Miao Xie reported the following issue:

The filesystem was corrupted after we did a device replace.

Steps to reproduce:
 # mkfs.btrfs -f -m single -d raid10 <device0>..<device3>
 # mount <device0> <mnt>
 # btrfs replace start -rfB 1 <device4> <mnt>
 # umount <mnt>
 # btrfsck <device4>

The reason for the issue is that we changed the write offset by mistake,
introduced by commit 625f1c8d.

We read the data from the source device at first, and then write the
data into the corresponding place of the new device. In order to
implement the "-r" option, the source location is remapped using
btrfs_map_block(). The read takes place on the mapped location, and
the write needs to take place on the unmapped location. Currently
the write is using the mapped location, and this commit changes it
back by undoing the change to the write address that the aforementioned
commit added by mistake.
Reported-by: NMiao Xie <miaox@cn.fujitsu.com>
Cc: <stable@vger.kernel.org> # 3.10+
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

- It makes no sense that we deal with a inode in the dead tree.
- fix the race between dio and page copy by waiting the dio completion
- avoid the page copy vs truncate/punch hole
- check if the page is in the page cache or not
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

- It make no sense that we continue to do something after the error
  happened, just go back with this patch.
- remove some check of copy_nocow_pages_for_inode(), such as page check
  after write, inode check in the end of the function, because we are
  sure they exist.
- remove the unnecessary goto in the return value check of the write
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

We get oops while running btrfs replace start test,
------------[ cut here ]------------
kernel BUG at mm/filemap.c:608!
[SNIP]
Call Trace:
  [<ffffffffa04b36c7>] copy_nocow_pages_for_inode+0x217/0x3f0 [btrfs]
  [<ffffffffa04b34b0>] ? scrub_print_warning_inode+0x230/0x230 [btrfs]
  [<ffffffffa04b34b0>] ? scrub_print_warning_inode+0x230/0x230 [btrfs]
  [<ffffffffa04bb8ce>] iterate_extent_inodes+0x1ae/0x300 [btrfs]
  [<ffffffffa04bbab2>] iterate_inodes_from_logical+0x92/0xb0 [btrfs]
  [<ffffffffa04b34b0>] ? scrub_print_warning_inode+0x230/0x230 [btrfs]
  [<ffffffffa04b3b07>] copy_nocow_pages_worker+0x97/0x150 [btrfs]
  [<ffffffffa048eed4>] worker_loop+0x134/0x540 [btrfs]
  [<ffffffff816274ea>] ? __schedule+0x3ca/0x7f0
  [<ffffffffa048eda0>] ? btrfs_queue_worker+0x300/0x300 [btrfs]
  [<ffffffff8106f2f0>] kthread+0xc0/0xd0
  [<ffffffff8106f230>] ? flush_kthread_worker+0x80/0x80
  [<ffffffff8163181c>] ret_from_fork+0x7c/0xb0
  [<ffffffff8106f230>] ? flush_kthread_worker+0x80/0x80
[SNIP]
 RIP  [<ffffffff8111f4c5>] unlock_page+0x35/0x40
  RSP <ffff88010316bb98>
 ---[ end trace 421e79ad0dd72c7d ]---

it is because we forgot to lock the page again after we read data to
the page. Fix it.
Signed-off-by: NLin Feng <linfeng@cn.fujitsu.com>
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Using the structure btrfs_sector_sum to keep the checksum value is
unnecessary, because the extents that btrfs_sector_sum points to are
continuous, we can find out the expected checksums by btrfs_ordered_sum's
bytenr and the offset, so we can remove btrfs_sector_sum's bytenr. After
removing bytenr, there is only one member in the structure, so it makes
no sense to keep the structure, just remove it, and use a u32 array to
store the checksum value.

By this change, we don't use the while loop to get the checksums one by
one. Now, we can get several checksum value at one time, it improved the
performance by ~74% on my SSD (31MB/s -> 54MB/s).

test command:
 # dd if=/dev/zero of=/mnt/btrfs/file0 bs=1M count=1024 oflag=sync
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

A user reported scrub taking up an unreasonable amount of ram as it ran.  This
is because we lookup the csums for the extent we're scrubbing but don't free it
up until after we're done with the scrub, which means we can take up a whole lot
of ram.  This patch fixes this by dropping the csums once we're done with the
extent we've scrubbed.  The user reported this to fix their problem.  Thanks,
Reported-and-tested-by: NRemco Hosman <remco@hosman.xs4all.nl>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Btrfs has been pointer tagging bi_private and using bi_bdev
to store the stripe index and mirror number of failed IOs.

As bios bubble back up through the call chain, we use these
to decide if and how to retry our IOs.  They are also used
to count IO failures on a per device basis.

Recently a bio tracepoint was added lead to crashes because
we were abusing bi_bdev.

This commit adds a btrfs bioset, and creates explicit fields
for the mirror number and stripe index.  The plan is to
extend this structure for all of the fields currently in
struct btrfs_bio, which will mean one less kmalloc in
our IO path.
Signed-off-by: NChris Mason <chris.mason@fusionio.com>
Reported-by: NTejun Heo <tj@kernel.org>

1) Right now scrub_stripe() is looping in some unnecessary cases:
* when the found extent item's objectid has been out of the dev extent's range
  but we haven't finish scanning all the range within the dev extent
* when all the items has been processed but we haven't finish scanning all the
  range within the dev extent

In both cases, we can just finish the loop to save costs.

2) Besides, when the found extent item's length is larger than the stripe
len(64k), we don't have to release the path and search again as it'll get at the
same key used in the last loop, we can instead increase the logical cursor in
place till all space of the extent is scanned.

3) And we use 0 as the key's offset to search btree, then get to previous item
to find a smaller item, and again have to move to the next one to get the right
item.  Setting offset=-1 and previous_item() is the correct way.

4) As we won't find any checksum at offset unless this 'offset' is in a data
extent, we can just find checksum when we're really going to scrub an extent.
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Big patch, but all it does is add statics to functions which
are in fact static, then remove the associated dead-code fallout.

removed functions:

btrfs_iref_to_path()
__btrfs_lookup_delayed_deletion_item()
__btrfs_search_delayed_insertion_item()
__btrfs_search_delayed_deletion_item()
find_eb_for_page()
btrfs_find_block_group()
range_straddles_pages()
extent_range_uptodate()
btrfs_file_extent_length()
btrfs_scrub_cancel_devid()
btrfs_start_transaction_lflush()

btrfs_print_tree() is left because it is used for debugging.
btrfs_start_transaction_lflush() and btrfs_reada_detach() are
left for symmetry.

ulist.c functions are left, another patch will take care of those.
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

We currently store the first key of the tree block inside the reference for the
tree block in the extent tree.  This takes up quite a bit of space.  Make a new
key type for metadata which holds the level as the offset and completely removes
storing the btrfs_tree_block_info inside the extent ref.  This reduces the size
from 51 bytes to 33 bytes per extent reference for each tree block.  In practice
this results in a 30-35% decrease in the size of our extent tree, which means we
COW less and can keep more of the extent tree in memory which makes our heavy
metadata operations go much faster.  This is not an automatic format change, you
must enable it at mkfs time or with btrfstune.  This patch deals with having
metadata stored as either the old format or the new format so it is easy to
convert.  Thanks,
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

Argument 'root' is no more used in btrfs_csum_data().
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

A user reported a panic where we were panicing somewhere in
tree_backref_for_extent from scrub_print_warning.  He only captured the trace
but looking at scrub_print_warning we drop the path right before we mess with
the extent buffer to print out a bunch of stuff, which isn't right.  So fix this
by dropping the path after we use the eb if we need to.  Thanks,

Cc: stable@vger.kernel.org
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

There is no lock to protect fs_info->fs_state, it will introduce
some problems, such as the value may be covered by the other task
when several tasks modify it. For example:
	Task0 - CPU0		Task1 - CPU1
	mov %fs_state rax
	or $0x1 rax
				mov %fs_state rax
				or $0x2 rax
	mov rax %fs_state
				mov rax %fs_state
The expected value is 3, but in fact, it is 2.

Though this problem doesn't happen now (because there is only one
flag currently), the code is error prone, if we add other flags,
the above problem will happen to a certainty.

Now we use bit operation for it to fix the above problem.
In this way, we can make the code more robust and be easy to
add new flags.
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

While running snapshot testscript created by Mitch and David,
the race between autodefrag and snapshot deletion can lead to
corruption of dead_root list so that we can get crash on
btrfs_clean_old_snapshots().

And besides autodefrag, scrub also does the same thing, ie. read
root first and get inode.

Here is the story(take autodefrag as an example):
(1) when we delete a snapshot or subvolume, it will set its root's
refs to zero and do a iput() on its own inode, and if this inode happens
to be the only active in-meory one in root's inode rbtree, it will add
itself to the global dead_roots list for later cleanup.

(2) after (1), the autodefrag thread may read another inode for defrag
and the inode is just in the deleted snapshot/subvolume, but all of these
are without checking if the root is still valid(refs > 0).  So the end up
result is adding the deleted snapshot/subvolume's root to the global
dead_roots list AGAIN.

Fortunately, we already have a srcu lock to avoid the race, ie. subvol_srcu.

So all we need to do is to take the lock to protect 'read root and get inode',
since we synchronize to wait for the rcu grace period before adding something
to the global dead_roots list.
Reported-by: NMitch Harder <mitch.harder@sabayonlinux.org>
Signed-off-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NJosef Bacik <jbacik@fusionio.com>

This builds on David Woodhouse's original Btrfs raid5/6 implementation.
The code has changed quite a bit, blame Chris Mason for any bugs.

Read/modify/write is done after the higher levels of the filesystem have
prepared a given bio.  This means the higher layers are not responsible
for building full stripes, and they don't need to query for the topology
of the extents that may get allocated during delayed allocation runs.
It also means different files can easily share the same stripe.

But, it does expose us to incorrect parity if we crash or lose power
while doing a read/modify/write cycle.  This will be addressed in a
later commit.

Scrub is unable to repair crc errors on raid5/6 chunks.

Discard does not work on raid5/6 (yet)

The stripe size is fixed at 64KiB per disk.  This will be tunable
in a later commit.
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This fixes a very special case that can be reproduced by just
disconnecting a disk at runtime, and without unmounting the
filesystem first, start scrub on the filesystem with the
disconnected disk. All read and write EIOs are handled
correctly, only the first superblock is an exception and gives
a BUG() in a subfunction. The BUG() is correct, it would crash
later otherwise. The subfunction must not be called for
superblocks and this is what the fix changes.
Reported-by: NJoeri Vanthienen <mail@joerivanthienen.be>
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This regression was introduced by the device-replace patches.
Scrub immediately stops checking those disks that have write errors.
This is nothing that happens in the real world, but it is wrong
since scrub is the tool to detect and repair defects. Fix it.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Before this commit, btrfs_map_block() was called with REQ_WRITE
in order to retrieve the list of mirrors for a disk block.
This needs to be changed for the device replace procedure since
it makes a difference whether you are asking for read mirrors
or for locations to write to.
GET_READ_MIRRORS is introduced as a new interface to call
btrfs_map_block().
In the current commit, the functionality is not yet changed,
only the interface for GET_READ_MIRRORS is introduced and all
the places that should use this new interface are adapted.

The reason that REQ_WRITE cannot be abused anymore to retrieve
a list of read mirrors is that during a running dev replace
operation all write requests to the live filesystem are
duplicated to also write to the target drive.
Keep in mind that the target disk is only partially a valid
copy of the source disk while the operation is ongoing. All
writes go to the target disk, but not all reads would return
valid data on the target disk. Therefore it is not possible
anymore to abuse a REQ_WRITE interface to find valid mirrors
for a REQ_READ.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This commit contains all the essential changes to the core code
of Btrfs for support of the device replace procedure.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The device replace procedure makes use of the scrub code. The scrub
code is the most efficient code to read the allocated data of a disk,
i.e. it reads sequentially in order to avoid disk head movements, it
skips unallocated blocks, it uses read ahead mechanisms, and it
contains all the code to detect and repair defects.
This commit adds code to scrub to allow the scrub code to copy read
data to another disk.
One goal is to be able to perform as fast as possible. Therefore the
write requests are collected until huge bios are built, and the
write process is decoupled from the read process with some kind of
flow control, of course, in order to limit the allocated memory.
The best performance on spinning disks could by reached when the
head movements are avoided as much as possible. Therefore a single
worker is used to interface the read process with the write process.
The regular scrub operation works as fast as before, it is not
negatively influenced and actually it is more or less unchanged.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This patch adds some code to disallow operations on the device that
is used as the target for the device replace operation.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

A small number of functions that are used in a device replace
procedure when the operation is resumed at mount time are unable
to pass the same root pointer that would be used in the regular
(ioctl) context. And since the root pointer is not required, only
the fs_info is, the root pointer argument is replaced with the
fs_info pointer argument.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

This is required for the device replace procedure in a later step.
Two calling functions also had to be changed to have the fs_info
pointer: repair_io_failure() and scrub_setup_recheck_block().
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

Just move some code into functions to make everything more readable.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

In the scrub repair code, the code is changed to handle memory
allocation errors a little bit smarter. The change is to handle
it just like a read error. This simplifies the code and removes
a couple of lines of code, since the code to handle read errors
is there anyway.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

In case that disk blocks need to be repaired (rewritten), the
current code at first (for simplicity reasons) reads all alternate
mirrors in the first step, afterwards selects the best one in a
second step. This is now changed to read one alternate mirror
after the other and to leave the loop early when a perfect mirror
is found.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

With the modified design (in order to support the devive replace
procedure) it is necessary to alloc the page array dynamically.
The reason is that pages are reused. At first a page is used for
the bio to read the data from the filesystem, then the same page
is reused for the bio that writes the data to the target disk.
Since the read process and the write process are completely
decoupled, this requires a new concept of refcounts and get/put
functions for pages, and it requires to use newly created pages
for each read bio which are freed after the write operation
is finished.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The block device is removed from the scrub context state structure.
The scrub code as it is used for the device replace procedure reads
the source data from whereever it is optimal. The source device might
even be gone (disconnected, for instance due to a hardware failure).
Or the drive can be so faulty so that the device replace procedure
tries to avoid access to the faulty source drive as much as possible,
and only if all other mirrors are damaged, as a last resort, the
source disk is accessed.
The modified scrub code operates as if it would handle the source
drive and thereby generates an exact copy of the source disk on the
target disk, even if the source disk is not present at all. Therefore
the block device pointer to the source disk is removed in the scrub
context struct and moved into the lower level scope of scrub_bio,
fixup and page structures where the block device context is known.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

The device replace procedure makes use of the scrub code. The scrub
code is the most efficient code to read the allocated data of a disk,
i.e. it reads sequentially in order to avoid disk head movements, it
skips unallocated blocks, it uses read ahead mechanisms, and it
contains all the code to detect and repair defects.
This commit is a first preparation step to adapt the scrub code to
be shareable for the device replace procedure.
The block device will be removed from the scrub context state
structure in a later step. It used to be the source block device.
The scrub code as it is used for the device replace procedure reads
the source data from whereever it is optimal. The source device might
even be gone (disconnected, for instance due to a hardware failure).
Or the drive can be so faulty so that the device replace procedure
tries to avoid access to the faulty source drive as much as possible,
and only if all other mirrors are damaged, as a last resort, the
source disk is accessed.
The modified scrub code operates as if it would handle the source
drive and thereby generates an exact copy of the source disk on the
target disk, even if the source disk is not present at all. Therefore
the block device pointer to the source disk is removed in a later
patch, and therefore the context structure is renamed (this is the
goal of the current patch) to reflect that no source block device
scope is there anymore.

Summary:
This first preparation step consists of a textual substitution of the
term "dev" to the term "ctx" whereever the scrub context is used.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@fusionio.com>

For immutable bio vecs, I've been auditing and removing bi_idx
references. These were harmless, but removing them will make auditing
easier.

scrub_bio_end_io_worker() was open coding a bio_reset() - but this
doesn't appear to have been needed for anything as right after it does a
bio_put(), and perusing the code it doesn't appear anything else was
holding a reference to the bio.

The other use end_bio_extent_readpage() was just for a pr_debug() -
changed it to something that might be a bit more useful.
Signed-off-by: NKent Overstreet <koverstreet@google.com>
CC: Chris Mason <chris.mason@oracle.com>
CC: Stefan Behrens <sbehrens@giantdisaster.de>

In logical resolve, we parse extent_from_logical()'s 'ret' as a kind of flag.

It is possible to lose our errors because
(-EXXXX & BTRFS_EXTENT_FLAG_TREE_BLOCK) is true.

I'm not sure if it is on purpose, it just looks too hacky if it is.
I'd rather use a real flag and a 'ret' to catch errors.
Acked-by: NJan Schmidt <list.btrfs@jan-o-sch.net>
Signed-off-by: NLiu Bo <liub.liubo@gmail.com>

bbio has been malloced in btrfs_map_block() and should be
freed before leaving from the error handling cases.

spatch with a semantic match is used to found this problem.
(http://coccinelle.lip6.fr/)
Signed-off-by: NWei Yongjun <yongjun_wei@trendmicro.com.cn>

Al pointed out that we can just toss out the old name on a device and add a
new one arbitrarily, so anybody who uses device->name in printk could
possibly use free'd memory.  Instead of adding locking around all of this he
suggested doing it with RCU, so I've introduced a struct rcu_string that
does just that and have gone through and protected all accesses to
device->name that aren't under the uuid_mutex with rcu_read_lock().  This
protects us and I will use it for dealing with removing the device that we
used to mount the file system in a later patch.  Thanks,
Reviewed-by: NDavid Sterba <dsterba@suse.cz>
Signed-off-by: NJosef Bacik <josef@redhat.com>

The goal is to detect when drives start to get an increased error rate,
when drives should be replaced soon. Therefore statistic counters are
added that count IO errors (read, write and flush). Additionally, the
software detected errors like checksum errors and corrupted blocks are
counted.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>

Fix that when scrub tries to repair an I/O or checksum error and one of
the devices containing the mirror is missing, it crashes in bio_add_page
because the bdev is a NULL pointer for missing devices.
Reported-by: NMarco L. Crociani <marco.crociani@gmail.com>
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Each CRC or header error was counted twice, this is now fixed.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>

bio_alloc() has the possibility of returning NULL.
So, it is necessary to check the return value.
Signed-off-by: NTsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Scrub used to be coded for nodesize == leafsize == sectorsize == PAGE_SIZE.
This is now changed to support sizes for nodesize and leafsize which are
N * PAGE_SIZE.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Just a minor cleanup commit in preparation for the big block changes.
Signed-off-by: NStefan Behrens <sbehrens@giantdisaster.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

In commit 4692cf58 we introduced new backref walking code for btrfs. This
assumes we're searching live roots, which requires a transaction context.
While scrubbing, however, we must not join a transaction because this could
deadlock with the commit path. Additionally, what scrub really wants to do
is resolving a logical address in the commit root it's currently checking.

This patch adds support for logical to path resolving on commit roots and
makes scrub use that.
Signed-off-by: NJan Schmidt <list.btrfs@jan-o-sch.net>