Slight improvements for OOM/Futex
Subsystems (12)
cover Slight improvements for OOM/Futex
The following 3 patches were developed while working on the OOM/Futex fix.
Patch 1: This is a nonfunctional change. The vma_is_anonymous function originally led to some confusion about what the oom reaper is checking for in __oom_reap_task_mm. This patch makes that function name more verbose.
Patch 2: Futex cleanup was designed with silent failures. Printing this failure would have led to more quickly finding the issue. This introduces a pr_info if the exit path has any issues walking the userspace list.
Patch 3: During the debug process I noticed that the oom_reap_task_mm function was always running twice for the same mm_struct; Once in the exit path and once in the oom_reaper. By checking the MMF_OOM_SKIP we can prevent these unnecissary double calls. I'd like some input from David Rientjes here with regards to CVE-2018-1000200, I want to make sure we are not reintroducing that CVE.
Nico Pache (3):
mm: change vma_is_anonymous to vma_is_private_anon
futex: exit: Print a warning when futex_cleanup fails
exit: Check for MMF_OOM_SKIP in exit_mmap
arch/powerpc/mm/book3s64/pgtable.c | 2 +- fs/userfaultfd.c | 2 +- include/linux/huge_mm.h | 2 +- include/linux/mm.h | 2 +- kernel/futex/core.c | 44 ++++++++++++++++++------------ mm/gup.c | 4 +-- mm/huge_memory.c | 10 +++---- mm/madvise.c | 4 +-- mm/memory.c | 10 +++---- mm/migrate_device.c | 6 ++-- mm/mincore.c | 2 +- mm/mmap.c | 7 +++-- mm/oom_kill.c | 2 +- mm/userfaultfd.c | 8 +++--- 14 files changed, 57 insertions(+), 48 deletions(-)
Signature · 1 line
2.35.1
RFC 1/3 mm: change vma_is_anonymous to vma_is_private_anon
The vma_is_anonymous function isn't fully indicative of what it checks.
Without having full knowledge of the mmap process, one may incorrectly assume this covers all types of anonymous memory; which is not the case.
No functional changes.
--- arch/powerpc/mm/book3s64/pgtable.c | 2 +- fs/userfaultfd.c | 2 +- include/linux/huge_mm.h | 2 +- include/linux/mm.h | 2 +- mm/gup.c | 4 ++-- mm/huge_memory.c | 10 +++++----- mm/madvise.c | 4 ++-- mm/memory.c | 10 +++++----- mm/migrate_device.c | 6 +++--- mm/mincore.c | 2 +- mm/mmap.c | 4 ++-- mm/oom_kill.c | 2 +- mm/userfaultfd.c | 8 ++++---- 13 files changed, 29 insertions(+), 29 deletions(-)
arch/powerpc/mm/book3s64/pgtable.c +1 -1
@@ -479,7 +479,7 @@ int pmd_move_must_withdraw(struct spinlock *new_pmd_ptl, struct vm_area_struct *vma) {if (radix_enabled ()) -return (new_pmd_ptl != old_pmd_ptl) &&vma_is_anonymous (vma); +return (new_pmd_ptl != old_pmd_ptl) &&vma_is_private_anon (vma);return true; }
fs/userfaultfd.c +1 -1
@@ -1269,7 +1269,7 @@ static inline bool vma_can_userfault(struct vm_area_struct *vma,return false; } -return vma_is_anonymous (vma) ||is_vm_hugetlb_page (vma) || +return vma_is_private_anon (vma) ||is_vm_hugetlb_page (vma) ||vma_is_shmem (vma); }
include/linux/huge_mm.h +1 -1
@@ -119,7 +119,7 @@ static inline bool transhuge_vma_suitable(struct vm_area_struct *vma, unsigned long haddr) {/* Don't have to check pgoff for anonymous vma */ -if (!vma_is_anonymous (vma)) { +if (!vma_is_private_anon (vma)) {if (!IS_ALIGNED ((vma->vm_start >>PAGE_SHIFT ) - vma->vm_pgoff,HPAGE_PMD_NR ))return false;
include/linux/mm.h +1 -1
@@ -623,7 +623,7 @@ static inline void vma_set_anonymous(struct vm_area_struct *vma) vma->vm_ops =NULL ; } -static inline bool vma_is_anonymous (struct vm_area_struct *vma) +static inline bool vma_is_private_anon (struct vm_area_struct *vma) {return !vma->vm_ops; }
mm/gup.c +2 -2
@@ -398,7 +398,7 @@ static struct page *no_page_table(struct vm_area_struct *vma,* be zero-filled if handle_mm_fault() actually did handle it. */ if ((flags &FOLL_DUMP ) && - (vma_is_anonymous (vma) || !vma->vm_ops->fault)) + (vma_is_private_anon (vma) || !vma->vm_ops->fault))return ERR_PTR (-EFAULT );return NULL ; }
@@ -913,7 +913,7 @@ static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)if (vm_flags & (VM_IO |VM_PFNMAP ))return -EFAULT ; -if (gup_flags &FOLL_ANON && !vma_is_anonymous (vma)) +if (gup_flags &FOLL_ANON && !vma_is_private_anon (vma))return -EFAULT ;if ((gup_flags &FOLL_LONGTERM ) &&vma_is_fsdax (vma))
mm/huge_memory.c +5 -5
@@ -82,7 +82,7 @@ bool transparent_hugepage_active(struct vm_area_struct *vma)if (!transhuge_vma_suitable (vma, addr))return false; -if (vma_is_anonymous (vma)) +if (vma_is_private_anon (vma))return __transparent_hugepage_enabled (vma);if (vma_is_shmem (vma))return shmem_huge_enabled (vma);
@@ -1035,7 +1035,7 @@ int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,int ret = -ENOMEM ;/* Skip if can be re-fill on fault */ -if (!vma_is_anonymous (dst_vma)) +if (!vma_is_private_anon (dst_vma))return 0 ; pgtable =pte_alloc_one (dst_mm);
@@ -1622,7 +1622,7 @@ static inline int pmd_move_must_withdraw(spinlock_t *new_pmd_ptl,* * We also don't deposit and withdraw tables for file pages. */ - return (new_pmd_ptl != old_pmd_ptl) && vma_is_anonymous(vma); + return (new_pmd_ptl != old_pmd_ptl) && vma_is_private_anon(vma); }#endif
@@ -1799,7 +1799,7 @@ int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd, } ret =HPAGE_PMD_NR ;set_pmd_at (mm, addr, pmd, entry); -BUG_ON (vma_is_anonymous (vma) && !preserve_write &&pmd_write (entry)); +BUG_ON (vma_is_private_anon (vma) && !preserve_write &&pmd_write (entry)); unlock:spin_unlock (ptl);return ret;
@@ -1957,7 +1957,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,count_vm_event (THP_SPLIT_PMD ); -if (!vma_is_anonymous (vma)) { +if (!vma_is_private_anon (vma)) { old_pmd =pmdp_huge_clear_flush_notify (vma, haddr, pmd);/* * We are going to unmap this huge page. So
mm/madvise.c +2 -2
@@ -543,7 +543,7 @@ static void madvise_pageout_page_range(struct mmu_gather *tlb,static inline bool can_do_pageout (struct vm_area_struct *vma) { -if (vma_is_anonymous (vma)) +if (vma_is_private_anon (vma))return true;if (!vma->vm_file)return false;
@@ -725,7 +725,7 @@ static int madvise_free_single_vma(struct vm_area_struct *vma,struct mmu_gather tlb;/* MADV_FREE works for only anon vma at the moment */ -if (!vma_is_anonymous (vma)) +if (!vma_is_private_anon (vma))return -EINVAL ; range.start =max (vma->vm_start, start_addr);
mm/memory.c +5 -5
@@ -4494,7 +4494,7 @@ static vm_fault_t do_numa_page(struct vm_fault *vmf)static inline vm_fault_t create_huge_pmd (struct vm_fault *vmf) { -if (vma_is_anonymous (vmf->vma)) +if (vma_is_private_anon (vmf->vma))return do_huge_pmd_anonymous_page (vmf);if (vmf->vma->vm_ops->huge_fault)return vmf->vma->vm_ops->huge_fault (vmf,PE_SIZE_PMD );
@@ -4504,7 +4504,7 @@ static inline vm_fault_t create_huge_pmd(struct vm_fault *vmf)/* `inline' is required to avoid gcc 4.1.2 build error */ static inline vm_fault_t wp_huge_pmd (struct vm_fault *vmf) { -if (vma_is_anonymous (vmf->vma)) { +if (vma_is_private_anon (vmf->vma)) {if (userfaultfd_huge_pmd_wp (vmf->vma, vmf->orig_pmd))return handle_userfault (vmf,VM_UFFD_WP );return do_huge_pmd_wp_page (vmf);
@@ -4527,7 +4527,7 @@ static vm_fault_t create_huge_pud(struct vm_fault *vmf)#if defined(CONFIG_TRANSPARENT_HUGEPAGE ) && \ defined(CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD )/* No support for anonymous transparent PUD pages yet */ -if (vma_is_anonymous (vmf->vma)) +if (vma_is_private_anon (vmf->vma)) goto split;if (vmf->vma->vm_ops->huge_fault) {vm_fault_t ret = vmf->vma->vm_ops->huge_fault (vmf,PE_SIZE_PUD );
@@ -4546,7 +4546,7 @@ static vm_fault_t wp_huge_pud(struct vm_fault *vmf, pud_t orig_pud) {#ifdef CONFIG_TRANSPARENT_HUGEPAGE /* No support for anonymous transparent PUD pages yet */ -if (vma_is_anonymous (vmf->vma)) +if (vma_is_private_anon (vmf->vma))return VM_FAULT_FALLBACK ;if (vmf->vma->vm_ops->huge_fault)return vmf->vma->vm_ops->huge_fault (vmf,PE_SIZE_PUD );
@@ -4621,7 +4621,7 @@ static vm_fault_t handle_pte_fault(struct vm_fault *vmf) }if (!vmf->pte) { -if (vma_is_anonymous (vmf->vma)) +if (vma_is_private_anon (vmf->vma))return do_anonymous_page (vmf);else return do_fault (vmf);
mm/migrate_device.c +3 -3
@@ -40,7 +40,7 @@ static int migrate_vma_collect_hole(unsigned long start,unsigned long addr;/* Only allow populating anonymous memory. */ -if (!vma_is_anonymous (walk->vma)) +if (!vma_is_private_anon (walk->vma))return migrate_vma_collect_skip (start, end, walk);for (addr = start; addr < end; addr +=PAGE_SIZE ) {
@@ -120,7 +120,7 @@ static int migrate_vma_collect_pmd(pmd_t *pmdp, pte = *ptep;if (pte_none (pte)) { -if (vma_is_anonymous (vma)) { +if (vma_is_private_anon (vma)) { mpfn =MIGRATE_PFN_MIGRATE ; migrate->cpages++; }
@@ -518,7 +518,7 @@ static void migrate_vma_insert_page(struct migrate_vma *migrate,pte_t *ptep;/* Only allow populating anonymous memory */ -if (!vma_is_anonymous (vma)) +if (!vma_is_private_anon (vma)) goto abort; pgdp =pgd_offset (mm, addr);
mm/mincore.c +1 -1
@@ -156,7 +156,7 @@ static int mincore_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,static inline bool can_do_mincore (struct vm_area_struct *vma) { -if (vma_is_anonymous (vma)) +if (vma_is_private_anon (vma))return true;if (!vma->vm_file)return false;
mm/mmap.c +2 -2
@@ -3189,7 +3189,7 @@ int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)* using the existing file pgoff checks and manipulations. * Similarly in do_mmap and in do_brk_flags. */ -if (vma_is_anonymous (vma)) { +if (vma_is_private_anon (vma)) {BUG_ON (vma->anon_vma); vma->vm_pgoff = vma->vm_start >>PAGE_SHIFT ; }
@@ -3217,7 +3217,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,* If anonymous vma has not yet been faulted, update new pgoff * to match new location, to increase its chance of merging. */ -if (unlikely (vma_is_anonymous (vma) && !vma->anon_vma)) { +if (unlikely (vma_is_private_anon (vma) && !vma->anon_vma)) { pgoff = addr >>PAGE_SHIFT ; faulted_in_anon_vma = false; }
mm/oom_kill.c +1 -1
@@ -536,7 +536,7 @@ bool __oom_reap_task_mm(struct mm_struct *mm)* we do not want to block exit_mmap by keeping mm ref * count elevated without a good reason. */ -if (vma_is_anonymous (vma) || !(vma->vm_flags &VM_SHARED )) { +if (vma_is_private_anon (vma) || !(vma->vm_flags &VM_SHARED )) {struct mmu_notifier_range range;struct mmu_gather tlb;
mm/userfaultfd.c +4 -4
@@ -540,9 +540,9 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm, err = -EINVAL ;/* * shmem_zero_setup is invoked in mmap for MAP_ANONYMOUS|MAP_SHARED but -* it will overwrite vm_ops, so vma_is_anonymous must return false. +* it will overwrite vm_ops, so vma_is_private_anon must return false. */ -if (WARN_ON_ONCE (vma_is_anonymous (dst_vma) && +if (WARN_ON_ONCE (vma_is_private_anon (dst_vma) && dst_vma->vm_flags &VM_SHARED )) goto out_unlock;
@@ -561,7 +561,7 @@ static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm,return __mcopy_atomic_hugetlb (dst_mm, dst_vma, dst_start, src_start, len, mcopy_mode); -if (!vma_is_anonymous (dst_vma) && !vma_is_shmem (dst_vma)) +if (!vma_is_private_anon (dst_vma) && !vma_is_shmem (dst_vma)) goto out_unlock;if (!vma_is_shmem (dst_vma) && mcopy_mode ==MCOPY_ATOMIC_CONTINUE ) goto out_unlock;
@@ -717,7 +717,7 @@ int mwriteprotect_range(struct mm_struct *dst_mm, unsigned long start, goto out_unlock;if (!userfaultfd_wp (dst_vma)) goto out_unlock; -if (!vma_is_anonymous (dst_vma)) +if (!vma_is_private_anon (dst_vma)) goto out_unlock;if (enable_wp)
Signature · 1 line
2.35.1
Re: [RFC 1/3] mm: change vma_is_anonymous to vma_is_private_anon
Quoted 4 lines · Nico Pache
The
vma_is_anonymousfunction isn't fully indicative of what it checks.Without having full knowledge of the mmap process, one may incorrectly assume this covers all types of anonymous memory; which is not the case.
Is your complaint that anonymous memory can also be found in file VMAs that were mapped with MAP_PRIVATE? ie COWed pages?
I don't think renaming this function is appropriate. It's whether the VMA is anonymous, not whether the VMA can contain anonymous pages.
Re: [RFC 1/3] mm: change vma_is_anonymous to vma_is_private_anon
Quoted 8 lines · Matthew Wilcox
Quoted 4 lines · Nico Pache
The
vma_is_anonymousfunction isn't fully indicative of what it checks.Without having full knowledge of the mmap process, one may incorrectly assume this covers all types of anonymous memory; which is not the case.
Is your complaint that anonymous memory can also be found in file VMAs that were mapped with
MAP_PRIVATE? ie COWed pages?
I should have been more descriptive in my commit msg about how I came to this conclusion.
From my understanding of the mmap process, a vma->vm_ops field is only NULL when mmapped as !file and !shared:
if (file){
...
} else if (vm_flags & VM_SHARED) { //ANON SHARED
error = shmem_zero_setup(vma);
if (error)
goto free_vma;
} else { //ANON PRIVATE
vma_set_anonymous(vma); //set vma->vm_ops= NULL
}To me this means that the VMA is PRIVATE ANON memory. The vma_is_anonymous function returns true when vm_ops == NULL. So my intentions were to more accurately describe what we are checking for. I could be wrong though thats why I started with an RFC :)
There could be some aspect of COW that I dont fully understand. It is not something I've looked into much.
Cheers,
-- Nico
Re: [RFC 1/3] mm: change vma_is_anonymous to vma_is_private_anon
Quoted 27 lines · Nico Pache
Quoted 8 lines · Matthew Wilcox
Quoted 4 lines · Nico Pache
The
vma_is_anonymousfunction isn't fully indicative of what it checks.Without having full knowledge of the mmap process, one may incorrectly assume this covers all types of anonymous memory; which is not the case.
Is your complaint that anonymous memory can also be found in file VMAs that were mapped with
MAP_PRIVATE? ie COWed pages?I should have been more descriptive in my commit msg about how I came to this conclusion.
From my understanding of the mmap process, a
vma->vm_opsfield is only NULL when mmapped as !file and !shared:if (file){ ... } else if (vm_flags & VM_SHARED) { //ANON SHARED error = shmem_zero_setup(vma); if (error) goto free_vma; } else { //ANON PRIVATE vma_set_anonymous(vma); //set vma->vm_ops= NULL }
Quoted 4 lines
To me this means that the VMA is PRIVATE ANON memory. The
vma_is_anonymousfunction returns true whenvm_ops== NULL. So my intentions were to more accurately describe what we are checking for. I could be wrong though thats why I started with an RFC :)
Shared anon in the kernel is really just shmem. The user space notion is MAP_ANON|MAP_SHARED, but that really just maps to shmem and the kernel doesn't really call that thing anonymous memory.
So I agree, renaming this is not appropriate.
Signature · 3 lines
Thanks, David / dhildenb
RFC 2/3 futex: exit: Print a warning when futex_cleanup fails
In reply to Nico PacheThe futex_cleanup routines currently fails silently.
Allow the futex_cleanup routines to fail more verbosely so we can leave a message behind for easier debugging/error detecting.
--- kernel/futex/core.c | 44 ++++++++++++++++++++++++++------------------ 1 file changed, 26 insertions(+), 18 deletions(-)
kernel/futex/core.c +26 -18
@@ -760,9 +760,9 @@ static inline int fetch_robust_entry(struct robust_list __user **entry,* Walk curr->robust_list (very carefully, it's a userspace list!) * and mark any locks found there dead, and notify any waiters. * -* We silently return on any sign of list-walking problem. +* We return false on any sign of list-walking problem. */ -static void exit_robust_list (struct task_struct *curr) +static bool exit_robust_list (struct task_struct *curr) {struct robust_list_head __user *head = curr->robust_list;struct robust_list __user *entry, *next_entry, *pending;
@@ -771,23 +771,25 @@ static void exit_robust_list(struct task_struct *curr)unsigned long futex_offset;int rc; +if (!futex_cmpxchg_enabled) +return false;/* * Fetch the list head (which was registered earlier, via * sys_set_robust_list()): */ if (fetch_robust_entry (&entry, &head->list.next, &pi)) -return ; +return false;/* * Fetch the relative futex offset: */ if (get_user (futex_offset, &head->futex_offset)) -return ; +return false;/* * Fetch any possibly pending lock-add first, and handle it * if it exists: */ if (fetch_robust_entry (&pending, &head->list_op_pending, &pip)) -return ; +return false; next_entry =NULL ;/* avoid warning with gcc */ while (entry != &head->list) {
@@ -803,10 +805,10 @@ static void exit_robust_list(struct task_struct *curr)if (entry != pending) {if (handle_futex_death ((void __user *)entry + futex_offset, curr, pi,HANDLE_DEATH_LIST )) -return ; +return false; }if (rc) -return ; +return false; entry = next_entry; pi = next_pi;/*
@@ -819,9 +821,10 @@ static void exit_robust_list(struct task_struct *curr) }if (pending) { -handle_futex_death ((void __user *)pending + futex_offset, +return handle_futex_death ((void __user *)pending + futex_offset, curr, pip,HANDLE_DEATH_PENDING ); } +return true; }#ifdef CONFIG_COMPAT
@@ -854,9 +857,9 @@ compat_fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **ent* Walk curr->robust_list (very carefully, it's a userspace list!) * and mark any locks found there dead, and notify any waiters. * -* We silently return on any sign of list-walking problem. +* We return false on any sign of list-walking problem. */ -static void compat_exit_robust_list (struct task_struct *curr) +static bool compat_exit_robust_list (struct task_struct *curr) {struct compat_robust_list_head __user *head = curr->compat_robust_list;struct robust_list __user *entry, *next_entry, *pending;
@@ -866,24 +869,26 @@ static void compat_exit_robust_list(struct task_struct *curr)compat_long_t futex_offset;int rc; +if (!futex_cmpxchg_enabled) +return false;/* * Fetch the list head (which was registered earlier, via * sys_set_robust_list()): */ if (compat_fetch_robust_entry (&uentry, &entry, &head->list.next, &pi)) -return ; +return false;/* * Fetch the relative futex offset: */ if (get_user (futex_offset, &head->futex_offset)) -return ; +return false;/* * Fetch any possibly pending lock-add first, and handle it * if it exists: */ if (compat_fetch_robust_entry (&upending, &pending, &head->list_op_pending, &pip)) -return ; +return false; next_entry =NULL ;/* avoid warning with gcc */ while (entry != (struct robust_list __user *) &head->list) {
@@ -902,10 +907,10 @@ static void compat_exit_robust_list(struct task_struct *curr)if (handle_futex_death (uaddr, curr, pi,HANDLE_DEATH_LIST )) -return ; +return false; }if (rc) -return ; +return false; uentry = next_uentry; entry = next_entry; pi = next_pi;
@@ -920,8 +925,9 @@ static void compat_exit_robust_list(struct task_struct *curr)if (pending) {void __user *uaddr =futex_uaddr (pending, futex_offset); -handle_futex_death (uaddr, curr, pip,HANDLE_DEATH_PENDING ); +return handle_futex_death (uaddr, curr, pip,HANDLE_DEATH_PENDING ); } + return true; }#endif
@@ -1007,13 +1013,15 @@ static inline void exit_pi_state_list(struct task_struct *curr) { }static void futex_cleanup (struct task_struct *tsk) {if (unlikely (tsk->robust_list)) { -exit_robust_list (tsk); +if (!exit_robust_list (tsk)) +pr_info ("futex: exit_robust_list failed" ); tsk->robust_list =NULL ; }#ifdef CONFIG_COMPAT if (unlikely (tsk->compat_robust_list)) { -compat_exit_robust_list (tsk); +if (!compat_exit_robust_list (tsk)) +pr_info ("futex: compat_exit_robust_list failed" ); tsk->compat_robust_list =NULL ; }#endif
Signature · 1 line
2.35.1
Re: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
kernel/futex/core.c static inline void exit_pi_state_list(struct task_struct *curr) { } · line 1017
Quoted 7 lines · Nico Pache
kernel/futex/core.c +2
-1@@ -1007,13 +1013,15 @@ static inline void exit_pi_state_list(struct task_struct *curr) { }static void futex_cleanup (struct task_struct *tsk) {if (unlikely (tsk->robust_list)) { -exit_robust_list (tsk); +if (!exit_robust_list (tsk)) +pr_info ("futex: exit_robust_list failed" );
Doesn't this allow a malicious user process to spam the kernel logs with messages? There needs to be a ratelimit on this, at least.
Re: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
Quoted 11 lines · Matthew Wilcox
kernel/futex/core.c static inline void exit_pi_state_list(struct task_struct *curr) { } · line 1017
Quoted 7 lines · Nico Pache
kernel/futex/core.c +2
-1@@ -1007,13 +1013,15 @@ static inline void exit_pi_state_list(struct task_struct *curr) { }static void futex_cleanup (struct task_struct *tsk) {if (unlikely (tsk->robust_list)) { -exit_robust_list (tsk); +if (!exit_robust_list (tsk)) +pr_info ("futex: exit_robust_list failed" );Doesn't this allow a malicious user process to spam the kernel logs with messages? There needs to be a ratelimit on this, at least.
Fair point, we'd need a ratelimited print if we want to continue forward with this. Additionally we may want to limit this print to debug kernels, but thats just a thought.
Thanks for the review :)
-- Nico
Re: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
In reply to Nico PacheQuoted 4 lines · Nico Pache
The
futex_cleanuproutines currently fails silently.Allow the
futex_cleanuproutines to fail more verbosely so we can leave a message behind for easier debugging/error detecting.
This is a free ticket to spam dmesg for any unpriviledged user.
Quoted 1 line
Fixes 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core")
There is nothing to fix, really.
Robust futexes are best effort as I explained you before and spamming dmesg won't help anything.
Thanks,
tglx
Re: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
Quoted 8 lines · Thomas Gleixner
Quoted 4 lines · Nico Pache
The
futex_cleanuproutines currently fails silently.Allow the
futex_cleanuproutines to fail more verbosely so we can leave a message behind for easier debugging/error detecting.This is a free ticket to spam dmesg for any unpriviledged user.
As Matthew pointed out, we'd want to rate limit this. I was also thinking we can limit this to debug kernels so unpriviledged users can not cause dmesg spamming in production configs.
Quoted 4 lines
Quoted 2 lines
Fixes 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core")There is nothing to fix, really.
Fair enough, its been part of the design since inceptions.
Quoted 3 lines
Robust futexes are best effort as I explained you before and spamming dmesg won't help anything.
It would have helped find the OOM/Robust Futex issue more quickly. It may also help detect breakages in the pthread code (or any other users of robust futex) if someone is doing something incorrectly.
Just a thought though. If you think its pointless to proceed down this path I wont push forward.
Cheers,
-- Nico
Re: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
Quoted 7 lines · Nico Pache
Quoted 3 lines · Thomas Gleixner
On Thu, Apr 21 2022 at 15:05, Nico Pache wrote:
Robust futexes are best effort as I explained you before and spamming dmesg won't help anything.It would have helped find the OOM/Robust Futex issue more quickly. It may also help detect breakages in the pthread code (or any other users of robust futex) if someone is doing something incorrectly.
If we follow that line of argumentation then we have to sprinkle printks en masse all over the place as there are a gazillion ways to create hard to debug issues.
Thanks,
tglx
kernel test robot
build botRe: [RFC 2/3] futex: exit: Print a warning when futex_cleanup fails
In reply to Nico Pache[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in https://git-scm.com/docs/git-format-patch]
Hi Nico,
[FYI, it's a private test report for your RFC patch.]
[auto build test ERROR on powerpc/next]
[also build test ERROR on tip/locking/core linus/master v5.18-rc3] [cannot apply to hnaz-mm/master linux/master next-20220421] [If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]
url: https://github.com/intel-lab-lkp/linux/commits/Nico-Pache/Slight-improvements-for-OOM-Futex/20220422-030740 base: https://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux.git next config: hexagon-randconfig-r025-20220421 (https://download.01.org/0day-ci/archive/20220422/[email protected]/config) compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project 5bd87350a5ae429baf8f373cb226a57b62f87280)
reproduce (this is a W=1 build):
wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross chmod +x ~/bin/make.cross
# https://github.com/intel-lab-lkp/linux/commit/cd9fef152153fe60c67bb633118dd8adef51cadc
git remote add linux-review https://github.com/intel-lab-lkp/linux
git fetch --no-tags linux-review Nico-Pache/Slight-improvements-for-OOM-Futex/20220422-030740 git checkout cd9fef152153fe60c67bb633118dd8adef51cadc
# save the config file
mkdir build_dir && cp config build_dir/.config
COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=hexagon SHELL=/bin/bash
If you fix the issue, kindly add following tag as appropriate
All errors (new ones prefixed by >>):
>> kernel/futex/core.c:774:7: error: use of undeclared identifier 'futex_cmpxchg_enabled'
if (!futex_cmpxchg_enabled)
^1 error generated.
vim +/futex_cmpxchg_enabled +774 kernel/futex/core.c
758
759 /*
760 * Walk curr->robust_list (very carefully, it's a userspace list!)
761 * and mark any locks found there dead, and notify any waiters.
762 *
763 * We return false on any sign of list-walking problem.
764 */
765 static bool exit_robust_list(struct task_struct *curr)
766 {
767 struct robust_list_head __user *head = curr->robust_list;
768 struct robust_list __user *entry, *next_entry, *pending;
769 unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
770 unsigned int next_pi;
771 unsigned long futex_offset;
772 int rc;
773
> 774 if (!futex_cmpxchg_enabled)
775 return false;
776 /*
777 * Fetch the list head (which was registered earlier, via
778 * sys_set_robust_list()):
779 */
780 if (fetch_robust_entry(&entry, &head->list.next, &pi))
781 return false;
782 /*
783 * Fetch the relative futex offset:
784 */
785 if (get_user(futex_offset, &head->futex_offset))
786 return false;
787 /*
788 * Fetch any possibly pending lock-add first, and handle it
789 * if it exists:
790 */
791 if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
792 return false;
793
794 next_entry = NULL; /* avoid warning with gcc */
795 while (entry != &head->list) {
796 /*
797 * Fetch the next entry in the list before calling
798 * handle_futex_death:
799 */
800 rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
801 /*
802 * A pending lock might already be on the list, so
803 * don't process it twice:
804 */
805 if (entry != pending) {
806 if (handle_futex_death((void __user *)entry + futex_offset,
807 curr, pi, HANDLE_DEATH_LIST))
808 return false;
809 }
810 if (rc)
811 return false;
812 entry = next_entry;
813 pi = next_pi;
814 /*
815 * Avoid excessively long or circular lists:
816 */
817 if (!--limit)
818 break;
819
820 cond_resched();
821 }
822
823 if (pending) {
824 return handle_futex_death((void __user *)pending + futex_offset,
825 curr, pip, HANDLE_DEATH_PENDING);
826 }
827 return true;
828 }
829
Signature · 2 lines
0-DAY CI Kernel Test Service https://01.org/lkp
RFC 3/3 exit: Check for MMF_OOM_SKIP in exit_mmap
In reply to Nico PacheThe MMF_OOM_SKIP bit is used to indicate weather a mm_struct can not be invalided or has already been invalided. exit_mmap currently calls __oom_reap_task_mm unconditionally despite the fact that the oom reaper may have already called this.
Add a check for the MMF_OOM_SKIP bit being set in exit_mmap to avoid unnessary calls to the invalidate code.
A slight race can occur on the MMF_OOM_SKIP bit that will still allow this to run twice. My testing has shown an ~66% decrease in double calls to _oom_reap_task_mm.
--- mm/mmap.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
mm/mmap.c +2 -1
@@ -3113,7 +3113,8 @@ void exit_mmap(struct mm_struct *mm)/* mm's last user has gone, and its about to be pulled down */ mmu_notifier_release (mm); -if (unlikely (mm_is_oom_victim (mm))) { +if (unlikely (mm_is_oom_victim (mm)) && + !test_bit (MMF_OOM_SKIP , &mm->flags)) {/* * Manually reap the mm to free as much memory as possible. * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
Signature · 1 line
2.35.1
Michal Hocko
maintainerRe: [RFC 3/3] exit: Check for MMF_OOM_SKIP in exit_mmap
Quoted 7 lines · Nico Pache
The
MMF_OOM_SKIPbit is used to indicate weather amm_structcan not be invalided or has already been invalided.exit_mmapcurrently calls__oom_reap_task_mmunconditionally despite the fact that the oom reaper may have already called this.Add a check for the
MMF_OOM_SKIPbit being set inexit_mmapto avoid unnessary calls to the invalidate code.
Why do we care about this?
Quoted 5 lines
A slight race can occur on the
MMF_OOM_SKIPbit that will still allow this to run twice. My testing has shown an ~66% decrease in double calls to_oom_reap_task_mm.Fixes 27ae357fa82b ("mm, oom: fix concurrent munlock and oom reaper unmap, v3")
I do not see this would be fixing anything.
Quoted 24 lines
Cc David Rientjes <[email protected]>Cc Michal Hocko <[email protected]>Cc Andrea Arcangeli <[email protected]>Signed-off-by Nico Pache <[email protected]>--- mm/mmap.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)mm/mmap.c +2
-1@@ -3113,7 +3113,8 @@ void exit_mmap(struct mm_struct *mm)/* mm's last user has gone, and its about to be pulled down */ mmu_notifier_release (mm); -if (unlikely (mm_is_oom_victim (mm))) { +if (unlikely (mm_is_oom_victim (mm)) && + !test_bit (MMF_OOM_SKIP , &mm->flags)) {/* * Manually reap the mm to free as much memory as possible. * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard Signature · 1 line
2.35.1
Signature · 2 lines
Michal Hocko SUSE Labs
Re: [RFC 3/3] exit: Check for MMF_OOM_SKIP in exit_mmap
Quoted 10 lines · Michal Hocko
Quoted 7 lines · Nico Pache
The
MMF_OOM_SKIPbit is used to indicate weather amm_structcan not be invalided or has already been invalided.exit_mmapcurrently calls__oom_reap_task_mmunconditionally despite the fact that the oom reaper may have already called this.Add a check for the
MMF_OOM_SKIPbit being set inexit_mmapto avoid unnessary calls to the invalidate code.Why do we care about this?
Is there no cost to the MMU/TLB invalidation? The MMU notifier contains a lock too so perhaps we can also avoids some unnecessary MMU notifier lock contention.
Quoted 8 lines
Quoted 6 lines
A slight race can occur on the
MMF_OOM_SKIPbit that will still allow this to run twice. My testing has shown an ~66% decrease in double calls to_oom_reap_task_mm.Fixes 27ae357fa82b ("mm, oom: fix concurrent munlock and oom reaper unmap, v3")I do not see this would be fixing anything.
Ok im just trying to make sure we are keeping an eye on what introduced this double call. Davids commit above is what introduced the oom_reap_task_mm in the exit_mmap code. It goes along with some other changes that I dont fully understand without more studying, so that's why I was hoping he could provide some input around that CVE (the main thing im concerned about re-introducing).
Quoted 26 lines
Quoted 25 lines
Cc David Rientjes <[email protected]>Cc Michal Hocko <[email protected]>Cc Andrea Arcangeli <[email protected]>Signed-off-by Nico Pache <[email protected]>--- mm/mmap.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)mm/mmap.c +2
-1@@ -3113,7 +3113,8 @@ void exit_mmap(struct mm_struct *mm)/* mm's last user has gone, and its about to be pulled down */ mmu_notifier_release (mm); -if (unlikely (mm_is_oom_victim (mm))) { +if (unlikely (mm_is_oom_victim (mm)) && + !test_bit (MMF_OOM_SKIP , &mm->flags)) {/* * Manually reap the mm to free as much memory as possible. * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard Signature · 1 line
2.35.1
Michal Hocko
maintainerRe: [RFC 3/3] exit: Check for MMF_OOM_SKIP in exit_mmap
Quoted 9 lines · Nico Pache
Quoted 6 lines · Michal Hocko
Quoted 5 lines · Nico Pache
The
MMF_OOM_SKIPbit is used to indicate weather amm_structcan not be invalided or has already been invalided.exit_mmapcurrently calls__oom_reap_task_mmunconditionally despite the fact that the oom reaper may have already called this.
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
Add a check for the
MMF_OOM_SKIPbit being set inexit_mmapto avoid unnessary calls to the invalidate code.Why do we care about this?
Is there no cost to the MMU/TLB invalidation? The MMU notifier contains a lock too so perhaps we can also avoids some unnecessary MMU notifier lock contention.
I am pretty sure that this area can be micro optimized but I do not really see a strong reason for that. OOM victims are/should be really rare so I do not think that any performance optimization would be really visible.
If you want to improve the code then I think a much better plan would be to get rid of the whole oom special case altogether. This might be much closer than ever after Hugh's recent m{un}lock changes. I didn't have time to think that through though. I believe Suren Baghdasaryan has been looking into that as well.
Signature · 2 lines
Michal Hocko SUSE Labs