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