2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
5 #include <linux/time.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>
24 int reiserfs_commit_write(struct file *f, struct page *page,
25 unsigned from, unsigned to);
27 void reiserfs_evict_inode(struct inode *inode)
30 * We need blocks for transaction + (user+group) quota
31 * update (possibly delete)
34 JOURNAL_PER_BALANCE_CNT * 2 +
35 2 * REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb);
36 struct reiserfs_transaction_handle th;
39 if (!inode->i_nlink && !is_bad_inode(inode))
40 dquot_initialize(inode);
42 truncate_inode_pages_final(&inode->i_data);
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
51 if (!(inode->i_state & I_NEW) && INODE_PKEY(inode)->k_objectid != 0) {
53 reiserfs_delete_xattrs(inode);
55 reiserfs_write_lock(inode->i_sb);
57 if (journal_begin(&th, inode->i_sb, jbegin_count))
59 reiserfs_update_inode_transaction(inode);
61 reiserfs_discard_prealloc(&th, inode);
63 err = reiserfs_delete_object(&th, inode);
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
71 int depth = reiserfs_write_unlock_nested(inode->i_sb);
72 dquot_free_inode(inode);
73 reiserfs_write_lock_nested(inode->i_sb, depth);
80 * check return value from reiserfs_delete_object after
81 * ending the transaction
87 * all items of file are deleted, so we can remove
89 * we can't do anything about an error here
91 remove_save_link(inode, 0 /* not truncate */);
93 reiserfs_write_unlock(inode->i_sb);
95 /* no object items are in the tree */
99 /* note this must go after the journal_end to prevent deadlock */
111 static void _make_cpu_key(struct cpu_key *key, int version, __u32 dirid,
112 __u32 objectid, loff_t offset, int type, int length)
114 key->version = version;
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;
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
127 void make_cpu_key(struct cpu_key *key, struct inode *inode, loff_t offset,
128 int type, int length)
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,
136 /* when key is 0, do not set version and short key */
137 inline void make_le_item_head(struct item_head *ih, const struct cpu_key *key,
139 loff_t offset, int type, int length,
140 int entry_count /*or ih_free_space */ )
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);
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); */
153 * for directory items it is entry count, for directs and stat
154 * datas - 0xffff, for indirects - 0
156 put_ih_entry_count(ih, entry_count);
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....
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.
176 * The size of the cache could be dynamic based on the size of the file.
178 * I'd also like to see us cache the location the stat data item, since
179 * we are needlessly researching for that frequently.
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
193 static inline void fix_tail_page_for_writing(struct page *page)
195 struct buffer_head *head, *next, *bh;
197 if (page && page_has_buffers(page)) {
198 head = page_buffers(page);
201 next = bh->b_this_page;
202 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
203 reiserfs_unmap_buffer(bh);
206 } while (bh != head);
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
214 static inline int allocation_needed(int retval, b_blocknr_t allocated,
215 struct item_head *ih,
216 __le32 * item, int pos_in_item)
220 if (retval == POSITION_FOUND && is_indirect_le_ih(ih) &&
221 get_block_num(item, pos_in_item))
226 static inline int indirect_item_found(int retval, struct item_head *ih)
228 return (retval == POSITION_FOUND) && is_indirect_le_ih(ih);
231 static inline void set_block_dev_mapped(struct buffer_head *bh,
232 b_blocknr_t block, struct inode *inode)
234 map_bh(bh, inode->i_sb, block);
238 * files which were created in the earlier version can not be longer,
241 static int file_capable(struct inode *inode, sector_t block)
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)))
252 static int restart_transaction(struct reiserfs_transaction_handle *th,
253 struct inode *inode, struct treepath *path)
255 struct super_block *s = th->t_super;
258 BUG_ON(!th->t_trans_id);
259 BUG_ON(!th->t_refcount);
263 /* we cannot restart while nested */
264 if (th->t_refcount > 1) {
267 reiserfs_update_sd(th, inode);
268 err = journal_end(th);
270 err = journal_begin(th, s, JOURNAL_PER_BALANCE_CNT * 6);
272 reiserfs_update_inode_transaction(inode);
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.
285 static int _get_block_create_0(struct inode *inode, sector_t block,
286 struct buffer_head *bh_result, int args)
288 INITIALIZE_PATH(path);
290 struct buffer_head *bh;
291 struct item_head *ih, tmp_ih;
298 unsigned long offset;
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,
305 result = search_for_position_by_key(inode->i_sb, &key, &path);
306 if (result != POSITION_FOUND) {
308 if (result == IO_ERROR)
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.
315 if ((args & GET_BLOCK_NO_HOLE)
316 && !PageUptodate(bh_result->b_page)) {
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);
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
332 blocknr = get_block_num(ind_item, path.pos_in_item);
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);
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.
347 if ((args & GET_BLOCK_NO_HOLE)
348 && !PageUptodate(bh_result->b_page)) {
355 /* requested data are in direct item(s) */
356 if (!(args & GET_BLOCK_READ_DIRECT)) {
358 * we are called by bmap. FIXME: we can not map block of file
359 * when it is stored in direct item(s)
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
370 if (buffer_uptodate(bh_result)) {
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
379 if (!bh_result->b_page || PageUptodate(bh_result->b_page)) {
380 set_buffer_uptodate(bh_result);
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);
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
393 p = (char *)kmap(bh_result->b_page);
395 memset(p, 0, inode->i_sb->s_blocksize);
397 if (!is_direct_le_ih(ih)) {
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
406 if ((le_ih_k_offset(ih) + path.pos_in_item) > inode->i_size)
408 if ((le_ih_k_offset(ih) - 1 + ih_item_len(ih)) > inode->i_size) {
410 inode->i_size - (le_ih_k_offset(ih) - 1) -
414 chars = ih_item_len(ih) - path.pos_in_item;
416 memcpy(p, ih_item_body(bh, ih) + path.pos_in_item, chars);
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
429 if (PATH_LAST_POSITION(&path) != (B_NR_ITEMS(bh) - 1))
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 */
438 bh = get_last_bh(&path);
439 ih = tp_item_head(&path);
442 flush_dcache_page(bh_result->b_page);
443 kunmap(bh_result->b_page);
448 if (result == IO_ERROR)
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
455 map_bh(bh_result, inode->i_sb, 0);
456 set_buffer_uptodate(bh_result);
461 * this is called to create file map. So, _get_block_create_0 will not
464 static int reiserfs_bmap(struct inode *inode, sector_t block,
465 struct buffer_head *bh_result, int create)
467 if (!file_capable(inode, block))
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);
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
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.
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.
495 static int reiserfs_get_block_create_0(struct inode *inode, sector_t block,
496 struct buffer_head *bh_result,
499 return reiserfs_get_block(inode, block, bh_result, GET_BLOCK_NO_HOLE);
503 * This is special helper for reiserfs_get_block in case we are executing
506 static int reiserfs_get_blocks_direct_io(struct inode *inode,
508 struct buffer_head *bh_result,
513 bh_result->b_page = NULL;
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()
520 bh_result->b_size = i_blocksize(inode);
522 ret = reiserfs_get_block(inode, iblock, bh_result,
523 create | GET_BLOCK_NO_DANGLE);
527 /* don't allow direct io onto tail pages */
528 if (buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
530 * make sure future calls to the direct io funcs for this
531 * offset in the file fail by unmapping the buffer
533 clear_buffer_mapped(bh_result);
538 * Possible unpacked tail. Flush the data before pages have
541 if (REISERFS_I(inode)->i_flags & i_pack_on_close_mask) {
544 reiserfs_write_lock(inode->i_sb);
546 err = reiserfs_commit_for_inode(inode);
547 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
549 reiserfs_write_unlock(inode->i_sb);
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
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
568 static int convert_tail_for_hole(struct inode *inode,
569 struct buffer_head *bh_result,
573 unsigned long tail_end;
574 unsigned long tail_start;
575 struct page *tail_page;
576 struct page *hole_page = bh_result->b_page;
579 if ((tail_offset & (bh_result->b_size - 1)) != 1)
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;
586 index = tail_offset >> PAGE_SHIFT;
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
591 if (!hole_page || index != hole_page->index) {
592 tail_page = grab_cache_page(inode->i_mapping, index);
598 tail_page = hole_page;
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.
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.
611 fix_tail_page_for_writing(tail_page);
612 retval = __reiserfs_write_begin(tail_page, tail_start,
613 tail_end - tail_start);
617 /* tail conversion might change the data in the page */
618 flush_dcache_page(tail_page);
620 retval = reiserfs_commit_write(NULL, tail_page, tail_start, tail_end);
623 if (tail_page != hole_page) {
624 unlock_page(tail_page);
631 static inline int _allocate_block(struct reiserfs_transaction_handle *th,
634 b_blocknr_t * allocated_block_nr,
635 struct treepath *path, int flags)
637 BUG_ON(!th->t_trans_id);
639 #ifdef REISERFS_PREALLOCATE
640 if (!(flags & GET_BLOCK_NO_IMUX)) {
641 return reiserfs_new_unf_blocknrs2(th, inode, allocated_block_nr,
645 return reiserfs_new_unf_blocknrs(th, inode, allocated_block_nr, path,
649 int reiserfs_get_block(struct inode *inode, sector_t block,
650 struct buffer_head *bh_result, int create)
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);
658 struct buffer_head *bh, *unbh = NULL;
659 struct item_head *ih, tmp_ih;
663 struct reiserfs_transaction_handle *th = NULL;
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
673 JOURNAL_PER_BALANCE_CNT * 3 + 1 +
674 2 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
678 (((loff_t) block) << inode->i_sb->s_blocksize_bits) + 1;
680 reiserfs_write_lock(inode->i_sb);
681 version = get_inode_item_key_version(inode);
683 if (!file_capable(inode, block)) {
684 reiserfs_write_unlock(inode->i_sb);
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
692 if (!(create & GET_BLOCK_CREATE)) {
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);
702 * if we're already in a transaction, make sure to close
703 * any new transactions we start in this func
705 if ((create & GET_BLOCK_NO_DANGLE) ||
706 reiserfs_transaction_running(inode->i_sb))
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
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;
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) {
724 th = reiserfs_persistent_transaction(inode->i_sb, jbegin_count);
729 reiserfs_update_inode_transaction(inode);
733 retval = search_for_position_by_key(inode->i_sb, &key, &path);
734 if (retval == IO_ERROR) {
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;
744 fs_gen = get_generation(inode->i_sb);
745 copy_item_head(&tmp_ih, ih);
747 if (allocation_needed
748 (retval, allocated_block_nr, ih, item, pos_in_item)) {
749 /* we have to allocate block for the unformatted node */
756 _allocate_block(th, block, inode, &allocated_block_nr,
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
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);
770 _allocate_block(th, block, inode,
771 &allocated_block_nr, NULL, create);
773 if (repeat != NO_DISK_SPACE && repeat != QUOTA_EXCEEDED) {
776 if (repeat == QUOTA_EXCEEDED)
783 if (fs_changed(fs_gen, inode->i_sb)
784 && item_moved(&tmp_ih, &path)) {
789 if (indirect_item_found(retval, ih)) {
790 b_blocknr_t unfm_ptr;
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)
796 unfm_ptr = get_block_num(item, pos_in_item);
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,
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);
815 set_block_dev_mapped(bh_result, unfm_ptr, inode);
819 retval = reiserfs_end_persistent_transaction(th);
821 reiserfs_write_unlock(inode->i_sb);
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
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
843 if (is_statdata_le_ih(ih)) {
845 struct cpu_key tmp_key;
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 */ );
853 * we are going to add 'block'-th block to the file.
854 * Use allocated block for that
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);
863 tmp_key = key; /* ;) */
864 set_cpu_key_k_offset(&tmp_key, 1);
865 PATH_LAST_POSITION(&path)++;
868 reiserfs_insert_item(th, &path, &tmp_key, &tmp_ih,
869 inode, (char *)&unp);
871 reiserfs_free_block(th, inode,
872 allocated_block_nr, 1);
874 * retval == -ENOSPC, -EDQUOT or -EIO
879 } else if (is_direct_le_ih(ih)) {
880 /* direct item has to be converted */
884 ((le_ih_k_offset(ih) -
885 1) & ~(inode->i_sb->s_blocksize - 1)) + 1;
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
892 if (tail_offset == cpu_key_k_offset(&key)) {
893 set_block_dev_mapped(bh_result,
894 allocated_block_nr, inode);
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
907 * ugly, but we can only end the transaction if
910 BUG_ON(!th->t_refcount);
911 if (th->t_refcount == 1) {
913 reiserfs_end_persistent_transaction
921 convert_tail_for_hole(inode, bh_result,
924 if (retval != -ENOSPC)
925 reiserfs_error(inode->i_sb,
927 "convert tail failed "
928 "inode %lu, error %d",
931 if (allocated_block_nr) {
933 * the bitmap, the super,
934 * and the stat data == 3
937 th = reiserfs_persistent_transaction(inode->i_sb, 3);
939 reiserfs_free_block(th,
949 direct2indirect(th, inode, &path, unbh,
952 reiserfs_unmap_buffer(unbh);
953 reiserfs_free_block(th, inode,
954 allocated_block_nr, 1);
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
967 set_buffer_uptodate(unbh);
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
976 * we've converted the tail, so we must
977 * flush unbh before the transaction commits
979 reiserfs_add_tail_list(inode, unbh);
982 * mark it dirty now to prevent commit_write
983 * from adding this buffer to the inode's
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())
994 mark_buffer_dirty(unbh);
998 * append indirect item with holes if needed, when
999 * appending pointer to 'block'-th block use block,
1000 * which is already allocated
1002 struct cpu_key tmp_key;
1004 * We use this in case we need to allocate
1005 * only one block which is a fastpath
1007 unp_t unf_single = 0;
1009 __u64 max_to_insert =
1010 MAX_ITEM_LEN(inode->i_sb->s_blocksize) /
1012 __u64 blocks_needed;
1014 RFALSE(pos_in_item != ih_item_len(ih) / UNFM_P_SIZE,
1015 "vs-804: invalid position for append");
1017 * indirect item has to be appended,
1018 * set up key of that position
1019 * (key type is unimportant)
1021 make_cpu_key(&tmp_key, inode,
1022 le_key_k_offset(version,
1025 inode->i_sb->s_blocksize),
1028 RFALSE(cpu_key_k_offset(&tmp_key) > cpu_key_k_offset(&key),
1029 "green-805: invalid offset");
1032 ((cpu_key_k_offset(&key) -
1033 cpu_key_k_offset(&tmp_key)) >> inode->i_sb->
1036 if (blocks_needed == 1) {
1039 un = kcalloc(min(blocks_needed, max_to_insert),
1040 UNFM_P_SIZE, GFP_NOFS);
1047 if (blocks_needed <= max_to_insert) {
1049 * we are going to add target block to
1050 * the file. Use allocated block for that
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);
1059 /* paste hole to the indirect item */
1061 * If kcalloc failed, max_to_insert becomes
1062 * zero and it means we only have space for
1066 max_to_insert ? max_to_insert : 1;
1069 reiserfs_paste_into_item(th, &path, &tmp_key, inode,
1074 if (blocks_needed != 1)
1078 reiserfs_free_block(th, inode,
1079 allocated_block_nr, 1);
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
1090 inode->i_sb->s_blocksize * blocks_needed;
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.
1105 if (journal_transaction_should_end(th, th->t_blocks_allocated)) {
1106 retval = restart_transaction(th, inode, &path);
1111 * inserting indirect pointers for a hole can take a
1112 * long time. reschedule if needed and also release the write
1115 reiserfs_cond_resched(inode->i_sb);
1117 retval = search_for_position_by_key(inode->i_sb, &key, &path);
1118 if (retval == IO_ERROR) {
1122 if (retval == POSITION_FOUND) {
1123 reiserfs_warning(inode->i_sb, "vs-825",
1124 "%K should not be found", &key);
1126 if (allocated_block_nr)
1127 reiserfs_free_block(th, inode,
1128 allocated_block_nr, 1);
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;
1141 if (th && (!dangle || (retval && !th->t_trans_id))) {
1144 reiserfs_update_sd(th, inode);
1145 err = reiserfs_end_persistent_transaction(th);
1150 reiserfs_write_unlock(inode->i_sb);
1151 reiserfs_check_path(&path);
1155 static void reiserfs_readahead(struct readahead_control *rac)
1157 mpage_readahead(rac, reiserfs_get_block);
1161 * Compute real number of used bytes by file
1162 * Following three functions can go away when we'll have enough space in
1165 static int real_space_diff(struct inode *inode, int sd_size)
1168 loff_t blocksize = inode->i_sb->s_blocksize;
1170 if (S_ISLNK(inode->i_mode) || S_ISDIR(inode->i_mode))
1174 * End of file is also in full block with indirect reference, so round
1175 * up to the next block.
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
1184 (blocksize - 1)) >> inode->i_sb->s_blocksize_bits) * UNFM_P_SIZE +
1189 static inline loff_t to_real_used_space(struct inode *inode, ulong blocks,
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));
1196 return ((loff_t) real_space_diff(inode, sd_size)) +
1197 (((loff_t) blocks) << 9);
1200 /* Compute number of blocks used by file in ReiserFS counting */
1201 static inline ulong to_fake_used_blocks(struct inode *inode, int sd_size)
1203 loff_t bytes = inode_get_bytes(inode);
1204 loff_t real_space = real_space_diff(inode, sd_size);
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;
1212 * files from before the quota patch might i_blocks such that
1213 * bytes < real_space. Deal with that here to prevent it from
1216 if (bytes < real_space)
1218 return (bytes - real_space) >> 9;
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
1228 /* called by read_locked_inode */
1229 static void init_inode(struct inode *inode, struct treepath *path)
1231 struct buffer_head *bh;
1232 struct item_head *ih;
1235 bh = PATH_PLAST_BUFFER(path);
1236 ih = tp_item_head(path);
1238 copy_key(INODE_PKEY(inode), &ih->ih_key);
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);
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;
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->i_atime.tv_sec = sd_v1_atime(sd);
1261 inode->i_mtime.tv_sec = sd_v1_mtime(sd);
1262 inode->i_ctime.tv_sec = sd_v1_ctime(sd);
1263 inode->i_atime.tv_nsec = 0;
1264 inode->i_ctime.tv_nsec = 0;
1265 inode->i_mtime.tv_nsec = 0;
1267 inode->i_blocks = sd_v1_blocks(sd);
1268 inode->i_generation = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1269 blocks = (inode->i_size + 511) >> 9;
1270 blocks = _ROUND_UP(blocks, inode->i_sb->s_blocksize >> 9);
1273 * there was a bug in <=3.5.23 when i_blocks could take
1274 * negative values. Starting from 3.5.17 this value could
1275 * even be stored in stat data. For such files we set
1276 * i_blocks based on file size. Just 2 notes: this can be
1277 * wrong for sparse files. On-disk value will be only
1278 * updated if file's inode will ever change
1280 if (inode->i_blocks > blocks) {
1281 inode->i_blocks = blocks;
1284 rdev = sd_v1_rdev(sd);
1285 REISERFS_I(inode)->i_first_direct_byte =
1286 sd_v1_first_direct_byte(sd);
1289 * an early bug in the quota code can give us an odd
1290 * number for the block count. This is incorrect, fix it here.
1292 if (inode->i_blocks & 1) {
1295 inode_set_bytes(inode,
1296 to_real_used_space(inode, inode->i_blocks,
1299 * nopack is initially zero for v1 objects. For v2 objects,
1300 * nopack is initialised from sd_attrs
1302 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
1305 * new stat data found, but object may have old items
1306 * (directories and symlinks)
1308 struct stat_data *sd = (struct stat_data *)ih_item_body(bh, ih);
1310 inode->i_mode = sd_v2_mode(sd);
1311 set_nlink(inode, sd_v2_nlink(sd));
1312 i_uid_write(inode, sd_v2_uid(sd));
1313 inode->i_size = sd_v2_size(sd);
1314 i_gid_write(inode, sd_v2_gid(sd));
1315 inode->i_mtime.tv_sec = sd_v2_mtime(sd);
1316 inode->i_atime.tv_sec = sd_v2_atime(sd);
1317 inode->i_ctime.tv_sec = sd_v2_ctime(sd);
1318 inode->i_ctime.tv_nsec = 0;
1319 inode->i_mtime.tv_nsec = 0;
1320 inode->i_atime.tv_nsec = 0;
1321 inode->i_blocks = sd_v2_blocks(sd);
1322 rdev = sd_v2_rdev(sd);
1323 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1324 inode->i_generation =
1325 le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1327 inode->i_generation = sd_v2_generation(sd);
1329 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
1330 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
1332 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
1333 REISERFS_I(inode)->i_first_direct_byte = 0;
1334 set_inode_sd_version(inode, STAT_DATA_V2);
1335 inode_set_bytes(inode,
1336 to_real_used_space(inode, inode->i_blocks,
1339 * read persistent inode attributes from sd and initialise
1340 * generic inode flags from them
1342 REISERFS_I(inode)->i_attrs = sd_v2_attrs(sd);
1343 sd_attrs_to_i_attrs(sd_v2_attrs(sd), inode);
1347 if (S_ISREG(inode->i_mode)) {
1348 inode->i_op = &reiserfs_file_inode_operations;
1349 inode->i_fop = &reiserfs_file_operations;
1350 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1351 } else if (S_ISDIR(inode->i_mode)) {
1352 inode->i_op = &reiserfs_dir_inode_operations;
1353 inode->i_fop = &reiserfs_dir_operations;
1354 } else if (S_ISLNK(inode->i_mode)) {
1355 inode->i_op = &reiserfs_symlink_inode_operations;
1356 inode_nohighmem(inode);
1357 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1359 inode->i_blocks = 0;
1360 inode->i_op = &reiserfs_special_inode_operations;
1361 init_special_inode(inode, inode->i_mode, new_decode_dev(rdev));
1365 /* update new stat data with inode fields */
1366 static void inode2sd(void *sd, struct inode *inode, loff_t size)
1368 struct stat_data *sd_v2 = (struct stat_data *)sd;
1370 set_sd_v2_mode(sd_v2, inode->i_mode);
1371 set_sd_v2_nlink(sd_v2, inode->i_nlink);
1372 set_sd_v2_uid(sd_v2, i_uid_read(inode));
1373 set_sd_v2_size(sd_v2, size);
1374 set_sd_v2_gid(sd_v2, i_gid_read(inode));
1375 set_sd_v2_mtime(sd_v2, inode->i_mtime.tv_sec);
1376 set_sd_v2_atime(sd_v2, inode->i_atime.tv_sec);
1377 set_sd_v2_ctime(sd_v2, inode->i_ctime.tv_sec);
1378 set_sd_v2_blocks(sd_v2, to_fake_used_blocks(inode, SD_V2_SIZE));
1379 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1380 set_sd_v2_rdev(sd_v2, new_encode_dev(inode->i_rdev));
1382 set_sd_v2_generation(sd_v2, inode->i_generation);
1383 set_sd_v2_attrs(sd_v2, REISERFS_I(inode)->i_attrs);
1386 /* used to copy inode's fields to old stat data */
1387 static void inode2sd_v1(void *sd, struct inode *inode, loff_t size)
1389 struct stat_data_v1 *sd_v1 = (struct stat_data_v1 *)sd;
1391 set_sd_v1_mode(sd_v1, inode->i_mode);
1392 set_sd_v1_uid(sd_v1, i_uid_read(inode));
1393 set_sd_v1_gid(sd_v1, i_gid_read(inode));
1394 set_sd_v1_nlink(sd_v1, inode->i_nlink);
1395 set_sd_v1_size(sd_v1, size);
1396 set_sd_v1_atime(sd_v1, inode->i_atime.tv_sec);
1397 set_sd_v1_ctime(sd_v1, inode->i_ctime.tv_sec);
1398 set_sd_v1_mtime(sd_v1, inode->i_mtime.tv_sec);
1400 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
1401 set_sd_v1_rdev(sd_v1, new_encode_dev(inode->i_rdev));
1403 set_sd_v1_blocks(sd_v1, to_fake_used_blocks(inode, SD_V1_SIZE));
1405 /* Sigh. i_first_direct_byte is back */
1406 set_sd_v1_first_direct_byte(sd_v1,
1407 REISERFS_I(inode)->i_first_direct_byte);
1411 * NOTE, you must prepare the buffer head before sending it here,
1412 * and then log it after the call
1414 static void update_stat_data(struct treepath *path, struct inode *inode,
1417 struct buffer_head *bh;
1418 struct item_head *ih;
1420 bh = PATH_PLAST_BUFFER(path);
1421 ih = tp_item_head(path);
1423 if (!is_statdata_le_ih(ih))
1424 reiserfs_panic(inode->i_sb, "vs-13065", "key %k, found item %h",
1425 INODE_PKEY(inode), ih);
1427 /* path points to old stat data */
1428 if (stat_data_v1(ih)) {
1429 inode2sd_v1(ih_item_body(bh, ih), inode, size);
1431 inode2sd(ih_item_body(bh, ih), inode, size);
1437 void reiserfs_update_sd_size(struct reiserfs_transaction_handle *th,
1438 struct inode *inode, loff_t size)
1441 INITIALIZE_PATH(path);
1442 struct buffer_head *bh;
1444 struct item_head *ih, tmp_ih;
1447 BUG_ON(!th->t_trans_id);
1449 /* key type is unimportant */
1450 make_cpu_key(&key, inode, SD_OFFSET, TYPE_STAT_DATA, 3);
1454 /* look for the object's stat data */
1455 retval = search_item(inode->i_sb, &key, &path);
1456 if (retval == IO_ERROR) {
1457 reiserfs_error(inode->i_sb, "vs-13050",
1458 "i/o failure occurred trying to "
1459 "update %K stat data", &key);
1462 if (retval == ITEM_NOT_FOUND) {
1463 pos = PATH_LAST_POSITION(&path);
1465 if (inode->i_nlink == 0) {
1466 /*reiserfs_warning (inode->i_sb, "vs-13050: reiserfs_update_sd: i_nlink == 0, stat data not found"); */
1469 reiserfs_warning(inode->i_sb, "vs-13060",
1470 "stat data of object %k (nlink == %d) "
1471 "not found (pos %d)",
1472 INODE_PKEY(inode), inode->i_nlink,
1474 reiserfs_check_path(&path);
1479 * sigh, prepare_for_journal might schedule. When it
1480 * schedules the FS might change. We have to detect that,
1481 * and loop back to the search if the stat data item has moved
1483 bh = get_last_bh(&path);
1484 ih = tp_item_head(&path);
1485 copy_item_head(&tmp_ih, ih);
1486 fs_gen = get_generation(inode->i_sb);
1487 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
1489 /* Stat_data item has been moved after scheduling. */
1490 if (fs_changed(fs_gen, inode->i_sb)
1491 && item_moved(&tmp_ih, &path)) {
1492 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
1497 update_stat_data(&path, inode, size);
1498 journal_mark_dirty(th, bh);
1504 * reiserfs_read_locked_inode is called to read the inode off disk, and it
1505 * does a make_bad_inode when things go wrong. But, we need to make sure
1506 * and clear the key in the private portion of the inode, otherwise a
1507 * corresponding iput might try to delete whatever object the inode last
1510 static void reiserfs_make_bad_inode(struct inode *inode)
1512 memset(INODE_PKEY(inode), 0, KEY_SIZE);
1513 make_bad_inode(inode);
1517 * initially this function was derived from minix or ext2's analog and
1518 * evolved as the prototype did
1520 int reiserfs_init_locked_inode(struct inode *inode, void *p)
1522 struct reiserfs_iget_args *args = (struct reiserfs_iget_args *)p;
1523 inode->i_ino = args->objectid;
1524 INODE_PKEY(inode)->k_dir_id = cpu_to_le32(args->dirid);
1529 * looks for stat data in the tree, and fills up the fields of in-core
1530 * inode stat data fields
1532 void reiserfs_read_locked_inode(struct inode *inode,
1533 struct reiserfs_iget_args *args)
1535 INITIALIZE_PATH(path_to_sd);
1537 unsigned long dirino;
1540 dirino = args->dirid;
1543 * set version 1, version 2 could be used too, because stat data
1544 * key is the same in both versions
1546 _make_cpu_key(&key, KEY_FORMAT_3_5, dirino, inode->i_ino, 0, 0, 3);
1548 /* look for the object's stat data */
1549 retval = search_item(inode->i_sb, &key, &path_to_sd);
1550 if (retval == IO_ERROR) {
1551 reiserfs_error(inode->i_sb, "vs-13070",
1552 "i/o failure occurred trying to find "
1553 "stat data of %K", &key);
1554 reiserfs_make_bad_inode(inode);
1558 /* a stale NFS handle can trigger this without it being an error */
1559 if (retval != ITEM_FOUND) {
1560 pathrelse(&path_to_sd);
1561 reiserfs_make_bad_inode(inode);
1566 init_inode(inode, &path_to_sd);
1569 * It is possible that knfsd is trying to access inode of a file
1570 * that is being removed from the disk by some other thread. As we
1571 * update sd on unlink all that is required is to check for nlink
1572 * here. This bug was first found by Sizif when debugging
1573 * SquidNG/Butterfly, forgotten, and found again after Philippe
1574 * Gramoulle <philippe.gramoulle@mmania.com> reproduced it.
1576 * More logical fix would require changes in fs/inode.c:iput() to
1577 * remove inode from hash-table _after_ fs cleaned disk stuff up and
1578 * in iget() to return NULL if I_FREEING inode is found in
1583 * Currently there is one place where it's ok to meet inode with
1584 * nlink==0: processing of open-unlinked and half-truncated files
1585 * during mount (fs/reiserfs/super.c:finish_unfinished()).
1587 if ((inode->i_nlink == 0) &&
1588 !REISERFS_SB(inode->i_sb)->s_is_unlinked_ok) {
1589 reiserfs_warning(inode->i_sb, "vs-13075",
1590 "dead inode read from disk %K. "
1591 "This is likely to be race with knfsd. Ignore",
1593 reiserfs_make_bad_inode(inode);
1596 /* init inode should be relsing */
1597 reiserfs_check_path(&path_to_sd);
1600 * Stat data v1 doesn't support ACLs.
1602 if (get_inode_sd_version(inode) == STAT_DATA_V1)
1603 cache_no_acl(inode);
1607 * reiserfs_find_actor() - "find actor" reiserfs supplies to iget5_locked().
1609 * @inode: inode from hash table to check
1610 * @opaque: "cookie" passed to iget5_locked(). This is &reiserfs_iget_args.
1612 * This function is called by iget5_locked() to distinguish reiserfs inodes
1613 * having the same inode numbers. Such inodes can only exist due to some
1614 * error condition. One of them should be bad. Inodes with identical
1615 * inode numbers (objectids) are distinguished by parent directory ids.
1618 int reiserfs_find_actor(struct inode *inode, void *opaque)
1620 struct reiserfs_iget_args *args;
1623 /* args is already in CPU order */
1624 return (inode->i_ino == args->objectid) &&
1625 (le32_to_cpu(INODE_PKEY(inode)->k_dir_id) == args->dirid);
1628 struct inode *reiserfs_iget(struct super_block *s, const struct cpu_key *key)
1630 struct inode *inode;
1631 struct reiserfs_iget_args args;
1634 args.objectid = key->on_disk_key.k_objectid;
1635 args.dirid = key->on_disk_key.k_dir_id;
1636 depth = reiserfs_write_unlock_nested(s);
1637 inode = iget5_locked(s, key->on_disk_key.k_objectid,
1638 reiserfs_find_actor, reiserfs_init_locked_inode,
1640 reiserfs_write_lock_nested(s, depth);
1642 return ERR_PTR(-ENOMEM);
1644 if (inode->i_state & I_NEW) {
1645 reiserfs_read_locked_inode(inode, &args);
1646 unlock_new_inode(inode);
1649 if (comp_short_keys(INODE_PKEY(inode), key) || is_bad_inode(inode)) {
1650 /* either due to i/o error or a stale NFS handle */
1657 static struct dentry *reiserfs_get_dentry(struct super_block *sb,
1658 u32 objectid, u32 dir_id, u32 generation)
1662 struct inode *inode;
1664 key.on_disk_key.k_objectid = objectid;
1665 key.on_disk_key.k_dir_id = dir_id;
1666 reiserfs_write_lock(sb);
1667 inode = reiserfs_iget(sb, &key);
1668 if (inode && !IS_ERR(inode) && generation != 0 &&
1669 generation != inode->i_generation) {
1673 reiserfs_write_unlock(sb);
1675 return d_obtain_alias(inode);
1678 struct dentry *reiserfs_fh_to_dentry(struct super_block *sb, struct fid *fid,
1679 int fh_len, int fh_type)
1682 * fhtype happens to reflect the number of u32s encoded.
1683 * due to a bug in earlier code, fhtype might indicate there
1684 * are more u32s then actually fitted.
1685 * so if fhtype seems to be more than len, reduce fhtype.
1687 * 2 - objectid + dir_id - legacy support
1688 * 3 - objectid + dir_id + generation
1689 * 4 - objectid + dir_id + objectid and dirid of parent - legacy
1690 * 5 - objectid + dir_id + generation + objectid and dirid of parent
1691 * 6 - as above plus generation of directory
1692 * 6 does not fit in NFSv2 handles
1694 if (fh_type > fh_len) {
1695 if (fh_type != 6 || fh_len != 5)
1696 reiserfs_warning(sb, "reiserfs-13077",
1697 "nfsd/reiserfs, fhtype=%d, len=%d - odd",
1704 return reiserfs_get_dentry(sb, fid->raw[0], fid->raw[1],
1705 (fh_type == 3 || fh_type >= 5) ? fid->raw[2] : 0);
1708 struct dentry *reiserfs_fh_to_parent(struct super_block *sb, struct fid *fid,
1709 int fh_len, int fh_type)
1711 if (fh_type > fh_len)
1716 return reiserfs_get_dentry(sb,
1717 (fh_type >= 5) ? fid->raw[3] : fid->raw[2],
1718 (fh_type >= 5) ? fid->raw[4] : fid->raw[3],
1719 (fh_type == 6) ? fid->raw[5] : 0);
1722 int reiserfs_encode_fh(struct inode *inode, __u32 * data, int *lenp,
1723 struct inode *parent)
1727 if (parent && (maxlen < 5)) {
1729 return FILEID_INVALID;
1730 } else if (maxlen < 3) {
1732 return FILEID_INVALID;
1735 data[0] = inode->i_ino;
1736 data[1] = le32_to_cpu(INODE_PKEY(inode)->k_dir_id);
1737 data[2] = inode->i_generation;
1740 data[3] = parent->i_ino;
1741 data[4] = le32_to_cpu(INODE_PKEY(parent)->k_dir_id);
1744 data[5] = parent->i_generation;
1752 * looks for stat data, then copies fields to it, marks the buffer
1753 * containing stat data as dirty
1756 * reiserfs inodes are never really dirty, since the dirty inode call
1757 * always logs them. This call allows the VFS inode marking routines
1758 * to properly mark inodes for datasync and such, but only actually
1759 * does something when called for a synchronous update.
1761 int reiserfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1763 struct reiserfs_transaction_handle th;
1764 int jbegin_count = 1;
1766 if (sb_rdonly(inode->i_sb))
1769 * memory pressure can sometimes initiate write_inode calls with
1771 * these cases are just when the system needs ram, not when the
1772 * inode needs to reach disk for safety, and they can safely be
1773 * ignored because the altered inode has already been logged.
1775 if (wbc->sync_mode == WB_SYNC_ALL && !(current->flags & PF_MEMALLOC)) {
1776 reiserfs_write_lock(inode->i_sb);
1777 if (!journal_begin(&th, inode->i_sb, jbegin_count)) {
1778 reiserfs_update_sd(&th, inode);
1779 journal_end_sync(&th);
1781 reiserfs_write_unlock(inode->i_sb);
1787 * stat data of new object is inserted already, this inserts the item
1788 * containing "." and ".." entries
1790 static int reiserfs_new_directory(struct reiserfs_transaction_handle *th,
1791 struct inode *inode,
1792 struct item_head *ih, struct treepath *path,
1795 struct super_block *sb = th->t_super;
1796 char empty_dir[EMPTY_DIR_SIZE];
1797 char *body = empty_dir;
1801 BUG_ON(!th->t_trans_id);
1803 _make_cpu_key(&key, KEY_FORMAT_3_5, le32_to_cpu(ih->ih_key.k_dir_id),
1804 le32_to_cpu(ih->ih_key.k_objectid), DOT_OFFSET,
1805 TYPE_DIRENTRY, 3 /*key length */ );
1808 * compose item head for new item. Directories consist of items of
1809 * old type (ITEM_VERSION_1). Do not set key (second arg is 0), it
1810 * is done by reiserfs_new_inode
1812 if (old_format_only(sb)) {
1813 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1814 TYPE_DIRENTRY, EMPTY_DIR_SIZE_V1, 2);
1816 make_empty_dir_item_v1(body, ih->ih_key.k_dir_id,
1817 ih->ih_key.k_objectid,
1818 INODE_PKEY(dir)->k_dir_id,
1819 INODE_PKEY(dir)->k_objectid);
1821 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, DOT_OFFSET,
1822 TYPE_DIRENTRY, EMPTY_DIR_SIZE, 2);
1824 make_empty_dir_item(body, ih->ih_key.k_dir_id,
1825 ih->ih_key.k_objectid,
1826 INODE_PKEY(dir)->k_dir_id,
1827 INODE_PKEY(dir)->k_objectid);
1830 /* look for place in the tree for new item */
1831 retval = search_item(sb, &key, path);
1832 if (retval == IO_ERROR) {
1833 reiserfs_error(sb, "vs-13080",
1834 "i/o failure occurred creating new directory");
1837 if (retval == ITEM_FOUND) {
1839 reiserfs_warning(sb, "vs-13070",
1840 "object with this key exists (%k)",
1845 /* insert item, that is empty directory item */
1846 return reiserfs_insert_item(th, path, &key, ih, inode, body);
1850 * stat data of object has been inserted, this inserts the item
1851 * containing the body of symlink
1853 static int reiserfs_new_symlink(struct reiserfs_transaction_handle *th,
1854 struct inode *inode,
1855 struct item_head *ih,
1856 struct treepath *path, const char *symname,
1859 struct super_block *sb = th->t_super;
1863 BUG_ON(!th->t_trans_id);
1865 _make_cpu_key(&key, KEY_FORMAT_3_5,
1866 le32_to_cpu(ih->ih_key.k_dir_id),
1867 le32_to_cpu(ih->ih_key.k_objectid),
1868 1, TYPE_DIRECT, 3 /*key length */ );
1870 make_le_item_head(ih, NULL, KEY_FORMAT_3_5, 1, TYPE_DIRECT, item_len,
1871 0 /*free_space */ );
1873 /* look for place in the tree for new item */
1874 retval = search_item(sb, &key, path);
1875 if (retval == IO_ERROR) {
1876 reiserfs_error(sb, "vs-13080",
1877 "i/o failure occurred creating new symlink");
1880 if (retval == ITEM_FOUND) {
1882 reiserfs_warning(sb, "vs-13080",
1883 "object with this key exists (%k)",
1888 /* insert item, that is body of symlink */
1889 return reiserfs_insert_item(th, path, &key, ih, inode, symname);
1893 * inserts the stat data into the tree, and then calls
1894 * reiserfs_new_directory (to insert ".", ".." item if new object is
1895 * directory) or reiserfs_new_symlink (to insert symlink body if new
1896 * object is symlink) or nothing (if new object is regular file)
1898 * NOTE! uid and gid must already be set in the inode. If we return
1899 * non-zero due to an error, we have to drop the quota previously allocated
1900 * for the fresh inode. This can only be done outside a transaction, so
1901 * if we return non-zero, we also end the transaction.
1903 * @th: active transaction handle
1904 * @dir: parent directory for new inode
1905 * @mode: mode of new inode
1906 * @symname: symlink contents if inode is symlink
1907 * @isize: 0 for regular file, EMPTY_DIR_SIZE for dirs, strlen(symname) for
1909 * @inode: inode to be filled
1910 * @security: optional security context to associate with this inode
1912 int reiserfs_new_inode(struct reiserfs_transaction_handle *th,
1913 struct inode *dir, umode_t mode, const char *symname,
1914 /* 0 for regular, EMTRY_DIR_SIZE for dirs,
1915 strlen (symname) for symlinks) */
1916 loff_t i_size, struct dentry *dentry,
1917 struct inode *inode,
1918 struct reiserfs_security_handle *security)
1920 struct super_block *sb = dir->i_sb;
1921 struct reiserfs_iget_args args;
1922 INITIALIZE_PATH(path_to_key);
1924 struct item_head ih;
1925 struct stat_data sd;
1930 BUG_ON(!th->t_trans_id);
1932 depth = reiserfs_write_unlock_nested(sb);
1933 err = dquot_alloc_inode(inode);
1934 reiserfs_write_lock_nested(sb, depth);
1937 if (!dir->i_nlink) {
1942 /* item head of new item */
1943 ih.ih_key.k_dir_id = reiserfs_choose_packing(dir);
1944 ih.ih_key.k_objectid = cpu_to_le32(reiserfs_get_unused_objectid(th));
1945 if (!ih.ih_key.k_objectid) {
1949 args.objectid = inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
1950 if (old_format_only(sb))
1951 make_le_item_head(&ih, NULL, KEY_FORMAT_3_5, SD_OFFSET,
1952 TYPE_STAT_DATA, SD_V1_SIZE, MAX_US_INT);
1954 make_le_item_head(&ih, NULL, KEY_FORMAT_3_6, SD_OFFSET,
1955 TYPE_STAT_DATA, SD_SIZE, MAX_US_INT);
1956 memcpy(INODE_PKEY(inode), &ih.ih_key, KEY_SIZE);
1957 args.dirid = le32_to_cpu(ih.ih_key.k_dir_id);
1959 depth = reiserfs_write_unlock_nested(inode->i_sb);
1960 err = insert_inode_locked4(inode, args.objectid,
1961 reiserfs_find_actor, &args);
1962 reiserfs_write_lock_nested(inode->i_sb, depth);
1968 if (old_format_only(sb))
1970 * not a perfect generation count, as object ids can be reused,
1971 * but this is as good as reiserfs can do right now.
1972 * note that the private part of inode isn't filled in yet,
1973 * we have to use the directory.
1975 inode->i_generation = le32_to_cpu(INODE_PKEY(dir)->k_objectid);
1977 #if defined( USE_INODE_GENERATION_COUNTER )
1978 inode->i_generation =
1979 le32_to_cpu(REISERFS_SB(sb)->s_rs->s_inode_generation);
1981 inode->i_generation = ++event;
1984 /* fill stat data */
1985 set_nlink(inode, (S_ISDIR(mode) ? 2 : 1));
1987 /* uid and gid must already be set by the caller for quota init */
1989 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1990 inode->i_size = i_size;
1991 inode->i_blocks = 0;
1993 REISERFS_I(inode)->i_first_direct_byte = S_ISLNK(mode) ? 1 :
1994 U32_MAX /*NO_BYTES_IN_DIRECT_ITEM */ ;
1996 INIT_LIST_HEAD(&REISERFS_I(inode)->i_prealloc_list);
1997 REISERFS_I(inode)->i_flags = 0;
1998 REISERFS_I(inode)->i_prealloc_block = 0;
1999 REISERFS_I(inode)->i_prealloc_count = 0;
2000 REISERFS_I(inode)->i_trans_id = 0;
2001 REISERFS_I(inode)->i_jl = NULL;
2002 REISERFS_I(inode)->i_attrs =
2003 REISERFS_I(dir)->i_attrs & REISERFS_INHERIT_MASK;
2004 sd_attrs_to_i_attrs(REISERFS_I(inode)->i_attrs, inode);
2005 reiserfs_init_xattr_rwsem(inode);
2007 /* key to search for correct place for new stat data */
2008 _make_cpu_key(&key, KEY_FORMAT_3_6, le32_to_cpu(ih.ih_key.k_dir_id),
2009 le32_to_cpu(ih.ih_key.k_objectid), SD_OFFSET,
2010 TYPE_STAT_DATA, 3 /*key length */ );
2012 /* find proper place for inserting of stat data */
2013 retval = search_item(sb, &key, &path_to_key);
2014 if (retval == IO_ERROR) {
2018 if (retval == ITEM_FOUND) {
2019 pathrelse(&path_to_key);
2023 if (old_format_only(sb)) {
2024 /* i_uid or i_gid is too big to be stored in stat data v3.5 */
2025 if (i_uid_read(inode) & ~0xffff || i_gid_read(inode) & ~0xffff) {
2026 pathrelse(&path_to_key);
2030 inode2sd_v1(&sd, inode, inode->i_size);
2032 inode2sd(&sd, inode, inode->i_size);
2035 * store in in-core inode the key of stat data and version all
2036 * object items will have (directory items will have old offset
2037 * format, other new objects will consist of new items)
2039 if (old_format_only(sb) || S_ISDIR(mode) || S_ISLNK(mode))
2040 set_inode_item_key_version(inode, KEY_FORMAT_3_5);
2042 set_inode_item_key_version(inode, KEY_FORMAT_3_6);
2043 if (old_format_only(sb))
2044 set_inode_sd_version(inode, STAT_DATA_V1);
2046 set_inode_sd_version(inode, STAT_DATA_V2);
2048 /* insert the stat data into the tree */
2049 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2050 if (REISERFS_I(dir)->new_packing_locality)
2051 th->displace_new_blocks = 1;
2054 reiserfs_insert_item(th, &path_to_key, &key, &ih, inode,
2058 reiserfs_check_path(&path_to_key);
2061 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
2062 if (!th->displace_new_blocks)
2063 REISERFS_I(dir)->new_packing_locality = 0;
2065 if (S_ISDIR(mode)) {
2066 /* insert item with "." and ".." */
2068 reiserfs_new_directory(th, inode, &ih, &path_to_key, dir);
2071 if (S_ISLNK(mode)) {
2072 /* insert body of symlink */
2073 if (!old_format_only(sb))
2074 i_size = ROUND_UP(i_size);
2076 reiserfs_new_symlink(th, inode, &ih, &path_to_key, symname,
2081 reiserfs_check_path(&path_to_key);
2083 goto out_inserted_sd;
2087 * Mark it private if we're creating the privroot
2088 * or something under it.
2090 if (IS_PRIVATE(dir) || dentry == REISERFS_SB(sb)->priv_root)
2091 reiserfs_init_priv_inode(inode);
2093 if (reiserfs_posixacl(inode->i_sb)) {
2094 reiserfs_write_unlock(inode->i_sb);
2095 retval = reiserfs_inherit_default_acl(th, dir, dentry, inode);
2096 reiserfs_write_lock(inode->i_sb);
2099 reiserfs_check_path(&path_to_key);
2101 goto out_inserted_sd;
2103 } else if (inode->i_sb->s_flags & SB_POSIXACL) {
2104 reiserfs_warning(inode->i_sb, "jdm-13090",
2105 "ACLs aren't enabled in the fs, "
2106 "but vfs thinks they are!");
2109 if (security->name) {
2110 reiserfs_write_unlock(inode->i_sb);
2111 retval = reiserfs_security_write(th, inode, security);
2112 reiserfs_write_lock(inode->i_sb);
2115 reiserfs_check_path(&path_to_key);
2116 retval = journal_end(th);
2119 goto out_inserted_sd;
2123 reiserfs_update_sd(th, inode);
2124 reiserfs_check_path(&path_to_key);
2129 /* Invalidate the object, nothing was inserted yet */
2130 INODE_PKEY(inode)->k_objectid = 0;
2132 /* Quota change must be inside a transaction for journaling */
2133 depth = reiserfs_write_unlock_nested(inode->i_sb);
2134 dquot_free_inode(inode);
2135 reiserfs_write_lock_nested(inode->i_sb, depth);
2140 * Drop can be outside and it needs more credits so it's better
2141 * to have it outside
2143 depth = reiserfs_write_unlock_nested(inode->i_sb);
2145 reiserfs_write_lock_nested(inode->i_sb, depth);
2146 inode->i_flags |= S_NOQUOTA;
2147 make_bad_inode(inode);
2151 th->t_trans_id = 0; /* so the caller can't use this handle later */
2152 if (inode->i_state & I_NEW)
2153 unlock_new_inode(inode);
2159 * finds the tail page in the page cache,
2160 * reads the last block in.
2162 * On success, page_result is set to a locked, pinned page, and bh_result
2163 * is set to an up to date buffer for the last block in the file. returns 0.
2165 * tail conversion is not done, so bh_result might not be valid for writing
2166 * check buffer_mapped(bh_result) and bh_result->b_blocknr != 0 before
2167 * trying to write the block.
2169 * on failure, nonzero is returned, page_result and bh_result are untouched.
2171 static int grab_tail_page(struct inode *inode,
2172 struct page **page_result,
2173 struct buffer_head **bh_result)
2177 * we want the page with the last byte in the file,
2178 * not the page that will hold the next byte for appending
2180 unsigned long index = (inode->i_size - 1) >> PAGE_SHIFT;
2181 unsigned long pos = 0;
2182 unsigned long start = 0;
2183 unsigned long blocksize = inode->i_sb->s_blocksize;
2184 unsigned long offset = (inode->i_size) & (PAGE_SIZE - 1);
2185 struct buffer_head *bh;
2186 struct buffer_head *head;
2191 * we know that we are only called with inode->i_size > 0.
2192 * we also know that a file tail can never be as big as a block
2193 * If i_size % blocksize == 0, our file is currently block aligned
2194 * and it won't need converting or zeroing after a truncate.
2196 if ((offset & (blocksize - 1)) == 0) {
2199 page = grab_cache_page(inode->i_mapping, index);
2204 /* start within the page of the last block in the file */
2205 start = (offset / blocksize) * blocksize;
2207 error = __block_write_begin(page, start, offset - start,
2208 reiserfs_get_block_create_0);
2212 head = page_buffers(page);
2218 bh = bh->b_this_page;
2220 } while (bh != head);
2222 if (!buffer_uptodate(bh)) {
2224 * note, this should never happen, prepare_write should be
2225 * taking care of this for us. If the buffer isn't up to
2226 * date, I've screwed up the code to find the buffer, or the
2227 * code to call prepare_write
2229 reiserfs_error(inode->i_sb, "clm-6000",
2230 "error reading block %lu", bh->b_blocknr);
2235 *page_result = page;
2247 * vfs version of truncate file. Must NOT be called with
2248 * a transaction already started.
2250 * some code taken from block_truncate_page
2252 int reiserfs_truncate_file(struct inode *inode, int update_timestamps)
2254 struct reiserfs_transaction_handle th;
2255 /* we want the offset for the first byte after the end of the file */
2256 unsigned long offset = inode->i_size & (PAGE_SIZE - 1);
2257 unsigned blocksize = inode->i_sb->s_blocksize;
2259 struct page *page = NULL;
2261 struct buffer_head *bh = NULL;
2264 reiserfs_write_lock(inode->i_sb);
2266 if (inode->i_size > 0) {
2267 error = grab_tail_page(inode, &page, &bh);
2270 * -ENOENT means we truncated past the end of the
2271 * file, and get_block_create_0 could not find a
2272 * block to read in, which is ok.
2274 if (error != -ENOENT)
2275 reiserfs_error(inode->i_sb, "clm-6001",
2276 "grab_tail_page failed %d",
2284 * so, if page != NULL, we have a buffer head for the offset at
2285 * the end of the file. if the bh is mapped, and bh->b_blocknr != 0,
2286 * then we have an unformatted node. Otherwise, we have a direct item,
2287 * and no zeroing is required on disk. We zero after the truncate,
2288 * because the truncate might pack the item anyway
2289 * (it will unmap bh if it packs).
2291 * it is enough to reserve space in transaction for 2 balancings:
2292 * one for "save" link adding and another for the first
2293 * cut_from_item. 1 is for update_sd
2295 error = journal_begin(&th, inode->i_sb,
2296 JOURNAL_PER_BALANCE_CNT * 2 + 1);
2299 reiserfs_update_inode_transaction(inode);
2300 if (update_timestamps)
2302 * we are doing real truncate: if the system crashes
2303 * before the last transaction of truncating gets committed
2304 * - on reboot the file either appears truncated properly
2305 * or not truncated at all
2307 add_save_link(&th, inode, 1);
2308 err2 = reiserfs_do_truncate(&th, inode, page, update_timestamps);
2309 error = journal_end(&th);
2313 /* check reiserfs_do_truncate after ending the transaction */
2319 if (update_timestamps) {
2320 error = remove_save_link(inode, 1 /* truncate */);
2326 length = offset & (blocksize - 1);
2327 /* if we are not on a block boundary */
2329 length = blocksize - length;
2330 zero_user(page, offset, length);
2331 if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2332 mark_buffer_dirty(bh);
2339 reiserfs_write_unlock(inode->i_sb);
2348 reiserfs_write_unlock(inode->i_sb);
2353 static int map_block_for_writepage(struct inode *inode,
2354 struct buffer_head *bh_result,
2355 unsigned long block)
2357 struct reiserfs_transaction_handle th;
2359 struct item_head tmp_ih;
2360 struct item_head *ih;
2361 struct buffer_head *bh;
2364 INITIALIZE_PATH(path);
2366 int jbegin_count = JOURNAL_PER_BALANCE_CNT;
2367 loff_t byte_offset = ((loff_t)block << inode->i_sb->s_blocksize_bits)+1;
2369 int use_get_block = 0;
2370 int bytes_copied = 0;
2372 int trans_running = 0;
2375 * catch places below that try to log something without
2380 if (!buffer_uptodate(bh_result)) {
2384 kmap(bh_result->b_page);
2386 reiserfs_write_lock(inode->i_sb);
2387 make_cpu_key(&key, inode, byte_offset, TYPE_ANY, 3);
2390 retval = search_for_position_by_key(inode->i_sb, &key, &path);
2391 if (retval != POSITION_FOUND) {
2396 bh = get_last_bh(&path);
2397 ih = tp_item_head(&path);
2398 item = tp_item_body(&path);
2399 pos_in_item = path.pos_in_item;
2401 /* we've found an unformatted node */
2402 if (indirect_item_found(retval, ih)) {
2403 if (bytes_copied > 0) {
2404 reiserfs_warning(inode->i_sb, "clm-6002",
2405 "bytes_copied %d", bytes_copied);
2407 if (!get_block_num(item, pos_in_item)) {
2408 /* crap, we are writing to a hole */
2412 set_block_dev_mapped(bh_result,
2413 get_block_num(item, pos_in_item), inode);
2414 } else if (is_direct_le_ih(ih)) {
2416 p = page_address(bh_result->b_page);
2417 p += (byte_offset - 1) & (PAGE_SIZE - 1);
2418 copy_size = ih_item_len(ih) - pos_in_item;
2420 fs_gen = get_generation(inode->i_sb);
2421 copy_item_head(&tmp_ih, ih);
2423 if (!trans_running) {
2424 /* vs-3050 is gone, no need to drop the path */
2425 retval = journal_begin(&th, inode->i_sb, jbegin_count);
2428 reiserfs_update_inode_transaction(inode);
2430 if (fs_changed(fs_gen, inode->i_sb)
2431 && item_moved(&tmp_ih, &path)) {
2432 reiserfs_restore_prepared_buffer(inode->i_sb,
2438 reiserfs_prepare_for_journal(inode->i_sb, bh, 1);
2440 if (fs_changed(fs_gen, inode->i_sb)
2441 && item_moved(&tmp_ih, &path)) {
2442 reiserfs_restore_prepared_buffer(inode->i_sb, bh);
2446 memcpy(ih_item_body(bh, ih) + pos_in_item, p + bytes_copied,
2449 journal_mark_dirty(&th, bh);
2450 bytes_copied += copy_size;
2451 set_block_dev_mapped(bh_result, 0, inode);
2453 /* are there still bytes left? */
2454 if (bytes_copied < bh_result->b_size &&
2455 (byte_offset + bytes_copied) < inode->i_size) {
2456 set_cpu_key_k_offset(&key,
2457 cpu_key_k_offset(&key) +
2462 reiserfs_warning(inode->i_sb, "clm-6003",
2463 "bad item inode %lu", inode->i_ino);
2471 if (trans_running) {
2472 int err = journal_end(&th);
2477 reiserfs_write_unlock(inode->i_sb);
2479 /* this is where we fill in holes in the file. */
2480 if (use_get_block) {
2481 retval = reiserfs_get_block(inode, block, bh_result,
2482 GET_BLOCK_CREATE | GET_BLOCK_NO_IMUX
2483 | GET_BLOCK_NO_DANGLE);
2485 if (!buffer_mapped(bh_result)
2486 || bh_result->b_blocknr == 0) {
2487 /* get_block failed to find a mapped unformatted node. */
2493 kunmap(bh_result->b_page);
2495 if (!retval && buffer_mapped(bh_result) && bh_result->b_blocknr == 0) {
2497 * we've copied data from the page into the direct item, so the
2498 * buffer in the page is now clean, mark it to reflect that.
2500 lock_buffer(bh_result);
2501 clear_buffer_dirty(bh_result);
2502 unlock_buffer(bh_result);
2508 * mason@suse.com: updated in 2.5.54 to follow the same general io
2509 * start/recovery path as __block_write_full_page, along with special
2510 * code to handle reiserfs tails.
2512 static int reiserfs_write_full_page(struct page *page,
2513 struct writeback_control *wbc)
2515 struct inode *inode = page->mapping->host;
2516 unsigned long end_index = inode->i_size >> PAGE_SHIFT;
2518 unsigned long block;
2519 sector_t last_block;
2520 struct buffer_head *head, *bh;
2523 int checked = PageChecked(page);
2524 struct reiserfs_transaction_handle th;
2525 struct super_block *s = inode->i_sb;
2526 int bh_per_page = PAGE_SIZE / s->s_blocksize;
2529 /* no logging allowed when nonblocking or from PF_MEMALLOC */
2530 if (checked && (current->flags & PF_MEMALLOC)) {
2531 redirty_page_for_writepage(wbc, page);
2537 * The page dirty bit is cleared before writepage is called, which
2538 * means we have to tell create_empty_buffers to make dirty buffers
2539 * The page really should be up to date at this point, so tossing
2540 * in the BH_Uptodate is just a sanity check.
2542 if (!page_has_buffers(page)) {
2543 create_empty_buffers(page, s->s_blocksize,
2544 (1 << BH_Dirty) | (1 << BH_Uptodate));
2546 head = page_buffers(page);
2549 * last page in the file, zero out any contents past the
2550 * last byte in the file
2552 if (page->index >= end_index) {
2553 unsigned last_offset;
2555 last_offset = inode->i_size & (PAGE_SIZE - 1);
2556 /* no file contents in this page */
2557 if (page->index >= end_index + 1 || !last_offset) {
2561 zero_user_segment(page, last_offset, PAGE_SIZE);
2564 block = page->index << (PAGE_SHIFT - s->s_blocksize_bits);
2565 last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
2566 /* first map all the buffers, logging any direct items we find */
2568 if (block > last_block) {
2570 * This can happen when the block size is less than
2571 * the page size. The corresponding bytes in the page
2572 * were zero filled above
2574 clear_buffer_dirty(bh);
2575 set_buffer_uptodate(bh);
2576 } else if ((checked || buffer_dirty(bh)) &&
2577 (!buffer_mapped(bh) || bh->b_blocknr == 0)) {
2579 * not mapped yet, or it points to a direct item, search
2580 * the btree for the mapping info, and log any direct
2583 if ((error = map_block_for_writepage(inode, bh, block))) {
2587 bh = bh->b_this_page;
2589 } while (bh != head);
2592 * we start the transaction after map_block_for_writepage,
2593 * because it can create holes in the file (an unbounded operation).
2594 * starting it here, we can make a reliable estimate for how many
2595 * blocks we're going to log
2598 ClearPageChecked(page);
2599 reiserfs_write_lock(s);
2600 error = journal_begin(&th, s, bh_per_page + 1);
2602 reiserfs_write_unlock(s);
2605 reiserfs_update_inode_transaction(inode);
2607 /* now go through and lock any dirty buffers on the page */
2610 if (!buffer_mapped(bh))
2612 if (buffer_mapped(bh) && bh->b_blocknr == 0)
2616 reiserfs_prepare_for_journal(s, bh, 1);
2617 journal_mark_dirty(&th, bh);
2621 * from this point on, we know the buffer is mapped to a
2622 * real block and not a direct item
2624 if (wbc->sync_mode != WB_SYNC_NONE) {
2627 if (!trylock_buffer(bh)) {
2628 redirty_page_for_writepage(wbc, page);
2632 if (test_clear_buffer_dirty(bh)) {
2633 mark_buffer_async_write(bh);
2637 } while ((bh = bh->b_this_page) != head);
2640 error = journal_end(&th);
2641 reiserfs_write_unlock(s);
2645 BUG_ON(PageWriteback(page));
2646 set_page_writeback(page);
2650 * since any buffer might be the only dirty buffer on the page,
2651 * the first submit_bh can bring the page out of writeback.
2652 * be careful with the buffers.
2655 struct buffer_head *next = bh->b_this_page;
2656 if (buffer_async_write(bh)) {
2657 submit_bh(REQ_OP_WRITE, bh);
2662 } while (bh != head);
2668 * if this page only had a direct item, it is very possible for
2669 * no io to be required without there being an error. Or,
2670 * someone else could have locked them and sent them down the
2671 * pipe without locking the page
2675 if (!buffer_uptodate(bh)) {
2679 bh = bh->b_this_page;
2680 } while (bh != head);
2682 SetPageUptodate(page);
2683 end_page_writeback(page);
2689 * catches various errors, we need to make sure any valid dirty blocks
2690 * get to the media. The page is currently locked and not marked for
2693 ClearPageUptodate(page);
2697 if (buffer_mapped(bh) && buffer_dirty(bh) && bh->b_blocknr) {
2699 mark_buffer_async_write(bh);
2702 * clear any dirty bits that might have come from
2703 * getting attached to a dirty page
2705 clear_buffer_dirty(bh);
2707 bh = bh->b_this_page;
2708 } while (bh != head);
2710 BUG_ON(PageWriteback(page));
2711 set_page_writeback(page);
2714 struct buffer_head *next = bh->b_this_page;
2715 if (buffer_async_write(bh)) {
2716 clear_buffer_dirty(bh);
2717 submit_bh(REQ_OP_WRITE, bh);
2722 } while (bh != head);
2726 static int reiserfs_read_folio(struct file *f, struct folio *folio)
2728 return block_read_full_folio(folio, reiserfs_get_block);
2731 static int reiserfs_writepage(struct page *page, struct writeback_control *wbc)
2733 struct inode *inode = page->mapping->host;
2734 reiserfs_wait_on_write_block(inode->i_sb);
2735 return reiserfs_write_full_page(page, wbc);
2738 static void reiserfs_truncate_failed_write(struct inode *inode)
2740 truncate_inode_pages(inode->i_mapping, inode->i_size);
2741 reiserfs_truncate_file(inode, 0);
2744 static int reiserfs_write_begin(struct file *file,
2745 struct address_space *mapping,
2746 loff_t pos, unsigned len,
2747 struct page **pagep, void **fsdata)
2749 struct inode *inode;
2755 inode = mapping->host;
2756 index = pos >> PAGE_SHIFT;
2757 page = grab_cache_page_write_begin(mapping, index);
2762 reiserfs_wait_on_write_block(inode->i_sb);
2763 fix_tail_page_for_writing(page);
2764 if (reiserfs_transaction_running(inode->i_sb)) {
2765 struct reiserfs_transaction_handle *th;
2766 th = (struct reiserfs_transaction_handle *)current->
2768 BUG_ON(!th->t_refcount);
2769 BUG_ON(!th->t_trans_id);
2770 old_ref = th->t_refcount;
2773 ret = __block_write_begin(page, pos, len, reiserfs_get_block);
2774 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2775 struct reiserfs_transaction_handle *th = current->journal_info;
2777 * this gets a little ugly. If reiserfs_get_block returned an
2778 * error and left a transacstion running, we've got to close
2779 * it, and we've got to free handle if it was a persistent
2782 * But, if we had nested into an existing transaction, we need
2783 * to just drop the ref count on the handle.
2785 * If old_ref == 0, the transaction is from reiserfs_get_block,
2786 * and it was a persistent trans. Otherwise, it was nested
2789 if (th->t_refcount > old_ref) {
2794 reiserfs_write_lock(inode->i_sb);
2795 err = reiserfs_end_persistent_transaction(th);
2796 reiserfs_write_unlock(inode->i_sb);
2805 /* Truncate allocated blocks */
2806 reiserfs_truncate_failed_write(inode);
2811 int __reiserfs_write_begin(struct page *page, unsigned from, unsigned len)
2813 struct inode *inode = page->mapping->host;
2818 depth = reiserfs_write_unlock_nested(inode->i_sb);
2819 reiserfs_wait_on_write_block(inode->i_sb);
2820 reiserfs_write_lock_nested(inode->i_sb, depth);
2822 fix_tail_page_for_writing(page);
2823 if (reiserfs_transaction_running(inode->i_sb)) {
2824 struct reiserfs_transaction_handle *th;
2825 th = (struct reiserfs_transaction_handle *)current->
2827 BUG_ON(!th->t_refcount);
2828 BUG_ON(!th->t_trans_id);
2829 old_ref = th->t_refcount;
2833 ret = __block_write_begin(page, from, len, reiserfs_get_block);
2834 if (ret && reiserfs_transaction_running(inode->i_sb)) {
2835 struct reiserfs_transaction_handle *th = current->journal_info;
2837 * this gets a little ugly. If reiserfs_get_block returned an
2838 * error and left a transacstion running, we've got to close
2839 * it, and we've got to free handle if it was a persistent
2842 * But, if we had nested into an existing transaction, we need
2843 * to just drop the ref count on the handle.
2845 * If old_ref == 0, the transaction is from reiserfs_get_block,
2846 * and it was a persistent trans. Otherwise, it was nested
2849 if (th->t_refcount > old_ref) {
2854 reiserfs_write_lock(inode->i_sb);
2855 err = reiserfs_end_persistent_transaction(th);
2856 reiserfs_write_unlock(inode->i_sb);
2866 static sector_t reiserfs_aop_bmap(struct address_space *as, sector_t block)
2868 return generic_block_bmap(as, block, reiserfs_bmap);
2871 static int reiserfs_write_end(struct file *file, struct address_space *mapping,
2872 loff_t pos, unsigned len, unsigned copied,
2873 struct page *page, void *fsdata)
2875 struct inode *inode = page->mapping->host;
2878 struct reiserfs_transaction_handle *th;
2880 bool locked = false;
2882 reiserfs_wait_on_write_block(inode->i_sb);
2883 if (reiserfs_transaction_running(inode->i_sb))
2884 th = current->journal_info;
2888 start = pos & (PAGE_SIZE - 1);
2889 if (unlikely(copied < len)) {
2890 if (!PageUptodate(page))
2893 page_zero_new_buffers(page, start + copied, start + len);
2895 flush_dcache_page(page);
2897 reiserfs_commit_page(inode, page, start, start + copied);
2900 * generic_commit_write does this for us, but does not update the
2901 * transaction tracking stuff when the size changes. So, we have
2902 * to do the i_size updates here.
2904 if (pos + copied > inode->i_size) {
2905 struct reiserfs_transaction_handle myth;
2906 reiserfs_write_lock(inode->i_sb);
2909 * If the file have grown beyond the border where it
2910 * can have a tail, unmark it as needing a tail
2913 if ((have_large_tails(inode->i_sb)
2914 && inode->i_size > i_block_size(inode) * 4)
2915 || (have_small_tails(inode->i_sb)
2916 && inode->i_size > i_block_size(inode)))
2917 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
2919 ret = journal_begin(&myth, inode->i_sb, 1);
2923 reiserfs_update_inode_transaction(inode);
2924 inode->i_size = pos + copied;
2926 * this will just nest into our transaction. It's important
2927 * to use mark_inode_dirty so the inode gets pushed around on
2928 * the dirty lists, and so that O_SYNC works as expected
2930 mark_inode_dirty(inode);
2931 reiserfs_update_sd(&myth, inode);
2933 ret = journal_end(&myth);
2939 reiserfs_write_lock(inode->i_sb);
2943 mark_inode_dirty(inode);
2944 ret = reiserfs_end_persistent_transaction(th);
2951 reiserfs_write_unlock(inode->i_sb);
2955 if (pos + len > inode->i_size)
2956 reiserfs_truncate_failed_write(inode);
2958 return ret == 0 ? copied : ret;
2961 reiserfs_write_unlock(inode->i_sb);
2965 reiserfs_update_sd(th, inode);
2966 ret = reiserfs_end_persistent_transaction(th);
2971 int reiserfs_commit_write(struct file *f, struct page *page,
2972 unsigned from, unsigned to)
2974 struct inode *inode = page->mapping->host;
2975 loff_t pos = ((loff_t) page->index << PAGE_SHIFT) + to;
2978 struct reiserfs_transaction_handle *th = NULL;
2981 depth = reiserfs_write_unlock_nested(inode->i_sb);
2982 reiserfs_wait_on_write_block(inode->i_sb);
2983 reiserfs_write_lock_nested(inode->i_sb, depth);
2985 if (reiserfs_transaction_running(inode->i_sb)) {
2986 th = current->journal_info;
2988 reiserfs_commit_page(inode, page, from, to);
2991 * generic_commit_write does this for us, but does not update the
2992 * transaction tracking stuff when the size changes. So, we have
2993 * to do the i_size updates here.
2995 if (pos > inode->i_size) {
2996 struct reiserfs_transaction_handle myth;
2998 * If the file have grown beyond the border where it
2999 * can have a tail, unmark it as needing a tail
3002 if ((have_large_tails(inode->i_sb)
3003 && inode->i_size > i_block_size(inode) * 4)
3004 || (have_small_tails(inode->i_sb)
3005 && inode->i_size > i_block_size(inode)))
3006 REISERFS_I(inode)->i_flags &= ~i_pack_on_close_mask;
3008 ret = journal_begin(&myth, inode->i_sb, 1);
3012 reiserfs_update_inode_transaction(inode);
3013 inode->i_size = pos;
3015 * this will just nest into our transaction. It's important
3016 * to use mark_inode_dirty so the inode gets pushed around
3017 * on the dirty lists, and so that O_SYNC works as expected
3019 mark_inode_dirty(inode);
3020 reiserfs_update_sd(&myth, inode);
3022 ret = journal_end(&myth);
3028 mark_inode_dirty(inode);
3029 ret = reiserfs_end_persistent_transaction(th);
3040 reiserfs_update_sd(th, inode);
3041 ret = reiserfs_end_persistent_transaction(th);
3047 void sd_attrs_to_i_attrs(__u16 sd_attrs, struct inode *inode)
3049 if (reiserfs_attrs(inode->i_sb)) {
3050 if (sd_attrs & REISERFS_SYNC_FL)
3051 inode->i_flags |= S_SYNC;
3053 inode->i_flags &= ~S_SYNC;
3054 if (sd_attrs & REISERFS_IMMUTABLE_FL)
3055 inode->i_flags |= S_IMMUTABLE;
3057 inode->i_flags &= ~S_IMMUTABLE;
3058 if (sd_attrs & REISERFS_APPEND_FL)
3059 inode->i_flags |= S_APPEND;
3061 inode->i_flags &= ~S_APPEND;
3062 if (sd_attrs & REISERFS_NOATIME_FL)
3063 inode->i_flags |= S_NOATIME;
3065 inode->i_flags &= ~S_NOATIME;
3066 if (sd_attrs & REISERFS_NOTAIL_FL)
3067 REISERFS_I(inode)->i_flags |= i_nopack_mask;
3069 REISERFS_I(inode)->i_flags &= ~i_nopack_mask;
3074 * decide if this buffer needs to stay around for data logging or ordered
3077 static int invalidate_folio_can_drop(struct inode *inode, struct buffer_head *bh)
3080 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3083 spin_lock(&j->j_dirty_buffers_lock);
3084 if (!buffer_mapped(bh)) {
3088 * the page is locked, and the only places that log a data buffer
3089 * also lock the page.
3091 if (reiserfs_file_data_log(inode)) {
3093 * very conservative, leave the buffer pinned if
3094 * anyone might need it.
3096 if (buffer_journaled(bh) || buffer_journal_dirty(bh)) {
3099 } else if (buffer_dirty(bh)) {
3100 struct reiserfs_journal_list *jl;
3101 struct reiserfs_jh *jh = bh->b_private;
3105 * reiserfs_setattr updates i_size in the on disk
3106 * stat data before allowing vmtruncate to be called.
3108 * If buffer was put onto the ordered list for this
3109 * transaction, we know for sure either this transaction
3110 * or an older one already has updated i_size on disk,
3111 * and this ordered data won't be referenced in the file
3114 * if the buffer was put onto the ordered list for an older
3115 * transaction, we need to leave it around
3117 if (jh && (jl = jh->jl)
3118 && jl != SB_JOURNAL(inode->i_sb)->j_current_jl)
3122 if (ret && bh->b_private) {
3123 reiserfs_free_jh(bh);
3125 spin_unlock(&j->j_dirty_buffers_lock);
3130 /* clm -- taken from fs/buffer.c:block_invalidate_folio */
3131 static void reiserfs_invalidate_folio(struct folio *folio, size_t offset,
3134 struct buffer_head *head, *bh, *next;
3135 struct inode *inode = folio->mapping->host;
3136 unsigned int curr_off = 0;
3137 unsigned int stop = offset + length;
3138 int partial_page = (offset || length < folio_size(folio));
3141 BUG_ON(!folio_test_locked(folio));
3144 folio_clear_checked(folio);
3146 head = folio_buffers(folio);
3152 unsigned int next_off = curr_off + bh->b_size;
3153 next = bh->b_this_page;
3155 if (next_off > stop)
3159 * is this block fully invalidated?
3161 if (offset <= curr_off) {
3162 if (invalidate_folio_can_drop(inode, bh))
3163 reiserfs_unmap_buffer(bh);
3167 curr_off = next_off;
3169 } while (bh != head);
3172 * We release buffers only if the entire page is being invalidated.
3173 * The get_block cached value has been unconditionally invalidated,
3174 * so real IO is not possible anymore.
3176 if (!partial_page && ret) {
3177 ret = filemap_release_folio(folio, 0);
3178 /* maybe should BUG_ON(!ret); - neilb */
3184 static bool reiserfs_dirty_folio(struct address_space *mapping,
3185 struct folio *folio)
3187 if (reiserfs_file_data_log(mapping->host)) {
3188 folio_set_checked(folio);
3189 return filemap_dirty_folio(mapping, folio);
3191 return block_dirty_folio(mapping, folio);
3195 * Returns true if the folio's buffers were dropped. The folio is locked.
3197 * Takes j_dirty_buffers_lock to protect the b_assoc_buffers list_heads
3198 * in the buffers at folio_buffers(folio).
3200 * even in -o notail mode, we can't be sure an old mount without -o notail
3201 * didn't create files with tails.
3203 static bool reiserfs_release_folio(struct folio *folio, gfp_t unused_gfp_flags)
3205 struct inode *inode = folio->mapping->host;
3206 struct reiserfs_journal *j = SB_JOURNAL(inode->i_sb);
3207 struct buffer_head *head;
3208 struct buffer_head *bh;
3211 WARN_ON(folio_test_checked(folio));
3212 spin_lock(&j->j_dirty_buffers_lock);
3213 head = folio_buffers(folio);
3216 if (bh->b_private) {
3217 if (!buffer_dirty(bh) && !buffer_locked(bh)) {
3218 reiserfs_free_jh(bh);
3224 bh = bh->b_this_page;
3225 } while (bh != head);
3227 ret = try_to_free_buffers(folio);
3228 spin_unlock(&j->j_dirty_buffers_lock);
3233 * We thank Mingming Cao for helping us understand in great detail what
3234 * to do in this section of the code.
3236 static ssize_t reiserfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3238 struct file *file = iocb->ki_filp;
3239 struct inode *inode = file->f_mapping->host;
3240 size_t count = iov_iter_count(iter);
3243 ret = blockdev_direct_IO(iocb, inode, iter,
3244 reiserfs_get_blocks_direct_io);
3247 * In case of error extending write may have instantiated a few
3248 * blocks outside i_size. Trim these off again.
3250 if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
3251 loff_t isize = i_size_read(inode);
3252 loff_t end = iocb->ki_pos + count;
3254 if ((end > isize) && inode_newsize_ok(inode, isize) == 0) {
3255 truncate_setsize(inode, isize);
3256 reiserfs_vfs_truncate_file(inode);
3263 int reiserfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
3266 struct inode *inode = d_inode(dentry);
3267 unsigned int ia_valid;
3270 error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
3274 /* must be turned off for recursive notify_change calls */
3275 ia_valid = attr->ia_valid &= ~(ATTR_KILL_SUID|ATTR_KILL_SGID);
3277 if (is_quota_modification(&nop_mnt_idmap, inode, attr)) {
3278 error = dquot_initialize(inode);
3282 reiserfs_write_lock(inode->i_sb);
3283 if (attr->ia_valid & ATTR_SIZE) {
3285 * version 2 items will be caught by the s_maxbytes check
3286 * done for us in vmtruncate
3288 if (get_inode_item_key_version(inode) == KEY_FORMAT_3_5 &&
3289 attr->ia_size > MAX_NON_LFS) {
3290 reiserfs_write_unlock(inode->i_sb);
3295 inode_dio_wait(inode);
3297 /* fill in hole pointers in the expanding truncate case. */
3298 if (attr->ia_size > inode->i_size) {
3299 loff_t pos = attr->ia_size;
3301 if ((pos & (inode->i_sb->s_blocksize - 1)) == 0)
3303 error = generic_cont_expand_simple(inode, pos);
3304 if (REISERFS_I(inode)->i_prealloc_count > 0) {
3306 struct reiserfs_transaction_handle th;
3307 /* we're changing at most 2 bitmaps, inode + super */
3308 err = journal_begin(&th, inode->i_sb, 4);
3310 reiserfs_discard_prealloc(&th, inode);
3311 err = journal_end(&th);
3317 reiserfs_write_unlock(inode->i_sb);
3321 * file size is changed, ctime and mtime are
3324 attr->ia_valid |= (ATTR_MTIME | ATTR_CTIME);
3327 reiserfs_write_unlock(inode->i_sb);
3329 if ((((attr->ia_valid & ATTR_UID) && (from_kuid(&init_user_ns, attr->ia_uid) & ~0xffff)) ||
3330 ((attr->ia_valid & ATTR_GID) && (from_kgid(&init_user_ns, attr->ia_gid) & ~0xffff))) &&
3331 (get_inode_sd_version(inode) == STAT_DATA_V1)) {
3332 /* stat data of format v3.5 has 16 bit uid and gid */
3337 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
3338 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
3339 struct reiserfs_transaction_handle th;
3342 (REISERFS_QUOTA_INIT_BLOCKS(inode->i_sb) +
3343 REISERFS_QUOTA_DEL_BLOCKS(inode->i_sb)) +
3346 error = reiserfs_chown_xattrs(inode, attr);
3352 * (user+group)*(old+new) structure - we count quota
3353 * info and , inode write (sb, inode)
3355 reiserfs_write_lock(inode->i_sb);
3356 error = journal_begin(&th, inode->i_sb, jbegin_count);
3357 reiserfs_write_unlock(inode->i_sb);
3360 error = dquot_transfer(&nop_mnt_idmap, inode, attr);
3361 reiserfs_write_lock(inode->i_sb);
3364 reiserfs_write_unlock(inode->i_sb);
3369 * Update corresponding info in inode so that everything
3370 * is in one transaction
3372 if (attr->ia_valid & ATTR_UID)
3373 inode->i_uid = attr->ia_uid;
3374 if (attr->ia_valid & ATTR_GID)
3375 inode->i_gid = attr->ia_gid;
3376 mark_inode_dirty(inode);
3377 error = journal_end(&th);
3378 reiserfs_write_unlock(inode->i_sb);
3383 if ((attr->ia_valid & ATTR_SIZE) &&
3384 attr->ia_size != i_size_read(inode)) {
3385 error = inode_newsize_ok(inode, attr->ia_size);
3388 * Could race against reiserfs_file_release
3389 * if called from NFS, so take tailpack mutex.
3391 mutex_lock(&REISERFS_I(inode)->tailpack);
3392 truncate_setsize(inode, attr->ia_size);
3393 reiserfs_truncate_file(inode, 1);
3394 mutex_unlock(&REISERFS_I(inode)->tailpack);
3399 setattr_copy(&nop_mnt_idmap, inode, attr);
3400 mark_inode_dirty(inode);
3403 if (!error && reiserfs_posixacl(inode->i_sb)) {
3404 if (attr->ia_valid & ATTR_MODE)
3405 error = reiserfs_acl_chmod(dentry);
3412 const struct address_space_operations reiserfs_address_space_operations = {
3413 .writepage = reiserfs_writepage,
3414 .read_folio = reiserfs_read_folio,
3415 .readahead = reiserfs_readahead,
3416 .release_folio = reiserfs_release_folio,
3417 .invalidate_folio = reiserfs_invalidate_folio,
3418 .write_begin = reiserfs_write_begin,
3419 .write_end = reiserfs_write_end,
3420 .bmap = reiserfs_aop_bmap,
3421 .direct_IO = reiserfs_direct_IO,
3422 .dirty_folio = reiserfs_dirty_folio,