4.19 wifi new #50
Merged
4.19 wifi new #50
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
frank-w
pushed a commit
that referenced
this pull request
Dec 17, 2018
Increase kasan instrumented kernel stack size from 32k to 64k. Other architectures seems to get away with just doubling kernel stack size under kasan, but on s390 this appears to be not enough due to bigger frame size. The particular pain point is kasan inlined checks (CONFIG_KASAN_INLINE vs CONFIG_KASAN_OUTLINE). With inlined checks one particular case hitting stack overflow is fs sync on xfs filesystem: #0 [9a0681e8] 704 bytes check_usage at 34b1fc #1 [9a0684a8] 432 bytes check_usage at 34c710 #2 [9a068658] 1048 bytes validate_chain at 35044a #3 [9a068a70] 312 bytes __lock_acquire at 3559fe #4 [9a068ba8] 440 bytes lock_acquire at 3576ee #5 [9a068d60] 104 bytes _raw_spin_lock at 21b44e0 #6 [9a068dc8] 1992 bytes enqueue_entity at 2dbf72 #7 [9a069590] 1496 bytes enqueue_task_fair at 2df5f0 #8 [9a069b68] 64 bytes ttwu_do_activate at 28f438 #9 [9a069ba8] 552 bytes try_to_wake_up at 298c4c #10 [9a069dd0] 168 bytes wake_up_worker at 23f97c #11 [9a069e78] 200 bytes insert_work at 23fc2e #12 [9a069f40] 648 bytes __queue_work at 2487c0 #13 [9a06a1c8] 200 bytes __queue_delayed_work at 24db28 #14 [9a06a290] 248 bytes mod_delayed_work_on at 24de84 #15 [9a06a388] 24 bytes kblockd_mod_delayed_work_on at 153e2a0 #16 [9a06a3a0] 288 bytes __blk_mq_delay_run_hw_queue at 158168c #17 [9a06a4c0] 192 bytes blk_mq_run_hw_queue at 1581a3c #18 [9a06a580] 184 bytes blk_mq_sched_insert_requests at 15a2192 #19 [9a06a638] 1024 bytes blk_mq_flush_plug_list at 1590f3a #20 [9a06aa38] 704 bytes blk_flush_plug_list at 1555028 #21 [9a06acf8] 320 bytes schedule at 219e476 #22 [9a06ae38] 760 bytes schedule_timeout at 21b0aac #23 [9a06b130] 408 bytes wait_for_common at 21a1706 #24 [9a06b2c8] 360 bytes xfs_buf_iowait at fa1540 #25 [9a06b430] 256 bytes __xfs_buf_submit at fadae6 #26 [9a06b530] 264 bytes xfs_buf_read_map at fae3f6 #27 [9a06b638] 656 bytes xfs_trans_read_buf_map at 10ac9a8 #28 [9a06b8c8] 304 bytes xfs_btree_kill_root at e72426 #29 [9a06b9f8] 288 bytes xfs_btree_lookup_get_block at e7bc5e #30 [9a06bb18] 624 bytes xfs_btree_lookup at e7e1a6 #31 [9a06bd88] 2664 bytes xfs_alloc_ag_vextent_near at dfa070 #32 [9a06c7f0] 144 bytes xfs_alloc_ag_vextent at dff3ca #33 [9a06c880] 1128 bytes xfs_alloc_vextent at e05fce #34 [9a06cce8] 584 bytes xfs_bmap_btalloc at e58342 #35 [9a06cf30] 1336 bytes xfs_bmapi_write at e618de #36 [9a06d468] 776 bytes xfs_iomap_write_allocate at ff678e #37 [9a06d770] 720 bytes xfs_map_blocks at f82af8 #38 [9a06da40] 928 bytes xfs_writepage_map at f83cd6 #39 [9a06dde0] 320 bytes xfs_do_writepage at f85872 #40 [9a06df20] 1320 bytes write_cache_pages at 73dfe8 #41 [9a06e448] 208 bytes xfs_vm_writepages at f7f892 #42 [9a06e518] 88 bytes do_writepages at 73fe6a #43 [9a06e570] 872 bytes __writeback_single_inode at a20cb6 #44 [9a06e8d8] 664 bytes writeback_sb_inodes at a23be2 #45 [9a06eb70] 296 bytes __writeback_inodes_wb at a242e0 #46 [9a06ec98] 928 bytes wb_writeback at a2500e #47 [9a06f038] 848 bytes wb_do_writeback at a260ae #48 [9a06f388] 536 bytes wb_workfn at a28228 #49 [9a06f5a0] 1088 bytes process_one_work at 24a234 #50 [9a06f9e0] 1120 bytes worker_thread at 24ba26 #51 [9a06fe40] 104 bytes kthread at 26545a #52 [9a06fea8] kernel_thread_starter at 21b6b62 To be able to increase the stack size to 64k reuse LLILL instruction in __switch_to function to load 64k - STACK_FRAME_OVERHEAD - __PT_SIZE (65192) value as unsigned. Reported-by: Benjamin Block <bblock@linux.ibm.com> Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com> Signed-off-by: Vasily Gorbik <gor@linux.ibm.com> Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
frank-w
pushed a commit
that referenced
this pull request
Mar 4, 2019
scsi_device_quiesce() and scsi_device_resume() are called during system-wide suspend and resume. scsi_device_quiesce() only succeeds for SCSI devices that are in one of the RUNNING, OFFLINE or TRANSPORT_OFFLINE states (see also scsi_set_device_state()). This patch avoids that the following warning is triggered when resuming a system for which quiescing a SCSI device failed: WARNING: CPU: 2 PID: 11303 at drivers/scsi/scsi_lib.c:2600 scsi_device_resume+0x4f/0x58 CPU: 2 PID: 11303 Comm: kworker/u8:70 Not tainted 5.0.0-rc1+ #50 Hardware name: LENOVO 80E3/Lancer 5B2, BIOS A2CN45WW(V2.13) 08/04/2016 Workqueue: events_unbound async_run_entry_fn Call Trace: scsi_dev_type_resume+0x2e/0x60 async_run_entry_fn+0x32/0xd8 process_one_work+0x1f4/0x420 worker_thread+0x28/0x3c0 kthread+0x118/0x130 ret_from_fork+0x22/0x40 Cc: Przemek Socha <soprwa@gmail.com> Reported-by: Przemek Socha <soprwa@gmail.com> Fixes: 3a0a529 ("block, scsi: Make SCSI quiesce and resume work reliably") # v4.15 Signed-off-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
frank-w
pushed a commit
that referenced
this pull request
May 8, 2019
[ Upstream commit 388b4e6 ] scsi_device_quiesce() and scsi_device_resume() are called during system-wide suspend and resume. scsi_device_quiesce() only succeeds for SCSI devices that are in one of the RUNNING, OFFLINE or TRANSPORT_OFFLINE states (see also scsi_set_device_state()). This patch avoids that the following warning is triggered when resuming a system for which quiescing a SCSI device failed: WARNING: CPU: 2 PID: 11303 at drivers/scsi/scsi_lib.c:2600 scsi_device_resume+0x4f/0x58 CPU: 2 PID: 11303 Comm: kworker/u8:70 Not tainted 5.0.0-rc1+ #50 Hardware name: LENOVO 80E3/Lancer 5B2, BIOS A2CN45WW(V2.13) 08/04/2016 Workqueue: events_unbound async_run_entry_fn Call Trace: scsi_dev_type_resume+0x2e/0x60 async_run_entry_fn+0x32/0xd8 process_one_work+0x1f4/0x420 worker_thread+0x28/0x3c0 kthread+0x118/0x130 ret_from_fork+0x22/0x40 Cc: Przemek Socha <soprwa@gmail.com> Reported-by: Przemek Socha <soprwa@gmail.com> Fixes: 3a0a529 ("block, scsi: Make SCSI quiesce and resume work reliably") # v4.15 Signed-off-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 2, 2019
[ Upstream commit 3f167e1 ] ipv4_pdp_add() is called in RCU read-side critical section. So GFP_KERNEL should not be used in the function. This patch make ipv4_pdp_add() to use GFP_ATOMIC instead of GFP_KERNEL. Test commands: gtp-link add gtp1 & gtp-tunnel add gtp1 v1 100 200 1.1.1.1 2.2.2.2 Splat looks like: [ 130.618881] ============================= [ 130.626382] WARNING: suspicious RCU usage [ 130.626994] 5.2.0-rc6+ #50 Not tainted [ 130.627622] ----------------------------- [ 130.628223] ./include/linux/rcupdate.h:266 Illegal context switch in RCU read-side critical section! [ 130.629684] [ 130.629684] other info that might help us debug this: [ 130.629684] [ 130.631022] [ 130.631022] rcu_scheduler_active = 2, debug_locks = 1 [ 130.632136] 4 locks held by gtp-tunnel/1025: [ 130.632925] #0: 000000002b93c8b7 (cb_lock){++++}, at: genl_rcv+0x15/0x40 [ 130.634159] #1: 00000000f17bc999 (genl_mutex){+.+.}, at: genl_rcv_msg+0xfb/0x130 [ 130.635487] #2: 00000000c644ed8e (rtnl_mutex){+.+.}, at: gtp_genl_new_pdp+0x18c/0x1150 [gtp] [ 130.636936] #3: 0000000007a1cde7 (rcu_read_lock){....}, at: gtp_genl_new_pdp+0x187/0x1150 [gtp] [ 130.638348] [ 130.638348] stack backtrace: [ 130.639062] CPU: 1 PID: 1025 Comm: gtp-tunnel Not tainted 5.2.0-rc6+ #50 [ 130.641318] Call Trace: [ 130.641707] dump_stack+0x7c/0xbb [ 130.642252] ___might_sleep+0x2c0/0x3b0 [ 130.642862] kmem_cache_alloc_trace+0x1cd/0x2b0 [ 130.643591] gtp_genl_new_pdp+0x6c5/0x1150 [gtp] [ 130.644371] genl_family_rcv_msg+0x63a/0x1030 [ 130.645074] ? mutex_lock_io_nested+0x1090/0x1090 [ 130.645845] ? genl_unregister_family+0x630/0x630 [ 130.646592] ? debug_show_all_locks+0x2d0/0x2d0 [ 130.647293] ? check_flags.part.40+0x440/0x440 [ 130.648099] genl_rcv_msg+0xa3/0x130 [ ... ] Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 2, 2019
[ Upstream commit 1788b85 ] gtp_encap_destroy() is called twice. 1. When interface is deleted. 2. When udp socket is destroyed. either gtp->sk0 or gtp->sk1u could be freed by sock_put() in gtp_encap_destroy(). so, when gtp_encap_destroy() is called again, it would uses freed sk pointer. patch makes gtp_encap_destroy() to set either gtp->sk0 or gtp->sk1u to null. in addition, both gtp->sk0 and gtp->sk1u pointer are protected by rtnl_lock. so, rtnl_lock() is added. Test command: gtp-link add gtp1 & killall gtp-link ip link del gtp1 Splat looks like: [ 83.182767] BUG: KASAN: use-after-free in __lock_acquire+0x3a20/0x46a0 [ 83.184128] Read of size 8 at addr ffff8880cc7d5360 by task ip/1008 [ 83.185567] CPU: 1 PID: 1008 Comm: ip Not tainted 5.2.0-rc6+ #50 [ 83.188469] Call Trace: [ ... ] [ 83.200126] lock_acquire+0x141/0x380 [ 83.200575] ? lock_sock_nested+0x3a/0xf0 [ 83.201069] _raw_spin_lock_bh+0x38/0x70 [ 83.201551] ? lock_sock_nested+0x3a/0xf0 [ 83.202044] lock_sock_nested+0x3a/0xf0 [ 83.202520] gtp_encap_destroy+0x18/0xe0 [gtp] [ 83.203065] gtp_encap_disable.isra.14+0x13/0x50 [gtp] [ 83.203687] gtp_dellink+0x56/0x170 [gtp] [ 83.204190] rtnl_delete_link+0xb4/0x100 [ ... ] [ 83.236513] Allocated by task 976: [ 83.236925] save_stack+0x19/0x80 [ 83.237332] __kasan_kmalloc.constprop.3+0xa0/0xd0 [ 83.237894] kmem_cache_alloc+0xd8/0x280 [ 83.238360] sk_prot_alloc.isra.42+0x50/0x200 [ 83.238874] sk_alloc+0x32/0x940 [ 83.239264] inet_create+0x283/0xc20 [ 83.239684] __sock_create+0x2dd/0x540 [ 83.240136] __sys_socket+0xca/0x1a0 [ 83.240550] __x64_sys_socket+0x6f/0xb0 [ 83.240998] do_syscall_64+0x9c/0x450 [ 83.241466] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 83.242061] [ 83.242249] Freed by task 0: [ 83.242616] save_stack+0x19/0x80 [ 83.243013] __kasan_slab_free+0x111/0x150 [ 83.243498] kmem_cache_free+0x89/0x250 [ 83.244444] __sk_destruct+0x38f/0x5a0 [ 83.245366] rcu_core+0x7e9/0x1c20 [ 83.245766] __do_softirq+0x213/0x8fa Fixes: 1e3a3ab ("gtp: make GTP sockets in gtp_newlink optional") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 2, 2019
[ Upstream commit a2bed90 ] Current gtp_newlink() could be called after unregister_pernet_subsys(). gtp_newlink() uses gtp_net but it can be destroyed by unregister_pernet_subsys(). So unregister_pernet_subsys() should be called after rtnl_link_unregister(). Test commands: #SHELL 1 while : do for i in {1..5} do ./gtp-link add gtp$i & done killall gtp-link done #SHELL 2 while : do modprobe -rv gtp done Splat looks like: [ 753.176631] BUG: KASAN: use-after-free in gtp_newlink+0x9b4/0xa5c [gtp] [ 753.177722] Read of size 8 at addr ffff8880d48f2458 by task gtp-link/7126 [ 753.179082] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G W 5.2.0-rc6+ #50 [ 753.185801] Call Trace: [ 753.186264] dump_stack+0x7c/0xbb [ 753.186863] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.187583] print_address_description+0xc7/0x240 [ 753.188382] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189097] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189846] __kasan_report+0x12a/0x16f [ 753.190542] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.191298] kasan_report+0xe/0x20 [ 753.191893] gtp_newlink+0x9b4/0xa5c [gtp] [ 753.192580] ? __netlink_ns_capable+0xc3/0xf0 [ 753.193370] __rtnl_newlink+0xb9f/0x11b0 [ ... ] [ 753.241201] Allocated by task 7186: [ 753.241844] save_stack+0x19/0x80 [ 753.242399] __kasan_kmalloc.constprop.3+0xa0/0xd0 [ 753.243192] __kmalloc+0x13e/0x300 [ 753.243764] ops_init+0xd6/0x350 [ 753.244314] register_pernet_operations+0x249/0x6f0 [ ... ] [ 753.251770] Freed by task 7178: [ 753.252288] save_stack+0x19/0x80 [ 753.252833] __kasan_slab_free+0x111/0x150 [ 753.253962] kfree+0xc7/0x280 [ 753.254509] ops_free_list.part.11+0x1c4/0x2d0 [ 753.255241] unregister_pernet_operations+0x262/0x390 [ ... ] [ 753.285883] list_add corruption. next->prev should be prev (ffff8880d48f2458), but was ffff8880d497d878. (next. [ 753.287241] ------------[ cut here ]------------ [ 753.287794] kernel BUG at lib/list_debug.c:25! [ 753.288364] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 753.289099] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G B W 5.2.0-rc6+ #50 [ 753.291036] RIP: 0010:__list_add_valid+0x74/0xd0 [ 753.291589] Code: 48 39 da 75 27 48 39 f5 74 36 48 39 dd 74 31 48 83 c4 08 b8 01 00 00 00 5b 5d c3 48 89 d9 48b [ 753.293779] RSP: 0018:ffff8880cae8f398 EFLAGS: 00010286 [ 753.294401] RAX: 0000000000000075 RBX: ffff8880d497d878 RCX: 0000000000000000 [ 753.296260] RDX: 0000000000000075 RSI: 0000000000000008 RDI: ffffed10195d1e69 [ 753.297070] RBP: ffff8880cd250ae0 R08: ffffed101b4bff21 R09: ffffed101b4bff21 [ 753.297899] R10: 0000000000000001 R11: ffffed101b4bff20 R12: ffff8880d497d878 [ 753.298703] R13: 0000000000000000 R14: ffff8880cd250ae0 R15: ffff8880d48f2458 [ 753.299564] FS: 00007f5f79805740(0000) GS:ffff8880da400000(0000) knlGS:0000000000000000 [ 753.300533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 753.301231] CR2: 00007fe8c7ef4f10 CR3: 00000000b71a6006 CR4: 00000000000606f0 [ 753.302183] Call Trace: [ 753.302530] gtp_newlink+0x5f6/0xa5c [gtp] [ 753.303037] ? __netlink_ns_capable+0xc3/0xf0 [ 753.303576] __rtnl_newlink+0xb9f/0x11b0 [ 753.304092] ? rtnl_link_unregister+0x230/0x230 Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 5, 2019
[ Upstream commit 3f167e1 ] ipv4_pdp_add() is called in RCU read-side critical section. So GFP_KERNEL should not be used in the function. This patch make ipv4_pdp_add() to use GFP_ATOMIC instead of GFP_KERNEL. Test commands: gtp-link add gtp1 & gtp-tunnel add gtp1 v1 100 200 1.1.1.1 2.2.2.2 Splat looks like: [ 130.618881] ============================= [ 130.626382] WARNING: suspicious RCU usage [ 130.626994] 5.2.0-rc6+ #50 Not tainted [ 130.627622] ----------------------------- [ 130.628223] ./include/linux/rcupdate.h:266 Illegal context switch in RCU read-side critical section! [ 130.629684] [ 130.629684] other info that might help us debug this: [ 130.629684] [ 130.631022] [ 130.631022] rcu_scheduler_active = 2, debug_locks = 1 [ 130.632136] 4 locks held by gtp-tunnel/1025: [ 130.632925] #0: 000000002b93c8b7 (cb_lock){++++}, at: genl_rcv+0x15/0x40 [ 130.634159] #1: 00000000f17bc999 (genl_mutex){+.+.}, at: genl_rcv_msg+0xfb/0x130 [ 130.635487] #2: 00000000c644ed8e (rtnl_mutex){+.+.}, at: gtp_genl_new_pdp+0x18c/0x1150 [gtp] [ 130.636936] #3: 0000000007a1cde7 (rcu_read_lock){....}, at: gtp_genl_new_pdp+0x187/0x1150 [gtp] [ 130.638348] [ 130.638348] stack backtrace: [ 130.639062] CPU: 1 PID: 1025 Comm: gtp-tunnel Not tainted 5.2.0-rc6+ #50 [ 130.641318] Call Trace: [ 130.641707] dump_stack+0x7c/0xbb [ 130.642252] ___might_sleep+0x2c0/0x3b0 [ 130.642862] kmem_cache_alloc_trace+0x1cd/0x2b0 [ 130.643591] gtp_genl_new_pdp+0x6c5/0x1150 [gtp] [ 130.644371] genl_family_rcv_msg+0x63a/0x1030 [ 130.645074] ? mutex_lock_io_nested+0x1090/0x1090 [ 130.645845] ? genl_unregister_family+0x630/0x630 [ 130.646592] ? debug_show_all_locks+0x2d0/0x2d0 [ 130.647293] ? check_flags.part.40+0x440/0x440 [ 130.648099] genl_rcv_msg+0xa3/0x130 [ ... ] Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 5, 2019
[ Upstream commit 1788b85 ] gtp_encap_destroy() is called twice. 1. When interface is deleted. 2. When udp socket is destroyed. either gtp->sk0 or gtp->sk1u could be freed by sock_put() in gtp_encap_destroy(). so, when gtp_encap_destroy() is called again, it would uses freed sk pointer. patch makes gtp_encap_destroy() to set either gtp->sk0 or gtp->sk1u to null. in addition, both gtp->sk0 and gtp->sk1u pointer are protected by rtnl_lock. so, rtnl_lock() is added. Test command: gtp-link add gtp1 & killall gtp-link ip link del gtp1 Splat looks like: [ 83.182767] BUG: KASAN: use-after-free in __lock_acquire+0x3a20/0x46a0 [ 83.184128] Read of size 8 at addr ffff8880cc7d5360 by task ip/1008 [ 83.185567] CPU: 1 PID: 1008 Comm: ip Not tainted 5.2.0-rc6+ #50 [ 83.188469] Call Trace: [ ... ] [ 83.200126] lock_acquire+0x141/0x380 [ 83.200575] ? lock_sock_nested+0x3a/0xf0 [ 83.201069] _raw_spin_lock_bh+0x38/0x70 [ 83.201551] ? lock_sock_nested+0x3a/0xf0 [ 83.202044] lock_sock_nested+0x3a/0xf0 [ 83.202520] gtp_encap_destroy+0x18/0xe0 [gtp] [ 83.203065] gtp_encap_disable.isra.14+0x13/0x50 [gtp] [ 83.203687] gtp_dellink+0x56/0x170 [gtp] [ 83.204190] rtnl_delete_link+0xb4/0x100 [ ... ] [ 83.236513] Allocated by task 976: [ 83.236925] save_stack+0x19/0x80 [ 83.237332] __kasan_kmalloc.constprop.3+0xa0/0xd0 [ 83.237894] kmem_cache_alloc+0xd8/0x280 [ 83.238360] sk_prot_alloc.isra.42+0x50/0x200 [ 83.238874] sk_alloc+0x32/0x940 [ 83.239264] inet_create+0x283/0xc20 [ 83.239684] __sock_create+0x2dd/0x540 [ 83.240136] __sys_socket+0xca/0x1a0 [ 83.240550] __x64_sys_socket+0x6f/0xb0 [ 83.240998] do_syscall_64+0x9c/0x450 [ 83.241466] entry_SYSCALL_64_after_hwframe+0x49/0xbe [ 83.242061] [ 83.242249] Freed by task 0: [ 83.242616] save_stack+0x19/0x80 [ 83.243013] __kasan_slab_free+0x111/0x150 [ 83.243498] kmem_cache_free+0x89/0x250 [ 83.244444] __sk_destruct+0x38f/0x5a0 [ 83.245366] rcu_core+0x7e9/0x1c20 [ 83.245766] __do_softirq+0x213/0x8fa Fixes: 1e3a3ab ("gtp: make GTP sockets in gtp_newlink optional") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 5, 2019
[ Upstream commit a2bed90 ] Current gtp_newlink() could be called after unregister_pernet_subsys(). gtp_newlink() uses gtp_net but it can be destroyed by unregister_pernet_subsys(). So unregister_pernet_subsys() should be called after rtnl_link_unregister(). Test commands: #SHELL 1 while : do for i in {1..5} do ./gtp-link add gtp$i & done killall gtp-link done #SHELL 2 while : do modprobe -rv gtp done Splat looks like: [ 753.176631] BUG: KASAN: use-after-free in gtp_newlink+0x9b4/0xa5c [gtp] [ 753.177722] Read of size 8 at addr ffff8880d48f2458 by task gtp-link/7126 [ 753.179082] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G W 5.2.0-rc6+ #50 [ 753.185801] Call Trace: [ 753.186264] dump_stack+0x7c/0xbb [ 753.186863] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.187583] print_address_description+0xc7/0x240 [ 753.188382] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189097] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189846] __kasan_report+0x12a/0x16f [ 753.190542] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.191298] kasan_report+0xe/0x20 [ 753.191893] gtp_newlink+0x9b4/0xa5c [gtp] [ 753.192580] ? __netlink_ns_capable+0xc3/0xf0 [ 753.193370] __rtnl_newlink+0xb9f/0x11b0 [ ... ] [ 753.241201] Allocated by task 7186: [ 753.241844] save_stack+0x19/0x80 [ 753.242399] __kasan_kmalloc.constprop.3+0xa0/0xd0 [ 753.243192] __kmalloc+0x13e/0x300 [ 753.243764] ops_init+0xd6/0x350 [ 753.244314] register_pernet_operations+0x249/0x6f0 [ ... ] [ 753.251770] Freed by task 7178: [ 753.252288] save_stack+0x19/0x80 [ 753.252833] __kasan_slab_free+0x111/0x150 [ 753.253962] kfree+0xc7/0x280 [ 753.254509] ops_free_list.part.11+0x1c4/0x2d0 [ 753.255241] unregister_pernet_operations+0x262/0x390 [ ... ] [ 753.285883] list_add corruption. next->prev should be prev (ffff8880d48f2458), but was ffff8880d497d878. (next. [ 753.287241] ------------[ cut here ]------------ [ 753.287794] kernel BUG at lib/list_debug.c:25! [ 753.288364] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 753.289099] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G B W 5.2.0-rc6+ #50 [ 753.291036] RIP: 0010:__list_add_valid+0x74/0xd0 [ 753.291589] Code: 48 39 da 75 27 48 39 f5 74 36 48 39 dd 74 31 48 83 c4 08 b8 01 00 00 00 5b 5d c3 48 89 d9 48b [ 753.293779] RSP: 0018:ffff8880cae8f398 EFLAGS: 00010286 [ 753.294401] RAX: 0000000000000075 RBX: ffff8880d497d878 RCX: 0000000000000000 [ 753.296260] RDX: 0000000000000075 RSI: 0000000000000008 RDI: ffffed10195d1e69 [ 753.297070] RBP: ffff8880cd250ae0 R08: ffffed101b4bff21 R09: ffffed101b4bff21 [ 753.297899] R10: 0000000000000001 R11: ffffed101b4bff20 R12: ffff8880d497d878 [ 753.298703] R13: 0000000000000000 R14: ffff8880cd250ae0 R15: ffff8880d48f2458 [ 753.299564] FS: 00007f5f79805740(0000) GS:ffff8880da400000(0000) knlGS:0000000000000000 [ 753.300533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 753.301231] CR2: 00007fe8c7ef4f10 CR3: 00000000b71a6006 CR4: 00000000000606f0 [ 753.302183] Call Trace: [ 753.302530] gtp_newlink+0x5f6/0xa5c [gtp] [ 753.303037] ? __netlink_ns_capable+0xc3/0xf0 [ 753.303576] __rtnl_newlink+0xb9f/0x11b0 [ 753.304092] ? rtnl_link_unregister+0x230/0x230 Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 10, 2019
[ Upstream commit 3f167e1 ] ipv4_pdp_add() is called in RCU read-side critical section. So GFP_KERNEL should not be used in the function. This patch make ipv4_pdp_add() to use GFP_ATOMIC instead of GFP_KERNEL. Test commands: gtp-link add gtp1 & gtp-tunnel add gtp1 v1 100 200 1.1.1.1 2.2.2.2 Splat looks like: [ 130.618881] ============================= [ 130.626382] WARNING: suspicious RCU usage [ 130.626994] 5.2.0-rc6+ #50 Not tainted [ 130.627622] ----------------------------- [ 130.628223] ./include/linux/rcupdate.h:266 Illegal context switch in RCU read-side critical section! [ 130.629684] [ 130.629684] other info that might help us debug this: [ 130.629684] [ 130.631022] [ 130.631022] rcu_scheduler_active = 2, debug_locks = 1 [ 130.632136] 4 locks held by gtp-tunnel/1025: [ 130.632925] #0: 000000002b93c8b7 (cb_lock){++++}, at: genl_rcv+0x15/0x40 [ 130.634159] #1: 00000000f17bc999 (genl_mutex){+.+.}, at: genl_rcv_msg+0xfb/0x130 [ 130.635487] #2: 00000000c644ed8e (rtnl_mutex){+.+.}, at: gtp_genl_new_pdp+0x18c/0x1150 [gtp] [ 130.636936] #3: 0000000007a1cde7 (rcu_read_lock){....}, at: gtp_genl_new_pdp+0x187/0x1150 [gtp] [ 130.638348] [ 130.638348] stack backtrace: [ 130.639062] CPU: 1 PID: 1025 Comm: gtp-tunnel Not tainted 5.2.0-rc6+ #50 [ 130.641318] Call Trace: [ 130.641707] dump_stack+0x7c/0xbb [ 130.642252] ___might_sleep+0x2c0/0x3b0 [ 130.642862] kmem_cache_alloc_trace+0x1cd/0x2b0 [ 130.643591] gtp_genl_new_pdp+0x6c5/0x1150 [gtp] [ 130.644371] genl_family_rcv_msg+0x63a/0x1030 [ 130.645074] ? mutex_lock_io_nested+0x1090/0x1090 [ 130.645845] ? genl_unregister_family+0x630/0x630 [ 130.646592] ? debug_show_all_locks+0x2d0/0x2d0 [ 130.647293] ? check_flags.part.40+0x440/0x440 [ 130.648099] genl_rcv_msg+0xa3/0x130 [ ... ] Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Aug 10, 2019
[ Upstream commit a2bed90 ] Current gtp_newlink() could be called after unregister_pernet_subsys(). gtp_newlink() uses gtp_net but it can be destroyed by unregister_pernet_subsys(). So unregister_pernet_subsys() should be called after rtnl_link_unregister(). Test commands: #SHELL 1 while : do for i in {1..5} do ./gtp-link add gtp$i & done killall gtp-link done #SHELL 2 while : do modprobe -rv gtp done Splat looks like: [ 753.176631] BUG: KASAN: use-after-free in gtp_newlink+0x9b4/0xa5c [gtp] [ 753.177722] Read of size 8 at addr ffff8880d48f2458 by task gtp-link/7126 [ 753.179082] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G W 5.2.0-rc6+ #50 [ 753.185801] Call Trace: [ 753.186264] dump_stack+0x7c/0xbb [ 753.186863] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.187583] print_address_description+0xc7/0x240 [ 753.188382] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189097] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.189846] __kasan_report+0x12a/0x16f [ 753.190542] ? gtp_newlink+0x9b4/0xa5c [gtp] [ 753.191298] kasan_report+0xe/0x20 [ 753.191893] gtp_newlink+0x9b4/0xa5c [gtp] [ 753.192580] ? __netlink_ns_capable+0xc3/0xf0 [ 753.193370] __rtnl_newlink+0xb9f/0x11b0 [ ... ] [ 753.241201] Allocated by task 7186: [ 753.241844] save_stack+0x19/0x80 [ 753.242399] __kasan_kmalloc.constprop.3+0xa0/0xd0 [ 753.243192] __kmalloc+0x13e/0x300 [ 753.243764] ops_init+0xd6/0x350 [ 753.244314] register_pernet_operations+0x249/0x6f0 [ ... ] [ 753.251770] Freed by task 7178: [ 753.252288] save_stack+0x19/0x80 [ 753.252833] __kasan_slab_free+0x111/0x150 [ 753.253962] kfree+0xc7/0x280 [ 753.254509] ops_free_list.part.11+0x1c4/0x2d0 [ 753.255241] unregister_pernet_operations+0x262/0x390 [ ... ] [ 753.285883] list_add corruption. next->prev should be prev (ffff8880d48f2458), but was ffff8880d497d878. (next. [ 753.287241] ------------[ cut here ]------------ [ 753.287794] kernel BUG at lib/list_debug.c:25! [ 753.288364] invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN PTI [ 753.289099] CPU: 0 PID: 7126 Comm: gtp-link Tainted: G B W 5.2.0-rc6+ #50 [ 753.291036] RIP: 0010:__list_add_valid+0x74/0xd0 [ 753.291589] Code: 48 39 da 75 27 48 39 f5 74 36 48 39 dd 74 31 48 83 c4 08 b8 01 00 00 00 5b 5d c3 48 89 d9 48b [ 753.293779] RSP: 0018:ffff8880cae8f398 EFLAGS: 00010286 [ 753.294401] RAX: 0000000000000075 RBX: ffff8880d497d878 RCX: 0000000000000000 [ 753.296260] RDX: 0000000000000075 RSI: 0000000000000008 RDI: ffffed10195d1e69 [ 753.297070] RBP: ffff8880cd250ae0 R08: ffffed101b4bff21 R09: ffffed101b4bff21 [ 753.297899] R10: 0000000000000001 R11: ffffed101b4bff20 R12: ffff8880d497d878 [ 753.298703] R13: 0000000000000000 R14: ffff8880cd250ae0 R15: ffff8880d48f2458 [ 753.299564] FS: 00007f5f79805740(0000) GS:ffff8880da400000(0000) knlGS:0000000000000000 [ 753.300533] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 753.301231] CR2: 00007fe8c7ef4f10 CR3: 00000000b71a6006 CR4: 00000000000606f0 [ 753.302183] Call Trace: [ 753.302530] gtp_newlink+0x5f6/0xa5c [gtp] [ 753.303037] ? __netlink_ns_capable+0xc3/0xf0 [ 753.303576] __rtnl_newlink+0xb9f/0x11b0 [ 753.304092] ? rtnl_link_unregister+0x230/0x230 Fixes: 459aa66 ("gtp: add initial driver for datapath of GPRS Tunneling Protocol (GTP-U)") Signed-off-by: Taehee Yoo <ap420073@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Mar 5, 2020
The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by
this change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
Lastly, fix the following checkpatch warning:
CHECK: Prefer kernel type 'u8' over 'uint8_t'
#50: FILE: net/l2tp/l2tp_core.h:119:
+ uint8_t priv[]; /* private data */
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] KSPP/linux#21
[3] commit 7649773 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
frank-w
pushed a commit
that referenced
this pull request
Jan 20, 2021
commit d68b295 upstream. Commit 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") made it possible for some code paths in dm-crypt to be executed in softirq context, when the underlying driver processes IO requests in interrupt/softirq context. In this case sometimes when allocating a new crypto request we may get a stacktrace like below: [ 210.103008][ C0] BUG: sleeping function called from invalid context at mm/mempool.c:381 [ 210.104746][ C0] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 2602, name: fio [ 210.106599][ C0] CPU: 0 PID: 2602 Comm: fio Tainted: G W 5.10.0+ #50 [ 210.108331][ C0] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 [ 210.110212][ C0] Call Trace: [ 210.110921][ C0] <IRQ> [ 210.111527][ C0] dump_stack+0x7d/0xa3 [ 210.112411][ C0] ___might_sleep.cold+0x122/0x151 [ 210.113527][ C0] mempool_alloc+0x16b/0x2f0 [ 210.114524][ C0] ? __queue_work+0x515/0xde0 [ 210.115553][ C0] ? mempool_resize+0x700/0x700 [ 210.116586][ C0] ? crypt_endio+0x91/0x180 [ 210.117479][ C0] ? blk_update_request+0x757/0x1150 [ 210.118513][ C0] ? blk_mq_end_request+0x4b/0x480 [ 210.119572][ C0] ? blk_done_softirq+0x21d/0x340 [ 210.120628][ C0] ? __do_softirq+0x190/0x611 [ 210.121626][ C0] crypt_convert+0x29f9/0x4c00 [ 210.122668][ C0] ? _raw_spin_lock_irqsave+0x87/0xe0 [ 210.123824][ C0] ? kasan_set_track+0x1c/0x30 [ 210.124858][ C0] ? crypt_iv_tcw_ctr+0x4a0/0x4a0 [ 210.125930][ C0] ? kmem_cache_free+0x104/0x470 [ 210.126973][ C0] ? crypt_endio+0x91/0x180 [ 210.127947][ C0] kcryptd_crypt_read_convert+0x30e/0x420 [ 210.129165][ C0] blk_update_request+0x757/0x1150 [ 210.130231][ C0] blk_mq_end_request+0x4b/0x480 [ 210.131294][ C0] blk_done_softirq+0x21d/0x340 [ 210.132332][ C0] ? _raw_spin_lock+0x81/0xd0 [ 210.133289][ C0] ? blk_mq_stop_hw_queue+0x30/0x30 [ 210.134399][ C0] ? _raw_read_lock_irq+0x40/0x40 [ 210.135458][ C0] __do_softirq+0x190/0x611 [ 210.136409][ C0] ? handle_edge_irq+0x221/0xb60 [ 210.137447][ C0] asm_call_irq_on_stack+0x12/0x20 [ 210.138507][ C0] </IRQ> [ 210.139118][ C0] do_softirq_own_stack+0x37/0x40 [ 210.140191][ C0] irq_exit_rcu+0x110/0x1b0 [ 210.141151][ C0] common_interrupt+0x74/0x120 [ 210.142171][ C0] asm_common_interrupt+0x1e/0x40 Fix this by allocating crypto requests with GFP_ATOMIC mask in interrupt context. Fixes: 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") Cc: stable@vger.kernel.org # v5.9+ Reported-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name> Signed-off-by: Ignat Korchagin <ignat@cloudflare.com> Acked-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
frank-w
pushed a commit
that referenced
this pull request
Jan 20, 2021
…tirq commit 8abec36 upstream. Commit 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") made it possible for some code paths in dm-crypt to be executed in softirq context, when the underlying driver processes IO requests in interrupt/softirq context. When Crypto API backlogs a crypto request, dm-crypt uses wait_for_completion to avoid sending further requests to an already overloaded crypto driver. However, if the code is executing in softirq context, we might get the following stacktrace: [ 210.235213][ C0] BUG: scheduling while atomic: fio/2602/0x00000102 [ 210.236701][ C0] Modules linked in: [ 210.237566][ C0] CPU: 0 PID: 2602 Comm: fio Tainted: G W 5.10.0+ #50 [ 210.239292][ C0] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 [ 210.241233][ C0] Call Trace: [ 210.241946][ C0] <IRQ> [ 210.242561][ C0] dump_stack+0x7d/0xa3 [ 210.243466][ C0] __schedule_bug.cold+0xb3/0xc2 [ 210.244539][ C0] __schedule+0x156f/0x20d0 [ 210.245518][ C0] ? io_schedule_timeout+0x140/0x140 [ 210.246660][ C0] schedule+0xd0/0x270 [ 210.247541][ C0] schedule_timeout+0x1fb/0x280 [ 210.248586][ C0] ? usleep_range+0x150/0x150 [ 210.249624][ C0] ? unpoison_range+0x3a/0x60 [ 210.250632][ C0] ? ____kasan_kmalloc.constprop.0+0x82/0xa0 [ 210.251949][ C0] ? unpoison_range+0x3a/0x60 [ 210.252958][ C0] ? __prepare_to_swait+0xa7/0x190 [ 210.254067][ C0] do_wait_for_common+0x2ab/0x370 [ 210.255158][ C0] ? usleep_range+0x150/0x150 [ 210.256192][ C0] ? bit_wait_io_timeout+0x160/0x160 [ 210.257358][ C0] ? blk_update_request+0x757/0x1150 [ 210.258582][ C0] ? _raw_spin_lock_irq+0x82/0xd0 [ 210.259674][ C0] ? _raw_read_unlock_irqrestore+0x30/0x30 [ 210.260917][ C0] wait_for_completion+0x4c/0x90 [ 210.261971][ C0] crypt_convert+0x19a6/0x4c00 [ 210.263033][ C0] ? _raw_spin_lock_irqsave+0x87/0xe0 [ 210.264193][ C0] ? kasan_set_track+0x1c/0x30 [ 210.265191][ C0] ? crypt_iv_tcw_ctr+0x4a0/0x4a0 [ 210.266283][ C0] ? kmem_cache_free+0x104/0x470 [ 210.267363][ C0] ? crypt_endio+0x91/0x180 [ 210.268327][ C0] kcryptd_crypt_read_convert+0x30e/0x420 [ 210.269565][ C0] blk_update_request+0x757/0x1150 [ 210.270563][ C0] blk_mq_end_request+0x4b/0x480 [ 210.271680][ C0] blk_done_softirq+0x21d/0x340 [ 210.272775][ C0] ? _raw_spin_lock+0x81/0xd0 [ 210.273847][ C0] ? blk_mq_stop_hw_queue+0x30/0x30 [ 210.275031][ C0] ? _raw_read_lock_irq+0x40/0x40 [ 210.276182][ C0] __do_softirq+0x190/0x611 [ 210.277203][ C0] ? handle_edge_irq+0x221/0xb60 [ 210.278340][ C0] asm_call_irq_on_stack+0x12/0x20 [ 210.279514][ C0] </IRQ> [ 210.280164][ C0] do_softirq_own_stack+0x37/0x40 [ 210.281281][ C0] irq_exit_rcu+0x110/0x1b0 [ 210.282286][ C0] common_interrupt+0x74/0x120 [ 210.283376][ C0] asm_common_interrupt+0x1e/0x40 [ 210.284496][ C0] RIP: 0010:_aesni_enc1+0x65/0xb0 Fix this by making crypt_convert function reentrant from the point of a single bio and make dm-crypt defer further bio processing to a workqueue, if Crypto API backlogs a request in interrupt context. Fixes: 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") Cc: stable@vger.kernel.org # v5.9+ Signed-off-by: Ignat Korchagin <ignat@cloudflare.com> Acked-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
frank-w
pushed a commit
that referenced
this pull request
Jan 22, 2021
…tirq Commit 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") made it possible for some code paths in dm-crypt to be executed in softirq context, when the underlying driver processes IO requests in interrupt/softirq context. When Crypto API backlogs a crypto request, dm-crypt uses wait_for_completion to avoid sending further requests to an already overloaded crypto driver. However, if the code is executing in softirq context, we might get the following stacktrace: [ 210.235213][ C0] BUG: scheduling while atomic: fio/2602/0x00000102 [ 210.236701][ C0] Modules linked in: [ 210.237566][ C0] CPU: 0 PID: 2602 Comm: fio Tainted: G W 5.10.0+ #50 [ 210.239292][ C0] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 [ 210.241233][ C0] Call Trace: [ 210.241946][ C0] <IRQ> [ 210.242561][ C0] dump_stack+0x7d/0xa3 [ 210.243466][ C0] __schedule_bug.cold+0xb3/0xc2 [ 210.244539][ C0] __schedule+0x156f/0x20d0 [ 210.245518][ C0] ? io_schedule_timeout+0x140/0x140 [ 210.246660][ C0] schedule+0xd0/0x270 [ 210.247541][ C0] schedule_timeout+0x1fb/0x280 [ 210.248586][ C0] ? usleep_range+0x150/0x150 [ 210.249624][ C0] ? unpoison_range+0x3a/0x60 [ 210.250632][ C0] ? ____kasan_kmalloc.constprop.0+0x82/0xa0 [ 210.251949][ C0] ? unpoison_range+0x3a/0x60 [ 210.252958][ C0] ? __prepare_to_swait+0xa7/0x190 [ 210.254067][ C0] do_wait_for_common+0x2ab/0x370 [ 210.255158][ C0] ? usleep_range+0x150/0x150 [ 210.256192][ C0] ? bit_wait_io_timeout+0x160/0x160 [ 210.257358][ C0] ? blk_update_request+0x757/0x1150 [ 210.258582][ C0] ? _raw_spin_lock_irq+0x82/0xd0 [ 210.259674][ C0] ? _raw_read_unlock_irqrestore+0x30/0x30 [ 210.260917][ C0] wait_for_completion+0x4c/0x90 [ 210.261971][ C0] crypt_convert+0x19a6/0x4c00 [ 210.263033][ C0] ? _raw_spin_lock_irqsave+0x87/0xe0 [ 210.264193][ C0] ? kasan_set_track+0x1c/0x30 [ 210.265191][ C0] ? crypt_iv_tcw_ctr+0x4a0/0x4a0 [ 210.266283][ C0] ? kmem_cache_free+0x104/0x470 [ 210.267363][ C0] ? crypt_endio+0x91/0x180 [ 210.268327][ C0] kcryptd_crypt_read_convert+0x30e/0x420 [ 210.269565][ C0] blk_update_request+0x757/0x1150 [ 210.270563][ C0] blk_mq_end_request+0x4b/0x480 [ 210.271680][ C0] blk_done_softirq+0x21d/0x340 [ 210.272775][ C0] ? _raw_spin_lock+0x81/0xd0 [ 210.273847][ C0] ? blk_mq_stop_hw_queue+0x30/0x30 [ 210.275031][ C0] ? _raw_read_lock_irq+0x40/0x40 [ 210.276182][ C0] __do_softirq+0x190/0x611 [ 210.277203][ C0] ? handle_edge_irq+0x221/0xb60 [ 210.278340][ C0] asm_call_irq_on_stack+0x12/0x20 [ 210.279514][ C0] </IRQ> [ 210.280164][ C0] do_softirq_own_stack+0x37/0x40 [ 210.281281][ C0] irq_exit_rcu+0x110/0x1b0 [ 210.282286][ C0] common_interrupt+0x74/0x120 [ 210.283376][ C0] asm_common_interrupt+0x1e/0x40 [ 210.284496][ C0] RIP: 0010:_aesni_enc1+0x65/0xb0 Fix this by making crypt_convert function reentrant from the point of a single bio and make dm-crypt defer further bio processing to a workqueue, if Crypto API backlogs a request in interrupt context. Fixes: 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") Cc: stable@vger.kernel.org # v5.9+ Signed-off-by: Ignat Korchagin <ignat@cloudflare.com> Acked-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
frank-w
pushed a commit
that referenced
this pull request
Jan 22, 2021
Commit 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") made it possible for some code paths in dm-crypt to be executed in softirq context, when the underlying driver processes IO requests in interrupt/softirq context. In this case sometimes when allocating a new crypto request we may get a stacktrace like below: [ 210.103008][ C0] BUG: sleeping function called from invalid context at mm/mempool.c:381 [ 210.104746][ C0] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 2602, name: fio [ 210.106599][ C0] CPU: 0 PID: 2602 Comm: fio Tainted: G W 5.10.0+ #50 [ 210.108331][ C0] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015 [ 210.110212][ C0] Call Trace: [ 210.110921][ C0] <IRQ> [ 210.111527][ C0] dump_stack+0x7d/0xa3 [ 210.112411][ C0] ___might_sleep.cold+0x122/0x151 [ 210.113527][ C0] mempool_alloc+0x16b/0x2f0 [ 210.114524][ C0] ? __queue_work+0x515/0xde0 [ 210.115553][ C0] ? mempool_resize+0x700/0x700 [ 210.116586][ C0] ? crypt_endio+0x91/0x180 [ 210.117479][ C0] ? blk_update_request+0x757/0x1150 [ 210.118513][ C0] ? blk_mq_end_request+0x4b/0x480 [ 210.119572][ C0] ? blk_done_softirq+0x21d/0x340 [ 210.120628][ C0] ? __do_softirq+0x190/0x611 [ 210.121626][ C0] crypt_convert+0x29f9/0x4c00 [ 210.122668][ C0] ? _raw_spin_lock_irqsave+0x87/0xe0 [ 210.123824][ C0] ? kasan_set_track+0x1c/0x30 [ 210.124858][ C0] ? crypt_iv_tcw_ctr+0x4a0/0x4a0 [ 210.125930][ C0] ? kmem_cache_free+0x104/0x470 [ 210.126973][ C0] ? crypt_endio+0x91/0x180 [ 210.127947][ C0] kcryptd_crypt_read_convert+0x30e/0x420 [ 210.129165][ C0] blk_update_request+0x757/0x1150 [ 210.130231][ C0] blk_mq_end_request+0x4b/0x480 [ 210.131294][ C0] blk_done_softirq+0x21d/0x340 [ 210.132332][ C0] ? _raw_spin_lock+0x81/0xd0 [ 210.133289][ C0] ? blk_mq_stop_hw_queue+0x30/0x30 [ 210.134399][ C0] ? _raw_read_lock_irq+0x40/0x40 [ 210.135458][ C0] __do_softirq+0x190/0x611 [ 210.136409][ C0] ? handle_edge_irq+0x221/0xb60 [ 210.137447][ C0] asm_call_irq_on_stack+0x12/0x20 [ 210.138507][ C0] </IRQ> [ 210.139118][ C0] do_softirq_own_stack+0x37/0x40 [ 210.140191][ C0] irq_exit_rcu+0x110/0x1b0 [ 210.141151][ C0] common_interrupt+0x74/0x120 [ 210.142171][ C0] asm_common_interrupt+0x1e/0x40 Fix this by allocating crypto requests with GFP_ATOMIC mask in interrupt context. Fixes: 39d42fa ("dm crypt: add flags to optionally bypass kcryptd workqueues") Cc: stable@vger.kernel.org # v5.9+ Reported-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name> Signed-off-by: Ignat Korchagin <ignat@cloudflare.com> Acked-by: Mikulas Patocka <mpatocka@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com>
frank-w
pushed a commit
that referenced
this pull request
May 19, 2021
[ Upstream commit 5bbf219 ] An out of bounds write happens when setting the default power state. KASAN sees this as: [drm] radeon: 512M of GTT memory ready. [drm] GART: num cpu pages 131072, num gpu pages 131072 ================================================================== BUG: KASAN: slab-out-of-bounds in radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] Write of size 4 at addr ffff88810178d858 by task systemd-udevd/157 CPU: 0 PID: 157 Comm: systemd-udevd Not tainted 5.12.0-E620 #50 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 Call Trace: dump_stack+0xa5/0xe6 print_address_description.constprop.0+0x18/0x239 kasan_report+0x170/0x1a8 radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] radeon_atombios_get_power_modes+0x144/0x1888 [radeon] radeon_pm_init+0x1019/0x1904 [radeon] rs690_init+0x76e/0x84a [radeon] radeon_device_init+0x1c1a/0x21e5 [radeon] radeon_driver_load_kms+0xf5/0x30b [radeon] drm_dev_register+0x255/0x4a0 [drm] radeon_pci_probe+0x246/0x2f6 [radeon] pci_device_probe+0x1aa/0x294 really_probe+0x30e/0x850 driver_probe_device+0xe6/0x135 device_driver_attach+0xc1/0xf8 __driver_attach+0x13f/0x146 bus_for_each_dev+0xfa/0x146 bus_add_driver+0x2b3/0x447 driver_register+0x242/0x2c1 do_one_initcall+0x149/0x2fd do_init_module+0x1ae/0x573 load_module+0x4dee/0x5cca __do_sys_finit_module+0xf1/0x140 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xae Without KASAN, this will manifest later when the kernel attempts to allocate memory that was stomped, since it collides with the inline slab freelist pointer: invalid opcode: 0000 [#1] SMP NOPTI CPU: 0 PID: 781 Comm: openrc-run.sh Tainted: G W 5.10.12-gentoo-E620 #2 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 RIP: 0010:kfree+0x115/0x230 Code: 89 c5 e8 75 ea ff ff 48 8b 00 0f ba e0 09 72 63 e8 1f f4 ff ff 41 89 c4 48 8b 45 00 0f ba e0 10 72 0a 48 8b 45 08 a8 01 75 02 <0f> 0b 44 89 e1 48 c7 c2 00 f0 ff ff be 06 00 00 00 48 d3 e2 48 c7 RSP: 0018:ffffb42f40267e10 EFLAGS: 00010246 RAX: ffffd61280ee8d88 RBX: 0000000000000004 RCX: 000000008010000d RDX: 4000000000000000 RSI: ffffffffba1360b0 RDI: ffffd61280ee8d80 RBP: ffffd61280ee8d80 R08: ffffffffb91bebdf R09: 0000000000000000 R10: ffff8fe2c1047ac8 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000100 FS: 00007fe80eff6b68(0000) GS:ffff8fe339c00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fe80eec7bc0 CR3: 0000000038012000 CR4: 00000000000006f0 Call Trace: __free_fdtable+0x16/0x1f put_files_struct+0x81/0x9b do_exit+0x433/0x94d do_group_exit+0xa6/0xa6 __x64_sys_exit_group+0xf/0xf do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fe80ef64bea Code: Unable to access opcode bytes at RIP 0x7fe80ef64bc0. RSP: 002b:00007ffdb1c47528 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7 RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007fe80ef64bea RDX: 00007fe80ef64f60 RSI: 0000000000000000 RDI: 0000000000000000 RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000 R10: 00007fe80ee2c620 R11: 0000000000000246 R12: 00007fe80eff41e0 R13: 00000000ffffffff R14: 0000000000000024 R15: 00007fe80edf9cd0 Modules linked in: radeon(+) ath5k(+) snd_hda_codec_realtek ... Use a valid power_state index when initializing the "flags" and "misc" and "misc2" fields. Bug: https://bugzilla.kernel.org/show_bug.cgi?id=211537 Reported-by: Erhard F. <erhard_f@mailbox.org> Fixes: a48b9b4 ("drm/radeon/kms/pm: add asic specific callbacks for getting power state (v2)") Fixes: 79daedc ("drm/radeon/kms: minor pm cleanups") Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 19, 2021
[ Upstream commit 5bbf219 ] An out of bounds write happens when setting the default power state. KASAN sees this as: [drm] radeon: 512M of GTT memory ready. [drm] GART: num cpu pages 131072, num gpu pages 131072 ================================================================== BUG: KASAN: slab-out-of-bounds in radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] Write of size 4 at addr ffff88810178d858 by task systemd-udevd/157 CPU: 0 PID: 157 Comm: systemd-udevd Not tainted 5.12.0-E620 #50 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 Call Trace: dump_stack+0xa5/0xe6 print_address_description.constprop.0+0x18/0x239 kasan_report+0x170/0x1a8 radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] radeon_atombios_get_power_modes+0x144/0x1888 [radeon] radeon_pm_init+0x1019/0x1904 [radeon] rs690_init+0x76e/0x84a [radeon] radeon_device_init+0x1c1a/0x21e5 [radeon] radeon_driver_load_kms+0xf5/0x30b [radeon] drm_dev_register+0x255/0x4a0 [drm] radeon_pci_probe+0x246/0x2f6 [radeon] pci_device_probe+0x1aa/0x294 really_probe+0x30e/0x850 driver_probe_device+0xe6/0x135 device_driver_attach+0xc1/0xf8 __driver_attach+0x13f/0x146 bus_for_each_dev+0xfa/0x146 bus_add_driver+0x2b3/0x447 driver_register+0x242/0x2c1 do_one_initcall+0x149/0x2fd do_init_module+0x1ae/0x573 load_module+0x4dee/0x5cca __do_sys_finit_module+0xf1/0x140 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xae Without KASAN, this will manifest later when the kernel attempts to allocate memory that was stomped, since it collides with the inline slab freelist pointer: invalid opcode: 0000 [#1] SMP NOPTI CPU: 0 PID: 781 Comm: openrc-run.sh Tainted: G W 5.10.12-gentoo-E620 #2 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 RIP: 0010:kfree+0x115/0x230 Code: 89 c5 e8 75 ea ff ff 48 8b 00 0f ba e0 09 72 63 e8 1f f4 ff ff 41 89 c4 48 8b 45 00 0f ba e0 10 72 0a 48 8b 45 08 a8 01 75 02 <0f> 0b 44 89 e1 48 c7 c2 00 f0 ff ff be 06 00 00 00 48 d3 e2 48 c7 RSP: 0018:ffffb42f40267e10 EFLAGS: 00010246 RAX: ffffd61280ee8d88 RBX: 0000000000000004 RCX: 000000008010000d RDX: 4000000000000000 RSI: ffffffffba1360b0 RDI: ffffd61280ee8d80 RBP: ffffd61280ee8d80 R08: ffffffffb91bebdf R09: 0000000000000000 R10: ffff8fe2c1047ac8 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000100 FS: 00007fe80eff6b68(0000) GS:ffff8fe339c00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fe80eec7bc0 CR3: 0000000038012000 CR4: 00000000000006f0 Call Trace: __free_fdtable+0x16/0x1f put_files_struct+0x81/0x9b do_exit+0x433/0x94d do_group_exit+0xa6/0xa6 __x64_sys_exit_group+0xf/0xf do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fe80ef64bea Code: Unable to access opcode bytes at RIP 0x7fe80ef64bc0. RSP: 002b:00007ffdb1c47528 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7 RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007fe80ef64bea RDX: 00007fe80ef64f60 RSI: 0000000000000000 RDI: 0000000000000000 RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000 R10: 00007fe80ee2c620 R11: 0000000000000246 R12: 00007fe80eff41e0 R13: 00000000ffffffff R14: 0000000000000024 R15: 00007fe80edf9cd0 Modules linked in: radeon(+) ath5k(+) snd_hda_codec_realtek ... Use a valid power_state index when initializing the "flags" and "misc" and "misc2" fields. Bug: https://bugzilla.kernel.org/show_bug.cgi?id=211537 Reported-by: Erhard F. <erhard_f@mailbox.org> Fixes: a48b9b4 ("drm/radeon/kms/pm: add asic specific callbacks for getting power state (v2)") Fixes: 79daedc ("drm/radeon/kms: minor pm cleanups") Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit d5027ca ] Ritesh reported a bug [1] against UML, noting that it crashed on startup. The backtrace shows the following (heavily redacted): (gdb) bt ... #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72 ... #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359 ... #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486 #45 0x00007f8990968b85 in __getgrnam_r [...] #46 0x00007f89909d6b77 in grantpt [...] #47 0x00007f8990a9394e in __GI_openpty [...] #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144 indicating that the UML function openpty_cb() calls openpty(), which internally calls __getgrnam_r(), which causes the nsswitch machinery to get started. This loads, through lots of indirection that I snipped, the libcom_err.so.2 library, which (in an unknown function, "??") calls sem_init(). Now, of course it wants to get libpthread's sem_init(), since it's linked against libpthread. However, the dynamic linker looks up that symbol against the binary first, and gets the kernel's sem_init(). Hajime Tazaki noted that "objcopy -L" can localize a symbol, so the dynamic linker wouldn't do the lookup this way. I tried, but for some reason that didn't seem to work. Doing the same thing in the linker script instead does seem to work, though I cannot entirely explain - it *also* works if I just add "VERSION { { global: *; }; }" instead, indicating that something else is happening that I don't really understand. It may be that explicitly doing that marks them with some kind of empty version, and that's different from the default. Explicitly marking them with a version breaks kallsyms, so that doesn't seem to be possible. Marking all the symbols as local seems correct, and does seem to address the issue, so do that. Also do it for static link, nsswitch libraries could still be loaded there. [1] https://bugs.debian.org/983379 Reported-by: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com> Tested-By: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit d5027ca ] Ritesh reported a bug [1] against UML, noting that it crashed on startup. The backtrace shows the following (heavily redacted): (gdb) bt ... #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72 ... #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359 ... #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486 #45 0x00007f8990968b85 in __getgrnam_r [...] #46 0x00007f89909d6b77 in grantpt [...] #47 0x00007f8990a9394e in __GI_openpty [...] #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144 indicating that the UML function openpty_cb() calls openpty(), which internally calls __getgrnam_r(), which causes the nsswitch machinery to get started. This loads, through lots of indirection that I snipped, the libcom_err.so.2 library, which (in an unknown function, "??") calls sem_init(). Now, of course it wants to get libpthread's sem_init(), since it's linked against libpthread. However, the dynamic linker looks up that symbol against the binary first, and gets the kernel's sem_init(). Hajime Tazaki noted that "objcopy -L" can localize a symbol, so the dynamic linker wouldn't do the lookup this way. I tried, but for some reason that didn't seem to work. Doing the same thing in the linker script instead does seem to work, though I cannot entirely explain - it *also* works if I just add "VERSION { { global: *; }; }" instead, indicating that something else is happening that I don't really understand. It may be that explicitly doing that marks them with some kind of empty version, and that's different from the default. Explicitly marking them with a version breaks kallsyms, so that doesn't seem to be possible. Marking all the symbols as local seems correct, and does seem to address the issue, so do that. Also do it for static link, nsswitch libraries could still be loaded there. [1] https://bugs.debian.org/983379 Reported-by: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com> Tested-By: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit 5bbf219 ] An out of bounds write happens when setting the default power state. KASAN sees this as: [drm] radeon: 512M of GTT memory ready. [drm] GART: num cpu pages 131072, num gpu pages 131072 ================================================================== BUG: KASAN: slab-out-of-bounds in radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] Write of size 4 at addr ffff88810178d858 by task systemd-udevd/157 CPU: 0 PID: 157 Comm: systemd-udevd Not tainted 5.12.0-E620 #50 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 Call Trace: dump_stack+0xa5/0xe6 print_address_description.constprop.0+0x18/0x239 kasan_report+0x170/0x1a8 radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] radeon_atombios_get_power_modes+0x144/0x1888 [radeon] radeon_pm_init+0x1019/0x1904 [radeon] rs690_init+0x76e/0x84a [radeon] radeon_device_init+0x1c1a/0x21e5 [radeon] radeon_driver_load_kms+0xf5/0x30b [radeon] drm_dev_register+0x255/0x4a0 [drm] radeon_pci_probe+0x246/0x2f6 [radeon] pci_device_probe+0x1aa/0x294 really_probe+0x30e/0x850 driver_probe_device+0xe6/0x135 device_driver_attach+0xc1/0xf8 __driver_attach+0x13f/0x146 bus_for_each_dev+0xfa/0x146 bus_add_driver+0x2b3/0x447 driver_register+0x242/0x2c1 do_one_initcall+0x149/0x2fd do_init_module+0x1ae/0x573 load_module+0x4dee/0x5cca __do_sys_finit_module+0xf1/0x140 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xae Without KASAN, this will manifest later when the kernel attempts to allocate memory that was stomped, since it collides with the inline slab freelist pointer: invalid opcode: 0000 [#1] SMP NOPTI CPU: 0 PID: 781 Comm: openrc-run.sh Tainted: G W 5.10.12-gentoo-E620 #2 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 RIP: 0010:kfree+0x115/0x230 Code: 89 c5 e8 75 ea ff ff 48 8b 00 0f ba e0 09 72 63 e8 1f f4 ff ff 41 89 c4 48 8b 45 00 0f ba e0 10 72 0a 48 8b 45 08 a8 01 75 02 <0f> 0b 44 89 e1 48 c7 c2 00 f0 ff ff be 06 00 00 00 48 d3 e2 48 c7 RSP: 0018:ffffb42f40267e10 EFLAGS: 00010246 RAX: ffffd61280ee8d88 RBX: 0000000000000004 RCX: 000000008010000d RDX: 4000000000000000 RSI: ffffffffba1360b0 RDI: ffffd61280ee8d80 RBP: ffffd61280ee8d80 R08: ffffffffb91bebdf R09: 0000000000000000 R10: ffff8fe2c1047ac8 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000100 FS: 00007fe80eff6b68(0000) GS:ffff8fe339c00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fe80eec7bc0 CR3: 0000000038012000 CR4: 00000000000006f0 Call Trace: __free_fdtable+0x16/0x1f put_files_struct+0x81/0x9b do_exit+0x433/0x94d do_group_exit+0xa6/0xa6 __x64_sys_exit_group+0xf/0xf do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fe80ef64bea Code: Unable to access opcode bytes at RIP 0x7fe80ef64bc0. RSP: 002b:00007ffdb1c47528 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7 RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007fe80ef64bea RDX: 00007fe80ef64f60 RSI: 0000000000000000 RDI: 0000000000000000 RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000 R10: 00007fe80ee2c620 R11: 0000000000000246 R12: 00007fe80eff41e0 R13: 00000000ffffffff R14: 0000000000000024 R15: 00007fe80edf9cd0 Modules linked in: radeon(+) ath5k(+) snd_hda_codec_realtek ... Use a valid power_state index when initializing the "flags" and "misc" and "misc2" fields. Bug: https://bugzilla.kernel.org/show_bug.cgi?id=211537 Reported-by: Erhard F. <erhard_f@mailbox.org> Fixes: a48b9b4 ("drm/radeon/kms/pm: add asic specific callbacks for getting power state (v2)") Fixes: 79daedc ("drm/radeon/kms: minor pm cleanups") Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit d5027ca ] Ritesh reported a bug [1] against UML, noting that it crashed on startup. The backtrace shows the following (heavily redacted): (gdb) bt ... #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72 ... #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359 ... #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486 #45 0x00007f8990968b85 in __getgrnam_r [...] #46 0x00007f89909d6b77 in grantpt [...] #47 0x00007f8990a9394e in __GI_openpty [...] #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144 indicating that the UML function openpty_cb() calls openpty(), which internally calls __getgrnam_r(), which causes the nsswitch machinery to get started. This loads, through lots of indirection that I snipped, the libcom_err.so.2 library, which (in an unknown function, "??") calls sem_init(). Now, of course it wants to get libpthread's sem_init(), since it's linked against libpthread. However, the dynamic linker looks up that symbol against the binary first, and gets the kernel's sem_init(). Hajime Tazaki noted that "objcopy -L" can localize a symbol, so the dynamic linker wouldn't do the lookup this way. I tried, but for some reason that didn't seem to work. Doing the same thing in the linker script instead does seem to work, though I cannot entirely explain - it *also* works if I just add "VERSION { { global: *; }; }" instead, indicating that something else is happening that I don't really understand. It may be that explicitly doing that marks them with some kind of empty version, and that's different from the default. Explicitly marking them with a version breaks kallsyms, so that doesn't seem to be possible. Marking all the symbols as local seems correct, and does seem to address the issue, so do that. Also do it for static link, nsswitch libraries could still be loaded there. [1] https://bugs.debian.org/983379 Reported-by: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com> Tested-By: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit 5bbf219 ] An out of bounds write happens when setting the default power state. KASAN sees this as: [drm] radeon: 512M of GTT memory ready. [drm] GART: num cpu pages 131072, num gpu pages 131072 ================================================================== BUG: KASAN: slab-out-of-bounds in radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] Write of size 4 at addr ffff88810178d858 by task systemd-udevd/157 CPU: 0 PID: 157 Comm: systemd-udevd Not tainted 5.12.0-E620 #50 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 Call Trace: dump_stack+0xa5/0xe6 print_address_description.constprop.0+0x18/0x239 kasan_report+0x170/0x1a8 radeon_atombios_parse_power_table_1_3+0x1837/0x1998 [radeon] radeon_atombios_get_power_modes+0x144/0x1888 [radeon] radeon_pm_init+0x1019/0x1904 [radeon] rs690_init+0x76e/0x84a [radeon] radeon_device_init+0x1c1a/0x21e5 [radeon] radeon_driver_load_kms+0xf5/0x30b [radeon] drm_dev_register+0x255/0x4a0 [drm] radeon_pci_probe+0x246/0x2f6 [radeon] pci_device_probe+0x1aa/0x294 really_probe+0x30e/0x850 driver_probe_device+0xe6/0x135 device_driver_attach+0xc1/0xf8 __driver_attach+0x13f/0x146 bus_for_each_dev+0xfa/0x146 bus_add_driver+0x2b3/0x447 driver_register+0x242/0x2c1 do_one_initcall+0x149/0x2fd do_init_module+0x1ae/0x573 load_module+0x4dee/0x5cca __do_sys_finit_module+0xf1/0x140 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xae Without KASAN, this will manifest later when the kernel attempts to allocate memory that was stomped, since it collides with the inline slab freelist pointer: invalid opcode: 0000 [#1] SMP NOPTI CPU: 0 PID: 781 Comm: openrc-run.sh Tainted: G W 5.10.12-gentoo-E620 #2 Hardware name: eMachines eMachines E620 /Nile , BIOS V1.03 09/30/2008 RIP: 0010:kfree+0x115/0x230 Code: 89 c5 e8 75 ea ff ff 48 8b 00 0f ba e0 09 72 63 e8 1f f4 ff ff 41 89 c4 48 8b 45 00 0f ba e0 10 72 0a 48 8b 45 08 a8 01 75 02 <0f> 0b 44 89 e1 48 c7 c2 00 f0 ff ff be 06 00 00 00 48 d3 e2 48 c7 RSP: 0018:ffffb42f40267e10 EFLAGS: 00010246 RAX: ffffd61280ee8d88 RBX: 0000000000000004 RCX: 000000008010000d RDX: 4000000000000000 RSI: ffffffffba1360b0 RDI: ffffd61280ee8d80 RBP: ffffd61280ee8d80 R08: ffffffffb91bebdf R09: 0000000000000000 R10: ffff8fe2c1047ac8 R11: 0000000000000000 R12: 0000000000000000 R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000100 FS: 00007fe80eff6b68(0000) GS:ffff8fe339c00000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fe80eec7bc0 CR3: 0000000038012000 CR4: 00000000000006f0 Call Trace: __free_fdtable+0x16/0x1f put_files_struct+0x81/0x9b do_exit+0x433/0x94d do_group_exit+0xa6/0xa6 __x64_sys_exit_group+0xf/0xf do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fe80ef64bea Code: Unable to access opcode bytes at RIP 0x7fe80ef64bc0. RSP: 002b:00007ffdb1c47528 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7 RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007fe80ef64bea RDX: 00007fe80ef64f60 RSI: 0000000000000000 RDI: 0000000000000000 RBP: 0000000000000000 R08: 0000000000000001 R09: 0000000000000000 R10: 00007fe80ee2c620 R11: 0000000000000246 R12: 00007fe80eff41e0 R13: 00000000ffffffff R14: 0000000000000024 R15: 00007fe80edf9cd0 Modules linked in: radeon(+) ath5k(+) snd_hda_codec_realtek ... Use a valid power_state index when initializing the "flags" and "misc" and "misc2" fields. Bug: https://bugzilla.kernel.org/show_bug.cgi?id=211537 Reported-by: Erhard F. <erhard_f@mailbox.org> Fixes: a48b9b4 ("drm/radeon/kms/pm: add asic specific callbacks for getting power state (v2)") Fixes: 79daedc ("drm/radeon/kms: minor pm cleanups") Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 3, 2021
[ Upstream commit d5027ca ] Ritesh reported a bug [1] against UML, noting that it crashed on startup. The backtrace shows the following (heavily redacted): (gdb) bt ... #26 0x0000000060015b5d in sem_init () at ipc/sem.c:268 #27 0x00007f89906d92f7 in ?? () from /lib/x86_64-linux-gnu/libcom_err.so.2 #28 0x00007f8990ab8fb2 in call_init (...) at dl-init.c:72 ... #40 0x00007f89909bf3a6 in nss_load_library (...) at nsswitch.c:359 ... #44 0x00007f8990895e35 in _nss_compat_getgrnam_r (...) at nss_compat/compat-grp.c:486 #45 0x00007f8990968b85 in __getgrnam_r [...] #46 0x00007f89909d6b77 in grantpt [...] #47 0x00007f8990a9394e in __GI_openpty [...] #48 0x00000000604a1f65 in openpty_cb (...) at arch/um/os-Linux/sigio.c:407 #49 0x00000000604a58d0 in start_idle_thread (...) at arch/um/os-Linux/skas/process.c:598 #50 0x0000000060004a3d in start_uml () at arch/um/kernel/skas/process.c:45 #51 0x00000000600047b2 in linux_main (...) at arch/um/kernel/um_arch.c:334 #52 0x000000006000574f in main (...) at arch/um/os-Linux/main.c:144 indicating that the UML function openpty_cb() calls openpty(), which internally calls __getgrnam_r(), which causes the nsswitch machinery to get started. This loads, through lots of indirection that I snipped, the libcom_err.so.2 library, which (in an unknown function, "??") calls sem_init(). Now, of course it wants to get libpthread's sem_init(), since it's linked against libpthread. However, the dynamic linker looks up that symbol against the binary first, and gets the kernel's sem_init(). Hajime Tazaki noted that "objcopy -L" can localize a symbol, so the dynamic linker wouldn't do the lookup this way. I tried, but for some reason that didn't seem to work. Doing the same thing in the linker script instead does seem to work, though I cannot entirely explain - it *also* works if I just add "VERSION { { global: *; }; }" instead, indicating that something else is happening that I don't really understand. It may be that explicitly doing that marks them with some kind of empty version, and that's different from the default. Explicitly marking them with a version breaks kallsyms, so that doesn't seem to be possible. Marking all the symbols as local seems correct, and does seem to address the issue, so do that. Also do it for static link, nsswitch libraries could still be loaded there. [1] https://bugs.debian.org/983379 Reported-by: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Johannes Berg <johannes.berg@intel.com> Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com> Tested-By: Ritesh Raj Sarraf <rrs@debian.org> Signed-off-by: Richard Weinberger <richard@nod.at> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 11, 2022
Problem:
=======
Userspace might read the zero-page instead of actual data from a direct IO
read on a block device if the buffers have been called madvise(MADV_FREE)
on earlier (this is discussed below) due to a race between page reclaim on
MADV_FREE and blkdev direct IO read.
Race condition:
==============
During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks
if the page is not dirty, then discards its PTE (vs remap it back if the
page is dirty).
However, after try_to_unmap_one() returns to shrink_page_list(), it might
keep the page _anyway_ if page_ref_freeze() fails (it expects a single
page ref from the isolation).
Well, blkdev_direct_IO() gets references for all pages, and on READ
operations it sets them dirty later.
So, if MADV_FREE pages (i.e., not dirty) are used as buffers (more later)
for direct IO read from block devices and page reclaim runs during
__blkdev_direct_IO[_simple]() AFTER bio_iov_iter_get_pages() but BEFORE it
sets pages dirty, that situation happens.
The direct IO read eventually completes. Now, when userspace reads the
buffers, the PTE is no longer there and the page fault handler
do_anonymous_page() services that with the zero-page, NOT the data!
A synthetic reproducer is provided.
Page faults:
===========
The data read from the block device probably won't generate faults due to
DMA (no MMU) but even in the case it wouldn't use DMA, that happens on
different virtual addresses (not user-mapped addresses) because `struct
bio_vec` stores `struct page` to figure addresses out (which are different
from/unrelated to user-mapped addresses) for the data read.
Thus userspace reads (to user-mapped addresses) still fault, then
do_anonymous_page() gets another `struct page` that would address/ map to
other memory than the `struct page` used by `struct bio_vec` for the read
(which runs correctly as the page wasn't freed due to page_ref_freeze(),
and is reclaimed later) -- but even if the page addresses matched, that
handler maps the zero-page in the PTE, not that page's memory (on read
faults.)
If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't
happen, because that faults-in all pages as writeable, so
do_anonymous_page() sets up a new page/rmap/PTE, and that is used by
direct IO. The userspace reads don't fault as the PTE is there (thus
zero-page is not used.)
Solution:
========
One solution is to check for the expected page reference count in
try_to_unmap_one() too, which should be exactly two: one from the
isolation (checked by shrink_page_list()), and the other from the rmap
(dropped by the discard: label). If that doesn't match, then remap the
PTE back, just like page dirty does.
The new check in try_to_unmap_one() should be safe in races with
bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's
done under the PTE lock. The fast path doesn't take that lock but it
checks the PTE has changed, then drops the reference and leaves the page
for the slow path (which does take that lock).
- try_to_unmap_one()
- page_vma_mapped_walk()
- map_pte() # see pte_offset_map_lock():
pte_offset_map()
spin_lock()
- page_ref_count() # new check
- page_vma_mapped_walk_done() # see pte_unmap_unlock():
pte_unmap()
spin_unlock()
- bio_iov_iter_get_pages()
- __bio_iov_iter_get_pages()
- iov_iter_get_pages()
- get_user_pages_fast()
- internal_get_user_pages_fast()
# fast path
- lockless_pages_from_mm()
- gup_{pgd,p4d,pud,pmd,pte}_range()
ptep = pte_offset_map() # not _lock()
pte = ptep_get_lockless(ptep)
page = pte_page(pte)
try_grab_compound_head(page) # get ref
if (pte_val(pte) != pte_val(*ptep))
put_compound_head(page) # put ref
# leave page for slow path
# slow path
- __gup_longterm_unlocked()
- get_user_pages_unlocked()
- __get_user_pages_locked()
- __get_user_pages()
- follow_{page,p4d,pud,pmd}_mask()
- follow_page_pte()
ptep = pte_offset_map_lock()
pte = *ptep
page = vm_normal_page(pte)
try_grab_page(page) # get ref
pte_unmap_unlock()
Regarding transparent hugepages, that number shouldn't change, as
MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked()
(madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn())
thus should reach shrink_page_list() -> split_huge_page_to_list() before
try_to_unmap[_one](), so it deals with normal pages only.
(And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens,
which it should not or be rare, the page refcount is not two, as the head
page counts tail pages, and tail pages have zero. That also prevents
checking the head `page` then incorrectly call page_remove_rmap(subpage)
for a tail page, that isn't even in the shrink_page_list()'s page_list (an
effect of split huge pmd/pmvw), as it might happen today in this unlikely
scenario.)
MADV_FREE'd buffers:
===================
So, back to the "if MADV_FREE pages are used as buffers" note. The case
is arguable, and subject to multiple interpretations.
The madvise(2) manual page on the MADV_FREE advice value says:
- 'After a successful MADV_FREE ... data will be lost when
the kernel frees the pages.'
- 'the free operation will be canceled if the caller writes
into the page' / 'subsequent writes ... will succeed and
then [the] kernel cannot free those dirtied pages'
- 'If there is no subsequent write, the kernel can free the
pages at any time.'
Thoughts, questions, considerations...
- Since the kernel didn't actually free the page (page_ref_freeze()
failed), should the data not have been lost? (on userspace read.)
- Should writes performed by the direct IO read be able to cancel
the free operation?
- Should the direct IO read be considered as 'the caller' too,
as it's been requested by 'the caller'?
- Should the bio technique to dirty pages on return to userspace
(bio_check_pages_dirty() is called/used by __blkdev_direct_IO())
be considered in another/special way here?
- Should an upcoming write from a previously requested direct IO
read be considered as a subsequent write, so the kernel should
not free the pages? (as it's known at the time of page reclaim.)
Technically, the last point would seem a reasonable consideration and
balance, as the madvise(2) manual page apparently (and fairly) seem to
assume that 'writes' are memory access from the userspace process (not
explicitly considering writes from the kernel or its corner cases; again,
fairly).. plus the kernel fix implementation for the corner case of the
largely 'non-atomic write' encompassed by a direct IO read operation, is
relatively simple; and it helps.
Reproducer:
==========
@ test.c (simplified, but works)
#define _GNU_SOURCE
#include <fcntl.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
int main() {
int fd, i;
char *buf;
fd = open(DEV, O_RDONLY | O_DIRECT);
buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
buf[i] = 1; // init to non-zero
madvise(buf, BUF_SIZE, MADV_FREE);
read(fd, buf, BUF_SIZE);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
printf("%p: 0x%x
", &buf[i], buf[i]);
return 0;
}
@ block/fops.c (formerly fs/block_dev.c)
+#include <linux/swap.h>
...
... __blkdev_direct_IO[_simple](...)
{
...
+ if (!strcmp(current->comm, "good"))
+ shrink_all_memory(ULONG_MAX);
+
ret = bio_iov_iter_get_pages(...);
+
+ if (!strcmp(current->comm, "bad"))
+ shrink_all_memory(ULONG_MAX);
...
}
@ shell
# yes | dd of=test.img bs=1k count=16
# DEV=$(losetup -f --show test.img)
# gcc -DDEV=\"$DEV\" -DBUF_SIZE=16384 -DPAGE_SIZE=4096 test.c -o test
# od -tx1 $DEV
0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a
*
0040000
# mv test good
# ./good
0x7f1509206000: 0x79
0x7f1509207000: 0x79
0x7f1509208000: 0x79
0x7f1509209000: 0x79
# mv good bad
# ./bad
0x7fd87272f000: 0x0
0x7fd872730000: 0x0
0x7fd872731000: 0x0
0x7fd872732000: 0x0
Ceph/TCMalloc:
=============
For documentation purposes, the use case driving the analysis/fix is Ceph
on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to
release unused memory to the system from the mmap'ed page heap (might be
committed back/used again; it's not munmap'ed.)
- PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise()
- PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing.
Note: TCMalloc switched back to MADV_DONTNEED a few commits after the
release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue
just 'disappeared' on Ceph on later Ubuntu releases but is still present
in the kernel, and can be hit by other use cases.
The observed issue seems to be the old Ceph bug #22464 [1], where checksum
mismatches are observed (and instrumentation with buffer dumps shows
zero-pages read from mmap'ed/MADV_FREE'd page ranges).
The issue in Ceph was reasonably deemed a kernel bug (comment #50) and
mostly worked around with a retry mechanism, but other parts of Ceph could
still hit that (rocksdb). Anyway, it's less likely to be hit again as
TCMalloc switched out of MADV_FREE by default.
(Some kernel versions/reports from the Ceph bug, and relation with
the MADV_FREE introduction/changes; TCMalloc versions not checked.)
- 4.4 good
- 4.5 (madv_free: introduction)
- 4.9 bad
- 4.10 good? maybe a swapless system
- 4.12 (madv_free: no longer free instantly on swapless systems)
- 4.13 bad
[1] https://tracker.ceph.com/issues/22464
Thanks:
======
Several people contributed to analysis/discussions/tests/reproducers
in the first stages when drilling down on ceph/tcmalloc/linux kernel:
- Dan Hill
- Dan Streetman
- Dongdong Tao
- Gavin Guo
- Gerald Yang
- Heitor Alves de Siqueira
- Ioanna Alifieraki
- Jay Vosburgh
- Matthew Ruffell
- Ponnuvel Palaniyappan
Link: https://lkml.kernel.org/r/20211211022115.1547617-1-mfo@canonical.com
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Dan Hill <daniel.hill@canonical.com>
Cc: Dan Streetman <dan.streetman@canonical.com>
Cc: Dongdong Tao <dongdong.tao@canonical.com>
Cc: Gavin Guo <gavin.guo@canonical.com>
Cc: Gerald Yang <gerald.yang@canonical.com>
Cc: Heitor Alves de Siqueira <halves@canonical.com>
Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com>
Cc: Jay Vosburgh <jay.vosburgh@canonical.com>
Cc: Matthew Ruffell <matthew.ruffell@canonical.com>
Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
frank-w
pushed a commit
that referenced
this pull request
May 11, 2022
Problem:
=======
Userspace might read the zero-page instead of actual data from a direct IO
read on a block device if the buffers have been called madvise(MADV_FREE)
on earlier (this is discussed below) due to a race between page reclaim on
MADV_FREE and blkdev direct IO read.
Race condition:
==============
During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks
if the page is not dirty, then discards its PTE (vs remap it back if the
page is dirty).
However, after try_to_unmap_one() returns to shrink_page_list(), it might
keep the page _anyway_ if page_ref_freeze() fails (it expects a single
page ref from the isolation).
Well, blkdev_direct_IO() gets references for all pages, and on READ
operations it sets them dirty later.
So, if MADV_FREE pages (i.e., not dirty) are used as buffers (more later)
for direct IO read from block devices and page reclaim runs during
__blkdev_direct_IO[_simple]() AFTER bio_iov_iter_get_pages() but BEFORE it
sets pages dirty, that situation happens.
The direct IO read eventually completes. Now, when userspace reads the
buffers, the PTE is no longer there and the page fault handler
do_anonymous_page() services that with the zero-page, NOT the data!
A synthetic reproducer is provided.
Page faults:
===========
The data read from the block device probably won't generate faults due to
DMA (no MMU) but even in the case it wouldn't use DMA, that happens on
different virtual addresses (not user-mapped addresses) because `struct
bio_vec` stores `struct page` to figure addresses out (which are different
from/unrelated to user-mapped addresses) for the data read.
Thus userspace reads (to user-mapped addresses) still fault, then
do_anonymous_page() gets another `struct page` that would address/ map to
other memory than the `struct page` used by `struct bio_vec` for the read
(which runs correctly as the page wasn't freed due to page_ref_freeze(),
and is reclaimed later) -- but even if the page addresses matched, that
handler maps the zero-page in the PTE, not that page's memory (on read
faults.)
If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't
happen, because that faults-in all pages as writeable, so
do_anonymous_page() sets up a new page/rmap/PTE, and that is used by
direct IO. The userspace reads don't fault as the PTE is there (thus
zero-page is not used.)
Solution:
========
One solution is to check for the expected page reference count in
try_to_unmap_one() too, which should be exactly two: one from the
isolation (checked by shrink_page_list()), and the other from the rmap
(dropped by the discard: label). If that doesn't match, then remap the
PTE back, just like page dirty does.
The new check in try_to_unmap_one() should be safe in races with
bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's
done under the PTE lock. The fast path doesn't take that lock but it
checks the PTE has changed, then drops the reference and leaves the page
for the slow path (which does take that lock).
- try_to_unmap_one()
- page_vma_mapped_walk()
- map_pte() # see pte_offset_map_lock():
pte_offset_map()
spin_lock()
- page_ref_count() # new check
- page_vma_mapped_walk_done() # see pte_unmap_unlock():
pte_unmap()
spin_unlock()
- bio_iov_iter_get_pages()
- __bio_iov_iter_get_pages()
- iov_iter_get_pages()
- get_user_pages_fast()
- internal_get_user_pages_fast()
# fast path
- lockless_pages_from_mm()
- gup_{pgd,p4d,pud,pmd,pte}_range()
ptep = pte_offset_map() # not _lock()
pte = ptep_get_lockless(ptep)
page = pte_page(pte)
try_grab_compound_head(page) # get ref
if (pte_val(pte) != pte_val(*ptep))
put_compound_head(page) # put ref
# leave page for slow path
# slow path
- __gup_longterm_unlocked()
- get_user_pages_unlocked()
- __get_user_pages_locked()
- __get_user_pages()
- follow_{page,p4d,pud,pmd}_mask()
- follow_page_pte()
ptep = pte_offset_map_lock()
pte = *ptep
page = vm_normal_page(pte)
try_grab_page(page) # get ref
pte_unmap_unlock()
Regarding transparent hugepages, that number shouldn't change, as
MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked()
(madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn())
thus should reach shrink_page_list() -> split_huge_page_to_list() before
try_to_unmap[_one](), so it deals with normal pages only.
(And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens,
which it should not or be rare, the page refcount is not two, as the head
page counts tail pages, and tail pages have zero. That also prevents
checking the head `page` then incorrectly call page_remove_rmap(subpage)
for a tail page, that isn't even in the shrink_page_list()'s page_list (an
effect of split huge pmd/pmvw), as it might happen today in this unlikely
scenario.)
MADV_FREE'd buffers:
===================
So, back to the "if MADV_FREE pages are used as buffers" note. The case
is arguable, and subject to multiple interpretations.
The madvise(2) manual page on the MADV_FREE advice value says:
- 'After a successful MADV_FREE ... data will be lost when
the kernel frees the pages.'
- 'the free operation will be canceled if the caller writes
into the page' / 'subsequent writes ... will succeed and
then [the] kernel cannot free those dirtied pages'
- 'If there is no subsequent write, the kernel can free the
pages at any time.'
Thoughts, questions, considerations...
- Since the kernel didn't actually free the page (page_ref_freeze()
failed), should the data not have been lost? (on userspace read.)
- Should writes performed by the direct IO read be able to cancel
the free operation?
- Should the direct IO read be considered as 'the caller' too,
as it's been requested by 'the caller'?
- Should the bio technique to dirty pages on return to userspace
(bio_check_pages_dirty() is called/used by __blkdev_direct_IO())
be considered in another/special way here?
- Should an upcoming write from a previously requested direct IO
read be considered as a subsequent write, so the kernel should
not free the pages? (as it's known at the time of page reclaim.)
Technically, the last point would seem a reasonable consideration and
balance, as the madvise(2) manual page apparently (and fairly) seem to
assume that 'writes' are memory access from the userspace process (not
explicitly considering writes from the kernel or its corner cases; again,
fairly).. plus the kernel fix implementation for the corner case of the
largely 'non-atomic write' encompassed by a direct IO read operation, is
relatively simple; and it helps.
Reproducer:
==========
@ test.c (simplified, but works)
#define _GNU_SOURCE
#include <fcntl.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
int main() {
int fd, i;
char *buf;
fd = open(DEV, O_RDONLY | O_DIRECT);
buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
buf[i] = 1; // init to non-zero
madvise(buf, BUF_SIZE, MADV_FREE);
read(fd, buf, BUF_SIZE);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
printf("%p: 0x%x
", &buf[i], buf[i]);
return 0;
}
@ block/fops.c (formerly fs/block_dev.c)
+#include <linux/swap.h>
...
... __blkdev_direct_IO[_simple](...)
{
...
+ if (!strcmp(current->comm, "good"))
+ shrink_all_memory(ULONG_MAX);
+
ret = bio_iov_iter_get_pages(...);
+
+ if (!strcmp(current->comm, "bad"))
+ shrink_all_memory(ULONG_MAX);
...
}
@ shell
# yes | dd of=test.img bs=1k count=16
# DEV=$(losetup -f --show test.img)
# gcc -DDEV=\"$DEV\" -DBUF_SIZE=16384 -DPAGE_SIZE=4096 test.c -o test
# od -tx1 $DEV
0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a
*
0040000
# mv test good
# ./good
0x7f1509206000: 0x79
0x7f1509207000: 0x79
0x7f1509208000: 0x79
0x7f1509209000: 0x79
# mv good bad
# ./bad
0x7fd87272f000: 0x0
0x7fd872730000: 0x0
0x7fd872731000: 0x0
0x7fd872732000: 0x0
Ceph/TCMalloc:
=============
For documentation purposes, the use case driving the analysis/fix is Ceph
on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to
release unused memory to the system from the mmap'ed page heap (might be
committed back/used again; it's not munmap'ed.)
- PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise()
- PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing.
Note: TCMalloc switched back to MADV_DONTNEED a few commits after the
release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue
just 'disappeared' on Ceph on later Ubuntu releases but is still present
in the kernel, and can be hit by other use cases.
The observed issue seems to be the old Ceph bug #22464 [1], where checksum
mismatches are observed (and instrumentation with buffer dumps shows
zero-pages read from mmap'ed/MADV_FREE'd page ranges).
The issue in Ceph was reasonably deemed a kernel bug (comment #50) and
mostly worked around with a retry mechanism, but other parts of Ceph could
still hit that (rocksdb). Anyway, it's less likely to be hit again as
TCMalloc switched out of MADV_FREE by default.
(Some kernel versions/reports from the Ceph bug, and relation with
the MADV_FREE introduction/changes; TCMalloc versions not checked.)
- 4.4 good
- 4.5 (madv_free: introduction)
- 4.9 bad
- 4.10 good? maybe a swapless system
- 4.12 (madv_free: no longer free instantly on swapless systems)
- 4.13 bad
[1] https://tracker.ceph.com/issues/22464
Thanks:
======
Several people contributed to analysis/discussions/tests/reproducers
in the first stages when drilling down on ceph/tcmalloc/linux kernel:
- Dan Hill
- Dan Streetman
- Dongdong Tao
- Gavin Guo
- Gerald Yang
- Heitor Alves de Siqueira
- Ioanna Alifieraki
- Jay Vosburgh
- Matthew Ruffell
- Ponnuvel Palaniyappan
Link: https://lkml.kernel.org/r/20211211022115.1547617-1-mfo@canonical.com
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Dan Hill <daniel.hill@canonical.com>
Cc: Dan Streetman <dan.streetman@canonical.com>
Cc: Dongdong Tao <dongdong.tao@canonical.com>
Cc: Gavin Guo <gavin.guo@canonical.com>
Cc: Gerald Yang <gerald.yang@canonical.com>
Cc: Heitor Alves de Siqueira <halves@canonical.com>
Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com>
Cc: Jay Vosburgh <jay.vosburgh@canonical.com>
Cc: Matthew Ruffell <matthew.ruffell@canonical.com>
Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
frank-w
pushed a commit
that referenced
this pull request
May 25, 2022
commit 6c8e2a2 upstream. Problem: ======= Userspace might read the zero-page instead of actual data from a direct IO read on a block device if the buffers have been called madvise(MADV_FREE) on earlier (this is discussed below) due to a race between page reclaim on MADV_FREE and blkdev direct IO read. - Race condition: ============== During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks if the page is not dirty, then discards its rmap PTE(s) (vs. remap back if the page is dirty). However, after try_to_unmap_one() returns to shrink_page_list(), it might keep the page _anyway_ if page_ref_freeze() fails (it expects exactly _one_ page reference, from the isolation for page reclaim). Well, blkdev_direct_IO() gets references for all pages, and on READ operations it only sets them dirty _later_. So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct IO read from block devices, and page reclaim happens during __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages() returns, but BEFORE the pages are set dirty, the situation happens. The direct IO read eventually completes. Now, when userspace reads the buffers, the PTE is no longer there and the page fault handler do_anonymous_page() services that with the zero-page, NOT the data! A synthetic reproducer is provided. - Page faults: =========== If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't happen, because that faults-in all pages as writeable, so do_anonymous_page() sets up a new page/rmap/PTE, and that is used by direct IO. The userspace reads don't fault as the PTE is there (thus zero-page is not used/setup). But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE is no longer there; the subsequent page faults can't help: The data-read from the block device probably won't generate faults due to DMA (no MMU) but even in the case it wouldn't use DMA, that happens on different virtual addresses (not user-mapped addresses) because `struct bio_vec` stores `struct page` to figure addresses out (which are different from user-mapped addresses) for the read. Thus userspace reads (to user-mapped addresses) still fault, then do_anonymous_page() gets another `struct page` that would address/ map to other memory than the `struct page` used by `struct bio_vec` for the read. (The original `struct page` is not available, since it wasn't freed, as page_ref_freeze() failed due to more page refs. And even if it were available, its data cannot be trusted anymore.) Solution: ======== One solution is to check for the expected page reference count in try_to_unmap_one(). There should be one reference from the isolation (that is also checked in shrink_page_list() with page_ref_freeze()) plus one or more references from page mapping(s) (put in discard: label). Further references mean that rmap/PTE cannot be unmapped/nuked. (Note: there might be more than one reference from mapping due to fork()/clone() without CLONE_VM, which use the same `struct page` for references, until the copy-on-write page gets copied.) So, additional page references (e.g., from direct IO read) now prevent the rmap/PTE from being unmapped/dropped; similarly to the page is not freed per shrink_page_list()/page_ref_freeze()). - Races and Barriers: ================== The new check in try_to_unmap_one() should be safe in races with bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's done under the PTE lock. The fast path doesn't take the lock, but it checks if the PTE has changed and if so, it drops the reference and leaves the page for the slow path (which does take that lock). The fast path requires synchronization w/ full memory barrier: it writes the page reference count first then it reads the PTE later, while try_to_unmap() writes PTE first then it reads page refcount. And a second barrier is needed, as the page dirty flag should not be read before the page reference count (as in __remove_mapping()). (This can be a load memory barrier only; no writes are involved.) Call stack/comments: - try_to_unmap_one() - page_vma_mapped_walk() - map_pte() # see pte_offset_map_lock(): pte_offset_map() spin_lock() - ptep_get_and_clear() # write PTE - smp_mb() # (new barrier) GUP fast path - page_ref_count() # (new check) read refcount - page_vma_mapped_walk_done() # see pte_unmap_unlock(): pte_unmap() spin_unlock() - bio_iov_iter_get_pages() - __bio_iov_iter_get_pages() - iov_iter_get_pages() - get_user_pages_fast() - internal_get_user_pages_fast() # fast path - lockless_pages_from_mm() - gup_{pgd,p4d,pud,pmd,pte}_range() ptep = pte_offset_map() # not _lock() pte = ptep_get_lockless(ptep) page = pte_page(pte) try_grab_compound_head(page) # inc refcount # (RMW/barrier # on success) if (pte_val(pte) != pte_val(*ptep)) # read PTE put_compound_head(page) # dec refcount # go slow path # slow path - __gup_longterm_unlocked() - get_user_pages_unlocked() - __get_user_pages_locked() - __get_user_pages() - follow_{page,p4d,pud,pmd}_mask() - follow_page_pte() ptep = pte_offset_map_lock() pte = *ptep page = vm_normal_page(pte) try_grab_page(page) # inc refcount pte_unmap_unlock() - Huge Pages: ========== Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked() (madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn()) thus should reach shrink_page_list() -> split_huge_page_to_list() before try_to_unmap[_one](), so it deals with normal pages only. (And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens, which should not or be rare, the page refcount should be greater than mapcount: the head page is referenced by tail pages. That also prevents checking the head `page` then incorrectly call page_remove_rmap(subpage) for a tail page, that isn't even in the shrink_page_list()'s page_list (an effect of split huge pmd/pmvw), as it might happen today in this unlikely scenario.) MADV_FREE'd buffers: =================== So, back to the "if MADV_FREE pages are used as buffers" note. The case is arguable, and subject to multiple interpretations. The madvise(2) manual page on the MADV_FREE advice value says: 1) 'After a successful MADV_FREE ... data will be lost when the kernel frees the pages.' 2) 'the free operation will be canceled if the caller writes into the page' / 'subsequent writes ... will succeed and then [the] kernel cannot free those dirtied pages' 3) 'If there is no subsequent write, the kernel can free the pages at any time.' Thoughts, questions, considerations... respectively: 1) Since the kernel didn't actually free the page (page_ref_freeze() failed), should the data not have been lost? (on userspace read.) 2) Should writes performed by the direct IO read be able to cancel the free operation? - Should the direct IO read be considered as 'the caller' too, as it's been requested by 'the caller'? - Should the bio technique to dirty pages on return to userspace (bio_check_pages_dirty() is called/used by __blkdev_direct_IO()) be considered in another/special way here? 3) Should an upcoming write from a previously requested direct IO read be considered as a subsequent write, so the kernel should not free the pages? (as it's known at the time of page reclaim.) And lastly: Technically, the last point would seem a reasonable consideration and balance, as the madvise(2) manual page apparently (and fairly) seem to assume that 'writes' are memory access from the userspace process (not explicitly considering writes from the kernel or its corner cases; again, fairly).. plus the kernel fix implementation for the corner case of the largely 'non-atomic write' encompassed by a direct IO read operation, is relatively simple; and it helps. Reproducer: ========== @ test.c (simplified, but works) #define _GNU_SOURCE #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> int main() { int fd, i; char *buf; fd = open(DEV, O_RDONLY | O_DIRECT); buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) buf[i] = 1; // init to non-zero madvise(buf, BUF_SIZE, MADV_FREE); read(fd, buf, BUF_SIZE); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) printf("%p: 0x%x\n", &buf[i], buf[i]); return 0; } @ block/fops.c (formerly fs/block_dev.c) +#include <linux/swap.h> ... ... __blkdev_direct_IO[_simple](...) { ... + if (!strcmp(current->comm, "good")) + shrink_all_memory(ULONG_MAX); + ret = bio_iov_iter_get_pages(...); + + if (!strcmp(current->comm, "bad")) + shrink_all_memory(ULONG_MAX); ... } @ shell # NUM_PAGES=4 # PAGE_SIZE=$(getconf PAGE_SIZE) # yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES} # DEV=$(losetup -f --show test.img) # gcc -DDEV=\"$DEV\" \ -DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \ -DPAGE_SIZE=${PAGE_SIZE} \ test.c -o test # od -tx1 $DEV 0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a * 0040000 # mv test good # ./good 0x7f7c10418000: 0x79 0x7f7c10419000: 0x79 0x7f7c1041a000: 0x79 0x7f7c1041b000: 0x79 # mv good bad # ./bad 0x7fa1b8050000: 0x0 0x7fa1b8051000: 0x0 0x7fa1b8052000: 0x0 0x7fa1b8053000: 0x0 Note: the issue is consistent on v5.17-rc3, but it's intermittent with the support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c]. - v5.17-rc3: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x0 # free | grep Swap Swap: 0 0 0 - v4.5: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 2702 0x0 1298 0x79 # swapoff -av swapoff /swap # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 Ceph/TCMalloc: ============= For documentation purposes, the use case driving the analysis/fix is Ceph on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to release unused memory to the system from the mmap'ed page heap (might be committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing. Note: TCMalloc switched back to MADV_DONTNEED a few commits after the release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue just 'disappeared' on Ceph on later Ubuntu releases but is still present in the kernel, and can be hit by other use cases. The observed issue seems to be the old Ceph bug #22464 [1], where checksum mismatches are observed (and instrumentation with buffer dumps shows zero-pages read from mmap'ed/MADV_FREE'd page ranges). The issue in Ceph was reasonably deemed a kernel bug (comment #50) and mostly worked around with a retry mechanism, but other parts of Ceph could still hit that (rocksdb). Anyway, it's less likely to be hit again as TCMalloc switched out of MADV_FREE by default. (Some kernel versions/reports from the Ceph bug, and relation with the MADV_FREE introduction/changes; TCMalloc versions not checked.) - 4.4 good - 4.5 (madv_free: introduction) - 4.9 bad - 4.10 good? maybe a swapless system - 4.12 (madv_free: no longer free instantly on swapless systems) - 4.13 bad [1] https://tracker.ceph.com/issues/22464 Thanks: ====== Several people contributed to analysis/discussions/tests/reproducers in the first stages when drilling down on ceph/tcmalloc/linux kernel: - Dan Hill - Dan Streetman - Dongdong Tao - Gavin Guo - Gerald Yang - Heitor Alves de Siqueira - Ioanna Alifieraki - Jay Vosburgh - Matthew Ruffell - Ponnuvel Palaniyappan Reviews, suggestions, corrections, comments: - Minchan Kim - Yu Zhao - Huang, Ying - John Hubbard - Christoph Hellwig [mfo@canonical.com: v4] Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com Fixes: 802a3a9 ("mm: reclaim MADV_FREE pages") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Dan Hill <daniel.hill@canonical.com> Cc: Dan Streetman <dan.streetman@canonical.com> Cc: Dongdong Tao <dongdong.tao@canonical.com> Cc: Gavin Guo <gavin.guo@canonical.com> Cc: Gerald Yang <gerald.yang@canonical.com> Cc: Heitor Alves de Siqueira <halves@canonical.com> Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Cc: Matthew Ruffell <matthew.ruffell@canonical.com> Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com> Cc: <stable@vger.kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [mfo: backport: replace folio/test_flag with page/flag equivalents; real Fixes: 854e9ed ("mm: support madvise(MADV_FREE)") in v4.] Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 25, 2022
commit 6c8e2a2 upstream. Problem: ======= Userspace might read the zero-page instead of actual data from a direct IO read on a block device if the buffers have been called madvise(MADV_FREE) on earlier (this is discussed below) due to a race between page reclaim on MADV_FREE and blkdev direct IO read. - Race condition: ============== During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks if the page is not dirty, then discards its rmap PTE(s) (vs. remap back if the page is dirty). However, after try_to_unmap_one() returns to shrink_page_list(), it might keep the page _anyway_ if page_ref_freeze() fails (it expects exactly _one_ page reference, from the isolation for page reclaim). Well, blkdev_direct_IO() gets references for all pages, and on READ operations it only sets them dirty _later_. So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct IO read from block devices, and page reclaim happens during __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages() returns, but BEFORE the pages are set dirty, the situation happens. The direct IO read eventually completes. Now, when userspace reads the buffers, the PTE is no longer there and the page fault handler do_anonymous_page() services that with the zero-page, NOT the data! A synthetic reproducer is provided. - Page faults: =========== If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't happen, because that faults-in all pages as writeable, so do_anonymous_page() sets up a new page/rmap/PTE, and that is used by direct IO. The userspace reads don't fault as the PTE is there (thus zero-page is not used/setup). But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE is no longer there; the subsequent page faults can't help: The data-read from the block device probably won't generate faults due to DMA (no MMU) but even in the case it wouldn't use DMA, that happens on different virtual addresses (not user-mapped addresses) because `struct bio_vec` stores `struct page` to figure addresses out (which are different from user-mapped addresses) for the read. Thus userspace reads (to user-mapped addresses) still fault, then do_anonymous_page() gets another `struct page` that would address/ map to other memory than the `struct page` used by `struct bio_vec` for the read. (The original `struct page` is not available, since it wasn't freed, as page_ref_freeze() failed due to more page refs. And even if it were available, its data cannot be trusted anymore.) Solution: ======== One solution is to check for the expected page reference count in try_to_unmap_one(). There should be one reference from the isolation (that is also checked in shrink_page_list() with page_ref_freeze()) plus one or more references from page mapping(s) (put in discard: label). Further references mean that rmap/PTE cannot be unmapped/nuked. (Note: there might be more than one reference from mapping due to fork()/clone() without CLONE_VM, which use the same `struct page` for references, until the copy-on-write page gets copied.) So, additional page references (e.g., from direct IO read) now prevent the rmap/PTE from being unmapped/dropped; similarly to the page is not freed per shrink_page_list()/page_ref_freeze()). - Races and Barriers: ================== The new check in try_to_unmap_one() should be safe in races with bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's done under the PTE lock. The fast path doesn't take the lock, but it checks if the PTE has changed and if so, it drops the reference and leaves the page for the slow path (which does take that lock). The fast path requires synchronization w/ full memory barrier: it writes the page reference count first then it reads the PTE later, while try_to_unmap() writes PTE first then it reads page refcount. And a second barrier is needed, as the page dirty flag should not be read before the page reference count (as in __remove_mapping()). (This can be a load memory barrier only; no writes are involved.) Call stack/comments: - try_to_unmap_one() - page_vma_mapped_walk() - map_pte() # see pte_offset_map_lock(): pte_offset_map() spin_lock() - ptep_get_and_clear() # write PTE - smp_mb() # (new barrier) GUP fast path - page_ref_count() # (new check) read refcount - page_vma_mapped_walk_done() # see pte_unmap_unlock(): pte_unmap() spin_unlock() - bio_iov_iter_get_pages() - __bio_iov_iter_get_pages() - iov_iter_get_pages() - get_user_pages_fast() - internal_get_user_pages_fast() # fast path - lockless_pages_from_mm() - gup_{pgd,p4d,pud,pmd,pte}_range() ptep = pte_offset_map() # not _lock() pte = ptep_get_lockless(ptep) page = pte_page(pte) try_grab_compound_head(page) # inc refcount # (RMW/barrier # on success) if (pte_val(pte) != pte_val(*ptep)) # read PTE put_compound_head(page) # dec refcount # go slow path # slow path - __gup_longterm_unlocked() - get_user_pages_unlocked() - __get_user_pages_locked() - __get_user_pages() - follow_{page,p4d,pud,pmd}_mask() - follow_page_pte() ptep = pte_offset_map_lock() pte = *ptep page = vm_normal_page(pte) try_grab_page(page) # inc refcount pte_unmap_unlock() - Huge Pages: ========== Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked() (madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn()) thus should reach shrink_page_list() -> split_huge_page_to_list() before try_to_unmap[_one](), so it deals with normal pages only. (And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens, which should not or be rare, the page refcount should be greater than mapcount: the head page is referenced by tail pages. That also prevents checking the head `page` then incorrectly call page_remove_rmap(subpage) for a tail page, that isn't even in the shrink_page_list()'s page_list (an effect of split huge pmd/pmvw), as it might happen today in this unlikely scenario.) MADV_FREE'd buffers: =================== So, back to the "if MADV_FREE pages are used as buffers" note. The case is arguable, and subject to multiple interpretations. The madvise(2) manual page on the MADV_FREE advice value says: 1) 'After a successful MADV_FREE ... data will be lost when the kernel frees the pages.' 2) 'the free operation will be canceled if the caller writes into the page' / 'subsequent writes ... will succeed and then [the] kernel cannot free those dirtied pages' 3) 'If there is no subsequent write, the kernel can free the pages at any time.' Thoughts, questions, considerations... respectively: 1) Since the kernel didn't actually free the page (page_ref_freeze() failed), should the data not have been lost? (on userspace read.) 2) Should writes performed by the direct IO read be able to cancel the free operation? - Should the direct IO read be considered as 'the caller' too, as it's been requested by 'the caller'? - Should the bio technique to dirty pages on return to userspace (bio_check_pages_dirty() is called/used by __blkdev_direct_IO()) be considered in another/special way here? 3) Should an upcoming write from a previously requested direct IO read be considered as a subsequent write, so the kernel should not free the pages? (as it's known at the time of page reclaim.) And lastly: Technically, the last point would seem a reasonable consideration and balance, as the madvise(2) manual page apparently (and fairly) seem to assume that 'writes' are memory access from the userspace process (not explicitly considering writes from the kernel or its corner cases; again, fairly).. plus the kernel fix implementation for the corner case of the largely 'non-atomic write' encompassed by a direct IO read operation, is relatively simple; and it helps. Reproducer: ========== @ test.c (simplified, but works) #define _GNU_SOURCE #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> int main() { int fd, i; char *buf; fd = open(DEV, O_RDONLY | O_DIRECT); buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) buf[i] = 1; // init to non-zero madvise(buf, BUF_SIZE, MADV_FREE); read(fd, buf, BUF_SIZE); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) printf("%p: 0x%x\n", &buf[i], buf[i]); return 0; } @ block/fops.c (formerly fs/block_dev.c) +#include <linux/swap.h> ... ... __blkdev_direct_IO[_simple](...) { ... + if (!strcmp(current->comm, "good")) + shrink_all_memory(ULONG_MAX); + ret = bio_iov_iter_get_pages(...); + + if (!strcmp(current->comm, "bad")) + shrink_all_memory(ULONG_MAX); ... } @ shell # NUM_PAGES=4 # PAGE_SIZE=$(getconf PAGE_SIZE) # yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES} # DEV=$(losetup -f --show test.img) # gcc -DDEV=\"$DEV\" \ -DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \ -DPAGE_SIZE=${PAGE_SIZE} \ test.c -o test # od -tx1 $DEV 0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a * 0040000 # mv test good # ./good 0x7f7c10418000: 0x79 0x7f7c10419000: 0x79 0x7f7c1041a000: 0x79 0x7f7c1041b000: 0x79 # mv good bad # ./bad 0x7fa1b8050000: 0x0 0x7fa1b8051000: 0x0 0x7fa1b8052000: 0x0 0x7fa1b8053000: 0x0 Note: the issue is consistent on v5.17-rc3, but it's intermittent with the support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c]. - v5.17-rc3: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x0 # free | grep Swap Swap: 0 0 0 - v4.5: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 2702 0x0 1298 0x79 # swapoff -av swapoff /swap # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 Ceph/TCMalloc: ============= For documentation purposes, the use case driving the analysis/fix is Ceph on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to release unused memory to the system from the mmap'ed page heap (might be committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing. Note: TCMalloc switched back to MADV_DONTNEED a few commits after the release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue just 'disappeared' on Ceph on later Ubuntu releases but is still present in the kernel, and can be hit by other use cases. The observed issue seems to be the old Ceph bug #22464 [1], where checksum mismatches are observed (and instrumentation with buffer dumps shows zero-pages read from mmap'ed/MADV_FREE'd page ranges). The issue in Ceph was reasonably deemed a kernel bug (comment #50) and mostly worked around with a retry mechanism, but other parts of Ceph could still hit that (rocksdb). Anyway, it's less likely to be hit again as TCMalloc switched out of MADV_FREE by default. (Some kernel versions/reports from the Ceph bug, and relation with the MADV_FREE introduction/changes; TCMalloc versions not checked.) - 4.4 good - 4.5 (madv_free: introduction) - 4.9 bad - 4.10 good? maybe a swapless system - 4.12 (madv_free: no longer free instantly on swapless systems) - 4.13 bad [1] https://tracker.ceph.com/issues/22464 Thanks: ====== Several people contributed to analysis/discussions/tests/reproducers in the first stages when drilling down on ceph/tcmalloc/linux kernel: - Dan Hill - Dan Streetman - Dongdong Tao - Gavin Guo - Gerald Yang - Heitor Alves de Siqueira - Ioanna Alifieraki - Jay Vosburgh - Matthew Ruffell - Ponnuvel Palaniyappan Reviews, suggestions, corrections, comments: - Minchan Kim - Yu Zhao - Huang, Ying - John Hubbard - Christoph Hellwig [mfo@canonical.com: v4] Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com Fixes: 802a3a9 ("mm: reclaim MADV_FREE pages") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Dan Hill <daniel.hill@canonical.com> Cc: Dan Streetman <dan.streetman@canonical.com> Cc: Dongdong Tao <dongdong.tao@canonical.com> Cc: Gavin Guo <gavin.guo@canonical.com> Cc: Gerald Yang <gerald.yang@canonical.com> Cc: Heitor Alves de Siqueira <halves@canonical.com> Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Cc: Matthew Ruffell <matthew.ruffell@canonical.com> Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com> Cc: <stable@vger.kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [mfo: backport: replace folio/test_flag with page/flag equivalents; real Fixes: 854e9ed ("mm: support madvise(MADV_FREE)") in v4.] Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 25, 2022
commit 6c8e2a2 upstream. Problem: ======= Userspace might read the zero-page instead of actual data from a direct IO read on a block device if the buffers have been called madvise(MADV_FREE) on earlier (this is discussed below) due to a race between page reclaim on MADV_FREE and blkdev direct IO read. - Race condition: ============== During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks if the page is not dirty, then discards its rmap PTE(s) (vs. remap back if the page is dirty). However, after try_to_unmap_one() returns to shrink_page_list(), it might keep the page _anyway_ if page_ref_freeze() fails (it expects exactly _one_ page reference, from the isolation for page reclaim). Well, blkdev_direct_IO() gets references for all pages, and on READ operations it only sets them dirty _later_. So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct IO read from block devices, and page reclaim happens during __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages() returns, but BEFORE the pages are set dirty, the situation happens. The direct IO read eventually completes. Now, when userspace reads the buffers, the PTE is no longer there and the page fault handler do_anonymous_page() services that with the zero-page, NOT the data! A synthetic reproducer is provided. - Page faults: =========== If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't happen, because that faults-in all pages as writeable, so do_anonymous_page() sets up a new page/rmap/PTE, and that is used by direct IO. The userspace reads don't fault as the PTE is there (thus zero-page is not used/setup). But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE is no longer there; the subsequent page faults can't help: The data-read from the block device probably won't generate faults due to DMA (no MMU) but even in the case it wouldn't use DMA, that happens on different virtual addresses (not user-mapped addresses) because `struct bio_vec` stores `struct page` to figure addresses out (which are different from user-mapped addresses) for the read. Thus userspace reads (to user-mapped addresses) still fault, then do_anonymous_page() gets another `struct page` that would address/ map to other memory than the `struct page` used by `struct bio_vec` for the read. (The original `struct page` is not available, since it wasn't freed, as page_ref_freeze() failed due to more page refs. And even if it were available, its data cannot be trusted anymore.) Solution: ======== One solution is to check for the expected page reference count in try_to_unmap_one(). There should be one reference from the isolation (that is also checked in shrink_page_list() with page_ref_freeze()) plus one or more references from page mapping(s) (put in discard: label). Further references mean that rmap/PTE cannot be unmapped/nuked. (Note: there might be more than one reference from mapping due to fork()/clone() without CLONE_VM, which use the same `struct page` for references, until the copy-on-write page gets copied.) So, additional page references (e.g., from direct IO read) now prevent the rmap/PTE from being unmapped/dropped; similarly to the page is not freed per shrink_page_list()/page_ref_freeze()). - Races and Barriers: ================== The new check in try_to_unmap_one() should be safe in races with bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's done under the PTE lock. The fast path doesn't take the lock, but it checks if the PTE has changed and if so, it drops the reference and leaves the page for the slow path (which does take that lock). The fast path requires synchronization w/ full memory barrier: it writes the page reference count first then it reads the PTE later, while try_to_unmap() writes PTE first then it reads page refcount. And a second barrier is needed, as the page dirty flag should not be read before the page reference count (as in __remove_mapping()). (This can be a load memory barrier only; no writes are involved.) Call stack/comments: - try_to_unmap_one() - page_vma_mapped_walk() - map_pte() # see pte_offset_map_lock(): pte_offset_map() spin_lock() - ptep_get_and_clear() # write PTE - smp_mb() # (new barrier) GUP fast path - page_ref_count() # (new check) read refcount - page_vma_mapped_walk_done() # see pte_unmap_unlock(): pte_unmap() spin_unlock() - bio_iov_iter_get_pages() - __bio_iov_iter_get_pages() - iov_iter_get_pages() - get_user_pages_fast() - internal_get_user_pages_fast() # fast path - lockless_pages_from_mm() - gup_{pgd,p4d,pud,pmd,pte}_range() ptep = pte_offset_map() # not _lock() pte = ptep_get_lockless(ptep) page = pte_page(pte) try_grab_compound_head(page) # inc refcount # (RMW/barrier # on success) if (pte_val(pte) != pte_val(*ptep)) # read PTE put_compound_head(page) # dec refcount # go slow path # slow path - __gup_longterm_unlocked() - get_user_pages_unlocked() - __get_user_pages_locked() - __get_user_pages() - follow_{page,p4d,pud,pmd}_mask() - follow_page_pte() ptep = pte_offset_map_lock() pte = *ptep page = vm_normal_page(pte) try_grab_page(page) # inc refcount pte_unmap_unlock() - Huge Pages: ========== Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked() (madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn()) thus should reach shrink_page_list() -> split_huge_page_to_list() before try_to_unmap[_one](), so it deals with normal pages only. (And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens, which should not or be rare, the page refcount should be greater than mapcount: the head page is referenced by tail pages. That also prevents checking the head `page` then incorrectly call page_remove_rmap(subpage) for a tail page, that isn't even in the shrink_page_list()'s page_list (an effect of split huge pmd/pmvw), as it might happen today in this unlikely scenario.) MADV_FREE'd buffers: =================== So, back to the "if MADV_FREE pages are used as buffers" note. The case is arguable, and subject to multiple interpretations. The madvise(2) manual page on the MADV_FREE advice value says: 1) 'After a successful MADV_FREE ... data will be lost when the kernel frees the pages.' 2) 'the free operation will be canceled if the caller writes into the page' / 'subsequent writes ... will succeed and then [the] kernel cannot free those dirtied pages' 3) 'If there is no subsequent write, the kernel can free the pages at any time.' Thoughts, questions, considerations... respectively: 1) Since the kernel didn't actually free the page (page_ref_freeze() failed), should the data not have been lost? (on userspace read.) 2) Should writes performed by the direct IO read be able to cancel the free operation? - Should the direct IO read be considered as 'the caller' too, as it's been requested by 'the caller'? - Should the bio technique to dirty pages on return to userspace (bio_check_pages_dirty() is called/used by __blkdev_direct_IO()) be considered in another/special way here? 3) Should an upcoming write from a previously requested direct IO read be considered as a subsequent write, so the kernel should not free the pages? (as it's known at the time of page reclaim.) And lastly: Technically, the last point would seem a reasonable consideration and balance, as the madvise(2) manual page apparently (and fairly) seem to assume that 'writes' are memory access from the userspace process (not explicitly considering writes from the kernel or its corner cases; again, fairly).. plus the kernel fix implementation for the corner case of the largely 'non-atomic write' encompassed by a direct IO read operation, is relatively simple; and it helps. Reproducer: ========== @ test.c (simplified, but works) #define _GNU_SOURCE #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> int main() { int fd, i; char *buf; fd = open(DEV, O_RDONLY | O_DIRECT); buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) buf[i] = 1; // init to non-zero madvise(buf, BUF_SIZE, MADV_FREE); read(fd, buf, BUF_SIZE); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) printf("%p: 0x%x\n", &buf[i], buf[i]); return 0; } @ block/fops.c (formerly fs/block_dev.c) +#include <linux/swap.h> ... ... __blkdev_direct_IO[_simple](...) { ... + if (!strcmp(current->comm, "good")) + shrink_all_memory(ULONG_MAX); + ret = bio_iov_iter_get_pages(...); + + if (!strcmp(current->comm, "bad")) + shrink_all_memory(ULONG_MAX); ... } @ shell # NUM_PAGES=4 # PAGE_SIZE=$(getconf PAGE_SIZE) # yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES} # DEV=$(losetup -f --show test.img) # gcc -DDEV=\"$DEV\" \ -DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \ -DPAGE_SIZE=${PAGE_SIZE} \ test.c -o test # od -tx1 $DEV 0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a * 0040000 # mv test good # ./good 0x7f7c10418000: 0x79 0x7f7c10419000: 0x79 0x7f7c1041a000: 0x79 0x7f7c1041b000: 0x79 # mv good bad # ./bad 0x7fa1b8050000: 0x0 0x7fa1b8051000: 0x0 0x7fa1b8052000: 0x0 0x7fa1b8053000: 0x0 Note: the issue is consistent on v5.17-rc3, but it's intermittent with the support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c]. - v5.17-rc3: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x0 # free | grep Swap Swap: 0 0 0 - v4.5: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 2702 0x0 1298 0x79 # swapoff -av swapoff /swap # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 Ceph/TCMalloc: ============= For documentation purposes, the use case driving the analysis/fix is Ceph on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to release unused memory to the system from the mmap'ed page heap (might be committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing. Note: TCMalloc switched back to MADV_DONTNEED a few commits after the release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue just 'disappeared' on Ceph on later Ubuntu releases but is still present in the kernel, and can be hit by other use cases. The observed issue seems to be the old Ceph bug #22464 [1], where checksum mismatches are observed (and instrumentation with buffer dumps shows zero-pages read from mmap'ed/MADV_FREE'd page ranges). The issue in Ceph was reasonably deemed a kernel bug (comment #50) and mostly worked around with a retry mechanism, but other parts of Ceph could still hit that (rocksdb). Anyway, it's less likely to be hit again as TCMalloc switched out of MADV_FREE by default. (Some kernel versions/reports from the Ceph bug, and relation with the MADV_FREE introduction/changes; TCMalloc versions not checked.) - 4.4 good - 4.5 (madv_free: introduction) - 4.9 bad - 4.10 good? maybe a swapless system - 4.12 (madv_free: no longer free instantly on swapless systems) - 4.13 bad [1] https://tracker.ceph.com/issues/22464 Thanks: ====== Several people contributed to analysis/discussions/tests/reproducers in the first stages when drilling down on ceph/tcmalloc/linux kernel: - Dan Hill - Dan Streetman - Dongdong Tao - Gavin Guo - Gerald Yang - Heitor Alves de Siqueira - Ioanna Alifieraki - Jay Vosburgh - Matthew Ruffell - Ponnuvel Palaniyappan Reviews, suggestions, corrections, comments: - Minchan Kim - Yu Zhao - Huang, Ying - John Hubbard - Christoph Hellwig [mfo@canonical.com: v4] Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com Fixes: 802a3a9 ("mm: reclaim MADV_FREE pages") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Dan Hill <daniel.hill@canonical.com> Cc: Dan Streetman <dan.streetman@canonical.com> Cc: Dongdong Tao <dongdong.tao@canonical.com> Cc: Gavin Guo <gavin.guo@canonical.com> Cc: Gerald Yang <gerald.yang@canonical.com> Cc: Heitor Alves de Siqueira <halves@canonical.com> Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Cc: Matthew Ruffell <matthew.ruffell@canonical.com> Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com> Cc: <stable@vger.kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [mfo: backport: replace folio/test_flag with page/flag equivalents; real Fixes: 854e9ed ("mm: support madvise(MADV_FREE)") in v4.] Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 25, 2022
commit 6c8e2a2 upstream. Problem: ======= Userspace might read the zero-page instead of actual data from a direct IO read on a block device if the buffers have been called madvise(MADV_FREE) on earlier (this is discussed below) due to a race between page reclaim on MADV_FREE and blkdev direct IO read. - Race condition: ============== During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks if the page is not dirty, then discards its rmap PTE(s) (vs. remap back if the page is dirty). However, after try_to_unmap_one() returns to shrink_page_list(), it might keep the page _anyway_ if page_ref_freeze() fails (it expects exactly _one_ page reference, from the isolation for page reclaim). Well, blkdev_direct_IO() gets references for all pages, and on READ operations it only sets them dirty _later_. So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct IO read from block devices, and page reclaim happens during __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages() returns, but BEFORE the pages are set dirty, the situation happens. The direct IO read eventually completes. Now, when userspace reads the buffers, the PTE is no longer there and the page fault handler do_anonymous_page() services that with the zero-page, NOT the data! A synthetic reproducer is provided. - Page faults: =========== If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't happen, because that faults-in all pages as writeable, so do_anonymous_page() sets up a new page/rmap/PTE, and that is used by direct IO. The userspace reads don't fault as the PTE is there (thus zero-page is not used/setup). But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE is no longer there; the subsequent page faults can't help: The data-read from the block device probably won't generate faults due to DMA (no MMU) but even in the case it wouldn't use DMA, that happens on different virtual addresses (not user-mapped addresses) because `struct bio_vec` stores `struct page` to figure addresses out (which are different from user-mapped addresses) for the read. Thus userspace reads (to user-mapped addresses) still fault, then do_anonymous_page() gets another `struct page` that would address/ map to other memory than the `struct page` used by `struct bio_vec` for the read. (The original `struct page` is not available, since it wasn't freed, as page_ref_freeze() failed due to more page refs. And even if it were available, its data cannot be trusted anymore.) Solution: ======== One solution is to check for the expected page reference count in try_to_unmap_one(). There should be one reference from the isolation (that is also checked in shrink_page_list() with page_ref_freeze()) plus one or more references from page mapping(s) (put in discard: label). Further references mean that rmap/PTE cannot be unmapped/nuked. (Note: there might be more than one reference from mapping due to fork()/clone() without CLONE_VM, which use the same `struct page` for references, until the copy-on-write page gets copied.) So, additional page references (e.g., from direct IO read) now prevent the rmap/PTE from being unmapped/dropped; similarly to the page is not freed per shrink_page_list()/page_ref_freeze()). - Races and Barriers: ================== The new check in try_to_unmap_one() should be safe in races with bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's done under the PTE lock. The fast path doesn't take the lock, but it checks if the PTE has changed and if so, it drops the reference and leaves the page for the slow path (which does take that lock). The fast path requires synchronization w/ full memory barrier: it writes the page reference count first then it reads the PTE later, while try_to_unmap() writes PTE first then it reads page refcount. And a second barrier is needed, as the page dirty flag should not be read before the page reference count (as in __remove_mapping()). (This can be a load memory barrier only; no writes are involved.) Call stack/comments: - try_to_unmap_one() - page_vma_mapped_walk() - map_pte() # see pte_offset_map_lock(): pte_offset_map() spin_lock() - ptep_get_and_clear() # write PTE - smp_mb() # (new barrier) GUP fast path - page_ref_count() # (new check) read refcount - page_vma_mapped_walk_done() # see pte_unmap_unlock(): pte_unmap() spin_unlock() - bio_iov_iter_get_pages() - __bio_iov_iter_get_pages() - iov_iter_get_pages() - get_user_pages_fast() - internal_get_user_pages_fast() # fast path - lockless_pages_from_mm() - gup_{pgd,p4d,pud,pmd,pte}_range() ptep = pte_offset_map() # not _lock() pte = ptep_get_lockless(ptep) page = pte_page(pte) try_grab_compound_head(page) # inc refcount # (RMW/barrier # on success) if (pte_val(pte) != pte_val(*ptep)) # read PTE put_compound_head(page) # dec refcount # go slow path # slow path - __gup_longterm_unlocked() - get_user_pages_unlocked() - __get_user_pages_locked() - __get_user_pages() - follow_{page,p4d,pud,pmd}_mask() - follow_page_pte() ptep = pte_offset_map_lock() pte = *ptep page = vm_normal_page(pte) try_grab_page(page) # inc refcount pte_unmap_unlock() - Huge Pages: ========== Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked() (madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn()) thus should reach shrink_page_list() -> split_huge_page_to_list() before try_to_unmap[_one](), so it deals with normal pages only. (And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens, which should not or be rare, the page refcount should be greater than mapcount: the head page is referenced by tail pages. That also prevents checking the head `page` then incorrectly call page_remove_rmap(subpage) for a tail page, that isn't even in the shrink_page_list()'s page_list (an effect of split huge pmd/pmvw), as it might happen today in this unlikely scenario.) MADV_FREE'd buffers: =================== So, back to the "if MADV_FREE pages are used as buffers" note. The case is arguable, and subject to multiple interpretations. The madvise(2) manual page on the MADV_FREE advice value says: 1) 'After a successful MADV_FREE ... data will be lost when the kernel frees the pages.' 2) 'the free operation will be canceled if the caller writes into the page' / 'subsequent writes ... will succeed and then [the] kernel cannot free those dirtied pages' 3) 'If there is no subsequent write, the kernel can free the pages at any time.' Thoughts, questions, considerations... respectively: 1) Since the kernel didn't actually free the page (page_ref_freeze() failed), should the data not have been lost? (on userspace read.) 2) Should writes performed by the direct IO read be able to cancel the free operation? - Should the direct IO read be considered as 'the caller' too, as it's been requested by 'the caller'? - Should the bio technique to dirty pages on return to userspace (bio_check_pages_dirty() is called/used by __blkdev_direct_IO()) be considered in another/special way here? 3) Should an upcoming write from a previously requested direct IO read be considered as a subsequent write, so the kernel should not free the pages? (as it's known at the time of page reclaim.) And lastly: Technically, the last point would seem a reasonable consideration and balance, as the madvise(2) manual page apparently (and fairly) seem to assume that 'writes' are memory access from the userspace process (not explicitly considering writes from the kernel or its corner cases; again, fairly).. plus the kernel fix implementation for the corner case of the largely 'non-atomic write' encompassed by a direct IO read operation, is relatively simple; and it helps. Reproducer: ========== @ test.c (simplified, but works) #define _GNU_SOURCE #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> int main() { int fd, i; char *buf; fd = open(DEV, O_RDONLY | O_DIRECT); buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) buf[i] = 1; // init to non-zero madvise(buf, BUF_SIZE, MADV_FREE); read(fd, buf, BUF_SIZE); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) printf("%p: 0x%x\n", &buf[i], buf[i]); return 0; } @ block/fops.c (formerly fs/block_dev.c) +#include <linux/swap.h> ... ... __blkdev_direct_IO[_simple](...) { ... + if (!strcmp(current->comm, "good")) + shrink_all_memory(ULONG_MAX); + ret = bio_iov_iter_get_pages(...); + + if (!strcmp(current->comm, "bad")) + shrink_all_memory(ULONG_MAX); ... } @ shell # NUM_PAGES=4 # PAGE_SIZE=$(getconf PAGE_SIZE) # yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES} # DEV=$(losetup -f --show test.img) # gcc -DDEV=\"$DEV\" \ -DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \ -DPAGE_SIZE=${PAGE_SIZE} \ test.c -o test # od -tx1 $DEV 0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a * 0040000 # mv test good # ./good 0x7f7c10418000: 0x79 0x7f7c10419000: 0x79 0x7f7c1041a000: 0x79 0x7f7c1041b000: 0x79 # mv good bad # ./bad 0x7fa1b8050000: 0x0 0x7fa1b8051000: 0x0 0x7fa1b8052000: 0x0 0x7fa1b8053000: 0x0 Note: the issue is consistent on v5.17-rc3, but it's intermittent with the support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c]. - v5.17-rc3: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x0 # free | grep Swap Swap: 0 0 0 - v4.5: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 2702 0x0 1298 0x79 # swapoff -av swapoff /swap # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 Ceph/TCMalloc: ============= For documentation purposes, the use case driving the analysis/fix is Ceph on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to release unused memory to the system from the mmap'ed page heap (might be committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing. Note: TCMalloc switched back to MADV_DONTNEED a few commits after the release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue just 'disappeared' on Ceph on later Ubuntu releases but is still present in the kernel, and can be hit by other use cases. The observed issue seems to be the old Ceph bug #22464 [1], where checksum mismatches are observed (and instrumentation with buffer dumps shows zero-pages read from mmap'ed/MADV_FREE'd page ranges). The issue in Ceph was reasonably deemed a kernel bug (comment #50) and mostly worked around with a retry mechanism, but other parts of Ceph could still hit that (rocksdb). Anyway, it's less likely to be hit again as TCMalloc switched out of MADV_FREE by default. (Some kernel versions/reports from the Ceph bug, and relation with the MADV_FREE introduction/changes; TCMalloc versions not checked.) - 4.4 good - 4.5 (madv_free: introduction) - 4.9 bad - 4.10 good? maybe a swapless system - 4.12 (madv_free: no longer free instantly on swapless systems) - 4.13 bad [1] https://tracker.ceph.com/issues/22464 Thanks: ====== Several people contributed to analysis/discussions/tests/reproducers in the first stages when drilling down on ceph/tcmalloc/linux kernel: - Dan Hill - Dan Streetman - Dongdong Tao - Gavin Guo - Gerald Yang - Heitor Alves de Siqueira - Ioanna Alifieraki - Jay Vosburgh - Matthew Ruffell - Ponnuvel Palaniyappan Reviews, suggestions, corrections, comments: - Minchan Kim - Yu Zhao - Huang, Ying - John Hubbard - Christoph Hellwig [mfo@canonical.com: v4] Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com Fixes: 802a3a9 ("mm: reclaim MADV_FREE pages") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Dan Hill <daniel.hill@canonical.com> Cc: Dan Streetman <dan.streetman@canonical.com> Cc: Dongdong Tao <dongdong.tao@canonical.com> Cc: Gavin Guo <gavin.guo@canonical.com> Cc: Gerald Yang <gerald.yang@canonical.com> Cc: Heitor Alves de Siqueira <halves@canonical.com> Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Cc: Matthew Ruffell <matthew.ruffell@canonical.com> Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com> Cc: <stable@vger.kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [mfo: backport: replace folio/test_flag with page/flag equivalents; real Fixes: 854e9ed ("mm: support madvise(MADV_FREE)") in v4.] Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
May 25, 2022
commit 6c8e2a2 upstream. Problem: ======= Userspace might read the zero-page instead of actual data from a direct IO read on a block device if the buffers have been called madvise(MADV_FREE) on earlier (this is discussed below) due to a race between page reclaim on MADV_FREE and blkdev direct IO read. - Race condition: ============== During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks if the page is not dirty, then discards its rmap PTE(s) (vs. remap back if the page is dirty). However, after try_to_unmap_one() returns to shrink_page_list(), it might keep the page _anyway_ if page_ref_freeze() fails (it expects exactly _one_ page reference, from the isolation for page reclaim). Well, blkdev_direct_IO() gets references for all pages, and on READ operations it only sets them dirty _later_. So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct IO read from block devices, and page reclaim happens during __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages() returns, but BEFORE the pages are set dirty, the situation happens. The direct IO read eventually completes. Now, when userspace reads the buffers, the PTE is no longer there and the page fault handler do_anonymous_page() services that with the zero-page, NOT the data! A synthetic reproducer is provided. - Page faults: =========== If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't happen, because that faults-in all pages as writeable, so do_anonymous_page() sets up a new page/rmap/PTE, and that is used by direct IO. The userspace reads don't fault as the PTE is there (thus zero-page is not used/setup). But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE is no longer there; the subsequent page faults can't help: The data-read from the block device probably won't generate faults due to DMA (no MMU) but even in the case it wouldn't use DMA, that happens on different virtual addresses (not user-mapped addresses) because `struct bio_vec` stores `struct page` to figure addresses out (which are different from user-mapped addresses) for the read. Thus userspace reads (to user-mapped addresses) still fault, then do_anonymous_page() gets another `struct page` that would address/ map to other memory than the `struct page` used by `struct bio_vec` for the read. (The original `struct page` is not available, since it wasn't freed, as page_ref_freeze() failed due to more page refs. And even if it were available, its data cannot be trusted anymore.) Solution: ======== One solution is to check for the expected page reference count in try_to_unmap_one(). There should be one reference from the isolation (that is also checked in shrink_page_list() with page_ref_freeze()) plus one or more references from page mapping(s) (put in discard: label). Further references mean that rmap/PTE cannot be unmapped/nuked. (Note: there might be more than one reference from mapping due to fork()/clone() without CLONE_VM, which use the same `struct page` for references, until the copy-on-write page gets copied.) So, additional page references (e.g., from direct IO read) now prevent the rmap/PTE from being unmapped/dropped; similarly to the page is not freed per shrink_page_list()/page_ref_freeze()). - Races and Barriers: ================== The new check in try_to_unmap_one() should be safe in races with bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's done under the PTE lock. The fast path doesn't take the lock, but it checks if the PTE has changed and if so, it drops the reference and leaves the page for the slow path (which does take that lock). The fast path requires synchronization w/ full memory barrier: it writes the page reference count first then it reads the PTE later, while try_to_unmap() writes PTE first then it reads page refcount. And a second barrier is needed, as the page dirty flag should not be read before the page reference count (as in __remove_mapping()). (This can be a load memory barrier only; no writes are involved.) Call stack/comments: - try_to_unmap_one() - page_vma_mapped_walk() - map_pte() # see pte_offset_map_lock(): pte_offset_map() spin_lock() - ptep_get_and_clear() # write PTE - smp_mb() # (new barrier) GUP fast path - page_ref_count() # (new check) read refcount - page_vma_mapped_walk_done() # see pte_unmap_unlock(): pte_unmap() spin_unlock() - bio_iov_iter_get_pages() - __bio_iov_iter_get_pages() - iov_iter_get_pages() - get_user_pages_fast() - internal_get_user_pages_fast() # fast path - lockless_pages_from_mm() - gup_{pgd,p4d,pud,pmd,pte}_range() ptep = pte_offset_map() # not _lock() pte = ptep_get_lockless(ptep) page = pte_page(pte) try_grab_compound_head(page) # inc refcount # (RMW/barrier # on success) if (pte_val(pte) != pte_val(*ptep)) # read PTE put_compound_head(page) # dec refcount # go slow path # slow path - __gup_longterm_unlocked() - get_user_pages_unlocked() - __get_user_pages_locked() - __get_user_pages() - follow_{page,p4d,pud,pmd}_mask() - follow_page_pte() ptep = pte_offset_map_lock() pte = *ptep page = vm_normal_page(pte) try_grab_page(page) # inc refcount pte_unmap_unlock() - Huge Pages: ========== Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE (aka lazyfree) pages are PageAnon() && !PageSwapBacked() (madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn()) thus should reach shrink_page_list() -> split_huge_page_to_list() before try_to_unmap[_one](), so it deals with normal pages only. (And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens, which should not or be rare, the page refcount should be greater than mapcount: the head page is referenced by tail pages. That also prevents checking the head `page` then incorrectly call page_remove_rmap(subpage) for a tail page, that isn't even in the shrink_page_list()'s page_list (an effect of split huge pmd/pmvw), as it might happen today in this unlikely scenario.) MADV_FREE'd buffers: =================== So, back to the "if MADV_FREE pages are used as buffers" note. The case is arguable, and subject to multiple interpretations. The madvise(2) manual page on the MADV_FREE advice value says: 1) 'After a successful MADV_FREE ... data will be lost when the kernel frees the pages.' 2) 'the free operation will be canceled if the caller writes into the page' / 'subsequent writes ... will succeed and then [the] kernel cannot free those dirtied pages' 3) 'If there is no subsequent write, the kernel can free the pages at any time.' Thoughts, questions, considerations... respectively: 1) Since the kernel didn't actually free the page (page_ref_freeze() failed), should the data not have been lost? (on userspace read.) 2) Should writes performed by the direct IO read be able to cancel the free operation? - Should the direct IO read be considered as 'the caller' too, as it's been requested by 'the caller'? - Should the bio technique to dirty pages on return to userspace (bio_check_pages_dirty() is called/used by __blkdev_direct_IO()) be considered in another/special way here? 3) Should an upcoming write from a previously requested direct IO read be considered as a subsequent write, so the kernel should not free the pages? (as it's known at the time of page reclaim.) And lastly: Technically, the last point would seem a reasonable consideration and balance, as the madvise(2) manual page apparently (and fairly) seem to assume that 'writes' are memory access from the userspace process (not explicitly considering writes from the kernel or its corner cases; again, fairly).. plus the kernel fix implementation for the corner case of the largely 'non-atomic write' encompassed by a direct IO read operation, is relatively simple; and it helps. Reproducer: ========== @ test.c (simplified, but works) #define _GNU_SOURCE #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> int main() { int fd, i; char *buf; fd = open(DEV, O_RDONLY | O_DIRECT); buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) buf[i] = 1; // init to non-zero madvise(buf, BUF_SIZE, MADV_FREE); read(fd, buf, BUF_SIZE); for (i = 0; i < BUF_SIZE; i += PAGE_SIZE) printf("%p: 0x%x\n", &buf[i], buf[i]); return 0; } @ block/fops.c (formerly fs/block_dev.c) +#include <linux/swap.h> ... ... __blkdev_direct_IO[_simple](...) { ... + if (!strcmp(current->comm, "good")) + shrink_all_memory(ULONG_MAX); + ret = bio_iov_iter_get_pages(...); + + if (!strcmp(current->comm, "bad")) + shrink_all_memory(ULONG_MAX); ... } @ shell # NUM_PAGES=4 # PAGE_SIZE=$(getconf PAGE_SIZE) # yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES} # DEV=$(losetup -f --show test.img) # gcc -DDEV=\"$DEV\" \ -DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \ -DPAGE_SIZE=${PAGE_SIZE} \ test.c -o test # od -tx1 $DEV 0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a * 0040000 # mv test good # ./good 0x7f7c10418000: 0x79 0x7f7c10419000: 0x79 0x7f7c1041a000: 0x79 0x7f7c1041b000: 0x79 # mv good bad # ./bad 0x7fa1b8050000: 0x0 0x7fa1b8051000: 0x0 0x7fa1b8052000: 0x0 0x7fa1b8053000: 0x0 Note: the issue is consistent on v5.17-rc3, but it's intermittent with the support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c]. - v5.17-rc3: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x0 # free | grep Swap Swap: 0 0 0 - v4.5: # for i in {1..1000}; do ./good; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 # mv good bad # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 2702 0x0 1298 0x79 # swapoff -av swapoff /swap # for i in {1..1000}; do ./bad; done \ | cut -d: -f2 | sort | uniq -c 4000 0x79 Ceph/TCMalloc: ============= For documentation purposes, the use case driving the analysis/fix is Ceph on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to release unused memory to the system from the mmap'ed page heap (might be committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan() -> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() -> TCMalloc_SystemCommit() -> do nothing. Note: TCMalloc switched back to MADV_DONTNEED a few commits after the release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue just 'disappeared' on Ceph on later Ubuntu releases but is still present in the kernel, and can be hit by other use cases. The observed issue seems to be the old Ceph bug #22464 [1], where checksum mismatches are observed (and instrumentation with buffer dumps shows zero-pages read from mmap'ed/MADV_FREE'd page ranges). The issue in Ceph was reasonably deemed a kernel bug (comment #50) and mostly worked around with a retry mechanism, but other parts of Ceph could still hit that (rocksdb). Anyway, it's less likely to be hit again as TCMalloc switched out of MADV_FREE by default. (Some kernel versions/reports from the Ceph bug, and relation with the MADV_FREE introduction/changes; TCMalloc versions not checked.) - 4.4 good - 4.5 (madv_free: introduction) - 4.9 bad - 4.10 good? maybe a swapless system - 4.12 (madv_free: no longer free instantly on swapless systems) - 4.13 bad [1] https://tracker.ceph.com/issues/22464 Thanks: ====== Several people contributed to analysis/discussions/tests/reproducers in the first stages when drilling down on ceph/tcmalloc/linux kernel: - Dan Hill - Dan Streetman - Dongdong Tao - Gavin Guo - Gerald Yang - Heitor Alves de Siqueira - Ioanna Alifieraki - Jay Vosburgh - Matthew Ruffell - Ponnuvel Palaniyappan Reviews, suggestions, corrections, comments: - Minchan Kim - Yu Zhao - Huang, Ying - John Hubbard - Christoph Hellwig [mfo@canonical.com: v4] Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com Fixes: 802a3a9 ("mm: reclaim MADV_FREE pages") Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Reviewed-by: "Huang, Ying" <ying.huang@intel.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Yu Zhao <yuzhao@google.com> Cc: Yang Shi <shy828301@gmail.com> Cc: Miaohe Lin <linmiaohe@huawei.com> Cc: Dan Hill <daniel.hill@canonical.com> Cc: Dan Streetman <dan.streetman@canonical.com> Cc: Dongdong Tao <dongdong.tao@canonical.com> Cc: Gavin Guo <gavin.guo@canonical.com> Cc: Gerald Yang <gerald.yang@canonical.com> Cc: Heitor Alves de Siqueira <halves@canonical.com> Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com> Cc: Jay Vosburgh <jay.vosburgh@canonical.com> Cc: Matthew Ruffell <matthew.ruffell@canonical.com> Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com> Cc: <stable@vger.kernel.org> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> [mfo: backport: replace folio/test_flag with page/flag equivalents; mmu_notifier_invalidate_range() is called in the 'discard:' label; real Fixes: 854e9ed ("mm: support madvise(MADV_FREE)") in v4.] Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
frank-w
pushed a commit
that referenced
this pull request
Jun 24, 2023
The commit 4f7e723 ("cgroup: Fix threadgroup_rwsem <-> cpus_read_lock() deadlock") fixed the deadlock between cgroup_threadgroup_rwsem and cpus_read_lock() by introducing cgroup_attach_{lock,unlock}() and removing cpus_read_{lock,unlock}() from cpuset_attach(). But cgroup_transfer_tasks() was missed and not handled, which will cause th following warning: WARNING: CPU: 0 PID: 589 at kernel/cpu.c:526 lockdep_assert_cpus_held+0x32/0x40 CPU: 0 PID: 589 Comm: kworker/1:4 Not tainted 6.4.0-rc2-next-20230517 #50 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.14.0-2 04/01/2014 Workqueue: events cpuset_hotplug_workfn RIP: 0010:lockdep_assert_cpus_held+0x32/0x40 <...> Call Trace: <TASK> cpuset_attach+0x40/0x240 cgroup_migrate_execute+0x452/0x5e0 ? _raw_spin_unlock_irq+0x28/0x40 cgroup_transfer_tasks+0x1f3/0x360 ? find_held_lock+0x32/0x90 ? cpuset_hotplug_workfn+0xc81/0xed0 cpuset_hotplug_workfn+0xcb1/0xed0 ? process_one_work+0x248/0x5b0 process_one_work+0x2b9/0x5b0 worker_thread+0x56/0x3b0 ? process_one_work+0x5b0/0x5b0 kthread+0xf1/0x120 ? kthread_complete_and_exit+0x20/0x20 ret_from_fork+0x1f/0x30 </TASK> So just use the cgroup_attach_{lock,unlock}() helper to fix it. Reported-by: Zhao Gongyi <zhaogongyi@bytedance.com> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com> Acked-by: Muchun Song <songmuchun@bytedance.com> Fixes: 05c7b7a ("cgroup/cpuset: Fix a race between cpuset_attach() and cpu hotplug") Cc: stable@vger.kernel.org # v5.17+ Signed-off-by: Tejun Heo <tj@kernel.org>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.