4 * As should be obvious for Linux kernel code, license is GPLv2
6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
8 * Generally contains mount/umount code and also serves as a dump area for
9 * any functions that don't fit elsewhere and neither justify a file of their
13 #include <linux/bio.h>
14 #include <linux/slab.h>
15 #include <linux/blkdev.h>
16 #include <linux/mtd/mtd.h>
17 #include <linux/statfs.h>
18 #include <linux/buffer_head.h>
20 static DEFINE_MUTEX(emergency_mutex);
21 static struct page *emergency_page;
23 struct page *emergency_read_begin(struct address_space *mapping, pgoff_t index)
25 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
29 page = read_cache_page(mapping, index, filler, NULL);
33 /* No more pages available, switch to emergency page */
34 printk(KERN_INFO"Logfs: Using emergency page\n");
35 mutex_lock(&emergency_mutex);
36 err = filler(NULL, emergency_page);
38 mutex_unlock(&emergency_mutex);
39 printk(KERN_EMERG"Logfs: Error reading emergency page\n");
42 return emergency_page;
45 void emergency_read_end(struct page *page)
47 if (page == emergency_page)
48 mutex_unlock(&emergency_mutex);
50 page_cache_release(page);
53 static void dump_segfile(struct super_block *sb)
55 struct logfs_super *super = logfs_super(sb);
56 struct logfs_segment_entry se;
59 for (segno = 0; segno < super->s_no_segs; segno++) {
60 logfs_get_segment_entry(sb, segno, &se);
61 printk("%3x: %6x %8x", segno, be32_to_cpu(se.ec_level),
62 be32_to_cpu(se.valid));
63 if (++segno < super->s_no_segs) {
64 logfs_get_segment_entry(sb, segno, &se);
65 printk(" %6x %8x", be32_to_cpu(se.ec_level),
66 be32_to_cpu(se.valid));
68 if (++segno < super->s_no_segs) {
69 logfs_get_segment_entry(sb, segno, &se);
70 printk(" %6x %8x", be32_to_cpu(se.ec_level),
71 be32_to_cpu(se.valid));
73 if (++segno < super->s_no_segs) {
74 logfs_get_segment_entry(sb, segno, &se);
75 printk(" %6x %8x", be32_to_cpu(se.ec_level),
76 be32_to_cpu(se.valid));
83 * logfs_crash_dump - dump debug information to device
85 * The LogFS superblock only occupies part of a segment. This function will
86 * write as much debug information as it can gather into the spare space.
88 void logfs_crash_dump(struct super_block *sb)
94 * FIXME: There should be a reserve for root, similar to ext2.
96 int logfs_statfs(struct dentry *dentry, struct kstatfs *stats)
98 struct super_block *sb = dentry->d_sb;
99 struct logfs_super *super = logfs_super(sb);
101 stats->f_type = LOGFS_MAGIC_U32;
102 stats->f_bsize = sb->s_blocksize;
103 stats->f_blocks = super->s_size >> LOGFS_BLOCK_BITS >> 3;
104 stats->f_bfree = super->s_free_bytes >> sb->s_blocksize_bits;
105 stats->f_bavail = super->s_free_bytes >> sb->s_blocksize_bits;
108 stats->f_namelen = LOGFS_MAX_NAMELEN;
112 static int logfs_sb_set(struct super_block *sb, void *_super)
114 struct logfs_super *super = _super;
116 sb->s_fs_info = super;
117 sb->s_mtd = super->s_mtd;
118 sb->s_bdev = super->s_bdev;
121 sb->s_bdi = &bdev_get_queue(sb->s_bdev)->backing_dev_info;
125 sb->s_bdi = sb->s_mtd->backing_dev_info;
130 static int logfs_sb_test(struct super_block *sb, void *_super)
132 struct logfs_super *super = _super;
133 struct mtd_info *mtd = super->s_mtd;
135 if (mtd && sb->s_mtd == mtd)
137 if (super->s_bdev && sb->s_bdev == super->s_bdev)
142 static void set_segment_header(struct logfs_segment_header *sh, u8 type,
143 u8 level, u32 segno, u32 ec)
148 sh->segno = cpu_to_be32(segno);
149 sh->ec = cpu_to_be32(ec);
150 sh->gec = cpu_to_be64(segno);
151 sh->crc = logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4);
154 static void logfs_write_ds(struct super_block *sb, struct logfs_disk_super *ds,
157 struct logfs_super *super = logfs_super(sb);
158 struct logfs_segment_header *sh = &ds->ds_sh;
161 memset(ds, 0, sizeof(*ds));
162 set_segment_header(sh, SEG_SUPER, 0, segno, ec);
164 ds->ds_ifile_levels = super->s_ifile_levels;
165 ds->ds_iblock_levels = super->s_iblock_levels;
166 ds->ds_data_levels = super->s_data_levels; /* XXX: Remove */
167 ds->ds_segment_shift = super->s_segshift;
168 ds->ds_block_shift = sb->s_blocksize_bits;
169 ds->ds_write_shift = super->s_writeshift;
170 ds->ds_filesystem_size = cpu_to_be64(super->s_size);
171 ds->ds_segment_size = cpu_to_be32(super->s_segsize);
172 ds->ds_bad_seg_reserve = cpu_to_be32(super->s_bad_seg_reserve);
173 ds->ds_feature_incompat = cpu_to_be64(super->s_feature_incompat);
174 ds->ds_feature_ro_compat= cpu_to_be64(super->s_feature_ro_compat);
175 ds->ds_feature_compat = cpu_to_be64(super->s_feature_compat);
176 ds->ds_feature_flags = cpu_to_be64(super->s_feature_flags);
177 ds->ds_root_reserve = cpu_to_be64(super->s_root_reserve);
178 ds->ds_speed_reserve = cpu_to_be64(super->s_speed_reserve);
180 ds->ds_journal_seg[i] = cpu_to_be32(super->s_journal_seg[i]);
181 ds->ds_magic = cpu_to_be64(LOGFS_MAGIC);
182 ds->ds_crc = logfs_crc32(ds, sizeof(*ds),
183 LOGFS_SEGMENT_HEADERSIZE + 12);
186 static int write_one_sb(struct super_block *sb,
187 struct page *(*find_sb)(struct super_block *sb, u64 *ofs))
189 struct logfs_super *super = logfs_super(sb);
190 struct logfs_disk_super *ds;
191 struct logfs_segment_entry se;
197 page = find_sb(sb, &ofs);
200 ds = page_address(page);
201 segno = seg_no(sb, ofs);
202 logfs_get_segment_entry(sb, segno, &se);
203 ec = be32_to_cpu(se.ec_level) >> 4;
205 logfs_set_segment_erased(sb, segno, ec, 0);
206 logfs_write_ds(sb, ds, segno, ec);
207 err = super->s_devops->write_sb(sb, page);
208 page_cache_release(page);
212 int logfs_write_sb(struct super_block *sb)
214 struct logfs_super *super = logfs_super(sb);
217 /* First superblock */
218 err = write_one_sb(sb, super->s_devops->find_first_sb);
222 /* Last superblock */
223 err = write_one_sb(sb, super->s_devops->find_last_sb);
229 static int ds_cmp(const void *ds0, const void *ds1)
231 size_t len = sizeof(struct logfs_disk_super);
233 /* We know the segment headers differ, so ignore them */
234 len -= LOGFS_SEGMENT_HEADERSIZE;
235 ds0 += LOGFS_SEGMENT_HEADERSIZE;
236 ds1 += LOGFS_SEGMENT_HEADERSIZE;
237 return memcmp(ds0, ds1, len);
240 static int logfs_recover_sb(struct super_block *sb)
242 struct logfs_super *super = logfs_super(sb);
243 struct logfs_disk_super _ds0, *ds0 = &_ds0;
244 struct logfs_disk_super _ds1, *ds1 = &_ds1;
245 int err, valid0, valid1;
247 /* read first superblock */
248 err = wbuf_read(sb, super->s_sb_ofs[0], sizeof(*ds0), ds0);
251 /* read last superblock */
252 err = wbuf_read(sb, super->s_sb_ofs[1], sizeof(*ds1), ds1);
255 valid0 = logfs_check_ds(ds0) == 0;
256 valid1 = logfs_check_ds(ds1) == 0;
258 if (!valid0 && valid1) {
259 printk(KERN_INFO"First superblock is invalid - fixing.\n");
260 return write_one_sb(sb, super->s_devops->find_first_sb);
262 if (valid0 && !valid1) {
263 printk(KERN_INFO"Last superblock is invalid - fixing.\n");
264 return write_one_sb(sb, super->s_devops->find_last_sb);
266 if (valid0 && valid1 && ds_cmp(ds0, ds1)) {
267 printk(KERN_INFO"Superblocks don't match - fixing.\n");
268 return logfs_write_sb(sb);
270 /* If neither is valid now, something's wrong. Didn't we properly
271 * check them before?!? */
272 BUG_ON(!valid0 && !valid1);
276 static int logfs_make_writeable(struct super_block *sb)
280 err = logfs_open_segfile(sb);
284 /* Repair any broken superblock copies */
285 err = logfs_recover_sb(sb);
289 /* Check areas for trailing unaccounted data */
290 err = logfs_check_areas(sb);
294 /* Do one GC pass before any data gets dirtied */
297 /* after all initializations are done, replay the journal
298 * for rw-mounts, if necessary */
299 err = logfs_replay_journal(sb);
306 static int logfs_get_sb_final(struct super_block *sb)
308 struct logfs_super *super = logfs_super(sb);
309 struct inode *rootdir;
313 rootdir = logfs_iget(sb, LOGFS_INO_ROOT);
317 sb->s_root = d_alloc_root(rootdir);
323 /* at that point we know that ->put_super() will be called */
324 super->s_erase_page = alloc_pages(GFP_KERNEL, 0);
325 if (!super->s_erase_page)
327 memset(page_address(super->s_erase_page), 0xFF, PAGE_SIZE);
329 /* FIXME: check for read-only mounts */
330 err = logfs_make_writeable(sb);
332 __free_page(super->s_erase_page);
336 log_super("LogFS: Finished mounting\n");
340 iput(super->s_master_inode);
341 iput(super->s_segfile_inode);
342 iput(super->s_mapping_inode);
346 int logfs_check_ds(struct logfs_disk_super *ds)
348 struct logfs_segment_header *sh = &ds->ds_sh;
350 if (ds->ds_magic != cpu_to_be64(LOGFS_MAGIC))
352 if (sh->crc != logfs_crc32(sh, LOGFS_SEGMENT_HEADERSIZE, 4))
354 if (ds->ds_crc != logfs_crc32(ds, sizeof(*ds),
355 LOGFS_SEGMENT_HEADERSIZE + 12))
360 static struct page *find_super_block(struct super_block *sb)
362 struct logfs_super *super = logfs_super(sb);
363 struct page *first, *last;
365 first = super->s_devops->find_first_sb(sb, &super->s_sb_ofs[0]);
366 if (!first || IS_ERR(first))
368 last = super->s_devops->find_last_sb(sb, &super->s_sb_ofs[1]);
369 if (!last || IS_ERR(last)) {
370 page_cache_release(first);
374 if (!logfs_check_ds(page_address(first))) {
375 page_cache_release(last);
379 /* First one didn't work, try the second superblock */
380 if (!logfs_check_ds(page_address(last))) {
381 page_cache_release(first);
385 /* Neither worked, sorry folks */
386 page_cache_release(first);
387 page_cache_release(last);
391 static int __logfs_read_sb(struct super_block *sb)
393 struct logfs_super *super = logfs_super(sb);
395 struct logfs_disk_super *ds;
398 page = find_super_block(sb);
402 ds = page_address(page);
403 super->s_size = be64_to_cpu(ds->ds_filesystem_size);
404 super->s_root_reserve = be64_to_cpu(ds->ds_root_reserve);
405 super->s_speed_reserve = be64_to_cpu(ds->ds_speed_reserve);
406 super->s_bad_seg_reserve = be32_to_cpu(ds->ds_bad_seg_reserve);
407 super->s_segsize = 1 << ds->ds_segment_shift;
408 super->s_segmask = (1 << ds->ds_segment_shift) - 1;
409 super->s_segshift = ds->ds_segment_shift;
410 sb->s_blocksize = 1 << ds->ds_block_shift;
411 sb->s_blocksize_bits = ds->ds_block_shift;
412 super->s_writesize = 1 << ds->ds_write_shift;
413 super->s_writeshift = ds->ds_write_shift;
414 super->s_no_segs = super->s_size >> super->s_segshift;
415 super->s_no_blocks = super->s_segsize >> sb->s_blocksize_bits;
416 super->s_feature_incompat = be64_to_cpu(ds->ds_feature_incompat);
417 super->s_feature_ro_compat = be64_to_cpu(ds->ds_feature_ro_compat);
418 super->s_feature_compat = be64_to_cpu(ds->ds_feature_compat);
419 super->s_feature_flags = be64_to_cpu(ds->ds_feature_flags);
422 super->s_journal_seg[i] = be32_to_cpu(ds->ds_journal_seg[i]);
424 super->s_ifile_levels = ds->ds_ifile_levels;
425 super->s_iblock_levels = ds->ds_iblock_levels;
426 super->s_data_levels = ds->ds_data_levels;
427 super->s_total_levels = super->s_ifile_levels + super->s_iblock_levels
428 + super->s_data_levels;
429 page_cache_release(page);
433 static int logfs_read_sb(struct super_block *sb, int read_only)
435 struct logfs_super *super = logfs_super(sb);
438 super->s_btree_pool = mempool_create(32, btree_alloc, btree_free, NULL);
439 if (!super->s_btree_pool)
442 btree_init_mempool64(&super->s_shadow_tree.new, super->s_btree_pool);
443 btree_init_mempool64(&super->s_shadow_tree.old, super->s_btree_pool);
444 btree_init_mempool32(&super->s_shadow_tree.segment_map,
445 super->s_btree_pool);
447 ret = logfs_init_mapping(sb);
451 ret = __logfs_read_sb(sb);
455 if (super->s_feature_incompat & ~LOGFS_FEATURES_INCOMPAT)
457 if ((super->s_feature_ro_compat & ~LOGFS_FEATURES_RO_COMPAT) &&
461 ret = logfs_init_rw(sb);
465 ret = logfs_init_areas(sb);
469 ret = logfs_init_gc(sb);
473 ret = logfs_init_journal(sb);
480 static void logfs_kill_sb(struct super_block *sb)
482 struct logfs_super *super = logfs_super(sb);
484 log_super("LogFS: Start unmounting\n");
485 /* Alias entries slow down mount, so evict as many as possible */
487 logfs_write_anchor(sb);
490 * From this point on alias entries are simply dropped - and any
491 * writes to the object store are considered bugs.
493 super->s_flags |= LOGFS_SB_FLAG_SHUTDOWN;
494 log_super("LogFS: Now in shutdown\n");
495 generic_shutdown_super(sb);
497 BUG_ON(super->s_dirty_used_bytes || super->s_dirty_free_bytes);
499 logfs_cleanup_gc(sb);
500 logfs_cleanup_journal(sb);
501 logfs_cleanup_areas(sb);
502 logfs_cleanup_rw(sb);
503 if (super->s_erase_page)
504 __free_page(super->s_erase_page);
505 super->s_devops->put_device(super);
506 logfs_mempool_destroy(super->s_btree_pool);
507 logfs_mempool_destroy(super->s_alias_pool);
509 log_super("LogFS: Finished unmounting\n");
512 static struct dentry *logfs_get_sb_device(struct logfs_super *super,
513 struct file_system_type *type, int flags)
515 struct super_block *sb;
517 static int mount_count;
519 log_super("LogFS: Start mount %x\n", mount_count++);
522 sb = sget(type, logfs_sb_test, logfs_sb_set, super);
524 super->s_devops->put_device(super);
530 /* Device is already in use */
531 super->s_devops->put_device(super);
533 return dget(sb->s_root);
537 * sb->s_maxbytes is limited to 8TB. On 32bit systems, the page cache
538 * only covers 16TB and the upper 8TB are used for indirect blocks.
539 * On 64bit system we could bump up the limit, but that would make
540 * the filesystem incompatible with 32bit systems.
542 sb->s_maxbytes = (1ull << 43) - 1;
543 sb->s_op = &logfs_super_operations;
544 sb->s_flags = flags | MS_NOATIME;
546 err = logfs_read_sb(sb, sb->s_flags & MS_RDONLY);
550 sb->s_flags |= MS_ACTIVE;
551 err = logfs_get_sb_final(sb);
553 deactivate_locked_super(sb);
556 return dget(sb->s_root);
559 /* no ->s_root, no ->put_super() */
560 iput(super->s_master_inode);
561 iput(super->s_segfile_inode);
562 iput(super->s_mapping_inode);
563 deactivate_locked_super(sb);
567 static struct dentry *logfs_mount(struct file_system_type *type, int flags,
568 const char *devname, void *data)
571 struct logfs_super *super;
574 super = kzalloc(sizeof(*super), GFP_KERNEL);
576 return ERR_PTR(-ENOMEM);
578 mutex_init(&super->s_dirop_mutex);
579 mutex_init(&super->s_object_alias_mutex);
580 INIT_LIST_HEAD(&super->s_freeing_list);
583 err = logfs_get_sb_bdev(super, type, devname);
584 else if (strncmp(devname, "mtd", 3))
585 err = logfs_get_sb_bdev(super, type, devname);
588 mtdnr = simple_strtoul(devname+3, &garbage, 0);
592 err = logfs_get_sb_mtd(super, mtdnr);
600 return logfs_get_sb_device(super, type, flags);
603 static struct file_system_type logfs_fs_type = {
604 .owner = THIS_MODULE,
606 .mount = logfs_mount,
607 .kill_sb = logfs_kill_sb,
608 .fs_flags = FS_REQUIRES_DEV,
612 static int __init logfs_init(void)
616 emergency_page = alloc_pages(GFP_KERNEL, 0);
620 ret = logfs_compr_init();
624 ret = logfs_init_inode_cache();
628 return register_filesystem(&logfs_fs_type);
632 __free_pages(emergency_page, 0);
636 static void __exit logfs_exit(void)
638 unregister_filesystem(&logfs_fs_type);
639 logfs_destroy_inode_cache();
641 __free_pages(emergency_page, 0);
644 module_init(logfs_init);
645 module_exit(logfs_exit);
647 MODULE_LICENSE("GPL v2");
648 MODULE_AUTHOR("Joern Engel <joern@logfs.org>");
649 MODULE_DESCRIPTION("scalable flash filesystem");