Paul Blakey [Fri, 9 Jun 2023 12:22:59 +0000 (15:22 +0300)]
net/sched: act_ct: Fix promotion of offloaded unreplied tuple
Currently UNREPLIED and UNASSURED connections are added to the nf flow
table. This causes the following connection packets to be processed
by the flow table which then skips conntrack_in(), and thus such the
connections will remain UNREPLIED and UNASSURED even if reply traffic
is then seen. Even still, the unoffloaded reply packets are the ones
triggering hardware update from new to established state, and if
there aren't any to triger an update and/or previous update was
missed, hardware can get out of sync with sw and still mark
packets as new.
Fix the above by:
1) Not skipping conntrack_in() for UNASSURED packets, but still
refresh for hardware, as before the cited patch.
2) Try and force a refresh by reply-direction packets that update
the hardware rules from new to established state.
3) Remove any bidirectional flows that didn't failed to update in
hardware for re-insertion as bidrectional once any new packet
arrives.
Hugh Dickins [Fri, 9 Jun 2023 21:29:39 +0000 (14:29 -0700)]
wifi: iwlwifi: mvm: spin_lock_bh() to fix lockdep regression
Lockdep on 6.4-rc on ThinkPad X1 Carbon 5th says
=====================================================
WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
6.4.0-rc5 #1 Not tainted
-----------------------------------------------------
kworker/3:1/49 [HC0[0]:SC0[4]:HE1:SE0] is trying to acquire: ffff8881066fa368 (&mvm_sta->deflink.lq_sta.rs_drv.pers.lock){+.+.}-{2:2}, at: rs_drv_get_rate+0x46/0xe7
and this task is already holding: ffff8881066f80a8 (&sta->rate_ctrl_lock){+.-.}-{2:2}, at: rate_control_get_rate+0xbd/0x126
which would create a new lock dependency:
(&sta->rate_ctrl_lock){+.-.}-{2:2} -> (&mvm_sta->deflink.lq_sta.rs_drv.pers.lock){+.+.}-{2:2}
but this new dependency connects a SOFTIRQ-irq-safe lock:
(&sta->rate_ctrl_lock){+.-.}-{2:2}
etc. etc. etc.
Changing the spin_lock() in rs_drv_get_rate() to spin_lock_bh() was not
enough to pacify lockdep, but changing them all on pers.lock has worked.
Vlad Buslov [Mon, 12 Jun 2023 07:57:12 +0000 (09:57 +0200)]
selftests/tc-testing: Remove configs that no longer exist
Some qdiscs and classifiers have recently been retired from kernel.
However, tc-testing config is still cluttered with them which causes noise
when using merge_config.sh script to update existing config for tc-testing
compatibility. Remove the config settings for affected qdiscs and
classifiers.
Vlad Buslov [Mon, 12 Jun 2023 07:57:11 +0000 (09:57 +0200)]
selftests/tc-testing: Fix SFB db test
Setting very small value of db like 10ms introduces rounding errors when
converting to/from jiffies on some kernel configs. For example, on 250hz
the actual value will be set to 12ms which causes the test to fail:
# $ sudo ./tdc.py -d eth2 -e 3410
# -- ns/SubPlugin.__init__
# Test 3410: Create SFB with db setting
#
# All test results:
#
# 1..1
# not ok 1 3410 - Create SFB with db setting
# Could not match regex pattern. Verify command output:
# qdisc sfb 1: root refcnt 2 rehash 600s db 12ms limit 1000p max 25p target 20p increment 0.000503548 decrement 4.57771e-05 penalty_rate 10pps penalty_burst 20p
Set the value to 100ms instead which currently seem to work on 100hz,
250hz, 300hz and 1000hz kernel configs.
Fixes: 6ad92dc56fca ("selftests/tc-testing: add selftests for sfb qdisc") Signed-off-by: Vlad Buslov <[email protected]> Reviewed-by: Pedro Tammela <[email protected]> Signed-off-by: Jakub Kicinski <[email protected]>
Jens Axboe [Mon, 12 Jun 2023 19:51:36 +0000 (13:51 -0600)]
io_uring/net: save msghdr->msg_control for retries
If the application sets ->msg_control and we have to later retry this
command, or if it got queued with IOSQE_ASYNC to begin with, then we
need to retain the original msg_control value. This is due to the net
stack overwriting this field with an in-kernel pointer, to copy it
in. Hitting that path for the second time will now fail the copy from
user, as it's attempting to copy from a non-user address.
Merge branch 'bpf: fix NULL dereference during extable search'
Krister Johansen says:
====================
Hi,
Enclosed are a pair of patches for an oops that can occur if an exception is
generated while a bpf subprogram is running. One of the bpf_prog_aux entries
for the subprograms are missing an extable. This can lead to an exception that
would otherwise be handled turning into a NULL pointer bug.
These changes were tested via the verifier and progs selftests and no
regressions were observed.
Changes from v4:
- Ensure that num_exentries is copied to prog->aux from func[0] (Feedback from
Ilya Leoshkevich)
Changes from v3:
- Selftest style fixups (Feedback from Yonghong Song)
- Selftest needs to assert that test bpf program executed (Feedback from
Yonghong Song)
- Selftest should combine open and load using open_and_load (Feedback from
Yonghong Song)
Changes from v2:
- Insert only the main program's kallsyms (Feedback from Yonghong Song and
Alexei Starovoitov)
- Selftest should use ASSERT instead of CHECK (Feedback from Yonghong Song)
- Selftest needs some cleanup (Feedback from Yonghong Song)
- Switch patch order (Feedback from Alexei Starovoitov)
Changes from v1:
- Add a selftest (Feedback From Alexei Starovoitov)
- Move to a 1-line verifier change instead of searching multiple extables
====================
Krister Johansen [Tue, 13 Jun 2023 00:44:54 +0000 (17:44 -0700)]
selftests/bpf: add a test for subprogram extables
In certain situations a program with subprograms may have a NULL
extable entry. This should not happen, and when it does, it turns a
single trap into multiple. Add a test case for further debugging and to
prevent regressions.
The test-case contains three essentially identical versions of the same
test because just one program may not be sufficient to trigger the oops.
This is due to the fact that the items are stored in a binary tree and
have identical values so it's possible to sometimes find the ksym with
the extable. With 3 copies, this has been reliable on this author's
test systems.
When triggered out of this test case, the oops looks like this:
Krister Johansen [Tue, 13 Jun 2023 00:44:40 +0000 (17:44 -0700)]
bpf: ensure main program has an extable
When subprograms are in use, the main program is not jit'd after the
subprograms because jit_subprogs sets a value for prog->bpf_func upon
success. Subsequent calls to the JIT are bypassed when this value is
non-NULL. This leads to a situation where the main program and its
func[0] counterpart are both in the bpf kallsyms tree, but only func[0]
has an extable. Extables are only created during JIT. Now there are
two nearly identical program ksym entries in the tree, but only one has
an extable. Depending upon how the entries are placed, there's a chance
that a fault will call search_extable on the aux with the NULL entry.
Since jit_subprogs already copies state from func[0] to the main
program, include the extable pointer in this state duplication.
Additionally, ensure that the copy of the main program in func[0] is not
added to the bpf_prog_kallsyms table. Instead, let the main program get
added later in bpf_prog_load(). This ensures there is only a single
copy of the main program in the kallsyms table, and that its tag matches
the tag observed by tooling like bpftool.
Janne Grunau [Sun, 12 Feb 2023 12:16:32 +0000 (13:16 +0100)]
nios2: dts: Fix tse_mac "max-frame-size" property
The given value of 1518 seems to refer to the layer 2 ethernet frame
size without 802.1Q tag. Actual use of the "max-frame-size" including in
the consumer of the "altr,tse-1.0" compatible is the MTU.
This patch disables the TOPDOWN flag for APU and few dGPU cards
which has the VRAM size equal to the BAR size.
When we enable the TOPDOWN flag, we get the free blocks at
the highest available memory region and we don't split the
lower order blocks. This change is required to keep off
the fragmentation related issues particularly in ASIC
which has VRAM space <= 500MiB
Sonny Jiang [Tue, 6 Jun 2023 21:18:52 +0000 (17:18 -0400)]
drm/amdgpu: vcn_4_0 set instance 0 init sched score to 1
Only vcn0 can process AV1 codecx. In order to use both vcn0 and
vcn1 in h264/265 transcode to AV1 cases, set vcn0 sched score to 1
at initialization time.
The buffer object backing the fbdev emulation got pinned twice: by the
fb_probe helper radeon_fbdev_create_pinned_object() and the modesetting
code when the framebuffer got displayed. It only got unpinned once by
the fbdev helper radeon_fbdev_destroy_pinned_object(). Hence TTM's BO-
release function complains about the pin counter. Forcing the outputs
off also undoes the modesettings pin increment.
Kenneth Feng [Thu, 8 Jun 2023 14:07:11 +0000 (22:07 +0800)]
drm/amd/pm: workaround for compute workload type on some skus
On smu 13.0.0, the compute workload type cannot be set on all the skus
due to some other problems. This workaround is to make sure compute workload type
can also run on some specific skus.
drm/amd: Tighten permissions on VBIOS flashing attributes
Non-root users shouldn't be able to try to trigger a VBIOS flash
or query the flashing status. This should be reserved for users with the
appropriate permissions.
drm/amd: Make sure image is written to trigger VBIOS image update flow
The VBIOS image update flow requires userspace to:
1) Write the image to `psp_vbflash`
2) Read `psp_vbflash`
3) Poll `psp_vbflash_status` to check for completion
If userspace reads `psp_vbflash` before writing an image, it's
possible that it causes problems that can put the dGPU into an invalid
state.
Explicitly check that an image has been written before letting a read
succeed.
Jiadong Zhu [Thu, 25 May 2023 08:52:55 +0000 (16:52 +0800)]
drm/amdgpu: Modify indirect buffer packages for resubmission
When the preempted IB frame resubmitted to cp, we need to modify the frame
data including:
1. set PRE_RESUME 1 in CONTEXT_CONTROL.
2. use meta data(DE and CE) read from CSA in WRITE_DATA.
Add functions to save the location the first time IBs emitted and callback
to patch the package when resubmission happens.
Natalia Petrova [Fri, 12 May 2023 10:33:20 +0000 (13:33 +0300)]
drm/nouveau: add nv_encoder pointer check for NULL
Pointer nv_encoder could be dereferenced at nouveau_connector.c
in case it's equal to NULL by jumping to goto label.
This patch adds a NULL-check to avoid it.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Jiadong Zhu [Wed, 24 May 2023 03:42:19 +0000 (11:42 +0800)]
drm/amdgpu: Reset CP_VMID_PREEMPT after trailing fence signaled
When MEC executes unmap_queue for mid command buffer preemption, it will
kick the write pointer of the gfx ring, set CP_VMID_PREEMPT to trigger the
preemption and wait for CP_VMID_PREEMPT becomes zero after the preemption
done. There is a race condition that PFP may excute the resetting command
before MEC set CP_VMID_PREEMPT. As a result, hang happens as
CP_VMID_PREEMPT is always 0xffff.
To avoid this, we send resetting CP_VMID_PREEMPT command after the trailing
fence is siganled and update gfx write pointer explicitly.
Natalia Petrova [Fri, 12 May 2023 11:15:26 +0000 (14:15 +0300)]
drm/nouveau/dp: check for NULL nv_connector->native_mode
Add checking for NULL before calling nouveau_connector_detect_depth() in
nouveau_connector_get_modes() function because nv_connector->native_mode
could be dereferenced there since connector pointer passed to
nouveau_connector_detect_depth() and the same value of
nv_connector->native_mode is used there.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Abe Kohandel [Tue, 13 Jun 2023 16:21:03 +0000 (09:21 -0700)]
spi: dw: Replace incorrect spi_get_chipselect with set
Commit 445164e8c136 ("spi: dw: Replace spi->chip_select references with
function calls") replaced direct access to spi.chip_select with
spi_*_chipselect calls but incorrectly replaced a set instance with a
get instance, replace the incorrect instance.
mmc: mmci: stm32: fix max busy timeout calculation
The way that the timeout is currently calculated could lead to a u64
timeout value in mmci_start_command(). This value is then cast in a u32
register that leads to mmc erase failed issue with some SD cards.
Rob Herring [Tue, 18 Apr 2023 15:06:06 +0000 (10:06 -0500)]
dt-bindings: pinctrl: qcom,pmic-mpp: Fix schema for "qcom,paired"
The "qcom,paired" schema is all wrong. First, it's a list rather than an
object(dictionary). Second, it is missing a required type. The meta-schema
normally catches this, but schemas under "$defs" was not getting checked.
A fix for that is pending.
In case of gpio-regmap, IRQ chip is added by regmap-irq and associated with
GPIO chip by gpiochip_irqchip_add_domain(). The initialization flag was not
added in gpiochip_irqchip_add_domain(), causing gpiochip_to_irq() to return
-EPROBE_DEFER.
Fixes: 5467801f1fcb ("gpio: Restrict usage of GPIO chip irq members before initialization") Signed-off-by: Jiawen Wu <[email protected]> Reviewed-by: Andy Shevchenko <[email protected]> Reviewed-by: Linus Walleij <[email protected]> Signed-off-by: Bartosz Golaszewski <[email protected]>
Takashi Iwai [Tue, 13 Jun 2023 11:22:40 +0000 (13:22 +0200)]
regmap: regcache: Don't sync read-only registers
regcache_maple_sync() tries to sync all cached values no matter
whether it's writable or not. OTOH, regache_sync_val() does care the
wrtability and returns -EIO for a read-only register. This results in
an error message like:
snd_hda_codec_realtek hdaudioC0D0: Unable to sync register 0x2f0009. -5
and the sync loop is aborted incompletely.
This patch adds the writable register check to regcache_sync_val() for
addressing the bug above.
Note that, although we may add the check in the caller side
(regcache_maple_sync()), here we put in regcache_sync_val(), so that a
similar case like this can be avoided in future.
Jon Hunter [Tue, 13 Jun 2023 09:34:53 +0000 (10:34 +0100)]
ASoC: tegra: Fix Master Volume Control
Commit 3ed2b549b39f ("ALSA: pcm: fix wait_time calculations") corrected
the PCM wait_time calculations and in doing so reduced the calculated
wait_time. This exposed an issue with the Tegra Master Volume Control
(MVC) device where the reduced wait_time caused the MVC to fail. For now
fix this by setting the default wait_time for Tegra to be 500ms.
Robert Hodaszi [Fri, 9 Jun 2023 12:13:34 +0000 (14:13 +0200)]
tty: serial: fsl_lpuart: reduce RX watermark to 0 on LS1028A
LS1028A is using DMA with LPUART. Having RX watermark set to 1, means
DMA transactions are started only after receiving the second character.
On other platforms with newer LPUART IP, Receiver Idle Empty function
initiates the DMA request after the receiver is idling for 4 characters.
But this feature is missing on LS1028A, which is causing a 1-character
delay in the RX direction on this platform.
Set RX watermark to 0 to initiate RX DMA after each character.
drm/nouveau: don't detect DSM for non-NVIDIA device
The call site of nouveau_dsm_pci_probe() uses single set of output
variables for all invocations. So, we must not write anything to them
unless it's an NVIDIA device. Otherwise, if we are called with another
device after the NVIDIA device, we'll clober the result of the NVIDIA
device.
For example, if the other device doesn't have _PR3 resources, the
detection later would miss the presence of power resource support, and
the rest of the code will keep using Optimus DSM, breaking power
management for that machine.
Also, because we're detecting NVIDIA's DSM, it doesn't make sense to run
this detection on a non-NVIDIA device anyway. Thus, check at the
beginning of the detection code if this is an NVIDIA card, and just
return if it isn't.
This, together with commit d22915d22ded ("drm/nouveau/devinit/tu102-:
wait for GFW_BOOT_PROGRESS == COMPLETED") developed independently and
landed earlier, fixes runtime power management of the NVIDIA card in
Lenovo Legion 5-15ARH05. Without this patch, the GPU resumption code
will "timeout", sometimes hanging userspace.
As a bonus, we'll also stop preventing _PR3 usage from the bridge for
unrelated devices, which is always nice, I guess.
usb_udc_connect_control(), soft_connect_store() and
usb_gadget_deactivate() can potentially race against each other to invoke
usb_gadget_connect()/usb_gadget_disconnect(). To prevent this, guard
udc->started, gadget->allow_connect, gadget->deactivate and
gadget->connect with connect_lock so that ->pullup() is only invoked when
the gadget is bound, started and not deactivated. The routines
usb_gadget_connect_locked(), usb_gadget_disconnect_locked(),
usb_udc_connect_control_locked(), usb_gadget_udc_start_locked(),
usb_gadget_udc_stop_locked() are called with this lock held.
An earlier version of this commit was reverted due to the crash reported in
https://lore.kernel.org/all/[email protected]/.
commit 16737e78d190 ("usb: gadget: udc: core: Offload usb_udc_vbus_handler processing")
addresses the crash reported.
usb_udc_vbus_handler() can be invoked from interrupt context by irq
handlers of the gadget drivers, however, usb_udc_connect_control() has
to run in non-atomic context due to the following:
a. Some of the gadget driver implementations expect the ->pullup
callback to be invoked in non-atomic context.
b. usb_gadget_disconnect() acquires udc_lock which is a mutex.
Hence offload invocation of usb_udc_connect_control()
to workqueue.
UDC should not be pulled up unless gadget driver is bound. The new flag
"allow_connect" is now set by gadget_bind_driver() and cleared by
gadget_unbind_driver(). This prevents work item to pull up the gadget
even if queued when the gadget driver is already unbound.
Pavan Holla [Wed, 7 Jun 2023 19:33:26 +0000 (19:33 +0000)]
usb: typec: Fix fast_role_swap_current show function
The current implementation mistakenly performs a & operation on
the output of sysfs_emit. This patch performs the & operation before
calling sysfs_emit.
Johan Hovold [Wed, 7 Jun 2023 10:05:40 +0000 (12:05 +0200)]
USB: dwc3: fix use-after-free on core driver unbind
Some dwc3 glue drivers are currently accessing the driver data of the
child core device directly, which is clearly a bad idea as the child may
not have probed yet or may have been unbound from its driver.
As a workaround until the glue drivers have been fixed, clear the driver
data pointer before allowing the glue parent device to runtime suspend
to prevent its driver from accessing data that has been freed during
unbind.
Johan Hovold [Wed, 7 Jun 2023 10:05:39 +0000 (12:05 +0200)]
USB: dwc3: qcom: fix NULL-deref on suspend
The Qualcomm dwc3 glue driver is currently accessing the driver data of
the child core device during suspend and on wakeup interrupts. This is
clearly a bad idea as the child may not have probed yet or could have
been unbound from its driver.
The first such layering violation was part of the initial version of the
driver, but this was later made worse when the hack that accesses the
driver data of the grand child xhci device to configure the wakeup
interrupts was added.
Fixing this properly is not that easily done, so add a sanity check to
make sure that the child driver data is non-NULL before dereferencing it
for now.
Note that this relies on subtleties like the fact that driver core is
making sure that the parent is not suspended while the child is probing.
usb: dwc3: gadget: Reset num TRBs before giving back the request
Consider a scenario where cable disconnect happens when there is an active
usb reqest queued to the UDC. As part of the disconnect we would issue an
end transfer with no interrupt-on-completion before giving back this
request. Since we are giving back the request without skipping TRBs the
num_trbs field of dwc3_request still holds the stale value previously used.
Function drivers re-use same request for a given bind-unbind session and
hence their dwc3_request context gets preserved across cable
disconnect/connect. When such a request gets re-queued after cable connect,
we would increase the num_trbs field on top of the previous stale value
thus incorrectly representing the number of TRBs used. Fix this by
resetting num_trbs field before giving back the request.
The reason is, it is trying to register an interrupt handler for the same
IRQ twice. The devm_request_irq() was called with the parent device.
So the interrupt handler won't be unregistered when the usb3-peri device
is unbound.
Fix this issue by replacing "parent dev"->"dev" as the irq resource
is managed by this driver.
Binbin Zhou [Wed, 24 May 2023 01:49:24 +0000 (09:49 +0800)]
clk: clk-loongson2: Zero init clk_init_data
As clk_core_populate_parent_map() checks clk_init_data.num_parents
first, and checks clk_init_data.parent_names[] before
clk_init_data.parent_data[] and clk_init_data.parent_hws[].
Therefore the clk_init_data structure needs to be explicitly initialised
to prevent an unexpected crash if clk_init_data.parent_names[] is a
random value.
The given operations are inverted for the wrong registers which makes
multiple of the mt8365 hardware units unusable. In my setup at least usb
did not work.
Fixed by swapping the operations with the inverted ones.
ULONG_MAX is used by a few drivers to figure out the highest available
clock rate via clk_round_rate(clk, ULONG_MAX). Since abs() takes a
signed value as input, the current logic effectively calculates with
ULONG_MAX = -1, which results in the worst parent clock being chosen
instead of the best one.
For example on Rockchip RK3588 the eMMC driver tries to figure out
the highest available clock rate. There are three parent clocks
available resulting in the following rate diffs with the existing
logic:
As a result the clock framework will promote a maximum supported
clock rate of 24 MHz, even though 1.5GHz are possible. With the
updated logic any casting between signed and unsigned is avoided
and the numbers look like this instead:
Jakub Kicinski [Mon, 12 Jun 2023 23:55:45 +0000 (16:55 -0700)]
Merge branch 'mptcp-fixes'
Matthieu Baerts says:
====================
selftests: mptcp: skip tests not supported by old kernels (part 3)
After a few years of increasing test coverage in the MPTCP selftests, we
realised [1] the last version of the selftests is supposed to run on old
kernels without issues.
Supporting older versions is not that easy for this MPTCP case: these
selftests are often validating the internals by checking packets that
are exchanged, when some MIB counters are incremented after some
actions, how connections are getting opened and closed in some cases,
etc. In other words, it is not limited to the socket interface between
the userspace and the kernelspace.
In addition to that, the current MPTCP selftests run a lot of different
sub-tests but the TAP13 protocol used in the selftests don't support
sub-tests: one failure in sub-tests implies that the whole selftest is
seen as failed at the end because sub-tests are not tracked. It is then
important to skip sub-tests not supported by old kernels.
To minimise the modifications and reduce the complexity to support old
versions, the idea is to look at external signs and skip the whole
selftest or just some sub-tests before starting them. This cannot be
applied in all cases.
Similar to the second part, this third one focuses on marking different
sub-tests as skipped if some MPTCP features are not supported. This
time, only in "mptcp_join.sh" selftest, the remaining one, is modified.
Several techniques are used here to achieve this task:
- Before starting some tests:
- Check if a file (sysctl knob) is present: that's what patch 12/17 is
doing for the userspace PM feature.
- Check if a required kernel symbol is present in /proc/kallsyms:
patches 9, 10, 14 and 15/17 are using this technique.
- Check if it is possible to setup a particular network environment
requiring Netfilter or TC: if the preparation step fail, the linked
sub-test is marked as skipped. Patch 5/17 is doing that.
- Check if a MIB counter is available: patches 7 and 13/17 do that.
- Check if the kernel version is newer than a specific one: patch 1/17
adds some helpers in mptcp_lib.sh to ease its use. That's not ideal
and it is only used as last resort but as mentioned above, it is
important to skip tests if they are not supported not to have the
whole selftest always being marked as failed on old kernels. Patches
11 and 17/17 are checking the kernel version. An alternative would
be to ignore the results for some sub-tests but that's not ideal
too. Note that SELFTESTS_MPTCP_LIB_NO_KVERSION_CHECK env var can be
set to 1 not to skip these tests if the running kernel doesn't have
a supported version.
- After having launched the tests:
- Adapt the expectations depending on the presence of a kernel symbol
(patch 6/17) or a kernel version (patch 8/17).
- Check is a MIB counter is available and skip the verification if
not. Patch 4/17 is using this technique.
Before skipping tests, SELFTESTS_MPTCP_LIB_EXPECT_ALL_FEATURES env var
value is checked: if it is set to 1, the test is marked as "failed"
instead of "skipped". MPTCP public CI expects to have all features
supported and it sets this env var to 1 to catch regressions in these
new checks.
Patch 2/17 uses 'iptables-legacy' if available because it might be
needed when using an older kernel not supporting iptables-nft.
Patch 3/17 adds some helpers used in the other patches mentioned to
easily mark sub-tests as skipped.
Patch 16/17 uniforms MPTCP Join "listener" tests: it was imported code
from userspace_pm.sh but without using the "code style" and ways of
using tools and printing messages from MPTCP Join selftest.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:52 +0000 (18:11 +0200)]
selftests: mptcp: join: skip mixed tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of a mix of subflows in v4 and v6 by the
in-kernel PM introduced by commit b9d69db87fb7 ("mptcp: let the
in-kernel PM use mixed IPv4 and IPv6 addresses").
It looks like there is no external sign we can use to predict the
expected behaviour. Instead of accepting different behaviours and thus
not really checking for the expected behaviour, we are looking here for
a specific kernel version. That's not ideal but it looks better than
removing the test because it cannot support older kernel versions.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:51 +0000 (18:11 +0200)]
selftests: mptcp: join: uniform listener tests
The alignment was different from the other tests because tabs were used
instead of spaces.
While at it, also use 'echo' instead of 'printf' to print the result to
keep the same style as done in the other sub-tests. And, even if it
should be better with, also remove 'stdbuf' and sed's '--unbuffered'
option because they are not used in the other subtests and they are not
available when using a minimal environment with busybox.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:49 +0000 (18:11 +0200)]
selftests: mptcp: join: skip MPC backups tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of sending an MP_PRIO signal for the initial
subflow, introduced by commit c157bbe776b7 ("mptcp: allow the in kernel
PM to set MPC subflow priority").
It is possible to look for "mptcp_subflow_send_ack" in kallsyms because
it was needed to introduce the mentioned feature. So we can know in
advance if the feature is supported instead of trying and accepting any
results.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:48 +0000 (18:11 +0200)]
selftests: mptcp: join: skip fail tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of the MP_FAIL / infinite mapping introduced
by commit 1e39e5a32ad7 ("mptcp: infinite mapping sending") and the
following ones.
It is possible to look for one of the infinite mapping counters to know
in advance if the this feature is available.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:47 +0000 (18:11 +0200)]
selftests: mptcp: join: skip userspace PM tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of the userspace PM introduced by commit 4638de5aefe5 ("mptcp: handle local addrs announced by userspace PMs")
and the following ones.
It is possible to look for the MPTCP pm_type's sysctl knob to know in
advance if the userspace PM is available.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:46 +0000 (18:11 +0200)]
selftests: mptcp: join: skip fullmesh flag tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of the fullmesh flag for the in-kernel PM
introduced by commit 2843ff6f36db ("mptcp: remote addresses fullmesh")
and commit 1a0d6136c5f0 ("mptcp: local addresses fullmesh").
It looks like there is no easy external sign we can use to predict the
expected behaviour. We could add the flag and then check if it has been
added but for that, and for each fullmesh test, we would need to setup a
new environment, do the checks, clean it and then only start the test
from yet another clean environment. To keep it simple and avoid
introducing new issues, we look for a specific kernel version. That's
not ideal but an acceptable solution for this case.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:45 +0000 (18:11 +0200)]
selftests: mptcp: join: skip backup if set flag on ID not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
Commit bccefb762439 ("selftests: mptcp: simplify pm_nl_change_endpoint")
has simplified the way the backup flag is set on an endpoint. Instead of
doing:
./pm_nl_ctl set 10.0.2.1 flags backup
Now we do:
./pm_nl_ctl set id 1 flags backup
The new way is easier to maintain but it is also incompatible with older
kernels not supporting the implicit endpoints putting in place the
infrastructure to set flags per ID, hence the second Fixes tag.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:44 +0000 (18:11 +0200)]
selftests: mptcp: join: skip implicit tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of the implicit endpoints introduced by
commit d045b9eb95a9 ("mptcp: introduce implicit endpoints").
It is possible to look for "mptcp_subflow_send_ack" in kallsyms because
it was needed to introduce the mentioned feature. So we can know in
advance if the feature is supported instead of trying and accepting any
results.
Note that here and in the following commits, we re-do the same check for
each sub-test of the same function for a few reasons. The main one is
not to break the ID assign to each test in order to be able to easily
compare results between different kernel versions. Also, we can still
run a specific test even if it is skipped. Another reason is that it
makes it clear during the review that a specific subtest will be skipped
or not under certain conditions. At the end, it looks OK to call the
exact same helper multiple times: it is not a critical path and it is
the same code that is executed, not really more cases to maintain.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:43 +0000 (18:11 +0200)]
selftests: mptcp: join: support RM_ADDR for used endpoints or not
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
At some points, a new feature caused internal behaviour changes we are
verifying in the selftests, see the Fixes tag below. It was not a UAPI
change but because in these selftests, we check some internal
behaviours, it is normal we have to adapt them from time to time after
having added some features.
It looks like there is no external sign we can use to predict the
expected behaviour. Instead of accepting different behaviours and thus
not really checking for the expected behaviour, we are looking here for
a specific kernel version. That's not ideal but it looks better than
removing the test because it cannot support older kernel versions.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:42 +0000 (18:11 +0200)]
selftests: mptcp: join: skip Fastclose tests if not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the support of MP_FASTCLOSE introduced in commit f284c0c77321 ("mptcp: implement fastclose xmit path").
If the MIB counter is not available, the test cannot be verified and the
behaviour will not be the expected one. So we can skip the test if the
counter is missing.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:41 +0000 (18:11 +0200)]
selftests: mptcp: join: support local endpoint being tracked or not
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
At some points, a new feature caused internal behaviour changes we are
verifying in the selftests, see the Fixes tag below. It was not a uAPI
change but because in these selftests, we check some internal
behaviours, it is normal we have to adapt them from time to time after
having added some features.
It is possible to look for "mptcp_pm_subflow_check_next" in kallsyms
because it was needed to introduce the mentioned feature. So we can know
in advance what the behaviour we are expecting here instead of
supporting the two behaviours.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:40 +0000 (18:11 +0200)]
selftests: mptcp: join: skip test if iptables/tc cmds fail
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
Some tests are using IPTables and/or TC commands to force some
behaviours. If one of these commands fails -- likely because some
features are not available due to missing kernel config -- we should
intercept the error and skip the tests requiring these features.
Note that if we expect to have these features available and if
SELFTESTS_MPTCP_LIB_EXPECT_ALL_FEATURES env var is set to 1, the tests
will be marked as failed instead of skipped.
This patch also replaces the 'exit 1' by 'return 1' not to stop the
selftest in the middle without the conclusion if there is an issue with
NF or TC.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:39 +0000 (18:11 +0200)]
selftests: mptcp: join: skip check if MIB counter not supported
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
One of them is the MPTCP MIB counters introduced in commit fc518953bc9c
("mptcp: add and use MIB counter infrastructure") and more later. The
MPTCP Join selftest heavily relies on these counters.
If a counter is not supported by the kernel, it is not displayed when
using 'nstat -z'. We can then detect that and skip the verification. A
new helper (get_counter()) has been added to do the required checks and
return an error if the counter is not available.
Note that if we expect to have these features available and if
SELFTESTS_MPTCP_LIB_EXPECT_ALL_FEATURES env var is set to 1, the tests
will be marked as failed instead of skipped.
This new helper also makes sure we get the exact counter we want to
avoid issues we had in the past, e.g. with MPTcpExtRmAddr and
MPTcpExtRmAddrDrop sharing the same prefix. While at it, we uniform the
way we fetch a MIB counter.
Note for the backports: we rarely change these modified blocks so if
there is are conflicts, it is very likely because a counter is not used
in the older kernels and we don't need that chunk.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:38 +0000 (18:11 +0200)]
selftests: mptcp: join: helpers to skip tests
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
Here are some helpers that will be used to mark subtests as skipped if a
feature is not supported. Marking as a fix for the commit introducing
this selftest to help with the backports.
While at it, also check if kallsyms feature is available as it will also
be used in the following commits to check if MPTCP features are
available before starting a test.
Matthieu Baerts [Sat, 10 Jun 2023 16:11:36 +0000 (18:11 +0200)]
selftests: mptcp: lib: skip if not below kernel version
Selftests are supposed to run on any kernels, including the old ones not
supporting all MPTCP features.
A new function is now available to easily detect if a feature is
missing by looking at the kernel version. That's clearly not ideal and
this kind of check should be avoided as soon as possible. But sometimes,
there are no external sign that a "feature" is available or not:
internal behaviours can change without modifying the uAPI and these
selftests are verifying the internal behaviours. Sometimes, the only
(easy) way to verify if the feature is present is to run the test but
then the validation cannot determine if there is a failure with the
feature or if the feature is missing. Then it looks better to check the
kernel version instead of having tests that can never fail. In any case,
we need a solution not to have a whole selftest being marked as failed
just because one sub-test has failed.
Note that this env var car be set to 1 not to do such check and run the
linked sub-test: SELFTESTS_MPTCP_LIB_NO_KVERSION_CHECK.
This new helper is going to be used in the following commits. In order
to ease the backport of such future patches, it would be good if this
patch is backported up to the introduction of MPTCP selftests, hence the
Fixes tag below: this type of check was supposed to be done from the
beginning.
====================
fixes for Q-USGMII speeds and autoneg
This is the second version of a small changeset for QUSGMII support,
fixing inconsistencies in reported max speed and control word parsing.
As reported here [1], there are some inconsistencies for the Q-USGMII
mode speeds and configuration. The first patch in this fixup series
makes so that we correctly report the max speed of 1Gbps for this mode.
The second patch uses a dedicated helper to decode the control word.
This is necessary as although USGMII control words are close to USXGMII,
they don't support the same speeds.
net: phylink: use a dedicated helper to parse usgmii control word
Q-USGMII is a derivative of USGMII, that uses a specific formatting for
the control word. The layout is close to the USXGMII control word, but
doesn't support speeds over 1Gbps. Use a dedicated decoding logic for
the USGMII control word, re-using USXGMII definitions but only considering
10/100/1000Mbps speeds
Fixes: 5e61fe157a27 ("net: phy: Introduce QUSGMII PHY mode") Signed-off-by: Maxime Chevallier <[email protected]> Reviewed-by: Russell King (Oracle) <[email protected]> Signed-off-by: Jakub Kicinski <[email protected]>
net: phylink: report correct max speed for QUSGMII
Q-USGMII is the quad port version of USGMII, and supports a max speed of
1Gbps on each line. Make so that phylink_interface_max_speed() reports
this information correctly.
Fixes: ae0e4bb2a0e0 ("net: phylink: Adjust link settings based on rate matching") Signed-off-by: Maxime Chevallier <[email protected]> Reviewed-by: Russell King (Oracle) <[email protected]> Signed-off-by: Jakub Kicinski <[email protected]>
[CAUSE]
Since the introduction of global roots, we handle
csum/extent/free-space-tree roots as global roots, even if no
extent-tree-v2 feature is enabled.
So for regular csum/extent/fst roots, we load them into
fs_info::global_root_tree rb tree.
And we should not expect any conflicts in that rb tree, thus we have an
ASSERT() inside btrfs_global_root_insert().
But rescue=usebackuproot can break the assumption, as we will try to
load those trees again and again as long as we have bad roots and have
backup roots slot remaining.
So in that case we can have conflicting roots in the rb tree, and
triggering the ASSERT() crash.
[FIX]
We can safely remove that ASSERT(), as the caller will properly put the
offending root.
To make further debugging easier, also add two explicit error messages:
- Error message for conflicting global roots
- Error message when using backup roots slot
Linus Torvalds [Mon, 12 Jun 2023 23:14:34 +0000 (16:14 -0700)]
Merge tag 'mm-hotfixes-stable-2023-06-12-12-22' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
Pull misc fixes from Andrew Morton:
"19 hotfixes. 14 are cc:stable and the remainder address issues which
were introduced during this development cycle or which were considered
inappropriate for a backport"
* tag 'mm-hotfixes-stable-2023-06-12-12-22' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
zswap: do not shrink if cgroup may not zswap
page cache: fix page_cache_next/prev_miss off by one
ocfs2: check new file size on fallocate call
mailmap: add entry for John Keeping
mm/damon/core: fix divide error in damon_nr_accesses_to_accesses_bp()
epoll: ep_autoremove_wake_function should use list_del_init_careful
mm/gup_test: fix ioctl fail for compat task
nilfs2: reject devices with insufficient block count
ocfs2: fix use-after-free when unmounting read-only filesystem
lib/test_vmalloc.c: avoid garbage in page array
nilfs2: fix possible out-of-bounds segment allocation in resize ioctl
riscv/purgatory: remove PGO flags
powerpc/purgatory: remove PGO flags
x86/purgatory: remove PGO flags
kexec: support purgatories with .text.hot sections
mm/uffd: allow vma to merge as much as possible
mm/uffd: fix vma operation where start addr cuts part of vma
radix-tree: move declarations to header
nilfs2: fix incomplete buffer cleanup in nilfs_btnode_abort_change_key()
Chris Mason [Fri, 9 Jun 2023 17:53:41 +0000 (10:53 -0700)]
btrfs: can_nocow_file_extent should pass down args->strict from callers
Commit 619104ba453ad0 ("btrfs: move common NOCOW checks against a file
extent into a helper") changed our call to btrfs_cross_ref_exist() to
always pass false for the 'strict' parameter. We're passing this down
through the stack so that we can do a full check for cross references
during swapfile activation.
can_nocow_extent can reduce the len passed in, which needs to be
propagated to btrfs_dio_iomap_begin so that iomap does not submit
more data then is mapped.
This problems exists since the btrfs_get_blocks_direct helper was added
in commit c5794e51784a ("btrfs: Factor out write portion of
btrfs_get_blocks_direct"), but the ordered_extent splitting added in
commit b73a6fd1b1ef ("btrfs: split partial dio bios before submit")
added a WARN_ON that made a syzkaller test fail.
igc: Fix possible system crash when loading module
Guarantee that when probe() is run again, PTM and PCI busmaster will be
in the same state as it was if the driver was never loaded.
Avoid an i225/i226 hardware issue that PTM requests can be made even
though PCI bus mastering is not enabled. These unexpected PTM requests
can crash some systems.
So, "force" disable PTM and busmastering before removing the driver,
so they can be re-enabled in the right order during probe(). This is
more like a workaround and should be applicable for i225 and i226, in
any platform.
There could be a race condition during link down where interrupt
being generated and igc_clean_tx_irq() been called to perform the
TX completion. Properly clear the TX buffer/descriptor ring and
disable the TX Queue ring in igc_free_tx_resources() to avoid that.
Before the patch [1], the clock probe was done directly in the
clk-mt8365 driver. In this probe function, the array which stores the
data clocks is sized using the higher defined numbers (*_NR_CLOCK) in
the clock lists [2]. Currently, with the patch [1], the specific
clk-mt8365 probe function is replaced by the mtk generic one [3], which
size the clock data array by adding all the clock descriptor array size
provided by the clk-mt8365 driver.
Actually, all clock indexes come from the header file [2], that mean, if
there are more clock (then more index) in the header file [2] than the
number of clock declared in the clock descriptor arrays (which is the
case currently), the clock data array will be undersized and then the
generic probe function will overflow when it will try to write in
"clk_data[CLK_INDEX]". Actually, instead of crashing at boot, the probe
function returns an error in the log which looks like:
"of_clk_hw_onecell_get: invalid index 135", then this clock isn't
enabled.
Solve this issue by adding in the driver the missing clocks declared in
the header clock file [2].
Nhat Pham [Tue, 30 May 2023 22:24:40 +0000 (15:24 -0700)]
zswap: do not shrink if cgroup may not zswap
Before storing a page, zswap first checks if the number of stored pages
exceeds the limit specified by memory.zswap.max, for each cgroup in the
hierarchy. If this limit is reached or exceeded, then zswap shrinking is
triggered and short-circuits the store attempt.
However, since the zswap's LRU is not memcg-aware, this can create the
following pathological behavior: the cgroup whose zswap limit is 0 will
evict pages from other cgroups continually, without lowering its own zswap
usage. This means the shrinking will continue until the need for swap
ceases or the pool becomes empty.
As a result of this, we observe a disproportionate amount of zswap
writeback and a perpetually small zswap pool in our experiments, even
though the pool limit is never hit.
More generally, a cgroup might unnecessarily evict pages from other
cgroups before we drive the memcg back below its limit.
This patch fixes the issue by rejecting zswap store attempt without
shrinking the pool when obj_cgroup_may_zswap() returns false.
Mike Kravetz [Fri, 2 Jun 2023 22:57:47 +0000 (15:57 -0700)]
page cache: fix page_cache_next/prev_miss off by one
Ackerley Tng reported an issue with hugetlbfs fallocate here[1]. The
issue showed up after the conversion of hugetlb page cache lookup code to
use page_cache_next_miss. Code in hugetlb fallocate, userfaultfd and GUP
is now using page_cache_next_miss to determine if a page is present the
page cache. The following statement is used.
There are two issues with page_cache_next_miss when used in this way.
1) If the passed value for index is equal to the 'wrap-around' value,
the same index will always be returned. This wrap-around value is 0,
so 0 will be returned even if page is present at index 0.
2) If there is no gap in the range passed, the last index in the range
will be returned. When passed a range of 1 as above, the passed
index value will be returned even if the page is present.
The end result is the statement above will NEVER indicate a page is
present in the cache, even if it is.
As noted by Ackerley in [1], users can see this by hugetlb fallocate
incorrectly returning EEXIST if pages are already present in the file. In
addition, hugetlb pages will not be included in core dumps if they need to
be brought in via GUP. userfaultfd UFFDIO_COPY also uses this code and
will not notice pages already present in the cache. It may try to
allocate a new page and potentially return ENOMEM as opposed to EEXIST.
Both page_cache_next_miss and page_cache_prev_miss have similar issues.
Fix by:
- Check for index equal to 'wrap-around' value and do not exit early.
- If no gap is found in range, return index outside range.
- Update function description to say 'wrap-around' value could be
returned if passed as index.
Luís Henriques [Mon, 29 May 2023 15:26:45 +0000 (16:26 +0100)]
ocfs2: check new file size on fallocate call
When changing a file size with fallocate() the new size isn't being
checked. In particular, the FSIZE ulimit isn't being checked, which makes
fstest generic/228 fail. Simply adding a call to inode_newsize_ok() fixes
this issue.
Kefeng Wang [Sat, 27 May 2023 03:21:01 +0000 (11:21 +0800)]
mm/damon/core: fix divide error in damon_nr_accesses_to_accesses_bp()
If 'aggr_interval' is smaller than 'sample_interval', max_nr_accesses in
damon_nr_accesses_to_accesses_bp() becomes zero which leads to divide
error, let's validate the values of them in damon_set_attrs() to fix it,
which similar to others attrs check.
Benjamin Segall [Tue, 30 May 2023 18:32:28 +0000 (11:32 -0700)]
epoll: ep_autoremove_wake_function should use list_del_init_careful
autoremove_wake_function uses list_del_init_careful, so should epoll's
more aggressive variant. It only doesn't because it was copied from an
older wait.c rather than the most recent.
Ryusuke Konishi [Fri, 26 May 2023 02:13:32 +0000 (11:13 +0900)]
nilfs2: reject devices with insufficient block count
The current sanity check for nilfs2 geometry information lacks checks for
the number of segments stored in superblocks, so even for device images
that have been destructively truncated or have an unusually high number of
segments, the mount operation may succeed.
This causes out-of-bounds block I/O on file system block reads or log
writes to the segments, the latter in particular causing
"a_ops->writepages" to repeatedly fail, resulting in sync_inodes_sb() to
hang.
Fix this issue by checking the number of segments stored in the superblock
and avoiding mounting devices that can cause out-of-bounds accesses. To
eliminate the possibility of overflow when calculating the number of
blocks required for the device from the number of segments, this also adds
a helper function to calculate the upper bound on the number of segments
and inserts a check using it.
Luís Henriques [Mon, 22 May 2023 10:21:12 +0000 (11:21 +0100)]
ocfs2: fix use-after-free when unmounting read-only filesystem
It's trivial to trigger a use-after-free bug in the ocfs2 quotas code using
fstest generic/452. After a read-only remount, quotas are suspended and
ocfs2_mem_dqinfo is freed through ->ocfs2_local_free_info(). When unmounting
the filesystem, an UAF access to the oinfo will eventually cause a crash.
BUG: KASAN: slab-use-after-free in timer_delete+0x54/0xc0
Read of size 8 at addr ffff8880389a8208 by task umount/669
...
Call Trace:
<TASK>
...
timer_delete+0x54/0xc0
try_to_grab_pending+0x31/0x230
__cancel_work_timer+0x6c/0x270
ocfs2_disable_quotas.isra.0+0x3e/0xf0 [ocfs2]
ocfs2_dismount_volume+0xdd/0x450 [ocfs2]
generic_shutdown_super+0xaa/0x280
kill_block_super+0x46/0x70
deactivate_locked_super+0x4d/0xb0
cleanup_mnt+0x135/0x1f0
...
</TASK>
Lorenzo Stoakes [Wed, 24 May 2023 08:24:24 +0000 (09:24 +0100)]
lib/test_vmalloc.c: avoid garbage in page array
It turns out that alloc_pages_bulk_array() does not treat the page_array
parameter as an output parameter, but rather reads the array and skips any
entries that have already been allocated.
This is somewhat unexpected and breaks this test, as we allocate the pages
array uninitialised on the assumption it will be overwritten.
As a result, the test was referencing uninitialised data and causing the
PFN to not be valid and thus a WARN_ON() followed by a null pointer deref
and panic.
In addition, this is an array of pointers not of struct page objects, so we
need only allocate an array with elements of pointer size.
We solve both problems by simply using kcalloc() and referencing
sizeof(struct page *) rather than sizeof(struct page).
Ryusuke Konishi [Wed, 24 May 2023 09:43:48 +0000 (18:43 +0900)]
nilfs2: fix possible out-of-bounds segment allocation in resize ioctl
Syzbot reports that in its stress test for resize ioctl, the log writing
function nilfs_segctor_do_construct hits a WARN_ON in
nilfs_segctor_truncate_segments().
It turned out that there is a problem with the current implementation of
the resize ioctl, which changes the writable range on the device (the
range of allocatable segments) at the end of the resize process.
This order is necessary for file system expansion to avoid corrupting the
superblock at trailing edge. However, in the case of a file system
shrink, if log writes occur after truncating out-of-bounds trailing
segments and before the resize is complete, segments may be allocated from
the truncated space.
The userspace resize tool was fine as it limits the range of allocatable
segments before performing the resize, but it can run into this issue if
the resize ioctl is called alone.
Fix this issue by changing nilfs_sufile_resize() to update the range of
allocatable segments immediately after successful truncation of segment
space in case of file system shrink.
And both of them have their range [sh_addr ... sh_addr+sh_size] on the
area pointed by `e_entry`.
This causes that image->start is calculated twice, once for .text and
another time for .text.hot. The second calculation leaves image->start
in a random location.
Because of this, the system crashes immediately after:
Peter Xu [Wed, 17 May 2023 19:09:16 +0000 (15:09 -0400)]
mm/uffd: allow vma to merge as much as possible
We used to not pass in the pgoff correctly when register/unregister uffd
regions, it caused incorrect behavior on vma merging and can cause
mergeable vmas being separate after ioctls return.
For example, when we have:
vma1(range 0-9, with uffd), vma2(range 10-19, no uffd)
Then someone unregisters uffd on range (5-9), it should logically become:
vma1(range 0-4, with uffd), vma2(range 5-19, no uffd)
But with current code we'll have:
vma1(range 0-4, with uffd), vma3(range 5-9, no uffd), vma2(range 10-19, no uffd)
This patch allows such merge to happen correctly before ioctl returns.
This behavior seems to have existed since the 1st day of uffd. Since
pgoff for vma_merge() is only used to identify the possibility of vma
merging, meanwhile here what we did was always passing in a pgoff smaller
than what we should, so there should have no other side effect besides not
merging it. Let's still tentatively copy stable for this, even though I
don't see anything will go wrong besides vma being split (which is mostly
not user visible).
Peter Xu [Wed, 17 May 2023 19:09:15 +0000 (15:09 -0400)]
mm/uffd: fix vma operation where start addr cuts part of vma
Patch series "mm/uffd: Fix vma merge/split", v2.
This series contains two patches that fix vma merge/split for userfaultfd
on two separate issues.
Patch 1 fixes a regression since 6.1+ due to something we overlooked when
converting to maple tree apis. The plan is we use patch 1 to replace the
commit "2f628010799e (mm: userfaultfd: avoid passing an invalid range to
vma_merge())" in mm-hostfixes-unstable tree if possible, so as to bring
uffd vma operations back aligned with the rest code again.
Patch 2 fixes a long standing issue that vma can be left unmerged even if
we can for either uffd register or unregister.
Many thanks to Lorenzo on either noticing this issue from the assert
movement patch, looking at this problem, and also provided a reproducer on
the unmerged vma issue [1].
It seems vma merging with uffd paths is broken with either
register/unregister, where right now we can feed wrong parameters to
vma_merge() and it's found by recent patch which moved asserts upwards in
vma_merge() by Lorenzo Stoakes:
It's possible that "start" is contained within vma but not clamped to its
start. We need to convert this into either "cannot merge" case or "can
merge" case 4 which permits subdivision of prev by assigning vma to prev.
As we loop, each subsequent VMA will be clamped to the start.
This patch will eliminate the report and make sure vma_merge() calls will
become legal again.
One thing to mention is that the "Fixes: 29417d292bd0" below is there only
to help explain where the warning can start to trigger, the real commit to
fix should be 69dbe6daf104. Commit 29417d292bd0 helps us to identify the
issue, but unfortunately we may want to keep it in Fixes too just to ease
kernel backporters for easier tracking.
Arnd Bergmann [Tue, 16 May 2023 19:41:54 +0000 (21:41 +0200)]
radix-tree: move declarations to header
The xarray.c file contains the only call to radix_tree_node_rcu_free(),
and it comes with its own extern declaration for it. This means the
function definition causes a missing-prototype warning:
lib/radix-tree.c:288:6: error: no previous prototype for 'radix_tree_node_rcu_free' [-Werror=missing-prototypes]
Instead, move the declaration for this function to a new header that can
be included by both, and do the same for the radix_tree_node_cachep
variable that has the same underlying problem but does not cause a warning
with gcc.
projects / linux.git / log