| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
mm/huge_memory: fix folio isn't locked in softleaf_to_folio()
On arm64 server, we found folio that get from migration entry isn't locked
in softleaf_to_folio(). This issue triggers when mTHP splitting and
zap_nonpresent_ptes() races, and the root cause is lack of memory barrier
in softleaf_to_folio(). The race is as follows:
CPU0 CPU1
deferred_split_scan() zap_nonpresent_ptes()
lock folio
split_folio()
unmap_folio()
change ptes to migration entries
__split_folio_to_order() softleaf_to_folio()
set flags(including PG_locked) for tail pages folio = pfn_folio(softleaf_to_pfn(entry))
smp_wmb() VM_WARN_ON_ONCE(!folio_test_locked(folio))
prep_compound_page() for tail pages
In __split_folio_to_order(), smp_wmb() guarantees page flags of tail pages
are visible before the tail page becomes non-compound. smp_wmb() should
be paired with smp_rmb() in softleaf_to_folio(), which is missed. As a
result, if zap_nonpresent_ptes() accesses migration entry that stores tail
pfn, softleaf_to_folio() may see the updated compound_head of tail page
before page->flags.
This issue will trigger VM_WARN_ON_ONCE() in pfn_swap_entry_folio()
because of the race between folio split and zap_nonpresent_ptes()
leading to a folio incorrectly undergoing modification without a folio
lock being held.
This is a BUG_ON() before commit 93976a20345b ("mm: eliminate further
swapops predicates"), which in merged in v6.19-rc1.
To fix it, add missing smp_rmb() if the softleaf entry is migration entry
in softleaf_to_folio() and softleaf_to_page().
[tujinjiang@huawei.com: update function name and comments] |
| In the Linux kernel, the following vulnerability has been resolved:
net: annotate data-races around sk->sk_{data_ready,write_space}
skmsg (and probably other layers) are changing these pointers
while other cpus might read them concurrently.
Add corresponding READ_ONCE()/WRITE_ONCE() annotations
for UDP, TCP and AF_UNIX. |
| In the Linux kernel, the following vulnerability has been resolved:
net: add proper RCU protection to /proc/net/ptype
Yin Fengwei reported an RCU stall in ptype_seq_show() and provided
a patch.
Real issue is that ptype_seq_next() and ptype_seq_show() violate
RCU rules.
ptype_seq_show() runs under rcu_read_lock(), and reads pt->dev
to get device name without any barrier.
At the same time, concurrent writers can remove a packet_type structure
(which is correctly freed after an RCU grace period) and clear pt->dev
without an RCU grace period.
Define ptype_iter_state to carry a dev pointer along seq_net_private:
struct ptype_iter_state {
struct seq_net_private p;
struct net_device *dev; // added in this patch
};
We need to record the device pointer in ptype_get_idx() and
ptype_seq_next() so that ptype_seq_show() is safe against
concurrent pt->dev changes.
We also need to add full RCU protection in ptype_seq_next().
(Missing READ_ONCE() when reading list.next values)
Many thanks to Dong Chenchen for providing a repro. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: ctnetlink: remove refcounting in expectation dumpers
Same pattern as previous patch: do not keep the expectation object
alive via refcount, only store a cookie value and then use that
as the skip hint for dump resumption.
AFAICS this has the same issue as the one resolved in the conntrack
dumper, when we do
if (!refcount_inc_not_zero(&exp->use))
to increment the refcount, there is a chance that exp == last, which
causes a double-increment of the refcount and subsequent memory leak. |
| An out-of-bounds read flaw was found in the Linux kernel’s TeleTYpe subsystem. The issue occurs in how a user triggers a race condition using ioctls TIOCSPTLCK and TIOCGPTPEER and TIOCSTI and TCXONC with leakage of memory in the flush_to_ldisc function. This flaw allows a local user to crash the system or read unauthorized random data from memory. |
| In the Linux kernel, the following vulnerability has been resolved:
tcp: Don't call skb_clone_and_charge_r() for close()d listener in tcp_v6_do_rcv().
tcp_v6_do_rcv() no longer calls skb_clone_and_charge_r() for
TCP_LISTEN since commit 073d89808c06 ("net: fix data-races around
sk->sk_forward_alloc").
However, there is still a small race window between tcp_v6_rcv()
and tcp_v6_do_rcv(), where concurrent close() changes TCP_LISTEN
to TCP_CLOSE, causing skb_clone_and_charge_r() to be called
locklessly and resulting in the splat below. [0]
Let's avoid calling skb_clone_and_charge_r() for TCP_CLOSE as well.
This is fine for non-listeners because tcp_rcv_state_process()
drops skb for TCP_CLOSE and opt_skb was freed immediately anyway.
[0]:
sk->sk_forward_alloc
WARNING: net/ipv4/af_inet.c:162 at inet_sock_destruct+0x64d/0x810 net/ipv4/af_inet.c:162, CPU#1: ksoftirqd/1/28
Modules linked in:
CPU: 1 UID: 0 PID: 28 Comm: ksoftirqd/1 Not tainted 7.2.0 #17 PREEMPT(full)
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.17.0-debian-1.17.0-1 04/01/2014
RIP: 0010:inet_sock_destruct+0x64d/0x810 net/ipv4/af_inet.c:162
Code: 3d 49 ff e9 06 fd ff ff e8 d0 5b 83 f8 90 0f 0b 90 e9 35 fe ff ff e8 c2 5b 83 f8 90 0f 0b 90 e9 c5 fe ff ff e8 b4 5b 83 f8 90 <0f> 0b 90 e9 04 ff ff ff e8 a6 5b 83 f8 90 0f 0b 90 e9 65 fe ff ff
RSP: 0018:ffffc90000677bb8 EFLAGS: 00010246
RAX: 0000000000000000 RBX: ffff8880117bde80 RCX: ffffffff8957eb41
RDX: ffff88801dad5d00 RSI: ffffffff8957ec3c RDI: 0000000000000005
RBP: 00000000fffff000 R08: ffffffff8957eb41 R09: 00000000fffff000
R10: 0000000000000005 R11: 0000000000000000 R12: dffffc0000000000
R13: ffff8880117bdf10 R14: ffffffff81c08eb7 R15: 0000000000000003
FS: 0000000000000000(0000) GS:ffff8880d7ae5000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007f93a1021138 CR3: 00000000207a9000 CR4: 0000000000350ef0
Call Trace:
<TASK>
__sk_destruct+0x82/0xae0 net/core/sock.c:2356
rcu_do_batch kernel/rcu/tree.c:2645 [inline]
rcu_core+0x59c/0x1100 kernel/rcu/tree.c:2897
handle_softirqs+0x1e4/0x9b0 kernel/softirq.c:622
run_ksoftirqd kernel/softirq.c:1076 [inline]
run_ksoftirqd+0x38/0x60 kernel/softirq.c:1068
smpboot_thread_fn+0x458/0xc80 kernel/smpboot.c:160
kthread+0x396/0x4a0 kernel/kthread.c:436
ret_from_fork+0x8e0/0xe40 arch/x86/kernel/process.c:158
ret_from_fork_asm+0x1a/0x30 arch/x86/entry/entry_64.S:245
</TASK> |
| In the Linux kernel, the following vulnerability has been resolved:
wifi: mac80211: don't start a ROC while scanning
The ROC work can be pending when a scan starts (which requires
ROC list to be empty, but that's possible), and then a new ROC
can be added to the list and the work will pick it up.
Avoid starting that ROC if a scan made it between things, as
otherwise we'll hit a warning later:
WARNING: net/mac80211/offchannel.c:404 at ieee80211_start_next_roc+0x256/0x2d0
Workqueue: events_unbound cfg80211_wiphy_work
Call Trace:
__ieee80211_scan_completed+0x4fd/0xe40 net/mac80211/scan.c:537
ieee80211_scan_work+0x472/0x1ff0 net/mac80211/scan.c:1193
cfg80211_wiphy_work+0x410/0x570 net/wireless/core.c:513 |
| In the Linux kernel, the following vulnerability has been resolved:
wifi: cfg80211: ibss: ref BSS entry for joined event
When the IBSS is joined, we only record the BSSID/channel in the event
and look up the BSS entry when processing it. However, that's racy,
e.g. a new scan with NL80211_SCAN_FLAG_FLUSH can remove it, causing a
warning in the event work:
!bss
WARNING: net/wireless/ibss.c:37 at __cfg80211_ibss_joined+0x3d3/0x440
Workqueue: cfg80211 cfg80211_event_work
cfg80211_process_wdev_events+0x39f/0x5b0 net/wireless/util.c:1144
cfg80211_process_rdev_events+0xa1/0x110 net/wireless/util.c:1179
cfg80211_event_work+0x2f/0x40 net/wireless/core.c:393
Do the lookup early (the driver is expected to only join an IBSS that
has a BSS entry) and keep a reference to it. |
| In the Linux kernel, the following vulnerability has been resolved:
RDMA/core: fix refcount bug in iwpm_get_nlmsg_request()
iwpm_get_nlmsg_request() initializes refcount _after_ list_add_tail()
making it accessible to global list where another CPU can kref_get()
on nlmsg_request causing a refcount "addition on 0" bug. Fix this
by initializing kref _before_ list_add_tail() so refcount for
nlmsg_request can be incremented/decremented normally. In addition,
also initialize every field before list_add_tail(). |
| In the Linux kernel, the following vulnerability has been resolved:
RDMA/core: Reject unregistering netdevs in ib_get_eth_speed
ib_device_get_netdev() intentionally returns a referenced net_device even
when it is unregistering, so matching and cleanup callers can still find
the association. The reference keeps struct net_device allocated, but does
not guarantee that the device remains operational.
ib_get_eth_speed() uses the returned device operationally by invoking its
ethtool callback. Although that call is made under RTNL, the function does
not verify the registration state first. An asynchronous RDMA port query
can therefore call into a netdev after NETDEV_UNREGISTER and ndo_uninit
have completed.
Check for NETREG_REGISTERED while holding RTNL and return -ENODEV for a
device which is being unregistered. Keeping RTNL across the check and the
ethtool operation prevents unregister from starting between them.
Keep the speed fallback and warning under RTNL as well, so the warning can
safely read netdev->name. Drop the netdev reference before releasing RTNL
once all accesses to the device are complete. |
| In the Linux kernel, the following vulnerability has been resolved:
openvswitch: avoid reallocating confirmed conntrack labels
ovs_ct_get_conn_labels() adds the labels extension when a conntrack
entry does not have one. Confirmed conntracks can be read locklessly,
so adding an extension may reallocate and free the extension block
while another CPU accesses it.
Only add the extension for unconfirmed conntracks. A confirmed
conntrack without labels now fails the caller's label operation instead
of reallocating its extension storage. |
| In the Linux kernel, the following vulnerability has been resolved:
ipv6: mcast: use copy-on-write RCU updates in ip6_mc_source()
pmc->sflist is read locklessly under rcu_read_lock() by
inet6_mc_check() during packet reception in the UDP and RAW
multicast receive paths.
ip6_mc_source() mutated psl->sl_addr and psl->sl_count in-place
when adding or removing a source filter. Additionally, when expanding
the filter buffer, newpsl was published via rcu_assign_pointer()
before writing the new source into the array.
Because 16-byte struct in6_addr writes are not atomic and array
shifting is not synchronized with RCU readers, concurrent readers in
inet6_mc_check() could read torn IPv6 addresses or observe
duplicated/missed source entries.
Fix this by switching ip6_mc_source() to copy-on-write RCU updates:
allocate and fully populate newpsl before publishing it via
rcu_assign_pointer(), and reclaim the old filter via kfree_rcu(),
matching ip6_mc_msfilter().
Also remove the now unused IP6_SFBLOCK macro. |
| In the Linux kernel, the following vulnerability has been resolved:
cgroup: Avoid iteration of dying tasks with zero refcount
The commit 260fbcb92bbea ("cgroup: Move dying_tasks cleanup from
cgroup_task_release() to cgroup_task_free()") extended the lifetime of
tasks on the dying_tasks list.
The iterators have provision to go through dying_tasks because of
dying threadgroup leaders or explicit CSS_TASK_ITER_WITH_DEAD, however,
it was expected that such tasks can obtain a new reference (that is
possible before cgroup_task_release()/put_task_struct_rcu_user()).
The tasks after cgroup_task_release() and before cgroup_task_free()
are subject to race when they may or may not have ->usage count > 0.
The race window is between css_task_iter_next() invocations
when css_set_lock is released and we may arrive at a new ->task_pos.
The iterator should not attempt to resurrect tasks whose ->usage count
dropped to zero. (When that happens, __put_task_struct_rcu_cb() is
already imminent and the returned task_struct would could be used
after free.)
As for the fix, we cannot simply check the signal->live count of a task
on the dying list because that won't distinguish regular zombies waiting
to be reaped from RCU remnant tasks that are going to be free'd.
Therefore add an extra check to rule out ->usage==0 tasks from any
iteration.
The repeat: loop in css_task_iter_advance() doesn't consider ->usage
count, so add a new loop to css_task_iter_next() to skip de-used tasks
on the dying_list.
Rough illustration of the possible race
R (reader of cgroup.procs) T (thread) L (group leader)
--------------------------------- -------------------------------- --------------------------------
L exits, signal->live > 0
cgroup_task_dead(L)
css_set_skip_task_iters() // skips only cset->tasks
list_add_tail(&L->cg_list, &cset->dying_tasks)
css_task_iter_next()
take css_set_lock
css_task_iter_advance()
leader && signal->live != 0
=> it->task_pos = &L->cg_list
release css_set_lock
T exits
--signal->live == 0
cgroup_task_dead(T) // css_set_lock
release_task(T)
cgroup_task_release(T)
release_task(L) // zap_leader
cgroup_task_release(L)
put_task_struct_rcu_user(L)
...RCU...
put_task_struct(L)
L->usage = 0
/* L still on dying_tasks */
...RCU...
__put_task_struct(L)
css_task_iter_next() // another iteration
take css_set_lock
it->task_pos = &L->cg_list
get_task_struct(L)
=> addition on 0
drop css_set_lock
cgroup_task_free(L)
css_set_skip_task_iters() // dying skip comes too late
free_task(L)
cgroup_procs_show()
task_pid_vnr(L) |
| Race condition within a thread vulnerability in Apache Tomcat Native allowed client certificate verification requirements to be down-graded for some configurations.
This issue affects Apache Tomcat Native: from 2.0.0 through 2.0.15, from 1.3.0 through 1.3.8. Unsupported versions may also be affected.
Users are recommended to upgrade to version 2.0.16 or 1.3.9, which fixes the issue. |
| In the Linux kernel, the following vulnerability has been resolved:
mlxsw: spectrum_ptp: Fix napi_gro_receive() call from GC workqueue context
Currently mlxsw_sp1_ptp_ht_gc_collect() is run from the PTP
garbage-collection workqueue, rather than the NAPI poll context. For any
unmatched PTP entries carrying an SKB, it calls
mlxsw_sp1_ptp_unmatched_finish() -> mlxsw_sp1_ptp_packet_finish(). For
ingress packets, this calls mlxsw_sp_rx_listener_no_mark_func(). The end
of that function is the following:
skb->protocol = eth_type_trans(skb, skb->dev);
napi_gro_receive(mlxsw_skb_cb(skb)->rx_md_info.napi, skb);
The napi pointer is one that was placed in the SKB control block when the
trapped packet was received in the NAPI context. Later, when the GC reaps
the unmatched entry (up to MLXSW_SP1_PTP_HT_GC_TIMEOUT later), the call to
napi_gro_receive() mutates the NAPI instance's GRO list, which is unsafe
if the poll is running concurrently on another CPU.
In mlxsw_sp1_ptp_ht_gc_collect(), local_bh_disable() is called to prevent
softirq processing, but this only applies to the local CPU. Additionally,
its comment is stale. It states that mlxsw_sp1_ptp_unmatched_finish()
invokes netif_receive_skb(). This has not been accurate since the
referenced commit; this patch makes that comment accurate again.
mlxsw_pci_napi_devs_init() calls netif_threaded_enable() on the NAPI RX
net_device without any conditions. The NAPI instance's poll, which may be
running concurrent to the GC, is running as an independently-scheduled
kthread which may be on a different CPU. The call to local_bh_disable()
does not guard against this.
If a tx-timestamp timeout produces an unmatched entry (which can be easily
reproduced by running ptp4l and waiting for a port to reach the
UNCALIBRATED/SLAVE state) while the owning NAPI thread is in the middle of
a poll on another CPU, both sides mutate the GRO list concurrently, as
shown below:
[39.846] port 1 (swp1): MASTER to UNCALIBRATED on RS_SLAVE
list_add corruption. next->prev should be prev (ffff8d620faf4138), but was ffff8d624150f700. (next=ffff8d620faf4138).
kernel BUG at lib/list_debug.c:29!
Oops: invalid opcode: 0000 [#1] SMP PTI
CPU: 1 UID: 0 PID: 539 Comm: napi/mlxsw_rx-0 Not tainted 6.18.48 #1-NixOS PREEMPT(lazy)
Hardware name: Mellanox Technologies Ltd. MSN2410/VMOD0001, BIOS 4.6.5 09/13/2018
RIP: 0010:__list_add_valid_or_report+0x79/0xb0
RSP: 0018:ffffcdf8c0f27c08 EFLAGS: 00010246
RAX: 0000000000000075 RBX: ffff8d624150fd00 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000001 RDI: ffff8d6315d1e540
RBP: ffff8d620faf4070 R08: 0000000000000000 R09: 00000000ffffdfff
R10: ffffffffa5c60fe0 R11: ffffcdf8c0f27ab8 R12: 0000000000000003
R13: 000000000000003d R14: 00000000000001bc R15: 0000000000000001
FS: 0000000000000000(0000) GS:ffff8d636f63f000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000562689a60c24 CR3: 000000015f224004 CR4: 00000000001726f0
Call Trace:
<TASK>
gro_receive_skb+0xee/0x230
mlxsw_sp1_ptp_got_packet+0x61/0x140 [mlxsw_spectrum]
mlxsw_core_skb_receive+0xdf/0x1b0 [mlxsw_core]
mlxsw_pci_napi_poll_cq_rx+0x780/0x9d0 [mlxsw_pci]
__napi_poll+0x31/0x1e0
napi_threaded_poll_loop+0x16b/0x1c0
napi_threaded_poll+0x71/0xa0
kthread+0xfb/0x260
ret_from_fork+0x22d/0x260
ret_from_fork_asm+0x1a/0x30
</TASK>
Kernel panic - not syncing: Fatal exception in interrupt
The machinery that leads to this kernel panic has not been changed between
6.18.48 and mainline.
This patch adds an ingress-delivery helper for the PTP packet_finish()
path that calls netif_receive_skb() instead of napi_gro_receive().
netif_receive_skb(), unlike napi_gro_receive(), can be called from outside
of the NAPI instance's poll context, which can occur at the call site for
this path. RX stats accounting and the skb->dev assignment are still
preserved; the only change is the delivery call itself.
This removes GR
---truncated--- |
| A malicious external BuildKit frontend can send requests using the internal API that can create conditions for a data race that can cause the BuildKit daemon to panic. |
| Race condition in the DOM: Content Processes component. This vulnerability was fixed in Firefox 156, Firefox ESR 153.3, Thunderbird 156, and Thunderbird 153.3. |
| In the Linux kernel, the following vulnerability has been resolved:
net/rds: clear cp_flags bits individually in rds_conn_path_reset()
rds_conn_path_reset() wipes the whole flag word with a plain
cp->cp_flags = 0 store. Every other accessor of that word uses
atomic bitops, and some of them can run concurrently with the reset:
RDS_LL_SEND_FULL is set from rds_send_xmit() and cleared from the
transport completion paths, neither of which holds anything that
excludes the shutdown worker. A plain store racing an atomic
read-modify-write on the same word is a data race, and whichever
side loses has its update silently discarded.
Clear the two bits the reset is actually responsible for instead.
RDS_IN_XMIT and RDS_RECV_REFILL need no store at all here: they
belong to the caller, rds_conn_shutdown(), which waits for both to be
clear before calling the transport shutdown and this reset.
This also gives every bit in cp_flags a single well-defined writer
discipline, which the following patches rely on when they turn
RDS_IN_XMIT and RDS_RECV_REFILL into bit locks held across the
teardown: a blanket store mid-teardown would destroy lock ownership
that an atomic clear preserves.
Oracle UEK carries the same conversion ("net/rds: Preserve essential
connection state flags"), motivated by its asynchronous shutdown
state machine, whose progress and destroy flags must survive the
reset. UEK's variant also clears RDS_IN_XMIT and RDS_RECV_REFILL
because there the reset runs as the final step of a teardown that
owns both bits, making those clears its unlock. Upstream that
release belongs in rds_conn_shutdown(): once a later patch in this
series turns the two bits into locks held across the teardown, ending
ownership needs release semantics and a wake-up that a plain clear
inside the reset would not provide.
Based on Oracle UEK commit "net/rds: Preserve essential connection
state flags" by Gerd Rausch. |
| In the Linux kernel, the following vulnerability has been resolved:
net/rds: acquire RDS_IN_XMIT in rds_tcp_reset_callbacks()
rds_tcp_reset_callbacks() quiesces the transmit path by setting the
path state to RDS_CONN_RESETTING and then waiting for RDS_IN_XMIT to
be sampled clear before swapping the underlying socket and calling
rds_send_path_reset().
Sampling the bit clear is not the same as owning it: rds_send_xmit()
can re-acquire RDS_IN_XMIT right after the wait_event() returns. Its
state recheck after taking the lock is a store-buffering pattern (the
resetter writes the state and reads the bit, the sender writes the
bit and reads the state) and acquire_in_xmit() is only an acquire
operation, so on weakly ordered architectures both sides can miss
each other's write and the transmit path then runs concurrently with
rds_send_path_reset() rewriting cp_xmit_* state - which is exactly
what the comment above rds_send_path_reset() tells its callers to
prevent.
Take the lock instead, hold it across the socket swap and
rds_send_path_reset(), and release it with a wake-up at the end. The
lock-ordering constraint documented above the wait still holds: the
lock is acquired before lock_sock(), so a sender inside tcp_sendmsg()
can never be waited on while we hold the socket lock.
Two details of the old code go away with the same change:
- t_sock is now read only after the lock is acquired. The old code
cached it before waiting; the teardown in rds_conn_shutdown()
releases that socket and clears t_sock, so a pointer cached before
the wait can be stale by the time the accept path resumes. Reading
it under RDS_IN_XMIT is what makes the exclusion complete once the
teardown owns the same lock, which the next patch arranges; until
then the teardown still only samples the bit, and the two paths
remain as exposed to each other as they are today.
- The old !osock early path called rds_send_path_reset() with no
serialization at all. It now runs under the lock like the normal
path. The conditional RDS_CONN_RESETTING transition of the
previous patch happens before the socket check either way: a path
found without a socket is either still connecting (its reconnect
worker blocked on t_conn_path_lock) and legitimately goes
RESETTING -> UP on the new socket, or it has been torn down
meanwhile and is dropped.
The in-function comment describing the old wait-based quiesce is
rewritten to describe the lock-based one, and the stale block comment
above the function (which still described a return value and an
incomplete list of t_sock writers) is refreshed to name all four
writers - the connect, accept, teardown and swap paths - and what
serializes each of them. |
| In the Linux kernel, the following vulnerability has been resolved:
net/rds: acquire the fastpath locks in rds_conn_shutdown()
rds_conn_shutdown() quiesces the transmit and receive-refill paths by
waiting for RDS_IN_XMIT and RDS_RECV_REFILL to be sampled clear, and
then runs the transport shutdown and rds_conn_path_reset(). Sampling
the bits clear is not the same as owning them: the moment after the
wait_event() returns, rds_send_xmit() can re-acquire RDS_IN_XMIT (or
rds_ib_recv_refill() can re-acquire RDS_RECV_REFILL) and run
concurrently with the teardown.
The sender does recheck the connection state after taking the lock,
but that recheck is a classic store-buffering pattern: teardown writes
the state and reads the bit while the sender writes the bit and reads
the state. acquire_in_xmit() is only an acquire operation, so on
weakly ordered architectures both sides can miss each other's write,
and the transmit path then runs while the transport zeroes its rings
(e.g. rds_ib_ring_init()) and rds_send_path_reset() rewrites the
transmit state under it.
Oracle UEK fixed the same class of crashes - a 14-year tail of
BUG_ON()s in rds_ib_sub_signaled(), unexpected op-codes and NULL
dereferences in rds_ib_send_cqe_handler() during failover testing -
by making the teardown path *acquire* the fastpath bit locks instead
of testing them ("rds: Make sure transmit path and connection
tear-down does not run concurrently"). Ownership of a single word is
decided by RMW atomicity, so no cross-variable ordering is needed.
Do the same here: take both locks before calling the transport
shutdown, hold them across rds_conn_path_reset(), and release them
explicitly with a wake-up afterwards. Both are released with
clear_bit_unlock(), so that the ring re-initialization done by the
transport shutdown and the transmit state rewritten by
rds_send_path_reset() are ordered before either bit is seen clear by
the next acquire_in_xmit() or acquire_refill().
The fastpath users of these bits - rds_send_xmit() and
rds_ib_recv_refill() - are trylock style and back off while teardown
owns the locks, so no new lock dependency is introduced for them.
rds_tcp_reset_callbacks() is different: since the previous patch it
acquires RDS_IN_XMIT as well, and it blocks doing so, so its wait now
spans the teardown instead of at most one send batch. That waiter
runs from rds_tcp_accept_one() on the single-threaded krdsd workqueue
and holds rds_tcp_accept_lock and t_conn_path_lock while it waits, so
a duelling SYN accepted while its path is being torn down parks
accept processing for the duration of the teardown - for TCP bounded
by the (up to 5 s) drain loop in rds_tcp_conn_path_shutdown(). An IB
path's drain in rds_ib_conn_path_shutdown() has no round cap, but no
blocking waiter either: rds_tcp_reset_callbacks() is the only blocking
acquirer of these bits and waits only on its own TCP path, and the
fastpaths are trylock-and-back-off on both transports, so a long IB
drain lengthens only that path's own quiesce. The
window is narrow: the accept-side state check has to pass before the
teardown moves the path to RDS_CONN_DISCONNECTING.
Because krdsd is a single global workqueue, everything else queued
there - accept processing for other connections and network
namespaces, and the flush_workqueue(rds_wq) in rds_tcp_listen_stop()
during namespace teardown - waits behind the parked accept worker for
that time. It cannot deadlock, although the waits do point at each
other: the teardown blocks until the bit's holder releases it, and
the holder may be that krdsd accept worker. The holder finishes
without needing anything the teardown owns: the sync cancels
rds_tcp_reset_callbacks() issues target cp_send_w and cp_recv_w on
the path's ordered cp_wq, whose only execution slot is occupied by
the blocked cp_down_w itself, so they are pending at most and cancel
without flushing - a reliance on cp_wq being ordered that is now
noted next to those cancels (on
---truncated--- |
