Max Reitz [Tue, 18 Nov 2014 11:21:14 +0000 (12:21 +0100)]
block: Lift more functions into BlockBackend
There are already some blk_aio_* functions, so we might as well have
blk_co_* functions (as far as we need them). This patch adds
blk_co_flush(), blk_co_discard(), and also blk_invalidate_cache() (which
is not a blk_co_* function but is needed nonetheless).
Stefan Hajnoczi [Thu, 13 Nov 2014 10:24:40 +0000 (10:24 +0000)]
ahci: avoid #ifdef DEBUG_AHCI bitrot
Debug code using #ifdef is susceptible to bitrot because the compiler
never checks the debug code.
This is easy to avoid, change the DPRINTF() macro to use if (DEBUG_AHCI)
and always give it a 0 or 1 value.
This also allows us to drop an #ifdef DEBUG_AHCI in ahci_start_dma()
since the compiler can now see the local variable is used.
The motivation for this change is a recent DEBUG_AHCI build failure due
to an outdated DPRINTF() format string. From now on the compiler will
catch these errors.
Max Reitz [Tue, 11 Nov 2014 09:23:45 +0000 (10:23 +0100)]
iotests: Plain blkdebug filename generation
Add one test whether blkdebug is able to generate a plain filename if
given a configuration file and a file to be tested only; and add another
test whether blkdebug is able to do the same without being given a
configuration file.
Max Reitz [Tue, 11 Nov 2014 09:23:44 +0000 (10:23 +0100)]
blkdebug: Simplify and improve filename generation
Instead of actually recreating the options from scratch, just reuse the
options given for creating the BDS, which are the configuration file
name and additional options. In case there are no additional options we
can thus create a plain filename.
This obviously results in a different output for qemu-iotest 099 which
exactly tests this filename generation. Fix it up as well.
Kevin Wolf [Mon, 15 Sep 2014 10:19:14 +0000 (12:19 +0200)]
block/hmp: Allow node-name in 'info block'
The optional parameter specifying a block device allows now to use a
node-name instead of a drive name (and therefore to inspect any node in
the graph). The new -n options allows listing all named nodes instead of
BlockBackends.
Stefan Hajnoczi [Wed, 19 Nov 2014 14:19:45 +0000 (14:19 +0000)]
blockdev: acquire AioContext in change-backing-file
Add dataplane support to the change-backing-file QMP commands. By
acquiring the AioContext we avoid race conditions with the dataplane
thread which may also be accessing the BlockDriverState.
Note that this command operates on both bs and a node in its chain
(image_bs). The bdrv_chain_contains(bs, image_bs) check guarantees that
bs and image_bs are in the same AioContext.
Stefan Hajnoczi [Wed, 19 Nov 2014 14:19:44 +0000 (14:19 +0000)]
blockdev: acquire AioContext in eject, change, and block_passwd
By acquiring the AioContext we avoid race conditions with the dataplane
thread which may also be accessing the BlockDriverState.
Fix up eject, change, and block_passwd in a single patch because
qmp_eject() and qmp_change_blockdev() both call eject_device(). Also
fix block_passwd while we're tackling a command that takes a block
encryption password.
Stefan Hajnoczi [Wed, 19 Nov 2014 14:19:42 +0000 (14:19 +0000)]
blockdev: acquire AioContext in blockdev-snapshot-delete-internal-sync
Add dataplane support to the blockdev-snapshot-delete-internal-sync QMP
command. By acquiring the AioContext we avoid race conditions with the
dataplane thread which may also be accessing the BlockDriverState.
Max Reitz [Mon, 17 Nov 2014 12:31:06 +0000 (13:31 +0100)]
iotests: Use -qmp-pretty in 067
067 invokes query-block, resulting in a reference output with really
long lines (which may pose a problem in email patches and always poses a
problem when the output changes, because it is hard to see what has
actually changed). Use -qmp-pretty to mitigate this issue.
Max Reitz [Fri, 21 Nov 2014 09:29:59 +0000 (10:29 +0100)]
qjson: Drop trailing space for pretty formatting
For the pretty formatting, the functions converting QDicts and QLists to
JSON should not print a space after the comma separating objects,
because a newline will emitted immediately afterwards, making the
whitespace superfluous.
Fam Zheng [Fri, 31 Oct 2014 03:32:57 +0000 (11:32 +0800)]
qmp: Add optional switch "query-nodes" in query-blockstats
This bool option will allow query all the node names. It iterates all
the BDSes that are assigned a name, also in this case don't query up the
backing chain.
David Gibson [Thu, 27 Nov 2014 05:48:10 +0000 (16:48 +1100)]
Fix for crash after migration in virtio-rng on bi-endian targets
VirtIO devices now remember which endianness they're operating in in order
to support targets which may have guests of either endianness, such as
powerpc. This endianness state is transferred in a subsection of the
virtio device's information.
With virtio-rng this can lead to an abort after a loadvm hitting the
assert() in virtio_is_big_endian(). This can be reproduced by doing a
migrate and load from file on a bi-endian target with a virtio-rng device.
The actual guest state isn't particularly important to triggering this.
The cause is that virtio_rng_load_device() calls virtio_rng_process() which
accesses the ring and thus needs the endianness. However,
virtio_rng_process() is called via virtio_load() before it loads the
subsections. Essentially the ->load callback in VirtioDeviceClass should
only be used for actually reading the device state from the stream, not for
post-load re-initialization.
This patch fixes the bug by moving the virtio_rng_process() after the call
to virtio_load(). Better yet would be to convert virtio to use vmsd and
have the virtio_rng_process() as a post_load callback, but that's a bigger
project for another day.
This is bugfix, and should be considered for the 2.2 branch.
Jason Wang [Thu, 27 Nov 2014 10:04:03 +0000 (18:04 +0800)]
virtio-net: fix unmap leak
virtio_net_handle_ctrl() and other functions that process control vq
request call iov_discard_front() which will shorten the iov. This will
lead unmapping in virtqueue_push() leaks mapping.
Fixes this by keeping the original iov untouched and using a temp variable
in those functions.
Marcel Apfelbaum [Wed, 26 Nov 2014 11:50:01 +0000 (13:50 +0200)]
hmp: fix regression of HMP device_del auto-completion
The commits:
- 6a1fa9f5 (monitor: add del completion for peripheral device)
- 66e56b13 (qdev: add qdev_build_hotpluggable_device_list helper)
cause a QEMU crash when trying to use HMP device_del auto-completion.
It can be easily reproduced by:
<qemu-bin> -enable-kvm ~/images/fedora.qcow2 -monitor stdio -device virtio-net-pci,id=vnet
(qemu) device_del
/home/mapfelba/git/upstream/qemu/hw/core/qdev.c:941:qdev_build_hotpluggable_device_list: Object 0x7f6ce04e4fe0 is not an instance of type device
Aborted (core dumped)
The root cause is qdev_build_hotpluggable_device_list going recursively over
all peripherals and their children assuming all are devices. It doesn't work
since PCI devices have at least on child which is a memory region (bus master).
Solved by observing that all devices appear as direct children of
/machine/peripheral container. No need of going recursively
over all the children.
Peter Maydell [Tue, 25 Nov 2014 18:21:45 +0000 (18:21 +0000)]
qemu-timer: Avoid overflows when converting timeout to struct timespec
In qemu_poll_ns(), when we convert an int64_t nanosecond timeout into
a struct timespec, we may accidentally run into overflow problems if
the timeout is very long. This happens because the tv_sec field is a
time_t, which is signed, so we might end up setting it to a negative
value by mistake. This will result in what was intended to be a
near-infinite timeout turning into an instantaneous timeout, and we'll
busy loop. Cap the maximum timeout at INT32_MAX seconds (about 68 years)
to avoid this problem.
This specifically manifested on ARM hosts as an extreme slowdown on
guest shutdown (when the guest reprogrammed the PL031 RTC to not
generate alarms using a very long timeout) but could happen on other
hosts and guests too.
Peter Maydell [Wed, 26 Nov 2014 12:18:00 +0000 (12:18 +0000)]
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
The final 2.2 patches from me.
# gpg: Signature made Wed 26 Nov 2014 11:12:25 GMT using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <[email protected]>"
# gpg: aka "Paolo Bonzini <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
* remotes/bonzini/tags/for-upstream:
s390x/kvm: Fix compile error
fw_cfg: fix boot order bug when dynamically modified via QOM
-machine vmport=auto: Fix handling of VMWare ioport emulation for xen
Gonglei [Tue, 25 Nov 2014 04:38:19 +0000 (12:38 +0800)]
fw_cfg: fix boot order bug when dynamically modified via QOM
When we dynamically modify boot order, the length of
boot order will be changed, but we don't update
s->files->f[i].size with new length. This casuse
seabios read a wrong vale of qemu cfg file about
bootorder.
Gerd Hoffmann [Tue, 25 Nov 2014 13:54:17 +0000 (14:54 +0100)]
input: move input-send-event into experimental namespace
Ongoing discussions on how we are going to specify the console,
so tag the command as experiental so we can refine things in
the 2.3 development cycle.
Peter Maydell [Mon, 24 Nov 2014 19:31:50 +0000 (19:31 +0000)]
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
pc, pci, misc bugfixes
A bunch of bugfixes for 2.2.
Signed-off-by: Michael S. Tsirkin <[email protected]>
# gpg: Signature made Mon 24 Nov 2014 18:59:47 GMT using RSA key ID D28D5469
# gpg: Good signature from "Michael S. Tsirkin <[email protected]>"
# gpg: aka "Michael S. Tsirkin <[email protected]>"
* remotes/mst/tags/for_upstream:
pc: acpi: mark all possible CPUs as enabled in SRAT
pcie: fix improper use of negative value
pcie: fix typo in pcie_cap_deverr_init()
target-i386: move generic memory hotplug methods to DSDTs
acpi-build: mark RAM dirty on table update
hw/pci: fix crash on shpc error flow
pc: count in 1Gb hugepage alignment when sizing hotplug-memory container
pc: explicitly check maxmem limit when adding DIMM
pc: pc-dimm: use backend alignment during address auto allocation
pc: align DIMM's address/size by backend's alignment value
memory: expose alignment used for allocating RAM as MemoryRegion API
pc: limit DIMM address and size to page aligned values
pc: make pc_dimm_plug() more readble
pc: kvm: check if KVM has free memory slots to avoid abort()
qemu-char: fix tcp_get_fds
Igor Mammedov [Mon, 10 Nov 2014 16:20:50 +0000 (16:20 +0000)]
pc: acpi: mark all possible CPUs as enabled in SRAT
If QEMU is started with -numa ... Windows only notices that
CPU has been hot-added but it will not online such CPUs.
It's caused by the fact that possible CPUs are flagged as
not enabled in SRAT and Windows honoring that information
doesn't use corresponding CPU.
ACPI 5.0 Spec regarding to flag says:
"
Table 5-47 Local APIC Flags
...
Enabled: if zero, this processor is unusable, and the operating system
support will not attempt to use it.
"
Fix QEMU to adhere to spec and mark possible CPUs as enabled
in SRAT.
With that Windows onlines hot-added CPUs as expected.
acpi build modifies internal FW CFG RAM on first access
but we forgot to mark it dirty.
If this RAM has been migrated already, it won't be
migrated again, returning corrupted tables to guest.
If the pci bridge enters in error flow as part
of init process it will only delete the shpc mmio
subregion but not remove it from the properties list,
resulting in segmentation fault when the bridge runs
the exit function.
Example: add a pci bridge without specifing the chassis number:
<qemu-bin> ... -device pci-bridge,id=p1
Result:
(qemu) qemu-system-x86_64: -device pci-bridge,id=p1: Bridge chassis not specified. Each bridge is required to be assigned a unique chassis id > 0.
qemu-system-x86_64: -device pci-bridge,id=p1: Device
initialization failed.
Segmentation fault (core dumped)
if (child->class->unparent) {
#0 0x00005555558d629b in object_finalize_child_property (obj=0x555556d2e830, name=0x555556d30630 "shpc-mmio[0]", opaque=0x555556a42fc8) at qom/object.c:1078
#1 0x00005555558d4b1f in object_property_del_all (obj=0x555556d2e830) at qom/object.c:367
#2 0x00005555558d4ca1 in object_finalize (data=0x555556d2e830) at qom/object.c:412
#3 0x00005555558d55a1 in object_unref (obj=0x555556d2e830) at qom/object.c:720
#4 0x000055555572c907 in qdev_device_add (opts=0x5555563544f0) at qdev-monitor.c:566
#5 0x0000555555744f16 in device_init_func (opts=0x5555563544f0, opaque=0x0) at vl.c:2213
#6 0x00005555559cf5f0 in qemu_opts_foreach (list=0x555555e0f8e0 <qemu_device_opts>, func=0x555555744efa <device_init_func>, opaque=0x0, abort_on_failure=1) at util/qemu-option.c:1057
#7 0x000055555574a11b in main (argc=16, argv=0x7fffffffdde8, envp=0x7fffffffde70) at vl.c:423
Unparent the shpc mmio region as part of shpc cleanup.
Igor Mammedov [Fri, 31 Oct 2014 16:38:42 +0000 (16:38 +0000)]
pc: count in 1Gb hugepage alignment when sizing hotplug-memory container
if DIMMs with different size/alignment are interleaved
in creation order, it could lead to hotplug-memory
container fragmentation and following inability to use
all RAM upto maxmem.
For example:
-m 4G,slots=3,maxmem=7G
-object memory-backend-file,id=mem-1,size=256M,mem-path=/pagesize-2MB
-device pc-dimm,id=mem1,memdev=mem-1
-object memory-backend-file,id=mem-2,size=1G,mem-path=/pagesize-1GB
-device pc-dimm,id=mem2,memdev=mem-2
-object memory-backend-file,id=mem-3,size=256M,mem-path=/pagesize-2MB
-device pc-dimm,id=mem3,memdev=mem-3
fragments hotplug-memory container and doesn't allow
to use 1GB hugepage backend to consume remainig 1Gb.
To ease managment factor count in max 1Gb alignment for
each memory slot when sizing hotplug-memory region so
that regadless of fragmentaion it would be possible to
add max aligned DIMM.
Igor Mammedov [Fri, 31 Oct 2014 16:38:41 +0000 (16:38 +0000)]
pc: explicitly check maxmem limit when adding DIMM
Currently maxmem limit is not checked and depends on
hotplug region container not being able to fit more RAM
than maxmem. Do check explicitly so that it would
be possible to change hotplug container size later
to deal with fragmentation.
Peter Maydell [Mon, 24 Nov 2014 13:50:22 +0000 (13:50 +0000)]
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
Three patches to fix ExtINT for the QEMU implementation of the local APIC.
# gpg: Signature made Mon 24 Nov 2014 13:38:36 GMT using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <[email protected]>"
# gpg: aka "Paolo Bonzini <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
* remotes/bonzini/tags/for-upstream:
apic: fix incorrect handling of ExtINT interrupts wrt processor priority
apic: fix loss of IPI due to masked ExtINT
apic: avoid getting out of halted state on masked PIC interrupts
Paolo Bonzini [Tue, 11 Nov 2014 12:14:18 +0000 (13:14 +0100)]
apic: fix incorrect handling of ExtINT interrupts wrt processor priority
This fixes another failure with ExtINT, demonstrated by QNX. The failure
mode is as follows:
- IPI sent to cpu 0 (bit set in APIC irr)
- IPI accepted by cpu 0 (bit cleared in irr, set in isr)
- IPI sent to cpu 0 (bit set in both irr and isr)
- PIC interrupt sent to cpu 0
The PIC interrupt causes CPU_INTERRUPT_HARD to be set, but
apic_irq_pending observes that the highest pending APIC interrupt priority
(the IPI) is the same as the processor priority (since the IPI is still
being handled), so apic_get_interrupt returns a spurious interrupt rather
than the pending PIC interrupt. The result is an endless sequence of
spurious interrupts, since nothing will clear CPU_INTERRUPT_HARD.
Instead, ExtINT interrupts should have ignored the processor priority.
Calling apic_check_pic early in apic_get_interrupt ensures that
apic_deliver_pic_intr is called instead of delivering the spurious
interrupt. apic_deliver_pic_intr then clears CPU_INTERRUPT_HARD if needed.
Paolo Bonzini [Tue, 11 Nov 2014 12:14:14 +0000 (13:14 +0100)]
apic: fix loss of IPI due to masked ExtINT
This patch fixes an obscure failure of the QNX kernel on QEMU x86 SMP.
In QNX, all hardware interrupts come via the PIC, and are delivered by
the cpu 0 LAPIC in ExtINT mode, while IPIs are delivered by the LAPIC
in fixed mode.
This bug happens as follows:
- cpu 0 masks a particular PIC interrupt
- IPI sent to cpu 0 (CPU_INTERRUPT_HARD is set)
- before the IPI is accepted, the masked interrupt line is asserted by the
device
Since the interrupt is masked, apic_deliver_pic_intr will clear
CPU_INTERRUPT_HARD. The IPI will still be set in the APIC irr, but since
CPU_INTERRUPT_HARD is not set the cpu will not notice. Depending on the
scenario this can cause a system hang, i.e. if cpu 0 is expected to unmask
the interrupt.
In order to fix this, do a full check of the APIC before an EXTINT
is acknowledged. This can result in clearing CPU_INTERRUPT_HARD, but
can also result in delivering the lost IPI.
Paolo Bonzini [Tue, 11 Nov 2014 12:14:05 +0000 (13:14 +0100)]
apic: avoid getting out of halted state on masked PIC interrupts
After the next patch, if a masked PIC interrupts causes CPU_INTERRUPT_POLL
to be set, the CPU will spuriously get out of halted state. While this
is technically valid, we should avoid that.
Make CPU_INTERRUPT_POLL run apic_update_irq in the right thread and then
look at CPU_INTERRUPT_HARD. If CPU_INTERRUPT_HARD does not get set,
do not report the CPU as having work.
Also move the handling of software-disabled APIC from apic_update_irq
to apic_irq_pending, and always trigger CPU_INTERRUPT_POLL. This will
be important once we will add a case that resets CPU_INTERRUPT_HARD
from apic_update_irq. We want to run it even if we go through
CPU_INTERRUPT_POLL, and even if the local APIC is software disabled.
The main reason for reverting this commit before the 2.2 release is that
it adds a QAPI interface that we don't want to keep: The 'nocow' flag
doesn't generally make sense for block nodes, but only for the raw-posix
driver. It should therefore be part of ImageInfoSpecific rather than
ImageInfo.
The commit contains more problems, but unlike the API stability issue
they wouldn't justify reverting it.
Igor Mammedov [Fri, 31 Oct 2014 16:38:36 +0000 (16:38 +0000)]
pc: limit DIMM address and size to page aligned values
When running in KVM mode, kvm_set_phys_mem() will silently
fail if registered MemoryRegion address/size is not page
aligned. Causing memory hotplug failure in guest.
Mapping non aligned MemoryRegion in TCG mode 'works', but
sane guest OS still expects page aligned memory module
and fails to initialize it if it's not aligned.
So do not allow non aligned (i.e. valid) address/size
values for DIMM to avoid either KVM failure or guest
issues caused by it.
Gonglei [Thu, 20 Nov 2014 11:35:02 +0000 (19:35 +0800)]
pcnet: fix Negative array index read
s->xmit_pos maybe assigned to a negative value (-1),
but in this branch variable s->xmit_pos as an index to
array s->buffer. Let's add a check for s->xmit_pos.
Gonglei [Thu, 20 Nov 2014 11:35:00 +0000 (19:35 +0800)]
net/slirp: fix memory leak
commit b412eb61 introduce 'cmd:' target for guestfwd,
and fwd don't be used in this scenario, and will leak
memory in true branch with 'cmd:'. Let's allocate memory
for fwd variable just in else statement.
Leif Lindholm [Wed, 19 Nov 2014 11:08:45 +0000 (11:08 +0000)]
hw/arm/virt: set stdout-path instead of linux,stdout-path
ePAPR 1.1 defines the stdout-path property, making the os-specific
linux,stdout-path property redundant. Change the DT setup for ARM virt
to use the generic property - supported by Linux since 3.15.
The old QEMU behaviour was not present in any released version of
QEMU, and was only added to QEMU after the kernel changed, so
this should not break any existing setups.
Signed-off-by: Leif Lindholm <[email protected]>
[PMM: add note to commit about the old behaviour never hving been
in a released version of QEMU] Signed-off-by: Peter Maydell <[email protected]>
The Move to Vector Status and Control Register (mtvscr) instruction
uses VRB as the source register. Fix the code generator to correctly
decode the VRB field. That is, use "rB(ctx->opcode)" instead of
"rD(ctx->opcode)".
Alexander Graf [Fri, 7 Nov 2014 21:12:48 +0000 (22:12 +0100)]
kvm: Fix memory slot page alignment logic
Memory slots have to be page aligned to get entered into KVM. There
is existing logic that tries to ensure that we pad memory slots that
are not page aligned to the biggest region that would still fit in the
alignment requirements.
Unfortunately, that logic is broken. It tries to calculate the start
offset based on the region size.
Fix up the logic to do the thing it was intended to do and document it
properly in the comment above it.
With this patch applied, I can successfully run an e500 guest with more
than 3GB RAM (at which point RAM starts overlapping subpage memory regions).
Peter Maydell [Thu, 20 Nov 2014 13:00:28 +0000 (13:00 +0000)]
Merge remote-tracking branch 'remotes/amit-migration/tags/for-2.2-2' into staging
Fix from a while back that unfortunately got ignored. Dave Gilbert says
it may actually fix a case where autoconverge would break on a repeat
migration (and not just fix stats).
Tracing: Fix simpletrace.py error on tcg enabled binary traces
simpletrace.py does not recognize the tcg option while reading trace-events file. In result simpletrace does not work on binary traces and tcg enabled events. Moved transformation of tcg enabled events to _read_events() which is used by simpletrace.
Peter Maydell [Tue, 18 Nov 2014 13:43:37 +0000 (13:43 +0000)]
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block patches for 2.2.0-rc2
# gpg: Signature made Tue 18 Nov 2014 11:32:55 GMT using RSA key ID C88F2FD6
# gpg: Good signature from "Kevin Wolf <[email protected]>"
* remotes/kevin/tags/for-upstream:
block/raw-posix: Catch fsync() errors
block/raw-posix: Only sync after successful preallocation
block/raw-posix: Fix preallocating write() loop
raw-posix: The SEEK_HOLE code is flawed, rewrite it
raw-posix: SEEK_HOLE suffices, get rid of FIEMAP
raw-posix: Fix comment for raw_co_get_block_status()
During migration, the values read from migration stream during ram load
are not validated. Especially offset in host_from_stream_offset() and
also the length of the writes in the callers of said function.
To fix this, we need to make sure that the [offset, offset + length]
range fits into one of the allocated memory regions.
Validating addr < len should be sufficient since data seems to always be
managed in TARGET_PAGE_SIZE chunks.
Fixes: CVE-2014-7840
Note: follow-up patches add extra checks on each block->host access.
Max Reitz [Tue, 18 Nov 2014 10:23:04 +0000 (11:23 +0100)]
block/raw-posix: Fix preallocating write() loop
write() may write less bytes than requested; in this case, the number of
bytes written is returned. This is the byte count we should be
subtracting from the number of bytes still to be written, and not the
byte count we requested to write.
Peter Maydell [Sun, 16 Nov 2014 19:44:21 +0000 (19:44 +0000)]
exec: Handle multipage ranges in invalidate_and_set_dirty()
The code in invalidate_and_set_dirty() needs to handle addr/length
combinations which cross guest physical page boundaries. This can happen,
for example, when disk I/O reads large blocks into guest RAM which previously
held code that we have cached translations for. Unfortunately we were only
checking the clean/dirty status of the first page in the range, and then
were calling a tb_invalidate function which only handles ranges that don't
cross page boundaries. Fix the function to deal with multipage ranges.
The symptoms of this bug were that guest code would misbehave (eg segfault),
in particular after a guest reboot but potentially any time the guest
reused a page of its physical RAM for new code.
Kevin Wolf [Tue, 18 Nov 2014 10:01:05 +0000 (11:01 +0100)]
Merge remote-tracking branch 'mreitz/block' into queue-block
* mreitz/block:
raw-posix: The SEEK_HOLE code is flawed, rewrite it
raw-posix: SEEK_HOLE suffices, get rid of FIEMAP
raw-posix: Fix comment for raw_co_get_block_status()
raw-posix: The SEEK_HOLE code is flawed, rewrite it
On systems where SEEK_HOLE in a trailing hole seeks to EOF (Solaris,
but not Linux), try_seek_hole() reports trailing data instead.
Additionally, unlikely lseek() failures are treated badly:
* When SEEK_HOLE fails, try_seek_hole() reports trailing data. For
-ENXIO, there's in fact a trailing hole. Can happen only when
something truncated the file since we opened it.
* When SEEK_HOLE succeeds, SEEK_DATA fails, and SEEK_END succeeds,
then try_seek_hole() reports a trailing hole. This is okay only
when SEEK_DATA failed with -ENXIO (which means the non-trailing hole
found by SEEK_HOLE has since become trailing somehow). For other
failures (unlikely), it's wrong.
* When SEEK_HOLE succeeds, SEEK_DATA fails, SEEK_END fails (unlikely),
then try_seek_hole() reports bogus data [-1,start), which its caller
raw_co_get_block_status() turns into zero sectors of data. Could
theoretically lead to infinite loops in code that attempts to scan
data vs. hole forward.
Commit 5500316 (May 2012) implemented raw_co_is_allocated() as
follows:
1. If defined(CONFIG_FIEMAP), use the FS_IOC_FIEMAP ioctl
2. Else if defined(SEEK_HOLE) && defined(SEEK_DATA), use lseek()
3. Else pretend there are no holes
Later on, raw_co_is_allocated() was generalized to
raw_co_get_block_status().
Commit 4f11aa8 (May 2014) changed it to try the three methods in order
until success, because "there may be implementations which support
[SEEK_HOLE/SEEK_DATA] but not [FIEMAP] (e.g., NFSv4.2) as well as vice
versa."
Unfortunately, we used FIEMAP incorrectly: we lacked FIEMAP_FLAG_SYNC.
Commit 38c4d0a (Sep 2014) added it. Because that's a significant
speed hit, the next commit 7c159037 put SEEK_HOLE/SEEK_DATA first.
As you see, the obvious use of FIEMAP is wrong, and the correct use is
slow. I guess this puts it somewhere between -7 "The obvious use is
wrong" and -10 "It's impossible to get right" on Rusty Russel's Hard
to Misuse scale[*].
"Fortunately", the FIEMAP code is used only when
* SEEK_HOLE/SEEK_DATA aren't defined, but CONFIG_FIEMAP is
Uncommon. SEEK_HOLE had no XFS implementation between 2011 (when it
was introduced for ext4 and btrfs) and 2012.
* SEEK_HOLE/SEEK_DATA and CONFIG_FIEMAP are defined, but lseek() fails
Unlikely.
Thus, the FIEMAP code executes rarely. Makes it a nice hidey-hole for
bugs. Worse, bugs hiding there can theoretically bite even on a host
that has SEEK_HOLE/SEEK_DATA.
I don't want to worry about this crap, not even theoretically. Get
rid of it.
Peter Maydell [Thu, 13 Nov 2014 14:56:09 +0000 (14:56 +0000)]
target-arm: handle address translations that start at level 3
The ARMv8 address translation system defines that a page table walk
starts at a level which depends on the translation granule size
and the number of bits of virtual address that need to be resolved.
Where the translation granule is 64KB and the guest sets the
TCR.TxSZ field to between 35 and 39, it's actually possible to
start at level 3 (the final level). QEMU's implementation failed
to handle this case, and so we would set level to 2 and behave
incorrectly (including invoking the C undefined behaviour of
shifting left by a negative number). Correct the code that
determines the starting level to deal with the start-at-3 case,
by replacing the if-else ladder with an expression derived from
the ARM ARM pseudocode version.
This error was detected by the Coverity scan, which spotted
the potential shift by a negative number.
Peter Maydell [Mon, 17 Nov 2014 17:22:03 +0000 (17:22 +0000)]
Merge remote-tracking branch 'remotes/bonzini/tags/for-upstream' into staging
A smattering of fixes for problems that Coverity reported.
# gpg: Signature made Mon 17 Nov 2014 17:03:25 GMT using RSA key ID 78C7AE83
# gpg: Good signature from "Paolo Bonzini <[email protected]>"
# gpg: aka "Paolo Bonzini <[email protected]>"
# gpg: WARNING: This key is not certified with sufficiently trusted signatures!
# gpg: It is not certain that the signature belongs to the owner.
# Primary key fingerprint: 46F5 9FBD 57D6 12E7 BFD4 E2F7 7E15 100C CD36 69B1
# Subkey fingerprint: F133 3857 4B66 2389 866C 7682 BFFB D25F 78C7 AE83
projects / qemu.git / log