2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
5 #include <linux/time.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/smp_lock.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <asm/uaccess.h>
15 #include <asm/unaligned.h>
16 #include <linux/buffer_head.h>
17 #include <linux/mpage.h>
18 #include <linux/writeback.h>
19 #include <linux/quotaops.h>
20 #include <linux/swap.h>
22 int reiserfs_commit_write(struct file *f, struct page *page,
23 unsigned from, unsigned to);
24 int reiserfs_prepare_write(struct file *f, struct page *page,
25 unsigned from, unsigned to);
27 void reiserfs_delete_inode(struct inode *inode)
29 /* We need blocks for transaction + (user+group) quota update (possibly delete) */
31 JOURNAL_PER_BALANCE_CNT * 2 +
32 2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
33 struct reiserfs_transaction_handle th;
36 truncate_inode_pages(&inode->i_data, 0);
38 reiserfs_write_lock(inode->i_sb);
40 /* The = 0 happens when we abort creating a new inode for some reason like lack of space.. */
41 if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) { /* also handles bad_inode case */
42 reiserfs_delete_xattrs(inode);
44 if (journal_begin(&th, inode->i_sb, jbegin_count))
46 reiserfs_update_inode_transaction(inode);
48 reiserfs_discard_prealloc(&th, inode);
50 err = reiserfs_delete_object(&th, inode);
52 /* Do quota update inside a transaction for journaled quotas. We must do that
53 * after delete_object so that quota updates go into the same transaction as
54 * stat data deletion */
56 DQUOT_FREE_INODE(inode);
58 if (journal_end(&th, inode->i_sb, jbegin_count))
61 /* check return value from reiserfs_delete_object after
62 * ending the transaction
67 /* all items of file are deleted, so we can remove "save" link */
68 remove_save_link(inode, 0 /* not truncate */ ); /* we can't do anything
69 * about an error here */
71 /* no object items are in the tree */
75 clear_inode(inode); /* note this must go after the journal_end to prevent deadlock */
77 reiserfs_write_unlock(inode->i_sb);
80 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
81 __u32 objectid, loff_t offset, int type, int length)
83 key->version = version;
85 key->on_disk_key.k_dir_id = dirid;
86 key->on_disk_key.k_objectid = objectid;
87 set_cpu_key_k_offset(key, offset);
88 set_cpu_key_k_type(key, type);
89 key->key_length = length;
92 /* take base of inode_key (it comes from inode always) (dirid, objectid) and version from an inode, set
93 offset and type of key */
94 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
97 _make_cpu_key(key, get_inode_item_key_version(inode),
98 le32_to_cpu(INODE_PKEY(inode)->k_dir_id),
99 le32_to_cpu(INODE_PKEY(inode)->k_objectid), offset, type,
104 // when key is 0, do not set version and short key
106 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
108 loff_t offset, int type, int length,
109 int entry_count /*or ih_free_space */ )
112 ih->ih_key.k_dir_id = cpu_to_le32(key->on_disk_key.k_dir_id);
113 ih->ih_key.k_objectid =
114 cpu_to_le32(key->on_disk_key.k_objectid);
116 put_ih_version(ih, version);
117 set_le_ih_k_offset(ih, offset);
118 set_le_ih_k_type(ih, type);
119 put_ih_item_len(ih, length);
120 /* set_ih_free_space (ih, 0); */
121 // for directory items it is entry count, for directs and stat
122 // datas - 0xffff, for indirects - 0
123 put_ih_entry_count(ih, entry_count);
127 // FIXME: we might cache recently accessed indirect item
129 // Ugh. Not too eager for that....
130 // I cut the code until such time as I see a convincing argument (benchmark).
131 // I don't want a bloated inode struct..., and I don't like code complexity....
133 /* cutting the code is fine, since it really isn't in use yet and is easy
134 ** to add back in. But, Vladimir has a really good idea here. Think
135 ** about what happens for reading a file. For each page,
136 ** The VFS layer calls reiserfs_readpage, who searches the tree to find
137 ** an indirect item. This indirect item has X number of pointers, where
138 ** X is a big number if we've done the block allocation right. But,
139 ** we only use one or two of these pointers during each call to readpage,
140 ** needlessly researching again later on.
142 ** The size of the cache could be dynamic based on the size of the file.
144 ** I'd also like to see us cache the location the stat data item, since
145 ** we are needlessly researching for that frequently.
150 /* If this page has a file tail in it, and
151 ** it was read in by get_block_create_0, the page data is valid,
152 ** but tail is still sitting in a direct item, and we can't write to
153 ** it. So, look through this page, and check all the mapped buffers
154 ** to make sure they have valid block numbers. Any that don't need
155 ** to be unmapped, so that block_prepare_write will correctly call
156 ** reiserfs_get_block to convert the tail into an unformatted node
158 static inline void fix_tail_page_for_writing(struct page *page)
160 struct buffer_head *head, *next, *bh;
162 if (page && page_has_buffers(page)) {
163 head = page_buffers(page);
166 next = bh->b_this_page;
167 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
168 reiserfs_unmap_buffer(bh);
171 } while (bh != head);
175 /* reiserfs_get_block does not need to allocate a block only if it has been
176 done already or non-hole position has been found in the indirect item */
177 static inline int allocation_needed(int retval, b_blocknr_t allocated,
178 struct item_head *ih,
179 __le32 * item, int pos_in_item)
183 if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
184 get_block_num(item, pos_in_item))
189 static inline int indirect_item_found(int retval, struct item_head *ih)
191 return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
194 static inline void set_block_dev_mapped(struct buffer_head *bh,
195 b_blocknr_t block, struct inode *inode)
197 map_bh(bh, inode->i_sb, block);
201 // files which were created in the earlier version can not be longer,
204 static int file_capable(struct inode *inode, sector_t block)
206 if (get_inode_item_key_version(inode) != KEY_FORMAT_3_5 || // it is new file.
207 block < (1 << (31 - inode->i_sb->s_blocksize_bits))) // old file, but 'block' is inside of 2gb
213 static int restart_transaction(struct reiserfs_transaction_handle *th,
214 struct inode *inode, struct treepath *path)
216 struct super_block *s = th->t_super;
217 int len = th->t_blocks_allocated;
220 BUG_ON(!th->t_trans_id);
221 BUG_ON(!th->t_refcount);
225 /* we cannot restart while nested */
226 if (th->t_refcount > 1) {
229 reiserfs_update_sd(th, inode);
230 err = journal_end(th, s, len);
232 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
234 reiserfs_update_inode_transaction(inode);
239 // it is called by get_block when create == 0. Returns block number
240 // for 'block'-th logical block of file. When it hits direct item it
241 // returns 0 (being called from bmap) or read direct item into piece
242 // of page (bh_result)
244 // Please improve the english/clarity in the comment above, as it is
245 // hard to understand.
247 static int _get_block_create_0(struct inode *inode, sector_t block,
248 struct buffer_head *bh_result, int args)
250 INITIALIZE_PATH(path);
252 struct buffer_head *bh;
253 struct item_head *ih, tmp_ih;
261 unsigned long offset;
263 // prepare the key to look for the 'block'-th block of file
264 make_cpu_key(&key, inode,
265 (loff_t) block * inode->i_sb->s_blocksize + 1, TYPE_ANY,
269 result = search_for_position_by_key(inode->i_sb, &key, &path);
270 if (result != POSITION_FOUND) {
273 kunmap(bh_result->b_page);
274 if (result == IO_ERROR)
276 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
277 // That there is some MMAPED data associated with it that is yet to be written to disk.
278 if ((args & GET_BLOCK_NO_HOLE)
279 && !PageUptodate(bh_result->b_page)) {
285 bh = get_last_bh(&path);
287 if (is_indirect_le_ih(ih)) {
288 __le32 *ind_item = (__le32 *) B_I_PITEM(bh, ih);
290 /* FIXME: here we could cache indirect item or part of it in
291 the inode to avoid search_by_key in case of subsequent
293 blocknr = get_block_num(ind_item, path.pos_in_item);
296 map_bh(bh_result, inode->i_sb, blocknr);
297 if (path.pos_in_item ==
298 ((ih_item_len(ih) / UNFM_P_SIZE) - 1)) {
299 set_buffer_boundary(bh_result);
302 // We do not return -ENOENT if there is a hole but page is uptodate, because it means
303 // That there is some MMAPED data associated with it that is yet to be written to disk.
304 if ((args & GET_BLOCK_NO_HOLE)
305 && !PageUptodate(bh_result->b_page)) {
311 kunmap(bh_result->b_page);
314 // requested data are in direct item(s)
315 if (!(args & GET_BLOCK_READ_DIRECT)) {
316 // we are called by bmap. FIXME: we can not map block of file
317 // when it is stored in direct item(s)
320 kunmap(bh_result->b_page);
324 /* if we've got a direct item, and the buffer or page was uptodate,
325 ** we don't want to pull data off disk again. skip to the
326 ** end, where we map the buffer and return
328 if (buffer_uptodate(bh_result)) {
332 ** grab_tail_page can trigger calls to reiserfs_get_block on up to date
333 ** pages without any buffers. If the page is up to date, we don't want
334 ** read old data off disk. Set the up to date bit on the buffer instead
335 ** and jump to the end
337 if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
338 set_buffer_uptodate(bh_result);
341 // read file tail into part of page
342 offset = (cpu_key_k_offset(&key) - 1) & (PAGE_CACHE_SIZE - 1);
343 fs_gen = get_generation(inode->i_sb);
344 copy_item_head(&tmp_ih, ih);
346 /* we only want to kmap if we are reading the tail into the page.
347 ** this is not the common case, so we don't kmap until we are
348 ** sure we need to. But, this means the item might move if
352 p = (char *)kmap(bh_result->b_page);
353 if (fs_changed(fs_gen, inode->i_sb)
354 && item_moved(&tmp_ih, &path)) {
359 memset(p, 0, inode->i_sb->s_blocksize);
361 if (!is_direct_le_ih(ih)) {
364 /* make sure we don't read more bytes than actually exist in
365 ** the file. This can happen in odd cases where i_size isn't
366 ** correct, and when direct item padding results in a few
367 ** extra bytes at the end of the direct item
369 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
371 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
373 inode->i_size - (le_ih_k_offset(ih) - 1) -
377 chars = ih_item_len(ih) - path.pos_in_item;
379 memcpy(p, B_I_PITEM(bh, ih) + path.pos_in_item, chars);
386 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
387 // we done, if read direct item is not the last item of
388 // node FIXME: we could try to check right delimiting key
389 // to see whether direct item continues in the right
390 // neighbor or rely on i_size
393 // update key to look for the next piece
394 set_cpu_key_k_offset(&key, cpu_key_k_offset(&key) + chars);
395 result = search_for_position_by_key(inode->i_sb, &key, &path);
396 if (result != POSITION_FOUND)
397 // i/o error most likely
399 bh = get_last_bh(&path);
403 flush_dcache_page(bh_result->b_page);
404 kunmap(bh_result->b_page);
409 if (result == IO_ERROR)
412 /* this buffer has valid data, but isn't valid for io. mapping it to
413 * block #0 tells the rest of reiserfs it just has a tail in it
415 map_bh(bh_result, inode->i_sb, 0);
416 set_buffer_uptodate(bh_result);
420 // this is called to create file map. So, _get_block_create_0 will not
422 static int reiserfs_bmap(struct inode *inode, sector_t block,
423 struct buffer_head *bh_result, int create)
425 if (!file_capable(inode, block))
428 reiserfs_write_lock(inode->i_sb);
429 /* do not read the direct item */
430 _get_block_create_0(inode, block, bh_result, 0);
431 reiserfs_write_unlock(inode->i_sb);
435 /* special version of get_block that is only used by grab_tail_page right
436 ** now. It is sent to block_prepare_write, and when you try to get a
437 ** block past the end of the file (or a block from a hole) it returns
438 ** -ENOENT instead of a valid buffer. block_prepare_write expects to
439 ** be able to do i/o on the buffers returned, unless an error value
442 ** So, this allows block_prepare_write to be used for reading a single block
443 ** in a page. Where it does not produce a valid page for holes, or past the
444 ** end of the file. This turns out to be exactly what we need for reading
445 ** tails for conversion.
447 ** The point of the wrapper is forcing a certain value for create, even
448 ** though the VFS layer is calling this function with create==1. If you
449 ** don't want to send create == GET_BLOCK_NO_HOLE to reiserfs_get_block,
450 ** don't use this function.
452 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
453 struct buffer_head *bh_result,
456 return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
459 /* This is special helper for reiserfs_get_block in case we are executing
460 direct_IO request. */
461 static int reiserfs_get_blocks_direct_io(struct inode *inode,
463 struct buffer_head *bh_result,
468 bh_result->b_page = NULL;
470 /* We set the b_size before reiserfs_get_block call since it is
471 referenced in convert_tail_for_hole() that may be called from
472 reiserfs_get_block() */
473 bh_result->b_size = (1 << inode->i_blkbits);
475 ret = reiserfs_get_block(inode, iblock, bh_result,
476 create | GET_BLOCK_NO_DANGLE);
480 /* don't allow direct io onto tail pages */
481 if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
482 /* make sure future calls to the direct io funcs for this offset
483 ** in the file fail by unmapping the buffer
485 clear_buffer_mapped(bh_result);
488 /* Possible unpacked tail. Flush the data before pages have
490 if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
493 err = reiserfs_commit_for_inode(inode);
494 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
504 ** helper function for when reiserfs_get_block is called for a hole
505 ** but the file tail is still in a direct item
506 ** bh_result is the buffer head for the hole
507 ** tail_offset is the offset of the start of the tail in the file
509 ** This calls prepare_write, which will start a new transaction
510 ** you should not be in a transaction, or have any paths held when you
513 static int convert_tail_for_hole(struct inode *inode,
514 struct buffer_head *bh_result,
518 unsigned long tail_end;
519 unsigned long tail_start;
520 struct page *tail_page;
521 struct page *hole_page = bh_result->b_page;
524 if ((tail_offset & (bh_result->b_size - 1)) != 1)
527 /* always try to read until the end of the block */
528 tail_start = tail_offset & (PAGE_CACHE_SIZE - 1);
529 tail_end = (tail_start | (bh_result->b_size - 1)) + 1;
531 index = tail_offset >> PAGE_CACHE_SHIFT;
532 /* hole_page can be zero in case of direct_io, we are sure
533 that we cannot get here if we write with O_DIRECT into
535 if (!hole_page || index != hole_page->index) {
536 tail_page = grab_cache_page(inode->i_mapping, index);
542 tail_page = hole_page;
545 /* we don't have to make sure the conversion did not happen while
546 ** we were locking the page because anyone that could convert
547 ** must first take i_mutex.
549 ** We must fix the tail page for writing because it might have buffers
550 ** that are mapped, but have a block number of 0. This indicates tail
551 ** data that has been read directly into the page, and block_prepare_write
552 ** won't trigger a get_block in this case.
554 fix_tail_page_for_writing(tail_page);
555 retval = reiserfs_prepare_write(NULL, tail_page, tail_start, tail_end);
559 /* tail conversion might change the data in the page */
560 flush_dcache_page(tail_page);
562 retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
565 if (tail_page != hole_page) {
566 unlock_page(tail_page);
567 page_cache_release(tail_page);
573 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
576 b_blocknr_t * allocated_block_nr,
577 struct treepath *path, int flags)
579 BUG_ON(!th->t_trans_id);
581 #ifdef REISERFS_PREALLOCATE
582 if (!(flags & GET_BLOCK_NO_IMUX)) {
583 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
587 return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
591 int reiserfs_get_block(struct inode *inode, sector_t block,
592 struct buffer_head *bh_result, int create)
594 int repeat, retval = 0;
595 b_blocknr_t allocated_block_nr = 0; // b_blocknr_t is (unsigned) 32 bit int
596 INITIALIZE_PATH(path);
599 struct buffer_head *bh, *unbh = NULL;
600 struct item_head *ih, tmp_ih;
604 struct reiserfs_transaction_handle *th = NULL;
605 /* space reserved in transaction batch:
606 . 3 balancings in direct->indirect conversion
607 . 1 block involved into reiserfs_update_sd()
608 XXX in practically impossible worst case direct2indirect()
609 can incur (much) more than 3 balancings.
610 quota update for user, group */
612 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
613 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
617 (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
620 reiserfs_write_lock(inode->i_sb);
621 version = get_inode_item_key_version(inode);
623 if (!file_capable(inode, block)) {
624 reiserfs_write_unlock(inode->i_sb);
628 /* if !create, we aren't changing the FS, so we don't need to
629 ** log anything, so we don't need to start a transaction
631 if (!(create & GET_BLOCK_CREATE)) {
633 /* find number of block-th logical block of the file */
634 ret = _get_block_create_0(inode, block, bh_result,
635 create | GET_BLOCK_READ_DIRECT);
636 reiserfs_write_unlock(inode->i_sb);
640 * if we're already in a transaction, make sure to close
641 * any new transactions we start in this func
643 if ((create & GET_BLOCK_NO_DANGLE) ||
644 reiserfs_transaction_running(inode->i_sb))
647 /* If file is of such a size, that it might have a tail and tails are enabled
648 ** we should mark it as possibly needing tail packing on close
650 if ((have_large_tails(inode->i_sb)
651 && inode->i_size < i_block_size(inode) * 4)
652 || (have_small_tails(inode->i_sb)
653 && inode->i_size < i_block_size(inode)))
654 REISERFS_I(inode)->i_flags |= i_pack_on_close_mask;
656 /* set the key of the first byte in the 'block'-th block of file */
657 make_cpu_key(&key, inode, new_offset, TYPE_ANY, 3 /*key length */ );
658 if ((new_offset + inode->i_sb->s_blocksize - 1) > inode->i_size) {
660 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
665 reiserfs_update_inode_transaction(inode);
669 retval = search_for_position_by_key(inode->i_sb, &key, &path);
670 if (retval == IO_ERROR) {
675 bh = get_last_bh(&path);
677 item = get_item(&path);
678 pos_in_item = path.pos_in_item;
680 fs_gen = get_generation(inode->i_sb);
681 copy_item_head(&tmp_ih, ih);
683 if (allocation_needed
684 (retval, allocated_block_nr, ih, item, pos_in_item)) {
685 /* we have to allocate block for the unformatted node */
692 _allocate_block(th, block, inode, &allocated_block_nr,
695 if (repeat == NO_DISK_SPACE || repeat == QUOTA_EXCEEDED) {
696 /* restart the transaction to give the journal a chance to free
697 ** some blocks. releases the path, so we have to go back to
698 ** research if we succeed on the second try
700 SB_JOURNAL(inode->i_sb)->j_next_async_flush = 1;
701 retval = restart_transaction(th, inode, &path);
705 _allocate_block(th, block, inode,
706 &allocated_block_nr, NULL, create);
708 if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
711 if (repeat == QUOTA_EXCEEDED)
718 if (fs_changed(fs_gen, inode->i_sb)
719 && item_moved(&tmp_ih, &path)) {
724 if (indirect_item_found(retval, ih)) {
725 b_blocknr_t unfm_ptr;
726 /* 'block'-th block is in the file already (there is
727 corresponding cell in some indirect item). But it may be
728 zero unformatted node pointer (hole) */
729 unfm_ptr = get_block_num(item, pos_in_item);
731 /* use allocated block to plug the hole */
732 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
733 if (fs_changed(fs_gen, inode->i_sb)
734 && item_moved(&tmp_ih, &path)) {
735 reiserfs_restore_prepared_buffer(inode->i_sb,
739 set_buffer_new(bh_result);
740 if (buffer_dirty(bh_result)
741 && reiserfs_data_ordered(inode->i_sb))
742 reiserfs_add_ordered_list(inode, bh_result);
743 put_block_num(item, pos_in_item, allocated_block_nr);
744 unfm_ptr = allocated_block_nr;
745 journal_mark_dirty(th, inode->i_sb, bh);
746 reiserfs_update_sd(th, inode);
748 set_block_dev_mapped(bh_result, unfm_ptr, inode);
752 retval = reiserfs_end_persistent_transaction(th);
754 reiserfs_write_unlock(inode->i_sb);
756 /* the item was found, so new blocks were not added to the file
757 ** there is no need to make sure the inode is updated with this
768 /* desired position is not found or is in the direct item. We have
769 to append file with holes up to 'block'-th block converting
770 direct items to indirect one if necessary */
773 if (is_statdata_le_ih(ih)) {
775 struct cpu_key tmp_key;
777 /* indirect item has to be inserted */
778 make_le_item_head(&tmp_ih, &key, version, 1,
779 TYPE_INDIRECT, UNFM_P_SIZE,
780 0 /* free_space */ );
782 if (cpu_key_k_offset(&key) == 1) {
783 /* we are going to add 'block'-th block to the file. Use
784 allocated block for that */
785 unp = cpu_to_le32(allocated_block_nr);
786 set_block_dev_mapped(bh_result,
787 allocated_block_nr, inode);
788 set_buffer_new(bh_result);
792 set_cpu_key_k_offset(&tmp_key, 1);
793 PATH_LAST_POSITION(&path)++;
796 reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
797 inode, (char *)&unp);
799 reiserfs_free_block(th, inode,
800 allocated_block_nr, 1);
801 goto failure; // retval == -ENOSPC, -EDQUOT or -EIO or -EEXIST
803 //mark_tail_converted (inode);
804 } else if (is_direct_le_ih(ih)) {
805 /* direct item has to be converted */
809 ((le_ih_k_offset(ih) -
810 1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
811 if (tail_offset == cpu_key_k_offset(&key)) {
812 /* direct item we just found fits into block we have
813 to map. Convert it into unformatted node: use
814 bh_result for the conversion */
815 set_block_dev_mapped(bh_result,
816 allocated_block_nr, inode);
820 /* we have to padd file tail stored in direct item(s)
821 up to block size and convert it to unformatted
822 node. FIXME: this should also get into page cache */
826 * ugly, but we can only end the transaction if
829 BUG_ON(!th->t_refcount);
830 if (th->t_refcount == 1) {
832 reiserfs_end_persistent_transaction
840 convert_tail_for_hole(inode, bh_result,
843 if (retval != -ENOSPC)
844 reiserfs_error(inode->i_sb,
846 "convert tail failed "
847 "inode %lu, error %d",
850 if (allocated_block_nr) {
851 /* the bitmap, the super, and the stat data == 3 */
853 th = reiserfs_persistent_transaction(inode->i_sb, 3);
855 reiserfs_free_block(th,
865 direct2indirect(th, inode, &path, unbh,
868 reiserfs_unmap_buffer(unbh);
869 reiserfs_free_block(th, inode,
870 allocated_block_nr, 1);
873 /* it is important the set_buffer_uptodate is done after
874 ** the direct2indirect. The buffer might contain valid
875 ** data newer than the data on disk (read by readpage, changed,
876 ** and then sent here by writepage). direct2indirect needs
877 ** to know if unbh was already up to date, so it can decide
878 ** if the data in unbh needs to be replaced with data from
881 set_buffer_uptodate(unbh);
883 /* unbh->b_page == NULL in case of DIRECT_IO request, this means
884 buffer will disappear shortly, so it should not be added to
887 /* we've converted the tail, so we must
888 ** flush unbh before the transaction commits
890 reiserfs_add_tail_list(inode, unbh);
892 /* mark it dirty now to prevent commit_write from adding
893 ** this buffer to the inode's dirty buffer list
896 * AKPM: changed __mark_buffer_dirty to mark_buffer_dirty().
897 * It's still atomic, but it sets the page dirty too,
898 * which makes it eligible for writeback at any time by the
899 * VM (which was also the case with __mark_buffer_dirty())
901 mark_buffer_dirty(unbh);
904 /* append indirect item with holes if needed, when appending
905 pointer to 'block'-th block use block, which is already
907 struct cpu_key tmp_key;
908 unp_t unf_single = 0; // We use this in case we need to allocate only
909 // one block which is a fastpath
911 __u64 max_to_insert =
912 MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
916 RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
917 "vs-804: invalid position for append");
918 /* indirect item has to be appended, set up key of that position */
919 make_cpu_key(&tmp_key, inode,
920 le_key_k_offset(version,
923 inode->i_sb->s_blocksize),
924 //pos_in_item * inode->i_sb->s_blocksize,
925 TYPE_INDIRECT, 3); // key type is unimportant
927 RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
928 "green-805: invalid offset");
931 ((cpu_key_k_offset(&key) -
932 cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
935 if (blocks_needed == 1) {
938 un = kzalloc(min(blocks_needed, max_to_insert) * UNFM_P_SIZE, GFP_ATOMIC); // We need to avoid scheduling.
945 if (blocks_needed <= max_to_insert) {
946 /* we are going to add target block to the file. Use allocated
948 un[blocks_needed - 1] =
949 cpu_to_le32(allocated_block_nr);
950 set_block_dev_mapped(bh_result,
951 allocated_block_nr, inode);
952 set_buffer_new(bh_result);
955 /* paste hole to the indirect item */
956 /* If kmalloc failed, max_to_insert becomes zero and it means we
957 only have space for one block */
959 max_to_insert ? max_to_insert : 1;
962 reiserfs_paste_into_item(th, &path, &tmp_key, inode,
967 if (blocks_needed != 1)
971 reiserfs_free_block(th, inode,
972 allocated_block_nr, 1);
976 /* We need to mark new file size in case this function will be
977 interrupted/aborted later on. And we may do this only for
980 inode->i_sb->s_blocksize * blocks_needed;
987 /* this loop could log more blocks than we had originally asked
988 ** for. So, we have to allow the transaction to end if it is
989 ** too big or too full. Update the inode so things are
990 ** consistent if we crash before the function returns
992 ** release the path so that anybody waiting on the path before
993 ** ending their transaction will be able to continue.
995 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
996 retval = restart_transaction(th, inode, &path);
1000 /* inserting indirect pointers for a hole can take a
1001 ** long time. reschedule if needed
1005 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1006 if (retval == IO_ERROR) {
1010 if (retval == POSITION_FOUND) {
1011 reiserfs_warning(inode->i_sb, "vs-825",
1012 "%K should not be found", &key);
1014 if (allocated_block_nr)
1015 reiserfs_free_block(th, inode,
1016 allocated_block_nr, 1);
1020 bh = get_last_bh(&path);
1022 item = get_item(&path);
1023 pos_in_item = path.pos_in_item;
1029 if (th && (!dangle || (retval && !th->t_trans_id))) {
1032 reiserfs_update_sd(th, inode);
1033 err = reiserfs_end_persistent_transaction(th);
1038 reiserfs_write_unlock(inode->i_sb);
1039 reiserfs_check_path(&path);
1044 reiserfs_readpages(struct file *file, struct address_space *mapping,
1045 struct list_head *pages, unsigned nr_pages)
1047 return mpage_readpages(mapping, pages, nr_pages, reiserfs_get_block);
1050 /* Compute real number of used bytes by file
1051 * Following three functions can go away when we'll have enough space in stat item
1053 static int real_space_diff(struct inode *inode, int sd_size)
1056 loff_t blocksize = inode->i_sb->s_blocksize;
1058 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1061 /* End of file is also in full block with indirect reference, so round
1062 ** up to the next block.
1064 ** there is just no way to know if the tail is actually packed
1065 ** on the file, so we have to assume it isn't. When we pack the
1066 ** tail, we add 4 bytes to pretend there really is an unformatted
1071 (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1076 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
1079 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1080 return inode->i_size +
1081 (loff_t) (real_space_diff(inode, sd_size));
1083 return ((loff_t) real_space_diff(inode, sd_size)) +
1084 (((loff_t) blocks) << 9);
1087 /* Compute number of blocks used by file in ReiserFS counting */
1088 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1090 loff_t bytes = inode_get_bytes(inode);
1091 loff_t real_space = real_space_diff(inode, sd_size);
1093 /* keeps fsck and non-quota versions of reiserfs happy */
1094 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode)) {
1095 bytes += (loff_t) 511;
1098 /* files from before the quota patch might i_blocks such that
1099 ** bytes < real_space. Deal with that here to prevent it from
1102 if (bytes < real_space)
1104 return (bytes - real_space) >> 9;
1108 // BAD: new directories have stat data of new type and all other items
1109 // of old type. Version stored in the inode says about body items, so
1110 // in update_stat_data we can not rely on inode, but have to check
1111 // item version directly
1114 // called by read_locked_inode
1115 static void init_inode(struct inode *inode, struct treepath *path)
1117 struct buffer_head *bh;
1118 struct item_head *ih;
1120 //int version = ITEM_VERSION_1;
1122 bh = PATH_PLAST_BUFFER(path);
1123 ih = PATH_PITEM_HEAD(path);
1125 copy_key(INODE_PKEY(inode), &(ih->ih_key));
1127 INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1128 REISERFS_I(inode)->i_flags = 0;
1129 REISERFS_I(inode)->i_prealloc_block = 0;
1130 REISERFS_I(inode)->i_prealloc_count = 0;
1131 REISERFS_I(inode)->i_trans_id = 0;
1132 REISERFS_I(inode)->i_jl = NULL;
1133 mutex_init(&(REISERFS_I(inode)->i_mmap));
1134 reiserfs_init_acl_access(inode);
1135 reiserfs_init_acl_default(inode);
1136 reiserfs_init_xattr_rwsem(inode);
1138 if (stat_data_v1(ih)) {
1139 struct stat_data_v1 *sd =
1140 (struct stat_data_v1 *)B_I_PITEM(bh, ih);
1141 unsigned long blocks;
1143 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1144 set_inode_sd_version(inode, STAT_DATA_V1);
1145 inode->i_mode = sd_v1_mode(sd);
1146 inode->i_nlink = sd_v1_nlink(sd);
1147 inode->i_uid = sd_v1_uid(sd);
1148 inode->i_gid = sd_v1_gid(sd);
1149 inode->i_size = sd_v1_size(sd);
1150 inode->i_atime.tv_sec = sd_v1_atime(sd);
1151 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1152 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1153 inode->i_atime.tv_nsec = 0;
1154 inode->i_ctime.tv_nsec = 0;
1155 inode->i_mtime.tv_nsec = 0;
1157 inode->i_blocks = sd_v1_blocks(sd);
1158 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1159 blocks = (inode->i_size + 511) >> 9;
1160 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1161 if (inode->i_blocks > blocks) {
1162 // there was a bug in <=3.5.23 when i_blocks could take negative
1163 // values. Starting from 3.5.17 this value could even be stored in
1164 // stat data. For such files we set i_blocks based on file
1165 // size. Just 2 notes: this can be wrong for sparce files. On-disk value will be
1166 // only updated if file's inode will ever change
1167 inode->i_blocks = blocks;
1170 rdev = sd_v1_rdev(sd);
1171 REISERFS_I(inode)->i_first_direct_byte =
1172 sd_v1_first_direct_byte(sd);
1173 /* an early bug in the quota code can give us an odd number for the
1174 ** block count. This is incorrect, fix it here.
1176 if (inode->i_blocks & 1) {
1179 inode_set_bytes(inode,
1180 to_real_used_space(inode, inode->i_blocks,
1182 /* nopack is initially zero for v1 objects. For v2 objects,
1183 nopack is initialised from sd_attrs */
1184 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1186 // new stat data found, but object may have old items
1187 // (directories and symlinks)
1188 struct stat_data *sd = (struct stat_data *)B_I_PITEM(bh, ih);
1190 inode->i_mode = sd_v2_mode(sd);
1191 inode->i_nlink = sd_v2_nlink(sd);
1192 inode->i_uid = sd_v2_uid(sd);
1193 inode->i_size = sd_v2_size(sd);
1194 inode->i_gid = sd_v2_gid(sd);
1195 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1196 inode->i_atime.tv_sec = sd_v2_atime(sd);
1197 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1198 inode->i_ctime.tv_nsec = 0;
1199 inode->i_mtime.tv_nsec = 0;
1200 inode->i_atime.tv_nsec = 0;
1201 inode->i_blocks = sd_v2_blocks(sd);
1202 rdev = sd_v2_rdev(sd);
1203 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1204 inode->i_generation =
1205 le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1207 inode->i_generation = sd_v2_generation(sd);
1209 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1210 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1212 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1213 REISERFS_I(inode)->i_first_direct_byte = 0;
1214 set_inode_sd_version(inode, STAT_DATA_V2);
1215 inode_set_bytes(inode,
1216 to_real_used_space(inode, inode->i_blocks,
1218 /* read persistent inode attributes from sd and initalise
1219 generic inode flags from them */
1220 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1221 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1225 if (S_ISREG(inode->i_mode)) {
1226 inode->i_op = &reiserfs_file_inode_operations;
1227 inode->i_fop = &reiserfs_file_operations;
1228 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1229 } else if (S_ISDIR(inode->i_mode)) {
1230 inode->i_op = &reiserfs_dir_inode_operations;
1231 inode->i_fop = &reiserfs_dir_operations;
1232 } else if (S_ISLNK(inode->i_mode)) {
1233 inode->i_op = &reiserfs_symlink_inode_operations;
1234 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1236 inode->i_blocks = 0;
1237 inode->i_op = &reiserfs_special_inode_operations;
1238 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1242 // update new stat data with inode fields
1243 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1245 struct stat_data *sd_v2 = (struct stat_data *)sd;
1248 set_sd_v2_mode(sd_v2, inode->i_mode);
1249 set_sd_v2_nlink(sd_v2, inode->i_nlink);
1250 set_sd_v2_uid(sd_v2, inode->i_uid);
1251 set_sd_v2_size(sd_v2, size);
1252 set_sd_v2_gid(sd_v2, inode->i_gid);
1253 set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1254 set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1255 set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1256 set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1257 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1258 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1260 set_sd_v2_generation(sd_v2, inode->i_generation);
1261 flags = REISERFS_I(inode)->i_attrs;
1262 i_attrs_to_sd_attrs(inode, &flags);
1263 set_sd_v2_attrs(sd_v2, flags);
1266 // used to copy inode's fields to old stat data
1267 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1269 struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1271 set_sd_v1_mode(sd_v1, inode->i_mode);
1272 set_sd_v1_uid(sd_v1, inode->i_uid);
1273 set_sd_v1_gid(sd_v1, inode->i_gid);
1274 set_sd_v1_nlink(sd_v1, inode->i_nlink);
1275 set_sd_v1_size(sd_v1, size);
1276 set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1277 set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1278 set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1280 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1281 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1283 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1285 // Sigh. i_first_direct_byte is back
1286 set_sd_v1_first_direct_byte(sd_v1,
1287 REISERFS_I(inode)->i_first_direct_byte);
1290 /* NOTE, you must prepare the buffer head before sending it here,
1291 ** and then log it after the call
1293 static void update_stat_data(struct treepath *path, struct inode *inode,
1296 struct buffer_head *bh;
1297 struct item_head *ih;
1299 bh = PATH_PLAST_BUFFER(path);
1300 ih = PATH_PITEM_HEAD(path);
1302 if (!is_statdata_le_ih(ih))
1303 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1304 INODE_PKEY(inode), ih);
1306 if (stat_data_v1(ih)) {
1307 // path points to old stat data
1308 inode2sd_v1(B_I_PITEM(bh, ih), inode, size);
1310 inode2sd(B_I_PITEM(bh, ih), inode, size);
1316 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1317 struct inode *inode, loff_t size)
1320 INITIALIZE_PATH(path);
1321 struct buffer_head *bh;
1323 struct item_head *ih, tmp_ih;
1326 BUG_ON(!th->t_trans_id);
1328 make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3); //key type is unimportant
1332 /* look for the object's stat data */
1333 retval = search_item(inode->i_sb, &key, &path);
1334 if (retval == IO_ERROR) {
1335 reiserfs_error(inode->i_sb, "vs-13050",
1336 "i/o failure occurred trying to "
1337 "update %K stat data", &key);
1340 if (retval == ITEM_NOT_FOUND) {
1341 pos = PATH_LAST_POSITION(&path);
1343 if (inode->i_nlink == 0) {
1344 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1347 reiserfs_warning(inode->i_sb, "vs-13060",
1348 "stat data of object %k (nlink == %d) "
1349 "not found (pos %d)",
1350 INODE_PKEY(inode), inode->i_nlink,
1352 reiserfs_check_path(&path);
1356 /* sigh, prepare_for_journal might schedule. When it schedules the
1357 ** FS might change. We have to detect that, and loop back to the
1358 ** search if the stat data item has moved
1360 bh = get_last_bh(&path);
1362 copy_item_head(&tmp_ih, ih);
1363 fs_gen = get_generation(inode->i_sb);
1364 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1365 if (fs_changed(fs_gen, inode->i_sb)
1366 && item_moved(&tmp_ih, &path)) {
1367 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1368 continue; /* Stat_data item has been moved after scheduling. */
1372 update_stat_data(&path, inode, size);
1373 journal_mark_dirty(th, th->t_super, bh);
1378 /* reiserfs_read_locked_inode is called to read the inode off disk, and it
1379 ** does a make_bad_inode when things go wrong. But, we need to make sure
1380 ** and clear the key in the private portion of the inode, otherwise a
1381 ** corresponding iput might try to delete whatever object the inode last
1384 static void reiserfs_make_bad_inode(struct inode *inode)
1386 memset(INODE_PKEY(inode), 0, KEY_SIZE);
1387 make_bad_inode(inode);
1391 // initially this function was derived from minix or ext2's analog and
1392 // evolved as the prototype did
1395 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1397 struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1398 inode->i_ino = args->objectid;
1399 INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1403 /* looks for stat data in the tree, and fills up the fields of in-core
1404 inode stat data fields */
1405 void reiserfs_read_locked_inode(struct inode *inode,
1406 struct reiserfs_iget_args *args)
1408 INITIALIZE_PATH(path_to_sd);
1410 unsigned long dirino;
1413 dirino = args->dirid;
1415 /* set version 1, version 2 could be used too, because stat data
1416 key is the same in both versions */
1417 key.version = KEY_FORMAT_3_5;
1418 key.on_disk_key.k_dir_id = dirino;
1419 key.on_disk_key.k_objectid = inode->i_ino;
1420 key.on_disk_key.k_offset = 0;
1421 key.on_disk_key.k_type = 0;
1423 /* look for the object's stat data */
1424 retval = search_item(inode->i_sb, &key, &path_to_sd);
1425 if (retval == IO_ERROR) {
1426 reiserfs_error(inode->i_sb, "vs-13070",
1427 "i/o failure occurred trying to find "
1428 "stat data of %K", &key);
1429 reiserfs_make_bad_inode(inode);
1432 if (retval != ITEM_FOUND) {
1433 /* a stale NFS handle can trigger this without it being an error */
1434 pathrelse(&path_to_sd);
1435 reiserfs_make_bad_inode(inode);
1440 init_inode(inode, &path_to_sd);
1442 /* It is possible that knfsd is trying to access inode of a file
1443 that is being removed from the disk by some other thread. As we
1444 update sd on unlink all that is required is to check for nlink
1445 here. This bug was first found by Sizif when debugging
1446 SquidNG/Butterfly, forgotten, and found again after Philippe
1447 Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1449 More logical fix would require changes in fs/inode.c:iput() to
1450 remove inode from hash-table _after_ fs cleaned disk stuff up and
1451 in iget() to return NULL if I_FREEING inode is found in
1453 /* Currently there is one place where it's ok to meet inode with
1454 nlink==0: processing of open-unlinked and half-truncated files
1455 during mount (fs/reiserfs/super.c:finish_unfinished()). */
1456 if ((inode->i_nlink == 0) &&
1457 !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1458 reiserfs_warning(inode->i_sb, "vs-13075",
1459 "dead inode read from disk %K. "
1460 "This is likely to be race with knfsd. Ignore",
1462 reiserfs_make_bad_inode(inode);
1465 reiserfs_check_path(&path_to_sd); /* init inode should be relsing */
1470 * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1472 * @inode: inode from hash table to check
1473 * @opaque: "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1475 * This function is called by iget5_locked() to distinguish reiserfs inodes
1476 * having the same inode numbers. Such inodes can only exist due to some
1477 * error condition. One of them should be bad. Inodes with identical
1478 * inode numbers (objectids) are distinguished by parent directory ids.
1481 int reiserfs_find_actor(struct inode *inode, void *opaque)
1483 struct reiserfs_iget_args *args;
1486 /* args is already in CPU order */
1487 return (inode->i_ino == args->objectid) &&
1488 (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1491 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1493 struct inode *inode;
1494 struct reiserfs_iget_args args;
1496 args.objectid = key->on_disk_key.k_objectid;
1497 args.dirid = key->on_disk_key.k_dir_id;
1498 inode = iget5_locked(s, key->on_disk_key.k_objectid,
1499 reiserfs_find_actor, reiserfs_init_locked_inode,
1502 return ERR_PTR(-ENOMEM);
1504 if (inode->i_state & I_NEW) {
1505 reiserfs_read_locked_inode(inode, &args);
1506 unlock_new_inode(inode);
1509 if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1510 /* either due to i/o error or a stale NFS handle */
1517 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1518 u32 objectid, u32 dir_id, u32 generation)
1522 struct inode *inode;
1524 key.on_disk_key.k_objectid = objectid;
1525 key.on_disk_key.k_dir_id = dir_id;
1526 reiserfs_write_lock(sb);
1527 inode = reiserfs_iget(sb, &key);
1528 if (inode && !IS_ERR(inode) && generation != 0 &&
1529 generation != inode->i_generation) {
1533 reiserfs_write_unlock(sb);
1535 return d_obtain_alias(inode);
1538 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1539 int fh_len, int fh_type)
1541 /* fhtype happens to reflect the number of u32s encoded.
1542 * due to a bug in earlier code, fhtype might indicate there
1543 * are more u32s then actually fitted.
1544 * so if fhtype seems to be more than len, reduce fhtype.
1546 * 2 - objectid + dir_id - legacy support
1547 * 3 - objectid + dir_id + generation
1548 * 4 - objectid + dir_id + objectid and dirid of parent - legacy
1549 * 5 - objectid + dir_id + generation + objectid and dirid of parent
1550 * 6 - as above plus generation of directory
1551 * 6 does not fit in NFSv2 handles
1553 if (fh_type > fh_len) {
1554 if (fh_type != 6 || fh_len != 5)
1555 reiserfs_warning(sb, "reiserfs-13077",
1556 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1561 return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1562 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1565 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1566 int fh_len, int fh_type)
1571 return reiserfs_get_dentry(sb,
1572 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1573 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1574 (fh_type == 6) ? fid->raw[5] : 0);
1577 int reiserfs_encode_fh(struct dentry *dentry, __u32 * data, int *lenp,
1580 struct inode *inode = dentry->d_inode;
1586 data[0] = inode->i_ino;
1587 data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1588 data[2] = inode->i_generation;
1590 /* no room for directory info? return what we've stored so far */
1591 if (maxlen < 5 || !need_parent)
1594 spin_lock(&dentry->d_lock);
1595 inode = dentry->d_parent->d_inode;
1596 data[3] = inode->i_ino;
1597 data[4] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1600 data[5] = inode->i_generation;
1603 spin_unlock(&dentry->d_lock);
1607 /* looks for stat data, then copies fields to it, marks the buffer
1608 containing stat data as dirty */
1609 /* reiserfs inodes are never really dirty, since the dirty inode call
1610 ** always logs them. This call allows the VFS inode marking routines
1611 ** to properly mark inodes for datasync and such, but only actually
1612 ** does something when called for a synchronous update.
1614 int reiserfs_write_inode(struct inode *inode, int do_sync)
1616 struct reiserfs_transaction_handle th;
1617 int jbegin_count = 1;
1619 if (inode->i_sb->s_flags & MS_RDONLY)
1621 /* memory pressure can sometimes initiate write_inode calls with sync == 1,
1622 ** these cases are just when the system needs ram, not when the
1623 ** inode needs to reach disk for safety, and they can safely be
1624 ** ignored because the altered inode has already been logged.
1626 if (do_sync && !(current->flags & PF_MEMALLOC)) {
1627 reiserfs_write_lock(inode->i_sb);
1628 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1629 reiserfs_update_sd(&th, inode);
1630 journal_end_sync(&th, inode->i_sb, jbegin_count);
1632 reiserfs_write_unlock(inode->i_sb);
1637 /* stat data of new object is inserted already, this inserts the item
1638 containing "." and ".." entries */
1639 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1640 struct inode *inode,
1641 struct item_head *ih, struct treepath *path,
1644 struct super_block *sb = th->t_super;
1645 char empty_dir[EMPTY_DIR_SIZE];
1646 char *body = empty_dir;
1650 BUG_ON(!th->t_trans_id);
1652 _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1653 le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1654 TYPE_DIRENTRY, 3 /*key length */ );
1656 /* compose item head for new item. Directories consist of items of
1657 old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1658 is done by reiserfs_new_inode */
1659 if (old_format_only(sb)) {
1660 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1661 TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1663 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1664 ih->ih_key.k_objectid,
1665 INODE_PKEY(dir)->k_dir_id,
1666 INODE_PKEY(dir)->k_objectid);
1668 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1669 TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1671 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1672 ih->ih_key.k_objectid,
1673 INODE_PKEY(dir)->k_dir_id,
1674 INODE_PKEY(dir)->k_objectid);
1677 /* look for place in the tree for new item */
1678 retval = search_item(sb, &key, path);
1679 if (retval == IO_ERROR) {
1680 reiserfs_error(sb, "vs-13080",
1681 "i/o failure occurred creating new directory");
1684 if (retval == ITEM_FOUND) {
1686 reiserfs_warning(sb, "vs-13070",
1687 "object with this key exists (%k)",
1692 /* insert item, that is empty directory item */
1693 return reiserfs_insert_item(th, path, &key, ih, inode, body);
1696 /* stat data of object has been inserted, this inserts the item
1697 containing the body of symlink */
1698 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th, struct inode *inode, /* Inode of symlink */
1699 struct item_head *ih,
1700 struct treepath *path, const char *symname,
1703 struct super_block *sb = th->t_super;
1707 BUG_ON(!th->t_trans_id);
1709 _make_cpu_key(&key, KEY_FORMAT_3_5,
1710 le32_to_cpu(ih->ih_key.k_dir_id),
1711 le32_to_cpu(ih->ih_key.k_objectid),
1712 1, TYPE_DIRECT, 3 /*key length */ );
1714 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1715 0 /*free_space */ );
1717 /* look for place in the tree for new item */
1718 retval = search_item(sb, &key, path);
1719 if (retval == IO_ERROR) {
1720 reiserfs_error(sb, "vs-13080",
1721 "i/o failure occurred creating new symlink");
1724 if (retval == ITEM_FOUND) {
1726 reiserfs_warning(sb, "vs-13080",
1727 "object with this key exists (%k)",
1732 /* insert item, that is body of symlink */
1733 return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1736 /* inserts the stat data into the tree, and then calls
1737 reiserfs_new_directory (to insert ".", ".." item if new object is
1738 directory) or reiserfs_new_symlink (to insert symlink body if new
1739 object is symlink) or nothing (if new object is regular file)
1741 NOTE! uid and gid must already be set in the inode. If we return
1742 non-zero due to an error, we have to drop the quota previously allocated
1743 for the fresh inode. This can only be done outside a transaction, so
1744 if we return non-zero, we also end the transaction. */
1745 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1746 struct inode *dir, int mode, const char *symname,
1747 /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1748 strlen (symname) for symlinks) */
1749 loff_t i_size, struct dentry *dentry,
1750 struct inode *inode)
1752 struct super_block *sb;
1753 struct reiserfs_iget_args args;
1754 INITIALIZE_PATH(path_to_key);
1756 struct item_head ih;
1757 struct stat_data sd;
1761 BUG_ON(!th->t_trans_id);
1763 if (DQUOT_ALLOC_INODE(inode)) {
1767 if (!dir->i_nlink) {
1774 /* item head of new item */
1775 ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1776 ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1777 if (!ih.ih_key.k_objectid) {
1781 args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1782 if (old_format_only(sb))
1783 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1784 TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1786 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1787 TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1788 memcpy(INODE_PKEY(inode), &(ih.ih_key), KEY_SIZE);
1789 args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1790 if (insert_inode_locked4(inode, args.objectid,
1791 reiserfs_find_actor, &args) < 0) {
1795 if (old_format_only(sb))
1796 /* not a perfect generation count, as object ids can be reused, but
1797 ** this is as good as reiserfs can do right now.
1798 ** note that the private part of inode isn't filled in yet, we have
1799 ** to use the directory.
1801 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1803 #if defined( USE_INODE_GENERATION_COUNTER )
1804 inode->i_generation =
1805 le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1807 inode->i_generation = ++event;
1810 /* fill stat data */
1811 inode->i_nlink = (S_ISDIR(mode) ? 2 : 1);
1813 /* uid and gid must already be set by the caller for quota init */
1815 /* symlink cannot be immutable or append only, right? */
1816 if (S_ISLNK(inode->i_mode))
1817 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND);
1819 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
1820 inode->i_size = i_size;
1821 inode->i_blocks = 0;
1823 REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1824 U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1826 INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
1827 REISERFS_I(inode)->i_flags = 0;
1828 REISERFS_I(inode)->i_prealloc_block = 0;
1829 REISERFS_I(inode)->i_prealloc_count = 0;
1830 REISERFS_I(inode)->i_trans_id = 0;
1831 REISERFS_I(inode)->i_jl = NULL;
1832 REISERFS_I(inode)->i_attrs =
1833 REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
1834 sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
1835 mutex_init(&(REISERFS_I(inode)->i_mmap));
1836 reiserfs_init_acl_access(inode);
1837 reiserfs_init_acl_default(inode);
1838 reiserfs_init_xattr_rwsem(inode);
1840 /* key to search for correct place for new stat data */
1841 _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
1842 le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
1843 TYPE_STAT_DATA, 3 /*key length */ );
1845 /* find proper place for inserting of stat data */
1846 retval = search_item(sb, &key, &path_to_key);
1847 if (retval == IO_ERROR) {
1851 if (retval == ITEM_FOUND) {
1852 pathrelse(&path_to_key);
1856 if (old_format_only(sb)) {
1857 if (inode->i_uid & ~0xffff || inode->i_gid & ~0xffff) {
1858 pathrelse(&path_to_key);
1859 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
1863 inode2sd_v1(&sd, inode, inode->i_size);
1865 inode2sd(&sd, inode, inode->i_size);
1867 // store in in-core inode the key of stat data and version all
1868 // object items will have (directory items will have old offset
1869 // format, other new objects will consist of new items)
1870 if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
1871 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1873 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1874 if (old_format_only(sb))
1875 set_inode_sd_version(inode, STAT_DATA_V1);
1877 set_inode_sd_version(inode, STAT_DATA_V2);
1879 /* insert the stat data into the tree */
1880 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1881 if (REISERFS_I(dir)->new_packing_locality)
1882 th->displace_new_blocks = 1;
1885 reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
1889 reiserfs_check_path(&path_to_key);
1892 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
1893 if (!th->displace_new_blocks)
1894 REISERFS_I(dir)->new_packing_locality = 0;
1896 if (S_ISDIR(mode)) {
1897 /* insert item with "." and ".." */
1899 reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
1902 if (S_ISLNK(mode)) {
1903 /* insert body of symlink */
1904 if (!old_format_only(sb))
1905 i_size = ROUND_UP(i_size);
1907 reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
1912 reiserfs_check_path(&path_to_key);
1913 journal_end(th, th->t_super, th->t_blocks_allocated);
1914 goto out_inserted_sd;
1917 /* XXX CHECK THIS */
1918 if (reiserfs_posixacl(inode->i_sb)) {
1919 retval = reiserfs_inherit_default_acl(dir, dentry, inode);
1922 reiserfs_check_path(&path_to_key);
1923 journal_end(th, th->t_super, th->t_blocks_allocated);
1924 goto out_inserted_sd;
1926 } else if (inode->i_sb->s_flags & MS_POSIXACL) {
1927 reiserfs_warning(inode->i_sb, "jdm-13090",
1928 "ACLs aren't enabled in the fs, "
1929 "but vfs thinks they are!");
1930 } else if (IS_PRIVATE(dir))
1931 inode->i_flags |= S_PRIVATE;
1933 reiserfs_update_sd(th, inode);
1934 reiserfs_check_path(&path_to_key);
1938 /* it looks like you can easily compress these two goto targets into
1939 * one. Keeping it like this doesn't actually hurt anything, and they
1940 * are place holders for what the quota code actually needs.
1943 /* Invalidate the object, nothing was inserted yet */
1944 INODE_PKEY(inode)->k_objectid = 0;
1946 /* Quota change must be inside a transaction for journaling */
1947 DQUOT_FREE_INODE(inode);
1950 journal_end(th, th->t_super, th->t_blocks_allocated);
1951 /* Drop can be outside and it needs more credits so it's better to have it outside */
1953 inode->i_flags |= S_NOQUOTA;
1954 make_bad_inode(inode);
1958 th->t_trans_id = 0; /* so the caller can't use this handle later */
1959 unlock_new_inode(inode); /* OK to do even if we hadn't locked it */
1965 ** finds the tail page in the page cache,
1966 ** reads the last block in.
1968 ** On success, page_result is set to a locked, pinned page, and bh_result
1969 ** is set to an up to date buffer for the last block in the file. returns 0.
1971 ** tail conversion is not done, so bh_result might not be valid for writing
1972 ** check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
1973 ** trying to write the block.
1975 ** on failure, nonzero is returned, page_result and bh_result are untouched.
1977 static int grab_tail_page(struct inode *p_s_inode,
1978 struct page **page_result,
1979 struct buffer_head **bh_result)
1982 /* we want the page with the last byte in the file,
1983 ** not the page that will hold the next byte for appending
1985 unsigned long index = (p_s_inode->i_size - 1) >> PAGE_CACHE_SHIFT;
1986 unsigned long pos = 0;
1987 unsigned long start = 0;
1988 unsigned long blocksize = p_s_inode->i_sb->s_blocksize;
1989 unsigned long offset = (p_s_inode->i_size) & (PAGE_CACHE_SIZE - 1);
1990 struct buffer_head *bh;
1991 struct buffer_head *head;
1995 /* we know that we are only called with inode->i_size > 0.
1996 ** we also know that a file tail can never be as big as a block
1997 ** If i_size % blocksize == 0, our file is currently block aligned
1998 ** and it won't need converting or zeroing after a truncate.
2000 if ((offset & (blocksize - 1)) == 0) {
2003 page = grab_cache_page(p_s_inode->i_mapping, index);
2008 /* start within the page of the last block in the file */
2009 start = (offset / blocksize) * blocksize;
2011 error = block_prepare_write(page, start, offset,
2012 reiserfs_get_block_create_0);
2016 head = page_buffers(page);
2022 bh = bh->b_this_page;
2024 } while (bh != head);
2026 if (!buffer_uptodate(bh)) {
2027 /* note, this should never happen, prepare_write should
2028 ** be taking care of this for us. If the buffer isn't up to date,
2029 ** I've screwed up the code to find the buffer, or the code to
2030 ** call prepare_write
2032 reiserfs_error(p_s_inode->i_sb, "clm-6000",
2033 "error reading block %lu", bh->b_blocknr);
2038 *page_result = page;
2045 page_cache_release(page);
2050 ** vfs version of truncate file. Must NOT be called with
2051 ** a transaction already started.
2053 ** some code taken from block_truncate_page
2055 int reiserfs_truncate_file(struct inode *p_s_inode, int update_timestamps)
2057 struct reiserfs_transaction_handle th;
2058 /* we want the offset for the first byte after the end of the file */
2059 unsigned long offset = p_s_inode->i_size & (PAGE_CACHE_SIZE - 1);
2060 unsigned blocksize = p_s_inode->i_sb->s_blocksize;
2062 struct page *page = NULL;
2064 struct buffer_head *bh = NULL;
2067 reiserfs_write_lock(p_s_inode->i_sb);
2069 if (p_s_inode->i_size > 0) {
2070 if ((error = grab_tail_page(p_s_inode, &page, &bh))) {
2071 // -ENOENT means we truncated past the end of the file,
2072 // and get_block_create_0 could not find a block to read in,
2074 if (error != -ENOENT)
2075 reiserfs_error(p_s_inode->i_sb, "clm-6001",
2076 "grab_tail_page failed %d",
2083 /* so, if page != NULL, we have a buffer head for the offset at
2084 ** the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2085 ** then we have an unformatted node. Otherwise, we have a direct item,
2086 ** and no zeroing is required on disk. We zero after the truncate,
2087 ** because the truncate might pack the item anyway
2088 ** (it will unmap bh if it packs).
2090 /* it is enough to reserve space in transaction for 2 balancings:
2091 one for "save" link adding and another for the first
2092 cut_from_item. 1 is for update_sd */
2093 error = journal_begin(&th, p_s_inode->i_sb,
2094 JOURNAL_PER_BALANCE_CNT * 2 + 1);
2097 reiserfs_update_inode_transaction(p_s_inode);
2098 if (update_timestamps)
2099 /* we are doing real truncate: if the system crashes before the last
2100 transaction of truncating gets committed - on reboot the file
2101 either appears truncated properly or not truncated at all */
2102 add_save_link(&th, p_s_inode, 1);
2103 err2 = reiserfs_do_truncate(&th, p_s_inode, page, update_timestamps);
2105 journal_end(&th, p_s_inode->i_sb, JOURNAL_PER_BALANCE_CNT * 2 + 1);
2109 /* check reiserfs_do_truncate after ending the transaction */
2115 if (update_timestamps) {
2116 error = remove_save_link(p_s_inode, 1 /* truncate */ );
2122 length = offset & (blocksize - 1);
2123 /* if we are not on a block boundary */
2125 length = blocksize - length;
2126 zero_user(page, offset, length);
2127 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2128 mark_buffer_dirty(bh);
2132 page_cache_release(page);
2135 reiserfs_write_unlock(p_s_inode->i_sb);
2140 page_cache_release(page);
2142 reiserfs_write_unlock(p_s_inode->i_sb);
2146 static int map_block_for_writepage(struct inode *inode,
2147 struct buffer_head *bh_result,
2148 unsigned long block)
2150 struct reiserfs_transaction_handle th;
2152 struct item_head tmp_ih;
2153 struct item_head *ih;
2154 struct buffer_head *bh;
2157 INITIALIZE_PATH(path);
2159 int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2160 loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2162 int use_get_block = 0;
2163 int bytes_copied = 0;
2165 int trans_running = 0;
2167 /* catch places below that try to log something without starting a trans */
2170 if (!buffer_uptodate(bh_result)) {
2174 kmap(bh_result->b_page);
2176 reiserfs_write_lock(inode->i_sb);
2177 make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2180 retval = search_for_position_by_key(inode->i_sb, &key, &path);
2181 if (retval != POSITION_FOUND) {
2186 bh = get_last_bh(&path);
2188 item = get_item(&path);
2189 pos_in_item = path.pos_in_item;
2191 /* we've found an unformatted node */
2192 if (indirect_item_found(retval, ih)) {
2193 if (bytes_copied > 0) {
2194 reiserfs_warning(inode->i_sb, "clm-6002",
2195 "bytes_copied %d", bytes_copied);
2197 if (!get_block_num(item, pos_in_item)) {
2198 /* crap, we are writing to a hole */
2202 set_block_dev_mapped(bh_result,
2203 get_block_num(item, pos_in_item), inode);
2204 } else if (is_direct_le_ih(ih)) {
2206 p = page_address(bh_result->b_page);
2207 p += (byte_offset - 1) & (PAGE_CACHE_SIZE - 1);
2208 copy_size = ih_item_len(ih) - pos_in_item;
2210 fs_gen = get_generation(inode->i_sb);
2211 copy_item_head(&tmp_ih, ih);
2213 if (!trans_running) {
2214 /* vs-3050 is gone, no need to drop the path */
2215 retval = journal_begin(&th, inode->i_sb, jbegin_count);
2218 reiserfs_update_inode_transaction(inode);
2220 if (fs_changed(fs_gen, inode->i_sb)
2221 && item_moved(&tmp_ih, &path)) {
2222 reiserfs_restore_prepared_buffer(inode->i_sb,
2228 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2230 if (fs_changed(fs_gen, inode->i_sb)
2231 && item_moved(&tmp_ih, &path)) {
2232 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2236 memcpy(B_I_PITEM(bh, ih) + pos_in_item, p + bytes_copied,
2239 journal_mark_dirty(&th, inode->i_sb, bh);
2240 bytes_copied += copy_size;
2241 set_block_dev_mapped(bh_result, 0, inode);
2243 /* are there still bytes left? */
2244 if (bytes_copied < bh_result->b_size &&
2245 (byte_offset + bytes_copied) < inode->i_size) {
2246 set_cpu_key_k_offset(&key,
2247 cpu_key_k_offset(&key) +
2252 reiserfs_warning(inode->i_sb, "clm-6003",
2253 "bad item inode %lu", inode->i_ino);
2261 if (trans_running) {
2262 int err = journal_end(&th, inode->i_sb, jbegin_count);
2267 reiserfs_write_unlock(inode->i_sb);
2269 /* this is where we fill in holes in the file. */
2270 if (use_get_block) {
2271 retval = reiserfs_get_block(inode, block, bh_result,
2272 GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2273 | GET_BLOCK_NO_DANGLE);
2275 if (!buffer_mapped(bh_result)
2276 || bh_result->b_blocknr == 0) {
2277 /* get_block failed to find a mapped unformatted node. */
2283 kunmap(bh_result->b_page);
2285 if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2286 /* we've copied data from the page into the direct item, so the
2287 * buffer in the page is now clean, mark it to reflect that.
2289 lock_buffer(bh_result);
2290 clear_buffer_dirty(bh_result);
2291 unlock_buffer(bh_result);
2297 * mason@suse.com: updated in 2.5.54 to follow the same general io
2298 * start/recovery path as __block_write_full_page, along with special
2299 * code to handle reiserfs tails.
2301 static int reiserfs_write_full_page(struct page *page,
2302 struct writeback_control *wbc)
2304 struct inode *inode = page->mapping->host;
2305 unsigned long end_index = inode->i_size >> PAGE_CACHE_SHIFT;
2307 unsigned long block;
2308 sector_t last_block;
2309 struct buffer_head *head, *bh;
2312 int checked = PageChecked(page);
2313 struct reiserfs_transaction_handle th;
2314 struct super_block *s = inode->i_sb;
2315 int bh_per_page = PAGE_CACHE_SIZE / s->s_blocksize;
2318 /* no logging allowed when nonblocking or from PF_MEMALLOC */
2319 if (checked && (current->flags & PF_MEMALLOC)) {
2320 redirty_page_for_writepage(wbc, page);
2325 /* The page dirty bit is cleared before writepage is called, which
2326 * means we have to tell create_empty_buffers to make dirty buffers
2327 * The page really should be up to date at this point, so tossing
2328 * in the BH_Uptodate is just a sanity check.
2330 if (!page_has_buffers(page)) {
2331 create_empty_buffers(page, s->s_blocksize,
2332 (1 << BH_Dirty) | (1 << BH_Uptodate));
2334 head = page_buffers(page);
2336 /* last page in the file, zero out any contents past the
2337 ** last byte in the file
2339 if (page->index >= end_index) {
2340 unsigned last_offset;
2342 last_offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
2343 /* no file contents in this page */
2344 if (page->index >= end_index + 1 || !last_offset) {
2348 zero_user_segment(page, last_offset, PAGE_CACHE_SIZE);
2351 block = page->index << (PAGE_CACHE_SHIFT - s->s_blocksize_bits);
2352 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2353 /* first map all the buffers, logging any direct items we find */
2355 if (block > last_block) {
2357 * This can happen when the block size is less than
2358 * the page size. The corresponding bytes in the page
2359 * were zero filled above
2361 clear_buffer_dirty(bh);
2362 set_buffer_uptodate(bh);
2363 } else if ((checked || buffer_dirty(bh)) &&
2364 (!buffer_mapped(bh) || (buffer_mapped(bh)
2367 /* not mapped yet, or it points to a direct item, search
2368 * the btree for the mapping info, and log any direct
2371 if ((error = map_block_for_writepage(inode, bh, block))) {
2375 bh = bh->b_this_page;
2377 } while (bh != head);
2380 * we start the transaction after map_block_for_writepage,
2381 * because it can create holes in the file (an unbounded operation).
2382 * starting it here, we can make a reliable estimate for how many
2383 * blocks we're going to log
2386 ClearPageChecked(page);
2387 reiserfs_write_lock(s);
2388 error = journal_begin(&th, s, bh_per_page + 1);
2390 reiserfs_write_unlock(s);
2393 reiserfs_update_inode_transaction(inode);
2395 /* now go through and lock any dirty buffers on the page */
2398 if (!buffer_mapped(bh))
2400 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2404 reiserfs_prepare_for_journal(s, bh, 1);
2405 journal_mark_dirty(&th, s, bh);
2408 /* from this point on, we know the buffer is mapped to a
2409 * real block and not a direct item
2411 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
2414 if (!trylock_buffer(bh)) {
2415 redirty_page_for_writepage(wbc, page);
2419 if (test_clear_buffer_dirty(bh)) {
2420 mark_buffer_async_write(bh);
2424 } while ((bh = bh->b_this_page) != head);
2427 error = journal_end(&th, s, bh_per_page + 1);
2428 reiserfs_write_unlock(s);
2432 BUG_ON(PageWriteback(page));
2433 set_page_writeback(page);
2437 * since any buffer might be the only dirty buffer on the page,
2438 * the first submit_bh can bring the page out of writeback.
2439 * be careful with the buffers.
2442 struct buffer_head *next = bh->b_this_page;
2443 if (buffer_async_write(bh)) {
2444 submit_bh(WRITE, bh);
2449 } while (bh != head);
2455 * if this page only had a direct item, it is very possible for
2456 * no io to be required without there being an error. Or,
2457 * someone else could have locked them and sent them down the
2458 * pipe without locking the page
2462 if (!buffer_uptodate(bh)) {
2466 bh = bh->b_this_page;
2467 } while (bh != head);
2469 SetPageUptodate(page);
2470 end_page_writeback(page);
2475 /* catches various errors, we need to make sure any valid dirty blocks
2476 * get to the media. The page is currently locked and not marked for
2479 ClearPageUptodate(page);
2483 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2485 mark_buffer_async_write(bh);
2488 * clear any dirty bits that might have come from getting
2489 * attached to a dirty page
2491 clear_buffer_dirty(bh);
2493 bh = bh->b_this_page;
2494 } while (bh != head);
2496 BUG_ON(PageWriteback(page));
2497 set_page_writeback(page);
2500 struct buffer_head *next = bh->b_this_page;
2501 if (buffer_async_write(bh)) {
2502 clear_buffer_dirty(bh);
2503 submit_bh(WRITE, bh);
2508 } while (bh != head);
2512 static int reiserfs_readpage(struct file *f, struct page *page)
2514 return block_read_full_page(page, reiserfs_get_block);
2517 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2519 struct inode *inode = page->mapping->host;
2520 reiserfs_wait_on_write_block(inode->i_sb);
2521 return reiserfs_write_full_page(page, wbc);
2524 static int reiserfs_write_begin(struct file *file,
2525 struct address_space *mapping,
2526 loff_t pos, unsigned len, unsigned flags,
2527 struct page **pagep, void **fsdata)
2529 struct inode *inode;
2535 inode = mapping->host;
2537 if (flags & AOP_FLAG_CONT_EXPAND &&
2538 (pos & (inode->i_sb->s_blocksize - 1)) == 0) {
2540 *fsdata = (void *)(unsigned long)flags;
2543 index = pos >> PAGE_CACHE_SHIFT;
2544 page = grab_cache_page_write_begin(mapping, index, flags);
2549 reiserfs_wait_on_write_block(inode->i_sb);
2550 fix_tail_page_for_writing(page);
2551 if (reiserfs_transaction_running(inode->i_sb)) {
2552 struct reiserfs_transaction_handle *th;
2553 th = (struct reiserfs_transaction_handle *)current->
2555 BUG_ON(!th->t_refcount);
2556 BUG_ON(!th->t_trans_id);
2557 old_ref = th->t_refcount;
2560 ret = block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
2561 reiserfs_get_block);
2562 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2563 struct reiserfs_transaction_handle *th = current->journal_info;
2564 /* this gets a little ugly. If reiserfs_get_block returned an
2565 * error and left a transacstion running, we've got to close it,
2566 * and we've got to free handle if it was a persistent transaction.
2568 * But, if we had nested into an existing transaction, we need
2569 * to just drop the ref count on the handle.
2571 * If old_ref == 0, the transaction is from reiserfs_get_block,
2572 * and it was a persistent trans. Otherwise, it was nested above.
2574 if (th->t_refcount > old_ref) {
2579 reiserfs_write_lock(inode->i_sb);
2580 err = reiserfs_end_persistent_transaction(th);
2581 reiserfs_write_unlock(inode->i_sb);
2589 page_cache_release(page);
2594 int reiserfs_prepare_write(struct file *f, struct page *page,
2595 unsigned from, unsigned to)
2597 struct inode *inode = page->mapping->host;
2601 reiserfs_wait_on_write_block(inode->i_sb);
2602 fix_tail_page_for_writing(page);
2603 if (reiserfs_transaction_running(inode->i_sb)) {
2604 struct reiserfs_transaction_handle *th;
2605 th = (struct reiserfs_transaction_handle *)current->
2607 BUG_ON(!th->t_refcount);
2608 BUG_ON(!th->t_trans_id);
2609 old_ref = th->t_refcount;
2613 ret = block_prepare_write(page, from, to, reiserfs_get_block);
2614 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2615 struct reiserfs_transaction_handle *th = current->journal_info;
2616 /* this gets a little ugly. If reiserfs_get_block returned an
2617 * error and left a transacstion running, we've got to close it,
2618 * and we've got to free handle if it was a persistent transaction.
2620 * But, if we had nested into an existing transaction, we need
2621 * to just drop the ref count on the handle.
2623 * If old_ref == 0, the transaction is from reiserfs_get_block,
2624 * and it was a persistent trans. Otherwise, it was nested above.
2626 if (th->t_refcount > old_ref) {
2631 reiserfs_write_lock(inode->i_sb);
2632 err = reiserfs_end_persistent_transaction(th);
2633 reiserfs_write_unlock(inode->i_sb);
2643 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2645 return generic_block_bmap(as, block, reiserfs_bmap);
2648 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2649 loff_t pos, unsigned len, unsigned copied,
2650 struct page *page, void *fsdata)
2652 struct inode *inode = page->mapping->host;
2655 struct reiserfs_transaction_handle *th;
2658 if ((unsigned long)fsdata & AOP_FLAG_CONT_EXPAND)
2661 reiserfs_wait_on_write_block(inode->i_sb);
2662 if (reiserfs_transaction_running(inode->i_sb))
2663 th = current->journal_info;
2667 start = pos & (PAGE_CACHE_SIZE - 1);
2668 if (unlikely(copied < len)) {
2669 if (!PageUptodate(page))
2672 page_zero_new_buffers(page, start + copied, start + len);
2674 flush_dcache_page(page);
2676 reiserfs_commit_page(inode, page, start, start + copied);
2678 /* generic_commit_write does this for us, but does not update the
2679 ** transaction tracking stuff when the size changes. So, we have
2680 ** to do the i_size updates here.
2683 if (pos > inode->i_size) {
2684 struct reiserfs_transaction_handle myth;
2685 reiserfs_write_lock(inode->i_sb);
2686 /* If the file have grown beyond the border where it
2687 can have a tail, unmark it as needing a tail
2689 if ((have_large_tails(inode->i_sb)
2690 && inode->i_size > i_block_size(inode) * 4)
2691 || (have_small_tails(inode->i_sb)
2692 && inode->i_size > i_block_size(inode)))
2693 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2695 ret = journal_begin(&myth, inode->i_sb, 1);
2697 reiserfs_write_unlock(inode->i_sb);
2700 reiserfs_update_inode_transaction(inode);
2701 inode->i_size = pos;
2703 * this will just nest into our transaction. It's important
2704 * to use mark_inode_dirty so the inode gets pushed around on the
2705 * dirty lists, and so that O_SYNC works as expected
2707 mark_inode_dirty(inode);
2708 reiserfs_update_sd(&myth, inode);
2710 ret = journal_end(&myth, inode->i_sb, 1);
2711 reiserfs_write_unlock(inode->i_sb);
2716 reiserfs_write_lock(inode->i_sb);
2718 mark_inode_dirty(inode);
2719 ret = reiserfs_end_persistent_transaction(th);
2720 reiserfs_write_unlock(inode->i_sb);
2727 page_cache_release(page);
2728 return ret == 0 ? copied : ret;
2732 reiserfs_write_lock(inode->i_sb);
2734 reiserfs_update_sd(th, inode);
2735 ret = reiserfs_end_persistent_transaction(th);
2736 reiserfs_write_unlock(inode->i_sb);
2742 int reiserfs_commit_write(struct file *f, struct page *page,
2743 unsigned from, unsigned to)
2745 struct inode *inode = page->mapping->host;
2746 loff_t pos = ((loff_t) page->index << PAGE_CACHE_SHIFT) + to;
2749 struct reiserfs_transaction_handle *th = NULL;
2751 reiserfs_wait_on_write_block(inode->i_sb);
2752 if (reiserfs_transaction_running(inode->i_sb)) {
2753 th = current->journal_info;
2755 reiserfs_commit_page(inode, page, from, to);
2757 /* generic_commit_write does this for us, but does not update the
2758 ** transaction tracking stuff when the size changes. So, we have
2759 ** to do the i_size updates here.
2761 if (pos > inode->i_size) {
2762 struct reiserfs_transaction_handle myth;
2763 reiserfs_write_lock(inode->i_sb);
2764 /* If the file have grown beyond the border where it
2765 can have a tail, unmark it as needing a tail
2767 if ((have_large_tails(inode->i_sb)
2768 && inode->i_size > i_block_size(inode) * 4)
2769 || (have_small_tails(inode->i_sb)
2770 && inode->i_size > i_block_size(inode)))
2771 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2773 ret = journal_begin(&myth, inode->i_sb, 1);
2775 reiserfs_write_unlock(inode->i_sb);
2778 reiserfs_update_inode_transaction(inode);
2779 inode->i_size = pos;
2781 * this will just nest into our transaction. It's important
2782 * to use mark_inode_dirty so the inode gets pushed around on the
2783 * dirty lists, and so that O_SYNC works as expected
2785 mark_inode_dirty(inode);
2786 reiserfs_update_sd(&myth, inode);
2788 ret = journal_end(&myth, inode->i_sb, 1);
2789 reiserfs_write_unlock(inode->i_sb);
2794 reiserfs_write_lock(inode->i_sb);
2796 mark_inode_dirty(inode);
2797 ret = reiserfs_end_persistent_transaction(th);
2798 reiserfs_write_unlock(inode->i_sb);
2808 reiserfs_write_lock(inode->i_sb);
2810 reiserfs_update_sd(th, inode);
2811 ret = reiserfs_end_persistent_transaction(th);
2812 reiserfs_write_unlock(inode->i_sb);
2818 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
2820 if (reiserfs_attrs(inode->i_sb)) {
2821 if (sd_attrs & REISERFS_SYNC_FL)
2822 inode->i_flags |= S_SYNC;
2824 inode->i_flags &= ~S_SYNC;
2825 if (sd_attrs & REISERFS_IMMUTABLE_FL)
2826 inode->i_flags |= S_IMMUTABLE;
2828 inode->i_flags &= ~S_IMMUTABLE;
2829 if (sd_attrs & REISERFS_APPEND_FL)
2830 inode->i_flags |= S_APPEND;
2832 inode->i_flags &= ~S_APPEND;
2833 if (sd_attrs & REISERFS_NOATIME_FL)
2834 inode->i_flags |= S_NOATIME;
2836 inode->i_flags &= ~S_NOATIME;
2837 if (sd_attrs & REISERFS_NOTAIL_FL)
2838 REISERFS_I(inode)->i_flags |= i_nopack_mask;
2840 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
2844 void i_attrs_to_sd_attrs(struct inode *inode, __u16 * sd_attrs)
2846 if (reiserfs_attrs(inode->i_sb)) {
2847 if (inode->i_flags & S_IMMUTABLE)
2848 *sd_attrs |= REISERFS_IMMUTABLE_FL;
2850 *sd_attrs &= ~REISERFS_IMMUTABLE_FL;
2851 if (inode->i_flags & S_SYNC)
2852 *sd_attrs |= REISERFS_SYNC_FL;
2854 *sd_attrs &= ~REISERFS_SYNC_FL;
2855 if (inode->i_flags & S_NOATIME)
2856 *sd_attrs |= REISERFS_NOATIME_FL;
2858 *sd_attrs &= ~REISERFS_NOATIME_FL;
2859 if (REISERFS_I(inode)->i_flags & i_nopack_mask)
2860 *sd_attrs |= REISERFS_NOTAIL_FL;
2862 *sd_attrs &= ~REISERFS_NOTAIL_FL;
2866 /* decide if this buffer needs to stay around for data logging or ordered
2869 static int invalidatepage_can_drop(struct inode *inode, struct buffer_head *bh)
2872 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2875 spin_lock(&j->j_dirty_buffers_lock);
2876 if (!buffer_mapped(bh)) {
2879 /* the page is locked, and the only places that log a data buffer
2880 * also lock the page.
2882 if (reiserfs_file_data_log(inode)) {
2884 * very conservative, leave the buffer pinned if
2885 * anyone might need it.
2887 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
2890 } else if (buffer_dirty(bh)) {
2891 struct reiserfs_journal_list *jl;
2892 struct reiserfs_jh *jh = bh->b_private;
2894 /* why is this safe?
2895 * reiserfs_setattr updates i_size in the on disk
2896 * stat data before allowing vmtruncate to be called.
2898 * If buffer was put onto the ordered list for this
2899 * transaction, we know for sure either this transaction
2900 * or an older one already has updated i_size on disk,
2901 * and this ordered data won't be referenced in the file
2904 * if the buffer was put onto the ordered list for an older
2905 * transaction, we need to leave it around
2907 if (jh && (jl = jh->jl)
2908 && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
2912 if (ret && bh->b_private) {
2913 reiserfs_free_jh(bh);
2915 spin_unlock(&j->j_dirty_buffers_lock);
2920 /* clm -- taken from fs/buffer.c:block_invalidate_page */
2921 static void reiserfs_invalidatepage(struct page *page, unsigned long offset)
2923 struct buffer_head *head, *bh, *next;
2924 struct inode *inode = page->mapping->host;
2925 unsigned int curr_off = 0;
2928 BUG_ON(!PageLocked(page));
2931 ClearPageChecked(page);
2933 if (!page_has_buffers(page))
2936 head = page_buffers(page);
2939 unsigned int next_off = curr_off + bh->b_size;
2940 next = bh->b_this_page;
2943 * is this block fully invalidated?
2945 if (offset <= curr_off) {
2946 if (invalidatepage_can_drop(inode, bh))
2947 reiserfs_unmap_buffer(bh);
2951 curr_off = next_off;
2953 } while (bh != head);
2956 * We release buffers only if the entire page is being invalidated.
2957 * The get_block cached value has been unconditionally invalidated,
2958 * so real IO is not possible anymore.
2960 if (!offset && ret) {
2961 ret = try_to_release_page(page, 0);
2962 /* maybe should BUG_ON(!ret); - neilb */
2968 static int reiserfs_set_page_dirty(struct page *page)
2970 struct inode *inode = page->mapping->host;
2971 if (reiserfs_file_data_log(inode)) {
2972 SetPageChecked(page);
2973 return __set_page_dirty_nobuffers(page);
2975 return __set_page_dirty_buffers(page);
2979 * Returns 1 if the page's buffers were dropped. The page is locked.
2981 * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
2982 * in the buffers at page_buffers(page).
2984 * even in -o notail mode, we can't be sure an old mount without -o notail
2985 * didn't create files with tails.
2987 static int reiserfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2989 struct inode *inode = page->mapping->host;
2990 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
2991 struct buffer_head *head;
2992 struct buffer_head *bh;
2995 WARN_ON(PageChecked(page));
2996 spin_lock(&j->j_dirty_buffers_lock);
2997 head = page_buffers(page);
3000 if (bh->b_private) {
3001 if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3002 reiserfs_free_jh(bh);
3008 bh = bh->b_this_page;
3009 } while (bh != head);
3011 ret = try_to_free_buffers(page);
3012 spin_unlock(&j->j_dirty_buffers_lock);
3016 /* We thank Mingming Cao for helping us understand in great detail what
3017 to do in this section of the code. */
3018 static ssize_t reiserfs_direct_IO(int rw, struct kiocb *iocb,
3019 const struct iovec *iov, loff_t offset,
3020 unsigned long nr_segs)
3022 struct file *file = iocb->ki_filp;
3023 struct inode *inode = file->f_mapping->host;
3025 return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3027 reiserfs_get_blocks_direct_io, NULL);
3030 int reiserfs_setattr(struct dentry *dentry, struct iattr *attr)
3032 struct inode *inode = dentry->d_inode;
3034 unsigned int ia_valid;
3036 /* must be turned off for recursive notify_change calls */
3037 ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3039 reiserfs_write_lock(inode->i_sb);
3040 if (attr->ia_valid & ATTR_SIZE) {
3041 /* version 2 items will be caught by the s_maxbytes check
3042 ** done for us in vmtruncate
3044 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3045 attr->ia_size > MAX_NON_LFS) {
3049 /* fill in hole pointers in the expanding truncate case. */
3050 if (attr->ia_size > inode->i_size) {
3051 error = generic_cont_expand_simple(inode, attr->ia_size);
3052 if (REISERFS_I(inode)->i_prealloc_count > 0) {
3054 struct reiserfs_transaction_handle th;
3055 /* we're changing at most 2 bitmaps, inode + super */
3056 err = journal_begin(&th, inode->i_sb, 4);
3058 reiserfs_discard_prealloc(&th, inode);
3059 err = journal_end(&th, inode->i_sb, 4);
3067 * file size is changed, ctime and mtime are
3070 attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3074 if ((((attr->ia_valid & ATTR_UID) && (attr->ia_uid & ~0xffff)) ||
3075 ((attr->ia_valid & ATTR_GID) && (attr->ia_gid & ~0xffff))) &&
3076 (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3077 /* stat data of format v3.5 has 16 bit uid and gid */
3082 error = inode_change_ok(inode, attr);
3084 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
3085 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
3086 error = reiserfs_chown_xattrs(inode, attr);
3089 struct reiserfs_transaction_handle th;
3092 (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3093 REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3096 /* (user+group)*(old+new) structure - we count quota info and , inode write (sb, inode) */
3098 journal_begin(&th, inode->i_sb,
3103 DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
3105 journal_end(&th, inode->i_sb,
3109 /* Update corresponding info in inode so that everything is in
3110 * one transaction */
3111 if (attr->ia_valid & ATTR_UID)
3112 inode->i_uid = attr->ia_uid;
3113 if (attr->ia_valid & ATTR_GID)
3114 inode->i_gid = attr->ia_gid;
3115 mark_inode_dirty(inode);
3117 journal_end(&th, inode->i_sb, jbegin_count);
3121 error = inode_setattr(inode, attr);
3124 if (!error && reiserfs_posixacl(inode->i_sb)) {
3125 if (attr->ia_valid & ATTR_MODE)
3126 error = reiserfs_acl_chmod(inode);
3130 reiserfs_write_unlock(inode->i_sb);
3134 const struct address_space_operations reiserfs_address_space_operations = {
3135 .writepage = reiserfs_writepage,
3136 .readpage = reiserfs_readpage,
3137 .readpages = reiserfs_readpages,
3138 .releasepage = reiserfs_releasepage,
3139 .invalidatepage = reiserfs_invalidatepage,
3140 .sync_page = block_sync_page,
3141 .write_begin = reiserfs_write_begin,
3142 .write_end = reiserfs_write_end,
3143 .bmap = reiserfs_aop_bmap,
3144 .direct_IO = reiserfs_direct_IO,
3145 .set_page_dirty = reiserfs_set_page_dirty,