fix(mount): Fix long EC writes

This commit also fixes the Ganesha file_corruption test.

The idea behind the change was that some already erased ChunkData instances were supposed to be "alive" at the moment of inserting data, causing segmentation fault errors in long EC writes.

src/mount/writedata.cc

@@ -136,6 +136,7 @@ struct ChunkData {
 	bool workerWaitingForData;
 	ChunkData *next; // for iterating through the chunkdata of an inode
 	std::unique_ptr<WriteChunkLocator> locator;
+	std::atomic<int> beingWritten;
 
 	ChunkData(uint32_t chunkIndex, inodedata *parent)
 	    : chunkIndex(chunkIndex),
@@ -144,6 +145,7 @@ struct ChunkData {
 	      inQueue(false),
 	      workerWaitingForData(false),
 	      next(nullptr),
+	      beingWritten(0),
 	      parent_(parent) {
 #ifdef _WIN32
 		// We don't use inodeData->waitingworker and inodeData->pipe on Cygwin because
@@ -198,8 +200,8 @@ struct ChunkData {
 
 	void updateParentStatus(int8_t status) { parent_->status = status; }
 
-	int8_t getParentStatus() { 
-		return parent_->status; 
+	int8_t getParentStatus() {
+		return parent_->status;
 	}
 
 	bool requiresFlushing() const {
@@ -455,7 +457,9 @@ void write_job_delayed_end(ChunkData* chunkData, int status, int seconds, Glock
 		// Don't sleep if we have to write all the data immediately
 		seconds = 0;
 	}
-	if (!chunkData->dataChain.empty() && status == SAUNAFS_STATUS_OK) { // still have some work to do
+	if (chunkData->beingWritten) {
+		write_enqueue(chunkData, lock);
+	} else if (!chunkData->dataChain.empty() && status == SAUNAFS_STATUS_OK) { // still have some work to do
 		chunkData->tryCounter = 0; // on good write reset try counter
 		write_delayed_enqueue(chunkData, seconds, lock);
 	} else {        // no more work or error occurred
@@ -802,7 +806,7 @@ void write_data_term(void) {
 		pthread_join(write_worker_th[i], NULL);
 	}
 	pthread_join(delayed_queue_worker_th, NULL);
-	queue_delete(jqueue, queue_deleter_delete<inodedata>);
+	queue_delete(jqueue, queue_deleter_delete<ChunkData>);
 	for (i = 0; i < IDHASHSIZE; i++) {
 		for (id = idhash[i]; id; id = idn) {
 			idn = id->next;
@@ -862,8 +866,10 @@ int write_blocks(inodedata *id, uint64_t offset, uint32_t size,
 	uint32_t from = offset & SFSBLOCKMASK;
 	while (size > 0) {
 		ChunkData *chunkData = write_get_chunkdata(chindx, id);
+		chunkData->beingWritten++;
 		if (size > SFSBLOCKSIZE - from) {
 			if (write_block(chunkData, pos, from, SFSBLOCKSIZE, data) < 0) {
+				chunkData->beingWritten--;
 				return SAUNAFS_ERROR_IO;
 			}
 			size -= (SFSBLOCKSIZE - from);
@@ -876,10 +882,12 @@ int write_blocks(inodedata *id, uint64_t offset, uint32_t size,
 			}
 		} else {
 			if (write_block(chunkData, pos, from, from + size, data) < 0) {
+				chunkData->beingWritten--;
 				return SAUNAFS_ERROR_IO;
 			}
 			size = 0;
 		}
+		chunkData->beingWritten--;
 	}
 	return 0;
 }