[SERVER] Verify checksums of all images on SIGUSR1HEADmaster

6 files changed, 92 insertions, 14 deletions

diff --git a/src/server/globals.h b/src/server/globals.h
index bde1184..5129108 100644
--- a/src/server/globals.h
+++ b/src/server/globals.h
@@ -159,6 +159,7 @@ struct _dnbd3_image
 	} problem;
 	uint16_t rid;          // revision of image
 	bool accessed;         // image was accessed since .meta was written
+	bool wantCheck;        // true if the entire image should be checked as soon as the according thread is idle
 	pthread_mutex_t lock;
 };
 #define PIMG(x) (x)->name, (int)(x)->rid
diff --git a/src/server/image.c b/src/server/image.c
index 51fd5b6..f438c18 100644
--- a/src/server/image.c
+++ b/src/server/image.c
@@ -2125,3 +2125,38 @@ static void loadImageMeta(dnbd3_image_t *image)
 	timing_gets( &image->atime, offset );
 }
 
+void image_checkForNextFullCheck(void)
+{
+	int i;
+	dnbd3_image_t *check = NULL;
+
+	mutex_lock( &imageListLock );
+	for (i = 0; i < _num_images; ++i) {
+		dnbd3_image_t * const image = _images[i];
+		if ( image != NULL && image->wantCheck ) {
+			image->wantCheck = false;
+			check = image;
+			break;
+		}
+	}
+	mutex_unlock( &imageListLock );
+	if ( check != NULL ) {
+		logadd( LOG_DEBUG1, "Queueing next full image check" );
+		integrity_check( check, -1, false );
+	}
+}
+
+void image_hashAllImages(void)
+{
+	int i;
+
+	mutex_lock( &imageListLock );
+	for (i = 0; i < _num_images; ++i) {
+		dnbd3_image_t * const image = _images[i];
+		if ( image != NULL ) {
+			image->wantCheck = true;
+		}
+	}
+	mutex_unlock( &imageListLock );
+	integrity_trigger();
+}
diff --git a/src/server/image.h b/src/server/image.h
index 7b6583c..df8ac87 100644
--- a/src/server/image.h
+++ b/src/server/image.h
@@ -51,6 +51,10 @@ bool image_ensureDiskSpaceLocked(uint64_t size, bool force);
 
 bool image_saveCacheMap(dnbd3_image_t *image);
 
+void image_checkForNextFullCheck(void);
+
+void image_hashAllImages(void);
+
 /**
  * Check if given range is cached. Be careful when using this function because:
  * 1) you need to hold a reference to the cache map
diff --git a/src/server/integrity.c b/src/server/integrity.c
index 5a436ee..346e3e2 100644
--- a/src/server/integrity.c
+++ b/src/server/integrity.c
@@ -40,7 +40,7 @@ static void flushFileRange(dnbd3_image_t *image, uint64_t start, uint64_t end);
 /**
  * Initialize the integrity check thread
  */
-void integrity_init()
+void integrity_init(void)
 {
 	assert( queueLen == -1 );
 	mutex_init( &integrityQueueLock, LOCK_INTEGRITY_QUEUE );
@@ -56,7 +56,7 @@ void integrity_init()
 	}
 }
 
-void integrity_shutdown()
+void integrity_shutdown(void)
 {
 	assert( queueLen != -1 );
 	if ( !bRunning )
@@ -98,11 +98,16 @@ start_over:
 		}
 		// There is an existing check request for the given image, see if we can merge
 		if ( block == -1 ) {
-			// New request is supposed to check entire image, reset existing queue item
-			checkQueue[i].block = 0;
-			checkQueue[i].count = CHECK_ALL;
-			mutex_unlock( &integrityQueueLock );
-			return;
+			// New request is supposed to check entire image
+			if ( checkQueue[i].block == 0 && checkQueue[i].count == CHECK_ALL ) {
+				// Existing full check that didn't start yet, bail out
+				mutex_unlock( &integrityQueueLock );
+				return;
+			}
+			// Mark existing queue item as void
+			checkQueue[i].block = -1;
+			checkQueue[i].count = -1;
+			continue;
 		}
 		if ( checkQueue[i].block <= block ) {
 			// The block to check is after the block to check in queue
@@ -148,9 +153,20 @@ start_over:
 	mutex_unlock( &integrityQueueLock );
 }
 
+void integrity_trigger(void)
+{
+	if ( !bRunning )
+		return;
+	mutex_lock( &integrityQueueLock );
+	pthread_cond_signal( &queueSignal );
+	mutex_unlock( &integrityQueueLock );
+}
+
 static void* integrity_main(void * data UNUSED)
 {
 	int i;
+	bool queueAnother;
+
 	setThreadName( "image-check" );
 	blockNoncriticalSignals();
 #if defined(__linux__)
@@ -165,6 +181,7 @@ static void* integrity_main(void * data UNUSED)
 		if ( queueLen == 0 ) {
 			mutex_cond_wait( &queueSignal, &integrityQueueLock );
 		}
+		queueAnother = true;
 		for (i = queueLen - 1; i >= 0; --i) {
 			if ( _shutdown ) break;
 			dnbd3_image_t * const image = image_lock( checkQueue[i].image );
@@ -202,7 +219,7 @@ static void* integrity_main(void * data UNUSED)
 					if ( _shutdown )
 						break;
 					// Open for direct I/O if possible; this prevents polluting the fs cache
-					if ( directFd == -1 && ( end % DNBD3_BLOCK_SIZE ) == 0 ) {
+					if ( directFd == -1 && ( MIN( end, fileSize ) % DNBD3_BLOCK_SIZE ) == 0 ) {
 						// Use direct I/O only if read length is multiple of 4096 to be on the safe side
 						directFd = open( image->path, O_RDONLY | O_DIRECT );
 						if ( directFd == -1 ) {
@@ -231,7 +248,7 @@ static void* integrity_main(void * data UNUSED)
 						// If this is not a full check, queue one
 						if ( qCount != CHECK_ALL ) {
 							logadd( LOG_INFO, "Queueing full check for %s", image->name );
-							integrity_check( image, -1, false );
+							image->wantCheck = true;
 						}
 						foundCorrupted = true;
 					}
@@ -244,12 +261,17 @@ static void* integrity_main(void * data UNUSED)
 				}
 				mutex_lock( &integrityQueueLock );
 				assert( checkQueue[i].image == image );
-				if ( qCount != CHECK_ALL ) {
+				if ( checkQueue[i].block == -1 && checkQueue[i].count == -1 ) {
+					// Marked as dominated while we were checking - discard silently
+				} else if ( qCount != CHECK_ALL ) {
 					// Not a full check; update the counter
+					assert( checkQueue[i].count != CHECK_ALL );
 					checkQueue[i].count -= ( blocks[0] - checkQueue[i].block );
 					if ( checkQueue[i].count < 0 ) {
 						logadd( LOG_WARNING, "BUG! checkQueue counter ran negative" );
 					}
+				} else {
+					assert( checkQueue[i].count == CHECK_ALL );
 				}
 				if ( checkCount > 0 || checkQueue[i].count <= 0 ) {
 					// Done with this task as nothing left
@@ -258,6 +280,7 @@ static void* integrity_main(void * data UNUSED)
 				} else {
 					// Still more blocks to go...
 					checkQueue[i].block = blocks[0];
+					queueAnother = false; // Still busy
 				}
 			}
 			if ( foundCorrupted && !_shutdown ) {
@@ -267,6 +290,12 @@ static void* integrity_main(void * data UNUSED)
 			// Release :-)
 			image_release( image );
 		}
+		// See if there's another image queued for check
+		if ( queueAnother ) {
+			mutex_unlock( &integrityQueueLock );
+			image_checkForNextFullCheck();
+			mutex_lock( &integrityQueueLock );
+		}
 	}
 	mutex_unlock( &integrityQueueLock );
 	bRunning = false;
diff --git a/src/server/integrity.h b/src/server/integrity.h
index 09d3785..7a1b8d7 100644
--- a/src/server/integrity.h
+++ b/src/server/integrity.h
@@ -3,10 +3,12 @@
 
 #include "globals.h"
 
-void integrity_init();
+void integrity_init(void);
 
-void integrity_shutdown();
+void integrity_shutdown(void);
 
 void integrity_check(dnbd3_image_t *image, int block, bool blocking);
 
+void integrity_trigger(void);
+
 #endif /* INTEGRITY_H_ */
diff --git a/src/server/server.c b/src/server/server.c
index 0f75935..d086930 100644
--- a/src/server/server.c
+++ b/src/server/server.c
@@ -75,7 +75,7 @@ static poll_list_t *listeners = NULL;
  * Time the server was started
  */
 static ticks startupTime;
-static bool sigReload = false, sigLogCycle = false;
+static bool sigReload = false, sigLogCycle = false, sigHashAll = false;
 
 /**
  * Copied to in signal handler so we can print info
@@ -445,6 +445,11 @@ int main(int argc, char *argv[])
 			else
 				logadd( LOG_WARNING, "Could not cycle log file." );
 		}
+		if ( sigHashAll ) {
+			sigHashAll = false;
+			logadd( LOG_INFO, "SIGUSR1 received, verifying checksum of all images..." );
+			image_hashAllImages();
+		}
 		//
 		len = sizeof(client);
 		fd = sock_accept( listeners, &client, &len );
@@ -543,8 +548,10 @@ static void dnbd3_handleSignal(int signum)
 	if ( _shutdown ) return;
 	if ( signum == SIGINT || signum == SIGTERM ) {
 		_shutdown = true;
-	} else if ( signum == SIGUSR1 || signum == SIGHUP ) {
+	} else if ( signum == SIGHUP ) {
 		sigReload = true;
+	} else if ( signum == SIGUSR1 ) {
+		sigHashAll = true;
 	} else if ( signum == SIGUSR2 ) {
 		sigLogCycle = true;
 	}