Merge branch 'linux-2.6'
[linux-2.6] / 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/smp_lock.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <asm/uaccess.h>
14 #include <asm/unaligned.h>
15 #include <linux/buffer_head.h>
16 #include <linux/mpage.h>
17 #include <linux/writeback.h>
18 #include <linux/quotaops.h>
19
20 static int reiserfs_commit_write(struct file *f, struct page *page,
21                                  unsigned from, unsigned to);
22 static int reiserfs_prepare_write(struct file *f, struct page *page,
23                                   unsigned from, unsigned to);
24
25 void reiserfs_delete_inode(struct inode *inode)
26 {
27         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
28         int jbegin_count =
29             JOURNAL_PER_BALANCE_CNT * 2 +
30             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
31         struct reiserfs_transaction_handle th;
32         int err;
33
34         truncate_inode_pages(&inode->i_data, 0);
35
36         reiserfs_write_lock(inode->i_sb);
37
38         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
39         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
40                 reiserfs_delete_xattrs(inode);
41
42                 if (journal_begin(&th, inode->i_sb, jbegin_count))
43                         goto out;
44                 reiserfs_update_inode_transaction(inode);
45
46                 err = reiserfs_delete_object(&th, inode);
47
48                 /* Do quota update inside a transaction for journaled quotas. We must do that
49                  * after delete_object so that quota updates go into the same transaction as
50                  * stat data deletion */
51                 if (!err) 
52                         DQUOT_FREE_INODE(inode);
53
54                 if (journal_end(&th, inode->i_sb, jbegin_count))
55                         goto out;
56
57                 /* check return value from reiserfs_delete_object after
58                  * ending the transaction
59                  */
60                 if (err)
61                     goto out;
62
63                 /* all items of file are deleted, so we can remove "save" link */
64                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
65                                                                  * about an error here */
66         } else {
67                 /* no object items are in the tree */
68                 ;
69         }
70       out:
71         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
72         inode->i_blocks = 0;
73         reiserfs_write_unlock(inode->i_sb);
74 }
75
76 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
77                           __u32 objectid, loff_t offset, int type, int length)
78 {
79         key->version = version;
80
81         key->on_disk_key.k_dir_id = dirid;
82         key->on_disk_key.k_objectid = objectid;
83         set_cpu_key_k_offset(key, offset);
84         set_cpu_key_k_type(key, type);
85         key->key_length = length;
86 }
87
88 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
89    offset and type of key */
90 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
91                   int type, int length)
92 {
93         _make_cpu_key(key, get_inode_item_key_version(inode),
94                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
95                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
96                       length);
97 }
98
99 //
100 // when key is 0, do not set version and short key
101 //
102 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
103                               int version,
104                               loff_t offset, int type, int length,
105                               int entry_count /*or ih_free_space */ )
106 {
107         if (key) {
108                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
109                 ih->ih_key.k_objectid =
110                     cpu_to_le32(key->on_disk_key.k_objectid);
111         }
112         put_ih_version(ih, version);
113         set_le_ih_k_offset(ih, offset);
114         set_le_ih_k_type(ih, type);
115         put_ih_item_len(ih, length);
116         /*    set_ih_free_space (ih, 0); */
117         // for directory items it is entry count, for directs and stat
118         // datas - 0xffff, for indirects - 0
119         put_ih_entry_count(ih, entry_count);
120 }
121
122 //
123 // FIXME: we might cache recently accessed indirect item
124
125 // Ugh.  Not too eager for that....
126 //  I cut the code until such time as I see a convincing argument (benchmark).
127 // I don't want a bloated inode struct..., and I don't like code complexity....
128
129 /* cutting the code is fine, since it really isn't in use yet and is easy
130 ** to add back in.  But, Vladimir has a really good idea here.  Think
131 ** about what happens for reading a file.  For each page,
132 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
133 ** an indirect item.  This indirect item has X number of pointers, where
134 ** X is a big number if we've done the block allocation right.  But,
135 ** we only use one or two of these pointers during each call to readpage,
136 ** needlessly researching again later on.
137 **
138 ** The size of the cache could be dynamic based on the size of the file.
139 **
140 ** I'd also like to see us cache the location the stat data item, since
141 ** we are needlessly researching for that frequently.
142 **
143 ** --chris
144 */
145
146 /* If this page has a file tail in it, and
147 ** it was read in by get_block_create_0, the page data is valid,
148 ** but tail is still sitting in a direct item, and we can't write to
149 ** it.  So, look through this page, and check all the mapped buffers
150 ** to make sure they have valid block numbers.  Any that don't need
151 ** to be unmapped, so that block_prepare_write will correctly call
152 ** reiserfs_get_block to convert the tail into an unformatted node
153 */
154 static inline void fix_tail_page_for_writing(struct page *page)
155 {
156         struct buffer_head *head, *next, *bh;
157
158         if (page && page_has_buffers(page)) {
159                 head = page_buffers(page);
160                 bh = head;
161                 do {
162                         next = bh->b_this_page;
163                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
164                                 reiserfs_unmap_buffer(bh);
165                         }
166                         bh = next;
167                 } while (bh != head);
168         }
169 }
170
171 /* reiserfs_get_block does not need to allocate a block only if it has been
172    done already or non-hole position has been found in the indirect item */
173 static inline int allocation_needed(int retval, b_blocknr_t allocated,
174                                     struct item_head *ih,
175                                     __le32 * item, int pos_in_item)
176 {
177         if (allocated)
178                 return 0;
179         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
180             get_block_num(item, pos_in_item))
181                 return 0;
182         return 1;
183 }
184
185 static inline int indirect_item_found(int retval, struct item_head *ih)
186 {
187         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
188 }
189
190 static inline void set_block_dev_mapped(struct buffer_head *bh,
191                                         b_blocknr_t block, struct inode *inode)
192 {
193         map_bh(bh, inode->i_sb, block);
194 }
195
196 //
197 // files which were created in the earlier version can not be longer,
198 // than 2 gb
199 //
200 static int file_capable(struct inode *inode, long block)
201 {
202         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
203             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
204                 return 1;
205
206         return 0;
207 }
208
209 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
210                                    struct inode *inode, struct treepath *path)
211 {
212         struct super_block *s = th->t_super;
213         int len = th->t_blocks_allocated;
214         int err;
215
216         BUG_ON(!th->t_trans_id);
217         BUG_ON(!th->t_refcount);
218
219         pathrelse(path);
220
221         /* we cannot restart while nested */
222         if (th->t_refcount > 1) {
223                 return 0;
224         }
225         reiserfs_update_sd(th, inode);
226         err = journal_end(th, s, len);
227         if (!err) {
228                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
229                 if (!err)
230                         reiserfs_update_inode_transaction(inode);
231         }
232         return err;
233 }
234
235 // it is called by get_block when create == 0. Returns block number
236 // for 'block'-th logical block of file. When it hits direct item it
237 // returns 0 (being called from bmap) or read direct item into piece
238 // of page (bh_result)
239
240 // Please improve the english/clarity in the comment above, as it is
241 // hard to understand.
242
243 static int _get_block_create_0(struct inode *inode, long block,
244                                struct buffer_head *bh_result, int args)
245 {
246         INITIALIZE_PATH(path);
247         struct cpu_key key;
248         struct buffer_head *bh;
249         struct item_head *ih, tmp_ih;
250         int fs_gen;
251         int blocknr;
252         char *p = NULL;
253         int chars;
254         int ret;
255         int result;
256         int done = 0;
257         unsigned long offset;
258
259         // prepare the key to look for the 'block'-th block of file
260         make_cpu_key(&key, inode,
261                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
262                      3);
263
264       research:
265         result = search_for_position_by_key(inode->i_sb, &key, &path);
266         if (result != POSITION_FOUND) {
267                 pathrelse(&path);
268                 if (p)
269                         kunmap(bh_result->b_page);
270                 if (result == IO_ERROR)
271                         return -EIO;
272                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
273                 // That there is some MMAPED data associated with it that is yet to be written to disk.
274                 if ((args & GET_BLOCK_NO_HOLE)
275                     && !PageUptodate(bh_result->b_page)) {
276                         return -ENOENT;
277                 }
278                 return 0;
279         }
280         //
281         bh = get_last_bh(&path);
282         ih = get_ih(&path);
283         if (is_indirect_le_ih(ih)) {
284                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
285
286                 /* FIXME: here we could cache indirect item or part of it in
287                    the inode to avoid search_by_key in case of subsequent
288                    access to file */
289                 blocknr = get_block_num(ind_item, path.pos_in_item);
290                 ret = 0;
291                 if (blocknr) {
292                         map_bh(bh_result, inode->i_sb, blocknr);
293                         if (path.pos_in_item ==
294                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
295                                 set_buffer_boundary(bh_result);
296                         }
297                 } else
298                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
299                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
300                 if ((args & GET_BLOCK_NO_HOLE)
301                             && !PageUptodate(bh_result->b_page)) {
302                         ret = -ENOENT;
303                 }
304
305                 pathrelse(&path);
306                 if (p)
307                         kunmap(bh_result->b_page);
308                 return ret;
309         }
310         // requested data are in direct item(s)
311         if (!(args & GET_BLOCK_READ_DIRECT)) {
312                 // we are called by bmap. FIXME: we can not map block of file
313                 // when it is stored in direct item(s)
314                 pathrelse(&path);
315                 if (p)
316                         kunmap(bh_result->b_page);
317                 return -ENOENT;
318         }
319
320         /* if we've got a direct item, and the buffer or page was uptodate,
321          ** we don't want to pull data off disk again.  skip to the
322          ** end, where we map the buffer and return
323          */
324         if (buffer_uptodate(bh_result)) {
325                 goto finished;
326         } else
327                 /*
328                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
329                  ** pages without any buffers.  If the page is up to date, we don't want
330                  ** read old data off disk.  Set the up to date bit on the buffer instead
331                  ** and jump to the end
332                  */
333         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
334                 set_buffer_uptodate(bh_result);
335                 goto finished;
336         }
337         // read file tail into part of page
338         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
339         fs_gen = get_generation(inode->i_sb);
340         copy_item_head(&tmp_ih, ih);
341
342         /* we only want to kmap if we are reading the tail into the page.
343          ** this is not the common case, so we don't kmap until we are
344          ** sure we need to.  But, this means the item might move if
345          ** kmap schedules
346          */
347         if (!p) {
348                 p = (char *)kmap(bh_result->b_page);
349                 if (fs_changed(fs_gen, inode->i_sb)
350                     && item_moved(&tmp_ih, &path)) {
351                         goto research;
352                 }
353         }
354         p += offset;
355         memset(p, 0, inode->i_sb->s_blocksize);
356         do {
357                 if (!is_direct_le_ih(ih)) {
358                         BUG();
359                 }
360                 /* make sure we don't read more bytes than actually exist in
361                  ** the file.  This can happen in odd cases where i_size isn't
362                  ** correct, and when direct item padding results in a few 
363                  ** extra bytes at the end of the direct item
364                  */
365                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
366                         break;
367                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
368                         chars =
369                             inode->i_size - (le_ih_k_offset(ih) - 1) -
370                             path.pos_in_item;
371                         done = 1;
372                 } else {
373                         chars = ih_item_len(ih) - path.pos_in_item;
374                 }
375                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
376
377                 if (done)
378                         break;
379
380                 p += chars;
381
382                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
383                         // we done, if read direct item is not the last item of
384                         // node FIXME: we could try to check right delimiting key
385                         // to see whether direct item continues in the right
386                         // neighbor or rely on i_size
387                         break;
388
389                 // update key to look for the next piece
390                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
391                 result = search_for_position_by_key(inode->i_sb, &key, &path);
392                 if (result != POSITION_FOUND)
393                         // i/o error most likely
394                         break;
395                 bh = get_last_bh(&path);
396                 ih = get_ih(&path);
397         } while (1);
398
399         flush_dcache_page(bh_result->b_page);
400         kunmap(bh_result->b_page);
401
402       finished:
403         pathrelse(&path);
404
405         if (result == IO_ERROR)
406                 return -EIO;
407
408         /* this buffer has valid data, but isn't valid for io.  mapping it to
409          * block #0 tells the rest of reiserfs it just has a tail in it
410          */
411         map_bh(bh_result, inode->i_sb, 0);
412         set_buffer_uptodate(bh_result);
413         return 0;
414 }
415
416 // this is called to create file map. So, _get_block_create_0 will not
417 // read direct item
418 static int reiserfs_bmap(struct inode *inode, sector_t block,
419                          struct buffer_head *bh_result, int create)
420 {
421         if (!file_capable(inode, block))
422                 return -EFBIG;
423
424         reiserfs_write_lock(inode->i_sb);
425         /* do not read the direct item */
426         _get_block_create_0(inode, block, bh_result, 0);
427         reiserfs_write_unlock(inode->i_sb);
428         return 0;
429 }
430
431 /* special version of get_block that is only used by grab_tail_page right
432 ** now.  It is sent to block_prepare_write, and when you try to get a
433 ** block past the end of the file (or a block from a hole) it returns
434 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
435 ** be able to do i/o on the buffers returned, unless an error value
436 ** is also returned.
437 ** 
438 ** So, this allows block_prepare_write to be used for reading a single block
439 ** in a page.  Where it does not produce a valid page for holes, or past the
440 ** end of the file.  This turns out to be exactly what we need for reading
441 ** tails for conversion.
442 **
443 ** The point of the wrapper is forcing a certain value for create, even
444 ** though the VFS layer is calling this function with create==1.  If you 
445 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block, 
446 ** don't use this function.
447 */
448 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
449                                        struct buffer_head *bh_result,
450                                        int create)
451 {
452         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
453 }
454
455 /* This is special helper for reiserfs_get_block in case we are executing
456    direct_IO request. */
457 static int reiserfs_get_blocks_direct_io(struct inode *inode,
458                                          sector_t iblock,
459                                          struct buffer_head *bh_result,
460                                          int create)
461 {
462         int ret;
463
464         bh_result->b_page = NULL;
465
466         /* We set the b_size before reiserfs_get_block call since it is
467            referenced in convert_tail_for_hole() that may be called from
468            reiserfs_get_block() */
469         bh_result->b_size = (1 << inode->i_blkbits);
470
471         ret = reiserfs_get_block(inode, iblock, bh_result,
472                                  create | GET_BLOCK_NO_DANGLE);
473         if (ret)
474                 goto out;
475
476         /* don't allow direct io onto tail pages */
477         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
478                 /* make sure future calls to the direct io funcs for this offset
479                  ** in the file fail by unmapping the buffer
480                  */
481                 clear_buffer_mapped(bh_result);
482                 ret = -EINVAL;
483         }
484         /* Possible unpacked tail. Flush the data before pages have
485            disappeared */
486         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
487                 int err;
488                 lock_kernel();
489                 err = reiserfs_commit_for_inode(inode);
490                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
491                 unlock_kernel();
492                 if (err < 0)
493                         ret = err;
494         }
495       out:
496         return ret;
497 }
498
499 /*
500 ** helper function for when reiserfs_get_block is called for a hole
501 ** but the file tail is still in a direct item
502 ** bh_result is the buffer head for the hole
503 ** tail_offset is the offset of the start of the tail in the file
504 **
505 ** This calls prepare_write, which will start a new transaction
506 ** you should not be in a transaction, or have any paths held when you
507 ** call this.
508 */
509 static int convert_tail_for_hole(struct inode *inode,
510                                  struct buffer_head *bh_result,
511                                  loff_t tail_offset)
512 {
513         unsigned long index;
514         unsigned long tail_end;
515         unsigned long tail_start;
516         struct page *tail_page;
517         struct page *hole_page = bh_result->b_page;
518         int retval = 0;
519
520         if ((tail_offset & (bh_result->b_size - 1)) != 1)
521                 return -EIO;
522
523         /* always try to read until the end of the block */
524         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
525         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
526
527         index = tail_offset >> PAGE_CACHE_SHIFT;
528         /* hole_page can be zero in case of direct_io, we are sure
529            that we cannot get here if we write with O_DIRECT into
530            tail page */
531         if (!hole_page || index != hole_page->index) {
532                 tail_page = grab_cache_page(inode->i_mapping, index);
533                 retval = -ENOMEM;
534                 if (!tail_page) {
535                         goto out;
536                 }
537         } else {
538                 tail_page = hole_page;
539         }
540
541         /* we don't have to make sure the conversion did not happen while
542          ** we were locking the page because anyone that could convert
543          ** must first take i_mutex.
544          **
545          ** We must fix the tail page for writing because it might have buffers
546          ** that are mapped, but have a block number of 0.  This indicates tail
547          ** data that has been read directly into the page, and block_prepare_write
548          ** won't trigger a get_block in this case.
549          */
550         fix_tail_page_for_writing(tail_page);
551         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
552         if (retval)
553                 goto unlock;
554
555         /* tail conversion might change the data in the page */
556         flush_dcache_page(tail_page);
557
558         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
559
560       unlock:
561         if (tail_page != hole_page) {
562                 unlock_page(tail_page);
563                 page_cache_release(tail_page);
564         }
565       out:
566         return retval;
567 }
568
569 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
570                                   long block,
571                                   struct inode *inode,
572                                   b_blocknr_t * allocated_block_nr,
573                                   struct treepath *path, int flags)
574 {
575         BUG_ON(!th->t_trans_id);
576
577 #ifdef REISERFS_PREALLOCATE
578         if (!(flags & GET_BLOCK_NO_IMUX)) {
579                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
580                                                   path, block);
581         }
582 #endif
583         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
584                                          block);
585 }
586
587 int reiserfs_get_block(struct inode *inode, sector_t block,
588                        struct buffer_head *bh_result, int create)
589 {
590         int repeat, retval = 0;
591         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
592         INITIALIZE_PATH(path);
593         int pos_in_item;
594         struct cpu_key key;
595         struct buffer_head *bh, *unbh = NULL;
596         struct item_head *ih, tmp_ih;
597         __le32 *item;
598         int done;
599         int fs_gen;
600         struct reiserfs_transaction_handle *th = NULL;
601         /* space reserved in transaction batch: 
602            . 3 balancings in direct->indirect conversion
603            . 1 block involved into reiserfs_update_sd()
604            XXX in practically impossible worst case direct2indirect()
605            can incur (much) more than 3 balancings.
606            quota update for user, group */
607         int jbegin_count =
608             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
609             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
610         int version;
611         int dangle = 1;
612         loff_t new_offset =
613             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
614
615         /* bad.... */
616         reiserfs_write_lock(inode->i_sb);
617         version = get_inode_item_key_version(inode);
618
619         if (!file_capable(inode, block)) {
620                 reiserfs_write_unlock(inode->i_sb);
621                 return -EFBIG;
622         }
623
624         /* if !create, we aren't changing the FS, so we don't need to
625          ** log anything, so we don't need to start a transaction
626          */
627         if (!(create & GET_BLOCK_CREATE)) {
628                 int ret;
629                 /* find number of block-th logical block of the file */
630                 ret = _get_block_create_0(inode, block, bh_result,
631                                           create | GET_BLOCK_READ_DIRECT);
632                 reiserfs_write_unlock(inode->i_sb);
633                 return ret;
634         }
635         /*
636          * if we're already in a transaction, make sure to close
637          * any new transactions we start in this func
638          */
639         if ((create & GET_BLOCK_NO_DANGLE) ||
640             reiserfs_transaction_running(inode->i_sb))
641                 dangle = 0;
642
643         /* If file is of such a size, that it might have a tail and tails are enabled
644          ** we should mark it as possibly needing tail packing on close
645          */
646         if ((have_large_tails(inode->i_sb)
647              && inode->i_size < i_block_size(inode) * 4)
648             || (have_small_tails(inode->i_sb)
649                 && inode->i_size < i_block_size(inode)))
650                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
651
652         /* set the key of the first byte in the 'block'-th block of file */
653         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
654         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
655               start_trans:
656                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
657                 if (!th) {
658                         retval = -ENOMEM;
659                         goto failure;
660                 }
661                 reiserfs_update_inode_transaction(inode);
662         }
663       research:
664
665         retval = search_for_position_by_key(inode->i_sb, &key, &path);
666         if (retval == IO_ERROR) {
667                 retval = -EIO;
668                 goto failure;
669         }
670
671         bh = get_last_bh(&path);
672         ih = get_ih(&path);
673         item = get_item(&path);
674         pos_in_item = path.pos_in_item;
675
676         fs_gen = get_generation(inode->i_sb);
677         copy_item_head(&tmp_ih, ih);
678
679         if (allocation_needed
680             (retval, allocated_block_nr, ih, item, pos_in_item)) {
681                 /* we have to allocate block for the unformatted node */
682                 if (!th) {
683                         pathrelse(&path);
684                         goto start_trans;
685                 }
686
687                 repeat =
688                     _allocate_block(th, block, inode, &allocated_block_nr,
689                                     &path, create);
690
691                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
692                         /* restart the transaction to give the journal a chance to free
693                          ** some blocks.  releases the path, so we have to go back to
694                          ** research if we succeed on the second try
695                          */
696                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
697                         retval = restart_transaction(th, inode, &path);
698                         if (retval)
699                                 goto failure;
700                         repeat =
701                             _allocate_block(th, block, inode,
702                                             &allocated_block_nr, NULL, create);
703
704                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
705                                 goto research;
706                         }
707                         if (repeat == QUOTA_EXCEEDED)
708                                 retval = -EDQUOT;
709                         else
710                                 retval = -ENOSPC;
711                         goto failure;
712                 }
713
714                 if (fs_changed(fs_gen, inode->i_sb)
715                     && item_moved(&tmp_ih, &path)) {
716                         goto research;
717                 }
718         }
719
720         if (indirect_item_found(retval, ih)) {
721                 b_blocknr_t unfm_ptr;
722                 /* 'block'-th block is in the file already (there is
723                    corresponding cell in some indirect item). But it may be
724                    zero unformatted node pointer (hole) */
725                 unfm_ptr = get_block_num(item, pos_in_item);
726                 if (unfm_ptr == 0) {
727                         /* use allocated block to plug the hole */
728                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
729                         if (fs_changed(fs_gen, inode->i_sb)
730                             && item_moved(&tmp_ih, &path)) {
731                                 reiserfs_restore_prepared_buffer(inode->i_sb,
732                                                                  bh);
733                                 goto research;
734                         }
735                         set_buffer_new(bh_result);
736                         if (buffer_dirty(bh_result)
737                             && reiserfs_data_ordered(inode->i_sb))
738                                 reiserfs_add_ordered_list(inode, bh_result);
739                         put_block_num(item, pos_in_item, allocated_block_nr);
740                         unfm_ptr = allocated_block_nr;
741                         journal_mark_dirty(th, inode->i_sb, bh);
742                         reiserfs_update_sd(th, inode);
743                 }
744                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
745                 pathrelse(&path);
746                 retval = 0;
747                 if (!dangle && th)
748                         retval = reiserfs_end_persistent_transaction(th);
749
750                 reiserfs_write_unlock(inode->i_sb);
751
752                 /* the item was found, so new blocks were not added to the file
753                  ** there is no need to make sure the inode is updated with this 
754                  ** transaction
755                  */
756                 return retval;
757         }
758
759         if (!th) {
760                 pathrelse(&path);
761                 goto start_trans;
762         }
763
764         /* desired position is not found or is in the direct item. We have
765            to append file with holes up to 'block'-th block converting
766            direct items to indirect one if necessary */
767         done = 0;
768         do {
769                 if (is_statdata_le_ih(ih)) {
770                         __le32 unp = 0;
771                         struct cpu_key tmp_key;
772
773                         /* indirect item has to be inserted */
774                         make_le_item_head(&tmp_ih, &key, version, 1,
775                                           TYPE_INDIRECT, UNFM_P_SIZE,
776                                           0 /* free_space */ );
777
778                         if (cpu_key_k_offset(&key) == 1) {
779                                 /* we are going to add 'block'-th block to the file. Use
780                                    allocated block for that */
781                                 unp = cpu_to_le32(allocated_block_nr);
782                                 set_block_dev_mapped(bh_result,
783                                                      allocated_block_nr, inode);
784                                 set_buffer_new(bh_result);
785                                 done = 1;
786                         }
787                         tmp_key = key;  // ;)
788                         set_cpu_key_k_offset(&tmp_key, 1);
789                         PATH_LAST_POSITION(&path)++;
790
791                         retval =
792                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
793                                                  inode, (char *)&unp);
794                         if (retval) {
795                                 reiserfs_free_block(th, inode,
796                                                     allocated_block_nr, 1);
797                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
798                         }
799                         //mark_tail_converted (inode);
800                 } else if (is_direct_le_ih(ih)) {
801                         /* direct item has to be converted */
802                         loff_t tail_offset;
803
804                         tail_offset =
805                             ((le_ih_k_offset(ih) -
806                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
807                         if (tail_offset == cpu_key_k_offset(&key)) {
808                                 /* direct item we just found fits into block we have
809                                    to map. Convert it into unformatted node: use
810                                    bh_result for the conversion */
811                                 set_block_dev_mapped(bh_result,
812                                                      allocated_block_nr, inode);
813                                 unbh = bh_result;
814                                 done = 1;
815                         } else {
816                                 /* we have to padd file tail stored in direct item(s)
817                                    up to block size and convert it to unformatted
818                                    node. FIXME: this should also get into page cache */
819
820                                 pathrelse(&path);
821                                 /*
822                                  * ugly, but we can only end the transaction if
823                                  * we aren't nested
824                                  */
825                                 BUG_ON(!th->t_refcount);
826                                 if (th->t_refcount == 1) {
827                                         retval =
828                                             reiserfs_end_persistent_transaction
829                                             (th);
830                                         th = NULL;
831                                         if (retval)
832                                                 goto failure;
833                                 }
834
835                                 retval =
836                                     convert_tail_for_hole(inode, bh_result,
837                                                           tail_offset);
838                                 if (retval) {
839                                         if (retval != -ENOSPC)
840                                                 reiserfs_warning(inode->i_sb,
841                                                                  "clm-6004: convert tail failed inode %lu, error %d",
842                                                                  inode->i_ino,
843                                                                  retval);
844                                         if (allocated_block_nr) {
845                                                 /* the bitmap, the super, and the stat data == 3 */
846                                                 if (!th)
847                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
848                                                 if (th)
849                                                         reiserfs_free_block(th,
850                                                                             inode,
851                                                                             allocated_block_nr,
852                                                                             1);
853                                         }
854                                         goto failure;
855                                 }
856                                 goto research;
857                         }
858                         retval =
859                             direct2indirect(th, inode, &path, unbh,
860                                             tail_offset);
861                         if (retval) {
862                                 reiserfs_unmap_buffer(unbh);
863                                 reiserfs_free_block(th, inode,
864                                                     allocated_block_nr, 1);
865                                 goto failure;
866                         }
867                         /* it is important the set_buffer_uptodate is done after
868                          ** the direct2indirect.  The buffer might contain valid
869                          ** data newer than the data on disk (read by readpage, changed,
870                          ** and then sent here by writepage).  direct2indirect needs
871                          ** to know if unbh was already up to date, so it can decide
872                          ** if the data in unbh needs to be replaced with data from
873                          ** the disk
874                          */
875                         set_buffer_uptodate(unbh);
876
877                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
878                            buffer will disappear shortly, so it should not be added to
879                          */
880                         if (unbh->b_page) {
881                                 /* we've converted the tail, so we must
882                                  ** flush unbh before the transaction commits
883                                  */
884                                 reiserfs_add_tail_list(inode, unbh);
885
886                                 /* mark it dirty now to prevent commit_write from adding
887                                  ** this buffer to the inode's dirty buffer list
888                                  */
889                                 /*
890                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
891                                  * It's still atomic, but it sets the page dirty too,
892                                  * which makes it eligible for writeback at any time by the
893                                  * VM (which was also the case with __mark_buffer_dirty())
894                                  */
895                                 mark_buffer_dirty(unbh);
896                         }
897                 } else {
898                         /* append indirect item with holes if needed, when appending
899                            pointer to 'block'-th block use block, which is already
900                            allocated */
901                         struct cpu_key tmp_key;
902                         unp_t unf_single = 0;   // We use this in case we need to allocate only
903                         // one block which is a fastpath
904                         unp_t *un;
905                         __u64 max_to_insert =
906                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
907                             UNFM_P_SIZE;
908                         __u64 blocks_needed;
909
910                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
911                                "vs-804: invalid position for append");
912                         /* indirect item has to be appended, set up key of that position */
913                         make_cpu_key(&tmp_key, inode,
914                                      le_key_k_offset(version,
915                                                      &(ih->ih_key)) +
916                                      op_bytes_number(ih,
917                                                      inode->i_sb->s_blocksize),
918                                      //pos_in_item * inode->i_sb->s_blocksize,
919                                      TYPE_INDIRECT, 3); // key type is unimportant
920
921                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
922                                "green-805: invalid offset");
923                         blocks_needed =
924                             1 +
925                             ((cpu_key_k_offset(&key) -
926                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
927                              s_blocksize_bits);
928
929                         if (blocks_needed == 1) {
930                                 un = &unf_single;
931                         } else {
932                                 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
933                                 if (!un) {
934                                         un = &unf_single;
935                                         blocks_needed = 1;
936                                         max_to_insert = 0;
937                                 }
938                         }
939                         if (blocks_needed <= max_to_insert) {
940                                 /* we are going to add target block to the file. Use allocated
941                                    block for that */
942                                 un[blocks_needed - 1] =
943                                     cpu_to_le32(allocated_block_nr);
944                                 set_block_dev_mapped(bh_result,
945                                                      allocated_block_nr, inode);
946                                 set_buffer_new(bh_result);
947                                 done = 1;
948                         } else {
949                                 /* paste hole to the indirect item */
950                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
951                                    only have space for one block */
952                                 blocks_needed =
953                                     max_to_insert ? max_to_insert : 1;
954                         }
955                         retval =
956                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
957                                                      (char *)un,
958                                                      UNFM_P_SIZE *
959                                                      blocks_needed);
960
961                         if (blocks_needed != 1)
962                                 kfree(un);
963
964                         if (retval) {
965                                 reiserfs_free_block(th, inode,
966                                                     allocated_block_nr, 1);
967                                 goto failure;
968                         }
969                         if (!done) {
970                                 /* We need to mark new file size in case this function will be
971                                    interrupted/aborted later on. And we may do this only for
972                                    holes. */
973                                 inode->i_size +=
974                                     inode->i_sb->s_blocksize * blocks_needed;
975                         }
976                 }
977
978                 if (done == 1)
979                         break;
980
981                 /* this loop could log more blocks than we had originally asked
982                  ** for.  So, we have to allow the transaction to end if it is
983                  ** too big or too full.  Update the inode so things are 
984                  ** consistent if we crash before the function returns
985                  **
986                  ** release the path so that anybody waiting on the path before
987                  ** ending their transaction will be able to continue.
988                  */
989                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
990                         retval = restart_transaction(th, inode, &path);
991                         if (retval)
992                                 goto failure;
993                 }
994                 /* inserting indirect pointers for a hole can take a 
995                  ** long time.  reschedule if needed
996                  */
997                 cond_resched();
998
999                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1000                 if (retval == IO_ERROR) {
1001                         retval = -EIO;
1002                         goto failure;
1003                 }
1004                 if (retval == POSITION_FOUND) {
1005                         reiserfs_warning(inode->i_sb,
1006                                          "vs-825: reiserfs_get_block: "
1007                                          "%K should not be found", &key);
1008                         retval = -EEXIST;
1009                         if (allocated_block_nr)
1010                                 reiserfs_free_block(th, inode,
1011                                                     allocated_block_nr, 1);
1012                         pathrelse(&path);
1013                         goto failure;
1014                 }
1015                 bh = get_last_bh(&path);
1016                 ih = get_ih(&path);
1017                 item = get_item(&path);
1018                 pos_in_item = path.pos_in_item;
1019         } while (1);
1020
1021         retval = 0;
1022
1023       failure:
1024         if (th && (!dangle || (retval && !th->t_trans_id))) {
1025                 int err;
1026                 if (th->t_trans_id)
1027                         reiserfs_update_sd(th, inode);
1028                 err = reiserfs_end_persistent_transaction(th);
1029                 if (err)
1030                         retval = err;
1031         }
1032
1033         reiserfs_write_unlock(inode->i_sb);
1034         reiserfs_check_path(&path);
1035         return retval;
1036 }
1037
1038 static int
1039 reiserfs_readpages(struct file *file, struct address_space *mapping,
1040                    struct list_head *pages, unsigned nr_pages)
1041 {
1042         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1043 }
1044
1045 /* Compute real number of used bytes by file
1046  * Following three functions can go away when we'll have enough space in stat item
1047  */
1048 static int real_space_diff(struct inode *inode, int sd_size)
1049 {
1050         int bytes;
1051         loff_t blocksize = inode->i_sb->s_blocksize;
1052
1053         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1054                 return sd_size;
1055
1056         /* End of file is also in full block with indirect reference, so round
1057          ** up to the next block.
1058          **
1059          ** there is just no way to know if the tail is actually packed
1060          ** on the file, so we have to assume it isn't.  When we pack the
1061          ** tail, we add 4 bytes to pretend there really is an unformatted
1062          ** node pointer
1063          */
1064         bytes =
1065             ((inode->i_size +
1066               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1067             sd_size;
1068         return bytes;
1069 }
1070
1071 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1072                                         int sd_size)
1073 {
1074         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1075                 return inode->i_size +
1076                     (loff_t) (real_space_diff(inode, sd_size));
1077         }
1078         return ((loff_t) real_space_diff(inode, sd_size)) +
1079             (((loff_t) blocks) << 9);
1080 }
1081
1082 /* Compute number of blocks used by file in ReiserFS counting */
1083 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1084 {
1085         loff_t bytes = inode_get_bytes(inode);
1086         loff_t real_space = real_space_diff(inode, sd_size);
1087
1088         /* keeps fsck and non-quota versions of reiserfs happy */
1089         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1090                 bytes += (loff_t) 511;
1091         }
1092
1093         /* files from before the quota patch might i_blocks such that
1094          ** bytes < real_space.  Deal with that here to prevent it from
1095          ** going negative.
1096          */
1097         if (bytes < real_space)
1098                 return 0;
1099         return (bytes - real_space) >> 9;
1100 }
1101
1102 //
1103 // BAD: new directories have stat data of new type and all other items
1104 // of old type. Version stored in the inode says about body items, so
1105 // in update_stat_data we can not rely on inode, but have to check
1106 // item version directly
1107 //
1108
1109 // called by read_locked_inode
1110 static void init_inode(struct inode *inode, struct treepath *path)
1111 {
1112         struct buffer_head *bh;
1113         struct item_head *ih;
1114         __u32 rdev;
1115         //int version = ITEM_VERSION_1;
1116
1117         bh = PATH_PLAST_BUFFER(path);
1118         ih = PATH_PITEM_HEAD(path);
1119
1120         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1121
1122         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1123         REISERFS_I(inode)->i_flags = 0;
1124         REISERFS_I(inode)->i_prealloc_block = 0;
1125         REISERFS_I(inode)->i_prealloc_count = 0;
1126         REISERFS_I(inode)->i_trans_id = 0;
1127         REISERFS_I(inode)->i_jl = NULL;
1128         mutex_init(&(REISERFS_I(inode)->i_mmap));
1129         reiserfs_init_acl_access(inode);
1130         reiserfs_init_acl_default(inode);
1131         reiserfs_init_xattr_rwsem(inode);
1132
1133         if (stat_data_v1(ih)) {
1134                 struct stat_data_v1 *sd =
1135                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1136                 unsigned long blocks;
1137
1138                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1139                 set_inode_sd_version(inode, STAT_DATA_V1);
1140                 inode->i_mode = sd_v1_mode(sd);
1141                 inode->i_nlink = sd_v1_nlink(sd);
1142                 inode->i_uid = sd_v1_uid(sd);
1143                 inode->i_gid = sd_v1_gid(sd);
1144                 inode->i_size = sd_v1_size(sd);
1145                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1146                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1147                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1148                 inode->i_atime.tv_nsec = 0;
1149                 inode->i_ctime.tv_nsec = 0;
1150                 inode->i_mtime.tv_nsec = 0;
1151
1152                 inode->i_blocks = sd_v1_blocks(sd);
1153                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1154                 blocks = (inode->i_size + 511) >> 9;
1155                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1156                 if (inode->i_blocks > blocks) {
1157                         // there was a bug in <=3.5.23 when i_blocks could take negative
1158                         // values. Starting from 3.5.17 this value could even be stored in
1159                         // stat data. For such files we set i_blocks based on file
1160                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1161                         // only updated if file's inode will ever change
1162                         inode->i_blocks = blocks;
1163                 }
1164
1165                 rdev = sd_v1_rdev(sd);
1166                 REISERFS_I(inode)->i_first_direct_byte =
1167                     sd_v1_first_direct_byte(sd);
1168                 /* an early bug in the quota code can give us an odd number for the
1169                  ** block count.  This is incorrect, fix it here.
1170                  */
1171                 if (inode->i_blocks & 1) {
1172                         inode->i_blocks++;
1173                 }
1174                 inode_set_bytes(inode,
1175                                 to_real_used_space(inode, inode->i_blocks,
1176                                                    SD_V1_SIZE));
1177                 /* nopack is initially zero for v1 objects. For v2 objects,
1178                    nopack is initialised from sd_attrs */
1179                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1180         } else {
1181                 // new stat data found, but object may have old items
1182                 // (directories and symlinks)
1183                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1184
1185                 inode->i_mode = sd_v2_mode(sd);
1186                 inode->i_nlink = sd_v2_nlink(sd);
1187                 inode->i_uid = sd_v2_uid(sd);
1188                 inode->i_size = sd_v2_size(sd);
1189                 inode->i_gid = sd_v2_gid(sd);
1190                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1191                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1192                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1193                 inode->i_ctime.tv_nsec = 0;
1194                 inode->i_mtime.tv_nsec = 0;
1195                 inode->i_atime.tv_nsec = 0;
1196                 inode->i_blocks = sd_v2_blocks(sd);
1197                 rdev = sd_v2_rdev(sd);
1198                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1199                         inode->i_generation =
1200                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1201                 else
1202                         inode->i_generation = sd_v2_generation(sd);
1203
1204                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1205                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1206                 else
1207                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1208                 REISERFS_I(inode)->i_first_direct_byte = 0;
1209                 set_inode_sd_version(inode, STAT_DATA_V2);
1210                 inode_set_bytes(inode,
1211                                 to_real_used_space(inode, inode->i_blocks,
1212                                                    SD_V2_SIZE));
1213                 /* read persistent inode attributes from sd and initalise
1214                    generic inode flags from them */
1215                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1216                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1217         }
1218
1219         pathrelse(path);
1220         if (S_ISREG(inode->i_mode)) {
1221                 inode->i_op = &reiserfs_file_inode_operations;
1222                 inode->i_fop = &reiserfs_file_operations;
1223                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1224         } else if (S_ISDIR(inode->i_mode)) {
1225                 inode->i_op = &reiserfs_dir_inode_operations;
1226                 inode->i_fop = &reiserfs_dir_operations;
1227         } else if (S_ISLNK(inode->i_mode)) {
1228                 inode->i_op = &reiserfs_symlink_inode_operations;
1229                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1230         } else {
1231                 inode->i_blocks = 0;
1232                 inode->i_op = &reiserfs_special_inode_operations;
1233                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1234         }
1235 }
1236
1237 // update new stat data with inode fields
1238 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1239 {
1240         struct stat_data *sd_v2 = (struct stat_data *)sd;
1241         __u16 flags;
1242
1243         set_sd_v2_mode(sd_v2, inode->i_mode);
1244         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1245         set_sd_v2_uid(sd_v2, inode->i_uid);
1246         set_sd_v2_size(sd_v2, size);
1247         set_sd_v2_gid(sd_v2, inode->i_gid);
1248         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1249         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1250         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1251         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1252         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1253                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1254         else
1255                 set_sd_v2_generation(sd_v2, inode->i_generation);
1256         flags = REISERFS_I(inode)->i_attrs;
1257         i_attrs_to_sd_attrs(inode, &flags);
1258         set_sd_v2_attrs(sd_v2, flags);
1259 }
1260
1261 // used to copy inode's fields to old stat data
1262 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1263 {
1264         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1265
1266         set_sd_v1_mode(sd_v1, inode->i_mode);
1267         set_sd_v1_uid(sd_v1, inode->i_uid);
1268         set_sd_v1_gid(sd_v1, inode->i_gid);
1269         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1270         set_sd_v1_size(sd_v1, size);
1271         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1272         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1273         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1274
1275         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1276                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1277         else
1278                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1279
1280         // Sigh. i_first_direct_byte is back
1281         set_sd_v1_first_direct_byte(sd_v1,
1282                                     REISERFS_I(inode)->i_first_direct_byte);
1283 }
1284
1285 /* NOTE, you must prepare the buffer head before sending it here,
1286 ** and then log it after the call
1287 */
1288 static void update_stat_data(struct treepath *path, struct inode *inode,
1289                              loff_t size)
1290 {
1291         struct buffer_head *bh;
1292         struct item_head *ih;
1293
1294         bh = PATH_PLAST_BUFFER(path);
1295         ih = PATH_PITEM_HEAD(path);
1296
1297         if (!is_statdata_le_ih(ih))
1298                 reiserfs_panic(inode->i_sb,
1299                                "vs-13065: update_stat_data: key %k, found item %h",
1300                                INODE_PKEY(inode), ih);
1301
1302         if (stat_data_v1(ih)) {
1303                 // path points to old stat data
1304                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1305         } else {
1306                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1307         }
1308
1309         return;
1310 }
1311
1312 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1313                              struct inode *inode, loff_t size)
1314 {
1315         struct cpu_key key;
1316         INITIALIZE_PATH(path);
1317         struct buffer_head *bh;
1318         int fs_gen;
1319         struct item_head *ih, tmp_ih;
1320         int retval;
1321
1322         BUG_ON(!th->t_trans_id);
1323
1324         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1325
1326         for (;;) {
1327                 int pos;
1328                 /* look for the object's stat data */
1329                 retval = search_item(inode->i_sb, &key, &path);
1330                 if (retval == IO_ERROR) {
1331                         reiserfs_warning(inode->i_sb,
1332                                          "vs-13050: reiserfs_update_sd: "
1333                                          "i/o failure occurred trying to update %K stat data",
1334                                          &key);
1335                         return;
1336                 }
1337                 if (retval == ITEM_NOT_FOUND) {
1338                         pos = PATH_LAST_POSITION(&path);
1339                         pathrelse(&path);
1340                         if (inode->i_nlink == 0) {
1341                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1342                                 return;
1343                         }
1344                         reiserfs_warning(inode->i_sb,
1345                                          "vs-13060: reiserfs_update_sd: "
1346                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1347                                          INODE_PKEY(inode), inode->i_nlink,
1348                                          pos);
1349                         reiserfs_check_path(&path);
1350                         return;
1351                 }
1352
1353                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1354                  ** FS might change.  We have to detect that, and loop back to the
1355                  ** search if the stat data item has moved
1356                  */
1357                 bh = get_last_bh(&path);
1358                 ih = get_ih(&path);
1359                 copy_item_head(&tmp_ih, ih);
1360                 fs_gen = get_generation(inode->i_sb);
1361                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1362                 if (fs_changed(fs_gen, inode->i_sb)
1363                     && item_moved(&tmp_ih, &path)) {
1364                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1365                         continue;       /* Stat_data item has been moved after scheduling. */
1366                 }
1367                 break;
1368         }
1369         update_stat_data(&path, inode, size);
1370         journal_mark_dirty(th, th->t_super, bh);
1371         pathrelse(&path);
1372         return;
1373 }
1374
1375 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1376 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1377 ** and clear the key in the private portion of the inode, otherwise a
1378 ** corresponding iput might try to delete whatever object the inode last
1379 ** represented.
1380 */
1381 static void reiserfs_make_bad_inode(struct inode *inode)
1382 {
1383         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1384         make_bad_inode(inode);
1385 }
1386
1387 //
1388 // initially this function was derived from minix or ext2's analog and
1389 // evolved as the prototype did
1390 //
1391
1392 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1393 {
1394         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1395         inode->i_ino = args->objectid;
1396         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1397         return 0;
1398 }
1399
1400 /* looks for stat data in the tree, and fills up the fields of in-core
1401    inode stat data fields */
1402 void reiserfs_read_locked_inode(struct inode *inode,
1403                                 struct reiserfs_iget_args *args)
1404 {
1405         INITIALIZE_PATH(path_to_sd);
1406         struct cpu_key key;
1407         unsigned long dirino;
1408         int retval;
1409
1410         dirino = args->dirid;
1411
1412         /* set version 1, version 2 could be used too, because stat data
1413            key is the same in both versions */
1414         key.version = KEY_FORMAT_3_5;
1415         key.on_disk_key.k_dir_id = dirino;
1416         key.on_disk_key.k_objectid = inode->i_ino;
1417         key.on_disk_key.k_offset = 0;
1418         key.on_disk_key.k_type = 0;
1419
1420         /* look for the object's stat data */
1421         retval = search_item(inode->i_sb, &key, &path_to_sd);
1422         if (retval == IO_ERROR) {
1423                 reiserfs_warning(inode->i_sb,
1424                                  "vs-13070: reiserfs_read_locked_inode: "
1425                                  "i/o failure occurred trying to find stat data of %K",
1426                                  &key);
1427                 reiserfs_make_bad_inode(inode);
1428                 return;
1429         }
1430         if (retval != ITEM_FOUND) {
1431                 /* a stale NFS handle can trigger this without it being an error */
1432                 pathrelse(&path_to_sd);
1433                 reiserfs_make_bad_inode(inode);
1434                 inode->i_nlink = 0;
1435                 return;
1436         }
1437
1438         init_inode(inode, &path_to_sd);
1439
1440         /* It is possible that knfsd is trying to access inode of a file
1441            that is being removed from the disk by some other thread. As we
1442            update sd on unlink all that is required is to check for nlink
1443            here. This bug was first found by Sizif when debugging
1444            SquidNG/Butterfly, forgotten, and found again after Philippe
1445            Gramoulle <philippe.gramoulle@mmania.com> reproduced it. 
1446
1447            More logical fix would require changes in fs/inode.c:iput() to
1448            remove inode from hash-table _after_ fs cleaned disk stuff up and
1449            in iget() to return NULL if I_FREEING inode is found in
1450            hash-table. */
1451         /* Currently there is one place where it's ok to meet inode with
1452            nlink==0: processing of open-unlinked and half-truncated files
1453            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1454         if ((inode->i_nlink == 0) &&
1455             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1456                 reiserfs_warning(inode->i_sb,
1457                                  "vs-13075: reiserfs_read_locked_inode: "
1458                                  "dead inode read from disk %K. "
1459                                  "This is likely to be race with knfsd. Ignore",
1460                                  &key);
1461                 reiserfs_make_bad_inode(inode);
1462         }
1463
1464         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1465
1466 }
1467
1468 /**
1469  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1470  *
1471  * @inode:    inode from hash table to check
1472  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1473  *
1474  * This function is called by iget5_locked() to distinguish reiserfs inodes
1475  * having the same inode numbers. Such inodes can only exist due to some
1476  * error condition. One of them should be bad. Inodes with identical
1477  * inode numbers (objectids) are distinguished by parent directory ids.
1478  *
1479  */
1480 int reiserfs_find_actor(struct inode *inode, void *opaque)
1481 {
1482         struct reiserfs_iget_args *args;
1483
1484         args = opaque;
1485         /* args is already in CPU order */
1486         return (inode->i_ino == args->objectid) &&
1487             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1488 }
1489
1490 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1491 {
1492         struct inode *inode;
1493         struct reiserfs_iget_args args;
1494
1495         args.objectid = key->on_disk_key.k_objectid;
1496         args.dirid = key->on_disk_key.k_dir_id;
1497         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1498                              reiserfs_find_actor, reiserfs_init_locked_inode,
1499                              (void *)(&args));
1500         if (!inode)
1501                 return ERR_PTR(-ENOMEM);
1502
1503         if (inode->i_state & I_NEW) {
1504                 reiserfs_read_locked_inode(inode, &args);
1505                 unlock_new_inode(inode);
1506         }
1507
1508         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1509                 /* either due to i/o error or a stale NFS handle */
1510                 iput(inode);
1511                 inode = NULL;
1512         }
1513         return inode;
1514 }
1515
1516 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1517 {
1518         __u32 *data = vobjp;
1519         struct cpu_key key;
1520         struct dentry *result;
1521         struct inode *inode;
1522
1523         key.on_disk_key.k_objectid = data[0];
1524         key.on_disk_key.k_dir_id = data[1];
1525         reiserfs_write_lock(sb);
1526         inode = reiserfs_iget(sb, &key);
1527         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1528             data[2] != inode->i_generation) {
1529                 iput(inode);
1530                 inode = NULL;
1531         }
1532         reiserfs_write_unlock(sb);
1533         if (!inode)
1534                 inode = ERR_PTR(-ESTALE);
1535         if (IS_ERR(inode))
1536                 return ERR_PTR(PTR_ERR(inode));
1537         result = d_alloc_anon(inode);
1538         if (!result) {
1539                 iput(inode);
1540                 return ERR_PTR(-ENOMEM);
1541         }
1542         return result;
1543 }
1544
1545 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1546                                   int len, int fhtype,
1547                                   int (*acceptable) (void *contect,
1548                                                      struct dentry * de),
1549                                   void *context)
1550 {
1551         __u32 obj[3], parent[3];
1552
1553         /* fhtype happens to reflect the number of u32s encoded.
1554          * due to a bug in earlier code, fhtype might indicate there
1555          * are more u32s then actually fitted.
1556          * so if fhtype seems to be more than len, reduce fhtype.
1557          * Valid types are:
1558          *   2 - objectid + dir_id - legacy support
1559          *   3 - objectid + dir_id + generation
1560          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1561          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1562          *   6 - as above plus generation of directory
1563          * 6 does not fit in NFSv2 handles
1564          */
1565         if (fhtype > len) {
1566                 if (fhtype != 6 || len != 5)
1567                         reiserfs_warning(sb,
1568                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1569                                          fhtype, len);
1570                 fhtype = 5;
1571         }
1572
1573         obj[0] = data[0];
1574         obj[1] = data[1];
1575         if (fhtype == 3 || fhtype >= 5)
1576                 obj[2] = data[2];
1577         else
1578                 obj[2] = 0;     /* generation number */
1579
1580         if (fhtype >= 4) {
1581                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1582                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1583                 if (fhtype == 6)
1584                         parent[2] = data[5];
1585                 else
1586                         parent[2] = 0;
1587         }
1588         return sb->s_export_op->find_exported_dentry(sb, obj,
1589                                                      fhtype < 4 ? NULL : parent,
1590                                                      acceptable, context);
1591 }
1592
1593 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1594                        int need_parent)
1595 {
1596         struct inode *inode = dentry->d_inode;
1597         int maxlen = *lenp;
1598
1599         if (maxlen < 3)
1600                 return 255;
1601
1602         data[0] = inode->i_ino;
1603         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1604         data[2] = inode->i_generation;
1605         *lenp = 3;
1606         /* no room for directory info? return what we've stored so far */
1607         if (maxlen < 5 || !need_parent)
1608                 return 3;
1609
1610         spin_lock(&dentry->d_lock);
1611         inode = dentry->d_parent->d_inode;
1612         data[3] = inode->i_ino;
1613         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1614         *lenp = 5;
1615         if (maxlen >= 6) {
1616                 data[5] = inode->i_generation;
1617                 *lenp = 6;
1618         }
1619         spin_unlock(&dentry->d_lock);
1620         return *lenp;
1621 }
1622
1623 /* looks for stat data, then copies fields to it, marks the buffer
1624    containing stat data as dirty */
1625 /* reiserfs inodes are never really dirty, since the dirty inode call
1626 ** always logs them.  This call allows the VFS inode marking routines
1627 ** to properly mark inodes for datasync and such, but only actually
1628 ** does something when called for a synchronous update.
1629 */
1630 int reiserfs_write_inode(struct inode *inode, int do_sync)
1631 {
1632         struct reiserfs_transaction_handle th;
1633         int jbegin_count = 1;
1634
1635         if (inode->i_sb->s_flags & MS_RDONLY)
1636                 return -EROFS;
1637         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1638          ** these cases are just when the system needs ram, not when the 
1639          ** inode needs to reach disk for safety, and they can safely be
1640          ** ignored because the altered inode has already been logged.
1641          */
1642         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1643                 reiserfs_write_lock(inode->i_sb);
1644                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1645                         reiserfs_update_sd(&th, inode);
1646                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1647                 }
1648                 reiserfs_write_unlock(inode->i_sb);
1649         }
1650         return 0;
1651 }
1652
1653 /* stat data of new object is inserted already, this inserts the item
1654    containing "." and ".." entries */
1655 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1656                                   struct inode *inode,
1657                                   struct item_head *ih, struct treepath *path,
1658                                   struct inode *dir)
1659 {
1660         struct super_block *sb = th->t_super;
1661         char empty_dir[EMPTY_DIR_SIZE];
1662         char *body = empty_dir;
1663         struct cpu_key key;
1664         int retval;
1665
1666         BUG_ON(!th->t_trans_id);
1667
1668         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1669                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1670                       TYPE_DIRENTRY, 3 /*key length */ );
1671
1672         /* compose item head for new item. Directories consist of items of
1673            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1674            is done by reiserfs_new_inode */
1675         if (old_format_only(sb)) {
1676                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1677                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1678
1679                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1680                                        ih->ih_key.k_objectid,
1681                                        INODE_PKEY(dir)->k_dir_id,
1682                                        INODE_PKEY(dir)->k_objectid);
1683         } else {
1684                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1685                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1686
1687                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1688                                     ih->ih_key.k_objectid,
1689                                     INODE_PKEY(dir)->k_dir_id,
1690                                     INODE_PKEY(dir)->k_objectid);
1691         }
1692
1693         /* look for place in the tree for new item */
1694         retval = search_item(sb, &key, path);
1695         if (retval == IO_ERROR) {
1696                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1697                                  "i/o failure occurred creating new directory");
1698                 return -EIO;
1699         }
1700         if (retval == ITEM_FOUND) {
1701                 pathrelse(path);
1702                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1703                                  "object with this key exists (%k)",
1704                                  &(ih->ih_key));
1705                 return -EEXIST;
1706         }
1707
1708         /* insert item, that is empty directory item */
1709         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1710 }
1711
1712 /* stat data of object has been inserted, this inserts the item
1713    containing the body of symlink */
1714 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1715                                 struct item_head *ih,
1716                                 struct treepath *path, const char *symname,
1717                                 int item_len)
1718 {
1719         struct super_block *sb = th->t_super;
1720         struct cpu_key key;
1721         int retval;
1722
1723         BUG_ON(!th->t_trans_id);
1724
1725         _make_cpu_key(&key, KEY_FORMAT_3_5,
1726                       le32_to_cpu(ih->ih_key.k_dir_id),
1727                       le32_to_cpu(ih->ih_key.k_objectid),
1728                       1, TYPE_DIRECT, 3 /*key length */ );
1729
1730         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1731                           0 /*free_space */ );
1732
1733         /* look for place in the tree for new item */
1734         retval = search_item(sb, &key, path);
1735         if (retval == IO_ERROR) {
1736                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1737                                  "i/o failure occurred creating new symlink");
1738                 return -EIO;
1739         }
1740         if (retval == ITEM_FOUND) {
1741                 pathrelse(path);
1742                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1743                                  "object with this key exists (%k)",
1744                                  &(ih->ih_key));
1745                 return -EEXIST;
1746         }
1747
1748         /* insert item, that is body of symlink */
1749         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1750 }
1751
1752 /* inserts the stat data into the tree, and then calls
1753    reiserfs_new_directory (to insert ".", ".." item if new object is
1754    directory) or reiserfs_new_symlink (to insert symlink body if new
1755    object is symlink) or nothing (if new object is regular file) 
1756
1757    NOTE! uid and gid must already be set in the inode.  If we return
1758    non-zero due to an error, we have to drop the quota previously allocated
1759    for the fresh inode.  This can only be done outside a transaction, so
1760    if we return non-zero, we also end the transaction.  */
1761 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1762                        struct inode *dir, int mode, const char *symname,
1763                        /* 0 for regular, EMTRY_DIR_SIZE for dirs, 
1764                           strlen (symname) for symlinks) */
1765                        loff_t i_size, struct dentry *dentry,
1766                        struct inode *inode)
1767 {
1768         struct super_block *sb;
1769         INITIALIZE_PATH(path_to_key);
1770         struct cpu_key key;
1771         struct item_head ih;
1772         struct stat_data sd;
1773         int retval;
1774         int err;
1775
1776         BUG_ON(!th->t_trans_id);
1777
1778         if (DQUOT_ALLOC_INODE(inode)) {
1779                 err = -EDQUOT;
1780                 goto out_end_trans;
1781         }
1782         if (!dir->i_nlink) {
1783                 err = -EPERM;
1784                 goto out_bad_inode;
1785         }
1786
1787         sb = dir->i_sb;
1788
1789         /* item head of new item */
1790         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1791         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1792         if (!ih.ih_key.k_objectid) {
1793                 err = -ENOMEM;
1794                 goto out_bad_inode;
1795         }
1796         if (old_format_only(sb))
1797                 /* not a perfect generation count, as object ids can be reused, but 
1798                  ** this is as good as reiserfs can do right now.
1799                  ** note that the private part of inode isn't filled in yet, we have
1800                  ** to use the directory.
1801                  */
1802                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1803         else
1804 #if defined( USE_INODE_GENERATION_COUNTER )
1805                 inode->i_generation =
1806                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1807 #else
1808                 inode->i_generation = ++event;
1809 #endif
1810
1811         /* fill stat data */
1812         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1813
1814         /* uid and gid must already be set by the caller for quota init */
1815
1816         /* symlink cannot be immutable or append only, right? */
1817         if (S_ISLNK(inode->i_mode))
1818                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1819
1820         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1821         inode->i_size = i_size;
1822         inode->i_blocks = 0;
1823         inode->i_bytes = 0;
1824         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1825             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1826
1827         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1828         REISERFS_I(inode)->i_flags = 0;
1829         REISERFS_I(inode)->i_prealloc_block = 0;
1830         REISERFS_I(inode)->i_prealloc_count = 0;
1831         REISERFS_I(inode)->i_trans_id = 0;
1832         REISERFS_I(inode)->i_jl = NULL;
1833         REISERFS_I(inode)->i_attrs =
1834             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1835         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1836         mutex_init(&(REISERFS_I(inode)->i_mmap));
1837         reiserfs_init_acl_access(inode);
1838         reiserfs_init_acl_default(inode);
1839         reiserfs_init_xattr_rwsem(inode);
1840
1841         if (old_format_only(sb))
1842                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1843                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1844         else
1845                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1846                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1847
1848         /* key to search for correct place for new stat data */
1849         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1850                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1851                       TYPE_STAT_DATA, 3 /*key length */ );
1852
1853         /* find proper place for inserting of stat data */
1854         retval = search_item(sb, &key, &path_to_key);
1855         if (retval == IO_ERROR) {
1856                 err = -EIO;
1857                 goto out_bad_inode;
1858         }
1859         if (retval == ITEM_FOUND) {
1860                 pathrelse(&path_to_key);
1861                 err = -EEXIST;
1862                 goto out_bad_inode;
1863         }
1864         if (old_format_only(sb)) {
1865                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1866                         pathrelse(&path_to_key);
1867                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1868                         err = -EINVAL;
1869                         goto out_bad_inode;
1870                 }
1871                 inode2sd_v1(&sd, inode, inode->i_size);
1872         } else {
1873                 inode2sd(&sd, inode, inode->i_size);
1874         }
1875         // these do not go to on-disk stat data
1876         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1877
1878         // store in in-core inode the key of stat data and version all
1879         // object items will have (directory items will have old offset
1880         // format, other new objects will consist of new items)
1881         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1882         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1883                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1884         else
1885                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1886         if (old_format_only(sb))
1887                 set_inode_sd_version(inode, STAT_DATA_V1);
1888         else
1889                 set_inode_sd_version(inode, STAT_DATA_V2);
1890
1891         /* insert the stat data into the tree */
1892 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1893         if (REISERFS_I(dir)->new_packing_locality)
1894                 th->displace_new_blocks = 1;
1895 #endif
1896         retval =
1897             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1898                                  (char *)(&sd));
1899         if (retval) {
1900                 err = retval;
1901                 reiserfs_check_path(&path_to_key);
1902                 goto out_bad_inode;
1903         }
1904 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1905         if (!th->displace_new_blocks)
1906                 REISERFS_I(dir)->new_packing_locality = 0;
1907 #endif
1908         if (S_ISDIR(mode)) {
1909                 /* insert item with "." and ".." */
1910                 retval =
1911                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1912         }
1913
1914         if (S_ISLNK(mode)) {
1915                 /* insert body of symlink */
1916                 if (!old_format_only(sb))
1917                         i_size = ROUND_UP(i_size);
1918                 retval =
1919                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1920                                          i_size);
1921         }
1922         if (retval) {
1923                 err = retval;
1924                 reiserfs_check_path(&path_to_key);
1925                 journal_end(th, th->t_super, th->t_blocks_allocated);
1926                 goto out_inserted_sd;
1927         }
1928
1929         /* XXX CHECK THIS */
1930         if (reiserfs_posixacl(inode->i_sb)) {
1931                 retval = reiserfs_inherit_default_acl(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, "ACLs aren't enabled in the fs, "
1940                                  "but vfs thinks they are!");
1941         } else if (is_reiserfs_priv_object(dir)) {
1942                 reiserfs_mark_inode_private(inode);
1943         }
1944
1945         insert_inode_hash(inode);
1946         reiserfs_update_sd(th, inode);
1947         reiserfs_check_path(&path_to_key);
1948
1949         return 0;
1950
1951 /* it looks like you can easily compress these two goto targets into
1952  * one.  Keeping it like this doesn't actually hurt anything, and they
1953  * are place holders for what the quota code actually needs.
1954  */
1955       out_bad_inode:
1956         /* Invalidate the object, nothing was inserted yet */
1957         INODE_PKEY(inode)->k_objectid = 0;
1958
1959         /* Quota change must be inside a transaction for journaling */
1960         DQUOT_FREE_INODE(inode);
1961
1962       out_end_trans:
1963         journal_end(th, th->t_super, th->t_blocks_allocated);
1964         /* Drop can be outside and it needs more credits so it's better to have it outside */
1965         DQUOT_DROP(inode);
1966         inode->i_flags |= S_NOQUOTA;
1967         make_bad_inode(inode);
1968
1969       out_inserted_sd:
1970         inode->i_nlink = 0;
1971         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1972
1973         /* If we were inheriting an ACL, we need to release the lock so that
1974          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1975          * code really needs to be reworked, but this will take care of it
1976          * for now. -jeffm */
1977 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1978         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1979                 reiserfs_write_unlock_xattrs(dir->i_sb);
1980                 iput(inode);
1981                 reiserfs_write_lock_xattrs(dir->i_sb);
1982         } else
1983 #endif
1984                 iput(inode);
1985         return err;
1986 }
1987
1988 /*
1989 ** finds the tail page in the page cache,
1990 ** reads the last block in.
1991 **
1992 ** On success, page_result is set to a locked, pinned page, and bh_result
1993 ** is set to an up to date buffer for the last block in the file.  returns 0.
1994 **
1995 ** tail conversion is not done, so bh_result might not be valid for writing
1996 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1997 ** trying to write the block.
1998 **
1999 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2000 */
2001 static int grab_tail_page(struct inode *p_s_inode,
2002                           struct page **page_result,
2003                           struct buffer_head **bh_result)
2004 {
2005
2006         /* we want the page with the last byte in the file,
2007          ** not the page that will hold the next byte for appending
2008          */
2009         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2010         unsigned long pos = 0;
2011         unsigned long start = 0;
2012         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2013         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2014         struct buffer_head *bh;
2015         struct buffer_head *head;
2016         struct page *page;
2017         int error;
2018
2019         /* we know that we are only called with inode->i_size > 0.
2020          ** we also know that a file tail can never be as big as a block
2021          ** If i_size % blocksize == 0, our file is currently block aligned
2022          ** and it won't need converting or zeroing after a truncate.
2023          */
2024         if ((offset & (blocksize - 1)) == 0) {
2025                 return -ENOENT;
2026         }
2027         page = grab_cache_page(p_s_inode->i_mapping, index);
2028         error = -ENOMEM;
2029         if (!page) {
2030                 goto out;
2031         }
2032         /* start within the page of the last block in the file */
2033         start = (offset / blocksize) * blocksize;
2034
2035         error = block_prepare_write(page, start, offset,
2036                                     reiserfs_get_block_create_0);
2037         if (error)
2038                 goto unlock;
2039
2040         head = page_buffers(page);
2041         bh = head;
2042         do {
2043                 if (pos >= start) {
2044                         break;
2045                 }
2046                 bh = bh->b_this_page;
2047                 pos += blocksize;
2048         } while (bh != head);
2049
2050         if (!buffer_uptodate(bh)) {
2051                 /* note, this should never happen, prepare_write should
2052                  ** be taking care of this for us.  If the buffer isn't up to date,
2053                  ** I've screwed up the code to find the buffer, or the code to
2054                  ** call prepare_write
2055                  */
2056                 reiserfs_warning(p_s_inode->i_sb,
2057                                  "clm-6000: error reading block %lu on dev %s",
2058                                  bh->b_blocknr,
2059                                  reiserfs_bdevname(p_s_inode->i_sb));
2060                 error = -EIO;
2061                 goto unlock;
2062         }
2063         *bh_result = bh;
2064         *page_result = page;
2065
2066       out:
2067         return error;
2068
2069       unlock:
2070         unlock_page(page);
2071         page_cache_release(page);
2072         return error;
2073 }
2074
2075 /*
2076 ** vfs version of truncate file.  Must NOT be called with
2077 ** a transaction already started.
2078 **
2079 ** some code taken from block_truncate_page
2080 */
2081 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2082 {
2083         struct reiserfs_transaction_handle th;
2084         /* we want the offset for the first byte after the end of the file */
2085         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2086         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2087         unsigned length;
2088         struct page *page = NULL;
2089         int error;
2090         struct buffer_head *bh = NULL;
2091         int err2;
2092
2093         reiserfs_write_lock(p_s_inode->i_sb);
2094
2095         if (p_s_inode->i_size > 0) {
2096                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2097                         // -ENOENT means we truncated past the end of the file, 
2098                         // and get_block_create_0 could not find a block to read in,
2099                         // which is ok.
2100                         if (error != -ENOENT)
2101                                 reiserfs_warning(p_s_inode->i_sb,
2102                                                  "clm-6001: grab_tail_page failed %d",
2103                                                  error);
2104                         page = NULL;
2105                         bh = NULL;
2106                 }
2107         }
2108
2109         /* so, if page != NULL, we have a buffer head for the offset at 
2110          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0, 
2111          ** then we have an unformatted node.  Otherwise, we have a direct item, 
2112          ** and no zeroing is required on disk.  We zero after the truncate, 
2113          ** because the truncate might pack the item anyway 
2114          ** (it will unmap bh if it packs).
2115          */
2116         /* it is enough to reserve space in transaction for 2 balancings:
2117            one for "save" link adding and another for the first
2118            cut_from_item. 1 is for update_sd */
2119         error = journal_begin(&th, p_s_inode->i_sb,
2120                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2121         if (error)
2122                 goto out;
2123         reiserfs_update_inode_transaction(p_s_inode);
2124         if (update_timestamps)
2125                 /* we are doing real truncate: if the system crashes before the last
2126                    transaction of truncating gets committed - on reboot the file
2127                    either appears truncated properly or not truncated at all */
2128                 add_save_link(&th, p_s_inode, 1);
2129         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2130         error =
2131             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2132         if (error)
2133                 goto out;
2134
2135         /* check reiserfs_do_truncate after ending the transaction */
2136         if (err2) {
2137                 error = err2;
2138                 goto out;
2139         }
2140         
2141         if (update_timestamps) {
2142                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2143                 if (error)
2144                         goto out;
2145         }
2146
2147         if (page) {
2148                 length = offset & (blocksize - 1);
2149                 /* if we are not on a block boundary */
2150                 if (length) {
2151                         length = blocksize - length;
2152                         zero_user_page(page, offset, length, KM_USER0);
2153                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2154                                 mark_buffer_dirty(bh);
2155                         }
2156                 }
2157                 unlock_page(page);
2158                 page_cache_release(page);
2159         }
2160
2161         reiserfs_write_unlock(p_s_inode->i_sb);
2162         return 0;
2163       out:
2164         if (page) {
2165                 unlock_page(page);
2166                 page_cache_release(page);
2167         }
2168         reiserfs_write_unlock(p_s_inode->i_sb);
2169         return error;
2170 }
2171
2172 static int map_block_for_writepage(struct inode *inode,
2173                                    struct buffer_head *bh_result,
2174                                    unsigned long block)
2175 {
2176         struct reiserfs_transaction_handle th;
2177         int fs_gen;
2178         struct item_head tmp_ih;
2179         struct item_head *ih;
2180         struct buffer_head *bh;
2181         __le32 *item;
2182         struct cpu_key key;
2183         INITIALIZE_PATH(path);
2184         int pos_in_item;
2185         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2186         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2187         int retval;
2188         int use_get_block = 0;
2189         int bytes_copied = 0;
2190         int copy_size;
2191         int trans_running = 0;
2192
2193         /* catch places below that try to log something without starting a trans */
2194         th.t_trans_id = 0;
2195
2196         if (!buffer_uptodate(bh_result)) {
2197                 return -EIO;
2198         }
2199
2200         kmap(bh_result->b_page);
2201       start_over:
2202         reiserfs_write_lock(inode->i_sb);
2203         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2204
2205       research:
2206         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2207         if (retval != POSITION_FOUND) {
2208                 use_get_block = 1;
2209                 goto out;
2210         }
2211
2212         bh = get_last_bh(&path);
2213         ih = get_ih(&path);
2214         item = get_item(&path);
2215         pos_in_item = path.pos_in_item;
2216
2217         /* we've found an unformatted node */
2218         if (indirect_item_found(retval, ih)) {
2219                 if (bytes_copied > 0) {
2220                         reiserfs_warning(inode->i_sb,
2221                                          "clm-6002: bytes_copied %d",
2222                                          bytes_copied);
2223                 }
2224                 if (!get_block_num(item, pos_in_item)) {
2225                         /* crap, we are writing to a hole */
2226                         use_get_block = 1;
2227                         goto out;
2228                 }
2229                 set_block_dev_mapped(bh_result,
2230                                      get_block_num(item, pos_in_item), inode);
2231         } else if (is_direct_le_ih(ih)) {
2232                 char *p;
2233                 p = page_address(bh_result->b_page);
2234                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2235                 copy_size = ih_item_len(ih) - pos_in_item;
2236
2237                 fs_gen = get_generation(inode->i_sb);
2238                 copy_item_head(&tmp_ih, ih);
2239
2240                 if (!trans_running) {
2241                         /* vs-3050 is gone, no need to drop the path */
2242                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2243                         if (retval)
2244                                 goto out;
2245                         reiserfs_update_inode_transaction(inode);
2246                         trans_running = 1;
2247                         if (fs_changed(fs_gen, inode->i_sb)
2248                             && item_moved(&tmp_ih, &path)) {
2249                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2250                                                                  bh);
2251                                 goto research;
2252                         }
2253                 }
2254
2255                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2256
2257                 if (fs_changed(fs_gen, inode->i_sb)
2258                     && item_moved(&tmp_ih, &path)) {
2259                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2260                         goto research;
2261                 }
2262
2263                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2264                        copy_size);
2265
2266                 journal_mark_dirty(&th, inode->i_sb, bh);
2267                 bytes_copied += copy_size;
2268                 set_block_dev_mapped(bh_result, 0, inode);
2269
2270                 /* are there still bytes left? */
2271                 if (bytes_copied < bh_result->b_size &&
2272                     (byte_offset + bytes_copied) < inode->i_size) {
2273                         set_cpu_key_k_offset(&key,
2274                                              cpu_key_k_offset(&key) +
2275                                              copy_size);
2276                         goto research;
2277                 }
2278         } else {
2279                 reiserfs_warning(inode->i_sb,
2280                                  "clm-6003: bad item inode %lu, device %s",
2281                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2282                 retval = -EIO;
2283                 goto out;
2284         }
2285         retval = 0;
2286
2287       out:
2288         pathrelse(&path);
2289         if (trans_running) {
2290                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2291                 if (err)
2292                         retval = err;
2293                 trans_running = 0;
2294         }
2295         reiserfs_write_unlock(inode->i_sb);
2296
2297         /* this is where we fill in holes in the file. */
2298         if (use_get_block) {
2299                 retval = reiserfs_get_block(inode, block, bh_result,
2300                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2301                                             | GET_BLOCK_NO_DANGLE);
2302                 if (!retval) {
2303                         if (!buffer_mapped(bh_result)
2304                             || bh_result->b_blocknr == 0) {
2305                                 /* get_block failed to find a mapped unformatted node. */
2306                                 use_get_block = 0;
2307                                 goto start_over;
2308                         }
2309                 }
2310         }
2311         kunmap(bh_result->b_page);
2312
2313         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2314                 /* we've copied data from the page into the direct item, so the
2315                  * buffer in the page is now clean, mark it to reflect that.
2316                  */
2317                 lock_buffer(bh_result);
2318                 clear_buffer_dirty(bh_result);
2319                 unlock_buffer(bh_result);
2320         }
2321         return retval;
2322 }
2323
2324 /* 
2325  * mason@suse.com: updated in 2.5.54 to follow the same general io 
2326  * start/recovery path as __block_write_full_page, along with special
2327  * code to handle reiserfs tails.
2328  */
2329 static int reiserfs_write_full_page(struct page *page,
2330                                     struct writeback_control *wbc)
2331 {
2332         struct inode *inode = page->mapping->host;
2333         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2334         int error = 0;
2335         unsigned long block;
2336         sector_t last_block;
2337         struct buffer_head *head, *bh;
2338         int partial = 0;
2339         int nr = 0;
2340         int checked = PageChecked(page);
2341         struct reiserfs_transaction_handle th;
2342         struct super_block *s = inode->i_sb;
2343         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2344         th.t_trans_id = 0;
2345
2346         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2347         if (checked && (current->flags & PF_MEMALLOC)) {
2348                 redirty_page_for_writepage(wbc, page);
2349                 unlock_page(page);
2350                 return 0;
2351         }
2352
2353         /* The page dirty bit is cleared before writepage is called, which
2354          * means we have to tell create_empty_buffers to make dirty buffers
2355          * The page really should be up to date at this point, so tossing
2356          * in the BH_Uptodate is just a sanity check.
2357          */
2358         if (!page_has_buffers(page)) {
2359                 create_empty_buffers(page, s->s_blocksize,
2360                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2361         }
2362         head = page_buffers(page);
2363
2364         /* last page in the file, zero out any contents past the
2365          ** last byte in the file
2366          */
2367         if (page->index >= end_index) {
2368                 unsigned last_offset;
2369
2370                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2371                 /* no file contents in this page */
2372                 if (page->index >= end_index + 1 || !last_offset) {
2373                         unlock_page(page);
2374                         return 0;
2375                 }
2376                 zero_user_page(page, last_offset, PAGE_CACHE_SIZE - last_offset, KM_USER0);
2377         }
2378         bh = head;
2379         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2380         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2381         /* first map all the buffers, logging any direct items we find */
2382         do {
2383                 if (block > last_block) {
2384                         /*
2385                          * This can happen when the block size is less than
2386                          * the page size.  The corresponding bytes in the page
2387                          * were zero filled above
2388                          */
2389                         clear_buffer_dirty(bh);
2390                         set_buffer_uptodate(bh);
2391                 } else if ((checked || buffer_dirty(bh)) &&
2392                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2393                                                        && bh->b_blocknr ==
2394                                                        0))) {
2395                         /* not mapped yet, or it points to a direct item, search
2396                          * the btree for the mapping info, and log any direct
2397                          * items found
2398                          */
2399                         if ((error = map_block_for_writepage(inode, bh, block))) {
2400                                 goto fail;
2401                         }
2402                 }
2403                 bh = bh->b_this_page;
2404                 block++;
2405         } while (bh != head);
2406
2407         /*
2408          * we start the transaction after map_block_for_writepage,
2409          * because it can create holes in the file (an unbounded operation).
2410          * starting it here, we can make a reliable estimate for how many
2411          * blocks we're going to log
2412          */
2413         if (checked) {
2414                 ClearPageChecked(page);
2415                 reiserfs_write_lock(s);
2416                 error = journal_begin(&th, s, bh_per_page + 1);
2417                 if (error) {
2418                         reiserfs_write_unlock(s);
2419                         goto fail;
2420                 }
2421                 reiserfs_update_inode_transaction(inode);
2422         }
2423         /* now go through and lock any dirty buffers on the page */
2424         do {
2425                 get_bh(bh);
2426                 if (!buffer_mapped(bh))
2427                         continue;
2428                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2429                         continue;
2430
2431                 if (checked) {
2432                         reiserfs_prepare_for_journal(s, bh, 1);
2433                         journal_mark_dirty(&th, s, bh);
2434                         continue;
2435                 }
2436                 /* from this point on, we know the buffer is mapped to a
2437                  * real block and not a direct item
2438                  */
2439                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2440                         lock_buffer(bh);
2441                 } else {
2442                         if (test_set_buffer_locked(bh)) {
2443                                 redirty_page_for_writepage(wbc, page);
2444                                 continue;
2445                         }
2446                 }
2447                 if (test_clear_buffer_dirty(bh)) {
2448                         mark_buffer_async_write(bh);
2449                 } else {
2450                         unlock_buffer(bh);
2451                 }
2452         } while ((bh = bh->b_this_page) != head);
2453
2454         if (checked) {
2455                 error = journal_end(&th, s, bh_per_page + 1);
2456                 reiserfs_write_unlock(s);
2457                 if (error)
2458                         goto fail;
2459         }
2460         BUG_ON(PageWriteback(page));
2461         set_page_writeback(page);
2462         unlock_page(page);
2463
2464         /*
2465          * since any buffer might be the only dirty buffer on the page, 
2466          * the first submit_bh can bring the page out of writeback.
2467          * be careful with the buffers.
2468          */
2469         do {
2470                 struct buffer_head *next = bh->b_this_page;
2471                 if (buffer_async_write(bh)) {
2472                         submit_bh(WRITE, bh);
2473                         nr++;
2474                 }
2475                 put_bh(bh);
2476                 bh = next;
2477         } while (bh != head);
2478
2479         error = 0;
2480       done:
2481         if (nr == 0) {
2482                 /*
2483                  * if this page only had a direct item, it is very possible for
2484                  * no io to be required without there being an error.  Or, 
2485                  * someone else could have locked them and sent them down the 
2486                  * pipe without locking the page
2487                  */
2488                 bh = head;
2489                 do {
2490                         if (!buffer_uptodate(bh)) {
2491                                 partial = 1;
2492                                 break;
2493                         }
2494                         bh = bh->b_this_page;
2495                 } while (bh != head);
2496                 if (!partial)
2497                         SetPageUptodate(page);
2498                 end_page_writeback(page);
2499         }
2500         return error;
2501
2502       fail:
2503         /* catches various errors, we need to make sure any valid dirty blocks
2504          * get to the media.  The page is currently locked and not marked for 
2505          * writeback
2506          */
2507         ClearPageUptodate(page);
2508         bh = head;
2509         do {
2510                 get_bh(bh);
2511                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2512                         lock_buffer(bh);
2513                         mark_buffer_async_write(bh);
2514                 } else {
2515                         /*
2516                          * clear any dirty bits that might have come from getting
2517                          * attached to a dirty page
2518                          */
2519                         clear_buffer_dirty(bh);
2520                 }
2521                 bh = bh->b_this_page;
2522         } while (bh != head);
2523         SetPageError(page);
2524         BUG_ON(PageWriteback(page));
2525         set_page_writeback(page);
2526         unlock_page(page);
2527         do {
2528                 struct buffer_head *next = bh->b_this_page;
2529                 if (buffer_async_write(bh)) {
2530                         clear_buffer_dirty(bh);
2531                         submit_bh(WRITE, bh);
2532                         nr++;
2533                 }
2534                 put_bh(bh);
2535                 bh = next;
2536         } while (bh != head);
2537         goto done;
2538 }
2539
2540 static int reiserfs_readpage(struct file *f, struct page *page)
2541 {
2542         return block_read_full_page(page, reiserfs_get_block);
2543 }
2544
2545 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2546 {
2547         struct inode *inode = page->mapping->host;
2548         reiserfs_wait_on_write_block(inode->i_sb);
2549         return reiserfs_write_full_page(page, wbc);
2550 }
2551
2552 static int reiserfs_prepare_write(struct file *f, struct page *page,
2553                                   unsigned from, unsigned to)
2554 {
2555         struct inode *inode = page->mapping->host;
2556         int ret;
2557         int old_ref = 0;
2558
2559         reiserfs_wait_on_write_block(inode->i_sb);
2560         fix_tail_page_for_writing(page);
2561         if (reiserfs_transaction_running(inode->i_sb)) {
2562                 struct reiserfs_transaction_handle *th;
2563                 th = (struct reiserfs_transaction_handle *)current->
2564                     journal_info;
2565                 BUG_ON(!th->t_refcount);
2566                 BUG_ON(!th->t_trans_id);
2567                 old_ref = th->t_refcount;
2568                 th->t_refcount++;
2569         }
2570
2571         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2572         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2573                 struct reiserfs_transaction_handle *th = current->journal_info;
2574                 /* this gets a little ugly.  If reiserfs_get_block returned an
2575                  * error and left a transacstion running, we've got to close it,
2576                  * and we've got to free handle if it was a persistent transaction.
2577                  *
2578                  * But, if we had nested into an existing transaction, we need
2579                  * to just drop the ref count on the handle.
2580                  *
2581                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2582                  * and it was a persistent trans.  Otherwise, it was nested above.
2583                  */
2584                 if (th->t_refcount > old_ref) {
2585                         if (old_ref)
2586                                 th->t_refcount--;
2587                         else {
2588                                 int err;
2589                                 reiserfs_write_lock(inode->i_sb);
2590                                 err = reiserfs_end_persistent_transaction(th);
2591                                 reiserfs_write_unlock(inode->i_sb);
2592                                 if (err)
2593                                         ret = err;
2594                         }
2595                 }
2596         }
2597         return ret;
2598
2599 }
2600
2601 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2602 {
2603         return generic_block_bmap(as, block, reiserfs_bmap);
2604 }
2605
2606 static int reiserfs_commit_write(struct file *f, struct page *page,
2607                                  unsigned from, unsigned to)
2608 {
2609         struct inode *inode = page->mapping->host;
2610         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2611         int ret = 0;
2612         int update_sd = 0;
2613         struct reiserfs_transaction_handle *th = NULL;
2614
2615         reiserfs_wait_on_write_block(inode->i_sb);
2616         if (reiserfs_transaction_running(inode->i_sb)) {
2617                 th = current->journal_info;
2618         }
2619         reiserfs_commit_page(inode, page, from, to);
2620
2621         /* generic_commit_write does this for us, but does not update the
2622          ** transaction tracking stuff when the size changes.  So, we have
2623          ** to do the i_size updates here.
2624          */
2625         if (pos > inode->i_size) {
2626                 struct reiserfs_transaction_handle myth;
2627                 reiserfs_write_lock(inode->i_sb);
2628                 /* If the file have grown beyond the border where it
2629                    can have a tail, unmark it as needing a tail
2630                    packing */
2631                 if ((have_large_tails(inode->i_sb)
2632                      && inode->i_size > i_block_size(inode) * 4)
2633                     || (have_small_tails(inode->i_sb)
2634                         && inode->i_size > i_block_size(inode)))
2635                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2636
2637                 ret = journal_begin(&myth, inode->i_sb, 1);
2638                 if (ret) {
2639                         reiserfs_write_unlock(inode->i_sb);
2640                         goto journal_error;
2641                 }
2642                 reiserfs_update_inode_transaction(inode);
2643                 inode->i_size = pos;
2644                 /*
2645                  * this will just nest into our transaction.  It's important
2646                  * to use mark_inode_dirty so the inode gets pushed around on the
2647                  * dirty lists, and so that O_SYNC works as expected
2648                  */
2649                 mark_inode_dirty(inode);
2650                 reiserfs_update_sd(&myth, inode);
2651                 update_sd = 1;
2652                 ret = journal_end(&myth, inode->i_sb, 1);
2653                 reiserfs_write_unlock(inode->i_sb);
2654                 if (ret)
2655                         goto journal_error;
2656         }
2657         if (th) {
2658                 reiserfs_write_lock(inode->i_sb);
2659                 if (!update_sd)
2660                         mark_inode_dirty(inode);
2661                 ret = reiserfs_end_persistent_transaction(th);
2662                 reiserfs_write_unlock(inode->i_sb);
2663                 if (ret)
2664                         goto out;
2665         }
2666
2667       out:
2668         return ret;
2669
2670       journal_error:
2671         if (th) {
2672                 reiserfs_write_lock(inode->i_sb);
2673                 if (!update_sd)
2674                         reiserfs_update_sd(th, inode);
2675                 ret = reiserfs_end_persistent_transaction(th);
2676                 reiserfs_write_unlock(inode->i_sb);
2677         }
2678
2679         return ret;
2680 }
2681
2682 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2683 {
2684         if (reiserfs_attrs(inode->i_sb)) {
2685                 if (sd_attrs & REISERFS_SYNC_FL)
2686                         inode->i_flags |= S_SYNC;
2687                 else
2688                         inode->i_flags &= ~S_SYNC;
2689                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2690                         inode->i_flags |= S_IMMUTABLE;
2691                 else
2692                         inode->i_flags &= ~S_IMMUTABLE;
2693                 if (sd_attrs & REISERFS_APPEND_FL)
2694                         inode->i_flags |= S_APPEND;
2695                 else
2696                         inode->i_flags &= ~S_APPEND;
2697                 if (sd_attrs & REISERFS_NOATIME_FL)
2698                         inode->i_flags |= S_NOATIME;
2699                 else
2700                         inode->i_flags &= ~S_NOATIME;
2701                 if (sd_attrs & REISERFS_NOTAIL_FL)
2702                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2703                 else
2704                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2705         }
2706 }
2707
2708 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2709 {
2710         if (reiserfs_attrs(inode->i_sb)) {
2711                 if (inode->i_flags & S_IMMUTABLE)
2712                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2713                 else
2714                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2715                 if (inode->i_flags & S_SYNC)
2716                         *sd_attrs |= REISERFS_SYNC_FL;
2717                 else
2718                         *sd_attrs &= ~REISERFS_SYNC_FL;
2719                 if (inode->i_flags & S_NOATIME)
2720                         *sd_attrs |= REISERFS_NOATIME_FL;
2721                 else
2722                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2723                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2724                         *sd_attrs |= REISERFS_NOTAIL_FL;
2725                 else
2726                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2727         }
2728 }
2729
2730 /* decide if this buffer needs to stay around for data logging or ordered
2731 ** write purposes
2732 */
2733 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2734 {
2735         int ret = 1;
2736         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2737
2738         lock_buffer(bh);
2739         spin_lock(&j->j_dirty_buffers_lock);
2740         if (!buffer_mapped(bh)) {
2741                 goto free_jh;
2742         }
2743         /* the page is locked, and the only places that log a data buffer
2744          * also lock the page.
2745          */
2746         if (reiserfs_file_data_log(inode)) {
2747                 /*
2748                  * very conservative, leave the buffer pinned if
2749                  * anyone might need it.
2750                  */
2751                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2752                         ret = 0;
2753                 }
2754         } else  if (buffer_dirty(bh)) {
2755                 struct reiserfs_journal_list *jl;
2756                 struct reiserfs_jh *jh = bh->b_private;
2757
2758                 /* why is this safe?
2759                  * reiserfs_setattr updates i_size in the on disk
2760                  * stat data before allowing vmtruncate to be called.
2761                  *
2762                  * If buffer was put onto the ordered list for this
2763                  * transaction, we know for sure either this transaction
2764                  * or an older one already has updated i_size on disk,
2765                  * and this ordered data won't be referenced in the file
2766                  * if we crash.
2767                  *
2768                  * if the buffer was put onto the ordered list for an older
2769                  * transaction, we need to leave it around
2770                  */
2771                 if (jh && (jl = jh->jl)
2772                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2773                         ret = 0;
2774         }
2775       free_jh:
2776         if (ret && bh->b_private) {
2777                 reiserfs_free_jh(bh);
2778         }
2779         spin_unlock(&j->j_dirty_buffers_lock);
2780         unlock_buffer(bh);
2781         return ret;
2782 }
2783
2784 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2785 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2786 {
2787         struct buffer_head *head, *bh, *next;
2788         struct inode *inode = page->mapping->host;
2789         unsigned int curr_off = 0;
2790         int ret = 1;
2791
2792         BUG_ON(!PageLocked(page));
2793
2794         if (offset == 0)
2795                 ClearPageChecked(page);
2796
2797         if (!page_has_buffers(page))
2798                 goto out;
2799
2800         head = page_buffers(page);
2801         bh = head;
2802         do {
2803                 unsigned int next_off = curr_off + bh->b_size;
2804                 next = bh->b_this_page;
2805
2806                 /*
2807                  * is this block fully invalidated?
2808                  */
2809                 if (offset <= curr_off) {
2810                         if (invalidatepage_can_drop(inode, bh))
2811                                 reiserfs_unmap_buffer(bh);
2812                         else
2813                                 ret = 0;
2814                 }
2815                 curr_off = next_off;
2816                 bh = next;
2817         } while (bh != head);
2818
2819         /*
2820          * We release buffers only if the entire page is being invalidated.
2821          * The get_block cached value has been unconditionally invalidated,
2822          * so real IO is not possible anymore.
2823          */
2824         if (!offset && ret) {
2825                 ret = try_to_release_page(page, 0);
2826                 /* maybe should BUG_ON(!ret); - neilb */
2827         }
2828       out:
2829         return;
2830 }
2831
2832 static int reiserfs_set_page_dirty(struct page *page)
2833 {
2834         struct inode *inode = page->mapping->host;
2835         if (reiserfs_file_data_log(inode)) {
2836                 SetPageChecked(page);
2837                 return __set_page_dirty_nobuffers(page);
2838         }
2839         return __set_page_dirty_buffers(page);
2840 }
2841
2842 /*
2843  * Returns 1 if the page's buffers were dropped.  The page is locked.
2844  *
2845  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2846  * in the buffers at page_buffers(page).
2847  *
2848  * even in -o notail mode, we can't be sure an old mount without -o notail
2849  * didn't create files with tails.
2850  */
2851 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2852 {
2853         struct inode *inode = page->mapping->host;
2854         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2855         struct buffer_head *head;
2856         struct buffer_head *bh;
2857         int ret = 1;
2858
2859         WARN_ON(PageChecked(page));
2860         spin_lock(&j->j_dirty_buffers_lock);
2861         head = page_buffers(page);
2862         bh = head;
2863         do {
2864                 if (bh->b_private) {
2865                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2866                                 reiserfs_free_jh(bh);
2867                         } else {
2868                                 ret = 0;
2869                                 break;
2870                         }
2871                 }
2872                 bh = bh->b_this_page;
2873         } while (bh != head);
2874         if (ret)
2875                 ret = try_to_free_buffers(page);
2876         spin_unlock(&j->j_dirty_buffers_lock);
2877         return ret;
2878 }
2879
2880 /* We thank Mingming Cao for helping us understand in great detail what
2881    to do in this section of the code. */
2882 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2883                                   const struct iovec *iov, loff_t offset,
2884                                   unsigned long nr_segs)
2885 {
2886         struct file *file = iocb->ki_filp;
2887         struct inode *inode = file->f_mapping->host;
2888
2889         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2890                                   offset, nr_segs,
2891                                   reiserfs_get_blocks_direct_io, NULL);
2892 }
2893
2894 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2895 {
2896         struct inode *inode = dentry->d_inode;
2897         int error;
2898         unsigned int ia_valid = attr->ia_valid;
2899         reiserfs_write_lock(inode->i_sb);
2900         if (attr->ia_valid & ATTR_SIZE) {
2901                 /* version 2 items will be caught by the s_maxbytes check
2902                  ** done for us in vmtruncate
2903                  */
2904                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2905                     attr->ia_size > MAX_NON_LFS) {
2906                         error = -EFBIG;
2907                         goto out;
2908                 }
2909                 /* fill in hole pointers in the expanding truncate case. */
2910                 if (attr->ia_size > inode->i_size) {
2911                         error = generic_cont_expand(inode, attr->ia_size);
2912                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2913                                 int err;
2914                                 struct reiserfs_transaction_handle th;
2915                                 /* we're changing at most 2 bitmaps, inode + super */
2916                                 err = journal_begin(&th, inode->i_sb, 4);
2917                                 if (!err) {
2918                                         reiserfs_discard_prealloc(&th, inode);
2919                                         err = journal_end(&th, inode->i_sb, 4);
2920                                 }
2921                                 if (err)
2922                                         error = err;
2923                         }
2924                         if (error)
2925                                 goto out;
2926                         /*
2927                          * file size is changed, ctime and mtime are
2928                          * to be updated
2929                          */
2930                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
2931                 }
2932         }
2933
2934         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2935              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2936             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2937                 /* stat data of format v3.5 has 16 bit uid and gid */
2938                 error = -EINVAL;
2939                 goto out;
2940         }
2941
2942         error = inode_change_ok(inode, attr);
2943         if (!error) {
2944                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2945                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2946                         error = reiserfs_chown_xattrs(inode, attr);
2947
2948                         if (!error) {
2949                                 struct reiserfs_transaction_handle th;
2950                                 int jbegin_count =
2951                                     2 *
2952                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2953                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2954                                     2;
2955
2956                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2957                                 error =
2958                                     journal_begin(&th, inode->i_sb,
2959                                                   jbegin_count);
2960                                 if (error)
2961                                         goto out;
2962                                 error =
2963                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2964                                 if (error) {
2965                                         journal_end(&th, inode->i_sb,
2966                                                     jbegin_count);
2967                                         goto out;
2968                                 }
2969                                 /* Update corresponding info in inode so that everything is in
2970                                  * one transaction */
2971                                 if (attr->ia_valid & ATTR_UID)
2972                                         inode->i_uid = attr->ia_uid;
2973                                 if (attr->ia_valid & ATTR_GID)
2974                                         inode->i_gid = attr->ia_gid;
2975                                 mark_inode_dirty(inode);
2976                                 error =
2977                                     journal_end(&th, inode->i_sb, jbegin_count);
2978                         }
2979                 }
2980                 if (!error)
2981                         error = inode_setattr(inode, attr);
2982         }
2983
2984         if (!error && reiserfs_posixacl(inode->i_sb)) {
2985                 if (attr->ia_valid & ATTR_MODE)
2986                         error = reiserfs_acl_chmod(inode);
2987         }
2988
2989       out:
2990         reiserfs_write_unlock(inode->i_sb);
2991         return error;
2992 }
2993
2994 const struct address_space_operations reiserfs_address_space_operations = {
2995         .writepage = reiserfs_writepage,
2996         .readpage = reiserfs_readpage,
2997         .readpages = reiserfs_readpages,
2998         .releasepage = reiserfs_releasepage,
2999         .invalidatepage = reiserfs_invalidatepage,
3000         .sync_page = block_sync_page,
3001         .prepare_write = reiserfs_prepare_write,
3002         .commit_write = reiserfs_commit_write,
3003         .bmap = reiserfs_aop_bmap,
3004         .direct_IO = reiserfs_direct_IO,
3005         .set_page_dirty = reiserfs_set_page_dirty,
3006 };