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>
21 #include <linux/slab.h>
23 static u64
adjust_bix(u64 bix
, level_t level
)
29 return max_t(u64
, bix
, I0_BLOCKS
);
31 return max_t(u64
, bix
, I1_BLOCKS
);
33 return max_t(u64
, bix
, I2_BLOCKS
);
35 return max_t(u64
, bix
, I3_BLOCKS
);
37 return max_t(u64
, bix
, I4_BLOCKS
);
44 static inline u64
maxbix(u8 height
)
46 return 1ULL << (LOGFS_BLOCK_BITS
* height
);
50 * The inode address space is cut in two halves. Lower half belongs to data
51 * pages, upper half to indirect blocks. If the high bit (INDIRECT_BIT) is
52 * set, the actual block index (bix) and level can be derived from the page
55 * The lowest three bits of the block index are set to 0 after packing and
56 * unpacking. Since the lowest n bits (9 for 4KiB blocksize) are ignored
57 * anyway this is harmless.
59 #define ARCH_SHIFT (BITS_PER_LONG - 32)
60 #define INDIRECT_BIT (0x80000000UL << ARCH_SHIFT)
61 #define LEVEL_SHIFT (28 + ARCH_SHIFT)
62 static inline pgoff_t
first_indirect_block(void)
64 return INDIRECT_BIT
| (1ULL << LEVEL_SHIFT
);
67 pgoff_t
logfs_pack_index(u64 bix
, level_t level
)
71 BUG_ON(bix
>= INDIRECT_BIT
);
76 index
|= (__force
long)level
<< LEVEL_SHIFT
;
77 index
|= bix
>> ((__force u8
)level
* LOGFS_BLOCK_BITS
);
81 void logfs_unpack_index(pgoff_t index
, u64
*bix
, level_t
*level
)
85 if (!(index
& INDIRECT_BIT
)) {
91 __level
= (index
& ~INDIRECT_BIT
) >> LEVEL_SHIFT
;
92 *level
= LEVEL(__level
);
93 *bix
= (index
<< (__level
* LOGFS_BLOCK_BITS
)) & ~INDIRECT_BIT
;
94 *bix
= adjust_bix(*bix
, *level
);
102 * Time is stored as nanoseconds since the epoch.
104 static struct timespec
be64_to_timespec(__be64 betime
)
106 return ns_to_timespec(be64_to_cpu(betime
));
109 static __be64
timespec_to_be64(struct timespec tsp
)
111 return cpu_to_be64((u64
)tsp
.tv_sec
* NSEC_PER_SEC
+ tsp
.tv_nsec
);
114 static void logfs_disk_to_inode(struct logfs_disk_inode
*di
, struct inode
*inode
)
116 struct logfs_inode
*li
= logfs_inode(inode
);
119 inode
->i_mode
= be16_to_cpu(di
->di_mode
);
120 li
->li_height
= di
->di_height
;
121 li
->li_flags
= be32_to_cpu(di
->di_flags
);
122 inode
->i_uid
= be32_to_cpu(di
->di_uid
);
123 inode
->i_gid
= be32_to_cpu(di
->di_gid
);
124 inode
->i_size
= be64_to_cpu(di
->di_size
);
125 logfs_set_blocks(inode
, be64_to_cpu(di
->di_used_bytes
));
126 inode
->i_atime
= be64_to_timespec(di
->di_atime
);
127 inode
->i_ctime
= be64_to_timespec(di
->di_ctime
);
128 inode
->i_mtime
= be64_to_timespec(di
->di_mtime
);
129 inode
->i_nlink
= be32_to_cpu(di
->di_refcount
);
130 inode
->i_generation
= be32_to_cpu(di
->di_generation
);
132 switch (inode
->i_mode
& S_IFMT
) {
133 case S_IFSOCK
: /* fall through */
134 case S_IFBLK
: /* fall through */
135 case S_IFCHR
: /* fall through */
137 inode
->i_rdev
= be64_to_cpu(di
->di_data
[0]);
139 case S_IFDIR
: /* fall through */
140 case S_IFREG
: /* fall through */
142 for (i
= 0; i
< LOGFS_EMBEDDED_FIELDS
; i
++)
143 li
->li_data
[i
] = be64_to_cpu(di
->di_data
[i
]);
150 static void logfs_inode_to_disk(struct inode
*inode
, struct logfs_disk_inode
*di
)
152 struct logfs_inode
*li
= logfs_inode(inode
);
155 di
->di_mode
= cpu_to_be16(inode
->i_mode
);
156 di
->di_height
= li
->li_height
;
158 di
->di_flags
= cpu_to_be32(li
->li_flags
);
159 di
->di_uid
= cpu_to_be32(inode
->i_uid
);
160 di
->di_gid
= cpu_to_be32(inode
->i_gid
);
161 di
->di_size
= cpu_to_be64(i_size_read(inode
));
162 di
->di_used_bytes
= cpu_to_be64(li
->li_used_bytes
);
163 di
->di_atime
= timespec_to_be64(inode
->i_atime
);
164 di
->di_ctime
= timespec_to_be64(inode
->i_ctime
);
165 di
->di_mtime
= timespec_to_be64(inode
->i_mtime
);
166 di
->di_refcount
= cpu_to_be32(inode
->i_nlink
);
167 di
->di_generation
= cpu_to_be32(inode
->i_generation
);
169 switch (inode
->i_mode
& S_IFMT
) {
170 case S_IFSOCK
: /* fall through */
171 case S_IFBLK
: /* fall through */
172 case S_IFCHR
: /* fall through */
174 di
->di_data
[0] = cpu_to_be64(inode
->i_rdev
);
176 case S_IFDIR
: /* fall through */
177 case S_IFREG
: /* fall through */
179 for (i
= 0; i
< LOGFS_EMBEDDED_FIELDS
; i
++)
180 di
->di_data
[i
] = cpu_to_be64(li
->li_data
[i
]);
187 static void __logfs_set_blocks(struct inode
*inode
)
189 struct super_block
*sb
= inode
->i_sb
;
190 struct logfs_inode
*li
= logfs_inode(inode
);
192 inode
->i_blocks
= ULONG_MAX
;
193 if (li
->li_used_bytes
>> sb
->s_blocksize_bits
< ULONG_MAX
)
194 inode
->i_blocks
= ALIGN(li
->li_used_bytes
, 512) >> 9;
197 void logfs_set_blocks(struct inode
*inode
, u64 bytes
)
199 struct logfs_inode
*li
= logfs_inode(inode
);
201 li
->li_used_bytes
= bytes
;
202 __logfs_set_blocks(inode
);
205 static void prelock_page(struct super_block
*sb
, struct page
*page
, int lock
)
207 struct logfs_super
*super
= logfs_super(sb
);
209 BUG_ON(!PageLocked(page
));
211 BUG_ON(PagePreLocked(page
));
212 SetPagePreLocked(page
);
214 /* We are in GC path. */
215 if (PagePreLocked(page
))
216 super
->s_lock_count
++;
218 SetPagePreLocked(page
);
222 static void preunlock_page(struct super_block
*sb
, struct page
*page
, int lock
)
224 struct logfs_super
*super
= logfs_super(sb
);
226 BUG_ON(!PageLocked(page
));
228 ClearPagePreLocked(page
);
230 /* We are in GC path. */
231 BUG_ON(!PagePreLocked(page
));
232 if (super
->s_lock_count
)
233 super
->s_lock_count
--;
235 ClearPagePreLocked(page
);
240 * Logfs is prone to an AB-BA deadlock where one task tries to acquire
241 * s_write_mutex with a locked page and GC tries to get that page while holding
243 * To solve this issue logfs will ignore the page lock iff the page in question
244 * is waiting for s_write_mutex. We annotate this fact by setting PG_pre_locked
245 * in addition to PG_locked.
247 static void logfs_get_wblocks(struct super_block
*sb
, struct page
*page
,
250 struct logfs_super
*super
= logfs_super(sb
);
253 prelock_page(sb
, page
, lock
);
256 mutex_lock(&super
->s_write_mutex
);
258 /* FIXME: We also have to check for shadowed space
259 * and mempool fill grade */
263 static void logfs_put_wblocks(struct super_block
*sb
, struct page
*page
,
266 struct logfs_super
*super
= logfs_super(sb
);
269 preunlock_page(sb
, page
, lock
);
270 /* Order matters - we must clear PG_pre_locked before releasing
271 * s_write_mutex or we could race against another task. */
273 mutex_unlock(&super
->s_write_mutex
);
276 static struct page
*logfs_get_read_page(struct inode
*inode
, u64 bix
,
279 return find_or_create_page(inode
->i_mapping
,
280 logfs_pack_index(bix
, level
), GFP_NOFS
);
283 static void logfs_put_read_page(struct page
*page
)
286 page_cache_release(page
);
289 static void logfs_lock_write_page(struct page
*page
)
293 while (unlikely(!trylock_page(page
))) {
294 if (loop
++ > 0x1000) {
295 /* Has been observed once so far... */
296 printk(KERN_ERR
"stack at %p\n", &loop
);
299 if (PagePreLocked(page
)) {
300 /* Holder of page lock is waiting for us, it
301 * is safe to use this page. */
304 /* Some other process has this page locked and has
305 * nothing to do with us. Wait for it to finish.
309 BUG_ON(!PageLocked(page
));
312 static struct page
*logfs_get_write_page(struct inode
*inode
, u64 bix
,
315 struct address_space
*mapping
= inode
->i_mapping
;
316 pgoff_t index
= logfs_pack_index(bix
, level
);
321 page
= find_get_page(mapping
, index
);
323 page
= __page_cache_alloc(GFP_NOFS
);
326 err
= add_to_page_cache_lru(page
, mapping
, index
, GFP_NOFS
);
328 page_cache_release(page
);
333 } else logfs_lock_write_page(page
);
334 BUG_ON(!PageLocked(page
));
338 static void logfs_unlock_write_page(struct page
*page
)
340 if (!PagePreLocked(page
))
344 static void logfs_put_write_page(struct page
*page
)
346 logfs_unlock_write_page(page
);
347 page_cache_release(page
);
350 static struct page
*logfs_get_page(struct inode
*inode
, u64 bix
, level_t level
,
354 return logfs_get_read_page(inode
, bix
, level
);
356 return logfs_get_write_page(inode
, bix
, level
);
359 static void logfs_put_page(struct page
*page
, int rw
)
362 logfs_put_read_page(page
);
364 logfs_put_write_page(page
);
367 static unsigned long __get_bits(u64 val
, int skip
, int no
)
377 static unsigned long get_bits(u64 val
, level_t skip
)
379 return __get_bits(val
, (__force
int)skip
, LOGFS_BLOCK_BITS
);
382 static inline void init_shadow_tree(struct super_block
*sb
,
383 struct shadow_tree
*tree
)
385 struct logfs_super
*super
= logfs_super(sb
);
387 btree_init_mempool64(&tree
->new, super
->s_btree_pool
);
388 btree_init_mempool64(&tree
->old
, super
->s_btree_pool
);
391 static void indirect_write_block(struct logfs_block
*block
)
398 inode
= page
->mapping
->host
;
399 logfs_lock_write_page(page
);
400 ret
= logfs_write_buf(inode
, page
, 0);
401 logfs_unlock_write_page(page
);
403 * This needs some rework. Unless you want your filesystem to run
404 * completely synchronously (you don't), the filesystem will always
405 * report writes as 'successful' before the actual work has been
406 * done. The actual work gets done here and this is where any errors
407 * will show up. And there isn't much we can do about it, really.
409 * Some attempts to fix the errors (move from bad blocks, retry io,...)
410 * have already been done, so anything left should be either a broken
411 * device or a bug somewhere in logfs itself. Being relatively new,
412 * the odds currently favor a bug, so for now the line below isn't
418 static void inode_write_block(struct logfs_block
*block
)
423 inode
= block
->inode
;
424 if (inode
->i_ino
== LOGFS_INO_MASTER
)
425 logfs_write_anchor(inode
->i_sb
);
427 ret
= __logfs_write_inode(inode
, 0);
428 /* see indirect_write_block comment */
433 static gc_level_t
inode_block_level(struct logfs_block
*block
)
435 BUG_ON(block
->inode
->i_ino
== LOGFS_INO_MASTER
);
436 return GC_LEVEL(LOGFS_MAX_LEVELS
);
439 static gc_level_t
indirect_block_level(struct logfs_block
*block
)
447 inode
= page
->mapping
->host
;
448 logfs_unpack_index(page
->index
, &bix
, &level
);
449 return expand_level(inode
->i_ino
, level
);
453 * This silences a false, yet annoying gcc warning. I hate it when my editor
454 * jumps into bitops.h each time I recompile this file.
455 * TODO: Complain to gcc folks about this and upgrade compiler.
457 static unsigned long fnb(const unsigned long *addr
,
458 unsigned long size
, unsigned long offset
)
460 return find_next_bit(addr
, size
, offset
);
463 static __be64
inode_val0(struct inode
*inode
)
465 struct logfs_inode
*li
= logfs_inode(inode
);
469 * Explicit shifting generates good code, but must match the format
470 * of the structure. Add some paranoia just in case.
472 BUILD_BUG_ON(offsetof(struct logfs_disk_inode
, di_mode
) != 0);
473 BUILD_BUG_ON(offsetof(struct logfs_disk_inode
, di_height
) != 2);
474 BUILD_BUG_ON(offsetof(struct logfs_disk_inode
, di_flags
) != 4);
476 val
= (u64
)inode
->i_mode
<< 48 |
477 (u64
)li
->li_height
<< 40 |
479 return cpu_to_be64(val
);
482 static int inode_write_alias(struct super_block
*sb
,
483 struct logfs_block
*block
, write_alias_t
*write_one_alias
)
485 struct inode
*inode
= block
->inode
;
486 struct logfs_inode
*li
= logfs_inode(inode
);
493 for (pos
= 0; ; pos
++) {
494 pos
= fnb(block
->alias_map
, LOGFS_BLOCK_FACTOR
, pos
);
495 if (pos
>= LOGFS_EMBEDDED_FIELDS
+ INODE_POINTER_OFS
)
499 case INODE_HEIGHT_OFS
:
500 val
= inode_val0(inode
);
503 val
= cpu_to_be64(li
->li_used_bytes
);;
506 val
= cpu_to_be64(i_size_read(inode
));
508 case INODE_POINTER_OFS
... INODE_POINTER_OFS
+ LOGFS_EMBEDDED_FIELDS
- 1:
509 val
= cpu_to_be64(li
->li_data
[pos
- INODE_POINTER_OFS
]);
515 ino
= LOGFS_INO_MASTER
;
518 err
= write_one_alias(sb
, ino
, bix
, level
, pos
, val
);
524 static int indirect_write_alias(struct super_block
*sb
,
525 struct logfs_block
*block
, write_alias_t
*write_one_alias
)
528 struct page
*page
= block
->page
;
534 for (pos
= 0; ; pos
++) {
535 pos
= fnb(block
->alias_map
, LOGFS_BLOCK_FACTOR
, pos
);
536 if (pos
>= LOGFS_BLOCK_FACTOR
)
539 ino
= page
->mapping
->host
->i_ino
;
540 logfs_unpack_index(page
->index
, &bix
, &level
);
541 child
= kmap_atomic(page
, KM_USER0
);
543 kunmap_atomic(child
, KM_USER0
);
544 err
= write_one_alias(sb
, ino
, bix
, level
, pos
, val
);
550 int logfs_write_obj_aliases_pagecache(struct super_block
*sb
)
552 struct logfs_super
*super
= logfs_super(sb
);
553 struct logfs_block
*block
;
556 list_for_each_entry(block
, &super
->s_object_alias
, alias_list
) {
557 err
= block
->ops
->write_alias(sb
, block
, write_alias_journal
);
564 void __free_block(struct super_block
*sb
, struct logfs_block
*block
)
566 BUG_ON(!list_empty(&block
->item_list
));
567 list_del(&block
->alias_list
);
568 mempool_free(block
, logfs_super(sb
)->s_block_pool
);
571 static void inode_free_block(struct super_block
*sb
, struct logfs_block
*block
)
573 struct inode
*inode
= block
->inode
;
575 logfs_inode(inode
)->li_block
= NULL
;
576 __free_block(sb
, block
);
579 static void indirect_free_block(struct super_block
*sb
,
580 struct logfs_block
*block
)
582 ClearPagePrivate(block
->page
);
583 block
->page
->private = 0;
584 __free_block(sb
, block
);
588 static struct logfs_block_ops inode_block_ops
= {
589 .write_block
= inode_write_block
,
590 .block_level
= inode_block_level
,
591 .free_block
= inode_free_block
,
592 .write_alias
= inode_write_alias
,
595 struct logfs_block_ops indirect_block_ops
= {
596 .write_block
= indirect_write_block
,
597 .block_level
= indirect_block_level
,
598 .free_block
= indirect_free_block
,
599 .write_alias
= indirect_write_alias
,
602 struct logfs_block
*__alloc_block(struct super_block
*sb
,
603 u64 ino
, u64 bix
, level_t level
)
605 struct logfs_super
*super
= logfs_super(sb
);
606 struct logfs_block
*block
;
608 block
= mempool_alloc(super
->s_block_pool
, GFP_NOFS
);
609 memset(block
, 0, sizeof(*block
));
610 INIT_LIST_HEAD(&block
->alias_list
);
611 INIT_LIST_HEAD(&block
->item_list
);
615 block
->level
= level
;
619 static void alloc_inode_block(struct inode
*inode
)
621 struct logfs_inode
*li
= logfs_inode(inode
);
622 struct logfs_block
*block
;
627 block
= __alloc_block(inode
->i_sb
, LOGFS_INO_MASTER
, inode
->i_ino
, 0);
628 block
->inode
= inode
;
629 li
->li_block
= block
;
630 block
->ops
= &inode_block_ops
;
633 void initialize_block_counters(struct page
*page
, struct logfs_block
*block
,
634 __be64
*array
, int page_is_empty
)
642 if (page
->index
< first_indirect_block()) {
643 /* Counters are pointless on level 0 */
646 if (page
->index
== first_indirect_block()) {
647 /* Skip unused pointers */
649 block
->full
= I0_BLOCKS
;
651 if (!page_is_empty
) {
652 for (i
= start
; i
< LOGFS_BLOCK_FACTOR
; i
++) {
653 ptr
= be64_to_cpu(array
[i
]);
656 if (ptr
& LOGFS_FULLY_POPULATED
)
662 static void alloc_data_block(struct inode
*inode
, struct page
*page
)
664 struct logfs_block
*block
;
668 if (PagePrivate(page
))
671 logfs_unpack_index(page
->index
, &bix
, &level
);
672 block
= __alloc_block(inode
->i_sb
, inode
->i_ino
, bix
, level
);
674 SetPagePrivate(page
);
675 page
->private = (unsigned long)block
;
676 block
->ops
= &indirect_block_ops
;
679 static void alloc_indirect_block(struct inode
*inode
, struct page
*page
,
682 struct logfs_block
*block
;
685 if (PagePrivate(page
))
688 alloc_data_block(inode
, page
);
690 block
= logfs_block(page
);
691 array
= kmap_atomic(page
, KM_USER0
);
692 initialize_block_counters(page
, block
, array
, page_is_empty
);
693 kunmap_atomic(array
, KM_USER0
);
696 static void block_set_pointer(struct page
*page
, int index
, u64 ptr
)
698 struct logfs_block
*block
= logfs_block(page
);
703 array
= kmap_atomic(page
, KM_USER0
);
704 oldptr
= be64_to_cpu(array
[index
]);
705 array
[index
] = cpu_to_be64(ptr
);
706 kunmap_atomic(array
, KM_USER0
);
707 SetPageUptodate(page
);
709 block
->full
+= !!(ptr
& LOGFS_FULLY_POPULATED
)
710 - !!(oldptr
& LOGFS_FULLY_POPULATED
);
711 block
->partial
+= !!ptr
- !!oldptr
;
714 static u64
block_get_pointer(struct page
*page
, int index
)
719 block
= kmap_atomic(page
, KM_USER0
);
720 ptr
= be64_to_cpu(block
[index
]);
721 kunmap_atomic(block
, KM_USER0
);
725 static int logfs_read_empty(struct page
*page
)
727 zero_user_segment(page
, 0, PAGE_CACHE_SIZE
);
731 static int logfs_read_direct(struct inode
*inode
, struct page
*page
)
733 struct logfs_inode
*li
= logfs_inode(inode
);
734 pgoff_t index
= page
->index
;
737 block
= li
->li_data
[index
];
739 return logfs_read_empty(page
);
741 return logfs_segment_read(inode
, page
, block
, index
, 0);
744 static int logfs_read_loop(struct inode
*inode
, struct page
*page
,
747 struct logfs_inode
*li
= logfs_inode(inode
);
748 u64 bix
, bofs
= li
->li_data
[INDIRECT_INDEX
];
749 level_t level
, target_level
;
753 logfs_unpack_index(page
->index
, &bix
, &target_level
);
755 return logfs_read_empty(page
);
757 if (bix
>= maxbix(li
->li_height
))
758 return logfs_read_empty(page
);
760 for (level
= LEVEL(li
->li_height
);
761 (__force u8
)level
> (__force u8
)target_level
;
762 level
= SUBLEVEL(level
)){
763 ipage
= logfs_get_page(inode
, bix
, level
, rw_context
);
767 ret
= logfs_segment_read(inode
, ipage
, bofs
, bix
, level
);
769 logfs_put_read_page(ipage
);
773 bofs
= block_get_pointer(ipage
, get_bits(bix
, SUBLEVEL(level
)));
774 logfs_put_page(ipage
, rw_context
);
776 return logfs_read_empty(page
);
779 return logfs_segment_read(inode
, page
, bofs
, bix
, 0);
782 static int logfs_read_block(struct inode
*inode
, struct page
*page
,
785 pgoff_t index
= page
->index
;
787 if (index
< I0_BLOCKS
)
788 return logfs_read_direct(inode
, page
);
789 return logfs_read_loop(inode
, page
, rw_context
);
792 static int logfs_exist_loop(struct inode
*inode
, u64 bix
)
794 struct logfs_inode
*li
= logfs_inode(inode
);
795 u64 bofs
= li
->li_data
[INDIRECT_INDEX
];
802 if (bix
>= maxbix(li
->li_height
))
805 for (level
= LEVEL(li
->li_height
); level
!= 0; level
= SUBLEVEL(level
)) {
806 ipage
= logfs_get_read_page(inode
, bix
, level
);
810 ret
= logfs_segment_read(inode
, ipage
, bofs
, bix
, level
);
812 logfs_put_read_page(ipage
);
816 bofs
= block_get_pointer(ipage
, get_bits(bix
, SUBLEVEL(level
)));
817 logfs_put_read_page(ipage
);
825 int logfs_exist_block(struct inode
*inode
, u64 bix
)
827 struct logfs_inode
*li
= logfs_inode(inode
);
830 return !!li
->li_data
[bix
];
831 return logfs_exist_loop(inode
, bix
);
834 static u64
seek_holedata_direct(struct inode
*inode
, u64 bix
, int data
)
836 struct logfs_inode
*li
= logfs_inode(inode
);
838 for (; bix
< I0_BLOCKS
; bix
++)
839 if (data
^ (li
->li_data
[bix
] == 0))
844 static u64
seek_holedata_loop(struct inode
*inode
, u64 bix
, int data
)
846 struct logfs_inode
*li
= logfs_inode(inode
);
848 u64 increment
, bofs
= li
->li_data
[INDIRECT_INDEX
];
855 for (level
= LEVEL(li
->li_height
); level
!= 0; level
= SUBLEVEL(level
)) {
856 increment
= 1 << (LOGFS_BLOCK_BITS
* ((__force u8
)level
-1));
857 page
= logfs_get_read_page(inode
, bix
, level
);
861 ret
= logfs_segment_read(inode
, page
, bofs
, bix
, level
);
863 logfs_put_read_page(page
);
867 slot
= get_bits(bix
, SUBLEVEL(level
));
868 rblock
= kmap_atomic(page
, KM_USER0
);
869 while (slot
< LOGFS_BLOCK_FACTOR
) {
870 if (data
&& (rblock
[slot
] != 0))
872 if (!data
&& !(be64_to_cpu(rblock
[slot
]) & LOGFS_FULLY_POPULATED
))
876 bix
&= ~(increment
- 1);
878 if (slot
>= LOGFS_BLOCK_FACTOR
) {
879 kunmap_atomic(rblock
, KM_USER0
);
880 logfs_put_read_page(page
);
883 bofs
= be64_to_cpu(rblock
[slot
]);
884 kunmap_atomic(rblock
, KM_USER0
);
885 logfs_put_read_page(page
);
895 * logfs_seek_hole - find next hole starting at a given block index
896 * @inode: inode to search in
897 * @bix: block index to start searching
899 * Returns next hole. If the file doesn't contain any further holes, the
900 * block address next to eof is returned instead.
902 u64
logfs_seek_hole(struct inode
*inode
, u64 bix
)
904 struct logfs_inode
*li
= logfs_inode(inode
);
906 if (bix
< I0_BLOCKS
) {
907 bix
= seek_holedata_direct(inode
, bix
, 0);
912 if (!li
->li_data
[INDIRECT_INDEX
])
914 else if (li
->li_data
[INDIRECT_INDEX
] & LOGFS_FULLY_POPULATED
)
915 bix
= maxbix(li
->li_height
);
917 bix
= seek_holedata_loop(inode
, bix
, 0);
918 if (bix
< maxbix(li
->li_height
))
920 /* Should not happen anymore. But if some port writes semi-
921 * corrupt images (as this one used to) we might run into it.
923 WARN_ON_ONCE(bix
== maxbix(li
->li_height
));
929 static u64
__logfs_seek_data(struct inode
*inode
, u64 bix
)
931 struct logfs_inode
*li
= logfs_inode(inode
);
933 if (bix
< I0_BLOCKS
) {
934 bix
= seek_holedata_direct(inode
, bix
, 1);
939 if (bix
< maxbix(li
->li_height
)) {
940 if (!li
->li_data
[INDIRECT_INDEX
])
941 bix
= maxbix(li
->li_height
);
943 return seek_holedata_loop(inode
, bix
, 1);
950 * logfs_seek_data - find next data block after a given block index
951 * @inode: inode to search in
952 * @bix: block index to start searching
954 * Returns next data block. If the file doesn't contain any further data
955 * blocks, the last block in the file is returned instead.
957 u64
logfs_seek_data(struct inode
*inode
, u64 bix
)
959 struct super_block
*sb
= inode
->i_sb
;
962 ret
= __logfs_seek_data(inode
, bix
);
963 end
= i_size_read(inode
) >> sb
->s_blocksize_bits
;
969 static int logfs_is_valid_direct(struct logfs_inode
*li
, u64 bix
, u64 ofs
)
971 return pure_ofs(li
->li_data
[bix
]) == ofs
;
974 static int __logfs_is_valid_loop(struct inode
*inode
, u64 bix
,
977 struct logfs_inode
*li
= logfs_inode(inode
);
982 for (level
= LEVEL(li
->li_height
); level
!= 0; level
= SUBLEVEL(level
)){
983 page
= logfs_get_write_page(inode
, bix
, level
);
986 ret
= logfs_segment_read(inode
, page
, bofs
, bix
, level
);
988 logfs_put_write_page(page
);
992 bofs
= block_get_pointer(page
, get_bits(bix
, SUBLEVEL(level
)));
993 logfs_put_write_page(page
);
997 if (pure_ofs(bofs
) == ofs
)
1003 static int logfs_is_valid_loop(struct inode
*inode
, u64 bix
, u64 ofs
)
1005 struct logfs_inode
*li
= logfs_inode(inode
);
1006 u64 bofs
= li
->li_data
[INDIRECT_INDEX
];
1011 if (bix
>= maxbix(li
->li_height
))
1014 if (pure_ofs(bofs
) == ofs
)
1017 return __logfs_is_valid_loop(inode
, bix
, ofs
, bofs
);
1020 static int __logfs_is_valid_block(struct inode
*inode
, u64 bix
, u64 ofs
)
1022 struct logfs_inode
*li
= logfs_inode(inode
);
1024 if ((inode
->i_nlink
== 0) && atomic_read(&inode
->i_count
) == 1)
1027 if (bix
< I0_BLOCKS
)
1028 return logfs_is_valid_direct(li
, bix
, ofs
);
1029 return logfs_is_valid_loop(inode
, bix
, ofs
);
1033 * logfs_is_valid_block - check whether this block is still valid
1036 * @ofs - block physical offset
1037 * @ino - block inode number
1038 * @bix - block index
1039 * @level - block level
1041 * Returns 0 if the block is invalid, 1 if it is valid and 2 if it will
1042 * become invalid once the journal is written.
1044 int logfs_is_valid_block(struct super_block
*sb
, u64 ofs
, u64 ino
, u64 bix
,
1045 gc_level_t gc_level
)
1047 struct logfs_super
*super
= logfs_super(sb
);
1048 struct inode
*inode
;
1051 /* Umount closes a segment with free blocks remaining. Those
1052 * blocks are by definition invalid. */
1056 LOGFS_BUG_ON((u64
)(u_long
)ino
!= ino
, sb
);
1058 inode
= logfs_safe_iget(sb
, ino
, &cookie
);
1062 ret
= __logfs_is_valid_block(inode
, bix
, ofs
);
1063 logfs_safe_iput(inode
, cookie
);
1068 /* Block is nominally invalid, but may still sit in the shadow tree,
1069 * waiting for a journal commit.
1071 if (btree_lookup64(&super
->s_shadow_tree
.old
, ofs
))
1076 int logfs_readpage_nolock(struct page
*page
)
1078 struct inode
*inode
= page
->mapping
->host
;
1081 ret
= logfs_read_block(inode
, page
, READ
);
1084 ClearPageUptodate(page
);
1087 SetPageUptodate(page
);
1088 ClearPageError(page
);
1090 flush_dcache_page(page
);
1095 static int logfs_reserve_bytes(struct inode
*inode
, int bytes
)
1097 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1098 u64 available
= super
->s_free_bytes
+ super
->s_dirty_free_bytes
1099 - super
->s_dirty_used_bytes
- super
->s_dirty_pages
;
1104 if (available
< bytes
)
1107 if (available
< bytes
+ super
->s_root_reserve
&&
1108 !capable(CAP_SYS_RESOURCE
))
1114 int get_page_reserve(struct inode
*inode
, struct page
*page
)
1116 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1119 if (logfs_block(page
) && logfs_block(page
)->reserved_bytes
)
1122 logfs_get_wblocks(inode
->i_sb
, page
, WF_LOCK
);
1123 ret
= logfs_reserve_bytes(inode
, 6 * LOGFS_MAX_OBJECTSIZE
);
1125 alloc_data_block(inode
, page
);
1126 logfs_block(page
)->reserved_bytes
+= 6 * LOGFS_MAX_OBJECTSIZE
;
1127 super
->s_dirty_pages
+= 6 * LOGFS_MAX_OBJECTSIZE
;
1129 logfs_put_wblocks(inode
->i_sb
, page
, WF_LOCK
);
1134 * We are protected by write lock. Push victims up to superblock level
1135 * and release transaction when appropriate.
1137 /* FIXME: This is currently called from the wrong spots. */
1138 static void logfs_handle_transaction(struct inode
*inode
,
1139 struct logfs_transaction
*ta
)
1141 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1145 logfs_inode(inode
)->li_block
->ta
= NULL
;
1147 if (inode
->i_ino
!= LOGFS_INO_MASTER
) {
1148 BUG(); /* FIXME: Yes, this needs more thought */
1149 /* just remember the transaction until inode is written */
1150 //BUG_ON(logfs_inode(inode)->li_transaction);
1151 //logfs_inode(inode)->li_transaction = ta;
1155 switch (ta
->state
) {
1156 case CREATE_1
: /* fall through */
1158 BUG_ON(super
->s_victim_ino
);
1159 super
->s_victim_ino
= ta
->ino
;
1161 case CREATE_2
: /* fall through */
1163 BUG_ON(super
->s_victim_ino
!= ta
->ino
);
1164 super
->s_victim_ino
= 0;
1165 /* transaction ends here - free it */
1168 case CROSS_RENAME_1
:
1169 BUG_ON(super
->s_rename_dir
);
1170 BUG_ON(super
->s_rename_pos
);
1171 super
->s_rename_dir
= ta
->dir
;
1172 super
->s_rename_pos
= ta
->pos
;
1174 case CROSS_RENAME_2
:
1175 BUG_ON(super
->s_rename_dir
!= ta
->dir
);
1176 BUG_ON(super
->s_rename_pos
!= ta
->pos
);
1177 super
->s_rename_dir
= 0;
1178 super
->s_rename_pos
= 0;
1181 case TARGET_RENAME_1
:
1182 BUG_ON(super
->s_rename_dir
);
1183 BUG_ON(super
->s_rename_pos
);
1184 BUG_ON(super
->s_victim_ino
);
1185 super
->s_rename_dir
= ta
->dir
;
1186 super
->s_rename_pos
= ta
->pos
;
1187 super
->s_victim_ino
= ta
->ino
;
1189 case TARGET_RENAME_2
:
1190 BUG_ON(super
->s_rename_dir
!= ta
->dir
);
1191 BUG_ON(super
->s_rename_pos
!= ta
->pos
);
1192 BUG_ON(super
->s_victim_ino
!= ta
->ino
);
1193 super
->s_rename_dir
= 0;
1194 super
->s_rename_pos
= 0;
1196 case TARGET_RENAME_3
:
1197 BUG_ON(super
->s_rename_dir
);
1198 BUG_ON(super
->s_rename_pos
);
1199 BUG_ON(super
->s_victim_ino
!= ta
->ino
);
1200 super
->s_victim_ino
= 0;
1209 * Not strictly a reservation, but rather a check that we still have enough
1210 * space to satisfy the write.
1212 static int logfs_reserve_blocks(struct inode
*inode
, int blocks
)
1214 return logfs_reserve_bytes(inode
, blocks
* LOGFS_MAX_OBJECTSIZE
);
1217 struct write_control
{
1222 static struct logfs_shadow
*alloc_shadow(struct inode
*inode
, u64 bix
,
1223 level_t level
, u64 old_ofs
)
1225 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1226 struct logfs_shadow
*shadow
;
1228 shadow
= mempool_alloc(super
->s_shadow_pool
, GFP_NOFS
);
1229 memset(shadow
, 0, sizeof(*shadow
));
1230 shadow
->ino
= inode
->i_ino
;
1232 shadow
->gc_level
= expand_level(inode
->i_ino
, level
);
1233 shadow
->old_ofs
= old_ofs
& ~LOGFS_FULLY_POPULATED
;
1237 static void free_shadow(struct inode
*inode
, struct logfs_shadow
*shadow
)
1239 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1241 mempool_free(shadow
, super
->s_shadow_pool
);
1245 * fill_shadow_tree - Propagate shadow tree changes due to a write
1246 * @inode: Inode owning the page
1247 * @page: Struct page that was written
1248 * @shadow: Shadow for the current write
1250 * Writes in logfs can result in two semi-valid objects. The old object
1251 * is still valid as long as it can be reached by following pointers on
1252 * the medium. Only when writes propagate all the way up to the journal
1253 * has the new object safely replaced the old one.
1255 * To handle this problem, a struct logfs_shadow is used to represent
1256 * every single write. It is attached to the indirect block, which is
1257 * marked dirty. When the indirect block is written, its shadows are
1258 * handed up to the next indirect block (or inode). Untimately they
1259 * will reach the master inode and be freed upon journal commit.
1261 * This function handles a single step in the propagation. It adds the
1262 * shadow for the current write to the tree, along with any shadows in
1263 * the page's tree, in case it was an indirect block. If a page is
1264 * written, the inode parameter is left NULL, if an inode is written,
1265 * the page parameter is left NULL.
1267 static void fill_shadow_tree(struct inode
*inode
, struct page
*page
,
1268 struct logfs_shadow
*shadow
)
1270 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1271 struct logfs_block
*block
= logfs_block(page
);
1272 struct shadow_tree
*tree
= &super
->s_shadow_tree
;
1274 if (PagePrivate(page
)) {
1275 if (block
->alias_map
)
1276 super
->s_no_object_aliases
-= bitmap_weight(
1277 block
->alias_map
, LOGFS_BLOCK_FACTOR
);
1278 logfs_handle_transaction(inode
, block
->ta
);
1279 block
->ops
->free_block(inode
->i_sb
, block
);
1282 if (shadow
->old_ofs
)
1283 btree_insert64(&tree
->old
, shadow
->old_ofs
, shadow
,
1286 btree_insert64(&tree
->new, shadow
->new_ofs
, shadow
,
1289 super
->s_dirty_used_bytes
+= shadow
->new_len
;
1290 super
->s_dirty_free_bytes
+= shadow
->old_len
;
1294 static void logfs_set_alias(struct super_block
*sb
, struct logfs_block
*block
,
1297 struct logfs_super
*super
= logfs_super(sb
);
1299 if (block
->inode
&& block
->inode
->i_ino
== LOGFS_INO_MASTER
) {
1300 /* Aliases in the master inode are pointless. */
1304 if (!test_bit(child_no
, block
->alias_map
)) {
1305 set_bit(child_no
, block
->alias_map
);
1306 super
->s_no_object_aliases
++;
1308 list_move_tail(&block
->alias_list
, &super
->s_object_alias
);
1312 * Object aliases can and often do change the size and occupied space of a
1313 * file. So not only do we have to change the pointers, we also have to
1314 * change inode->i_size and li->li_used_bytes. Which is done by setting
1315 * another two object aliases for the inode itself.
1317 static void set_iused(struct inode
*inode
, struct logfs_shadow
*shadow
)
1319 struct logfs_inode
*li
= logfs_inode(inode
);
1321 if (shadow
->new_len
== shadow
->old_len
)
1324 alloc_inode_block(inode
);
1325 li
->li_used_bytes
+= shadow
->new_len
- shadow
->old_len
;
1326 __logfs_set_blocks(inode
);
1327 logfs_set_alias(inode
->i_sb
, li
->li_block
, INODE_USED_OFS
);
1328 logfs_set_alias(inode
->i_sb
, li
->li_block
, INODE_SIZE_OFS
);
1331 static int logfs_write_i0(struct inode
*inode
, struct page
*page
,
1332 struct write_control
*wc
)
1334 struct logfs_shadow
*shadow
;
1339 logfs_unpack_index(page
->index
, &bix
, &level
);
1341 if (logfs_reserve_blocks(inode
, 1))
1344 shadow
= alloc_shadow(inode
, bix
, level
, wc
->ofs
);
1345 if (wc
->flags
& WF_WRITE
)
1346 err
= logfs_segment_write(inode
, page
, shadow
);
1347 if (wc
->flags
& WF_DELETE
)
1348 logfs_segment_delete(inode
, shadow
);
1350 free_shadow(inode
, shadow
);
1354 set_iused(inode
, shadow
);
1357 alloc_indirect_block(inode
, page
, 0);
1358 full
= logfs_block(page
)->full
== LOGFS_BLOCK_FACTOR
;
1360 fill_shadow_tree(inode
, page
, shadow
);
1361 wc
->ofs
= shadow
->new_ofs
;
1362 if (wc
->ofs
&& full
)
1363 wc
->ofs
|= LOGFS_FULLY_POPULATED
;
1367 static int logfs_write_direct(struct inode
*inode
, struct page
*page
,
1370 struct logfs_inode
*li
= logfs_inode(inode
);
1371 struct write_control wc
= {
1372 .ofs
= li
->li_data
[page
->index
],
1377 alloc_inode_block(inode
);
1379 err
= logfs_write_i0(inode
, page
, &wc
);
1383 li
->li_data
[page
->index
] = wc
.ofs
;
1384 logfs_set_alias(inode
->i_sb
, li
->li_block
,
1385 page
->index
+ INODE_POINTER_OFS
);
1389 static int ptr_change(u64 ofs
, struct page
*page
)
1391 struct logfs_block
*block
= logfs_block(page
);
1392 int empty0
, empty1
, full0
, full1
;
1395 empty1
= block
->partial
== 0;
1396 if (empty0
!= empty1
)
1399 /* The !! is necessary to shrink result to int */
1400 full0
= !!(ofs
& LOGFS_FULLY_POPULATED
);
1401 full1
= block
->full
== LOGFS_BLOCK_FACTOR
;
1407 static int __logfs_write_rec(struct inode
*inode
, struct page
*page
,
1408 struct write_control
*this_wc
,
1409 pgoff_t bix
, level_t target_level
, level_t level
)
1411 int ret
, page_empty
= 0;
1412 int child_no
= get_bits(bix
, SUBLEVEL(level
));
1414 struct write_control child_wc
= {
1415 .flags
= this_wc
->flags
,
1418 ipage
= logfs_get_write_page(inode
, bix
, level
);
1423 ret
= logfs_segment_read(inode
, ipage
, this_wc
->ofs
, bix
, level
);
1426 } else if (!PageUptodate(ipage
)) {
1428 logfs_read_empty(ipage
);
1431 child_wc
.ofs
= block_get_pointer(ipage
, child_no
);
1433 if ((__force u8
)level
-1 > (__force u8
)target_level
)
1434 ret
= __logfs_write_rec(inode
, page
, &child_wc
, bix
,
1435 target_level
, SUBLEVEL(level
));
1437 ret
= logfs_write_i0(inode
, page
, &child_wc
);
1442 alloc_indirect_block(inode
, ipage
, page_empty
);
1443 block_set_pointer(ipage
, child_no
, child_wc
.ofs
);
1444 /* FIXME: first condition seems superfluous */
1445 if (child_wc
.ofs
|| logfs_block(ipage
)->partial
)
1446 this_wc
->flags
|= WF_WRITE
;
1447 /* the condition on this_wc->ofs ensures that we won't consume extra
1448 * space for indirect blocks in the future, which we cannot reserve */
1449 if (!this_wc
->ofs
|| ptr_change(this_wc
->ofs
, ipage
))
1450 ret
= logfs_write_i0(inode
, ipage
, this_wc
);
1452 logfs_set_alias(inode
->i_sb
, logfs_block(ipage
), child_no
);
1454 logfs_put_write_page(ipage
);
1458 static int logfs_write_rec(struct inode
*inode
, struct page
*page
,
1459 pgoff_t bix
, level_t target_level
, long flags
)
1461 struct logfs_inode
*li
= logfs_inode(inode
);
1462 struct write_control wc
= {
1463 .ofs
= li
->li_data
[INDIRECT_INDEX
],
1468 alloc_inode_block(inode
);
1470 if (li
->li_height
> (__force u8
)target_level
)
1471 ret
= __logfs_write_rec(inode
, page
, &wc
, bix
, target_level
,
1472 LEVEL(li
->li_height
));
1474 ret
= logfs_write_i0(inode
, page
, &wc
);
1478 if (li
->li_data
[INDIRECT_INDEX
] != wc
.ofs
) {
1479 li
->li_data
[INDIRECT_INDEX
] = wc
.ofs
;
1480 logfs_set_alias(inode
->i_sb
, li
->li_block
,
1481 INDIRECT_INDEX
+ INODE_POINTER_OFS
);
1486 void logfs_add_transaction(struct inode
*inode
, struct logfs_transaction
*ta
)
1488 alloc_inode_block(inode
);
1489 logfs_inode(inode
)->li_block
->ta
= ta
;
1492 void logfs_del_transaction(struct inode
*inode
, struct logfs_transaction
*ta
)
1494 struct logfs_block
*block
= logfs_inode(inode
)->li_block
;
1496 if (block
&& block
->ta
)
1500 static int grow_inode(struct inode
*inode
, u64 bix
, level_t level
)
1502 struct logfs_inode
*li
= logfs_inode(inode
);
1503 u8 height
= (__force u8
)level
;
1505 struct write_control wc
= {
1510 BUG_ON(height
> 5 || li
->li_height
> 5);
1511 while (height
> li
->li_height
|| bix
>= maxbix(li
->li_height
)) {
1512 page
= logfs_get_write_page(inode
, I0_BLOCKS
+ 1,
1513 LEVEL(li
->li_height
+ 1));
1516 logfs_read_empty(page
);
1517 alloc_indirect_block(inode
, page
, 1);
1518 block_set_pointer(page
, 0, li
->li_data
[INDIRECT_INDEX
]);
1519 err
= logfs_write_i0(inode
, page
, &wc
);
1520 logfs_put_write_page(page
);
1523 li
->li_data
[INDIRECT_INDEX
] = wc
.ofs
;
1526 logfs_set_alias(inode
->i_sb
, li
->li_block
, INODE_HEIGHT_OFS
);
1531 static int __logfs_write_buf(struct inode
*inode
, struct page
*page
, long flags
)
1533 struct logfs_super
*super
= logfs_super(inode
->i_sb
);
1534 pgoff_t index
= page
->index
;
1539 flags
|= WF_WRITE
| WF_DELETE
;
1540 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1542 logfs_unpack_index(index
, &bix
, &level
);
1543 if (logfs_block(page
) && logfs_block(page
)->reserved_bytes
)
1544 super
->s_dirty_pages
-= logfs_block(page
)->reserved_bytes
;
1546 if (index
< I0_BLOCKS
)
1547 return logfs_write_direct(inode
, page
, flags
);
1549 bix
= adjust_bix(bix
, level
);
1550 err
= grow_inode(inode
, bix
, level
);
1553 return logfs_write_rec(inode
, page
, bix
, level
, flags
);
1556 int logfs_write_buf(struct inode
*inode
, struct page
*page
, long flags
)
1558 struct super_block
*sb
= inode
->i_sb
;
1561 logfs_get_wblocks(sb
, page
, flags
& WF_LOCK
);
1562 ret
= __logfs_write_buf(inode
, page
, flags
);
1563 logfs_put_wblocks(sb
, page
, flags
& WF_LOCK
);
1567 static int __logfs_delete(struct inode
*inode
, struct page
*page
)
1569 long flags
= WF_DELETE
;
1571 inode
->i_ctime
= inode
->i_mtime
= CURRENT_TIME
;
1573 if (page
->index
< I0_BLOCKS
)
1574 return logfs_write_direct(inode
, page
, flags
);
1575 return logfs_write_rec(inode
, page
, page
->index
, 0, flags
);
1578 int logfs_delete(struct inode
*inode
, pgoff_t index
,
1579 struct shadow_tree
*shadow_tree
)
1581 struct super_block
*sb
= inode
->i_sb
;
1585 page
= logfs_get_read_page(inode
, index
, 0);
1589 logfs_get_wblocks(sb
, page
, 1);
1590 ret
= __logfs_delete(inode
, page
);
1591 logfs_put_wblocks(sb
, page
, 1);
1593 logfs_put_read_page(page
);
1598 /* Rewrite cannot mark the inode dirty but has to write it immediatly. */
1599 int logfs_rewrite_block(struct inode
*inode
, u64 bix
, u64 ofs
,
1600 gc_level_t gc_level
, long flags
)
1602 level_t level
= shrink_level(gc_level
);
1606 page
= logfs_get_write_page(inode
, bix
, level
);
1610 err
= logfs_segment_read(inode
, page
, ofs
, bix
, level
);
1613 alloc_indirect_block(inode
, page
, 0);
1614 err
= logfs_write_buf(inode
, page
, flags
);
1616 logfs_put_write_page(page
);
1620 static int truncate_data_block(struct inode
*inode
, struct page
*page
,
1621 u64 ofs
, struct logfs_shadow
*shadow
, u64 size
)
1623 loff_t pageofs
= page
->index
<< inode
->i_sb
->s_blocksize_bits
;
1628 /* Does truncation happen within this page? */
1629 if (size
<= pageofs
|| size
- pageofs
>= PAGE_SIZE
)
1632 logfs_unpack_index(page
->index
, &bix
, &level
);
1635 err
= logfs_segment_read(inode
, page
, ofs
, bix
, level
);
1639 zero_user_segment(page
, size
- pageofs
, PAGE_CACHE_SIZE
);
1640 return logfs_segment_write(inode
, page
, shadow
);
1643 static int logfs_truncate_i0(struct inode
*inode
, struct page
*page
,
1644 struct write_control
*wc
, u64 size
)
1646 struct logfs_shadow
*shadow
;
1651 logfs_unpack_index(page
->index
, &bix
, &level
);
1653 shadow
= alloc_shadow(inode
, bix
, level
, wc
->ofs
);
1655 err
= truncate_data_block(inode
, page
, wc
->ofs
, shadow
, size
);
1657 free_shadow(inode
, shadow
);
1661 logfs_segment_delete(inode
, shadow
);
1662 set_iused(inode
, shadow
);
1663 fill_shadow_tree(inode
, page
, shadow
);
1664 wc
->ofs
= shadow
->new_ofs
;
1668 static int logfs_truncate_direct(struct inode
*inode
, u64 size
)
1670 struct logfs_inode
*li
= logfs_inode(inode
);
1671 struct write_control wc
;
1676 alloc_inode_block(inode
);
1678 for (e
= I0_BLOCKS
- 1; e
>= 0; e
--) {
1679 if (size
> (e
+1) * LOGFS_BLOCKSIZE
)
1682 wc
.ofs
= li
->li_data
[e
];
1686 page
= logfs_get_write_page(inode
, e
, 0);
1689 err
= logfs_segment_read(inode
, page
, wc
.ofs
, e
, 0);
1691 logfs_put_write_page(page
);
1694 err
= logfs_truncate_i0(inode
, page
, &wc
, size
);
1695 logfs_put_write_page(page
);
1699 li
->li_data
[e
] = wc
.ofs
;
1704 /* FIXME: these need to become per-sb once we support different blocksizes */
1705 static u64 __logfs_step
[] = {
1712 static u64 __logfs_start_index
[] = {
1719 static inline u64
logfs_step(level_t level
)
1721 return __logfs_step
[(__force u8
)level
];
1724 static inline u64
logfs_factor(u8 level
)
1726 return __logfs_step
[level
] * LOGFS_BLOCKSIZE
;
1729 static inline u64
logfs_start_index(level_t level
)
1731 return __logfs_start_index
[(__force u8
)level
];
1734 static void logfs_unpack_raw_index(pgoff_t index
, u64
*bix
, level_t
*level
)
1736 logfs_unpack_index(index
, bix
, level
);
1737 if (*bix
<= logfs_start_index(SUBLEVEL(*level
)))
1741 static int __logfs_truncate_rec(struct inode
*inode
, struct page
*ipage
,
1742 struct write_control
*this_wc
, u64 size
)
1744 int truncate_happened
= 0;
1746 u64 bix
, child_bix
, next_bix
;
1749 struct write_control child_wc
= { /* FIXME: flags */ };
1751 logfs_unpack_raw_index(ipage
->index
, &bix
, &level
);
1752 err
= logfs_segment_read(inode
, ipage
, this_wc
->ofs
, bix
, level
);
1756 for (e
= LOGFS_BLOCK_FACTOR
- 1; e
>= 0; e
--) {
1757 child_bix
= bix
+ e
* logfs_step(SUBLEVEL(level
));
1758 next_bix
= child_bix
+ logfs_step(SUBLEVEL(level
));
1759 if (size
> next_bix
* LOGFS_BLOCKSIZE
)
1762 child_wc
.ofs
= pure_ofs(block_get_pointer(ipage
, e
));
1766 page
= logfs_get_write_page(inode
, child_bix
, SUBLEVEL(level
));
1770 if ((__force u8
)level
> 1)
1771 err
= __logfs_truncate_rec(inode
, page
, &child_wc
, size
);
1773 err
= logfs_truncate_i0(inode
, page
, &child_wc
, size
);
1774 logfs_put_write_page(page
);
1778 truncate_happened
= 1;
1779 alloc_indirect_block(inode
, ipage
, 0);
1780 block_set_pointer(ipage
, e
, child_wc
.ofs
);
1783 if (!truncate_happened
) {
1784 printk("ineffectual truncate (%lx, %lx, %llx)\n", inode
->i_ino
, ipage
->index
, size
);
1788 this_wc
->flags
= WF_DELETE
;
1789 if (logfs_block(ipage
)->partial
)
1790 this_wc
->flags
|= WF_WRITE
;
1792 return logfs_write_i0(inode
, ipage
, this_wc
);
1795 static int logfs_truncate_rec(struct inode
*inode
, u64 size
)
1797 struct logfs_inode
*li
= logfs_inode(inode
);
1798 struct write_control wc
= {
1799 .ofs
= li
->li_data
[INDIRECT_INDEX
],
1804 alloc_inode_block(inode
);
1809 page
= logfs_get_write_page(inode
, 0, LEVEL(li
->li_height
));
1813 err
= __logfs_truncate_rec(inode
, page
, &wc
, size
);
1814 logfs_put_write_page(page
);
1818 if (li
->li_data
[INDIRECT_INDEX
] != wc
.ofs
)
1819 li
->li_data
[INDIRECT_INDEX
] = wc
.ofs
;
1823 static int __logfs_truncate(struct inode
*inode
, u64 size
)
1827 if (size
>= logfs_factor(logfs_inode(inode
)->li_height
))
1830 ret
= logfs_truncate_rec(inode
, size
);
1834 return logfs_truncate_direct(inode
, size
);
1837 int logfs_truncate(struct inode
*inode
, u64 size
)
1839 struct super_block
*sb
= inode
->i_sb
;
1842 logfs_get_wblocks(sb
, NULL
, 1);
1843 err
= __logfs_truncate(inode
, size
);
1845 err
= __logfs_write_inode(inode
, 0);
1846 logfs_put_wblocks(sb
, NULL
, 1);
1849 err
= vmtruncate(inode
, size
);
1851 /* I don't trust error recovery yet. */
1856 static void move_page_to_inode(struct inode
*inode
, struct page
*page
)
1858 struct logfs_inode
*li
= logfs_inode(inode
);
1859 struct logfs_block
*block
= logfs_block(page
);
1864 log_blockmove("move_page_to_inode(%llx, %llx, %x)\n",
1865 block
->ino
, block
->bix
, block
->level
);
1866 BUG_ON(li
->li_block
);
1867 block
->ops
= &inode_block_ops
;
1868 block
->inode
= inode
;
1869 li
->li_block
= block
;
1873 ClearPagePrivate(page
);
1876 static void move_inode_to_page(struct page
*page
, struct inode
*inode
)
1878 struct logfs_inode
*li
= logfs_inode(inode
);
1879 struct logfs_block
*block
= li
->li_block
;
1884 log_blockmove("move_inode_to_page(%llx, %llx, %x)\n",
1885 block
->ino
, block
->bix
, block
->level
);
1886 BUG_ON(PagePrivate(page
));
1887 block
->ops
= &indirect_block_ops
;
1889 page
->private = (unsigned long)block
;
1890 SetPagePrivate(page
);
1892 block
->inode
= NULL
;
1893 li
->li_block
= NULL
;
1896 int logfs_read_inode(struct inode
*inode
)
1898 struct super_block
*sb
= inode
->i_sb
;
1899 struct logfs_super
*super
= logfs_super(sb
);
1900 struct inode
*master_inode
= super
->s_master_inode
;
1902 struct logfs_disk_inode
*di
;
1903 u64 ino
= inode
->i_ino
;
1905 if (ino
<< sb
->s_blocksize_bits
> i_size_read(master_inode
))
1907 if (!logfs_exist_block(master_inode
, ino
))
1910 page
= read_cache_page(master_inode
->i_mapping
, ino
,
1911 (filler_t
*)logfs_readpage
, NULL
);
1913 return PTR_ERR(page
);
1915 di
= kmap_atomic(page
, KM_USER0
);
1916 logfs_disk_to_inode(di
, inode
);
1917 kunmap_atomic(di
, KM_USER0
);
1918 move_page_to_inode(inode
, page
);
1919 page_cache_release(page
);
1923 /* Caller must logfs_put_write_page(page); */
1924 static struct page
*inode_to_page(struct inode
*inode
)
1926 struct inode
*master_inode
= logfs_super(inode
->i_sb
)->s_master_inode
;
1927 struct logfs_disk_inode
*di
;
1930 BUG_ON(inode
->i_ino
== LOGFS_INO_MASTER
);
1932 page
= logfs_get_write_page(master_inode
, inode
->i_ino
, 0);
1936 di
= kmap_atomic(page
, KM_USER0
);
1937 logfs_inode_to_disk(inode
, di
);
1938 kunmap_atomic(di
, KM_USER0
);
1939 move_inode_to_page(page
, inode
);
1943 /* Cheaper version of write_inode. All changes are concealed in
1944 * aliases, which are moved back. No write to the medium happens.
1946 void logfs_clear_inode(struct inode
*inode
)
1948 struct super_block
*sb
= inode
->i_sb
;
1949 struct logfs_inode
*li
= logfs_inode(inode
);
1950 struct logfs_block
*block
= li
->li_block
;
1953 /* Only deleted files may be dirty at this point */
1954 BUG_ON(inode
->i_state
& I_DIRTY
&& inode
->i_nlink
);
1957 if ((logfs_super(sb
)->s_flags
& LOGFS_SB_FLAG_SHUTDOWN
)) {
1958 block
->ops
->free_block(inode
->i_sb
, block
);
1962 BUG_ON(inode
->i_ino
< LOGFS_RESERVED_INOS
);
1963 page
= inode_to_page(inode
);
1964 BUG_ON(!page
); /* FIXME: Use emergency page */
1965 logfs_put_write_page(page
);
1968 static int do_write_inode(struct inode
*inode
)
1970 struct super_block
*sb
= inode
->i_sb
;
1971 struct inode
*master_inode
= logfs_super(sb
)->s_master_inode
;
1972 loff_t size
= (inode
->i_ino
+ 1) << inode
->i_sb
->s_blocksize_bits
;
1976 BUG_ON(inode
->i_ino
== LOGFS_INO_MASTER
);
1977 /* FIXME: lock inode */
1979 if (i_size_read(master_inode
) < size
)
1980 i_size_write(master_inode
, size
);
1982 /* TODO: Tell vfs this inode is clean now */
1984 page
= inode_to_page(inode
);
1988 /* FIXME: transaction is part of logfs_block now. Is that enough? */
1989 err
= logfs_write_buf(master_inode
, page
, 0);
1990 logfs_put_write_page(page
);
1994 static void logfs_mod_segment_entry(struct super_block
*sb
, u32 segno
,
1996 void (*change_se
)(struct logfs_segment_entry
*, long),
1999 struct logfs_super
*super
= logfs_super(sb
);
2000 struct inode
*inode
;
2002 struct logfs_segment_entry
*se
;
2006 page_no
= segno
>> (sb
->s_blocksize_bits
- 3);
2007 child_no
= segno
& ((sb
->s_blocksize
>> 3) - 1);
2009 inode
= super
->s_segfile_inode
;
2010 page
= logfs_get_write_page(inode
, page_no
, 0);
2011 BUG_ON(!page
); /* FIXME: We need some reserve page for this case */
2012 if (!PageUptodate(page
))
2013 logfs_read_block(inode
, page
, WRITE
);
2016 alloc_indirect_block(inode
, page
, 0);
2017 se
= kmap_atomic(page
, KM_USER0
);
2018 change_se(se
+ child_no
, arg
);
2020 logfs_set_alias(sb
, logfs_block(page
), child_no
);
2021 BUG_ON((int)be32_to_cpu(se
[child_no
].valid
) > super
->s_segsize
);
2023 kunmap_atomic(se
, KM_USER0
);
2025 logfs_put_write_page(page
);
2028 static void __get_segment_entry(struct logfs_segment_entry
*se
, long _target
)
2030 struct logfs_segment_entry
*target
= (void *)_target
;
2035 void logfs_get_segment_entry(struct super_block
*sb
, u32 segno
,
2036 struct logfs_segment_entry
*se
)
2038 logfs_mod_segment_entry(sb
, segno
, 0, __get_segment_entry
, (long)se
);
2041 static void __set_segment_used(struct logfs_segment_entry
*se
, long increment
)
2045 valid
= be32_to_cpu(se
->valid
);
2047 se
->valid
= cpu_to_be32(valid
);
2050 void logfs_set_segment_used(struct super_block
*sb
, u64 ofs
, int increment
)
2052 struct logfs_super
*super
= logfs_super(sb
);
2053 u32 segno
= ofs
>> super
->s_segshift
;
2058 logfs_mod_segment_entry(sb
, segno
, 1, __set_segment_used
, increment
);
2061 static void __set_segment_erased(struct logfs_segment_entry
*se
, long ec_level
)
2063 se
->ec_level
= cpu_to_be32(ec_level
);
2066 void logfs_set_segment_erased(struct super_block
*sb
, u32 segno
, u32 ec
,
2067 gc_level_t gc_level
)
2069 u32 ec_level
= ec
<< 4 | (__force u8
)gc_level
;
2071 logfs_mod_segment_entry(sb
, segno
, 1, __set_segment_erased
, ec_level
);
2074 static void __set_segment_reserved(struct logfs_segment_entry
*se
, long ignore
)
2076 se
->valid
= cpu_to_be32(RESERVED
);
2079 void logfs_set_segment_reserved(struct super_block
*sb
, u32 segno
)
2081 logfs_mod_segment_entry(sb
, segno
, 1, __set_segment_reserved
, 0);
2084 static void __set_segment_unreserved(struct logfs_segment_entry
*se
,
2088 se
->ec_level
= cpu_to_be32(ec_level
);
2091 void logfs_set_segment_unreserved(struct super_block
*sb
, u32 segno
, u32 ec
)
2093 u32 ec_level
= ec
<< 4;
2095 logfs_mod_segment_entry(sb
, segno
, 1, __set_segment_unreserved
,
2099 int __logfs_write_inode(struct inode
*inode
, long flags
)
2101 struct super_block
*sb
= inode
->i_sb
;
2104 logfs_get_wblocks(sb
, NULL
, flags
& WF_LOCK
);
2105 ret
= do_write_inode(inode
);
2106 logfs_put_wblocks(sb
, NULL
, flags
& WF_LOCK
);
2110 static int do_delete_inode(struct inode
*inode
)
2112 struct super_block
*sb
= inode
->i_sb
;
2113 struct inode
*master_inode
= logfs_super(sb
)->s_master_inode
;
2117 page
= logfs_get_write_page(master_inode
, inode
->i_ino
, 0);
2121 move_inode_to_page(page
, inode
);
2123 logfs_get_wblocks(sb
, page
, 1);
2124 ret
= __logfs_delete(master_inode
, page
);
2125 logfs_put_wblocks(sb
, page
, 1);
2127 logfs_put_write_page(page
);
2132 * ZOMBIE inodes have already been deleted before and should remain dead,
2133 * if it weren't for valid checking. No need to kill them again here.
2135 void logfs_delete_inode(struct inode
*inode
)
2137 struct logfs_inode
*li
= logfs_inode(inode
);
2139 if (!(li
->li_flags
& LOGFS_IF_ZOMBIE
)) {
2140 li
->li_flags
|= LOGFS_IF_ZOMBIE
;
2141 if (i_size_read(inode
) > 0)
2142 logfs_truncate(inode
, 0);
2143 do_delete_inode(inode
);
2145 truncate_inode_pages(&inode
->i_data
, 0);
2149 void btree_write_block(struct logfs_block
*block
)
2151 struct inode
*inode
;
2155 inode
= logfs_safe_iget(block
->sb
, block
->ino
, &cookie
);
2156 page
= logfs_get_write_page(inode
, block
->bix
, block
->level
);
2158 err
= logfs_readpage_nolock(page
);
2160 BUG_ON(!PagePrivate(page
));
2161 BUG_ON(logfs_block(page
) != block
);
2162 err
= __logfs_write_buf(inode
, page
, 0);
2164 BUG_ON(PagePrivate(page
) || page
->private);
2166 logfs_put_write_page(page
);
2167 logfs_safe_iput(inode
, cookie
);
2171 * logfs_inode_write - write inode or dentry objects
2173 * @inode: parent inode (ifile or directory)
2174 * @buf: object to write (inode or dentry)
2176 * @_pos: object number (file position in blocks/objects)
2177 * @flags: write flags
2178 * @lock: 0 if write lock is already taken, 1 otherwise
2179 * @shadow_tree: shadow below this inode
2181 * FIXME: All caller of this put a 200-300 byte variable on the stack,
2182 * only to call here and do a memcpy from that stack variable. A good
2183 * example of wasted performance and stack space.
2185 int logfs_inode_write(struct inode
*inode
, const void *buf
, size_t count
,
2186 loff_t bix
, long flags
, struct shadow_tree
*shadow_tree
)
2188 loff_t pos
= bix
<< inode
->i_sb
->s_blocksize_bits
;
2193 BUG_ON(pos
& (LOGFS_BLOCKSIZE
-1));
2194 BUG_ON(count
> LOGFS_BLOCKSIZE
);
2195 page
= logfs_get_write_page(inode
, bix
, 0);
2199 pagebuf
= kmap_atomic(page
, KM_USER0
);
2200 memcpy(pagebuf
, buf
, count
);
2201 flush_dcache_page(page
);
2202 kunmap_atomic(pagebuf
, KM_USER0
);
2204 if (i_size_read(inode
) < pos
+ LOGFS_BLOCKSIZE
)
2205 i_size_write(inode
, pos
+ LOGFS_BLOCKSIZE
);
2207 err
= logfs_write_buf(inode
, page
, flags
);
2208 logfs_put_write_page(page
);
2212 int logfs_open_segfile(struct super_block
*sb
)
2214 struct logfs_super
*super
= logfs_super(sb
);
2215 struct inode
*inode
;
2217 inode
= logfs_read_meta_inode(sb
, LOGFS_INO_SEGFILE
);
2219 return PTR_ERR(inode
);
2220 super
->s_segfile_inode
= inode
;
2224 int logfs_init_rw(struct super_block
*sb
)
2226 struct logfs_super
*super
= logfs_super(sb
);
2227 int min_fill
= 3 * super
->s_no_blocks
;
2229 INIT_LIST_HEAD(&super
->s_object_alias
);
2230 mutex_init(&super
->s_write_mutex
);
2231 super
->s_block_pool
= mempool_create_kmalloc_pool(min_fill
,
2232 sizeof(struct logfs_block
));
2233 super
->s_shadow_pool
= mempool_create_kmalloc_pool(min_fill
,
2234 sizeof(struct logfs_shadow
));
2238 void logfs_cleanup_rw(struct super_block
*sb
)
2240 struct logfs_super
*super
= logfs_super(sb
);
2242 destroy_meta_inode(super
->s_segfile_inode
);
2243 if (super
->s_block_pool
)
2244 mempool_destroy(super
->s_block_pool
);
2245 if (super
->s_shadow_pool
)
2246 mempool_destroy(super
->s_shadow_pool
);