[PATCH] reiserfs: ifdef ACL stuff from inode
[linux-2.6.git] / fs / reiserfs / inode.c
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4
5 #include <linux/time.h>
6 #include <linux/fs.h>
7 #include <linux/reiserfs_fs.h>
8 #include <linux/reiserfs_acl.h>
9 #include <linux/reiserfs_xattr.h>
10 #include <linux/smp_lock.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <asm/uaccess.h>
14 #include <asm/unaligned.h>
15 #include <linux/buffer_head.h>
16 #include <linux/mpage.h>
17 #include <linux/writeback.h>
18 #include <linux/quotaops.h>
19
20 static int reiserfs_commit_write(struct file *f, struct page *page,
21                                  unsigned from, unsigned to);
22 static int reiserfs_prepare_write(struct file *f, struct page *page,
23                                   unsigned from, unsigned to);
24
25 void reiserfs_delete_inode(struct inode *inode)
26 {
27         /* We need blocks for transaction + (user+group) quota update (possibly delete) */
28         int jbegin_count =
29             JOURNAL_PER_BALANCE_CNT * 2 +
30             2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
31         struct reiserfs_transaction_handle th;
32         int err;
33
34         truncate_inode_pages(&inode->i_data, 0);
35
36         reiserfs_write_lock(inode->i_sb);
37
38         /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
39         if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {  /* also handles bad_inode case */
40                 reiserfs_delete_xattrs(inode);
41
42                 if (journal_begin(&th, inode->i_sb, jbegin_count))
43                         goto out;
44                 reiserfs_update_inode_transaction(inode);
45
46                 err = reiserfs_delete_object(&th, inode);
47
48                 /* Do quota update inside a transaction for journaled quotas. We must do that
49                  * after delete_object so that quota updates go into the same transaction as
50                  * stat data deletion */
51                 if (!err) 
52                         DQUOT_FREE_INODE(inode);
53
54                 if (journal_end(&th, inode->i_sb, jbegin_count))
55                         goto out;
56
57                 /* check return value from reiserfs_delete_object after
58                  * ending the transaction
59                  */
60                 if (err)
61                     goto out;
62
63                 /* all items of file are deleted, so we can remove "save" link */
64                 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
65                                                                  * about an error here */
66         } else {
67                 /* no object items are in the tree */
68                 ;
69         }
70       out:
71         clear_inode(inode);     /* note this must go after the journal_end to prevent deadlock */
72         inode->i_blocks = 0;
73         reiserfs_write_unlock(inode->i_sb);
74 }
75
76 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
77                           __u32 objectid, loff_t offset, int type, int length)
78 {
79         key->version = version;
80
81         key->on_disk_key.k_dir_id = dirid;
82         key->on_disk_key.k_objectid = objectid;
83         set_cpu_key_k_offset(key, offset);
84         set_cpu_key_k_type(key, type);
85         key->key_length = length;
86 }
87
88 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
89    offset and type of key */
90 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
91                   int type, int length)
92 {
93         _make_cpu_key(key, get_inode_item_key_version(inode),
94                       le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
95                       le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
96                       length);
97 }
98
99 //
100 // when key is 0, do not set version and short key
101 //
102 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
103                               int version,
104                               loff_t offset, int type, int length,
105                               int entry_count /*or ih_free_space */ )
106 {
107         if (key) {
108                 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
109                 ih->ih_key.k_objectid =
110                     cpu_to_le32(key->on_disk_key.k_objectid);
111         }
112         put_ih_version(ih, version);
113         set_le_ih_k_offset(ih, offset);
114         set_le_ih_k_type(ih, type);
115         put_ih_item_len(ih, length);
116         /*    set_ih_free_space (ih, 0); */
117         // for directory items it is entry count, for directs and stat
118         // datas - 0xffff, for indirects - 0
119         put_ih_entry_count(ih, entry_count);
120 }
121
122 //
123 // FIXME: we might cache recently accessed indirect item
124
125 // Ugh.  Not too eager for that....
126 //  I cut the code until such time as I see a convincing argument (benchmark).
127 // I don't want a bloated inode struct..., and I don't like code complexity....
128
129 /* cutting the code is fine, since it really isn't in use yet and is easy
130 ** to add back in.  But, Vladimir has a really good idea here.  Think
131 ** about what happens for reading a file.  For each page,
132 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
133 ** an indirect item.  This indirect item has X number of pointers, where
134 ** X is a big number if we've done the block allocation right.  But,
135 ** we only use one or two of these pointers during each call to readpage,
136 ** needlessly researching again later on.
137 **
138 ** The size of the cache could be dynamic based on the size of the file.
139 **
140 ** I'd also like to see us cache the location the stat data item, since
141 ** we are needlessly researching for that frequently.
142 **
143 ** --chris
144 */
145
146 /* If this page has a file tail in it, and
147 ** it was read in by get_block_create_0, the page data is valid,
148 ** but tail is still sitting in a direct item, and we can't write to
149 ** it.  So, look through this page, and check all the mapped buffers
150 ** to make sure they have valid block numbers.  Any that don't need
151 ** to be unmapped, so that block_prepare_write will correctly call
152 ** reiserfs_get_block to convert the tail into an unformatted node
153 */
154 static inline void fix_tail_page_for_writing(struct page *page)
155 {
156         struct buffer_head *head, *next, *bh;
157
158         if (page && page_has_buffers(page)) {
159                 head = page_buffers(page);
160                 bh = head;
161                 do {
162                         next = bh->b_this_page;
163                         if (buffer_mapped(bh) && bh->b_blocknr == 0) {
164                                 reiserfs_unmap_buffer(bh);
165                         }
166                         bh = next;
167                 } while (bh != head);
168         }
169 }
170
171 /* reiserfs_get_block does not need to allocate a block only if it has been
172    done already or non-hole position has been found in the indirect item */
173 static inline int allocation_needed(int retval, b_blocknr_t allocated,
174                                     struct item_head *ih,
175                                     __le32 * item, int pos_in_item)
176 {
177         if (allocated)
178                 return 0;
179         if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
180             get_block_num(item, pos_in_item))
181                 return 0;
182         return 1;
183 }
184
185 static inline int indirect_item_found(int retval, struct item_head *ih)
186 {
187         return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
188 }
189
190 static inline void set_block_dev_mapped(struct buffer_head *bh,
191                                         b_blocknr_t block, struct inode *inode)
192 {
193         map_bh(bh, inode->i_sb, block);
194 }
195
196 //
197 // files which were created in the earlier version can not be longer,
198 // than 2 gb
199 //
200 static int file_capable(struct inode *inode, long block)
201 {
202         if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||      // it is new file.
203             block < (1 << (31 - inode->i_sb->s_blocksize_bits)))        // old file, but 'block' is inside of 2gb
204                 return 1;
205
206         return 0;
207 }
208
209 /*static*/ int restart_transaction(struct reiserfs_transaction_handle *th,
210                                    struct inode *inode, struct path *path)
211 {
212         struct super_block *s = th->t_super;
213         int len = th->t_blocks_allocated;
214         int err;
215
216         BUG_ON(!th->t_trans_id);
217         BUG_ON(!th->t_refcount);
218
219         /* we cannot restart while nested */
220         if (th->t_refcount > 1) {
221                 return 0;
222         }
223         pathrelse(path);
224         reiserfs_update_sd(th, inode);
225         err = journal_end(th, s, len);
226         if (!err) {
227                 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
228                 if (!err)
229                         reiserfs_update_inode_transaction(inode);
230         }
231         return err;
232 }
233
234 // it is called by get_block when create == 0. Returns block number
235 // for 'block'-th logical block of file. When it hits direct item it
236 // returns 0 (being called from bmap) or read direct item into piece
237 // of page (bh_result)
238
239 // Please improve the english/clarity in the comment above, as it is
240 // hard to understand.
241
242 static int _get_block_create_0(struct inode *inode, long block,
243                                struct buffer_head *bh_result, int args)
244 {
245         INITIALIZE_PATH(path);
246         struct cpu_key key;
247         struct buffer_head *bh;
248         struct item_head *ih, tmp_ih;
249         int fs_gen;
250         int blocknr;
251         char *p = NULL;
252         int chars;
253         int ret;
254         int result;
255         int done = 0;
256         unsigned long offset;
257
258         // prepare the key to look for the 'block'-th block of file
259         make_cpu_key(&key, inode,
260                      (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
261                      3);
262
263       research:
264         result = search_for_position_by_key(inode->i_sb, &key, &path);
265         if (result != POSITION_FOUND) {
266                 pathrelse(&path);
267                 if (p)
268                         kunmap(bh_result->b_page);
269                 if (result == IO_ERROR)
270                         return -EIO;
271                 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
272                 // That there is some MMAPED data associated with it that is yet to be written to disk.
273                 if ((args & GET_BLOCK_NO_HOLE)
274                     && !PageUptodate(bh_result->b_page)) {
275                         return -ENOENT;
276                 }
277                 return 0;
278         }
279         //
280         bh = get_last_bh(&path);
281         ih = get_ih(&path);
282         if (is_indirect_le_ih(ih)) {
283                 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
284
285                 /* FIXME: here we could cache indirect item or part of it in
286                    the inode to avoid search_by_key in case of subsequent
287                    access to file */
288                 blocknr = get_block_num(ind_item, path.pos_in_item);
289                 ret = 0;
290                 if (blocknr) {
291                         map_bh(bh_result, inode->i_sb, blocknr);
292                         if (path.pos_in_item ==
293                             ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
294                                 set_buffer_boundary(bh_result);
295                         }
296                 } else
297                         // We do not return -ENOENT if there is a hole but page is uptodate, because it means
298                         // That there is some MMAPED data associated with it that is yet to  be written to disk.
299                 if ((args & GET_BLOCK_NO_HOLE)
300                             && !PageUptodate(bh_result->b_page)) {
301                         ret = -ENOENT;
302                 }
303
304                 pathrelse(&path);
305                 if (p)
306                         kunmap(bh_result->b_page);
307                 return ret;
308         }
309         // requested data are in direct item(s)
310         if (!(args & GET_BLOCK_READ_DIRECT)) {
311                 // we are called by bmap. FIXME: we can not map block of file
312                 // when it is stored in direct item(s)
313                 pathrelse(&path);
314                 if (p)
315                         kunmap(bh_result->b_page);
316                 return -ENOENT;
317         }
318
319         /* if we've got a direct item, and the buffer or page was uptodate,
320          ** we don't want to pull data off disk again.  skip to the
321          ** end, where we map the buffer and return
322          */
323         if (buffer_uptodate(bh_result)) {
324                 goto finished;
325         } else
326                 /*
327                  ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
328                  ** pages without any buffers.  If the page is up to date, we don't want
329                  ** read old data off disk.  Set the up to date bit on the buffer instead
330                  ** and jump to the end
331                  */
332         if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
333                 set_buffer_uptodate(bh_result);
334                 goto finished;
335         }
336         // read file tail into part of page
337         offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
338         fs_gen = get_generation(inode->i_sb);
339         copy_item_head(&tmp_ih, ih);
340
341         /* we only want to kmap if we are reading the tail into the page.
342          ** this is not the common case, so we don't kmap until we are
343          ** sure we need to.  But, this means the item might move if
344          ** kmap schedules
345          */
346         if (!p) {
347                 p = (char *)kmap(bh_result->b_page);
348                 if (fs_changed(fs_gen, inode->i_sb)
349                     && item_moved(&tmp_ih, &path)) {
350                         goto research;
351                 }
352         }
353         p += offset;
354         memset(p, 0, inode->i_sb->s_blocksize);
355         do {
356                 if (!is_direct_le_ih(ih)) {
357                         BUG();
358                 }
359                 /* make sure we don't read more bytes than actually exist in
360                  ** the file.  This can happen in odd cases where i_size isn't
361                  ** correct, and when direct item padding results in a few 
362                  ** extra bytes at the end of the direct item
363                  */
364                 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
365                         break;
366                 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
367                         chars =
368                             inode->i_size - (le_ih_k_offset(ih) - 1) -
369                             path.pos_in_item;
370                         done = 1;
371                 } else {
372                         chars = ih_item_len(ih) - path.pos_in_item;
373                 }
374                 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
375
376                 if (done)
377                         break;
378
379                 p += chars;
380
381                 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
382                         // we done, if read direct item is not the last item of
383                         // node FIXME: we could try to check right delimiting key
384                         // to see whether direct item continues in the right
385                         // neighbor or rely on i_size
386                         break;
387
388                 // update key to look for the next piece
389                 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
390                 result = search_for_position_by_key(inode->i_sb, &key, &path);
391                 if (result != POSITION_FOUND)
392                         // i/o error most likely
393                         break;
394                 bh = get_last_bh(&path);
395                 ih = get_ih(&path);
396         } while (1);
397
398         flush_dcache_page(bh_result->b_page);
399         kunmap(bh_result->b_page);
400
401       finished:
402         pathrelse(&path);
403
404         if (result == IO_ERROR)
405                 return -EIO;
406
407         /* this buffer has valid data, but isn't valid for io.  mapping it to
408          * block #0 tells the rest of reiserfs it just has a tail in it
409          */
410         map_bh(bh_result, inode->i_sb, 0);
411         set_buffer_uptodate(bh_result);
412         return 0;
413 }
414
415 // this is called to create file map. So, _get_block_create_0 will not
416 // read direct item
417 static int reiserfs_bmap(struct inode *inode, sector_t block,
418                          struct buffer_head *bh_result, int create)
419 {
420         if (!file_capable(inode, block))
421                 return -EFBIG;
422
423         reiserfs_write_lock(inode->i_sb);
424         /* do not read the direct item */
425         _get_block_create_0(inode, block, bh_result, 0);
426         reiserfs_write_unlock(inode->i_sb);
427         return 0;
428 }
429
430 /* special version of get_block that is only used by grab_tail_page right
431 ** now.  It is sent to block_prepare_write, and when you try to get a
432 ** block past the end of the file (or a block from a hole) it returns
433 ** -ENOENT instead of a valid buffer.  block_prepare_write expects to
434 ** be able to do i/o on the buffers returned, unless an error value
435 ** is also returned.
436 ** 
437 ** So, this allows block_prepare_write to be used for reading a single block
438 ** in a page.  Where it does not produce a valid page for holes, or past the
439 ** end of the file.  This turns out to be exactly what we need for reading
440 ** tails for conversion.
441 **
442 ** The point of the wrapper is forcing a certain value for create, even
443 ** though the VFS layer is calling this function with create==1.  If you 
444 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block, 
445 ** don't use this function.
446 */
447 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
448                                        struct buffer_head *bh_result,
449                                        int create)
450 {
451         return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
452 }
453
454 /* This is special helper for reiserfs_get_block in case we are executing
455    direct_IO request. */
456 static int reiserfs_get_blocks_direct_io(struct inode *inode,
457                                          sector_t iblock,
458                                          struct buffer_head *bh_result,
459                                          int create)
460 {
461         int ret;
462
463         bh_result->b_page = NULL;
464
465         /* We set the b_size before reiserfs_get_block call since it is
466            referenced in convert_tail_for_hole() that may be called from
467            reiserfs_get_block() */
468         bh_result->b_size = (1 << inode->i_blkbits);
469
470         ret = reiserfs_get_block(inode, iblock, bh_result,
471                                  create | GET_BLOCK_NO_DANGLE);
472         if (ret)
473                 goto out;
474
475         /* don't allow direct io onto tail pages */
476         if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
477                 /* make sure future calls to the direct io funcs for this offset
478                  ** in the file fail by unmapping the buffer
479                  */
480                 clear_buffer_mapped(bh_result);
481                 ret = -EINVAL;
482         }
483         /* Possible unpacked tail. Flush the data before pages have
484            disappeared */
485         if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
486                 int err;
487                 lock_kernel();
488                 err = reiserfs_commit_for_inode(inode);
489                 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
490                 unlock_kernel();
491                 if (err < 0)
492                         ret = err;
493         }
494       out:
495         return ret;
496 }
497
498 /*
499 ** helper function for when reiserfs_get_block is called for a hole
500 ** but the file tail is still in a direct item
501 ** bh_result is the buffer head for the hole
502 ** tail_offset is the offset of the start of the tail in the file
503 **
504 ** This calls prepare_write, which will start a new transaction
505 ** you should not be in a transaction, or have any paths held when you
506 ** call this.
507 */
508 static int convert_tail_for_hole(struct inode *inode,
509                                  struct buffer_head *bh_result,
510                                  loff_t tail_offset)
511 {
512         unsigned long index;
513         unsigned long tail_end;
514         unsigned long tail_start;
515         struct page *tail_page;
516         struct page *hole_page = bh_result->b_page;
517         int retval = 0;
518
519         if ((tail_offset & (bh_result->b_size - 1)) != 1)
520                 return -EIO;
521
522         /* always try to read until the end of the block */
523         tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
524         tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
525
526         index = tail_offset >> PAGE_CACHE_SHIFT;
527         /* hole_page can be zero in case of direct_io, we are sure
528            that we cannot get here if we write with O_DIRECT into
529            tail page */
530         if (!hole_page || index != hole_page->index) {
531                 tail_page = grab_cache_page(inode->i_mapping, index);
532                 retval = -ENOMEM;
533                 if (!tail_page) {
534                         goto out;
535                 }
536         } else {
537                 tail_page = hole_page;
538         }
539
540         /* we don't have to make sure the conversion did not happen while
541          ** we were locking the page because anyone that could convert
542          ** must first take i_mutex.
543          **
544          ** We must fix the tail page for writing because it might have buffers
545          ** that are mapped, but have a block number of 0.  This indicates tail
546          ** data that has been read directly into the page, and block_prepare_write
547          ** won't trigger a get_block in this case.
548          */
549         fix_tail_page_for_writing(tail_page);
550         retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
551         if (retval)
552                 goto unlock;
553
554         /* tail conversion might change the data in the page */
555         flush_dcache_page(tail_page);
556
557         retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
558
559       unlock:
560         if (tail_page != hole_page) {
561                 unlock_page(tail_page);
562                 page_cache_release(tail_page);
563         }
564       out:
565         return retval;
566 }
567
568 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
569                                   long block,
570                                   struct inode *inode,
571                                   b_blocknr_t * allocated_block_nr,
572                                   struct path *path, int flags)
573 {
574         BUG_ON(!th->t_trans_id);
575
576 #ifdef REISERFS_PREALLOCATE
577         if (!(flags & GET_BLOCK_NO_IMUX)) {
578                 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
579                                                   path, block);
580         }
581 #endif
582         return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
583                                          block);
584 }
585
586 int reiserfs_get_block(struct inode *inode, sector_t block,
587                        struct buffer_head *bh_result, int create)
588 {
589         int repeat, retval = 0;
590         b_blocknr_t allocated_block_nr = 0;     // b_blocknr_t is (unsigned) 32 bit int
591         INITIALIZE_PATH(path);
592         int pos_in_item;
593         struct cpu_key key;
594         struct buffer_head *bh, *unbh = NULL;
595         struct item_head *ih, tmp_ih;
596         __le32 *item;
597         int done;
598         int fs_gen;
599         struct reiserfs_transaction_handle *th = NULL;
600         /* space reserved in transaction batch: 
601            . 3 balancings in direct->indirect conversion
602            . 1 block involved into reiserfs_update_sd()
603            XXX in practically impossible worst case direct2indirect()
604            can incur (much) more than 3 balancings.
605            quota update for user, group */
606         int jbegin_count =
607             JOURNAL_PER_BALANCE_CNT * 3 + 1 +
608             2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
609         int version;
610         int dangle = 1;
611         loff_t new_offset =
612             (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
613
614         /* bad.... */
615         reiserfs_write_lock(inode->i_sb);
616         version = get_inode_item_key_version(inode);
617
618         if (!file_capable(inode, block)) {
619                 reiserfs_write_unlock(inode->i_sb);
620                 return -EFBIG;
621         }
622
623         /* if !create, we aren't changing the FS, so we don't need to
624          ** log anything, so we don't need to start a transaction
625          */
626         if (!(create & GET_BLOCK_CREATE)) {
627                 int ret;
628                 /* find number of block-th logical block of the file */
629                 ret = _get_block_create_0(inode, block, bh_result,
630                                           create | GET_BLOCK_READ_DIRECT);
631                 reiserfs_write_unlock(inode->i_sb);
632                 return ret;
633         }
634         /*
635          * if we're already in a transaction, make sure to close
636          * any new transactions we start in this func
637          */
638         if ((create & GET_BLOCK_NO_DANGLE) ||
639             reiserfs_transaction_running(inode->i_sb))
640                 dangle = 0;
641
642         /* If file is of such a size, that it might have a tail and tails are enabled
643          ** we should mark it as possibly needing tail packing on close
644          */
645         if ((have_large_tails(inode->i_sb)
646              && inode->i_size < i_block_size(inode) * 4)
647             || (have_small_tails(inode->i_sb)
648                 && inode->i_size < i_block_size(inode)))
649                 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
650
651         /* set the key of the first byte in the 'block'-th block of file */
652         make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
653         if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
654               start_trans:
655                 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
656                 if (!th) {
657                         retval = -ENOMEM;
658                         goto failure;
659                 }
660                 reiserfs_update_inode_transaction(inode);
661         }
662       research:
663
664         retval = search_for_position_by_key(inode->i_sb, &key, &path);
665         if (retval == IO_ERROR) {
666                 retval = -EIO;
667                 goto failure;
668         }
669
670         bh = get_last_bh(&path);
671         ih = get_ih(&path);
672         item = get_item(&path);
673         pos_in_item = path.pos_in_item;
674
675         fs_gen = get_generation(inode->i_sb);
676         copy_item_head(&tmp_ih, ih);
677
678         if (allocation_needed
679             (retval, allocated_block_nr, ih, item, pos_in_item)) {
680                 /* we have to allocate block for the unformatted node */
681                 if (!th) {
682                         pathrelse(&path);
683                         goto start_trans;
684                 }
685
686                 repeat =
687                     _allocate_block(th, block, inode, &allocated_block_nr,
688                                     &path, create);
689
690                 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
691                         /* restart the transaction to give the journal a chance to free
692                          ** some blocks.  releases the path, so we have to go back to
693                          ** research if we succeed on the second try
694                          */
695                         SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
696                         retval = restart_transaction(th, inode, &path);
697                         if (retval)
698                                 goto failure;
699                         repeat =
700                             _allocate_block(th, block, inode,
701                                             &allocated_block_nr, NULL, create);
702
703                         if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
704                                 goto research;
705                         }
706                         if (repeat == QUOTA_EXCEEDED)
707                                 retval = -EDQUOT;
708                         else
709                                 retval = -ENOSPC;
710                         goto failure;
711                 }
712
713                 if (fs_changed(fs_gen, inode->i_sb)
714                     && item_moved(&tmp_ih, &path)) {
715                         goto research;
716                 }
717         }
718
719         if (indirect_item_found(retval, ih)) {
720                 b_blocknr_t unfm_ptr;
721                 /* 'block'-th block is in the file already (there is
722                    corresponding cell in some indirect item). But it may be
723                    zero unformatted node pointer (hole) */
724                 unfm_ptr = get_block_num(item, pos_in_item);
725                 if (unfm_ptr == 0) {
726                         /* use allocated block to plug the hole */
727                         reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
728                         if (fs_changed(fs_gen, inode->i_sb)
729                             && item_moved(&tmp_ih, &path)) {
730                                 reiserfs_restore_prepared_buffer(inode->i_sb,
731                                                                  bh);
732                                 goto research;
733                         }
734                         set_buffer_new(bh_result);
735                         if (buffer_dirty(bh_result)
736                             && reiserfs_data_ordered(inode->i_sb))
737                                 reiserfs_add_ordered_list(inode, bh_result);
738                         put_block_num(item, pos_in_item, allocated_block_nr);
739                         unfm_ptr = allocated_block_nr;
740                         journal_mark_dirty(th, inode->i_sb, bh);
741                         reiserfs_update_sd(th, inode);
742                 }
743                 set_block_dev_mapped(bh_result, unfm_ptr, inode);
744                 pathrelse(&path);
745                 retval = 0;
746                 if (!dangle && th)
747                         retval = reiserfs_end_persistent_transaction(th);
748
749                 reiserfs_write_unlock(inode->i_sb);
750
751                 /* the item was found, so new blocks were not added to the file
752                  ** there is no need to make sure the inode is updated with this 
753                  ** transaction
754                  */
755                 return retval;
756         }
757
758         if (!th) {
759                 pathrelse(&path);
760                 goto start_trans;
761         }
762
763         /* desired position is not found or is in the direct item. We have
764            to append file with holes up to 'block'-th block converting
765            direct items to indirect one if necessary */
766         done = 0;
767         do {
768                 if (is_statdata_le_ih(ih)) {
769                         __le32 unp = 0;
770                         struct cpu_key tmp_key;
771
772                         /* indirect item has to be inserted */
773                         make_le_item_head(&tmp_ih, &key, version, 1,
774                                           TYPE_INDIRECT, UNFM_P_SIZE,
775                                           0 /* free_space */ );
776
777                         if (cpu_key_k_offset(&key) == 1) {
778                                 /* we are going to add 'block'-th block to the file. Use
779                                    allocated block for that */
780                                 unp = cpu_to_le32(allocated_block_nr);
781                                 set_block_dev_mapped(bh_result,
782                                                      allocated_block_nr, inode);
783                                 set_buffer_new(bh_result);
784                                 done = 1;
785                         }
786                         tmp_key = key;  // ;)
787                         set_cpu_key_k_offset(&tmp_key, 1);
788                         PATH_LAST_POSITION(&path)++;
789
790                         retval =
791                             reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
792                                                  inode, (char *)&unp);
793                         if (retval) {
794                                 reiserfs_free_block(th, inode,
795                                                     allocated_block_nr, 1);
796                                 goto failure;   // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
797                         }
798                         //mark_tail_converted (inode);
799                 } else if (is_direct_le_ih(ih)) {
800                         /* direct item has to be converted */
801                         loff_t tail_offset;
802
803                         tail_offset =
804                             ((le_ih_k_offset(ih) -
805                               1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
806                         if (tail_offset == cpu_key_k_offset(&key)) {
807                                 /* direct item we just found fits into block we have
808                                    to map. Convert it into unformatted node: use
809                                    bh_result for the conversion */
810                                 set_block_dev_mapped(bh_result,
811                                                      allocated_block_nr, inode);
812                                 unbh = bh_result;
813                                 done = 1;
814                         } else {
815                                 /* we have to padd file tail stored in direct item(s)
816                                    up to block size and convert it to unformatted
817                                    node. FIXME: this should also get into page cache */
818
819                                 pathrelse(&path);
820                                 /*
821                                  * ugly, but we can only end the transaction if
822                                  * we aren't nested
823                                  */
824                                 BUG_ON(!th->t_refcount);
825                                 if (th->t_refcount == 1) {
826                                         retval =
827                                             reiserfs_end_persistent_transaction
828                                             (th);
829                                         th = NULL;
830                                         if (retval)
831                                                 goto failure;
832                                 }
833
834                                 retval =
835                                     convert_tail_for_hole(inode, bh_result,
836                                                           tail_offset);
837                                 if (retval) {
838                                         if (retval != -ENOSPC)
839                                                 reiserfs_warning(inode->i_sb,
840                                                                  "clm-6004: convert tail failed inode %lu, error %d",
841                                                                  inode->i_ino,
842                                                                  retval);
843                                         if (allocated_block_nr) {
844                                                 /* the bitmap, the super, and the stat data == 3 */
845                                                 if (!th)
846                                                         th = reiserfs_persistent_transaction(inode->i_sb, 3);
847                                                 if (th)
848                                                         reiserfs_free_block(th,
849                                                                             inode,
850                                                                             allocated_block_nr,
851                                                                             1);
852                                         }
853                                         goto failure;
854                                 }
855                                 goto research;
856                         }
857                         retval =
858                             direct2indirect(th, inode, &path, unbh,
859                                             tail_offset);
860                         if (retval) {
861                                 reiserfs_unmap_buffer(unbh);
862                                 reiserfs_free_block(th, inode,
863                                                     allocated_block_nr, 1);
864                                 goto failure;
865                         }
866                         /* it is important the set_buffer_uptodate is done after
867                          ** the direct2indirect.  The buffer might contain valid
868                          ** data newer than the data on disk (read by readpage, changed,
869                          ** and then sent here by writepage).  direct2indirect needs
870                          ** to know if unbh was already up to date, so it can decide
871                          ** if the data in unbh needs to be replaced with data from
872                          ** the disk
873                          */
874                         set_buffer_uptodate(unbh);
875
876                         /* unbh->b_page == NULL in case of DIRECT_IO request, this means
877                            buffer will disappear shortly, so it should not be added to
878                          */
879                         if (unbh->b_page) {
880                                 /* we've converted the tail, so we must
881                                  ** flush unbh before the transaction commits
882                                  */
883                                 reiserfs_add_tail_list(inode, unbh);
884
885                                 /* mark it dirty now to prevent commit_write from adding
886                                  ** this buffer to the inode's dirty buffer list
887                                  */
888                                 /*
889                                  * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
890                                  * It's still atomic, but it sets the page dirty too,
891                                  * which makes it eligible for writeback at any time by the
892                                  * VM (which was also the case with __mark_buffer_dirty())
893                                  */
894                                 mark_buffer_dirty(unbh);
895                         }
896                 } else {
897                         /* append indirect item with holes if needed, when appending
898                            pointer to 'block'-th block use block, which is already
899                            allocated */
900                         struct cpu_key tmp_key;
901                         unp_t unf_single = 0;   // We use this in case we need to allocate only
902                         // one block which is a fastpath
903                         unp_t *un;
904                         __u64 max_to_insert =
905                             MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
906                             UNFM_P_SIZE;
907                         __u64 blocks_needed;
908
909                         RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
910                                "vs-804: invalid position for append");
911                         /* indirect item has to be appended, set up key of that position */
912                         make_cpu_key(&tmp_key, inode,
913                                      le_key_k_offset(version,
914                                                      &(ih->ih_key)) +
915                                      op_bytes_number(ih,
916                                                      inode->i_sb->s_blocksize),
917                                      //pos_in_item * inode->i_sb->s_blocksize,
918                                      TYPE_INDIRECT, 3); // key type is unimportant
919
920                         RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
921                                "green-805: invalid offset");
922                         blocks_needed =
923                             1 +
924                             ((cpu_key_k_offset(&key) -
925                               cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
926                              s_blocksize_bits);
927
928                         if (blocks_needed == 1) {
929                                 un = &unf_single;
930                         } else {
931                                 un = kmalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC);      // We need to avoid scheduling.
932                                 if (!un) {
933                                         un = &unf_single;
934                                         blocks_needed = 1;
935                                         max_to_insert = 0;
936                                 } else
937                                         memset(un, 0,
938                                                UNFM_P_SIZE * min(blocks_needed,
939                                                                  max_to_insert));
940                         }
941                         if (blocks_needed <= max_to_insert) {
942                                 /* we are going to add target block to the file. Use allocated
943                                    block for that */
944                                 un[blocks_needed - 1] =
945                                     cpu_to_le32(allocated_block_nr);
946                                 set_block_dev_mapped(bh_result,
947                                                      allocated_block_nr, inode);
948                                 set_buffer_new(bh_result);
949                                 done = 1;
950                         } else {
951                                 /* paste hole to the indirect item */
952                                 /* If kmalloc failed, max_to_insert becomes zero and it means we
953                                    only have space for one block */
954                                 blocks_needed =
955                                     max_to_insert ? max_to_insert : 1;
956                         }
957                         retval =
958                             reiserfs_paste_into_item(th, &path, &tmp_key, inode,
959                                                      (char *)un,
960                                                      UNFM_P_SIZE *
961                                                      blocks_needed);
962
963                         if (blocks_needed != 1)
964                                 kfree(un);
965
966                         if (retval) {
967                                 reiserfs_free_block(th, inode,
968                                                     allocated_block_nr, 1);
969                                 goto failure;
970                         }
971                         if (!done) {
972                                 /* We need to mark new file size in case this function will be
973                                    interrupted/aborted later on. And we may do this only for
974                                    holes. */
975                                 inode->i_size +=
976                                     inode->i_sb->s_blocksize * blocks_needed;
977                         }
978                 }
979
980                 if (done == 1)
981                         break;
982
983                 /* this loop could log more blocks than we had originally asked
984                  ** for.  So, we have to allow the transaction to end if it is
985                  ** too big or too full.  Update the inode so things are 
986                  ** consistent if we crash before the function returns
987                  **
988                  ** release the path so that anybody waiting on the path before
989                  ** ending their transaction will be able to continue.
990                  */
991                 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
992                         retval = restart_transaction(th, inode, &path);
993                         if (retval)
994                                 goto failure;
995                 }
996                 /* inserting indirect pointers for a hole can take a 
997                  ** long time.  reschedule if needed
998                  */
999                 cond_resched();
1000
1001                 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1002                 if (retval == IO_ERROR) {
1003                         retval = -EIO;
1004                         goto failure;
1005                 }
1006                 if (retval == POSITION_FOUND) {
1007                         reiserfs_warning(inode->i_sb,
1008                                          "vs-825: reiserfs_get_block: "
1009                                          "%K should not be found", &key);
1010                         retval = -EEXIST;
1011                         if (allocated_block_nr)
1012                                 reiserfs_free_block(th, inode,
1013                                                     allocated_block_nr, 1);
1014                         pathrelse(&path);
1015                         goto failure;
1016                 }
1017                 bh = get_last_bh(&path);
1018                 ih = get_ih(&path);
1019                 item = get_item(&path);
1020                 pos_in_item = path.pos_in_item;
1021         } while (1);
1022
1023         retval = 0;
1024
1025       failure:
1026         if (th && (!dangle || (retval && !th->t_trans_id))) {
1027                 int err;
1028                 if (th->t_trans_id)
1029                         reiserfs_update_sd(th, inode);
1030                 err = reiserfs_end_persistent_transaction(th);
1031                 if (err)
1032                         retval = err;
1033         }
1034
1035         reiserfs_write_unlock(inode->i_sb);
1036         reiserfs_check_path(&path);
1037         return retval;
1038 }
1039
1040 static int
1041 reiserfs_readpages(struct file *file, struct address_space *mapping,
1042                    struct list_head *pages, unsigned nr_pages)
1043 {
1044         return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1045 }
1046
1047 /* Compute real number of used bytes by file
1048  * Following three functions can go away when we'll have enough space in stat item
1049  */
1050 static int real_space_diff(struct inode *inode, int sd_size)
1051 {
1052         int bytes;
1053         loff_t blocksize = inode->i_sb->s_blocksize;
1054
1055         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1056                 return sd_size;
1057
1058         /* End of file is also in full block with indirect reference, so round
1059          ** up to the next block.
1060          **
1061          ** there is just no way to know if the tail is actually packed
1062          ** on the file, so we have to assume it isn't.  When we pack the
1063          ** tail, we add 4 bytes to pretend there really is an unformatted
1064          ** node pointer
1065          */
1066         bytes =
1067             ((inode->i_size +
1068               (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1069             sd_size;
1070         return bytes;
1071 }
1072
1073 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1074                                         int sd_size)
1075 {
1076         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1077                 return inode->i_size +
1078                     (loff_t) (real_space_diff(inode, sd_size));
1079         }
1080         return ((loff_t) real_space_diff(inode, sd_size)) +
1081             (((loff_t) blocks) << 9);
1082 }
1083
1084 /* Compute number of blocks used by file in ReiserFS counting */
1085 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1086 {
1087         loff_t bytes = inode_get_bytes(inode);
1088         loff_t real_space = real_space_diff(inode, sd_size);
1089
1090         /* keeps fsck and non-quota versions of reiserfs happy */
1091         if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1092                 bytes += (loff_t) 511;
1093         }
1094
1095         /* files from before the quota patch might i_blocks such that
1096          ** bytes < real_space.  Deal with that here to prevent it from
1097          ** going negative.
1098          */
1099         if (bytes < real_space)
1100                 return 0;
1101         return (bytes - real_space) >> 9;
1102 }
1103
1104 //
1105 // BAD: new directories have stat data of new type and all other items
1106 // of old type. Version stored in the inode says about body items, so
1107 // in update_stat_data we can not rely on inode, but have to check
1108 // item version directly
1109 //
1110
1111 // called by read_locked_inode
1112 static void init_inode(struct inode *inode, struct path *path)
1113 {
1114         struct buffer_head *bh;
1115         struct item_head *ih;
1116         __u32 rdev;
1117         //int version = ITEM_VERSION_1;
1118
1119         bh = PATH_PLAST_BUFFER(path);
1120         ih = PATH_PITEM_HEAD(path);
1121
1122         copy_key(INODE_PKEY(inode), &(ih->ih_key));
1123
1124         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1125         REISERFS_I(inode)->i_flags = 0;
1126         REISERFS_I(inode)->i_prealloc_block = 0;
1127         REISERFS_I(inode)->i_prealloc_count = 0;
1128         REISERFS_I(inode)->i_trans_id = 0;
1129         REISERFS_I(inode)->i_jl = NULL;
1130         reiserfs_init_acl_access(inode);
1131         reiserfs_init_acl_default(inode);
1132         reiserfs_init_xattr_rwsem(inode);
1133
1134         if (stat_data_v1(ih)) {
1135                 struct stat_data_v1 *sd =
1136                     (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1137                 unsigned long blocks;
1138
1139                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1140                 set_inode_sd_version(inode, STAT_DATA_V1);
1141                 inode->i_mode = sd_v1_mode(sd);
1142                 inode->i_nlink = sd_v1_nlink(sd);
1143                 inode->i_uid = sd_v1_uid(sd);
1144                 inode->i_gid = sd_v1_gid(sd);
1145                 inode->i_size = sd_v1_size(sd);
1146                 inode->i_atime.tv_sec = sd_v1_atime(sd);
1147                 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1148                 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1149                 inode->i_atime.tv_nsec = 0;
1150                 inode->i_ctime.tv_nsec = 0;
1151                 inode->i_mtime.tv_nsec = 0;
1152
1153                 inode->i_blocks = sd_v1_blocks(sd);
1154                 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1155                 blocks = (inode->i_size + 511) >> 9;
1156                 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1157                 if (inode->i_blocks > blocks) {
1158                         // there was a bug in <=3.5.23 when i_blocks could take negative
1159                         // values. Starting from 3.5.17 this value could even be stored in
1160                         // stat data. For such files we set i_blocks based on file
1161                         // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1162                         // only updated if file's inode will ever change
1163                         inode->i_blocks = blocks;
1164                 }
1165
1166                 rdev = sd_v1_rdev(sd);
1167                 REISERFS_I(inode)->i_first_direct_byte =
1168                     sd_v1_first_direct_byte(sd);
1169                 /* an early bug in the quota code can give us an odd number for the
1170                  ** block count.  This is incorrect, fix it here.
1171                  */
1172                 if (inode->i_blocks & 1) {
1173                         inode->i_blocks++;
1174                 }
1175                 inode_set_bytes(inode,
1176                                 to_real_used_space(inode, inode->i_blocks,
1177                                                    SD_V1_SIZE));
1178                 /* nopack is initially zero for v1 objects. For v2 objects,
1179                    nopack is initialised from sd_attrs */
1180                 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1181         } else {
1182                 // new stat data found, but object may have old items
1183                 // (directories and symlinks)
1184                 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1185
1186                 inode->i_mode = sd_v2_mode(sd);
1187                 inode->i_nlink = sd_v2_nlink(sd);
1188                 inode->i_uid = sd_v2_uid(sd);
1189                 inode->i_size = sd_v2_size(sd);
1190                 inode->i_gid = sd_v2_gid(sd);
1191                 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1192                 inode->i_atime.tv_sec = sd_v2_atime(sd);
1193                 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1194                 inode->i_ctime.tv_nsec = 0;
1195                 inode->i_mtime.tv_nsec = 0;
1196                 inode->i_atime.tv_nsec = 0;
1197                 inode->i_blocks = sd_v2_blocks(sd);
1198                 rdev = sd_v2_rdev(sd);
1199                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1200                         inode->i_generation =
1201                             le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1202                 else
1203                         inode->i_generation = sd_v2_generation(sd);
1204
1205                 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1206                         set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1207                 else
1208                         set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1209                 REISERFS_I(inode)->i_first_direct_byte = 0;
1210                 set_inode_sd_version(inode, STAT_DATA_V2);
1211                 inode_set_bytes(inode,
1212                                 to_real_used_space(inode, inode->i_blocks,
1213                                                    SD_V2_SIZE));
1214                 /* read persistent inode attributes from sd and initalise
1215                    generic inode flags from them */
1216                 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1217                 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1218         }
1219
1220         pathrelse(path);
1221         if (S_ISREG(inode->i_mode)) {
1222                 inode->i_op = &reiserfs_file_inode_operations;
1223                 inode->i_fop = &reiserfs_file_operations;
1224                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1225         } else if (S_ISDIR(inode->i_mode)) {
1226                 inode->i_op = &reiserfs_dir_inode_operations;
1227                 inode->i_fop = &reiserfs_dir_operations;
1228         } else if (S_ISLNK(inode->i_mode)) {
1229                 inode->i_op = &reiserfs_symlink_inode_operations;
1230                 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1231         } else {
1232                 inode->i_blocks = 0;
1233                 inode->i_op = &reiserfs_special_inode_operations;
1234                 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1235         }
1236 }
1237
1238 // update new stat data with inode fields
1239 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1240 {
1241         struct stat_data *sd_v2 = (struct stat_data *)sd;
1242         __u16 flags;
1243
1244         set_sd_v2_mode(sd_v2, inode->i_mode);
1245         set_sd_v2_nlink(sd_v2, inode->i_nlink);
1246         set_sd_v2_uid(sd_v2, inode->i_uid);
1247         set_sd_v2_size(sd_v2, size);
1248         set_sd_v2_gid(sd_v2, inode->i_gid);
1249         set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1250         set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1251         set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1252         set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1253         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1254                 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1255         else
1256                 set_sd_v2_generation(sd_v2, inode->i_generation);
1257         flags = REISERFS_I(inode)->i_attrs;
1258         i_attrs_to_sd_attrs(inode, &flags);
1259         set_sd_v2_attrs(sd_v2, flags);
1260 }
1261
1262 // used to copy inode's fields to old stat data
1263 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1264 {
1265         struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1266
1267         set_sd_v1_mode(sd_v1, inode->i_mode);
1268         set_sd_v1_uid(sd_v1, inode->i_uid);
1269         set_sd_v1_gid(sd_v1, inode->i_gid);
1270         set_sd_v1_nlink(sd_v1, inode->i_nlink);
1271         set_sd_v1_size(sd_v1, size);
1272         set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1273         set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1274         set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1275
1276         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1277                 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1278         else
1279                 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1280
1281         // Sigh. i_first_direct_byte is back
1282         set_sd_v1_first_direct_byte(sd_v1,
1283                                     REISERFS_I(inode)->i_first_direct_byte);
1284 }
1285
1286 /* NOTE, you must prepare the buffer head before sending it here,
1287 ** and then log it after the call
1288 */
1289 static void update_stat_data(struct path *path, struct inode *inode,
1290                              loff_t size)
1291 {
1292         struct buffer_head *bh;
1293         struct item_head *ih;
1294
1295         bh = PATH_PLAST_BUFFER(path);
1296         ih = PATH_PITEM_HEAD(path);
1297
1298         if (!is_statdata_le_ih(ih))
1299                 reiserfs_panic(inode->i_sb,
1300                                "vs-13065: update_stat_data: key %k, found item %h",
1301                                INODE_PKEY(inode), ih);
1302
1303         if (stat_data_v1(ih)) {
1304                 // path points to old stat data
1305                 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1306         } else {
1307                 inode2sd(B_I_PITEM(bh, ih), inode, size);
1308         }
1309
1310         return;
1311 }
1312
1313 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1314                              struct inode *inode, loff_t size)
1315 {
1316         struct cpu_key key;
1317         INITIALIZE_PATH(path);
1318         struct buffer_head *bh;
1319         int fs_gen;
1320         struct item_head *ih, tmp_ih;
1321         int retval;
1322
1323         BUG_ON(!th->t_trans_id);
1324
1325         make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);        //key type is unimportant
1326
1327         for (;;) {
1328                 int pos;
1329                 /* look for the object's stat data */
1330                 retval = search_item(inode->i_sb, &key, &path);
1331                 if (retval == IO_ERROR) {
1332                         reiserfs_warning(inode->i_sb,
1333                                          "vs-13050: reiserfs_update_sd: "
1334                                          "i/o failure occurred trying to update %K stat data",
1335                                          &key);
1336                         return;
1337                 }
1338                 if (retval == ITEM_NOT_FOUND) {
1339                         pos = PATH_LAST_POSITION(&path);
1340                         pathrelse(&path);
1341                         if (inode->i_nlink == 0) {
1342                                 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1343                                 return;
1344                         }
1345                         reiserfs_warning(inode->i_sb,
1346                                          "vs-13060: reiserfs_update_sd: "
1347                                          "stat data of object %k (nlink == %d) not found (pos %d)",
1348                                          INODE_PKEY(inode), inode->i_nlink,
1349                                          pos);
1350                         reiserfs_check_path(&path);
1351                         return;
1352                 }
1353
1354                 /* sigh, prepare_for_journal might schedule.  When it schedules the
1355                  ** FS might change.  We have to detect that, and loop back to the
1356                  ** search if the stat data item has moved
1357                  */
1358                 bh = get_last_bh(&path);
1359                 ih = get_ih(&path);
1360                 copy_item_head(&tmp_ih, ih);
1361                 fs_gen = get_generation(inode->i_sb);
1362                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1363                 if (fs_changed(fs_gen, inode->i_sb)
1364                     && item_moved(&tmp_ih, &path)) {
1365                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1366                         continue;       /* Stat_data item has been moved after scheduling. */
1367                 }
1368                 break;
1369         }
1370         update_stat_data(&path, inode, size);
1371         journal_mark_dirty(th, th->t_super, bh);
1372         pathrelse(&path);
1373         return;
1374 }
1375
1376 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1377 ** does a make_bad_inode when things go wrong.  But, we need to make sure
1378 ** and clear the key in the private portion of the inode, otherwise a
1379 ** corresponding iput might try to delete whatever object the inode last
1380 ** represented.
1381 */
1382 static void reiserfs_make_bad_inode(struct inode *inode)
1383 {
1384         memset(INODE_PKEY(inode), 0, KEY_SIZE);
1385         make_bad_inode(inode);
1386 }
1387
1388 //
1389 // initially this function was derived from minix or ext2's analog and
1390 // evolved as the prototype did
1391 //
1392
1393 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1394 {
1395         struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1396         inode->i_ino = args->objectid;
1397         INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1398         return 0;
1399 }
1400
1401 /* looks for stat data in the tree, and fills up the fields of in-core
1402    inode stat data fields */
1403 void reiserfs_read_locked_inode(struct inode *inode,
1404                                 struct reiserfs_iget_args *args)
1405 {
1406         INITIALIZE_PATH(path_to_sd);
1407         struct cpu_key key;
1408         unsigned long dirino;
1409         int retval;
1410
1411         dirino = args->dirid;
1412
1413         /* set version 1, version 2 could be used too, because stat data
1414            key is the same in both versions */
1415         key.version = KEY_FORMAT_3_5;
1416         key.on_disk_key.k_dir_id = dirino;
1417         key.on_disk_key.k_objectid = inode->i_ino;
1418         key.on_disk_key.k_offset = 0;
1419         key.on_disk_key.k_type = 0;
1420
1421         /* look for the object's stat data */
1422         retval = search_item(inode->i_sb, &key, &path_to_sd);
1423         if (retval == IO_ERROR) {
1424                 reiserfs_warning(inode->i_sb,
1425                                  "vs-13070: reiserfs_read_locked_inode: "
1426                                  "i/o failure occurred trying to find stat data of %K",
1427                                  &key);
1428                 reiserfs_make_bad_inode(inode);
1429                 return;
1430         }
1431         if (retval != ITEM_FOUND) {
1432                 /* a stale NFS handle can trigger this without it being an error */
1433                 pathrelse(&path_to_sd);
1434                 reiserfs_make_bad_inode(inode);
1435                 inode->i_nlink = 0;
1436                 return;
1437         }
1438
1439         init_inode(inode, &path_to_sd);
1440
1441         /* It is possible that knfsd is trying to access inode of a file
1442            that is being removed from the disk by some other thread. As we
1443            update sd on unlink all that is required is to check for nlink
1444            here. This bug was first found by Sizif when debugging
1445            SquidNG/Butterfly, forgotten, and found again after Philippe
1446            Gramoulle <philippe.gramoulle@mmania.com> reproduced it. 
1447
1448            More logical fix would require changes in fs/inode.c:iput() to
1449            remove inode from hash-table _after_ fs cleaned disk stuff up and
1450            in iget() to return NULL if I_FREEING inode is found in
1451            hash-table. */
1452         /* Currently there is one place where it's ok to meet inode with
1453            nlink==0: processing of open-unlinked and half-truncated files
1454            during mount (fs/reiserfs/super.c:finish_unfinished()). */
1455         if ((inode->i_nlink == 0) &&
1456             !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1457                 reiserfs_warning(inode->i_sb,
1458                                  "vs-13075: reiserfs_read_locked_inode: "
1459                                  "dead inode read from disk %K. "
1460                                  "This is likely to be race with knfsd. Ignore",
1461                                  &key);
1462                 reiserfs_make_bad_inode(inode);
1463         }
1464
1465         reiserfs_check_path(&path_to_sd);       /* init inode should be relsing */
1466
1467 }
1468
1469 /**
1470  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1471  *
1472  * @inode:    inode from hash table to check
1473  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1474  *
1475  * This function is called by iget5_locked() to distinguish reiserfs inodes
1476  * having the same inode numbers. Such inodes can only exist due to some
1477  * error condition. One of them should be bad. Inodes with identical
1478  * inode numbers (objectids) are distinguished by parent directory ids.
1479  *
1480  */
1481 int reiserfs_find_actor(struct inode *inode, void *opaque)
1482 {
1483         struct reiserfs_iget_args *args;
1484
1485         args = opaque;
1486         /* args is already in CPU order */
1487         return (inode->i_ino == args->objectid) &&
1488             (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1489 }
1490
1491 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1492 {
1493         struct inode *inode;
1494         struct reiserfs_iget_args args;
1495
1496         args.objectid = key->on_disk_key.k_objectid;
1497         args.dirid = key->on_disk_key.k_dir_id;
1498         inode = iget5_locked(s, key->on_disk_key.k_objectid,
1499                              reiserfs_find_actor, reiserfs_init_locked_inode,
1500                              (void *)(&args));
1501         if (!inode)
1502                 return ERR_PTR(-ENOMEM);
1503
1504         if (inode->i_state & I_NEW) {
1505                 reiserfs_read_locked_inode(inode, &args);
1506                 unlock_new_inode(inode);
1507         }
1508
1509         if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1510                 /* either due to i/o error or a stale NFS handle */
1511                 iput(inode);
1512                 inode = NULL;
1513         }
1514         return inode;
1515 }
1516
1517 struct dentry *reiserfs_get_dentry(struct super_block *sb, void *vobjp)
1518 {
1519         __u32 *data = vobjp;
1520         struct cpu_key key;
1521         struct dentry *result;
1522         struct inode *inode;
1523
1524         key.on_disk_key.k_objectid = data[0];
1525         key.on_disk_key.k_dir_id = data[1];
1526         reiserfs_write_lock(sb);
1527         inode = reiserfs_iget(sb, &key);
1528         if (inode && !IS_ERR(inode) && data[2] != 0 &&
1529             data[2] != inode->i_generation) {
1530                 iput(inode);
1531                 inode = NULL;
1532         }
1533         reiserfs_write_unlock(sb);
1534         if (!inode)
1535                 inode = ERR_PTR(-ESTALE);
1536         if (IS_ERR(inode))
1537                 return ERR_PTR(PTR_ERR(inode));
1538         result = d_alloc_anon(inode);
1539         if (!result) {
1540                 iput(inode);
1541                 return ERR_PTR(-ENOMEM);
1542         }
1543         return result;
1544 }
1545
1546 struct dentry *reiserfs_decode_fh(struct super_block *sb, __u32 * data,
1547                                   int len, int fhtype,
1548                                   int (*acceptable) (void *contect,
1549                                                      struct dentry * de),
1550                                   void *context)
1551 {
1552         __u32 obj[3], parent[3];
1553
1554         /* fhtype happens to reflect the number of u32s encoded.
1555          * due to a bug in earlier code, fhtype might indicate there
1556          * are more u32s then actually fitted.
1557          * so if fhtype seems to be more than len, reduce fhtype.
1558          * Valid types are:
1559          *   2 - objectid + dir_id - legacy support
1560          *   3 - objectid + dir_id + generation
1561          *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1562          *   5 - objectid + dir_id + generation + objectid and dirid of parent
1563          *   6 - as above plus generation of directory
1564          * 6 does not fit in NFSv2 handles
1565          */
1566         if (fhtype > len) {
1567                 if (fhtype != 6 || len != 5)
1568                         reiserfs_warning(sb,
1569                                          "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1570                                          fhtype, len);
1571                 fhtype = 5;
1572         }
1573
1574         obj[0] = data[0];
1575         obj[1] = data[1];
1576         if (fhtype == 3 || fhtype >= 5)
1577                 obj[2] = data[2];
1578         else
1579                 obj[2] = 0;     /* generation number */
1580
1581         if (fhtype >= 4) {
1582                 parent[0] = data[fhtype >= 5 ? 3 : 2];
1583                 parent[1] = data[fhtype >= 5 ? 4 : 3];
1584                 if (fhtype == 6)
1585                         parent[2] = data[5];
1586                 else
1587                         parent[2] = 0;
1588         }
1589         return sb->s_export_op->find_exported_dentry(sb, obj,
1590                                                      fhtype < 4 ? NULL : parent,
1591                                                      acceptable, context);
1592 }
1593
1594 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1595                        int need_parent)
1596 {
1597         struct inode *inode = dentry->d_inode;
1598         int maxlen = *lenp;
1599
1600         if (maxlen < 3)
1601                 return 255;
1602
1603         data[0] = inode->i_ino;
1604         data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1605         data[2] = inode->i_generation;
1606         *lenp = 3;
1607         /* no room for directory info? return what we've stored so far */
1608         if (maxlen < 5 || !need_parent)
1609                 return 3;
1610
1611         spin_lock(&dentry->d_lock);
1612         inode = dentry->d_parent->d_inode;
1613         data[3] = inode->i_ino;
1614         data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1615         *lenp = 5;
1616         if (maxlen >= 6) {
1617                 data[5] = inode->i_generation;
1618                 *lenp = 6;
1619         }
1620         spin_unlock(&dentry->d_lock);
1621         return *lenp;
1622 }
1623
1624 /* looks for stat data, then copies fields to it, marks the buffer
1625    containing stat data as dirty */
1626 /* reiserfs inodes are never really dirty, since the dirty inode call
1627 ** always logs them.  This call allows the VFS inode marking routines
1628 ** to properly mark inodes for datasync and such, but only actually
1629 ** does something when called for a synchronous update.
1630 */
1631 int reiserfs_write_inode(struct inode *inode, int do_sync)
1632 {
1633         struct reiserfs_transaction_handle th;
1634         int jbegin_count = 1;
1635
1636         if (inode->i_sb->s_flags & MS_RDONLY)
1637                 return -EROFS;
1638         /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1639          ** these cases are just when the system needs ram, not when the 
1640          ** inode needs to reach disk for safety, and they can safely be
1641          ** ignored because the altered inode has already been logged.
1642          */
1643         if (do_sync && !(current->flags & PF_MEMALLOC)) {
1644                 reiserfs_write_lock(inode->i_sb);
1645                 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1646                         reiserfs_update_sd(&th, inode);
1647                         journal_end_sync(&th, inode->i_sb, jbegin_count);
1648                 }
1649                 reiserfs_write_unlock(inode->i_sb);
1650         }
1651         return 0;
1652 }
1653
1654 /* stat data of new object is inserted already, this inserts the item
1655    containing "." and ".." entries */
1656 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1657                                   struct inode *inode,
1658                                   struct item_head *ih, struct path *path,
1659                                   struct inode *dir)
1660 {
1661         struct super_block *sb = th->t_super;
1662         char empty_dir[EMPTY_DIR_SIZE];
1663         char *body = empty_dir;
1664         struct cpu_key key;
1665         int retval;
1666
1667         BUG_ON(!th->t_trans_id);
1668
1669         _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1670                       le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1671                       TYPE_DIRENTRY, 3 /*key length */ );
1672
1673         /* compose item head for new item. Directories consist of items of
1674            old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1675            is done by reiserfs_new_inode */
1676         if (old_format_only(sb)) {
1677                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1678                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1679
1680                 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1681                                        ih->ih_key.k_objectid,
1682                                        INODE_PKEY(dir)->k_dir_id,
1683                                        INODE_PKEY(dir)->k_objectid);
1684         } else {
1685                 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1686                                   TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1687
1688                 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1689                                     ih->ih_key.k_objectid,
1690                                     INODE_PKEY(dir)->k_dir_id,
1691                                     INODE_PKEY(dir)->k_objectid);
1692         }
1693
1694         /* look for place in the tree for new item */
1695         retval = search_item(sb, &key, path);
1696         if (retval == IO_ERROR) {
1697                 reiserfs_warning(sb, "vs-13080: reiserfs_new_directory: "
1698                                  "i/o failure occurred creating new directory");
1699                 return -EIO;
1700         }
1701         if (retval == ITEM_FOUND) {
1702                 pathrelse(path);
1703                 reiserfs_warning(sb, "vs-13070: reiserfs_new_directory: "
1704                                  "object with this key exists (%k)",
1705                                  &(ih->ih_key));
1706                 return -EEXIST;
1707         }
1708
1709         /* insert item, that is empty directory item */
1710         return reiserfs_insert_item(th, path, &key, ih, inode, body);
1711 }
1712
1713 /* stat data of object has been inserted, this inserts the item
1714    containing the body of symlink */
1715 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode,    /* Inode of symlink */
1716                                 struct item_head *ih,
1717                                 struct path *path, const char *symname,
1718                                 int item_len)
1719 {
1720         struct super_block *sb = th->t_super;
1721         struct cpu_key key;
1722         int retval;
1723
1724         BUG_ON(!th->t_trans_id);
1725
1726         _make_cpu_key(&key, KEY_FORMAT_3_5,
1727                       le32_to_cpu(ih->ih_key.k_dir_id),
1728                       le32_to_cpu(ih->ih_key.k_objectid),
1729                       1, TYPE_DIRECT, 3 /*key length */ );
1730
1731         make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1732                           0 /*free_space */ );
1733
1734         /* look for place in the tree for new item */
1735         retval = search_item(sb, &key, path);
1736         if (retval == IO_ERROR) {
1737                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlinik: "
1738                                  "i/o failure occurred creating new symlink");
1739                 return -EIO;
1740         }
1741         if (retval == ITEM_FOUND) {
1742                 pathrelse(path);
1743                 reiserfs_warning(sb, "vs-13080: reiserfs_new_symlink: "
1744                                  "object with this key exists (%k)",
1745                                  &(ih->ih_key));
1746                 return -EEXIST;
1747         }
1748
1749         /* insert item, that is body of symlink */
1750         return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1751 }
1752
1753 /* inserts the stat data into the tree, and then calls
1754    reiserfs_new_directory (to insert ".", ".." item if new object is
1755    directory) or reiserfs_new_symlink (to insert symlink body if new
1756    object is symlink) or nothing (if new object is regular file) 
1757
1758    NOTE! uid and gid must already be set in the inode.  If we return
1759    non-zero due to an error, we have to drop the quota previously allocated
1760    for the fresh inode.  This can only be done outside a transaction, so
1761    if we return non-zero, we also end the transaction.  */
1762 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1763                        struct inode *dir, int mode, const char *symname,
1764                        /* 0 for regular, EMTRY_DIR_SIZE for dirs, 
1765                           strlen (symname) for symlinks) */
1766                        loff_t i_size, struct dentry *dentry,
1767                        struct inode *inode)
1768 {
1769         struct super_block *sb;
1770         INITIALIZE_PATH(path_to_key);
1771         struct cpu_key key;
1772         struct item_head ih;
1773         struct stat_data sd;
1774         int retval;
1775         int err;
1776
1777         BUG_ON(!th->t_trans_id);
1778
1779         if (DQUOT_ALLOC_INODE(inode)) {
1780                 err = -EDQUOT;
1781                 goto out_end_trans;
1782         }
1783         if (!dir || !dir->i_nlink) {
1784                 err = -EPERM;
1785                 goto out_bad_inode;
1786         }
1787
1788         sb = dir->i_sb;
1789
1790         /* item head of new item */
1791         ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1792         ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1793         if (!ih.ih_key.k_objectid) {
1794                 err = -ENOMEM;
1795                 goto out_bad_inode;
1796         }
1797         if (old_format_only(sb))
1798                 /* not a perfect generation count, as object ids can be reused, but 
1799                  ** this is as good as reiserfs can do right now.
1800                  ** note that the private part of inode isn't filled in yet, we have
1801                  ** to use the directory.
1802                  */
1803                 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1804         else
1805 #if defined( USE_INODE_GENERATION_COUNTER )
1806                 inode->i_generation =
1807                     le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1808 #else
1809                 inode->i_generation = ++event;
1810 #endif
1811
1812         /* fill stat data */
1813         inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1814
1815         /* uid and gid must already be set by the caller for quota init */
1816
1817         /* symlink cannot be immutable or append only, right? */
1818         if (S_ISLNK(inode->i_mode))
1819                 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1820
1821         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1822         inode->i_size = i_size;
1823         inode->i_blocks = 0;
1824         inode->i_bytes = 0;
1825         REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1826             U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1827
1828         INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1829         REISERFS_I(inode)->i_flags = 0;
1830         REISERFS_I(inode)->i_prealloc_block = 0;
1831         REISERFS_I(inode)->i_prealloc_count = 0;
1832         REISERFS_I(inode)->i_trans_id = 0;
1833         REISERFS_I(inode)->i_jl = NULL;
1834         REISERFS_I(inode)->i_attrs =
1835             REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1836         sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1837         reiserfs_init_acl_access(inode);
1838         reiserfs_init_acl_default(inode);
1839         reiserfs_init_xattr_rwsem(inode);
1840
1841         if (old_format_only(sb))
1842                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1843                                   TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1844         else
1845                 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1846                                   TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1847
1848         /* key to search for correct place for new stat data */
1849         _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1850                       le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1851                       TYPE_STAT_DATA, 3 /*key length */ );
1852
1853         /* find proper place for inserting of stat data */
1854         retval = search_item(sb, &key, &path_to_key);
1855         if (retval == IO_ERROR) {
1856                 err = -EIO;
1857                 goto out_bad_inode;
1858         }
1859         if (retval == ITEM_FOUND) {
1860                 pathrelse(&path_to_key);
1861                 err = -EEXIST;
1862                 goto out_bad_inode;
1863         }
1864         if (old_format_only(sb)) {
1865                 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1866                         pathrelse(&path_to_key);
1867                         /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1868                         err = -EINVAL;
1869                         goto out_bad_inode;
1870                 }
1871                 inode2sd_v1(&sd, inode, inode->i_size);
1872         } else {
1873                 inode2sd(&sd, inode, inode->i_size);
1874         }
1875         // these do not go to on-disk stat data
1876         inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1877
1878         // store in in-core inode the key of stat data and version all
1879         // object items will have (directory items will have old offset
1880         // format, other new objects will consist of new items)
1881         memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1882         if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1883                 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1884         else
1885                 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1886         if (old_format_only(sb))
1887                 set_inode_sd_version(inode, STAT_DATA_V1);
1888         else
1889                 set_inode_sd_version(inode, STAT_DATA_V2);
1890
1891         /* insert the stat data into the tree */
1892 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1893         if (REISERFS_I(dir)->new_packing_locality)
1894                 th->displace_new_blocks = 1;
1895 #endif
1896         retval =
1897             reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1898                                  (char *)(&sd));
1899         if (retval) {
1900                 err = retval;
1901                 reiserfs_check_path(&path_to_key);
1902                 goto out_bad_inode;
1903         }
1904 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1905         if (!th->displace_new_blocks)
1906                 REISERFS_I(dir)->new_packing_locality = 0;
1907 #endif
1908         if (S_ISDIR(mode)) {
1909                 /* insert item with "." and ".." */
1910                 retval =
1911                     reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1912         }
1913
1914         if (S_ISLNK(mode)) {
1915                 /* insert body of symlink */
1916                 if (!old_format_only(sb))
1917                         i_size = ROUND_UP(i_size);
1918                 retval =
1919                     reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1920                                          i_size);
1921         }
1922         if (retval) {
1923                 err = retval;
1924                 reiserfs_check_path(&path_to_key);
1925                 journal_end(th, th->t_super, th->t_blocks_allocated);
1926                 goto out_inserted_sd;
1927         }
1928
1929         /* XXX CHECK THIS */
1930         if (reiserfs_posixacl(inode->i_sb)) {
1931                 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1932                 if (retval) {
1933                         err = retval;
1934                         reiserfs_check_path(&path_to_key);
1935                         journal_end(th, th->t_super, th->t_blocks_allocated);
1936                         goto out_inserted_sd;
1937                 }
1938         } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1939                 reiserfs_warning(inode->i_sb, "ACLs aren't enabled in the fs, "
1940                                  "but vfs thinks they are!");
1941         } else if (is_reiserfs_priv_object(dir)) {
1942                 reiserfs_mark_inode_private(inode);
1943         }
1944
1945         insert_inode_hash(inode);
1946         reiserfs_update_sd(th, inode);
1947         reiserfs_check_path(&path_to_key);
1948
1949         return 0;
1950
1951 /* it looks like you can easily compress these two goto targets into
1952  * one.  Keeping it like this doesn't actually hurt anything, and they
1953  * are place holders for what the quota code actually needs.
1954  */
1955       out_bad_inode:
1956         /* Invalidate the object, nothing was inserted yet */
1957         INODE_PKEY(inode)->k_objectid = 0;
1958
1959         /* Quota change must be inside a transaction for journaling */
1960         DQUOT_FREE_INODE(inode);
1961
1962       out_end_trans:
1963         journal_end(th, th->t_super, th->t_blocks_allocated);
1964         /* Drop can be outside and it needs more credits so it's better to have it outside */
1965         DQUOT_DROP(inode);
1966         inode->i_flags |= S_NOQUOTA;
1967         make_bad_inode(inode);
1968
1969       out_inserted_sd:
1970         inode->i_nlink = 0;
1971         th->t_trans_id = 0;     /* so the caller can't use this handle later */
1972
1973         /* If we were inheriting an ACL, we need to release the lock so that
1974          * iput doesn't deadlock in reiserfs_delete_xattrs. The locking
1975          * code really needs to be reworked, but this will take care of it
1976          * for now. -jeffm */
1977 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
1978         if (REISERFS_I(dir)->i_acl_default && !IS_ERR(REISERFS_I(dir)->i_acl_default)) {
1979                 reiserfs_write_unlock_xattrs(dir->i_sb);
1980                 iput(inode);
1981                 reiserfs_write_lock_xattrs(dir->i_sb);
1982         } else
1983 #endif
1984                 iput(inode);
1985         return err;
1986 }
1987
1988 /*
1989 ** finds the tail page in the page cache,
1990 ** reads the last block in.
1991 **
1992 ** On success, page_result is set to a locked, pinned page, and bh_result
1993 ** is set to an up to date buffer for the last block in the file.  returns 0.
1994 **
1995 ** tail conversion is not done, so bh_result might not be valid for writing
1996 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1997 ** trying to write the block.
1998 **
1999 ** on failure, nonzero is returned, page_result and bh_result are untouched.
2000 */
2001 static int grab_tail_page(struct inode *p_s_inode,
2002                           struct page **page_result,
2003                           struct buffer_head **bh_result)
2004 {
2005
2006         /* we want the page with the last byte in the file,
2007          ** not the page that will hold the next byte for appending
2008          */
2009         unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2010         unsigned long pos = 0;
2011         unsigned long start = 0;
2012         unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
2013         unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
2014         struct buffer_head *bh;
2015         struct buffer_head *head;
2016         struct page *page;
2017         int error;
2018
2019         /* we know that we are only called with inode->i_size > 0.
2020          ** we also know that a file tail can never be as big as a block
2021          ** If i_size % blocksize == 0, our file is currently block aligned
2022          ** and it won't need converting or zeroing after a truncate.
2023          */
2024         if ((offset & (blocksize - 1)) == 0) {
2025                 return -ENOENT;
2026         }
2027         page = grab_cache_page(p_s_inode->i_mapping, index);
2028         error = -ENOMEM;
2029         if (!page) {
2030                 goto out;
2031         }
2032         /* start within the page of the last block in the file */
2033         start = (offset / blocksize) * blocksize;
2034
2035         error = block_prepare_write(page, start, offset,
2036                                     reiserfs_get_block_create_0);
2037         if (error)
2038                 goto unlock;
2039
2040         head = page_buffers(page);
2041         bh = head;
2042         do {
2043                 if (pos >= start) {
2044                         break;
2045                 }
2046                 bh = bh->b_this_page;
2047                 pos += blocksize;
2048         } while (bh != head);
2049
2050         if (!buffer_uptodate(bh)) {
2051                 /* note, this should never happen, prepare_write should
2052                  ** be taking care of this for us.  If the buffer isn't up to date,
2053                  ** I've screwed up the code to find the buffer, or the code to
2054                  ** call prepare_write
2055                  */
2056                 reiserfs_warning(p_s_inode->i_sb,
2057                                  "clm-6000: error reading block %lu on dev %s",
2058                                  bh->b_blocknr,
2059                                  reiserfs_bdevname(p_s_inode->i_sb));
2060                 error = -EIO;
2061                 goto unlock;
2062         }
2063         *bh_result = bh;
2064         *page_result = page;
2065
2066       out:
2067         return error;
2068
2069       unlock:
2070         unlock_page(page);
2071         page_cache_release(page);
2072         return error;
2073 }
2074
2075 /*
2076 ** vfs version of truncate file.  Must NOT be called with
2077 ** a transaction already started.
2078 **
2079 ** some code taken from block_truncate_page
2080 */
2081 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2082 {
2083         struct reiserfs_transaction_handle th;
2084         /* we want the offset for the first byte after the end of the file */
2085         unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2086         unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2087         unsigned length;
2088         struct page *page = NULL;
2089         int error;
2090         struct buffer_head *bh = NULL;
2091         int err2;
2092
2093         reiserfs_write_lock(p_s_inode->i_sb);
2094
2095         if (p_s_inode->i_size > 0) {
2096                 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2097                         // -ENOENT means we truncated past the end of the file, 
2098                         // and get_block_create_0 could not find a block to read in,
2099                         // which is ok.
2100                         if (error != -ENOENT)
2101                                 reiserfs_warning(p_s_inode->i_sb,
2102                                                  "clm-6001: grab_tail_page failed %d",
2103                                                  error);
2104                         page = NULL;
2105                         bh = NULL;
2106                 }
2107         }
2108
2109         /* so, if page != NULL, we have a buffer head for the offset at 
2110          ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0, 
2111          ** then we have an unformatted node.  Otherwise, we have a direct item, 
2112          ** and no zeroing is required on disk.  We zero after the truncate, 
2113          ** because the truncate might pack the item anyway 
2114          ** (it will unmap bh if it packs).
2115          */
2116         /* it is enough to reserve space in transaction for 2 balancings:
2117            one for "save" link adding and another for the first
2118            cut_from_item. 1 is for update_sd */
2119         error = journal_begin(&th, p_s_inode->i_sb,
2120                               JOURNAL_PER_BALANCE_CNT * 2 + 1);
2121         if (error)
2122                 goto out;
2123         reiserfs_update_inode_transaction(p_s_inode);
2124         if (update_timestamps)
2125                 /* we are doing real truncate: if the system crashes before the last
2126                    transaction of truncating gets committed - on reboot the file
2127                    either appears truncated properly or not truncated at all */
2128                 add_save_link(&th, p_s_inode, 1);
2129         err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2130         error =
2131             journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2132         if (error)
2133                 goto out;
2134
2135         /* check reiserfs_do_truncate after ending the transaction */
2136         if (err2) {
2137                 error = err2;
2138                 goto out;
2139         }
2140         
2141         if (update_timestamps) {
2142                 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2143                 if (error)
2144                         goto out;
2145         }
2146
2147         if (page) {
2148                 length = offset & (blocksize - 1);
2149                 /* if we are not on a block boundary */
2150                 if (length) {
2151                         char *kaddr;
2152
2153                         length = blocksize - length;
2154                         kaddr = kmap_atomic(page, KM_USER0);
2155                         memset(kaddr + offset, 0, length);
2156                         flush_dcache_page(page);
2157                         kunmap_atomic(kaddr, KM_USER0);
2158                         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2159                                 mark_buffer_dirty(bh);
2160                         }
2161                 }
2162                 unlock_page(page);
2163                 page_cache_release(page);
2164         }
2165
2166         reiserfs_write_unlock(p_s_inode->i_sb);
2167         return 0;
2168       out:
2169         if (page) {
2170                 unlock_page(page);
2171                 page_cache_release(page);
2172         }
2173         reiserfs_write_unlock(p_s_inode->i_sb);
2174         return error;
2175 }
2176
2177 static int map_block_for_writepage(struct inode *inode,
2178                                    struct buffer_head *bh_result,
2179                                    unsigned long block)
2180 {
2181         struct reiserfs_transaction_handle th;
2182         int fs_gen;
2183         struct item_head tmp_ih;
2184         struct item_head *ih;
2185         struct buffer_head *bh;
2186         __le32 *item;
2187         struct cpu_key key;
2188         INITIALIZE_PATH(path);
2189         int pos_in_item;
2190         int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2191         loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2192         int retval;
2193         int use_get_block = 0;
2194         int bytes_copied = 0;
2195         int copy_size;
2196         int trans_running = 0;
2197
2198         /* catch places below that try to log something without starting a trans */
2199         th.t_trans_id = 0;
2200
2201         if (!buffer_uptodate(bh_result)) {
2202                 return -EIO;
2203         }
2204
2205         kmap(bh_result->b_page);
2206       start_over:
2207         reiserfs_write_lock(inode->i_sb);
2208         make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2209
2210       research:
2211         retval = search_for_position_by_key(inode->i_sb, &key, &path);
2212         if (retval != POSITION_FOUND) {
2213                 use_get_block = 1;
2214                 goto out;
2215         }
2216
2217         bh = get_last_bh(&path);
2218         ih = get_ih(&path);
2219         item = get_item(&path);
2220         pos_in_item = path.pos_in_item;
2221
2222         /* we've found an unformatted node */
2223         if (indirect_item_found(retval, ih)) {
2224                 if (bytes_copied > 0) {
2225                         reiserfs_warning(inode->i_sb,
2226                                          "clm-6002: bytes_copied %d",
2227                                          bytes_copied);
2228                 }
2229                 if (!get_block_num(item, pos_in_item)) {
2230                         /* crap, we are writing to a hole */
2231                         use_get_block = 1;
2232                         goto out;
2233                 }
2234                 set_block_dev_mapped(bh_result,
2235                                      get_block_num(item, pos_in_item), inode);
2236         } else if (is_direct_le_ih(ih)) {
2237                 char *p;
2238                 p = page_address(bh_result->b_page);
2239                 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2240                 copy_size = ih_item_len(ih) - pos_in_item;
2241
2242                 fs_gen = get_generation(inode->i_sb);
2243                 copy_item_head(&tmp_ih, ih);
2244
2245                 if (!trans_running) {
2246                         /* vs-3050 is gone, no need to drop the path */
2247                         retval = journal_begin(&th, inode->i_sb, jbegin_count);
2248                         if (retval)
2249                                 goto out;
2250                         reiserfs_update_inode_transaction(inode);
2251                         trans_running = 1;
2252                         if (fs_changed(fs_gen, inode->i_sb)
2253                             && item_moved(&tmp_ih, &path)) {
2254                                 reiserfs_restore_prepared_buffer(inode->i_sb,
2255                                                                  bh);
2256                                 goto research;
2257                         }
2258                 }
2259
2260                 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2261
2262                 if (fs_changed(fs_gen, inode->i_sb)
2263                     && item_moved(&tmp_ih, &path)) {
2264                         reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2265                         goto research;
2266                 }
2267
2268                 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2269                        copy_size);
2270
2271                 journal_mark_dirty(&th, inode->i_sb, bh);
2272                 bytes_copied += copy_size;
2273                 set_block_dev_mapped(bh_result, 0, inode);
2274
2275                 /* are there still bytes left? */
2276                 if (bytes_copied < bh_result->b_size &&
2277                     (byte_offset + bytes_copied) < inode->i_size) {
2278                         set_cpu_key_k_offset(&key,
2279                                              cpu_key_k_offset(&key) +
2280                                              copy_size);
2281                         goto research;
2282                 }
2283         } else {
2284                 reiserfs_warning(inode->i_sb,
2285                                  "clm-6003: bad item inode %lu, device %s",
2286                                  inode->i_ino, reiserfs_bdevname(inode->i_sb));
2287                 retval = -EIO;
2288                 goto out;
2289         }
2290         retval = 0;
2291
2292       out:
2293         pathrelse(&path);
2294         if (trans_running) {
2295                 int err = journal_end(&th, inode->i_sb, jbegin_count);
2296                 if (err)
2297                         retval = err;
2298                 trans_running = 0;
2299         }
2300         reiserfs_write_unlock(inode->i_sb);
2301
2302         /* this is where we fill in holes in the file. */
2303         if (use_get_block) {
2304                 retval = reiserfs_get_block(inode, block, bh_result,
2305                                             GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2306                                             | GET_BLOCK_NO_DANGLE);
2307                 if (!retval) {
2308                         if (!buffer_mapped(bh_result)
2309                             || bh_result->b_blocknr == 0) {
2310                                 /* get_block failed to find a mapped unformatted node. */
2311                                 use_get_block = 0;
2312                                 goto start_over;
2313                         }
2314                 }
2315         }
2316         kunmap(bh_result->b_page);
2317
2318         if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2319                 /* we've copied data from the page into the direct item, so the
2320                  * buffer in the page is now clean, mark it to reflect that.
2321                  */
2322                 lock_buffer(bh_result);
2323                 clear_buffer_dirty(bh_result);
2324                 unlock_buffer(bh_result);
2325         }
2326         return retval;
2327 }
2328
2329 /* 
2330  * mason@suse.com: updated in 2.5.54 to follow the same general io 
2331  * start/recovery path as __block_write_full_page, along with special
2332  * code to handle reiserfs tails.
2333  */
2334 static int reiserfs_write_full_page(struct page *page,
2335                                     struct writeback_control *wbc)
2336 {
2337         struct inode *inode = page->mapping->host;
2338         unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2339         int error = 0;
2340         unsigned long block;
2341         sector_t last_block;
2342         struct buffer_head *head, *bh;
2343         int partial = 0;
2344         int nr = 0;
2345         int checked = PageChecked(page);
2346         struct reiserfs_transaction_handle th;
2347         struct super_block *s = inode->i_sb;
2348         int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2349         th.t_trans_id = 0;
2350
2351         /* no logging allowed when nonblocking or from PF_MEMALLOC */
2352         if (checked && (current->flags & PF_MEMALLOC)) {
2353                 redirty_page_for_writepage(wbc, page);
2354                 unlock_page(page);
2355                 return 0;
2356         }
2357
2358         /* The page dirty bit is cleared before writepage is called, which
2359          * means we have to tell create_empty_buffers to make dirty buffers
2360          * The page really should be up to date at this point, so tossing
2361          * in the BH_Uptodate is just a sanity check.
2362          */
2363         if (!page_has_buffers(page)) {
2364                 create_empty_buffers(page, s->s_blocksize,
2365                                      (1 << BH_Dirty) | (1 << BH_Uptodate));
2366         }
2367         head = page_buffers(page);
2368
2369         /* last page in the file, zero out any contents past the
2370          ** last byte in the file
2371          */
2372         if (page->index >= end_index) {
2373                 char *kaddr;
2374                 unsigned last_offset;
2375
2376                 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2377                 /* no file contents in this page */
2378                 if (page->index >= end_index + 1 || !last_offset) {
2379                         unlock_page(page);
2380                         return 0;
2381                 }
2382                 kaddr = kmap_atomic(page, KM_USER0);
2383                 memset(kaddr + last_offset, 0, PAGE_CACHE_SIZE - last_offset);
2384                 flush_dcache_page(page);
2385                 kunmap_atomic(kaddr, KM_USER0);
2386         }
2387         bh = head;
2388         block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2389         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2390         /* first map all the buffers, logging any direct items we find */
2391         do {
2392                 if (block > last_block) {
2393                         /*
2394                          * This can happen when the block size is less than
2395                          * the page size.  The corresponding bytes in the page
2396                          * were zero filled above
2397                          */
2398                         clear_buffer_dirty(bh);
2399                         set_buffer_uptodate(bh);
2400                 } else if ((checked || buffer_dirty(bh)) &&
2401                            (!buffer_mapped(bh) || (buffer_mapped(bh)
2402                                                        && bh->b_blocknr ==
2403                                                        0))) {
2404                         /* not mapped yet, or it points to a direct item, search
2405                          * the btree for the mapping info, and log any direct
2406                          * items found
2407                          */
2408                         if ((error = map_block_for_writepage(inode, bh, block))) {
2409                                 goto fail;
2410                         }
2411                 }
2412                 bh = bh->b_this_page;
2413                 block++;
2414         } while (bh != head);
2415
2416         /*
2417          * we start the transaction after map_block_for_writepage,
2418          * because it can create holes in the file (an unbounded operation).
2419          * starting it here, we can make a reliable estimate for how many
2420          * blocks we're going to log
2421          */
2422         if (checked) {
2423                 ClearPageChecked(page);
2424                 reiserfs_write_lock(s);
2425                 error = journal_begin(&th, s, bh_per_page + 1);
2426                 if (error) {
2427                         reiserfs_write_unlock(s);
2428                         goto fail;
2429                 }
2430                 reiserfs_update_inode_transaction(inode);
2431         }
2432         /* now go through and lock any dirty buffers on the page */
2433         do {
2434                 get_bh(bh);
2435                 if (!buffer_mapped(bh))
2436                         continue;
2437                 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2438                         continue;
2439
2440                 if (checked) {
2441                         reiserfs_prepare_for_journal(s, bh, 1);
2442                         journal_mark_dirty(&th, s, bh);
2443                         continue;
2444                 }
2445                 /* from this point on, we know the buffer is mapped to a
2446                  * real block and not a direct item
2447                  */
2448                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2449                         lock_buffer(bh);
2450                 } else {
2451                         if (test_set_buffer_locked(bh)) {
2452                                 redirty_page_for_writepage(wbc, page);
2453                                 continue;
2454                         }
2455                 }
2456                 if (test_clear_buffer_dirty(bh)) {
2457                         mark_buffer_async_write(bh);
2458                 } else {
2459                         unlock_buffer(bh);
2460                 }
2461         } while ((bh = bh->b_this_page) != head);
2462
2463         if (checked) {
2464                 error = journal_end(&th, s, bh_per_page + 1);
2465                 reiserfs_write_unlock(s);
2466                 if (error)
2467                         goto fail;
2468         }
2469         BUG_ON(PageWriteback(page));
2470         set_page_writeback(page);
2471         unlock_page(page);
2472
2473         /*
2474          * since any buffer might be the only dirty buffer on the page, 
2475          * the first submit_bh can bring the page out of writeback.
2476          * be careful with the buffers.
2477          */
2478         do {
2479                 struct buffer_head *next = bh->b_this_page;
2480                 if (buffer_async_write(bh)) {
2481                         submit_bh(WRITE, bh);
2482                         nr++;
2483                 }
2484                 put_bh(bh);
2485                 bh = next;
2486         } while (bh != head);
2487
2488         error = 0;
2489       done:
2490         if (nr == 0) {
2491                 /*
2492                  * if this page only had a direct item, it is very possible for
2493                  * no io to be required without there being an error.  Or, 
2494                  * someone else could have locked them and sent them down the 
2495                  * pipe without locking the page
2496                  */
2497                 bh = head;
2498                 do {
2499                         if (!buffer_uptodate(bh)) {
2500                                 partial = 1;
2501                                 break;
2502                         }
2503                         bh = bh->b_this_page;
2504                 } while (bh != head);
2505                 if (!partial)
2506                         SetPageUptodate(page);
2507                 end_page_writeback(page);
2508         }
2509         return error;
2510
2511       fail:
2512         /* catches various errors, we need to make sure any valid dirty blocks
2513          * get to the media.  The page is currently locked and not marked for 
2514          * writeback
2515          */
2516         ClearPageUptodate(page);
2517         bh = head;
2518         do {
2519                 get_bh(bh);
2520                 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2521                         lock_buffer(bh);
2522                         mark_buffer_async_write(bh);
2523                 } else {
2524                         /*
2525                          * clear any dirty bits that might have come from getting
2526                          * attached to a dirty page
2527                          */
2528                         clear_buffer_dirty(bh);
2529                 }
2530                 bh = bh->b_this_page;
2531         } while (bh != head);
2532         SetPageError(page);
2533         BUG_ON(PageWriteback(page));
2534         set_page_writeback(page);
2535         unlock_page(page);
2536         do {
2537                 struct buffer_head *next = bh->b_this_page;
2538                 if (buffer_async_write(bh)) {
2539                         clear_buffer_dirty(bh);
2540                         submit_bh(WRITE, bh);
2541                         nr++;
2542                 }
2543                 put_bh(bh);
2544                 bh = next;
2545         } while (bh != head);
2546         goto done;
2547 }
2548
2549 static int reiserfs_readpage(struct file *f, struct page *page)
2550 {
2551         return block_read_full_page(page, reiserfs_get_block);
2552 }
2553
2554 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2555 {
2556         struct inode *inode = page->mapping->host;
2557         reiserfs_wait_on_write_block(inode->i_sb);
2558         return reiserfs_write_full_page(page, wbc);
2559 }
2560
2561 static int reiserfs_prepare_write(struct file *f, struct page *page,
2562                                   unsigned from, unsigned to)
2563 {
2564         struct inode *inode = page->mapping->host;
2565         int ret;
2566         int old_ref = 0;
2567
2568         reiserfs_wait_on_write_block(inode->i_sb);
2569         fix_tail_page_for_writing(page);
2570         if (reiserfs_transaction_running(inode->i_sb)) {
2571                 struct reiserfs_transaction_handle *th;
2572                 th = (struct reiserfs_transaction_handle *)current->
2573                     journal_info;
2574                 BUG_ON(!th->t_refcount);
2575                 BUG_ON(!th->t_trans_id);
2576                 old_ref = th->t_refcount;
2577                 th->t_refcount++;
2578         }
2579
2580         ret = block_prepare_write(page, from, to, reiserfs_get_block);
2581         if (ret && reiserfs_transaction_running(inode->i_sb)) {
2582                 struct reiserfs_transaction_handle *th = current->journal_info;
2583                 /* this gets a little ugly.  If reiserfs_get_block returned an
2584                  * error and left a transacstion running, we've got to close it,
2585                  * and we've got to free handle if it was a persistent transaction.
2586                  *
2587                  * But, if we had nested into an existing transaction, we need
2588                  * to just drop the ref count on the handle.
2589                  *
2590                  * If old_ref == 0, the transaction is from reiserfs_get_block,
2591                  * and it was a persistent trans.  Otherwise, it was nested above.
2592                  */
2593                 if (th->t_refcount > old_ref) {
2594                         if (old_ref)
2595                                 th->t_refcount--;
2596                         else {
2597                                 int err;
2598                                 reiserfs_write_lock(inode->i_sb);
2599                                 err = reiserfs_end_persistent_transaction(th);
2600                                 reiserfs_write_unlock(inode->i_sb);
2601                                 if (err)
2602                                         ret = err;
2603                         }
2604                 }
2605         }
2606         return ret;
2607
2608 }
2609
2610 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2611 {
2612         return generic_block_bmap(as, block, reiserfs_bmap);
2613 }
2614
2615 static int reiserfs_commit_write(struct file *f, struct page *page,
2616                                  unsigned from, unsigned to)
2617 {
2618         struct inode *inode = page->mapping->host;
2619         loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2620         int ret = 0;
2621         int update_sd = 0;
2622         struct reiserfs_transaction_handle *th = NULL;
2623
2624         reiserfs_wait_on_write_block(inode->i_sb);
2625         if (reiserfs_transaction_running(inode->i_sb)) {
2626                 th = current->journal_info;
2627         }
2628         reiserfs_commit_page(inode, page, from, to);
2629
2630         /* generic_commit_write does this for us, but does not update the
2631          ** transaction tracking stuff when the size changes.  So, we have
2632          ** to do the i_size updates here.
2633          */
2634         if (pos > inode->i_size) {
2635                 struct reiserfs_transaction_handle myth;
2636                 reiserfs_write_lock(inode->i_sb);
2637                 /* If the file have grown beyond the border where it
2638                    can have a tail, unmark it as needing a tail
2639                    packing */
2640                 if ((have_large_tails(inode->i_sb)
2641                      && inode->i_size > i_block_size(inode) * 4)
2642                     || (have_small_tails(inode->i_sb)
2643                         && inode->i_size > i_block_size(inode)))
2644                         REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2645
2646                 ret = journal_begin(&myth, inode->i_sb, 1);
2647                 if (ret) {
2648                         reiserfs_write_unlock(inode->i_sb);
2649                         goto journal_error;
2650                 }
2651                 reiserfs_update_inode_transaction(inode);
2652                 inode->i_size = pos;
2653                 /*
2654                  * this will just nest into our transaction.  It's important
2655                  * to use mark_inode_dirty so the inode gets pushed around on the
2656                  * dirty lists, and so that O_SYNC works as expected
2657                  */
2658                 mark_inode_dirty(inode);
2659                 reiserfs_update_sd(&myth, inode);
2660                 update_sd = 1;
2661                 ret = journal_end(&myth, inode->i_sb, 1);
2662                 reiserfs_write_unlock(inode->i_sb);
2663                 if (ret)
2664                         goto journal_error;
2665         }
2666         if (th) {
2667                 reiserfs_write_lock(inode->i_sb);
2668                 if (!update_sd)
2669                         mark_inode_dirty(inode);
2670                 ret = reiserfs_end_persistent_transaction(th);
2671                 reiserfs_write_unlock(inode->i_sb);
2672                 if (ret)
2673                         goto out;
2674         }
2675
2676       out:
2677         return ret;
2678
2679       journal_error:
2680         if (th) {
2681                 reiserfs_write_lock(inode->i_sb);
2682                 if (!update_sd)
2683                         reiserfs_update_sd(th, inode);
2684                 ret = reiserfs_end_persistent_transaction(th);
2685                 reiserfs_write_unlock(inode->i_sb);
2686         }
2687
2688         return ret;
2689 }
2690
2691 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2692 {
2693         if (reiserfs_attrs(inode->i_sb)) {
2694                 if (sd_attrs & REISERFS_SYNC_FL)
2695                         inode->i_flags |= S_SYNC;
2696                 else
2697                         inode->i_flags &= ~S_SYNC;
2698                 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2699                         inode->i_flags |= S_IMMUTABLE;
2700                 else
2701                         inode->i_flags &= ~S_IMMUTABLE;
2702                 if (sd_attrs & REISERFS_APPEND_FL)
2703                         inode->i_flags |= S_APPEND;
2704                 else
2705                         inode->i_flags &= ~S_APPEND;
2706                 if (sd_attrs & REISERFS_NOATIME_FL)
2707                         inode->i_flags |= S_NOATIME;
2708                 else
2709                         inode->i_flags &= ~S_NOATIME;
2710                 if (sd_attrs & REISERFS_NOTAIL_FL)
2711                         REISERFS_I(inode)->i_flags |= i_nopack_mask;
2712                 else
2713                         REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2714         }
2715 }
2716
2717 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2718 {
2719         if (reiserfs_attrs(inode->i_sb)) {
2720                 if (inode->i_flags & S_IMMUTABLE)
2721                         *sd_attrs |= REISERFS_IMMUTABLE_FL;
2722                 else
2723                         *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2724                 if (inode->i_flags & S_SYNC)
2725                         *sd_attrs |= REISERFS_SYNC_FL;
2726                 else
2727                         *sd_attrs &= ~REISERFS_SYNC_FL;
2728                 if (inode->i_flags & S_NOATIME)
2729                         *sd_attrs |= REISERFS_NOATIME_FL;
2730                 else
2731                         *sd_attrs &= ~REISERFS_NOATIME_FL;
2732                 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2733                         *sd_attrs |= REISERFS_NOTAIL_FL;
2734                 else
2735                         *sd_attrs &= ~REISERFS_NOTAIL_FL;
2736         }
2737 }
2738
2739 /* decide if this buffer needs to stay around for data logging or ordered
2740 ** write purposes
2741 */
2742 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2743 {
2744         int ret = 1;
2745         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2746
2747         lock_buffer(bh);
2748         spin_lock(&j->j_dirty_buffers_lock);
2749         if (!buffer_mapped(bh)) {
2750                 goto free_jh;
2751         }
2752         /* the page is locked, and the only places that log a data buffer
2753          * also lock the page.
2754          */
2755         if (reiserfs_file_data_log(inode)) {
2756                 /*
2757                  * very conservative, leave the buffer pinned if
2758                  * anyone might need it.
2759                  */
2760                 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2761                         ret = 0;
2762                 }
2763         } else  if (buffer_dirty(bh)) {
2764                 struct reiserfs_journal_list *jl;
2765                 struct reiserfs_jh *jh = bh->b_private;
2766
2767                 /* why is this safe?
2768                  * reiserfs_setattr updates i_size in the on disk
2769                  * stat data before allowing vmtruncate to be called.
2770                  *
2771                  * If buffer was put onto the ordered list for this
2772                  * transaction, we know for sure either this transaction
2773                  * or an older one already has updated i_size on disk,
2774                  * and this ordered data won't be referenced in the file
2775                  * if we crash.
2776                  *
2777                  * if the buffer was put onto the ordered list for an older
2778                  * transaction, we need to leave it around
2779                  */
2780                 if (jh && (jl = jh->jl)
2781                     && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2782                         ret = 0;
2783         }
2784       free_jh:
2785         if (ret && bh->b_private) {
2786                 reiserfs_free_jh(bh);
2787         }
2788         spin_unlock(&j->j_dirty_buffers_lock);
2789         unlock_buffer(bh);
2790         return ret;
2791 }
2792
2793 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2794 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2795 {
2796         struct buffer_head *head, *bh, *next;
2797         struct inode *inode = page->mapping->host;
2798         unsigned int curr_off = 0;
2799         int ret = 1;
2800
2801         BUG_ON(!PageLocked(page));
2802
2803         if (offset == 0)
2804                 ClearPageChecked(page);
2805
2806         if (!page_has_buffers(page))
2807                 goto out;
2808
2809         head = page_buffers(page);
2810         bh = head;
2811         do {
2812                 unsigned int next_off = curr_off + bh->b_size;
2813                 next = bh->b_this_page;
2814
2815                 /*
2816                  * is this block fully invalidated?
2817                  */
2818                 if (offset <= curr_off) {
2819                         if (invalidatepage_can_drop(inode, bh))
2820                                 reiserfs_unmap_buffer(bh);
2821                         else
2822                                 ret = 0;
2823                 }
2824                 curr_off = next_off;
2825                 bh = next;
2826         } while (bh != head);
2827
2828         /*
2829          * We release buffers only if the entire page is being invalidated.
2830          * The get_block cached value has been unconditionally invalidated,
2831          * so real IO is not possible anymore.
2832          */
2833         if (!offset && ret) {
2834                 ret = try_to_release_page(page, 0);
2835                 /* maybe should BUG_ON(!ret); - neilb */
2836         }
2837       out:
2838         return;
2839 }
2840
2841 static int reiserfs_set_page_dirty(struct page *page)
2842 {
2843         struct inode *inode = page->mapping->host;
2844         if (reiserfs_file_data_log(inode)) {
2845                 SetPageChecked(page);
2846                 return __set_page_dirty_nobuffers(page);
2847         }
2848         return __set_page_dirty_buffers(page);
2849 }
2850
2851 /*
2852  * Returns 1 if the page's buffers were dropped.  The page is locked.
2853  *
2854  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2855  * in the buffers at page_buffers(page).
2856  *
2857  * even in -o notail mode, we can't be sure an old mount without -o notail
2858  * didn't create files with tails.
2859  */
2860 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2861 {
2862         struct inode *inode = page->mapping->host;
2863         struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2864         struct buffer_head *head;
2865         struct buffer_head *bh;
2866         int ret = 1;
2867
2868         WARN_ON(PageChecked(page));
2869         spin_lock(&j->j_dirty_buffers_lock);
2870         head = page_buffers(page);
2871         bh = head;
2872         do {
2873                 if (bh->b_private) {
2874                         if (!buffer_dirty(bh) && !buffer_locked(bh)) {
2875                                 reiserfs_free_jh(bh);
2876                         } else {
2877                                 ret = 0;
2878                                 break;
2879                         }
2880                 }
2881                 bh = bh->b_this_page;
2882         } while (bh != head);
2883         if (ret)
2884                 ret = try_to_free_buffers(page);
2885         spin_unlock(&j->j_dirty_buffers_lock);
2886         return ret;
2887 }
2888
2889 /* We thank Mingming Cao for helping us understand in great detail what
2890    to do in this section of the code. */
2891 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
2892                                   const struct iovec *iov, loff_t offset,
2893                                   unsigned long nr_segs)
2894 {
2895         struct file *file = iocb->ki_filp;
2896         struct inode *inode = file->f_mapping->host;
2897
2898         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2899                                   offset, nr_segs,
2900                                   reiserfs_get_blocks_direct_io, NULL);
2901 }
2902
2903 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
2904 {
2905         struct inode *inode = dentry->d_inode;
2906         int error;
2907         unsigned int ia_valid = attr->ia_valid;
2908         reiserfs_write_lock(inode->i_sb);
2909         if (attr->ia_valid & ATTR_SIZE) {
2910                 /* version 2 items will be caught by the s_maxbytes check
2911                  ** done for us in vmtruncate
2912                  */
2913                 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
2914                     attr->ia_size > MAX_NON_LFS) {
2915                         error = -EFBIG;
2916                         goto out;
2917                 }
2918                 /* fill in hole pointers in the expanding truncate case. */
2919                 if (attr->ia_size > inode->i_size) {
2920                         error = generic_cont_expand(inode, attr->ia_size);
2921                         if (REISERFS_I(inode)->i_prealloc_count > 0) {
2922                                 int err;
2923                                 struct reiserfs_transaction_handle th;
2924                                 /* we're changing at most 2 bitmaps, inode + super */
2925                                 err = journal_begin(&th, inode->i_sb, 4);
2926                                 if (!err) {
2927                                         reiserfs_discard_prealloc(&th, inode);
2928                                         err = journal_end(&th, inode->i_sb, 4);
2929                                 }
2930                                 if (err)
2931                                         error = err;
2932                         }
2933                         if (error)
2934                                 goto out;
2935                         /*
2936                          * file size is changed, ctime and mtime are
2937                          * to be updated
2938                          */
2939                         attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
2940                 }
2941         }
2942
2943         if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
2944              ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
2945             (get_inode_sd_version(inode) == STAT_DATA_V1)) {
2946                 /* stat data of format v3.5 has 16 bit uid and gid */
2947                 error = -EINVAL;
2948                 goto out;
2949         }
2950
2951         error = inode_change_ok(inode, attr);
2952         if (!error) {
2953                 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2954                     (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
2955                         error = reiserfs_chown_xattrs(inode, attr);
2956
2957                         if (!error) {
2958                                 struct reiserfs_transaction_handle th;
2959                                 int jbegin_count =
2960                                     2 *
2961                                     (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
2962                                      REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
2963                                     2;
2964
2965                                 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
2966                                 error =
2967                                     journal_begin(&th, inode->i_sb,
2968                                                   jbegin_count);
2969                                 if (error)
2970                                         goto out;
2971                                 error =
2972                                     DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2973                                 if (error) {
2974                                         journal_end(&th, inode->i_sb,
2975                                                     jbegin_count);
2976                                         goto out;
2977                                 }
2978                                 /* Update corresponding info in inode so that everything is in
2979                                  * one transaction */
2980                                 if (attr->ia_valid & ATTR_UID)
2981                                         inode->i_uid = attr->ia_uid;
2982                                 if (attr->ia_valid & ATTR_GID)
2983                                         inode->i_gid = attr->ia_gid;
2984                                 mark_inode_dirty(inode);
2985                                 error =
2986                                     journal_end(&th, inode->i_sb, jbegin_count);
2987                         }
2988                 }
2989                 if (!error)
2990                         error = inode_setattr(inode, attr);
2991         }
2992
2993         if (!error && reiserfs_posixacl(inode->i_sb)) {
2994                 if (attr->ia_valid & ATTR_MODE)
2995                         error = reiserfs_acl_chmod(inode);
2996         }
2997
2998       out:
2999         reiserfs_write_unlock(inode->i_sb);
3000         return error;
3001 }
3002
3003 const struct address_space_operations reiserfs_address_space_operations = {
3004         .writepage = reiserfs_writepage,
3005         .readpage = reiserfs_readpage,
3006         .readpages = reiserfs_readpages,
3007         .releasepage = reiserfs_releasepage,
3008         .invalidatepage = reiserfs_invalidatepage,
3009         .sync_page = block_sync_page,
3010         .prepare_write = reiserfs_prepare_write,
3011         .commit_write = reiserfs_commit_write,
3012         .bmap = reiserfs_aop_bmap,
3013         .direct_IO = reiserfs_direct_IO,
3014         .set_page_dirty = reiserfs_set_page_dirty,
3015 };