• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
gfs2_tune_init(struct gfs2_tune * gt)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 
free_sbd(struct gfs2_sbd * sdp)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 
init_sbd(struct super_block * sb)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 
gfs2_check_sb(struct gfs2_sbd * sdp,int silent)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 (sb->sb_fs_format != GFS2_FORMAT_FS ||
173 	    sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
174 		fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
175 		return -EINVAL;
176 	}
177 
178 	if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
179 	    (sb->sb_bsize & (sb->sb_bsize - 1))) {
180 		pr_warn("Invalid superblock size\n");
181 		return -EINVAL;
182 	}
183 	if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
184 		pr_warn("Invalid block size shift\n");
185 		return -EINVAL;
186 	}
187 	return 0;
188 }
189 
end_bio_io_page(struct bio * bio)190 static void end_bio_io_page(struct bio *bio)
191 {
192 	struct page *page = bio->bi_private;
193 
194 	if (!bio->bi_status)
195 		SetPageUptodate(page);
196 	else
197 		pr_warn("error %d reading superblock\n", bio->bi_status);
198 	unlock_page(page);
199 }
200 
gfs2_sb_in(struct gfs2_sbd * sdp,const void * buf)201 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
202 {
203 	struct gfs2_sb_host *sb = &sdp->sd_sb;
204 	struct super_block *s = sdp->sd_vfs;
205 	const struct gfs2_sb *str = buf;
206 
207 	sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
208 	sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
209 	sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
210 	sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
211 	sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
212 	sb->sb_bsize = be32_to_cpu(str->sb_bsize);
213 	sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
214 	sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
215 	sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
216 	sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
217 	sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
218 
219 	memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
220 	memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
221 	memcpy(&s->s_uuid, str->sb_uuid, 16);
222 }
223 
224 /**
225  * gfs2_read_super - Read the gfs2 super block from disk
226  * @sdp: The GFS2 super block
227  * @sector: The location of the super block
228  * @error: The error code to return
229  *
230  * This uses the bio functions to read the super block from disk
231  * because we want to be 100% sure that we never read cached data.
232  * A super block is read twice only during each GFS2 mount and is
233  * never written to by the filesystem. The first time its read no
234  * locks are held, and the only details which are looked at are those
235  * relating to the locking protocol. Once locking is up and working,
236  * the sb is read again under the lock to establish the location of
237  * the master directory (contains pointers to journals etc) and the
238  * root directory.
239  *
240  * Returns: 0 on success or error
241  */
242 
gfs2_read_super(struct gfs2_sbd * sdp,sector_t sector,int silent)243 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
244 {
245 	struct super_block *sb = sdp->sd_vfs;
246 	struct gfs2_sb *p;
247 	struct page *page;
248 	struct bio *bio;
249 
250 	page = alloc_page(GFP_NOFS);
251 	if (unlikely(!page))
252 		return -ENOMEM;
253 
254 	ClearPageUptodate(page);
255 	ClearPageDirty(page);
256 	lock_page(page);
257 
258 	bio = bio_alloc(GFP_NOFS, 1);
259 	bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
260 	bio_set_dev(bio, sb->s_bdev);
261 	bio_add_page(bio, page, PAGE_SIZE, 0);
262 
263 	bio->bi_end_io = end_bio_io_page;
264 	bio->bi_private = page;
265 	bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
266 	submit_bio(bio);
267 	wait_on_page_locked(page);
268 	bio_put(bio);
269 	if (!PageUptodate(page)) {
270 		__free_page(page);
271 		return -EIO;
272 	}
273 	p = kmap(page);
274 	gfs2_sb_in(sdp, p);
275 	kunmap(page);
276 	__free_page(page);
277 	return gfs2_check_sb(sdp, silent);
278 }
279 
280 /**
281  * gfs2_read_sb - Read super block
282  * @sdp: The GFS2 superblock
283  * @silent: Don't print message if mount fails
284  *
285  */
286 
gfs2_read_sb(struct gfs2_sbd * sdp,int silent)287 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
288 {
289 	u32 hash_blocks, ind_blocks, leaf_blocks;
290 	u32 tmp_blocks;
291 	unsigned int x;
292 	int error;
293 
294 	error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
295 	if (error) {
296 		if (!silent)
297 			fs_err(sdp, "can't read superblock\n");
298 		return error;
299 	}
300 
301 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
302 			       GFS2_BASIC_BLOCK_SHIFT;
303 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
304 	sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
305 			  sizeof(struct gfs2_dinode)) / sizeof(u64);
306 	sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
307 			  sizeof(struct gfs2_meta_header)) / sizeof(u64);
308 	sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
309 	sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
310 	sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
311 	sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
312 	sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
313 				sizeof(struct gfs2_meta_header)) /
314 			        sizeof(struct gfs2_quota_change);
315 	sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
316 				     sizeof(struct gfs2_meta_header))
317 		* GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
318 
319 	/* Compute maximum reservation required to add a entry to a directory */
320 
321 	hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
322 			     sdp->sd_jbsize);
323 
324 	ind_blocks = 0;
325 	for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
326 		tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
327 		ind_blocks += tmp_blocks;
328 	}
329 
330 	leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
331 
332 	sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
333 
334 	sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
335 				sizeof(struct gfs2_dinode);
336 	sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
337 	for (x = 2;; x++) {
338 		u64 space, d;
339 		u32 m;
340 
341 		space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
342 		d = space;
343 		m = do_div(d, sdp->sd_inptrs);
344 
345 		if (d != sdp->sd_heightsize[x - 1] || m)
346 			break;
347 		sdp->sd_heightsize[x] = space;
348 	}
349 	sdp->sd_max_height = x;
350 	sdp->sd_heightsize[x] = ~0;
351 	gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
352 
353 	sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
354 				      sizeof(struct gfs2_leaf)) /
355 				     GFS2_MIN_DIRENT_SIZE;
356 	return 0;
357 }
358 
init_names(struct gfs2_sbd * sdp,int silent)359 static int init_names(struct gfs2_sbd *sdp, int silent)
360 {
361 	char *proto, *table;
362 	int error = 0;
363 
364 	proto = sdp->sd_args.ar_lockproto;
365 	table = sdp->sd_args.ar_locktable;
366 
367 	/*  Try to autodetect  */
368 
369 	if (!proto[0] || !table[0]) {
370 		error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
371 		if (error)
372 			return error;
373 
374 		if (!proto[0])
375 			proto = sdp->sd_sb.sb_lockproto;
376 		if (!table[0])
377 			table = sdp->sd_sb.sb_locktable;
378 	}
379 
380 	if (!table[0])
381 		table = sdp->sd_vfs->s_id;
382 
383 	BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
384 
385 	strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
386 	strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
387 
388 	table = sdp->sd_table_name;
389 	while ((table = strchr(table, '/')))
390 		*table = '_';
391 
392 	return error;
393 }
394 
init_locking(struct gfs2_sbd * sdp,struct gfs2_holder * mount_gh,int undo)395 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
396 			int undo)
397 {
398 	int error = 0;
399 
400 	if (undo)
401 		goto fail_trans;
402 
403 	error = gfs2_glock_nq_num(sdp,
404 				  GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
405 				  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
406 				  mount_gh);
407 	if (error) {
408 		fs_err(sdp, "can't acquire mount glock: %d\n", error);
409 		goto fail;
410 	}
411 
412 	error = gfs2_glock_nq_num(sdp,
413 				  GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
414 				  LM_ST_SHARED,
415 				  LM_FLAG_NOEXP | GL_EXACT,
416 				  &sdp->sd_live_gh);
417 	if (error) {
418 		fs_err(sdp, "can't acquire live glock: %d\n", error);
419 		goto fail_mount;
420 	}
421 
422 	error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
423 			       CREATE, &sdp->sd_rename_gl);
424 	if (error) {
425 		fs_err(sdp, "can't create rename glock: %d\n", error);
426 		goto fail_live;
427 	}
428 
429 	error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
430 			       CREATE, &sdp->sd_freeze_gl);
431 	if (error) {
432 		fs_err(sdp, "can't create transaction glock: %d\n", error);
433 		goto fail_rename;
434 	}
435 
436 	return 0;
437 
438 fail_trans:
439 	gfs2_glock_put(sdp->sd_freeze_gl);
440 fail_rename:
441 	gfs2_glock_put(sdp->sd_rename_gl);
442 fail_live:
443 	gfs2_glock_dq_uninit(&sdp->sd_live_gh);
444 fail_mount:
445 	gfs2_glock_dq_uninit(mount_gh);
446 fail:
447 	return error;
448 }
449 
gfs2_lookup_root(struct super_block * sb,struct dentry ** dptr,u64 no_addr,const char * name)450 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
451 			    u64 no_addr, const char *name)
452 {
453 	struct gfs2_sbd *sdp = sb->s_fs_info;
454 	struct dentry *dentry;
455 	struct inode *inode;
456 
457 	inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
458 				  GFS2_BLKST_FREE /* ignore */);
459 	if (IS_ERR(inode)) {
460 		fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
461 		return PTR_ERR(inode);
462 	}
463 	dentry = d_make_root(inode);
464 	if (!dentry) {
465 		fs_err(sdp, "can't alloc %s dentry\n", name);
466 		return -ENOMEM;
467 	}
468 	*dptr = dentry;
469 	return 0;
470 }
471 
init_sb(struct gfs2_sbd * sdp,int silent)472 static int init_sb(struct gfs2_sbd *sdp, int silent)
473 {
474 	struct super_block *sb = sdp->sd_vfs;
475 	struct gfs2_holder sb_gh;
476 	u64 no_addr;
477 	int ret;
478 
479 	ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
480 				LM_ST_SHARED, 0, &sb_gh);
481 	if (ret) {
482 		fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
483 		return ret;
484 	}
485 
486 	ret = gfs2_read_sb(sdp, silent);
487 	if (ret) {
488 		fs_err(sdp, "can't read superblock: %d\n", ret);
489 		goto out;
490 	}
491 
492 	/* Set up the buffer cache and SB for real */
493 	if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
494 		ret = -EINVAL;
495 		fs_err(sdp, "FS block size (%u) is too small for device "
496 		       "block size (%u)\n",
497 		       sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
498 		goto out;
499 	}
500 	if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
501 		ret = -EINVAL;
502 		fs_err(sdp, "FS block size (%u) is too big for machine "
503 		       "page size (%u)\n",
504 		       sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
505 		goto out;
506 	}
507 	sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
508 
509 	/* Get the root inode */
510 	no_addr = sdp->sd_sb.sb_root_dir.no_addr;
511 	ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
512 	if (ret)
513 		goto out;
514 
515 	/* Get the master inode */
516 	no_addr = sdp->sd_sb.sb_master_dir.no_addr;
517 	ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
518 	if (ret) {
519 		dput(sdp->sd_root_dir);
520 		goto out;
521 	}
522 	sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
523 out:
524 	gfs2_glock_dq_uninit(&sb_gh);
525 	return ret;
526 }
527 
gfs2_others_may_mount(struct gfs2_sbd * sdp)528 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
529 {
530 	char *message = "FIRSTMOUNT=Done";
531 	char *envp[] = { message, NULL };
532 
533 	fs_info(sdp, "first mount done, others may mount\n");
534 
535 	if (sdp->sd_lockstruct.ls_ops->lm_first_done)
536 		sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
537 
538 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
539 }
540 
541 /**
542  * gfs2_jindex_hold - Grab a lock on the jindex
543  * @sdp: The GFS2 superblock
544  * @ji_gh: the holder for the jindex glock
545  *
546  * Returns: errno
547  */
548 
gfs2_jindex_hold(struct gfs2_sbd * sdp,struct gfs2_holder * ji_gh)549 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
550 {
551 	struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
552 	struct qstr name;
553 	char buf[20];
554 	struct gfs2_jdesc *jd;
555 	int error;
556 
557 	name.name = buf;
558 
559 	mutex_lock(&sdp->sd_jindex_mutex);
560 
561 	for (;;) {
562 		error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
563 		if (error)
564 			break;
565 
566 		name.len = sprintf(buf, "journal%u", sdp->sd_journals);
567 		name.hash = gfs2_disk_hash(name.name, name.len);
568 
569 		error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
570 		if (error == -ENOENT) {
571 			error = 0;
572 			break;
573 		}
574 
575 		gfs2_glock_dq_uninit(ji_gh);
576 
577 		if (error)
578 			break;
579 
580 		error = -ENOMEM;
581 		jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
582 		if (!jd)
583 			break;
584 
585 		INIT_LIST_HEAD(&jd->extent_list);
586 		INIT_LIST_HEAD(&jd->jd_revoke_list);
587 
588 		INIT_WORK(&jd->jd_work, gfs2_recover_func);
589 		jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
590 		if (IS_ERR_OR_NULL(jd->jd_inode)) {
591 			if (!jd->jd_inode)
592 				error = -ENOENT;
593 			else
594 				error = PTR_ERR(jd->jd_inode);
595 			kfree(jd);
596 			break;
597 		}
598 
599 		spin_lock(&sdp->sd_jindex_spin);
600 		jd->jd_jid = sdp->sd_journals++;
601 		list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
602 		spin_unlock(&sdp->sd_jindex_spin);
603 	}
604 
605 	mutex_unlock(&sdp->sd_jindex_mutex);
606 
607 	return error;
608 }
609 
610 /**
611  * check_journal_clean - Make sure a journal is clean for a spectator mount
612  * @sdp: The GFS2 superblock
613  * @jd: The journal descriptor
614  *
615  * Returns: 0 if the journal is clean or locked, else an error
616  */
check_journal_clean(struct gfs2_sbd * sdp,struct gfs2_jdesc * jd)617 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
618 {
619 	int error;
620 	struct gfs2_holder j_gh;
621 	struct gfs2_log_header_host head;
622 	struct gfs2_inode *ip;
623 
624 	ip = GFS2_I(jd->jd_inode);
625 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
626 				   GL_EXACT | GL_NOCACHE, &j_gh);
627 	if (error) {
628 		fs_err(sdp, "Error locking journal for spectator mount.\n");
629 		return -EPERM;
630 	}
631 	error = gfs2_jdesc_check(jd);
632 	if (error) {
633 		fs_err(sdp, "Error checking journal for spectator mount.\n");
634 		goto out_unlock;
635 	}
636 	error = gfs2_find_jhead(jd, &head, false);
637 	if (error) {
638 		fs_err(sdp, "Error parsing journal for spectator mount.\n");
639 		goto out_unlock;
640 	}
641 	if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
642 		error = -EPERM;
643 		fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
644 		       "must not be a spectator.\n", jd->jd_jid);
645 	}
646 
647 out_unlock:
648 	gfs2_glock_dq_uninit(&j_gh);
649 	return error;
650 }
651 
init_journal(struct gfs2_sbd * sdp,int undo)652 static int init_journal(struct gfs2_sbd *sdp, int undo)
653 {
654 	struct inode *master = d_inode(sdp->sd_master_dir);
655 	struct gfs2_holder ji_gh;
656 	struct gfs2_inode *ip;
657 	int jindex = 1;
658 	int error = 0;
659 
660 	if (undo) {
661 		jindex = 0;
662 		goto fail_jinode_gh;
663 	}
664 
665 	sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
666 	if (IS_ERR(sdp->sd_jindex)) {
667 		fs_err(sdp, "can't lookup journal index: %d\n", error);
668 		return PTR_ERR(sdp->sd_jindex);
669 	}
670 
671 	/* Load in the journal index special file */
672 
673 	error = gfs2_jindex_hold(sdp, &ji_gh);
674 	if (error) {
675 		fs_err(sdp, "can't read journal index: %d\n", error);
676 		goto fail;
677 	}
678 
679 	error = -EUSERS;
680 	if (!gfs2_jindex_size(sdp)) {
681 		fs_err(sdp, "no journals!\n");
682 		goto fail_jindex;
683 	}
684 
685 	atomic_set(&sdp->sd_log_blks_needed, 0);
686 	if (sdp->sd_args.ar_spectator) {
687 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
688 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
689 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
690 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
691 	} else {
692 		if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
693 			fs_err(sdp, "can't mount journal #%u\n",
694 			       sdp->sd_lockstruct.ls_jid);
695 			fs_err(sdp, "there are only %u journals (0 - %u)\n",
696 			       gfs2_jindex_size(sdp),
697 			       gfs2_jindex_size(sdp) - 1);
698 			goto fail_jindex;
699 		}
700 		sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
701 
702 		error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
703 					  &gfs2_journal_glops,
704 					  LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
705 					  &sdp->sd_journal_gh);
706 		if (error) {
707 			fs_err(sdp, "can't acquire journal glock: %d\n", error);
708 			goto fail_jindex;
709 		}
710 
711 		ip = GFS2_I(sdp->sd_jdesc->jd_inode);
712 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
713 					   LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
714 					   &sdp->sd_jinode_gh);
715 		if (error) {
716 			fs_err(sdp, "can't acquire journal inode glock: %d\n",
717 			       error);
718 			goto fail_journal_gh;
719 		}
720 
721 		error = gfs2_jdesc_check(sdp->sd_jdesc);
722 		if (error) {
723 			fs_err(sdp, "my journal (%u) is bad: %d\n",
724 			       sdp->sd_jdesc->jd_jid, error);
725 			goto fail_jinode_gh;
726 		}
727 		atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
728 		atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
729 		atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
730 
731 		/* Map the extents for this journal's blocks */
732 		gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
733 	}
734 	trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
735 
736 	if (sdp->sd_lockstruct.ls_first) {
737 		unsigned int x;
738 		for (x = 0; x < sdp->sd_journals; x++) {
739 			struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
740 
741 			if (sdp->sd_args.ar_spectator) {
742 				error = check_journal_clean(sdp, jd);
743 				if (error)
744 					goto fail_jinode_gh;
745 				continue;
746 			}
747 			error = gfs2_recover_journal(jd, true);
748 			if (error) {
749 				fs_err(sdp, "error recovering journal %u: %d\n",
750 				       x, error);
751 				goto fail_jinode_gh;
752 			}
753 		}
754 
755 		gfs2_others_may_mount(sdp);
756 	} else if (!sdp->sd_args.ar_spectator) {
757 		error = gfs2_recover_journal(sdp->sd_jdesc, true);
758 		if (error) {
759 			fs_err(sdp, "error recovering my journal: %d\n", error);
760 			goto fail_jinode_gh;
761 		}
762 	}
763 
764 	sdp->sd_log_idle = 1;
765 	set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
766 	gfs2_glock_dq_uninit(&ji_gh);
767 	jindex = 0;
768 	INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
769 	return 0;
770 
771 fail_jinode_gh:
772 	if (!sdp->sd_args.ar_spectator)
773 		gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
774 fail_journal_gh:
775 	if (!sdp->sd_args.ar_spectator)
776 		gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
777 fail_jindex:
778 	gfs2_jindex_free(sdp);
779 	if (jindex)
780 		gfs2_glock_dq_uninit(&ji_gh);
781 fail:
782 	iput(sdp->sd_jindex);
783 	return error;
784 }
785 
786 static struct lock_class_key gfs2_quota_imutex_key;
787 
init_inodes(struct gfs2_sbd * sdp,int undo)788 static int init_inodes(struct gfs2_sbd *sdp, int undo)
789 {
790 	int error = 0;
791 	struct inode *master = d_inode(sdp->sd_master_dir);
792 
793 	if (undo)
794 		goto fail_qinode;
795 
796 	error = init_journal(sdp, undo);
797 	complete_all(&sdp->sd_journal_ready);
798 	if (error)
799 		goto fail;
800 
801 	/* Read in the master statfs inode */
802 	sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
803 	if (IS_ERR(sdp->sd_statfs_inode)) {
804 		error = PTR_ERR(sdp->sd_statfs_inode);
805 		fs_err(sdp, "can't read in statfs inode: %d\n", error);
806 		goto fail_journal;
807 	}
808 
809 	/* Read in the resource index inode */
810 	sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
811 	if (IS_ERR(sdp->sd_rindex)) {
812 		error = PTR_ERR(sdp->sd_rindex);
813 		fs_err(sdp, "can't get resource index inode: %d\n", error);
814 		goto fail_statfs;
815 	}
816 	sdp->sd_rindex_uptodate = 0;
817 
818 	/* Read in the quota inode */
819 	sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
820 	if (IS_ERR(sdp->sd_quota_inode)) {
821 		error = PTR_ERR(sdp->sd_quota_inode);
822 		fs_err(sdp, "can't get quota file inode: %d\n", error);
823 		goto fail_rindex;
824 	}
825 	/*
826 	 * i_rwsem on quota files is special. Since this inode is hidden system
827 	 * file, we are safe to define locking ourselves.
828 	 */
829 	lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
830 			  &gfs2_quota_imutex_key);
831 
832 	error = gfs2_rindex_update(sdp);
833 	if (error)
834 		goto fail_qinode;
835 
836 	return 0;
837 
838 fail_qinode:
839 	iput(sdp->sd_quota_inode);
840 fail_rindex:
841 	gfs2_clear_rgrpd(sdp);
842 	iput(sdp->sd_rindex);
843 fail_statfs:
844 	iput(sdp->sd_statfs_inode);
845 fail_journal:
846 	init_journal(sdp, UNDO);
847 fail:
848 	return error;
849 }
850 
init_per_node(struct gfs2_sbd * sdp,int undo)851 static int init_per_node(struct gfs2_sbd *sdp, int undo)
852 {
853 	struct inode *pn = NULL;
854 	char buf[30];
855 	int error = 0;
856 	struct gfs2_inode *ip;
857 	struct inode *master = d_inode(sdp->sd_master_dir);
858 
859 	if (sdp->sd_args.ar_spectator)
860 		return 0;
861 
862 	if (undo)
863 		goto fail_qc_gh;
864 
865 	pn = gfs2_lookup_simple(master, "per_node");
866 	if (IS_ERR(pn)) {
867 		error = PTR_ERR(pn);
868 		fs_err(sdp, "can't find per_node directory: %d\n", error);
869 		return error;
870 	}
871 
872 	sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
873 	sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
874 	if (IS_ERR(sdp->sd_sc_inode)) {
875 		error = PTR_ERR(sdp->sd_sc_inode);
876 		fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
877 		goto fail;
878 	}
879 
880 	sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
881 	sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
882 	if (IS_ERR(sdp->sd_qc_inode)) {
883 		error = PTR_ERR(sdp->sd_qc_inode);
884 		fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
885 		goto fail_ut_i;
886 	}
887 
888 	iput(pn);
889 	pn = NULL;
890 
891 	ip = GFS2_I(sdp->sd_sc_inode);
892 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
893 				   &sdp->sd_sc_gh);
894 	if (error) {
895 		fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
896 		goto fail_qc_i;
897 	}
898 
899 	ip = GFS2_I(sdp->sd_qc_inode);
900 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
901 				   &sdp->sd_qc_gh);
902 	if (error) {
903 		fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
904 		goto fail_ut_gh;
905 	}
906 
907 	return 0;
908 
909 fail_qc_gh:
910 	gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
911 fail_ut_gh:
912 	gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
913 fail_qc_i:
914 	iput(sdp->sd_qc_inode);
915 fail_ut_i:
916 	iput(sdp->sd_sc_inode);
917 fail:
918 	iput(pn);
919 	return error;
920 }
921 
922 static const match_table_t nolock_tokens = {
923 	{ Opt_jid, "jid=%d", },
924 	{ Opt_err, NULL },
925 };
926 
927 static const struct lm_lockops nolock_ops = {
928 	.lm_proto_name = "lock_nolock",
929 	.lm_put_lock = gfs2_glock_free,
930 	.lm_tokens = &nolock_tokens,
931 };
932 
933 /**
934  * gfs2_lm_mount - mount a locking protocol
935  * @sdp: the filesystem
936  * @args: mount arguments
937  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
938  *
939  * Returns: errno
940  */
941 
gfs2_lm_mount(struct gfs2_sbd * sdp,int silent)942 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
943 {
944 	const struct lm_lockops *lm;
945 	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
946 	struct gfs2_args *args = &sdp->sd_args;
947 	const char *proto = sdp->sd_proto_name;
948 	const char *table = sdp->sd_table_name;
949 	char *o, *options;
950 	int ret;
951 
952 	if (!strcmp("lock_nolock", proto)) {
953 		lm = &nolock_ops;
954 		sdp->sd_args.ar_localflocks = 1;
955 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
956 	} else if (!strcmp("lock_dlm", proto)) {
957 		lm = &gfs2_dlm_ops;
958 #endif
959 	} else {
960 		pr_info("can't find protocol %s\n", proto);
961 		return -ENOENT;
962 	}
963 
964 	fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
965 
966 	ls->ls_ops = lm;
967 	ls->ls_first = 1;
968 
969 	for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
970 		substring_t tmp[MAX_OPT_ARGS];
971 		int token, option;
972 
973 		if (!o || !*o)
974 			continue;
975 
976 		token = match_token(o, *lm->lm_tokens, tmp);
977 		switch (token) {
978 		case Opt_jid:
979 			ret = match_int(&tmp[0], &option);
980 			if (ret || option < 0)
981 				goto hostdata_error;
982 			if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
983 				ls->ls_jid = option;
984 			break;
985 		case Opt_id:
986 		case Opt_nodir:
987 			/* Obsolete, but left for backward compat purposes */
988 			break;
989 		case Opt_first:
990 			ret = match_int(&tmp[0], &option);
991 			if (ret || (option != 0 && option != 1))
992 				goto hostdata_error;
993 			ls->ls_first = option;
994 			break;
995 		case Opt_err:
996 		default:
997 hostdata_error:
998 			fs_info(sdp, "unknown hostdata (%s)\n", o);
999 			return -EINVAL;
1000 		}
1001 	}
1002 
1003 	if (lm->lm_mount == NULL) {
1004 		fs_info(sdp, "Now mounting FS...\n");
1005 		complete_all(&sdp->sd_locking_init);
1006 		return 0;
1007 	}
1008 	ret = lm->lm_mount(sdp, table);
1009 	if (ret == 0)
1010 		fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1011 	complete_all(&sdp->sd_locking_init);
1012 	return ret;
1013 }
1014 
gfs2_lm_unmount(struct gfs2_sbd * sdp)1015 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1016 {
1017 	const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1018 	if (likely(!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) &&
1019 	    lm->lm_unmount)
1020 		lm->lm_unmount(sdp);
1021 }
1022 
wait_on_journal(struct gfs2_sbd * sdp)1023 static int wait_on_journal(struct gfs2_sbd *sdp)
1024 {
1025 	if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1026 		return 0;
1027 
1028 	return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1029 		? -EINTR : 0;
1030 }
1031 
gfs2_online_uevent(struct gfs2_sbd * sdp)1032 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1033 {
1034 	struct super_block *sb = sdp->sd_vfs;
1035 	char ro[20];
1036 	char spectator[20];
1037 	char *envp[] = { ro, spectator, NULL };
1038 	sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1039 	sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1040 	kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1041 }
1042 
1043 /**
1044  * gfs2_fill_super - Read in superblock
1045  * @sb: The VFS superblock
1046  * @args: Mount options
1047  * @silent: Don't complain if it's not a GFS2 filesystem
1048  *
1049  * Returns: -errno
1050  */
gfs2_fill_super(struct super_block * sb,struct fs_context * fc)1051 static int gfs2_fill_super(struct super_block *sb, struct fs_context *fc)
1052 {
1053 	struct gfs2_args *args = fc->fs_private;
1054 	int silent = fc->sb_flags & SB_SILENT;
1055 	struct gfs2_sbd *sdp;
1056 	struct gfs2_holder mount_gh;
1057 	int error;
1058 
1059 	sdp = init_sbd(sb);
1060 	if (!sdp) {
1061 		pr_warn("can't alloc struct gfs2_sbd\n");
1062 		return -ENOMEM;
1063 	}
1064 	sdp->sd_args = *args;
1065 
1066 	if (sdp->sd_args.ar_spectator) {
1067                 sb->s_flags |= SB_RDONLY;
1068 		set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1069 	}
1070 	if (sdp->sd_args.ar_posix_acl)
1071 		sb->s_flags |= SB_POSIXACL;
1072 	if (sdp->sd_args.ar_nobarrier)
1073 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1074 
1075 	sb->s_flags |= SB_NOSEC;
1076 	sb->s_magic = GFS2_MAGIC;
1077 	sb->s_op = &gfs2_super_ops;
1078 	sb->s_d_op = &gfs2_dops;
1079 	sb->s_export_op = &gfs2_export_ops;
1080 	sb->s_xattr = gfs2_xattr_handlers;
1081 	sb->s_qcop = &gfs2_quotactl_ops;
1082 	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1083 	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1084 	sb->s_time_gran = 1;
1085 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1086 
1087 	/* Set up the buffer cache and fill in some fake block size values
1088 	   to allow us to read-in the on-disk superblock. */
1089 	sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1090 	sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1091 	sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1092                                GFS2_BASIC_BLOCK_SHIFT;
1093 	sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1094 
1095 	sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1096 	sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1097 	if (sdp->sd_args.ar_statfs_quantum) {
1098 		sdp->sd_tune.gt_statfs_slow = 0;
1099 		sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1100 	} else {
1101 		sdp->sd_tune.gt_statfs_slow = 1;
1102 		sdp->sd_tune.gt_statfs_quantum = 30;
1103 	}
1104 
1105 	error = init_names(sdp, silent);
1106 	if (error)
1107 		goto fail_free;
1108 
1109 	snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1110 
1111 	error = gfs2_sys_fs_add(sdp);
1112 	if (error)
1113 		goto fail_free;
1114 
1115 	gfs2_create_debugfs_file(sdp);
1116 
1117 	error = gfs2_lm_mount(sdp, silent);
1118 	if (error)
1119 		goto fail_debug;
1120 
1121 	error = init_locking(sdp, &mount_gh, DO);
1122 	if (error)
1123 		goto fail_lm;
1124 
1125 	error = init_sb(sdp, silent);
1126 	if (error)
1127 		goto fail_locking;
1128 
1129 	error = wait_on_journal(sdp);
1130 	if (error)
1131 		goto fail_sb;
1132 
1133 	/*
1134 	 * If user space has failed to join the cluster or some similar
1135 	 * failure has occurred, then the journal id will contain a
1136 	 * negative (error) number. This will then be returned to the
1137 	 * caller (of the mount syscall). We do this even for spectator
1138 	 * mounts (which just write a jid of 0 to indicate "ok" even though
1139 	 * the jid is unused in the spectator case)
1140 	 */
1141 	if (sdp->sd_lockstruct.ls_jid < 0) {
1142 		error = sdp->sd_lockstruct.ls_jid;
1143 		sdp->sd_lockstruct.ls_jid = 0;
1144 		goto fail_sb;
1145 	}
1146 
1147 	if (sdp->sd_args.ar_spectator)
1148 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1149 			 sdp->sd_table_name);
1150 	else
1151 		snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1152 			 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1153 
1154 	error = init_inodes(sdp, DO);
1155 	if (error)
1156 		goto fail_sb;
1157 
1158 	error = init_per_node(sdp, DO);
1159 	if (error)
1160 		goto fail_inodes;
1161 
1162 	error = gfs2_statfs_init(sdp);
1163 	if (error) {
1164 		fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1165 		goto fail_per_node;
1166 	}
1167 
1168 	if (sb_rdonly(sb)) {
1169 		struct gfs2_holder freeze_gh;
1170 
1171 		error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1172 					   GL_EXACT, &freeze_gh);
1173 		if (error) {
1174 			fs_err(sdp, "can't make FS RO: %d\n", error);
1175 			goto fail_per_node;
1176 		}
1177 		gfs2_glock_dq_uninit(&freeze_gh);
1178 	} else {
1179 		error = gfs2_make_fs_rw(sdp);
1180 		if (error) {
1181 			fs_err(sdp, "can't make FS RW: %d\n", error);
1182 			goto fail_per_node;
1183 		}
1184 	}
1185 
1186 	gfs2_glock_dq_uninit(&mount_gh);
1187 	gfs2_online_uevent(sdp);
1188 	return 0;
1189 
1190 fail_per_node:
1191 	init_per_node(sdp, UNDO);
1192 fail_inodes:
1193 	init_inodes(sdp, UNDO);
1194 fail_sb:
1195 	if (sdp->sd_root_dir)
1196 		dput(sdp->sd_root_dir);
1197 	if (sdp->sd_master_dir)
1198 		dput(sdp->sd_master_dir);
1199 	if (sb->s_root)
1200 		dput(sb->s_root);
1201 	sb->s_root = NULL;
1202 fail_locking:
1203 	init_locking(sdp, &mount_gh, UNDO);
1204 fail_lm:
1205 	complete_all(&sdp->sd_journal_ready);
1206 	gfs2_gl_hash_clear(sdp);
1207 	gfs2_lm_unmount(sdp);
1208 fail_debug:
1209 	gfs2_delete_debugfs_file(sdp);
1210 	gfs2_sys_fs_del(sdp);
1211 fail_free:
1212 	free_sbd(sdp);
1213 	sb->s_fs_info = NULL;
1214 	return error;
1215 }
1216 
1217 /**
1218  * gfs2_get_tree - Get the GFS2 superblock and root directory
1219  * @fc: The filesystem context
1220  *
1221  * Returns: 0 or -errno on error
1222  */
gfs2_get_tree(struct fs_context * fc)1223 static int gfs2_get_tree(struct fs_context *fc)
1224 {
1225 	struct gfs2_args *args = fc->fs_private;
1226 	struct gfs2_sbd *sdp;
1227 	int error;
1228 
1229 	error = get_tree_bdev(fc, gfs2_fill_super);
1230 	if (error)
1231 		return error;
1232 
1233 	sdp = fc->root->d_sb->s_fs_info;
1234 	dput(fc->root);
1235 	if (args->ar_meta)
1236 		fc->root = dget(sdp->sd_master_dir);
1237 	else
1238 		fc->root = dget(sdp->sd_root_dir);
1239 	return 0;
1240 }
1241 
gfs2_fc_free(struct fs_context * fc)1242 static void gfs2_fc_free(struct fs_context *fc)
1243 {
1244 	struct gfs2_args *args = fc->fs_private;
1245 
1246 	kfree(args);
1247 }
1248 
1249 enum gfs2_param {
1250 	Opt_lockproto,
1251 	Opt_locktable,
1252 	Opt_hostdata,
1253 	Opt_spectator,
1254 	Opt_ignore_local_fs,
1255 	Opt_localflocks,
1256 	Opt_localcaching,
1257 	Opt_debug,
1258 	Opt_upgrade,
1259 	Opt_acl,
1260 	Opt_quota,
1261 	Opt_suiddir,
1262 	Opt_data,
1263 	Opt_meta,
1264 	Opt_discard,
1265 	Opt_commit,
1266 	Opt_errors,
1267 	Opt_statfs_quantum,
1268 	Opt_statfs_percent,
1269 	Opt_quota_quantum,
1270 	Opt_barrier,
1271 	Opt_rgrplvb,
1272 	Opt_loccookie,
1273 };
1274 
1275 enum opt_quota {
1276 	Opt_quota_unset = 0,
1277 	Opt_quota_off,
1278 	Opt_quota_account,
1279 	Opt_quota_on,
1280 };
1281 
1282 static const unsigned int opt_quota_values[] = {
1283 	[Opt_quota_off]     = GFS2_QUOTA_OFF,
1284 	[Opt_quota_account] = GFS2_QUOTA_ACCOUNT,
1285 	[Opt_quota_on]      = GFS2_QUOTA_ON,
1286 };
1287 
1288 enum opt_data {
1289 	Opt_data_writeback = GFS2_DATA_WRITEBACK,
1290 	Opt_data_ordered   = GFS2_DATA_ORDERED,
1291 };
1292 
1293 enum opt_errors {
1294 	Opt_errors_withdraw = GFS2_ERRORS_WITHDRAW,
1295 	Opt_errors_panic    = GFS2_ERRORS_PANIC,
1296 };
1297 
1298 static const struct fs_parameter_spec gfs2_param_specs[] = {
1299 	fsparam_string ("lockproto",          Opt_lockproto),
1300 	fsparam_string ("locktable",          Opt_locktable),
1301 	fsparam_string ("hostdata",           Opt_hostdata),
1302 	fsparam_flag   ("spectator",          Opt_spectator),
1303 	fsparam_flag   ("norecovery",         Opt_spectator),
1304 	fsparam_flag   ("ignore_local_fs",    Opt_ignore_local_fs),
1305 	fsparam_flag   ("localflocks",        Opt_localflocks),
1306 	fsparam_flag   ("localcaching",       Opt_localcaching),
1307 	fsparam_flag_no("debug",              Opt_debug),
1308 	fsparam_flag   ("upgrade",            Opt_upgrade),
1309 	fsparam_flag_no("acl",                Opt_acl),
1310 	fsparam_flag_no("suiddir",            Opt_suiddir),
1311 	fsparam_enum   ("data",               Opt_data),
1312 	fsparam_flag   ("meta",               Opt_meta),
1313 	fsparam_flag_no("discard",            Opt_discard),
1314 	fsparam_s32    ("commit",             Opt_commit),
1315 	fsparam_enum   ("errors",             Opt_errors),
1316 	fsparam_s32    ("statfs_quantum",     Opt_statfs_quantum),
1317 	fsparam_s32    ("statfs_percent",     Opt_statfs_percent),
1318 	fsparam_s32    ("quota_quantum",      Opt_quota_quantum),
1319 	fsparam_flag_no("barrier",            Opt_barrier),
1320 	fsparam_flag_no("rgrplvb",            Opt_rgrplvb),
1321 	fsparam_flag_no("loccookie",          Opt_loccookie),
1322 	/* quota can be a flag or an enum so it gets special treatment */
1323 	__fsparam(fs_param_is_enum, "quota", Opt_quota, fs_param_neg_with_no|fs_param_v_optional),
1324 	{}
1325 };
1326 
1327 static const struct fs_parameter_enum gfs2_param_enums[] = {
1328 	{ Opt_quota,    "off",        Opt_quota_off },
1329 	{ Opt_quota,    "account",    Opt_quota_account },
1330 	{ Opt_quota,    "on",         Opt_quota_on },
1331 	{ Opt_data,     "writeback",  Opt_data_writeback },
1332 	{ Opt_data,     "ordered",    Opt_data_ordered },
1333 	{ Opt_errors,   "withdraw",   Opt_errors_withdraw },
1334 	{ Opt_errors,   "panic",      Opt_errors_panic },
1335 	{}
1336 };
1337 
1338 const struct fs_parameter_description gfs2_fs_parameters = {
1339 	.name = "gfs2",
1340 	.specs = gfs2_param_specs,
1341 	.enums = gfs2_param_enums,
1342 };
1343 
1344 /* Parse a single mount parameter */
gfs2_parse_param(struct fs_context * fc,struct fs_parameter * param)1345 static int gfs2_parse_param(struct fs_context *fc, struct fs_parameter *param)
1346 {
1347 	struct gfs2_args *args = fc->fs_private;
1348 	struct fs_parse_result result;
1349 	int o;
1350 
1351 	o = fs_parse(fc, &gfs2_fs_parameters, param, &result);
1352 	if (o < 0)
1353 		return o;
1354 
1355 	switch (o) {
1356 	case Opt_lockproto:
1357 		strscpy(args->ar_lockproto, param->string, GFS2_LOCKNAME_LEN);
1358 		break;
1359 	case Opt_locktable:
1360 		strscpy(args->ar_locktable, param->string, GFS2_LOCKNAME_LEN);
1361 		break;
1362 	case Opt_hostdata:
1363 		strscpy(args->ar_hostdata, param->string, GFS2_LOCKNAME_LEN);
1364 		break;
1365 	case Opt_spectator:
1366 		args->ar_spectator = 1;
1367 		break;
1368 	case Opt_ignore_local_fs:
1369 		/* Retained for backwards compat only */
1370 		break;
1371 	case Opt_localflocks:
1372 		args->ar_localflocks = 1;
1373 		break;
1374 	case Opt_localcaching:
1375 		/* Retained for backwards compat only */
1376 		break;
1377 	case Opt_debug:
1378 		if (result.boolean && args->ar_errors == GFS2_ERRORS_PANIC)
1379 			return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1380 		args->ar_debug = result.boolean;
1381 		break;
1382 	case Opt_upgrade:
1383 		/* Retained for backwards compat only */
1384 		break;
1385 	case Opt_acl:
1386 		args->ar_posix_acl = result.boolean;
1387 		break;
1388 	case Opt_quota:
1389 		/* The quota option can be a flag or an enum. A non-zero int_32
1390 		   result means that we have an enum index. Otherwise we have
1391 		   to rely on the 'negated' flag to tell us whether 'quota' or
1392 		   'noquota' was specified. */
1393 		if (result.negated)
1394 			args->ar_quota = GFS2_QUOTA_OFF;
1395 		else if (result.int_32 > 0)
1396 			args->ar_quota = opt_quota_values[result.int_32];
1397 		else
1398 			args->ar_quota = GFS2_QUOTA_ON;
1399 		break;
1400 	case Opt_suiddir:
1401 		args->ar_suiddir = result.boolean;
1402 		break;
1403 	case Opt_data:
1404 		/* The uint_32 result maps directly to GFS2_DATA_* */
1405 		args->ar_data = result.uint_32;
1406 		break;
1407 	case Opt_meta:
1408 		args->ar_meta = 1;
1409 		break;
1410 	case Opt_discard:
1411 		args->ar_discard = result.boolean;
1412 		break;
1413 	case Opt_commit:
1414 		if (result.int_32 <= 0)
1415 			return invalf(fc, "gfs2: commit mount option requires a positive numeric argument");
1416 		args->ar_commit = result.int_32;
1417 		break;
1418 	case Opt_statfs_quantum:
1419 		if (result.int_32 < 0)
1420 			return invalf(fc, "gfs2: statfs_quantum mount option requires a non-negative numeric argument");
1421 		args->ar_statfs_quantum = result.int_32;
1422 		break;
1423 	case Opt_quota_quantum:
1424 		if (result.int_32 <= 0)
1425 			return invalf(fc, "gfs2: quota_quantum mount option requires a positive numeric argument");
1426 		args->ar_quota_quantum = result.int_32;
1427 		break;
1428 	case Opt_statfs_percent:
1429 		if (result.int_32 < 0 || result.int_32 > 100)
1430 			return invalf(fc, "gfs2: statfs_percent mount option requires a numeric argument between 0 and 100");
1431 		args->ar_statfs_percent = result.int_32;
1432 		break;
1433 	case Opt_errors:
1434 		if (args->ar_debug && result.uint_32 == GFS2_ERRORS_PANIC)
1435 			return invalf(fc, "gfs2: -o debug and -o errors=panic are mutually exclusive");
1436 		args->ar_errors = result.uint_32;
1437 		break;
1438 	case Opt_barrier:
1439 		args->ar_nobarrier = result.boolean;
1440 		break;
1441 	case Opt_rgrplvb:
1442 		args->ar_rgrplvb = result.boolean;
1443 		break;
1444 	case Opt_loccookie:
1445 		args->ar_loccookie = result.boolean;
1446 		break;
1447 	default:
1448 		return invalf(fc, "gfs2: invalid mount option: %s", param->key);
1449 	}
1450 	return 0;
1451 }
1452 
gfs2_reconfigure(struct fs_context * fc)1453 static int gfs2_reconfigure(struct fs_context *fc)
1454 {
1455 	struct super_block *sb = fc->root->d_sb;
1456 	struct gfs2_sbd *sdp = sb->s_fs_info;
1457 	struct gfs2_args *oldargs = &sdp->sd_args;
1458 	struct gfs2_args *newargs = fc->fs_private;
1459 	struct gfs2_tune *gt = &sdp->sd_tune;
1460 	int error = 0;
1461 
1462 	sync_filesystem(sb);
1463 
1464 	spin_lock(&gt->gt_spin);
1465 	oldargs->ar_commit = gt->gt_logd_secs;
1466 	oldargs->ar_quota_quantum = gt->gt_quota_quantum;
1467 	if (gt->gt_statfs_slow)
1468 		oldargs->ar_statfs_quantum = 0;
1469 	else
1470 		oldargs->ar_statfs_quantum = gt->gt_statfs_quantum;
1471 	spin_unlock(&gt->gt_spin);
1472 
1473 	if (strcmp(newargs->ar_lockproto, oldargs->ar_lockproto)) {
1474 		errorf(fc, "gfs2: reconfiguration of locking protocol not allowed");
1475 		return -EINVAL;
1476 	}
1477 	if (strcmp(newargs->ar_locktable, oldargs->ar_locktable)) {
1478 		errorf(fc, "gfs2: reconfiguration of lock table not allowed");
1479 		return -EINVAL;
1480 	}
1481 	if (strcmp(newargs->ar_hostdata, oldargs->ar_hostdata)) {
1482 		errorf(fc, "gfs2: reconfiguration of host data not allowed");
1483 		return -EINVAL;
1484 	}
1485 	if (newargs->ar_spectator != oldargs->ar_spectator) {
1486 		errorf(fc, "gfs2: reconfiguration of spectator mode not allowed");
1487 		return -EINVAL;
1488 	}
1489 	if (newargs->ar_localflocks != oldargs->ar_localflocks) {
1490 		errorf(fc, "gfs2: reconfiguration of localflocks not allowed");
1491 		return -EINVAL;
1492 	}
1493 	if (newargs->ar_meta != oldargs->ar_meta) {
1494 		errorf(fc, "gfs2: switching between gfs2 and gfs2meta not allowed");
1495 		return -EINVAL;
1496 	}
1497 	if (oldargs->ar_spectator)
1498 		fc->sb_flags |= SB_RDONLY;
1499 
1500 	if ((sb->s_flags ^ fc->sb_flags) & SB_RDONLY) {
1501 		if (fc->sb_flags & SB_RDONLY) {
1502 			error = gfs2_make_fs_ro(sdp);
1503 			if (error)
1504 				errorf(fc, "gfs2: unable to remount read-only");
1505 		} else {
1506 			error = gfs2_make_fs_rw(sdp);
1507 			if (error)
1508 				errorf(fc, "gfs2: unable to remount read-write");
1509 		}
1510 	}
1511 	sdp->sd_args = *newargs;
1512 
1513 	if (sdp->sd_args.ar_posix_acl)
1514 		sb->s_flags |= SB_POSIXACL;
1515 	else
1516 		sb->s_flags &= ~SB_POSIXACL;
1517 	if (sdp->sd_args.ar_nobarrier)
1518 		set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1519 	else
1520 		clear_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1521 	spin_lock(&gt->gt_spin);
1522 	gt->gt_logd_secs = newargs->ar_commit;
1523 	gt->gt_quota_quantum = newargs->ar_quota_quantum;
1524 	if (newargs->ar_statfs_quantum) {
1525 		gt->gt_statfs_slow = 0;
1526 		gt->gt_statfs_quantum = newargs->ar_statfs_quantum;
1527 	}
1528 	else {
1529 		gt->gt_statfs_slow = 1;
1530 		gt->gt_statfs_quantum = 30;
1531 	}
1532 	spin_unlock(&gt->gt_spin);
1533 
1534 	gfs2_online_uevent(sdp);
1535 	return error;
1536 }
1537 
1538 static const struct fs_context_operations gfs2_context_ops = {
1539 	.free        = gfs2_fc_free,
1540 	.parse_param = gfs2_parse_param,
1541 	.get_tree    = gfs2_get_tree,
1542 	.reconfigure = gfs2_reconfigure,
1543 };
1544 
1545 /* Set up the filesystem mount context */
gfs2_init_fs_context(struct fs_context * fc)1546 static int gfs2_init_fs_context(struct fs_context *fc)
1547 {
1548 	struct gfs2_args *args;
1549 
1550 	args = kmalloc(sizeof(*args), GFP_KERNEL);
1551 	if (args == NULL)
1552 		return -ENOMEM;
1553 
1554 	if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
1555 		struct gfs2_sbd *sdp = fc->root->d_sb->s_fs_info;
1556 
1557 		*args = sdp->sd_args;
1558 	} else {
1559 		memset(args, 0, sizeof(*args));
1560 		args->ar_quota = GFS2_QUOTA_DEFAULT;
1561 		args->ar_data = GFS2_DATA_DEFAULT;
1562 		args->ar_commit = 30;
1563 		args->ar_statfs_quantum = 30;
1564 		args->ar_quota_quantum = 60;
1565 		args->ar_errors = GFS2_ERRORS_DEFAULT;
1566 	}
1567 	fc->fs_private = args;
1568 	fc->ops = &gfs2_context_ops;
1569 	return 0;
1570 }
1571 
set_meta_super(struct super_block * s,struct fs_context * fc)1572 static int set_meta_super(struct super_block *s, struct fs_context *fc)
1573 {
1574 	return -EINVAL;
1575 }
1576 
test_meta_super(struct super_block * s,struct fs_context * fc)1577 static int test_meta_super(struct super_block *s, struct fs_context *fc)
1578 {
1579 	return (fc->sget_key == s->s_bdev);
1580 }
1581 
gfs2_meta_get_tree(struct fs_context * fc)1582 static int gfs2_meta_get_tree(struct fs_context *fc)
1583 {
1584 	struct super_block *s;
1585 	struct gfs2_sbd *sdp;
1586 	struct path path;
1587 	int error;
1588 
1589 	if (!fc->source || !*fc->source)
1590 		return -EINVAL;
1591 
1592 	error = kern_path(fc->source, LOOKUP_FOLLOW, &path);
1593 	if (error) {
1594 		pr_warn("path_lookup on %s returned error %d\n",
1595 		        fc->source, error);
1596 		return error;
1597 	}
1598 	fc->fs_type = &gfs2_fs_type;
1599 	fc->sget_key = path.dentry->d_sb->s_bdev;
1600 	s = sget_fc(fc, test_meta_super, set_meta_super);
1601 	path_put(&path);
1602 	if (IS_ERR(s)) {
1603 		pr_warn("gfs2 mount does not exist\n");
1604 		return PTR_ERR(s);
1605 	}
1606 	if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) {
1607 		deactivate_locked_super(s);
1608 		return -EBUSY;
1609 	}
1610 	sdp = s->s_fs_info;
1611 	fc->root = dget(sdp->sd_master_dir);
1612 	return 0;
1613 }
1614 
1615 static const struct fs_context_operations gfs2_meta_context_ops = {
1616 	.free        = gfs2_fc_free,
1617 	.get_tree    = gfs2_meta_get_tree,
1618 };
1619 
gfs2_meta_init_fs_context(struct fs_context * fc)1620 static int gfs2_meta_init_fs_context(struct fs_context *fc)
1621 {
1622 	int ret = gfs2_init_fs_context(fc);
1623 
1624 	if (ret)
1625 		return ret;
1626 
1627 	fc->ops = &gfs2_meta_context_ops;
1628 	return 0;
1629 }
1630 
gfs2_kill_sb(struct super_block * sb)1631 static void gfs2_kill_sb(struct super_block *sb)
1632 {
1633 	struct gfs2_sbd *sdp = sb->s_fs_info;
1634 
1635 	if (sdp == NULL) {
1636 		kill_block_super(sb);
1637 		return;
1638 	}
1639 
1640 	gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1641 	dput(sdp->sd_root_dir);
1642 	dput(sdp->sd_master_dir);
1643 	sdp->sd_root_dir = NULL;
1644 	sdp->sd_master_dir = NULL;
1645 	shrink_dcache_sb(sb);
1646 	kill_block_super(sb);
1647 }
1648 
1649 struct file_system_type gfs2_fs_type = {
1650 	.name = "gfs2",
1651 	.fs_flags = FS_REQUIRES_DEV,
1652 	.init_fs_context = gfs2_init_fs_context,
1653 	.parameters = &gfs2_fs_parameters,
1654 	.kill_sb = gfs2_kill_sb,
1655 	.owner = THIS_MODULE,
1656 };
1657 MODULE_ALIAS_FS("gfs2");
1658 
1659 struct file_system_type gfs2meta_fs_type = {
1660 	.name = "gfs2meta",
1661 	.fs_flags = FS_REQUIRES_DEV,
1662 	.init_fs_context = gfs2_meta_init_fs_context,
1663 	.owner = THIS_MODULE,
1664 };
1665 MODULE_ALIAS_FS("gfs2meta");
1666