writeback: remove nonblocking/encountered_congestion references
[linux-2.6.git] / fs / reiserfs / inode.c
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4
5 #include <linux/time.h>
6 #include <linux/fs.h>
7 #include <linux/reiserfs_fs.h>
8 #include <linux/reiserfs_acl.h>
9 #include <linux/reiserfs_xattr.h>
10 #include <linux/exportfs.h>
11 #include <linux/smp_lock.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/slab.h>
15 #include <asm/uaccess.h>
16 #include <asm/unaligned.h>
17 #include <linux/buffer_head.h>
18 #include <linux/mpage.h>
19 #include <linux/writeback.h>
20 #include <linux/quotaops.h>
21 #include <linux/swap.h>
22
23 int reiserfs_commit_write(struct file *f, struct page *page,
24                           unsigned from, unsigned to);
25 int reiserfs_prepare_write(struct file *f, struct page *page,
26                            unsigned from, unsigned to);
27
28 void reiserfs_evict_inode(struct inode *inode)
29 {
30         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
31         int jbegin_count =
32             JOURNAL_PER_BALANCE_CNT * 2 +
33             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
34         struct reiserfs_transaction_handle th;
35         int depth;
36         int err;
37
38         if (!inode->i_nlink && !is_bad_inode(inode))
39                 dquot_initialize(inode);
40
41         truncate_inode_pages(&inode->i_data, 0);
42         if (inode->i_nlink)
43                 goto no_delete;
44
45         depth = reiserfs_write_lock_once(inode->i_sb);
46
47         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
48         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
49                 reiserfs_delete_xattrs(inode);
50
51                 if (journal_begin(&th, inode->i_sb, jbegin_count))
52                         goto out;
53                 reiserfs_update_inode_transaction(inode);
54
55                 reiserfs_discard_prealloc(&th, inode);
56
57                 err = reiserfs_delete_object(&th, inode);
58
59                 /* Do quota update inside a transaction for journaled quotas. We must do that
60                  * after delete_object so that quota updates go into the same transaction as
61                  * stat data deletion */
62                 if (!err) 
63                         dquot_free_inode(inode);
64
65                 if (journal_end(&th, inode->i_sb, jbegin_count))
66                         goto out;
67
68                 /* check return value from reiserfs_delete_object after
69                  * ending the transaction
70                  */
71                 if (err)
72                     goto out;
73
74                 /* all items of file are deleted, so we can remove "save" link */
75                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
76                                                                  * about an error here */
77         } else {
78                 /* no object items are in the tree */
79                 ;
80         }
81       out:
82         end_writeback(inode);   /* note this must go after the journal_end to prevent deadlock */
83         dquot_drop(inode);
84         inode->i_blocks = 0;
85         reiserfs_write_unlock_once(inode->i_sb, depth);
86         return;
87
88 no_delete:
89         end_writeback(inode);
90         dquot_drop(inode);
91 }
92
93 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
94                           __u32 objectid, loff_t offset, int type, int length)
95 {
96         key->version = version;
97
98         key->on_disk_key.k_dir_id = dirid;
99         key->on_disk_key.k_objectid = objectid;
100         set_cpu_key_k_offset(key, offset);
101         set_cpu_key_k_type(key, type);
102         key->key_length = length;
103 }
104
105 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
106    offset and type of key */
107 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
108                   int type, int length)
109 {
110         _make_cpu_key(key, get_inode_item_key_version(inode),
111                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
112                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
113                       length);
114 }
115
116 //
117 // when key is 0, do not set version and short key
118 //
119 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
120                               int version,
121                               loff_t offset, int type, int length,
122                               int entry_count /*or ih_free_space */ )
123 {
124         if (key) {
125                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
126                 ih->ih_key.k_objectid =
127                     cpu_to_le32(key->on_disk_key.k_objectid);
128         }
129         put_ih_version(ih, version);
130         set_le_ih_k_offset(ih, offset);
131         set_le_ih_k_type(ih, type);
132         put_ih_item_len(ih, length);
133         /*    set_ih_free_space (ih, 0); */
134         // for directory items it is entry count, for directs and stat
135         // datas - 0xffff, for indirects - 0
136         put_ih_entry_count(ih, entry_count);
137 }
138
139 //
140 // FIXME: we might cache recently accessed indirect item
141
142 // Ugh.  Not too eager for that....
143 //  I cut the code until such time as I see a convincing argument (benchmark).
144 // I don't want a bloated inode struct..., and I don't like code complexity....
145
146 /* cutting the code is fine, since it really isn't in use yet and is easy
147 ** to add back in.  But, Vladimir has a really good idea here.  Think
148 ** about what happens for reading a file.  For each page,
149 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
150 ** an indirect item.  This indirect item has X number of pointers, where
151 ** X is a big number if we've done the block allocation right.  But,
152 ** we only use one or two of these pointers during each call to readpage,
153 ** needlessly researching again later on.
154 **
155 ** The size of the cache could be dynamic based on the size of the file.
156 **
157 ** I'd also like to see us cache the location the stat data item, since
158 ** we are needlessly researching for that frequently.
159 **
160 ** --chris
161 */
162
163 /* If this page has a file tail in it, and
164 ** it was read in by get_block_create_0, the page data is valid,
165 ** but tail is still sitting in a direct item, and we can't write to
166 ** it.  So, look through this page, and check all the mapped buffers
167 ** to make sure they have valid block numbers.  Any that don't need
168 ** to be unmapped, so that block_prepare_write will correctly call
169 ** reiserfs_get_block to convert the tail into an unformatted node
170 */
171 static inline void fix_tail_page_for_writing(struct page *page)
172 {
173         struct buffer_head *head, *next, *bh;
174
175         if (page && page_has_buffers(page)) {
176                 head = page_buffers(page);
177                 bh = head;
178                 do {
179                         next = bh->b_this_page;
180                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
181                                 reiserfs_unmap_buffer(bh);
182                         }
183                         bh = next;
184                 } while (bh != head);
185         }
186 }
187
188 /* reiserfs_get_block does not need to allocate a block only if it has been
189    done already or non-hole position has been found in the indirect item */
190 static inline int allocation_needed(int retval, b_blocknr_t allocated,
191                                     struct item_head *ih,
192                                     __le32 * item, int pos_in_item)
193 {
194         if (allocated)
195                 return 0;
196         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
197             get_block_num(item, pos_in_item))
198                 return 0;
199         return 1;
200 }
201
202 static inline int indirect_item_found(int retval, struct item_head *ih)
203 {
204         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
205 }
206
207 static inline void set_block_dev_mapped(struct buffer_head *bh,
208                                         b_blocknr_t block, struct inode *inode)
209 {
210         map_bh(bh, inode->i_sb, block);
211 }
212
213 //
214 // files which were created in the earlier version can not be longer,
215 // than 2 gb
216 //
217 static int file_capable(struct inode *inode, sector_t block)
218 {
219         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
220             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
221                 return 1;
222
223         return 0;
224 }
225
226 static int restart_transaction(struct reiserfs_transaction_handle *th,
227                                struct inode *inode, struct treepath *path)
228 {
229         struct super_block *s = th->t_super;
230         int len = th->t_blocks_allocated;
231         int err;
232
233         BUG_ON(!th->t_trans_id);
234         BUG_ON(!th->t_refcount);
235
236         pathrelse(path);
237
238         /* we cannot restart while nested */
239         if (th->t_refcount > 1) {
240                 return 0;
241         }
242         reiserfs_update_sd(th, inode);
243         err = journal_end(th, s, len);
244         if (!err) {
245                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
246                 if (!err)
247                         reiserfs_update_inode_transaction(inode);
248         }
249         return err;
250 }
251
252 // it is called by get_block when create == 0. Returns block number
253 // for 'block'-th logical block of file. When it hits direct item it
254 // returns 0 (being called from bmap) or read direct item into piece
255 // of page (bh_result)
256
257 // Please improve the english/clarity in the comment above, as it is
258 // hard to understand.
259
260 static int _get_block_create_0(struct inode *inode, sector_t block,
261                                struct buffer_head *bh_result, int args)
262 {
263         INITIALIZE_PATH(path);
264         struct cpu_key key;
265         struct buffer_head *bh;
266         struct item_head *ih, tmp_ih;
267         b_blocknr_t blocknr;
268         char *p = NULL;
269         int chars;
270         int ret;
271         int result;
272         int done = 0;
273         unsigned long offset;
274
275         // prepare the key to look for the 'block'-th block of file
276         make_cpu_key(&key, inode,
277                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
278                      3);
279
280         result = search_for_position_by_key(inode->i_sb, &key, &path);
281         if (result != POSITION_FOUND) {
282                 pathrelse(&path);
283                 if (p)
284                         kunmap(bh_result->b_page);
285                 if (result == IO_ERROR)
286                         return -EIO;
287                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
288                 // That there is some MMAPED data associated with it that is yet to be written to disk.
289                 if ((args & GET_BLOCK_NO_HOLE)
290                     && !PageUptodate(bh_result->b_page)) {
291                         return -ENOENT;
292                 }
293                 return 0;
294         }
295         //
296         bh = get_last_bh(&path);
297         ih = get_ih(&path);
298         if (is_indirect_le_ih(ih)) {
299                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
300
301                 /* FIXME: here we could cache indirect item or part of it in
302                    the inode to avoid search_by_key in case of subsequent
303                    access to file */
304                 blocknr = get_block_num(ind_item, path.pos_in_item);
305                 ret = 0;
306                 if (blocknr) {
307                         map_bh(bh_result, inode->i_sb, blocknr);
308                         if (path.pos_in_item ==
309                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
310                                 set_buffer_boundary(bh_result);
311                         }
312                 } else
313                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
314                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
315                 if ((args & GET_BLOCK_NO_HOLE)
316                             && !PageUptodate(bh_result->b_page)) {
317                         ret = -ENOENT;
318                 }
319
320                 pathrelse(&path);
321                 if (p)
322                         kunmap(bh_result->b_page);
323                 return ret;
324         }
325         // requested data are in direct item(s)
326         if (!(args & GET_BLOCK_READ_DIRECT)) {
327                 // we are called by bmap. FIXME: we can not map block of file
328                 // when it is stored in direct item(s)
329                 pathrelse(&path);
330                 if (p)
331                         kunmap(bh_result->b_page);
332                 return -ENOENT;
333         }
334
335         /* if we've got a direct item, and the buffer or page was uptodate,
336          ** we don't want to pull data off disk again.  skip to the
337          ** end, where we map the buffer and return
338          */
339         if (buffer_uptodate(bh_result)) {
340                 goto finished;
341         } else
342                 /*
343                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
344                  ** pages without any buffers.  If the page is up to date, we don't want
345                  ** read old data off disk.  Set the up to date bit on the buffer instead
346                  ** and jump to the end
347                  */
348         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
349                 set_buffer_uptodate(bh_result);
350                 goto finished;
351         }
352         // read file tail into part of page
353         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
354         copy_item_head(&tmp_ih, ih);
355
356         /* we only want to kmap if we are reading the tail into the page.
357          ** this is not the common case, so we don't kmap until we are
358          ** sure we need to.  But, this means the item might move if
359          ** kmap schedules
360          */
361         if (!p)
362                 p = (char *)kmap(bh_result->b_page);
363
364         p += offset;
365         memset(p, 0, inode->i_sb->s_blocksize);
366         do {
367                 if (!is_direct_le_ih(ih)) {
368                         BUG();
369                 }
370                 /* make sure we don't read more bytes than actually exist in
371                  ** the file.  This can happen in odd cases where i_size isn't
372                  ** correct, and when direct item padding results in a few
373                  ** extra bytes at the end of the direct item
374                  */
375                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
376                         break;
377                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
378                         chars =
379                             inode->i_size - (le_ih_k_offset(ih) - 1) -
380                             path.pos_in_item;
381                         done = 1;
382                 } else {
383                         chars = ih_item_len(ih) - path.pos_in_item;
384                 }
385                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
386
387                 if (done)
388                         break;
389
390                 p += chars;
391
392                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
393                         // we done, if read direct item is not the last item of
394                         // node FIXME: we could try to check right delimiting key
395                         // to see whether direct item continues in the right
396                         // neighbor or rely on i_size
397                         break;
398
399                 // update key to look for the next piece
400                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
401                 result = search_for_position_by_key(inode->i_sb, &key, &path);
402                 if (result != POSITION_FOUND)
403                         // i/o error most likely
404                         break;
405                 bh = get_last_bh(&path);
406                 ih = get_ih(&path);
407         } while (1);
408
409         flush_dcache_page(bh_result->b_page);
410         kunmap(bh_result->b_page);
411
412       finished:
413         pathrelse(&path);
414
415         if (result == IO_ERROR)
416                 return -EIO;
417
418         /* this buffer has valid data, but isn't valid for io.  mapping it to
419          * block #0 tells the rest of reiserfs it just has a tail in it
420          */
421         map_bh(bh_result, inode->i_sb, 0);
422         set_buffer_uptodate(bh_result);
423         return 0;
424 }
425
426 // this is called to create file map. So, _get_block_create_0 will not
427 // read direct item
428 static int reiserfs_bmap(struct inode *inode, sector_t block,
429                          struct buffer_head *bh_result, int create)
430 {
431         if (!file_capable(inode, block))
432                 return -EFBIG;
433
434         reiserfs_write_lock(inode->i_sb);
435         /* do not read the direct item */
436         _get_block_create_0(inode, block, bh_result, 0);
437         reiserfs_write_unlock(inode->i_sb);
438         return 0;
439 }
440
441 /* special version of get_block that is only used by grab_tail_page right
442 ** now.  It is sent to block_prepare_write, and when you try to get a
443 ** block past the end of the file (or a block from a hole) it returns
444 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
445 ** be able to do i/o on the buffers returned, unless an error value
446 ** is also returned.
447 **
448 ** So, this allows block_prepare_write to be used for reading a single block
449 ** in a page.  Where it does not produce a valid page for holes, or past the
450 ** end of the file.  This turns out to be exactly what we need for reading
451 ** tails for conversion.
452 **
453 ** The point of the wrapper is forcing a certain value for create, even
454 ** though the VFS layer is calling this function with create==1.  If you
455 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
456 ** don't use this function.
457 */
458 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
459                                        struct buffer_head *bh_result,
460                                        int create)
461 {
462         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
463 }
464
465 /* This is special helper for reiserfs_get_block in case we are executing
466    direct_IO request. */
467 static int reiserfs_get_blocks_direct_io(struct inode *inode,
468                                          sector_t iblock,
469                                          struct buffer_head *bh_result,
470                                          int create)
471 {
472         int ret;
473
474         bh_result->b_page = NULL;
475
476         /* We set the b_size before reiserfs_get_block call since it is
477            referenced in convert_tail_for_hole() that may be called from
478            reiserfs_get_block() */
479         bh_result->b_size = (1 << inode->i_blkbits);
480
481         ret = reiserfs_get_block(inode, iblock, bh_result,
482                                  create | GET_BLOCK_NO_DANGLE);
483         if (ret)
484                 goto out;
485
486         /* don't allow direct io onto tail pages */
487         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
488                 /* make sure future calls to the direct io funcs for this offset
489                  ** in the file fail by unmapping the buffer
490                  */
491                 clear_buffer_mapped(bh_result);
492                 ret = -EINVAL;
493         }
494         /* Possible unpacked tail. Flush the data before pages have
495            disappeared */
496         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
497                 int err;
498
499                 reiserfs_write_lock(inode->i_sb);
500
501                 err = reiserfs_commit_for_inode(inode);
502                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
503
504                 reiserfs_write_unlock(inode->i_sb);
505
506                 if (err < 0)
507                         ret = err;
508         }
509       out:
510         return ret;
511 }
512
513 /*
514 ** helper function for when reiserfs_get_block is called for a hole
515 ** but the file tail is still in a direct item
516 ** bh_result is the buffer head for the hole
517 ** tail_offset is the offset of the start of the tail in the file
518 **
519 ** This calls prepare_write, which will start a new transaction
520 ** you should not be in a transaction, or have any paths held when you
521 ** call this.
522 */
523 static int convert_tail_for_hole(struct inode *inode,
524                                  struct buffer_head *bh_result,
525                                  loff_t tail_offset)
526 {
527         unsigned long index;
528         unsigned long tail_end;
529         unsigned long tail_start;
530         struct page *tail_page;
531         struct page *hole_page = bh_result->b_page;
532         int retval = 0;
533
534         if ((tail_offset & (bh_result->b_size - 1)) != 1)
535                 return -EIO;
536
537         /* always try to read until the end of the block */
538         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
539         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
540
541         index = tail_offset >> PAGE_CACHE_SHIFT;
542         /* hole_page can be zero in case of direct_io, we are sure
543            that we cannot get here if we write with O_DIRECT into
544            tail page */
545         if (!hole_page || index != hole_page->index) {
546                 tail_page = grab_cache_page(inode->i_mapping, index);
547                 retval = -ENOMEM;
548                 if (!tail_page) {
549                         goto out;
550                 }
551         } else {
552                 tail_page = hole_page;
553         }
554
555         /* we don't have to make sure the conversion did not happen while
556          ** we were locking the page because anyone that could convert
557          ** must first take i_mutex.
558          **
559          ** We must fix the tail page for writing because it might have buffers
560          ** that are mapped, but have a block number of 0.  This indicates tail
561          ** data that has been read directly into the page, and block_prepare_write
562          ** won't trigger a get_block in this case.
563          */
564         fix_tail_page_for_writing(tail_page);
565         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
566         if (retval)
567                 goto unlock;
568
569         /* tail conversion might change the data in the page */
570         flush_dcache_page(tail_page);
571
572         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
573
574       unlock:
575         if (tail_page != hole_page) {
576                 unlock_page(tail_page);
577                 page_cache_release(tail_page);
578         }
579       out:
580         return retval;
581 }
582
583 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
584                                   sector_t block,
585                                   struct inode *inode,
586                                   b_blocknr_t * allocated_block_nr,
587                                   struct treepath *path, int flags)
588 {
589         BUG_ON(!th->t_trans_id);
590
591 #ifdef REISERFS_PREALLOCATE
592         if (!(flags & GET_BLOCK_NO_IMUX)) {
593                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
594                                                   path, block);
595         }
596 #endif
597         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
598                                          block);
599 }
600
601 int reiserfs_get_block(struct inode *inode, sector_t block,
602                        struct buffer_head *bh_result, int create)
603 {
604         int repeat, retval = 0;
605         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
606         INITIALIZE_PATH(path);
607         int pos_in_item;
608         struct cpu_key key;
609         struct buffer_head *bh, *unbh = NULL;
610         struct item_head *ih, tmp_ih;
611         __le32 *item;
612         int done;
613         int fs_gen;
614         int lock_depth;
615         struct reiserfs_transaction_handle *th = NULL;
616         /* space reserved in transaction batch:
617            . 3 balancings in direct->indirect conversion
618            . 1 block involved into reiserfs_update_sd()
619            XXX in practically impossible worst case direct2indirect()
620            can incur (much) more than 3 balancings.
621            quota update for user, group */
622         int jbegin_count =
623             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
624             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
625         int version;
626         int dangle = 1;
627         loff_t new_offset =
628             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
629
630         lock_depth = reiserfs_write_lock_once(inode->i_sb);
631         version = get_inode_item_key_version(inode);
632
633         if (!file_capable(inode, block)) {
634                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
635                 return -EFBIG;
636         }
637
638         /* if !create, we aren't changing the FS, so we don't need to
639          ** log anything, so we don't need to start a transaction
640          */
641         if (!(create & GET_BLOCK_CREATE)) {
642                 int ret;
643                 /* find number of block-th logical block of the file */
644                 ret = _get_block_create_0(inode, block, bh_result,
645                                           create | GET_BLOCK_READ_DIRECT);
646                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
647                 return ret;
648         }
649         /*
650          * if we're already in a transaction, make sure to close
651          * any new transactions we start in this func
652          */
653         if ((create & GET_BLOCK_NO_DANGLE) ||
654             reiserfs_transaction_running(inode->i_sb))
655                 dangle = 0;
656
657         /* If file is of such a size, that it might have a tail and tails are enabled
658          ** we should mark it as possibly needing tail packing on close
659          */
660         if ((have_large_tails(inode->i_sb)
661              && inode->i_size < i_block_size(inode) * 4)
662             || (have_small_tails(inode->i_sb)
663                 && inode->i_size < i_block_size(inode)))
664                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
665
666         /* set the key of the first byte in the 'block'-th block of file */
667         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
668         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
669               start_trans:
670                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
671                 if (!th) {
672                         retval = -ENOMEM;
673                         goto failure;
674                 }
675                 reiserfs_update_inode_transaction(inode);
676         }
677       research:
678
679         retval = search_for_position_by_key(inode->i_sb, &key, &path);
680         if (retval == IO_ERROR) {
681                 retval = -EIO;
682                 goto failure;
683         }
684
685         bh = get_last_bh(&path);
686         ih = get_ih(&path);
687         item = get_item(&path);
688         pos_in_item = path.pos_in_item;
689
690         fs_gen = get_generation(inode->i_sb);
691         copy_item_head(&tmp_ih, ih);
692
693         if (allocation_needed
694             (retval, allocated_block_nr, ih, item, pos_in_item)) {
695                 /* we have to allocate block for the unformatted node */
696                 if (!th) {
697                         pathrelse(&path);
698                         goto start_trans;
699                 }
700
701                 repeat =
702                     _allocate_block(th, block, inode, &allocated_block_nr,
703                                     &path, create);
704
705                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
706                         /* restart the transaction to give the journal a chance to free
707                          ** some blocks.  releases the path, so we have to go back to
708                          ** research if we succeed on the second try
709                          */
710                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
711                         retval = restart_transaction(th, inode, &path);
712                         if (retval)
713                                 goto failure;
714                         repeat =
715                             _allocate_block(th, block, inode,
716                                             &allocated_block_nr, NULL, create);
717
718                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
719                                 goto research;
720                         }
721                         if (repeat == QUOTA_EXCEEDED)
722                                 retval = -EDQUOT;
723                         else
724                                 retval = -ENOSPC;
725                         goto failure;
726                 }
727
728                 if (fs_changed(fs_gen, inode->i_sb)
729                     && item_moved(&tmp_ih, &path)) {
730                         goto research;
731                 }
732         }
733
734         if (indirect_item_found(retval, ih)) {
735                 b_blocknr_t unfm_ptr;
736                 /* 'block'-th block is in the file already (there is
737                    corresponding cell in some indirect item). But it may be
738                    zero unformatted node pointer (hole) */
739                 unfm_ptr = get_block_num(item, pos_in_item);
740                 if (unfm_ptr == 0) {
741                         /* use allocated block to plug the hole */
742                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
743                         if (fs_changed(fs_gen, inode->i_sb)
744                             && item_moved(&tmp_ih, &path)) {
745                                 reiserfs_restore_prepared_buffer(inode->i_sb,
746                                                                  bh);
747                                 goto research;
748                         }
749                         set_buffer_new(bh_result);
750                         if (buffer_dirty(bh_result)
751                             && reiserfs_data_ordered(inode->i_sb))
752                                 reiserfs_add_ordered_list(inode, bh_result);
753                         put_block_num(item, pos_in_item, allocated_block_nr);
754                         unfm_ptr = allocated_block_nr;
755                         journal_mark_dirty(th, inode->i_sb, bh);
756                         reiserfs_update_sd(th, inode);
757                 }
758                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
759                 pathrelse(&path);
760                 retval = 0;
761                 if (!dangle && th)
762                         retval = reiserfs_end_persistent_transaction(th);
763
764                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
765
766                 /* the item was found, so new blocks were not added to the file
767                  ** there is no need to make sure the inode is updated with this
768                  ** transaction
769                  */
770                 return retval;
771         }
772
773         if (!th) {
774                 pathrelse(&path);
775                 goto start_trans;
776         }
777
778         /* desired position is not found or is in the direct item. We have
779            to append file with holes up to 'block'-th block converting
780            direct items to indirect one if necessary */
781         done = 0;
782         do {
783                 if (is_statdata_le_ih(ih)) {
784                         __le32 unp = 0;
785                         struct cpu_key tmp_key;
786
787                         /* indirect item has to be inserted */
788                         make_le_item_head(&tmp_ih, &key, version, 1,
789                                           TYPE_INDIRECT, UNFM_P_SIZE,
790                                           0 /* free_space */ );
791
792                         if (cpu_key_k_offset(&key) == 1) {
793                                 /* we are going to add 'block'-th block to the file. Use
794                                    allocated block for that */
795                                 unp = cpu_to_le32(allocated_block_nr);
796                                 set_block_dev_mapped(bh_result,
797                                                      allocated_block_nr, inode);
798                                 set_buffer_new(bh_result);
799                                 done = 1;
800                         }
801                         tmp_key = key;  // ;)
802                         set_cpu_key_k_offset(&tmp_key, 1);
803                         PATH_LAST_POSITION(&path)++;
804
805                         retval =
806                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
807                                                  inode, (char *)&unp);
808                         if (retval) {
809                                 reiserfs_free_block(th, inode,
810                                                     allocated_block_nr, 1);
811                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
812                         }
813                         //mark_tail_converted (inode);
814                 } else if (is_direct_le_ih(ih)) {
815                         /* direct item has to be converted */
816                         loff_t tail_offset;
817
818                         tail_offset =
819                             ((le_ih_k_offset(ih) -
820                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
821                         if (tail_offset == cpu_key_k_offset(&key)) {
822                                 /* direct item we just found fits into block we have
823                                    to map. Convert it into unformatted node: use
824                                    bh_result for the conversion */
825                                 set_block_dev_mapped(bh_result,
826                                                      allocated_block_nr, inode);
827                                 unbh = bh_result;
828                                 done = 1;
829                         } else {
830                                 /* we have to padd file tail stored in direct item(s)
831                                    up to block size and convert it to unformatted
832                                    node. FIXME: this should also get into page cache */
833
834                                 pathrelse(&path);
835                                 /*
836                                  * ugly, but we can only end the transaction if
837                                  * we aren't nested
838                                  */
839                                 BUG_ON(!th->t_refcount);
840                                 if (th->t_refcount == 1) {
841                                         retval =
842                                             reiserfs_end_persistent_transaction
843                                             (th);
844                                         th = NULL;
845                                         if (retval)
846                                                 goto failure;
847                                 }
848
849                                 retval =
850                                     convert_tail_for_hole(inode, bh_result,
851                                                           tail_offset);
852                                 if (retval) {
853                                         if (retval != -ENOSPC)
854                                                 reiserfs_error(inode->i_sb,
855                                                         "clm-6004",
856                                                         "convert tail failed "
857                                                         "inode %lu, error %d",
858                                                         inode->i_ino,
859                                                         retval);
860                                         if (allocated_block_nr) {
861                                                 /* the bitmap, the super, and the stat data == 3 */
862                                                 if (!th)
863                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
864                                                 if (th)
865                                                         reiserfs_free_block(th,
866                                                                             inode,
867                                                                             allocated_block_nr,
868                                                                             1);
869                                         }
870                                         goto failure;
871                                 }
872                                 goto research;
873                         }
874                         retval =
875                             direct2indirect(th, inode, &path, unbh,
876                                             tail_offset);
877                         if (retval) {
878                                 reiserfs_unmap_buffer(unbh);
879                                 reiserfs_free_block(th, inode,
880                                                     allocated_block_nr, 1);
881                                 goto failure;
882                         }
883                         /* it is important the set_buffer_uptodate is done after
884                          ** the direct2indirect.  The buffer might contain valid
885                          ** data newer than the data on disk (read by readpage, changed,
886                          ** and then sent here by writepage).  direct2indirect needs
887                          ** to know if unbh was already up to date, so it can decide
888                          ** if the data in unbh needs to be replaced with data from
889                          ** the disk
890                          */
891                         set_buffer_uptodate(unbh);
892
893                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
894                            buffer will disappear shortly, so it should not be added to
895                          */
896                         if (unbh->b_page) {
897                                 /* we've converted the tail, so we must
898                                  ** flush unbh before the transaction commits
899                                  */
900                                 reiserfs_add_tail_list(inode, unbh);
901
902                                 /* mark it dirty now to prevent commit_write from adding
903                                  ** this buffer to the inode's dirty buffer list
904                                  */
905                                 /*
906                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
907                                  * It's still atomic, but it sets the page dirty too,
908                                  * which makes it eligible for writeback at any time by the
909                                  * VM (which was also the case with __mark_buffer_dirty())
910                                  */
911                                 mark_buffer_dirty(unbh);
912                         }
913                 } else {
914                         /* append indirect item with holes if needed, when appending
915                            pointer to 'block'-th block use block, which is already
916                            allocated */
917                         struct cpu_key tmp_key;
918                         unp_t unf_single = 0;   // We use this in case we need to allocate only
919                         // one block which is a fastpath
920                         unp_t *un;
921                         __u64 max_to_insert =
922                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
923                             UNFM_P_SIZE;
924                         __u64 blocks_needed;
925
926                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
927                                "vs-804: invalid position for append");
928                         /* indirect item has to be appended, set up key of that position */
929                         make_cpu_key(&tmp_key, inode,
930                                      le_key_k_offset(version,
931                                                      &(ih->ih_key)) +
932                                      op_bytes_number(ih,
933                                                      inode->i_sb->s_blocksize),
934                                      //pos_in_item * inode->i_sb->s_blocksize,
935                                      TYPE_INDIRECT, 3); // key type is unimportant
936
937                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
938                                "green-805: invalid offset");
939                         blocks_needed =
940                             1 +
941                             ((cpu_key_k_offset(&key) -
942                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
943                              s_blocksize_bits);
944
945                         if (blocks_needed == 1) {
946                                 un = &unf_single;
947                         } else {
948                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
949                                 if (!un) {
950                                         un = &unf_single;
951                                         blocks_needed = 1;
952                                         max_to_insert = 0;
953                                 }
954                         }
955                         if (blocks_needed <= max_to_insert) {
956                                 /* we are going to add target block to the file. Use allocated
957                                    block for that */
958                                 un[blocks_needed - 1] =
959                                     cpu_to_le32(allocated_block_nr);
960                                 set_block_dev_mapped(bh_result,
961                                                      allocated_block_nr, inode);
962                                 set_buffer_new(bh_result);
963                                 done = 1;
964                         } else {
965                                 /* paste hole to the indirect item */
966                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
967                                    only have space for one block */
968                                 blocks_needed =
969                                     max_to_insert ? max_to_insert : 1;
970                         }
971                         retval =
972                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
973                                                      (char *)un,
974                                                      UNFM_P_SIZE *
975                                                      blocks_needed);
976
977                         if (blocks_needed != 1)
978                                 kfree(un);
979
980                         if (retval) {
981                                 reiserfs_free_block(th, inode,
982                                                     allocated_block_nr, 1);
983                                 goto failure;
984                         }
985                         if (!done) {
986                                 /* We need to mark new file size in case this function will be
987                                    interrupted/aborted later on. And we may do this only for
988                                    holes. */
989                                 inode->i_size +=
990                                     inode->i_sb->s_blocksize * blocks_needed;
991                         }
992                 }
993
994                 if (done == 1)
995                         break;
996
997                 /* this loop could log more blocks than we had originally asked
998                  ** for.  So, we have to allow the transaction to end if it is
999                  ** too big or too full.  Update the inode so things are
1000                  ** consistent if we crash before the function returns
1001                  **
1002                  ** release the path so that anybody waiting on the path before
1003                  ** ending their transaction will be able to continue.
1004                  */
1005                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1006                         retval = restart_transaction(th, inode, &path);
1007                         if (retval)
1008                                 goto failure;
1009                 }
1010                 /*
1011                  * inserting indirect pointers for a hole can take a
1012                  * long time.  reschedule if needed and also release the write
1013                  * lock for others.
1014                  */
1015                 if (need_resched()) {
1016                         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1017                         schedule();
1018                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
1019                 }
1020
1021                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1022                 if (retval == IO_ERROR) {
1023                         retval = -EIO;
1024                         goto failure;
1025                 }
1026                 if (retval == POSITION_FOUND) {
1027                         reiserfs_warning(inode->i_sb, "vs-825",
1028                                          "%K should not be found", &key);
1029                         retval = -EEXIST;
1030                         if (allocated_block_nr)
1031                                 reiserfs_free_block(th, inode,
1032                                                     allocated_block_nr, 1);
1033                         pathrelse(&path);
1034                         goto failure;
1035                 }
1036                 bh = get_last_bh(&path);
1037                 ih = get_ih(&path);
1038                 item = get_item(&path);
1039                 pos_in_item = path.pos_in_item;
1040         } while (1);
1041
1042         retval = 0;
1043
1044       failure:
1045         if (th && (!dangle || (retval && !th->t_trans_id))) {
1046                 int err;
1047                 if (th->t_trans_id)
1048                         reiserfs_update_sd(th, inode);
1049                 err = reiserfs_end_persistent_transaction(th);
1050                 if (err)
1051                         retval = err;
1052         }
1053
1054         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1055         reiserfs_check_path(&path);
1056         return retval;
1057 }
1058
1059 static int
1060 reiserfs_readpages(struct file *file, struct address_space *mapping,
1061                    struct list_head *pages, unsigned nr_pages)
1062 {
1063         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1064 }
1065
1066 /* Compute real number of used bytes by file
1067  * Following three functions can go away when we'll have enough space in stat item
1068  */
1069 static int real_space_diff(struct inode *inode, int sd_size)
1070 {
1071         int bytes;
1072         loff_t blocksize = inode->i_sb->s_blocksize;
1073
1074         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1075                 return sd_size;
1076
1077         /* End of file is also in full block with indirect reference, so round
1078          ** up to the next block.
1079          **
1080          ** there is just no way to know if the tail is actually packed
1081          ** on the file, so we have to assume it isn't.  When we pack the
1082          ** tail, we add 4 bytes to pretend there really is an unformatted
1083          ** node pointer
1084          */
1085         bytes =
1086             ((inode->i_size +
1087               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1088             sd_size;
1089         return bytes;
1090 }
1091
1092 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1093                                         int sd_size)
1094 {
1095         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1096                 return inode->i_size +
1097                     (loff_t) (real_space_diff(inode, sd_size));
1098         }
1099         return ((loff_t) real_space_diff(inode, sd_size)) +
1100             (((loff_t) blocks) << 9);
1101 }
1102
1103 /* Compute number of blocks used by file in ReiserFS counting */
1104 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1105 {
1106         loff_t bytes = inode_get_bytes(inode);
1107         loff_t real_space = real_space_diff(inode, sd_size);
1108
1109         /* keeps fsck and non-quota versions of reiserfs happy */
1110         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1111                 bytes += (loff_t) 511;
1112         }
1113
1114         /* files from before the quota patch might i_blocks such that
1115          ** bytes < real_space.  Deal with that here to prevent it from
1116          ** going negative.
1117          */
1118         if (bytes < real_space)
1119                 return 0;
1120         return (bytes - real_space) >> 9;
1121 }
1122
1123 //
1124 // BAD: new directories have stat data of new type and all other items
1125 // of old type. Version stored in the inode says about body items, so
1126 // in update_stat_data we can not rely on inode, but have to check
1127 // item version directly
1128 //
1129
1130 // called by read_locked_inode
1131 static void init_inode(struct inode *inode, struct treepath *path)
1132 {
1133         struct buffer_head *bh;
1134         struct item_head *ih;
1135         __u32 rdev;
1136         //int version = ITEM_VERSION_1;
1137
1138         bh = PATH_PLAST_BUFFER(path);
1139         ih = PATH_PITEM_HEAD(path);
1140
1141         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1142
1143         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1144         REISERFS_I(inode)->i_flags = 0;
1145         REISERFS_I(inode)->i_prealloc_block = 0;
1146         REISERFS_I(inode)->i_prealloc_count = 0;
1147         REISERFS_I(inode)->i_trans_id = 0;
1148         REISERFS_I(inode)->i_jl = NULL;
1149         reiserfs_init_xattr_rwsem(inode);
1150
1151         if (stat_data_v1(ih)) {
1152                 struct stat_data_v1 *sd =
1153                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1154                 unsigned long blocks;
1155
1156                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1157                 set_inode_sd_version(inode, STAT_DATA_V1);
1158                 inode->i_mode = sd_v1_mode(sd);
1159                 inode->i_nlink = sd_v1_nlink(sd);
1160                 inode->i_uid = sd_v1_uid(sd);
1161                 inode->i_gid = sd_v1_gid(sd);
1162                 inode->i_size = sd_v1_size(sd);
1163                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1164                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1165                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1166                 inode->i_atime.tv_nsec = 0;
1167                 inode->i_ctime.tv_nsec = 0;
1168                 inode->i_mtime.tv_nsec = 0;
1169
1170                 inode->i_blocks = sd_v1_blocks(sd);
1171                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1172                 blocks = (inode->i_size + 511) >> 9;
1173                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1174                 if (inode->i_blocks > blocks) {
1175                         // there was a bug in <=3.5.23 when i_blocks could take negative
1176                         // values. Starting from 3.5.17 this value could even be stored in
1177                         // stat data. For such files we set i_blocks based on file
1178                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1179                         // only updated if file's inode will ever change
1180                         inode->i_blocks = blocks;
1181                 }
1182
1183                 rdev = sd_v1_rdev(sd);
1184                 REISERFS_I(inode)->i_first_direct_byte =
1185                     sd_v1_first_direct_byte(sd);
1186                 /* an early bug in the quota code can give us an odd number for the
1187                  ** block count.  This is incorrect, fix it here.
1188                  */
1189                 if (inode->i_blocks & 1) {
1190                         inode->i_blocks++;
1191                 }
1192                 inode_set_bytes(inode,
1193                                 to_real_used_space(inode, inode->i_blocks,
1194                                                    SD_V1_SIZE));
1195                 /* nopack is initially zero for v1 objects. For v2 objects,
1196                    nopack is initialised from sd_attrs */
1197                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1198         } else {
1199                 // new stat data found, but object may have old items
1200                 // (directories and symlinks)
1201                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1202
1203                 inode->i_mode = sd_v2_mode(sd);
1204                 inode->i_nlink = sd_v2_nlink(sd);
1205                 inode->i_uid = sd_v2_uid(sd);
1206                 inode->i_size = sd_v2_size(sd);
1207                 inode->i_gid = sd_v2_gid(sd);
1208                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1209                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1210                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1211                 inode->i_ctime.tv_nsec = 0;
1212                 inode->i_mtime.tv_nsec = 0;
1213                 inode->i_atime.tv_nsec = 0;
1214                 inode->i_blocks = sd_v2_blocks(sd);
1215                 rdev = sd_v2_rdev(sd);
1216                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1217                         inode->i_generation =
1218                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1219                 else
1220                         inode->i_generation = sd_v2_generation(sd);
1221
1222                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1223                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1224                 else
1225                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1226                 REISERFS_I(inode)->i_first_direct_byte = 0;
1227                 set_inode_sd_version(inode, STAT_DATA_V2);
1228                 inode_set_bytes(inode,
1229                                 to_real_used_space(inode, inode->i_blocks,
1230                                                    SD_V2_SIZE));
1231                 /* read persistent inode attributes from sd and initialise
1232                    generic inode flags from them */
1233                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1234                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1235         }
1236
1237         pathrelse(path);
1238         if (S_ISREG(inode->i_mode)) {
1239                 inode->i_op = &reiserfs_file_inode_operations;
1240                 inode->i_fop = &reiserfs_file_operations;
1241                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1242         } else if (S_ISDIR(inode->i_mode)) {
1243                 inode->i_op = &reiserfs_dir_inode_operations;
1244                 inode->i_fop = &reiserfs_dir_operations;
1245         } else if (S_ISLNK(inode->i_mode)) {
1246                 inode->i_op = &reiserfs_symlink_inode_operations;
1247                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1248         } else {
1249                 inode->i_blocks = 0;
1250                 inode->i_op = &reiserfs_special_inode_operations;
1251                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1252         }
1253 }
1254
1255 // update new stat data with inode fields
1256 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1257 {
1258         struct stat_data *sd_v2 = (struct stat_data *)sd;
1259         __u16 flags;
1260
1261         set_sd_v2_mode(sd_v2, inode->i_mode);
1262         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1263         set_sd_v2_uid(sd_v2, inode->i_uid);
1264         set_sd_v2_size(sd_v2, size);
1265         set_sd_v2_gid(sd_v2, inode->i_gid);
1266         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1267         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1268         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1269         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1270         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1271                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1272         else
1273                 set_sd_v2_generation(sd_v2, inode->i_generation);
1274         flags = REISERFS_I(inode)->i_attrs;
1275         i_attrs_to_sd_attrs(inode, &flags);
1276         set_sd_v2_attrs(sd_v2, flags);
1277 }
1278
1279 // used to copy inode's fields to old stat data
1280 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1281 {
1282         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1283
1284         set_sd_v1_mode(sd_v1, inode->i_mode);
1285         set_sd_v1_uid(sd_v1, inode->i_uid);
1286         set_sd_v1_gid(sd_v1, inode->i_gid);
1287         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1288         set_sd_v1_size(sd_v1, size);
1289         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1290         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1291         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1292
1293         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1294                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1295         else
1296                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1297
1298         // Sigh. i_first_direct_byte is back
1299         set_sd_v1_first_direct_byte(sd_v1,
1300                                     REISERFS_I(inode)->i_first_direct_byte);
1301 }
1302
1303 /* NOTE, you must prepare the buffer head before sending it here,
1304 ** and then log it after the call
1305 */
1306 static void update_stat_data(struct treepath *path, struct inode *inode,
1307                              loff_t size)
1308 {
1309         struct buffer_head *bh;
1310         struct item_head *ih;
1311
1312         bh = PATH_PLAST_BUFFER(path);
1313         ih = PATH_PITEM_HEAD(path);
1314
1315         if (!is_statdata_le_ih(ih))
1316                 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1317                                INODE_PKEY(inode), ih);
1318
1319         if (stat_data_v1(ih)) {
1320                 // path points to old stat data
1321                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1322         } else {
1323                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1324         }
1325
1326         return;
1327 }
1328
1329 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1330                              struct inode *inode, loff_t size)
1331 {
1332         struct cpu_key key;
1333         INITIALIZE_PATH(path);
1334         struct buffer_head *bh;
1335         int fs_gen;
1336         struct item_head *ih, tmp_ih;
1337         int retval;
1338
1339         BUG_ON(!th->t_trans_id);
1340
1341         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1342
1343         for (;;) {
1344                 int pos;
1345                 /* look for the object's stat data */
1346                 retval = search_item(inode->i_sb, &key, &path);
1347                 if (retval == IO_ERROR) {
1348                         reiserfs_error(inode->i_sb, "vs-13050",
1349                                        "i/o failure occurred trying to "
1350                                        "update %K stat data", &key);
1351                         return;
1352                 }
1353                 if (retval == ITEM_NOT_FOUND) {
1354                         pos = PATH_LAST_POSITION(&path);
1355                         pathrelse(&path);
1356                         if (inode->i_nlink == 0) {
1357                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1358                                 return;
1359                         }
1360                         reiserfs_warning(inode->i_sb, "vs-13060",
1361                                          "stat data of object %k (nlink == %d) "
1362                                          "not found (pos %d)",
1363                                          INODE_PKEY(inode), inode->i_nlink,
1364                                          pos);
1365                         reiserfs_check_path(&path);
1366                         return;
1367                 }
1368
1369                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1370                  ** FS might change.  We have to detect that, and loop back to the
1371                  ** search if the stat data item has moved
1372                  */
1373                 bh = get_last_bh(&path);
1374                 ih = get_ih(&path);
1375                 copy_item_head(&tmp_ih, ih);
1376                 fs_gen = get_generation(inode->i_sb);
1377                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1378                 if (fs_changed(fs_gen, inode->i_sb)
1379                     && item_moved(&tmp_ih, &path)) {
1380                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1381                         continue;       /* Stat_data item has been moved after scheduling. */
1382                 }
1383                 break;
1384         }
1385         update_stat_data(&path, inode, size);
1386         journal_mark_dirty(th, th->t_super, bh);
1387         pathrelse(&path);
1388         return;
1389 }
1390
1391 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1392 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1393 ** and clear the key in the private portion of the inode, otherwise a
1394 ** corresponding iput might try to delete whatever object the inode last
1395 ** represented.
1396 */
1397 static void reiserfs_make_bad_inode(struct inode *inode)
1398 {
1399         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1400         make_bad_inode(inode);
1401 }
1402
1403 //
1404 // initially this function was derived from minix or ext2's analog and
1405 // evolved as the prototype did
1406 //
1407
1408 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1409 {
1410         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1411         inode->i_ino = args->objectid;
1412         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1413         return 0;
1414 }
1415
1416 /* looks for stat data in the tree, and fills up the fields of in-core
1417    inode stat data fields */
1418 void reiserfs_read_locked_inode(struct inode *inode,
1419                                 struct reiserfs_iget_args *args)
1420 {
1421         INITIALIZE_PATH(path_to_sd);
1422         struct cpu_key key;
1423         unsigned long dirino;
1424         int retval;
1425
1426         dirino = args->dirid;
1427
1428         /* set version 1, version 2 could be used too, because stat data
1429            key is the same in both versions */
1430         key.version = KEY_FORMAT_3_5;
1431         key.on_disk_key.k_dir_id = dirino;
1432         key.on_disk_key.k_objectid = inode->i_ino;
1433         key.on_disk_key.k_offset = 0;
1434         key.on_disk_key.k_type = 0;
1435
1436         /* look for the object's stat data */
1437         retval = search_item(inode->i_sb, &key, &path_to_sd);
1438         if (retval == IO_ERROR) {
1439                 reiserfs_error(inode->i_sb, "vs-13070",
1440                                "i/o failure occurred trying to find "
1441                                "stat data of %K", &key);
1442                 reiserfs_make_bad_inode(inode);
1443                 return;
1444         }
1445         if (retval != ITEM_FOUND) {
1446                 /* a stale NFS handle can trigger this without it being an error */
1447                 pathrelse(&path_to_sd);
1448                 reiserfs_make_bad_inode(inode);
1449                 inode->i_nlink = 0;
1450                 return;
1451         }
1452
1453         init_inode(inode, &path_to_sd);
1454
1455         /* It is possible that knfsd is trying to access inode of a file
1456            that is being removed from the disk by some other thread. As we
1457            update sd on unlink all that is required is to check for nlink
1458            here. This bug was first found by Sizif when debugging
1459            SquidNG/Butterfly, forgotten, and found again after Philippe
1460            Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1461
1462            More logical fix would require changes in fs/inode.c:iput() to
1463            remove inode from hash-table _after_ fs cleaned disk stuff up and
1464            in iget() to return NULL if I_FREEING inode is found in
1465            hash-table. */
1466         /* Currently there is one place where it's ok to meet inode with
1467            nlink==0: processing of open-unlinked and half-truncated files
1468            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1469         if ((inode->i_nlink == 0) &&
1470             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1471                 reiserfs_warning(inode->i_sb, "vs-13075",
1472                                  "dead inode read from disk %K. "
1473                                  "This is likely to be race with knfsd. Ignore",
1474                                  &key);
1475                 reiserfs_make_bad_inode(inode);
1476         }
1477
1478         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1479
1480 }
1481
1482 /**
1483  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1484  *
1485  * @inode:    inode from hash table to check
1486  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1487  *
1488  * This function is called by iget5_locked() to distinguish reiserfs inodes
1489  * having the same inode numbers. Such inodes can only exist due to some
1490  * error condition. One of them should be bad. Inodes with identical
1491  * inode numbers (objectids) are distinguished by parent directory ids.
1492  *
1493  */
1494 int reiserfs_find_actor(struct inode *inode, void *opaque)
1495 {
1496         struct reiserfs_iget_args *args;
1497
1498         args = opaque;
1499         /* args is already in CPU order */
1500         return (inode->i_ino == args->objectid) &&
1501             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1502 }
1503
1504 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1505 {
1506         struct inode *inode;
1507         struct reiserfs_iget_args args;
1508
1509         args.objectid = key->on_disk_key.k_objectid;
1510         args.dirid = key->on_disk_key.k_dir_id;
1511         reiserfs_write_unlock(s);
1512         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1513                              reiserfs_find_actor, reiserfs_init_locked_inode,
1514                              (void *)(&args));
1515         reiserfs_write_lock(s);
1516         if (!inode)
1517                 return ERR_PTR(-ENOMEM);
1518
1519         if (inode->i_state & I_NEW) {
1520                 reiserfs_read_locked_inode(inode, &args);
1521                 unlock_new_inode(inode);
1522         }
1523
1524         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1525                 /* either due to i/o error or a stale NFS handle */
1526                 iput(inode);
1527                 inode = NULL;
1528         }
1529         return inode;
1530 }
1531
1532 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1533         u32 objectid, u32 dir_id, u32 generation)
1534
1535 {
1536         struct cpu_key key;
1537         struct inode *inode;
1538
1539         key.on_disk_key.k_objectid = objectid;
1540         key.on_disk_key.k_dir_id = dir_id;
1541         reiserfs_write_lock(sb);
1542         inode = reiserfs_iget(sb, &key);
1543         if (inode && !IS_ERR(inode) && generation != 0 &&
1544             generation != inode->i_generation) {
1545                 iput(inode);
1546                 inode = NULL;
1547         }
1548         reiserfs_write_unlock(sb);
1549
1550         return d_obtain_alias(inode);
1551 }
1552
1553 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1554                 int fh_len, int fh_type)
1555 {
1556         /* fhtype happens to reflect the number of u32s encoded.
1557          * due to a bug in earlier code, fhtype might indicate there
1558          * are more u32s then actually fitted.
1559          * so if fhtype seems to be more than len, reduce fhtype.
1560          * Valid types are:
1561          *   2 - objectid + dir_id - legacy support
1562          *   3 - objectid + dir_id + generation
1563          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1564          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1565          *   6 - as above plus generation of directory
1566          * 6 does not fit in NFSv2 handles
1567          */
1568         if (fh_type > fh_len) {
1569                 if (fh_type != 6 || fh_len != 5)
1570                         reiserfs_warning(sb, "reiserfs-13077",
1571                                 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1572                                 fh_type, fh_len);
1573                 fh_type = 5;
1574         }
1575
1576         return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1577                 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1578 }
1579
1580 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1581                 int fh_len, int fh_type)
1582 {
1583         if (fh_type < 4)
1584                 return NULL;
1585
1586         return reiserfs_get_dentry(sb,
1587                 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1588                 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1589                 (fh_type == 6) ? fid->raw[5] : 0);
1590 }
1591
1592 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1593                        int need_parent)
1594 {
1595         struct inode *inode = dentry->d_inode;
1596         int maxlen = *lenp;
1597
1598         if (maxlen < 3)
1599                 return 255;
1600
1601         data[0] = inode->i_ino;
1602         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1603         data[2] = inode->i_generation;
1604         *lenp = 3;
1605         /* no room for directory info? return what we've stored so far */
1606         if (maxlen < 5 || !need_parent)
1607                 return 3;
1608
1609         spin_lock(&dentry->d_lock);
1610         inode = dentry->d_parent->d_inode;
1611         data[3] = inode->i_ino;
1612         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1613         *lenp = 5;
1614         if (maxlen >= 6) {
1615                 data[5] = inode->i_generation;
1616                 *lenp = 6;
1617         }
1618         spin_unlock(&dentry->d_lock);
1619         return *lenp;
1620 }
1621
1622 /* looks for stat data, then copies fields to it, marks the buffer
1623    containing stat data as dirty */
1624 /* reiserfs inodes are never really dirty, since the dirty inode call
1625 ** always logs them.  This call allows the VFS inode marking routines
1626 ** to properly mark inodes for datasync and such, but only actually
1627 ** does something when called for a synchronous update.
1628 */
1629 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1630 {
1631         struct reiserfs_transaction_handle th;
1632         int jbegin_count = 1;
1633
1634         if (inode->i_sb->s_flags & MS_RDONLY)
1635                 return -EROFS;
1636         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1637          ** these cases are just when the system needs ram, not when the
1638          ** inode needs to reach disk for safety, and they can safely be
1639          ** ignored because the altered inode has already been logged.
1640          */
1641         if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1642                 reiserfs_write_lock(inode->i_sb);
1643                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1644                         reiserfs_update_sd(&th, inode);
1645                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1646                 }
1647                 reiserfs_write_unlock(inode->i_sb);
1648         }
1649         return 0;
1650 }
1651
1652 /* stat data of new object is inserted already, this inserts the item
1653    containing "." and ".." entries */
1654 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1655                                   struct inode *inode,
1656                                   struct item_head *ih, struct treepath *path,
1657                                   struct inode *dir)
1658 {
1659         struct super_block *sb = th->t_super;
1660         char empty_dir[EMPTY_DIR_SIZE];
1661         char *body = empty_dir;
1662         struct cpu_key key;
1663         int retval;
1664
1665         BUG_ON(!th->t_trans_id);
1666
1667         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1668                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1669                       TYPE_DIRENTRY, 3 /*key length */ );
1670
1671         /* compose item head for new item. Directories consist of items of
1672            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1673            is done by reiserfs_new_inode */
1674         if (old_format_only(sb)) {
1675                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1676                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1677
1678                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1679                                        ih->ih_key.k_objectid,
1680                                        INODE_PKEY(dir)->k_dir_id,
1681                                        INODE_PKEY(dir)->k_objectid);
1682         } else {
1683                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1684                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1685
1686                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1687                                     ih->ih_key.k_objectid,
1688                                     INODE_PKEY(dir)->k_dir_id,
1689                                     INODE_PKEY(dir)->k_objectid);
1690         }
1691
1692         /* look for place in the tree for new item */
1693         retval = search_item(sb, &key, path);
1694         if (retval == IO_ERROR) {
1695                 reiserfs_error(sb, "vs-13080",
1696                                "i/o failure occurred creating new directory");
1697                 return -EIO;
1698         }
1699         if (retval == ITEM_FOUND) {
1700                 pathrelse(path);
1701                 reiserfs_warning(sb, "vs-13070",
1702                                  "object with this key exists (%k)",
1703                                  &(ih->ih_key));
1704                 return -EEXIST;
1705         }
1706
1707         /* insert item, that is empty directory item */
1708         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1709 }
1710
1711 /* stat data of object has been inserted, this inserts the item
1712    containing the body of symlink */
1713 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1714                                 struct item_head *ih,
1715                                 struct treepath *path, const char *symname,
1716                                 int item_len)
1717 {
1718         struct super_block *sb = th->t_super;
1719         struct cpu_key key;
1720         int retval;
1721
1722         BUG_ON(!th->t_trans_id);
1723
1724         _make_cpu_key(&key, KEY_FORMAT_3_5,
1725                       le32_to_cpu(ih->ih_key.k_dir_id),
1726                       le32_to_cpu(ih->ih_key.k_objectid),
1727                       1, TYPE_DIRECT, 3 /*key length */ );
1728
1729         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1730                           0 /*free_space */ );
1731
1732         /* look for place in the tree for new item */
1733         retval = search_item(sb, &key, path);
1734         if (retval == IO_ERROR) {
1735                 reiserfs_error(sb, "vs-13080",
1736                                "i/o failure occurred creating new symlink");
1737                 return -EIO;
1738         }
1739         if (retval == ITEM_FOUND) {
1740                 pathrelse(path);
1741                 reiserfs_warning(sb, "vs-13080",
1742                                  "object with this key exists (%k)",
1743                                  &(ih->ih_key));
1744                 return -EEXIST;
1745         }
1746
1747         /* insert item, that is body of symlink */
1748         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1749 }
1750
1751 /* inserts the stat data into the tree, and then calls
1752    reiserfs_new_directory (to insert ".", ".." item if new object is
1753    directory) or reiserfs_new_symlink (to insert symlink body if new
1754    object is symlink) or nothing (if new object is regular file)
1755
1756    NOTE! uid and gid must already be set in the inode.  If we return
1757    non-zero due to an error, we have to drop the quota previously allocated
1758    for the fresh inode.  This can only be done outside a transaction, so
1759    if we return non-zero, we also end the transaction.  */
1760 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1761                        struct inode *dir, int mode, const char *symname,
1762                        /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1763                           strlen (symname) for symlinks) */
1764                        loff_t i_size, struct dentry *dentry,
1765                        struct inode *inode,
1766                        struct reiserfs_security_handle *security)
1767 {
1768         struct super_block *sb;
1769         struct reiserfs_iget_args args;
1770         INITIALIZE_PATH(path_to_key);
1771         struct cpu_key key;
1772         struct item_head ih;
1773         struct stat_data sd;
1774         int retval;
1775         int err;
1776
1777         BUG_ON(!th->t_trans_id);
1778
1779         dquot_initialize(inode);
1780         err = dquot_alloc_inode(inode);
1781         if (err)
1782                 goto out_end_trans;
1783         if (!dir->i_nlink) {
1784                 err = -EPERM;
1785                 goto out_bad_inode;
1786         }
1787
1788         sb = dir->i_sb;
1789
1790         /* item head of new item */
1791         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1792         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1793         if (!ih.ih_key.k_objectid) {
1794                 err = -ENOMEM;
1795                 goto out_bad_inode;
1796         }
1797         args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1798         if (old_format_only(sb))
1799                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1800                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1801         else
1802                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1803                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1804         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1805         args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1806         if (insert_inode_locked4(inode, args.objectid,
1807                              reiserfs_find_actor, &args) < 0) {
1808                 err = -EINVAL;
1809                 goto out_bad_inode;
1810         }
1811         if (old_format_only(sb))
1812                 /* not a perfect generation count, as object ids can be reused, but
1813                  ** this is as good as reiserfs can do right now.
1814                  ** note that the private part of inode isn't filled in yet, we have
1815                  ** to use the directory.
1816                  */
1817                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1818         else
1819 #if defined( USE_INODE_GENERATION_COUNTER )
1820                 inode->i_generation =
1821                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1822 #else
1823                 inode->i_generation = ++event;
1824 #endif
1825
1826         /* fill stat data */
1827         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1828
1829         /* uid and gid must already be set by the caller for quota init */
1830
1831         /* symlink cannot be immutable or append only, right? */
1832         if (S_ISLNK(inode->i_mode))
1833                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1834
1835         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1836         inode->i_size = i_size;
1837         inode->i_blocks = 0;
1838         inode->i_bytes = 0;
1839         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1840             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1841
1842         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1843         REISERFS_I(inode)->i_flags = 0;
1844         REISERFS_I(inode)->i_prealloc_block = 0;
1845         REISERFS_I(inode)->i_prealloc_count = 0;
1846         REISERFS_I(inode)->i_trans_id = 0;
1847         REISERFS_I(inode)->i_jl = NULL;
1848         REISERFS_I(inode)->i_attrs =
1849             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1850         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1851         reiserfs_init_xattr_rwsem(inode);
1852
1853         /* key to search for correct place for new stat data */
1854         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1855                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1856                       TYPE_STAT_DATA, 3 /*key length */ );
1857
1858         /* find proper place for inserting of stat data */
1859         retval = search_item(sb, &key, &path_to_key);
1860         if (retval == IO_ERROR) {
1861                 err = -EIO;
1862                 goto out_bad_inode;
1863         }
1864         if (retval == ITEM_FOUND) {
1865                 pathrelse(&path_to_key);
1866                 err = -EEXIST;
1867                 goto out_bad_inode;
1868         }
1869         if (old_format_only(sb)) {
1870                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1871                         pathrelse(&path_to_key);
1872                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1873                         err = -EINVAL;
1874                         goto out_bad_inode;
1875                 }
1876                 inode2sd_v1(&sd, inode, inode->i_size);
1877         } else {
1878                 inode2sd(&sd, inode, inode->i_size);
1879         }
1880         // store in in-core inode the key of stat data and version all
1881         // object items will have (directory items will have old offset
1882         // format, other new objects will consist of new items)
1883         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1884                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1885         else
1886                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1887         if (old_format_only(sb))
1888                 set_inode_sd_version(inode, STAT_DATA_V1);
1889         else
1890                 set_inode_sd_version(inode, STAT_DATA_V2);
1891
1892         /* insert the stat data into the tree */
1893 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1894         if (REISERFS_I(dir)->new_packing_locality)
1895                 th->displace_new_blocks = 1;
1896 #endif
1897         retval =
1898             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1899                                  (char *)(&sd));
1900         if (retval) {
1901                 err = retval;
1902                 reiserfs_check_path(&path_to_key);
1903                 goto out_bad_inode;
1904         }
1905 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1906         if (!th->displace_new_blocks)
1907                 REISERFS_I(dir)->new_packing_locality = 0;
1908 #endif
1909         if (S_ISDIR(mode)) {
1910                 /* insert item with "." and ".." */
1911                 retval =
1912                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1913         }
1914
1915         if (S_ISLNK(mode)) {
1916                 /* insert body of symlink */
1917                 if (!old_format_only(sb))
1918                         i_size = ROUND_UP(i_size);
1919                 retval =
1920                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1921                                          i_size);
1922         }
1923         if (retval) {
1924                 err = retval;
1925                 reiserfs_check_path(&path_to_key);
1926                 journal_end(th, th->t_super, th->t_blocks_allocated);
1927                 goto out_inserted_sd;
1928         }
1929
1930         if (reiserfs_posixacl(inode->i_sb)) {
1931                 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1932                 if (retval) {
1933                         err = retval;
1934                         reiserfs_check_path(&path_to_key);
1935                         journal_end(th, th->t_super, th->t_blocks_allocated);
1936                         goto out_inserted_sd;
1937                 }
1938         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1939                 reiserfs_warning(inode->i_sb, "jdm-13090",
1940                                  "ACLs aren't enabled in the fs, "
1941                                  "but vfs thinks they are!");
1942         } else if (IS_PRIVATE(dir))
1943                 inode->i_flags |= S_PRIVATE;
1944
1945         if (security->name) {
1946                 retval = reiserfs_security_write(th, inode, security);
1947                 if (retval) {
1948                         err = retval;
1949                         reiserfs_check_path(&path_to_key);
1950                         retval = journal_end(th, th->t_super,
1951                                              th->t_blocks_allocated);
1952                         if (retval)
1953                                 err = retval;
1954                         goto out_inserted_sd;
1955                 }
1956         }
1957
1958         reiserfs_update_sd(th, inode);
1959         reiserfs_check_path(&path_to_key);
1960
1961         return 0;
1962
1963 /* it looks like you can easily compress these two goto targets into
1964  * one.  Keeping it like this doesn't actually hurt anything, and they
1965  * are place holders for what the quota code actually needs.
1966  */
1967       out_bad_inode:
1968         /* Invalidate the object, nothing was inserted yet */
1969         INODE_PKEY(inode)->k_objectid = 0;
1970
1971         /* Quota change must be inside a transaction for journaling */
1972         dquot_free_inode(inode);
1973
1974       out_end_trans:
1975         journal_end(th, th->t_super, th->t_blocks_allocated);
1976         /* Drop can be outside and it needs more credits so it's better to have it outside */
1977         dquot_drop(inode);
1978         inode->i_flags |= S_NOQUOTA;
1979         make_bad_inode(inode);
1980
1981       out_inserted_sd:
1982         inode->i_nlink = 0;
1983         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1984         unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1985         iput(inode);
1986         return err;
1987 }
1988
1989 /*
1990 ** finds the tail page in the page cache,
1991 ** reads the last block in.
1992 **
1993 ** On success, page_result is set to a locked, pinned page, and bh_result
1994 ** is set to an up to date buffer for the last block in the file.  returns 0.
1995 **
1996 ** tail conversion is not done, so bh_result might not be valid for writing
1997 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1998 ** trying to write the block.
1999 **
2000 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2001 */
2002 static int grab_tail_page(struct inode *inode,
2003                           struct page **page_result,
2004                           struct buffer_head **bh_result)
2005 {
2006
2007         /* we want the page with the last byte in the file,
2008          ** not the page that will hold the next byte for appending
2009          */
2010         unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2011         unsigned long pos = 0;
2012         unsigned long start = 0;
2013         unsigned long blocksize = inode->i_sb->s_blocksize;
2014         unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2015         struct buffer_head *bh;
2016         struct buffer_head *head;
2017         struct page *page;
2018         int error;
2019
2020         /* we know that we are only called with inode->i_size > 0.
2021          ** we also know that a file tail can never be as big as a block
2022          ** If i_size % blocksize == 0, our file is currently block aligned
2023          ** and it won't need converting or zeroing after a truncate.
2024          */
2025         if ((offset & (blocksize - 1)) == 0) {
2026                 return -ENOENT;
2027         }
2028         page = grab_cache_page(inode->i_mapping, index);
2029         error = -ENOMEM;
2030         if (!page) {
2031                 goto out;
2032         }
2033         /* start within the page of the last block in the file */
2034         start = (offset / blocksize) * blocksize;
2035
2036         error = block_prepare_write(page, start, offset,
2037                                     reiserfs_get_block_create_0);
2038         if (error)
2039                 goto unlock;
2040
2041         head = page_buffers(page);
2042         bh = head;
2043         do {
2044                 if (pos >= start) {
2045                         break;
2046                 }
2047                 bh = bh->b_this_page;
2048                 pos += blocksize;
2049         } while (bh != head);
2050
2051         if (!buffer_uptodate(bh)) {
2052                 /* note, this should never happen, prepare_write should
2053                  ** be taking care of this for us.  If the buffer isn't up to date,
2054                  ** I've screwed up the code to find the buffer, or the code to
2055                  ** call prepare_write
2056                  */
2057                 reiserfs_error(inode->i_sb, "clm-6000",
2058                                "error reading block %lu", bh->b_blocknr);
2059                 error = -EIO;
2060                 goto unlock;
2061         }
2062         *bh_result = bh;
2063         *page_result = page;
2064
2065       out:
2066         return error;
2067
2068       unlock:
2069         unlock_page(page);
2070         page_cache_release(page);
2071         return error;
2072 }
2073
2074 /*
2075 ** vfs version of truncate file.  Must NOT be called with
2076 ** a transaction already started.
2077 **
2078 ** some code taken from block_truncate_page
2079 */
2080 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2081 {
2082         struct reiserfs_transaction_handle th;
2083         /* we want the offset for the first byte after the end of the file */
2084         unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2085         unsigned blocksize = inode->i_sb->s_blocksize;
2086         unsigned length;
2087         struct page *page = NULL;
2088         int error;
2089         struct buffer_head *bh = NULL;
2090         int err2;
2091         int lock_depth;
2092
2093         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2094
2095         if (inode->i_size > 0) {
2096                 error = grab_tail_page(inode, &page, &bh);
2097                 if (error) {
2098                         // -ENOENT means we truncated past the end of the file,
2099                         // and get_block_create_0 could not find a block to read in,
2100                         // which is ok.
2101                         if (error != -ENOENT)
2102                                 reiserfs_error(inode->i_sb, "clm-6001",
2103                                                "grab_tail_page failed %d",
2104                                                error);
2105                         page = NULL;
2106                         bh = NULL;
2107                 }
2108         }
2109
2110         /* so, if page != NULL, we have a buffer head for the offset at
2111          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2112          ** then we have an unformatted node.  Otherwise, we have a direct item,
2113          ** and no zeroing is required on disk.  We zero after the truncate,
2114          ** because the truncate might pack the item anyway
2115          ** (it will unmap bh if it packs).
2116          */
2117         /* it is enough to reserve space in transaction for 2 balancings:
2118            one for "save" link adding and another for the first
2119            cut_from_item. 1 is for update_sd */
2120         error = journal_begin(&th, inode->i_sb,
2121                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2122         if (error)
2123                 goto out;
2124         reiserfs_update_inode_transaction(inode);
2125         if (update_timestamps)
2126                 /* we are doing real truncate: if the system crashes before the last
2127                    transaction of truncating gets committed - on reboot the file
2128                    either appears truncated properly or not truncated at all */
2129                 add_save_link(&th, inode, 1);
2130         err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2131         error =
2132             journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2133         if (error)
2134                 goto out;
2135
2136         /* check reiserfs_do_truncate after ending the transaction */
2137         if (err2) {
2138                 error = err2;
2139                 goto out;
2140         }
2141         
2142         if (update_timestamps) {
2143                 error = remove_save_link(inode, 1 /* truncate */);
2144                 if (error)
2145                         goto out;
2146         }
2147
2148         if (page) {
2149                 length = offset & (blocksize - 1);
2150                 /* if we are not on a block boundary */
2151                 if (length) {
2152                         length = blocksize - length;
2153                         zero_user(page, offset, length);
2154                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2155                                 mark_buffer_dirty(bh);
2156                         }
2157                 }
2158                 unlock_page(page);
2159                 page_cache_release(page);
2160         }
2161
2162         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2163
2164         return 0;
2165       out:
2166         if (page) {
2167                 unlock_page(page);
2168                 page_cache_release(page);
2169         }
2170
2171         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2172
2173         return error;
2174 }
2175
2176 static int map_block_for_writepage(struct inode *inode,
2177                                    struct buffer_head *bh_result,
2178                                    unsigned long block)
2179 {
2180         struct reiserfs_transaction_handle th;
2181         int fs_gen;
2182         struct item_head tmp_ih;
2183         struct item_head *ih;
2184         struct buffer_head *bh;
2185         __le32 *item;
2186         struct cpu_key key;
2187         INITIALIZE_PATH(path);
2188         int pos_in_item;
2189         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2190         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2191         int retval;
2192         int use_get_block = 0;
2193         int bytes_copied = 0;
2194         int copy_size;
2195         int trans_running = 0;
2196
2197         /* catch places below that try to log something without starting a trans */
2198         th.t_trans_id = 0;
2199
2200         if (!buffer_uptodate(bh_result)) {
2201                 return -EIO;
2202         }
2203
2204         kmap(bh_result->b_page);
2205       start_over:
2206         reiserfs_write_lock(inode->i_sb);
2207         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2208
2209       research:
2210         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2211         if (retval != POSITION_FOUND) {
2212                 use_get_block = 1;
2213                 goto out;
2214         }
2215
2216         bh = get_last_bh(&path);
2217         ih = get_ih(&path);
2218         item = get_item(&path);
2219         pos_in_item = path.pos_in_item;
2220
2221         /* we've found an unformatted node */
2222         if (indirect_item_found(retval, ih)) {
2223                 if (bytes_copied > 0) {
2224                         reiserfs_warning(inode->i_sb, "clm-6002",
2225                                          "bytes_copied %d", bytes_copied);
2226                 }
2227                 if (!get_block_num(item, pos_in_item)) {
2228                         /* crap, we are writing to a hole */
2229                         use_get_block = 1;
2230                         goto out;
2231                 }
2232                 set_block_dev_mapped(bh_result,
2233                                      get_block_num(item, pos_in_item), inode);
2234         } else if (is_direct_le_ih(ih)) {
2235                 char *p;
2236                 p = page_address(bh_result->b_page);
2237                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2238                 copy_size = ih_item_len(ih) - pos_in_item;
2239
2240                 fs_gen = get_generation(inode->i_sb);
2241                 copy_item_head(&tmp_ih, ih);
2242
2243                 if (!trans_running) {
2244                         /* vs-3050 is gone, no need to drop the path */
2245                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2246                         if (retval)
2247                                 goto out;
2248                         reiserfs_update_inode_transaction(inode);
2249                         trans_running = 1;
2250                         if (fs_changed(fs_gen, inode->i_sb)
2251                             && item_moved(&tmp_ih, &path)) {
2252                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2253                                                                  bh);
2254                                 goto research;
2255                         }
2256                 }
2257
2258                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2259
2260                 if (fs_changed(fs_gen, inode->i_sb)
2261                     && item_moved(&tmp_ih, &path)) {
2262                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2263                         goto research;
2264                 }
2265
2266                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2267                        copy_size);
2268
2269                 journal_mark_dirty(&th, inode->i_sb, bh);
2270                 bytes_copied += copy_size;
2271                 set_block_dev_mapped(bh_result, 0, inode);
2272
2273                 /* are there still bytes left? */
2274                 if (bytes_copied < bh_result->b_size &&
2275                     (byte_offset + bytes_copied) < inode->i_size) {
2276                         set_cpu_key_k_offset(&key,
2277                                              cpu_key_k_offset(&key) +
2278                                              copy_size);
2279                         goto research;
2280                 }
2281         } else {
2282                 reiserfs_warning(inode->i_sb, "clm-6003",
2283                                  "bad item inode %lu", inode->i_ino);
2284                 retval = -EIO;
2285                 goto out;
2286         }
2287         retval = 0;
2288
2289       out:
2290         pathrelse(&path);
2291         if (trans_running) {
2292                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2293                 if (err)
2294                         retval = err;
2295                 trans_running = 0;
2296         }
2297         reiserfs_write_unlock(inode->i_sb);
2298
2299         /* this is where we fill in holes in the file. */
2300         if (use_get_block) {
2301                 retval = reiserfs_get_block(inode, block, bh_result,
2302                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2303                                             | GET_BLOCK_NO_DANGLE);
2304                 if (!retval) {
2305                         if (!buffer_mapped(bh_result)
2306                             || bh_result->b_blocknr == 0) {
2307                                 /* get_block failed to find a mapped unformatted node. */
2308                                 use_get_block = 0;
2309                                 goto start_over;
2310                         }
2311                 }
2312         }
2313         kunmap(bh_result->b_page);
2314
2315         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2316                 /* we've copied data from the page into the direct item, so the
2317                  * buffer in the page is now clean, mark it to reflect that.
2318                  */
2319                 lock_buffer(bh_result);
2320                 clear_buffer_dirty(bh_result);
2321                 unlock_buffer(bh_result);
2322         }
2323         return retval;
2324 }
2325
2326 /*
2327  * mason@suse.com: updated in 2.5.54 to follow the same general io
2328  * start/recovery path as __block_write_full_page, along with special
2329  * code to handle reiserfs tails.
2330  */
2331 static int reiserfs_write_full_page(struct page *page,
2332                                     struct writeback_control *wbc)
2333 {
2334         struct inode *inode = page->mapping->host;
2335         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2336         int error = 0;
2337         unsigned long block;
2338         sector_t last_block;
2339         struct buffer_head *head, *bh;
2340         int partial = 0;
2341         int nr = 0;
2342         int checked = PageChecked(page);
2343         struct reiserfs_transaction_handle th;
2344         struct super_block *s = inode->i_sb;
2345         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2346         th.t_trans_id = 0;
2347
2348         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2349         if (checked && (current->flags & PF_MEMALLOC)) {
2350                 redirty_page_for_writepage(wbc, page);
2351                 unlock_page(page);
2352                 return 0;
2353         }
2354
2355         /* The page dirty bit is cleared before writepage is called, which
2356          * means we have to tell create_empty_buffers to make dirty buffers
2357          * The page really should be up to date at this point, so tossing
2358          * in the BH_Uptodate is just a sanity check.
2359          */
2360         if (!page_has_buffers(page)) {
2361                 create_empty_buffers(page, s->s_blocksize,
2362                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2363         }
2364         head = page_buffers(page);
2365
2366         /* last page in the file, zero out any contents past the
2367          ** last byte in the file
2368          */
2369         if (page->index >= end_index) {
2370                 unsigned last_offset;
2371
2372                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2373                 /* no file contents in this page */
2374                 if (page->index >= end_index + 1 || !last_offset) {
2375                         unlock_page(page);
2376                         return 0;
2377                 }
2378                 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2379         }
2380         bh = head;
2381         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2382         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2383         /* first map all the buffers, logging any direct items we find */
2384         do {
2385                 if (block > last_block) {
2386                         /*
2387                          * This can happen when the block size is less than
2388                          * the page size.  The corresponding bytes in the page
2389                          * were zero filled above
2390                          */
2391                         clear_buffer_dirty(bh);
2392                         set_buffer_uptodate(bh);
2393                 } else if ((checked || buffer_dirty(bh)) &&
2394                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2395                                                        && bh->b_blocknr ==
2396                                                        0))) {
2397                         /* not mapped yet, or it points to a direct item, search
2398                          * the btree for the mapping info, and log any direct
2399                          * items found
2400                          */
2401                         if ((error = map_block_for_writepage(inode, bh, block))) {
2402                                 goto fail;
2403                         }
2404                 }
2405                 bh = bh->b_this_page;
2406                 block++;
2407         } while (bh != head);
2408
2409         /*
2410          * we start the transaction after map_block_for_writepage,
2411          * because it can create holes in the file (an unbounded operation).
2412          * starting it here, we can make a reliable estimate for how many
2413          * blocks we're going to log
2414          */
2415         if (checked) {
2416                 ClearPageChecked(page);
2417                 reiserfs_write_lock(s);
2418                 error = journal_begin(&th, s, bh_per_page + 1);
2419                 if (error) {
2420                         reiserfs_write_unlock(s);
2421                         goto fail;
2422                 }
2423                 reiserfs_update_inode_transaction(inode);
2424         }
2425         /* now go through and lock any dirty buffers on the page */
2426         do {
2427                 get_bh(bh);
2428                 if (!buffer_mapped(bh))
2429                         continue;
2430                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2431                         continue;
2432
2433                 if (checked) {
2434                         reiserfs_prepare_for_journal(s, bh, 1);
2435                         journal_mark_dirty(&th, s, bh);
2436                         continue;
2437                 }
2438                 /* from this point on, we know the buffer is mapped to a
2439                  * real block and not a direct item
2440                  */
2441                 if (wbc->sync_mode != WB_SYNC_NONE) {
2442                         lock_buffer(bh);
2443                 } else {
2444                         if (!trylock_buffer(bh)) {
2445                                 redirty_page_for_writepage(wbc, page);
2446                                 continue;
2447                         }
2448                 }
2449                 if (test_clear_buffer_dirty(bh)) {
2450                         mark_buffer_async_write(bh);
2451                 } else {
2452                         unlock_buffer(bh);
2453                 }
2454         } while ((bh = bh->b_this_page) != head);
2455
2456         if (checked) {
2457                 error = journal_end(&th, s, bh_per_page + 1);
2458                 reiserfs_write_unlock(s);
2459                 if (error)
2460                         goto fail;
2461         }
2462         BUG_ON(PageWriteback(page));
2463         set_page_writeback(page);
2464         unlock_page(page);
2465
2466         /*
2467          * since any buffer might be the only dirty buffer on the page,
2468          * the first submit_bh can bring the page out of writeback.
2469          * be careful with the buffers.
2470          */
2471         do {
2472                 struct buffer_head *next = bh->b_this_page;
2473                 if (buffer_async_write(bh)) {
2474                         submit_bh(WRITE, bh);
2475                         nr++;
2476                 }
2477                 put_bh(bh);
2478                 bh = next;
2479         } while (bh != head);
2480
2481         error = 0;
2482       done:
2483         if (nr == 0) {
2484                 /*
2485                  * if this page only had a direct item, it is very possible for
2486                  * no io to be required without there being an error.  Or,
2487                  * someone else could have locked them and sent them down the
2488                  * pipe without locking the page
2489                  */
2490                 bh = head;
2491                 do {
2492                         if (!buffer_uptodate(bh)) {
2493                                 partial = 1;
2494                                 break;
2495                         }
2496                         bh = bh->b_this_page;
2497                 } while (bh != head);
2498                 if (!partial)
2499                         SetPageUptodate(page);
2500                 end_page_writeback(page);
2501         }
2502         return error;
2503
2504       fail:
2505         /* catches various errors, we need to make sure any valid dirty blocks
2506          * get to the media.  The page is currently locked and not marked for
2507          * writeback
2508          */
2509         ClearPageUptodate(page);
2510         bh = head;
2511         do {
2512                 get_bh(bh);
2513                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2514                         lock_buffer(bh);
2515                         mark_buffer_async_write(bh);
2516                 } else {
2517                         /*
2518                          * clear any dirty bits that might have come from getting
2519                          * attached to a dirty page
2520                          */
2521                         clear_buffer_dirty(bh);
2522                 }
2523                 bh = bh->b_this_page;
2524         } while (bh != head);
2525         SetPageError(page);
2526         BUG_ON(PageWriteback(page));
2527         set_page_writeback(page);
2528         unlock_page(page);
2529         do {
2530                 struct buffer_head *next = bh->b_this_page;
2531                 if (buffer_async_write(bh)) {
2532                         clear_buffer_dirty(bh);
2533                         submit_bh(WRITE, bh);
2534                         nr++;
2535                 }
2536                 put_bh(bh);
2537                 bh = next;
2538         } while (bh != head);
2539         goto done;
2540 }
2541
2542 static int reiserfs_readpage(struct file *f, struct page *page)
2543 {
2544         return block_read_full_page(page, reiserfs_get_block);
2545 }
2546
2547 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2548 {
2549         struct inode *inode = page->mapping->host;
2550         reiserfs_wait_on_write_block(inode->i_sb);
2551         return reiserfs_write_full_page(page, wbc);
2552 }
2553
2554 static void reiserfs_truncate_failed_write(struct inode *inode)
2555 {
2556         truncate_inode_pages(inode->i_mapping, inode->i_size);
2557         reiserfs_truncate_file(inode, 0);
2558 }
2559
2560 static int reiserfs_write_begin(struct file *file,
2561                                 struct address_space *mapping,
2562                                 loff_t pos, unsigned len, unsigned flags,
2563                                 struct page **pagep, void **fsdata)
2564 {
2565         struct inode *inode;
2566         struct page *page;
2567         pgoff_t index;
2568         int ret;
2569         int old_ref = 0;
2570
2571         inode = mapping->host;
2572         *fsdata = 0;
2573         if (flags & AOP_FLAG_CONT_EXPAND &&
2574             (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2575                 pos ++;
2576                 *fsdata = (void *)(unsigned long)flags;
2577         }
2578
2579         index = pos >> PAGE_CACHE_SHIFT;
2580         page = grab_cache_page_write_begin(mapping, index, flags);
2581         if (!page)
2582                 return -ENOMEM;
2583         *pagep = page;
2584
2585         reiserfs_wait_on_write_block(inode->i_sb);
2586         fix_tail_page_for_writing(page);
2587         if (reiserfs_transaction_running(inode->i_sb)) {
2588                 struct reiserfs_transaction_handle *th;
2589                 th = (struct reiserfs_transaction_handle *)current->
2590                     journal_info;
2591                 BUG_ON(!th->t_refcount);
2592                 BUG_ON(!th->t_trans_id);
2593                 old_ref = th->t_refcount;
2594                 th->t_refcount++;
2595         }
2596         ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2597         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2598                 struct reiserfs_transaction_handle *th = current->journal_info;
2599                 /* this gets a little ugly.  If reiserfs_get_block returned an
2600                  * error and left a transacstion running, we've got to close it,
2601                  * and we've got to free handle if it was a persistent transaction.
2602                  *
2603                  * But, if we had nested into an existing transaction, we need
2604                  * to just drop the ref count on the handle.
2605                  *
2606                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2607                  * and it was a persistent trans.  Otherwise, it was nested above.
2608                  */
2609                 if (th->t_refcount > old_ref) {
2610                         if (old_ref)
2611                                 th->t_refcount--;
2612                         else {
2613                                 int err;
2614                                 reiserfs_write_lock(inode->i_sb);
2615                                 err = reiserfs_end_persistent_transaction(th);
2616                                 reiserfs_write_unlock(inode->i_sb);
2617                                 if (err)
2618                                         ret = err;
2619                         }
2620                 }
2621         }
2622         if (ret) {
2623                 unlock_page(page);
2624                 page_cache_release(page);
2625                 /* Truncate allocated blocks */
2626                 reiserfs_truncate_failed_write(inode);
2627         }
2628         return ret;
2629 }
2630
2631 int reiserfs_prepare_write(struct file *f, struct page *page,
2632                            unsigned from, unsigned to)
2633 {
2634         struct inode *inode = page->mapping->host;
2635         int ret;
2636         int old_ref = 0;
2637
2638         reiserfs_write_unlock(inode->i_sb);
2639         reiserfs_wait_on_write_block(inode->i_sb);
2640         reiserfs_write_lock(inode->i_sb);
2641
2642         fix_tail_page_for_writing(page);
2643         if (reiserfs_transaction_running(inode->i_sb)) {
2644                 struct reiserfs_transaction_handle *th;
2645                 th = (struct reiserfs_transaction_handle *)current->
2646                     journal_info;
2647                 BUG_ON(!th->t_refcount);
2648                 BUG_ON(!th->t_trans_id);
2649                 old_ref = th->t_refcount;
2650                 th->t_refcount++;
2651         }
2652
2653         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2654         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2655                 struct reiserfs_transaction_handle *th = current->journal_info;
2656                 /* this gets a little ugly.  If reiserfs_get_block returned an
2657                  * error and left a transacstion running, we've got to close it,
2658                  * and we've got to free handle if it was a persistent transaction.
2659                  *
2660                  * But, if we had nested into an existing transaction, we need
2661                  * to just drop the ref count on the handle.
2662                  *
2663                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2664                  * and it was a persistent trans.  Otherwise, it was nested above.
2665                  */
2666                 if (th->t_refcount > old_ref) {
2667                         if (old_ref)
2668                                 th->t_refcount--;
2669                         else {
2670                                 int err;
2671                                 reiserfs_write_lock(inode->i_sb);
2672                                 err = reiserfs_end_persistent_transaction(th);
2673                                 reiserfs_write_unlock(inode->i_sb);
2674                                 if (err)
2675                                         ret = err;
2676                         }
2677                 }
2678         }
2679         return ret;
2680
2681 }
2682
2683 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2684 {
2685         return generic_block_bmap(as, block, reiserfs_bmap);
2686 }
2687
2688 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2689                               loff_t pos, unsigned len, unsigned copied,
2690                               struct page *page, void *fsdata)
2691 {
2692         struct inode *inode = page->mapping->host;
2693         int ret = 0;
2694         int update_sd = 0;
2695         struct reiserfs_transaction_handle *th;
2696         unsigned start;
2697         int lock_depth = 0;
2698         bool locked = false;
2699
2700         if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2701                 pos ++;
2702
2703         reiserfs_wait_on_write_block(inode->i_sb);
2704         if (reiserfs_transaction_running(inode->i_sb))
2705                 th = current->journal_info;
2706         else
2707                 th = NULL;
2708
2709         start = pos & (PAGE_CACHE_SIZE - 1);
2710         if (unlikely(copied < len)) {
2711                 if (!PageUptodate(page))
2712                         copied = 0;
2713
2714                 page_zero_new_buffers(page, start + copied, start + len);
2715         }
2716         flush_dcache_page(page);
2717
2718         reiserfs_commit_page(inode, page, start, start + copied);
2719
2720         /* generic_commit_write does this for us, but does not update the
2721          ** transaction tracking stuff when the size changes.  So, we have
2722          ** to do the i_size updates here.
2723          */
2724         if (pos + copied > inode->i_size) {
2725                 struct reiserfs_transaction_handle myth;
2726                 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2727                 locked = true;
2728                 /* If the file have grown beyond the border where it
2729                    can have a tail, unmark it as needing a tail
2730                    packing */
2731                 if ((have_large_tails(inode->i_sb)
2732                      && inode->i_size > i_block_size(inode) * 4)
2733                     || (have_small_tails(inode->i_sb)
2734                         && inode->i_size > i_block_size(inode)))
2735                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2736
2737                 ret = journal_begin(&myth, inode->i_sb, 1);
2738                 if (ret)
2739                         goto journal_error;
2740
2741                 reiserfs_update_inode_transaction(inode);
2742                 inode->i_size = pos + copied;
2743                 /*
2744                  * this will just nest into our transaction.  It's important
2745                  * to use mark_inode_dirty so the inode gets pushed around on the
2746                  * dirty lists, and so that O_SYNC works as expected
2747                  */
2748                 mark_inode_dirty(inode);
2749                 reiserfs_update_sd(&myth, inode);
2750                 update_sd = 1;
2751                 ret = journal_end(&myth, inode->i_sb, 1);
2752                 if (ret)
2753                         goto journal_error;
2754         }
2755         if (th) {
2756                 if (!locked) {
2757                         lock_depth = reiserfs_write_lock_once(inode->i_sb);
2758                         locked = true;
2759                 }
2760                 if (!update_sd)
2761                         mark_inode_dirty(inode);
2762                 ret = reiserfs_end_persistent_transaction(th);
2763                 if (ret)
2764                         goto out;
2765         }
2766
2767       out:
2768         if (locked)
2769                 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2770         unlock_page(page);
2771         page_cache_release(page);
2772
2773         if (pos + len > inode->i_size)
2774                 reiserfs_truncate_failed_write(inode);
2775
2776         return ret == 0 ? copied : ret;
2777
2778       journal_error:
2779         reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2780         locked = false;
2781         if (th) {
2782                 if (!update_sd)
2783                         reiserfs_update_sd(th, inode);
2784                 ret = reiserfs_end_persistent_transaction(th);
2785         }
2786         goto out;
2787 }
2788
2789 int reiserfs_commit_write(struct file *f, struct page *page,
2790                           unsigned from, unsigned to)
2791 {
2792         struct inode *inode = page->mapping->host;
2793         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2794         int ret = 0;
2795         int update_sd = 0;
2796         struct reiserfs_transaction_handle *th = NULL;
2797
2798         reiserfs_write_unlock(inode->i_sb);
2799         reiserfs_wait_on_write_block(inode->i_sb);
2800         reiserfs_write_lock(inode->i_sb);
2801
2802         if (reiserfs_transaction_running(inode->i_sb)) {
2803                 th = current->journal_info;
2804         }
2805         reiserfs_commit_page(inode, page, from, to);
2806
2807         /* generic_commit_write does this for us, but does not update the
2808          ** transaction tracking stuff when the size changes.  So, we have
2809          ** to do the i_size updates here.
2810          */
2811         if (pos > inode->i_size) {
2812                 struct reiserfs_transaction_handle myth;
2813                 /* If the file have grown beyond the border where it
2814                    can have a tail, unmark it as needing a tail
2815                    packing */
2816                 if ((have_large_tails(inode->i_sb)
2817                      && inode->i_size > i_block_size(inode) * 4)
2818                     || (have_small_tails(inode->i_sb)
2819                         && inode->i_size > i_block_size(inode)))
2820                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2821
2822                 ret = journal_begin(&myth, inode->i_sb, 1);
2823                 if (ret)
2824                         goto journal_error;
2825
2826                 reiserfs_update_inode_transaction(inode);
2827                 inode->i_size = pos;
2828                 /*
2829                  * this will just nest into our transaction.  It's important
2830                  * to use mark_inode_dirty so the inode gets pushed around on the
2831                  * dirty lists, and so that O_SYNC works as expected
2832                  */
2833                 mark_inode_dirty(inode);
2834                 reiserfs_update_sd(&myth, inode);
2835                 update_sd = 1;
2836                 ret = journal_end(&myth, inode->i_sb, 1);
2837                 if (ret)
2838                         goto journal_error;
2839         }
2840         if (th) {
2841                 if (!update_sd)
2842                         mark_inode_dirty(inode);
2843                 ret = reiserfs_end_persistent_transaction(th);
2844                 if (ret)
2845                         goto out;
2846         }
2847
2848       out:
2849         return ret;
2850
2851       journal_error:
2852         if (th) {
2853                 if (!update_sd)
2854                         reiserfs_update_sd(th, inode);
2855                 ret = reiserfs_end_persistent_transaction(th);
2856         }
2857
2858         return ret;
2859 }
2860
2861 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2862 {
2863         if (reiserfs_attrs(inode->i_sb)) {
2864                 if (sd_attrs & REISERFS_SYNC_FL)
2865                         inode->i_flags |= S_SYNC;
2866                 else
2867                         inode->i_flags &= ~S_SYNC;
2868                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2869                         inode->i_flags |= S_IMMUTABLE;
2870                 else
2871                         inode->i_flags &= ~S_IMMUTABLE;
2872                 if (sd_attrs & REISERFS_APPEND_FL)
2873                         inode->i_flags |= S_APPEND;
2874                 else
2875                         inode->i_flags &= ~S_APPEND;
2876                 if (sd_attrs & REISERFS_NOATIME_FL)
2877                         inode->i_flags |= S_NOATIME;
2878                 else
2879                         inode->i_flags &= ~S_NOATIME;
2880                 if (sd_attrs & REISERFS_NOTAIL_FL)
2881                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2882                 else
2883                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2884         }
2885 }
2886
2887 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2888 {
2889         if (reiserfs_attrs(inode->i_sb)) {
2890                 if (inode->i_flags & S_IMMUTABLE)
2891                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2892                 else
2893                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2894                 if (inode->i_flags & S_SYNC)
2895                         *sd_attrs |= REISERFS_SYNC_FL;
2896                 else
2897                         *sd_attrs &= ~REISERFS_SYNC_FL;
2898                 if (inode->i_flags & S_NOATIME)
2899                         *sd_attrs |= REISERFS_NOATIME_FL;
2900                 else
2901                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2902                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2903                         *sd_attrs |= REISERFS_NOTAIL_FL;
2904                 else
2905                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2906         }
2907 }
2908
2909 /* decide if this buffer needs to stay around for data logging or ordered
2910 ** write purposes
2911 */
2912 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2913 {
2914         int ret = 1;
2915         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2916
2917         lock_buffer(bh);
2918         spin_lock(&j->j_dirty_buffers_lock);
2919         if (!buffer_mapped(bh)) {
2920                 goto free_jh;
2921         }
2922         /* the page is locked, and the only places that log a data buffer
2923          * also lock the page.
2924          */
2925         if (reiserfs_file_data_log(inode)) {
2926                 /*
2927                  * very conservative, leave the buffer pinned if
2928                  * anyone might need it.
2929                  */
2930                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2931                         ret = 0;
2932                 }
2933         } else  if (buffer_dirty(bh)) {
2934                 struct reiserfs_journal_list *jl;
2935                 struct reiserfs_jh *jh = bh->b_private;
2936
2937                 /* why is this safe?
2938                  * reiserfs_setattr updates i_size in the on disk
2939                  * stat data before allowing vmtruncate to be called.
2940                  *
2941                  * If buffer was put onto the ordered list for this
2942                  * transaction, we know for sure either this transaction
2943                  * or an older one already has updated i_size on disk,
2944                  * and this ordered data won't be referenced in the file
2945                  * if we crash.
2946                  *
2947                  * if the buffer was put onto the ordered list for an older
2948                  * transaction, we need to leave it around
2949                  */
2950                 if (jh && (jl = jh->jl)
2951                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2952                         ret = 0;
2953         }
2954       free_jh:
2955         if (ret && bh->b_private) {
2956                 reiserfs_free_jh(bh);
2957         }
2958         spin_unlock(&j->j_dirty_buffers_lock);
2959         unlock_buffer(bh);
2960         return ret;
2961 }
2962
2963 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2964 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2965 {
2966         struct buffer_head *head, *bh, *next;
2967         struct inode *inode = page->mapping->host;
2968         unsigned int curr_off = 0;
2969         int ret = 1;
2970
2971         BUG_ON(!PageLocked(page));
2972
2973         if (offset == 0)
2974                 ClearPageChecked(page);
2975
2976         if (!page_has_buffers(page))
2977                 goto out;
2978
2979         head = page_buffers(page);
2980         bh = head;
2981         do {
2982                 unsigned int next_off = curr_off + bh->b_size;
2983                 next = bh->b_this_page;
2984
2985                 /*
2986                  * is this block fully invalidated?
2987                  */
2988                 if (offset <= curr_off) {
2989                         if (invalidatepage_can_drop(inode, bh))
2990                                 reiserfs_unmap_buffer(bh);
2991                         else
2992                                 ret = 0;
2993                 }
2994                 curr_off = next_off;
2995                 bh = next;
2996         } while (bh != head);
2997
2998         /*
2999          * We release buffers only if the entire page is being invalidated.
3000          * The get_block cached value has been unconditionally invalidated,
3001          * so real IO is not possible anymore.
3002          */
3003         if (!offset && ret) {
3004                 ret = try_to_release_page(page, 0);
3005                 /* maybe should BUG_ON(!ret); - neilb */
3006         }
3007       out:
3008         return;
3009 }
3010
3011 static int reiserfs_set_page_dirty(struct page *page)
3012 {
3013         struct inode *inode = page->mapping->host;
3014         if (reiserfs_file_data_log(inode)) {
3015                 SetPageChecked(page);
3016                 return __set_page_dirty_nobuffers(page);
3017         }
3018         return __set_page_dirty_buffers(page);
3019 }
3020
3021 /*
3022  * Returns 1 if the page's buffers were dropped.  The page is locked.
3023  *
3024  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3025  * in the buffers at page_buffers(page).
3026  *
3027  * even in -o notail mode, we can't be sure an old mount without -o notail
3028  * didn't create files with tails.
3029  */
3030 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3031 {
3032         struct inode *inode = page->mapping->host;
3033         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3034         struct buffer_head *head;
3035         struct buffer_head *bh;
3036         int ret = 1;
3037
3038         WARN_ON(PageChecked(page));
3039         spin_lock(&j->j_dirty_buffers_lock);
3040         head = page_buffers(page);
3041         bh = head;
3042         do {
3043                 if (bh->b_private) {
3044                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3045                                 reiserfs_free_jh(bh);
3046                         } else {
3047                                 ret = 0;
3048                                 break;
3049                         }
3050                 }
3051                 bh = bh->b_this_page;
3052         } while (bh != head);
3053         if (ret)
3054                 ret = try_to_free_buffers(page);
3055         spin_unlock(&j->j_dirty_buffers_lock);
3056         return ret;
3057 }
3058
3059 /* We thank Mingming Cao for helping us understand in great detail what
3060    to do in this section of the code. */
3061 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3062                                   const struct iovec *iov, loff_t offset,
3063                                   unsigned long nr_segs)
3064 {
3065         struct file *file = iocb->ki_filp;
3066         struct inode *inode = file->f_mapping->host;
3067         ssize_t ret;
3068
3069         ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3070                                   offset, nr_segs,
3071                                   reiserfs_get_blocks_direct_io, NULL);
3072
3073         /*
3074          * In case of error extending write may have instantiated a few
3075          * blocks outside i_size. Trim these off again.
3076          */
3077         if (unlikely((rw & WRITE) && ret < 0)) {
3078                 loff_t isize = i_size_read(inode);
3079                 loff_t end = offset + iov_length(iov, nr_segs);
3080
3081                 if (end > isize)
3082                         vmtruncate(inode, isize);
3083         }
3084
3085         return ret;
3086 }
3087
3088 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3089 {
3090         struct inode *inode = dentry->d_inode;
3091         unsigned int ia_valid;
3092         int depth;
3093         int error;
3094
3095         error = inode_change_ok(inode, attr);
3096         if (error)
3097                 return error;
3098
3099         /* must be turned off for recursive notify_change calls */
3100         ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3101
3102         depth = reiserfs_write_lock_once(inode->i_sb);
3103         if (is_quota_modification(inode, attr))
3104                 dquot_initialize(inode);
3105
3106         if (attr->ia_valid & ATTR_SIZE) {
3107                 /* version 2 items will be caught by the s_maxbytes check
3108                  ** done for us in vmtruncate
3109                  */
3110                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3111                     attr->ia_size > MAX_NON_LFS) {
3112                         error = -EFBIG;
3113                         goto out;
3114                 }
3115                 /* fill in hole pointers in the expanding truncate case. */
3116                 if (attr->ia_size > inode->i_size) {
3117                         error = generic_cont_expand_simple(inode, attr->ia_size);
3118                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
3119                                 int err;
3120                                 struct reiserfs_transaction_handle th;
3121                                 /* we're changing at most 2 bitmaps, inode + super */
3122                                 err = journal_begin(&th, inode->i_sb, 4);
3123                                 if (!err) {
3124                                         reiserfs_discard_prealloc(&th, inode);
3125                                         err = journal_end(&th, inode->i_sb, 4);
3126                                 }
3127                                 if (err)
3128                                         error = err;
3129                         }
3130                         if (error)
3131                                 goto out;
3132                         /*
3133                          * file size is changed, ctime and mtime are
3134                          * to be updated
3135                          */
3136                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3137                 }
3138         }
3139
3140         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3141              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3142             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3143                 /* stat data of format v3.5 has 16 bit uid and gid */
3144                 error = -EINVAL;
3145                 goto out;
3146         }
3147
3148         if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3149             (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3150                 struct reiserfs_transaction_handle th;
3151                 int jbegin_count =
3152                     2 *
3153                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3154                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3155                     2;
3156
3157                 error = reiserfs_chown_xattrs(inode, attr);
3158
3159                 if (error)
3160                         return error;
3161
3162                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3163                 error = journal_begin(&th, inode->i_sb, jbegin_count);
3164                 if (error)
3165                         goto out;
3166                 error = dquot_transfer(inode, attr);
3167                 if (error) {
3168                         journal_end(&th, inode->i_sb, jbegin_count);
3169                         goto out;
3170                 }
3171
3172                 /* Update corresponding info in inode so that everything is in
3173                  * one transaction */
3174                 if (attr->ia_valid & ATTR_UID)
3175                         inode->i_uid = attr->ia_uid;
3176                 if (attr->ia_valid & ATTR_GID)
3177                         inode->i_gid = attr->ia_gid;
3178                 mark_inode_dirty(inode);
3179                 error = journal_end(&th, inode->i_sb, jbegin_count);
3180                 if (error)
3181                         goto out;
3182         }
3183
3184         /*
3185          * Relax the lock here, as it might truncate the
3186          * inode pages and wait for inode pages locks.
3187          * To release such page lock, the owner needs the
3188          * reiserfs lock
3189          */
3190         reiserfs_write_unlock_once(inode->i_sb, depth);
3191         if ((attr->ia_valid & ATTR_SIZE) &&
3192             attr->ia_size != i_size_read(inode))
3193                 error = vmtruncate(inode, attr->ia_size);
3194
3195         if (!error) {
3196                 setattr_copy(inode, attr);
3197                 mark_inode_dirty(inode);
3198         }
3199         depth = reiserfs_write_lock_once(inode->i_sb);
3200
3201         if (!error && reiserfs_posixacl(inode->i_sb)) {
3202                 if (attr->ia_valid & ATTR_MODE)
3203                         error = reiserfs_acl_chmod(inode);
3204         }
3205
3206       out:
3207         reiserfs_write_unlock_once(inode->i_sb, depth);
3208
3209         return error;
3210 }
3211
3212 const struct address_space_operations reiserfs_address_space_operations = {
3213         .writepage = reiserfs_writepage,
3214         .readpage = reiserfs_readpage,
3215         .readpages = reiserfs_readpages,
3216         .releasepage = reiserfs_releasepage,
3217         .invalidatepage = reiserfs_invalidatepage,
3218         .sync_page = block_sync_page,
3219         .write_begin = reiserfs_write_begin,
3220         .write_end = reiserfs_write_end,
3221         .bmap = reiserfs_aop_bmap,
3222         .direct_IO = reiserfs_direct_IO,
3223         .set_page_dirty = reiserfs_set_page_dirty,
3224 };