CVE Database

vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.26.0, the /v1/completions CompletionRequest.prompt field in vllm/entrypoints/openai/completion/protocol.py accepts an unbounded list[str] or list[list[int]], prompt_to_seq() in vllm/renderers/inputs/preprocess.py and OnlineRenderer.preprocess_completion() in vllm/renderers/online_renderer.py expand every element, and vllm/entrypoints/openai/completion/serving.py creates one engine generator and response slot per prompt, allowing an authenticated API client to exhaust CPU, memory, async scheduling capacity, engine request slots, and response buffering with one request. This issue is fixed in version 0.26.0.

Ninja Tables Pro 5.2.11 contains an embedded malicious code vulnerability introduced via a tampered plugin build served through a decommissioned update server. The tampered build introduced a rogue PHP file (app/Library/updater/NinjaTableDataSync.php) that established a backdoor REST API endpoint, dropped persistent PHP files in mu-plugins and uploads directories, installed a passwordless administrator account, and registered scheduled tasks that survived plugin removal.

Fluent Forms Pro 6.2.7 contains an embedded malicious code vulnerability introduced via a tampered plugin build served through a decommissioned update server. The tampered build introduced a rogue PHP file (libs/class-license-sync.php), loaded via a require_once directive added to fluentformpro.php, that established a backdoor REST API endpoint, dropped persistent PHP files in mu-plugins and uploads directories, installed a passwordless administrator account, and registered scheduled tasks that survived plugin removal.

PostGIS before 3.7.0beta2 contains an out-of-bounds read vulnerability that allows attackers to cause memory disclosure or a server crash by supplying a malformed FlatGeobuf buffer. The FlatGeobuf property metadata decoder verifies that a string length field is present but fails to verify that the subsequent string body is contained within the supplied buffer before materializing it into a SQL-visible value, enabling memory disclosure or denial of service.

The address_standardizer extension for PostGIS through 3.7.0, fixed in commit 423570b, contains an out-of-bounds write vulnerability that allows a database user with the ability to supply caller-controlled relation names to standardize_address() to trigger memory corruption by providing a rules table with a classification Type value exceeding the fixed class range. Attackers can craft a malicious rules table entry with an oversized rule type value that is used without bounds checking as an index into an internal output-link table, resulting in an out-of-bounds write.

CVE-2026-55401 is a null dereference vulnerability on the load-balancing sub-system of Secure Access servers prior to 14.57. Attackers can send an unauthenticated packet to a Secure Access server with load balancing enabled, which results in the internal load balancer crashing. After a successful attack, the Secure Access server is still able to accept connections and is still able to issue a failover to connected clients. ‍ https://www.first.org/cvss/calculator/4.0#CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:L/SC:N/SI:N/SA:L

CVE-2026-55400 is an integer underflow in Secure Access servers prior to version 14.57. Attackers with an authenticated session can send specially crafted traffic to a server in a non-default configuration and cause a persistent denial of service.

Cross-site Scripting in the Markdown renderer in maalfer Pentestify before 2.3.2 allows authenticated users to execute arbitrary JavaScript in the application origin via a Markdown link whose URL contains a double quote, which closes the anchor's href attribute because the renderer's sanitization step does not escape quotes

A vulnerability was found in SourceCodester Simple Student Information System. Affected by this vulnerability is an unknown functionality of the file app/admin/departments/view_department.php. Performing a manipulation of the argument ID results in sql injection. The attack is possible to be carried out remotely. The exploit has been made public and could be used.

Perl versions from 5.9.4 before 5.41.9 produce incorrect regular expression match results when a stale failure flag ends the Aho-Corasick prescan early in S_find_byclass. The prescan walks the subject for positions where the full pattern could match, and the engine tries it from the leftmost one recorded. A failing transition sets the failed flag, and a later successful transition does not clear it, so the prescan reads the stale flag as a failure and stops before it can record a candidate that starts earlier. It takes a subject where one candidate is recorded and a later character then forces a fallback through a fail link that succeeds. Example: "ABCDE" =~ m/ABCF|BCDE|C/; # matches C at offset 2, not BCDE "ABCDE" =~ m/ABCF|BCDE|C(G)/; # no match, BCDE missed An alternation like this can miss input it should match, or match it on the wrong branch, so an access or filtering decision made from the result can be wrong.

vLLM is an inference and serving engine for large language models. Prior to 0.27.0, an integer overflow in blockIdx.x * 2 * d in activation_kernels.cu can cause act_and_mul_kernel to consume another batched user's input, allowing a request processed in the same inference batch to receive a partial or complete copy of another user's inference result. This issue is fixed in version 0.27.0.

vLLM is an inference and serving engine for large language models. From 0.20.2rc0 until 0.26.0, safe_load_prompt_embeds in vllm/renderers/embed_utils.py uses torch.sparse.check_sparse_tensor_invariants, whose process-global save, enable, and restore state can be raced by concurrent prompt_embeds parts submitted to POST /v1/chat/completions through AsyncMultiModalItemTracker.resolve_items, asyncio.gather, and the default executor, allowing an invalid sparse tensor to reach tensor.to_dense despite the CVE-2025-62164 guard when enable_prompt_embeds is enabled. This issue is fixed in version 0.26.0.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the structured_outputs.regex parameter in vllm/v1/structured_output/backend_lm_format_enforcer.py is passed to lmformatenforcer.RegexParser without compile_regex_with_timeout or validation in validate_structured_output_request_lm_format_enforcer, allowing an unauthenticated /v1/completions request against the lm-format-enforcer backend to consume a CPU core and stall the structured-output engine path with a catastrophic regular expression. This issue is fixed in version 0.26.0.

vLLM is an inference and serving engine for large language models. Prior to 0.26.0, the validation_exception_handler in vllm/entrypoints/openai/server_utils.py converts FastAPI RequestValidationError objects with str(exc), and sanitize_message in vllm/entrypoints/utils.py does not remove traceback-style file paths, allowing unauthenticated malformed JSON requests to /v1/chat/completions, /v1/completions, /tokenize, and /detokenize to disclose the OS username, home and virtual-environment paths, Python version, internal package structure, line numbers, and endpoint handler names. This issue is fixed in version 0.26.0.

OpenList a file list program that supports multiple storage. Prior to 4.2.4, the authenticated /api/fs/batch_rename handler in server/handles/fsbatch.go authorizes only the source directory produced by user.JoinPath(req.SrcDir) and validates renameObject.NewName with checkRelativePath, but does not validate attacker-controlled renameObject.SrcName, supplied as src_name, before concatenating it with the authorized path and passing the result to fs.Rename. A user with rename permission can use traversal segments in src_name to make path normalization select a file outside the authorized directory and configured base path, resulting in cross-user file integrity loss, limited availability impact, and file-existence disclosure through success or error responses. This issue is fixed in version 4.2.4.

Netty is an asynchronous, event-driven network application framework. Prior to 4.1.136.Final and 4.2.16.Final, io.netty.handler.codec.dns.AbstractDnsRecord, io.netty.handler.codec.dns.DefaultDnsRecordDecoder.decodeRecord(), and io.netty.handler.codec.dns.DnsCodecUtil.decompressDomainName() failed to release retained or newly allocated ByteBuf objects when IDN.toASCII() or encodeDomainName() rejected a malformed domain name, allowing unauthenticated remote DNS packets to leak direct memory incrementally until denial of service. This issue is fixed in versions 4.1.136.Final and 4.2.16.Final.

Netty is an asynchronous, event-driven network application framework. Prior to 4.1.136.Final and 4.2.16.Final, io.netty.handler.codec.xml.XmlFrameDecoder.decode() failed to preserve closing-tag parser state across invocations, so an unauthenticated remote attacker could trickle-feed repeated </ sequences that repeatedly rescanned the accumulated buffer and exhausted an EventLoop thread's CPU, causing denial of service with a maxFrameLength of 1 MB. This issue is fixed in versions 4.1.136.Final and 4.2.16.Final.

Oh My Posh is the most customisable and low-latency cross platform/shell prompt renderer. Prior to 29.35.1, write(s rune) in src/terminal/writer.go emitted attacker-controlled current directory names and Git metadata, including Commit.Subject, Commit.Author.Name, Commit.Author.Email, and RawUpstreamURL, without removing C0/C1 terminal control characters such as ESC, BEL, CSI, and OSC, allowing terminal escape sequence injection during prompt rendering that could overwrite the clipboard, spoof the prompt or screen, manipulate the window title, or disrupt the terminal. This issue is fixed in version 29.35.1.

Oh My Posh is the most customisable and low-latency cross platform/shell prompt renderer. Prior to 29.35.1, the setStyle() function in src/segments/path.go passed pt.Path, which includes raw folder names, to template.Render, whose function map exposes cmd, so an attacker-controlled directory name containing a Go template expression could execute arbitrary operating system commands as the current user whenever the prompt rendered inside that directory or a descendant. This issue is fixed in version 29.35.1.

rsync daemon 2.0.0 before 3.5.0 contains a denial of service vulnerability that allows unauthenticated remote attackers to exhaust daemon connection slots by stalling the handshake process before or after module selection without triggering the I/O timeout. Attackers can open many simultaneous connections and trickle data at the minimum rate to avoid timeout, or stall entirely before module selection where no timeout applies, consuming all available connection slots and denying service to legitimate clients.

rsync 3.1.0 before 3.5.0 contains an authorization bypass in auth users directive parsing. The auth users parser uses comma-only tokenization when splitting the user list, which fails to correctly handle entries of the form @Group Name where the group name contains a space. The space within the group name causes the parser to split the entry at the space boundary, discarding the deny rule associated with the group. An authenticated user whose username or group membership would be denied by an @Group Name auth users entry can connect to a restricted module because the deny rule is silently discarded during parsing.

rsync 3.1.0 before 3.5.0 contains a signed integer overflow vulnerability in the I/O timeout implementation that allows attackers to permanently disable connection timeouts by injecting MSG_IO_TIMEOUT messages carrying non-positive (zero or negative) values. Attackers can craft malicious MSG_IO_TIMEOUT messages that cause the timeout variable to wrap to a non-positive value, preventing the timeout check from firing and enabling idle or stalled connections to hold daemon slots indefinitely, leading to resource exhaustion.

rsync 3.2.5 before 3.5.0 contains a heap out-of-bounds write vulnerability that allows remote unauthenticated attackers to write one attacker-controlled byte past the end of a heap allocation by supplying a crafted files-from entry. Attackers can trigger the vulnerability against a read-only rsync daemon module by providing a files-from entry containing both an interior and trailing backslash, causing the add_implied_include() function to under-count the trailing backslash when sizing the destination buffer.

rsync 2.3.3 before 3.5.0 contains a path traversal vulnerability that allows a malicious sender to escape the module root by exploiting symlinks within the module file tree when using --partial-dir or --backup-dir options. Attackers with write access to place a symlink under the module root, or who can exploit a pre-existing trusted symlink, can direct file writes to locations outside the intended module root, achieving arbitrary file write relative to the module root parent.

rsync 3.0.0 before 3.5.0 contains a null pointer dereference vulnerability in the daemon child process that allows remote attackers to crash the daemon by sending a file list whose first entry is a dot entry not typed as a directory. The daemon dereferences the first file list entry as a directory structure pointer without verifying the entry type, resulting in an invalid or uninitialized pointer dereference that terminates the client connection.

rsync 3.0.0 before 3.5.0 contains an out-of-bounds write vulnerability that allows attackers to corrupt memory by triggering HLINK_BUMP processing on file entries with the FLAG_HLINKED flag set while the hard-link preservation option is inactive. Attackers can exploit the missing F_SUM field in the file_struct layout to access memory past the end of the allocated structure, corrupting adjacent heap or stack data.

rsync 3.2.3 before 3.5.0 contains an out-of-bounds write in parse_size_arg() where the return value of snprintf() is used directly as an index into a .bss-segment array without bounds checking. When snprintf truncates the formatted size string, the return value equals the number of characters that would have been written including the truncated portion, and this value may exceed the array length. The subsequent indexed write targets memory outside the intended array bounds, corrupting .bss memory.

rsync 3.0.1 before 3.5.0 contains an out-of-bounds write vulnerability in the read_args() function that allows a malicious sender to corrupt adjacent heap memory by sending a crafted argument list. When the argument count causes the argv allocation to be exactly full, the trailing NULL terminator is written one slot beyond the allocation boundary, corrupting adjacent heap memory.

rsync 3.4.2 before 3.5.0 contains a denial of service vulnerability that allows a remote sender to exhaust system resources by specifying the --zt short alias for --compress-threads, which bypasses the refuse options directive's string matching on long option names. Attackers can specify --zt=N with a large value to spawn an unbounded number of Zstandard worker threads on the receiver, exhausting available thread and memory resources.

rsync 3.2.0 through 3.2.3 (openssl mode) and rsync-ssl through 3.4.4 (stunnel mode) contain a TLS certificate validation vulnerability that allows on-path attackers to intercept encrypted sessions by presenting self-signed or otherwise invalid certificates. Attackers can exploit the failure to validate server TLS certificates against a trusted CA or verify certificate hostname matching to decrypt or tamper with rsync session content without detection by the client.