aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--fs/btrfs/compression.c150
-rw-r--r--fs/btrfs/ctree.h4
-rw-r--r--fs/btrfs/disk-io.c27
-rw-r--r--fs/btrfs/extent-tree.c6
-rw-r--r--fs/btrfs/extent_io.c140
-rw-r--r--fs/btrfs/extent_io.h13
-rw-r--r--fs/btrfs/file.c53
-rw-r--r--fs/btrfs/inode.c643
-rw-r--r--fs/btrfs/ordered-data.c13
-rw-r--r--fs/btrfs/super.c4
-rw-r--r--fs/btrfs/zlib.c3
11 files changed, 849 insertions, 207 deletions
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index 354913177ba..284f21025bc 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -33,6 +33,7 @@
#include <linux/writeback.h>
#include <linux/bit_spinlock.h>
#include <linux/version.h>
+#include <linux/pagevec.h>
#include "ctree.h"
#include "disk-io.h"
#include "transaction.h"
@@ -145,9 +146,9 @@ static void end_compressed_bio_read(struct bio *bio, int err)
}
/* do io completion on the original bio */
- if (cb->errors)
+ if (cb->errors) {
bio_io_error(cb->orig_bio);
- else
+ } else
bio_endio(cb->orig_bio, 0);
/* finally free the cb struct */
@@ -333,6 +334,7 @@ int btrfs_submit_compressed_write(struct inode *inode, u64 start,
}
bytes_left -= PAGE_CACHE_SIZE;
first_byte += PAGE_CACHE_SIZE;
+ cond_resched();
}
bio_get(bio);
@@ -346,6 +348,130 @@ int btrfs_submit_compressed_write(struct inode *inode, u64 start,
return 0;
}
+static noinline int add_ra_bio_pages(struct inode *inode,
+ u64 compressed_end,
+ struct compressed_bio *cb)
+{
+ unsigned long end_index;
+ unsigned long page_index;
+ u64 last_offset;
+ u64 isize = i_size_read(inode);
+ int ret;
+ struct page *page;
+ unsigned long nr_pages = 0;
+ struct extent_map *em;
+ struct address_space *mapping = inode->i_mapping;
+ struct pagevec pvec;
+ struct extent_map_tree *em_tree;
+ struct extent_io_tree *tree;
+ u64 end;
+ int misses = 0;
+
+ page = cb->orig_bio->bi_io_vec[cb->orig_bio->bi_vcnt - 1].bv_page;
+ last_offset = (page_offset(page) + PAGE_CACHE_SIZE);
+ em_tree = &BTRFS_I(inode)->extent_tree;
+ tree = &BTRFS_I(inode)->io_tree;
+
+ if (isize == 0)
+ return 0;
+
+ end_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
+
+ pagevec_init(&pvec, 0);
+ while(last_offset < compressed_end) {
+ page_index = last_offset >> PAGE_CACHE_SHIFT;
+
+ if (page_index > end_index)
+ break;
+
+ rcu_read_lock();
+ page = radix_tree_lookup(&mapping->page_tree, page_index);
+ rcu_read_unlock();
+ if (page) {
+ misses++;
+ if (misses > 4)
+ break;
+ goto next;
+ }
+
+ page = alloc_page(mapping_gfp_mask(mapping) | GFP_NOFS);
+ if (!page)
+ break;
+
+ page->index = page_index;
+ /*
+ * what we want to do here is call add_to_page_cache_lru,
+ * but that isn't exported, so we reproduce it here
+ */
+ if (add_to_page_cache(page, mapping,
+ page->index, GFP_NOFS)) {
+ page_cache_release(page);
+ goto next;
+ }
+
+ /* open coding of lru_cache_add, also not exported */
+ page_cache_get(page);
+ if (!pagevec_add(&pvec, page))
+ __pagevec_lru_add(&pvec);
+
+ end = last_offset + PAGE_CACHE_SIZE - 1;
+ /*
+ * at this point, we have a locked page in the page cache
+ * for these bytes in the file. But, we have to make
+ * sure they map to this compressed extent on disk.
+ */
+ set_page_extent_mapped(page);
+ lock_extent(tree, last_offset, end, GFP_NOFS);
+ spin_lock(&em_tree->lock);
+ em = lookup_extent_mapping(em_tree, last_offset,
+ PAGE_CACHE_SIZE);
+ spin_unlock(&em_tree->lock);
+
+ if (!em || last_offset < em->start ||
+ (last_offset + PAGE_CACHE_SIZE > extent_map_end(em)) ||
+ (em->block_start >> 9) != cb->orig_bio->bi_sector) {
+ free_extent_map(em);
+ unlock_extent(tree, last_offset, end, GFP_NOFS);
+ unlock_page(page);
+ page_cache_release(page);
+ break;
+ }
+ free_extent_map(em);
+
+ if (page->index == end_index) {
+ char *userpage;
+ size_t zero_offset = isize & (PAGE_CACHE_SIZE - 1);
+
+ if (zero_offset) {
+ int zeros;
+ zeros = PAGE_CACHE_SIZE - zero_offset;
+ userpage = kmap_atomic(page, KM_USER0);
+ memset(userpage + zero_offset, 0, zeros);
+ flush_dcache_page(page);
+ kunmap_atomic(userpage, KM_USER0);
+ }
+ }
+
+ ret = bio_add_page(cb->orig_bio, page,
+ PAGE_CACHE_SIZE, 0);
+
+ if (ret == PAGE_CACHE_SIZE) {
+ nr_pages++;
+ page_cache_release(page);
+ } else {
+ unlock_extent(tree, last_offset, end, GFP_NOFS);
+ unlock_page(page);
+ page_cache_release(page);
+ break;
+ }
+next:
+ last_offset += PAGE_CACHE_SIZE;
+ }
+ if (pagevec_count(&pvec))
+ __pagevec_lru_add(&pvec);
+ return 0;
+}
+
/*
* for a compressed read, the bio we get passed has all the inode pages
* in it. We don't actually do IO on those pages but allocate new ones
@@ -373,6 +499,7 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
struct block_device *bdev;
struct bio *comp_bio;
u64 cur_disk_byte = (u64)bio->bi_sector << 9;
+ u64 em_len;
struct extent_map *em;
int ret;
@@ -393,6 +520,7 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
cb->start = em->start;
compressed_len = em->block_len;
+ em_len = em->len;
free_extent_map(em);
cb->len = uncompressed_len;
@@ -411,6 +539,17 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
}
cb->nr_pages = nr_pages;
+ add_ra_bio_pages(inode, cb->start + em_len, cb);
+
+ if (!btrfs_test_opt(root, NODATASUM) &&
+ !btrfs_test_flag(inode, NODATASUM)) {
+ btrfs_lookup_bio_sums(root, inode, cb->orig_bio);
+ }
+
+ /* include any pages we added in add_ra-bio_pages */
+ uncompressed_len = bio->bi_vcnt * PAGE_CACHE_SIZE;
+ cb->len = uncompressed_len;
+
comp_bio = compressed_bio_alloc(bdev, cur_disk_byte, GFP_NOFS);
comp_bio->bi_private = cb;
comp_bio->bi_end_io = end_compressed_bio_read;
@@ -442,9 +581,10 @@ int btrfs_submit_compressed_read(struct inode *inode, struct bio *bio,
comp_bio = compressed_bio_alloc(bdev, cur_disk_byte,
GFP_NOFS);
atomic_inc(&cb->pending_bios);
- bio->bi_private = cb;
- bio->bi_end_io = end_compressed_bio_write;
- bio_add_page(bio, page, PAGE_CACHE_SIZE, 0);
+ comp_bio->bi_private = cb;
+ comp_bio->bi_end_io = end_compressed_bio_read;
+
+ bio_add_page(comp_bio, page, PAGE_CACHE_SIZE, 0);
}
cur_disk_byte += PAGE_CACHE_SIZE;
}
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 689df070c8e..c83cc5b2ded 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -625,8 +625,8 @@ struct btrfs_fs_info {
struct btrfs_transaction *running_transaction;
wait_queue_head_t transaction_throttle;
wait_queue_head_t transaction_wait;
- wait_queue_head_t async_submit_wait;
+ wait_queue_head_t async_submit_wait;
wait_queue_head_t tree_log_wait;
struct btrfs_super_block super_copy;
@@ -653,6 +653,7 @@ struct btrfs_fs_info {
atomic_t nr_async_submits;
atomic_t async_submit_draining;
atomic_t nr_async_bios;
+ atomic_t async_delalloc_pages;
atomic_t tree_log_writers;
atomic_t tree_log_commit;
unsigned long tree_log_batch;
@@ -677,6 +678,7 @@ struct btrfs_fs_info {
* two
*/
struct btrfs_workers workers;
+ struct btrfs_workers delalloc_workers;
struct btrfs_workers endio_workers;
struct btrfs_workers endio_write_workers;
struct btrfs_workers submit_workers;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index e0a28f705a6..8efc123d222 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -539,6 +539,13 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
(atomic_read(&fs_info->nr_async_bios) < limit),
HZ/10);
}
+
+ while(atomic_read(&fs_info->async_submit_draining) &&
+ atomic_read(&fs_info->nr_async_submits)) {
+ wait_event(fs_info->async_submit_wait,
+ (atomic_read(&fs_info->nr_async_submits) == 0));
+ }
+
return 0;
}
@@ -1437,6 +1444,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
INIT_LIST_HEAD(&fs_info->space_info);
btrfs_mapping_init(&fs_info->mapping_tree);
atomic_set(&fs_info->nr_async_submits, 0);
+ atomic_set(&fs_info->async_delalloc_pages, 0);
atomic_set(&fs_info->async_submit_draining, 0);
atomic_set(&fs_info->nr_async_bios, 0);
atomic_set(&fs_info->throttles, 0);
@@ -1550,6 +1558,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
btrfs_init_workers(&fs_info->workers, "worker",
fs_info->thread_pool_size);
+ btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
+ fs_info->thread_pool_size);
+
btrfs_init_workers(&fs_info->submit_workers, "submit",
min_t(u64, fs_devices->num_devices,
fs_info->thread_pool_size));
@@ -1560,15 +1571,12 @@ struct btrfs_root *open_ctree(struct super_block *sb,
*/
fs_info->submit_workers.idle_thresh = 64;
- /* fs_info->workers is responsible for checksumming file data
- * blocks and metadata. Using a larger idle thresh allows each
- * worker thread to operate on things in roughly the order they
- * were sent by the writeback daemons, improving overall locality
- * of the IO going down the pipe.
- */
- fs_info->workers.idle_thresh = 8;
+ fs_info->workers.idle_thresh = 16;
fs_info->workers.ordered = 1;
+ fs_info->delalloc_workers.idle_thresh = 2;
+ fs_info->delalloc_workers.ordered = 1;
+
btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
btrfs_init_workers(&fs_info->endio_workers, "endio",
fs_info->thread_pool_size);
@@ -1584,6 +1592,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
btrfs_start_workers(&fs_info->workers, 1);
btrfs_start_workers(&fs_info->submit_workers, 1);
+ btrfs_start_workers(&fs_info->delalloc_workers, 1);
btrfs_start_workers(&fs_info->fixup_workers, 1);
btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
btrfs_start_workers(&fs_info->endio_write_workers,
@@ -1732,6 +1741,7 @@ fail_tree_root:
fail_sys_array:
fail_sb_buffer:
btrfs_stop_workers(&fs_info->fixup_workers);
+ btrfs_stop_workers(&fs_info->delalloc_workers);
btrfs_stop_workers(&fs_info->workers);
btrfs_stop_workers(&fs_info->endio_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
@@ -1988,6 +1998,7 @@ int close_ctree(struct btrfs_root *root)
truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
btrfs_stop_workers(&fs_info->fixup_workers);
+ btrfs_stop_workers(&fs_info->delalloc_workers);
btrfs_stop_workers(&fs_info->workers);
btrfs_stop_workers(&fs_info->endio_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
@@ -2062,7 +2073,7 @@ void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
struct extent_io_tree *tree;
u64 num_dirty;
u64 start = 0;
- unsigned long thresh = 96 * 1024 * 1024;
+ unsigned long thresh = 32 * 1024 * 1024;
tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
if (current_is_pdflush() || current->flags & PF_MEMALLOC)
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 8af39521eb7..ebd8275a193 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -768,7 +768,11 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
l = path->nodes[0];
btrfs_item_key_to_cpu(l, &key, path->slots[0]);
- BUG_ON(key.objectid != bytenr);
+ if (key.objectid != bytenr) {
+ btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
+ printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
+ BUG();
+ }
BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index 9b37ce6e516..bbe3bcfcf4a 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -47,6 +47,11 @@ struct extent_page_data {
struct bio *bio;
struct extent_io_tree *tree;
get_extent_t *get_extent;
+
+ /* tells writepage not to lock the state bits for this range
+ * it still does the unlocking
+ */
+ int extent_locked;
};
int __init extent_io_init(void)
@@ -1198,11 +1203,18 @@ static noinline int lock_delalloc_pages(struct inode *inode,
* the caller is taking responsibility for
* locked_page
*/
- if (pages[i] != locked_page)
+ if (pages[i] != locked_page) {
lock_page(pages[i]);
+ if (pages[i]->mapping != inode->i_mapping) {
+ ret = -EAGAIN;
+ unlock_page(pages[i]);
+ page_cache_release(pages[i]);
+ goto done;
+ }
+ }
page_cache_release(pages[i]);
+ pages_locked++;
}
- pages_locked += ret;
nrpages -= ret;
index += ret;
cond_resched();
@@ -1262,8 +1274,7 @@ again:
* if we're looping.
*/
if (delalloc_end + 1 - delalloc_start > max_bytes && loops) {
- delalloc_end = (delalloc_start + PAGE_CACHE_SIZE - 1) &
- ~((u64)PAGE_CACHE_SIZE - 1);
+ delalloc_end = delalloc_start + PAGE_CACHE_SIZE - 1;
}
/* step two, lock all the pages after the page that has start */
ret = lock_delalloc_pages(inode, locked_page,
@@ -1306,7 +1317,10 @@ out_failed:
int extent_clear_unlock_delalloc(struct inode *inode,
struct extent_io_tree *tree,
u64 start, u64 end, struct page *locked_page,
- int clear_dirty, int set_writeback,
+ int unlock_pages,
+ int clear_unlock,
+ int clear_delalloc, int clear_dirty,
+ int set_writeback,
int end_writeback)
{
int ret;
@@ -1315,12 +1329,19 @@ int extent_clear_unlock_delalloc(struct inode *inode,
unsigned long end_index = end >> PAGE_CACHE_SHIFT;
unsigned long nr_pages = end_index - index + 1;
int i;
- int clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC;
+ int clear_bits = 0;
+ if (clear_unlock)
+ clear_bits |= EXTENT_LOCKED;
if (clear_dirty)
clear_bits |= EXTENT_DIRTY;
+ if (clear_delalloc)
+ clear_bits |= EXTENT_DELALLOC;
+
clear_extent_bit(tree, start, end, clear_bits, 1, 0, GFP_NOFS);
+ if (!(unlock_pages || clear_dirty || set_writeback || end_writeback))
+ return 0;
while(nr_pages > 0) {
ret = find_get_pages_contig(inode->i_mapping, index,
@@ -1336,7 +1357,8 @@ int extent_clear_unlock_delalloc(struct inode *inode,
set_page_writeback(pages[i]);
if (end_writeback)
end_page_writeback(pages[i]);
- unlock_page(pages[i]);
+ if (unlock_pages)
+ unlock_page(pages[i]);
page_cache_release(pages[i]);
}
nr_pages -= ret;
@@ -1741,9 +1763,10 @@ static void end_bio_extent_readpage(struct bio *bio, int err)
}
}
- if (uptodate)
+ if (uptodate) {
set_extent_uptodate(tree, start, end,
GFP_ATOMIC);
+ }
unlock_extent(tree, start, end, GFP_ATOMIC);
if (whole_page) {
@@ -1925,6 +1948,7 @@ void set_page_extent_mapped(struct page *page)
set_page_private(page, EXTENT_PAGE_PRIVATE);
}
}
+EXPORT_SYMBOL(set_page_extent_mapped);
void set_page_extent_head(struct page *page, unsigned long len)
{
@@ -2143,12 +2167,17 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
u64 delalloc_end;
int page_started;
int compressed;
+ unsigned long nr_written = 0;
WARN_ON(!PageLocked(page));
pg_offset = i_size & (PAGE_CACHE_SIZE - 1);
if (page->index > end_index ||
(page->index == end_index && !pg_offset)) {
- page->mapping->a_ops->invalidatepage(page, 0);
+ if (epd->extent_locked) {
+ if (tree->ops && tree->ops->writepage_end_io_hook)
+ tree->ops->writepage_end_io_hook(page, start,
+ page_end, NULL, 1);
+ }
unlock_page(page);
return 0;
}
@@ -2169,27 +2198,33 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
delalloc_start = start;
delalloc_end = 0;
page_started = 0;
- while(delalloc_end < page_end) {
- nr_delalloc = find_lock_delalloc_range(inode, tree,
+ if (!epd->extent_locked) {
+ while(delalloc_end < page_end) {
+ nr_delalloc = find_lock_delalloc_range(inode, tree,
page,
&delalloc_start,
&delalloc_end,
128 * 1024 * 1024);
- if (nr_delalloc == 0) {
+ if (nr_delalloc == 0) {
+ delalloc_start = delalloc_end + 1;
+ continue;
+ }
+ tree->ops->fill_delalloc(inode, page, delalloc_start,
+ delalloc_end, &page_started,
+ &nr_written);
delalloc_start = delalloc_end + 1;
- continue;
}
- tree->ops->fill_delalloc(inode, page, delalloc_start,
- delalloc_end, &page_started);
- delalloc_start = delalloc_end + 1;
- }
- /* did the fill delalloc function already unlock and start the IO? */
- if (page_started) {
- return 0;
+ /* did the fill delalloc function already unlock and start
+ * the IO?
+ */
+ if (page_started) {
+ ret = 0;
+ goto update_nr_written;
+ }
}
-
lock_extent(tree, start, page_end, GFP_NOFS);
+
unlock_start = start;
if (tree->ops && tree->ops->writepage_start_hook) {
@@ -2199,10 +2234,13 @@ static int __extent_writepage(struct page *page, struct writeback_control *wbc,
unlock_extent(tree, start, page_end, GFP_NOFS);
redirty_page_for_writepage(wbc, page);
unlock_page(page);
- return 0;
+ ret = 0;
+ goto update_nr_written;
}
}
+ nr_written++;
+
end = page_end;
if (test_range_bit(tree, start, page_end, EXTENT_DELALLOC, 0)) {
printk("found delalloc bits after lock_extent\n");
@@ -2333,6 +2371,12 @@ done:
if (unlock_start <= page_end)
unlock_extent(tree, unlock_start, page_end, GFP_NOFS);
unlock_page(page);
+
+update_nr_written:
+ wbc->nr_to_write -= nr_written;
+ if (wbc->range_cyclic || (wbc->nr_to_write > 0 &&
+ wbc->range_start == 0 && wbc->range_end == LLONG_MAX))
+ page->mapping->writeback_index = page->index + nr_written;
return 0;
}
@@ -2431,7 +2475,7 @@ retry:
unlock_page(page);
ret = 0;
}
- if (ret || (--(wbc->nr_to_write) <= 0))
+ if (ret || wbc->nr_to_write <= 0)
done = 1;
if (wbc->nonblocking && bdi_write_congested(bdi)) {
wbc->encountered_congestion = 1;
@@ -2452,6 +2496,8 @@ retry:
}
if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
mapping->writeback_index = index;
+ if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
+ range_whole = 1;
if (wbc->range_cont)
wbc->range_start = index << PAGE_CACHE_SHIFT;
@@ -2469,6 +2515,7 @@ int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
.bio = NULL,
.tree = tree,
.get_extent = get_extent,
+ .extent_locked = 0,
};
struct writeback_control wbc_writepages = {
.bdi = wbc->bdi,
@@ -2491,6 +2538,52 @@ int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
}
EXPORT_SYMBOL(extent_write_full_page);
+int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
+ u64 start, u64 end, get_extent_t *get_extent,
+ int mode)
+{
+ int ret = 0;
+ struct address_space *mapping = inode->i_mapping;
+ struct page *page;
+ unsigned long nr_pages = (end - start + PAGE_CACHE_SIZE) >>
+ PAGE_CACHE_SHIFT;
+
+ struct extent_page_data epd = {
+ .bio = NULL,
+ .tree = tree,
+ .get_extent = get_extent,
+ .extent_locked = 1,
+ };
+ struct writeback_control wbc_writepages = {
+ .bdi = inode->i_mapping->backing_dev_info,
+ .sync_mode = mode,
+ .older_than_this = NULL,
+ .nr_to_write = nr_pages * 2,
+ .range_start = start,
+ .range_end = end + 1,
+ };
+
+ while(start <= end) {
+ page = find_get_page(mapping, start >> PAGE_CACHE_SHIFT);
+ if (clear_page_dirty_for_io(page))
+ ret = __extent_writepage(page, &wbc_writepages, &epd);
+ else {
+ if (tree->ops && tree->ops->writepage_end_io_hook)
+ tree->ops->writepage_end_io_hook(page, start,
+ start + PAGE_CACHE_SIZE - 1,
+ NULL, 1);
+ unlock_page(page);
+ }
+ page_cache_release(page);
+ start += PAGE_CACHE_SIZE;
+ }
+
+ if (epd.bio)
+ submit_one_bio(WRITE, epd.bio, 0, 0);
+ return ret;
+}
+EXPORT_SYMBOL(extent_write_locked_range);
+
int extent_writepages(struct extent_io_tree *tree,
struct address_space *mapping,
@@ -2502,6 +2595,7 @@ int extent_writepages(struct extent_io_tree *tree,
.bio = NULL,
.tree = tree,
.get_extent = get_extent,
+ .extent_locked = 0,
};
ret = extent_write_cache_pages(tree, mapping, wbc,
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 283110ec4ee..2d5f67065b6 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -35,7 +35,8 @@ typedef int (extent_submit_bio_hook_t)(struct inode *inode, int rw,
unsigned long bio_flags);
struct extent_io_ops {
int (*fill_delalloc)(struct inode *inode, struct page *locked_page,
- u64 start, u64 end, int *page_started);
+ u64 start, u64 end, int *page_started,
+ unsigned long *nr_written);
int (*writepage_start_hook)(struct page *page, u64 start, u64 end);
int (*writepage_io_hook)(struct page *page, u64 start, u64 end);
extent_submit_bio_hook_t *submit_bio_hook;
@@ -172,6 +173,9 @@ int extent_invalidatepage(struct extent_io_tree *tree,
int extent_write_full_page(struct extent_io_tree *tree, struct page *page,
get_extent_t *get_extent,
struct writeback_control *wbc);
+int extent_write_locked_range(struct extent_io_tree *tree, struct inode *inode,
+ u64 start, u64 end, get_extent_t *get_extent,
+ int mode);
int extent_writepages(struct extent_io_tree *tree,
struct address_space *mapping,
get_extent_t *get_extent,
@@ -256,6 +260,9 @@ int extent_range_uptodate(struct extent_io_tree *tree,
int extent_clear_unlock_delalloc(struct inode *inode,
struct extent_io_tree *tree,
u64 start, u64 end, struct page *locked_page,
- int clear_dirty, int set_writeback,
- int clear_writeback);
+ int unlock_page,
+ int clear_unlock,
+ int clear_delalloc, int clear_dirty,
+ int set_writeback,
+ int end_writeback);
#endif
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 0c8cc35a8b9..337221ecca2 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -368,6 +368,8 @@ int noinline btrfs_drop_extents(struct btrfs_trans_handle *trans,
u64 search_start = start;
u64 leaf_start;
u64 ram_bytes = 0;
+ u64 orig_parent = 0;
+ u64 disk_bytenr = 0;
u8 compression;
u8 encryption;
u16 other_encoding = 0;
@@ -500,17 +502,31 @@ next_slot:
keep = 1;
}
- if (bookend && found_extent && locked_end < extent_end) {
- ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
- locked_end, extent_end - 1, GFP_NOFS);
- if (!ret) {
- btrfs_release_path(root, path);
- lock_extent(&BTRFS_I(inode)->io_tree,
- locked_end, extent_end - 1, GFP_NOFS);
+ if (bookend && found_extent) {
+ if (locked_end < extent_end) {
+ ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
+ locked_end, extent_end - 1,
+ GFP_NOFS);
+ if (!ret) {
+ btrfs_release_path(root, path);
+ lock_extent(&BTRFS_I(inode)->io_tree,
+ locked_end, extent_end - 1,
+ GFP_NOFS);
+ locked_end = extent_end;
+ continue;
+ }
locked_end = extent_end;
- continue;
}
- locked_end = extent_end;
+ orig_parent = path->nodes[0]->start;
+ disk_bytenr = le64_to_cpu(old.disk_bytenr);
+ if (disk_bytenr != 0) {
+ ret = btrfs_inc_extent_ref(trans, root,
+ disk_bytenr,
+ le64_to_cpu(old.disk_num_bytes),
+ orig_parent, root->root_key.objectid,
+ trans->transid, inode->i_ino);
+ BUG_ON(ret);
+ }
}
if (found_inline) {
@@ -537,8 +553,12 @@ next_slot:
inode_sub_bytes(inode, old_num -
new_num);
}
- btrfs_set_file_extent_num_bytes(leaf, extent,
- new_num);
+ if (!compression && !encryption) {
+ btrfs_set_file_extent_ram_bytes(leaf,
+ extent, new_num);
+ }
+ btrfs_set_file_extent_num_bytes(leaf,
+ extent, new_num);
btrfs_mark_buffer_dirty(leaf);
} else if (key.offset < inline_limit &&
(end > extent_end) &&
@@ -582,11 +602,11 @@ next_slot:
}
/* create bookend, splitting the extent in two */
if (bookend && found_extent) {
- u64 disk_bytenr;
struct btrfs_key ins;
ins.objectid = inode->i_ino;
ins.offset = end;
btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
+
btrfs_release_path(root, path);
ret = btrfs_insert_empty_item(trans, root, path, &ins,
sizeof(*extent));
@@ -623,14 +643,13 @@ next_slot:
btrfs_mark_buffer_dirty(path->nodes[0]);
- disk_bytenr = le64_to_cpu(old.disk_bytenr);
if (disk_bytenr != 0) {
- ret = btrfs_inc_extent_ref(trans, root,
- disk_bytenr,
- le64_to_cpu(old.disk_num_bytes),
- leaf->start,
+ ret = btrfs_update_extent_ref(trans, root,
+ disk_bytenr, orig_parent,
+ leaf->start,
root->root_key.objectid,
trans->transid, ins.objectid);
+
BUG_ON(ret);
}
btrfs_release_path(root, path);
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 3df0ffad976..e01c0d0310a 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -86,6 +86,10 @@ static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
static void btrfs_truncate(struct inode *inode);
static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
+static noinline int cow_file_range(struct inode *inode,
+ struct page *locked_page,
+ u64 start, u64 end, int *page_started,
+ unsigned long *nr_written, int unlock);
/*
* a very lame attempt at stopping writes when the FS is 85% full. There
@@ -262,35 +266,72 @@ static int cow_file_range_inline(struct btrfs_trans_handle *trans,
return 0;
}
+struct async_extent {
+ u64 start;
+ u64 ram_size;
+ u64 compressed_size;
+ struct page **pages;
+ unsigned long nr_pages;
+ struct list_head list;
+};
+
+struct async_cow {
+ struct inode *inode;
+ struct btrfs_root *root;
+ struct page *locked_page;
+ u64 start;
+ u64 end;
+ struct list_head extents;
+ struct btrfs_work work;
+};
+
+static noinline int add_async_extent(struct async_cow *cow,
+ u64 start, u64 ram_size,
+ u64 compressed_size,
+ struct page **pages,
+ unsigned long nr_pages)
+{
+ struct async_extent *async_extent;
+
+ async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
+ async_extent->start = start;
+ async_extent->ram_size = ram_size;
+ async_extent->compressed_size = compressed_size;
+ async_extent->pages = pages;
+ async_extent->nr_pages = nr_pages;
+ list_add_tail(&async_extent->list, &cow->extents);
+ return 0;
+}
+
/*
- * when extent_io.c finds a delayed allocation range in the file,
- * the call backs end up in this code. The basic idea is to
- * allocate extents on disk for the range, and create ordered data structs
- * in ram to track those extents.
+ * we create compressed extents in two phases. The first
+ * phase compresses a range of pages that have already been
+ * locked (both pages and state bits are locked).
*
- * locked_page is the page that writepage had locked already. We use
- * it to make sure we don't do extra locks or unlocks.
+ * This is done inside an ordered work queue, and the compression
+ * is spread across many cpus. The actual IO submission is step
+ * two, and the ordered work queue takes care of making sure that
+ * happens in the same order things were put onto the queue by
+ * writepages and friends.
*
- * *page_started is set to one if we unlock locked_page and do everything
- * required to start IO on it. It may be clean and already done with
- * IO when we return.
+ * If this code finds it can't get good compression, it puts an
+ * entry onto the work queue to write the uncompressed bytes. This
+ * makes sure that both compressed inodes and uncompressed inodes
+ * are written in the same order that pdflush sent them down.
*/
-static int cow_file_range(struct inode *inode, struct page *locked_page,
- u64 start, u64 end, int *page_started)
+static noinline int compress_file_range(struct inode *inode,
+ struct page *locked_page,
+ u64 start, u64 end,
+ struct async_cow *async_cow,
+ int *num_added)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
- u64 alloc_hint = 0;
u64 num_bytes;
- unsigned long ram_size;
u64 orig_start;
u64 disk_num_bytes;
- u64 cur_alloc_size;
u64 blocksize = root->sectorsize;
u64 actual_end;
- struct btrfs_key ins;
- struct extent_map *em;
- struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
int ret = 0;
struct page **pages = NULL;
unsigned long nr_pages;
@@ -298,22 +339,12 @@ static int cow_file_range(struct inode *inode, struct page *locked_page,
unsigned long total_compressed = 0;
unsigned long total_in = 0;
unsigned long max_compressed = 128 * 1024;
- unsigned long max_uncompressed = 256 * 1024;
+ unsigned long max_uncompressed = 128 * 1024;
int i;
- int ordered_type;
int will_compress;
- trans = btrfs_join_transaction(root, 1);
- BUG_ON(!trans);
- btrfs_set_trans_block_group(trans, inode);
orig_start = start;
- /*
- * compression made this loop a bit ugly, but the basic idea is to
- * compress some pages but keep the total size of the compressed
- * extent relatively small. If compression is off, this goto target
- * is never used.
- */
again:
will_compress = 0;
nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
@@ -324,7 +355,13 @@ again:
/* we want to make sure that amount of ram required to uncompress
* an extent is reasonable, so we limit the total size in ram
- * of a compressed extent to 256k
+ * of a compressed extent to 128k. This is a crucial number
+ * because it also controls how easily we can spread reads across
+ * cpus for decompression.
+ *
+ * We also want to make sure the amount of IO required to do
+ * a random read is reasonably small, so we limit the size of
+ * a compressed extent to 128k.
*/
total_compressed = min(total_compressed, max_uncompressed);
num_bytes = (end - start + blocksize) & ~(blocksize - 1);
@@ -333,18 +370,16 @@ again:
total_in = 0;
ret = 0;
- /* we do compression for mount -o compress and when the
- * inode has not been flagged as nocompress
+ /*
+ * we do compression for mount -o compress and when the
+ * inode has not been flagged as nocompress. This flag can
+ * change at any time if we discover bad compression ratios.
*/
if (!btrfs_test_flag(inode, NOCOMPRESS) &&
btrfs_test_opt(root, COMPRESS)) {
WARN_ON(pages);
pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
- /* we want to make sure the amount of IO required to satisfy
- * a random read is reasonably small, so we limit the size
- * of a compressed extent to 128k
- */
ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
total_compressed, pages,
nr_pages, &nr_pages_ret,
@@ -371,26 +406,34 @@ again:
}
}
if (start == 0) {
+ trans = btrfs_join_transaction(root, 1);
+ BUG_ON(!trans);
+ btrfs_set_trans_block_group(trans, inode);
+
/* lets try to make an inline extent */
- if (ret || total_in < (end - start + 1)) {
+ if (ret || total_in < (actual_end - start)) {
/* we didn't compress the entire range, try
- * to make an uncompressed inline extent. This
- * is almost sure to fail, but maybe inline sizes
- * will get bigger later
+ * to make an uncompressed inline extent.
*/
ret = cow_file_range_inline(trans, root, inode,
start, end, 0, NULL);
} else {
+ /* try making a compressed inline extent */
ret = cow_file_range_inline(trans, root, inode,
start, end,
total_compressed, pages);
}
+ btrfs_end_transaction(trans, root);
if (ret == 0) {
+ /*
+ * inline extent creation worked, we don't need
+ * to create any more async work items. Unlock
+ * and free up our temp pages.
+ */
extent_clear_unlock_delalloc(inode,
&BTRFS_I(inode)->io_tree,
- start, end, NULL,
- 1, 1, 1);
- *page_started = 1;
+ start, end, NULL, 1, 0,
+ 0, 1, 1, 1);
ret = 0;
goto free_pages_out;
}
@@ -435,53 +478,280 @@ again:
/* flag the file so we don't compress in the future */
btrfs_set_flag(inode, NOCOMPRESS);
}
+ if (will_compress) {
+ *num_added += 1;
- BUG_ON(disk_num_bytes >
- btrfs_super_total_bytes(&root->fs_info->super_copy));
+ /* the async work queues will take care of doing actual
+ * allocation on disk for these compressed pages,
+ * and will submit them to the elevator.
+ */
+ add_async_extent(async_cow, start, num_bytes,
+ total_compressed, pages, nr_pages_ret);
- btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
+ if (start + num_bytes < end) {
+ start += num_bytes;
+ pages = NULL;
+ cond_resched();
+ goto again;
+ }
+ } else {
+ /*
+ * No compression, but we still need to write the pages in
+ * the file we've been given so far. redirty the locked
+ * page if it corresp