GFS2: Send some sensible sysfs stuff
[linux-2.6-block.git] / fs / gfs2 / ops_fstype.c
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/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/lm_interface.h>
21
22 #include "gfs2.h"
23 #include "incore.h"
24 #include "bmap.h"
25 #include "glock.h"
26 #include "glops.h"
27 #include "inode.h"
28 #include "mount.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
37 #define DO 0
38 #define UNDO 1
39
40 static const u32 gfs2_old_fs_formats[] = {
41         0
42 };
43
44 static const u32 gfs2_old_multihost_formats[] = {
45         0
46 };
47
48 /**
49  * gfs2_tune_init - Fill a gfs2_tune structure with default values
50  * @gt: tune
51  *
52  */
53
54 static void gfs2_tune_init(struct gfs2_tune *gt)
55 {
56         spin_lock_init(&gt->gt_spin);
57
58         gt->gt_incore_log_blocks = 1024;
59         gt->gt_log_flush_secs = 60;
60         gt->gt_recoverd_secs = 60;
61         gt->gt_logd_secs = 1;
62         gt->gt_quota_simul_sync = 64;
63         gt->gt_quota_warn_period = 10;
64         gt->gt_quota_scale_num = 1;
65         gt->gt_quota_scale_den = 1;
66         gt->gt_quota_cache_secs = 300;
67         gt->gt_quota_quantum = 60;
68         gt->gt_new_files_jdata = 0;
69         gt->gt_max_readahead = 1 << 18;
70         gt->gt_stall_secs = 600;
71         gt->gt_complain_secs = 10;
72         gt->gt_statfs_quantum = 30;
73         gt->gt_statfs_slow = 0;
74 }
75
76 static struct gfs2_sbd *init_sbd(struct super_block *sb)
77 {
78         struct gfs2_sbd *sdp;
79
80         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
81         if (!sdp)
82                 return NULL;
83
84         sb->s_fs_info = sdp;
85         sdp->sd_vfs = sb;
86
87         gfs2_tune_init(&sdp->sd_tune);
88
89         mutex_init(&sdp->sd_inum_mutex);
90         spin_lock_init(&sdp->sd_statfs_spin);
91
92         spin_lock_init(&sdp->sd_rindex_spin);
93         mutex_init(&sdp->sd_rindex_mutex);
94         INIT_LIST_HEAD(&sdp->sd_rindex_list);
95         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
96
97         INIT_LIST_HEAD(&sdp->sd_jindex_list);
98         spin_lock_init(&sdp->sd_jindex_spin);
99         mutex_init(&sdp->sd_jindex_mutex);
100
101         INIT_LIST_HEAD(&sdp->sd_quota_list);
102         spin_lock_init(&sdp->sd_quota_spin);
103         mutex_init(&sdp->sd_quota_mutex);
104         init_waitqueue_head(&sdp->sd_quota_wait);
105         INIT_LIST_HEAD(&sdp->sd_trunc_list);
106         spin_lock_init(&sdp->sd_trunc_lock);
107
108         spin_lock_init(&sdp->sd_log_lock);
109
110         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
111         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
112         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
113         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
114         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
115
116         mutex_init(&sdp->sd_log_reserve_mutex);
117         INIT_LIST_HEAD(&sdp->sd_ail1_list);
118         INIT_LIST_HEAD(&sdp->sd_ail2_list);
119
120         init_rwsem(&sdp->sd_log_flush_lock);
121         atomic_set(&sdp->sd_log_in_flight, 0);
122         init_waitqueue_head(&sdp->sd_log_flush_wait);
123
124         INIT_LIST_HEAD(&sdp->sd_revoke_list);
125
126         mutex_init(&sdp->sd_freeze_lock);
127
128         return sdp;
129 }
130
131
132 /**
133  * gfs2_check_sb - Check superblock
134  * @sdp: the filesystem
135  * @sb: The superblock
136  * @silent: Don't print a message if the check fails
137  *
138  * Checks the version code of the FS is one that we understand how to
139  * read and that the sizes of the various on-disk structures have not
140  * changed.
141  */
142
143 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
144 {
145         unsigned int x;
146
147         if (sb->sb_magic != GFS2_MAGIC ||
148             sb->sb_type != GFS2_METATYPE_SB) {
149                 if (!silent)
150                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
151                 return -EINVAL;
152         }
153
154         /*  If format numbers match exactly, we're done.  */
155
156         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
157             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
158                 return 0;
159
160         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
161                 for (x = 0; gfs2_old_fs_formats[x]; x++)
162                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
163                                 break;
164
165                 if (!gfs2_old_fs_formats[x]) {
166                         printk(KERN_WARNING
167                                "GFS2: code version (%u, %u) is incompatible "
168                                "with ondisk format (%u, %u)\n",
169                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
170                                sb->sb_fs_format, sb->sb_multihost_format);
171                         printk(KERN_WARNING
172                                "GFS2: I don't know how to upgrade this FS\n");
173                         return -EINVAL;
174                 }
175         }
176
177         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
178                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
179                         if (gfs2_old_multihost_formats[x] ==
180                             sb->sb_multihost_format)
181                                 break;
182
183                 if (!gfs2_old_multihost_formats[x]) {
184                         printk(KERN_WARNING
185                                "GFS2: code version (%u, %u) is incompatible "
186                                "with ondisk format (%u, %u)\n",
187                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
188                                sb->sb_fs_format, sb->sb_multihost_format);
189                         printk(KERN_WARNING
190                                "GFS2: I don't know how to upgrade this FS\n");
191                         return -EINVAL;
192                 }
193         }
194
195         if (!sdp->sd_args.ar_upgrade) {
196                 printk(KERN_WARNING
197                        "GFS2: code version (%u, %u) is incompatible "
198                        "with ondisk format (%u, %u)\n",
199                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
200                        sb->sb_fs_format, sb->sb_multihost_format);
201                 printk(KERN_INFO
202                        "GFS2: Use the \"upgrade\" mount option to upgrade "
203                        "the FS\n");
204                 printk(KERN_INFO "GFS2: See the manual for more details\n");
205                 return -EINVAL;
206         }
207
208         return 0;
209 }
210
211 static void end_bio_io_page(struct bio *bio, int error)
212 {
213         struct page *page = bio->bi_private;
214
215         if (!error)
216                 SetPageUptodate(page);
217         else
218                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
219         unlock_page(page);
220 }
221
222 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
223 {
224         const struct gfs2_sb *str = buf;
225
226         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
227         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
228         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
229         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
230         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
231         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
232         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
233         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
234         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
235         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
236         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
237
238         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
239         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
240 }
241
242 /**
243  * gfs2_read_super - Read the gfs2 super block from disk
244  * @sdp: The GFS2 super block
245  * @sector: The location of the super block
246  * @error: The error code to return
247  *
248  * This uses the bio functions to read the super block from disk
249  * because we want to be 100% sure that we never read cached data.
250  * A super block is read twice only during each GFS2 mount and is
251  * never written to by the filesystem. The first time its read no
252  * locks are held, and the only details which are looked at are those
253  * relating to the locking protocol. Once locking is up and working,
254  * the sb is read again under the lock to establish the location of
255  * the master directory (contains pointers to journals etc) and the
256  * root directory.
257  *
258  * Returns: 0 on success or error
259  */
260
261 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
262 {
263         struct super_block *sb = sdp->sd_vfs;
264         struct gfs2_sb *p;
265         struct page *page;
266         struct bio *bio;
267
268         page = alloc_page(GFP_NOFS);
269         if (unlikely(!page))
270                 return -ENOBUFS;
271
272         ClearPageUptodate(page);
273         ClearPageDirty(page);
274         lock_page(page);
275
276         bio = bio_alloc(GFP_NOFS, 1);
277         if (unlikely(!bio)) {
278                 __free_page(page);
279                 return -ENOBUFS;
280         }
281
282         bio->bi_sector = sector * (sb->s_blocksize >> 9);
283         bio->bi_bdev = sb->s_bdev;
284         bio_add_page(bio, page, PAGE_SIZE, 0);
285
286         bio->bi_end_io = end_bio_io_page;
287         bio->bi_private = page;
288         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
289         wait_on_page_locked(page);
290         bio_put(bio);
291         if (!PageUptodate(page)) {
292                 __free_page(page);
293                 return -EIO;
294         }
295         p = kmap(page);
296         gfs2_sb_in(&sdp->sd_sb, p);
297         kunmap(page);
298         __free_page(page);
299         return 0;
300 }
301 /**
302  * gfs2_read_sb - Read super block
303  * @sdp: The GFS2 superblock
304  * @gl: the glock for the superblock (assumed to be held)
305  * @silent: Don't print message if mount fails
306  *
307  */
308
309 static int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
310 {
311         u32 hash_blocks, ind_blocks, leaf_blocks;
312         u32 tmp_blocks;
313         unsigned int x;
314         int error;
315
316         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
317         if (error) {
318                 if (!silent)
319                         fs_err(sdp, "can't read superblock\n");
320                 return error;
321         }
322
323         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
324         if (error)
325                 return error;
326
327         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
328                                GFS2_BASIC_BLOCK_SHIFT;
329         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
330         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
331                           sizeof(struct gfs2_dinode)) / sizeof(u64);
332         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
333                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
334         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
335         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
336         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
337         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
338         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
339                                 sizeof(struct gfs2_meta_header)) /
340                                 sizeof(struct gfs2_quota_change);
341
342         /* Compute maximum reservation required to add a entry to a directory */
343
344         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
345                              sdp->sd_jbsize);
346
347         ind_blocks = 0;
348         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
349                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
350                 ind_blocks += tmp_blocks;
351         }
352
353         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
354
355         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
356
357         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
358                                 sizeof(struct gfs2_dinode);
359         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
360         for (x = 2;; x++) {
361                 u64 space, d;
362                 u32 m;
363
364                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
365                 d = space;
366                 m = do_div(d, sdp->sd_inptrs);
367
368                 if (d != sdp->sd_heightsize[x - 1] || m)
369                         break;
370                 sdp->sd_heightsize[x] = space;
371         }
372         sdp->sd_max_height = x;
373         sdp->sd_heightsize[x] = ~0;
374         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
375
376         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
377                                  sizeof(struct gfs2_dinode);
378         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
379         for (x = 2;; x++) {
380                 u64 space, d;
381                 u32 m;
382
383                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
384                 d = space;
385                 m = do_div(d, sdp->sd_inptrs);
386
387                 if (d != sdp->sd_jheightsize[x - 1] || m)
388                         break;
389                 sdp->sd_jheightsize[x] = space;
390         }
391         sdp->sd_max_jheight = x;
392         sdp->sd_jheightsize[x] = ~0;
393         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
394
395         return 0;
396 }
397
398 static int init_names(struct gfs2_sbd *sdp, int silent)
399 {
400         char *proto, *table;
401         int error = 0;
402
403         proto = sdp->sd_args.ar_lockproto;
404         table = sdp->sd_args.ar_locktable;
405
406         /*  Try to autodetect  */
407
408         if (!proto[0] || !table[0]) {
409                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
410                 if (error)
411                         return error;
412
413                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
414                 if (error)
415                         goto out;
416
417                 if (!proto[0])
418                         proto = sdp->sd_sb.sb_lockproto;
419                 if (!table[0])
420                         table = sdp->sd_sb.sb_locktable;
421         }
422
423         if (!table[0])
424                 table = sdp->sd_vfs->s_id;
425
426         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
427         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
428
429         table = sdp->sd_table_name;
430         while ((table = strchr(table, '/')))
431                 *table = '_';
432
433 out:
434         return error;
435 }
436
437 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
438                         int undo)
439 {
440         int error = 0;
441
442         if (undo)
443                 goto fail_trans;
444
445         error = gfs2_glock_nq_num(sdp,
446                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
447                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
448                                   mount_gh);
449         if (error) {
450                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
451                 goto fail;
452         }
453
454         error = gfs2_glock_nq_num(sdp,
455                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
456                                   LM_ST_SHARED,
457                                   LM_FLAG_NOEXP | GL_EXACT,
458                                   &sdp->sd_live_gh);
459         if (error) {
460                 fs_err(sdp, "can't acquire live glock: %d\n", error);
461                 goto fail_mount;
462         }
463
464         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
465                                CREATE, &sdp->sd_rename_gl);
466         if (error) {
467                 fs_err(sdp, "can't create rename glock: %d\n", error);
468                 goto fail_live;
469         }
470
471         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
472                                CREATE, &sdp->sd_trans_gl);
473         if (error) {
474                 fs_err(sdp, "can't create transaction glock: %d\n", error);
475                 goto fail_rename;
476         }
477
478         return 0;
479
480 fail_trans:
481         gfs2_glock_put(sdp->sd_trans_gl);
482 fail_rename:
483         gfs2_glock_put(sdp->sd_rename_gl);
484 fail_live:
485         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
486 fail_mount:
487         gfs2_glock_dq_uninit(mount_gh);
488 fail:
489         return error;
490 }
491
492 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
493                             u64 no_addr, const char *name)
494 {
495         struct gfs2_sbd *sdp = sb->s_fs_info;
496         struct dentry *dentry;
497         struct inode *inode;
498
499         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
500         if (IS_ERR(inode)) {
501                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
502                 return PTR_ERR(inode);
503         }
504         dentry = d_alloc_root(inode);
505         if (!dentry) {
506                 fs_err(sdp, "can't alloc %s dentry\n", name);
507                 iput(inode);
508                 return -ENOMEM;
509         }
510         dentry->d_op = &gfs2_dops;
511         *dptr = dentry;
512         return 0;
513 }
514
515 static int init_sb(struct gfs2_sbd *sdp, int silent)
516 {
517         struct super_block *sb = sdp->sd_vfs;
518         struct gfs2_holder sb_gh;
519         u64 no_addr;
520         int ret;
521
522         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
523                                 LM_ST_SHARED, 0, &sb_gh);
524         if (ret) {
525                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
526                 return ret;
527         }
528
529         ret = gfs2_read_sb(sdp, sb_gh.gh_gl, silent);
530         if (ret) {
531                 fs_err(sdp, "can't read superblock: %d\n", ret);
532                 goto out;
533         }
534
535         /* Set up the buffer cache and SB for real */
536         if (sdp->sd_sb.sb_bsize < bdev_hardsect_size(sb->s_bdev)) {
537                 ret = -EINVAL;
538                 fs_err(sdp, "FS block size (%u) is too small for device "
539                        "block size (%u)\n",
540                        sdp->sd_sb.sb_bsize, bdev_hardsect_size(sb->s_bdev));
541                 goto out;
542         }
543         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
544                 ret = -EINVAL;
545                 fs_err(sdp, "FS block size (%u) is too big for machine "
546                        "page size (%u)\n",
547                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
548                 goto out;
549         }
550         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
551
552         /* Get the root inode */
553         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
554         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
555         if (ret)
556                 goto out;
557
558         /* Get the master inode */
559         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
560         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
561         if (ret) {
562                 dput(sdp->sd_root_dir);
563                 goto out;
564         }
565         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
566 out:
567         gfs2_glock_dq_uninit(&sb_gh);
568         return ret;
569 }
570
571 /**
572  * map_journal_extents - create a reusable "extent" mapping from all logical
573  * blocks to all physical blocks for the given journal.  This will save
574  * us time when writing journal blocks.  Most journals will have only one
575  * extent that maps all their logical blocks.  That's because gfs2.mkfs
576  * arranges the journal blocks sequentially to maximize performance.
577  * So the extent would map the first block for the entire file length.
578  * However, gfs2_jadd can happen while file activity is happening, so
579  * those journals may not be sequential.  Less likely is the case where
580  * the users created their own journals by mounting the metafs and
581  * laying it out.  But it's still possible.  These journals might have
582  * several extents.
583  *
584  * TODO: This should be done in bigger chunks rather than one block at a time,
585  *       but since it's only done at mount time, I'm not worried about the
586  *       time it takes.
587  */
588 static int map_journal_extents(struct gfs2_sbd *sdp)
589 {
590         struct gfs2_jdesc *jd = sdp->sd_jdesc;
591         unsigned int lb;
592         u64 db, prev_db; /* logical block, disk block, prev disk block */
593         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
594         struct gfs2_journal_extent *jext = NULL;
595         struct buffer_head bh;
596         int rc = 0;
597
598         prev_db = 0;
599
600         for (lb = 0; lb < ip->i_disksize >> sdp->sd_sb.sb_bsize_shift; lb++) {
601                 bh.b_state = 0;
602                 bh.b_blocknr = 0;
603                 bh.b_size = 1 << ip->i_inode.i_blkbits;
604                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
605                 db = bh.b_blocknr;
606                 if (rc || !db) {
607                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
608                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
609                         break;
610                 }
611                 if (!prev_db || db != prev_db + 1) {
612                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
613                                        GFP_KERNEL);
614                         if (!jext) {
615                                 printk(KERN_INFO "GFS2 error: out of memory "
616                                        "mapping journal extents.\n");
617                                 rc = -ENOMEM;
618                                 break;
619                         }
620                         jext->dblock = db;
621                         jext->lblock = lb;
622                         jext->blocks = 1;
623                         list_add_tail(&jext->extent_list, &jd->extent_list);
624                 } else {
625                         jext->blocks++;
626                 }
627                 prev_db = db;
628         }
629         return rc;
630 }
631
632 static void gfs2_lm_others_may_mount(struct gfs2_sbd *sdp)
633 {
634         if (!sdp->sd_lockstruct.ls_ops->lm_others_may_mount)
635                 return;
636         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
637                 sdp->sd_lockstruct.ls_ops->lm_others_may_mount(
638                                         sdp->sd_lockstruct.ls_lockspace);
639 }
640
641 static int init_journal(struct gfs2_sbd *sdp, int undo)
642 {
643         struct inode *master = sdp->sd_master_dir->d_inode;
644         struct gfs2_holder ji_gh;
645         struct task_struct *p;
646         struct gfs2_inode *ip;
647         int jindex = 1;
648         int error = 0;
649
650         if (undo) {
651                 jindex = 0;
652                 goto fail_recoverd;
653         }
654
655         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
656         if (IS_ERR(sdp->sd_jindex)) {
657                 fs_err(sdp, "can't lookup journal index: %d\n", error);
658                 return PTR_ERR(sdp->sd_jindex);
659         }
660         ip = GFS2_I(sdp->sd_jindex);
661
662         /* Load in the journal index special file */
663
664         error = gfs2_jindex_hold(sdp, &ji_gh);
665         if (error) {
666                 fs_err(sdp, "can't read journal index: %d\n", error);
667                 goto fail;
668         }
669
670         error = -EINVAL;
671         if (!gfs2_jindex_size(sdp)) {
672                 fs_err(sdp, "no journals!\n");
673                 goto fail_jindex;
674         }
675
676         if (sdp->sd_args.ar_spectator) {
677                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
678                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
679         } else {
680                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
681                         fs_err(sdp, "can't mount journal #%u\n",
682                                sdp->sd_lockstruct.ls_jid);
683                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
684                                gfs2_jindex_size(sdp),
685                                gfs2_jindex_size(sdp) - 1);
686                         goto fail_jindex;
687                 }
688                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
689
690                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
691                                           &gfs2_journal_glops,
692                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
693                                           &sdp->sd_journal_gh);
694                 if (error) {
695                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
696                         goto fail_jindex;
697                 }
698
699                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
700                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
701                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
702                                            &sdp->sd_jinode_gh);
703                 if (error) {
704                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
705                                error);
706                         goto fail_journal_gh;
707                 }
708
709                 error = gfs2_jdesc_check(sdp->sd_jdesc);
710                 if (error) {
711                         fs_err(sdp, "my journal (%u) is bad: %d\n",
712                                sdp->sd_jdesc->jd_jid, error);
713                         goto fail_jinode_gh;
714                 }
715                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
716
717                 /* Map the extents for this journal's blocks */
718                 map_journal_extents(sdp);
719         }
720
721         if (sdp->sd_lockstruct.ls_first) {
722                 unsigned int x;
723                 for (x = 0; x < sdp->sd_journals; x++) {
724                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
725                         if (error) {
726                                 fs_err(sdp, "error recovering journal %u: %d\n",
727                                        x, error);
728                                 goto fail_jinode_gh;
729                         }
730                 }
731
732                 gfs2_lm_others_may_mount(sdp);
733         } else if (!sdp->sd_args.ar_spectator) {
734                 error = gfs2_recover_journal(sdp->sd_jdesc);
735                 if (error) {
736                         fs_err(sdp, "error recovering my journal: %d\n", error);
737                         goto fail_jinode_gh;
738                 }
739         }
740
741         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
742         gfs2_glock_dq_uninit(&ji_gh);
743         jindex = 0;
744
745         p = kthread_run(gfs2_recoverd, sdp, "gfs2_recoverd");
746         error = IS_ERR(p);
747         if (error) {
748                 fs_err(sdp, "can't start recoverd thread: %d\n", error);
749                 goto fail_jinode_gh;
750         }
751         sdp->sd_recoverd_process = p;
752
753         return 0;
754
755 fail_recoverd:
756         kthread_stop(sdp->sd_recoverd_process);
757 fail_jinode_gh:
758         if (!sdp->sd_args.ar_spectator)
759                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
760 fail_journal_gh:
761         if (!sdp->sd_args.ar_spectator)
762                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
763 fail_jindex:
764         gfs2_jindex_free(sdp);
765         if (jindex)
766                 gfs2_glock_dq_uninit(&ji_gh);
767 fail:
768         iput(sdp->sd_jindex);
769         return error;
770 }
771
772
773 static int init_inodes(struct gfs2_sbd *sdp, int undo)
774 {
775         int error = 0;
776         struct gfs2_inode *ip;
777         struct inode *master = sdp->sd_master_dir->d_inode;
778
779         if (undo)
780                 goto fail_qinode;
781
782         error = init_journal(sdp, undo);
783         if (error)
784                 goto fail;
785
786         /* Read in the master inode number inode */
787         sdp->sd_inum_inode = gfs2_lookup_simple(master, "inum");
788         if (IS_ERR(sdp->sd_inum_inode)) {
789                 error = PTR_ERR(sdp->sd_inum_inode);
790                 fs_err(sdp, "can't read in inum inode: %d\n", error);
791                 goto fail_journal;
792         }
793
794
795         /* Read in the master statfs inode */
796         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
797         if (IS_ERR(sdp->sd_statfs_inode)) {
798                 error = PTR_ERR(sdp->sd_statfs_inode);
799                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
800                 goto fail_inum;
801         }
802
803         /* Read in the resource index inode */
804         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
805         if (IS_ERR(sdp->sd_rindex)) {
806                 error = PTR_ERR(sdp->sd_rindex);
807                 fs_err(sdp, "can't get resource index inode: %d\n", error);
808                 goto fail_statfs;
809         }
810         ip = GFS2_I(sdp->sd_rindex);
811         sdp->sd_rindex_uptodate = 0;
812
813         /* Read in the quota inode */
814         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
815         if (IS_ERR(sdp->sd_quota_inode)) {
816                 error = PTR_ERR(sdp->sd_quota_inode);
817                 fs_err(sdp, "can't get quota file inode: %d\n", error);
818                 goto fail_rindex;
819         }
820         return 0;
821
822 fail_qinode:
823         iput(sdp->sd_quota_inode);
824 fail_rindex:
825         gfs2_clear_rgrpd(sdp);
826         iput(sdp->sd_rindex);
827 fail_statfs:
828         iput(sdp->sd_statfs_inode);
829 fail_inum:
830         iput(sdp->sd_inum_inode);
831 fail_journal:
832         init_journal(sdp, UNDO);
833 fail:
834         return error;
835 }
836
837 static int init_per_node(struct gfs2_sbd *sdp, int undo)
838 {
839         struct inode *pn = NULL;
840         char buf[30];
841         int error = 0;
842         struct gfs2_inode *ip;
843         struct inode *master = sdp->sd_master_dir->d_inode;
844
845         if (sdp->sd_args.ar_spectator)
846                 return 0;
847
848         if (undo)
849                 goto fail_qc_gh;
850
851         pn = gfs2_lookup_simple(master, "per_node");
852         if (IS_ERR(pn)) {
853                 error = PTR_ERR(pn);
854                 fs_err(sdp, "can't find per_node directory: %d\n", error);
855                 return error;
856         }
857
858         sprintf(buf, "inum_range%u", sdp->sd_jdesc->jd_jid);
859         sdp->sd_ir_inode = gfs2_lookup_simple(pn, buf);
860         if (IS_ERR(sdp->sd_ir_inode)) {
861                 error = PTR_ERR(sdp->sd_ir_inode);
862                 fs_err(sdp, "can't find local \"ir\" file: %d\n", error);
863                 goto fail;
864         }
865
866         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
867         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
868         if (IS_ERR(sdp->sd_sc_inode)) {
869                 error = PTR_ERR(sdp->sd_sc_inode);
870                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
871                 goto fail_ir_i;
872         }
873
874         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
875         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
876         if (IS_ERR(sdp->sd_qc_inode)) {
877                 error = PTR_ERR(sdp->sd_qc_inode);
878                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
879                 goto fail_ut_i;
880         }
881
882         iput(pn);
883         pn = NULL;
884
885         ip = GFS2_I(sdp->sd_ir_inode);
886         error = gfs2_glock_nq_init(ip->i_gl,
887                                    LM_ST_EXCLUSIVE, 0,
888                                    &sdp->sd_ir_gh);
889         if (error) {
890                 fs_err(sdp, "can't lock local \"ir\" file: %d\n", error);
891                 goto fail_qc_i;
892         }
893
894         ip = GFS2_I(sdp->sd_sc_inode);
895         error = gfs2_glock_nq_init(ip->i_gl,
896                                    LM_ST_EXCLUSIVE, 0,
897                                    &sdp->sd_sc_gh);
898         if (error) {
899                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
900                 goto fail_ir_gh;
901         }
902
903         ip = GFS2_I(sdp->sd_qc_inode);
904         error = gfs2_glock_nq_init(ip->i_gl,
905                                    LM_ST_EXCLUSIVE, 0,
906                                    &sdp->sd_qc_gh);
907         if (error) {
908                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
909                 goto fail_ut_gh;
910         }
911
912         return 0;
913
914 fail_qc_gh:
915         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
916 fail_ut_gh:
917         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
918 fail_ir_gh:
919         gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
920 fail_qc_i:
921         iput(sdp->sd_qc_inode);
922 fail_ut_i:
923         iput(sdp->sd_sc_inode);
924 fail_ir_i:
925         iput(sdp->sd_ir_inode);
926 fail:
927         if (pn)
928                 iput(pn);
929         return error;
930 }
931
932 static int init_threads(struct gfs2_sbd *sdp, int undo)
933 {
934         struct task_struct *p;
935         int error = 0;
936
937         if (undo)
938                 goto fail_quotad;
939
940         sdp->sd_log_flush_time = jiffies;
941         sdp->sd_jindex_refresh_time = jiffies;
942
943         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
944         error = IS_ERR(p);
945         if (error) {
946                 fs_err(sdp, "can't start logd thread: %d\n", error);
947                 return error;
948         }
949         sdp->sd_logd_process = p;
950
951         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
952         error = IS_ERR(p);
953         if (error) {
954                 fs_err(sdp, "can't start quotad thread: %d\n", error);
955                 goto fail;
956         }
957         sdp->sd_quotad_process = p;
958
959         return 0;
960
961
962 fail_quotad:
963         kthread_stop(sdp->sd_quotad_process);
964 fail:
965         kthread_stop(sdp->sd_logd_process);
966         return error;
967 }
968
969 /**
970  * gfs2_lm_mount - mount a locking protocol
971  * @sdp: the filesystem
972  * @args: mount arguements
973  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
974  *
975  * Returns: errno
976  */
977
978 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
979 {
980         char *proto = sdp->sd_proto_name;
981         char *table = sdp->sd_table_name;
982         int flags = LM_MFLAG_CONV_NODROP;
983         int error;
984
985         if (sdp->sd_args.ar_spectator)
986                 flags |= LM_MFLAG_SPECTATOR;
987
988         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
989
990         error = gfs2_mount_lockproto(proto, table, sdp->sd_args.ar_hostdata,
991                                      gfs2_glock_cb, sdp,
992                                      GFS2_MIN_LVB_SIZE, flags,
993                                      &sdp->sd_lockstruct, &sdp->sd_kobj);
994         if (error) {
995                 fs_info(sdp, "can't mount proto=%s, table=%s, hostdata=%s\n",
996                         proto, table, sdp->sd_args.ar_hostdata);
997                 goto out;
998         }
999
1000         if (gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_ops) ||
1001             gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lvb_size >=
1002                                   GFS2_MIN_LVB_SIZE)) {
1003                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
1004                 goto out;
1005         }
1006
1007         if (sdp->sd_args.ar_spectator)
1008                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1009         else
1010                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1011                          sdp->sd_lockstruct.ls_jid);
1012
1013         fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1014
1015         if ((sdp->sd_lockstruct.ls_flags & LM_LSFLAG_LOCAL) &&
1016             !sdp->sd_args.ar_ignore_local_fs) {
1017                 sdp->sd_args.ar_localflocks = 1;
1018                 sdp->sd_args.ar_localcaching = 1;
1019         }
1020
1021 out:
1022         return error;
1023 }
1024
1025 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1026 {
1027         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1028                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
1029 }
1030
1031 /**
1032  * fill_super - Read in superblock
1033  * @sb: The VFS superblock
1034  * @data: Mount options
1035  * @silent: Don't complain if it's not a GFS2 filesystem
1036  *
1037  * Returns: errno
1038  */
1039
1040 static int fill_super(struct super_block *sb, void *data, int silent)
1041 {
1042         struct gfs2_sbd *sdp;
1043         struct gfs2_holder mount_gh;
1044         int error;
1045
1046         sdp = init_sbd(sb);
1047         if (!sdp) {
1048                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1049                 return -ENOMEM;
1050         }
1051
1052         error = gfs2_mount_args(sdp, (char *)data, 0);
1053         if (error) {
1054                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1055                 goto fail;
1056         }
1057
1058         sb->s_magic = GFS2_MAGIC;
1059         sb->s_op = &gfs2_super_ops;
1060         sb->s_export_op = &gfs2_export_ops;
1061         sb->s_time_gran = 1;
1062         sb->s_maxbytes = MAX_LFS_FILESIZE;
1063
1064         /* Set up the buffer cache and fill in some fake block size values
1065            to allow us to read-in the on-disk superblock. */
1066         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1067         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1068         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1069                                GFS2_BASIC_BLOCK_SHIFT;
1070         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1071
1072         error = init_names(sdp, silent);
1073         if (error)
1074                 goto fail;
1075
1076         gfs2_create_debugfs_file(sdp);
1077
1078         error = gfs2_sys_fs_add(sdp);
1079         if (error)
1080                 goto fail;
1081
1082         error = gfs2_lm_mount(sdp, silent);
1083         if (error)
1084                 goto fail_sys;
1085
1086         error = init_locking(sdp, &mount_gh, DO);
1087         if (error)
1088                 goto fail_lm;
1089
1090         error = init_sb(sdp, silent);
1091         if (error)
1092                 goto fail_locking;
1093
1094         error = init_inodes(sdp, DO);
1095         if (error)
1096                 goto fail_sb;
1097
1098         error = init_per_node(sdp, DO);
1099         if (error)
1100                 goto fail_inodes;
1101
1102         error = gfs2_statfs_init(sdp);
1103         if (error) {
1104                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1105                 goto fail_per_node;
1106         }
1107
1108         error = init_threads(sdp, DO);
1109         if (error)
1110                 goto fail_per_node;
1111
1112         if (!(sb->s_flags & MS_RDONLY)) {
1113                 error = gfs2_make_fs_rw(sdp);
1114                 if (error) {
1115                         fs_err(sdp, "can't make FS RW: %d\n", error);
1116                         goto fail_threads;
1117                 }
1118         }
1119
1120         gfs2_glock_dq_uninit(&mount_gh);
1121
1122         return 0;
1123
1124 fail_threads:
1125         init_threads(sdp, UNDO);
1126 fail_per_node:
1127         init_per_node(sdp, UNDO);
1128 fail_inodes:
1129         init_inodes(sdp, UNDO);
1130 fail_sb:
1131         if (sdp->sd_root_dir)
1132                 dput(sdp->sd_root_dir);
1133         if (sdp->sd_master_dir)
1134                 dput(sdp->sd_master_dir);
1135         sb->s_root = NULL;
1136 fail_locking:
1137         init_locking(sdp, &mount_gh, UNDO);
1138 fail_lm:
1139         gfs2_gl_hash_clear(sdp);
1140         gfs2_lm_unmount(sdp);
1141         while (invalidate_inodes(sb))
1142                 yield();
1143 fail_sys:
1144         gfs2_sys_fs_del(sdp);
1145 fail:
1146         gfs2_delete_debugfs_file(sdp);
1147         kfree(sdp);
1148         sb->s_fs_info = NULL;
1149         return error;
1150 }
1151
1152 static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
1153                        const char *dev_name, void *data, struct vfsmount *mnt)
1154 {
1155         return get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
1156 }
1157
1158 static struct super_block *get_gfs2_sb(const char *dev_name)
1159 {
1160         struct super_block *sb;
1161         struct nameidata nd;
1162         int error;
1163
1164         error = path_lookup(dev_name, LOOKUP_FOLLOW, &nd);
1165         if (error) {
1166                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1167                        dev_name, error);
1168                 return NULL;
1169         }
1170         sb = nd.path.dentry->d_inode->i_sb;
1171         if (sb && (sb->s_type == &gfs2_fs_type))
1172                 atomic_inc(&sb->s_active);
1173         else
1174                 sb = NULL;
1175         path_put(&nd.path);
1176         return sb;
1177 }
1178
1179 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
1180                             const char *dev_name, void *data, struct vfsmount *mnt)
1181 {
1182         struct super_block *sb = NULL;
1183         struct gfs2_sbd *sdp;
1184
1185         sb = get_gfs2_sb(dev_name);
1186         if (!sb) {
1187                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1188                 return -ENOENT;
1189         }
1190         sdp = sb->s_fs_info;
1191         mnt->mnt_sb = sb;
1192         mnt->mnt_root = dget(sdp->sd_master_dir);
1193         return 0;
1194 }
1195
1196 static void gfs2_kill_sb(struct super_block *sb)
1197 {
1198         struct gfs2_sbd *sdp = sb->s_fs_info;
1199         if (sdp) {
1200                 gfs2_meta_syncfs(sdp);
1201                 dput(sdp->sd_root_dir);
1202                 dput(sdp->sd_master_dir);
1203                 sdp->sd_root_dir = NULL;
1204                 sdp->sd_master_dir = NULL;
1205         }
1206         shrink_dcache_sb(sb);
1207         kill_block_super(sb);
1208         if (sdp)
1209                 gfs2_delete_debugfs_file(sdp);
1210 }
1211
1212 struct file_system_type gfs2_fs_type = {
1213         .name = "gfs2",
1214         .fs_flags = FS_REQUIRES_DEV,
1215         .get_sb = gfs2_get_sb,
1216         .kill_sb = gfs2_kill_sb,
1217         .owner = THIS_MODULE,
1218 };
1219
1220 struct file_system_type gfs2meta_fs_type = {
1221         .name = "gfs2meta",
1222         .fs_flags = FS_REQUIRES_DEV,
1223         .get_sb = gfs2_get_sb_meta,
1224         .owner = THIS_MODULE,
1225 };
1226