The read stream callback apw_read_stream_next_block() advances p->pos
through the block_info array. When processing the last block, it
increments p->pos to prewarm_stop_idx before returning. The callback
itself is safe because it checks bounds before accessing the array.
However, the caller assigned blk from block_info[i] at the end of the
loop body, before the loop condition was re-evaluated. When i equaled
prewarm_stop_idx, this accessed memory beyond the allocated DSM segment,
causing a segfault.
Restructure the loop to check bounds at the top and assign blk at the
beginning of the loop body, where it is always safe. This avoids the
need for an explicit bounds check at the end.
Backpatch to 18, where the bug was introduced by commit
6acab8bdbcda.
Author: Matheus Alcantara <mths.dev@pm.me>
Reported-by: Glauber Batista <glauberrbatista@gmail.com>
Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Backpatch-through: 18
Discussion: https://www.postgresql.org/message-id/CAO%2B_mTQgQyTYwDh%3DU8iTnsDmOGyWsZJjUV31SmEYwmw6_xY6Bw%40mail.gmail.com
* valid forks or run out of options, we'll close the relation and
* move on.
*/
- while (i < apw_state->prewarm_stop_idx &&
- blk.tablespace == tablespace &&
- blk.filenumber == filenumber)
+ while (i < apw_state->prewarm_stop_idx)
{
- ForkNumber forknum = blk.forknum;
+ ForkNumber forknum;
BlockNumber nblocks;
struct AutoPrewarmReadStreamData p;
ReadStream *stream;
Buffer buf;
+ blk = block_info[i];
+
+ /* Stop when we reach a different relation. */
+ if (blk.tablespace != tablespace ||
+ blk.filenumber != filenumber)
+ break;
+
+ forknum = blk.forknum;
+
/*
* smgrexists is not safe for illegal forknum, hence check whether
* the passed forknum is valid before using it in smgrexists.
read_stream_end(stream);
- /* Advance i past all the blocks just prewarmed. */
+ /*
+ * Advance i past all the blocks just prewarmed. Note that the
+ * callback might have advanced the index beyond the last valid
+ * block, so don't access block_info[i] yet.
+ */
i = p.pos;
- blk = block_info[i];
}
relation_close(rel, AccessShareLock);