Linux 4.3-rc5
[linux/fpc-iii.git] / fs / ext2 / inode.c
blobc60a248c640cb9f23494ed3442ad161639922e7b
1 /*
2 * linux/fs/ext2/inode.c
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)
9 * from
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
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)
22 * Assorted race fixes, rewrite of ext2_get_block() by Al Viro, 2000
25 #include <linux/time.h>
26 #include <linux/highuid.h>
27 #include <linux/pagemap.h>
28 #include <linux/dax.h>
29 #include <linux/quotaops.h>
30 #include <linux/writeback.h>
31 #include <linux/buffer_head.h>
32 #include <linux/mpage.h>
33 #include <linux/fiemap.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include "ext2.h"
37 #include "acl.h"
38 #include "xattr.h"
40 static int __ext2_write_inode(struct inode *inode, int do_sync);
43 * Test whether an inode is a fast symlink.
45 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
47 int ea_blocks = EXT2_I(inode)->i_file_acl ?
48 (inode->i_sb->s_blocksize >> 9) : 0;
50 return (S_ISLNK(inode->i_mode) &&
51 inode->i_blocks - ea_blocks == 0);
54 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
56 static void ext2_write_failed(struct address_space *mapping, loff_t to)
58 struct inode *inode = mapping->host;
60 if (to > inode->i_size) {
61 truncate_pagecache(inode, inode->i_size);
62 ext2_truncate_blocks(inode, inode->i_size);
67 * Called at the last iput() if i_nlink is zero.
69 void ext2_evict_inode(struct inode * inode)
71 struct ext2_block_alloc_info *rsv;
72 int want_delete = 0;
74 if (!inode->i_nlink && !is_bad_inode(inode)) {
75 want_delete = 1;
76 dquot_initialize(inode);
77 } else {
78 dquot_drop(inode);
81 truncate_inode_pages_final(&inode->i_data);
83 if (want_delete) {
84 sb_start_intwrite(inode->i_sb);
85 /* set dtime */
86 EXT2_I(inode)->i_dtime = get_seconds();
87 mark_inode_dirty(inode);
88 __ext2_write_inode(inode, inode_needs_sync(inode));
89 /* truncate to 0 */
90 inode->i_size = 0;
91 if (inode->i_blocks)
92 ext2_truncate_blocks(inode, 0);
93 ext2_xattr_delete_inode(inode);
96 invalidate_inode_buffers(inode);
97 clear_inode(inode);
99 ext2_discard_reservation(inode);
100 rsv = EXT2_I(inode)->i_block_alloc_info;
101 EXT2_I(inode)->i_block_alloc_info = NULL;
102 if (unlikely(rsv))
103 kfree(rsv);
105 if (want_delete) {
106 ext2_free_inode(inode);
107 sb_end_intwrite(inode->i_sb);
111 typedef struct {
112 __le32 *p;
113 __le32 key;
114 struct buffer_head *bh;
115 } Indirect;
117 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
119 p->key = *(p->p = v);
120 p->bh = bh;
123 static inline int verify_chain(Indirect *from, Indirect *to)
125 while (from <= to && from->key == *from->p)
126 from++;
127 return (from > to);
131 * ext2_block_to_path - parse the block number into array of offsets
132 * @inode: inode in question (we are only interested in its superblock)
133 * @i_block: block number to be parsed
134 * @offsets: array to store the offsets in
135 * @boundary: set this non-zero if the referred-to block is likely to be
136 * followed (on disk) by an indirect block.
137 * To store the locations of file's data ext2 uses a data structure common
138 * for UNIX filesystems - tree of pointers anchored in the inode, with
139 * data blocks at leaves and indirect blocks in intermediate nodes.
140 * This function translates the block number into path in that tree -
141 * return value is the path length and @offsets[n] is the offset of
142 * pointer to (n+1)th node in the nth one. If @block is out of range
143 * (negative or too large) warning is printed and zero returned.
145 * Note: function doesn't find node addresses, so no IO is needed. All
146 * we need to know is the capacity of indirect blocks (taken from the
147 * inode->i_sb).
151 * Portability note: the last comparison (check that we fit into triple
152 * indirect block) is spelled differently, because otherwise on an
153 * architecture with 32-bit longs and 8Kb pages we might get into trouble
154 * if our filesystem had 8Kb blocks. We might use long long, but that would
155 * kill us on x86. Oh, well, at least the sign propagation does not matter -
156 * i_block would have to be negative in the very beginning, so we would not
157 * get there at all.
160 static int ext2_block_to_path(struct inode *inode,
161 long i_block, int offsets[4], int *boundary)
163 int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
164 int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
165 const long direct_blocks = EXT2_NDIR_BLOCKS,
166 indirect_blocks = ptrs,
167 double_blocks = (1 << (ptrs_bits * 2));
168 int n = 0;
169 int final = 0;
171 if (i_block < 0) {
172 ext2_msg(inode->i_sb, KERN_WARNING,
173 "warning: %s: block < 0", __func__);
174 } else if (i_block < direct_blocks) {
175 offsets[n++] = i_block;
176 final = direct_blocks;
177 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
178 offsets[n++] = EXT2_IND_BLOCK;
179 offsets[n++] = i_block;
180 final = ptrs;
181 } else if ((i_block -= indirect_blocks) < double_blocks) {
182 offsets[n++] = EXT2_DIND_BLOCK;
183 offsets[n++] = i_block >> ptrs_bits;
184 offsets[n++] = i_block & (ptrs - 1);
185 final = ptrs;
186 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
187 offsets[n++] = EXT2_TIND_BLOCK;
188 offsets[n++] = i_block >> (ptrs_bits * 2);
189 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
190 offsets[n++] = i_block & (ptrs - 1);
191 final = ptrs;
192 } else {
193 ext2_msg(inode->i_sb, KERN_WARNING,
194 "warning: %s: block is too big", __func__);
196 if (boundary)
197 *boundary = final - 1 - (i_block & (ptrs - 1));
199 return n;
203 * ext2_get_branch - read the chain of indirect blocks leading to data
204 * @inode: inode in question
205 * @depth: depth of the chain (1 - direct pointer, etc.)
206 * @offsets: offsets of pointers in inode/indirect blocks
207 * @chain: place to store the result
208 * @err: here we store the error value
210 * Function fills the array of triples <key, p, bh> and returns %NULL
211 * if everything went OK or the pointer to the last filled triple
212 * (incomplete one) otherwise. Upon the return chain[i].key contains
213 * the number of (i+1)-th block in the chain (as it is stored in memory,
214 * i.e. little-endian 32-bit), chain[i].p contains the address of that
215 * number (it points into struct inode for i==0 and into the bh->b_data
216 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
217 * block for i>0 and NULL for i==0. In other words, it holds the block
218 * numbers of the chain, addresses they were taken from (and where we can
219 * verify that chain did not change) and buffer_heads hosting these
220 * numbers.
222 * Function stops when it stumbles upon zero pointer (absent block)
223 * (pointer to last triple returned, *@err == 0)
224 * or when it gets an IO error reading an indirect block
225 * (ditto, *@err == -EIO)
226 * or when it notices that chain had been changed while it was reading
227 * (ditto, *@err == -EAGAIN)
228 * or when it reads all @depth-1 indirect blocks successfully and finds
229 * the whole chain, all way to the data (returns %NULL, *err == 0).
231 static Indirect *ext2_get_branch(struct inode *inode,
232 int depth,
233 int *offsets,
234 Indirect chain[4],
235 int *err)
237 struct super_block *sb = inode->i_sb;
238 Indirect *p = chain;
239 struct buffer_head *bh;
241 *err = 0;
242 /* i_data is not going away, no lock needed */
243 add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
244 if (!p->key)
245 goto no_block;
246 while (--depth) {
247 bh = sb_bread(sb, le32_to_cpu(p->key));
248 if (!bh)
249 goto failure;
250 read_lock(&EXT2_I(inode)->i_meta_lock);
251 if (!verify_chain(chain, p))
252 goto changed;
253 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
254 read_unlock(&EXT2_I(inode)->i_meta_lock);
255 if (!p->key)
256 goto no_block;
258 return NULL;
260 changed:
261 read_unlock(&EXT2_I(inode)->i_meta_lock);
262 brelse(bh);
263 *err = -EAGAIN;
264 goto no_block;
265 failure:
266 *err = -EIO;
267 no_block:
268 return p;
272 * ext2_find_near - find a place for allocation with sufficient locality
273 * @inode: owner
274 * @ind: descriptor of indirect block.
276 * This function returns the preferred place for block allocation.
277 * It is used when heuristic for sequential allocation fails.
278 * Rules are:
279 * + if there is a block to the left of our position - allocate near it.
280 * + if pointer will live in indirect block - allocate near that block.
281 * + if pointer will live in inode - allocate in the same cylinder group.
283 * In the latter case we colour the starting block by the callers PID to
284 * prevent it from clashing with concurrent allocations for a different inode
285 * in the same block group. The PID is used here so that functionally related
286 * files will be close-by on-disk.
288 * Caller must make sure that @ind is valid and will stay that way.
291 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
293 struct ext2_inode_info *ei = EXT2_I(inode);
294 __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
295 __le32 *p;
296 ext2_fsblk_t bg_start;
297 ext2_fsblk_t colour;
299 /* Try to find previous block */
300 for (p = ind->p - 1; p >= start; p--)
301 if (*p)
302 return le32_to_cpu(*p);
304 /* No such thing, so let's try location of indirect block */
305 if (ind->bh)
306 return ind->bh->b_blocknr;
309 * It is going to be referred from inode itself? OK, just put it into
310 * the same cylinder group then.
312 bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
313 colour = (current->pid % 16) *
314 (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
315 return bg_start + colour;
319 * ext2_find_goal - find a preferred place for allocation.
320 * @inode: owner
321 * @block: block we want
322 * @partial: pointer to the last triple within a chain
324 * Returns preferred place for a block (the goal).
327 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
328 Indirect *partial)
330 struct ext2_block_alloc_info *block_i;
332 block_i = EXT2_I(inode)->i_block_alloc_info;
335 * try the heuristic for sequential allocation,
336 * failing that at least try to get decent locality.
338 if (block_i && (block == block_i->last_alloc_logical_block + 1)
339 && (block_i->last_alloc_physical_block != 0)) {
340 return block_i->last_alloc_physical_block + 1;
343 return ext2_find_near(inode, partial);
347 * ext2_blks_to_allocate: Look up the block map and count the number
348 * of direct blocks need to be allocated for the given branch.
350 * @branch: chain of indirect blocks
351 * @k: number of blocks need for indirect blocks
352 * @blks: number of data blocks to be mapped.
353 * @blocks_to_boundary: the offset in the indirect block
355 * return the total number of blocks to be allocate, including the
356 * direct and indirect blocks.
358 static int
359 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
360 int blocks_to_boundary)
362 unsigned long count = 0;
365 * Simple case, [t,d]Indirect block(s) has not allocated yet
366 * then it's clear blocks on that path have not allocated
368 if (k > 0) {
369 /* right now don't hanel cross boundary allocation */
370 if (blks < blocks_to_boundary + 1)
371 count += blks;
372 else
373 count += blocks_to_boundary + 1;
374 return count;
377 count++;
378 while (count < blks && count <= blocks_to_boundary
379 && le32_to_cpu(*(branch[0].p + count)) == 0) {
380 count++;
382 return count;
386 * ext2_alloc_blocks: multiple allocate blocks needed for a branch
387 * @indirect_blks: the number of blocks need to allocate for indirect
388 * blocks
390 * @new_blocks: on return it will store the new block numbers for
391 * the indirect blocks(if needed) and the first direct block,
392 * @blks: on return it will store the total number of allocated
393 * direct blocks
395 static int ext2_alloc_blocks(struct inode *inode,
396 ext2_fsblk_t goal, int indirect_blks, int blks,
397 ext2_fsblk_t new_blocks[4], int *err)
399 int target, i;
400 unsigned long count = 0;
401 int index = 0;
402 ext2_fsblk_t current_block = 0;
403 int ret = 0;
406 * Here we try to allocate the requested multiple blocks at once,
407 * on a best-effort basis.
408 * To build a branch, we should allocate blocks for
409 * the indirect blocks(if not allocated yet), and at least
410 * the first direct block of this branch. That's the
411 * minimum number of blocks need to allocate(required)
413 target = blks + indirect_blks;
415 while (1) {
416 count = target;
417 /* allocating blocks for indirect blocks and direct blocks */
418 current_block = ext2_new_blocks(inode,goal,&count,err);
419 if (*err)
420 goto failed_out;
422 target -= count;
423 /* allocate blocks for indirect blocks */
424 while (index < indirect_blks && count) {
425 new_blocks[index++] = current_block++;
426 count--;
429 if (count > 0)
430 break;
433 /* save the new block number for the first direct block */
434 new_blocks[index] = current_block;
436 /* total number of blocks allocated for direct blocks */
437 ret = count;
438 *err = 0;
439 return ret;
440 failed_out:
441 for (i = 0; i <index; i++)
442 ext2_free_blocks(inode, new_blocks[i], 1);
443 if (index)
444 mark_inode_dirty(inode);
445 return ret;
449 * ext2_alloc_branch - allocate and set up a chain of blocks.
450 * @inode: owner
451 * @num: depth of the chain (number of blocks to allocate)
452 * @offsets: offsets (in the blocks) to store the pointers to next.
453 * @branch: place to store the chain in.
455 * This function allocates @num blocks, zeroes out all but the last one,
456 * links them into chain and (if we are synchronous) writes them to disk.
457 * In other words, it prepares a branch that can be spliced onto the
458 * inode. It stores the information about that chain in the branch[], in
459 * the same format as ext2_get_branch() would do. We are calling it after
460 * we had read the existing part of chain and partial points to the last
461 * triple of that (one with zero ->key). Upon the exit we have the same
462 * picture as after the successful ext2_get_block(), except that in one
463 * place chain is disconnected - *branch->p is still zero (we did not
464 * set the last link), but branch->key contains the number that should
465 * be placed into *branch->p to fill that gap.
467 * If allocation fails we free all blocks we've allocated (and forget
468 * their buffer_heads) and return the error value the from failed
469 * ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
470 * as described above and return 0.
473 static int ext2_alloc_branch(struct inode *inode,
474 int indirect_blks, int *blks, ext2_fsblk_t goal,
475 int *offsets, Indirect *branch)
477 int blocksize = inode->i_sb->s_blocksize;
478 int i, n = 0;
479 int err = 0;
480 struct buffer_head *bh;
481 int num;
482 ext2_fsblk_t new_blocks[4];
483 ext2_fsblk_t current_block;
485 num = ext2_alloc_blocks(inode, goal, indirect_blks,
486 *blks, new_blocks, &err);
487 if (err)
488 return err;
490 branch[0].key = cpu_to_le32(new_blocks[0]);
492 * metadata blocks and data blocks are allocated.
494 for (n = 1; n <= indirect_blks; n++) {
496 * Get buffer_head for parent block, zero it out
497 * and set the pointer to new one, then send
498 * parent to disk.
500 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
501 if (unlikely(!bh)) {
502 err = -ENOMEM;
503 goto failed;
505 branch[n].bh = bh;
506 lock_buffer(bh);
507 memset(bh->b_data, 0, blocksize);
508 branch[n].p = (__le32 *) bh->b_data + offsets[n];
509 branch[n].key = cpu_to_le32(new_blocks[n]);
510 *branch[n].p = branch[n].key;
511 if ( n == indirect_blks) {
512 current_block = new_blocks[n];
514 * End of chain, update the last new metablock of
515 * the chain to point to the new allocated
516 * data blocks numbers
518 for (i=1; i < num; i++)
519 *(branch[n].p + i) = cpu_to_le32(++current_block);
521 set_buffer_uptodate(bh);
522 unlock_buffer(bh);
523 mark_buffer_dirty_inode(bh, inode);
524 /* We used to sync bh here if IS_SYNC(inode).
525 * But we now rely upon generic_write_sync()
526 * and b_inode_buffers. But not for directories.
528 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
529 sync_dirty_buffer(bh);
531 *blks = num;
532 return err;
534 failed:
535 for (i = 1; i < n; i++)
536 bforget(branch[i].bh);
537 for (i = 0; i < indirect_blks; i++)
538 ext2_free_blocks(inode, new_blocks[i], 1);
539 ext2_free_blocks(inode, new_blocks[i], num);
540 return err;
544 * ext2_splice_branch - splice the allocated branch onto inode.
545 * @inode: owner
546 * @block: (logical) number of block we are adding
547 * @where: location of missing link
548 * @num: number of indirect blocks we are adding
549 * @blks: number of direct blocks we are adding
551 * This function fills the missing link and does all housekeeping needed in
552 * inode (->i_blocks, etc.). In case of success we end up with the full
553 * chain to new block and return 0.
555 static void ext2_splice_branch(struct inode *inode,
556 long block, Indirect *where, int num, int blks)
558 int i;
559 struct ext2_block_alloc_info *block_i;
560 ext2_fsblk_t current_block;
562 block_i = EXT2_I(inode)->i_block_alloc_info;
564 /* XXX LOCKING probably should have i_meta_lock ?*/
565 /* That's it */
567 *where->p = where->key;
570 * Update the host buffer_head or inode to point to more just allocated
571 * direct blocks blocks
573 if (num == 0 && blks > 1) {
574 current_block = le32_to_cpu(where->key) + 1;
575 for (i = 1; i < blks; i++)
576 *(where->p + i ) = cpu_to_le32(current_block++);
580 * update the most recently allocated logical & physical block
581 * in i_block_alloc_info, to assist find the proper goal block for next
582 * allocation
584 if (block_i) {
585 block_i->last_alloc_logical_block = block + blks - 1;
586 block_i->last_alloc_physical_block =
587 le32_to_cpu(where[num].key) + blks - 1;
590 /* We are done with atomic stuff, now do the rest of housekeeping */
592 /* had we spliced it onto indirect block? */
593 if (where->bh)
594 mark_buffer_dirty_inode(where->bh, inode);
596 inode->i_ctime = CURRENT_TIME_SEC;
597 mark_inode_dirty(inode);
601 * Allocation strategy is simple: if we have to allocate something, we will
602 * have to go the whole way to leaf. So let's do it before attaching anything
603 * to tree, set linkage between the newborn blocks, write them if sync is
604 * required, recheck the path, free and repeat if check fails, otherwise
605 * set the last missing link (that will protect us from any truncate-generated
606 * removals - all blocks on the path are immune now) and possibly force the
607 * write on the parent block.
608 * That has a nice additional property: no special recovery from the failed
609 * allocations is needed - we simply release blocks and do not touch anything
610 * reachable from inode.
612 * `handle' can be NULL if create == 0.
614 * return > 0, # of blocks mapped or allocated.
615 * return = 0, if plain lookup failed.
616 * return < 0, error case.
618 static int ext2_get_blocks(struct inode *inode,
619 sector_t iblock, unsigned long maxblocks,
620 struct buffer_head *bh_result,
621 int create)
623 int err = -EIO;
624 int offsets[4];
625 Indirect chain[4];
626 Indirect *partial;
627 ext2_fsblk_t goal;
628 int indirect_blks;
629 int blocks_to_boundary = 0;
630 int depth;
631 struct ext2_inode_info *ei = EXT2_I(inode);
632 int count = 0;
633 ext2_fsblk_t first_block = 0;
635 BUG_ON(maxblocks == 0);
637 depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
639 if (depth == 0)
640 return (err);
642 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
643 /* Simplest case - block found, no allocation needed */
644 if (!partial) {
645 first_block = le32_to_cpu(chain[depth - 1].key);
646 clear_buffer_new(bh_result); /* What's this do? */
647 count++;
648 /*map more blocks*/
649 while (count < maxblocks && count <= blocks_to_boundary) {
650 ext2_fsblk_t blk;
652 if (!verify_chain(chain, chain + depth - 1)) {
654 * Indirect block might be removed by
655 * truncate while we were reading it.
656 * Handling of that case: forget what we've
657 * got now, go to reread.
659 err = -EAGAIN;
660 count = 0;
661 break;
663 blk = le32_to_cpu(*(chain[depth-1].p + count));
664 if (blk == first_block + count)
665 count++;
666 else
667 break;
669 if (err != -EAGAIN)
670 goto got_it;
673 /* Next simple case - plain lookup or failed read of indirect block */
674 if (!create || err == -EIO)
675 goto cleanup;
677 mutex_lock(&ei->truncate_mutex);
679 * If the indirect block is missing while we are reading
680 * the chain(ext2_get_branch() returns -EAGAIN err), or
681 * if the chain has been changed after we grab the semaphore,
682 * (either because another process truncated this branch, or
683 * another get_block allocated this branch) re-grab the chain to see if
684 * the request block has been allocated or not.
686 * Since we already block the truncate/other get_block
687 * at this point, we will have the current copy of the chain when we
688 * splice the branch into the tree.
690 if (err == -EAGAIN || !verify_chain(chain, partial)) {
691 while (partial > chain) {
692 brelse(partial->bh);
693 partial--;
695 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
696 if (!partial) {
697 count++;
698 mutex_unlock(&ei->truncate_mutex);
699 if (err)
700 goto cleanup;
701 clear_buffer_new(bh_result);
702 goto got_it;
707 * Okay, we need to do block allocation. Lazily initialize the block
708 * allocation info here if necessary
710 if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
711 ext2_init_block_alloc_info(inode);
713 goal = ext2_find_goal(inode, iblock, partial);
715 /* the number of blocks need to allocate for [d,t]indirect blocks */
716 indirect_blks = (chain + depth) - partial - 1;
718 * Next look up the indirect map to count the totoal number of
719 * direct blocks to allocate for this branch.
721 count = ext2_blks_to_allocate(partial, indirect_blks,
722 maxblocks, blocks_to_boundary);
724 * XXX ???? Block out ext2_truncate while we alter the tree
726 err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
727 offsets + (partial - chain), partial);
729 if (err) {
730 mutex_unlock(&ei->truncate_mutex);
731 goto cleanup;
734 if (IS_DAX(inode)) {
736 * block must be initialised before we put it in the tree
737 * so that it's not found by another thread before it's
738 * initialised
740 err = dax_clear_blocks(inode, le32_to_cpu(chain[depth-1].key),
741 1 << inode->i_blkbits);
742 if (err) {
743 mutex_unlock(&ei->truncate_mutex);
744 goto cleanup;
748 ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
749 mutex_unlock(&ei->truncate_mutex);
750 set_buffer_new(bh_result);
751 got_it:
752 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
753 if (count > blocks_to_boundary)
754 set_buffer_boundary(bh_result);
755 err = count;
756 /* Clean up and exit */
757 partial = chain + depth - 1; /* the whole chain */
758 cleanup:
759 while (partial > chain) {
760 brelse(partial->bh);
761 partial--;
763 return err;
766 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
768 unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
769 int ret = ext2_get_blocks(inode, iblock, max_blocks,
770 bh_result, create);
771 if (ret > 0) {
772 bh_result->b_size = (ret << inode->i_blkbits);
773 ret = 0;
775 return ret;
779 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
780 u64 start, u64 len)
782 return generic_block_fiemap(inode, fieinfo, start, len,
783 ext2_get_block);
786 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
788 return block_write_full_page(page, ext2_get_block, wbc);
791 static int ext2_readpage(struct file *file, struct page *page)
793 return mpage_readpage(page, ext2_get_block);
796 static int
797 ext2_readpages(struct file *file, struct address_space *mapping,
798 struct list_head *pages, unsigned nr_pages)
800 return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
803 static int
804 ext2_write_begin(struct file *file, struct address_space *mapping,
805 loff_t pos, unsigned len, unsigned flags,
806 struct page **pagep, void **fsdata)
808 int ret;
810 ret = block_write_begin(mapping, pos, len, flags, pagep,
811 ext2_get_block);
812 if (ret < 0)
813 ext2_write_failed(mapping, pos + len);
814 return ret;
817 static int ext2_write_end(struct file *file, struct address_space *mapping,
818 loff_t pos, unsigned len, unsigned copied,
819 struct page *page, void *fsdata)
821 int ret;
823 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
824 if (ret < len)
825 ext2_write_failed(mapping, pos + len);
826 return ret;
829 static int
830 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
831 loff_t pos, unsigned len, unsigned flags,
832 struct page **pagep, void **fsdata)
834 int ret;
836 ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
837 ext2_get_block);
838 if (ret < 0)
839 ext2_write_failed(mapping, pos + len);
840 return ret;
843 static int ext2_nobh_writepage(struct page *page,
844 struct writeback_control *wbc)
846 return nobh_writepage(page, ext2_get_block, wbc);
849 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
851 return generic_block_bmap(mapping,block,ext2_get_block);
854 static ssize_t
855 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
857 struct file *file = iocb->ki_filp;
858 struct address_space *mapping = file->f_mapping;
859 struct inode *inode = mapping->host;
860 size_t count = iov_iter_count(iter);
861 ssize_t ret;
863 if (IS_DAX(inode))
864 ret = dax_do_io(iocb, inode, iter, offset, ext2_get_block, NULL,
865 DIO_LOCKING);
866 else
867 ret = blockdev_direct_IO(iocb, inode, iter, offset,
868 ext2_get_block);
869 if (ret < 0 && iov_iter_rw(iter) == WRITE)
870 ext2_write_failed(mapping, offset + count);
871 return ret;
874 static int
875 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
877 return mpage_writepages(mapping, wbc, ext2_get_block);
880 const struct address_space_operations ext2_aops = {
881 .readpage = ext2_readpage,
882 .readpages = ext2_readpages,
883 .writepage = ext2_writepage,
884 .write_begin = ext2_write_begin,
885 .write_end = ext2_write_end,
886 .bmap = ext2_bmap,
887 .direct_IO = ext2_direct_IO,
888 .writepages = ext2_writepages,
889 .migratepage = buffer_migrate_page,
890 .is_partially_uptodate = block_is_partially_uptodate,
891 .error_remove_page = generic_error_remove_page,
894 const struct address_space_operations ext2_nobh_aops = {
895 .readpage = ext2_readpage,
896 .readpages = ext2_readpages,
897 .writepage = ext2_nobh_writepage,
898 .write_begin = ext2_nobh_write_begin,
899 .write_end = nobh_write_end,
900 .bmap = ext2_bmap,
901 .direct_IO = ext2_direct_IO,
902 .writepages = ext2_writepages,
903 .migratepage = buffer_migrate_page,
904 .error_remove_page = generic_error_remove_page,
908 * Probably it should be a library function... search for first non-zero word
909 * or memcmp with zero_page, whatever is better for particular architecture.
910 * Linus?
912 static inline int all_zeroes(__le32 *p, __le32 *q)
914 while (p < q)
915 if (*p++)
916 return 0;
917 return 1;
921 * ext2_find_shared - find the indirect blocks for partial truncation.
922 * @inode: inode in question
923 * @depth: depth of the affected branch
924 * @offsets: offsets of pointers in that branch (see ext2_block_to_path)
925 * @chain: place to store the pointers to partial indirect blocks
926 * @top: place to the (detached) top of branch
928 * This is a helper function used by ext2_truncate().
930 * When we do truncate() we may have to clean the ends of several indirect
931 * blocks but leave the blocks themselves alive. Block is partially
932 * truncated if some data below the new i_size is referred from it (and
933 * it is on the path to the first completely truncated data block, indeed).
934 * We have to free the top of that path along with everything to the right
935 * of the path. Since no allocation past the truncation point is possible
936 * until ext2_truncate() finishes, we may safely do the latter, but top
937 * of branch may require special attention - pageout below the truncation
938 * point might try to populate it.
940 * We atomically detach the top of branch from the tree, store the block
941 * number of its root in *@top, pointers to buffer_heads of partially
942 * truncated blocks - in @chain[].bh and pointers to their last elements
943 * that should not be removed - in @chain[].p. Return value is the pointer
944 * to last filled element of @chain.
946 * The work left to caller to do the actual freeing of subtrees:
947 * a) free the subtree starting from *@top
948 * b) free the subtrees whose roots are stored in
949 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
950 * c) free the subtrees growing from the inode past the @chain[0].p
951 * (no partially truncated stuff there).
954 static Indirect *ext2_find_shared(struct inode *inode,
955 int depth,
956 int offsets[4],
957 Indirect chain[4],
958 __le32 *top)
960 Indirect *partial, *p;
961 int k, err;
963 *top = 0;
964 for (k = depth; k > 1 && !offsets[k-1]; k--)
966 partial = ext2_get_branch(inode, k, offsets, chain, &err);
967 if (!partial)
968 partial = chain + k-1;
970 * If the branch acquired continuation since we've looked at it -
971 * fine, it should all survive and (new) top doesn't belong to us.
973 write_lock(&EXT2_I(inode)->i_meta_lock);
974 if (!partial->key && *partial->p) {
975 write_unlock(&EXT2_I(inode)->i_meta_lock);
976 goto no_top;
978 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
981 * OK, we've found the last block that must survive. The rest of our
982 * branch should be detached before unlocking. However, if that rest
983 * of branch is all ours and does not grow immediately from the inode
984 * it's easier to cheat and just decrement partial->p.
986 if (p == chain + k - 1 && p > chain) {
987 p->p--;
988 } else {
989 *top = *p->p;
990 *p->p = 0;
992 write_unlock(&EXT2_I(inode)->i_meta_lock);
994 while(partial > p)
996 brelse(partial->bh);
997 partial--;
999 no_top:
1000 return partial;
1004 * ext2_free_data - free a list of data blocks
1005 * @inode: inode we are dealing with
1006 * @p: array of block numbers
1007 * @q: points immediately past the end of array
1009 * We are freeing all blocks referred from that array (numbers are
1010 * stored as little-endian 32-bit) and updating @inode->i_blocks
1011 * appropriately.
1013 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1015 unsigned long block_to_free = 0, count = 0;
1016 unsigned long nr;
1018 for ( ; p < q ; p++) {
1019 nr = le32_to_cpu(*p);
1020 if (nr) {
1021 *p = 0;
1022 /* accumulate blocks to free if they're contiguous */
1023 if (count == 0)
1024 goto free_this;
1025 else if (block_to_free == nr - count)
1026 count++;
1027 else {
1028 ext2_free_blocks (inode, block_to_free, count);
1029 mark_inode_dirty(inode);
1030 free_this:
1031 block_to_free = nr;
1032 count = 1;
1036 if (count > 0) {
1037 ext2_free_blocks (inode, block_to_free, count);
1038 mark_inode_dirty(inode);
1043 * ext2_free_branches - free an array of branches
1044 * @inode: inode we are dealing with
1045 * @p: array of block numbers
1046 * @q: pointer immediately past the end of array
1047 * @depth: depth of the branches to free
1049 * We are freeing all blocks referred from these branches (numbers are
1050 * stored as little-endian 32-bit) and updating @inode->i_blocks
1051 * appropriately.
1053 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1055 struct buffer_head * bh;
1056 unsigned long nr;
1058 if (depth--) {
1059 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1060 for ( ; p < q ; p++) {
1061 nr = le32_to_cpu(*p);
1062 if (!nr)
1063 continue;
1064 *p = 0;
1065 bh = sb_bread(inode->i_sb, nr);
1067 * A read failure? Report error and clear slot
1068 * (should be rare).
1070 if (!bh) {
1071 ext2_error(inode->i_sb, "ext2_free_branches",
1072 "Read failure, inode=%ld, block=%ld",
1073 inode->i_ino, nr);
1074 continue;
1076 ext2_free_branches(inode,
1077 (__le32*)bh->b_data,
1078 (__le32*)bh->b_data + addr_per_block,
1079 depth);
1080 bforget(bh);
1081 ext2_free_blocks(inode, nr, 1);
1082 mark_inode_dirty(inode);
1084 } else
1085 ext2_free_data(inode, p, q);
1088 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1090 __le32 *i_data = EXT2_I(inode)->i_data;
1091 struct ext2_inode_info *ei = EXT2_I(inode);
1092 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1093 int offsets[4];
1094 Indirect chain[4];
1095 Indirect *partial;
1096 __le32 nr = 0;
1097 int n;
1098 long iblock;
1099 unsigned blocksize;
1100 blocksize = inode->i_sb->s_blocksize;
1101 iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1103 n = ext2_block_to_path(inode, iblock, offsets, NULL);
1104 if (n == 0)
1105 return;
1108 * From here we block out all ext2_get_block() callers who want to
1109 * modify the block allocation tree.
1111 mutex_lock(&ei->truncate_mutex);
1113 if (n == 1) {
1114 ext2_free_data(inode, i_data+offsets[0],
1115 i_data + EXT2_NDIR_BLOCKS);
1116 goto do_indirects;
1119 partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1120 /* Kill the top of shared branch (already detached) */
1121 if (nr) {
1122 if (partial == chain)
1123 mark_inode_dirty(inode);
1124 else
1125 mark_buffer_dirty_inode(partial->bh, inode);
1126 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1128 /* Clear the ends of indirect blocks on the shared branch */
1129 while (partial > chain) {
1130 ext2_free_branches(inode,
1131 partial->p + 1,
1132 (__le32*)partial->bh->b_data+addr_per_block,
1133 (chain+n-1) - partial);
1134 mark_buffer_dirty_inode(partial->bh, inode);
1135 brelse (partial->bh);
1136 partial--;
1138 do_indirects:
1139 /* Kill the remaining (whole) subtrees */
1140 switch (offsets[0]) {
1141 default:
1142 nr = i_data[EXT2_IND_BLOCK];
1143 if (nr) {
1144 i_data[EXT2_IND_BLOCK] = 0;
1145 mark_inode_dirty(inode);
1146 ext2_free_branches(inode, &nr, &nr+1, 1);
1148 case EXT2_IND_BLOCK:
1149 nr = i_data[EXT2_DIND_BLOCK];
1150 if (nr) {
1151 i_data[EXT2_DIND_BLOCK] = 0;
1152 mark_inode_dirty(inode);
1153 ext2_free_branches(inode, &nr, &nr+1, 2);
1155 case EXT2_DIND_BLOCK:
1156 nr = i_data[EXT2_TIND_BLOCK];
1157 if (nr) {
1158 i_data[EXT2_TIND_BLOCK] = 0;
1159 mark_inode_dirty(inode);
1160 ext2_free_branches(inode, &nr, &nr+1, 3);
1162 case EXT2_TIND_BLOCK:
1166 ext2_discard_reservation(inode);
1168 mutex_unlock(&ei->truncate_mutex);
1171 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1174 * XXX: it seems like a bug here that we don't allow
1175 * IS_APPEND inode to have blocks-past-i_size trimmed off.
1176 * review and fix this.
1178 * Also would be nice to be able to handle IO errors and such,
1179 * but that's probably too much to ask.
1181 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 S_ISLNK(inode->i_mode)))
1183 return;
1184 if (ext2_inode_is_fast_symlink(inode))
1185 return;
1186 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1187 return;
1188 __ext2_truncate_blocks(inode, offset);
1191 static int ext2_setsize(struct inode *inode, loff_t newsize)
1193 int error;
1195 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1196 S_ISLNK(inode->i_mode)))
1197 return -EINVAL;
1198 if (ext2_inode_is_fast_symlink(inode))
1199 return -EINVAL;
1200 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1201 return -EPERM;
1203 inode_dio_wait(inode);
1205 if (IS_DAX(inode))
1206 error = dax_truncate_page(inode, newsize, ext2_get_block);
1207 else if (test_opt(inode->i_sb, NOBH))
1208 error = nobh_truncate_page(inode->i_mapping,
1209 newsize, ext2_get_block);
1210 else
1211 error = block_truncate_page(inode->i_mapping,
1212 newsize, ext2_get_block);
1213 if (error)
1214 return error;
1216 truncate_setsize(inode, newsize);
1217 __ext2_truncate_blocks(inode, newsize);
1219 inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1220 if (inode_needs_sync(inode)) {
1221 sync_mapping_buffers(inode->i_mapping);
1222 sync_inode_metadata(inode, 1);
1223 } else {
1224 mark_inode_dirty(inode);
1227 return 0;
1230 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1231 struct buffer_head **p)
1233 struct buffer_head * bh;
1234 unsigned long block_group;
1235 unsigned long block;
1236 unsigned long offset;
1237 struct ext2_group_desc * gdp;
1239 *p = NULL;
1240 if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1241 ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1242 goto Einval;
1244 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1245 gdp = ext2_get_group_desc(sb, block_group, NULL);
1246 if (!gdp)
1247 goto Egdp;
1249 * Figure out the offset within the block group inode table
1251 offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1252 block = le32_to_cpu(gdp->bg_inode_table) +
1253 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1254 if (!(bh = sb_bread(sb, block)))
1255 goto Eio;
1257 *p = bh;
1258 offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1259 return (struct ext2_inode *) (bh->b_data + offset);
1261 Einval:
1262 ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1263 (unsigned long) ino);
1264 return ERR_PTR(-EINVAL);
1265 Eio:
1266 ext2_error(sb, "ext2_get_inode",
1267 "unable to read inode block - inode=%lu, block=%lu",
1268 (unsigned long) ino, block);
1269 Egdp:
1270 return ERR_PTR(-EIO);
1273 void ext2_set_inode_flags(struct inode *inode)
1275 unsigned int flags = EXT2_I(inode)->i_flags;
1277 inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1278 S_DIRSYNC | S_DAX);
1279 if (flags & EXT2_SYNC_FL)
1280 inode->i_flags |= S_SYNC;
1281 if (flags & EXT2_APPEND_FL)
1282 inode->i_flags |= S_APPEND;
1283 if (flags & EXT2_IMMUTABLE_FL)
1284 inode->i_flags |= S_IMMUTABLE;
1285 if (flags & EXT2_NOATIME_FL)
1286 inode->i_flags |= S_NOATIME;
1287 if (flags & EXT2_DIRSYNC_FL)
1288 inode->i_flags |= S_DIRSYNC;
1289 if (test_opt(inode->i_sb, DAX))
1290 inode->i_flags |= S_DAX;
1293 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1294 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1296 unsigned int flags = ei->vfs_inode.i_flags;
1298 ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1299 EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1300 if (flags & S_SYNC)
1301 ei->i_flags |= EXT2_SYNC_FL;
1302 if (flags & S_APPEND)
1303 ei->i_flags |= EXT2_APPEND_FL;
1304 if (flags & S_IMMUTABLE)
1305 ei->i_flags |= EXT2_IMMUTABLE_FL;
1306 if (flags & S_NOATIME)
1307 ei->i_flags |= EXT2_NOATIME_FL;
1308 if (flags & S_DIRSYNC)
1309 ei->i_flags |= EXT2_DIRSYNC_FL;
1312 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1314 struct ext2_inode_info *ei;
1315 struct buffer_head * bh;
1316 struct ext2_inode *raw_inode;
1317 struct inode *inode;
1318 long ret = -EIO;
1319 int n;
1320 uid_t i_uid;
1321 gid_t i_gid;
1323 inode = iget_locked(sb, ino);
1324 if (!inode)
1325 return ERR_PTR(-ENOMEM);
1326 if (!(inode->i_state & I_NEW))
1327 return inode;
1329 ei = EXT2_I(inode);
1330 ei->i_block_alloc_info = NULL;
1332 raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1333 if (IS_ERR(raw_inode)) {
1334 ret = PTR_ERR(raw_inode);
1335 goto bad_inode;
1338 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1339 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1340 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1341 if (!(test_opt (inode->i_sb, NO_UID32))) {
1342 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1343 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1345 i_uid_write(inode, i_uid);
1346 i_gid_write(inode, i_gid);
1347 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1348 inode->i_size = le32_to_cpu(raw_inode->i_size);
1349 inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1350 inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1351 inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1352 inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1353 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1354 /* We now have enough fields to check if the inode was active or not.
1355 * This is needed because nfsd might try to access dead inodes
1356 * the test is that same one that e2fsck uses
1357 * NeilBrown 1999oct15
1359 if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1360 /* this inode is deleted */
1361 brelse (bh);
1362 ret = -ESTALE;
1363 goto bad_inode;
1365 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1366 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1367 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1368 ei->i_frag_no = raw_inode->i_frag;
1369 ei->i_frag_size = raw_inode->i_fsize;
1370 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1371 ei->i_dir_acl = 0;
1372 if (S_ISREG(inode->i_mode))
1373 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1374 else
1375 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1376 ei->i_dtime = 0;
1377 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1378 ei->i_state = 0;
1379 ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1380 ei->i_dir_start_lookup = 0;
1383 * NOTE! The in-memory inode i_data array is in little-endian order
1384 * even on big-endian machines: we do NOT byteswap the block numbers!
1386 for (n = 0; n < EXT2_N_BLOCKS; n++)
1387 ei->i_data[n] = raw_inode->i_block[n];
1389 if (S_ISREG(inode->i_mode)) {
1390 inode->i_op = &ext2_file_inode_operations;
1391 if (test_opt(inode->i_sb, NOBH)) {
1392 inode->i_mapping->a_ops = &ext2_nobh_aops;
1393 inode->i_fop = &ext2_file_operations;
1394 } else {
1395 inode->i_mapping->a_ops = &ext2_aops;
1396 inode->i_fop = &ext2_file_operations;
1398 } else if (S_ISDIR(inode->i_mode)) {
1399 inode->i_op = &ext2_dir_inode_operations;
1400 inode->i_fop = &ext2_dir_operations;
1401 if (test_opt(inode->i_sb, NOBH))
1402 inode->i_mapping->a_ops = &ext2_nobh_aops;
1403 else
1404 inode->i_mapping->a_ops = &ext2_aops;
1405 } else if (S_ISLNK(inode->i_mode)) {
1406 if (ext2_inode_is_fast_symlink(inode)) {
1407 inode->i_link = (char *)ei->i_data;
1408 inode->i_op = &ext2_fast_symlink_inode_operations;
1409 nd_terminate_link(ei->i_data, inode->i_size,
1410 sizeof(ei->i_data) - 1);
1411 } else {
1412 inode->i_op = &ext2_symlink_inode_operations;
1413 if (test_opt(inode->i_sb, NOBH))
1414 inode->i_mapping->a_ops = &ext2_nobh_aops;
1415 else
1416 inode->i_mapping->a_ops = &ext2_aops;
1418 } else {
1419 inode->i_op = &ext2_special_inode_operations;
1420 if (raw_inode->i_block[0])
1421 init_special_inode(inode, inode->i_mode,
1422 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1423 else
1424 init_special_inode(inode, inode->i_mode,
1425 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1427 brelse (bh);
1428 ext2_set_inode_flags(inode);
1429 unlock_new_inode(inode);
1430 return inode;
1432 bad_inode:
1433 iget_failed(inode);
1434 return ERR_PTR(ret);
1437 static int __ext2_write_inode(struct inode *inode, int do_sync)
1439 struct ext2_inode_info *ei = EXT2_I(inode);
1440 struct super_block *sb = inode->i_sb;
1441 ino_t ino = inode->i_ino;
1442 uid_t uid = i_uid_read(inode);
1443 gid_t gid = i_gid_read(inode);
1444 struct buffer_head * bh;
1445 struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1446 int n;
1447 int err = 0;
1449 if (IS_ERR(raw_inode))
1450 return -EIO;
1452 /* For fields not not tracking in the in-memory inode,
1453 * initialise them to zero for new inodes. */
1454 if (ei->i_state & EXT2_STATE_NEW)
1455 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1457 ext2_get_inode_flags(ei);
1458 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1459 if (!(test_opt(sb, NO_UID32))) {
1460 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1461 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1463 * Fix up interoperability with old kernels. Otherwise, old inodes get
1464 * re-used with the upper 16 bits of the uid/gid intact
1466 if (!ei->i_dtime) {
1467 raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1468 raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1469 } else {
1470 raw_inode->i_uid_high = 0;
1471 raw_inode->i_gid_high = 0;
1473 } else {
1474 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1475 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1476 raw_inode->i_uid_high = 0;
1477 raw_inode->i_gid_high = 0;
1479 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1480 raw_inode->i_size = cpu_to_le32(inode->i_size);
1481 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1482 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1483 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1485 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1486 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1487 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1488 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1489 raw_inode->i_frag = ei->i_frag_no;
1490 raw_inode->i_fsize = ei->i_frag_size;
1491 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1492 if (!S_ISREG(inode->i_mode))
1493 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1494 else {
1495 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1496 if (inode->i_size > 0x7fffffffULL) {
1497 if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1498 EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1499 EXT2_SB(sb)->s_es->s_rev_level ==
1500 cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1501 /* If this is the first large file
1502 * created, add a flag to the superblock.
1504 spin_lock(&EXT2_SB(sb)->s_lock);
1505 ext2_update_dynamic_rev(sb);
1506 EXT2_SET_RO_COMPAT_FEATURE(sb,
1507 EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1508 spin_unlock(&EXT2_SB(sb)->s_lock);
1509 ext2_write_super(sb);
1514 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1515 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1516 if (old_valid_dev(inode->i_rdev)) {
1517 raw_inode->i_block[0] =
1518 cpu_to_le32(old_encode_dev(inode->i_rdev));
1519 raw_inode->i_block[1] = 0;
1520 } else {
1521 raw_inode->i_block[0] = 0;
1522 raw_inode->i_block[1] =
1523 cpu_to_le32(new_encode_dev(inode->i_rdev));
1524 raw_inode->i_block[2] = 0;
1526 } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1527 raw_inode->i_block[n] = ei->i_data[n];
1528 mark_buffer_dirty(bh);
1529 if (do_sync) {
1530 sync_dirty_buffer(bh);
1531 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1532 printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1533 sb->s_id, (unsigned long) ino);
1534 err = -EIO;
1537 ei->i_state &= ~EXT2_STATE_NEW;
1538 brelse (bh);
1539 return err;
1542 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1544 return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1547 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1549 struct inode *inode = d_inode(dentry);
1550 int error;
1552 error = inode_change_ok(inode, iattr);
1553 if (error)
1554 return error;
1556 if (is_quota_modification(inode, iattr)) {
1557 error = dquot_initialize(inode);
1558 if (error)
1559 return error;
1561 if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1562 (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1563 error = dquot_transfer(inode, iattr);
1564 if (error)
1565 return error;
1567 if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1568 error = ext2_setsize(inode, iattr->ia_size);
1569 if (error)
1570 return error;
1572 setattr_copy(inode, iattr);
1573 if (iattr->ia_valid & ATTR_MODE)
1574 error = posix_acl_chmod(inode, inode->i_mode);
1575 mark_inode_dirty(inode);
1577 return error;