]>
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> | |
5a0e3ad6 | 21 | #include <linux/slab.h> |
5db53f3e JE |
22 | |
23 | static u64 adjust_bix(u64 bix, level_t level) | |
24 | { | |
25 | switch (level) { | |
26 | case 0: | |
27 | return bix; | |
28 | case LEVEL(1): | |
29 | return max_t(u64, bix, I0_BLOCKS); | |
30 | case LEVEL(2): | |
31 | return max_t(u64, bix, I1_BLOCKS); | |
32 | case LEVEL(3): | |
33 | return max_t(u64, bix, I2_BLOCKS); | |
34 | case LEVEL(4): | |
35 | return max_t(u64, bix, I3_BLOCKS); | |
36 | case LEVEL(5): | |
37 | return max_t(u64, bix, I4_BLOCKS); | |
38 | default: | |
39 | WARN_ON(1); | |
40 | return bix; | |
41 | } | |
42 | } | |
43 | ||
44 | static inline u64 maxbix(u8 height) | |
45 | { | |
46 | return 1ULL << (LOGFS_BLOCK_BITS * height); | |
47 | } | |
48 | ||
49 | /** | |
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 | |
53 | * index. | |
54 | * | |
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. | |
58 | */ | |
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) | |
63 | { | |
64 | return INDIRECT_BIT | (1ULL << LEVEL_SHIFT); | |
65 | } | |
66 | ||
67 | pgoff_t logfs_pack_index(u64 bix, level_t level) | |
68 | { | |
69 | pgoff_t index; | |
70 | ||
71 | BUG_ON(bix >= INDIRECT_BIT); | |
72 | if (level == 0) | |
73 | return bix; | |
74 | ||
75 | index = INDIRECT_BIT; | |
76 | index |= (__force long)level << LEVEL_SHIFT; | |
77 | index |= bix >> ((__force u8)level * LOGFS_BLOCK_BITS); | |
78 | return index; | |
79 | } | |
80 | ||
81 | void logfs_unpack_index(pgoff_t index, u64 *bix, level_t *level) | |
82 | { | |
83 | u8 __level; | |
84 | ||
85 | if (!(index & INDIRECT_BIT)) { | |
86 | *bix = index; | |
87 | *level = 0; | |
88 | return; | |
89 | } | |
90 | ||
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); | |
95 | return; | |
96 | } | |
97 | #undef ARCH_SHIFT | |
98 | #undef INDIRECT_BIT | |
99 | #undef LEVEL_SHIFT | |
100 | ||
101 | /* | |
102 | * Time is stored as nanoseconds since the epoch. | |
103 | */ | |
104 | static struct timespec be64_to_timespec(__be64 betime) | |
105 | { | |
106 | return ns_to_timespec(be64_to_cpu(betime)); | |
107 | } | |
108 | ||
109 | static __be64 timespec_to_be64(struct timespec tsp) | |
110 | { | |
111 | return cpu_to_be64((u64)tsp.tv_sec * NSEC_PER_SEC + tsp.tv_nsec); | |
112 | } | |
113 | ||
114 | static void logfs_disk_to_inode(struct logfs_disk_inode *di, struct inode*inode) | |
115 | { | |
116 | struct logfs_inode *li = logfs_inode(inode); | |
117 | int i; | |
118 | ||
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); | |
131 | ||
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 */ | |
136 | case S_IFIFO: | |
137 | inode->i_rdev = be64_to_cpu(di->di_data[0]); | |
138 | break; | |
139 | case S_IFDIR: /* fall through */ | |
140 | case S_IFREG: /* fall through */ | |
141 | case S_IFLNK: | |
142 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
143 | li->li_data[i] = be64_to_cpu(di->di_data[i]); | |
144 | break; | |
145 | default: | |
146 | BUG(); | |
147 | } | |
148 | } | |
149 | ||
150 | static void logfs_inode_to_disk(struct inode *inode, struct logfs_disk_inode*di) | |
151 | { | |
152 | struct logfs_inode *li = logfs_inode(inode); | |
153 | int i; | |
154 | ||
155 | di->di_mode = cpu_to_be16(inode->i_mode); | |
156 | di->di_height = li->li_height; | |
157 | di->di_pad = 0; | |
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); | |
168 | ||
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 */ | |
173 | case S_IFIFO: | |
174 | di->di_data[0] = cpu_to_be64(inode->i_rdev); | |
175 | break; | |
176 | case S_IFDIR: /* fall through */ | |
177 | case S_IFREG: /* fall through */ | |
178 | case S_IFLNK: | |
179 | for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++) | |
180 | di->di_data[i] = cpu_to_be64(li->li_data[i]); | |
181 | break; | |
182 | default: | |
183 | BUG(); | |
184 | } | |
185 | } | |
186 | ||
187 | static void __logfs_set_blocks(struct inode *inode) | |
188 | { | |
189 | struct super_block *sb = inode->i_sb; | |
190 | struct logfs_inode *li = logfs_inode(inode); | |
191 | ||
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; | |
195 | } | |
196 | ||
197 | void logfs_set_blocks(struct inode *inode, u64 bytes) | |
198 | { | |
199 | struct logfs_inode *li = logfs_inode(inode); | |
200 | ||
201 | li->li_used_bytes = bytes; | |
202 | __logfs_set_blocks(inode); | |
203 | } | |
204 | ||
205 | static void prelock_page(struct super_block *sb, struct page *page, int lock) | |
206 | { | |
207 | struct logfs_super *super = logfs_super(sb); | |
208 | ||
209 | BUG_ON(!PageLocked(page)); | |
210 | if (lock) { | |
211 | BUG_ON(PagePreLocked(page)); | |
212 | SetPagePreLocked(page); | |
213 | } else { | |
214 | /* We are in GC path. */ | |
215 | if (PagePreLocked(page)) | |
216 | super->s_lock_count++; | |
217 | else | |
218 | SetPagePreLocked(page); | |
219 | } | |
220 | } | |
221 | ||
222 | static void preunlock_page(struct super_block *sb, struct page *page, int lock) | |
223 | { | |
224 | struct logfs_super *super = logfs_super(sb); | |
225 | ||
226 | BUG_ON(!PageLocked(page)); | |
227 | if (lock) | |
228 | ClearPagePreLocked(page); | |
229 | else { | |
230 | /* We are in GC path. */ | |
231 | BUG_ON(!PagePreLocked(page)); | |
232 | if (super->s_lock_count) | |
233 | super->s_lock_count--; | |
234 | else | |
235 | ClearPagePreLocked(page); | |
236 | } | |
237 | } | |
238 | ||
239 | /* | |
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 | |
242 | * s_write_mutex. | |
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. | |
246 | */ | |
247 | static void logfs_get_wblocks(struct super_block *sb, struct page *page, | |
248 | int lock) | |
249 | { | |
250 | struct logfs_super *super = logfs_super(sb); | |
251 | ||
252 | if (page) | |
253 | prelock_page(sb, page, lock); | |
254 | ||
255 | if (lock) { | |
256 | mutex_lock(&super->s_write_mutex); | |
257 | logfs_gc_pass(sb); | |
258 | /* FIXME: We also have to check for shadowed space | |
259 | * and mempool fill grade */ | |
260 | } | |
261 | } | |
262 | ||
263 | static void logfs_put_wblocks(struct super_block *sb, struct page *page, | |
264 | int lock) | |
265 | { | |
266 | struct logfs_super *super = logfs_super(sb); | |
267 | ||
268 | if (page) | |
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. */ | |
272 | if (lock) | |
273 | mutex_unlock(&super->s_write_mutex); | |
274 | } | |
275 | ||
276 | static struct page *logfs_get_read_page(struct inode *inode, u64 bix, | |
277 | level_t level) | |
278 | { | |
279 | return find_or_create_page(inode->i_mapping, | |
280 | logfs_pack_index(bix, level), GFP_NOFS); | |
281 | } | |
282 | ||
283 | static void logfs_put_read_page(struct page *page) | |
284 | { | |
285 | unlock_page(page); | |
286 | page_cache_release(page); | |
287 | } | |
288 | ||
289 | static void logfs_lock_write_page(struct page *page) | |
290 | { | |
291 | int loop = 0; | |
292 | ||
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); | |
297 | BUG(); | |
298 | } | |
299 | if (PagePreLocked(page)) { | |
300 | /* Holder of page lock is waiting for us, it | |
301 | * is safe to use this page. */ | |
302 | break; | |
303 | } | |
304 | /* Some other process has this page locked and has | |
305 | * nothing to do with us. Wait for it to finish. | |
306 | */ | |
307 | schedule(); | |
308 | } | |
309 | BUG_ON(!PageLocked(page)); | |
310 | } | |
311 | ||
312 | static struct page *logfs_get_write_page(struct inode *inode, u64 bix, | |
313 | level_t level) | |
314 | { | |
315 | struct address_space *mapping = inode->i_mapping; | |
316 | pgoff_t index = logfs_pack_index(bix, level); | |
317 | struct page *page; | |
318 | int err; | |
319 | ||
320 | repeat: | |
321 | page = find_get_page(mapping, index); | |
322 | if (!page) { | |
323 | page = __page_cache_alloc(GFP_NOFS); | |
324 | if (!page) | |
325 | return NULL; | |
326 | err = add_to_page_cache_lru(page, mapping, index, GFP_NOFS); | |
327 | if (unlikely(err)) { | |
328 | page_cache_release(page); | |
329 | if (err == -EEXIST) | |
330 | goto repeat; | |
331 | return NULL; | |
332 | } | |
333 | } else logfs_lock_write_page(page); | |
334 | BUG_ON(!PageLocked(page)); | |
335 | return page; | |
336 | } | |
337 | ||
338 | static void logfs_unlock_write_page(struct page *page) | |
339 | { | |
340 | if (!PagePreLocked(page)) | |
341 | unlock_page(page); | |
342 | } | |
343 | ||
344 | static void logfs_put_write_page(struct page *page) | |
345 | { | |
346 | logfs_unlock_write_page(page); | |
347 | page_cache_release(page); | |
348 | } | |
349 | ||
350 | static struct page *logfs_get_page(struct inode *inode, u64 bix, level_t level, | |
351 | int rw) | |
352 | { | |
353 | if (rw == READ) | |
354 | return logfs_get_read_page(inode, bix, level); | |
355 | else | |
356 | return logfs_get_write_page(inode, bix, level); | |
357 | } | |
358 | ||
359 | static void logfs_put_page(struct page *page, int rw) | |
360 | { | |
361 | if (rw == READ) | |
362 | logfs_put_read_page(page); | |
363 | else | |
364 | logfs_put_write_page(page); | |
365 | } | |
366 | ||
367 | static unsigned long __get_bits(u64 val, int skip, int no) | |
368 | { | |
369 | u64 ret = val; | |
370 | ||
371 | ret >>= skip * no; | |
372 | ret <<= 64 - no; | |
373 | ret >>= 64 - no; | |
374 | return ret; | |
375 | } | |
376 | ||
377 | static unsigned long get_bits(u64 val, level_t skip) | |
378 | { | |
379 | return __get_bits(val, (__force int)skip, LOGFS_BLOCK_BITS); | |
380 | } | |
381 | ||
382 | static inline void init_shadow_tree(struct super_block *sb, | |
383 | struct shadow_tree *tree) | |
384 | { | |
385 | struct logfs_super *super = logfs_super(sb); | |
386 | ||
387 | btree_init_mempool64(&tree->new, super->s_btree_pool); | |
388 | btree_init_mempool64(&tree->old, super->s_btree_pool); | |
389 | } | |
390 | ||
391 | static void indirect_write_block(struct logfs_block *block) | |
392 | { | |
393 | struct page *page; | |
394 | struct inode *inode; | |
395 | int ret; | |
396 | ||
397 | page = block->page; | |
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); | |
402 | /* | |
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. | |
408 | * | |
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 | |
413 | * entirely tasteles. | |
414 | */ | |
415 | BUG_ON(ret); | |
416 | } | |
417 | ||
418 | static void inode_write_block(struct logfs_block *block) | |
419 | { | |
420 | struct inode *inode; | |
421 | int ret; | |
422 | ||
423 | inode = block->inode; | |
424 | if (inode->i_ino == LOGFS_INO_MASTER) | |
c6d38301 | 425 | logfs_write_anchor(inode->i_sb); |
5db53f3e JE |
426 | else { |
427 | ret = __logfs_write_inode(inode, 0); | |
428 | /* see indirect_write_block comment */ | |
429 | BUG_ON(ret); | |
430 | } | |
431 | } | |
432 | ||
5db53f3e JE |
433 | /* |
434 | * This silences a false, yet annoying gcc warning. I hate it when my editor | |
435 | * jumps into bitops.h each time I recompile this file. | |
436 | * TODO: Complain to gcc folks about this and upgrade compiler. | |
437 | */ | |
438 | static unsigned long fnb(const unsigned long *addr, | |
439 | unsigned long size, unsigned long offset) | |
440 | { | |
441 | return find_next_bit(addr, size, offset); | |
442 | } | |
443 | ||
444 | static __be64 inode_val0(struct inode *inode) | |
445 | { | |
446 | struct logfs_inode *li = logfs_inode(inode); | |
447 | u64 val; | |
448 | ||
449 | /* | |
450 | * Explicit shifting generates good code, but must match the format | |
451 | * of the structure. Add some paranoia just in case. | |
452 | */ | |
453 | BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_mode) != 0); | |
454 | BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_height) != 2); | |
455 | BUILD_BUG_ON(offsetof(struct logfs_disk_inode, di_flags) != 4); | |
456 | ||
457 | val = (u64)inode->i_mode << 48 | | |
458 | (u64)li->li_height << 40 | | |
459 | (u64)li->li_flags; | |
460 | return cpu_to_be64(val); | |
461 | } | |
462 | ||
463 | static int inode_write_alias(struct super_block *sb, | |
464 | struct logfs_block *block, write_alias_t *write_one_alias) | |
465 | { | |
466 | struct inode *inode = block->inode; | |
467 | struct logfs_inode *li = logfs_inode(inode); | |
468 | unsigned long pos; | |
469 | u64 ino , bix; | |
470 | __be64 val; | |
471 | level_t level; | |
472 | int err; | |
473 | ||
474 | for (pos = 0; ; pos++) { | |
475 | pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos); | |
476 | if (pos >= LOGFS_EMBEDDED_FIELDS + INODE_POINTER_OFS) | |
477 | return 0; | |
478 | ||
479 | switch (pos) { | |
480 | case INODE_HEIGHT_OFS: | |
481 | val = inode_val0(inode); | |
482 | break; | |
483 | case INODE_USED_OFS: | |
484 | val = cpu_to_be64(li->li_used_bytes);; | |
485 | break; | |
486 | case INODE_SIZE_OFS: | |
487 | val = cpu_to_be64(i_size_read(inode)); | |
488 | break; | |
489 | case INODE_POINTER_OFS ... INODE_POINTER_OFS + LOGFS_EMBEDDED_FIELDS - 1: | |
490 | val = cpu_to_be64(li->li_data[pos - INODE_POINTER_OFS]); | |
491 | break; | |
492 | default: | |
493 | BUG(); | |
494 | } | |
495 | ||
496 | ino = LOGFS_INO_MASTER; | |
497 | bix = inode->i_ino; | |
498 | level = LEVEL(0); | |
499 | err = write_one_alias(sb, ino, bix, level, pos, val); | |
500 | if (err) | |
501 | return err; | |
502 | } | |
503 | } | |
504 | ||
505 | static int indirect_write_alias(struct super_block *sb, | |
506 | struct logfs_block *block, write_alias_t *write_one_alias) | |
507 | { | |
508 | unsigned long pos; | |
509 | struct page *page = block->page; | |
510 | u64 ino , bix; | |
511 | __be64 *child, val; | |
512 | level_t level; | |
513 | int err; | |
514 | ||
515 | for (pos = 0; ; pos++) { | |
516 | pos = fnb(block->alias_map, LOGFS_BLOCK_FACTOR, pos); | |
517 | if (pos >= LOGFS_BLOCK_FACTOR) | |
518 | return 0; | |
519 | ||
520 | ino = page->mapping->host->i_ino; | |
521 | logfs_unpack_index(page->index, &bix, &level); | |
522 | child = kmap_atomic(page, KM_USER0); | |
523 | val = child[pos]; | |
524 | kunmap_atomic(child, KM_USER0); | |
525 | err = write_one_alias(sb, ino, bix, level, pos, val); | |
526 | if (err) | |
527 | return err; | |
528 | } | |
529 | } | |
530 | ||
531 | int logfs_write_obj_aliases_pagecache(struct super_block *sb) | |
532 | { | |
533 | struct logfs_super *super = logfs_super(sb); | |
534 | struct logfs_block *block; | |
535 | int err; | |
536 | ||
537 | list_for_each_entry(block, &super->s_object_alias, alias_list) { | |
538 | err = block->ops->write_alias(sb, block, write_alias_journal); | |
539 | if (err) | |
540 | return err; | |
541 | } | |
542 | return 0; | |
543 | } | |
544 | ||
545 | void __free_block(struct super_block *sb, struct logfs_block *block) | |
546 | { | |
547 | BUG_ON(!list_empty(&block->item_list)); | |
548 | list_del(&block->alias_list); | |
549 | mempool_free(block, logfs_super(sb)->s_block_pool); | |
550 | } | |
551 | ||
552 | static void inode_free_block(struct super_block *sb, struct logfs_block *block) | |
553 | { | |
554 | struct inode *inode = block->inode; | |
555 | ||
556 | logfs_inode(inode)->li_block = NULL; | |
557 | __free_block(sb, block); | |
558 | } | |
559 | ||
560 | static void indirect_free_block(struct super_block *sb, | |
561 | struct logfs_block *block) | |
562 | { | |
563 | ClearPagePrivate(block->page); | |
564 | block->page->private = 0; | |
565 | __free_block(sb, block); | |
566 | } | |
567 | ||
568 | ||
569 | static struct logfs_block_ops inode_block_ops = { | |
570 | .write_block = inode_write_block, | |
5db53f3e JE |
571 | .free_block = inode_free_block, |
572 | .write_alias = inode_write_alias, | |
573 | }; | |
574 | ||
575 | struct logfs_block_ops indirect_block_ops = { | |
576 | .write_block = indirect_write_block, | |
5db53f3e JE |
577 | .free_block = indirect_free_block, |
578 | .write_alias = indirect_write_alias, | |
579 | }; | |
580 | ||
581 | struct logfs_block *__alloc_block(struct super_block *sb, | |
582 | u64 ino, u64 bix, level_t level) | |
583 | { | |
584 | struct logfs_super *super = logfs_super(sb); | |
585 | struct logfs_block *block; | |
586 | ||
587 | block = mempool_alloc(super->s_block_pool, GFP_NOFS); | |
588 | memset(block, 0, sizeof(*block)); | |
589 | INIT_LIST_HEAD(&block->alias_list); | |
590 | INIT_LIST_HEAD(&block->item_list); | |
591 | block->sb = sb; | |
592 | block->ino = ino; | |
593 | block->bix = bix; | |
594 | block->level = level; | |
595 | return block; | |
596 | } | |
597 | ||
598 | static void alloc_inode_block(struct inode *inode) | |
599 | { | |
600 | struct logfs_inode *li = logfs_inode(inode); | |
601 | struct logfs_block *block; | |
602 | ||
603 | if (li->li_block) | |
604 | return; | |
605 | ||
606 | block = __alloc_block(inode->i_sb, LOGFS_INO_MASTER, inode->i_ino, 0); | |
607 | block->inode = inode; | |
608 | li->li_block = block; | |
609 | block->ops = &inode_block_ops; | |
610 | } | |
611 | ||
612 | void initialize_block_counters(struct page *page, struct logfs_block *block, | |
613 | __be64 *array, int page_is_empty) | |
614 | { | |
615 | u64 ptr; | |
616 | int i, start; | |
617 | ||
618 | block->partial = 0; | |
619 | block->full = 0; | |
620 | start = 0; | |
621 | if (page->index < first_indirect_block()) { | |
622 | /* Counters are pointless on level 0 */ | |
623 | return; | |
624 | } | |
625 | if (page->index == first_indirect_block()) { | |
626 | /* Skip unused pointers */ | |
627 | start = I0_BLOCKS; | |
628 | block->full = I0_BLOCKS; | |
629 | } | |
630 | if (!page_is_empty) { | |
631 | for (i = start; i < LOGFS_BLOCK_FACTOR; i++) { | |
632 | ptr = be64_to_cpu(array[i]); | |
633 | if (ptr) | |
634 | block->partial++; | |
635 | if (ptr & LOGFS_FULLY_POPULATED) | |
636 | block->full++; | |
637 | } | |
638 | } | |
639 | } | |
640 | ||
641 | static void alloc_data_block(struct inode *inode, struct page *page) | |
642 | { | |
643 | struct logfs_block *block; | |
644 | u64 bix; | |
645 | level_t level; | |
646 | ||
647 | if (PagePrivate(page)) | |
648 | return; | |
649 | ||
650 | logfs_unpack_index(page->index, &bix, &level); | |
651 | block = __alloc_block(inode->i_sb, inode->i_ino, bix, level); | |
652 | block->page = page; | |
653 | SetPagePrivate(page); | |
654 | page->private = (unsigned long)block; | |
655 | block->ops = &indirect_block_ops; | |
656 | } | |
657 | ||
658 | static void alloc_indirect_block(struct inode *inode, struct page *page, | |
659 | int page_is_empty) | |
660 | { | |
661 | struct logfs_block *block; | |
662 | __be64 *array; | |
663 | ||
664 | if (PagePrivate(page)) | |
665 | return; | |
666 | ||
667 | alloc_data_block(inode, page); | |
668 | ||
669 | block = logfs_block(page); | |
670 | array = kmap_atomic(page, KM_USER0); | |
671 | initialize_block_counters(page, block, array, page_is_empty); | |
672 | kunmap_atomic(array, KM_USER0); | |
673 | } | |
674 | ||
675 | static void block_set_pointer(struct page *page, int index, u64 ptr) | |
676 | { | |
677 | struct logfs_block *block = logfs_block(page); | |
678 | __be64 *array; | |
679 | u64 oldptr; | |
680 | ||
681 | BUG_ON(!block); | |
682 | array = kmap_atomic(page, KM_USER0); | |
683 | oldptr = be64_to_cpu(array[index]); | |
684 | array[index] = cpu_to_be64(ptr); | |
685 | kunmap_atomic(array, KM_USER0); | |
686 | SetPageUptodate(page); | |
687 | ||
688 | block->full += !!(ptr & LOGFS_FULLY_POPULATED) | |
689 | - !!(oldptr & LOGFS_FULLY_POPULATED); | |
690 | block->partial += !!ptr - !!oldptr; | |
691 | } | |
692 | ||
693 | static u64 block_get_pointer(struct page *page, int index) | |
694 | { | |
695 | __be64 *block; | |
696 | u64 ptr; | |
697 | ||
698 | block = kmap_atomic(page, KM_USER0); | |
699 | ptr = be64_to_cpu(block[index]); | |
700 | kunmap_atomic(block, KM_USER0); | |
701 | return ptr; | |
702 | } | |
703 | ||
704 | static int logfs_read_empty(struct page *page) | |
705 | { | |
706 | zero_user_segment(page, 0, PAGE_CACHE_SIZE); | |
707 | return 0; | |
708 | } | |
709 | ||
710 | static int logfs_read_direct(struct inode *inode, struct page *page) | |
711 | { | |
712 | struct logfs_inode *li = logfs_inode(inode); | |
713 | pgoff_t index = page->index; | |
714 | u64 block; | |
715 | ||
716 | block = li->li_data[index]; | |
717 | if (!block) | |
718 | return logfs_read_empty(page); | |
719 | ||
720 | return logfs_segment_read(inode, page, block, index, 0); | |
721 | } | |
722 | ||
723 | static int logfs_read_loop(struct inode *inode, struct page *page, | |
724 | int rw_context) | |
725 | { | |
726 | struct logfs_inode *li = logfs_inode(inode); | |
727 | u64 bix, bofs = li->li_data[INDIRECT_INDEX]; | |
728 | level_t level, target_level; | |
729 | int ret; | |
730 | struct page *ipage; | |
731 | ||
732 | logfs_unpack_index(page->index, &bix, &target_level); | |
733 | if (!bofs) | |
734 | return logfs_read_empty(page); | |
735 | ||
736 | if (bix >= maxbix(li->li_height)) | |
737 | return logfs_read_empty(page); | |
738 | ||
739 | for (level = LEVEL(li->li_height); | |
740 | (__force u8)level > (__force u8)target_level; | |
741 | level = SUBLEVEL(level)){ | |
742 | ipage = logfs_get_page(inode, bix, level, rw_context); | |
743 | if (!ipage) | |
744 | return -ENOMEM; | |
745 | ||
746 | ret = logfs_segment_read(inode, ipage, bofs, bix, level); | |
747 | if (ret) { | |
748 | logfs_put_read_page(ipage); | |
749 | return ret; | |
750 | } | |
751 | ||
752 | bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level))); | |
753 | logfs_put_page(ipage, rw_context); | |
754 | if (!bofs) | |
755 | return logfs_read_empty(page); | |
756 | } | |
757 | ||
758 | return logfs_segment_read(inode, page, bofs, bix, 0); | |
759 | } | |
760 | ||
761 | static int logfs_read_block(struct inode *inode, struct page *page, | |
762 | int rw_context) | |
763 | { | |
764 | pgoff_t index = page->index; | |
765 | ||
766 | if (index < I0_BLOCKS) | |
767 | return logfs_read_direct(inode, page); | |
768 | return logfs_read_loop(inode, page, rw_context); | |
769 | } | |
770 | ||
771 | static int logfs_exist_loop(struct inode *inode, u64 bix) | |
772 | { | |
773 | struct logfs_inode *li = logfs_inode(inode); | |
774 | u64 bofs = li->li_data[INDIRECT_INDEX]; | |
775 | level_t level; | |
776 | int ret; | |
777 | struct page *ipage; | |
778 | ||
779 | if (!bofs) | |
780 | return 0; | |
781 | if (bix >= maxbix(li->li_height)) | |
782 | return 0; | |
783 | ||
784 | for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)) { | |
785 | ipage = logfs_get_read_page(inode, bix, level); | |
786 | if (!ipage) | |
787 | return -ENOMEM; | |
788 | ||
789 | ret = logfs_segment_read(inode, ipage, bofs, bix, level); | |
790 | if (ret) { | |
791 | logfs_put_read_page(ipage); | |
792 | return ret; | |
793 | } | |
794 | ||
795 | bofs = block_get_pointer(ipage, get_bits(bix, SUBLEVEL(level))); | |
796 | logfs_put_read_page(ipage); | |
797 | if (!bofs) | |
798 | return 0; | |
799 | } | |
800 | ||
801 | return 1; | |
802 | } | |
803 | ||
804 | int logfs_exist_block(struct inode *inode, u64 bix) | |
805 | { | |
806 | struct logfs_inode *li = logfs_inode(inode); | |
807 | ||
808 | if (bix < I0_BLOCKS) | |
809 | return !!li->li_data[bix]; | |
810 | return logfs_exist_loop(inode, bix); | |
811 | } | |
812 | ||
813 | static u64 seek_holedata_direct(struct inode *inode, u64 bix, int data) | |
814 | { | |
815 | struct logfs_inode *li = logfs_inode(inode); | |
816 | ||
817 | for (; bix < I0_BLOCKS; bix++) | |
818 | if (data ^ (li->li_data[bix] == 0)) | |
819 | return bix; | |
820 | return I0_BLOCKS; | |
821 | } | |
822 | ||
823 | static u64 seek_holedata_loop(struct inode *inode, u64 bix, int data) | |
824 | { | |
825 | struct logfs_inode *li = logfs_inode(inode); | |
826 | __be64 *rblock; | |
827 | u64 increment, bofs = li->li_data[INDIRECT_INDEX]; | |
828 | level_t level; | |
829 | int ret, slot; | |
830 | struct page *page; | |
831 | ||
832 | BUG_ON(!bofs); | |
833 | ||
834 | for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)) { | |
835 | increment = 1 << (LOGFS_BLOCK_BITS * ((__force u8)level-1)); | |
836 | page = logfs_get_read_page(inode, bix, level); | |
837 | if (!page) | |
838 | return bix; | |
839 | ||
840 | ret = logfs_segment_read(inode, page, bofs, bix, level); | |
841 | if (ret) { | |
842 | logfs_put_read_page(page); | |
843 | return bix; | |
844 | } | |
845 | ||
846 | slot = get_bits(bix, SUBLEVEL(level)); | |
847 | rblock = kmap_atomic(page, KM_USER0); | |
848 | while (slot < LOGFS_BLOCK_FACTOR) { | |
849 | if (data && (rblock[slot] != 0)) | |
850 | break; | |
851 | if (!data && !(be64_to_cpu(rblock[slot]) & LOGFS_FULLY_POPULATED)) | |
852 | break; | |
853 | slot++; | |
854 | bix += increment; | |
855 | bix &= ~(increment - 1); | |
856 | } | |
857 | if (slot >= LOGFS_BLOCK_FACTOR) { | |
858 | kunmap_atomic(rblock, KM_USER0); | |
859 | logfs_put_read_page(page); | |
860 | return bix; | |
861 | } | |
862 | bofs = be64_to_cpu(rblock[slot]); | |
863 | kunmap_atomic(rblock, KM_USER0); | |
864 | logfs_put_read_page(page); | |
865 | if (!bofs) { | |
866 | BUG_ON(data); | |
867 | return bix; | |
868 | } | |
869 | } | |
870 | return bix; | |
871 | } | |
872 | ||
873 | /** | |
874 | * logfs_seek_hole - find next hole starting at a given block index | |
875 | * @inode: inode to search in | |
876 | * @bix: block index to start searching | |
877 | * | |
878 | * Returns next hole. If the file doesn't contain any further holes, the | |
879 | * block address next to eof is returned instead. | |
880 | */ | |
881 | u64 logfs_seek_hole(struct inode *inode, u64 bix) | |
882 | { | |
883 | struct logfs_inode *li = logfs_inode(inode); | |
884 | ||
885 | if (bix < I0_BLOCKS) { | |
886 | bix = seek_holedata_direct(inode, bix, 0); | |
887 | if (bix < I0_BLOCKS) | |
888 | return bix; | |
889 | } | |
890 | ||
891 | if (!li->li_data[INDIRECT_INDEX]) | |
892 | return bix; | |
893 | else if (li->li_data[INDIRECT_INDEX] & LOGFS_FULLY_POPULATED) | |
894 | bix = maxbix(li->li_height); | |
895 | else { | |
896 | bix = seek_holedata_loop(inode, bix, 0); | |
897 | if (bix < maxbix(li->li_height)) | |
898 | return bix; | |
899 | /* Should not happen anymore. But if some port writes semi- | |
900 | * corrupt images (as this one used to) we might run into it. | |
901 | */ | |
902 | WARN_ON_ONCE(bix == maxbix(li->li_height)); | |
903 | } | |
904 | ||
905 | return bix; | |
906 | } | |
907 | ||
908 | static u64 __logfs_seek_data(struct inode *inode, u64 bix) | |
909 | { | |
910 | struct logfs_inode *li = logfs_inode(inode); | |
911 | ||
912 | if (bix < I0_BLOCKS) { | |
913 | bix = seek_holedata_direct(inode, bix, 1); | |
914 | if (bix < I0_BLOCKS) | |
915 | return bix; | |
916 | } | |
917 | ||
918 | if (bix < maxbix(li->li_height)) { | |
919 | if (!li->li_data[INDIRECT_INDEX]) | |
920 | bix = maxbix(li->li_height); | |
921 | else | |
922 | return seek_holedata_loop(inode, bix, 1); | |
923 | } | |
924 | ||
925 | return bix; | |
926 | } | |
927 | ||
928 | /** | |
929 | * logfs_seek_data - find next data block after a given block index | |
930 | * @inode: inode to search in | |
931 | * @bix: block index to start searching | |
932 | * | |
933 | * Returns next data block. If the file doesn't contain any further data | |
934 | * blocks, the last block in the file is returned instead. | |
935 | */ | |
936 | u64 logfs_seek_data(struct inode *inode, u64 bix) | |
937 | { | |
938 | struct super_block *sb = inode->i_sb; | |
939 | u64 ret, end; | |
940 | ||
941 | ret = __logfs_seek_data(inode, bix); | |
942 | end = i_size_read(inode) >> sb->s_blocksize_bits; | |
943 | if (ret >= end) | |
944 | ret = max(bix, end); | |
945 | return ret; | |
946 | } | |
947 | ||
948 | static int logfs_is_valid_direct(struct logfs_inode *li, u64 bix, u64 ofs) | |
949 | { | |
950 | return pure_ofs(li->li_data[bix]) == ofs; | |
951 | } | |
952 | ||
953 | static int __logfs_is_valid_loop(struct inode *inode, u64 bix, | |
954 | u64 ofs, u64 bofs) | |
955 | { | |
956 | struct logfs_inode *li = logfs_inode(inode); | |
957 | level_t level; | |
958 | int ret; | |
959 | struct page *page; | |
960 | ||
961 | for (level = LEVEL(li->li_height); level != 0; level = SUBLEVEL(level)){ | |
962 | page = logfs_get_write_page(inode, bix, level); | |
963 | BUG_ON(!page); | |
964 | ||
965 | ret = logfs_segment_read(inode, page, bofs, bix, level); | |
966 | if (ret) { | |
967 | logfs_put_write_page(page); | |
968 | return 0; | |
969 | } | |
970 | ||
971 | bofs = block_get_pointer(page, get_bits(bix, SUBLEVEL(level))); | |
972 | logfs_put_write_page(page); | |
973 | if (!bofs) | |
974 | return 0; | |
975 | ||
976 | if (pure_ofs(bofs) == ofs) | |
977 | return 1; | |
978 | } | |
979 | return 0; | |
980 | } | |
981 | ||
982 | static int logfs_is_valid_loop(struct inode *inode, u64 bix, u64 ofs) | |
983 | { | |
984 | struct logfs_inode *li = logfs_inode(inode); | |
985 | u64 bofs = li->li_data[INDIRECT_INDEX]; | |
986 | ||
987 | if (!bofs) | |
988 | return 0; | |
989 | ||
990 | if (bix >= maxbix(li->li_height)) | |
991 | return 0; | |
992 | ||
993 | if (pure_ofs(bofs) == ofs) | |
994 | return 1; | |
995 | ||
996 | return __logfs_is_valid_loop(inode, bix, ofs, bofs); | |
997 | } | |
998 | ||
999 | static int __logfs_is_valid_block(struct inode *inode, u64 bix, u64 ofs) | |
1000 | { | |
1001 | struct logfs_inode *li = logfs_inode(inode); | |
1002 | ||
1003 | if ((inode->i_nlink == 0) && atomic_read(&inode->i_count) == 1) | |
1004 | return 0; | |
1005 | ||
1006 | if (bix < I0_BLOCKS) | |
1007 | return logfs_is_valid_direct(li, bix, ofs); | |
1008 | return logfs_is_valid_loop(inode, bix, ofs); | |
1009 | } | |
1010 | ||
1011 | /** | |
1012 | * logfs_is_valid_block - check whether this block is still valid | |
1013 | * | |
1014 | * @sb - superblock | |
1015 | * @ofs - block physical offset | |
1016 | * @ino - block inode number | |
1017 | * @bix - block index | |
1018 | * @level - block level | |
1019 | * | |
1020 | * Returns 0 if the block is invalid, 1 if it is valid and 2 if it will | |
1021 | * become invalid once the journal is written. | |
1022 | */ | |
1023 | int logfs_is_valid_block(struct super_block *sb, u64 ofs, u64 ino, u64 bix, | |
1024 | gc_level_t gc_level) | |
1025 | { | |
1026 | struct logfs_super *super = logfs_super(sb); | |
1027 | struct inode *inode; | |
1028 | int ret, cookie; | |
1029 | ||
1030 | /* Umount closes a segment with free blocks remaining. Those | |
1031 | * blocks are by definition invalid. */ | |
1032 | if (ino == -1) | |
1033 | return 0; | |
1034 | ||
1035 | LOGFS_BUG_ON((u64)(u_long)ino != ino, sb); | |
1036 | ||
1037 | inode = logfs_safe_iget(sb, ino, &cookie); | |
1038 | if (IS_ERR(inode)) | |
1039 | goto invalid; | |
1040 | ||
1041 | ret = __logfs_is_valid_block(inode, bix, ofs); | |
1042 | logfs_safe_iput(inode, cookie); | |
1043 | if (ret) | |
1044 | return ret; | |
1045 | ||
1046 | invalid: | |
1047 | /* Block is nominally invalid, but may still sit in the shadow tree, | |
1048 | * waiting for a journal commit. | |
1049 | */ | |
1050 | if (btree_lookup64(&super->s_shadow_tree.old, ofs)) | |
1051 | return 2; | |
1052 | return 0; | |
1053 | } | |
1054 | ||
1055 | int logfs_readpage_nolock(struct page *page) | |
1056 | { | |
1057 | struct inode *inode = page->mapping->host; | |
1058 | int ret = -EIO; | |
1059 | ||
1060 | ret = logfs_read_block(inode, page, READ); | |
1061 | ||
1062 | if (ret) { | |
1063 | ClearPageUptodate(page); | |
1064 | SetPageError(page); | |
1065 | } else { | |
1066 | SetPageUptodate(page); | |
1067 | ClearPageError(page); | |
1068 | } | |
1069 | flush_dcache_page(page); | |
1070 | ||
1071 | return ret; | |
1072 | } | |
1073 | ||
1074 | static int logfs_reserve_bytes(struct inode *inode, int bytes) | |
1075 | { | |
1076 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1077 | u64 available = super->s_free_bytes + super->s_dirty_free_bytes | |
1078 | - super->s_dirty_used_bytes - super->s_dirty_pages; | |
1079 | ||
1080 | if (!bytes) | |
1081 | return 0; | |
1082 | ||
1083 | if (available < bytes) | |
1084 | return -ENOSPC; | |
1085 | ||
1086 | if (available < bytes + super->s_root_reserve && | |
1087 | !capable(CAP_SYS_RESOURCE)) | |
1088 | return -ENOSPC; | |
1089 | ||
1090 | return 0; | |
1091 | } | |
1092 | ||
1093 | int get_page_reserve(struct inode *inode, struct page *page) | |
1094 | { | |
1095 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1096 | int ret; | |
1097 | ||
1098 | if (logfs_block(page) && logfs_block(page)->reserved_bytes) | |
1099 | return 0; | |
1100 | ||
1101 | logfs_get_wblocks(inode->i_sb, page, WF_LOCK); | |
1102 | ret = logfs_reserve_bytes(inode, 6 * LOGFS_MAX_OBJECTSIZE); | |
1103 | if (!ret) { | |
1104 | alloc_data_block(inode, page); | |
1105 | logfs_block(page)->reserved_bytes += 6 * LOGFS_MAX_OBJECTSIZE; | |
1106 | super->s_dirty_pages += 6 * LOGFS_MAX_OBJECTSIZE; | |
1107 | } | |
1108 | logfs_put_wblocks(inode->i_sb, page, WF_LOCK); | |
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * We are protected by write lock. Push victims up to superblock level | |
1114 | * and release transaction when appropriate. | |
1115 | */ | |
1116 | /* FIXME: This is currently called from the wrong spots. */ | |
1117 | static void logfs_handle_transaction(struct inode *inode, | |
1118 | struct logfs_transaction *ta) | |
1119 | { | |
1120 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1121 | ||
1122 | if (!ta) | |
1123 | return; | |
1124 | logfs_inode(inode)->li_block->ta = NULL; | |
1125 | ||
1126 | if (inode->i_ino != LOGFS_INO_MASTER) { | |
1127 | BUG(); /* FIXME: Yes, this needs more thought */ | |
1128 | /* just remember the transaction until inode is written */ | |
1129 | //BUG_ON(logfs_inode(inode)->li_transaction); | |
1130 | //logfs_inode(inode)->li_transaction = ta; | |
1131 | return; | |
1132 | } | |
1133 | ||
1134 | switch (ta->state) { | |
1135 | case CREATE_1: /* fall through */ | |
1136 | case UNLINK_1: | |
1137 | BUG_ON(super->s_victim_ino); | |
1138 | super->s_victim_ino = ta->ino; | |
1139 | break; | |
1140 | case CREATE_2: /* fall through */ | |
1141 | case UNLINK_2: | |
1142 | BUG_ON(super->s_victim_ino != ta->ino); | |
1143 | super->s_victim_ino = 0; | |
1144 | /* transaction ends here - free it */ | |
1145 | kfree(ta); | |
1146 | break; | |
1147 | case CROSS_RENAME_1: | |
1148 | BUG_ON(super->s_rename_dir); | |
1149 | BUG_ON(super->s_rename_pos); | |
1150 | super->s_rename_dir = ta->dir; | |
1151 | super->s_rename_pos = ta->pos; | |
1152 | break; | |
1153 | case CROSS_RENAME_2: | |
1154 | BUG_ON(super->s_rename_dir != ta->dir); | |
1155 | BUG_ON(super->s_rename_pos != ta->pos); | |
1156 | super->s_rename_dir = 0; | |
1157 | super->s_rename_pos = 0; | |
1158 | kfree(ta); | |
1159 | break; | |
1160 | case TARGET_RENAME_1: | |
1161 | BUG_ON(super->s_rename_dir); | |
1162 | BUG_ON(super->s_rename_pos); | |
1163 | BUG_ON(super->s_victim_ino); | |
1164 | super->s_rename_dir = ta->dir; | |
1165 | super->s_rename_pos = ta->pos; | |
1166 | super->s_victim_ino = ta->ino; | |
1167 | break; | |
1168 | case TARGET_RENAME_2: | |
1169 | BUG_ON(super->s_rename_dir != ta->dir); | |
1170 | BUG_ON(super->s_rename_pos != ta->pos); | |
1171 | BUG_ON(super->s_victim_ino != ta->ino); | |
1172 | super->s_rename_dir = 0; | |
1173 | super->s_rename_pos = 0; | |
1174 | break; | |
1175 | case TARGET_RENAME_3: | |
1176 | BUG_ON(super->s_rename_dir); | |
1177 | BUG_ON(super->s_rename_pos); | |
1178 | BUG_ON(super->s_victim_ino != ta->ino); | |
1179 | super->s_victim_ino = 0; | |
1180 | kfree(ta); | |
1181 | break; | |
1182 | default: | |
1183 | BUG(); | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * Not strictly a reservation, but rather a check that we still have enough | |
1189 | * space to satisfy the write. | |
1190 | */ | |
1191 | static int logfs_reserve_blocks(struct inode *inode, int blocks) | |
1192 | { | |
1193 | return logfs_reserve_bytes(inode, blocks * LOGFS_MAX_OBJECTSIZE); | |
1194 | } | |
1195 | ||
1196 | struct write_control { | |
1197 | u64 ofs; | |
1198 | long flags; | |
1199 | }; | |
1200 | ||
1201 | static struct logfs_shadow *alloc_shadow(struct inode *inode, u64 bix, | |
1202 | level_t level, u64 old_ofs) | |
1203 | { | |
1204 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1205 | struct logfs_shadow *shadow; | |
1206 | ||
1207 | shadow = mempool_alloc(super->s_shadow_pool, GFP_NOFS); | |
1208 | memset(shadow, 0, sizeof(*shadow)); | |
1209 | shadow->ino = inode->i_ino; | |
1210 | shadow->bix = bix; | |
1211 | shadow->gc_level = expand_level(inode->i_ino, level); | |
1212 | shadow->old_ofs = old_ofs & ~LOGFS_FULLY_POPULATED; | |
1213 | return shadow; | |
1214 | } | |
1215 | ||
1216 | static void free_shadow(struct inode *inode, struct logfs_shadow *shadow) | |
1217 | { | |
1218 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1219 | ||
1220 | mempool_free(shadow, super->s_shadow_pool); | |
1221 | } | |
1222 | ||
032d8f72 JE |
1223 | static void mark_segment(struct shadow_tree *tree, u32 segno) |
1224 | { | |
1225 | int err; | |
1226 | ||
1227 | if (!btree_lookup32(&tree->segment_map, segno)) { | |
1228 | err = btree_insert32(&tree->segment_map, segno, (void *)1, | |
1229 | GFP_NOFS); | |
1230 | BUG_ON(err); | |
1231 | tree->no_shadowed_segments++; | |
1232 | } | |
1233 | } | |
1234 | ||
5db53f3e JE |
1235 | /** |
1236 | * fill_shadow_tree - Propagate shadow tree changes due to a write | |
1237 | * @inode: Inode owning the page | |
1238 | * @page: Struct page that was written | |
1239 | * @shadow: Shadow for the current write | |
1240 | * | |
1241 | * Writes in logfs can result in two semi-valid objects. The old object | |
1242 | * is still valid as long as it can be reached by following pointers on | |
1243 | * the medium. Only when writes propagate all the way up to the journal | |
1244 | * has the new object safely replaced the old one. | |
1245 | * | |
1246 | * To handle this problem, a struct logfs_shadow is used to represent | |
1247 | * every single write. It is attached to the indirect block, which is | |
1248 | * marked dirty. When the indirect block is written, its shadows are | |
1249 | * handed up to the next indirect block (or inode). Untimately they | |
1250 | * will reach the master inode and be freed upon journal commit. | |
1251 | * | |
1252 | * This function handles a single step in the propagation. It adds the | |
1253 | * shadow for the current write to the tree, along with any shadows in | |
1254 | * the page's tree, in case it was an indirect block. If a page is | |
1255 | * written, the inode parameter is left NULL, if an inode is written, | |
1256 | * the page parameter is left NULL. | |
1257 | */ | |
1258 | static void fill_shadow_tree(struct inode *inode, struct page *page, | |
1259 | struct logfs_shadow *shadow) | |
1260 | { | |
1261 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1262 | struct logfs_block *block = logfs_block(page); | |
1263 | struct shadow_tree *tree = &super->s_shadow_tree; | |
1264 | ||
1265 | if (PagePrivate(page)) { | |
1266 | if (block->alias_map) | |
1267 | super->s_no_object_aliases -= bitmap_weight( | |
1268 | block->alias_map, LOGFS_BLOCK_FACTOR); | |
1269 | logfs_handle_transaction(inode, block->ta); | |
1270 | block->ops->free_block(inode->i_sb, block); | |
1271 | } | |
1272 | if (shadow) { | |
1273 | if (shadow->old_ofs) | |
1274 | btree_insert64(&tree->old, shadow->old_ofs, shadow, | |
1275 | GFP_NOFS); | |
1276 | else | |
1277 | btree_insert64(&tree->new, shadow->new_ofs, shadow, | |
1278 | GFP_NOFS); | |
1279 | ||
1280 | super->s_dirty_used_bytes += shadow->new_len; | |
1281 | super->s_dirty_free_bytes += shadow->old_len; | |
032d8f72 JE |
1282 | mark_segment(tree, shadow->old_ofs >> super->s_segshift); |
1283 | mark_segment(tree, shadow->new_ofs >> super->s_segshift); | |
5db53f3e JE |
1284 | } |
1285 | } | |
1286 | ||
1287 | static void logfs_set_alias(struct super_block *sb, struct logfs_block *block, | |
1288 | long child_no) | |
1289 | { | |
1290 | struct logfs_super *super = logfs_super(sb); | |
1291 | ||
1292 | if (block->inode && block->inode->i_ino == LOGFS_INO_MASTER) { | |
1293 | /* Aliases in the master inode are pointless. */ | |
1294 | return; | |
1295 | } | |
1296 | ||
1297 | if (!test_bit(child_no, block->alias_map)) { | |
1298 | set_bit(child_no, block->alias_map); | |
1299 | super->s_no_object_aliases++; | |
1300 | } | |
1301 | list_move_tail(&block->alias_list, &super->s_object_alias); | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Object aliases can and often do change the size and occupied space of a | |
1306 | * file. So not only do we have to change the pointers, we also have to | |
1307 | * change inode->i_size and li->li_used_bytes. Which is done by setting | |
1308 | * another two object aliases for the inode itself. | |
1309 | */ | |
1310 | static void set_iused(struct inode *inode, struct logfs_shadow *shadow) | |
1311 | { | |
1312 | struct logfs_inode *li = logfs_inode(inode); | |
1313 | ||
1314 | if (shadow->new_len == shadow->old_len) | |
1315 | return; | |
1316 | ||
1317 | alloc_inode_block(inode); | |
1318 | li->li_used_bytes += shadow->new_len - shadow->old_len; | |
1319 | __logfs_set_blocks(inode); | |
1320 | logfs_set_alias(inode->i_sb, li->li_block, INODE_USED_OFS); | |
1321 | logfs_set_alias(inode->i_sb, li->li_block, INODE_SIZE_OFS); | |
1322 | } | |
1323 | ||
1324 | static int logfs_write_i0(struct inode *inode, struct page *page, | |
1325 | struct write_control *wc) | |
1326 | { | |
1327 | struct logfs_shadow *shadow; | |
1328 | u64 bix; | |
1329 | level_t level; | |
1330 | int full, err = 0; | |
1331 | ||
1332 | logfs_unpack_index(page->index, &bix, &level); | |
1333 | if (wc->ofs == 0) | |
1334 | if (logfs_reserve_blocks(inode, 1)) | |
1335 | return -ENOSPC; | |
1336 | ||
1337 | shadow = alloc_shadow(inode, bix, level, wc->ofs); | |
1338 | if (wc->flags & WF_WRITE) | |
1339 | err = logfs_segment_write(inode, page, shadow); | |
1340 | if (wc->flags & WF_DELETE) | |
1341 | logfs_segment_delete(inode, shadow); | |
1342 | if (err) { | |
1343 | free_shadow(inode, shadow); | |
1344 | return err; | |
1345 | } | |
1346 | ||
1347 | set_iused(inode, shadow); | |
1348 | full = 1; | |
1349 | if (level != 0) { | |
1350 | alloc_indirect_block(inode, page, 0); | |
1351 | full = logfs_block(page)->full == LOGFS_BLOCK_FACTOR; | |
1352 | } | |
1353 | fill_shadow_tree(inode, page, shadow); | |
1354 | wc->ofs = shadow->new_ofs; | |
1355 | if (wc->ofs && full) | |
1356 | wc->ofs |= LOGFS_FULLY_POPULATED; | |
1357 | return 0; | |
1358 | } | |
1359 | ||
1360 | static int logfs_write_direct(struct inode *inode, struct page *page, | |
1361 | long flags) | |
1362 | { | |
1363 | struct logfs_inode *li = logfs_inode(inode); | |
1364 | struct write_control wc = { | |
1365 | .ofs = li->li_data[page->index], | |
1366 | .flags = flags, | |
1367 | }; | |
1368 | int err; | |
1369 | ||
1370 | alloc_inode_block(inode); | |
1371 | ||
1372 | err = logfs_write_i0(inode, page, &wc); | |
1373 | if (err) | |
1374 | return err; | |
1375 | ||
1376 | li->li_data[page->index] = wc.ofs; | |
1377 | logfs_set_alias(inode->i_sb, li->li_block, | |
1378 | page->index + INODE_POINTER_OFS); | |
1379 | return 0; | |
1380 | } | |
1381 | ||
1382 | static int ptr_change(u64 ofs, struct page *page) | |
1383 | { | |
1384 | struct logfs_block *block = logfs_block(page); | |
1385 | int empty0, empty1, full0, full1; | |
1386 | ||
1387 | empty0 = ofs == 0; | |
1388 | empty1 = block->partial == 0; | |
1389 | if (empty0 != empty1) | |
1390 | return 1; | |
1391 | ||
1392 | /* The !! is necessary to shrink result to int */ | |
1393 | full0 = !!(ofs & LOGFS_FULLY_POPULATED); | |
1394 | full1 = block->full == LOGFS_BLOCK_FACTOR; | |
1395 | if (full0 != full1) | |
1396 | return 1; | |
1397 | return 0; | |
1398 | } | |
1399 | ||
1400 | static int __logfs_write_rec(struct inode *inode, struct page *page, | |
1401 | struct write_control *this_wc, | |
1402 | pgoff_t bix, level_t target_level, level_t level) | |
1403 | { | |
1404 | int ret, page_empty = 0; | |
1405 | int child_no = get_bits(bix, SUBLEVEL(level)); | |
1406 | struct page *ipage; | |
1407 | struct write_control child_wc = { | |
1408 | .flags = this_wc->flags, | |
1409 | }; | |
1410 | ||
1411 | ipage = logfs_get_write_page(inode, bix, level); | |
1412 | if (!ipage) | |
1413 | return -ENOMEM; | |
1414 | ||
1415 | if (this_wc->ofs) { | |
1416 | ret = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level); | |
1417 | if (ret) | |
1418 | goto out; | |
1419 | } else if (!PageUptodate(ipage)) { | |
1420 | page_empty = 1; | |
1421 | logfs_read_empty(ipage); | |
1422 | } | |
1423 | ||
1424 | child_wc.ofs = block_get_pointer(ipage, child_no); | |
1425 | ||
1426 | if ((__force u8)level-1 > (__force u8)target_level) | |
1427 | ret = __logfs_write_rec(inode, page, &child_wc, bix, | |
1428 | target_level, SUBLEVEL(level)); | |
1429 | else | |
1430 | ret = logfs_write_i0(inode, page, &child_wc); | |
1431 | ||
1432 | if (ret) | |
1433 | goto out; | |
1434 | ||
1435 | alloc_indirect_block(inode, ipage, page_empty); | |
1436 | block_set_pointer(ipage, child_no, child_wc.ofs); | |
1437 | /* FIXME: first condition seems superfluous */ | |
1438 | if (child_wc.ofs || logfs_block(ipage)->partial) | |
1439 | this_wc->flags |= WF_WRITE; | |
1440 | /* the condition on this_wc->ofs ensures that we won't consume extra | |
1441 | * space for indirect blocks in the future, which we cannot reserve */ | |
1442 | if (!this_wc->ofs || ptr_change(this_wc->ofs, ipage)) | |
1443 | ret = logfs_write_i0(inode, ipage, this_wc); | |
1444 | else | |
1445 | logfs_set_alias(inode->i_sb, logfs_block(ipage), child_no); | |
1446 | out: | |
1447 | logfs_put_write_page(ipage); | |
1448 | return ret; | |
1449 | } | |
1450 | ||
1451 | static int logfs_write_rec(struct inode *inode, struct page *page, | |
1452 | pgoff_t bix, level_t target_level, long flags) | |
1453 | { | |
1454 | struct logfs_inode *li = logfs_inode(inode); | |
1455 | struct write_control wc = { | |
1456 | .ofs = li->li_data[INDIRECT_INDEX], | |
1457 | .flags = flags, | |
1458 | }; | |
1459 | int ret; | |
1460 | ||
1461 | alloc_inode_block(inode); | |
1462 | ||
1463 | if (li->li_height > (__force u8)target_level) | |
1464 | ret = __logfs_write_rec(inode, page, &wc, bix, target_level, | |
1465 | LEVEL(li->li_height)); | |
1466 | else | |
1467 | ret = logfs_write_i0(inode, page, &wc); | |
1468 | if (ret) | |
1469 | return ret; | |
1470 | ||
1471 | if (li->li_data[INDIRECT_INDEX] != wc.ofs) { | |
1472 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1473 | logfs_set_alias(inode->i_sb, li->li_block, | |
1474 | INDIRECT_INDEX + INODE_POINTER_OFS); | |
1475 | } | |
1476 | return ret; | |
1477 | } | |
1478 | ||
1479 | void logfs_add_transaction(struct inode *inode, struct logfs_transaction *ta) | |
1480 | { | |
1481 | alloc_inode_block(inode); | |
1482 | logfs_inode(inode)->li_block->ta = ta; | |
1483 | } | |
1484 | ||
1485 | void logfs_del_transaction(struct inode *inode, struct logfs_transaction *ta) | |
1486 | { | |
1487 | struct logfs_block *block = logfs_inode(inode)->li_block; | |
1488 | ||
1489 | if (block && block->ta) | |
1490 | block->ta = NULL; | |
1491 | } | |
1492 | ||
1493 | static int grow_inode(struct inode *inode, u64 bix, level_t level) | |
1494 | { | |
1495 | struct logfs_inode *li = logfs_inode(inode); | |
1496 | u8 height = (__force u8)level; | |
1497 | struct page *page; | |
1498 | struct write_control wc = { | |
1499 | .flags = WF_WRITE, | |
1500 | }; | |
1501 | int err; | |
1502 | ||
1503 | BUG_ON(height > 5 || li->li_height > 5); | |
1504 | while (height > li->li_height || bix >= maxbix(li->li_height)) { | |
1505 | page = logfs_get_write_page(inode, I0_BLOCKS + 1, | |
1506 | LEVEL(li->li_height + 1)); | |
1507 | if (!page) | |
1508 | return -ENOMEM; | |
1509 | logfs_read_empty(page); | |
1510 | alloc_indirect_block(inode, page, 1); | |
1511 | block_set_pointer(page, 0, li->li_data[INDIRECT_INDEX]); | |
1512 | err = logfs_write_i0(inode, page, &wc); | |
1513 | logfs_put_write_page(page); | |
1514 | if (err) | |
1515 | return err; | |
1516 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1517 | wc.ofs = 0; | |
1518 | li->li_height++; | |
1519 | logfs_set_alias(inode->i_sb, li->li_block, INODE_HEIGHT_OFS); | |
1520 | } | |
1521 | return 0; | |
1522 | } | |
1523 | ||
1524 | static int __logfs_write_buf(struct inode *inode, struct page *page, long flags) | |
1525 | { | |
1526 | struct logfs_super *super = logfs_super(inode->i_sb); | |
1527 | pgoff_t index = page->index; | |
1528 | u64 bix; | |
1529 | level_t level; | |
1530 | int err; | |
1531 | ||
1532 | flags |= WF_WRITE | WF_DELETE; | |
1533 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
1534 | ||
1535 | logfs_unpack_index(index, &bix, &level); | |
1536 | if (logfs_block(page) && logfs_block(page)->reserved_bytes) | |
1537 | super->s_dirty_pages -= logfs_block(page)->reserved_bytes; | |
1538 | ||
1539 | if (index < I0_BLOCKS) | |
1540 | return logfs_write_direct(inode, page, flags); | |
1541 | ||
1542 | bix = adjust_bix(bix, level); | |
1543 | err = grow_inode(inode, bix, level); | |
1544 | if (err) | |
1545 | return err; | |
1546 | return logfs_write_rec(inode, page, bix, level, flags); | |
1547 | } | |
1548 | ||
1549 | int logfs_write_buf(struct inode *inode, struct page *page, long flags) | |
1550 | { | |
1551 | struct super_block *sb = inode->i_sb; | |
1552 | int ret; | |
1553 | ||
1554 | logfs_get_wblocks(sb, page, flags & WF_LOCK); | |
1555 | ret = __logfs_write_buf(inode, page, flags); | |
1556 | logfs_put_wblocks(sb, page, flags & WF_LOCK); | |
1557 | return ret; | |
1558 | } | |
1559 | ||
1560 | static int __logfs_delete(struct inode *inode, struct page *page) | |
1561 | { | |
1562 | long flags = WF_DELETE; | |
1563 | ||
1564 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
1565 | ||
1566 | if (page->index < I0_BLOCKS) | |
1567 | return logfs_write_direct(inode, page, flags); | |
1568 | return logfs_write_rec(inode, page, page->index, 0, flags); | |
1569 | } | |
1570 | ||
1571 | int logfs_delete(struct inode *inode, pgoff_t index, | |
1572 | struct shadow_tree *shadow_tree) | |
1573 | { | |
1574 | struct super_block *sb = inode->i_sb; | |
1575 | struct page *page; | |
1576 | int ret; | |
1577 | ||
1578 | page = logfs_get_read_page(inode, index, 0); | |
1579 | if (!page) | |
1580 | return -ENOMEM; | |
1581 | ||
1582 | logfs_get_wblocks(sb, page, 1); | |
1583 | ret = __logfs_delete(inode, page); | |
1584 | logfs_put_wblocks(sb, page, 1); | |
1585 | ||
1586 | logfs_put_read_page(page); | |
1587 | ||
1588 | return ret; | |
1589 | } | |
1590 | ||
5db53f3e JE |
1591 | int logfs_rewrite_block(struct inode *inode, u64 bix, u64 ofs, |
1592 | gc_level_t gc_level, long flags) | |
1593 | { | |
1594 | level_t level = shrink_level(gc_level); | |
1595 | struct page *page; | |
1596 | int err; | |
1597 | ||
1598 | page = logfs_get_write_page(inode, bix, level); | |
1599 | if (!page) | |
1600 | return -ENOMEM; | |
1601 | ||
1602 | err = logfs_segment_read(inode, page, ofs, bix, level); | |
1603 | if (!err) { | |
1604 | if (level != 0) | |
1605 | alloc_indirect_block(inode, page, 0); | |
1606 | err = logfs_write_buf(inode, page, flags); | |
19321917 JE |
1607 | if (!err && shrink_level(gc_level) == 0) { |
1608 | /* Rewrite cannot mark the inode dirty but has to | |
1609 | * write it immediatly. | |
1610 | * Q: Can't we just create an alias for the inode | |
1611 | * instead? And if not, why not? | |
1612 | */ | |
1613 | if (inode->i_ino == LOGFS_INO_MASTER) | |
1614 | logfs_write_anchor(inode->i_sb); | |
1615 | else { | |
1616 | err = __logfs_write_inode(inode, flags); | |
1617 | } | |
1618 | } | |
5db53f3e JE |
1619 | } |
1620 | logfs_put_write_page(page); | |
1621 | return err; | |
1622 | } | |
1623 | ||
1624 | static int truncate_data_block(struct inode *inode, struct page *page, | |
1625 | u64 ofs, struct logfs_shadow *shadow, u64 size) | |
1626 | { | |
1627 | loff_t pageofs = page->index << inode->i_sb->s_blocksize_bits; | |
1628 | u64 bix; | |
1629 | level_t level; | |
1630 | int err; | |
1631 | ||
1632 | /* Does truncation happen within this page? */ | |
1633 | if (size <= pageofs || size - pageofs >= PAGE_SIZE) | |
1634 | return 0; | |
1635 | ||
1636 | logfs_unpack_index(page->index, &bix, &level); | |
1637 | BUG_ON(level != 0); | |
1638 | ||
1639 | err = logfs_segment_read(inode, page, ofs, bix, level); | |
1640 | if (err) | |
1641 | return err; | |
1642 | ||
1643 | zero_user_segment(page, size - pageofs, PAGE_CACHE_SIZE); | |
1644 | return logfs_segment_write(inode, page, shadow); | |
1645 | } | |
1646 | ||
1647 | static int logfs_truncate_i0(struct inode *inode, struct page *page, | |
1648 | struct write_control *wc, u64 size) | |
1649 | { | |
1650 | struct logfs_shadow *shadow; | |
1651 | u64 bix; | |
1652 | level_t level; | |
1653 | int err = 0; | |
1654 | ||
1655 | logfs_unpack_index(page->index, &bix, &level); | |
1656 | BUG_ON(level != 0); | |
1657 | shadow = alloc_shadow(inode, bix, level, wc->ofs); | |
1658 | ||
1659 | err = truncate_data_block(inode, page, wc->ofs, shadow, size); | |
1660 | if (err) { | |
1661 | free_shadow(inode, shadow); | |
1662 | return err; | |
1663 | } | |
1664 | ||
1665 | logfs_segment_delete(inode, shadow); | |
1666 | set_iused(inode, shadow); | |
1667 | fill_shadow_tree(inode, page, shadow); | |
1668 | wc->ofs = shadow->new_ofs; | |
1669 | return 0; | |
1670 | } | |
1671 | ||
1672 | static int logfs_truncate_direct(struct inode *inode, u64 size) | |
1673 | { | |
1674 | struct logfs_inode *li = logfs_inode(inode); | |
1675 | struct write_control wc; | |
1676 | struct page *page; | |
1677 | int e; | |
1678 | int err; | |
1679 | ||
1680 | alloc_inode_block(inode); | |
1681 | ||
1682 | for (e = I0_BLOCKS - 1; e >= 0; e--) { | |
1683 | if (size > (e+1) * LOGFS_BLOCKSIZE) | |
1684 | break; | |
1685 | ||
1686 | wc.ofs = li->li_data[e]; | |
1687 | if (!wc.ofs) | |
1688 | continue; | |
1689 | ||
1690 | page = logfs_get_write_page(inode, e, 0); | |
1691 | if (!page) | |
1692 | return -ENOMEM; | |
1693 | err = logfs_segment_read(inode, page, wc.ofs, e, 0); | |
1694 | if (err) { | |
1695 | logfs_put_write_page(page); | |
1696 | return err; | |
1697 | } | |
1698 | err = logfs_truncate_i0(inode, page, &wc, size); | |
1699 | logfs_put_write_page(page); | |
1700 | if (err) | |
1701 | return err; | |
1702 | ||
1703 | li->li_data[e] = wc.ofs; | |
1704 | } | |
1705 | return 0; | |
1706 | } | |
1707 | ||
1708 | /* FIXME: these need to become per-sb once we support different blocksizes */ | |
1709 | static u64 __logfs_step[] = { | |
1710 | 1, | |
1711 | I1_BLOCKS, | |
1712 | I2_BLOCKS, | |
1713 | I3_BLOCKS, | |
1714 | }; | |
1715 | ||
1716 | static u64 __logfs_start_index[] = { | |
1717 | I0_BLOCKS, | |
1718 | I1_BLOCKS, | |
1719 | I2_BLOCKS, | |
1720 | I3_BLOCKS | |
1721 | }; | |
1722 | ||
1723 | static inline u64 logfs_step(level_t level) | |
1724 | { | |
1725 | return __logfs_step[(__force u8)level]; | |
1726 | } | |
1727 | ||
1728 | static inline u64 logfs_factor(u8 level) | |
1729 | { | |
1730 | return __logfs_step[level] * LOGFS_BLOCKSIZE; | |
1731 | } | |
1732 | ||
1733 | static inline u64 logfs_start_index(level_t level) | |
1734 | { | |
1735 | return __logfs_start_index[(__force u8)level]; | |
1736 | } | |
1737 | ||
1738 | static void logfs_unpack_raw_index(pgoff_t index, u64 *bix, level_t *level) | |
1739 | { | |
1740 | logfs_unpack_index(index, bix, level); | |
1741 | if (*bix <= logfs_start_index(SUBLEVEL(*level))) | |
1742 | *bix = 0; | |
1743 | } | |
1744 | ||
1745 | static int __logfs_truncate_rec(struct inode *inode, struct page *ipage, | |
1746 | struct write_control *this_wc, u64 size) | |
1747 | { | |
1748 | int truncate_happened = 0; | |
1749 | int e, err = 0; | |
1750 | u64 bix, child_bix, next_bix; | |
1751 | level_t level; | |
1752 | struct page *page; | |
1753 | struct write_control child_wc = { /* FIXME: flags */ }; | |
1754 | ||
1755 | logfs_unpack_raw_index(ipage->index, &bix, &level); | |
1756 | err = logfs_segment_read(inode, ipage, this_wc->ofs, bix, level); | |
1757 | if (err) | |
1758 | return err; | |
1759 | ||
1760 | for (e = LOGFS_BLOCK_FACTOR - 1; e >= 0; e--) { | |
1761 | child_bix = bix + e * logfs_step(SUBLEVEL(level)); | |
1762 | next_bix = child_bix + logfs_step(SUBLEVEL(level)); | |
1763 | if (size > next_bix * LOGFS_BLOCKSIZE) | |
1764 | break; | |
1765 | ||
1766 | child_wc.ofs = pure_ofs(block_get_pointer(ipage, e)); | |
1767 | if (!child_wc.ofs) | |
1768 | continue; | |
1769 | ||
1770 | page = logfs_get_write_page(inode, child_bix, SUBLEVEL(level)); | |
1771 | if (!page) | |
1772 | return -ENOMEM; | |
1773 | ||
1774 | if ((__force u8)level > 1) | |
1775 | err = __logfs_truncate_rec(inode, page, &child_wc, size); | |
1776 | else | |
1777 | err = logfs_truncate_i0(inode, page, &child_wc, size); | |
1778 | logfs_put_write_page(page); | |
1779 | if (err) | |
1780 | return err; | |
1781 | ||
1782 | truncate_happened = 1; | |
1783 | alloc_indirect_block(inode, ipage, 0); | |
1784 | block_set_pointer(ipage, e, child_wc.ofs); | |
1785 | } | |
1786 | ||
1787 | if (!truncate_happened) { | |
1788 | printk("ineffectual truncate (%lx, %lx, %llx)\n", inode->i_ino, ipage->index, size); | |
1789 | return 0; | |
1790 | } | |
1791 | ||
1792 | this_wc->flags = WF_DELETE; | |
1793 | if (logfs_block(ipage)->partial) | |
1794 | this_wc->flags |= WF_WRITE; | |
1795 | ||
1796 | return logfs_write_i0(inode, ipage, this_wc); | |
1797 | } | |
1798 | ||
1799 | static int logfs_truncate_rec(struct inode *inode, u64 size) | |
1800 | { | |
1801 | struct logfs_inode *li = logfs_inode(inode); | |
1802 | struct write_control wc = { | |
1803 | .ofs = li->li_data[INDIRECT_INDEX], | |
1804 | }; | |
1805 | struct page *page; | |
1806 | int err; | |
1807 | ||
1808 | alloc_inode_block(inode); | |
1809 | ||
1810 | if (!wc.ofs) | |
1811 | return 0; | |
1812 | ||
1813 | page = logfs_get_write_page(inode, 0, LEVEL(li->li_height)); | |
1814 | if (!page) | |
1815 | return -ENOMEM; | |
1816 | ||
1817 | err = __logfs_truncate_rec(inode, page, &wc, size); | |
1818 | logfs_put_write_page(page); | |
1819 | if (err) | |
1820 | return err; | |
1821 | ||
1822 | if (li->li_data[INDIRECT_INDEX] != wc.ofs) | |
1823 | li->li_data[INDIRECT_INDEX] = wc.ofs; | |
1824 | return 0; | |
1825 | } | |
1826 | ||
1827 | static int __logfs_truncate(struct inode *inode, u64 size) | |
1828 | { | |
1829 | int ret; | |
1830 | ||
1831 | if (size >= logfs_factor(logfs_inode(inode)->li_height)) | |
1832 | return 0; | |
1833 | ||
1834 | ret = logfs_truncate_rec(inode, size); | |
1835 | if (ret) | |
1836 | return ret; | |
1837 | ||
1838 | return logfs_truncate_direct(inode, size); | |
1839 | } | |
1840 | ||
b6349ac8 JE |
1841 | /* |
1842 | * Truncate, by changing the segment file, can consume a fair amount | |
1843 | * of resources. So back off from time to time and do some GC. | |
1844 | * 8 or 2048 blocks should be well within safety limits even if | |
1845 | * every single block resided in a different segment. | |
1846 | */ | |
1847 | #define TRUNCATE_STEP (8 * 1024 * 1024) | |
1848 | int logfs_truncate(struct inode *inode, u64 target) | |
5db53f3e JE |
1849 | { |
1850 | struct super_block *sb = inode->i_sb; | |
b6349ac8 JE |
1851 | u64 size = i_size_read(inode); |
1852 | int err = 0; | |
5db53f3e | 1853 | |
b6349ac8 JE |
1854 | size = ALIGN(size, TRUNCATE_STEP); |
1855 | while (size > target) { | |
1856 | if (size > TRUNCATE_STEP) | |
1857 | size -= TRUNCATE_STEP; | |
1858 | else | |
1859 | size = 0; | |
1860 | if (size < target) | |
1861 | size = target; | |
1862 | ||
1863 | logfs_get_wblocks(sb, NULL, 1); | |
2e531fa0 | 1864 | err = __logfs_truncate(inode, size); |
b6349ac8 JE |
1865 | if (!err) |
1866 | err = __logfs_write_inode(inode, 0); | |
1867 | logfs_put_wblocks(sb, NULL, 1); | |
1868 | } | |
5db53f3e JE |
1869 | |
1870 | if (!err) | |
b6349ac8 | 1871 | err = vmtruncate(inode, target); |
5db53f3e JE |
1872 | |
1873 | /* I don't trust error recovery yet. */ | |
1874 | WARN_ON(err); | |
1875 | return err; | |
1876 | } | |
1877 | ||
1878 | static void move_page_to_inode(struct inode *inode, struct page *page) | |
1879 | { | |
1880 | struct logfs_inode *li = logfs_inode(inode); | |
1881 | struct logfs_block *block = logfs_block(page); | |
1882 | ||
1883 | if (!block) | |
1884 | return; | |
1885 | ||
1886 | log_blockmove("move_page_to_inode(%llx, %llx, %x)\n", | |
1887 | block->ino, block->bix, block->level); | |
1888 | BUG_ON(li->li_block); | |
1889 | block->ops = &inode_block_ops; | |
1890 | block->inode = inode; | |
1891 | li->li_block = block; | |
1892 | ||
1893 | block->page = NULL; | |
1894 | page->private = 0; | |
1895 | ClearPagePrivate(page); | |
1896 | } | |
1897 | ||
1898 | static void move_inode_to_page(struct page *page, struct inode *inode) | |
1899 | { | |
1900 | struct logfs_inode *li = logfs_inode(inode); | |
1901 | struct logfs_block *block = li->li_block; | |
1902 | ||
1903 | if (!block) | |
1904 | return; | |
1905 | ||
1906 | log_blockmove("move_inode_to_page(%llx, %llx, %x)\n", | |
1907 | block->ino, block->bix, block->level); | |
1908 | BUG_ON(PagePrivate(page)); | |
1909 | block->ops = &indirect_block_ops; | |
1910 | block->page = page; | |
1911 | page->private = (unsigned long)block; | |
1912 | SetPagePrivate(page); | |
1913 | ||
1914 | block->inode = NULL; | |
1915 | li->li_block = NULL; | |
1916 | } | |
1917 | ||
1918 | int logfs_read_inode(struct inode *inode) | |
1919 | { | |
1920 | struct super_block *sb = inode->i_sb; | |
1921 | struct logfs_super *super = logfs_super(sb); | |
1922 | struct inode *master_inode = super->s_master_inode; | |
1923 | struct page *page; | |
1924 | struct logfs_disk_inode *di; | |
1925 | u64 ino = inode->i_ino; | |
1926 | ||
1927 | if (ino << sb->s_blocksize_bits > i_size_read(master_inode)) | |
1928 | return -ENODATA; | |
1929 | if (!logfs_exist_block(master_inode, ino)) | |
1930 | return -ENODATA; | |
1931 | ||
1932 | page = read_cache_page(master_inode->i_mapping, ino, | |
1933 | (filler_t *)logfs_readpage, NULL); | |
1934 | if (IS_ERR(page)) | |
1935 | return PTR_ERR(page); | |
1936 | ||
1937 | di = kmap_atomic(page, KM_USER0); | |
1938 | logfs_disk_to_inode(di, inode); | |
1939 | kunmap_atomic(di, KM_USER0); | |
1940 | move_page_to_inode(inode, page); | |
1941 | page_cache_release(page); | |
1942 | return 0; | |
1943 | } | |
1944 | ||
1945 | /* Caller must logfs_put_write_page(page); */ | |
1946 | static struct page *inode_to_page(struct inode *inode) | |
1947 | { | |
1948 | struct inode *master_inode = logfs_super(inode->i_sb)->s_master_inode; | |
1949 | struct logfs_disk_inode *di; | |
1950 | struct page *page; | |
1951 | ||
1952 | BUG_ON(inode->i_ino == LOGFS_INO_MASTER); | |
1953 | ||
1954 | page = logfs_get_write_page(master_inode, inode->i_ino, 0); | |
1955 | if (!page) | |
1956 | return NULL; | |
1957 | ||
1958 | di = kmap_atomic(page, KM_USER0); | |
1959 | logfs_inode_to_disk(inode, di); | |
1960 | kunmap_atomic(di, KM_USER0); | |
1961 | move_inode_to_page(page, inode); | |
1962 | return page; | |
1963 | } | |
1964 | ||
1965 | /* Cheaper version of write_inode. All changes are concealed in | |
1966 | * aliases, which are moved back. No write to the medium happens. | |
1967 | */ | |
1968 | void logfs_clear_inode(struct inode *inode) | |
1969 | { | |
1970 | struct super_block *sb = inode->i_sb; | |
1971 | struct logfs_inode *li = logfs_inode(inode); | |
1972 | struct logfs_block *block = li->li_block; | |
1973 | struct page *page; | |
1974 | ||
1975 | /* Only deleted files may be dirty at this point */ | |
1976 | BUG_ON(inode->i_state & I_DIRTY && inode->i_nlink); | |
1977 | if (!block) | |
1978 | return; | |
1979 | if ((logfs_super(sb)->s_flags & LOGFS_SB_FLAG_SHUTDOWN)) { | |
1980 | block->ops->free_block(inode->i_sb, block); | |
1981 | return; | |
1982 | } | |
1983 | ||
1984 | BUG_ON(inode->i_ino < LOGFS_RESERVED_INOS); | |
1985 | page = inode_to_page(inode); | |
1986 | BUG_ON(!page); /* FIXME: Use emergency page */ | |
1987 | logfs_put_write_page(page); | |
1988 | } | |
1989 | ||
1990 | static int do_write_inode(struct inode *inode) | |
1991 | { | |
1992 | struct super_block *sb = inode->i_sb; | |
1993 | struct inode *master_inode = logfs_super(sb)->s_master_inode; | |
1994 | loff_t size = (inode->i_ino + 1) << inode->i_sb->s_blocksize_bits; | |
1995 | struct page *page; | |
1996 | int err; | |
1997 | ||
1998 | BUG_ON(inode->i_ino == LOGFS_INO_MASTER); | |
1999 | /* FIXME: lock inode */ | |
2000 | ||
2001 | if (i_size_read(master_inode) < size) | |
2002 | i_size_write(master_inode, size); | |
2003 | ||
2004 | /* TODO: Tell vfs this inode is clean now */ | |
2005 | ||
2006 | page = inode_to_page(inode); | |
2007 | if (!page) | |
2008 | return -ENOMEM; | |
2009 | ||
2010 | /* FIXME: transaction is part of logfs_block now. Is that enough? */ | |
2011 | err = logfs_write_buf(master_inode, page, 0); | |
2012 | logfs_put_write_page(page); | |
2013 | return err; | |
2014 | } | |
2015 | ||
2016 | static void logfs_mod_segment_entry(struct super_block *sb, u32 segno, | |
2017 | int write, | |
2018 | void (*change_se)(struct logfs_segment_entry *, long), | |
2019 | long arg) | |
2020 | { | |
2021 | struct logfs_super *super = logfs_super(sb); | |
2022 | struct inode *inode; | |
2023 | struct page *page; | |
2024 | struct logfs_segment_entry *se; | |
2025 | pgoff_t page_no; | |
2026 | int child_no; | |
2027 | ||
2028 | page_no = segno >> (sb->s_blocksize_bits - 3); | |
2029 | child_no = segno & ((sb->s_blocksize >> 3) - 1); | |
2030 | ||
2031 | inode = super->s_segfile_inode; | |
2032 | page = logfs_get_write_page(inode, page_no, 0); | |
2033 | BUG_ON(!page); /* FIXME: We need some reserve page for this case */ | |
2034 | if (!PageUptodate(page)) | |
2035 | logfs_read_block(inode, page, WRITE); | |
2036 | ||
2037 | if (write) | |
2038 | alloc_indirect_block(inode, page, 0); | |
2039 | se = kmap_atomic(page, KM_USER0); | |
2040 | change_se(se + child_no, arg); | |
2041 | if (write) { | |
2042 | logfs_set_alias(sb, logfs_block(page), child_no); | |
2043 | BUG_ON((int)be32_to_cpu(se[child_no].valid) > super->s_segsize); | |
2044 | } | |
2045 | kunmap_atomic(se, KM_USER0); | |
2046 | ||
2047 | logfs_put_write_page(page); | |
2048 | } | |
2049 | ||
2050 | static void __get_segment_entry(struct logfs_segment_entry *se, long _target) | |
2051 | { | |
2052 | struct logfs_segment_entry *target = (void *)_target; | |
2053 | ||
2054 | *target = *se; | |
2055 | } | |
2056 | ||
2057 | void logfs_get_segment_entry(struct super_block *sb, u32 segno, | |
2058 | struct logfs_segment_entry *se) | |
2059 | { | |
2060 | logfs_mod_segment_entry(sb, segno, 0, __get_segment_entry, (long)se); | |
2061 | } | |
2062 | ||
2063 | static void __set_segment_used(struct logfs_segment_entry *se, long increment) | |
2064 | { | |
2065 | u32 valid; | |
2066 | ||
2067 | valid = be32_to_cpu(se->valid); | |
2068 | valid += increment; | |
2069 | se->valid = cpu_to_be32(valid); | |
2070 | } | |
2071 | ||
2072 | void logfs_set_segment_used(struct super_block *sb, u64 ofs, int increment) | |
2073 | { | |
2074 | struct logfs_super *super = logfs_super(sb); | |
2075 | u32 segno = ofs >> super->s_segshift; | |
2076 | ||
2077 | if (!increment) | |
2078 | return; | |
2079 | ||
2080 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_used, increment); | |
2081 | } | |
2082 | ||
2083 | static void __set_segment_erased(struct logfs_segment_entry *se, long ec_level) | |
2084 | { | |
2085 | se->ec_level = cpu_to_be32(ec_level); | |
2086 | } | |
2087 | ||
2088 | void logfs_set_segment_erased(struct super_block *sb, u32 segno, u32 ec, | |
2089 | gc_level_t gc_level) | |
2090 | { | |
2091 | u32 ec_level = ec << 4 | (__force u8)gc_level; | |
2092 | ||
2093 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_erased, ec_level); | |
2094 | } | |
2095 | ||
2096 | static void __set_segment_reserved(struct logfs_segment_entry *se, long ignore) | |
2097 | { | |
2098 | se->valid = cpu_to_be32(RESERVED); | |
2099 | } | |
2100 | ||
2101 | void logfs_set_segment_reserved(struct super_block *sb, u32 segno) | |
2102 | { | |
2103 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_reserved, 0); | |
2104 | } | |
2105 | ||
2106 | static void __set_segment_unreserved(struct logfs_segment_entry *se, | |
2107 | long ec_level) | |
2108 | { | |
2109 | se->valid = 0; | |
2110 | se->ec_level = cpu_to_be32(ec_level); | |
2111 | } | |
2112 | ||
2113 | void logfs_set_segment_unreserved(struct super_block *sb, u32 segno, u32 ec) | |
2114 | { | |
2115 | u32 ec_level = ec << 4; | |
2116 | ||
2117 | logfs_mod_segment_entry(sb, segno, 1, __set_segment_unreserved, | |
2118 | ec_level); | |
2119 | } | |
2120 | ||
2121 | int __logfs_write_inode(struct inode *inode, long flags) | |
2122 | { | |
2123 | struct super_block *sb = inode->i_sb; | |
2124 | int ret; | |
2125 | ||
2126 | logfs_get_wblocks(sb, NULL, flags & WF_LOCK); | |
2127 | ret = do_write_inode(inode); | |
2128 | logfs_put_wblocks(sb, NULL, flags & WF_LOCK); | |
2129 | return ret; | |
2130 | } | |
2131 | ||
2132 | static int do_delete_inode(struct inode *inode) | |
2133 | { | |
2134 | struct super_block *sb = inode->i_sb; | |
2135 | struct inode *master_inode = logfs_super(sb)->s_master_inode; | |
2136 | struct page *page; | |
2137 | int ret; | |
2138 | ||
2139 | page = logfs_get_write_page(master_inode, inode->i_ino, 0); | |
2140 | if (!page) | |
2141 | return -ENOMEM; | |
2142 | ||
2143 | move_inode_to_page(page, inode); | |
2144 | ||
2145 | logfs_get_wblocks(sb, page, 1); | |
2146 | ret = __logfs_delete(master_inode, page); | |
2147 | logfs_put_wblocks(sb, page, 1); | |
2148 | ||
2149 | logfs_put_write_page(page); | |
2150 | return ret; | |
2151 | } | |
2152 | ||
2153 | /* | |
2154 | * ZOMBIE inodes have already been deleted before and should remain dead, | |
2155 | * if it weren't for valid checking. No need to kill them again here. | |
2156 | */ | |
2157 | void logfs_delete_inode(struct inode *inode) | |
2158 | { | |
2159 | struct logfs_inode *li = logfs_inode(inode); | |
2160 | ||
2161 | if (!(li->li_flags & LOGFS_IF_ZOMBIE)) { | |
2162 | li->li_flags |= LOGFS_IF_ZOMBIE; | |
2163 | if (i_size_read(inode) > 0) | |
2164 | logfs_truncate(inode, 0); | |
2165 | do_delete_inode(inode); | |
2166 | } | |
2167 | truncate_inode_pages(&inode->i_data, 0); | |
2168 | clear_inode(inode); | |
2169 | } | |
2170 | ||
2171 | void btree_write_block(struct logfs_block *block) | |
2172 | { | |
2173 | struct inode *inode; | |
2174 | struct page *page; | |
2175 | int err, cookie; | |
2176 | ||
2177 | inode = logfs_safe_iget(block->sb, block->ino, &cookie); | |
2178 | page = logfs_get_write_page(inode, block->bix, block->level); | |
2179 | ||
2180 | err = logfs_readpage_nolock(page); | |
2181 | BUG_ON(err); | |
2182 | BUG_ON(!PagePrivate(page)); | |
2183 | BUG_ON(logfs_block(page) != block); | |
2184 | err = __logfs_write_buf(inode, page, 0); | |
2185 | BUG_ON(err); | |
2186 | BUG_ON(PagePrivate(page) || page->private); | |
2187 | ||
2188 | logfs_put_write_page(page); | |
2189 | logfs_safe_iput(inode, cookie); | |
2190 | } | |
2191 | ||
2192 | /** | |
2193 | * logfs_inode_write - write inode or dentry objects | |
2194 | * | |
2195 | * @inode: parent inode (ifile or directory) | |
2196 | * @buf: object to write (inode or dentry) | |
2197 | * @n: object size | |
2198 | * @_pos: object number (file position in blocks/objects) | |
2199 | * @flags: write flags | |
2200 | * @lock: 0 if write lock is already taken, 1 otherwise | |
2201 | * @shadow_tree: shadow below this inode | |
2202 | * | |
2203 | * FIXME: All caller of this put a 200-300 byte variable on the stack, | |
2204 | * only to call here and do a memcpy from that stack variable. A good | |
2205 | * example of wasted performance and stack space. | |
2206 | */ | |
2207 | int logfs_inode_write(struct inode *inode, const void *buf, size_t count, | |
2208 | loff_t bix, long flags, struct shadow_tree *shadow_tree) | |
2209 | { | |
2210 | loff_t pos = bix << inode->i_sb->s_blocksize_bits; | |
2211 | int err; | |
2212 | struct page *page; | |
2213 | void *pagebuf; | |
2214 | ||
2215 | BUG_ON(pos & (LOGFS_BLOCKSIZE-1)); | |
2216 | BUG_ON(count > LOGFS_BLOCKSIZE); | |
2217 | page = logfs_get_write_page(inode, bix, 0); | |
2218 | if (!page) | |
2219 | return -ENOMEM; | |
2220 | ||
2221 | pagebuf = kmap_atomic(page, KM_USER0); | |
2222 | memcpy(pagebuf, buf, count); | |
2223 | flush_dcache_page(page); | |
2224 | kunmap_atomic(pagebuf, KM_USER0); | |
2225 | ||
2226 | if (i_size_read(inode) < pos + LOGFS_BLOCKSIZE) | |
2227 | i_size_write(inode, pos + LOGFS_BLOCKSIZE); | |
2228 | ||
2229 | err = logfs_write_buf(inode, page, flags); | |
2230 | logfs_put_write_page(page); | |
2231 | return err; | |
2232 | } | |
2233 | ||
2234 | int logfs_open_segfile(struct super_block *sb) | |
2235 | { | |
2236 | struct logfs_super *super = logfs_super(sb); | |
2237 | struct inode *inode; | |
2238 | ||
2239 | inode = logfs_read_meta_inode(sb, LOGFS_INO_SEGFILE); | |
2240 | if (IS_ERR(inode)) | |
2241 | return PTR_ERR(inode); | |
2242 | super->s_segfile_inode = inode; | |
2243 | return 0; | |
2244 | } | |
2245 | ||
2246 | int logfs_init_rw(struct super_block *sb) | |
2247 | { | |
2248 | struct logfs_super *super = logfs_super(sb); | |
2249 | int min_fill = 3 * super->s_no_blocks; | |
2250 | ||
2251 | INIT_LIST_HEAD(&super->s_object_alias); | |
2252 | mutex_init(&super->s_write_mutex); | |
2253 | super->s_block_pool = mempool_create_kmalloc_pool(min_fill, | |
2254 | sizeof(struct logfs_block)); | |
2255 | super->s_shadow_pool = mempool_create_kmalloc_pool(min_fill, | |
2256 | sizeof(struct logfs_shadow)); | |
2257 | return 0; | |
2258 | } | |
2259 | ||
2260 | void logfs_cleanup_rw(struct super_block *sb) | |
2261 | { | |
2262 | struct logfs_super *super = logfs_super(sb); | |
2263 | ||
2264 | destroy_meta_inode(super->s_segfile_inode); | |
1f1b0008 JE |
2265 | logfs_mempool_destroy(super->s_block_pool); |
2266 | logfs_mempool_destroy(super->s_shadow_pool); | |
5db53f3e | 2267 | } |