Re: [PATCH] perf session: Remap buf if there is no space for event

From: James Clark <james.clark@arm.com>
To: Denis Nikitin <denik@chromium.org>,
	acme@kernel.org, linux-perf-users@vger.kernel.org
Cc: jolsa@kernel.org, alexey.budankov@linux.intel.com,
	alexander.shishkin@linux.intel.com, namhyung@kernel.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] perf session: Remap buf if there is no space for event
Date: Thu, 31 Mar 2022 15:20:00 +0100	[thread overview]
Message-ID: <77fa14e8-6630-c0e9-21fc-2603f7383f5f@arm.com> (raw)
In-Reply-To: <20220330031130.2152327-1-denik@chromium.org>

On 30/03/2022 04:11, Denis Nikitin wrote:
> If a perf event doesn't fit into remaining buffer space return NULL to
> remap buf and fetch the event again.
> Keep the logic to error out on inadequate input from fuzzing.
> 
> This fixes perf failing on ChromeOS (with 32b userspace):
> 
>   $ perf report -v -i perf.data
>   ...
>   prefetch_event: head=0x1fffff8 event->header_size=0x30, mmap_size=0x2000000: fuzzed or compressed perf.data?
>   Error:
>   failed to process sample
> 
> Fixes: 57fc032ad643 ("perf session: Avoid infinite loop when seeing invalid header.size")
> Signed-off-by: Denis Nikitin <denik@chromium.org>
> ---
>  tools/perf/util/session.c | 15 ++++++++++++---
>  1 file changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
> index 3b8dfe603e50..45a30040ec8d 100644
> --- a/tools/perf/util/session.c
> +++ b/tools/perf/util/session.c
> @@ -2095,6 +2095,7 @@ prefetch_event(char *buf, u64 head, size_t mmap_size,
>  	       bool needs_swap, union perf_event *error)
>  {
>  	union perf_event *event;
> +	u16 event_size;
>  
>  	/*
>  	 * Ensure we have enough space remaining to read
> @@ -2107,15 +2108,23 @@ prefetch_event(char *buf, u64 head, size_t mmap_size,
>  	if (needs_swap)
>  		perf_event_header__bswap(&event->header);
>  
> -	if (head + event->header.size <= mmap_size)
> +	event_size = event->header.size;
> +	if (head + event_size <= mmap_size)
>  		return event;
>  
>  	/* We're not fetching the event so swap back again */
>  	if (needs_swap)
>  		perf_event_header__bswap(&event->header);
>  
> -	pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
> -		 " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
> +	/* Check if the event fits into the next mmapped buf. */
> +	if (event_size <= mmap_size - head % page_size) {
> +		/* Remap buf and fetch again. */
> +		return NULL;

Hi Denis,

I tested this and it does fix the issue with a 32bit build. One concern is that the calculation to
see if it will fit in the next map is dependent on the implementation of reader__mmap(). I think it
would be possible for that to change slightly and then you could still get an infinite loop.

But I can't really see a better way to do it, and it's unlikely for reader__mmap() to be modified
to map data in a way to waste part of the buffer so it's probably fine.

Maybe you could extract a function to calculate where the new offset would be in the buffer and share
it between here and reader__mmap(). That would also make it more obvious what the 'head % page_size'
bit is for.

Either way:

Reviewed-by: James Clark <james.clark@arm.com>

> +	}
> +
> +	/* Invalid input. Event size should never exceed mmap_size. */
> +	pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
> +		 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
>  
>  	return error;
>  }