From: Chris Mason <chris.mason@oracle.com>
Date: Sun, 6 Nov 2011 08:07:10 +0000 (-0500)
Subject: Merge git://git.jan-o-sch.net/btrfs-unstable into integration
X-Git-Url: http://git.openwrt.org/?a=commitdiff_plain;h=806468f8bf76a3cb2b626dd282946a6c9c0a50f0;p=openwrt%2Fstaging%2Fblogic.git
Merge git://git.jan-o-sch.net/btrfs-unstable into integration
Conflicts:
fs/btrfs/Makefile
fs/btrfs/extent_io.c
fs/btrfs/extent_io.h
fs/btrfs/scrub.c
Signed-off-by: Chris Mason <chris.mason@oracle.com>
---
806468f8bf76a3cb2b626dd282946a6c9c0a50f0
diff --cc fs/btrfs/Makefile
index bdd6fb238ce1,89b6ce3634fd..c0ddfd29c5e5
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@@ -7,7 -7,7 +7,7 @@@ btrfs-y += super.o ctree.o extent-tree.
extent_map.o sysfs.o struct-funcs.o xattr.o ordered-data.o \
extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
export.o tree-log.o free-space-cache.o zlib.o lzo.o \
- compression.o delayed-ref.o relocation.o delayed-inode.o backref.o \
- scrub.o
+ compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \
- reada.o
++ reada.o backref.o
btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o
diff --cc fs/btrfs/disk-io.c
index cedfbfb278eb,dc0343802535..0eb1f0951251
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@@ -618,37 -613,6 +618,37 @@@ out
return ret;
}
+static int btree_io_failed_hook(struct bio *failed_bio,
+ struct page *page, u64 start, u64 end,
- struct extent_state *state)
++ u64 mirror_num, struct extent_state *state)
+{
+ struct extent_io_tree *tree;
+ unsigned long len;
+ struct extent_buffer *eb;
+ struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
+
+ tree = &BTRFS_I(page->mapping->host)->io_tree;
+ if (page->private == EXTENT_PAGE_PRIVATE)
+ goto out;
+ if (!page->private)
+ goto out;
+
+ len = page->private >> 2;
+ WARN_ON(len == 0);
+
+ eb = alloc_extent_buffer(tree, start, len, page);
+ if (eb == NULL)
+ goto out;
+
+ if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
+ clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
+ btree_readahead_hook(root, eb, eb->start, -EIO);
+ }
+
+out:
+ return -EIO; /* we fixed nothing */
+}
+
static void end_workqueue_bio(struct bio *bio, int err)
{
struct end_io_wq *end_io_wq = bio->bi_private;
diff --cc fs/btrfs/extent_io.c
index c12705682c65,624ef10d36cc..1f87c4d0e7a0
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@@ -1915,11 -2093,19 +2281,19 @@@ static void end_bio_extent_readpage(str
state);
if (ret)
uptodate = 0;
+ else
+ clean_io_failure(start, page);
}
- if (!uptodate && tree->ops &&
- tree->ops->readpage_io_failed_hook) {
- ret = tree->ops->readpage_io_failed_hook(bio, page,
- start, end, state);
+ if (!uptodate) {
+ u64 failed_mirror;
+ failed_mirror = (u64)bio->bi_bdev;
+ if (tree->ops && tree->ops->readpage_io_failed_hook)
+ ret = tree->ops->readpage_io_failed_hook(
+ bio, page, start, end,
- failed_mirror, NULL);
++ failed_mirror, state);
+ else
+ ret = bio_readpage_error(bio, page, start, end,
+ failed_mirror, NULL);
if (ret == 0) {
uptodate =
test_bit(BIO_UPTODATE, &bio->bi_flags);
diff --cc fs/btrfs/extent_io.h
index 697570eed9e8,a8e20b672922..feb9be0e23bc
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@@ -17,7 -17,7 +17,8 @@@
#define EXTENT_NODATASUM (1 << 10)
#define EXTENT_DO_ACCOUNTING (1 << 11)
#define EXTENT_FIRST_DELALLOC (1 << 12)
-#define EXTENT_DAMAGED (1 << 13)
+#define EXTENT_NEED_WAIT (1 << 13)
++#define EXTENT_DAMAGED (1 << 14)
#define EXTENT_IOBITS (EXTENT_LOCKED | EXTENT_WRITEBACK)
#define EXTENT_CTLBITS (EXTENT_DO_ACCOUNTING | EXTENT_FIRST_DELALLOC)
diff --cc fs/btrfs/ioctl.c
index 8f6e14279409,7f57efa76d11..cc9893990341
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@@ -2890,6 -2856,144 +2891,144 @@@ static long btrfs_ioctl_scrub_progress(
return ret;
}
+ static long btrfs_ioctl_ino_to_path(struct btrfs_root *root, void __user *arg)
+ {
+ int ret = 0;
+ int i;
+ unsigned long rel_ptr;
+ int size;
- struct btrfs_ioctl_ino_path_args *ipa;
++ struct btrfs_ioctl_ino_path_args *ipa = NULL;
+ struct inode_fs_paths *ipath = NULL;
+ struct btrfs_path *path;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EPERM;
+
+ path = btrfs_alloc_path();
+ if (!path) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ ipa = memdup_user(arg, sizeof(*ipa));
+ if (IS_ERR(ipa)) {
+ ret = PTR_ERR(ipa);
+ ipa = NULL;
+ goto out;
+ }
+
+ size = min_t(u32, ipa->size, 4096);
+ ipath = init_ipath(size, root, path);
+ if (IS_ERR(ipath)) {
+ ret = PTR_ERR(ipath);
+ ipath = NULL;
+ goto out;
+ }
+
+ ret = paths_from_inode(ipa->inum, ipath);
+ if (ret < 0)
+ goto out;
+
+ for (i = 0; i < ipath->fspath->elem_cnt; ++i) {
+ rel_ptr = ipath->fspath->str[i] - (char *)ipath->fspath->str;
+ ipath->fspath->str[i] = (void *)rel_ptr;
+ }
+
+ ret = copy_to_user(ipa->fspath, ipath->fspath, size);
+ if (ret) {
+ ret = -EFAULT;
+ goto out;
+ }
+
+ out:
+ btrfs_free_path(path);
+ free_ipath(ipath);
+ kfree(ipa);
+
+ return ret;
+ }
+
+ static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
+ {
+ struct btrfs_data_container *inodes = ctx;
+ const size_t c = 3 * sizeof(u64);
+
+ if (inodes->bytes_left >= c) {
+ inodes->bytes_left -= c;
+ inodes->val[inodes->elem_cnt] = inum;
+ inodes->val[inodes->elem_cnt + 1] = offset;
+ inodes->val[inodes->elem_cnt + 2] = root;
+ inodes->elem_cnt += 3;
+ } else {
+ inodes->bytes_missing += c - inodes->bytes_left;
+ inodes->bytes_left = 0;
+ inodes->elem_missed += 3;
+ }
+
+ return 0;
+ }
+
+ static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
+ void __user *arg)
+ {
+ int ret = 0;
+ int size;
+ u64 extent_offset;
+ struct btrfs_ioctl_logical_ino_args *loi;
+ struct btrfs_data_container *inodes = NULL;
+ struct btrfs_path *path = NULL;
+ struct btrfs_key key;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EPERM;
+
+ loi = memdup_user(arg, sizeof(*loi));
+ if (IS_ERR(loi)) {
+ ret = PTR_ERR(loi);
+ loi = NULL;
+ goto out;
+ }
+
+ path = btrfs_alloc_path();
+ if (!path) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ size = min_t(u32, loi->size, 4096);
+ inodes = init_data_container(size);
+ if (IS_ERR(inodes)) {
+ ret = PTR_ERR(inodes);
+ inodes = NULL;
+ goto out;
+ }
+
+ ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
+
+ if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
+ ret = -ENOENT;
+ if (ret < 0)
+ goto out;
+
+ extent_offset = loi->logical - key.objectid;
+ ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
+ extent_offset, build_ino_list, inodes);
+
+ if (ret < 0)
+ goto out;
+
+ ret = copy_to_user(loi->inodes, inodes, size);
+ if (ret)
+ ret = -EFAULT;
+
+ out:
+ btrfs_free_path(path);
+ kfree(inodes);
+ kfree(loi);
+
+ return ret;
+ }
+
long btrfs_ioctl(struct file *file, unsigned int
cmd, unsigned long arg)
{
diff --cc fs/btrfs/reada.c
index 2b701d082227,000000000000..cd857119ba8a
mode 100644,000000..100644
--- a/fs/btrfs/reada.c
+++ b/fs/btrfs/reada.c
@@@ -1,949 -1,0 +1,949 @@@
+/*
+ * Copyright (C) 2011 STRATO. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/rbtree.h>
+#include <linux/slab.h>
+#include <linux/workqueue.h>
+#include "ctree.h"
+#include "volumes.h"
+#include "disk-io.h"
+#include "transaction.h"
+
+#undef DEBUG
+
+/*
+ * This is the implementation for the generic read ahead framework.
+ *
+ * To trigger a readahead, btrfs_reada_add must be called. It will start
+ * a read ahead for the given range [start, end) on tree root. The returned
+ * handle can either be used to wait on the readahead to finish
+ * (btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
+ *
+ * The read ahead works as follows:
+ * On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
+ * reada_start_machine will then search for extents to prefetch and trigger
+ * some reads. When a read finishes for a node, all contained node/leaf
+ * pointers that lie in the given range will also be enqueued. The reads will
+ * be triggered in sequential order, thus giving a big win over a naive
+ * enumeration. It will also make use of multi-device layouts. Each disk
+ * will have its on read pointer and all disks will by utilized in parallel.
+ * Also will no two disks read both sides of a mirror simultaneously, as this
+ * would waste seeking capacity. Instead both disks will read different parts
+ * of the filesystem.
+ * Any number of readaheads can be started in parallel. The read order will be
+ * determined globally, i.e. 2 parallel readaheads will normally finish faster
+ * than the 2 started one after another.
+ */
+
+#define MAX_MIRRORS 2
+#define MAX_IN_FLIGHT 6
+
+struct reada_extctl {
+ struct list_head list;
+ struct reada_control *rc;
+ u64 generation;
+};
+
+struct reada_extent {
+ u64 logical;
+ struct btrfs_key top;
+ u32 blocksize;
+ int err;
+ struct list_head extctl;
+ struct kref refcnt;
+ spinlock_t lock;
+ struct reada_zone *zones[MAX_MIRRORS];
+ int nzones;
+ struct btrfs_device *scheduled_for;
+};
+
+struct reada_zone {
+ u64 start;
+ u64 end;
+ u64 elems;
+ struct list_head list;
+ spinlock_t lock;
+ int locked;
+ struct btrfs_device *device;
+ struct btrfs_device *devs[MAX_MIRRORS]; /* full list, incl self */
+ int ndevs;
+ struct kref refcnt;
+};
+
+struct reada_machine_work {
+ struct btrfs_work work;
+ struct btrfs_fs_info *fs_info;
+};
+
+static void reada_extent_put(struct btrfs_fs_info *, struct reada_extent *);
+static void reada_control_release(struct kref *kref);
+static void reada_zone_release(struct kref *kref);
+static void reada_start_machine(struct btrfs_fs_info *fs_info);
+static void __reada_start_machine(struct btrfs_fs_info *fs_info);
+
+static int reada_add_block(struct reada_control *rc, u64 logical,
+ struct btrfs_key *top, int level, u64 generation);
+
+/* recurses */
+/* in case of err, eb might be NULL */
+static int __readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
+ u64 start, int err)
+{
+ int level = 0;
+ int nritems;
+ int i;
+ u64 bytenr;
+ u64 generation;
+ struct reada_extent *re;
+ struct btrfs_fs_info *fs_info = root->fs_info;
+ struct list_head list;
+ unsigned long index = start >> PAGE_CACHE_SHIFT;
+ struct btrfs_device *for_dev;
+
+ if (eb)
+ level = btrfs_header_level(eb);
+
+ /* find extent */
+ spin_lock(&fs_info->reada_lock);
+ re = radix_tree_lookup(&fs_info->reada_tree, index);
+ if (re)
+ kref_get(&re->refcnt);
+ spin_unlock(&fs_info->reada_lock);
+
+ if (!re)
+ return -1;
+
+ spin_lock(&re->lock);
+ /*
+ * just take the full list from the extent. afterwards we
+ * don't need the lock anymore
+ */
+ list_replace_init(&re->extctl, &list);
+ for_dev = re->scheduled_for;
+ re->scheduled_for = NULL;
+ spin_unlock(&re->lock);
+
+ if (err == 0) {
+ nritems = level ? btrfs_header_nritems(eb) : 0;
+ generation = btrfs_header_generation(eb);
+ /*
+ * FIXME: currently we just set nritems to 0 if this is a leaf,
+ * effectively ignoring the content. In a next step we could
+ * trigger more readahead depending from the content, e.g.
+ * fetch the checksums for the extents in the leaf.
+ */
+ } else {
+ /*
+ * this is the error case, the extent buffer has not been
+ * read correctly. We won't access anything from it and
+ * just cleanup our data structures. Effectively this will
+ * cut the branch below this node from read ahead.
+ */
+ nritems = 0;
+ generation = 0;
+ }
+
+ for (i = 0; i < nritems; i++) {
+ struct reada_extctl *rec;
+ u64 n_gen;
+ struct btrfs_key key;
+ struct btrfs_key next_key;
+
+ btrfs_node_key_to_cpu(eb, &key, i);
+ if (i + 1 < nritems)
+ btrfs_node_key_to_cpu(eb, &next_key, i + 1);
+ else
+ next_key = re->top;
+ bytenr = btrfs_node_blockptr(eb, i);
+ n_gen = btrfs_node_ptr_generation(eb, i);
+
+ list_for_each_entry(rec, &list, list) {
+ struct reada_control *rc = rec->rc;
+
+ /*
+ * if the generation doesn't match, just ignore this
+ * extctl. This will probably cut off a branch from
+ * prefetch. Alternatively one could start a new (sub-)
+ * prefetch for this branch, starting again from root.
+ * FIXME: move the generation check out of this loop
+ */
+#ifdef DEBUG
+ if (rec->generation != generation) {
+ printk(KERN_DEBUG "generation mismatch for "
+ "(%llu,%d,%llu) %llu != %llu\n",
+ key.objectid, key.type, key.offset,
+ rec->generation, generation);
+ }
+#endif
+ if (rec->generation == generation &&
+ btrfs_comp_cpu_keys(&key, &rc->key_end) < 0 &&
+ btrfs_comp_cpu_keys(&next_key, &rc->key_start) > 0)
+ reada_add_block(rc, bytenr, &next_key,
+ level - 1, n_gen);
+ }
+ }
+ /*
+ * free extctl records
+ */
+ while (!list_empty(&list)) {
+ struct reada_control *rc;
+ struct reada_extctl *rec;
+
+ rec = list_first_entry(&list, struct reada_extctl, list);
+ list_del(&rec->list);
+ rc = rec->rc;
+ kfree(rec);
+
+ kref_get(&rc->refcnt);
+ if (atomic_dec_and_test(&rc->elems)) {
+ kref_put(&rc->refcnt, reada_control_release);
+ wake_up(&rc->wait);
+ }
+ kref_put(&rc->refcnt, reada_control_release);
+
+ reada_extent_put(fs_info, re); /* one ref for each entry */
+ }
+ reada_extent_put(fs_info, re); /* our ref */
+ if (for_dev)
+ atomic_dec(&for_dev->reada_in_flight);
+
+ return 0;
+}
+
+/*
+ * start is passed separately in case eb in NULL, which may be the case with
+ * failed I/O
+ */
+int btree_readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
+ u64 start, int err)
+{
+ int ret;
+
+ ret = __readahead_hook(root, eb, start, err);
+
+ reada_start_machine(root->fs_info);
+
+ return ret;
+}
+
+static struct reada_zone *reada_find_zone(struct btrfs_fs_info *fs_info,
+ struct btrfs_device *dev, u64 logical,
- struct btrfs_multi_bio *multi)
++ struct btrfs_bio *multi)
+{
+ int ret;
+ int looped = 0;
+ struct reada_zone *zone;
+ struct btrfs_block_group_cache *cache = NULL;
+ u64 start;
+ u64 end;
+ int i;
+
+again:
+ zone = NULL;
+ spin_lock(&fs_info->reada_lock);
+ ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
+ logical >> PAGE_CACHE_SHIFT, 1);
+ if (ret == 1)
+ kref_get(&zone->refcnt);
+ spin_unlock(&fs_info->reada_lock);
+
+ if (ret == 1) {
+ if (logical >= zone->start && logical < zone->end)
+ return zone;
+ spin_lock(&fs_info->reada_lock);
+ kref_put(&zone->refcnt, reada_zone_release);
+ spin_unlock(&fs_info->reada_lock);
+ }
+
+ if (looped)
+ return NULL;
+
+ cache = btrfs_lookup_block_group(fs_info, logical);
+ if (!cache)
+ return NULL;
+
+ start = cache->key.objectid;
+ end = start + cache->key.offset - 1;
+ btrfs_put_block_group(cache);
+
+ zone = kzalloc(sizeof(*zone), GFP_NOFS);
+ if (!zone)
+ return NULL;
+
+ zone->start = start;
+ zone->end = end;
+ INIT_LIST_HEAD(&zone->list);
+ spin_lock_init(&zone->lock);
+ zone->locked = 0;
+ kref_init(&zone->refcnt);
+ zone->elems = 0;
+ zone->device = dev; /* our device always sits at index 0 */
+ for (i = 0; i < multi->num_stripes; ++i) {
+ /* bounds have already been checked */
+ zone->devs[i] = multi->stripes[i].dev;
+ }
+ zone->ndevs = multi->num_stripes;
+
+ spin_lock(&fs_info->reada_lock);
+ ret = radix_tree_insert(&dev->reada_zones,
+ (unsigned long)zone->end >> PAGE_CACHE_SHIFT,
+ zone);
+ spin_unlock(&fs_info->reada_lock);
+
+ if (ret) {
+ kfree(zone);
+ looped = 1;
+ goto again;
+ }
+
+ return zone;
+}
+
+static struct reada_extent *reada_find_extent(struct btrfs_root *root,
+ u64 logical,
+ struct btrfs_key *top, int level)
+{
+ int ret;
+ int looped = 0;
+ struct reada_extent *re = NULL;
+ struct btrfs_fs_info *fs_info = root->fs_info;
+ struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
- struct btrfs_multi_bio *multi = NULL;
++ struct btrfs_bio *multi = NULL;
+ struct btrfs_device *dev;
+ u32 blocksize;
+ u64 length;
+ int nzones = 0;
+ int i;
+ unsigned long index = logical >> PAGE_CACHE_SHIFT;
+
+again:
+ spin_lock(&fs_info->reada_lock);
+ re = radix_tree_lookup(&fs_info->reada_tree, index);
+ if (re)
+ kref_get(&re->refcnt);
+ spin_unlock(&fs_info->reada_lock);
+
+ if (re || looped)
+ return re;
+
+ re = kzalloc(sizeof(*re), GFP_NOFS);
+ if (!re)
+ return NULL;
+
+ blocksize = btrfs_level_size(root, level);
+ re->logical = logical;
+ re->blocksize = blocksize;
+ re->top = *top;
+ INIT_LIST_HEAD(&re->extctl);
+ spin_lock_init(&re->lock);
+ kref_init(&re->refcnt);
+
+ /*
+ * map block
+ */
+ length = blocksize;
+ ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length, &multi, 0);
+ if (ret || !multi || length < blocksize)
+ goto error;
+
+ if (multi->num_stripes > MAX_MIRRORS) {
+ printk(KERN_ERR "btrfs readahead: more than %d copies not "
+ "supported", MAX_MIRRORS);
+ goto error;
+ }
+
+ for (nzones = 0; nzones < multi->num_stripes; ++nzones) {
+ struct reada_zone *zone;
+
+ dev = multi->stripes[nzones].dev;
+ zone = reada_find_zone(fs_info, dev, logical, multi);
+ if (!zone)
+ break;
+
+ re->zones[nzones] = zone;
+ spin_lock(&zone->lock);
+ if (!zone->elems)
+ kref_get(&zone->refcnt);
+ ++zone->elems;
+ spin_unlock(&zone->lock);
+ spin_lock(&fs_info->reada_lock);
+ kref_put(&zone->refcnt, reada_zone_release);
+ spin_unlock(&fs_info->reada_lock);
+ }
+ re->nzones = nzones;
+ if (nzones == 0) {
+ /* not a single zone found, error and out */
+ goto error;
+ }
+
+ /* insert extent in reada_tree + all per-device trees, all or nothing */
+ spin_lock(&fs_info->reada_lock);
+ ret = radix_tree_insert(&fs_info->reada_tree, index, re);
+ if (ret) {
+ spin_unlock(&fs_info->reada_lock);
+ if (ret != -ENOMEM) {
+ /* someone inserted the extent in the meantime */
+ looped = 1;
+ }
+ goto error;
+ }
+ for (i = 0; i < nzones; ++i) {
+ dev = multi->stripes[i].dev;
+ ret = radix_tree_insert(&dev->reada_extents, index, re);
+ if (ret) {
+ while (--i >= 0) {
+ dev = multi->stripes[i].dev;
+ BUG_ON(dev == NULL);
+ radix_tree_delete(&dev->reada_extents, index);
+ }
+ BUG_ON(fs_info == NULL);
+ radix_tree_delete(&fs_info->reada_tree, index);
+ spin_unlock(&fs_info->reada_lock);
+ goto error;
+ }
+ }
+ spin_unlock(&fs_info->reada_lock);
+
+ return re;
+
+error:
+ while (nzones) {
+ struct reada_zone *zone;
+
+ --nzones;
+ zone = re->zones[nzones];
+ kref_get(&zone->refcnt);
+ spin_lock(&zone->lock);
+ --zone->elems;
+ if (zone->elems == 0) {
+ /*
+ * no fs_info->reada_lock needed, as this can't be
+ * the last ref
+ */
+ kref_put(&zone->refcnt, reada_zone_release);
+ }
+ spin_unlock(&zone->lock);
+
+ spin_lock(&fs_info->reada_lock);
+ kref_put(&zone->refcnt, reada_zone_release);
+ spin_unlock(&fs_info->reada_lock);
+ }
+ kfree(re);
+ if (looped)
+ goto again;
+ return NULL;
+}
+
+static void reada_kref_dummy(struct kref *kr)
+{
+}
+
+static void reada_extent_put(struct btrfs_fs_info *fs_info,
+ struct reada_extent *re)
+{
+ int i;
+ unsigned long index = re->logical >> PAGE_CACHE_SHIFT;
+
+ spin_lock(&fs_info->reada_lock);
+ if (!kref_put(&re->refcnt, reada_kref_dummy)) {
+ spin_unlock(&fs_info->reada_lock);
+ return;
+ }
+
+ radix_tree_delete(&fs_info->reada_tree, index);
+ for (i = 0; i < re->nzones; ++i) {
+ struct reada_zone *zone = re->zones[i];
+
+ radix_tree_delete(&zone->device->reada_extents, index);
+ }
+
+ spin_unlock(&fs_info->reada_lock);
+
+ for (i = 0; i < re->nzones; ++i) {
+ struct reada_zone *zone = re->zones[i];
+
+ kref_get(&zone->refcnt);
+ spin_lock(&zone->lock);
+ --zone->elems;
+ if (zone->elems == 0) {
+ /* no fs_info->reada_lock needed, as this can't be
+ * the last ref */
+ kref_put(&zone->refcnt, reada_zone_release);
+ }
+ spin_unlock(&zone->lock);
+
+ spin_lock(&fs_info->reada_lock);
+ kref_put(&zone->refcnt, reada_zone_release);
+ spin_unlock(&fs_info->reada_lock);
+ }
+ if (re->scheduled_for)
+ atomic_dec(&re->scheduled_for->reada_in_flight);
+
+ kfree(re);
+}
+
+static void reada_zone_release(struct kref *kref)
+{
+ struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);
+
+ radix_tree_delete(&zone->device->reada_zones,
+ zone->end >> PAGE_CACHE_SHIFT);
+
+ kfree(zone);
+}
+
+static void reada_control_release(struct kref *kref)
+{
+ struct reada_control *rc = container_of(kref, struct reada_control,
+ refcnt);
+
+ kfree(rc);
+}
+
+static int reada_add_block(struct reada_control *rc, u64 logical,
+ struct btrfs_key *top, int level, u64 generation)
+{
+ struct btrfs_root *root = rc->root;
+ struct reada_extent *re;
+ struct reada_extctl *rec;
+
+ re = reada_find_extent(root, logical, top, level); /* takes one ref */
+ if (!re)
+ return -1;
+
+ rec = kzalloc(sizeof(*rec), GFP_NOFS);
+ if (!rec) {
+ reada_extent_put(root->fs_info, re);
+ return -1;
+ }
+
+ rec->rc = rc;
+ rec->generation = generation;
+ atomic_inc(&rc->elems);
+
+ spin_lock(&re->lock);
+ list_add_tail(&rec->list, &re->extctl);
+ spin_unlock(&re->lock);
+
+ /* leave the ref on the extent */
+
+ return 0;
+}
+
+/*
+ * called with fs_info->reada_lock held
+ */
+static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
+{
+ int i;
+ unsigned long index = zone->end >> PAGE_CACHE_SHIFT;
+
+ for (i = 0; i < zone->ndevs; ++i) {
+ struct reada_zone *peer;
+ peer = radix_tree_lookup(&zone->devs[i]->reada_zones, index);
+ if (peer && peer->device != zone->device)
+ peer->locked = lock;
+ }
+}
+
+/*
+ * called with fs_info->reada_lock held
+ */
+static int reada_pick_zone(struct btrfs_device *dev)
+{
+ struct reada_zone *top_zone = NULL;
+ struct reada_zone *top_locked_zone = NULL;
+ u64 top_elems = 0;
+ u64 top_locked_elems = 0;
+ unsigned long index = 0;
+ int ret;
+
+ if (dev->reada_curr_zone) {
+ reada_peer_zones_set_lock(dev->reada_curr_zone, 0);
+ kref_put(&dev->reada_curr_zone->refcnt, reada_zone_release);
+ dev->reada_curr_zone = NULL;
+ }
+ /* pick the zone with the most elements */
+ while (1) {
+ struct reada_zone *zone;
+
+ ret = radix_tree_gang_lookup(&dev->reada_zones,
+ (void **)&zone, index, 1);
+ if (ret == 0)
+ break;
+ index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
+ if (zone->locked) {
+ if (zone->elems > top_locked_elems) {
+ top_locked_elems = zone->elems;
+ top_locked_zone = zone;
+ }
+ } else {
+ if (zone->elems > top_elems) {
+ top_elems = zone->elems;
+ top_zone = zone;
+ }
+ }
+ }
+ if (top_zone)
+ dev->reada_curr_zone = top_zone;
+ else if (top_locked_zone)
+ dev->reada_curr_zone = top_locked_zone;
+ else
+ return 0;
+
+ dev->reada_next = dev->reada_curr_zone->start;
+ kref_get(&dev->reada_curr_zone->refcnt);
+ reada_peer_zones_set_lock(dev->reada_curr_zone, 1);
+
+ return 1;
+}
+
+static int reada_start_machine_dev(struct btrfs_fs_info *fs_info,
+ struct btrfs_device *dev)
+{
+ struct reada_extent *re = NULL;
+ int mirror_num = 0;
+ struct extent_buffer *eb = NULL;
+ u64 logical;
+ u32 blocksize;
+ int ret;
+ int i;
+ int need_kick = 0;
+
+ spin_lock(&fs_info->reada_lock);
+ if (dev->reada_curr_zone == NULL) {
+ ret = reada_pick_zone(dev);
+ if (!ret) {
+ spin_unlock(&fs_info->reada_lock);
+ return 0;
+ }
+ }
+ /*
+ * FIXME currently we issue the reads one extent at a time. If we have
+ * a contiguous block of extents, we could also coagulate them or use
+ * plugging to speed things up
+ */
+ ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
+ dev->reada_next >> PAGE_CACHE_SHIFT, 1);
+ if (ret == 0 || re->logical >= dev->reada_curr_zone->end) {
+ ret = reada_pick_zone(dev);
+ if (!ret) {
+ spin_unlock(&fs_info->reada_lock);
+ return 0;
+ }
+ re = NULL;
+ ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
+ dev->reada_next >> PAGE_CACHE_SHIFT, 1);
+ }
+ if (ret == 0) {
+ spin_unlock(&fs_info->reada_lock);
+ return 0;
+ }
+ dev->reada_next = re->logical + re->blocksize;
+ kref_get(&re->refcnt);
+
+ spin_unlock(&fs_info->reada_lock);
+
+ /*
+ * find mirror num
+ */
+ for (i = 0; i < re->nzones; ++i) {
+ if (re->zones[i]->device == dev) {
+ mirror_num = i + 1;
+ break;
+ }
+ }
+ logical = re->logical;
+ blocksize = re->blocksize;
+
+ spin_lock(&re->lock);
+ if (re->scheduled_for == NULL) {
+ re->scheduled_for = dev;
+ need_kick = 1;
+ }
+ spin_unlock(&re->lock);
+
+ reada_extent_put(fs_info, re);
+
+ if (!need_kick)
+ return 0;
+
+ atomic_inc(&dev->reada_in_flight);
+ ret = reada_tree_block_flagged(fs_info->extent_root, logical, blocksize,
+ mirror_num, &eb);
+ if (ret)
+ __readahead_hook(fs_info->extent_root, NULL, logical, ret);
+ else if (eb)
+ __readahead_hook(fs_info->extent_root, eb, eb->start, ret);
+
+ if (eb)
+ free_extent_buffer(eb);
+
+ return 1;
+
+}
+
+static void reada_start_machine_worker(struct btrfs_work *work)
+{
+ struct reada_machine_work *rmw;
+ struct btrfs_fs_info *fs_info;
+
+ rmw = container_of(work, struct reada_machine_work, work);
+ fs_info = rmw->fs_info;
+
+ kfree(rmw);
+
+ __reada_start_machine(fs_info);
+}
+
+static void __reada_start_machine(struct btrfs_fs_info *fs_info)
+{
+ struct btrfs_device *device;
+ struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+ u64 enqueued;
+ u64 total = 0;
+ int i;
+
+ do {
+ enqueued = 0;
+ list_for_each_entry(device, &fs_devices->devices, dev_list) {
+ if (atomic_read(&device->reada_in_flight) <
+ MAX_IN_FLIGHT)
+ enqueued += reada_start_machine_dev(fs_info,
+ device);
+ }
+ total += enqueued;
+ } while (enqueued && total < 10000);
+
+ if (enqueued == 0)
+ return;
+
+ /*
+ * If everything is already in the cache, this is effectively single
+ * threaded. To a) not hold the caller for too long and b) to utilize
+ * more cores, we broke the loop above after 10000 iterations and now
+ * enqueue to workers to finish it. This will distribute the load to
+ * the cores.
+ */
+ for (i = 0; i < 2; ++i)
+ reada_start_machine(fs_info);
+}
+
+static void reada_start_machine(struct btrfs_fs_info *fs_info)
+{
+ struct reada_machine_work *rmw;
+
+ rmw = kzalloc(sizeof(*rmw), GFP_NOFS);
+ if (!rmw) {
+ /* FIXME we cannot handle this properly right now */
+ BUG();
+ }
+ rmw->work.func = reada_start_machine_worker;
+ rmw->fs_info = fs_info;
+
+ btrfs_queue_worker(&fs_info->readahead_workers, &rmw->work);
+}
+
+#ifdef DEBUG
+static void dump_devs(struct btrfs_fs_info *fs_info, int all)
+{
+ struct btrfs_device *device;
+ struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+ unsigned long index;
+ int ret;
+ int i;
+ int j;
+ int cnt;
+
+ spin_lock(&fs_info->reada_lock);
+ list_for_each_entry(device, &fs_devices->devices, dev_list) {
+ printk(KERN_DEBUG "dev %lld has %d in flight\n", device->devid,
+ atomic_read(&device->reada_in_flight));
+ index = 0;
+ while (1) {
+ struct reada_zone *zone;
+ ret = radix_tree_gang_lookup(&device->reada_zones,
+ (void **)&zone, index, 1);
+ if (ret == 0)
+ break;
+ printk(KERN_DEBUG " zone %llu-%llu elems %llu locked "
+ "%d devs", zone->start, zone->end, zone->elems,
+ zone->locked);
+ for (j = 0; j < zone->ndevs; ++j) {
+ printk(KERN_CONT " %lld",
+ zone->devs[j]->devid);
+ }
+ if (device->reada_curr_zone == zone)
+ printk(KERN_CONT " curr off %llu",
+ device->reada_next - zone->start);
+ printk(KERN_CONT "\n");
+ index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
+ }
+ cnt = 0;
+ index = 0;
+ while (all) {
+ struct reada_extent *re = NULL;
+
+ ret = radix_tree_gang_lookup(&device->reada_extents,
+ (void **)&re, index, 1);
+ if (ret == 0)
+ break;
+ printk(KERN_DEBUG
+ " re: logical %llu size %u empty %d for %lld",
+ re->logical, re->blocksize,
+ list_empty(&re->extctl), re->scheduled_for ?
+ re->scheduled_for->devid : -1);
+
+ for (i = 0; i < re->nzones; ++i) {
+ printk(KERN_CONT " zone %llu-%llu devs",
+ re->zones[i]->start,
+ re->zones[i]->end);
+ for (j = 0; j < re->zones[i]->ndevs; ++j) {
+ printk(KERN_CONT " %lld",
+ re->zones[i]->devs[j]->devid);
+ }
+ }
+ printk(KERN_CONT "\n");
+ index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+ if (++cnt > 15)
+ break;
+ }
+ }
+
+ index = 0;
+ cnt = 0;
+ while (all) {
+ struct reada_extent *re = NULL;
+
+ ret = radix_tree_gang_lookup(&fs_info->reada_tree, (void **)&re,
+ index, 1);
+ if (ret == 0)
+ break;
+ if (!re->scheduled_for) {
+ index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+ continue;
+ }
+ printk(KERN_DEBUG
+ "re: logical %llu size %u list empty %d for %lld",
+ re->logical, re->blocksize, list_empty(&re->extctl),
+ re->scheduled_for ? re->scheduled_for->devid : -1);
+ for (i = 0; i < re->nzones; ++i) {
+ printk(KERN_CONT " zone %llu-%llu devs",
+ re->zones[i]->start,
+ re->zones[i]->end);
+ for (i = 0; i < re->nzones; ++i) {
+ printk(KERN_CONT " zone %llu-%llu devs",
+ re->zones[i]->start,
+ re->zones[i]->end);
+ for (j = 0; j < re->zones[i]->ndevs; ++j) {
+ printk(KERN_CONT " %lld",
+ re->zones[i]->devs[j]->devid);
+ }
+ }
+ }
+ printk(KERN_CONT "\n");
+ index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+ }
+ spin_unlock(&fs_info->reada_lock);
+}
+#endif
+
+/*
+ * interface
+ */
+struct reada_control *btrfs_reada_add(struct btrfs_root *root,
+ struct btrfs_key *key_start, struct btrfs_key *key_end)
+{
+ struct reada_control *rc;
+ u64 start;
+ u64 generation;
+ int level;
+ struct extent_buffer *node;
+ static struct btrfs_key max_key = {
+ .objectid = (u64)-1,
+ .type = (u8)-1,
+ .offset = (u64)-1
+ };
+
+ rc = kzalloc(sizeof(*rc), GFP_NOFS);
+ if (!rc)
+ return ERR_PTR(-ENOMEM);
+
+ rc->root = root;
+ rc->key_start = *key_start;
+ rc->key_end = *key_end;
+ atomic_set(&rc->elems, 0);
+ init_waitqueue_head(&rc->wait);
+ kref_init(&rc->refcnt);
+ kref_get(&rc->refcnt); /* one ref for having elements */
+
+ node = btrfs_root_node(root);
+ start = node->start;
+ level = btrfs_header_level(node);
+ generation = btrfs_header_generation(node);
+ free_extent_buffer(node);
+
+ reada_add_block(rc, start, &max_key, level, generation);
+
+ reada_start_machine(root->fs_info);
+
+ return rc;
+}
+
+#ifdef DEBUG
+int btrfs_reada_wait(void *handle)
+{
+ struct reada_control *rc = handle;
+
+ while (atomic_read(&rc->elems)) {
+ wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
+ 5 * HZ);
+ dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
+ }
+
+ dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
+
+ kref_put(&rc->refcnt, reada_control_release);
+
+ return 0;
+}
+#else
+int btrfs_reada_wait(void *handle)
+{
+ struct reada_control *rc = handle;
+
+ while (atomic_read(&rc->elems)) {
+ wait_event(rc->wait, atomic_read(&rc->elems) == 0);
+ }
+
+ kref_put(&rc->refcnt, reada_control_release);
+
+ return 0;
+}
+#endif
+
+void btrfs_reada_detach(void *handle)
+{
+ struct reada_control *rc = handle;
+
+ kref_put(&rc->refcnt, reada_control_release);
+}
diff --cc fs/btrfs/scrub.c
index 5bc4ec827b3d,eba42e5fd5fd..94cd3a19e9c8
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@@ -178,8 -207,9 +204,9 @@@ struct scrub_dev *scrub_setup_dev(struc
sdev->first_free = 0;
sdev->curr = -1;
atomic_set(&sdev->in_flight, 0);
+ atomic_set(&sdev->fixup_cnt, 0);
atomic_set(&sdev->cancel_req, 0);
- sdev->csum_size = btrfs_super_csum_size(&fs_info->super_copy);
+ sdev->csum_size = btrfs_super_csum_size(fs_info->super_copy);
INIT_LIST_HEAD(&sdev->csum_list);
spin_lock_init(&sdev->list_lock);
@@@ -743,11 -1136,7 +1132,11 @@@ static noinline_for_stack int scrub_str
u64 physical;
u64 logical;
u64 generation;
- u64 mirror_num;
+ int mirror_num;
+ struct reada_control *reada1;
+ struct reada_control *reada2;
+ struct btrfs_key key_start;
+ struct btrfs_key key_end;
u64 increment = map->stripe_len;
u64 offset;