xref: /openbmc/linux/fs/reiserfs/inode.c (revision 239480ab)
1 /*
2  * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3  */
4 
5 #include <linux/time.h>
6 #include <linux/fs.h>
7 #include "reiserfs.h"
8 #include "acl.h"
9 #include "xattr.h"
10 #include <linux/exportfs.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/slab.h>
14 #include <linux/uaccess.h>
15 #include <asm/unaligned.h>
16 #include <linux/buffer_head.h>
17 #include <linux/mpage.h>
18 #include <linux/writeback.h>
19 #include <linux/quotaops.h>
20 #include <linux/swap.h>
21 #include <linux/uio.h>
22 #include <linux/bio.h>
23 
24 int reiserfs_commit_write(struct file *f, struct page *page,
25 			  unsigned from, unsigned to);
26 
27 void reiserfs_evict_inode(struct inode *inode)
28 {
29 	/*
30 	 * We need blocks for transaction + (user+group) quota
31 	 * update (possibly delete)
32 	 */
33 	int jbegin_count =
34 	    JOURNAL_PER_BALANCE_CNT * 2 +
35 	    2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
36 	struct reiserfs_transaction_handle th;
37 	int err;
38 
39 	if (!inode->i_nlink && !is_bad_inode(inode))
40 		dquot_initialize(inode);
41 
42 	truncate_inode_pages_final(&inode->i_data);
43 	if (inode->i_nlink)
44 		goto no_delete;
45 
46 	/*
47 	 * The = 0 happens when we abort creating a new inode
48 	 * for some reason like lack of space..
49 	 * also handles bad_inode case
50 	 */
51 	if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {
52 
53 		reiserfs_delete_xattrs(inode);
54 
55 		reiserfs_write_lock(inode->i_sb);
56 
57 		if (journal_begin(&th, inode->i_sb, jbegin_count))
58 			goto out;
59 		reiserfs_update_inode_transaction(inode);
60 
61 		reiserfs_discard_prealloc(&th, inode);
62 
63 		err = reiserfs_delete_object(&th, inode);
64 
65 		/*
66 		 * Do quota update inside a transaction for journaled quotas.
67 		 * We must do that after delete_object so that quota updates
68 		 * go into the same transaction as stat data deletion
69 		 */
70 		if (!err) {
71 			int depth = reiserfs_write_unlock_nested(inode->i_sb);
72 			dquot_free_inode(inode);
73 			reiserfs_write_lock_nested(inode->i_sb, depth);
74 		}
75 
76 		if (journal_end(&th))
77 			goto out;
78 
79 		/*
80 		 * check return value from reiserfs_delete_object after
81 		 * ending the transaction
82 		 */
83 		if (err)
84 		    goto out;
85 
86 		/*
87 		 * all items of file are deleted, so we can remove
88 		 * "save" link
89 		 * we can't do anything about an error here
90 		 */
91 		remove_save_link(inode, 0 /* not truncate */);
92 out:
93 		reiserfs_write_unlock(inode->i_sb);
94 	} else {
95 		/* no object items are in the tree */
96 		;
97 	}
98 
99 	/* note this must go after the journal_end to prevent deadlock */
100 	clear_inode(inode);
101 
102 	dquot_drop(inode);
103 	inode->i_blocks = 0;
104 	return;
105 
106 no_delete:
107 	clear_inode(inode);
108 	dquot_drop(inode);
109 }
110 
111 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
112 			  __u32 objectid, loff_t offset, int type, int length)
113 {
114 	key->version = version;
115 
116 	key->on_disk_key.k_dir_id = dirid;
117 	key->on_disk_key.k_objectid = objectid;
118 	set_cpu_key_k_offset(key, offset);
119 	set_cpu_key_k_type(key, type);
120 	key->key_length = length;
121 }
122 
123 /*
124  * take base of inode_key (it comes from inode always) (dirid, objectid)
125  * and version from an inode, set offset and type of key
126  */
127 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
128 		  int type, int length)
129 {
130 	_make_cpu_key(key, get_inode_item_key_version(inode),
131 		      le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
132 		      le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
133 		      length);
134 }
135 
136 /* when key is 0, do not set version and short key */
137 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
138 			      int version,
139 			      loff_t offset, int type, int length,
140 			      int entry_count /*or ih_free_space */ )
141 {
142 	if (key) {
143 		ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
144 		ih->ih_key.k_objectid =
145 		    cpu_to_le32(key->on_disk_key.k_objectid);
146 	}
147 	put_ih_version(ih, version);
148 	set_le_ih_k_offset(ih, offset);
149 	set_le_ih_k_type(ih, type);
150 	put_ih_item_len(ih, length);
151 	/*    set_ih_free_space (ih, 0); */
152 	/*
153 	 * for directory items it is entry count, for directs and stat
154 	 * datas - 0xffff, for indirects - 0
155 	 */
156 	put_ih_entry_count(ih, entry_count);
157 }
158 
159 /*
160  * FIXME: we might cache recently accessed indirect item
161  * Ugh.  Not too eager for that....
162  * I cut the code until such time as I see a convincing argument (benchmark).
163  * I don't want a bloated inode struct..., and I don't like code complexity....
164  */
165 
166 /*
167  * cutting the code is fine, since it really isn't in use yet and is easy
168  * to add back in.  But, Vladimir has a really good idea here.  Think
169  * about what happens for reading a file.  For each page,
170  * The VFS layer calls reiserfs_readpage, who searches the tree to find
171  * an indirect item.  This indirect item has X number of pointers, where
172  * X is a big number if we've done the block allocation right.  But,
173  * we only use one or two of these pointers during each call to readpage,
174  * needlessly researching again later on.
175  *
176  * The size of the cache could be dynamic based on the size of the file.
177  *
178  * I'd also like to see us cache the location the stat data item, since
179  * we are needlessly researching for that frequently.
180  *
181  * --chris
182  */
183 
184 /*
185  * If this page has a file tail in it, and
186  * it was read in by get_block_create_0, the page data is valid,
187  * but tail is still sitting in a direct item, and we can't write to
188  * it.  So, look through this page, and check all the mapped buffers
189  * to make sure they have valid block numbers.  Any that don't need
190  * to be unmapped, so that __block_write_begin will correctly call
191  * reiserfs_get_block to convert the tail into an unformatted node
192  */
193 static inline void fix_tail_page_for_writing(struct page *page)
194 {
195 	struct buffer_head *head, *next, *bh;
196 
197 	if (page && page_has_buffers(page)) {
198 		head = page_buffers(page);
199 		bh = head;
200 		do {
201 			next = bh->b_this_page;
202 			if (buffer_mapped(bh) && bh->b_blocknr == 0) {
203 				reiserfs_unmap_buffer(bh);
204 			}
205 			bh = next;
206 		} while (bh != head);
207 	}
208 }
209 
210 /*
211  * reiserfs_get_block does not need to allocate a block only if it has been
212  * done already or non-hole position has been found in the indirect item
213  */
214 static inline int allocation_needed(int retval, b_blocknr_t allocated,
215 				    struct item_head *ih,
216 				    __le32 * item, int pos_in_item)
217 {
218 	if (allocated)
219 		return 0;
220 	if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
221 	    get_block_num(item, pos_in_item))
222 		return 0;
223 	return 1;
224 }
225 
226 static inline int indirect_item_found(int retval, struct item_head *ih)
227 {
228 	return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
229 }
230 
231 static inline void set_block_dev_mapped(struct buffer_head *bh,
232 					b_blocknr_t block, struct inode *inode)
233 {
234 	map_bh(bh, inode->i_sb, block);
235 }
236 
237 /*
238  * files which were created in the earlier version can not be longer,
239  * than 2 gb
240  */
241 static int file_capable(struct inode *inode, sector_t block)
242 {
243 	/* it is new file. */
244 	if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 ||
245 	    /* old file, but 'block' is inside of 2gb */
246 	    block < (1 << (31 - inode->i_sb->s_blocksize_bits)))
247 		return 1;
248 
249 	return 0;
250 }
251 
252 static int restart_transaction(struct reiserfs_transaction_handle *th,
253 			       struct inode *inode, struct treepath *path)
254 {
255 	struct super_block *s = th->t_super;
256 	int err;
257 
258 	BUG_ON(!th->t_trans_id);
259 	BUG_ON(!th->t_refcount);
260 
261 	pathrelse(path);
262 
263 	/* we cannot restart while nested */
264 	if (th->t_refcount > 1) {
265 		return 0;
266 	}
267 	reiserfs_update_sd(th, inode);
268 	err = journal_end(th);
269 	if (!err) {
270 		err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
271 		if (!err)
272 			reiserfs_update_inode_transaction(inode);
273 	}
274 	return err;
275 }
276 
277 /*
278  * it is called by get_block when create == 0. Returns block number
279  * for 'block'-th logical block of file. When it hits direct item it
280  * returns 0 (being called from bmap) or read direct item into piece
281  * of page (bh_result)
282  * Please improve the english/clarity in the comment above, as it is
283  * hard to understand.
284  */
285 static int _get_block_create_0(struct inode *inode, sector_t block,
286 			       struct buffer_head *bh_result, int args)
287 {
288 	INITIALIZE_PATH(path);
289 	struct cpu_key key;
290 	struct buffer_head *bh;
291 	struct item_head *ih, tmp_ih;
292 	b_blocknr_t blocknr;
293 	char *p = NULL;
294 	int chars;
295 	int ret;
296 	int result;
297 	int done = 0;
298 	unsigned long offset;
299 
300 	/* prepare the key to look for the 'block'-th block of file */
301 	make_cpu_key(&key, inode,
302 		     (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
303 		     3);
304 
305 	result = search_for_position_by_key(inode->i_sb, &key, &path);
306 	if (result != POSITION_FOUND) {
307 		pathrelse(&path);
308 		if (p)
309 			kunmap(bh_result->b_page);
310 		if (result == IO_ERROR)
311 			return -EIO;
312 		/*
313 		 * We do not return -ENOENT if there is a hole but page is
314 		 * uptodate, because it means that there is some MMAPED data
315 		 * associated with it that is yet to be written to disk.
316 		 */
317 		if ((args & GET_BLOCK_NO_HOLE)
318 		    && !PageUptodate(bh_result->b_page)) {
319 			return -ENOENT;
320 		}
321 		return 0;
322 	}
323 
324 	bh = get_last_bh(&path);
325 	ih = tp_item_head(&path);
326 	if (is_indirect_le_ih(ih)) {
327 		__le32 *ind_item = (__le32 *) ih_item_body(bh, ih);
328 
329 		/*
330 		 * FIXME: here we could cache indirect item or part of it in
331 		 * the inode to avoid search_by_key in case of subsequent
332 		 * access to file
333 		 */
334 		blocknr = get_block_num(ind_item, path.pos_in_item);
335 		ret = 0;
336 		if (blocknr) {
337 			map_bh(bh_result, inode->i_sb, blocknr);
338 			if (path.pos_in_item ==
339 			    ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
340 				set_buffer_boundary(bh_result);
341 			}
342 		} else
343 			/*
344 			 * We do not return -ENOENT if there is a hole but
345 			 * page is uptodate, because it means that there is
346 			 * some MMAPED data associated with it that is
347 			 * yet to be written to disk.
348 			 */
349 		if ((args & GET_BLOCK_NO_HOLE)
350 			    && !PageUptodate(bh_result->b_page)) {
351 			ret = -ENOENT;
352 		}
353 
354 		pathrelse(&path);
355 		if (p)
356 			kunmap(bh_result->b_page);
357 		return ret;
358 	}
359 	/* requested data are in direct item(s) */
360 	if (!(args & GET_BLOCK_READ_DIRECT)) {
361 		/*
362 		 * we are called by bmap. FIXME: we can not map block of file
363 		 * when it is stored in direct item(s)
364 		 */
365 		pathrelse(&path);
366 		if (p)
367 			kunmap(bh_result->b_page);
368 		return -ENOENT;
369 	}
370 
371 	/*
372 	 * if we've got a direct item, and the buffer or page was uptodate,
373 	 * we don't want to pull data off disk again.  skip to the
374 	 * end, where we map the buffer and return
375 	 */
376 	if (buffer_uptodate(bh_result)) {
377 		goto finished;
378 	} else
379 		/*
380 		 * grab_tail_page can trigger calls to reiserfs_get_block on
381 		 * up to date pages without any buffers.  If the page is up
382 		 * to date, we don't want read old data off disk.  Set the up
383 		 * to date bit on the buffer instead and jump to the end
384 		 */
385 	if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
386 		set_buffer_uptodate(bh_result);
387 		goto finished;
388 	}
389 	/* read file tail into part of page */
390 	offset = (cpu_key_k_offset(&key) - 1) & (PAGE_SIZE - 1);
391 	copy_item_head(&tmp_ih, ih);
392 
393 	/*
394 	 * we only want to kmap if we are reading the tail into the page.
395 	 * this is not the common case, so we don't kmap until we are
396 	 * sure we need to.  But, this means the item might move if
397 	 * kmap schedules
398 	 */
399 	if (!p)
400 		p = (char *)kmap(bh_result->b_page);
401 
402 	p += offset;
403 	memset(p, 0, inode->i_sb->s_blocksize);
404 	do {
405 		if (!is_direct_le_ih(ih)) {
406 			BUG();
407 		}
408 		/*
409 		 * make sure we don't read more bytes than actually exist in
410 		 * the file.  This can happen in odd cases where i_size isn't
411 		 * correct, and when direct item padding results in a few
412 		 * extra bytes at the end of the direct item
413 		 */
414 		if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
415 			break;
416 		if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
417 			chars =
418 			    inode->i_size - (le_ih_k_offset(ih) - 1) -
419 			    path.pos_in_item;
420 			done = 1;
421 		} else {
422 			chars = ih_item_len(ih) - path.pos_in_item;
423 		}
424 		memcpy(p, ih_item_body(bh, ih) + path.pos_in_item, chars);
425 
426 		if (done)
427 			break;
428 
429 		p += chars;
430 
431 		/*
432 		 * we done, if read direct item is not the last item of
433 		 * node FIXME: we could try to check right delimiting key
434 		 * to see whether direct item continues in the right
435 		 * neighbor or rely on i_size
436 		 */
437 		if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
438 			break;
439 
440 		/* update key to look for the next piece */
441 		set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
442 		result = search_for_position_by_key(inode->i_sb, &key, &path);
443 		if (result != POSITION_FOUND)
444 			/* i/o error most likely */
445 			break;
446 		bh = get_last_bh(&path);
447 		ih = tp_item_head(&path);
448 	} while (1);
449 
450 	flush_dcache_page(bh_result->b_page);
451 	kunmap(bh_result->b_page);
452 
453 finished:
454 	pathrelse(&path);
455 
456 	if (result == IO_ERROR)
457 		return -EIO;
458 
459 	/*
460 	 * this buffer has valid data, but isn't valid for io.  mapping it to
461 	 * block #0 tells the rest of reiserfs it just has a tail in it
462 	 */
463 	map_bh(bh_result, inode->i_sb, 0);
464 	set_buffer_uptodate(bh_result);
465 	return 0;
466 }
467 
468 /*
469  * this is called to create file map. So, _get_block_create_0 will not
470  * read direct item
471  */
472 static int reiserfs_bmap(struct inode *inode, sector_t block,
473 			 struct buffer_head *bh_result, int create)
474 {
475 	if (!file_capable(inode, block))
476 		return -EFBIG;
477 
478 	reiserfs_write_lock(inode->i_sb);
479 	/* do not read the direct item */
480 	_get_block_create_0(inode, block, bh_result, 0);
481 	reiserfs_write_unlock(inode->i_sb);
482 	return 0;
483 }
484 
485 /*
486  * special version of get_block that is only used by grab_tail_page right
487  * now.  It is sent to __block_write_begin, and when you try to get a
488  * block past the end of the file (or a block from a hole) it returns
489  * -ENOENT instead of a valid buffer.  __block_write_begin expects to
490  * be able to do i/o on the buffers returned, unless an error value
491  * is also returned.
492  *
493  * So, this allows __block_write_begin to be used for reading a single block
494  * in a page.  Where it does not produce a valid page for holes, or past the
495  * end of the file.  This turns out to be exactly what we need for reading
496  * tails for conversion.
497  *
498  * The point of the wrapper is forcing a certain value for create, even
499  * though the VFS layer is calling this function with create==1.  If you
500  * don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
501  * don't use this function.
502 */
503 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
504 				       struct buffer_head *bh_result,
505 				       int create)
506 {
507 	return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
508 }
509 
510 /*
511  * This is special helper for reiserfs_get_block in case we are executing
512  * direct_IO request.
513  */
514 static int reiserfs_get_blocks_direct_io(struct inode *inode,
515 					 sector_t iblock,
516 					 struct buffer_head *bh_result,
517 					 int create)
518 {
519 	int ret;
520 
521 	bh_result->b_page = NULL;
522 
523 	/*
524 	 * We set the b_size before reiserfs_get_block call since it is
525 	 * referenced in convert_tail_for_hole() that may be called from
526 	 * reiserfs_get_block()
527 	 */
528 	bh_result->b_size = i_blocksize(inode);
529 
530 	ret = reiserfs_get_block(inode, iblock, bh_result,
531 				 create | GET_BLOCK_NO_DANGLE);
532 	if (ret)
533 		goto out;
534 
535 	/* don't allow direct io onto tail pages */
536 	if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
537 		/*
538 		 * make sure future calls to the direct io funcs for this
539 		 * offset in the file fail by unmapping the buffer
540 		 */
541 		clear_buffer_mapped(bh_result);
542 		ret = -EINVAL;
543 	}
544 
545 	/*
546 	 * Possible unpacked tail. Flush the data before pages have
547 	 * disappeared
548 	 */
549 	if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
550 		int err;
551 
552 		reiserfs_write_lock(inode->i_sb);
553 
554 		err = reiserfs_commit_for_inode(inode);
555 		REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
556 
557 		reiserfs_write_unlock(inode->i_sb);
558 
559 		if (err < 0)
560 			ret = err;
561 	}
562 out:
563 	return ret;
564 }
565 
566 /*
567  * helper function for when reiserfs_get_block is called for a hole
568  * but the file tail is still in a direct item
569  * bh_result is the buffer head for the hole
570  * tail_offset is the offset of the start of the tail in the file
571  *
572  * This calls prepare_write, which will start a new transaction
573  * you should not be in a transaction, or have any paths held when you
574  * call this.
575  */
576 static int convert_tail_for_hole(struct inode *inode,
577 				 struct buffer_head *bh_result,
578 				 loff_t tail_offset)
579 {
580 	unsigned long index;
581 	unsigned long tail_end;
582 	unsigned long tail_start;
583 	struct page *tail_page;
584 	struct page *hole_page = bh_result->b_page;
585 	int retval = 0;
586 
587 	if ((tail_offset & (bh_result->b_size - 1)) != 1)
588 		return -EIO;
589 
590 	/* always try to read until the end of the block */
591 	tail_start = tail_offset & (PAGE_SIZE - 1);
592 	tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
593 
594 	index = tail_offset >> PAGE_SHIFT;
595 	/*
596 	 * hole_page can be zero in case of direct_io, we are sure
597 	 * that we cannot get here if we write with O_DIRECT into tail page
598 	 */
599 	if (!hole_page || index != hole_page->index) {
600 		tail_page = grab_cache_page(inode->i_mapping, index);
601 		retval = -ENOMEM;
602 		if (!tail_page) {
603 			goto out;
604 		}
605 	} else {
606 		tail_page = hole_page;
607 	}
608 
609 	/*
610 	 * we don't have to make sure the conversion did not happen while
611 	 * we were locking the page because anyone that could convert
612 	 * must first take i_mutex.
613 	 *
614 	 * We must fix the tail page for writing because it might have buffers
615 	 * that are mapped, but have a block number of 0.  This indicates tail
616 	 * data that has been read directly into the page, and
617 	 * __block_write_begin won't trigger a get_block in this case.
618 	 */
619 	fix_tail_page_for_writing(tail_page);
620 	retval = __reiserfs_write_begin(tail_page, tail_start,
621 				      tail_end - tail_start);
622 	if (retval)
623 		goto unlock;
624 
625 	/* tail conversion might change the data in the page */
626 	flush_dcache_page(tail_page);
627 
628 	retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
629 
630 unlock:
631 	if (tail_page != hole_page) {
632 		unlock_page(tail_page);
633 		put_page(tail_page);
634 	}
635 out:
636 	return retval;
637 }
638 
639 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
640 				  sector_t block,
641 				  struct inode *inode,
642 				  b_blocknr_t * allocated_block_nr,
643 				  struct treepath *path, int flags)
644 {
645 	BUG_ON(!th->t_trans_id);
646 
647 #ifdef REISERFS_PREALLOCATE
648 	if (!(flags & GET_BLOCK_NO_IMUX)) {
649 		return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
650 						  path, block);
651 	}
652 #endif
653 	return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
654 					 block);
655 }
656 
657 int reiserfs_get_block(struct inode *inode, sector_t block,
658 		       struct buffer_head *bh_result, int create)
659 {
660 	int repeat, retval = 0;
661 	/* b_blocknr_t is (unsigned) 32 bit int*/
662 	b_blocknr_t allocated_block_nr = 0;
663 	INITIALIZE_PATH(path);
664 	int pos_in_item;
665 	struct cpu_key key;
666 	struct buffer_head *bh, *unbh = NULL;
667 	struct item_head *ih, tmp_ih;
668 	__le32 *item;
669 	int done;
670 	int fs_gen;
671 	struct reiserfs_transaction_handle *th = NULL;
672 	/*
673 	 * space reserved in transaction batch:
674 	 * . 3 balancings in direct->indirect conversion
675 	 * . 1 block involved into reiserfs_update_sd()
676 	 * XXX in practically impossible worst case direct2indirect()
677 	 * can incur (much) more than 3 balancings.
678 	 * quota update for user, group
679 	 */
680 	int jbegin_count =
681 	    JOURNAL_PER_BALANCE_CNT * 3 + 1 +
682 	    2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
683 	int version;
684 	int dangle = 1;
685 	loff_t new_offset =
686 	    (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
687 
688 	reiserfs_write_lock(inode->i_sb);
689 	version = get_inode_item_key_version(inode);
690 
691 	if (!file_capable(inode, block)) {
692 		reiserfs_write_unlock(inode->i_sb);
693 		return -EFBIG;
694 	}
695 
696 	/*
697 	 * if !create, we aren't changing the FS, so we don't need to
698 	 * log anything, so we don't need to start a transaction
699 	 */
700 	if (!(create & GET_BLOCK_CREATE)) {
701 		int ret;
702 		/* find number of block-th logical block of the file */
703 		ret = _get_block_create_0(inode, block, bh_result,
704 					  create | GET_BLOCK_READ_DIRECT);
705 		reiserfs_write_unlock(inode->i_sb);
706 		return ret;
707 	}
708 
709 	/*
710 	 * if we're already in a transaction, make sure to close
711 	 * any new transactions we start in this func
712 	 */
713 	if ((create & GET_BLOCK_NO_DANGLE) ||
714 	    reiserfs_transaction_running(inode->i_sb))
715 		dangle = 0;
716 
717 	/*
718 	 * If file is of such a size, that it might have a tail and
719 	 * tails are enabled  we should mark it as possibly needing
720 	 * tail packing on close
721 	 */
722 	if ((have_large_tails(inode->i_sb)
723 	     && inode->i_size < i_block_size(inode) * 4)
724 	    || (have_small_tails(inode->i_sb)
725 		&& inode->i_size < i_block_size(inode)))
726 		REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
727 
728 	/* set the key of the first byte in the 'block'-th block of file */
729 	make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
730 	if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
731 start_trans:
732 		th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
733 		if (!th) {
734 			retval = -ENOMEM;
735 			goto failure;
736 		}
737 		reiserfs_update_inode_transaction(inode);
738 	}
739 research:
740 
741 	retval = search_for_position_by_key(inode->i_sb, &key, &path);
742 	if (retval == IO_ERROR) {
743 		retval = -EIO;
744 		goto failure;
745 	}
746 
747 	bh = get_last_bh(&path);
748 	ih = tp_item_head(&path);
749 	item = tp_item_body(&path);
750 	pos_in_item = path.pos_in_item;
751 
752 	fs_gen = get_generation(inode->i_sb);
753 	copy_item_head(&tmp_ih, ih);
754 
755 	if (allocation_needed
756 	    (retval, allocated_block_nr, ih, item, pos_in_item)) {
757 		/* we have to allocate block for the unformatted node */
758 		if (!th) {
759 			pathrelse(&path);
760 			goto start_trans;
761 		}
762 
763 		repeat =
764 		    _allocate_block(th, block, inode, &allocated_block_nr,
765 				    &path, create);
766 
767 		/*
768 		 * restart the transaction to give the journal a chance to free
769 		 * some blocks.  releases the path, so we have to go back to
770 		 * research if we succeed on the second try
771 		 */
772 		if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
773 			SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
774 			retval = restart_transaction(th, inode, &path);
775 			if (retval)
776 				goto failure;
777 			repeat =
778 			    _allocate_block(th, block, inode,
779 					    &allocated_block_nr, NULL, create);
780 
781 			if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
782 				goto research;
783 			}
784 			if (repeat == QUOTA_EXCEEDED)
785 				retval = -EDQUOT;
786 			else
787 				retval = -ENOSPC;
788 			goto failure;
789 		}
790 
791 		if (fs_changed(fs_gen, inode->i_sb)
792 		    && item_moved(&tmp_ih, &path)) {
793 			goto research;
794 		}
795 	}
796 
797 	if (indirect_item_found(retval, ih)) {
798 		b_blocknr_t unfm_ptr;
799 		/*
800 		 * 'block'-th block is in the file already (there is
801 		 * corresponding cell in some indirect item). But it may be
802 		 * zero unformatted node pointer (hole)
803 		 */
804 		unfm_ptr = get_block_num(item, pos_in_item);
805 		if (unfm_ptr == 0) {
806 			/* use allocated block to plug the hole */
807 			reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
808 			if (fs_changed(fs_gen, inode->i_sb)
809 			    && item_moved(&tmp_ih, &path)) {
810 				reiserfs_restore_prepared_buffer(inode->i_sb,
811 								 bh);
812 				goto research;
813 			}
814 			set_buffer_new(bh_result);
815 			if (buffer_dirty(bh_result)
816 			    && reiserfs_data_ordered(inode->i_sb))
817 				reiserfs_add_ordered_list(inode, bh_result);
818 			put_block_num(item, pos_in_item, allocated_block_nr);
819 			unfm_ptr = allocated_block_nr;
820 			journal_mark_dirty(th, bh);
821 			reiserfs_update_sd(th, inode);
822 		}
823 		set_block_dev_mapped(bh_result, unfm_ptr, inode);
824 		pathrelse(&path);
825 		retval = 0;
826 		if (!dangle && th)
827 			retval = reiserfs_end_persistent_transaction(th);
828 
829 		reiserfs_write_unlock(inode->i_sb);
830 
831 		/*
832 		 * the item was found, so new blocks were not added to the file
833 		 * there is no need to make sure the inode is updated with this
834 		 * transaction
835 		 */
836 		return retval;
837 	}
838 
839 	if (!th) {
840 		pathrelse(&path);
841 		goto start_trans;
842 	}
843 
844 	/*
845 	 * desired position is not found or is in the direct item. We have
846 	 * to append file with holes up to 'block'-th block converting
847 	 * direct items to indirect one if necessary
848 	 */
849 	done = 0;
850 	do {
851 		if (is_statdata_le_ih(ih)) {
852 			__le32 unp = 0;
853 			struct cpu_key tmp_key;
854 
855 			/* indirect item has to be inserted */
856 			make_le_item_head(&tmp_ih, &key, version, 1,
857 					  TYPE_INDIRECT, UNFM_P_SIZE,
858 					  0 /* free_space */ );
859 
860 			/*
861 			 * we are going to add 'block'-th block to the file.
862 			 * Use allocated block for that
863 			 */
864 			if (cpu_key_k_offset(&key) == 1) {
865 				unp = cpu_to_le32(allocated_block_nr);
866 				set_block_dev_mapped(bh_result,
867 						     allocated_block_nr, inode);
868 				set_buffer_new(bh_result);
869 				done = 1;
870 			}
871 			tmp_key = key;	/* ;) */
872 			set_cpu_key_k_offset(&tmp_key, 1);
873 			PATH_LAST_POSITION(&path)++;
874 
875 			retval =
876 			    reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
877 						 inode, (char *)&unp);
878 			if (retval) {
879 				reiserfs_free_block(th, inode,
880 						    allocated_block_nr, 1);
881 				/*
882 				 * retval == -ENOSPC, -EDQUOT or -EIO
883 				 * or -EEXIST
884 				 */
885 				goto failure;
886 			}
887 		} else if (is_direct_le_ih(ih)) {
888 			/* direct item has to be converted */
889 			loff_t tail_offset;
890 
891 			tail_offset =
892 			    ((le_ih_k_offset(ih) -
893 			      1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
894 
895 			/*
896 			 * direct item we just found fits into block we have
897 			 * to map. Convert it into unformatted node: use
898 			 * bh_result for the conversion
899 			 */
900 			if (tail_offset == cpu_key_k_offset(&key)) {
901 				set_block_dev_mapped(bh_result,
902 						     allocated_block_nr, inode);
903 				unbh = bh_result;
904 				done = 1;
905 			} else {
906 				/*
907 				 * we have to pad file tail stored in direct
908 				 * item(s) up to block size and convert it
909 				 * to unformatted node. FIXME: this should
910 				 * also get into page cache
911 				 */
912 
913 				pathrelse(&path);
914 				/*
915 				 * ugly, but we can only end the transaction if
916 				 * we aren't nested
917 				 */
918 				BUG_ON(!th->t_refcount);
919 				if (th->t_refcount == 1) {
920 					retval =
921 					    reiserfs_end_persistent_transaction
922 					    (th);
923 					th = NULL;
924 					if (retval)
925 						goto failure;
926 				}
927 
928 				retval =
929 				    convert_tail_for_hole(inode, bh_result,
930 							  tail_offset);
931 				if (retval) {
932 					if (retval != -ENOSPC)
933 						reiserfs_error(inode->i_sb,
934 							"clm-6004",
935 							"convert tail failed "
936 							"inode %lu, error %d",
937 							inode->i_ino,
938 							retval);
939 					if (allocated_block_nr) {
940 						/*
941 						 * the bitmap, the super,
942 						 * and the stat data == 3
943 						 */
944 						if (!th)
945 							th = reiserfs_persistent_transaction(inode->i_sb, 3);
946 						if (th)
947 							reiserfs_free_block(th,
948 									    inode,
949 									    allocated_block_nr,
950 									    1);
951 					}
952 					goto failure;
953 				}
954 				goto research;
955 			}
956 			retval =
957 			    direct2indirect(th, inode, &path, unbh,
958 					    tail_offset);
959 			if (retval) {
960 				reiserfs_unmap_buffer(unbh);
961 				reiserfs_free_block(th, inode,
962 						    allocated_block_nr, 1);
963 				goto failure;
964 			}
965 			/*
966 			 * it is important the set_buffer_uptodate is done
967 			 * after the direct2indirect.  The buffer might
968 			 * contain valid data newer than the data on disk
969 			 * (read by readpage, changed, and then sent here by
970 			 * writepage).  direct2indirect needs to know if unbh
971 			 * was already up to date, so it can decide if the
972 			 * data in unbh needs to be replaced with data from
973 			 * the disk
974 			 */
975 			set_buffer_uptodate(unbh);
976 
977 			/*
978 			 * unbh->b_page == NULL in case of DIRECT_IO request,
979 			 * this means buffer will disappear shortly, so it
980 			 * should not be added to
981 			 */
982 			if (unbh->b_page) {
983 				/*
984 				 * we've converted the tail, so we must
985 				 * flush unbh before the transaction commits
986 				 */
987 				reiserfs_add_tail_list(inode, unbh);
988 
989 				/*
990 				 * mark it dirty now to prevent commit_write
991 				 * from adding this buffer to the inode's
992 				 * dirty buffer list
993 				 */
994 				/*
995 				 * AKPM: changed __mark_buffer_dirty to
996 				 * mark_buffer_dirty().  It's still atomic,
997 				 * but it sets the page dirty too, which makes
998 				 * it eligible for writeback at any time by the
999 				 * VM (which was also the case with
1000 				 * __mark_buffer_dirty())
1001 				 */
1002 				mark_buffer_dirty(unbh);
1003 			}
1004 		} else {
1005 			/*
1006 			 * append indirect item with holes if needed, when
1007 			 * appending pointer to 'block'-th block use block,
1008 			 * which is already allocated
1009 			 */
1010 			struct cpu_key tmp_key;
1011 			/*
1012 			 * We use this in case we need to allocate
1013 			 * only one block which is a fastpath
1014 			 */
1015 			unp_t unf_single = 0;
1016 			unp_t *un;
1017 			__u64 max_to_insert =
1018 			    MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
1019 			    UNFM_P_SIZE;
1020 			__u64 blocks_needed;
1021 
1022 			RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
1023 			       "vs-804: invalid position for append");
1024 			/*
1025 			 * indirect item has to be appended,
1026 			 * set up key of that position
1027 			 * (key type is unimportant)
1028 			 */
1029 			make_cpu_key(&tmp_key, inode,
1030 				     le_key_k_offset(version,
1031 						     &ih->ih_key) +
1032 				     op_bytes_number(ih,
1033 						     inode->i_sb->s_blocksize),
1034 				     TYPE_INDIRECT, 3);
1035 
1036 			RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
1037 			       "green-805: invalid offset");
1038 			blocks_needed =
1039 			    1 +
1040 			    ((cpu_key_k_offset(&key) -
1041 			      cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
1042 			     s_blocksize_bits);
1043 
1044 			if (blocks_needed == 1) {
1045 				un = &unf_single;
1046 			} else {
1047 				un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
1048 				if (!un) {
1049 					un = &unf_single;
1050 					blocks_needed = 1;
1051 					max_to_insert = 0;
1052 				}
1053 			}
1054 			if (blocks_needed <= max_to_insert) {
1055 				/*
1056 				 * we are going to add target block to
1057 				 * the file. Use allocated block for that
1058 				 */
1059 				un[blocks_needed - 1] =
1060 				    cpu_to_le32(allocated_block_nr);
1061 				set_block_dev_mapped(bh_result,
1062 						     allocated_block_nr, inode);
1063 				set_buffer_new(bh_result);
1064 				done = 1;
1065 			} else {
1066 				/* paste hole to the indirect item */
1067 				/*
1068 				 * If kmalloc failed, max_to_insert becomes
1069 				 * zero and it means we only have space for
1070 				 * one block
1071 				 */
1072 				blocks_needed =
1073 				    max_to_insert ? max_to_insert : 1;
1074 			}
1075 			retval =
1076 			    reiserfs_paste_into_item(th, &path, &tmp_key, inode,
1077 						     (char *)un,
1078 						     UNFM_P_SIZE *
1079 						     blocks_needed);
1080 
1081 			if (blocks_needed != 1)
1082 				kfree(un);
1083 
1084 			if (retval) {
1085 				reiserfs_free_block(th, inode,
1086 						    allocated_block_nr, 1);
1087 				goto failure;
1088 			}
1089 			if (!done) {
1090 				/*
1091 				 * We need to mark new file size in case
1092 				 * this function will be interrupted/aborted
1093 				 * later on. And we may do this only for
1094 				 * holes.
1095 				 */
1096 				inode->i_size +=
1097 				    inode->i_sb->s_blocksize * blocks_needed;
1098 			}
1099 		}
1100 
1101 		if (done == 1)
1102 			break;
1103 
1104 		/*
1105 		 * this loop could log more blocks than we had originally
1106 		 * asked for.  So, we have to allow the transaction to end
1107 		 * if it is too big or too full.  Update the inode so things
1108 		 * are consistent if we crash before the function returns
1109 		 * release the path so that anybody waiting on the path before
1110 		 * ending their transaction will be able to continue.
1111 		 */
1112 		if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1113 			retval = restart_transaction(th, inode, &path);
1114 			if (retval)
1115 				goto failure;
1116 		}
1117 		/*
1118 		 * inserting indirect pointers for a hole can take a
1119 		 * long time.  reschedule if needed and also release the write
1120 		 * lock for others.
1121 		 */
1122 		reiserfs_cond_resched(inode->i_sb);
1123 
1124 		retval = search_for_position_by_key(inode->i_sb, &key, &path);
1125 		if (retval == IO_ERROR) {
1126 			retval = -EIO;
1127 			goto failure;
1128 		}
1129 		if (retval == POSITION_FOUND) {
1130 			reiserfs_warning(inode->i_sb, "vs-825",
1131 					 "%K should not be found", &key);
1132 			retval = -EEXIST;
1133 			if (allocated_block_nr)
1134 				reiserfs_free_block(th, inode,
1135 						    allocated_block_nr, 1);
1136 			pathrelse(&path);
1137 			goto failure;
1138 		}
1139 		bh = get_last_bh(&path);
1140 		ih = tp_item_head(&path);
1141 		item = tp_item_body(&path);
1142 		pos_in_item = path.pos_in_item;
1143 	} while (1);
1144 
1145 	retval = 0;
1146 
1147 failure:
1148 	if (th && (!dangle || (retval && !th->t_trans_id))) {
1149 		int err;
1150 		if (th->t_trans_id)
1151 			reiserfs_update_sd(th, inode);
1152 		err = reiserfs_end_persistent_transaction(th);
1153 		if (err)
1154 			retval = err;
1155 	}
1156 
1157 	reiserfs_write_unlock(inode->i_sb);
1158 	reiserfs_check_path(&path);
1159 	return retval;
1160 }
1161 
1162 static int
1163 reiserfs_readpages(struct file *file, struct address_space *mapping,
1164 		   struct list_head *pages, unsigned nr_pages)
1165 {
1166 	return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1167 }
1168 
1169 /*
1170  * Compute real number of used bytes by file
1171  * Following three functions can go away when we'll have enough space in
1172  * stat item
1173  */
1174 static int real_space_diff(struct inode *inode, int sd_size)
1175 {
1176 	int bytes;
1177 	loff_t blocksize = inode->i_sb->s_blocksize;
1178 
1179 	if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1180 		return sd_size;
1181 
1182 	/*
1183 	 * End of file is also in full block with indirect reference, so round
1184 	 * up to the next block.
1185 	 *
1186 	 * there is just no way to know if the tail is actually packed
1187 	 * on the file, so we have to assume it isn't.  When we pack the
1188 	 * tail, we add 4 bytes to pretend there really is an unformatted
1189 	 * node pointer
1190 	 */
1191 	bytes =
1192 	    ((inode->i_size +
1193 	      (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1194 	    sd_size;
1195 	return bytes;
1196 }
1197 
1198 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1199 					int sd_size)
1200 {
1201 	if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1202 		return inode->i_size +
1203 		    (loff_t) (real_space_diff(inode, sd_size));
1204 	}
1205 	return ((loff_t) real_space_diff(inode, sd_size)) +
1206 	    (((loff_t) blocks) << 9);
1207 }
1208 
1209 /* Compute number of blocks used by file in ReiserFS counting */
1210 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1211 {
1212 	loff_t bytes = inode_get_bytes(inode);
1213 	loff_t real_space = real_space_diff(inode, sd_size);
1214 
1215 	/* keeps fsck and non-quota versions of reiserfs happy */
1216 	if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1217 		bytes += (loff_t) 511;
1218 	}
1219 
1220 	/*
1221 	 * files from before the quota patch might i_blocks such that
1222 	 * bytes < real_space.  Deal with that here to prevent it from
1223 	 * going negative.
1224 	 */
1225 	if (bytes < real_space)
1226 		return 0;
1227 	return (bytes - real_space) >> 9;
1228 }
1229 
1230 /*
1231  * BAD: new directories have stat data of new type and all other items
1232  * of old type. Version stored in the inode says about body items, so
1233  * in update_stat_data we can not rely on inode, but have to check
1234  * item version directly
1235  */
1236 
1237 /* called by read_locked_inode */
1238 static void init_inode(struct inode *inode, struct treepath *path)
1239 {
1240 	struct buffer_head *bh;
1241 	struct item_head *ih;
1242 	__u32 rdev;
1243 
1244 	bh = PATH_PLAST_BUFFER(path);
1245 	ih = tp_item_head(path);
1246 
1247 	copy_key(INODE_PKEY(inode), &ih->ih_key);
1248 
1249 	INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
1250 	REISERFS_I(inode)->i_flags = 0;
1251 	REISERFS_I(inode)->i_prealloc_block = 0;
1252 	REISERFS_I(inode)->i_prealloc_count = 0;
1253 	REISERFS_I(inode)->i_trans_id = 0;
1254 	REISERFS_I(inode)->i_jl = NULL;
1255 	reiserfs_init_xattr_rwsem(inode);
1256 
1257 	if (stat_data_v1(ih)) {
1258 		struct stat_data_v1 *sd =
1259 		    (struct stat_data_v1 *)ih_item_body(bh, ih);
1260 		unsigned long blocks;
1261 
1262 		set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1263 		set_inode_sd_version(inode, STAT_DATA_V1);
1264 		inode->i_mode = sd_v1_mode(sd);
1265 		set_nlink(inode, sd_v1_nlink(sd));
1266 		i_uid_write(inode, sd_v1_uid(sd));
1267 		i_gid_write(inode, sd_v1_gid(sd));
1268 		inode->i_size = sd_v1_size(sd);
1269 		inode->i_atime.tv_sec = sd_v1_atime(sd);
1270 		inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1271 		inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1272 		inode->i_atime.tv_nsec = 0;
1273 		inode->i_ctime.tv_nsec = 0;
1274 		inode->i_mtime.tv_nsec = 0;
1275 
1276 		inode->i_blocks = sd_v1_blocks(sd);
1277 		inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1278 		blocks = (inode->i_size + 511) >> 9;
1279 		blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1280 
1281 		/*
1282 		 * there was a bug in <=3.5.23 when i_blocks could take
1283 		 * negative values. Starting from 3.5.17 this value could
1284 		 * even be stored in stat data. For such files we set
1285 		 * i_blocks based on file size. Just 2 notes: this can be
1286 		 * wrong for sparse files. On-disk value will be only
1287 		 * updated if file's inode will ever change
1288 		 */
1289 		if (inode->i_blocks > blocks) {
1290 			inode->i_blocks = blocks;
1291 		}
1292 
1293 		rdev = sd_v1_rdev(sd);
1294 		REISERFS_I(inode)->i_first_direct_byte =
1295 		    sd_v1_first_direct_byte(sd);
1296 
1297 		/*
1298 		 * an early bug in the quota code can give us an odd
1299 		 * number for the block count.  This is incorrect, fix it here.
1300 		 */
1301 		if (inode->i_blocks & 1) {
1302 			inode->i_blocks++;
1303 		}
1304 		inode_set_bytes(inode,
1305 				to_real_used_space(inode, inode->i_blocks,
1306 						   SD_V1_SIZE));
1307 		/*
1308 		 * nopack is initially zero for v1 objects. For v2 objects,
1309 		 * nopack is initialised from sd_attrs
1310 		 */
1311 		REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1312 	} else {
1313 		/*
1314 		 * new stat data found, but object may have old items
1315 		 * (directories and symlinks)
1316 		 */
1317 		struct stat_data *sd = (struct stat_data *)ih_item_body(bh, ih);
1318 
1319 		inode->i_mode = sd_v2_mode(sd);
1320 		set_nlink(inode, sd_v2_nlink(sd));
1321 		i_uid_write(inode, sd_v2_uid(sd));
1322 		inode->i_size = sd_v2_size(sd);
1323 		i_gid_write(inode, sd_v2_gid(sd));
1324 		inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1325 		inode->i_atime.tv_sec = sd_v2_atime(sd);
1326 		inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1327 		inode->i_ctime.tv_nsec = 0;
1328 		inode->i_mtime.tv_nsec = 0;
1329 		inode->i_atime.tv_nsec = 0;
1330 		inode->i_blocks = sd_v2_blocks(sd);
1331 		rdev = sd_v2_rdev(sd);
1332 		if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1333 			inode->i_generation =
1334 			    le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1335 		else
1336 			inode->i_generation = sd_v2_generation(sd);
1337 
1338 		if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1339 			set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1340 		else
1341 			set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1342 		REISERFS_I(inode)->i_first_direct_byte = 0;
1343 		set_inode_sd_version(inode, STAT_DATA_V2);
1344 		inode_set_bytes(inode,
1345 				to_real_used_space(inode, inode->i_blocks,
1346 						   SD_V2_SIZE));
1347 		/*
1348 		 * read persistent inode attributes from sd and initialise
1349 		 * generic inode flags from them
1350 		 */
1351 		REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1352 		sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1353 	}
1354 
1355 	pathrelse(path);
1356 	if (S_ISREG(inode->i_mode)) {
1357 		inode->i_op = &reiserfs_file_inode_operations;
1358 		inode->i_fop = &reiserfs_file_operations;
1359 		inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1360 	} else if (S_ISDIR(inode->i_mode)) {
1361 		inode->i_op = &reiserfs_dir_inode_operations;
1362 		inode->i_fop = &reiserfs_dir_operations;
1363 	} else if (S_ISLNK(inode->i_mode)) {
1364 		inode->i_op = &reiserfs_symlink_inode_operations;
1365 		inode_nohighmem(inode);
1366 		inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1367 	} else {
1368 		inode->i_blocks = 0;
1369 		inode->i_op = &reiserfs_special_inode_operations;
1370 		init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1371 	}
1372 }
1373 
1374 /* update new stat data with inode fields */
1375 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1376 {
1377 	struct stat_data *sd_v2 = (struct stat_data *)sd;
1378 
1379 	set_sd_v2_mode(sd_v2, inode->i_mode);
1380 	set_sd_v2_nlink(sd_v2, inode->i_nlink);
1381 	set_sd_v2_uid(sd_v2, i_uid_read(inode));
1382 	set_sd_v2_size(sd_v2, size);
1383 	set_sd_v2_gid(sd_v2, i_gid_read(inode));
1384 	set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1385 	set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1386 	set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1387 	set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1388 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1389 		set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1390 	else
1391 		set_sd_v2_generation(sd_v2, inode->i_generation);
1392 	set_sd_v2_attrs(sd_v2, REISERFS_I(inode)->i_attrs);
1393 }
1394 
1395 /* used to copy inode's fields to old stat data */
1396 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1397 {
1398 	struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1399 
1400 	set_sd_v1_mode(sd_v1, inode->i_mode);
1401 	set_sd_v1_uid(sd_v1, i_uid_read(inode));
1402 	set_sd_v1_gid(sd_v1, i_gid_read(inode));
1403 	set_sd_v1_nlink(sd_v1, inode->i_nlink);
1404 	set_sd_v1_size(sd_v1, size);
1405 	set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1406 	set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1407 	set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1408 
1409 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1410 		set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1411 	else
1412 		set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1413 
1414 	/* Sigh. i_first_direct_byte is back */
1415 	set_sd_v1_first_direct_byte(sd_v1,
1416 				    REISERFS_I(inode)->i_first_direct_byte);
1417 }
1418 
1419 /*
1420  * NOTE, you must prepare the buffer head before sending it here,
1421  * and then log it after the call
1422  */
1423 static void update_stat_data(struct treepath *path, struct inode *inode,
1424 			     loff_t size)
1425 {
1426 	struct buffer_head *bh;
1427 	struct item_head *ih;
1428 
1429 	bh = PATH_PLAST_BUFFER(path);
1430 	ih = tp_item_head(path);
1431 
1432 	if (!is_statdata_le_ih(ih))
1433 		reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1434 			       INODE_PKEY(inode), ih);
1435 
1436 	/* path points to old stat data */
1437 	if (stat_data_v1(ih)) {
1438 		inode2sd_v1(ih_item_body(bh, ih), inode, size);
1439 	} else {
1440 		inode2sd(ih_item_body(bh, ih), inode, size);
1441 	}
1442 
1443 	return;
1444 }
1445 
1446 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1447 			     struct inode *inode, loff_t size)
1448 {
1449 	struct cpu_key key;
1450 	INITIALIZE_PATH(path);
1451 	struct buffer_head *bh;
1452 	int fs_gen;
1453 	struct item_head *ih, tmp_ih;
1454 	int retval;
1455 
1456 	BUG_ON(!th->t_trans_id);
1457 
1458 	/* key type is unimportant */
1459 	make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1460 
1461 	for (;;) {
1462 		int pos;
1463 		/* look for the object's stat data */
1464 		retval = search_item(inode->i_sb, &key, &path);
1465 		if (retval == IO_ERROR) {
1466 			reiserfs_error(inode->i_sb, "vs-13050",
1467 				       "i/o failure occurred trying to "
1468 				       "update %K stat data", &key);
1469 			return;
1470 		}
1471 		if (retval == ITEM_NOT_FOUND) {
1472 			pos = PATH_LAST_POSITION(&path);
1473 			pathrelse(&path);
1474 			if (inode->i_nlink == 0) {
1475 				/*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1476 				return;
1477 			}
1478 			reiserfs_warning(inode->i_sb, "vs-13060",
1479 					 "stat data of object %k (nlink == %d) "
1480 					 "not found (pos %d)",
1481 					 INODE_PKEY(inode), inode->i_nlink,
1482 					 pos);
1483 			reiserfs_check_path(&path);
1484 			return;
1485 		}
1486 
1487 		/*
1488 		 * sigh, prepare_for_journal might schedule.  When it
1489 		 * schedules the FS might change.  We have to detect that,
1490 		 * and loop back to the search if the stat data item has moved
1491 		 */
1492 		bh = get_last_bh(&path);
1493 		ih = tp_item_head(&path);
1494 		copy_item_head(&tmp_ih, ih);
1495 		fs_gen = get_generation(inode->i_sb);
1496 		reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1497 
1498 		/* Stat_data item has been moved after scheduling. */
1499 		if (fs_changed(fs_gen, inode->i_sb)
1500 		    && item_moved(&tmp_ih, &path)) {
1501 			reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1502 			continue;
1503 		}
1504 		break;
1505 	}
1506 	update_stat_data(&path, inode, size);
1507 	journal_mark_dirty(th, bh);
1508 	pathrelse(&path);
1509 	return;
1510 }
1511 
1512 /*
1513  * reiserfs_read_locked_inode is called to read the inode off disk, and it
1514  * does a make_bad_inode when things go wrong.  But, we need to make sure
1515  * and clear the key in the private portion of the inode, otherwise a
1516  * corresponding iput might try to delete whatever object the inode last
1517  * represented.
1518  */
1519 static void reiserfs_make_bad_inode(struct inode *inode)
1520 {
1521 	memset(INODE_PKEY(inode), 0, KEY_SIZE);
1522 	make_bad_inode(inode);
1523 }
1524 
1525 /*
1526  * initially this function was derived from minix or ext2's analog and
1527  * evolved as the prototype did
1528  */
1529 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1530 {
1531 	struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1532 	inode->i_ino = args->objectid;
1533 	INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1534 	return 0;
1535 }
1536 
1537 /*
1538  * looks for stat data in the tree, and fills up the fields of in-core
1539  * inode stat data fields
1540  */
1541 void reiserfs_read_locked_inode(struct inode *inode,
1542 				struct reiserfs_iget_args *args)
1543 {
1544 	INITIALIZE_PATH(path_to_sd);
1545 	struct cpu_key key;
1546 	unsigned long dirino;
1547 	int retval;
1548 
1549 	dirino = args->dirid;
1550 
1551 	/*
1552 	 * set version 1, version 2 could be used too, because stat data
1553 	 * key is the same in both versions
1554 	 */
1555 	key.version = KEY_FORMAT_3_5;
1556 	key.on_disk_key.k_dir_id = dirino;
1557 	key.on_disk_key.k_objectid = inode->i_ino;
1558 	key.on_disk_key.k_offset = 0;
1559 	key.on_disk_key.k_type = 0;
1560 
1561 	/* look for the object's stat data */
1562 	retval = search_item(inode->i_sb, &key, &path_to_sd);
1563 	if (retval == IO_ERROR) {
1564 		reiserfs_error(inode->i_sb, "vs-13070",
1565 			       "i/o failure occurred trying to find "
1566 			       "stat data of %K", &key);
1567 		reiserfs_make_bad_inode(inode);
1568 		return;
1569 	}
1570 
1571 	/* a stale NFS handle can trigger this without it being an error */
1572 	if (retval != ITEM_FOUND) {
1573 		pathrelse(&path_to_sd);
1574 		reiserfs_make_bad_inode(inode);
1575 		clear_nlink(inode);
1576 		return;
1577 	}
1578 
1579 	init_inode(inode, &path_to_sd);
1580 
1581 	/*
1582 	 * It is possible that knfsd is trying to access inode of a file
1583 	 * that is being removed from the disk by some other thread. As we
1584 	 * update sd on unlink all that is required is to check for nlink
1585 	 * here. This bug was first found by Sizif when debugging
1586 	 * SquidNG/Butterfly, forgotten, and found again after Philippe
1587 	 * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1588 
1589 	 * More logical fix would require changes in fs/inode.c:iput() to
1590 	 * remove inode from hash-table _after_ fs cleaned disk stuff up and
1591 	 * in iget() to return NULL if I_FREEING inode is found in
1592 	 * hash-table.
1593 	 */
1594 
1595 	/*
1596 	 * Currently there is one place where it's ok to meet inode with
1597 	 * nlink==0: processing of open-unlinked and half-truncated files
1598 	 * during mount (fs/reiserfs/super.c:finish_unfinished()).
1599 	 */
1600 	if ((inode->i_nlink == 0) &&
1601 	    !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1602 		reiserfs_warning(inode->i_sb, "vs-13075",
1603 				 "dead inode read from disk %K. "
1604 				 "This is likely to be race with knfsd. Ignore",
1605 				 &key);
1606 		reiserfs_make_bad_inode(inode);
1607 	}
1608 
1609 	/* init inode should be relsing */
1610 	reiserfs_check_path(&path_to_sd);
1611 
1612 	/*
1613 	 * Stat data v1 doesn't support ACLs.
1614 	 */
1615 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
1616 		cache_no_acl(inode);
1617 }
1618 
1619 /*
1620  * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1621  *
1622  * @inode:    inode from hash table to check
1623  * @opaque:   "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1624  *
1625  * This function is called by iget5_locked() to distinguish reiserfs inodes
1626  * having the same inode numbers. Such inodes can only exist due to some
1627  * error condition. One of them should be bad. Inodes with identical
1628  * inode numbers (objectids) are distinguished by parent directory ids.
1629  *
1630  */
1631 int reiserfs_find_actor(struct inode *inode, void *opaque)
1632 {
1633 	struct reiserfs_iget_args *args;
1634 
1635 	args = opaque;
1636 	/* args is already in CPU order */
1637 	return (inode->i_ino == args->objectid) &&
1638 	    (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1639 }
1640 
1641 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1642 {
1643 	struct inode *inode;
1644 	struct reiserfs_iget_args args;
1645 	int depth;
1646 
1647 	args.objectid = key->on_disk_key.k_objectid;
1648 	args.dirid = key->on_disk_key.k_dir_id;
1649 	depth = reiserfs_write_unlock_nested(s);
1650 	inode = iget5_locked(s, key->on_disk_key.k_objectid,
1651 			     reiserfs_find_actor, reiserfs_init_locked_inode,
1652 			     (void *)(&args));
1653 	reiserfs_write_lock_nested(s, depth);
1654 	if (!inode)
1655 		return ERR_PTR(-ENOMEM);
1656 
1657 	if (inode->i_state & I_NEW) {
1658 		reiserfs_read_locked_inode(inode, &args);
1659 		unlock_new_inode(inode);
1660 	}
1661 
1662 	if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1663 		/* either due to i/o error or a stale NFS handle */
1664 		iput(inode);
1665 		inode = NULL;
1666 	}
1667 	return inode;
1668 }
1669 
1670 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1671 	u32 objectid, u32 dir_id, u32 generation)
1672 
1673 {
1674 	struct cpu_key key;
1675 	struct inode *inode;
1676 
1677 	key.on_disk_key.k_objectid = objectid;
1678 	key.on_disk_key.k_dir_id = dir_id;
1679 	reiserfs_write_lock(sb);
1680 	inode = reiserfs_iget(sb, &key);
1681 	if (inode && !IS_ERR(inode) && generation != 0 &&
1682 	    generation != inode->i_generation) {
1683 		iput(inode);
1684 		inode = NULL;
1685 	}
1686 	reiserfs_write_unlock(sb);
1687 
1688 	return d_obtain_alias(inode);
1689 }
1690 
1691 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1692 		int fh_len, int fh_type)
1693 {
1694 	/*
1695 	 * fhtype happens to reflect the number of u32s encoded.
1696 	 * due to a bug in earlier code, fhtype might indicate there
1697 	 * are more u32s then actually fitted.
1698 	 * so if fhtype seems to be more than len, reduce fhtype.
1699 	 * Valid types are:
1700 	 *   2 - objectid + dir_id - legacy support
1701 	 *   3 - objectid + dir_id + generation
1702 	 *   4 - objectid + dir_id + objectid and dirid of parent - legacy
1703 	 *   5 - objectid + dir_id + generation + objectid and dirid of parent
1704 	 *   6 - as above plus generation of directory
1705 	 * 6 does not fit in NFSv2 handles
1706 	 */
1707 	if (fh_type > fh_len) {
1708 		if (fh_type != 6 || fh_len != 5)
1709 			reiserfs_warning(sb, "reiserfs-13077",
1710 				"nfsd/reiserfs, fhtype=%d, len=%d - odd",
1711 				fh_type, fh_len);
1712 		fh_type = fh_len;
1713 	}
1714 	if (fh_len < 2)
1715 		return NULL;
1716 
1717 	return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1718 		(fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1719 }
1720 
1721 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1722 		int fh_len, int fh_type)
1723 {
1724 	if (fh_type > fh_len)
1725 		fh_type = fh_len;
1726 	if (fh_type < 4)
1727 		return NULL;
1728 
1729 	return reiserfs_get_dentry(sb,
1730 		(fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1731 		(fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1732 		(fh_type == 6) ? fid->raw[5] : 0);
1733 }
1734 
1735 int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1736 		       struct inode *parent)
1737 {
1738 	int maxlen = *lenp;
1739 
1740 	if (parent && (maxlen < 5)) {
1741 		*lenp = 5;
1742 		return FILEID_INVALID;
1743 	} else if (maxlen < 3) {
1744 		*lenp = 3;
1745 		return FILEID_INVALID;
1746 	}
1747 
1748 	data[0] = inode->i_ino;
1749 	data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1750 	data[2] = inode->i_generation;
1751 	*lenp = 3;
1752 	if (parent) {
1753 		data[3] = parent->i_ino;
1754 		data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1755 		*lenp = 5;
1756 		if (maxlen >= 6) {
1757 			data[5] = parent->i_generation;
1758 			*lenp = 6;
1759 		}
1760 	}
1761 	return *lenp;
1762 }
1763 
1764 /*
1765  * looks for stat data, then copies fields to it, marks the buffer
1766  * containing stat data as dirty
1767  */
1768 /*
1769  * reiserfs inodes are never really dirty, since the dirty inode call
1770  * always logs them.  This call allows the VFS inode marking routines
1771  * to properly mark inodes for datasync and such, but only actually
1772  * does something when called for a synchronous update.
1773  */
1774 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1775 {
1776 	struct reiserfs_transaction_handle th;
1777 	int jbegin_count = 1;
1778 
1779 	if (inode->i_sb->s_flags & MS_RDONLY)
1780 		return -EROFS;
1781 	/*
1782 	 * memory pressure can sometimes initiate write_inode calls with
1783 	 * sync == 1,
1784 	 * these cases are just when the system needs ram, not when the
1785 	 * inode needs to reach disk for safety, and they can safely be
1786 	 * ignored because the altered inode has already been logged.
1787 	 */
1788 	if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1789 		reiserfs_write_lock(inode->i_sb);
1790 		if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1791 			reiserfs_update_sd(&th, inode);
1792 			journal_end_sync(&th);
1793 		}
1794 		reiserfs_write_unlock(inode->i_sb);
1795 	}
1796 	return 0;
1797 }
1798 
1799 /*
1800  * stat data of new object is inserted already, this inserts the item
1801  * containing "." and ".." entries
1802  */
1803 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1804 				  struct inode *inode,
1805 				  struct item_head *ih, struct treepath *path,
1806 				  struct inode *dir)
1807 {
1808 	struct super_block *sb = th->t_super;
1809 	char empty_dir[EMPTY_DIR_SIZE];
1810 	char *body = empty_dir;
1811 	struct cpu_key key;
1812 	int retval;
1813 
1814 	BUG_ON(!th->t_trans_id);
1815 
1816 	_make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1817 		      le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1818 		      TYPE_DIRENTRY, 3 /*key length */ );
1819 
1820 	/*
1821 	 * compose item head for new item. Directories consist of items of
1822 	 * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1823 	 * is done by reiserfs_new_inode
1824 	 */
1825 	if (old_format_only(sb)) {
1826 		make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1827 				  TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1828 
1829 		make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1830 				       ih->ih_key.k_objectid,
1831 				       INODE_PKEY(dir)->k_dir_id,
1832 				       INODE_PKEY(dir)->k_objectid);
1833 	} else {
1834 		make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1835 				  TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1836 
1837 		make_empty_dir_item(body, ih->ih_key.k_dir_id,
1838 				    ih->ih_key.k_objectid,
1839 				    INODE_PKEY(dir)->k_dir_id,
1840 				    INODE_PKEY(dir)->k_objectid);
1841 	}
1842 
1843 	/* look for place in the tree for new item */
1844 	retval = search_item(sb, &key, path);
1845 	if (retval == IO_ERROR) {
1846 		reiserfs_error(sb, "vs-13080",
1847 			       "i/o failure occurred creating new directory");
1848 		return -EIO;
1849 	}
1850 	if (retval == ITEM_FOUND) {
1851 		pathrelse(path);
1852 		reiserfs_warning(sb, "vs-13070",
1853 				 "object with this key exists (%k)",
1854 				 &(ih->ih_key));
1855 		return -EEXIST;
1856 	}
1857 
1858 	/* insert item, that is empty directory item */
1859 	return reiserfs_insert_item(th, path, &key, ih, inode, body);
1860 }
1861 
1862 /*
1863  * stat data of object has been inserted, this inserts the item
1864  * containing the body of symlink
1865  */
1866 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1867 				struct inode *inode,
1868 				struct item_head *ih,
1869 				struct treepath *path, const char *symname,
1870 				int item_len)
1871 {
1872 	struct super_block *sb = th->t_super;
1873 	struct cpu_key key;
1874 	int retval;
1875 
1876 	BUG_ON(!th->t_trans_id);
1877 
1878 	_make_cpu_key(&key, KEY_FORMAT_3_5,
1879 		      le32_to_cpu(ih->ih_key.k_dir_id),
1880 		      le32_to_cpu(ih->ih_key.k_objectid),
1881 		      1, TYPE_DIRECT, 3 /*key length */ );
1882 
1883 	make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1884 			  0 /*free_space */ );
1885 
1886 	/* look for place in the tree for new item */
1887 	retval = search_item(sb, &key, path);
1888 	if (retval == IO_ERROR) {
1889 		reiserfs_error(sb, "vs-13080",
1890 			       "i/o failure occurred creating new symlink");
1891 		return -EIO;
1892 	}
1893 	if (retval == ITEM_FOUND) {
1894 		pathrelse(path);
1895 		reiserfs_warning(sb, "vs-13080",
1896 				 "object with this key exists (%k)",
1897 				 &(ih->ih_key));
1898 		return -EEXIST;
1899 	}
1900 
1901 	/* insert item, that is body of symlink */
1902 	return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1903 }
1904 
1905 /*
1906  * inserts the stat data into the tree, and then calls
1907  * reiserfs_new_directory (to insert ".", ".." item if new object is
1908  * directory) or reiserfs_new_symlink (to insert symlink body if new
1909  * object is symlink) or nothing (if new object is regular file)
1910 
1911  * NOTE! uid and gid must already be set in the inode.  If we return
1912  * non-zero due to an error, we have to drop the quota previously allocated
1913  * for the fresh inode.  This can only be done outside a transaction, so
1914  * if we return non-zero, we also end the transaction.
1915  *
1916  * @th: active transaction handle
1917  * @dir: parent directory for new inode
1918  * @mode: mode of new inode
1919  * @symname: symlink contents if inode is symlink
1920  * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1921  *         symlinks
1922  * @inode: inode to be filled
1923  * @security: optional security context to associate with this inode
1924  */
1925 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1926 		       struct inode *dir, umode_t mode, const char *symname,
1927 		       /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1928 		          strlen (symname) for symlinks) */
1929 		       loff_t i_size, struct dentry *dentry,
1930 		       struct inode *inode,
1931 		       struct reiserfs_security_handle *security)
1932 {
1933 	struct super_block *sb = dir->i_sb;
1934 	struct reiserfs_iget_args args;
1935 	INITIALIZE_PATH(path_to_key);
1936 	struct cpu_key key;
1937 	struct item_head ih;
1938 	struct stat_data sd;
1939 	int retval;
1940 	int err;
1941 	int depth;
1942 
1943 	BUG_ON(!th->t_trans_id);
1944 
1945 	depth = reiserfs_write_unlock_nested(sb);
1946 	err = dquot_alloc_inode(inode);
1947 	reiserfs_write_lock_nested(sb, depth);
1948 	if (err)
1949 		goto out_end_trans;
1950 	if (!dir->i_nlink) {
1951 		err = -EPERM;
1952 		goto out_bad_inode;
1953 	}
1954 
1955 	/* item head of new item */
1956 	ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1957 	ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1958 	if (!ih.ih_key.k_objectid) {
1959 		err = -ENOMEM;
1960 		goto out_bad_inode;
1961 	}
1962 	args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1963 	if (old_format_only(sb))
1964 		make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1965 				  TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1966 	else
1967 		make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1968 				  TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1969 	memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1970 	args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1971 
1972 	depth = reiserfs_write_unlock_nested(inode->i_sb);
1973 	err = insert_inode_locked4(inode, args.objectid,
1974 			     reiserfs_find_actor, &args);
1975 	reiserfs_write_lock_nested(inode->i_sb, depth);
1976 	if (err) {
1977 		err = -EINVAL;
1978 		goto out_bad_inode;
1979 	}
1980 
1981 	if (old_format_only(sb))
1982 		/*
1983 		 * not a perfect generation count, as object ids can be reused,
1984 		 * but this is as good as reiserfs can do right now.
1985 		 * note that the private part of inode isn't filled in yet,
1986 		 * we have to use the directory.
1987 		 */
1988 		inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1989 	else
1990 #if defined( USE_INODE_GENERATION_COUNTER )
1991 		inode->i_generation =
1992 		    le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1993 #else
1994 		inode->i_generation = ++event;
1995 #endif
1996 
1997 	/* fill stat data */
1998 	set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
1999 
2000 	/* uid and gid must already be set by the caller for quota init */
2001 
2002 	inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
2003 	inode->i_size = i_size;
2004 	inode->i_blocks = 0;
2005 	inode->i_bytes = 0;
2006 	REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
2007 	    U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
2008 
2009 	INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
2010 	REISERFS_I(inode)->i_flags = 0;
2011 	REISERFS_I(inode)->i_prealloc_block = 0;
2012 	REISERFS_I(inode)->i_prealloc_count = 0;
2013 	REISERFS_I(inode)->i_trans_id = 0;
2014 	REISERFS_I(inode)->i_jl = NULL;
2015 	REISERFS_I(inode)->i_attrs =
2016 	    REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2017 	sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2018 	reiserfs_init_xattr_rwsem(inode);
2019 
2020 	/* key to search for correct place for new stat data */
2021 	_make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2022 		      le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2023 		      TYPE_STAT_DATA, 3 /*key length */ );
2024 
2025 	/* find proper place for inserting of stat data */
2026 	retval = search_item(sb, &key, &path_to_key);
2027 	if (retval == IO_ERROR) {
2028 		err = -EIO;
2029 		goto out_bad_inode;
2030 	}
2031 	if (retval == ITEM_FOUND) {
2032 		pathrelse(&path_to_key);
2033 		err = -EEXIST;
2034 		goto out_bad_inode;
2035 	}
2036 	if (old_format_only(sb)) {
2037 		/* i_uid or i_gid is too big to be stored in stat data v3.5 */
2038 		if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2039 			pathrelse(&path_to_key);
2040 			err = -EINVAL;
2041 			goto out_bad_inode;
2042 		}
2043 		inode2sd_v1(&sd, inode, inode->i_size);
2044 	} else {
2045 		inode2sd(&sd, inode, inode->i_size);
2046 	}
2047 	/*
2048 	 * store in in-core inode the key of stat data and version all
2049 	 * object items will have (directory items will have old offset
2050 	 * format, other new objects will consist of new items)
2051 	 */
2052 	if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2053 		set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2054 	else
2055 		set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2056 	if (old_format_only(sb))
2057 		set_inode_sd_version(inode, STAT_DATA_V1);
2058 	else
2059 		set_inode_sd_version(inode, STAT_DATA_V2);
2060 
2061 	/* insert the stat data into the tree */
2062 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2063 	if (REISERFS_I(dir)->new_packing_locality)
2064 		th->displace_new_blocks = 1;
2065 #endif
2066 	retval =
2067 	    reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2068 				 (char *)(&sd));
2069 	if (retval) {
2070 		err = retval;
2071 		reiserfs_check_path(&path_to_key);
2072 		goto out_bad_inode;
2073 	}
2074 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2075 	if (!th->displace_new_blocks)
2076 		REISERFS_I(dir)->new_packing_locality = 0;
2077 #endif
2078 	if (S_ISDIR(mode)) {
2079 		/* insert item with "." and ".." */
2080 		retval =
2081 		    reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2082 	}
2083 
2084 	if (S_ISLNK(mode)) {
2085 		/* insert body of symlink */
2086 		if (!old_format_only(sb))
2087 			i_size = ROUND_UP(i_size);
2088 		retval =
2089 		    reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2090 					 i_size);
2091 	}
2092 	if (retval) {
2093 		err = retval;
2094 		reiserfs_check_path(&path_to_key);
2095 		journal_end(th);
2096 		goto out_inserted_sd;
2097 	}
2098 
2099 	if (reiserfs_posixacl(inode->i_sb)) {
2100 		reiserfs_write_unlock(inode->i_sb);
2101 		retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2102 		reiserfs_write_lock(inode->i_sb);
2103 		if (retval) {
2104 			err = retval;
2105 			reiserfs_check_path(&path_to_key);
2106 			journal_end(th);
2107 			goto out_inserted_sd;
2108 		}
2109 	} else if (inode->i_sb->s_flags & MS_POSIXACL) {
2110 		reiserfs_warning(inode->i_sb, "jdm-13090",
2111 				 "ACLs aren't enabled in the fs, "
2112 				 "but vfs thinks they are!");
2113 	} else if (IS_PRIVATE(dir))
2114 		inode->i_flags |= S_PRIVATE;
2115 
2116 	if (security->name) {
2117 		reiserfs_write_unlock(inode->i_sb);
2118 		retval = reiserfs_security_write(th, inode, security);
2119 		reiserfs_write_lock(inode->i_sb);
2120 		if (retval) {
2121 			err = retval;
2122 			reiserfs_check_path(&path_to_key);
2123 			retval = journal_end(th);
2124 			if (retval)
2125 				err = retval;
2126 			goto out_inserted_sd;
2127 		}
2128 	}
2129 
2130 	reiserfs_update_sd(th, inode);
2131 	reiserfs_check_path(&path_to_key);
2132 
2133 	return 0;
2134 
2135 out_bad_inode:
2136 	/* Invalidate the object, nothing was inserted yet */
2137 	INODE_PKEY(inode)->k_objectid = 0;
2138 
2139 	/* Quota change must be inside a transaction for journaling */
2140 	depth = reiserfs_write_unlock_nested(inode->i_sb);
2141 	dquot_free_inode(inode);
2142 	reiserfs_write_lock_nested(inode->i_sb, depth);
2143 
2144 out_end_trans:
2145 	journal_end(th);
2146 	/*
2147 	 * Drop can be outside and it needs more credits so it's better
2148 	 * to have it outside
2149 	 */
2150 	depth = reiserfs_write_unlock_nested(inode->i_sb);
2151 	dquot_drop(inode);
2152 	reiserfs_write_lock_nested(inode->i_sb, depth);
2153 	inode->i_flags |= S_NOQUOTA;
2154 	make_bad_inode(inode);
2155 
2156 out_inserted_sd:
2157 	clear_nlink(inode);
2158 	th->t_trans_id = 0;	/* so the caller can't use this handle later */
2159 	unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
2160 	iput(inode);
2161 	return err;
2162 }
2163 
2164 /*
2165  * finds the tail page in the page cache,
2166  * reads the last block in.
2167  *
2168  * On success, page_result is set to a locked, pinned page, and bh_result
2169  * is set to an up to date buffer for the last block in the file.  returns 0.
2170  *
2171  * tail conversion is not done, so bh_result might not be valid for writing
2172  * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2173  * trying to write the block.
2174  *
2175  * on failure, nonzero is returned, page_result and bh_result are untouched.
2176  */
2177 static int grab_tail_page(struct inode *inode,
2178 			  struct page **page_result,
2179 			  struct buffer_head **bh_result)
2180 {
2181 
2182 	/*
2183 	 * we want the page with the last byte in the file,
2184 	 * not the page that will hold the next byte for appending
2185 	 */
2186 	unsigned long index = (inode->i_size - 1) >> PAGE_SHIFT;
2187 	unsigned long pos = 0;
2188 	unsigned long start = 0;
2189 	unsigned long blocksize = inode->i_sb->s_blocksize;
2190 	unsigned long offset = (inode->i_size) & (PAGE_SIZE - 1);
2191 	struct buffer_head *bh;
2192 	struct buffer_head *head;
2193 	struct page *page;
2194 	int error;
2195 
2196 	/*
2197 	 * we know that we are only called with inode->i_size > 0.
2198 	 * we also know that a file tail can never be as big as a block
2199 	 * If i_size % blocksize == 0, our file is currently block aligned
2200 	 * and it won't need converting or zeroing after a truncate.
2201 	 */
2202 	if ((offset & (blocksize - 1)) == 0) {
2203 		return -ENOENT;
2204 	}
2205 	page = grab_cache_page(inode->i_mapping, index);
2206 	error = -ENOMEM;
2207 	if (!page) {
2208 		goto out;
2209 	}
2210 	/* start within the page of the last block in the file */
2211 	start = (offset / blocksize) * blocksize;
2212 
2213 	error = __block_write_begin(page, start, offset - start,
2214 				    reiserfs_get_block_create_0);
2215 	if (error)
2216 		goto unlock;
2217 
2218 	head = page_buffers(page);
2219 	bh = head;
2220 	do {
2221 		if (pos >= start) {
2222 			break;
2223 		}
2224 		bh = bh->b_this_page;
2225 		pos += blocksize;
2226 	} while (bh != head);
2227 
2228 	if (!buffer_uptodate(bh)) {
2229 		/*
2230 		 * note, this should never happen, prepare_write should be
2231 		 * taking care of this for us.  If the buffer isn't up to
2232 		 * date, I've screwed up the code to find the buffer, or the
2233 		 * code to call prepare_write
2234 		 */
2235 		reiserfs_error(inode->i_sb, "clm-6000",
2236 			       "error reading block %lu", bh->b_blocknr);
2237 		error = -EIO;
2238 		goto unlock;
2239 	}
2240 	*bh_result = bh;
2241 	*page_result = page;
2242 
2243 out:
2244 	return error;
2245 
2246 unlock:
2247 	unlock_page(page);
2248 	put_page(page);
2249 	return error;
2250 }
2251 
2252 /*
2253  * vfs version of truncate file.  Must NOT be called with
2254  * a transaction already started.
2255  *
2256  * some code taken from block_truncate_page
2257  */
2258 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2259 {
2260 	struct reiserfs_transaction_handle th;
2261 	/* we want the offset for the first byte after the end of the file */
2262 	unsigned long offset = inode->i_size & (PAGE_SIZE - 1);
2263 	unsigned blocksize = inode->i_sb->s_blocksize;
2264 	unsigned length;
2265 	struct page *page = NULL;
2266 	int error;
2267 	struct buffer_head *bh = NULL;
2268 	int err2;
2269 
2270 	reiserfs_write_lock(inode->i_sb);
2271 
2272 	if (inode->i_size > 0) {
2273 		error = grab_tail_page(inode, &page, &bh);
2274 		if (error) {
2275 			/*
2276 			 * -ENOENT means we truncated past the end of the
2277 			 * file, and get_block_create_0 could not find a
2278 			 * block to read in, which is ok.
2279 			 */
2280 			if (error != -ENOENT)
2281 				reiserfs_error(inode->i_sb, "clm-6001",
2282 					       "grab_tail_page failed %d",
2283 					       error);
2284 			page = NULL;
2285 			bh = NULL;
2286 		}
2287 	}
2288 
2289 	/*
2290 	 * so, if page != NULL, we have a buffer head for the offset at
2291 	 * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2292 	 * then we have an unformatted node.  Otherwise, we have a direct item,
2293 	 * and no zeroing is required on disk.  We zero after the truncate,
2294 	 * because the truncate might pack the item anyway
2295 	 * (it will unmap bh if it packs).
2296 	 *
2297 	 * it is enough to reserve space in transaction for 2 balancings:
2298 	 * one for "save" link adding and another for the first
2299 	 * cut_from_item. 1 is for update_sd
2300 	 */
2301 	error = journal_begin(&th, inode->i_sb,
2302 			      JOURNAL_PER_BALANCE_CNT * 2 + 1);
2303 	if (error)
2304 		goto out;
2305 	reiserfs_update_inode_transaction(inode);
2306 	if (update_timestamps)
2307 		/*
2308 		 * we are doing real truncate: if the system crashes
2309 		 * before the last transaction of truncating gets committed
2310 		 * - on reboot the file either appears truncated properly
2311 		 * or not truncated at all
2312 		 */
2313 		add_save_link(&th, inode, 1);
2314 	err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2315 	error = journal_end(&th);
2316 	if (error)
2317 		goto out;
2318 
2319 	/* check reiserfs_do_truncate after ending the transaction */
2320 	if (err2) {
2321 		error = err2;
2322   		goto out;
2323 	}
2324 
2325 	if (update_timestamps) {
2326 		error = remove_save_link(inode, 1 /* truncate */);
2327 		if (error)
2328 			goto out;
2329 	}
2330 
2331 	if (page) {
2332 		length = offset & (blocksize - 1);
2333 		/* if we are not on a block boundary */
2334 		if (length) {
2335 			length = blocksize - length;
2336 			zero_user(page, offset, length);
2337 			if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2338 				mark_buffer_dirty(bh);
2339 			}
2340 		}
2341 		unlock_page(page);
2342 		put_page(page);
2343 	}
2344 
2345 	reiserfs_write_unlock(inode->i_sb);
2346 
2347 	return 0;
2348 out:
2349 	if (page) {
2350 		unlock_page(page);
2351 		put_page(page);
2352 	}
2353 
2354 	reiserfs_write_unlock(inode->i_sb);
2355 
2356 	return error;
2357 }
2358 
2359 static int map_block_for_writepage(struct inode *inode,
2360 				   struct buffer_head *bh_result,
2361 				   unsigned long block)
2362 {
2363 	struct reiserfs_transaction_handle th;
2364 	int fs_gen;
2365 	struct item_head tmp_ih;
2366 	struct item_head *ih;
2367 	struct buffer_head *bh;
2368 	__le32 *item;
2369 	struct cpu_key key;
2370 	INITIALIZE_PATH(path);
2371 	int pos_in_item;
2372 	int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2373 	loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2374 	int retval;
2375 	int use_get_block = 0;
2376 	int bytes_copied = 0;
2377 	int copy_size;
2378 	int trans_running = 0;
2379 
2380 	/*
2381 	 * catch places below that try to log something without
2382 	 * starting a trans
2383 	 */
2384 	th.t_trans_id = 0;
2385 
2386 	if (!buffer_uptodate(bh_result)) {
2387 		return -EIO;
2388 	}
2389 
2390 	kmap(bh_result->b_page);
2391 start_over:
2392 	reiserfs_write_lock(inode->i_sb);
2393 	make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2394 
2395 research:
2396 	retval = search_for_position_by_key(inode->i_sb, &key, &path);
2397 	if (retval != POSITION_FOUND) {
2398 		use_get_block = 1;
2399 		goto out;
2400 	}
2401 
2402 	bh = get_last_bh(&path);
2403 	ih = tp_item_head(&path);
2404 	item = tp_item_body(&path);
2405 	pos_in_item = path.pos_in_item;
2406 
2407 	/* we've found an unformatted node */
2408 	if (indirect_item_found(retval, ih)) {
2409 		if (bytes_copied > 0) {
2410 			reiserfs_warning(inode->i_sb, "clm-6002",
2411 					 "bytes_copied %d", bytes_copied);
2412 		}
2413 		if (!get_block_num(item, pos_in_item)) {
2414 			/* crap, we are writing to a hole */
2415 			use_get_block = 1;
2416 			goto out;
2417 		}
2418 		set_block_dev_mapped(bh_result,
2419 				     get_block_num(item, pos_in_item), inode);
2420 	} else if (is_direct_le_ih(ih)) {
2421 		char *p;
2422 		p = page_address(bh_result->b_page);
2423 		p += (byte_offset - 1) & (PAGE_SIZE - 1);
2424 		copy_size = ih_item_len(ih) - pos_in_item;
2425 
2426 		fs_gen = get_generation(inode->i_sb);
2427 		copy_item_head(&tmp_ih, ih);
2428 
2429 		if (!trans_running) {
2430 			/* vs-3050 is gone, no need to drop the path */
2431 			retval = journal_begin(&th, inode->i_sb, jbegin_count);
2432 			if (retval)
2433 				goto out;
2434 			reiserfs_update_inode_transaction(inode);
2435 			trans_running = 1;
2436 			if (fs_changed(fs_gen, inode->i_sb)
2437 			    && item_moved(&tmp_ih, &path)) {
2438 				reiserfs_restore_prepared_buffer(inode->i_sb,
2439 								 bh);
2440 				goto research;
2441 			}
2442 		}
2443 
2444 		reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2445 
2446 		if (fs_changed(fs_gen, inode->i_sb)
2447 		    && item_moved(&tmp_ih, &path)) {
2448 			reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2449 			goto research;
2450 		}
2451 
2452 		memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2453 		       copy_size);
2454 
2455 		journal_mark_dirty(&th, bh);
2456 		bytes_copied += copy_size;
2457 		set_block_dev_mapped(bh_result, 0, inode);
2458 
2459 		/* are there still bytes left? */
2460 		if (bytes_copied < bh_result->b_size &&
2461 		    (byte_offset + bytes_copied) < inode->i_size) {
2462 			set_cpu_key_k_offset(&key,
2463 					     cpu_key_k_offset(&key) +
2464 					     copy_size);
2465 			goto research;
2466 		}
2467 	} else {
2468 		reiserfs_warning(inode->i_sb, "clm-6003",
2469 				 "bad item inode %lu", inode->i_ino);
2470 		retval = -EIO;
2471 		goto out;
2472 	}
2473 	retval = 0;
2474 
2475 out:
2476 	pathrelse(&path);
2477 	if (trans_running) {
2478 		int err = journal_end(&th);
2479 		if (err)
2480 			retval = err;
2481 		trans_running = 0;
2482 	}
2483 	reiserfs_write_unlock(inode->i_sb);
2484 
2485 	/* this is where we fill in holes in the file. */
2486 	if (use_get_block) {
2487 		retval = reiserfs_get_block(inode, block, bh_result,
2488 					    GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2489 					    | GET_BLOCK_NO_DANGLE);
2490 		if (!retval) {
2491 			if (!buffer_mapped(bh_result)
2492 			    || bh_result->b_blocknr == 0) {
2493 				/* get_block failed to find a mapped unformatted node. */
2494 				use_get_block = 0;
2495 				goto start_over;
2496 			}
2497 		}
2498 	}
2499 	kunmap(bh_result->b_page);
2500 
2501 	if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2502 		/*
2503 		 * we've copied data from the page into the direct item, so the
2504 		 * buffer in the page is now clean, mark it to reflect that.
2505 		 */
2506 		lock_buffer(bh_result);
2507 		clear_buffer_dirty(bh_result);
2508 		unlock_buffer(bh_result);
2509 	}
2510 	return retval;
2511 }
2512 
2513 /*
2514  * mason@suse.com: updated in 2.5.54 to follow the same general io
2515  * start/recovery path as __block_write_full_page, along with special
2516  * code to handle reiserfs tails.
2517  */
2518 static int reiserfs_write_full_page(struct page *page,
2519 				    struct writeback_control *wbc)
2520 {
2521 	struct inode *inode = page->mapping->host;
2522 	unsigned long end_index = inode->i_size >> PAGE_SHIFT;
2523 	int error = 0;
2524 	unsigned long block;
2525 	sector_t last_block;
2526 	struct buffer_head *head, *bh;
2527 	int partial = 0;
2528 	int nr = 0;
2529 	int checked = PageChecked(page);
2530 	struct reiserfs_transaction_handle th;
2531 	struct super_block *s = inode->i_sb;
2532 	int bh_per_page = PAGE_SIZE / s->s_blocksize;
2533 	th.t_trans_id = 0;
2534 
2535 	/* no logging allowed when nonblocking or from PF_MEMALLOC */
2536 	if (checked && (current->flags & PF_MEMALLOC)) {
2537 		redirty_page_for_writepage(wbc, page);
2538 		unlock_page(page);
2539 		return 0;
2540 	}
2541 
2542 	/*
2543 	 * The page dirty bit is cleared before writepage is called, which
2544 	 * means we have to tell create_empty_buffers to make dirty buffers
2545 	 * The page really should be up to date at this point, so tossing
2546 	 * in the BH_Uptodate is just a sanity check.
2547 	 */
2548 	if (!page_has_buffers(page)) {
2549 		create_empty_buffers(page, s->s_blocksize,
2550 				     (1 << BH_Dirty) | (1 << BH_Uptodate));
2551 	}
2552 	head = page_buffers(page);
2553 
2554 	/*
2555 	 * last page in the file, zero out any contents past the
2556 	 * last byte in the file
2557 	 */
2558 	if (page->index >= end_index) {
2559 		unsigned last_offset;
2560 
2561 		last_offset = inode->i_size & (PAGE_SIZE - 1);
2562 		/* no file contents in this page */
2563 		if (page->index >= end_index + 1 || !last_offset) {
2564 			unlock_page(page);
2565 			return 0;
2566 		}
2567 		zero_user_segment(page, last_offset, PAGE_SIZE);
2568 	}
2569 	bh = head;
2570 	block = page->index << (PAGE_SHIFT - s->s_blocksize_bits);
2571 	last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2572 	/* first map all the buffers, logging any direct items we find */
2573 	do {
2574 		if (block > last_block) {
2575 			/*
2576 			 * This can happen when the block size is less than
2577 			 * the page size.  The corresponding bytes in the page
2578 			 * were zero filled above
2579 			 */
2580 			clear_buffer_dirty(bh);
2581 			set_buffer_uptodate(bh);
2582 		} else if ((checked || buffer_dirty(bh)) &&
2583 		           (!buffer_mapped(bh) || (buffer_mapped(bh)
2584 						       && bh->b_blocknr ==
2585 						       0))) {
2586 			/*
2587 			 * not mapped yet, or it points to a direct item, search
2588 			 * the btree for the mapping info, and log any direct
2589 			 * items found
2590 			 */
2591 			if ((error = map_block_for_writepage(inode, bh, block))) {
2592 				goto fail;
2593 			}
2594 		}
2595 		bh = bh->b_this_page;
2596 		block++;
2597 	} while (bh != head);
2598 
2599 	/*
2600 	 * we start the transaction after map_block_for_writepage,
2601 	 * because it can create holes in the file (an unbounded operation).
2602 	 * starting it here, we can make a reliable estimate for how many
2603 	 * blocks we're going to log
2604 	 */
2605 	if (checked) {
2606 		ClearPageChecked(page);
2607 		reiserfs_write_lock(s);
2608 		error = journal_begin(&th, s, bh_per_page + 1);
2609 		if (error) {
2610 			reiserfs_write_unlock(s);
2611 			goto fail;
2612 		}
2613 		reiserfs_update_inode_transaction(inode);
2614 	}
2615 	/* now go through and lock any dirty buffers on the page */
2616 	do {
2617 		get_bh(bh);
2618 		if (!buffer_mapped(bh))
2619 			continue;
2620 		if (buffer_mapped(bh) && bh->b_blocknr == 0)
2621 			continue;
2622 
2623 		if (checked) {
2624 			reiserfs_prepare_for_journal(s, bh, 1);
2625 			journal_mark_dirty(&th, bh);
2626 			continue;
2627 		}
2628 		/*
2629 		 * from this point on, we know the buffer is mapped to a
2630 		 * real block and not a direct item
2631 		 */
2632 		if (wbc->sync_mode != WB_SYNC_NONE) {
2633 			lock_buffer(bh);
2634 		} else {
2635 			if (!trylock_buffer(bh)) {
2636 				redirty_page_for_writepage(wbc, page);
2637 				continue;
2638 			}
2639 		}
2640 		if (test_clear_buffer_dirty(bh)) {
2641 			mark_buffer_async_write(bh);
2642 		} else {
2643 			unlock_buffer(bh);
2644 		}
2645 	} while ((bh = bh->b_this_page) != head);
2646 
2647 	if (checked) {
2648 		error = journal_end(&th);
2649 		reiserfs_write_unlock(s);
2650 		if (error)
2651 			goto fail;
2652 	}
2653 	BUG_ON(PageWriteback(page));
2654 	set_page_writeback(page);
2655 	unlock_page(page);
2656 
2657 	/*
2658 	 * since any buffer might be the only dirty buffer on the page,
2659 	 * the first submit_bh can bring the page out of writeback.
2660 	 * be careful with the buffers.
2661 	 */
2662 	do {
2663 		struct buffer_head *next = bh->b_this_page;
2664 		if (buffer_async_write(bh)) {
2665 			submit_bh(REQ_OP_WRITE, 0, bh);
2666 			nr++;
2667 		}
2668 		put_bh(bh);
2669 		bh = next;
2670 	} while (bh != head);
2671 
2672 	error = 0;
2673 done:
2674 	if (nr == 0) {
2675 		/*
2676 		 * if this page only had a direct item, it is very possible for
2677 		 * no io to be required without there being an error.  Or,
2678 		 * someone else could have locked them and sent them down the
2679 		 * pipe without locking the page
2680 		 */
2681 		bh = head;
2682 		do {
2683 			if (!buffer_uptodate(bh)) {
2684 				partial = 1;
2685 				break;
2686 			}
2687 			bh = bh->b_this_page;
2688 		} while (bh != head);
2689 		if (!partial)
2690 			SetPageUptodate(page);
2691 		end_page_writeback(page);
2692 	}
2693 	return error;
2694 
2695 fail:
2696 	/*
2697 	 * catches various errors, we need to make sure any valid dirty blocks
2698 	 * get to the media.  The page is currently locked and not marked for
2699 	 * writeback
2700 	 */
2701 	ClearPageUptodate(page);
2702 	bh = head;
2703 	do {
2704 		get_bh(bh);
2705 		if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2706 			lock_buffer(bh);
2707 			mark_buffer_async_write(bh);
2708 		} else {
2709 			/*
2710 			 * clear any dirty bits that might have come from
2711 			 * getting attached to a dirty page
2712 			 */
2713 			clear_buffer_dirty(bh);
2714 		}
2715 		bh = bh->b_this_page;
2716 	} while (bh != head);
2717 	SetPageError(page);
2718 	BUG_ON(PageWriteback(page));
2719 	set_page_writeback(page);
2720 	unlock_page(page);
2721 	do {
2722 		struct buffer_head *next = bh->b_this_page;
2723 		if (buffer_async_write(bh)) {
2724 			clear_buffer_dirty(bh);
2725 			submit_bh(REQ_OP_WRITE, 0, bh);
2726 			nr++;
2727 		}
2728 		put_bh(bh);
2729 		bh = next;
2730 	} while (bh != head);
2731 	goto done;
2732 }
2733 
2734 static int reiserfs_readpage(struct file *f, struct page *page)
2735 {
2736 	return block_read_full_page(page, reiserfs_get_block);
2737 }
2738 
2739 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2740 {
2741 	struct inode *inode = page->mapping->host;
2742 	reiserfs_wait_on_write_block(inode->i_sb);
2743 	return reiserfs_write_full_page(page, wbc);
2744 }
2745 
2746 static void reiserfs_truncate_failed_write(struct inode *inode)
2747 {
2748 	truncate_inode_pages(inode->i_mapping, inode->i_size);
2749 	reiserfs_truncate_file(inode, 0);
2750 }
2751 
2752 static int reiserfs_write_begin(struct file *file,
2753 				struct address_space *mapping,
2754 				loff_t pos, unsigned len, unsigned flags,
2755 				struct page **pagep, void **fsdata)
2756 {
2757 	struct inode *inode;
2758 	struct page *page;
2759 	pgoff_t index;
2760 	int ret;
2761 	int old_ref = 0;
2762 
2763  	inode = mapping->host;
2764 	*fsdata = NULL;
2765  	if (flags & AOP_FLAG_CONT_EXPAND &&
2766  	    (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2767  		pos ++;
2768 		*fsdata = (void *)(unsigned long)flags;
2769 	}
2770 
2771 	index = pos >> PAGE_SHIFT;
2772 	page = grab_cache_page_write_begin(mapping, index, flags);
2773 	if (!page)
2774 		return -ENOMEM;
2775 	*pagep = page;
2776 
2777 	reiserfs_wait_on_write_block(inode->i_sb);
2778 	fix_tail_page_for_writing(page);
2779 	if (reiserfs_transaction_running(inode->i_sb)) {
2780 		struct reiserfs_transaction_handle *th;
2781 		th = (struct reiserfs_transaction_handle *)current->
2782 		    journal_info;
2783 		BUG_ON(!th->t_refcount);
2784 		BUG_ON(!th->t_trans_id);
2785 		old_ref = th->t_refcount;
2786 		th->t_refcount++;
2787 	}
2788 	ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2789 	if (ret && reiserfs_transaction_running(inode->i_sb)) {
2790 		struct reiserfs_transaction_handle *th = current->journal_info;
2791 		/*
2792 		 * this gets a little ugly.  If reiserfs_get_block returned an
2793 		 * error and left a transacstion running, we've got to close
2794 		 * it, and we've got to free handle if it was a persistent
2795 		 * transaction.
2796 		 *
2797 		 * But, if we had nested into an existing transaction, we need
2798 		 * to just drop the ref count on the handle.
2799 		 *
2800 		 * If old_ref == 0, the transaction is from reiserfs_get_block,
2801 		 * and it was a persistent trans.  Otherwise, it was nested
2802 		 * above.
2803 		 */
2804 		if (th->t_refcount > old_ref) {
2805 			if (old_ref)
2806 				th->t_refcount--;
2807 			else {
2808 				int err;
2809 				reiserfs_write_lock(inode->i_sb);
2810 				err = reiserfs_end_persistent_transaction(th);
2811 				reiserfs_write_unlock(inode->i_sb);
2812 				if (err)
2813 					ret = err;
2814 			}
2815 		}
2816 	}
2817 	if (ret) {
2818 		unlock_page(page);
2819 		put_page(page);
2820 		/* Truncate allocated blocks */
2821 		reiserfs_truncate_failed_write(inode);
2822 	}
2823 	return ret;
2824 }
2825 
2826 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2827 {
2828 	struct inode *inode = page->mapping->host;
2829 	int ret;
2830 	int old_ref = 0;
2831 	int depth;
2832 
2833 	depth = reiserfs_write_unlock_nested(inode->i_sb);
2834 	reiserfs_wait_on_write_block(inode->i_sb);
2835 	reiserfs_write_lock_nested(inode->i_sb, depth);
2836 
2837 	fix_tail_page_for_writing(page);
2838 	if (reiserfs_transaction_running(inode->i_sb)) {
2839 		struct reiserfs_transaction_handle *th;
2840 		th = (struct reiserfs_transaction_handle *)current->
2841 		    journal_info;
2842 		BUG_ON(!th->t_refcount);
2843 		BUG_ON(!th->t_trans_id);
2844 		old_ref = th->t_refcount;
2845 		th->t_refcount++;
2846 	}
2847 
2848 	ret = __block_write_begin(page, from, len, reiserfs_get_block);
2849 	if (ret && reiserfs_transaction_running(inode->i_sb)) {
2850 		struct reiserfs_transaction_handle *th = current->journal_info;
2851 		/*
2852 		 * this gets a little ugly.  If reiserfs_get_block returned an
2853 		 * error and left a transacstion running, we've got to close
2854 		 * it, and we've got to free handle if it was a persistent
2855 		 * transaction.
2856 		 *
2857 		 * But, if we had nested into an existing transaction, we need
2858 		 * to just drop the ref count on the handle.
2859 		 *
2860 		 * If old_ref == 0, the transaction is from reiserfs_get_block,
2861 		 * and it was a persistent trans.  Otherwise, it was nested
2862 		 * above.
2863 		 */
2864 		if (th->t_refcount > old_ref) {
2865 			if (old_ref)
2866 				th->t_refcount--;
2867 			else {
2868 				int err;
2869 				reiserfs_write_lock(inode->i_sb);
2870 				err = reiserfs_end_persistent_transaction(th);
2871 				reiserfs_write_unlock(inode->i_sb);
2872 				if (err)
2873 					ret = err;
2874 			}
2875 		}
2876 	}
2877 	return ret;
2878 
2879 }
2880 
2881 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2882 {
2883 	return generic_block_bmap(as, block, reiserfs_bmap);
2884 }
2885 
2886 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2887 			      loff_t pos, unsigned len, unsigned copied,
2888 			      struct page *page, void *fsdata)
2889 {
2890 	struct inode *inode = page->mapping->host;
2891 	int ret = 0;
2892 	int update_sd = 0;
2893 	struct reiserfs_transaction_handle *th;
2894 	unsigned start;
2895 	bool locked = false;
2896 
2897 	if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2898 		pos ++;
2899 
2900 	reiserfs_wait_on_write_block(inode->i_sb);
2901 	if (reiserfs_transaction_running(inode->i_sb))
2902 		th = current->journal_info;
2903 	else
2904 		th = NULL;
2905 
2906 	start = pos & (PAGE_SIZE - 1);
2907 	if (unlikely(copied < len)) {
2908 		if (!PageUptodate(page))
2909 			copied = 0;
2910 
2911 		page_zero_new_buffers(page, start + copied, start + len);
2912 	}
2913 	flush_dcache_page(page);
2914 
2915 	reiserfs_commit_page(inode, page, start, start + copied);
2916 
2917 	/*
2918 	 * generic_commit_write does this for us, but does not update the
2919 	 * transaction tracking stuff when the size changes.  So, we have
2920 	 * to do the i_size updates here.
2921 	 */
2922 	if (pos + copied > inode->i_size) {
2923 		struct reiserfs_transaction_handle myth;
2924 		reiserfs_write_lock(inode->i_sb);
2925 		locked = true;
2926 		/*
2927 		 * If the file have grown beyond the border where it
2928 		 * can have a tail, unmark it as needing a tail
2929 		 * packing
2930 		 */
2931 		if ((have_large_tails(inode->i_sb)
2932 		     && inode->i_size > i_block_size(inode) * 4)
2933 		    || (have_small_tails(inode->i_sb)
2934 			&& inode->i_size > i_block_size(inode)))
2935 			REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2936 
2937 		ret = journal_begin(&myth, inode->i_sb, 1);
2938 		if (ret)
2939 			goto journal_error;
2940 
2941 		reiserfs_update_inode_transaction(inode);
2942 		inode->i_size = pos + copied;
2943 		/*
2944 		 * this will just nest into our transaction.  It's important
2945 		 * to use mark_inode_dirty so the inode gets pushed around on
2946 		 * the dirty lists, and so that O_SYNC works as expected
2947 		 */
2948 		mark_inode_dirty(inode);
2949 		reiserfs_update_sd(&myth, inode);
2950 		update_sd = 1;
2951 		ret = journal_end(&myth);
2952 		if (ret)
2953 			goto journal_error;
2954 	}
2955 	if (th) {
2956 		if (!locked) {
2957 			reiserfs_write_lock(inode->i_sb);
2958 			locked = true;
2959 		}
2960 		if (!update_sd)
2961 			mark_inode_dirty(inode);
2962 		ret = reiserfs_end_persistent_transaction(th);
2963 		if (ret)
2964 			goto out;
2965 	}
2966 
2967 out:
2968 	if (locked)
2969 		reiserfs_write_unlock(inode->i_sb);
2970 	unlock_page(page);
2971 	put_page(page);
2972 
2973 	if (pos + len > inode->i_size)
2974 		reiserfs_truncate_failed_write(inode);
2975 
2976 	return ret == 0 ? copied : ret;
2977 
2978 journal_error:
2979 	reiserfs_write_unlock(inode->i_sb);
2980 	locked = false;
2981 	if (th) {
2982 		if (!update_sd)
2983 			reiserfs_update_sd(th, inode);
2984 		ret = reiserfs_end_persistent_transaction(th);
2985 	}
2986 	goto out;
2987 }
2988 
2989 int reiserfs_commit_write(struct file *f, struct page *page,
2990 			  unsigned from, unsigned to)
2991 {
2992 	struct inode *inode = page->mapping->host;
2993 	loff_t pos = ((loff_t) page->index << PAGE_SHIFT) + to;
2994 	int ret = 0;
2995 	int update_sd = 0;
2996 	struct reiserfs_transaction_handle *th = NULL;
2997 	int depth;
2998 
2999 	depth = reiserfs_write_unlock_nested(inode->i_sb);
3000 	reiserfs_wait_on_write_block(inode->i_sb);
3001 	reiserfs_write_lock_nested(inode->i_sb, depth);
3002 
3003 	if (reiserfs_transaction_running(inode->i_sb)) {
3004 		th = current->journal_info;
3005 	}
3006 	reiserfs_commit_page(inode, page, from, to);
3007 
3008 	/*
3009 	 * generic_commit_write does this for us, but does not update the
3010 	 * transaction tracking stuff when the size changes.  So, we have
3011 	 * to do the i_size updates here.
3012 	 */
3013 	if (pos > inode->i_size) {
3014 		struct reiserfs_transaction_handle myth;
3015 		/*
3016 		 * If the file have grown beyond the border where it
3017 		 * can have a tail, unmark it as needing a tail
3018 		 * packing
3019 		 */
3020 		if ((have_large_tails(inode->i_sb)
3021 		     && inode->i_size > i_block_size(inode) * 4)
3022 		    || (have_small_tails(inode->i_sb)
3023 			&& inode->i_size > i_block_size(inode)))
3024 			REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3025 
3026 		ret = journal_begin(&myth, inode->i_sb, 1);
3027 		if (ret)
3028 			goto journal_error;
3029 
3030 		reiserfs_update_inode_transaction(inode);
3031 		inode->i_size = pos;
3032 		/*
3033 		 * this will just nest into our transaction.  It's important
3034 		 * to use mark_inode_dirty so the inode gets pushed around
3035 		 * on the dirty lists, and so that O_SYNC works as expected
3036 		 */
3037 		mark_inode_dirty(inode);
3038 		reiserfs_update_sd(&myth, inode);
3039 		update_sd = 1;
3040 		ret = journal_end(&myth);
3041 		if (ret)
3042 			goto journal_error;
3043 	}
3044 	if (th) {
3045 		if (!update_sd)
3046 			mark_inode_dirty(inode);
3047 		ret = reiserfs_end_persistent_transaction(th);
3048 		if (ret)
3049 			goto out;
3050 	}
3051 
3052 out:
3053 	return ret;
3054 
3055 journal_error:
3056 	if (th) {
3057 		if (!update_sd)
3058 			reiserfs_update_sd(th, inode);
3059 		ret = reiserfs_end_persistent_transaction(th);
3060 	}
3061 
3062 	return ret;
3063 }
3064 
3065 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3066 {
3067 	if (reiserfs_attrs(inode->i_sb)) {
3068 		if (sd_attrs & REISERFS_SYNC_FL)
3069 			inode->i_flags |= S_SYNC;
3070 		else
3071 			inode->i_flags &= ~S_SYNC;
3072 		if (sd_attrs & REISERFS_IMMUTABLE_FL)
3073 			inode->i_flags |= S_IMMUTABLE;
3074 		else
3075 			inode->i_flags &= ~S_IMMUTABLE;
3076 		if (sd_attrs & REISERFS_APPEND_FL)
3077 			inode->i_flags |= S_APPEND;
3078 		else
3079 			inode->i_flags &= ~S_APPEND;
3080 		if (sd_attrs & REISERFS_NOATIME_FL)
3081 			inode->i_flags |= S_NOATIME;
3082 		else
3083 			inode->i_flags &= ~S_NOATIME;
3084 		if (sd_attrs & REISERFS_NOTAIL_FL)
3085 			REISERFS_I(inode)->i_flags |= i_nopack_mask;
3086 		else
3087 			REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3088 	}
3089 }
3090 
3091 /*
3092  * decide if this buffer needs to stay around for data logging or ordered
3093  * write purposes
3094  */
3095 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
3096 {
3097 	int ret = 1;
3098 	struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3099 
3100 	lock_buffer(bh);
3101 	spin_lock(&j->j_dirty_buffers_lock);
3102 	if (!buffer_mapped(bh)) {
3103 		goto free_jh;
3104 	}
3105 	/*
3106 	 * the page is locked, and the only places that log a data buffer
3107 	 * also lock the page.
3108 	 */
3109 	if (reiserfs_file_data_log(inode)) {
3110 		/*
3111 		 * very conservative, leave the buffer pinned if
3112 		 * anyone might need it.
3113 		 */
3114 		if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3115 			ret = 0;
3116 		}
3117 	} else  if (buffer_dirty(bh)) {
3118 		struct reiserfs_journal_list *jl;
3119 		struct reiserfs_jh *jh = bh->b_private;
3120 
3121 		/*
3122 		 * why is this safe?
3123 		 * reiserfs_setattr updates i_size in the on disk
3124 		 * stat data before allowing vmtruncate to be called.
3125 		 *
3126 		 * If buffer was put onto the ordered list for this
3127 		 * transaction, we know for sure either this transaction
3128 		 * or an older one already has updated i_size on disk,
3129 		 * and this ordered data won't be referenced in the file
3130 		 * if we crash.
3131 		 *
3132 		 * if the buffer was put onto the ordered list for an older
3133 		 * transaction, we need to leave it around
3134 		 */
3135 		if (jh && (jl = jh->jl)
3136 		    && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3137 			ret = 0;
3138 	}
3139 free_jh:
3140 	if (ret && bh->b_private) {
3141 		reiserfs_free_jh(bh);
3142 	}
3143 	spin_unlock(&j->j_dirty_buffers_lock);
3144 	unlock_buffer(bh);
3145 	return ret;
3146 }
3147 
3148 /* clm -- taken from fs/buffer.c:block_invalidate_page */
3149 static void reiserfs_invalidatepage(struct page *page, unsigned int offset,
3150 				    unsigned int length)
3151 {
3152 	struct buffer_head *head, *bh, *next;
3153 	struct inode *inode = page->mapping->host;
3154 	unsigned int curr_off = 0;
3155 	unsigned int stop = offset + length;
3156 	int partial_page = (offset || length < PAGE_SIZE);
3157 	int ret = 1;
3158 
3159 	BUG_ON(!PageLocked(page));
3160 
3161 	if (!partial_page)
3162 		ClearPageChecked(page);
3163 
3164 	if (!page_has_buffers(page))
3165 		goto out;
3166 
3167 	head = page_buffers(page);
3168 	bh = head;
3169 	do {
3170 		unsigned int next_off = curr_off + bh->b_size;
3171 		next = bh->b_this_page;
3172 
3173 		if (next_off > stop)
3174 			goto out;
3175 
3176 		/*
3177 		 * is this block fully invalidated?
3178 		 */
3179 		if (offset <= curr_off) {
3180 			if (invalidatepage_can_drop(inode, bh))
3181 				reiserfs_unmap_buffer(bh);
3182 			else
3183 				ret = 0;
3184 		}
3185 		curr_off = next_off;
3186 		bh = next;
3187 	} while (bh != head);
3188 
3189 	/*
3190 	 * We release buffers only if the entire page is being invalidated.
3191 	 * The get_block cached value has been unconditionally invalidated,
3192 	 * so real IO is not possible anymore.
3193 	 */
3194 	if (!partial_page && ret) {
3195 		ret = try_to_release_page(page, 0);
3196 		/* maybe should BUG_ON(!ret); - neilb */
3197 	}
3198 out:
3199 	return;
3200 }
3201 
3202 static int reiserfs_set_page_dirty(struct page *page)
3203 {
3204 	struct inode *inode = page->mapping->host;
3205 	if (reiserfs_file_data_log(inode)) {
3206 		SetPageChecked(page);
3207 		return __set_page_dirty_nobuffers(page);
3208 	}
3209 	return __set_page_dirty_buffers(page);
3210 }
3211 
3212 /*
3213  * Returns 1 if the page's buffers were dropped.  The page is locked.
3214  *
3215  * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3216  * in the buffers at page_buffers(page).
3217  *
3218  * even in -o notail mode, we can't be sure an old mount without -o notail
3219  * didn't create files with tails.
3220  */
3221 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3222 {
3223 	struct inode *inode = page->mapping->host;
3224 	struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3225 	struct buffer_head *head;
3226 	struct buffer_head *bh;
3227 	int ret = 1;
3228 
3229 	WARN_ON(PageChecked(page));
3230 	spin_lock(&j->j_dirty_buffers_lock);
3231 	head = page_buffers(page);
3232 	bh = head;
3233 	do {
3234 		if (bh->b_private) {
3235 			if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3236 				reiserfs_free_jh(bh);
3237 			} else {
3238 				ret = 0;
3239 				break;
3240 			}
3241 		}
3242 		bh = bh->b_this_page;
3243 	} while (bh != head);
3244 	if (ret)
3245 		ret = try_to_free_buffers(page);
3246 	spin_unlock(&j->j_dirty_buffers_lock);
3247 	return ret;
3248 }
3249 
3250 /*
3251  * We thank Mingming Cao for helping us understand in great detail what
3252  * to do in this section of the code.
3253  */
3254 static ssize_t reiserfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3255 {
3256 	struct file *file = iocb->ki_filp;
3257 	struct inode *inode = file->f_mapping->host;
3258 	size_t count = iov_iter_count(iter);
3259 	ssize_t ret;
3260 
3261 	ret = blockdev_direct_IO(iocb, inode, iter,
3262 				 reiserfs_get_blocks_direct_io);
3263 
3264 	/*
3265 	 * In case of error extending write may have instantiated a few
3266 	 * blocks outside i_size. Trim these off again.
3267 	 */
3268 	if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
3269 		loff_t isize = i_size_read(inode);
3270 		loff_t end = iocb->ki_pos + count;
3271 
3272 		if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3273 			truncate_setsize(inode, isize);
3274 			reiserfs_vfs_truncate_file(inode);
3275 		}
3276 	}
3277 
3278 	return ret;
3279 }
3280 
3281 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3282 {
3283 	struct inode *inode = d_inode(dentry);
3284 	unsigned int ia_valid;
3285 	int error;
3286 
3287 	error = setattr_prepare(dentry, attr);
3288 	if (error)
3289 		return error;
3290 
3291 	/* must be turned off for recursive notify_change calls */
3292 	ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3293 
3294 	if (is_quota_modification(inode, attr)) {
3295 		error = dquot_initialize(inode);
3296 		if (error)
3297 			return error;
3298 	}
3299 	reiserfs_write_lock(inode->i_sb);
3300 	if (attr->ia_valid & ATTR_SIZE) {
3301 		/*
3302 		 * version 2 items will be caught by the s_maxbytes check
3303 		 * done for us in vmtruncate
3304 		 */
3305 		if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3306 		    attr->ia_size > MAX_NON_LFS) {
3307 			reiserfs_write_unlock(inode->i_sb);
3308 			error = -EFBIG;
3309 			goto out;
3310 		}
3311 
3312 		inode_dio_wait(inode);
3313 
3314 		/* fill in hole pointers in the expanding truncate case. */
3315 		if (attr->ia_size > inode->i_size) {
3316 			error = generic_cont_expand_simple(inode, attr->ia_size);
3317 			if (REISERFS_I(inode)->i_prealloc_count > 0) {
3318 				int err;
3319 				struct reiserfs_transaction_handle th;
3320 				/* we're changing at most 2 bitmaps, inode + super */
3321 				err = journal_begin(&th, inode->i_sb, 4);
3322 				if (!err) {
3323 					reiserfs_discard_prealloc(&th, inode);
3324 					err = journal_end(&th);
3325 				}
3326 				if (err)
3327 					error = err;
3328 			}
3329 			if (error) {
3330 				reiserfs_write_unlock(inode->i_sb);
3331 				goto out;
3332 			}
3333 			/*
3334 			 * file size is changed, ctime and mtime are
3335 			 * to be updated
3336 			 */
3337 			attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3338 		}
3339 	}
3340 	reiserfs_write_unlock(inode->i_sb);
3341 
3342 	if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3343 	     ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3344 	    (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3345 		/* stat data of format v3.5 has 16 bit uid and gid */
3346 		error = -EINVAL;
3347 		goto out;
3348 	}
3349 
3350 	if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3351 	    (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3352 		struct reiserfs_transaction_handle th;
3353 		int jbegin_count =
3354 		    2 *
3355 		    (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3356 		     REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3357 		    2;
3358 
3359 		error = reiserfs_chown_xattrs(inode, attr);
3360 
3361 		if (error)
3362 			return error;
3363 
3364 		/*
3365 		 * (user+group)*(old+new) structure - we count quota
3366 		 * info and , inode write (sb, inode)
3367 		 */
3368 		reiserfs_write_lock(inode->i_sb);
3369 		error = journal_begin(&th, inode->i_sb, jbegin_count);
3370 		reiserfs_write_unlock(inode->i_sb);
3371 		if (error)
3372 			goto out;
3373 		error = dquot_transfer(inode, attr);
3374 		reiserfs_write_lock(inode->i_sb);
3375 		if (error) {
3376 			journal_end(&th);
3377 			reiserfs_write_unlock(inode->i_sb);
3378 			goto out;
3379 		}
3380 
3381 		/*
3382 		 * Update corresponding info in inode so that everything
3383 		 * is in one transaction
3384 		 */
3385 		if (attr->ia_valid & ATTR_UID)
3386 			inode->i_uid = attr->ia_uid;
3387 		if (attr->ia_valid & ATTR_GID)
3388 			inode->i_gid = attr->ia_gid;
3389 		mark_inode_dirty(inode);
3390 		error = journal_end(&th);
3391 		reiserfs_write_unlock(inode->i_sb);
3392 		if (error)
3393 			goto out;
3394 	}
3395 
3396 	if ((attr->ia_valid & ATTR_SIZE) &&
3397 	    attr->ia_size != i_size_read(inode)) {
3398 		error = inode_newsize_ok(inode, attr->ia_size);
3399 		if (!error) {
3400 			/*
3401 			 * Could race against reiserfs_file_release
3402 			 * if called from NFS, so take tailpack mutex.
3403 			 */
3404 			mutex_lock(&REISERFS_I(inode)->tailpack);
3405 			truncate_setsize(inode, attr->ia_size);
3406 			reiserfs_truncate_file(inode, 1);
3407 			mutex_unlock(&REISERFS_I(inode)->tailpack);
3408 		}
3409 	}
3410 
3411 	if (!error) {
3412 		setattr_copy(inode, attr);
3413 		mark_inode_dirty(inode);
3414 	}
3415 
3416 	if (!error && reiserfs_posixacl(inode->i_sb)) {
3417 		if (attr->ia_valid & ATTR_MODE)
3418 			error = reiserfs_acl_chmod(inode);
3419 	}
3420 
3421 out:
3422 	return error;
3423 }
3424 
3425 const struct address_space_operations reiserfs_address_space_operations = {
3426 	.writepage = reiserfs_writepage,
3427 	.readpage = reiserfs_readpage,
3428 	.readpages = reiserfs_readpages,
3429 	.releasepage = reiserfs_releasepage,
3430 	.invalidatepage = reiserfs_invalidatepage,
3431 	.write_begin = reiserfs_write_begin,
3432 	.write_end = reiserfs_write_end,
3433 	.bmap = reiserfs_aop_bmap,
3434 	.direct_IO = reiserfs_direct_IO,
3435 	.set_page_dirty = reiserfs_set_page_dirty,
3436 };
3437