xref: /linux/fs/gfs2/ops_fstype.c (revision 52338415)
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_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
302 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
303 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
304 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
305 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
306 				sizeof(struct gfs2_meta_header)) /
307 			        sizeof(struct gfs2_quota_change);
308 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
309 				     sizeof(struct gfs2_meta_header))
310 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
311 
312 	/* Compute maximum reservation required to add a entry to a directory */
313 
314 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
315 			     sdp->sd_jbsize);
316 
317 	ind_blocks = 0;
318 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
319 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
320 		ind_blocks += tmp_blocks;
321 	}
322 
323 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
324 
325 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
326 
327 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
328 				sizeof(struct gfs2_dinode);
329 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
330 	for (x = 2;; x++) {
331 		u64 space, d;
332 		u32 m;
333 
334 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
335 		d = space;
336 		m = do_div(d, sdp->sd_inptrs);
337 
338 		if (d != sdp->sd_heightsize[x - 1] || m)
339 			break;
340 		sdp->sd_heightsize[x] = space;
341 	}
342 	sdp->sd_max_height = x;
343 	sdp->sd_heightsize[x] = ~0;
344 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
345 
346 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
347 				      sizeof(struct gfs2_leaf)) /
348 				     GFS2_MIN_DIRENT_SIZE;
349 	return 0;
350 }
351 
352 static int init_names(struct gfs2_sbd *sdp, int silent)
353 {
354 	char *proto, *table;
355 	int error = 0;
356 
357 	proto = sdp->sd_args.ar_lockproto;
358 	table = sdp->sd_args.ar_locktable;
359 
360 	/*  Try to autodetect  */
361 
362 	if (!proto[0] || !table[0]) {
363 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
364 		if (error)
365 			return error;
366 
367 		if (!proto[0])
368 			proto = sdp->sd_sb.sb_lockproto;
369 		if (!table[0])
370 			table = sdp->sd_sb.sb_locktable;
371 	}
372 
373 	if (!table[0])
374 		table = sdp->sd_vfs->s_id;
375 
376 	strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
377 	strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
378 
379 	table = sdp->sd_table_name;
380 	while ((table = strchr(table, '/')))
381 		*table = '_';
382 
383 	return error;
384 }
385 
386 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
387 			int undo)
388 {
389 	int error = 0;
390 
391 	if (undo)
392 		goto fail_trans;
393 
394 	error = gfs2_glock_nq_num(sdp,
395 				  GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
396 				  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
397 				  mount_gh);
398 	if (error) {
399 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
400 		goto fail;
401 	}
402 
403 	error = gfs2_glock_nq_num(sdp,
404 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
405 				  LM_ST_SHARED,
406 				  LM_FLAG_NOEXP | GL_EXACT,
407 				  &sdp->sd_live_gh);
408 	if (error) {
409 		fs_err(sdp, "can't acquire live glock: %d\n", error);
410 		goto fail_mount;
411 	}
412 
413 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
414 			       CREATE, &sdp->sd_rename_gl);
415 	if (error) {
416 		fs_err(sdp, "can't create rename glock: %d\n", error);
417 		goto fail_live;
418 	}
419 
420 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
421 			       CREATE, &sdp->sd_freeze_gl);
422 	if (error) {
423 		fs_err(sdp, "can't create transaction glock: %d\n", error);
424 		goto fail_rename;
425 	}
426 
427 	return 0;
428 
429 fail_trans:
430 	gfs2_glock_put(sdp->sd_freeze_gl);
431 fail_rename:
432 	gfs2_glock_put(sdp->sd_rename_gl);
433 fail_live:
434 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
435 fail_mount:
436 	gfs2_glock_dq_uninit(mount_gh);
437 fail:
438 	return error;
439 }
440 
441 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
442 			    u64 no_addr, const char *name)
443 {
444 	struct gfs2_sbd *sdp = sb->s_fs_info;
445 	struct dentry *dentry;
446 	struct inode *inode;
447 
448 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
449 				  GFS2_BLKST_FREE /* ignore */);
450 	if (IS_ERR(inode)) {
451 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
452 		return PTR_ERR(inode);
453 	}
454 	dentry = d_make_root(inode);
455 	if (!dentry) {
456 		fs_err(sdp, "can't alloc %s dentry\n", name);
457 		return -ENOMEM;
458 	}
459 	*dptr = dentry;
460 	return 0;
461 }
462 
463 static int init_sb(struct gfs2_sbd *sdp, int silent)
464 {
465 	struct super_block *sb = sdp->sd_vfs;
466 	struct gfs2_holder sb_gh;
467 	u64 no_addr;
468 	int ret;
469 
470 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
471 				LM_ST_SHARED, 0, &sb_gh);
472 	if (ret) {
473 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
474 		return ret;
475 	}
476 
477 	ret = gfs2_read_sb(sdp, silent);
478 	if (ret) {
479 		fs_err(sdp, "can't read superblock: %d\n", ret);
480 		goto out;
481 	}
482 
483 	/* Set up the buffer cache and SB for real */
484 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
485 		ret = -EINVAL;
486 		fs_err(sdp, "FS block size (%u) is too small for device "
487 		       "block size (%u)\n",
488 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
489 		goto out;
490 	}
491 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
492 		ret = -EINVAL;
493 		fs_err(sdp, "FS block size (%u) is too big for machine "
494 		       "page size (%u)\n",
495 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
496 		goto out;
497 	}
498 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
499 
500 	/* Get the root inode */
501 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
502 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
503 	if (ret)
504 		goto out;
505 
506 	/* Get the master inode */
507 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
508 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
509 	if (ret) {
510 		dput(sdp->sd_root_dir);
511 		goto out;
512 	}
513 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
514 out:
515 	gfs2_glock_dq_uninit(&sb_gh);
516 	return ret;
517 }
518 
519 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
520 {
521 	char *message = "FIRSTMOUNT=Done";
522 	char *envp[] = { message, NULL };
523 
524 	fs_info(sdp, "first mount done, others may mount\n");
525 
526 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
527 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
528 
529 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
530 }
531 
532 /**
533  * gfs2_jindex_hold - Grab a lock on the jindex
534  * @sdp: The GFS2 superblock
535  * @ji_gh: the holder for the jindex glock
536  *
537  * Returns: errno
538  */
539 
540 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
541 {
542 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
543 	struct qstr name;
544 	char buf[20];
545 	struct gfs2_jdesc *jd;
546 	int error;
547 
548 	name.name = buf;
549 
550 	mutex_lock(&sdp->sd_jindex_mutex);
551 
552 	for (;;) {
553 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
554 		if (error)
555 			break;
556 
557 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
558 		name.hash = gfs2_disk_hash(name.name, name.len);
559 
560 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
561 		if (error == -ENOENT) {
562 			error = 0;
563 			break;
564 		}
565 
566 		gfs2_glock_dq_uninit(ji_gh);
567 
568 		if (error)
569 			break;
570 
571 		error = -ENOMEM;
572 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
573 		if (!jd)
574 			break;
575 
576 		INIT_LIST_HEAD(&jd->extent_list);
577 		INIT_LIST_HEAD(&jd->jd_revoke_list);
578 
579 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
580 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
581 		if (IS_ERR_OR_NULL(jd->jd_inode)) {
582 			if (!jd->jd_inode)
583 				error = -ENOENT;
584 			else
585 				error = PTR_ERR(jd->jd_inode);
586 			kfree(jd);
587 			break;
588 		}
589 
590 		spin_lock(&sdp->sd_jindex_spin);
591 		jd->jd_jid = sdp->sd_journals++;
592 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
593 		spin_unlock(&sdp->sd_jindex_spin);
594 	}
595 
596 	mutex_unlock(&sdp->sd_jindex_mutex);
597 
598 	return error;
599 }
600 
601 /**
602  * check_journal_clean - Make sure a journal is clean for a spectator mount
603  * @sdp: The GFS2 superblock
604  * @jd: The journal descriptor
605  *
606  * Returns: 0 if the journal is clean or locked, else an error
607  */
608 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
609 {
610 	int error;
611 	struct gfs2_holder j_gh;
612 	struct gfs2_log_header_host head;
613 	struct gfs2_inode *ip;
614 
615 	ip = GFS2_I(jd->jd_inode);
616 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
617 				   GL_EXACT | GL_NOCACHE, &j_gh);
618 	if (error) {
619 		fs_err(sdp, "Error locking journal for spectator mount.\n");
620 		return -EPERM;
621 	}
622 	error = gfs2_jdesc_check(jd);
623 	if (error) {
624 		fs_err(sdp, "Error checking journal for spectator mount.\n");
625 		goto out_unlock;
626 	}
627 	error = gfs2_find_jhead(jd, &head, false);
628 	if (error) {
629 		fs_err(sdp, "Error parsing journal for spectator mount.\n");
630 		goto out_unlock;
631 	}
632 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
633 		error = -EPERM;
634 		fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
635 		       "must not be a spectator.\n", jd->jd_jid);
636 	}
637 
638 out_unlock:
639 	gfs2_glock_dq_uninit(&j_gh);
640 	return error;
641 }
642 
643 static int init_journal(struct gfs2_sbd *sdp, int undo)
644 {
645 	struct inode *master = d_inode(sdp->sd_master_dir);
646 	struct gfs2_holder ji_gh;
647 	struct gfs2_inode *ip;
648 	int jindex = 1;
649 	int error = 0;
650 
651 	if (undo) {
652 		jindex = 0;
653 		goto fail_jinode_gh;
654 	}
655 
656 	sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
657 	if (IS_ERR(sdp->sd_jindex)) {
658 		fs_err(sdp, "can't lookup journal index: %d\n", error);
659 		return PTR_ERR(sdp->sd_jindex);
660 	}
661 
662 	/* Load in the journal index special file */
663 
664 	error = gfs2_jindex_hold(sdp, &ji_gh);
665 	if (error) {
666 		fs_err(sdp, "can't read journal index: %d\n", error);
667 		goto fail;
668 	}
669 
670 	error = -EUSERS;
671 	if (!gfs2_jindex_size(sdp)) {
672 		fs_err(sdp, "no journals!\n");
673 		goto fail_jindex;
674 	}
675 
676 	atomic_set(&sdp->sd_log_blks_needed, 0);
677 	if (sdp->sd_args.ar_spectator) {
678 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
679 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
680 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
681 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
682 	} else {
683 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
684 			fs_err(sdp, "can't mount journal #%u\n",
685 			       sdp->sd_lockstruct.ls_jid);
686 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
687 			       gfs2_jindex_size(sdp),
688 			       gfs2_jindex_size(sdp) - 1);
689 			goto fail_jindex;
690 		}
691 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
692 
693 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
694 					  &gfs2_journal_glops,
695 					  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
696 					  &sdp->sd_journal_gh);
697 		if (error) {
698 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
699 			goto fail_jindex;
700 		}
701 
702 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
703 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
704 					   LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
705 					   &sdp->sd_jinode_gh);
706 		if (error) {
707 			fs_err(sdp, "can't acquire journal inode glock: %d\n",
708 			       error);
709 			goto fail_journal_gh;
710 		}
711 
712 		error = gfs2_jdesc_check(sdp->sd_jdesc);
713 		if (error) {
714 			fs_err(sdp, "my journal (%u) is bad: %d\n",
715 			       sdp->sd_jdesc->jd_jid, error);
716 			goto fail_jinode_gh;
717 		}
718 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
719 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
720 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
721 
722 		/* Map the extents for this journal's blocks */
723 		gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
724 	}
725 	trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
726 
727 	if (sdp->sd_lockstruct.ls_first) {
728 		unsigned int x;
729 		for (x = 0; x < sdp->sd_journals; x++) {
730 			struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
731 
732 			if (sdp->sd_args.ar_spectator) {
733 				error = check_journal_clean(sdp, jd);
734 				if (error)
735 					goto fail_jinode_gh;
736 				continue;
737 			}
738 			error = gfs2_recover_journal(jd, true);
739 			if (error) {
740 				fs_err(sdp, "error recovering journal %u: %d\n",
741 				       x, error);
742 				goto fail_jinode_gh;
743 			}
744 		}
745 
746 		gfs2_others_may_mount(sdp);
747 	} else if (!sdp->sd_args.ar_spectator) {
748 		error = gfs2_recover_journal(sdp->sd_jdesc, true);
749 		if (error) {
750 			fs_err(sdp, "error recovering my journal: %d\n", error);
751 			goto fail_jinode_gh;
752 		}
753 	}
754 
755 	sdp->sd_log_idle = 1;
756 	set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
757 	gfs2_glock_dq_uninit(&ji_gh);
758 	jindex = 0;
759 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
760 	return 0;
761 
762 fail_jinode_gh:
763 	if (!sdp->sd_args.ar_spectator)
764 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
765 fail_journal_gh:
766 	if (!sdp->sd_args.ar_spectator)
767 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
768 fail_jindex:
769 	gfs2_jindex_free(sdp);
770 	if (jindex)
771 		gfs2_glock_dq_uninit(&ji_gh);
772 fail:
773 	iput(sdp->sd_jindex);
774 	return error;
775 }
776 
777 static struct lock_class_key gfs2_quota_imutex_key;
778 
779 static int init_inodes(struct gfs2_sbd *sdp, int undo)
780 {
781 	int error = 0;
782 	struct inode *master = d_inode(sdp->sd_master_dir);
783 
784 	if (undo)
785 		goto fail_qinode;
786 
787 	error = init_journal(sdp, undo);
788 	complete_all(&sdp->sd_journal_ready);
789 	if (error)
790 		goto fail;
791 
792 	/* Read in the master statfs inode */
793 	sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
794 	if (IS_ERR(sdp->sd_statfs_inode)) {
795 		error = PTR_ERR(sdp->sd_statfs_inode);
796 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
797 		goto fail_journal;
798 	}
799 
800 	/* Read in the resource index inode */
801 	sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
802 	if (IS_ERR(sdp->sd_rindex)) {
803 		error = PTR_ERR(sdp->sd_rindex);
804 		fs_err(sdp, "can't get resource index inode: %d\n", error);
805 		goto fail_statfs;
806 	}
807 	sdp->sd_rindex_uptodate = 0;
808 
809 	/* Read in the quota inode */
810 	sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
811 	if (IS_ERR(sdp->sd_quota_inode)) {
812 		error = PTR_ERR(sdp->sd_quota_inode);
813 		fs_err(sdp, "can't get quota file inode: %d\n", error);
814 		goto fail_rindex;
815 	}
816 	/*
817 	 * i_rwsem on quota files is special. Since this inode is hidden system
818 	 * file, we are safe to define locking ourselves.
819 	 */
820 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
821 			  &gfs2_quota_imutex_key);
822 
823 	error = gfs2_rindex_update(sdp);
824 	if (error)
825 		goto fail_qinode;
826 
827 	return 0;
828 
829 fail_qinode:
830 	iput(sdp->sd_quota_inode);
831 fail_rindex:
832 	gfs2_clear_rgrpd(sdp);
833 	iput(sdp->sd_rindex);
834 fail_statfs:
835 	iput(sdp->sd_statfs_inode);
836 fail_journal:
837 	init_journal(sdp, UNDO);
838 fail:
839 	return error;
840 }
841 
842 static int init_per_node(struct gfs2_sbd *sdp, int undo)
843 {
844 	struct inode *pn = NULL;
845 	char buf[30];
846 	int error = 0;
847 	struct gfs2_inode *ip;
848 	struct inode *master = d_inode(sdp->sd_master_dir);
849 
850 	if (sdp->sd_args.ar_spectator)
851 		return 0;
852 
853 	if (undo)
854 		goto fail_qc_gh;
855 
856 	pn = gfs2_lookup_simple(master, "per_node");
857 	if (IS_ERR(pn)) {
858 		error = PTR_ERR(pn);
859 		fs_err(sdp, "can't find per_node directory: %d\n", error);
860 		return error;
861 	}
862 
863 	sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
864 	sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
865 	if (IS_ERR(sdp->sd_sc_inode)) {
866 		error = PTR_ERR(sdp->sd_sc_inode);
867 		fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
868 		goto fail;
869 	}
870 
871 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
872 	sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
873 	if (IS_ERR(sdp->sd_qc_inode)) {
874 		error = PTR_ERR(sdp->sd_qc_inode);
875 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
876 		goto fail_ut_i;
877 	}
878 
879 	iput(pn);
880 	pn = NULL;
881 
882 	ip = GFS2_I(sdp->sd_sc_inode);
883 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
884 				   &sdp->sd_sc_gh);
885 	if (error) {
886 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
887 		goto fail_qc_i;
888 	}
889 
890 	ip = GFS2_I(sdp->sd_qc_inode);
891 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
892 				   &sdp->sd_qc_gh);
893 	if (error) {
894 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
895 		goto fail_ut_gh;
896 	}
897 
898 	return 0;
899 
900 fail_qc_gh:
901 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
902 fail_ut_gh:
903 	gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
904 fail_qc_i:
905 	iput(sdp->sd_qc_inode);
906 fail_ut_i:
907 	iput(sdp->sd_sc_inode);
908 fail:
909 	iput(pn);
910 	return error;
911 }
912 
913 static const match_table_t nolock_tokens = {
914 	{ Opt_jid, "jid=%d\n", },
915 	{ Opt_err, NULL },
916 };
917 
918 static const struct lm_lockops nolock_ops = {
919 	.lm_proto_name = "lock_nolock",
920 	.lm_put_lock = gfs2_glock_free,
921 	.lm_tokens = &nolock_tokens,
922 };
923 
924 /**
925  * gfs2_lm_mount - mount a locking protocol
926  * @sdp: the filesystem
927  * @args: mount arguments
928  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
929  *
930  * Returns: errno
931  */
932 
933 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
934 {
935 	const struct lm_lockops *lm;
936 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
937 	struct gfs2_args *args = &sdp->sd_args;
938 	const char *proto = sdp->sd_proto_name;
939 	const char *table = sdp->sd_table_name;
940 	char *o, *options;
941 	int ret;
942 
943 	if (!strcmp("lock_nolock", proto)) {
944 		lm = &nolock_ops;
945 		sdp->sd_args.ar_localflocks = 1;
946 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
947 	} else if (!strcmp("lock_dlm", proto)) {
948 		lm = &gfs2_dlm_ops;
949 #endif
950 	} else {
951 		pr_info("can't find protocol %s\n", proto);
952 		return -ENOENT;
953 	}
954 
955 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
956 
957 	ls->ls_ops = lm;
958 	ls->ls_first = 1;
959 
960 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
961 		substring_t tmp[MAX_OPT_ARGS];
962 		int token, option;
963 
964 		if (!o || !*o)
965 			continue;
966 
967 		token = match_token(o, *lm->lm_tokens, tmp);
968 		switch (token) {
969 		case Opt_jid:
970 			ret = match_int(&tmp[0], &option);
971 			if (ret || option < 0)
972 				goto hostdata_error;
973 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
974 				ls->ls_jid = option;
975 			break;
976 		case Opt_id:
977 		case Opt_nodir:
978 			/* Obsolete, but left for backward compat purposes */
979 			break;
980 		case Opt_first:
981 			ret = match_int(&tmp[0], &option);
982 			if (ret || (option != 0 && option != 1))
983 				goto hostdata_error;
984 			ls->ls_first = option;
985 			break;
986 		case Opt_err:
987 		default:
988 hostdata_error:
989 			fs_info(sdp, "unknown hostdata (%s)\n", o);
990 			return -EINVAL;
991 		}
992 	}
993 
994 	if (lm->lm_mount == NULL) {
995 		fs_info(sdp, "Now mounting FS...\n");
996 		complete_all(&sdp->sd_locking_init);
997 		return 0;
998 	}
999 	ret = lm->lm_mount(sdp, table);
1000 	if (ret == 0)
1001 		fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1002 	complete_all(&sdp->sd_locking_init);
1003 	return ret;
1004 }
1005 
1006 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1007 {
1008 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1009 	if (likely(!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) &&
1010 	    lm->lm_unmount)
1011 		lm->lm_unmount(sdp);
1012 }
1013 
1014 static int wait_on_journal(struct gfs2_sbd *sdp)
1015 {
1016 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1017 		return 0;
1018 
1019 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1020 		? -EINTR : 0;
1021 }
1022 
1023 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1024 {
1025 	struct super_block *sb = sdp->sd_vfs;
1026 	char ro[20];
1027 	char spectator[20];
1028 	char *envp[] = { ro, spectator, NULL };
1029 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1030 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1031 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1032 }
1033 
1034 /**
1035  * gfs2_fill_super - Read in superblock
1036  * @sb: The VFS superblock
1037  * @args: Mount options
1038  * @silent: Don't complain if it's not a GFS2 filesystem
1039  *
1040  * Returns: -errno
1041  */
1042 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1043 {
1044 	struct gfs2_args *args = fc->fs_private;
1045 	int silent = fc->sb_flags & SB_SILENT;
1046 	struct gfs2_sbd *sdp;
1047 	struct gfs2_holder mount_gh;
1048 	int error;
1049 
1050 	sdp = init_sbd(sb);
1051 	if (!sdp) {
1052 		pr_warn("can't alloc struct gfs2_sbd\n");
1053 		return -ENOMEM;
1054 	}
1055 	sdp->sd_args = *args;
1056 
1057 	if (sdp->sd_args.ar_spectator) {
1058                 sb->s_flags |= SB_RDONLY;
1059 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1060 	}
1061 	if (sdp->sd_args.ar_posix_acl)
1062 		sb->s_flags |= SB_POSIXACL;
1063 	if (sdp->sd_args.ar_nobarrier)
1064 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1065 
1066 	sb->s_flags |= SB_NOSEC;
1067 	sb->s_magic = GFS2_MAGIC;
1068 	sb->s_op = &gfs2_super_ops;
1069 	sb->s_d_op = &gfs2_dops;
1070 	sb->s_export_op = &gfs2_export_ops;
1071 	sb->s_xattr = gfs2_xattr_handlers;
1072 	sb->s_qcop = &gfs2_quotactl_ops;
1073 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1074 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1075 	sb->s_time_gran = 1;
1076 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1077 
1078 	/* Set up the buffer cache and fill in some fake block size values
1079 	   to allow us to read-in the on-disk superblock. */
1080 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1081 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1082 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1083                                GFS2_BASIC_BLOCK_SHIFT;
1084 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1085 
1086 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1087 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1088 	if (sdp->sd_args.ar_statfs_quantum) {
1089 		sdp->sd_tune.gt_statfs_slow = 0;
1090 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1091 	} else {
1092 		sdp->sd_tune.gt_statfs_slow = 1;
1093 		sdp->sd_tune.gt_statfs_quantum = 30;
1094 	}
1095 
1096 	error = init_names(sdp, silent);
1097 	if (error) {
1098 		/* In this case, we haven't initialized sysfs, so we have to
1099 		   manually free the sdp. */
1100 		free_sbd(sdp);
1101 		sb->s_fs_info = NULL;
1102 		return error;
1103 	}
1104 
1105 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1106 
1107 	error = gfs2_sys_fs_add(sdp);
1108 	/*
1109 	 * If we hit an error here, gfs2_sys_fs_add will have called function
1110 	 * kobject_put which causes the sysfs usage count to go to zero, which
1111 	 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1112 	 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1113 	 * kobject_put to free sdp.
1114 	 */
1115 	if (error)
1116 		return error;
1117 
1118 	gfs2_create_debugfs_file(sdp);
1119 
1120 	error = gfs2_lm_mount(sdp, silent);
1121 	if (error)
1122 		goto fail_debug;
1123 
1124 	error = init_locking(sdp, &mount_gh, DO);
1125 	if (error)
1126 		goto fail_lm;
1127 
1128 	error = init_sb(sdp, silent);
1129 	if (error)
1130 		goto fail_locking;
1131 
1132 	error = wait_on_journal(sdp);
1133 	if (error)
1134 		goto fail_sb;
1135 
1136 	/*
1137 	 * If user space has failed to join the cluster or some similar
1138 	 * failure has occurred, then the journal id will contain a
1139 	 * negative (error) number. This will then be returned to the
1140 	 * caller (of the mount syscall). We do this even for spectator
1141 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1142 	 * the jid is unused in the spectator case)
1143 	 */
1144 	if (sdp->sd_lockstruct.ls_jid < 0) {
1145 		error = sdp->sd_lockstruct.ls_jid;
1146 		sdp->sd_lockstruct.ls_jid = 0;
1147 		goto fail_sb;
1148 	}
1149 
1150 	if (sdp->sd_args.ar_spectator)
1151 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1152 			 sdp->sd_table_name);
1153 	else
1154 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1155 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1156 
1157 	error = init_inodes(sdp, DO);
1158 	if (error)
1159 		goto fail_sb;
1160 
1161 	error = init_per_node(sdp, DO);
1162 	if (error)
1163 		goto fail_inodes;
1164 
1165 	error = gfs2_statfs_init(sdp);
1166 	if (error) {
1167 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1168 		goto fail_per_node;
1169 	}
1170 
1171 	if (!sb_rdonly(sb)) {
1172 		error = gfs2_make_fs_rw(sdp);
1173 		if (error) {
1174 			fs_err(sdp, "can't make FS RW: %d\n", error);
1175 			goto fail_per_node;
1176 		}
1177 	}
1178 
1179 	gfs2_glock_dq_uninit(&mount_gh);
1180 	gfs2_online_uevent(sdp);
1181 	return 0;
1182 
1183 fail_per_node:
1184 	init_per_node(sdp, UNDO);
1185 fail_inodes:
1186 	init_inodes(sdp, UNDO);
1187 fail_sb:
1188 	if (sdp->sd_root_dir)
1189 		dput(sdp->sd_root_dir);
1190 	if (sdp->sd_master_dir)
1191 		dput(sdp->sd_master_dir);
1192 	if (sb->s_root)
1193 		dput(sb->s_root);
1194 	sb->s_root = NULL;
1195 fail_locking:
1196 	init_locking(sdp, &mount_gh, UNDO);
1197 fail_lm:
1198 	complete_all(&sdp->sd_journal_ready);
1199 	gfs2_gl_hash_clear(sdp);
1200 	gfs2_lm_unmount(sdp);
1201 fail_debug:
1202 	gfs2_delete_debugfs_file(sdp);
1203 	/* gfs2_sys_fs_del must be the last thing we do, since it causes
1204 	 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1205 	gfs2_sys_fs_del(sdp);
1206 	sb->s_fs_info = NULL;
1207 	return error;
1208 }
1209 
1210 /**
1211  * gfs2_get_tree - Get the GFS2 superblock and root directory
1212  * @fc: The filesystem context
1213  *
1214  * Returns: 0 or -errno on error
1215  */
1216 static int gfs2_get_tree(struct fs_context *fc)
1217 {
1218 	struct gfs2_args *args = fc->fs_private;
1219 	struct gfs2_sbd *sdp;
1220 	int error;
1221 
1222 	error = get_tree_bdev(fc, gfs2_fill_super);
1223 	if (error)
1224 		return error;
1225 
1226 	sdp = fc->root->d_sb->s_fs_info;
1227 	dput(fc->root);
1228 	if (args->ar_meta)
1229 		fc->root = dget(sdp->sd_master_dir);
1230 	else
1231 		fc->root = dget(sdp->sd_root_dir);
1232 	return 0;
1233 }
1234 
1235 static void gfs2_fc_free(struct fs_context *fc)
1236 {
1237 	struct gfs2_args *args = fc->fs_private;
1238 
1239 	kfree(args);
1240 }
1241 
1242 enum gfs2_param {
1243 	Opt_lockproto,
1244 	Opt_locktable,
1245 	Opt_hostdata,
1246 	Opt_spectator,
1247 	Opt_ignore_local_fs,
1248 	Opt_localflocks,
1249 	Opt_localcaching,
1250 	Opt_debug,
1251 	Opt_upgrade,
1252 	Opt_acl,
1253 	Opt_quota,
1254 	Opt_suiddir,
1255 	Opt_data,
1256 	Opt_meta,
1257 	Opt_discard,
1258 	Opt_commit,
1259 	Opt_errors,
1260 	Opt_statfs_quantum,
1261 	Opt_statfs_percent,
1262 	Opt_quota_quantum,
1263 	Opt_barrier,
1264 	Opt_rgrplvb,
1265 	Opt_loccookie,
1266 };
1267 
1268 enum opt_quota {
1269 	Opt_quota_unset = 0,
1270 	Opt_quota_off,
1271 	Opt_quota_account,
1272 	Opt_quota_on,
1273 };
1274 
1275 static const unsigned int opt_quota_values[] = {
1276 	[Opt_quota_off]     = GFS2_QUOTA_OFF,
1277 	[Opt_quota_account] = GFS2_QUOTA_ACCOUNT,
1278 	[Opt_quota_on]      = GFS2_QUOTA_ON,
1279 };
1280 
1281 enum opt_data {
1282 	Opt_data_writeback = GFS2_DATA_WRITEBACK,
1283 	Opt_data_ordered   = GFS2_DATA_ORDERED,
1284 };
1285 
1286 enum opt_errors {
1287 	Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1288 	Opt_errors_panic    = GFS2_ERRORS_PANIC,
1289 };
1290 
1291 static const struct fs_parameter_spec gfs2_param_specs[] = {
1292 	fsparam_string ("lockproto",          Opt_lockproto),
1293 	fsparam_string ("locktable",          Opt_locktable),
1294 	fsparam_string ("hostdata",           Opt_hostdata),
1295 	fsparam_flag   ("spectator",          Opt_spectator),
1296 	fsparam_flag   ("norecovery",         Opt_spectator),
1297 	fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1298 	fsparam_flag   ("localflocks",        Opt_localflocks),
1299 	fsparam_flag   ("localcaching",       Opt_localcaching),
1300 	fsparam_flag_no("debug",              Opt_debug),
1301 	fsparam_flag   ("upgrade",            Opt_upgrade),
1302 	fsparam_flag_no("acl",                Opt_acl),
1303 	fsparam_flag_no("suiddir",            Opt_suiddir),
1304 	fsparam_enum   ("data",               Opt_data),
1305 	fsparam_flag   ("meta",               Opt_meta),
1306 	fsparam_flag_no("discard",            Opt_discard),
1307 	fsparam_s32    ("commit",             Opt_commit),
1308 	fsparam_enum   ("errors",             Opt_errors),
1309 	fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1310 	fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1311 	fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1312 	fsparam_flag_no("barrier",            Opt_barrier),
1313 	fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1314 	fsparam_flag_no("loccookie",          Opt_loccookie),
1315 	/* quota can be a flag or an enum so it gets special treatment */
1316 	__fsparam(fs_param_is_enum, "quota", Opt_quota, fs_param_neg_with_no|fs_param_v_optional),
1317 	{}
1318 };
1319 
1320 static const struct fs_parameter_enum gfs2_param_enums[] = {
1321 	{ Opt_quota,    "off",        Opt_quota_off },
1322 	{ Opt_quota,    "account",    Opt_quota_account },
1323 	{ Opt_quota,    "on",         Opt_quota_on },
1324 	{ Opt_data,     "writeback",  Opt_data_writeback },
1325 	{ Opt_data,     "ordered",    Opt_data_ordered },
1326 	{ Opt_errors,   "withdraw",   Opt_errors_withdraw },
1327 	{ Opt_errors,   "panic",      Opt_errors_panic },
1328 	{}
1329 };
1330 
1331 const struct fs_parameter_description gfs2_fs_parameters = {
1332 	.name = "gfs2",
1333 	.specs = gfs2_param_specs,
1334 	.enums = gfs2_param_enums,
1335 };
1336 
1337 /* Parse a single mount parameter */
1338 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1339 {
1340 	struct gfs2_args *args = fc->fs_private;
1341 	struct fs_parse_result result;
1342 	int o;
1343 
1344 	o = fs_parse(fc, &gfs2_fs_parameters, param, &result);
1345 	if (o < 0)
1346 		return o;
1347 
1348 	switch (o) {
1349 	case Opt_lockproto:
1350 		strlcpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1351 		break;
1352 	case Opt_locktable:
1353 		strlcpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1354 		break;
1355 	case Opt_hostdata:
1356 		strlcpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1357 		break;
1358 	case Opt_spectator:
1359 		args->ar_spectator = 1;
1360 		break;
1361 	case Opt_ignore_local_fs:
1362 		/* Retained for backwards compat only */
1363 		break;
1364 	case Opt_localflocks:
1365 		args->ar_localflocks = 1;
1366 		break;
1367 	case Opt_localcaching:
1368 		/* Retained for backwards compat only */
1369 		break;
1370 	case Opt_debug:
1371 		if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1372 			return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1373 		args->ar_debug = result.boolean;
1374 		break;
1375 	case Opt_upgrade:
1376 		/* Retained for backwards compat only */
1377 		break;
1378 	case Opt_acl:
1379 		args->ar_posix_acl = result.boolean;
1380 		break;
1381 	case Opt_quota:
1382 		/* The quota option can be a flag or an enum. A non-zero int_32
1383 		   result means that we have an enum index. Otherwise we have
1384 		   to rely on the 'negated' flag to tell us whether 'quota' or
1385 		   'noquota' was specified. */
1386 		if (result.negated)
1387 			args->ar_quota = GFS2_QUOTA_OFF;
1388 		else if (result.int_32 > 0)
1389 			args->ar_quota = opt_quota_values[result.int_32];
1390 		else
1391 			args->ar_quota = GFS2_QUOTA_ON;
1392 		break;
1393 	case Opt_suiddir:
1394 		args->ar_suiddir = result.boolean;
1395 		break;
1396 	case Opt_data:
1397 		/* The uint_32 result maps directly to GFS2_DATA_* */
1398 		args->ar_data = result.uint_32;
1399 		break;
1400 	case Opt_meta:
1401 		args->ar_meta = 1;
1402 		break;
1403 	case Opt_discard:
1404 		args->ar_discard = result.boolean;
1405 		break;
1406 	case Opt_commit:
1407 		if (result.int_32 <= 0)
1408 			return invalf(fc, "gfs2: commit mount option requires a positive numeric argument");
1409 		args->ar_commit = result.int_32;
1410 		break;
1411 	case Opt_statfs_quantum:
1412 		if (result.int_32 < 0)
1413 			return invalf(fc, "gfs2: statfs_quantum mount option requires a non-negative numeric argument");
1414 		args->ar_statfs_quantum = result.int_32;
1415 		break;
1416 	case Opt_quota_quantum:
1417 		if (result.int_32 <= 0)
1418 			return invalf(fc, "gfs2: quota_quantum mount option requires a positive numeric argument");
1419 		args->ar_quota_quantum = result.int_32;
1420 		break;
1421 	case Opt_statfs_percent:
1422 		if (result.int_32 < 0 || result.int_32 > 100)
1423 			return invalf(fc, "gfs2: statfs_percent mount option requires a numeric argument between 0 and 100");
1424 		args->ar_statfs_percent = result.int_32;
1425 		break;
1426 	case Opt_errors:
1427 		if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1428 			return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1429 		args->ar_errors = result.uint_32;
1430 		break;
1431 	case Opt_barrier:
1432 		args->ar_nobarrier = result.boolean;
1433 		break;
1434 	case Opt_rgrplvb:
1435 		args->ar_rgrplvb = result.boolean;
1436 		break;
1437 	case Opt_loccookie:
1438 		args->ar_loccookie = result.boolean;
1439 		break;
1440 	default:
1441 		return invalf(fc, "gfs2: invalid mount option: %s", param->key);
1442 	}
1443 	return 0;
1444 }
1445 
1446 static int gfs2_reconfigure(struct fs_context *fc)
1447 {
1448 	struct super_block *sb = fc->root->d_sb;
1449 	struct gfs2_sbd *sdp = sb->s_fs_info;
1450 	struct gfs2_args *oldargs = &sdp->sd_args;
1451 	struct gfs2_args *newargs = fc->fs_private;
1452 	struct gfs2_tune *gt = &sdp->sd_tune;
1453 	int error = 0;
1454 
1455 	sync_filesystem(sb);
1456 
1457 	spin_lock(&gt->gt_spin);
1458 	oldargs->ar_commit = gt->gt_logd_secs;
1459 	oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1460 	if (gt->gt_statfs_slow)
1461 		oldargs->ar_statfs_quantum = 0;
1462 	else
1463 		oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1464 	spin_unlock(&gt->gt_spin);
1465 
1466 	if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1467 		errorf(fc, "gfs2: reconfiguration of locking protocol not allowed");
1468 		return -EINVAL;
1469 	}
1470 	if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1471 		errorf(fc, "gfs2: reconfiguration of lock table not allowed");
1472 		return -EINVAL;
1473 	}
1474 	if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1475 		errorf(fc, "gfs2: reconfiguration of host data not allowed");
1476 		return -EINVAL;
1477 	}
1478 	if (newargs->ar_spectator != oldargs->ar_spectator) {
1479 		errorf(fc, "gfs2: reconfiguration of spectator mode not allowed");
1480 		return -EINVAL;
1481 	}
1482 	if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1483 		errorf(fc, "gfs2: reconfiguration of localflocks not allowed");
1484 		return -EINVAL;
1485 	}
1486 	if (newargs->ar_meta != oldargs->ar_meta) {
1487 		errorf(fc, "gfs2: switching between gfs2 and gfs2meta not allowed");
1488 		return -EINVAL;
1489 	}
1490 	if (oldargs->ar_spectator)
1491 		fc->sb_flags |= SB_RDONLY;
1492 
1493 	if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1494 		if (fc->sb_flags & SB_RDONLY) {
1495 			error = gfs2_make_fs_ro(sdp);
1496 			if (error)
1497 				errorf(fc, "gfs2: unable to remount read-only");
1498 		} else {
1499 			error = gfs2_make_fs_rw(sdp);
1500 			if (error)
1501 				errorf(fc, "gfs2: unable to remount read-write");
1502 		}
1503 	}
1504 	sdp->sd_args = *newargs;
1505 
1506 	if (sdp->sd_args.ar_posix_acl)
1507 		sb->s_flags |= SB_POSIXACL;
1508 	else
1509 		sb->s_flags &= ~SB_POSIXACL;
1510 	if (sdp->sd_args.ar_nobarrier)
1511 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1512 	else
1513 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1514 	spin_lock(&gt->gt_spin);
1515 	gt->gt_logd_secs = newargs->ar_commit;
1516 	gt->gt_quota_quantum = newargs->ar_quota_quantum;
1517 	if (newargs->ar_statfs_quantum) {
1518 		gt->gt_statfs_slow = 0;
1519 		gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1520 	}
1521 	else {
1522 		gt->gt_statfs_slow = 1;
1523 		gt->gt_statfs_quantum = 30;
1524 	}
1525 	spin_unlock(&gt->gt_spin);
1526 
1527 	gfs2_online_uevent(sdp);
1528 	return error;
1529 }
1530 
1531 static const struct fs_context_operations gfs2_context_ops = {
1532 	.free        = gfs2_fc_free,
1533 	.parse_param = gfs2_parse_param,
1534 	.get_tree    = gfs2_get_tree,
1535 	.reconfigure = gfs2_reconfigure,
1536 };
1537 
1538 /* Set up the filesystem mount context */
1539 static int gfs2_init_fs_context(struct fs_context *fc)
1540 {
1541 	struct gfs2_args *args;
1542 
1543 	args = kmalloc(sizeof(*args), GFP_KERNEL);
1544 	if (args == NULL)
1545 		return -ENOMEM;
1546 
1547 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1548 		struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1549 
1550 		*args = sdp->sd_args;
1551 	} else {
1552 		memset(args, 0, sizeof(*args));
1553 		args->ar_quota = GFS2_QUOTA_DEFAULT;
1554 		args->ar_data = GFS2_DATA_DEFAULT;
1555 		args->ar_commit = 30;
1556 		args->ar_statfs_quantum = 30;
1557 		args->ar_quota_quantum = 60;
1558 		args->ar_errors = GFS2_ERRORS_DEFAULT;
1559 	}
1560 	fc->fs_private = args;
1561 	fc->ops = &gfs2_context_ops;
1562 	return 0;
1563 }
1564 
1565 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1566 {
1567 	return -EINVAL;
1568 }
1569 
1570 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1571 {
1572 	return (fc->sget_key == s->s_bdev);
1573 }
1574 
1575 static int gfs2_meta_get_tree(struct fs_context *fc)
1576 {
1577 	struct super_block *s;
1578 	struct gfs2_sbd *sdp;
1579 	struct path path;
1580 	int error;
1581 
1582 	if (!fc->source || !*fc->source)
1583 		return -EINVAL;
1584 
1585 	error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1586 	if (error) {
1587 		pr_warn("path_lookup on %s returned error %d\n",
1588 		        fc->source, error);
1589 		return error;
1590 	}
1591 	fc->fs_type = &gfs2_fs_type;
1592 	fc->sget_key = path.dentry->d_sb->s_bdev;
1593 	s = sget_fc(fc, test_meta_super, set_meta_super);
1594 	path_put(&path);
1595 	if (IS_ERR(s)) {
1596 		pr_warn("gfs2 mount does not exist\n");
1597 		return PTR_ERR(s);
1598 	}
1599 	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1600 		deactivate_locked_super(s);
1601 		return -EBUSY;
1602 	}
1603 	sdp = s->s_fs_info;
1604 	fc->root = dget(sdp->sd_master_dir);
1605 	return 0;
1606 }
1607 
1608 static const struct fs_context_operations gfs2_meta_context_ops = {
1609 	.free        = gfs2_fc_free,
1610 	.get_tree    = gfs2_meta_get_tree,
1611 };
1612 
1613 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1614 {
1615 	int ret = gfs2_init_fs_context(fc);
1616 
1617 	if (ret)
1618 		return ret;
1619 
1620 	fc->ops = &gfs2_meta_context_ops;
1621 	return 0;
1622 }
1623 
1624 static void gfs2_kill_sb(struct super_block *sb)
1625 {
1626 	struct gfs2_sbd *sdp = sb->s_fs_info;
1627 
1628 	if (sdp == NULL) {
1629 		kill_block_super(sb);
1630 		return;
1631 	}
1632 
1633 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1634 	dput(sdp->sd_root_dir);
1635 	dput(sdp->sd_master_dir);
1636 	sdp->sd_root_dir = NULL;
1637 	sdp->sd_master_dir = NULL;
1638 	shrink_dcache_sb(sb);
1639 	kill_block_super(sb);
1640 }
1641 
1642 struct file_system_type gfs2_fs_type = {
1643 	.name = "gfs2",
1644 	.fs_flags = FS_REQUIRES_DEV,
1645 	.init_fs_context = gfs2_init_fs_context,
1646 	.parameters = &gfs2_fs_parameters,
1647 	.kill_sb = gfs2_kill_sb,
1648 	.owner = THIS_MODULE,
1649 };
1650 MODULE_ALIAS_FS("gfs2");
1651 
1652 struct file_system_type gfs2meta_fs_type = {
1653 	.name = "gfs2meta",
1654 	.fs_flags = FS_REQUIRES_DEV,
1655 	.init_fs_context = gfs2_meta_init_fs_context,
1656 	.owner = THIS_MODULE,
1657 };
1658 MODULE_ALIAS_FS("gfs2meta");
1659