Aleksandr Nogikh

This commit introduces a **performance optimization** to the **`prog` package's memory allocation subsystem**, specifically within the `alloc` function in `prog/alloc.go`. It **refactors** the allocation logic to track the last known free position per alignment, thereby avoiding costly re-traversals of large memory arrays. This change significantly improves the efficiency of allocating program arguments, addressing a bottleneck identified in `-race` benchmarks where memory allocation was a major time consumer. By leveraging the system's "never free" memory model, this optimization streamlines memory management and enhances overall performance for the `prog` subsystem.

This commit introduces a **new capability** to restrict **AI clients** to specific namespaces for job processing, addressing the inconvenience of managing unique client names for each namespace in multi-tenant `syz-agent` deployments. It modifies the **API client configuration** in `dashboard/app/config.go` by adding an `AIJobNamespaces` field, and updates the **AI job handling logic** in `dashboard/app/ai.go` to enforce these restrictions during job polling, completion, and trajectory logging via a new `checkAiJobAccess` function. Furthermore, database retrieval functions in `dashboard/app/aidb/crud.go` are adjusted to filter jobs by namespace, ensuring that `syz-agents` can now efficiently serve multiple namespaces concurrently. This enhancement simplifies the management of AI jobs across different namespaces. Comprehensive tests in `dashboard/app/ai_test.go` verify this new namespace-based access control.

This commit introduces a new, dedicated **Kubernetes service account**, `kernel-fetcher-ksa`, within the **`syz-cluster`** module to enhance security. This **security enhancement** refines permissions for the **kernel fetching workflow** by assigning the newly defined `kernel-fetcher-ksa` to the `process-tree` step in the `fetch-kernels-template.yaml`. Previously, the broader `argo-executor-ksa` was used for this operation, but this change ensures that kernel fetching now operates with more granular, least-privilege permissions. This **improves the overall security posture** by isolating the permissions required for kernel fetching from the general workflow execution.

This commit introduces a **test performance optimization** for `TestChooseProgram` within the **`pkg/corpus` module**. It achieves this by **restricting the set of enabled syscalls** used during program generation. Specifically, the `generateRangedInput` helper function in `corpus_test.go` is modified to reduce the diversity of generated programs. This change **accelerates test execution** by simplifying the input, as extensive program diversity is not critical for the correctness of this particular test.

This commit introduces a **new test case** to validate the system's handling of **pointer recursion**, specifically preventing stack overflows. It adds `TestPtrRecursion` to the **`prog` module's testing suite** (`prog/prog_test.go`) and defines new types (`test_list_entry`, `test_recur_struct`) along with a `test$recur_ptr` syscall in `sys/test/test.txt`. This **new test** acts as a crucial regression check, ensuring the system can correctly process complex, recursive data structures. The addition verifies that previous fixes effectively mitigate stack overflow issues when dealing with such recursive pointers, enhancing the overall stability of the **system call interface**.

This commit **enhances the program generation pruning logic** within the `prog` module, specifically in `prog/rand.go`. It introduces a new `leafType` helper function and `pruneRecursion` limits to accurately determine the base type of complex data structures by following pointers and arrays. This **feature improvement** also implements **pruning for nested pointers**, which prevents "object explosion" and ensures more efficient and manageable program generation by avoiding excessively deep or resource-intensive data structures. The changes to the `generate` function will result in **more robust and controlled program generation**.

This commit **refactors** the **recursion pruning logic** within the **`prog`** package, specifically in `prog/rand.go`. It transitions from using less reliable string names to more robust and efficient **Type object pointers** as keys for tracking recursion depth. This **refactoring** improves the reliability and accuracy of the `pruneRecursion` mechanism, particularly for generic types like pointers where string names are insufficient. The change ensures more consistent and correct program generation by providing a more reliable identifier for type-based recursion.

This commit introduces a **feature enhancement** to the **dashboard application's bug details page**, refining the display of **AI job history**. It implements a new filtering logic, encapsulated in the `compactAIJobs` function, which ensures that only **finished AI jobs** are shown, alongside the **most recent aborted job for each workflow type**. This modification, applied within the `handleBug` function, **reduces visual clutter** and improves the user experience by presenting a more concise and relevant overview of AI workflow outcomes directly on the bug page.

This commit implements **temporary workarounds** within the `prog` package's **testing infrastructure** to mitigate frequent timeouts. It modifies tests like `TestSerializeDeserializeRandom`, `TestMutateRandom`, `testCrossTarget`, and `TestDeterminism` to **skip large generated programs**, which were causing quadratic time processing. This **maintenance fix** addresses performance regressions observed after adding FreeBSD descriptions, preventing numerous syzkaller test errors on syzbot until the program generation and mutation logic can be permanently improved.

This commit introduces a **panic recovery mechanism** within the **`syz-agent`** to enhance its robustness during job processing. Specifically, the `executeJob` function in `syz-agent/agent.go` is updated to gracefully handle panics. This **bug fix** prevents the agent from crashing and entering a restart loop when encountering a problematic job. Instead, the agent will now report the job failure to the dashboard, ensuring continuous operation and proper error reporting for the **`syz-agent`** subsystem.

This commit **fixes a panic** that previously occurred in the `prog` package when `AUTO` was specified for `CsumType` arguments. Although `AUTO` was accepted during initial parsing, the `fixupAutos` function would panic during the subsequent value resolution. The change introduces specific handling for `CsumType` within `fixupAutos` to correctly process `AUTO` values, preventing the crash and improving the **robustness of program encoding/decoding**. A new test case in `prog/encoding_test.go` verifies that deserializing `CsumType` with `AUTO` now functions as expected, ensuring the stability of program generation when using automatic checksum types.

This commit significantly **improves the manual reproduction request handling** workflow by introducing a new **API endpoint `ReproTaskDone`** in the **dashboard**. This allows `syz-managers` to explicitly notify the dashboard upon completion of a reproduction task, enabling the dashboard to **stop retrying successful requests** and manage task states more efficiently. Furthermore, the system now **records and displays the reproduction log** for each task, providing users with direct access to the results via the `manager.html` UI. This **feature enhancement** streamlines the interaction between the dashboard and `syz-manager`, reducing redundant processing and offering better transparency into the reproduction process.

This commit **enhances the build system's error parsing** within the `pkg/build` module. It introduces new logic to **match global compiler errors** that occur without a specific source file being mentioned, addressing cases where compilation fails due to broader issues. By treating generic `error: ` messages as "weak" matches, this **bug fix** improves the robustness of error reporting, ensuring all types of compilation failures are properly recognized and handled.

This commit refines the **build error reporting logic** within the `pkg/build` module. It introduces a preference for **strong build reasons**, ensuring that if a definitive error is identified, any less critical "weak" reasons are suppressed from the output. This **enhancement** improves the clarity and relevance of build error messages, helping developers more quickly identify and address the most significant issues. The change streamlines error feedback by prioritizing actionable information.

This commit introduces a **new feature** to the **dashboard application**, enabling users to view the last five reproducer tasks sent to a specific manager directly on its dedicated page. It involves updating the `ReproTask` data structure with a `Created` timestamp and adding a corresponding datastore index for efficient time-based retrieval. The backend logic in `main.go` and `api.go` was modified to load and process these tasks, which are then rendered in a new table section on the `manager.html` template. This enhancement to the **manager page** provides immediate visibility into recent reproducer activity, improving monitoring capabilities.

This commit **enhances error reporting** within the `pkg/aflow` module, specifically for **kernel build failures**. It **improves the kernel build process** by ensuring that detailed build error reports, including the root cause, are extracted and included in the error message when a failure occurs. This **bug fix** addresses a previous limitation where diagnosing kernel build issues was difficult due to a lack of underlying error context. The change, primarily affecting the `BuildKernel` function in `pkg/aflow/action/kernel/build.go`, significantly **improves the debuggability and reliability of kernel compilation** within the `aflow` system.

This commit **exports** the `ExtractRootCause` function within the **`pkg/build`** package. This **API change** makes the utility for extracting the root cause of build-related errors, previously internal, publicly available. Downstream consumers can now directly leverage `ExtractRootCause` for improved error analysis and reporting, enhancing the diagnostic capabilities of the build system.

This commit **improves bug categorization** within the **`pkg/report` module** by filtering out generic stack frames from **Linux kernel crash reports**. Previously, the common `mark_buffer_dirty` frame led to unrelated bugs from different filesystems being incorrectly grouped together, hindering effective triaging. This **bug fix** adds `mark_buffer_dirty` to the list of ignored stack lines in `pkg/report/linux.go`, ensuring more precise deduplication of kernel issues. The change enhances the overall accuracy of syzkaller's reporting and is validated by a new test case.

This commit **refactors** the **AI Jobs dashboard** to significantly improve data retrieval efficiency. It **moves workflow type and aborted status filtering** from in-memory Go application logic, specifically within `handleAIJobsPage`, directly into the SQL database query. The `LoadNamespaceJobs` function in `dashboard/app/aidb/crud.go` now accepts a `JobFilter` struct, enabling the database to perform filtering, which results in **more efficient data loading** and reduced application-side processing for the dashboard.

This commit introduces a **new feature** to manage and display **aborted AI jobs** within the dashboard. It extends the `Job` entity with an `Aborted` boolean field, accompanied by necessary database migrations, and automatically **marks stale jobs as aborted** if their failure is attributed to an agent restart. The **AI job management** module now supports filtering based on this status, and the **dashboard UI** for AI jobs (`ai_jobs.html`) gains a new control, allowing users to easily filter out these interrupted jobs for improved clarity and job tracking.