Loading...
1/*
2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3 */
4
5#include <linux/time.h>
6#include <linux/fs.h>
7#include <linux/reiserfs_fs.h>
8#include <linux/reiserfs_acl.h>
9#include <linux/reiserfs_xattr.h>
10#include <linux/exportfs.h>
11#include <linux/pagemap.h>
12#include <linux/highmem.h>
13#include <linux/slab.h>
14#include <asm/uaccess.h>
15#include <asm/unaligned.h>
16#include <linux/buffer_head.h>
17#include <linux/mpage.h>
18#include <linux/writeback.h>
19#include <linux/quotaops.h>
20#include <linux/swap.h>
21
22int reiserfs_commit_write(struct file *f, struct page *page,
23 unsigned from, unsigned to);
24
25void reiserfs_evict_inode(struct inode *inode)
26{
27 /* We need blocks for transaction + (user+group) quota update (possibly delete) */
28 int jbegin_count =
29 JOURNAL_PER_BALANCE_CNT * 2 +
30 2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
31 struct reiserfs_transaction_handle th;
32 int depth;
33 int err;
34
35 if (!inode->i_nlink && !is_bad_inode(inode))
36 dquot_initialize(inode);
37
38 truncate_inode_pages(&inode->i_data, 0);
39 if (inode->i_nlink)
40 goto no_delete;
41
42 depth = reiserfs_write_lock_once(inode->i_sb);
43
44 /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
45 if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) { /* also handles bad_inode case */
46 reiserfs_delete_xattrs(inode);
47
48 if (journal_begin(&th, inode->i_sb, jbegin_count))
49 goto out;
50 reiserfs_update_inode_transaction(inode);
51
52 reiserfs_discard_prealloc(&th, inode);
53
54 err = reiserfs_delete_object(&th, inode);
55
56 /* Do quota update inside a transaction for journaled quotas. We must do that
57 * after delete_object so that quota updates go into the same transaction as
58 * stat data deletion */
59 if (!err)
60 dquot_free_inode(inode);
61
62 if (journal_end(&th, inode->i_sb, jbegin_count))
63 goto out;
64
65 /* check return value from reiserfs_delete_object after
66 * ending the transaction
67 */
68 if (err)
69 goto out;
70
71 /* all items of file are deleted, so we can remove "save" link */
72 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
73 * about an error here */
74 } else {
75 /* no object items are in the tree */
76 ;
77 }
78 out:
79 end_writeback(inode); /* note this must go after the journal_end to prevent deadlock */
80 dquot_drop(inode);
81 inode->i_blocks = 0;
82 reiserfs_write_unlock_once(inode->i_sb, depth);
83 return;
84
85no_delete:
86 end_writeback(inode);
87 dquot_drop(inode);
88}
89
90static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
91 __u32 objectid, loff_t offset, int type, int length)
92{
93 key->version = version;
94
95 key->on_disk_key.k_dir_id = dirid;
96 key->on_disk_key.k_objectid = objectid;
97 set_cpu_key_k_offset(key, offset);
98 set_cpu_key_k_type(key, type);
99 key->key_length = length;
100}
101
102/* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
103 offset and type of key */
104void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
105 int type, int length)
106{
107 _make_cpu_key(key, get_inode_item_key_version(inode),
108 le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
109 le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
110 length);
111}
112
113//
114// when key is 0, do not set version and short key
115//
116inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
117 int version,
118 loff_t offset, int type, int length,
119 int entry_count /*or ih_free_space */ )
120{
121 if (key) {
122 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
123 ih->ih_key.k_objectid =
124 cpu_to_le32(key->on_disk_key.k_objectid);
125 }
126 put_ih_version(ih, version);
127 set_le_ih_k_offset(ih, offset);
128 set_le_ih_k_type(ih, type);
129 put_ih_item_len(ih, length);
130 /* set_ih_free_space (ih, 0); */
131 // for directory items it is entry count, for directs and stat
132 // datas - 0xffff, for indirects - 0
133 put_ih_entry_count(ih, entry_count);
134}
135
136//
137// FIXME: we might cache recently accessed indirect item
138
139// Ugh. Not too eager for that....
140// I cut the code until such time as I see a convincing argument (benchmark).
141// I don't want a bloated inode struct..., and I don't like code complexity....
142
143/* cutting the code is fine, since it really isn't in use yet and is easy
144** to add back in. But, Vladimir has a really good idea here. Think
145** about what happens for reading a file. For each page,
146** The VFS layer calls reiserfs_readpage, who searches the tree to find
147** an indirect item. This indirect item has X number of pointers, where
148** X is a big number if we've done the block allocation right. But,
149** we only use one or two of these pointers during each call to readpage,
150** needlessly researching again later on.
151**
152** The size of the cache could be dynamic based on the size of the file.
153**
154** I'd also like to see us cache the location the stat data item, since
155** we are needlessly researching for that frequently.
156**
157** --chris
158*/
159
160/* If this page has a file tail in it, and
161** it was read in by get_block_create_0, the page data is valid,
162** but tail is still sitting in a direct item, and we can't write to
163** it. So, look through this page, and check all the mapped buffers
164** to make sure they have valid block numbers. Any that don't need
165** to be unmapped, so that __block_write_begin will correctly call
166** reiserfs_get_block to convert the tail into an unformatted node
167*/
168static inline void fix_tail_page_for_writing(struct page *page)
169{
170 struct buffer_head *head, *next, *bh;
171
172 if (page && page_has_buffers(page)) {
173 head = page_buffers(page);
174 bh = head;
175 do {
176 next = bh->b_this_page;
177 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
178 reiserfs_unmap_buffer(bh);
179 }
180 bh = next;
181 } while (bh != head);
182 }
183}
184
185/* reiserfs_get_block does not need to allocate a block only if it has been
186 done already or non-hole position has been found in the indirect item */
187static inline int allocation_needed(int retval, b_blocknr_t allocated,
188 struct item_head *ih,
189 __le32 * item, int pos_in_item)
190{
191 if (allocated)
192 return 0;
193 if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
194 get_block_num(item, pos_in_item))
195 return 0;
196 return 1;
197}
198
199static inline int indirect_item_found(int retval, struct item_head *ih)
200{
201 return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
202}
203
204static inline void set_block_dev_mapped(struct buffer_head *bh,
205 b_blocknr_t block, struct inode *inode)
206{
207 map_bh(bh, inode->i_sb, block);
208}
209
210//
211// files which were created in the earlier version can not be longer,
212// than 2 gb
213//
214static int file_capable(struct inode *inode, sector_t block)
215{
216 if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 || // it is new file.
217 block < (1 << (31 - inode->i_sb->s_blocksize_bits))) // old file, but 'block' is inside of 2gb
218 return 1;
219
220 return 0;
221}
222
223static int restart_transaction(struct reiserfs_transaction_handle *th,
224 struct inode *inode, struct treepath *path)
225{
226 struct super_block *s = th->t_super;
227 int len = th->t_blocks_allocated;
228 int err;
229
230 BUG_ON(!th->t_trans_id);
231 BUG_ON(!th->t_refcount);
232
233 pathrelse(path);
234
235 /* we cannot restart while nested */
236 if (th->t_refcount > 1) {
237 return 0;
238 }
239 reiserfs_update_sd(th, inode);
240 err = journal_end(th, s, len);
241 if (!err) {
242 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
243 if (!err)
244 reiserfs_update_inode_transaction(inode);
245 }
246 return err;
247}
248
249// it is called by get_block when create == 0. Returns block number
250// for 'block'-th logical block of file. When it hits direct item it
251// returns 0 (being called from bmap) or read direct item into piece
252// of page (bh_result)
253
254// Please improve the english/clarity in the comment above, as it is
255// hard to understand.
256
257static int _get_block_create_0(struct inode *inode, sector_t block,
258 struct buffer_head *bh_result, int args)
259{
260 INITIALIZE_PATH(path);
261 struct cpu_key key;
262 struct buffer_head *bh;
263 struct item_head *ih, tmp_ih;
264 b_blocknr_t blocknr;
265 char *p = NULL;
266 int chars;
267 int ret;
268 int result;
269 int done = 0;
270 unsigned long offset;
271
272 // prepare the key to look for the 'block'-th block of file
273 make_cpu_key(&key, inode,
274 (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
275 3);
276
277 result = search_for_position_by_key(inode->i_sb, &key, &path);
278 if (result != POSITION_FOUND) {
279 pathrelse(&path);
280 if (p)
281 kunmap(bh_result->b_page);
282 if (result == IO_ERROR)
283 return -EIO;
284 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
285 // That there is some MMAPED data associated with it that is yet to be written to disk.
286 if ((args & GET_BLOCK_NO_HOLE)
287 && !PageUptodate(bh_result->b_page)) {
288 return -ENOENT;
289 }
290 return 0;
291 }
292 //
293 bh = get_last_bh(&path);
294 ih = get_ih(&path);
295 if (is_indirect_le_ih(ih)) {
296 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
297
298 /* FIXME: here we could cache indirect item or part of it in
299 the inode to avoid search_by_key in case of subsequent
300 access to file */
301 blocknr = get_block_num(ind_item, path.pos_in_item);
302 ret = 0;
303 if (blocknr) {
304 map_bh(bh_result, inode->i_sb, blocknr);
305 if (path.pos_in_item ==
306 ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
307 set_buffer_boundary(bh_result);
308 }
309 } else
310 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
311 // That there is some MMAPED data associated with it that is yet to be written to disk.
312 if ((args & GET_BLOCK_NO_HOLE)
313 && !PageUptodate(bh_result->b_page)) {
314 ret = -ENOENT;
315 }
316
317 pathrelse(&path);
318 if (p)
319 kunmap(bh_result->b_page);
320 return ret;
321 }
322 // requested data are in direct item(s)
323 if (!(args & GET_BLOCK_READ_DIRECT)) {
324 // we are called by bmap. FIXME: we can not map block of file
325 // when it is stored in direct item(s)
326 pathrelse(&path);
327 if (p)
328 kunmap(bh_result->b_page);
329 return -ENOENT;
330 }
331
332 /* if we've got a direct item, and the buffer or page was uptodate,
333 ** we don't want to pull data off disk again. skip to the
334 ** end, where we map the buffer and return
335 */
336 if (buffer_uptodate(bh_result)) {
337 goto finished;
338 } else
339 /*
340 ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
341 ** pages without any buffers. If the page is up to date, we don't want
342 ** read old data off disk. Set the up to date bit on the buffer instead
343 ** and jump to the end
344 */
345 if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
346 set_buffer_uptodate(bh_result);
347 goto finished;
348 }
349 // read file tail into part of page
350 offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
351 copy_item_head(&tmp_ih, ih);
352
353 /* we only want to kmap if we are reading the tail into the page.
354 ** this is not the common case, so we don't kmap until we are
355 ** sure we need to. But, this means the item might move if
356 ** kmap schedules
357 */
358 if (!p)
359 p = (char *)kmap(bh_result->b_page);
360
361 p += offset;
362 memset(p, 0, inode->i_sb->s_blocksize);
363 do {
364 if (!is_direct_le_ih(ih)) {
365 BUG();
366 }
367 /* make sure we don't read more bytes than actually exist in
368 ** the file. This can happen in odd cases where i_size isn't
369 ** correct, and when direct item padding results in a few
370 ** extra bytes at the end of the direct item
371 */
372 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
373 break;
374 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
375 chars =
376 inode->i_size - (le_ih_k_offset(ih) - 1) -
377 path.pos_in_item;
378 done = 1;
379 } else {
380 chars = ih_item_len(ih) - path.pos_in_item;
381 }
382 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
383
384 if (done)
385 break;
386
387 p += chars;
388
389 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
390 // we done, if read direct item is not the last item of
391 // node FIXME: we could try to check right delimiting key
392 // to see whether direct item continues in the right
393 // neighbor or rely on i_size
394 break;
395
396 // update key to look for the next piece
397 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
398 result = search_for_position_by_key(inode->i_sb, &key, &path);
399 if (result != POSITION_FOUND)
400 // i/o error most likely
401 break;
402 bh = get_last_bh(&path);
403 ih = get_ih(&path);
404 } while (1);
405
406 flush_dcache_page(bh_result->b_page);
407 kunmap(bh_result->b_page);
408
409 finished:
410 pathrelse(&path);
411
412 if (result == IO_ERROR)
413 return -EIO;
414
415 /* this buffer has valid data, but isn't valid for io. mapping it to
416 * block #0 tells the rest of reiserfs it just has a tail in it
417 */
418 map_bh(bh_result, inode->i_sb, 0);
419 set_buffer_uptodate(bh_result);
420 return 0;
421}
422
423// this is called to create file map. So, _get_block_create_0 will not
424// read direct item
425static int reiserfs_bmap(struct inode *inode, sector_t block,
426 struct buffer_head *bh_result, int create)
427{
428 if (!file_capable(inode, block))
429 return -EFBIG;
430
431 reiserfs_write_lock(inode->i_sb);
432 /* do not read the direct item */
433 _get_block_create_0(inode, block, bh_result, 0);
434 reiserfs_write_unlock(inode->i_sb);
435 return 0;
436}
437
438/* special version of get_block that is only used by grab_tail_page right
439** now. It is sent to __block_write_begin, and when you try to get a
440** block past the end of the file (or a block from a hole) it returns
441** -ENOENT instead of a valid buffer. __block_write_begin expects to
442** be able to do i/o on the buffers returned, unless an error value
443** is also returned.
444**
445** So, this allows __block_write_begin to be used for reading a single block
446** in a page. Where it does not produce a valid page for holes, or past the
447** end of the file. This turns out to be exactly what we need for reading
448** tails for conversion.
449**
450** The point of the wrapper is forcing a certain value for create, even
451** though the VFS layer is calling this function with create==1. If you
452** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
453** don't use this function.
454*/
455static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
456 struct buffer_head *bh_result,
457 int create)
458{
459 return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
460}
461
462/* This is special helper for reiserfs_get_block in case we are executing
463 direct_IO request. */
464static int reiserfs_get_blocks_direct_io(struct inode *inode,
465 sector_t iblock,
466 struct buffer_head *bh_result,
467 int create)
468{
469 int ret;
470
471 bh_result->b_page = NULL;
472
473 /* We set the b_size before reiserfs_get_block call since it is
474 referenced in convert_tail_for_hole() that may be called from
475 reiserfs_get_block() */
476 bh_result->b_size = (1 << inode->i_blkbits);
477
478 ret = reiserfs_get_block(inode, iblock, bh_result,
479 create | GET_BLOCK_NO_DANGLE);
480 if (ret)
481 goto out;
482
483 /* don't allow direct io onto tail pages */
484 if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
485 /* make sure future calls to the direct io funcs for this offset
486 ** in the file fail by unmapping the buffer
487 */
488 clear_buffer_mapped(bh_result);
489 ret = -EINVAL;
490 }
491 /* Possible unpacked tail. Flush the data before pages have
492 disappeared */
493 if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
494 int err;
495
496 reiserfs_write_lock(inode->i_sb);
497
498 err = reiserfs_commit_for_inode(inode);
499 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
500
501 reiserfs_write_unlock(inode->i_sb);
502
503 if (err < 0)
504 ret = err;
505 }
506 out:
507 return ret;
508}
509
510/*
511** helper function for when reiserfs_get_block is called for a hole
512** but the file tail is still in a direct item
513** bh_result is the buffer head for the hole
514** tail_offset is the offset of the start of the tail in the file
515**
516** This calls prepare_write, which will start a new transaction
517** you should not be in a transaction, or have any paths held when you
518** call this.
519*/
520static int convert_tail_for_hole(struct inode *inode,
521 struct buffer_head *bh_result,
522 loff_t tail_offset)
523{
524 unsigned long index;
525 unsigned long tail_end;
526 unsigned long tail_start;
527 struct page *tail_page;
528 struct page *hole_page = bh_result->b_page;
529 int retval = 0;
530
531 if ((tail_offset & (bh_result->b_size - 1)) != 1)
532 return -EIO;
533
534 /* always try to read until the end of the block */
535 tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
536 tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
537
538 index = tail_offset >> PAGE_CACHE_SHIFT;
539 /* hole_page can be zero in case of direct_io, we are sure
540 that we cannot get here if we write with O_DIRECT into
541 tail page */
542 if (!hole_page || index != hole_page->index) {
543 tail_page = grab_cache_page(inode->i_mapping, index);
544 retval = -ENOMEM;
545 if (!tail_page) {
546 goto out;
547 }
548 } else {
549 tail_page = hole_page;
550 }
551
552 /* we don't have to make sure the conversion did not happen while
553 ** we were locking the page because anyone that could convert
554 ** must first take i_mutex.
555 **
556 ** We must fix the tail page for writing because it might have buffers
557 ** that are mapped, but have a block number of 0. This indicates tail
558 ** data that has been read directly into the page, and
559 ** __block_write_begin won't trigger a get_block in this case.
560 */
561 fix_tail_page_for_writing(tail_page);
562 retval = __reiserfs_write_begin(tail_page, tail_start,
563 tail_end - tail_start);
564 if (retval)
565 goto unlock;
566
567 /* tail conversion might change the data in the page */
568 flush_dcache_page(tail_page);
569
570 retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
571
572 unlock:
573 if (tail_page != hole_page) {
574 unlock_page(tail_page);
575 page_cache_release(tail_page);
576 }
577 out:
578 return retval;
579}
580
581static inline int _allocate_block(struct reiserfs_transaction_handle *th,
582 sector_t block,
583 struct inode *inode,
584 b_blocknr_t * allocated_block_nr,
585 struct treepath *path, int flags)
586{
587 BUG_ON(!th->t_trans_id);
588
589#ifdef REISERFS_PREALLOCATE
590 if (!(flags & GET_BLOCK_NO_IMUX)) {
591 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
592 path, block);
593 }
594#endif
595 return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
596 block);
597}
598
599int reiserfs_get_block(struct inode *inode, sector_t block,
600 struct buffer_head *bh_result, int create)
601{
602 int repeat, retval = 0;
603 b_blocknr_t allocated_block_nr = 0; // b_blocknr_t is (unsigned) 32 bit int
604 INITIALIZE_PATH(path);
605 int pos_in_item;
606 struct cpu_key key;
607 struct buffer_head *bh, *unbh = NULL;
608 struct item_head *ih, tmp_ih;
609 __le32 *item;
610 int done;
611 int fs_gen;
612 int lock_depth;
613 struct reiserfs_transaction_handle *th = NULL;
614 /* space reserved in transaction batch:
615 . 3 balancings in direct->indirect conversion
616 . 1 block involved into reiserfs_update_sd()
617 XXX in practically impossible worst case direct2indirect()
618 can incur (much) more than 3 balancings.
619 quota update for user, group */
620 int jbegin_count =
621 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
622 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
623 int version;
624 int dangle = 1;
625 loff_t new_offset =
626 (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
627
628 lock_depth = reiserfs_write_lock_once(inode->i_sb);
629 version = get_inode_item_key_version(inode);
630
631 if (!file_capable(inode, block)) {
632 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
633 return -EFBIG;
634 }
635
636 /* if !create, we aren't changing the FS, so we don't need to
637 ** log anything, so we don't need to start a transaction
638 */
639 if (!(create & GET_BLOCK_CREATE)) {
640 int ret;
641 /* find number of block-th logical block of the file */
642 ret = _get_block_create_0(inode, block, bh_result,
643 create | GET_BLOCK_READ_DIRECT);
644 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
645 return ret;
646 }
647 /*
648 * if we're already in a transaction, make sure to close
649 * any new transactions we start in this func
650 */
651 if ((create & GET_BLOCK_NO_DANGLE) ||
652 reiserfs_transaction_running(inode->i_sb))
653 dangle = 0;
654
655 /* If file is of such a size, that it might have a tail and tails are enabled
656 ** we should mark it as possibly needing tail packing on close
657 */
658 if ((have_large_tails(inode->i_sb)
659 && inode->i_size < i_block_size(inode) * 4)
660 || (have_small_tails(inode->i_sb)
661 && inode->i_size < i_block_size(inode)))
662 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
663
664 /* set the key of the first byte in the 'block'-th block of file */
665 make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
666 if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
667 start_trans:
668 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
669 if (!th) {
670 retval = -ENOMEM;
671 goto failure;
672 }
673 reiserfs_update_inode_transaction(inode);
674 }
675 research:
676
677 retval = search_for_position_by_key(inode->i_sb, &key, &path);
678 if (retval == IO_ERROR) {
679 retval = -EIO;
680 goto failure;
681 }
682
683 bh = get_last_bh(&path);
684 ih = get_ih(&path);
685 item = get_item(&path);
686 pos_in_item = path.pos_in_item;
687
688 fs_gen = get_generation(inode->i_sb);
689 copy_item_head(&tmp_ih, ih);
690
691 if (allocation_needed
692 (retval, allocated_block_nr, ih, item, pos_in_item)) {
693 /* we have to allocate block for the unformatted node */
694 if (!th) {
695 pathrelse(&path);
696 goto start_trans;
697 }
698
699 repeat =
700 _allocate_block(th, block, inode, &allocated_block_nr,
701 &path, create);
702
703 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
704 /* restart the transaction to give the journal a chance to free
705 ** some blocks. releases the path, so we have to go back to
706 ** research if we succeed on the second try
707 */
708 SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
709 retval = restart_transaction(th, inode, &path);
710 if (retval)
711 goto failure;
712 repeat =
713 _allocate_block(th, block, inode,
714 &allocated_block_nr, NULL, create);
715
716 if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
717 goto research;
718 }
719 if (repeat == QUOTA_EXCEEDED)
720 retval = -EDQUOT;
721 else
722 retval = -ENOSPC;
723 goto failure;
724 }
725
726 if (fs_changed(fs_gen, inode->i_sb)
727 && item_moved(&tmp_ih, &path)) {
728 goto research;
729 }
730 }
731
732 if (indirect_item_found(retval, ih)) {
733 b_blocknr_t unfm_ptr;
734 /* 'block'-th block is in the file already (there is
735 corresponding cell in some indirect item). But it may be
736 zero unformatted node pointer (hole) */
737 unfm_ptr = get_block_num(item, pos_in_item);
738 if (unfm_ptr == 0) {
739 /* use allocated block to plug the hole */
740 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
741 if (fs_changed(fs_gen, inode->i_sb)
742 && item_moved(&tmp_ih, &path)) {
743 reiserfs_restore_prepared_buffer(inode->i_sb,
744 bh);
745 goto research;
746 }
747 set_buffer_new(bh_result);
748 if (buffer_dirty(bh_result)
749 && reiserfs_data_ordered(inode->i_sb))
750 reiserfs_add_ordered_list(inode, bh_result);
751 put_block_num(item, pos_in_item, allocated_block_nr);
752 unfm_ptr = allocated_block_nr;
753 journal_mark_dirty(th, inode->i_sb, bh);
754 reiserfs_update_sd(th, inode);
755 }
756 set_block_dev_mapped(bh_result, unfm_ptr, inode);
757 pathrelse(&path);
758 retval = 0;
759 if (!dangle && th)
760 retval = reiserfs_end_persistent_transaction(th);
761
762 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
763
764 /* the item was found, so new blocks were not added to the file
765 ** there is no need to make sure the inode is updated with this
766 ** transaction
767 */
768 return retval;
769 }
770
771 if (!th) {
772 pathrelse(&path);
773 goto start_trans;
774 }
775
776 /* desired position is not found or is in the direct item. We have
777 to append file with holes up to 'block'-th block converting
778 direct items to indirect one if necessary */
779 done = 0;
780 do {
781 if (is_statdata_le_ih(ih)) {
782 __le32 unp = 0;
783 struct cpu_key tmp_key;
784
785 /* indirect item has to be inserted */
786 make_le_item_head(&tmp_ih, &key, version, 1,
787 TYPE_INDIRECT, UNFM_P_SIZE,
788 0 /* free_space */ );
789
790 if (cpu_key_k_offset(&key) == 1) {
791 /* we are going to add 'block'-th block to the file. Use
792 allocated block for that */
793 unp = cpu_to_le32(allocated_block_nr);
794 set_block_dev_mapped(bh_result,
795 allocated_block_nr, inode);
796 set_buffer_new(bh_result);
797 done = 1;
798 }
799 tmp_key = key; // ;)
800 set_cpu_key_k_offset(&tmp_key, 1);
801 PATH_LAST_POSITION(&path)++;
802
803 retval =
804 reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
805 inode, (char *)&unp);
806 if (retval) {
807 reiserfs_free_block(th, inode,
808 allocated_block_nr, 1);
809 goto failure; // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
810 }
811 //mark_tail_converted (inode);
812 } else if (is_direct_le_ih(ih)) {
813 /* direct item has to be converted */
814 loff_t tail_offset;
815
816 tail_offset =
817 ((le_ih_k_offset(ih) -
818 1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
819 if (tail_offset == cpu_key_k_offset(&key)) {
820 /* direct item we just found fits into block we have
821 to map. Convert it into unformatted node: use
822 bh_result for the conversion */
823 set_block_dev_mapped(bh_result,
824 allocated_block_nr, inode);
825 unbh = bh_result;
826 done = 1;
827 } else {
828 /* we have to padd file tail stored in direct item(s)
829 up to block size and convert it to unformatted
830 node. FIXME: this should also get into page cache */
831
832 pathrelse(&path);
833 /*
834 * ugly, but we can only end the transaction if
835 * we aren't nested
836 */
837 BUG_ON(!th->t_refcount);
838 if (th->t_refcount == 1) {
839 retval =
840 reiserfs_end_persistent_transaction
841 (th);
842 th = NULL;
843 if (retval)
844 goto failure;
845 }
846
847 retval =
848 convert_tail_for_hole(inode, bh_result,
849 tail_offset);
850 if (retval) {
851 if (retval != -ENOSPC)
852 reiserfs_error(inode->i_sb,
853 "clm-6004",
854 "convert tail failed "
855 "inode %lu, error %d",
856 inode->i_ino,
857 retval);
858 if (allocated_block_nr) {
859 /* the bitmap, the super, and the stat data == 3 */
860 if (!th)
861 th = reiserfs_persistent_transaction(inode->i_sb, 3);
862 if (th)
863 reiserfs_free_block(th,
864 inode,
865 allocated_block_nr,
866 1);
867 }
868 goto failure;
869 }
870 goto research;
871 }
872 retval =
873 direct2indirect(th, inode, &path, unbh,
874 tail_offset);
875 if (retval) {
876 reiserfs_unmap_buffer(unbh);
877 reiserfs_free_block(th, inode,
878 allocated_block_nr, 1);
879 goto failure;
880 }
881 /* it is important the set_buffer_uptodate is done after
882 ** the direct2indirect. The buffer might contain valid
883 ** data newer than the data on disk (read by readpage, changed,
884 ** and then sent here by writepage). direct2indirect needs
885 ** to know if unbh was already up to date, so it can decide
886 ** if the data in unbh needs to be replaced with data from
887 ** the disk
888 */
889 set_buffer_uptodate(unbh);
890
891 /* unbh->b_page == NULL in case of DIRECT_IO request, this means
892 buffer will disappear shortly, so it should not be added to
893 */
894 if (unbh->b_page) {
895 /* we've converted the tail, so we must
896 ** flush unbh before the transaction commits
897 */
898 reiserfs_add_tail_list(inode, unbh);
899
900 /* mark it dirty now to prevent commit_write from adding
901 ** this buffer to the inode's dirty buffer list
902 */
903 /*
904 * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
905 * It's still atomic, but it sets the page dirty too,
906 * which makes it eligible for writeback at any time by the
907 * VM (which was also the case with __mark_buffer_dirty())
908 */
909 mark_buffer_dirty(unbh);
910 }
911 } else {
912 /* append indirect item with holes if needed, when appending
913 pointer to 'block'-th block use block, which is already
914 allocated */
915 struct cpu_key tmp_key;
916 unp_t unf_single = 0; // We use this in case we need to allocate only
917 // one block which is a fastpath
918 unp_t *un;
919 __u64 max_to_insert =
920 MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
921 UNFM_P_SIZE;
922 __u64 blocks_needed;
923
924 RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
925 "vs-804: invalid position for append");
926 /* indirect item has to be appended, set up key of that position */
927 make_cpu_key(&tmp_key, inode,
928 le_key_k_offset(version,
929 &(ih->ih_key)) +
930 op_bytes_number(ih,
931 inode->i_sb->s_blocksize),
932 //pos_in_item * inode->i_sb->s_blocksize,
933 TYPE_INDIRECT, 3); // key type is unimportant
934
935 RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
936 "green-805: invalid offset");
937 blocks_needed =
938 1 +
939 ((cpu_key_k_offset(&key) -
940 cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
941 s_blocksize_bits);
942
943 if (blocks_needed == 1) {
944 un = &unf_single;
945 } else {
946 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_NOFS);
947 if (!un) {
948 un = &unf_single;
949 blocks_needed = 1;
950 max_to_insert = 0;
951 }
952 }
953 if (blocks_needed <= max_to_insert) {
954 /* we are going to add target block to the file. Use allocated
955 block for that */
956 un[blocks_needed - 1] =
957 cpu_to_le32(allocated_block_nr);
958 set_block_dev_mapped(bh_result,
959 allocated_block_nr, inode);
960 set_buffer_new(bh_result);
961 done = 1;
962 } else {
963 /* paste hole to the indirect item */
964 /* If kmalloc failed, max_to_insert becomes zero and it means we
965 only have space for one block */
966 blocks_needed =
967 max_to_insert ? max_to_insert : 1;
968 }
969 retval =
970 reiserfs_paste_into_item(th, &path, &tmp_key, inode,
971 (char *)un,
972 UNFM_P_SIZE *
973 blocks_needed);
974
975 if (blocks_needed != 1)
976 kfree(un);
977
978 if (retval) {
979 reiserfs_free_block(th, inode,
980 allocated_block_nr, 1);
981 goto failure;
982 }
983 if (!done) {
984 /* We need to mark new file size in case this function will be
985 interrupted/aborted later on. And we may do this only for
986 holes. */
987 inode->i_size +=
988 inode->i_sb->s_blocksize * blocks_needed;
989 }
990 }
991
992 if (done == 1)
993 break;
994
995 /* this loop could log more blocks than we had originally asked
996 ** for. So, we have to allow the transaction to end if it is
997 ** too big or too full. Update the inode so things are
998 ** consistent if we crash before the function returns
999 **
1000 ** release the path so that anybody waiting on the path before
1001 ** ending their transaction will be able to continue.
1002 */
1003 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1004 retval = restart_transaction(th, inode, &path);
1005 if (retval)
1006 goto failure;
1007 }
1008 /*
1009 * inserting indirect pointers for a hole can take a
1010 * long time. reschedule if needed and also release the write
1011 * lock for others.
1012 */
1013 if (need_resched()) {
1014 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1015 schedule();
1016 lock_depth = reiserfs_write_lock_once(inode->i_sb);
1017 }
1018
1019 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1020 if (retval == IO_ERROR) {
1021 retval = -EIO;
1022 goto failure;
1023 }
1024 if (retval == POSITION_FOUND) {
1025 reiserfs_warning(inode->i_sb, "vs-825",
1026 "%K should not be found", &key);
1027 retval = -EEXIST;
1028 if (allocated_block_nr)
1029 reiserfs_free_block(th, inode,
1030 allocated_block_nr, 1);
1031 pathrelse(&path);
1032 goto failure;
1033 }
1034 bh = get_last_bh(&path);
1035 ih = get_ih(&path);
1036 item = get_item(&path);
1037 pos_in_item = path.pos_in_item;
1038 } while (1);
1039
1040 retval = 0;
1041
1042 failure:
1043 if (th && (!dangle || (retval && !th->t_trans_id))) {
1044 int err;
1045 if (th->t_trans_id)
1046 reiserfs_update_sd(th, inode);
1047 err = reiserfs_end_persistent_transaction(th);
1048 if (err)
1049 retval = err;
1050 }
1051
1052 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
1053 reiserfs_check_path(&path);
1054 return retval;
1055}
1056
1057static int
1058reiserfs_readpages(struct file *file, struct address_space *mapping,
1059 struct list_head *pages, unsigned nr_pages)
1060{
1061 return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1062}
1063
1064/* Compute real number of used bytes by file
1065 * Following three functions can go away when we'll have enough space in stat item
1066 */
1067static int real_space_diff(struct inode *inode, int sd_size)
1068{
1069 int bytes;
1070 loff_t blocksize = inode->i_sb->s_blocksize;
1071
1072 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1073 return sd_size;
1074
1075 /* End of file is also in full block with indirect reference, so round
1076 ** up to the next block.
1077 **
1078 ** there is just no way to know if the tail is actually packed
1079 ** on the file, so we have to assume it isn't. When we pack the
1080 ** tail, we add 4 bytes to pretend there really is an unformatted
1081 ** node pointer
1082 */
1083 bytes =
1084 ((inode->i_size +
1085 (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1086 sd_size;
1087 return bytes;
1088}
1089
1090static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1091 int sd_size)
1092{
1093 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1094 return inode->i_size +
1095 (loff_t) (real_space_diff(inode, sd_size));
1096 }
1097 return ((loff_t) real_space_diff(inode, sd_size)) +
1098 (((loff_t) blocks) << 9);
1099}
1100
1101/* Compute number of blocks used by file in ReiserFS counting */
1102static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1103{
1104 loff_t bytes = inode_get_bytes(inode);
1105 loff_t real_space = real_space_diff(inode, sd_size);
1106
1107 /* keeps fsck and non-quota versions of reiserfs happy */
1108 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1109 bytes += (loff_t) 511;
1110 }
1111
1112 /* files from before the quota patch might i_blocks such that
1113 ** bytes < real_space. Deal with that here to prevent it from
1114 ** going negative.
1115 */
1116 if (bytes < real_space)
1117 return 0;
1118 return (bytes - real_space) >> 9;
1119}
1120
1121//
1122// BAD: new directories have stat data of new type and all other items
1123// of old type. Version stored in the inode says about body items, so
1124// in update_stat_data we can not rely on inode, but have to check
1125// item version directly
1126//
1127
1128// called by read_locked_inode
1129static void init_inode(struct inode *inode, struct treepath *path)
1130{
1131 struct buffer_head *bh;
1132 struct item_head *ih;
1133 __u32 rdev;
1134 //int version = ITEM_VERSION_1;
1135
1136 bh = PATH_PLAST_BUFFER(path);
1137 ih = PATH_PITEM_HEAD(path);
1138
1139 copy_key(INODE_PKEY(inode), &(ih->ih_key));
1140
1141 INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1142 REISERFS_I(inode)->i_flags = 0;
1143 REISERFS_I(inode)->i_prealloc_block = 0;
1144 REISERFS_I(inode)->i_prealloc_count = 0;
1145 REISERFS_I(inode)->i_trans_id = 0;
1146 REISERFS_I(inode)->i_jl = NULL;
1147 reiserfs_init_xattr_rwsem(inode);
1148
1149 if (stat_data_v1(ih)) {
1150 struct stat_data_v1 *sd =
1151 (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1152 unsigned long blocks;
1153
1154 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1155 set_inode_sd_version(inode, STAT_DATA_V1);
1156 inode->i_mode = sd_v1_mode(sd);
1157 inode->i_nlink = sd_v1_nlink(sd);
1158 inode->i_uid = sd_v1_uid(sd);
1159 inode->i_gid = sd_v1_gid(sd);
1160 inode->i_size = sd_v1_size(sd);
1161 inode->i_atime.tv_sec = sd_v1_atime(sd);
1162 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1163 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1164 inode->i_atime.tv_nsec = 0;
1165 inode->i_ctime.tv_nsec = 0;
1166 inode->i_mtime.tv_nsec = 0;
1167
1168 inode->i_blocks = sd_v1_blocks(sd);
1169 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1170 blocks = (inode->i_size + 511) >> 9;
1171 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1172 if (inode->i_blocks > blocks) {
1173 // there was a bug in <=3.5.23 when i_blocks could take negative
1174 // values. Starting from 3.5.17 this value could even be stored in
1175 // stat data. For such files we set i_blocks based on file
1176 // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1177 // only updated if file's inode will ever change
1178 inode->i_blocks = blocks;
1179 }
1180
1181 rdev = sd_v1_rdev(sd);
1182 REISERFS_I(inode)->i_first_direct_byte =
1183 sd_v1_first_direct_byte(sd);
1184 /* an early bug in the quota code can give us an odd number for the
1185 ** block count. This is incorrect, fix it here.
1186 */
1187 if (inode->i_blocks & 1) {
1188 inode->i_blocks++;
1189 }
1190 inode_set_bytes(inode,
1191 to_real_used_space(inode, inode->i_blocks,
1192 SD_V1_SIZE));
1193 /* nopack is initially zero for v1 objects. For v2 objects,
1194 nopack is initialised from sd_attrs */
1195 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1196 } else {
1197 // new stat data found, but object may have old items
1198 // (directories and symlinks)
1199 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1200
1201 inode->i_mode = sd_v2_mode(sd);
1202 inode->i_nlink = sd_v2_nlink(sd);
1203 inode->i_uid = sd_v2_uid(sd);
1204 inode->i_size = sd_v2_size(sd);
1205 inode->i_gid = sd_v2_gid(sd);
1206 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1207 inode->i_atime.tv_sec = sd_v2_atime(sd);
1208 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1209 inode->i_ctime.tv_nsec = 0;
1210 inode->i_mtime.tv_nsec = 0;
1211 inode->i_atime.tv_nsec = 0;
1212 inode->i_blocks = sd_v2_blocks(sd);
1213 rdev = sd_v2_rdev(sd);
1214 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1215 inode->i_generation =
1216 le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1217 else
1218 inode->i_generation = sd_v2_generation(sd);
1219
1220 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1221 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1222 else
1223 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1224 REISERFS_I(inode)->i_first_direct_byte = 0;
1225 set_inode_sd_version(inode, STAT_DATA_V2);
1226 inode_set_bytes(inode,
1227 to_real_used_space(inode, inode->i_blocks,
1228 SD_V2_SIZE));
1229 /* read persistent inode attributes from sd and initialise
1230 generic inode flags from them */
1231 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1232 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1233 }
1234
1235 pathrelse(path);
1236 if (S_ISREG(inode->i_mode)) {
1237 inode->i_op = &reiserfs_file_inode_operations;
1238 inode->i_fop = &reiserfs_file_operations;
1239 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1240 } else if (S_ISDIR(inode->i_mode)) {
1241 inode->i_op = &reiserfs_dir_inode_operations;
1242 inode->i_fop = &reiserfs_dir_operations;
1243 } else if (S_ISLNK(inode->i_mode)) {
1244 inode->i_op = &reiserfs_symlink_inode_operations;
1245 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1246 } else {
1247 inode->i_blocks = 0;
1248 inode->i_op = &reiserfs_special_inode_operations;
1249 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1250 }
1251}
1252
1253// update new stat data with inode fields
1254static void inode2sd(void *sd, struct inode *inode, loff_t size)
1255{
1256 struct stat_data *sd_v2 = (struct stat_data *)sd;
1257 __u16 flags;
1258
1259 set_sd_v2_mode(sd_v2, inode->i_mode);
1260 set_sd_v2_nlink(sd_v2, inode->i_nlink);
1261 set_sd_v2_uid(sd_v2, inode->i_uid);
1262 set_sd_v2_size(sd_v2, size);
1263 set_sd_v2_gid(sd_v2, inode->i_gid);
1264 set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1265 set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1266 set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1267 set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1268 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1269 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1270 else
1271 set_sd_v2_generation(sd_v2, inode->i_generation);
1272 flags = REISERFS_I(inode)->i_attrs;
1273 i_attrs_to_sd_attrs(inode, &flags);
1274 set_sd_v2_attrs(sd_v2, flags);
1275}
1276
1277// used to copy inode's fields to old stat data
1278static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1279{
1280 struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1281
1282 set_sd_v1_mode(sd_v1, inode->i_mode);
1283 set_sd_v1_uid(sd_v1, inode->i_uid);
1284 set_sd_v1_gid(sd_v1, inode->i_gid);
1285 set_sd_v1_nlink(sd_v1, inode->i_nlink);
1286 set_sd_v1_size(sd_v1, size);
1287 set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1288 set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1289 set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1290
1291 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1292 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1293 else
1294 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1295
1296 // Sigh. i_first_direct_byte is back
1297 set_sd_v1_first_direct_byte(sd_v1,
1298 REISERFS_I(inode)->i_first_direct_byte);
1299}
1300
1301/* NOTE, you must prepare the buffer head before sending it here,
1302** and then log it after the call
1303*/
1304static void update_stat_data(struct treepath *path, struct inode *inode,
1305 loff_t size)
1306{
1307 struct buffer_head *bh;
1308 struct item_head *ih;
1309
1310 bh = PATH_PLAST_BUFFER(path);
1311 ih = PATH_PITEM_HEAD(path);
1312
1313 if (!is_statdata_le_ih(ih))
1314 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1315 INODE_PKEY(inode), ih);
1316
1317 if (stat_data_v1(ih)) {
1318 // path points to old stat data
1319 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1320 } else {
1321 inode2sd(B_I_PITEM(bh, ih), inode, size);
1322 }
1323
1324 return;
1325}
1326
1327void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1328 struct inode *inode, loff_t size)
1329{
1330 struct cpu_key key;
1331 INITIALIZE_PATH(path);
1332 struct buffer_head *bh;
1333 int fs_gen;
1334 struct item_head *ih, tmp_ih;
1335 int retval;
1336
1337 BUG_ON(!th->t_trans_id);
1338
1339 make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3); //key type is unimportant
1340
1341 for (;;) {
1342 int pos;
1343 /* look for the object's stat data */
1344 retval = search_item(inode->i_sb, &key, &path);
1345 if (retval == IO_ERROR) {
1346 reiserfs_error(inode->i_sb, "vs-13050",
1347 "i/o failure occurred trying to "
1348 "update %K stat data", &key);
1349 return;
1350 }
1351 if (retval == ITEM_NOT_FOUND) {
1352 pos = PATH_LAST_POSITION(&path);
1353 pathrelse(&path);
1354 if (inode->i_nlink == 0) {
1355 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1356 return;
1357 }
1358 reiserfs_warning(inode->i_sb, "vs-13060",
1359 "stat data of object %k (nlink == %d) "
1360 "not found (pos %d)",
1361 INODE_PKEY(inode), inode->i_nlink,
1362 pos);
1363 reiserfs_check_path(&path);
1364 return;
1365 }
1366
1367 /* sigh, prepare_for_journal might schedule. When it schedules the
1368 ** FS might change. We have to detect that, and loop back to the
1369 ** search if the stat data item has moved
1370 */
1371 bh = get_last_bh(&path);
1372 ih = get_ih(&path);
1373 copy_item_head(&tmp_ih, ih);
1374 fs_gen = get_generation(inode->i_sb);
1375 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1376 if (fs_changed(fs_gen, inode->i_sb)
1377 && item_moved(&tmp_ih, &path)) {
1378 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1379 continue; /* Stat_data item has been moved after scheduling. */
1380 }
1381 break;
1382 }
1383 update_stat_data(&path, inode, size);
1384 journal_mark_dirty(th, th->t_super, bh);
1385 pathrelse(&path);
1386 return;
1387}
1388
1389/* reiserfs_read_locked_inode is called to read the inode off disk, and it
1390** does a make_bad_inode when things go wrong. But, we need to make sure
1391** and clear the key in the private portion of the inode, otherwise a
1392** corresponding iput might try to delete whatever object the inode last
1393** represented.
1394*/
1395static void reiserfs_make_bad_inode(struct inode *inode)
1396{
1397 memset(INODE_PKEY(inode), 0, KEY_SIZE);
1398 make_bad_inode(inode);
1399}
1400
1401//
1402// initially this function was derived from minix or ext2's analog and
1403// evolved as the prototype did
1404//
1405
1406int reiserfs_init_locked_inode(struct inode *inode, void *p)
1407{
1408 struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1409 inode->i_ino = args->objectid;
1410 INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1411 return 0;
1412}
1413
1414/* looks for stat data in the tree, and fills up the fields of in-core
1415 inode stat data fields */
1416void reiserfs_read_locked_inode(struct inode *inode,
1417 struct reiserfs_iget_args *args)
1418{
1419 INITIALIZE_PATH(path_to_sd);
1420 struct cpu_key key;
1421 unsigned long dirino;
1422 int retval;
1423
1424 dirino = args->dirid;
1425
1426 /* set version 1, version 2 could be used too, because stat data
1427 key is the same in both versions */
1428 key.version = KEY_FORMAT_3_5;
1429 key.on_disk_key.k_dir_id = dirino;
1430 key.on_disk_key.k_objectid = inode->i_ino;
1431 key.on_disk_key.k_offset = 0;
1432 key.on_disk_key.k_type = 0;
1433
1434 /* look for the object's stat data */
1435 retval = search_item(inode->i_sb, &key, &path_to_sd);
1436 if (retval == IO_ERROR) {
1437 reiserfs_error(inode->i_sb, "vs-13070",
1438 "i/o failure occurred trying to find "
1439 "stat data of %K", &key);
1440 reiserfs_make_bad_inode(inode);
1441 return;
1442 }
1443 if (retval != ITEM_FOUND) {
1444 /* a stale NFS handle can trigger this without it being an error */
1445 pathrelse(&path_to_sd);
1446 reiserfs_make_bad_inode(inode);
1447 inode->i_nlink = 0;
1448 return;
1449 }
1450
1451 init_inode(inode, &path_to_sd);
1452
1453 /* It is possible that knfsd is trying to access inode of a file
1454 that is being removed from the disk by some other thread. As we
1455 update sd on unlink all that is required is to check for nlink
1456 here. This bug was first found by Sizif when debugging
1457 SquidNG/Butterfly, forgotten, and found again after Philippe
1458 Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1459
1460 More logical fix would require changes in fs/inode.c:iput() to
1461 remove inode from hash-table _after_ fs cleaned disk stuff up and
1462 in iget() to return NULL if I_FREEING inode is found in
1463 hash-table. */
1464 /* Currently there is one place where it's ok to meet inode with
1465 nlink==0: processing of open-unlinked and half-truncated files
1466 during mount (fs/reiserfs/super.c:finish_unfinished()). */
1467 if ((inode->i_nlink == 0) &&
1468 !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1469 reiserfs_warning(inode->i_sb, "vs-13075",
1470 "dead inode read from disk %K. "
1471 "This is likely to be race with knfsd. Ignore",
1472 &key);
1473 reiserfs_make_bad_inode(inode);
1474 }
1475
1476 reiserfs_check_path(&path_to_sd); /* init inode should be relsing */
1477
1478 /*
1479 * Stat data v1 doesn't support ACLs.
1480 */
1481 if (get_inode_sd_version(inode) == STAT_DATA_V1)
1482 cache_no_acl(inode);
1483}
1484
1485/**
1486 * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1487 *
1488 * @inode: inode from hash table to check
1489 * @opaque: "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1490 *
1491 * This function is called by iget5_locked() to distinguish reiserfs inodes
1492 * having the same inode numbers. Such inodes can only exist due to some
1493 * error condition. One of them should be bad. Inodes with identical
1494 * inode numbers (objectids) are distinguished by parent directory ids.
1495 *
1496 */
1497int reiserfs_find_actor(struct inode *inode, void *opaque)
1498{
1499 struct reiserfs_iget_args *args;
1500
1501 args = opaque;
1502 /* args is already in CPU order */
1503 return (inode->i_ino == args->objectid) &&
1504 (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1505}
1506
1507struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1508{
1509 struct inode *inode;
1510 struct reiserfs_iget_args args;
1511
1512 args.objectid = key->on_disk_key.k_objectid;
1513 args.dirid = key->on_disk_key.k_dir_id;
1514 reiserfs_write_unlock(s);
1515 inode = iget5_locked(s, key->on_disk_key.k_objectid,
1516 reiserfs_find_actor, reiserfs_init_locked_inode,
1517 (void *)(&args));
1518 reiserfs_write_lock(s);
1519 if (!inode)
1520 return ERR_PTR(-ENOMEM);
1521
1522 if (inode->i_state & I_NEW) {
1523 reiserfs_read_locked_inode(inode, &args);
1524 unlock_new_inode(inode);
1525 }
1526
1527 if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1528 /* either due to i/o error or a stale NFS handle */
1529 iput(inode);
1530 inode = NULL;
1531 }
1532 return inode;
1533}
1534
1535static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1536 u32 objectid, u32 dir_id, u32 generation)
1537
1538{
1539 struct cpu_key key;
1540 struct inode *inode;
1541
1542 key.on_disk_key.k_objectid = objectid;
1543 key.on_disk_key.k_dir_id = dir_id;
1544 reiserfs_write_lock(sb);
1545 inode = reiserfs_iget(sb, &key);
1546 if (inode && !IS_ERR(inode) && generation != 0 &&
1547 generation != inode->i_generation) {
1548 iput(inode);
1549 inode = NULL;
1550 }
1551 reiserfs_write_unlock(sb);
1552
1553 return d_obtain_alias(inode);
1554}
1555
1556struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1557 int fh_len, int fh_type)
1558{
1559 /* fhtype happens to reflect the number of u32s encoded.
1560 * due to a bug in earlier code, fhtype might indicate there
1561 * are more u32s then actually fitted.
1562 * so if fhtype seems to be more than len, reduce fhtype.
1563 * Valid types are:
1564 * 2 - objectid + dir_id - legacy support
1565 * 3 - objectid + dir_id + generation
1566 * 4 - objectid + dir_id + objectid and dirid of parent - legacy
1567 * 5 - objectid + dir_id + generation + objectid and dirid of parent
1568 * 6 - as above plus generation of directory
1569 * 6 does not fit in NFSv2 handles
1570 */
1571 if (fh_type > fh_len) {
1572 if (fh_type != 6 || fh_len != 5)
1573 reiserfs_warning(sb, "reiserfs-13077",
1574 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1575 fh_type, fh_len);
1576 fh_type = 5;
1577 }
1578
1579 return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1580 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1581}
1582
1583struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1584 int fh_len, int fh_type)
1585{
1586 if (fh_type < 4)
1587 return NULL;
1588
1589 return reiserfs_get_dentry(sb,
1590 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1591 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1592 (fh_type == 6) ? fid->raw[5] : 0);
1593}
1594
1595int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1596 int need_parent)
1597{
1598 struct inode *inode = dentry->d_inode;
1599 int maxlen = *lenp;
1600
1601 if (need_parent && (maxlen < 5)) {
1602 *lenp = 5;
1603 return 255;
1604 } else if (maxlen < 3) {
1605 *lenp = 3;
1606 return 255;
1607 }
1608
1609 data[0] = inode->i_ino;
1610 data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1611 data[2] = inode->i_generation;
1612 *lenp = 3;
1613 /* no room for directory info? return what we've stored so far */
1614 if (maxlen < 5 || !need_parent)
1615 return 3;
1616
1617 spin_lock(&dentry->d_lock);
1618 inode = dentry->d_parent->d_inode;
1619 data[3] = inode->i_ino;
1620 data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1621 *lenp = 5;
1622 if (maxlen >= 6) {
1623 data[5] = inode->i_generation;
1624 *lenp = 6;
1625 }
1626 spin_unlock(&dentry->d_lock);
1627 return *lenp;
1628}
1629
1630/* looks for stat data, then copies fields to it, marks the buffer
1631 containing stat data as dirty */
1632/* reiserfs inodes are never really dirty, since the dirty inode call
1633** always logs them. This call allows the VFS inode marking routines
1634** to properly mark inodes for datasync and such, but only actually
1635** does something when called for a synchronous update.
1636*/
1637int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1638{
1639 struct reiserfs_transaction_handle th;
1640 int jbegin_count = 1;
1641
1642 if (inode->i_sb->s_flags & MS_RDONLY)
1643 return -EROFS;
1644 /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1645 ** these cases are just when the system needs ram, not when the
1646 ** inode needs to reach disk for safety, and they can safely be
1647 ** ignored because the altered inode has already been logged.
1648 */
1649 if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1650 reiserfs_write_lock(inode->i_sb);
1651 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1652 reiserfs_update_sd(&th, inode);
1653 journal_end_sync(&th, inode->i_sb, jbegin_count);
1654 }
1655 reiserfs_write_unlock(inode->i_sb);
1656 }
1657 return 0;
1658}
1659
1660/* stat data of new object is inserted already, this inserts the item
1661 containing "." and ".." entries */
1662static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1663 struct inode *inode,
1664 struct item_head *ih, struct treepath *path,
1665 struct inode *dir)
1666{
1667 struct super_block *sb = th->t_super;
1668 char empty_dir[EMPTY_DIR_SIZE];
1669 char *body = empty_dir;
1670 struct cpu_key key;
1671 int retval;
1672
1673 BUG_ON(!th->t_trans_id);
1674
1675 _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1676 le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1677 TYPE_DIRENTRY, 3 /*key length */ );
1678
1679 /* compose item head for new item. Directories consist of items of
1680 old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1681 is done by reiserfs_new_inode */
1682 if (old_format_only(sb)) {
1683 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1684 TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1685
1686 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1687 ih->ih_key.k_objectid,
1688 INODE_PKEY(dir)->k_dir_id,
1689 INODE_PKEY(dir)->k_objectid);
1690 } else {
1691 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1692 TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1693
1694 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1695 ih->ih_key.k_objectid,
1696 INODE_PKEY(dir)->k_dir_id,
1697 INODE_PKEY(dir)->k_objectid);
1698 }
1699
1700 /* look for place in the tree for new item */
1701 retval = search_item(sb, &key, path);
1702 if (retval == IO_ERROR) {
1703 reiserfs_error(sb, "vs-13080",
1704 "i/o failure occurred creating new directory");
1705 return -EIO;
1706 }
1707 if (retval == ITEM_FOUND) {
1708 pathrelse(path);
1709 reiserfs_warning(sb, "vs-13070",
1710 "object with this key exists (%k)",
1711 &(ih->ih_key));
1712 return -EEXIST;
1713 }
1714
1715 /* insert item, that is empty directory item */
1716 return reiserfs_insert_item(th, path, &key, ih, inode, body);
1717}
1718
1719/* stat data of object has been inserted, this inserts the item
1720 containing the body of symlink */
1721static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode, /* Inode of symlink */
1722 struct item_head *ih,
1723 struct treepath *path, const char *symname,
1724 int item_len)
1725{
1726 struct super_block *sb = th->t_super;
1727 struct cpu_key key;
1728 int retval;
1729
1730 BUG_ON(!th->t_trans_id);
1731
1732 _make_cpu_key(&key, KEY_FORMAT_3_5,
1733 le32_to_cpu(ih->ih_key.k_dir_id),
1734 le32_to_cpu(ih->ih_key.k_objectid),
1735 1, TYPE_DIRECT, 3 /*key length */ );
1736
1737 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1738 0 /*free_space */ );
1739
1740 /* look for place in the tree for new item */
1741 retval = search_item(sb, &key, path);
1742 if (retval == IO_ERROR) {
1743 reiserfs_error(sb, "vs-13080",
1744 "i/o failure occurred creating new symlink");
1745 return -EIO;
1746 }
1747 if (retval == ITEM_FOUND) {
1748 pathrelse(path);
1749 reiserfs_warning(sb, "vs-13080",
1750 "object with this key exists (%k)",
1751 &(ih->ih_key));
1752 return -EEXIST;
1753 }
1754
1755 /* insert item, that is body of symlink */
1756 return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1757}
1758
1759/* inserts the stat data into the tree, and then calls
1760 reiserfs_new_directory (to insert ".", ".." item if new object is
1761 directory) or reiserfs_new_symlink (to insert symlink body if new
1762 object is symlink) or nothing (if new object is regular file)
1763
1764 NOTE! uid and gid must already be set in the inode. If we return
1765 non-zero due to an error, we have to drop the quota previously allocated
1766 for the fresh inode. This can only be done outside a transaction, so
1767 if we return non-zero, we also end the transaction. */
1768int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1769 struct inode *dir, int mode, const char *symname,
1770 /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1771 strlen (symname) for symlinks) */
1772 loff_t i_size, struct dentry *dentry,
1773 struct inode *inode,
1774 struct reiserfs_security_handle *security)
1775{
1776 struct super_block *sb;
1777 struct reiserfs_iget_args args;
1778 INITIALIZE_PATH(path_to_key);
1779 struct cpu_key key;
1780 struct item_head ih;
1781 struct stat_data sd;
1782 int retval;
1783 int err;
1784
1785 BUG_ON(!th->t_trans_id);
1786
1787 dquot_initialize(inode);
1788 err = dquot_alloc_inode(inode);
1789 if (err)
1790 goto out_end_trans;
1791 if (!dir->i_nlink) {
1792 err = -EPERM;
1793 goto out_bad_inode;
1794 }
1795
1796 sb = dir->i_sb;
1797
1798 /* item head of new item */
1799 ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1800 ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1801 if (!ih.ih_key.k_objectid) {
1802 err = -ENOMEM;
1803 goto out_bad_inode;
1804 }
1805 args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1806 if (old_format_only(sb))
1807 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1808 TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1809 else
1810 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1811 TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1812 memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1813 args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1814 if (insert_inode_locked4(inode, args.objectid,
1815 reiserfs_find_actor, &args) < 0) {
1816 err = -EINVAL;
1817 goto out_bad_inode;
1818 }
1819 if (old_format_only(sb))
1820 /* not a perfect generation count, as object ids can be reused, but
1821 ** this is as good as reiserfs can do right now.
1822 ** note that the private part of inode isn't filled in yet, we have
1823 ** to use the directory.
1824 */
1825 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1826 else
1827#if defined( USE_INODE_GENERATION_COUNTER )
1828 inode->i_generation =
1829 le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1830#else
1831 inode->i_generation = ++event;
1832#endif
1833
1834 /* fill stat data */
1835 inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1836
1837 /* uid and gid must already be set by the caller for quota init */
1838
1839 /* symlink cannot be immutable or append only, right? */
1840 if (S_ISLNK(inode->i_mode))
1841 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1842
1843 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1844 inode->i_size = i_size;
1845 inode->i_blocks = 0;
1846 inode->i_bytes = 0;
1847 REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1848 U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1849
1850 INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1851 REISERFS_I(inode)->i_flags = 0;
1852 REISERFS_I(inode)->i_prealloc_block = 0;
1853 REISERFS_I(inode)->i_prealloc_count = 0;
1854 REISERFS_I(inode)->i_trans_id = 0;
1855 REISERFS_I(inode)->i_jl = NULL;
1856 REISERFS_I(inode)->i_attrs =
1857 REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1858 sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1859 reiserfs_init_xattr_rwsem(inode);
1860
1861 /* key to search for correct place for new stat data */
1862 _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1863 le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1864 TYPE_STAT_DATA, 3 /*key length */ );
1865
1866 /* find proper place for inserting of stat data */
1867 retval = search_item(sb, &key, &path_to_key);
1868 if (retval == IO_ERROR) {
1869 err = -EIO;
1870 goto out_bad_inode;
1871 }
1872 if (retval == ITEM_FOUND) {
1873 pathrelse(&path_to_key);
1874 err = -EEXIST;
1875 goto out_bad_inode;
1876 }
1877 if (old_format_only(sb)) {
1878 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1879 pathrelse(&path_to_key);
1880 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1881 err = -EINVAL;
1882 goto out_bad_inode;
1883 }
1884 inode2sd_v1(&sd, inode, inode->i_size);
1885 } else {
1886 inode2sd(&sd, inode, inode->i_size);
1887 }
1888 // store in in-core inode the key of stat data and version all
1889 // object items will have (directory items will have old offset
1890 // format, other new objects will consist of new items)
1891 if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1892 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1893 else
1894 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1895 if (old_format_only(sb))
1896 set_inode_sd_version(inode, STAT_DATA_V1);
1897 else
1898 set_inode_sd_version(inode, STAT_DATA_V2);
1899
1900 /* insert the stat data into the tree */
1901#ifdef DISPLACE_NEW_PACKING_LOCALITIES
1902 if (REISERFS_I(dir)->new_packing_locality)
1903 th->displace_new_blocks = 1;
1904#endif
1905 retval =
1906 reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1907 (char *)(&sd));
1908 if (retval) {
1909 err = retval;
1910 reiserfs_check_path(&path_to_key);
1911 goto out_bad_inode;
1912 }
1913#ifdef DISPLACE_NEW_PACKING_LOCALITIES
1914 if (!th->displace_new_blocks)
1915 REISERFS_I(dir)->new_packing_locality = 0;
1916#endif
1917 if (S_ISDIR(mode)) {
1918 /* insert item with "." and ".." */
1919 retval =
1920 reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1921 }
1922
1923 if (S_ISLNK(mode)) {
1924 /* insert body of symlink */
1925 if (!old_format_only(sb))
1926 i_size = ROUND_UP(i_size);
1927 retval =
1928 reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1929 i_size);
1930 }
1931 if (retval) {
1932 err = retval;
1933 reiserfs_check_path(&path_to_key);
1934 journal_end(th, th->t_super, th->t_blocks_allocated);
1935 goto out_inserted_sd;
1936 }
1937
1938 if (reiserfs_posixacl(inode->i_sb)) {
1939 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
1940 if (retval) {
1941 err = retval;
1942 reiserfs_check_path(&path_to_key);
1943 journal_end(th, th->t_super, th->t_blocks_allocated);
1944 goto out_inserted_sd;
1945 }
1946 } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1947 reiserfs_warning(inode->i_sb, "jdm-13090",
1948 "ACLs aren't enabled in the fs, "
1949 "but vfs thinks they are!");
1950 } else if (IS_PRIVATE(dir))
1951 inode->i_flags |= S_PRIVATE;
1952
1953 if (security->name) {
1954 retval = reiserfs_security_write(th, inode, security);
1955 if (retval) {
1956 err = retval;
1957 reiserfs_check_path(&path_to_key);
1958 retval = journal_end(th, th->t_super,
1959 th->t_blocks_allocated);
1960 if (retval)
1961 err = retval;
1962 goto out_inserted_sd;
1963 }
1964 }
1965
1966 reiserfs_update_sd(th, inode);
1967 reiserfs_check_path(&path_to_key);
1968
1969 return 0;
1970
1971/* it looks like you can easily compress these two goto targets into
1972 * one. Keeping it like this doesn't actually hurt anything, and they
1973 * are place holders for what the quota code actually needs.
1974 */
1975 out_bad_inode:
1976 /* Invalidate the object, nothing was inserted yet */
1977 INODE_PKEY(inode)->k_objectid = 0;
1978
1979 /* Quota change must be inside a transaction for journaling */
1980 dquot_free_inode(inode);
1981
1982 out_end_trans:
1983 journal_end(th, th->t_super, th->t_blocks_allocated);
1984 /* Drop can be outside and it needs more credits so it's better to have it outside */
1985 dquot_drop(inode);
1986 inode->i_flags |= S_NOQUOTA;
1987 make_bad_inode(inode);
1988
1989 out_inserted_sd:
1990 inode->i_nlink = 0;
1991 th->t_trans_id = 0; /* so the caller can't use this handle later */
1992 unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1993 iput(inode);
1994 return err;
1995}
1996
1997/*
1998** finds the tail page in the page cache,
1999** reads the last block in.
2000**
2001** On success, page_result is set to a locked, pinned page, and bh_result
2002** is set to an up to date buffer for the last block in the file. returns 0.
2003**
2004** tail conversion is not done, so bh_result might not be valid for writing
2005** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2006** trying to write the block.
2007**
2008** on failure, nonzero is returned, page_result and bh_result are untouched.
2009*/
2010static int grab_tail_page(struct inode *inode,
2011 struct page **page_result,
2012 struct buffer_head **bh_result)
2013{
2014
2015 /* we want the page with the last byte in the file,
2016 ** not the page that will hold the next byte for appending
2017 */
2018 unsigned long index = (inode->i_size - 1) >> PAGE_CACHE_SHIFT;
2019 unsigned long pos = 0;
2020 unsigned long start = 0;
2021 unsigned long blocksize = inode->i_sb->s_blocksize;
2022 unsigned long offset = (inode->i_size) & (PAGE_CACHE_SIZE - 1);
2023 struct buffer_head *bh;
2024 struct buffer_head *head;
2025 struct page *page;
2026 int error;
2027
2028 /* we know that we are only called with inode->i_size > 0.
2029 ** we also know that a file tail can never be as big as a block
2030 ** If i_size % blocksize == 0, our file is currently block aligned
2031 ** and it won't need converting or zeroing after a truncate.
2032 */
2033 if ((offset & (blocksize - 1)) == 0) {
2034 return -ENOENT;
2035 }
2036 page = grab_cache_page(inode->i_mapping, index);
2037 error = -ENOMEM;
2038 if (!page) {
2039 goto out;
2040 }
2041 /* start within the page of the last block in the file */
2042 start = (offset / blocksize) * blocksize;
2043
2044 error = __block_write_begin(page, start, offset - start,
2045 reiserfs_get_block_create_0);
2046 if (error)
2047 goto unlock;
2048
2049 head = page_buffers(page);
2050 bh = head;
2051 do {
2052 if (pos >= start) {
2053 break;
2054 }
2055 bh = bh->b_this_page;
2056 pos += blocksize;
2057 } while (bh != head);
2058
2059 if (!buffer_uptodate(bh)) {
2060 /* note, this should never happen, prepare_write should
2061 ** be taking care of this for us. If the buffer isn't up to date,
2062 ** I've screwed up the code to find the buffer, or the code to
2063 ** call prepare_write
2064 */
2065 reiserfs_error(inode->i_sb, "clm-6000",
2066 "error reading block %lu", bh->b_blocknr);
2067 error = -EIO;
2068 goto unlock;
2069 }
2070 *bh_result = bh;
2071 *page_result = page;
2072
2073 out:
2074 return error;
2075
2076 unlock:
2077 unlock_page(page);
2078 page_cache_release(page);
2079 return error;
2080}
2081
2082/*
2083** vfs version of truncate file. Must NOT be called with
2084** a transaction already started.
2085**
2086** some code taken from block_truncate_page
2087*/
2088int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2089{
2090 struct reiserfs_transaction_handle th;
2091 /* we want the offset for the first byte after the end of the file */
2092 unsigned long offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2093 unsigned blocksize = inode->i_sb->s_blocksize;
2094 unsigned length;
2095 struct page *page = NULL;
2096 int error;
2097 struct buffer_head *bh = NULL;
2098 int err2;
2099 int lock_depth;
2100
2101 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2102
2103 if (inode->i_size > 0) {
2104 error = grab_tail_page(inode, &page, &bh);
2105 if (error) {
2106 // -ENOENT means we truncated past the end of the file,
2107 // and get_block_create_0 could not find a block to read in,
2108 // which is ok.
2109 if (error != -ENOENT)
2110 reiserfs_error(inode->i_sb, "clm-6001",
2111 "grab_tail_page failed %d",
2112 error);
2113 page = NULL;
2114 bh = NULL;
2115 }
2116 }
2117
2118 /* so, if page != NULL, we have a buffer head for the offset at
2119 ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2120 ** then we have an unformatted node. Otherwise, we have a direct item,
2121 ** and no zeroing is required on disk. We zero after the truncate,
2122 ** because the truncate might pack the item anyway
2123 ** (it will unmap bh if it packs).
2124 */
2125 /* it is enough to reserve space in transaction for 2 balancings:
2126 one for "save" link adding and another for the first
2127 cut_from_item. 1 is for update_sd */
2128 error = journal_begin(&th, inode->i_sb,
2129 JOURNAL_PER_BALANCE_CNT * 2 + 1);
2130 if (error)
2131 goto out;
2132 reiserfs_update_inode_transaction(inode);
2133 if (update_timestamps)
2134 /* we are doing real truncate: if the system crashes before the last
2135 transaction of truncating gets committed - on reboot the file
2136 either appears truncated properly or not truncated at all */
2137 add_save_link(&th, inode, 1);
2138 err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2139 error =
2140 journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2141 if (error)
2142 goto out;
2143
2144 /* check reiserfs_do_truncate after ending the transaction */
2145 if (err2) {
2146 error = err2;
2147 goto out;
2148 }
2149
2150 if (update_timestamps) {
2151 error = remove_save_link(inode, 1 /* truncate */);
2152 if (error)
2153 goto out;
2154 }
2155
2156 if (page) {
2157 length = offset & (blocksize - 1);
2158 /* if we are not on a block boundary */
2159 if (length) {
2160 length = blocksize - length;
2161 zero_user(page, offset, length);
2162 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2163 mark_buffer_dirty(bh);
2164 }
2165 }
2166 unlock_page(page);
2167 page_cache_release(page);
2168 }
2169
2170 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2171
2172 return 0;
2173 out:
2174 if (page) {
2175 unlock_page(page);
2176 page_cache_release(page);
2177 }
2178
2179 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2180
2181 return error;
2182}
2183
2184static int map_block_for_writepage(struct inode *inode,
2185 struct buffer_head *bh_result,
2186 unsigned long block)
2187{
2188 struct reiserfs_transaction_handle th;
2189 int fs_gen;
2190 struct item_head tmp_ih;
2191 struct item_head *ih;
2192 struct buffer_head *bh;
2193 __le32 *item;
2194 struct cpu_key key;
2195 INITIALIZE_PATH(path);
2196 int pos_in_item;
2197 int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2198 loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2199 int retval;
2200 int use_get_block = 0;
2201 int bytes_copied = 0;
2202 int copy_size;
2203 int trans_running = 0;
2204
2205 /* catch places below that try to log something without starting a trans */
2206 th.t_trans_id = 0;
2207
2208 if (!buffer_uptodate(bh_result)) {
2209 return -EIO;
2210 }
2211
2212 kmap(bh_result->b_page);
2213 start_over:
2214 reiserfs_write_lock(inode->i_sb);
2215 make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2216
2217 research:
2218 retval = search_for_position_by_key(inode->i_sb, &key, &path);
2219 if (retval != POSITION_FOUND) {
2220 use_get_block = 1;
2221 goto out;
2222 }
2223
2224 bh = get_last_bh(&path);
2225 ih = get_ih(&path);
2226 item = get_item(&path);
2227 pos_in_item = path.pos_in_item;
2228
2229 /* we've found an unformatted node */
2230 if (indirect_item_found(retval, ih)) {
2231 if (bytes_copied > 0) {
2232 reiserfs_warning(inode->i_sb, "clm-6002",
2233 "bytes_copied %d", bytes_copied);
2234 }
2235 if (!get_block_num(item, pos_in_item)) {
2236 /* crap, we are writing to a hole */
2237 use_get_block = 1;
2238 goto out;
2239 }
2240 set_block_dev_mapped(bh_result,
2241 get_block_num(item, pos_in_item), inode);
2242 } else if (is_direct_le_ih(ih)) {
2243 char *p;
2244 p = page_address(bh_result->b_page);
2245 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2246 copy_size = ih_item_len(ih) - pos_in_item;
2247
2248 fs_gen = get_generation(inode->i_sb);
2249 copy_item_head(&tmp_ih, ih);
2250
2251 if (!trans_running) {
2252 /* vs-3050 is gone, no need to drop the path */
2253 retval = journal_begin(&th, inode->i_sb, jbegin_count);
2254 if (retval)
2255 goto out;
2256 reiserfs_update_inode_transaction(inode);
2257 trans_running = 1;
2258 if (fs_changed(fs_gen, inode->i_sb)
2259 && item_moved(&tmp_ih, &path)) {
2260 reiserfs_restore_prepared_buffer(inode->i_sb,
2261 bh);
2262 goto research;
2263 }
2264 }
2265
2266 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2267
2268 if (fs_changed(fs_gen, inode->i_sb)
2269 && item_moved(&tmp_ih, &path)) {
2270 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2271 goto research;
2272 }
2273
2274 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2275 copy_size);
2276
2277 journal_mark_dirty(&th, inode->i_sb, bh);
2278 bytes_copied += copy_size;
2279 set_block_dev_mapped(bh_result, 0, inode);
2280
2281 /* are there still bytes left? */
2282 if (bytes_copied < bh_result->b_size &&
2283 (byte_offset + bytes_copied) < inode->i_size) {
2284 set_cpu_key_k_offset(&key,
2285 cpu_key_k_offset(&key) +
2286 copy_size);
2287 goto research;
2288 }
2289 } else {
2290 reiserfs_warning(inode->i_sb, "clm-6003",
2291 "bad item inode %lu", inode->i_ino);
2292 retval = -EIO;
2293 goto out;
2294 }
2295 retval = 0;
2296
2297 out:
2298 pathrelse(&path);
2299 if (trans_running) {
2300 int err = journal_end(&th, inode->i_sb, jbegin_count);
2301 if (err)
2302 retval = err;
2303 trans_running = 0;
2304 }
2305 reiserfs_write_unlock(inode->i_sb);
2306
2307 /* this is where we fill in holes in the file. */
2308 if (use_get_block) {
2309 retval = reiserfs_get_block(inode, block, bh_result,
2310 GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2311 | GET_BLOCK_NO_DANGLE);
2312 if (!retval) {
2313 if (!buffer_mapped(bh_result)
2314 || bh_result->b_blocknr == 0) {
2315 /* get_block failed to find a mapped unformatted node. */
2316 use_get_block = 0;
2317 goto start_over;
2318 }
2319 }
2320 }
2321 kunmap(bh_result->b_page);
2322
2323 if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2324 /* we've copied data from the page into the direct item, so the
2325 * buffer in the page is now clean, mark it to reflect that.
2326 */
2327 lock_buffer(bh_result);
2328 clear_buffer_dirty(bh_result);
2329 unlock_buffer(bh_result);
2330 }
2331 return retval;
2332}
2333
2334/*
2335 * mason@suse.com: updated in 2.5.54 to follow the same general io
2336 * start/recovery path as __block_write_full_page, along with special
2337 * code to handle reiserfs tails.
2338 */
2339static int reiserfs_write_full_page(struct page *page,
2340 struct writeback_control *wbc)
2341{
2342 struct inode *inode = page->mapping->host;
2343 unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2344 int error = 0;
2345 unsigned long block;
2346 sector_t last_block;
2347 struct buffer_head *head, *bh;
2348 int partial = 0;
2349 int nr = 0;
2350 int checked = PageChecked(page);
2351 struct reiserfs_transaction_handle th;
2352 struct super_block *s = inode->i_sb;
2353 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2354 th.t_trans_id = 0;
2355
2356 /* no logging allowed when nonblocking or from PF_MEMALLOC */
2357 if (checked && (current->flags & PF_MEMALLOC)) {
2358 redirty_page_for_writepage(wbc, page);
2359 unlock_page(page);
2360 return 0;
2361 }
2362
2363 /* The page dirty bit is cleared before writepage is called, which
2364 * means we have to tell create_empty_buffers to make dirty buffers
2365 * The page really should be up to date at this point, so tossing
2366 * in the BH_Uptodate is just a sanity check.
2367 */
2368 if (!page_has_buffers(page)) {
2369 create_empty_buffers(page, s->s_blocksize,
2370 (1 << BH_Dirty) | (1 << BH_Uptodate));
2371 }
2372 head = page_buffers(page);
2373
2374 /* last page in the file, zero out any contents past the
2375 ** last byte in the file
2376 */
2377 if (page->index >= end_index) {
2378 unsigned last_offset;
2379
2380 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2381 /* no file contents in this page */
2382 if (page->index >= end_index + 1 || !last_offset) {
2383 unlock_page(page);
2384 return 0;
2385 }
2386 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2387 }
2388 bh = head;
2389 block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2390 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2391 /* first map all the buffers, logging any direct items we find */
2392 do {
2393 if (block > last_block) {
2394 /*
2395 * This can happen when the block size is less than
2396 * the page size. The corresponding bytes in the page
2397 * were zero filled above
2398 */
2399 clear_buffer_dirty(bh);
2400 set_buffer_uptodate(bh);
2401 } else if ((checked || buffer_dirty(bh)) &&
2402 (!buffer_mapped(bh) || (buffer_mapped(bh)
2403 && bh->b_blocknr ==
2404 0))) {
2405 /* not mapped yet, or it points to a direct item, search
2406 * the btree for the mapping info, and log any direct
2407 * items found
2408 */
2409 if ((error = map_block_for_writepage(inode, bh, block))) {
2410 goto fail;
2411 }
2412 }
2413 bh = bh->b_this_page;
2414 block++;
2415 } while (bh != head);
2416
2417 /*
2418 * we start the transaction after map_block_for_writepage,
2419 * because it can create holes in the file (an unbounded operation).
2420 * starting it here, we can make a reliable estimate for how many
2421 * blocks we're going to log
2422 */
2423 if (checked) {
2424 ClearPageChecked(page);
2425 reiserfs_write_lock(s);
2426 error = journal_begin(&th, s, bh_per_page + 1);
2427 if (error) {
2428 reiserfs_write_unlock(s);
2429 goto fail;
2430 }
2431 reiserfs_update_inode_transaction(inode);
2432 }
2433 /* now go through and lock any dirty buffers on the page */
2434 do {
2435 get_bh(bh);
2436 if (!buffer_mapped(bh))
2437 continue;
2438 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2439 continue;
2440
2441 if (checked) {
2442 reiserfs_prepare_for_journal(s, bh, 1);
2443 journal_mark_dirty(&th, s, bh);
2444 continue;
2445 }
2446 /* from this point on, we know the buffer is mapped to a
2447 * real block and not a direct item
2448 */
2449 if (wbc->sync_mode != WB_SYNC_NONE) {
2450 lock_buffer(bh);
2451 } else {
2452 if (!trylock_buffer(bh)) {
2453 redirty_page_for_writepage(wbc, page);
2454 continue;
2455 }
2456 }
2457 if (test_clear_buffer_dirty(bh)) {
2458 mark_buffer_async_write(bh);
2459 } else {
2460 unlock_buffer(bh);
2461 }
2462 } while ((bh = bh->b_this_page) != head);
2463
2464 if (checked) {
2465 error = journal_end(&th, s, bh_per_page + 1);
2466 reiserfs_write_unlock(s);
2467 if (error)
2468 goto fail;
2469 }
2470 BUG_ON(PageWriteback(page));
2471 set_page_writeback(page);
2472 unlock_page(page);
2473
2474 /*
2475 * since any buffer might be the only dirty buffer on the page,
2476 * the first submit_bh can bring the page out of writeback.
2477 * be careful with the buffers.
2478 */
2479 do {
2480 struct buffer_head *next = bh->b_this_page;
2481 if (buffer_async_write(bh)) {
2482 submit_bh(WRITE, bh);
2483 nr++;
2484 }
2485 put_bh(bh);
2486 bh = next;
2487 } while (bh != head);
2488
2489 error = 0;
2490 done:
2491 if (nr == 0) {
2492 /*
2493 * if this page only had a direct item, it is very possible for
2494 * no io to be required without there being an error. Or,
2495 * someone else could have locked them and sent them down the
2496 * pipe without locking the page
2497 */
2498 bh = head;
2499 do {
2500 if (!buffer_uptodate(bh)) {
2501 partial = 1;
2502 break;
2503 }
2504 bh = bh->b_this_page;
2505 } while (bh != head);
2506 if (!partial)
2507 SetPageUptodate(page);
2508 end_page_writeback(page);
2509 }
2510 return error;
2511
2512 fail:
2513 /* catches various errors, we need to make sure any valid dirty blocks
2514 * get to the media. The page is currently locked and not marked for
2515 * writeback
2516 */
2517 ClearPageUptodate(page);
2518 bh = head;
2519 do {
2520 get_bh(bh);
2521 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2522 lock_buffer(bh);
2523 mark_buffer_async_write(bh);
2524 } else {
2525 /*
2526 * clear any dirty bits that might have come from getting
2527 * attached to a dirty page
2528 */
2529 clear_buffer_dirty(bh);
2530 }
2531 bh = bh->b_this_page;
2532 } while (bh != head);
2533 SetPageError(page);
2534 BUG_ON(PageWriteback(page));
2535 set_page_writeback(page);
2536 unlock_page(page);
2537 do {
2538 struct buffer_head *next = bh->b_this_page;
2539 if (buffer_async_write(bh)) {
2540 clear_buffer_dirty(bh);
2541 submit_bh(WRITE, bh);
2542 nr++;
2543 }
2544 put_bh(bh);
2545 bh = next;
2546 } while (bh != head);
2547 goto done;
2548}
2549
2550static int reiserfs_readpage(struct file *f, struct page *page)
2551{
2552 return block_read_full_page(page, reiserfs_get_block);
2553}
2554
2555static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2556{
2557 struct inode *inode = page->mapping->host;
2558 reiserfs_wait_on_write_block(inode->i_sb);
2559 return reiserfs_write_full_page(page, wbc);
2560}
2561
2562static void reiserfs_truncate_failed_write(struct inode *inode)
2563{
2564 truncate_inode_pages(inode->i_mapping, inode->i_size);
2565 reiserfs_truncate_file(inode, 0);
2566}
2567
2568static int reiserfs_write_begin(struct file *file,
2569 struct address_space *mapping,
2570 loff_t pos, unsigned len, unsigned flags,
2571 struct page **pagep, void **fsdata)
2572{
2573 struct inode *inode;
2574 struct page *page;
2575 pgoff_t index;
2576 int ret;
2577 int old_ref = 0;
2578
2579 inode = mapping->host;
2580 *fsdata = 0;
2581 if (flags & AOP_FLAG_CONT_EXPAND &&
2582 (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2583 pos ++;
2584 *fsdata = (void *)(unsigned long)flags;
2585 }
2586
2587 index = pos >> PAGE_CACHE_SHIFT;
2588 page = grab_cache_page_write_begin(mapping, index, flags);
2589 if (!page)
2590 return -ENOMEM;
2591 *pagep = page;
2592
2593 reiserfs_wait_on_write_block(inode->i_sb);
2594 fix_tail_page_for_writing(page);
2595 if (reiserfs_transaction_running(inode->i_sb)) {
2596 struct reiserfs_transaction_handle *th;
2597 th = (struct reiserfs_transaction_handle *)current->
2598 journal_info;
2599 BUG_ON(!th->t_refcount);
2600 BUG_ON(!th->t_trans_id);
2601 old_ref = th->t_refcount;
2602 th->t_refcount++;
2603 }
2604 ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2605 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2606 struct reiserfs_transaction_handle *th = current->journal_info;
2607 /* this gets a little ugly. If reiserfs_get_block returned an
2608 * error and left a transacstion running, we've got to close it,
2609 * and we've got to free handle if it was a persistent transaction.
2610 *
2611 * But, if we had nested into an existing transaction, we need
2612 * to just drop the ref count on the handle.
2613 *
2614 * If old_ref == 0, the transaction is from reiserfs_get_block,
2615 * and it was a persistent trans. Otherwise, it was nested above.
2616 */
2617 if (th->t_refcount > old_ref) {
2618 if (old_ref)
2619 th->t_refcount--;
2620 else {
2621 int err;
2622 reiserfs_write_lock(inode->i_sb);
2623 err = reiserfs_end_persistent_transaction(th);
2624 reiserfs_write_unlock(inode->i_sb);
2625 if (err)
2626 ret = err;
2627 }
2628 }
2629 }
2630 if (ret) {
2631 unlock_page(page);
2632 page_cache_release(page);
2633 /* Truncate allocated blocks */
2634 reiserfs_truncate_failed_write(inode);
2635 }
2636 return ret;
2637}
2638
2639int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2640{
2641 struct inode *inode = page->mapping->host;
2642 int ret;
2643 int old_ref = 0;
2644
2645 reiserfs_write_unlock(inode->i_sb);
2646 reiserfs_wait_on_write_block(inode->i_sb);
2647 reiserfs_write_lock(inode->i_sb);
2648
2649 fix_tail_page_for_writing(page);
2650 if (reiserfs_transaction_running(inode->i_sb)) {
2651 struct reiserfs_transaction_handle *th;
2652 th = (struct reiserfs_transaction_handle *)current->
2653 journal_info;
2654 BUG_ON(!th->t_refcount);
2655 BUG_ON(!th->t_trans_id);
2656 old_ref = th->t_refcount;
2657 th->t_refcount++;
2658 }
2659
2660 ret = __block_write_begin(page, from, len, reiserfs_get_block);
2661 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2662 struct reiserfs_transaction_handle *th = current->journal_info;
2663 /* this gets a little ugly. If reiserfs_get_block returned an
2664 * error and left a transacstion running, we've got to close it,
2665 * and we've got to free handle if it was a persistent transaction.
2666 *
2667 * But, if we had nested into an existing transaction, we need
2668 * to just drop the ref count on the handle.
2669 *
2670 * If old_ref == 0, the transaction is from reiserfs_get_block,
2671 * and it was a persistent trans. Otherwise, it was nested above.
2672 */
2673 if (th->t_refcount > old_ref) {
2674 if (old_ref)
2675 th->t_refcount--;
2676 else {
2677 int err;
2678 reiserfs_write_lock(inode->i_sb);
2679 err = reiserfs_end_persistent_transaction(th);
2680 reiserfs_write_unlock(inode->i_sb);
2681 if (err)
2682 ret = err;
2683 }
2684 }
2685 }
2686 return ret;
2687
2688}
2689
2690static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2691{
2692 return generic_block_bmap(as, block, reiserfs_bmap);
2693}
2694
2695static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2696 loff_t pos, unsigned len, unsigned copied,
2697 struct page *page, void *fsdata)
2698{
2699 struct inode *inode = page->mapping->host;
2700 int ret = 0;
2701 int update_sd = 0;
2702 struct reiserfs_transaction_handle *th;
2703 unsigned start;
2704 int lock_depth = 0;
2705 bool locked = false;
2706
2707 if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2708 pos ++;
2709
2710 reiserfs_wait_on_write_block(inode->i_sb);
2711 if (reiserfs_transaction_running(inode->i_sb))
2712 th = current->journal_info;
2713 else
2714 th = NULL;
2715
2716 start = pos & (PAGE_CACHE_SIZE - 1);
2717 if (unlikely(copied < len)) {
2718 if (!PageUptodate(page))
2719 copied = 0;
2720
2721 page_zero_new_buffers(page, start + copied, start + len);
2722 }
2723 flush_dcache_page(page);
2724
2725 reiserfs_commit_page(inode, page, start, start + copied);
2726
2727 /* generic_commit_write does this for us, but does not update the
2728 ** transaction tracking stuff when the size changes. So, we have
2729 ** to do the i_size updates here.
2730 */
2731 if (pos + copied > inode->i_size) {
2732 struct reiserfs_transaction_handle myth;
2733 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2734 locked = true;
2735 /* If the file have grown beyond the border where it
2736 can have a tail, unmark it as needing a tail
2737 packing */
2738 if ((have_large_tails(inode->i_sb)
2739 && inode->i_size > i_block_size(inode) * 4)
2740 || (have_small_tails(inode->i_sb)
2741 && inode->i_size > i_block_size(inode)))
2742 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2743
2744 ret = journal_begin(&myth, inode->i_sb, 1);
2745 if (ret)
2746 goto journal_error;
2747
2748 reiserfs_update_inode_transaction(inode);
2749 inode->i_size = pos + copied;
2750 /*
2751 * this will just nest into our transaction. It's important
2752 * to use mark_inode_dirty so the inode gets pushed around on the
2753 * dirty lists, and so that O_SYNC works as expected
2754 */
2755 mark_inode_dirty(inode);
2756 reiserfs_update_sd(&myth, inode);
2757 update_sd = 1;
2758 ret = journal_end(&myth, inode->i_sb, 1);
2759 if (ret)
2760 goto journal_error;
2761 }
2762 if (th) {
2763 if (!locked) {
2764 lock_depth = reiserfs_write_lock_once(inode->i_sb);
2765 locked = true;
2766 }
2767 if (!update_sd)
2768 mark_inode_dirty(inode);
2769 ret = reiserfs_end_persistent_transaction(th);
2770 if (ret)
2771 goto out;
2772 }
2773
2774 out:
2775 if (locked)
2776 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2777 unlock_page(page);
2778 page_cache_release(page);
2779
2780 if (pos + len > inode->i_size)
2781 reiserfs_truncate_failed_write(inode);
2782
2783 return ret == 0 ? copied : ret;
2784
2785 journal_error:
2786 reiserfs_write_unlock_once(inode->i_sb, lock_depth);
2787 locked = false;
2788 if (th) {
2789 if (!update_sd)
2790 reiserfs_update_sd(th, inode);
2791 ret = reiserfs_end_persistent_transaction(th);
2792 }
2793 goto out;
2794}
2795
2796int reiserfs_commit_write(struct file *f, struct page *page,
2797 unsigned from, unsigned to)
2798{
2799 struct inode *inode = page->mapping->host;
2800 loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2801 int ret = 0;
2802 int update_sd = 0;
2803 struct reiserfs_transaction_handle *th = NULL;
2804
2805 reiserfs_write_unlock(inode->i_sb);
2806 reiserfs_wait_on_write_block(inode->i_sb);
2807 reiserfs_write_lock(inode->i_sb);
2808
2809 if (reiserfs_transaction_running(inode->i_sb)) {
2810 th = current->journal_info;
2811 }
2812 reiserfs_commit_page(inode, page, from, to);
2813
2814 /* generic_commit_write does this for us, but does not update the
2815 ** transaction tracking stuff when the size changes. So, we have
2816 ** to do the i_size updates here.
2817 */
2818 if (pos > inode->i_size) {
2819 struct reiserfs_transaction_handle myth;
2820 /* If the file have grown beyond the border where it
2821 can have a tail, unmark it as needing a tail
2822 packing */
2823 if ((have_large_tails(inode->i_sb)
2824 && inode->i_size > i_block_size(inode) * 4)
2825 || (have_small_tails(inode->i_sb)
2826 && inode->i_size > i_block_size(inode)))
2827 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2828
2829 ret = journal_begin(&myth, inode->i_sb, 1);
2830 if (ret)
2831 goto journal_error;
2832
2833 reiserfs_update_inode_transaction(inode);
2834 inode->i_size = pos;
2835 /*
2836 * this will just nest into our transaction. It's important
2837 * to use mark_inode_dirty so the inode gets pushed around on the
2838 * dirty lists, and so that O_SYNC works as expected
2839 */
2840 mark_inode_dirty(inode);
2841 reiserfs_update_sd(&myth, inode);
2842 update_sd = 1;
2843 ret = journal_end(&myth, inode->i_sb, 1);
2844 if (ret)
2845 goto journal_error;
2846 }
2847 if (th) {
2848 if (!update_sd)
2849 mark_inode_dirty(inode);
2850 ret = reiserfs_end_persistent_transaction(th);
2851 if (ret)
2852 goto out;
2853 }
2854
2855 out:
2856 return ret;
2857
2858 journal_error:
2859 if (th) {
2860 if (!update_sd)
2861 reiserfs_update_sd(th, inode);
2862 ret = reiserfs_end_persistent_transaction(th);
2863 }
2864
2865 return ret;
2866}
2867
2868void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2869{
2870 if (reiserfs_attrs(inode->i_sb)) {
2871 if (sd_attrs & REISERFS_SYNC_FL)
2872 inode->i_flags |= S_SYNC;
2873 else
2874 inode->i_flags &= ~S_SYNC;
2875 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2876 inode->i_flags |= S_IMMUTABLE;
2877 else
2878 inode->i_flags &= ~S_IMMUTABLE;
2879 if (sd_attrs & REISERFS_APPEND_FL)
2880 inode->i_flags |= S_APPEND;
2881 else
2882 inode->i_flags &= ~S_APPEND;
2883 if (sd_attrs & REISERFS_NOATIME_FL)
2884 inode->i_flags |= S_NOATIME;
2885 else
2886 inode->i_flags &= ~S_NOATIME;
2887 if (sd_attrs & REISERFS_NOTAIL_FL)
2888 REISERFS_I(inode)->i_flags |= i_nopack_mask;
2889 else
2890 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2891 }
2892}
2893
2894void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2895{
2896 if (reiserfs_attrs(inode->i_sb)) {
2897 if (inode->i_flags & S_IMMUTABLE)
2898 *sd_attrs |= REISERFS_IMMUTABLE_FL;
2899 else
2900 *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2901 if (inode->i_flags & S_SYNC)
2902 *sd_attrs |= REISERFS_SYNC_FL;
2903 else
2904 *sd_attrs &= ~REISERFS_SYNC_FL;
2905 if (inode->i_flags & S_NOATIME)
2906 *sd_attrs |= REISERFS_NOATIME_FL;
2907 else
2908 *sd_attrs &= ~REISERFS_NOATIME_FL;
2909 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2910 *sd_attrs |= REISERFS_NOTAIL_FL;
2911 else
2912 *sd_attrs &= ~REISERFS_NOTAIL_FL;
2913 }
2914}
2915
2916/* decide if this buffer needs to stay around for data logging or ordered
2917** write purposes
2918*/
2919static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2920{
2921 int ret = 1;
2922 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2923
2924 lock_buffer(bh);
2925 spin_lock(&j->j_dirty_buffers_lock);
2926 if (!buffer_mapped(bh)) {
2927 goto free_jh;
2928 }
2929 /* the page is locked, and the only places that log a data buffer
2930 * also lock the page.
2931 */
2932 if (reiserfs_file_data_log(inode)) {
2933 /*
2934 * very conservative, leave the buffer pinned if
2935 * anyone might need it.
2936 */
2937 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2938 ret = 0;
2939 }
2940 } else if (buffer_dirty(bh)) {
2941 struct reiserfs_journal_list *jl;
2942 struct reiserfs_jh *jh = bh->b_private;
2943
2944 /* why is this safe?
2945 * reiserfs_setattr updates i_size in the on disk
2946 * stat data before allowing vmtruncate to be called.
2947 *
2948 * If buffer was put onto the ordered list for this
2949 * transaction, we know for sure either this transaction
2950 * or an older one already has updated i_size on disk,
2951 * and this ordered data won't be referenced in the file
2952 * if we crash.
2953 *
2954 * if the buffer was put onto the ordered list for an older
2955 * transaction, we need to leave it around
2956 */
2957 if (jh && (jl = jh->jl)
2958 && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2959 ret = 0;
2960 }
2961 free_jh:
2962 if (ret && bh->b_private) {
2963 reiserfs_free_jh(bh);
2964 }
2965 spin_unlock(&j->j_dirty_buffers_lock);
2966 unlock_buffer(bh);
2967 return ret;
2968}
2969
2970/* clm -- taken from fs/buffer.c:block_invalidate_page */
2971static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2972{
2973 struct buffer_head *head, *bh, *next;
2974 struct inode *inode = page->mapping->host;
2975 unsigned int curr_off = 0;
2976 int ret = 1;
2977
2978 BUG_ON(!PageLocked(page));
2979
2980 if (offset == 0)
2981 ClearPageChecked(page);
2982
2983 if (!page_has_buffers(page))
2984 goto out;
2985
2986 head = page_buffers(page);
2987 bh = head;
2988 do {
2989 unsigned int next_off = curr_off + bh->b_size;
2990 next = bh->b_this_page;
2991
2992 /*
2993 * is this block fully invalidated?
2994 */
2995 if (offset <= curr_off) {
2996 if (invalidatepage_can_drop(inode, bh))
2997 reiserfs_unmap_buffer(bh);
2998 else
2999 ret = 0;
3000 }
3001 curr_off = next_off;
3002 bh = next;
3003 } while (bh != head);
3004
3005 /*
3006 * We release buffers only if the entire page is being invalidated.
3007 * The get_block cached value has been unconditionally invalidated,
3008 * so real IO is not possible anymore.
3009 */
3010 if (!offset && ret) {
3011 ret = try_to_release_page(page, 0);
3012 /* maybe should BUG_ON(!ret); - neilb */
3013 }
3014 out:
3015 return;
3016}
3017
3018static int reiserfs_set_page_dirty(struct page *page)
3019{
3020 struct inode *inode = page->mapping->host;
3021 if (reiserfs_file_data_log(inode)) {
3022 SetPageChecked(page);
3023 return __set_page_dirty_nobuffers(page);
3024 }
3025 return __set_page_dirty_buffers(page);
3026}
3027
3028/*
3029 * Returns 1 if the page's buffers were dropped. The page is locked.
3030 *
3031 * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3032 * in the buffers at page_buffers(page).
3033 *
3034 * even in -o notail mode, we can't be sure an old mount without -o notail
3035 * didn't create files with tails.
3036 */
3037static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3038{
3039 struct inode *inode = page->mapping->host;
3040 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3041 struct buffer_head *head;
3042 struct buffer_head *bh;
3043 int ret = 1;
3044
3045 WARN_ON(PageChecked(page));
3046 spin_lock(&j->j_dirty_buffers_lock);
3047 head = page_buffers(page);
3048 bh = head;
3049 do {
3050 if (bh->b_private) {
3051 if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3052 reiserfs_free_jh(bh);
3053 } else {
3054 ret = 0;
3055 break;
3056 }
3057 }
3058 bh = bh->b_this_page;
3059 } while (bh != head);
3060 if (ret)
3061 ret = try_to_free_buffers(page);
3062 spin_unlock(&j->j_dirty_buffers_lock);
3063 return ret;
3064}
3065
3066/* We thank Mingming Cao for helping us understand in great detail what
3067 to do in this section of the code. */
3068static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3069 const struct iovec *iov, loff_t offset,
3070 unsigned long nr_segs)
3071{
3072 struct file *file = iocb->ki_filp;
3073 struct inode *inode = file->f_mapping->host;
3074 ssize_t ret;
3075
3076 ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
3077 reiserfs_get_blocks_direct_io);
3078
3079 /*
3080 * In case of error extending write may have instantiated a few
3081 * blocks outside i_size. Trim these off again.
3082 */
3083 if (unlikely((rw & WRITE) && ret < 0)) {
3084 loff_t isize = i_size_read(inode);
3085 loff_t end = offset + iov_length(iov, nr_segs);
3086
3087 if (end > isize)
3088 vmtruncate(inode, isize);
3089 }
3090
3091 return ret;
3092}
3093
3094int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3095{
3096 struct inode *inode = dentry->d_inode;
3097 unsigned int ia_valid;
3098 int depth;
3099 int error;
3100
3101 error = inode_change_ok(inode, attr);
3102 if (error)
3103 return error;
3104
3105 /* must be turned off for recursive notify_change calls */
3106 ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3107
3108 depth = reiserfs_write_lock_once(inode->i_sb);
3109 if (is_quota_modification(inode, attr))
3110 dquot_initialize(inode);
3111
3112 if (attr->ia_valid & ATTR_SIZE) {
3113 /* version 2 items will be caught by the s_maxbytes check
3114 ** done for us in vmtruncate
3115 */
3116 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3117 attr->ia_size > MAX_NON_LFS) {
3118 error = -EFBIG;
3119 goto out;
3120 }
3121
3122 inode_dio_wait(inode);
3123
3124 /* fill in hole pointers in the expanding truncate case. */
3125 if (attr->ia_size > inode->i_size) {
3126 error = generic_cont_expand_simple(inode, attr->ia_size);
3127 if (REISERFS_I(inode)->i_prealloc_count > 0) {
3128 int err;
3129 struct reiserfs_transaction_handle th;
3130 /* we're changing at most 2 bitmaps, inode + super */
3131 err = journal_begin(&th, inode->i_sb, 4);
3132 if (!err) {
3133 reiserfs_discard_prealloc(&th, inode);
3134 err = journal_end(&th, inode->i_sb, 4);
3135 }
3136 if (err)
3137 error = err;
3138 }
3139 if (error)
3140 goto out;
3141 /*
3142 * file size is changed, ctime and mtime are
3143 * to be updated
3144 */
3145 attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3146 }
3147 }
3148
3149 if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3150 ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3151 (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3152 /* stat data of format v3.5 has 16 bit uid and gid */
3153 error = -EINVAL;
3154 goto out;
3155 }
3156
3157 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3158 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3159 struct reiserfs_transaction_handle th;
3160 int jbegin_count =
3161 2 *
3162 (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3163 REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3164 2;
3165
3166 error = reiserfs_chown_xattrs(inode, attr);
3167
3168 if (error)
3169 return error;
3170
3171 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3172 error = journal_begin(&th, inode->i_sb, jbegin_count);
3173 if (error)
3174 goto out;
3175 error = dquot_transfer(inode, attr);
3176 if (error) {
3177 journal_end(&th, inode->i_sb, jbegin_count);
3178 goto out;
3179 }
3180
3181 /* Update corresponding info in inode so that everything is in
3182 * one transaction */
3183 if (attr->ia_valid & ATTR_UID)
3184 inode->i_uid = attr->ia_uid;
3185 if (attr->ia_valid & ATTR_GID)
3186 inode->i_gid = attr->ia_gid;
3187 mark_inode_dirty(inode);
3188 error = journal_end(&th, inode->i_sb, jbegin_count);
3189 if (error)
3190 goto out;
3191 }
3192
3193 /*
3194 * Relax the lock here, as it might truncate the
3195 * inode pages and wait for inode pages locks.
3196 * To release such page lock, the owner needs the
3197 * reiserfs lock
3198 */
3199 reiserfs_write_unlock_once(inode->i_sb, depth);
3200 if ((attr->ia_valid & ATTR_SIZE) &&
3201 attr->ia_size != i_size_read(inode))
3202 error = vmtruncate(inode, attr->ia_size);
3203
3204 if (!error) {
3205 setattr_copy(inode, attr);
3206 mark_inode_dirty(inode);
3207 }
3208 depth = reiserfs_write_lock_once(inode->i_sb);
3209
3210 if (!error && reiserfs_posixacl(inode->i_sb)) {
3211 if (attr->ia_valid & ATTR_MODE)
3212 error = reiserfs_acl_chmod(inode);
3213 }
3214
3215 out:
3216 reiserfs_write_unlock_once(inode->i_sb, depth);
3217
3218 return error;
3219}
3220
3221const struct address_space_operations reiserfs_address_space_operations = {
3222 .writepage = reiserfs_writepage,
3223 .readpage = reiserfs_readpage,
3224 .readpages = reiserfs_readpages,
3225 .releasepage = reiserfs_releasepage,
3226 .invalidatepage = reiserfs_invalidatepage,
3227 .write_begin = reiserfs_write_begin,
3228 .write_end = reiserfs_write_end,
3229 .bmap = reiserfs_aop_bmap,
3230 .direct_IO = reiserfs_direct_IO,
3231 .set_page_dirty = reiserfs_set_page_dirty,
3232};
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
24int reiserfs_commit_write(struct file *f, struct page *page,
25 unsigned from, unsigned to);
26
27void 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 */);
92out:
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
106no_delete:
107 clear_inode(inode);
108 dquot_drop(inode);
109}
110
111static 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 */
127void 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 */
137inline 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 */
193static 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 */
214static 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
226static inline int indirect_item_found(int retval, struct item_head *ih)
227{
228 return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
229}
230
231static 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 */
241static 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
252static 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 */
285static 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
453finished:
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 */
472static 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*/
503static 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 */
514static 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 = (1 << inode->i_blkbits);
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 }
562out:
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 */
576static 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
630unlock:
631 if (tail_page != hole_page) {
632 unlock_page(tail_page);
633 put_page(tail_page);
634 }
635out:
636 return retval;
637}
638
639static 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
657int 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) {
731start_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 }
739research:
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
1147failure:
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
1162static int
1163reiserfs_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 */
1174static 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
1198static 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 */
1210static 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 */
1238static 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 */
1375static void inode2sd(void *sd, struct inode *inode, loff_t size)
1376{
1377 struct stat_data *sd_v2 = (struct stat_data *)sd;
1378 __u16 flags;
1379
1380 set_sd_v2_mode(sd_v2, inode->i_mode);
1381 set_sd_v2_nlink(sd_v2, inode->i_nlink);
1382 set_sd_v2_uid(sd_v2, i_uid_read(inode));
1383 set_sd_v2_size(sd_v2, size);
1384 set_sd_v2_gid(sd_v2, i_gid_read(inode));
1385 set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1386 set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1387 set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1388 set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1389 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1390 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1391 else
1392 set_sd_v2_generation(sd_v2, inode->i_generation);
1393 flags = REISERFS_I(inode)->i_attrs;
1394 i_attrs_to_sd_attrs(inode, &flags);
1395 set_sd_v2_attrs(sd_v2, flags);
1396}
1397
1398/* used to copy inode's fields to old stat data */
1399static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1400{
1401 struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1402
1403 set_sd_v1_mode(sd_v1, inode->i_mode);
1404 set_sd_v1_uid(sd_v1, i_uid_read(inode));
1405 set_sd_v1_gid(sd_v1, i_gid_read(inode));
1406 set_sd_v1_nlink(sd_v1, inode->i_nlink);
1407 set_sd_v1_size(sd_v1, size);
1408 set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1409 set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1410 set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1411
1412 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1413 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1414 else
1415 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1416
1417 /* Sigh. i_first_direct_byte is back */
1418 set_sd_v1_first_direct_byte(sd_v1,
1419 REISERFS_I(inode)->i_first_direct_byte);
1420}
1421
1422/*
1423 * NOTE, you must prepare the buffer head before sending it here,
1424 * and then log it after the call
1425 */
1426static void update_stat_data(struct treepath *path, struct inode *inode,
1427 loff_t size)
1428{
1429 struct buffer_head *bh;
1430 struct item_head *ih;
1431
1432 bh = PATH_PLAST_BUFFER(path);
1433 ih = tp_item_head(path);
1434
1435 if (!is_statdata_le_ih(ih))
1436 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1437 INODE_PKEY(inode), ih);
1438
1439 /* path points to old stat data */
1440 if (stat_data_v1(ih)) {
1441 inode2sd_v1(ih_item_body(bh, ih), inode, size);
1442 } else {
1443 inode2sd(ih_item_body(bh, ih), inode, size);
1444 }
1445
1446 return;
1447}
1448
1449void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1450 struct inode *inode, loff_t size)
1451{
1452 struct cpu_key key;
1453 INITIALIZE_PATH(path);
1454 struct buffer_head *bh;
1455 int fs_gen;
1456 struct item_head *ih, tmp_ih;
1457 int retval;
1458
1459 BUG_ON(!th->t_trans_id);
1460
1461 /* key type is unimportant */
1462 make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1463
1464 for (;;) {
1465 int pos;
1466 /* look for the object's stat data */
1467 retval = search_item(inode->i_sb, &key, &path);
1468 if (retval == IO_ERROR) {
1469 reiserfs_error(inode->i_sb, "vs-13050",
1470 "i/o failure occurred trying to "
1471 "update %K stat data", &key);
1472 return;
1473 }
1474 if (retval == ITEM_NOT_FOUND) {
1475 pos = PATH_LAST_POSITION(&path);
1476 pathrelse(&path);
1477 if (inode->i_nlink == 0) {
1478 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1479 return;
1480 }
1481 reiserfs_warning(inode->i_sb, "vs-13060",
1482 "stat data of object %k (nlink == %d) "
1483 "not found (pos %d)",
1484 INODE_PKEY(inode), inode->i_nlink,
1485 pos);
1486 reiserfs_check_path(&path);
1487 return;
1488 }
1489
1490 /*
1491 * sigh, prepare_for_journal might schedule. When it
1492 * schedules the FS might change. We have to detect that,
1493 * and loop back to the search if the stat data item has moved
1494 */
1495 bh = get_last_bh(&path);
1496 ih = tp_item_head(&path);
1497 copy_item_head(&tmp_ih, ih);
1498 fs_gen = get_generation(inode->i_sb);
1499 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1500
1501 /* Stat_data item has been moved after scheduling. */
1502 if (fs_changed(fs_gen, inode->i_sb)
1503 && item_moved(&tmp_ih, &path)) {
1504 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1505 continue;
1506 }
1507 break;
1508 }
1509 update_stat_data(&path, inode, size);
1510 journal_mark_dirty(th, bh);
1511 pathrelse(&path);
1512 return;
1513}
1514
1515/*
1516 * reiserfs_read_locked_inode is called to read the inode off disk, and it
1517 * does a make_bad_inode when things go wrong. But, we need to make sure
1518 * and clear the key in the private portion of the inode, otherwise a
1519 * corresponding iput might try to delete whatever object the inode last
1520 * represented.
1521 */
1522static void reiserfs_make_bad_inode(struct inode *inode)
1523{
1524 memset(INODE_PKEY(inode), 0, KEY_SIZE);
1525 make_bad_inode(inode);
1526}
1527
1528/*
1529 * initially this function was derived from minix or ext2's analog and
1530 * evolved as the prototype did
1531 */
1532int reiserfs_init_locked_inode(struct inode *inode, void *p)
1533{
1534 struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1535 inode->i_ino = args->objectid;
1536 INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1537 return 0;
1538}
1539
1540/*
1541 * looks for stat data in the tree, and fills up the fields of in-core
1542 * inode stat data fields
1543 */
1544void reiserfs_read_locked_inode(struct inode *inode,
1545 struct reiserfs_iget_args *args)
1546{
1547 INITIALIZE_PATH(path_to_sd);
1548 struct cpu_key key;
1549 unsigned long dirino;
1550 int retval;
1551
1552 dirino = args->dirid;
1553
1554 /*
1555 * set version 1, version 2 could be used too, because stat data
1556 * key is the same in both versions
1557 */
1558 key.version = KEY_FORMAT_3_5;
1559 key.on_disk_key.k_dir_id = dirino;
1560 key.on_disk_key.k_objectid = inode->i_ino;
1561 key.on_disk_key.k_offset = 0;
1562 key.on_disk_key.k_type = 0;
1563
1564 /* look for the object's stat data */
1565 retval = search_item(inode->i_sb, &key, &path_to_sd);
1566 if (retval == IO_ERROR) {
1567 reiserfs_error(inode->i_sb, "vs-13070",
1568 "i/o failure occurred trying to find "
1569 "stat data of %K", &key);
1570 reiserfs_make_bad_inode(inode);
1571 return;
1572 }
1573
1574 /* a stale NFS handle can trigger this without it being an error */
1575 if (retval != ITEM_FOUND) {
1576 pathrelse(&path_to_sd);
1577 reiserfs_make_bad_inode(inode);
1578 clear_nlink(inode);
1579 return;
1580 }
1581
1582 init_inode(inode, &path_to_sd);
1583
1584 /*
1585 * It is possible that knfsd is trying to access inode of a file
1586 * that is being removed from the disk by some other thread. As we
1587 * update sd on unlink all that is required is to check for nlink
1588 * here. This bug was first found by Sizif when debugging
1589 * SquidNG/Butterfly, forgotten, and found again after Philippe
1590 * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1591
1592 * More logical fix would require changes in fs/inode.c:iput() to
1593 * remove inode from hash-table _after_ fs cleaned disk stuff up and
1594 * in iget() to return NULL if I_FREEING inode is found in
1595 * hash-table.
1596 */
1597
1598 /*
1599 * Currently there is one place where it's ok to meet inode with
1600 * nlink==0: processing of open-unlinked and half-truncated files
1601 * during mount (fs/reiserfs/super.c:finish_unfinished()).
1602 */
1603 if ((inode->i_nlink == 0) &&
1604 !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1605 reiserfs_warning(inode->i_sb, "vs-13075",
1606 "dead inode read from disk %K. "
1607 "This is likely to be race with knfsd. Ignore",
1608 &key);
1609 reiserfs_make_bad_inode(inode);
1610 }
1611
1612 /* init inode should be relsing */
1613 reiserfs_check_path(&path_to_sd);
1614
1615 /*
1616 * Stat data v1 doesn't support ACLs.
1617 */
1618 if (get_inode_sd_version(inode) == STAT_DATA_V1)
1619 cache_no_acl(inode);
1620}
1621
1622/*
1623 * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1624 *
1625 * @inode: inode from hash table to check
1626 * @opaque: "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1627 *
1628 * This function is called by iget5_locked() to distinguish reiserfs inodes
1629 * having the same inode numbers. Such inodes can only exist due to some
1630 * error condition. One of them should be bad. Inodes with identical
1631 * inode numbers (objectids) are distinguished by parent directory ids.
1632 *
1633 */
1634int reiserfs_find_actor(struct inode *inode, void *opaque)
1635{
1636 struct reiserfs_iget_args *args;
1637
1638 args = opaque;
1639 /* args is already in CPU order */
1640 return (inode->i_ino == args->objectid) &&
1641 (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1642}
1643
1644struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1645{
1646 struct inode *inode;
1647 struct reiserfs_iget_args args;
1648 int depth;
1649
1650 args.objectid = key->on_disk_key.k_objectid;
1651 args.dirid = key->on_disk_key.k_dir_id;
1652 depth = reiserfs_write_unlock_nested(s);
1653 inode = iget5_locked(s, key->on_disk_key.k_objectid,
1654 reiserfs_find_actor, reiserfs_init_locked_inode,
1655 (void *)(&args));
1656 reiserfs_write_lock_nested(s, depth);
1657 if (!inode)
1658 return ERR_PTR(-ENOMEM);
1659
1660 if (inode->i_state & I_NEW) {
1661 reiserfs_read_locked_inode(inode, &args);
1662 unlock_new_inode(inode);
1663 }
1664
1665 if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1666 /* either due to i/o error or a stale NFS handle */
1667 iput(inode);
1668 inode = NULL;
1669 }
1670 return inode;
1671}
1672
1673static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1674 u32 objectid, u32 dir_id, u32 generation)
1675
1676{
1677 struct cpu_key key;
1678 struct inode *inode;
1679
1680 key.on_disk_key.k_objectid = objectid;
1681 key.on_disk_key.k_dir_id = dir_id;
1682 reiserfs_write_lock(sb);
1683 inode = reiserfs_iget(sb, &key);
1684 if (inode && !IS_ERR(inode) && generation != 0 &&
1685 generation != inode->i_generation) {
1686 iput(inode);
1687 inode = NULL;
1688 }
1689 reiserfs_write_unlock(sb);
1690
1691 return d_obtain_alias(inode);
1692}
1693
1694struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1695 int fh_len, int fh_type)
1696{
1697 /*
1698 * fhtype happens to reflect the number of u32s encoded.
1699 * due to a bug in earlier code, fhtype might indicate there
1700 * are more u32s then actually fitted.
1701 * so if fhtype seems to be more than len, reduce fhtype.
1702 * Valid types are:
1703 * 2 - objectid + dir_id - legacy support
1704 * 3 - objectid + dir_id + generation
1705 * 4 - objectid + dir_id + objectid and dirid of parent - legacy
1706 * 5 - objectid + dir_id + generation + objectid and dirid of parent
1707 * 6 - as above plus generation of directory
1708 * 6 does not fit in NFSv2 handles
1709 */
1710 if (fh_type > fh_len) {
1711 if (fh_type != 6 || fh_len != 5)
1712 reiserfs_warning(sb, "reiserfs-13077",
1713 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1714 fh_type, fh_len);
1715 fh_type = fh_len;
1716 }
1717 if (fh_len < 2)
1718 return NULL;
1719
1720 return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1721 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1722}
1723
1724struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1725 int fh_len, int fh_type)
1726{
1727 if (fh_type > fh_len)
1728 fh_type = fh_len;
1729 if (fh_type < 4)
1730 return NULL;
1731
1732 return reiserfs_get_dentry(sb,
1733 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1734 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1735 (fh_type == 6) ? fid->raw[5] : 0);
1736}
1737
1738int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1739 struct inode *parent)
1740{
1741 int maxlen = *lenp;
1742
1743 if (parent && (maxlen < 5)) {
1744 *lenp = 5;
1745 return FILEID_INVALID;
1746 } else if (maxlen < 3) {
1747 *lenp = 3;
1748 return FILEID_INVALID;
1749 }
1750
1751 data[0] = inode->i_ino;
1752 data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1753 data[2] = inode->i_generation;
1754 *lenp = 3;
1755 if (parent) {
1756 data[3] = parent->i_ino;
1757 data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1758 *lenp = 5;
1759 if (maxlen >= 6) {
1760 data[5] = parent->i_generation;
1761 *lenp = 6;
1762 }
1763 }
1764 return *lenp;
1765}
1766
1767/*
1768 * looks for stat data, then copies fields to it, marks the buffer
1769 * containing stat data as dirty
1770 */
1771/*
1772 * reiserfs inodes are never really dirty, since the dirty inode call
1773 * always logs them. This call allows the VFS inode marking routines
1774 * to properly mark inodes for datasync and such, but only actually
1775 * does something when called for a synchronous update.
1776 */
1777int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1778{
1779 struct reiserfs_transaction_handle th;
1780 int jbegin_count = 1;
1781
1782 if (inode->i_sb->s_flags & MS_RDONLY)
1783 return -EROFS;
1784 /*
1785 * memory pressure can sometimes initiate write_inode calls with
1786 * sync == 1,
1787 * these cases are just when the system needs ram, not when the
1788 * inode needs to reach disk for safety, and they can safely be
1789 * ignored because the altered inode has already been logged.
1790 */
1791 if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1792 reiserfs_write_lock(inode->i_sb);
1793 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1794 reiserfs_update_sd(&th, inode);
1795 journal_end_sync(&th);
1796 }
1797 reiserfs_write_unlock(inode->i_sb);
1798 }
1799 return 0;
1800}
1801
1802/*
1803 * stat data of new object is inserted already, this inserts the item
1804 * containing "." and ".." entries
1805 */
1806static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1807 struct inode *inode,
1808 struct item_head *ih, struct treepath *path,
1809 struct inode *dir)
1810{
1811 struct super_block *sb = th->t_super;
1812 char empty_dir[EMPTY_DIR_SIZE];
1813 char *body = empty_dir;
1814 struct cpu_key key;
1815 int retval;
1816
1817 BUG_ON(!th->t_trans_id);
1818
1819 _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1820 le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1821 TYPE_DIRENTRY, 3 /*key length */ );
1822
1823 /*
1824 * compose item head for new item. Directories consist of items of
1825 * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1826 * is done by reiserfs_new_inode
1827 */
1828 if (old_format_only(sb)) {
1829 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1830 TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1831
1832 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1833 ih->ih_key.k_objectid,
1834 INODE_PKEY(dir)->k_dir_id,
1835 INODE_PKEY(dir)->k_objectid);
1836 } else {
1837 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1838 TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1839
1840 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1841 ih->ih_key.k_objectid,
1842 INODE_PKEY(dir)->k_dir_id,
1843 INODE_PKEY(dir)->k_objectid);
1844 }
1845
1846 /* look for place in the tree for new item */
1847 retval = search_item(sb, &key, path);
1848 if (retval == IO_ERROR) {
1849 reiserfs_error(sb, "vs-13080",
1850 "i/o failure occurred creating new directory");
1851 return -EIO;
1852 }
1853 if (retval == ITEM_FOUND) {
1854 pathrelse(path);
1855 reiserfs_warning(sb, "vs-13070",
1856 "object with this key exists (%k)",
1857 &(ih->ih_key));
1858 return -EEXIST;
1859 }
1860
1861 /* insert item, that is empty directory item */
1862 return reiserfs_insert_item(th, path, &key, ih, inode, body);
1863}
1864
1865/*
1866 * stat data of object has been inserted, this inserts the item
1867 * containing the body of symlink
1868 */
1869static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1870 struct inode *inode,
1871 struct item_head *ih,
1872 struct treepath *path, const char *symname,
1873 int item_len)
1874{
1875 struct super_block *sb = th->t_super;
1876 struct cpu_key key;
1877 int retval;
1878
1879 BUG_ON(!th->t_trans_id);
1880
1881 _make_cpu_key(&key, KEY_FORMAT_3_5,
1882 le32_to_cpu(ih->ih_key.k_dir_id),
1883 le32_to_cpu(ih->ih_key.k_objectid),
1884 1, TYPE_DIRECT, 3 /*key length */ );
1885
1886 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1887 0 /*free_space */ );
1888
1889 /* look for place in the tree for new item */
1890 retval = search_item(sb, &key, path);
1891 if (retval == IO_ERROR) {
1892 reiserfs_error(sb, "vs-13080",
1893 "i/o failure occurred creating new symlink");
1894 return -EIO;
1895 }
1896 if (retval == ITEM_FOUND) {
1897 pathrelse(path);
1898 reiserfs_warning(sb, "vs-13080",
1899 "object with this key exists (%k)",
1900 &(ih->ih_key));
1901 return -EEXIST;
1902 }
1903
1904 /* insert item, that is body of symlink */
1905 return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1906}
1907
1908/*
1909 * inserts the stat data into the tree, and then calls
1910 * reiserfs_new_directory (to insert ".", ".." item if new object is
1911 * directory) or reiserfs_new_symlink (to insert symlink body if new
1912 * object is symlink) or nothing (if new object is regular file)
1913
1914 * NOTE! uid and gid must already be set in the inode. If we return
1915 * non-zero due to an error, we have to drop the quota previously allocated
1916 * for the fresh inode. This can only be done outside a transaction, so
1917 * if we return non-zero, we also end the transaction.
1918 *
1919 * @th: active transaction handle
1920 * @dir: parent directory for new inode
1921 * @mode: mode of new inode
1922 * @symname: symlink contents if inode is symlink
1923 * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1924 * symlinks
1925 * @inode: inode to be filled
1926 * @security: optional security context to associate with this inode
1927 */
1928int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1929 struct inode *dir, umode_t mode, const char *symname,
1930 /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1931 strlen (symname) for symlinks) */
1932 loff_t i_size, struct dentry *dentry,
1933 struct inode *inode,
1934 struct reiserfs_security_handle *security)
1935{
1936 struct super_block *sb = dir->i_sb;
1937 struct reiserfs_iget_args args;
1938 INITIALIZE_PATH(path_to_key);
1939 struct cpu_key key;
1940 struct item_head ih;
1941 struct stat_data sd;
1942 int retval;
1943 int err;
1944 int depth;
1945
1946 BUG_ON(!th->t_trans_id);
1947
1948 depth = reiserfs_write_unlock_nested(sb);
1949 err = dquot_alloc_inode(inode);
1950 reiserfs_write_lock_nested(sb, depth);
1951 if (err)
1952 goto out_end_trans;
1953 if (!dir->i_nlink) {
1954 err = -EPERM;
1955 goto out_bad_inode;
1956 }
1957
1958 /* item head of new item */
1959 ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1960 ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1961 if (!ih.ih_key.k_objectid) {
1962 err = -ENOMEM;
1963 goto out_bad_inode;
1964 }
1965 args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1966 if (old_format_only(sb))
1967 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1968 TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1969 else
1970 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1971 TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1972 memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1973 args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1974
1975 depth = reiserfs_write_unlock_nested(inode->i_sb);
1976 err = insert_inode_locked4(inode, args.objectid,
1977 reiserfs_find_actor, &args);
1978 reiserfs_write_lock_nested(inode->i_sb, depth);
1979 if (err) {
1980 err = -EINVAL;
1981 goto out_bad_inode;
1982 }
1983
1984 if (old_format_only(sb))
1985 /*
1986 * not a perfect generation count, as object ids can be reused,
1987 * but this is as good as reiserfs can do right now.
1988 * note that the private part of inode isn't filled in yet,
1989 * we have to use the directory.
1990 */
1991 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1992 else
1993#if defined( USE_INODE_GENERATION_COUNTER )
1994 inode->i_generation =
1995 le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1996#else
1997 inode->i_generation = ++event;
1998#endif
1999
2000 /* fill stat data */
2001 set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
2002
2003 /* uid and gid must already be set by the caller for quota init */
2004
2005 /* symlink cannot be immutable or append only, right? */
2006 if (S_ISLNK(inode->i_mode))
2007 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
2008
2009 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
2010 inode->i_size = i_size;
2011 inode->i_blocks = 0;
2012 inode->i_bytes = 0;
2013 REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
2014 U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
2015
2016 INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
2017 REISERFS_I(inode)->i_flags = 0;
2018 REISERFS_I(inode)->i_prealloc_block = 0;
2019 REISERFS_I(inode)->i_prealloc_count = 0;
2020 REISERFS_I(inode)->i_trans_id = 0;
2021 REISERFS_I(inode)->i_jl = NULL;
2022 REISERFS_I(inode)->i_attrs =
2023 REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2024 sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2025 reiserfs_init_xattr_rwsem(inode);
2026
2027 /* key to search for correct place for new stat data */
2028 _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2029 le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2030 TYPE_STAT_DATA, 3 /*key length */ );
2031
2032 /* find proper place for inserting of stat data */
2033 retval = search_item(sb, &key, &path_to_key);
2034 if (retval == IO_ERROR) {
2035 err = -EIO;
2036 goto out_bad_inode;
2037 }
2038 if (retval == ITEM_FOUND) {
2039 pathrelse(&path_to_key);
2040 err = -EEXIST;
2041 goto out_bad_inode;
2042 }
2043 if (old_format_only(sb)) {
2044 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
2045 if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2046 pathrelse(&path_to_key);
2047 err = -EINVAL;
2048 goto out_bad_inode;
2049 }
2050 inode2sd_v1(&sd, inode, inode->i_size);
2051 } else {
2052 inode2sd(&sd, inode, inode->i_size);
2053 }
2054 /*
2055 * store in in-core inode the key of stat data and version all
2056 * object items will have (directory items will have old offset
2057 * format, other new objects will consist of new items)
2058 */
2059 if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2060 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2061 else
2062 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2063 if (old_format_only(sb))
2064 set_inode_sd_version(inode, STAT_DATA_V1);
2065 else
2066 set_inode_sd_version(inode, STAT_DATA_V2);
2067
2068 /* insert the stat data into the tree */
2069#ifdef DISPLACE_NEW_PACKING_LOCALITIES
2070 if (REISERFS_I(dir)->new_packing_locality)
2071 th->displace_new_blocks = 1;
2072#endif
2073 retval =
2074 reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2075 (char *)(&sd));
2076 if (retval) {
2077 err = retval;
2078 reiserfs_check_path(&path_to_key);
2079 goto out_bad_inode;
2080 }
2081#ifdef DISPLACE_NEW_PACKING_LOCALITIES
2082 if (!th->displace_new_blocks)
2083 REISERFS_I(dir)->new_packing_locality = 0;
2084#endif
2085 if (S_ISDIR(mode)) {
2086 /* insert item with "." and ".." */
2087 retval =
2088 reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2089 }
2090
2091 if (S_ISLNK(mode)) {
2092 /* insert body of symlink */
2093 if (!old_format_only(sb))
2094 i_size = ROUND_UP(i_size);
2095 retval =
2096 reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2097 i_size);
2098 }
2099 if (retval) {
2100 err = retval;
2101 reiserfs_check_path(&path_to_key);
2102 journal_end(th);
2103 goto out_inserted_sd;
2104 }
2105
2106 if (reiserfs_posixacl(inode->i_sb)) {
2107 reiserfs_write_unlock(inode->i_sb);
2108 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2109 reiserfs_write_lock(inode->i_sb);
2110 if (retval) {
2111 err = retval;
2112 reiserfs_check_path(&path_to_key);
2113 journal_end(th);
2114 goto out_inserted_sd;
2115 }
2116 } else if (inode->i_sb->s_flags & MS_POSIXACL) {
2117 reiserfs_warning(inode->i_sb, "jdm-13090",
2118 "ACLs aren't enabled in the fs, "
2119 "but vfs thinks they are!");
2120 } else if (IS_PRIVATE(dir))
2121 inode->i_flags |= S_PRIVATE;
2122
2123 if (security->name) {
2124 reiserfs_write_unlock(inode->i_sb);
2125 retval = reiserfs_security_write(th, inode, security);
2126 reiserfs_write_lock(inode->i_sb);
2127 if (retval) {
2128 err = retval;
2129 reiserfs_check_path(&path_to_key);
2130 retval = journal_end(th);
2131 if (retval)
2132 err = retval;
2133 goto out_inserted_sd;
2134 }
2135 }
2136
2137 reiserfs_update_sd(th, inode);
2138 reiserfs_check_path(&path_to_key);
2139
2140 return 0;
2141
2142out_bad_inode:
2143 /* Invalidate the object, nothing was inserted yet */
2144 INODE_PKEY(inode)->k_objectid = 0;
2145
2146 /* Quota change must be inside a transaction for journaling */
2147 depth = reiserfs_write_unlock_nested(inode->i_sb);
2148 dquot_free_inode(inode);
2149 reiserfs_write_lock_nested(inode->i_sb, depth);
2150
2151out_end_trans:
2152 journal_end(th);
2153 /*
2154 * Drop can be outside and it needs more credits so it's better
2155 * to have it outside
2156 */
2157 depth = reiserfs_write_unlock_nested(inode->i_sb);
2158 dquot_drop(inode);
2159 reiserfs_write_lock_nested(inode->i_sb, depth);
2160 inode->i_flags |= S_NOQUOTA;
2161 make_bad_inode(inode);
2162
2163out_inserted_sd:
2164 clear_nlink(inode);
2165 th->t_trans_id = 0; /* so the caller can't use this handle later */
2166 unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
2167 iput(inode);
2168 return err;
2169}
2170
2171/*
2172 * finds the tail page in the page cache,
2173 * reads the last block in.
2174 *
2175 * On success, page_result is set to a locked, pinned page, and bh_result
2176 * is set to an up to date buffer for the last block in the file. returns 0.
2177 *
2178 * tail conversion is not done, so bh_result might not be valid for writing
2179 * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2180 * trying to write the block.
2181 *
2182 * on failure, nonzero is returned, page_result and bh_result are untouched.
2183 */
2184static int grab_tail_page(struct inode *inode,
2185 struct page **page_result,
2186 struct buffer_head **bh_result)
2187{
2188
2189 /*
2190 * we want the page with the last byte in the file,
2191 * not the page that will hold the next byte for appending
2192 */
2193 unsigned long index = (inode->i_size - 1) >> PAGE_SHIFT;
2194 unsigned long pos = 0;
2195 unsigned long start = 0;
2196 unsigned long blocksize = inode->i_sb->s_blocksize;
2197 unsigned long offset = (inode->i_size) & (PAGE_SIZE - 1);
2198 struct buffer_head *bh;
2199 struct buffer_head *head;
2200 struct page *page;
2201 int error;
2202
2203 /*
2204 * we know that we are only called with inode->i_size > 0.
2205 * we also know that a file tail can never be as big as a block
2206 * If i_size % blocksize == 0, our file is currently block aligned
2207 * and it won't need converting or zeroing after a truncate.
2208 */
2209 if ((offset & (blocksize - 1)) == 0) {
2210 return -ENOENT;
2211 }
2212 page = grab_cache_page(inode->i_mapping, index);
2213 error = -ENOMEM;
2214 if (!page) {
2215 goto out;
2216 }
2217 /* start within the page of the last block in the file */
2218 start = (offset / blocksize) * blocksize;
2219
2220 error = __block_write_begin(page, start, offset - start,
2221 reiserfs_get_block_create_0);
2222 if (error)
2223 goto unlock;
2224
2225 head = page_buffers(page);
2226 bh = head;
2227 do {
2228 if (pos >= start) {
2229 break;
2230 }
2231 bh = bh->b_this_page;
2232 pos += blocksize;
2233 } while (bh != head);
2234
2235 if (!buffer_uptodate(bh)) {
2236 /*
2237 * note, this should never happen, prepare_write should be
2238 * taking care of this for us. If the buffer isn't up to
2239 * date, I've screwed up the code to find the buffer, or the
2240 * code to call prepare_write
2241 */
2242 reiserfs_error(inode->i_sb, "clm-6000",
2243 "error reading block %lu", bh->b_blocknr);
2244 error = -EIO;
2245 goto unlock;
2246 }
2247 *bh_result = bh;
2248 *page_result = page;
2249
2250out:
2251 return error;
2252
2253unlock:
2254 unlock_page(page);
2255 put_page(page);
2256 return error;
2257}
2258
2259/*
2260 * vfs version of truncate file. Must NOT be called with
2261 * a transaction already started.
2262 *
2263 * some code taken from block_truncate_page
2264 */
2265int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2266{
2267 struct reiserfs_transaction_handle th;
2268 /* we want the offset for the first byte after the end of the file */
2269 unsigned long offset = inode->i_size & (PAGE_SIZE - 1);
2270 unsigned blocksize = inode->i_sb->s_blocksize;
2271 unsigned length;
2272 struct page *page = NULL;
2273 int error;
2274 struct buffer_head *bh = NULL;
2275 int err2;
2276
2277 reiserfs_write_lock(inode->i_sb);
2278
2279 if (inode->i_size > 0) {
2280 error = grab_tail_page(inode, &page, &bh);
2281 if (error) {
2282 /*
2283 * -ENOENT means we truncated past the end of the
2284 * file, and get_block_create_0 could not find a
2285 * block to read in, which is ok.
2286 */
2287 if (error != -ENOENT)
2288 reiserfs_error(inode->i_sb, "clm-6001",
2289 "grab_tail_page failed %d",
2290 error);
2291 page = NULL;
2292 bh = NULL;
2293 }
2294 }
2295
2296 /*
2297 * so, if page != NULL, we have a buffer head for the offset at
2298 * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2299 * then we have an unformatted node. Otherwise, we have a direct item,
2300 * and no zeroing is required on disk. We zero after the truncate,
2301 * because the truncate might pack the item anyway
2302 * (it will unmap bh if it packs).
2303 *
2304 * it is enough to reserve space in transaction for 2 balancings:
2305 * one for "save" link adding and another for the first
2306 * cut_from_item. 1 is for update_sd
2307 */
2308 error = journal_begin(&th, inode->i_sb,
2309 JOURNAL_PER_BALANCE_CNT * 2 + 1);
2310 if (error)
2311 goto out;
2312 reiserfs_update_inode_transaction(inode);
2313 if (update_timestamps)
2314 /*
2315 * we are doing real truncate: if the system crashes
2316 * before the last transaction of truncating gets committed
2317 * - on reboot the file either appears truncated properly
2318 * or not truncated at all
2319 */
2320 add_save_link(&th, inode, 1);
2321 err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2322 error = journal_end(&th);
2323 if (error)
2324 goto out;
2325
2326 /* check reiserfs_do_truncate after ending the transaction */
2327 if (err2) {
2328 error = err2;
2329 goto out;
2330 }
2331
2332 if (update_timestamps) {
2333 error = remove_save_link(inode, 1 /* truncate */);
2334 if (error)
2335 goto out;
2336 }
2337
2338 if (page) {
2339 length = offset & (blocksize - 1);
2340 /* if we are not on a block boundary */
2341 if (length) {
2342 length = blocksize - length;
2343 zero_user(page, offset, length);
2344 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2345 mark_buffer_dirty(bh);
2346 }
2347 }
2348 unlock_page(page);
2349 put_page(page);
2350 }
2351
2352 reiserfs_write_unlock(inode->i_sb);
2353
2354 return 0;
2355out:
2356 if (page) {
2357 unlock_page(page);
2358 put_page(page);
2359 }
2360
2361 reiserfs_write_unlock(inode->i_sb);
2362
2363 return error;
2364}
2365
2366static int map_block_for_writepage(struct inode *inode,
2367 struct buffer_head *bh_result,
2368 unsigned long block)
2369{
2370 struct reiserfs_transaction_handle th;
2371 int fs_gen;
2372 struct item_head tmp_ih;
2373 struct item_head *ih;
2374 struct buffer_head *bh;
2375 __le32 *item;
2376 struct cpu_key key;
2377 INITIALIZE_PATH(path);
2378 int pos_in_item;
2379 int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2380 loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2381 int retval;
2382 int use_get_block = 0;
2383 int bytes_copied = 0;
2384 int copy_size;
2385 int trans_running = 0;
2386
2387 /*
2388 * catch places below that try to log something without
2389 * starting a trans
2390 */
2391 th.t_trans_id = 0;
2392
2393 if (!buffer_uptodate(bh_result)) {
2394 return -EIO;
2395 }
2396
2397 kmap(bh_result->b_page);
2398start_over:
2399 reiserfs_write_lock(inode->i_sb);
2400 make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2401
2402research:
2403 retval = search_for_position_by_key(inode->i_sb, &key, &path);
2404 if (retval != POSITION_FOUND) {
2405 use_get_block = 1;
2406 goto out;
2407 }
2408
2409 bh = get_last_bh(&path);
2410 ih = tp_item_head(&path);
2411 item = tp_item_body(&path);
2412 pos_in_item = path.pos_in_item;
2413
2414 /* we've found an unformatted node */
2415 if (indirect_item_found(retval, ih)) {
2416 if (bytes_copied > 0) {
2417 reiserfs_warning(inode->i_sb, "clm-6002",
2418 "bytes_copied %d", bytes_copied);
2419 }
2420 if (!get_block_num(item, pos_in_item)) {
2421 /* crap, we are writing to a hole */
2422 use_get_block = 1;
2423 goto out;
2424 }
2425 set_block_dev_mapped(bh_result,
2426 get_block_num(item, pos_in_item), inode);
2427 } else if (is_direct_le_ih(ih)) {
2428 char *p;
2429 p = page_address(bh_result->b_page);
2430 p += (byte_offset - 1) & (PAGE_SIZE - 1);
2431 copy_size = ih_item_len(ih) - pos_in_item;
2432
2433 fs_gen = get_generation(inode->i_sb);
2434 copy_item_head(&tmp_ih, ih);
2435
2436 if (!trans_running) {
2437 /* vs-3050 is gone, no need to drop the path */
2438 retval = journal_begin(&th, inode->i_sb, jbegin_count);
2439 if (retval)
2440 goto out;
2441 reiserfs_update_inode_transaction(inode);
2442 trans_running = 1;
2443 if (fs_changed(fs_gen, inode->i_sb)
2444 && item_moved(&tmp_ih, &path)) {
2445 reiserfs_restore_prepared_buffer(inode->i_sb,
2446 bh);
2447 goto research;
2448 }
2449 }
2450
2451 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2452
2453 if (fs_changed(fs_gen, inode->i_sb)
2454 && item_moved(&tmp_ih, &path)) {
2455 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2456 goto research;
2457 }
2458
2459 memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2460 copy_size);
2461
2462 journal_mark_dirty(&th, bh);
2463 bytes_copied += copy_size;
2464 set_block_dev_mapped(bh_result, 0, inode);
2465
2466 /* are there still bytes left? */
2467 if (bytes_copied < bh_result->b_size &&
2468 (byte_offset + bytes_copied) < inode->i_size) {
2469 set_cpu_key_k_offset(&key,
2470 cpu_key_k_offset(&key) +
2471 copy_size);
2472 goto research;
2473 }
2474 } else {
2475 reiserfs_warning(inode->i_sb, "clm-6003",
2476 "bad item inode %lu", inode->i_ino);
2477 retval = -EIO;
2478 goto out;
2479 }
2480 retval = 0;
2481
2482out:
2483 pathrelse(&path);
2484 if (trans_running) {
2485 int err = journal_end(&th);
2486 if (err)
2487 retval = err;
2488 trans_running = 0;
2489 }
2490 reiserfs_write_unlock(inode->i_sb);
2491
2492 /* this is where we fill in holes in the file. */
2493 if (use_get_block) {
2494 retval = reiserfs_get_block(inode, block, bh_result,
2495 GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2496 | GET_BLOCK_NO_DANGLE);
2497 if (!retval) {
2498 if (!buffer_mapped(bh_result)
2499 || bh_result->b_blocknr == 0) {
2500 /* get_block failed to find a mapped unformatted node. */
2501 use_get_block = 0;
2502 goto start_over;
2503 }
2504 }
2505 }
2506 kunmap(bh_result->b_page);
2507
2508 if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2509 /*
2510 * we've copied data from the page into the direct item, so the
2511 * buffer in the page is now clean, mark it to reflect that.
2512 */
2513 lock_buffer(bh_result);
2514 clear_buffer_dirty(bh_result);
2515 unlock_buffer(bh_result);
2516 }
2517 return retval;
2518}
2519
2520/*
2521 * mason@suse.com: updated in 2.5.54 to follow the same general io
2522 * start/recovery path as __block_write_full_page, along with special
2523 * code to handle reiserfs tails.
2524 */
2525static int reiserfs_write_full_page(struct page *page,
2526 struct writeback_control *wbc)
2527{
2528 struct inode *inode = page->mapping->host;
2529 unsigned long end_index = inode->i_size >> PAGE_SHIFT;
2530 int error = 0;
2531 unsigned long block;
2532 sector_t last_block;
2533 struct buffer_head *head, *bh;
2534 int partial = 0;
2535 int nr = 0;
2536 int checked = PageChecked(page);
2537 struct reiserfs_transaction_handle th;
2538 struct super_block *s = inode->i_sb;
2539 int bh_per_page = PAGE_SIZE / s->s_blocksize;
2540 th.t_trans_id = 0;
2541
2542 /* no logging allowed when nonblocking or from PF_MEMALLOC */
2543 if (checked && (current->flags & PF_MEMALLOC)) {
2544 redirty_page_for_writepage(wbc, page);
2545 unlock_page(page);
2546 return 0;
2547 }
2548
2549 /*
2550 * The page dirty bit is cleared before writepage is called, which
2551 * means we have to tell create_empty_buffers to make dirty buffers
2552 * The page really should be up to date at this point, so tossing
2553 * in the BH_Uptodate is just a sanity check.
2554 */
2555 if (!page_has_buffers(page)) {
2556 create_empty_buffers(page, s->s_blocksize,
2557 (1 << BH_Dirty) | (1 << BH_Uptodate));
2558 }
2559 head = page_buffers(page);
2560
2561 /*
2562 * last page in the file, zero out any contents past the
2563 * last byte in the file
2564 */
2565 if (page->index >= end_index) {
2566 unsigned last_offset;
2567
2568 last_offset = inode->i_size & (PAGE_SIZE - 1);
2569 /* no file contents in this page */
2570 if (page->index >= end_index + 1 || !last_offset) {
2571 unlock_page(page);
2572 return 0;
2573 }
2574 zero_user_segment(page, last_offset, PAGE_SIZE);
2575 }
2576 bh = head;
2577 block = page->index << (PAGE_SHIFT - s->s_blocksize_bits);
2578 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2579 /* first map all the buffers, logging any direct items we find */
2580 do {
2581 if (block > last_block) {
2582 /*
2583 * This can happen when the block size is less than
2584 * the page size. The corresponding bytes in the page
2585 * were zero filled above
2586 */
2587 clear_buffer_dirty(bh);
2588 set_buffer_uptodate(bh);
2589 } else if ((checked || buffer_dirty(bh)) &&
2590 (!buffer_mapped(bh) || (buffer_mapped(bh)
2591 && bh->b_blocknr ==
2592 0))) {
2593 /*
2594 * not mapped yet, or it points to a direct item, search
2595 * the btree for the mapping info, and log any direct
2596 * items found
2597 */
2598 if ((error = map_block_for_writepage(inode, bh, block))) {
2599 goto fail;
2600 }
2601 }
2602 bh = bh->b_this_page;
2603 block++;
2604 } while (bh != head);
2605
2606 /*
2607 * we start the transaction after map_block_for_writepage,
2608 * because it can create holes in the file (an unbounded operation).
2609 * starting it here, we can make a reliable estimate for how many
2610 * blocks we're going to log
2611 */
2612 if (checked) {
2613 ClearPageChecked(page);
2614 reiserfs_write_lock(s);
2615 error = journal_begin(&th, s, bh_per_page + 1);
2616 if (error) {
2617 reiserfs_write_unlock(s);
2618 goto fail;
2619 }
2620 reiserfs_update_inode_transaction(inode);
2621 }
2622 /* now go through and lock any dirty buffers on the page */
2623 do {
2624 get_bh(bh);
2625 if (!buffer_mapped(bh))
2626 continue;
2627 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2628 continue;
2629
2630 if (checked) {
2631 reiserfs_prepare_for_journal(s, bh, 1);
2632 journal_mark_dirty(&th, bh);
2633 continue;
2634 }
2635 /*
2636 * from this point on, we know the buffer is mapped to a
2637 * real block and not a direct item
2638 */
2639 if (wbc->sync_mode != WB_SYNC_NONE) {
2640 lock_buffer(bh);
2641 } else {
2642 if (!trylock_buffer(bh)) {
2643 redirty_page_for_writepage(wbc, page);
2644 continue;
2645 }
2646 }
2647 if (test_clear_buffer_dirty(bh)) {
2648 mark_buffer_async_write(bh);
2649 } else {
2650 unlock_buffer(bh);
2651 }
2652 } while ((bh = bh->b_this_page) != head);
2653
2654 if (checked) {
2655 error = journal_end(&th);
2656 reiserfs_write_unlock(s);
2657 if (error)
2658 goto fail;
2659 }
2660 BUG_ON(PageWriteback(page));
2661 set_page_writeback(page);
2662 unlock_page(page);
2663
2664 /*
2665 * since any buffer might be the only dirty buffer on the page,
2666 * the first submit_bh can bring the page out of writeback.
2667 * be careful with the buffers.
2668 */
2669 do {
2670 struct buffer_head *next = bh->b_this_page;
2671 if (buffer_async_write(bh)) {
2672 submit_bh(REQ_OP_WRITE, 0, bh);
2673 nr++;
2674 }
2675 put_bh(bh);
2676 bh = next;
2677 } while (bh != head);
2678
2679 error = 0;
2680done:
2681 if (nr == 0) {
2682 /*
2683 * if this page only had a direct item, it is very possible for
2684 * no io to be required without there being an error. Or,
2685 * someone else could have locked them and sent them down the
2686 * pipe without locking the page
2687 */
2688 bh = head;
2689 do {
2690 if (!buffer_uptodate(bh)) {
2691 partial = 1;
2692 break;
2693 }
2694 bh = bh->b_this_page;
2695 } while (bh != head);
2696 if (!partial)
2697 SetPageUptodate(page);
2698 end_page_writeback(page);
2699 }
2700 return error;
2701
2702fail:
2703 /*
2704 * catches various errors, we need to make sure any valid dirty blocks
2705 * get to the media. The page is currently locked and not marked for
2706 * writeback
2707 */
2708 ClearPageUptodate(page);
2709 bh = head;
2710 do {
2711 get_bh(bh);
2712 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2713 lock_buffer(bh);
2714 mark_buffer_async_write(bh);
2715 } else {
2716 /*
2717 * clear any dirty bits that might have come from
2718 * getting attached to a dirty page
2719 */
2720 clear_buffer_dirty(bh);
2721 }
2722 bh = bh->b_this_page;
2723 } while (bh != head);
2724 SetPageError(page);
2725 BUG_ON(PageWriteback(page));
2726 set_page_writeback(page);
2727 unlock_page(page);
2728 do {
2729 struct buffer_head *next = bh->b_this_page;
2730 if (buffer_async_write(bh)) {
2731 clear_buffer_dirty(bh);
2732 submit_bh(REQ_OP_WRITE, 0, bh);
2733 nr++;
2734 }
2735 put_bh(bh);
2736 bh = next;
2737 } while (bh != head);
2738 goto done;
2739}
2740
2741static int reiserfs_readpage(struct file *f, struct page *page)
2742{
2743 return block_read_full_page(page, reiserfs_get_block);
2744}
2745
2746static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2747{
2748 struct inode *inode = page->mapping->host;
2749 reiserfs_wait_on_write_block(inode->i_sb);
2750 return reiserfs_write_full_page(page, wbc);
2751}
2752
2753static void reiserfs_truncate_failed_write(struct inode *inode)
2754{
2755 truncate_inode_pages(inode->i_mapping, inode->i_size);
2756 reiserfs_truncate_file(inode, 0);
2757}
2758
2759static int reiserfs_write_begin(struct file *file,
2760 struct address_space *mapping,
2761 loff_t pos, unsigned len, unsigned flags,
2762 struct page **pagep, void **fsdata)
2763{
2764 struct inode *inode;
2765 struct page *page;
2766 pgoff_t index;
2767 int ret;
2768 int old_ref = 0;
2769
2770 inode = mapping->host;
2771 *fsdata = NULL;
2772 if (flags & AOP_FLAG_CONT_EXPAND &&
2773 (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2774 pos ++;
2775 *fsdata = (void *)(unsigned long)flags;
2776 }
2777
2778 index = pos >> PAGE_SHIFT;
2779 page = grab_cache_page_write_begin(mapping, index, flags);
2780 if (!page)
2781 return -ENOMEM;
2782 *pagep = page;
2783
2784 reiserfs_wait_on_write_block(inode->i_sb);
2785 fix_tail_page_for_writing(page);
2786 if (reiserfs_transaction_running(inode->i_sb)) {
2787 struct reiserfs_transaction_handle *th;
2788 th = (struct reiserfs_transaction_handle *)current->
2789 journal_info;
2790 BUG_ON(!th->t_refcount);
2791 BUG_ON(!th->t_trans_id);
2792 old_ref = th->t_refcount;
2793 th->t_refcount++;
2794 }
2795 ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2796 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2797 struct reiserfs_transaction_handle *th = current->journal_info;
2798 /*
2799 * this gets a little ugly. If reiserfs_get_block returned an
2800 * error and left a transacstion running, we've got to close
2801 * it, and we've got to free handle if it was a persistent
2802 * transaction.
2803 *
2804 * But, if we had nested into an existing transaction, we need
2805 * to just drop the ref count on the handle.
2806 *
2807 * If old_ref == 0, the transaction is from reiserfs_get_block,
2808 * and it was a persistent trans. Otherwise, it was nested
2809 * above.
2810 */
2811 if (th->t_refcount > old_ref) {
2812 if (old_ref)
2813 th->t_refcount--;
2814 else {
2815 int err;
2816 reiserfs_write_lock(inode->i_sb);
2817 err = reiserfs_end_persistent_transaction(th);
2818 reiserfs_write_unlock(inode->i_sb);
2819 if (err)
2820 ret = err;
2821 }
2822 }
2823 }
2824 if (ret) {
2825 unlock_page(page);
2826 put_page(page);
2827 /* Truncate allocated blocks */
2828 reiserfs_truncate_failed_write(inode);
2829 }
2830 return ret;
2831}
2832
2833int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2834{
2835 struct inode *inode = page->mapping->host;
2836 int ret;
2837 int old_ref = 0;
2838 int depth;
2839
2840 depth = reiserfs_write_unlock_nested(inode->i_sb);
2841 reiserfs_wait_on_write_block(inode->i_sb);
2842 reiserfs_write_lock_nested(inode->i_sb, depth);
2843
2844 fix_tail_page_for_writing(page);
2845 if (reiserfs_transaction_running(inode->i_sb)) {
2846 struct reiserfs_transaction_handle *th;
2847 th = (struct reiserfs_transaction_handle *)current->
2848 journal_info;
2849 BUG_ON(!th->t_refcount);
2850 BUG_ON(!th->t_trans_id);
2851 old_ref = th->t_refcount;
2852 th->t_refcount++;
2853 }
2854
2855 ret = __block_write_begin(page, from, len, reiserfs_get_block);
2856 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2857 struct reiserfs_transaction_handle *th = current->journal_info;
2858 /*
2859 * this gets a little ugly. If reiserfs_get_block returned an
2860 * error and left a transacstion running, we've got to close
2861 * it, and we've got to free handle if it was a persistent
2862 * transaction.
2863 *
2864 * But, if we had nested into an existing transaction, we need
2865 * to just drop the ref count on the handle.
2866 *
2867 * If old_ref == 0, the transaction is from reiserfs_get_block,
2868 * and it was a persistent trans. Otherwise, it was nested
2869 * above.
2870 */
2871 if (th->t_refcount > old_ref) {
2872 if (old_ref)
2873 th->t_refcount--;
2874 else {
2875 int err;
2876 reiserfs_write_lock(inode->i_sb);
2877 err = reiserfs_end_persistent_transaction(th);
2878 reiserfs_write_unlock(inode->i_sb);
2879 if (err)
2880 ret = err;
2881 }
2882 }
2883 }
2884 return ret;
2885
2886}
2887
2888static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2889{
2890 return generic_block_bmap(as, block, reiserfs_bmap);
2891}
2892
2893static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2894 loff_t pos, unsigned len, unsigned copied,
2895 struct page *page, void *fsdata)
2896{
2897 struct inode *inode = page->mapping->host;
2898 int ret = 0;
2899 int update_sd = 0;
2900 struct reiserfs_transaction_handle *th;
2901 unsigned start;
2902 bool locked = false;
2903
2904 if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2905 pos ++;
2906
2907 reiserfs_wait_on_write_block(inode->i_sb);
2908 if (reiserfs_transaction_running(inode->i_sb))
2909 th = current->journal_info;
2910 else
2911 th = NULL;
2912
2913 start = pos & (PAGE_SIZE - 1);
2914 if (unlikely(copied < len)) {
2915 if (!PageUptodate(page))
2916 copied = 0;
2917
2918 page_zero_new_buffers(page, start + copied, start + len);
2919 }
2920 flush_dcache_page(page);
2921
2922 reiserfs_commit_page(inode, page, start, start + copied);
2923
2924 /*
2925 * generic_commit_write does this for us, but does not update the
2926 * transaction tracking stuff when the size changes. So, we have
2927 * to do the i_size updates here.
2928 */
2929 if (pos + copied > inode->i_size) {
2930 struct reiserfs_transaction_handle myth;
2931 reiserfs_write_lock(inode->i_sb);
2932 locked = true;
2933 /*
2934 * If the file have grown beyond the border where it
2935 * can have a tail, unmark it as needing a tail
2936 * packing
2937 */
2938 if ((have_large_tails(inode->i_sb)
2939 && inode->i_size > i_block_size(inode) * 4)
2940 || (have_small_tails(inode->i_sb)
2941 && inode->i_size > i_block_size(inode)))
2942 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2943
2944 ret = journal_begin(&myth, inode->i_sb, 1);
2945 if (ret)
2946 goto journal_error;
2947
2948 reiserfs_update_inode_transaction(inode);
2949 inode->i_size = pos + copied;
2950 /*
2951 * this will just nest into our transaction. It's important
2952 * to use mark_inode_dirty so the inode gets pushed around on
2953 * the dirty lists, and so that O_SYNC works as expected
2954 */
2955 mark_inode_dirty(inode);
2956 reiserfs_update_sd(&myth, inode);
2957 update_sd = 1;
2958 ret = journal_end(&myth);
2959 if (ret)
2960 goto journal_error;
2961 }
2962 if (th) {
2963 if (!locked) {
2964 reiserfs_write_lock(inode->i_sb);
2965 locked = true;
2966 }
2967 if (!update_sd)
2968 mark_inode_dirty(inode);
2969 ret = reiserfs_end_persistent_transaction(th);
2970 if (ret)
2971 goto out;
2972 }
2973
2974out:
2975 if (locked)
2976 reiserfs_write_unlock(inode->i_sb);
2977 unlock_page(page);
2978 put_page(page);
2979
2980 if (pos + len > inode->i_size)
2981 reiserfs_truncate_failed_write(inode);
2982
2983 return ret == 0 ? copied : ret;
2984
2985journal_error:
2986 reiserfs_write_unlock(inode->i_sb);
2987 locked = false;
2988 if (th) {
2989 if (!update_sd)
2990 reiserfs_update_sd(th, inode);
2991 ret = reiserfs_end_persistent_transaction(th);
2992 }
2993 goto out;
2994}
2995
2996int reiserfs_commit_write(struct file *f, struct page *page,
2997 unsigned from, unsigned to)
2998{
2999 struct inode *inode = page->mapping->host;
3000 loff_t pos = ((loff_t) page->index << PAGE_SHIFT) + to;
3001 int ret = 0;
3002 int update_sd = 0;
3003 struct reiserfs_transaction_handle *th = NULL;
3004 int depth;
3005
3006 depth = reiserfs_write_unlock_nested(inode->i_sb);
3007 reiserfs_wait_on_write_block(inode->i_sb);
3008 reiserfs_write_lock_nested(inode->i_sb, depth);
3009
3010 if (reiserfs_transaction_running(inode->i_sb)) {
3011 th = current->journal_info;
3012 }
3013 reiserfs_commit_page(inode, page, from, to);
3014
3015 /*
3016 * generic_commit_write does this for us, but does not update the
3017 * transaction tracking stuff when the size changes. So, we have
3018 * to do the i_size updates here.
3019 */
3020 if (pos > inode->i_size) {
3021 struct reiserfs_transaction_handle myth;
3022 /*
3023 * If the file have grown beyond the border where it
3024 * can have a tail, unmark it as needing a tail
3025 * packing
3026 */
3027 if ((have_large_tails(inode->i_sb)
3028 && inode->i_size > i_block_size(inode) * 4)
3029 || (have_small_tails(inode->i_sb)
3030 && inode->i_size > i_block_size(inode)))
3031 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3032
3033 ret = journal_begin(&myth, inode->i_sb, 1);
3034 if (ret)
3035 goto journal_error;
3036
3037 reiserfs_update_inode_transaction(inode);
3038 inode->i_size = pos;
3039 /*
3040 * this will just nest into our transaction. It's important
3041 * to use mark_inode_dirty so the inode gets pushed around
3042 * on the dirty lists, and so that O_SYNC works as expected
3043 */
3044 mark_inode_dirty(inode);
3045 reiserfs_update_sd(&myth, inode);
3046 update_sd = 1;
3047 ret = journal_end(&myth);
3048 if (ret)
3049 goto journal_error;
3050 }
3051 if (th) {
3052 if (!update_sd)
3053 mark_inode_dirty(inode);
3054 ret = reiserfs_end_persistent_transaction(th);
3055 if (ret)
3056 goto out;
3057 }
3058
3059out:
3060 return ret;
3061
3062journal_error:
3063 if (th) {
3064 if (!update_sd)
3065 reiserfs_update_sd(th, inode);
3066 ret = reiserfs_end_persistent_transaction(th);
3067 }
3068
3069 return ret;
3070}
3071
3072void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3073{
3074 if (reiserfs_attrs(inode->i_sb)) {
3075 if (sd_attrs & REISERFS_SYNC_FL)
3076 inode->i_flags |= S_SYNC;
3077 else
3078 inode->i_flags &= ~S_SYNC;
3079 if (sd_attrs & REISERFS_IMMUTABLE_FL)
3080 inode->i_flags |= S_IMMUTABLE;
3081 else
3082 inode->i_flags &= ~S_IMMUTABLE;
3083 if (sd_attrs & REISERFS_APPEND_FL)
3084 inode->i_flags |= S_APPEND;
3085 else
3086 inode->i_flags &= ~S_APPEND;
3087 if (sd_attrs & REISERFS_NOATIME_FL)
3088 inode->i_flags |= S_NOATIME;
3089 else
3090 inode->i_flags &= ~S_NOATIME;
3091 if (sd_attrs & REISERFS_NOTAIL_FL)
3092 REISERFS_I(inode)->i_flags |= i_nopack_mask;
3093 else
3094 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3095 }
3096}
3097
3098void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
3099{
3100 if (reiserfs_attrs(inode->i_sb)) {
3101 if (inode->i_flags & S_IMMUTABLE)
3102 *sd_attrs |= REISERFS_IMMUTABLE_FL;
3103 else
3104 *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
3105 if (inode->i_flags & S_SYNC)
3106 *sd_attrs |= REISERFS_SYNC_FL;
3107 else
3108 *sd_attrs &= ~REISERFS_SYNC_FL;
3109 if (inode->i_flags & S_NOATIME)
3110 *sd_attrs |= REISERFS_NOATIME_FL;
3111 else
3112 *sd_attrs &= ~REISERFS_NOATIME_FL;
3113 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
3114 *sd_attrs |= REISERFS_NOTAIL_FL;
3115 else
3116 *sd_attrs &= ~REISERFS_NOTAIL_FL;
3117 }
3118}
3119
3120/*
3121 * decide if this buffer needs to stay around for data logging or ordered
3122 * write purposes
3123 */
3124static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
3125{
3126 int ret = 1;
3127 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3128
3129 lock_buffer(bh);
3130 spin_lock(&j->j_dirty_buffers_lock);
3131 if (!buffer_mapped(bh)) {
3132 goto free_jh;
3133 }
3134 /*
3135 * the page is locked, and the only places that log a data buffer
3136 * also lock the page.
3137 */
3138 if (reiserfs_file_data_log(inode)) {
3139 /*
3140 * very conservative, leave the buffer pinned if
3141 * anyone might need it.
3142 */
3143 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3144 ret = 0;
3145 }
3146 } else if (buffer_dirty(bh)) {
3147 struct reiserfs_journal_list *jl;
3148 struct reiserfs_jh *jh = bh->b_private;
3149
3150 /*
3151 * why is this safe?
3152 * reiserfs_setattr updates i_size in the on disk
3153 * stat data before allowing vmtruncate to be called.
3154 *
3155 * If buffer was put onto the ordered list for this
3156 * transaction, we know for sure either this transaction
3157 * or an older one already has updated i_size on disk,
3158 * and this ordered data won't be referenced in the file
3159 * if we crash.
3160 *
3161 * if the buffer was put onto the ordered list for an older
3162 * transaction, we need to leave it around
3163 */
3164 if (jh && (jl = jh->jl)
3165 && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3166 ret = 0;
3167 }
3168free_jh:
3169 if (ret && bh->b_private) {
3170 reiserfs_free_jh(bh);
3171 }
3172 spin_unlock(&j->j_dirty_buffers_lock);
3173 unlock_buffer(bh);
3174 return ret;
3175}
3176
3177/* clm -- taken from fs/buffer.c:block_invalidate_page */
3178static void reiserfs_invalidatepage(struct page *page, unsigned int offset,
3179 unsigned int length)
3180{
3181 struct buffer_head *head, *bh, *next;
3182 struct inode *inode = page->mapping->host;
3183 unsigned int curr_off = 0;
3184 unsigned int stop = offset + length;
3185 int partial_page = (offset || length < PAGE_SIZE);
3186 int ret = 1;
3187
3188 BUG_ON(!PageLocked(page));
3189
3190 if (!partial_page)
3191 ClearPageChecked(page);
3192
3193 if (!page_has_buffers(page))
3194 goto out;
3195
3196 head = page_buffers(page);
3197 bh = head;
3198 do {
3199 unsigned int next_off = curr_off + bh->b_size;
3200 next = bh->b_this_page;
3201
3202 if (next_off > stop)
3203 goto out;
3204
3205 /*
3206 * is this block fully invalidated?
3207 */
3208 if (offset <= curr_off) {
3209 if (invalidatepage_can_drop(inode, bh))
3210 reiserfs_unmap_buffer(bh);
3211 else
3212 ret = 0;
3213 }
3214 curr_off = next_off;
3215 bh = next;
3216 } while (bh != head);
3217
3218 /*
3219 * We release buffers only if the entire page is being invalidated.
3220 * The get_block cached value has been unconditionally invalidated,
3221 * so real IO is not possible anymore.
3222 */
3223 if (!partial_page && ret) {
3224 ret = try_to_release_page(page, 0);
3225 /* maybe should BUG_ON(!ret); - neilb */
3226 }
3227out:
3228 return;
3229}
3230
3231static int reiserfs_set_page_dirty(struct page *page)
3232{
3233 struct inode *inode = page->mapping->host;
3234 if (reiserfs_file_data_log(inode)) {
3235 SetPageChecked(page);
3236 return __set_page_dirty_nobuffers(page);
3237 }
3238 return __set_page_dirty_buffers(page);
3239}
3240
3241/*
3242 * Returns 1 if the page's buffers were dropped. The page is locked.
3243 *
3244 * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3245 * in the buffers at page_buffers(page).
3246 *
3247 * even in -o notail mode, we can't be sure an old mount without -o notail
3248 * didn't create files with tails.
3249 */
3250static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
3251{
3252 struct inode *inode = page->mapping->host;
3253 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3254 struct buffer_head *head;
3255 struct buffer_head *bh;
3256 int ret = 1;
3257
3258 WARN_ON(PageChecked(page));
3259 spin_lock(&j->j_dirty_buffers_lock);
3260 head = page_buffers(page);
3261 bh = head;
3262 do {
3263 if (bh->b_private) {
3264 if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3265 reiserfs_free_jh(bh);
3266 } else {
3267 ret = 0;
3268 break;
3269 }
3270 }
3271 bh = bh->b_this_page;
3272 } while (bh != head);
3273 if (ret)
3274 ret = try_to_free_buffers(page);
3275 spin_unlock(&j->j_dirty_buffers_lock);
3276 return ret;
3277}
3278
3279/*
3280 * We thank Mingming Cao for helping us understand in great detail what
3281 * to do in this section of the code.
3282 */
3283static ssize_t reiserfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3284{
3285 struct file *file = iocb->ki_filp;
3286 struct inode *inode = file->f_mapping->host;
3287 size_t count = iov_iter_count(iter);
3288 ssize_t ret;
3289
3290 ret = blockdev_direct_IO(iocb, inode, iter,
3291 reiserfs_get_blocks_direct_io);
3292
3293 /*
3294 * In case of error extending write may have instantiated a few
3295 * blocks outside i_size. Trim these off again.
3296 */
3297 if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
3298 loff_t isize = i_size_read(inode);
3299 loff_t end = iocb->ki_pos + count;
3300
3301 if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3302 truncate_setsize(inode, isize);
3303 reiserfs_vfs_truncate_file(inode);
3304 }
3305 }
3306
3307 return ret;
3308}
3309
3310int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3311{
3312 struct inode *inode = d_inode(dentry);
3313 unsigned int ia_valid;
3314 int error;
3315
3316 error = setattr_prepare(dentry, attr);
3317 if (error)
3318 return error;
3319
3320 /* must be turned off for recursive notify_change calls */
3321 ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3322
3323 if (is_quota_modification(inode, attr)) {
3324 error = dquot_initialize(inode);
3325 if (error)
3326 return error;
3327 }
3328 reiserfs_write_lock(inode->i_sb);
3329 if (attr->ia_valid & ATTR_SIZE) {
3330 /*
3331 * version 2 items will be caught by the s_maxbytes check
3332 * done for us in vmtruncate
3333 */
3334 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3335 attr->ia_size > MAX_NON_LFS) {
3336 reiserfs_write_unlock(inode->i_sb);
3337 error = -EFBIG;
3338 goto out;
3339 }
3340
3341 inode_dio_wait(inode);
3342
3343 /* fill in hole pointers in the expanding truncate case. */
3344 if (attr->ia_size > inode->i_size) {
3345 error = generic_cont_expand_simple(inode, attr->ia_size);
3346 if (REISERFS_I(inode)->i_prealloc_count > 0) {
3347 int err;
3348 struct reiserfs_transaction_handle th;
3349 /* we're changing at most 2 bitmaps, inode + super */
3350 err = journal_begin(&th, inode->i_sb, 4);
3351 if (!err) {
3352 reiserfs_discard_prealloc(&th, inode);
3353 err = journal_end(&th);
3354 }
3355 if (err)
3356 error = err;
3357 }
3358 if (error) {
3359 reiserfs_write_unlock(inode->i_sb);
3360 goto out;
3361 }
3362 /*
3363 * file size is changed, ctime and mtime are
3364 * to be updated
3365 */
3366 attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3367 }
3368 }
3369 reiserfs_write_unlock(inode->i_sb);
3370
3371 if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3372 ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3373 (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3374 /* stat data of format v3.5 has 16 bit uid and gid */
3375 error = -EINVAL;
3376 goto out;
3377 }
3378
3379 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3380 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3381 struct reiserfs_transaction_handle th;
3382 int jbegin_count =
3383 2 *
3384 (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3385 REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3386 2;
3387
3388 error = reiserfs_chown_xattrs(inode, attr);
3389
3390 if (error)
3391 return error;
3392
3393 /*
3394 * (user+group)*(old+new) structure - we count quota
3395 * info and , inode write (sb, inode)
3396 */
3397 reiserfs_write_lock(inode->i_sb);
3398 error = journal_begin(&th, inode->i_sb, jbegin_count);
3399 reiserfs_write_unlock(inode->i_sb);
3400 if (error)
3401 goto out;
3402 error = dquot_transfer(inode, attr);
3403 reiserfs_write_lock(inode->i_sb);
3404 if (error) {
3405 journal_end(&th);
3406 reiserfs_write_unlock(inode->i_sb);
3407 goto out;
3408 }
3409
3410 /*
3411 * Update corresponding info in inode so that everything
3412 * is in one transaction
3413 */
3414 if (attr->ia_valid & ATTR_UID)
3415 inode->i_uid = attr->ia_uid;
3416 if (attr->ia_valid & ATTR_GID)
3417 inode->i_gid = attr->ia_gid;
3418 mark_inode_dirty(inode);
3419 error = journal_end(&th);
3420 reiserfs_write_unlock(inode->i_sb);
3421 if (error)
3422 goto out;
3423 }
3424
3425 if ((attr->ia_valid & ATTR_SIZE) &&
3426 attr->ia_size != i_size_read(inode)) {
3427 error = inode_newsize_ok(inode, attr->ia_size);
3428 if (!error) {
3429 /*
3430 * Could race against reiserfs_file_release
3431 * if called from NFS, so take tailpack mutex.
3432 */
3433 mutex_lock(&REISERFS_I(inode)->tailpack);
3434 truncate_setsize(inode, attr->ia_size);
3435 reiserfs_truncate_file(inode, 1);
3436 mutex_unlock(&REISERFS_I(inode)->tailpack);
3437 }
3438 }
3439
3440 if (!error) {
3441 setattr_copy(inode, attr);
3442 mark_inode_dirty(inode);
3443 }
3444
3445 if (!error && reiserfs_posixacl(inode->i_sb)) {
3446 if (attr->ia_valid & ATTR_MODE)
3447 error = reiserfs_acl_chmod(inode);
3448 }
3449
3450out:
3451 return error;
3452}
3453
3454const struct address_space_operations reiserfs_address_space_operations = {
3455 .writepage = reiserfs_writepage,
3456 .readpage = reiserfs_readpage,
3457 .readpages = reiserfs_readpages,
3458 .releasepage = reiserfs_releasepage,
3459 .invalidatepage = reiserfs_invalidatepage,
3460 .write_begin = reiserfs_write_begin,
3461 .write_end = reiserfs_write_end,
3462 .bmap = reiserfs_aop_bmap,
3463 .direct_IO = reiserfs_direct_IO,
3464 .set_page_dirty = reiserfs_set_page_dirty,
3465};