Linux Audio

Check our new training course

Loading...
v4.17
  1/*
  2 * the_nilfs.c - the_nilfs shared structure.
  3 *
  4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  5 *
  6 * This program is free software; you can redistribute it and/or modify
  7 * it under the terms of the GNU General Public License as published by
  8 * the Free Software Foundation; either version 2 of the License, or
  9 * (at your option) any later version.
 10 *
 11 * This program is distributed in the hope that it will be useful,
 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 14 * GNU General Public License for more details.
 15 *
 16 * Written by Ryusuke Konishi.
 
 
 
 
 17 *
 18 */
 19
 20#include <linux/buffer_head.h>
 21#include <linux/slab.h>
 22#include <linux/blkdev.h>
 23#include <linux/backing-dev.h>
 24#include <linux/random.h>
 25#include <linux/crc32.h>
 26#include "nilfs.h"
 27#include "segment.h"
 28#include "alloc.h"
 29#include "cpfile.h"
 30#include "sufile.h"
 31#include "dat.h"
 32#include "segbuf.h"
 33
 34
 35static int nilfs_valid_sb(struct nilfs_super_block *sbp);
 36
 37void nilfs_set_last_segment(struct the_nilfs *nilfs,
 38			    sector_t start_blocknr, u64 seq, __u64 cno)
 39{
 40	spin_lock(&nilfs->ns_last_segment_lock);
 41	nilfs->ns_last_pseg = start_blocknr;
 42	nilfs->ns_last_seq = seq;
 43	nilfs->ns_last_cno = cno;
 44
 45	if (!nilfs_sb_dirty(nilfs)) {
 46		if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
 47			goto stay_cursor;
 48
 49		set_nilfs_sb_dirty(nilfs);
 50	}
 51	nilfs->ns_prev_seq = nilfs->ns_last_seq;
 52
 53 stay_cursor:
 54	spin_unlock(&nilfs->ns_last_segment_lock);
 55}
 56
 57/**
 58 * alloc_nilfs - allocate a nilfs object
 59 * @sb: super block instance
 60 *
 61 * Return Value: On success, pointer to the_nilfs is returned.
 62 * On error, NULL is returned.
 63 */
 64struct the_nilfs *alloc_nilfs(struct super_block *sb)
 65{
 66	struct the_nilfs *nilfs;
 67
 68	nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
 69	if (!nilfs)
 70		return NULL;
 71
 72	nilfs->ns_sb = sb;
 73	nilfs->ns_bdev = sb->s_bdev;
 74	atomic_set(&nilfs->ns_ndirtyblks, 0);
 75	init_rwsem(&nilfs->ns_sem);
 76	mutex_init(&nilfs->ns_snapshot_mount_mutex);
 77	INIT_LIST_HEAD(&nilfs->ns_dirty_files);
 78	INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
 79	spin_lock_init(&nilfs->ns_inode_lock);
 80	spin_lock_init(&nilfs->ns_next_gen_lock);
 81	spin_lock_init(&nilfs->ns_last_segment_lock);
 82	nilfs->ns_cptree = RB_ROOT;
 83	spin_lock_init(&nilfs->ns_cptree_lock);
 84	init_rwsem(&nilfs->ns_segctor_sem);
 85	nilfs->ns_sb_update_freq = NILFS_SB_FREQ;
 86
 87	return nilfs;
 88}
 89
 90/**
 91 * destroy_nilfs - destroy nilfs object
 92 * @nilfs: nilfs object to be released
 93 */
 94void destroy_nilfs(struct the_nilfs *nilfs)
 95{
 96	might_sleep();
 97	if (nilfs_init(nilfs)) {
 98		nilfs_sysfs_delete_device_group(nilfs);
 99		brelse(nilfs->ns_sbh[0]);
100		brelse(nilfs->ns_sbh[1]);
101	}
102	kfree(nilfs);
103}
104
105static int nilfs_load_super_root(struct the_nilfs *nilfs,
106				 struct super_block *sb, sector_t sr_block)
107{
108	struct buffer_head *bh_sr;
109	struct nilfs_super_root *raw_sr;
110	struct nilfs_super_block **sbp = nilfs->ns_sbp;
111	struct nilfs_inode *rawi;
112	unsigned int dat_entry_size, segment_usage_size, checkpoint_size;
113	unsigned int inode_size;
114	int err;
115
116	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
117	if (unlikely(err))
118		return err;
119
120	down_read(&nilfs->ns_sem);
121	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
122	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
123	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
124	up_read(&nilfs->ns_sem);
125
126	inode_size = nilfs->ns_inode_size;
127
128	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
129	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
130	if (err)
131		goto failed;
132
133	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
134	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
135	if (err)
136		goto failed_dat;
137
138	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
139	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
140				&nilfs->ns_sufile);
141	if (err)
142		goto failed_cpfile;
143
144	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
145	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
146
147 failed:
148	brelse(bh_sr);
149	return err;
150
151 failed_cpfile:
152	iput(nilfs->ns_cpfile);
153
154 failed_dat:
155	iput(nilfs->ns_dat);
156	goto failed;
157}
158
159static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
160{
161	memset(ri, 0, sizeof(*ri));
162	INIT_LIST_HEAD(&ri->ri_used_segments);
163}
164
165static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
166{
167	nilfs_dispose_segment_list(&ri->ri_used_segments);
168}
169
170/**
171 * nilfs_store_log_cursor - load log cursor from a super block
172 * @nilfs: nilfs object
173 * @sbp: buffer storing super block to be read
174 *
175 * nilfs_store_log_cursor() reads the last position of the log
176 * containing a super root from a given super block, and initializes
177 * relevant information on the nilfs object preparatory for log
178 * scanning and recovery.
179 */
180static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
181				  struct nilfs_super_block *sbp)
182{
183	int ret = 0;
184
185	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
186	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
187	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
188
189	nilfs->ns_prev_seq = nilfs->ns_last_seq;
190	nilfs->ns_seg_seq = nilfs->ns_last_seq;
191	nilfs->ns_segnum =
192		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
193	nilfs->ns_cno = nilfs->ns_last_cno + 1;
194	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
195		nilfs_msg(nilfs->ns_sb, KERN_ERR,
196			  "pointed segment number is out of range: segnum=%llu, nsegments=%lu",
197			  (unsigned long long)nilfs->ns_segnum,
198			  nilfs->ns_nsegments);
199		ret = -EINVAL;
200	}
201	return ret;
202}
203
204/**
205 * load_nilfs - load and recover the nilfs
206 * @nilfs: the_nilfs structure to be released
207 * @sb: super block isntance used to recover past segment
208 *
209 * load_nilfs() searches and load the latest super root,
210 * attaches the last segment, and does recovery if needed.
211 * The caller must call this exclusively for simultaneous mounts.
212 */
213int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
214{
215	struct nilfs_recovery_info ri;
216	unsigned int s_flags = sb->s_flags;
217	int really_read_only = bdev_read_only(nilfs->ns_bdev);
218	int valid_fs = nilfs_valid_fs(nilfs);
219	int err;
220
221	if (!valid_fs) {
222		nilfs_msg(sb, KERN_WARNING, "mounting unchecked fs");
223		if (s_flags & SB_RDONLY) {
224			nilfs_msg(sb, KERN_INFO,
225				  "recovery required for readonly filesystem");
226			nilfs_msg(sb, KERN_INFO,
227				  "write access will be enabled during recovery");
228		}
229	}
230
231	nilfs_init_recovery_info(&ri);
232
233	err = nilfs_search_super_root(nilfs, &ri);
234	if (unlikely(err)) {
235		struct nilfs_super_block **sbp = nilfs->ns_sbp;
236		int blocksize;
237
238		if (err != -EINVAL)
239			goto scan_error;
240
241		if (!nilfs_valid_sb(sbp[1])) {
242			nilfs_msg(sb, KERN_WARNING,
243				  "unable to fall back to spare super block");
 
244			goto scan_error;
245		}
246		nilfs_msg(sb, KERN_INFO,
247			  "trying rollback from an earlier position");
248
249		/*
250		 * restore super block with its spare and reconfigure
251		 * relevant states of the nilfs object.
252		 */
253		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
254		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
255		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
256
257		/* verify consistency between two super blocks */
258		blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
259		if (blocksize != nilfs->ns_blocksize) {
260			nilfs_msg(sb, KERN_WARNING,
261				  "blocksize differs between two super blocks (%d != %d)",
262				  blocksize, nilfs->ns_blocksize);
 
263			goto scan_error;
264		}
265
266		err = nilfs_store_log_cursor(nilfs, sbp[0]);
267		if (err)
268			goto scan_error;
269
270		/* drop clean flag to allow roll-forward and recovery */
271		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
272		valid_fs = 0;
273
274		err = nilfs_search_super_root(nilfs, &ri);
275		if (err)
276			goto scan_error;
277	}
278
279	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
280	if (unlikely(err)) {
281		nilfs_msg(sb, KERN_ERR, "error %d while loading super root",
282			  err);
283		goto failed;
284	}
285
286	if (valid_fs)
287		goto skip_recovery;
288
289	if (s_flags & SB_RDONLY) {
290		__u64 features;
291
292		if (nilfs_test_opt(nilfs, NORECOVERY)) {
293			nilfs_msg(sb, KERN_INFO,
294				  "norecovery option specified, skipping roll-forward recovery");
295			goto skip_recovery;
296		}
297		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
298			~NILFS_FEATURE_COMPAT_RO_SUPP;
299		if (features) {
300			nilfs_msg(sb, KERN_ERR,
301				  "couldn't proceed with recovery because of unsupported optional features (%llx)",
302				  (unsigned long long)features);
 
303			err = -EROFS;
304			goto failed_unload;
305		}
306		if (really_read_only) {
307			nilfs_msg(sb, KERN_ERR,
308				  "write access unavailable, cannot proceed");
309			err = -EROFS;
310			goto failed_unload;
311		}
312		sb->s_flags &= ~SB_RDONLY;
313	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
314		nilfs_msg(sb, KERN_ERR,
315			  "recovery cancelled because norecovery option was specified for a read/write mount");
316		err = -EINVAL;
317		goto failed_unload;
318	}
319
320	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
321	if (err)
322		goto failed_unload;
323
324	down_write(&nilfs->ns_sem);
325	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
326	err = nilfs_cleanup_super(sb);
327	up_write(&nilfs->ns_sem);
328
329	if (err) {
330		nilfs_msg(sb, KERN_ERR,
331			  "error %d updating super block. recovery unfinished.",
332			  err);
333		goto failed_unload;
334	}
335	nilfs_msg(sb, KERN_INFO, "recovery complete");
336
337 skip_recovery:
338	nilfs_clear_recovery_info(&ri);
339	sb->s_flags = s_flags;
340	return 0;
341
342 scan_error:
343	nilfs_msg(sb, KERN_ERR, "error %d while searching super root", err);
344	goto failed;
345
346 failed_unload:
347	iput(nilfs->ns_cpfile);
348	iput(nilfs->ns_sufile);
349	iput(nilfs->ns_dat);
350
351 failed:
352	nilfs_clear_recovery_info(&ri);
353	sb->s_flags = s_flags;
354	return err;
355}
356
357static unsigned long long nilfs_max_size(unsigned int blkbits)
358{
359	unsigned int max_bits;
360	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
361
362	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
363	if (max_bits < 64)
364		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
365	return res;
366}
367
368/**
369 * nilfs_nrsvsegs - calculate the number of reserved segments
370 * @nilfs: nilfs object
371 * @nsegs: total number of segments
372 */
373unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
374{
375	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
376		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
377				  100));
378}
379
380void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
381{
382	nilfs->ns_nsegments = nsegs;
383	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
384}
385
386static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
387				   struct nilfs_super_block *sbp)
388{
389	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
390		nilfs_msg(nilfs->ns_sb, KERN_ERR,
391			  "unsupported revision (superblock rev.=%d.%d, current rev.=%d.%d). Please check the version of mkfs.nilfs(2).",
392			  le32_to_cpu(sbp->s_rev_level),
393			  le16_to_cpu(sbp->s_minor_rev_level),
394			  NILFS_CURRENT_REV, NILFS_MINOR_REV);
 
395		return -EINVAL;
396	}
397	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
398	if (nilfs->ns_sbsize > BLOCK_SIZE)
399		return -EINVAL;
400
401	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
402	if (nilfs->ns_inode_size > nilfs->ns_blocksize) {
403		nilfs_msg(nilfs->ns_sb, KERN_ERR,
404			  "too large inode size: %d bytes",
405			  nilfs->ns_inode_size);
406		return -EINVAL;
407	} else if (nilfs->ns_inode_size < NILFS_MIN_INODE_SIZE) {
408		nilfs_msg(nilfs->ns_sb, KERN_ERR,
409			  "too small inode size: %d bytes",
410			  nilfs->ns_inode_size);
411		return -EINVAL;
412	}
413
414	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
415
416	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
417	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
418		nilfs_msg(nilfs->ns_sb, KERN_ERR,
419			  "too short segment: %lu blocks",
420			  nilfs->ns_blocks_per_segment);
421		return -EINVAL;
422	}
423
424	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
425	nilfs->ns_r_segments_percentage =
426		le32_to_cpu(sbp->s_r_segments_percentage);
427	if (nilfs->ns_r_segments_percentage < 1 ||
428	    nilfs->ns_r_segments_percentage > 99) {
429		nilfs_msg(nilfs->ns_sb, KERN_ERR,
430			  "invalid reserved segments percentage: %lu",
431			  nilfs->ns_r_segments_percentage);
432		return -EINVAL;
433	}
434
435	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
436	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
437	return 0;
438}
439
440static int nilfs_valid_sb(struct nilfs_super_block *sbp)
441{
442	static unsigned char sum[4];
443	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
444	size_t bytes;
445	u32 crc;
446
447	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
448		return 0;
449	bytes = le16_to_cpu(sbp->s_bytes);
450	if (bytes < sumoff + 4 || bytes > BLOCK_SIZE)
451		return 0;
452	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
453		       sumoff);
454	crc = crc32_le(crc, sum, 4);
455	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
456		       bytes - sumoff - 4);
457	return crc == le32_to_cpu(sbp->s_sum);
458}
459
460static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
461{
462	return offset < ((le64_to_cpu(sbp->s_nsegments) *
463			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
464			 (le32_to_cpu(sbp->s_log_block_size) + 10));
465}
466
467static void nilfs_release_super_block(struct the_nilfs *nilfs)
468{
469	int i;
470
471	for (i = 0; i < 2; i++) {
472		if (nilfs->ns_sbp[i]) {
473			brelse(nilfs->ns_sbh[i]);
474			nilfs->ns_sbh[i] = NULL;
475			nilfs->ns_sbp[i] = NULL;
476		}
477	}
478}
479
480void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
481{
482	brelse(nilfs->ns_sbh[0]);
483	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
484	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
485	nilfs->ns_sbh[1] = NULL;
486	nilfs->ns_sbp[1] = NULL;
487}
488
489void nilfs_swap_super_block(struct the_nilfs *nilfs)
490{
491	struct buffer_head *tsbh = nilfs->ns_sbh[0];
492	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
493
494	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
495	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
496	nilfs->ns_sbh[1] = tsbh;
497	nilfs->ns_sbp[1] = tsbp;
498}
499
500static int nilfs_load_super_block(struct the_nilfs *nilfs,
501				  struct super_block *sb, int blocksize,
502				  struct nilfs_super_block **sbpp)
503{
504	struct nilfs_super_block **sbp = nilfs->ns_sbp;
505	struct buffer_head **sbh = nilfs->ns_sbh;
506	u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
507	int valid[2], swp = 0;
508
509	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
510					&sbh[0]);
511	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
512
513	if (!sbp[0]) {
514		if (!sbp[1]) {
515			nilfs_msg(sb, KERN_ERR, "unable to read superblock");
516			return -EIO;
517		}
518		nilfs_msg(sb, KERN_WARNING,
519			  "unable to read primary superblock (blocksize = %d)",
520			  blocksize);
521	} else if (!sbp[1]) {
522		nilfs_msg(sb, KERN_WARNING,
523			  "unable to read secondary superblock (blocksize = %d)",
524			  blocksize);
525	}
526
527	/*
528	 * Compare two super blocks and set 1 in swp if the secondary
529	 * super block is valid and newer.  Otherwise, set 0 in swp.
530	 */
531	valid[0] = nilfs_valid_sb(sbp[0]);
532	valid[1] = nilfs_valid_sb(sbp[1]);
533	swp = valid[1] && (!valid[0] ||
534			   le64_to_cpu(sbp[1]->s_last_cno) >
535			   le64_to_cpu(sbp[0]->s_last_cno));
536
537	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
538		brelse(sbh[1]);
539		sbh[1] = NULL;
540		sbp[1] = NULL;
541		valid[1] = 0;
542		swp = 0;
543	}
544	if (!valid[swp]) {
545		nilfs_release_super_block(nilfs);
546		nilfs_msg(sb, KERN_ERR, "couldn't find nilfs on the device");
 
547		return -EINVAL;
548	}
549
550	if (!valid[!swp])
551		nilfs_msg(sb, KERN_WARNING,
552			  "broken superblock, retrying with spare superblock (blocksize = %d)",
553			  blocksize);
554	if (swp)
555		nilfs_swap_super_block(nilfs);
556
557	nilfs->ns_sbwcount = 0;
558	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
559	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
560	*sbpp = sbp[0];
561	return 0;
562}
563
564/**
565 * init_nilfs - initialize a NILFS instance.
566 * @nilfs: the_nilfs structure
567 * @sb: super block
568 * @data: mount options
569 *
570 * init_nilfs() performs common initialization per block device (e.g.
571 * reading the super block, getting disk layout information, initializing
572 * shared fields in the_nilfs).
573 *
574 * Return Value: On success, 0 is returned. On error, a negative error
575 * code is returned.
576 */
577int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
578{
579	struct nilfs_super_block *sbp;
580	int blocksize;
581	int err;
582
583	down_write(&nilfs->ns_sem);
584
585	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
586	if (!blocksize) {
587		nilfs_msg(sb, KERN_ERR, "unable to set blocksize");
588		err = -EINVAL;
589		goto out;
590	}
591	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
592	if (err)
593		goto out;
594
595	err = nilfs_store_magic_and_option(sb, sbp, data);
596	if (err)
597		goto failed_sbh;
598
599	err = nilfs_check_feature_compatibility(sb, sbp);
600	if (err)
601		goto failed_sbh;
602
603	blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
604	if (blocksize < NILFS_MIN_BLOCK_SIZE ||
605	    blocksize > NILFS_MAX_BLOCK_SIZE) {
606		nilfs_msg(sb, KERN_ERR,
607			  "couldn't mount because of unsupported filesystem blocksize %d",
608			  blocksize);
609		err = -EINVAL;
610		goto failed_sbh;
611	}
612	if (sb->s_blocksize != blocksize) {
613		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
614
615		if (blocksize < hw_blocksize) {
616			nilfs_msg(sb, KERN_ERR,
617				  "blocksize %d too small for device (sector-size = %d)",
618				  blocksize, hw_blocksize);
 
619			err = -EINVAL;
620			goto failed_sbh;
621		}
622		nilfs_release_super_block(nilfs);
623		sb_set_blocksize(sb, blocksize);
624
625		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
626		if (err)
627			goto out;
628			/*
629			 * Not to failed_sbh; sbh is released automatically
630			 * when reloading fails.
631			 */
632	}
633	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
634	nilfs->ns_blocksize = blocksize;
635
636	get_random_bytes(&nilfs->ns_next_generation,
637			 sizeof(nilfs->ns_next_generation));
638
639	err = nilfs_store_disk_layout(nilfs, sbp);
640	if (err)
641		goto failed_sbh;
642
643	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
644
645	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
646
647	err = nilfs_store_log_cursor(nilfs, sbp);
648	if (err)
649		goto failed_sbh;
650
651	err = nilfs_sysfs_create_device_group(sb);
652	if (err)
653		goto failed_sbh;
654
655	set_nilfs_init(nilfs);
656	err = 0;
657 out:
658	up_write(&nilfs->ns_sem);
659	return err;
660
661 failed_sbh:
662	nilfs_release_super_block(nilfs);
663	goto out;
664}
665
666int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
667			    size_t nsegs)
668{
669	sector_t seg_start, seg_end;
670	sector_t start = 0, nblocks = 0;
671	unsigned int sects_per_block;
672	__u64 *sn;
673	int ret = 0;
674
675	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
676		bdev_logical_block_size(nilfs->ns_bdev);
677	for (sn = segnump; sn < segnump + nsegs; sn++) {
678		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
679
680		if (!nblocks) {
681			start = seg_start;
682			nblocks = seg_end - seg_start + 1;
683		} else if (start + nblocks == seg_start) {
684			nblocks += seg_end - seg_start + 1;
685		} else {
686			ret = blkdev_issue_discard(nilfs->ns_bdev,
687						   start * sects_per_block,
688						   nblocks * sects_per_block,
689						   GFP_NOFS, 0);
690			if (ret < 0)
691				return ret;
692			nblocks = 0;
693		}
694	}
695	if (nblocks)
696		ret = blkdev_issue_discard(nilfs->ns_bdev,
697					   start * sects_per_block,
698					   nblocks * sects_per_block,
699					   GFP_NOFS, 0);
700	return ret;
701}
702
703int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
704{
705	unsigned long ncleansegs;
706
707	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
708	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
709	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
710	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
711	return 0;
712}
713
714int nilfs_near_disk_full(struct the_nilfs *nilfs)
715{
716	unsigned long ncleansegs, nincsegs;
717
718	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
719	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
720		nilfs->ns_blocks_per_segment + 1;
721
722	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
723}
724
725struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
726{
727	struct rb_node *n;
728	struct nilfs_root *root;
729
730	spin_lock(&nilfs->ns_cptree_lock);
731	n = nilfs->ns_cptree.rb_node;
732	while (n) {
733		root = rb_entry(n, struct nilfs_root, rb_node);
734
735		if (cno < root->cno) {
736			n = n->rb_left;
737		} else if (cno > root->cno) {
738			n = n->rb_right;
739		} else {
740			refcount_inc(&root->count);
741			spin_unlock(&nilfs->ns_cptree_lock);
742			return root;
743		}
744	}
745	spin_unlock(&nilfs->ns_cptree_lock);
746
747	return NULL;
748}
749
750struct nilfs_root *
751nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
752{
753	struct rb_node **p, *parent;
754	struct nilfs_root *root, *new;
755	int err;
756
757	root = nilfs_lookup_root(nilfs, cno);
758	if (root)
759		return root;
760
761	new = kzalloc(sizeof(*root), GFP_KERNEL);
762	if (!new)
763		return NULL;
764
765	spin_lock(&nilfs->ns_cptree_lock);
766
767	p = &nilfs->ns_cptree.rb_node;
768	parent = NULL;
769
770	while (*p) {
771		parent = *p;
772		root = rb_entry(parent, struct nilfs_root, rb_node);
773
774		if (cno < root->cno) {
775			p = &(*p)->rb_left;
776		} else if (cno > root->cno) {
777			p = &(*p)->rb_right;
778		} else {
779			refcount_inc(&root->count);
780			spin_unlock(&nilfs->ns_cptree_lock);
781			kfree(new);
782			return root;
783		}
784	}
785
786	new->cno = cno;
787	new->ifile = NULL;
788	new->nilfs = nilfs;
789	refcount_set(&new->count, 1);
790	atomic64_set(&new->inodes_count, 0);
791	atomic64_set(&new->blocks_count, 0);
792
793	rb_link_node(&new->rb_node, parent, p);
794	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
795
796	spin_unlock(&nilfs->ns_cptree_lock);
797
798	err = nilfs_sysfs_create_snapshot_group(new);
799	if (err) {
800		kfree(new);
801		new = NULL;
802	}
803
804	return new;
805}
806
807void nilfs_put_root(struct nilfs_root *root)
808{
809	if (refcount_dec_and_test(&root->count)) {
810		struct the_nilfs *nilfs = root->nilfs;
811
812		nilfs_sysfs_delete_snapshot_group(root);
813
814		spin_lock(&nilfs->ns_cptree_lock);
815		rb_erase(&root->rb_node, &nilfs->ns_cptree);
816		spin_unlock(&nilfs->ns_cptree_lock);
817		iput(root->ifile);
 
818
819		kfree(root);
820	}
821}
v3.1
  1/*
  2 * the_nilfs.c - the_nilfs shared structure.
  3 *
  4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
  5 *
  6 * This program is free software; you can redistribute it and/or modify
  7 * it under the terms of the GNU General Public License as published by
  8 * the Free Software Foundation; either version 2 of the License, or
  9 * (at your option) any later version.
 10 *
 11 * This program is distributed in the hope that it will be useful,
 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 14 * GNU General Public License for more details.
 15 *
 16 * You should have received a copy of the GNU General Public License
 17 * along with this program; if not, write to the Free Software
 18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 19 *
 20 * Written by Ryusuke Konishi <ryusuke@osrg.net>
 21 *
 22 */
 23
 24#include <linux/buffer_head.h>
 25#include <linux/slab.h>
 26#include <linux/blkdev.h>
 27#include <linux/backing-dev.h>
 28#include <linux/random.h>
 29#include <linux/crc32.h>
 30#include "nilfs.h"
 31#include "segment.h"
 32#include "alloc.h"
 33#include "cpfile.h"
 34#include "sufile.h"
 35#include "dat.h"
 36#include "segbuf.h"
 37
 38
 39static int nilfs_valid_sb(struct nilfs_super_block *sbp);
 40
 41void nilfs_set_last_segment(struct the_nilfs *nilfs,
 42			    sector_t start_blocknr, u64 seq, __u64 cno)
 43{
 44	spin_lock(&nilfs->ns_last_segment_lock);
 45	nilfs->ns_last_pseg = start_blocknr;
 46	nilfs->ns_last_seq = seq;
 47	nilfs->ns_last_cno = cno;
 48
 49	if (!nilfs_sb_dirty(nilfs)) {
 50		if (nilfs->ns_prev_seq == nilfs->ns_last_seq)
 51			goto stay_cursor;
 52
 53		set_nilfs_sb_dirty(nilfs);
 54	}
 55	nilfs->ns_prev_seq = nilfs->ns_last_seq;
 56
 57 stay_cursor:
 58	spin_unlock(&nilfs->ns_last_segment_lock);
 59}
 60
 61/**
 62 * alloc_nilfs - allocate a nilfs object
 63 * @bdev: block device to which the_nilfs is related
 64 *
 65 * Return Value: On success, pointer to the_nilfs is returned.
 66 * On error, NULL is returned.
 67 */
 68struct the_nilfs *alloc_nilfs(struct block_device *bdev)
 69{
 70	struct the_nilfs *nilfs;
 71
 72	nilfs = kzalloc(sizeof(*nilfs), GFP_KERNEL);
 73	if (!nilfs)
 74		return NULL;
 75
 76	nilfs->ns_bdev = bdev;
 
 77	atomic_set(&nilfs->ns_ndirtyblks, 0);
 78	init_rwsem(&nilfs->ns_sem);
 
 79	INIT_LIST_HEAD(&nilfs->ns_dirty_files);
 80	INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
 81	spin_lock_init(&nilfs->ns_inode_lock);
 82	spin_lock_init(&nilfs->ns_next_gen_lock);
 83	spin_lock_init(&nilfs->ns_last_segment_lock);
 84	nilfs->ns_cptree = RB_ROOT;
 85	spin_lock_init(&nilfs->ns_cptree_lock);
 86	init_rwsem(&nilfs->ns_segctor_sem);
 
 87
 88	return nilfs;
 89}
 90
 91/**
 92 * destroy_nilfs - destroy nilfs object
 93 * @nilfs: nilfs object to be released
 94 */
 95void destroy_nilfs(struct the_nilfs *nilfs)
 96{
 97	might_sleep();
 98	if (nilfs_init(nilfs)) {
 
 99		brelse(nilfs->ns_sbh[0]);
100		brelse(nilfs->ns_sbh[1]);
101	}
102	kfree(nilfs);
103}
104
105static int nilfs_load_super_root(struct the_nilfs *nilfs,
106				 struct super_block *sb, sector_t sr_block)
107{
108	struct buffer_head *bh_sr;
109	struct nilfs_super_root *raw_sr;
110	struct nilfs_super_block **sbp = nilfs->ns_sbp;
111	struct nilfs_inode *rawi;
112	unsigned dat_entry_size, segment_usage_size, checkpoint_size;
113	unsigned inode_size;
114	int err;
115
116	err = nilfs_read_super_root_block(nilfs, sr_block, &bh_sr, 1);
117	if (unlikely(err))
118		return err;
119
120	down_read(&nilfs->ns_sem);
121	dat_entry_size = le16_to_cpu(sbp[0]->s_dat_entry_size);
122	checkpoint_size = le16_to_cpu(sbp[0]->s_checkpoint_size);
123	segment_usage_size = le16_to_cpu(sbp[0]->s_segment_usage_size);
124	up_read(&nilfs->ns_sem);
125
126	inode_size = nilfs->ns_inode_size;
127
128	rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
129	err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
130	if (err)
131		goto failed;
132
133	rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
134	err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
135	if (err)
136		goto failed_dat;
137
138	rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
139	err = nilfs_sufile_read(sb, segment_usage_size, rawi,
140				&nilfs->ns_sufile);
141	if (err)
142		goto failed_cpfile;
143
144	raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
145	nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
146
147 failed:
148	brelse(bh_sr);
149	return err;
150
151 failed_cpfile:
152	iput(nilfs->ns_cpfile);
153
154 failed_dat:
155	iput(nilfs->ns_dat);
156	goto failed;
157}
158
159static void nilfs_init_recovery_info(struct nilfs_recovery_info *ri)
160{
161	memset(ri, 0, sizeof(*ri));
162	INIT_LIST_HEAD(&ri->ri_used_segments);
163}
164
165static void nilfs_clear_recovery_info(struct nilfs_recovery_info *ri)
166{
167	nilfs_dispose_segment_list(&ri->ri_used_segments);
168}
169
170/**
171 * nilfs_store_log_cursor - load log cursor from a super block
172 * @nilfs: nilfs object
173 * @sbp: buffer storing super block to be read
174 *
175 * nilfs_store_log_cursor() reads the last position of the log
176 * containing a super root from a given super block, and initializes
177 * relevant information on the nilfs object preparatory for log
178 * scanning and recovery.
179 */
180static int nilfs_store_log_cursor(struct the_nilfs *nilfs,
181				  struct nilfs_super_block *sbp)
182{
183	int ret = 0;
184
185	nilfs->ns_last_pseg = le64_to_cpu(sbp->s_last_pseg);
186	nilfs->ns_last_cno = le64_to_cpu(sbp->s_last_cno);
187	nilfs->ns_last_seq = le64_to_cpu(sbp->s_last_seq);
188
189	nilfs->ns_prev_seq = nilfs->ns_last_seq;
190	nilfs->ns_seg_seq = nilfs->ns_last_seq;
191	nilfs->ns_segnum =
192		nilfs_get_segnum_of_block(nilfs, nilfs->ns_last_pseg);
193	nilfs->ns_cno = nilfs->ns_last_cno + 1;
194	if (nilfs->ns_segnum >= nilfs->ns_nsegments) {
195		printk(KERN_ERR "NILFS invalid last segment number.\n");
 
 
 
196		ret = -EINVAL;
197	}
198	return ret;
199}
200
201/**
202 * load_nilfs - load and recover the nilfs
203 * @nilfs: the_nilfs structure to be released
204 * @sb: super block isntance used to recover past segment
205 *
206 * load_nilfs() searches and load the latest super root,
207 * attaches the last segment, and does recovery if needed.
208 * The caller must call this exclusively for simultaneous mounts.
209 */
210int load_nilfs(struct the_nilfs *nilfs, struct super_block *sb)
211{
212	struct nilfs_recovery_info ri;
213	unsigned int s_flags = sb->s_flags;
214	int really_read_only = bdev_read_only(nilfs->ns_bdev);
215	int valid_fs = nilfs_valid_fs(nilfs);
216	int err;
217
218	if (!valid_fs) {
219		printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
220		if (s_flags & MS_RDONLY) {
221			printk(KERN_INFO "NILFS: INFO: recovery "
222			       "required for readonly filesystem.\n");
223			printk(KERN_INFO "NILFS: write access will "
224			       "be enabled during recovery.\n");
225		}
226	}
227
228	nilfs_init_recovery_info(&ri);
229
230	err = nilfs_search_super_root(nilfs, &ri);
231	if (unlikely(err)) {
232		struct nilfs_super_block **sbp = nilfs->ns_sbp;
233		int blocksize;
234
235		if (err != -EINVAL)
236			goto scan_error;
237
238		if (!nilfs_valid_sb(sbp[1])) {
239			printk(KERN_WARNING
240			       "NILFS warning: unable to fall back to spare"
241			       "super block\n");
242			goto scan_error;
243		}
244		printk(KERN_INFO
245		       "NILFS: try rollback from an earlier position\n");
246
247		/*
248		 * restore super block with its spare and reconfigure
249		 * relevant states of the nilfs object.
250		 */
251		memcpy(sbp[0], sbp[1], nilfs->ns_sbsize);
252		nilfs->ns_crc_seed = le32_to_cpu(sbp[0]->s_crc_seed);
253		nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
254
255		/* verify consistency between two super blocks */
256		blocksize = BLOCK_SIZE << le32_to_cpu(sbp[0]->s_log_block_size);
257		if (blocksize != nilfs->ns_blocksize) {
258			printk(KERN_WARNING
259			       "NILFS warning: blocksize differs between "
260			       "two super blocks (%d != %d)\n",
261			       blocksize, nilfs->ns_blocksize);
262			goto scan_error;
263		}
264
265		err = nilfs_store_log_cursor(nilfs, sbp[0]);
266		if (err)
267			goto scan_error;
268
269		/* drop clean flag to allow roll-forward and recovery */
270		nilfs->ns_mount_state &= ~NILFS_VALID_FS;
271		valid_fs = 0;
272
273		err = nilfs_search_super_root(nilfs, &ri);
274		if (err)
275			goto scan_error;
276	}
277
278	err = nilfs_load_super_root(nilfs, sb, ri.ri_super_root);
279	if (unlikely(err)) {
280		printk(KERN_ERR "NILFS: error loading super root.\n");
 
281		goto failed;
282	}
283
284	if (valid_fs)
285		goto skip_recovery;
286
287	if (s_flags & MS_RDONLY) {
288		__u64 features;
289
290		if (nilfs_test_opt(nilfs, NORECOVERY)) {
291			printk(KERN_INFO "NILFS: norecovery option specified. "
292			       "skipping roll-forward recovery\n");
293			goto skip_recovery;
294		}
295		features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) &
296			~NILFS_FEATURE_COMPAT_RO_SUPP;
297		if (features) {
298			printk(KERN_ERR "NILFS: couldn't proceed with "
299			       "recovery because of unsupported optional "
300			       "features (%llx)\n",
301			       (unsigned long long)features);
302			err = -EROFS;
303			goto failed_unload;
304		}
305		if (really_read_only) {
306			printk(KERN_ERR "NILFS: write access "
307			       "unavailable, cannot proceed.\n");
308			err = -EROFS;
309			goto failed_unload;
310		}
311		sb->s_flags &= ~MS_RDONLY;
312	} else if (nilfs_test_opt(nilfs, NORECOVERY)) {
313		printk(KERN_ERR "NILFS: recovery cancelled because norecovery "
314		       "option was specified for a read/write mount\n");
315		err = -EINVAL;
316		goto failed_unload;
317	}
318
319	err = nilfs_salvage_orphan_logs(nilfs, sb, &ri);
320	if (err)
321		goto failed_unload;
322
323	down_write(&nilfs->ns_sem);
324	nilfs->ns_mount_state |= NILFS_VALID_FS; /* set "clean" flag */
325	err = nilfs_cleanup_super(sb);
326	up_write(&nilfs->ns_sem);
327
328	if (err) {
329		printk(KERN_ERR "NILFS: failed to update super block. "
330		       "recovery unfinished.\n");
 
331		goto failed_unload;
332	}
333	printk(KERN_INFO "NILFS: recovery complete.\n");
334
335 skip_recovery:
336	nilfs_clear_recovery_info(&ri);
337	sb->s_flags = s_flags;
338	return 0;
339
340 scan_error:
341	printk(KERN_ERR "NILFS: error searching super root.\n");
342	goto failed;
343
344 failed_unload:
345	iput(nilfs->ns_cpfile);
346	iput(nilfs->ns_sufile);
347	iput(nilfs->ns_dat);
348
349 failed:
350	nilfs_clear_recovery_info(&ri);
351	sb->s_flags = s_flags;
352	return err;
353}
354
355static unsigned long long nilfs_max_size(unsigned int blkbits)
356{
357	unsigned int max_bits;
358	unsigned long long res = MAX_LFS_FILESIZE; /* page cache limit */
359
360	max_bits = blkbits + NILFS_BMAP_KEY_BIT; /* bmap size limit */
361	if (max_bits < 64)
362		res = min_t(unsigned long long, res, (1ULL << max_bits) - 1);
363	return res;
364}
365
366/**
367 * nilfs_nrsvsegs - calculate the number of reserved segments
368 * @nilfs: nilfs object
369 * @nsegs: total number of segments
370 */
371unsigned long nilfs_nrsvsegs(struct the_nilfs *nilfs, unsigned long nsegs)
372{
373	return max_t(unsigned long, NILFS_MIN_NRSVSEGS,
374		     DIV_ROUND_UP(nsegs * nilfs->ns_r_segments_percentage,
375				  100));
376}
377
378void nilfs_set_nsegments(struct the_nilfs *nilfs, unsigned long nsegs)
379{
380	nilfs->ns_nsegments = nsegs;
381	nilfs->ns_nrsvsegs = nilfs_nrsvsegs(nilfs, nsegs);
382}
383
384static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
385				   struct nilfs_super_block *sbp)
386{
387	if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
388		printk(KERN_ERR "NILFS: unsupported revision "
389		       "(superblock rev.=%d.%d, current rev.=%d.%d). "
390		       "Please check the version of mkfs.nilfs.\n",
391		       le32_to_cpu(sbp->s_rev_level),
392		       le16_to_cpu(sbp->s_minor_rev_level),
393		       NILFS_CURRENT_REV, NILFS_MINOR_REV);
394		return -EINVAL;
395	}
396	nilfs->ns_sbsize = le16_to_cpu(sbp->s_bytes);
397	if (nilfs->ns_sbsize > BLOCK_SIZE)
398		return -EINVAL;
399
400	nilfs->ns_inode_size = le16_to_cpu(sbp->s_inode_size);
 
 
 
 
 
 
 
 
 
 
 
 
401	nilfs->ns_first_ino = le32_to_cpu(sbp->s_first_ino);
402
403	nilfs->ns_blocks_per_segment = le32_to_cpu(sbp->s_blocks_per_segment);
404	if (nilfs->ns_blocks_per_segment < NILFS_SEG_MIN_BLOCKS) {
405		printk(KERN_ERR "NILFS: too short segment.\n");
 
 
406		return -EINVAL;
407	}
408
409	nilfs->ns_first_data_block = le64_to_cpu(sbp->s_first_data_block);
410	nilfs->ns_r_segments_percentage =
411		le32_to_cpu(sbp->s_r_segments_percentage);
 
 
 
 
 
 
 
 
412	nilfs_set_nsegments(nilfs, le64_to_cpu(sbp->s_nsegments));
413	nilfs->ns_crc_seed = le32_to_cpu(sbp->s_crc_seed);
414	return 0;
415}
416
417static int nilfs_valid_sb(struct nilfs_super_block *sbp)
418{
419	static unsigned char sum[4];
420	const int sumoff = offsetof(struct nilfs_super_block, s_sum);
421	size_t bytes;
422	u32 crc;
423
424	if (!sbp || le16_to_cpu(sbp->s_magic) != NILFS_SUPER_MAGIC)
425		return 0;
426	bytes = le16_to_cpu(sbp->s_bytes);
427	if (bytes > BLOCK_SIZE)
428		return 0;
429	crc = crc32_le(le32_to_cpu(sbp->s_crc_seed), (unsigned char *)sbp,
430		       sumoff);
431	crc = crc32_le(crc, sum, 4);
432	crc = crc32_le(crc, (unsigned char *)sbp + sumoff + 4,
433		       bytes - sumoff - 4);
434	return crc == le32_to_cpu(sbp->s_sum);
435}
436
437static int nilfs_sb2_bad_offset(struct nilfs_super_block *sbp, u64 offset)
438{
439	return offset < ((le64_to_cpu(sbp->s_nsegments) *
440			  le32_to_cpu(sbp->s_blocks_per_segment)) <<
441			 (le32_to_cpu(sbp->s_log_block_size) + 10));
442}
443
444static void nilfs_release_super_block(struct the_nilfs *nilfs)
445{
446	int i;
447
448	for (i = 0; i < 2; i++) {
449		if (nilfs->ns_sbp[i]) {
450			brelse(nilfs->ns_sbh[i]);
451			nilfs->ns_sbh[i] = NULL;
452			nilfs->ns_sbp[i] = NULL;
453		}
454	}
455}
456
457void nilfs_fall_back_super_block(struct the_nilfs *nilfs)
458{
459	brelse(nilfs->ns_sbh[0]);
460	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
461	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
462	nilfs->ns_sbh[1] = NULL;
463	nilfs->ns_sbp[1] = NULL;
464}
465
466void nilfs_swap_super_block(struct the_nilfs *nilfs)
467{
468	struct buffer_head *tsbh = nilfs->ns_sbh[0];
469	struct nilfs_super_block *tsbp = nilfs->ns_sbp[0];
470
471	nilfs->ns_sbh[0] = nilfs->ns_sbh[1];
472	nilfs->ns_sbp[0] = nilfs->ns_sbp[1];
473	nilfs->ns_sbh[1] = tsbh;
474	nilfs->ns_sbp[1] = tsbp;
475}
476
477static int nilfs_load_super_block(struct the_nilfs *nilfs,
478				  struct super_block *sb, int blocksize,
479				  struct nilfs_super_block **sbpp)
480{
481	struct nilfs_super_block **sbp = nilfs->ns_sbp;
482	struct buffer_head **sbh = nilfs->ns_sbh;
483	u64 sb2off = NILFS_SB2_OFFSET_BYTES(nilfs->ns_bdev->bd_inode->i_size);
484	int valid[2], swp = 0;
485
486	sbp[0] = nilfs_read_super_block(sb, NILFS_SB_OFFSET_BYTES, blocksize,
487					&sbh[0]);
488	sbp[1] = nilfs_read_super_block(sb, sb2off, blocksize, &sbh[1]);
489
490	if (!sbp[0]) {
491		if (!sbp[1]) {
492			printk(KERN_ERR "NILFS: unable to read superblock\n");
493			return -EIO;
494		}
495		printk(KERN_WARNING
496		       "NILFS warning: unable to read primary superblock "
497		       "(blocksize = %d)\n", blocksize);
498	} else if (!sbp[1]) {
499		printk(KERN_WARNING
500		       "NILFS warning: unable to read secondary superblock "
501		       "(blocksize = %d)\n", blocksize);
502	}
503
504	/*
505	 * Compare two super blocks and set 1 in swp if the secondary
506	 * super block is valid and newer.  Otherwise, set 0 in swp.
507	 */
508	valid[0] = nilfs_valid_sb(sbp[0]);
509	valid[1] = nilfs_valid_sb(sbp[1]);
510	swp = valid[1] && (!valid[0] ||
511			   le64_to_cpu(sbp[1]->s_last_cno) >
512			   le64_to_cpu(sbp[0]->s_last_cno));
513
514	if (valid[swp] && nilfs_sb2_bad_offset(sbp[swp], sb2off)) {
515		brelse(sbh[1]);
516		sbh[1] = NULL;
517		sbp[1] = NULL;
 
518		swp = 0;
519	}
520	if (!valid[swp]) {
521		nilfs_release_super_block(nilfs);
522		printk(KERN_ERR "NILFS: Can't find nilfs on dev %s.\n",
523		       sb->s_id);
524		return -EINVAL;
525	}
526
527	if (!valid[!swp])
528		printk(KERN_WARNING "NILFS warning: broken superblock. "
529		       "using spare superblock (blocksize = %d).\n", blocksize);
 
530	if (swp)
531		nilfs_swap_super_block(nilfs);
532
533	nilfs->ns_sbwcount = 0;
534	nilfs->ns_sbwtime = le64_to_cpu(sbp[0]->s_wtime);
535	nilfs->ns_prot_seq = le64_to_cpu(sbp[valid[1] & !swp]->s_last_seq);
536	*sbpp = sbp[0];
537	return 0;
538}
539
540/**
541 * init_nilfs - initialize a NILFS instance.
542 * @nilfs: the_nilfs structure
543 * @sb: super block
544 * @data: mount options
545 *
546 * init_nilfs() performs common initialization per block device (e.g.
547 * reading the super block, getting disk layout information, initializing
548 * shared fields in the_nilfs).
549 *
550 * Return Value: On success, 0 is returned. On error, a negative error
551 * code is returned.
552 */
553int init_nilfs(struct the_nilfs *nilfs, struct super_block *sb, char *data)
554{
555	struct nilfs_super_block *sbp;
556	int blocksize;
557	int err;
558
559	down_write(&nilfs->ns_sem);
560
561	blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
562	if (!blocksize) {
563		printk(KERN_ERR "NILFS: unable to set blocksize\n");
564		err = -EINVAL;
565		goto out;
566	}
567	err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
568	if (err)
569		goto out;
570
571	err = nilfs_store_magic_and_option(sb, sbp, data);
572	if (err)
573		goto failed_sbh;
574
575	err = nilfs_check_feature_compatibility(sb, sbp);
576	if (err)
577		goto failed_sbh;
578
579	blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
580	if (blocksize < NILFS_MIN_BLOCK_SIZE ||
581	    blocksize > NILFS_MAX_BLOCK_SIZE) {
582		printk(KERN_ERR "NILFS: couldn't mount because of unsupported "
583		       "filesystem blocksize %d\n", blocksize);
 
584		err = -EINVAL;
585		goto failed_sbh;
586	}
587	if (sb->s_blocksize != blocksize) {
588		int hw_blocksize = bdev_logical_block_size(sb->s_bdev);
589
590		if (blocksize < hw_blocksize) {
591			printk(KERN_ERR
592			       "NILFS: blocksize %d too small for device "
593			       "(sector-size = %d).\n",
594			       blocksize, hw_blocksize);
595			err = -EINVAL;
596			goto failed_sbh;
597		}
598		nilfs_release_super_block(nilfs);
599		sb_set_blocksize(sb, blocksize);
600
601		err = nilfs_load_super_block(nilfs, sb, blocksize, &sbp);
602		if (err)
603			goto out;
604			/* not failed_sbh; sbh is released automatically
605			   when reloading fails. */
 
 
606	}
607	nilfs->ns_blocksize_bits = sb->s_blocksize_bits;
608	nilfs->ns_blocksize = blocksize;
609
610	get_random_bytes(&nilfs->ns_next_generation,
611			 sizeof(nilfs->ns_next_generation));
612
613	err = nilfs_store_disk_layout(nilfs, sbp);
614	if (err)
615		goto failed_sbh;
616
617	sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
618
619	nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
620
621	err = nilfs_store_log_cursor(nilfs, sbp);
622	if (err)
623		goto failed_sbh;
624
 
 
 
 
625	set_nilfs_init(nilfs);
626	err = 0;
627 out:
628	up_write(&nilfs->ns_sem);
629	return err;
630
631 failed_sbh:
632	nilfs_release_super_block(nilfs);
633	goto out;
634}
635
636int nilfs_discard_segments(struct the_nilfs *nilfs, __u64 *segnump,
637			    size_t nsegs)
638{
639	sector_t seg_start, seg_end;
640	sector_t start = 0, nblocks = 0;
641	unsigned int sects_per_block;
642	__u64 *sn;
643	int ret = 0;
644
645	sects_per_block = (1 << nilfs->ns_blocksize_bits) /
646		bdev_logical_block_size(nilfs->ns_bdev);
647	for (sn = segnump; sn < segnump + nsegs; sn++) {
648		nilfs_get_segment_range(nilfs, *sn, &seg_start, &seg_end);
649
650		if (!nblocks) {
651			start = seg_start;
652			nblocks = seg_end - seg_start + 1;
653		} else if (start + nblocks == seg_start) {
654			nblocks += seg_end - seg_start + 1;
655		} else {
656			ret = blkdev_issue_discard(nilfs->ns_bdev,
657						   start * sects_per_block,
658						   nblocks * sects_per_block,
659						   GFP_NOFS, 0);
660			if (ret < 0)
661				return ret;
662			nblocks = 0;
663		}
664	}
665	if (nblocks)
666		ret = blkdev_issue_discard(nilfs->ns_bdev,
667					   start * sects_per_block,
668					   nblocks * sects_per_block,
669					   GFP_NOFS, 0);
670	return ret;
671}
672
673int nilfs_count_free_blocks(struct the_nilfs *nilfs, sector_t *nblocks)
674{
675	unsigned long ncleansegs;
676
677	down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
678	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
679	up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
680	*nblocks = (sector_t)ncleansegs * nilfs->ns_blocks_per_segment;
681	return 0;
682}
683
684int nilfs_near_disk_full(struct the_nilfs *nilfs)
685{
686	unsigned long ncleansegs, nincsegs;
687
688	ncleansegs = nilfs_sufile_get_ncleansegs(nilfs->ns_sufile);
689	nincsegs = atomic_read(&nilfs->ns_ndirtyblks) /
690		nilfs->ns_blocks_per_segment + 1;
691
692	return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
693}
694
695struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
696{
697	struct rb_node *n;
698	struct nilfs_root *root;
699
700	spin_lock(&nilfs->ns_cptree_lock);
701	n = nilfs->ns_cptree.rb_node;
702	while (n) {
703		root = rb_entry(n, struct nilfs_root, rb_node);
704
705		if (cno < root->cno) {
706			n = n->rb_left;
707		} else if (cno > root->cno) {
708			n = n->rb_right;
709		} else {
710			atomic_inc(&root->count);
711			spin_unlock(&nilfs->ns_cptree_lock);
712			return root;
713		}
714	}
715	spin_unlock(&nilfs->ns_cptree_lock);
716
717	return NULL;
718}
719
720struct nilfs_root *
721nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
722{
723	struct rb_node **p, *parent;
724	struct nilfs_root *root, *new;
 
725
726	root = nilfs_lookup_root(nilfs, cno);
727	if (root)
728		return root;
729
730	new = kmalloc(sizeof(*root), GFP_KERNEL);
731	if (!new)
732		return NULL;
733
734	spin_lock(&nilfs->ns_cptree_lock);
735
736	p = &nilfs->ns_cptree.rb_node;
737	parent = NULL;
738
739	while (*p) {
740		parent = *p;
741		root = rb_entry(parent, struct nilfs_root, rb_node);
742
743		if (cno < root->cno) {
744			p = &(*p)->rb_left;
745		} else if (cno > root->cno) {
746			p = &(*p)->rb_right;
747		} else {
748			atomic_inc(&root->count);
749			spin_unlock(&nilfs->ns_cptree_lock);
750			kfree(new);
751			return root;
752		}
753	}
754
755	new->cno = cno;
756	new->ifile = NULL;
757	new->nilfs = nilfs;
758	atomic_set(&new->count, 1);
759	atomic_set(&new->inodes_count, 0);
760	atomic_set(&new->blocks_count, 0);
761
762	rb_link_node(&new->rb_node, parent, p);
763	rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
764
765	spin_unlock(&nilfs->ns_cptree_lock);
766
 
 
 
 
 
 
767	return new;
768}
769
770void nilfs_put_root(struct nilfs_root *root)
771{
772	if (atomic_dec_and_test(&root->count)) {
773		struct the_nilfs *nilfs = root->nilfs;
774
 
 
775		spin_lock(&nilfs->ns_cptree_lock);
776		rb_erase(&root->rb_node, &nilfs->ns_cptree);
777		spin_unlock(&nilfs->ns_cptree_lock);
778		if (root->ifile)
779			iput(root->ifile);
780
781		kfree(root);
782	}
783}