GFS2: Fix "truncate in progress" hang
[linux-2.6.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 "daemon.h"
26 #include "glock.h"
27 #include "glops.h"
28 #include "inode.h"
29 #include "mount.h"
30 #include "recovery.h"
31 #include "rgrp.h"
32 #include "super.h"
33 #include "sys.h"
34 #include "util.h"
35 #include "log.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_demote_secs = 300;
59         gt->gt_incore_log_blocks = 1024;
60         gt->gt_log_flush_secs = 60;
61         gt->gt_recoverd_secs = 60;
62         gt->gt_logd_secs = 1;
63         gt->gt_quota_simul_sync = 64;
64         gt->gt_quota_warn_period = 10;
65         gt->gt_quota_scale_num = 1;
66         gt->gt_quota_scale_den = 1;
67         gt->gt_quota_cache_secs = 300;
68         gt->gt_quota_quantum = 60;
69         gt->gt_new_files_jdata = 0;
70         gt->gt_max_readahead = 1 << 18;
71         gt->gt_stall_secs = 600;
72         gt->gt_complain_secs = 10;
73         gt->gt_statfs_quantum = 30;
74         gt->gt_statfs_slow = 0;
75 }
76
77 static struct gfs2_sbd *init_sbd(struct super_block *sb)
78 {
79         struct gfs2_sbd *sdp;
80
81         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
82         if (!sdp)
83                 return NULL;
84
85         sb->s_fs_info = sdp;
86         sdp->sd_vfs = sb;
87
88         gfs2_tune_init(&sdp->sd_tune);
89
90         INIT_LIST_HEAD(&sdp->sd_reclaim_list);
91         spin_lock_init(&sdp->sd_reclaim_lock);
92         init_waitqueue_head(&sdp->sd_reclaim_wq);
93
94         mutex_init(&sdp->sd_inum_mutex);
95         spin_lock_init(&sdp->sd_statfs_spin);
96
97         spin_lock_init(&sdp->sd_rindex_spin);
98         mutex_init(&sdp->sd_rindex_mutex);
99         INIT_LIST_HEAD(&sdp->sd_rindex_list);
100         INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
101
102         INIT_LIST_HEAD(&sdp->sd_jindex_list);
103         spin_lock_init(&sdp->sd_jindex_spin);
104         mutex_init(&sdp->sd_jindex_mutex);
105
106         INIT_LIST_HEAD(&sdp->sd_quota_list);
107         spin_lock_init(&sdp->sd_quota_spin);
108         mutex_init(&sdp->sd_quota_mutex);
109         init_waitqueue_head(&sdp->sd_quota_wait);
110         INIT_LIST_HEAD(&sdp->sd_trunc_list);
111         spin_lock_init(&sdp->sd_trunc_lock);
112
113         spin_lock_init(&sdp->sd_log_lock);
114
115         INIT_LIST_HEAD(&sdp->sd_log_le_buf);
116         INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
117         INIT_LIST_HEAD(&sdp->sd_log_le_rg);
118         INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
119         INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
120
121         mutex_init(&sdp->sd_log_reserve_mutex);
122         INIT_LIST_HEAD(&sdp->sd_ail1_list);
123         INIT_LIST_HEAD(&sdp->sd_ail2_list);
124
125         init_rwsem(&sdp->sd_log_flush_lock);
126         atomic_set(&sdp->sd_log_in_flight, 0);
127         init_waitqueue_head(&sdp->sd_log_flush_wait);
128
129         INIT_LIST_HEAD(&sdp->sd_revoke_list);
130
131         mutex_init(&sdp->sd_freeze_lock);
132
133         return sdp;
134 }
135
136
137 /**
138  * gfs2_check_sb - Check superblock
139  * @sdp: the filesystem
140  * @sb: The superblock
141  * @silent: Don't print a message if the check fails
142  *
143  * Checks the version code of the FS is one that we understand how to
144  * read and that the sizes of the various on-disk structures have not
145  * changed.
146  */
147
148 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
149 {
150         unsigned int x;
151
152         if (sb->sb_magic != GFS2_MAGIC ||
153             sb->sb_type != GFS2_METATYPE_SB) {
154                 if (!silent)
155                         printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
156                 return -EINVAL;
157         }
158
159         /*  If format numbers match exactly, we're done.  */
160
161         if (sb->sb_fs_format == GFS2_FORMAT_FS &&
162             sb->sb_multihost_format == GFS2_FORMAT_MULTI)
163                 return 0;
164
165         if (sb->sb_fs_format != GFS2_FORMAT_FS) {
166                 for (x = 0; gfs2_old_fs_formats[x]; x++)
167                         if (gfs2_old_fs_formats[x] == sb->sb_fs_format)
168                                 break;
169
170                 if (!gfs2_old_fs_formats[x]) {
171                         printk(KERN_WARNING
172                                "GFS2: code version (%u, %u) is incompatible "
173                                "with ondisk format (%u, %u)\n",
174                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
175                                sb->sb_fs_format, sb->sb_multihost_format);
176                         printk(KERN_WARNING
177                                "GFS2: I don't know how to upgrade this FS\n");
178                         return -EINVAL;
179                 }
180         }
181
182         if (sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
183                 for (x = 0; gfs2_old_multihost_formats[x]; x++)
184                         if (gfs2_old_multihost_formats[x] ==
185                             sb->sb_multihost_format)
186                                 break;
187
188                 if (!gfs2_old_multihost_formats[x]) {
189                         printk(KERN_WARNING
190                                "GFS2: code version (%u, %u) is incompatible "
191                                "with ondisk format (%u, %u)\n",
192                                GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
193                                sb->sb_fs_format, sb->sb_multihost_format);
194                         printk(KERN_WARNING
195                                "GFS2: I don't know how to upgrade this FS\n");
196                         return -EINVAL;
197                 }
198         }
199
200         if (!sdp->sd_args.ar_upgrade) {
201                 printk(KERN_WARNING
202                        "GFS2: code version (%u, %u) is incompatible "
203                        "with ondisk format (%u, %u)\n",
204                        GFS2_FORMAT_FS, GFS2_FORMAT_MULTI,
205                        sb->sb_fs_format, sb->sb_multihost_format);
206                 printk(KERN_INFO
207                        "GFS2: Use the \"upgrade\" mount option to upgrade "
208                        "the FS\n");
209                 printk(KERN_INFO "GFS2: See the manual for more details\n");
210                 return -EINVAL;
211         }
212
213         return 0;
214 }
215
216 static void end_bio_io_page(struct bio *bio, int error)
217 {
218         struct page *page = bio->bi_private;
219
220         if (!error)
221                 SetPageUptodate(page);
222         else
223                 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
224         unlock_page(page);
225 }
226
227 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
228 {
229         const struct gfs2_sb *str = buf;
230
231         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
232         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
233         sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
234         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
235         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
236         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
237         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
238         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
239         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
240         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
241         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
242
243         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
244         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
245 }
246
247 /**
248  * gfs2_read_super - Read the gfs2 super block from disk
249  * @sdp: The GFS2 super block
250  * @sector: The location of the super block
251  * @error: The error code to return
252  *
253  * This uses the bio functions to read the super block from disk
254  * because we want to be 100% sure that we never read cached data.
255  * A super block is read twice only during each GFS2 mount and is
256  * never written to by the filesystem. The first time its read no
257  * locks are held, and the only details which are looked at are those
258  * relating to the locking protocol. Once locking is up and working,
259  * the sb is read again under the lock to establish the location of
260  * the master directory (contains pointers to journals etc) and the
261  * root directory.
262  *
263  * Returns: 0 on success or error
264  */
265
266 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
267 {
268         struct super_block *sb = sdp->sd_vfs;
269         struct gfs2_sb *p;
270         struct page *page;
271         struct bio *bio;
272
273         page = alloc_page(GFP_NOFS);
274         if (unlikely(!page))
275                 return -ENOBUFS;
276
277         ClearPageUptodate(page);
278         ClearPageDirty(page);
279         lock_page(page);
280
281         bio = bio_alloc(GFP_NOFS, 1);
282         if (unlikely(!bio)) {
283                 __free_page(page);
284                 return -ENOBUFS;
285         }
286
287         bio->bi_sector = sector * (sb->s_blocksize >> 9);
288         bio->bi_bdev = sb->s_bdev;
289         bio_add_page(bio, page, PAGE_SIZE, 0);
290
291         bio->bi_end_io = end_bio_io_page;
292         bio->bi_private = page;
293         submit_bio(READ_SYNC | (1 << BIO_RW_META), bio);
294         wait_on_page_locked(page);
295         bio_put(bio);
296         if (!PageUptodate(page)) {
297                 __free_page(page);
298                 return -EIO;
299         }
300         p = kmap(page);
301         gfs2_sb_in(&sdp->sd_sb, p);
302         kunmap(page);
303         __free_page(page);
304         return 0;
305 }
306 /**
307  * gfs2_read_sb - Read super block
308  * @sdp: The GFS2 superblock
309  * @gl: the glock for the superblock (assumed to be held)
310  * @silent: Don't print message if mount fails
311  *
312  */
313
314 static int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
315 {
316         u32 hash_blocks, ind_blocks, leaf_blocks;
317         u32 tmp_blocks;
318         unsigned int x;
319         int error;
320
321         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
322         if (error) {
323                 if (!silent)
324                         fs_err(sdp, "can't read superblock\n");
325                 return error;
326         }
327
328         error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
329         if (error)
330                 return error;
331
332         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
333                                GFS2_BASIC_BLOCK_SHIFT;
334         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
335         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
336                           sizeof(struct gfs2_dinode)) / sizeof(u64);
337         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
338                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
339         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
340         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
341         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
342         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
343         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
344                                 sizeof(struct gfs2_meta_header)) /
345                                 sizeof(struct gfs2_quota_change);
346
347         /* Compute maximum reservation required to add a entry to a directory */
348
349         hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
350                              sdp->sd_jbsize);
351
352         ind_blocks = 0;
353         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
354                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
355                 ind_blocks += tmp_blocks;
356         }
357
358         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
359
360         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
361
362         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
363                                 sizeof(struct gfs2_dinode);
364         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
365         for (x = 2;; x++) {
366                 u64 space, d;
367                 u32 m;
368
369                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
370                 d = space;
371                 m = do_div(d, sdp->sd_inptrs);
372
373                 if (d != sdp->sd_heightsize[x - 1] || m)
374                         break;
375                 sdp->sd_heightsize[x] = space;
376         }
377         sdp->sd_max_height = x;
378         sdp->sd_heightsize[x] = ~0;
379         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
380
381         sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
382                                  sizeof(struct gfs2_dinode);
383         sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
384         for (x = 2;; x++) {
385                 u64 space, d;
386                 u32 m;
387
388                 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
389                 d = space;
390                 m = do_div(d, sdp->sd_inptrs);
391
392                 if (d != sdp->sd_jheightsize[x - 1] || m)
393                         break;
394                 sdp->sd_jheightsize[x] = space;
395         }
396         sdp->sd_max_jheight = x;
397         sdp->sd_jheightsize[x] = ~0;
398         gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
399
400         return 0;
401 }
402
403 static int init_names(struct gfs2_sbd *sdp, int silent)
404 {
405         char *proto, *table;
406         int error = 0;
407
408         proto = sdp->sd_args.ar_lockproto;
409         table = sdp->sd_args.ar_locktable;
410
411         /*  Try to autodetect  */
412
413         if (!proto[0] || !table[0]) {
414                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
415                 if (error)
416                         return error;
417
418                 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
419                 if (error)
420                         goto out;
421
422                 if (!proto[0])
423                         proto = sdp->sd_sb.sb_lockproto;
424                 if (!table[0])
425                         table = sdp->sd_sb.sb_locktable;
426         }
427
428         if (!table[0])
429                 table = sdp->sd_vfs->s_id;
430
431         strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
432         strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
433
434         table = sdp->sd_table_name;
435         while ((table = strchr(table, '/')))
436                 *table = '_';
437
438 out:
439         return error;
440 }
441
442 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
443                         int undo)
444 {
445         struct task_struct *p;
446         int error = 0;
447
448         if (undo)
449                 goto fail_trans;
450
451         for (sdp->sd_glockd_num = 0;
452              sdp->sd_glockd_num < sdp->sd_args.ar_num_glockd;
453              sdp->sd_glockd_num++) {
454                 p = kthread_run(gfs2_glockd, sdp, "gfs2_glockd");
455                 error = IS_ERR(p);
456                 if (error) {
457                         fs_err(sdp, "can't start glockd thread: %d\n", error);
458                         goto fail;
459                 }
460                 sdp->sd_glockd_process[sdp->sd_glockd_num] = p;
461         }
462
463         error = gfs2_glock_nq_num(sdp,
464                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
465                                   LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
466                                   mount_gh);
467         if (error) {
468                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
469                 goto fail;
470         }
471
472         error = gfs2_glock_nq_num(sdp,
473                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
474                                   LM_ST_SHARED,
475                                   LM_FLAG_NOEXP | GL_EXACT,
476                                   &sdp->sd_live_gh);
477         if (error) {
478                 fs_err(sdp, "can't acquire live glock: %d\n", error);
479                 goto fail_mount;
480         }
481
482         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
483                                CREATE, &sdp->sd_rename_gl);
484         if (error) {
485                 fs_err(sdp, "can't create rename glock: %d\n", error);
486                 goto fail_live;
487         }
488
489         error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
490                                CREATE, &sdp->sd_trans_gl);
491         if (error) {
492                 fs_err(sdp, "can't create transaction glock: %d\n", error);
493                 goto fail_rename;
494         }
495         set_bit(GLF_STICKY, &sdp->sd_trans_gl->gl_flags);
496
497         return 0;
498
499 fail_trans:
500         gfs2_glock_put(sdp->sd_trans_gl);
501 fail_rename:
502         gfs2_glock_put(sdp->sd_rename_gl);
503 fail_live:
504         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
505 fail_mount:
506         gfs2_glock_dq_uninit(mount_gh);
507 fail:
508         while (sdp->sd_glockd_num--)
509                 kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]);
510
511         return error;
512 }
513
514 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
515                             u64 no_addr, const char *name)
516 {
517         struct gfs2_sbd *sdp = sb->s_fs_info;
518         struct dentry *dentry;
519         struct inode *inode;
520
521         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
522         if (IS_ERR(inode)) {
523                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
524                 return PTR_ERR(inode);
525         }
526         dentry = d_alloc_root(inode);
527         if (!dentry) {
528                 fs_err(sdp, "can't alloc %s dentry\n", name);
529                 iput(inode);
530                 return -ENOMEM;
531         }
532         dentry->d_op = &gfs2_dops;
533         *dptr = dentry;
534         return 0;
535 }
536
537 static int init_sb(struct gfs2_sbd *sdp, int silent)
538 {
539         struct super_block *sb = sdp->sd_vfs;
540         struct gfs2_holder sb_gh;
541         u64 no_addr;
542         int ret;
543
544         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
545                                 LM_ST_SHARED, 0, &sb_gh);
546         if (ret) {
547                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
548                 return ret;
549         }
550
551         ret = gfs2_read_sb(sdp, sb_gh.gh_gl, silent);
552         if (ret) {
553                 fs_err(sdp, "can't read superblock: %d\n", ret);
554                 goto out;
555         }
556
557         /* Set up the buffer cache and SB for real */
558         if (sdp->sd_sb.sb_bsize < bdev_hardsect_size(sb->s_bdev)) {
559                 ret = -EINVAL;
560                 fs_err(sdp, "FS block size (%u) is too small for device "
561                        "block size (%u)\n",
562                        sdp->sd_sb.sb_bsize, bdev_hardsect_size(sb->s_bdev));
563                 goto out;
564         }
565         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
566                 ret = -EINVAL;
567                 fs_err(sdp, "FS block size (%u) is too big for machine "
568                        "page size (%u)\n",
569                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
570                 goto out;
571         }
572         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
573
574         /* Get the root inode */
575         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
576         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
577         if (ret)
578                 goto out;
579
580         /* Get the master inode */
581         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
582         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
583         if (ret) {
584                 dput(sdp->sd_root_dir);
585                 goto out;
586         }
587         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
588 out:
589         gfs2_glock_dq_uninit(&sb_gh);
590         return ret;
591 }
592
593 /**
594  * map_journal_extents - create a reusable "extent" mapping from all logical
595  * blocks to all physical blocks for the given journal.  This will save
596  * us time when writing journal blocks.  Most journals will have only one
597  * extent that maps all their logical blocks.  That's because gfs2.mkfs
598  * arranges the journal blocks sequentially to maximize performance.
599  * So the extent would map the first block for the entire file length.
600  * However, gfs2_jadd can happen while file activity is happening, so
601  * those journals may not be sequential.  Less likely is the case where
602  * the users created their own journals by mounting the metafs and
603  * laying it out.  But it's still possible.  These journals might have
604  * several extents.
605  *
606  * TODO: This should be done in bigger chunks rather than one block at a time,
607  *       but since it's only done at mount time, I'm not worried about the
608  *       time it takes.
609  */
610 static int map_journal_extents(struct gfs2_sbd *sdp)
611 {
612         struct gfs2_jdesc *jd = sdp->sd_jdesc;
613         unsigned int lb;
614         u64 db, prev_db; /* logical block, disk block, prev disk block */
615         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
616         struct gfs2_journal_extent *jext = NULL;
617         struct buffer_head bh;
618         int rc = 0;
619
620         prev_db = 0;
621
622         for (lb = 0; lb < ip->i_disksize >> sdp->sd_sb.sb_bsize_shift; lb++) {
623                 bh.b_state = 0;
624                 bh.b_blocknr = 0;
625                 bh.b_size = 1 << ip->i_inode.i_blkbits;
626                 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
627                 db = bh.b_blocknr;
628                 if (rc || !db) {
629                         printk(KERN_INFO "GFS2 journal mapping error %d: lb="
630                                "%u db=%llu\n", rc, lb, (unsigned long long)db);
631                         break;
632                 }
633                 if (!prev_db || db != prev_db + 1) {
634                         jext = kzalloc(sizeof(struct gfs2_journal_extent),
635                                        GFP_KERNEL);
636                         if (!jext) {
637                                 printk(KERN_INFO "GFS2 error: out of memory "
638                                        "mapping journal extents.\n");
639                                 rc = -ENOMEM;
640                                 break;
641                         }
642                         jext->dblock = db;
643                         jext->lblock = lb;
644                         jext->blocks = 1;
645                         list_add_tail(&jext->extent_list, &jd->extent_list);
646                 } else {
647                         jext->blocks++;
648                 }
649                 prev_db = db;
650         }
651         return rc;
652 }
653
654 static void gfs2_lm_others_may_mount(struct gfs2_sbd *sdp)
655 {
656         if (!sdp->sd_lockstruct.ls_ops->lm_others_may_mount)
657                 return;
658         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
659                 sdp->sd_lockstruct.ls_ops->lm_others_may_mount(
660                                         sdp->sd_lockstruct.ls_lockspace);
661 }
662
663 static int init_journal(struct gfs2_sbd *sdp, int undo)
664 {
665         struct inode *master = sdp->sd_master_dir->d_inode;
666         struct gfs2_holder ji_gh;
667         struct task_struct *p;
668         struct gfs2_inode *ip;
669         int jindex = 1;
670         int error = 0;
671
672         if (undo) {
673                 jindex = 0;
674                 goto fail_recoverd;
675         }
676
677         sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
678         if (IS_ERR(sdp->sd_jindex)) {
679                 fs_err(sdp, "can't lookup journal index: %d\n", error);
680                 return PTR_ERR(sdp->sd_jindex);
681         }
682         ip = GFS2_I(sdp->sd_jindex);
683         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
684
685         /* Load in the journal index special file */
686
687         error = gfs2_jindex_hold(sdp, &ji_gh);
688         if (error) {
689                 fs_err(sdp, "can't read journal index: %d\n", error);
690                 goto fail;
691         }
692
693         error = -EINVAL;
694         if (!gfs2_jindex_size(sdp)) {
695                 fs_err(sdp, "no journals!\n");
696                 goto fail_jindex;
697         }
698
699         if (sdp->sd_args.ar_spectator) {
700                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
701                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
702         } else {
703                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
704                         fs_err(sdp, "can't mount journal #%u\n",
705                                sdp->sd_lockstruct.ls_jid);
706                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
707                                gfs2_jindex_size(sdp),
708                                gfs2_jindex_size(sdp) - 1);
709                         goto fail_jindex;
710                 }
711                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
712
713                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
714                                           &gfs2_journal_glops,
715                                           LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
716                                           &sdp->sd_journal_gh);
717                 if (error) {
718                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
719                         goto fail_jindex;
720                 }
721
722                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
723                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
724                                            LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
725                                            &sdp->sd_jinode_gh);
726                 if (error) {
727                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
728                                error);
729                         goto fail_journal_gh;
730                 }
731
732                 error = gfs2_jdesc_check(sdp->sd_jdesc);
733                 if (error) {
734                         fs_err(sdp, "my journal (%u) is bad: %d\n",
735                                sdp->sd_jdesc->jd_jid, error);
736                         goto fail_jinode_gh;
737                 }
738                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
739
740                 /* Map the extents for this journal's blocks */
741                 map_journal_extents(sdp);
742         }
743
744         if (sdp->sd_lockstruct.ls_first) {
745                 unsigned int x;
746                 for (x = 0; x < sdp->sd_journals; x++) {
747                         error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x));
748                         if (error) {
749                                 fs_err(sdp, "error recovering journal %u: %d\n",
750                                        x, error);
751                                 goto fail_jinode_gh;
752                         }
753                 }
754
755                 gfs2_lm_others_may_mount(sdp);
756         } else if (!sdp->sd_args.ar_spectator) {
757                 error = gfs2_recover_journal(sdp->sd_jdesc);
758                 if (error) {
759                         fs_err(sdp, "error recovering my journal: %d\n", error);
760                         goto fail_jinode_gh;
761                 }
762         }
763
764         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
765         gfs2_glock_dq_uninit(&ji_gh);
766         jindex = 0;
767
768         p = kthread_run(gfs2_recoverd, sdp, "gfs2_recoverd");
769         error = IS_ERR(p);
770         if (error) {
771                 fs_err(sdp, "can't start recoverd thread: %d\n", error);
772                 goto fail_jinode_gh;
773         }
774         sdp->sd_recoverd_process = p;
775
776         return 0;
777
778 fail_recoverd:
779         kthread_stop(sdp->sd_recoverd_process);
780 fail_jinode_gh:
781         if (!sdp->sd_args.ar_spectator)
782                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
783 fail_journal_gh:
784         if (!sdp->sd_args.ar_spectator)
785                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
786 fail_jindex:
787         gfs2_jindex_free(sdp);
788         if (jindex)
789                 gfs2_glock_dq_uninit(&ji_gh);
790 fail:
791         iput(sdp->sd_jindex);
792         return error;
793 }
794
795
796 static int init_inodes(struct gfs2_sbd *sdp, int undo)
797 {
798         int error = 0;
799         struct gfs2_inode *ip;
800         struct inode *master = sdp->sd_master_dir->d_inode;
801
802         if (undo)
803                 goto fail_qinode;
804
805         error = init_journal(sdp, undo);
806         if (error)
807                 goto fail;
808
809         /* Read in the master inode number inode */
810         sdp->sd_inum_inode = gfs2_lookup_simple(master, "inum");
811         if (IS_ERR(sdp->sd_inum_inode)) {
812                 error = PTR_ERR(sdp->sd_inum_inode);
813                 fs_err(sdp, "can't read in inum inode: %d\n", error);
814                 goto fail_journal;
815         }
816
817
818         /* Read in the master statfs inode */
819         sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
820         if (IS_ERR(sdp->sd_statfs_inode)) {
821                 error = PTR_ERR(sdp->sd_statfs_inode);
822                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
823                 goto fail_inum;
824         }
825
826         /* Read in the resource index inode */
827         sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
828         if (IS_ERR(sdp->sd_rindex)) {
829                 error = PTR_ERR(sdp->sd_rindex);
830                 fs_err(sdp, "can't get resource index inode: %d\n", error);
831                 goto fail_statfs;
832         }
833         ip = GFS2_I(sdp->sd_rindex);
834         set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
835         sdp->sd_rindex_uptodate = 0;
836
837         /* Read in the quota inode */
838         sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
839         if (IS_ERR(sdp->sd_quota_inode)) {
840                 error = PTR_ERR(sdp->sd_quota_inode);
841                 fs_err(sdp, "can't get quota file inode: %d\n", error);
842                 goto fail_rindex;
843         }
844         return 0;
845
846 fail_qinode:
847         iput(sdp->sd_quota_inode);
848 fail_rindex:
849         gfs2_clear_rgrpd(sdp);
850         iput(sdp->sd_rindex);
851 fail_statfs:
852         iput(sdp->sd_statfs_inode);
853 fail_inum:
854         iput(sdp->sd_inum_inode);
855 fail_journal:
856         init_journal(sdp, UNDO);
857 fail:
858         return error;
859 }
860
861 static int init_per_node(struct gfs2_sbd *sdp, int undo)
862 {
863         struct inode *pn = NULL;
864         char buf[30];
865         int error = 0;
866         struct gfs2_inode *ip;
867         struct inode *master = sdp->sd_master_dir->d_inode;
868
869         if (sdp->sd_args.ar_spectator)
870                 return 0;
871
872         if (undo)
873                 goto fail_qc_gh;
874
875         pn = gfs2_lookup_simple(master, "per_node");
876         if (IS_ERR(pn)) {
877                 error = PTR_ERR(pn);
878                 fs_err(sdp, "can't find per_node directory: %d\n", error);
879                 return error;
880         }
881
882         sprintf(buf, "inum_range%u", sdp->sd_jdesc->jd_jid);
883         sdp->sd_ir_inode = gfs2_lookup_simple(pn, buf);
884         if (IS_ERR(sdp->sd_ir_inode)) {
885                 error = PTR_ERR(sdp->sd_ir_inode);
886                 fs_err(sdp, "can't find local \"ir\" file: %d\n", error);
887                 goto fail;
888         }
889
890         sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
891         sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
892         if (IS_ERR(sdp->sd_sc_inode)) {
893                 error = PTR_ERR(sdp->sd_sc_inode);
894                 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
895                 goto fail_ir_i;
896         }
897
898         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
899         sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
900         if (IS_ERR(sdp->sd_qc_inode)) {
901                 error = PTR_ERR(sdp->sd_qc_inode);
902                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
903                 goto fail_ut_i;
904         }
905
906         iput(pn);
907         pn = NULL;
908
909         ip = GFS2_I(sdp->sd_ir_inode);
910         error = gfs2_glock_nq_init(ip->i_gl,
911                                    LM_ST_EXCLUSIVE, 0,
912                                    &sdp->sd_ir_gh);
913         if (error) {
914                 fs_err(sdp, "can't lock local \"ir\" file: %d\n", error);
915                 goto fail_qc_i;
916         }
917
918         ip = GFS2_I(sdp->sd_sc_inode);
919         error = gfs2_glock_nq_init(ip->i_gl,
920                                    LM_ST_EXCLUSIVE, 0,
921                                    &sdp->sd_sc_gh);
922         if (error) {
923                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
924                 goto fail_ir_gh;
925         }
926
927         ip = GFS2_I(sdp->sd_qc_inode);
928         error = gfs2_glock_nq_init(ip->i_gl,
929                                    LM_ST_EXCLUSIVE, 0,
930                                    &sdp->sd_qc_gh);
931         if (error) {
932                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
933                 goto fail_ut_gh;
934         }
935
936         return 0;
937
938 fail_qc_gh:
939         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
940 fail_ut_gh:
941         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
942 fail_ir_gh:
943         gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
944 fail_qc_i:
945         iput(sdp->sd_qc_inode);
946 fail_ut_i:
947         iput(sdp->sd_sc_inode);
948 fail_ir_i:
949         iput(sdp->sd_ir_inode);
950 fail:
951         if (pn)
952                 iput(pn);
953         return error;
954 }
955
956 static int init_threads(struct gfs2_sbd *sdp, int undo)
957 {
958         struct task_struct *p;
959         int error = 0;
960
961         if (undo)
962                 goto fail_quotad;
963
964         sdp->sd_log_flush_time = jiffies;
965         sdp->sd_jindex_refresh_time = jiffies;
966
967         p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
968         error = IS_ERR(p);
969         if (error) {
970                 fs_err(sdp, "can't start logd thread: %d\n", error);
971                 return error;
972         }
973         sdp->sd_logd_process = p;
974
975         p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
976         error = IS_ERR(p);
977         if (error) {
978                 fs_err(sdp, "can't start quotad thread: %d\n", error);
979                 goto fail;
980         }
981         sdp->sd_quotad_process = p;
982
983         return 0;
984
985
986 fail_quotad:
987         kthread_stop(sdp->sd_quotad_process);
988 fail:
989         kthread_stop(sdp->sd_logd_process);
990         return error;
991 }
992
993 /**
994  * gfs2_lm_mount - mount a locking protocol
995  * @sdp: the filesystem
996  * @args: mount arguements
997  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
998  *
999  * Returns: errno
1000  */
1001
1002 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
1003 {
1004         char *proto = sdp->sd_proto_name;
1005         char *table = sdp->sd_table_name;
1006         int flags = LM_MFLAG_CONV_NODROP;
1007         int error;
1008
1009         if (sdp->sd_args.ar_spectator)
1010                 flags |= LM_MFLAG_SPECTATOR;
1011
1012         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1013
1014         error = gfs2_mount_lockproto(proto, table, sdp->sd_args.ar_hostdata,
1015                                      gfs2_glock_cb, sdp,
1016                                      GFS2_MIN_LVB_SIZE, flags,
1017                                      &sdp->sd_lockstruct, &sdp->sd_kobj);
1018         if (error) {
1019                 fs_info(sdp, "can't mount proto=%s, table=%s, hostdata=%s\n",
1020                         proto, table, sdp->sd_args.ar_hostdata);
1021                 goto out;
1022         }
1023
1024         if (gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_ops) ||
1025             gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lvb_size >=
1026                                   GFS2_MIN_LVB_SIZE)) {
1027                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
1028                 goto out;
1029         }
1030
1031         if (sdp->sd_args.ar_spectator)
1032                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1033         else
1034                 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1035                          sdp->sd_lockstruct.ls_jid);
1036
1037         fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1038
1039         if ((sdp->sd_lockstruct.ls_flags & LM_LSFLAG_LOCAL) &&
1040             !sdp->sd_args.ar_ignore_local_fs) {
1041                 sdp->sd_args.ar_localflocks = 1;
1042                 sdp->sd_args.ar_localcaching = 1;
1043         }
1044
1045 out:
1046         return error;
1047 }
1048
1049 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1050 {
1051         if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
1052                 gfs2_unmount_lockproto(&sdp->sd_lockstruct);
1053 }
1054
1055 /**
1056  * fill_super - Read in superblock
1057  * @sb: The VFS superblock
1058  * @data: Mount options
1059  * @silent: Don't complain if it's not a GFS2 filesystem
1060  *
1061  * Returns: errno
1062  */
1063
1064 static int fill_super(struct super_block *sb, void *data, int silent)
1065 {
1066         struct gfs2_sbd *sdp;
1067         struct gfs2_holder mount_gh;
1068         int error;
1069
1070         sdp = init_sbd(sb);
1071         if (!sdp) {
1072                 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1073                 return -ENOMEM;
1074         }
1075
1076         error = gfs2_mount_args(sdp, (char *)data, 0);
1077         if (error) {
1078                 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1079                 goto fail;
1080         }
1081
1082         sb->s_magic = GFS2_MAGIC;
1083         sb->s_op = &gfs2_super_ops;
1084         sb->s_export_op = &gfs2_export_ops;
1085         sb->s_time_gran = 1;
1086         sb->s_maxbytes = MAX_LFS_FILESIZE;
1087
1088         /* Set up the buffer cache and fill in some fake block size values
1089            to allow us to read-in the on-disk superblock. */
1090         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1091         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1092         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1093                                GFS2_BASIC_BLOCK_SHIFT;
1094         sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1095
1096         error = init_names(sdp, silent);
1097         if (error)
1098                 goto fail;
1099
1100         gfs2_create_debugfs_file(sdp);
1101
1102         error = gfs2_sys_fs_add(sdp);
1103         if (error)
1104                 goto fail;
1105
1106         error = gfs2_lm_mount(sdp, silent);
1107         if (error)
1108                 goto fail_sys;
1109
1110         error = init_locking(sdp, &mount_gh, DO);
1111         if (error)
1112                 goto fail_lm;
1113
1114         error = init_sb(sdp, silent);
1115         if (error)
1116                 goto fail_locking;
1117
1118         error = init_inodes(sdp, DO);
1119         if (error)
1120                 goto fail_sb;
1121
1122         error = init_per_node(sdp, DO);
1123         if (error)
1124                 goto fail_inodes;
1125
1126         error = gfs2_statfs_init(sdp);
1127         if (error) {
1128                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1129                 goto fail_per_node;
1130         }
1131
1132         error = init_threads(sdp, DO);
1133         if (error)
1134                 goto fail_per_node;
1135
1136         if (!(sb->s_flags & MS_RDONLY)) {
1137                 error = gfs2_make_fs_rw(sdp);
1138                 if (error) {
1139                         fs_err(sdp, "can't make FS RW: %d\n", error);
1140                         goto fail_threads;
1141                 }
1142         }
1143
1144         gfs2_glock_dq_uninit(&mount_gh);
1145
1146         return 0;
1147
1148 fail_threads:
1149         init_threads(sdp, UNDO);
1150 fail_per_node:
1151         init_per_node(sdp, UNDO);
1152 fail_inodes:
1153         init_inodes(sdp, UNDO);
1154 fail_sb:
1155         if (sdp->sd_root_dir)
1156                 dput(sdp->sd_root_dir);
1157         if (sdp->sd_master_dir)
1158                 dput(sdp->sd_master_dir);
1159         sb->s_root = NULL;
1160 fail_locking:
1161         init_locking(sdp, &mount_gh, UNDO);
1162 fail_lm:
1163         gfs2_gl_hash_clear(sdp);
1164         gfs2_lm_unmount(sdp);
1165         while (invalidate_inodes(sb))
1166                 yield();
1167 fail_sys:
1168         gfs2_sys_fs_del(sdp);
1169 fail:
1170         gfs2_delete_debugfs_file(sdp);
1171         kfree(sdp);
1172         sb->s_fs_info = NULL;
1173         return error;
1174 }
1175
1176 static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
1177                        const char *dev_name, void *data, struct vfsmount *mnt)
1178 {
1179         return get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
1180 }
1181
1182 static struct super_block *get_gfs2_sb(const char *dev_name)
1183 {
1184         struct super_block *sb;
1185         struct nameidata nd;
1186         int error;
1187
1188         error = path_lookup(dev_name, LOOKUP_FOLLOW, &nd);
1189         if (error) {
1190                 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1191                        dev_name, error);
1192                 return NULL;
1193         }
1194         sb = nd.path.dentry->d_inode->i_sb;
1195         if (sb && (sb->s_type == &gfs2_fs_type))
1196                 atomic_inc(&sb->s_active);
1197         else
1198                 sb = NULL;
1199         path_put(&nd.path);
1200         return sb;
1201 }
1202
1203 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
1204                             const char *dev_name, void *data, struct vfsmount *mnt)
1205 {
1206         struct super_block *sb = NULL;
1207         struct gfs2_sbd *sdp;
1208
1209         sb = get_gfs2_sb(dev_name);
1210         if (!sb) {
1211                 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1212                 return -ENOENT;
1213         }
1214         sdp = sb->s_fs_info;
1215         mnt->mnt_sb = sb;
1216         mnt->mnt_root = dget(sdp->sd_master_dir);
1217         return 0;
1218 }
1219
1220 static void gfs2_kill_sb(struct super_block *sb)
1221 {
1222         struct gfs2_sbd *sdp = sb->s_fs_info;
1223         if (sdp) {
1224                 gfs2_meta_syncfs(sdp);
1225                 dput(sdp->sd_root_dir);
1226                 dput(sdp->sd_master_dir);
1227                 sdp->sd_root_dir = NULL;
1228                 sdp->sd_master_dir = NULL;
1229         }
1230         shrink_dcache_sb(sb);
1231         kill_block_super(sb);
1232         if (sdp)
1233                 gfs2_delete_debugfs_file(sdp);
1234 }
1235
1236 struct file_system_type gfs2_fs_type = {
1237         .name = "gfs2",
1238         .fs_flags = FS_REQUIRES_DEV,
1239         .get_sb = gfs2_get_sb,
1240         .kill_sb = gfs2_kill_sb,
1241         .owner = THIS_MODULE,
1242 };
1243
1244 struct file_system_type gfs2meta_fs_type = {
1245         .name = "gfs2meta",
1246         .fs_flags = FS_REQUIRES_DEV,
1247         .get_sb = gfs2_get_sb_meta,
1248         .owner = THIS_MODULE,
1249 };
1250