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