Skip to contentSkip to conversation

Memory leak in tpm_ascii_bios_measurements_open() fix.

PATCH Reiner Sailer 3 participants 4 messages Jul 20, 2007 TPM device driver

lore
Share & tools
lore
#0 lore raw

Jesper,

thank you very much for finding this error and for posting a patch proposal. Since Kylene is not online, I am responding. Please see my inlines and an alternative patch proposal below.

Jesper Juhl <[email protected]> wrote on 07/18/2007 07:11:54 PM:

Quoted 22 lines

Ehlo,

Coverity found a memory leak in tpm_ascii_bios_measurements_open().

If "read_log(log)" fails, then we may leak 'log' and 'log->bios_event_log'.

This patch should fix it, but please double check it. I don't know this code very well and the patch has only been compile tested.

---

 drivers/char/tpm/tpm_bios.c |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)
drivers/char/tpm/tpm_bios.c +0 -0
index 4eba32b..4b26ce4 100644

drivers/char/tpm/tpm_bios.c static int · line 430

Quoted 5 lines
drivers/char/tpm/tpm_bios.c +1 -1
@@ -427,7 +427,7 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       return -ENOMEM;
 >     if ((err = read_log(log)))
-      return err;
+      goto out_free;

log->bios_event_log should not be pointing to allocated memory here (seems cleaner if read_log cleans its allocated memory in the error case) ---> just free log

Quoted 13 lines
drivers/char/tpm/tpm_bios.c +1 -2
    /* now register seq file */
    err = seq_open(file, &tpm_ascii_b_measurments_seqops);
@@ -435,10 +435,15 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       seq = file->private_data;
       seq->private = log;
    } else {
-      kfree(log->bios_event_log);
-      kfree(log);
+      goto out_free;

drivers/char/tpm/tpm_bios.c static int · line 449

Quoted 10 lines
drivers/char/tpm/tpm_bios.c +6 -0
    }
+
+out:
    return err;
+out_free:
+   kfree(log->bios_event_log);
+   kfree(log);
+   goto out;
 }
 >  const struct file_operations tpm_ascii_bios_measurements_ops = {

The following patch should be sufficient to fix the problem you discovered:
---
drivers/char/tpm/tpm_bios.c | 5 +++--

1 file changed, 3 insertions(+), 2 deletions(-)
drivers/char/tpm/tpm_bios.c +3 -2
Index: linux-2.6.22-rc7/drivers/char/tpm/tpm_bios.c
===================================================================
@@ -426,9 +426,10 @@ static int tpm_ascii_bios_measurements_o
    if (!log)
        return -ENOMEM;
 
-    if ((err = read_log(log)))
+    if ((err = read_log(log))) {
+        kfree(log);
        return err;
-
+    }
    /* now register seq file */
    err = seq_open(file, &tpm_ascii_b_measurments_seqops);
    if (!err) {

Do you agree?

Greetings
Reiner

#1 lore raw
Quoted 54 lines · Reiner Sailer

Jesper Juhl <[email protected]> wrote on 07/18/2007 07:11:54 PM:

Quoted 30 lines

+++ b/drivers/char/tpm/tpm_bios.c

(quoted diff) +1 -1
@@ -427,7 +427,7 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       return -ENOMEM;
 
    if ((err = read_log(log)))
-      return err;
+      goto out_free;
 
    /* now register seq file */
    err = seq_open(file, &tpm_ascii_b_measurments_seqops);
(quoted diff) +7 -2
@@ -435,10 +435,15 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       seq = file->private_data;
       seq->private = log;
    } else {
-      kfree(log->bios_event_log);
-      kfree(log);
+      goto out_free;
    }
+
+out:
    return err;
+out_free:
+   kfree(log->bios_event_log);
+   kfree(log);
+   goto out;
 }
 
Quoted 3 lines

const struct file_operations tpm_ascii_bios_measurements_ops = {

Index: linux-2.6.22-rc7/drivers/char/tpm/tpm_bios.c
===================================================================

drivers/char/tpm/tpm_bios.c +3 -2
@@ -426,9 +426,10 @@ static int tpm_ascii_bios_measurements_o
     if (!log)
         return -ENOMEM;
 
-    if ((err = read_log(log)))
+    if ((err = read_log(log))) {
+        kfree(log);
         return err;
-
+    }
     /* now register seq file */
     err = seq_open(file, &tpm_ascii_b_measurments_seqops);
     if (!err) {
Quoted 1 line

Do you agree?

Jesper's patch makes the code conforming to the kernel coding style (kfree()'s in common error paths, single "return" call, etc). Not a big issue, just a consistency thing, so feel free to take your pick between the two alternatives.

BTW please also fix an exactly similar memory leak 30 lines later in the same file -- in tpm_binary_bios_measurements_open() -- while you're at it.

Thanks,
Satyam

#2 lore raw
Quoted 25 lines · Satyam Sharma
Quoted 24 lines · Reiner Sailer

Jesper Juhl <[email protected]> wrote on 07/18/2007 07:11:54 PM:

Quoted 22 lines

+++ b/drivers/char/tpm/tpm_bios.c

(quoted diff) +1 -1
@@ -427,7 +427,7 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       return -ENOMEM;
 
    if ((err = read_log(log)))
-      return err;
+      goto out_free;
 
    /* now register seq file */
    err = seq_open(file, &tpm_ascii_b_measurments_seqops);
(quoted diff) +2 -2
@@ -435,10 +435,15 @@ static int
 tpm_ascii_bios_measurements_open(struct inode *inode,
       seq = file->private_data;
       seq->private = log;
    } else {
-      kfree(log->bios_event_log);
-      kfree(log);
+      goto out_free;
    }
+
Quoted 8 lines
Quoted 8 lines

Review static int · line 449

Quoted 8 lines
(quoted diff) +5 -0
+out:
    return err;
+out_free:
+   kfree(log->bios_event_log);
+   kfree(log);
+   goto out;
 }
 

===>
out_free:
kfree(log->bios_event_log);
kfree(log);
out:
return err;

could kill one extra goto.

YH

#3 lore raw
Quoted 7 lines · Yinghai Lu

out_free:

   kfree(log->bios_event_log);
   kfree(log);
out:
   return err;

could kill one extra goto.

Such constructs disallow the success path from falling through to the same "return err;" (with err = 0 for success obviously)