Loading...
1/*
2 * linux/fs/ext2/inode.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@dcs.ed.ac.uk), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
21 *
22 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
23 */
24
25#include <linux/time.h>
26#include <linux/highuid.h>
27#include <linux/pagemap.h>
28#include <linux/quotaops.h>
29#include <linux/writeback.h>
30#include <linux/buffer_head.h>
31#include <linux/mpage.h>
32#include <linux/fiemap.h>
33#include <linux/namei.h>
34#include "ext2.h"
35#include "acl.h"
36#include "xip.h"
37
38static int __ext2_write_inode(struct inode *inode, int do_sync);
39
40/*
41 * Test whether an inode is a fast symlink.
42 */
43static inline int ext2_inode_is_fast_symlink(struct inode *inode)
44{
45 int ea_blocks = EXT2_I(inode)->i_file_acl ?
46 (inode->i_sb->s_blocksize >> 9) : 0;
47
48 return (S_ISLNK(inode->i_mode) &&
49 inode->i_blocks - ea_blocks == 0);
50}
51
52static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
53
54static void ext2_write_failed(struct address_space *mapping, loff_t to)
55{
56 struct inode *inode = mapping->host;
57
58 if (to > inode->i_size) {
59 truncate_pagecache(inode, to, inode->i_size);
60 ext2_truncate_blocks(inode, inode->i_size);
61 }
62}
63
64/*
65 * Called at the last iput() if i_nlink is zero.
66 */
67void ext2_evict_inode(struct inode * inode)
68{
69 struct ext2_block_alloc_info *rsv;
70 int want_delete = 0;
71
72 if (!inode->i_nlink && !is_bad_inode(inode)) {
73 want_delete = 1;
74 dquot_initialize(inode);
75 } else {
76 dquot_drop(inode);
77 }
78
79 truncate_inode_pages(&inode->i_data, 0);
80
81 if (want_delete) {
82 /* set dtime */
83 EXT2_I(inode)->i_dtime = get_seconds();
84 mark_inode_dirty(inode);
85 __ext2_write_inode(inode, inode_needs_sync(inode));
86 /* truncate to 0 */
87 inode->i_size = 0;
88 if (inode->i_blocks)
89 ext2_truncate_blocks(inode, 0);
90 }
91
92 invalidate_inode_buffers(inode);
93 clear_inode(inode);
94
95 ext2_discard_reservation(inode);
96 rsv = EXT2_I(inode)->i_block_alloc_info;
97 EXT2_I(inode)->i_block_alloc_info = NULL;
98 if (unlikely(rsv))
99 kfree(rsv);
100
101 if (want_delete)
102 ext2_free_inode(inode);
103}
104
105typedef struct {
106 __le32 *p;
107 __le32 key;
108 struct buffer_head *bh;
109} Indirect;
110
111static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
112{
113 p->key = *(p->p = v);
114 p->bh = bh;
115}
116
117static inline int verify_chain(Indirect *from, Indirect *to)
118{
119 while (from <= to && from->key == *from->p)
120 from++;
121 return (from > to);
122}
123
124/**
125 * ext2_block_to_path - parse the block number into array of offsets
126 * @inode: inode in question (we are only interested in its superblock)
127 * @i_block: block number to be parsed
128 * @offsets: array to store the offsets in
129 * @boundary: set this non-zero if the referred-to block is likely to be
130 * followed (on disk) by an indirect block.
131 * To store the locations of file's data ext2 uses a data structure common
132 * for UNIX filesystems - tree of pointers anchored in the inode, with
133 * data blocks at leaves and indirect blocks in intermediate nodes.
134 * This function translates the block number into path in that tree -
135 * return value is the path length and @offsets[n] is the offset of
136 * pointer to (n+1)th node in the nth one. If @block is out of range
137 * (negative or too large) warning is printed and zero returned.
138 *
139 * Note: function doesn't find node addresses, so no IO is needed. All
140 * we need to know is the capacity of indirect blocks (taken from the
141 * inode->i_sb).
142 */
143
144/*
145 * Portability note: the last comparison (check that we fit into triple
146 * indirect block) is spelled differently, because otherwise on an
147 * architecture with 32-bit longs and 8Kb pages we might get into trouble
148 * if our filesystem had 8Kb blocks. We might use long long, but that would
149 * kill us on x86. Oh, well, at least the sign propagation does not matter -
150 * i_block would have to be negative in the very beginning, so we would not
151 * get there at all.
152 */
153
154static int ext2_block_to_path(struct inode *inode,
155 long i_block, int offsets[4], int *boundary)
156{
157 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
158 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
159 const long direct_blocks = EXT2_NDIR_BLOCKS,
160 indirect_blocks = ptrs,
161 double_blocks = (1 << (ptrs_bits * 2));
162 int n = 0;
163 int final = 0;
164
165 if (i_block < 0) {
166 ext2_msg(inode->i_sb, KERN_WARNING,
167 "warning: %s: block < 0", __func__);
168 } else if (i_block < direct_blocks) {
169 offsets[n++] = i_block;
170 final = direct_blocks;
171 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
172 offsets[n++] = EXT2_IND_BLOCK;
173 offsets[n++] = i_block;
174 final = ptrs;
175 } else if ((i_block -= indirect_blocks) < double_blocks) {
176 offsets[n++] = EXT2_DIND_BLOCK;
177 offsets[n++] = i_block >> ptrs_bits;
178 offsets[n++] = i_block & (ptrs - 1);
179 final = ptrs;
180 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
181 offsets[n++] = EXT2_TIND_BLOCK;
182 offsets[n++] = i_block >> (ptrs_bits * 2);
183 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
184 offsets[n++] = i_block & (ptrs - 1);
185 final = ptrs;
186 } else {
187 ext2_msg(inode->i_sb, KERN_WARNING,
188 "warning: %s: block is too big", __func__);
189 }
190 if (boundary)
191 *boundary = final - 1 - (i_block & (ptrs - 1));
192
193 return n;
194}
195
196/**
197 * ext2_get_branch - read the chain of indirect blocks leading to data
198 * @inode: inode in question
199 * @depth: depth of the chain (1 - direct pointer, etc.)
200 * @offsets: offsets of pointers in inode/indirect blocks
201 * @chain: place to store the result
202 * @err: here we store the error value
203 *
204 * Function fills the array of triples <key, p, bh> and returns %NULL
205 * if everything went OK or the pointer to the last filled triple
206 * (incomplete one) otherwise. Upon the return chain[i].key contains
207 * the number of (i+1)-th block in the chain (as it is stored in memory,
208 * i.e. little-endian 32-bit), chain[i].p contains the address of that
209 * number (it points into struct inode for i==0 and into the bh->b_data
210 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
211 * block for i>0 and NULL for i==0. In other words, it holds the block
212 * numbers of the chain, addresses they were taken from (and where we can
213 * verify that chain did not change) and buffer_heads hosting these
214 * numbers.
215 *
216 * Function stops when it stumbles upon zero pointer (absent block)
217 * (pointer to last triple returned, *@err == 0)
218 * or when it gets an IO error reading an indirect block
219 * (ditto, *@err == -EIO)
220 * or when it notices that chain had been changed while it was reading
221 * (ditto, *@err == -EAGAIN)
222 * or when it reads all @depth-1 indirect blocks successfully and finds
223 * the whole chain, all way to the data (returns %NULL, *err == 0).
224 */
225static Indirect *ext2_get_branch(struct inode *inode,
226 int depth,
227 int *offsets,
228 Indirect chain[4],
229 int *err)
230{
231 struct super_block *sb = inode->i_sb;
232 Indirect *p = chain;
233 struct buffer_head *bh;
234
235 *err = 0;
236 /* i_data is not going away, no lock needed */
237 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
238 if (!p->key)
239 goto no_block;
240 while (--depth) {
241 bh = sb_bread(sb, le32_to_cpu(p->key));
242 if (!bh)
243 goto failure;
244 read_lock(&EXT2_I(inode)->i_meta_lock);
245 if (!verify_chain(chain, p))
246 goto changed;
247 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
248 read_unlock(&EXT2_I(inode)->i_meta_lock);
249 if (!p->key)
250 goto no_block;
251 }
252 return NULL;
253
254changed:
255 read_unlock(&EXT2_I(inode)->i_meta_lock);
256 brelse(bh);
257 *err = -EAGAIN;
258 goto no_block;
259failure:
260 *err = -EIO;
261no_block:
262 return p;
263}
264
265/**
266 * ext2_find_near - find a place for allocation with sufficient locality
267 * @inode: owner
268 * @ind: descriptor of indirect block.
269 *
270 * This function returns the preferred place for block allocation.
271 * It is used when heuristic for sequential allocation fails.
272 * Rules are:
273 * + if there is a block to the left of our position - allocate near it.
274 * + if pointer will live in indirect block - allocate near that block.
275 * + if pointer will live in inode - allocate in the same cylinder group.
276 *
277 * In the latter case we colour the starting block by the callers PID to
278 * prevent it from clashing with concurrent allocations for a different inode
279 * in the same block group. The PID is used here so that functionally related
280 * files will be close-by on-disk.
281 *
282 * Caller must make sure that @ind is valid and will stay that way.
283 */
284
285static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
286{
287 struct ext2_inode_info *ei = EXT2_I(inode);
288 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
289 __le32 *p;
290 ext2_fsblk_t bg_start;
291 ext2_fsblk_t colour;
292
293 /* Try to find previous block */
294 for (p = ind->p - 1; p >= start; p--)
295 if (*p)
296 return le32_to_cpu(*p);
297
298 /* No such thing, so let's try location of indirect block */
299 if (ind->bh)
300 return ind->bh->b_blocknr;
301
302 /*
303 * It is going to be referred from inode itself? OK, just put it into
304 * the same cylinder group then.
305 */
306 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
307 colour = (current->pid % 16) *
308 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
309 return bg_start + colour;
310}
311
312/**
313 * ext2_find_goal - find a preferred place for allocation.
314 * @inode: owner
315 * @block: block we want
316 * @partial: pointer to the last triple within a chain
317 *
318 * Returns preferred place for a block (the goal).
319 */
320
321static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
322 Indirect *partial)
323{
324 struct ext2_block_alloc_info *block_i;
325
326 block_i = EXT2_I(inode)->i_block_alloc_info;
327
328 /*
329 * try the heuristic for sequential allocation,
330 * failing that at least try to get decent locality.
331 */
332 if (block_i && (block == block_i->last_alloc_logical_block + 1)
333 && (block_i->last_alloc_physical_block != 0)) {
334 return block_i->last_alloc_physical_block + 1;
335 }
336
337 return ext2_find_near(inode, partial);
338}
339
340/**
341 * ext2_blks_to_allocate: Look up the block map and count the number
342 * of direct blocks need to be allocated for the given branch.
343 *
344 * @branch: chain of indirect blocks
345 * @k: number of blocks need for indirect blocks
346 * @blks: number of data blocks to be mapped.
347 * @blocks_to_boundary: the offset in the indirect block
348 *
349 * return the total number of blocks to be allocate, including the
350 * direct and indirect blocks.
351 */
352static int
353ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
354 int blocks_to_boundary)
355{
356 unsigned long count = 0;
357
358 /*
359 * Simple case, [t,d]Indirect block(s) has not allocated yet
360 * then it's clear blocks on that path have not allocated
361 */
362 if (k > 0) {
363 /* right now don't hanel cross boundary allocation */
364 if (blks < blocks_to_boundary + 1)
365 count += blks;
366 else
367 count += blocks_to_boundary + 1;
368 return count;
369 }
370
371 count++;
372 while (count < blks && count <= blocks_to_boundary
373 && le32_to_cpu(*(branch[0].p + count)) == 0) {
374 count++;
375 }
376 return count;
377}
378
379/**
380 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
381 * @indirect_blks: the number of blocks need to allocate for indirect
382 * blocks
383 *
384 * @new_blocks: on return it will store the new block numbers for
385 * the indirect blocks(if needed) and the first direct block,
386 * @blks: on return it will store the total number of allocated
387 * direct blocks
388 */
389static int ext2_alloc_blocks(struct inode *inode,
390 ext2_fsblk_t goal, int indirect_blks, int blks,
391 ext2_fsblk_t new_blocks[4], int *err)
392{
393 int target, i;
394 unsigned long count = 0;
395 int index = 0;
396 ext2_fsblk_t current_block = 0;
397 int ret = 0;
398
399 /*
400 * Here we try to allocate the requested multiple blocks at once,
401 * on a best-effort basis.
402 * To build a branch, we should allocate blocks for
403 * the indirect blocks(if not allocated yet), and at least
404 * the first direct block of this branch. That's the
405 * minimum number of blocks need to allocate(required)
406 */
407 target = blks + indirect_blks;
408
409 while (1) {
410 count = target;
411 /* allocating blocks for indirect blocks and direct blocks */
412 current_block = ext2_new_blocks(inode,goal,&count,err);
413 if (*err)
414 goto failed_out;
415
416 target -= count;
417 /* allocate blocks for indirect blocks */
418 while (index < indirect_blks && count) {
419 new_blocks[index++] = current_block++;
420 count--;
421 }
422
423 if (count > 0)
424 break;
425 }
426
427 /* save the new block number for the first direct block */
428 new_blocks[index] = current_block;
429
430 /* total number of blocks allocated for direct blocks */
431 ret = count;
432 *err = 0;
433 return ret;
434failed_out:
435 for (i = 0; i <index; i++)
436 ext2_free_blocks(inode, new_blocks[i], 1);
437 if (index)
438 mark_inode_dirty(inode);
439 return ret;
440}
441
442/**
443 * ext2_alloc_branch - allocate and set up a chain of blocks.
444 * @inode: owner
445 * @num: depth of the chain (number of blocks to allocate)
446 * @offsets: offsets (in the blocks) to store the pointers to next.
447 * @branch: place to store the chain in.
448 *
449 * This function allocates @num blocks, zeroes out all but the last one,
450 * links them into chain and (if we are synchronous) writes them to disk.
451 * In other words, it prepares a branch that can be spliced onto the
452 * inode. It stores the information about that chain in the branch[], in
453 * the same format as ext2_get_branch() would do. We are calling it after
454 * we had read the existing part of chain and partial points to the last
455 * triple of that (one with zero ->key). Upon the exit we have the same
456 * picture as after the successful ext2_get_block(), except that in one
457 * place chain is disconnected - *branch->p is still zero (we did not
458 * set the last link), but branch->key contains the number that should
459 * be placed into *branch->p to fill that gap.
460 *
461 * If allocation fails we free all blocks we've allocated (and forget
462 * their buffer_heads) and return the error value the from failed
463 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
464 * as described above and return 0.
465 */
466
467static int ext2_alloc_branch(struct inode *inode,
468 int indirect_blks, int *blks, ext2_fsblk_t goal,
469 int *offsets, Indirect *branch)
470{
471 int blocksize = inode->i_sb->s_blocksize;
472 int i, n = 0;
473 int err = 0;
474 struct buffer_head *bh;
475 int num;
476 ext2_fsblk_t new_blocks[4];
477 ext2_fsblk_t current_block;
478
479 num = ext2_alloc_blocks(inode, goal, indirect_blks,
480 *blks, new_blocks, &err);
481 if (err)
482 return err;
483
484 branch[0].key = cpu_to_le32(new_blocks[0]);
485 /*
486 * metadata blocks and data blocks are allocated.
487 */
488 for (n = 1; n <= indirect_blks; n++) {
489 /*
490 * Get buffer_head for parent block, zero it out
491 * and set the pointer to new one, then send
492 * parent to disk.
493 */
494 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
495 branch[n].bh = bh;
496 lock_buffer(bh);
497 memset(bh->b_data, 0, blocksize);
498 branch[n].p = (__le32 *) bh->b_data + offsets[n];
499 branch[n].key = cpu_to_le32(new_blocks[n]);
500 *branch[n].p = branch[n].key;
501 if ( n == indirect_blks) {
502 current_block = new_blocks[n];
503 /*
504 * End of chain, update the last new metablock of
505 * the chain to point to the new allocated
506 * data blocks numbers
507 */
508 for (i=1; i < num; i++)
509 *(branch[n].p + i) = cpu_to_le32(++current_block);
510 }
511 set_buffer_uptodate(bh);
512 unlock_buffer(bh);
513 mark_buffer_dirty_inode(bh, inode);
514 /* We used to sync bh here if IS_SYNC(inode).
515 * But we now rely upon generic_write_sync()
516 * and b_inode_buffers. But not for directories.
517 */
518 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
519 sync_dirty_buffer(bh);
520 }
521 *blks = num;
522 return err;
523}
524
525/**
526 * ext2_splice_branch - splice the allocated branch onto inode.
527 * @inode: owner
528 * @block: (logical) number of block we are adding
529 * @where: location of missing link
530 * @num: number of indirect blocks we are adding
531 * @blks: number of direct blocks we are adding
532 *
533 * This function fills the missing link and does all housekeeping needed in
534 * inode (->i_blocks, etc.). In case of success we end up with the full
535 * chain to new block and return 0.
536 */
537static void ext2_splice_branch(struct inode *inode,
538 long block, Indirect *where, int num, int blks)
539{
540 int i;
541 struct ext2_block_alloc_info *block_i;
542 ext2_fsblk_t current_block;
543
544 block_i = EXT2_I(inode)->i_block_alloc_info;
545
546 /* XXX LOCKING probably should have i_meta_lock ?*/
547 /* That's it */
548
549 *where->p = where->key;
550
551 /*
552 * Update the host buffer_head or inode to point to more just allocated
553 * direct blocks blocks
554 */
555 if (num == 0 && blks > 1) {
556 current_block = le32_to_cpu(where->key) + 1;
557 for (i = 1; i < blks; i++)
558 *(where->p + i ) = cpu_to_le32(current_block++);
559 }
560
561 /*
562 * update the most recently allocated logical & physical block
563 * in i_block_alloc_info, to assist find the proper goal block for next
564 * allocation
565 */
566 if (block_i) {
567 block_i->last_alloc_logical_block = block + blks - 1;
568 block_i->last_alloc_physical_block =
569 le32_to_cpu(where[num].key) + blks - 1;
570 }
571
572 /* We are done with atomic stuff, now do the rest of housekeeping */
573
574 /* had we spliced it onto indirect block? */
575 if (where->bh)
576 mark_buffer_dirty_inode(where->bh, inode);
577
578 inode->i_ctime = CURRENT_TIME_SEC;
579 mark_inode_dirty(inode);
580}
581
582/*
583 * Allocation strategy is simple: if we have to allocate something, we will
584 * have to go the whole way to leaf. So let's do it before attaching anything
585 * to tree, set linkage between the newborn blocks, write them if sync is
586 * required, recheck the path, free and repeat if check fails, otherwise
587 * set the last missing link (that will protect us from any truncate-generated
588 * removals - all blocks on the path are immune now) and possibly force the
589 * write on the parent block.
590 * That has a nice additional property: no special recovery from the failed
591 * allocations is needed - we simply release blocks and do not touch anything
592 * reachable from inode.
593 *
594 * `handle' can be NULL if create == 0.
595 *
596 * return > 0, # of blocks mapped or allocated.
597 * return = 0, if plain lookup failed.
598 * return < 0, error case.
599 */
600static int ext2_get_blocks(struct inode *inode,
601 sector_t iblock, unsigned long maxblocks,
602 struct buffer_head *bh_result,
603 int create)
604{
605 int err = -EIO;
606 int offsets[4];
607 Indirect chain[4];
608 Indirect *partial;
609 ext2_fsblk_t goal;
610 int indirect_blks;
611 int blocks_to_boundary = 0;
612 int depth;
613 struct ext2_inode_info *ei = EXT2_I(inode);
614 int count = 0;
615 ext2_fsblk_t first_block = 0;
616
617 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
618
619 if (depth == 0)
620 return (err);
621
622 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
623 /* Simplest case - block found, no allocation needed */
624 if (!partial) {
625 first_block = le32_to_cpu(chain[depth - 1].key);
626 clear_buffer_new(bh_result); /* What's this do? */
627 count++;
628 /*map more blocks*/
629 while (count < maxblocks && count <= blocks_to_boundary) {
630 ext2_fsblk_t blk;
631
632 if (!verify_chain(chain, chain + depth - 1)) {
633 /*
634 * Indirect block might be removed by
635 * truncate while we were reading it.
636 * Handling of that case: forget what we've
637 * got now, go to reread.
638 */
639 err = -EAGAIN;
640 count = 0;
641 break;
642 }
643 blk = le32_to_cpu(*(chain[depth-1].p + count));
644 if (blk == first_block + count)
645 count++;
646 else
647 break;
648 }
649 if (err != -EAGAIN)
650 goto got_it;
651 }
652
653 /* Next simple case - plain lookup or failed read of indirect block */
654 if (!create || err == -EIO)
655 goto cleanup;
656
657 mutex_lock(&ei->truncate_mutex);
658 /*
659 * If the indirect block is missing while we are reading
660 * the chain(ext2_get_branch() returns -EAGAIN err), or
661 * if the chain has been changed after we grab the semaphore,
662 * (either because another process truncated this branch, or
663 * another get_block allocated this branch) re-grab the chain to see if
664 * the request block has been allocated or not.
665 *
666 * Since we already block the truncate/other get_block
667 * at this point, we will have the current copy of the chain when we
668 * splice the branch into the tree.
669 */
670 if (err == -EAGAIN || !verify_chain(chain, partial)) {
671 while (partial > chain) {
672 brelse(partial->bh);
673 partial--;
674 }
675 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
676 if (!partial) {
677 count++;
678 mutex_unlock(&ei->truncate_mutex);
679 if (err)
680 goto cleanup;
681 clear_buffer_new(bh_result);
682 goto got_it;
683 }
684 }
685
686 /*
687 * Okay, we need to do block allocation. Lazily initialize the block
688 * allocation info here if necessary
689 */
690 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
691 ext2_init_block_alloc_info(inode);
692
693 goal = ext2_find_goal(inode, iblock, partial);
694
695 /* the number of blocks need to allocate for [d,t]indirect blocks */
696 indirect_blks = (chain + depth) - partial - 1;
697 /*
698 * Next look up the indirect map to count the totoal number of
699 * direct blocks to allocate for this branch.
700 */
701 count = ext2_blks_to_allocate(partial, indirect_blks,
702 maxblocks, blocks_to_boundary);
703 /*
704 * XXX ???? Block out ext2_truncate while we alter the tree
705 */
706 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
707 offsets + (partial - chain), partial);
708
709 if (err) {
710 mutex_unlock(&ei->truncate_mutex);
711 goto cleanup;
712 }
713
714 if (ext2_use_xip(inode->i_sb)) {
715 /*
716 * we need to clear the block
717 */
718 err = ext2_clear_xip_target (inode,
719 le32_to_cpu(chain[depth-1].key));
720 if (err) {
721 mutex_unlock(&ei->truncate_mutex);
722 goto cleanup;
723 }
724 }
725
726 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
727 mutex_unlock(&ei->truncate_mutex);
728 set_buffer_new(bh_result);
729got_it:
730 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
731 if (count > blocks_to_boundary)
732 set_buffer_boundary(bh_result);
733 err = count;
734 /* Clean up and exit */
735 partial = chain + depth - 1; /* the whole chain */
736cleanup:
737 while (partial > chain) {
738 brelse(partial->bh);
739 partial--;
740 }
741 return err;
742}
743
744int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
745{
746 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
747 int ret = ext2_get_blocks(inode, iblock, max_blocks,
748 bh_result, create);
749 if (ret > 0) {
750 bh_result->b_size = (ret << inode->i_blkbits);
751 ret = 0;
752 }
753 return ret;
754
755}
756
757int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
758 u64 start, u64 len)
759{
760 return generic_block_fiemap(inode, fieinfo, start, len,
761 ext2_get_block);
762}
763
764static int ext2_writepage(struct page *page, struct writeback_control *wbc)
765{
766 return block_write_full_page(page, ext2_get_block, wbc);
767}
768
769static int ext2_readpage(struct file *file, struct page *page)
770{
771 return mpage_readpage(page, ext2_get_block);
772}
773
774static int
775ext2_readpages(struct file *file, struct address_space *mapping,
776 struct list_head *pages, unsigned nr_pages)
777{
778 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
779}
780
781static int
782ext2_write_begin(struct file *file, struct address_space *mapping,
783 loff_t pos, unsigned len, unsigned flags,
784 struct page **pagep, void **fsdata)
785{
786 int ret;
787
788 ret = block_write_begin(mapping, pos, len, flags, pagep,
789 ext2_get_block);
790 if (ret < 0)
791 ext2_write_failed(mapping, pos + len);
792 return ret;
793}
794
795static int ext2_write_end(struct file *file, struct address_space *mapping,
796 loff_t pos, unsigned len, unsigned copied,
797 struct page *page, void *fsdata)
798{
799 int ret;
800
801 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
802 if (ret < len)
803 ext2_write_failed(mapping, pos + len);
804 return ret;
805}
806
807static int
808ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
809 loff_t pos, unsigned len, unsigned flags,
810 struct page **pagep, void **fsdata)
811{
812 int ret;
813
814 ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
815 ext2_get_block);
816 if (ret < 0)
817 ext2_write_failed(mapping, pos + len);
818 return ret;
819}
820
821static int ext2_nobh_writepage(struct page *page,
822 struct writeback_control *wbc)
823{
824 return nobh_writepage(page, ext2_get_block, wbc);
825}
826
827static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
828{
829 return generic_block_bmap(mapping,block,ext2_get_block);
830}
831
832static ssize_t
833ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
834 loff_t offset, unsigned long nr_segs)
835{
836 struct file *file = iocb->ki_filp;
837 struct address_space *mapping = file->f_mapping;
838 struct inode *inode = mapping->host;
839 ssize_t ret;
840
841 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
842 ext2_get_block);
843 if (ret < 0 && (rw & WRITE))
844 ext2_write_failed(mapping, offset + iov_length(iov, nr_segs));
845 return ret;
846}
847
848static int
849ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
850{
851 return mpage_writepages(mapping, wbc, ext2_get_block);
852}
853
854const struct address_space_operations ext2_aops = {
855 .readpage = ext2_readpage,
856 .readpages = ext2_readpages,
857 .writepage = ext2_writepage,
858 .write_begin = ext2_write_begin,
859 .write_end = ext2_write_end,
860 .bmap = ext2_bmap,
861 .direct_IO = ext2_direct_IO,
862 .writepages = ext2_writepages,
863 .migratepage = buffer_migrate_page,
864 .is_partially_uptodate = block_is_partially_uptodate,
865 .error_remove_page = generic_error_remove_page,
866};
867
868const struct address_space_operations ext2_aops_xip = {
869 .bmap = ext2_bmap,
870 .get_xip_mem = ext2_get_xip_mem,
871};
872
873const struct address_space_operations ext2_nobh_aops = {
874 .readpage = ext2_readpage,
875 .readpages = ext2_readpages,
876 .writepage = ext2_nobh_writepage,
877 .write_begin = ext2_nobh_write_begin,
878 .write_end = nobh_write_end,
879 .bmap = ext2_bmap,
880 .direct_IO = ext2_direct_IO,
881 .writepages = ext2_writepages,
882 .migratepage = buffer_migrate_page,
883 .error_remove_page = generic_error_remove_page,
884};
885
886/*
887 * Probably it should be a library function... search for first non-zero word
888 * or memcmp with zero_page, whatever is better for particular architecture.
889 * Linus?
890 */
891static inline int all_zeroes(__le32 *p, __le32 *q)
892{
893 while (p < q)
894 if (*p++)
895 return 0;
896 return 1;
897}
898
899/**
900 * ext2_find_shared - find the indirect blocks for partial truncation.
901 * @inode: inode in question
902 * @depth: depth of the affected branch
903 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
904 * @chain: place to store the pointers to partial indirect blocks
905 * @top: place to the (detached) top of branch
906 *
907 * This is a helper function used by ext2_truncate().
908 *
909 * When we do truncate() we may have to clean the ends of several indirect
910 * blocks but leave the blocks themselves alive. Block is partially
911 * truncated if some data below the new i_size is referred from it (and
912 * it is on the path to the first completely truncated data block, indeed).
913 * We have to free the top of that path along with everything to the right
914 * of the path. Since no allocation past the truncation point is possible
915 * until ext2_truncate() finishes, we may safely do the latter, but top
916 * of branch may require special attention - pageout below the truncation
917 * point might try to populate it.
918 *
919 * We atomically detach the top of branch from the tree, store the block
920 * number of its root in *@top, pointers to buffer_heads of partially
921 * truncated blocks - in @chain[].bh and pointers to their last elements
922 * that should not be removed - in @chain[].p. Return value is the pointer
923 * to last filled element of @chain.
924 *
925 * The work left to caller to do the actual freeing of subtrees:
926 * a) free the subtree starting from *@top
927 * b) free the subtrees whose roots are stored in
928 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
929 * c) free the subtrees growing from the inode past the @chain[0].p
930 * (no partially truncated stuff there).
931 */
932
933static Indirect *ext2_find_shared(struct inode *inode,
934 int depth,
935 int offsets[4],
936 Indirect chain[4],
937 __le32 *top)
938{
939 Indirect *partial, *p;
940 int k, err;
941
942 *top = 0;
943 for (k = depth; k > 1 && !offsets[k-1]; k--)
944 ;
945 partial = ext2_get_branch(inode, k, offsets, chain, &err);
946 if (!partial)
947 partial = chain + k-1;
948 /*
949 * If the branch acquired continuation since we've looked at it -
950 * fine, it should all survive and (new) top doesn't belong to us.
951 */
952 write_lock(&EXT2_I(inode)->i_meta_lock);
953 if (!partial->key && *partial->p) {
954 write_unlock(&EXT2_I(inode)->i_meta_lock);
955 goto no_top;
956 }
957 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
958 ;
959 /*
960 * OK, we've found the last block that must survive. The rest of our
961 * branch should be detached before unlocking. However, if that rest
962 * of branch is all ours and does not grow immediately from the inode
963 * it's easier to cheat and just decrement partial->p.
964 */
965 if (p == chain + k - 1 && p > chain) {
966 p->p--;
967 } else {
968 *top = *p->p;
969 *p->p = 0;
970 }
971 write_unlock(&EXT2_I(inode)->i_meta_lock);
972
973 while(partial > p)
974 {
975 brelse(partial->bh);
976 partial--;
977 }
978no_top:
979 return partial;
980}
981
982/**
983 * ext2_free_data - free a list of data blocks
984 * @inode: inode we are dealing with
985 * @p: array of block numbers
986 * @q: points immediately past the end of array
987 *
988 * We are freeing all blocks referred from that array (numbers are
989 * stored as little-endian 32-bit) and updating @inode->i_blocks
990 * appropriately.
991 */
992static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
993{
994 unsigned long block_to_free = 0, count = 0;
995 unsigned long nr;
996
997 for ( ; p < q ; p++) {
998 nr = le32_to_cpu(*p);
999 if (nr) {
1000 *p = 0;
1001 /* accumulate blocks to free if they're contiguous */
1002 if (count == 0)
1003 goto free_this;
1004 else if (block_to_free == nr - count)
1005 count++;
1006 else {
1007 ext2_free_blocks (inode, block_to_free, count);
1008 mark_inode_dirty(inode);
1009 free_this:
1010 block_to_free = nr;
1011 count = 1;
1012 }
1013 }
1014 }
1015 if (count > 0) {
1016 ext2_free_blocks (inode, block_to_free, count);
1017 mark_inode_dirty(inode);
1018 }
1019}
1020
1021/**
1022 * ext2_free_branches - free an array of branches
1023 * @inode: inode we are dealing with
1024 * @p: array of block numbers
1025 * @q: pointer immediately past the end of array
1026 * @depth: depth of the branches to free
1027 *
1028 * We are freeing all blocks referred from these branches (numbers are
1029 * stored as little-endian 32-bit) and updating @inode->i_blocks
1030 * appropriately.
1031 */
1032static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1033{
1034 struct buffer_head * bh;
1035 unsigned long nr;
1036
1037 if (depth--) {
1038 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1039 for ( ; p < q ; p++) {
1040 nr = le32_to_cpu(*p);
1041 if (!nr)
1042 continue;
1043 *p = 0;
1044 bh = sb_bread(inode->i_sb, nr);
1045 /*
1046 * A read failure? Report error and clear slot
1047 * (should be rare).
1048 */
1049 if (!bh) {
1050 ext2_error(inode->i_sb, "ext2_free_branches",
1051 "Read failure, inode=%ld, block=%ld",
1052 inode->i_ino, nr);
1053 continue;
1054 }
1055 ext2_free_branches(inode,
1056 (__le32*)bh->b_data,
1057 (__le32*)bh->b_data + addr_per_block,
1058 depth);
1059 bforget(bh);
1060 ext2_free_blocks(inode, nr, 1);
1061 mark_inode_dirty(inode);
1062 }
1063 } else
1064 ext2_free_data(inode, p, q);
1065}
1066
1067static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1068{
1069 __le32 *i_data = EXT2_I(inode)->i_data;
1070 struct ext2_inode_info *ei = EXT2_I(inode);
1071 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1072 int offsets[4];
1073 Indirect chain[4];
1074 Indirect *partial;
1075 __le32 nr = 0;
1076 int n;
1077 long iblock;
1078 unsigned blocksize;
1079 blocksize = inode->i_sb->s_blocksize;
1080 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1081
1082 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1083 if (n == 0)
1084 return;
1085
1086 /*
1087 * From here we block out all ext2_get_block() callers who want to
1088 * modify the block allocation tree.
1089 */
1090 mutex_lock(&ei->truncate_mutex);
1091
1092 if (n == 1) {
1093 ext2_free_data(inode, i_data+offsets[0],
1094 i_data + EXT2_NDIR_BLOCKS);
1095 goto do_indirects;
1096 }
1097
1098 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1099 /* Kill the top of shared branch (already detached) */
1100 if (nr) {
1101 if (partial == chain)
1102 mark_inode_dirty(inode);
1103 else
1104 mark_buffer_dirty_inode(partial->bh, inode);
1105 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1106 }
1107 /* Clear the ends of indirect blocks on the shared branch */
1108 while (partial > chain) {
1109 ext2_free_branches(inode,
1110 partial->p + 1,
1111 (__le32*)partial->bh->b_data+addr_per_block,
1112 (chain+n-1) - partial);
1113 mark_buffer_dirty_inode(partial->bh, inode);
1114 brelse (partial->bh);
1115 partial--;
1116 }
1117do_indirects:
1118 /* Kill the remaining (whole) subtrees */
1119 switch (offsets[0]) {
1120 default:
1121 nr = i_data[EXT2_IND_BLOCK];
1122 if (nr) {
1123 i_data[EXT2_IND_BLOCK] = 0;
1124 mark_inode_dirty(inode);
1125 ext2_free_branches(inode, &nr, &nr+1, 1);
1126 }
1127 case EXT2_IND_BLOCK:
1128 nr = i_data[EXT2_DIND_BLOCK];
1129 if (nr) {
1130 i_data[EXT2_DIND_BLOCK] = 0;
1131 mark_inode_dirty(inode);
1132 ext2_free_branches(inode, &nr, &nr+1, 2);
1133 }
1134 case EXT2_DIND_BLOCK:
1135 nr = i_data[EXT2_TIND_BLOCK];
1136 if (nr) {
1137 i_data[EXT2_TIND_BLOCK] = 0;
1138 mark_inode_dirty(inode);
1139 ext2_free_branches(inode, &nr, &nr+1, 3);
1140 }
1141 case EXT2_TIND_BLOCK:
1142 ;
1143 }
1144
1145 ext2_discard_reservation(inode);
1146
1147 mutex_unlock(&ei->truncate_mutex);
1148}
1149
1150static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1151{
1152 /*
1153 * XXX: it seems like a bug here that we don't allow
1154 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1155 * review and fix this.
1156 *
1157 * Also would be nice to be able to handle IO errors and such,
1158 * but that's probably too much to ask.
1159 */
1160 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1161 S_ISLNK(inode->i_mode)))
1162 return;
1163 if (ext2_inode_is_fast_symlink(inode))
1164 return;
1165 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1166 return;
1167 __ext2_truncate_blocks(inode, offset);
1168}
1169
1170static int ext2_setsize(struct inode *inode, loff_t newsize)
1171{
1172 int error;
1173
1174 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1175 S_ISLNK(inode->i_mode)))
1176 return -EINVAL;
1177 if (ext2_inode_is_fast_symlink(inode))
1178 return -EINVAL;
1179 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1180 return -EPERM;
1181
1182 inode_dio_wait(inode);
1183
1184 if (mapping_is_xip(inode->i_mapping))
1185 error = xip_truncate_page(inode->i_mapping, newsize);
1186 else if (test_opt(inode->i_sb, NOBH))
1187 error = nobh_truncate_page(inode->i_mapping,
1188 newsize, ext2_get_block);
1189 else
1190 error = block_truncate_page(inode->i_mapping,
1191 newsize, ext2_get_block);
1192 if (error)
1193 return error;
1194
1195 truncate_setsize(inode, newsize);
1196 __ext2_truncate_blocks(inode, newsize);
1197
1198 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1199 if (inode_needs_sync(inode)) {
1200 sync_mapping_buffers(inode->i_mapping);
1201 sync_inode_metadata(inode, 1);
1202 } else {
1203 mark_inode_dirty(inode);
1204 }
1205
1206 return 0;
1207}
1208
1209static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1210 struct buffer_head **p)
1211{
1212 struct buffer_head * bh;
1213 unsigned long block_group;
1214 unsigned long block;
1215 unsigned long offset;
1216 struct ext2_group_desc * gdp;
1217
1218 *p = NULL;
1219 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1220 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1221 goto Einval;
1222
1223 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1224 gdp = ext2_get_group_desc(sb, block_group, NULL);
1225 if (!gdp)
1226 goto Egdp;
1227 /*
1228 * Figure out the offset within the block group inode table
1229 */
1230 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1231 block = le32_to_cpu(gdp->bg_inode_table) +
1232 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1233 if (!(bh = sb_bread(sb, block)))
1234 goto Eio;
1235
1236 *p = bh;
1237 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1238 return (struct ext2_inode *) (bh->b_data + offset);
1239
1240Einval:
1241 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1242 (unsigned long) ino);
1243 return ERR_PTR(-EINVAL);
1244Eio:
1245 ext2_error(sb, "ext2_get_inode",
1246 "unable to read inode block - inode=%lu, block=%lu",
1247 (unsigned long) ino, block);
1248Egdp:
1249 return ERR_PTR(-EIO);
1250}
1251
1252void ext2_set_inode_flags(struct inode *inode)
1253{
1254 unsigned int flags = EXT2_I(inode)->i_flags;
1255
1256 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1257 if (flags & EXT2_SYNC_FL)
1258 inode->i_flags |= S_SYNC;
1259 if (flags & EXT2_APPEND_FL)
1260 inode->i_flags |= S_APPEND;
1261 if (flags & EXT2_IMMUTABLE_FL)
1262 inode->i_flags |= S_IMMUTABLE;
1263 if (flags & EXT2_NOATIME_FL)
1264 inode->i_flags |= S_NOATIME;
1265 if (flags & EXT2_DIRSYNC_FL)
1266 inode->i_flags |= S_DIRSYNC;
1267}
1268
1269/* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1270void ext2_get_inode_flags(struct ext2_inode_info *ei)
1271{
1272 unsigned int flags = ei->vfs_inode.i_flags;
1273
1274 ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1275 EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1276 if (flags & S_SYNC)
1277 ei->i_flags |= EXT2_SYNC_FL;
1278 if (flags & S_APPEND)
1279 ei->i_flags |= EXT2_APPEND_FL;
1280 if (flags & S_IMMUTABLE)
1281 ei->i_flags |= EXT2_IMMUTABLE_FL;
1282 if (flags & S_NOATIME)
1283 ei->i_flags |= EXT2_NOATIME_FL;
1284 if (flags & S_DIRSYNC)
1285 ei->i_flags |= EXT2_DIRSYNC_FL;
1286}
1287
1288struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1289{
1290 struct ext2_inode_info *ei;
1291 struct buffer_head * bh;
1292 struct ext2_inode *raw_inode;
1293 struct inode *inode;
1294 long ret = -EIO;
1295 int n;
1296 uid_t i_uid;
1297 gid_t i_gid;
1298
1299 inode = iget_locked(sb, ino);
1300 if (!inode)
1301 return ERR_PTR(-ENOMEM);
1302 if (!(inode->i_state & I_NEW))
1303 return inode;
1304
1305 ei = EXT2_I(inode);
1306 ei->i_block_alloc_info = NULL;
1307
1308 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1309 if (IS_ERR(raw_inode)) {
1310 ret = PTR_ERR(raw_inode);
1311 goto bad_inode;
1312 }
1313
1314 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1315 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1316 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1317 if (!(test_opt (inode->i_sb, NO_UID32))) {
1318 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1319 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1320 }
1321 i_uid_write(inode, i_uid);
1322 i_gid_write(inode, i_gid);
1323 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1324 inode->i_size = le32_to_cpu(raw_inode->i_size);
1325 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1326 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1327 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1328 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1329 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1330 /* We now have enough fields to check if the inode was active or not.
1331 * This is needed because nfsd might try to access dead inodes
1332 * the test is that same one that e2fsck uses
1333 * NeilBrown 1999oct15
1334 */
1335 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1336 /* this inode is deleted */
1337 brelse (bh);
1338 ret = -ESTALE;
1339 goto bad_inode;
1340 }
1341 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1342 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1343 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1344 ei->i_frag_no = raw_inode->i_frag;
1345 ei->i_frag_size = raw_inode->i_fsize;
1346 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1347 ei->i_dir_acl = 0;
1348 if (S_ISREG(inode->i_mode))
1349 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1350 else
1351 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1352 ei->i_dtime = 0;
1353 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1354 ei->i_state = 0;
1355 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1356 ei->i_dir_start_lookup = 0;
1357
1358 /*
1359 * NOTE! The in-memory inode i_data array is in little-endian order
1360 * even on big-endian machines: we do NOT byteswap the block numbers!
1361 */
1362 for (n = 0; n < EXT2_N_BLOCKS; n++)
1363 ei->i_data[n] = raw_inode->i_block[n];
1364
1365 if (S_ISREG(inode->i_mode)) {
1366 inode->i_op = &ext2_file_inode_operations;
1367 if (ext2_use_xip(inode->i_sb)) {
1368 inode->i_mapping->a_ops = &ext2_aops_xip;
1369 inode->i_fop = &ext2_xip_file_operations;
1370 } else if (test_opt(inode->i_sb, NOBH)) {
1371 inode->i_mapping->a_ops = &ext2_nobh_aops;
1372 inode->i_fop = &ext2_file_operations;
1373 } else {
1374 inode->i_mapping->a_ops = &ext2_aops;
1375 inode->i_fop = &ext2_file_operations;
1376 }
1377 } else if (S_ISDIR(inode->i_mode)) {
1378 inode->i_op = &ext2_dir_inode_operations;
1379 inode->i_fop = &ext2_dir_operations;
1380 if (test_opt(inode->i_sb, NOBH))
1381 inode->i_mapping->a_ops = &ext2_nobh_aops;
1382 else
1383 inode->i_mapping->a_ops = &ext2_aops;
1384 } else if (S_ISLNK(inode->i_mode)) {
1385 if (ext2_inode_is_fast_symlink(inode)) {
1386 inode->i_op = &ext2_fast_symlink_inode_operations;
1387 nd_terminate_link(ei->i_data, inode->i_size,
1388 sizeof(ei->i_data) - 1);
1389 } else {
1390 inode->i_op = &ext2_symlink_inode_operations;
1391 if (test_opt(inode->i_sb, NOBH))
1392 inode->i_mapping->a_ops = &ext2_nobh_aops;
1393 else
1394 inode->i_mapping->a_ops = &ext2_aops;
1395 }
1396 } else {
1397 inode->i_op = &ext2_special_inode_operations;
1398 if (raw_inode->i_block[0])
1399 init_special_inode(inode, inode->i_mode,
1400 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1401 else
1402 init_special_inode(inode, inode->i_mode,
1403 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1404 }
1405 brelse (bh);
1406 ext2_set_inode_flags(inode);
1407 unlock_new_inode(inode);
1408 return inode;
1409
1410bad_inode:
1411 iget_failed(inode);
1412 return ERR_PTR(ret);
1413}
1414
1415static int __ext2_write_inode(struct inode *inode, int do_sync)
1416{
1417 struct ext2_inode_info *ei = EXT2_I(inode);
1418 struct super_block *sb = inode->i_sb;
1419 ino_t ino = inode->i_ino;
1420 uid_t uid = i_uid_read(inode);
1421 gid_t gid = i_gid_read(inode);
1422 struct buffer_head * bh;
1423 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1424 int n;
1425 int err = 0;
1426
1427 if (IS_ERR(raw_inode))
1428 return -EIO;
1429
1430 /* For fields not not tracking in the in-memory inode,
1431 * initialise them to zero for new inodes. */
1432 if (ei->i_state & EXT2_STATE_NEW)
1433 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1434
1435 ext2_get_inode_flags(ei);
1436 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1437 if (!(test_opt(sb, NO_UID32))) {
1438 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1439 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1440/*
1441 * Fix up interoperability with old kernels. Otherwise, old inodes get
1442 * re-used with the upper 16 bits of the uid/gid intact
1443 */
1444 if (!ei->i_dtime) {
1445 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1446 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1447 } else {
1448 raw_inode->i_uid_high = 0;
1449 raw_inode->i_gid_high = 0;
1450 }
1451 } else {
1452 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1453 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1454 raw_inode->i_uid_high = 0;
1455 raw_inode->i_gid_high = 0;
1456 }
1457 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1458 raw_inode->i_size = cpu_to_le32(inode->i_size);
1459 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1460 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1461 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1462
1463 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1464 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1465 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1466 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1467 raw_inode->i_frag = ei->i_frag_no;
1468 raw_inode->i_fsize = ei->i_frag_size;
1469 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1470 if (!S_ISREG(inode->i_mode))
1471 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1472 else {
1473 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1474 if (inode->i_size > 0x7fffffffULL) {
1475 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1476 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1477 EXT2_SB(sb)->s_es->s_rev_level ==
1478 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1479 /* If this is the first large file
1480 * created, add a flag to the superblock.
1481 */
1482 spin_lock(&EXT2_SB(sb)->s_lock);
1483 ext2_update_dynamic_rev(sb);
1484 EXT2_SET_RO_COMPAT_FEATURE(sb,
1485 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1486 spin_unlock(&EXT2_SB(sb)->s_lock);
1487 ext2_write_super(sb);
1488 }
1489 }
1490 }
1491
1492 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1493 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1494 if (old_valid_dev(inode->i_rdev)) {
1495 raw_inode->i_block[0] =
1496 cpu_to_le32(old_encode_dev(inode->i_rdev));
1497 raw_inode->i_block[1] = 0;
1498 } else {
1499 raw_inode->i_block[0] = 0;
1500 raw_inode->i_block[1] =
1501 cpu_to_le32(new_encode_dev(inode->i_rdev));
1502 raw_inode->i_block[2] = 0;
1503 }
1504 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1505 raw_inode->i_block[n] = ei->i_data[n];
1506 mark_buffer_dirty(bh);
1507 if (do_sync) {
1508 sync_dirty_buffer(bh);
1509 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1510 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1511 sb->s_id, (unsigned long) ino);
1512 err = -EIO;
1513 }
1514 }
1515 ei->i_state &= ~EXT2_STATE_NEW;
1516 brelse (bh);
1517 return err;
1518}
1519
1520int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1521{
1522 return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1523}
1524
1525int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1526{
1527 struct inode *inode = dentry->d_inode;
1528 int error;
1529
1530 error = inode_change_ok(inode, iattr);
1531 if (error)
1532 return error;
1533
1534 if (is_quota_modification(inode, iattr))
1535 dquot_initialize(inode);
1536 if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1537 (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1538 error = dquot_transfer(inode, iattr);
1539 if (error)
1540 return error;
1541 }
1542 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1543 error = ext2_setsize(inode, iattr->ia_size);
1544 if (error)
1545 return error;
1546 }
1547 setattr_copy(inode, iattr);
1548 if (iattr->ia_valid & ATTR_MODE)
1549 error = ext2_acl_chmod(inode);
1550 mark_inode_dirty(inode);
1551
1552 return error;
1553}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext2/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Goal-directed block allocation by Stephen Tweedie
17 * (sct@dcs.ed.ac.uk), 1993, 1998
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * 64-bit file support on 64-bit platforms by Jakub Jelinek
21 * (jj@sunsite.ms.mff.cuni.cz)
22 *
23 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
24 */
25
26#include <linux/time.h>
27#include <linux/highuid.h>
28#include <linux/pagemap.h>
29#include <linux/dax.h>
30#include <linux/blkdev.h>
31#include <linux/quotaops.h>
32#include <linux/writeback.h>
33#include <linux/buffer_head.h>
34#include <linux/mpage.h>
35#include <linux/fiemap.h>
36#include <linux/iomap.h>
37#include <linux/namei.h>
38#include <linux/uio.h>
39#include "ext2.h"
40#include "acl.h"
41#include "xattr.h"
42
43static int __ext2_write_inode(struct inode *inode, int do_sync);
44
45/*
46 * Test whether an inode is a fast symlink.
47 */
48static inline int ext2_inode_is_fast_symlink(struct inode *inode)
49{
50 int ea_blocks = EXT2_I(inode)->i_file_acl ?
51 (inode->i_sb->s_blocksize >> 9) : 0;
52
53 return (S_ISLNK(inode->i_mode) &&
54 inode->i_blocks - ea_blocks == 0);
55}
56
57static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
58
59void ext2_write_failed(struct address_space *mapping, loff_t to)
60{
61 struct inode *inode = mapping->host;
62
63 if (to > inode->i_size) {
64 truncate_pagecache(inode, inode->i_size);
65 ext2_truncate_blocks(inode, inode->i_size);
66 }
67}
68
69/*
70 * Called at the last iput() if i_nlink is zero.
71 */
72void ext2_evict_inode(struct inode * inode)
73{
74 struct ext2_block_alloc_info *rsv;
75 int want_delete = 0;
76
77 if (!inode->i_nlink && !is_bad_inode(inode)) {
78 want_delete = 1;
79 dquot_initialize(inode);
80 } else {
81 dquot_drop(inode);
82 }
83
84 truncate_inode_pages_final(&inode->i_data);
85
86 if (want_delete) {
87 sb_start_intwrite(inode->i_sb);
88 /* set dtime */
89 EXT2_I(inode)->i_dtime = ktime_get_real_seconds();
90 mark_inode_dirty(inode);
91 __ext2_write_inode(inode, inode_needs_sync(inode));
92 /* truncate to 0 */
93 inode->i_size = 0;
94 if (inode->i_blocks)
95 ext2_truncate_blocks(inode, 0);
96 ext2_xattr_delete_inode(inode);
97 }
98
99 invalidate_inode_buffers(inode);
100 clear_inode(inode);
101
102 ext2_discard_reservation(inode);
103 rsv = EXT2_I(inode)->i_block_alloc_info;
104 EXT2_I(inode)->i_block_alloc_info = NULL;
105 if (unlikely(rsv))
106 kfree(rsv);
107
108 if (want_delete) {
109 ext2_free_inode(inode);
110 sb_end_intwrite(inode->i_sb);
111 }
112}
113
114typedef struct {
115 __le32 *p;
116 __le32 key;
117 struct buffer_head *bh;
118} Indirect;
119
120static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
121{
122 p->key = *(p->p = v);
123 p->bh = bh;
124}
125
126static inline int verify_chain(Indirect *from, Indirect *to)
127{
128 while (from <= to && from->key == *from->p)
129 from++;
130 return (from > to);
131}
132
133/**
134 * ext2_block_to_path - parse the block number into array of offsets
135 * @inode: inode in question (we are only interested in its superblock)
136 * @i_block: block number to be parsed
137 * @offsets: array to store the offsets in
138 * @boundary: set this non-zero if the referred-to block is likely to be
139 * followed (on disk) by an indirect block.
140 * To store the locations of file's data ext2 uses a data structure common
141 * for UNIX filesystems - tree of pointers anchored in the inode, with
142 * data blocks at leaves and indirect blocks in intermediate nodes.
143 * This function translates the block number into path in that tree -
144 * return value is the path length and @offsets[n] is the offset of
145 * pointer to (n+1)th node in the nth one. If @block is out of range
146 * (negative or too large) warning is printed and zero returned.
147 *
148 * Note: function doesn't find node addresses, so no IO is needed. All
149 * we need to know is the capacity of indirect blocks (taken from the
150 * inode->i_sb).
151 */
152
153/*
154 * Portability note: the last comparison (check that we fit into triple
155 * indirect block) is spelled differently, because otherwise on an
156 * architecture with 32-bit longs and 8Kb pages we might get into trouble
157 * if our filesystem had 8Kb blocks. We might use long long, but that would
158 * kill us on x86. Oh, well, at least the sign propagation does not matter -
159 * i_block would have to be negative in the very beginning, so we would not
160 * get there at all.
161 */
162
163static int ext2_block_to_path(struct inode *inode,
164 long i_block, int offsets[4], int *boundary)
165{
166 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
167 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
168 const long direct_blocks = EXT2_NDIR_BLOCKS,
169 indirect_blocks = ptrs,
170 double_blocks = (1 << (ptrs_bits * 2));
171 int n = 0;
172 int final = 0;
173
174 if (i_block < 0) {
175 ext2_msg(inode->i_sb, KERN_WARNING,
176 "warning: %s: block < 0", __func__);
177 } else if (i_block < direct_blocks) {
178 offsets[n++] = i_block;
179 final = direct_blocks;
180 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
181 offsets[n++] = EXT2_IND_BLOCK;
182 offsets[n++] = i_block;
183 final = ptrs;
184 } else if ((i_block -= indirect_blocks) < double_blocks) {
185 offsets[n++] = EXT2_DIND_BLOCK;
186 offsets[n++] = i_block >> ptrs_bits;
187 offsets[n++] = i_block & (ptrs - 1);
188 final = ptrs;
189 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
190 offsets[n++] = EXT2_TIND_BLOCK;
191 offsets[n++] = i_block >> (ptrs_bits * 2);
192 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
193 offsets[n++] = i_block & (ptrs - 1);
194 final = ptrs;
195 } else {
196 ext2_msg(inode->i_sb, KERN_WARNING,
197 "warning: %s: block is too big", __func__);
198 }
199 if (boundary)
200 *boundary = final - 1 - (i_block & (ptrs - 1));
201
202 return n;
203}
204
205/**
206 * ext2_get_branch - read the chain of indirect blocks leading to data
207 * @inode: inode in question
208 * @depth: depth of the chain (1 - direct pointer, etc.)
209 * @offsets: offsets of pointers in inode/indirect blocks
210 * @chain: place to store the result
211 * @err: here we store the error value
212 *
213 * Function fills the array of triples <key, p, bh> and returns %NULL
214 * if everything went OK or the pointer to the last filled triple
215 * (incomplete one) otherwise. Upon the return chain[i].key contains
216 * the number of (i+1)-th block in the chain (as it is stored in memory,
217 * i.e. little-endian 32-bit), chain[i].p contains the address of that
218 * number (it points into struct inode for i==0 and into the bh->b_data
219 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
220 * block for i>0 and NULL for i==0. In other words, it holds the block
221 * numbers of the chain, addresses they were taken from (and where we can
222 * verify that chain did not change) and buffer_heads hosting these
223 * numbers.
224 *
225 * Function stops when it stumbles upon zero pointer (absent block)
226 * (pointer to last triple returned, *@err == 0)
227 * or when it gets an IO error reading an indirect block
228 * (ditto, *@err == -EIO)
229 * or when it notices that chain had been changed while it was reading
230 * (ditto, *@err == -EAGAIN)
231 * or when it reads all @depth-1 indirect blocks successfully and finds
232 * the whole chain, all way to the data (returns %NULL, *err == 0).
233 */
234static Indirect *ext2_get_branch(struct inode *inode,
235 int depth,
236 int *offsets,
237 Indirect chain[4],
238 int *err)
239{
240 struct super_block *sb = inode->i_sb;
241 Indirect *p = chain;
242 struct buffer_head *bh;
243
244 *err = 0;
245 /* i_data is not going away, no lock needed */
246 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
247 if (!p->key)
248 goto no_block;
249 while (--depth) {
250 bh = sb_bread(sb, le32_to_cpu(p->key));
251 if (!bh)
252 goto failure;
253 read_lock(&EXT2_I(inode)->i_meta_lock);
254 if (!verify_chain(chain, p))
255 goto changed;
256 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
257 read_unlock(&EXT2_I(inode)->i_meta_lock);
258 if (!p->key)
259 goto no_block;
260 }
261 return NULL;
262
263changed:
264 read_unlock(&EXT2_I(inode)->i_meta_lock);
265 brelse(bh);
266 *err = -EAGAIN;
267 goto no_block;
268failure:
269 *err = -EIO;
270no_block:
271 return p;
272}
273
274/**
275 * ext2_find_near - find a place for allocation with sufficient locality
276 * @inode: owner
277 * @ind: descriptor of indirect block.
278 *
279 * This function returns the preferred place for block allocation.
280 * It is used when heuristic for sequential allocation fails.
281 * Rules are:
282 * + if there is a block to the left of our position - allocate near it.
283 * + if pointer will live in indirect block - allocate near that block.
284 * + if pointer will live in inode - allocate in the same cylinder group.
285 *
286 * In the latter case we colour the starting block by the callers PID to
287 * prevent it from clashing with concurrent allocations for a different inode
288 * in the same block group. The PID is used here so that functionally related
289 * files will be close-by on-disk.
290 *
291 * Caller must make sure that @ind is valid and will stay that way.
292 */
293
294static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
295{
296 struct ext2_inode_info *ei = EXT2_I(inode);
297 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
298 __le32 *p;
299 ext2_fsblk_t bg_start;
300 ext2_fsblk_t colour;
301
302 /* Try to find previous block */
303 for (p = ind->p - 1; p >= start; p--)
304 if (*p)
305 return le32_to_cpu(*p);
306
307 /* No such thing, so let's try location of indirect block */
308 if (ind->bh)
309 return ind->bh->b_blocknr;
310
311 /*
312 * It is going to be referred from inode itself? OK, just put it into
313 * the same cylinder group then.
314 */
315 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
316 colour = (current->pid % 16) *
317 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
318 return bg_start + colour;
319}
320
321/**
322 * ext2_find_goal - find a preferred place for allocation.
323 * @inode: owner
324 * @block: block we want
325 * @partial: pointer to the last triple within a chain
326 *
327 * Returns preferred place for a block (the goal).
328 */
329
330static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
331 Indirect *partial)
332{
333 struct ext2_block_alloc_info *block_i;
334
335 block_i = EXT2_I(inode)->i_block_alloc_info;
336
337 /*
338 * try the heuristic for sequential allocation,
339 * failing that at least try to get decent locality.
340 */
341 if (block_i && (block == block_i->last_alloc_logical_block + 1)
342 && (block_i->last_alloc_physical_block != 0)) {
343 return block_i->last_alloc_physical_block + 1;
344 }
345
346 return ext2_find_near(inode, partial);
347}
348
349/**
350 * ext2_blks_to_allocate: Look up the block map and count the number
351 * of direct blocks need to be allocated for the given branch.
352 *
353 * @branch: chain of indirect blocks
354 * @k: number of blocks need for indirect blocks
355 * @blks: number of data blocks to be mapped.
356 * @blocks_to_boundary: the offset in the indirect block
357 *
358 * return the number of direct blocks to allocate.
359 */
360static int
361ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
362 int blocks_to_boundary)
363{
364 unsigned long count = 0;
365
366 /*
367 * Simple case, [t,d]Indirect block(s) has not allocated yet
368 * then it's clear blocks on that path have not allocated
369 */
370 if (k > 0) {
371 /* right now don't hanel cross boundary allocation */
372 if (blks < blocks_to_boundary + 1)
373 count += blks;
374 else
375 count += blocks_to_boundary + 1;
376 return count;
377 }
378
379 count++;
380 while (count < blks && count <= blocks_to_boundary
381 && le32_to_cpu(*(branch[0].p + count)) == 0) {
382 count++;
383 }
384 return count;
385}
386
387/**
388 * ext2_alloc_blocks: Allocate multiple blocks needed for a branch.
389 * @inode: Owner.
390 * @goal: Preferred place for allocation.
391 * @indirect_blks: The number of blocks needed to allocate for indirect blocks.
392 * @blks: The number of blocks need to allocate for direct blocks.
393 * @new_blocks: On return it will store the new block numbers for
394 * the indirect blocks(if needed) and the first direct block.
395 * @err: Error pointer.
396 *
397 * Return: Number of blocks allocated.
398 */
399static int ext2_alloc_blocks(struct inode *inode,
400 ext2_fsblk_t goal, int indirect_blks, int blks,
401 ext2_fsblk_t new_blocks[4], int *err)
402{
403 int target, i;
404 unsigned long count = 0;
405 int index = 0;
406 ext2_fsblk_t current_block = 0;
407 int ret = 0;
408
409 /*
410 * Here we try to allocate the requested multiple blocks at once,
411 * on a best-effort basis.
412 * To build a branch, we should allocate blocks for
413 * the indirect blocks(if not allocated yet), and at least
414 * the first direct block of this branch. That's the
415 * minimum number of blocks need to allocate(required)
416 */
417 target = blks + indirect_blks;
418
419 while (1) {
420 count = target;
421 /* allocating blocks for indirect blocks and direct blocks */
422 current_block = ext2_new_blocks(inode, goal, &count, err, 0);
423 if (*err)
424 goto failed_out;
425
426 target -= count;
427 /* allocate blocks for indirect blocks */
428 while (index < indirect_blks && count) {
429 new_blocks[index++] = current_block++;
430 count--;
431 }
432
433 if (count > 0)
434 break;
435 }
436
437 /* save the new block number for the first direct block */
438 new_blocks[index] = current_block;
439
440 /* total number of blocks allocated for direct blocks */
441 ret = count;
442 *err = 0;
443 return ret;
444failed_out:
445 for (i = 0; i <index; i++)
446 ext2_free_blocks(inode, new_blocks[i], 1);
447 if (index)
448 mark_inode_dirty(inode);
449 return ret;
450}
451
452/**
453 * ext2_alloc_branch - allocate and set up a chain of blocks.
454 * @inode: owner
455 * @indirect_blks: depth of the chain (number of blocks to allocate)
456 * @blks: number of allocated direct blocks
457 * @goal: preferred place for allocation
458 * @offsets: offsets (in the blocks) to store the pointers to next.
459 * @branch: place to store the chain in.
460 *
461 * This function allocates @num blocks, zeroes out all but the last one,
462 * links them into chain and (if we are synchronous) writes them to disk.
463 * In other words, it prepares a branch that can be spliced onto the
464 * inode. It stores the information about that chain in the branch[], in
465 * the same format as ext2_get_branch() would do. We are calling it after
466 * we had read the existing part of chain and partial points to the last
467 * triple of that (one with zero ->key). Upon the exit we have the same
468 * picture as after the successful ext2_get_block(), except that in one
469 * place chain is disconnected - *branch->p is still zero (we did not
470 * set the last link), but branch->key contains the number that should
471 * be placed into *branch->p to fill that gap.
472 *
473 * If allocation fails we free all blocks we've allocated (and forget
474 * their buffer_heads) and return the error value the from failed
475 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
476 * as described above and return 0.
477 */
478
479static int ext2_alloc_branch(struct inode *inode,
480 int indirect_blks, int *blks, ext2_fsblk_t goal,
481 int *offsets, Indirect *branch)
482{
483 int blocksize = inode->i_sb->s_blocksize;
484 int i, n = 0;
485 int err = 0;
486 struct buffer_head *bh;
487 int num;
488 ext2_fsblk_t new_blocks[4];
489 ext2_fsblk_t current_block;
490
491 num = ext2_alloc_blocks(inode, goal, indirect_blks,
492 *blks, new_blocks, &err);
493 if (err)
494 return err;
495
496 branch[0].key = cpu_to_le32(new_blocks[0]);
497 /*
498 * metadata blocks and data blocks are allocated.
499 */
500 for (n = 1; n <= indirect_blks; n++) {
501 /*
502 * Get buffer_head for parent block, zero it out
503 * and set the pointer to new one, then send
504 * parent to disk.
505 */
506 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
507 if (unlikely(!bh)) {
508 err = -ENOMEM;
509 goto failed;
510 }
511 branch[n].bh = bh;
512 lock_buffer(bh);
513 memset(bh->b_data, 0, blocksize);
514 branch[n].p = (__le32 *) bh->b_data + offsets[n];
515 branch[n].key = cpu_to_le32(new_blocks[n]);
516 *branch[n].p = branch[n].key;
517 if ( n == indirect_blks) {
518 current_block = new_blocks[n];
519 /*
520 * End of chain, update the last new metablock of
521 * the chain to point to the new allocated
522 * data blocks numbers
523 */
524 for (i=1; i < num; i++)
525 *(branch[n].p + i) = cpu_to_le32(++current_block);
526 }
527 set_buffer_uptodate(bh);
528 unlock_buffer(bh);
529 mark_buffer_dirty_inode(bh, inode);
530 /* We used to sync bh here if IS_SYNC(inode).
531 * But we now rely upon generic_write_sync()
532 * and b_inode_buffers. But not for directories.
533 */
534 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
535 sync_dirty_buffer(bh);
536 }
537 *blks = num;
538 return err;
539
540failed:
541 for (i = 1; i < n; i++)
542 bforget(branch[i].bh);
543 for (i = 0; i < indirect_blks; i++)
544 ext2_free_blocks(inode, new_blocks[i], 1);
545 ext2_free_blocks(inode, new_blocks[i], num);
546 return err;
547}
548
549/**
550 * ext2_splice_branch - splice the allocated branch onto inode.
551 * @inode: owner
552 * @block: (logical) number of block we are adding
553 * @where: location of missing link
554 * @num: number of indirect blocks we are adding
555 * @blks: number of direct blocks we are adding
556 *
557 * This function fills the missing link and does all housekeeping needed in
558 * inode (->i_blocks, etc.). In case of success we end up with the full
559 * chain to new block and return 0.
560 */
561static void ext2_splice_branch(struct inode *inode,
562 long block, Indirect *where, int num, int blks)
563{
564 int i;
565 struct ext2_block_alloc_info *block_i;
566 ext2_fsblk_t current_block;
567
568 block_i = EXT2_I(inode)->i_block_alloc_info;
569
570 /* XXX LOCKING probably should have i_meta_lock ?*/
571 /* That's it */
572
573 *where->p = where->key;
574
575 /*
576 * Update the host buffer_head or inode to point to more just allocated
577 * direct blocks blocks
578 */
579 if (num == 0 && blks > 1) {
580 current_block = le32_to_cpu(where->key) + 1;
581 for (i = 1; i < blks; i++)
582 *(where->p + i ) = cpu_to_le32(current_block++);
583 }
584
585 /*
586 * update the most recently allocated logical & physical block
587 * in i_block_alloc_info, to assist find the proper goal block for next
588 * allocation
589 */
590 if (block_i) {
591 block_i->last_alloc_logical_block = block + blks - 1;
592 block_i->last_alloc_physical_block =
593 le32_to_cpu(where[num].key) + blks - 1;
594 }
595
596 /* We are done with atomic stuff, now do the rest of housekeeping */
597
598 /* had we spliced it onto indirect block? */
599 if (where->bh)
600 mark_buffer_dirty_inode(where->bh, inode);
601
602 inode_set_ctime_current(inode);
603 mark_inode_dirty(inode);
604}
605
606/*
607 * Allocation strategy is simple: if we have to allocate something, we will
608 * have to go the whole way to leaf. So let's do it before attaching anything
609 * to tree, set linkage between the newborn blocks, write them if sync is
610 * required, recheck the path, free and repeat if check fails, otherwise
611 * set the last missing link (that will protect us from any truncate-generated
612 * removals - all blocks on the path are immune now) and possibly force the
613 * write on the parent block.
614 * That has a nice additional property: no special recovery from the failed
615 * allocations is needed - we simply release blocks and do not touch anything
616 * reachable from inode.
617 *
618 * `handle' can be NULL if create == 0.
619 *
620 * return > 0, # of blocks mapped or allocated.
621 * return = 0, if plain lookup failed.
622 * return < 0, error case.
623 */
624static int ext2_get_blocks(struct inode *inode,
625 sector_t iblock, unsigned long maxblocks,
626 u32 *bno, bool *new, bool *boundary,
627 int create)
628{
629 int err;
630 int offsets[4];
631 Indirect chain[4];
632 Indirect *partial;
633 ext2_fsblk_t goal;
634 int indirect_blks;
635 int blocks_to_boundary = 0;
636 int depth;
637 struct ext2_inode_info *ei = EXT2_I(inode);
638 int count = 0;
639 ext2_fsblk_t first_block = 0;
640
641 BUG_ON(maxblocks == 0);
642
643 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
644
645 if (depth == 0)
646 return -EIO;
647
648 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
649 /* Simplest case - block found, no allocation needed */
650 if (!partial) {
651 first_block = le32_to_cpu(chain[depth - 1].key);
652 count++;
653 /*map more blocks*/
654 while (count < maxblocks && count <= blocks_to_boundary) {
655 ext2_fsblk_t blk;
656
657 if (!verify_chain(chain, chain + depth - 1)) {
658 /*
659 * Indirect block might be removed by
660 * truncate while we were reading it.
661 * Handling of that case: forget what we've
662 * got now, go to reread.
663 */
664 err = -EAGAIN;
665 count = 0;
666 partial = chain + depth - 1;
667 break;
668 }
669 blk = le32_to_cpu(*(chain[depth-1].p + count));
670 if (blk == first_block + count)
671 count++;
672 else
673 break;
674 }
675 if (err != -EAGAIN)
676 goto got_it;
677 }
678
679 /* Next simple case - plain lookup or failed read of indirect block */
680 if (!create || err == -EIO)
681 goto cleanup;
682
683 mutex_lock(&ei->truncate_mutex);
684 /*
685 * If the indirect block is missing while we are reading
686 * the chain(ext2_get_branch() returns -EAGAIN err), or
687 * if the chain has been changed after we grab the semaphore,
688 * (either because another process truncated this branch, or
689 * another get_block allocated this branch) re-grab the chain to see if
690 * the request block has been allocated or not.
691 *
692 * Since we already block the truncate/other get_block
693 * at this point, we will have the current copy of the chain when we
694 * splice the branch into the tree.
695 */
696 if (err == -EAGAIN || !verify_chain(chain, partial)) {
697 while (partial > chain) {
698 brelse(partial->bh);
699 partial--;
700 }
701 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
702 if (!partial) {
703 count++;
704 mutex_unlock(&ei->truncate_mutex);
705 goto got_it;
706 }
707
708 if (err) {
709 mutex_unlock(&ei->truncate_mutex);
710 goto cleanup;
711 }
712 }
713
714 /*
715 * Okay, we need to do block allocation. Lazily initialize the block
716 * allocation info here if necessary
717 */
718 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
719 ext2_init_block_alloc_info(inode);
720
721 goal = ext2_find_goal(inode, iblock, partial);
722
723 /* the number of blocks need to allocate for [d,t]indirect blocks */
724 indirect_blks = (chain + depth) - partial - 1;
725 /*
726 * Next look up the indirect map to count the total number of
727 * direct blocks to allocate for this branch.
728 */
729 count = ext2_blks_to_allocate(partial, indirect_blks,
730 maxblocks, blocks_to_boundary);
731 /*
732 * XXX ???? Block out ext2_truncate while we alter the tree
733 */
734 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
735 offsets + (partial - chain), partial);
736
737 if (err) {
738 mutex_unlock(&ei->truncate_mutex);
739 goto cleanup;
740 }
741
742 if (IS_DAX(inode)) {
743 /*
744 * We must unmap blocks before zeroing so that writeback cannot
745 * overwrite zeros with stale data from block device page cache.
746 */
747 clean_bdev_aliases(inode->i_sb->s_bdev,
748 le32_to_cpu(chain[depth-1].key),
749 count);
750 /*
751 * block must be initialised before we put it in the tree
752 * so that it's not found by another thread before it's
753 * initialised
754 */
755 err = sb_issue_zeroout(inode->i_sb,
756 le32_to_cpu(chain[depth-1].key), count,
757 GFP_KERNEL);
758 if (err) {
759 mutex_unlock(&ei->truncate_mutex);
760 goto cleanup;
761 }
762 }
763 *new = true;
764
765 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
766 mutex_unlock(&ei->truncate_mutex);
767got_it:
768 if (count > blocks_to_boundary)
769 *boundary = true;
770 err = count;
771 /* Clean up and exit */
772 partial = chain + depth - 1; /* the whole chain */
773cleanup:
774 while (partial > chain) {
775 brelse(partial->bh);
776 partial--;
777 }
778 if (err > 0)
779 *bno = le32_to_cpu(chain[depth-1].key);
780 return err;
781}
782
783int ext2_get_block(struct inode *inode, sector_t iblock,
784 struct buffer_head *bh_result, int create)
785{
786 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
787 bool new = false, boundary = false;
788 u32 bno;
789 int ret;
790
791 ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
792 create);
793 if (ret <= 0)
794 return ret;
795
796 map_bh(bh_result, inode->i_sb, bno);
797 bh_result->b_size = (ret << inode->i_blkbits);
798 if (new)
799 set_buffer_new(bh_result);
800 if (boundary)
801 set_buffer_boundary(bh_result);
802 return 0;
803
804}
805
806static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
807 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
808{
809 unsigned int blkbits = inode->i_blkbits;
810 unsigned long first_block = offset >> blkbits;
811 unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
812 struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
813 bool new = false, boundary = false;
814 u32 bno;
815 int ret;
816 bool create = flags & IOMAP_WRITE;
817
818 /*
819 * For writes that could fill holes inside i_size on a
820 * DIO_SKIP_HOLES filesystem we forbid block creations: only
821 * overwrites are permitted.
822 */
823 if ((flags & IOMAP_DIRECT) &&
824 (first_block << blkbits) < i_size_read(inode))
825 create = 0;
826
827 /*
828 * Writes that span EOF might trigger an IO size update on completion,
829 * so consider them to be dirty for the purposes of O_DSYNC even if
830 * there is no other metadata changes pending or have been made here.
831 */
832 if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode))
833 iomap->flags |= IOMAP_F_DIRTY;
834
835 ret = ext2_get_blocks(inode, first_block, max_blocks,
836 &bno, &new, &boundary, create);
837 if (ret < 0)
838 return ret;
839
840 iomap->flags = 0;
841 iomap->offset = (u64)first_block << blkbits;
842 if (flags & IOMAP_DAX)
843 iomap->dax_dev = sbi->s_daxdev;
844 else
845 iomap->bdev = inode->i_sb->s_bdev;
846
847 if (ret == 0) {
848 /*
849 * Switch to buffered-io for writing to holes in a non-extent
850 * based filesystem to avoid stale data exposure problem.
851 */
852 if (!create && (flags & IOMAP_WRITE) && (flags & IOMAP_DIRECT))
853 return -ENOTBLK;
854 iomap->type = IOMAP_HOLE;
855 iomap->addr = IOMAP_NULL_ADDR;
856 iomap->length = 1 << blkbits;
857 } else {
858 iomap->type = IOMAP_MAPPED;
859 iomap->addr = (u64)bno << blkbits;
860 if (flags & IOMAP_DAX)
861 iomap->addr += sbi->s_dax_part_off;
862 iomap->length = (u64)ret << blkbits;
863 iomap->flags |= IOMAP_F_MERGED;
864 }
865
866 if (new)
867 iomap->flags |= IOMAP_F_NEW;
868 return 0;
869}
870
871static int
872ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
873 ssize_t written, unsigned flags, struct iomap *iomap)
874{
875 /*
876 * Switch to buffered-io in case of any error.
877 * Blocks allocated can be used by the buffered-io path.
878 */
879 if ((flags & IOMAP_DIRECT) && (flags & IOMAP_WRITE) && written == 0)
880 return -ENOTBLK;
881
882 if (iomap->type == IOMAP_MAPPED &&
883 written < length &&
884 (flags & IOMAP_WRITE))
885 ext2_write_failed(inode->i_mapping, offset + length);
886 return 0;
887}
888
889const struct iomap_ops ext2_iomap_ops = {
890 .iomap_begin = ext2_iomap_begin,
891 .iomap_end = ext2_iomap_end,
892};
893
894int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
895 u64 start, u64 len)
896{
897 int ret;
898
899 inode_lock(inode);
900 len = min_t(u64, len, i_size_read(inode));
901 ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops);
902 inode_unlock(inode);
903
904 return ret;
905}
906
907static int ext2_read_folio(struct file *file, struct folio *folio)
908{
909 return mpage_read_folio(folio, ext2_get_block);
910}
911
912static void ext2_readahead(struct readahead_control *rac)
913{
914 mpage_readahead(rac, ext2_get_block);
915}
916
917static int
918ext2_write_begin(struct file *file, struct address_space *mapping,
919 loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
920{
921 int ret;
922
923 ret = block_write_begin(mapping, pos, len, foliop, ext2_get_block);
924 if (ret < 0)
925 ext2_write_failed(mapping, pos + len);
926 return ret;
927}
928
929static int ext2_write_end(struct file *file, struct address_space *mapping,
930 loff_t pos, unsigned len, unsigned copied,
931 struct folio *folio, void *fsdata)
932{
933 int ret;
934
935 ret = generic_write_end(file, mapping, pos, len, copied, folio, fsdata);
936 if (ret < len)
937 ext2_write_failed(mapping, pos + len);
938 return ret;
939}
940
941static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
942{
943 return generic_block_bmap(mapping,block,ext2_get_block);
944}
945
946static int
947ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
948{
949 return mpage_writepages(mapping, wbc, ext2_get_block);
950}
951
952static int
953ext2_dax_writepages(struct address_space *mapping, struct writeback_control *wbc)
954{
955 struct ext2_sb_info *sbi = EXT2_SB(mapping->host->i_sb);
956
957 return dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
958}
959
960const struct address_space_operations ext2_aops = {
961 .dirty_folio = block_dirty_folio,
962 .invalidate_folio = block_invalidate_folio,
963 .read_folio = ext2_read_folio,
964 .readahead = ext2_readahead,
965 .write_begin = ext2_write_begin,
966 .write_end = ext2_write_end,
967 .bmap = ext2_bmap,
968 .writepages = ext2_writepages,
969 .migrate_folio = buffer_migrate_folio,
970 .is_partially_uptodate = block_is_partially_uptodate,
971 .error_remove_folio = generic_error_remove_folio,
972};
973
974static const struct address_space_operations ext2_dax_aops = {
975 .writepages = ext2_dax_writepages,
976 .dirty_folio = noop_dirty_folio,
977};
978
979/*
980 * Probably it should be a library function... search for first non-zero word
981 * or memcmp with zero_page, whatever is better for particular architecture.
982 * Linus?
983 */
984static inline int all_zeroes(__le32 *p, __le32 *q)
985{
986 while (p < q)
987 if (*p++)
988 return 0;
989 return 1;
990}
991
992/**
993 * ext2_find_shared - find the indirect blocks for partial truncation.
994 * @inode: inode in question
995 * @depth: depth of the affected branch
996 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
997 * @chain: place to store the pointers to partial indirect blocks
998 * @top: place to the (detached) top of branch
999 *
1000 * This is a helper function used by ext2_truncate().
1001 *
1002 * When we do truncate() we may have to clean the ends of several indirect
1003 * blocks but leave the blocks themselves alive. Block is partially
1004 * truncated if some data below the new i_size is referred from it (and
1005 * it is on the path to the first completely truncated data block, indeed).
1006 * We have to free the top of that path along with everything to the right
1007 * of the path. Since no allocation past the truncation point is possible
1008 * until ext2_truncate() finishes, we may safely do the latter, but top
1009 * of branch may require special attention - pageout below the truncation
1010 * point might try to populate it.
1011 *
1012 * We atomically detach the top of branch from the tree, store the block
1013 * number of its root in *@top, pointers to buffer_heads of partially
1014 * truncated blocks - in @chain[].bh and pointers to their last elements
1015 * that should not be removed - in @chain[].p. Return value is the pointer
1016 * to last filled element of @chain.
1017 *
1018 * The work left to caller to do the actual freeing of subtrees:
1019 * a) free the subtree starting from *@top
1020 * b) free the subtrees whose roots are stored in
1021 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1022 * c) free the subtrees growing from the inode past the @chain[0].p
1023 * (no partially truncated stuff there).
1024 */
1025
1026static Indirect *ext2_find_shared(struct inode *inode,
1027 int depth,
1028 int offsets[4],
1029 Indirect chain[4],
1030 __le32 *top)
1031{
1032 Indirect *partial, *p;
1033 int k, err;
1034
1035 *top = 0;
1036 for (k = depth; k > 1 && !offsets[k-1]; k--)
1037 ;
1038 partial = ext2_get_branch(inode, k, offsets, chain, &err);
1039 if (!partial)
1040 partial = chain + k-1;
1041 /*
1042 * If the branch acquired continuation since we've looked at it -
1043 * fine, it should all survive and (new) top doesn't belong to us.
1044 */
1045 write_lock(&EXT2_I(inode)->i_meta_lock);
1046 if (!partial->key && *partial->p) {
1047 write_unlock(&EXT2_I(inode)->i_meta_lock);
1048 goto no_top;
1049 }
1050 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1051 ;
1052 /*
1053 * OK, we've found the last block that must survive. The rest of our
1054 * branch should be detached before unlocking. However, if that rest
1055 * of branch is all ours and does not grow immediately from the inode
1056 * it's easier to cheat and just decrement partial->p.
1057 */
1058 if (p == chain + k - 1 && p > chain) {
1059 p->p--;
1060 } else {
1061 *top = *p->p;
1062 *p->p = 0;
1063 }
1064 write_unlock(&EXT2_I(inode)->i_meta_lock);
1065
1066 while(partial > p)
1067 {
1068 brelse(partial->bh);
1069 partial--;
1070 }
1071no_top:
1072 return partial;
1073}
1074
1075/**
1076 * ext2_free_data - free a list of data blocks
1077 * @inode: inode we are dealing with
1078 * @p: array of block numbers
1079 * @q: points immediately past the end of array
1080 *
1081 * We are freeing all blocks referred from that array (numbers are
1082 * stored as little-endian 32-bit) and updating @inode->i_blocks
1083 * appropriately.
1084 */
1085static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1086{
1087 ext2_fsblk_t block_to_free = 0, count = 0;
1088 ext2_fsblk_t nr;
1089
1090 for ( ; p < q ; p++) {
1091 nr = le32_to_cpu(*p);
1092 if (nr) {
1093 *p = 0;
1094 /* accumulate blocks to free if they're contiguous */
1095 if (count == 0)
1096 goto free_this;
1097 else if (block_to_free == nr - count)
1098 count++;
1099 else {
1100 ext2_free_blocks (inode, block_to_free, count);
1101 mark_inode_dirty(inode);
1102 free_this:
1103 block_to_free = nr;
1104 count = 1;
1105 }
1106 }
1107 }
1108 if (count > 0) {
1109 ext2_free_blocks (inode, block_to_free, count);
1110 mark_inode_dirty(inode);
1111 }
1112}
1113
1114/**
1115 * ext2_free_branches - free an array of branches
1116 * @inode: inode we are dealing with
1117 * @p: array of block numbers
1118 * @q: pointer immediately past the end of array
1119 * @depth: depth of the branches to free
1120 *
1121 * We are freeing all blocks referred from these branches (numbers are
1122 * stored as little-endian 32-bit) and updating @inode->i_blocks
1123 * appropriately.
1124 */
1125static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1126{
1127 struct buffer_head * bh;
1128 ext2_fsblk_t nr;
1129
1130 if (depth--) {
1131 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1132 for ( ; p < q ; p++) {
1133 nr = le32_to_cpu(*p);
1134 if (!nr)
1135 continue;
1136 *p = 0;
1137 bh = sb_bread(inode->i_sb, nr);
1138 /*
1139 * A read failure? Report error and clear slot
1140 * (should be rare).
1141 */
1142 if (!bh) {
1143 ext2_error(inode->i_sb, "ext2_free_branches",
1144 "Read failure, inode=%ld, block=%ld",
1145 inode->i_ino, nr);
1146 continue;
1147 }
1148 ext2_free_branches(inode,
1149 (__le32*)bh->b_data,
1150 (__le32*)bh->b_data + addr_per_block,
1151 depth);
1152 bforget(bh);
1153 ext2_free_blocks(inode, nr, 1);
1154 mark_inode_dirty(inode);
1155 }
1156 } else
1157 ext2_free_data(inode, p, q);
1158}
1159
1160/* mapping->invalidate_lock must be held when calling this function */
1161static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1162{
1163 __le32 *i_data = EXT2_I(inode)->i_data;
1164 struct ext2_inode_info *ei = EXT2_I(inode);
1165 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1166 int offsets[4];
1167 Indirect chain[4];
1168 Indirect *partial;
1169 __le32 nr = 0;
1170 int n;
1171 long iblock;
1172 unsigned blocksize;
1173 blocksize = inode->i_sb->s_blocksize;
1174 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1175
1176#ifdef CONFIG_FS_DAX
1177 WARN_ON(!rwsem_is_locked(&inode->i_mapping->invalidate_lock));
1178#endif
1179
1180 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1181 if (n == 0)
1182 return;
1183
1184 /*
1185 * From here we block out all ext2_get_block() callers who want to
1186 * modify the block allocation tree.
1187 */
1188 mutex_lock(&ei->truncate_mutex);
1189
1190 if (n == 1) {
1191 ext2_free_data(inode, i_data+offsets[0],
1192 i_data + EXT2_NDIR_BLOCKS);
1193 goto do_indirects;
1194 }
1195
1196 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1197 /* Kill the top of shared branch (already detached) */
1198 if (nr) {
1199 if (partial == chain)
1200 mark_inode_dirty(inode);
1201 else
1202 mark_buffer_dirty_inode(partial->bh, inode);
1203 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1204 }
1205 /* Clear the ends of indirect blocks on the shared branch */
1206 while (partial > chain) {
1207 ext2_free_branches(inode,
1208 partial->p + 1,
1209 (__le32*)partial->bh->b_data+addr_per_block,
1210 (chain+n-1) - partial);
1211 mark_buffer_dirty_inode(partial->bh, inode);
1212 brelse (partial->bh);
1213 partial--;
1214 }
1215do_indirects:
1216 /* Kill the remaining (whole) subtrees */
1217 switch (offsets[0]) {
1218 default:
1219 nr = i_data[EXT2_IND_BLOCK];
1220 if (nr) {
1221 i_data[EXT2_IND_BLOCK] = 0;
1222 mark_inode_dirty(inode);
1223 ext2_free_branches(inode, &nr, &nr+1, 1);
1224 }
1225 fallthrough;
1226 case EXT2_IND_BLOCK:
1227 nr = i_data[EXT2_DIND_BLOCK];
1228 if (nr) {
1229 i_data[EXT2_DIND_BLOCK] = 0;
1230 mark_inode_dirty(inode);
1231 ext2_free_branches(inode, &nr, &nr+1, 2);
1232 }
1233 fallthrough;
1234 case EXT2_DIND_BLOCK:
1235 nr = i_data[EXT2_TIND_BLOCK];
1236 if (nr) {
1237 i_data[EXT2_TIND_BLOCK] = 0;
1238 mark_inode_dirty(inode);
1239 ext2_free_branches(inode, &nr, &nr+1, 3);
1240 }
1241 break;
1242 case EXT2_TIND_BLOCK:
1243 ;
1244 }
1245
1246 ext2_discard_reservation(inode);
1247
1248 mutex_unlock(&ei->truncate_mutex);
1249}
1250
1251static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1252{
1253 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1254 S_ISLNK(inode->i_mode)))
1255 return;
1256 if (ext2_inode_is_fast_symlink(inode))
1257 return;
1258
1259 filemap_invalidate_lock(inode->i_mapping);
1260 __ext2_truncate_blocks(inode, offset);
1261 filemap_invalidate_unlock(inode->i_mapping);
1262}
1263
1264static int ext2_setsize(struct inode *inode, loff_t newsize)
1265{
1266 int error;
1267
1268 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1269 S_ISLNK(inode->i_mode)))
1270 return -EINVAL;
1271 if (ext2_inode_is_fast_symlink(inode))
1272 return -EINVAL;
1273 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1274 return -EPERM;
1275
1276 inode_dio_wait(inode);
1277
1278 if (IS_DAX(inode))
1279 error = dax_truncate_page(inode, newsize, NULL,
1280 &ext2_iomap_ops);
1281 else
1282 error = block_truncate_page(inode->i_mapping,
1283 newsize, ext2_get_block);
1284 if (error)
1285 return error;
1286
1287 filemap_invalidate_lock(inode->i_mapping);
1288 truncate_setsize(inode, newsize);
1289 __ext2_truncate_blocks(inode, newsize);
1290 filemap_invalidate_unlock(inode->i_mapping);
1291
1292 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1293 if (inode_needs_sync(inode)) {
1294 sync_mapping_buffers(inode->i_mapping);
1295 sync_inode_metadata(inode, 1);
1296 } else {
1297 mark_inode_dirty(inode);
1298 }
1299
1300 return 0;
1301}
1302
1303static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1304 struct buffer_head **p)
1305{
1306 struct buffer_head * bh;
1307 unsigned long block_group;
1308 unsigned long block;
1309 unsigned long offset;
1310 struct ext2_group_desc * gdp;
1311
1312 *p = NULL;
1313 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1314 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1315 goto Einval;
1316
1317 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1318 gdp = ext2_get_group_desc(sb, block_group, NULL);
1319 if (!gdp)
1320 goto Egdp;
1321 /*
1322 * Figure out the offset within the block group inode table
1323 */
1324 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1325 block = le32_to_cpu(gdp->bg_inode_table) +
1326 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1327 if (!(bh = sb_bread(sb, block)))
1328 goto Eio;
1329
1330 *p = bh;
1331 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1332 return (struct ext2_inode *) (bh->b_data + offset);
1333
1334Einval:
1335 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1336 (unsigned long) ino);
1337 return ERR_PTR(-EINVAL);
1338Eio:
1339 ext2_error(sb, "ext2_get_inode",
1340 "unable to read inode block - inode=%lu, block=%lu",
1341 (unsigned long) ino, block);
1342Egdp:
1343 return ERR_PTR(-EIO);
1344}
1345
1346void ext2_set_inode_flags(struct inode *inode)
1347{
1348 unsigned int flags = EXT2_I(inode)->i_flags;
1349
1350 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1351 S_DIRSYNC | S_DAX);
1352 if (flags & EXT2_SYNC_FL)
1353 inode->i_flags |= S_SYNC;
1354 if (flags & EXT2_APPEND_FL)
1355 inode->i_flags |= S_APPEND;
1356 if (flags & EXT2_IMMUTABLE_FL)
1357 inode->i_flags |= S_IMMUTABLE;
1358 if (flags & EXT2_NOATIME_FL)
1359 inode->i_flags |= S_NOATIME;
1360 if (flags & EXT2_DIRSYNC_FL)
1361 inode->i_flags |= S_DIRSYNC;
1362 if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1363 inode->i_flags |= S_DAX;
1364}
1365
1366void ext2_set_file_ops(struct inode *inode)
1367{
1368 inode->i_op = &ext2_file_inode_operations;
1369 inode->i_fop = &ext2_file_operations;
1370 if (IS_DAX(inode))
1371 inode->i_mapping->a_ops = &ext2_dax_aops;
1372 else
1373 inode->i_mapping->a_ops = &ext2_aops;
1374}
1375
1376struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1377{
1378 struct ext2_inode_info *ei;
1379 struct buffer_head * bh = NULL;
1380 struct ext2_inode *raw_inode;
1381 struct inode *inode;
1382 long ret = -EIO;
1383 int n;
1384 uid_t i_uid;
1385 gid_t i_gid;
1386
1387 inode = iget_locked(sb, ino);
1388 if (!inode)
1389 return ERR_PTR(-ENOMEM);
1390 if (!(inode->i_state & I_NEW))
1391 return inode;
1392
1393 ei = EXT2_I(inode);
1394 ei->i_block_alloc_info = NULL;
1395
1396 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1397 if (IS_ERR(raw_inode)) {
1398 ret = PTR_ERR(raw_inode);
1399 goto bad_inode;
1400 }
1401
1402 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1403 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1404 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1405 if (!(test_opt (inode->i_sb, NO_UID32))) {
1406 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1407 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1408 }
1409 i_uid_write(inode, i_uid);
1410 i_gid_write(inode, i_gid);
1411 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1412 inode->i_size = le32_to_cpu(raw_inode->i_size);
1413 inode_set_atime(inode, (signed)le32_to_cpu(raw_inode->i_atime), 0);
1414 inode_set_ctime(inode, (signed)le32_to_cpu(raw_inode->i_ctime), 0);
1415 inode_set_mtime(inode, (signed)le32_to_cpu(raw_inode->i_mtime), 0);
1416 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1417 /* We now have enough fields to check if the inode was active or not.
1418 * This is needed because nfsd might try to access dead inodes
1419 * the test is that same one that e2fsck uses
1420 * NeilBrown 1999oct15
1421 */
1422 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1423 /* this inode is deleted */
1424 ret = -ESTALE;
1425 goto bad_inode;
1426 }
1427 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1428 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1429 ext2_set_inode_flags(inode);
1430 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1431 ei->i_frag_no = raw_inode->i_frag;
1432 ei->i_frag_size = raw_inode->i_fsize;
1433 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1434 ei->i_dir_acl = 0;
1435
1436 if (ei->i_file_acl &&
1437 !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1438 ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1439 ei->i_file_acl);
1440 ret = -EFSCORRUPTED;
1441 goto bad_inode;
1442 }
1443
1444 if (S_ISREG(inode->i_mode))
1445 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1446 else
1447 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1448 if (i_size_read(inode) < 0) {
1449 ret = -EFSCORRUPTED;
1450 goto bad_inode;
1451 }
1452 ei->i_dtime = 0;
1453 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1454 ei->i_state = 0;
1455 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1456 ei->i_dir_start_lookup = 0;
1457
1458 /*
1459 * NOTE! The in-memory inode i_data array is in little-endian order
1460 * even on big-endian machines: we do NOT byteswap the block numbers!
1461 */
1462 for (n = 0; n < EXT2_N_BLOCKS; n++)
1463 ei->i_data[n] = raw_inode->i_block[n];
1464
1465 if (S_ISREG(inode->i_mode)) {
1466 ext2_set_file_ops(inode);
1467 } else if (S_ISDIR(inode->i_mode)) {
1468 inode->i_op = &ext2_dir_inode_operations;
1469 inode->i_fop = &ext2_dir_operations;
1470 inode->i_mapping->a_ops = &ext2_aops;
1471 } else if (S_ISLNK(inode->i_mode)) {
1472 if (ext2_inode_is_fast_symlink(inode)) {
1473 inode->i_link = (char *)ei->i_data;
1474 inode->i_op = &ext2_fast_symlink_inode_operations;
1475 nd_terminate_link(ei->i_data, inode->i_size,
1476 sizeof(ei->i_data) - 1);
1477 } else {
1478 inode->i_op = &ext2_symlink_inode_operations;
1479 inode_nohighmem(inode);
1480 inode->i_mapping->a_ops = &ext2_aops;
1481 }
1482 } else {
1483 inode->i_op = &ext2_special_inode_operations;
1484 if (raw_inode->i_block[0])
1485 init_special_inode(inode, inode->i_mode,
1486 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1487 else
1488 init_special_inode(inode, inode->i_mode,
1489 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1490 }
1491 brelse (bh);
1492 unlock_new_inode(inode);
1493 return inode;
1494
1495bad_inode:
1496 brelse(bh);
1497 iget_failed(inode);
1498 return ERR_PTR(ret);
1499}
1500
1501static int __ext2_write_inode(struct inode *inode, int do_sync)
1502{
1503 struct ext2_inode_info *ei = EXT2_I(inode);
1504 struct super_block *sb = inode->i_sb;
1505 ino_t ino = inode->i_ino;
1506 uid_t uid = i_uid_read(inode);
1507 gid_t gid = i_gid_read(inode);
1508 struct buffer_head * bh;
1509 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1510 int n;
1511 int err = 0;
1512
1513 if (IS_ERR(raw_inode))
1514 return -EIO;
1515
1516 /* For fields not tracking in the in-memory inode,
1517 * initialise them to zero for new inodes. */
1518 if (ei->i_state & EXT2_STATE_NEW)
1519 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1520
1521 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1522 if (!(test_opt(sb, NO_UID32))) {
1523 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1524 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1525/*
1526 * Fix up interoperability with old kernels. Otherwise, old inodes get
1527 * re-used with the upper 16 bits of the uid/gid intact
1528 */
1529 if (!ei->i_dtime) {
1530 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1531 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1532 } else {
1533 raw_inode->i_uid_high = 0;
1534 raw_inode->i_gid_high = 0;
1535 }
1536 } else {
1537 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1538 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1539 raw_inode->i_uid_high = 0;
1540 raw_inode->i_gid_high = 0;
1541 }
1542 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1543 raw_inode->i_size = cpu_to_le32(inode->i_size);
1544 raw_inode->i_atime = cpu_to_le32(inode_get_atime_sec(inode));
1545 raw_inode->i_ctime = cpu_to_le32(inode_get_ctime_sec(inode));
1546 raw_inode->i_mtime = cpu_to_le32(inode_get_mtime_sec(inode));
1547
1548 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1549 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1550 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1551 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1552 raw_inode->i_frag = ei->i_frag_no;
1553 raw_inode->i_fsize = ei->i_frag_size;
1554 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1555 if (!S_ISREG(inode->i_mode))
1556 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1557 else {
1558 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1559 if (inode->i_size > 0x7fffffffULL) {
1560 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1561 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1562 EXT2_SB(sb)->s_es->s_rev_level ==
1563 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1564 /* If this is the first large file
1565 * created, add a flag to the superblock.
1566 */
1567 spin_lock(&EXT2_SB(sb)->s_lock);
1568 ext2_update_dynamic_rev(sb);
1569 EXT2_SET_RO_COMPAT_FEATURE(sb,
1570 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1571 spin_unlock(&EXT2_SB(sb)->s_lock);
1572 ext2_sync_super(sb, EXT2_SB(sb)->s_es, 1);
1573 }
1574 }
1575 }
1576
1577 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1578 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1579 if (old_valid_dev(inode->i_rdev)) {
1580 raw_inode->i_block[0] =
1581 cpu_to_le32(old_encode_dev(inode->i_rdev));
1582 raw_inode->i_block[1] = 0;
1583 } else {
1584 raw_inode->i_block[0] = 0;
1585 raw_inode->i_block[1] =
1586 cpu_to_le32(new_encode_dev(inode->i_rdev));
1587 raw_inode->i_block[2] = 0;
1588 }
1589 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1590 raw_inode->i_block[n] = ei->i_data[n];
1591 mark_buffer_dirty(bh);
1592 if (do_sync) {
1593 sync_dirty_buffer(bh);
1594 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1595 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1596 sb->s_id, (unsigned long) ino);
1597 err = -EIO;
1598 }
1599 }
1600 ei->i_state &= ~EXT2_STATE_NEW;
1601 brelse (bh);
1602 return err;
1603}
1604
1605int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1606{
1607 return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1608}
1609
1610int ext2_getattr(struct mnt_idmap *idmap, const struct path *path,
1611 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1612{
1613 struct inode *inode = d_inode(path->dentry);
1614 struct ext2_inode_info *ei = EXT2_I(inode);
1615 unsigned int flags;
1616
1617 flags = ei->i_flags & EXT2_FL_USER_VISIBLE;
1618 if (flags & EXT2_APPEND_FL)
1619 stat->attributes |= STATX_ATTR_APPEND;
1620 if (flags & EXT2_COMPR_FL)
1621 stat->attributes |= STATX_ATTR_COMPRESSED;
1622 if (flags & EXT2_IMMUTABLE_FL)
1623 stat->attributes |= STATX_ATTR_IMMUTABLE;
1624 if (flags & EXT2_NODUMP_FL)
1625 stat->attributes |= STATX_ATTR_NODUMP;
1626 stat->attributes_mask |= (STATX_ATTR_APPEND |
1627 STATX_ATTR_COMPRESSED |
1628 STATX_ATTR_ENCRYPTED |
1629 STATX_ATTR_IMMUTABLE |
1630 STATX_ATTR_NODUMP);
1631
1632 generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1633 return 0;
1634}
1635
1636int ext2_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1637 struct iattr *iattr)
1638{
1639 struct inode *inode = d_inode(dentry);
1640 int error;
1641
1642 error = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
1643 if (error)
1644 return error;
1645
1646 if (is_quota_modification(&nop_mnt_idmap, inode, iattr)) {
1647 error = dquot_initialize(inode);
1648 if (error)
1649 return error;
1650 }
1651 if (i_uid_needs_update(&nop_mnt_idmap, iattr, inode) ||
1652 i_gid_needs_update(&nop_mnt_idmap, iattr, inode)) {
1653 error = dquot_transfer(&nop_mnt_idmap, inode, iattr);
1654 if (error)
1655 return error;
1656 }
1657 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1658 error = ext2_setsize(inode, iattr->ia_size);
1659 if (error)
1660 return error;
1661 }
1662 setattr_copy(&nop_mnt_idmap, inode, iattr);
1663 if (iattr->ia_valid & ATTR_MODE)
1664 error = posix_acl_chmod(&nop_mnt_idmap, dentry, inode->i_mode);
1665 mark_inode_dirty(inode);
1666
1667 return error;
1668}