ext3: Fix format string issues
[linux-2.6.git] / fs / ext2 / inode.c
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
38 static int __ext2_write_inode(struct inode *inode, int do_sync);
39
40 /*
41  * Test whether an inode is a fast symlink.
42  */
43 static 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
52 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
53
54 static 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  */
67 void 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         end_writeback(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
105 typedef struct {
106         __le32  *p;
107         __le32  key;
108         struct buffer_head *bh;
109 } Indirect;
110
111 static 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
117 static 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
154 static 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  */
225 static 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
254 changed:
255         read_unlock(&EXT2_I(inode)->i_meta_lock);
256         brelse(bh);
257         *err = -EAGAIN;
258         goto no_block;
259 failure:
260         *err = -EIO;
261 no_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
285 static 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
321 static 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  */
352 static int
353 ext2_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  */
389 static 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;
434 failed_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
467 static 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  */
537 static 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  */
600 static 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);
729 got_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 */
736 cleanup:
737         while (partial > chain) {
738                 brelse(partial->bh);
739                 partial--;
740         }
741         return err;
742 }
743
744 int 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
757 int 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
764 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
765 {
766         return block_write_full_page(page, ext2_get_block, wbc);
767 }
768
769 static int ext2_readpage(struct file *file, struct page *page)
770 {
771         return mpage_readpage(page, ext2_get_block);
772 }
773
774 static int
775 ext2_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
781 static int
782 ext2_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
795 static 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
807 static int
808 ext2_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
821 static int ext2_nobh_writepage(struct page *page,
822                         struct writeback_control *wbc)
823 {
824         return nobh_writepage(page, ext2_get_block, wbc);
825 }
826
827 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
828 {
829         return generic_block_bmap(mapping,block,ext2_get_block);
830 }
831
832 static ssize_t
833 ext2_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
848 static int
849 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
850 {
851         return mpage_writepages(mapping, wbc, ext2_get_block);
852 }
853
854 const 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
868 const struct address_space_operations ext2_aops_xip = {
869         .bmap                   = ext2_bmap,
870         .get_xip_mem            = ext2_get_xip_mem,
871 };
872
873 const 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  */
891 static 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
933 static 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         }
978 no_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  */
992 static 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  */
1032 static 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
1067 static 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         }
1117 do_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
1150 static 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
1170 static 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
1209 static 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
1240 Einval:
1241         ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1242                    (unsigned long) ino);
1243         return ERR_PTR(-EINVAL);
1244 Eio:
1245         ext2_error(sb, "ext2_get_inode",
1246                    "unable to read inode block - inode=%lu, block=%lu",
1247                    (unsigned long) ino, block);
1248 Egdp:
1249         return ERR_PTR(-EIO);
1250 }
1251
1252 void 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 */
1270 void 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
1288 struct 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
1297         inode = iget_locked(sb, ino);
1298         if (!inode)
1299                 return ERR_PTR(-ENOMEM);
1300         if (!(inode->i_state & I_NEW))
1301                 return inode;
1302
1303         ei = EXT2_I(inode);
1304         ei->i_block_alloc_info = NULL;
1305
1306         raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1307         if (IS_ERR(raw_inode)) {
1308                 ret = PTR_ERR(raw_inode);
1309                 goto bad_inode;
1310         }
1311
1312         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1313         inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1314         inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1315         if (!(test_opt (inode->i_sb, NO_UID32))) {
1316                 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1317                 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1318         }
1319         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1320         inode->i_size = le32_to_cpu(raw_inode->i_size);
1321         inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1322         inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1323         inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1324         inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1325         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1326         /* We now have enough fields to check if the inode was active or not.
1327          * This is needed because nfsd might try to access dead inodes
1328          * the test is that same one that e2fsck uses
1329          * NeilBrown 1999oct15
1330          */
1331         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1332                 /* this inode is deleted */
1333                 brelse (bh);
1334                 ret = -ESTALE;
1335                 goto bad_inode;
1336         }
1337         inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1338         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1339         ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1340         ei->i_frag_no = raw_inode->i_frag;
1341         ei->i_frag_size = raw_inode->i_fsize;
1342         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1343         ei->i_dir_acl = 0;
1344         if (S_ISREG(inode->i_mode))
1345                 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1346         else
1347                 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1348         ei->i_dtime = 0;
1349         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1350         ei->i_state = 0;
1351         ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1352         ei->i_dir_start_lookup = 0;
1353
1354         /*
1355          * NOTE! The in-memory inode i_data array is in little-endian order
1356          * even on big-endian machines: we do NOT byteswap the block numbers!
1357          */
1358         for (n = 0; n < EXT2_N_BLOCKS; n++)
1359                 ei->i_data[n] = raw_inode->i_block[n];
1360
1361         if (S_ISREG(inode->i_mode)) {
1362                 inode->i_op = &ext2_file_inode_operations;
1363                 if (ext2_use_xip(inode->i_sb)) {
1364                         inode->i_mapping->a_ops = &ext2_aops_xip;
1365                         inode->i_fop = &ext2_xip_file_operations;
1366                 } else if (test_opt(inode->i_sb, NOBH)) {
1367                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1368                         inode->i_fop = &ext2_file_operations;
1369                 } else {
1370                         inode->i_mapping->a_ops = &ext2_aops;
1371                         inode->i_fop = &ext2_file_operations;
1372                 }
1373         } else if (S_ISDIR(inode->i_mode)) {
1374                 inode->i_op = &ext2_dir_inode_operations;
1375                 inode->i_fop = &ext2_dir_operations;
1376                 if (test_opt(inode->i_sb, NOBH))
1377                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1378                 else
1379                         inode->i_mapping->a_ops = &ext2_aops;
1380         } else if (S_ISLNK(inode->i_mode)) {
1381                 if (ext2_inode_is_fast_symlink(inode)) {
1382                         inode->i_op = &ext2_fast_symlink_inode_operations;
1383                         nd_terminate_link(ei->i_data, inode->i_size,
1384                                 sizeof(ei->i_data) - 1);
1385                 } else {
1386                         inode->i_op = &ext2_symlink_inode_operations;
1387                         if (test_opt(inode->i_sb, NOBH))
1388                                 inode->i_mapping->a_ops = &ext2_nobh_aops;
1389                         else
1390                                 inode->i_mapping->a_ops = &ext2_aops;
1391                 }
1392         } else {
1393                 inode->i_op = &ext2_special_inode_operations;
1394                 if (raw_inode->i_block[0])
1395                         init_special_inode(inode, inode->i_mode,
1396                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1397                 else 
1398                         init_special_inode(inode, inode->i_mode,
1399                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1400         }
1401         brelse (bh);
1402         ext2_set_inode_flags(inode);
1403         unlock_new_inode(inode);
1404         return inode;
1405         
1406 bad_inode:
1407         iget_failed(inode);
1408         return ERR_PTR(ret);
1409 }
1410
1411 static int __ext2_write_inode(struct inode *inode, int do_sync)
1412 {
1413         struct ext2_inode_info *ei = EXT2_I(inode);
1414         struct super_block *sb = inode->i_sb;
1415         ino_t ino = inode->i_ino;
1416         uid_t uid = inode->i_uid;
1417         gid_t gid = inode->i_gid;
1418         struct buffer_head * bh;
1419         struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1420         int n;
1421         int err = 0;
1422
1423         if (IS_ERR(raw_inode))
1424                 return -EIO;
1425
1426         /* For fields not not tracking in the in-memory inode,
1427          * initialise them to zero for new inodes. */
1428         if (ei->i_state & EXT2_STATE_NEW)
1429                 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1430
1431         ext2_get_inode_flags(ei);
1432         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1433         if (!(test_opt(sb, NO_UID32))) {
1434                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1435                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1436 /*
1437  * Fix up interoperability with old kernels. Otherwise, old inodes get
1438  * re-used with the upper 16 bits of the uid/gid intact
1439  */
1440                 if (!ei->i_dtime) {
1441                         raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1442                         raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1443                 } else {
1444                         raw_inode->i_uid_high = 0;
1445                         raw_inode->i_gid_high = 0;
1446                 }
1447         } else {
1448                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1449                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1450                 raw_inode->i_uid_high = 0;
1451                 raw_inode->i_gid_high = 0;
1452         }
1453         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1454         raw_inode->i_size = cpu_to_le32(inode->i_size);
1455         raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1456         raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1457         raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1458
1459         raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1460         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1461         raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1462         raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1463         raw_inode->i_frag = ei->i_frag_no;
1464         raw_inode->i_fsize = ei->i_frag_size;
1465         raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1466         if (!S_ISREG(inode->i_mode))
1467                 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1468         else {
1469                 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1470                 if (inode->i_size > 0x7fffffffULL) {
1471                         if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1472                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1473                             EXT2_SB(sb)->s_es->s_rev_level ==
1474                                         cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1475                                /* If this is the first large file
1476                                 * created, add a flag to the superblock.
1477                                 */
1478                                 spin_lock(&EXT2_SB(sb)->s_lock);
1479                                 ext2_update_dynamic_rev(sb);
1480                                 EXT2_SET_RO_COMPAT_FEATURE(sb,
1481                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1482                                 spin_unlock(&EXT2_SB(sb)->s_lock);
1483                                 ext2_write_super(sb);
1484                         }
1485                 }
1486         }
1487         
1488         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1489         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1490                 if (old_valid_dev(inode->i_rdev)) {
1491                         raw_inode->i_block[0] =
1492                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1493                         raw_inode->i_block[1] = 0;
1494                 } else {
1495                         raw_inode->i_block[0] = 0;
1496                         raw_inode->i_block[1] =
1497                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1498                         raw_inode->i_block[2] = 0;
1499                 }
1500         } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1501                 raw_inode->i_block[n] = ei->i_data[n];
1502         mark_buffer_dirty(bh);
1503         if (do_sync) {
1504                 sync_dirty_buffer(bh);
1505                 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1506                         printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1507                                 sb->s_id, (unsigned long) ino);
1508                         err = -EIO;
1509                 }
1510         }
1511         ei->i_state &= ~EXT2_STATE_NEW;
1512         brelse (bh);
1513         return err;
1514 }
1515
1516 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1517 {
1518         return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1519 }
1520
1521 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1522 {
1523         struct inode *inode = dentry->d_inode;
1524         int error;
1525
1526         error = inode_change_ok(inode, iattr);
1527         if (error)
1528                 return error;
1529
1530         if (is_quota_modification(inode, iattr))
1531                 dquot_initialize(inode);
1532         if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1533             (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1534                 error = dquot_transfer(inode, iattr);
1535                 if (error)
1536                         return error;
1537         }
1538         if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1539                 error = ext2_setsize(inode, iattr->ia_size);
1540                 if (error)
1541                         return error;
1542         }
1543         setattr_copy(inode, iattr);
1544         if (iattr->ia_valid & ATTR_MODE)
1545                 error = ext2_acl_chmod(inode);
1546         mark_inode_dirty(inode);
1547
1548         return error;
1549 }