release /dev/nvmeX if nvme_probe failed

Changes since V1:
* drop patch of remove whitespace
* release iomap and prp_pools in nvme_pci_free_ctrl
* direct use nvme_put_ctrl to release resource if failed after nvme_init_ctrl in nvme_probe.

Weiping Zhang (2):
nvme-pci: release iomap and prp_pools in nvme_pci_free_ctrl
nvme-pci: release nvme_ctrl if failed to alloc memory in nvme_probe

 drivers/nvme/host/pci.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

2.14.1

nvme_free_ctrl is the final release function all allocated resource should be free safety in this function. Release iomap and prp_pools in nvme_pci_free_ctrl, otherwise these two resource cann't be released if init failed in nvme_reset_work.

---
 drivers/nvme/host/pci.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index ba943f211687..92a8b5b2b39e 100644
@@ -2279,6 +2279,8 @@ static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
 	struct nvme_dev *dev = to_nvme_dev(ctrl);
 
 	nvme_dbbuf_dma_free(dev);
+	nvme_release_prp_pools(dev);
+	nvme_dev_unmap(dev);
 	put_device(dev->dev);
 	if (dev->tagset.tags)
 		blk_mq_free_tag_set(&dev->tagset);

@@ -2642,8 +2644,6 @@ static void nvme_remove(struct pci_dev *pdev)
 	nvme_dev_remove_admin(dev);
 	nvme_free_queues(dev, 0);
 	nvme_uninit_ctrl(&dev->ctrl);
-	nvme_release_prp_pools(dev);
-	nvme_dev_unmap(dev);
 	nvme_put_ctrl(&dev->ctrl);
 }
 

2.14.1

I'm getting the following warning with this patch:

[  164.864948] ------------[ cut here ]------------
[  164.864951] kernfs: can not remove 'pools', no directory
[  164.864980] WARNING: CPU: 31 PID: 2904 at fs/kernfs/dir.c:1481 kernfs_remove_by_name_ns+0x84/0x90
[  164.864982] Modules linked in: ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ip_set nfnetlink ebtable_nat ebtable_broute bridge stp llc ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_raw ip6table_security iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_raw iptable_security ebtable_filter ebtables ip6table_filter ip6_tables sunrpc vfat fat intel_rapl x86_pkg_temp_thermal intel_powerclamp coretemp kvm_intel kvm irqbypass crct10dif_pclmul crc32_pclmul ghash_clmulni_intel intel_uncore intel_rapl_perf ipmi_ssif iTCO_wdt joydev iTCO_vendor_support ipmi_si mei_me ipmi_devintf mei ioatdma lpc_ich i2c_i801 ipmi_msghandler dca wmi acpi_power_meter acpi_pad xfs libcrc32c ast i2c_algo_bit drm_kms_helper ttm drm i40e
[  164.865059]  nvme nvme_core crc32c_intel ptp pps_core
[  164.865069] CPU: 31 PID: 2904 Comm: umount Not tainted 4.18.0-rc2+ #197
[  164.865071] Hardware name: Intel Corporation S2600STB/S2600STB, BIOS SE5C620.86B.00.01.0010.010920180151 01/09/2018
[  164.865074] RIP: 0010:kernfs_remove_by_name_ns+0x84/0x90
[  164.865076] Code: 00 31 c0 5b 5d 41 5c c3 48 c7 c7 80 47 34 94 e8 52 0a 5b 00 b8 fe ff ff ff 5b 5d 41 5c c3 48 c7 c7 e8 c9 0d 94 e8 4c 2d d8 ff <0f> 0b b8 fe ff ff ff eb d0 0f 1f 00 0f 1f 44 00 00 41 57 41 56 41
[  164.865128] RSP: 0018:ffffa3af8bc03d60 EFLAGS: 00010282
[  164.865131] RAX: 0000000000000000 RBX: ffff8d54f101b198 RCX: 0000000000000006
[  164.865133] RDX: 0000000000000007 RSI: 0000000000000096 RDI: ffff8d64fe8d6a30
[  164.865134] RBP: ffffffff940d13f5 R08: 000000000000069f R09: 0000000000000004
[  164.865136] R10: ffffed0a00caa640 R11: 0000000000000001 R12: ffff8d54f101b198
[  164.865137] R13: ffff8d54fa0e7af8 R14: ffff8d54fa320800 R15: ffff8d64a3145538
[  164.865140] FS:  00007fec629c5fc0(0000) GS:ffff8d64fe8c0000(0000) knlGS:0000000000000000
[  164.865141] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  164.865143] CR2: 00007efdf3d6b160 CR3: 0000002001c6a005 CR4: 00000000007606e0
[  164.865145] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  164.865147] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  164.865148] PKRU: 55555554
[  164.865149] Call Trace:
[  164.865160]  dma_pool_destroy+0x18c/0x1b0
[  164.865169]  nvme_pci_free_ctrl+0x31/0xa0 [nvme]
[  164.865179]  nvme_free_ctrl+0xba/0x100 [nvme_core]
[  164.865190]  device_release+0x2d/0x80
[  164.865199]  kobject_put+0x81/0x1a0
[  164.865206]  nvme_free_ns+0xc9/0xe0 [nvme_core]
[  164.865214]  __blkdev_put+0x17d/0x1d0
[  164.865221]  ? generic_shutdown_super+0x69/0x110
[  164.865225]  deactivate_locked_super+0x39/0x70
[  164.865230]  cleanup_mnt+0x3b/0x70
[  164.865237]  task_work_run+0x97/0xc0
[  164.865246]  exit_to_usermode_loop+0xd8/0xe0
[  164.865250]  do_syscall_64+0x158/0x160
[  164.865255]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  164.865259] RIP: 0033:0x7fec61a12957
[  164.865260] Code: 55 2c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 31 f6 e9 09 00 00 00 66 0f 1f 84 00 00 00 00 00 b8 a6 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 19 55 2c 00 f7 d8 64 89 01 48
[  164.865312] RSP: 002b:00007ffc1c6070c8 EFLAGS: 00000246 ORIG_RAX: 00000000000000a6
[  164.865315] RAX: 0000000000000000 RBX: 00005608458eb2a0 RCX: 00007fec61a12957
[  164.865316] RDX: 0000000000000001 RSI: 0000000000000000 RDI: 00005608458eb480
[  164.865318] RBP: 00005608458eb480 R08: 00005608458eb4a0 R09: 0000000000000003
[  164.865320] R10: 00000000fffffffd R11: 0000000000000246 R12: 00007fec627b1184
[  164.865321] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000
[  164.865324] ---[ end trace 45a702454817613a ]---

Keith Busch <keith.busch at linux.intel.com> ?2018?7?10??? ??10:46???

Hi Keith,

How to reproduce this problem, I cann't reproduce it (apply these two patch on 4.18.0-rc4) as following steps:
1. rmmod nvme
2. git am these 2 patchs
3. make modules SUBDIRS=drivers/nvme
4. insmod drivers/nvme/host/nvme.ko
5. fdisk /dev/nvme0n1 and create a partition
6. mkfs.ext4 /dev/nvme0n1p1
7. mount /dev/nvme0n1p1 /mnt
8. echo aaaaaa > /mnt/b.log
9. umount /mnt

8.1: Hot remove /dev/nvme0

The reference counting is to ensure that resources needed to close a a holder (the mount point, in your example) aren't released during the 'remove'. We have to release the prp pools and the io memory before returning from the driver's 'remove' because the pci driver is going to release their dependencies.

Anyway, I think the right thing to do is reorder the allocations to something more appropriate:

---

index ba943f211687..ddd441b1516a 100644
@@ -2556,11 +2556,6 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 
 	quirks |= check_vendor_combination_bug(pdev);
 
-	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
-			quirks);
-	if (result)
-		goto release_pools;
-
 	/*
 	 * Double check that our mempool alloc size will cover the biggest
 	 * command we support.

@@ -2578,6 +2573,11 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto release_pools;
 	}
 
+	result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops,
+			quirks);
+	if (result)
+		goto release_mempool;
+
 	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));
 
 	nvme_get_ctrl(&dev->ctrl);

@@ -2585,6 +2585,8 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 
 	return 0;
 
+ release_mempool:
+	mempool_destroy(dev->iod_mempool);
  release_pools:
 	nvme_release_prp_pools(dev);
  unmap:

--

Keith Busch <keith.busch at linux.intel.com> ?2018?7?11??? ??11:03???

Do you means "rm -f /dev/nvme0" ?

It seems no dependency between these two functions, I agree this approach.

I mean remove the device from the pci topology, either physically or simulate with 'echo 1 > /sys/class/nvme/nvme0/device/remove'.

release nvme_ctrl if failed to alloc memory for iod_mempool, otherwise /dev/nvmeX cann't be removed for this nvme controller, any ioctl to this device doesn't work.

---
 drivers/nvme/host/pci.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

index 92a8b5b2b39e..29aff33c4b4c 100644
@@ -2576,8 +2576,10 @@ static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 						(void *) alloc_size,
 						GFP_KERNEL, node);
 	if (!dev->iod_mempool) {
-		result = -ENOMEM;
-		goto release_pools;
+		nvme_uninit_ctrl(&dev->ctrl);
+		/* all above resource will be released after put ctrl */
+		nvme_put_ctrl(&dev->ctrl);
+		return -ENOMEM;
 	}
 
 	dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev));

2.14.1

Jens, Keith: any opinion on these? They seems to look fine to me, so I'd be tempted to merge them for 4.18.

---end quoted text---

Looks fine to me, I'd be OK with queueing these up for 4.18.

Jens Axboe