Skip to contentSkip to conversation

libbpf: synchronize access to print function pointer

PATCH Applied JP Kobryn 4 participants 5 messages Mar 25, 2023 – Mar 27, 2023

Subsystems (2)
lore
Share & tools
lore

JP Kobryn

author
#0 lore raw

PATCH 1/1 libbpf: synchronize access to print function pointer

This patch prevents races on the print function pointer, allowing the libbpf_set_print() function to become thread safe.

---
 tools/lib/bpf/libbpf.c | 9 ++++++---
 tools/lib/bpf/libbpf.h | 3 +++
 2 files changed, 9 insertions(+), 3 deletions(-)
tools/lib/bpf/libbpf.c +6 -3
index f6a071db5c6e..15737d7b5a28 100644
@@ -216,9 +216,10 @@ static libbpf_print_fn_t __libbpf_pr = __base_pr;
 
 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
 {
-	libbpf_print_fn_t old_print_fn = __libbpf_pr;
+	libbpf_print_fn_t old_print_fn;
+
+	old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
 
-	__libbpf_pr = fn;
 	return old_print_fn;
 }
 
@@ -227,8 +228,10 @@ void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 {
 	va_list args;
 	int old_errno;
+	libbpf_print_fn_t print_fn;
 
-	if (!__libbpf_pr)
+	print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
+	if (!print_fn)
 		return;
 
 	old_errno = errno;
tools/lib/bpf/libbpf.h +3 -0
index 1615e55e2e79..4478809ff9ca 100644
@@ -99,6 +99,9 @@ typedef int (*libbpf_print_fn_t)(enum libbpf_print_level level,
 /**
  * @brief **libbpf_set_print()** sets user-provided log callback function to
  * be used for libbpf warnings and informational messages.
+ *
+ * This function is thread safe.
+ *
  * @param fn The log print function. If NULL, libbpf won't print anything.
  * @return Pointer to old print function.
  */
Signature · 1 line
2.39.2
#1 lore raw
Quoted 2 lines · JP Kobryn

This patch prevents races on the print function pointer, allowing the libbpf_set_print() function to become thread safe.

Why does it have to be thread-safe? The rest of the APIs aren't, so why can't use solve it on your side by wrapping those calls with a mutex?

(is there some context I'm missing?)

Quoted 55 lines
---
  tools/lib/bpf/libbpf.c | 9 ++++++---
  tools/lib/bpf/libbpf.h | 3 +++
  2 files changed, 9 insertions(+), 3 deletions(-)
tools/lib/bpf/libbpf.c +6 -3
index f6a071db5c6e..15737d7b5a28 100644
@@ -216,9 +216,10 @@ static libbpf_print_fn_t __libbpf_pr = __base_pr;
 
  libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  {
-	libbpf_print_fn_t old_print_fn = __libbpf_pr;
+	libbpf_print_fn_t old_print_fn;
+
+	old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
 
-	__libbpf_pr = fn;
  	return old_print_fn;
  }
 
@@ -227,8 +228,10 @@ void libbpf_print(enum libbpf_print_level level,  
 const char *format, ...)
  {
  	va_list args;
  	int old_errno;
+	libbpf_print_fn_t print_fn;
 
-	if (!__libbpf_pr)
+	print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
+	if (!print_fn)
  		return;
 
  	old_errno = errno;
tools/lib/bpf/libbpf.h +0 -0
index 1615e55e2e79..4478809ff9ca 100644
@@ -99,6 +99,9 @@ typedef int (*libbpf_print_fn_t)(enum  
 libbpf_print_level level,
  /**
   * @brief **libbpf_set_print()** sets user-provided log callback  

function to
* be used for libbpf warnings and informational messages.
+ *
+ * This function is thread safe.
+ *
* @param fn The log print function. If NULL, libbpf won't print
anything.
* @return Pointer to old print function.
*/
--
2.39.2

#2 lore raw
Quoted 6 lines · Stanislav Fomichev
Quoted 2 lines · JP Kobryn

This patch prevents races on the print function pointer, allowing the libbpf_set_print() function to become thread safe.

Why does it have to be thread-safe? The rest of the APIs aren't, so

It doesn't have to, but if it can be made thread-safe trivially, it probably should be. Rust users are especially sensitive to this (see discussion in [0] for example).

Generally speaking, yep, libbpf APIs are not thread-safe by default (we don't do explicit locking anywhere inside libbpf), but there are a bunch of APIs that are inherently thread-safe as they are stateless (like feature probing, string conversion APIs, I suspect upon inspection we'll see that all bpf_program__attach_*() APIs are also probably thread-safe already). So we should start marking them as such to avoid confusion and uncertainty for users. I'd also like to document somewhere that two independent bpf_objects can be used safely on two separate threads, because whatever state they are sharing (like feature detection cache) is designed in such a way to be thread-safe and shareable with no locking.

[0] https://github.com/libbpf/libbpf-rs/pull/374#issuecomment-1462565778

Quoted 2 lines

why can't use solve it on your side by wrapping those calls with a mutex?

It would be very unfortunate to wrap libbpf_set_print and *all other libbpf API* in mutex.

Quoted 59 lines

(is there some context I'm missing?)

Quoted 55 lines
---
  tools/lib/bpf/libbpf.c | 9 ++++++---
  tools/lib/bpf/libbpf.h | 3 +++
  2 files changed, 9 insertions(+), 3 deletions(-)
tools/lib/bpf/libbpf.c +6 -3
index f6a071db5c6e..15737d7b5a28 100644
@@ -216,9 +216,10 @@ static libbpf_print_fn_t __libbpf_pr = __base_pr;
 
  libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
  {
-     libbpf_print_fn_t old_print_fn = __libbpf_pr;
+     libbpf_print_fn_t old_print_fn;
+
+     old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
 
-     __libbpf_pr = fn;
      return old_print_fn;
  }
 
@@ -227,8 +228,10 @@ void libbpf_print(enum libbpf_print_level level,
 const char *format, ...)
  {
      va_list args;
      int old_errno;
+     libbpf_print_fn_t print_fn;
 
-     if (!__libbpf_pr)
+     print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
+     if (!print_fn)
              return;
 
      old_errno = errno;
tools/lib/bpf/libbpf.h +0 -0
index 1615e55e2e79..4478809ff9ca 100644
@@ -99,6 +99,9 @@ typedef int (*libbpf_print_fn_t)(enum
 libbpf_print_level level,
  /**
   * @brief **libbpf_set_print()** sets user-provided log callback

function to
* be used for libbpf warnings and informational messages.
+ *
+ * This function is thread safe.
+ *
* @param fn The log print function. If NULL, libbpf won't print
anything.
* @return Pointer to old print function.
*/
--
2.39.2

#3 lore raw
In reply to JP Kobryn
Quoted 6 lines · JP Kobryn

This patch prevents races on the print function pointer, allowing the libbpf_set_print() function to become thread safe.

---

I reworked the patch subject to "libbpf: Ensure print callback usage is thread-safe", it felt more on point. Also changed to "thread-safe", which seems to be a proper spelling for this term.

Applied to bpf-next, thanks!

Quoted 38 lines
 tools/lib/bpf/libbpf.c | 9 ++++++---
 tools/lib/bpf/libbpf.h | 3 +++
 2 files changed, 9 insertions(+), 3 deletions(-)
tools/lib/bpf/libbpf.c +6 -3
index f6a071db5c6e..15737d7b5a28 100644
@@ -216,9 +216,10 @@ static libbpf_print_fn_t __libbpf_pr = __base_pr;
 
 libbpf_print_fn_t libbpf_set_print(libbpf_print_fn_t fn)
 {
-       libbpf_print_fn_t old_print_fn = __libbpf_pr;
+       libbpf_print_fn_t old_print_fn;
+
+       old_print_fn = __atomic_exchange_n(&__libbpf_pr, fn, __ATOMIC_RELAXED);
 
-       __libbpf_pr = fn;
        return old_print_fn;
 }
 
@@ -227,8 +228,10 @@ void libbpf_print(enum libbpf_print_level level, const char *format, ...)
 {
        va_list args;
        int old_errno;
+       libbpf_print_fn_t print_fn;
 
-       if (!__libbpf_pr)
+       print_fn = __atomic_load_n(&__libbpf_pr, __ATOMIC_RELAXED);
+       if (!print_fn)
                return;
 
        old_errno = errno;
tools/lib/bpf/libbpf.h +0 -0
index 1615e55e2e79..4478809ff9ca 100644

tools/lib/bpf/libbpf.h typedef int (*libbpf_print_fn_t)(enum libbpf_print_level level, · line 104

Quoted 6 lines
tools/lib/bpf/libbpf.h +3 -0
@@ -99,6 +99,9 @@ typedef int (*libbpf_print_fn_t)(enum libbpf_print_level level,
 /**
  * @brief **libbpf_set_print()** sets user-provided log callback function to
  * be used for libbpf warnings and informational messages.
+ *
+ * This function is thread safe.
+ *

I moved this part to after @return spec, where details about functions are normally put.

Quoted 6 lines
tools/lib/bpf/libbpf.h +0 -1
  * @param fn The log print function. If NULL, libbpf won't print anything.
  * @return Pointer to old print function.
  */
--

2.39.2

#4 lore raw
In reply to JP Kobryn
This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Hello:

This patch was applied to bpf/bpf-next.git (master) by Andrii Nakryiko <[email protected]>:

Quoted 8 lines · On Fri, 24 Mar 2023 18:08:45 -0700 you wrote:

This patch prevents races on the print function pointer, allowing the libbpf_set_print() function to become thread safe.

---
 tools/lib/bpf/libbpf.c | 9 ++++++---
 tools/lib/bpf/libbpf.h | 3 +++
 2 files changed, 9 insertions(+), 3 deletions(-)

Here is the summary with links:
- [bpf-next] libbpf: synchronize access to print function pointer

    https://git.kernel.org/bpf/bpf-next/c/f1cb927cdb62

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html