threads / patch / 62003

patchbundle-uri: plug leak in unbundle_from_file()

Subject: [PATCH] bundle-uri: plug leak in unbundle_from_file()

## tl;dr

9 messages between Aug 26, 2024 and Oct 10, 2024. Diffs are folded; open one to read it.

replies: 8people: 3as markdown or json

Toon Claes· Aug 26, 2024, 08:30 UTC · lore

When the function returns early, the variable bundle_ref is not released through strbuf_release().

Fix this leak. And while at it, remove assignments in the conditions of the "if" statements as suggested in the CodingGuidelines.

Signed-off-by: Toon Claes <toon@iotcl.com>
---
 bundle-uri.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)
Show changes to bundle-uri.c +11 −6
diff --git a/bundle-uri.c b/bundle-uri.c
index 1e0ee156ba..eb49cba182 100644
--- a/bundle-uri.c
+++ b/bundle-uri.c
@@ -367,17 +367,21 @@ static int unbundle_from_file(struct repository *r, const char *file)
 	struct strbuf bundle_ref = STRBUF_INIT;
 	size_t bundle_prefix_len;

-	if ((bundle_fd = read_bundle_header(file, &header)) < 0)
-		return 1;
+	bundle_fd = read_bundle_header(file, &header);
+	if (bundle_fd < 0) {
+		result = 1;
+		goto cleanup;
+	}

 	/*
 	 * Skip the reachability walk here, since we will be adding
 	 * a reachable ref pointing to the new tips, which will reach
 	 * the prerequisite commits.
 	 */
-	if ((result = unbundle(r, &header, bundle_fd, NULL,
-			       VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0))))
-		return 1;
+	result = unbundle(r, &header, bundle_fd, NULL,
+			  VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0));
+	if (result)
+		goto cleanup;

 	/*
 	 * Convert all refs/heads/ from the bundle into refs/bundles/
@@ -406,6 +410,8 @@ static int unbundle_from_file(struct repository *r, const char *file)
 				0, UPDATE_REFS_MSG_ON_ERR);
 	}

+cleanup:
+	strbuf_release(&bundle_ref);
 	bundle_header_release(&header);
 	return result;
 }
--
2.46.0
Patrick Steinhardt· Aug 26, 2024, 09:51 UTC · re: Toon Claes · lore

Re: [PATCH] bundle-uri: plug leak in unbundle_from_file()

On Mon, Aug 26, 2024 at 10:30:52AM +0200, Toon Claes wrote:
Show 39 quoted lines
> When the function returns early, the variable bundle_ref is not released
> through strbuf_release().
> 
> Fix this leak. And while at it, remove assignments in the conditions of
> the "if" statements as suggested in the CodingGuidelines.
> 
> Signed-off-by: Toon Claes <toon@iotcl.com>
> ---
>  bundle-uri.c | 16 +++++++++++-----
>  1 file changed, 11 insertions(+), 5 deletions(-)
> 
> diff --git a/bundle-uri.c b/bundle-uri.c
> index 1e0ee156ba..eb49cba182 100644
> --- a/bundle-uri.c
> +++ b/bundle-uri.c
> @@ -367,17 +367,21 @@ static int unbundle_from_file(struct repository *r, const char *file)
>  	struct strbuf bundle_ref = STRBUF_INIT;
>  	size_t bundle_prefix_len;
> 
> -	if ((bundle_fd = read_bundle_header(file, &header)) < 0)
> -		return 1;
> +	bundle_fd = read_bundle_header(file, &header);
> +	if (bundle_fd < 0) {
> +		result = 1;
> +		goto cleanup;
> +	}
> 
>  	/*
>  	 * Skip the reachability walk here, since we will be adding
>  	 * a reachable ref pointing to the new tips, which will reach
>  	 * the prerequisite commits.
>  	 */
> -	if ((result = unbundle(r, &header, bundle_fd, NULL,
> -			       VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0))))
> -		return 1;
> +	result = unbundle(r, &header, bundle_fd, NULL,
> +			  VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0));
> +	if (result)
> +		goto cleanup;

This changes the returned error code from `1` to whatever `unbundle()` returns. Is this intentional? If so, the commit message should explain why this change is safe.

Other than that this looks good to me, and the fix does not conflict with any of my leak-plugging series.

Thanks!
Patrick
Junio C Hamano· Aug 26, 2024, 16:06 UTC · re: Patrick Steinhardt · lore

Re: [PATCH] bundle-uri: plug leak in unbundle_from_file()

Patrick Steinhardt <ps@pks.im> writes:
Show 18 quoted lines
> On Mon, Aug 26, 2024 at 10:30:52AM +0200, Toon Claes wrote:
>> When the function returns early, the variable bundle_ref is not released
>> through strbuf_release().
>> 
>> Fix this leak. And while at it, remove assignments in the conditions of
>> the "if" statements as suggested in the CodingGuidelines.
> ...
>> -	if ((result = unbundle(r, &header, bundle_fd, NULL,
>> -			       VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0))))
>> -		return 1;
>> +	result = unbundle(r, &header, bundle_fd, NULL,
>> +			  VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0));
>> +	if (result)
>> +		goto cleanup;
>
> This changes the returned error code from `1` to whatever `unbundle()`
> returns. Is this intentional? If so, the commit message should explain
> why this change is safe.
Thanks for reviewing carefully.

Both of two callers of unbundle_from_file() are used as the condition of an if() statement, so unbundle() that signals an error with -1 wouldn't be a problem, I would think.

It may not be a bad idea as a #leftoverbits item, after the dust settles, to clean up the calling convention in this file (may not be limited to the code path that reaches this function) to follow the usual "signal success with 0, failures are signalled with a negative value". Then we can just return the value we got from a failing read_bundle_header(), just the same way we return the value we got from a failing unbundle().

> Other than that this looks good to me, and the fix does not conflict
> with any of my leak-plugging series.
Yup.  Thanks, both.
Toon Claes· Oct 1, 2024, 18:58 UTC · re: Junio C Hamano · lore

Re: [PATCH] bundle-uri: plug leak in unbundle_from_file()

Junio C Hamano <gitster@pobox.com> writes:
Show 5 quoted lines
> Thanks for reviewing carefully.
>
> Both of two callers of unbundle_from_file() are used as the
> condition of an if() statement, so unbundle() that signals an error
> with -1 wouldn't be a problem, I would think.
Hi Junio,

I've noticed this patch wasn't picked up yet. Is there anything you want me to change and have me sent another version, or is it good to go in?

-- Toon

Junio C Hamano· Oct 1, 2024, 19:29 UTC · re: Toon Claes · lore

Re: [PATCH] bundle-uri: plug leak in unbundle_from_file()

Toon Claes <toon@iotcl.com> writes:
Show 12 quoted lines
> Junio C Hamano <gitster@pobox.com> writes:
>
>> Thanks for reviewing carefully.
>>
>> Both of two callers of unbundle_from_file() are used as the
>> condition of an if() statement, so unbundle() that signals an error
>> with -1 wouldn't be a problem, I would think.
>
> Hi Junio,
>
> I've noticed this patch wasn't picked up yet. Is there anything you want
> me to change and have me sent another version, or is it good to go in?

I am waiting for a reroll with an updated log message, i.e, what Patrick pointed out in his review. I only said "yeah, this looks safe", and never meant "it is so obvious there is no need for extra explanation in the log message".

Thanks.
Toon Claes· Oct 10, 2024, 09:15 UTC · re: Junio C Hamano · lore

Re: [PATCH] bundle-uri: plug leak in unbundle_from_file()

Junio C Hamano <gitster@pobox.com> writes:
> Both of two callers of unbundle_from_file() are used as the
> condition of an if() statement, so unbundle() that signals an error
> with -1 wouldn't be a problem, I would think.
Show 7 quoted lines
> It may not be a bad idea as a #leftoverbits item, after the dust
> settles, to clean up the calling convention in this file (may not be
> limited to the code path that reaches this function) to follow the
> usual "signal success with 0, failures are signalled with a negative
> value".  Then we can just return the value we got from a failing
> read_bundle_header(), just the same way we return the value we got
> from a failing unbundle().

I just submitted v2 which preserves the original return values as before. I'll leave changing any return value to a #leftoverbits series to addresses them together.

-- Toon

Toon Claes· Oct 10, 2024, 09:12 UTC · re: Toon Claes · lore

[PATCH v2] bundle-uri: plug leak in unbundle_from_file()

The function `unbundle_from_file()` has two memory leaks:
  - We do not release the `struct bundle_header header` when hitting
    errors because we return early without any cleanup.
  - We do not release the `struct strbuf bundle_ref` at all.

Plug these leaks by creating a common exit path where both of these variables are released.

While at it, refactor the code such that the variable assignments do not happen inside the conditional statement itself according to our coding style.

Signed-off-by: Toon Claes <toon@iotcl.com>
---
 bundle-uri.c | 18 +++++++++++++-----
 1 file changed, 13 insertions(+), 5 deletions(-)
Show changes to bundle-uri.c +13 −5
diff --git a/bundle-uri.c b/bundle-uri.c
index 4b1a2e2937..0df66e2872 100644
--- a/bundle-uri.c
+++ b/bundle-uri.c
@@ -368,17 +368,23 @@ static int unbundle_from_file(struct repository *r, const char *file)
 	struct strbuf bundle_ref = STRBUF_INIT;
 	size_t bundle_prefix_len;
 
-	if ((bundle_fd = read_bundle_header(file, &header)) < 0)
-		return 1;
+	bundle_fd = read_bundle_header(file, &header);
+	if (bundle_fd < 0) {
+		result = 1;
+		goto cleanup;
+	}
 
 	/*
 	 * Skip the reachability walk here, since we will be adding
 	 * a reachable ref pointing to the new tips, which will reach
 	 * the prerequisite commits.
 	 */
-	if ((result = unbundle(r, &header, bundle_fd, NULL,
-			       VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0))))
-		return 1;
+	result = unbundle(r, &header, bundle_fd, NULL,
+			  VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0));
+	if (result) {
+		result = 1;
+		goto cleanup;
+	}
 
 	/*
 	 * Convert all refs/heads/ from the bundle into refs/bundles/
@@ -407,6 +413,8 @@ static int unbundle_from_file(struct repository *r, const char *file)
 				0, UPDATE_REFS_MSG_ON_ERR);
 	}
 
+cleanup:
+	strbuf_release(&bundle_ref);
 	bundle_header_release(&header);
 	return result;
 }

Range-diff against v1:
1:  f30f393e05 ! 1:  19714d860c bundle-uri: plug leak in unbundle_from_file()
    @@ Metadata
      ## Commit message ##
         bundle-uri: plug leak in unbundle_from_file()
     
    -    When the function returns early, the variable bundle_ref is not released
    -    through strbuf_release().
    +    The function `unbundle_from_file()` has two memory leaks:
     
    -    Fix this leak. And while at it, remove assignments in the conditions of
    -    the "if" statements as suggested in the CodingGuidelines.
    +      - We do not release the `struct bundle_header header` when hitting
    +        errors because we return early without any cleanup.
    +
    +      - We do not release the `struct strbuf bundle_ref` at all.
    +
    +    Plug these leaks by creating a common exit path where both of these
    +    variables are released.
    +
    +    While at it, refactor the code such that the variable assignments do not
    +    happen inside the conditional statement itself according to our coding
    +    style.
     
         Signed-off-by: Toon Claes <toon@iotcl.com>
     
    @@ bundle-uri.c: static int unbundle_from_file(struct repository *r, const char *fi
     -		return 1;
     +	result = unbundle(r, &header, bundle_fd, NULL,
     +			  VERIFY_BUNDLE_QUIET | (fetch_pack_fsck_objects() ? VERIFY_BUNDLE_FSCK : 0));
    -+	if (result)
    ++	if (result) {
    ++		result = 1;
     +		goto cleanup;
    ++	}
      
      	/*
      	 * Convert all refs/heads/ from the bundle into refs/bundles/
-- 
2.46.0.46.g406f326d27
Patrick Steinhardt· Oct 10, 2024, 12:55 UTC · re: Toon Claes · lore

Re: [PATCH v2] bundle-uri: plug leak in unbundle_from_file()

On Thu, Oct 10, 2024 at 11:12:49AM +0200, Toon Claes wrote:
Show 13 quoted lines
> The function `unbundle_from_file()` has two memory leaks:
> 
>   - We do not release the `struct bundle_header header` when hitting
>     errors because we return early without any cleanup.
> 
>   - We do not release the `struct strbuf bundle_ref` at all.
> 
> Plug these leaks by creating a common exit path where both of these
> variables are released.
> 
> While at it, refactor the code such that the variable assignments do not
> happen inside the conditional statement itself according to our coding
> style.

Thanks, this version looks good to me. We now avoid any discussion around the changed error code completely, and the commit message seems reasonable to me.

Thanks!
Patrick
Junio C Hamano· Oct 10, 2024, 18:47 UTC · re: Patrick Steinhardt · lore

Re: [PATCH v2] bundle-uri: plug leak in unbundle_from_file()

Patrick Steinhardt <ps@pks.im> writes:
Show 22 quoted lines
> On Thu, Oct 10, 2024 at 11:12:49AM +0200, Toon Claes wrote:
>> The function `unbundle_from_file()` has two memory leaks:
>> 
>>   - We do not release the `struct bundle_header header` when hitting
>>     errors because we return early without any cleanup.
>> 
>>   - We do not release the `struct strbuf bundle_ref` at all.
>> 
>> Plug these leaks by creating a common exit path where both of these
>> variables are released.
>> 
>> While at it, refactor the code such that the variable assignments do not
>> happen inside the conditional statement itself according to our coding
>> style.
>
> Thanks, this version looks good to me. We now avoid any discussion
> around the changed error code completely, and the commit message seems
> reasonable to me.
>
> Thanks!
>
> Patrick
Thanks, both of you.
Will queue.

← back to recent threads