Fix phpGH-13071: Copying large files using mmap-able source streams may exhaust available memory and fail

Commit 5cbe5a5 disabled chunking for all writes to streams. However,
user streams have a callback where code is executed on data that is
subject to the memory limit. Therefore, when using large writes or
stream_copy_to_stream/copy the memory limit can easily be hit with large
enough data.

To solve this, we reintroduce chunking for userspace streams.
Users have control over the chunk size, which is neat because
they can improve the performance by setting the chunk size if
that turns out to be a bottleneck.

In an ideal world, we add an option so we can "ask" the stream whether
it "prefers" chunked writes, similar to how we have
php_stream_mmap_supported & friends. However, that cannot be done on
stable branches.

Closes phpGH-13136.

Author: ndossche

NEWS

@@ -30,6 +30,10 @@ PHP                                                                        NEWS
   . Fixed bug GH-13138 (Randomizer::pickArrayKeys() does not detect broken
     engines). (timwolla)
 
+- Streams:
+  . Fixed bug GH-13071 (Copying large files using mmap-able source streams may
+    exhaust available memory and fail). (nielsdos)
+
 18 Jan 2024, PHP 8.2.15
 
 - Core:

ext/standard/tests/file/gh13136.phpt

@@ -0,0 +1,55 @@
+--TEST--
+GH-13071 (Copying large files using mmap-able source streams may exhaust available memory and fail)
+--FILE--
+<?php
+
+class CustomStream {
+    public $context;
+    protected $file;
+    protected $seekable;
+    public static int $writes = 0;
+
+    public function stream_open($path, $mode, $options, &$opened_path) {
+        $path = $this->trim_path($path);
+        $this->file = fopen($path, $mode);
+        return true;
+    }
+
+    public function stream_close() {
+        fclose($this->file);
+        return true;
+    }
+
+    public function stream_write($data) {
+        self::$writes++;
+        return fwrite($this->file, $data);
+    }
+
+    public function url_stat($path, $flags) {
+        return false;
+    }
+
+    private function trim_path(string $path): string {
+        return substr($path, strlen("up://"));
+    }
+}
+
+file_put_contents(__DIR__ . "/gh13071.tmp", str_repeat("a", 1024 * 1024 * 8));
+
+stream_wrapper_register("up", CustomStream::class, STREAM_IS_URL);
+
+$old_limit = ini_get("memory_limit");
+ini_set("memory_limit", memory_get_usage(true) + 5 * 1024 * 1024);
+copy(__DIR__ . "/gh13071.tmp", "up://" . __DIR__ . "/gh13071.out.tmp");
+ini_set("memory_limit", $old_limit);
+
+echo "Done ", CustomStream::$writes, " writes\n";
+
+?>
+--CLEAN--
+<?php
+@unlink(__DIR__ . "/gh13071.tmp");
+@unlink(__DIR__ . "/gh13071.out.tmp");
+?>
+--EXPECT--
+Done 1024 writes

ext/standard/tests/file/userstreams_006.phpt

@@ -34,5 +34,6 @@ bool(true)
 option: 3, 2, 50
 int(-1)
 int(8192)
-size: 70
+size: 42
+size: 28
 int(70)

ext/standard/tests/streams/set_file_buffer.phpt

@@ -39,4 +39,5 @@ option: %d, %d, %d
 int(%i)
 int(%d)
 size: %d
+size: 28
 int(%d)

ext/standard/tests/streams/stream_set_chunk_size.phpt

@@ -35,7 +35,7 @@ echo "should return previous chunk size (8192)\n";
 var_dump(stream_set_chunk_size($f, 1));
 echo "should be read without buffer (\$count == 10000)\n";
 var_dump(strlen(fread($f, 10000)));
-echo "should have no effect on writes\n";
+echo "should elicit 3 writes\n";
 var_dump(fwrite($f, str_repeat('b', 3)));
 
 echo "should return previous chunk size (1)\n";
@@ -46,7 +46,7 @@ echo "should elicit one read of size 100 (chunk size)\n";
 var_dump(strlen(fread($f, 50)));
 echo "should elicit no read because there is sufficient cached data\n";
 var_dump(strlen(fread($f, 50)));
-echo "should have no effect on writes\n";
+echo "should elicit 3 writes\n";
 var_dump(strlen(fwrite($f, str_repeat('b', 250))));
 
 echo "\nerror conditions\n";
@@ -68,8 +68,10 @@ int(8192)
 should be read without buffer ($count == 10000)
 read with size: 10000
 int(10000)
-should have no effect on writes
-write with size: 3
+should elicit 3 writes
+write with size: 1
+write with size: 1
+write with size: 1
 int(3)
 should return previous chunk size (1)
 int(1)
@@ -81,8 +83,10 @@ read with size: 100
 int(50)
 should elicit no read because there is sufficient cached data
 int(50)
-should have no effect on writes
-write with size: 250
+should elicit 3 writes
+write with size: 100
+write with size: 100
+write with size: 50
 int(3)
 
 error conditions

main/streams/streams.c

@@ -1133,8 +1133,15 @@ static ssize_t _php_stream_write_buffer(php_stream *stream, const char *buf, siz
 		stream->ops->seek(stream, stream->position, SEEK_SET, &stream->position);
 	}
 
+	/* See GH-13071: userspace stream is subject to the memory limit. */
+	size_t chunk_size = count;
+	if (php_stream_is(stream, PHP_STREAM_IS_USERSPACE)) {
+		/* If the stream is unbuffered, we can only write one byte at a time. */
+		chunk_size = stream->chunk_size;
+	}
+
 	while (count > 0) {
-		ssize_t justwrote = stream->ops->write(stream, buf, count);
+		ssize_t justwrote = stream->ops->write(stream, buf, MIN(chunk_size, count));
 		if (justwrote <= 0) {
 			/* If we already successfully wrote some bytes and a write error occurred
 			 * later, report the successfully written bytes. */