kernel/git/stable/linux.git/drivers/block/nbd.c, branch linux-5.17.y

nbd: fix io hung while disconnecting device

2022-06-14T16:41:40+00:00

[ Upstream commit 09dadb5985023e27d4740ebd17e6fea4640110e5 ] In our tests, "qemu-nbd" triggers a io hung: INFO: task qemu-nbd:11445 blocked for more than 368 seconds. Not tainted 5.18.0-rc3-next-20220422-00003-g2176915513ca #884 "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. task:qemu-nbd state:D stack: 0 pid:11445 ppid: 1 flags:0x00000000 Call Trace: __schedule+0x480/0x1050 ? _raw_spin_lock_irqsave+0x3e/0xb0 schedule+0x9c/0x1b0 blk_mq_freeze_queue_wait+0x9d/0xf0 ? ipi_rseq+0x70/0x70 blk_mq_freeze_queue+0x2b/0x40 nbd_add_socket+0x6b/0x270 [nbd] nbd_ioctl+0x383/0x510 [nbd] blkdev_ioctl+0x18e/0x3e0 __x64_sys_ioctl+0xac/0x120 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7fd8ff706577 RSP: 002b:00007fd8fcdfebf8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 0000000040000000 RCX: 00007fd8ff706577 RDX: 000000000000000d RSI: 000000000000ab00 RDI: 000000000000000f RBP: 000000000000000f R08: 000000000000fbe8 R09: 000055fe497c62b0 R10: 00000002aff20000 R11: 0000000000000246 R12: 000000000000006d R13: 0000000000000000 R14: 00007ffe82dc5e70 R15: 00007fd8fcdff9c0 "qemu-ndb -d" will call ioctl 'NBD_DISCONNECT' first, however, following message was found: block nbd0: Send disconnect failed -32 Which indicate that something is wrong with the server. Then, "qemu-nbd -d" will call ioctl 'NBD_CLEAR_SOCK', however ioctl can't clear requests after commit 2516ab1543fd("nbd: only clear the queue on device teardown"). And in the meantime, request can't complete through timeout because nbd_xmit_timeout() will always return 'BLK_EH_RESET_TIMER', which means such request will never be completed in this situation. Now that the flag 'NBD_CMD_INFLIGHT' can make sure requests won't complete multiple times, switch back to call nbd_clear_sock() in nbd_clear_sock_ioctl(), so that inflight requests can be cleared. Signed-off-by: Yu Kuai Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20220521073749.3146892-5-yukuai3@huawei.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

nbd: fix race between nbd_alloc_config() and module removal

2022-06-14T16:41:40+00:00

[ Upstream commit c55b2b983b0fa012942c3eb16384b2b722caa810 ] When nbd module is being removing, nbd_alloc_config() may be called concurrently by nbd_genl_connect(), although try_module_get() will return false, but nbd_alloc_config() doesn't handle it. The race may lead to the leak of nbd_config and its related resources (e.g, recv_workq) and oops in nbd_read_stat() due to the unload of nbd module as shown below: BUG: kernel NULL pointer dereference, address: 0000000000000040 Oops: 0000 [#1] SMP PTI CPU: 5 PID: 13840 Comm: kworker/u17:33 Not tainted 5.14.0+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996) Workqueue: knbd16-recv recv_work [nbd] RIP: 0010:nbd_read_stat.cold+0x130/0x1a4 [nbd] Call Trace: recv_work+0x3b/0xb0 [nbd] process_one_work+0x1ed/0x390 worker_thread+0x4a/0x3d0 kthread+0x12a/0x150 ret_from_fork+0x22/0x30 Fixing it by checking the return value of try_module_get() in nbd_alloc_config(). As nbd_alloc_config() may return ERR_PTR(-ENODEV), assign nbd->config only when nbd_alloc_config() succeeds to ensure the value of nbd->config is binary (valid or NULL). Also adding a debug message to check the reference counter of nbd_config during module removal. Signed-off-by: Hou Tao Signed-off-by: Yu Kuai Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20220521073749.3146892-3-yukuai3@huawei.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

nbd: call genl_unregister_family() first in nbd_cleanup()

2022-06-14T16:41:39+00:00

[ Upstream commit 06c4da89c24e7023ea448cadf8e9daf06a0aae6e ] Otherwise there may be race between module removal and the handling of netlink command, which can lead to the oops as shown below: BUG: kernel NULL pointer dereference, address: 0000000000000098 Oops: 0002 [#1] SMP PTI CPU: 1 PID: 31299 Comm: nbd-client Tainted: G E 5.14.0-rc4 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996) RIP: 0010:down_write+0x1a/0x50 Call Trace: start_creating+0x89/0x130 debugfs_create_dir+0x1b/0x130 nbd_start_device+0x13d/0x390 [nbd] nbd_genl_connect+0x42f/0x748 [nbd] genl_family_rcv_msg_doit.isra.0+0xec/0x150 genl_rcv_msg+0xe5/0x1e0 netlink_rcv_skb+0x55/0x100 genl_rcv+0x29/0x40 netlink_unicast+0x1a8/0x250 netlink_sendmsg+0x21b/0x430 ____sys_sendmsg+0x2a4/0x2d0 ___sys_sendmsg+0x81/0xc0 __sys_sendmsg+0x62/0xb0 __x64_sys_sendmsg+0x1f/0x30 do_syscall_64+0x3b/0xc0 entry_SYSCALL_64_after_hwframe+0x44/0xae Modules linked in: nbd(E-) Signed-off-by: Hou Tao Signed-off-by: Yu Kuai Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20220521073749.3146892-2-yukuai3@huawei.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

nbd: don't clear 'NBD_CMD_INFLIGHT' flag if request is not completed

2022-06-14T16:41:22+00:00

[ Upstream commit 2895f1831e911ca87d4efdf43e35eb72a0c7e66e ] Otherwise io will hung because request will only be completed if the cmd has the flag 'NBD_CMD_INFLIGHT'. Fixes: 07175cb1baf4 ("nbd: make sure request completion won't concurrent") Signed-off-by: Yu Kuai Link: https://lore.kernel.org/r/20220521073749.3146892-4-yukuai3@huawei.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

nbd: Fix hung on disconnect request if socket is closed before

2022-06-09T08:25:28+00:00

[ Upstream commit 491bf8f236fdeec698fa6744993f1ecf3fafd1a5 ] When userspace closes the socket before sending a disconnect request, the following I/O requests will be blocked in wait_for_reconnect() until dead timeout. This will cause the following disconnect request also hung on blk_mq_quiesce_queue(). That means we have no way to disconnect a nbd device if there are some I/O requests waiting for reconnecting until dead timeout. It's not expected. So let's wake up the thread waiting for reconnecting directly when a disconnect request is sent. Reported-by: Xu Jianhai Signed-off-by: Xie Yongji Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20220322080639.142-1-xieyongji@bytedance.com Signed-off-by: Jens Axboe Signed-off-by: Sasha Levin

Revert "nbd: fix possible overflow on 'first_minor' in nbd_dev_add()"

2022-04-08T11:59:04+00:00

commit 7198bfc2017644c6b92d2ecef9b8b8e0363bb5fd upstream. This reverts commit 6d35d04a9e18990040e87d2bbf72689252669d54. Both Gabriel and Borislav report that this commit casues a regression with nbd: sysfs: cannot create duplicate filename '/dev/block/43:0' Revert it before 5.18-rc1 and we'll investigage this separately in due time. Link: https://lore.kernel.org/all/YkiJTnFOt9bTv6A2@zn.tnic/ Reported-by: Gabriel L. Somlo Reported-by: Borislav Petkov Signed-off-by: Jens Axboe Signed-off-by: Greg Kroah-Hartman

nbd: fix possible overflow on 'first_minor' in nbd_dev_add()

2022-04-08T11:59:00+00:00

commit 6d35d04a9e18990040e87d2bbf72689252669d54 upstream. When 'index' is a big numbers, it may become negative which forced to 'int'. then 'index << part_shift' might overflow to a positive value that is not greater than '0xfffff', then sysfs might complains about duplicate creation. Because of this, move the 'index' judgment to the front will fix it and be better. Fixes: b0d9111a2d53 ("nbd: use an idr to keep track of nbd devices") Fixes: 940c264984fd ("nbd: fix possible overflow for 'first_minor' in nbd_dev_add()") Signed-off-by: Zhang Wensheng Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20220310093224.4002895-1-zhangwensheng5@huawei.com Signed-off-by: Jens Axboe Signed-off-by: Greg Kroah-Hartman

Merge tag 'for-5.16/drivers-2021-11-09' of git://git.kernel.dk/linux-block

2021-11-09T19:24:08+00:00

Pull more block driver updates from Jens Axboe: - Last series adding error handling support for add_disk() in drivers. After this one, and once the SCSI side has been merged, we can finally annotate add_disk() as must_check. (Luis) - bcache fixes (Coly) - zram fixes (Ming) - ataflop locking fix (Tetsuo) - nbd fixes (Ye, Yu) - MD merge via Song - Cleanup (Yang) - sysfs fix (Guoqing) - Misc fixes (Geert, Wu, luo) * tag 'for-5.16/drivers-2021-11-09' of git://git.kernel.dk/linux-block: (34 commits) bcache: Revert "bcache: use bvec_virt" ataflop: Add missing semicolon to return statement floppy: address add_disk() error handling on probe ataflop: address add_disk() error handling on probe block: update __register_blkdev() probe documentation ataflop: remove ataflop_probe_lock mutex mtd/ubi/block: add error handling support for add_disk() block/sunvdc: add error handling support for add_disk() z2ram: add error handling support for add_disk() nvdimm/pmem: use add_disk() error handling nvdimm/pmem: cleanup the disk if pmem_release_disk() is yet assigned nvdimm/blk: add error handling support for add_disk() nvdimm/blk: avoid calling del_gendisk() on early failures nvdimm/btt: add error handling support for add_disk() nvdimm/btt: use goto error labels on btt_blk_init() loop: Remove duplicate assignments drbd: Fix double free problem in drbd_create_device nvdimm/btt: do not call del_gendisk() if not needed bcache: fix use-after-free problem in bcache_device_free() zram: replace fsync_bdev with sync_blockdev ...

nbd: error out if socket index doesn't match in nbd_handle_reply()

2021-11-02T20:42:41+00:00

commit fcf3d633d8e1 ("nbd: check sock index in nbd_read_stat()") just add error message when socket index doesn't match. Since the request and reply must be transmitted over the same socket, it's ok to error out in such situation. Signed-off-by: Yu Kuai Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20211101092538.1155842-1-yukuai3@huawei.com Signed-off-by: Jens Axboe

nbd: Fix hungtask when nbd_config_put

2021-11-02T16:50:27+00:00

I got follow issue: [ 247.381177] INFO: task kworker/u10:0:47 blocked for more than 120 seconds. [ 247.382644] Not tainted 4.19.90-dirty #140 [ 247.383502] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.385027] Call Trace: [ 247.388384] schedule+0xb8/0x3c0 [ 247.388966] schedule_timeout+0x2b4/0x380 [ 247.392815] wait_for_completion+0x367/0x510 [ 247.397713] flush_workqueue+0x32b/0x1340 [ 247.402700] drain_workqueue+0xda/0x3c0 [ 247.403442] destroy_workqueue+0x7b/0x690 [ 247.405014] nbd_config_put.cold+0x2f9/0x5b6 [ 247.405823] recv_work+0x1fd/0x2b0 [ 247.406485] process_one_work+0x70b/0x1610 [ 247.407262] worker_thread+0x5a9/0x1060 [ 247.408699] kthread+0x35e/0x430 [ 247.410918] ret_from_fork+0x1f/0x30 We can reproduce issue as follows: 1. Inject memory fault in nbd_start_device -1244,10 +1248,18 @@ static int nbd_start_device(struct nbd_device *nbd) nbd_dev_dbg_init(nbd); for (i = 0; i < num_connections; i++) { struct recv_thread_args *args; - - args = kzalloc(sizeof(*args), GFP_KERNEL); + + if (i == 1) { + args = NULL; + printk("%s: inject malloc error\n", __func__); + } + else + args = kzalloc(sizeof(*args), GFP_KERNEL); 2. Inject delay in recv_work -757,6 +760,8 @@ static void recv_work(struct work_struct *work) blk_mq_complete_request(blk_mq_rq_from_pdu(cmd)); } + printk("%s: comm=%s pid=%d\n", __func__, current->comm, current->pid); + mdelay(5 * 1000); nbd_config_put(nbd); atomic_dec(&config->recv_threads); wake_up(&config->recv_wq); 3. Create nbd server nbd-server 8000 /tmp/disk 4. Create nbd client nbd-client localhost 8000 /dev/nbd1 Then will trigger above issue. Reason is when add delay in recv_work, lead to release the last reference of 'nbd->config_refs'. nbd_config_put will call flush_workqueue to make all work finish. Obviously, it will lead to deadloop. To solve this issue, according to Josef's suggestion move 'recv_work' init from start device to nbd_dev_add, then destroy 'recv_work'when nbd device teardown. Signed-off-by: Ye Bin Reviewed-by: Josef Bacik Link: https://lore.kernel.org/r/20211102015237.2309763-5-yebin10@huawei.com Signed-off-by: Jens Axboe