help: add id suboption to -iscsi

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

scsi-disk: fix WRITE SAME with large non-zero payload

Due to a thinko in the patch that implemented WRITE SAME.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

block/iscsi: introduce bdrv_co_{readv, writev, flush_to_disk}

this converts read, write and flush functions from aio to coroutines
eliminating almost 200 lines of code.

The requirement for libiscsi is bumped to version 1.4.0 which was
released in may 2012.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

scsi-disk: fix VERIFY emulation

VERIFY emulation was completely botched (and remained botched through
all the refactorings).  The command must be emulated both in check-medium
mode (BYTCHK=00, which we implement by doing nothing) and in check-bytes
mode (which we do not implement yet).  Unlike WRITE AND VERIFY (which we
treat simply as WRITE with FUA bit set), VERIFY cannot be handled like
READ.  In fact the device is _receiving_ data for VERIFY, not _sending_
it like READ.

Cc: qemu-stable@nongnu.org
Tested-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

scsi-bus: fix transfer length and direction for VERIFY command

The amount of bytes to transfer depends on the BYTCHK field.
If any data is transferred, it is sent to the device.

Cc: qemu-stable@nongnu.org
Tested-by: Hervé Poussineau <hpoussin@reactos.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

qemu-iotests: filter QEMU monitor \r\n

SMTP does not preserve newlines.  This is normally not a problem if the
email body uses DOS or UNIX newlines consistently.  In 051.out we mix
UNIX newlines with DOS newlines (since QEMU monitor output uses \r\n).

This patch filters the QEMU monitor output so the golden master file
uses UNIX newlines exclusively.

The result is that patches touching 051.out will apply cleanly without
mangling newlines after this commit.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

aio: make aio_poll(ctx, true) block with no fds

This patch drops a special case where aio_poll(ctx, true) returns false
instead of blocking if no file descriptors are waiting on I/O.  Now it
is possible to block in aio_poll() to wait for aio_notify().

This change eliminates busy waiting.  bdrv_drain_all() used to rely on
busy waiting to completed throttled I/O requests but this is no longer
required so we can simplify aio_poll().

Note that aio_poll() still returns false when aio_notify() was used.  In
other words, stopping a blocking aio_poll() wait is not considered
making progress.

Adjust test-aio /aio/bh/callback-delete/one which assumed aio_poll(ctx,
true) would immediately return false instead of blocking.

Reviewed-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: clean up bdrv_drain_all() throttling comments

Since cc0681c45430a1f1a4c2d06e9499b7775afc9a18 ("block: Enable the new
throttling code in the block layer.") bdrv_drain_all() no longer spins.
The code used to look as follows:

  do {
      busy = qemu_aio_wait();

      /* FIXME: We do not have timer support here, so this is effectively
       * a busy wait.
       */
      QTAILQ_FOREACH(bs, &bdrv_states, list) {
          while (qemu_co_enter_next(&bs->throttled_reqs)) {
              busy = true;
          }
      }
  } while (busy);

Note that throttle requests are kicked but I/O throttling limits are
still in effect.  The loop spins until the vm_clock time allows the
request to make progress and complete.

The new throttling code introduced bdrv_start_throttled_reqs().  This
function not only kicks throttled requests but also temporarily disables
throttling so requests can run.

The outdated FIXME comment can be removed.  Also drop the busy = true
assignment since we overwrite it immediately afterwards.

Reviewed-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qcow2: use start_of_cluster() and offset_into_cluster() everywhere

Signed-off-by: Hu Tao <hutao@cn.fujitsu.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: decrease progress update interval on convert

when doing very large jobs updating the progress only every 2%
is too rare.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: round down request length to an aligned sector

this patch shortens requests to end at an aligned sector so that
the next request starts aligned.

[Squashed Peter's fix for bdrv_get_info() failure discussed on the
mailing list.
--Stefan]

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: dynamically adjust iobuffer size during convert

since the convert process is basically a sync operation it might
be benificial in some case to change the hardcoded I/O buffer
size to a greater value.

This patch increases the I/O buffer size if the output
driver advertises an optimal transfer length or discard alignment
that is greater than the default buffer size of 2M.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: set bs->bl.opt_transfer_length

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add opt_transfer_length to BlockLimits

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: set bdi->cluster_size

this patch aims to set bdi->cluster_size to the internal page size
of the iscsi target so that enabled callers can align requests
properly.

Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: fix usage instruction for qemu-img convert

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: add support for skipping zeroes in input during convert

we currently do not check if a sector is allocated during convert.
This means if a sector is unallocated that we allocate a bounce
buffer of zeroes, find out its zero later and do not write it
in the best case. In the worst case this can lead to reading
blocks from a raw device (like iSCSI) altough we could easily
know via get_block_status that they are zero and simply skip them.

This patch also fixes the progress output not being at 100% after
a successful conversion.

Signed-off-by: Peter Lieven <pl@kamp.de>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-nbd: add doc for option -f

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: add test for snapshot in qemu-img convert

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: add -l for snapshot in convert

Now qemu-img convert have similar options as qemu-nbd for internal
snapshot.

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: add 058 internal snapshot export with qemu-nbd case

This case can't run when IMGPROTO=nbd, since it needs to create some
internal snapshot which would fail for EOF write request, even when
TEST_IMG is exported with "-f raw" in common.rc, so set _supported_proto
to file.

_require_command() is changed to tip what util is missing, instead
of printing a blank.

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-nbd: support internal snapshot export

Now it is possible to directly export an internal snapshot, which
can be used to probe the snapshot's contents without qemu-img
convert.

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

snapshot: distinguish id and name in load_tmp

Since later this function will be used so improve it. The only caller of it
now is qemu-img, and it is not impacted by introduce function
bdrv_snapshot_load_tmp_by_id_or_name() that call bdrv_snapshot_load_tmp()
twice to keep old search logic. bdrv_snapshot_load_tmp_by_id_or_name() return
int to let caller know the errno, and errno will be used later.
Also fix a typo in comments of bdrv_snapshot_delete().

Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Split qcow2 only cases in 048

Format "raw" doesn't always work on certain file systems (e.g. tmpfs).
Use qcow2 to make the allocation status explicit and split into a new
case.

[Resolved merge conflict due to "qemu-io> " prompt filter, added 074 to
group file, and fixed up s/048/074/ copy-paste mistake.
--Stefan]

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Clean up spaces in usage output

Whitespace changes to align columns.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Change default cache mode to "writeback"

So that the tests can run faster.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Add _default_cache_mode and _supported_cache_modes

This replaces _unsupported_qemu_io_options and check for support of
current cache mode, and allow to provide a default if user didn't
specify.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Honour cache mode in iotests.py

This will allow overriding cache mode from the "-c mode" option.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Add "-c <cache-mode>" option

The option sets cache mode used in the tests. "-nocache" is changed to
an alias to "-c none", and internally passes "-t none" to qemu-io.

Python scripts will make use of option this in the next commit.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qcow2: Zero-initialise first cluster for new images

Strictly speaking, this is only required for has_zero_init() == false,
but it's easy enough to just do a cluster-aligned write that is padded
with zeros after the header.

This fixes that after 'qemu-img create' header extensions are attempted
to be parsed that are really just random leftover data.

Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: Close backing file early in bdrv_img_create

Leaving the backing file open although it is not needed anymore can
cause problems if it is opened through a block driver which allows
exclusive access only and if the create function of the block driver
used for the top image (the one being created) tries to close and reopen
the image file (which will include opening the backing file a second
time).

In particular, this will happen with a backing file opened through
qemu-nbd and using qcow2 as the top image file format (which reopens the
image to flush it to disk).

In addition, the BlockDriverState in bdrv_img_create() is used for the
backing file only; it should therefore be made local to the respective
block.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

scsi-disk: correctly implement WRITE SAME

Fetch the data to be written from the input buffer.  If it is all zeroes,
we can use the write_zeroes call (possibly with the new MAY_UNMAP flag).
Otherwise, do as many write cycles as needed, writing 512k at a time.

Strictly speaking, this is still incorrect because a zero cluster should
only be written if the MAY_UNMAP flag is set.  But this is a bug in qcow2
and the other formats, not in the SCSI code.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

scsi-disk: reject ANCHOR=1 for UNMAP and WRITE SAME commands

Since we report ANC_SUP==0 in VPD page B2h, we need to return
an error (ILLEGAL REQUEST/INVALID FIELD IN CDB) for all WRITE SAME
requests with ANCHOR==1.

Inspired by a similar patch to the LIO in-kernel target.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

scsi-disk: catch write protection errors in UNMAP

This is the same that is already done for WRITE SAME.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: 033 is fast

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

raw-posix: add support for write_zeroes on XFS and block devices

The code is similar to the implementation of discard and write_zeroes
with UNMAP.  However, failure must be propagated up to block.c.

The stale page cache problem can be reproduced as follows:

    # modprobe scsi-debug lbpws=1 lbprz=1
    # ./qemu-io /dev/sdXX
    qemu-io> write -P 0xcc 0 2M
    qemu-io> write -z 0 1M
    qemu-io> read -P 0x00 0 512
    Pattern verification failed at offset 0, 512 bytes
    qemu-io> read -v 0 512
    00000000:  cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc  ................
    ...

    # ./qemu-io --cache=none /dev/sdXX
    qemu-io> write -P 0xcc 0 2M
    qemu-io> write -z 0 1M
    qemu-io> read -P 0x00 0 512
    qemu-io> read -v 0 512
    00000000:  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    ...

And similarly with discard instead of "write -z".

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

raw-posix: implement write_zeroes with MAY_UNMAP for block devices

See the next commit for the description of the Linux kernel problem
that is worked around in raw_open_common.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

raw-posix: implement write_zeroes with MAY_UNMAP for files

Writing zeroes to a file can be done by punching a hole if
MAY_UNMAP is set.

Note that in this case ENOTSUP is not ignored, but makes
the block layer fall back to the generic implementation.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: check WRITE SAME support differently depending on MAY_UNMAP

The current check is right for MAY_UNMAP=1.  For MAY_UNMAP=0, just
try and fall back to regular writes as soon as a WRITE SAME command
fails.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: updated copyright

added myself to reflect recent work on the iscsi block driver.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: remove .bdrv_has_zero_init

since commit 3ac21627 the default value changed to 0.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block drivers: expose requirement for write same alignment from formats

This will let misaligned but large requests use zero clusters.  This
is important because the cluster size is not guest visible.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block drivers: add discard/write_zeroes properties to bdrv_get_info implementation

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

vpc, vhdx: add get_info

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: make bdrv_co_do_write_zeroes stricter in producing aligned requests

Right now, bdrv_co_do_write_zeroes will only try to align the
beginning of the request.  However, it is simpler for many
formats to expect the block layer to separate both the head *and*
the tail.  This makes sure that the format's bdrv_co_write_zeroes
function will be called with aligned sector_num and nb_sectors for
the bulk of the request.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: handle ENOTSUP from discard in generic code

Similar to write_zeroes, let the generic code receive a ENOTSUP for
discard operations.  Since bdrv_discard has advisory semantics,
we can just swallow the error.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add bdrv_aio_write_zeroes

This will be used by the SCSI layer.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add flags argument to bdrv_co_write_zeroes tracepoint

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add flags to BlockRequest

This lets bdrv_co_do_rw receive flags, so that it can be used for
zero writes.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: generalize BlockLimits handling to cover bdrv_aio_discard too

bdrv_co_discard is only covering drivers which have a .bdrv_co_discard()
implementation, but not those with .bdrv_aio_discard(). Not very nice,
and easy to avoid.

Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

vmdk: Fix creating big description file

The buffer for description file was 4096 which only covers a few
hundred of extents. This changes the buffer to dynamic allocated with
g_strdup_printf in order to support bigger cases.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

coroutine: remove unused CoQueue AioContext

The AioContext ctx field is apparently unused in qemu codebase since
02ffb504485.

Signed-off-by: Marc-André Lureau <marcandre.lureau@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

coroutine: remove qemu_co_queue_wait_insert_head

qemu_co_queue_wait_insert_head() is unused in qemu code base now.

Signed-off-by: Marc-André Lureau <marcandre.lureau@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Add sample image and test for VMDK version 3

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vmdk: Allow read only open of VMDK version 3

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Filter out 'qemu-io> ' prompt

This removes "qemu-io> " prompt from qemu-io output in _filter_qemu_io,
and updates all the output files with the following command:

 cd tests/qemu-iotests && sed -i "s/qemu-io> //g" *.out

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Filter qemu-io output in 025

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Use BDRV_O_NO_BACKING where appropriate

If you open an image temporarily just because you want to check its size
or get it flushed, there's no real reason to open the whole backing file
chain.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>

qemu-iotests: Test snapshot mode

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>

block: Enable BDRV_O_SNAPSHOT with driver-specific options

In the case of snapshot=on, don't rely on the backing file path in the
temporary image any more, but override the backing file with the given
set of options. This way, block drivers that don't use a file name can
be accessed with snapshot=on, for example:

    -drive file.driver=nbd,file.host=localhost,snapshot=on

Which becomes internally something like:

    file.filename=/tmp/vl.AWQZCu,backing.file.driver=nbd,backing.file.host=localhost

Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qemu-iotests: Make test case 030, 040 and 055 deterministic

Pause the drive and start the block job, so we won't miss the block job.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotest: Add pause_drive and resume_drive methods

They wrap blkdebug "break" and "remove_break".

Add optional argument "resume" to cancel_and_wait().

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

blkdebug: add "remove_break" command

This adds "remove_break" command which is the reverse of blkdebug
command "break": it removes all breakpoints with given tag and resumes
all the requests.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-iotests: Drop local version of cancel_and_wait from 040

iotests.py already has one.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

sheepdog: support user-defined redundancy option

Sheepdog support two kinds of redundancy, full replication and erasure coding.

# create a fully replicated vdi with x copies
 -o redundancy=x (1 <= x <= SD_MAX_COPIES)

# create a erasure coded vdi with x data strips and y parity strips
 -o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)

E.g, to convert a vdi into sheepdog vdi 'test' with 8:3 erasure coding scheme

$ qemu-img convert -o redundancy=8:3 linux-0.2.img sheepdog:test

Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

sheepdog: refactor do_sd_create()

We can actually use BDRVSheepdogState *s to pass most of the parameters.

Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qdict: Optimise qdict_do_flatten()

Nested QDicts used to be both entered recursively in order to move their
entries to the target QDict and also be moved themselves to the target
QDict like all other objects. This is harmless because for the top
level, qdict_do_flatten() will encounter the (now empty) QDict for a
second time and then delete it, but at the same time it's obviously
unnecessary overhead. Just delete nested QDicts directly after moving
all of their entries.

Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qdict: Fix memory leak in qdict_do_flatten()

Reported-by: Laszlo Ersek <lersek@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

MAINTAINERS: add sheepdog development mailing list

This will help people find mailing list relevant to sheepdog.

Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

COW: Extend checking allocated bits to beyond one sector

cow_co_is_allocated() only checks one sector's worth of allocated bits
before returning. This is allowed but (slightly) inefficient, so extend
it to check all of the file's metadata sectors.

Signed-off-by: Charlie Shepherd <charlie@ctshepherd.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
[kwolf: silenced compiler warning (-Wmaybe-uninitialized for changed)]
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

COW: Speed up writes

Process a whole sector's worth of COW bits by reading a sector, setting
the bits after skipping any already set bits, then writing it out again.
Make sure we only flush once before writing metadata, and only if we
need to write metadata.

Signed-off-by: Charlie Shepherd <charlie@ctshepherd.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qapi: Change BlockDirtyInfo to list

We have multiple dirty bitmaps in BDS now, switch QAPI to allow query
it (BlockInfo.dirty_bitmaps), and also drop old BlockInfo.dirty.

Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: per caller dirty bitmap

Previously a BlockDriverState has only one dirty bitmap, so only one
caller (e.g. a block job) can keep track of writing. This changes the
dirty bitmap to a list and creates a BdrvDirtyBitmap for each caller, the
lifecycle is managed with these new functions:

    bdrv_create_dirty_bitmap
    bdrv_release_dirty_bitmap

Where BdrvDirtyBitmap is a linked list wrapper structure of HBitmap.

In place of bdrv_set_dirty_tracking, a BdrvDirtyBitmap pointer argument
is added to these functions, since each caller has its own dirty bitmap:

    bdrv_get_dirty
    bdrv_dirty_iter_init
    bdrv_get_dirty_count

bdrv_set_dirty and bdrv_reset_dirty prototypes are unchanged but will
internally walk the list of all dirty bitmaps and set them one by one.

Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/stream: Don't stream unbacked devices

If a block device is unbacked, a streaming blockjob should immediately
finish instead of beginning to try to stream, then noticing the backing
file does not contain even the first sector (since it does not exist)
and then finishing normally.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

sheepdog: implement .bdrv_get_allocated_file_size

With this patch, qemu-img info sheepdog:image will show disk size for sheepdog
images.

Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
Reviewed-by: MORITA Kazutaka <morita.kazutaka@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Test coroutine execution order

This patch adds a test for coroutine execution order in test-coroutine -
this catches a bug in the CPC coroutine implementation.

Signed-off-by: Charlie Shepherd <charlie@ctshepherd.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

util/error: Save errno from clobbering

There may be calls to error_setg() and especially error_setg_errno()
which blindly (and until now wrongly) assume these functions not to
clobber errno (e.g., they pass errno to error_setg_errno() and return
-errno afterwards). Instead of trying to find and fix all of these
constructs, just make sure error_setg() and error_setg_errno() indeed do
not clobber errno.

Suggested-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: conditionally zero out target on convert

If the target has_zero_init = 0, but supports efficiently
writing zeroes by unmapping we call bdrv_make_zero to
avoid fully allocating the target. This currently works
only for iscsi.  It can be extended to raw with
BLKDISCARDZEROES for example.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

qemu-img: add support for fully allocated images

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/get_block_status: fix BDRV_BLOCK_ZERO for unallocated blocks

this patch does 2 things:
a) only do additional call outs if BDRV_BLOCK_ZERO is not already set.
b) use the newly introduced bdrv_unallocated_blocks_are_zero()
   to return the zero state of an unallocated block. the used callout
   to bdrv_has_zero_init() is only valid right after bdrv_create.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: introduce bdrv_make_zero

this patch adds a call to completely zero out a block device.
the operation is sped up by checking the block status and
only writing zeroes to the device if they currently do not
return zeroes. optionally the zero writing can be sped up
by setting the flag BDRV_REQ_MAY_UNMAP to emulate the zero
write by unmapping if the driver supports it.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iscsi: add bdrv_co_write_zeroes

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iscsi: simplify iscsi_co_discard

now that bdrv_co_discard can handle limits we do not need
the request split logic here anymore.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

iscsi: set limits in BlockDriverState

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: honour BlockLimits in bdrv_co_discard

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: honour BlockLimits in bdrv_co_do_write_zeroes

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/raw: copy BlockLimits on raw_open

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add BlockLimits structure to BlockDriverState

this patch adds BlockLimits which introduces discard and write_zeroes
limits and alignment information to the BlockDriverState.

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block/iscsi: add .bdrv_get_info

Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add wrappers for logical block provisioning information

This adds 2 wrappers to read the unallocated_blocks_are_zero and
can_write_zeroes_with_unmap info from the BDI. The wrappers are
required to check for the existence of a backing_hd and
if the devices are opened with the correct flags.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add logical block provisioning info to BlockDriverInfo

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: introduce BDRV_REQ_MAY_UNMAP request flag

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: add flags to bdrv_*_write_zeroes

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

block: make BdrvRequestFlags public

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

Open 2.0 development tree

Signed-off-by: Anthony Liguori <aliguori@amazon.com>

Update version for 1.7.0 release

Signed-off-by: Anthony Liguori <aliguori@amazon.com>

qemu-iotests: Fix test 041

Performing multiple drive-mirror blockjobs on the same qemu instance
results in the image file used for the block device being replaced by
the newly mirrored file, which is not what we want.

Fix this by performing one dedicated test per sync mode.

Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1385407736-13941-3-git-send-email-mreitz@redhat.com
Signed-off-by: Anthony Liguori <aliguori@amazon.com>

block/drive-mirror: Reuse backing HD for sync=none

For "none" sync mode in "absolute-paths" mode, the current image should
be used as the backing file for the newly created image.

The current behavior is:
a) If the image to be mirrored has a backing file, use that (which is
   wrong, since the operations recorded by "none" are applied to the
   image itself, not to its backing file).
b) If the image to be mirrored lacks a backing file, the target doesn't
   have one either (which is not really wrong, but not really right,
   either; "none" records a set of operations executed on the image
   file, therefore having no backing file to apply these operations on
   seems rather pointless).

For a, this is clearly a bugfix. For b, it is still a bugfix, although
it might break existing API - but since that case crashed qemu just
three weeks ago (before 1452686495922b81d6cf43edf025c1aef15965c0), we
can safely assume there is no such API relying on that case yet.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1385407736-13941-2-git-send-email-mreitz@redhat.com
Signed-off-by: Anthony Liguori <aliguori@amazon.com>

Update version for v1.7.0-rc2 release

curses: fixup SIGWINCH handler mess

Don't run code in the signal handler, only set a flag.
Use sigaction(2) to avoid non-portable signal(2) semantics.
Make #ifdefs less messy.

Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Message-id: 1385130903-20531-1-git-send-email-kraxel@redhat.com
Signed-off-by: Anthony Liguori <aliguori@amazon.com>