4 * As should be obvious for Linux kernel code, license is GPLv2
6 * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
9 * Actually contains five sets of very similar functions:
10 * read read blocks from a file
11 * seek_hole find next hole
12 * seek_data find next data block
13 * valid check whether a block still belongs to a file
14 * write write blocks to a file
15 * delete delete a block (for directories and ifile)
16 * rewrite move existing blocks of a file to a new location (gc helper)
17 * truncate truncate a file
20 #include <linux/sched.h>
22 static u64 adjust_bix(u64 bix, level_t level)
28 return max_t(u64, bix, I0_BLOCKS);
30 return max_t(u64, bix, I1_BLOCKS);
32 return max_t(u64, bix, I2_BLOCKS);
34 return max_t(u64, bix, I3_BLOCKS);
36 return max_t(u64, bix, I4_BLOCKS);
43 static inline u64 maxbix(u8 height)
45 return 1ULL << (LOGFS_BLOCK_BITS * height);
49 * The inode address space is cut in two halves. Lower half belongs to data
50 * pages, upper half to indirect blocks. If the high bit (INDIRECT_BIT) is
51 * set, the actual block index (bix) and level can be derived from the page
54 * The lowest three bits of the block index are set to 0 after packing and
55 * unpacking. Since the lowest n bits (9 for 4KiB blocksize) are ignored
56 * anyway this is harmless.
58 #define ARCH_SHIFT (BITS_PER_LONG - 32)
59 #define INDIRECT_BIT (0x80000000UL << ARCH_SHIFT)
60 #define LEVEL_SHIFT (28 + ARCH_SHIFT)
61 static inline pgoff_t first_indirect_block(void)
63 return INDIRECT_BIT | (1ULL << LEVEL_SHIFT);
66 pgoff_t logfs_pack_index(u64 bix, level_t level)
70 BUG_ON(bix >= INDIRECT_BIT);
75 index |= (__force long)level << LEVEL_SHIFT;
76 index |= bix >> ((__force u8)level * LOGFS_BLOCK_BITS);
80 void logfs_unpack_index(pgoff_t index, u64 *bix, level_t *level)
84 if (!(index & INDIRECT_BIT)) {
90 __level = (index & ~INDIRECT_BIT) >> LEVEL_SHIFT;
91 *level = LEVEL(__level);
92 *bix = (index << (__level * LOGFS_BLOCK_BITS)) & ~INDIRECT_BIT;
93 *bix = adjust_bix(*bix, *level);
101 * Time is stored as nanoseconds since the epoch.
103 static struct timespec be64_to_timespec(__be64 betime)
105 return ns_to_timespec(be64_to_cpu(betime));
108 static __be64 timespec_to_be64(struct timespec tsp)
110 return cpu_to_be64((u64)tsp.tv_sec * NSEC_PER_SEC + tsp.tv_nsec);
113 static void logfs_disk_to_inode(struct logfs_disk_inode *di, struct inode*inode)
115 struct logfs_inode *li = logfs_inode(inode);
118 inode->i_mode = be16_to_cpu(di->di_mode);
119 li->li_height = di->di_height;
120 li->li_flags = be32_to_cpu(di->di_flags);
121 inode->i_uid = be32_to_cpu(di->di_uid);
122 inode->i_gid = be32_to_cpu(di->di_gid);
123 inode->i_size = be64_to_cpu(di->di_size);
124 logfs_set_blocks(inode, be64_to_cpu(di->di_used_bytes));
125 inode->i_atime = be64_to_timespec(di->di_atime);
126 inode->i_ctime = be64_to_timespec(di->di_ctime);
127 inode->i_mtime = be64_to_timespec(di->di_mtime);
128 inode->i_nlink = be32_to_cpu(di->di_refcount);
129 inode->i_generation = be32_to_cpu(di->di_generation);
131 switch (inode->i_mode & S_IFMT) {
132 case S_IFSOCK: /* fall through */
133 case S_IFBLK: /* fall through */
134 case S_IFCHR: /* fall through */
136 inode->i_rdev = be64_to_cpu(di->di_data[0]);
138 case S_IFDIR: /* fall through */
139 case S_IFREG: /* fall through */
141 for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
142 li->li_data[i] = be64_to_cpu(di->di_data[i]);
149 static void logfs_inode_to_disk(struct inode *inode, struct logfs_disk_inode*di)
151 struct logfs_inode *li = logfs_inode(inode);
154 di->di_mode = cpu_to_be16(inode->i_mode);
155 di->di_height = li->li_height;
157 di->di_flags = cpu_to_be32(li->li_flags);
158 di->di_uid = cpu_to_be32(inode->i_uid);
159 di->di_gid = cpu_to_be32(inode->i_gid);
160 di->di_size = cpu_to_be64(i_size_read(inode));
161 di->di_used_bytes = cpu_to_be64(li->li_used_bytes);
162 di->di_atime = timespec_to_be64(inode->i_atime);
163 di->di_ctime = timespec_to_be64(inode->i_ctime);
164 di->di_mtime = timespec_to_be64(inode->i_mtime);
165 di->di_refcount = cpu_to_be32(inode->i_nlink);
166 di->di_generation = cpu_to_be32(inode->i_generation);
168 switch (inode->i_mode & S_IFMT) {
169 case S_IFSOCK: /* fall through */
170 case S_IFBLK: /* fall through */
171 case S_IFCHR: /* fall through */
173 di->di_data[0] = cpu_to_be64(inode->i_rdev);
175 case S_IFDIR: /* fall through */
176 case S_IFREG: /* fall through */
178 for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
179 di->di_data[i] = cpu_to_be64(li->li_data[i]);
186 static void __logfs_set_blocks(struct inode *inode)
188 struct super_block *sb = inode->i_sb;
189 struct logfs_inode *li = logfs_inode(inode);
191 inode->i_blocks = ULONG_MAX;
192 if (li->li_used_bytes >> sb->s_blocksize_bits < ULONG_MAX)
193 inode->i_blocks = ALIGN(li->li_used_bytes, 512) >> 9;
196 void logfs_set_blocks(struct inode *inode, u64 bytes)
198 struct logfs_inode *li = logfs_inode(inode);
200 li->li_used_bytes = bytes;
201 __logfs_set_blocks(inode);
204 static void prelock_page(struct super_block *sb, struct page *page, int lock)
206 struct logfs_super *super = logfs_super(sb);
208 BUG_ON(!PageLocked(page));
210 BUG_ON(PagePreLocked(page));
211 SetPagePreLocked(page);
213 /* We are in GC path. */
214 if (PagePreLocked(page))
215 super->s_lock_count++;
217 SetPagePreLocked(page);
221 static void preunlock_page(struct super_block *sb, struct page *page, int lock)
223 struct logfs_super *super = logfs_super(sb);
225 BUG_ON(!PageLocked(page));
227 ClearPagePreLocked(page);
229 /* We are in GC path. */
230 BUG_ON(!PagePreLocked(page));
231 if (super->s_lock_count)
232 super->s_lock_count--;
234 ClearPagePreLocked(page);
239 * Logfs is prone to an AB-BA deadlock where one task tries to acquire
240 * s_write_mutex with a locked page and GC tries to get that page while holding
242 * To solve this issue logfs will ignore the page lock iff the page in question
243 * is waiting for s_write_mutex. We annotate this fact by setting PG_pre_locked
244 * in addition to PG_locked.
246 static void logfs_get_wblocks(struct super_block *sb, struct page *page,
249 struct logfs_super *super = logfs_super(sb);
252 prelock_page(sb, page, lock);
255 mutex_lock(&super->s_write_mutex);
257 /* FIXME: We also have to check for shadowed space
258 * and mempool fill grade */
262 static void logfs_put_wblocks(struct super_block *sb, struct page *page,
265 struct logfs_super *super = logfs_super(sb);
268 preunlock_page(sb, page, lock);
269 /* Order matters - we must clear PG_pre_locked before releasing
270 * s_write_mutex or we could race against another task. */
272 mutex_unlock(&super->s_write_mutex);
275 static struct page *logfs_get_read_page(struct inode *inode, u64 bix,
278 return find_or_create_page(inode->i_mapping,
279 logfs_pack_index(bix, level), GFP_NOFS);
282 static void logfs_put_read_page(struct page *page)
285 page_cache_release(page);
288 static void logfs_lock_write_page(struct page *page)
292 while (unlikely(!trylock_page(page))) {
293 if (loop++ > 0x1000) {
294 /* Has been observed once so far... */
295 printk(KERN_ERR "stack at %p\n", &loop);
298 if (PagePreLocked(page)) {
299 /* Holder of page lock is waiting for us, it
300 * is safe to use this page. */
303 /* Some other process has this page locked and has
304 * nothing to do with us. Wait for it to finish.
308 BUG_ON(!PageLocked(page));
311 static struct page *logfs_get_write_page(struct inode *inode, u64 bix,
314 struct address_space *mapping = inode->i_mapping;
315 pgoff_t index = logfs_pack_index(bix, level);
320 page = find_get_page(mapping, index);
322 page = __page_cache_alloc(GFP_NOFS);
325 err = add_to_page_cache_lru(page, mapping, index, GFP_NOFS);
327 page_cache_release(page);
332 } else logfs_lock_write_page(page);
333 BUG_ON(!PageLocked(page));
337 static void logfs_unlock_write_page(struct page *page)
339 if (!PagePreLocked(page))
343 static void logfs_put_write_page(struct page *page)
345 logfs_unlock_write_page(page);
346 page_cache_release(page);
349 static struct page *logfs_get_page(struct inode *inode, u64 bix, level_t level,
353 return logfs_get_read_page(inode, bix, level);
355 return logfs_get_write_page(inode, bix, level);
358 static void logfs_put_page(struct page *page, int rw)
361 logfs_put_read_page(page);
363 logfs_put_write_page(page);
366 static unsigned long __get_bits(u64 val, int skip, int no)
376 static unsigned long get_bits(u64 val, level_t skip)
378 return __get_bits(val, (__force int)skip, LOGFS_BLOCK_BITS);
381 static inline void init_shadow_tree(struct super_block *sb,
382 struct shadow_tree *tree)
384 struct logfs_super *super = logfs_super(sb);
386 btree_init_mempool64(&tree->new, super->s_btree_pool);
387 btree_init_mempool64(&tree->old, super->s_btree_pool);
390 static void indirect_write_block(struct logfs_block *block)
397 inode = page->mapping->host;
398 logfs_lock_write_page(page);
399 ret = logfs_write_buf(inode, page, 0);
400 logfs_unlock_write_page(page);
402 * This needs some rework. Unless you want your filesystem to run
403 * completely synchronously (you don't), the filesystem will always
404 * report writes as 'successful' before the actual work has been
405 * done. The actual work gets done here and this is where any errors
406 * will show up. And there isn't much we can do about it, really.
408 * Some attempts to fix the errors (move from bad blocks, retry io,...)
409 * have already been done, so anything left should be either a broken
410 * device or a bug somewhere in logfs itself. Being relatively new,
411 * the odds currently favor a bug, so for now the line below isn't
417 static void inode_write_block(struct logfs_block *block)
422 inode = block->inode;
423 if (inode->i_ino == LOGFS_INO_MASTER)
424 logfs_write_anchor(inode->i_sb);
426 ret = __logfs_write_inode(inode, 0);
427 /* see indirect_write_block comment */
433 * This silences a false, yet annoying gcc warning. I hate it when my editor
434 * jumps into bitops.h each time I recompile this file.
435 * TODO: Complain to gcc folks about this and upgrade compiler.
437 static unsigned long fnb(const unsigned long *addr,
438 unsigned long size, unsigned long offset)
440 return find_next_bit(addr, size, offset);
443 static __be64 inode_val0(struct inode *inode)
445 struct logfs_inode *li = logfs_inode(inode);
449 * Explicit shifting generates good code, but must match the format
450 * of the structure. Add some paranoia just in case.
452 BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_mode) != 0);
453 BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_height) != 2);
454 BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_flags) != 4);
456 val = (u64)inode->i_mode << 48 |
457 (u64)li->li_height << 40 |
459 return cpu_to_be64(val);
462 static int inode_write_alias(struct super_block *sb,
463 struct logfs_block *block, write_alias_t *write_one_alias)
465 struct inode *inode = block->inode;
466 struct logfs_inode *li = logfs_inode(inode);
473 for (pos = 0; ; pos++) {
474 pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos);
475 if (pos >= LOGFS_EMBEDDED_FIELDS + INODE_POINTER_OFS)
479 case INODE_HEIGHT_OFS:
480 val = inode_val0(inode);
483 val = cpu_to_be64(li->li_used_bytes);;
486 val = cpu_to_be64(i_size_read(inode));
488 case INODE_POINTER_OFS ... INODE_POINTER_OFS + LOGFS_EMBEDDED_FIELDS - 1:
489 val = cpu_to_be64(li->li_data[pos - INODE_POINTER_OFS]);
495 ino = LOGFS_INO_MASTER;
498 err = write_one_alias(sb, ino, bix, level, pos, val);
504 static int indirect_write_alias(struct super_block *sb,
505 struct logfs_block *block, write_alias_t *write_one_alias)
508 struct page *page = block->page;
514 for (pos = 0; ; pos++) {
515 pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos);
516 if (pos >= LOGFS_BLOCK_FACTOR)
519 ino = page->mapping->host->i_ino;
520 logfs_unpack_index(page->index, &bix, &level);
521 child = kmap_atomic(page, KM_USER0);
523 kunmap_atomic(child, KM_USER0);
524 err = write_one_alias(sb, ino, bix, level, pos, val);
530 int logfs_write_obj_aliases_pagecache(struct super_block *sb)
532 struct logfs_super *super = logfs_super(sb);
533 struct logfs_block *block;
536 list_for_each_entry(block, &super->s_object_alias, alias_list) {
537 err = block->ops->write_alias(sb, block, write_alias_journal);
544 void __free_block(struct super_block *sb, struct logfs_block *block)
546 BUG_ON(!list_empty(&block->item_list));
547 list_del(&block->alias_list);
548 mempool_free(block, logfs_super(sb)->s_block_pool);
551 static void inode_free_block(struct super_block *sb, struct logfs_block *block)
553 struct inode *inode = block->inode;
555 logfs_inode(inode)->li_block = NULL;
556 __free_block(sb, block);
559 static void indirect_free_block(struct super_block *sb,
560 struct logfs_block *block)
562 ClearPagePrivate(block->page);
563 block->page->private = 0;
564 __free_block(sb, block);
568 static struct logfs_block_ops inode_block_ops = {
569 .write_block = inode_write_block,
570 .free_block = inode_free_block,
571 .write_alias = inode_write_alias,
574 struct logfs_block_ops indirect_block_ops = {
575 .write_block = indirect_write_block,
576 .free_block = indirect_free_block,
577 .write_alias = indirect_write_alias,
580 struct logfs_block *__alloc_block(struct super_block *sb,
581 u64 ino, u64 bix, level_t level)
583 struct logfs_super *super = logfs_super(sb);
584 struct logfs_block *block;
586 block = mempool_alloc(super->s_block_pool, GFP_NOFS);
587 memset(block, 0, sizeof(*block));
588 INIT_LIST_HEAD(&block->alias_list);
589 INIT_LIST_HEAD(&block->item_list);
593 block->level = level;
597 static void alloc_inode_block(struct inode *inode)
599 struct logfs_inode *li = logfs_inode(inode);
600 struct logfs_block *block;
605 block = __alloc_block(inode->i_sb, LOGFS_INO_MASTER, inode->i_ino, 0);
606 block->inode = inode;
607 li->li_block = block;
608 block->ops = &inode_block_ops;
611 void initialize_block_counters(struct page *page, struct logfs_block *block,
612 __be64 *array, int page_is_empty)
620 if (page->index < first_indirect_block()) {
621 /* Counters are pointless on level 0 */
624 if (page->index == first_indirect_block()) {
625 /* Skip unused pointers */
627 block->full = I0_BLOCKS;
629 if (!page_is_empty) {
630 for (i = start; i < LOGFS_BLOCK_FACTOR; i++) {
631 ptr = be64_to_cpu(array[i]);
634 if (ptr & LOGFS_FULLY_POPULATED)
640 static void alloc_data_block(struct inode *inode, struct page *page)
642 struct logfs_block *block;
646 if (PagePrivate(page))
649 logfs_unpack_index(page->index, &bix, &level);
650 block = __alloc_block(inode->i_sb, inode->i_ino, bix, level);
652 SetPagePrivate(page);
653 page->private = (unsigned long)block;
654 block->ops = &indirect_block_ops;
657 static void alloc_indirect_block(struct inode *inode, struct page *page,
660 struct logfs_block *block;
663 if (PagePrivate(page))
666 alloc_data_block(inode, page);
668 block = logfs_block(page);
669 array = kmap_atomic(page, KM_USER0);
670 initialize_block_counters(page, block, array, page_is_empty);
671 kunmap_atomic(array, KM_USER0);
674 static void block_set_pointer(struct page *page, int index, u64 ptr)
676 struct logfs_block *block = logfs_block(page);
681 array = kmap_atomic(page, KM_USER0);
682 oldptr = be64_to_cpu(array[index]);
683 array[index] = cpu_to_be64(ptr);
684 kunmap_atomic(array, KM_USER0);
685 SetPageUptodate(page);
687 block->full += !!(ptr & LOGFS_FULLY_POPULATED)
688 - !!(oldptr & LOGFS_FULLY_POPULATED);
689 block->partial += !!ptr - !!oldptr;
692 static u64 block_get_pointer(struct page *page, int index)
697 block = kmap_atomic(page, KM_USER0);
698 ptr = be64_to_cpu(block[index]);
699 kunmap_atomic(block, KM_USER0);
703 static int logfs_read_empty(struct page *page)
705 zero_user_segment(page, 0, PAGE_CACHE_SIZE);
709 static int logfs_read_direct(struct inode *inode, struct page *page)
711 struct logfs_inode *li = logfs_inode(inode);
712 pgoff_t index = page->index;
715 block = li->li_data[index];
717 return logfs_read_empty(page);
719 return logfs_segment_read(inode, page, block, index, 0);
722 static int logfs_read_loop(struct inode *inode, struct page *page,
725 struct logfs_inode *li = logfs_inode(inode);
726 u64 bix, bofs = li->li_data[INDIRECT_INDEX];
727 level_t level, target_level;
731 logfs_unpack_index(page->index, &bix, &target_level);
733 return logfs_read_empty(page);
735 if (bix >= maxbix(li->li_height))
736 return logfs_read_empty(page);
738 for (level = LEVEL(li->li_height);
739 (__force u8)level > (__force u8)target_level;
740 level = SUBLEVEL(level)){
741 ipage = logfs_get_page(inode, bix, level, rw_context);
745 ret = logfs_segment_read(inode, ipage, bofs, bix, level);
747 logfs_put_read_page(ipage);
751 bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level)));
752 logfs_put_page(ipage, rw_context);
754 return logfs_read_empty(page);
757 return logfs_segment_read(inode, page, bofs, bix, 0);
760 static int logfs_read_block(struct inode *inode, struct page *page,
763 pgoff_t index = page->index;
765 if (index < I0_BLOCKS)
766 return logfs_read_direct(inode, page);
767 return logfs_read_loop(inode, page, rw_context);
770 static int logfs_exist_loop(struct inode *inode, u64 bix)
772 struct logfs_inode *li = logfs_inode(inode);
773 u64 bofs = li->li_data[INDIRECT_INDEX];
780 if (bix >= maxbix(li->li_height))
783 for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)) {
784 ipage = logfs_get_read_page(inode, bix, level);
788 ret = logfs_segment_read(inode, ipage, bofs, bix, level);
790 logfs_put_read_page(ipage);
794 bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level)));
795 logfs_put_read_page(ipage);
803 int logfs_exist_block(struct inode *inode, u64 bix)
805 struct logfs_inode *li = logfs_inode(inode);
808 return !!li->li_data[bix];
809 return logfs_exist_loop(inode, bix);
812 static u64 seek_holedata_direct(struct inode *inode, u64 bix, int data)
814 struct logfs_inode *li = logfs_inode(inode);
816 for (; bix < I0_BLOCKS; bix++)
817 if (data ^ (li->li_data[bix] == 0))
822 static u64 seek_holedata_loop(struct inode *inode, u64 bix, int data)
824 struct logfs_inode *li = logfs_inode(inode);
826 u64 increment, bofs = li->li_data[INDIRECT_INDEX];
833 for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)) {
834 increment = 1 << (LOGFS_BLOCK_BITS * ((__force u8)level-1));
835 page = logfs_get_read_page(inode, bix, level);
839 ret = logfs_segment_read(inode, page, bofs, bix, level);
841 logfs_put_read_page(page);
845 slot = get_bits(bix, SUBLEVEL(level));
846 rblock = kmap_atomic(page, KM_USER0);
847 while (slot < LOGFS_BLOCK_FACTOR) {
848 if (data && (rblock[slot] != 0))
850 if (!data && !(be64_to_cpu(rblock[slot]) & LOGFS_FULLY_POPULATED))
854 bix &= ~(increment - 1);
856 if (slot >= LOGFS_BLOCK_FACTOR) {
857 kunmap_atomic(rblock, KM_USER0);
858 logfs_put_read_page(page);
861 bofs = be64_to_cpu(rblock[slot]);
862 kunmap_atomic(rblock, KM_USER0);
863 logfs_put_read_page(page);
873 * logfs_seek_hole - find next hole starting at a given block index
874 * @inode: inode to search in
875 * @bix: block index to start searching
877 * Returns next hole. If the file doesn't contain any further holes, the
878 * block address next to eof is returned instead.
880 u64 logfs_seek_hole(struct inode *inode, u64 bix)
882 struct logfs_inode *li = logfs_inode(inode);
884 if (bix < I0_BLOCKS) {
885 bix = seek_holedata_direct(inode, bix, 0);
890 if (!li->li_data[INDIRECT_INDEX])
892 else if (li->li_data[INDIRECT_INDEX] & LOGFS_FULLY_POPULATED)
893 bix = maxbix(li->li_height);
895 bix = seek_holedata_loop(inode, bix, 0);
896 if (bix < maxbix(li->li_height))
898 /* Should not happen anymore. But if some port writes semi-
899 * corrupt images (as this one used to) we might run into it.
901 WARN_ON_ONCE(bix == maxbix(li->li_height));
907 static u64 __logfs_seek_data(struct inode *inode, u64 bix)
909 struct logfs_inode *li = logfs_inode(inode);
911 if (bix < I0_BLOCKS) {
912 bix = seek_holedata_direct(inode, bix, 1);
917 if (bix < maxbix(li->li_height)) {
918 if (!li->li_data[INDIRECT_INDEX])
919 bix = maxbix(li->li_height);
921 return seek_holedata_loop(inode, bix, 1);
928 * logfs_seek_data - find next data block after a given block index
929 * @inode: inode to search in
930 * @bix: block index to start searching
932 * Returns next data block. If the file doesn't contain any further data
933 * blocks, the last block in the file is returned instead.
935 u64 logfs_seek_data(struct inode *inode, u64 bix)
937 struct super_block *sb = inode->i_sb;
940 ret = __logfs_seek_data(inode, bix);
941 end = i_size_read(inode) >> sb->s_blocksize_bits;
947 static int logfs_is_valid_direct(struct logfs_inode *li, u64 bix, u64 ofs)
949 return pure_ofs(li->li_data[bix]) == ofs;
952 static int __logfs_is_valid_loop(struct inode *inode, u64 bix,
955 struct logfs_inode *li = logfs_inode(inode);
960 for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)){
961 page = logfs_get_write_page(inode, bix, level);
964 ret = logfs_segment_read(inode, page, bofs, bix, level);
966 logfs_put_write_page(page);
970 bofs = block_get_pointer(page, get_bits(bix, SUBLEVEL(level)));
971 logfs_put_write_page(page);
975 if (pure_ofs(bofs) == ofs)
981 static int logfs_is_valid_loop(struct inode *inode, u64 bix, u64 ofs)
983 struct logfs_inode *li = logfs_inode(inode);
984 u64 bofs = li->li_data[INDIRECT_INDEX];
989 if (bix >= maxbix(li->li_height))
992 if (pure_ofs(bofs) == ofs)
995 return __logfs_is_valid_loop(inode, bix, ofs, bofs);
998 static int __logfs_is_valid_block(struct inode *inode, u64 bix, u64 ofs)
1000 struct logfs_inode *li = logfs_inode(inode);
1002 if ((inode->i_nlink == 0) && atomic_read(&inode->i_count) == 1)
1005 if (bix < I0_BLOCKS)
1006 return logfs_is_valid_direct(li, bix, ofs);
1007 return logfs_is_valid_loop(inode, bix, ofs);
1011 * logfs_is_valid_block - check whether this block is still valid
1014 * @ofs - block physical offset
1015 * @ino - block inode number
1016 * @bix - block index
1017 * @level - block level
1019 * Returns 0 if the block is invalid, 1 if it is valid and 2 if it will
1020 * become invalid once the journal is written.
1022 int logfs_is_valid_block(struct super_block *sb, u64 ofs, u64 ino, u64 bix,
1023 gc_level_t gc_level)
1025 struct logfs_super *super = logfs_super(sb);
1026 struct inode *inode;
1029 /* Umount closes a segment with free blocks remaining. Those
1030 * blocks are by definition invalid. */
1034 LOGFS_BUG_ON((u64)(u_long)ino != ino, sb);
1036 inode = logfs_safe_iget(sb, ino, &cookie);
1040 ret = __logfs_is_valid_block(inode, bix, ofs);
1041 logfs_safe_iput(inode, cookie);
1046 /* Block is nominally invalid, but may still sit in the shadow tree,
1047 * waiting for a journal commit.
1049 if (btree_lookup64(&super->s_shadow_tree.old, ofs))
1054 int logfs_readpage_nolock(struct page *page)
1056 struct inode *inode = page->mapping->host;
1059 ret = logfs_read_block(inode, page, READ);
1062 ClearPageUptodate(page);
1065 SetPageUptodate(page);
1066 ClearPageError(page);
1068 flush_dcache_page(page);
1073 static int logfs_reserve_bytes(struct inode *inode, int bytes)
1075 struct logfs_super *super = logfs_super(inode->i_sb);
1076 u64 available = super->s_free_bytes + super->s_dirty_free_bytes
1077 - super->s_dirty_used_bytes - super->s_dirty_pages;
1082 if (available < bytes)
1085 if (available < bytes + super->s_root_reserve &&
1086 !capable(CAP_SYS_RESOURCE))
1092 int get_page_reserve(struct inode *inode, struct page *page)
1094 struct logfs_super *super = logfs_super(inode->i_sb);
1097 if (logfs_block(page) && logfs_block(page)->reserved_bytes)
1100 logfs_get_wblocks(inode->i_sb, page, WF_LOCK);
1101 ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE);
1103 alloc_data_block(inode, page);
1104 logfs_block(page)->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE;
1105 super->s_dirty_pages += 6 * LOGFS_MAX_OBJECTSIZE;
1107 logfs_put_wblocks(inode->i_sb, page, WF_LOCK);
1112 * We are protected by write lock. Push victims up to superblock level
1113 * and release transaction when appropriate.
1115 /* FIXME: This is currently called from the wrong spots. */
1116 static void logfs_handle_transaction(struct inode *inode,
1117 struct logfs_transaction *ta)
1119 struct logfs_super *super = logfs_super(inode->i_sb);
1123 logfs_inode(inode)->li_block->ta = NULL;
1125 if (inode->i_ino != LOGFS_INO_MASTER) {
1126 BUG(); /* FIXME: Yes, this needs more thought */
1127 /* just remember the transaction until inode is written */
1128 //BUG_ON(logfs_inode(inode)->li_transaction);
1129 //logfs_inode(inode)->li_transaction = ta;
1133 switch (ta->state) {
1134 case CREATE_1: /* fall through */
1136 BUG_ON(super->s_victim_ino);
1137 super->s_victim_ino = ta->ino;
1139 case CREATE_2: /* fall through */
1141 BUG_ON(super->s_victim_ino != ta->ino);
1142 super->s_victim_ino = 0;
1143 /* transaction ends here - free it */
1146 case CROSS_RENAME_1:
1147 BUG_ON(super->s_rename_dir);
1148 BUG_ON(super->s_rename_pos);
1149 super->s_rename_dir = ta->dir;
1150 super->s_rename_pos = ta->pos;
1152 case CROSS_RENAME_2:
1153 BUG_ON(super->s_rename_dir != ta->dir);
1154 BUG_ON(super->s_rename_pos != ta->pos);
1155 super->s_rename_dir = 0;
1156 super->s_rename_pos = 0;
1159 case TARGET_RENAME_1:
1160 BUG_ON(super->s_rename_dir);
1161 BUG_ON(super->s_rename_pos);
1162 BUG_ON(super->s_victim_ino);
1163 super->s_rename_dir = ta->dir;
1164 super->s_rename_pos = ta->pos;
1165 super->s_victim_ino = ta->ino;
1167 case TARGET_RENAME_2:
1168 BUG_ON(super->s_rename_dir != ta->dir);
1169 BUG_ON(super->s_rename_pos != ta->pos);
1170 BUG_ON(super->s_victim_ino != ta->ino);
1171 super->s_rename_dir = 0;
1172 super->s_rename_pos = 0;
1174 case TARGET_RENAME_3:
1175 BUG_ON(super->s_rename_dir);
1176 BUG_ON(super->s_rename_pos);
1177 BUG_ON(super->s_victim_ino != ta->ino);
1178 super->s_victim_ino = 0;
1187 * Not strictly a reservation, but rather a check that we still have enough
1188 * space to satisfy the write.
1190 static int logfs_reserve_blocks(struct inode *inode, int blocks)
1192 return logfs_reserve_bytes(inode, blocks * LOGFS_MAX_OBJECTSIZE);
1195 struct write_control {
1200 static struct logfs_shadow *alloc_shadow(struct inode *inode, u64 bix,
1201 level_t level, u64 old_ofs)
1203 struct logfs_super *super = logfs_super(inode->i_sb);
1204 struct logfs_shadow *shadow;
1206 shadow = mempool_alloc(super->s_shadow_pool, GFP_NOFS);
1207 memset(shadow, 0, sizeof(*shadow));
1208 shadow->ino = inode->i_ino;
1210 shadow->gc_level = expand_level(inode->i_ino, level);
1211 shadow->old_ofs = old_ofs & ~LOGFS_FULLY_POPULATED;
1215 static void free_shadow(struct inode *inode, struct logfs_shadow *shadow)
1217 struct logfs_super *super = logfs_super(inode->i_sb);
1219 mempool_free(shadow, super->s_shadow_pool);
1222 static void mark_segment(struct shadow_tree *tree, u32 segno)
1226 if (!btree_lookup32(&tree->segment_map, segno)) {
1227 err = btree_insert32(&tree->segment_map, segno, (void *)1,
1230 tree->no_shadowed_segments++;
1235 * fill_shadow_tree - Propagate shadow tree changes due to a write
1236 * @inode: Inode owning the page
1237 * @page: Struct page that was written
1238 * @shadow: Shadow for the current write
1240 * Writes in logfs can result in two semi-valid objects. The old object
1241 * is still valid as long as it can be reached by following pointers on
1242 * the medium. Only when writes propagate all the way up to the journal
1243 * has the new object safely replaced the old one.
1245 * To handle this problem, a struct logfs_shadow is used to represent
1246 * every single write. It is attached to the indirect block, which is
1247 * marked dirty. When the indirect block is written, its shadows are
1248 * handed up to the next indirect block (or inode). Untimately they
1249 * will reach the master inode and be freed upon journal commit.
1251 * This function handles a single step in the propagation. It adds the
1252 * shadow for the current write to the tree, along with any shadows in
1253 * the page's tree, in case it was an indirect block. If a page is
1254 * written, the inode parameter is left NULL, if an inode is written,
1255 * the page parameter is left NULL.
1257 static void fill_shadow_tree(struct inode *inode, struct page *page,
1258 struct logfs_shadow *shadow)
1260 struct logfs_super *super = logfs_super(inode->i_sb);
1261 struct logfs_block *block = logfs_block(page);
1262 struct shadow_tree *tree = &super->s_shadow_tree;
1264 if (PagePrivate(page)) {
1265 if (block->alias_map)
1266 super->s_no_object_aliases -= bitmap_weight(
1267 block->alias_map, LOGFS_BLOCK_FACTOR);
1268 logfs_handle_transaction(inode, block->ta);
1269 block->ops->free_block(inode->i_sb, block);
1272 if (shadow->old_ofs)
1273 btree_insert64(&tree->old, shadow->old_ofs, shadow,
1276 btree_insert64(&tree->new, shadow->new_ofs, shadow,
1279 super->s_dirty_used_bytes += shadow->new_len;
1280 super->s_dirty_free_bytes += shadow->old_len;
1281 mark_segment(tree, shadow->old_ofs >> super->s_segshift);
1282 mark_segment(tree, shadow->new_ofs >> super->s_segshift);
1286 static void logfs_set_alias(struct super_block *sb, struct logfs_block *block,
1289 struct logfs_super *super = logfs_super(sb);
1291 if (block->inode && block->inode->i_ino == LOGFS_INO_MASTER) {
1292 /* Aliases in the master inode are pointless. */
1296 if (!test_bit(child_no, block->alias_map)) {
1297 set_bit(child_no, block->alias_map);
1298 super->s_no_object_aliases++;
1300 list_move_tail(&block->alias_list, &super->s_object_alias);
1304 * Object aliases can and often do change the size and occupied space of a
1305 * file. So not only do we have to change the pointers, we also have to
1306 * change inode->i_size and li->li_used_bytes. Which is done by setting
1307 * another two object aliases for the inode itself.
1309 static void set_iused(struct inode *inode, struct logfs_shadow *shadow)
1311 struct logfs_inode *li = logfs_inode(inode);
1313 if (shadow->new_len == shadow->old_len)
1316 alloc_inode_block(inode);
1317 li->li_used_bytes += shadow->new_len - shadow->old_len;
1318 __logfs_set_blocks(inode);
1319 logfs_set_alias(inode->i_sb, li->li_block, INODE_USED_OFS);
1320 logfs_set_alias(inode->i_sb, li->li_block, INODE_SIZE_OFS);
1323 static int logfs_write_i0(struct inode *inode, struct page *page,
1324 struct write_control *wc)
1326 struct logfs_shadow *shadow;
1331 logfs_unpack_index(page->index, &bix, &level);
1333 if (logfs_reserve_blocks(inode, 1))
1336 shadow = alloc_shadow(inode, bix, level, wc->ofs);
1337 if (wc->flags & WF_WRITE)
1338 err = logfs_segment_write(inode, page, shadow);
1339 if (wc->flags & WF_DELETE)
1340 logfs_segment_delete(inode, shadow);
1342 free_shadow(inode, shadow);
1346 set_iused(inode, shadow);
1349 alloc_indirect_block(inode, page, 0);
1350 full = logfs_block(page)->full == LOGFS_BLOCK_FACTOR;
1352 fill_shadow_tree(inode, page, shadow);
1353 wc->ofs = shadow->new_ofs;
1354 if (wc->ofs && full)
1355 wc->ofs |= LOGFS_FULLY_POPULATED;
1359 static int logfs_write_direct(struct inode *inode, struct page *page,
1362 struct logfs_inode *li = logfs_inode(inode);
1363 struct write_control wc = {
1364 .ofs = li->li_data[page->index],
1369 alloc_inode_block(inode);
1371 err = logfs_write_i0(inode, page, &wc);
1375 li->li_data[page->index] = wc.ofs;
1376 logfs_set_alias(inode->i_sb, li->li_block,
1377 page->index + INODE_POINTER_OFS);
1381 static int ptr_change(u64 ofs, struct page *page)
1383 struct logfs_block *block = logfs_block(page);
1384 int empty0, empty1, full0, full1;
1387 empty1 = block->partial == 0;
1388 if (empty0 != empty1)
1391 /* The !! is necessary to shrink result to int */
1392 full0 = !!(ofs & LOGFS_FULLY_POPULATED);
1393 full1 = block->full == LOGFS_BLOCK_FACTOR;
1399 static int __logfs_write_rec(struct inode *inode, struct page *page,
1400 struct write_control *this_wc,
1401 pgoff_t bix, level_t target_level, level_t level)
1403 int ret, page_empty = 0;
1404 int child_no = get_bits(bix, SUBLEVEL(level));
1406 struct write_control child_wc = {
1407 .flags = this_wc->flags,
1410 ipage = logfs_get_write_page(inode, bix, level);
1415 ret = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level);
1418 } else if (!PageUptodate(ipage)) {
1420 logfs_read_empty(ipage);
1423 child_wc.ofs = block_get_pointer(ipage, child_no);
1425 if ((__force u8)level-1 > (__force u8)target_level)
1426 ret = __logfs_write_rec(inode, page, &child_wc, bix,
1427 target_level, SUBLEVEL(level));
1429 ret = logfs_write_i0(inode, page, &child_wc);
1434 alloc_indirect_block(inode, ipage, page_empty);
1435 block_set_pointer(ipage, child_no, child_wc.ofs);
1436 /* FIXME: first condition seems superfluous */
1437 if (child_wc.ofs || logfs_block(ipage)->partial)
1438 this_wc->flags |= WF_WRITE;
1439 /* the condition on this_wc->ofs ensures that we won't consume extra
1440 * space for indirect blocks in the future, which we cannot reserve */
1441 if (!this_wc->ofs || ptr_change(this_wc->ofs, ipage))
1442 ret = logfs_write_i0(inode, ipage, this_wc);
1444 logfs_set_alias(inode->i_sb, logfs_block(ipage), child_no);
1446 logfs_put_write_page(ipage);
1450 static int logfs_write_rec(struct inode *inode, struct page *page,
1451 pgoff_t bix, level_t target_level, long flags)
1453 struct logfs_inode *li = logfs_inode(inode);
1454 struct write_control wc = {
1455 .ofs = li->li_data[INDIRECT_INDEX],
1460 alloc_inode_block(inode);
1462 if (li->li_height > (__force u8)target_level)
1463 ret = __logfs_write_rec(inode, page, &wc, bix, target_level,
1464 LEVEL(li->li_height));
1466 ret = logfs_write_i0(inode, page, &wc);
1470 if (li->li_data[INDIRECT_INDEX] != wc.ofs) {
1471 li->li_data[INDIRECT_INDEX] = wc.ofs;
1472 logfs_set_alias(inode->i_sb, li->li_block,
1473 INDIRECT_INDEX + INODE_POINTER_OFS);
1478 void logfs_add_transaction(struct inode *inode, struct logfs_transaction *ta)
1480 alloc_inode_block(inode);
1481 logfs_inode(inode)->li_block->ta = ta;
1484 void logfs_del_transaction(struct inode *inode, struct logfs_transaction *ta)
1486 struct logfs_block *block = logfs_inode(inode)->li_block;
1488 if (block && block->ta)
1492 static int grow_inode(struct inode *inode, u64 bix, level_t level)
1494 struct logfs_inode *li = logfs_inode(inode);
1495 u8 height = (__force u8)level;
1497 struct write_control wc = {
1502 BUG_ON(height > 5 || li->li_height > 5);
1503 while (height > li->li_height || bix >= maxbix(li->li_height)) {
1504 page = logfs_get_write_page(inode, I0_BLOCKS + 1,
1505 LEVEL(li->li_height + 1));
1508 logfs_read_empty(page);
1509 alloc_indirect_block(inode, page, 1);
1510 block_set_pointer(page, 0, li->li_data[INDIRECT_INDEX]);
1511 err = logfs_write_i0(inode, page, &wc);
1512 logfs_put_write_page(page);
1515 li->li_data[INDIRECT_INDEX] = wc.ofs;
1518 logfs_set_alias(inode->i_sb, li->li_block, INODE_HEIGHT_OFS);
1523 static int __logfs_write_buf(struct inode *inode, struct page *page, long flags)
1525 struct logfs_super *super = logfs_super(inode->i_sb);
1526 pgoff_t index = page->index;
1531 flags |= WF_WRITE | WF_DELETE;
1532 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1534 logfs_unpack_index(index, &bix, &level);
1535 if (logfs_block(page) && logfs_block(page)->reserved_bytes)
1536 super->s_dirty_pages -= logfs_block(page)->reserved_bytes;
1538 if (index < I0_BLOCKS)
1539 return logfs_write_direct(inode, page, flags);
1541 bix = adjust_bix(bix, level);
1542 err = grow_inode(inode, bix, level);
1545 return logfs_write_rec(inode, page, bix, level, flags);
1548 int logfs_write_buf(struct inode *inode, struct page *page, long flags)
1550 struct super_block *sb = inode->i_sb;
1553 logfs_get_wblocks(sb, page, flags & WF_LOCK);
1554 ret = __logfs_write_buf(inode, page, flags);
1555 logfs_put_wblocks(sb, page, flags & WF_LOCK);
1559 static int __logfs_delete(struct inode *inode, struct page *page)
1561 long flags = WF_DELETE;
1563 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1565 if (page->index < I0_BLOCKS)
1566 return logfs_write_direct(inode, page, flags);
1567 return logfs_write_rec(inode, page, page->index, 0, flags);
1570 int logfs_delete(struct inode *inode, pgoff_t index,
1571 struct shadow_tree *shadow_tree)
1573 struct super_block *sb = inode->i_sb;
1577 page = logfs_get_read_page(inode, index, 0);
1581 logfs_get_wblocks(sb, page, 1);
1582 ret = __logfs_delete(inode, page);
1583 logfs_put_wblocks(sb, page, 1);
1585 logfs_put_read_page(page);
1590 int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs,
1591 gc_level_t gc_level, long flags)
1593 level_t level = shrink_level(gc_level);
1597 page = logfs_get_write_page(inode, bix, level);
1601 err = logfs_segment_read(inode, page, ofs, bix, level);
1604 alloc_indirect_block(inode, page, 0);
1605 err = logfs_write_buf(inode, page, flags);
1606 if (!err && shrink_level(gc_level) == 0) {
1607 /* Rewrite cannot mark the inode dirty but has to
1608 * write it immediatly.
1609 * Q: Can't we just create an alias for the inode
1610 * instead? And if not, why not?
1612 if (inode->i_ino == LOGFS_INO_MASTER)
1613 logfs_write_anchor(inode->i_sb);
1615 err = __logfs_write_inode(inode, flags);
1619 logfs_put_write_page(page);
1623 static int truncate_data_block(struct inode *inode, struct page *page,
1624 u64 ofs, struct logfs_shadow *shadow, u64 size)
1626 loff_t pageofs = page->index << inode->i_sb->s_blocksize_bits;
1631 /* Does truncation happen within this page? */
1632 if (size <= pageofs || size - pageofs >= PAGE_SIZE)
1635 logfs_unpack_index(page->index, &bix, &level);
1638 err = logfs_segment_read(inode, page, ofs, bix, level);
1642 zero_user_segment(page, size - pageofs, PAGE_CACHE_SIZE);
1643 return logfs_segment_write(inode, page, shadow);
1646 static int logfs_truncate_i0(struct inode *inode, struct page *page,
1647 struct write_control *wc, u64 size)
1649 struct logfs_shadow *shadow;
1654 logfs_unpack_index(page->index, &bix, &level);
1656 shadow = alloc_shadow(inode, bix, level, wc->ofs);
1658 err = truncate_data_block(inode, page, wc->ofs, shadow, size);
1660 free_shadow(inode, shadow);
1664 logfs_segment_delete(inode, shadow);
1665 set_iused(inode, shadow);
1666 fill_shadow_tree(inode, page, shadow);
1667 wc->ofs = shadow->new_ofs;
1671 static int logfs_truncate_direct(struct inode *inode, u64 size)
1673 struct logfs_inode *li = logfs_inode(inode);
1674 struct write_control wc;
1679 alloc_inode_block(inode);
1681 for (e = I0_BLOCKS - 1; e >= 0; e--) {
1682 if (size > (e+1) * LOGFS_BLOCKSIZE)
1685 wc.ofs = li->li_data[e];
1689 page = logfs_get_write_page(inode, e, 0);
1692 err = logfs_segment_read(inode, page, wc.ofs, e, 0);
1694 logfs_put_write_page(page);
1697 err = logfs_truncate_i0(inode, page, &wc, size);
1698 logfs_put_write_page(page);
1702 li->li_data[e] = wc.ofs;
1707 /* FIXME: these need to become per-sb once we support different blocksizes */
1708 static u64 __logfs_step[] = {
1715 static u64 __logfs_start_index[] = {
1722 static inline u64 logfs_step(level_t level)
1724 return __logfs_step[(__force u8)level];
1727 static inline u64 logfs_factor(u8 level)
1729 return __logfs_step[level] * LOGFS_BLOCKSIZE;
1732 static inline u64 logfs_start_index(level_t level)
1734 return __logfs_start_index[(__force u8)level];
1737 static void logfs_unpack_raw_index(pgoff_t index, u64 *bix, level_t *level)
1739 logfs_unpack_index(index, bix, level);
1740 if (*bix <= logfs_start_index(SUBLEVEL(*level)))
1744 static int __logfs_truncate_rec(struct inode *inode, struct page *ipage,
1745 struct write_control *this_wc, u64 size)
1747 int truncate_happened = 0;
1749 u64 bix, child_bix, next_bix;
1752 struct write_control child_wc = { /* FIXME: flags */ };
1754 logfs_unpack_raw_index(ipage->index, &bix, &level);
1755 err = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level);
1759 for (e = LOGFS_BLOCK_FACTOR - 1; e >= 0; e--) {
1760 child_bix = bix + e * logfs_step(SUBLEVEL(level));
1761 next_bix = child_bix + logfs_step(SUBLEVEL(level));
1762 if (size > next_bix * LOGFS_BLOCKSIZE)
1765 child_wc.ofs = pure_ofs(block_get_pointer(ipage, e));
1769 page = logfs_get_write_page(inode, child_bix, SUBLEVEL(level));
1773 if ((__force u8)level > 1)
1774 err = __logfs_truncate_rec(inode, page, &child_wc, size);
1776 err = logfs_truncate_i0(inode, page, &child_wc, size);
1777 logfs_put_write_page(page);
1781 truncate_happened = 1;
1782 alloc_indirect_block(inode, ipage, 0);
1783 block_set_pointer(ipage, e, child_wc.ofs);
1786 if (!truncate_happened) {
1787 printk("ineffectual truncate (%lx, %lx, %llx)\n", inode->i_ino, ipage->index, size);
1791 this_wc->flags = WF_DELETE;
1792 if (logfs_block(ipage)->partial)
1793 this_wc->flags |= WF_WRITE;
1795 return logfs_write_i0(inode, ipage, this_wc);
1798 static int logfs_truncate_rec(struct inode *inode, u64 size)
1800 struct logfs_inode *li = logfs_inode(inode);
1801 struct write_control wc = {
1802 .ofs = li->li_data[INDIRECT_INDEX],
1807 alloc_inode_block(inode);
1812 page = logfs_get_write_page(inode, 0, LEVEL(li->li_height));
1816 err = __logfs_truncate_rec(inode, page, &wc, size);
1817 logfs_put_write_page(page);
1821 if (li->li_data[INDIRECT_INDEX] != wc.ofs)
1822 li->li_data[INDIRECT_INDEX] = wc.ofs;
1826 static int __logfs_truncate(struct inode *inode, u64 size)
1830 if (size >= logfs_factor(logfs_inode(inode)->li_height))
1833 ret = logfs_truncate_rec(inode, size);
1837 return logfs_truncate_direct(inode, size);
1840 int logfs_truncate(struct inode *inode, u64 size)
1842 struct super_block *sb = inode->i_sb;
1845 logfs_get_wblocks(sb, NULL, 1);
1846 err = __logfs_truncate(inode, size);
1848 err = __logfs_write_inode(inode, 0);
1849 logfs_put_wblocks(sb, NULL, 1);
1852 err = vmtruncate(inode, size);
1854 /* I don't trust error recovery yet. */
1859 static void move_page_to_inode(struct inode *inode, struct page *page)
1861 struct logfs_inode *li = logfs_inode(inode);
1862 struct logfs_block *block = logfs_block(page);
1867 log_blockmove("move_page_to_inode(%llx, %llx, %x)\n",
1868 block->ino, block->bix, block->level);
1869 BUG_ON(li->li_block);
1870 block->ops = &inode_block_ops;
1871 block->inode = inode;
1872 li->li_block = block;
1876 ClearPagePrivate(page);
1879 static void move_inode_to_page(struct page *page, struct inode *inode)
1881 struct logfs_inode *li = logfs_inode(inode);
1882 struct logfs_block *block = li->li_block;
1887 log_blockmove("move_inode_to_page(%llx, %llx, %x)\n",
1888 block->ino, block->bix, block->level);
1889 BUG_ON(PagePrivate(page));
1890 block->ops = &indirect_block_ops;
1892 page->private = (unsigned long)block;
1893 SetPagePrivate(page);
1895 block->inode = NULL;
1896 li->li_block = NULL;
1899 int logfs_read_inode(struct inode *inode)
1901 struct super_block *sb = inode->i_sb;
1902 struct logfs_super *super = logfs_super(sb);
1903 struct inode *master_inode = super->s_master_inode;
1905 struct logfs_disk_inode *di;
1906 u64 ino = inode->i_ino;
1908 if (ino << sb->s_blocksize_bits > i_size_read(master_inode))
1910 if (!logfs_exist_block(master_inode, ino))
1913 page = read_cache_page(master_inode->i_mapping, ino,
1914 (filler_t *)logfs_readpage, NULL);
1916 return PTR_ERR(page);
1918 di = kmap_atomic(page, KM_USER0);
1919 logfs_disk_to_inode(di, inode);
1920 kunmap_atomic(di, KM_USER0);
1921 move_page_to_inode(inode, page);
1922 page_cache_release(page);
1926 /* Caller must logfs_put_write_page(page); */
1927 static struct page *inode_to_page(struct inode *inode)
1929 struct inode *master_inode = logfs_super(inode->i_sb)->s_master_inode;
1930 struct logfs_disk_inode *di;
1933 BUG_ON(inode->i_ino == LOGFS_INO_MASTER);
1935 page = logfs_get_write_page(master_inode, inode->i_ino, 0);
1939 di = kmap_atomic(page, KM_USER0);
1940 logfs_inode_to_disk(inode, di);
1941 kunmap_atomic(di, KM_USER0);
1942 move_inode_to_page(page, inode);
1946 /* Cheaper version of write_inode. All changes are concealed in
1947 * aliases, which are moved back. No write to the medium happens.
1949 void logfs_clear_inode(struct inode *inode)
1951 struct super_block *sb = inode->i_sb;
1952 struct logfs_inode *li = logfs_inode(inode);
1953 struct logfs_block *block = li->li_block;
1956 /* Only deleted files may be dirty at this point */
1957 BUG_ON(inode->i_state & I_DIRTY && inode->i_nlink);
1960 if ((logfs_super(sb)->s_flags & LOGFS_SB_FLAG_SHUTDOWN)) {
1961 block->ops->free_block(inode->i_sb, block);
1965 BUG_ON(inode->i_ino < LOGFS_RESERVED_INOS);
1966 page = inode_to_page(inode);
1967 BUG_ON(!page); /* FIXME: Use emergency page */
1968 logfs_put_write_page(page);
1971 static int do_write_inode(struct inode *inode)
1973 struct super_block *sb = inode->i_sb;
1974 struct inode *master_inode = logfs_super(sb)->s_master_inode;
1975 loff_t size = (inode->i_ino + 1) << inode->i_sb->s_blocksize_bits;
1979 BUG_ON(inode->i_ino == LOGFS_INO_MASTER);
1980 /* FIXME: lock inode */
1982 if (i_size_read(master_inode) < size)
1983 i_size_write(master_inode, size);
1985 /* TODO: Tell vfs this inode is clean now */
1987 page = inode_to_page(inode);
1991 /* FIXME: transaction is part of logfs_block now. Is that enough? */
1992 err = logfs_write_buf(master_inode, page, 0);
1993 logfs_put_write_page(page);
1997 static void logfs_mod_segment_entry(struct super_block *sb, u32 segno,
1999 void (*change_se)(struct logfs_segment_entry *, long),
2002 struct logfs_super *super = logfs_super(sb);
2003 struct inode *inode;
2005 struct logfs_segment_entry *se;
2009 page_no = segno >> (sb->s_blocksize_bits - 3);
2010 child_no = segno & ((sb->s_blocksize >> 3) - 1);
2012 inode = super->s_segfile_inode;
2013 page = logfs_get_write_page(inode, page_no, 0);
2014 BUG_ON(!page); /* FIXME: We need some reserve page for this case */
2015 if (!PageUptodate(page))
2016 logfs_read_block(inode, page, WRITE);
2019 alloc_indirect_block(inode, page, 0);
2020 se = kmap_atomic(page, KM_USER0);
2021 change_se(se + child_no, arg);
2023 logfs_set_alias(sb, logfs_block(page), child_no);
2024 BUG_ON((int)be32_to_cpu(se[child_no].valid) > super->s_segsize);
2026 kunmap_atomic(se, KM_USER0);
2028 logfs_put_write_page(page);
2031 static void __get_segment_entry(struct logfs_segment_entry *se, long _target)
2033 struct logfs_segment_entry *target = (void *)_target;
2038 void logfs_get_segment_entry(struct super_block *sb, u32 segno,
2039 struct logfs_segment_entry *se)
2041 logfs_mod_segment_entry(sb, segno, 0, __get_segment_entry, (long)se);
2044 static void __set_segment_used(struct logfs_segment_entry *se, long increment)
2048 valid = be32_to_cpu(se->valid);
2050 se->valid = cpu_to_be32(valid);
2053 void logfs_set_segment_used(struct super_block *sb, u64 ofs, int increment)
2055 struct logfs_super *super = logfs_super(sb);
2056 u32 segno = ofs >> super->s_segshift;
2061 logfs_mod_segment_entry(sb, segno, 1, __set_segment_used, increment);
2064 static void __set_segment_erased(struct logfs_segment_entry *se, long ec_level)
2066 se->ec_level = cpu_to_be32(ec_level);
2069 void logfs_set_segment_erased(struct super_block *sb, u32 segno, u32 ec,
2070 gc_level_t gc_level)
2072 u32 ec_level = ec << 4 | (__force u8)gc_level;
2074 logfs_mod_segment_entry(sb, segno, 1, __set_segment_erased, ec_level);
2077 static void __set_segment_reserved(struct logfs_segment_entry *se, long ignore)
2079 se->valid = cpu_to_be32(RESERVED);
2082 void logfs_set_segment_reserved(struct super_block *sb, u32 segno)
2084 logfs_mod_segment_entry(sb, segno, 1, __set_segment_reserved, 0);
2087 static void __set_segment_unreserved(struct logfs_segment_entry *se,
2091 se->ec_level = cpu_to_be32(ec_level);
2094 void logfs_set_segment_unreserved(struct super_block *sb, u32 segno, u32 ec)
2096 u32 ec_level = ec << 4;
2098 logfs_mod_segment_entry(sb, segno, 1, __set_segment_unreserved,
2102 int __logfs_write_inode(struct inode *inode, long flags)
2104 struct super_block *sb = inode->i_sb;
2107 logfs_get_wblocks(sb, NULL, flags & WF_LOCK);
2108 ret = do_write_inode(inode);
2109 logfs_put_wblocks(sb, NULL, flags & WF_LOCK);
2113 static int do_delete_inode(struct inode *inode)
2115 struct super_block *sb = inode->i_sb;
2116 struct inode *master_inode = logfs_super(sb)->s_master_inode;
2120 page = logfs_get_write_page(master_inode, inode->i_ino, 0);
2124 move_inode_to_page(page, inode);
2126 logfs_get_wblocks(sb, page, 1);
2127 ret = __logfs_delete(master_inode, page);
2128 logfs_put_wblocks(sb, page, 1);
2130 logfs_put_write_page(page);
2135 * ZOMBIE inodes have already been deleted before and should remain dead,
2136 * if it weren't for valid checking. No need to kill them again here.
2138 void logfs_delete_inode(struct inode *inode)
2140 struct logfs_inode *li = logfs_inode(inode);
2142 if (!(li->li_flags & LOGFS_IF_ZOMBIE)) {
2143 li->li_flags |= LOGFS_IF_ZOMBIE;
2144 if (i_size_read(inode) > 0)
2145 logfs_truncate(inode, 0);
2146 do_delete_inode(inode);
2148 truncate_inode_pages(&inode->i_data, 0);
2152 void btree_write_block(struct logfs_block *block)
2154 struct inode *inode;
2158 inode = logfs_safe_iget(block->sb, block->ino, &cookie);
2159 page = logfs_get_write_page(inode, block->bix, block->level);
2161 err = logfs_readpage_nolock(page);
2163 BUG_ON(!PagePrivate(page));
2164 BUG_ON(logfs_block(page) != block);
2165 err = __logfs_write_buf(inode, page, 0);
2167 BUG_ON(PagePrivate(page) || page->private);
2169 logfs_put_write_page(page);
2170 logfs_safe_iput(inode, cookie);
2174 * logfs_inode_write - write inode or dentry objects
2176 * @inode: parent inode (ifile or directory)
2177 * @buf: object to write (inode or dentry)
2179 * @_pos: object number (file position in blocks/objects)
2180 * @flags: write flags
2181 * @lock: 0 if write lock is already taken, 1 otherwise
2182 * @shadow_tree: shadow below this inode
2184 * FIXME: All caller of this put a 200-300 byte variable on the stack,
2185 * only to call here and do a memcpy from that stack variable. A good
2186 * example of wasted performance and stack space.
2188 int logfs_inode_write(struct inode *inode, const void *buf, size_t count,
2189 loff_t bix, long flags, struct shadow_tree *shadow_tree)
2191 loff_t pos = bix << inode->i_sb->s_blocksize_bits;
2196 BUG_ON(pos & (LOGFS_BLOCKSIZE-1));
2197 BUG_ON(count > LOGFS_BLOCKSIZE);
2198 page = logfs_get_write_page(inode, bix, 0);
2202 pagebuf = kmap_atomic(page, KM_USER0);
2203 memcpy(pagebuf, buf, count);
2204 flush_dcache_page(page);
2205 kunmap_atomic(pagebuf, KM_USER0);
2207 if (i_size_read(inode) < pos + LOGFS_BLOCKSIZE)
2208 i_size_write(inode, pos + LOGFS_BLOCKSIZE);
2210 err = logfs_write_buf(inode, page, flags);
2211 logfs_put_write_page(page);
2215 int logfs_open_segfile(struct super_block *sb)
2217 struct logfs_super *super = logfs_super(sb);
2218 struct inode *inode;
2220 inode = logfs_read_meta_inode(sb, LOGFS_INO_SEGFILE);
2222 return PTR_ERR(inode);
2223 super->s_segfile_inode = inode;
2227 int logfs_init_rw(struct super_block *sb)
2229 struct logfs_super *super = logfs_super(sb);
2230 int min_fill = 3 * super->s_no_blocks;
2232 INIT_LIST_HEAD(&super->s_object_alias);
2233 mutex_init(&super->s_write_mutex);
2234 super->s_block_pool = mempool_create_kmalloc_pool(min_fill,
2235 sizeof(struct logfs_block));
2236 super->s_shadow_pool = mempool_create_kmalloc_pool(min_fill,
2237 sizeof(struct logfs_shadow));
2241 void logfs_cleanup_rw(struct super_block *sb)
2243 struct logfs_super *super = logfs_super(sb);
2245 destroy_meta_inode(super->s_segfile_inode);
2246 if (super->s_block_pool)
2247 mempool_destroy(super->s_block_pool);
2248 if (super->s_shadow_pool)
2249 mempool_destroy(super->s_shadow_pool);