| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| pypdf is a free and open-source pure-python PDF library. Prior to 6.21.0, an attacker can craft a PDF containing a content stream with many small operations, causing parsing to consume large amounts of memory. This issue is fixed in version 6.21.0. |
| External control of file name or path and uncontrolled resource consumption in Apache bRPC (version < 1.18.0) on all platforms allows remote attackers who can reach the builtin service port to create directories and truncate/overwrite files with the privileges of the server process, or to hold the process-wide CPU/contention profiler for an arbitrarily long time.
Root Cause:
1. The builtin /flags service allows gflags carrying a validator to be modified at runtime via ?setvalue=. The bvar/mbvar dump path gflags (bvar_dump_file, bvar_dump_tabs, mbvar_dump_file) were registered with an accept-anything validator whose only purpose was to wake the dumping thread. A remote attacker could therefore repoint the dump path (e.g. GET /flags/bvar_dump_file?setvalue=/victim/dir/target followed by GET /flags/bvar_dump?setvalue=true, or the equivalent for mbvar_dump_file, which is used verbatim without a forced suffix), making the dumping thread recursively create directories and open/truncate files at attacker-chosen locations. Tab names in bvar_dump_tabs were appended to the path without sanitization, allowing '..' to escape the configured directory.
2. The /pprof builtin service did not clamp the ?seconds= parameter to FLAGS_max_profiling_seconds (a limit that /hotspots enforced). A single request such as ?seconds=2147483647 made the handler sleep for decades while holding the process-wide gperftools CPU profiler or bthread contention profiler, so /hotspots/cpu and /hotspots/contention returned 503 for everyone else during that time.
Affected Scenarios:
Any Apache bRPC server exposing the builtin services (enabled by default on the same port as normal RPC traffic) to untrusted networks.
How to Fix:
(Choose one of the following options)
1. Upgrade bRPC to version 1.18.0, which fixes this issue.
2. Apply this patch: https://github.com/apache/brpc/pull/3523
Note:
After the fix, bvar_dump_file, bvar_dump_tabs and mbvar_dump_file can no longer be modified at runtime through the /flags builtin service. They can still be changed programmatically via google::SetCommandLineOption(), taking effect on the next dump cycle. The profiling duration of /pprof is now capped by the -max_profiling_seconds gflag (default 300 seconds), which can be adjusted by the server owner. |
| Uncontrolled resource consumption in the streaming RPC component of Apache bRPC (version < 1.18.0) allows a remote attacker to exhaust server memory and cause denial of service by sending a single RPC request that declares an excessive number of streams.
Root Cause:
In the streaming RPC feature, a client can declare multiple streams in one RPC request by putting extra stream identifiers into the repeated field `stream_settings.extra_stream_ids` of the RPC meta. On the server side, `StreamAccept` creates one response stream for every stream identifier declared by the request, without limiting the number of streams accepted in a single RPC. The request body size is only bounded by `max_body_size` (default 64MB), which is enough to carry millions of stream identifiers. Each accepted stream allocates its own bthread execution queue and associated resources, so a single crafted request can make the server create a huge number of streams and exhaust memory, leading to denial of service.
Affected Scenarios:
Use a bRPC server that serves streaming RPC and calls `StreamAccept` to accept streams from untrusted clients.
How to Fix:
(Choose one of the following options)
1. Upgrade bRPC to the 1.18.0 version.
2. Apply this patch: https://github.com/apache/brpc/pull/3493
Note:
The fix adds a new gflag `stream_max_streams_per_request` with a default value of 64. Requests declaring more streams than the limit are rejected before any response stream is created. Applications that legitimately need to accept more than 64 streams in one RPC must raise this flag; otherwise such requests will fail after applying the fix. |
| vLLM is an inference and serving engine for large language models. From 0.16.0 until 0.32.0, the /v1/responses endpoint allows an authenticated caller to place an attacker-controlled template in chat_template_kwargs[chat_template] when no server template is configured. ChatParams.get_apply_chat_template_kwargs() preserves that value without enforcing trust_request_chat_template, so the request can bypass the intended trust policy and cause CPU or memory exhaustion through Jinja rendering. This issue is fixed in version 0.32.0. |
| vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0. |
| zip is a zip library for Rust that supports reading and writing simple ZIP files. From 5.1.0 until 9.0.0, applications built with the legacy-zip feature can reserve output capacity from an attacker-controlled ZIP64 uncompressed size before validating the compressed stream. The ShrinkDecoder path in src/legacy/shrink.rs can panic with a capacity overflow during ZipArchive::extract, and the Reduce and Implode decoders contain the same unbounded preallocation pattern. An unhandled panic can terminate an archive-processing service. This issue is fixed in version 9.0.0. |
| Twisted is an event-based framework for internet applications, supporting Python 3.6+. Prior to 26.10.0rc1, twisted.words.protocols.irc.IRC.dataReceived retains all bytes received without a line feed in its per-connection buffer and applies no maximum line length. The twisted.words.service.IRCUser server subclass inherits this behavior, and buffering occurs before PASS, NICK, or USER authentication commands. An unauthenticated connection can continuously grow the buffer until the shared server process runs out of memory and drops all users. This issue is fixed in version 26.10.0rc1. |
| toml-node is a TOML parser for Node.js and the browser. Prior to 5.0.1, toml.parse() passes unbounded hexadecimal, octal, binary, and decimal digit sequences from lib/parser.js to BigInt() before reduceInteger() in lib/compiler.js checks whether the resulting value fits the TOML 64-bit integer range. Converting a sufficiently long integer string consumes super-linear CPU time and synchronously blocks the Node.js event loop, allowing one attacker-supplied TOML document to deny service to concurrent work in the process. The later range check cannot prevent the conversion cost because it executes after BigInt(). This issue is fixed in version 5.0.1. |
| Apache bRPC versions before 1.18.0 do not consistently enforce the configured max_body_size limit in non-progressive HTTP/1 and HTTP/2 parsing and in memcache and couchbase response parsing.
A remote unauthenticated peer can cause an affected process to buffer message data beyond the configured limit, which may lead to memory exhaustion and denial of service.
Users should upgrade to 1.18.0 or apply the patch from PR https://github.com/apache/brpc/pull/3382 . The fix rejects oversized buffered HTTP bodies, returns HTTP 413 for HTTP/1 requests, resets oversized HTTP/2 streams, and rejects oversized memcache and couchbase responses from their declared lengths. Progressive HTTP reading remains supported because it uses bounded buffering. |
| In the Linux kernel, the following vulnerability has been resolved:
HID: alps: unregister DualPoint Stick input device on remove
alps_input_configured() allocates a second input device ("DualPoint
Stick") with input_allocate_device() and registers it, but the
alps_driver struct has no .remove handler and input2 is not tracked in
hdev->inputs. The default remove path (hid_hw_stop -> hidinput_disconnect)
only iterates hdev->inputs, so input2 is never unregistered and leaks
on every device removal.
Add a .remove handler that stops the device first (preventing URB
callbacks from touching input2 during teardown) and then unregisters
input2. |
| In the Linux kernel, the following vulnerability has been resolved:
ipmr: account multicast table and route memory
A netadmin in a user+net namespace can create many IPv4 and IPv6
multicast routing tables with MRT_TABLE and MRT6_TABLE. Each unseen
id allocates an mr_table via the shared mr_table_alloc(), links it
into the per-net list, and leaves it until netns teardown. Those
objects were not charged to memcg, so the host unreclaimable slab
grows with the table count.
Account mr_table allocations with GFP_KERNEL_ACCOUNT and mark the
IPv4/IPv6 MFC caches SLAB_ACCOUNT. This matches the established
handling of IP addresses, routes and alternate interface names.
Unresolved MFC entries are still allocated from softIRQ with
GFP_ATOMIC and are not charged. They expire after 10 seconds and are
bounded by the socket receive queue; see commit 0079ad8e8dc3
("ipmr: remove hard code cache_resolve_queue_len limit"). |
| In the Linux kernel, the following vulnerability has been resolved:
mm/secretmem: properly account locked pages
secretmem accounts folios by treating memory as if it were mlock()'d and
thus limited by the RLIMIT_MEMLOCK limit.
However the folios are unevictable and remain so until the inode is
evicted, eliminating usual mlock() semantics - mapping folios then
unmapping them does not clear their unevictable state, since it depends on
AS_UNEVICTABLE, not PG_mlocked.
A user can therefore easily work around the RLIMIT_MEMLOCK limit - simply
map then unmap and VmLck no longer counts the secretmem range. Worse,
folios are not accounted in the process's RSS, meaning the OOM killer
won't know to kill the process.
Repeatedly mapping/unmapping (or forking) can then result in the
consumption of all available system memory with unevictable folios and
cause system instability.
A secretmem fd can be passed between processes and over fork so a
per-process limit simply does not make sense, so follow the precedent set
by io_uring, perf, skbuff, iommufd and xdp by tracking the number of
locked pages in user_struct->locked_vm.
Since the scope tracked is actually inode lifetime, the RLIMIT_MEMLOCK
applies per-user not per-process, so it doesn't make sense to bypass for
users with CAP_IPC_LOCK, therefore remove this bypass.
There is simply no reason to carry on marking the mapping as mlock()'d
since it's misleading and the lifecycle is now correctly handled, so
remove this too.
Note that secretmem does not support any form of truncation (including
hole punching) and the folios are unreclaimable, so the folios need only
be accounted on fault and unaccounted on inode destruction.
__secretmem_account_pages() is more or less a duplicate of the code that
io_uring etc. use, but since this is a bug fix that needs backporting,
defer any de-duplication efforts to a follow-up.
test_mlock_limit() asserts mlock_future_ok() on mmap(), however this has
been removed, so remove the test altogether for the fix. A new test will
be sent separately for upstream. |
| In the Linux kernel, the following vulnerability has been resolved:
RDMA/erdma: complete object teardown when the destroy command fails
erdma_destroy_qp(), erdma_destroy_cq(), erdma_dereg_mr(), and
erdma_destroy_ah() returned early when erdma_post_cmd_wait() failed,
leaking the queue buffers, MTTs, doorbells and the STAG, QPN, CQN and AHN
identifiers. A command timeout clears ERDMA_CMDQ_STATE_OK_BIT and
permanently disables the command queue, so no retry can succeed; the RDMA
core keeps the object after a failed destructor and forced uverbs cleanup
then nulls the pointers, making the resources unreachable.
Warn on failure but release every software-owned resource and return
success, since during terminal destruction the hardware command result is
only diagnostic. |
| In the Linux kernel, the following vulnerability has been resolved:
bpf: Fix mmap_lock leak in irq_work path
stack_map_get_build_id_offset() introduced a per-CPU irq_work to defer
mmap_read_unlock() from NMI context, and bpf_find_vma() later reused the
same mmap_unlock_work. Both callers only check whether the work is busy
before taking mmap_lock, so a nested caller can reuse the slot before the
first caller queues it. Two read locks may then be acquired while only one
deferred unlock runs, leaking a read lock and blocking exit_mmap().
Reserve the per-CPU slot before mmap_read_trylock(). Use the same wrapper
in stackmap and bpf_find_vma() so both callers release the reservation on
trylock failure. Keep rejecting the slot while the irq_work remains busy.
Release it after the irq_work callback unlocks the mm. |
| In the Linux kernel, the following vulnerability has been resolved:
net/sched: account classifier filter allocations to memcg
Allocations in the tc classifier *_change() paths (filter objects,
per-CPU counters, and per-filter aux data) use plain GFP_KERNEL without
__GFP_ACCOUNT, allowing unprivileged users to pin kernel memory outside
memcg charging. The shared tcf_exts_init_ex() action array allocation in
cls_api.c was also uncharged; this patch closes it along with the
per-classifier filter-object/percpu/aux allocations that remain
unaccounted.
Add GFP_KERNEL_ACCOUNT to:
- the shared tcf_exts_init_ex() action array (cls_api.c), common to every
filter of every classifier (32 pointers, 256 bytes);
- the filter-object, per-CPU-counter, and per-filter aux allocations in
cls_basic, cls_bpf, cls_cgroup, cls_flow, cls_flower, cls_fw,
cls_matchall, cls_route and cls_u32;
- the u32_init_knode() replace-path knode allocation (cls_u32.c), which
allocates the same struct tc_u_knode + sel.keys on every replace of an
existing knode and was missed by the create-path-only conversion.
Also fix the cls_basic error path: basic_change() inserts fnew into the
IDR before allocating the per-CPU counter. If alloc_percpu() fails the
errout path kfree'd fnew without idr_remove, leaving a dangling pointer
in the IDR. With GFP_KERNEL_ACCOUNT the percpu alloc becomes failable
on demand (memcg at memory.max), making the dead path attacker-reachable
and burning the handle permanently. Add the idr_remove on the percpu
failure path, matching the basic_set_parms failure-path pattern.
Note: vega@nebusec.ai provided a poc for basic_cls, but it was easy to
extend to the other classifiers.
Conditions to recreate the bug:
- CONFIG_NET_SCHED, CONFIG_NET_CLS_* (the classifier being used),
CONFIG_NET_CLS_ACT, CONFIG_MEMCG, CONFIG_USER_NS, CONFIG_NET_NS.
- Unprivileged user in a fresh user+network namespace (unshare -Urn),
or root with CAP_NET_ADMIN.
- Create a large number of tc filters (e.g. tc filter add dev lo
ingress ... <classifier> ...) while watching a memcg-limited cgroup:
system slab grows far faster than memory.current, pinning kernel
memory outside memcg charging. |
| In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: force complete the KIQ ring fences on reset
Like the MES scheduler ring, the KIQ ring sets no_scheduler = true and uses a
polling fence, so it is skipped by the force-completion loop in
amdgpu_device_pre_asic_reset(). Its hw fence value lives in wb (GTT) memory and
survives a MODE1 reset while fence_drv.sync_seq keeps advancing, so after a
reset the first KIQ submission can poll forever on a seq that is never written
back.
Force complete the KIQ ring fences too so their hw fence is realigned to
sync_seq. |
| In the Linux kernel, the following vulnerability has been resolved:
Bluetooth: hci_sync: Set HCI_CMD_DRAIN_WORKQUEUE during device close
Since hci_dev_close_sync() can now be called during the reset path, we
should also set HCI_CMD_DRAIN_WORKQUEUE. This avoids queuing timeouts
while the hdev workqueue is being drained. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: nfnetlink_queue: do shared-unconfirmed check before segmentation
Ulrich reports a regression with nfqueue:
If an application did not set the 'F_GSO' capability flag and a gso
packet with an unconfirmed nf_conn entry is received all packets are
now dropped instead of queued, because the check happens after
skb_gso_segment(). In that case, we did have exclusive ownership
of the skb and its associated conntrack entry. The elevated use
count is due to skb_clone happening via skb_gso_segment().
Move the check so that its peformed vs. the aggregated packet.
Then, annotate the individual segments except the first one so we
can do a 2nd check at reinject time.
For the normal case, where userspace does in-order reinjects, this avoids
packet drops: first reinjected segment continues traversal and confirms
entry, remaining segments observe the confirmed entry.
While at it, simplify nf_ct_drop_unconfirmed(): We only care about
unconfirmed entries with a refcnt > 1, there is no need to special-case
dying entries.
This only happens with UDP. With TCP, the only unconfirmed packet will
be the TCP SYN, those aren't aggregated by GRO.
Next patch adds a udpgro test case to cover this scenario. |
| In the Linux kernel, the following vulnerability has been resolved:
drm/amd/display: Prevent crash when disable stream
[Why]
Disabling stream encoder invokes a function that no longer exists.
[How]
Check if the function declaration is NULL in disable stream encoder. |
| A flaw was found in glib-networking. When establishing connections using Transport Layer Security (TLS) version 1.3, the client-side session cache does not limit the number of session tickets retained for a single host. A malicious server can exploit this by sending excessive session ticket messages, causing the client application to continuously consume memory. This unbounded memory growth can exhaust system memory, resulting in a Denial of Service (DoS). |
