Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs
[sfrench/cifs-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/dax.h>
29 #include <linux/blkdev.h>
30 #include <linux/quotaops.h>
31 #include <linux/writeback.h>
32 #include <linux/buffer_head.h>
33 #include <linux/mpage.h>
34 #include <linux/fiemap.h>
35 #include <linux/iomap.h>
36 #include <linux/namei.h>
37 #include <linux/uio.h>
38 #include "ext2.h"
39 #include "acl.h"
40 #include "xattr.h"
41
42 static int __ext2_write_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 static void ext2_truncate_blocks(struct inode *inode, loff_t offset);
57
58 static void ext2_write_failed(struct address_space *mapping, loff_t to)
59 {
60         struct inode *inode = mapping->host;
61
62         if (to > inode->i_size) {
63                 truncate_pagecache(inode, inode->i_size);
64                 ext2_truncate_blocks(inode, inode->i_size);
65         }
66 }
67
68 /*
69  * Called at the last iput() if i_nlink is zero.
70  */
71 void ext2_evict_inode(struct inode * inode)
72 {
73         struct ext2_block_alloc_info *rsv;
74         int want_delete = 0;
75
76         if (!inode->i_nlink && !is_bad_inode(inode)) {
77                 want_delete = 1;
78                 dquot_initialize(inode);
79         } else {
80                 dquot_drop(inode);
81         }
82
83         truncate_inode_pages_final(&inode->i_data);
84
85         if (want_delete) {
86                 sb_start_intwrite(inode->i_sb);
87                 /* set dtime */
88                 EXT2_I(inode)->i_dtime  = get_seconds();
89                 mark_inode_dirty(inode);
90                 __ext2_write_inode(inode, inode_needs_sync(inode));
91                 /* truncate to 0 */
92                 inode->i_size = 0;
93                 if (inode->i_blocks)
94                         ext2_truncate_blocks(inode, 0);
95                 ext2_xattr_delete_inode(inode);
96         }
97
98         invalidate_inode_buffers(inode);
99         clear_inode(inode);
100
101         ext2_discard_reservation(inode);
102         rsv = EXT2_I(inode)->i_block_alloc_info;
103         EXT2_I(inode)->i_block_alloc_info = NULL;
104         if (unlikely(rsv))
105                 kfree(rsv);
106
107         if (want_delete) {
108                 ext2_free_inode(inode);
109                 sb_end_intwrite(inode->i_sb);
110         }
111 }
112
113 typedef struct {
114         __le32  *p;
115         __le32  key;
116         struct buffer_head *bh;
117 } Indirect;
118
119 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
120 {
121         p->key = *(p->p = v);
122         p->bh = bh;
123 }
124
125 static inline int verify_chain(Indirect *from, Indirect *to)
126 {
127         while (from <= to && from->key == *from->p)
128                 from++;
129         return (from > to);
130 }
131
132 /**
133  *      ext2_block_to_path - parse the block number into array of offsets
134  *      @inode: inode in question (we are only interested in its superblock)
135  *      @i_block: block number to be parsed
136  *      @offsets: array to store the offsets in
137  *      @boundary: set this non-zero if the referred-to block is likely to be
138  *             followed (on disk) by an indirect block.
139  *      To store the locations of file's data ext2 uses a data structure common
140  *      for UNIX filesystems - tree of pointers anchored in the inode, with
141  *      data blocks at leaves and indirect blocks in intermediate nodes.
142  *      This function translates the block number into path in that tree -
143  *      return value is the path length and @offsets[n] is the offset of
144  *      pointer to (n+1)th node in the nth one. If @block is out of range
145  *      (negative or too large) warning is printed and zero returned.
146  *
147  *      Note: function doesn't find node addresses, so no IO is needed. All
148  *      we need to know is the capacity of indirect blocks (taken from the
149  *      inode->i_sb).
150  */
151
152 /*
153  * Portability note: the last comparison (check that we fit into triple
154  * indirect block) is spelled differently, because otherwise on an
155  * architecture with 32-bit longs and 8Kb pages we might get into trouble
156  * if our filesystem had 8Kb blocks. We might use long long, but that would
157  * kill us on x86. Oh, well, at least the sign propagation does not matter -
158  * i_block would have to be negative in the very beginning, so we would not
159  * get there at all.
160  */
161
162 static int ext2_block_to_path(struct inode *inode,
163                         long i_block, int offsets[4], int *boundary)
164 {
165         int ptrs = EXT2_ADDR_PER_BLOCK(inode->i_sb);
166         int ptrs_bits = EXT2_ADDR_PER_BLOCK_BITS(inode->i_sb);
167         const long direct_blocks = EXT2_NDIR_BLOCKS,
168                 indirect_blocks = ptrs,
169                 double_blocks = (1 << (ptrs_bits * 2));
170         int n = 0;
171         int final = 0;
172
173         if (i_block < 0) {
174                 ext2_msg(inode->i_sb, KERN_WARNING,
175                         "warning: %s: block < 0", __func__);
176         } else if (i_block < direct_blocks) {
177                 offsets[n++] = i_block;
178                 final = direct_blocks;
179         } else if ( (i_block -= direct_blocks) < indirect_blocks) {
180                 offsets[n++] = EXT2_IND_BLOCK;
181                 offsets[n++] = i_block;
182                 final = ptrs;
183         } else if ((i_block -= indirect_blocks) < double_blocks) {
184                 offsets[n++] = EXT2_DIND_BLOCK;
185                 offsets[n++] = i_block >> ptrs_bits;
186                 offsets[n++] = i_block & (ptrs - 1);
187                 final = ptrs;
188         } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
189                 offsets[n++] = EXT2_TIND_BLOCK;
190                 offsets[n++] = i_block >> (ptrs_bits * 2);
191                 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
192                 offsets[n++] = i_block & (ptrs - 1);
193                 final = ptrs;
194         } else {
195                 ext2_msg(inode->i_sb, KERN_WARNING,
196                         "warning: %s: block is too big", __func__);
197         }
198         if (boundary)
199                 *boundary = final - 1 - (i_block & (ptrs - 1));
200
201         return n;
202 }
203
204 /**
205  *      ext2_get_branch - read the chain of indirect blocks leading to data
206  *      @inode: inode in question
207  *      @depth: depth of the chain (1 - direct pointer, etc.)
208  *      @offsets: offsets of pointers in inode/indirect blocks
209  *      @chain: place to store the result
210  *      @err: here we store the error value
211  *
212  *      Function fills the array of triples <key, p, bh> and returns %NULL
213  *      if everything went OK or the pointer to the last filled triple
214  *      (incomplete one) otherwise. Upon the return chain[i].key contains
215  *      the number of (i+1)-th block in the chain (as it is stored in memory,
216  *      i.e. little-endian 32-bit), chain[i].p contains the address of that
217  *      number (it points into struct inode for i==0 and into the bh->b_data
218  *      for i>0) and chain[i].bh points to the buffer_head of i-th indirect
219  *      block for i>0 and NULL for i==0. In other words, it holds the block
220  *      numbers of the chain, addresses they were taken from (and where we can
221  *      verify that chain did not change) and buffer_heads hosting these
222  *      numbers.
223  *
224  *      Function stops when it stumbles upon zero pointer (absent block)
225  *              (pointer to last triple returned, *@err == 0)
226  *      or when it gets an IO error reading an indirect block
227  *              (ditto, *@err == -EIO)
228  *      or when it notices that chain had been changed while it was reading
229  *              (ditto, *@err == -EAGAIN)
230  *      or when it reads all @depth-1 indirect blocks successfully and finds
231  *      the whole chain, all way to the data (returns %NULL, *err == 0).
232  */
233 static Indirect *ext2_get_branch(struct inode *inode,
234                                  int depth,
235                                  int *offsets,
236                                  Indirect chain[4],
237                                  int *err)
238 {
239         struct super_block *sb = inode->i_sb;
240         Indirect *p = chain;
241         struct buffer_head *bh;
242
243         *err = 0;
244         /* i_data is not going away, no lock needed */
245         add_chain (chain, NULL, EXT2_I(inode)->i_data + *offsets);
246         if (!p->key)
247                 goto no_block;
248         while (--depth) {
249                 bh = sb_bread(sb, le32_to_cpu(p->key));
250                 if (!bh)
251                         goto failure;
252                 read_lock(&EXT2_I(inode)->i_meta_lock);
253                 if (!verify_chain(chain, p))
254                         goto changed;
255                 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
256                 read_unlock(&EXT2_I(inode)->i_meta_lock);
257                 if (!p->key)
258                         goto no_block;
259         }
260         return NULL;
261
262 changed:
263         read_unlock(&EXT2_I(inode)->i_meta_lock);
264         brelse(bh);
265         *err = -EAGAIN;
266         goto no_block;
267 failure:
268         *err = -EIO;
269 no_block:
270         return p;
271 }
272
273 /**
274  *      ext2_find_near - find a place for allocation with sufficient locality
275  *      @inode: owner
276  *      @ind: descriptor of indirect block.
277  *
278  *      This function returns the preferred place for block allocation.
279  *      It is used when heuristic for sequential allocation fails.
280  *      Rules are:
281  *        + if there is a block to the left of our position - allocate near it.
282  *        + if pointer will live in indirect block - allocate near that block.
283  *        + if pointer will live in inode - allocate in the same cylinder group.
284  *
285  * In the latter case we colour the starting block by the callers PID to
286  * prevent it from clashing with concurrent allocations for a different inode
287  * in the same block group.   The PID is used here so that functionally related
288  * files will be close-by on-disk.
289  *
290  *      Caller must make sure that @ind is valid and will stay that way.
291  */
292
293 static ext2_fsblk_t ext2_find_near(struct inode *inode, Indirect *ind)
294 {
295         struct ext2_inode_info *ei = EXT2_I(inode);
296         __le32 *start = ind->bh ? (__le32 *) ind->bh->b_data : ei->i_data;
297         __le32 *p;
298         ext2_fsblk_t bg_start;
299         ext2_fsblk_t colour;
300
301         /* Try to find previous block */
302         for (p = ind->p - 1; p >= start; p--)
303                 if (*p)
304                         return le32_to_cpu(*p);
305
306         /* No such thing, so let's try location of indirect block */
307         if (ind->bh)
308                 return ind->bh->b_blocknr;
309
310         /*
311          * It is going to be referred from inode itself? OK, just put it into
312          * the same cylinder group then.
313          */
314         bg_start = ext2_group_first_block_no(inode->i_sb, ei->i_block_group);
315         colour = (current->pid % 16) *
316                         (EXT2_BLOCKS_PER_GROUP(inode->i_sb) / 16);
317         return bg_start + colour;
318 }
319
320 /**
321  *      ext2_find_goal - find a preferred place for allocation.
322  *      @inode: owner
323  *      @block:  block we want
324  *      @partial: pointer to the last triple within a chain
325  *
326  *      Returns preferred place for a block (the goal).
327  */
328
329 static inline ext2_fsblk_t ext2_find_goal(struct inode *inode, long block,
330                                           Indirect *partial)
331 {
332         struct ext2_block_alloc_info *block_i;
333
334         block_i = EXT2_I(inode)->i_block_alloc_info;
335
336         /*
337          * try the heuristic for sequential allocation,
338          * failing that at least try to get decent locality.
339          */
340         if (block_i && (block == block_i->last_alloc_logical_block + 1)
341                 && (block_i->last_alloc_physical_block != 0)) {
342                 return block_i->last_alloc_physical_block + 1;
343         }
344
345         return ext2_find_near(inode, partial);
346 }
347
348 /**
349  *      ext2_blks_to_allocate: Look up the block map and count the number
350  *      of direct blocks need to be allocated for the given branch.
351  *
352  *      @branch: chain of indirect blocks
353  *      @k: number of blocks need for indirect blocks
354  *      @blks: number of data blocks to be mapped.
355  *      @blocks_to_boundary:  the offset in the indirect block
356  *
357  *      return the total number of blocks to be allocate, including the
358  *      direct and indirect blocks.
359  */
360 static int
361 ext2_blks_to_allocate(Indirect * branch, int k, unsigned long blks,
362                 int blocks_to_boundary)
363 {
364         unsigned long count = 0;
365
366         /*
367          * Simple case, [t,d]Indirect block(s) has not allocated yet
368          * then it's clear blocks on that path have not allocated
369          */
370         if (k > 0) {
371                 /* right now don't hanel cross boundary allocation */
372                 if (blks < blocks_to_boundary + 1)
373                         count += blks;
374                 else
375                         count += blocks_to_boundary + 1;
376                 return count;
377         }
378
379         count++;
380         while (count < blks && count <= blocks_to_boundary
381                 && le32_to_cpu(*(branch[0].p + count)) == 0) {
382                 count++;
383         }
384         return count;
385 }
386
387 /**
388  *      ext2_alloc_blocks: multiple allocate blocks needed for a branch
389  *      @indirect_blks: the number of blocks need to allocate for indirect
390  *                      blocks
391  *
392  *      @new_blocks: on return it will store the new block numbers for
393  *      the indirect blocks(if needed) and the first direct block,
394  *      @blks:  on return it will store the total number of allocated
395  *              direct blocks
396  */
397 static int ext2_alloc_blocks(struct inode *inode,
398                         ext2_fsblk_t goal, int indirect_blks, int blks,
399                         ext2_fsblk_t new_blocks[4], int *err)
400 {
401         int target, i;
402         unsigned long count = 0;
403         int index = 0;
404         ext2_fsblk_t current_block = 0;
405         int ret = 0;
406
407         /*
408          * Here we try to allocate the requested multiple blocks at once,
409          * on a best-effort basis.
410          * To build a branch, we should allocate blocks for
411          * the indirect blocks(if not allocated yet), and at least
412          * the first direct block of this branch.  That's the
413          * minimum number of blocks need to allocate(required)
414          */
415         target = blks + indirect_blks;
416
417         while (1) {
418                 count = target;
419                 /* allocating blocks for indirect blocks and direct blocks */
420                 current_block = ext2_new_blocks(inode,goal,&count,err);
421                 if (*err)
422                         goto failed_out;
423
424                 target -= count;
425                 /* allocate blocks for indirect blocks */
426                 while (index < indirect_blks && count) {
427                         new_blocks[index++] = current_block++;
428                         count--;
429                 }
430
431                 if (count > 0)
432                         break;
433         }
434
435         /* save the new block number for the first direct block */
436         new_blocks[index] = current_block;
437
438         /* total number of blocks allocated for direct blocks */
439         ret = count;
440         *err = 0;
441         return ret;
442 failed_out:
443         for (i = 0; i <index; i++)
444                 ext2_free_blocks(inode, new_blocks[i], 1);
445         if (index)
446                 mark_inode_dirty(inode);
447         return ret;
448 }
449
450 /**
451  *      ext2_alloc_branch - allocate and set up a chain of blocks.
452  *      @inode: owner
453  *      @num: depth of the chain (number of blocks to allocate)
454  *      @offsets: offsets (in the blocks) to store the pointers to next.
455  *      @branch: place to store the chain in.
456  *
457  *      This function allocates @num blocks, zeroes out all but the last one,
458  *      links them into chain and (if we are synchronous) writes them to disk.
459  *      In other words, it prepares a branch that can be spliced onto the
460  *      inode. It stores the information about that chain in the branch[], in
461  *      the same format as ext2_get_branch() would do. We are calling it after
462  *      we had read the existing part of chain and partial points to the last
463  *      triple of that (one with zero ->key). Upon the exit we have the same
464  *      picture as after the successful ext2_get_block(), except that in one
465  *      place chain is disconnected - *branch->p is still zero (we did not
466  *      set the last link), but branch->key contains the number that should
467  *      be placed into *branch->p to fill that gap.
468  *
469  *      If allocation fails we free all blocks we've allocated (and forget
470  *      their buffer_heads) and return the error value the from failed
471  *      ext2_alloc_block() (normally -ENOSPC). Otherwise we set the chain
472  *      as described above and return 0.
473  */
474
475 static int ext2_alloc_branch(struct inode *inode,
476                         int indirect_blks, int *blks, ext2_fsblk_t goal,
477                         int *offsets, Indirect *branch)
478 {
479         int blocksize = inode->i_sb->s_blocksize;
480         int i, n = 0;
481         int err = 0;
482         struct buffer_head *bh;
483         int num;
484         ext2_fsblk_t new_blocks[4];
485         ext2_fsblk_t current_block;
486
487         num = ext2_alloc_blocks(inode, goal, indirect_blks,
488                                 *blks, new_blocks, &err);
489         if (err)
490                 return err;
491
492         branch[0].key = cpu_to_le32(new_blocks[0]);
493         /*
494          * metadata blocks and data blocks are allocated.
495          */
496         for (n = 1; n <= indirect_blks;  n++) {
497                 /*
498                  * Get buffer_head for parent block, zero it out
499                  * and set the pointer to new one, then send
500                  * parent to disk.
501                  */
502                 bh = sb_getblk(inode->i_sb, new_blocks[n-1]);
503                 if (unlikely(!bh)) {
504                         err = -ENOMEM;
505                         goto failed;
506                 }
507                 branch[n].bh = bh;
508                 lock_buffer(bh);
509                 memset(bh->b_data, 0, blocksize);
510                 branch[n].p = (__le32 *) bh->b_data + offsets[n];
511                 branch[n].key = cpu_to_le32(new_blocks[n]);
512                 *branch[n].p = branch[n].key;
513                 if ( n == indirect_blks) {
514                         current_block = new_blocks[n];
515                         /*
516                          * End of chain, update the last new metablock of
517                          * the chain to point to the new allocated
518                          * data blocks numbers
519                          */
520                         for (i=1; i < num; i++)
521                                 *(branch[n].p + i) = cpu_to_le32(++current_block);
522                 }
523                 set_buffer_uptodate(bh);
524                 unlock_buffer(bh);
525                 mark_buffer_dirty_inode(bh, inode);
526                 /* We used to sync bh here if IS_SYNC(inode).
527                  * But we now rely upon generic_write_sync()
528                  * and b_inode_buffers.  But not for directories.
529                  */
530                 if (S_ISDIR(inode->i_mode) && IS_DIRSYNC(inode))
531                         sync_dirty_buffer(bh);
532         }
533         *blks = num;
534         return err;
535
536 failed:
537         for (i = 1; i < n; i++)
538                 bforget(branch[i].bh);
539         for (i = 0; i < indirect_blks; i++)
540                 ext2_free_blocks(inode, new_blocks[i], 1);
541         ext2_free_blocks(inode, new_blocks[i], num);
542         return err;
543 }
544
545 /**
546  * ext2_splice_branch - splice the allocated branch onto inode.
547  * @inode: owner
548  * @block: (logical) number of block we are adding
549  * @where: location of missing link
550  * @num:   number of indirect blocks we are adding
551  * @blks:  number of direct blocks we are adding
552  *
553  * This function fills the missing link and does all housekeeping needed in
554  * inode (->i_blocks, etc.). In case of success we end up with the full
555  * chain to new block and return 0.
556  */
557 static void ext2_splice_branch(struct inode *inode,
558                         long block, Indirect *where, int num, int blks)
559 {
560         int i;
561         struct ext2_block_alloc_info *block_i;
562         ext2_fsblk_t current_block;
563
564         block_i = EXT2_I(inode)->i_block_alloc_info;
565
566         /* XXX LOCKING probably should have i_meta_lock ?*/
567         /* That's it */
568
569         *where->p = where->key;
570
571         /*
572          * Update the host buffer_head or inode to point to more just allocated
573          * direct blocks blocks
574          */
575         if (num == 0 && blks > 1) {
576                 current_block = le32_to_cpu(where->key) + 1;
577                 for (i = 1; i < blks; i++)
578                         *(where->p + i ) = cpu_to_le32(current_block++);
579         }
580
581         /*
582          * update the most recently allocated logical & physical block
583          * in i_block_alloc_info, to assist find the proper goal block for next
584          * allocation
585          */
586         if (block_i) {
587                 block_i->last_alloc_logical_block = block + blks - 1;
588                 block_i->last_alloc_physical_block =
589                                 le32_to_cpu(where[num].key) + blks - 1;
590         }
591
592         /* We are done with atomic stuff, now do the rest of housekeeping */
593
594         /* had we spliced it onto indirect block? */
595         if (where->bh)
596                 mark_buffer_dirty_inode(where->bh, inode);
597
598         inode->i_ctime = current_time(inode);
599         mark_inode_dirty(inode);
600 }
601
602 /*
603  * Allocation strategy is simple: if we have to allocate something, we will
604  * have to go the whole way to leaf. So let's do it before attaching anything
605  * to tree, set linkage between the newborn blocks, write them if sync is
606  * required, recheck the path, free and repeat if check fails, otherwise
607  * set the last missing link (that will protect us from any truncate-generated
608  * removals - all blocks on the path are immune now) and possibly force the
609  * write on the parent block.
610  * That has a nice additional property: no special recovery from the failed
611  * allocations is needed - we simply release blocks and do not touch anything
612  * reachable from inode.
613  *
614  * `handle' can be NULL if create == 0.
615  *
616  * return > 0, # of blocks mapped or allocated.
617  * return = 0, if plain lookup failed.
618  * return < 0, error case.
619  */
620 static int ext2_get_blocks(struct inode *inode,
621                            sector_t iblock, unsigned long maxblocks,
622                            u32 *bno, bool *new, bool *boundary,
623                            int create)
624 {
625         int err;
626         int offsets[4];
627         Indirect chain[4];
628         Indirect *partial;
629         ext2_fsblk_t goal;
630         int indirect_blks;
631         int blocks_to_boundary = 0;
632         int depth;
633         struct ext2_inode_info *ei = EXT2_I(inode);
634         int count = 0;
635         ext2_fsblk_t first_block = 0;
636
637         BUG_ON(maxblocks == 0);
638
639         depth = ext2_block_to_path(inode,iblock,offsets,&blocks_to_boundary);
640
641         if (depth == 0)
642                 return -EIO;
643
644         partial = ext2_get_branch(inode, depth, offsets, chain, &err);
645         /* Simplest case - block found, no allocation needed */
646         if (!partial) {
647                 first_block = le32_to_cpu(chain[depth - 1].key);
648                 count++;
649                 /*map more blocks*/
650                 while (count < maxblocks && count <= blocks_to_boundary) {
651                         ext2_fsblk_t blk;
652
653                         if (!verify_chain(chain, chain + depth - 1)) {
654                                 /*
655                                  * Indirect block might be removed by
656                                  * truncate while we were reading it.
657                                  * Handling of that case: forget what we've
658                                  * got now, go to reread.
659                                  */
660                                 err = -EAGAIN;
661                                 count = 0;
662                                 break;
663                         }
664                         blk = le32_to_cpu(*(chain[depth-1].p + count));
665                         if (blk == first_block + count)
666                                 count++;
667                         else
668                                 break;
669                 }
670                 if (err != -EAGAIN)
671                         goto got_it;
672         }
673
674         /* Next simple case - plain lookup or failed read of indirect block */
675         if (!create || err == -EIO)
676                 goto cleanup;
677
678         mutex_lock(&ei->truncate_mutex);
679         /*
680          * If the indirect block is missing while we are reading
681          * the chain(ext2_get_branch() returns -EAGAIN err), or
682          * if the chain has been changed after we grab the semaphore,
683          * (either because another process truncated this branch, or
684          * another get_block allocated this branch) re-grab the chain to see if
685          * the request block has been allocated or not.
686          *
687          * Since we already block the truncate/other get_block
688          * at this point, we will have the current copy of the chain when we
689          * splice the branch into the tree.
690          */
691         if (err == -EAGAIN || !verify_chain(chain, partial)) {
692                 while (partial > chain) {
693                         brelse(partial->bh);
694                         partial--;
695                 }
696                 partial = ext2_get_branch(inode, depth, offsets, chain, &err);
697                 if (!partial) {
698                         count++;
699                         mutex_unlock(&ei->truncate_mutex);
700                         if (err)
701                                 goto cleanup;
702                         goto got_it;
703                 }
704         }
705
706         /*
707          * Okay, we need to do block allocation.  Lazily initialize the block
708          * allocation info here if necessary
709         */
710         if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info))
711                 ext2_init_block_alloc_info(inode);
712
713         goal = ext2_find_goal(inode, iblock, partial);
714
715         /* the number of blocks need to allocate for [d,t]indirect blocks */
716         indirect_blks = (chain + depth) - partial - 1;
717         /*
718          * Next look up the indirect map to count the totoal number of
719          * direct blocks to allocate for this branch.
720          */
721         count = ext2_blks_to_allocate(partial, indirect_blks,
722                                         maxblocks, blocks_to_boundary);
723         /*
724          * XXX ???? Block out ext2_truncate while we alter the tree
725          */
726         err = ext2_alloc_branch(inode, indirect_blks, &count, goal,
727                                 offsets + (partial - chain), partial);
728
729         if (err) {
730                 mutex_unlock(&ei->truncate_mutex);
731                 goto cleanup;
732         }
733
734         if (IS_DAX(inode)) {
735                 /*
736                  * We must unmap blocks before zeroing so that writeback cannot
737                  * overwrite zeros with stale data from block device page cache.
738                  */
739                 clean_bdev_aliases(inode->i_sb->s_bdev,
740                                    le32_to_cpu(chain[depth-1].key),
741                                    count);
742                 /*
743                  * block must be initialised before we put it in the tree
744                  * so that it's not found by another thread before it's
745                  * initialised
746                  */
747                 err = sb_issue_zeroout(inode->i_sb,
748                                 le32_to_cpu(chain[depth-1].key), count,
749                                 GFP_NOFS);
750                 if (err) {
751                         mutex_unlock(&ei->truncate_mutex);
752                         goto cleanup;
753                 }
754         } else {
755                 *new = true;
756         }
757
758         ext2_splice_branch(inode, iblock, partial, indirect_blks, count);
759         mutex_unlock(&ei->truncate_mutex);
760 got_it:
761         if (count > blocks_to_boundary)
762                 *boundary = true;
763         err = count;
764         /* Clean up and exit */
765         partial = chain + depth - 1;    /* the whole chain */
766 cleanup:
767         while (partial > chain) {
768                 brelse(partial->bh);
769                 partial--;
770         }
771         if (err > 0)
772                 *bno = le32_to_cpu(chain[depth-1].key);
773         return err;
774 }
775
776 int ext2_get_block(struct inode *inode, sector_t iblock,
777                 struct buffer_head *bh_result, int create)
778 {
779         unsigned max_blocks = bh_result->b_size >> inode->i_blkbits;
780         bool new = false, boundary = false;
781         u32 bno;
782         int ret;
783
784         ret = ext2_get_blocks(inode, iblock, max_blocks, &bno, &new, &boundary,
785                         create);
786         if (ret <= 0)
787                 return ret;
788
789         map_bh(bh_result, inode->i_sb, bno);
790         bh_result->b_size = (ret << inode->i_blkbits);
791         if (new)
792                 set_buffer_new(bh_result);
793         if (boundary)
794                 set_buffer_boundary(bh_result);
795         return 0;
796
797 }
798
799 #ifdef CONFIG_FS_DAX
800 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
801                 unsigned flags, struct iomap *iomap)
802 {
803         unsigned int blkbits = inode->i_blkbits;
804         unsigned long first_block = offset >> blkbits;
805         unsigned long max_blocks = (length + (1 << blkbits) - 1) >> blkbits;
806         bool new = false, boundary = false;
807         u32 bno;
808         int ret;
809
810         ret = ext2_get_blocks(inode, first_block, max_blocks,
811                         &bno, &new, &boundary, flags & IOMAP_WRITE);
812         if (ret < 0)
813                 return ret;
814
815         iomap->flags = 0;
816         iomap->bdev = inode->i_sb->s_bdev;
817         iomap->offset = (u64)first_block << blkbits;
818
819         if (ret == 0) {
820                 iomap->type = IOMAP_HOLE;
821                 iomap->blkno = IOMAP_NULL_BLOCK;
822                 iomap->length = 1 << blkbits;
823         } else {
824                 iomap->type = IOMAP_MAPPED;
825                 iomap->blkno = (sector_t)bno << (blkbits - 9);
826                 iomap->length = (u64)ret << blkbits;
827                 iomap->flags |= IOMAP_F_MERGED;
828         }
829
830         if (new)
831                 iomap->flags |= IOMAP_F_NEW;
832         return 0;
833 }
834
835 static int
836 ext2_iomap_end(struct inode *inode, loff_t offset, loff_t length,
837                 ssize_t written, unsigned flags, struct iomap *iomap)
838 {
839         if (iomap->type == IOMAP_MAPPED &&
840             written < length &&
841             (flags & IOMAP_WRITE))
842                 ext2_write_failed(inode->i_mapping, offset + length);
843         return 0;
844 }
845
846 struct iomap_ops ext2_iomap_ops = {
847         .iomap_begin            = ext2_iomap_begin,
848         .iomap_end              = ext2_iomap_end,
849 };
850 #else
851 /* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
852 struct iomap_ops ext2_iomap_ops;
853 #endif /* CONFIG_FS_DAX */
854
855 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
856                 u64 start, u64 len)
857 {
858         return generic_block_fiemap(inode, fieinfo, start, len,
859                                     ext2_get_block);
860 }
861
862 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
863 {
864         return block_write_full_page(page, ext2_get_block, wbc);
865 }
866
867 static int ext2_readpage(struct file *file, struct page *page)
868 {
869         return mpage_readpage(page, ext2_get_block);
870 }
871
872 static int
873 ext2_readpages(struct file *file, struct address_space *mapping,
874                 struct list_head *pages, unsigned nr_pages)
875 {
876         return mpage_readpages(mapping, pages, nr_pages, ext2_get_block);
877 }
878
879 static int
880 ext2_write_begin(struct file *file, struct address_space *mapping,
881                 loff_t pos, unsigned len, unsigned flags,
882                 struct page **pagep, void **fsdata)
883 {
884         int ret;
885
886         ret = block_write_begin(mapping, pos, len, flags, pagep,
887                                 ext2_get_block);
888         if (ret < 0)
889                 ext2_write_failed(mapping, pos + len);
890         return ret;
891 }
892
893 static int ext2_write_end(struct file *file, struct address_space *mapping,
894                         loff_t pos, unsigned len, unsigned copied,
895                         struct page *page, void *fsdata)
896 {
897         int ret;
898
899         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
900         if (ret < len)
901                 ext2_write_failed(mapping, pos + len);
902         return ret;
903 }
904
905 static int
906 ext2_nobh_write_begin(struct file *file, struct address_space *mapping,
907                 loff_t pos, unsigned len, unsigned flags,
908                 struct page **pagep, void **fsdata)
909 {
910         int ret;
911
912         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
913                                ext2_get_block);
914         if (ret < 0)
915                 ext2_write_failed(mapping, pos + len);
916         return ret;
917 }
918
919 static int ext2_nobh_writepage(struct page *page,
920                         struct writeback_control *wbc)
921 {
922         return nobh_writepage(page, ext2_get_block, wbc);
923 }
924
925 static sector_t ext2_bmap(struct address_space *mapping, sector_t block)
926 {
927         return generic_block_bmap(mapping,block,ext2_get_block);
928 }
929
930 static ssize_t
931 ext2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
932 {
933         struct file *file = iocb->ki_filp;
934         struct address_space *mapping = file->f_mapping;
935         struct inode *inode = mapping->host;
936         size_t count = iov_iter_count(iter);
937         loff_t offset = iocb->ki_pos;
938         ssize_t ret;
939
940         if (WARN_ON_ONCE(IS_DAX(inode)))
941                 return -EIO;
942
943         ret = blockdev_direct_IO(iocb, inode, iter, ext2_get_block);
944         if (ret < 0 && iov_iter_rw(iter) == WRITE)
945                 ext2_write_failed(mapping, offset + count);
946         return ret;
947 }
948
949 static int
950 ext2_writepages(struct address_space *mapping, struct writeback_control *wbc)
951 {
952 #ifdef CONFIG_FS_DAX
953         if (dax_mapping(mapping)) {
954                 return dax_writeback_mapping_range(mapping,
955                                                    mapping->host->i_sb->s_bdev,
956                                                    wbc);
957         }
958 #endif
959
960         return mpage_writepages(mapping, wbc, ext2_get_block);
961 }
962
963 const struct address_space_operations ext2_aops = {
964         .readpage               = ext2_readpage,
965         .readpages              = ext2_readpages,
966         .writepage              = ext2_writepage,
967         .write_begin            = ext2_write_begin,
968         .write_end              = ext2_write_end,
969         .bmap                   = ext2_bmap,
970         .direct_IO              = ext2_direct_IO,
971         .writepages             = ext2_writepages,
972         .migratepage            = buffer_migrate_page,
973         .is_partially_uptodate  = block_is_partially_uptodate,
974         .error_remove_page      = generic_error_remove_page,
975 };
976
977 const struct address_space_operations ext2_nobh_aops = {
978         .readpage               = ext2_readpage,
979         .readpages              = ext2_readpages,
980         .writepage              = ext2_nobh_writepage,
981         .write_begin            = ext2_nobh_write_begin,
982         .write_end              = nobh_write_end,
983         .bmap                   = ext2_bmap,
984         .direct_IO              = ext2_direct_IO,
985         .writepages             = ext2_writepages,
986         .migratepage            = buffer_migrate_page,
987         .error_remove_page      = generic_error_remove_page,
988 };
989
990 /*
991  * Probably it should be a library function... search for first non-zero word
992  * or memcmp with zero_page, whatever is better for particular architecture.
993  * Linus?
994  */
995 static inline int all_zeroes(__le32 *p, __le32 *q)
996 {
997         while (p < q)
998                 if (*p++)
999                         return 0;
1000         return 1;
1001 }
1002
1003 /**
1004  *      ext2_find_shared - find the indirect blocks for partial truncation.
1005  *      @inode:   inode in question
1006  *      @depth:   depth of the affected branch
1007  *      @offsets: offsets of pointers in that branch (see ext2_block_to_path)
1008  *      @chain:   place to store the pointers to partial indirect blocks
1009  *      @top:     place to the (detached) top of branch
1010  *
1011  *      This is a helper function used by ext2_truncate().
1012  *
1013  *      When we do truncate() we may have to clean the ends of several indirect
1014  *      blocks but leave the blocks themselves alive. Block is partially
1015  *      truncated if some data below the new i_size is referred from it (and
1016  *      it is on the path to the first completely truncated data block, indeed).
1017  *      We have to free the top of that path along with everything to the right
1018  *      of the path. Since no allocation past the truncation point is possible
1019  *      until ext2_truncate() finishes, we may safely do the latter, but top
1020  *      of branch may require special attention - pageout below the truncation
1021  *      point might try to populate it.
1022  *
1023  *      We atomically detach the top of branch from the tree, store the block
1024  *      number of its root in *@top, pointers to buffer_heads of partially
1025  *      truncated blocks - in @chain[].bh and pointers to their last elements
1026  *      that should not be removed - in @chain[].p. Return value is the pointer
1027  *      to last filled element of @chain.
1028  *
1029  *      The work left to caller to do the actual freeing of subtrees:
1030  *              a) free the subtree starting from *@top
1031  *              b) free the subtrees whose roots are stored in
1032  *                      (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1033  *              c) free the subtrees growing from the inode past the @chain[0].p
1034  *                      (no partially truncated stuff there).
1035  */
1036
1037 static Indirect *ext2_find_shared(struct inode *inode,
1038                                 int depth,
1039                                 int offsets[4],
1040                                 Indirect chain[4],
1041                                 __le32 *top)
1042 {
1043         Indirect *partial, *p;
1044         int k, err;
1045
1046         *top = 0;
1047         for (k = depth; k > 1 && !offsets[k-1]; k--)
1048                 ;
1049         partial = ext2_get_branch(inode, k, offsets, chain, &err);
1050         if (!partial)
1051                 partial = chain + k-1;
1052         /*
1053          * If the branch acquired continuation since we've looked at it -
1054          * fine, it should all survive and (new) top doesn't belong to us.
1055          */
1056         write_lock(&EXT2_I(inode)->i_meta_lock);
1057         if (!partial->key && *partial->p) {
1058                 write_unlock(&EXT2_I(inode)->i_meta_lock);
1059                 goto no_top;
1060         }
1061         for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1062                 ;
1063         /*
1064          * OK, we've found the last block that must survive. The rest of our
1065          * branch should be detached before unlocking. However, if that rest
1066          * of branch is all ours and does not grow immediately from the inode
1067          * it's easier to cheat and just decrement partial->p.
1068          */
1069         if (p == chain + k - 1 && p > chain) {
1070                 p->p--;
1071         } else {
1072                 *top = *p->p;
1073                 *p->p = 0;
1074         }
1075         write_unlock(&EXT2_I(inode)->i_meta_lock);
1076
1077         while(partial > p)
1078         {
1079                 brelse(partial->bh);
1080                 partial--;
1081         }
1082 no_top:
1083         return partial;
1084 }
1085
1086 /**
1087  *      ext2_free_data - free a list of data blocks
1088  *      @inode: inode we are dealing with
1089  *      @p:     array of block numbers
1090  *      @q:     points immediately past the end of array
1091  *
1092  *      We are freeing all blocks referred from that array (numbers are
1093  *      stored as little-endian 32-bit) and updating @inode->i_blocks
1094  *      appropriately.
1095  */
1096 static inline void ext2_free_data(struct inode *inode, __le32 *p, __le32 *q)
1097 {
1098         unsigned long block_to_free = 0, count = 0;
1099         unsigned long nr;
1100
1101         for ( ; p < q ; p++) {
1102                 nr = le32_to_cpu(*p);
1103                 if (nr) {
1104                         *p = 0;
1105                         /* accumulate blocks to free if they're contiguous */
1106                         if (count == 0)
1107                                 goto free_this;
1108                         else if (block_to_free == nr - count)
1109                                 count++;
1110                         else {
1111                                 ext2_free_blocks (inode, block_to_free, count);
1112                                 mark_inode_dirty(inode);
1113                         free_this:
1114                                 block_to_free = nr;
1115                                 count = 1;
1116                         }
1117                 }
1118         }
1119         if (count > 0) {
1120                 ext2_free_blocks (inode, block_to_free, count);
1121                 mark_inode_dirty(inode);
1122         }
1123 }
1124
1125 /**
1126  *      ext2_free_branches - free an array of branches
1127  *      @inode: inode we are dealing with
1128  *      @p:     array of block numbers
1129  *      @q:     pointer immediately past the end of array
1130  *      @depth: depth of the branches to free
1131  *
1132  *      We are freeing all blocks referred from these branches (numbers are
1133  *      stored as little-endian 32-bit) and updating @inode->i_blocks
1134  *      appropriately.
1135  */
1136 static void ext2_free_branches(struct inode *inode, __le32 *p, __le32 *q, int depth)
1137 {
1138         struct buffer_head * bh;
1139         unsigned long nr;
1140
1141         if (depth--) {
1142                 int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1143                 for ( ; p < q ; p++) {
1144                         nr = le32_to_cpu(*p);
1145                         if (!nr)
1146                                 continue;
1147                         *p = 0;
1148                         bh = sb_bread(inode->i_sb, nr);
1149                         /*
1150                          * A read failure? Report error and clear slot
1151                          * (should be rare).
1152                          */ 
1153                         if (!bh) {
1154                                 ext2_error(inode->i_sb, "ext2_free_branches",
1155                                         "Read failure, inode=%ld, block=%ld",
1156                                         inode->i_ino, nr);
1157                                 continue;
1158                         }
1159                         ext2_free_branches(inode,
1160                                            (__le32*)bh->b_data,
1161                                            (__le32*)bh->b_data + addr_per_block,
1162                                            depth);
1163                         bforget(bh);
1164                         ext2_free_blocks(inode, nr, 1);
1165                         mark_inode_dirty(inode);
1166                 }
1167         } else
1168                 ext2_free_data(inode, p, q);
1169 }
1170
1171 /* dax_sem must be held when calling this function */
1172 static void __ext2_truncate_blocks(struct inode *inode, loff_t offset)
1173 {
1174         __le32 *i_data = EXT2_I(inode)->i_data;
1175         struct ext2_inode_info *ei = EXT2_I(inode);
1176         int addr_per_block = EXT2_ADDR_PER_BLOCK(inode->i_sb);
1177         int offsets[4];
1178         Indirect chain[4];
1179         Indirect *partial;
1180         __le32 nr = 0;
1181         int n;
1182         long iblock;
1183         unsigned blocksize;
1184         blocksize = inode->i_sb->s_blocksize;
1185         iblock = (offset + blocksize-1) >> EXT2_BLOCK_SIZE_BITS(inode->i_sb);
1186
1187 #ifdef CONFIG_FS_DAX
1188         WARN_ON(!rwsem_is_locked(&ei->dax_sem));
1189 #endif
1190
1191         n = ext2_block_to_path(inode, iblock, offsets, NULL);
1192         if (n == 0)
1193                 return;
1194
1195         /*
1196          * From here we block out all ext2_get_block() callers who want to
1197          * modify the block allocation tree.
1198          */
1199         mutex_lock(&ei->truncate_mutex);
1200
1201         if (n == 1) {
1202                 ext2_free_data(inode, i_data+offsets[0],
1203                                         i_data + EXT2_NDIR_BLOCKS);
1204                 goto do_indirects;
1205         }
1206
1207         partial = ext2_find_shared(inode, n, offsets, chain, &nr);
1208         /* Kill the top of shared branch (already detached) */
1209         if (nr) {
1210                 if (partial == chain)
1211                         mark_inode_dirty(inode);
1212                 else
1213                         mark_buffer_dirty_inode(partial->bh, inode);
1214                 ext2_free_branches(inode, &nr, &nr+1, (chain+n-1) - partial);
1215         }
1216         /* Clear the ends of indirect blocks on the shared branch */
1217         while (partial > chain) {
1218                 ext2_free_branches(inode,
1219                                    partial->p + 1,
1220                                    (__le32*)partial->bh->b_data+addr_per_block,
1221                                    (chain+n-1) - partial);
1222                 mark_buffer_dirty_inode(partial->bh, inode);
1223                 brelse (partial->bh);
1224                 partial--;
1225         }
1226 do_indirects:
1227         /* Kill the remaining (whole) subtrees */
1228         switch (offsets[0]) {
1229                 default:
1230                         nr = i_data[EXT2_IND_BLOCK];
1231                         if (nr) {
1232                                 i_data[EXT2_IND_BLOCK] = 0;
1233                                 mark_inode_dirty(inode);
1234                                 ext2_free_branches(inode, &nr, &nr+1, 1);
1235                         }
1236                 case EXT2_IND_BLOCK:
1237                         nr = i_data[EXT2_DIND_BLOCK];
1238                         if (nr) {
1239                                 i_data[EXT2_DIND_BLOCK] = 0;
1240                                 mark_inode_dirty(inode);
1241                                 ext2_free_branches(inode, &nr, &nr+1, 2);
1242                         }
1243                 case EXT2_DIND_BLOCK:
1244                         nr = i_data[EXT2_TIND_BLOCK];
1245                         if (nr) {
1246                                 i_data[EXT2_TIND_BLOCK] = 0;
1247                                 mark_inode_dirty(inode);
1248                                 ext2_free_branches(inode, &nr, &nr+1, 3);
1249                         }
1250                 case EXT2_TIND_BLOCK:
1251                         ;
1252         }
1253
1254         ext2_discard_reservation(inode);
1255
1256         mutex_unlock(&ei->truncate_mutex);
1257 }
1258
1259 static void ext2_truncate_blocks(struct inode *inode, loff_t offset)
1260 {
1261         /*
1262          * XXX: it seems like a bug here that we don't allow
1263          * IS_APPEND inode to have blocks-past-i_size trimmed off.
1264          * review and fix this.
1265          *
1266          * Also would be nice to be able to handle IO errors and such,
1267          * but that's probably too much to ask.
1268          */
1269         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1270             S_ISLNK(inode->i_mode)))
1271                 return;
1272         if (ext2_inode_is_fast_symlink(inode))
1273                 return;
1274         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1275                 return;
1276
1277         dax_sem_down_write(EXT2_I(inode));
1278         __ext2_truncate_blocks(inode, offset);
1279         dax_sem_up_write(EXT2_I(inode));
1280 }
1281
1282 static int ext2_setsize(struct inode *inode, loff_t newsize)
1283 {
1284         int error;
1285
1286         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1287             S_ISLNK(inode->i_mode)))
1288                 return -EINVAL;
1289         if (ext2_inode_is_fast_symlink(inode))
1290                 return -EINVAL;
1291         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1292                 return -EPERM;
1293
1294         inode_dio_wait(inode);
1295
1296         if (IS_DAX(inode)) {
1297                 error = iomap_zero_range(inode, newsize,
1298                                          PAGE_ALIGN(newsize) - newsize, NULL,
1299                                          &ext2_iomap_ops);
1300         } else if (test_opt(inode->i_sb, NOBH))
1301                 error = nobh_truncate_page(inode->i_mapping,
1302                                 newsize, ext2_get_block);
1303         else
1304                 error = block_truncate_page(inode->i_mapping,
1305                                 newsize, ext2_get_block);
1306         if (error)
1307                 return error;
1308
1309         dax_sem_down_write(EXT2_I(inode));
1310         truncate_setsize(inode, newsize);
1311         __ext2_truncate_blocks(inode, newsize);
1312         dax_sem_up_write(EXT2_I(inode));
1313
1314         inode->i_mtime = inode->i_ctime = current_time(inode);
1315         if (inode_needs_sync(inode)) {
1316                 sync_mapping_buffers(inode->i_mapping);
1317                 sync_inode_metadata(inode, 1);
1318         } else {
1319                 mark_inode_dirty(inode);
1320         }
1321
1322         return 0;
1323 }
1324
1325 static struct ext2_inode *ext2_get_inode(struct super_block *sb, ino_t ino,
1326                                         struct buffer_head **p)
1327 {
1328         struct buffer_head * bh;
1329         unsigned long block_group;
1330         unsigned long block;
1331         unsigned long offset;
1332         struct ext2_group_desc * gdp;
1333
1334         *p = NULL;
1335         if ((ino != EXT2_ROOT_INO && ino < EXT2_FIRST_INO(sb)) ||
1336             ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
1337                 goto Einval;
1338
1339         block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
1340         gdp = ext2_get_group_desc(sb, block_group, NULL);
1341         if (!gdp)
1342                 goto Egdp;
1343         /*
1344          * Figure out the offset within the block group inode table
1345          */
1346         offset = ((ino - 1) % EXT2_INODES_PER_GROUP(sb)) * EXT2_INODE_SIZE(sb);
1347         block = le32_to_cpu(gdp->bg_inode_table) +
1348                 (offset >> EXT2_BLOCK_SIZE_BITS(sb));
1349         if (!(bh = sb_bread(sb, block)))
1350                 goto Eio;
1351
1352         *p = bh;
1353         offset &= (EXT2_BLOCK_SIZE(sb) - 1);
1354         return (struct ext2_inode *) (bh->b_data + offset);
1355
1356 Einval:
1357         ext2_error(sb, "ext2_get_inode", "bad inode number: %lu",
1358                    (unsigned long) ino);
1359         return ERR_PTR(-EINVAL);
1360 Eio:
1361         ext2_error(sb, "ext2_get_inode",
1362                    "unable to read inode block - inode=%lu, block=%lu",
1363                    (unsigned long) ino, block);
1364 Egdp:
1365         return ERR_PTR(-EIO);
1366 }
1367
1368 void ext2_set_inode_flags(struct inode *inode)
1369 {
1370         unsigned int flags = EXT2_I(inode)->i_flags;
1371
1372         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
1373                                 S_DIRSYNC | S_DAX);
1374         if (flags & EXT2_SYNC_FL)
1375                 inode->i_flags |= S_SYNC;
1376         if (flags & EXT2_APPEND_FL)
1377                 inode->i_flags |= S_APPEND;
1378         if (flags & EXT2_IMMUTABLE_FL)
1379                 inode->i_flags |= S_IMMUTABLE;
1380         if (flags & EXT2_NOATIME_FL)
1381                 inode->i_flags |= S_NOATIME;
1382         if (flags & EXT2_DIRSYNC_FL)
1383                 inode->i_flags |= S_DIRSYNC;
1384         if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode))
1385                 inode->i_flags |= S_DAX;
1386 }
1387
1388 /* Propagate flags from i_flags to EXT2_I(inode)->i_flags */
1389 void ext2_get_inode_flags(struct ext2_inode_info *ei)
1390 {
1391         unsigned int flags = ei->vfs_inode.i_flags;
1392
1393         ei->i_flags &= ~(EXT2_SYNC_FL|EXT2_APPEND_FL|
1394                         EXT2_IMMUTABLE_FL|EXT2_NOATIME_FL|EXT2_DIRSYNC_FL);
1395         if (flags & S_SYNC)
1396                 ei->i_flags |= EXT2_SYNC_FL;
1397         if (flags & S_APPEND)
1398                 ei->i_flags |= EXT2_APPEND_FL;
1399         if (flags & S_IMMUTABLE)
1400                 ei->i_flags |= EXT2_IMMUTABLE_FL;
1401         if (flags & S_NOATIME)
1402                 ei->i_flags |= EXT2_NOATIME_FL;
1403         if (flags & S_DIRSYNC)
1404                 ei->i_flags |= EXT2_DIRSYNC_FL;
1405 }
1406
1407 struct inode *ext2_iget (struct super_block *sb, unsigned long ino)
1408 {
1409         struct ext2_inode_info *ei;
1410         struct buffer_head * bh;
1411         struct ext2_inode *raw_inode;
1412         struct inode *inode;
1413         long ret = -EIO;
1414         int n;
1415         uid_t i_uid;
1416         gid_t i_gid;
1417
1418         inode = iget_locked(sb, ino);
1419         if (!inode)
1420                 return ERR_PTR(-ENOMEM);
1421         if (!(inode->i_state & I_NEW))
1422                 return inode;
1423
1424         ei = EXT2_I(inode);
1425         ei->i_block_alloc_info = NULL;
1426
1427         raw_inode = ext2_get_inode(inode->i_sb, ino, &bh);
1428         if (IS_ERR(raw_inode)) {
1429                 ret = PTR_ERR(raw_inode);
1430                 goto bad_inode;
1431         }
1432
1433         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
1434         i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
1435         i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
1436         if (!(test_opt (inode->i_sb, NO_UID32))) {
1437                 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
1438                 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
1439         }
1440         i_uid_write(inode, i_uid);
1441         i_gid_write(inode, i_gid);
1442         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
1443         inode->i_size = le32_to_cpu(raw_inode->i_size);
1444         inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime);
1445         inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime);
1446         inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime);
1447         inode->i_atime.tv_nsec = inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1448         ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
1449         /* We now have enough fields to check if the inode was active or not.
1450          * This is needed because nfsd might try to access dead inodes
1451          * the test is that same one that e2fsck uses
1452          * NeilBrown 1999oct15
1453          */
1454         if (inode->i_nlink == 0 && (inode->i_mode == 0 || ei->i_dtime)) {
1455                 /* this inode is deleted */
1456                 brelse (bh);
1457                 ret = -ESTALE;
1458                 goto bad_inode;
1459         }
1460         inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
1461         ei->i_flags = le32_to_cpu(raw_inode->i_flags);
1462         ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
1463         ei->i_frag_no = raw_inode->i_frag;
1464         ei->i_frag_size = raw_inode->i_fsize;
1465         ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
1466         ei->i_dir_acl = 0;
1467
1468         if (ei->i_file_acl &&
1469             !ext2_data_block_valid(EXT2_SB(sb), ei->i_file_acl, 1)) {
1470                 ext2_error(sb, "ext2_iget", "bad extended attribute block %u",
1471                            ei->i_file_acl);
1472                 brelse(bh);
1473                 ret = -EFSCORRUPTED;
1474                 goto bad_inode;
1475         }
1476
1477         if (S_ISREG(inode->i_mode))
1478                 inode->i_size |= ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
1479         else
1480                 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
1481         if (i_size_read(inode) < 0) {
1482                 ret = -EFSCORRUPTED;
1483                 goto bad_inode;
1484         }
1485         ei->i_dtime = 0;
1486         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
1487         ei->i_state = 0;
1488         ei->i_block_group = (ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
1489         ei->i_dir_start_lookup = 0;
1490
1491         /*
1492          * NOTE! The in-memory inode i_data array is in little-endian order
1493          * even on big-endian machines: we do NOT byteswap the block numbers!
1494          */
1495         for (n = 0; n < EXT2_N_BLOCKS; n++)
1496                 ei->i_data[n] = raw_inode->i_block[n];
1497
1498         if (S_ISREG(inode->i_mode)) {
1499                 inode->i_op = &ext2_file_inode_operations;
1500                 if (test_opt(inode->i_sb, NOBH)) {
1501                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1502                         inode->i_fop = &ext2_file_operations;
1503                 } else {
1504                         inode->i_mapping->a_ops = &ext2_aops;
1505                         inode->i_fop = &ext2_file_operations;
1506                 }
1507         } else if (S_ISDIR(inode->i_mode)) {
1508                 inode->i_op = &ext2_dir_inode_operations;
1509                 inode->i_fop = &ext2_dir_operations;
1510                 if (test_opt(inode->i_sb, NOBH))
1511                         inode->i_mapping->a_ops = &ext2_nobh_aops;
1512                 else
1513                         inode->i_mapping->a_ops = &ext2_aops;
1514         } else if (S_ISLNK(inode->i_mode)) {
1515                 if (ext2_inode_is_fast_symlink(inode)) {
1516                         inode->i_link = (char *)ei->i_data;
1517                         inode->i_op = &ext2_fast_symlink_inode_operations;
1518                         nd_terminate_link(ei->i_data, inode->i_size,
1519                                 sizeof(ei->i_data) - 1);
1520                 } else {
1521                         inode->i_op = &ext2_symlink_inode_operations;
1522                         inode_nohighmem(inode);
1523                         if (test_opt(inode->i_sb, NOBH))
1524                                 inode->i_mapping->a_ops = &ext2_nobh_aops;
1525                         else
1526                                 inode->i_mapping->a_ops = &ext2_aops;
1527                 }
1528         } else {
1529                 inode->i_op = &ext2_special_inode_operations;
1530                 if (raw_inode->i_block[0])
1531                         init_special_inode(inode, inode->i_mode,
1532                            old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
1533                 else 
1534                         init_special_inode(inode, inode->i_mode,
1535                            new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
1536         }
1537         brelse (bh);
1538         ext2_set_inode_flags(inode);
1539         unlock_new_inode(inode);
1540         return inode;
1541         
1542 bad_inode:
1543         iget_failed(inode);
1544         return ERR_PTR(ret);
1545 }
1546
1547 static int __ext2_write_inode(struct inode *inode, int do_sync)
1548 {
1549         struct ext2_inode_info *ei = EXT2_I(inode);
1550         struct super_block *sb = inode->i_sb;
1551         ino_t ino = inode->i_ino;
1552         uid_t uid = i_uid_read(inode);
1553         gid_t gid = i_gid_read(inode);
1554         struct buffer_head * bh;
1555         struct ext2_inode * raw_inode = ext2_get_inode(sb, ino, &bh);
1556         int n;
1557         int err = 0;
1558
1559         if (IS_ERR(raw_inode))
1560                 return -EIO;
1561
1562         /* For fields not not tracking in the in-memory inode,
1563          * initialise them to zero for new inodes. */
1564         if (ei->i_state & EXT2_STATE_NEW)
1565                 memset(raw_inode, 0, EXT2_SB(sb)->s_inode_size);
1566
1567         ext2_get_inode_flags(ei);
1568         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
1569         if (!(test_opt(sb, NO_UID32))) {
1570                 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
1571                 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
1572 /*
1573  * Fix up interoperability with old kernels. Otherwise, old inodes get
1574  * re-used with the upper 16 bits of the uid/gid intact
1575  */
1576                 if (!ei->i_dtime) {
1577                         raw_inode->i_uid_high = cpu_to_le16(high_16_bits(uid));
1578                         raw_inode->i_gid_high = cpu_to_le16(high_16_bits(gid));
1579                 } else {
1580                         raw_inode->i_uid_high = 0;
1581                         raw_inode->i_gid_high = 0;
1582                 }
1583         } else {
1584                 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(uid));
1585                 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(gid));
1586                 raw_inode->i_uid_high = 0;
1587                 raw_inode->i_gid_high = 0;
1588         }
1589         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
1590         raw_inode->i_size = cpu_to_le32(inode->i_size);
1591         raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
1592         raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
1593         raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
1594
1595         raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
1596         raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
1597         raw_inode->i_flags = cpu_to_le32(ei->i_flags);
1598         raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
1599         raw_inode->i_frag = ei->i_frag_no;
1600         raw_inode->i_fsize = ei->i_frag_size;
1601         raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
1602         if (!S_ISREG(inode->i_mode))
1603                 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
1604         else {
1605                 raw_inode->i_size_high = cpu_to_le32(inode->i_size >> 32);
1606                 if (inode->i_size > 0x7fffffffULL) {
1607                         if (!EXT2_HAS_RO_COMPAT_FEATURE(sb,
1608                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE) ||
1609                             EXT2_SB(sb)->s_es->s_rev_level ==
1610                                         cpu_to_le32(EXT2_GOOD_OLD_REV)) {
1611                                /* If this is the first large file
1612                                 * created, add a flag to the superblock.
1613                                 */
1614                                 spin_lock(&EXT2_SB(sb)->s_lock);
1615                                 ext2_update_dynamic_rev(sb);
1616                                 EXT2_SET_RO_COMPAT_FEATURE(sb,
1617                                         EXT2_FEATURE_RO_COMPAT_LARGE_FILE);
1618                                 spin_unlock(&EXT2_SB(sb)->s_lock);
1619                                 ext2_write_super(sb);
1620                         }
1621                 }
1622         }
1623         
1624         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
1625         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
1626                 if (old_valid_dev(inode->i_rdev)) {
1627                         raw_inode->i_block[0] =
1628                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
1629                         raw_inode->i_block[1] = 0;
1630                 } else {
1631                         raw_inode->i_block[0] = 0;
1632                         raw_inode->i_block[1] =
1633                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
1634                         raw_inode->i_block[2] = 0;
1635                 }
1636         } else for (n = 0; n < EXT2_N_BLOCKS; n++)
1637                 raw_inode->i_block[n] = ei->i_data[n];
1638         mark_buffer_dirty(bh);
1639         if (do_sync) {
1640                 sync_dirty_buffer(bh);
1641                 if (buffer_req(bh) && !buffer_uptodate(bh)) {
1642                         printk ("IO error syncing ext2 inode [%s:%08lx]\n",
1643                                 sb->s_id, (unsigned long) ino);
1644                         err = -EIO;
1645                 }
1646         }
1647         ei->i_state &= ~EXT2_STATE_NEW;
1648         brelse (bh);
1649         return err;
1650 }
1651
1652 int ext2_write_inode(struct inode *inode, struct writeback_control *wbc)
1653 {
1654         return __ext2_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1655 }
1656
1657 int ext2_setattr(struct dentry *dentry, struct iattr *iattr)
1658 {
1659         struct inode *inode = d_inode(dentry);
1660         int error;
1661
1662         error = setattr_prepare(dentry, iattr);
1663         if (error)
1664                 return error;
1665
1666         if (is_quota_modification(inode, iattr)) {
1667                 error = dquot_initialize(inode);
1668                 if (error)
1669                         return error;
1670         }
1671         if ((iattr->ia_valid & ATTR_UID && !uid_eq(iattr->ia_uid, inode->i_uid)) ||
1672             (iattr->ia_valid & ATTR_GID && !gid_eq(iattr->ia_gid, inode->i_gid))) {
1673                 error = dquot_transfer(inode, iattr);
1674                 if (error)
1675                         return error;
1676         }
1677         if (iattr->ia_valid & ATTR_SIZE && iattr->ia_size != inode->i_size) {
1678                 error = ext2_setsize(inode, iattr->ia_size);
1679                 if (error)
1680                         return error;
1681         }
1682         setattr_copy(inode, iattr);
1683         if (iattr->ia_valid & ATTR_MODE)
1684                 error = posix_acl_chmod(inode, inode->i_mode);
1685         mark_inode_dirty(inode);
1686
1687         return error;
1688 }