From de4da7bd5c5172f6362b5994f541d830119840dc Mon Sep 17 00:00:00 2001 From: Thomas Huth Date: Thu, 21 Aug 2025 17:23:09 +0200 Subject: KVM: s390: Fix access to unavailable adapter indicator pages during postcopy When you run a KVM guest with vhost-net and migrate that guest to another host, and you immediately enable postcopy after starting the migration, there is a big chance that the network connection of the guest won't work anymore on the destination side after the migration. With a debug kernel v6.16.0, there is also a call trace that looks like this: FAULT_FLAG_ALLOW_RETRY missing 881 CPU: 6 UID: 0 PID: 549 Comm: kworker/6:2 Kdump: loaded Not tainted 6.16.0 #56 NONE Hardware name: IBM 3931 LA1 400 (LPAR) Workqueue: events irqfd_inject [kvm] Call Trace: [<00003173cbecc634>] dump_stack_lvl+0x104/0x168 [<00003173cca69588>] handle_userfault+0xde8/0x1310 [<00003173cc756f0c>] handle_pte_fault+0x4fc/0x760 [<00003173cc759212>] __handle_mm_fault+0x452/0xa00 [<00003173cc7599ba>] handle_mm_fault+0x1fa/0x6a0 [<00003173cc73409a>] __get_user_pages+0x4aa/0xba0 [<00003173cc7349e8>] get_user_pages_remote+0x258/0x770 [<000031734be6f052>] get_map_page+0xe2/0x190 [kvm] [<000031734be6f910>] adapter_indicators_set+0x50/0x4a0 [kvm] [<000031734be7f674>] set_adapter_int+0xc4/0x170 [kvm] [<000031734be2f268>] kvm_set_irq+0x228/0x3f0 [kvm] [<000031734be27000>] irqfd_inject+0xd0/0x150 [kvm] [<00003173cc00c9ec>] process_one_work+0x87c/0x1490 [<00003173cc00dda6>] worker_thread+0x7a6/0x1010 [<00003173cc02dc36>] kthread+0x3b6/0x710 [<00003173cbed2f0c>] __ret_from_fork+0xdc/0x7f0 [<00003173cdd737ca>] ret_from_fork+0xa/0x30 3 locks held by kworker/6:2/549: #0: 00000000800bc958 ((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x7ee/0x1490 #1: 000030f3d527fbd0 ((work_completion)(&irqfd->inject)){+.+.}-{0:0}, at: process_one_work+0x81c/0x1490 #2: 00000000f99862b0 (&mm->mmap_lock){++++}-{3:3}, at: get_map_page+0xa8/0x190 [kvm] The "FAULT_FLAG_ALLOW_RETRY missing" indicates that handle_userfaultfd() saw a page fault request without ALLOW_RETRY flag set, hence userfaultfd cannot remotely resolve it (because the caller was asking for an immediate resolution, aka, FAULT_FLAG_NOWAIT, while remote faults can take time). With that, get_map_page() failed and the irq was lost. We should not be strictly in an atomic environment here and the worker should be sleepable (the call is done during an ioctl from userspace), so we can allow adapter_indicators_set() to just sleep waiting for the remote fault instead. Link: https://issues.redhat.com/browse/RHEL-42486 Signed-off-by: Peter Xu [thuth: Assembled patch description and fixed some cosmetical issues] Signed-off-by: Thomas Huth Reviewed-by: Claudio Imbrenda Acked-by: Janosch Frank Fixes: f65470661f36 ("KVM: s390/interrupt: do not pin adapter interrupt pages") [frankja: Added fixes tag] Signed-off-by: Janosch Frank --- arch/s390/kvm/interrupt.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c index 2a92a8b9e4c2..9384572ffa7b 100644 --- a/arch/s390/kvm/interrupt.c +++ b/arch/s390/kvm/interrupt.c @@ -2778,12 +2778,19 @@ static unsigned long get_ind_bit(__u64 addr, unsigned long bit_nr, bool swap) static struct page *get_map_page(struct kvm *kvm, u64 uaddr) { + struct mm_struct *mm = kvm->mm; struct page *page = NULL; + int locked = 1; + + if (mmget_not_zero(mm)) { + mmap_read_lock(mm); + get_user_pages_remote(mm, uaddr, 1, FOLL_WRITE, + &page, &locked); + if (locked) + mmap_read_unlock(mm); + mmput(mm); + } - mmap_read_lock(kvm->mm); - get_user_pages_remote(kvm->mm, uaddr, 1, FOLL_WRITE, - &page, NULL); - mmap_read_unlock(kvm->mm); return page; } -- cgit v1.2.3 From 185d903064f8680c2de43514c94fddc0d75ed00a Mon Sep 17 00:00:00 2001 From: Claudio Imbrenda Date: Mon, 25 Aug 2025 17:18:30 +0200 Subject: KVM: s390: Fix incorrect usage of mmu_notifier_register() If mmu_notifier_register() fails, for example because a signal was pending, the mmu_notifier will not be registered. But when the VM gets destroyed, it will get unregistered anyway and that will cause one extra mmdrop(), which will eventually cause the mm of the process to be freed too early, and cause a use-after free. This bug happens rarely, and only when secure guests are involved. The solution is to check the return value of mmu_notifier_register() and return it to the caller (ultimately it will be propagated all the way to userspace). In case of -EINTR, userspace will try again. Fixes: ca2fd0609b5d ("KVM: s390: pv: add mmu_notifier") Signed-off-by: Claudio Imbrenda Reviewed-by: Christian Borntraeger Reviewed-by: David Hildenbrand Reviewed-by: Steffen Eiden Reviewed-by: Christoph Schlameuss Signed-off-by: Janosch Frank --- arch/s390/kvm/pv.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) diff --git a/arch/s390/kvm/pv.c b/arch/s390/kvm/pv.c index 25ede8354514..6ba5a0305e25 100644 --- a/arch/s390/kvm/pv.c +++ b/arch/s390/kvm/pv.c @@ -624,6 +624,17 @@ int kvm_s390_pv_init_vm(struct kvm *kvm, u16 *rc, u16 *rrc) int cc, ret; u16 dummy; + /* Add the notifier only once. No races because we hold kvm->lock */ + if (kvm->arch.pv.mmu_notifier.ops != &kvm_s390_pv_mmu_notifier_ops) { + /* The notifier will be unregistered when the VM is destroyed */ + kvm->arch.pv.mmu_notifier.ops = &kvm_s390_pv_mmu_notifier_ops; + ret = mmu_notifier_register(&kvm->arch.pv.mmu_notifier, kvm->mm); + if (ret) { + kvm->arch.pv.mmu_notifier.ops = NULL; + return ret; + } + } + ret = kvm_s390_pv_alloc_vm(kvm); if (ret) return ret; @@ -659,11 +670,6 @@ int kvm_s390_pv_init_vm(struct kvm *kvm, u16 *rc, u16 *rrc) return -EIO; } kvm->arch.gmap->guest_handle = uvcb.guest_handle; - /* Add the notifier only once. No races because we hold kvm->lock */ - if (kvm->arch.pv.mmu_notifier.ops != &kvm_s390_pv_mmu_notifier_ops) { - kvm->arch.pv.mmu_notifier.ops = &kvm_s390_pv_mmu_notifier_ops; - mmu_notifier_register(&kvm->arch.pv.mmu_notifier, kvm->mm); - } return 0; } -- cgit v1.2.3 From 5f9df945d4e862979b50e4ecaba3dc81fb06e8ed Mon Sep 17 00:00:00 2001 From: Claudio Imbrenda Date: Mon, 25 Aug 2025 17:18:31 +0200 Subject: KVM: s390: Fix FOLL_*/FAULT_FLAG_* confusion Pass the right type of flag to vcpu_dat_fault_handler(); it expects a FOLL_* flag (in particular FOLL_WRITE), but FAULT_FLAG_WRITE is passed instead. This still works because they happen to have the same integer value, but it's a mistake, thus the fix. Signed-off-by: Claudio Imbrenda Fixes: 05066cafa925 ("s390/mm/fault: Handle guest-related program interrupts in KVM") Acked-by: Christian Borntraeger Reviewed-by: David Hildenbrand Reviewed-by: Steffen Eiden Reviewed-by: Christoph Schlameuss Signed-off-by: Janosch Frank --- arch/s390/kvm/kvm-s390.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-) diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c index bf6fa8b9ca73..6d51aa5f66be 100644 --- a/arch/s390/kvm/kvm-s390.c +++ b/arch/s390/kvm/kvm-s390.c @@ -4864,12 +4864,12 @@ static void kvm_s390_assert_primary_as(struct kvm_vcpu *vcpu) * @vcpu: the vCPU whose gmap is to be fixed up * @gfn: the guest frame number used for memslots (including fake memslots) * @gaddr: the gmap address, does not have to match @gfn for ucontrol gmaps - * @flags: FOLL_* flags + * @foll: FOLL_* flags * * Return: 0 on success, < 0 in case of error. * Context: The mm lock must not be held before calling. May sleep. */ -int __kvm_s390_handle_dat_fault(struct kvm_vcpu *vcpu, gfn_t gfn, gpa_t gaddr, unsigned int flags) +int __kvm_s390_handle_dat_fault(struct kvm_vcpu *vcpu, gfn_t gfn, gpa_t gaddr, unsigned int foll) { struct kvm_memory_slot *slot; unsigned int fault_flags; @@ -4883,13 +4883,13 @@ int __kvm_s390_handle_dat_fault(struct kvm_vcpu *vcpu, gfn_t gfn, gpa_t gaddr, u if (!slot || slot->flags & KVM_MEMSLOT_INVALID) return vcpu_post_run_addressing_exception(vcpu); - fault_flags = flags & FOLL_WRITE ? FAULT_FLAG_WRITE : 0; + fault_flags = foll & FOLL_WRITE ? FAULT_FLAG_WRITE : 0; if (vcpu->arch.gmap->pfault_enabled) - flags |= FOLL_NOWAIT; + foll |= FOLL_NOWAIT; vmaddr = __gfn_to_hva_memslot(slot, gfn); try_again: - pfn = __kvm_faultin_pfn(slot, gfn, flags, &writable, &page); + pfn = __kvm_faultin_pfn(slot, gfn, foll, &writable, &page); /* Access outside memory, inject addressing exception */ if (is_noslot_pfn(pfn)) @@ -4905,7 +4905,7 @@ try_again: return 0; vcpu->stat.pfault_sync++; /* Could not setup async pfault, try again synchronously */ - flags &= ~FOLL_NOWAIT; + foll &= ~FOLL_NOWAIT; goto try_again; } /* Any other error */ @@ -4925,7 +4925,7 @@ try_again: return rc; } -static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, unsigned long gaddr, unsigned int flags) +static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, unsigned long gaddr, unsigned int foll) { unsigned long gaddr_tmp; gfn_t gfn; @@ -4950,18 +4950,18 @@ static int vcpu_dat_fault_handler(struct kvm_vcpu *vcpu, unsigned long gaddr, un } gfn = gpa_to_gfn(gaddr_tmp); } - return __kvm_s390_handle_dat_fault(vcpu, gfn, gaddr, flags); + return __kvm_s390_handle_dat_fault(vcpu, gfn, gaddr, foll); } static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu) { - unsigned int flags = 0; + unsigned int foll = 0; unsigned long gaddr; int rc; gaddr = current->thread.gmap_teid.addr * PAGE_SIZE; if (kvm_s390_cur_gmap_fault_is_write()) - flags = FAULT_FLAG_WRITE; + foll = FOLL_WRITE; switch (current->thread.gmap_int_code & PGM_INT_CODE_MASK) { case 0: @@ -5003,7 +5003,7 @@ static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu) send_sig(SIGSEGV, current, 0); if (rc != -ENXIO) break; - flags = FAULT_FLAG_WRITE; + foll = FOLL_WRITE; fallthrough; case PGM_PROTECTION: case PGM_SEGMENT_TRANSLATION: @@ -5013,7 +5013,7 @@ static int vcpu_post_run_handle_fault(struct kvm_vcpu *vcpu) case PGM_REGION_SECOND_TRANS: case PGM_REGION_THIRD_TRANS: kvm_s390_assert_primary_as(vcpu); - return vcpu_dat_fault_handler(vcpu, gaddr, flags); + return vcpu_dat_fault_handler(vcpu, gaddr, foll); default: KVM_BUG(1, vcpu->kvm, "Unexpected program interrupt 0x%x, TEID 0x%016lx", current->thread.gmap_int_code, current->thread.gmap_teid.val); -- cgit v1.2.3