提交 0b9b6fff 编写于 作者: K Kirill A. Shutemov 提交者: Linus Torvalds

thp: fix interrupt unsafe locking in split_huge_page()

split_queue_lock can be taken from interrupt context in some cases, but
I forgot to convert locking in split_huge_page() to interrupt-safe
primitives.

Let's fix this.

lockdep output:

  ======================================================
  [ INFO: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected ]
  4.4.0+ #259 Tainted: G        W
  ------------------------------------------------------
  syz-executor/18183 [HC0[0]:SC0[2]:HE0:SE0] is trying to acquire:
   (split_queue_lock){+.+...}, at: free_transhuge_page+0x24/0x90 mm/huge_memory.c:3436

  and this task is already holding:
   (slock-AF_INET){+.-...}, at: spin_lock_bh include/linux/spinlock.h:307
   (slock-AF_INET){+.-...}, at: lock_sock_fast+0x45/0x120 net/core/sock.c:2462
  which would create a new lock dependency:
   (slock-AF_INET){+.-...} -> (split_queue_lock){+.+...}

  but this new dependency connects a SOFTIRQ-irq-safe lock:
   (slock-AF_INET){+.-...}
  ... which became SOFTIRQ-irq-safe at:
     mark_irqflags kernel/locking/lockdep.c:2799
     __lock_acquire+0xfd8/0x4700 kernel/locking/lockdep.c:3162
     lock_acquire+0x1dc/0x430 kernel/locking/lockdep.c:3585
     __raw_spin_lock include/linux/spinlock_api_smp.h:144
     _raw_spin_lock+0x33/0x50 kernel/locking/spinlock.c:151
     spin_lock include/linux/spinlock.h:302
     udp_queue_rcv_skb+0x781/0x1550 net/ipv4/udp.c:1680
     flush_stack+0x50/0x330 net/ipv6/udp.c:799
     __udp4_lib_mcast_deliver+0x694/0x7f0 net/ipv4/udp.c:1798
     __udp4_lib_rcv+0x17dc/0x23e0 net/ipv4/udp.c:1888
     udp_rcv+0x21/0x30 net/ipv4/udp.c:2108
     ip_local_deliver_finish+0x2b3/0xa50 net/ipv4/ip_input.c:216
     NF_HOOK_THRESH include/linux/netfilter.h:226
     NF_HOOK include/linux/netfilter.h:249
     ip_local_deliver+0x1c4/0x2f0 net/ipv4/ip_input.c:257
     dst_input include/net/dst.h:498
     ip_rcv_finish+0x5ec/0x1730 net/ipv4/ip_input.c:365
     NF_HOOK_THRESH include/linux/netfilter.h:226
     NF_HOOK include/linux/netfilter.h:249
     ip_rcv+0x963/0x1080 net/ipv4/ip_input.c:455
     __netif_receive_skb_core+0x1620/0x2f80 net/core/dev.c:4154
     __netif_receive_skb+0x2a/0x160 net/core/dev.c:4189
     netif_receive_skb_internal+0x1b5/0x390 net/core/dev.c:4217
     napi_skb_finish net/core/dev.c:4542
     napi_gro_receive+0x2bd/0x3c0 net/core/dev.c:4572
     e1000_clean_rx_irq+0x4e2/0x1100 drivers/net/ethernet/intel/e1000e/netdev.c:1038
     e1000_clean+0xa08/0x24a0 drivers/net/ethernet/intel/e1000/e1000_main.c:3819
     napi_poll net/core/dev.c:5074
     net_rx_action+0x7eb/0xdf0 net/core/dev.c:5139
     __do_softirq+0x26a/0x920 kernel/softirq.c:273
     invoke_softirq kernel/softirq.c:350
     irq_exit+0x18f/0x1d0 kernel/softirq.c:391
     exiting_irq ./arch/x86/include/asm/apic.h:659
     do_IRQ+0x86/0x1a0 arch/x86/kernel/irq.c:252
     ret_from_intr+0x0/0x20 arch/x86/entry/entry_64.S:520
     arch_safe_halt ./arch/x86/include/asm/paravirt.h:117
     default_idle+0x52/0x2e0 arch/x86/kernel/process.c:304
     arch_cpu_idle+0xa/0x10 arch/x86/kernel/process.c:295
     default_idle_call+0x48/0xa0 kernel/sched/idle.c:92
     cpuidle_idle_call kernel/sched/idle.c:156
     cpu_idle_loop kernel/sched/idle.c:252
     cpu_startup_entry+0x554/0x710 kernel/sched/idle.c:300
     rest_init+0x192/0x1a0 init/main.c:412
     start_kernel+0x678/0x69e init/main.c:683
     x86_64_start_reservations+0x2a/0x2c arch/x86/kernel/head64.c:195
     x86_64_start_kernel+0x158/0x167 arch/x86/kernel/head64.c:184

  to a SOFTIRQ-irq-unsafe lock:
   (split_queue_lock){+.+...}
   which became SOFTIRQ-irq-unsafe at:
     mark_irqflags kernel/locking/lockdep.c:2817
     __lock_acquire+0x146e/0x4700 kernel/locking/lockdep.c:3162
     lock_acquire+0x1dc/0x430 kernel/locking/lockdep.c:3585
     __raw_spin_lock include/linux/spinlock_api_smp.h:144
     _raw_spin_lock+0x33/0x50 kernel/locking/spinlock.c:151
     spin_lock include/linux/spinlock.h:302
     split_huge_page_to_list+0xcc0/0x1c50 mm/huge_memory.c:3399
     split_huge_page include/linux/huge_mm.h:99
     queue_pages_pte_range+0xa38/0xef0 mm/mempolicy.c:507
     walk_pmd_range mm/pagewalk.c:50
     walk_pud_range mm/pagewalk.c:90
     walk_pgd_range mm/pagewalk.c:116
     __walk_page_range+0x653/0xcd0 mm/pagewalk.c:204
     walk_page_range+0xfe/0x2b0 mm/pagewalk.c:281
     queue_pages_range+0xfb/0x130 mm/mempolicy.c:687
     migrate_to_node mm/mempolicy.c:1004
     do_migrate_pages+0x370/0x4e0 mm/mempolicy.c:1109
     SYSC_migrate_pages mm/mempolicy.c:1453
     SyS_migrate_pages+0x640/0x730 mm/mempolicy.c:1374
     entry_SYSCALL_64_fastpath+0x16/0x7a arch/x86/entry/entry_64.S:185

  other info that might help us debug this:

   Possible interrupt unsafe locking scenario:

         CPU0                    CPU1
         ----                    ----
    lock(split_queue_lock);
                                 local_irq_disable();
                                 lock(slock-AF_INET);
                                 lock(split_queue_lock);
    <Interrupt>
      lock(slock-AF_INET);
Signed-off-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: NDmitry Vyukov <dvyukov@google.com>
Acked-by: NDavid Rientjes <rientjes@google.com>
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
上级 290e0e0f
...@@ -3357,6 +3357,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) ...@@ -3357,6 +3357,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
struct anon_vma *anon_vma; struct anon_vma *anon_vma;
int count, mapcount, ret; int count, mapcount, ret;
bool mlocked; bool mlocked;
unsigned long flags;
VM_BUG_ON_PAGE(is_huge_zero_page(page), page); VM_BUG_ON_PAGE(is_huge_zero_page(page), page);
VM_BUG_ON_PAGE(!PageAnon(page), page); VM_BUG_ON_PAGE(!PageAnon(page), page);
...@@ -3396,7 +3397,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) ...@@ -3396,7 +3397,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
lru_add_drain(); lru_add_drain();
/* Prevent deferred_split_scan() touching ->_count */ /* Prevent deferred_split_scan() touching ->_count */
spin_lock(&split_queue_lock); spin_lock_irqsave(&split_queue_lock, flags);
count = page_count(head); count = page_count(head);
mapcount = total_mapcount(head); mapcount = total_mapcount(head);
if (!mapcount && count == 1) { if (!mapcount && count == 1) {
...@@ -3404,11 +3405,11 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) ...@@ -3404,11 +3405,11 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
split_queue_len--; split_queue_len--;
list_del(page_deferred_list(head)); list_del(page_deferred_list(head));
} }
spin_unlock(&split_queue_lock); spin_unlock_irqrestore(&split_queue_lock, flags);
__split_huge_page(page, list); __split_huge_page(page, list);
ret = 0; ret = 0;
} else if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) { } else if (IS_ENABLED(CONFIG_DEBUG_VM) && mapcount) {
spin_unlock(&split_queue_lock); spin_unlock_irqrestore(&split_queue_lock, flags);
pr_alert("total_mapcount: %u, page_count(): %u\n", pr_alert("total_mapcount: %u, page_count(): %u\n",
mapcount, count); mapcount, count);
if (PageTail(page)) if (PageTail(page))
...@@ -3416,7 +3417,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list) ...@@ -3416,7 +3417,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
dump_page(page, "total_mapcount(head) > 0"); dump_page(page, "total_mapcount(head) > 0");
BUG(); BUG();
} else { } else {
spin_unlock(&split_queue_lock); spin_unlock_irqrestore(&split_queue_lock, flags);
unfreeze_page(anon_vma, head); unfreeze_page(anon_vma, head);
ret = -EBUSY; ret = -EBUSY;
} }
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册