linux-stable.git/include, branch v6.7.7 Linux kernel stable tree iommu: Add mm_get_enqcmd_pasid() helper function 2024-03-01T12:41:59+00:00 Tina Zhang tina.zhang@intel.com 2023-10-27T00:05:22+00:00 7c122e02b32f92f9b0ef06fa90d5f1172afeea96 [ Upstream commit 2396046d75d3c0b2cfead852a77efd023f8539dc ] mm_get_enqcmd_pasid() should be used by architecture code and closely related to learn the PASID value that the x86 ENQCMD operation should use for the mm. For the moment SMMUv3 uses this without any connection to ENQCMD, it will be cleaned up similar to how the prior patch made VT-d use the PASID argument of set_dev_pasid(). The motivation is to replace mm->pasid with an iommu private data structure that is introduced in a later patch. Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com> Reviewed-by: Jason Gunthorpe <jgg@nvidia.com> Tested-by: Nicolin Chen <nicolinc@nvidia.com> Signed-off-by: Tina Zhang <tina.zhang@intel.com> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com> Link: https://lore.kernel.org/r/20231027000525.1278806-4-tina.zhang@intel.com Signed-off-by: Joerg Roedel <jroedel@suse.de> Stable-dep-of: b5bf7778b722 ("iommu/arm-smmu-v3: Do not use GFP_KERNEL under as spinlock") Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 2396046d75d3c0b2cfead852a77efd023f8539dc ]

mm_get_enqcmd_pasid() should be used by architecture code and closely
related to learn the PASID value that the x86 ENQCMD operation should
use for the mm.

For the moment SMMUv3 uses this without any connection to ENQCMD, it
will be cleaned up similar to how the prior patch made VT-d use the
PASID argument of set_dev_pasid().

The motivation is to replace mm->pasid with an iommu private data
structure that is introduced in a later patch.

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Tested-by: Nicolin Chen <nicolinc@nvidia.com>
Signed-off-by: Tina Zhang <tina.zhang@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/20231027000525.1278806-4-tina.zhang@intel.com
Signed-off-by: Joerg Roedel <jroedel@suse.de>
Stable-dep-of: b5bf7778b722 ("iommu/arm-smmu-v3: Do not use GFP_KERNEL under as spinlock")
Signed-off-by: Sasha Levin <sashal@kernel.org>
netfilter: nft_flow_offload: reset dst in route object after setting up flow 2024-03-01T12:41:58+00:00 Pablo Neira Ayuso pablo@netfilter.org 2024-02-21T11:32:58+00:00 670548c8db44d76e40e1dfc06812bca36a61e9ae [ Upstream commit 9e0f0430389be7696396c62f037be4bf72cf93e3 ] dst is transferred to the flow object, route object does not own it anymore. Reset dst in route object, otherwise if flow_offload_add() fails, error path releases dst twice, leading to a refcount underflow. Fixes: a3c90f7a2323 ("netfilter: nf_tables: flow offload expression") Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 9e0f0430389be7696396c62f037be4bf72cf93e3 ]

dst is transferred to the flow object, route object does not own it
anymore.  Reset dst in route object, otherwise if flow_offload_add()
fails, error path releases dst twice, leading to a refcount underflow.

Fixes: a3c90f7a2323 ("netfilter: nf_tables: flow offload expression")
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
x86/numa: Fix the address overlap check in numa_fill_memblks() 2024-03-01T12:41:54+00:00 Alison Schofield alison.schofield@intel.com 2024-01-12T20:09:50+00:00 6b725eab0967839ec393d8caa827cec2886c2aa5 [ Upstream commit 9b99c17f7510bed2adbe17751fb8abddba5620bc ] numa_fill_memblks() fills in the gaps in numa_meminfo memblks over a physical address range. To do so, it first creates a list of existing memblks that overlap that address range. The issue is that it is off by one when comparing to the end of the address range, so memblks that do not overlap are selected. The impact of selecting a memblk that does not actually overlap is that an existing memblk may be filled when the expected action is to do nothing and return NUMA_NO_MEMBLK to the caller. The caller can then add a new NUMA node and memblk. Replace the broken open-coded search for address overlap with the memblock helper memblock_addrs_overlap(). Update the kernel doc and in code comments. Suggested by: "Huang, Ying" <ying.huang@intel.com> Fixes: 8f012db27c95 ("x86/numa: Introduce numa_fill_memblks()") Signed-off-by: Alison Schofield <alison.schofield@intel.com> Acked-by: Mike Rapoport (IBM) <rppt@kernel.org> Acked-by: Dave Hansen <dave.hansen@linux.intel.com> Reviewed-by: Dan Williams <dan.j.williams@intel.com> Link: https://lore.kernel.org/r/10a3e6109c34c21a8dd4c513cf63df63481a2b07.1705085543.git.alison.schofield@intel.com Signed-off-by: Dan Williams <dan.j.williams@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 9b99c17f7510bed2adbe17751fb8abddba5620bc ]

numa_fill_memblks() fills in the gaps in numa_meminfo memblks over a
physical address range. To do so, it first creates a list of existing
memblks that overlap that address range. The issue is that it is off
by one when comparing to the end of the address range, so memblks
that do not overlap are selected.

The impact of selecting a memblk that does not actually overlap is
that an existing memblk may be filled when the expected action is to
do nothing and return NUMA_NO_MEMBLK to the caller. The caller can
then add a new NUMA node and memblk.

Replace the broken open-coded search for address overlap with the
memblock helper memblock_addrs_overlap(). Update the kernel doc
and in code comments.

Suggested by: "Huang, Ying" <ying.huang@intel.com>

Fixes: 8f012db27c95 ("x86/numa: Introduce numa_fill_memblks()")
Signed-off-by: Alison Schofield <alison.schofield@intel.com>
Acked-by: Mike Rapoport (IBM) <rppt@kernel.org>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Link: https://lore.kernel.org/r/10a3e6109c34c21a8dd4c513cf63df63481a2b07.1705085543.git.alison.schofield@intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net: bridge: switchdev: Skip MDB replays of deferred events on offload 2024-03-01T12:41:54+00:00 Tobias Waldekranz tobias@waldekranz.com 2024-02-14T21:40:03+00:00 e0b4c5b1d760008f1dd18c07c35af0442e54f9c8 [ Upstream commit dc489f86257cab5056e747344f17a164f63bff4b ] Before this change, generation of the list of MDB events to replay would race against the creation of new group memberships, either from the IGMP/MLD snooping logic or from user configuration. While new memberships are immediately visible to walkers of br->mdb_list, the notification of their existence to switchdev event subscribers is deferred until a later point in time. So if a replay list was generated during a time that overlapped with such a window, it would also contain a replay of the not-yet-delivered event. The driver would thus receive two copies of what the bridge internally considered to be one single event. On destruction of the bridge, only a single membership deletion event was therefore sent. As a consequence of this, drivers which reference count memberships (at least DSA), would be left with orphan groups in their hardware database when the bridge was destroyed. This is only an issue when replaying additions. While deletion events may still be pending on the deferred queue, they will already have been removed from br->mdb_list, so no duplicates can be generated in that scenario. To a user this meant that old group memberships, from a bridge in which a port was previously attached, could be reanimated (in hardware) when the port joined a new bridge, without the new bridge's knowledge. For example, on an mv88e6xxx system, create a snooping bridge and immediately add a port to it: root@infix-06-0b-00:~$ ip link add dev br0 up type bridge mcast_snooping 1 && \ > ip link set dev x3 up master br0 And then destroy the bridge: root@infix-06-0b-00:~$ ip link del dev br0 root@infix-06-0b-00:~$ mvls atu ADDRESS FID STATE Q F 0 1 2 3 4 5 6 7 8 9 a DEV:0 Marvell 88E6393X 33:33:00:00:00:6a 1 static - - 0 . . . . . . . . . . 33:33:ff:87:e4:3f 1 static - - 0 . . . . . . . . . . ff:ff:ff:ff:ff:ff 1 static - - 0 1 2 3 4 5 6 7 8 9 a root@infix-06-0b-00:~$ The two IPv6 groups remain in the hardware database because the port (x3) is notified of the host's membership twice: once via the original event and once via a replay. Since only a single delete notification is sent, the count remains at 1 when the bridge is destroyed. Then add the same port (or another port belonging to the same hardware domain) to a new bridge, this time with snooping disabled: root@infix-06-0b-00:~$ ip link add dev br1 up type bridge mcast_snooping 0 && \ > ip link set dev x3 up master br1 All multicast, including the two IPv6 groups from br0, should now be flooded, according to the policy of br1. But instead the old memberships are still active in the hardware database, causing the switch to only forward traffic to those groups towards the CPU (port 0). Eliminate the race in two steps: 1. Grab the write-side lock of the MDB while generating the replay list. This prevents new memberships from showing up while we are generating the replay list. But it leaves the scenario in which a deferred event was already generated, but not delivered, before we grabbed the lock. Therefore: 2. Make sure that no deferred version of a replay event is already enqueued to the switchdev deferred queue, before adding it to the replay list, when replaying additions. Fixes: 4f2673b3a2b6 ("net: bridge: add helper to replay port and host-joined mdb entries") Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com> Reviewed-by: Vladimir Oltean <olteanv@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit dc489f86257cab5056e747344f17a164f63bff4b ]

Before this change, generation of the list of MDB events to replay
would race against the creation of new group memberships, either from
the IGMP/MLD snooping logic or from user configuration.

While new memberships are immediately visible to walkers of
br->mdb_list, the notification of their existence to switchdev event
subscribers is deferred until a later point in time. So if a replay
list was generated during a time that overlapped with such a window,
it would also contain a replay of the not-yet-delivered event.

The driver would thus receive two copies of what the bridge internally
considered to be one single event. On destruction of the bridge, only
a single membership deletion event was therefore sent. As a
consequence of this, drivers which reference count memberships (at
least DSA), would be left with orphan groups in their hardware
database when the bridge was destroyed.

This is only an issue when replaying additions. While deletion events
may still be pending on the deferred queue, they will already have
been removed from br->mdb_list, so no duplicates can be generated in
that scenario.

To a user this meant that old group memberships, from a bridge in
which a port was previously attached, could be reanimated (in
hardware) when the port joined a new bridge, without the new bridge's
knowledge.

For example, on an mv88e6xxx system, create a snooping bridge and
immediately add a port to it:

    root@infix-06-0b-00:~$ ip link add dev br0 up type bridge mcast_snooping 1 && \
    > ip link set dev x3 up master br0

And then destroy the bridge:

    root@infix-06-0b-00:~$ ip link del dev br0
    root@infix-06-0b-00:~$ mvls atu
    ADDRESS             FID  STATE      Q  F  0  1  2  3  4  5  6  7  8  9  a
    DEV:0 Marvell 88E6393X
    33:33:00:00:00:6a     1  static     -  -  0  .  .  .  .  .  .  .  .  .  .
    33:33:ff:87:e4:3f     1  static     -  -  0  .  .  .  .  .  .  .  .  .  .
    ff:ff:ff:ff:ff:ff     1  static     -  -  0  1  2  3  4  5  6  7  8  9  a
    root@infix-06-0b-00:~$

The two IPv6 groups remain in the hardware database because the
port (x3) is notified of the host's membership twice: once via the
original event and once via a replay. Since only a single delete
notification is sent, the count remains at 1 when the bridge is
destroyed.

Then add the same port (or another port belonging to the same hardware
domain) to a new bridge, this time with snooping disabled:

    root@infix-06-0b-00:~$ ip link add dev br1 up type bridge mcast_snooping 0 && \
    > ip link set dev x3 up master br1

All multicast, including the two IPv6 groups from br0, should now be
flooded, according to the policy of br1. But instead the old
memberships are still active in the hardware database, causing the
switch to only forward traffic to those groups towards the CPU (port
0).

Eliminate the race in two steps:

1. Grab the write-side lock of the MDB while generating the replay
   list.

This prevents new memberships from showing up while we are generating
the replay list. But it leaves the scenario in which a deferred event
was already generated, but not delivered, before we grabbed the
lock. Therefore:

2. Make sure that no deferred version of a replay event is already
   enqueued to the switchdev deferred queue, before adding it to the
   replay list, when replaying additions.

Fixes: 4f2673b3a2b6 ("net: bridge: add helper to replay port and host-joined mdb entries")
Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
Reviewed-by: Vladimir Oltean <olteanv@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
IB/mlx5: Don't expose debugfs entries for RRoCE general parameters if not supported 2024-03-01T12:41:52+00:00 Mark Zhang markzhang@nvidia.com 2024-01-28T09:29:12+00:00 35dae0bd7ea51d13861fddeefc1a8abf1e34c0cd [ Upstream commit 43fdbd140238d44e7e847232719fef7d20f9d326 ] debugfs entries for RRoCE general CC parameters must be exposed only when they are supported, otherwise when accessing them there may be a syndrome error in kernel log, for example: $ cat /sys/kernel/debug/mlx5/0000:08:00.1/cc_params/rtt_resp_dscp cat: '/sys/kernel/debug/mlx5/0000:08:00.1/cc_params/rtt_resp_dscp': Invalid argument $ dmesg mlx5_core 0000:08:00.1: mlx5_cmd_out_err:805:(pid 1253): QUERY_CONG_PARAMS(0x824) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0x325a82), err(-22) Fixes: 66fb1d5df6ac ("IB/mlx5: Extend debug control for CC parameters") Reviewed-by: Edward Srouji <edwards@nvidia.com> Signed-off-by: Mark Zhang <markzhang@nvidia.com> Link: https://lore.kernel.org/r/e7ade70bad52b7468bdb1de4d41d5fad70c8b71c.1706433934.git.leon@kernel.org Signed-off-by: Leon Romanovsky <leon@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org> 
[ Upstream commit 43fdbd140238d44e7e847232719fef7d20f9d326 ]

debugfs entries for RRoCE general CC parameters must be exposed only when
they are supported, otherwise when accessing them there may be a syndrome
error in kernel log, for example:

$ cat /sys/kernel/debug/mlx5/0000:08:00.1/cc_params/rtt_resp_dscp
cat: '/sys/kernel/debug/mlx5/0000:08:00.1/cc_params/rtt_resp_dscp': Invalid argument
$ dmesg
 mlx5_core 0000:08:00.1: mlx5_cmd_out_err:805:(pid 1253): QUERY_CONG_PARAMS(0x824) op_mod(0x0) failed, status bad parameter(0x3), syndrome (0x325a82), err(-22)

Fixes: 66fb1d5df6ac ("IB/mlx5: Extend debug control for CC parameters")
Reviewed-by: Edward Srouji <edwards@nvidia.com>
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Link: https://lore.kernel.org/r/e7ade70bad52b7468bdb1de4d41d5fad70c8b71c.1706433934.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
mptcp: fix lockless access in subflow ULP diag 2024-03-01T12:41:50+00:00 Paolo Abeni pabeni@redhat.com 2024-02-15T18:25:30+00:00 298ac00da8e6bc2dcd690b45108acbd944c347d3 commit b8adb69a7d29c2d33eb327bca66476fb6066516b upstream. Since the introduction of the subflow ULP diag interface, the dump callback accessed all the subflow data with lockless. We need either to annotate all the read and write operation accordingly, or acquire the subflow socket lock. Let's do latter, even if slower, to avoid a diffstat havoc. Fixes: 5147dfb50832 ("mptcp: allow dumping subflow context to userspace") Cc: stable@vger.kernel.org Signed-off-by: Paolo Abeni <pabeni@redhat.com> Reviewed-by: Mat Martineau <martineau@kernel.org> Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit b8adb69a7d29c2d33eb327bca66476fb6066516b upstream.

Since the introduction of the subflow ULP diag interface, the
dump callback accessed all the subflow data with lockless.

We need either to annotate all the read and write operation accordingly,
or acquire the subflow socket lock. Let's do latter, even if slower, to
avoid a diffstat havoc.

Fixes: 5147dfb50832 ("mptcp: allow dumping subflow context to userspace")
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/swap: fix race when skipping swapcache 2024-03-01T12:41:47+00:00 Kairui Song kasong@tencent.com 2024-02-06T18:25:59+00:00 d183a4631acfc7af955c02a02e739cec15f5234d commit 13ddaf26be324a7f951891ecd9ccd04466d27458 upstream. When skipping swapcache for SWP_SYNCHRONOUS_IO, if two or more threads swapin the same entry at the same time, they get different pages (A, B). Before one thread (T0) finishes the swapin and installs page (A) to the PTE, another thread (T1) could finish swapin of page (B), swap_free the entry, then swap out the possibly modified page reusing the same entry. It breaks the pte_same check in (T0) because PTE value is unchanged, causing ABA problem. Thread (T0) will install a stalled page (A) into the PTE and cause data corruption. One possible callstack is like this: CPU0 CPU1 ---- ---- do_swap_page() do_swap_page() with same entry <direct swapin path> <direct swapin path> <alloc page A> <alloc page B> swap_read_folio() <- read to page A swap_read_folio() <- read to page B <slow on later locks or interrupt> <finished swapin first> ... set_pte_at() swap_free() <- entry is free <write to page B, now page A stalled> <swap out page B to same swap entry> pte_same() <- Check pass, PTE seems unchanged, but page A is stalled! swap_free() <- page B content lost! set_pte_at() <- staled page A installed! And besides, for ZRAM, swap_free() allows the swap device to discard the entry content, so even if page (B) is not modified, if swap_read_folio() on CPU0 happens later than swap_free() on CPU1, it may also cause data loss. To fix this, reuse swapcache_prepare which will pin the swap entry using the cache flag, and allow only one thread to swap it in, also prevent any parallel code from putting the entry in the cache. Release the pin after PT unlocked. Racers just loop and wait since it's a rare and very short event. A schedule_timeout_uninterruptible(1) call is added to avoid repeated page faults wasting too much CPU, causing livelock or adding too much noise to perf statistics. A similar livelock issue was described in commit 029c4628b2eb ("mm: swap: get rid of livelock in swapin readahead") Reproducer: This race issue can be triggered easily using a well constructed reproducer and patched brd (with a delay in read path) [1]: With latest 6.8 mainline, race caused data loss can be observed easily: $ gcc -g -lpthread test-thread-swap-race.c && ./a.out Polulating 32MB of memory region... Keep swapping out... Starting round 0... Spawning 65536 workers... 32746 workers spawned, wait for done... Round 0: Error on 0x5aa00, expected 32746, got 32743, 3 data loss! Round 0: Error on 0x395200, expected 32746, got 32743, 3 data loss! Round 0: Error on 0x3fd000, expected 32746, got 32737, 9 data loss! Round 0 Failed, 15 data loss! This reproducer spawns multiple threads sharing the same memory region using a small swap device. Every two threads updates mapped pages one by one in opposite direction trying to create a race, with one dedicated thread keep swapping out the data out using madvise. The reproducer created a reproduce rate of about once every 5 minutes, so the race should be totally possible in production. After this patch, I ran the reproducer for over a few hundred rounds and no data loss observed. Performance overhead is minimal, microbenchmark swapin 10G from 32G zram: Before: 10934698 us After: 11157121 us Cached: 13155355 us (Dropping SWP_SYNCHRONOUS_IO flag) [kasong@tencent.com: v4] Link: https://lkml.kernel.org/r/20240219082040.7495-1-ryncsn@gmail.com Link: https://lkml.kernel.org/r/20240206182559.32264-1-ryncsn@gmail.com Fixes: 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device") Reported-by: "Huang, Ying" <ying.huang@intel.com> Closes: https://lore.kernel.org/lkml/87bk92gqpx.fsf_-_@yhuang6-desk2.ccr.corp.intel.com/ Link: https://github.com/ryncsn/emm-test-project/tree/master/swap-stress-race [1] Signed-off-by: Kairui Song <kasong@tencent.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Acked-by: Yu Zhao <yuzhao@google.com> Acked-by: David Hildenbrand <david@redhat.com> Acked-by: Chris Li <chrisl@kernel.org> Cc: Hugh Dickins <hughd@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Matthew Wilcox (Oracle) <willy@infradead.org> Cc: Michal Hocko <mhocko@suse.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yosry Ahmed <yosryahmed@google.com> Cc: Yu Zhao <yuzhao@google.com> Cc: Barry Song <21cnbao@gmail.com> Cc: SeongJae Park <sj@kernel.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 13ddaf26be324a7f951891ecd9ccd04466d27458 upstream.

When skipping swapcache for SWP_SYNCHRONOUS_IO, if two or more threads
swapin the same entry at the same time, they get different pages (A, B).
Before one thread (T0) finishes the swapin and installs page (A) to the
PTE, another thread (T1) could finish swapin of page (B), swap_free the
entry, then swap out the possibly modified page reusing the same entry.
It breaks the pte_same check in (T0) because PTE value is unchanged,
causing ABA problem.  Thread (T0) will install a stalled page (A) into the
PTE and cause data corruption.

One possible callstack is like this:

CPU0                                 CPU1
----                                 ----
do_swap_page()                       do_swap_page() with same entry
<direct swapin path>                 <direct swapin path>
<alloc page A>                       <alloc page B>
swap_read_folio() <- read to page A  swap_read_folio() <- read to page B
<slow on later locks or interrupt>   <finished swapin first>
...                                  set_pte_at()
                                     swap_free() <- entry is free
                                     <write to page B, now page A stalled>
                                     <swap out page B to same swap entry>
pte_same() <- Check pass, PTE seems
              unchanged, but page A
              is stalled!
swap_free() <- page B content lost!
set_pte_at() <- staled page A installed!

And besides, for ZRAM, swap_free() allows the swap device to discard the
entry content, so even if page (B) is not modified, if swap_read_folio()
on CPU0 happens later than swap_free() on CPU1, it may also cause data
loss.

To fix this, reuse swapcache_prepare which will pin the swap entry using
the cache flag, and allow only one thread to swap it in, also prevent any
parallel code from putting the entry in the cache.  Release the pin after
PT unlocked.

Racers just loop and wait since it's a rare and very short event.  A
schedule_timeout_uninterruptible(1) call is added to avoid repeated page
faults wasting too much CPU, causing livelock or adding too much noise to
perf statistics.  A similar livelock issue was described in commit
029c4628b2eb ("mm: swap: get rid of livelock in swapin readahead")

Reproducer:

This race issue can be triggered easily using a well constructed
reproducer and patched brd (with a delay in read path) [1]:

With latest 6.8 mainline, race caused data loss can be observed easily:
$ gcc -g -lpthread test-thread-swap-race.c && ./a.out
  Polulating 32MB of memory region...
  Keep swapping out...
  Starting round 0...
  Spawning 65536 workers...
  32746 workers spawned, wait for done...
  Round 0: Error on 0x5aa00, expected 32746, got 32743, 3 data loss!
  Round 0: Error on 0x395200, expected 32746, got 32743, 3 data loss!
  Round 0: Error on 0x3fd000, expected 32746, got 32737, 9 data loss!
  Round 0 Failed, 15 data loss!

This reproducer spawns multiple threads sharing the same memory region
using a small swap device.  Every two threads updates mapped pages one by
one in opposite direction trying to create a race, with one dedicated
thread keep swapping out the data out using madvise.

The reproducer created a reproduce rate of about once every 5 minutes, so
the race should be totally possible in production.

After this patch, I ran the reproducer for over a few hundred rounds and
no data loss observed.

Performance overhead is minimal, microbenchmark swapin 10G from 32G
zram:

Before:     10934698 us
After:      11157121 us
Cached:     13155355 us (Dropping SWP_SYNCHRONOUS_IO flag)

[kasong@tencent.com: v4]
  Link: https://lkml.kernel.org/r/20240219082040.7495-1-ryncsn@gmail.com
Link: https://lkml.kernel.org/r/20240206182559.32264-1-ryncsn@gmail.com
Fixes: 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device")
Reported-by: "Huang, Ying" <ying.huang@intel.com>
Closes: https://lore.kernel.org/lkml/87bk92gqpx.fsf_-_@yhuang6-desk2.ccr.corp.intel.com/
Link: https://github.com/ryncsn/emm-test-project/tree/master/swap-stress-race [1]
Signed-off-by: Kairui Song <kasong@tencent.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Acked-by: Yu Zhao <yuzhao@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Chris Li <chrisl@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Yosry Ahmed <yosryahmed@google.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Barry Song <21cnbao@gmail.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
scsi: core: Consult supported VPD page list prior to fetching page 2024-03-01T12:41:46+00:00 Martin K. Petersen martin.petersen@oracle.com 2024-02-14T22:14:11+00:00 fa658e37b13419e768e7c6befc32fac48d059522 commit b5fc07a5fb56216a49e6c1d0b172d5464d99a89b upstream. Commit c92a6b5d6335 ("scsi: core: Query VPD size before getting full page") removed the logic which checks whether a VPD page is present on the supported pages list before asking for the page itself. That was done because SPC helpfully states "The Supported VPD Pages VPD page list may or may not include all the VPD pages that are able to be returned by the device server". Testing had revealed a few devices that supported some of the 0xBn pages but didn't actually list them in page 0. Julian Sikorski bisected a problem with his drive resetting during discovery to the commit above. As it turns out, this particular drive firmware will crash if we attempt to fetch page 0xB9. Various approaches were attempted to work around this. In the end, reinstating the logic that consults VPD page 0 before fetching any other page was the path of least resistance. A firmware update for the devices which originally compelled us to remove the check has since been released. Link: https://lore.kernel.org/r/20240214221411.2888112-1-martin.petersen@oracle.com Fixes: c92a6b5d6335 ("scsi: core: Query VPD size before getting full page") Cc: stable@vger.kernel.org Cc: Bart Van Assche <bvanassche@acm.org> Reported-by: Julian Sikorski <belegdol@gmail.com> Tested-by: Julian Sikorski <belegdol@gmail.com> Reviewed-by: Lee Duncan <lee.duncan@suse.com> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit b5fc07a5fb56216a49e6c1d0b172d5464d99a89b upstream.

Commit c92a6b5d6335 ("scsi: core: Query VPD size before getting full
page") removed the logic which checks whether a VPD page is present on
the supported pages list before asking for the page itself. That was
done because SPC helpfully states "The Supported VPD Pages VPD page
list may or may not include all the VPD pages that are able to be
returned by the device server". Testing had revealed a few devices
that supported some of the 0xBn pages but didn't actually list them in
page 0.

Julian Sikorski bisected a problem with his drive resetting during
discovery to the commit above. As it turns out, this particular drive
firmware will crash if we attempt to fetch page 0xB9.

Various approaches were attempted to work around this. In the end,
reinstating the logic that consults VPD page 0 before fetching any
other page was the path of least resistance. A firmware update for the
devices which originally compelled us to remove the check has since
been released.

Link: https://lore.kernel.org/r/20240214221411.2888112-1-martin.petersen@oracle.com
Fixes: c92a6b5d6335 ("scsi: core: Query VPD size before getting full page")
Cc: stable@vger.kernel.org
Cc: Bart Van Assche <bvanassche@acm.org>
Reported-by: Julian Sikorski <belegdol@gmail.com>
Tested-by: Julian Sikorski <belegdol@gmail.com>
Reviewed-by: Lee Duncan <lee.duncan@suse.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
scsi: sd: usb_storage: uas: Access media prior to querying device properties 2024-03-01T12:41:46+00:00 Martin K. Petersen martin.petersen@oracle.com 2024-02-13T14:33:06+00:00 d97e1c3602240bd35c48ef9aa978e0c47a511d03 commit 321da3dc1f3c92a12e3c5da934090d2992a8814c upstream. It has been observed that some USB/UAS devices return generic properties hardcoded in firmware for mode pages for a period of time after a device has been discovered. The reported properties are either garbage or they do not accurately reflect the characteristics of the physical storage device attached in the case of a bridge. Prior to commit 1e029397d12f ("scsi: sd: Reorganize DIF/DIX code to avoid calling revalidate twice") we would call revalidate several times during device discovery. As a result, incorrect values would eventually get replaced with ones accurately describing the attached storage. When we did away with the redundant revalidate pass, several cases were reported where devices reported nonsensical values or would end up in write-protected state. An initial attempt at addressing this issue involved introducing a delayed second revalidate invocation. However, this approach still left some devices reporting incorrect characteristics. Tasos Sahanidis debugged the problem further and identified that introducing a READ operation prior to MODE SENSE fixed the problem and that it wasn't a timing issue. Issuing a READ appears to cause the devices to update their state to reflect the actual properties of the storage media. Device properties like vendor, model, and storage capacity appear to be correctly reported from the get-go. It is unclear why these devices defer populating the remaining characteristics. Match the behavior of a well known commercial operating system and trigger a READ operation prior to querying device characteristics to force the device to populate the mode pages. The additional READ is triggered by a flag set in the USB storage and UAS drivers. We avoid issuing the READ for other transport classes since some storage devices identify Linux through our particular discovery command sequence. Link: https://lore.kernel.org/r/20240213143306.2194237-1-martin.petersen@oracle.com Fixes: 1e029397d12f ("scsi: sd: Reorganize DIF/DIX code to avoid calling revalidate twice") Cc: stable@vger.kernel.org Reported-by: Tasos Sahanidis <tasos@tasossah.com> Reviewed-by: Ewan D. Milne <emilne@redhat.com> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Tasos Sahanidis <tasos@tasossah.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit 321da3dc1f3c92a12e3c5da934090d2992a8814c upstream.

It has been observed that some USB/UAS devices return generic properties
hardcoded in firmware for mode pages for a period of time after a device
has been discovered. The reported properties are either garbage or they do
not accurately reflect the characteristics of the physical storage device
attached in the case of a bridge.

Prior to commit 1e029397d12f ("scsi: sd: Reorganize DIF/DIX code to
avoid calling revalidate twice") we would call revalidate several
times during device discovery. As a result, incorrect values would
eventually get replaced with ones accurately describing the attached
storage. When we did away with the redundant revalidate pass, several
cases were reported where devices reported nonsensical values or would
end up in write-protected state.

An initial attempt at addressing this issue involved introducing a
delayed second revalidate invocation. However, this approach still
left some devices reporting incorrect characteristics.

Tasos Sahanidis debugged the problem further and identified that
introducing a READ operation prior to MODE SENSE fixed the problem and that
it wasn't a timing issue. Issuing a READ appears to cause the devices to
update their state to reflect the actual properties of the storage
media. Device properties like vendor, model, and storage capacity appear to
be correctly reported from the get-go. It is unclear why these devices
defer populating the remaining characteristics.

Match the behavior of a well known commercial operating system and
trigger a READ operation prior to querying device characteristics to
force the device to populate the mode pages.

The additional READ is triggered by a flag set in the USB storage and
UAS drivers. We avoid issuing the READ for other transport classes
since some storage devices identify Linux through our particular
discovery command sequence.

Link: https://lore.kernel.org/r/20240213143306.2194237-1-martin.petersen@oracle.com
Fixes: 1e029397d12f ("scsi: sd: Reorganize DIF/DIX code to avoid calling revalidate twice")
Cc: stable@vger.kernel.org
Reported-by: Tasos Sahanidis <tasos@tasossah.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Tested-by: Tasos Sahanidis <tasos@tasossah.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/aio: Restrict kiocb_set_cancel_fn() to I/O submitted via libaio 2024-03-01T12:41:46+00:00 Bart Van Assche bvanassche@acm.org 2024-02-15T20:47:38+00:00 1dc7d74fe456944a9b1c57bd776280249f441ac6 commit b820de741ae48ccf50dd95e297889c286ff4f760 upstream. If kiocb_set_cancel_fn() is called for I/O submitted via io_uring, the following kernel warning appears: WARNING: CPU: 3 PID: 368 at fs/aio.c:598 kiocb_set_cancel_fn+0x9c/0xa8 Call trace: kiocb_set_cancel_fn+0x9c/0xa8 ffs_epfile_read_iter+0x144/0x1d0 io_read+0x19c/0x498 io_issue_sqe+0x118/0x27c io_submit_sqes+0x25c/0x5fc __arm64_sys_io_uring_enter+0x104/0xab0 invoke_syscall+0x58/0x11c el0_svc_common+0xb4/0xf4 do_el0_svc+0x2c/0xb0 el0_svc+0x2c/0xa4 el0t_64_sync_handler+0x68/0xb4 el0t_64_sync+0x1a4/0x1a8 Fix this by setting the IOCB_AIO_RW flag for read and write I/O that is submitted by libaio. Suggested-by: Jens Axboe <axboe@kernel.dk> Cc: Christoph Hellwig <hch@lst.de> Cc: Avi Kivity <avi@scylladb.com> Cc: Sandeep Dhavale <dhavale@google.com> Cc: Jens Axboe <axboe@kernel.dk> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Kent Overstreet <kent.overstreet@linux.dev> Cc: stable@vger.kernel.org Signed-off-by: Bart Van Assche <bvanassche@acm.org> Link: https://lore.kernel.org/r/20240215204739.2677806-2-bvanassche@acm.org Signed-off-by: Christian Brauner <brauner@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> 
commit b820de741ae48ccf50dd95e297889c286ff4f760 upstream.

If kiocb_set_cancel_fn() is called for I/O submitted via io_uring, the
following kernel warning appears:

WARNING: CPU: 3 PID: 368 at fs/aio.c:598 kiocb_set_cancel_fn+0x9c/0xa8
Call trace:
 kiocb_set_cancel_fn+0x9c/0xa8
 ffs_epfile_read_iter+0x144/0x1d0
 io_read+0x19c/0x498
 io_issue_sqe+0x118/0x27c
 io_submit_sqes+0x25c/0x5fc
 __arm64_sys_io_uring_enter+0x104/0xab0
 invoke_syscall+0x58/0x11c
 el0_svc_common+0xb4/0xf4
 do_el0_svc+0x2c/0xb0
 el0_svc+0x2c/0xa4
 el0t_64_sync_handler+0x68/0xb4
 el0t_64_sync+0x1a4/0x1a8

Fix this by setting the IOCB_AIO_RW flag for read and write I/O that is
submitted by libaio.

Suggested-by: Jens Axboe <axboe@kernel.dk>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Avi Kivity <avi@scylladb.com>
Cc: Sandeep Dhavale <dhavale@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: stable@vger.kernel.org
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240215204739.2677806-2-bvanassche@acm.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>