mtd: spi-nor: fix memory leak when using debugfs_lookup()

PATCH In review Greg Kroah-Hartman 4 participants 8 messages Feb 8, 2023 – Mar 17, 2023

Subsystems (2)

Memory technology devices (MTD) SPI NOR

lore

Share & tools

lore

Greg Kroah-Hartman

author

2023-02-08 16:02 UTC #0 lore raw

PATCH v4 1/1 mtd: spi-nor: fix memory leak when using debugfs_lookup()

When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.

Cc Tudor Ambarus <[email protected]>

Cc Pratyush Yadav <[email protected]>

Cc Michael Walle <[email protected]>

Cc Miquel Raynal <[email protected]>

Cc Richard Weinberger <[email protected]>

Cc Vignesh Raghavendra <[email protected]>

Cc [email protected]

Signed-off-by Greg Kroah-Hartman <[email protected]>

---
v4: reverse the teardown order when the module is unloaded to ensure
that the debugfs directory is still around when the device directory is removed.
v3: create the directory the first time it is used, and then remove it
if it is present when the module is unloaded. More complex than v2, but correct.
v2: fix up to work when module is removed and added, making the fix
much simpler.

 drivers/mtd/spi-nor/core.c    | 14 +++++++++++++-
 drivers/mtd/spi-nor/core.h    |  2 ++
 drivers/mtd/spi-nor/debugfs.c | 12 +++++++++---
 3 files changed, 24 insertions(+), 4 deletions(-)

drivers/mtd/spi-nor/core.c +13 -1

index d67c926bca8b..1452ed6fe5d1 100644
@@ -3335,7 +3335,19 @@ static struct spi_mem_driver spi_nor_driver = {
 	.remove = spi_nor_remove,
 	.shutdown = spi_nor_shutdown,
 };
-module_spi_mem_driver(spi_nor_driver);
+
+static int __init spi_nor_module_init(void)
+{
+	return spi_mem_driver_register(&spi_nor_driver);
+}
+module_init(spi_nor_module_init);
+
+static void __exit spi_nor_module_exit(void)
+{
+	spi_mem_driver_unregister(&spi_nor_driver);
+	spi_nor_debugfs_shutdown();
+}
+module_exit(spi_nor_module_exit);
 
 MODULE_LICENSE("GPL v2");
 MODULE_AUTHOR("Huang Shijie <[email protected]>");

drivers/mtd/spi-nor/core.h +2 -0

index f03b55cf7e6f..e62cd9964456 100644
@@ -713,8 +713,10 @@ static inline struct spi_nor *mtd_to_spi_nor(struct mtd_info *mtd)
 
 #ifdef CONFIG_DEBUG_FS
 void spi_nor_debugfs_register(struct spi_nor *nor);
+void spi_nor_debugfs_shutdown(void);
 #else
 static inline void spi_nor_debugfs_register(struct spi_nor *nor) {}
+static inline void spi_nor_debugfs_shutdown(void) {}
 #endif
 
 #endif /* __LINUX_MTD_SPI_NOR_INTERNAL_H */

drivers/mtd/spi-nor/debugfs.c +9 -3

index ff895f6758ea..cd9c2dc07509 100644
@@ -226,13 +226,13 @@ static void spi_nor_debugfs_unregister(void *data)
 	nor->debugfs_root = NULL;
 }
 
+static struct dentry *rootdir;
+
 void spi_nor_debugfs_register(struct spi_nor *nor)
 {
-	struct dentry *rootdir, *d;
+	struct dentry *d;
 	int ret;
 
-	/* Create rootdir once. Will never be deleted again. */
-	rootdir = debugfs_lookup(SPI_NOR_DEBUGFS_ROOT, NULL);
 	if (!rootdir)
 		rootdir = debugfs_create_dir(SPI_NOR_DEBUGFS_ROOT, NULL);

@@ -247,3 +247,9 @@ void spi_nor_debugfs_register(struct spi_nor *nor)
 	debugfs_create_file("capabilities", 0444, d, nor,
 			    &spi_nor_capabilities_fops);
 }
+
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);
+}

Signature · 1 line

2.39.1

Michael Walle

Reviewed-by on v4 1/1

2023-02-08 16:15 UTC #1 lore raw

Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:

Quoted 5 lines

When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.

Quoted 8 lines

Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>

Reviewed-by Michael Walle <[email protected]>

one nit below I didn't notice earlier, no need to send a new patch version just for that.

drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254

Quoted 4 lines

drivers/mtd/spi-nor/debugfs.c +4 -0

+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);

debugfs_remove() already has a check for NULL.

Greg Kroah-Hartman

author

2023-02-08 17:31 UTC #2 lore raw

Quoted 20 lines · Michael Walle

Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.

Quoted 9 lines

..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);
debugfs_remove() already has a check for NULL.

Argh, I knew that, for some reason I went back and added that check as I was thinking this would have been an issue if rootdir was never created. You are right, it isn't a problem, oh well.

The simplest patches sometimes take the longest time to get right, I'll stop now :)

thanks so much for the reviews,

greg k-h

Greg Kroah-Hartman

author

2023-03-06 06:52 UTC #3 lore raw

In reply to Michael Walle

Quoted 20 lines · Michael Walle

Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.

Quoted 9 lines

..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);
debugfs_remove() already has a check for NULL.

Ah, good catch, I merged this in when I applied it to my tree, thanks!

greg k-h

Miquel Raynal

maintainer

2023-03-06 08:33 UTC #4 lore raw

Hi Greg,

[email protected] wrote on Mon, 6 Mar 2023 07:52:38 +0100:

Quoted 31 lines · Michael Walle

Quoted 20 lines
Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.
Quoted 9 lines
..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);  
debugfs_remove() already has a check for NULL.

Quoted 1 line

Ah, good catch, I merged this in when I applied it to my tree, thanks!

Any reasons why you did apply this patch to your tree? It is a spi-nor fix, I would have expected it to go through mtd.

Thanks,
Miquèl

Greg Kroah-Hartman

author

2023-03-06 09:13 UTC #5 lore raw

Quoted 35 lines · Miquel Raynal

Hi Greg,
[email protected] wrote on Mon, 6 Mar 2023 07:52:38 +0100:
Quoted 31 lines · Michael Walle
Quoted 29 lines
Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.
..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);  
debugfs_remove() already has a check for NULL.

Quoted 4 lines

Quoted 1 line
Ah, good catch, I merged this in when I applied it to my tree, thanks!
Any reasons why you did apply this patch to your tree? It is a spi-nor fix, I would have expected it to go through mtd.

It's been sitting around for a month, I assumed it was lost, so I picked it up. I can revert it if you don't want me to take it for 6.3-final through my driver core tree.

thanks,

greg k-h

Miquel Raynal

maintainer

2023-03-06 11:06 UTC #6 lore raw

Hi Greg,

[email protected] wrote on Mon, 6 Mar 2023 10:13:33 +0100:

Quoted 36 lines · Miquel Raynal

Quoted 35 lines
Hi Greg,
[email protected] wrote on Mon, 6 Mar 2023 07:52:38 +0100:
Quoted 31 lines · Michael Walle
Quoted 29 lines
Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.
..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);    
debugfs_remove() already has a check for NULL.

Quoted 7 lines

Quoted 4 lines
Quoted 1 line
Ah, good catch, I merged this in when I applied it to my tree, thanks!
Any reasons why you did apply this patch to your tree? It is a spi-nor fix, I would have expected it to go through mtd.
It's been sitting around for a month, I assumed it was lost, so I picked it up.

Sorry if it took too long, the merge window also happened during that time, we are collecting patches now that 6.3-rc1 has been released. Next time don't hesitate to ping first ;-)

Quoted 2 lines

I can revert it if you don't want me to take it for 6.3-final through my driver core tree.

I'll let spi-nor maintainers decide what they prefer.

Thanks,
Miquèl

Tudor Ambarus

2023-03-17 08:44 UTC #7 lore raw

Quoted 45 lines · Miquel Raynal

Hi Greg,
[email protected] wrote on Mon, 6 Mar 2023 10:13:33 +0100:
Quoted 41 lines · Miquel Raynal
Quoted 39 lines
Hi Greg,
[email protected] wrote on Mon, 6 Mar 2023 07:52:38 +0100:
Quoted 31 lines · Michael Walle
Quoted 20 lines
Am 2023-02-08 17:02, schrieb Greg Kroah-Hartman:
Quoted 5 lines
When calling debugfs_lookup() the result must have dput() called on it, otherwise the memory will leak over time. To solve this, remove the lookup and create the directory on the first device found, and then remove it when the module is unloaded.
Quoted 8 lines
Cc Tudor Ambarus <[email protected]>
Cc Pratyush Yadav <[email protected]>
Cc Michael Walle <[email protected]>
Cc Miquel Raynal <[email protected]>
Cc Richard Weinberger <[email protected]>
Cc Vignesh Raghavendra <[email protected]>
Cc [email protected]
Signed-off-by Greg Kroah-Hartman <[email protected]>
Reviewed-by Michael Walle <[email protected]>
one nit below I didn't notice earlier, no need to send a new patch version just for that.
Quoted 9 lines
..
drivers/mtd/spi-nor/debugfs.c void spi_nor_debugfs_register(struct spi_nor *nor) · line 254
Quoted 4 lines
drivers/mtd/spi-nor/debugfs.c +4 -0
+void spi_nor_debugfs_shutdown(void)
+{
+	if (rootdir)
+		debugfs_remove(rootdir);    
debugfs_remove() already has a check for NULL.
Quoted 1 line
Ah, good catch, I merged this in when I applied it to my tree, thanks!
Any reasons why you did apply this patch to your tree? It is a spi-nor fix, I would have expected it to go through mtd.

Quoted 10 lines

Quoted 2 lines
It's been sitting around for a month, I assumed it was lost, so I picked it up.
Sorry if it took too long, the merge window also happened during that time, we are collecting patches now that 6.3-rc1 has been released. Next time don't hesitate to ping first ;-)
Quoted 2 lines
I can revert it if you don't want me to take it for 6.3-final through my driver core tree.

It's fine, I don't expect other changes on these chunks of code. Anyway, if there will be merge conflicts I'll solve them locally by merging the -rc containing the fix.

BTW, you forgot to cc stable and add a fixes tag :).

Quoted 2 lines

I'll let spi-nor maintainers decide what they prefer.

If it'll be a next time, a ping would be better.

Cheers,
ta

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/