Message ID | 20210605202051.14783-1-paskripkin@gmail.com (mailing list archive) |
---|---|
State | Changes Requested |
Headers | show |
Series | infiniband: core: fix memory leak | expand |
On Sat, Jun 05, 2021 at 11:20:51PM +0300, Pavel Skripkin wrote: > My local syzbot instance hit memory leak in > copy_process(). The problem was in unputted task > struct in _destroy_id(). > > Simple reproducer: > > int main(void) > { > struct rdma_ucm_cmd_hdr *hdr; > struct rdma_ucm_create_id *cmd_id; > char cmd[sizeof(*hdr) + sizeof(*cmd_id)] = {0}; > int fd; > > hdr = (struct rdma_ucm_cmd_hdr *)cmd; > cmd_id = (struct rdma_ucm_create_id *) (cmd + sizeof(*hdr)); > > hdr->cmd = 0; > hdr->in = 0x18; > hdr->out = 0xfa00; > > cmd_id->uid = 0x3; > cmd_id->response = 0x0; > cmd_id->ps = 0x106; > > fd = open("/dev/infiniband/rdma_cm", O_WRONLY); > write(fd, cmd, sizeof(cmd)); > } > > Ftrace log: > > ucma_open(); > ucma_write() { > ucma_create_id() { > ucma_alloc_ctx(); > rdma_create_user_id() { > rdma_restrack_new(); > rdma_restrack_set_name() { > rdma_restrack_attach_task.part.0(); <--- task_struct getted > } > } > ucma_destroy_private_ctx() { > ucma_put_ctx(); > rdma_destroy_id() { > _destroy_id() <--- id_priv freed > } > } > } > } > ucma_close(); > > From previous log it's easy to undertand that > _destroy_id() is the last place, where task_struct > can be putted, because at the end of this function > id_priv is freed. > > With this patch applied, above reproducer doesn't hit memory > leak anymore. > > Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager") > Cc: <stable@vger.kernel.org> > Signed-off-by: Pavel Skripkin <paskripkin@gmail.com> > --- > drivers/infiniband/core/cma.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > index ab148a696c0c..2760352261b3 100644 > --- a/drivers/infiniband/core/cma.c > +++ b/drivers/infiniband/core/cma.c > @@ -1874,6 +1874,7 @@ static void _destroy_id(struct rdma_id_private *id_priv, > > kfree(id_priv->id.route.path_rec); > > + rdma_restrack_put(&id_priv->res); This change will cause to use-after-free [ 2478.766905] ------------[ cut here ]------------ [ 2478.768721] refcount_t: underflow; use-after-free. [ 2478.770577] WARNING: CPU: 1 PID: 56454 at lib/refcount.c:28 refcount_warn_saturate+0x17a/0x190 [ 2478.773787] Modules linked in: bonding ib_umad ip_gre geneve nf_tables ipip tunnel4 ip6_gre gre ib_ipoib rdma_ucm mlx5_ib ib_uverbs mlx5_core ptp pps_core ip6_tunnel tunnel6 openvswitch nsh xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink rpcrdma xt_addrtype ib_iser libiscsi scsi_transport_iscsi iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 rdma_cm iw_cm br_netfilter ib_cm ib_core overlay fuse [last unloaded: nf_tables] [ 2478.783501] CPU: 1 PID: 56454 Comm: python3 Tainted: G W 5.13.0-rc4_2021_06_05_20_55_37_ #1 [ 2478.785418] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014 [ 2478.787566] RIP: 0010:refcount_warn_saturate+0x17a/0x190 [ 2478.788612] Code: 08 e8 00 0f 0b e9 10 ff ff ff 48 89 df e8 ee c2 5d ff e9 d4 fe ff ff 48 c7 c7 40 6d a4 83 c6 05 3b f7 6f 02 01 e8 65 08 e8 00 <0f> 0b e9 e9 fe ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 [ 2478.792014] RSP: 0018:ffff8881277f7b90 EFLAGS: 00010282 [ 2478.793081] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000 [ 2478.794418] RDX: 0000000000000004 RSI: 0000000000000008 RDI: ffffed1024efef64 [ 2478.795761] RBP: 0000000000000003 R08: 0000000000000001 R09: ffff8882a40bfd47 [ 2478.797141] R10: ffffed1054817fa8 R11: ffff88812ab60898 R12: 0000000000000000 [ 2478.798502] R13: dffffc0000000000 R14: dead000000000100 R15: ffff888102099000 [ 2478.799863] FS: 00007f5a6492a740(0000) GS:ffff8882a4080000(0000) knlGS:0000000000000000 [ 2478.801538] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 2478.802671] CR2: 00007f5a56e60900 CR3: 00000001aa7d8002 CR4: 0000000000370ea0 [ 2478.804030] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 2478.805428] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 2478.806802] Call Trace: [ 2478.807380] rdma_restrack_put+0x5b/0x70 [ib_core] [ 2478.808397] _destroy_id+0x41c/0x760 [rdma_cm] [ 2478.809342] ? ucma_close+0x180/0x180 [rdma_ucm] [ 2478.810295] ? ucma_close+0x180/0x180 [rdma_ucm] [ 2478.811252] ucma_destroy_private_ctx+0x7e4/0xa80 [rdma_ucm] [ 2478.812371] ? ucma_destroy_id+0x18e/0x220 [rdma_ucm] [ 2478.813422] ? ucma_query_route+0xd70/0xd70 [rdma_ucm] [ 2478.814464] ? ucma_close+0x180/0x180 [rdma_ucm] [ 2478.815400] ucma_destroy_id+0x196/0x220 [rdma_ucm] [ 2478.816366] ? ucma_close+0x180/0x180 [rdma_ucm] [ 2478.817313] ucma_write+0x1d2/0x2f0 [rdma_ucm] [ 2478.818207] ? ucma_close_id+0x90/0x90 [rdma_ucm] [ 2478.819159] ? __fget_files+0x1e0/0x2f0 [ 2478.819958] vfs_write+0x1c9/0x850 [ 2478.820681] ksys_write+0x176/0x1d0 [ 2478.821436] ? __x64_sys_read+0xb0/0xb0 [ 2478.822234] ? lockdep_hardirqs_on_prepare+0x286/0x400 [ 2478.823256] ? syscall_enter_from_user_mode+0x1d/0x50 [ 2478.824247] do_syscall_64+0x3f/0x80 [ 2478.825019] entry_SYSCALL_64_after_hwframe+0x44/0xae [ 2478.826015] RIP: 0033:0x7f5a64ecf32f [ 2478.826773] Code: 89 54 24 18 48 89 74 24 10 89 7c 24 08 e8 19 4b f9 ff 48 8b 54 24 18 48 8b 74 24 10 41 89 c0 8b 7c 24 08 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 31 44 89 c7 48 89 44 24 08 e8 4c 4b f9 ff 48 [ 2478.830248] RSP: 002b:00007ffe9ea8d730 EFLAGS: 00000293 ORIG_RAX: 0000000000000001 [ 2478.831779] RAX: ffffffffffffffda RBX: 00007ffe9ea8d764 RCX: 00007f5a64ecf32f [ 2478.833166] RDX: 0000000000000018 RSI: 00007ffe9ea8d770 RDI: 000000000000000a [ 2478.834529] RBP: 0000563b16f9f620 R08: 0000000000000000 R09: 00007f5a64db6c60 [ 2478.835897] R10: 0000000000000001 R11: 0000000000000293 R12: 00007f5a574dea60 [ 2478.837278] R13: 00007f5a64db49e0 R14: 00007f5a64dba840 R15: 00007f5a565c33a0 [ 2478.838649] irq event stamp: 23055 [ 2478.839381] hardirqs last enabled at (23063): [<ffffffff8146354b>] console_unlock+0x5fb/0x8e0 [ 2478.841098] hardirqs last disabled at (23072): [<ffffffff81463541>] console_unlock+0x5f1/0x8e0 [ 2478.842805] softirqs last enabled at (22670): [<ffffffff81346ba6>] __irq_exit_rcu+0x126/0x1b0 [ 2478.844498] softirqs last disabled at (22665): [<ffffffff81346ba6>] __irq_exit_rcu+0x126/0x1b0 [ 2478.846230] ---[ end trace f33482f6d20ccf0e ]--- I think that better fix is: https://lore.kernel.org/linux-rdma/f72e27d5c82cd9beec7670141afa62786836c569.1622956637.git.leonro@nvidia.com/T/#u > put_net(id_priv->id.route.addr.dev_addr.net); > kfree(id_priv); > } > -- > 2.31.1 >
On Sun, 6 Jun 2021 08:21:51 +0300 Leon Romanovsky <leon@kernel.org> wrote: > On Sat, Jun 05, 2021 at 11:20:51PM +0300, Pavel Skripkin wrote: > > My local syzbot instance hit memory leak in > > copy_process(). The problem was in unputted task > > struct in _destroy_id(). > > > > Simple reproducer: > > > > int main(void) > > { > > struct rdma_ucm_cmd_hdr *hdr; > > struct rdma_ucm_create_id *cmd_id; > > char cmd[sizeof(*hdr) + sizeof(*cmd_id)] = {0}; > > int fd; > > > > hdr = (struct rdma_ucm_cmd_hdr *)cmd; > > cmd_id = (struct rdma_ucm_create_id *) (cmd + sizeof(*hdr)); > > > > hdr->cmd = 0; > > hdr->in = 0x18; > > hdr->out = 0xfa00; > > > > cmd_id->uid = 0x3; > > cmd_id->response = 0x0; > > cmd_id->ps = 0x106; > > > > fd = open("/dev/infiniband/rdma_cm", O_WRONLY); > > write(fd, cmd, sizeof(cmd)); > > } > > > > Ftrace log: > > > > ucma_open(); > > ucma_write() { > > ucma_create_id() { > > ucma_alloc_ctx(); > > rdma_create_user_id() { > > rdma_restrack_new(); > > rdma_restrack_set_name() { > > rdma_restrack_attach_task.part.0(); <--- task_struct getted > > } > > } > > ucma_destroy_private_ctx() { > > ucma_put_ctx(); > > rdma_destroy_id() { > > _destroy_id() <--- id_priv freed > > } > > } > > } > > } > > ucma_close(); > > > > From previous log it's easy to undertand that > > _destroy_id() is the last place, where task_struct > > can be putted, because at the end of this function > > id_priv is freed. > > > > With this patch applied, above reproducer doesn't hit memory > > leak anymore. > > > > Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager") > > Cc: <stable@vger.kernel.org> > > Signed-off-by: Pavel Skripkin <paskripkin@gmail.com> > > --- > > drivers/infiniband/core/cma.c | 1 + > > 1 file changed, 1 insertion(+) > > > > diff --git a/drivers/infiniband/core/cma.c > > b/drivers/infiniband/core/cma.c index ab148a696c0c..2760352261b3 > > 100644 --- a/drivers/infiniband/core/cma.c > > +++ b/drivers/infiniband/core/cma.c > > @@ -1874,6 +1874,7 @@ static void _destroy_id(struct > > rdma_id_private *id_priv, > > kfree(id_priv->id.route.path_rec); > > > > + rdma_restrack_put(&id_priv->res); > > This change will cause to use-after-free > > [ 2478.766905] ------------[ cut here ]------------ > [ 2478.768721] refcount_t: underflow; use-after-free. > [ 2478.770577] WARNING: CPU: 1 PID: 56454 at lib/refcount.c:28 > refcount_warn_saturate+0x17a/0x190 [ 2478.773787] Modules linked in: > bonding ib_umad ip_gre geneve nf_tables ipip tunnel4 ip6_gre gre > ib_ipoib rdma_ucm mlx5_ib ib_uverbs mlx5_core ptp pps_core ip6_tunnel > tunnel6 openvswitch nsh xt_conntrack xt_MASQUERADE > nf_conntrack_netlink nfnetlink rpcrdma xt_addrtype ib_iser libiscsi > scsi_transport_iscsi iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 > nf_defrag_ipv4 rdma_cm iw_cm br_netfilter ib_cm ib_core overlay fuse > [last unloaded: nf_tables] [ 2478.783501] CPU: 1 PID: 56454 Comm: > python3 Tainted: G W 5.13.0-rc4_2021_06_05_20_55_37_ > #1 [ 2478.785418] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), > BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014 [ > 2478.787566] RIP: 0010:refcount_warn_saturate+0x17a/0x190 [ > 2478.788612] Code: 08 e8 00 0f 0b e9 10 ff ff ff 48 89 df e8 ee c2 5d > ff e9 d4 fe ff ff 48 c7 c7 40 6d a4 83 c6 05 3b f7 6f 02 01 e8 65 08 > e8 00 <0f> 0b e9 e9 fe ff ff 66 66 2e 0f 1f 84 00 00 00 00 00 0f 1f > 40 00 [ 2478.792014] RSP: 0018:ffff8881277f7b90 EFLAGS: 00010282 [ > 2478.793081] RAX: 0000000000000000 RBX: 0000000000000000 RCX: > 0000000000000000 [ 2478.794418] RDX: 0000000000000004 RSI: > 0000000000000008 RDI: ffffed1024efef64 [ 2478.795761] RBP: > 0000000000000003 R08: 0000000000000001 R09: ffff8882a40bfd47 [ > 2478.797141] R10: ffffed1054817fa8 R11: ffff88812ab60898 R12: > 0000000000000000 [ 2478.798502] R13: dffffc0000000000 R14: > dead000000000100 R15: ffff888102099000 [ 2478.799863] FS: > 00007f5a6492a740(0000) GS:ffff8882a4080000(0000) > knlGS:0000000000000000 [ 2478.801538] CS: 0010 DS: 0000 ES: 0000 > CR0: 0000000080050033 [ 2478.802671] CR2: 00007f5a56e60900 CR3: > 00000001aa7d8002 CR4: 0000000000370ea0 [ 2478.804030] DR0: > 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ > 2478.805428] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: > 0000000000000400 [ 2478.806802] Call Trace: [ 2478.807380] > rdma_restrack_put+0x5b/0x70 [ib_core] [ 2478.808397] > _destroy_id+0x41c/0x760 [rdma_cm] [ 2478.809342] ? > ucma_close+0x180/0x180 [rdma_ucm] [ 2478.810295] ? > ucma_close+0x180/0x180 [rdma_ucm] [ 2478.811252] > ucma_destroy_private_ctx+0x7e4/0xa80 [rdma_ucm] [ 2478.812371] ? > ucma_destroy_id+0x18e/0x220 [rdma_ucm] [ 2478.813422] ? > ucma_query_route+0xd70/0xd70 [rdma_ucm] [ 2478.814464] ? > ucma_close+0x180/0x180 [rdma_ucm] [ 2478.815400] > ucma_destroy_id+0x196/0x220 [rdma_ucm] [ 2478.816366] ? > ucma_close+0x180/0x180 [rdma_ucm] [ 2478.817313] > ucma_write+0x1d2/0x2f0 [rdma_ucm] [ 2478.818207] ? > ucma_close_id+0x90/0x90 [rdma_ucm] [ 2478.819159] ? > __fget_files+0x1e0/0x2f0 [ 2478.819958] vfs_write+0x1c9/0x850 [ > 2478.820681] ksys_write+0x176/0x1d0 [ 2478.821436] ? > __x64_sys_read+0xb0/0xb0 [ 2478.822234] ? > lockdep_hardirqs_on_prepare+0x286/0x400 [ 2478.823256] ? > syscall_enter_from_user_mode+0x1d/0x50 [ 2478.824247] > do_syscall_64+0x3f/0x80 [ 2478.825019] > entry_SYSCALL_64_after_hwframe+0x44/0xae [ 2478.826015] RIP: > 0033:0x7f5a64ecf32f [ 2478.826773] Code: 89 54 24 18 48 89 74 24 10 > 89 7c 24 08 e8 19 4b f9 ff 48 8b 54 24 18 48 8b 74 24 10 41 89 c0 8b > 7c 24 08 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 31 44 89 c7 48 > 89 44 24 08 e8 4c 4b f9 ff 48 [ 2478.830248] RSP: > 002b:00007ffe9ea8d730 EFLAGS: 00000293 ORIG_RAX: 0000000000000001 [ > 2478.831779] RAX: ffffffffffffffda RBX: 00007ffe9ea8d764 RCX: > 00007f5a64ecf32f [ 2478.833166] RDX: 0000000000000018 RSI: > 00007ffe9ea8d770 RDI: 000000000000000a [ 2478.834529] RBP: > 0000563b16f9f620 R08: 0000000000000000 R09: 00007f5a64db6c60 [ > 2478.835897] R10: 0000000000000001 R11: 0000000000000293 R12: > 00007f5a574dea60 [ 2478.837278] R13: 00007f5a64db49e0 R14: > 00007f5a64dba840 R15: 00007f5a565c33a0 [ 2478.838649] irq event > stamp: 23055 [ 2478.839381] hardirqs last enabled at (23063): > [<ffffffff8146354b>] console_unlock+0x5fb/0x8e0 [ 2478.841098] > hardirqs last disabled at (23072): [<ffffffff81463541>] > console_unlock+0x5f1/0x8e0 [ 2478.842805] softirqs last enabled at > (22670): [<ffffffff81346ba6>] __irq_exit_rcu+0x126/0x1b0 [ > 2478.844498] softirqs last disabled at (22665): [<ffffffff81346ba6>] > __irq_exit_rcu+0x126/0x1b0 [ 2478.846230] ---[ end trace > f33482f6d20ccf0e ]--- > > I think that better fix is: > https://lore.kernel.org/linux-rdma/f72e27d5c82cd9beec7670141afa62786836c569.1622956637.git.leonro@nvidia.com/T/#u > Thanks for testing. I ran some random tests, but didn't hit this use-after-free. > > > put_net(id_priv->id.route.addr.dev_addr.net); > > kfree(id_priv); > > } > > -- > > 2.31.1 > > With regards, Pavel Skripkin
On Sun, Jun 06, 2021 at 08:21:51AM +0300, Leon Romanovsky wrote: > I think that better fix is: > https://lore.kernel.org/linux-rdma/f72e27d5c82cd9beec7670141afa62786836c569.1622956637.git.leonro@nvidia.com/T/#u Can you resend that with the bug report in the comment? What is the issue? Some error path missed a restrack_del? which one? And I still dislike that patch for adding restrack_del and undoing the previous patch which was trying to get rid of it, but if it actually fixes a bug... Jason
On Mon, Jun 07, 2021 at 08:44:49PM -0300, Jason Gunthorpe wrote: > On Sun, Jun 06, 2021 at 08:21:51AM +0300, Leon Romanovsky wrote: > > > I think that better fix is: > > https://lore.kernel.org/linux-rdma/f72e27d5c82cd9beec7670141afa62786836c569.1622956637.git.leonro@nvidia.com/T/#u > > Can you resend that with the bug report in the comment? Done, https://lore.kernel.org/linux-rdma/e2eed941f912b2068e371fd37f43b8cf5082a0e6.1623129597.git.leonro@nvidia.com/T/#u > > What is the issue? Some error path missed a restrack_del? which one? All paths that called to rdma_restrack_new(), but didn't set "id_priv->cma_dev" yet. For example, cma_ib_new_conn_id() -> __rdma_create_id() -> do something -> exit on error -> miss call to rdma_restrack_del() > > And I still dislike that patch for adding restrack_del and undoing the > previous patch which was trying to get rid of it, but if it actually > fixes a bug... And I didn't like the previous patch :). Thanks > > Jason
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index ab148a696c0c..2760352261b3 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -1874,6 +1874,7 @@ static void _destroy_id(struct rdma_id_private *id_priv, kfree(id_priv->id.route.path_rec); + rdma_restrack_put(&id_priv->res); put_net(id_priv->id.route.addr.dev_addr.net); kfree(id_priv); }
My local syzbot instance hit memory leak in copy_process(). The problem was in unputted task struct in _destroy_id(). Simple reproducer: int main(void) { struct rdma_ucm_cmd_hdr *hdr; struct rdma_ucm_create_id *cmd_id; char cmd[sizeof(*hdr) + sizeof(*cmd_id)] = {0}; int fd; hdr = (struct rdma_ucm_cmd_hdr *)cmd; cmd_id = (struct rdma_ucm_create_id *) (cmd + sizeof(*hdr)); hdr->cmd = 0; hdr->in = 0x18; hdr->out = 0xfa00; cmd_id->uid = 0x3; cmd_id->response = 0x0; cmd_id->ps = 0x106; fd = open("/dev/infiniband/rdma_cm", O_WRONLY); write(fd, cmd, sizeof(cmd)); } Ftrace log: ucma_open(); ucma_write() { ucma_create_id() { ucma_alloc_ctx(); rdma_create_user_id() { rdma_restrack_new(); rdma_restrack_set_name() { rdma_restrack_attach_task.part.0(); <--- task_struct getted } } ucma_destroy_private_ctx() { ucma_put_ctx(); rdma_destroy_id() { _destroy_id() <--- id_priv freed } } } } ucma_close(); From previous log it's easy to undertand that _destroy_id() is the last place, where task_struct can be putted, because at the end of this function id_priv is freed. With this patch applied, above reproducer doesn't hit memory leak anymore. Fixes: e51060f08a61 ("IB: IP address based RDMA connection manager") Cc: <stable@vger.kernel.org> Signed-off-by: Pavel Skripkin <paskripkin@gmail.com> --- drivers/infiniband/core/cma.c | 1 + 1 file changed, 1 insertion(+)