The following pattern is unsafe:
char buf[32];
ret = read(fd, buf, sizeof(buf));
...
buf[ret] = 0;
If read(2) returns 32 then a byte beyond the end of the buffer is
zeroed.
In practice this buffer overflow does not occur because the sysfs
max_segments file only contains an unsigned short + '\n'. The string is
always shorter than 32 bytes.
Regardless, avoid this pattern because static analysis tools might
complain and it could lead to real buffer overflows if copy-pasted
elsewhere in the codebase.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
BlockLimits.max_transfer can be too high without this fix, guest will
encounter I/O error or even get paused with werror=stop or rerror=stop. The
cause is explained below.
Linux has a separate limit, /sys/block/.../queue/max_segments, which in
the worst case can be more restrictive than the BLKSECTGET which we
already consider (note that they are two different things). So, the
failure scenario before this patch is:
1) host device has max_sectors_kb = 4096 and max_segments = 64;
2) guest learns max_sectors_kb limit from QEMU, but doesn't know
max_segments;
3) guest issues e.g. a 512KB request thinking it's okay, but actually
it's not, because it will be passed through to host device as an
SG_IO req that has niov > 64;
4) host kernel doesn't like the segmenting of the request, and returns
-EINVAL;
This patch checks the max_segments sysfs entry for the host device and
calculates a "conservative" bytes limit using the page size, which is
then merged into the existing max_transfer limit. Guest will discover
this from the usual virtual block device interfaces. (In the case of
scsi-generic, it will be done in the INQUIRY reply interception in
device model.)
The other possibility is to actually propagate it as a separate limit,
but it's not better. On the one hand, there is a big complication: the
limit is per-LUN in QEMU PoV (because we can attach LUNs from different
host HBAs to the same virtio-scsi bus), but the channel to communicate
it in a per-LUN manner is missing down the stack; on the other hand,
two limits versus one doesn't change much about the valid size of I/O
(because guest has no control over host segmenting).
Also, the idea to fall back to bounce buffering in QEMU, upon -EINVAL,
was explored. Unfortunately there is no neat way to ensure the bounce
buffer is less segmented (in terms of DMA addr) than the guest buffer.
Practically, this bug is not very common. It is only reported on a
Emulex (lpfc), so it's okay to get it fixed in the easier way.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that we are truncating the file in both PREALLOC_MODE_FULL and
PREALLOC_MODE_OFF, not truncating in PREALLOC_MODE_FALLOC looks odd.
Add a comment explaining why we do not truncate in this case.
Signed-off-by: Nir Soffer <nirsof@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In a previous commit (qemu-img: Do not truncate before preallocation) we
moved truncate to the PREALLOC_MODE_OFF branch to avoid slowdown in
posix_fallocate().
However this change is not optimal when using PREALLOC_MODE_FULL, since
knowing the final size from the beginning could allow the file system
driver to do less allocations and possibly avoid fragmentation of the
file.
Now we truncate also before doing full preallocation.
Signed-off-by: Nir Soffer <nirsof@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When using file system that does not support fallocate() (e.g. NFS <
4.2), truncating the file only when preallocation=OFF speeds up creating
raw file.
Here is example run, tested on Fedora 24 machine, creating raw file on
NFS version 3 server.
$ time ./qemu-img-master create -f raw -o preallocation=falloc mnt/test 1g
Formatting 'mnt/test', fmt=raw size=1073741824 preallocation=falloc
real 0m21.185s
user 0m0.022s
sys 0m0.574s
$ time ./qemu-img-fix create -f raw -o preallocation=falloc mnt/test 1g
Formatting 'mnt/test', fmt=raw size=1073741824 preallocation=falloc
real 0m11.601s
user 0m0.016s
sys 0m0.525s
$ time dd if=/dev/zero of=mnt/test bs=1M count=1024 oflag=direct
1024+0 records in
1024+0 records out
1073741824 bytes (1.1 GB, 1.0 GiB) copied, 15.6627 s, 68.6 MB/s
real 0m16.104s
user 0m0.009s
sys 0m0.220s
Running with strace we can see that without this change we do one
pread() and one pwrite() for each block. With this change, we do only
one pwrite() per block.
$ strace ./qemu-img-master create -f raw -o preallocation=falloc mnt/test 8192
...
pread64(9, "\0", 1, 4095) = 1
pwrite64(9, "\0", 1, 4095) = 1
pread64(9, "\0", 1, 8191) = 1
pwrite64(9, "\0", 1, 8191) = 1
$ strace ./qemu-img-fix create -f raw -o preallocation=falloc mnt/test 8192
...
pwrite64(9, "\0", 1, 4095) = 1
pwrite64(9, "\0", 1, 8191) = 1
This happens because posix_fallocate is checking if each block is
allocated before writing a byte to the block, and when truncating the
file before preallocation, all blocks are unallocated.
Signed-off-by: Nir Soffer <nirsof@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We can get the maximum number of bytes for a single I/O transfer
from the BLKSECTGET ioctl, but we only perform this for block
devices. scsi-generic devices are represented as character devices,
and so do not issue this today. Update this, so that virtio-scsi
devices using the scsi-generic interface can return the same data.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Message-Id: <20170120162527.66075-4-farman@linux.vnet.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Commit 6f6071745b ("raw-posix: Fetch max sectors for host block device")
introduced a routine to call the kernel BLKSECTGET ioctl, which stores the
result back to user space. However, the size of the data returned depends
on the routine handling the ioctl. The (compat_)blkdev_ioctl returns a
short, while sg_ioctl returns an int. Thus, on big-endian systems, we can
find ourselves accidentally shifting the result to a much larger value.
(On s390x, a short is 16 bits while an int is 32 bits.)
Also, the two ioctl handlers return values in different scales (block
returns sectors, while sg returns bytes), so some tweaking of the outputs
is required such that hdev_get_max_transfer_length returns a value in a
consistent set of units.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Message-Id: <20170120162527.66075-3-farman@linux.vnet.ibm.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
These files deal with the file protocol, not the raw format (the
file protocol is often used with other formats, and the raw
format is not forced to use the file protocol). Rename things
to make it a bit easier to follow.
Suggested-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Stefan Hajnoczi
Fam Zheng
Nir Soffer
Eric Farman
Eric Blake