vfs: pagecache usage optimization for pagesize!=blocksize
[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/smp_lock.h>
26 #include <linux/time.h>
27 #include <linux/highuid.h>
28 #include <linux/pagemap.h>
29 #include <linux/quotaops.h>
30 #include <linux/module.h>
31 #include <linux/writeback.h>
32 #include <linux/buffer_head.h>
33 #include <linux/mpage.h>
34 #include "ext2.h"
35 #include "acl.h"
36 #include "xip.h"
37
38 MODULE_AUTHOR("Remy Card and others");
39 MODULE_DESCRIPTION("Second Extended Filesystem");
40 MODULE_LICENSE("GPL");
41
42 static int ext2_update_inode(struct inode * inode, int do_sync);
43
44 /*
45  * Test whether an inode is a fast symlink.
46  */
47 static inline int ext2_inode_is_fast_symlink(struct inode *inode)
48 {
49         int ea_blocks = EXT2_I(inode)->i_file_acl ?
50                 (inode->i_sb->s_blocksize >> 9) : 0;
51
52         return (S_ISLNK(inode->i_mode) &&
53                 inode->i_blocks - ea_blocks == 0);
54 }
55
56 /*
57  * Called at the last iput() if i_nlink is zero.
58  */
59 void ext2_delete_inode (struct inode * inode)
60 {
61         truncate_inode_pages(&inode->i_data, 0);
62
63         if (is_bad_inode(inode))
64                 goto no_delete;
65         EXT2_I(inode)->i_dtime  = get_seconds();
66         mark_inode_dirty(inode);
67         ext2_update_inode(inode, inode_needs_sync(inode));
68
69         inode->i_size = 0;
70         if (inode->i_blocks)
71                 ext2_truncate (inode);
72         ext2_free_inode (inode);
73
74         return;
75 no_delete:
76         clear_inode(inode);     /* We must guarantee clearing of inode... */
77 }
78
79 typedef struct {
80         __le32  *p;
81         __le32  key;
82         struct buffer_head *bh;
83 } Indirect;
84
85 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
86 {
87         p->key = *(p->p = v);
88         p->bh = bh;
89 }
90
91 static inline int verify_chain(Indirect *from, Indirect *to)
92 {
93         while (from <= to && from->key == *from->p)
94                 from++;
95         return (from > to);
96 }
97
98 /**
99  *      ext2_block_to_path - parse the block number into array of offsets
100  *      @inode: inode in question (we are only interested in its superblock)
101  *      @i_block: block number to be parsed
102  *      @offsets: array to store the offsets in
103  *      @boundary: set this non-zero if the referred-to block is likely to be
104  *             followed (on disk) by an indirect block.
105  *      To store the locations of file's data ext2 uses a data structure common
106  *      for UNIX filesystems - tree of pointers anchored in the inode, with
107  *      data blocks at leaves and indirect blocks in intermediate nodes.
108  *      This function translates the block number into path in that tree -
109  *      return value is the path length and @offsets[n] is the offset of
110  *      pointer to (n+1)th node in the nth one. If @block is out of range
111  *      (negative or too large) warning is printed and zero returned.
112  *
113  *      Note: function doesn't find node addresses, so no IO is needed. All
114  *      we need to know is the capacity of indirect blocks (taken from the
115  *      inode->i_sb).
116  */
117
118 /*
119  * Portability note: the last comparison (check that we fit into triple
120  * indirect block) is spelled differently, because otherwise on an
121  * architecture with 32-bit longs and 8Kb pages we might get into trouble
122  * if our filesystem had 8Kb blocks. We might use long long, but that would
123  * kill us on x86. Oh, well, at least the sign propagation does not matter -
124  * i_block would have to be negative in the very beginning, so we would not
125  * get there at all.
126  */
127
128 static int ext2_block_to_path(struct inode *inode,
129                         long i_block, int offsets[4], int *boundary)
130 {
131         int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
132         int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
133         const long direct_blocks = EXT2_NDIR_BLOCKS,
134                 indirect_blocks = ptrs,
135                 double_blocks = (1 << (ptrs_bits * 2));
136         int n = 0;
137         int final = 0;
138
139         if (i_block < 0) {
140                 ext2_warning (inode->i_sb, "ext2_block_to_path", "block < 0");
141         } else if (i_block < direct_blocks) {
142                 offsets[n++] = i_block;
143                 final = direct_blocks;
144         } else if ( (i_block -= direct_blocks) < indirect_blocks) {
145                 offsets[n++] = EXT2_IND_BLOCK;
146                 offsets[n++] = i_block;
147                 final = ptrs;
148         } else if ((i_block -= indirect_blocks) < double_blocks) {
149                 offsets[n++] = EXT2_DIND_BLOCK;
150                 offsets[n++] = i_block >> ptrs_bits;
151                 offsets[n++] = i_block & (ptrs - 1);
152                 final = ptrs;
153         } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
154                 offsets[n++] = EXT2_TIND_BLOCK;
155                 offsets[n++] = i_block >> (ptrs_bits * 2);
156                 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
157                 offsets[n++] = i_block & (ptrs - 1);
158                 final = ptrs;
159         } else {
160                 ext2_warning (inode->i_sb, "ext2_block_to_path", "block > big");
161         }
162         if (boundary)
163                 *boundary = final - 1 - (i_block & (ptrs - 1));
164
165         return n;
166 }
167
168 /**
169  *      ext2_get_branch - read the chain of indirect blocks leading to data
170  *      @inode: inode in question
171  *      @depth: depth of the chain (1 - direct pointer, etc.)
172  *      @offsets: offsets of pointers in inode/indirect blocks
173  *      @chain: place to store the result
174  *      @err: here we store the error value
175  *
176  *      Function fills the array of triples <key, p, bh> and returns %NULL
177  *      if everything went OK or the pointer to the last filled triple
178  *      (incomplete one) otherwise. Upon the return chain[i].key contains
179  *      the number of (i+1)-th block in the chain (as it is stored in memory,
180  *      i.e. little-endian 32-bit), chain[i].p contains the address of that
181  *      number (it points into struct inode for i==0 and into the bh->b_data
182  *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
183  *      block for i>0 and NULL for i==0. In other words, it holds the block
184  *      numbers of the chain, addresses they were taken from (and where we can
185  *      verify that chain did not change) and buffer_heads hosting these
186  *      numbers.
187  *
188  *      Function stops when it stumbles upon zero pointer (absent block)
189  *              (pointer to last triple returned, *@err == 0)
190  *      or when it gets an IO error reading an indirect block
191  *              (ditto, *@err == -EIO)
192  *      or when it notices that chain had been changed while it was reading
193  *              (ditto, *@err == -EAGAIN)
194  *      or when it reads all @depth-1 indirect blocks successfully and finds
195  *      the whole chain, all way to the data (returns %NULL, *err == 0).
196  */
197 static Indirect *ext2_get_branch(struct inode *inode,
198                                  int depth,
199                                  int *offsets,
200                                  Indirect chain[4],
201                                  int *err)
202 {
203         struct super_block *sb = inode->i_sb;
204         Indirect *p = chain;
205         struct buffer_head *bh;
206
207         *err = 0;
208         /* i_data is not going away, no lock needed */
209         add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
210         if (!p->key)
211                 goto no_block;
212         while (--depth) {
213                 bh = sb_bread(sb, le32_to_cpu(p->key));
214                 if (!bh)
215                         goto failure;
216                 read_lock(&EXT2_I(inode)->i_meta_lock);
217                 if (!verify_chain(chain, p))
218                         goto changed;
219                 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
220                 read_unlock(&EXT2_I(inode)->i_meta_lock);
221                 if (!p->key)
222                         goto no_block;
223         }
224         return NULL;
225
226 changed:
227         read_unlock(&EXT2_I(inode)->i_meta_lock);
228         brelse(bh);
229         *err = -EAGAIN;
230         goto no_block;
231 failure:
232         *err = -EIO;
233 no_block:
234         return p;
235 }
236
237 /**
238  *      ext2_find_near - find a place for allocation with sufficient locality
239  *      @inode: owner
240  *      @ind: descriptor of indirect block.
241  *
242  *      This function returns the preferred place for block allocation.
243  *      It is used when heuristic for sequential allocation fails.
244  *      Rules are:
245  *        + if there is a block to the left of our position - allocate near it.
246  *        + if pointer will live in indirect block - allocate near that block.
247  *        + if pointer will live in inode - allocate in the same cylinder group.
248  *
249  * In the latter case we colour the starting block by the callers PID to
250  * prevent it from clashing with concurrent allocations for a different inode
251  * in the same block group.   The PID is used here so that functionally related
252  * files will be close-by on-disk.
253  *
254  *      Caller must make sure that @ind is valid and will stay that way.
255  */
256
257 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
258 {
259         struct ext2_inode_info *ei = EXT2_I(inode);
260         __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
261         __le32 *p;
262         ext2_fsblk_t bg_start;
263         ext2_fsblk_t colour;
264
265         /* Try to find previous block */
266         for (p = ind->p - 1; p >= start; p--)
267                 if (*p)
268                         return le32_to_cpu(*p);
269
270         /* No such thing, so let's try location of indirect block */
271         if (ind->bh)
272                 return ind->bh->b_blocknr;
273
274         /*
275          * It is going to be refered from inode itself? OK, just put it into
276          * the same cylinder group then.
277          */
278         bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
279         colour = (current->pid % 16) *
280                         (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
281         return bg_start + colour;
282 }
283
284 /**
285  *      ext2_find_goal - find a preferred place for allocation.
286  *      @inode: owner
287  *      @block:  block we want
288  *      @partial: pointer to the last triple within a chain
289  *
290  *      Returns preferred place for a block (the goal).
291  */
292
293 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
294                                           Indirect *partial)
295 {
296         struct ext2_block_alloc_info *block_i;
297
298         block_i = EXT2_I(inode)->i_block_alloc_info;
299
300         /*
301          * try the heuristic for sequential allocation,
302          * failing that at least try to get decent locality.
303          */
304         if (block_i && (block == block_i->last_alloc_logical_block + 1)
305                 && (block_i->last_alloc_physical_block != 0)) {
306                 return block_i->last_alloc_physical_block + 1;
307         }
308
309         return ext2_find_near(inode, partial);
310 }
311
312 /**
313  *      ext2_blks_to_allocate: Look up the block map and count the number
314  *      of direct blocks need to be allocated for the given branch.
315  *
316  *      @branch: chain of indirect blocks
317  *      @k: number of blocks need for indirect blocks
318  *      @blks: number of data blocks to be mapped.
319  *      @blocks_to_boundary:  the offset in the indirect block
320  *
321  *      return the total number of blocks to be allocate, including the
322  *      direct and indirect blocks.
323  */
324 static int
325 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
326                 int blocks_to_boundary)
327 {
328         unsigned long count = 0;
329
330         /*
331          * Simple case, [t,d]Indirect block(s) has not allocated yet
332          * then it's clear blocks on that path have not allocated
333          */
334         if (k > 0) {
335                 /* right now don't hanel cross boundary allocation */
336                 if (blks < blocks_to_boundary + 1)
337                         count += blks;
338                 else
339                         count += blocks_to_boundary + 1;
340                 return count;
341         }
342
343         count++;
344         while (count < blks && count <= blocks_to_boundary
345                 && le32_to_cpu(*(branch[0].p + count)) == 0) {
346                 count++;
347         }
348         return count;
349 }
350
351 /**
352  *      ext2_alloc_blocks: multiple allocate blocks needed for a branch
353  *      @indirect_blks: the number of blocks need to allocate for indirect
354  *                      blocks
355  *
356  *      @new_blocks: on return it will store the new block numbers for
357  *      the indirect blocks(if needed) and the first direct block,
358  *      @blks:  on return it will store the total number of allocated
359  *              direct blocks
360  */
361 static int ext2_alloc_blocks(struct inode *inode,
362                         ext2_fsblk_t goal, int indirect_blks, int blks,
363                         ext2_fsblk_t new_blocks[4], int *err)
364 {
365         int target, i;
366         unsigned long count = 0;
367         int index = 0;
368         ext2_fsblk_t current_block = 0;
369         int ret = 0;
370
371         /*
372          * Here we try to allocate the requested multiple blocks at once,
373          * on a best-effort basis.
374          * To build a branch, we should allocate blocks for
375          * the indirect blocks(if not allocated yet), and at least
376          * the first direct block of this branch.  That's the
377          * minimum number of blocks need to allocate(required)
378          */
379         target = blks + indirect_blks;
380
381         while (1) {
382                 count = target;
383                 /* allocating blocks for indirect blocks and direct blocks */
384                 current_block = ext2_new_blocks(inode,goal,&count,err);
385                 if (*err)
386                         goto failed_out;
387
388                 target -= count;
389                 /* allocate blocks for indirect blocks */
390                 while (index < indirect_blks && count) {
391                         new_blocks[index++] = current_block++;
392                         count--;
393                 }
394
395                 if (count > 0)
396                         break;
397         }
398
399         /* save the new block number for the first direct block */
400         new_blocks[index] = current_block;
401
402         /* total number of blocks allocated for direct blocks */
403         ret = count;
404         *err = 0;
405         return ret;
406 failed_out:
407         for (i = 0; i <index; i++)
408                 ext2_free_blocks(inode, new_blocks[i], 1);
409         return ret;
410 }
411
412 /**
413  *      ext2_alloc_branch - allocate and set up a chain of blocks.
414  *      @inode: owner
415  *      @num: depth of the chain (number of blocks to allocate)
416  *      @offsets: offsets (in the blocks) to store the pointers to next.
417  *      @branch: place to store the chain in.
418  *
419  *      This function allocates @num blocks, zeroes out all but the last one,
420  *      links them into chain and (if we are synchronous) writes them to disk.
421  *      In other words, it prepares a branch that can be spliced onto the
422  *      inode. It stores the information about that chain in the branch[], in
423  *      the same format as ext2_get_branch() would do. We are calling it after
424  *      we had read the existing part of chain and partial points to the last
425  *      triple of that (one with zero ->key). Upon the exit we have the same
426  *      picture as after the successful ext2_get_block(), excpet that in one
427  *      place chain is disconnected - *branch->p is still zero (we did not
428  *      set the last link), but branch->key contains the number that should
429  *      be placed into *branch->p to fill that gap.
430  *
431  *      If allocation fails we free all blocks we've allocated (and forget
432  *      their buffer_heads) and return the error value the from failed
433  *      ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
434  *      as described above and return 0.
435  */
436
437 static int ext2_alloc_branch(struct inode *inode,
438                         int indirect_blks, int *blks, ext2_fsblk_t goal,
439                         int *offsets, Indirect *branch)
440 {
441         int blocksize = inode->i_sb->s_blocksize;
442         int i, n = 0;
443         int err = 0;
444         struct buffer_head *bh;
445         int num;
446         ext2_fsblk_t new_blocks[4];
447         ext2_fsblk_t current_block;
448
449         num = ext2_alloc_blocks(inode, goal, indirect_blks,
450                                 *blks, new_blocks, &err);
451         if (err)
452                 return err;
453
454         branch[0].key = cpu_to_le32(new_blocks[0]);
455         /*
456          * metadata blocks and data blocks are allocated.
457          */
458         for (n = 1; n <= indirect_blks;  n++) {
459                 /*
460                  * Get buffer_head for parent block, zero it out
461                  * and set the pointer to new one, then send
462                  * parent to disk.
463                  */
464                 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
465                 branch[n].bh = bh;
466                 lock_buffer(bh);
467                 memset(bh->b_data, 0, blocksize);
468                 branch[n].p = (__le32 *) bh->b_data + offsets[n];
469                 branch[n].key = cpu_to_le32(new_blocks[n]);
470                 *branch[n].p = branch[n].key;
471                 if ( n == indirect_blks) {
472                         current_block = new_blocks[n];
473                         /*
474                          * End of chain, update the last new metablock of
475                          * the chain to point to the new allocated
476                          * data blocks numbers
477                          */
478                         for (i=1; i < num; i++)
479                                 *(branch[n].p + i) = cpu_to_le32(++current_block);
480                 }
481                 set_buffer_uptodate(bh);
482                 unlock_buffer(bh);
483                 mark_buffer_dirty_inode(bh, inode);
484                 /* We used to sync bh here if IS_SYNC(inode).
485                  * But we now rely upon generic_osync_inode()
486                  * and b_inode_buffers.  But not for directories.
487                  */
488                 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
489                         sync_dirty_buffer(bh);
490         }
491         *blks = num;
492         return err;
493 }
494
495 /**
496  * ext2_splice_branch - splice the allocated branch onto inode.
497  * @inode: owner
498  * @block: (logical) number of block we are adding
499  * @chain: chain of indirect blocks (with a missing link - see
500  *      ext2_alloc_branch)
501  * @where: location of missing link
502  * @num:   number of indirect blocks we are adding
503  * @blks:  number of direct blocks we are adding
504  *
505  * This function fills the missing link and does all housekeeping needed in
506  * inode (->i_blocks, etc.). In case of success we end up with the full
507  * chain to new block and return 0.
508  */
509 static void ext2_splice_branch(struct inode *inode,
510                         long block, Indirect *where, int num, int blks)
511 {
512         int i;
513         struct ext2_block_alloc_info *block_i;
514         ext2_fsblk_t current_block;
515
516         block_i = EXT2_I(inode)->i_block_alloc_info;
517
518         /* XXX LOCKING probably should have i_meta_lock ?*/
519         /* That's it */
520
521         *where->p = where->key;
522
523         /*
524          * Update the host buffer_head or inode to point to more just allocated
525          * direct blocks blocks
526          */
527         if (num == 0 && blks > 1) {
528                 current_block = le32_to_cpu(where->key) + 1;
529                 for (i = 1; i < blks; i++)
530                         *(where->p + i ) = cpu_to_le32(current_block++);
531         }
532
533         /*
534          * update the most recently allocated logical & physical block
535          * in i_block_alloc_info, to assist find the proper goal block for next
536          * allocation
537          */
538         if (block_i) {
539                 block_i->last_alloc_logical_block = block + blks - 1;
540                 block_i->last_alloc_physical_block =
541                                 le32_to_cpu(where[num].key) + blks - 1;
542         }
543
544         /* We are done with atomic stuff, now do the rest of housekeeping */
545
546         /* had we spliced it onto indirect block? */
547         if (where->bh)
548                 mark_buffer_dirty_inode(where->bh, inode);
549
550         inode->i_ctime = CURRENT_TIME_SEC;
551         mark_inode_dirty(inode);
552 }
553
554 /*
555  * Allocation strategy is simple: if we have to allocate something, we will
556  * have to go the whole way to leaf. So let's do it before attaching anything
557  * to tree, set linkage between the newborn blocks, write them if sync is
558  * required, recheck the path, free and repeat if check fails, otherwise
559  * set the last missing link (that will protect us from any truncate-generated
560  * removals - all blocks on the path are immune now) and possibly force the
561  * write on the parent block.
562  * That has a nice additional property: no special recovery from the failed
563  * allocations is needed - we simply release blocks and do not touch anything
564  * reachable from inode.
565  *
566  * `handle' can be NULL if create == 0.
567  *
568  * return > 0, # of blocks mapped or allocated.
569  * return = 0, if plain lookup failed.
570  * return < 0, error case.
571  */
572 static int ext2_get_blocks(struct inode *inode,
573                            sector_t iblock, unsigned long maxblocks,
574                            struct buffer_head *bh_result,
575                            int create)
576 {
577         int err = -EIO;
578         int offsets[4];
579         Indirect chain[4];
580         Indirect *partial;
581         ext2_fsblk_t goal;
582         int indirect_blks;
583         int blocks_to_boundary = 0;
584         int depth;
585         struct ext2_inode_info *ei = EXT2_I(inode);
586         int count = 0;
587         ext2_fsblk_t first_block = 0;
588
589         depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
590
591         if (depth == 0)
592                 return (err);
593 reread:
594         partial = ext2_get_branch(inode, depth, offsets, chain, &err);
595
596         /* Simplest case - block found, no allocation needed */
597         if (!partial) {
598                 first_block = le32_to_cpu(chain[depth - 1].key);
599                 clear_buffer_new(bh_result); /* What's this do? */
600                 count++;
601                 /*map more blocks*/
602                 while (count < maxblocks && count <= blocks_to_boundary) {
603                         ext2_fsblk_t blk;
604
605                         if (!verify_chain(chain, partial)) {
606                                 /*
607                                  * Indirect block might be removed by
608                                  * truncate while we were reading it.
609                                  * Handling of that case: forget what we've
610                                  * got now, go to reread.
611                                  */
612                                 count = 0;
613                                 goto changed;
614                         }
615                         blk = le32_to_cpu(*(chain[depth-1].p + count));
616                         if (blk == first_block + count)
617                                 count++;
618                         else
619                                 break;
620                 }
621                 goto got_it;
622         }
623
624         /* Next simple case - plain lookup or failed read of indirect block */
625         if (!create || err == -EIO)
626                 goto cleanup;
627
628         mutex_lock(&ei->truncate_mutex);
629
630         /*
631          * Okay, we need to do block allocation.  Lazily initialize the block
632          * allocation info here if necessary
633         */
634         if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
635                 ext2_init_block_alloc_info(inode);
636
637         goal = ext2_find_goal(inode, iblock, partial);
638
639         /* the number of blocks need to allocate for [d,t]indirect blocks */
640         indirect_blks = (chain + depth) - partial - 1;
641         /*
642          * Next look up the indirect map to count the totoal number of
643          * direct blocks to allocate for this branch.
644          */
645         count = ext2_blks_to_allocate(partial, indirect_blks,
646                                         maxblocks, blocks_to_boundary);
647         /*
648          * XXX ???? Block out ext2_truncate while we alter the tree
649          */
650         err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
651                                 offsets + (partial - chain), partial);
652
653         if (err) {
654                 mutex_unlock(&ei->truncate_mutex);
655                 goto cleanup;
656         }
657
658         if (ext2_use_xip(inode->i_sb)) {
659                 /*
660                  * we need to clear the block
661                  */
662                 err = ext2_clear_xip_target (inode,
663                         le32_to_cpu(chain[depth-1].key));
664                 if (err) {
665                         mutex_unlock(&ei->truncate_mutex);
666                         goto cleanup;
667                 }
668         }
669
670         ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
671         mutex_unlock(&ei->truncate_mutex);
672         set_buffer_new(bh_result);
673 got_it:
674         map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
675         if (count > blocks_to_boundary)
676                 set_buffer_boundary(bh_result);
677         err = count;
678         /* Clean up and exit */
679         partial = chain + depth - 1;    /* the whole chain */
680 cleanup:
681         while (partial > chain) {
682                 brelse(partial->bh);
683                 partial--;
684         }
685         return err;
686 changed:
687         while (partial > chain) {
688                 brelse(partial->bh);
689                 partial--;
690         }
691         goto reread;
692 }
693
694 int ext2_get_block(struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create)
695 {
696         unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
697         int ret = ext2_get_blocks(inode, iblock, max_blocks,
698                               bh_result, create);
699         if (ret > 0) {
700                 bh_result->b_size = (ret << inode->i_blkbits);
701                 ret = 0;
702         }
703         return ret;
704
705 }
706
707 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
708 {
709         return block_write_full_page(page, ext2_get_block, wbc);
710 }
711
712 static int ext2_readpage(struct file *file, struct page *page)
713 {
714         return mpage_readpage(page, ext2_get_block);
715 }
716
717 static int
718 ext2_readpages(struct file *file, struct address_space *mapping,
719                 struct list_head *pages, unsigned nr_pages)
720 {
721         return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
722 }
723
724 int __ext2_write_begin(struct file *file, struct address_space *mapping,
725                 loff_t pos, unsigned len, unsigned flags,
726                 struct page **pagep, void **fsdata)
727 {
728         return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
729                                                         ext2_get_block);
730 }
731
732 static int
733 ext2_write_begin(struct file *file, struct address_space *mapping,
734                 loff_t pos, unsigned len, unsigned flags,
735                 struct page **pagep, void **fsdata)
736 {
737         *pagep = NULL;
738         return __ext2_write_begin(file, mapping, pos, len, flags, pagep,fsdata);
739 }
740
741 static int
742 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
743                 loff_t pos, unsigned len, unsigned flags,
744                 struct page **pagep, void **fsdata)
745 {
746         /*
747          * Dir-in-pagecache still uses ext2_write_begin. Would have to rework
748          * directory handling code to pass around offsets rather than struct
749          * pages in order to make this work easily.
750          */
751         return nobh_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
752                                                         ext2_get_block);
753 }
754
755 static int ext2_nobh_writepage(struct page *page,
756                         struct writeback_control *wbc)
757 {
758         return nobh_writepage(page, ext2_get_block, wbc);
759 }
760
761 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
762 {
763         return generic_block_bmap(mapping,block,ext2_get_block);
764 }
765
766 static ssize_t
767 ext2_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
768                         loff_t offset, unsigned long nr_segs)
769 {
770         struct file *file = iocb->ki_filp;
771         struct inode *inode = file->f_mapping->host;
772
773         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
774                                 offset, nr_segs, ext2_get_block, NULL);
775 }
776
777 static int
778 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
779 {
780         return mpage_writepages(mapping, wbc, ext2_get_block);
781 }
782
783 const struct address_space_operations ext2_aops = {
784         .readpage               = ext2_readpage,
785         .readpages              = ext2_readpages,
786         .writepage              = ext2_writepage,
787         .sync_page              = block_sync_page,
788         .write_begin            = ext2_write_begin,
789         .write_end              = generic_write_end,
790         .bmap                   = ext2_bmap,
791         .direct_IO              = ext2_direct_IO,
792         .writepages             = ext2_writepages,
793         .migratepage            = buffer_migrate_page,
794         .is_partially_uptodate  = block_is_partially_uptodate,
795 };
796
797 const struct address_space_operations ext2_aops_xip = {
798         .bmap                   = ext2_bmap,
799         .get_xip_mem            = ext2_get_xip_mem,
800 };
801
802 const struct address_space_operations ext2_nobh_aops = {
803         .readpage               = ext2_readpage,
804         .readpages              = ext2_readpages,
805         .writepage              = ext2_nobh_writepage,
806         .sync_page              = block_sync_page,
807         .write_begin            = ext2_nobh_write_begin,
808         .write_end              = nobh_write_end,
809         .bmap                   = ext2_bmap,
810         .direct_IO              = ext2_direct_IO,
811         .writepages             = ext2_writepages,
812         .migratepage            = buffer_migrate_page,
813 };
814
815 /*
816  * Probably it should be a library function... search for first non-zero word
817  * or memcmp with zero_page, whatever is better for particular architecture.
818  * Linus?
819  */
820 static inline int all_zeroes(__le32 *p, __le32 *q)
821 {
822         while (p < q)
823                 if (*p++)
824                         return 0;
825         return 1;
826 }
827
828 /**
829  *      ext2_find_shared - find the indirect blocks for partial truncation.
830  *      @inode:   inode in question
831  *      @depth:   depth of the affected branch
832  *      @offsets: offsets of pointers in that branch (see ext2_block_to_path)
833  *      @chain:   place to store the pointers to partial indirect blocks
834  *      @top:     place to the (detached) top of branch
835  *
836  *      This is a helper function used by ext2_truncate().
837  *
838  *      When we do truncate() we may have to clean the ends of several indirect
839  *      blocks but leave the blocks themselves alive. Block is partially
840  *      truncated if some data below the new i_size is refered from it (and
841  *      it is on the path to the first completely truncated data block, indeed).
842  *      We have to free the top of that path along with everything to the right
843  *      of the path. Since no allocation past the truncation point is possible
844  *      until ext2_truncate() finishes, we may safely do the latter, but top
845  *      of branch may require special attention - pageout below the truncation
846  *      point might try to populate it.
847  *
848  *      We atomically detach the top of branch from the tree, store the block
849  *      number of its root in *@top, pointers to buffer_heads of partially
850  *      truncated blocks - in @chain[].bh and pointers to their last elements
851  *      that should not be removed - in @chain[].p. Return value is the pointer
852  *      to last filled element of @chain.
853  *
854  *      The work left to caller to do the actual freeing of subtrees:
855  *              a) free the subtree starting from *@top
856  *              b) free the subtrees whose roots are stored in
857  *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
858  *              c) free the subtrees growing from the inode past the @chain[0].p
859  *                      (no partially truncated stuff there).
860  */
861
862 static Indirect *ext2_find_shared(struct inode *inode,
863                                 int depth,
864                                 int offsets[4],
865                                 Indirect chain[4],
866                                 __le32 *top)
867 {
868         Indirect *partial, *p;
869         int k, err;
870
871         *top = 0;
872         for (k = depth; k > 1 && !offsets[k-1]; k--)
873                 ;
874         partial = ext2_get_branch(inode, k, offsets, chain, &err);
875         if (!partial)
876                 partial = chain + k-1;
877         /*
878          * If the branch acquired continuation since we've looked at it -
879          * fine, it should all survive and (new) top doesn't belong to us.
880          */
881         write_lock(&EXT2_I(inode)->i_meta_lock);
882         if (!partial->key && *partial->p) {
883                 write_unlock(&EXT2_I(inode)->i_meta_lock);
884                 goto no_top;
885         }
886         for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
887                 ;
888         /*
889          * OK, we've found the last block that must survive. The rest of our
890          * branch should be detached before unlocking. However, if that rest
891          * of branch is all ours and does not grow immediately from the inode
892          * it's easier to cheat and just decrement partial->p.
893          */
894         if (p == chain + k - 1 && p > chain) {
895                 p->p--;
896         } else {
897                 *top = *p->p;
898                 *p->p = 0;
899         }
900         write_unlock(&EXT2_I(inode)->i_meta_lock);
901
902         while(partial > p)
903         {
904                 brelse(partial->bh);
905                 partial--;
906         }
907 no_top:
908         return partial;
909 }
910
911 /**
912  *      ext2_free_data - free a list of data blocks
913  *      @inode: inode we are dealing with
914  *      @p:     array of block numbers
915  *      @q:     points immediately past the end of array
916  *
917  *      We are freeing all blocks refered from that array (numbers are
918  *      stored as little-endian 32-bit) and updating @inode->i_blocks
919  *      appropriately.
920  */
921 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
922 {
923         unsigned long block_to_free = 0, count = 0;
924         unsigned long nr;
925
926         for ( ; p < q ; p++) {
927                 nr = le32_to_cpu(*p);
928                 if (nr) {
929                         *p = 0;
930                         /* accumulate blocks to free if they're contiguous */
931                         if (count == 0)
932                                 goto free_this;
933                         else if (block_to_free == nr - count)
934                                 count++;
935                         else {
936                                 mark_inode_dirty(inode);
937                                 ext2_free_blocks (inode, block_to_free, count);
938                         free_this:
939                                 block_to_free = nr;
940                                 count = 1;
941                         }
942                 }
943         }
944         if (count > 0) {
945                 mark_inode_dirty(inode);
946                 ext2_free_blocks (inode, block_to_free, count);
947         }
948 }
949
950 /**
951  *      ext2_free_branches - free an array of branches
952  *      @inode: inode we are dealing with
953  *      @p:     array of block numbers
954  *      @q:     pointer immediately past the end of array
955  *      @depth: depth of the branches to free
956  *
957  *      We are freeing all blocks refered from these branches (numbers are
958  *      stored as little-endian 32-bit) and updating @inode->i_blocks
959  *      appropriately.
960  */
961 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
962 {
963         struct buffer_head * bh;
964         unsigned long nr;
965
966         if (depth--) {
967                 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
968                 for ( ; p < q ; p++) {
969                         nr = le32_to_cpu(*p);
970                         if (!nr)
971                                 continue;
972                         *p = 0;
973                         bh = sb_bread(inode->i_sb, nr);
974                         /*
975                          * A read failure? Report error and clear slot
976                          * (should be rare).
977                          */ 
978                         if (!bh) {
979                                 ext2_error(inode->i_sb, "ext2_free_branches",
980                                         "Read failure, inode=%ld, block=%ld",
981                                         inode->i_ino, nr);
982                                 continue;
983                         }
984                         ext2_free_branches(inode,
985                                            (__le32*)bh->b_data,
986                                            (__le32*)bh->b_data + addr_per_block,
987                                            depth);
988                         bforget(bh);
989                         ext2_free_blocks(inode, nr, 1);
990                         mark_inode_dirty(inode);
991                 }
992         } else
993                 ext2_free_data(inode, p, q);
994 }
995
996 void ext2_truncate(struct inode *inode)
997 {
998         __le32 *i_data = EXT2_I(inode)->i_data;
999         struct ext2_inode_info *ei = EXT2_I(inode);
1000         int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1001         int offsets[4];
1002         Indirect chain[4];
1003         Indirect *partial;
1004         __le32 nr = 0;
1005         int n;
1006         long iblock;
1007         unsigned blocksize;
1008
1009         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1010             S_ISLNK(inode->i_mode)))
1011                 return;
1012         if (ext2_inode_is_fast_symlink(inode))
1013                 return;
1014         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1015                 return;
1016
1017         blocksize = inode->i_sb->s_blocksize;
1018         iblock = (inode->i_size + blocksize-1)
1019                                         >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1020
1021         if (mapping_is_xip(inode->i_mapping))
1022                 xip_truncate_page(inode->i_mapping, inode->i_size);
1023         else if (test_opt(inode->i_sb, NOBH))
1024                 nobh_truncate_page(inode->i_mapping,
1025                                 inode->i_size, ext2_get_block);
1026         else
1027                 block_truncate_page(inode->i_mapping,
1028                                 inode->i_size, ext2_get_block);
1029
1030         n = ext2_block_to_path(inode, iblock, offsets, NULL);
1031         if (n == 0)
1032                 return;
1033
1034         /*
1035          * From here we block out all ext2_get_block() callers who want to
1036          * modify the block allocation tree.
1037          */
1038         mutex_lock(&ei->truncate_mutex);
1039
1040         if (n == 1) {
1041                 ext2_free_data(inode, i_data+offsets[0],
1042                                         i_data + EXT2_NDIR_BLOCKS);
1043                 goto do_indirects;
1044         }
1045
1046         partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1047         /* Kill the top of shared branch (already detached) */
1048         if (nr) {
1049                 if (partial == chain)
1050                         mark_inode_dirty(inode);
1051                 else
1052                         mark_buffer_dirty_inode(partial->bh, inode);
1053                 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1054         }
1055         /* Clear the ends of indirect blocks on the shared branch */
1056         while (partial > chain) {
1057                 ext2_free_branches(inode,
1058                                    partial->p + 1,
1059                                    (__le32*)partial->bh->b_data+addr_per_block,
1060                                    (chain+n-1) - partial);
1061                 mark_buffer_dirty_inode(partial->bh, inode);
1062                 brelse (partial->bh);
1063                 partial--;
1064         }
1065 do_indirects:
1066         /* Kill the remaining (whole) subtrees */
1067         switch (offsets[0]) {
1068                 default:
1069                         nr = i_data[EXT2_IND_BLOCK];
1070                         if (nr) {
1071                                 i_data[EXT2_IND_BLOCK] = 0;
1072                                 mark_inode_dirty(inode);
1073                                 ext2_free_branches(inode, &nr, &nr+1, 1);
1074                         }
1075                 case EXT2_IND_BLOCK:
1076                         nr = i_data[EXT2_DIND_BLOCK];
1077                         if (nr) {
1078                                 i_data[EXT2_DIND_BLOCK] = 0;
1079                                 mark_inode_dirty(inode);
1080                                 ext2_free_branches(inode, &nr, &nr+1, 2);
1081                         }
1082                 case EXT2_DIND_BLOCK:
1083                         nr = i_data[EXT2_TIND_BLOCK];
1084                         if (nr) {
1085                                 i_data[EXT2_TIND_BLOCK] = 0;
1086                                 mark_inode_dirty(inode);
1087                                 ext2_free_branches(inode, &nr, &nr+1, 3);
1088                         }
1089                 case EXT2_TIND_BLOCK:
1090                         ;
1091         }
1092
1093         ext2_discard_reservation(inode);
1094
1095         mutex_unlock(&ei->truncate_mutex);
1096         inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
1097         if (inode_needs_sync(inode)) {
1098                 sync_mapping_buffers(inode->i_mapping);
1099                 ext2_sync_inode (inode);
1100         } else {
1101                 mark_inode_dirty(inode);
1102         }
1103 }
1104
1105 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1106                                         struct buffer_head **p)
1107 {
1108         struct buffer_head * bh;
1109         unsigned long block_group;
1110         unsigned long block;
1111         unsigned long offset;
1112         struct ext2_group_desc * gdp;
1113
1114         *p = NULL;
1115         if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1116             ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1117                 goto Einval;
1118
1119         block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1120         gdp = ext2_get_group_desc(sb, block_group, NULL);
1121         if (!gdp)
1122                 goto Egdp;
1123         /*
1124          * Figure out the offset within the block group inode table
1125          */
1126         offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1127         block = le32_to_cpu(gdp->bg_inode_table) +
1128                 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1129         if (!(bh = sb_bread(sb, block)))
1130                 goto Eio;
1131
1132         *p = bh;
1133         offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1134         return (struct ext2_inode *) (bh->b_data + offset);
1135
1136 Einval:
1137         ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1138                    (unsigned long) ino);
1139         return ERR_PTR(-EINVAL);
1140 Eio:
1141         ext2_error(sb, "ext2_get_inode",
1142                    "unable to read inode block - inode=%lu, block=%lu",
1143                    (unsigned long) ino, block);
1144 Egdp:
1145         return ERR_PTR(-EIO);
1146 }
1147
1148 void ext2_set_inode_flags(struct inode *inode)
1149 {
1150         unsigned int flags = EXT2_I(inode)->i_flags;
1151
1152         inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
1153         if (flags & EXT2_SYNC_FL)
1154                 inode->i_flags |= S_SYNC;
1155         if (flags & EXT2_APPEND_FL)
1156                 inode->i_flags |= S_APPEND;
1157         if (flags & EXT2_IMMUTABLE_FL)
1158                 inode->i_flags |= S_IMMUTABLE;
1159         if (flags & EXT2_NOATIME_FL)
1160                 inode->i_flags |= S_NOATIME;
1161         if (flags & EXT2_DIRSYNC_FL)
1162                 inode->i_flags |= S_DIRSYNC;
1163 }
1164
1165 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1166 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1167 {
1168         unsigned int flags = ei->vfs_inode.i_flags;
1169
1170         ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1171                         EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1172         if (flags & S_SYNC)
1173                 ei->i_flags |= EXT2_SYNC_FL;
1174         if (flags & S_APPEND)
1175                 ei->i_flags |= EXT2_APPEND_FL;
1176         if (flags & S_IMMUTABLE)
1177                 ei->i_flags |= EXT2_IMMUTABLE_FL;
1178         if (flags & S_NOATIME)
1179                 ei->i_flags |= EXT2_NOATIME_FL;
1180         if (flags & S_DIRSYNC)
1181                 ei->i_flags |= EXT2_DIRSYNC_FL;
1182 }
1183
1184 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1185 {
1186         struct ext2_inode_info *ei;
1187         struct buffer_head * bh;
1188         struct ext2_inode *raw_inode;
1189         struct inode *inode;
1190         long ret = -EIO;
1191         int n;
1192
1193         inode = iget_locked(sb, ino);
1194         if (!inode)
1195                 return ERR_PTR(-ENOMEM);
1196         if (!(inode->i_state & I_NEW))
1197                 return inode;
1198
1199         ei = EXT2_I(inode);
1200 #ifdef CONFIG_EXT2_FS_POSIX_ACL
1201         ei->i_acl = EXT2_ACL_NOT_CACHED;
1202         ei->i_default_acl = EXT2_ACL_NOT_CACHED;
1203 #endif
1204         ei->i_block_alloc_info = NULL;
1205
1206         raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1207         if (IS_ERR(raw_inode)) {
1208                 ret = PTR_ERR(raw_inode);
1209                 goto bad_inode;
1210         }
1211
1212         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1213         inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1214         inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1215         if (!(test_opt (inode->i_sb, NO_UID32))) {
1216                 inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1217                 inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1218         }
1219         inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
1220         inode->i_size = le32_to_cpu(raw_inode->i_size);
1221         inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1222         inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1223         inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1224         inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1225         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1226         /* We now have enough fields to check if the inode was active or not.
1227          * This is needed because nfsd might try to access dead inodes
1228          * the test is that same one that e2fsck uses
1229          * NeilBrown 1999oct15
1230          */
1231         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1232                 /* this inode is deleted */
1233                 brelse (bh);
1234                 ret = -ESTALE;
1235                 goto bad_inode;
1236         }
1237         inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1238         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1239         ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1240         ei->i_frag_no = raw_inode->i_frag;
1241         ei->i_frag_size = raw_inode->i_fsize;
1242         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1243         ei->i_dir_acl = 0;
1244         if (S_ISREG(inode->i_mode))
1245                 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1246         else
1247                 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1248         ei->i_dtime = 0;
1249         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1250         ei->i_state = 0;
1251         ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1252         ei->i_dir_start_lookup = 0;
1253
1254         /*
1255          * NOTE! The in-memory inode i_data array is in little-endian order
1256          * even on big-endian machines: we do NOT byteswap the block numbers!
1257          */
1258         for (n = 0; n < EXT2_N_BLOCKS; n++)
1259                 ei->i_data[n] = raw_inode->i_block[n];
1260
1261         if (S_ISREG(inode->i_mode)) {
1262                 inode->i_op = &ext2_file_inode_operations;
1263                 if (ext2_use_xip(inode->i_sb)) {
1264                         inode->i_mapping->a_ops = &ext2_aops_xip;
1265                         inode->i_fop = &ext2_xip_file_operations;
1266                 } else if (test_opt(inode->i_sb, NOBH)) {
1267                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1268                         inode->i_fop = &ext2_file_operations;
1269                 } else {
1270                         inode->i_mapping->a_ops = &ext2_aops;
1271                         inode->i_fop = &ext2_file_operations;
1272                 }
1273         } else if (S_ISDIR(inode->i_mode)) {
1274                 inode->i_op = &ext2_dir_inode_operations;
1275                 inode->i_fop = &ext2_dir_operations;
1276                 if (test_opt(inode->i_sb, NOBH))
1277                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1278                 else
1279                         inode->i_mapping->a_ops = &ext2_aops;
1280         } else if (S_ISLNK(inode->i_mode)) {
1281                 if (ext2_inode_is_fast_symlink(inode))
1282                         inode->i_op = &ext2_fast_symlink_inode_operations;
1283                 else {
1284                         inode->i_op = &ext2_symlink_inode_operations;
1285                         if (test_opt(inode->i_sb, NOBH))
1286                                 inode->i_mapping->a_ops = &ext2_nobh_aops;
1287                         else
1288                                 inode->i_mapping->a_ops = &ext2_aops;
1289                 }
1290         } else {
1291                 inode->i_op = &ext2_special_inode_operations;
1292                 if (raw_inode->i_block[0])
1293                         init_special_inode(inode, inode->i_mode,
1294                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1295                 else 
1296                         init_special_inode(inode, inode->i_mode,
1297                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1298         }
1299         brelse (bh);
1300         ext2_set_inode_flags(inode);
1301         unlock_new_inode(inode);
1302         return inode;
1303         
1304 bad_inode:
1305         iget_failed(inode);
1306         return ERR_PTR(ret);
1307 }
1308
1309 static int ext2_update_inode(struct inode * inode, int do_sync)
1310 {
1311         struct ext2_inode_info *ei = EXT2_I(inode);
1312         struct super_block *sb = inode->i_sb;
1313         ino_t ino = inode->i_ino;
1314         uid_t uid = inode->i_uid;
1315         gid_t gid = inode->i_gid;
1316         struct buffer_head * bh;
1317         struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1318         int n;
1319         int err = 0;
1320
1321         if (IS_ERR(raw_inode))
1322                 return -EIO;
1323
1324         /* For fields not not tracking in the in-memory inode,
1325          * initialise them to zero for new inodes. */
1326         if (ei->i_state & EXT2_STATE_NEW)
1327                 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1328
1329         ext2_get_inode_flags(ei);
1330         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1331         if (!(test_opt(sb, NO_UID32))) {
1332                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1333                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1334 /*
1335  * Fix up interoperability with old kernels. Otherwise, old inodes get
1336  * re-used with the upper 16 bits of the uid/gid intact
1337  */
1338                 if (!ei->i_dtime) {
1339                         raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1340                         raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1341                 } else {
1342                         raw_inode->i_uid_high = 0;
1343                         raw_inode->i_gid_high = 0;
1344                 }
1345         } else {
1346                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1347                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1348                 raw_inode->i_uid_high = 0;
1349                 raw_inode->i_gid_high = 0;
1350         }
1351         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1352         raw_inode->i_size = cpu_to_le32(inode->i_size);
1353         raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1354         raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1355         raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1356
1357         raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1358         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1359         raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1360         raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1361         raw_inode->i_frag = ei->i_frag_no;
1362         raw_inode->i_fsize = ei->i_frag_size;
1363         raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1364         if (!S_ISREG(inode->i_mode))
1365                 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1366         else {
1367                 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1368                 if (inode->i_size > 0x7fffffffULL) {
1369                         if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1370                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1371                             EXT2_SB(sb)->s_es->s_rev_level ==
1372                                         cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1373                                /* If this is the first large file
1374                                 * created, add a flag to the superblock.
1375                                 */
1376                                 lock_kernel();
1377                                 ext2_update_dynamic_rev(sb);
1378                                 EXT2_SET_RO_COMPAT_FEATURE(sb,
1379                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1380                                 unlock_kernel();
1381                                 ext2_write_super(sb);
1382                         }
1383                 }
1384         }
1385         
1386         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1387         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1388                 if (old_valid_dev(inode->i_rdev)) {
1389                         raw_inode->i_block[0] =
1390                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1391                         raw_inode->i_block[1] = 0;
1392                 } else {
1393                         raw_inode->i_block[0] = 0;
1394                         raw_inode->i_block[1] =
1395                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1396                         raw_inode->i_block[2] = 0;
1397                 }
1398         } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1399                 raw_inode->i_block[n] = ei->i_data[n];
1400         mark_buffer_dirty(bh);
1401         if (do_sync) {
1402                 sync_dirty_buffer(bh);
1403                 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1404                         printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1405                                 sb->s_id, (unsigned long) ino);
1406                         err = -EIO;
1407                 }
1408         }
1409         ei->i_state &= ~EXT2_STATE_NEW;
1410         brelse (bh);
1411         return err;
1412 }
1413
1414 int ext2_write_inode(struct inode *inode, int wait)
1415 {
1416         return ext2_update_inode(inode, wait);
1417 }
1418
1419 int ext2_sync_inode(struct inode *inode)
1420 {
1421         struct writeback_control wbc = {
1422                 .sync_mode = WB_SYNC_ALL,
1423                 .nr_to_write = 0,       /* sys_fsync did this */
1424         };
1425         return sync_inode(inode, &wbc);
1426 }
1427
1428 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1429 {
1430         struct inode *inode = dentry->d_inode;
1431         int error;
1432
1433         error = inode_change_ok(inode, iattr);
1434         if (error)
1435                 return error;
1436         if ((iattr->ia_valid & ATTR_UID && iattr->ia_uid != inode->i_uid) ||
1437             (iattr->ia_valid & ATTR_GID && iattr->ia_gid != inode->i_gid)) {
1438                 error = DQUOT_TRANSFER(inode, iattr) ? -EDQUOT : 0;
1439                 if (error)
1440                         return error;
1441         }
1442         error = inode_setattr(inode, iattr);
1443         if (!error && (iattr->ia_valid & ATTR_MODE))
1444                 error = ext2_acl_chmod(inode);
1445         return error;
1446 }