| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
ice: change XDP RxQ frag_size from DMA write length to xdp.frame_sz
The only user of frag_size field in XDP RxQ info is
bpf_xdp_frags_increase_tail(). It clearly expects whole buff size instead
of DMA write size. Different assumptions in ice driver configuration lead
to negative tailroom.
This allows to trigger kernel panic, when using
XDP_ADJUST_TAIL_GROW_MULTI_BUFF xskxceiver test and changing packet size to
6912 and the requested offset to a huge value, e.g.
XSK_UMEM__MAX_FRAME_SIZE * 100.
Due to other quirks of the ZC configuration in ice, panic is not observed
in ZC mode, but tailroom growing still fails when it should not.
Use fill queue buffer truesize instead of DMA write size in XDP RxQ info.
Fix ZC mode too by using the new helper. |
| In the Linux kernel, the following vulnerability has been resolved:
net/sched: act_ife: Fix metalist update behavior
Whenever an ife action replace changes the metalist, instead of
replacing the old data on the metalist, the current ife code is appending
the new metadata. Aside from being innapropriate behavior, this may lead
to an unbounded addition of metadata to the metalist which might cause an
out of bounds error when running the encode op:
[ 138.423369][ C1] ==================================================================
[ 138.424317][ C1] BUG: KASAN: slab-out-of-bounds in ife_tlv_meta_encode (net/ife/ife.c:168)
[ 138.424906][ C1] Write of size 4 at addr ffff8880077f4ffe by task ife_out_out_bou/255
[ 138.425778][ C1] CPU: 1 UID: 0 PID: 255 Comm: ife_out_out_bou Not tainted 7.0.0-rc1-00169-gfbdfa8da05b6 #624 PREEMPT(full)
[ 138.425795][ C1] Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
[ 138.425800][ C1] Call Trace:
[ 138.425804][ C1] <IRQ>
[ 138.425808][ C1] dump_stack_lvl (lib/dump_stack.c:122)
[ 138.425828][ C1] print_report (mm/kasan/report.c:379 mm/kasan/report.c:482)
[ 138.425839][ C1] ? srso_alias_return_thunk (arch/x86/lib/retpoline.S:221)
[ 138.425844][ C1] ? __virt_addr_valid (./arch/x86/include/asm/preempt.h:95 (discriminator 1) ./include/linux/rcupdate.h:975 (discriminator 1) ./include/linux/mmzone.h:2207 (discriminator 1) arch/x86/mm/physaddr.c:54 (discriminator 1))
[ 138.425853][ C1] ? ife_tlv_meta_encode (net/ife/ife.c:168)
[ 138.425859][ C1] kasan_report (mm/kasan/report.c:221 mm/kasan/report.c:597)
[ 138.425868][ C1] ? ife_tlv_meta_encode (net/ife/ife.c:168)
[ 138.425878][ C1] kasan_check_range (mm/kasan/generic.c:186 (discriminator 1) mm/kasan/generic.c:200 (discriminator 1))
[ 138.425884][ C1] __asan_memset (mm/kasan/shadow.c:84 (discriminator 2))
[ 138.425889][ C1] ife_tlv_meta_encode (net/ife/ife.c:168)
[ 138.425893][ C1] ? ife_tlv_meta_encode (net/ife/ife.c:171)
[ 138.425898][ C1] ? srso_alias_return_thunk (arch/x86/lib/retpoline.S:221)
[ 138.425903][ C1] ife_encode_meta_u16 (net/sched/act_ife.c:57)
[ 138.425910][ C1] ? __pfx_do_raw_spin_lock (kernel/locking/spinlock_debug.c:114)
[ 138.425916][ C1] ? __asan_memcpy (mm/kasan/shadow.c:105 (discriminator 3))
[ 138.425921][ C1] ? __pfx_ife_encode_meta_u16 (net/sched/act_ife.c:45)
[ 138.425927][ C1] ? srso_alias_return_thunk (arch/x86/lib/retpoline.S:221)
[ 138.425931][ C1] tcf_ife_act (net/sched/act_ife.c:847 net/sched/act_ife.c:879)
To solve this issue, fix the replace behavior by adding the metalist to
the ife rcu data structure. |
| In the Linux kernel, the following vulnerability has been resolved:
net/sched: ets: fix divide by zero in the offload path
Offloading ETS requires computing each class' WRR weight: this is done by
averaging over the sums of quanta as 'q_sum' and 'q_psum'. Using unsigned
int, the same integer size as the individual DRR quanta, can overflow and
even cause division by zero, like it happened in the following splat:
Oops: divide error: 0000 [#1] SMP PTI
CPU: 13 UID: 0 PID: 487 Comm: tc Tainted: G E 6.19.0-virtme #45 PREEMPT(full)
Tainted: [E]=UNSIGNED_MODULE
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
RIP: 0010:ets_offload_change+0x11f/0x290 [sch_ets]
Code: e4 45 31 ff eb 03 41 89 c7 41 89 cb 89 ce 83 f9 0f 0f 87 b7 00 00 00 45 8b 08 31 c0 45 01 cc 45 85 c9 74 09 41 6b c4 64 31 d2 <41> f7 f2 89 c2 44 29 fa 45 89 df 41 83 fb 0f 0f 87 c7 00 00 00 44
RSP: 0018:ffffd0a180d77588 EFLAGS: 00010246
RAX: 00000000ffffff38 RBX: ffff8d3d482ca000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffd0a180d77660
RBP: ffffd0a180d77690 R08: ffff8d3d482ca2d8 R09: 00000000fffffffe
R10: 0000000000000000 R11: 0000000000000000 R12: 00000000fffffffe
R13: ffff8d3d472f2000 R14: 0000000000000003 R15: 0000000000000000
FS: 00007f440b6c2740(0000) GS:ffff8d3dc9803000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000003cdd2000 CR3: 0000000007b58002 CR4: 0000000000172ef0
Call Trace:
<TASK>
ets_qdisc_change+0x870/0xf40 [sch_ets]
qdisc_create+0x12b/0x540
tc_modify_qdisc+0x6d7/0xbd0
rtnetlink_rcv_msg+0x168/0x6b0
netlink_rcv_skb+0x5c/0x110
netlink_unicast+0x1d6/0x2b0
netlink_sendmsg+0x22e/0x470
____sys_sendmsg+0x38a/0x3c0
___sys_sendmsg+0x99/0xe0
__sys_sendmsg+0x8a/0xf0
do_syscall_64+0x111/0xf80
entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7f440b81c77e
Code: 4d 89 d8 e8 d4 bc 00 00 4c 8b 5d f8 41 8b 93 08 03 00 00 59 5e 48 83 f8 fc 74 11 c9 c3 0f 1f 80 00 00 00 00 48 8b 45 10 0f 05 <c9> c3 83 e2 39 83 fa 08 75 e7 e8 13 ff ff ff 0f 1f 00 f3 0f 1e fa
RSP: 002b:00007fff951e4c10 EFLAGS: 00000202 ORIG_RAX: 000000000000002e
RAX: ffffffffffffffda RBX: 0000000000481820 RCX: 00007f440b81c77e
RDX: 0000000000000000 RSI: 00007fff951e4cd0 RDI: 0000000000000003
RBP: 00007fff951e4c20 R08: 0000000000000000 R09: 0000000000000000
R10: 0000000000000000 R11: 0000000000000202 R12: 00007fff951f4fa8
R13: 00000000699ddede R14: 00007f440bb01000 R15: 0000000000486980
</TASK>
Modules linked in: sch_ets(E) netdevsim(E)
---[ end trace 0000000000000000 ]---
RIP: 0010:ets_offload_change+0x11f/0x290 [sch_ets]
Code: e4 45 31 ff eb 03 41 89 c7 41 89 cb 89 ce 83 f9 0f 0f 87 b7 00 00 00 45 8b 08 31 c0 45 01 cc 45 85 c9 74 09 41 6b c4 64 31 d2 <41> f7 f2 89 c2 44 29 fa 45 89 df 41 83 fb 0f 0f 87 c7 00 00 00 44
RSP: 0018:ffffd0a180d77588 EFLAGS: 00010246
RAX: 00000000ffffff38 RBX: ffff8d3d482ca000 RCX: 0000000000000000
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffd0a180d77660
RBP: ffffd0a180d77690 R08: ffff8d3d482ca2d8 R09: 00000000fffffffe
R10: 0000000000000000 R11: 0000000000000000 R12: 00000000fffffffe
R13: ffff8d3d472f2000 R14: 0000000000000003 R15: 0000000000000000
FS: 00007f440b6c2740(0000) GS:ffff8d3dc9803000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000003cdd2000 CR3: 0000000007b58002 CR4: 0000000000172ef0
Kernel panic - not syncing: Fatal exception
Kernel Offset: 0x30000000 from 0xffffffff81000000 (relocation range: 0xffffffff80000000-0xffffffffbfffffff)
---[ end Kernel panic - not syncing: Fatal exception ]---
Fix this using 64-bit integers for 'q_sum' and 'q_psum'. |
| In the Linux kernel, the following vulnerability has been resolved:
tracing: Fix WARN_ON in tracing_buffers_mmap_close
When a process forks, the child process copies the parent's VMAs but the
user_mapped reference count is not incremented. As a result, when both the
parent and child processes exit, tracing_buffers_mmap_close() is called
twice. On the second call, user_mapped is already 0, causing the function to
return -ENODEV and triggering a WARN_ON.
Normally, this isn't an issue as the memory is mapped with VM_DONTCOPY set.
But this is only a hint, and the application can call
madvise(MADVISE_DOFORK) which resets the VM_DONTCOPY flag. When the
application does that, it can trigger this issue on fork.
Fix it by incrementing the user_mapped reference count without re-mapping
the pages in the VMA's open callback. |
| A logging issue was addressed with improved data redaction. This issue is fixed in macOS Sequoia 15.7.5, macOS Sonoma 14.8.5, macOS Tahoe 26.4. An app may be able to access sensitive user data. |
| In the Linux kernel, the following vulnerability has been resolved:
net: bridge: fix nd_tbl NULL dereference when IPv6 is disabled
When booting with the 'ipv6.disable=1' parameter, the nd_tbl is never
initialized because inet6_init() exits before ndisc_init() is called
which initializes it. Then, if neigh_suppress is enabled and an ICMPv6
Neighbor Discovery packet reaches the bridge, br_do_suppress_nd() will
dereference ipv6_stub->nd_tbl which is NULL, passing it to
neigh_lookup(). This causes a kernel NULL pointer dereference.
BUG: kernel NULL pointer dereference, address: 0000000000000268
Oops: 0000 [#1] PREEMPT SMP NOPTI
[...]
RIP: 0010:neigh_lookup+0x16/0xe0
[...]
Call Trace:
<IRQ>
? neigh_lookup+0x16/0xe0
br_do_suppress_nd+0x160/0x290 [bridge]
br_handle_frame_finish+0x500/0x620 [bridge]
br_handle_frame+0x353/0x440 [bridge]
__netif_receive_skb_core.constprop.0+0x298/0x1110
__netif_receive_skb_one_core+0x3d/0xa0
process_backlog+0xa0/0x140
__napi_poll+0x2c/0x170
net_rx_action+0x2c4/0x3a0
handle_softirqs+0xd0/0x270
do_softirq+0x3f/0x60
Fix this by replacing IS_ENABLED(IPV6) call with ipv6_mod_enabled() in
the callers. This is in essence disabling NS/NA suppression when IPv6 is
disabled. |
| In the Linux kernel, the following vulnerability has been resolved:
HID: Add HID_CLAIMED_INPUT guards in raw_event callbacks missing them
In commit 2ff5baa9b527 ("HID: appleir: Fix potential NULL dereference at
raw event handle"), we handle the fact that raw event callbacks
can happen even for a HID device that has not been "claimed" causing a
crash if a broken device were attempted to be connected to the system.
Fix up the remaining in-tree HID drivers that forgot to add this same
check to resolve the same issue. |
| In the Linux kernel, the following vulnerability has been resolved:
bpf, arm64: Force 8-byte alignment for JIT buffer to prevent atomic tearing
struct bpf_plt contains a u64 target field. Currently, the BPF JIT
allocator requests an alignment of 4 bytes (sizeof(u32)) for the JIT
buffer.
Because the base address of the JIT buffer can be 4-byte aligned (e.g.,
ending in 0x4 or 0xc), the relative padding logic in build_plt() fails
to ensure that target lands on an 8-byte boundary.
This leads to two issues:
1. UBSAN reports misaligned-access warnings when dereferencing the
structure.
2. More critically, target is updated concurrently via WRITE_ONCE() in
bpf_arch_text_poke() while the JIT'd code executes ldr. On arm64,
64-bit loads/stores are only guaranteed to be single-copy atomic if
they are 64-bit aligned. A misaligned target risks a torn read,
causing the JIT to jump to a corrupted address.
Fix this by increasing the allocation alignment requirement to 8 bytes
(sizeof(u64)) in bpf_jit_binary_pack_alloc(). This anchors the base of
the JIT buffer to an 8-byte boundary, allowing the relative padding math
in build_plt() to correctly align the target field. |
| In the Linux kernel, the following vulnerability has been resolved:
RDMA/ionic: Fix kernel stack leak in ionic_create_cq()
struct ionic_cq_resp resp {
__u32 cqid[2]; // offset 0 - PARTIALLY SET (see below)
__u8 udma_mask; // offset 8 - SET (resp.udma_mask = vcq->udma_mask)
__u8 rsvd[7]; // offset 9 - NEVER SET <- LEAK
};
rsvd[7]: 7 bytes of stack memory leaked unconditionally.
cqid[2]: The loop at line 1256 iterates over udma_idx but skips indices
where !(vcq->udma_mask & BIT(udma_idx)). The array has 2 entries but
udma_count could be 1, meaning cqid[1] might never be written via
ionic_create_cq_common(). If udma_mask only has bit 0 set, cqid[1] (4
bytes) is also leaked. So potentially 11 bytes leaked. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: nf_tables: clone set on flush only
Syzbot with fault injection triggered a failing memory allocation with
GFP_KERNEL which results in a WARN splat:
iter.err
WARNING: net/netfilter/nf_tables_api.c:845 at nft_map_deactivate+0x34e/0x3c0 net/netfilter/nf_tables_api.c:845, CPU#0: syz.0.17/5992
Modules linked in:
CPU: 0 UID: 0 PID: 5992 Comm: syz.0.17 Not tainted syzkaller #0 PREEMPT(full)
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 02/12/2026
RIP: 0010:nft_map_deactivate+0x34e/0x3c0 net/netfilter/nf_tables_api.c:845
Code: 8b 05 86 5a 4e 09 48 3b 84 24 a0 00 00 00 75 62 48 8d 65 d8 5b 41 5c 41 5d 41 5e 41 5f 5d c3 cc cc cc cc cc e8 63 6d fa f7 90 <0f> 0b 90 43
+80 7c 35 00 00 0f 85 23 fe ff ff e9 26 fe ff ff 89 d9
RSP: 0018:ffffc900045af780 EFLAGS: 00010293
RAX: ffffffff89ca45bd RBX: 00000000fffffff4 RCX: ffff888028111e40
RDX: 0000000000000000 RSI: 00000000fffffff4 RDI: 0000000000000000
RBP: ffffc900045af870 R08: 0000000000400dc0 R09: 00000000ffffffff
R10: dffffc0000000000 R11: fffffbfff1d141db R12: ffffc900045af7e0
R13: 1ffff920008b5f24 R14: dffffc0000000000 R15: ffffc900045af920
FS: 000055557a6a5500(0000) GS:ffff888125496000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fb5ea271fc0 CR3: 000000003269e000 CR4: 00000000003526f0
Call Trace:
<TASK>
__nft_release_table+0xceb/0x11f0 net/netfilter/nf_tables_api.c:12115
nft_rcv_nl_event+0xc25/0xdb0 net/netfilter/nf_tables_api.c:12187
notifier_call_chain+0x19d/0x3a0 kernel/notifier.c:85
blocking_notifier_call_chain+0x6a/0x90 kernel/notifier.c:380
netlink_release+0x123b/0x1ad0 net/netlink/af_netlink.c:761
__sock_release net/socket.c:662 [inline]
sock_close+0xc3/0x240 net/socket.c:1455
Restrict set clone to the flush set command in the preparation phase.
Add NFT_ITER_UPDATE_CLONE and use it for this purpose, update the rbtree
and pipapo backends to only clone the set when this iteration type is
used.
As for the existing NFT_ITER_UPDATE type, update the pipapo backend to
use the existing set clone if available, otherwise use the existing set
representation. After this update, there is no need to clone a set that
is being deleted, this includes bound anonymous set.
An alternative approach to NFT_ITER_UPDATE_CLONE is to add a .clone
interface and call it from the flush set path. |
| In the Linux kernel, the following vulnerability has been resolved:
gve: fix incorrect buffer cleanup in gve_tx_clean_pending_packets for QPL
In DQ-QPL mode, gve_tx_clean_pending_packets() incorrectly uses the RDA
buffer cleanup path. It iterates num_bufs times and attempts to unmap
entries in the dma array.
This leads to two issues:
1. The dma array shares storage with tx_qpl_buf_ids (union).
Interpreting buffer IDs as DMA addresses results in attempting to
unmap incorrect memory locations.
2. num_bufs in QPL mode (counting 2K chunks) can significantly exceed
the size of the dma array, causing out-of-bounds access warnings
(trace below is how we noticed this issue).
UBSAN: array-index-out-of-bounds in
drivers/net/ethernet/drivers/net/ethernet/google/gve/gve_tx_dqo.c:178:5 index 18 is out of
range for type 'dma_addr_t[18]' (aka 'unsigned long long[18]')
Workqueue: gve gve_service_task [gve]
Call Trace:
<TASK>
dump_stack_lvl+0x33/0xa0
__ubsan_handle_out_of_bounds+0xdc/0x110
gve_tx_stop_ring_dqo+0x182/0x200 [gve]
gve_close+0x1be/0x450 [gve]
gve_reset+0x99/0x120 [gve]
gve_service_task+0x61/0x100 [gve]
process_scheduled_works+0x1e9/0x380
Fix this by properly checking for QPL mode and delegating to
gve_free_tx_qpl_bufs() to reclaim the buffers. |
| In the Linux kernel, the following vulnerability has been resolved:
pinctrl: cirrus: cs42l43: Fix double-put in cs42l43_pin_probe()
devm_add_action_or_reset() already invokes the action on failure,
so the explicit put causes a double-put. |
| This issue was addressed with improved checks. This issue is fixed in macOS Tahoe 26.4. An app may be able to access sensitive user data. |
| In the Linux kernel, the following vulnerability has been resolved:
Squashfs: check metadata block offset is within range
Syzkaller reports a "general protection fault in squashfs_copy_data"
This is ultimately caused by a corrupted index look-up table, which
produces a negative metadata block offset.
This is subsequently passed to squashfs_copy_data (via
squashfs_read_metadata) where the negative offset causes an out of bounds
access.
The fix is to check that the offset is within range in
squashfs_read_metadata. This will trap this and other cases. |
| In the Linux kernel, the following vulnerability has been resolved:
ice: Fix memory leak in ice_set_ringparam()
In ice_set_ringparam, tx_rings and xdp_rings are allocated before
rx_rings. If the allocation of rx_rings fails, the code jumps to
the done label leaking both tx_rings and xdp_rings. Furthermore, if
the setup of an individual Rx ring fails during the loop, the code jumps
to the free_tx label which releases tx_rings but leaks xdp_rings.
Fix this by introducing a free_xdp label and updating the error paths to
ensure both xdp_rings and tx_rings are properly freed if rx_rings
allocation or setup fails.
Compile tested only. Issue found using a prototype static analysis tool
and code review. |
| In the Linux kernel, the following vulnerability has been resolved:
tracing/dma: Cap dma_map_sg tracepoint arrays to prevent buffer overflow
The dma_map_sg tracepoint can trigger a perf buffer overflow when
tracing large scatter-gather lists. With devices like virtio-gpu
creating large DRM buffers, nents can exceed 1000 entries, resulting
in:
phys_addrs: 1000 * 8 bytes = 8,000 bytes
dma_addrs: 1000 * 8 bytes = 8,000 bytes
lengths: 1000 * 4 bytes = 4,000 bytes
Total: ~20,000 bytes
This exceeds PERF_MAX_TRACE_SIZE (8192 bytes), causing:
WARNING: CPU: 0 PID: 5497 at kernel/trace/trace_event_perf.c:405
perf buffer not large enough, wanted 24620, have 8192
Cap all three dynamic arrays at 128 entries using min() in the array
size calculation. This ensures arrays are only as large as needed
(up to the cap), avoiding unnecessary memory allocation for small
operations while preventing overflow for large ones.
The tracepoint now records the full nents/ents counts and a truncated
flag so users can see when data has been capped.
Changes in v2:
- Use min(nents, DMA_TRACE_MAX_ENTRIES) for dynamic array sizing
instead of fixed DMA_TRACE_MAX_ENTRIES allocation (feedback from
Steven Rostedt)
- This allocates only what's needed up to the cap, avoiding waste
for small operations
Reviwed-by: Sean Anderson <sean.anderson@linux.dev> |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: xt_CT: drop pending enqueued packets on template removal
Templates refer to objects that can go away while packets are sitting in
nfqueue refer to:
- helper, this can be an issue on module removal.
- timeout policy, nfnetlink_cttimeout might remove it.
The use of templates with zone and event cache filter are safe, since
this just copies values.
Flush these enqueued packets in case the template rule gets removed. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: nf_tables: release flowtable after rcu grace period on error
Call synchronize_rcu() after unregistering the hooks from error path,
since a hook that already refers to this flowtable can be already
registered, exposing this flowtable to packet path and nfnetlink_hook
control plane.
This error path is rare, it should only happen by reaching the maximum
number hooks or by failing to set up to hardware offload, just call
synchronize_rcu().
There is a check for already used device hooks by different flowtable
that could result in EEXIST at this late stage. The hook parser can be
updated to perform this check earlier to this error path really becomes
rarely exercised.
Uncovered by KASAN reported as use-after-free from nfnetlink_hook path
when dumping hooks. |
| In the Linux kernel, the following vulnerability has been resolved:
bridge: cfm: Fix race condition in peer_mep deletion
When a peer MEP is being deleted, cancel_delayed_work_sync() is called
on ccm_rx_dwork before freeing. However, br_cfm_frame_rx() runs in
softirq context under rcu_read_lock (without RTNL) and can re-schedule
ccm_rx_dwork via ccm_rx_timer_start() between cancel_delayed_work_sync()
returning and kfree_rcu() being called.
The following is a simple race scenario:
cpu0 cpu1
mep_delete_implementation()
cancel_delayed_work_sync(ccm_rx_dwork);
br_cfm_frame_rx()
// peer_mep still in hlist
if (peer_mep->ccm_defect)
ccm_rx_timer_start()
queue_delayed_work(ccm_rx_dwork)
hlist_del_rcu(&peer_mep->head);
kfree_rcu(peer_mep, rcu);
ccm_rx_work_expired()
// on freed peer_mep
To prevent this, cancel_delayed_work_sync() is replaced with
disable_delayed_work_sync() in both peer MEP deletion paths, so
that subsequent queue_delayed_work() calls from br_cfm_frame_rx()
are silently rejected.
The cc_peer_disable() helper retains cancel_delayed_work_sync()
because it is also used for the CC enable/disable toggle path where
the work must remain re-schedulable. |
| In the Linux kernel, the following vulnerability has been resolved:
af_unix: Give up GC if MSG_PEEK intervened.
Igor Ushakov reported that GC purged the receive queue of
an alive socket due to a race with MSG_PEEK with a nice repro.
This is the exact same issue previously fixed by commit
cbcf01128d0a ("af_unix: fix garbage collect vs MSG_PEEK").
After GC was replaced with the current algorithm, the cited
commit removed the locking dance in unix_peek_fds() and
reintroduced the same issue.
The problem is that MSG_PEEK bumps a file refcount without
interacting with GC.
Consider an SCC containing sk-A and sk-B, where sk-A is
close()d but can be recv()ed via sk-B.
The bad thing happens if sk-A is recv()ed with MSG_PEEK from
sk-B and sk-B is close()d while GC is checking unix_vertex_dead()
for sk-A and sk-B.
GC thread User thread
--------- -----------
unix_vertex_dead(sk-A)
-> true <------.
\
`------ recv(sk-B, MSG_PEEK)
invalidate !! -> sk-A's file refcount : 1 -> 2
close(sk-B)
-> sk-B's file refcount : 2 -> 1
unix_vertex_dead(sk-B)
-> true
Initially, sk-A's file refcount is 1 by the inflight fd in sk-B
recvq. GC thinks sk-A is dead because the file refcount is the
same as the number of its inflight fds.
However, sk-A's file refcount is bumped silently by MSG_PEEK,
which invalidates the previous evaluation.
At this moment, sk-B's file refcount is 2; one by the open fd,
and one by the inflight fd in sk-A. The subsequent close()
releases one refcount by the former.
Finally, GC incorrectly concludes that both sk-A and sk-B are dead.
One option is to restore the locking dance in unix_peek_fds(),
but we can resolve this more elegantly thanks to the new algorithm.
The point is that the issue does not occur without the subsequent
close() and we actually do not need to synchronise MSG_PEEK with
the dead SCC detection.
When the issue occurs, close() and GC touch the same file refcount.
If GC sees the refcount being decremented by close(), it can just
give up garbage-collecting the SCC.
Therefore, we only need to signal the race during MSG_PEEK with
a proper memory barrier to make it visible to the GC.
Let's use seqcount_t to notify GC when MSG_PEEK occurs and let
it defer the SCC to the next run.
This way no locking is needed on the MSG_PEEK side, and we can
avoid imposing a penalty on every MSG_PEEK unnecessarily.
Note that we can retry within unix_scc_dead() if MSG_PEEK is
detected, but we do not do so to avoid hung task splat from
abusive MSG_PEEK calls. |
