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