]> Git Repo - linux.git/blob - fs/gfs2/ops_fstype.c
cifs: Add a tracepoint to track credits involved in R/W requests
[linux.git] / fs / gfs2 / ops_fstype.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
4  * Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
5  */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
24 #include <linux/fs_parser.h>
25
26 #include "gfs2.h"
27 #include "incore.h"
28 #include "bmap.h"
29 #include "glock.h"
30 #include "glops.h"
31 #include "inode.h"
32 #include "recovery.h"
33 #include "rgrp.h"
34 #include "super.h"
35 #include "sys.h"
36 #include "util.h"
37 #include "log.h"
38 #include "quota.h"
39 #include "dir.h"
40 #include "meta_io.h"
41 #include "trace_gfs2.h"
42 #include "lops.h"
43
44 #define DO 0
45 #define UNDO 1
46
47 /**
48  * gfs2_tune_init - Fill a gfs2_tune structure with default values
49  * @gt: tune
50  *
51  */
52
53 static void gfs2_tune_init(struct gfs2_tune *gt)
54 {
55         spin_lock_init(&gt->gt_spin);
56
57         gt->gt_quota_warn_period = 10;
58         gt->gt_quota_scale_num = 1;
59         gt->gt_quota_scale_den = 1;
60         gt->gt_new_files_jdata = 0;
61         gt->gt_max_readahead = BIT(18);
62         gt->gt_complain_secs = 10;
63 }
64
65 void free_sbd(struct gfs2_sbd *sdp)
66 {
67         if (sdp->sd_lkstats)
68                 free_percpu(sdp->sd_lkstats);
69         kfree(sdp);
70 }
71
72 static struct gfs2_sbd *init_sbd(struct super_block *sb)
73 {
74         struct gfs2_sbd *sdp;
75         struct address_space *mapping;
76
77         sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
78         if (!sdp)
79                 return NULL;
80
81         sdp->sd_vfs = sb;
82         sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
83         if (!sdp->sd_lkstats)
84                 goto fail;
85         sb->s_fs_info = sdp;
86
87         set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
88         gfs2_tune_init(&sdp->sd_tune);
89
90         init_waitqueue_head(&sdp->sd_kill_wait);
91         init_waitqueue_head(&sdp->sd_async_glock_wait);
92         atomic_set(&sdp->sd_glock_disposal, 0);
93         init_completion(&sdp->sd_locking_init);
94         init_completion(&sdp->sd_wdack);
95         spin_lock_init(&sdp->sd_statfs_spin);
96
97         spin_lock_init(&sdp->sd_rindex_spin);
98         sdp->sd_rindex_tree.rb_node = NULL;
99
100         INIT_LIST_HEAD(&sdp->sd_jindex_list);
101         spin_lock_init(&sdp->sd_jindex_spin);
102         mutex_init(&sdp->sd_jindex_mutex);
103         init_completion(&sdp->sd_journal_ready);
104
105         INIT_LIST_HEAD(&sdp->sd_quota_list);
106         mutex_init(&sdp->sd_quota_mutex);
107         mutex_init(&sdp->sd_quota_sync_mutex);
108         init_waitqueue_head(&sdp->sd_quota_wait);
109         spin_lock_init(&sdp->sd_bitmap_lock);
110
111         INIT_LIST_HEAD(&sdp->sd_sc_inodes_list);
112
113         mapping = &sdp->sd_aspace;
114
115         address_space_init_once(mapping);
116         mapping->a_ops = &gfs2_rgrp_aops;
117         mapping->host = sb->s_bdev->bd_mapping->host;
118         mapping->flags = 0;
119         mapping_set_gfp_mask(mapping, GFP_NOFS);
120         mapping->i_private_data = NULL;
121         mapping->writeback_index = 0;
122
123         spin_lock_init(&sdp->sd_log_lock);
124         atomic_set(&sdp->sd_log_pinned, 0);
125         INIT_LIST_HEAD(&sdp->sd_log_revokes);
126         INIT_LIST_HEAD(&sdp->sd_log_ordered);
127         spin_lock_init(&sdp->sd_ordered_lock);
128
129         init_waitqueue_head(&sdp->sd_log_waitq);
130         init_waitqueue_head(&sdp->sd_logd_waitq);
131         spin_lock_init(&sdp->sd_ail_lock);
132         INIT_LIST_HEAD(&sdp->sd_ail1_list);
133         INIT_LIST_HEAD(&sdp->sd_ail2_list);
134
135         init_rwsem(&sdp->sd_log_flush_lock);
136         atomic_set(&sdp->sd_log_in_flight, 0);
137         init_waitqueue_head(&sdp->sd_log_flush_wait);
138         mutex_init(&sdp->sd_freeze_mutex);
139         INIT_LIST_HEAD(&sdp->sd_dead_glocks);
140
141         return sdp;
142
143 fail:
144         free_sbd(sdp);
145         return NULL;
146 }
147
148 /**
149  * gfs2_check_sb - Check superblock
150  * @sdp: the filesystem
151  * @silent: Don't print a message if the check fails
152  *
153  * Checks the version code of the FS is one that we understand how to
154  * read and that the sizes of the various on-disk structures have not
155  * changed.
156  */
157
158 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
159 {
160         struct gfs2_sb_host *sb = &sdp->sd_sb;
161
162         if (sb->sb_magic != GFS2_MAGIC ||
163             sb->sb_type != GFS2_METATYPE_SB) {
164                 if (!silent)
165                         pr_warn("not a GFS2 filesystem\n");
166                 return -EINVAL;
167         }
168
169         if (sb->sb_fs_format < GFS2_FS_FORMAT_MIN ||
170             sb->sb_fs_format > GFS2_FS_FORMAT_MAX ||
171             sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
172                 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
173                 return -EINVAL;
174         }
175
176         if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
177             (sb->sb_bsize & (sb->sb_bsize - 1))) {
178                 pr_warn("Invalid block size\n");
179                 return -EINVAL;
180         }
181         if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
182                 pr_warn("Invalid block size shift\n");
183                 return -EINVAL;
184         }
185         return 0;
186 }
187
188 static void gfs2_sb_in(struct gfs2_sbd *sdp, const struct gfs2_sb *str)
189 {
190         struct gfs2_sb_host *sb = &sdp->sd_sb;
191         struct super_block *s = sdp->sd_vfs;
192
193         sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
194         sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
195         sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
196         sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
197         sb->sb_bsize = be32_to_cpu(str->sb_bsize);
198         sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
199         sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
200         sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
201         sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
202         sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
203
204         memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
205         memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
206         super_set_uuid(s, str->sb_uuid, 16);
207 }
208
209 /**
210  * gfs2_read_super - Read the gfs2 super block from disk
211  * @sdp: The GFS2 super block
212  * @sector: The location of the super block
213  * @silent: Don't print a message if the check fails
214  *
215  * This uses the bio functions to read the super block from disk
216  * because we want to be 100% sure that we never read cached data.
217  * A super block is read twice only during each GFS2 mount and is
218  * never written to by the filesystem. The first time its read no
219  * locks are held, and the only details which are looked at are those
220  * relating to the locking protocol. Once locking is up and working,
221  * the sb is read again under the lock to establish the location of
222  * the master directory (contains pointers to journals etc) and the
223  * root directory.
224  *
225  * Returns: 0 on success or error
226  */
227
228 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
229 {
230         struct super_block *sb = sdp->sd_vfs;
231         struct page *page;
232         struct bio_vec bvec;
233         struct bio bio;
234         int err;
235
236         page = alloc_page(GFP_KERNEL);
237         if (unlikely(!page))
238                 return -ENOMEM;
239
240         bio_init(&bio, sb->s_bdev, &bvec, 1, REQ_OP_READ | REQ_META);
241         bio.bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
242         __bio_add_page(&bio, page, PAGE_SIZE, 0);
243
244         err = submit_bio_wait(&bio);
245         if (err) {
246                 pr_warn("error %d reading superblock\n", err);
247                 __free_page(page);
248                 return err;
249         }
250         gfs2_sb_in(sdp, page_address(page));
251         __free_page(page);
252         return gfs2_check_sb(sdp, silent);
253 }
254
255 /**
256  * gfs2_read_sb - Read super block
257  * @sdp: The GFS2 superblock
258  * @silent: Don't print message if mount fails
259  *
260  */
261
262 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
263 {
264         u32 hash_blocks, ind_blocks, leaf_blocks;
265         u32 tmp_blocks;
266         unsigned int x;
267         int error;
268
269         error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
270         if (error) {
271                 if (!silent)
272                         fs_err(sdp, "can't read superblock\n");
273                 return error;
274         }
275
276         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
277         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
278         sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
279                           sizeof(struct gfs2_dinode)) / sizeof(u64);
280         sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
281                           sizeof(struct gfs2_meta_header)) / sizeof(u64);
282         sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
283                           sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
284         sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
285         sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
286         sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
287         sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
288         sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
289                                 sizeof(struct gfs2_meta_header)) /
290                                 sizeof(struct gfs2_quota_change);
291         sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
292                                      sizeof(struct gfs2_meta_header))
293                 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
294
295         /*
296          * We always keep at least one block reserved for revokes in
297          * transactions.  This greatly simplifies allocating additional
298          * revoke blocks.
299          */
300         atomic_set(&sdp->sd_log_revokes_available, sdp->sd_ldptrs);
301
302         /* Compute maximum reservation required to add a entry to a directory */
303
304         hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
305                              sdp->sd_jbsize);
306
307         ind_blocks = 0;
308         for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
309                 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
310                 ind_blocks += tmp_blocks;
311         }
312
313         leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
314
315         sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
316
317         sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
318                                 sizeof(struct gfs2_dinode);
319         sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
320         for (x = 2;; x++) {
321                 u64 space, d;
322                 u32 m;
323
324                 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
325                 d = space;
326                 m = do_div(d, sdp->sd_inptrs);
327
328                 if (d != sdp->sd_heightsize[x - 1] || m)
329                         break;
330                 sdp->sd_heightsize[x] = space;
331         }
332         sdp->sd_max_height = x;
333         sdp->sd_heightsize[x] = ~0;
334         gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
335
336         sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
337                                       sizeof(struct gfs2_leaf)) /
338                                      GFS2_MIN_DIRENT_SIZE;
339         return 0;
340 }
341
342 static int init_names(struct gfs2_sbd *sdp, int silent)
343 {
344         char *proto, *table;
345         int error = 0;
346
347         proto = sdp->sd_args.ar_lockproto;
348         table = sdp->sd_args.ar_locktable;
349
350         /*  Try to autodetect  */
351
352         if (!proto[0] || !table[0]) {
353                 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
354                 if (error)
355                         return error;
356
357                 if (!proto[0])
358                         proto = sdp->sd_sb.sb_lockproto;
359                 if (!table[0])
360                         table = sdp->sd_sb.sb_locktable;
361         }
362
363         if (!table[0])
364                 table = sdp->sd_vfs->s_id;
365
366         BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
367
368         strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
369         strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
370
371         table = sdp->sd_table_name;
372         while ((table = strchr(table, '/')))
373                 *table = '_';
374
375         return error;
376 }
377
378 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
379                         int undo)
380 {
381         int error = 0;
382
383         if (undo)
384                 goto fail_trans;
385
386         error = gfs2_glock_nq_num(sdp,
387                                   GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
388                                   LM_ST_EXCLUSIVE,
389                                   LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
390                                   mount_gh);
391         if (error) {
392                 fs_err(sdp, "can't acquire mount glock: %d\n", error);
393                 goto fail;
394         }
395
396         error = gfs2_glock_nq_num(sdp,
397                                   GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
398                                   LM_ST_SHARED,
399                                   LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
400                                   &sdp->sd_live_gh);
401         if (error) {
402                 fs_err(sdp, "can't acquire live glock: %d\n", error);
403                 goto fail_mount;
404         }
405
406         error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
407                                CREATE, &sdp->sd_rename_gl);
408         if (error) {
409                 fs_err(sdp, "can't create rename glock: %d\n", error);
410                 goto fail_live;
411         }
412
413         error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
414                                CREATE, &sdp->sd_freeze_gl);
415         if (error) {
416                 fs_err(sdp, "can't create freeze glock: %d\n", error);
417                 goto fail_rename;
418         }
419
420         return 0;
421
422 fail_trans:
423         gfs2_glock_put(sdp->sd_freeze_gl);
424 fail_rename:
425         gfs2_glock_put(sdp->sd_rename_gl);
426 fail_live:
427         gfs2_glock_dq_uninit(&sdp->sd_live_gh);
428 fail_mount:
429         gfs2_glock_dq_uninit(mount_gh);
430 fail:
431         return error;
432 }
433
434 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
435                             u64 no_addr, const char *name)
436 {
437         struct gfs2_sbd *sdp = sb->s_fs_info;
438         struct dentry *dentry;
439         struct inode *inode;
440
441         inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
442                                   GFS2_BLKST_FREE /* ignore */);
443         if (IS_ERR(inode)) {
444                 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
445                 return PTR_ERR(inode);
446         }
447         dentry = d_make_root(inode);
448         if (!dentry) {
449                 fs_err(sdp, "can't alloc %s dentry\n", name);
450                 return -ENOMEM;
451         }
452         *dptr = dentry;
453         return 0;
454 }
455
456 static int init_sb(struct gfs2_sbd *sdp, int silent)
457 {
458         struct super_block *sb = sdp->sd_vfs;
459         struct gfs2_holder sb_gh;
460         u64 no_addr;
461         int ret;
462
463         ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
464                                 LM_ST_SHARED, 0, &sb_gh);
465         if (ret) {
466                 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
467                 return ret;
468         }
469
470         ret = gfs2_read_sb(sdp, silent);
471         if (ret) {
472                 fs_err(sdp, "can't read superblock: %d\n", ret);
473                 goto out;
474         }
475
476         switch(sdp->sd_sb.sb_fs_format) {
477         case GFS2_FS_FORMAT_MAX:
478                 sb->s_xattr = gfs2_xattr_handlers_max;
479                 break;
480
481         case GFS2_FS_FORMAT_MIN:
482                 sb->s_xattr = gfs2_xattr_handlers_min;
483                 break;
484
485         default:
486                 BUG();
487         }
488
489         /* Set up the buffer cache and SB for real */
490         if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
491                 ret = -EINVAL;
492                 fs_err(sdp, "FS block size (%u) is too small for device "
493                        "block size (%u)\n",
494                        sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
495                 goto out;
496         }
497         if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
498                 ret = -EINVAL;
499                 fs_err(sdp, "FS block size (%u) is too big for machine "
500                        "page size (%u)\n",
501                        sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
502                 goto out;
503         }
504         sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
505
506         /* Get the root inode */
507         no_addr = sdp->sd_sb.sb_root_dir.no_addr;
508         ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
509         if (ret)
510                 goto out;
511
512         /* Get the master inode */
513         no_addr = sdp->sd_sb.sb_master_dir.no_addr;
514         ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
515         if (ret) {
516                 dput(sdp->sd_root_dir);
517                 goto out;
518         }
519         sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
520 out:
521         gfs2_glock_dq_uninit(&sb_gh);
522         return ret;
523 }
524
525 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
526 {
527         char *message = "FIRSTMOUNT=Done";
528         char *envp[] = { message, NULL };
529
530         fs_info(sdp, "first mount done, others may mount\n");
531
532         if (sdp->sd_lockstruct.ls_ops->lm_first_done)
533                 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
534
535         kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
536 }
537
538 /**
539  * gfs2_jindex_hold - Grab a lock on the jindex
540  * @sdp: The GFS2 superblock
541  * @ji_gh: the holder for the jindex glock
542  *
543  * Returns: errno
544  */
545
546 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
547 {
548         struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
549         struct qstr name;
550         char buf[20];
551         struct gfs2_jdesc *jd;
552         int error;
553
554         name.name = buf;
555
556         mutex_lock(&sdp->sd_jindex_mutex);
557
558         for (;;) {
559                 struct gfs2_inode *jip;
560
561                 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
562                 if (error)
563                         break;
564
565                 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
566                 name.hash = gfs2_disk_hash(name.name, name.len);
567
568                 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
569                 if (error == -ENOENT) {
570                         error = 0;
571                         break;
572                 }
573
574                 gfs2_glock_dq_uninit(ji_gh);
575
576                 if (error)
577                         break;
578
579                 error = -ENOMEM;
580                 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
581                 if (!jd)
582                         break;
583
584                 INIT_LIST_HEAD(&jd->extent_list);
585                 INIT_LIST_HEAD(&jd->jd_revoke_list);
586
587                 INIT_WORK(&jd->jd_work, gfs2_recover_func);
588                 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
589                 if (IS_ERR_OR_NULL(jd->jd_inode)) {
590                         if (!jd->jd_inode)
591                                 error = -ENOENT;
592                         else
593                                 error = PTR_ERR(jd->jd_inode);
594                         kfree(jd);
595                         break;
596                 }
597
598                 d_mark_dontcache(jd->jd_inode);
599                 spin_lock(&sdp->sd_jindex_spin);
600                 jd->jd_jid = sdp->sd_journals++;
601                 jip = GFS2_I(jd->jd_inode);
602                 jd->jd_no_addr = jip->i_no_addr;
603                 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
604                 spin_unlock(&sdp->sd_jindex_spin);
605         }
606
607         mutex_unlock(&sdp->sd_jindex_mutex);
608
609         return error;
610 }
611
612 /**
613  * init_statfs - look up and initialize master and local (per node) statfs inodes
614  * @sdp: The GFS2 superblock
615  *
616  * This should be called after the jindex is initialized in init_journal() and
617  * before gfs2_journal_recovery() is called because we need to be able to write
618  * to these inodes during recovery.
619  *
620  * Returns: errno
621  */
622 static int init_statfs(struct gfs2_sbd *sdp)
623 {
624         int error = 0;
625         struct inode *master = d_inode(sdp->sd_master_dir);
626         struct inode *pn = NULL;
627         char buf[30];
628         struct gfs2_jdesc *jd;
629         struct gfs2_inode *ip;
630
631         sdp->sd_statfs_inode = gfs2_lookup_meta(master, "statfs");
632         if (IS_ERR(sdp->sd_statfs_inode)) {
633                 error = PTR_ERR(sdp->sd_statfs_inode);
634                 fs_err(sdp, "can't read in statfs inode: %d\n", error);
635                 goto out;
636         }
637         if (sdp->sd_args.ar_spectator)
638                 goto out;
639
640         pn = gfs2_lookup_meta(master, "per_node");
641         if (IS_ERR(pn)) {
642                 error = PTR_ERR(pn);
643                 fs_err(sdp, "can't find per_node directory: %d\n", error);
644                 goto put_statfs;
645         }
646
647         /* For each jid, lookup the corresponding local statfs inode in the
648          * per_node metafs directory and save it in the sdp->sd_sc_inodes_list. */
649         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
650                 struct local_statfs_inode *lsi =
651                         kmalloc(sizeof(struct local_statfs_inode), GFP_NOFS);
652                 if (!lsi) {
653                         error = -ENOMEM;
654                         goto free_local;
655                 }
656                 sprintf(buf, "statfs_change%u", jd->jd_jid);
657                 lsi->si_sc_inode = gfs2_lookup_meta(pn, buf);
658                 if (IS_ERR(lsi->si_sc_inode)) {
659                         error = PTR_ERR(lsi->si_sc_inode);
660                         fs_err(sdp, "can't find local \"sc\" file#%u: %d\n",
661                                jd->jd_jid, error);
662                         kfree(lsi);
663                         goto free_local;
664                 }
665                 lsi->si_jid = jd->jd_jid;
666                 if (jd->jd_jid == sdp->sd_jdesc->jd_jid)
667                         sdp->sd_sc_inode = lsi->si_sc_inode;
668
669                 list_add_tail(&lsi->si_list, &sdp->sd_sc_inodes_list);
670         }
671
672         iput(pn);
673         pn = NULL;
674         ip = GFS2_I(sdp->sd_sc_inode);
675         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
676                                    &sdp->sd_sc_gh);
677         if (error) {
678                 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
679                 goto free_local;
680         }
681         /* read in the local statfs buffer - other nodes don't change it. */
682         error = gfs2_meta_inode_buffer(ip, &sdp->sd_sc_bh);
683         if (error) {
684                 fs_err(sdp, "Cannot read in local statfs: %d\n", error);
685                 goto unlock_sd_gh;
686         }
687         return 0;
688
689 unlock_sd_gh:
690         gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
691 free_local:
692         free_local_statfs_inodes(sdp);
693         iput(pn);
694 put_statfs:
695         iput(sdp->sd_statfs_inode);
696 out:
697         return error;
698 }
699
700 /* Uninitialize and free up memory used by the list of statfs inodes */
701 static void uninit_statfs(struct gfs2_sbd *sdp)
702 {
703         if (!sdp->sd_args.ar_spectator) {
704                 brelse(sdp->sd_sc_bh);
705                 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
706                 free_local_statfs_inodes(sdp);
707         }
708         iput(sdp->sd_statfs_inode);
709 }
710
711 static int init_journal(struct gfs2_sbd *sdp, int undo)
712 {
713         struct inode *master = d_inode(sdp->sd_master_dir);
714         struct gfs2_holder ji_gh;
715         struct gfs2_inode *ip;
716         int error = 0;
717
718         gfs2_holder_mark_uninitialized(&ji_gh);
719         if (undo)
720                 goto fail_statfs;
721
722         sdp->sd_jindex = gfs2_lookup_meta(master, "jindex");
723         if (IS_ERR(sdp->sd_jindex)) {
724                 fs_err(sdp, "can't lookup journal index: %d\n", error);
725                 return PTR_ERR(sdp->sd_jindex);
726         }
727
728         /* Load in the journal index special file */
729
730         error = gfs2_jindex_hold(sdp, &ji_gh);
731         if (error) {
732                 fs_err(sdp, "can't read journal index: %d\n", error);
733                 goto fail;
734         }
735
736         error = -EUSERS;
737         if (!gfs2_jindex_size(sdp)) {
738                 fs_err(sdp, "no journals!\n");
739                 goto fail_jindex;
740         }
741
742         atomic_set(&sdp->sd_log_blks_needed, 0);
743         if (sdp->sd_args.ar_spectator) {
744                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
745                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
746                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
747                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
748         } else {
749                 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
750                         fs_err(sdp, "can't mount journal #%u\n",
751                                sdp->sd_lockstruct.ls_jid);
752                         fs_err(sdp, "there are only %u journals (0 - %u)\n",
753                                gfs2_jindex_size(sdp),
754                                gfs2_jindex_size(sdp) - 1);
755                         goto fail_jindex;
756                 }
757                 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
758
759                 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
760                                           &gfs2_journal_glops,
761                                           LM_ST_EXCLUSIVE,
762                                           LM_FLAG_NOEXP | GL_NOCACHE | GL_NOPID,
763                                           &sdp->sd_journal_gh);
764                 if (error) {
765                         fs_err(sdp, "can't acquire journal glock: %d\n", error);
766                         goto fail_jindex;
767                 }
768
769                 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
770                 sdp->sd_jinode_gl = ip->i_gl;
771                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
772                                            LM_FLAG_NOEXP | GL_EXACT |
773                                            GL_NOCACHE | GL_NOPID,
774                                            &sdp->sd_jinode_gh);
775                 if (error) {
776                         fs_err(sdp, "can't acquire journal inode glock: %d\n",
777                                error);
778                         goto fail_journal_gh;
779                 }
780
781                 error = gfs2_jdesc_check(sdp->sd_jdesc);
782                 if (error) {
783                         fs_err(sdp, "my journal (%u) is bad: %d\n",
784                                sdp->sd_jdesc->jd_jid, error);
785                         goto fail_jinode_gh;
786                 }
787                 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
788                 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
789                 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
790
791                 /* Map the extents for this journal's blocks */
792                 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
793         }
794         trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
795
796         /* Lookup statfs inodes here so journal recovery can use them. */
797         error = init_statfs(sdp);
798         if (error)
799                 goto fail_jinode_gh;
800
801         if (sdp->sd_lockstruct.ls_first) {
802                 unsigned int x;
803                 for (x = 0; x < sdp->sd_journals; x++) {
804                         struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
805
806                         if (sdp->sd_args.ar_spectator) {
807                                 error = check_journal_clean(sdp, jd, true);
808                                 if (error)
809                                         goto fail_statfs;
810                                 continue;
811                         }
812                         error = gfs2_recover_journal(jd, true);
813                         if (error) {
814                                 fs_err(sdp, "error recovering journal %u: %d\n",
815                                        x, error);
816                                 goto fail_statfs;
817                         }
818                 }
819
820                 gfs2_others_may_mount(sdp);
821         } else if (!sdp->sd_args.ar_spectator) {
822                 error = gfs2_recover_journal(sdp->sd_jdesc, true);
823                 if (error) {
824                         fs_err(sdp, "error recovering my journal: %d\n", error);
825                         goto fail_statfs;
826                 }
827         }
828
829         sdp->sd_log_idle = 1;
830         set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
831         gfs2_glock_dq_uninit(&ji_gh);
832         INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
833         return 0;
834
835 fail_statfs:
836         uninit_statfs(sdp);
837 fail_jinode_gh:
838         /* A withdraw may have done dq/uninit so now we need to check it */
839         if (!sdp->sd_args.ar_spectator &&
840             gfs2_holder_initialized(&sdp->sd_jinode_gh))
841                 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
842 fail_journal_gh:
843         if (!sdp->sd_args.ar_spectator &&
844             gfs2_holder_initialized(&sdp->sd_journal_gh))
845                 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
846 fail_jindex:
847         gfs2_jindex_free(sdp);
848         if (gfs2_holder_initialized(&ji_gh))
849                 gfs2_glock_dq_uninit(&ji_gh);
850 fail:
851         iput(sdp->sd_jindex);
852         return error;
853 }
854
855 static struct lock_class_key gfs2_quota_imutex_key;
856
857 static int init_inodes(struct gfs2_sbd *sdp, int undo)
858 {
859         int error = 0;
860         struct inode *master = d_inode(sdp->sd_master_dir);
861
862         if (undo)
863                 goto fail_qinode;
864
865         error = init_journal(sdp, undo);
866         complete_all(&sdp->sd_journal_ready);
867         if (error)
868                 goto fail;
869
870         /* Read in the resource index inode */
871         sdp->sd_rindex = gfs2_lookup_meta(master, "rindex");
872         if (IS_ERR(sdp->sd_rindex)) {
873                 error = PTR_ERR(sdp->sd_rindex);
874                 fs_err(sdp, "can't get resource index inode: %d\n", error);
875                 goto fail_journal;
876         }
877         sdp->sd_rindex_uptodate = 0;
878
879         /* Read in the quota inode */
880         sdp->sd_quota_inode = gfs2_lookup_meta(master, "quota");
881         if (IS_ERR(sdp->sd_quota_inode)) {
882                 error = PTR_ERR(sdp->sd_quota_inode);
883                 fs_err(sdp, "can't get quota file inode: %d\n", error);
884                 goto fail_rindex;
885         }
886         /*
887          * i_rwsem on quota files is special. Since this inode is hidden system
888          * file, we are safe to define locking ourselves.
889          */
890         lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
891                           &gfs2_quota_imutex_key);
892
893         error = gfs2_rindex_update(sdp);
894         if (error)
895                 goto fail_qinode;
896
897         return 0;
898
899 fail_qinode:
900         iput(sdp->sd_quota_inode);
901 fail_rindex:
902         gfs2_clear_rgrpd(sdp);
903         iput(sdp->sd_rindex);
904 fail_journal:
905         init_journal(sdp, UNDO);
906 fail:
907         return error;
908 }
909
910 static int init_per_node(struct gfs2_sbd *sdp, int undo)
911 {
912         struct inode *pn = NULL;
913         char buf[30];
914         int error = 0;
915         struct gfs2_inode *ip;
916         struct inode *master = d_inode(sdp->sd_master_dir);
917
918         if (sdp->sd_args.ar_spectator)
919                 return 0;
920
921         if (undo)
922                 goto fail_qc_gh;
923
924         pn = gfs2_lookup_meta(master, "per_node");
925         if (IS_ERR(pn)) {
926                 error = PTR_ERR(pn);
927                 fs_err(sdp, "can't find per_node directory: %d\n", error);
928                 return error;
929         }
930
931         sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
932         sdp->sd_qc_inode = gfs2_lookup_meta(pn, buf);
933         if (IS_ERR(sdp->sd_qc_inode)) {
934                 error = PTR_ERR(sdp->sd_qc_inode);
935                 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
936                 goto fail_ut_i;
937         }
938
939         iput(pn);
940         pn = NULL;
941
942         ip = GFS2_I(sdp->sd_qc_inode);
943         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_NOPID,
944                                    &sdp->sd_qc_gh);
945         if (error) {
946                 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
947                 goto fail_qc_i;
948         }
949
950         return 0;
951
952 fail_qc_gh:
953         gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
954 fail_qc_i:
955         iput(sdp->sd_qc_inode);
956 fail_ut_i:
957         iput(pn);
958         return error;
959 }
960
961 static const match_table_t nolock_tokens = {
962         { Opt_jid, "jid=%d", },
963         { Opt_err, NULL },
964 };
965
966 static const struct lm_lockops nolock_ops = {
967         .lm_proto_name = "lock_nolock",
968         .lm_put_lock = gfs2_glock_free,
969         .lm_tokens = &nolock_tokens,
970 };
971
972 /**
973  * gfs2_lm_mount - mount a locking protocol
974  * @sdp: the filesystem
975  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
976  *
977  * Returns: errno
978  */
979
980 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
981 {
982         const struct lm_lockops *lm;
983         struct lm_lockstruct *ls = &sdp->sd_lockstruct;
984         struct gfs2_args *args = &sdp->sd_args;
985         const char *proto = sdp->sd_proto_name;
986         const char *table = sdp->sd_table_name;
987         char *o, *options;
988         int ret;
989
990         if (!strcmp("lock_nolock", proto)) {
991                 lm = &nolock_ops;
992                 sdp->sd_args.ar_localflocks = 1;
993 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
994         } else if (!strcmp("lock_dlm", proto)) {
995                 lm = &gfs2_dlm_ops;
996 #endif
997         } else {
998                 pr_info("can't find protocol %s\n", proto);
999                 return -ENOENT;
1000         }
1001
1002         fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
1003
1004         ls->ls_ops = lm;
1005         ls->ls_first = 1;
1006
1007         for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
1008                 substring_t tmp[MAX_OPT_ARGS];
1009                 int token, option;
1010
1011                 if (!o || !*o)
1012                         continue;
1013
1014                 token = match_token(o, *lm->lm_tokens, tmp);
1015                 switch (token) {
1016                 case Opt_jid:
1017                         ret = match_int(&tmp[0], &option);
1018                         if (ret || option < 0) 
1019                                 goto hostdata_error;
1020                         if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1021                                 ls->ls_jid = option;
1022                         break;
1023                 case Opt_id:
1024                 case Opt_nodir:
1025                         /* Obsolete, but left for backward compat purposes */
1026                         break;
1027                 case Opt_first:
1028                         ret = match_int(&tmp[0], &option);
1029                         if (ret || (option != 0 && option != 1))
1030                                 goto hostdata_error;
1031                         ls->ls_first = option;
1032                         break;
1033                 case Opt_err:
1034                 default:
1035 hostdata_error:
1036                         fs_info(sdp, "unknown hostdata (%s)\n", o);
1037                         return -EINVAL;
1038                 }
1039         }
1040
1041         if (lm->lm_mount == NULL) {
1042                 fs_info(sdp, "Now mounting FS (format %u)...\n", sdp->sd_sb.sb_fs_format);
1043                 complete_all(&sdp->sd_locking_init);
1044                 return 0;
1045         }
1046         ret = lm->lm_mount(sdp, table);
1047         if (ret == 0)
1048                 fs_info(sdp, "Joined cluster. Now mounting FS (format %u)...\n",
1049                         sdp->sd_sb.sb_fs_format);
1050         complete_all(&sdp->sd_locking_init);
1051         return ret;
1052 }
1053
1054 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1055 {
1056         const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1057         if (!gfs2_withdrawing_or_withdrawn(sdp) && lm->lm_unmount)
1058                 lm->lm_unmount(sdp);
1059 }
1060
1061 static int wait_on_journal(struct gfs2_sbd *sdp)
1062 {
1063         if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1064                 return 0;
1065
1066         return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1067                 ? -EINTR : 0;
1068 }
1069
1070 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1071 {
1072         struct super_block *sb = sdp->sd_vfs;
1073         char ro[20];
1074         char spectator[20];
1075         char *envp[] = { ro, spectator, NULL };
1076         sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1077         sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1078         kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1079 }
1080
1081 static int init_threads(struct gfs2_sbd *sdp)
1082 {
1083         struct task_struct *p;
1084         int error = 0;
1085
1086         p = kthread_create(gfs2_logd, sdp, "gfs2_logd/%s", sdp->sd_fsname);
1087         if (IS_ERR(p)) {
1088                 error = PTR_ERR(p);
1089                 fs_err(sdp, "can't create logd thread: %d\n", error);
1090                 return error;
1091         }
1092         get_task_struct(p);
1093         sdp->sd_logd_process = p;
1094
1095         p = kthread_create(gfs2_quotad, sdp, "gfs2_quotad/%s", sdp->sd_fsname);
1096         if (IS_ERR(p)) {
1097                 error = PTR_ERR(p);
1098                 fs_err(sdp, "can't create quotad thread: %d\n", error);
1099                 goto fail;
1100         }
1101         get_task_struct(p);
1102         sdp->sd_quotad_process = p;
1103
1104         wake_up_process(sdp->sd_logd_process);
1105         wake_up_process(sdp->sd_quotad_process);
1106         return 0;
1107
1108 fail:
1109         kthread_stop_put(sdp->sd_logd_process);
1110         sdp->sd_logd_process = NULL;
1111         return error;
1112 }
1113
1114 void gfs2_destroy_threads(struct gfs2_sbd *sdp)
1115 {
1116         if (sdp->sd_logd_process) {
1117                 kthread_stop_put(sdp->sd_logd_process);
1118                 sdp->sd_logd_process = NULL;
1119         }
1120         if (sdp->sd_quotad_process) {
1121                 kthread_stop_put(sdp->sd_quotad_process);
1122                 sdp->sd_quotad_process = NULL;
1123         }
1124 }
1125
1126 /**
1127  * gfs2_fill_super - Read in superblock
1128  * @sb: The VFS superblock
1129  * @fc: Mount options and flags
1130  *
1131  * Returns: -errno
1132  */
1133 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1134 {
1135         struct gfs2_args *args = fc->fs_private;
1136         int silent = fc->sb_flags & SB_SILENT;
1137         struct gfs2_sbd *sdp;
1138         struct gfs2_holder mount_gh;
1139         int error;
1140
1141         sdp = init_sbd(sb);
1142         if (!sdp) {
1143                 pr_warn("can't alloc struct gfs2_sbd\n");
1144                 return -ENOMEM;
1145         }
1146         sdp->sd_args = *args;
1147
1148         if (sdp->sd_args.ar_spectator) {
1149                 sb->s_flags |= SB_RDONLY;
1150                 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1151         }
1152         if (sdp->sd_args.ar_posix_acl)
1153                 sb->s_flags |= SB_POSIXACL;
1154         if (sdp->sd_args.ar_nobarrier)
1155                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1156
1157         sb->s_flags |= SB_NOSEC;
1158         sb->s_magic = GFS2_MAGIC;
1159         sb->s_op = &gfs2_super_ops;
1160         sb->s_d_op = &gfs2_dops;
1161         sb->s_export_op = &gfs2_export_ops;
1162         sb->s_qcop = &gfs2_quotactl_ops;
1163         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1164         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1165         sb->s_time_gran = 1;
1166         sb->s_maxbytes = MAX_LFS_FILESIZE;
1167
1168         /* Set up the buffer cache and fill in some fake block size values
1169            to allow us to read-in the on-disk superblock. */
1170         sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, 512);
1171         sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1172         sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift - 9;
1173         sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1174
1175         sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1176         sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1177         if (sdp->sd_args.ar_statfs_quantum) {
1178                 sdp->sd_tune.gt_statfs_slow = 0;
1179                 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1180         } else {
1181                 sdp->sd_tune.gt_statfs_slow = 1;
1182                 sdp->sd_tune.gt_statfs_quantum = 30;
1183         }
1184
1185         error = init_names(sdp, silent);
1186         if (error)
1187                 goto fail_free;
1188
1189         snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1190
1191         sdp->sd_delete_wq = alloc_workqueue("gfs2-delete/%s",
1192                         WQ_MEM_RECLAIM | WQ_FREEZABLE, 0, sdp->sd_fsname);
1193         error = -ENOMEM;
1194         if (!sdp->sd_delete_wq)
1195                 goto fail_free;
1196
1197         error = gfs2_sys_fs_add(sdp);
1198         if (error)
1199                 goto fail_delete_wq;
1200
1201         gfs2_create_debugfs_file(sdp);
1202
1203         error = gfs2_lm_mount(sdp, silent);
1204         if (error)
1205                 goto fail_debug;
1206
1207         error = init_locking(sdp, &mount_gh, DO);
1208         if (error)
1209                 goto fail_lm;
1210
1211         error = init_sb(sdp, silent);
1212         if (error)
1213                 goto fail_locking;
1214
1215         /* Turn rgrplvb on by default if fs format is recent enough */
1216         if (!sdp->sd_args.ar_got_rgrplvb && sdp->sd_sb.sb_fs_format > 1801)
1217                 sdp->sd_args.ar_rgrplvb = 1;
1218
1219         error = wait_on_journal(sdp);
1220         if (error)
1221                 goto fail_sb;
1222
1223         /*
1224          * If user space has failed to join the cluster or some similar
1225          * failure has occurred, then the journal id will contain a
1226          * negative (error) number. This will then be returned to the
1227          * caller (of the mount syscall). We do this even for spectator
1228          * mounts (which just write a jid of 0 to indicate "ok" even though
1229          * the jid is unused in the spectator case)
1230          */
1231         if (sdp->sd_lockstruct.ls_jid < 0) {
1232                 error = sdp->sd_lockstruct.ls_jid;
1233                 sdp->sd_lockstruct.ls_jid = 0;
1234                 goto fail_sb;
1235         }
1236
1237         if (sdp->sd_args.ar_spectator)
1238                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1239                          sdp->sd_table_name);
1240         else
1241                 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1242                          sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1243
1244         error = init_inodes(sdp, DO);
1245         if (error)
1246                 goto fail_sb;
1247
1248         error = init_per_node(sdp, DO);
1249         if (error)
1250                 goto fail_inodes;
1251
1252         error = gfs2_statfs_init(sdp);
1253         if (error) {
1254                 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1255                 goto fail_per_node;
1256         }
1257
1258         if (!sb_rdonly(sb)) {
1259                 error = init_threads(sdp);
1260                 if (error)
1261                         goto fail_per_node;
1262         }
1263
1264         error = gfs2_freeze_lock_shared(sdp);
1265         if (error)
1266                 goto fail_per_node;
1267
1268         if (!sb_rdonly(sb))
1269                 error = gfs2_make_fs_rw(sdp);
1270
1271         if (error) {
1272                 gfs2_freeze_unlock(sdp);
1273                 gfs2_destroy_threads(sdp);
1274                 fs_err(sdp, "can't make FS RW: %d\n", error);
1275                 goto fail_per_node;
1276         }
1277         gfs2_glock_dq_uninit(&mount_gh);
1278         gfs2_online_uevent(sdp);
1279         return 0;
1280
1281 fail_per_node:
1282         init_per_node(sdp, UNDO);
1283 fail_inodes:
1284         init_inodes(sdp, UNDO);
1285 fail_sb:
1286         if (sdp->sd_root_dir)
1287                 dput(sdp->sd_root_dir);
1288         if (sdp->sd_master_dir)
1289                 dput(sdp->sd_master_dir);
1290         if (sb->s_root)
1291                 dput(sb->s_root);
1292         sb->s_root = NULL;
1293 fail_locking:
1294         init_locking(sdp, &mount_gh, UNDO);
1295 fail_lm:
1296         complete_all(&sdp->sd_journal_ready);
1297         gfs2_gl_hash_clear(sdp);
1298         gfs2_lm_unmount(sdp);
1299 fail_debug:
1300         gfs2_delete_debugfs_file(sdp);
1301         gfs2_sys_fs_del(sdp);
1302 fail_delete_wq:
1303         destroy_workqueue(sdp->sd_delete_wq);
1304 fail_free:
1305         free_sbd(sdp);
1306         sb->s_fs_info = NULL;
1307         return error;
1308 }
1309
1310 /**
1311  * gfs2_get_tree - Get the GFS2 superblock and root directory
1312  * @fc: The filesystem context
1313  *
1314  * Returns: 0 or -errno on error
1315  */
1316 static int gfs2_get_tree(struct fs_context *fc)
1317 {
1318         struct gfs2_args *args = fc->fs_private;
1319         struct gfs2_sbd *sdp;
1320         int error;
1321
1322         error = get_tree_bdev(fc, gfs2_fill_super);
1323         if (error)
1324                 return error;
1325
1326         sdp = fc->root->d_sb->s_fs_info;
1327         dput(fc->root);
1328         if (args->ar_meta)
1329                 fc->root = dget(sdp->sd_master_dir);
1330         else
1331                 fc->root = dget(sdp->sd_root_dir);
1332         return 0;
1333 }
1334
1335 static void gfs2_fc_free(struct fs_context *fc)
1336 {
1337         struct gfs2_args *args = fc->fs_private;
1338
1339         kfree(args);
1340 }
1341
1342 enum gfs2_param {
1343         Opt_lockproto,
1344         Opt_locktable,
1345         Opt_hostdata,
1346         Opt_spectator,
1347         Opt_ignore_local_fs,
1348         Opt_localflocks,
1349         Opt_localcaching,
1350         Opt_debug,
1351         Opt_upgrade,
1352         Opt_acl,
1353         Opt_quota,
1354         Opt_quota_flag,
1355         Opt_suiddir,
1356         Opt_data,
1357         Opt_meta,
1358         Opt_discard,
1359         Opt_commit,
1360         Opt_errors,
1361         Opt_statfs_quantum,
1362         Opt_statfs_percent,
1363         Opt_quota_quantum,
1364         Opt_barrier,
1365         Opt_rgrplvb,
1366         Opt_loccookie,
1367 };
1368
1369 static const struct constant_table gfs2_param_quota[] = {
1370         {"off",        GFS2_QUOTA_OFF},
1371         {"account",    GFS2_QUOTA_ACCOUNT},
1372         {"on",         GFS2_QUOTA_ON},
1373         {"quiet",      GFS2_QUOTA_QUIET},
1374         {}
1375 };
1376
1377 enum opt_data {
1378         Opt_data_writeback = GFS2_DATA_WRITEBACK,
1379         Opt_data_ordered   = GFS2_DATA_ORDERED,
1380 };
1381
1382 static const struct constant_table gfs2_param_data[] = {
1383         {"writeback",  Opt_data_writeback },
1384         {"ordered",    Opt_data_ordered },
1385         {}
1386 };
1387
1388 enum opt_errors {
1389         Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1390         Opt_errors_panic    = GFS2_ERRORS_PANIC,
1391 };
1392
1393 static const struct constant_table gfs2_param_errors[] = {
1394         {"withdraw",   Opt_errors_withdraw },
1395         {"panic",      Opt_errors_panic },
1396         {}
1397 };
1398
1399 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1400         fsparam_string ("lockproto",          Opt_lockproto),
1401         fsparam_string ("locktable",          Opt_locktable),
1402         fsparam_string ("hostdata",           Opt_hostdata),
1403         fsparam_flag   ("spectator",          Opt_spectator),
1404         fsparam_flag   ("norecovery",         Opt_spectator),
1405         fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1406         fsparam_flag   ("localflocks",        Opt_localflocks),
1407         fsparam_flag   ("localcaching",       Opt_localcaching),
1408         fsparam_flag_no("debug",              Opt_debug),
1409         fsparam_flag   ("upgrade",            Opt_upgrade),
1410         fsparam_flag_no("acl",                Opt_acl),
1411         fsparam_flag_no("suiddir",            Opt_suiddir),
1412         fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1413         fsparam_flag   ("meta",               Opt_meta),
1414         fsparam_flag_no("discard",            Opt_discard),
1415         fsparam_s32    ("commit",             Opt_commit),
1416         fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1417         fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1418         fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1419         fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1420         fsparam_flag_no("barrier",            Opt_barrier),
1421         fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1422         fsparam_flag_no("loccookie",          Opt_loccookie),
1423         /* quota can be a flag or an enum so it gets special treatment */
1424         fsparam_flag_no("quota",              Opt_quota_flag),
1425         fsparam_enum("quota",                 Opt_quota, gfs2_param_quota),
1426         {}
1427 };
1428
1429 /* Parse a single mount parameter */
1430 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1431 {
1432         struct gfs2_args *args = fc->fs_private;
1433         struct fs_parse_result result;
1434         int o;
1435
1436         o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1437         if (o < 0)
1438                 return o;
1439
1440         switch (o) {
1441         case Opt_lockproto:
1442                 strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1443                 break;
1444         case Opt_locktable:
1445                 strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1446                 break;
1447         case Opt_hostdata:
1448                 strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1449                 break;
1450         case Opt_spectator:
1451                 args->ar_spectator = 1;
1452                 break;
1453         case Opt_ignore_local_fs:
1454                 /* Retained for backwards compat only */
1455                 break;
1456         case Opt_localflocks:
1457                 args->ar_localflocks = 1;
1458                 break;
1459         case Opt_localcaching:
1460                 /* Retained for backwards compat only */
1461                 break;
1462         case Opt_debug:
1463                 if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1464                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1465                 args->ar_debug = result.boolean;
1466                 break;
1467         case Opt_upgrade:
1468                 /* Retained for backwards compat only */
1469                 break;
1470         case Opt_acl:
1471                 args->ar_posix_acl = result.boolean;
1472                 break;
1473         case Opt_quota_flag:
1474                 args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1475                 break;
1476         case Opt_quota:
1477                 args->ar_quota = result.int_32;
1478                 break;
1479         case Opt_suiddir:
1480                 args->ar_suiddir = result.boolean;
1481                 break;
1482         case Opt_data:
1483                 /* The uint_32 result maps directly to GFS2_DATA_* */
1484                 args->ar_data = result.uint_32;
1485                 break;
1486         case Opt_meta:
1487                 args->ar_meta = 1;
1488                 break;
1489         case Opt_discard:
1490                 args->ar_discard = result.boolean;
1491                 break;
1492         case Opt_commit:
1493                 if (result.int_32 <= 0)
1494                         return invalfc(fc, "commit mount option requires a positive numeric argument");
1495                 args->ar_commit = result.int_32;
1496                 break;
1497         case Opt_statfs_quantum:
1498                 if (result.int_32 < 0)
1499                         return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1500                 args->ar_statfs_quantum = result.int_32;
1501                 break;
1502         case Opt_quota_quantum:
1503                 if (result.int_32 <= 0)
1504                         return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1505                 args->ar_quota_quantum = result.int_32;
1506                 break;
1507         case Opt_statfs_percent:
1508                 if (result.int_32 < 0 || result.int_32 > 100)
1509                         return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1510                 args->ar_statfs_percent = result.int_32;
1511                 break;
1512         case Opt_errors:
1513                 if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1514                         return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1515                 args->ar_errors = result.uint_32;
1516                 break;
1517         case Opt_barrier:
1518                 args->ar_nobarrier = result.boolean;
1519                 break;
1520         case Opt_rgrplvb:
1521                 args->ar_rgrplvb = result.boolean;
1522                 args->ar_got_rgrplvb = 1;
1523                 break;
1524         case Opt_loccookie:
1525                 args->ar_loccookie = result.boolean;
1526                 break;
1527         default:
1528                 return invalfc(fc, "invalid mount option: %s", param->key);
1529         }
1530         return 0;
1531 }
1532
1533 static int gfs2_reconfigure(struct fs_context *fc)
1534 {
1535         struct super_block *sb = fc->root->d_sb;
1536         struct gfs2_sbd *sdp = sb->s_fs_info;
1537         struct gfs2_args *oldargs = &sdp->sd_args;
1538         struct gfs2_args *newargs = fc->fs_private;
1539         struct gfs2_tune *gt = &sdp->sd_tune;
1540         int error = 0;
1541
1542         sync_filesystem(sb);
1543
1544         spin_lock(&gt->gt_spin);
1545         oldargs->ar_commit = gt->gt_logd_secs;
1546         oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1547         if (gt->gt_statfs_slow)
1548                 oldargs->ar_statfs_quantum = 0;
1549         else
1550                 oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1551         spin_unlock(&gt->gt_spin);
1552
1553         if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1554                 errorfc(fc, "reconfiguration of locking protocol not allowed");
1555                 return -EINVAL;
1556         }
1557         if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1558                 errorfc(fc, "reconfiguration of lock table not allowed");
1559                 return -EINVAL;
1560         }
1561         if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1562                 errorfc(fc, "reconfiguration of host data not allowed");
1563                 return -EINVAL;
1564         }
1565         if (newargs->ar_spectator != oldargs->ar_spectator) {
1566                 errorfc(fc, "reconfiguration of spectator mode not allowed");
1567                 return -EINVAL;
1568         }
1569         if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1570                 errorfc(fc, "reconfiguration of localflocks not allowed");
1571                 return -EINVAL;
1572         }
1573         if (newargs->ar_meta != oldargs->ar_meta) {
1574                 errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1575                 return -EINVAL;
1576         }
1577         if (oldargs->ar_spectator)
1578                 fc->sb_flags |= SB_RDONLY;
1579
1580         if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1581                 if (fc->sb_flags & SB_RDONLY) {
1582                         gfs2_make_fs_ro(sdp);
1583                 } else {
1584                         error = gfs2_make_fs_rw(sdp);
1585                         if (error)
1586                                 errorfc(fc, "unable to remount read-write");
1587                 }
1588         }
1589         sdp->sd_args = *newargs;
1590
1591         if (sdp->sd_args.ar_posix_acl)
1592                 sb->s_flags |= SB_POSIXACL;
1593         else
1594                 sb->s_flags &= ~SB_POSIXACL;
1595         if (sdp->sd_args.ar_nobarrier)
1596                 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1597         else
1598                 clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1599         spin_lock(&gt->gt_spin);
1600         gt->gt_logd_secs = newargs->ar_commit;
1601         gt->gt_quota_quantum = newargs->ar_quota_quantum;
1602         if (newargs->ar_statfs_quantum) {
1603                 gt->gt_statfs_slow = 0;
1604                 gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1605         }
1606         else {
1607                 gt->gt_statfs_slow = 1;
1608                 gt->gt_statfs_quantum = 30;
1609         }
1610         spin_unlock(&gt->gt_spin);
1611
1612         gfs2_online_uevent(sdp);
1613         return error;
1614 }
1615
1616 static const struct fs_context_operations gfs2_context_ops = {
1617         .free        = gfs2_fc_free,
1618         .parse_param = gfs2_parse_param,
1619         .get_tree    = gfs2_get_tree,
1620         .reconfigure = gfs2_reconfigure,
1621 };
1622
1623 /* Set up the filesystem mount context */
1624 static int gfs2_init_fs_context(struct fs_context *fc)
1625 {
1626         struct gfs2_args *args;
1627
1628         args = kmalloc(sizeof(*args), GFP_KERNEL);
1629         if (args == NULL)
1630                 return -ENOMEM;
1631
1632         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1633                 struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1634
1635                 *args = sdp->sd_args;
1636         } else {
1637                 memset(args, 0, sizeof(*args));
1638                 args->ar_quota = GFS2_QUOTA_DEFAULT;
1639                 args->ar_data = GFS2_DATA_DEFAULT;
1640                 args->ar_commit = 30;
1641                 args->ar_statfs_quantum = 30;
1642                 args->ar_quota_quantum = 60;
1643                 args->ar_errors = GFS2_ERRORS_DEFAULT;
1644         }
1645         fc->fs_private = args;
1646         fc->ops = &gfs2_context_ops;
1647         return 0;
1648 }
1649
1650 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1651 {
1652         return -EINVAL;
1653 }
1654
1655 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1656 {
1657         return (fc->sget_key == s->s_bdev);
1658 }
1659
1660 static int gfs2_meta_get_tree(struct fs_context *fc)
1661 {
1662         struct super_block *s;
1663         struct gfs2_sbd *sdp;
1664         struct path path;
1665         int error;
1666
1667         if (!fc->source || !*fc->source)
1668                 return -EINVAL;
1669
1670         error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1671         if (error) {
1672                 pr_warn("path_lookup on %s returned error %d\n",
1673                         fc->source, error);
1674                 return error;
1675         }
1676         fc->fs_type = &gfs2_fs_type;
1677         fc->sget_key = path.dentry->d_sb->s_bdev;
1678         s = sget_fc(fc, test_meta_super, set_meta_super);
1679         path_put(&path);
1680         if (IS_ERR(s)) {
1681                 pr_warn("gfs2 mount does not exist\n");
1682                 return PTR_ERR(s);
1683         }
1684         if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1685                 deactivate_locked_super(s);
1686                 return -EBUSY;
1687         }
1688         sdp = s->s_fs_info;
1689         fc->root = dget(sdp->sd_master_dir);
1690         return 0;
1691 }
1692
1693 static const struct fs_context_operations gfs2_meta_context_ops = {
1694         .free        = gfs2_fc_free,
1695         .get_tree    = gfs2_meta_get_tree,
1696 };
1697
1698 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1699 {
1700         int ret = gfs2_init_fs_context(fc);
1701
1702         if (ret)
1703                 return ret;
1704
1705         fc->ops = &gfs2_meta_context_ops;
1706         return 0;
1707 }
1708
1709 /**
1710  * gfs2_evict_inodes - evict inodes cooperatively
1711  * @sb: the superblock
1712  *
1713  * When evicting an inode with a zero link count, we are trying to upgrade the
1714  * inode's iopen glock from SH to EX mode in order to determine if we can
1715  * delete the inode.  The other nodes are supposed to evict the inode from
1716  * their caches if they can, and to poke the inode's inode glock if they cannot
1717  * do so.  Either behavior allows gfs2_upgrade_iopen_glock() to proceed
1718  * quickly, but if the other nodes are not cooperating, the lock upgrading
1719  * attempt will time out.  Since inodes are evicted sequentially, this can add
1720  * up quickly.
1721  *
1722  * Function evict_inodes() tries to keep the s_inode_list_lock list locked over
1723  * a long time, which prevents other inodes from being evicted concurrently.
1724  * This precludes the cooperative behavior we are looking for.  This special
1725  * version of evict_inodes() avoids that.
1726  *
1727  * Modeled after drop_pagecache_sb().
1728  */
1729 static void gfs2_evict_inodes(struct super_block *sb)
1730 {
1731         struct inode *inode, *toput_inode = NULL;
1732         struct gfs2_sbd *sdp = sb->s_fs_info;
1733
1734         set_bit(SDF_EVICTING, &sdp->sd_flags);
1735
1736         spin_lock(&sb->s_inode_list_lock);
1737         list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1738                 spin_lock(&inode->i_lock);
1739                 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) &&
1740                     !need_resched()) {
1741                         spin_unlock(&inode->i_lock);
1742                         continue;
1743                 }
1744                 atomic_inc(&inode->i_count);
1745                 spin_unlock(&inode->i_lock);
1746                 spin_unlock(&sb->s_inode_list_lock);
1747
1748                 iput(toput_inode);
1749                 toput_inode = inode;
1750
1751                 cond_resched();
1752                 spin_lock(&sb->s_inode_list_lock);
1753         }
1754         spin_unlock(&sb->s_inode_list_lock);
1755         iput(toput_inode);
1756 }
1757
1758 static void gfs2_kill_sb(struct super_block *sb)
1759 {
1760         struct gfs2_sbd *sdp = sb->s_fs_info;
1761
1762         if (sdp == NULL) {
1763                 kill_block_super(sb);
1764                 return;
1765         }
1766
1767         gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1768         dput(sdp->sd_root_dir);
1769         dput(sdp->sd_master_dir);
1770         sdp->sd_root_dir = NULL;
1771         sdp->sd_master_dir = NULL;
1772         shrink_dcache_sb(sb);
1773
1774         gfs2_evict_inodes(sb);
1775
1776         /*
1777          * Flush and then drain the delete workqueue here (via
1778          * destroy_workqueue()) to ensure that any delete work that
1779          * may be running will also see the SDF_KILL flag.
1780          */
1781         set_bit(SDF_KILL, &sdp->sd_flags);
1782         gfs2_flush_delete_work(sdp);
1783         destroy_workqueue(sdp->sd_delete_wq);
1784
1785         kill_block_super(sb);
1786 }
1787
1788 struct file_system_type gfs2_fs_type = {
1789         .name = "gfs2",
1790         .fs_flags = FS_REQUIRES_DEV,
1791         .init_fs_context = gfs2_init_fs_context,
1792         .parameters = gfs2_fs_parameters,
1793         .kill_sb = gfs2_kill_sb,
1794         .owner = THIS_MODULE,
1795 };
1796 MODULE_ALIAS_FS("gfs2");
1797
1798 struct file_system_type gfs2meta_fs_type = {
1799         .name = "gfs2meta",
1800         .fs_flags = FS_REQUIRES_DEV,
1801         .init_fs_context = gfs2_meta_init_fs_context,
1802         .owner = THIS_MODULE,
1803 };
1804 MODULE_ALIAS_FS("gfs2meta");
This page took 0.133748 seconds and 4 git commands to generate.