CVE Feed

The VobSub subtitle decoder libsub_vob (part of the Amlogic DroidLogic SubTitleService), as shipped on Nexbox A95X (Amlogic S905X) TV boxes, contains an infinite loop vulnerability in get_vob_spu() that allows unauthenticated network attackers or local users loading malicious files to trigger an unreachable loop exit condition via mismatched length fields. Attackers can send crafted AMLU data blocks over the unauthenticated tcp/10100 listener or via a malicious VobSub (.sub) file to cause the subtitle decode thread to spin at 100% CPU indefinitely, resulting in denial of service.

The pppoe_wrapper vendor daemon from the Amlogic Android SDK, as shipped on Nexbox A95X (Amlogic S905/S905X) TV boxes, contains a command injection vulnerability that allows local attackers to execute arbitrary commands as root by sending crafted messages to a world-writable AF_UNIX datagram socket at /dev/socket/pppoe_wrapper. Attackers can send a message beginning with 'pppd pty' followed by shell metacharacters and arbitrary commands, which are passed unfiltered to system(3) and executed by /system/bin/sh -c as root, resulting in full privilege escalation.

HDF5 through 2.2.0 contains an uncontrolled recursion vulnerability in H5B__iterate_helper() in src/H5B.c. When iterating a version 1 B-tree, the function does not verify that a child node’s level matches the expected level. That check is performed only in H5B__cache_deserialize() when a node is first loaded into the metadata cache. A crafted HDF5 file whose B-tree child pointer refers back to a node already in the cache, such as a node that points to itself, bypasses the check and causes unbounded recursion, leading to stack exhaustion and a crash. This can be triggered by calling H5Gget_info() or H5Gget_num_objs() on a group in the crafted file, which reaches H5B_iterate() through H5G__obj_info() and H5G__stab_count(). An attacker who can get a user or application to open a malicious file can cause a denial of service.

The Ultra Addons for Contact Form 7 plugin for WordPress is vulnerable to arbitrary file deletion due to insufficient file path validation in the _getObjAttr function in all versions up to, and including, 3.5.54. This makes it possible for unauthenticated attackers to delete arbitrary files on the server, which can easily lead to remote code execution when the right file is deleted (such as wp-config.php). This requires the PDF Generator add-on to be active and a Contact Form 7 form to be configured for PDF generation, as the WPCF7_Submission POP gadget chain relies on CF7 being present to complete the file deletion via __destruct(). Note: This vulnerability can only be exploited when the PDF generator feature is enabled.

For updatable cursors, which are enabled by default, psqlODBC adds hidden ctid and, for tables with OIDs, oid columns to the query. It copies each value into a 32-byte stack buffer using the length the server sent, without checking it against the buffer size, so a value of 32 bytes or more writes past the buffer. A server not trusted under current SSL settings can send such a value. So can any database user who can create a view with its own column named ctid that the application reads with an updatable cursor. Builds with _FORTIFY_SOURCE abort; other builds overwrite adjacent stack memory. Versions before psqlODBC 18.00.0006 are affected.

For an updatable cursor over a table without OIDs that has an auto-increment int4 column, psqlODBC uses that column to identify rows. It places the column's name, as reported by the server, between double quotes in its keyset queries and in the WHERE clause of its positioned SELECT, UPDATE, and DELETE statements, without doubling any double quote in the name. A column name that contains a double quote therefore changes the SQL that the driver runs with the application's privileges, including which rows a positioned UPDATE or DELETE affects. Any database user who can create a table that the application reads with an updatable cursor can choose such a name, and so can a server not trusted under current SSL settings. Versions before psqlODBC 18.00.0006 are affected.

On Unix-like systems, when logging is enabled, psqlODBC creates its log files in /tmp by default, named from the program name, user name, and process ID, with the default permissions, typically world-readable, and follows symbolic links when opening them. A local user can read query text and data from another user's logs, and can place a symbolic link at the predictable path so that the driver appends log data to a file of the attacker's choosing with the victim's privileges. Versions before psqlODBC 18.00.0006 are affected.

Several psqlODBC entry points use an application-supplied column number or row index as an array index without validating it. SQLBindCol with a column number above 32767 wraps the binding count negative and a later copy runs off the array. A negative SQL_DESC_COUNT on an application row descriptor unbinds columns at negative indexes. SQLColAttribute with a very large column number uses a negative index. SQLSetPos(SQL_ADD) does not check the row number against the rowset size. An application that forwards untrusted column or row numbers exposes its process to out-of-bounds reads and writes. Versions before psqlODBC 18.00.0006 are affected.

When psqlODBC quotes a schema or table name into a 128-byte stack buffer, its length check lets it write up to two bytes past the end of the buffer for a long name. The names come from the application's SQL text: the table of a statement the driver parses to describe its columns, and the target of the last INSERT when the driver rewrites SELECT @@IDENTITY. In the @@IDENTITY rewrite, a column name sent by the server can also write one byte past the buffer. The bytes written past the buffer are a double quote and a NUL. An application that forwards untrusted SQL text to the driver exposes its process to stack corruption. Versions before psqlODBC 18.00.0006 are affected.

SQLSpecialColumns in psqlODBC splices the table and schema names it receives into its catalog query without escaping them, while the other catalog functions escape them. An application that passes an untrusted table or schema name to SQLSpecialColumns lets the caller execute arbitrary SQL with the privileges of the connection. Versions before psqlODBC 18.00.0006 are affected.

The psqlODBC parser for ODBC escape sequences in application SQL text writes the argument offsets of a {fn} escape into a fixed 16-entry stack array without a bound and does not initialize the array. An escape with too many arguments overflows the stack, and a reference to an unsupplied argument copies arbitrary bytes into the query. The copy of a {d}, {t}, {ts}, or {escape} body also writes one byte past a 128-byte stack buffer. An application that forwards untrusted SQL text to the driver exposes its process to memory corruption. Versions before psqlODBC 18.00.0006 are affected.

psqlODBC trusts counts and lengths in server data in three places. SQLBulkOperations uses a server-supplied row count in a size subtraction that can underflow. SQLStatistics iterates index keys using a server-supplied key count without bounding it to the fixed storage. Bytea octal decoding reads up to two bytes past the end of a truncated escape sequence. A server not trusted under current SSL settings can crash the client or make it read out of bounds. Versions before psqlODBC 18.00.0006 are affected.

For an updatable cursor over a table without OIDs that has an auto-increment int4 column, psqlODBC places that column's name, as reported by the server, into a printf-style format string. It later expands that string with one argument while it builds the WHERE clause of its positioned SELECT, UPDATE, and DELETE statements. A column name that contains format directives makes the driver read unrelated memory, which can crash the application or copy client memory into query text sent to the server. The name can come from a server not trusted under current SSL settings, or from any database user who can create a table that the application reads with an updatable cursor. Versions before psqlODBC 18.00.0006 are affected.

psqlODBC sizes its row cache from the column count of a cursor's first result, but recomputes the column count from each later FETCH reply without checking that it matches. A server not trusted under current SSL settings can answer a later FETCH with more columns and make the driver write tuple records past the end of the heap allocation. A keyset reply with fewer columns than the key fields also underflows the unsigned field count. This is reachable with UseDeclareFetch enabled. Versions before psqlODBC 18.00.0006 are affected.

psqlODBC copies the reply to a large-object read (loread) into the application's buffer using the length sent by the server, without comparing it to the buffer size. A server not trusted under current SSL settings, or a man-in-the-middle, can return more bytes than the buffer holds and overwrite adjacent memory in the client process with server-chosen bytes. Fetching any large-object column is sufficient to trigger it. Versions before psqlODBC 18.00.0006 are affected.

Stored cross-site scripting (XSS) in the HTML report generator in the aidlc-traceability CLI in AWS AI-DLC Workflows before 2026-07-03 might allow an actor who can modify a project's AI-DLC artifact files to execute arbitrary script in the browser of a user who opens a generated HTML traceability report via crafted artifact titles or descriptions. To remediate this issue, users should remove any local copy of the aidlc-traceability CLI obtained before 2026-07-03, or update it to commit 96cf0ea or later. The CLI was removed from the repository on 2026-07-16.

Attendo (formerly Petpooja Payroll) Admin Android application version 3.0.13 and earlier contains a hard-coded credentials vulnerability that allows attackers to extract production MQTT broker credentials by decompiling the APK's bundled Flutter configuration assets such as production.json and device-command configuration files. Attackers can use the recovered credentials to authenticate to the production MQTT broker over TCP 1883, access messaging topics containing employee attendance records, personal identifiers, Base64-encoded photographs, device telemetry, and authentication material, and publish messages to those topics, resulting in exposure of sensitive data and potential unauthorized message injection.

Joomla Extension - joomlaboat.com - Unauthenticated arbitrary file upload in Custom Tables < 3.7.8 - An unauthenticated attacker can upload a polyglot file with an attacker-controlled extension to the web-accessible tmp directory of any Joomla site that exposes a public Custom Tables submission form, and the stored path is disclosed in the response. By default the uploaded file is served as static content, so the primary impact is an unauthenticated arbitrary file write to a web-accessible directory, which allows hosting of attacker-controlled content (for example phishing or malware pages) under the victim's domain.

Munshi (by Petpooja) before 2026-07-13 contained a stored cross-site scripting vulnerability that allows authenticated users to inject malicious markup through item descriptions or messages submitted in the Purchase Inquiry thread (/purchase-inquiry/thread/{thread-id}), which is assigned to innerHTML during rendering before text cleaning completes. Attackers can store a payload that executes in the browser of another user who opens the affected thread, achieving JavaScript execution and full-page defacement that re-triggers whenever the thread is reopened.

Munshi (by Petpooja) before 2026-06-17 contained a broken object level authorization vulnerability that allows authenticated attackers to access another user's purchase inquiry by manipulating the thread identifier in the GET /purchase-inquiry/thread/{thread-id} endpoint. Attackers can modify the thread ID to read supplier communication and quotation-related content belonging to other users, resulting in unauthorized disclosure of sensitive business information outside their access scope.

Munshi (by Petpooja) before 2026-07-07 contained a missing authorization vulnerability in its company deletion workflow at DELETE /api/companies/{company_id}, where the application fails to enforce the dedicated company_delete permission or verify company ownership. An authenticated invited sub-user with mere access to a company can exploit this flaw to perform a hard delete of the company record, causing irreversible loss of associated business data including invoices, configurations, and chat sessions.

Munshi (by Petpooja) before 2026-07-28 contained a broken object level authorization vulnerability that allows authenticated sub-users to delete invoices belonging to the tenant owner by exploiting the DELETE /api/invoice/{invoice_id} and POST /api/invoices/bulk-delete endpoints, which only verify tenant membership without enforcing object-level ownership or permission checks. Attackers can invoke single or bulk deletion requests with arbitrary invoice identifiers to delete invoices they do not own, resulting in unauthorized data loss.

Munshi (by Petpooja) before 2026-06-30 contained a payment bypass vulnerability in its subscription payment webhook at POST /api/payment/order/webhook, which lacks webhook authentication, cryptographic signature verification, source validation, and independent server-side confirmation with the payment provider. A remote, unauthenticated attacker can submit a forged payment-success notification referencing an order obtained from the payment-status redirect URL, causing the backend to mark the subscription as paid/active and granting paid access without an actual successful payment.

OpenKruise kruise-manager exposes an unauthenticated pprof endpoint, enabled by default, opening it on 0.0.0.0:8090, which can enable a DoS attack triggerable by any pod on the cluster network.

Joomla Extension - dj-extensions.com - IDOR vulnerability allowing cross-account ad access in DJ-Classifieds < 3.11.3 - A “duplicate my ad” convenience feature on the new-ad form (site/models/additem.php) loaded an existing advert by its raw numeric ID with no check that the ad belonged to the requesting user, and no check of whether it was published, blocked, or still a pending draft. Any logged-in user, regardless of role, could request the new-ad form with another ad’s ID and have that ad’s full content, including the seller’s phone number, address, email, and price, rendered into the response, even for ads that had never been published. Advert IDs are small sequential integers.

Joomla Extension - dj-extensions.com - Unauthenticated SQL injection in DJ-Classifieds < 3.11.3 - The Joomla extension DJ-Classifieds is vulnerable to an unauthenticated SQL injection vector in the frontend search.

Joomla Extension - dj-extensions.com - Authenticated arbitrary file upload in DJ-Classifieds < 3.11.3 - The Joomla extension DJ-Classifieds is vulnerable to an unprivileged file upload, leading to full RCE.

Grafana Mimir's Alertmanager rejects tenant-supplied receiver configurations that reference local files, but the check did not cover two HTTP client settings: http_headers.<name>.files and basic_auth.username_file. A tenant permitted to upload an Alertmanager configuration could point these fields at any file readable by the Alertmanager process. When a notification was sent, Alertmanager read the file and put its contents in a header of the outgoing request. By directing the receiver at a server they control, the tenant could read files from the Alertmanager host, such as credentials and configuration. Exploitation requires permission to write the tenant's Alertmanager configuration. Grafana Enterprise Metrics (GEM), which is built on Mimir, is also affected. The issue is fixed in Mimir 3.1.7 and 3.2.2 and in GEM 2.17.20. GEM 2.11 through 2.16 will not receive a fix; users on those versions should upgrade to 2.17.20 or later.

A security vulnerability has been detected in AndyFactory newspaper4k up to 0.9.6. This affects the function Article.parse of the file newspaper/extractors/image_extractor.py of the component Image Extractor. Such manipulation leads to server-side request forgery. It is possible to launch the attack remotely. Upgrading to version 0.9.6.1 mitigates this issue. The name of the patch is eec5fab21897c14667c98f5ba1831ebd3a692dd4. Upgrading the affected component is advised. The vendor was contacted early, responded in a very professional manner and quickly released a fixed version of the affected product.

A command injection vulnerability was found in Argo CD Image Updater, where a proxy URL configured on an SSH Git repository is passed unsanitized into a shell-interpreted SSH command. An attacker able to set a malicious proxy value can inject shell metacharacters to execute arbitrary commands and access credentials in the context of the Image Updater process.

FOSSBilling is an open source billing and client-management platform. From 0.1.0 until 0.8.9, the admin/support/note_create endpoint stores the note parameter without sanitization in the support_ticket_note table, and src/modules/Support/templates/admin/mod_support_tickets.html.twig renders each stored support ticket note through innerHTML in the admin ticket list. A staff user with only the support.manage_tickets permission can store HTML or JavaScript that executes when another staff member opens the ticket notes. The script runs with the viewing staff member's session and privileges, can read a CSRF token and accessible client or invoice data, and can perform administrative actions available to the viewer, including a Super Administrator. The ticket detail page is not affected because its Twig rendering uses autoescaping. This issue is fixed in version 0.8.9.

Improper neutralization of special elements used in a template engine in the publishing component of Amazon Virtual Engineering Workbench (VEW) could allow a remote authenticated third party to run arbitrary code with the permissions of the publishing Lambda's IAM role. This could allow access to data and resources across all tenant projects. The code is reachable through a crafted CloudFormation template body containing template engine directives submitted through the product versioning API. To remediate this issue, users should redeploy from the latest version of awslabs/virtual-engineering-workbench.

CrewAI contains an SSRF vulnerability caused by its SSRF guard and HTTP client, which both miss malicious content in URLs by checking opposite content passed to it, enabling any request that injects a malicious URL to perform SSRF, such as with the ScrapWebsiteTool.

The error handling implementation fails to properly sanitize or validate user-supplied input within URL parameters under specific conditions. This allows end-users to manipulate the error messages that are displayed by the application. By exploiting this vulnerability, malicious actors can inject arbitrary content into error messages. This enables them to perform social engineering attacks, such as phishing or credential theft, by replacing legitimate error messages with malicious content, misleading users into compromising sensitive information or taking harmful actions.

The Private Internet Access Extension for Firefox contains an unguarded onAuthRequired authentication handler, leading to leakage of PIA proxy credentials when a user visits any malicious website that issues a Basic auth challenge. This issue was fixed in version 3.2.5, but reintroduced in later versions.

A remotely exploitable denial-of-service vulnerability exists in the TCP options parser of the Nut/Net TCP/IP stack included in Nut/OS, developed by Ethernut/Egnite. Nut/OS releases up to version 5.1 are affected. An unauthenticated attacker can send a single TCP segment containing an unknown TCP option with a length byte of zero to any open TCP port on a vulnerable device. When processed, this malformed option triggers an infinite loop in the function NutTcpInputOptions, permanently blocking the TCP processing thread and rendering all TCP-based services unresponsive until the device is rebooted.

Furl versions through 3.15 for Perl allow memory exhaustion via unbounded decompression of HTTP responses. Furl inflates a response body sent with a gzip, x-gzip or deflate Content-Encoding and appends the output to the response string with no size limit. It does this whether or not the request asked for compression, and no option caps the decompressed size. Responses read with write_file or write_code are streamed rather than held in memory. A malicious server, or an attacker on a plain HTTP connection, can send a small compressed response that expands to exhaust the client's memory and end the process.

kaniko before 1.25.20 can send a Microsoft Entra ID (Azure AD) access token for Azure Resource Manager to an attacker-controlled registry host. kaniko uses the Azure Container Registry credential helper github.com/chrismellard/docker-credential-acr-env, which decides whether a host is an ACR registry with an unanchored regular expression, so any hostname containing .azurecr.io, .azurecr.cn, .azurecr.de or .azurecr.us is accepted. kaniko consults this helper for any registry that has no credentials in the Docker config and is not a Google or Amazon ECR registry, before contacting the registry. A Dockerfile that references an image on a lookalike host, such as FROM x.azurecr.io.attacker.com/img or COPY --from=<image>, causes the helper to obtain a token from the build's Azure identity (service principal environment variables, federated workload identity token, or managed identity) and POST it to https://<host>/oauth2/exchange. The attacker can use the token against Azure Resource Manager with the identity's role assignments until it expires, and can exchange it for credentials to any ACR registry the identity can access. No RUN instruction or push is required. This issue is fixed in 1.25.20.

Missing filtering in the SUSE Rancher Manager meta proxy could cause passed in rancher credentials to be leaked to cloud providers. This affects SUSE Rancher 2.15.0 up to 2.15.3, 2.14.0 up to 2.14.7, 2.13.0 up to 2.13.11,

Improper neutralization of quoting syntax in SQL identifier construction in the Apache Iceberg write component in AWS SDK for pandas (also published as awswrangler) 3.5.2 through 3.17.0 might allow an authenticated remote user with permission to modify AWS Glue Data Catalog metadata to read any table that a separate, more privileged principal can access and write to the table that principal is updating via a crafted column name. To remediate this issue, users should upgrade to version 3.17.1 or later.

A privilege escalation issue was discovered in Rancher Manager's role-based access control. The inheritedNamespacedRules field of a GlobalRole propagates permissions into named namespaces of every downstream cluster managed by the Rancher instance, but it was not subject to the privilege-escalation validation applied to the other permission-carrying fields of a GlobalRole. As a result, a non-administrative user could be granted, through a global role  binding, permissions in downstream clusters that exceed the permissions that user already holds. This affected SUSE Rancher 2.15.2 only, and SUSE Rancher Webhook 0.11.0 up to 0.11.4, and 0.12.0 up to 0.12.1.

In the Linux kernel, the following vulnerability has been resolved: ntfs: fix race between fallocate and mmap reads The fallocate implementation only takes invalidate_lock for punch hole, collapse range, and insert range operations. For standard allocation modes (mode == 0, FALLOC_FL_KEEP_SIZE), the lock is not held. During ntfs_attr_fallocate(), new clusters are mapped to the runlist via ntfs_attr_map_cluster() before being zeroed by ntfs_dio_zero_range(). This creates a window where concurrent mmap page faults can read uninitialized disk data. Since mmap uses filemap_fault() which takes invalidate_lock in shared mode, it can fault in pages during this window and expose old disk contents to userspace. This is an information leak and data integrity issue. Fix by taking invalidate_lock for all fallocate operations, not just for punch/collapse/insert modes. This prevents concurrent page faults from accessing unzeroed clusters during the allocation window.

In the Linux kernel, the following vulnerability has been resolved: ntfs: take invalidate_lock in ntfs_filemap_page_mkwrite() ntfs_filemap_page_mkwrite() calls iomap_page_mkwrite() without holding mapping->invalidate_lock, so a concurrent truncate or fallocate can be in the middle of invalidating pagecache and rewriting the runlist while the write fault maps blocks and dirties the folio. This races with ntfs_attr_fallocate(), which merges clusters into the in-memory runlist, drops the runlist lock, and only afterwards zeroes the newly allocated clusters on disk; and with the punch-hole/insert/collapse paths that free clusters after truncating the cache. Per Documentation/filesystems/locking.rst, ->page_mkwrite() must ensure there are no truncate/invalidate races, "usually mapping->invalidate_lock is suitable for proper serialization". xfs takes its mmaplock (= the invalidate_lock rwsem) shared in exactly this path. Take invalidate_lock shared around iomap_page_mkwrite(). The read-only fault path is already covered because filemap_fault() itself grabs invalidate_lock shared on instantiation/read paths; only page_mkwrite was bypassing it in this driver.

In the Linux kernel, the following vulnerability has been resolved: mm/damon/core: error damos_commit_quota_goal() for zero target_value Patch series "mm/damon: move zero damos quota target_value handling to the core layer". Having zero DAMOS quota target value can cause division by zero. DAMON API callers are checking the target value parameters to avoid that. It is easy to make mistakes in some of the multiple API callers. Move that to the core layer. Patch 1 adds the corner case handling into the core layer DAMON parameters validation logic. Patches 2 and 3 remove no more needed DAMON API callers side handling of the corner case in DAMON_LRU_SORT and DMON_SAMPLIE_MTIER, respectively. This patch (of 3): If a DAMOS scheme has a damos_quota_goal of zero target_value, damos_quota_goal() could trigger division-by-zero error. Hence each DAMON API callers should do the zero target_value validation. It is easy to make mistakes. Actually such bugs in DAMON_LRU_SORT and DAMON_SAMPLE_MTIER were found and fixed [1]. It is better to handle the corner case only once in the core layer, instead of multiple places in all DAMON API callers. One straightforward option is using an alternative denominator for the corner case in the damos_quota_goal(). However, the zero target_value is meaningless. In this case, the quota goal is always evaluated as achieved or over-achieved. The quota will only keep being reduced. Simply avoid using zero target_value by adding a check in the core layer DAMOS quota goal parameters validation/commit path, damos_commit_quota_goal(). Update it to return an error in the case. Also update its caller to propagate the error.

In the Linux kernel, the following vulnerability has been resolved: Bluetooth: L2CAP: refuse __l2cap_chan_add if chan already has conn l2cap_chan may be linked to l2cap_conn at most once. This is assumed in several places, eg l2cap_chan_del cleanup. There is a TOCTOU race where the invariant is violated: [Task 1] [Task 2] l2cap_chan_connect l2cap_sock_bind l2cap_chan_lock lock_sock l2cap_state_change if (sk->sk_state != BT_OPEN) chan->state = BT_CONNECT l2cap_sock_state_change_cb chan->state = BT_BOUND sk->sk_state = BT_BOUND lock_sock <------------------ release_sock sk->sk_state = BT_CONNECT l2cap_sock_connect() does not check sk->sk_state, so since chan->state is now BT_BOUND, subsequent connect() ends up with second __l2cap_chan_add. Explicitly document and check the invariant in __l2cap_chan_add with WARN_ON_ONCE. The only callsite where it could be hit is l2cap_chan_connect, so add pre-check there to avoid relying on chan->state. chan->state read/write is not properly guarded currently so there can be other TOCTOUC problems. Add l2cap_lock_chan in l2cap_sock_bind() to guard chan->state write.

In the Linux kernel, the following vulnerability has been resolved: tracing: Hold trace_types_lock when resetting last boot data from readers update_last_data_if_empty() is called from tracing_open() with O_TRUNC, tracing_read_pipe() and tracing_buffers_read() without trace_types_lock. Two readers can both pass the LAST_BOOT test and free module_delta twice, and a module notifier can update module_delta while it is being freed. Take trace_types_lock in those paths and keep the callers that already hold it on an unlocked variant.

In the Linux kernel, the following vulnerability has been resolved: tracing: Hold trace_types_lock when updating modules in persistent instances trace_module_record() walks ftrace_trace_arrays from the module notifier without trace_types_lock. A concurrent rmdir of an instance removes it from the list and frees it under that lock, so a module load or unload can walk into a freed trace_array.

In the Linux kernel, the following vulnerability has been resolved: tracing: Unregister field variable histograms when the command fails An action that names a field on another event needs that field as a variable, so create_field_var_hist() registers a second hist trigger on that event to provide it, and records it in target_hist_data->field_var_hists[]. If a later part of the same command fails, create_actions() returns an error and event_hist_trigger_parse() jumps to out_free. The write fails with -EINVAL and sched_switch gets no trigger, but: # echo 'hist:keys=pid:ts0=common_timestamp.usecs' >> \ events/sched/sched_waking/trigger # echo 'my_synth u64 lat; int a; int b' >> synthetic_events # echo 'hist:keys=next_pid:wakeup_lat=common_timestamp.usecs-$ts0:\ onmatch(sched.sched_waking).my_synth($wakeup_lat,prio,nosuchfld)'\ >> events/sched/sched_switch/trigger # cat events/sched/sched_waking/trigger hist:keys=pid:vals=hitcount:ts0=common_timestamp.usecs:sort=hitcount:size=2048:clock=global [active] hist:keys=pid:vals=hitcount:synthetic_prio=prio:sort=hitcount:size=2048 [active] The second histogram is the one created for 'prio'. It is still active and still counting. out_free reaches destroy_field_var_hists(), which frees the field_var_hists[] entries, but never unregister_field_var_hists(), which is what removes the triggers and runs only from event_hist_trigger_free(). The entry is freed and the trigger is left with nothing tracking it. Unregister the field variable histograms in out_free as well, and record each one as soon as its trigger is registered rather than after the lookup that follows it, so no exit can leave a registered trigger untracked. The entry then belongs to field_var_hists[] and is released by destroy_field_var_hists(), so freeing it on that exit would be a double free.

In the Linux kernel, the following vulnerability has been resolved: tracing: Save the event file instead of the compatible histogram create_field_var_hist() creates a histogram on the matched event to supply a field variable, and records it on the target so it can be torn down later: hist_data = find_compatible_hist(target_hist_data, file); ... /* Save the compatible histogram information */ var_hist->hist_data = hist_data; hist_data is not the histogram it creates. It is one that was already there, whose keys the new one copies. The saved pointer has a single user, unregister_field_var_hists(), which runs when the target is removed: file = hist_data->field_var_hists[i]->hist_data->event_file; find_compatible_hist() matches on keys alone, so it can pick one with no variables of its own. check_var_refs() then returns false and the user can remove it while the target still points at it: BUG: KASAN: slab-use-after-free in event_hist_trigger_free+0x2b2/0x320 Read of size 8 at addr ffff8880093b90e0 by task init/1 Freed by task 1: kfree+0x154/0x420 event_hist_trigger_free+0x1d5/0x320 event_hist_trigger_parse+0x35f3/0x69e0 trigger_process_regex+0x1a6/0x250 Save the event file instead. It is all the dereference ever wanted, and it does not go away when the user removes a trigger.

In the Linux kernel, the following vulnerability has been resolved: tracing: Reject incomplete event filter expressions A non-empty expression normally finishes in predicate_parse()'s inner loop, which resolves the pending logical operators before jumping to finalization. If the input ends after an operator, possibly followed by unary '!' tokens, the outer loop instead falls through with unresolved operator state. For example, "common_pid == 1 || common_pid == 2 &&" leaves prog[0].target equal to -1. The second pass then reads prog[-1] before the forward-jump check can reject it. Depending on adjacent slab contents, it can also copy an out-of-bounds word into a branch target that passes the forward-only check. A self-targeted perf tracepoint counting event can reach this path through PERF_EVENT_IOC_SET_FILTER. Reject the outer-loop fallthrough when a predicate has already been parsed. This ensures that non-empty expressions only reach finalization after the inner loop has resolved their operators. Keep the existing no-filter error for input that has no predicates. Add a regression test for trailing logical operators and unary negations.