Skip to contentSkip to conversation

rr: nvme-pci: fix races in nvme_setup_io_queues() and UAF introduced by nvme_dev_remove_admin()

PATCH In review Casey Chen 2 participants 10 messages Jun 22, 2021 – Jun 23, 2021 NVMe

lore
Share & tools
lore

Casey Chen

author
#0 lore raw

Found two bugs while power-cycling PCIe NVMe drive for hours:
- Races in nvme_setup_io_queues()
- UAF introduced by nvme_dev_remove_admin(), which is found after fixing the races. Without fixing the races, the system just crashes and cannot survive to reproduce the UAF.

The proposed fixes have been tested for several days for correctness.

0. Code baseline

Tag nvme-5.14-2021-06-08 of repo http://git.infradead.org/nvme.git

1. Testing method

while :; power off one drive;

    sleep $((RANDOM%3)).$((RANDOM%10));
    power on the same drive;
    sleep $((RANDOM%3)).$((RANDOM%10));

done

2. Sample crash trace due to races in nvme_setup_io_queues() (task ID shown after the timestamps)

[11668.533431][  T716] pcieport 0000:87:08.0: pciehp: Slot(402): Card present
...
[11668.681298][T251231] nvme nvme12: pci function 0000:8c:00.0
[11668.681354][T26714] nvme 0000:8c:00.0: enabling device (0100 -> 0102)
[11669.046119][   C31] pcieport 0000:87:08.0: pciehp: pending interrupts 0x0108 from Slot Status
[11669.046142][  T716] pcieport 0000:87:08.0: pciehp: Slot(402): Link Down
[11669.046146][  T716] pcieport 0000:87:08.0: pciehp: Slot(402): Card not present
[11669.046149][  T716] pcieport 0000:87:08.0: pciehp: pciehp_unconfigure_device: domain:bus:dev = 0000:8c:00
[11669.077428][  T716] ------------[ cut here ]------------
[11669.077431][  T716] kernel BUG at drivers/pci/msi.c:348!
[11669.077555][  T716] invalid opcode: 0000 [#1] SMP KASAN
[11669.077658][  T716] CPU: 31 PID: 716 Comm: irq/127-pciehp Not tainted 5.13.0-rc3+
[11669.078022][  T716] RIP: 0010:free_msi_irqs+0x28a/0x2d0
...
[11669.093982][  T716] Call Trace:
[11669.096850][  T716]  pci_free_irq_vectors+0xe/0x20
[11669.099695][  T716]  nvme_dev_disable+0x140/0x760 [nvme]
[11669.102503][  T716]  ? _raw_spin_lock_irqsave+0x9c/0x100
[11669.105271][  T716]  ? trace_hardirqs_on+0x2c/0xe0
[11669.107994][  T716]  nvme_remove+0x191/0x1e0 [nvme]
[11669.110689][  T716]  pci_device_remove+0x6b/0x110
[11669.113316][  T716]  device_release_driver_internal+0x14f/0x280
[11669.115939][  T716]  pci_stop_bus_device+0xcb/0x100
[11669.118515][  T716]  pci_stop_and_remove_bus_device+0xe/0x20
[11669.121079][  T716]  pciehp_unconfigure_device+0xfa/0x200
[11669.123597][  T716]  ? pciehp_configure_device+0x1c0/0x1c0
[11669.126049][  T716]  ? trace_hardirqs_on+0x2c/0xe0
[11669.128444][  T716]  pciehp_disable_slot+0xc4/0x1a0
[11669.130771][  T716]  ? pciehp_runtime_suspend+0x40/0x40
[11669.133054][  T716]  ? __mutex_lock_slowpath+0x10/0x10
[11669.135289][  T716]  ? trace_hardirqs_on+0x2c/0xe0
[11669.137462][  T716]  pciehp_handle_presence_or_link_change+0x15c/0x4f0
[11669.139632][  T716]  ? down_read+0x11f/0x1a0
[11669.141731][  T716]  ? pciehp_handle_disable_request+0x80/0x80
[11669.143817][  T716]  ? rwsem_down_read_slowpath+0x600/0x600
[11669.145851][  T716]  ? __radix_tree_lookup+0xb2/0x130
[11669.147842][  T716]  pciehp_ist+0x19d/0x1a0
[11669.149790][  T716]  ? pciehp_set_indicators+0xe0/0xe0
[11669.151704][  T716]  ? irq_finalize_oneshot.part.46+0x1d0/0x1d0
[11669.153588][  T716]  irq_thread_fn+0x3f/0xa0
[11669.155407][  T716]  irq_thread+0x195/0x290
[11669.157147][  T716]  ? irq_thread_check_affinity.part.49+0xe0/0xe0
[11669.158883][  T716]  ? _raw_read_lock_irq+0x50/0x50
[11669.160611][  T716]  ? _raw_read_lock_irq+0x50/0x50
[11669.162320][  T716]  ? irq_forced_thread_fn+0xf0/0xf0
[11669.164032][  T716]  ? trace_hardirqs_on+0x2c/0xe0
[11669.165731][  T716]  ? irq_thread_check_affinity.part.49+0xe0/0xe0
[11669.167461][  T716]  kthread+0x1c8/0x1f0
[11669.169173][  T716]  ? kthread_parkme+0x40/0x40
[11669.170883][  T716]  ret_from_fork+0x22/0x30

3. KASAN report for the UAF introduced by nvme_dev_remove_admin() (task ID is shown after the timestamp)

[18319.015748][T246989] nvme nvme13: pci function 0000:8c:00.0
[18319.015795][T215541] nvme 0000:8c:00.0: enabling device (0100 -> 0102)
[18319.369086][   C31] pcieport 0000:87:08.0: pciehp: pending interrupts 0x0108 from Slot Status
[18319.369107][  T716] pcieport 0000:87:08.0: pciehp: Slot(402): Link Down
[18319.369111][  T716] pcieport 0000:87:08.0: pciehp: Slot(402): Card not present
[18319.369116][  T716] pcieport 0000:87:08.0: pciehp: pciehp_unconfigure_device: domain:bus:dev = 0000:8c:00
[18320.452045][T215541] nvme nvme13: 88/0/0 default/read/poll queues
[18320.469475][T215541] nvme nvme13: failed to mark controller live state
[18320.469483][T215541] nvme nvme13: Removing after probe failure status: -19
[18320.551295][T215541] ==================================================================
[18320.551299][T215541] BUG: KASAN: use-after-free in __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551311][T215541] Read of size 4 at addr ffff888897904d04 by task kworker/u178:2/215541
[18320.551315][T215541] 
[18320.551318][T215541] CPU: 86 PID: 215541 Comm: kworker/u178:2 Not tainted 5.13.0-rc3+
[18320.551327][T215541] Workqueue: nvme-reset-wq nvme_reset_work [nvme]
[18320.551339][T215541] Call Trace:
[18320.551343][T215541]  dump_stack+0xa4/0xdb
[18320.551354][T215541]  ? __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551359][T215541]  print_address_description.constprop.10+0x3a/0x60
[18320.551366][T215541]  ? __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551372][T215541]  ? __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551377][T215541]  ? blk_mq_update_nr_requests+0x2a0/0x2a0
[18320.551382][T215541]  kasan_report.cold.15+0x7c/0xd8
[18320.551390][T215541]  ? __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551395][T215541]  __blk_mq_all_tag_iter+0x9c/0x3f0
[18320.551401][T215541]  ? blk_mq_update_nr_requests+0x2a0/0x2a0
[18320.551407][T215541]  ? bt_iter+0xf0/0xf0
[18320.551412][T215541]  ? __blk_mq_all_tag_iter+0x2c9/0x3f0
[18320.551417][T215541]  ? blk_mq_update_nr_requests+0x2a0/0x2a0
[18320.551422][T215541]  ? bt_iter+0xf0/0xf0
[18320.551427][T215541]  ? dev_printk_emit+0x95/0xbb
[18320.551436][T215541]  blk_mq_tagset_busy_iter+0x75/0xa0
[18320.551441][T215541]  ? blk_mq_update_nr_requests+0x2a0/0x2a0
[18320.551446][T215541]  ? blk_mq_update_nr_requests+0x2a0/0x2a0
[18320.551451][T215541]  blk_mq_tagset_wait_completed_request+0x86/0xc0
[18320.551457][T215541]  ? blk_mq_tagset_busy_iter+0xa0/0xa0
[18320.551463][T215541]  ? blk_mq_tagset_busy_iter+0x80/0xa0
[18320.551469][T215541]  ? trace_event_raw_event_nvme_setup_cmd+0x2d0/0x2d0 [nvme_core]
[18320.551493][T215541]  nvme_dev_disable+0x4f6/0x760 [nvme]
[18320.551502][T215541]  ? trace_hardirqs_on+0x2c/0xe0
[18320.551510][T215541]  nvme_reset_work+0x226/0x2060 [nvme]
[18320.551520][T215541]  ? nvme_remove+0x1e0/0x1e0 [nvme]
[18320.551528][T215541]  ? __update_load_avg_cfs_rq+0x1d8/0x550
[18320.551537][T215541]  ? down_read+0x11f/0x1a0
[18320.551545][T215541]  ? newidle_balance+0x444/0x690
[18320.551552][T215541]  ? update_load_avg+0x626/0xbe0
[18320.551557][T215541]  ? update_cfs_group+0x1e/0x150
[18320.551562][T215541]  ? load_balance+0x11d0/0x11d0
[18320.551567][T215541]  ? dequeue_entity+0x150/0x730
[18320.551573][T215541]  ? nvme_irq_check+0x60/0x60 [nvme]
[18320.551581][T215541]  ? finish_task_switch+0x101/0x3d0
[18320.551588][T215541]  ? read_word_at_a_time+0xe/0x20
[18320.551594][T215541]  ? strscpy+0xc1/0x1d0
[18320.551598][T215541]  process_one_work+0x4b9/0x7b0
[18320.551604][T215541]  worker_thread+0x72/0x710
[18320.551610][T215541]  ? process_one_work+0x7b0/0x7b0
[18320.551614][T215541]  kthread+0x1c8/0x1f0
[18320.551618][T215541]  ? kthread_parkme+0x40/0x40
[18320.551622][T215541]  ret_from_fork+0x22/0x30
[18320.551630][T215541] 
[18320.551632][T215541] Allocated by task 215541:
[18320.551635][T215541]  kasan_save_stack+0x19/0x40
[18320.551639][T215541]  __kasan_kmalloc+0x7f/0xa0
[18320.551642][T215541]  kmem_cache_alloc_node_trace+0x187/0x2b0
[18320.551648][T215541]  blk_mq_init_tags+0x47/0x100
[18320.551651][T215541]  blk_mq_alloc_rq_map+0x44/0xf0
[18320.551656][T215541]  __blk_mq_alloc_map_and_request+0x7f/0x140
[18320.551661][T215541]  blk_mq_alloc_tag_set+0x25e/0x510
[18320.551666][T215541]  nvme_reset_work+0x14f9/0x2060 [nvme]
[18320.551674][T215541]  process_one_work+0x4b9/0x7b0
[18320.551678][T215541]  worker_thread+0x72/0x710
[18320.551682][T215541]  kthread+0x1c8/0x1f0
[18320.551685][T215541]  ret_from_fork+0x22/0x30
[18320.551689][T215541] 
[18320.551690][T215541] Freed by task 716:
[18320.551693][T215541]  kasan_save_stack+0x19/0x40
[18320.551696][T215541]  kasan_set_track+0x1c/0x30
[18320.551699][T215541]  kasan_set_free_info+0x20/0x30
[18320.551704][T215541]  __kasan_slab_free+0xec/0x130
[18320.551707][T215541]  kfree+0xa8/0x460
[18320.551712][T215541]  blk_mq_free_map_and_requests+0x8d/0xc0
[18320.551717][T215541]  blk_mq_free_tag_set+0x30/0xf0
[18320.551721][T215541]  nvme_remove+0x199/0x1e0 [nvme]
[18320.551729][T215541]  pci_device_remove+0x6b/0x110
[18320.551735][T215541]  device_release_driver_internal+0x14f/0x280
[18320.551744][T215541]  pci_stop_bus_device+0xcb/0x100
[18320.551750][T215541]  pci_stop_and_remove_bus_device+0xe/0x20
[18320.551754][T215541]  pciehp_unconfigure_device+0xfa/0x200
[18320.551761][T215541]  pciehp_disable_slot+0xc4/0x1a0
[18320.551765][T215541]  pciehp_handle_presence_or_link_change+0x15c/0x4f0
[18320.551770][T215541]  pciehp_ist+0x19d/0x1a0
[18320.551774][T215541]  irq_thread_fn+0x3f/0xa0
[18320.551780][T215541]  irq_thread+0x195/0x290
[18320.551783][T215541]  kthread+0x1c8/0x1f0
[18320.551786][T215541]  ret_from_fork+0x22/0x30
[18320.551791][T215541] 
[18320.551792][T215541] The buggy address belongs to the object at ffff888897904d00
[18320.551792][T215541]  which belongs to the cache kmalloc-192 of size 192
[18320.551795][T215541] The buggy address is located 4 bytes inside of
[18320.551795][T215541]  192-byte region [ffff888897904d00, ffff888897904dc0)
[18320.551800][T215541] The buggy address belongs to the page:
[18320.551802][T215541] page:000000002f3df664 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x897904
[18320.551807][T215541] head:000000002f3df664 order:1 compound_mapcount:0
[18320.551810][T215541] flags: 0x200000000010200(slab|head|node=0|zone=2)
[18320.551819][T215541] raw: 0200000000010200 dead000000000100 dead000000000122 ffff88810004ca00
[18320.551824][T215541] raw: 0000000000000000 0000000000200020 00000001ffffffff 0000000000000000
[18320.551826][T215541] page dumped because: kasan: bad access detected
[18320.551828][T215541] 
[18320.551829][T215541] Memory state around the buggy address:
[18320.551832][T215541]  ffff888897904c00: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[18320.551835][T215541]  ffff888897904c80: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
[18320.551838][T215541] >ffff888897904d00: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[18320.551841][T215541]                    ^
[18320.551843][T215541]  ffff888897904d80: fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc fc
[18320.551846][T215541]  ffff888897904e00: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[18320.551848][T215541] ==================================================================
[18320.551850][T215541] Disabling lock debugging due to kernel taint

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

author
#1 lore raw

PATCH 1/2 nvme-pci: Fix multiple races in nvme_setup_io_queues()

Below two paths could overlap each other if we power off a drive quickly after power it on. There are multiple races in nvme_setup_io_queues() because of shutdown_lock missing and improper use of NVMEQ_ENABLED bit.

nvme_reset_work() nvme_remove()
nvme_setup_io_queues() nvme_dev_disable()
... ...

A1  clear NVMEQ_ENABLED bit for admin queue          lock
    retry:                                       B1  nvme_suspend_io_queues()
A2    pci_free_irq() admin queue                 B2  nvme_suspend_queue() admin queue
A3    pci_free_irq_vectors()                         nvme_pci_disable()

A4 nvme_setup_irqs(); B3 pci_free_irq_vectors()
... unlock
A5 queue_request_irq() for admin queue
set NVMEQ_ENABLED bit
...
nvme_create_io_queues()
A6 result = queue_request_irq();
set NVMEQ_ENABLED bit
...
fail to allocate enough IO queues:
A7 nvme_suspend_io_queues()
goto retry

1). If B3 runs in between A1 and A2, it will crash if irqaction haven't
been freed by A2. B2 is supposed to free admin queue IRQ but it simply can't fulfill the job as A1 has cleared NVMEQ_ENABLED bit.

Fix: combine A1 A2 so IRQ get freed as soon as NVMEQ_ENABLED bit get cleared.

2). After solved #1, A2 could race with B3 if A2 is freeing IRQ while B3
is checking irqaction. A3 also could race with B2 if B2 is freeing IRQ while A3 is checking irqaction.

Fix: A2 and A3 take lock for mutual exclusion.

3) A3 could race with B3 since they could run free_msi_irqs() in parallel.

Fix: A3 takes lock for mutual exclusion.

4). A4 could fail to allocate all needed IRQ vectors if A3 and A4 are
interrupted by B3.

Fix: A4 takes lock for mutual exclusion.

5). If A5/A6 happened after B2/B1, B3 will crash since irqaction is not NULL.
They are just allocated by A5/A6.

Fix: Lock queue_request_irq() and setting of NVMEQ_ENABLED bit.

6). A7 could get chance to pci_free_irq() for certain IO queue while B3
is checking irqaction.

Fix: A7 takes lock.

7). nvme_dev->online_queues need to be protected by shutdown_lock. Since it
is not atomic, both paths could modify it using its own copy.

---
 drivers/nvme/host/pci.c | 37 +++++++++++++++++++++++++++++++------
 1 file changed, 31 insertions(+), 6 deletions(-)
drivers/nvme/host/pci.c +31 -6
index 3aa7245a505f..5abc4c3be454 100644
@@ -1590,8 +1590,13 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 		goto release_cq;
 
 	nvmeq->cq_vector = vector;
-	nvme_init_queue(nvmeq, qid);
 
+	/*
+	 * Return early if locking fails as device is being disabled.
+	 */
+	if (!mutex_trylock(&dev->shutdown_lock))
+		return -ENODEV;
+	nvme_init_queue(nvmeq, qid);
 	if (!polled) {
 		result = queue_request_irq(nvmeq);
 		if (result < 0)
@@ -1599,10 +1604,12 @@ static int nvme_create_queue(struct nvme_queue *nvmeq, int qid, bool polled)
 	}
 
 	set_bit(NVMEQ_ENABLED, &nvmeq->flags);
+	mutex_unlock(&dev->shutdown_lock);
 	return result;
 
 release_sq:
 	dev->online_queues--;
+	mutex_unlock(&dev->shutdown_lock);
 	adapter_delete_sq(dev, qid);
 release_cq:
 	adapter_delete_cq(dev, qid);
@@ -2176,7 +2183,16 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	if (nr_io_queues == 0)
 		return 0;
 
-	clear_bit(NVMEQ_ENABLED, &adminq->flags);
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock);
+	if (test_and_clear_bit(NVMEQ_ENABLED, &adminq->flags))
+		pci_free_irq(pdev, 0, adminq);
 
 	if (dev->cmb_use_sqes) {
 		result = nvme_cmb_qdepth(dev, nr_io_queues,
@@ -2192,14 +2208,17 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 		result = nvme_remap_bar(dev, size);
 		if (!result)
 			break;
-		if (!--nr_io_queues)
+		if (!--nr_io_queues) {
+			mutex_unlock(&dev->shutdown_lock);
 			return -ENOMEM;
+		}
 	} while (1);
 	adminq->q_db = dev->dbs;
 
  retry:
 	/* Deregister the admin queue's interrupt */
-	pci_free_irq(pdev, 0, adminq);
+	if (test_and_clear_bit(NVMEQ_ENABLED, &adminq->flags))
+		pci_free_irq(pdev, 0, adminq);
 
 	/*
 	 * If we enable msix early due to not intx, disable it again before
@@ -2208,8 +2227,10 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	pci_free_irq_vectors(pdev);
 
 	result = nvme_setup_irqs(dev, nr_io_queues);
-	if (result <= 0)
+	if (result <= 0) {
+		mutex_unlock(&dev->shutdown_lock);
 		return -EIO;
+	}
 
 	dev->num_vecs = result;
 	result = max(result - 1, 1);
@@ -2222,9 +2243,12 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	 * number of interrupts.
 	 */
 	result = queue_request_irq(adminq);
-	if (result)
+	if (result) {
+		mutex_unlock(&dev->shutdown_lock);
 		return result;
+	}
 	set_bit(NVMEQ_ENABLED, &adminq->flags);
+	mutex_unlock(&dev->shutdown_lock);
 
 	result = nvme_create_io_queues(dev);
 	if (result || dev->online_queues < 2)
@@ -2233,6 +2257,7 @@ static int nvme_setup_io_queues(struct nvme_dev *dev)
 	if (dev->online_queues - 1 < dev->max_qid) {
 		nr_io_queues = dev->online_queues - 1;
 		nvme_disable_io_queues(dev);
+		mutex_lock(&dev->shutdown_lock);
 		nvme_suspend_io_queues(dev);
 		goto retry;
 	}
Signature · 7 lines
2.17.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
maintainer
#2 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

drivers/nvme/host/pci.c static int nvme_setup_io_queues(struct nvme_dev *dev)

Quoted 8 lines · Casey Chen
drivers/nvme/host/pci.c +8 -0
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock)

Sorry, I wasn't clear in previous review. All of the shutdown_locks taken after initialization need to by mutex_trylock()'s. If you have to wait to get the lock, the device is not going to be suitable for continuing initialization after the lock is available.

And looking at this again, if trylock is successful, I think you need to verify controller state is suitable for continuing the initialization.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

author
#3 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

Quoted 14 lines · Keith Busch

drivers/nvme/host/pci.c static int nvme_setup_io_queues(struct nvme_dev *dev)

Quoted 8 lines · Casey Chen
drivers/nvme/host/pci.c +8 -0
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock)

Sorry, I wasn't clear in previous review. All of the shutdown_locks taken after initialization need to by mutex_trylock()'s. If you have to wait to get the lock, the device is not going to be suitable for continuing initialization after the lock is available.

Quoted 2 lines

And looking at this again, if trylock is successful, I think you need to verify controller state is suitable for continuing the initialization.

I assume you mean all the mutex_lock() except the one in nvme_dev_disable(), right ? I agree with replacing all mutex_lock() with mutex_trylock() except the one in nvme_dev_disable(). But I doubt about checking controller state after lock being acquired. Two reasons:

1. What state can use to check ? Suppose there is a state set on surprise removal, what if the state check happen before it is set while surprise removal is on its way ?

2. Even we don't check controller state after acquiring lock, the initialization still fail fairly soon if surprise removal happens. We don't sacrifice any performance.

Casey

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

maintainer
#4 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

Quoted 15 lines · Casey Chen
Quoted 14 lines · Keith Busch

drivers/nvme/host/pci.c static int nvme_setup_io_queues(struct nvme_dev *dev)

Quoted 8 lines · Casey Chen
drivers/nvme/host/pci.c +8 -0
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock)

Sorry, I wasn't clear in previous review. All of the shutdown_locks taken after initialization need to by mutex_trylock()'s. If you have to wait to get the lock, the device is not going to be suitable for continuing initialization after the lock is available.

Quoted 5 lines
Quoted 2 lines

And looking at this again, if trylock is successful, I think you need to verify controller state is suitable for continuing the initialization.

I assume you mean all the mutex_lock() except the one in nvme_dev_disable(), right ? I agree with replacing all mutex_lock() with mutex_trylock() except the one in nvme_dev_disable().

I only mean the locks you are adding. The existing usages are correct as-is.

Quoted 4 lines

But I doubt about checking controller state
after lock being acquired. Two reasons:

1. What state can use to check ?

If the state is not NVME_CTRL_CONNECTING after successfully taking the shutdown_lock, then the controller is not in a state that can proceed with IO queue setup.

Quoted 7 lines

Suppose there is a state set on surprise
removal, what if the state check happen before it is set while surprise removal is on its way ?

2. Even we don't check controller state after acquiring lock, the initialization still fail fairly soon if surprise removal happens. We don't sacrifice any performance.

Nothing to do with performance; we're not in io path. The suggestion is about avoiding access to a non-existent device. The timing of a surprise removal can always defeat such a check, so my suggestion is not critical since we expect that to be handled anyway, but in my experience, platforms work better if the OS prevents such access when it knows it will fail. It's not full-proof.

But as I said, everything should work even without the check, so feel free to omit it.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

author
#5 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

Quoted 46 lines · Keith Busch
Quoted 15 lines · Casey Chen
Quoted 14 lines · Keith Busch

drivers/nvme/host/pci.c static int nvme_setup_io_queues(struct nvme_dev *dev)

Quoted 8 lines · Casey Chen
drivers/nvme/host/pci.c +8 -0
+	/*
+	 * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions
+	 * from set to unset. If there is a window to it is truely freed,
+	 * pci_free_irq_vectors() jumping into this window will crash.
+	 * And take lock to avoid racing with pci_free_irq_vectors() in
+	 * nvme_dev_disable() path.
+	 */
+	mutex_lock(&dev->shutdown_lock)

Sorry, I wasn't clear in previous review. All of the shutdown_locks taken after initialization need to by mutex_trylock()'s. If you have to wait to get the lock, the device is not going to be suitable for continuing initialization after the lock is available.

Quoted 5 lines
Quoted 2 lines

And looking at this again, if trylock is successful, I think you need to verify controller state is suitable for continuing the initialization.

I assume you mean all the mutex_lock() except the one in nvme_dev_disable(), right ? I agree with replacing all mutex_lock() with mutex_trylock() except the one in nvme_dev_disable().

I only mean the locks you are adding. The existing usages are correct as-is.

Quoted 4 lines

But I doubt about checking controller state
after lock being acquired. Two reasons:

1. What state can use to check ?

If the state is not NVME_CTRL_CONNECTING after successfully taking the shutdown_lock, then the controller is not in a state that can proceed with IO queue setup.

Quoted 7 lines

Suppose there is a state set on surprise
removal, what if the state check happen before it is set while surprise removal is on its way ?

2. Even we don't check controller state after acquiring lock, the initialization still fail fairly soon if surprise removal happens. We don't sacrifice any performance.

Nothing to do with performance; we're not in io path. The suggestion is about avoiding access to a non-existent device. The timing of a surprise removal can always defeat such a check, so my suggestion is not critical since we expect that to be handled anyway, but in my experience, platforms work better if the OS prevents such access when it knows it will fail. It's not full-proof.

Quoted 2 lines

But as I said, everything should work even without the check, so feel free to omit it.

Thanks for explanation. I prefer not having the check even I had came up with a solution as below.

This looks cumbersome. If you can make it better, I might adopt it. Otherwise, I will send new patches without the check. :)

+/*
+ * Try getting shutdown_lock while setting up IO queues.
+ * Caller remember to unlock.
+ */
+static int nvme_setup_io_queues_trylock(struct nvme_dev *dev) +{
+       /*
+       * Lock is being held by nvme_dev_disable(), fail early. +       */
+       if (!mutex_trylock(&dev->shutdown_lock))
+               return -ENODEV;
+
+       /*
+       * Controller is in wrong state, fail early.
+        */
+       if (dev->ctrl.state == NVME_CTRL_CONNECTING) +               return -ENODEV;
+
+       return 0;
+}
+

Sample call:

-       mutex_lock(&dev->shutdown_lock);
+       if ((result = nvme_setup_io_queues_trylock(dev))) +               return result;

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

author
#6 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

Quoted 78 lines · Casey Chen
Quoted 47 lines · Keith Busch
Quoted 16 lines · Casey Chen
Quoted 15 lines · Keith Busch
Quoted 9 lines · Casey Chen

+    /*
+     * Free IRQ resources as soon as NVMEQ_ENABLED bit transitions +     * from set to unset. If there is a window to it is truely freed,
+     * pci_free_irq_vectors() jumping into this window will crash. +     * And take lock to avoid racing with pci_free_irq_vectors() in +     * nvme_dev_disable() path.
+     */
+    mutex_lock(&dev->shutdown_lock)

Sorry, I wasn't clear in previous review. All of the shutdown_locks taken after initialization need to by mutex_trylock()'s. If you have to wait to get the lock, the device is not going to be suitable for continuing initialization after the lock is available.

Quoted 5 lines
Quoted 2 lines

And looking at this again, if trylock is successful, I think you need to verify controller state is suitable for continuing the initialization.

I assume you mean all the mutex_lock() except the one in nvme_dev_disable(), right ? I agree with replacing all mutex_lock() with mutex_trylock() except the one in nvme_dev_disable().

I only mean the locks you are adding. The existing usages are correct as-is.

Quoted 4 lines

But I doubt about checking controller state
after lock being acquired. Two reasons:

1. What state can use to check ?

If the state is not NVME_CTRL_CONNECTING after successfully taking the shutdown_lock, then the controller is not in a state that can proceed with IO queue setup.

Quoted 7 lines

Suppose there is a state set on surprise
removal, what if the state check happen before it is set while surprise removal is on its way ?

2. Even we don't check controller state after acquiring lock, the initialization still fail fairly soon if surprise removal happens. We don't sacrifice any performance.

Nothing to do with performance; we're not in io path. The suggestion is about avoiding access to a non-existent device. The timing of a surprise removal can always defeat such a check, so my suggestion is not critical since we expect that to be handled anyway, but in my experience, platforms work better if the OS prevents such access when it knows it will fail. It's not full-proof.

Quoted 2 lines

But as I said, everything should work even without the check, so feel free to omit it.

Thanks for explanation. I prefer not having the check even I had came up with a solution as below.

This looks cumbersome. If you can make it better, I might adopt it. Otherwise, I will send new patches without the check. :)

(quoted diff) +21 -0
+/*
+ * Try getting shutdown_lock while setting up IO queues.
+ * Caller remember to unlock.
+ */
+static int nvme_setup_io_queues_trylock(struct nvme_dev *dev)
+{
+       /*
+       * Lock is being held by nvme_dev_disable(), fail early.
+       */
+       if (!mutex_trylock(&dev->shutdown_lock))
+               return -ENODEV;
+
+       /*
+       * Controller is in wrong state, fail early.
+        */
+       if (dev->ctrl.state == NVME_CTRL_CONNECTING)
+               return -ENODEV;
+
+       return 0;
+}
+
Quoted 6 lines

Sample call:

-       mutex_lock(&dev->shutdown_lock);
+       if ((result = nvme_setup_io_queues_trylock(dev))) +               return result;

Sorry I sent in a rush, should be:

+/*
+ * Try getting shutdown_lock while setting up IO queues.
+ * Caller remember to unlock if success.
+ */
+static int nvme_setup_io_queues_trylock(struct nvme_dev *dev) +{
+       /*
+       * Lock is being held by nvme_dev_disable(), fail early. +       */
+       if (!mutex_trylock(&dev->shutdown_lock))
+               return -ENODEV;
+
+       /*
+       * Controller is in wrong state, fail early.
+        */
+       if (dev->ctrl.state != NVME_CTRL_CONNECTING) { +               mutex_unlock(&dev->shutdown_lock);
+               return -ENODEV;
+       }
+
+       return 0;
+}
+

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

maintainer
#7 lore raw

Re: [PATCH 1/2] nvme-pci: Fix multiple races in nvme_setup_io_queues()

Quoted 13 lines · Casey Chen
(quoted diff) +13 -0
+/*
+ * Try getting shutdown_lock while setting up IO queues.
+ * Caller remember to unlock if success.
+ */
+static int nvme_setup_io_queues_trylock(struct nvme_dev *dev)
+{
+       /*
+       * Lock is being held by nvme_dev_disable(), fail early.
+       */
+       if (!mutex_trylock(&dev->shutdown_lock))
+               return -ENODEV;
+
+       /*

Review

Quoted 10 lines
(quoted diff) +10 -0
+       * Controller is in wrong state, fail early.
+        */
+       if (dev->ctrl.state != NVME_CTRL_CONNECTING) {
+               mutex_unlock(&dev->shutdown_lock);
+               return -ENODEV;
+       }
+
+       return 0;
+}
+

Yes, thank you, this is aligned with what I suggested.

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme

author
#8 lore raw

PATCH 2/2 nvme-pci: Fix UAF introduced by nvme_dev_remove_admin()

In reply to Casey Chen

nvme_dev_remove_admin() could free admin_q and admin_tagset while they are being accessed by nvme_dev_disable(), which could come from nvme_remove_dead_ctrl() by nvme_reset_work() during cleanup.

Commit cb4bfda62afa ("nvme-pci: fix hot removal during error handling") was to avoid requests being stuck on a removed controller by killing admin queue. But the later fix c8e9e9b7646e ("nvme-pci: unquiesce admin queue on shutdown"), together with nvme_dev_disable(dev, true) right before nvme_dev_remove_admin() could help dispatch requests and fail them early, so we don't need nvme_dev_remove_admin() any more.

Fixes cb4bfda62afa ("nvme-pci: fix hot removal during error handling")

upstream_status: upstream

---
 drivers/nvme/host/pci.c | 1 -
 1 file changed, 1 deletion(-)
drivers/nvme/host/pci.c +0 -1
index 5abc4c3be454..49ef2027bcb0 100644
@@ -3023,7 +3023,6 @@ static void nvme_remove(struct pci_dev *pdev)
 	if (!pci_device_is_present(pdev)) {
 		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
 		nvme_dev_disable(dev, true);
-		nvme_dev_remove_admin(dev);
 	}
 
 	flush_work(&dev->ctrl.reset_work);
Signature · 7 lines
2.17.1


_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme
maintainer Reviewed-by on v1 2/2
#9 lore raw

Re: [PATCH 2/2] nvme-pci: Fix UAF introduced by nvme_dev_remove_admin()

Quoted 4 lines · Casey Chen

nvme_dev_remove_admin() could free admin_q and admin_tagset while they are being accessed by nvme_dev_disable(), which could come from nvme_remove_dead_ctrl() by nvme_reset_work() during cleanup.

Quoted 8 lines

Commit cb4bfda62afa ("nvme-pci: fix hot removal during error handling") was to avoid requests being stuck on a removed controller by killing admin queue. But the later fix c8e9e9b7646e ("nvme-pci: unquiesce admin queue on shutdown"), together with nvme_dev_disable(dev, true) right before nvme_dev_remove_admin() could help dispatch requests and fail them early, so we don't need nvme_dev_remove_admin() any more.

Fixes cb4bfda62afa ("nvme-pci: fix hot removal during error handling")

Thanks, this change looks good.

Quoted 19 lines

upstream_status: upstream

---
 drivers/nvme/host/pci.c | 1 -
 1 file changed, 1 deletion(-)
drivers/nvme/host/pci.c +0 -1
index 5abc4c3be454..49ef2027bcb0 100644
@@ -3023,7 +3023,6 @@ static void nvme_remove(struct pci_dev *pdev)
 	if (!pci_device_is_present(pdev)) {
 		nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD);
 		nvme_dev_disable(dev, true);
-		nvme_dev_remove_admin(dev);
 	}
 
 	flush_work(&dev->ctrl.reset_work);
Signature · 1 line
2.17.1

_______________________________________________
Linux-nvme mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-nvme