8d75c338f0: BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map
Subsystems (5)
kernel test robot
build botHello,
kernel test robot noticed "BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map" on:
commit: 8d75c338f0bcecaa6c9af67f86c176b67b6acf3e ("sysctl: remove CONFIG_PROC_SYSCTL, it just mirrors CONFIG_SYSCTL") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
[test failed on linux-next/master b810335a0699d1810683796b0e3ffe8827ddfe6c]
in testcase: boot
config: x86_64-randconfig-2300-20261014
compiler: gcc-14
test machine: qemu-system-x86_64 -enable-kvm -cpu Haswell-noTSX,enforce -smp 4 -m 16G
(please refer to attached dmesg/kmsg for entire log/backtrace)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-lkp/[email protected]
[ 35.688893][ T1] BUG: KASAN: slab-use-after-free in touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] Read of size 8 at addr ffff8881259a2968 by task swapper/0/1 [ 35.688893][ T1] [ 35.688893][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.688893][ T1] Call Trace: [ 35.688893][ T1] <TASK> [ 35.688893][ T1] dump_stack_lvl (dump_stack.c:94 dump_stack.c:120) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_address_description+0x51/0x300 [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_report (kasan/report.c:482) [ 35.688893][ T1] ? kmem_cache_debug_flags (slab.h:503) [ 35.688893][ T1] ? fixup_red_left (slub.c:250) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] kasan_report (kasan/report.c:595) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] __flush_workqueue (workqueue.c:4079) [ 35.688893][ T1] ? xa_load (xarray.c:1621) [ 35.688893][ T1] ? tracer_preempt_on (trace/trace_irqsoff.c:686 (discriminator 1)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4043) [ 35.688893][ T1] ? __up_write (locking/rwsem.c:1422 (discriminator 4)) [ 35.688893][ T1] rds_ib_init (rds/ib.c:618) [ 35.688893][ T1] rds_rdma_init (rds/rdma_transport.c:300) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] ? __raw_spin_unlock_irqrestore (linux/spinlock_api_smp.h:179 (discriminator 4)) [ 35.688893][ T1] ? add_device_randomness (linux/spinlock.h:408 char/random.c:936) [ 35.688893][ T1] ? random_get_entropy (x86/include/asm/timex.h:11) [ 35.688893][ T1] ? sctp_diag_init (diag.c:?) [ 35.688893][ T1] ? rds_init (rds/af_rds.c:977) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] ? trace_initcall_finish_cb (main.c:1289 (discriminator 1)) [ 35.688893][ T1] ? __kmalloc_noprof (slub.c:5363 slub.c:5387) [ 35.688893][ T1] ? do_initcalls (linux/slab.h:973 linux/slab.h:1290 main.c:1418) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ? exit_thread (x86/kernel/process.c:138) [ 35.688893][ T1] ? __switch_to (x86/kernel/process_64.c:674) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] </TASK> [ 35.688893][ T1] [ 35.688893][ T1] Allocated by task 1 on cpu 1 at 35.661878s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] __kasan_kmalloc (kasan/common.c:415) [ 35.688893][ T1] __kmalloc_noprof (linux/kasan.h:263 slub.c:5362 slub.c:5387) [ 35.688893][ T1] __alloc_workqueue (workqueue.c:5827) [ 35.688893][ T1] alloc_workqueue_va (workqueue.c:5946) [ 35.688893][ T1] alloc_workqueue_noprof (workqueue.c:5962) [ 35.688893][ T1] rds_threads_init (rds/threads.c:259 (discriminator 2)) [ 35.688893][ T1] rds_init (rds/af_rds.c:961) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] Freed by task 0 on cpu 3 at 35.674112s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] kasan_save_free_info (kasan/generic.c:584) [ 35.688893][ T1] __kasan_slab_free (kasan/common.c:285) [ 35.688893][ T1] kfree (linux/kasan.h:235 slub.c:2705 slub.c:6405 slub.c:6720) [ 35.688893][ T1] rcu_do_batch (rcu/tree.c:2645) [ 35.688893][ T1] rcu_core (rcu/tree.c:2897) [ 35.688893][ T1] handle_softirqs (softirq.c:622) [ 35.688893][ T1] __irq_exit_rcu (softirq.c:496 softirq.c:735) [ 35.688893][ T1] irq_exit_rcu (softirq.c:752) [ 35.688893][ T1] sysvec_apic_timer_interrupt (x86/kernel/apic/apic.c:1062 (discriminator 35) x86/kernel/apic/apic.c:1062 (discriminator 35)) [ 35.688893][ T1] asm_sysvec_apic_timer_interrupt (x86/include/asm/idtentry.h:546) [ 35.688893][ T1] [ 35.688893][ T1] Last potentially related work creation: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_record_aux_stack (kasan/generic.c:556) [ 35.688893][ T1] __call_rcu_common+0xe6/0x480 [ 35.688893][ T1] kthread_worker_fn (kthread.c:1025) [ 35.688893][ T1] kthread (kthread.c:436) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the physical page: [ 35.688893][ T1] page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1259a0 [ 35.688893][ T1] head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 [ 35.688893][ T1] flags: 0x4000000000000040(head|zone=2) [ 35.688893][ T1] page_type: f5(slab) [ 35.688893][ T1] raw: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] raw: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] head: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000003 ffffea0004966801 00000000ffffffff 00000000ffffffff [ 35.688893][ T1] head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008 [ 35.688893][ T1] page dumped because: kasan: bad access detected [ 35.688893][ T1] [ 35.688893][ T1] Memory state around the buggy address: [ 35.688893][ T1] ffff8881259a2800: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2880: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] >ffff8881259a2900: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ^ [ 35.688893][ T1] ffff8881259a2980: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2a00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ================================================================== [ 35.984299][ T1] Disabling lock debugging due to kernel taint [ 35.987548][ T1] Oops: general protection fault, probably for non-canonical address 0xe0a07c0ec0000008: 0000 [#1] SMP KASAN [ 35.988482][ T1] KASAN: maybe wild-memory-access in range [0x0504007600000040-0x0504007600000047] [ 35.988482][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Tainted: G B 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.988482][ T1] Tainted: [B]=BAD_PAGE [ 35.988482][ T1] RIP: 0010:do_raw_spin_lock (locking/spinlock_debug.c:86 (discriminator 2) locking/spinlock_debug.c:115 (discriminator 2)) [ 35.988482][ T1] Code: 3b 58 81 49 8d 14 04 c7 02 f1 f1 f1 f1 c7 42 04 04 f3 f3 f3 65 48 8b 15 f0 53 f7 08 48 89 54 24 60 31 d2 48 89 fa 48 c1 ea 03 <8a> 14 02 48 89 f8 83 e0 07 83 c0 03 38 d0 7c 09 84 d2 74 05 e8 39 All code ======== 0: 3b 58 81 cmp -0x7f(%rax),%ebx 3: 49 8d 14 04 lea (%r12,%rax,1),%rdx 7: c7 02 f1 f1 f1 f1 movl $0xf1f1f1f1,(%rdx) d: c7 42 04 04 f3 f3 f3 movl $0xf3f3f304,0x4(%rdx) 14: 65 48 8b 15 f0 53 f7 mov %gs:0x8f753f0(%rip),%rdx # 0x8f7540c 1b: 08 1c: 48 89 54 24 60 mov %rdx,0x60(%rsp) 21: 31 d2 xor %edx,%edx 23: 48 89 fa mov %rdi,%rdx 26: 48 c1 ea 03 shr $0x3,%rdx 2a:* 8a 14 02 mov (%rdx,%rax,1),%dl <-- trapping instruction 2d: 48 89 f8 mov %rdi,%rax 30: 83 e0 07 and $0x7,%eax
The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20261019/[email protected]
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
Quoted 7 lines · kernel test robot
Hello,
kernel test robot noticed "BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map" on:
commit:
8d75c338f0bcecaa6c9af67f86c176b67b6acf3e("sysctl: removeCONFIG_PROC_SYSCTL, it just mirrorsCONFIG_SYSCTL") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
I don't think that commit could cause such a problem. Please re-check.
Oleg.
Quoted 179 lines
[test failed on linux-next/master
b810335a0699d1810683796b0e3ffe8827ddfe6c]in testcase: boot
config: x86_64-randconfig-2300-20261014
compiler: gcc-14
test machine: qemu-system-x86_64 -enable-kvm -cpu Haswell-noTSX,enforce -smp 4 -m 16G(please refer to attached dmesg/kmsg for entire log/backtrace)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-lkp/[email protected][ 35.688893][ T1] BUG: KASAN: slab-use-after-free in touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] Read of size 8 at addr ffff8881259a2968 by task swapper/0/1 [ 35.688893][ T1] [ 35.688893][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.688893][ T1] Call Trace: [ 35.688893][ T1] <TASK> [ 35.688893][ T1] dump_stack_lvl (dump_stack.c:94 dump_stack.c:120) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_address_description+0x51/0x300 [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_report (kasan/report.c:482) [ 35.688893][ T1] ? kmem_cache_debug_flags (slab.h:503) [ 35.688893][ T1] ? fixup_red_left (slub.c:250) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] kasan_report (kasan/report.c:595) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] __flush_workqueue (workqueue.c:4079) [ 35.688893][ T1] ? xa_load (xarray.c:1621) [ 35.688893][ T1] ? tracer_preempt_on (trace/trace_irqsoff.c:686 (discriminator 1)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4043) [ 35.688893][ T1] ? __up_write (locking/rwsem.c:1422 (discriminator 4)) [ 35.688893][ T1] rds_ib_init (rds/ib.c:618) [ 35.688893][ T1] rds_rdma_init (rds/rdma_transport.c:300) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] ? __raw_spin_unlock_irqrestore (linux/spinlock_api_smp.h:179 (discriminator 4)) [ 35.688893][ T1] ? add_device_randomness (linux/spinlock.h:408 char/random.c:936) [ 35.688893][ T1] ? random_get_entropy (x86/include/asm/timex.h:11) [ 35.688893][ T1] ? sctp_diag_init (diag.c:?) [ 35.688893][ T1] ? rds_init (rds/af_rds.c:977) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] ? trace_initcall_finish_cb (main.c:1289 (discriminator 1)) [ 35.688893][ T1] ? __kmalloc_noprof (slub.c:5363 slub.c:5387) [ 35.688893][ T1] ? do_initcalls (linux/slab.h:973 linux/slab.h:1290 main.c:1418) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ? exit_thread (x86/kernel/process.c:138) [ 35.688893][ T1] ? __switch_to (x86/kernel/process_64.c:674) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] </TASK> [ 35.688893][ T1] [ 35.688893][ T1] Allocated by task 1 on cpu 1 at 35.661878s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] __kasan_kmalloc (kasan/common.c:415) [ 35.688893][ T1] __kmalloc_noprof (linux/kasan.h:263 slub.c:5362 slub.c:5387) [ 35.688893][ T1] __alloc_workqueue (workqueue.c:5827) [ 35.688893][ T1] alloc_workqueue_va (workqueue.c:5946) [ 35.688893][ T1] alloc_workqueue_noprof (workqueue.c:5962) [ 35.688893][ T1] rds_threads_init (rds/threads.c:259 (discriminator 2)) [ 35.688893][ T1] rds_init (rds/af_rds.c:961) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] Freed by task 0 on cpu 3 at 35.674112s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] kasan_save_free_info (kasan/generic.c:584) [ 35.688893][ T1] __kasan_slab_free (kasan/common.c:285) [ 35.688893][ T1] kfree (linux/kasan.h:235 slub.c:2705 slub.c:6405 slub.c:6720) [ 35.688893][ T1] rcu_do_batch (rcu/tree.c:2645) [ 35.688893][ T1] rcu_core (rcu/tree.c:2897) [ 35.688893][ T1] handle_softirqs (softirq.c:622) [ 35.688893][ T1] __irq_exit_rcu (softirq.c:496 softirq.c:735) [ 35.688893][ T1] irq_exit_rcu (softirq.c:752) [ 35.688893][ T1] sysvec_apic_timer_interrupt (x86/kernel/apic/apic.c:1062 (discriminator 35) x86/kernel/apic/apic.c:1062 (discriminator 35)) [ 35.688893][ T1] asm_sysvec_apic_timer_interrupt (x86/include/asm/idtentry.h:546) [ 35.688893][ T1] [ 35.688893][ T1] Last potentially related work creation: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_record_aux_stack (kasan/generic.c:556) [ 35.688893][ T1] __call_rcu_common+0xe6/0x480 [ 35.688893][ T1] kthread_worker_fn (kthread.c:1025) [ 35.688893][ T1] kthread (kthread.c:436) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the physical page: [ 35.688893][ T1] page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1259a0 [ 35.688893][ T1] head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 [ 35.688893][ T1] flags: 0x4000000000000040(head|zone=2) [ 35.688893][ T1] page_type: f5(slab) [ 35.688893][ T1] raw: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] raw: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] head: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000003 ffffea0004966801 00000000ffffffff 00000000ffffffff [ 35.688893][ T1] head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008 [ 35.688893][ T1] page dumped because: kasan: bad access detected [ 35.688893][ T1] [ 35.688893][ T1] Memory state around the buggy address: [ 35.688893][ T1] ffff8881259a2800: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2880: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] >ffff8881259a2900: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ^ [ 35.688893][ T1] ffff8881259a2980: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2a00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ================================================================== [ 35.984299][ T1] Disabling lock debugging due to kernel taint [ 35.987548][ T1] Oops: general protection fault, probably for non-canonical address 0xe0a07c0ec0000008: 0000 [#1] SMP KASAN [ 35.988482][ T1] KASAN: maybe wild-memory-access in range [0x0504007600000040-0x0504007600000047] [ 35.988482][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Tainted: G B 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.988482][ T1] Tainted: [B]=BAD_PAGE [ 35.988482][ T1] RIP: 0010:do_raw_spin_lock (locking/spinlock_debug.c:86 (discriminator 2) locking/spinlock_debug.c:115 (discriminator 2)) [ 35.988482][ T1] Code: 3b 58 81 49 8d 14 04 c7 02 f1 f1 f1 f1 c7 42 04 04 f3 f3 f3 65 48 8b 15 f0 53 f7 08 48 89 54 24 60 31 d2 48 89 fa 48 c1 ea 03 <8a> 14 02 48 89 f8 83 e0 07 83 c0 03 38 d0 7c 09 84 d2 74 05 e8 39 All code ======== 0: 3b 58 81 cmp -0x7f(%rax),%ebx 3: 49 8d 14 04 lea (%r12,%rax,1),%rdx 7: c7 02 f1 f1 f1 f1 movl $0xf1f1f1f1,(%rdx) d: c7 42 04 04 f3 f3 f3 movl $0xf3f3f304,0x4(%rdx) 14: 65 48 8b 15 f0 53 f7 mov %gs:0x8f753f0(%rip),%rdx # 0x8f7540c 1b: 08 1c: 48 89 54 24 60 mov %rdx,0x60(%rsp) 21: 31 d2 xor %edx,%edx 23: 48 89 fa mov %rdi,%rdx 26: 48 c1 ea 03 shr $0x3,%rdx 2a:* 8a 14 02 mov (%rdx,%rax,1),%dl <-- trapping instruction 2d: 48 89 f8 mov %rdi,%rax 30: 83 e0 07 and $0x7,%eaxThe kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20261019/[email protected]--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
Quoted 5 lines · Oleg Nesterov
Quoted 4 lines · kernel test robot
Hello,
Quoted 7 lines
Quoted 3 lines
kernel test robot noticed "BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map" on:
commit:
8d75c338f0bcecaa6c9af67f86c176b67b6acf3e("sysctl: removeCONFIG_PROC_SYSCTL, it just mirrorsCONFIG_SYSCTL") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git masterI don't think that commit could cause such a problem. Please re-check.
Oleg.
That commit might not be the one that introduces the UAF, but it might be the one that exposes it?
You have CONFIG_SYSCTL=n & CONFIG_RDS=y & CONFIG_RDS_RDMA=y. Therefore rds_sysctl_init() returns -ENOMEM and rds_ib_sysctl_init returns -ENOMEM.
If rds_ib_sysctl_init is called after a failed rds_sysctl_init, then rds_ib_sysctl_init would call a flush_workqueue on a rds_wq that has already been freed by the failed rds_sysctl_init. Right?
Please let me know if I missed anything.
Best
Quoted 181 lines
Quoted 180 lines
[test failed on linux-next/master
b810335a0699d1810683796b0e3ffe8827ddfe6c]in testcase: boot
config: x86_64-randconfig-2300-20261014
compiler: gcc-14
test machine: qemu-system-x86_64 -enable-kvm -cpu Haswell-noTSX,enforce -smp 4 -m 16G(please refer to attached dmesg/kmsg for entire log/backtrace)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-lkp/[email protected][ 35.688893][ T1] BUG: KASAN: slab-use-after-free in touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] Read of size 8 at addr ffff8881259a2968 by task swapper/0/1 [ 35.688893][ T1] [ 35.688893][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Not tainted 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.688893][ T1] Call Trace: [ 35.688893][ T1] <TASK> [ 35.688893][ T1] dump_stack_lvl (dump_stack.c:94 dump_stack.c:120) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_address_description+0x51/0x300 [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] print_report (kasan/report.c:482) [ 35.688893][ T1] ? kmem_cache_debug_flags (slab.h:503) [ 35.688893][ T1] ? fixup_red_left (slub.c:250) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] kasan_report (kasan/report.c:595) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] touch_wq_lockdep_map (workqueue.c:4031 (discriminator 1)) [ 35.688893][ T1] __flush_workqueue (workqueue.c:4079) [ 35.688893][ T1] ? xa_load (xarray.c:1621) [ 35.688893][ T1] ? tracer_preempt_on (trace/trace_irqsoff.c:686 (discriminator 1)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? preempt_count_sub (sched/core.c:6019) [ 35.688893][ T1] ? touch_wq_lockdep_map (workqueue.c:4043) [ 35.688893][ T1] ? __up_write (locking/rwsem.c:1422 (discriminator 4)) [ 35.688893][ T1] rds_ib_init (rds/ib.c:618) [ 35.688893][ T1] rds_rdma_init (rds/rdma_transport.c:300) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] ? __raw_spin_unlock_irqrestore (linux/spinlock_api_smp.h:179 (discriminator 4)) [ 35.688893][ T1] ? add_device_randomness (linux/spinlock.h:408 char/random.c:936) [ 35.688893][ T1] ? random_get_entropy (x86/include/asm/timex.h:11) [ 35.688893][ T1] ? sctp_diag_init (diag.c:?) [ 35.688893][ T1] ? rds_init (rds/af_rds.c:977) [ 35.688893][ T1] ? rcu_is_watching (rcu/tree.c:753 (discriminator 3)) [ 35.688893][ T1] ? rds_init (af_rds.c:?) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] ? trace_initcall_finish_cb (main.c:1289 (discriminator 1)) [ 35.688893][ T1] ? __kmalloc_noprof (slub.c:5363 slub.c:5387) [ 35.688893][ T1] ? do_initcalls (linux/slab.h:973 linux/slab.h:1290 main.c:1418) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ? exit_thread (x86/kernel/process.c:138) [ 35.688893][ T1] ? __switch_to (x86/kernel/process_64.c:674) [ 35.688893][ T1] ? rest_init (main.c:717) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] </TASK> [ 35.688893][ T1] [ 35.688893][ T1] Allocated by task 1 on cpu 1 at 35.661878s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] __kasan_kmalloc (kasan/common.c:415) [ 35.688893][ T1] __kmalloc_noprof (linux/kasan.h:263 slub.c:5362 slub.c:5387) [ 35.688893][ T1] __alloc_workqueue (workqueue.c:5827) [ 35.688893][ T1] alloc_workqueue_va (workqueue.c:5946) [ 35.688893][ T1] alloc_workqueue_noprof (workqueue.c:5962) [ 35.688893][ T1] rds_threads_init (rds/threads.c:259 (discriminator 2)) [ 35.688893][ T1] rds_init (rds/af_rds.c:961) [ 35.688893][ T1] do_one_initcall (main.c:1347) [ 35.688893][ T1] do_initcalls (main.c:1409 (discriminator 1) main.c:1425 (discriminator 1)) [ 35.688893][ T1] kernel_init_freeable (main.c:1445 main.c:1658) [ 35.688893][ T1] kernel_init (main.c:1548) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] Freed by task 0 on cpu 3 at 35.674112s: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_save_track (kasan/common.c:78) [ 35.688893][ T1] kasan_save_free_info (kasan/generic.c:584) [ 35.688893][ T1] __kasan_slab_free (kasan/common.c:285) [ 35.688893][ T1] kfree (linux/kasan.h:235 slub.c:2705 slub.c:6405 slub.c:6720) [ 35.688893][ T1] rcu_do_batch (rcu/tree.c:2645) [ 35.688893][ T1] rcu_core (rcu/tree.c:2897) [ 35.688893][ T1] handle_softirqs (softirq.c:622) [ 35.688893][ T1] __irq_exit_rcu (softirq.c:496 softirq.c:735) [ 35.688893][ T1] irq_exit_rcu (softirq.c:752) [ 35.688893][ T1] sysvec_apic_timer_interrupt (x86/kernel/apic/apic.c:1062 (discriminator 35) x86/kernel/apic/apic.c:1062 (discriminator 35)) [ 35.688893][ T1] asm_sysvec_apic_timer_interrupt (x86/include/asm/idtentry.h:546) [ 35.688893][ T1] [ 35.688893][ T1] Last potentially related work creation: [ 35.688893][ T1] stack_trace_save (stacktrace.c:122 (discriminator 1)) [ 35.688893][ T1] kasan_save_stack (kasan/common.c:57) [ 35.688893][ T1] kasan_record_aux_stack (kasan/generic.c:556) [ 35.688893][ T1] __call_rcu_common+0xe6/0x480 [ 35.688893][ T1] kthread_worker_fn (kthread.c:1025) [ 35.688893][ T1] kthread (kthread.c:436) [ 35.688893][ T1] ret_from_fork (x86/kernel/process.c:158) [ 35.688893][ T1] ret_from_fork_asm (x86/entry/entry_64.S:245) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address belongs to the object at ffff8881259a2800 [ 35.688893][ T1] which belongs to the cache kmalloc-1k of size 1024 [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] The buggy address is located 360 bytes inside of [ 35.688893][ T1] freed 1024-byte region [ffff8881259a2800, ffff8881259a2c00) [ 35.688893][ T1] [ 35.688893][ T1] The buggy address belongs to the physical page: [ 35.688893][ T1] page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1259a0 [ 35.688893][ T1] head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 [ 35.688893][ T1] flags: 0x4000000000000040(head|zone=2) [ 35.688893][ T1] page_type: f5(slab) [ 35.688893][ T1] raw: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] raw: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000040 ffff888100042dc0 dead000000000100 dead000000000122 [ 35.688893][ T1] head: 0000000000000000 0000000800100010 00000000f5000000 0000000000000000 [ 35.688893][ T1] head: 4000000000000003 ffffea0004966801 00000000ffffffff 00000000ffffffff [ 35.688893][ T1] head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008 [ 35.688893][ T1] page dumped because: kasan: bad access detected [ 35.688893][ T1] [ 35.688893][ T1] Memory state around the buggy address: [ 35.688893][ T1] ffff8881259a2800: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2880: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] >ffff8881259a2900: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ^ [ 35.688893][ T1] ffff8881259a2980: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ffff8881259a2a00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb [ 35.688893][ T1] ================================================================== [ 35.984299][ T1] Disabling lock debugging due to kernel taint [ 35.987548][ T1] Oops: general protection fault, probably for non-canonical address 0xe0a07c0ec0000008: 0000 [#1] SMP KASAN [ 35.988482][ T1] KASAN: maybe wild-memory-access in range [0x0504007600000040-0x0504007600000047] [ 35.988482][ T1] CPU: 1 UID: 0 PID: 1 Comm: swapper/0 Tainted: G B 7.2.0-rc2-00010-g8d75c338f0bc #1 PREEMPT [ 35.988482][ T1] Tainted: [B]=BAD_PAGE [ 35.988482][ T1] RIP: 0010:do_raw_spin_lock (locking/spinlock_debug.c:86 (discriminator 2) locking/spinlock_debug.c:115 (discriminator 2)) [ 35.988482][ T1] Code: 3b 58 81 49 8d 14 04 c7 02 f1 f1 f1 f1 c7 42 04 04 f3 f3 f3 65 48 8b 15 f0 53 f7 08 48 89 54 24 60 31 d2 48 89 fa 48 c1 ea 03 <8a> 14 02 48 89 f8 83 e0 07 83 c0 03 38 d0 7c 09 84 d2 74 05 e8 39 All code ======== 0: 3b 58 81 cmp -0x7f(%rax),%ebx 3: 49 8d 14 04 lea (%r12,%rax,1),%rdx 7: c7 02 f1 f1 f1 f1 movl $0xf1f1f1f1,(%rdx) d: c7 42 04 04 f3 f3 f3 movl $0xf3f3f304,0x4(%rdx) 14: 65 48 8b 15 f0 53 f7 mov %gs:0x8f753f0(%rip),%rdx # 0x8f7540c 1b: 08 1c: 48 89 54 24 60 mov %rdx,0x60(%rsp) 21: 31 d2 xor %edx,%edx 23: 48 89 fa mov %rdi,%rdx 26: 48 c1 ea 03 shr $0x3,%rdx 2a:* 8a 14 02 mov (%rdx,%rax,1),%dl <-- trapping instruction 2d: 48 89 f8 mov %rdi,%rax 30: 83 e0 07 and $0x7,%eaxThe kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20261019/[email protected]--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
Quoted 22 lines · Joel Granados
Quoted 5 lines · Oleg Nesterov
Quoted 4 lines · kernel test robot
Hello,
Quoted 7 lines
Quoted 3 lines
kernel test robot noticed "BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map" on:
commit:
8d75c338f0bcecaa6c9af67f86c176b67b6acf3e("sysctl: removeCONFIG_PROC_SYSCTL, it just mirrorsCONFIG_SYSCTL") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git masterI don't think that commit could cause such a problem. Please re-check.
Oleg.
That commit might not be the one that introduces the UAF, but it might be the one that exposes it?
You have
CONFIG_SYSCTL=n&CONFIG_RDS=y&CONFIG_RDS_RDMA=y. Thereforerds_sysctl_init()returns-ENOMEMandrds_ib_sysctl_initreturns-ENOMEM.
Quoted 5 lines
If
rds_ib_sysctl_initis called after a failedrds_sysctl_init, thenrds_ib_sysctl_initwould call aflush_workqueueon ards_wqthat has already been freed by the failedrds_sysctl_init. Right?Please let me know if I missed anything.
I think you are right...
rds_init() does
rds_threads_init()
rds_wq = create_singlethread_workqueue("krdsd");
rds_sysctl_init()
fails due to CONFIG_SYSCTL=n
out_threads:
destroy_workqueue(rds_wq);After that any access to rds_wq is wrong.
If rds_rdma_init() runs after that, rds_ib_init() will call rds_ib_unregister_client() -> flush_workqueue(rds_wq).
And this will equally happen before the blamed commit 8d75c338f0bc, just you need CONFIG_PROC_SYSCTL=n instead of CONFIG_SYSCTL=n.
But. rds_init() can also fail even if CONFIG_SYSCTL=y, after rds_sysctl_init(), so to me this looks like a bug in net/rds/.
In any case, I guess net/rds/ init paths should not fail if CONFIG_SYSCTL=n ?
Oleg.
Quoted 7 lines · Oleg Nesterov
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.In any case, I guess
net/rds/init paths should not fail ifCONFIG_SYSCTL=n?
So I think that something like the uncompiled/untested patch below makes sense anyway even if doesn't fix all the potential problems with rds_wq.
What do you think?
Oleg.
---
net/rds/ib_sysctl.c +22 -9
@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. *
@@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",
@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)
@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif
net/rds/sysctl.c +18 -5
@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;
@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",
@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);
@@ -107,3 +108,15 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_sysctl_exit (void ) +{ +} + +int rds_sysctl_init (void ) +{ +return 0 ; +} +#endif
Allison Henderson
maintainerQuoted 1 line · Oleg Nesterov
On 10/19, Oleg Nesterov wrote:
Hi Oleg, Joel,
Thanks for running this down. Agreed we shouldn't just fail to init because sysctl is off. Some comments below:
Quoted 4 lines
Quoted 4 lines
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.
Yes that's a separate bug, so I'll send a separate fix for that.
Quoted 61 lines
Quoted 3 lines
In any case, I guess
net/rds/init paths should not fail ifCONFIG_SYSCTL=n?So I think that something like the uncompiled/untested patch below makes sense anyway even if doesn't fix all the potential problems with
rds_wq.What do you think?
Oleg.
---net/rds/ib_sysctl.c +10
-9@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. * @@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)
net/rds/ib_sysctl.c int rds_ib_sysctl_init(void) · line 121
Quoted 3 lines
net/rds/ib_sysctl.c +0
-0@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ;return 0 ; }
The real rds_sysctl_init() also does the runtime setup of rds_sysctl_reconnect_min_jiffies (msecs_to_jiffies(1)). Without it, the reconnect back off is lost.
Also, if register_net_sysctl_sz() returns NULL in rds_tcp_init_net(), it fails as an -ENOMEM. So RDS/TCP would still fail to come up with CONFIG_SYSCTL=n. A simpler fix might be to just handle the NULL at the three call sites, e.g.:
rds_sysctl_reg_table =
register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table);
if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL))
return -ENOMEM;And same for rds_ib_sysctl_init() and rds_tcp_init_net(). The exit paths are ok, since unregister_net_sysctl_table() is already a no-op stub when sysctl is off.
Let me know if you want to respin or if you'd rather I send it. Either way is fine with me.
Thanks
Allison
Quoted 65 lines
net/rds/ib_sysctl.c +12
-0+ +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif net/rds/sysctl.c +18
-5@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);@@ -107,3 +108,15 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_sysctl_exit (void ) +{ +} + +int rds_sysctl_init (void ) +{ +return 0 ; +} +#endif
Hi Allison,
Quoted 4 lines · Allison Henderson
The real
rds_sysctl_init()also does the runtime setup ofrds_sysctl_reconnect_min_jiffies(msecs_to_jiffies(1)). Without it, the reconnect back off is lost.
Yes, thanks, I missed that.
Quoted 3 lines
Also, if
register_net_sysctl_sz()returns NULL inrds_tcp_init_net(), it fails as an-ENOMEM. So RDS/TCP would still fail to come up withCONFIG_SYSCTL=n.
Yes, I didn't even try to grep other files.
Hmm. I am m just curious how difficult it is to create a "complete" patch which puts all the sysctl-related code under ifdef(SYSCTL)... May be something like below but I don't understand net/rds, most probably it is wrong and I am not sure it makes a lot of sense. Plus I have not idea how to test it.
Quoted 7 lines
A simpler fix might be to just handle the NULL at the three call sites, e.g.:
rds_sysctl_reg_table = register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table); if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL)) return -ENOMEM;
Yes, I thought about a simpler patch too. Perhaps just
--- x/net/rds/Kconfig +++ x/net/rds/Kconfig @@ -2,7 +2,7 @@ config RDS tristate "The Reliable Datagram Sockets Protocol" - depends on INET + depends on INET && SYSCTL help The RDS (Reliable Datagram Sockets) protocol provides reliable, sequenced delivery of datagrams over Infiniband or TCP.
for the start? CONFIG_SYSCTL=n is not a common case.
Quoted 2 lines
Let me know if you want to respin or if you'd rather I send it. Either way is fine with me.
Allison, up to you. Whatever you prefer.
Oleg.
---
net/rds/ib_sysctl.c +22 -9
@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. *
@@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",
@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)
@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif
net/rds/sysctl.c +19 -5
@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;
@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",
@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);
@@ -107,3 +108,16 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +int rds_sysctl_init (void ) +{ + rds_sysctl_reconnect_min_jiffies =msecs_to_jiffies (1 ); +return 0 ; +} + +void rds_sysctl_exit (void ) +{ +} +#endif
net/rds/tcp.c +123 -107
281 lines not on this page · raw message
net/rds/tcp.h +2 -0
@@ -13,8 +13,10 @@ struct rds_tcp_net {struct socket *rds_tcp_listen_sock;struct socket *rds_tcp_accepted_sock;struct work_struct rds_tcp_accept_w; +#ifdef CONFIG_SYSCTL struct ctl_table_header *rds_tcp_sysctl;const struct ctl_table *ctl_table; +#endif int sndbuf_size;int rcvbuf_size; };
Quoted 43 lines · Oleg Nesterov
Hi Allison,
Quoted 4 lines · Allison Henderson
The real
rds_sysctl_init()also does the runtime setup ofrds_sysctl_reconnect_min_jiffies(msecs_to_jiffies(1)). Without it, the reconnect back off is lost.Yes, thanks, I missed that.
Quoted 3 lines
Also, if
register_net_sysctl_sz()returns NULL inrds_tcp_init_net(), it fails as an-ENOMEM. So RDS/TCP would still fail to come up withCONFIG_SYSCTL=n.Yes, I didn't even try to grep other files.
Hmm. I am m just curious how difficult it is to create a "complete" patch which puts all the sysctl-related code under ifdef(SYSCTL)... May be something like below but I don't understand net/rds, most probably it is wrong and I am not sure it makes a lot of sense. Plus I have not idea how to test it.
Quoted 7 lines
A simpler fix might be to just handle the NULL at the three call sites, e.g.:
rds_sysctl_reg_table = register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table); if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL)) return -ENOMEM;Yes, I thought about a simpler patch too. Perhaps just
--- x/net/rds/Kconfig +++ x/net/rds/Kconfig @@ -2,7 +2,7 @@ config RDS tristate "The Reliable Datagram Sockets Protocol" - depends on INET + depends on INET && SYSCTL help The RDS (Reliable Datagram Sockets) protocol provides reliable, sequenced delivery of datagrams over Infiniband or TCP.
Quoted 1 line
for the start?
CONFIG_SYSCTL=nis not a common case.
Indeed CONFIG_SYSCTL=n is not common. On the other hand why would RDS depend on SYSCTL being configured in? Shouldn't RDS just work with default values when CONFIG_SYSCTL=n?
Best
Quoted 434 lines
Quoted 3 lines
Let me know if you want to respin or if you'd rather I send it. Either way is fine with me.
Allison, up to you. Whatever you prefer.
Oleg.
---net/rds/ib_sysctl.c +22
-9@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. * @@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif net/rds/sysctl.c +19
-5@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);@@ -107,3 +108,16 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +int rds_sysctl_init (void ) +{ + rds_sysctl_reconnect_min_jiffies =msecs_to_jiffies (1 ); +return 0 ; +} + +void rds_sysctl_exit (void ) +{ +} +#endif net/rds/tcp.c +123
-107@@ -53,13 +53,71 @@ static atomic_t rds_tcp_unloading = ATOMIC_INIT(0);static struct kmem_cache *rds_tcp_conn_slab; +#ifdef CONFIG_SYSCTL +static int rds_tcp_min_sndbuf =SOCK_MIN_SNDBUF ; +static int rds_tcp_min_rcvbuf =SOCK_MIN_RCVBUF ; + +/* when sysctl is used to modify some kernel socket parameters,this +* function resets the RDS connections in that netns so that we can +* restart with new parameters. The assumption is that such reset +* events are few and far-between. +*/ +static void rds_tcp_sysctl_reset (struct net *net) +{ +struct rds_tcp_connection *tc, *_tc; + +spin_lock_irq (&rds_tcp_conn_lock); +list_for_each_entry_safe (tc, _tc, &rds_tcp_conn_list, t_tcp_node) { +struct net *c_net =read_pnet (&tc->t_cpath->cp_conn->c_net); + +if (net != c_net || !tc->t_sock) +continue ; + +/* reconnect with new parameters */ +rds_conn_path_drop (tc->t_cpath, false); + } +spin_unlock_irq (&rds_tcp_conn_lock); +} + +static int rds_tcp_skbuf_handler (struct rds_tcp_net *rtn, +const struct ctl_table *ctl,int write, +void *buffer,size_t *lenp,loff_t *fpos) +{ +int err; + + err =proc_dointvec_minmax (ctl, write, buffer, lenp, fpos); +if (err <0 ) { +pr_warn ("Invalid input. Must be >= %d\n" , + *(int *)(ctl->extra1)); +return err; + } + +if (write && rtn->rds_tcp_listen_sock && rtn->rds_tcp_listen_sock->sk) { +struct net *net =sock_net (rtn->rds_tcp_listen_sock->sk); + +rds_tcp_sysctl_reset (net); + } + +return 0 ; +} +static int rds_tcp_sndbuf_handler (const struct ctl_table *ctl,int write, -void *buffer,size_t *lenp,loff_t *fpos); +void *buffer,size_t *lenp,loff_t *fpos) +{ + struct rds_tcp_net *rtn = container_of(ctl->data, struct rds_tcp_net, + sndbuf_size); + +return rds_tcp_skbuf_handler (rtn, ctl, write, buffer, lenp, fpos); +} +static int rds_tcp_rcvbuf_handler (const struct ctl_table *ctl,int write, -void *buffer,size_t *lenp,loff_t *fpos); +void *buffer,size_t *lenp,loff_t *fpos) +{ + struct rds_tcp_net *rtn = container_of(ctl->data, struct rds_tcp_net, + rcvbuf_size); -static int rds_tcp_min_sndbuf =SOCK_MIN_SNDBUF ; -static int rds_tcp_min_rcvbuf =SOCK_MIN_RCVBUF ; +return rds_tcp_skbuf_handler (rtn, ctl, write, buffer, lenp, fpos); +} static struct ctl_table rds_tcp_sysctl_table[] = {#define RDS_TCP_SNDBUF 0@@ -82,6 +140,61 @@ static struct ctl_table rds_tcp_sysctl_table[] = { }, }; +static __net_init int rds_tcp_init_net_sysctl (struct net *net) +{ +struct rds_tcp_net *rtn =net_generic (net, rds_tcp_netid); +struct ctl_table *tbl; + +/* {snd, rcv}buf_size default to 0, which implies we let the +* stack pick the value, and permit auto-tuning of buffer size. +*/ +if (net == &init_net) { + tbl = rds_tcp_sysctl_table; + }else { + tbl =kmemdup (rds_tcp_sysctl_table, +sizeof (rds_tcp_sysctl_table),GFP_KERNEL ); +if (!tbl) { +pr_warn ("could not set allocate sysctl table\n" ); +return -ENOMEM ; + } + rtn->ctl_table = tbl; + } + tbl[RDS_TCP_SNDBUF ].data = &rtn->sndbuf_size; + tbl[RDS_TCP_RCVBUF ].data = &rtn->rcvbuf_size; + rtn->rds_tcp_sysctl =register_net_sysctl_sz (net,"net/rds/tcp" , tbl, +ARRAY_SIZE (rds_tcp_sysctl_table)); +if (!rtn->rds_tcp_sysctl) { +pr_warn ("could not register sysctl\n" ); +if (net != &init_net) +kfree (tbl); +return -ENOMEM ; + } + +return 0 ; +} + +static void __net_exit rds_tcp_exit_net_sysctl(struct net *net) +{ +struct rds_tcp_net *rtn =net_generic (net, rds_tcp_netid); + +if (rtn->rds_tcp_sysctl) { +unregister_net_sysctl_table (rtn->rds_tcp_sysctl); + rtn->rds_tcp_sysctl =NULL ; + } + +if (net != &init_net) +kfree (rtn->ctl_table); +} +#else +static inline int rds_tcp_init_net_sysctl (struct net *net) +{ +return 0 ; +} +static inline void rds_tcp_exit_net_sysctl (struct net *net) +{ +} +#endif +u32 rds_tcp_write_seq (struct rds_tcp_connection *tc) {/* seq# of the last byte of data in tcp send buffer */ @@ -612,36 +725,15 @@ void rds_tcp_accept_work(struct rds_tcp_net *rtn)static __net_init int rds_tcp_init_net (struct net *net) {struct rds_tcp_net *rtn =net_generic (net, rds_tcp_netid); -struct ctl_table *tbl;int err =0 ;memset (rtn,0 ,sizeof (*rtn));mutex_init (&rtn->rds_tcp_accept_lock); -/* {snd, rcv}buf_size default to 0, which implies we let the -* stack pick the value, and permit auto-tuning of buffer size. -*/ -if (net == &init_net) { - tbl = rds_tcp_sysctl_table; - }else { - tbl =kmemdup (rds_tcp_sysctl_table, -sizeof (rds_tcp_sysctl_table),GFP_KERNEL ); -if (!tbl) { -pr_warn ("could not set allocate sysctl table\n" ); -return -ENOMEM ; - } - rtn->ctl_table = tbl; - } - tbl[RDS_TCP_SNDBUF ].data = &rtn->sndbuf_size; - tbl[RDS_TCP_RCVBUF ].data = &rtn->rcvbuf_size; - rtn->rds_tcp_sysctl =register_net_sysctl_sz (net,"net/rds/tcp" , tbl, -ARRAY_SIZE (rds_tcp_sysctl_table)); -if (!rtn->rds_tcp_sysctl) { -pr_warn ("could not register sysctl\n" ); - err = -ENOMEM ; - goto fail; - } + err =rds_tcp_init_net_sysctl (net); +if (err) +return err;#if IS_ENABLED (CONFIG_IPV6 ) rtn->rds_tcp_listen_sock =rds_tcp_listen_init (net, true);@@ -656,21 +748,14 @@ static __net_init int rds_tcp_init_net(struct net *net) rtn->rds_tcp_listen_sock =rds_tcp_listen_init (net, false);if (!rtn->rds_tcp_listen_sock) {#endif -unregister_net_sysctl_table (rtn->rds_tcp_sysctl); - rtn->rds_tcp_sysctl =NULL ; - err = -EAFNOSUPPORT ; - goto fail; +rds_tcp_exit_net_sysctl (net); +return -EAFNOSUPPORT ;#if IS_ENABLED (CONFIG_IPV6 ) }#endif }INIT_WORK (&rtn->rds_tcp_accept_w, rds_tcp_accept_worker);return 0 ; - -fail: -if (net != &init_net) -kfree (tbl); -return err; }static void rds_tcp_kill_sock (struct net *net)@@ -704,14 +789,7 @@ static void rds_tcp_kill_sock(struct net *net) static void __net_exit rds_tcp_exit_net(struct net *net) { -struct rds_tcp_net *rtn =net_generic (net, rds_tcp_netid); - -if (rtn->rds_tcp_sysctl) -unregister_net_sysctl_table (rtn->rds_tcp_sysctl); - -if (net != &init_net) -kfree (rtn->ctl_table); - +rds_tcp_exit_net_sysctl (net);rds_tcp_kill_sock (net); }@@ -733,68 +811,6 @@ void *rds_tcp_listen_sock_def_readable(struct net *net)return lsock->sk->sk_user_data; } -/* when sysctl is used to modify some kernel socket parameters,this -* function resets the RDS connections in that netns so that we can -* restart with new parameters. The assumption is that such reset -* events are few and far-between. -*/ -static void rds_tcp_sysctl_reset (struct net *net) -{ -struct rds_tcp_connection *tc, *_tc; - -spin_lock_irq (&rds_tcp_conn_lock); -list_for_each_entry_safe (tc, _tc, &rds_tcp_conn_list, t_tcp_node) { -struct net *c_net =read_pnet (&tc->t_cpath->cp_conn->c_net); - -if (net != c_net || !tc->t_sock) -continue ; - -/* reconnect with new parameters */ -rds_conn_path_drop (tc->t_cpath, false); - } -spin_unlock_irq (&rds_tcp_conn_lock); -} - -static int rds_tcp_skbuf_handler (struct rds_tcp_net *rtn, -const struct ctl_table *ctl,int write, -void *buffer,size_t *lenp,loff_t *fpos) -{ -int err; - - err =proc_dointvec_minmax (ctl, write, buffer, lenp, fpos); -if (err <0 ) { -pr_warn ("Invalid input. Must be >= %d\n" , - *(int *)(ctl->extra1)); -return err; - } - -if (write && rtn->rds_tcp_listen_sock && rtn->rds_tcp_listen_sock->sk) { -struct net *net =sock_net (rtn->rds_tcp_listen_sock->sk); - -rds_tcp_sysctl_reset (net); - } - -return 0 ; -} - -static int rds_tcp_sndbuf_handler (const struct ctl_table *ctl,int write, -void *buffer,size_t *lenp,loff_t *fpos) -{ - struct rds_tcp_net *rtn = container_of(ctl->data, struct rds_tcp_net, - sndbuf_size); - -return rds_tcp_skbuf_handler (rtn, ctl, write, buffer, lenp, fpos); -} - -static int rds_tcp_rcvbuf_handler (const struct ctl_table *ctl,int write, -void *buffer,size_t *lenp,loff_t *fpos) -{ - struct rds_tcp_net *rtn = container_of(ctl->data, struct rds_tcp_net, - rcvbuf_size); - -return rds_tcp_skbuf_handler (rtn, ctl, write, buffer, lenp, fpos); -} -static void rds_tcp_exit (void ) {rds_tcp_set_unloading ();net/rds/tcp.h +2
-0@@ -13,8 +13,10 @@ struct rds_tcp_net {struct socket *rds_tcp_listen_sock;struct socket *rds_tcp_accepted_sock;struct work_struct rds_tcp_accept_w; +#ifdef CONFIG_SYSCTL struct ctl_table_header *rds_tcp_sysctl;const struct ctl_table *ctl_table; +#endif int sndbuf_size;int rcvbuf_size; };
Quoted 90 lines · Allison Henderson
Quoted 1 line · Oleg Nesterov
On 10/19, Oleg Nesterov wrote:
Hi Oleg, Joel,
Thanks for running this down. Agreed we shouldn't just fail to init because sysctl is off. Some comments below:
Quoted 4 lines
Quoted 4 lines
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.Yes that's a separate bug, so I'll send a separate fix for that.
Quoted 61 lines
Quoted 3 lines
In any case, I guess
net/rds/init paths should not fail ifCONFIG_SYSCTL=n?So I think that something like the uncompiled/untested patch below makes sense anyway even if doesn't fix all the potential problems with
rds_wq.What do you think?
Oleg.
---net/rds/ib_sysctl.c +10
-9@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. * @@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)net/rds/ib_sysctl.c int rds_ib_sysctl_init(void) · line 121
Quoted 3 lines
net/rds/ib_sysctl.c +0
-0@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ;return 0 ; }The real
rds_sysctl_init()also does the runtime setup ofrds_sysctl_reconnect_min_jiffies(msecs_to_jiffies(1)). Without it, the reconnect back off is lost.Also, if
register_net_sysctl_sz()returns NULL inrds_tcp_init_net(), it fails as an-ENOMEM. So RDS/TCP would still fail to come up withCONFIG_SYSCTL=n. A simpler fix might be to just handle the NULL at the three call sites, e.g.:
Quoted 5 lines
rds_sysctl_reg_table =
register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table);
if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL))
return -ENOMEM;Remember to NULL check rds_sysctl_reg_table in rds_sysctl_exit(). You might hit another UAF on exit if you don't.
Quoted 76 lines
And same for
rds_ib_sysctl_init()andrds_tcp_init_net(). The exit paths are ok, sinceunregister_net_sysctl_table()is already a no-op stub when sysctl is off.Let me know if you want to respin or if you'd rather I send it. Either way is fine with me.
Thanks
AllisonQuoted 65 lines
net/rds/ib_sysctl.c +12
-0+ +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif net/rds/sysctl.c +18
-5@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);@@ -107,3 +108,15 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_sysctl_exit (void ) +{ +} + +int rds_sysctl_init (void ) +{ +return 0 ; +} +#endif
Quoted 6 lines · Joel Granados
Quoted 4 lines · Allison Henderson
CONFIG_SYSCTL=n. A simpler fix might be to just handle the NULL at the three call sites, e.g.:
Quoted 8 lines
Quoted 5 lines
rds_sysctl_reg_table = register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table); if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL)) return -ENOMEM;Remember to NULL check
rds_sysctl_reg_tableinrds_sysctl_exit(). You might hit another UAF on exit if you don't.
Why? unregister_sysctl_table(header) does nothing if header == NULL.
Oleg.
Quoted 13 lines · Oleg Nesterov
Quoted 12 lines · Joel Granados
Quoted 9 lines · Allison Henderson
CONFIG_SYSCTL=n. A simpler fix might be to just handle the NULL at the three call sites, e.g.:rds_sysctl_reg_table = register_net_sysctl(&init_net, "net/rds", rds_sysctl_rds_table); if (!rds_sysctl_reg_table && IS_ENABLED(CONFIG_SYSCTL)) return -ENOMEM;
Quoted 4 lines
Quoted 2 lines
Remember to NULL check
rds_sysctl_reg_tableinrds_sysctl_exit(). You might hit another UAF on exit if you don't.Why?
unregister_sysctl_table(header)does nothing if header == NULL.
Nice catch! You can forgo the check as it is the first thing that happens in unregister_sysctl_table.
Best
rds_wq UAF
In reply to Allison Henderson(change subject)
Quoted 8 lines · Allison Henderson
Quoted 4 lines · Oleg Nesterov
Quoted 4 lines
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.Yes that's a separate bug, so I'll send a separate fix for that.
Let me try to summarize.
If rds_init() fails after rds_threads_init() for any reason, rds_threads_exit() will destroy/free rds_wq.
If rds_rdma_init() is called after that, and rds_ib_init() fails after rds_ib_mr_init(), rds_ib_unregister_client() will try to flush the already freed rds_wq.
But we have
module_init(rds_init); // net/rds/af_rds.c module_init(rds_rdma_init); // net/rds/rdma_transport.c
and IIUC the order of these initcalls is not defined. So if rds_rdma_init() -> rds_ib_init() is called before rds_init(), it can do flush_workqueue(rds_wq) when rds_wq == NULL on failure.
With or without CONFIG_SYSCTL. Right?
But it seems that rds_ib_init() doesn't need to flush rds_wq on failure, so how about the patch below? Of course, I have no idea if net/rds has more similar issues with rds_wq.
Oleg.
---
net/rds/ib.c +1 -1
@@ -619,7 +619,7 @@ int rds_ib_init(void) out_sysctl:rds_ib_sysctl_exit (); out_ibreg: -rds_ib_unregister_client (); +ib_unregister_client (&rds_ib_client); out_mr_exit:rds_ib_mr_exit (); out:
Allison Henderson
maintainerRe: rds_wq UAF
Quoted 21 lines · Oleg Nesterov
(change subject)
Quoted 8 lines · Allison Henderson
Quoted 4 lines · Oleg Nesterov
Quoted 4 lines
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.Yes that's a separate bug, so I'll send a separate fix for that.
Let me try to summarize.
If
rds_init()fails afterrds_threads_init()for any reason,rds_threads_exit()will destroy/freerds_wq.If
rds_rdma_init()is called after that, andrds_ib_init()fails afterrds_ib_mr_init(),rds_ib_unregister_client()will try to flush the already freedrds_wq.
Quoted 8 lines
But we have
module_init(rds_init); // net/rds/af_rds.c module_init(rds_rdma_init); // net/rds/rdma_transport.cand IIUC the order of these initcalls is not defined. So if
rds_rdma_init()->rds_ib_init()is called beforerds_init(), it can doflush_workqueue(rds_wq)whenrds_wq== NULL on failure.
It is sort of defined. When built as modules the order is enforced by symbol dependencies. rds_rdma and rds_tcp need symbols exported by rds (rds_wq itself is one of them), so they can't be loaded unless rds_init() succeeded. So the rds_wq == NULL case can't happen.
What can happen is what was reported: built in, rds_init() fails, do_initcalls() keeps going, and rds_rdma_init()/rds_tcp_init() run against a core that has already been torn down.
Quoted 15 lines
With or without
CONFIG_SYSCTL. Right?But it seems that
rds_ib_init()doesn't need to flushrds_wqon failure, so how about the patch below? Of course, I have no idea if net/rds has more similar issues withrds_wq.Oleg.
---net/rds/ib.c +0
-0
net/rds/ib.c int rds_ib_init(void) · line 622
Quoted 5 lines
net/rds/ib.c +1
-1@@ -619,7 +619,7 @@ int rds_ib_init(void) out_sysctl:rds_ib_sysctl_exit (); out_ibreg: -rds_ib_unregister_client (); +ib_unregister_client (&rds_ib_client);
This works for when there are no IB devices, as in the report, but the flush is what ensures rds_ib_dev_free() has finished before rds_rdma.ko's text goes away.
If there is a device, ib_register_client() has already called rds_ib_add_one(). Then ib_unregister_client() calls rds_ib_remove_one(), which drops the last reference. And then rds_ib_dev_put() does queue_work(rds_wq, ...) on the freed workqueue anyway.
So I think the fix belongs at the transport init: if the core didn't come up, rds_rdma_init() and rds_tcp_init() should return early, before they register anything. I can send that along with the sysctl change tomorrow.
Thanks for working on this!
Allison
Quoted 4 lines
net/rds/ib.c +0
-0out_mr_exit:rds_ib_mr_exit (); out:
Re: rds_wq UAF
Quoted 17 lines · Allison Henderson
Quoted 15 lines · Oleg Nesterov
Let me try to summarize.
If
rds_init()fails afterrds_threads_init()for any reason,rds_threads_exit()will destroy/freerds_wq.If
rds_rdma_init()is called after that, andrds_ib_init()fails afterrds_ib_mr_init(),rds_ib_unregister_client()will try to flush the already freedrds_wq.But we have
module_init(rds_init); // net/rds/af_rds.c module_init(rds_rdma_init); // net/rds/rdma_transport.c
Quoted 8 lines
Quoted 3 lines
and IIUC the order of these initcalls is not defined. So if
rds_rdma_init()->rds_ib_init()is called beforerds_init(), it can doflush_workqueue(rds_wq)whenrds_wq== NULL on failure.It is sort of defined. When built as modules the order is enforced by symbol dependencies.
rds_rdmaandrds_tcpneed symbols exported by rds (rds_wqitself is one of them), so they can't be loaded unlessrds_init()succeeded.
Yes, somehow I forgot that rds_rdma depends on rds ...
Quoted 1 line
So the
rds_wq== NULL case can't happen.
Well, in theory rds_threads_init() -> create_singlethread_workqueue() can fail too and leave rds_wq = NULL. Nevermind.
Quoted 4 lines
So I think the fix belongs at the transport init: if the core didn't come up,
rds_rdma_init()andrds_tcp_init()should return early, before they register anything. I can send that along with the sysctl change tomorrow.
OK, thanks.
Oleg.
[email protected]
AI reviewRe: rds_wq UAF
In reply to Oleg NesterovThank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [Critical] rds: ib: skipping flush_workqueue() in rds_ib_init() causes UAF on error - [Critical] rds…
Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [Critical] rds: ib: skipping flush_workqueue() in rds_ib_init() causes UAF on error
- [Critical] rds: rdma: rds_rdma_init() still accesses invalid rds_wq when rds_init() fails --
commit 1ca141aafc7f21d383b77362aeece7748105d3b9
Author: Oleg Nesterov <[email protected]>
rds_wq UAF
This commit replaces rds_ib_unregister_client() with ib_unregister_client(&rds_ib_client) on the rds_ib_init() error path to avoid flushing a NULL or already-destroyed rds_wq.
net/rds/ib.c int rds_ib_init(void) · line 622
Quoted 10 lines
net/rds/ib.c +1
-1@@ -619,7 +619,7 @@ int rds_ib_init(void) out_sysctl:rds_ib_sysctl_exit (); out_ibreg: -rds_ib_unregister_client (); +ib_unregister_client (&rds_ib_client);
[Severity: Critical]
Could skipping flush_workqueue(rds_wq) here cause a use-after-free of rds_ib_mr_wq or the rds_rdma module?
In rds_ib_init(), ib_register_client(&rds_ib_client) synchronously calls rds_ib_add_one() for every registered InfiniBand device.
If rds_ib_add_one() fails after allocating rds_ibdev (for example at ib_alloc_pd() or rds_ib_create_mr_pool()), its put_dev path drops the reference count to zero and queues rds_ib_dev_free() onto rds_wq:
net/rds/ib.c:rds_ib_dev_put()
if (refcount_dec_and_test(&rds_ibdev->refcount)) queue_work(rds_wq, &rds_ibdev->free_work);
Similarly, if rds_ib_add_one() succeeds for one or more devices and a later step in rds_ib_init() fails (rds_ib_sysctl_init() or rds_ib_recv_init()), ib_unregister_client(&rds_ib_client) calls rds_ib_remove_one() for each added device, which also drops the final reference and queues rds_ibdev->free_work (rds_ib_dev_free()) onto rds_wq.
Commit 24fa163a4bae74 added flush_workqueue(rds_wq) in rds_ib_unregister_client() specifically to wait for rds_ib_dev_free() to complete:
static void rds_ib_unregister_client(void)
{
ib_unregister_client(&rds_ib_client);
/* wait for rds_ib_dev_free() to complete */
flush_workqueue(rds_wq);
}Without that flush, rds_ib_init() proceeds directly to out_mr_exit and calls rds_ib_mr_exit() -> destroy_workqueue(rds_ib_mr_wq) while rds_ib_dev_free() is still pending or running on rds_wq and about to tear down rds_ibdev->mr_8k_pool and rds_ibdev->mr_1m_pool (which flushes and cancels delayed work on rds_ib_mr_wq).
In addition, when CONFIG_RDS_RDMA=m, returning from rds_ib_init() causes the module loader to unmap rds_rdma.ko while rds_ib_dev_free() is still queued or executing on rds_wq.
[Severity: Critical]
Also, if rds_init() has failed and destroyed rds_wq (or has not run yet so rds_wq is NULL), doesn't rds_rdma_init() still access the invalid rds_wq even with this change?
1. During rds_ib_init(), if an IB device is present and either rds_ib_add_one() fails at put_dev or ib_unregister_client(&rds_ib_client) calls rds_ib_remove_one(), rds_ib_dev_put() will still call queue_work(rds_wq, &rds_ibdev->free_work) on the destroyed or NULL rds_wq.
2. In rds_rdma_init(), if rds_ib_init() succeeds and rds_rdma_listen_init() fails, rds_ib_exit() is called, which still calls rds_ib_unregister_client() -> flush_workqueue(rds_wq):
net/rds/rdma_transport.c:rds_rdma_init()
ret = rds_ib_init(); if (ret) goto out; ret = rds_rdma_listen_init(); if (ret) rds_ib_exit();
3. If rds_rdma_init() (or rds_tcp_init()) succeeds after rds_init() has failed and torn down RDS core resources in rds_threads_exit(), the transport remains registered and active callbacks will queue work onto the destroyed rds_wq.
Quoted 3 lines
net/rds/ib.c +0
-0out_mr_exit:rds_ib_mr_exit (); out:
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Quoted 6 lines · Oleg Nesterov
Quoted 5 lines · Oleg Nesterov
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.
Quoted 7 lines
Quoted 2 lines
In any case, I guess
net/rds/init paths should not fail ifCONFIG_SYSCTL=n?So I think that something like the uncompiled/untested patch below makes sense anyway even if doesn't fix all the potential problems with
rds_wq.What do you think?
That is a good option. I also did not compile not test it, but guarding the error inside an ifdef would work and is a pattern that is used elsewhere.
You can also do this (also untested and not compiled):
net/rds/ib_sysctl.c +1 -1
@@ -114,7 +114,7 @@ void rds_ib_sysctl_exit(void)int rds_ib_sysctl_init (void ) { rds_ib_sysctl_hdr =register_net_sysctl (&init_net,"net/rds/ib" , rds_ib_sysctl_table); -if (!rds_ib_sysctl_hdr) +if (!rds_ib_sysctl_hdr &&IS_ENABLED (CONFIG_SYSCTL ))return -ENOMEM ;return 0 ; }
net/rds/sysctl.c +3 -2
@@ -93,7 +93,8 @@ static struct ctl_table rds_sysctl_rds_table[] = {void rds_sysctl_exit (void ) { -unregister_net_sysctl_table (rds_sysctl_reg_table); +if (rds_sysctl_reg_table) +unregister_net_sysctl_table (rds_sysctl_reg_table); }int rds_sysctl_init (void )
@@ -103,7 +104,7 @@ int rds_sysctl_init(void) rds_sysctl_reg_table =register_net_sysctl (&init_net,"net/rds" , rds_sysctl_rds_table); -if (!rds_sysctl_reg_table) +if (!rds_sysctl_reg_table &&IS_ENABLED (CONFIG_SYSCTL ))return -ENOMEM ;return 0 ; }
Both are ok IMO.
Best
Quoted 121 lines
Oleg.
---net/rds/ib_sysctl.c +22
-9@@ -36,19 +36,10 @@#include "ib.h" -static struct ctl_table_header *rds_ib_sysctl_hdr; -unsigned long rds_ib_sysctl_max_send_wr =RDS_IB_DEFAULT_SEND_WR ;unsigned long rds_ib_sysctl_max_recv_wr =RDS_IB_DEFAULT_RECV_WR ;unsigned long rds_ib_sysctl_max_recv_allocation = (128 *1024 *1024 ) /RDS_FRAG_SIZE ; -static unsigned long rds_ib_sysctl_max_wr_min =1 ; -/* hardware will fail CQ creation long before this */ -static unsigned long rds_ib_sysctl_max_wr_max = (u32 )~0 ; -unsigned long rds_ib_sysctl_max_unsig_wrs =16 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_min =1 ; -static unsigned long rds_ib_sysctl_max_unsig_wr_max =64 ; -/* * This sysctl does nothing. * @@ -61,6 +52,14 @@ static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64;*/ unsigned int rds_ib_sysctl_flow_control =0 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_ib_sysctl_max_wr_min = 1; +/* hardware will fail CQ creation long before this */ +static unsigned long rds_ib_sysctl_max_wr_max = (u32)~0; + +static unsigned long rds_ib_sysctl_max_unsig_wr_min = 1; +static unsigned long rds_ib_sysctl_max_unsig_wr_max = 64; + static struct ctl_table rds_ib_sysctl_table[] = { { .procname = "max_send_wr",@@ -105,6 +104,8 @@ static struct ctl_table rds_ib_sysctl_table[] = { }, }; +static struct ctl_table_header *rds_ib_sysctl_hdr; +void rds_ib_sysctl_exit (void ) {if (rds_ib_sysctl_hdr)@@ -118,3 +119,15 @@ int rds_ib_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_ib_sysctl_exit (void ) +{ +} + +int rds_ib_sysctl_init (void ) +{ +return 0 ; +} +#endif net/rds/sysctl.c +18
-5@@ -36,11 +36,6 @@#include "rds.h" -static struct ctl_table_header *rds_sysctl_reg_table; - -static unsigned long rds_sysctl_reconnect_min =1 ; -static unsigned long rds_sysctl_reconnect_max = ~0UL ; -unsigned long rds_sysctl_reconnect_min_jiffies;unsigned long rds_sysctl_reconnect_max_jiffies =HZ ;@@ -49,6 +44,10 @@ unsigned int rds_sysctl_max_unacked_bytes = (16 << 20);unsigned int rds_sysctl_ping_enable =1 ; +#ifdef CONFIG_SYSCTL +static unsigned long rds_sysctl_reconnect_min = 1; +static unsigned long rds_sysctl_reconnect_max = ~0UL; + static struct ctl_table rds_sysctl_rds_table[] = { { .procname = "reconnect_min_delay_ms",@@ -91,6 +90,8 @@ static struct ctl_table rds_sysctl_rds_table[] = { }, }; +static struct ctl_table_header *rds_sysctl_reg_table; +void rds_sysctl_exit (void ) {unregister_net_sysctl_table (rds_sysctl_reg_table);@@ -107,3 +108,15 @@ int rds_sysctl_init(void)return -ENOMEM ; return 0; } + +#else + +void rds_sysctl_exit (void ) +{ +} + +int rds_sysctl_init (void ) +{ +return 0 ; +} +#endif
Quoted 21 lines · Joel Granados
You can also do this (also untested and not compiled):
net/rds/ib_sysctl.c +1
-1@@ -114,7 +114,7 @@ void rds_ib_sysctl_exit(void)int rds_ib_sysctl_init (void ) { rds_ib_sysctl_hdr =register_net_sysctl (&init_net,"net/rds/ib" , rds_ib_sysctl_table); -if (!rds_ib_sysctl_hdr) +if (!rds_ib_sysctl_hdr &&IS_ENABLED (CONFIG_SYSCTL ))return -ENOMEM ;return 0 ; }net/rds/sysctl.c +0
-0
net/rds/sysctl.c static struct ctl_table rds_sysctl_rds_table[] = { · line 97
Quoted 6 lines
net/rds/sysctl.c +2
-1@@ -93,7 +93,8 @@ static struct ctl_table rds_sysctl_rds_table[] = {void rds_sysctl_exit (void ) { -unregister_net_sysctl_table (rds_sysctl_reg_table); +if (rds_sysctl_reg_table) +unregister_net_sysctl_table (rds_sysctl_reg_table);
Yes, this is what Allison suggested too. But he also pointed out that the changes in _exit() are not strictly needed, unregister_net_sysctl_table is nop if !CONFIG_SYSCTL. Plus we also need to change rds_tcp_init_net() the same way, see register_net_sysctl_sz().
So, let me repeat, I agree in advance with anything you and Allison prefer.
Oleg.
Quoted 51 lines · Oleg Nesterov
Quoted 22 lines · Joel Granados
Quoted 5 lines · Oleg Nesterov
Quoted 4 lines · kernel test robot
Hello,
Quoted 7 lines
Quoted 3 lines
kernel test robot noticed "BUG:KASAN:slab-use-after-free_in_touch_wq_lockdep_map" on:
commit:
8d75c338f0bcecaa6c9af67f86c176b67b6acf3e("sysctl: removeCONFIG_PROC_SYSCTL, it just mirrorsCONFIG_SYSCTL") https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git masterI don't think that commit could cause such a problem. Please re-check.
Oleg.
That commit might not be the one that introduces the UAF, but it might be the one that exposes it?
You have
CONFIG_SYSCTL=n&CONFIG_RDS=y&CONFIG_RDS_RDMA=y. Thereforerds_sysctl_init()returns-ENOMEMandrds_ib_sysctl_initreturns-ENOMEM.Quoted 5 lines
If
rds_ib_sysctl_initis called after a failedrds_sysctl_init, thenrds_ib_sysctl_initwould call aflush_workqueueon ards_wqthat has already been freed by the failedrds_sysctl_init. Right?Please let me know if I missed anything.
I think you are right...
rds_init()doesrds_threads_init() rds_wq = create_singlethread_workqueue("krdsd"); rds_sysctl_init() fails due to CONFIG_SYSCTL=n out_threads: destroy_workqueue(rds_wq);After that any access to
rds_wqis wrong.If
rds_rdma_init()runs after that,rds_ib_init()will callrds_ib_unregister_client()->flush_workqueue(rds_wq).And this will equally happen before the blamed commit
8d75c338f0bc, just you needCONFIG_PROC_SYSCTL=ninstead ofCONFIG_SYSCTL=n.
Quoted 6 lines
But.
rds_init()can also fail even ifCONFIG_SYSCTL=y, afterrds_sysctl_init(), so to me this looks like a bug innet/rds/.In any case, I guess
net/rds/init paths should not fail ifCONFIG_SYSCTL=n?
That is up to net/rds.
The question is: do you have to fail if a sysctl registration fails? Or another way to ask it is: Can't it (net/rds or others) run while not being tuned or queried through sysctl?
If the answer to these questions is that the sysctl path is necessary, then the failure is reasonable. But if they can work without the sysctl, the failure is an overreaction (IMO).
Best
Quoted 3 lines
Oleg.