aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs
diff options
context:
space:
mode:
authorChris Mason <chris.mason@oracle.com>2008-04-09 16:28:12 -0400
committerChris Mason <chris.mason@oracle.com>2008-09-25 11:04:01 -0400
commitce9adaa5a792c2099a83246265eb4055bc38b6b8 (patch)
tree3ac03bb73cd3aad0c0b194d387cbb5b136c92fa7 /fs/btrfs
parent728131d8e40c2a47c59ca91a806299c4708029f9 (diff)
Btrfs: Do metadata checksums for reads via a workqueue
Before, metadata checksumming was done by the callers of read_tree_block, which would set EXTENT_CSUM bits in the extent tree to show that a given range of pages was already checksummed and didn't need to be verified again. But, those bits could go away via try_to_releasepage, and the end result was bogus checksum failures on pages that never left the cache. The new code validates checksums when the page is read. It is a little tricky because metadata blocks can span pages and a single read may end up going via multiple bios. Signed-off-by: Chris Mason <chris.mason@oracle.com>
Diffstat (limited to 'fs/btrfs')
-rw-r--r--fs/btrfs/ctree.c5
-rw-r--r--fs/btrfs/ctree.h4
-rw-r--r--fs/btrfs/disk-io.c258
-rw-r--r--fs/btrfs/extent-tree.c4
-rw-r--r--fs/btrfs/extent_io.c100
-rw-r--r--fs/btrfs/extent_io.h4
-rw-r--r--fs/btrfs/transaction.c2
7 files changed, 310 insertions, 67 deletions
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index c92c6b0ee582..efce173a935c 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -158,9 +158,6 @@ int __btrfs_cow_block(struct btrfs_trans_handle *trans,
158 } else { 158 } else {
159 root_gen = 0; 159 root_gen = 0;
160 } 160 }
161 if (!(buf->flags & EXTENT_CSUM))
162 WARN_ON(1);
163
164 WARN_ON(root->ref_cows && trans->transid != 161 WARN_ON(root->ref_cows && trans->transid !=
165 root->fs_info->running_transaction->transid); 162 root->fs_info->running_transaction->transid);
166 WARN_ON(root->ref_cows && trans->transid != root->last_trans); 163 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
@@ -247,8 +244,6 @@ int btrfs_cow_block(struct btrfs_trans_handle *trans,
247 root->fs_info->generation); 244 root->fs_info->generation);
248 WARN_ON(1); 245 WARN_ON(1);
249 } 246 }
250 if (!(buf->flags & EXTENT_CSUM))
251 WARN_ON(1);
252 247
253 header_trans = btrfs_header_generation(buf); 248 header_trans = btrfs_header_generation(buf);
254 spin_lock(&root->fs_info->hash_lock); 249 spin_lock(&root->fs_info->hash_lock);
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 4b3b20459f6e..e803c4daad21 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -485,6 +485,10 @@ struct btrfs_fs_info {
485 struct list_head trans_list; 485 struct list_head trans_list;
486 struct list_head hashers; 486 struct list_head hashers;
487 struct list_head dead_roots; 487 struct list_head dead_roots;
488 struct list_head end_io_work_list;
489 struct work_struct end_io_work;
490 spinlock_t end_io_work_lock;
491
488#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) 492#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
489 struct work_struct trans_work; 493 struct work_struct trans_work;
490#else 494#else
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e444b99e02da..82109204788c 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -24,6 +24,7 @@
24#include <linux/radix-tree.h> 24#include <linux/radix-tree.h>
25#include <linux/writeback.h> 25#include <linux/writeback.h>
26#include <linux/buffer_head.h> // for block_sync_page 26#include <linux/buffer_head.h> // for block_sync_page
27#include <linux/workqueue.h>
27#include "ctree.h" 28#include "ctree.h"
28#include "disk-io.h" 29#include "disk-io.h"
29#include "transaction.h" 30#include "transaction.h"
@@ -45,6 +46,16 @@ static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
45#endif 46#endif
46 47
47static struct extent_io_ops btree_extent_io_ops; 48static struct extent_io_ops btree_extent_io_ops;
49static struct workqueue_struct *end_io_workqueue;
50
51struct end_io_wq {
52 struct bio *bio;
53 bio_end_io_t *end_io;
54 void *private;
55 struct btrfs_fs_info *info;
56 int error;
57 struct list_head list;
58};
48 59
49struct extent_map *btree_get_extent(struct inode *inode, struct page *page, 60struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
50 size_t page_offset, u64 start, u64 len, 61 size_t page_offset, u64 start, u64 len,
@@ -219,11 +230,108 @@ static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
219 return 0; 230 return 0;
220} 231}
221 232
233int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
234 struct extent_state *state)
235{
236 struct extent_io_tree *tree;
237 u64 found_start;
238 int found_level;
239 unsigned long len;
240 struct extent_buffer *eb;
241 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
242 int ret;
243
244 tree = &BTRFS_I(page->mapping->host)->io_tree;
245 if (page->private == EXTENT_PAGE_PRIVATE)
246 goto out;
247 if (!page->private)
248 goto out;
249 len = page->private >> 2;
250 if (len == 0) {
251 WARN_ON(1);
252 }
253 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
254 read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1,
255 btree_get_extent);
256 btrfs_clear_buffer_defrag(eb);
257 found_start = btrfs_header_bytenr(eb);
258 if (found_start != start) {
259 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
260 start, found_start, len);
261 WARN_ON(1);
262 goto err;
263 }
264 if (eb->first_page != page) {
265 printk("bad first page %lu %lu\n", eb->first_page->index,
266 page->index);
267 WARN_ON(1);
268 goto err;
269 }
270 found_level = btrfs_header_level(eb);
271
272 ret = csum_tree_block(root, eb, 1);
273
274 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
275 end = eb->start + end - 1;
276 release_extent_buffer_tail_pages(eb);
277err:
278 free_extent_buffer(eb);
279out:
280 return 0;
281}
282
283#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
284static void end_workqueue_bio(struct bio *bio, int err)
285#else
286static int end_workqueue_bio(struct bio *bio,
287 unsigned int bytes_done, int err)
288#endif
289{
290 struct end_io_wq *end_io_wq = bio->bi_private;
291 struct btrfs_fs_info *fs_info;
292 unsigned long flags;
293
294#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
295 if (bio->bi_size)
296 return 1;
297#endif
298
299 fs_info = end_io_wq->info;
300 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
301 end_io_wq->error = err;
302 list_add_tail(&end_io_wq->list, &fs_info->end_io_work_list);
303 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
304 queue_work(end_io_workqueue, &fs_info->end_io_work);
305
306#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
307 return 0;
308#endif
309}
310
222static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio) 311static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
223{ 312{
224 struct btrfs_root *root = BTRFS_I(inode)->root; 313 struct btrfs_root *root = BTRFS_I(inode)->root;
314 struct end_io_wq *end_io_wq;
225 u64 offset; 315 u64 offset;
226 offset = bio->bi_sector << 9; 316 offset = bio->bi_sector << 9;
317
318 if (rw & (1 << BIO_RW)) {
319 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio);
320 }
321
322 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
323 if (!end_io_wq)
324 return -ENOMEM;
325
326 end_io_wq->private = bio->bi_private;
327 end_io_wq->end_io = bio->bi_end_io;
328 end_io_wq->info = root->fs_info;
329 end_io_wq->error = 0;
330 end_io_wq->bio = bio;
331
332 bio->bi_private = end_io_wq;
333 bio->bi_end_io = end_workqueue_bio;
334
227 if (offset == BTRFS_SUPER_INFO_OFFSET) { 335 if (offset == BTRFS_SUPER_INFO_OFFSET) {
228 bio->bi_bdev = root->fs_info->sb->s_bdev; 336 bio->bi_bdev = root->fs_info->sb->s_bdev;
229 submit_bio(rw, bio); 337 submit_bio(rw, bio);
@@ -363,36 +471,7 @@ static int close_all_devices(struct btrfs_fs_info *fs_info)
363int btrfs_verify_block_csum(struct btrfs_root *root, 471int btrfs_verify_block_csum(struct btrfs_root *root,
364 struct extent_buffer *buf) 472 struct extent_buffer *buf)
365{ 473{
366 struct extent_io_tree *io_tree; 474 return btrfs_buffer_uptodate(buf);
367 u64 end;
368 int ret;
369
370 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
371 if (buf->flags & EXTENT_CSUM)
372 return 0;
373
374 end = min_t(u64, buf->len, PAGE_CACHE_SIZE);
375 end = buf->start + end - 1;
376 if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
377 buf->flags |= EXTENT_CSUM;
378 return 0;
379 }
380 lock_extent(io_tree, buf->start, end, GFP_NOFS);
381
382 if (test_range_bit(io_tree, buf->start, end, EXTENT_CSUM, 1)) {
383 buf->flags |= EXTENT_CSUM;
384 ret = 0;
385 goto out_unlock;
386 }
387WARN_ON(buf->flags & EXTENT_CSUM);
388
389 ret = csum_tree_block(root, buf, 1);
390 set_extent_bits(io_tree, buf->start, end, EXTENT_CSUM, GFP_NOFS);
391 buf->flags |= EXTENT_CSUM;
392
393out_unlock:
394 unlock_extent(io_tree, buf->start, end, GFP_NOFS);
395 return ret;
396} 475}
397 476
398struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 477struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
@@ -430,11 +509,15 @@ struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
430 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 509 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
431 if (!buf) 510 if (!buf)
432 return NULL; 511 return NULL;
433 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0, 1,
434 btree_get_extent);
435 512
436 ret = btrfs_verify_block_csum(root, buf); 513 ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf, 0,
514 1, btree_get_extent);
515
516 if (ret == 0) {
517 buf->flags |= EXTENT_UPTODATE;
518 }
437 return buf; 519 return buf;
520
438} 521}
439 522
440int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 523int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
@@ -724,6 +807,99 @@ static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
724 return 0; 807 return 0;
725} 808}
726 809
810static int bio_ready_for_csum(struct bio *bio)
811{
812 u64 length = 0;
813 u64 buf_len = 0;
814 u64 start = 0;
815 struct page *page;
816 struct extent_io_tree *io_tree = NULL;
817 struct btrfs_fs_info *info = NULL;
818 struct bio_vec *bvec;
819 int i;
820 int ret;
821
822 bio_for_each_segment(bvec, bio, i) {
823 page = bvec->bv_page;
824 if (page->private == EXTENT_PAGE_PRIVATE) {
825 length += bvec->bv_len;
826 continue;
827 }
828 if (!page->private) {
829 length += bvec->bv_len;
830 continue;
831 }
832 length = bvec->bv_len;
833 buf_len = page->private >> 2;
834 start = page_offset(page) + bvec->bv_offset;
835 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
836 info = BTRFS_I(page->mapping->host)->root->fs_info;
837 }
838 /* are we fully contained in this bio? */
839 if (buf_len <= length)
840 return 1;
841
842 ret = extent_range_uptodate(io_tree, start + length,
843 start + buf_len - 1);
844 if (ret == 1)
845 return ret;
846 return ret;
847}
848
849#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
850void btrfs_end_io_csum(void *p)
851#else
852void btrfs_end_io_csum(struct work_struct *work)
853#endif
854{
855#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
856 struct btrfs_fs_info *fs_info = p;
857#else
858 struct btrfs_fs_info *fs_info = container_of(work,
859 struct btrfs_fs_info,
860 end_io_work);
861#endif
862 unsigned long flags;
863 struct end_io_wq *end_io_wq;
864 struct bio *bio;
865 struct list_head *next;
866 int error;
867 int was_empty;
868
869 while(1) {
870 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
871 if (list_empty(&fs_info->end_io_work_list)) {
872 spin_unlock_irqrestore(&fs_info->end_io_work_lock,
873 flags);
874 return;
875 }
876 next = fs_info->end_io_work_list.next;
877 list_del(next);
878 spin_unlock_irqrestore(&fs_info->end_io_work_lock, flags);
879
880 end_io_wq = list_entry(next, struct end_io_wq, list);
881
882 bio = end_io_wq->bio;
883 if (!bio_ready_for_csum(bio)) {
884 spin_lock_irqsave(&fs_info->end_io_work_lock, flags);
885 was_empty = list_empty(&fs_info->end_io_work_list);
886 list_add_tail(&end_io_wq->list,
887 &fs_info->end_io_work_list);
888 spin_unlock_irqrestore(&fs_info->end_io_work_lock,
889 flags);
890 if (was_empty)
891 return;
892 continue;
893 }
894 error = end_io_wq->error;
895 bio->bi_private = end_io_wq->private;
896 bio->bi_end_io = end_io_wq->end_io;
897 kfree(end_io_wq);
898 bio_endio(bio, error);
899 }
900}
901
902
727struct btrfs_root *open_ctree(struct super_block *sb, 903struct btrfs_root *open_ctree(struct super_block *sb,
728 struct btrfs_fs_devices *fs_devices) 904 struct btrfs_fs_devices *fs_devices)
729{ 905{
@@ -750,11 +926,16 @@ struct btrfs_root *open_ctree(struct super_block *sb,
750 err = -ENOMEM; 926 err = -ENOMEM;
751 goto fail; 927 goto fail;
752 } 928 }
929 end_io_workqueue = create_workqueue("btrfs-end-io");
930 BUG_ON(!end_io_workqueue);
931
753 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS); 932 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
754 INIT_LIST_HEAD(&fs_info->trans_list); 933 INIT_LIST_HEAD(&fs_info->trans_list);
755 INIT_LIST_HEAD(&fs_info->dead_roots); 934 INIT_LIST_HEAD(&fs_info->dead_roots);
756 INIT_LIST_HEAD(&fs_info->hashers); 935 INIT_LIST_HEAD(&fs_info->hashers);
936 INIT_LIST_HEAD(&fs_info->end_io_work_list);
757 spin_lock_init(&fs_info->hash_lock); 937 spin_lock_init(&fs_info->hash_lock);
938 spin_lock_init(&fs_info->end_io_work_lock);
758 spin_lock_init(&fs_info->delalloc_lock); 939 spin_lock_init(&fs_info->delalloc_lock);
759 spin_lock_init(&fs_info->new_trans_lock); 940 spin_lock_init(&fs_info->new_trans_lock);
760 941
@@ -799,6 +980,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
799 fs_info->btree_inode->i_mapping, GFP_NOFS); 980 fs_info->btree_inode->i_mapping, GFP_NOFS);
800 fs_info->do_barriers = 1; 981 fs_info->do_barriers = 1;
801 982
983 INIT_WORK(&fs_info->end_io_work, btrfs_end_io_csum);
802#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18) 984#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
803 INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info); 985 INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
804#else 986#else
@@ -1044,6 +1226,8 @@ int close_ctree(struct btrfs_root *root)
1044 extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree); 1226 extent_io_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->io_tree);
1045 1227
1046 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0); 1228 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1229 flush_workqueue(end_io_workqueue);
1230 destroy_workqueue(end_io_workqueue);
1047 1231
1048 iput(fs_info->btree_inode); 1232 iput(fs_info->btree_inode);
1049#if 0 1233#if 0
@@ -1171,12 +1355,18 @@ int btrfs_read_buffer(struct extent_buffer *buf)
1171{ 1355{
1172 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root; 1356 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1173 struct inode *btree_inode = root->fs_info->btree_inode; 1357 struct inode *btree_inode = root->fs_info->btree_inode;
1174 return read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1358 int ret;
1359 ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1175 buf, 0, 1, btree_get_extent); 1360 buf, 0, 1, btree_get_extent);
1361 if (ret == 0) {
1362 buf->flags |= EXTENT_UPTODATE;
1363 }
1364 return ret;
1176} 1365}
1177 1366
1178static struct extent_io_ops btree_extent_io_ops = { 1367static struct extent_io_ops btree_extent_io_ops = {
1179 .writepage_io_hook = btree_writepage_io_hook, 1368 .writepage_io_hook = btree_writepage_io_hook,
1369 .readpage_end_io_hook = btree_readpage_end_io_hook,
1180 .submit_bio_hook = btree_submit_bio_hook, 1370 .submit_bio_hook = btree_submit_bio_hook,
1181 /* note we're sharing with inode.c for the merge bio hook */ 1371 /* note we're sharing with inode.c for the merge bio hook */
1182 .merge_bio_hook = btrfs_merge_bio_hook, 1372 .merge_bio_hook = btrfs_merge_bio_hook,
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 0cb742884cdf..283b08a32a43 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -1898,10 +1898,6 @@ struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1898 1898
1899 set_extent_dirty(&trans->transaction->dirty_pages, buf->start, 1899 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1900 buf->start + buf->len - 1, GFP_NOFS); 1900 buf->start + buf->len - 1, GFP_NOFS);
1901 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1902 buf->start, buf->start + buf->len - 1,
1903 EXTENT_CSUM, GFP_NOFS);
1904 buf->flags |= EXTENT_CSUM;
1905 if (!btrfs_test_opt(root, SSD)) 1901 if (!btrfs_test_opt(root, SSD))
1906 btrfs_set_buffer_defrag(buf); 1902 btrfs_set_buffer_defrag(buf);
1907 trans->blocks_used++; 1903 trans->blocks_used++;
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 13cc2360e375..cfc383c17a3a 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -2592,6 +2592,22 @@ static inline struct page *extent_buffer_page(struct extent_buffer *eb,
2592 return p; 2592 return p;
2593} 2593}
2594 2594
2595int release_extent_buffer_tail_pages(struct extent_buffer *eb)
2596{
2597 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
2598 struct page *page;
2599 unsigned long i;
2600
2601 if (num_pages == 1)
2602 return 0;
2603 for (i = 1; i < num_pages; i++) {
2604 page = extent_buffer_page(eb, i);
2605 page_cache_release(page);
2606 }
2607 return 0;
2608}
2609
2610
2595int invalidate_extent_lru(struct extent_io_tree *tree, u64 start, 2611int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2596 unsigned long len) 2612 unsigned long len)
2597{ 2613{
@@ -2609,9 +2625,6 @@ int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
2609 if (eb->start <= start && eb->start + eb->len > start) { 2625 if (eb->start <= start && eb->start + eb->len > start) {
2610 eb->flags &= ~EXTENT_UPTODATE; 2626 eb->flags &= ~EXTENT_UPTODATE;
2611 } 2627 }
2612 if (eb->start == start) {
2613 eb->flags &= ~EXTENT_CSUM;
2614 }
2615 cur = cur->next; 2628 cur = cur->next;
2616 } while (cur != lru); 2629 } while (cur != lru);
2617out: 2630out:
@@ -2682,7 +2695,6 @@ struct extent_buffer *alloc_extent_buffer(struct extent_io_tree *tree,
2682 page_cache_get(page0); 2695 page_cache_get(page0);
2683 mark_page_accessed(page0); 2696 mark_page_accessed(page0);
2684 set_page_extent_mapped(page0); 2697 set_page_extent_mapped(page0);
2685 WARN_ON(!PageUptodate(page0));
2686 set_page_extent_head(page0, len); 2698 set_page_extent_head(page0, len);
2687 } else { 2699 } else {
2688 i = 0; 2700 i = 0;
@@ -2933,13 +2945,39 @@ int set_extent_buffer_uptodate(struct extent_io_tree *tree,
2933} 2945}
2934EXPORT_SYMBOL(set_extent_buffer_uptodate); 2946EXPORT_SYMBOL(set_extent_buffer_uptodate);
2935 2947
2948int extent_range_uptodate(struct extent_io_tree *tree,
2949 u64 start, u64 end)
2950{
2951 struct page *page;
2952 int ret;
2953 int pg_uptodate = 1;
2954 int uptodate;
2955 unsigned long index;
2956
2957 ret = test_range_bit(tree, start, end, EXTENT_UPTODATE, 1);
2958 if (ret)
2959 return 1;
2960 while(start <= end) {
2961 index = start >> PAGE_CACHE_SHIFT;
2962 page = find_get_page(tree->mapping, index);
2963 uptodate = PageUptodate(page);
2964 page_cache_release(page);
2965 if (!uptodate) {
2966 pg_uptodate = 0;
2967 break;
2968 }
2969 start += PAGE_CACHE_SIZE;
2970 }
2971 return pg_uptodate;
2972}
2973
2936int extent_buffer_uptodate(struct extent_io_tree *tree, 2974int extent_buffer_uptodate(struct extent_io_tree *tree,
2937 struct extent_buffer *eb) 2975 struct extent_buffer *eb)
2938{ 2976{
2939 int ret = 0; 2977 int ret = 0;
2940 int ret2; 2978 int ret2;
2941 int num_pages; 2979 unsigned long num_pages;
2942 int i; 2980 unsigned long i;
2943 struct page *page; 2981 struct page *page;
2944 int pg_uptodate = 1; 2982 int pg_uptodate = 1;
2945 2983
@@ -2975,13 +3013,16 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
2975 struct page *page; 3013 struct page *page;
2976 int err; 3014 int err;
2977 int ret = 0; 3015 int ret = 0;
3016 int locked_pages = 0;
3017 int all_uptodate = 1;
3018 int inc_all_pages = 0;
2978 unsigned long num_pages; 3019 unsigned long num_pages;
2979 struct bio *bio = NULL; 3020 struct bio *bio = NULL;
2980 3021
2981 if (eb->flags & EXTENT_UPTODATE) 3022 if (eb->flags & EXTENT_UPTODATE)
2982 return 0; 3023 return 0;
2983 3024
2984 if (0 && test_range_bit(tree, eb->start, eb->start + eb->len - 1, 3025 if (test_range_bit(tree, eb->start, eb->start + eb->len - 1,
2985 EXTENT_UPTODATE, 1)) { 3026 EXTENT_UPTODATE, 1)) {
2986 return 0; 3027 return 0;
2987 } 3028 }
@@ -2997,17 +3038,30 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
2997 num_pages = num_extent_pages(eb->start, eb->len); 3038 num_pages = num_extent_pages(eb->start, eb->len);
2998 for (i = start_i; i < num_pages; i++) { 3039 for (i = start_i; i < num_pages; i++) {
2999 page = extent_buffer_page(eb, i); 3040 page = extent_buffer_page(eb, i);
3000 if (PageUptodate(page)) {
3001 continue;
3002 }
3003 if (!wait) { 3041 if (!wait) {
3004 if (TestSetPageLocked(page)) { 3042 if (TestSetPageLocked(page))
3005 continue; 3043 goto unlock_exit;
3006 }
3007 } else { 3044 } else {
3008 lock_page(page); 3045 lock_page(page);
3009 } 3046 }
3047 locked_pages++;
3048 if (!PageUptodate(page)) {
3049 all_uptodate = 0;
3050 }
3051 }
3052 if (all_uptodate) {
3053 if (start_i == 0)
3054 eb->flags |= EXTENT_UPTODATE;
3055 goto unlock_exit;
3056 }
3057
3058 for (i = start_i; i < num_pages; i++) {
3059 page = extent_buffer_page(eb, i);
3060 if (inc_all_pages)
3061 page_cache_get(page);
3010 if (!PageUptodate(page)) { 3062 if (!PageUptodate(page)) {
3063 if (start_i == 0)
3064 inc_all_pages = 1;
3011 err = __extent_read_full_page(tree, page, 3065 err = __extent_read_full_page(tree, page,
3012 get_extent, &bio); 3066 get_extent, &bio);
3013 if (err) { 3067 if (err) {
@@ -3034,6 +3088,16 @@ int read_extent_buffer_pages(struct extent_io_tree *tree,
3034 if (!ret) 3088 if (!ret)
3035 eb->flags |= EXTENT_UPTODATE; 3089 eb->flags |= EXTENT_UPTODATE;
3036 return ret; 3090 return ret;
3091
3092unlock_exit:
3093 i = start_i;
3094 while(locked_pages > 0) {
3095 page = extent_buffer_page(eb, i);
3096 i++;
3097 unlock_page(page);
3098 locked_pages--;
3099 }
3100 return ret;
3037} 3101}
3038EXPORT_SYMBOL(read_extent_buffer_pages); 3102EXPORT_SYMBOL(read_extent_buffer_pages);
3039 3103
@@ -3048,7 +3112,6 @@ void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3048 char *dst = (char *)dstv; 3112 char *dst = (char *)dstv;
3049 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1); 3113 size_t start_offset = eb->start & ((u64)PAGE_CACHE_SIZE - 1);
3050 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT; 3114 unsigned long i = (start_offset + start) >> PAGE_CACHE_SHIFT;
3051 unsigned long num_pages = num_extent_pages(eb->start, eb->len);
3052 3115
3053 WARN_ON(start > eb->len); 3116 WARN_ON(start > eb->len);
3054 WARN_ON(start + len > eb->start + eb->len); 3117 WARN_ON(start + len > eb->start + eb->len);
@@ -3057,11 +3120,6 @@ void read_extent_buffer(struct extent_buffer *eb, void *dstv,
3057 3120
3058 while(len > 0) { 3121 while(len > 0) {
3059 page = extent_buffer_page(eb, i); 3122 page = extent_buffer_page(eb, i);
3060 if (!PageUptodate(page)) {
3061 printk("page %lu not up to date i %lu, total %lu, len %lu\n", page->index, i, num_pages, eb->len);
3062 WARN_ON(1);
3063 }
3064 WARN_ON(!PageUptodate(page));
3065 3123
3066 cur = min(len, (PAGE_CACHE_SIZE - offset)); 3124 cur = min(len, (PAGE_CACHE_SIZE - offset));
3067 kaddr = kmap_atomic(page, KM_USER1); 3125 kaddr = kmap_atomic(page, KM_USER1);
@@ -3105,7 +3163,6 @@ printk("bad mapping eb start %Lu len %lu, wanted %lu %lu\n", eb->start, eb->len,
3105 } 3163 }
3106 3164
3107 p = extent_buffer_page(eb, i); 3165 p = extent_buffer_page(eb, i);
3108 WARN_ON(!PageUptodate(p));
3109 kaddr = kmap_atomic(p, km); 3166 kaddr = kmap_atomic(p, km);
3110 *token = kaddr; 3167 *token = kaddr;
3111 *map = kaddr + offset; 3168 *map = kaddr + offset;
@@ -3165,7 +3222,6 @@ int memcmp_extent_buffer(struct extent_buffer *eb, const void *ptrv,
3165 3222
3166 while(len > 0) { 3223 while(len > 0) {
3167 page = extent_buffer_page(eb, i); 3224 page = extent_buffer_page(eb, i);
3168 WARN_ON(!PageUptodate(page));
3169 3225
3170 cur = min(len, (PAGE_CACHE_SIZE - offset)); 3226 cur = min(len, (PAGE_CACHE_SIZE - offset));
3171 3227
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index d9f5bc4dbd7f..9d2991d1d3ce 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -13,7 +13,6 @@
13#define EXTENT_DEFRAG (1 << 6) 13#define EXTENT_DEFRAG (1 << 6)
14#define EXTENT_DEFRAG_DONE (1 << 7) 14#define EXTENT_DEFRAG_DONE (1 << 7)
15#define EXTENT_BUFFER_FILLED (1 << 8) 15#define EXTENT_BUFFER_FILLED (1 << 8)
16#define EXTENT_CSUM (1 << 9)
17#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK) 16#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
18 17
19/* 18/*
@@ -218,4 +217,7 @@ int map_private_extent_buffer(struct extent_buffer *eb, unsigned long offset,
218void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km); 217void unmap_extent_buffer(struct extent_buffer *eb, char *token, int km);
219int invalidate_extent_lru(struct extent_io_tree *tree, u64 start, 218int invalidate_extent_lru(struct extent_io_tree *tree, u64 start,
220 unsigned long len); 219 unsigned long len);
220int release_extent_buffer_tail_pages(struct extent_buffer *eb);
221int extent_range_uptodate(struct extent_io_tree *tree,
222 u64 start, u64 end);
221#endif 223#endif
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 5e9f69244f9f..1ed179c020af 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -843,7 +843,7 @@ void btrfs_transaction_flush_work(struct btrfs_root *root)
843 843
844void __init btrfs_init_transaction_sys(void) 844void __init btrfs_init_transaction_sys(void)
845{ 845{
846 trans_wq = create_workqueue("btrfs"); 846 trans_wq = create_workqueue("btrfs-transaction");
847} 847}
848 848
849void btrfs_exit_transaction_sys(void) 849void btrfs_exit_transaction_sys(void)