1a8a1bf2154d10475a0bacc89b085070587d48d6
[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
36         truncate_inode_pages(&inode->i_data, 0);
37
38         reiserfs_write_lock(inode->i_sb);
39
40         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
41         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
42                 down(&inode->i_sem);
43
44                 reiserfs_delete_xattrs(inode);
45
46                 if (journal_begin(&th, inode->i_sb, jbegin_count)) {
47                         up(&inode->i_sem);
48                         goto out;
49                 }
50                 reiserfs_update_inode_transaction(inode);
51
52                 if (reiserfs_delete_object(&th, inode)) {
53                         up(&inode->i_sem);
54                         goto out;
55                 }
56
57                 /* Do quota update inside a transaction for journaled quotas. We must do that
58                  * after delete_object so that quota updates go into the same transaction as
59                  * stat data deletion */
60                 DQUOT_FREE_INODE(inode);
61
62                 if (journal_end(&th, inode->i_sb, jbegin_count)) {
63                         up(&inode->i_sem);
64                         goto out;
65                 }
66
67                 up(&inode->i_sem);
68
69                 /* all items of file are deleted, so we can remove "save" link */
70                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
71                                                                  * about an error here */
72         } else {
73                 /* no object items are in the tree */
74                 ;
75         }
76       out:
77         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
78         inode->i_blocks = 0;
79         reiserfs_write_unlock(inode->i_sb);
80 }
81
82 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
83                           __u32 objectid, loff_t offset, int type, int length)
84 {
85         key->version = version;
86
87         key->on_disk_key.k_dir_id = dirid;
88         key->on_disk_key.k_objectid = objectid;
89         set_cpu_key_k_offset(key, offset);
90         set_cpu_key_k_type(key, type);
91         key->key_length = length;
92 }
93
94 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
95    offset and type of key */
96 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
97                   int type, int length)
98 {
99         _make_cpu_key(key, get_inode_item_key_version(inode),
100                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
101                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
102                       length);
103 }
104
105 //
106 // when key is 0, do not set version and short key
107 //
108 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
109                               int version,
110                               loff_t offset, int type, int length,
111                               int entry_count /*or ih_free_space */ )
112 {
113         if (key) {
114                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
115                 ih->ih_key.k_objectid =
116                     cpu_to_le32(key->on_disk_key.k_objectid);
117         }
118         put_ih_version(ih, version);
119         set_le_ih_k_offset(ih, offset);
120         set_le_ih_k_type(ih, type);
121         put_ih_item_len(ih, length);
122         /*    set_ih_free_space (ih, 0); */
123         // for directory items it is entry count, for directs and stat
124         // datas - 0xffff, for indirects - 0
125         put_ih_entry_count(ih, entry_count);
126 }
127
128 //
129 // FIXME: we might cache recently accessed indirect item
130
131 // Ugh.  Not too eager for that....
132 //  I cut the code until such time as I see a convincing argument (benchmark).
133 // I don't want a bloated inode struct..., and I don't like code complexity....
134
135 /* cutting the code is fine, since it really isn't in use yet and is easy
136 ** to add back in.  But, Vladimir has a really good idea here.  Think
137 ** about what happens for reading a file.  For each page,
138 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
139 ** an indirect item.  This indirect item has X number of pointers, where
140 ** X is a big number if we've done the block allocation right.  But,
141 ** we only use one or two of these pointers during each call to readpage,
142 ** needlessly researching again later on.
143 **
144 ** The size of the cache could be dynamic based on the size of the file.
145 **
146 ** I'd also like to see us cache the location the stat data item, since
147 ** we are needlessly researching for that frequently.
148 **
149 ** --chris
150 */
151
152 /* If this page has a file tail in it, and
153 ** it was read in by get_block_create_0, the page data is valid,
154 ** but tail is still sitting in a direct item, and we can't write to
155 ** it.  So, look through this page, and check all the mapped buffers
156 ** to make sure they have valid block numbers.  Any that don't need
157 ** to be unmapped, so that block_prepare_write will correctly call
158 ** reiserfs_get_block to convert the tail into an unformatted node
159 */
160 static inline void fix_tail_page_for_writing(struct page *page)
161 {
162         struct buffer_head *head, *next, *bh;
163
164         if (page && page_has_buffers(page)) {
165                 head = page_buffers(page);
166                 bh = head;
167                 do {
168                         next = bh->b_this_page;
169                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
170                                 reiserfs_unmap_buffer(bh);
171                         }
172                         bh = next;
173                 } while (bh != head);
174         }
175 }
176
177 /* reiserfs_get_block does not need to allocate a block only if it has been
178    done already or non-hole position has been found in the indirect item */
179 static inline int allocation_needed(int retval, b_blocknr_t allocated,
180                                     struct item_head *ih,
181                                     __le32 * item, int pos_in_item)
182 {
183         if (allocated)
184                 return 0;
185         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
186             get_block_num(item, pos_in_item))
187                 return 0;
188         return 1;
189 }
190
191 static inline int indirect_item_found(int retval, struct item_head *ih)
192 {
193         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
194 }
195
196 static inline void set_block_dev_mapped(struct buffer_head *bh,
197                                         b_blocknr_t block, struct inode *inode)
198 {
199         map_bh(bh, inode->i_sb, block);
200 }
201
202 //
203 // files which were created in the earlier version can not be longer,
204 // than 2 gb
205 //
206 static int file_capable(struct inode *inode, long block)
207 {
208         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
209             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
210                 return 1;
211
212         return 0;
213 }
214
215 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
216                                    struct inode *inode, struct path *path)
217 {
218         struct super_block *s = th->t_super;
219         int len = th->t_blocks_allocated;
220         int err;
221
222         BUG_ON(!th->t_trans_id);
223         BUG_ON(!th->t_refcount);
224
225         /* we cannot restart while nested */
226         if (th->t_refcount > 1) {
227                 return 0;
228         }
229         pathrelse(path);
230         reiserfs_update_sd(th, inode);
231         err = journal_end(th, s, len);
232         if (!err) {
233                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
234                 if (!err)
235                         reiserfs_update_inode_transaction(inode);
236         }
237         return err;
238 }
239
240 // it is called by get_block when create == 0. Returns block number
241 // for 'block'-th logical block of file. When it hits direct item it
242 // returns 0 (being called from bmap) or read direct item into piece
243 // of page (bh_result)
244
245 // Please improve the english/clarity in the comment above, as it is
246 // hard to understand.
247
248 static int _get_block_create_0(struct inode *inode, long block,
249                                struct buffer_head *bh_result, int args)
250 {
251         INITIALIZE_PATH(path);
252         struct cpu_key key;
253         struct buffer_head *bh;
254         struct item_head *ih, tmp_ih;
255         int fs_gen;
256         int blocknr;
257         char *p = NULL;
258         int chars;
259         int ret;
260         int result;
261         int done = 0;
262         unsigned long offset;
263
264         // prepare the key to look for the 'block'-th block of file
265         make_cpu_key(&key, inode,
266                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
267                      3);
268
269       research:
270         result = search_for_position_by_key(inode->i_sb, &key, &path);
271         if (result != POSITION_FOUND) {
272                 pathrelse(&path);
273                 if (p)
274                         kunmap(bh_result->b_page);
275                 if (result == IO_ERROR)
276                         return -EIO;
277                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
278                 // That there is some MMAPED data associated with it that is yet to be written to disk.
279                 if ((args & GET_BLOCK_NO_HOLE)
280                     && !PageUptodate(bh_result->b_page)) {
281                         return -ENOENT;
282                 }
283                 return 0;
284         }
285         //
286         bh = get_last_bh(&path);
287         ih = get_ih(&path);
288         if (is_indirect_le_ih(ih)) {
289                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
290
291                 /* FIXME: here we could cache indirect item or part of it in
292                    the inode to avoid search_by_key in case of subsequent
293                    access to file */
294                 blocknr = get_block_num(ind_item, path.pos_in_item);
295                 ret = 0;
296                 if (blocknr) {
297                         map_bh(bh_result, inode->i_sb, blocknr);
298                         if (path.pos_in_item ==
299                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
300                                 set_buffer_boundary(bh_result);
301                         }
302                 } else
303                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
304                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
305                 if ((args & GET_BLOCK_NO_HOLE)
306                             && !PageUptodate(bh_result->b_page)) {
307                         ret = -ENOENT;
308                 }
309
310                 pathrelse(&path);
311                 if (p)
312                         kunmap(bh_result->b_page);
313                 return ret;
314         }
315         // requested data are in direct item(s)
316         if (!(args & GET_BLOCK_READ_DIRECT)) {
317                 // we are called by bmap. FIXME: we can not map block of file
318                 // when it is stored in direct item(s)
319                 pathrelse(&path);
320                 if (p)
321                         kunmap(bh_result->b_page);
322                 return -ENOENT;
323         }
324
325         /* if we've got a direct item, and the buffer or page was uptodate,
326          ** we don't want to pull data off disk again.  skip to the
327          ** end, where we map the buffer and return
328          */
329         if (buffer_uptodate(bh_result)) {
330                 goto finished;
331         } else
332                 /*
333                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
334                  ** pages without any buffers.  If the page is up to date, we don't want
335                  ** read old data off disk.  Set the up to date bit on the buffer instead
336                  ** and jump to the end
337                  */
338         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
339                 set_buffer_uptodate(bh_result);
340                 goto finished;
341         }
342         // read file tail into part of page
343         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
344         fs_gen = get_generation(inode->i_sb);
345         copy_item_head(&tmp_ih, ih);
346
347         /* we only want to kmap if we are reading the tail into the page.
348          ** this is not the common case, so we don't kmap until we are
349          ** sure we need to.  But, this means the item might move if
350          ** kmap schedules
351          */
352         if (!p) {
353                 p = (char *)kmap(bh_result->b_page);
354                 if (fs_changed(fs_gen, inode->i_sb)
355                     && item_moved(&tmp_ih, &path)) {
356                         goto research;
357                 }
358         }
359         p += offset;
360         memset(p, 0, inode->i_sb->s_blocksize);
361         do {
362                 if (!is_direct_le_ih(ih)) {
363                         BUG();
364                 }
365                 /* make sure we don't read more bytes than actually exist in
366                  ** the file.  This can happen in odd cases where i_size isn't
367                  ** correct, and when direct item padding results in a few 
368                  ** extra bytes at the end of the direct item
369                  */
370                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
371                         break;
372                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
373                         chars =
374                             inode->i_size - (le_ih_k_offset(ih) - 1) -
375                             path.pos_in_item;
376                         done = 1;
377                 } else {
378                         chars = ih_item_len(ih) - path.pos_in_item;
379                 }
380                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
381
382                 if (done)
383                         break;
384
385                 p += chars;
386
387                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
388                         // we done, if read direct item is not the last item of
389                         // node FIXME: we could try to check right delimiting key
390                         // to see whether direct item continues in the right
391                         // neighbor or rely on i_size
392                         break;
393
394                 // update key to look for the next piece
395                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
396                 result = search_for_position_by_key(inode->i_sb, &key, &path);
397                 if (result != POSITION_FOUND)
398                         // i/o error most likely
399                         break;
400                 bh = get_last_bh(&path);
401                 ih = get_ih(&path);
402         } while (1);
403
404         flush_dcache_page(bh_result->b_page);
405         kunmap(bh_result->b_page);
406
407       finished:
408         pathrelse(&path);
409
410         if (result == IO_ERROR)
411                 return -EIO;
412
413         /* this buffer has valid data, but isn't valid for io.  mapping it to
414          * block #0 tells the rest of reiserfs it just has a tail in it
415          */
416         map_bh(bh_result, inode->i_sb, 0);
417         set_buffer_uptodate(bh_result);
418         return 0;
419 }
420
421 // this is called to create file map. So, _get_block_create_0 will not
422 // read direct item
423 static int reiserfs_bmap(struct inode *inode, sector_t block,
424                          struct buffer_head *bh_result, int create)
425 {
426         if (!file_capable(inode, block))
427                 return -EFBIG;
428
429         reiserfs_write_lock(inode->i_sb);
430         /* do not read the direct item */
431         _get_block_create_0(inode, block, bh_result, 0);
432         reiserfs_write_unlock(inode->i_sb);
433         return 0;
434 }
435
436 /* special version of get_block that is only used by grab_tail_page right
437 ** now.  It is sent to block_prepare_write, and when you try to get a
438 ** block past the end of the file (or a block from a hole) it returns
439 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
440 ** be able to do i/o on the buffers returned, unless an error value
441 ** is also returned.
442 ** 
443 ** So, this allows block_prepare_write to be used for reading a single block
444 ** in a page.  Where it does not produce a valid page for holes, or past the
445 ** end of the file.  This turns out to be exactly what we need for reading
446 ** tails for conversion.
447 **
448 ** The point of the wrapper is forcing a certain value for create, even
449 ** though the VFS layer is calling this function with create==1.  If you 
450 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block, 
451 ** don't use this function.
452 */
453 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
454                                        struct buffer_head *bh_result,
455                                        int create)
456 {
457         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
458 }
459
460 /* This is special helper for reiserfs_get_block in case we are executing
461    direct_IO request. */
462 static int reiserfs_get_blocks_direct_io(struct inode *inode,
463                                          sector_t iblock,
464                                          unsigned long max_blocks,
465                                          struct buffer_head *bh_result,
466                                          int create)
467 {
468         int ret;
469
470         bh_result->b_page = NULL;
471
472         /* We set the b_size before reiserfs_get_block call since it is
473            referenced in convert_tail_for_hole() that may be called from
474            reiserfs_get_block() */
475         bh_result->b_size = (1 << inode->i_blkbits);
476
477         ret = reiserfs_get_block(inode, iblock, bh_result,
478                                  create | GET_BLOCK_NO_DANGLE);
479         if (ret)
480                 goto out;
481
482         /* don't allow direct io onto tail pages */
483         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
484                 /* make sure future calls to the direct io funcs for this offset
485                  ** in the file fail by unmapping the buffer
486                  */
487                 clear_buffer_mapped(bh_result);
488                 ret = -EINVAL;
489         }
490         /* Possible unpacked tail. Flush the data before pages have
491            disappeared */
492         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
493                 int err;
494                 lock_kernel();
495                 err = reiserfs_commit_for_inode(inode);
496                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
497                 unlock_kernel();
498                 if (err < 0)
499                         ret = err;
500         }
501       out:
502         return ret;
503 }
504
505 /*
506 ** helper function for when reiserfs_get_block is called for a hole
507 ** but the file tail is still in a direct item
508 ** bh_result is the buffer head for the hole
509 ** tail_offset is the offset of the start of the tail in the file
510 **
511 ** This calls prepare_write, which will start a new transaction
512 ** you should not be in a transaction, or have any paths held when you
513 ** call this.
514 */
515 static int convert_tail_for_hole(struct inode *inode,
516                                  struct buffer_head *bh_result,
517                                  loff_t tail_offset)
518 {
519         unsigned long index;
520         unsigned long tail_end;
521         unsigned long tail_start;
522         struct page *tail_page;
523         struct page *hole_page = bh_result->b_page;
524         int retval = 0;
525
526         if ((tail_offset & (bh_result->b_size - 1)) != 1)
527                 return -EIO;
528
529         /* always try to read until the end of the block */
530         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
531         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
532
533         index = tail_offset >> PAGE_CACHE_SHIFT;
534         /* hole_page can be zero in case of direct_io, we are sure
535            that we cannot get here if we write with O_DIRECT into
536            tail page */
537         if (!hole_page || index != hole_page->index) {
538                 tail_page = grab_cache_page(inode->i_mapping, index);
539                 retval = -ENOMEM;
540                 if (!tail_page) {
541                         goto out;
542                 }
543         } else {
544                 tail_page = hole_page;
545         }
546
547         /* we don't have to make sure the conversion did not happen while
548          ** we were locking the page because anyone that could convert
549          ** must first take i_sem.
550          **
551          ** We must fix the tail page for writing because it might have buffers
552          ** that are mapped, but have a block number of 0.  This indicates tail
553          ** data that has been read directly into the page, and block_prepare_write
554          ** won't trigger a get_block in this case.
555          */
556         fix_tail_page_for_writing(tail_page);
557         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
558         if (retval)
559                 goto unlock;
560
561         /* tail conversion might change the data in the page */
562         flush_dcache_page(tail_page);
563
564         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
565
566       unlock:
567         if (tail_page != hole_page) {
568                 unlock_page(tail_page);
569                 page_cache_release(tail_page);
570         }
571       out:
572         return retval;
573 }
574
575 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
576                                   long block,
577                                   struct inode *inode,
578                                   b_blocknr_t * allocated_block_nr,
579                                   struct path *path, int flags)
580 {
581         BUG_ON(!th->t_trans_id);
582
583 #ifdef REISERFS_PREALLOCATE
584         if (!(flags & GET_BLOCK_NO_ISEM)) {
585                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
586                                                   path, block);
587         }
588 #endif
589         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
590                                          block);
591 }
592
593 int reiserfs_get_block(struct inode *inode, sector_t block,
594                        struct buffer_head *bh_result, int create)
595 {
596         int repeat, retval = 0;
597         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
598         INITIALIZE_PATH(path);
599         int pos_in_item;
600         struct cpu_key key;
601         struct buffer_head *bh, *unbh = NULL;
602         struct item_head *ih, tmp_ih;
603         __le32 *item;
604         int done;
605         int fs_gen;
606         struct reiserfs_transaction_handle *th = NULL;
607         /* space reserved in transaction batch: 
608            . 3 balancings in direct->indirect conversion
609            . 1 block involved into reiserfs_update_sd()
610            XXX in practically impossible worst case direct2indirect()
611            can incur (much) more than 3 balancings.
612            quota update for user, group */
613         int jbegin_count =
614             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
615             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
616         int version;
617         int dangle = 1;
618         loff_t new_offset =
619             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
620
621         /* bad.... */
622         reiserfs_write_lock(inode->i_sb);
623         version = get_inode_item_key_version(inode);
624
625         if (block < 0) {
626                 reiserfs_write_unlock(inode->i_sb);
627                 return -EIO;
628         }
629
630         if (!file_capable(inode, block)) {
631                 reiserfs_write_unlock(inode->i_sb);
632                 return -EFBIG;
633         }
634
635         /* if !create, we aren't changing the FS, so we don't need to
636          ** log anything, so we don't need to start a transaction
637          */
638         if (!(create & GET_BLOCK_CREATE)) {
639                 int ret;
640                 /* find number of block-th logical block of the file */
641                 ret = _get_block_create_0(inode, block, bh_result,
642                                           create | GET_BLOCK_READ_DIRECT);
643                 reiserfs_write_unlock(inode->i_sb);
644                 return ret;
645         }
646         /*
647          * if we're already in a transaction, make sure to close
648          * any new transactions we start in this func
649          */
650         if ((create & GET_BLOCK_NO_DANGLE) ||
651             reiserfs_transaction_running(inode->i_sb))
652                 dangle = 0;
653
654         /* If file is of such a size, that it might have a tail and tails are enabled
655          ** we should mark it as possibly needing tail packing on close
656          */
657         if ((have_large_tails(inode->i_sb)
658              && inode->i_size < i_block_size(inode) * 4)
659             || (have_small_tails(inode->i_sb)
660                 && inode->i_size < i_block_size(inode)))
661                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
662
663         /* set the key of the first byte in the 'block'-th block of file */
664         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
665         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
666               start_trans:
667                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
668                 if (!th) {
669                         retval = -ENOMEM;
670                         goto failure;
671                 }
672                 reiserfs_update_inode_transaction(inode);
673         }
674       research:
675
676         retval = search_for_position_by_key(inode->i_sb, &key, &path);
677         if (retval == IO_ERROR) {
678                 retval = -EIO;
679                 goto failure;
680         }
681
682         bh = get_last_bh(&path);
683         ih = get_ih(&path);
684         item = get_item(&path);
685         pos_in_item = path.pos_in_item;
686
687         fs_gen = get_generation(inode->i_sb);
688         copy_item_head(&tmp_ih, ih);
689
690         if (allocation_needed
691             (retval, allocated_block_nr, ih, item, pos_in_item)) {
692                 /* we have to allocate block for the unformatted node */
693                 if (!th) {
694                         pathrelse(&path);
695                         goto start_trans;
696                 }
697
698                 repeat =
699                     _allocate_block(th, block, inode, &allocated_block_nr,
700                                     &path, create);
701
702                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
703                         /* restart the transaction to give the journal a chance to free
704                          ** some blocks.  releases the path, so we have to go back to
705                          ** research if we succeed on the second try
706                          */
707                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
708                         retval = restart_transaction(th, inode, &path);
709                         if (retval)
710                                 goto failure;
711                         repeat =
712                             _allocate_block(th, block, inode,
713                                             &allocated_block_nr, NULL, create);
714
715                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
716                                 goto research;
717                         }
718                         if (repeat == QUOTA_EXCEEDED)
719                                 retval = -EDQUOT;
720                         else
721                                 retval = -ENOSPC;
722                         goto failure;
723                 }
724
725                 if (fs_changed(fs_gen, inode->i_sb)
726                     && item_moved(&tmp_ih, &path)) {
727                         goto research;
728                 }
729         }
730
731         if (indirect_item_found(retval, ih)) {
732                 b_blocknr_t unfm_ptr;
733                 /* 'block'-th block is in the file already (there is
734                    corresponding cell in some indirect item). But it may be
735                    zero unformatted node pointer (hole) */
736                 unfm_ptr = get_block_num(item, pos_in_item);
737                 if (unfm_ptr == 0) {
738                         /* use allocated block to plug the hole */
739                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
740                         if (fs_changed(fs_gen, inode->i_sb)
741                             && item_moved(&tmp_ih, &path)) {
742                                 reiserfs_restore_prepared_buffer(inode->i_sb,
743                                                                  bh);
744                                 goto research;
745                         }
746                         set_buffer_new(bh_result);
747                         if (buffer_dirty(bh_result)
748                             && reiserfs_data_ordered(inode->i_sb))
749                                 reiserfs_add_ordered_list(inode, bh_result);
750                         put_block_num(item, pos_in_item, allocated_block_nr);
751                         unfm_ptr = allocated_block_nr;
752                         journal_mark_dirty(th, inode->i_sb, bh);
753                         reiserfs_update_sd(th, inode);
754                 }
755                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
756                 pathrelse(&path);
757                 retval = 0;
758                 if (!dangle && th)
759                         retval = reiserfs_end_persistent_transaction(th);
760
761                 reiserfs_write_unlock(inode->i_sb);
762
763                 /* the item was found, so new blocks were not added to the file
764                  ** there is no need to make sure the inode is updated with this 
765                  ** transaction
766                  */
767                 return retval;
768         }
769
770         if (!th) {
771                 pathrelse(&path);
772                 goto start_trans;
773         }
774
775         /* desired position is not found or is in the direct item. We have
776            to append file with holes up to 'block'-th block converting
777            direct items to indirect one if necessary */
778         done = 0;
779         do {
780                 if (is_statdata_le_ih(ih)) {
781                         __le32 unp = 0;
782                         struct cpu_key tmp_key;
783
784                         /* indirect item has to be inserted */
785                         make_le_item_head(&tmp_ih, &key, version, 1,
786                                           TYPE_INDIRECT, UNFM_P_SIZE,
787                                           0 /* free_space */ );
788
789                         if (cpu_key_k_offset(&key) == 1) {
790                                 /* we are going to add 'block'-th block to the file. Use
791                                    allocated block for that */
792                                 unp = cpu_to_le32(allocated_block_nr);
793                                 set_block_dev_mapped(bh_result,
794                                                      allocated_block_nr, inode);
795                                 set_buffer_new(bh_result);
796                                 done = 1;
797                         }
798                         tmp_key = key;  // ;)
799                         set_cpu_key_k_offset(&tmp_key, 1);
800                         PATH_LAST_POSITION(&path)++;
801
802                         retval =
803                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
804                                                  inode, (char *)&unp);
805                         if (retval) {
806                                 reiserfs_free_block(th, inode,
807                                                     allocated_block_nr, 1);
808                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
809                         }
810                         //mark_tail_converted (inode);
811                 } else if (is_direct_le_ih(ih)) {
812                         /* direct item has to be converted */
813                         loff_t tail_offset;
814
815                         tail_offset =
816                             ((le_ih_k_offset(ih) -
817                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
818                         if (tail_offset == cpu_key_k_offset(&key)) {
819                                 /* direct item we just found fits into block we have
820                                    to map. Convert it into unformatted node: use
821                                    bh_result for the conversion */
822                                 set_block_dev_mapped(bh_result,
823                                                      allocated_block_nr, inode);
824                                 unbh = bh_result;
825                                 done = 1;
826                         } else {
827                                 /* we have to padd file tail stored in direct item(s)
828                                    up to block size and convert it to unformatted
829                                    node. FIXME: this should also get into page cache */
830
831                                 pathrelse(&path);
832                                 /*
833                                  * ugly, but we can only end the transaction if
834                                  * we aren't nested
835                                  */
836                                 BUG_ON(!th->t_refcount);
837                                 if (th->t_refcount == 1) {
838                                         retval =
839                                             reiserfs_end_persistent_transaction
840                                             (th);
841                                         th = NULL;
842                                         if (retval)
843                                                 goto failure;
844                                 }
845
846                                 retval =
847                                     convert_tail_for_hole(inode, bh_result,
848                                                           tail_offset);
849                                 if (retval) {
850                                         if (retval != -ENOSPC)
851                                                 reiserfs_warning(inode->i_sb,
852                                                                  "clm-6004: convert tail failed inode %lu, error %d",
853                                                                  inode->i_ino,
854                                                                  retval);
855                                         if (allocated_block_nr) {
856                                                 /* the bitmap, the super, and the stat data == 3 */
857                                                 if (!th)
858                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
859                                                 if (th)
860                                                         reiserfs_free_block(th,
861                                                                             inode,
862                                                                             allocated_block_nr,
863                                                                             1);
864                                         }
865                                         goto failure;
866                                 }
867                                 goto research;
868                         }
869                         retval =
870                             direct2indirect(th, inode, &path, unbh,
871                                             tail_offset);
872                         if (retval) {
873                                 reiserfs_unmap_buffer(unbh);
874                                 reiserfs_free_block(th, inode,
875                                                     allocated_block_nr, 1);
876                                 goto failure;
877                         }
878                         /* it is important the set_buffer_uptodate is done after
879                          ** the direct2indirect.  The buffer might contain valid
880                          ** data newer than the data on disk (read by readpage, changed,
881                          ** and then sent here by writepage).  direct2indirect needs
882                          ** to know if unbh was already up to date, so it can decide
883                          ** if the data in unbh needs to be replaced with data from
884                          ** the disk
885                          */
886                         set_buffer_uptodate(unbh);
887
888                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
889                            buffer will disappear shortly, so it should not be added to
890                          */
891                         if (unbh->b_page) {
892                                 /* we've converted the tail, so we must
893                                  ** flush unbh before the transaction commits
894                                  */
895                                 reiserfs_add_tail_list(inode, unbh);
896
897                                 /* mark it dirty now to prevent commit_write from adding
898                                  ** this buffer to the inode's dirty buffer list
899                                  */
900                                 /*
901                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
902                                  * It's still atomic, but it sets the page dirty too,
903                                  * which makes it eligible for writeback at any time by the
904                                  * VM (which was also the case with __mark_buffer_dirty())
905                                  */
906                                 mark_buffer_dirty(unbh);
907                         }
908                 } else {
909                         /* append indirect item with holes if needed, when appending
910                            pointer to 'block'-th block use block, which is already
911                            allocated */
912                         struct cpu_key tmp_key;
913                         unp_t unf_single = 0;   // We use this in case we need to allocate only
914                         // one block which is a fastpath
915                         unp_t *un;
916                         __u64 max_to_insert =
917                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
918                             UNFM_P_SIZE;
919                         __u64 blocks_needed;
920
921                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
922                                "vs-804: invalid position for append");
923                         /* indirect item has to be appended, set up key of that position */
924                         make_cpu_key(&tmp_key, inode,
925                                      le_key_k_offset(version,
926                                                      &(ih->ih_key)) +
927                                      op_bytes_number(ih,
928                                                      inode->i_sb->s_blocksize),
929                                      //pos_in_item * inode->i_sb->s_blocksize,
930                                      TYPE_INDIRECT, 3); // key type is unimportant
931
932                         blocks_needed =
933                             1 +
934                             ((cpu_key_k_offset(&key) -
935                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
936                              s_blocksize_bits);
937                         RFALSE(blocks_needed < 0, "green-805: invalid offset");
938
939                         if (blocks_needed == 1) {
940                                 un = &unf_single;
941                         } else {
942                                 un = kmalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
943                                 if (!un) {
944                                         un = &unf_single;
945                                         blocks_needed = 1;
946                                         max_to_insert = 0;
947                                 } else
948                                         memset(un, 0,
949                                                UNFM_P_SIZE * min(blocks_needed,
950                                                                  max_to_insert));
951                         }
952                         if (blocks_needed <= max_to_insert) {
953                                 /* we are going to add target block to the file. Use allocated
954                                    block for that */
955                                 un[blocks_needed - 1] =
956                                     cpu_to_le32(allocated_block_nr);
957                                 set_block_dev_mapped(bh_result,
958                                                      allocated_block_nr, inode);
959                                 set_buffer_new(bh_result);
960                                 done = 1;
961                         } else {
962                                 /* paste hole to the indirect item */
963                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
964                                    only have space for one block */
965                                 blocks_needed =
966                                     max_to_insert ? max_to_insert : 1;
967                         }
968                         retval =
969                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
970                                                      (char *)un,
971                                                      UNFM_P_SIZE *
972                                                      blocks_needed);
973
974                         if (blocks_needed != 1)
975                                 kfree(un);
976
977                         if (retval) {
978                                 reiserfs_free_block(th, inode,
979                                                     allocated_block_nr, 1);
980                                 goto failure;
981                         }
982                         if (!done) {
983                                 /* We need to mark new file size in case this function will be
984                                    interrupted/aborted later on. And we may do this only for
985                                    holes. */
986                                 inode->i_size +=
987                                     inode->i_sb->s_blocksize * blocks_needed;
988                         }
989                 }
990
991                 if (done == 1)
992                         break;
993
994                 /* this loop could log more blocks than we had originally asked
995                  ** for.  So, we have to allow the transaction to end if it is
996                  ** too big or too full.  Update the inode so things are 
997                  ** consistent if we crash before the function returns
998                  **
999                  ** release the path so that anybody waiting on the path before
1000                  ** ending their transaction will be able to continue.
1001                  */
1002                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1003                         retval = restart_transaction(th, inode, &path);
1004                         if (retval)
1005                                 goto failure;
1006                 }
1007                 /* inserting indirect pointers for a hole can take a 
1008                  ** long time.  reschedule if needed
1009                  */
1010                 cond_resched();
1011
1012                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1013                 if (retval == IO_ERROR) {
1014                         retval = -EIO;
1015                         goto failure;
1016                 }
1017                 if (retval == POSITION_FOUND) {
1018                         reiserfs_warning(inode->i_sb,
1019                                          "vs-825: reiserfs_get_block: "
1020                                          "%K should not be found", &key);
1021                         retval = -EEXIST;
1022                         if (allocated_block_nr)
1023                                 reiserfs_free_block(th, inode,
1024                                                     allocated_block_nr, 1);
1025                         pathrelse(&path);
1026                         goto failure;
1027                 }
1028                 bh = get_last_bh(&path);
1029                 ih = get_ih(&path);
1030                 item = get_item(&path);
1031                 pos_in_item = path.pos_in_item;
1032         } while (1);
1033
1034         retval = 0;
1035
1036       failure:
1037         if (th && (!dangle || (retval && !th->t_trans_id))) {
1038                 int err;
1039                 if (th->t_trans_id)
1040                         reiserfs_update_sd(th, inode);
1041                 err = reiserfs_end_persistent_transaction(th);
1042                 if (err)
1043                         retval = err;
1044         }
1045
1046         reiserfs_write_unlock(inode->i_sb);
1047         reiserfs_check_path(&path);
1048         return retval;
1049 }
1050
1051 static int
1052 reiserfs_readpages(struct file *file, struct address_space *mapping,
1053                    struct list_head *pages, unsigned nr_pages)
1054 {
1055         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1056 }
1057
1058 /* Compute real number of used bytes by file
1059  * Following three functions can go away when we'll have enough space in stat item
1060  */
1061 static int real_space_diff(struct inode *inode, int sd_size)
1062 {
1063         int bytes;
1064         loff_t blocksize = inode->i_sb->s_blocksize;
1065
1066         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1067                 return sd_size;
1068
1069         /* End of file is also in full block with indirect reference, so round
1070          ** up to the next block.
1071          **
1072          ** there is just no way to know if the tail is actually packed
1073          ** on the file, so we have to assume it isn't.  When we pack the
1074          ** tail, we add 4 bytes to pretend there really is an unformatted
1075          ** node pointer
1076          */
1077         bytes =
1078             ((inode->i_size +
1079               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1080             sd_size;
1081         return bytes;
1082 }
1083
1084 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1085                                         int sd_size)
1086 {
1087         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1088                 return inode->i_size +
1089                     (loff_t) (real_space_diff(inode, sd_size));
1090         }
1091         return ((loff_t) real_space_diff(inode, sd_size)) +
1092             (((loff_t) blocks) << 9);
1093 }
1094
1095 /* Compute number of blocks used by file in ReiserFS counting */
1096 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1097 {
1098         loff_t bytes = inode_get_bytes(inode);
1099         loff_t real_space = real_space_diff(inode, sd_size);
1100
1101         /* keeps fsck and non-quota versions of reiserfs happy */
1102         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1103                 bytes += (loff_t) 511;
1104         }
1105
1106         /* files from before the quota patch might i_blocks such that
1107          ** bytes < real_space.  Deal with that here to prevent it from
1108          ** going negative.
1109          */
1110         if (bytes < real_space)
1111                 return 0;
1112         return (bytes - real_space) >> 9;
1113 }
1114
1115 //
1116 // BAD: new directories have stat data of new type and all other items
1117 // of old type. Version stored in the inode says about body items, so
1118 // in update_stat_data we can not rely on inode, but have to check
1119 // item version directly
1120 //
1121
1122 // called by read_locked_inode
1123 static void init_inode(struct inode *inode, struct path *path)
1124 {
1125         struct buffer_head *bh;
1126         struct item_head *ih;
1127         __u32 rdev;
1128         //int version = ITEM_VERSION_1;
1129
1130         bh = PATH_PLAST_BUFFER(path);
1131         ih = PATH_PITEM_HEAD(path);
1132
1133         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1134         inode->i_blksize = reiserfs_default_io_size;
1135
1136         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1137         REISERFS_I(inode)->i_flags = 0;
1138         REISERFS_I(inode)->i_prealloc_block = 0;
1139         REISERFS_I(inode)->i_prealloc_count = 0;
1140         REISERFS_I(inode)->i_trans_id = 0;
1141         REISERFS_I(inode)->i_jl = NULL;
1142         REISERFS_I(inode)->i_acl_access = NULL;
1143         REISERFS_I(inode)->i_acl_default = NULL;
1144         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1145
1146         if (stat_data_v1(ih)) {
1147                 struct stat_data_v1 *sd =
1148                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1149                 unsigned long blocks;
1150
1151                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1152                 set_inode_sd_version(inode, STAT_DATA_V1);
1153                 inode->i_mode = sd_v1_mode(sd);
1154                 inode->i_nlink = sd_v1_nlink(sd);
1155                 inode->i_uid = sd_v1_uid(sd);
1156                 inode->i_gid = sd_v1_gid(sd);
1157                 inode->i_size = sd_v1_size(sd);
1158                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1159                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1160                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1161                 inode->i_atime.tv_nsec = 0;
1162                 inode->i_ctime.tv_nsec = 0;
1163                 inode->i_mtime.tv_nsec = 0;
1164
1165                 inode->i_blocks = sd_v1_blocks(sd);
1166                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1167                 blocks = (inode->i_size + 511) >> 9;
1168                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1169                 if (inode->i_blocks > blocks) {
1170                         // there was a bug in <=3.5.23 when i_blocks could take negative
1171                         // values. Starting from 3.5.17 this value could even be stored in
1172                         // stat data. For such files we set i_blocks based on file
1173                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1174                         // only updated if file's inode will ever change
1175                         inode->i_blocks = blocks;
1176                 }
1177
1178                 rdev = sd_v1_rdev(sd);
1179                 REISERFS_I(inode)->i_first_direct_byte =
1180                     sd_v1_first_direct_byte(sd);
1181                 /* an early bug in the quota code can give us an odd number for the
1182                  ** block count.  This is incorrect, fix it here.
1183                  */
1184                 if (inode->i_blocks & 1) {
1185                         inode->i_blocks++;
1186                 }
1187                 inode_set_bytes(inode,
1188                                 to_real_used_space(inode, inode->i_blocks,
1189                                                    SD_V1_SIZE));
1190                 /* nopack is initially zero for v1 objects. For v2 objects,
1191                    nopack is initialised from sd_attrs */
1192                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1193         } else {
1194                 // new stat data found, but object may have old items
1195                 // (directories and symlinks)
1196                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1197
1198                 inode->i_mode = sd_v2_mode(sd);
1199                 inode->i_nlink = sd_v2_nlink(sd);
1200                 inode->i_uid = sd_v2_uid(sd);
1201                 inode->i_size = sd_v2_size(sd);
1202                 inode->i_gid = sd_v2_gid(sd);
1203                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1204                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1205                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1206                 inode->i_ctime.tv_nsec = 0;
1207                 inode->i_mtime.tv_nsec = 0;
1208                 inode->i_atime.tv_nsec = 0;
1209                 inode->i_blocks = sd_v2_blocks(sd);
1210                 rdev = sd_v2_rdev(sd);
1211                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1212                         inode->i_generation =
1213                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1214                 else
1215                         inode->i_generation = sd_v2_generation(sd);
1216
1217                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1218                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1219                 else
1220                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1221                 REISERFS_I(inode)->i_first_direct_byte = 0;
1222                 set_inode_sd_version(inode, STAT_DATA_V2);
1223                 inode_set_bytes(inode,
1224                                 to_real_used_space(inode, inode->i_blocks,
1225                                                    SD_V2_SIZE));
1226                 /* read persistent inode attributes from sd and initalise
1227                    generic inode flags from them */
1228                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1229                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1230         }
1231
1232         pathrelse(path);
1233         if (S_ISREG(inode->i_mode)) {
1234                 inode->i_op = &reiserfs_file_inode_operations;
1235                 inode->i_fop = &reiserfs_file_operations;
1236                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1237         } else if (S_ISDIR(inode->i_mode)) {
1238                 inode->i_op = &reiserfs_dir_inode_operations;
1239                 inode->i_fop = &reiserfs_dir_operations;
1240         } else if (S_ISLNK(inode->i_mode)) {
1241                 inode->i_op = &reiserfs_symlink_inode_operations;
1242                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1243         } else {
1244                 inode->i_blocks = 0;
1245                 inode->i_op = &reiserfs_special_inode_operations;
1246                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1247         }
1248 }
1249
1250 // update new stat data with inode fields
1251 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1252 {
1253         struct stat_data *sd_v2 = (struct stat_data *)sd;
1254         __u16 flags;
1255
1256         set_sd_v2_mode(sd_v2, inode->i_mode);
1257         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1258         set_sd_v2_uid(sd_v2, inode->i_uid);
1259         set_sd_v2_size(sd_v2, size);
1260         set_sd_v2_gid(sd_v2, inode->i_gid);
1261         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1262         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1263         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1264         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1265         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1266                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1267         else
1268                 set_sd_v2_generation(sd_v2, inode->i_generation);
1269         flags = REISERFS_I(inode)->i_attrs;
1270         i_attrs_to_sd_attrs(inode, &flags);
1271         set_sd_v2_attrs(sd_v2, flags);
1272 }
1273
1274 // used to copy inode's fields to old stat data
1275 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1276 {
1277         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1278
1279         set_sd_v1_mode(sd_v1, inode->i_mode);
1280         set_sd_v1_uid(sd_v1, inode->i_uid);
1281         set_sd_v1_gid(sd_v1, inode->i_gid);
1282         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1283         set_sd_v1_size(sd_v1, size);
1284         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1285         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1286         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1287
1288         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1289                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1290         else
1291                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1292
1293         // Sigh. i_first_direct_byte is back
1294         set_sd_v1_first_direct_byte(sd_v1,
1295                                     REISERFS_I(inode)->i_first_direct_byte);
1296 }
1297
1298 /* NOTE, you must prepare the buffer head before sending it here,
1299 ** and then log it after the call
1300 */
1301 static void update_stat_data(struct path *path, struct inode *inode,
1302                              loff_t size)
1303 {
1304         struct buffer_head *bh;
1305         struct item_head *ih;
1306
1307         bh = PATH_PLAST_BUFFER(path);
1308         ih = PATH_PITEM_HEAD(path);
1309
1310         if (!is_statdata_le_ih(ih))
1311                 reiserfs_panic(inode->i_sb,
1312                                "vs-13065: update_stat_data: key %k, found item %h",
1313                                INODE_PKEY(inode), ih);
1314
1315         if (stat_data_v1(ih)) {
1316                 // path points to old stat data
1317                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1318         } else {
1319                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1320         }
1321
1322         return;
1323 }
1324
1325 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1326                              struct inode *inode, loff_t size)
1327 {
1328         struct cpu_key key;
1329         INITIALIZE_PATH(path);
1330         struct buffer_head *bh;
1331         int fs_gen;
1332         struct item_head *ih, tmp_ih;
1333         int retval;
1334
1335         BUG_ON(!th->t_trans_id);
1336
1337         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1338
1339         for (;;) {
1340                 int pos;
1341                 /* look for the object's stat data */
1342                 retval = search_item(inode->i_sb, &key, &path);
1343                 if (retval == IO_ERROR) {
1344                         reiserfs_warning(inode->i_sb,
1345                                          "vs-13050: reiserfs_update_sd: "
1346                                          "i/o failure occurred trying to update %K stat data",
1347                                          &key);
1348                         return;
1349                 }
1350                 if (retval == ITEM_NOT_FOUND) {
1351                         pos = PATH_LAST_POSITION(&path);
1352                         pathrelse(&path);
1353                         if (inode->i_nlink == 0) {
1354                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1355                                 return;
1356                         }
1357                         reiserfs_warning(inode->i_sb,
1358                                          "vs-13060: reiserfs_update_sd: "
1359                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1360                                          INODE_PKEY(inode), inode->i_nlink,
1361                                          pos);
1362                         reiserfs_check_path(&path);
1363                         return;
1364                 }
1365
1366                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1367                  ** FS might change.  We have to detect that, and loop back to the
1368                  ** search if the stat data item has moved
1369                  */
1370                 bh = get_last_bh(&path);
1371                 ih = get_ih(&path);
1372                 copy_item_head(&tmp_ih, ih);
1373                 fs_gen = get_generation(inode->i_sb);
1374                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1375                 if (fs_changed(fs_gen, inode->i_sb)
1376                     && item_moved(&tmp_ih, &path)) {
1377                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1378                         continue;       /* Stat_data item has been moved after scheduling. */
1379                 }
1380                 break;
1381         }
1382         update_stat_data(&path, inode, size);
1383         journal_mark_dirty(th, th->t_super, bh);
1384         pathrelse(&path);
1385         return;
1386 }
1387
1388 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1389 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1390 ** and clear the key in the private portion of the inode, otherwise a
1391 ** corresponding iput might try to delete whatever object the inode last
1392 ** represented.
1393 */
1394 static void reiserfs_make_bad_inode(struct inode *inode)
1395 {
1396         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1397         make_bad_inode(inode);
1398 }
1399
1400 //
1401 // initially this function was derived from minix or ext2's analog and
1402 // evolved as the prototype did
1403 //
1404
1405 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1406 {
1407         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1408         inode->i_ino = args->objectid;
1409         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1410         return 0;
1411 }
1412
1413 /* looks for stat data in the tree, and fills up the fields of in-core
1414    inode stat data fields */
1415 void reiserfs_read_locked_inode(struct inode *inode,
1416                                 struct reiserfs_iget_args *args)
1417 {
1418         INITIALIZE_PATH(path_to_sd);
1419         struct cpu_key key;
1420         unsigned long dirino;
1421         int retval;
1422
1423         dirino = args->dirid;
1424
1425         /* set version 1, version 2 could be used too, because stat data
1426            key is the same in both versions */
1427         key.version = KEY_FORMAT_3_5;
1428         key.on_disk_key.k_dir_id = dirino;
1429         key.on_disk_key.k_objectid = inode->i_ino;
1430         key.on_disk_key.k_offset = 0;
1431         key.on_disk_key.k_type = 0;
1432
1433         /* look for the object's stat data */
1434         retval = search_item(inode->i_sb, &key, &path_to_sd);
1435         if (retval == IO_ERROR) {
1436                 reiserfs_warning(inode->i_sb,
1437                                  "vs-13070: reiserfs_read_locked_inode: "
1438                                  "i/o failure occurred trying to find stat data of %K",
1439                                  &key);
1440                 reiserfs_make_bad_inode(inode);
1441                 return;
1442         }
1443         if (retval != ITEM_FOUND) {
1444                 /* a stale NFS handle can trigger this without it being an error */
1445                 pathrelse(&path_to_sd);
1446                 reiserfs_make_bad_inode(inode);
1447                 inode->i_nlink = 0;
1448                 return;
1449         }
1450
1451         init_inode(inode, &path_to_sd);
1452
1453         /* It is possible that knfsd is trying to access inode of a file
1454            that is being removed from the disk by some other thread. As we
1455            update sd on unlink all that is required is to check for nlink
1456            here. This bug was first found by Sizif when debugging
1457            SquidNG/Butterfly, forgotten, and found again after Philippe
1458            Gramoulle <philippe.gramoulle@mmania.com> reproduced it. 
1459
1460            More logical fix would require changes in fs/inode.c:iput() to
1461            remove inode from hash-table _after_ fs cleaned disk stuff up and
1462            in iget() to return NULL if I_FREEING inode is found in
1463            hash-table. */
1464         /* Currently there is one place where it's ok to meet inode with
1465            nlink==0: processing of open-unlinked and half-truncated files
1466            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1467         if ((inode->i_nlink == 0) &&
1468             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1469                 reiserfs_warning(inode->i_sb,
1470                                  "vs-13075: reiserfs_read_locked_inode: "
1471                                  "dead inode read from disk %K. "
1472                                  "This is likely to be race with knfsd. Ignore",
1473                                  &key);
1474                 reiserfs_make_bad_inode(inode);
1475         }
1476
1477         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1478
1479 }
1480
1481 /**
1482  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1483  *
1484  * @inode:    inode from hash table to check
1485  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1486  *
1487  * This function is called by iget5_locked() to distinguish reiserfs inodes
1488  * having the same inode numbers. Such inodes can only exist due to some
1489  * error condition. One of them should be bad. Inodes with identical
1490  * inode numbers (objectids) are distinguished by parent directory ids.
1491  *
1492  */
1493 int reiserfs_find_actor(struct inode *inode, void *opaque)
1494 {
1495         struct reiserfs_iget_args *args;
1496
1497         args = opaque;
1498         /* args is already in CPU order */
1499         return (inode->i_ino == args->objectid) &&
1500             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1501 }
1502
1503 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1504 {
1505         struct inode *inode;
1506         struct reiserfs_iget_args args;
1507
1508         args.objectid = key->on_disk_key.k_objectid;
1509         args.dirid = key->on_disk_key.k_dir_id;
1510         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1511                              reiserfs_find_actor, reiserfs_init_locked_inode,
1512                              (void *)(&args));
1513         if (!inode)
1514                 return ERR_PTR(-ENOMEM);
1515
1516         if (inode->i_state & I_NEW) {
1517                 reiserfs_read_locked_inode(inode, &args);
1518                 unlock_new_inode(inode);
1519         }
1520
1521         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1522                 /* either due to i/o error or a stale NFS handle */
1523                 iput(inode);
1524                 inode = NULL;
1525         }
1526         return inode;
1527 }
1528
1529 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1530 {
1531         __u32 *data = vobjp;
1532         struct cpu_key key;
1533         struct dentry *result;
1534         struct inode *inode;
1535
1536         key.on_disk_key.k_objectid = data[0];
1537         key.on_disk_key.k_dir_id = data[1];
1538         reiserfs_write_lock(sb);
1539         inode = reiserfs_iget(sb, &key);
1540         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1541             data[2] != inode->i_generation) {
1542                 iput(inode);
1543                 inode = NULL;
1544         }
1545         reiserfs_write_unlock(sb);
1546         if (!inode)
1547                 inode = ERR_PTR(-ESTALE);
1548         if (IS_ERR(inode))
1549                 return ERR_PTR(PTR_ERR(inode));
1550         result = d_alloc_anon(inode);
1551         if (!result) {
1552                 iput(inode);
1553                 return ERR_PTR(-ENOMEM);
1554         }
1555         return result;
1556 }
1557
1558 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1559                                   int len, int fhtype,
1560                                   int (*acceptable) (void *contect,
1561                                                      struct dentry * de),
1562                                   void *context)
1563 {
1564         __u32 obj[3], parent[3];
1565
1566         /* fhtype happens to reflect the number of u32s encoded.
1567          * due to a bug in earlier code, fhtype might indicate there
1568          * are more u32s then actually fitted.
1569          * so if fhtype seems to be more than len, reduce fhtype.
1570          * Valid types are:
1571          *   2 - objectid + dir_id - legacy support
1572          *   3 - objectid + dir_id + generation
1573          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1574          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1575          *   6 - as above plus generation of directory
1576          * 6 does not fit in NFSv2 handles
1577          */
1578         if (fhtype > len) {
1579                 if (fhtype != 6 || len != 5)
1580                         reiserfs_warning(sb,
1581                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1582                                          fhtype, len);
1583                 fhtype = 5;
1584         }
1585
1586         obj[0] = data[0];
1587         obj[1] = data[1];
1588         if (fhtype == 3 || fhtype >= 5)
1589                 obj[2] = data[2];
1590         else
1591                 obj[2] = 0;     /* generation number */
1592
1593         if (fhtype >= 4) {
1594                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1595                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1596                 if (fhtype == 6)
1597                         parent[2] = data[5];
1598                 else
1599                         parent[2] = 0;
1600         }
1601         return sb->s_export_op->find_exported_dentry(sb, obj,
1602                                                      fhtype < 4 ? NULL : parent,
1603                                                      acceptable, context);
1604 }
1605
1606 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1607                        int need_parent)
1608 {
1609         struct inode *inode = dentry->d_inode;
1610         int maxlen = *lenp;
1611
1612         if (maxlen < 3)
1613                 return 255;
1614
1615         data[0] = inode->i_ino;
1616         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1617         data[2] = inode->i_generation;
1618         *lenp = 3;
1619         /* no room for directory info? return what we've stored so far */
1620         if (maxlen < 5 || !need_parent)
1621                 return 3;
1622
1623         spin_lock(&dentry->d_lock);
1624         inode = dentry->d_parent->d_inode;
1625         data[3] = inode->i_ino;
1626         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1627         *lenp = 5;
1628         if (maxlen >= 6) {
1629                 data[5] = inode->i_generation;
1630                 *lenp = 6;
1631         }
1632         spin_unlock(&dentry->d_lock);
1633         return *lenp;
1634 }
1635
1636 /* looks for stat data, then copies fields to it, marks the buffer
1637    containing stat data as dirty */
1638 /* reiserfs inodes are never really dirty, since the dirty inode call
1639 ** always logs them.  This call allows the VFS inode marking routines
1640 ** to properly mark inodes for datasync and such, but only actually
1641 ** does something when called for a synchronous update.
1642 */
1643 int reiserfs_write_inode(struct inode *inode, int do_sync)
1644 {
1645         struct reiserfs_transaction_handle th;
1646         int jbegin_count = 1;
1647
1648         if (inode->i_sb->s_flags & MS_RDONLY)
1649                 return -EROFS;
1650         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1651          ** these cases are just when the system needs ram, not when the 
1652          ** inode needs to reach disk for safety, and they can safely be
1653          ** ignored because the altered inode has already been logged.
1654          */
1655         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1656                 reiserfs_write_lock(inode->i_sb);
1657                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1658                         reiserfs_update_sd(&th, inode);
1659                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1660                 }
1661                 reiserfs_write_unlock(inode->i_sb);
1662         }
1663         return 0;
1664 }
1665
1666 /* stat data of new object is inserted already, this inserts the item
1667    containing "." and ".." entries */
1668 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1669                                   struct inode *inode,
1670                                   struct item_head *ih, struct path *path,
1671                                   struct inode *dir)
1672 {
1673         struct super_block *sb = th->t_super;
1674         char empty_dir[EMPTY_DIR_SIZE];
1675         char *body = empty_dir;
1676         struct cpu_key key;
1677         int retval;
1678
1679         BUG_ON(!th->t_trans_id);
1680
1681         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1682                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1683                       TYPE_DIRENTRY, 3 /*key length */ );
1684
1685         /* compose item head for new item. Directories consist of items of
1686            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1687            is done by reiserfs_new_inode */
1688         if (old_format_only(sb)) {
1689                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1690                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1691
1692                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1693                                        ih->ih_key.k_objectid,
1694                                        INODE_PKEY(dir)->k_dir_id,
1695                                        INODE_PKEY(dir)->k_objectid);
1696         } else {
1697                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1698                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1699
1700                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1701                                     ih->ih_key.k_objectid,
1702                                     INODE_PKEY(dir)->k_dir_id,
1703                                     INODE_PKEY(dir)->k_objectid);
1704         }
1705
1706         /* look for place in the tree for new item */
1707         retval = search_item(sb, &key, path);
1708         if (retval == IO_ERROR) {
1709                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1710                                  "i/o failure occurred creating new directory");
1711                 return -EIO;
1712         }
1713         if (retval == ITEM_FOUND) {
1714                 pathrelse(path);
1715                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1716                                  "object with this key exists (%k)",
1717                                  &(ih->ih_key));
1718                 return -EEXIST;
1719         }
1720
1721         /* insert item, that is empty directory item */
1722         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1723 }
1724
1725 /* stat data of object has been inserted, this inserts the item
1726    containing the body of symlink */
1727 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1728                                 struct item_head *ih,
1729                                 struct path *path, const char *symname,
1730                                 int item_len)
1731 {
1732         struct super_block *sb = th->t_super;
1733         struct cpu_key key;
1734         int retval;
1735
1736         BUG_ON(!th->t_trans_id);
1737
1738         _make_cpu_key(&key, KEY_FORMAT_3_5,
1739                       le32_to_cpu(ih->ih_key.k_dir_id),
1740                       le32_to_cpu(ih->ih_key.k_objectid),
1741                       1, TYPE_DIRECT, 3 /*key length */ );
1742
1743         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1744                           0 /*free_space */ );
1745
1746         /* look for place in the tree for new item */
1747         retval = search_item(sb, &key, path);
1748         if (retval == IO_ERROR) {
1749                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1750                                  "i/o failure occurred creating new symlink");
1751                 return -EIO;
1752         }
1753         if (retval == ITEM_FOUND) {
1754                 pathrelse(path);
1755                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1756                                  "object with this key exists (%k)",
1757                                  &(ih->ih_key));
1758                 return -EEXIST;
1759         }
1760
1761         /* insert item, that is body of symlink */
1762         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1763 }
1764
1765 /* inserts the stat data into the tree, and then calls
1766    reiserfs_new_directory (to insert ".", ".." item if new object is
1767    directory) or reiserfs_new_symlink (to insert symlink body if new
1768    object is symlink) or nothing (if new object is regular file) 
1769
1770    NOTE! uid and gid must already be set in the inode.  If we return
1771    non-zero due to an error, we have to drop the quota previously allocated
1772    for the fresh inode.  This can only be done outside a transaction, so
1773    if we return non-zero, we also end the transaction.  */
1774 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1775                        struct inode *dir, int mode, const char *symname,
1776                        /* 0 for regular, EMTRY_DIR_SIZE for dirs, 
1777                           strlen (symname) for symlinks) */
1778                        loff_t i_size, struct dentry *dentry,
1779                        struct inode *inode)
1780 {
1781         struct super_block *sb;
1782         INITIALIZE_PATH(path_to_key);
1783         struct cpu_key key;
1784         struct item_head ih;
1785         struct stat_data sd;
1786         int retval;
1787         int err;
1788
1789         BUG_ON(!th->t_trans_id);
1790
1791         if (DQUOT_ALLOC_INODE(inode)) {
1792                 err = -EDQUOT;
1793                 goto out_end_trans;
1794         }
1795         if (!dir || !dir->i_nlink) {
1796                 err = -EPERM;
1797                 goto out_bad_inode;
1798         }
1799
1800         sb = dir->i_sb;
1801
1802         /* item head of new item */
1803         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1804         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1805         if (!ih.ih_key.k_objectid) {
1806                 err = -ENOMEM;
1807                 goto out_bad_inode;
1808         }
1809         if (old_format_only(sb))
1810                 /* not a perfect generation count, as object ids can be reused, but 
1811                  ** this is as good as reiserfs can do right now.
1812                  ** note that the private part of inode isn't filled in yet, we have
1813                  ** to use the directory.
1814                  */
1815                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1816         else
1817 #if defined( USE_INODE_GENERATION_COUNTER )
1818                 inode->i_generation =
1819                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1820 #else
1821                 inode->i_generation = ++event;
1822 #endif
1823
1824         /* fill stat data */
1825         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1826
1827         /* uid and gid must already be set by the caller for quota init */
1828
1829         /* symlink cannot be immutable or append only, right? */
1830         if (S_ISLNK(inode->i_mode))
1831                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1832
1833         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1834         inode->i_size = i_size;
1835         inode->i_blocks = 0;
1836         inode->i_bytes = 0;
1837         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1838             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1839
1840         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1841         REISERFS_I(inode)->i_flags = 0;
1842         REISERFS_I(inode)->i_prealloc_block = 0;
1843         REISERFS_I(inode)->i_prealloc_count = 0;
1844         REISERFS_I(inode)->i_trans_id = 0;
1845         REISERFS_I(inode)->i_jl = NULL;
1846         REISERFS_I(inode)->i_attrs =
1847             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1848         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1849         REISERFS_I(inode)->i_acl_access = NULL;
1850         REISERFS_I(inode)->i_acl_default = NULL;
1851         init_rwsem(&REISERFS_I(inode)->xattr_sem);
1852
1853         if (old_format_only(sb))
1854                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1855                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1856         else
1857                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1858                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1859
1860         /* key to search for correct place for new stat data */
1861         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1862                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1863                       TYPE_STAT_DATA, 3 /*key length */ );
1864
1865         /* find proper place for inserting of stat data */
1866         retval = search_item(sb, &key, &path_to_key);
1867         if (retval == IO_ERROR) {
1868                 err = -EIO;
1869                 goto out_bad_inode;
1870         }
1871         if (retval == ITEM_FOUND) {
1872                 pathrelse(&path_to_key);
1873                 err = -EEXIST;
1874                 goto out_bad_inode;
1875         }
1876         if (old_format_only(sb)) {
1877                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1878                         pathrelse(&path_to_key);
1879                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1880                         err = -EINVAL;
1881                         goto out_bad_inode;
1882                 }
1883                 inode2sd_v1(&sd, inode, inode->i_size);
1884         } else {
1885                 inode2sd(&sd, inode, inode->i_size);
1886         }
1887         // these do not go to on-disk stat data
1888         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1889         inode->i_blksize = reiserfs_default_io_size;
1890
1891         // store in in-core inode the key of stat data and version all
1892         // object items will have (directory items will have old offset
1893         // format, other new objects will consist of new items)
1894         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1895         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1896                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1897         else
1898                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1899         if (old_format_only(sb))
1900                 set_inode_sd_version(inode, STAT_DATA_V1);
1901         else
1902                 set_inode_sd_version(inode, STAT_DATA_V2);
1903
1904         /* insert the stat data into the tree */
1905 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1906         if (REISERFS_I(dir)->new_packing_locality)
1907                 th->displace_new_blocks = 1;
1908 #endif
1909         retval =
1910             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1911                                  (char *)(&sd));
1912         if (retval) {
1913                 err = retval;
1914                 reiserfs_check_path(&path_to_key);
1915                 goto out_bad_inode;
1916         }
1917 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1918         if (!th->displace_new_blocks)
1919                 REISERFS_I(dir)->new_packing_locality = 0;
1920 #endif
1921         if (S_ISDIR(mode)) {
1922                 /* insert item with "." and ".." */
1923                 retval =
1924                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1925         }
1926
1927         if (S_ISLNK(mode)) {
1928                 /* insert body of symlink */
1929                 if (!old_format_only(sb))
1930                         i_size = ROUND_UP(i_size);
1931                 retval =
1932                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1933                                          i_size);
1934         }
1935         if (retval) {
1936                 err = retval;
1937                 reiserfs_check_path(&path_to_key);
1938                 journal_end(th, th->t_super, th->t_blocks_allocated);
1939                 goto out_inserted_sd;
1940         }
1941
1942         /* XXX CHECK THIS */
1943         if (reiserfs_posixacl(inode->i_sb)) {
1944                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1945                 if (retval) {
1946                         err = retval;
1947                         reiserfs_check_path(&path_to_key);
1948                         journal_end(th, th->t_super, th->t_blocks_allocated);
1949                         goto out_inserted_sd;
1950                 }
1951         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1952                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1953                                  "but vfs thinks they are!");
1954         } else if (is_reiserfs_priv_object(dir)) {
1955                 reiserfs_mark_inode_private(inode);
1956         }
1957
1958         insert_inode_hash(inode);
1959         reiserfs_update_sd(th, inode);
1960         reiserfs_check_path(&path_to_key);
1961
1962         return 0;
1963
1964 /* it looks like you can easily compress these two goto targets into
1965  * one.  Keeping it like this doesn't actually hurt anything, and they
1966  * are place holders for what the quota code actually needs.
1967  */
1968       out_bad_inode:
1969         /* Invalidate the object, nothing was inserted yet */
1970         INODE_PKEY(inode)->k_objectid = 0;
1971
1972         /* Quota change must be inside a transaction for journaling */
1973         DQUOT_FREE_INODE(inode);
1974
1975       out_end_trans:
1976         journal_end(th, th->t_super, th->t_blocks_allocated);
1977         /* Drop can be outside and it needs more credits so it's better to have it outside */
1978         DQUOT_DROP(inode);
1979         inode->i_flags |= S_NOQUOTA;
1980         make_bad_inode(inode);
1981
1982       out_inserted_sd:
1983         inode->i_nlink = 0;
1984         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1985
1986         /* If we were inheriting an ACL, we need to release the lock so that
1987          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1988          * code really needs to be reworked, but this will take care of it
1989          * for now. -jeffm */
1990         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1991                 reiserfs_write_unlock_xattrs(dir->i_sb);
1992                 iput(inode);
1993                 reiserfs_write_lock_xattrs(dir->i_sb);
1994         } else
1995                 iput(inode);
1996         return err;
1997 }
1998
1999 /*
2000 ** finds the tail page in the page cache,
2001 ** reads the last block in.
2002 **
2003 ** On success, page_result is set to a locked, pinned page, and bh_result
2004 ** is set to an up to date buffer for the last block in the file.  returns 0.
2005 **
2006 ** tail conversion is not done, so bh_result might not be valid for writing
2007 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2008 ** trying to write the block.
2009 **
2010 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2011 */
2012 static int grab_tail_page(struct inode *p_s_inode,
2013                           struct page **page_result,
2014                           struct buffer_head **bh_result)
2015 {
2016
2017         /* we want the page with the last byte in the file,
2018          ** not the page that will hold the next byte for appending
2019          */
2020         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2021         unsigned long pos = 0;
2022         unsigned long start = 0;
2023         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2024         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2025         struct buffer_head *bh;
2026         struct buffer_head *head;
2027         struct page *page;
2028         int error;
2029
2030         /* we know that we are only called with inode->i_size > 0.
2031          ** we also know that a file tail can never be as big as a block
2032          ** If i_size % blocksize == 0, our file is currently block aligned
2033          ** and it won't need converting or zeroing after a truncate.
2034          */
2035         if ((offset & (blocksize - 1)) == 0) {
2036                 return -ENOENT;
2037         }
2038         page = grab_cache_page(p_s_inode->i_mapping, index);
2039         error = -ENOMEM;
2040         if (!page) {
2041                 goto out;
2042         }
2043         /* start within the page of the last block in the file */
2044         start = (offset / blocksize) * blocksize;
2045
2046         error = block_prepare_write(page, start, offset,
2047                                     reiserfs_get_block_create_0);
2048         if (error)
2049                 goto unlock;
2050
2051         head = page_buffers(page);
2052         bh = head;
2053         do {
2054                 if (pos >= start) {
2055                         break;
2056                 }
2057                 bh = bh->b_this_page;
2058                 pos += blocksize;
2059         } while (bh != head);
2060
2061         if (!buffer_uptodate(bh)) {
2062                 /* note, this should never happen, prepare_write should
2063                  ** be taking care of this for us.  If the buffer isn't up to date,
2064                  ** I've screwed up the code to find the buffer, or the code to
2065                  ** call prepare_write
2066                  */
2067                 reiserfs_warning(p_s_inode->i_sb,
2068                                  "clm-6000: error reading block %lu on dev %s",
2069                                  bh->b_blocknr,
2070                                  reiserfs_bdevname(p_s_inode->i_sb));
2071                 error = -EIO;
2072                 goto unlock;
2073         }
2074         *bh_result = bh;
2075         *page_result = page;
2076
2077       out:
2078         return error;
2079
2080       unlock:
2081         unlock_page(page);
2082         page_cache_release(page);
2083         return error;
2084 }
2085
2086 /*
2087 ** vfs version of truncate file.  Must NOT be called with
2088 ** a transaction already started.
2089 **
2090 ** some code taken from block_truncate_page
2091 */
2092 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2093 {
2094         struct reiserfs_transaction_handle th;
2095         /* we want the offset for the first byte after the end of the file */
2096         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2097         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2098         unsigned length;
2099         struct page *page = NULL;
2100         int error;
2101         struct buffer_head *bh = NULL;
2102
2103         reiserfs_write_lock(p_s_inode->i_sb);
2104
2105         if (p_s_inode->i_size > 0) {
2106                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2107                         // -ENOENT means we truncated past the end of the file, 
2108                         // and get_block_create_0 could not find a block to read in,
2109                         // which is ok.
2110                         if (error != -ENOENT)
2111                                 reiserfs_warning(p_s_inode->i_sb,
2112                                                  "clm-6001: grab_tail_page failed %d",
2113                                                  error);
2114                         page = NULL;
2115                         bh = NULL;
2116                 }
2117         }
2118
2119         /* so, if page != NULL, we have a buffer head for the offset at 
2120          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0, 
2121          ** then we have an unformatted node.  Otherwise, we have a direct item, 
2122          ** and no zeroing is required on disk.  We zero after the truncate, 
2123          ** because the truncate might pack the item anyway 
2124          ** (it will unmap bh if it packs).
2125          */
2126         /* it is enough to reserve space in transaction for 2 balancings:
2127            one for "save" link adding and another for the first
2128            cut_from_item. 1 is for update_sd */
2129         error = journal_begin(&th, p_s_inode->i_sb,
2130                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2131         if (error)
2132                 goto out;
2133         reiserfs_update_inode_transaction(p_s_inode);
2134         if (update_timestamps)
2135                 /* we are doing real truncate: if the system crashes before the last
2136                    transaction of truncating gets committed - on reboot the file
2137                    either appears truncated properly or not truncated at all */
2138                 add_save_link(&th, p_s_inode, 1);
2139         error = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2140         if (error)
2141                 goto out;
2142         error =
2143             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2144         if (error)
2145                 goto out;
2146
2147         if (update_timestamps) {
2148                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2149                 if (error)
2150                         goto out;
2151         }
2152
2153         if (page) {
2154                 length = offset & (blocksize - 1);
2155                 /* if we are not on a block boundary */
2156                 if (length) {
2157                         char *kaddr;
2158
2159                         length = blocksize - length;
2160                         kaddr = kmap_atomic(page, KM_USER0);
2161                         memset(kaddr + offset, 0, length);
2162                         flush_dcache_page(page);
2163                         kunmap_atomic(kaddr, KM_USER0);
2164                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2165                                 mark_buffer_dirty(bh);
2166                         }
2167                 }
2168                 unlock_page(page);
2169                 page_cache_release(page);
2170         }
2171
2172         reiserfs_write_unlock(p_s_inode->i_sb);
2173         return 0;
2174       out:
2175         if (page) {
2176                 unlock_page(page);
2177                 page_cache_release(page);
2178         }
2179         reiserfs_write_unlock(p_s_inode->i_sb);
2180         return error;
2181 }
2182
2183 static int map_block_for_writepage(struct inode *inode,
2184                                    struct buffer_head *bh_result,
2185                                    unsigned long block)
2186 {
2187         struct reiserfs_transaction_handle th;
2188         int fs_gen;
2189         struct item_head tmp_ih;
2190         struct item_head *ih;
2191         struct buffer_head *bh;
2192         __le32 *item;
2193         struct cpu_key key;
2194         INITIALIZE_PATH(path);
2195         int pos_in_item;
2196         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2197         loff_t byte_offset = (block << inode->i_sb->s_blocksize_bits) + 1;
2198         int retval;
2199         int use_get_block = 0;
2200         int bytes_copied = 0;
2201         int copy_size;
2202         int trans_running = 0;
2203
2204         /* catch places below that try to log something without starting a trans */
2205         th.t_trans_id = 0;
2206
2207         if (!buffer_uptodate(bh_result)) {
2208                 return -EIO;
2209         }
2210
2211         kmap(bh_result->b_page);
2212       start_over:
2213         reiserfs_write_lock(inode->i_sb);
2214         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2215
2216       research:
2217         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2218         if (retval != POSITION_FOUND) {
2219                 use_get_block = 1;
2220                 goto out;
2221         }
2222
2223         bh = get_last_bh(&path);
2224         ih = get_ih(&path);
2225         item = get_item(&path);
2226         pos_in_item = path.pos_in_item;
2227
2228         /* we've found an unformatted node */
2229         if (indirect_item_found(retval, ih)) {
2230                 if (bytes_copied > 0) {
2231                         reiserfs_warning(inode->i_sb,
2232                                          "clm-6002: bytes_copied %d",
2233                                          bytes_copied);
2234                 }
2235                 if (!get_block_num(item, pos_in_item)) {
2236                         /* crap, we are writing to a hole */
2237                         use_get_block = 1;
2238                         goto out;
2239                 }
2240                 set_block_dev_mapped(bh_result,
2241                                      get_block_num(item, pos_in_item), inode);
2242         } else if (is_direct_le_ih(ih)) {
2243                 char *p;
2244                 p = page_address(bh_result->b_page);
2245                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2246                 copy_size = ih_item_len(ih) - pos_in_item;
2247
2248                 fs_gen = get_generation(inode->i_sb);
2249                 copy_item_head(&tmp_ih, ih);
2250
2251                 if (!trans_running) {
2252                         /* vs-3050 is gone, no need to drop the path */
2253                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2254                         if (retval)
2255                                 goto out;
2256                         reiserfs_update_inode_transaction(inode);
2257                         trans_running = 1;
2258                         if (fs_changed(fs_gen, inode->i_sb)
2259                             && item_moved(&tmp_ih, &path)) {
2260                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2261                                                                  bh);
2262                                 goto research;
2263                         }
2264                 }
2265
2266                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2267
2268                 if (fs_changed(fs_gen, inode->i_sb)
2269                     && item_moved(&tmp_ih, &path)) {
2270                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2271                         goto research;
2272                 }
2273
2274                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2275                        copy_size);
2276
2277                 journal_mark_dirty(&th, inode->i_sb, bh);
2278                 bytes_copied += copy_size;
2279                 set_block_dev_mapped(bh_result, 0, inode);
2280
2281                 /* are there still bytes left? */
2282                 if (bytes_copied < bh_result->b_size &&
2283                     (byte_offset + bytes_copied) < inode->i_size) {
2284                         set_cpu_key_k_offset(&key,
2285                                              cpu_key_k_offset(&key) +
2286                                              copy_size);
2287                         goto research;
2288                 }
2289         } else {
2290                 reiserfs_warning(inode->i_sb,
2291                                  "clm-6003: bad item inode %lu, device %s",
2292                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2293                 retval = -EIO;
2294                 goto out;
2295         }
2296         retval = 0;
2297
2298       out:
2299         pathrelse(&path);
2300         if (trans_running) {
2301                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2302                 if (err)
2303                         retval = err;
2304                 trans_running = 0;
2305         }
2306         reiserfs_write_unlock(inode->i_sb);
2307
2308         /* this is where we fill in holes in the file. */
2309         if (use_get_block) {
2310                 retval = reiserfs_get_block(inode, block, bh_result,
2311                                             GET_BLOCK_CREATE | GET_BLOCK_NO_ISEM
2312                                             | GET_BLOCK_NO_DANGLE);
2313                 if (!retval) {
2314                         if (!buffer_mapped(bh_result)
2315                             || bh_result->b_blocknr == 0) {
2316                                 /* get_block failed to find a mapped unformatted node. */
2317                                 use_get_block = 0;
2318                                 goto start_over;
2319                         }
2320                 }
2321         }
2322         kunmap(bh_result->b_page);
2323
2324         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2325                 /* we've copied data from the page into the direct item, so the
2326                  * buffer in the page is now clean, mark it to reflect that.
2327                  */
2328                 lock_buffer(bh_result);
2329                 clear_buffer_dirty(bh_result);
2330                 unlock_buffer(bh_result);
2331         }
2332         return retval;
2333 }
2334
2335 /* 
2336  * mason@suse.com: updated in 2.5.54 to follow the same general io 
2337  * start/recovery path as __block_write_full_page, along with special
2338  * code to handle reiserfs tails.
2339  */
2340 static int reiserfs_write_full_page(struct page *page,
2341                                     struct writeback_control *wbc)
2342 {
2343         struct inode *inode = page->mapping->host;
2344         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2345         int error = 0;
2346         unsigned long block;
2347         struct buffer_head *head, *bh;
2348         int partial = 0;
2349         int nr = 0;
2350         int checked = PageChecked(page);
2351         struct reiserfs_transaction_handle th;
2352         struct super_block *s = inode->i_sb;
2353         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2354         th.t_trans_id = 0;
2355
2356         /* The page dirty bit is cleared before writepage is called, which
2357          * means we have to tell create_empty_buffers to make dirty buffers
2358          * The page really should be up to date at this point, so tossing
2359          * in the BH_Uptodate is just a sanity check.
2360          */
2361         if (!page_has_buffers(page)) {
2362                 create_empty_buffers(page, s->s_blocksize,
2363                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2364         }
2365         head = page_buffers(page);
2366
2367         /* last page in the file, zero out any contents past the
2368          ** last byte in the file
2369          */
2370         if (page->index >= end_index) {
2371                 char *kaddr;
2372                 unsigned last_offset;
2373
2374                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2375                 /* no file contents in this page */
2376                 if (page->index >= end_index + 1 || !last_offset) {
2377                         unlock_page(page);
2378                         return 0;
2379                 }
2380                 kaddr = kmap_atomic(page, KM_USER0);
2381                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2382                 flush_dcache_page(page);
2383                 kunmap_atomic(kaddr, KM_USER0);
2384         }
2385         bh = head;
2386         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2387         /* first map all the buffers, logging any direct items we find */
2388         do {
2389                 if ((checked || buffer_dirty(bh)) && (!buffer_mapped(bh) ||
2390                                                       (buffer_mapped(bh)
2391                                                        && bh->b_blocknr ==
2392                                                        0))) {
2393                         /* not mapped yet, or it points to a direct item, search
2394                          * the btree for the mapping info, and log any direct
2395                          * items found
2396                          */
2397                         if ((error = map_block_for_writepage(inode, bh, block))) {
2398                                 goto fail;
2399                         }
2400                 }
2401                 bh = bh->b_this_page;
2402                 block++;
2403         } while (bh != head);
2404
2405         /*
2406          * we start the transaction after map_block_for_writepage,
2407          * because it can create holes in the file (an unbounded operation).
2408          * starting it here, we can make a reliable estimate for how many
2409          * blocks we're going to log
2410          */
2411         if (checked) {
2412                 ClearPageChecked(page);
2413                 reiserfs_write_lock(s);
2414                 error = journal_begin(&th, s, bh_per_page + 1);
2415                 if (error) {
2416                         reiserfs_write_unlock(s);
2417                         goto fail;
2418                 }
2419                 reiserfs_update_inode_transaction(inode);
2420         }
2421         /* now go through and lock any dirty buffers on the page */
2422         do {
2423                 get_bh(bh);
2424                 if (!buffer_mapped(bh))
2425                         continue;
2426                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2427                         continue;
2428
2429                 if (checked) {
2430                         reiserfs_prepare_for_journal(s, bh, 1);
2431                         journal_mark_dirty(&th, s, bh);
2432                         continue;
2433                 }
2434                 /* from this point on, we know the buffer is mapped to a
2435                  * real block and not a direct item
2436                  */
2437                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2438                         lock_buffer(bh);
2439                 } else {
2440                         if (test_set_buffer_locked(bh)) {
2441                                 redirty_page_for_writepage(wbc, page);
2442                                 continue;
2443                         }
2444                 }
2445                 if (test_clear_buffer_dirty(bh)) {
2446                         mark_buffer_async_write(bh);
2447                 } else {
2448                         unlock_buffer(bh);
2449                 }
2450         } while ((bh = bh->b_this_page) != head);
2451
2452         if (checked) {
2453                 error = journal_end(&th, s, bh_per_page + 1);
2454                 reiserfs_write_unlock(s);
2455                 if (error)
2456                         goto fail;
2457         }
2458         BUG_ON(PageWriteback(page));
2459         set_page_writeback(page);
2460         unlock_page(page);
2461
2462         /*
2463          * since any buffer might be the only dirty buffer on the page, 
2464          * the first submit_bh can bring the page out of writeback.
2465          * be careful with the buffers.
2466          */
2467         do {
2468                 struct buffer_head *next = bh->b_this_page;
2469                 if (buffer_async_write(bh)) {
2470                         submit_bh(WRITE, bh);
2471                         nr++;
2472                 }
2473                 put_bh(bh);
2474                 bh = next;
2475         } while (bh != head);
2476
2477         error = 0;
2478       done:
2479         if (nr == 0) {
2480                 /*
2481                  * if this page only had a direct item, it is very possible for
2482                  * no io to be required without there being an error.  Or, 
2483                  * someone else could have locked them and sent them down the 
2484                  * pipe without locking the page
2485                  */
2486                 bh = head;
2487                 do {
2488                         if (!buffer_uptodate(bh)) {
2489                                 partial = 1;
2490                                 break;
2491                         }
2492                         bh = bh->b_this_page;
2493                 } while (bh != head);
2494                 if (!partial)
2495                         SetPageUptodate(page);
2496                 end_page_writeback(page);
2497         }
2498         return error;
2499
2500       fail:
2501         /* catches various errors, we need to make sure any valid dirty blocks
2502          * get to the media.  The page is currently locked and not marked for 
2503          * writeback
2504          */
2505         ClearPageUptodate(page);
2506         bh = head;
2507         do {
2508                 get_bh(bh);
2509                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2510                         lock_buffer(bh);
2511                         mark_buffer_async_write(bh);
2512                 } else {
2513                         /*
2514                          * clear any dirty bits that might have come from getting
2515                          * attached to a dirty page
2516                          */
2517                         clear_buffer_dirty(bh);
2518                 }
2519                 bh = bh->b_this_page;
2520         } while (bh != head);
2521         SetPageError(page);
2522         BUG_ON(PageWriteback(page));
2523         set_page_writeback(page);
2524         unlock_page(page);
2525         do {
2526                 struct buffer_head *next = bh->b_this_page;
2527                 if (buffer_async_write(bh)) {
2528                         clear_buffer_dirty(bh);
2529                         submit_bh(WRITE, bh);
2530                         nr++;
2531                 }
2532                 put_bh(bh);
2533                 bh = next;
2534         } while (bh != head);
2535         goto done;
2536 }
2537
2538 static int reiserfs_readpage(struct file *f, struct page *page)
2539 {
2540         return block_read_full_page(page, reiserfs_get_block);
2541 }
2542
2543 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2544 {
2545         struct inode *inode = page->mapping->host;
2546         reiserfs_wait_on_write_block(inode->i_sb);
2547         return reiserfs_write_full_page(page, wbc);
2548 }
2549
2550 static int reiserfs_prepare_write(struct file *f, struct page *page,
2551                                   unsigned from, unsigned to)
2552 {
2553         struct inode *inode = page->mapping->host;
2554         int ret;
2555         int old_ref = 0;
2556
2557         reiserfs_wait_on_write_block(inode->i_sb);
2558         fix_tail_page_for_writing(page);
2559         if (reiserfs_transaction_running(inode->i_sb)) {
2560                 struct reiserfs_transaction_handle *th;
2561                 th = (struct reiserfs_transaction_handle *)current->
2562                     journal_info;
2563                 BUG_ON(!th->t_refcount);
2564                 BUG_ON(!th->t_trans_id);
2565                 old_ref = th->t_refcount;
2566                 th->t_refcount++;
2567         }
2568
2569         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2570         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2571                 struct reiserfs_transaction_handle *th = current->journal_info;
2572                 /* this gets a little ugly.  If reiserfs_get_block returned an
2573                  * error and left a transacstion running, we've got to close it,
2574                  * and we've got to free handle if it was a persistent transaction.
2575                  *
2576                  * But, if we had nested into an existing transaction, we need
2577                  * to just drop the ref count on the handle.
2578                  *
2579                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2580                  * and it was a persistent trans.  Otherwise, it was nested above.
2581                  */
2582                 if (th->t_refcount > old_ref) {
2583                         if (old_ref)
2584                                 th->t_refcount--;
2585                         else {
2586                                 int err;
2587                                 reiserfs_write_lock(inode->i_sb);
2588                                 err = reiserfs_end_persistent_transaction(th);
2589                                 reiserfs_write_unlock(inode->i_sb);
2590                                 if (err)
2591                                         ret = err;
2592                         }
2593                 }
2594         }
2595         return ret;
2596
2597 }
2598
2599 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2600 {
2601         return generic_block_bmap(as, block, reiserfs_bmap);
2602 }
2603
2604 static int reiserfs_commit_write(struct file *f, struct page *page,
2605                                  unsigned from, unsigned to)
2606 {
2607         struct inode *inode = page->mapping->host;
2608         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2609         int ret = 0;
2610         int update_sd = 0;
2611         struct reiserfs_transaction_handle *th = NULL;
2612
2613         reiserfs_wait_on_write_block(inode->i_sb);
2614         if (reiserfs_transaction_running(inode->i_sb)) {
2615                 th = current->journal_info;
2616         }
2617         reiserfs_commit_page(inode, page, from, to);
2618
2619         /* generic_commit_write does this for us, but does not update the
2620          ** transaction tracking stuff when the size changes.  So, we have
2621          ** to do the i_size updates here.
2622          */
2623         if (pos > inode->i_size) {
2624                 struct reiserfs_transaction_handle myth;
2625                 reiserfs_write_lock(inode->i_sb);
2626                 /* If the file have grown beyond the border where it
2627                    can have a tail, unmark it as needing a tail
2628                    packing */
2629                 if ((have_large_tails(inode->i_sb)
2630                      && inode->i_size > i_block_size(inode) * 4)
2631                     || (have_small_tails(inode->i_sb)
2632                         && inode->i_size > i_block_size(inode)))
2633                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2634
2635                 ret = journal_begin(&myth, inode->i_sb, 1);
2636                 if (ret) {
2637                         reiserfs_write_unlock(inode->i_sb);
2638                         goto journal_error;
2639                 }
2640                 reiserfs_update_inode_transaction(inode);
2641                 inode->i_size = pos;
2642                 reiserfs_update_sd(&myth, inode);
2643                 update_sd = 1;
2644                 ret = journal_end(&myth, inode->i_sb, 1);
2645                 reiserfs_write_unlock(inode->i_sb);
2646                 if (ret)
2647                         goto journal_error;
2648         }
2649         if (th) {
2650                 reiserfs_write_lock(inode->i_sb);
2651                 if (!update_sd)
2652                         reiserfs_update_sd(th, inode);
2653                 ret = reiserfs_end_persistent_transaction(th);
2654                 reiserfs_write_unlock(inode->i_sb);
2655                 if (ret)
2656                         goto out;
2657         }
2658
2659         /* we test for O_SYNC here so we can commit the transaction
2660          ** for any packed tails the file might have had
2661          */
2662         if (f && (f->f_flags & O_SYNC)) {
2663                 reiserfs_write_lock(inode->i_sb);
2664                 ret = reiserfs_commit_for_inode(inode);
2665                 reiserfs_write_unlock(inode->i_sb);
2666         }
2667       out:
2668         return ret;
2669
2670       journal_error:
2671         if (th) {
2672                 reiserfs_write_lock(inode->i_sb);
2673                 if (!update_sd)
2674                         reiserfs_update_sd(th, inode);
2675                 ret = reiserfs_end_persistent_transaction(th);
2676                 reiserfs_write_unlock(inode->i_sb);
2677         }
2678
2679         return ret;
2680 }
2681
2682 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2683 {
2684         if (reiserfs_attrs(inode->i_sb)) {
2685                 if (sd_attrs & REISERFS_SYNC_FL)
2686                         inode->i_flags |= S_SYNC;
2687                 else
2688                         inode->i_flags &= ~S_SYNC;
2689                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2690                         inode->i_flags |= S_IMMUTABLE;
2691                 else
2692                         inode->i_flags &= ~S_IMMUTABLE;
2693                 if (sd_attrs & REISERFS_APPEND_FL)
2694                         inode->i_flags |= S_APPEND;
2695                 else
2696                         inode->i_flags &= ~S_APPEND;
2697                 if (sd_attrs & REISERFS_NOATIME_FL)
2698                         inode->i_flags |= S_NOATIME;
2699                 else
2700                         inode->i_flags &= ~S_NOATIME;
2701                 if (sd_attrs & REISERFS_NOTAIL_FL)
2702                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2703                 else
2704                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2705         }
2706 }
2707
2708 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2709 {
2710         if (reiserfs_attrs(inode->i_sb)) {
2711                 if (inode->i_flags & S_IMMUTABLE)
2712                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2713                 else
2714                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2715                 if (inode->i_flags & S_SYNC)
2716                         *sd_attrs |= REISERFS_SYNC_FL;
2717                 else
2718                         *sd_attrs &= ~REISERFS_SYNC_FL;
2719                 if (inode->i_flags & S_NOATIME)
2720                         *sd_attrs |= REISERFS_NOATIME_FL;
2721                 else
2722                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2723                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2724                         *sd_attrs |= REISERFS_NOTAIL_FL;
2725                 else
2726                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2727         }
2728 }
2729
2730 /* decide if this buffer needs to stay around for data logging or ordered
2731 ** write purposes
2732 */
2733 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2734 {
2735         int ret = 1;
2736         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2737
2738         spin_lock(&j->j_dirty_buffers_lock);
2739         if (!buffer_mapped(bh)) {
2740                 goto free_jh;
2741         }
2742         /* the page is locked, and the only places that log a data buffer
2743          * also lock the page.
2744          */
2745         if (reiserfs_file_data_log(inode)) {
2746                 /*
2747                  * very conservative, leave the buffer pinned if
2748                  * anyone might need it.
2749                  */
2750                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2751                         ret = 0;
2752                 }
2753         } else if (buffer_dirty(bh) || buffer_locked(bh)) {
2754                 struct reiserfs_journal_list *jl;
2755                 struct reiserfs_jh *jh = bh->b_private;
2756
2757                 /* why is this safe?
2758                  * reiserfs_setattr updates i_size in the on disk
2759                  * stat data before allowing vmtruncate to be called.
2760                  *
2761                  * If buffer was put onto the ordered list for this
2762                  * transaction, we know for sure either this transaction
2763                  * or an older one already has updated i_size on disk,
2764                  * and this ordered data won't be referenced in the file
2765                  * if we crash.
2766                  *
2767                  * if the buffer was put onto the ordered list for an older
2768                  * transaction, we need to leave it around
2769                  */
2770                 if (jh && (jl = jh->jl)
2771                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2772                         ret = 0;
2773         }
2774       free_jh:
2775         if (ret && bh->b_private) {
2776                 reiserfs_free_jh(bh);
2777         }
2778         spin_unlock(&j->j_dirty_buffers_lock);
2779         return ret;
2780 }
2781
2782 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2783 static int reiserfs_invalidatepage(struct page *page, unsigned long offset)
2784 {
2785         struct buffer_head *head, *bh, *next;
2786         struct inode *inode = page->mapping->host;
2787         unsigned int curr_off = 0;
2788         int ret = 1;
2789
2790         BUG_ON(!PageLocked(page));
2791
2792         if (offset == 0)
2793                 ClearPageChecked(page);
2794
2795         if (!page_has_buffers(page))
2796                 goto out;
2797
2798         head = page_buffers(page);
2799         bh = head;
2800         do {
2801                 unsigned int next_off = curr_off + bh->b_size;
2802                 next = bh->b_this_page;
2803
2804                 /*
2805                  * is this block fully invalidated?
2806                  */
2807                 if (offset <= curr_off) {
2808                         if (invalidatepage_can_drop(inode, bh))
2809                                 reiserfs_unmap_buffer(bh);
2810                         else
2811                                 ret = 0;
2812                 }
2813                 curr_off = next_off;
2814                 bh = next;
2815         } while (bh != head);
2816
2817         /*
2818          * We release buffers only if the entire page is being invalidated.
2819          * The get_block cached value has been unconditionally invalidated,
2820          * so real IO is not possible anymore.
2821          */
2822         if (!offset && ret)
2823                 ret = try_to_release_page(page, 0);
2824       out:
2825         return ret;
2826 }
2827
2828 static int reiserfs_set_page_dirty(struct page *page)
2829 {
2830         struct inode *inode = page->mapping->host;
2831         if (reiserfs_file_data_log(inode)) {
2832                 SetPageChecked(page);
2833                 return __set_page_dirty_nobuffers(page);
2834         }
2835         return __set_page_dirty_buffers(page);
2836 }
2837
2838 /*
2839  * Returns 1 if the page's buffers were dropped.  The page is locked.
2840  *
2841  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2842  * in the buffers at page_buffers(page).
2843  *
2844  * even in -o notail mode, we can't be sure an old mount without -o notail
2845  * didn't create files with tails.
2846  */
2847 static int reiserfs_releasepage(struct page *page, int unused_gfp_flags)
2848 {
2849         struct inode *inode = page->mapping->host;
2850         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2851         struct buffer_head *head;
2852         struct buffer_head *bh;
2853         int ret = 1;
2854
2855         WARN_ON(PageChecked(page));
2856         spin_lock(&j->j_dirty_buffers_lock);
2857         head = page_buffers(page);
2858         bh = head;
2859         do {
2860                 if (bh->b_private) {
2861                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2862                                 reiserfs_free_jh(bh);
2863                         } else {
2864                                 ret = 0;
2865                                 break;
2866                         }
2867                 }
2868                 bh = bh->b_this_page;
2869         } while (bh != head);
2870         if (ret)
2871                 ret = try_to_free_buffers(page);
2872         spin_unlock(&j->j_dirty_buffers_lock);
2873         return ret;
2874 }
2875
2876 /* We thank Mingming Cao for helping us understand in great detail what
2877    to do in this section of the code. */
2878 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2879                                   const struct iovec *iov, loff_t offset,
2880                                   unsigned long nr_segs)
2881 {
2882         struct file *file = iocb->ki_filp;
2883         struct inode *inode = file->f_mapping->host;
2884
2885         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2886                                   offset, nr_segs,
2887                                   reiserfs_get_blocks_direct_io, NULL);
2888 }
2889
2890 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2891 {
2892         struct inode *inode = dentry->d_inode;
2893         int error;
2894         unsigned int ia_valid = attr->ia_valid;
2895         reiserfs_write_lock(inode->i_sb);
2896         if (attr->ia_valid & ATTR_SIZE) {
2897                 /* version 2 items will be caught by the s_maxbytes check
2898                  ** done for us in vmtruncate
2899                  */
2900                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2901                     attr->ia_size > MAX_NON_LFS) {
2902                         error = -EFBIG;
2903                         goto out;
2904                 }
2905                 /* fill in hole pointers in the expanding truncate case. */
2906                 if (attr->ia_size > inode->i_size) {
2907                         error = generic_cont_expand(inode, attr->ia_size);
2908                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2909                                 int err;
2910                                 struct reiserfs_transaction_handle th;
2911                                 /* we're changing at most 2 bitmaps, inode + super */
2912                                 err = journal_begin(&th, inode->i_sb, 4);
2913                                 if (!err) {
2914                                         reiserfs_discard_prealloc(&th, inode);
2915                                         err = journal_end(&th, inode->i_sb, 4);
2916                                 }
2917                                 if (err)
2918                                         error = err;
2919                         }
2920                         if (error)
2921                                 goto out;
2922                 }
2923         }
2924
2925         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2926              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2927             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2928                 /* stat data of format v3.5 has 16 bit uid and gid */
2929                 error = -EINVAL;
2930                 goto out;
2931         }
2932
2933         error = inode_change_ok(inode, attr);
2934         if (!error) {
2935                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2936                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2937                         error = reiserfs_chown_xattrs(inode, attr);
2938
2939                         if (!error) {
2940                                 struct reiserfs_transaction_handle th;
2941                                 int jbegin_count =
2942                                     2 *
2943                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2944                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2945                                     2;
2946
2947                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2948                                 error =
2949                                     journal_begin(&th, inode->i_sb,
2950                                                   jbegin_count);
2951                                 if (error)
2952                                         goto out;
2953                                 error =
2954                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2955                                 if (error) {
2956                                         journal_end(&th, inode->i_sb,
2957                                                     jbegin_count);
2958                                         goto out;
2959                                 }
2960                                 /* Update corresponding info in inode so that everything is in
2961                                  * one transaction */
2962                                 if (attr->ia_valid & ATTR_UID)
2963                                         inode->i_uid = attr->ia_uid;
2964                                 if (attr->ia_valid & ATTR_GID)
2965                                         inode->i_gid = attr->ia_gid;
2966                                 mark_inode_dirty(inode);
2967                                 error =
2968                                     journal_end(&th, inode->i_sb, jbegin_count);
2969                         }
2970                 }
2971                 if (!error)
2972                         error = inode_setattr(inode, attr);
2973         }
2974
2975         if (!error && reiserfs_posixacl(inode->i_sb)) {
2976                 if (attr->ia_valid & ATTR_MODE)
2977                         error = reiserfs_acl_chmod(inode);
2978         }
2979
2980       out:
2981         reiserfs_write_unlock(inode->i_sb);
2982         return error;
2983 }
2984
2985 struct address_space_operations reiserfs_address_space_operations = {
2986         .writepage = reiserfs_writepage,
2987         .readpage = reiserfs_readpage,
2988         .readpages = reiserfs_readpages,
2989         .releasepage = reiserfs_releasepage,
2990         .invalidatepage = reiserfs_invalidatepage,
2991         .sync_page = block_sync_page,
2992         .prepare_write = reiserfs_prepare_write,
2993         .commit_write = reiserfs_commit_write,
2994         .bmap = reiserfs_aop_bmap,
2995         .direct_IO = reiserfs_direct_IO,
2996         .set_page_dirty = reiserfs_set_page_dirty,
2997 };