Skip to contentSkip to conversation

android: binder: print error message on failure of creating proc file

PATCH In review Leesoo Ahn 3 participants 8 messages Jul 12, 2024 Android drivers

lore
Share & tools
lore

Leesoo Ahn

author
#0 lore raw

PATCH 1/1 android: binder: print error message on failure of creating proc file

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

---
 drivers/android/binder.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/android/binder.c +11 -1
index b21a7b246a0d..eb0fd1443d69 100644
@@ -5673,6 +5673,7 @@ static int binder_open(struct inode *nodp, struct file *filp)
 
 	if (binder_debugfs_dir_entry_proc && !existing_pid) {
 		char strbuf[11];
+		struct dentry *debugfs_entry;
 
 		snprintf(strbuf, sizeof(strbuf), "%u", proc->pid);
 		/*
@@ -5681,10 +5682,19 @@ static int binder_open(struct inode *nodp, struct file *filp)
 		 * The printing code will anyway print all contexts for a given
 		 * PID so this is not a problem.
 		 */
-		proc->debugfs_entry = debugfs_create_file(strbuf, 0444,
+		debugfs_entry = debugfs_create_file(strbuf, 0444,
 			binder_debugfs_dir_entry_proc,
 			(void *)(unsigned long)proc->pid,
 			&proc_fops);
+		if (!IS_ERR(debugfs_entry)) {
+			proc->debugfs_entry = debugfs_entry;
+		} else {
+			int error;
+
+			error = PTR_ERR(debugfs_entry);
+			pr_warn("Unable to create file %s in debugfs (error %d)\n",
+				strbuf, error);
+		}
 	}
 
 	if (binder_binderfs_dir_entry_proc && !existing_pid) {
Signature · 1 line
2.34.1
maintainer
#1 lore raw
Quoted 6 lines · Leesoo Ahn

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

---

What are you trying to fix? My understanding is that users of the debugfs API can safely ignore any errors and move on. IMO it doesn't make sense to add this without a real reason.

--
Carlos Llamas

#2 lore raw
Quoted 11 lines · Carlos Llamas
Quoted 6 lines · Leesoo Ahn

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

---

What are you trying to fix? My understanding is that users of the debugfs API can safely ignore any errors and move on. IMO it doesn't make sense to add this without a real reason.

Agreed, no one should care if debugfs works or not.

thanks,

greg k-h

author
#3 lore raw
In reply to Carlos Llamas

2024년 7월 12일 (금) 오후 1:01, Carlos Llamas <[email protected]>님이 작성:

Quoted 6 lines · Leesoo Ahn
Quoted 4 lines

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

Quoted 6 lines
Quoted 2 lines

---

What are you trying to fix? My understanding is that users of the debugfs API can safely ignore any errors and move on. IMO it doesn't make sense to add this without a real reason.

What I was trying to say, users would predict that a file under debugfs will be created while they are opening a binder device. But if it failed for some reason without any debug message, they would get confused that the file doesn't exist and have no clue what happened without a message.

BR,
Leesoo

(send this again by missing CCs)

#4 lore raw
Quoted 14 lines · Leesoo Ahn

2024년 7월 12일 (금) 오후 1:01, Carlos Llamas <[email protected]>님이 작성:

Quoted 6 lines · Leesoo Ahn
Quoted 4 lines

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

Quoted 6 lines
Quoted 2 lines

---

What are you trying to fix? My understanding is that users of the debugfs API can safely ignore any errors and move on. IMO it doesn't make sense to add this without a real reason.

Quoted 5 lines

What I was trying to say, users would predict that a file under debugfs will be created while they are opening a binder device. But if it failed for some reason without any debug message, they would get confused that the file doesn't exist and have no clue what happened without a message.

And that's fine, again, the kernel does not care if debugfs is working or not. It's just a debugging help, it does not affect the normal operation of a system at all, and as such, userspace can't rely on it being present for any functionality other than debugging issues that might happen at times.

thanks,

greg k-h

author
#5 lore raw

2024년 7월 12일 (금) 오후 4:19, Greg Kroah-Hartman <[email protected]>님이 작성:

Quoted 22 lines · Leesoo Ahn
Quoted 14 lines

2024년 7월 12일 (금) 오후 1:01, Carlos Llamas <[email protected]>님이 작성:

Quoted 6 lines · Leesoo Ahn
Quoted 4 lines

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

Quoted 6 lines
Quoted 2 lines

---

What are you trying to fix? My understanding is that users of the debugfs API can safely ignore any errors and move on. IMO it doesn't make sense to add this without a real reason.

Quoted 5 lines

What I was trying to say, users would predict that a file under debugfs will be created while they are opening a binder device. But if it failed for some reason without any debug message, they would get confused that the file doesn't exist and have no clue what happened without a message.

Quoted 5 lines

And that's fine, again, the kernel does not care if debugfs is working or not. It's just a debugging help, it does not affect the normal operation of a system at all, and as such, userspace can't rely on it being present for any functionality other than debugging issues that might happen at times.

Thank you for explaining in detail.
I hadn't thought about it from that perspective.

Let me close this issue.

BR,
Leesoo

#6 lore raw
In reply to Leesoo Ahn
Quoted 28 lines · Leesoo Ahn

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

---
 drivers/android/binder.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/android/binder.c +2 -1
index b21a7b246a0d..eb0fd1443d69 100644
@@ -5673,6 +5673,7 @@ static int binder_open(struct inode *nodp, struct file *filp)
 
 	if (binder_debugfs_dir_entry_proc && !existing_pid) {
 		char strbuf[11];
+		struct dentry *debugfs_entry;
 
 		snprintf(strbuf, sizeof(strbuf), "%u", proc->pid);
 		/*
@@ -5681,10 +5682,19 @@ static int binder_open(struct inode *nodp, struct file *filp)
 		 * The printing code will anyway print all contexts for a given
 		 * PID so this is not a problem.
 		 */
-		proc->debugfs_entry = debugfs_create_file(strbuf, 0444,
+		debugfs_entry = debugfs_create_file(strbuf, 0444,
 			binder_debugfs_dir_entry_proc,

drivers/android/binder.c static int binder_open(struct inode *nodp, struct file *filp) · line 5696

Quoted 10 lines
drivers/android/binder.c +8 -0
 			(void *)(unsigned long)proc->pid,
 			&proc_fops);
+		if (!IS_ERR(debugfs_entry)) {
+			proc->debugfs_entry = debugfs_entry;
+		} else {
+			int error;
+
+			error = PTR_ERR(debugfs_entry);
+			pr_warn("Unable to create file %s in debugfs (error %d)\n",
+				strbuf, error);

Even if we wanted to warn about this (hint, you don't, see previous response), this way to check is incorrect and will fail if debugfs is not enabled, which you don't want to have happen.

So I'm guessing you did not test this with that config option disabled?

thanks,

greg k-h

author
#7 lore raw

2024년 7월 12일 (금) 오후 2:40, Greg Kroah-Hartman <[email protected]>님이 작성:

Quoted 45 lines · Leesoo Ahn
Quoted 28 lines

It better prints out an error message to give more information if calling debugfs_create_file() is failure and the return value has an error code.

---
 drivers/android/binder.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
drivers/android/binder.c +2 -1
index b21a7b246a0d..eb0fd1443d69 100644
@@ -5673,6 +5673,7 @@ static int binder_open(struct inode *nodp, struct file *filp)
 
      if (binder_debugfs_dir_entry_proc && !existing_pid) {
              char strbuf[11];
+             struct dentry *debugfs_entry;
 
              snprintf(strbuf, sizeof(strbuf), "%u", proc->pid);
              /*
@@ -5681,10 +5682,19 @@ static int binder_open(struct inode *nodp, struct file *filp)
               * The printing code will anyway print all contexts for a given
               * PID so this is not a problem.
               */
-             proc->debugfs_entry = debugfs_create_file(strbuf, 0444,
+             debugfs_entry = debugfs_create_file(strbuf, 0444,
                      binder_debugfs_dir_entry_proc,

drivers/android/binder.c static int binder_open(struct inode *nodp, struct file *filp) · line 5696

Quoted 10 lines
drivers/android/binder.c +8 -0
                      (void *)(unsigned long)proc->pid,
                      &proc_fops);
+             if (!IS_ERR(debugfs_entry)) {
+                     proc->debugfs_entry = debugfs_entry;
+             } else {
+                     int error;
+
+                     error = PTR_ERR(debugfs_entry);
+                     pr_warn("Unable to create file %s in debugfs (error %d)\n",
+                             strbuf, error);

Even if we wanted to warn about this (hint, you don't, see previous response), this way to check is incorrect and will fail if debugfs is not enabled, which you don't want to have happen.

Quoted 1 line

So I'm guessing you did not test this with that config option disabled?

Oh, I haven't thought about this and just figured out it would work weird. Thank you for mentioning it.

Quoted 4 lines

thanks,

greg k-h