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