]> git.ipfire.org Git - people/ms/linux.git/blob - fs/gfs2/ops_fstype.c
projects / people / ms / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'block-6.0-2022-08-19' of git://git.kernel.dk/linux-block
[people/ms/linux.git] / fs / gfs2 / ops_fstype.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
24 #include <linux/fs_parser.h>
25
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "recovery.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "sys.h"
36 #include "util.h"
37 #include "log.h"
38 #include "quota.h"
39 #include "dir.h"
40 #include "meta_io.h"
41 #include "trace_gfs2.h"
42 #include "lops.h"
43
44 #define DO 0
45 #define UNDO 1
46
47 /**
48 * gfs2_tune_init - Fill a gfs2_tune structure with default values
49 * @gt: tune
50 *
51 */
52
53 static void gfs2_tune_init(struct gfs2_tune *gt)
54 {
55 spin_lock_init(&gt->gt_spin);
56
57 gt->gt_quota_warn_period = 10;
58 gt->gt_quota_scale_num = 1;
59 gt->gt_quota_scale_den = 1;
60 gt->gt_new_files_jdata = 0;
61 gt->gt_max_readahead = BIT(18);
62 gt->gt_complain_secs = 10;
63 }
64
65 void free_sbd(struct gfs2_sbd *sdp)
66 {
67 if (sdp->sd_lkstats)
68 free_percpu(sdp->sd_lkstats);
69 kfree(sdp);
70 }
71
72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
73 {
74 struct gfs2_sbd *sdp;
75 struct address_space *mapping;
76
77 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
78 if (!sdp)
79 return NULL;
80
81 sdp->sd_vfs = sb;
82 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
83 if (!sdp->sd_lkstats)
84 goto fail;
85 sb->s_fs_info = sdp;
86
87 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
88 gfs2_tune_init(&sdp->sd_tune);
89
90 init_waitqueue_head(&sdp->sd_glock_wait);
91 init_waitqueue_head(&sdp->sd_async_glock_wait);
92 atomic_set(&sdp->sd_glock_disposal, 0);
93 init_completion(&sdp->sd_locking_init);
94 init_completion(&sdp->sd_wdack);
95 spin_lock_init(&sdp->sd_statfs_spin);
96
97 spin_lock_init(&sdp->sd_rindex_spin);
98 sdp->sd_rindex_tree.rb_node = NULL;
99
100 INIT_LIST_HEAD(&sdp->sd_jindex_list);
101 spin_lock_init(&sdp->sd_jindex_spin);
102 mutex_init(&sdp->sd_jindex_mutex);
103 init_completion(&sdp->sd_journal_ready);
104
105 INIT_LIST_HEAD(&sdp->sd_quota_list);
106 mutex_init(&sdp->sd_quota_mutex);
107 mutex_init(&sdp->sd_quota_sync_mutex);
108 init_waitqueue_head(&sdp->sd_quota_wait);
109 spin_lock_init(&sdp->sd_bitmap_lock);
110
111 INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
112
113 mapping = &sdp->sd_aspace;
114
115 address_space_init_once(mapping);
116 mapping->a_ops = &gfs2_rgrp_aops;
117 mapping->host = sb->s_bdev->bd_inode;
118 mapping->flags = 0;
119 mapping_set_gfp_mask(mapping, GFP_NOFS);
120 mapping->private_data = NULL;
121 mapping->writeback_index = 0;
122
123 spin_lock_init(&sdp->sd_log_lock);
124 atomic_set(&sdp->sd_log_pinned, 0);
125 INIT_LIST_HEAD(&sdp->sd_log_revokes);
126 INIT_LIST_HEAD(&sdp->sd_log_ordered);
127 spin_lock_init(&sdp->sd_ordered_lock);
128
129 init_waitqueue_head(&sdp->sd_log_waitq);
130 init_waitqueue_head(&sdp->sd_logd_waitq);
131 spin_lock_init(&sdp->sd_ail_lock);
132 INIT_LIST_HEAD(&sdp->sd_ail1_list);
133 INIT_LIST_HEAD(&sdp->sd_ail2_list);
134
135 init_rwsem(&sdp->sd_log_flush_lock);
136 atomic_set(&sdp->sd_log_in_flight, 0);
137 init_waitqueue_head(&sdp->sd_log_flush_wait);
138 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
139 mutex_init(&sdp->sd_freeze_mutex);
140
141 return sdp;
142
143 fail:
144 free_sbd(sdp);
145 return NULL;
146 }
147
148 /**
149 * gfs2_check_sb - Check superblock
150 * @sdp: the filesystem
151 * @silent: Don't print a message if the check fails
152 *
153 * Checks the version code of the FS is one that we understand how to
154 * read and that the sizes of the various on-disk structures have not
155 * changed.
156 */
157
158 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
159 {
160 struct gfs2_sb_host *sb = &sdp->sd_sb;
161
162 if (sb->sb_magic != GFS2_MAGIC ||
163 sb->sb_type != GFS2_METATYPE_SB) {
164 if (!silent)
165 pr_warn("not a GFS2 filesystem\n");
166 return -EINVAL;
167 }
168
169 if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
170 sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
171 sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
172 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
173 return -EINVAL;
174 }
175
176 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
177 (sb->sb_bsize & (sb->sb_bsize - 1))) {
178 pr_warn("Invalid block size\n");
179 return -EINVAL;
180 }
181
182 return 0;
183 }
184
185 static void end_bio_io_page(struct bio *bio)
186 {
187 struct page *page = bio->bi_private;
188
189 if (!bio->bi_status)
190 SetPageUptodate(page);
191 else
192 pr_warn("error %d reading superblock\n", bio->bi_status);
193 unlock_page(page);
194 }
195
196 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
197 {
198 struct gfs2_sb_host *sb = &sdp->sd_sb;
199 struct super_block *s = sdp->sd_vfs;
200 const struct gfs2_sb *str = buf;
201
202 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
203 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
204 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
205 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
206 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
207 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
208 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
209 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
210 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
211 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
212
213 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
214 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
215 memcpy(&s->s_uuid, str->sb_uuid, 16);
216 }
217
218 /**
219 * gfs2_read_super - Read the gfs2 super block from disk
220 * @sdp: The GFS2 super block
221 * @sector: The location of the super block
222 * @silent: Don't print a message if the check fails
223 *
224 * This uses the bio functions to read the super block from disk
225 * because we want to be 100% sure that we never read cached data.
226 * A super block is read twice only during each GFS2 mount and is
227 * never written to by the filesystem. The first time its read no
228 * locks are held, and the only details which are looked at are those
229 * relating to the locking protocol. Once locking is up and working,
230 * the sb is read again under the lock to establish the location of
231 * the master directory (contains pointers to journals etc) and the
232 * root directory.
233 *
234 * Returns: 0 on success or error
235 */
236
237 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
238 {
239 struct super_block *sb = sdp->sd_vfs;
240 struct gfs2_sb *p;
241 struct page *page;
242 struct bio *bio;
243
244 page = alloc_page(GFP_NOFS);
245 if (unlikely(!page))
246 return -ENOMEM;
247
248 ClearPageUptodate(page);
249 ClearPageDirty(page);
250 lock_page(page);
251
252 bio = bio_alloc(sb->s_bdev, 1, REQ_OP_READ | REQ_META, GFP_NOFS);
253 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
254 bio_add_page(bio, page, PAGE_SIZE, 0);
255
256 bio->bi_end_io = end_bio_io_page;
257 bio->bi_private = page;
258 submit_bio(bio);
259 wait_on_page_locked(page);
260 bio_put(bio);
261 if (!PageUptodate(page)) {
262 __free_page(page);
263 return -EIO;
264 }
265 p = kmap(page);
266 gfs2_sb_in(sdp, p);
267 kunmap(page);
268 __free_page(page);
269 return gfs2_check_sb(sdp, silent);
270 }
271
272 /**
273 * gfs2_read_sb - Read super block
274 * @sdp: The GFS2 superblock
275 * @silent: Don't print message if mount fails
276 *
277 */
278
279 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
280 {
281 u32 hash_blocks, ind_blocks, leaf_blocks;
282 u32 tmp_blocks;
283 unsigned int x;
284 int error;
285
286 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
287 if (error) {
288 if (!silent)
289 fs_err(sdp, "can't read superblock\n");
290 return error;
291 }
292
293 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
294 GFS2_BASIC_BLOCK_SHIFT;
295 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
296 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
297 sizeof(struct gfs2_dinode)) / sizeof(u64);
298 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
299 sizeof(struct gfs2_meta_header)) / sizeof(u64);
300 sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
301 sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
302 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
303 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
304 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
305 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
306 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
307 sizeof(struct gfs2_meta_header)) /
308 sizeof(struct gfs2_quota_change);
309 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
310 sizeof(struct gfs2_meta_header))
311 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
312
313 /*
314 * We always keep at least one block reserved for revokes in
315 * transactions. This greatly simplifies allocating additional
316 * revoke blocks.
317 */
318 atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
319
320 /* Compute maximum reservation required to add a entry to a directory */
321
322 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
323 sdp->sd_jbsize);
324
325 ind_blocks = 0;
326 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
327 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
328 ind_blocks += tmp_blocks;
329 }
330
331 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
332
333 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
334
335 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
336 sizeof(struct gfs2_dinode);
337 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
338 for (x = 2;; x++) {
339 u64 space, d;
340 u32 m;
341
342 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
343 d = space;
344 m = do_div(d, sdp->sd_inptrs);
345
346 if (d != sdp->sd_heightsize[x - 1] || m)
347 break;
348 sdp->sd_heightsize[x] = space;
349 }
350 sdp->sd_max_height = x;
351 sdp->sd_heightsize[x] = ~0;
352 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
353
354 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
355 sizeof(struct gfs2_leaf)) /
356 GFS2_MIN_DIRENT_SIZE;
357 return 0;
358 }
359
360 static int init_names(struct gfs2_sbd *sdp, int silent)
361 {
362 char *proto, *table;
363 int error = 0;
364
365 proto = sdp->sd_args.ar_lockproto;
366 table = sdp->sd_args.ar_locktable;
367
368 /* Try to autodetect */
369
370 if (!proto[0] || !table[0]) {
371 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
372 if (error)
373 return error;
374
375 if (!proto[0])
376 proto = sdp->sd_sb.sb_lockproto;
377 if (!table[0])
378 table = sdp->sd_sb.sb_locktable;
379 }
380
381 if (!table[0])
382 table = sdp->sd_vfs->s_id;
383
384 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
385 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
386
387 table = sdp->sd_table_name;
388 while ((table = strchr(table, '/')))
389 *table = '_';
390
391 return error;
392 }
393
394 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
395 int undo)
396 {
397 int error = 0;
398
399 if (undo)
400 goto fail_trans;
401
402 error = gfs2_glock_nq_num(sdp,
403 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
404 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
405 mount_gh);
406 if (error) {
407 fs_err(sdp, "can't acquire mount glock: %d\n", error);
408 goto fail;
409 }
410
411 error = gfs2_glock_nq_num(sdp,
412 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
413 LM_ST_SHARED,
414 LM_FLAG_NOEXP | GL_EXACT,
415 &sdp->sd_live_gh);
416 if (error) {
417 fs_err(sdp, "can't acquire live glock: %d\n", error);
418 goto fail_mount;
419 }
420
421 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
422 CREATE, &sdp->sd_rename_gl);
423 if (error) {
424 fs_err(sdp, "can't create rename glock: %d\n", error);
425 goto fail_live;
426 }
427
428 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
429 CREATE, &sdp->sd_freeze_gl);
430 if (error) {
431 fs_err(sdp, "can't create transaction glock: %d\n", error);
432 goto fail_rename;
433 }
434
435 return 0;
436
437 fail_trans:
438 gfs2_glock_put(sdp->sd_freeze_gl);
439 fail_rename:
440 gfs2_glock_put(sdp->sd_rename_gl);
441 fail_live:
442 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
443 fail_mount:
444 gfs2_glock_dq_uninit(mount_gh);
445 fail:
446 return error;
447 }
448
449 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
450 u64 no_addr, const char *name)
451 {
452 struct gfs2_sbd *sdp = sb->s_fs_info;
453 struct dentry *dentry;
454 struct inode *inode;
455
456 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
457 GFS2_BLKST_FREE /* ignore */);
458 if (IS_ERR(inode)) {
459 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
460 return PTR_ERR(inode);
461 }
462 dentry = d_make_root(inode);
463 if (!dentry) {
464 fs_err(sdp, "can't alloc %s dentry\n", name);
465 return -ENOMEM;
466 }
467 *dptr = dentry;
468 return 0;
469 }
470
471 static int init_sb(struct gfs2_sbd *sdp, int silent)
472 {
473 struct super_block *sb = sdp->sd_vfs;
474 struct gfs2_holder sb_gh;
475 u64 no_addr;
476 int ret;
477
478 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
479 LM_ST_SHARED, 0, &sb_gh);
480 if (ret) {
481 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
482 return ret;
483 }
484
485 ret = gfs2_read_sb(sdp, silent);
486 if (ret) {
487 fs_err(sdp, "can't read superblock: %d\n", ret);
488 goto out;
489 }
490
491 switch(sdp->sd_sb.sb_fs_format) {
492 case GFS2_FS_FORMAT_MAX:
493 sb->s_xattr = gfs2_xattr_handlers_max;
494 break;
495
496 case GFS2_FS_FORMAT_MIN:
497 sb->s_xattr = gfs2_xattr_handlers_min;
498 break;
499
500 default:
501 BUG();
502 }
503
504 /* Set up the buffer cache and SB for real */
505 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
506 ret = -EINVAL;
507 fs_err(sdp, "FS block size (%u) is too small for device "
508 "block size (%u)\n",
509 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
510 goto out;
511 }
512 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
513 ret = -EINVAL;
514 fs_err(sdp, "FS block size (%u) is too big for machine "
515 "page size (%u)\n",
516 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
517 goto out;
518 }
519 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
520
521 /* Get the root inode */
522 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
523 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
524 if (ret)
525 goto out;
526
527 /* Get the master inode */
528 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
529 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
530 if (ret) {
531 dput(sdp->sd_root_dir);
532 goto out;
533 }
534 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
535 out:
536 gfs2_glock_dq_uninit(&sb_gh);
537 return ret;
538 }
539
540 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
541 {
542 char *message = "FIRSTMOUNT=Done";
543 char *envp[] = { message, NULL };
544
545 fs_info(sdp, "first mount done, others may mount\n");
546
547 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
548 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
549
550 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
551 }
552
553 /**
554 * gfs2_jindex_hold - Grab a lock on the jindex
555 * @sdp: The GFS2 superblock
556 * @ji_gh: the holder for the jindex glock
557 *
558 * Returns: errno
559 */
560
561 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
562 {
563 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
564 struct qstr name;
565 char buf[20];
566 struct gfs2_jdesc *jd;
567 int error;
568
569 name.name = buf;
570
571 mutex_lock(&sdp->sd_jindex_mutex);
572
573 for (;;) {
574 struct gfs2_inode *jip;
575
576 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
577 if (error)
578 break;
579
580 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
581 name.hash = gfs2_disk_hash(name.name, name.len);
582
583 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
584 if (error == -ENOENT) {
585 error = 0;
586 break;
587 }
588
589 gfs2_glock_dq_uninit(ji_gh);
590
591 if (error)
592 break;
593
594 error = -ENOMEM;
595 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
596 if (!jd)
597 break;
598
599 INIT_LIST_HEAD(&jd->extent_list);
600 INIT_LIST_HEAD(&jd->jd_revoke_list);
601
602 INIT_WORK(&jd->jd_work, gfs2_recover_func);
603 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
604 if (IS_ERR_OR_NULL(jd->jd_inode)) {
605 if (!jd->jd_inode)
606 error = -ENOENT;
607 else
608 error = PTR_ERR(jd->jd_inode);
609 kfree(jd);
610 break;
611 }
612
613 d_mark_dontcache(jd->jd_inode);
614 spin_lock(&sdp->sd_jindex_spin);
615 jd->jd_jid = sdp->sd_journals++;
616 jip = GFS2_I(jd->jd_inode);
617 jd->jd_no_addr = jip->i_no_addr;
618 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
619 spin_unlock(&sdp->sd_jindex_spin);
620 }
621
622 mutex_unlock(&sdp->sd_jindex_mutex);
623
624 return error;
625 }
626
627 /**
628 * init_statfs - look up and initialize master and local (per node) statfs inodes
629 * @sdp: The GFS2 superblock
630 *
631 * This should be called after the jindex is initialized in init_journal() and
632 * before gfs2_journal_recovery() is called because we need to be able to write
633 * to these inodes during recovery.
634 *
635 * Returns: errno
636 */
637 static int init_statfs(struct gfs2_sbd *sdp)
638 {
639 int error = 0;
640 struct inode *master = d_inode(sdp->sd_master_dir);
641 struct inode *pn = NULL;
642 char buf[30];
643 struct gfs2_jdesc *jd;
644 struct gfs2_inode *ip;
645
646 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
647 if (IS_ERR(sdp->sd_statfs_inode)) {
648 error = PTR_ERR(sdp->sd_statfs_inode);
649 fs_err(sdp, "can't read in statfs inode: %d\n", error);
650 goto out;
651 }
652 if (sdp->sd_args.ar_spectator)
653 goto out;
654
655 pn = gfs2_lookup_simple(master, "per_node");
656 if (IS_ERR(pn)) {
657 error = PTR_ERR(pn);
658 fs_err(sdp, "can't find per_node directory: %d\n", error);
659 goto put_statfs;
660 }
661
662 /* For each jid, lookup the corresponding local statfs inode in the
663 * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
664 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
665 struct local_statfs_inode *lsi =
666 kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
667 if (!lsi) {
668 error = -ENOMEM;
669 goto free_local;
670 }
671 sprintf(buf, "statfs_change%u", jd->jd_jid);
672 lsi->si_sc_inode = gfs2_lookup_simple(pn, buf);
673 if (IS_ERR(lsi->si_sc_inode)) {
674 error = PTR_ERR(lsi->si_sc_inode);
675 fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
676 jd->jd_jid, error);
677 kfree(lsi);
678 goto free_local;
679 }
680 lsi->si_jid = jd->jd_jid;
681 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
682 sdp->sd_sc_inode = lsi->si_sc_inode;
683
684 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
685 }
686
687 iput(pn);
688 pn = NULL;
689 ip = GFS2_I(sdp->sd_sc_inode);
690 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
691 &sdp->sd_sc_gh);
692 if (error) {
693 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
694 goto free_local;
695 }
696 /* read in the local statfs buffer - other nodes don't change it. */
697 error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
698 if (error) {
699 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
700 goto unlock_sd_gh;
701 }
702 return 0;
703
704 unlock_sd_gh:
705 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
706 free_local:
707 free_local_statfs_inodes(sdp);
708 iput(pn);
709 put_statfs:
710 iput(sdp->sd_statfs_inode);
711 out:
712 return error;
713 }
714
715 /* Uninitialize and free up memory used by the list of statfs inodes */
716 static void uninit_statfs(struct gfs2_sbd *sdp)
717 {
718 if (!sdp->sd_args.ar_spectator) {
719 brelse(sdp->sd_sc_bh);
720 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
721 free_local_statfs_inodes(sdp);
722 }
723 iput(sdp->sd_statfs_inode);
724 }
725
726 static int init_journal(struct gfs2_sbd *sdp, int undo)
727 {
728 struct inode *master = d_inode(sdp->sd_master_dir);
729 struct gfs2_holder ji_gh;
730 struct gfs2_inode *ip;
731 int jindex = 1;
732 int error = 0;
733
734 if (undo) {
735 jindex = 0;
736 goto fail_statfs;
737 }
738
739 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
740 if (IS_ERR(sdp->sd_jindex)) {
741 fs_err(sdp, "can't lookup journal index: %d\n", error);
742 return PTR_ERR(sdp->sd_jindex);
743 }
744
745 /* Load in the journal index special file */
746
747 error = gfs2_jindex_hold(sdp, &ji_gh);
748 if (error) {
749 fs_err(sdp, "can't read journal index: %d\n", error);
750 goto fail;
751 }
752
753 error = -EUSERS;
754 if (!gfs2_jindex_size(sdp)) {
755 fs_err(sdp, "no journals!\n");
756 goto fail_jindex;
757 }
758
759 atomic_set(&sdp->sd_log_blks_needed, 0);
760 if (sdp->sd_args.ar_spectator) {
761 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
762 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
763 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
764 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
765 } else {
766 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
767 fs_err(sdp, "can't mount journal #%u\n",
768 sdp->sd_lockstruct.ls_jid);
769 fs_err(sdp, "there are only %u journals (0 - %u)\n",
770 gfs2_jindex_size(sdp),
771 gfs2_jindex_size(sdp) - 1);
772 goto fail_jindex;
773 }
774 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
775
776 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
777 &gfs2_journal_glops,
778 LM_ST_EXCLUSIVE,
779 LM_FLAG_NOEXP | GL_NOCACHE,
780 &sdp->sd_journal_gh);
781 if (error) {
782 fs_err(sdp, "can't acquire journal glock: %d\n", error);
783 goto fail_jindex;
784 }
785
786 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
787 sdp->sd_jinode_gl = ip->i_gl;
788 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
789 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
790 &sdp->sd_jinode_gh);
791 if (error) {
792 fs_err(sdp, "can't acquire journal inode glock: %d\n",
793 error);
794 goto fail_journal_gh;
795 }
796
797 error = gfs2_jdesc_check(sdp->sd_jdesc);
798 if (error) {
799 fs_err(sdp, "my journal (%u) is bad: %d\n",
800 sdp->sd_jdesc->jd_jid, error);
801 goto fail_jinode_gh;
802 }
803 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
804 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
805 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
806
807 /* Map the extents for this journal's blocks */
808 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
809 }
810 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
811
812 /* Lookup statfs inodes here so journal recovery can use them. */
813 error = init_statfs(sdp);
814 if (error)
815 goto fail_jinode_gh;
816
817 if (sdp->sd_lockstruct.ls_first) {
818 unsigned int x;
819 for (x = 0; x < sdp->sd_journals; x++) {
820 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
821
822 if (sdp->sd_args.ar_spectator) {
823 error = check_journal_clean(sdp, jd, true);
824 if (error)
825 goto fail_statfs;
826 continue;
827 }
828 error = gfs2_recover_journal(jd, true);
829 if (error) {
830 fs_err(sdp, "error recovering journal %u: %d\n",
831 x, error);
832 goto fail_statfs;
833 }
834 }
835
836 gfs2_others_may_mount(sdp);
837 } else if (!sdp->sd_args.ar_spectator) {
838 error = gfs2_recover_journal(sdp->sd_jdesc, true);
839 if (error) {
840 fs_err(sdp, "error recovering my journal: %d\n", error);
841 goto fail_statfs;
842 }
843 }
844
845 sdp->sd_log_idle = 1;
846 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
847 gfs2_glock_dq_uninit(&ji_gh);
848 jindex = 0;
849 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
850 return 0;
851
852 fail_statfs:
853 uninit_statfs(sdp);
854 fail_jinode_gh:
855 /* A withdraw may have done dq/uninit so now we need to check it */
856 if (!sdp->sd_args.ar_spectator &&
857 gfs2_holder_initialized(&sdp->sd_jinode_gh))
858 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
859 fail_journal_gh:
860 if (!sdp->sd_args.ar_spectator &&
861 gfs2_holder_initialized(&sdp->sd_journal_gh))
862 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
863 fail_jindex:
864 gfs2_jindex_free(sdp);
865 if (jindex)
866 gfs2_glock_dq_uninit(&ji_gh);
867 fail:
868 iput(sdp->sd_jindex);
869 return error;
870 }
871
872 static struct lock_class_key gfs2_quota_imutex_key;
873
874 static int init_inodes(struct gfs2_sbd *sdp, int undo)
875 {
876 int error = 0;
877 struct inode *master = d_inode(sdp->sd_master_dir);
878
879 if (undo)
880 goto fail_qinode;
881
882 error = init_journal(sdp, undo);
883 complete_all(&sdp->sd_journal_ready);
884 if (error)
885 goto fail;
886
887 /* Read in the resource index inode */
888 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
889 if (IS_ERR(sdp->sd_rindex)) {
890 error = PTR_ERR(sdp->sd_rindex);
891 fs_err(sdp, "can't get resource index inode: %d\n", error);
892 goto fail_journal;
893 }
894 sdp->sd_rindex_uptodate = 0;
895
896 /* Read in the quota inode */
897 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
898 if (IS_ERR(sdp->sd_quota_inode)) {
899 error = PTR_ERR(sdp->sd_quota_inode);
900 fs_err(sdp, "can't get quota file inode: %d\n", error);
901 goto fail_rindex;
902 }
903 /*
904 * i_rwsem on quota files is special. Since this inode is hidden system
905 * file, we are safe to define locking ourselves.
906 */
907 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
908 &gfs2_quota_imutex_key);
909
910 error = gfs2_rindex_update(sdp);
911 if (error)
912 goto fail_qinode;
913
914 return 0;
915
916 fail_qinode:
917 iput(sdp->sd_quota_inode);
918 fail_rindex:
919 gfs2_clear_rgrpd(sdp);
920 iput(sdp->sd_rindex);
921 fail_journal:
922 init_journal(sdp, UNDO);
923 fail:
924 return error;
925 }
926
927 static int init_per_node(struct gfs2_sbd *sdp, int undo)
928 {
929 struct inode *pn = NULL;
930 char buf[30];
931 int error = 0;
932 struct gfs2_inode *ip;
933 struct inode *master = d_inode(sdp->sd_master_dir);
934
935 if (sdp->sd_args.ar_spectator)
936 return 0;
937
938 if (undo)
939 goto fail_qc_gh;
940
941 pn = gfs2_lookup_simple(master, "per_node");
942 if (IS_ERR(pn)) {
943 error = PTR_ERR(pn);
944 fs_err(sdp, "can't find per_node directory: %d\n", error);
945 return error;
946 }
947
948 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
949 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
950 if (IS_ERR(sdp->sd_qc_inode)) {
951 error = PTR_ERR(sdp->sd_qc_inode);
952 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
953 goto fail_ut_i;
954 }
955
956 iput(pn);
957 pn = NULL;
958
959 ip = GFS2_I(sdp->sd_qc_inode);
960 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
961 &sdp->sd_qc_gh);
962 if (error) {
963 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
964 goto fail_qc_i;
965 }
966
967 return 0;
968
969 fail_qc_gh:
970 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
971 fail_qc_i:
972 iput(sdp->sd_qc_inode);
973 fail_ut_i:
974 iput(pn);
975 return error;
976 }
977
978 static const match_table_t nolock_tokens = {
979 { Opt_jid, "jid=%d", },
980 { Opt_err, NULL },
981 };
982
983 static const struct lm_lockops nolock_ops = {
984 .lm_proto_name = "lock_nolock",
985 .lm_put_lock = gfs2_glock_free,
986 .lm_tokens = &nolock_tokens,
987 };
988
989 /**
990 * gfs2_lm_mount - mount a locking protocol
991 * @sdp: the filesystem
992 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
993 *
994 * Returns: errno
995 */
996
997 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
998 {
999 const struct lm_lockops *lm;
1000 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
1001 struct gfs2_args *args = &sdp->sd_args;
1002 const char *proto = sdp->sd_proto_name;
1003 const char *table = sdp->sd_table_name;
1004 char *o, *options;
1005 int ret;
1006
1007 if (!strcmp("lock_nolock", proto)) {
1008 lm = &nolock_ops;
1009 sdp->sd_args.ar_localflocks = 1;
1010 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1011 } else if (!strcmp("lock_dlm", proto)) {
1012 lm = &gfs2_dlm_ops;
1013 #endif
1014 } else {
1015 pr_info("can't find protocol %s\n", proto);
1016 return -ENOENT;
1017 }
1018
1019 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1020
1021 ls->ls_ops = lm;
1022 ls->ls_first = 1;
1023
1024 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1025 substring_t tmp[MAX_OPT_ARGS];
1026 int token, option;
1027
1028 if (!o || !*o)
1029 continue;
1030
1031 token = match_token(o, *lm->lm_tokens, tmp);
1032 switch (token) {
1033 case Opt_jid:
1034 ret = match_int(&tmp[0], &option);
1035 if (ret || option < 0)
1036 goto hostdata_error;
1037 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1038 ls->ls_jid = option;
1039 break;
1040 case Opt_id:
1041 case Opt_nodir:
1042 /* Obsolete, but left for backward compat purposes */
1043 break;
1044 case Opt_first:
1045 ret = match_int(&tmp[0], &option);
1046 if (ret || (option != 0 && option != 1))
1047 goto hostdata_error;
1048 ls->ls_first = option;
1049 break;
1050 case Opt_err:
1051 default:
1052 hostdata_error:
1053 fs_info(sdp, "unknown hostdata (%s)\n", o);
1054 return -EINVAL;
1055 }
1056 }
1057
1058 if (lm->lm_mount == NULL) {
1059 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1060 complete_all(&sdp->sd_locking_init);
1061 return 0;
1062 }
1063 ret = lm->lm_mount(sdp, table);
1064 if (ret == 0)
1065 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1066 sdp->sd_sb.sb_fs_format);
1067 complete_all(&sdp->sd_locking_init);
1068 return ret;
1069 }
1070
1071 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1072 {
1073 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1074 if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1075 lm->lm_unmount(sdp);
1076 }
1077
1078 static int wait_on_journal(struct gfs2_sbd *sdp)
1079 {
1080 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1081 return 0;
1082
1083 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1084 ? -EINTR : 0;
1085 }
1086
1087 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1088 {
1089 struct super_block *sb = sdp->sd_vfs;
1090 char ro[20];
1091 char spectator[20];
1092 char *envp[] = { ro, spectator, NULL };
1093 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1094 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1095 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1096 }
1097
1098 static int init_threads(struct gfs2_sbd *sdp)
1099 {
1100 struct task_struct *p;
1101 int error = 0;
1102
1103 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
1104 if (IS_ERR(p)) {
1105 error = PTR_ERR(p);
1106 fs_err(sdp, "can't start logd thread: %d\n", error);
1107 return error;
1108 }
1109 sdp->sd_logd_process = p;
1110
1111 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
1112 if (IS_ERR(p)) {
1113 error = PTR_ERR(p);
1114 fs_err(sdp, "can't start quotad thread: %d\n", error);
1115 goto fail;
1116 }
1117 sdp->sd_quotad_process = p;
1118 return 0;
1119
1120 fail:
1121 kthread_stop(sdp->sd_logd_process);
1122 sdp->sd_logd_process = NULL;
1123 return error;
1124 }
1125
1126 /**
1127 * gfs2_fill_super - Read in superblock
1128 * @sb: The VFS superblock
1129 * @fc: Mount options and flags
1130 *
1131 * Returns: -errno
1132 */
1133 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1134 {
1135 struct gfs2_args *args = fc->fs_private;
1136 int silent = fc->sb_flags & SB_SILENT;
1137 struct gfs2_sbd *sdp;
1138 struct gfs2_holder mount_gh;
1139 struct gfs2_holder freeze_gh;
1140 int error;
1141
1142 sdp = init_sbd(sb);
1143 if (!sdp) {
1144 pr_warn("can't alloc struct gfs2_sbd\n");
1145 return -ENOMEM;
1146 }
1147 sdp->sd_args = *args;
1148
1149 if (sdp->sd_args.ar_spectator) {
1150 sb->s_flags |= SB_RDONLY;
1151 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1152 }
1153 if (sdp->sd_args.ar_posix_acl)
1154 sb->s_flags |= SB_POSIXACL;
1155 if (sdp->sd_args.ar_nobarrier)
1156 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1157
1158 sb->s_flags |= SB_NOSEC;
1159 sb->s_magic = GFS2_MAGIC;
1160 sb->s_op = &gfs2_super_ops;
1161 sb->s_d_op = &gfs2_dops;
1162 sb->s_export_op = &gfs2_export_ops;
1163 sb->s_qcop = &gfs2_quotactl_ops;
1164 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1165 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1166 sb->s_time_gran = 1;
1167 sb->s_maxbytes = MAX_LFS_FILESIZE;
1168
1169 /* Set up the buffer cache and fill in some fake block size values
1170 to allow us to read-in the on-disk superblock. */
1171 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1172 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1173 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1174 GFS2_BASIC_BLOCK_SHIFT;
1175 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1176
1177 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1178 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1179 if (sdp->sd_args.ar_statfs_quantum) {
1180 sdp->sd_tune.gt_statfs_slow = 0;
1181 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1182 } else {
1183 sdp->sd_tune.gt_statfs_slow = 1;
1184 sdp->sd_tune.gt_statfs_quantum = 30;
1185 }
1186
1187 error = init_names(sdp, silent);
1188 if (error)
1189 goto fail_free;
1190
1191 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1192
1193 error = gfs2_sys_fs_add(sdp);
1194 if (error)
1195 goto fail_free;
1196
1197 gfs2_create_debugfs_file(sdp);
1198
1199 error = gfs2_lm_mount(sdp, silent);
1200 if (error)
1201 goto fail_debug;
1202
1203 error = init_locking(sdp, &mount_gh, DO);
1204 if (error)
1205 goto fail_lm;
1206
1207 error = init_sb(sdp, silent);
1208 if (error)
1209 goto fail_locking;
1210
1211 /* Turn rgrplvb on by default if fs format is recent enough */
1212 if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1213 sdp->sd_args.ar_rgrplvb = 1;
1214
1215 error = wait_on_journal(sdp);
1216 if (error)
1217 goto fail_sb;
1218
1219 /*
1220 * If user space has failed to join the cluster or some similar
1221 * failure has occurred, then the journal id will contain a
1222 * negative (error) number. This will then be returned to the
1223 * caller (of the mount syscall). We do this even for spectator
1224 * mounts (which just write a jid of 0 to indicate "ok" even though
1225 * the jid is unused in the spectator case)
1226 */
1227 if (sdp->sd_lockstruct.ls_jid < 0) {
1228 error = sdp->sd_lockstruct.ls_jid;
1229 sdp->sd_lockstruct.ls_jid = 0;
1230 goto fail_sb;
1231 }
1232
1233 if (sdp->sd_args.ar_spectator)
1234 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1235 sdp->sd_table_name);
1236 else
1237 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1238 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1239
1240 error = init_inodes(sdp, DO);
1241 if (error)
1242 goto fail_sb;
1243
1244 error = init_per_node(sdp, DO);
1245 if (error)
1246 goto fail_inodes;
1247
1248 error = gfs2_statfs_init(sdp);
1249 if (error) {
1250 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1251 goto fail_per_node;
1252 }
1253
1254 if (!sb_rdonly(sb)) {
1255 error = init_threads(sdp);
1256 if (error) {
1257 gfs2_withdraw_delayed(sdp);
1258 goto fail_per_node;
1259 }
1260 }
1261
1262 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1263 if (error)
1264 goto fail_per_node;
1265
1266 if (!sb_rdonly(sb))
1267 error = gfs2_make_fs_rw(sdp);
1268
1269 gfs2_freeze_unlock(&freeze_gh);
1270 if (error) {
1271 if (sdp->sd_quotad_process)
1272 kthread_stop(sdp->sd_quotad_process);
1273 sdp->sd_quotad_process = NULL;
1274 if (sdp->sd_logd_process)
1275 kthread_stop(sdp->sd_logd_process);
1276 sdp->sd_logd_process = NULL;
1277 fs_err(sdp, "can't make FS RW: %d\n", error);
1278 goto fail_per_node;
1279 }
1280 gfs2_glock_dq_uninit(&mount_gh);
1281 gfs2_online_uevent(sdp);
1282 return 0;
1283
1284 fail_per_node:
1285 init_per_node(sdp, UNDO);
1286 fail_inodes:
1287 init_inodes(sdp, UNDO);
1288 fail_sb:
1289 if (sdp->sd_root_dir)
1290 dput(sdp->sd_root_dir);
1291 if (sdp->sd_master_dir)
1292 dput(sdp->sd_master_dir);
1293 if (sb->s_root)
1294 dput(sb->s_root);
1295 sb->s_root = NULL;
1296 fail_locking:
1297 init_locking(sdp, &mount_gh, UNDO);
1298 fail_lm:
1299 complete_all(&sdp->sd_journal_ready);
1300 gfs2_gl_hash_clear(sdp);
1301 gfs2_lm_unmount(sdp);
1302 fail_debug:
1303 gfs2_delete_debugfs_file(sdp);
1304 gfs2_sys_fs_del(sdp);
1305 fail_free:
1306 free_sbd(sdp);
1307 sb->s_fs_info = NULL;
1308 return error;
1309 }
1310
1311 /**
1312 * gfs2_get_tree - Get the GFS2 superblock and root directory
1313 * @fc: The filesystem context
1314 *
1315 * Returns: 0 or -errno on error
1316 */
1317 static int gfs2_get_tree(struct fs_context *fc)
1318 {
1319 struct gfs2_args *args = fc->fs_private;
1320 struct gfs2_sbd *sdp;
1321 int error;
1322
1323 error = get_tree_bdev(fc, gfs2_fill_super);
1324 if (error)
1325 return error;
1326
1327 sdp = fc->root->d_sb->s_fs_info;
1328 dput(fc->root);
1329 if (args->ar_meta)
1330 fc->root = dget(sdp->sd_master_dir);
1331 else
1332 fc->root = dget(sdp->sd_root_dir);
1333 return 0;
1334 }
1335
1336 static void gfs2_fc_free(struct fs_context *fc)
1337 {
1338 struct gfs2_args *args = fc->fs_private;
1339
1340 kfree(args);
1341 }
1342
1343 enum gfs2_param {
1344 Opt_lockproto,
1345 Opt_locktable,
1346 Opt_hostdata,
1347 Opt_spectator,
1348 Opt_ignore_local_fs,
1349 Opt_localflocks,
1350 Opt_localcaching,
1351 Opt_debug,
1352 Opt_upgrade,
1353 Opt_acl,
1354 Opt_quota,
1355 Opt_quota_flag,
1356 Opt_suiddir,
1357 Opt_data,
1358 Opt_meta,
1359 Opt_discard,
1360 Opt_commit,
1361 Opt_errors,
1362 Opt_statfs_quantum,
1363 Opt_statfs_percent,
1364 Opt_quota_quantum,
1365 Opt_barrier,
1366 Opt_rgrplvb,
1367 Opt_loccookie,
1368 };
1369
1370 static const struct constant_table gfs2_param_quota[] = {
1371 {"off", GFS2_QUOTA_OFF},
1372 {"account", GFS2_QUOTA_ACCOUNT},
1373 {"on", GFS2_QUOTA_ON},
1374 {}
1375 };
1376
1377 enum opt_data {
1378 Opt_data_writeback = GFS2_DATA_WRITEBACK,
1379 Opt_data_ordered = GFS2_DATA_ORDERED,
1380 };
1381
1382 static const struct constant_table gfs2_param_data[] = {
1383 {"writeback", Opt_data_writeback },
1384 {"ordered", Opt_data_ordered },
1385 {}
1386 };
1387
1388 enum opt_errors {
1389 Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1390 Opt_errors_panic = GFS2_ERRORS_PANIC,
1391 };
1392
1393 static const struct constant_table gfs2_param_errors[] = {
1394 {"withdraw", Opt_errors_withdraw },
1395 {"panic", Opt_errors_panic },
1396 {}
1397 };
1398
1399 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1400 fsparam_string ("lockproto", Opt_lockproto),
1401 fsparam_string ("locktable", Opt_locktable),
1402 fsparam_string ("hostdata", Opt_hostdata),
1403 fsparam_flag ("spectator", Opt_spectator),
1404 fsparam_flag ("norecovery", Opt_spectator),
1405 fsparam_flag ("ignore_local_fs", Opt_ignore_local_fs),
1406 fsparam_flag ("localflocks", Opt_localflocks),
1407 fsparam_flag ("localcaching", Opt_localcaching),
1408 fsparam_flag_no("debug", Opt_debug),
1409 fsparam_flag ("upgrade", Opt_upgrade),
1410 fsparam_flag_no("acl", Opt_acl),
1411 fsparam_flag_no("suiddir", Opt_suiddir),
1412 fsparam_enum ("data", Opt_data, gfs2_param_data),
1413 fsparam_flag ("meta", Opt_meta),
1414 fsparam_flag_no("discard", Opt_discard),
1415 fsparam_s32 ("commit", Opt_commit),
1416 fsparam_enum ("errors", Opt_errors, gfs2_param_errors),
1417 fsparam_s32 ("statfs_quantum", Opt_statfs_quantum),
1418 fsparam_s32 ("statfs_percent", Opt_statfs_percent),
1419 fsparam_s32 ("quota_quantum", Opt_quota_quantum),
1420 fsparam_flag_no("barrier", Opt_barrier),
1421 fsparam_flag_no("rgrplvb", Opt_rgrplvb),
1422 fsparam_flag_no("loccookie", Opt_loccookie),
1423 /* quota can be a flag or an enum so it gets special treatment */
1424 fsparam_flag_no("quota", Opt_quota_flag),
1425 fsparam_enum("quota", Opt_quota, gfs2_param_quota),
1426 {}
1427 };
1428
1429 /* Parse a single mount parameter */
1430 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1431 {
1432 struct gfs2_args *args = fc->fs_private;
1433 struct fs_parse_result result;
1434 int o;
1435
1436 o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1437 if (o < 0)
1438 return o;
1439
1440 switch (o) {
1441 case Opt_lockproto:
1442 strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1443 break;
1444 case Opt_locktable:
1445 strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1446 break;
1447 case Opt_hostdata:
1448 strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1449 break;
1450 case Opt_spectator:
1451 args->ar_spectator = 1;
1452 break;
1453 case Opt_ignore_local_fs:
1454 /* Retained for backwards compat only */
1455 break;
1456 case Opt_localflocks:
1457 args->ar_localflocks = 1;
1458 break;
1459 case Opt_localcaching:
1460 /* Retained for backwards compat only */
1461 break;
1462 case Opt_debug:
1463 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1464 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1465 args->ar_debug = result.boolean;
1466 break;
1467 case Opt_upgrade:
1468 /* Retained for backwards compat only */
1469 break;
1470 case Opt_acl:
1471 args->ar_posix_acl = result.boolean;
1472 break;
1473 case Opt_quota_flag:
1474 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1475 break;
1476 case Opt_quota:
1477 args->ar_quota = result.int_32;
1478 break;
1479 case Opt_suiddir:
1480 args->ar_suiddir = result.boolean;
1481 break;
1482 case Opt_data:
1483 /* The uint_32 result maps directly to GFS2_DATA_* */
1484 args->ar_data = result.uint_32;
1485 break;
1486 case Opt_meta:
1487 args->ar_meta = 1;
1488 break;
1489 case Opt_discard:
1490 args->ar_discard = result.boolean;
1491 break;
1492 case Opt_commit:
1493 if (result.int_32 <= 0)
1494 return invalfc(fc, "commit mount option requires a positive numeric argument");
1495 args->ar_commit = result.int_32;
1496 break;
1497 case Opt_statfs_quantum:
1498 if (result.int_32 < 0)
1499 return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1500 args->ar_statfs_quantum = result.int_32;
1501 break;
1502 case Opt_quota_quantum:
1503 if (result.int_32 <= 0)
1504 return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1505 args->ar_quota_quantum = result.int_32;
1506 break;
1507 case Opt_statfs_percent:
1508 if (result.int_32 < 0 || result.int_32 > 100)
1509 return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1510 args->ar_statfs_percent = result.int_32;
1511 break;
1512 case Opt_errors:
1513 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1514 return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1515 args->ar_errors = result.uint_32;
1516 break;
1517 case Opt_barrier:
1518 args->ar_nobarrier = result.boolean;
1519 break;
1520 case Opt_rgrplvb:
1521 args->ar_rgrplvb = result.boolean;
1522 args->ar_got_rgrplvb = 1;
1523 break;
1524 case Opt_loccookie:
1525 args->ar_loccookie = result.boolean;
1526 break;
1527 default:
1528 return invalfc(fc, "invalid mount option: %s", param->key);
1529 }
1530 return 0;
1531 }
1532
1533 static int gfs2_reconfigure(struct fs_context *fc)
1534 {
1535 struct super_block *sb = fc->root->d_sb;
1536 struct gfs2_sbd *sdp = sb->s_fs_info;
1537 struct gfs2_args *oldargs = &sdp->sd_args;
1538 struct gfs2_args *newargs = fc->fs_private;
1539 struct gfs2_tune *gt = &sdp->sd_tune;
1540 int error = 0;
1541
1542 sync_filesystem(sb);
1543
1544 spin_lock(&gt->gt_spin);
1545 oldargs->ar_commit = gt->gt_logd_secs;
1546 oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1547 if (gt->gt_statfs_slow)
1548 oldargs->ar_statfs_quantum = 0;
1549 else
1550 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1551 spin_unlock(&gt->gt_spin);
1552
1553 if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1554 errorfc(fc, "reconfiguration of locking protocol not allowed");
1555 return -EINVAL;
1556 }
1557 if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1558 errorfc(fc, "reconfiguration of lock table not allowed");
1559 return -EINVAL;
1560 }
1561 if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1562 errorfc(fc, "reconfiguration of host data not allowed");
1563 return -EINVAL;
1564 }
1565 if (newargs->ar_spectator != oldargs->ar_spectator) {
1566 errorfc(fc, "reconfiguration of spectator mode not allowed");
1567 return -EINVAL;
1568 }
1569 if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1570 errorfc(fc, "reconfiguration of localflocks not allowed");
1571 return -EINVAL;
1572 }
1573 if (newargs->ar_meta != oldargs->ar_meta) {
1574 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1575 return -EINVAL;
1576 }
1577 if (oldargs->ar_spectator)
1578 fc->sb_flags |= SB_RDONLY;
1579
1580 if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1581 struct gfs2_holder freeze_gh;
1582
1583 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
1584 if (error)
1585 return -EINVAL;
1586
1587 if (fc->sb_flags & SB_RDONLY) {
1588 gfs2_make_fs_ro(sdp);
1589 } else {
1590 error = gfs2_make_fs_rw(sdp);
1591 if (error)
1592 errorfc(fc, "unable to remount read-write");
1593 }
1594 gfs2_freeze_unlock(&freeze_gh);
1595 }
1596 sdp->sd_args = *newargs;
1597
1598 if (sdp->sd_args.ar_posix_acl)
1599 sb->s_flags |= SB_POSIXACL;
1600 else
1601 sb->s_flags &= ~SB_POSIXACL;
1602 if (sdp->sd_args.ar_nobarrier)
1603 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1604 else
1605 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1606 spin_lock(&gt->gt_spin);
1607 gt->gt_logd_secs = newargs->ar_commit;
1608 gt->gt_quota_quantum = newargs->ar_quota_quantum;
1609 if (newargs->ar_statfs_quantum) {
1610 gt->gt_statfs_slow = 0;
1611 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1612 }
1613 else {
1614 gt->gt_statfs_slow = 1;
1615 gt->gt_statfs_quantum = 30;
1616 }
1617 spin_unlock(&gt->gt_spin);
1618
1619 gfs2_online_uevent(sdp);
1620 return error;
1621 }
1622
1623 static const struct fs_context_operations gfs2_context_ops = {
1624 .free = gfs2_fc_free,
1625 .parse_param = gfs2_parse_param,
1626 .get_tree = gfs2_get_tree,
1627 .reconfigure = gfs2_reconfigure,
1628 };
1629
1630 /* Set up the filesystem mount context */
1631 static int gfs2_init_fs_context(struct fs_context *fc)
1632 {
1633 struct gfs2_args *args;
1634
1635 args = kmalloc(sizeof(*args), GFP_KERNEL);
1636 if (args == NULL)
1637 return -ENOMEM;
1638
1639 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1640 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1641
1642 *args = sdp->sd_args;
1643 } else {
1644 memset(args, 0, sizeof(*args));
1645 args->ar_quota = GFS2_QUOTA_DEFAULT;
1646 args->ar_data = GFS2_DATA_DEFAULT;
1647 args->ar_commit = 30;
1648 args->ar_statfs_quantum = 30;
1649 args->ar_quota_quantum = 60;
1650 args->ar_errors = GFS2_ERRORS_DEFAULT;
1651 }
1652 fc->fs_private = args;
1653 fc->ops = &gfs2_context_ops;
1654 return 0;
1655 }
1656
1657 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1658 {
1659 return -EINVAL;
1660 }
1661
1662 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1663 {
1664 return (fc->sget_key == s->s_bdev);
1665 }
1666
1667 static int gfs2_meta_get_tree(struct fs_context *fc)
1668 {
1669 struct super_block *s;
1670 struct gfs2_sbd *sdp;
1671 struct path path;
1672 int error;
1673
1674 if (!fc->source || !*fc->source)
1675 return -EINVAL;
1676
1677 error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1678 if (error) {
1679 pr_warn("path_lookup on %s returned error %d\n",
1680 fc->source, error);
1681 return error;
1682 }
1683 fc->fs_type = &gfs2_fs_type;
1684 fc->sget_key = path.dentry->d_sb->s_bdev;
1685 s = sget_fc(fc, test_meta_super, set_meta_super);
1686 path_put(&path);
1687 if (IS_ERR(s)) {
1688 pr_warn("gfs2 mount does not exist\n");
1689 return PTR_ERR(s);
1690 }
1691 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1692 deactivate_locked_super(s);
1693 return -EBUSY;
1694 }
1695 sdp = s->s_fs_info;
1696 fc->root = dget(sdp->sd_master_dir);
1697 return 0;
1698 }
1699
1700 static const struct fs_context_operations gfs2_meta_context_ops = {
1701 .free = gfs2_fc_free,
1702 .get_tree = gfs2_meta_get_tree,
1703 };
1704
1705 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1706 {
1707 int ret = gfs2_init_fs_context(fc);
1708
1709 if (ret)
1710 return ret;
1711
1712 fc->ops = &gfs2_meta_context_ops;
1713 return 0;
1714 }
1715
1716 static void gfs2_kill_sb(struct super_block *sb)
1717 {
1718 struct gfs2_sbd *sdp = sb->s_fs_info;
1719
1720 if (sdp == NULL) {
1721 kill_block_super(sb);
1722 return;
1723 }
1724
1725 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1726 dput(sdp->sd_root_dir);
1727 dput(sdp->sd_master_dir);
1728 sdp->sd_root_dir = NULL;
1729 sdp->sd_master_dir = NULL;
1730 shrink_dcache_sb(sb);
1731 kill_block_super(sb);
1732 }
1733
1734 struct file_system_type gfs2_fs_type = {
1735 .name = "gfs2",
1736 .fs_flags = FS_REQUIRES_DEV,
1737 .init_fs_context = gfs2_init_fs_context,
1738 .parameters = gfs2_fs_parameters,
1739 .kill_sb = gfs2_kill_sb,
1740 .owner = THIS_MODULE,
1741 };
1742 MODULE_ALIAS_FS("gfs2");
1743
1744 struct file_system_type gfs2meta_fs_type = {
1745 .name = "gfs2meta",
1746 .fs_flags = FS_REQUIRES_DEV,
1747 .init_fs_context = gfs2_meta_init_fs_context,
1748 .owner = THIS_MODULE,
1749 };
1750 MODULE_ALIAS_FS("gfs2meta");
Michael's Linux tree.