]> git.itanic.dy.fi Git - linux-stable/commitdiff
nvme: fix SRCU protection of nvme_ns_head list
authorCaleb Sander <csander@purestorage.com>
Fri, 18 Nov 2022 23:27:56 +0000 (16:27 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 8 Dec 2022 10:30:21 +0000 (11:30 +0100)
[ Upstream commit 899d2a05dc14733cfba6224083c6b0dd5a738590 ]

Walking the nvme_ns_head siblings list is protected by the head's srcu
in nvme_ns_head_submit_bio() but not nvme_mpath_revalidate_paths().
Removing namespaces from the list also fails to synchronize the srcu.
Concurrent scan work can therefore cause use-after-frees.

Hold the head's srcu lock in nvme_mpath_revalidate_paths() and
synchronize with the srcu, not the global RCU, in nvme_ns_remove().

Observed the following panic when making NVMe/RDMA connections
with native multipath on the Rocky Linux 8.6 kernel
(it seems the upstream kernel has the same race condition).
Disassembly shows the faulting instruction is cmp 0x50(%rdx),%rcx;
computing capacity != get_capacity(ns->disk).
Address 0x50 is dereferenced because ns->disk is NULL.
The NULL disk appears to be the result of concurrent scan work
freeing the namespace (note the log line in the middle of the panic).

[37314.206036] BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
[37314.206036] nvme0n3: detected capacity change from 0 to 11811160064
[37314.299753] PGD 0 P4D 0
[37314.299756] Oops: 0000 [#1] SMP PTI
[37314.299759] CPU: 29 PID: 322046 Comm: kworker/u98:3 Kdump: loaded Tainted: G        W      X --------- -  - 4.18.0-372.32.1.el8test86.x86_64 #1
[37314.299762] Hardware name: Dell Inc. PowerEdge R720/0JP31P, BIOS 2.7.0 05/23/2018
[37314.299763] Workqueue: nvme-wq nvme_scan_work [nvme_core]
[37314.299783] RIP: 0010:nvme_mpath_revalidate_paths+0x26/0xb0 [nvme_core]
[37314.299790] Code: 1f 44 00 00 66 66 66 66 90 55 53 48 8b 5f 50 48 8b 83 c8 c9 00 00 48 8b 13 48 8b 48 50 48 39 d3 74 20 48 8d 42 d0 48 8b 50 20 <48> 3b 4a 50 74 05 f0 80 60 70 ef 48 8b 50 30 48 8d 42 d0 48 39 d3
[37315.058803] RSP: 0018:ffffabe28f913d10 EFLAGS: 00010202
[37315.121316] RAX: ffff927a077da800 RBX: ffff92991dd70000 RCX: 0000000001600000
[37315.206704] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff92991b719800
[37315.292106] RBP: ffff929a6b70c000 R08: 000000010234cd4a R09: c0000000ffff7fff
[37315.377501] R10: 0000000000000001 R11: ffffabe28f913a30 R12: 0000000000000000
[37315.462889] R13: ffff92992716600c R14: ffff929964e6e030 R15: ffff92991dd70000
[37315.548286] FS:  0000000000000000(0000) GS:ffff92b87fb80000(0000) knlGS:0000000000000000
[37315.645111] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[37315.713871] CR2: 0000000000000050 CR3: 0000002208810006 CR4: 00000000000606e0
[37315.799267] Call Trace:
[37315.828515]  nvme_update_ns_info+0x1ac/0x250 [nvme_core]
[37315.892075]  nvme_validate_or_alloc_ns+0x2ff/0xa00 [nvme_core]
[37315.961871]  ? __blk_mq_free_request+0x6b/0x90
[37316.015021]  nvme_scan_work+0x151/0x240 [nvme_core]
[37316.073371]  process_one_work+0x1a7/0x360
[37316.121318]  ? create_worker+0x1a0/0x1a0
[37316.168227]  worker_thread+0x30/0x390
[37316.212024]  ? create_worker+0x1a0/0x1a0
[37316.258939]  kthread+0x10a/0x120
[37316.297557]  ? set_kthread_struct+0x50/0x50
[37316.347590]  ret_from_fork+0x35/0x40
[37316.390360] Modules linked in: nvme_rdma nvme_tcp(X) nvme_fabrics nvme_core netconsole iscsi_tcp libiscsi_tcp dm_queue_length dm_service_time nf_conntrack_netlink br_netfilter bridge stp llc overlay nft_chain_nat ipt_MASQUERADE nf_nat xt_addrtype xt_CT nft_counter xt_state xt_conntrack nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 xt_comment xt_multiport nft_compat nf_tables libcrc32c nfnetlink dm_multipath tg3 rpcrdma sunrpc rdma_ucm ib_srpt ib_isert iscsi_target_mod target_core_mod ib_iser libiscsi scsi_transport_iscsi ib_umad rdma_cm ib_ipoib iw_cm ib_cm intel_rapl_msr iTCO_wdt iTCO_vendor_support dcdbas intel_rapl_common sb_edac x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel ipmi_ssif kvm irqbypass crct10dif_pclmul crc32_pclmul mlx5_ib ghash_clmulni_intel ib_uverbs rapl intel_cstate intel_uncore ib_core ipmi_si joydev mei_me pcspkr ipmi_devintf mei lpc_ich wmi ipmi_msghandler acpi_power_meter ext4 mbcache jbd2 sd_mod t10_pi sg mgag200 mlx5_core drm_kms_helper syscopyarea
[37316.390419]  sysfillrect ahci sysimgblt fb_sys_fops libahci drm crc32c_intel libata mlxfw pci_hyperv_intf tls i2c_algo_bit psample dm_mirror dm_region_hash dm_log dm_mod fuse [last unloaded: nvme_core]
[37317.645908] CR2: 0000000000000050

Fixes: e7d65803e2bb ("nvme-multipath: revalidate paths during rescan")
Signed-off-by: Caleb Sander <csander@purestorage.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/nvme/host/core.c
drivers/nvme/host/multipath.c

index 01c36284e5428aaa614448655758c06e09a27e80..f612a0ba64d00d1594b8acf6b7715d8688ff6b89 100644 (file)
@@ -4297,7 +4297,7 @@ static void nvme_ns_remove(struct nvme_ns *ns)
        mutex_unlock(&ns->ctrl->subsys->lock);
 
        /* guarantee not available in head->list */
-       synchronize_rcu();
+       synchronize_srcu(&ns->head->srcu);
 
        if (!nvme_ns_head_multipath(ns->head))
                nvme_cdev_del(&ns->cdev, &ns->cdev_device);
index b9cf17cbbbd5d6bf1681355df47c9cc71f676050..114e2b9359f83f5564be723b9cc84fd12b76b4a5 100644 (file)
@@ -174,11 +174,14 @@ void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
        struct nvme_ns_head *head = ns->head;
        sector_t capacity = get_capacity(head->disk);
        int node;
+       int srcu_idx;
 
+       srcu_idx = srcu_read_lock(&head->srcu);
        list_for_each_entry_rcu(ns, &head->list, siblings) {
                if (capacity != get_capacity(ns->disk))
                        clear_bit(NVME_NS_READY, &ns->flags);
        }
+       srcu_read_unlock(&head->srcu, srcu_idx);
 
        for_each_node(node)
                rcu_assign_pointer(head->current_path[node], NULL);