Skip to contentSkip to conversation

nvme: init ctrl memleak and debugfs fix

PATCH Applied Chaitanya Kulkarni 6 participants 14 messages Apr 28, 2023 – May 15, 2023 NVMe

lore
Share & tools
lore
#0 lore raw

v2 cover nvme: init ctrl memleak and debugfs fix

Hi,

There is memleak in nvme_ctrl_dhchap_secret_store(), nvme_ctrl_dhchap_ctrl_secret_store(), and nvme_init_ctrl().

Fix dhchap memleaks by adding right kfree() :-

unreferenced object 0xffff8886376ea800 (size 64):
comm "check", pid 22048, jiffies 4344316705 (age 92.199s) hex dump (first 32 bytes):

    44 48 48 43 2d 31 3a 30 30 3a 6e 78 72 35 4b 67  DHHC-1:00:nxr5Kg
    75 58 34 75 6f 41 78 73 4a 61 34 63 2f 68 75 4c  uX4uoAxsJa4c/huL
  backtrace:
    [<0000000030ce5d4b>] __kmalloc+0x4b/0x130
    [<000000009be1cdc1>] nvme_ctrl_dhchap_secret_store+0x8f/0x160 [nvme_core]
    [<00000000ac06c96a>] kernfs_fop_write_iter+0x12b/0x1c0
    [<00000000437e7ced>] vfs_write+0x2ba/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc

Fix dev_pm_qos related memleak in nvme_init_ctrl() with adding dev_pm_qos_hide_latency_tolerance() in the error unwind code :-

nvme/044 (Test bi-directional authentication) [passed]

    runtime  2.111s  ...  2.124s
unreferenced object 0xffff888110c46240 (size 96):
  comm "nvme", pid 33461, jiffies 4345365353 (age 75.586s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<0000000069ac2cec>] kmalloc_trace+0x25/0x90
    [<000000006acc66d5>] dev_pm_qos_update_user_latency_tolerance+0x6f/0x100
    [<00000000cc376ea7>] nvme_init_ctrl+0x38e/0x410 [nvme_core]
    [<000000007df61b4b>] 0xffffffffc05e88b3
    [<00000000d152b985>] 0xffffffffc05744cb
    [<00000000f04a4041>] vfs_write+0xc5/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc

When blktests are running multiple times we get following messages consistently :-

[  147.853464] debugfs: Directory 'nvme1' with parent '/' already present!
[  147.853973] nvme1: failed to create debugfs attr
[  148.802490] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.803244] nvme1: failed to create debugfs attr
[  148.877304] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.877775] nvme1: failed to create debugfs attr
[  149.816652] debugfs: Directory 'nvme1' with parent '/' already present!
[  149.818011] nvme1: failed to create debugfs attr

Fix those by adding nvme_fault inject_fini() call in the error unwind code path.

Still trying to find out device_add() memleak, will send patch soon.

-ck

v2:-

1. In first two patches add kfree() call after strcmp() if block.
2. Add a patch to fix the nvme_ctrl_init() dev_pm_qos memleak.
3. Add a patch to fix the debugfs directory error message.
4. Add tested by tag on patch 1.

Chaitanya Kulkarni (4):
nvme-core: fix memory leak in dhchap_secret_store
nvme-core: fix memory leak in dhchap_ctrl_secret
nvme-core: add missing fault-injection cleanup
nvme-core: fix dev_pm_qos memleak

 drivers/nvme/host/core.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
Signature · 1 line
2.40.0
#1 lore raw

PATCH v2 1/4 nvme-core: fix memory leak in dhchap_secret_store

Free dhchap_secret in nvme_ctrl_dhchap_secret_store() before we return fix following kmemleack:-

unreferenced object 0xffff8886376ea800 (size 64):
comm "check", pid 22048, jiffies 4344316705 (age 92.199s) hex dump (first 32 bytes):

    44 48 48 43 2d 31 3a 30 30 3a 6e 78 72 35 4b 67  DHHC-1:00:nxr5Kg
    75 58 34 75 6f 41 78 73 4a 61 34 63 2f 68 75 4c  uX4uoAxsJa4c/huL
  backtrace:
    [<0000000030ce5d4b>] __kmalloc+0x4b/0x130
    [<000000009be1cdc1>] nvme_ctrl_dhchap_secret_store+0x8f/0x160 [nvme_core]
    [<00000000ac06c96a>] kernfs_fop_write_iter+0x12b/0x1c0
    [<00000000437e7ced>] vfs_write+0x2ba/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc
unreferenced object 0xffff8886376eaf00 (size 64):
  comm "check", pid 22048, jiffies 4344316736 (age 92.168s)
  hex dump (first 32 bytes):
    44 48 48 43 2d 31 3a 30 30 3a 6e 78 72 35 4b 67  DHHC-1:00:nxr5Kg
    75 58 34 75 6f 41 78 73 4a 61 34 63 2f 68 75 4c  uX4uoAxsJa4c/huL
  backtrace:
    [<0000000030ce5d4b>] __kmalloc+0x4b/0x130
    [<000000009be1cdc1>] nvme_ctrl_dhchap_secret_store+0x8f/0x160 [nvme_core]
    [<00000000ac06c96a>] kernfs_fop_write_iter+0x12b/0x1c0
    [<00000000437e7ced>] vfs_write+0x2ba/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc
Fixes f50fff73d620 ("nvme: implement In-Band authentication")
---
 drivers/nvme/host/core.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/nvme/host/core.c +5 -2
index 1bfd52eae2ee..c7793991faeb 100644
@@ -3825,8 +3825,10 @@ static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
 		int ret;
 
 		ret = nvme_auth_generate_key(dhchap_secret, &key);
-		if (ret)
+		if (ret) {
+			kfree(dhchap_secret);
 			return ret;
+		}
 		kfree(opts->dhchap_secret);
 		opts->dhchap_secret = dhchap_secret;
 		host_key = ctrl->host_key;
@@ -3834,7 +3836,8 @@ static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
 		ctrl->host_key = key;
 		mutex_unlock(&ctrl->dhchap_auth_mutex);
 		nvme_auth_free_key(host_key);
-	}
+	} else
+		kfree(dhchap_secret);
 	/* Start re-authentication */
 	dev_info(ctrl->device, "re-authenticating controller\n");
 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
Signature · 1 line
2.40.0
#3 lore raw

PATCH v2 2/4 nvme-core: fix memory leak in dhchap_ctrl_secret

In reply to Chaitanya Kulkarni

Free dhchap_secret in nvme_ctrl_dhchap_ctrl_secret_store() before we return when nvme_auth_generate_key() returns error.

Fixes f50fff73d620 ("nvme: implement In-Band authentication")
---
 drivers/nvme/host/core.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/nvme/host/core.c +5 -2
index c7793991faeb..42e90d00fc40 100644
@@ -3882,8 +3882,10 @@ static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
 		int ret;
 
 		ret = nvme_auth_generate_key(dhchap_secret, &key);
-		if (ret)
+		if (ret) {
+			kfree(dhchap_secret);
 			return ret;
+		}
 		kfree(opts->dhchap_ctrl_secret);
 		opts->dhchap_ctrl_secret = dhchap_secret;
 		ctrl_key = ctrl->ctrl_key;
@@ -3891,7 +3893,8 @@ static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
 		ctrl->ctrl_key = key;
 		mutex_unlock(&ctrl->dhchap_auth_mutex);
 		nvme_auth_free_key(ctrl_key);
-	}
+	} else
+		kfree(dhchap_secret);
 	/* Start re-authentication */
 	dev_info(ctrl->device, "re-authenticating controller\n");
 	queue_work(nvme_wq, &ctrl->dhchap_auth_work);
Signature · 1 line
2.40.0
#5 lore raw

PATCH v2 3/4 nvme-core: add missing fault-injection cleanup

In reply to Chaitanya Kulkarni

Add missing fault-injection cleanup in nvme_init_ctrl() in the error unwind path that also fixes following message for blktests:-

linux-block (for-next) # grep debugfs debugfs-err.log

[  147.853464] debugfs: Directory 'nvme1' with parent '/' already present!
[  147.853973] nvme1: failed to create debugfs attr
[  148.802490] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.803244] nvme1: failed to create debugfs attr
[  148.877304] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.877775] nvme1: failed to create debugfs attr
[  149.816652] debugfs: Directory 'nvme1' with parent '/' already present!
[  149.818011] nvme1: failed to create debugfs attr
---
 drivers/nvme/host/core.c | 1 +
 1 file changed, 1 insertion(+)
drivers/nvme/host/core.c +1 -0
index 42e90d00fc40..15607281baaa 100644
@@ -5201,6 +5201,7 @@ int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
 
 	return 0;
 out_free_cdev:
+	nvme_fault_inject_fini(&ctrl->fault_inject);
 	cdev_device_del(&ctrl->cdev, ctrl->device);
 out_free_name:
 	nvme_put_ctrl(ctrl);
Signature · 1 line
2.40.0
#8 lore raw

PATCH v2 4/4 nvme-core: fix dev_pm_qos memleak

In reply to Chaitanya Kulkarni

Call dev_pm_qos_hide_latency_tolerance() in the error unwind patch to avoid following kmemleak:-

blktests (master) # kmemleak-clear; ./check nvme/044; blktests (master) # kmemleak-scan ; kmemleak-show
nvme/044 (Test bi-directional authentication) [passed]

    runtime  2.111s  ...  2.124s
unreferenced object 0xffff888110c46240 (size 96):
  comm "nvme", pid 33461, jiffies 4345365353 (age 75.586s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<0000000069ac2cec>] kmalloc_trace+0x25/0x90
    [<000000006acc66d5>] dev_pm_qos_update_user_latency_tolerance+0x6f/0x100
    [<00000000cc376ea7>] nvme_init_ctrl+0x38e/0x410 [nvme_core]
    [<000000007df61b4b>] 0xffffffffc05e88b3
    [<00000000d152b985>] 0xffffffffc05744cb
    [<00000000f04a4041>] vfs_write+0xc5/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc
Fixes f50fff73d620 ("nvme: implement In-Band authentication")
---
 drivers/nvme/host/core.c | 1 +
 1 file changed, 1 insertion(+)
drivers/nvme/host/core.c +1 -0
index 15607281baaa..90dc4b832200 100644
@@ -5202,6 +5202,7 @@ int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
 	return 0;
 out_free_cdev:
 	nvme_fault_inject_fini(&ctrl->fault_inject);
+	dev_pm_qos_hide_latency_tolerance(ctrl->device);
 	cdev_device_del(&ctrl->cdev, ctrl->device);
 out_free_name:
 	nvme_put_ctrl(ctrl);
Signature · 1 line
2.40.0
#11 lore raw
In reply to Chaitanya Kulkarni
Quoted 54 lines · Chaitanya Kulkarni

Hi,

There is memleak in nvme_ctrl_dhchap_secret_store(), nvme_ctrl_dhchap_ctrl_secret_store(), and nvme_init_ctrl().

Fix dhchap memleaks by adding right kfree() :-

unreferenced object 0xffff8886376ea800 (size 64):

   comm "check", pid 22048, jiffies 4344316705 (age 92.199s)
   hex dump (first 32 bytes):
     44 48 48 43 2d 31 3a 30 30 3a 6e 78 72 35 4b 67  DHHC-1:00:nxr5Kg
     75 58 34 75 6f 41 78 73 4a 61 34 63 2f 68 75 4c  uX4uoAxsJa4c/huL
   backtrace:
     [<0000000030ce5d4b>] __kmalloc+0x4b/0x130
     [<000000009be1cdc1>] nvme_ctrl_dhchap_secret_store+0x8f/0x160 [nvme_core]
     [<00000000ac06c96a>] kernfs_fop_write_iter+0x12b/0x1c0
     [<00000000437e7ced>] vfs_write+0x2ba/0x3c0
     [<00000000f9491baf>] ksys_write+0x5f/0xe0
     [<000000001c46513d>] do_syscall_64+0x3b/0x90
     [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc

Fix dev_pm_qos related memleak in nvme_init_ctrl() with adding dev_pm_qos_hide_latency_tolerance() in the error unwind code :-

nvme/044 (Test bi-directional authentication) [passed]

     runtime  2.111s  ...  2.124s
unreferenced object 0xffff888110c46240 (size 96):
   comm "nvme", pid 33461, jiffies 4345365353 (age 75.586s)
   hex dump (first 32 bytes):
     00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
     00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
   backtrace:
     [<0000000069ac2cec>] kmalloc_trace+0x25/0x90
     [<000000006acc66d5>] dev_pm_qos_update_user_latency_tolerance+0x6f/0x100
     [<00000000cc376ea7>] nvme_init_ctrl+0x38e/0x410 [nvme_core]
     [<000000007df61b4b>] 0xffffffffc05e88b3
     [<00000000d152b985>] 0xffffffffc05744cb
     [<00000000f04a4041>] vfs_write+0xc5/0x3c0
     [<00000000f9491baf>] ksys_write+0x5f/0xe0
     [<000000001c46513d>] do_syscall_64+0x3b/0x90
     [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc

When blktests are running multiple times we get following messages consistently :-

[  147.853464] debugfs: Directory 'nvme1' with parent '/' already present!
[  147.853973] nvme1: failed to create debugfs attr
[  148.802490] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.803244] nvme1: failed to create debugfs attr
[  148.877304] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.877775] nvme1: failed to create debugfs attr
[  149.816652] debugfs: Directory 'nvme1' with parent '/' already present!
[  149.818011] nvme1: failed to create debugfs attr
Quoted 8 lines

Fix those by adding nvme_fault inject_fini() call in the error unwind code path.

Still trying to find out device_add() memleak, will send patch soon.

-ck

I've generated this series on the linux-block/for-next, forget to mentioned that in the cover-letter, plz let me know if I should re-post this on linux-block mailing list...

-ck

Tested-by on v2 4/4
#10 lore raw

Re: [PATCH V2 4/4] nvme-core: fix dev_pm_qos memleak

In reply to Chaitanya Kulkarni

Verified the kmemleak issue I reported:
https://lore.kernel.org/linux-nvme/CAHj4cs-nDaKzMx2txO4dbE+Mz9ePwLtU0e3egz+StmzOUgWUrA@mail.gmail.com/

Quoted 43 lines · Chaitanya Kulkarni

Call dev_pm_qos_hide_latency_tolerance() in the error unwind patch to avoid following kmemleak:-

blktests (master) # kmemleak-clear; ./check nvme/044; blktests (master) # kmemleak-scan ; kmemleak-show
nvme/044 (Test bi-directional authentication) [passed]

    runtime  2.111s  ...  2.124s
unreferenced object 0xffff888110c46240 (size 96):
  comm "nvme", pid 33461, jiffies 4345365353 (age 75.586s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
  backtrace:
    [<0000000069ac2cec>] kmalloc_trace+0x25/0x90
    [<000000006acc66d5>] dev_pm_qos_update_user_latency_tolerance+0x6f/0x100
    [<00000000cc376ea7>] nvme_init_ctrl+0x38e/0x410 [nvme_core]
    [<000000007df61b4b>] 0xffffffffc05e88b3
    [<00000000d152b985>] 0xffffffffc05744cb
    [<00000000f04a4041>] vfs_write+0xc5/0x3c0
    [<00000000f9491baf>] ksys_write+0x5f/0xe0
    [<000000001c46513d>] do_syscall_64+0x3b/0x90
    [<00000000ecf348fe>] entry_SYSCALL_64_after_hwframe+0x72/0xdc
Fixes f50fff73d620 ("nvme: implement In-Band authentication")
---
 drivers/nvme/host/core.c | 1 +
 1 file changed, 1 insertion(+)
drivers/nvme/host/core.c +1 -1
index 15607281baaa..90dc4b832200 100644
@@ -5202,6 +5202,7 @@ int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
        return 0;
 out_free_cdev:
        nvme_fault_inject_fini(&ctrl->fault_inject);
+       dev_pm_qos_hide_latency_tolerance(ctrl->device);
        cdev_device_del(&ctrl->cdev, ctrl->device);
 out_free_name:
        nvme_put_ctrl(ctrl);
--
Quoted 3 lines

2.40.0

--
Best Regards,
Yi Zhang

Tested-by on v2 3/4
#7 lore raw

Re: [PATCH V2 3/4] nvme-core: add missing fault-injection cleanup

In reply to Chaitanya Kulkarni

Confirmed the issue was fixed, feel free to add:

Quoted 32 lines · Chaitanya Kulkarni

Add missing fault-injection cleanup in nvme_init_ctrl() in the error unwind path that also fixes following message for blktests:-

linux-block (for-next) # grep debugfs debugfs-err.log

[  147.853464] debugfs: Directory 'nvme1' with parent '/' already present!
[  147.853973] nvme1: failed to create debugfs attr
[  148.802490] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.803244] nvme1: failed to create debugfs attr
[  148.877304] debugfs: Directory 'nvme1' with parent '/' already present!
[  148.877775] nvme1: failed to create debugfs attr
[  149.816652] debugfs: Directory 'nvme1' with parent '/' already present!
[  149.818011] nvme1: failed to create debugfs attr
---
 drivers/nvme/host/core.c | 1 +
 1 file changed, 1 insertion(+)
drivers/nvme/host/core.c +1 -1
index 42e90d00fc40..15607281baaa 100644
@@ -5201,6 +5201,7 @@ int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
 
        return 0;
 out_free_cdev:
+       nvme_fault_inject_fini(&ctrl->fault_inject);
        cdev_device_del(&ctrl->cdev, ctrl->device);
 out_free_name:
        nvme_put_ctrl(ctrl);
--
Quoted 3 lines

2.40.0

--
Best Regards,
Yi Zhang