]>
Commit | Line | Data |
---|---|---|
5db53f3e JE |
1 | /* |
2 | * fs/logfs/readwrite.c | |
3 | * | |
4 | * As should be obvious for Linux kernel code, license is GPLv2 | |
5 | * | |
6 | * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org> | |
7 | * | |
8 | * | |
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 | |
18 | */ | |
19 | #include "logfs.h" | |
20 | #include <linux/sched.h> | |
21 | ||
22 | static u64 adjust_bix(u64 bix, level_t level) | |
23 | { | |
24 | switch (level) { | |
25 | case 0: | |
26 | return bix; | |
27 | case LEVEL(1): | |
28 | return max_t(u64, bix, I0_BLOCKS); | |
29 | case LEVEL(2): | |
30 | return max_t(u64, bix, I1_BLOCKS); | |
31 | case LEVEL(3): | |
32 | return max_t(u64, bix, I2_BLOCKS); | |
33 | case LEVEL(4): | |
34 | return max_t(u64, bix, I3_BLOCKS); | |
35 | case LEVEL(5): | |
36 | return max_t(u64, bix, I4_BLOCKS); | |
37 | default: | |
38 | WARN_ON(1); | |
39 | return bix; | |
40 | } | |
41 | } | |
42 | ||
43 | static inline u64 maxbix(u8 height) | |
44 | { | |
45 | return 1ULL << (LOGFS_BLOCK_BITS * height); | |
46 | } | |
47 | ||
48 | /** | |
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 | |
52 | * index. | |
53 | * | |
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. | |
57 | */ | |
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) | |
62 | { | |
63 | return INDIRECT_BIT | (1ULL << LEVEL_SHIFT); | |
64 | } | |
65 | ||
66 | pgoff_t logfs_pack_index(u64 bix, level_t level) | |
67 | { | |
68 | pgoff_t index; | |
69 | ||
70 | BUG_ON(bix >= INDIRECT_BIT); | |
71 | if (level == 0) | |
72 | return bix; | |
73 | ||
74 | index = INDIRECT_BIT; | |
75 | index |= (__force long)level << LEVEL_SHIFT; | |
76 | index |= bix >> ((__force u8)level * LOGFS_BLOCK_BITS); | |
77 | return index; | |
78 | } | |
79 | ||
80 | void logfs_unpack_index(pgoff_t index, u64 *bix, level_t *level) | |
81 | { | |
82 | u8 __level; | |
83 | ||
84 | if (!(index & INDIRECT_BIT)) { | |
85 | *bix = index; | |
86 | *level = 0; | |
87 | return; | |
88 | } | |
89 | ||
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); | |
94 | return; | |
95 | } | |
96 | #undef ARCH_SHIFT | |
97 | #undef INDIRECT_BIT | |
98 | #undef LEVEL_SHIFT | |
99 | ||
100 | /* | |
101 | * Time is stored as nanoseconds since the epoch. | |
102 | */ | |
103 | static struct timespec be64_to_timespec(__be64 betime) | |
104 | { | |
105 | return ns_to_timespec(be64_to_cpu(betime)); | |
106 | } | |
107 | ||
108 | static __be64 timespec_to_be64(struct timespec tsp) | |
109 | { | |
110 | return cpu_to_be64((u64)tsp.tv_sec * NSEC_PER_SEC + tsp.tv_nsec); | |
111 | } | |
112 | ||
113 | static void logfs_disk_to_inode(struct logfs_disk_inode *di, struct inode*inode) | |
114 | { | |
115 | struct logfs_inode *li = logfs_inode(inode); | |
116 | int i; | |
117 | ||
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); | |
130 | ||
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 */ | |
135 | case S_IFIFO: | |
136 | inode->i_rdev = be64_to_cpu(di->di_data[0]); | |
137 | break; | |
138 | case S_IFDIR: /* fall through */ | |
139 | case S_IFREG: /* fall through */ | |
140 | case S_IFLNK: | |
141 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
142 | li->li_data[i] = be64_to_cpu(di->di_data[i]); | |
143 | break; | |
144 | default: | |
145 | BUG(); | |
146 | } | |
147 | } | |
148 | ||
149 | static void logfs_inode_to_disk(struct inode *inode, struct logfs_disk_inode*di) | |
150 | { | |
151 | struct logfs_inode *li = logfs_inode(inode); | |
152 | int i; | |
153 | ||
154 | di->di_mode = cpu_to_be16(inode->i_mode); | |
155 | di->di_height = li->li_height; | |
156 | di->di_pad = 0; | |
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); | |
167 | ||
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 */ | |
172 | case S_IFIFO: | |
173 | di->di_data[0] = cpu_to_be64(inode->i_rdev); | |
174 | break; | |
175 | case S_IFDIR: /* fall through */ | |
176 | case S_IFREG: /* fall through */ | |
177 | case S_IFLNK: | |
178 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
179 | di->di_data[i] = cpu_to_be64(li->li_data[i]); | |
180 | break; | |
181 | default: | |
182 | BUG(); | |
183 | } | |
184 | } | |
185 | ||
186 | static void __logfs_set_blocks(struct inode *inode) | |
187 | { | |
188 | struct super_block *sb = inode->i_sb; | |
189 | struct logfs_inode *li = logfs_inode(inode); | |
190 | ||
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; | |
194 | } | |
195 | ||
196 | void logfs_set_blocks(struct inode *inode, u64 bytes) | |
197 | { | |
198 | struct logfs_inode *li = logfs_inode(inode); | |
199 | ||
200 | li->li_used_bytes = bytes; | |
201 | __logfs_set_blocks(inode); | |
202 | } | |
203 | ||
204 | static void prelock_page(struct super_block *sb, struct page *page, int lock) | |
205 | { | |
206 | struct logfs_super *super = logfs_super(sb); | |
207 | ||
208 | BUG_ON(!PageLocked(page)); | |
209 | if (lock) { | |
210 | BUG_ON(PagePreLocked(page)); | |
211 | SetPagePreLocked(page); | |
212 | } else { | |
213 | /* We are in GC path. */ | |
214 | if (PagePreLocked(page)) | |
215 | super->s_lock_count++; | |
216 | else | |
217 | SetPagePreLocked(page); | |
218 | } | |
219 | } | |
220 | ||
221 | static void preunlock_page(struct super_block *sb, struct page *page, int lock) | |
222 | { | |
223 | struct logfs_super *super = logfs_super(sb); | |
224 | ||
225 | BUG_ON(!PageLocked(page)); | |
226 | if (lock) | |
227 | ClearPagePreLocked(page); | |
228 | else { | |
229 | /* We are in GC path. */ | |
230 | BUG_ON(!PagePreLocked(page)); | |
231 | if (super->s_lock_count) | |
232 | super->s_lock_count--; | |
233 | else | |
234 | ClearPagePreLocked(page); | |
235 | } | |
236 | } | |
237 | ||
238 | /* | |
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 | |
241 | * s_write_mutex. | |
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. | |
245 | */ | |
246 | static void logfs_get_wblocks(struct super_block *sb, struct page *page, | |
247 | int lock) | |
248 | { | |
249 | struct logfs_super *super = logfs_super(sb); | |
250 | ||
251 | if (page) | |
252 | prelock_page(sb, page, lock); | |
253 | ||
254 | if (lock) { | |
255 | mutex_lock(&super->s_write_mutex); | |
256 | logfs_gc_pass(sb); | |
257 | /* FIXME: We also have to check for shadowed space | |
258 | * and mempool fill grade */ | |
259 | } | |
260 | } | |
261 | ||
262 | static void logfs_put_wblocks(struct super_block *sb, struct page *page, | |
263 | int lock) | |
264 | { | |
265 | struct logfs_super *super = logfs_super(sb); | |
266 | ||
267 | if (page) | |
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. */ | |
271 | if (lock) | |
272 | mutex_unlock(&super->s_write_mutex); | |
273 | } | |
274 | ||
275 | static struct page *logfs_get_read_page(struct inode *inode, u64 bix, | |
276 | level_t level) | |
277 | { | |
278 | return find_or_create_page(inode->i_mapping, | |
279 | logfs_pack_index(bix, level), GFP_NOFS); | |
280 | } | |
281 | ||
282 | static void logfs_put_read_page(struct page *page) | |
283 | { | |
284 | unlock_page(page); | |
285 | page_cache_release(page); | |
286 | } | |
287 | ||
288 | static void logfs_lock_write_page(struct page *page) | |
289 | { | |
290 | int loop = 0; | |
291 | ||
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); | |
296 | BUG(); | |
297 | } | |
298 | if (PagePreLocked(page)) { | |
299 | /* Holder of page lock is waiting for us, it | |
300 | * is safe to use this page. */ | |
301 | break; | |
302 | } | |
303 | /* Some other process has this page locked and has | |
304 | * nothing to do with us. Wait for it to finish. | |
305 | */ | |
306 | schedule(); | |
307 | } | |
308 | BUG_ON(!PageLocked(page)); | |
309 | } | |
310 | ||
311 | static struct page *logfs_get_write_page(struct inode *inode, u64 bix, | |
312 | level_t level) | |
313 | { | |
314 | struct address_space *mapping = inode->i_mapping; | |
315 | pgoff_t index = logfs_pack_index(bix, level); | |
316 | struct page *page; | |
317 | int err; | |
318 | ||
319 | repeat: | |
320 | page = find_get_page(mapping, index); | |
321 | if (!page) { | |
322 | page = __page_cache_alloc(GFP_NOFS); | |
323 | if (!page) | |
324 | return NULL; | |
325 | err = add_to_page_cache_lru(page, mapping, index, GFP_NOFS); | |
326 | if (unlikely(err)) { | |
327 | page_cache_release(page); | |
328 | if (err == -EEXIST) | |
329 | goto repeat; | |
330 | return NULL; | |
331 | } | |
332 | } else logfs_lock_write_page(page); | |
333 | BUG_ON(!PageLocked(page)); | |
334 | return page; | |
335 | } | |
336 | ||
337 | static void logfs_unlock_write_page(struct page *page) | |
338 | { | |
339 | if (!PagePreLocked(page)) | |
340 | unlock_page(page); | |
341 | } | |
342 | ||
343 | static void logfs_put_write_page(struct page *page) | |
344 | { | |
345 | logfs_unlock_write_page(page); | |
346 | page_cache_release(page); | |
347 | } | |
348 | ||
349 | static struct page *logfs_get_page(struct inode *inode, u64 bix, level_t level, | |
350 | int rw) | |
351 | { | |
352 | if (rw == READ) | |
353 | return logfs_get_read_page(inode, bix, level); | |
354 | else | |
355 | return logfs_get_write_page(inode, bix, level); | |
356 | } | |
357 | ||
358 | static void logfs_put_page(struct page *page, int rw) | |
359 | { | |
360 | if (rw == READ) | |
361 | logfs_put_read_page(page); | |
362 | else | |
363 | logfs_put_write_page(page); | |
364 | } | |
365 | ||
366 | static unsigned long __get_bits(u64 val, int skip, int no) | |
367 | { | |
368 | u64 ret = val; | |
369 | ||
370 | ret >>= skip * no; | |
371 | ret <<= 64 - no; | |
372 | ret >>= 64 - no; | |
373 | return ret; | |
374 | } | |
375 | ||
376 | static unsigned long get_bits(u64 val, level_t skip) | |
377 | { | |
378 | return __get_bits(val, (__force int)skip, LOGFS_BLOCK_BITS); | |
379 | } | |
380 | ||
381 | static inline void init_shadow_tree(struct super_block *sb, | |
382 | struct shadow_tree *tree) | |
383 | { | |
384 | struct logfs_super *super = logfs_super(sb); | |
385 | ||
386 | btree_init_mempool64(&tree->new, super->s_btree_pool); | |
387 | btree_init_mempool64(&tree->old, super->s_btree_pool); | |
388 | } | |
389 | ||
390 | static void indirect_write_block(struct logfs_block *block) | |
391 | { | |
392 | struct page *page; | |
393 | struct inode *inode; | |
394 | int ret; | |
395 | ||
396 | page = block->page; | |
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); | |
401 | /* | |
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. | |
407 | * | |
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 | |
412 | * entirely tasteles. | |
413 | */ | |
414 | BUG_ON(ret); | |
415 | } | |
416 | ||
417 | static void inode_write_block(struct logfs_block *block) | |
418 | { | |
419 | struct inode *inode; | |
420 | int ret; | |
421 | ||
422 | inode = block->inode; | |
423 | if (inode->i_ino == LOGFS_INO_MASTER) | |
c6d38301 | 424 | logfs_write_anchor(inode->i_sb); |
5db53f3e JE |
425 | else { |
426 | ret = __logfs_write_inode(inode, 0); | |
427 | /* see indirect_write_block comment */ | |
428 | BUG_ON(ret); | |
429 | } | |
430 | } | |
431 | ||
5db53f3e JE |
432 | /* |
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. | |
436 | */ | |
437 | static unsigned long fnb(const unsigned long *addr, | |
438 | unsigned long size, unsigned long offset) | |
439 | { | |
440 | return find_next_bit(addr, size, offset); | |
441 | } | |
442 | ||
443 | static __be64 inode_val0(struct inode *inode) | |
444 | { | |
445 | struct logfs_inode *li = logfs_inode(inode); | |
446 | u64 val; | |
447 | ||
448 | /* | |
449 | * Explicit shifting generates good code, but must match the format | |
450 | * of the structure. Add some paranoia just in case. | |
451 | */ | |
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); | |
455 | ||
456 | val = (u64)inode->i_mode << 48 | | |
457 | (u64)li->li_height << 40 | | |
458 | (u64)li->li_flags; | |
459 | return cpu_to_be64(val); | |
460 | } | |
461 | ||
462 | static int inode_write_alias(struct super_block *sb, | |
463 | struct logfs_block *block, write_alias_t *write_one_alias) | |
464 | { | |
465 | struct inode *inode = block->inode; | |
466 | struct logfs_inode *li = logfs_inode(inode); | |
467 | unsigned long pos; | |
468 | u64 ino , bix; | |
469 | __be64 val; | |
470 | level_t level; | |
471 | int err; | |
472 | ||
473 | for (pos = 0; ; pos++) { | |
474 | pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos); | |
475 | if (pos >= LOGFS_EMBEDDED_FIELDS + INODE_POINTER_OFS) | |
476 | return 0; | |
477 | ||
478 | switch (pos) { | |
479 | case INODE_HEIGHT_OFS: | |
480 | val = inode_val0(inode); | |
481 | break; | |
482 | case INODE_USED_OFS: | |
483 | val = cpu_to_be64(li->li_used_bytes);; | |
484 | break; | |
485 | case INODE_SIZE_OFS: | |
486 | val = cpu_to_be64(i_size_read(inode)); | |
487 | break; | |
488 | case INODE_POINTER_OFS ... INODE_POINTER_OFS + LOGFS_EMBEDDED_FIELDS - 1: | |
489 | val = cpu_to_be64(li->li_data[pos - INODE_POINTER_OFS]); | |
490 | break; | |
491 | default: | |
492 | BUG(); | |
493 | } | |
494 | ||
495 | ino = LOGFS_INO_MASTER; | |
496 | bix = inode->i_ino; | |
497 | level = LEVEL(0); | |
498 | err = write_one_alias(sb, ino, bix, level, pos, val); | |
499 | if (err) | |
500 | return err; | |
501 | } | |
502 | } | |
503 | ||
504 | static int indirect_write_alias(struct super_block *sb, | |
505 | struct logfs_block *block, write_alias_t *write_one_alias) | |
506 | { | |
507 | unsigned long pos; | |
508 | struct page *page = block->page; | |
509 | u64 ino , bix; | |
510 | __be64 *child, val; | |
511 | level_t level; | |
512 | int err; | |
513 | ||
514 | for (pos = 0; ; pos++) { | |
515 | pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos); | |
516 | if (pos >= LOGFS_BLOCK_FACTOR) | |
517 | return 0; | |
518 | ||
519 | ino = page->mapping->host->i_ino; | |
520 | logfs_unpack_index(page->index, &bix, &level); | |
521 | child = kmap_atomic(page, KM_USER0); | |
522 | val = child[pos]; | |
523 | kunmap_atomic(child, KM_USER0); | |
524 | err = write_one_alias(sb, ino, bix, level, pos, val); | |
525 | if (err) | |
526 | return err; | |
527 | } | |
528 | } | |
529 | ||
530 | int logfs_write_obj_aliases_pagecache(struct super_block *sb) | |
531 | { | |
532 | struct logfs_super *super = logfs_super(sb); | |
533 | struct logfs_block *block; | |
534 | int err; | |
535 | ||
536 | list_for_each_entry(block, &super->s_object_alias, alias_list) { | |
537 | err = block->ops->write_alias(sb, block, write_alias_journal); | |
538 | if (err) | |
539 | return err; | |
540 | } | |
541 | return 0; | |
542 | } | |
543 | ||
544 | void __free_block(struct super_block *sb, struct logfs_block *block) | |
545 | { | |
546 | BUG_ON(!list_empty(&block->item_list)); | |
547 | list_del(&block->alias_list); | |
548 | mempool_free(block, logfs_super(sb)->s_block_pool); | |
549 | } | |
550 | ||
551 | static void inode_free_block(struct super_block *sb, struct logfs_block *block) | |
552 | { | |
553 | struct inode *inode = block->inode; | |
554 | ||
555 | logfs_inode(inode)->li_block = NULL; | |
556 | __free_block(sb, block); | |
557 | } | |
558 | ||
559 | static void indirect_free_block(struct super_block *sb, | |
560 | struct logfs_block *block) | |
561 | { | |
562 | ClearPagePrivate(block->page); | |
563 | block->page->private = 0; | |
564 | __free_block(sb, block); | |
565 | } | |
566 | ||
567 | ||
568 | static struct logfs_block_ops inode_block_ops = { | |
569 | .write_block = inode_write_block, | |
5db53f3e JE |
570 | .free_block = inode_free_block, |
571 | .write_alias = inode_write_alias, | |
572 | }; | |
573 | ||
574 | struct logfs_block_ops indirect_block_ops = { | |
575 | .write_block = indirect_write_block, | |
5db53f3e JE |
576 | .free_block = indirect_free_block, |
577 | .write_alias = indirect_write_alias, | |
578 | }; | |
579 | ||
580 | struct logfs_block *__alloc_block(struct super_block *sb, | |
581 | u64 ino, u64 bix, level_t level) | |
582 | { | |
583 | struct logfs_super *super = logfs_super(sb); | |
584 | struct logfs_block *block; | |
585 | ||
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); | |
590 | block->sb = sb; | |
591 | block->ino = ino; | |
592 | block->bix = bix; | |
593 | block->level = level; | |
594 | return block; | |
595 | } | |
596 | ||
597 | static void alloc_inode_block(struct inode *inode) | |
598 | { | |
599 | struct logfs_inode *li = logfs_inode(inode); | |
600 | struct logfs_block *block; | |
601 | ||
602 | if (li->li_block) | |
603 | return; | |
604 | ||
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; | |
609 | } | |
610 | ||
611 | void initialize_block_counters(struct page *page, struct logfs_block *block, | |
612 | __be64 *array, int page_is_empty) | |
613 | { | |
614 | u64 ptr; | |
615 | int i, start; | |
616 | ||
617 | block->partial = 0; | |
618 | block->full = 0; | |
619 | start = 0; | |
620 | if (page->index < first_indirect_block()) { | |
621 | /* Counters are pointless on level 0 */ | |
622 | return; | |
623 | } | |
624 | if (page->index == first_indirect_block()) { | |
625 | /* Skip unused pointers */ | |
626 | start = I0_BLOCKS; | |
627 | block->full = I0_BLOCKS; | |
628 | } | |
629 | if (!page_is_empty) { | |
630 | for (i = start; i < LOGFS_BLOCK_FACTOR; i++) { | |
631 | ptr = be64_to_cpu(array[i]); | |
632 | if (ptr) | |
633 | block->partial++; | |
634 | if (ptr & LOGFS_FULLY_POPULATED) | |
635 | block->full++; | |
636 | } | |
637 | } | |
638 | } | |
639 | ||
640 | static void alloc_data_block(struct inode *inode, struct page *page) | |
641 | { | |
642 | struct logfs_block *block; | |
643 | u64 bix; | |
644 | level_t level; | |
645 | ||
646 | if (PagePrivate(page)) | |
647 | return; | |
648 | ||
649 | logfs_unpack_index(page->index, &bix, &level); | |
650 | block = __alloc_block(inode->i_sb, inode->i_ino, bix, level); | |
651 | block->page = page; | |
652 | SetPagePrivate(page); | |
653 | page->private = (unsigned long)block; | |
654 | block->ops = &indirect_block_ops; | |
655 | } | |
656 | ||
657 | static void alloc_indirect_block(struct inode *inode, struct page *page, | |
658 | int page_is_empty) | |
659 | { | |
660 | struct logfs_block *block; | |
661 | __be64 *array; | |
662 | ||
663 | if (PagePrivate(page)) | |
664 | return; | |
665 | ||
666 | alloc_data_block(inode, page); | |
667 | ||
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); | |
672 | } | |
673 | ||
674 | static void block_set_pointer(struct page *page, int index, u64 ptr) | |
675 | { | |
676 | struct logfs_block *block = logfs_block(page); | |
677 | __be64 *array; | |
678 | u64 oldptr; | |
679 | ||
680 | BUG_ON(!block); | |
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); | |
686 | ||
687 | block->full += !!(ptr & LOGFS_FULLY_POPULATED) | |
688 | - !!(oldptr & LOGFS_FULLY_POPULATED); | |
689 | block->partial += !!ptr - !!oldptr; | |
690 | } | |
691 | ||
692 | static u64 block_get_pointer(struct page *page, int index) | |
693 | { | |
694 | __be64 *block; | |
695 | u64 ptr; | |
696 | ||
697 | block = kmap_atomic(page, KM_USER0); | |
698 | ptr = be64_to_cpu(block[index]); | |
699 | kunmap_atomic(block, KM_USER0); | |
700 | return ptr; | |
701 | } | |
702 | ||
703 | static int logfs_read_empty(struct page *page) | |
704 | { | |
705 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
706 | return 0; | |
707 | } | |
708 | ||
709 | static int logfs_read_direct(struct inode *inode, struct page *page) | |
710 | { | |
711 | struct logfs_inode *li = logfs_inode(inode); | |
712 | pgoff_t index = page->index; | |
713 | u64 block; | |
714 | ||
715 | block = li->li_data[index]; | |
716 | if (!block) | |
717 | return logfs_read_empty(page); | |
718 | ||
719 | return logfs_segment_read(inode, page, block, index, 0); | |
720 | } | |
721 | ||
722 | static int logfs_read_loop(struct inode *inode, struct page *page, | |
723 | int rw_context) | |
724 | { | |
725 | struct logfs_inode *li = logfs_inode(inode); | |
726 | u64 bix, bofs = li->li_data[INDIRECT_INDEX]; | |
727 | level_t level, target_level; | |
728 | int ret; | |
729 | struct page *ipage; | |
730 | ||
731 | logfs_unpack_index(page->index, &bix, &target_level); | |
732 | if (!bofs) | |
733 | return logfs_read_empty(page); | |
734 | ||
735 | if (bix >= maxbix(li->li_height)) | |
736 | return logfs_read_empty(page); | |
737 | ||
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); | |
742 | if (!ipage) | |
743 | return -ENOMEM; | |
744 | ||
745 | ret = logfs_segment_read(inode, ipage, bofs, bix, level); | |
746 | if (ret) { | |
747 | logfs_put_read_page(ipage); | |
748 | return ret; | |
749 | } | |
750 | ||
751 | bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level))); | |
752 | logfs_put_page(ipage, rw_context); | |
753 | if (!bofs) | |
754 | return logfs_read_empty(page); | |
755 | } | |
756 | ||
757 | return logfs_segment_read(inode, page, bofs, bix, 0); | |
758 | } | |
759 | ||
760 | static int logfs_read_block(struct inode *inode, struct page *page, | |
761 | int rw_context) | |
762 | { | |
763 | pgoff_t index = page->index; | |
764 | ||
765 | if (index < I0_BLOCKS) | |
766 | return logfs_read_direct(inode, page); | |
767 | return logfs_read_loop(inode, page, rw_context); | |
768 | } | |
769 | ||
770 | static int logfs_exist_loop(struct inode *inode, u64 bix) | |
771 | { | |
772 | struct logfs_inode *li = logfs_inode(inode); | |
773 | u64 bofs = li->li_data[INDIRECT_INDEX]; | |
774 | level_t level; | |
775 | int ret; | |
776 | struct page *ipage; | |
777 | ||
778 | if (!bofs) | |
779 | return 0; | |
780 | if (bix >= maxbix(li->li_height)) | |
781 | return 0; | |
782 | ||
783 | for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)) { | |
784 | ipage = logfs_get_read_page(inode, bix, level); | |
785 | if (!ipage) | |
786 | return -ENOMEM; | |
787 | ||
788 | ret = logfs_segment_read(inode, ipage, bofs, bix, level); | |
789 | if (ret) { | |
790 | logfs_put_read_page(ipage); | |
791 | return ret; | |
792 | } | |
793 | ||
794 | bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level))); | |
795 | logfs_put_read_page(ipage); | |
796 | if (!bofs) | |
797 | return 0; | |
798 | } | |
799 | ||
800 | return 1; | |
801 | } | |
802 | ||
803 | int logfs_exist_block(struct inode *inode, u64 bix) | |
804 | { | |
805 | struct logfs_inode *li = logfs_inode(inode); | |
806 | ||
807 | if (bix < I0_BLOCKS) | |
808 | return !!li->li_data[bix]; | |
809 | return logfs_exist_loop(inode, bix); | |
810 | } | |
811 | ||
812 | static u64 seek_holedata_direct(struct inode *inode, u64 bix, int data) | |
813 | { | |
814 | struct logfs_inode *li = logfs_inode(inode); | |
815 | ||
816 | for (; bix < I0_BLOCKS; bix++) | |
817 | if (data ^ (li->li_data[bix] == 0)) | |
818 | return bix; | |
819 | return I0_BLOCKS; | |
820 | } | |
821 | ||
822 | static u64 seek_holedata_loop(struct inode *inode, u64 bix, int data) | |
823 | { | |
824 | struct logfs_inode *li = logfs_inode(inode); | |
825 | __be64 *rblock; | |
826 | u64 increment, bofs = li->li_data[INDIRECT_INDEX]; | |
827 | level_t level; | |
828 | int ret, slot; | |
829 | struct page *page; | |
830 | ||
831 | BUG_ON(!bofs); | |
832 | ||
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); | |
836 | if (!page) | |
837 | return bix; | |
838 | ||
839 | ret = logfs_segment_read(inode, page, bofs, bix, level); | |
840 | if (ret) { | |
841 | logfs_put_read_page(page); | |
842 | return bix; | |
843 | } | |
844 | ||
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)) | |
849 | break; | |
850 | if (!data && !(be64_to_cpu(rblock[slot]) & LOGFS_FULLY_POPULATED)) | |
851 | break; | |
852 | slot++; | |
853 | bix += increment; | |
854 | bix &= ~(increment - 1); | |
855 | } | |
856 | if (slot >= LOGFS_BLOCK_FACTOR) { | |
857 | kunmap_atomic(rblock, KM_USER0); | |
858 | logfs_put_read_page(page); | |
859 | return bix; | |
860 | } | |
861 | bofs = be64_to_cpu(rblock[slot]); | |
862 | kunmap_atomic(rblock, KM_USER0); | |
863 | logfs_put_read_page(page); | |
864 | if (!bofs) { | |
865 | BUG_ON(data); | |
866 | return bix; | |
867 | } | |
868 | } | |
869 | return bix; | |
870 | } | |
871 | ||
872 | /** | |
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 | |
876 | * | |
877 | * Returns next hole. If the file doesn't contain any further holes, the | |
878 | * block address next to eof is returned instead. | |
879 | */ | |
880 | u64 logfs_seek_hole(struct inode *inode, u64 bix) | |
881 | { | |
882 | struct logfs_inode *li = logfs_inode(inode); | |
883 | ||
884 | if (bix < I0_BLOCKS) { | |
885 | bix = seek_holedata_direct(inode, bix, 0); | |
886 | if (bix < I0_BLOCKS) | |
887 | return bix; | |
888 | } | |
889 | ||
890 | if (!li->li_data[INDIRECT_INDEX]) | |
891 | return bix; | |
892 | else if (li->li_data[INDIRECT_INDEX] & LOGFS_FULLY_POPULATED) | |
893 | bix = maxbix(li->li_height); | |
894 | else { | |
895 | bix = seek_holedata_loop(inode, bix, 0); | |
896 | if (bix < maxbix(li->li_height)) | |
897 | return bix; | |
898 | /* Should not happen anymore. But if some port writes semi- | |
899 | * corrupt images (as this one used to) we might run into it. | |
900 | */ | |
901 | WARN_ON_ONCE(bix == maxbix(li->li_height)); | |
902 | } | |
903 | ||
904 | return bix; | |
905 | } | |
906 | ||
907 | static u64 __logfs_seek_data(struct inode *inode, u64 bix) | |
908 | { | |
909 | struct logfs_inode *li = logfs_inode(inode); | |
910 | ||
911 | if (bix < I0_BLOCKS) { | |
912 | bix = seek_holedata_direct(inode, bix, 1); | |
913 | if (bix < I0_BLOCKS) | |
914 | return bix; | |
915 | } | |
916 | ||
917 | if (bix < maxbix(li->li_height)) { | |
918 | if (!li->li_data[INDIRECT_INDEX]) | |
919 | bix = maxbix(li->li_height); | |
920 | else | |
921 | return seek_holedata_loop(inode, bix, 1); | |
922 | } | |
923 | ||
924 | return bix; | |
925 | } | |
926 | ||
927 | /** | |
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 | |
931 | * | |
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. | |
934 | */ | |
935 | u64 logfs_seek_data(struct inode *inode, u64 bix) | |
936 | { | |
937 | struct super_block *sb = inode->i_sb; | |
938 | u64 ret, end; | |
939 | ||
940 | ret = __logfs_seek_data(inode, bix); | |
941 | end = i_size_read(inode) >> sb->s_blocksize_bits; | |
942 | if (ret >= end) | |
943 | ret = max(bix, end); | |
944 | return ret; | |
945 | } | |
946 | ||
947 | static int logfs_is_valid_direct(struct logfs_inode *li, u64 bix, u64 ofs) | |
948 | { | |
949 | return pure_ofs(li->li_data[bix]) == ofs; | |
950 | } | |
951 | ||
952 | static int __logfs_is_valid_loop(struct inode *inode, u64 bix, | |
953 | u64 ofs, u64 bofs) | |
954 | { | |
955 | struct logfs_inode *li = logfs_inode(inode); | |
956 | level_t level; | |
957 | int ret; | |
958 | struct page *page; | |
959 | ||
960 | for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)){ | |
961 | page = logfs_get_write_page(inode, bix, level); | |
962 | BUG_ON(!page); | |
963 | ||
964 | ret = logfs_segment_read(inode, page, bofs, bix, level); | |
965 | if (ret) { | |
966 | logfs_put_write_page(page); | |
967 | return 0; | |
968 | } | |
969 | ||
970 | bofs = block_get_pointer(page, get_bits(bix, SUBLEVEL(level))); | |
971 | logfs_put_write_page(page); | |
972 | if (!bofs) | |
973 | return 0; | |
974 | ||
975 | if (pure_ofs(bofs) == ofs) | |
976 | return 1; | |
977 | } | |
978 | return 0; | |
979 | } | |
980 | ||
981 | static int logfs_is_valid_loop(struct inode *inode, u64 bix, u64 ofs) | |
982 | { | |
983 | struct logfs_inode *li = logfs_inode(inode); | |
984 | u64 bofs = li->li_data[INDIRECT_INDEX]; | |
985 | ||
986 | if (!bofs) | |
987 | return 0; | |
988 | ||
989 | if (bix >= maxbix(li->li_height)) | |
990 | return 0; | |
991 | ||
992 | if (pure_ofs(bofs) == ofs) | |
993 | return 1; | |
994 | ||
995 | return __logfs_is_valid_loop(inode, bix, ofs, bofs); | |
996 | } | |
997 | ||
998 | static int __logfs_is_valid_block(struct inode *inode, u64 bix, u64 ofs) | |
999 | { | |
1000 | struct logfs_inode *li = logfs_inode(inode); | |
1001 | ||
1002 | if ((inode->i_nlink == 0) && atomic_read(&inode->i_count) == 1) | |
1003 | return 0; | |
1004 | ||
1005 | if (bix < I0_BLOCKS) | |
1006 | return logfs_is_valid_direct(li, bix, ofs); | |
1007 | return logfs_is_valid_loop(inode, bix, ofs); | |
1008 | } | |
1009 | ||
1010 | /** | |
1011 | * logfs_is_valid_block - check whether this block is still valid | |
1012 | * | |
1013 | * @sb - superblock | |
1014 | * @ofs - block physical offset | |
1015 | * @ino - block inode number | |
1016 | * @bix - block index | |
1017 | * @level - block level | |
1018 | * | |
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. | |
1021 | */ | |
1022 | int logfs_is_valid_block(struct super_block *sb, u64 ofs, u64 ino, u64 bix, | |
1023 | gc_level_t gc_level) | |
1024 | { | |
1025 | struct logfs_super *super = logfs_super(sb); | |
1026 | struct inode *inode; | |
1027 | int ret, cookie; | |
1028 | ||
1029 | /* Umount closes a segment with free blocks remaining. Those | |
1030 | * blocks are by definition invalid. */ | |
1031 | if (ino == -1) | |
1032 | return 0; | |
1033 | ||
1034 | LOGFS_BUG_ON((u64)(u_long)ino != ino, sb); | |
1035 | ||
1036 | inode = logfs_safe_iget(sb, ino, &cookie); | |
1037 | if (IS_ERR(inode)) | |
1038 | goto invalid; | |
1039 | ||
1040 | ret = __logfs_is_valid_block(inode, bix, ofs); | |
1041 | logfs_safe_iput(inode, cookie); | |
1042 | if (ret) | |
1043 | return ret; | |
1044 | ||
1045 | invalid: | |
1046 | /* Block is nominally invalid, but may still sit in the shadow tree, | |
1047 | * waiting for a journal commit. | |
1048 | */ | |
1049 | if (btree_lookup64(&super->s_shadow_tree.old, ofs)) | |
1050 | return 2; | |
1051 | return 0; | |
1052 | } | |
1053 | ||
1054 | int logfs_readpage_nolock(struct page *page) | |
1055 | { | |
1056 | struct inode *inode = page->mapping->host; | |
1057 | int ret = -EIO; | |
1058 | ||
1059 | ret = logfs_read_block(inode, page, READ); | |
1060 | ||
1061 | if (ret) { | |
1062 | ClearPageUptodate(page); | |
1063 | SetPageError(page); | |
1064 | } else { | |
1065 | SetPageUptodate(page); | |
1066 | ClearPageError(page); | |
1067 | } | |
1068 | flush_dcache_page(page); | |
1069 | ||
1070 | return ret; | |
1071 | } | |
1072 | ||
1073 | static int logfs_reserve_bytes(struct inode *inode, int bytes) | |
1074 | { | |
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; | |
1078 | ||
1079 | if (!bytes) | |
1080 | return 0; | |
1081 | ||
1082 | if (available < bytes) | |
1083 | return -ENOSPC; | |
1084 | ||
1085 | if (available < bytes + super->s_root_reserve && | |
1086 | !capable(CAP_SYS_RESOURCE)) | |
1087 | return -ENOSPC; | |
1088 | ||
1089 | return 0; | |
1090 | } | |
1091 | ||
1092 | int get_page_reserve(struct inode *inode, struct page *page) | |
1093 | { | |
1094 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1095 | int ret; | |
1096 | ||
1097 | if (logfs_block(page) && logfs_block(page)->reserved_bytes) | |
1098 | return 0; | |
1099 | ||
1100 | logfs_get_wblocks(inode->i_sb, page, WF_LOCK); | |
1101 | ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE); | |
1102 | if (!ret) { | |
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; | |
1106 | } | |
1107 | logfs_put_wblocks(inode->i_sb, page, WF_LOCK); | |
1108 | return ret; | |
1109 | } | |
1110 | ||
1111 | /* | |
1112 | * We are protected by write lock. Push victims up to superblock level | |
1113 | * and release transaction when appropriate. | |
1114 | */ | |
1115 | /* FIXME: This is currently called from the wrong spots. */ | |
1116 | static void logfs_handle_transaction(struct inode *inode, | |
1117 | struct logfs_transaction *ta) | |
1118 | { | |
1119 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1120 | ||
1121 | if (!ta) | |
1122 | return; | |
1123 | logfs_inode(inode)->li_block->ta = NULL; | |
1124 | ||
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; | |
1130 | return; | |
1131 | } | |
1132 | ||
1133 | switch (ta->state) { | |
1134 | case CREATE_1: /* fall through */ | |
1135 | case UNLINK_1: | |
1136 | BUG_ON(super->s_victim_ino); | |
1137 | super->s_victim_ino = ta->ino; | |
1138 | break; | |
1139 | case CREATE_2: /* fall through */ | |
1140 | case UNLINK_2: | |
1141 | BUG_ON(super->s_victim_ino != ta->ino); | |
1142 | super->s_victim_ino = 0; | |
1143 | /* transaction ends here - free it */ | |
1144 | kfree(ta); | |
1145 | break; | |
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; | |
1151 | break; | |
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; | |
1157 | kfree(ta); | |
1158 | break; | |
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; | |
1166 | break; | |
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; | |
1173 | break; | |
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; | |
1179 | kfree(ta); | |
1180 | break; | |
1181 | default: | |
1182 | BUG(); | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * Not strictly a reservation, but rather a check that we still have enough | |
1188 | * space to satisfy the write. | |
1189 | */ | |
1190 | static int logfs_reserve_blocks(struct inode *inode, int blocks) | |
1191 | { | |
1192 | return logfs_reserve_bytes(inode, blocks * LOGFS_MAX_OBJECTSIZE); | |
1193 | } | |
1194 | ||
1195 | struct write_control { | |
1196 | u64 ofs; | |
1197 | long flags; | |
1198 | }; | |
1199 | ||
1200 | static struct logfs_shadow *alloc_shadow(struct inode *inode, u64 bix, | |
1201 | level_t level, u64 old_ofs) | |
1202 | { | |
1203 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1204 | struct logfs_shadow *shadow; | |
1205 | ||
1206 | shadow = mempool_alloc(super->s_shadow_pool, GFP_NOFS); | |
1207 | memset(shadow, 0, sizeof(*shadow)); | |
1208 | shadow->ino = inode->i_ino; | |
1209 | shadow->bix = bix; | |
1210 | shadow->gc_level = expand_level(inode->i_ino, level); | |
1211 | shadow->old_ofs = old_ofs & ~LOGFS_FULLY_POPULATED; | |
1212 | return shadow; | |
1213 | } | |
1214 | ||
1215 | static void free_shadow(struct inode *inode, struct logfs_shadow *shadow) | |
1216 | { | |
1217 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1218 | ||
1219 | mempool_free(shadow, super->s_shadow_pool); | |
1220 | } | |
1221 | ||
1222 | /** | |
1223 | * fill_shadow_tree - Propagate shadow tree changes due to a write | |
1224 | * @inode: Inode owning the page | |
1225 | * @page: Struct page that was written | |
1226 | * @shadow: Shadow for the current write | |
1227 | * | |
1228 | * Writes in logfs can result in two semi-valid objects. The old object | |
1229 | * is still valid as long as it can be reached by following pointers on | |
1230 | * the medium. Only when writes propagate all the way up to the journal | |
1231 | * has the new object safely replaced the old one. | |
1232 | * | |
1233 | * To handle this problem, a struct logfs_shadow is used to represent | |
1234 | * every single write. It is attached to the indirect block, which is | |
1235 | * marked dirty. When the indirect block is written, its shadows are | |
1236 | * handed up to the next indirect block (or inode). Untimately they | |
1237 | * will reach the master inode and be freed upon journal commit. | |
1238 | * | |
1239 | * This function handles a single step in the propagation. It adds the | |
1240 | * shadow for the current write to the tree, along with any shadows in | |
1241 | * the page's tree, in case it was an indirect block. If a page is | |
1242 | * written, the inode parameter is left NULL, if an inode is written, | |
1243 | * the page parameter is left NULL. | |
1244 | */ | |
1245 | static void fill_shadow_tree(struct inode *inode, struct page *page, | |
1246 | struct logfs_shadow *shadow) | |
1247 | { | |
1248 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1249 | struct logfs_block *block = logfs_block(page); | |
1250 | struct shadow_tree *tree = &super->s_shadow_tree; | |
1251 | ||
1252 | if (PagePrivate(page)) { | |
1253 | if (block->alias_map) | |
1254 | super->s_no_object_aliases -= bitmap_weight( | |
1255 | block->alias_map, LOGFS_BLOCK_FACTOR); | |
1256 | logfs_handle_transaction(inode, block->ta); | |
1257 | block->ops->free_block(inode->i_sb, block); | |
1258 | } | |
1259 | if (shadow) { | |
1260 | if (shadow->old_ofs) | |
1261 | btree_insert64(&tree->old, shadow->old_ofs, shadow, | |
1262 | GFP_NOFS); | |
1263 | else | |
1264 | btree_insert64(&tree->new, shadow->new_ofs, shadow, | |
1265 | GFP_NOFS); | |
1266 | ||
1267 | super->s_dirty_used_bytes += shadow->new_len; | |
1268 | super->s_dirty_free_bytes += shadow->old_len; | |
1269 | } | |
1270 | } | |
1271 | ||
1272 | static void logfs_set_alias(struct super_block *sb, struct logfs_block *block, | |
1273 | long child_no) | |
1274 | { | |
1275 | struct logfs_super *super = logfs_super(sb); | |
1276 | ||
1277 | if (block->inode && block->inode->i_ino == LOGFS_INO_MASTER) { | |
1278 | /* Aliases in the master inode are pointless. */ | |
1279 | return; | |
1280 | } | |
1281 | ||
1282 | if (!test_bit(child_no, block->alias_map)) { | |
1283 | set_bit(child_no, block->alias_map); | |
1284 | super->s_no_object_aliases++; | |
1285 | } | |
1286 | list_move_tail(&block->alias_list, &super->s_object_alias); | |
1287 | } | |
1288 | ||
1289 | /* | |
1290 | * Object aliases can and often do change the size and occupied space of a | |
1291 | * file. So not only do we have to change the pointers, we also have to | |
1292 | * change inode->i_size and li->li_used_bytes. Which is done by setting | |
1293 | * another two object aliases for the inode itself. | |
1294 | */ | |
1295 | static void set_iused(struct inode *inode, struct logfs_shadow *shadow) | |
1296 | { | |
1297 | struct logfs_inode *li = logfs_inode(inode); | |
1298 | ||
1299 | if (shadow->new_len == shadow->old_len) | |
1300 | return; | |
1301 | ||
1302 | alloc_inode_block(inode); | |
1303 | li->li_used_bytes += shadow->new_len - shadow->old_len; | |
1304 | __logfs_set_blocks(inode); | |
1305 | logfs_set_alias(inode->i_sb, li->li_block, INODE_USED_OFS); | |
1306 | logfs_set_alias(inode->i_sb, li->li_block, INODE_SIZE_OFS); | |
1307 | } | |
1308 | ||
1309 | static int logfs_write_i0(struct inode *inode, struct page *page, | |
1310 | struct write_control *wc) | |
1311 | { | |
1312 | struct logfs_shadow *shadow; | |
1313 | u64 bix; | |
1314 | level_t level; | |
1315 | int full, err = 0; | |
1316 | ||
1317 | logfs_unpack_index(page->index, &bix, &level); | |
1318 | if (wc->ofs == 0) | |
1319 | if (logfs_reserve_blocks(inode, 1)) | |
1320 | return -ENOSPC; | |
1321 | ||
1322 | shadow = alloc_shadow(inode, bix, level, wc->ofs); | |
1323 | if (wc->flags & WF_WRITE) | |
1324 | err = logfs_segment_write(inode, page, shadow); | |
1325 | if (wc->flags & WF_DELETE) | |
1326 | logfs_segment_delete(inode, shadow); | |
1327 | if (err) { | |
1328 | free_shadow(inode, shadow); | |
1329 | return err; | |
1330 | } | |
1331 | ||
1332 | set_iused(inode, shadow); | |
1333 | full = 1; | |
1334 | if (level != 0) { | |
1335 | alloc_indirect_block(inode, page, 0); | |
1336 | full = logfs_block(page)->full == LOGFS_BLOCK_FACTOR; | |
1337 | } | |
1338 | fill_shadow_tree(inode, page, shadow); | |
1339 | wc->ofs = shadow->new_ofs; | |
1340 | if (wc->ofs && full) | |
1341 | wc->ofs |= LOGFS_FULLY_POPULATED; | |
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | static int logfs_write_direct(struct inode *inode, struct page *page, | |
1346 | long flags) | |
1347 | { | |
1348 | struct logfs_inode *li = logfs_inode(inode); | |
1349 | struct write_control wc = { | |
1350 | .ofs = li->li_data[page->index], | |
1351 | .flags = flags, | |
1352 | }; | |
1353 | int err; | |
1354 | ||
1355 | alloc_inode_block(inode); | |
1356 | ||
1357 | err = logfs_write_i0(inode, page, &wc); | |
1358 | if (err) | |
1359 | return err; | |
1360 | ||
1361 | li->li_data[page->index] = wc.ofs; | |
1362 | logfs_set_alias(inode->i_sb, li->li_block, | |
1363 | page->index + INODE_POINTER_OFS); | |
1364 | return 0; | |
1365 | } | |
1366 | ||
1367 | static int ptr_change(u64 ofs, struct page *page) | |
1368 | { | |
1369 | struct logfs_block *block = logfs_block(page); | |
1370 | int empty0, empty1, full0, full1; | |
1371 | ||
1372 | empty0 = ofs == 0; | |
1373 | empty1 = block->partial == 0; | |
1374 | if (empty0 != empty1) | |
1375 | return 1; | |
1376 | ||
1377 | /* The !! is necessary to shrink result to int */ | |
1378 | full0 = !!(ofs & LOGFS_FULLY_POPULATED); | |
1379 | full1 = block->full == LOGFS_BLOCK_FACTOR; | |
1380 | if (full0 != full1) | |
1381 | return 1; | |
1382 | return 0; | |
1383 | } | |
1384 | ||
1385 | static int __logfs_write_rec(struct inode *inode, struct page *page, | |
1386 | struct write_control *this_wc, | |
1387 | pgoff_t bix, level_t target_level, level_t level) | |
1388 | { | |
1389 | int ret, page_empty = 0; | |
1390 | int child_no = get_bits(bix, SUBLEVEL(level)); | |
1391 | struct page *ipage; | |
1392 | struct write_control child_wc = { | |
1393 | .flags = this_wc->flags, | |
1394 | }; | |
1395 | ||
1396 | ipage = logfs_get_write_page(inode, bix, level); | |
1397 | if (!ipage) | |
1398 | return -ENOMEM; | |
1399 | ||
1400 | if (this_wc->ofs) { | |
1401 | ret = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level); | |
1402 | if (ret) | |
1403 | goto out; | |
1404 | } else if (!PageUptodate(ipage)) { | |
1405 | page_empty = 1; | |
1406 | logfs_read_empty(ipage); | |
1407 | } | |
1408 | ||
1409 | child_wc.ofs = block_get_pointer(ipage, child_no); | |
1410 | ||
1411 | if ((__force u8)level-1 > (__force u8)target_level) | |
1412 | ret = __logfs_write_rec(inode, page, &child_wc, bix, | |
1413 | target_level, SUBLEVEL(level)); | |
1414 | else | |
1415 | ret = logfs_write_i0(inode, page, &child_wc); | |
1416 | ||
1417 | if (ret) | |
1418 | goto out; | |
1419 | ||
1420 | alloc_indirect_block(inode, ipage, page_empty); | |
1421 | block_set_pointer(ipage, child_no, child_wc.ofs); | |
1422 | /* FIXME: first condition seems superfluous */ | |
1423 | if (child_wc.ofs || logfs_block(ipage)->partial) | |
1424 | this_wc->flags |= WF_WRITE; | |
1425 | /* the condition on this_wc->ofs ensures that we won't consume extra | |
1426 | * space for indirect blocks in the future, which we cannot reserve */ | |
1427 | if (!this_wc->ofs || ptr_change(this_wc->ofs, ipage)) | |
1428 | ret = logfs_write_i0(inode, ipage, this_wc); | |
1429 | else | |
1430 | logfs_set_alias(inode->i_sb, logfs_block(ipage), child_no); | |
1431 | out: | |
1432 | logfs_put_write_page(ipage); | |
1433 | return ret; | |
1434 | } | |
1435 | ||
1436 | static int logfs_write_rec(struct inode *inode, struct page *page, | |
1437 | pgoff_t bix, level_t target_level, long flags) | |
1438 | { | |
1439 | struct logfs_inode *li = logfs_inode(inode); | |
1440 | struct write_control wc = { | |
1441 | .ofs = li->li_data[INDIRECT_INDEX], | |
1442 | .flags = flags, | |
1443 | }; | |
1444 | int ret; | |
1445 | ||
1446 | alloc_inode_block(inode); | |
1447 | ||
1448 | if (li->li_height > (__force u8)target_level) | |
1449 | ret = __logfs_write_rec(inode, page, &wc, bix, target_level, | |
1450 | LEVEL(li->li_height)); | |
1451 | else | |
1452 | ret = logfs_write_i0(inode, page, &wc); | |
1453 | if (ret) | |
1454 | return ret; | |
1455 | ||
1456 | if (li->li_data[INDIRECT_INDEX] != wc.ofs) { | |
1457 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1458 | logfs_set_alias(inode->i_sb, li->li_block, | |
1459 | INDIRECT_INDEX + INODE_POINTER_OFS); | |
1460 | } | |
1461 | return ret; | |
1462 | } | |
1463 | ||
1464 | void logfs_add_transaction(struct inode *inode, struct logfs_transaction *ta) | |
1465 | { | |
1466 | alloc_inode_block(inode); | |
1467 | logfs_inode(inode)->li_block->ta = ta; | |
1468 | } | |
1469 | ||
1470 | void logfs_del_transaction(struct inode *inode, struct logfs_transaction *ta) | |
1471 | { | |
1472 | struct logfs_block *block = logfs_inode(inode)->li_block; | |
1473 | ||
1474 | if (block && block->ta) | |
1475 | block->ta = NULL; | |
1476 | } | |
1477 | ||
1478 | static int grow_inode(struct inode *inode, u64 bix, level_t level) | |
1479 | { | |
1480 | struct logfs_inode *li = logfs_inode(inode); | |
1481 | u8 height = (__force u8)level; | |
1482 | struct page *page; | |
1483 | struct write_control wc = { | |
1484 | .flags = WF_WRITE, | |
1485 | }; | |
1486 | int err; | |
1487 | ||
1488 | BUG_ON(height > 5 || li->li_height > 5); | |
1489 | while (height > li->li_height || bix >= maxbix(li->li_height)) { | |
1490 | page = logfs_get_write_page(inode, I0_BLOCKS + 1, | |
1491 | LEVEL(li->li_height + 1)); | |
1492 | if (!page) | |
1493 | return -ENOMEM; | |
1494 | logfs_read_empty(page); | |
1495 | alloc_indirect_block(inode, page, 1); | |
1496 | block_set_pointer(page, 0, li->li_data[INDIRECT_INDEX]); | |
1497 | err = logfs_write_i0(inode, page, &wc); | |
1498 | logfs_put_write_page(page); | |
1499 | if (err) | |
1500 | return err; | |
1501 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1502 | wc.ofs = 0; | |
1503 | li->li_height++; | |
1504 | logfs_set_alias(inode->i_sb, li->li_block, INODE_HEIGHT_OFS); | |
1505 | } | |
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | static int __logfs_write_buf(struct inode *inode, struct page *page, long flags) | |
1510 | { | |
1511 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1512 | pgoff_t index = page->index; | |
1513 | u64 bix; | |
1514 | level_t level; | |
1515 | int err; | |
1516 | ||
1517 | flags |= WF_WRITE | WF_DELETE; | |
1518 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
1519 | ||
1520 | logfs_unpack_index(index, &bix, &level); | |
1521 | if (logfs_block(page) && logfs_block(page)->reserved_bytes) | |
1522 | super->s_dirty_pages -= logfs_block(page)->reserved_bytes; | |
1523 | ||
1524 | if (index < I0_BLOCKS) | |
1525 | return logfs_write_direct(inode, page, flags); | |
1526 | ||
1527 | bix = adjust_bix(bix, level); | |
1528 | err = grow_inode(inode, bix, level); | |
1529 | if (err) | |
1530 | return err; | |
1531 | return logfs_write_rec(inode, page, bix, level, flags); | |
1532 | } | |
1533 | ||
1534 | int logfs_write_buf(struct inode *inode, struct page *page, long flags) | |
1535 | { | |
1536 | struct super_block *sb = inode->i_sb; | |
1537 | int ret; | |
1538 | ||
1539 | logfs_get_wblocks(sb, page, flags & WF_LOCK); | |
1540 | ret = __logfs_write_buf(inode, page, flags); | |
1541 | logfs_put_wblocks(sb, page, flags & WF_LOCK); | |
1542 | return ret; | |
1543 | } | |
1544 | ||
1545 | static int __logfs_delete(struct inode *inode, struct page *page) | |
1546 | { | |
1547 | long flags = WF_DELETE; | |
1548 | ||
1549 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
1550 | ||
1551 | if (page->index < I0_BLOCKS) | |
1552 | return logfs_write_direct(inode, page, flags); | |
1553 | return logfs_write_rec(inode, page, page->index, 0, flags); | |
1554 | } | |
1555 | ||
1556 | int logfs_delete(struct inode *inode, pgoff_t index, | |
1557 | struct shadow_tree *shadow_tree) | |
1558 | { | |
1559 | struct super_block *sb = inode->i_sb; | |
1560 | struct page *page; | |
1561 | int ret; | |
1562 | ||
1563 | page = logfs_get_read_page(inode, index, 0); | |
1564 | if (!page) | |
1565 | return -ENOMEM; | |
1566 | ||
1567 | logfs_get_wblocks(sb, page, 1); | |
1568 | ret = __logfs_delete(inode, page); | |
1569 | logfs_put_wblocks(sb, page, 1); | |
1570 | ||
1571 | logfs_put_read_page(page); | |
1572 | ||
1573 | return ret; | |
1574 | } | |
1575 | ||
5db53f3e JE |
1576 | int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs, |
1577 | gc_level_t gc_level, long flags) | |
1578 | { | |
1579 | level_t level = shrink_level(gc_level); | |
1580 | struct page *page; | |
1581 | int err; | |
1582 | ||
1583 | page = logfs_get_write_page(inode, bix, level); | |
1584 | if (!page) | |
1585 | return -ENOMEM; | |
1586 | ||
1587 | err = logfs_segment_read(inode, page, ofs, bix, level); | |
1588 | if (!err) { | |
1589 | if (level != 0) | |
1590 | alloc_indirect_block(inode, page, 0); | |
1591 | err = logfs_write_buf(inode, page, flags); | |
19321917 JE |
1592 | if (!err && shrink_level(gc_level) == 0) { |
1593 | /* Rewrite cannot mark the inode dirty but has to | |
1594 | * write it immediatly. | |
1595 | * Q: Can't we just create an alias for the inode | |
1596 | * instead? And if not, why not? | |
1597 | */ | |
1598 | if (inode->i_ino == LOGFS_INO_MASTER) | |
1599 | logfs_write_anchor(inode->i_sb); | |
1600 | else { | |
1601 | err = __logfs_write_inode(inode, flags); | |
1602 | } | |
1603 | } | |
5db53f3e JE |
1604 | } |
1605 | logfs_put_write_page(page); | |
1606 | return err; | |
1607 | } | |
1608 | ||
1609 | static int truncate_data_block(struct inode *inode, struct page *page, | |
1610 | u64 ofs, struct logfs_shadow *shadow, u64 size) | |
1611 | { | |
1612 | loff_t pageofs = page->index << inode->i_sb->s_blocksize_bits; | |
1613 | u64 bix; | |
1614 | level_t level; | |
1615 | int err; | |
1616 | ||
1617 | /* Does truncation happen within this page? */ | |
1618 | if (size <= pageofs || size - pageofs >= PAGE_SIZE) | |
1619 | return 0; | |
1620 | ||
1621 | logfs_unpack_index(page->index, &bix, &level); | |
1622 | BUG_ON(level != 0); | |
1623 | ||
1624 | err = logfs_segment_read(inode, page, ofs, bix, level); | |
1625 | if (err) | |
1626 | return err; | |
1627 | ||
1628 | zero_user_segment(page, size - pageofs, PAGE_CACHE_SIZE); | |
1629 | return logfs_segment_write(inode, page, shadow); | |
1630 | } | |
1631 | ||
1632 | static int logfs_truncate_i0(struct inode *inode, struct page *page, | |
1633 | struct write_control *wc, u64 size) | |
1634 | { | |
1635 | struct logfs_shadow *shadow; | |
1636 | u64 bix; | |
1637 | level_t level; | |
1638 | int err = 0; | |
1639 | ||
1640 | logfs_unpack_index(page->index, &bix, &level); | |
1641 | BUG_ON(level != 0); | |
1642 | shadow = alloc_shadow(inode, bix, level, wc->ofs); | |
1643 | ||
1644 | err = truncate_data_block(inode, page, wc->ofs, shadow, size); | |
1645 | if (err) { | |
1646 | free_shadow(inode, shadow); | |
1647 | return err; | |
1648 | } | |
1649 | ||
1650 | logfs_segment_delete(inode, shadow); | |
1651 | set_iused(inode, shadow); | |
1652 | fill_shadow_tree(inode, page, shadow); | |
1653 | wc->ofs = shadow->new_ofs; | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | static int logfs_truncate_direct(struct inode *inode, u64 size) | |
1658 | { | |
1659 | struct logfs_inode *li = logfs_inode(inode); | |
1660 | struct write_control wc; | |
1661 | struct page *page; | |
1662 | int e; | |
1663 | int err; | |
1664 | ||
1665 | alloc_inode_block(inode); | |
1666 | ||
1667 | for (e = I0_BLOCKS - 1; e >= 0; e--) { | |
1668 | if (size > (e+1) * LOGFS_BLOCKSIZE) | |
1669 | break; | |
1670 | ||
1671 | wc.ofs = li->li_data[e]; | |
1672 | if (!wc.ofs) | |
1673 | continue; | |
1674 | ||
1675 | page = logfs_get_write_page(inode, e, 0); | |
1676 | if (!page) | |
1677 | return -ENOMEM; | |
1678 | err = logfs_segment_read(inode, page, wc.ofs, e, 0); | |
1679 | if (err) { | |
1680 | logfs_put_write_page(page); | |
1681 | return err; | |
1682 | } | |
1683 | err = logfs_truncate_i0(inode, page, &wc, size); | |
1684 | logfs_put_write_page(page); | |
1685 | if (err) | |
1686 | return err; | |
1687 | ||
1688 | li->li_data[e] = wc.ofs; | |
1689 | } | |
1690 | return 0; | |
1691 | } | |
1692 | ||
1693 | /* FIXME: these need to become per-sb once we support different blocksizes */ | |
1694 | static u64 __logfs_step[] = { | |
1695 | 1, | |
1696 | I1_BLOCKS, | |
1697 | I2_BLOCKS, | |
1698 | I3_BLOCKS, | |
1699 | }; | |
1700 | ||
1701 | static u64 __logfs_start_index[] = { | |
1702 | I0_BLOCKS, | |
1703 | I1_BLOCKS, | |
1704 | I2_BLOCKS, | |
1705 | I3_BLOCKS | |
1706 | }; | |
1707 | ||
1708 | static inline u64 logfs_step(level_t level) | |
1709 | { | |
1710 | return __logfs_step[(__force u8)level]; | |
1711 | } | |
1712 | ||
1713 | static inline u64 logfs_factor(u8 level) | |
1714 | { | |
1715 | return __logfs_step[level] * LOGFS_BLOCKSIZE; | |
1716 | } | |
1717 | ||
1718 | static inline u64 logfs_start_index(level_t level) | |
1719 | { | |
1720 | return __logfs_start_index[(__force u8)level]; | |
1721 | } | |
1722 | ||
1723 | static void logfs_unpack_raw_index(pgoff_t index, u64 *bix, level_t *level) | |
1724 | { | |
1725 | logfs_unpack_index(index, bix, level); | |
1726 | if (*bix <= logfs_start_index(SUBLEVEL(*level))) | |
1727 | *bix = 0; | |
1728 | } | |
1729 | ||
1730 | static int __logfs_truncate_rec(struct inode *inode, struct page *ipage, | |
1731 | struct write_control *this_wc, u64 size) | |
1732 | { | |
1733 | int truncate_happened = 0; | |
1734 | int e, err = 0; | |
1735 | u64 bix, child_bix, next_bix; | |
1736 | level_t level; | |
1737 | struct page *page; | |
1738 | struct write_control child_wc = { /* FIXME: flags */ }; | |
1739 | ||
1740 | logfs_unpack_raw_index(ipage->index, &bix, &level); | |
1741 | err = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level); | |
1742 | if (err) | |
1743 | return err; | |
1744 | ||
1745 | for (e = LOGFS_BLOCK_FACTOR - 1; e >= 0; e--) { | |
1746 | child_bix = bix + e * logfs_step(SUBLEVEL(level)); | |
1747 | next_bix = child_bix + logfs_step(SUBLEVEL(level)); | |
1748 | if (size > next_bix * LOGFS_BLOCKSIZE) | |
1749 | break; | |
1750 | ||
1751 | child_wc.ofs = pure_ofs(block_get_pointer(ipage, e)); | |
1752 | if (!child_wc.ofs) | |
1753 | continue; | |
1754 | ||
1755 | page = logfs_get_write_page(inode, child_bix, SUBLEVEL(level)); | |
1756 | if (!page) | |
1757 | return -ENOMEM; | |
1758 | ||
1759 | if ((__force u8)level > 1) | |
1760 | err = __logfs_truncate_rec(inode, page, &child_wc, size); | |
1761 | else | |
1762 | err = logfs_truncate_i0(inode, page, &child_wc, size); | |
1763 | logfs_put_write_page(page); | |
1764 | if (err) | |
1765 | return err; | |
1766 | ||
1767 | truncate_happened = 1; | |
1768 | alloc_indirect_block(inode, ipage, 0); | |
1769 | block_set_pointer(ipage, e, child_wc.ofs); | |
1770 | } | |
1771 | ||
1772 | if (!truncate_happened) { | |
1773 | printk("ineffectual truncate (%lx, %lx, %llx)\n", inode->i_ino, ipage->index, size); | |
1774 | return 0; | |
1775 | } | |
1776 | ||
1777 | this_wc->flags = WF_DELETE; | |
1778 | if (logfs_block(ipage)->partial) | |
1779 | this_wc->flags |= WF_WRITE; | |
1780 | ||
1781 | return logfs_write_i0(inode, ipage, this_wc); | |
1782 | } | |
1783 | ||
1784 | static int logfs_truncate_rec(struct inode *inode, u64 size) | |
1785 | { | |
1786 | struct logfs_inode *li = logfs_inode(inode); | |
1787 | struct write_control wc = { | |
1788 | .ofs = li->li_data[INDIRECT_INDEX], | |
1789 | }; | |
1790 | struct page *page; | |
1791 | int err; | |
1792 | ||
1793 | alloc_inode_block(inode); | |
1794 | ||
1795 | if (!wc.ofs) | |
1796 | return 0; | |
1797 | ||
1798 | page = logfs_get_write_page(inode, 0, LEVEL(li->li_height)); | |
1799 | if (!page) | |
1800 | return -ENOMEM; | |
1801 | ||
1802 | err = __logfs_truncate_rec(inode, page, &wc, size); | |
1803 | logfs_put_write_page(page); | |
1804 | if (err) | |
1805 | return err; | |
1806 | ||
1807 | if (li->li_data[INDIRECT_INDEX] != wc.ofs) | |
1808 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1809 | return 0; | |
1810 | } | |
1811 | ||
1812 | static int __logfs_truncate(struct inode *inode, u64 size) | |
1813 | { | |
1814 | int ret; | |
1815 | ||
1816 | if (size >= logfs_factor(logfs_inode(inode)->li_height)) | |
1817 | return 0; | |
1818 | ||
1819 | ret = logfs_truncate_rec(inode, size); | |
1820 | if (ret) | |
1821 | return ret; | |
1822 | ||
1823 | return logfs_truncate_direct(inode, size); | |
1824 | } | |
1825 | ||
1826 | int logfs_truncate(struct inode *inode, u64 size) | |
1827 | { | |
1828 | struct super_block *sb = inode->i_sb; | |
1829 | int err; | |
1830 | ||
1831 | logfs_get_wblocks(sb, NULL, 1); | |
1832 | err = __logfs_truncate(inode, size); | |
1833 | if (!err) | |
1834 | err = __logfs_write_inode(inode, 0); | |
1835 | logfs_put_wblocks(sb, NULL, 1); | |
1836 | ||
1837 | if (!err) | |
1838 | err = vmtruncate(inode, size); | |
1839 | ||
1840 | /* I don't trust error recovery yet. */ | |
1841 | WARN_ON(err); | |
1842 | return err; | |
1843 | } | |
1844 | ||
1845 | static void move_page_to_inode(struct inode *inode, struct page *page) | |
1846 | { | |
1847 | struct logfs_inode *li = logfs_inode(inode); | |
1848 | struct logfs_block *block = logfs_block(page); | |
1849 | ||
1850 | if (!block) | |
1851 | return; | |
1852 | ||
1853 | log_blockmove("move_page_to_inode(%llx, %llx, %x)\n", | |
1854 | block->ino, block->bix, block->level); | |
1855 | BUG_ON(li->li_block); | |
1856 | block->ops = &inode_block_ops; | |
1857 | block->inode = inode; | |
1858 | li->li_block = block; | |
1859 | ||
1860 | block->page = NULL; | |
1861 | page->private = 0; | |
1862 | ClearPagePrivate(page); | |
1863 | } | |
1864 | ||
1865 | static void move_inode_to_page(struct page *page, struct inode *inode) | |
1866 | { | |
1867 | struct logfs_inode *li = logfs_inode(inode); | |
1868 | struct logfs_block *block = li->li_block; | |
1869 | ||
1870 | if (!block) | |
1871 | return; | |
1872 | ||
1873 | log_blockmove("move_inode_to_page(%llx, %llx, %x)\n", | |
1874 | block->ino, block->bix, block->level); | |
1875 | BUG_ON(PagePrivate(page)); | |
1876 | block->ops = &indirect_block_ops; | |
1877 | block->page = page; | |
1878 | page->private = (unsigned long)block; | |
1879 | SetPagePrivate(page); | |
1880 | ||
1881 | block->inode = NULL; | |
1882 | li->li_block = NULL; | |
1883 | } | |
1884 | ||
1885 | int logfs_read_inode(struct inode *inode) | |
1886 | { | |
1887 | struct super_block *sb = inode->i_sb; | |
1888 | struct logfs_super *super = logfs_super(sb); | |
1889 | struct inode *master_inode = super->s_master_inode; | |
1890 | struct page *page; | |
1891 | struct logfs_disk_inode *di; | |
1892 | u64 ino = inode->i_ino; | |
1893 | ||
1894 | if (ino << sb->s_blocksize_bits > i_size_read(master_inode)) | |
1895 | return -ENODATA; | |
1896 | if (!logfs_exist_block(master_inode, ino)) | |
1897 | return -ENODATA; | |
1898 | ||
1899 | page = read_cache_page(master_inode->i_mapping, ino, | |
1900 | (filler_t *)logfs_readpage, NULL); | |
1901 | if (IS_ERR(page)) | |
1902 | return PTR_ERR(page); | |
1903 | ||
1904 | di = kmap_atomic(page, KM_USER0); | |
1905 | logfs_disk_to_inode(di, inode); | |
1906 | kunmap_atomic(di, KM_USER0); | |
1907 | move_page_to_inode(inode, page); | |
1908 | page_cache_release(page); | |
1909 | return 0; | |
1910 | } | |
1911 | ||
1912 | /* Caller must logfs_put_write_page(page); */ | |
1913 | static struct page *inode_to_page(struct inode *inode) | |
1914 | { | |
1915 | struct inode *master_inode = logfs_super(inode->i_sb)->s_master_inode; | |
1916 | struct logfs_disk_inode *di; | |
1917 | struct page *page; | |
1918 | ||
1919 | BUG_ON(inode->i_ino == LOGFS_INO_MASTER); | |
1920 | ||
1921 | page = logfs_get_write_page(master_inode, inode->i_ino, 0); | |
1922 | if (!page) | |
1923 | return NULL; | |
1924 | ||
1925 | di = kmap_atomic(page, KM_USER0); | |
1926 | logfs_inode_to_disk(inode, di); | |
1927 | kunmap_atomic(di, KM_USER0); | |
1928 | move_inode_to_page(page, inode); | |
1929 | return page; | |
1930 | } | |
1931 | ||
1932 | /* Cheaper version of write_inode. All changes are concealed in | |
1933 | * aliases, which are moved back. No write to the medium happens. | |
1934 | */ | |
1935 | void logfs_clear_inode(struct inode *inode) | |
1936 | { | |
1937 | struct super_block *sb = inode->i_sb; | |
1938 | struct logfs_inode *li = logfs_inode(inode); | |
1939 | struct logfs_block *block = li->li_block; | |
1940 | struct page *page; | |
1941 | ||
1942 | /* Only deleted files may be dirty at this point */ | |
1943 | BUG_ON(inode->i_state & I_DIRTY && inode->i_nlink); | |
1944 | if (!block) | |
1945 | return; | |
1946 | if ((logfs_super(sb)->s_flags & LOGFS_SB_FLAG_SHUTDOWN)) { | |
1947 | block->ops->free_block(inode->i_sb, block); | |
1948 | return; | |
1949 | } | |
1950 | ||
1951 | BUG_ON(inode->i_ino < LOGFS_RESERVED_INOS); | |
1952 | page = inode_to_page(inode); | |
1953 | BUG_ON(!page); /* FIXME: Use emergency page */ | |
1954 | logfs_put_write_page(page); | |
1955 | } | |
1956 | ||
1957 | static int do_write_inode(struct inode *inode) | |
1958 | { | |
1959 | struct super_block *sb = inode->i_sb; | |
1960 | struct inode *master_inode = logfs_super(sb)->s_master_inode; | |
1961 | loff_t size = (inode->i_ino + 1) << inode->i_sb->s_blocksize_bits; | |
1962 | struct page *page; | |
1963 | int err; | |
1964 | ||
1965 | BUG_ON(inode->i_ino == LOGFS_INO_MASTER); | |
1966 | /* FIXME: lock inode */ | |
1967 | ||
1968 | if (i_size_read(master_inode) < size) | |
1969 | i_size_write(master_inode, size); | |
1970 | ||
1971 | /* TODO: Tell vfs this inode is clean now */ | |
1972 | ||
1973 | page = inode_to_page(inode); | |
1974 | if (!page) | |
1975 | return -ENOMEM; | |
1976 | ||
1977 | /* FIXME: transaction is part of logfs_block now. Is that enough? */ | |
1978 | err = logfs_write_buf(master_inode, page, 0); | |
1979 | logfs_put_write_page(page); | |
1980 | return err; | |
1981 | } | |
1982 | ||
1983 | static void logfs_mod_segment_entry(struct super_block *sb, u32 segno, | |
1984 | int write, | |
1985 | void (*change_se)(struct logfs_segment_entry *, long), | |
1986 | long arg) | |
1987 | { | |
1988 | struct logfs_super *super = logfs_super(sb); | |
1989 | struct inode *inode; | |
1990 | struct page *page; | |
1991 | struct logfs_segment_entry *se; | |
1992 | pgoff_t page_no; | |
1993 | int child_no; | |
1994 | ||
1995 | page_no = segno >> (sb->s_blocksize_bits - 3); | |
1996 | child_no = segno & ((sb->s_blocksize >> 3) - 1); | |
1997 | ||
1998 | inode = super->s_segfile_inode; | |
1999 | page = logfs_get_write_page(inode, page_no, 0); | |
2000 | BUG_ON(!page); /* FIXME: We need some reserve page for this case */ | |
2001 | if (!PageUptodate(page)) | |
2002 | logfs_read_block(inode, page, WRITE); | |
2003 | ||
2004 | if (write) | |
2005 | alloc_indirect_block(inode, page, 0); | |
2006 | se = kmap_atomic(page, KM_USER0); | |
2007 | change_se(se + child_no, arg); | |
2008 | if (write) { | |
2009 | logfs_set_alias(sb, logfs_block(page), child_no); | |
2010 | BUG_ON((int)be32_to_cpu(se[child_no].valid) > super->s_segsize); | |
2011 | } | |
2012 | kunmap_atomic(se, KM_USER0); | |
2013 | ||
2014 | logfs_put_write_page(page); | |
2015 | } | |
2016 | ||
2017 | static void __get_segment_entry(struct logfs_segment_entry *se, long _target) | |
2018 | { | |
2019 | struct logfs_segment_entry *target = (void *)_target; | |
2020 | ||
2021 | *target = *se; | |
2022 | } | |
2023 | ||
2024 | void logfs_get_segment_entry(struct super_block *sb, u32 segno, | |
2025 | struct logfs_segment_entry *se) | |
2026 | { | |
2027 | logfs_mod_segment_entry(sb, segno, 0, __get_segment_entry, (long)se); | |
2028 | } | |
2029 | ||
2030 | static void __set_segment_used(struct logfs_segment_entry *se, long increment) | |
2031 | { | |
2032 | u32 valid; | |
2033 | ||
2034 | valid = be32_to_cpu(se->valid); | |
2035 | valid += increment; | |
2036 | se->valid = cpu_to_be32(valid); | |
2037 | } | |
2038 | ||
2039 | void logfs_set_segment_used(struct super_block *sb, u64 ofs, int increment) | |
2040 | { | |
2041 | struct logfs_super *super = logfs_super(sb); | |
2042 | u32 segno = ofs >> super->s_segshift; | |
2043 | ||
2044 | if (!increment) | |
2045 | return; | |
2046 | ||
2047 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_used, increment); | |
2048 | } | |
2049 | ||
2050 | static void __set_segment_erased(struct logfs_segment_entry *se, long ec_level) | |
2051 | { | |
2052 | se->ec_level = cpu_to_be32(ec_level); | |
2053 | } | |
2054 | ||
2055 | void logfs_set_segment_erased(struct super_block *sb, u32 segno, u32 ec, | |
2056 | gc_level_t gc_level) | |
2057 | { | |
2058 | u32 ec_level = ec << 4 | (__force u8)gc_level; | |
2059 | ||
2060 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_erased, ec_level); | |
2061 | } | |
2062 | ||
2063 | static void __set_segment_reserved(struct logfs_segment_entry *se, long ignore) | |
2064 | { | |
2065 | se->valid = cpu_to_be32(RESERVED); | |
2066 | } | |
2067 | ||
2068 | void logfs_set_segment_reserved(struct super_block *sb, u32 segno) | |
2069 | { | |
2070 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_reserved, 0); | |
2071 | } | |
2072 | ||
2073 | static void __set_segment_unreserved(struct logfs_segment_entry *se, | |
2074 | long ec_level) | |
2075 | { | |
2076 | se->valid = 0; | |
2077 | se->ec_level = cpu_to_be32(ec_level); | |
2078 | } | |
2079 | ||
2080 | void logfs_set_segment_unreserved(struct super_block *sb, u32 segno, u32 ec) | |
2081 | { | |
2082 | u32 ec_level = ec << 4; | |
2083 | ||
2084 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_unreserved, | |
2085 | ec_level); | |
2086 | } | |
2087 | ||
2088 | int __logfs_write_inode(struct inode *inode, long flags) | |
2089 | { | |
2090 | struct super_block *sb = inode->i_sb; | |
2091 | int ret; | |
2092 | ||
2093 | logfs_get_wblocks(sb, NULL, flags & WF_LOCK); | |
2094 | ret = do_write_inode(inode); | |
2095 | logfs_put_wblocks(sb, NULL, flags & WF_LOCK); | |
2096 | return ret; | |
2097 | } | |
2098 | ||
2099 | static int do_delete_inode(struct inode *inode) | |
2100 | { | |
2101 | struct super_block *sb = inode->i_sb; | |
2102 | struct inode *master_inode = logfs_super(sb)->s_master_inode; | |
2103 | struct page *page; | |
2104 | int ret; | |
2105 | ||
2106 | page = logfs_get_write_page(master_inode, inode->i_ino, 0); | |
2107 | if (!page) | |
2108 | return -ENOMEM; | |
2109 | ||
2110 | move_inode_to_page(page, inode); | |
2111 | ||
2112 | logfs_get_wblocks(sb, page, 1); | |
2113 | ret = __logfs_delete(master_inode, page); | |
2114 | logfs_put_wblocks(sb, page, 1); | |
2115 | ||
2116 | logfs_put_write_page(page); | |
2117 | return ret; | |
2118 | } | |
2119 | ||
2120 | /* | |
2121 | * ZOMBIE inodes have already been deleted before and should remain dead, | |
2122 | * if it weren't for valid checking. No need to kill them again here. | |
2123 | */ | |
2124 | void logfs_delete_inode(struct inode *inode) | |
2125 | { | |
2126 | struct logfs_inode *li = logfs_inode(inode); | |
2127 | ||
2128 | if (!(li->li_flags & LOGFS_IF_ZOMBIE)) { | |
2129 | li->li_flags |= LOGFS_IF_ZOMBIE; | |
2130 | if (i_size_read(inode) > 0) | |
2131 | logfs_truncate(inode, 0); | |
2132 | do_delete_inode(inode); | |
2133 | } | |
2134 | truncate_inode_pages(&inode->i_data, 0); | |
2135 | clear_inode(inode); | |
2136 | } | |
2137 | ||
2138 | void btree_write_block(struct logfs_block *block) | |
2139 | { | |
2140 | struct inode *inode; | |
2141 | struct page *page; | |
2142 | int err, cookie; | |
2143 | ||
2144 | inode = logfs_safe_iget(block->sb, block->ino, &cookie); | |
2145 | page = logfs_get_write_page(inode, block->bix, block->level); | |
2146 | ||
2147 | err = logfs_readpage_nolock(page); | |
2148 | BUG_ON(err); | |
2149 | BUG_ON(!PagePrivate(page)); | |
2150 | BUG_ON(logfs_block(page) != block); | |
2151 | err = __logfs_write_buf(inode, page, 0); | |
2152 | BUG_ON(err); | |
2153 | BUG_ON(PagePrivate(page) || page->private); | |
2154 | ||
2155 | logfs_put_write_page(page); | |
2156 | logfs_safe_iput(inode, cookie); | |
2157 | } | |
2158 | ||
2159 | /** | |
2160 | * logfs_inode_write - write inode or dentry objects | |
2161 | * | |
2162 | * @inode: parent inode (ifile or directory) | |
2163 | * @buf: object to write (inode or dentry) | |
2164 | * @n: object size | |
2165 | * @_pos: object number (file position in blocks/objects) | |
2166 | * @flags: write flags | |
2167 | * @lock: 0 if write lock is already taken, 1 otherwise | |
2168 | * @shadow_tree: shadow below this inode | |
2169 | * | |
2170 | * FIXME: All caller of this put a 200-300 byte variable on the stack, | |
2171 | * only to call here and do a memcpy from that stack variable. A good | |
2172 | * example of wasted performance and stack space. | |
2173 | */ | |
2174 | int logfs_inode_write(struct inode *inode, const void *buf, size_t count, | |
2175 | loff_t bix, long flags, struct shadow_tree *shadow_tree) | |
2176 | { | |
2177 | loff_t pos = bix << inode->i_sb->s_blocksize_bits; | |
2178 | int err; | |
2179 | struct page *page; | |
2180 | void *pagebuf; | |
2181 | ||
2182 | BUG_ON(pos & (LOGFS_BLOCKSIZE-1)); | |
2183 | BUG_ON(count > LOGFS_BLOCKSIZE); | |
2184 | page = logfs_get_write_page(inode, bix, 0); | |
2185 | if (!page) | |
2186 | return -ENOMEM; | |
2187 | ||
2188 | pagebuf = kmap_atomic(page, KM_USER0); | |
2189 | memcpy(pagebuf, buf, count); | |
2190 | flush_dcache_page(page); | |
2191 | kunmap_atomic(pagebuf, KM_USER0); | |
2192 | ||
2193 | if (i_size_read(inode) < pos + LOGFS_BLOCKSIZE) | |
2194 | i_size_write(inode, pos + LOGFS_BLOCKSIZE); | |
2195 | ||
2196 | err = logfs_write_buf(inode, page, flags); | |
2197 | logfs_put_write_page(page); | |
2198 | return err; | |
2199 | } | |
2200 | ||
2201 | int logfs_open_segfile(struct super_block *sb) | |
2202 | { | |
2203 | struct logfs_super *super = logfs_super(sb); | |
2204 | struct inode *inode; | |
2205 | ||
2206 | inode = logfs_read_meta_inode(sb, LOGFS_INO_SEGFILE); | |
2207 | if (IS_ERR(inode)) | |
2208 | return PTR_ERR(inode); | |
2209 | super->s_segfile_inode = inode; | |
2210 | return 0; | |
2211 | } | |
2212 | ||
2213 | int logfs_init_rw(struct super_block *sb) | |
2214 | { | |
2215 | struct logfs_super *super = logfs_super(sb); | |
2216 | int min_fill = 3 * super->s_no_blocks; | |
2217 | ||
2218 | INIT_LIST_HEAD(&super->s_object_alias); | |
2219 | mutex_init(&super->s_write_mutex); | |
2220 | super->s_block_pool = mempool_create_kmalloc_pool(min_fill, | |
2221 | sizeof(struct logfs_block)); | |
2222 | super->s_shadow_pool = mempool_create_kmalloc_pool(min_fill, | |
2223 | sizeof(struct logfs_shadow)); | |
2224 | return 0; | |
2225 | } | |
2226 | ||
2227 | void logfs_cleanup_rw(struct super_block *sb) | |
2228 | { | |
2229 | struct logfs_super *super = logfs_super(sb); | |
2230 | ||
2231 | destroy_meta_inode(super->s_segfile_inode); | |
2232 | if (super->s_block_pool) | |
2233 | mempool_destroy(super->s_block_pool); | |
2234 | if (super->s_shadow_pool) | |
2235 | mempool_destroy(super->s_shadow_pool); | |
2236 | } |