On Wed, Jun 01, 2022 at 03:53:41PM -0400, Jason Andryuk wrote: > When a VBD is not fully created and then closed, the kernel can have a > NULL pointer dereference: > > The reproducer is trivial: > > [user@dom0 ~]$ sudo xl block-attach work backend=sys-usb vdev=xvdi target=/dev/sdz > [user@dom0 ~]$ xl block-list work > Vdev BE handle state evt-ch ring-ref BE-path > 51712 0 241 4 -1 -1 /local/domain/0/backend/vbd/241/51712 > 51728 0 241 4 -1 -1 /local/domain/0/backend/vbd/241/51728 > 51744 0 241 4 -1 -1 /local/domain/0/backend/vbd/241/51744 > 51760 0 241 4 -1 -1 /local/domain/0/backend/vbd/241/51760 > 51840 3 241 3 -1 -1 /local/domain/3/backend/vbd/241/51840 > ^ note state, the /dev/sdz doesn't exist in the backend > > [user@dom0 ~]$ sudo xl block-detach work xvdi > [user@dom0 ~]$ xl block-list work > Vdev BE handle state evt-ch ring-ref BE-path > work is an invalid domain identifier > > And its console has: > > BUG: kernel NULL pointer dereference, address: 0000000000000050 > PGD 80000000edebb067 P4D 80000000edebb067 PUD edec2067 PMD 0 > Oops: 0000 [#1] PREEMPT SMP PTI > CPU: 1 PID: 52 Comm: xenwatch Not tainted 5.16.18-2.43.fc32.qubes.x86_64 #1 > RIP: 0010:blk_mq_stop_hw_queues+0x5/0x40 > Code: 00 48 83 e0 fd 83 c3 01 48 89 85 a8 00 00 00 41 39 5c 24 50 77 c0 5b 5d 41 5c 41 5d c3 c3 0f 1f 80 00 00 00 00 0f 1f 44 00 00 <8b> 47 50 85 c0 74 32 41 54 49 89 fc 55 53 31 db 49 8b 44 24 48 48 > RSP: 0018:ffffc90000bcfe98 EFLAGS: 00010293 > RAX: ffffffffc0008370 RBX: 0000000000000005 RCX: 0000000000000000 > RDX: 0000000000000000 RSI: 0000000000000005 RDI: 0000000000000000 > RBP: ffff88800775f000 R08: 0000000000000001 R09: ffff888006e620b8 > R10: ffff888006e620b0 R11: f000000000000000 R12: ffff8880bff39000 > R13: ffff8880bff39000 R14: 0000000000000000 R15: ffff88800604be00 > FS: 0000000000000000(0000) GS:ffff8880f3300000(0000) knlGS:0000000000000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 0000000000000050 CR3: 00000000e932e002 CR4: 00000000003706e0 > Call Trace: > <TASK> > blkback_changed+0x95/0x137 [xen_blkfront] > ? read_reply+0x160/0x160 > xenwatch_thread+0xc0/0x1a0 > ? do_wait_intr_irq+0xa0/0xa0 > kthread+0x16b/0x190 > ? set_kthread_struct+0x40/0x40 > ret_from_fork+0x22/0x30 > </TASK> > Modules linked in: snd_seq_dummy snd_hrtimer snd_seq snd_seq_device snd_timer snd soundcore ipt_REJECT nf_reject_ipv4 xt_state xt_conntrack nft_counter nft_chain_nat xt_MASQUERADE nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 nft_compat nf_tables nfnetlink intel_rapl_msr intel_rapl_common crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel xen_netfront pcspkr xen_scsiback target_core_mod xen_netback xen_privcmd xen_gntdev xen_gntalloc xen_blkback xen_evtchn ipmi_devintf ipmi_msghandler fuse bpf_preload ip_tables overlay xen_blkfront > CR2: 0000000000000050 > ---[ end trace 7bc9597fd06ae89d ]--- > RIP: 0010:blk_mq_stop_hw_queues+0x5/0x40 > Code: 00 48 83 e0 fd 83 c3 01 48 89 85 a8 00 00 00 41 39 5c 24 50 77 c0 5b 5d 41 5c 41 5d c3 c3 0f 1f 80 00 00 00 00 0f 1f 44 00 00 <8b> 47 50 85 c0 74 32 41 54 49 89 fc 55 53 31 db 49 8b 44 24 48 48 > RSP: 0018:ffffc90000bcfe98 EFLAGS: 00010293 > RAX: ffffffffc0008370 RBX: 0000000000000005 RCX: 0000000000000000 > RDX: 0000000000000000 RSI: 0000000000000005 RDI: 0000000000000000 > RBP: ffff88800775f000 R08: 0000000000000001 R09: ffff888006e620b8 > R10: ffff888006e620b0 R11: f000000000000000 R12: ffff8880bff39000 > R13: ffff8880bff39000 R14: 0000000000000000 R15: ffff88800604be00 > FS: 0000000000000000(0000) GS:ffff8880f3300000(0000) knlGS:0000000000000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 0000000000000050 CR3: 00000000e932e002 CR4: 00000000003706e0 > Kernel panic - not syncing: Fatal exception > Kernel Offset: disabled > > info->rq and info->gd are only set in blkfront_connect(), which is > called for state 4 (XenbusStateConnected). Guard against using NULL > variables in blkfront_closing() to avoid the issue. > > The rest of blkfront_closing looks okay. If info->nr_rings is 0, then > for_each_rinfo won't do anything. > > blkfront_remove also needs to check for non-NULL pointers before > cleaning up the gendisk and request queue. > > Fixes: 05d69d950d9d "xen-blkfront: sanitize the removal state machine" > Reported-by: Marek Marczykowski-Górecki <marmarek@xxxxxxxxxxxxxxxxxxxxxx> > Signed-off-by: Jason Andryuk <jandryuk@xxxxxxxxx> Tis looks ok, but do we have anything that prevents races between blkfront_connect, blkfront_closing and blkfront_remove?
