xref: /linux/fs/f2fs/super.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * fs/f2fs/super.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/module.h>
12 #include <linux/init.h>
13 #include <linux/fs.h>
14 #include <linux/statfs.h>
15 #include <linux/buffer_head.h>
16 #include <linux/backing-dev.h>
17 #include <linux/kthread.h>
18 #include <linux/parser.h>
19 #include <linux/mount.h>
20 #include <linux/seq_file.h>
21 #include <linux/proc_fs.h>
22 #include <linux/random.h>
23 #include <linux/exportfs.h>
24 #include <linux/blkdev.h>
25 #include <linux/f2fs_fs.h>
26 #include <linux/sysfs.h>
27 
28 #include "f2fs.h"
29 #include "node.h"
30 #include "segment.h"
31 #include "xattr.h"
32 #include "gc.h"
33 
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/f2fs.h>
36 
37 static struct proc_dir_entry *f2fs_proc_root;
38 static struct kmem_cache *f2fs_inode_cachep;
39 static struct kset *f2fs_kset;
40 
41 enum {
42 	Opt_gc_background,
43 	Opt_disable_roll_forward,
44 	Opt_discard,
45 	Opt_noheap,
46 	Opt_user_xattr,
47 	Opt_nouser_xattr,
48 	Opt_acl,
49 	Opt_noacl,
50 	Opt_active_logs,
51 	Opt_disable_ext_identify,
52 	Opt_inline_xattr,
53 	Opt_inline_data,
54 	Opt_flush_merge,
55 	Opt_nobarrier,
56 	Opt_err,
57 };
58 
59 static match_table_t f2fs_tokens = {
60 	{Opt_gc_background, "background_gc=%s"},
61 	{Opt_disable_roll_forward, "disable_roll_forward"},
62 	{Opt_discard, "discard"},
63 	{Opt_noheap, "no_heap"},
64 	{Opt_user_xattr, "user_xattr"},
65 	{Opt_nouser_xattr, "nouser_xattr"},
66 	{Opt_acl, "acl"},
67 	{Opt_noacl, "noacl"},
68 	{Opt_active_logs, "active_logs=%u"},
69 	{Opt_disable_ext_identify, "disable_ext_identify"},
70 	{Opt_inline_xattr, "inline_xattr"},
71 	{Opt_inline_data, "inline_data"},
72 	{Opt_flush_merge, "flush_merge"},
73 	{Opt_nobarrier, "nobarrier"},
74 	{Opt_err, NULL},
75 };
76 
77 /* Sysfs support for f2fs */
78 enum {
79 	GC_THREAD,	/* struct f2fs_gc_thread */
80 	SM_INFO,	/* struct f2fs_sm_info */
81 	NM_INFO,	/* struct f2fs_nm_info */
82 	F2FS_SBI,	/* struct f2fs_sb_info */
83 };
84 
85 struct f2fs_attr {
86 	struct attribute attr;
87 	ssize_t (*show)(struct f2fs_attr *, struct f2fs_sb_info *, char *);
88 	ssize_t (*store)(struct f2fs_attr *, struct f2fs_sb_info *,
89 			 const char *, size_t);
90 	int struct_type;
91 	int offset;
92 };
93 
94 static unsigned char *__struct_ptr(struct f2fs_sb_info *sbi, int struct_type)
95 {
96 	if (struct_type == GC_THREAD)
97 		return (unsigned char *)sbi->gc_thread;
98 	else if (struct_type == SM_INFO)
99 		return (unsigned char *)SM_I(sbi);
100 	else if (struct_type == NM_INFO)
101 		return (unsigned char *)NM_I(sbi);
102 	else if (struct_type == F2FS_SBI)
103 		return (unsigned char *)sbi;
104 	return NULL;
105 }
106 
107 static ssize_t f2fs_sbi_show(struct f2fs_attr *a,
108 			struct f2fs_sb_info *sbi, char *buf)
109 {
110 	unsigned char *ptr = NULL;
111 	unsigned int *ui;
112 
113 	ptr = __struct_ptr(sbi, a->struct_type);
114 	if (!ptr)
115 		return -EINVAL;
116 
117 	ui = (unsigned int *)(ptr + a->offset);
118 
119 	return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
120 }
121 
122 static ssize_t f2fs_sbi_store(struct f2fs_attr *a,
123 			struct f2fs_sb_info *sbi,
124 			const char *buf, size_t count)
125 {
126 	unsigned char *ptr;
127 	unsigned long t;
128 	unsigned int *ui;
129 	ssize_t ret;
130 
131 	ptr = __struct_ptr(sbi, a->struct_type);
132 	if (!ptr)
133 		return -EINVAL;
134 
135 	ui = (unsigned int *)(ptr + a->offset);
136 
137 	ret = kstrtoul(skip_spaces(buf), 0, &t);
138 	if (ret < 0)
139 		return ret;
140 	*ui = t;
141 	return count;
142 }
143 
144 static ssize_t f2fs_attr_show(struct kobject *kobj,
145 				struct attribute *attr, char *buf)
146 {
147 	struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
148 								s_kobj);
149 	struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr);
150 
151 	return a->show ? a->show(a, sbi, buf) : 0;
152 }
153 
154 static ssize_t f2fs_attr_store(struct kobject *kobj, struct attribute *attr,
155 						const char *buf, size_t len)
156 {
157 	struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
158 									s_kobj);
159 	struct f2fs_attr *a = container_of(attr, struct f2fs_attr, attr);
160 
161 	return a->store ? a->store(a, sbi, buf, len) : 0;
162 }
163 
164 static void f2fs_sb_release(struct kobject *kobj)
165 {
166 	struct f2fs_sb_info *sbi = container_of(kobj, struct f2fs_sb_info,
167 								s_kobj);
168 	complete(&sbi->s_kobj_unregister);
169 }
170 
171 #define F2FS_ATTR_OFFSET(_struct_type, _name, _mode, _show, _store, _offset) \
172 static struct f2fs_attr f2fs_attr_##_name = {			\
173 	.attr = {.name = __stringify(_name), .mode = _mode },	\
174 	.show	= _show,					\
175 	.store	= _store,					\
176 	.struct_type = _struct_type,				\
177 	.offset = _offset					\
178 }
179 
180 #define F2FS_RW_ATTR(struct_type, struct_name, name, elname)	\
181 	F2FS_ATTR_OFFSET(struct_type, name, 0644,		\
182 		f2fs_sbi_show, f2fs_sbi_store,			\
183 		offsetof(struct struct_name, elname))
184 
185 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_min_sleep_time, min_sleep_time);
186 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_max_sleep_time, max_sleep_time);
187 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_no_gc_sleep_time, no_gc_sleep_time);
188 F2FS_RW_ATTR(GC_THREAD, f2fs_gc_kthread, gc_idle, gc_idle);
189 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, reclaim_segments, rec_prefree_segments);
190 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, max_small_discards, max_discards);
191 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, ipu_policy, ipu_policy);
192 F2FS_RW_ATTR(SM_INFO, f2fs_sm_info, min_ipu_util, min_ipu_util);
193 F2FS_RW_ATTR(NM_INFO, f2fs_nm_info, ram_thresh, ram_thresh);
194 F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, max_victim_search, max_victim_search);
195 F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, dir_level, dir_level);
196 
197 #define ATTR_LIST(name) (&f2fs_attr_##name.attr)
198 static struct attribute *f2fs_attrs[] = {
199 	ATTR_LIST(gc_min_sleep_time),
200 	ATTR_LIST(gc_max_sleep_time),
201 	ATTR_LIST(gc_no_gc_sleep_time),
202 	ATTR_LIST(gc_idle),
203 	ATTR_LIST(reclaim_segments),
204 	ATTR_LIST(max_small_discards),
205 	ATTR_LIST(ipu_policy),
206 	ATTR_LIST(min_ipu_util),
207 	ATTR_LIST(max_victim_search),
208 	ATTR_LIST(dir_level),
209 	ATTR_LIST(ram_thresh),
210 	NULL,
211 };
212 
213 static const struct sysfs_ops f2fs_attr_ops = {
214 	.show	= f2fs_attr_show,
215 	.store	= f2fs_attr_store,
216 };
217 
218 static struct kobj_type f2fs_ktype = {
219 	.default_attrs	= f2fs_attrs,
220 	.sysfs_ops	= &f2fs_attr_ops,
221 	.release	= f2fs_sb_release,
222 };
223 
224 void f2fs_msg(struct super_block *sb, const char *level, const char *fmt, ...)
225 {
226 	struct va_format vaf;
227 	va_list args;
228 
229 	va_start(args, fmt);
230 	vaf.fmt = fmt;
231 	vaf.va = &args;
232 	printk("%sF2FS-fs (%s): %pV\n", level, sb->s_id, &vaf);
233 	va_end(args);
234 }
235 
236 static void init_once(void *foo)
237 {
238 	struct f2fs_inode_info *fi = (struct f2fs_inode_info *) foo;
239 
240 	inode_init_once(&fi->vfs_inode);
241 }
242 
243 static int parse_options(struct super_block *sb, char *options)
244 {
245 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
246 	substring_t args[MAX_OPT_ARGS];
247 	char *p, *name;
248 	int arg = 0;
249 
250 	if (!options)
251 		return 0;
252 
253 	while ((p = strsep(&options, ",")) != NULL) {
254 		int token;
255 		if (!*p)
256 			continue;
257 		/*
258 		 * Initialize args struct so we know whether arg was
259 		 * found; some options take optional arguments.
260 		 */
261 		args[0].to = args[0].from = NULL;
262 		token = match_token(p, f2fs_tokens, args);
263 
264 		switch (token) {
265 		case Opt_gc_background:
266 			name = match_strdup(&args[0]);
267 
268 			if (!name)
269 				return -ENOMEM;
270 			if (strlen(name) == 2 && !strncmp(name, "on", 2))
271 				set_opt(sbi, BG_GC);
272 			else if (strlen(name) == 3 && !strncmp(name, "off", 3))
273 				clear_opt(sbi, BG_GC);
274 			else {
275 				kfree(name);
276 				return -EINVAL;
277 			}
278 			kfree(name);
279 			break;
280 		case Opt_disable_roll_forward:
281 			set_opt(sbi, DISABLE_ROLL_FORWARD);
282 			break;
283 		case Opt_discard:
284 			set_opt(sbi, DISCARD);
285 			break;
286 		case Opt_noheap:
287 			set_opt(sbi, NOHEAP);
288 			break;
289 #ifdef CONFIG_F2FS_FS_XATTR
290 		case Opt_user_xattr:
291 			set_opt(sbi, XATTR_USER);
292 			break;
293 		case Opt_nouser_xattr:
294 			clear_opt(sbi, XATTR_USER);
295 			break;
296 		case Opt_inline_xattr:
297 			set_opt(sbi, INLINE_XATTR);
298 			break;
299 #else
300 		case Opt_user_xattr:
301 			f2fs_msg(sb, KERN_INFO,
302 				"user_xattr options not supported");
303 			break;
304 		case Opt_nouser_xattr:
305 			f2fs_msg(sb, KERN_INFO,
306 				"nouser_xattr options not supported");
307 			break;
308 		case Opt_inline_xattr:
309 			f2fs_msg(sb, KERN_INFO,
310 				"inline_xattr options not supported");
311 			break;
312 #endif
313 #ifdef CONFIG_F2FS_FS_POSIX_ACL
314 		case Opt_acl:
315 			set_opt(sbi, POSIX_ACL);
316 			break;
317 		case Opt_noacl:
318 			clear_opt(sbi, POSIX_ACL);
319 			break;
320 #else
321 		case Opt_acl:
322 			f2fs_msg(sb, KERN_INFO, "acl options not supported");
323 			break;
324 		case Opt_noacl:
325 			f2fs_msg(sb, KERN_INFO, "noacl options not supported");
326 			break;
327 #endif
328 		case Opt_active_logs:
329 			if (args->from && match_int(args, &arg))
330 				return -EINVAL;
331 			if (arg != 2 && arg != 4 && arg != NR_CURSEG_TYPE)
332 				return -EINVAL;
333 			sbi->active_logs = arg;
334 			break;
335 		case Opt_disable_ext_identify:
336 			set_opt(sbi, DISABLE_EXT_IDENTIFY);
337 			break;
338 		case Opt_inline_data:
339 			set_opt(sbi, INLINE_DATA);
340 			break;
341 		case Opt_flush_merge:
342 			set_opt(sbi, FLUSH_MERGE);
343 			break;
344 		case Opt_nobarrier:
345 			set_opt(sbi, NOBARRIER);
346 			break;
347 		default:
348 			f2fs_msg(sb, KERN_ERR,
349 				"Unrecognized mount option \"%s\" or missing value",
350 				p);
351 			return -EINVAL;
352 		}
353 	}
354 	return 0;
355 }
356 
357 static struct inode *f2fs_alloc_inode(struct super_block *sb)
358 {
359 	struct f2fs_inode_info *fi;
360 
361 	fi = kmem_cache_alloc(f2fs_inode_cachep, GFP_F2FS_ZERO);
362 	if (!fi)
363 		return NULL;
364 
365 	init_once((void *) fi);
366 
367 	/* Initialize f2fs-specific inode info */
368 	fi->vfs_inode.i_version = 1;
369 	atomic_set(&fi->dirty_dents, 0);
370 	fi->i_current_depth = 1;
371 	fi->i_advise = 0;
372 	rwlock_init(&fi->ext.ext_lock);
373 	init_rwsem(&fi->i_sem);
374 
375 	set_inode_flag(fi, FI_NEW_INODE);
376 
377 	if (test_opt(F2FS_SB(sb), INLINE_XATTR))
378 		set_inode_flag(fi, FI_INLINE_XATTR);
379 
380 	/* Will be used by directory only */
381 	fi->i_dir_level = F2FS_SB(sb)->dir_level;
382 
383 	return &fi->vfs_inode;
384 }
385 
386 static int f2fs_drop_inode(struct inode *inode)
387 {
388 	/*
389 	 * This is to avoid a deadlock condition like below.
390 	 * writeback_single_inode(inode)
391 	 *  - f2fs_write_data_page
392 	 *    - f2fs_gc -> iput -> evict
393 	 *       - inode_wait_for_writeback(inode)
394 	 */
395 	if (!inode_unhashed(inode) && inode->i_state & I_SYNC)
396 		return 0;
397 	return generic_drop_inode(inode);
398 }
399 
400 /*
401  * f2fs_dirty_inode() is called from __mark_inode_dirty()
402  *
403  * We should call set_dirty_inode to write the dirty inode through write_inode.
404  */
405 static void f2fs_dirty_inode(struct inode *inode, int flags)
406 {
407 	set_inode_flag(F2FS_I(inode), FI_DIRTY_INODE);
408 }
409 
410 static void f2fs_i_callback(struct rcu_head *head)
411 {
412 	struct inode *inode = container_of(head, struct inode, i_rcu);
413 	kmem_cache_free(f2fs_inode_cachep, F2FS_I(inode));
414 }
415 
416 static void f2fs_destroy_inode(struct inode *inode)
417 {
418 	call_rcu(&inode->i_rcu, f2fs_i_callback);
419 }
420 
421 static void f2fs_put_super(struct super_block *sb)
422 {
423 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
424 
425 	if (sbi->s_proc) {
426 		remove_proc_entry("segment_info", sbi->s_proc);
427 		remove_proc_entry(sb->s_id, f2fs_proc_root);
428 	}
429 	kobject_del(&sbi->s_kobj);
430 
431 	f2fs_destroy_stats(sbi);
432 	stop_gc_thread(sbi);
433 
434 	/* We don't need to do checkpoint when it's clean */
435 	if (sbi->s_dirty)
436 		write_checkpoint(sbi, true);
437 
438 	/*
439 	 * normally superblock is clean, so we need to release this.
440 	 * In addition, EIO will skip do checkpoint, we need this as well.
441 	 */
442 	release_dirty_inode(sbi);
443 
444 	iput(sbi->node_inode);
445 	iput(sbi->meta_inode);
446 
447 	/* destroy f2fs internal modules */
448 	destroy_node_manager(sbi);
449 	destroy_segment_manager(sbi);
450 
451 	kfree(sbi->ckpt);
452 	kobject_put(&sbi->s_kobj);
453 	wait_for_completion(&sbi->s_kobj_unregister);
454 
455 	sb->s_fs_info = NULL;
456 	brelse(sbi->raw_super_buf);
457 	kfree(sbi);
458 }
459 
460 int f2fs_sync_fs(struct super_block *sb, int sync)
461 {
462 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
463 
464 	trace_f2fs_sync_fs(sb, sync);
465 
466 	if (sync) {
467 		mutex_lock(&sbi->gc_mutex);
468 		write_checkpoint(sbi, false);
469 		mutex_unlock(&sbi->gc_mutex);
470 	} else {
471 		f2fs_balance_fs(sbi);
472 	}
473 
474 	return 0;
475 }
476 
477 static int f2fs_freeze(struct super_block *sb)
478 {
479 	int err;
480 
481 	if (f2fs_readonly(sb))
482 		return 0;
483 
484 	err = f2fs_sync_fs(sb, 1);
485 	return err;
486 }
487 
488 static int f2fs_unfreeze(struct super_block *sb)
489 {
490 	return 0;
491 }
492 
493 static int f2fs_statfs(struct dentry *dentry, struct kstatfs *buf)
494 {
495 	struct super_block *sb = dentry->d_sb;
496 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
497 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
498 	block_t total_count, user_block_count, start_count, ovp_count;
499 
500 	total_count = le64_to_cpu(sbi->raw_super->block_count);
501 	user_block_count = sbi->user_block_count;
502 	start_count = le32_to_cpu(sbi->raw_super->segment0_blkaddr);
503 	ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
504 	buf->f_type = F2FS_SUPER_MAGIC;
505 	buf->f_bsize = sbi->blocksize;
506 
507 	buf->f_blocks = total_count - start_count;
508 	buf->f_bfree = buf->f_blocks - valid_user_blocks(sbi) - ovp_count;
509 	buf->f_bavail = user_block_count - valid_user_blocks(sbi);
510 
511 	buf->f_files = sbi->total_node_count - F2FS_RESERVED_NODE_NUM;
512 	buf->f_ffree = buf->f_files - valid_inode_count(sbi);
513 
514 	buf->f_namelen = F2FS_NAME_LEN;
515 	buf->f_fsid.val[0] = (u32)id;
516 	buf->f_fsid.val[1] = (u32)(id >> 32);
517 
518 	return 0;
519 }
520 
521 static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
522 {
523 	struct f2fs_sb_info *sbi = F2FS_SB(root->d_sb);
524 
525 	if (!f2fs_readonly(sbi->sb) && test_opt(sbi, BG_GC))
526 		seq_printf(seq, ",background_gc=%s", "on");
527 	else
528 		seq_printf(seq, ",background_gc=%s", "off");
529 	if (test_opt(sbi, DISABLE_ROLL_FORWARD))
530 		seq_puts(seq, ",disable_roll_forward");
531 	if (test_opt(sbi, DISCARD))
532 		seq_puts(seq, ",discard");
533 	if (test_opt(sbi, NOHEAP))
534 		seq_puts(seq, ",no_heap_alloc");
535 #ifdef CONFIG_F2FS_FS_XATTR
536 	if (test_opt(sbi, XATTR_USER))
537 		seq_puts(seq, ",user_xattr");
538 	else
539 		seq_puts(seq, ",nouser_xattr");
540 	if (test_opt(sbi, INLINE_XATTR))
541 		seq_puts(seq, ",inline_xattr");
542 #endif
543 #ifdef CONFIG_F2FS_FS_POSIX_ACL
544 	if (test_opt(sbi, POSIX_ACL))
545 		seq_puts(seq, ",acl");
546 	else
547 		seq_puts(seq, ",noacl");
548 #endif
549 	if (test_opt(sbi, DISABLE_EXT_IDENTIFY))
550 		seq_puts(seq, ",disable_ext_identify");
551 	if (test_opt(sbi, INLINE_DATA))
552 		seq_puts(seq, ",inline_data");
553 	if (!f2fs_readonly(sbi->sb) && test_opt(sbi, FLUSH_MERGE))
554 		seq_puts(seq, ",flush_merge");
555 	if (test_opt(sbi, NOBARRIER))
556 		seq_puts(seq, ",nobarrier");
557 	seq_printf(seq, ",active_logs=%u", sbi->active_logs);
558 
559 	return 0;
560 }
561 
562 static int segment_info_seq_show(struct seq_file *seq, void *offset)
563 {
564 	struct super_block *sb = seq->private;
565 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
566 	unsigned int total_segs =
567 			le32_to_cpu(sbi->raw_super->segment_count_main);
568 	int i;
569 
570 	seq_puts(seq, "format: segment_type|valid_blocks\n"
571 		"segment_type(0:HD, 1:WD, 2:CD, 3:HN, 4:WN, 5:CN)\n");
572 
573 	for (i = 0; i < total_segs; i++) {
574 		struct seg_entry *se = get_seg_entry(sbi, i);
575 
576 		if ((i % 10) == 0)
577 			seq_printf(seq, "%-5d", i);
578 		seq_printf(seq, "%d|%-3u", se->type,
579 					get_valid_blocks(sbi, i, 1));
580 		if ((i % 10) == 9 || i == (total_segs - 1))
581 			seq_putc(seq, '\n');
582 		else
583 			seq_putc(seq, ' ');
584 	}
585 
586 	return 0;
587 }
588 
589 static int segment_info_open_fs(struct inode *inode, struct file *file)
590 {
591 	return single_open(file, segment_info_seq_show, PDE_DATA(inode));
592 }
593 
594 static const struct file_operations f2fs_seq_segment_info_fops = {
595 	.owner = THIS_MODULE,
596 	.open = segment_info_open_fs,
597 	.read = seq_read,
598 	.llseek = seq_lseek,
599 	.release = single_release,
600 };
601 
602 static int f2fs_remount(struct super_block *sb, int *flags, char *data)
603 {
604 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
605 	struct f2fs_mount_info org_mount_opt;
606 	int err, active_logs;
607 	bool need_restart_gc = false;
608 	bool need_stop_gc = false;
609 
610 	sync_filesystem(sb);
611 
612 	/*
613 	 * Save the old mount options in case we
614 	 * need to restore them.
615 	 */
616 	org_mount_opt = sbi->mount_opt;
617 	active_logs = sbi->active_logs;
618 
619 	/* parse mount options */
620 	err = parse_options(sb, data);
621 	if (err)
622 		goto restore_opts;
623 
624 	/*
625 	 * Previous and new state of filesystem is RO,
626 	 * so skip checking GC and FLUSH_MERGE conditions.
627 	 */
628 	if (f2fs_readonly(sb) && (*flags & MS_RDONLY))
629 		goto skip;
630 
631 	/*
632 	 * We stop the GC thread if FS is mounted as RO
633 	 * or if background_gc = off is passed in mount
634 	 * option. Also sync the filesystem.
635 	 */
636 	if ((*flags & MS_RDONLY) || !test_opt(sbi, BG_GC)) {
637 		if (sbi->gc_thread) {
638 			stop_gc_thread(sbi);
639 			f2fs_sync_fs(sb, 1);
640 			need_restart_gc = true;
641 		}
642 	} else if (test_opt(sbi, BG_GC) && !sbi->gc_thread) {
643 		err = start_gc_thread(sbi);
644 		if (err)
645 			goto restore_opts;
646 		need_stop_gc = true;
647 	}
648 
649 	/*
650 	 * We stop issue flush thread if FS is mounted as RO
651 	 * or if flush_merge is not passed in mount option.
652 	 */
653 	if ((*flags & MS_RDONLY) || !test_opt(sbi, FLUSH_MERGE)) {
654 		destroy_flush_cmd_control(sbi);
655 	} else if (test_opt(sbi, FLUSH_MERGE) && !SM_I(sbi)->cmd_control_info) {
656 		err = create_flush_cmd_control(sbi);
657 		if (err)
658 			goto restore_gc;
659 	}
660 skip:
661 	/* Update the POSIXACL Flag */
662 	 sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
663 		(test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0);
664 	return 0;
665 restore_gc:
666 	if (need_restart_gc) {
667 		if (start_gc_thread(sbi))
668 			f2fs_msg(sbi->sb, KERN_WARNING,
669 				"background gc thread has stopped");
670 	} else if (need_stop_gc) {
671 		stop_gc_thread(sbi);
672 	}
673 restore_opts:
674 	sbi->mount_opt = org_mount_opt;
675 	sbi->active_logs = active_logs;
676 	return err;
677 }
678 
679 static struct super_operations f2fs_sops = {
680 	.alloc_inode	= f2fs_alloc_inode,
681 	.drop_inode	= f2fs_drop_inode,
682 	.destroy_inode	= f2fs_destroy_inode,
683 	.write_inode	= f2fs_write_inode,
684 	.dirty_inode	= f2fs_dirty_inode,
685 	.show_options	= f2fs_show_options,
686 	.evict_inode	= f2fs_evict_inode,
687 	.put_super	= f2fs_put_super,
688 	.sync_fs	= f2fs_sync_fs,
689 	.freeze_fs	= f2fs_freeze,
690 	.unfreeze_fs	= f2fs_unfreeze,
691 	.statfs		= f2fs_statfs,
692 	.remount_fs	= f2fs_remount,
693 };
694 
695 static struct inode *f2fs_nfs_get_inode(struct super_block *sb,
696 		u64 ino, u32 generation)
697 {
698 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
699 	struct inode *inode;
700 
701 	if (check_nid_range(sbi, ino))
702 		return ERR_PTR(-ESTALE);
703 
704 	/*
705 	 * f2fs_iget isn't quite right if the inode is currently unallocated!
706 	 * However f2fs_iget currently does appropriate checks to handle stale
707 	 * inodes so everything is OK.
708 	 */
709 	inode = f2fs_iget(sb, ino);
710 	if (IS_ERR(inode))
711 		return ERR_CAST(inode);
712 	if (unlikely(generation && inode->i_generation != generation)) {
713 		/* we didn't find the right inode.. */
714 		iput(inode);
715 		return ERR_PTR(-ESTALE);
716 	}
717 	return inode;
718 }
719 
720 static struct dentry *f2fs_fh_to_dentry(struct super_block *sb, struct fid *fid,
721 		int fh_len, int fh_type)
722 {
723 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
724 				    f2fs_nfs_get_inode);
725 }
726 
727 static struct dentry *f2fs_fh_to_parent(struct super_block *sb, struct fid *fid,
728 		int fh_len, int fh_type)
729 {
730 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
731 				    f2fs_nfs_get_inode);
732 }
733 
734 static const struct export_operations f2fs_export_ops = {
735 	.fh_to_dentry = f2fs_fh_to_dentry,
736 	.fh_to_parent = f2fs_fh_to_parent,
737 	.get_parent = f2fs_get_parent,
738 };
739 
740 static loff_t max_file_size(unsigned bits)
741 {
742 	loff_t result = (DEF_ADDRS_PER_INODE - F2FS_INLINE_XATTR_ADDRS);
743 	loff_t leaf_count = ADDRS_PER_BLOCK;
744 
745 	/* two direct node blocks */
746 	result += (leaf_count * 2);
747 
748 	/* two indirect node blocks */
749 	leaf_count *= NIDS_PER_BLOCK;
750 	result += (leaf_count * 2);
751 
752 	/* one double indirect node block */
753 	leaf_count *= NIDS_PER_BLOCK;
754 	result += leaf_count;
755 
756 	result <<= bits;
757 	return result;
758 }
759 
760 static int sanity_check_raw_super(struct super_block *sb,
761 			struct f2fs_super_block *raw_super)
762 {
763 	unsigned int blocksize;
764 
765 	if (F2FS_SUPER_MAGIC != le32_to_cpu(raw_super->magic)) {
766 		f2fs_msg(sb, KERN_INFO,
767 			"Magic Mismatch, valid(0x%x) - read(0x%x)",
768 			F2FS_SUPER_MAGIC, le32_to_cpu(raw_super->magic));
769 		return 1;
770 	}
771 
772 	/* Currently, support only 4KB page cache size */
773 	if (F2FS_BLKSIZE != PAGE_CACHE_SIZE) {
774 		f2fs_msg(sb, KERN_INFO,
775 			"Invalid page_cache_size (%lu), supports only 4KB\n",
776 			PAGE_CACHE_SIZE);
777 		return 1;
778 	}
779 
780 	/* Currently, support only 4KB block size */
781 	blocksize = 1 << le32_to_cpu(raw_super->log_blocksize);
782 	if (blocksize != F2FS_BLKSIZE) {
783 		f2fs_msg(sb, KERN_INFO,
784 			"Invalid blocksize (%u), supports only 4KB\n",
785 			blocksize);
786 		return 1;
787 	}
788 
789 	if (le32_to_cpu(raw_super->log_sectorsize) !=
790 					F2FS_LOG_SECTOR_SIZE) {
791 		f2fs_msg(sb, KERN_INFO, "Invalid log sectorsize");
792 		return 1;
793 	}
794 	if (le32_to_cpu(raw_super->log_sectors_per_block) !=
795 					F2FS_LOG_SECTORS_PER_BLOCK) {
796 		f2fs_msg(sb, KERN_INFO, "Invalid log sectors per block");
797 		return 1;
798 	}
799 	return 0;
800 }
801 
802 static int sanity_check_ckpt(struct f2fs_sb_info *sbi)
803 {
804 	unsigned int total, fsmeta;
805 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
806 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
807 
808 	total = le32_to_cpu(raw_super->segment_count);
809 	fsmeta = le32_to_cpu(raw_super->segment_count_ckpt);
810 	fsmeta += le32_to_cpu(raw_super->segment_count_sit);
811 	fsmeta += le32_to_cpu(raw_super->segment_count_nat);
812 	fsmeta += le32_to_cpu(ckpt->rsvd_segment_count);
813 	fsmeta += le32_to_cpu(raw_super->segment_count_ssa);
814 
815 	if (unlikely(fsmeta >= total))
816 		return 1;
817 
818 	if (unlikely(f2fs_cp_error(sbi))) {
819 		f2fs_msg(sbi->sb, KERN_ERR, "A bug case: need to run fsck");
820 		return 1;
821 	}
822 	return 0;
823 }
824 
825 static void init_sb_info(struct f2fs_sb_info *sbi)
826 {
827 	struct f2fs_super_block *raw_super = sbi->raw_super;
828 	int i;
829 
830 	sbi->log_sectors_per_block =
831 		le32_to_cpu(raw_super->log_sectors_per_block);
832 	sbi->log_blocksize = le32_to_cpu(raw_super->log_blocksize);
833 	sbi->blocksize = 1 << sbi->log_blocksize;
834 	sbi->log_blocks_per_seg = le32_to_cpu(raw_super->log_blocks_per_seg);
835 	sbi->blocks_per_seg = 1 << sbi->log_blocks_per_seg;
836 	sbi->segs_per_sec = le32_to_cpu(raw_super->segs_per_sec);
837 	sbi->secs_per_zone = le32_to_cpu(raw_super->secs_per_zone);
838 	sbi->total_sections = le32_to_cpu(raw_super->section_count);
839 	sbi->total_node_count =
840 		(le32_to_cpu(raw_super->segment_count_nat) / 2)
841 			* sbi->blocks_per_seg * NAT_ENTRY_PER_BLOCK;
842 	sbi->root_ino_num = le32_to_cpu(raw_super->root_ino);
843 	sbi->node_ino_num = le32_to_cpu(raw_super->node_ino);
844 	sbi->meta_ino_num = le32_to_cpu(raw_super->meta_ino);
845 	sbi->cur_victim_sec = NULL_SECNO;
846 	sbi->max_victim_search = DEF_MAX_VICTIM_SEARCH;
847 
848 	for (i = 0; i < NR_COUNT_TYPE; i++)
849 		atomic_set(&sbi->nr_pages[i], 0);
850 
851 	sbi->dir_level = DEF_DIR_LEVEL;
852 }
853 
854 /*
855  * Read f2fs raw super block.
856  * Because we have two copies of super block, so read the first one at first,
857  * if the first one is invalid, move to read the second one.
858  */
859 static int read_raw_super_block(struct super_block *sb,
860 			struct f2fs_super_block **raw_super,
861 			struct buffer_head **raw_super_buf)
862 {
863 	int block = 0;
864 
865 retry:
866 	*raw_super_buf = sb_bread(sb, block);
867 	if (!*raw_super_buf) {
868 		f2fs_msg(sb, KERN_ERR, "Unable to read %dth superblock",
869 				block + 1);
870 		if (block == 0) {
871 			block++;
872 			goto retry;
873 		} else {
874 			return -EIO;
875 		}
876 	}
877 
878 	*raw_super = (struct f2fs_super_block *)
879 		((char *)(*raw_super_buf)->b_data + F2FS_SUPER_OFFSET);
880 
881 	/* sanity checking of raw super */
882 	if (sanity_check_raw_super(sb, *raw_super)) {
883 		brelse(*raw_super_buf);
884 		f2fs_msg(sb, KERN_ERR,
885 			"Can't find valid F2FS filesystem in %dth superblock",
886 								block + 1);
887 		if (block == 0) {
888 			block++;
889 			goto retry;
890 		} else {
891 			return -EINVAL;
892 		}
893 	}
894 
895 	return 0;
896 }
897 
898 static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
899 {
900 	struct f2fs_sb_info *sbi;
901 	struct f2fs_super_block *raw_super;
902 	struct buffer_head *raw_super_buf;
903 	struct inode *root;
904 	long err = -EINVAL;
905 	bool retry = true;
906 	int i;
907 
908 try_onemore:
909 	/* allocate memory for f2fs-specific super block info */
910 	sbi = kzalloc(sizeof(struct f2fs_sb_info), GFP_KERNEL);
911 	if (!sbi)
912 		return -ENOMEM;
913 
914 	/* set a block size */
915 	if (unlikely(!sb_set_blocksize(sb, F2FS_BLKSIZE))) {
916 		f2fs_msg(sb, KERN_ERR, "unable to set blocksize");
917 		goto free_sbi;
918 	}
919 
920 	err = read_raw_super_block(sb, &raw_super, &raw_super_buf);
921 	if (err)
922 		goto free_sbi;
923 
924 	sb->s_fs_info = sbi;
925 	/* init some FS parameters */
926 	sbi->active_logs = NR_CURSEG_TYPE;
927 
928 	set_opt(sbi, BG_GC);
929 
930 #ifdef CONFIG_F2FS_FS_XATTR
931 	set_opt(sbi, XATTR_USER);
932 #endif
933 #ifdef CONFIG_F2FS_FS_POSIX_ACL
934 	set_opt(sbi, POSIX_ACL);
935 #endif
936 	/* parse mount options */
937 	err = parse_options(sb, (char *)data);
938 	if (err)
939 		goto free_sb_buf;
940 
941 	sb->s_maxbytes = max_file_size(le32_to_cpu(raw_super->log_blocksize));
942 	sb->s_max_links = F2FS_LINK_MAX;
943 	get_random_bytes(&sbi->s_next_generation, sizeof(u32));
944 
945 	sb->s_op = &f2fs_sops;
946 	sb->s_xattr = f2fs_xattr_handlers;
947 	sb->s_export_op = &f2fs_export_ops;
948 	sb->s_magic = F2FS_SUPER_MAGIC;
949 	sb->s_time_gran = 1;
950 	sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
951 		(test_opt(sbi, POSIX_ACL) ? MS_POSIXACL : 0);
952 	memcpy(sb->s_uuid, raw_super->uuid, sizeof(raw_super->uuid));
953 
954 	/* init f2fs-specific super block info */
955 	sbi->sb = sb;
956 	sbi->raw_super = raw_super;
957 	sbi->raw_super_buf = raw_super_buf;
958 	mutex_init(&sbi->gc_mutex);
959 	mutex_init(&sbi->writepages);
960 	mutex_init(&sbi->cp_mutex);
961 	init_rwsem(&sbi->node_write);
962 	sbi->por_doing = false;
963 	spin_lock_init(&sbi->stat_lock);
964 
965 	init_rwsem(&sbi->read_io.io_rwsem);
966 	sbi->read_io.sbi = sbi;
967 	sbi->read_io.bio = NULL;
968 	for (i = 0; i < NR_PAGE_TYPE; i++) {
969 		init_rwsem(&sbi->write_io[i].io_rwsem);
970 		sbi->write_io[i].sbi = sbi;
971 		sbi->write_io[i].bio = NULL;
972 	}
973 
974 	init_rwsem(&sbi->cp_rwsem);
975 	init_waitqueue_head(&sbi->cp_wait);
976 	init_sb_info(sbi);
977 
978 	/* get an inode for meta space */
979 	sbi->meta_inode = f2fs_iget(sb, F2FS_META_INO(sbi));
980 	if (IS_ERR(sbi->meta_inode)) {
981 		f2fs_msg(sb, KERN_ERR, "Failed to read F2FS meta data inode");
982 		err = PTR_ERR(sbi->meta_inode);
983 		goto free_sb_buf;
984 	}
985 
986 	err = get_valid_checkpoint(sbi);
987 	if (err) {
988 		f2fs_msg(sb, KERN_ERR, "Failed to get valid F2FS checkpoint");
989 		goto free_meta_inode;
990 	}
991 
992 	/* sanity checking of checkpoint */
993 	err = -EINVAL;
994 	if (sanity_check_ckpt(sbi)) {
995 		f2fs_msg(sb, KERN_ERR, "Invalid F2FS checkpoint");
996 		goto free_cp;
997 	}
998 
999 	sbi->total_valid_node_count =
1000 				le32_to_cpu(sbi->ckpt->valid_node_count);
1001 	sbi->total_valid_inode_count =
1002 				le32_to_cpu(sbi->ckpt->valid_inode_count);
1003 	sbi->user_block_count = le64_to_cpu(sbi->ckpt->user_block_count);
1004 	sbi->total_valid_block_count =
1005 				le64_to_cpu(sbi->ckpt->valid_block_count);
1006 	sbi->last_valid_block_count = sbi->total_valid_block_count;
1007 	sbi->alloc_valid_block_count = 0;
1008 	INIT_LIST_HEAD(&sbi->dir_inode_list);
1009 	spin_lock_init(&sbi->dir_inode_lock);
1010 
1011 	init_ino_entry_info(sbi);
1012 
1013 	/* setup f2fs internal modules */
1014 	err = build_segment_manager(sbi);
1015 	if (err) {
1016 		f2fs_msg(sb, KERN_ERR,
1017 			"Failed to initialize F2FS segment manager");
1018 		goto free_sm;
1019 	}
1020 	err = build_node_manager(sbi);
1021 	if (err) {
1022 		f2fs_msg(sb, KERN_ERR,
1023 			"Failed to initialize F2FS node manager");
1024 		goto free_nm;
1025 	}
1026 
1027 	build_gc_manager(sbi);
1028 
1029 	/* get an inode for node space */
1030 	sbi->node_inode = f2fs_iget(sb, F2FS_NODE_INO(sbi));
1031 	if (IS_ERR(sbi->node_inode)) {
1032 		f2fs_msg(sb, KERN_ERR, "Failed to read node inode");
1033 		err = PTR_ERR(sbi->node_inode);
1034 		goto free_nm;
1035 	}
1036 
1037 	/* if there are nt orphan nodes free them */
1038 	recover_orphan_inodes(sbi);
1039 
1040 	/* read root inode and dentry */
1041 	root = f2fs_iget(sb, F2FS_ROOT_INO(sbi));
1042 	if (IS_ERR(root)) {
1043 		f2fs_msg(sb, KERN_ERR, "Failed to read root inode");
1044 		err = PTR_ERR(root);
1045 		goto free_node_inode;
1046 	}
1047 	if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1048 		iput(root);
1049 		err = -EINVAL;
1050 		goto free_node_inode;
1051 	}
1052 
1053 	sb->s_root = d_make_root(root); /* allocate root dentry */
1054 	if (!sb->s_root) {
1055 		err = -ENOMEM;
1056 		goto free_root_inode;
1057 	}
1058 
1059 	err = f2fs_build_stats(sbi);
1060 	if (err)
1061 		goto free_root_inode;
1062 
1063 	if (f2fs_proc_root)
1064 		sbi->s_proc = proc_mkdir(sb->s_id, f2fs_proc_root);
1065 
1066 	if (sbi->s_proc)
1067 		proc_create_data("segment_info", S_IRUGO, sbi->s_proc,
1068 				 &f2fs_seq_segment_info_fops, sb);
1069 
1070 	if (test_opt(sbi, DISCARD)) {
1071 		struct request_queue *q = bdev_get_queue(sb->s_bdev);
1072 		if (!blk_queue_discard(q))
1073 			f2fs_msg(sb, KERN_WARNING,
1074 					"mounting with \"discard\" option, but "
1075 					"the device does not support discard");
1076 	}
1077 
1078 	sbi->s_kobj.kset = f2fs_kset;
1079 	init_completion(&sbi->s_kobj_unregister);
1080 	err = kobject_init_and_add(&sbi->s_kobj, &f2fs_ktype, NULL,
1081 							"%s", sb->s_id);
1082 	if (err)
1083 		goto free_proc;
1084 
1085 	/* recover fsynced data */
1086 	if (!test_opt(sbi, DISABLE_ROLL_FORWARD)) {
1087 		err = recover_fsync_data(sbi);
1088 		if (err) {
1089 			f2fs_msg(sb, KERN_ERR,
1090 				"Cannot recover all fsync data errno=%ld", err);
1091 			goto free_kobj;
1092 		}
1093 	}
1094 
1095 	/*
1096 	 * If filesystem is not mounted as read-only then
1097 	 * do start the gc_thread.
1098 	 */
1099 	if (!f2fs_readonly(sb)) {
1100 		/* After POR, we can run background GC thread.*/
1101 		err = start_gc_thread(sbi);
1102 		if (err)
1103 			goto free_kobj;
1104 	}
1105 	return 0;
1106 
1107 free_kobj:
1108 	kobject_del(&sbi->s_kobj);
1109 free_proc:
1110 	if (sbi->s_proc) {
1111 		remove_proc_entry("segment_info", sbi->s_proc);
1112 		remove_proc_entry(sb->s_id, f2fs_proc_root);
1113 	}
1114 	f2fs_destroy_stats(sbi);
1115 free_root_inode:
1116 	dput(sb->s_root);
1117 	sb->s_root = NULL;
1118 free_node_inode:
1119 	iput(sbi->node_inode);
1120 free_nm:
1121 	destroy_node_manager(sbi);
1122 free_sm:
1123 	destroy_segment_manager(sbi);
1124 free_cp:
1125 	kfree(sbi->ckpt);
1126 free_meta_inode:
1127 	make_bad_inode(sbi->meta_inode);
1128 	iput(sbi->meta_inode);
1129 free_sb_buf:
1130 	brelse(raw_super_buf);
1131 free_sbi:
1132 	kfree(sbi);
1133 
1134 	/* give only one another chance */
1135 	if (retry) {
1136 		retry = 0;
1137 		shrink_dcache_sb(sb);
1138 		goto try_onemore;
1139 	}
1140 	return err;
1141 }
1142 
1143 static struct dentry *f2fs_mount(struct file_system_type *fs_type, int flags,
1144 			const char *dev_name, void *data)
1145 {
1146 	return mount_bdev(fs_type, flags, dev_name, data, f2fs_fill_super);
1147 }
1148 
1149 static struct file_system_type f2fs_fs_type = {
1150 	.owner		= THIS_MODULE,
1151 	.name		= "f2fs",
1152 	.mount		= f2fs_mount,
1153 	.kill_sb	= kill_block_super,
1154 	.fs_flags	= FS_REQUIRES_DEV,
1155 };
1156 MODULE_ALIAS_FS("f2fs");
1157 
1158 static int __init init_inodecache(void)
1159 {
1160 	f2fs_inode_cachep = f2fs_kmem_cache_create("f2fs_inode_cache",
1161 			sizeof(struct f2fs_inode_info));
1162 	if (!f2fs_inode_cachep)
1163 		return -ENOMEM;
1164 	return 0;
1165 }
1166 
1167 static void destroy_inodecache(void)
1168 {
1169 	/*
1170 	 * Make sure all delayed rcu free inodes are flushed before we
1171 	 * destroy cache.
1172 	 */
1173 	rcu_barrier();
1174 	kmem_cache_destroy(f2fs_inode_cachep);
1175 }
1176 
1177 static int __init init_f2fs_fs(void)
1178 {
1179 	int err;
1180 
1181 	err = init_inodecache();
1182 	if (err)
1183 		goto fail;
1184 	err = create_node_manager_caches();
1185 	if (err)
1186 		goto free_inodecache;
1187 	err = create_segment_manager_caches();
1188 	if (err)
1189 		goto free_node_manager_caches;
1190 	err = create_gc_caches();
1191 	if (err)
1192 		goto free_segment_manager_caches;
1193 	err = create_checkpoint_caches();
1194 	if (err)
1195 		goto free_gc_caches;
1196 	f2fs_kset = kset_create_and_add("f2fs", NULL, fs_kobj);
1197 	if (!f2fs_kset) {
1198 		err = -ENOMEM;
1199 		goto free_checkpoint_caches;
1200 	}
1201 	err = register_filesystem(&f2fs_fs_type);
1202 	if (err)
1203 		goto free_kset;
1204 	f2fs_create_root_stats();
1205 	f2fs_proc_root = proc_mkdir("fs/f2fs", NULL);
1206 	return 0;
1207 
1208 free_kset:
1209 	kset_unregister(f2fs_kset);
1210 free_checkpoint_caches:
1211 	destroy_checkpoint_caches();
1212 free_gc_caches:
1213 	destroy_gc_caches();
1214 free_segment_manager_caches:
1215 	destroy_segment_manager_caches();
1216 free_node_manager_caches:
1217 	destroy_node_manager_caches();
1218 free_inodecache:
1219 	destroy_inodecache();
1220 fail:
1221 	return err;
1222 }
1223 
1224 static void __exit exit_f2fs_fs(void)
1225 {
1226 	remove_proc_entry("fs/f2fs", NULL);
1227 	f2fs_destroy_root_stats();
1228 	unregister_filesystem(&f2fs_fs_type);
1229 	destroy_checkpoint_caches();
1230 	destroy_gc_caches();
1231 	destroy_segment_manager_caches();
1232 	destroy_node_manager_caches();
1233 	destroy_inodecache();
1234 	kset_unregister(f2fs_kset);
1235 }
1236 
1237 module_init(init_f2fs_fs)
1238 module_exit(exit_f2fs_fs)
1239 
1240 MODULE_AUTHOR("Samsung Electronics's Praesto Team");
1241 MODULE_DESCRIPTION("Flash Friendly File System");
1242 MODULE_LICENSE("GPL");
1243