[PATCH] resierfs: fix reiserfs_invalidatepage race against data=ordered
[linux-2.6.git] / fs / reiserfs / inode.c
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4
5 #include <linux/config.h>
6 #include <linux/time.h>
7 #include <linux/fs.h>
8 #include <linux/reiserfs_fs.h>
9 #include <linux/reiserfs_acl.h>
10 #include <linux/reiserfs_xattr.h>
11 #include <linux/smp_lock.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <asm/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
21 extern int reiserfs_default_io_size;    /* default io size devuned in super.c */
22
23 static int reiserfs_commit_write(struct file *f, struct page *page,
24                                  unsigned from, unsigned to);
25 static int reiserfs_prepare_write(struct file *f, struct page *page,
26                                   unsigned from, unsigned to);
27
28 void reiserfs_delete_inode(struct inode *inode)
29 {
30         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
31         int jbegin_count =
32             JOURNAL_PER_BALANCE_CNT * 2 +
33             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
34         struct reiserfs_transaction_handle th;
35         int err;
36
37         truncate_inode_pages(&inode->i_data, 0);
38
39         reiserfs_write_lock(inode->i_sb);
40
41         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
42         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
43                 mutex_lock(&inode->i_mutex);
44
45                 reiserfs_delete_xattrs(inode);
46
47                 if (journal_begin(&th, inode->i_sb, jbegin_count)) {
48                         mutex_unlock(&inode->i_mutex);
49                         goto out;
50                 }
51                 reiserfs_update_inode_transaction(inode);
52
53                 err = reiserfs_delete_object(&th, inode);
54
55                 /* Do quota update inside a transaction for journaled quotas. We must do that
56                  * after delete_object so that quota updates go into the same transaction as
57                  * stat data deletion */
58                 if (!err) 
59                         DQUOT_FREE_INODE(inode);
60
61                 if (journal_end(&th, inode->i_sb, jbegin_count)) {
62                         mutex_unlock(&inode->i_mutex);
63                         goto out;
64                 }
65
66                 mutex_unlock(&inode->i_mutex);
67
68                 /* check return value from reiserfs_delete_object after
69                  * ending the transaction
70                  */
71                 if (err)
72                     goto out;
73
74                 /* all items of file are deleted, so we can remove "save" link */
75                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
76                                                                  * about an error here */
77         } else {
78                 /* no object items are in the tree */
79                 ;
80         }
81       out:
82         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
83         inode->i_blocks = 0;
84         reiserfs_write_unlock(inode->i_sb);
85 }
86
87 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
88                           __u32 objectid, loff_t offset, int type, int length)
89 {
90         key->version = version;
91
92         key->on_disk_key.k_dir_id = dirid;
93         key->on_disk_key.k_objectid = objectid;
94         set_cpu_key_k_offset(key, offset);
95         set_cpu_key_k_type(key, type);
96         key->key_length = length;
97 }
98
99 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
100    offset and type of key */
101 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
102                   int type, int length)
103 {
104         _make_cpu_key(key, get_inode_item_key_version(inode),
105                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
106                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
107                       length);
108 }
109
110 //
111 // when key is 0, do not set version and short key
112 //
113 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
114                               int version,
115                               loff_t offset, int type, int length,
116                               int entry_count /*or ih_free_space */ )
117 {
118         if (key) {
119                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
120                 ih->ih_key.k_objectid =
121                     cpu_to_le32(key->on_disk_key.k_objectid);
122         }
123         put_ih_version(ih, version);
124         set_le_ih_k_offset(ih, offset);
125         set_le_ih_k_type(ih, type);
126         put_ih_item_len(ih, length);
127         /*    set_ih_free_space (ih, 0); */
128         // for directory items it is entry count, for directs and stat
129         // datas - 0xffff, for indirects - 0
130         put_ih_entry_count(ih, entry_count);
131 }
132
133 //
134 // FIXME: we might cache recently accessed indirect item
135
136 // Ugh.  Not too eager for that....
137 //  I cut the code until such time as I see a convincing argument (benchmark).
138 // I don't want a bloated inode struct..., and I don't like code complexity....
139
140 /* cutting the code is fine, since it really isn't in use yet and is easy
141 ** to add back in.  But, Vladimir has a really good idea here.  Think
142 ** about what happens for reading a file.  For each page,
143 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
144 ** an indirect item.  This indirect item has X number of pointers, where
145 ** X is a big number if we've done the block allocation right.  But,
146 ** we only use one or two of these pointers during each call to readpage,
147 ** needlessly researching again later on.
148 **
149 ** The size of the cache could be dynamic based on the size of the file.
150 **
151 ** I'd also like to see us cache the location the stat data item, since
152 ** we are needlessly researching for that frequently.
153 **
154 ** --chris
155 */
156
157 /* If this page has a file tail in it, and
158 ** it was read in by get_block_create_0, the page data is valid,
159 ** but tail is still sitting in a direct item, and we can't write to
160 ** it.  So, look through this page, and check all the mapped buffers
161 ** to make sure they have valid block numbers.  Any that don't need
162 ** to be unmapped, so that block_prepare_write will correctly call
163 ** reiserfs_get_block to convert the tail into an unformatted node
164 */
165 static inline void fix_tail_page_for_writing(struct page *page)
166 {
167         struct buffer_head *head, *next, *bh;
168
169         if (page && page_has_buffers(page)) {
170                 head = page_buffers(page);
171                 bh = head;
172                 do {
173                         next = bh->b_this_page;
174                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
175                                 reiserfs_unmap_buffer(bh);
176                         }
177                         bh = next;
178                 } while (bh != head);
179         }
180 }
181
182 /* reiserfs_get_block does not need to allocate a block only if it has been
183    done already or non-hole position has been found in the indirect item */
184 static inline int allocation_needed(int retval, b_blocknr_t allocated,
185                                     struct item_head *ih,
186                                     __le32 * item, int pos_in_item)
187 {
188         if (allocated)
189                 return 0;
190         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
191             get_block_num(item, pos_in_item))
192                 return 0;
193         return 1;
194 }
195
196 static inline int indirect_item_found(int retval, struct item_head *ih)
197 {
198         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
199 }
200
201 static inline void set_block_dev_mapped(struct buffer_head *bh,
202                                         b_blocknr_t block, struct inode *inode)
203 {
204         map_bh(bh, inode->i_sb, block);
205 }
206
207 //
208 // files which were created in the earlier version can not be longer,
209 // than 2 gb
210 //
211 static int file_capable(struct inode *inode, long block)
212 {
213         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
214             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
215                 return 1;
216
217         return 0;
218 }
219
220 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
221                                    struct inode *inode, struct path *path)
222 {
223         struct super_block *s = th->t_super;
224         int len = th->t_blocks_allocated;
225         int err;
226
227         BUG_ON(!th->t_trans_id);
228         BUG_ON(!th->t_refcount);
229
230         /* we cannot restart while nested */
231         if (th->t_refcount > 1) {
232                 return 0;
233         }
234         pathrelse(path);
235         reiserfs_update_sd(th, inode);
236         err = journal_end(th, s, len);
237         if (!err) {
238                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
239                 if (!err)
240                         reiserfs_update_inode_transaction(inode);
241         }
242         return err;
243 }
244
245 // it is called by get_block when create == 0. Returns block number
246 // for 'block'-th logical block of file. When it hits direct item it
247 // returns 0 (being called from bmap) or read direct item into piece
248 // of page (bh_result)
249
250 // Please improve the english/clarity in the comment above, as it is
251 // hard to understand.
252
253 static int _get_block_create_0(struct inode *inode, long block,
254                                struct buffer_head *bh_result, int args)
255 {
256         INITIALIZE_PATH(path);
257         struct cpu_key key;
258         struct buffer_head *bh;
259         struct item_head *ih, tmp_ih;
260         int fs_gen;
261         int blocknr;
262         char *p = NULL;
263         int chars;
264         int ret;
265         int result;
266         int done = 0;
267         unsigned long offset;
268
269         // prepare the key to look for the 'block'-th block of file
270         make_cpu_key(&key, inode,
271                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
272                      3);
273
274       research:
275         result = search_for_position_by_key(inode->i_sb, &key, &path);
276         if (result != POSITION_FOUND) {
277                 pathrelse(&path);
278                 if (p)
279                         kunmap(bh_result->b_page);
280                 if (result == IO_ERROR)
281                         return -EIO;
282                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
283                 // That there is some MMAPED data associated with it that is yet to be written to disk.
284                 if ((args & GET_BLOCK_NO_HOLE)
285                     && !PageUptodate(bh_result->b_page)) {
286                         return -ENOENT;
287                 }
288                 return 0;
289         }
290         //
291         bh = get_last_bh(&path);
292         ih = get_ih(&path);
293         if (is_indirect_le_ih(ih)) {
294                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
295
296                 /* FIXME: here we could cache indirect item or part of it in
297                    the inode to avoid search_by_key in case of subsequent
298                    access to file */
299                 blocknr = get_block_num(ind_item, path.pos_in_item);
300                 ret = 0;
301                 if (blocknr) {
302                         map_bh(bh_result, inode->i_sb, blocknr);
303                         if (path.pos_in_item ==
304                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
305                                 set_buffer_boundary(bh_result);
306                         }
307                 } else
308                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
309                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
310                 if ((args & GET_BLOCK_NO_HOLE)
311                             && !PageUptodate(bh_result->b_page)) {
312                         ret = -ENOENT;
313                 }
314
315                 pathrelse(&path);
316                 if (p)
317                         kunmap(bh_result->b_page);
318                 return ret;
319         }
320         // requested data are in direct item(s)
321         if (!(args & GET_BLOCK_READ_DIRECT)) {
322                 // we are called by bmap. FIXME: we can not map block of file
323                 // when it is stored in direct item(s)
324                 pathrelse(&path);
325                 if (p)
326                         kunmap(bh_result->b_page);
327                 return -ENOENT;
328         }
329
330         /* if we've got a direct item, and the buffer or page was uptodate,
331          ** we don't want to pull data off disk again.  skip to the
332          ** end, where we map the buffer and return
333          */
334         if (buffer_uptodate(bh_result)) {
335                 goto finished;
336         } else
337                 /*
338                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
339                  ** pages without any buffers.  If the page is up to date, we don't want
340                  ** read old data off disk.  Set the up to date bit on the buffer instead
341                  ** and jump to the end
342                  */
343         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
344                 set_buffer_uptodate(bh_result);
345                 goto finished;
346         }
347         // read file tail into part of page
348         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
349         fs_gen = get_generation(inode->i_sb);
350         copy_item_head(&tmp_ih, ih);
351
352         /* we only want to kmap if we are reading the tail into the page.
353          ** this is not the common case, so we don't kmap until we are
354          ** sure we need to.  But, this means the item might move if
355          ** kmap schedules
356          */
357         if (!p) {
358                 p = (char *)kmap(bh_result->b_page);
359                 if (fs_changed(fs_gen, inode->i_sb)
360                     && item_moved(&tmp_ih, &path)) {
361                         goto research;
362                 }
363         }
364         p += offset;
365         memset(p, 0, inode->i_sb->s_blocksize);
366         do {
367                 if (!is_direct_le_ih(ih)) {
368                         BUG();
369                 }
370                 /* make sure we don't read more bytes than actually exist in
371                  ** the file.  This can happen in odd cases where i_size isn't
372                  ** correct, and when direct item padding results in a few 
373                  ** extra bytes at the end of the direct item
374                  */
375                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
376                         break;
377                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
378                         chars =
379                             inode->i_size - (le_ih_k_offset(ih) - 1) -
380                             path.pos_in_item;
381                         done = 1;
382                 } else {
383                         chars = ih_item_len(ih) - path.pos_in_item;
384                 }
385                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
386
387                 if (done)
388                         break;
389
390                 p += chars;
391
392                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
393                         // we done, if read direct item is not the last item of
394                         // node FIXME: we could try to check right delimiting key
395                         // to see whether direct item continues in the right
396                         // neighbor or rely on i_size
397                         break;
398
399                 // update key to look for the next piece
400                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
401                 result = search_for_position_by_key(inode->i_sb, &key, &path);
402                 if (result != POSITION_FOUND)
403                         // i/o error most likely
404                         break;
405                 bh = get_last_bh(&path);
406                 ih = get_ih(&path);
407         } while (1);
408
409         flush_dcache_page(bh_result->b_page);
410         kunmap(bh_result->b_page);
411
412       finished:
413         pathrelse(&path);
414
415         if (result == IO_ERROR)
416                 return -EIO;
417
418         /* this buffer has valid data, but isn't valid for io.  mapping it to
419          * block #0 tells the rest of reiserfs it just has a tail in it
420          */
421         map_bh(bh_result, inode->i_sb, 0);
422         set_buffer_uptodate(bh_result);
423         return 0;
424 }
425
426 // this is called to create file map. So, _get_block_create_0 will not
427 // read direct item
428 static int reiserfs_bmap(struct inode *inode, sector_t block,
429                          struct buffer_head *bh_result, int create)
430 {
431         if (!file_capable(inode, block))
432                 return -EFBIG;
433
434         reiserfs_write_lock(inode->i_sb);
435         /* do not read the direct item */
436         _get_block_create_0(inode, block, bh_result, 0);
437         reiserfs_write_unlock(inode->i_sb);
438         return 0;
439 }
440
441 /* special version of get_block that is only used by grab_tail_page right
442 ** now.  It is sent to block_prepare_write, and when you try to get a
443 ** block past the end of the file (or a block from a hole) it returns
444 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
445 ** be able to do i/o on the buffers returned, unless an error value
446 ** is also returned.
447 ** 
448 ** So, this allows block_prepare_write to be used for reading a single block
449 ** in a page.  Where it does not produce a valid page for holes, or past the
450 ** end of the file.  This turns out to be exactly what we need for reading
451 ** tails for conversion.
452 **
453 ** The point of the wrapper is forcing a certain value for create, even
454 ** though the VFS layer is calling this function with create==1.  If you 
455 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block, 
456 ** don't use this function.
457 */
458 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
459                                        struct buffer_head *bh_result,
460                                        int create)
461 {
462         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
463 }
464
465 /* This is special helper for reiserfs_get_block in case we are executing
466    direct_IO request. */
467 static int reiserfs_get_blocks_direct_io(struct inode *inode,
468                                          sector_t iblock,
469                                          unsigned long max_blocks,
470                                          struct buffer_head *bh_result,
471                                          int create)
472 {
473         int ret;
474
475         bh_result->b_page = NULL;
476
477         /* We set the b_size before reiserfs_get_block call since it is
478            referenced in convert_tail_for_hole() that may be called from
479            reiserfs_get_block() */
480         bh_result->b_size = (1 << inode->i_blkbits);
481
482         ret = reiserfs_get_block(inode, iblock, bh_result,
483                                  create | GET_BLOCK_NO_DANGLE);
484         if (ret)
485                 goto out;
486
487         /* don't allow direct io onto tail pages */
488         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
489                 /* make sure future calls to the direct io funcs for this offset
490                  ** in the file fail by unmapping the buffer
491                  */
492                 clear_buffer_mapped(bh_result);
493                 ret = -EINVAL;
494         }
495         /* Possible unpacked tail. Flush the data before pages have
496            disappeared */
497         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
498                 int err;
499                 lock_kernel();
500                 err = reiserfs_commit_for_inode(inode);
501                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
502                 unlock_kernel();
503                 if (err < 0)
504                         ret = err;
505         }
506       out:
507         return ret;
508 }
509
510 /*
511 ** helper function for when reiserfs_get_block is called for a hole
512 ** but the file tail is still in a direct item
513 ** bh_result is the buffer head for the hole
514 ** tail_offset is the offset of the start of the tail in the file
515 **
516 ** This calls prepare_write, which will start a new transaction
517 ** you should not be in a transaction, or have any paths held when you
518 ** call this.
519 */
520 static int convert_tail_for_hole(struct inode *inode,
521                                  struct buffer_head *bh_result,
522                                  loff_t tail_offset)
523 {
524         unsigned long index;
525         unsigned long tail_end;
526         unsigned long tail_start;
527         struct page *tail_page;
528         struct page *hole_page = bh_result->b_page;
529         int retval = 0;
530
531         if ((tail_offset & (bh_result->b_size - 1)) != 1)
532                 return -EIO;
533
534         /* always try to read until the end of the block */
535         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
536         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
537
538         index = tail_offset >> PAGE_CACHE_SHIFT;
539         /* hole_page can be zero in case of direct_io, we are sure
540            that we cannot get here if we write with O_DIRECT into
541            tail page */
542         if (!hole_page || index != hole_page->index) {
543                 tail_page = grab_cache_page(inode->i_mapping, index);
544                 retval = -ENOMEM;
545                 if (!tail_page) {
546                         goto out;
547                 }
548         } else {
549                 tail_page = hole_page;
550         }
551
552         /* we don't have to make sure the conversion did not happen while
553          ** we were locking the page because anyone that could convert
554          ** must first take i_mutex.
555          **
556          ** We must fix the tail page for writing because it might have buffers
557          ** that are mapped, but have a block number of 0.  This indicates tail
558          ** data that has been read directly into the page, and block_prepare_write
559          ** won't trigger a get_block in this case.
560          */
561         fix_tail_page_for_writing(tail_page);
562         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
563         if (retval)
564                 goto unlock;
565
566         /* tail conversion might change the data in the page */
567         flush_dcache_page(tail_page);
568
569         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
570
571       unlock:
572         if (tail_page != hole_page) {
573                 unlock_page(tail_page);
574                 page_cache_release(tail_page);
575         }
576       out:
577         return retval;
578 }
579
580 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
581                                   long block,
582                                   struct inode *inode,
583                                   b_blocknr_t * allocated_block_nr,
584                                   struct path *path, int flags)
585 {
586         BUG_ON(!th->t_trans_id);
587
588 #ifdef REISERFS_PREALLOCATE
589         if (!(flags & GET_BLOCK_NO_IMUX)) {
590                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
591                                                   path, block);
592         }
593 #endif
594         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
595                                          block);
596 }
597
598 int reiserfs_get_block(struct inode *inode, sector_t block,
599                        struct buffer_head *bh_result, int create)
600 {
601         int repeat, retval = 0;
602         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
603         INITIALIZE_PATH(path);
604         int pos_in_item;
605         struct cpu_key key;
606         struct buffer_head *bh, *unbh = NULL;
607         struct item_head *ih, tmp_ih;
608         __le32 *item;
609         int done;
610         int fs_gen;
611         struct reiserfs_transaction_handle *th = NULL;
612         /* space reserved in transaction batch: 
613            . 3 balancings in direct->indirect conversion
614            . 1 block involved into reiserfs_update_sd()
615            XXX in practically impossible worst case direct2indirect()
616            can incur (much) more than 3 balancings.
617            quota update for user, group */
618         int jbegin_count =
619             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
620             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
621         int version;
622         int dangle = 1;
623         loff_t new_offset =
624             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
625
626         /* bad.... */
627         reiserfs_write_lock(inode->i_sb);
628         version = get_inode_item_key_version(inode);
629
630         if (block < 0) {
631                 reiserfs_write_unlock(inode->i_sb);
632                 return -EIO;
633         }
634
635         if (!file_capable(inode, block)) {
636                 reiserfs_write_unlock(inode->i_sb);
637                 return -EFBIG;
638         }
639
640         /* if !create, we aren't changing the FS, so we don't need to
641          ** log anything, so we don't need to start a transaction
642          */
643         if (!(create & GET_BLOCK_CREATE)) {
644                 int ret;
645                 /* find number of block-th logical block of the file */
646                 ret = _get_block_create_0(inode, block, bh_result,
647                                           create | GET_BLOCK_READ_DIRECT);
648                 reiserfs_write_unlock(inode->i_sb);
649                 return ret;
650         }
651         /*
652          * if we're already in a transaction, make sure to close
653          * any new transactions we start in this func
654          */
655         if ((create & GET_BLOCK_NO_DANGLE) ||
656             reiserfs_transaction_running(inode->i_sb))
657                 dangle = 0;
658
659         /* If file is of such a size, that it might have a tail and tails are enabled
660          ** we should mark it as possibly needing tail packing on close
661          */
662         if ((have_large_tails(inode->i_sb)
663              && inode->i_size < i_block_size(inode) * 4)
664             || (have_small_tails(inode->i_sb)
665                 && inode->i_size < i_block_size(inode)))
666                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
667
668         /* set the key of the first byte in the 'block'-th block of file */
669         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
670         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
671               start_trans:
672                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
673                 if (!th) {
674                         retval = -ENOMEM;
675                         goto failure;
676                 }
677                 reiserfs_update_inode_transaction(inode);
678         }
679       research:
680
681         retval = search_for_position_by_key(inode->i_sb, &key, &path);
682         if (retval == IO_ERROR) {
683                 retval = -EIO;
684                 goto failure;
685         }
686
687         bh = get_last_bh(&path);
688         ih = get_ih(&path);
689         item = get_item(&path);
690         pos_in_item = path.pos_in_item;
691
692         fs_gen = get_generation(inode->i_sb);
693         copy_item_head(&tmp_ih, ih);
694
695         if (allocation_needed
696             (retval, allocated_block_nr, ih, item, pos_in_item)) {
697                 /* we have to allocate block for the unformatted node */
698                 if (!th) {
699                         pathrelse(&path);
700                         goto start_trans;
701                 }
702
703                 repeat =
704                     _allocate_block(th, block, inode, &allocated_block_nr,
705                                     &path, create);
706
707                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
708                         /* restart the transaction to give the journal a chance to free
709                          ** some blocks.  releases the path, so we have to go back to
710                          ** research if we succeed on the second try
711                          */
712                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
713                         retval = restart_transaction(th, inode, &path);
714                         if (retval)
715                                 goto failure;
716                         repeat =
717                             _allocate_block(th, block, inode,
718                                             &allocated_block_nr, NULL, create);
719
720                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
721                                 goto research;
722                         }
723                         if (repeat == QUOTA_EXCEEDED)
724                                 retval = -EDQUOT;
725                         else
726                                 retval = -ENOSPC;
727                         goto failure;
728                 }
729
730                 if (fs_changed(fs_gen, inode->i_sb)
731                     && item_moved(&tmp_ih, &path)) {
732                         goto research;
733                 }
734         }
735
736         if (indirect_item_found(retval, ih)) {
737                 b_blocknr_t unfm_ptr;
738                 /* 'block'-th block is in the file already (there is
739                    corresponding cell in some indirect item). But it may be
740                    zero unformatted node pointer (hole) */
741                 unfm_ptr = get_block_num(item, pos_in_item);
742                 if (unfm_ptr == 0) {
743                         /* use allocated block to plug the hole */
744                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
745                         if (fs_changed(fs_gen, inode->i_sb)
746                             && item_moved(&tmp_ih, &path)) {
747                                 reiserfs_restore_prepared_buffer(inode->i_sb,
748                                                                  bh);
749                                 goto research;
750                         }
751                         set_buffer_new(bh_result);
752                         if (buffer_dirty(bh_result)
753                             && reiserfs_data_ordered(inode->i_sb))
754                                 reiserfs_add_ordered_list(inode, bh_result);
755                         put_block_num(item, pos_in_item, allocated_block_nr);
756                         unfm_ptr = allocated_block_nr;
757                         journal_mark_dirty(th, inode->i_sb, bh);
758                         reiserfs_update_sd(th, inode);
759                 }
760                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
761                 pathrelse(&path);
762                 retval = 0;
763                 if (!dangle && th)
764                         retval = reiserfs_end_persistent_transaction(th);
765
766                 reiserfs_write_unlock(inode->i_sb);
767
768                 /* the item was found, so new blocks were not added to the file
769                  ** there is no need to make sure the inode is updated with this 
770                  ** transaction
771                  */
772                 return retval;
773         }
774
775         if (!th) {
776                 pathrelse(&path);
777                 goto start_trans;
778         }
779
780         /* desired position is not found or is in the direct item. We have
781            to append file with holes up to 'block'-th block converting
782            direct items to indirect one if necessary */
783         done = 0;
784         do {
785                 if (is_statdata_le_ih(ih)) {
786                         __le32 unp = 0;
787                         struct cpu_key tmp_key;
788
789                         /* indirect item has to be inserted */
790                         make_le_item_head(&tmp_ih, &key, version, 1,
791                                           TYPE_INDIRECT, UNFM_P_SIZE,
792                                           0 /* free_space */ );
793
794                         if (cpu_key_k_offset(&key) == 1) {
795                                 /* we are going to add 'block'-th block to the file. Use
796                                    allocated block for that */
797                                 unp = cpu_to_le32(allocated_block_nr);
798                                 set_block_dev_mapped(bh_result,
799                                                      allocated_block_nr, inode);
800                                 set_buffer_new(bh_result);
801                                 done = 1;
802                         }
803                         tmp_key = key;  // ;)
804                         set_cpu_key_k_offset(&tmp_key, 1);
805                         PATH_LAST_POSITION(&path)++;
806
807                         retval =
808                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
809                                                  inode, (char *)&unp);
810                         if (retval) {
811                                 reiserfs_free_block(th, inode,
812                                                     allocated_block_nr, 1);
813                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
814                         }
815                         //mark_tail_converted (inode);
816                 } else if (is_direct_le_ih(ih)) {
817                         /* direct item has to be converted */
818                         loff_t tail_offset;
819
820                         tail_offset =
821                             ((le_ih_k_offset(ih) -
822                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
823                         if (tail_offset == cpu_key_k_offset(&key)) {
824                                 /* direct item we just found fits into block we have
825                                    to map. Convert it into unformatted node: use
826                                    bh_result for the conversion */
827                                 set_block_dev_mapped(bh_result,
828                                                      allocated_block_nr, inode);
829                                 unbh = bh_result;
830                                 done = 1;
831                         } else {
832                                 /* we have to padd file tail stored in direct item(s)
833                                    up to block size and convert it to unformatted
834                                    node. FIXME: this should also get into page cache */
835
836                                 pathrelse(&path);
837                                 /*
838                                  * ugly, but we can only end the transaction if
839                                  * we aren't nested
840                                  */
841                                 BUG_ON(!th->t_refcount);
842                                 if (th->t_refcount == 1) {
843                                         retval =
844                                             reiserfs_end_persistent_transaction
845                                             (th);
846                                         th = NULL;
847                                         if (retval)
848                                                 goto failure;
849                                 }
850
851                                 retval =
852                                     convert_tail_for_hole(inode, bh_result,
853                                                           tail_offset);
854                                 if (retval) {
855                                         if (retval != -ENOSPC)
856                                                 reiserfs_warning(inode->i_sb,
857                                                                  "clm-6004: convert tail failed inode %lu, error %d",
858                                                                  inode->i_ino,
859                                                                  retval);
860                                         if (allocated_block_nr) {
861                                                 /* the bitmap, the super, and the stat data == 3 */
862                                                 if (!th)
863                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
864                                                 if (th)
865                                                         reiserfs_free_block(th,
866                                                                             inode,
867                                                                             allocated_block_nr,
868                                                                             1);
869                                         }
870                                         goto failure;
871                                 }
872                                 goto research;
873                         }
874                         retval =
875                             direct2indirect(th, inode, &path, unbh,
876                                             tail_offset);
877                         if (retval) {
878                                 reiserfs_unmap_buffer(unbh);
879                                 reiserfs_free_block(th, inode,
880                                                     allocated_block_nr, 1);
881                                 goto failure;
882                         }
883                         /* it is important the set_buffer_uptodate is done after
884                          ** the direct2indirect.  The buffer might contain valid
885                          ** data newer than the data on disk (read by readpage, changed,
886                          ** and then sent here by writepage).  direct2indirect needs
887                          ** to know if unbh was already up to date, so it can decide
888                          ** if the data in unbh needs to be replaced with data from
889                          ** the disk
890                          */
891                         set_buffer_uptodate(unbh);
892
893                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
894                            buffer will disappear shortly, so it should not be added to
895                          */
896                         if (unbh->b_page) {
897                                 /* we've converted the tail, so we must
898                                  ** flush unbh before the transaction commits
899                                  */
900                                 reiserfs_add_tail_list(inode, unbh);
901
902                                 /* mark it dirty now to prevent commit_write from adding
903                                  ** this buffer to the inode's dirty buffer list
904                                  */
905                                 /*
906                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
907                                  * It's still atomic, but it sets the page dirty too,
908                                  * which makes it eligible for writeback at any time by the
909                                  * VM (which was also the case with __mark_buffer_dirty())
910                                  */
911                                 mark_buffer_dirty(unbh);
912                         }
913                 } else {
914                         /* append indirect item with holes if needed, when appending
915                            pointer to 'block'-th block use block, which is already
916                            allocated */
917                         struct cpu_key tmp_key;
918                         unp_t unf_single = 0;   // We use this in case we need to allocate only
919                         // one block which is a fastpath
920                         unp_t *un;
921                         __u64 max_to_insert =
922                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
923                             UNFM_P_SIZE;
924                         __u64 blocks_needed;
925
926                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
927                                "vs-804: invalid position for append");
928                         /* indirect item has to be appended, set up key of that position */
929                         make_cpu_key(&tmp_key, inode,
930                                      le_key_k_offset(version,
931                                                      &(ih->ih_key)) +
932                                      op_bytes_number(ih,
933                                                      inode->i_sb->s_blocksize),
934                                      //pos_in_item * inode->i_sb->s_blocksize,
935                                      TYPE_INDIRECT, 3); // key type is unimportant
936
937                         blocks_needed =
938                             1 +
939                             ((cpu_key_k_offset(&key) -
940                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
941                              s_blocksize_bits);
942                         RFALSE(blocks_needed < 0, "green-805: invalid offset");
943
944                         if (blocks_needed == 1) {
945                                 un = &unf_single;
946                         } else {
947                                 un = kmalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
948                                 if (!un) {
949                                         un = &unf_single;
950                                         blocks_needed = 1;
951                                         max_to_insert = 0;
952                                 } else
953                                         memset(un, 0,
954                                                UNFM_P_SIZE * min(blocks_needed,
955                                                                  max_to_insert));
956                         }
957                         if (blocks_needed <= max_to_insert) {
958                                 /* we are going to add target block to the file. Use allocated
959                                    block for that */
960                                 un[blocks_needed - 1] =
961                                     cpu_to_le32(allocated_block_nr);
962                                 set_block_dev_mapped(bh_result,
963                                                      allocated_block_nr, inode);
964                                 set_buffer_new(bh_result);
965                                 done = 1;
966                         } else {
967                                 /* paste hole to the indirect item */
968                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
969                                    only have space for one block */
970                                 blocks_needed =
971                                     max_to_insert ? max_to_insert : 1;
972                         }
973                         retval =
974                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
975                                                      (char *)un,
976                                                      UNFM_P_SIZE *
977                                                      blocks_needed);
978
979                         if (blocks_needed != 1)
980                                 kfree(un);
981
982                         if (retval) {
983                                 reiserfs_free_block(th, inode,
984                                                     allocated_block_nr, 1);
985                                 goto failure;
986                         }
987                         if (!done) {
988                                 /* We need to mark new file size in case this function will be
989                                    interrupted/aborted later on. And we may do this only for
990                                    holes. */
991                                 inode->i_size +=
992                                     inode->i_sb->s_blocksize * blocks_needed;
993                         }
994                 }
995
996                 if (done == 1)
997                         break;
998
999                 /* this loop could log more blocks than we had originally asked
1000                  ** for.  So, we have to allow the transaction to end if it is
1001                  ** too big or too full.  Update the inode so things are 
1002                  ** consistent if we crash before the function returns
1003                  **
1004                  ** release the path so that anybody waiting on the path before
1005                  ** ending their transaction will be able to continue.
1006                  */
1007                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1008                         retval = restart_transaction(th, inode, &path);
1009                         if (retval)
1010                                 goto failure;
1011                 }
1012                 /* inserting indirect pointers for a hole can take a 
1013                  ** long time.  reschedule if needed
1014                  */
1015                 cond_resched();
1016
1017                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1018                 if (retval == IO_ERROR) {
1019                         retval = -EIO;
1020                         goto failure;
1021                 }
1022                 if (retval == POSITION_FOUND) {
1023                         reiserfs_warning(inode->i_sb,
1024                                          "vs-825: reiserfs_get_block: "
1025                                          "%K should not be found", &key);
1026                         retval = -EEXIST;
1027                         if (allocated_block_nr)
1028                                 reiserfs_free_block(th, inode,
1029                                                     allocated_block_nr, 1);
1030                         pathrelse(&path);
1031                         goto failure;
1032                 }
1033                 bh = get_last_bh(&path);
1034                 ih = get_ih(&path);
1035                 item = get_item(&path);
1036                 pos_in_item = path.pos_in_item;
1037         } while (1);
1038
1039         retval = 0;
1040
1041       failure:
1042         if (th && (!dangle || (retval && !th->t_trans_id))) {
1043                 int err;
1044                 if (th->t_trans_id)
1045                         reiserfs_update_sd(th, inode);
1046                 err = reiserfs_end_persistent_transaction(th);
1047                 if (err)
1048                         retval = err;
1049         }
1050
1051         reiserfs_write_unlock(inode->i_sb);
1052         reiserfs_check_path(&path);
1053         return retval;
1054 }
1055
1056 static int
1057 reiserfs_readpages(struct file *file, struct address_space *mapping,
1058                    struct list_head *pages, unsigned nr_pages)
1059 {
1060         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1061 }
1062
1063 /* Compute real number of used bytes by file
1064  * Following three functions can go away when we'll have enough space in stat item
1065  */
1066 static int real_space_diff(struct inode *inode, int sd_size)
1067 {
1068         int bytes;
1069         loff_t blocksize = inode->i_sb->s_blocksize;
1070
1071         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1072                 return sd_size;
1073
1074         /* End of file is also in full block with indirect reference, so round
1075          ** up to the next block.
1076          **
1077          ** there is just no way to know if the tail is actually packed
1078          ** on the file, so we have to assume it isn't.  When we pack the
1079          ** tail, we add 4 bytes to pretend there really is an unformatted
1080          ** node pointer
1081          */
1082         bytes =
1083             ((inode->i_size +
1084               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1085             sd_size;
1086         return bytes;
1087 }
1088
1089 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1090                                         int sd_size)
1091 {
1092         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1093                 return inode->i_size +
1094                     (loff_t) (real_space_diff(inode, sd_size));
1095         }
1096         return ((loff_t) real_space_diff(inode, sd_size)) +
1097             (((loff_t) blocks) << 9);
1098 }
1099
1100 /* Compute number of blocks used by file in ReiserFS counting */
1101 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1102 {
1103         loff_t bytes = inode_get_bytes(inode);
1104         loff_t real_space = real_space_diff(inode, sd_size);
1105
1106         /* keeps fsck and non-quota versions of reiserfs happy */
1107         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1108                 bytes += (loff_t) 511;
1109         }
1110
1111         /* files from before the quota patch might i_blocks such that
1112          ** bytes < real_space.  Deal with that here to prevent it from
1113          ** going negative.
1114          */
1115         if (bytes < real_space)
1116                 return 0;
1117         return (bytes - real_space) >> 9;
1118 }
1119
1120 //
1121 // BAD: new directories have stat data of new type and all other items
1122 // of old type. Version stored in the inode says about body items, so
1123 // in update_stat_data we can not rely on inode, but have to check
1124 // item version directly
1125 //
1126
1127 // called by read_locked_inode
1128 static void init_inode(struct inode *inode, struct path *path)
1129 {
1130         struct buffer_head *bh;
1131         struct item_head *ih;
1132         __u32 rdev;
1133         //int version = ITEM_VERSION_1;
1134
1135         bh = PATH_PLAST_BUFFER(path);
1136         ih = PATH_PITEM_HEAD(path);
1137
1138         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1139         inode->i_blksize = reiserfs_default_io_size;
1140
1141         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1142         REISERFS_I(inode)->i_flags = 0;
1143         REISERFS_I(inode)->i_prealloc_block = 0;
1144         REISERFS_I(inode)->i_prealloc_count = 0;
1145         REISERFS_I(inode)->i_trans_id = 0;
1146         REISERFS_I(inode)->i_jl = NULL;
1147         REISERFS_I(inode)->i_acl_access = NULL;
1148         REISERFS_I(inode)->i_acl_default = NULL;
1149         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1150
1151         if (stat_data_v1(ih)) {
1152                 struct stat_data_v1 *sd =
1153                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1154                 unsigned long blocks;
1155
1156                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1157                 set_inode_sd_version(inode, STAT_DATA_V1);
1158                 inode->i_mode = sd_v1_mode(sd);
1159                 inode->i_nlink = sd_v1_nlink(sd);
1160                 inode->i_uid = sd_v1_uid(sd);
1161                 inode->i_gid = sd_v1_gid(sd);
1162                 inode->i_size = sd_v1_size(sd);
1163                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1164                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1165                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1166                 inode->i_atime.tv_nsec = 0;
1167                 inode->i_ctime.tv_nsec = 0;
1168                 inode->i_mtime.tv_nsec = 0;
1169
1170                 inode->i_blocks = sd_v1_blocks(sd);
1171                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1172                 blocks = (inode->i_size + 511) >> 9;
1173                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1174                 if (inode->i_blocks > blocks) {
1175                         // there was a bug in <=3.5.23 when i_blocks could take negative
1176                         // values. Starting from 3.5.17 this value could even be stored in
1177                         // stat data. For such files we set i_blocks based on file
1178                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1179                         // only updated if file's inode will ever change
1180                         inode->i_blocks = blocks;
1181                 }
1182
1183                 rdev = sd_v1_rdev(sd);
1184                 REISERFS_I(inode)->i_first_direct_byte =
1185                     sd_v1_first_direct_byte(sd);
1186                 /* an early bug in the quota code can give us an odd number for the
1187                  ** block count.  This is incorrect, fix it here.
1188                  */
1189                 if (inode->i_blocks & 1) {
1190                         inode->i_blocks++;
1191                 }
1192                 inode_set_bytes(inode,
1193                                 to_real_used_space(inode, inode->i_blocks,
1194                                                    SD_V1_SIZE));
1195                 /* nopack is initially zero for v1 objects. For v2 objects,
1196                    nopack is initialised from sd_attrs */
1197                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1198         } else {
1199                 // new stat data found, but object may have old items
1200                 // (directories and symlinks)
1201                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1202
1203                 inode->i_mode = sd_v2_mode(sd);
1204                 inode->i_nlink = sd_v2_nlink(sd);
1205                 inode->i_uid = sd_v2_uid(sd);
1206                 inode->i_size = sd_v2_size(sd);
1207                 inode->i_gid = sd_v2_gid(sd);
1208                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1209                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1210                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1211                 inode->i_ctime.tv_nsec = 0;
1212                 inode->i_mtime.tv_nsec = 0;
1213                 inode->i_atime.tv_nsec = 0;
1214                 inode->i_blocks = sd_v2_blocks(sd);
1215                 rdev = sd_v2_rdev(sd);
1216                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1217                         inode->i_generation =
1218                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1219                 else
1220                         inode->i_generation = sd_v2_generation(sd);
1221
1222                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1223                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1224                 else
1225                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1226                 REISERFS_I(inode)->i_first_direct_byte = 0;
1227                 set_inode_sd_version(inode, STAT_DATA_V2);
1228                 inode_set_bytes(inode,
1229                                 to_real_used_space(inode, inode->i_blocks,
1230                                                    SD_V2_SIZE));
1231                 /* read persistent inode attributes from sd and initalise
1232                    generic inode flags from them */
1233                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1234                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1235         }
1236
1237         pathrelse(path);
1238         if (S_ISREG(inode->i_mode)) {
1239                 inode->i_op = &reiserfs_file_inode_operations;
1240                 inode->i_fop = &reiserfs_file_operations;
1241                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1242         } else if (S_ISDIR(inode->i_mode)) {
1243                 inode->i_op = &reiserfs_dir_inode_operations;
1244                 inode->i_fop = &reiserfs_dir_operations;
1245         } else if (S_ISLNK(inode->i_mode)) {
1246                 inode->i_op = &reiserfs_symlink_inode_operations;
1247                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1248         } else {
1249                 inode->i_blocks = 0;
1250                 inode->i_op = &reiserfs_special_inode_operations;
1251                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1252         }
1253 }
1254
1255 // update new stat data with inode fields
1256 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1257 {
1258         struct stat_data *sd_v2 = (struct stat_data *)sd;
1259         __u16 flags;
1260
1261         set_sd_v2_mode(sd_v2, inode->i_mode);
1262         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1263         set_sd_v2_uid(sd_v2, inode->i_uid);
1264         set_sd_v2_size(sd_v2, size);
1265         set_sd_v2_gid(sd_v2, inode->i_gid);
1266         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1267         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1268         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1269         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1270         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1271                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1272         else
1273                 set_sd_v2_generation(sd_v2, inode->i_generation);
1274         flags = REISERFS_I(inode)->i_attrs;
1275         i_attrs_to_sd_attrs(inode, &flags);
1276         set_sd_v2_attrs(sd_v2, flags);
1277 }
1278
1279 // used to copy inode's fields to old stat data
1280 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1281 {
1282         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1283
1284         set_sd_v1_mode(sd_v1, inode->i_mode);
1285         set_sd_v1_uid(sd_v1, inode->i_uid);
1286         set_sd_v1_gid(sd_v1, inode->i_gid);
1287         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1288         set_sd_v1_size(sd_v1, size);
1289         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1290         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1291         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1292
1293         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1294                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1295         else
1296                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1297
1298         // Sigh. i_first_direct_byte is back
1299         set_sd_v1_first_direct_byte(sd_v1,
1300                                     REISERFS_I(inode)->i_first_direct_byte);
1301 }
1302
1303 /* NOTE, you must prepare the buffer head before sending it here,
1304 ** and then log it after the call
1305 */
1306 static void update_stat_data(struct path *path, struct inode *inode,
1307                              loff_t size)
1308 {
1309         struct buffer_head *bh;
1310         struct item_head *ih;
1311
1312         bh = PATH_PLAST_BUFFER(path);
1313         ih = PATH_PITEM_HEAD(path);
1314
1315         if (!is_statdata_le_ih(ih))
1316                 reiserfs_panic(inode->i_sb,
1317                                "vs-13065: update_stat_data: key %k, found item %h",
1318                                INODE_PKEY(inode), ih);
1319
1320         if (stat_data_v1(ih)) {
1321                 // path points to old stat data
1322                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1323         } else {
1324                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1325         }
1326
1327         return;
1328 }
1329
1330 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1331                              struct inode *inode, loff_t size)
1332 {
1333         struct cpu_key key;
1334         INITIALIZE_PATH(path);
1335         struct buffer_head *bh;
1336         int fs_gen;
1337         struct item_head *ih, tmp_ih;
1338         int retval;
1339
1340         BUG_ON(!th->t_trans_id);
1341
1342         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1343
1344         for (;;) {
1345                 int pos;
1346                 /* look for the object's stat data */
1347                 retval = search_item(inode->i_sb, &key, &path);
1348                 if (retval == IO_ERROR) {
1349                         reiserfs_warning(inode->i_sb,
1350                                          "vs-13050: reiserfs_update_sd: "
1351                                          "i/o failure occurred trying to update %K stat data",
1352                                          &key);
1353                         return;
1354                 }
1355                 if (retval == ITEM_NOT_FOUND) {
1356                         pos = PATH_LAST_POSITION(&path);
1357                         pathrelse(&path);
1358                         if (inode->i_nlink == 0) {
1359                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1360                                 return;
1361                         }
1362                         reiserfs_warning(inode->i_sb,
1363                                          "vs-13060: reiserfs_update_sd: "
1364                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1365                                          INODE_PKEY(inode), inode->i_nlink,
1366                                          pos);
1367                         reiserfs_check_path(&path);
1368                         return;
1369                 }
1370
1371                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1372                  ** FS might change.  We have to detect that, and loop back to the
1373                  ** search if the stat data item has moved
1374                  */
1375                 bh = get_last_bh(&path);
1376                 ih = get_ih(&path);
1377                 copy_item_head(&tmp_ih, ih);
1378                 fs_gen = get_generation(inode->i_sb);
1379                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1380                 if (fs_changed(fs_gen, inode->i_sb)
1381                     && item_moved(&tmp_ih, &path)) {
1382                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1383                         continue;       /* Stat_data item has been moved after scheduling. */
1384                 }
1385                 break;
1386         }
1387         update_stat_data(&path, inode, size);
1388         journal_mark_dirty(th, th->t_super, bh);
1389         pathrelse(&path);
1390         return;
1391 }
1392
1393 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1394 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1395 ** and clear the key in the private portion of the inode, otherwise a
1396 ** corresponding iput might try to delete whatever object the inode last
1397 ** represented.
1398 */
1399 static void reiserfs_make_bad_inode(struct inode *inode)
1400 {
1401         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1402         make_bad_inode(inode);
1403 }
1404
1405 //
1406 // initially this function was derived from minix or ext2's analog and
1407 // evolved as the prototype did
1408 //
1409
1410 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1411 {
1412         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1413         inode->i_ino = args->objectid;
1414         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1415         return 0;
1416 }
1417
1418 /* looks for stat data in the tree, and fills up the fields of in-core
1419    inode stat data fields */
1420 void reiserfs_read_locked_inode(struct inode *inode,
1421                                 struct reiserfs_iget_args *args)
1422 {
1423         INITIALIZE_PATH(path_to_sd);
1424         struct cpu_key key;
1425         unsigned long dirino;
1426         int retval;
1427
1428         dirino = args->dirid;
1429
1430         /* set version 1, version 2 could be used too, because stat data
1431            key is the same in both versions */
1432         key.version = KEY_FORMAT_3_5;
1433         key.on_disk_key.k_dir_id = dirino;
1434         key.on_disk_key.k_objectid = inode->i_ino;
1435         key.on_disk_key.k_offset = 0;
1436         key.on_disk_key.k_type = 0;
1437
1438         /* look for the object's stat data */
1439         retval = search_item(inode->i_sb, &key, &path_to_sd);
1440         if (retval == IO_ERROR) {
1441                 reiserfs_warning(inode->i_sb,
1442                                  "vs-13070: reiserfs_read_locked_inode: "
1443                                  "i/o failure occurred trying to find stat data of %K",
1444                                  &key);
1445                 reiserfs_make_bad_inode(inode);
1446                 return;
1447         }
1448         if (retval != ITEM_FOUND) {
1449                 /* a stale NFS handle can trigger this without it being an error */
1450                 pathrelse(&path_to_sd);
1451                 reiserfs_make_bad_inode(inode);
1452                 inode->i_nlink = 0;
1453                 return;
1454         }
1455
1456         init_inode(inode, &path_to_sd);
1457
1458         /* It is possible that knfsd is trying to access inode of a file
1459            that is being removed from the disk by some other thread. As we
1460            update sd on unlink all that is required is to check for nlink
1461            here. This bug was first found by Sizif when debugging
1462            SquidNG/Butterfly, forgotten, and found again after Philippe
1463            Gramoulle <philippe.gramoulle@mmania.com> reproduced it. 
1464
1465            More logical fix would require changes in fs/inode.c:iput() to
1466            remove inode from hash-table _after_ fs cleaned disk stuff up and
1467            in iget() to return NULL if I_FREEING inode is found in
1468            hash-table. */
1469         /* Currently there is one place where it's ok to meet inode with
1470            nlink==0: processing of open-unlinked and half-truncated files
1471            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1472         if ((inode->i_nlink == 0) &&
1473             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1474                 reiserfs_warning(inode->i_sb,
1475                                  "vs-13075: reiserfs_read_locked_inode: "
1476                                  "dead inode read from disk %K. "
1477                                  "This is likely to be race with knfsd. Ignore",
1478                                  &key);
1479                 reiserfs_make_bad_inode(inode);
1480         }
1481
1482         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1483
1484 }
1485
1486 /**
1487  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1488  *
1489  * @inode:    inode from hash table to check
1490  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1491  *
1492  * This function is called by iget5_locked() to distinguish reiserfs inodes
1493  * having the same inode numbers. Such inodes can only exist due to some
1494  * error condition. One of them should be bad. Inodes with identical
1495  * inode numbers (objectids) are distinguished by parent directory ids.
1496  *
1497  */
1498 int reiserfs_find_actor(struct inode *inode, void *opaque)
1499 {
1500         struct reiserfs_iget_args *args;
1501
1502         args = opaque;
1503         /* args is already in CPU order */
1504         return (inode->i_ino == args->objectid) &&
1505             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1506 }
1507
1508 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1509 {
1510         struct inode *inode;
1511         struct reiserfs_iget_args args;
1512
1513         args.objectid = key->on_disk_key.k_objectid;
1514         args.dirid = key->on_disk_key.k_dir_id;
1515         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1516                              reiserfs_find_actor, reiserfs_init_locked_inode,
1517                              (void *)(&args));
1518         if (!inode)
1519                 return ERR_PTR(-ENOMEM);
1520
1521         if (inode->i_state & I_NEW) {
1522                 reiserfs_read_locked_inode(inode, &args);
1523                 unlock_new_inode(inode);
1524         }
1525
1526         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1527                 /* either due to i/o error or a stale NFS handle */
1528                 iput(inode);
1529                 inode = NULL;
1530         }
1531         return inode;
1532 }
1533
1534 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1535 {
1536         __u32 *data = vobjp;
1537         struct cpu_key key;
1538         struct dentry *result;
1539         struct inode *inode;
1540
1541         key.on_disk_key.k_objectid = data[0];
1542         key.on_disk_key.k_dir_id = data[1];
1543         reiserfs_write_lock(sb);
1544         inode = reiserfs_iget(sb, &key);
1545         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1546             data[2] != inode->i_generation) {
1547                 iput(inode);
1548                 inode = NULL;
1549         }
1550         reiserfs_write_unlock(sb);
1551         if (!inode)
1552                 inode = ERR_PTR(-ESTALE);
1553         if (IS_ERR(inode))
1554                 return ERR_PTR(PTR_ERR(inode));
1555         result = d_alloc_anon(inode);
1556         if (!result) {
1557                 iput(inode);
1558                 return ERR_PTR(-ENOMEM);
1559         }
1560         return result;
1561 }
1562
1563 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1564                                   int len, int fhtype,
1565                                   int (*acceptable) (void *contect,
1566                                                      struct dentry * de),
1567                                   void *context)
1568 {
1569         __u32 obj[3], parent[3];
1570
1571         /* fhtype happens to reflect the number of u32s encoded.
1572          * due to a bug in earlier code, fhtype might indicate there
1573          * are more u32s then actually fitted.
1574          * so if fhtype seems to be more than len, reduce fhtype.
1575          * Valid types are:
1576          *   2 - objectid + dir_id - legacy support
1577          *   3 - objectid + dir_id + generation
1578          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1579          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1580          *   6 - as above plus generation of directory
1581          * 6 does not fit in NFSv2 handles
1582          */
1583         if (fhtype > len) {
1584                 if (fhtype != 6 || len != 5)
1585                         reiserfs_warning(sb,
1586                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1587                                          fhtype, len);
1588                 fhtype = 5;
1589         }
1590
1591         obj[0] = data[0];
1592         obj[1] = data[1];
1593         if (fhtype == 3 || fhtype >= 5)
1594                 obj[2] = data[2];
1595         else
1596                 obj[2] = 0;     /* generation number */
1597
1598         if (fhtype >= 4) {
1599                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1600                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1601                 if (fhtype == 6)
1602                         parent[2] = data[5];
1603                 else
1604                         parent[2] = 0;
1605         }
1606         return sb->s_export_op->find_exported_dentry(sb, obj,
1607                                                      fhtype < 4 ? NULL : parent,
1608                                                      acceptable, context);
1609 }
1610
1611 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1612                        int need_parent)
1613 {
1614         struct inode *inode = dentry->d_inode;
1615         int maxlen = *lenp;
1616
1617         if (maxlen < 3)
1618                 return 255;
1619
1620         data[0] = inode->i_ino;
1621         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1622         data[2] = inode->i_generation;
1623         *lenp = 3;
1624         /* no room for directory info? return what we've stored so far */
1625         if (maxlen < 5 || !need_parent)
1626                 return 3;
1627
1628         spin_lock(&dentry->d_lock);
1629         inode = dentry->d_parent->d_inode;
1630         data[3] = inode->i_ino;
1631         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1632         *lenp = 5;
1633         if (maxlen >= 6) {
1634                 data[5] = inode->i_generation;
1635                 *lenp = 6;
1636         }
1637         spin_unlock(&dentry->d_lock);
1638         return *lenp;
1639 }
1640
1641 /* looks for stat data, then copies fields to it, marks the buffer
1642    containing stat data as dirty */
1643 /* reiserfs inodes are never really dirty, since the dirty inode call
1644 ** always logs them.  This call allows the VFS inode marking routines
1645 ** to properly mark inodes for datasync and such, but only actually
1646 ** does something when called for a synchronous update.
1647 */
1648 int reiserfs_write_inode(struct inode *inode, int do_sync)
1649 {
1650         struct reiserfs_transaction_handle th;
1651         int jbegin_count = 1;
1652
1653         if (inode->i_sb->s_flags & MS_RDONLY)
1654                 return -EROFS;
1655         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1656          ** these cases are just when the system needs ram, not when the 
1657          ** inode needs to reach disk for safety, and they can safely be
1658          ** ignored because the altered inode has already been logged.
1659          */
1660         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1661                 reiserfs_write_lock(inode->i_sb);
1662                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1663                         reiserfs_update_sd(&th, inode);
1664                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1665                 }
1666                 reiserfs_write_unlock(inode->i_sb);
1667         }
1668         return 0;
1669 }
1670
1671 /* stat data of new object is inserted already, this inserts the item
1672    containing "." and ".." entries */
1673 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1674                                   struct inode *inode,
1675                                   struct item_head *ih, struct path *path,
1676                                   struct inode *dir)
1677 {
1678         struct super_block *sb = th->t_super;
1679         char empty_dir[EMPTY_DIR_SIZE];
1680         char *body = empty_dir;
1681         struct cpu_key key;
1682         int retval;
1683
1684         BUG_ON(!th->t_trans_id);
1685
1686         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1687                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1688                       TYPE_DIRENTRY, 3 /*key length */ );
1689
1690         /* compose item head for new item. Directories consist of items of
1691            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1692            is done by reiserfs_new_inode */
1693         if (old_format_only(sb)) {
1694                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1695                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1696
1697                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1698                                        ih->ih_key.k_objectid,
1699                                        INODE_PKEY(dir)->k_dir_id,
1700                                        INODE_PKEY(dir)->k_objectid);
1701         } else {
1702                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1703                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1704
1705                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1706                                     ih->ih_key.k_objectid,
1707                                     INODE_PKEY(dir)->k_dir_id,
1708                                     INODE_PKEY(dir)->k_objectid);
1709         }
1710
1711         /* look for place in the tree for new item */
1712         retval = search_item(sb, &key, path);
1713         if (retval == IO_ERROR) {
1714                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1715                                  "i/o failure occurred creating new directory");
1716                 return -EIO;
1717         }
1718         if (retval == ITEM_FOUND) {
1719                 pathrelse(path);
1720                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1721                                  "object with this key exists (%k)",
1722                                  &(ih->ih_key));
1723                 return -EEXIST;
1724         }
1725
1726         /* insert item, that is empty directory item */
1727         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1728 }
1729
1730 /* stat data of object has been inserted, this inserts the item
1731    containing the body of symlink */
1732 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1733                                 struct item_head *ih,
1734                                 struct path *path, const char *symname,
1735                                 int item_len)
1736 {
1737         struct super_block *sb = th->t_super;
1738         struct cpu_key key;
1739         int retval;
1740
1741         BUG_ON(!th->t_trans_id);
1742
1743         _make_cpu_key(&key, KEY_FORMAT_3_5,
1744                       le32_to_cpu(ih->ih_key.k_dir_id),
1745                       le32_to_cpu(ih->ih_key.k_objectid),
1746                       1, TYPE_DIRECT, 3 /*key length */ );
1747
1748         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1749                           0 /*free_space */ );
1750
1751         /* look for place in the tree for new item */
1752         retval = search_item(sb, &key, path);
1753         if (retval == IO_ERROR) {
1754                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1755                                  "i/o failure occurred creating new symlink");
1756                 return -EIO;
1757         }
1758         if (retval == ITEM_FOUND) {
1759                 pathrelse(path);
1760                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1761                                  "object with this key exists (%k)",
1762                                  &(ih->ih_key));
1763                 return -EEXIST;
1764         }
1765
1766         /* insert item, that is body of symlink */
1767         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1768 }
1769
1770 /* inserts the stat data into the tree, and then calls
1771    reiserfs_new_directory (to insert ".", ".." item if new object is
1772    directory) or reiserfs_new_symlink (to insert symlink body if new
1773    object is symlink) or nothing (if new object is regular file) 
1774
1775    NOTE! uid and gid must already be set in the inode.  If we return
1776    non-zero due to an error, we have to drop the quota previously allocated
1777    for the fresh inode.  This can only be done outside a transaction, so
1778    if we return non-zero, we also end the transaction.  */
1779 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1780                        struct inode *dir, int mode, const char *symname,
1781                        /* 0 for regular, EMTRY_DIR_SIZE for dirs, 
1782                           strlen (symname) for symlinks) */
1783                        loff_t i_size, struct dentry *dentry,
1784                        struct inode *inode)
1785 {
1786         struct super_block *sb;
1787         INITIALIZE_PATH(path_to_key);
1788         struct cpu_key key;
1789         struct item_head ih;
1790         struct stat_data sd;
1791         int retval;
1792         int err;
1793
1794         BUG_ON(!th->t_trans_id);
1795
1796         if (DQUOT_ALLOC_INODE(inode)) {
1797                 err = -EDQUOT;
1798                 goto out_end_trans;
1799         }
1800         if (!dir || !dir->i_nlink) {
1801                 err = -EPERM;
1802                 goto out_bad_inode;
1803         }
1804
1805         sb = dir->i_sb;
1806
1807         /* item head of new item */
1808         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1809         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1810         if (!ih.ih_key.k_objectid) {
1811                 err = -ENOMEM;
1812                 goto out_bad_inode;
1813         }
1814         if (old_format_only(sb))
1815                 /* not a perfect generation count, as object ids can be reused, but 
1816                  ** this is as good as reiserfs can do right now.
1817                  ** note that the private part of inode isn't filled in yet, we have
1818                  ** to use the directory.
1819                  */
1820                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1821         else
1822 #if defined( USE_INODE_GENERATION_COUNTER )
1823                 inode->i_generation =
1824                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1825 #else
1826                 inode->i_generation = ++event;
1827 #endif
1828
1829         /* fill stat data */
1830         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1831
1832         /* uid and gid must already be set by the caller for quota init */
1833
1834         /* symlink cannot be immutable or append only, right? */
1835         if (S_ISLNK(inode->i_mode))
1836                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1837
1838         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1839         inode->i_size = i_size;
1840         inode->i_blocks = 0;
1841         inode->i_bytes = 0;
1842         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1843             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1844
1845         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1846         REISERFS_I(inode)->i_flags = 0;
1847         REISERFS_I(inode)->i_prealloc_block = 0;
1848         REISERFS_I(inode)->i_prealloc_count = 0;
1849         REISERFS_I(inode)->i_trans_id = 0;
1850         REISERFS_I(inode)->i_jl = NULL;
1851         REISERFS_I(inode)->i_attrs =
1852             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1853         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1854         REISERFS_I(inode)->i_acl_access = NULL;
1855         REISERFS_I(inode)->i_acl_default = NULL;
1856         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1857
1858         if (old_format_only(sb))
1859                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1860                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1861         else
1862                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1863                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1864
1865         /* key to search for correct place for new stat data */
1866         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1867                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1868                       TYPE_STAT_DATA, 3 /*key length */ );
1869
1870         /* find proper place for inserting of stat data */
1871         retval = search_item(sb, &key, &path_to_key);
1872         if (retval == IO_ERROR) {
1873                 err = -EIO;
1874                 goto out_bad_inode;
1875         }
1876         if (retval == ITEM_FOUND) {
1877                 pathrelse(&path_to_key);
1878                 err = -EEXIST;
1879                 goto out_bad_inode;
1880         }
1881         if (old_format_only(sb)) {
1882                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1883                         pathrelse(&path_to_key);
1884                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1885                         err = -EINVAL;
1886                         goto out_bad_inode;
1887                 }
1888                 inode2sd_v1(&sd, inode, inode->i_size);
1889         } else {
1890                 inode2sd(&sd, inode, inode->i_size);
1891         }
1892         // these do not go to on-disk stat data
1893         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1894         inode->i_blksize = reiserfs_default_io_size;
1895
1896         // store in in-core inode the key of stat data and version all
1897         // object items will have (directory items will have old offset
1898         // format, other new objects will consist of new items)
1899         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1900         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1901                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1902         else
1903                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1904         if (old_format_only(sb))
1905                 set_inode_sd_version(inode, STAT_DATA_V1);
1906         else
1907                 set_inode_sd_version(inode, STAT_DATA_V2);
1908
1909         /* insert the stat data into the tree */
1910 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1911         if (REISERFS_I(dir)->new_packing_locality)
1912                 th->displace_new_blocks = 1;
1913 #endif
1914         retval =
1915             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1916                                  (char *)(&sd));
1917         if (retval) {
1918                 err = retval;
1919                 reiserfs_check_path(&path_to_key);
1920                 goto out_bad_inode;
1921         }
1922 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1923         if (!th->displace_new_blocks)
1924                 REISERFS_I(dir)->new_packing_locality = 0;
1925 #endif
1926         if (S_ISDIR(mode)) {
1927                 /* insert item with "." and ".." */
1928                 retval =
1929                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1930         }
1931
1932         if (S_ISLNK(mode)) {
1933                 /* insert body of symlink */
1934                 if (!old_format_only(sb))
1935                         i_size = ROUND_UP(i_size);
1936                 retval =
1937                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1938                                          i_size);
1939         }
1940         if (retval) {
1941                 err = retval;
1942                 reiserfs_check_path(&path_to_key);
1943                 journal_end(th, th->t_super, th->t_blocks_allocated);
1944                 goto out_inserted_sd;
1945         }
1946
1947         /* XXX CHECK THIS */
1948         if (reiserfs_posixacl(inode->i_sb)) {
1949                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1950                 if (retval) {
1951                         err = retval;
1952                         reiserfs_check_path(&path_to_key);
1953                         journal_end(th, th->t_super, th->t_blocks_allocated);
1954                         goto out_inserted_sd;
1955                 }
1956         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1957                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1958                                  "but vfs thinks they are!");
1959         } else if (is_reiserfs_priv_object(dir)) {
1960                 reiserfs_mark_inode_private(inode);
1961         }
1962
1963         insert_inode_hash(inode);
1964         reiserfs_update_sd(th, inode);
1965         reiserfs_check_path(&path_to_key);
1966
1967         return 0;
1968
1969 /* it looks like you can easily compress these two goto targets into
1970  * one.  Keeping it like this doesn't actually hurt anything, and they
1971  * are place holders for what the quota code actually needs.
1972  */
1973       out_bad_inode:
1974         /* Invalidate the object, nothing was inserted yet */
1975         INODE_PKEY(inode)->k_objectid = 0;
1976
1977         /* Quota change must be inside a transaction for journaling */
1978         DQUOT_FREE_INODE(inode);
1979
1980       out_end_trans:
1981         journal_end(th, th->t_super, th->t_blocks_allocated);
1982         /* Drop can be outside and it needs more credits so it's better to have it outside */
1983         DQUOT_DROP(inode);
1984         inode->i_flags |= S_NOQUOTA;
1985         make_bad_inode(inode);
1986
1987       out_inserted_sd:
1988         inode->i_nlink = 0;
1989         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1990
1991         /* If we were inheriting an ACL, we need to release the lock so that
1992          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1993          * code really needs to be reworked, but this will take care of it
1994          * for now. -jeffm */
1995         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1996                 reiserfs_write_unlock_xattrs(dir->i_sb);
1997                 iput(inode);
1998                 reiserfs_write_lock_xattrs(dir->i_sb);
1999         } else
2000                 iput(inode);
2001         return err;
2002 }
2003
2004 /*
2005 ** finds the tail page in the page cache,
2006 ** reads the last block in.
2007 **
2008 ** On success, page_result is set to a locked, pinned page, and bh_result
2009 ** is set to an up to date buffer for the last block in the file.  returns 0.
2010 **
2011 ** tail conversion is not done, so bh_result might not be valid for writing
2012 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2013 ** trying to write the block.
2014 **
2015 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2016 */
2017 static int grab_tail_page(struct inode *p_s_inode,
2018                           struct page **page_result,
2019                           struct buffer_head **bh_result)
2020 {
2021
2022         /* we want the page with the last byte in the file,
2023          ** not the page that will hold the next byte for appending
2024          */
2025         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2026         unsigned long pos = 0;
2027         unsigned long start = 0;
2028         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2029         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2030         struct buffer_head *bh;
2031         struct buffer_head *head;
2032         struct page *page;
2033         int error;
2034
2035         /* we know that we are only called with inode->i_size > 0.
2036          ** we also know that a file tail can never be as big as a block
2037          ** If i_size % blocksize == 0, our file is currently block aligned
2038          ** and it won't need converting or zeroing after a truncate.
2039          */
2040         if ((offset & (blocksize - 1)) == 0) {
2041                 return -ENOENT;
2042         }
2043         page = grab_cache_page(p_s_inode->i_mapping, index);
2044         error = -ENOMEM;
2045         if (!page) {
2046                 goto out;
2047         }
2048         /* start within the page of the last block in the file */
2049         start = (offset / blocksize) * blocksize;
2050
2051         error = block_prepare_write(page, start, offset,
2052                                     reiserfs_get_block_create_0);
2053         if (error)
2054                 goto unlock;
2055
2056         head = page_buffers(page);
2057         bh = head;
2058         do {
2059                 if (pos >= start) {
2060                         break;
2061                 }
2062                 bh = bh->b_this_page;
2063                 pos += blocksize;
2064         } while (bh != head);
2065
2066         if (!buffer_uptodate(bh)) {
2067                 /* note, this should never happen, prepare_write should
2068                  ** be taking care of this for us.  If the buffer isn't up to date,
2069                  ** I've screwed up the code to find the buffer, or the code to
2070                  ** call prepare_write
2071                  */
2072                 reiserfs_warning(p_s_inode->i_sb,
2073                                  "clm-6000: error reading block %lu on dev %s",
2074                                  bh->b_blocknr,
2075                                  reiserfs_bdevname(p_s_inode->i_sb));
2076                 error = -EIO;
2077                 goto unlock;
2078         }
2079         *bh_result = bh;
2080         *page_result = page;
2081
2082       out:
2083         return error;
2084
2085       unlock:
2086         unlock_page(page);
2087         page_cache_release(page);
2088         return error;
2089 }
2090
2091 /*
2092 ** vfs version of truncate file.  Must NOT be called with
2093 ** a transaction already started.
2094 **
2095 ** some code taken from block_truncate_page
2096 */
2097 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2098 {
2099         struct reiserfs_transaction_handle th;
2100         /* we want the offset for the first byte after the end of the file */
2101         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2102         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2103         unsigned length;
2104         struct page *page = NULL;
2105         int error;
2106         struct buffer_head *bh = NULL;
2107         int err2;
2108
2109         reiserfs_write_lock(p_s_inode->i_sb);
2110
2111         if (p_s_inode->i_size > 0) {
2112                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2113                         // -ENOENT means we truncated past the end of the file, 
2114                         // and get_block_create_0 could not find a block to read in,
2115                         // which is ok.
2116                         if (error != -ENOENT)
2117                                 reiserfs_warning(p_s_inode->i_sb,
2118                                                  "clm-6001: grab_tail_page failed %d",
2119                                                  error);
2120                         page = NULL;
2121                         bh = NULL;
2122                 }
2123         }
2124
2125         /* so, if page != NULL, we have a buffer head for the offset at 
2126          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0, 
2127          ** then we have an unformatted node.  Otherwise, we have a direct item, 
2128          ** and no zeroing is required on disk.  We zero after the truncate, 
2129          ** because the truncate might pack the item anyway 
2130          ** (it will unmap bh if it packs).
2131          */
2132         /* it is enough to reserve space in transaction for 2 balancings:
2133            one for "save" link adding and another for the first
2134            cut_from_item. 1 is for update_sd */
2135         error = journal_begin(&th, p_s_inode->i_sb,
2136                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2137         if (error)
2138                 goto out;
2139         reiserfs_update_inode_transaction(p_s_inode);
2140         if (update_timestamps)
2141                 /* we are doing real truncate: if the system crashes before the last
2142                    transaction of truncating gets committed - on reboot the file
2143                    either appears truncated properly or not truncated at all */
2144                 add_save_link(&th, p_s_inode, 1);
2145         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2146         error =
2147             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2148         if (error)
2149                 goto out;
2150
2151         /* check reiserfs_do_truncate after ending the transaction */
2152         if (err2) {
2153                 error = err2;
2154                 goto out;
2155         }
2156         
2157         if (update_timestamps) {
2158                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2159                 if (error)
2160                         goto out;
2161         }
2162
2163         if (page) {
2164                 length = offset & (blocksize - 1);
2165                 /* if we are not on a block boundary */
2166                 if (length) {
2167                         char *kaddr;
2168
2169                         length = blocksize - length;
2170                         kaddr = kmap_atomic(page, KM_USER0);
2171                         memset(kaddr + offset, 0, length);
2172                         flush_dcache_page(page);
2173                         kunmap_atomic(kaddr, KM_USER0);
2174                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2175                                 mark_buffer_dirty(bh);
2176                         }
2177                 }
2178                 unlock_page(page);
2179                 page_cache_release(page);
2180         }
2181
2182         reiserfs_write_unlock(p_s_inode->i_sb);
2183         return 0;
2184       out:
2185         if (page) {
2186                 unlock_page(page);
2187                 page_cache_release(page);
2188         }
2189         reiserfs_write_unlock(p_s_inode->i_sb);
2190         return error;
2191 }
2192
2193 static int map_block_for_writepage(struct inode *inode,
2194                                    struct buffer_head *bh_result,
2195                                    unsigned long block)
2196 {
2197         struct reiserfs_transaction_handle th;
2198         int fs_gen;
2199         struct item_head tmp_ih;
2200         struct item_head *ih;
2201         struct buffer_head *bh;
2202         __le32 *item;
2203         struct cpu_key key;
2204         INITIALIZE_PATH(path);
2205         int pos_in_item;
2206         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2207         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2208         int retval;
2209         int use_get_block = 0;
2210         int bytes_copied = 0;
2211         int copy_size;
2212         int trans_running = 0;
2213
2214         /* catch places below that try to log something without starting a trans */
2215         th.t_trans_id = 0;
2216
2217         if (!buffer_uptodate(bh_result)) {
2218                 return -EIO;
2219         }
2220
2221         kmap(bh_result->b_page);
2222       start_over:
2223         reiserfs_write_lock(inode->i_sb);
2224         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2225
2226       research:
2227         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2228         if (retval != POSITION_FOUND) {
2229                 use_get_block = 1;
2230                 goto out;
2231         }
2232
2233         bh = get_last_bh(&path);
2234         ih = get_ih(&path);
2235         item = get_item(&path);
2236         pos_in_item = path.pos_in_item;
2237
2238         /* we've found an unformatted node */
2239         if (indirect_item_found(retval, ih)) {
2240                 if (bytes_copied > 0) {
2241                         reiserfs_warning(inode->i_sb,
2242                                          "clm-6002: bytes_copied %d",
2243                                          bytes_copied);
2244                 }
2245                 if (!get_block_num(item, pos_in_item)) {
2246                         /* crap, we are writing to a hole */
2247                         use_get_block = 1;
2248                         goto out;
2249                 }
2250                 set_block_dev_mapped(bh_result,
2251                                      get_block_num(item, pos_in_item), inode);
2252         } else if (is_direct_le_ih(ih)) {
2253                 char *p;
2254                 p = page_address(bh_result->b_page);
2255                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2256                 copy_size = ih_item_len(ih) - pos_in_item;
2257
2258                 fs_gen = get_generation(inode->i_sb);
2259                 copy_item_head(&tmp_ih, ih);
2260
2261                 if (!trans_running) {
2262                         /* vs-3050 is gone, no need to drop the path */
2263                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2264                         if (retval)
2265                                 goto out;
2266                         reiserfs_update_inode_transaction(inode);
2267                         trans_running = 1;
2268                         if (fs_changed(fs_gen, inode->i_sb)
2269                             && item_moved(&tmp_ih, &path)) {
2270                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2271                                                                  bh);
2272                                 goto research;
2273                         }
2274                 }
2275
2276                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2277
2278                 if (fs_changed(fs_gen, inode->i_sb)
2279                     && item_moved(&tmp_ih, &path)) {
2280                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2281                         goto research;
2282                 }
2283
2284                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2285                        copy_size);
2286
2287                 journal_mark_dirty(&th, inode->i_sb, bh);
2288                 bytes_copied += copy_size;
2289                 set_block_dev_mapped(bh_result, 0, inode);
2290
2291                 /* are there still bytes left? */
2292                 if (bytes_copied < bh_result->b_size &&
2293                     (byte_offset + bytes_copied) < inode->i_size) {
2294                         set_cpu_key_k_offset(&key,
2295                                              cpu_key_k_offset(&key) +
2296                                              copy_size);
2297                         goto research;
2298                 }
2299         } else {
2300                 reiserfs_warning(inode->i_sb,
2301                                  "clm-6003: bad item inode %lu, device %s",
2302                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2303                 retval = -EIO;
2304                 goto out;
2305         }
2306         retval = 0;
2307
2308       out:
2309         pathrelse(&path);
2310         if (trans_running) {
2311                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2312                 if (err)
2313                         retval = err;
2314                 trans_running = 0;
2315         }
2316         reiserfs_write_unlock(inode->i_sb);
2317
2318         /* this is where we fill in holes in the file. */
2319         if (use_get_block) {
2320                 retval = reiserfs_get_block(inode, block, bh_result,
2321                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2322                                             | GET_BLOCK_NO_DANGLE);
2323                 if (!retval) {
2324                         if (!buffer_mapped(bh_result)
2325                             || bh_result->b_blocknr == 0) {
2326                                 /* get_block failed to find a mapped unformatted node. */
2327                                 use_get_block = 0;
2328                                 goto start_over;
2329                         }
2330                 }
2331         }
2332         kunmap(bh_result->b_page);
2333
2334         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2335                 /* we've copied data from the page into the direct item, so the
2336                  * buffer in the page is now clean, mark it to reflect that.
2337                  */
2338                 lock_buffer(bh_result);
2339                 clear_buffer_dirty(bh_result);
2340                 unlock_buffer(bh_result);
2341         }
2342         return retval;
2343 }
2344
2345 /* 
2346  * mason@suse.com: updated in 2.5.54 to follow the same general io 
2347  * start/recovery path as __block_write_full_page, along with special
2348  * code to handle reiserfs tails.
2349  */
2350 static int reiserfs_write_full_page(struct page *page,
2351                                     struct writeback_control *wbc)
2352 {
2353         struct inode *inode = page->mapping->host;
2354         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2355         int error = 0;
2356         unsigned long block;
2357         struct buffer_head *head, *bh;
2358         int partial = 0;
2359         int nr = 0;
2360         int checked = PageChecked(page);
2361         struct reiserfs_transaction_handle th;
2362         struct super_block *s = inode->i_sb;
2363         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2364         th.t_trans_id = 0;
2365
2366         /* The page dirty bit is cleared before writepage is called, which
2367          * means we have to tell create_empty_buffers to make dirty buffers
2368          * The page really should be up to date at this point, so tossing
2369          * in the BH_Uptodate is just a sanity check.
2370          */
2371         if (!page_has_buffers(page)) {
2372                 create_empty_buffers(page, s->s_blocksize,
2373                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2374         }
2375         head = page_buffers(page);
2376
2377         /* last page in the file, zero out any contents past the
2378          ** last byte in the file
2379          */
2380         if (page->index >= end_index) {
2381                 char *kaddr;
2382                 unsigned last_offset;
2383
2384                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2385                 /* no file contents in this page */
2386                 if (page->index >= end_index + 1 || !last_offset) {
2387                         unlock_page(page);
2388                         return 0;
2389                 }
2390                 kaddr = kmap_atomic(page, KM_USER0);
2391                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2392                 flush_dcache_page(page);
2393                 kunmap_atomic(kaddr, KM_USER0);
2394         }
2395         bh = head;
2396         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2397         /* first map all the buffers, logging any direct items we find */
2398         do {
2399                 if ((checked || buffer_dirty(bh)) && (!buffer_mapped(bh) ||
2400                                                       (buffer_mapped(bh)
2401                                                        && bh->b_blocknr ==
2402                                                        0))) {
2403                         /* not mapped yet, or it points to a direct item, search
2404                          * the btree for the mapping info, and log any direct
2405                          * items found
2406                          */
2407                         if ((error = map_block_for_writepage(inode, bh, block))) {
2408                                 goto fail;
2409                         }
2410                 }
2411                 bh = bh->b_this_page;
2412                 block++;
2413         } while (bh != head);
2414
2415         /*
2416          * we start the transaction after map_block_for_writepage,
2417          * because it can create holes in the file (an unbounded operation).
2418          * starting it here, we can make a reliable estimate for how many
2419          * blocks we're going to log
2420          */
2421         if (checked) {
2422                 ClearPageChecked(page);
2423                 reiserfs_write_lock(s);
2424                 error = journal_begin(&th, s, bh_per_page + 1);
2425                 if (error) {
2426                         reiserfs_write_unlock(s);
2427                         goto fail;
2428                 }
2429                 reiserfs_update_inode_transaction(inode);
2430         }
2431         /* now go through and lock any dirty buffers on the page */
2432         do {
2433                 get_bh(bh);
2434                 if (!buffer_mapped(bh))
2435                         continue;
2436                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2437                         continue;
2438
2439                 if (checked) {
2440                         reiserfs_prepare_for_journal(s, bh, 1);
2441                         journal_mark_dirty(&th, s, bh);
2442                         continue;
2443                 }
2444                 /* from this point on, we know the buffer is mapped to a
2445                  * real block and not a direct item
2446                  */
2447                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2448                         lock_buffer(bh);
2449                 } else {
2450                         if (test_set_buffer_locked(bh)) {
2451                                 redirty_page_for_writepage(wbc, page);
2452                                 continue;
2453                         }
2454                 }
2455                 if (test_clear_buffer_dirty(bh)) {
2456                         mark_buffer_async_write(bh);
2457                 } else {
2458                         unlock_buffer(bh);
2459                 }
2460         } while ((bh = bh->b_this_page) != head);
2461
2462         if (checked) {
2463                 error = journal_end(&th, s, bh_per_page + 1);
2464                 reiserfs_write_unlock(s);
2465                 if (error)
2466                         goto fail;
2467         }
2468         BUG_ON(PageWriteback(page));
2469         set_page_writeback(page);
2470         unlock_page(page);
2471
2472         /*
2473          * since any buffer might be the only dirty buffer on the page, 
2474          * the first submit_bh can bring the page out of writeback.
2475          * be careful with the buffers.
2476          */
2477         do {
2478                 struct buffer_head *next = bh->b_this_page;
2479                 if (buffer_async_write(bh)) {
2480                         submit_bh(WRITE, bh);
2481                         nr++;
2482                 }
2483                 put_bh(bh);
2484                 bh = next;
2485         } while (bh != head);
2486
2487         error = 0;
2488       done:
2489         if (nr == 0) {
2490                 /*
2491                  * if this page only had a direct item, it is very possible for
2492                  * no io to be required without there being an error.  Or, 
2493                  * someone else could have locked them and sent them down the 
2494                  * pipe without locking the page
2495                  */
2496                 bh = head;
2497                 do {
2498                         if (!buffer_uptodate(bh)) {
2499                                 partial = 1;
2500                                 break;
2501                         }
2502                         bh = bh->b_this_page;
2503                 } while (bh != head);
2504                 if (!partial)
2505                         SetPageUptodate(page);
2506                 end_page_writeback(page);
2507         }
2508         return error;
2509
2510       fail:
2511         /* catches various errors, we need to make sure any valid dirty blocks
2512          * get to the media.  The page is currently locked and not marked for 
2513          * writeback
2514          */
2515         ClearPageUptodate(page);
2516         bh = head;
2517         do {
2518                 get_bh(bh);
2519                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2520                         lock_buffer(bh);
2521                         mark_buffer_async_write(bh);
2522                 } else {
2523                         /*
2524                          * clear any dirty bits that might have come from getting
2525                          * attached to a dirty page
2526                          */
2527                         clear_buffer_dirty(bh);
2528                 }
2529                 bh = bh->b_this_page;
2530         } while (bh != head);
2531         SetPageError(page);
2532         BUG_ON(PageWriteback(page));
2533         set_page_writeback(page);
2534         unlock_page(page);
2535         do {
2536                 struct buffer_head *next = bh->b_this_page;
2537                 if (buffer_async_write(bh)) {
2538                         clear_buffer_dirty(bh);
2539                         submit_bh(WRITE, bh);
2540                         nr++;
2541                 }
2542                 put_bh(bh);
2543                 bh = next;
2544         } while (bh != head);
2545         goto done;
2546 }
2547
2548 static int reiserfs_readpage(struct file *f, struct page *page)
2549 {
2550         return block_read_full_page(page, reiserfs_get_block);
2551 }
2552
2553 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2554 {
2555         struct inode *inode = page->mapping->host;
2556         reiserfs_wait_on_write_block(inode->i_sb);
2557         return reiserfs_write_full_page(page, wbc);
2558 }
2559
2560 static int reiserfs_prepare_write(struct file *f, struct page *page,
2561                                   unsigned from, unsigned to)
2562 {
2563         struct inode *inode = page->mapping->host;
2564         int ret;
2565         int old_ref = 0;
2566
2567         reiserfs_wait_on_write_block(inode->i_sb);
2568         fix_tail_page_for_writing(page);
2569         if (reiserfs_transaction_running(inode->i_sb)) {
2570                 struct reiserfs_transaction_handle *th;
2571                 th = (struct reiserfs_transaction_handle *)current->
2572                     journal_info;
2573                 BUG_ON(!th->t_refcount);
2574                 BUG_ON(!th->t_trans_id);
2575                 old_ref = th->t_refcount;
2576                 th->t_refcount++;
2577         }
2578
2579         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2580         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2581                 struct reiserfs_transaction_handle *th = current->journal_info;
2582                 /* this gets a little ugly.  If reiserfs_get_block returned an
2583                  * error and left a transacstion running, we've got to close it,
2584                  * and we've got to free handle if it was a persistent transaction.
2585                  *
2586                  * But, if we had nested into an existing transaction, we need
2587                  * to just drop the ref count on the handle.
2588                  *
2589                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2590                  * and it was a persistent trans.  Otherwise, it was nested above.
2591                  */
2592                 if (th->t_refcount > old_ref) {
2593                         if (old_ref)
2594                                 th->t_refcount--;
2595                         else {
2596                                 int err;
2597                                 reiserfs_write_lock(inode->i_sb);
2598                                 err = reiserfs_end_persistent_transaction(th);
2599                                 reiserfs_write_unlock(inode->i_sb);
2600                                 if (err)
2601                                         ret = err;
2602                         }
2603                 }
2604         }
2605         return ret;
2606
2607 }
2608
2609 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2610 {
2611         return generic_block_bmap(as, block, reiserfs_bmap);
2612 }
2613
2614 static int reiserfs_commit_write(struct file *f, struct page *page,
2615                                  unsigned from, unsigned to)
2616 {
2617         struct inode *inode = page->mapping->host;
2618         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2619         int ret = 0;
2620         int update_sd = 0;
2621         struct reiserfs_transaction_handle *th = NULL;
2622
2623         reiserfs_wait_on_write_block(inode->i_sb);
2624         if (reiserfs_transaction_running(inode->i_sb)) {
2625                 th = current->journal_info;
2626         }
2627         reiserfs_commit_page(inode, page, from, to);
2628
2629         /* generic_commit_write does this for us, but does not update the
2630          ** transaction tracking stuff when the size changes.  So, we have
2631          ** to do the i_size updates here.
2632          */
2633         if (pos > inode->i_size) {
2634                 struct reiserfs_transaction_handle myth;
2635                 reiserfs_write_lock(inode->i_sb);
2636                 /* If the file have grown beyond the border where it
2637                    can have a tail, unmark it as needing a tail
2638                    packing */
2639                 if ((have_large_tails(inode->i_sb)
2640                      && inode->i_size > i_block_size(inode) * 4)
2641                     || (have_small_tails(inode->i_sb)
2642                         && inode->i_size > i_block_size(inode)))
2643                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2644
2645                 ret = journal_begin(&myth, inode->i_sb, 1);
2646                 if (ret) {
2647                         reiserfs_write_unlock(inode->i_sb);
2648                         goto journal_error;
2649                 }
2650                 reiserfs_update_inode_transaction(inode);
2651                 inode->i_size = pos;
2652                 /*
2653                  * this will just nest into our transaction.  It's important
2654                  * to use mark_inode_dirty so the inode gets pushed around on the
2655                  * dirty lists, and so that O_SYNC works as expected
2656                  */
2657                 mark_inode_dirty(inode);
2658                 reiserfs_update_sd(&myth, inode);
2659                 update_sd = 1;
2660                 ret = journal_end(&myth, inode->i_sb, 1);
2661                 reiserfs_write_unlock(inode->i_sb);
2662                 if (ret)
2663                         goto journal_error;
2664         }
2665         if (th) {
2666                 reiserfs_write_lock(inode->i_sb);
2667                 if (!update_sd)
2668                         mark_inode_dirty(inode);
2669                 ret = reiserfs_end_persistent_transaction(th);
2670                 reiserfs_write_unlock(inode->i_sb);
2671                 if (ret)
2672                         goto out;
2673         }
2674
2675       out:
2676         return ret;
2677
2678       journal_error:
2679         if (th) {
2680                 reiserfs_write_lock(inode->i_sb);
2681                 if (!update_sd)
2682                         reiserfs_update_sd(th, inode);
2683                 ret = reiserfs_end_persistent_transaction(th);
2684                 reiserfs_write_unlock(inode->i_sb);
2685         }
2686
2687         return ret;
2688 }
2689
2690 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2691 {
2692         if (reiserfs_attrs(inode->i_sb)) {
2693                 if (sd_attrs & REISERFS_SYNC_FL)
2694                         inode->i_flags |= S_SYNC;
2695                 else
2696                         inode->i_flags &= ~S_SYNC;
2697                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2698                         inode->i_flags |= S_IMMUTABLE;
2699                 else
2700                         inode->i_flags &= ~S_IMMUTABLE;
2701                 if (sd_attrs & REISERFS_APPEND_FL)
2702                         inode->i_flags |= S_APPEND;
2703                 else
2704                         inode->i_flags &= ~S_APPEND;
2705                 if (sd_attrs & REISERFS_NOATIME_FL)
2706                         inode->i_flags |= S_NOATIME;
2707                 else
2708                         inode->i_flags &= ~S_NOATIME;
2709                 if (sd_attrs & REISERFS_NOTAIL_FL)
2710                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2711                 else
2712                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2713         }
2714 }
2715
2716 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2717 {
2718         if (reiserfs_attrs(inode->i_sb)) {
2719                 if (inode->i_flags & S_IMMUTABLE)
2720                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2721                 else
2722                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2723                 if (inode->i_flags & S_SYNC)
2724                         *sd_attrs |= REISERFS_SYNC_FL;
2725                 else
2726                         *sd_attrs &= ~REISERFS_SYNC_FL;
2727                 if (inode->i_flags & S_NOATIME)
2728                         *sd_attrs |= REISERFS_NOATIME_FL;
2729                 else
2730                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2731                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2732                         *sd_attrs |= REISERFS_NOTAIL_FL;
2733                 else
2734                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2735         }
2736 }
2737
2738 /* decide if this buffer needs to stay around for data logging or ordered
2739 ** write purposes
2740 */
2741 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2742 {
2743         int ret = 1;
2744         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2745
2746         lock_buffer(bh);
2747         spin_lock(&j->j_dirty_buffers_lock);
2748         if (!buffer_mapped(bh)) {
2749                 goto free_jh;
2750         }
2751         /* the page is locked, and the only places that log a data buffer
2752          * also lock the page.
2753          */
2754         if (reiserfs_file_data_log(inode)) {
2755                 /*
2756                  * very conservative, leave the buffer pinned if
2757                  * anyone might need it.
2758                  */
2759                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2760                         ret = 0;
2761                 }
2762         } else  if (buffer_dirty(bh)) {
2763                 struct reiserfs_journal_list *jl;
2764                 struct reiserfs_jh *jh = bh->b_private;
2765
2766                 /* why is this safe?
2767                  * reiserfs_setattr updates i_size in the on disk
2768                  * stat data before allowing vmtruncate to be called.
2769                  *
2770                  * If buffer was put onto the ordered list for this
2771                  * transaction, we know for sure either this transaction
2772                  * or an older one already has updated i_size on disk,
2773                  * and this ordered data won't be referenced in the file
2774                  * if we crash.
2775                  *
2776                  * if the buffer was put onto the ordered list for an older
2777                  * transaction, we need to leave it around
2778                  */
2779                 if (jh && (jl = jh->jl)
2780                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2781                         ret = 0;
2782         }
2783       free_jh:
2784         if (ret && bh->b_private) {
2785                 reiserfs_free_jh(bh);
2786         }
2787         spin_unlock(&j->j_dirty_buffers_lock);
2788         unlock_buffer(bh);
2789         return ret;
2790 }
2791
2792 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2793 static int reiserfs_invalidatepage(struct page *page, unsigned long offset)
2794 {
2795         struct buffer_head *head, *bh, *next;
2796         struct inode *inode = page->mapping->host;
2797         unsigned int curr_off = 0;
2798         int ret = 1;
2799
2800         BUG_ON(!PageLocked(page));
2801
2802         if (offset == 0)
2803                 ClearPageChecked(page);
2804
2805         if (!page_has_buffers(page))
2806                 goto out;
2807
2808         head = page_buffers(page);
2809         bh = head;
2810         do {
2811                 unsigned int next_off = curr_off + bh->b_size;
2812                 next = bh->b_this_page;
2813
2814                 /*
2815                  * is this block fully invalidated?
2816                  */
2817                 if (offset <= curr_off) {
2818                         if (invalidatepage_can_drop(inode, bh))
2819                                 reiserfs_unmap_buffer(bh);
2820                         else
2821                                 ret = 0;
2822                 }
2823                 curr_off = next_off;
2824                 bh = next;
2825         } while (bh != head);
2826
2827         /*
2828          * We release buffers only if the entire page is being invalidated.
2829          * The get_block cached value has been unconditionally invalidated,
2830          * so real IO is not possible anymore.
2831          */
2832         if (!offset && ret)
2833                 ret = try_to_release_page(page, 0);
2834       out:
2835         return ret;
2836 }
2837
2838 static int reiserfs_set_page_dirty(struct page *page)
2839 {
2840         struct inode *inode = page->mapping->host;
2841         if (reiserfs_file_data_log(inode)) {
2842                 SetPageChecked(page);
2843                 return __set_page_dirty_nobuffers(page);
2844         }
2845         return __set_page_dirty_buffers(page);
2846 }
2847
2848 /*
2849  * Returns 1 if the page's buffers were dropped.  The page is locked.
2850  *
2851  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2852  * in the buffers at page_buffers(page).
2853  *
2854  * even in -o notail mode, we can't be sure an old mount without -o notail
2855  * didn't create files with tails.
2856  */
2857 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2858 {
2859         struct inode *inode = page->mapping->host;
2860         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2861         struct buffer_head *head;
2862         struct buffer_head *bh;
2863         int ret = 1;
2864
2865         WARN_ON(PageChecked(page));
2866         spin_lock(&j->j_dirty_buffers_lock);
2867         head = page_buffers(page);
2868         bh = head;
2869         do {
2870                 if (bh->b_private) {
2871                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2872                                 reiserfs_free_jh(bh);
2873                         } else {
2874                                 ret = 0;
2875                                 break;
2876                         }
2877                 }
2878                 bh = bh->b_this_page;
2879         } while (bh != head);
2880         if (ret)
2881                 ret = try_to_free_buffers(page);
2882         spin_unlock(&j->j_dirty_buffers_lock);
2883         return ret;
2884 }
2885
2886 /* We thank Mingming Cao for helping us understand in great detail what
2887    to do in this section of the code. */
2888 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2889                                   const struct iovec *iov, loff_t offset,
2890                                   unsigned long nr_segs)
2891 {
2892         struct file *file = iocb->ki_filp;
2893         struct inode *inode = file->f_mapping->host;
2894
2895         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2896                                   offset, nr_segs,
2897                                   reiserfs_get_blocks_direct_io, NULL);
2898 }
2899
2900 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2901 {
2902         struct inode *inode = dentry->d_inode;
2903         int error;
2904         unsigned int ia_valid = attr->ia_valid;
2905         reiserfs_write_lock(inode->i_sb);
2906         if (attr->ia_valid & ATTR_SIZE) {
2907                 /* version 2 items will be caught by the s_maxbytes check
2908                  ** done for us in vmtruncate
2909                  */
2910                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2911                     attr->ia_size > MAX_NON_LFS) {
2912                         error = -EFBIG;
2913                         goto out;
2914                 }
2915                 /* fill in hole pointers in the expanding truncate case. */
2916                 if (attr->ia_size > inode->i_size) {
2917                         error = generic_cont_expand(inode, attr->ia_size);
2918                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2919                                 int err;
2920                                 struct reiserfs_transaction_handle th;
2921                                 /* we're changing at most 2 bitmaps, inode + super */
2922                                 err = journal_begin(&th, inode->i_sb, 4);
2923                                 if (!err) {
2924                                         reiserfs_discard_prealloc(&th, inode);
2925                                         err = journal_end(&th, inode->i_sb, 4);
2926                                 }
2927                                 if (err)
2928                                         error = err;
2929                         }
2930                         if (error)
2931                                 goto out;
2932                 }
2933         }
2934
2935         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2936              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2937             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2938                 /* stat data of format v3.5 has 16 bit uid and gid */
2939                 error = -EINVAL;
2940                 goto out;
2941         }
2942
2943         error = inode_change_ok(inode, attr);
2944         if (!error) {
2945                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2946                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2947                         error = reiserfs_chown_xattrs(inode, attr);
2948
2949                         if (!error) {
2950                                 struct reiserfs_transaction_handle th;
2951                                 int jbegin_count =
2952                                     2 *
2953                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2954                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2955                                     2;
2956
2957                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2958                                 error =
2959                                     journal_begin(&th, inode->i_sb,
2960                                                   jbegin_count);
2961                                 if (error)
2962                                         goto out;
2963                                 error =
2964                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2965                                 if (error) {
2966                                         journal_end(&th, inode->i_sb,
2967                                                     jbegin_count);
2968                                         goto out;
2969                                 }
2970                                 /* Update corresponding info in inode so that everything is in
2971                                  * one transaction */
2972                                 if (attr->ia_valid & ATTR_UID)
2973                                         inode->i_uid = attr->ia_uid;
2974                                 if (attr->ia_valid & ATTR_GID)
2975                                         inode->i_gid = attr->ia_gid;
2976                                 mark_inode_dirty(inode);
2977                                 error =
2978                                     journal_end(&th, inode->i_sb, jbegin_count);
2979                         }
2980                 }
2981                 if (!error)
2982                         error = inode_setattr(inode, attr);
2983         }
2984
2985         if (!error && reiserfs_posixacl(inode->i_sb)) {
2986                 if (attr->ia_valid & ATTR_MODE)
2987                         error = reiserfs_acl_chmod(inode);
2988         }
2989
2990       out:
2991         reiserfs_write_unlock(inode->i_sb);
2992         return error;
2993 }
2994
2995 struct address_space_operations reiserfs_address_space_operations = {
2996         .writepage = reiserfs_writepage,
2997         .readpage = reiserfs_readpage,
2998         .readpages = reiserfs_readpages,
2999         .releasepage = reiserfs_releasepage,
3000         .invalidatepage = reiserfs_invalidatepage,
3001         .sync_page = block_sync_page,
3002         .prepare_write = reiserfs_prepare_write,
3003         .commit_write = reiserfs_commit_write,
3004         .bmap = reiserfs_aop_bmap,
3005         .direct_IO = reiserfs_direct_IO,
3006         .set_page_dirty = reiserfs_set_page_dirty,
3007 };