xref: /openbmc/linux/fs/gfs2/ops_fstype.c (revision ba61bb17)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/blkdev.h>
18 #include <linux/kthread.h>
19 #include <linux/export.h>
20 #include <linux/namei.h>
21 #include <linux/mount.h>
22 #include <linux/gfs2_ondisk.h>
23 #include <linux/quotaops.h>
24 #include <linux/lockdep.h>
25 #include <linux/module.h>
26 #include <linux/backing-dev.h>
27 
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "glock.h"
32 #include "glops.h"
33 #include "inode.h"
34 #include "recovery.h"
35 #include "rgrp.h"
36 #include "super.h"
37 #include "sys.h"
38 #include "util.h"
39 #include "log.h"
40 #include "quota.h"
41 #include "dir.h"
42 #include "meta_io.h"
43 #include "trace_gfs2.h"
44 
45 #define DO 0
46 #define UNDO 1
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_quota_warn_period = 10;
59 	gt->gt_quota_scale_num = 1;
60 	gt->gt_quota_scale_den = 1;
61 	gt->gt_new_files_jdata = 0;
62 	gt->gt_max_readahead = BIT(18);
63 	gt->gt_complain_secs = 10;
64 }
65 
66 static struct gfs2_sbd *init_sbd(struct super_block *sb)
67 {
68 	struct gfs2_sbd *sdp;
69 	struct address_space *mapping;
70 
71 	sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
72 	if (!sdp)
73 		return NULL;
74 
75 	sb->s_fs_info = sdp;
76 	sdp->sd_vfs = sb;
77 	sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
78 	if (!sdp->sd_lkstats) {
79 		kfree(sdp);
80 		return NULL;
81 	}
82 
83 	set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
84 	gfs2_tune_init(&sdp->sd_tune);
85 
86 	init_waitqueue_head(&sdp->sd_glock_wait);
87 	atomic_set(&sdp->sd_glock_disposal, 0);
88 	init_completion(&sdp->sd_locking_init);
89 	init_completion(&sdp->sd_wdack);
90 	spin_lock_init(&sdp->sd_statfs_spin);
91 
92 	spin_lock_init(&sdp->sd_rindex_spin);
93 	sdp->sd_rindex_tree.rb_node = NULL;
94 
95 	INIT_LIST_HEAD(&sdp->sd_jindex_list);
96 	spin_lock_init(&sdp->sd_jindex_spin);
97 	mutex_init(&sdp->sd_jindex_mutex);
98 	init_completion(&sdp->sd_journal_ready);
99 
100 	INIT_LIST_HEAD(&sdp->sd_quota_list);
101 	mutex_init(&sdp->sd_quota_mutex);
102 	mutex_init(&sdp->sd_quota_sync_mutex);
103 	init_waitqueue_head(&sdp->sd_quota_wait);
104 	INIT_LIST_HEAD(&sdp->sd_trunc_list);
105 	spin_lock_init(&sdp->sd_trunc_lock);
106 	spin_lock_init(&sdp->sd_bitmap_lock);
107 
108 	mapping = &sdp->sd_aspace;
109 
110 	address_space_init_once(mapping);
111 	mapping->a_ops = &gfs2_rgrp_aops;
112 	mapping->host = sb->s_bdev->bd_inode;
113 	mapping->flags = 0;
114 	mapping_set_gfp_mask(mapping, GFP_NOFS);
115 	mapping->private_data = NULL;
116 	mapping->writeback_index = 0;
117 
118 	spin_lock_init(&sdp->sd_log_lock);
119 	atomic_set(&sdp->sd_log_pinned, 0);
120 	INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
121 	INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
122 	spin_lock_init(&sdp->sd_ordered_lock);
123 
124 	init_waitqueue_head(&sdp->sd_log_waitq);
125 	init_waitqueue_head(&sdp->sd_logd_waitq);
126 	spin_lock_init(&sdp->sd_ail_lock);
127 	INIT_LIST_HEAD(&sdp->sd_ail1_list);
128 	INIT_LIST_HEAD(&sdp->sd_ail2_list);
129 
130 	init_rwsem(&sdp->sd_log_flush_lock);
131 	atomic_set(&sdp->sd_log_in_flight, 0);
132 	atomic_set(&sdp->sd_reserving_log, 0);
133 	init_waitqueue_head(&sdp->sd_reserving_log_wait);
134 	init_waitqueue_head(&sdp->sd_log_flush_wait);
135 	atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
136 	mutex_init(&sdp->sd_freeze_mutex);
137 
138 	return sdp;
139 }
140 
141 
142 /**
143  * gfs2_check_sb - Check superblock
144  * @sdp: the filesystem
145  * @sb: The superblock
146  * @silent: Don't print a message if the check fails
147  *
148  * Checks the version code of the FS is one that we understand how to
149  * read and that the sizes of the various on-disk structures have not
150  * changed.
151  */
152 
153 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
154 {
155 	struct gfs2_sb_host *sb = &sdp->sd_sb;
156 
157 	if (sb->sb_magic != GFS2_MAGIC ||
158 	    sb->sb_type != GFS2_METATYPE_SB) {
159 		if (!silent)
160 			pr_warn("not a GFS2 filesystem\n");
161 		return -EINVAL;
162 	}
163 
164 	/*  If format numbers match exactly, we're done.  */
165 
166 	if (sb->sb_fs_format == GFS2_FORMAT_FS &&
167 	    sb->sb_multihost_format == GFS2_FORMAT_MULTI)
168 		return 0;
169 
170 	fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
171 
172 	return -EINVAL;
173 }
174 
175 static void end_bio_io_page(struct bio *bio)
176 {
177 	struct page *page = bio->bi_private;
178 
179 	if (!bio->bi_status)
180 		SetPageUptodate(page);
181 	else
182 		pr_warn("error %d reading superblock\n", bio->bi_status);
183 	unlock_page(page);
184 }
185 
186 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
187 {
188 	struct gfs2_sb_host *sb = &sdp->sd_sb;
189 	struct super_block *s = sdp->sd_vfs;
190 	const struct gfs2_sb *str = buf;
191 
192 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
193 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
194 	sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
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 	memcpy(&s->s_uuid, 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  * @error: The error code to return
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 gfs2_sb *p;
232 	struct page *page;
233 	struct bio *bio;
234 
235 	page = alloc_page(GFP_NOFS);
236 	if (unlikely(!page))
237 		return -ENOMEM;
238 
239 	ClearPageUptodate(page);
240 	ClearPageDirty(page);
241 	lock_page(page);
242 
243 	bio = bio_alloc(GFP_NOFS, 1);
244 	bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
245 	bio_set_dev(bio, sb->s_bdev);
246 	bio_add_page(bio, page, PAGE_SIZE, 0);
247 
248 	bio->bi_end_io = end_bio_io_page;
249 	bio->bi_private = page;
250 	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
251 	submit_bio(bio);
252 	wait_on_page_locked(page);
253 	bio_put(bio);
254 	if (!PageUptodate(page)) {
255 		__free_page(page);
256 		return -EIO;
257 	}
258 	p = kmap(page);
259 	gfs2_sb_in(sdp, p);
260 	kunmap(page);
261 	__free_page(page);
262 	return gfs2_check_sb(sdp, silent);
263 }
264 
265 /**
266  * gfs2_read_sb - Read super block
267  * @sdp: The GFS2 superblock
268  * @silent: Don't print message if mount fails
269  *
270  */
271 
272 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
273 {
274 	u32 hash_blocks, ind_blocks, leaf_blocks;
275 	u32 tmp_blocks;
276 	unsigned int x;
277 	int error;
278 
279 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
280 	if (error) {
281 		if (!silent)
282 			fs_err(sdp, "can't read superblock\n");
283 		return error;
284 	}
285 
286 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
287 			       GFS2_BASIC_BLOCK_SHIFT;
288 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
289 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
290 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
291 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
292 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
293 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
294 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
295 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
296 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
297 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
298 				sizeof(struct gfs2_meta_header)) /
299 			        sizeof(struct gfs2_quota_change);
300 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
301 				     sizeof(struct gfs2_meta_header))
302 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
303 
304 	/* Compute maximum reservation required to add a entry to a directory */
305 
306 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
307 			     sdp->sd_jbsize);
308 
309 	ind_blocks = 0;
310 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
311 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
312 		ind_blocks += tmp_blocks;
313 	}
314 
315 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
316 
317 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
318 
319 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
320 				sizeof(struct gfs2_dinode);
321 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
322 	for (x = 2;; x++) {
323 		u64 space, d;
324 		u32 m;
325 
326 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
327 		d = space;
328 		m = do_div(d, sdp->sd_inptrs);
329 
330 		if (d != sdp->sd_heightsize[x - 1] || m)
331 			break;
332 		sdp->sd_heightsize[x] = space;
333 	}
334 	sdp->sd_max_height = x;
335 	sdp->sd_heightsize[x] = ~0;
336 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
337 
338 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
339 				      sizeof(struct gfs2_leaf)) /
340 				     GFS2_MIN_DIRENT_SIZE;
341 	return 0;
342 }
343 
344 static int init_names(struct gfs2_sbd *sdp, int silent)
345 {
346 	char *proto, *table;
347 	int error = 0;
348 
349 	proto = sdp->sd_args.ar_lockproto;
350 	table = sdp->sd_args.ar_locktable;
351 
352 	/*  Try to autodetect  */
353 
354 	if (!proto[0] || !table[0]) {
355 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
356 		if (error)
357 			return error;
358 
359 		if (!proto[0])
360 			proto = sdp->sd_sb.sb_lockproto;
361 		if (!table[0])
362 			table = sdp->sd_sb.sb_locktable;
363 	}
364 
365 	if (!table[0])
366 		table = sdp->sd_vfs->s_id;
367 
368 	strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
369 	strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_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, LM_FLAG_NOEXP | GL_NOCACHE,
389 				  mount_gh);
390 	if (error) {
391 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
392 		goto fail;
393 	}
394 
395 	error = gfs2_glock_nq_num(sdp,
396 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
397 				  LM_ST_SHARED,
398 				  LM_FLAG_NOEXP | GL_EXACT,
399 				  &sdp->sd_live_gh);
400 	if (error) {
401 		fs_err(sdp, "can't acquire live glock: %d\n", error);
402 		goto fail_mount;
403 	}
404 
405 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
406 			       CREATE, &sdp->sd_rename_gl);
407 	if (error) {
408 		fs_err(sdp, "can't create rename glock: %d\n", error);
409 		goto fail_live;
410 	}
411 
412 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
413 			       CREATE, &sdp->sd_freeze_gl);
414 	if (error) {
415 		fs_err(sdp, "can't create transaction glock: %d\n", error);
416 		goto fail_rename;
417 	}
418 
419 	return 0;
420 
421 fail_trans:
422 	gfs2_glock_put(sdp->sd_freeze_gl);
423 fail_rename:
424 	gfs2_glock_put(sdp->sd_rename_gl);
425 fail_live:
426 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
427 fail_mount:
428 	gfs2_glock_dq_uninit(mount_gh);
429 fail:
430 	return error;
431 }
432 
433 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
434 			    u64 no_addr, const char *name)
435 {
436 	struct gfs2_sbd *sdp = sb->s_fs_info;
437 	struct dentry *dentry;
438 	struct inode *inode;
439 
440 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
441 				  GFS2_BLKST_FREE /* ignore */);
442 	if (IS_ERR(inode)) {
443 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
444 		return PTR_ERR(inode);
445 	}
446 	dentry = d_make_root(inode);
447 	if (!dentry) {
448 		fs_err(sdp, "can't alloc %s dentry\n", name);
449 		return -ENOMEM;
450 	}
451 	*dptr = dentry;
452 	return 0;
453 }
454 
455 static int init_sb(struct gfs2_sbd *sdp, int silent)
456 {
457 	struct super_block *sb = sdp->sd_vfs;
458 	struct gfs2_holder sb_gh;
459 	u64 no_addr;
460 	int ret;
461 
462 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
463 				LM_ST_SHARED, 0, &sb_gh);
464 	if (ret) {
465 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
466 		return ret;
467 	}
468 
469 	ret = gfs2_read_sb(sdp, silent);
470 	if (ret) {
471 		fs_err(sdp, "can't read superblock: %d\n", ret);
472 		goto out;
473 	}
474 
475 	/* Set up the buffer cache and SB for real */
476 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
477 		ret = -EINVAL;
478 		fs_err(sdp, "FS block size (%u) is too small for device "
479 		       "block size (%u)\n",
480 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
481 		goto out;
482 	}
483 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
484 		ret = -EINVAL;
485 		fs_err(sdp, "FS block size (%u) is too big for machine "
486 		       "page size (%u)\n",
487 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
488 		goto out;
489 	}
490 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
491 
492 	/* Get the root inode */
493 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
494 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
495 	if (ret)
496 		goto out;
497 
498 	/* Get the master inode */
499 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
500 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
501 	if (ret) {
502 		dput(sdp->sd_root_dir);
503 		goto out;
504 	}
505 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
506 out:
507 	gfs2_glock_dq_uninit(&sb_gh);
508 	return ret;
509 }
510 
511 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
512 {
513 	char *message = "FIRSTMOUNT=Done";
514 	char *envp[] = { message, NULL };
515 
516 	fs_info(sdp, "first mount done, others may mount\n");
517 
518 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
519 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
520 
521 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
522 }
523 
524 /**
525  * gfs2_jindex_hold - Grab a lock on the jindex
526  * @sdp: The GFS2 superblock
527  * @ji_gh: the holder for the jindex glock
528  *
529  * Returns: errno
530  */
531 
532 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
533 {
534 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
535 	struct qstr name;
536 	char buf[20];
537 	struct gfs2_jdesc *jd;
538 	int error;
539 
540 	name.name = buf;
541 
542 	mutex_lock(&sdp->sd_jindex_mutex);
543 
544 	for (;;) {
545 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
546 		if (error)
547 			break;
548 
549 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
550 		name.hash = gfs2_disk_hash(name.name, name.len);
551 
552 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
553 		if (error == -ENOENT) {
554 			error = 0;
555 			break;
556 		}
557 
558 		gfs2_glock_dq_uninit(ji_gh);
559 
560 		if (error)
561 			break;
562 
563 		error = -ENOMEM;
564 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
565 		if (!jd)
566 			break;
567 
568 		INIT_LIST_HEAD(&jd->extent_list);
569 		INIT_LIST_HEAD(&jd->jd_revoke_list);
570 
571 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
572 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
573 		if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
574 			if (!jd->jd_inode)
575 				error = -ENOENT;
576 			else
577 				error = PTR_ERR(jd->jd_inode);
578 			kfree(jd);
579 			break;
580 		}
581 
582 		spin_lock(&sdp->sd_jindex_spin);
583 		jd->jd_jid = sdp->sd_journals++;
584 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
585 		spin_unlock(&sdp->sd_jindex_spin);
586 	}
587 
588 	mutex_unlock(&sdp->sd_jindex_mutex);
589 
590 	return error;
591 }
592 
593 /**
594  * check_journal_clean - Make sure a journal is clean for a spectator mount
595  * @sdp: The GFS2 superblock
596  * @jd: The journal descriptor
597  *
598  * Returns: 0 if the journal is clean or locked, else an error
599  */
600 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
601 {
602 	int error;
603 	struct gfs2_holder j_gh;
604 	struct gfs2_log_header_host head;
605 	struct gfs2_inode *ip;
606 
607 	ip = GFS2_I(jd->jd_inode);
608 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
609 				   GL_EXACT | GL_NOCACHE, &j_gh);
610 	if (error) {
611 		fs_err(sdp, "Error locking journal for spectator mount.\n");
612 		return -EPERM;
613 	}
614 	error = gfs2_jdesc_check(jd);
615 	if (error) {
616 		fs_err(sdp, "Error checking journal for spectator mount.\n");
617 		goto out_unlock;
618 	}
619 	error = gfs2_find_jhead(jd, &head);
620 	if (error) {
621 		fs_err(sdp, "Error parsing journal for spectator mount.\n");
622 		goto out_unlock;
623 	}
624 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
625 		error = -EPERM;
626 		fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
627 		       "must not be a spectator.\n", jd->jd_jid);
628 	}
629 
630 out_unlock:
631 	gfs2_glock_dq_uninit(&j_gh);
632 	return error;
633 }
634 
635 static int init_journal(struct gfs2_sbd *sdp, int undo)
636 {
637 	struct inode *master = d_inode(sdp->sd_master_dir);
638 	struct gfs2_holder ji_gh;
639 	struct gfs2_inode *ip;
640 	int jindex = 1;
641 	int error = 0;
642 
643 	if (undo) {
644 		jindex = 0;
645 		goto fail_jinode_gh;
646 	}
647 
648 	sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
649 	if (IS_ERR(sdp->sd_jindex)) {
650 		fs_err(sdp, "can't lookup journal index: %d\n", error);
651 		return PTR_ERR(sdp->sd_jindex);
652 	}
653 
654 	/* Load in the journal index special file */
655 
656 	error = gfs2_jindex_hold(sdp, &ji_gh);
657 	if (error) {
658 		fs_err(sdp, "can't read journal index: %d\n", error);
659 		goto fail;
660 	}
661 
662 	error = -EUSERS;
663 	if (!gfs2_jindex_size(sdp)) {
664 		fs_err(sdp, "no journals!\n");
665 		goto fail_jindex;
666 	}
667 
668 	atomic_set(&sdp->sd_log_blks_needed, 0);
669 	if (sdp->sd_args.ar_spectator) {
670 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
671 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
672 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
673 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
674 	} else {
675 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
676 			fs_err(sdp, "can't mount journal #%u\n",
677 			       sdp->sd_lockstruct.ls_jid);
678 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
679 			       gfs2_jindex_size(sdp),
680 			       gfs2_jindex_size(sdp) - 1);
681 			goto fail_jindex;
682 		}
683 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
684 
685 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
686 					  &gfs2_journal_glops,
687 					  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
688 					  &sdp->sd_journal_gh);
689 		if (error) {
690 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
691 			goto fail_jindex;
692 		}
693 
694 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
695 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
696 					   LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
697 					   &sdp->sd_jinode_gh);
698 		if (error) {
699 			fs_err(sdp, "can't acquire journal inode glock: %d\n",
700 			       error);
701 			goto fail_journal_gh;
702 		}
703 
704 		error = gfs2_jdesc_check(sdp->sd_jdesc);
705 		if (error) {
706 			fs_err(sdp, "my journal (%u) is bad: %d\n",
707 			       sdp->sd_jdesc->jd_jid, error);
708 			goto fail_jinode_gh;
709 		}
710 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
711 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
712 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
713 
714 		/* Map the extents for this journal's blocks */
715 		gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
716 	}
717 	trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
718 
719 	if (sdp->sd_lockstruct.ls_first) {
720 		unsigned int x;
721 		for (x = 0; x < sdp->sd_journals; x++) {
722 			struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
723 
724 			if (sdp->sd_args.ar_spectator) {
725 				error = check_journal_clean(sdp, jd);
726 				if (error)
727 					goto fail_jinode_gh;
728 				continue;
729 			}
730 			error = gfs2_recover_journal(jd, true);
731 			if (error) {
732 				fs_err(sdp, "error recovering journal %u: %d\n",
733 				       x, error);
734 				goto fail_jinode_gh;
735 			}
736 		}
737 
738 		gfs2_others_may_mount(sdp);
739 	} else if (!sdp->sd_args.ar_spectator) {
740 		error = gfs2_recover_journal(sdp->sd_jdesc, true);
741 		if (error) {
742 			fs_err(sdp, "error recovering my journal: %d\n", error);
743 			goto fail_jinode_gh;
744 		}
745 	}
746 
747 	sdp->sd_log_idle = 1;
748 	set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
749 	gfs2_glock_dq_uninit(&ji_gh);
750 	jindex = 0;
751 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
752 	return 0;
753 
754 fail_jinode_gh:
755 	if (!sdp->sd_args.ar_spectator)
756 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
757 fail_journal_gh:
758 	if (!sdp->sd_args.ar_spectator)
759 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
760 fail_jindex:
761 	gfs2_jindex_free(sdp);
762 	if (jindex)
763 		gfs2_glock_dq_uninit(&ji_gh);
764 fail:
765 	iput(sdp->sd_jindex);
766 	return error;
767 }
768 
769 static struct lock_class_key gfs2_quota_imutex_key;
770 
771 static int init_inodes(struct gfs2_sbd *sdp, int undo)
772 {
773 	int error = 0;
774 	struct inode *master = d_inode(sdp->sd_master_dir);
775 
776 	if (undo)
777 		goto fail_qinode;
778 
779 	error = init_journal(sdp, undo);
780 	complete_all(&sdp->sd_journal_ready);
781 	if (error)
782 		goto fail;
783 
784 	/* Read in the master statfs inode */
785 	sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
786 	if (IS_ERR(sdp->sd_statfs_inode)) {
787 		error = PTR_ERR(sdp->sd_statfs_inode);
788 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
789 		goto fail_journal;
790 	}
791 
792 	/* Read in the resource index inode */
793 	sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
794 	if (IS_ERR(sdp->sd_rindex)) {
795 		error = PTR_ERR(sdp->sd_rindex);
796 		fs_err(sdp, "can't get resource index inode: %d\n", error);
797 		goto fail_statfs;
798 	}
799 	sdp->sd_rindex_uptodate = 0;
800 
801 	/* Read in the quota inode */
802 	sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
803 	if (IS_ERR(sdp->sd_quota_inode)) {
804 		error = PTR_ERR(sdp->sd_quota_inode);
805 		fs_err(sdp, "can't get quota file inode: %d\n", error);
806 		goto fail_rindex;
807 	}
808 	/*
809 	 * i_rwsem on quota files is special. Since this inode is hidden system
810 	 * file, we are safe to define locking ourselves.
811 	 */
812 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
813 			  &gfs2_quota_imutex_key);
814 
815 	error = gfs2_rindex_update(sdp);
816 	if (error)
817 		goto fail_qinode;
818 
819 	return 0;
820 
821 fail_qinode:
822 	iput(sdp->sd_quota_inode);
823 fail_rindex:
824 	gfs2_clear_rgrpd(sdp);
825 	iput(sdp->sd_rindex);
826 fail_statfs:
827 	iput(sdp->sd_statfs_inode);
828 fail_journal:
829 	init_journal(sdp, UNDO);
830 fail:
831 	return error;
832 }
833 
834 static int init_per_node(struct gfs2_sbd *sdp, int undo)
835 {
836 	struct inode *pn = NULL;
837 	char buf[30];
838 	int error = 0;
839 	struct gfs2_inode *ip;
840 	struct inode *master = d_inode(sdp->sd_master_dir);
841 
842 	if (sdp->sd_args.ar_spectator)
843 		return 0;
844 
845 	if (undo)
846 		goto fail_qc_gh;
847 
848 	pn = gfs2_lookup_simple(master, "per_node");
849 	if (IS_ERR(pn)) {
850 		error = PTR_ERR(pn);
851 		fs_err(sdp, "can't find per_node directory: %d\n", error);
852 		return error;
853 	}
854 
855 	sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
856 	sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
857 	if (IS_ERR(sdp->sd_sc_inode)) {
858 		error = PTR_ERR(sdp->sd_sc_inode);
859 		fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
860 		goto fail;
861 	}
862 
863 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
864 	sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
865 	if (IS_ERR(sdp->sd_qc_inode)) {
866 		error = PTR_ERR(sdp->sd_qc_inode);
867 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
868 		goto fail_ut_i;
869 	}
870 
871 	iput(pn);
872 	pn = NULL;
873 
874 	ip = GFS2_I(sdp->sd_sc_inode);
875 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
876 				   &sdp->sd_sc_gh);
877 	if (error) {
878 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
879 		goto fail_qc_i;
880 	}
881 
882 	ip = GFS2_I(sdp->sd_qc_inode);
883 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
884 				   &sdp->sd_qc_gh);
885 	if (error) {
886 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
887 		goto fail_ut_gh;
888 	}
889 
890 	return 0;
891 
892 fail_qc_gh:
893 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
894 fail_ut_gh:
895 	gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
896 fail_qc_i:
897 	iput(sdp->sd_qc_inode);
898 fail_ut_i:
899 	iput(sdp->sd_sc_inode);
900 fail:
901 	iput(pn);
902 	return error;
903 }
904 
905 static const match_table_t nolock_tokens = {
906 	{ Opt_jid, "jid=%d\n", },
907 	{ Opt_err, NULL },
908 };
909 
910 static const struct lm_lockops nolock_ops = {
911 	.lm_proto_name = "lock_nolock",
912 	.lm_put_lock = gfs2_glock_free,
913 	.lm_tokens = &nolock_tokens,
914 };
915 
916 /**
917  * gfs2_lm_mount - mount a locking protocol
918  * @sdp: the filesystem
919  * @args: mount arguments
920  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
921  *
922  * Returns: errno
923  */
924 
925 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
926 {
927 	const struct lm_lockops *lm;
928 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
929 	struct gfs2_args *args = &sdp->sd_args;
930 	const char *proto = sdp->sd_proto_name;
931 	const char *table = sdp->sd_table_name;
932 	char *o, *options;
933 	int ret;
934 
935 	if (!strcmp("lock_nolock", proto)) {
936 		lm = &nolock_ops;
937 		sdp->sd_args.ar_localflocks = 1;
938 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
939 	} else if (!strcmp("lock_dlm", proto)) {
940 		lm = &gfs2_dlm_ops;
941 #endif
942 	} else {
943 		pr_info("can't find protocol %s\n", proto);
944 		return -ENOENT;
945 	}
946 
947 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
948 
949 	ls->ls_ops = lm;
950 	ls->ls_first = 1;
951 
952 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
953 		substring_t tmp[MAX_OPT_ARGS];
954 		int token, option;
955 
956 		if (!o || !*o)
957 			continue;
958 
959 		token = match_token(o, *lm->lm_tokens, tmp);
960 		switch (token) {
961 		case Opt_jid:
962 			ret = match_int(&tmp[0], &option);
963 			if (ret || option < 0)
964 				goto hostdata_error;
965 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
966 				ls->ls_jid = option;
967 			break;
968 		case Opt_id:
969 		case Opt_nodir:
970 			/* Obsolete, but left for backward compat purposes */
971 			break;
972 		case Opt_first:
973 			ret = match_int(&tmp[0], &option);
974 			if (ret || (option != 0 && option != 1))
975 				goto hostdata_error;
976 			ls->ls_first = option;
977 			break;
978 		case Opt_err:
979 		default:
980 hostdata_error:
981 			fs_info(sdp, "unknown hostdata (%s)\n", o);
982 			return -EINVAL;
983 		}
984 	}
985 
986 	if (lm->lm_mount == NULL) {
987 		fs_info(sdp, "Now mounting FS...\n");
988 		complete_all(&sdp->sd_locking_init);
989 		return 0;
990 	}
991 	ret = lm->lm_mount(sdp, table);
992 	if (ret == 0)
993 		fs_info(sdp, "Joined cluster. Now mounting FS...\n");
994 	complete_all(&sdp->sd_locking_init);
995 	return ret;
996 }
997 
998 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
999 {
1000 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1001 	if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1002 	    lm->lm_unmount)
1003 		lm->lm_unmount(sdp);
1004 }
1005 
1006 static int wait_on_journal(struct gfs2_sbd *sdp)
1007 {
1008 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1009 		return 0;
1010 
1011 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1012 		? -EINTR : 0;
1013 }
1014 
1015 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1016 {
1017 	struct super_block *sb = sdp->sd_vfs;
1018 	char ro[20];
1019 	char spectator[20];
1020 	char *envp[] = { ro, spectator, NULL };
1021 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1022 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1023 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1024 }
1025 
1026 /**
1027  * fill_super - Read in superblock
1028  * @sb: The VFS superblock
1029  * @data: Mount options
1030  * @silent: Don't complain if it's not a GFS2 filesystem
1031  *
1032  * Returns: errno
1033  */
1034 
1035 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1036 {
1037 	struct gfs2_sbd *sdp;
1038 	struct gfs2_holder mount_gh;
1039 	int error;
1040 
1041 	sdp = init_sbd(sb);
1042 	if (!sdp) {
1043 		pr_warn("can't alloc struct gfs2_sbd\n");
1044 		return -ENOMEM;
1045 	}
1046 	sdp->sd_args = *args;
1047 
1048 	if (sdp->sd_args.ar_spectator) {
1049                 sb->s_flags |= SB_RDONLY;
1050 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1051 	}
1052 	if (sdp->sd_args.ar_posix_acl)
1053 		sb->s_flags |= SB_POSIXACL;
1054 	if (sdp->sd_args.ar_nobarrier)
1055 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1056 
1057 	sb->s_flags |= SB_NOSEC;
1058 	sb->s_magic = GFS2_MAGIC;
1059 	sb->s_op = &gfs2_super_ops;
1060 	sb->s_d_op = &gfs2_dops;
1061 	sb->s_export_op = &gfs2_export_ops;
1062 	sb->s_xattr = gfs2_xattr_handlers;
1063 	sb->s_qcop = &gfs2_quotactl_ops;
1064 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1065 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1066 	sb->s_time_gran = 1;
1067 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1068 
1069 	/* Set up the buffer cache and fill in some fake block size values
1070 	   to allow us to read-in the on-disk superblock. */
1071 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1072 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1073 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1074                                GFS2_BASIC_BLOCK_SHIFT;
1075 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1076 
1077 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1078 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1079 	if (sdp->sd_args.ar_statfs_quantum) {
1080 		sdp->sd_tune.gt_statfs_slow = 0;
1081 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1082 	} else {
1083 		sdp->sd_tune.gt_statfs_slow = 1;
1084 		sdp->sd_tune.gt_statfs_quantum = 30;
1085 	}
1086 
1087 	error = init_names(sdp, silent);
1088 	if (error) {
1089 		/* In this case, we haven't initialized sysfs, so we have to
1090 		   manually free the sdp. */
1091 		free_percpu(sdp->sd_lkstats);
1092 		kfree(sdp);
1093 		sb->s_fs_info = NULL;
1094 		return error;
1095 	}
1096 
1097 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1098 
1099 	error = gfs2_sys_fs_add(sdp);
1100 	/*
1101 	 * If we hit an error here, gfs2_sys_fs_add will have called function
1102 	 * kobject_put which causes the sysfs usage count to go to zero, which
1103 	 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1104 	 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1105 	 * kobject_put to free sdp.
1106 	 */
1107 	if (error)
1108 		return error;
1109 
1110 	gfs2_create_debugfs_file(sdp);
1111 
1112 	error = gfs2_lm_mount(sdp, silent);
1113 	if (error)
1114 		goto fail_debug;
1115 
1116 	error = init_locking(sdp, &mount_gh, DO);
1117 	if (error)
1118 		goto fail_lm;
1119 
1120 	error = init_sb(sdp, silent);
1121 	if (error)
1122 		goto fail_locking;
1123 
1124 	error = wait_on_journal(sdp);
1125 	if (error)
1126 		goto fail_sb;
1127 
1128 	/*
1129 	 * If user space has failed to join the cluster or some similar
1130 	 * failure has occurred, then the journal id will contain a
1131 	 * negative (error) number. This will then be returned to the
1132 	 * caller (of the mount syscall). We do this even for spectator
1133 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1134 	 * the jid is unused in the spectator case)
1135 	 */
1136 	if (sdp->sd_lockstruct.ls_jid < 0) {
1137 		error = sdp->sd_lockstruct.ls_jid;
1138 		sdp->sd_lockstruct.ls_jid = 0;
1139 		goto fail_sb;
1140 	}
1141 
1142 	if (sdp->sd_args.ar_spectator)
1143 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1144 			 sdp->sd_table_name);
1145 	else
1146 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1147 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1148 
1149 	error = init_inodes(sdp, DO);
1150 	if (error)
1151 		goto fail_sb;
1152 
1153 	error = init_per_node(sdp, DO);
1154 	if (error)
1155 		goto fail_inodes;
1156 
1157 	error = gfs2_statfs_init(sdp);
1158 	if (error) {
1159 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1160 		goto fail_per_node;
1161 	}
1162 
1163 	if (!sb_rdonly(sb)) {
1164 		error = gfs2_make_fs_rw(sdp);
1165 		if (error) {
1166 			fs_err(sdp, "can't make FS RW: %d\n", error);
1167 			goto fail_per_node;
1168 		}
1169 	}
1170 
1171 	gfs2_glock_dq_uninit(&mount_gh);
1172 	gfs2_online_uevent(sdp);
1173 	return 0;
1174 
1175 fail_per_node:
1176 	init_per_node(sdp, UNDO);
1177 fail_inodes:
1178 	init_inodes(sdp, UNDO);
1179 fail_sb:
1180 	if (sdp->sd_root_dir)
1181 		dput(sdp->sd_root_dir);
1182 	if (sdp->sd_master_dir)
1183 		dput(sdp->sd_master_dir);
1184 	if (sb->s_root)
1185 		dput(sb->s_root);
1186 	sb->s_root = NULL;
1187 fail_locking:
1188 	init_locking(sdp, &mount_gh, UNDO);
1189 fail_lm:
1190 	complete_all(&sdp->sd_journal_ready);
1191 	gfs2_gl_hash_clear(sdp);
1192 	gfs2_lm_unmount(sdp);
1193 fail_debug:
1194 	gfs2_delete_debugfs_file(sdp);
1195 	free_percpu(sdp->sd_lkstats);
1196 	/* gfs2_sys_fs_del must be the last thing we do, since it causes
1197 	 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1198 	gfs2_sys_fs_del(sdp);
1199 	sb->s_fs_info = NULL;
1200 	return error;
1201 }
1202 
1203 static int set_gfs2_super(struct super_block *s, void *data)
1204 {
1205 	s->s_bdev = data;
1206 	s->s_dev = s->s_bdev->bd_dev;
1207 	s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1208 	return 0;
1209 }
1210 
1211 static int test_gfs2_super(struct super_block *s, void *ptr)
1212 {
1213 	struct block_device *bdev = ptr;
1214 	return (bdev == s->s_bdev);
1215 }
1216 
1217 /**
1218  * gfs2_mount - Get the GFS2 superblock
1219  * @fs_type: The GFS2 filesystem type
1220  * @flags: Mount flags
1221  * @dev_name: The name of the device
1222  * @data: The mount arguments
1223  *
1224  * Q. Why not use get_sb_bdev() ?
1225  * A. We need to select one of two root directories to mount, independent
1226  *    of whether this is the initial, or subsequent, mount of this sb
1227  *
1228  * Returns: 0 or -ve on error
1229  */
1230 
1231 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1232 		       const char *dev_name, void *data)
1233 {
1234 	struct block_device *bdev;
1235 	struct super_block *s;
1236 	fmode_t mode = FMODE_READ | FMODE_EXCL;
1237 	int error;
1238 	struct gfs2_args args;
1239 	struct gfs2_sbd *sdp;
1240 
1241 	if (!(flags & SB_RDONLY))
1242 		mode |= FMODE_WRITE;
1243 
1244 	bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1245 	if (IS_ERR(bdev))
1246 		return ERR_CAST(bdev);
1247 
1248 	/*
1249 	 * once the super is inserted into the list by sget, s_umount
1250 	 * will protect the lockfs code from trying to start a snapshot
1251 	 * while we are mounting
1252 	 */
1253 	mutex_lock(&bdev->bd_fsfreeze_mutex);
1254 	if (bdev->bd_fsfreeze_count > 0) {
1255 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
1256 		error = -EBUSY;
1257 		goto error_bdev;
1258 	}
1259 	s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1260 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
1261 	error = PTR_ERR(s);
1262 	if (IS_ERR(s))
1263 		goto error_bdev;
1264 
1265 	if (s->s_root) {
1266 		/*
1267 		 * s_umount nests inside bd_mutex during
1268 		 * __invalidate_device().  blkdev_put() acquires
1269 		 * bd_mutex and can't be called under s_umount.  Drop
1270 		 * s_umount temporarily.  This is safe as we're
1271 		 * holding an active reference.
1272 		 */
1273 		up_write(&s->s_umount);
1274 		blkdev_put(bdev, mode);
1275 		down_write(&s->s_umount);
1276 	} else {
1277 		/* s_mode must be set before deactivate_locked_super calls */
1278 		s->s_mode = mode;
1279 	}
1280 
1281 	memset(&args, 0, sizeof(args));
1282 	args.ar_quota = GFS2_QUOTA_DEFAULT;
1283 	args.ar_data = GFS2_DATA_DEFAULT;
1284 	args.ar_commit = 30;
1285 	args.ar_statfs_quantum = 30;
1286 	args.ar_quota_quantum = 60;
1287 	args.ar_errors = GFS2_ERRORS_DEFAULT;
1288 
1289 	error = gfs2_mount_args(&args, data);
1290 	if (error) {
1291 		pr_warn("can't parse mount arguments\n");
1292 		goto error_super;
1293 	}
1294 
1295 	if (s->s_root) {
1296 		error = -EBUSY;
1297 		if ((flags ^ s->s_flags) & SB_RDONLY)
1298 			goto error_super;
1299 	} else {
1300 		snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1301 		sb_set_blocksize(s, block_size(bdev));
1302 		error = fill_super(s, &args, flags & SB_SILENT ? 1 : 0);
1303 		if (error)
1304 			goto error_super;
1305 		s->s_flags |= SB_ACTIVE;
1306 		bdev->bd_super = s;
1307 	}
1308 
1309 	sdp = s->s_fs_info;
1310 	if (args.ar_meta)
1311 		return dget(sdp->sd_master_dir);
1312 	else
1313 		return dget(sdp->sd_root_dir);
1314 
1315 error_super:
1316 	deactivate_locked_super(s);
1317 	return ERR_PTR(error);
1318 error_bdev:
1319 	blkdev_put(bdev, mode);
1320 	return ERR_PTR(error);
1321 }
1322 
1323 static int set_meta_super(struct super_block *s, void *ptr)
1324 {
1325 	return -EINVAL;
1326 }
1327 
1328 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1329 			int flags, const char *dev_name, void *data)
1330 {
1331 	struct super_block *s;
1332 	struct gfs2_sbd *sdp;
1333 	struct path path;
1334 	int error;
1335 
1336 	error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1337 	if (error) {
1338 		pr_warn("path_lookup on %s returned error %d\n",
1339 			dev_name, error);
1340 		return ERR_PTR(error);
1341 	}
1342 	s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1343 		 path.dentry->d_sb->s_bdev);
1344 	path_put(&path);
1345 	if (IS_ERR(s)) {
1346 		pr_warn("gfs2 mount does not exist\n");
1347 		return ERR_CAST(s);
1348 	}
1349 	if ((flags ^ s->s_flags) & SB_RDONLY) {
1350 		deactivate_locked_super(s);
1351 		return ERR_PTR(-EBUSY);
1352 	}
1353 	sdp = s->s_fs_info;
1354 	return dget(sdp->sd_master_dir);
1355 }
1356 
1357 static void gfs2_kill_sb(struct super_block *sb)
1358 {
1359 	struct gfs2_sbd *sdp = sb->s_fs_info;
1360 
1361 	if (sdp == NULL) {
1362 		kill_block_super(sb);
1363 		return;
1364 	}
1365 
1366 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1367 	dput(sdp->sd_root_dir);
1368 	dput(sdp->sd_master_dir);
1369 	sdp->sd_root_dir = NULL;
1370 	sdp->sd_master_dir = NULL;
1371 	shrink_dcache_sb(sb);
1372 	free_percpu(sdp->sd_lkstats);
1373 	kill_block_super(sb);
1374 }
1375 
1376 struct file_system_type gfs2_fs_type = {
1377 	.name = "gfs2",
1378 	.fs_flags = FS_REQUIRES_DEV,
1379 	.mount = gfs2_mount,
1380 	.kill_sb = gfs2_kill_sb,
1381 	.owner = THIS_MODULE,
1382 };
1383 MODULE_ALIAS_FS("gfs2");
1384 
1385 struct file_system_type gfs2meta_fs_type = {
1386 	.name = "gfs2meta",
1387 	.fs_flags = FS_REQUIRES_DEV,
1388 	.mount = gfs2_mount_meta,
1389 	.owner = THIS_MODULE,
1390 };
1391 MODULE_ALIAS_FS("gfs2meta");
1392