 if (undo)
403 goto fail_trans;
404
405 error = gfs2_glock_nq_num(sdp,
406 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
407 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
408 mount_gh);
409 if (error) {
410 fs_err(sdp, "can't acquire mount glock: %d\n", error);
411 goto fail;
412 }
413
414 error = gfs2_glock_nq_num(sdp,
415 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
416 LM_ST_SHARED,
417 LM_FLAG_NOEXP | GL_EXACT,
418 &sdp->sd_live_gh);
419 if (error) {
420 fs_err(sdp, "can't acquire live glock: %d\n", error);
421 goto fail_mount;
422 }
423
424 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
425 CREATE, &sdp->sd_rename_gl);
426 if (error) {
427 fs_err(sdp, "can't create rename glock: %d\n", error);
428 goto fail_live;
429 }
430
431 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
432 CREATE, &sdp->sd_freeze_gl);
433 if (error) {
434 fs_err(sdp, "can't create transaction glock: %d\n", error);
435 goto fail_rename;
436 }
437
438 return 0;
439
440fail_trans:
441 gfs2_glock_put(sdp->sd_freeze_gl);
442fail_rename:
443 gfs2_glock_put(sdp->sd_rename_gl);
444fail_live:
445 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
446fail_mount:
447 gfs2_glock_dq_uninit(mount_gh);
448fail:
449 return error;
450}
451
452static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
453 u64 no_addr, const char *name)
454{
455 struct gfs2_sbd *sdp = sb->s_fs_info;
456 struct dentry *dentry;
457 struct inode *inode;
458
459 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
460 GFS2_BLKST_FREE /* ignore */);
461 if (IS_ERR(inode)) {
462 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
463 return PTR_ERR(inode);
464 }
465 dentry = d_make_root(inode);
466 if (!dentry) {
467 fs_err(sdp, "can't alloc %s dentry\n", name);
468 return -ENOMEM;
469 }
470 *dptr = dentry;
471 return 0;
472}
473
474static int init_sb(struct gfs2_sbd *sdp, int silent)
475{
476 struct super_block *sb = sdp->sd_vfs;
477 struct gfs2_holder sb_gh;
478 u64 no_addr;
479 int ret;
480
481 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
482 LM_ST_SHARED, 0, &sb_gh);
483 if (ret) {
484 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
485 return ret;
486 }
487
488 ret = gfs2_read_sb(sdp, silent);
489 if (ret) {
490 fs_err(sdp, "can't read superblock: %d\n", ret);
491 goto out;
492 }
493
494 /* Set up the buffer cache and SB for real */
495 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
496 ret = -EINVAL;
497 fs_err(sdp, "FS block size (%u) is too small for device "
498 "block size (%u)\n",
499 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
500 goto out;
501 }
502 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
503 ret = -EINVAL;
504 fs_err(sdp, "FS block size (%u) is too big for machine "
505 "page size (%u)\n",
506 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
507 goto out;
508 }
509 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
510
511 /* Get the root inode */
512 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
513 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
514 if (ret)
515 goto out;
516
517 /* Get the master inode */
518 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
519 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
520 if (ret) {
521 dput(sdp->sd_root_dir);
522 goto out;
523 }
524 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
525out:
526 gfs2_glock_dq_uninit(&sb_gh);
527 return ret;
528}
529
530static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
531{
532 char *message = "FIRSTMOUNT=Done";
533 char *envp[] = { message, NULL };
534
535 fs_info(sdp, "first mount done, others may mount\n");
536
537 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
538 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
539
540 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
541}
542
543/**
544 * gfs2_jindex_hold - Grab a lock on the jindex
545 * @sdp: The GFS2 superblock
546 * @ji_gh: the holder for the jindex glock
547 *
548 * Returns: errno
549 */
550
551static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
552{
553 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
554 struct qstr name;
555 char buf[20];
556 struct gfs2_jdesc *jd;
557 int error;
558
559 name.name = buf;
560
561 mutex_lock(&sdp->sd_jindex_mutex);
562
563 for (;;) {
564 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
565 if (error)
566 break;
567
568 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
569 name.hash = gfs2_disk_hash(name.name, name.len);
570
571 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
572 if (error == -ENOENT) {
573 error = 0;
574 break;
575 }
576
577 gfs2_glock_dq_uninit(ji_gh);
578
579 if (error)
580 break;
581
582 error = -ENOMEM;
583 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
584 if (!jd)
585 break;
586
587 INIT_LIST_HEAD(&jd->extent_list);
588 INIT_LIST_HEAD(&jd->jd_revoke_list);
589
590 INIT_WORK(&jd->jd_work, gfs2_recover_func);
591 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
592 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
593 if (!jd->jd_inode)
594 error = -ENOENT;
595 else
596 error = PTR_ERR(jd->jd_inode);
597 kfree(jd);
598 break;
599 }
600
601 spin_lock(&sdp->sd_jindex_spin);
602 jd->jd_jid = sdp->sd_journals++;
603 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
604 spin_unlock(&sdp->sd_jindex_spin);
605 }
606
607 mutex_unlock(&sdp->sd_jindex_mutex);
608
609 return error;
610}
611
612/**
613 * check_journal_clean - Make sure a journal is clean for a spectator mount
614 * @sdp: The GFS2 superblock
615 * @jd: The journal descriptor
616 *
617 * Returns: 0 if the journal is clean or locked, else an error
618 */
619static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
620{
621 int error;
622 struct gfs2_holder j_gh;
623 struct gfs2_log_header_host head;
624 struct gfs2_inode *ip;
625
626 ip = GFS2_I(jd->jd_inode);
627 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
628 GL_EXACT | GL_NOCACHE, &j_gh);
629 if (error) {
630 fs_err(sdp, "Error locking journal for spectator mount.\n");
631 return -EPERM;
632 }
633 error = gfs2_jdesc_check(jd);
634 if (error) {
635 fs_err(sdp, "Error checking journal for spectator mount.\n");
636 goto out_unlock;
637 }
638 error = gfs2_find_jhead(jd, &head);
639 if (error) {
640 fs_err(sdp, "Error parsing journal for spectator mount.\n");
641 goto out_unlock;
642 }
643 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
644 error = -EPERM;
645 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
646 "must not be a spectator.\n", jd->jd_jid);
647 }
648
649out_unlock:
650 gfs2_glock_dq_uninit(&j_gh);
651 return error;
652}
653
654static int init_journal(struct gfs2_sbd *sdp, int undo)
655{
656 struct inode *master = d_inode(sdp->sd_master_dir);
657 struct gfs2_holder ji_gh;
658 struct gfs2_inode *ip;
659 int jindex = 1;
660 int error = 0;
661
662 if (undo) {
663 jindex = 0;
664 goto fail_jinode_gh;
665 }
666
667 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
668 if (IS_ERR(sdp->sd_jindex)) {
669 fs_err(sdp, "can't lookup journal index: %d\n", error);
670 return PTR_ERR(sdp->sd_jindex);
671 }
672
673 /* Load in the journal index special file */
674
675 error = gfs2_jindex_hold(sdp, &ji_gh);
676 if (error) {
677 fs_err(sdp, "can't read journal index: %d\n", error);
678 goto fail;
679 }
680
681 error = -EUSERS;
682 if (!gfs2_jindex_size(sdp)) {
683 fs_err(sdp, "no journals!\n");
684 goto fail_jindex;
685 }
686
687 atomic_set(&sdp->sd_log_blks_needed, 0);
688 if (sdp->sd_args.ar_spectator) {
689 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
690 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
691 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
692 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
693 } else {
694 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
695 fs_err(sdp, "can't mount journal #%u\n",
696 sdp->sd_lockstruct.ls_jid);
697 fs_err(sdp, "there are only %u journals (0 - %u)\n",
698 gfs2_jindex_size(sdp),
699 gfs2_jindex_size(sdp) - 1);
700 goto fail_jindex;
701 }
702 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
703
704 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
705 &gfs2_journal_glops,
706 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
707 &sdp->sd_journal_gh);
708 if (error) {
709 fs_err(sdp, "can't acquire journal glock: %d\n", error);
710 goto fail_jindex;
711 }
712
713 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
714 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
715 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
716 &sdp->sd_jinode_gh);
717 if (error) {
718 fs_err(sdp, "can't acquire journal inode glock: %d\n",
719 error);
720 goto fail_journal_gh;
721 }
722
723 error = gfs2_jdesc_check(sdp->sd_jdesc);
724 if (error) {
725 fs_err(sdp, "my journal (%u) is bad: %d\n",
726 sdp->sd_jdesc->jd_jid, error);
727 goto fail_jinode_gh;
728 }
729 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
730 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
731 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
732
733 /* Map the extents for this journal's blocks */
734 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
735 }
736 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
737
738 if (sdp->sd_lockstruct.ls_first) {
739 unsigned int x;
740 for (x = 0; x < sdp->sd_journals; x++) {
741 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
742
743 if (sdp->sd_args.ar_spectator) {
744 error = check_journal_clean(sdp, jd);
745 if (error)
746 goto fail_jinode_gh;
747 continue;
748 }
749 error = gfs2_recover_journal(jd, true);
750 if (error) {
751 fs_err(sdp, "error recovering journal %u: %d\n",
752 x, error);
753 goto fail_jinode_gh;
754 }
755 }
756
757 gfs2_others_may_mount(sdp);
758 } else if (!sdp->sd_args.ar_spectator) {
759 error = gfs2_recover_journal(sdp->sd_jdesc, true);
760 if (error) {
761 fs_err(sdp, "error recovering my journal: %d\n", error);
762 goto fail_jinode_gh;
763 }
764 }
765
766 sdp->sd_log_idle = 1;
767 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
768 gfs2_glock_dq_uninit(&ji_gh);
769 jindex = 0;
770 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
771 return 0;
772
773fail_jinode_gh:
774 if (!sdp->sd_args.ar_spectator)
775 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
776fail_journal_gh:
777 if (!sdp->sd_args.ar_spectator)
778 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
779fail_jindex:
780 gfs2_jindex_free(sdp);
781 if (jindex)
782 gfs2_glock_dq_uninit(&ji_gh);
783fail:
784 iput(sdp->sd_jindex);
785 return error;
786}
787
788static struct lock_class_key gfs2_quota_imutex_key;
789
790static int init_inodes(struct gfs2_sbd *sdp, int undo)
791{
792 int error = 0;
793 struct inode *master = d_inode(sdp->sd_master_dir);
794
795 if (undo)
796 goto fail_qinode;
797
798 error = init_journal(sdp, undo);
799 complete_all(&sdp->sd_journal_ready);
800 if (error)
801 goto fail;
802
803 /* Read in the master statfs inode */
804 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
805 if (IS_ERR(sdp->sd_statfs_inode)) {
806 error = PTR_ERR(sdp->sd_statfs_inode);
807 fs_err(sdp, "can't read in statfs inode: %d\n", error);
808 goto fail_journal;
809 }
810
811 /* Read in the resource index inode */
812 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
813 if (IS_ERR(sdp->sd_rindex)) {
814 error = PTR_ERR(sdp->sd_rindex);
815 fs_err(sdp, "can't get resource index inode: %d\n", error);
816 goto fail_statfs;
817 }
818 sdp->sd_rindex_uptodate = 0;
819
820 /* Read in the quota inode */
821 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
822 if (IS_ERR(sdp->sd_quota_inode)) {
823 error = PTR_ERR(sdp->sd_quota_inode);
824 fs_err(sdp, "can't get quota file inode: %d\n", error);
825 goto fail_rindex;
826 }
827 /*
828 * i_rwsem on quota files is special. Since this inode is hidden system
829 * file, we are safe to define locking ourselves.
830 */
831 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
832 &gfs2_quota_imutex_key);
833
834 error = gfs2_rindex_update(sdp);
835 if (error)
836 goto fail_qinode;
837
838 return 0;
839
840fail_qinode:
841 iput(sdp->sd_quota_inode);
842fail_rindex:
843 gfs2_clear_rgrpd(sdp);
844 iput(sdp->sd_rindex);
845fail_statfs:
846 iput(sdp->sd_statfs_inode);
847fail_journal:
848 init_journal(sdp, UNDO);
849fail:
850 return error;
851}
852
853static int init_per_node(struct gfs2_sbd *sdp, int undo)
854{
855 struct inode *pn = NULL;
856 char buf[30];
857 int error = 0;
858 struct gfs2_inode *ip;
859 struct inode *master = d_inode(sdp->sd_master_dir);
860
861 if (sdp->sd_args.ar_spectator)
862 return 0;
863
864 if (undo)
865 goto fail_qc_gh;
866
867 pn = gfs2_lookup_simple(master, "per_node");
868 if (IS_ERR(pn)) {
869 error = PTR_ERR(pn);
870 fs_err(sdp, "can't find per_node directory: %d\n", error);
871 return error;
872 }
873
874 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
875 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
876 if (IS_ERR(sdp->sd_sc_inode)) {
877 error = PTR_ERR(sdp->sd_sc_inode);
878 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
879 goto fail;
880 }
881
882 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
883 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
884 if (IS_ERR(sdp->sd_qc_inode)) {
885 error = PTR_ERR(sdp->sd_qc_inode);
886 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
887 goto fail_ut_i;
888 }
889
890 iput(pn);
891 pn = NULL;
892
893 ip = GFS2_I(sdp->sd_sc_inode);
894 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
895 &sdp->sd_sc_gh);
896 if (error) {
897 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
898 goto fail_qc_i;
899 }
900
901 ip = GFS2_I(sdp->sd_qc_inode);
902 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
903 &sdp->sd_qc_gh);
904 if (error) {
905 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
906 goto fail_ut_gh;
907 }
908
909 return 0;
910
911fail_qc_gh:
912 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
913fail_ut_gh:
914 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
915fail_qc_i:
916 iput(sdp->sd_qc_inode);
917fail_ut_i:
918 iput(sdp->sd_sc_inode);
919fail:
920 iput(pn);
921 return error;
922}
923
924static const match_table_t nolock_tokens = {
925 { Opt_jid, "jid=%d\n", },
926 { Opt_err, NULL },
927};
928
929static const struct lm_lockops nolock_ops = {
930 .lm_proto_name = "lock_nolock",
931 .lm_put_lock = gfs2_glock_free,
932 .lm_tokens = &nolock_tokens,
933};
934
935/**
936 * gfs2_lm_mount - mount a locking protocol
937 * @sdp: the filesystem
938 * @args: mount arguments
939 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
940 *
941 * Returns: errno
942 */
943
944static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
945{
946 const struct lm_lockops *lm;
947 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
948 struct gfs2_args *args = &sdp->sd_args;
949 const char *proto = sdp->sd_proto_name;
950 const char *table = sdp->sd_table_name;
951 char *o, *options;
952 int ret;
953
954 if (!strcmp("lock_nolock", proto)) {
955 lm = &nolock_ops;
956 sdp->sd_args.ar_localflocks = 1;
957#ifdef CONFIG_GFS2_FS_LOCKING_DLM
958 } else if (!strcmp("lock_dlm", proto)) {
959 lm = &gfs2_dlm_ops;
960#endif
961 } else {
962 pr_info("can't find protocol %s\n", proto);
963 return -ENOENT;
964 }
965
966 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
967
968 ls->ls_ops = lm;
969 ls->ls_first = 1;
970
971 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
972 substring_t tmp[MAX_OPT_ARGS];
973 int token, option;
974
975 if (!o || !*o)
976 continue;
977
978 token = match_token(o, *lm->lm_tokens, tmp);
979 switch (token) {
980 case Opt_jid:
981 ret = match_int(&tmp[0], &option);
982 if (ret || option < 0)
983 goto hostdata_error;
984 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
985 ls->ls_jid = option;
986 break;
987 case Opt_id:
988 case Opt_nodir:
989 /* Obsolete, but left for backward compat purposes */
990 break;
991 case Opt_first:
992 ret = match_int(&tmp[0], &option);
993 if (ret || (option != 0 && option != 1))
994 goto hostdata_error;
995 ls->ls_first = option;
996 break;
997 case Opt_err:
998 default:
999hostdata_error:
1000 fs_info(sdp, "unknown hostdata (%s)\n", o);
1001 return -EINVAL;
1002 }
1003 }
1004
1005 if (lm->lm_mount == NULL) {
1006 fs_info(sdp, "Now mounting FS...\n");
1007 complete_all(&sdp->sd_locking_init);
1008 return 0;
1009 }
1010 ret = lm->lm_mount(sdp, table);
1011 if (ret == 0)
1012 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1013 complete_all(&sdp->sd_locking_init);
1014 return ret;
1015}
1016
1017void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1018{
1019 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1020 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1021 lm->lm_unmount)
1022 lm->lm_unmount(sdp);
1023}
1024
1025static int wait_on_journal(struct gfs2_sbd *sdp)
1026{
1027 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1028 return 0;
1029
1030 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1031 ? -EINTR : 0;
1032}
1033
1034void gfs2_online_uevent(struct gfs2_sbd *sdp)
1035{
1036 struct super_block *sb = sdp->sd_vfs;
1037 char ro[20];
1038 char spectator[20];
1039 char *envp[] = { ro, spectator, NULL };
1040 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1041 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1042 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1043}
1044
1045/**
1046 * fill_super - Read in superblock
1047 * @sb: The VFS superblock
1048 * @data: Mount options
1049 * @silent: Don't complain if it's not a GFS2 filesystem
1050 *
1051 * Returns: errno
1052 */
1053
1054static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1055{
1056 struct gfs2_sbd *sdp;
1057 struct gfs2_holder mount_gh;
1058 int error;
1059
1060 sdp = init_sbd(sb);
1061 if (!sdp) {
1062 pr_warn("can't alloc struct gfs2_sbd\n");
1063 return -ENOMEM;
1064 }
1065 sdp->sd_args = *args;
1066
1067 if (sdp->sd_args.ar_spectator) {
1068 sb->s_flags |= SB_RDONLY;
1069 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1070 }
1071 if (sdp->sd_args.ar_posix_acl)
1072 sb->s_flags |= SB_POSIXACL;
1073 if (sdp->sd_args.ar_nobarrier)
1074 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1075
1076 sb->s_flags |= SB_NOSEC;
1077 sb->s_magic = GFS2_MAGIC;
1078 sb->s_op = &gfs2_super_ops;
1079 sb->s_d_op = &gfs2_dops;
1080 sb->s_export_op = &gfs2_export_ops;
1081 sb->s_xattr = gfs2_xattr_handlers;
1082 sb->s_qcop = &gfs2_quotactl_ops;
1083 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1084 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1085 sb->s_time_gran = 1;
1086 sb->s_maxbytes = MAX_LFS_FILESIZE;
1087
1088 /* Set up the buffer cache and fill in some fake block size values
1089 to allow us to read-in the on-disk superblock. */
1090 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1091 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1092 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1093 GFS2_BASIC_BLOCK_SHIFT;
1094 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1095
1096 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1097 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1098 if (sdp->sd_args.ar_statfs_quantum) {
1099 sdp->sd_tune.gt_statfs_slow = 0;
1100 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1101 } else {
1102 sdp->sd_tune.gt_statfs_slow = 1;
1103 sdp->sd_tune.gt_statfs_quantum = 30;
1104 }
1105
1106 error = init_names(sdp, silent);
1107 if (error) {
1108 /* In this case, we haven't initialized sysfs, so we have to
1109 manually free the sdp. */
1110 free_percpu(sdp->sd_lkstats);
1111 kfree(sdp);
1112 sb->s_fs_info = NULL;
1113 return error;
1114 }
1115
1116 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1117
1118 error = gfs2_sys_fs_add(sdp);
1119 /*
1120 * If we hit an error here, gfs2_sys_fs_add will have called function
1121 * kobject_put which causes the sysfs usage count to go to zero, which
1122 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1123 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1124 * kobject_put to free sdp.
1125 */
1126 if (error)
1127 return error;
1128
1129 gfs2_create_debugfs_file(sdp);
1130
1131 error = gfs2_lm_mount(sdp, silent);
1132 if (error)
1133 goto fail_debug;
1134
1135 error = init_locking(sdp, &mount_gh, DO);
1136 if (error)
1137 goto fail_lm;
1138
1139 error = init_sb(sdp, silent);
1140 if (error)
1141 goto fail_locking;
1142
1143 error = wait_on_journal(sdp);
1144 if (error)
1145 goto fail_sb;
1146
1147 /*
1148 * If user space has failed to join the cluster or some similar
1149 * failure has occurred, then the journal id will contain a
1150 * negative (error) number. This will then be returned to the
1151 * caller (of the mount syscall). We do this even for spectator
1152 * mounts (which just write a jid of 0 to indicate "ok" even though
1153 * the jid is unused in the spectator case)
1154 */
1155 if (sdp->sd_lockstruct.ls_jid < 0) {
1156 error = sdp->sd_lockstruct.ls_jid;
1157 sdp->sd_lockstruct.ls_jid = 0;
1158 goto fail_sb;
1159 }
1160
1161 if (sdp->sd_args.ar_spectator)
1162 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1163 sdp->sd_table_name);
1164 else
1165 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1166 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1167
1168 error = init_inodes(sdp, DO);
1169 if (error)
1170 goto fail_sb;
1171
1172 error = init_per_node(sdp, DO);
1173 if (error)
1174 goto fail_inodes;
1175
1176 error = gfs2_statfs_init(sdp);
1177 if (error) {
1178 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1179 goto fail_per_node;
1180 }
1181
1182 if (!sb_rdonly(sb)) {
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
1194fail_per_node:
1195 init_per_node(sdp, UNDO);
1196fail_inodes:
1197 init_inodes(sdp, UNDO);
1198fail_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;
1206fail_locking:
1207 init_locking(sdp, &mount_gh, UNDO);
1208fail_lm:
1209 complete_all(&sdp->sd_journal_ready);
1210 gfs2_gl_hash_clear(sdp);
1211 gfs2_lm_unmount(sdp);
1212fail_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
1222static 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
1230static 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
1250static 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
1334error_super:
1335 deactivate_locked_super(s);
1336 return ERR_PTR(error);
1337error_bdev:
1338 blkdev_put(bdev, mode);
1339 return ERR_PTR(error);
1340}
1341
1342static int set_meta_super(struct super_block *s, void *ptr)
1343{
1344 return -EINVAL;
1345}
1346
1347static 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 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1356 if (error) {
1357 pr_warn("path_lookup on %s returned error %d\n",
1358 dev_name, error);
1359 return ERR_PTR(error);
1360 }
1361 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1362 path.dentry->d_sb->s_bdev);
1363 path_put(&path);
1364 if (IS_ERR(s)) {
1365 pr_warn("gfs2 mount does not exist\n");
1366 return ERR_CAST(s);
1367 }
1368 if ((flags ^ s->s_flags) & SB_RDONLY) {
1369 deactivate_locked_super(s);
1370 return ERR_PTR(-EBUSY);
1371 }
1372 sdp = s->s_fs_info;
1373 return dget(sdp->sd_master_dir);
1374}
1375
1376static void gfs2_kill_sb(struct super_block *sb)
1377{
1378 struct gfs2_sbd *sdp = sb->s_fs_info;
1379
1380 if (sdp == NULL) {
1381 kill_block_super(sb);
1382 return;
1383 }
1384
1385 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1386 dput(sdp->sd_root_dir);
1387 dput(sdp->sd_master_dir);
1388 sdp->sd_root_dir = NULL;
1389 sdp->sd_master_dir = NULL;
1390 shrink_dcache_sb(sb);
1391 free_percpu(sdp->sd_lkstats);
1392 kill_block_super(sb);
1393}
1394
1395struct file_system_type gfs2_fs_type = {
1396 .name = "gfs2",
1397 .fs_flags = FS_REQUIRES_DEV,
1398 .mount = gfs2_mount,
1399 .kill_sb = gfs2_kill_sb,
1400 .owner = THIS_MODULE,
1401};
1402MODULE_ALIAS_FS("gfs2");
1403
1404struct file_system_type gfs2meta_fs_type = {
1405 .name = "gfs2meta",
1406 .fs_flags = FS_REQUIRES_DEV,
1407 .mount = gfs2_mount_meta,
1408 .owner = THIS_MODULE,
1409};
1410MODULE_ALIAS_FS("gfs2meta");