xref: /linux/fs/gfs2/ops_fstype.c (revision f86fd32d)
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_glock_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 	INIT_LIST_HEAD(&sdp->sd_trunc_list);
110 	spin_lock_init(&sdp->sd_trunc_lock);
111 	spin_lock_init(&sdp->sd_bitmap_lock);
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_inode;
118 	mapping->flags = 0;
119 	mapping_set_gfp_mask(mapping, GFP_NOFS);
120 	mapping->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 	atomic_set(&sdp->sd_reserving_log, 0);
138 	init_waitqueue_head(&sdp->sd_reserving_log_wait);
139 	init_waitqueue_head(&sdp->sd_log_flush_wait);
140 	atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
141 	mutex_init(&sdp->sd_freeze_mutex);
142 
143 	return sdp;
144 
145 fail:
146 	free_sbd(sdp);
147 	return NULL;
148 }
149 
150 /**
151  * gfs2_check_sb - Check superblock
152  * @sdp: the filesystem
153  * @sb: The superblock
154  * @silent: Don't print a message if the check fails
155  *
156  * Checks the version code of the FS is one that we understand how to
157  * read and that the sizes of the various on-disk structures have not
158  * changed.
159  */
160 
161 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
162 {
163 	struct gfs2_sb_host *sb = &sdp->sd_sb;
164 
165 	if (sb->sb_magic != GFS2_MAGIC ||
166 	    sb->sb_type != GFS2_METATYPE_SB) {
167 		if (!silent)
168 			pr_warn("not a GFS2 filesystem\n");
169 		return -EINVAL;
170 	}
171 
172 	/*  If format numbers match exactly, we're done.  */
173 
174 	if (sb->sb_fs_format == GFS2_FORMAT_FS &&
175 	    sb->sb_multihost_format == GFS2_FORMAT_MULTI)
176 		return 0;
177 
178 	fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
179 
180 	return -EINVAL;
181 }
182 
183 static void end_bio_io_page(struct bio *bio)
184 {
185 	struct page *page = bio->bi_private;
186 
187 	if (!bio->bi_status)
188 		SetPageUptodate(page);
189 	else
190 		pr_warn("error %d reading superblock\n", bio->bi_status);
191 	unlock_page(page);
192 }
193 
194 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
195 {
196 	struct gfs2_sb_host *sb = &sdp->sd_sb;
197 	struct super_block *s = sdp->sd_vfs;
198 	const struct gfs2_sb *str = buf;
199 
200 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
201 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
202 	sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
203 	sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
204 	sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
205 	sb->sb_bsize = be32_to_cpu(str->sb_bsize);
206 	sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
207 	sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
208 	sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
209 	sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
210 	sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
211 
212 	memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
213 	memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
214 	memcpy(&s->s_uuid, str->sb_uuid, 16);
215 }
216 
217 /**
218  * gfs2_read_super - Read the gfs2 super block from disk
219  * @sdp: The GFS2 super block
220  * @sector: The location of the super block
221  * @error: The error code to return
222  *
223  * This uses the bio functions to read the super block from disk
224  * because we want to be 100% sure that we never read cached data.
225  * A super block is read twice only during each GFS2 mount and is
226  * never written to by the filesystem. The first time its read no
227  * locks are held, and the only details which are looked at are those
228  * relating to the locking protocol. Once locking is up and working,
229  * the sb is read again under the lock to establish the location of
230  * the master directory (contains pointers to journals etc) and the
231  * root directory.
232  *
233  * Returns: 0 on success or error
234  */
235 
236 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
237 {
238 	struct super_block *sb = sdp->sd_vfs;
239 	struct gfs2_sb *p;
240 	struct page *page;
241 	struct bio *bio;
242 
243 	page = alloc_page(GFP_NOFS);
244 	if (unlikely(!page))
245 		return -ENOMEM;
246 
247 	ClearPageUptodate(page);
248 	ClearPageDirty(page);
249 	lock_page(page);
250 
251 	bio = bio_alloc(GFP_NOFS, 1);
252 	bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
253 	bio_set_dev(bio, sb->s_bdev);
254 	bio_add_page(bio, page, PAGE_SIZE, 0);
255 
256 	bio->bi_end_io = end_bio_io_page;
257 	bio->bi_private = page;
258 	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
259 	submit_bio(bio);
260 	wait_on_page_locked(page);
261 	bio_put(bio);
262 	if (!PageUptodate(page)) {
263 		__free_page(page);
264 		return -EIO;
265 	}
266 	p = kmap(page);
267 	gfs2_sb_in(sdp, p);
268 	kunmap(page);
269 	__free_page(page);
270 	return gfs2_check_sb(sdp, silent);
271 }
272 
273 /**
274  * gfs2_read_sb - Read super block
275  * @sdp: The GFS2 superblock
276  * @silent: Don't print message if mount fails
277  *
278  */
279 
280 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
281 {
282 	u32 hash_blocks, ind_blocks, leaf_blocks;
283 	u32 tmp_blocks;
284 	unsigned int x;
285 	int error;
286 
287 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
288 	if (error) {
289 		if (!silent)
290 			fs_err(sdp, "can't read superblock\n");
291 		return error;
292 	}
293 
294 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
295 			       GFS2_BASIC_BLOCK_SHIFT;
296 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
297 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
298 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
299 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
300 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
301 	sdp->sd_ldptrs = (sdp->sd_sb.sb_bsize -
302 			  sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
303 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
304 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
305 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
306 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
307 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
308 				sizeof(struct gfs2_meta_header)) /
309 			        sizeof(struct gfs2_quota_change);
310 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
311 				     sizeof(struct gfs2_meta_header))
312 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
313 
314 	/* Compute maximum reservation required to add a entry to a directory */
315 
316 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
317 			     sdp->sd_jbsize);
318 
319 	ind_blocks = 0;
320 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
321 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
322 		ind_blocks += tmp_blocks;
323 	}
324 
325 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
326 
327 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
328 
329 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
330 				sizeof(struct gfs2_dinode);
331 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
332 	for (x = 2;; x++) {
333 		u64 space, d;
334 		u32 m;
335 
336 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
337 		d = space;
338 		m = do_div(d, sdp->sd_inptrs);
339 
340 		if (d != sdp->sd_heightsize[x - 1] || m)
341 			break;
342 		sdp->sd_heightsize[x] = space;
343 	}
344 	sdp->sd_max_height = x;
345 	sdp->sd_heightsize[x] = ~0;
346 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
347 
348 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
349 				      sizeof(struct gfs2_leaf)) /
350 				     GFS2_MIN_DIRENT_SIZE;
351 	return 0;
352 }
353 
354 static int init_names(struct gfs2_sbd *sdp, int silent)
355 {
356 	char *proto, *table;
357 	int error = 0;
358 
359 	proto = sdp->sd_args.ar_lockproto;
360 	table = sdp->sd_args.ar_locktable;
361 
362 	/*  Try to autodetect  */
363 
364 	if (!proto[0] || !table[0]) {
365 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
366 		if (error)
367 			return error;
368 
369 		if (!proto[0])
370 			proto = sdp->sd_sb.sb_lockproto;
371 		if (!table[0])
372 			table = sdp->sd_sb.sb_locktable;
373 	}
374 
375 	if (!table[0])
376 		table = sdp->sd_vfs->s_id;
377 
378 	strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
379 	strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
380 
381 	table = sdp->sd_table_name;
382 	while ((table = strchr(table, '/')))
383 		*table = '_';
384 
385 	return error;
386 }
387 
388 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
389 			int undo)
390 {
391 	int error = 0;
392 
393 	if (undo)
394 		goto fail_trans;
395 
396 	error = gfs2_glock_nq_num(sdp,
397 				  GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
398 				  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
399 				  mount_gh);
400 	if (error) {
401 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
402 		goto fail;
403 	}
404 
405 	error = gfs2_glock_nq_num(sdp,
406 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
407 				  LM_ST_SHARED,
408 				  LM_FLAG_NOEXP | GL_EXACT,
409 				  &sdp->sd_live_gh);
410 	if (error) {
411 		fs_err(sdp, "can't acquire live glock: %d\n", error);
412 		goto fail_mount;
413 	}
414 
415 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
416 			       CREATE, &sdp->sd_rename_gl);
417 	if (error) {
418 		fs_err(sdp, "can't create rename glock: %d\n", error);
419 		goto fail_live;
420 	}
421 
422 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
423 			       CREATE, &sdp->sd_freeze_gl);
424 	if (error) {
425 		fs_err(sdp, "can't create transaction glock: %d\n", error);
426 		goto fail_rename;
427 	}
428 
429 	return 0;
430 
431 fail_trans:
432 	gfs2_glock_put(sdp->sd_freeze_gl);
433 fail_rename:
434 	gfs2_glock_put(sdp->sd_rename_gl);
435 fail_live:
436 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
437 fail_mount:
438 	gfs2_glock_dq_uninit(mount_gh);
439 fail:
440 	return error;
441 }
442 
443 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
444 			    u64 no_addr, const char *name)
445 {
446 	struct gfs2_sbd *sdp = sb->s_fs_info;
447 	struct dentry *dentry;
448 	struct inode *inode;
449 
450 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
451 				  GFS2_BLKST_FREE /* ignore */);
452 	if (IS_ERR(inode)) {
453 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
454 		return PTR_ERR(inode);
455 	}
456 	dentry = d_make_root(inode);
457 	if (!dentry) {
458 		fs_err(sdp, "can't alloc %s dentry\n", name);
459 		return -ENOMEM;
460 	}
461 	*dptr = dentry;
462 	return 0;
463 }
464 
465 static int init_sb(struct gfs2_sbd *sdp, int silent)
466 {
467 	struct super_block *sb = sdp->sd_vfs;
468 	struct gfs2_holder sb_gh;
469 	u64 no_addr;
470 	int ret;
471 
472 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
473 				LM_ST_SHARED, 0, &sb_gh);
474 	if (ret) {
475 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
476 		return ret;
477 	}
478 
479 	ret = gfs2_read_sb(sdp, silent);
480 	if (ret) {
481 		fs_err(sdp, "can't read superblock: %d\n", ret);
482 		goto out;
483 	}
484 
485 	/* Set up the buffer cache and SB for real */
486 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
487 		ret = -EINVAL;
488 		fs_err(sdp, "FS block size (%u) is too small for device "
489 		       "block size (%u)\n",
490 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
491 		goto out;
492 	}
493 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
494 		ret = -EINVAL;
495 		fs_err(sdp, "FS block size (%u) is too big for machine "
496 		       "page size (%u)\n",
497 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
498 		goto out;
499 	}
500 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
501 
502 	/* Get the root inode */
503 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
504 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
505 	if (ret)
506 		goto out;
507 
508 	/* Get the master inode */
509 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
510 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
511 	if (ret) {
512 		dput(sdp->sd_root_dir);
513 		goto out;
514 	}
515 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
516 out:
517 	gfs2_glock_dq_uninit(&sb_gh);
518 	return ret;
519 }
520 
521 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
522 {
523 	char *message = "FIRSTMOUNT=Done";
524 	char *envp[] = { message, NULL };
525 
526 	fs_info(sdp, "first mount done, others may mount\n");
527 
528 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
529 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
530 
531 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
532 }
533 
534 /**
535  * gfs2_jindex_hold - Grab a lock on the jindex
536  * @sdp: The GFS2 superblock
537  * @ji_gh: the holder for the jindex glock
538  *
539  * Returns: errno
540  */
541 
542 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
543 {
544 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
545 	struct qstr name;
546 	char buf[20];
547 	struct gfs2_jdesc *jd;
548 	int error;
549 
550 	name.name = buf;
551 
552 	mutex_lock(&sdp->sd_jindex_mutex);
553 
554 	for (;;) {
555 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
556 		if (error)
557 			break;
558 
559 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
560 		name.hash = gfs2_disk_hash(name.name, name.len);
561 
562 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
563 		if (error == -ENOENT) {
564 			error = 0;
565 			break;
566 		}
567 
568 		gfs2_glock_dq_uninit(ji_gh);
569 
570 		if (error)
571 			break;
572 
573 		error = -ENOMEM;
574 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
575 		if (!jd)
576 			break;
577 
578 		INIT_LIST_HEAD(&jd->extent_list);
579 		INIT_LIST_HEAD(&jd->jd_revoke_list);
580 
581 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
582 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
583 		if (IS_ERR_OR_NULL(jd->jd_inode)) {
584 			if (!jd->jd_inode)
585 				error = -ENOENT;
586 			else
587 				error = PTR_ERR(jd->jd_inode);
588 			kfree(jd);
589 			break;
590 		}
591 
592 		spin_lock(&sdp->sd_jindex_spin);
593 		jd->jd_jid = sdp->sd_journals++;
594 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
595 		spin_unlock(&sdp->sd_jindex_spin);
596 	}
597 
598 	mutex_unlock(&sdp->sd_jindex_mutex);
599 
600 	return error;
601 }
602 
603 /**
604  * check_journal_clean - Make sure a journal is clean for a spectator mount
605  * @sdp: The GFS2 superblock
606  * @jd: The journal descriptor
607  *
608  * Returns: 0 if the journal is clean or locked, else an error
609  */
610 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
611 {
612 	int error;
613 	struct gfs2_holder j_gh;
614 	struct gfs2_log_header_host head;
615 	struct gfs2_inode *ip;
616 
617 	ip = GFS2_I(jd->jd_inode);
618 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
619 				   GL_EXACT | GL_NOCACHE, &j_gh);
620 	if (error) {
621 		fs_err(sdp, "Error locking journal for spectator mount.\n");
622 		return -EPERM;
623 	}
624 	error = gfs2_jdesc_check(jd);
625 	if (error) {
626 		fs_err(sdp, "Error checking journal for spectator mount.\n");
627 		goto out_unlock;
628 	}
629 	error = gfs2_find_jhead(jd, &head, false);
630 	if (error) {
631 		fs_err(sdp, "Error parsing journal for spectator mount.\n");
632 		goto out_unlock;
633 	}
634 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
635 		error = -EPERM;
636 		fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
637 		       "must not be a spectator.\n", jd->jd_jid);
638 	}
639 
640 out_unlock:
641 	gfs2_glock_dq_uninit(&j_gh);
642 	return error;
643 }
644 
645 static int init_journal(struct gfs2_sbd *sdp, int undo)
646 {
647 	struct inode *master = d_inode(sdp->sd_master_dir);
648 	struct gfs2_holder ji_gh;
649 	struct gfs2_inode *ip;
650 	int jindex = 1;
651 	int error = 0;
652 
653 	if (undo) {
654 		jindex = 0;
655 		goto fail_jinode_gh;
656 	}
657 
658 	sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
659 	if (IS_ERR(sdp->sd_jindex)) {
660 		fs_err(sdp, "can't lookup journal index: %d\n", error);
661 		return PTR_ERR(sdp->sd_jindex);
662 	}
663 
664 	/* Load in the journal index special file */
665 
666 	error = gfs2_jindex_hold(sdp, &ji_gh);
667 	if (error) {
668 		fs_err(sdp, "can't read journal index: %d\n", error);
669 		goto fail;
670 	}
671 
672 	error = -EUSERS;
673 	if (!gfs2_jindex_size(sdp)) {
674 		fs_err(sdp, "no journals!\n");
675 		goto fail_jindex;
676 	}
677 
678 	atomic_set(&sdp->sd_log_blks_needed, 0);
679 	if (sdp->sd_args.ar_spectator) {
680 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
681 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
682 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
683 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
684 	} else {
685 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
686 			fs_err(sdp, "can't mount journal #%u\n",
687 			       sdp->sd_lockstruct.ls_jid);
688 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
689 			       gfs2_jindex_size(sdp),
690 			       gfs2_jindex_size(sdp) - 1);
691 			goto fail_jindex;
692 		}
693 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
694 
695 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
696 					  &gfs2_journal_glops,
697 					  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
698 					  &sdp->sd_journal_gh);
699 		if (error) {
700 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
701 			goto fail_jindex;
702 		}
703 
704 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
705 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
706 					   LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
707 					   &sdp->sd_jinode_gh);
708 		if (error) {
709 			fs_err(sdp, "can't acquire journal inode glock: %d\n",
710 			       error);
711 			goto fail_journal_gh;
712 		}
713 
714 		error = gfs2_jdesc_check(sdp->sd_jdesc);
715 		if (error) {
716 			fs_err(sdp, "my journal (%u) is bad: %d\n",
717 			       sdp->sd_jdesc->jd_jid, error);
718 			goto fail_jinode_gh;
719 		}
720 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
721 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
722 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
723 
724 		/* Map the extents for this journal's blocks */
725 		gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
726 	}
727 	trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
728 
729 	if (sdp->sd_lockstruct.ls_first) {
730 		unsigned int x;
731 		for (x = 0; x < sdp->sd_journals; x++) {
732 			struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
733 
734 			if (sdp->sd_args.ar_spectator) {
735 				error = check_journal_clean(sdp, jd);
736 				if (error)
737 					goto fail_jinode_gh;
738 				continue;
739 			}
740 			error = gfs2_recover_journal(jd, true);
741 			if (error) {
742 				fs_err(sdp, "error recovering journal %u: %d\n",
743 				       x, error);
744 				goto fail_jinode_gh;
745 			}
746 		}
747 
748 		gfs2_others_may_mount(sdp);
749 	} else if (!sdp->sd_args.ar_spectator) {
750 		error = gfs2_recover_journal(sdp->sd_jdesc, true);
751 		if (error) {
752 			fs_err(sdp, "error recovering my journal: %d\n", error);
753 			goto fail_jinode_gh;
754 		}
755 	}
756 
757 	sdp->sd_log_idle = 1;
758 	set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
759 	gfs2_glock_dq_uninit(&ji_gh);
760 	jindex = 0;
761 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
762 	return 0;
763 
764 fail_jinode_gh:
765 	if (!sdp->sd_args.ar_spectator)
766 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
767 fail_journal_gh:
768 	if (!sdp->sd_args.ar_spectator)
769 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
770 fail_jindex:
771 	gfs2_jindex_free(sdp);
772 	if (jindex)
773 		gfs2_glock_dq_uninit(&ji_gh);
774 fail:
775 	iput(sdp->sd_jindex);
776 	return error;
777 }
778 
779 static struct lock_class_key gfs2_quota_imutex_key;
780 
781 static int init_inodes(struct gfs2_sbd *sdp, int undo)
782 {
783 	int error = 0;
784 	struct inode *master = d_inode(sdp->sd_master_dir);
785 
786 	if (undo)
787 		goto fail_qinode;
788 
789 	error = init_journal(sdp, undo);
790 	complete_all(&sdp->sd_journal_ready);
791 	if (error)
792 		goto fail;
793 
794 	/* Read in the master statfs inode */
795 	sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
796 	if (IS_ERR(sdp->sd_statfs_inode)) {
797 		error = PTR_ERR(sdp->sd_statfs_inode);
798 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
799 		goto fail_journal;
800 	}
801 
802 	/* Read in the resource index inode */
803 	sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
804 	if (IS_ERR(sdp->sd_rindex)) {
805 		error = PTR_ERR(sdp->sd_rindex);
806 		fs_err(sdp, "can't get resource index inode: %d\n", error);
807 		goto fail_statfs;
808 	}
809 	sdp->sd_rindex_uptodate = 0;
810 
811 	/* Read in the quota inode */
812 	sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
813 	if (IS_ERR(sdp->sd_quota_inode)) {
814 		error = PTR_ERR(sdp->sd_quota_inode);
815 		fs_err(sdp, "can't get quota file inode: %d\n", error);
816 		goto fail_rindex;
817 	}
818 	/*
819 	 * i_rwsem on quota files is special. Since this inode is hidden system
820 	 * file, we are safe to define locking ourselves.
821 	 */
822 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
823 			  &gfs2_quota_imutex_key);
824 
825 	error = gfs2_rindex_update(sdp);
826 	if (error)
827 		goto fail_qinode;
828 
829 	return 0;
830 
831 fail_qinode:
832 	iput(sdp->sd_quota_inode);
833 fail_rindex:
834 	gfs2_clear_rgrpd(sdp);
835 	iput(sdp->sd_rindex);
836 fail_statfs:
837 	iput(sdp->sd_statfs_inode);
838 fail_journal:
839 	init_journal(sdp, UNDO);
840 fail:
841 	return error;
842 }
843 
844 static int init_per_node(struct gfs2_sbd *sdp, int undo)
845 {
846 	struct inode *pn = NULL;
847 	char buf[30];
848 	int error = 0;
849 	struct gfs2_inode *ip;
850 	struct inode *master = d_inode(sdp->sd_master_dir);
851 
852 	if (sdp->sd_args.ar_spectator)
853 		return 0;
854 
855 	if (undo)
856 		goto fail_qc_gh;
857 
858 	pn = gfs2_lookup_simple(master, "per_node");
859 	if (IS_ERR(pn)) {
860 		error = PTR_ERR(pn);
861 		fs_err(sdp, "can't find per_node directory: %d\n", error);
862 		return error;
863 	}
864 
865 	sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
866 	sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
867 	if (IS_ERR(sdp->sd_sc_inode)) {
868 		error = PTR_ERR(sdp->sd_sc_inode);
869 		fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
870 		goto fail;
871 	}
872 
873 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
874 	sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
875 	if (IS_ERR(sdp->sd_qc_inode)) {
876 		error = PTR_ERR(sdp->sd_qc_inode);
877 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
878 		goto fail_ut_i;
879 	}
880 
881 	iput(pn);
882 	pn = NULL;
883 
884 	ip = GFS2_I(sdp->sd_sc_inode);
885 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
886 				   &sdp->sd_sc_gh);
887 	if (error) {
888 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
889 		goto fail_qc_i;
890 	}
891 
892 	ip = GFS2_I(sdp->sd_qc_inode);
893 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
894 				   &sdp->sd_qc_gh);
895 	if (error) {
896 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
897 		goto fail_ut_gh;
898 	}
899 
900 	return 0;
901 
902 fail_qc_gh:
903 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
904 fail_ut_gh:
905 	gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
906 fail_qc_i:
907 	iput(sdp->sd_qc_inode);
908 fail_ut_i:
909 	iput(sdp->sd_sc_inode);
910 fail:
911 	iput(pn);
912 	return error;
913 }
914 
915 static const match_table_t nolock_tokens = {
916 	{ Opt_jid, "jid=%d\n", },
917 	{ Opt_err, NULL },
918 };
919 
920 static const struct lm_lockops nolock_ops = {
921 	.lm_proto_name = "lock_nolock",
922 	.lm_put_lock = gfs2_glock_free,
923 	.lm_tokens = &nolock_tokens,
924 };
925 
926 /**
927  * gfs2_lm_mount - mount a locking protocol
928  * @sdp: the filesystem
929  * @args: mount arguments
930  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
931  *
932  * Returns: errno
933  */
934 
935 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
936 {
937 	const struct lm_lockops *lm;
938 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
939 	struct gfs2_args *args = &sdp->sd_args;
940 	const char *proto = sdp->sd_proto_name;
941 	const char *table = sdp->sd_table_name;
942 	char *o, *options;
943 	int ret;
944 
945 	if (!strcmp("lock_nolock", proto)) {
946 		lm = &nolock_ops;
947 		sdp->sd_args.ar_localflocks = 1;
948 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
949 	} else if (!strcmp("lock_dlm", proto)) {
950 		lm = &gfs2_dlm_ops;
951 #endif
952 	} else {
953 		pr_info("can't find protocol %s\n", proto);
954 		return -ENOENT;
955 	}
956 
957 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
958 
959 	ls->ls_ops = lm;
960 	ls->ls_first = 1;
961 
962 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
963 		substring_t tmp[MAX_OPT_ARGS];
964 		int token, option;
965 
966 		if (!o || !*o)
967 			continue;
968 
969 		token = match_token(o, *lm->lm_tokens, tmp);
970 		switch (token) {
971 		case Opt_jid:
972 			ret = match_int(&tmp[0], &option);
973 			if (ret || option < 0)
974 				goto hostdata_error;
975 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
976 				ls->ls_jid = option;
977 			break;
978 		case Opt_id:
979 		case Opt_nodir:
980 			/* Obsolete, but left for backward compat purposes */
981 			break;
982 		case Opt_first:
983 			ret = match_int(&tmp[0], &option);
984 			if (ret || (option != 0 && option != 1))
985 				goto hostdata_error;
986 			ls->ls_first = option;
987 			break;
988 		case Opt_err:
989 		default:
990 hostdata_error:
991 			fs_info(sdp, "unknown hostdata (%s)\n", o);
992 			return -EINVAL;
993 		}
994 	}
995 
996 	if (lm->lm_mount == NULL) {
997 		fs_info(sdp, "Now mounting FS...\n");
998 		complete_all(&sdp->sd_locking_init);
999 		return 0;
1000 	}
1001 	ret = lm->lm_mount(sdp, table);
1002 	if (ret == 0)
1003 		fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1004 	complete_all(&sdp->sd_locking_init);
1005 	return ret;
1006 }
1007 
1008 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1009 {
1010 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1011 	if (likely(!gfs2_withdrawn(sdp)) && lm->lm_unmount)
1012 		lm->lm_unmount(sdp);
1013 }
1014 
1015 static int wait_on_journal(struct gfs2_sbd *sdp)
1016 {
1017 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1018 		return 0;
1019 
1020 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1021 		? -EINTR : 0;
1022 }
1023 
1024 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1025 {
1026 	struct super_block *sb = sdp->sd_vfs;
1027 	char ro[20];
1028 	char spectator[20];
1029 	char *envp[] = { ro, spectator, NULL };
1030 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1031 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1032 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1033 }
1034 
1035 /**
1036  * gfs2_fill_super - Read in superblock
1037  * @sb: The VFS superblock
1038  * @args: Mount options
1039  * @silent: Don't complain if it's not a GFS2 filesystem
1040  *
1041  * Returns: -errno
1042  */
1043 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1044 {
1045 	struct gfs2_args *args = fc->fs_private;
1046 	int silent = fc->sb_flags & SB_SILENT;
1047 	struct gfs2_sbd *sdp;
1048 	struct gfs2_holder mount_gh;
1049 	int error;
1050 
1051 	sdp = init_sbd(sb);
1052 	if (!sdp) {
1053 		pr_warn("can't alloc struct gfs2_sbd\n");
1054 		return -ENOMEM;
1055 	}
1056 	sdp->sd_args = *args;
1057 
1058 	if (sdp->sd_args.ar_spectator) {
1059                 sb->s_flags |= SB_RDONLY;
1060 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1061 	}
1062 	if (sdp->sd_args.ar_posix_acl)
1063 		sb->s_flags |= SB_POSIXACL;
1064 	if (sdp->sd_args.ar_nobarrier)
1065 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1066 
1067 	sb->s_flags |= SB_NOSEC;
1068 	sb->s_magic = GFS2_MAGIC;
1069 	sb->s_op = &gfs2_super_ops;
1070 	sb->s_d_op = &gfs2_dops;
1071 	sb->s_export_op = &gfs2_export_ops;
1072 	sb->s_xattr = gfs2_xattr_handlers;
1073 	sb->s_qcop = &gfs2_quotactl_ops;
1074 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1075 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1076 	sb->s_time_gran = 1;
1077 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1078 
1079 	/* Set up the buffer cache and fill in some fake block size values
1080 	   to allow us to read-in the on-disk superblock. */
1081 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1082 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1083 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1084                                GFS2_BASIC_BLOCK_SHIFT;
1085 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1086 
1087 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1088 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1089 	if (sdp->sd_args.ar_statfs_quantum) {
1090 		sdp->sd_tune.gt_statfs_slow = 0;
1091 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1092 	} else {
1093 		sdp->sd_tune.gt_statfs_slow = 1;
1094 		sdp->sd_tune.gt_statfs_quantum = 30;
1095 	}
1096 
1097 	error = init_names(sdp, silent);
1098 	if (error) {
1099 		/* In this case, we haven't initialized sysfs, so we have to
1100 		   manually free the sdp. */
1101 		free_sbd(sdp);
1102 		sb->s_fs_info = NULL;
1103 		return error;
1104 	}
1105 
1106 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1107 
1108 	error = gfs2_sys_fs_add(sdp);
1109 	/*
1110 	 * If we hit an error here, gfs2_sys_fs_add will have called function
1111 	 * kobject_put which causes the sysfs usage count to go to zero, which
1112 	 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1113 	 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1114 	 * kobject_put to free sdp.
1115 	 */
1116 	if (error)
1117 		return error;
1118 
1119 	gfs2_create_debugfs_file(sdp);
1120 
1121 	error = gfs2_lm_mount(sdp, silent);
1122 	if (error)
1123 		goto fail_debug;
1124 
1125 	error = init_locking(sdp, &mount_gh, DO);
1126 	if (error)
1127 		goto fail_lm;
1128 
1129 	error = init_sb(sdp, silent);
1130 	if (error)
1131 		goto fail_locking;
1132 
1133 	error = wait_on_journal(sdp);
1134 	if (error)
1135 		goto fail_sb;
1136 
1137 	/*
1138 	 * If user space has failed to join the cluster or some similar
1139 	 * failure has occurred, then the journal id will contain a
1140 	 * negative (error) number. This will then be returned to the
1141 	 * caller (of the mount syscall). We do this even for spectator
1142 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1143 	 * the jid is unused in the spectator case)
1144 	 */
1145 	if (sdp->sd_lockstruct.ls_jid < 0) {
1146 		error = sdp->sd_lockstruct.ls_jid;
1147 		sdp->sd_lockstruct.ls_jid = 0;
1148 		goto fail_sb;
1149 	}
1150 
1151 	if (sdp->sd_args.ar_spectator)
1152 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1153 			 sdp->sd_table_name);
1154 	else
1155 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1156 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1157 
1158 	error = init_inodes(sdp, DO);
1159 	if (error)
1160 		goto fail_sb;
1161 
1162 	error = init_per_node(sdp, DO);
1163 	if (error)
1164 		goto fail_inodes;
1165 
1166 	error = gfs2_statfs_init(sdp);
1167 	if (error) {
1168 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1169 		goto fail_per_node;
1170 	}
1171 
1172 	if (!sb_rdonly(sb)) {
1173 		error = gfs2_make_fs_rw(sdp);
1174 		if (error) {
1175 			fs_err(sdp, "can't make FS RW: %d\n", error);
1176 			goto fail_per_node;
1177 		}
1178 	}
1179 
1180 	gfs2_glock_dq_uninit(&mount_gh);
1181 	gfs2_online_uevent(sdp);
1182 	return 0;
1183 
1184 fail_per_node:
1185 	init_per_node(sdp, UNDO);
1186 fail_inodes:
1187 	init_inodes(sdp, UNDO);
1188 fail_sb:
1189 	if (sdp->sd_root_dir)
1190 		dput(sdp->sd_root_dir);
1191 	if (sdp->sd_master_dir)
1192 		dput(sdp->sd_master_dir);
1193 	if (sb->s_root)
1194 		dput(sb->s_root);
1195 	sb->s_root = NULL;
1196 fail_locking:
1197 	init_locking(sdp, &mount_gh, UNDO);
1198 fail_lm:
1199 	complete_all(&sdp->sd_journal_ready);
1200 	gfs2_gl_hash_clear(sdp);
1201 	gfs2_lm_unmount(sdp);
1202 fail_debug:
1203 	gfs2_delete_debugfs_file(sdp);
1204 	/* gfs2_sys_fs_del must be the last thing we do, since it causes
1205 	 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1206 	gfs2_sys_fs_del(sdp);
1207 	sb->s_fs_info = NULL;
1208 	return error;
1209 }
1210 
1211 /**
1212  * gfs2_get_tree - Get the GFS2 superblock and root directory
1213  * @fc: The filesystem context
1214  *
1215  * Returns: 0 or -errno on error
1216  */
1217 static int gfs2_get_tree(struct fs_context *fc)
1218 {
1219 	struct gfs2_args *args = fc->fs_private;
1220 	struct gfs2_sbd *sdp;
1221 	int error;
1222 
1223 	error = get_tree_bdev(fc, gfs2_fill_super);
1224 	if (error)
1225 		return error;
1226 
1227 	sdp = fc->root->d_sb->s_fs_info;
1228 	dput(fc->root);
1229 	if (args->ar_meta)
1230 		fc->root = dget(sdp->sd_master_dir);
1231 	else
1232 		fc->root = dget(sdp->sd_root_dir);
1233 	return 0;
1234 }
1235 
1236 static void gfs2_fc_free(struct fs_context *fc)
1237 {
1238 	struct gfs2_args *args = fc->fs_private;
1239 
1240 	kfree(args);
1241 }
1242 
1243 enum gfs2_param {
1244 	Opt_lockproto,
1245 	Opt_locktable,
1246 	Opt_hostdata,
1247 	Opt_spectator,
1248 	Opt_ignore_local_fs,
1249 	Opt_localflocks,
1250 	Opt_localcaching,
1251 	Opt_debug,
1252 	Opt_upgrade,
1253 	Opt_acl,
1254 	Opt_quota,
1255 	Opt_quota_flag,
1256 	Opt_suiddir,
1257 	Opt_data,
1258 	Opt_meta,
1259 	Opt_discard,
1260 	Opt_commit,
1261 	Opt_errors,
1262 	Opt_statfs_quantum,
1263 	Opt_statfs_percent,
1264 	Opt_quota_quantum,
1265 	Opt_barrier,
1266 	Opt_rgrplvb,
1267 	Opt_loccookie,
1268 };
1269 
1270 static const struct constant_table gfs2_param_quota[] = {
1271 	{"off",        GFS2_QUOTA_OFF},
1272 	{"account",    GFS2_QUOTA_ACCOUNT},
1273 	{"on",         GFS2_QUOTA_ON},
1274 	{}
1275 };
1276 
1277 enum opt_data {
1278 	Opt_data_writeback = GFS2_DATA_WRITEBACK,
1279 	Opt_data_ordered   = GFS2_DATA_ORDERED,
1280 };
1281 
1282 static const struct constant_table gfs2_param_data[] = {
1283 	{"writeback",  Opt_data_writeback },
1284 	{"ordered",    Opt_data_ordered },
1285 	{}
1286 };
1287 
1288 enum opt_errors {
1289 	Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1290 	Opt_errors_panic    = GFS2_ERRORS_PANIC,
1291 };
1292 
1293 static const struct constant_table gfs2_param_errors[] = {
1294 	{"withdraw",   Opt_errors_withdraw },
1295 	{"panic",      Opt_errors_panic },
1296 	{}
1297 };
1298 
1299 static const struct fs_parameter_spec gfs2_fs_parameters[] = {
1300 	fsparam_string ("lockproto",          Opt_lockproto),
1301 	fsparam_string ("locktable",          Opt_locktable),
1302 	fsparam_string ("hostdata",           Opt_hostdata),
1303 	fsparam_flag   ("spectator",          Opt_spectator),
1304 	fsparam_flag   ("norecovery",         Opt_spectator),
1305 	fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1306 	fsparam_flag   ("localflocks",        Opt_localflocks),
1307 	fsparam_flag   ("localcaching",       Opt_localcaching),
1308 	fsparam_flag_no("debug",              Opt_debug),
1309 	fsparam_flag   ("upgrade",            Opt_upgrade),
1310 	fsparam_flag_no("acl",                Opt_acl),
1311 	fsparam_flag_no("suiddir",            Opt_suiddir),
1312 	fsparam_enum   ("data",               Opt_data, gfs2_param_data),
1313 	fsparam_flag   ("meta",               Opt_meta),
1314 	fsparam_flag_no("discard",            Opt_discard),
1315 	fsparam_s32    ("commit",             Opt_commit),
1316 	fsparam_enum   ("errors",             Opt_errors, gfs2_param_errors),
1317 	fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1318 	fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1319 	fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1320 	fsparam_flag_no("barrier",            Opt_barrier),
1321 	fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1322 	fsparam_flag_no("loccookie",          Opt_loccookie),
1323 	/* quota can be a flag or an enum so it gets special treatment */
1324 	fsparam_flag_no("quota",	      Opt_quota_flag),
1325 	fsparam_enum("quota",		      Opt_quota, gfs2_param_quota),
1326 	{}
1327 };
1328 
1329 /* Parse a single mount parameter */
1330 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1331 {
1332 	struct gfs2_args *args = fc->fs_private;
1333 	struct fs_parse_result result;
1334 	int o;
1335 
1336 	o = fs_parse(fc, gfs2_fs_parameters, param, &result);
1337 	if (o < 0)
1338 		return o;
1339 
1340 	switch (o) {
1341 	case Opt_lockproto:
1342 		strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1343 		break;
1344 	case Opt_locktable:
1345 		strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1346 		break;
1347 	case Opt_hostdata:
1348 		strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1349 		break;
1350 	case Opt_spectator:
1351 		args->ar_spectator = 1;
1352 		break;
1353 	case Opt_ignore_local_fs:
1354 		/* Retained for backwards compat only */
1355 		break;
1356 	case Opt_localflocks:
1357 		args->ar_localflocks = 1;
1358 		break;
1359 	case Opt_localcaching:
1360 		/* Retained for backwards compat only */
1361 		break;
1362 	case Opt_debug:
1363 		if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1364 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1365 		args->ar_debug = result.boolean;
1366 		break;
1367 	case Opt_upgrade:
1368 		/* Retained for backwards compat only */
1369 		break;
1370 	case Opt_acl:
1371 		args->ar_posix_acl = result.boolean;
1372 		break;
1373 	case Opt_quota_flag:
1374 		args->ar_quota = result.negated ? GFS2_QUOTA_OFF : GFS2_QUOTA_ON;
1375 		break;
1376 	case Opt_quota:
1377 		args->ar_quota = result.int_32;
1378 		break;
1379 	case Opt_suiddir:
1380 		args->ar_suiddir = result.boolean;
1381 		break;
1382 	case Opt_data:
1383 		/* The uint_32 result maps directly to GFS2_DATA_* */
1384 		args->ar_data = result.uint_32;
1385 		break;
1386 	case Opt_meta:
1387 		args->ar_meta = 1;
1388 		break;
1389 	case Opt_discard:
1390 		args->ar_discard = result.boolean;
1391 		break;
1392 	case Opt_commit:
1393 		if (result.int_32 <= 0)
1394 			return invalfc(fc, "commit mount option requires a positive numeric argument");
1395 		args->ar_commit = result.int_32;
1396 		break;
1397 	case Opt_statfs_quantum:
1398 		if (result.int_32 < 0)
1399 			return invalfc(fc, "statfs_quantum mount option requires a non-negative numeric argument");
1400 		args->ar_statfs_quantum = result.int_32;
1401 		break;
1402 	case Opt_quota_quantum:
1403 		if (result.int_32 <= 0)
1404 			return invalfc(fc, "quota_quantum mount option requires a positive numeric argument");
1405 		args->ar_quota_quantum = result.int_32;
1406 		break;
1407 	case Opt_statfs_percent:
1408 		if (result.int_32 < 0 || result.int_32 > 100)
1409 			return invalfc(fc, "statfs_percent mount option requires a numeric argument between 0 and 100");
1410 		args->ar_statfs_percent = result.int_32;
1411 		break;
1412 	case Opt_errors:
1413 		if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1414 			return invalfc(fc, "-o debug and -o errors=panic are mutually exclusive");
1415 		args->ar_errors = result.uint_32;
1416 		break;
1417 	case Opt_barrier:
1418 		args->ar_nobarrier = result.boolean;
1419 		break;
1420 	case Opt_rgrplvb:
1421 		args->ar_rgrplvb = result.boolean;
1422 		break;
1423 	case Opt_loccookie:
1424 		args->ar_loccookie = result.boolean;
1425 		break;
1426 	default:
1427 		return invalfc(fc, "invalid mount option: %s", param->key);
1428 	}
1429 	return 0;
1430 }
1431 
1432 static int gfs2_reconfigure(struct fs_context *fc)
1433 {
1434 	struct super_block *sb = fc->root->d_sb;
1435 	struct gfs2_sbd *sdp = sb->s_fs_info;
1436 	struct gfs2_args *oldargs = &sdp->sd_args;
1437 	struct gfs2_args *newargs = fc->fs_private;
1438 	struct gfs2_tune *gt = &sdp->sd_tune;
1439 	int error = 0;
1440 
1441 	sync_filesystem(sb);
1442 
1443 	spin_lock(&gt->gt_spin);
1444 	oldargs->ar_commit = gt->gt_logd_secs;
1445 	oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1446 	if (gt->gt_statfs_slow)
1447 		oldargs->ar_statfs_quantum = 0;
1448 	else
1449 		oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1450 	spin_unlock(&gt->gt_spin);
1451 
1452 	if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1453 		errorfc(fc, "reconfiguration of locking protocol not allowed");
1454 		return -EINVAL;
1455 	}
1456 	if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1457 		errorfc(fc, "reconfiguration of lock table not allowed");
1458 		return -EINVAL;
1459 	}
1460 	if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1461 		errorfc(fc, "reconfiguration of host data not allowed");
1462 		return -EINVAL;
1463 	}
1464 	if (newargs->ar_spectator != oldargs->ar_spectator) {
1465 		errorfc(fc, "reconfiguration of spectator mode not allowed");
1466 		return -EINVAL;
1467 	}
1468 	if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1469 		errorfc(fc, "reconfiguration of localflocks not allowed");
1470 		return -EINVAL;
1471 	}
1472 	if (newargs->ar_meta != oldargs->ar_meta) {
1473 		errorfc(fc, "switching between gfs2 and gfs2meta not allowed");
1474 		return -EINVAL;
1475 	}
1476 	if (oldargs->ar_spectator)
1477 		fc->sb_flags |= SB_RDONLY;
1478 
1479 	if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1480 		if (fc->sb_flags & SB_RDONLY) {
1481 			error = gfs2_make_fs_ro(sdp);
1482 			if (error)
1483 				errorfc(fc, "unable to remount read-only");
1484 		} else {
1485 			error = gfs2_make_fs_rw(sdp);
1486 			if (error)
1487 				errorfc(fc, "unable to remount read-write");
1488 		}
1489 	}
1490 	sdp->sd_args = *newargs;
1491 
1492 	if (sdp->sd_args.ar_posix_acl)
1493 		sb->s_flags |= SB_POSIXACL;
1494 	else
1495 		sb->s_flags &= ~SB_POSIXACL;
1496 	if (sdp->sd_args.ar_nobarrier)
1497 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1498 	else
1499 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1500 	spin_lock(&gt->gt_spin);
1501 	gt->gt_logd_secs = newargs->ar_commit;
1502 	gt->gt_quota_quantum = newargs->ar_quota_quantum;
1503 	if (newargs->ar_statfs_quantum) {
1504 		gt->gt_statfs_slow = 0;
1505 		gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1506 	}
1507 	else {
1508 		gt->gt_statfs_slow = 1;
1509 		gt->gt_statfs_quantum = 30;
1510 	}
1511 	spin_unlock(&gt->gt_spin);
1512 
1513 	gfs2_online_uevent(sdp);
1514 	return error;
1515 }
1516 
1517 static const struct fs_context_operations gfs2_context_ops = {
1518 	.free        = gfs2_fc_free,
1519 	.parse_param = gfs2_parse_param,
1520 	.get_tree    = gfs2_get_tree,
1521 	.reconfigure = gfs2_reconfigure,
1522 };
1523 
1524 /* Set up the filesystem mount context */
1525 static int gfs2_init_fs_context(struct fs_context *fc)
1526 {
1527 	struct gfs2_args *args;
1528 
1529 	args = kmalloc(sizeof(*args), GFP_KERNEL);
1530 	if (args == NULL)
1531 		return -ENOMEM;
1532 
1533 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1534 		struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1535 
1536 		*args = sdp->sd_args;
1537 	} else {
1538 		memset(args, 0, sizeof(*args));
1539 		args->ar_quota = GFS2_QUOTA_DEFAULT;
1540 		args->ar_data = GFS2_DATA_DEFAULT;
1541 		args->ar_commit = 30;
1542 		args->ar_statfs_quantum = 30;
1543 		args->ar_quota_quantum = 60;
1544 		args->ar_errors = GFS2_ERRORS_DEFAULT;
1545 	}
1546 	fc->fs_private = args;
1547 	fc->ops = &gfs2_context_ops;
1548 	return 0;
1549 }
1550 
1551 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1552 {
1553 	return -EINVAL;
1554 }
1555 
1556 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1557 {
1558 	return (fc->sget_key == s->s_bdev);
1559 }
1560 
1561 static int gfs2_meta_get_tree(struct fs_context *fc)
1562 {
1563 	struct super_block *s;
1564 	struct gfs2_sbd *sdp;
1565 	struct path path;
1566 	int error;
1567 
1568 	if (!fc->source || !*fc->source)
1569 		return -EINVAL;
1570 
1571 	error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1572 	if (error) {
1573 		pr_warn("path_lookup on %s returned error %d\n",
1574 		        fc->source, error);
1575 		return error;
1576 	}
1577 	fc->fs_type = &gfs2_fs_type;
1578 	fc->sget_key = path.dentry->d_sb->s_bdev;
1579 	s = sget_fc(fc, test_meta_super, set_meta_super);
1580 	path_put(&path);
1581 	if (IS_ERR(s)) {
1582 		pr_warn("gfs2 mount does not exist\n");
1583 		return PTR_ERR(s);
1584 	}
1585 	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1586 		deactivate_locked_super(s);
1587 		return -EBUSY;
1588 	}
1589 	sdp = s->s_fs_info;
1590 	fc->root = dget(sdp->sd_master_dir);
1591 	return 0;
1592 }
1593 
1594 static const struct fs_context_operations gfs2_meta_context_ops = {
1595 	.free        = gfs2_fc_free,
1596 	.get_tree    = gfs2_meta_get_tree,
1597 };
1598 
1599 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1600 {
1601 	int ret = gfs2_init_fs_context(fc);
1602 
1603 	if (ret)
1604 		return ret;
1605 
1606 	fc->ops = &gfs2_meta_context_ops;
1607 	return 0;
1608 }
1609 
1610 static void gfs2_kill_sb(struct super_block *sb)
1611 {
1612 	struct gfs2_sbd *sdp = sb->s_fs_info;
1613 
1614 	if (sdp == NULL) {
1615 		kill_block_super(sb);
1616 		return;
1617 	}
1618 
1619 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1620 	dput(sdp->sd_root_dir);
1621 	dput(sdp->sd_master_dir);
1622 	sdp->sd_root_dir = NULL;
1623 	sdp->sd_master_dir = NULL;
1624 	shrink_dcache_sb(sb);
1625 	kill_block_super(sb);
1626 }
1627 
1628 struct file_system_type gfs2_fs_type = {
1629 	.name = "gfs2",
1630 	.fs_flags = FS_REQUIRES_DEV,
1631 	.init_fs_context = gfs2_init_fs_context,
1632 	.parameters = gfs2_fs_parameters,
1633 	.kill_sb = gfs2_kill_sb,
1634 	.owner = THIS_MODULE,
1635 };
1636 MODULE_ALIAS_FS("gfs2");
1637 
1638 struct file_system_type gfs2meta_fs_type = {
1639 	.name = "gfs2meta",
1640 	.fs_flags = FS_REQUIRES_DEV,
1641 	.init_fs_context = gfs2_meta_init_fs_context,
1642 	.owner = THIS_MODULE,
1643 };
1644 MODULE_ALIAS_FS("gfs2meta");
1645