Alexander Potapenko

dashboard/app: support testing C reproducers via handleTestReproCRequest Adds support for testing C reproducers on target kernel managers. Extends the `Job` entity with `CandidateReproC` to store C reproducer payloads, updates job creation and polling to transmit candidate repros, and updates bug status upon test completion. Implements deduplication in `handleTestReproCRequest` and adds `TestReproCJob` unit tests.

syz-manager, pkg/vminfo: auto-detect CPU vendor and nested virt for boot tests Introduce the vminfo.Capabilities structure and automatically detect CPU vendor ('intel' or 'amd') from /proc/cpuinfo and nested KVM virtualization status from /sys/module/kvm_*/parameters/nested during machine check. Pass the detected *vminfo.Capabilities through rpcserver.Manager.MachineChecked to syz-manager and runtest.Context so that boot tests can filter on environment capabilities via '# requires:' directives.

dashboard: record C/Syz reproducer timestamps and add migration tool Currently syzbot does not record the timestamp when a reproducer is found for a bug. Add FirstCReproTime and FirstSyzReproTime fields to the Bug entity, and update UpdateReproLevel to record these timestamps when reproducers arrive. Add a populateReproTime admin tool (/admin?action=populateReproTime) to populate historical reproducer timestamps based on the earliest crash associated with a reproducer. Add unit tests covering UpdateReproLevel and populateReproTime.

dashboard/app: throttle automatic repro-c jobs Throttle automatic creation of repro-c jobs to at most 5 jobs per 6 hours per namespace. Filter out repro-c from requested workflows in autoCreateAIJobs when the quota is reached.

pkg/aflow/action/actionsyzlang: use VM arch for kernel in execute-corpus execute-corpus did not consume TargetVMArch, so for 386 targets it built a linux/386/386 manager config (failing to find syz-execprog in bin/linux_386). It also used the 386 target for kernel-side work: symbolization of coverage PCs, the vmlinux path and PreviousInstructionPCs, although the kernel is amd64. Embed crash.Target into ExecuteCorpusArgs and pass it to TargetConfig. Use the target for the VM arch for all kernel-side work, and keep the TargetArch target only for program deserialization. Build TargetConfig once, and validate the VM arch upfront to return an error instead of panicking in build.LinuxKernelImage() on unknown arches.

dashboard/app: restrict automatic AI jobs to supported architectures Previously, supported TargetArch values were only listed on the manual workflow form in manualAIWorkflows(). Automatic AI job creation via pendingWorkflowsForBug() did not check the bug's crash build architecture, causing jobs (such as repro-c, moderation, and assessment-security) to be auto-created for unsupported architectures like riscv64 and s390. Those jobs immediately failed on syz-agent and consumed repro-c rate-limit quota slots. Extract aiSupportedArches (including arm alongside amd64, arm64, and 386) and filter out bugs on unsupported target architectures in pendingWorkflowsForBug().

sys/linux: describe device mapper control interface Describe the /dev/mapper/control ioctls (drivers/md/dm-ioctl.c) with typed table load payloads for the dm targets (linear, striped, zero, error, crypt, flakey, snapshot*, mirror, raid, thin-pool, thin, cache, writecache, clone, verity, integrity, multipath) and target messages (dm-stats, thin-pool, raid, mpath, cache, writecache, clone, crypt). Underlying devices are limited to per-proc brd ramdisks and rnull devices (a thin target's pool is a mknod'ed dm device). Since /dev/dm-N nodes only appear after a device is created, the dev number returned in struct dm_ioctl is used as a resource to mknod and open the mapped device. Previously only an auto-generated raw ioctl existed, and syzbot coverage of drivers/md/dm-* was ~4.5%.

dashboard/app: automatically create repro-c jobs for eligible bugs Allow automatic creation of repro-c AI jobs for open bugs without a C reproducer that have a crash report, have been open for more than 48 hours, and had their last crash within the last 30 days, omitting bugs whose titles start with "INFO:" and bugs that have recent patch candidates in BugDiscussionInfo. Restrict automatic creation to namespaces with AI.AutoReproC enabled. Cap automatic retries on error to at most 2 attempts, while keeping manual creation unconstrained.

dashboard/app: use NewSpannerCtx in tests polling for jobs Switch tests that poll for manager jobs from NewCtx to NewSpannerCtx in preparation for job polling querying Spanner. Also update urlRe in tree_test.go to match hyphens in domain names, and update test assertions in TestTreeOriginDownstreamEmail and TestBisectCause to use dynamic AppID.

dashboard/app: do not abort getNextJob on first handler error If a job handler encounters an error, log it and continue trying the remaining handlers rather than aborting immediately. This ensures that independent job generators (such as bisection and patch testing) are not starved if one handler fails. Return the last encountered error if no handler produced a job.