X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=iolog.c;h=961c9171f72f21654f17cc84f7ef07c8683b7991;hp=f9265b56dfa3cf83847d108fe0de60d89d5fb9d1;hb=52618a2a3565950a412739354be222f23e4fcdc5;hpb=97eabb2a4edcc237d5203db419728a1c598f061a

diff --git a/iolog.c b/iolog.c
index f9265b56..961c9171 100644
--- a/iolog.c
+++ b/iolog.c
@@ -18,7 +18,6 @@
 #include "verify.h"
 #include "trim.h"
 #include "filelock.h"
-#include "lib/tp.h"
 
 static const char iolog_ver2[] = "fio version 2 iolog";
 
@@ -28,19 +27,14 @@ void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
 	td->total_io_size += ipo->len;
 }
 
-void log_io_u(struct thread_data *td, struct io_u *io_u)
+void log_io_u(const struct thread_data *td, const struct io_u *io_u)
 {
-	const char *act[] = { "read", "write", "sync", "datasync",
-				"sync_file_range", "wait", "trim" };
-
-	assert(io_u->ddir <= 6);
-
 	if (!td->o.write_iolog_file)
 		return;
 
 	fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
-						act[io_u->ddir], io_u->offset,
-						io_u->buflen);
+						io_ddir_name(io_u->ddir),
+						io_u->offset, io_u->buflen);
 }
 
 void log_file(struct thread_data *td, struct fio_file *f,
@@ -65,20 +59,22 @@ void log_file(struct thread_data *td, struct fio_file *f,
 
 static void iolog_delay(struct thread_data *td, unsigned long delay)
 {
-	unsigned long usec = utime_since_now(&td->last_issue);
-	unsigned long this_delay;
+	uint64_t usec = utime_since_now(&td->last_issue);
+	uint64_t this_delay;
+	struct timeval tv;
+
+	if (delay < td->time_offset) {
+		td->time_offset = 0;
+		return;
+	}
 
+	delay -= td->time_offset;
 	if (delay < usec)
 		return;
 
 	delay -= usec;
 
-	/*
-	 * less than 100 usec delay, just regard it as noise
-	 */
-	if (delay < 100)
-		return;
-
+	fio_gettime(&tv, NULL);
 	while (delay && !td->terminate) {
 		this_delay = delay;
 		if (this_delay > 500000)
@@ -87,6 +83,12 @@ static void iolog_delay(struct thread_data *td, unsigned long delay)
 		usec_sleep(td, this_delay);
 		delay -= this_delay;
 	}
+
+	usec = utime_since_now(&tv);
+	if (usec > delay)
+		td->time_offset = usec - delay;
+	else
+		td->time_offset = 0;
 }
 
 static int ipo_special(struct thread_data *td, struct io_piece *ipo)
@@ -113,7 +115,7 @@ static int ipo_special(struct thread_data *td, struct io_piece *ipo)
 		td_io_close_file(td, f);
 		break;
 	case FIO_LOG_UNLINK_FILE:
-		unlink(f->file_name);
+		td_io_unlink_file(td, f);
 		break;
 	default:
 		log_err("fio: bad file action %d\n", ipo->file_action);
@@ -131,7 +133,7 @@ int read_iolog_get(struct thread_data *td, struct io_u *io_u)
 	while (!flist_empty(&td->io_log_list)) {
 		int ret;
 
-		ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
+		ipo = flist_first_entry(&td->io_log_list, struct io_piece, list);
 		flist_del(&ipo->list);
 		remove_trim_entry(td, ipo);
 
@@ -184,7 +186,7 @@ void prune_io_piece_log(struct thread_data *td)
 	}
 
 	while (!flist_empty(&td->io_hist_list)) {
-		ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
+		ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list);
 		flist_del(&ipo->list);
 		remove_trim_entry(td, ipo);
 		td->io_hist_len--;
@@ -247,6 +249,7 @@ restart:
 	p = &td->io_hist_tree.rb_node;
 	parent = NULL;
 	while (*p) {
+		int overlap = 0;
 		parent = *p;
 
 		__ipo = rb_entry(parent, struct io_piece, rb_node);
@@ -254,11 +257,18 @@ restart:
 			p = &(*p)->rb_left;
 		else if (ipo->file > __ipo->file)
 			p = &(*p)->rb_right;
-		else if (ipo->offset < __ipo->offset)
+		else if (ipo->offset < __ipo->offset) {
 			p = &(*p)->rb_left;
-		else if (ipo->offset > __ipo->offset)
+			overlap = ipo->offset + ipo->len > __ipo->offset;
+		}
+		else if (ipo->offset > __ipo->offset) {
 			p = &(*p)->rb_right;
-		else {
+			overlap = __ipo->offset + __ipo->len > ipo->offset;
+		}
+		else
+			overlap = 1;
+
+		if (overlap) {
 			dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu",
 				__ipo->offset, __ipo->len,
 				ipo->offset, ipo->len);
@@ -280,6 +290,18 @@ void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
 {
 	struct io_piece *ipo = io_u->ipo;
 
+	if (td->ts.nr_block_infos) {
+		uint32_t *info = io_u_block_info(td, io_u);
+		if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) {
+			if (io_u->ddir == DDIR_TRIM)
+				*info = BLOCK_INFO_SET_STATE(*info,
+						BLOCK_STATE_TRIM_FAILURE);
+			else if (io_u->ddir == DDIR_WRITE)
+				*info = BLOCK_INFO_SET_STATE(*info,
+						BLOCK_STATE_WRITE_FAILURE);
+		}
+	}
+
 	if (!ipo)
 		return;
 
@@ -293,7 +315,7 @@ void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
 	td->io_hist_len--;
 }
 
-void trim_io_piece(struct thread_data *td, struct io_u *io_u)
+void trim_io_piece(struct thread_data *td, const struct io_u *io_u)
 {
 	struct io_piece *ipo = io_u->ipo;
 
@@ -550,9 +572,9 @@ int init_iolog(struct thread_data *td)
 void setup_log(struct io_log **log, struct log_params *p,
 	       const char *filename)
 {
-	struct io_log *l = malloc(sizeof(*l));
+	struct io_log *l;
 
-	memset(l, 0, sizeof(*l));
+	l = calloc(1, sizeof(*l));
 	l->nr_samples = 0;
 	l->max_samples = 1024;
 	l->log_type = p->log_type;
@@ -565,13 +587,13 @@ void setup_log(struct io_log **log, struct log_params *p,
 	l->td = p->td;
 
 	if (l->log_offset)
-		l->log_ddir_mask = 0x80000000;
+		l->log_ddir_mask = LOG_OFFSET_SAMPLE_BIT;
 
 	INIT_FLIST_HEAD(&l->chunk_list);
 
 	if (l->log_gz && !p->td)
 		l->log_gz = 0;
-	else if (l->log_gz) {
+	else if (l->log_gz || l->log_gz_store) {
 		pthread_mutex_init(&l->chunk_lock, NULL);
 		p->td->flags |= TD_F_COMPRESS_LOG;
 	}
@@ -622,10 +644,7 @@ static void flush_samples(FILE *f, void *samples, uint64_t sample_size)
 		return;
 
 	s = __get_sample(samples, 0, 0);
-	if (s->__ddir & 0x80000000)
-		log_offset = 1;
-	else
-		log_offset = 0;
+	log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
 
 	nr_samples = sample_size / __log_entry_sz(log_offset);
 
@@ -652,7 +671,12 @@ static void flush_samples(FILE *f, void *samples, uint64_t sample_size)
 #ifdef CONFIG_ZLIB
 
 struct iolog_flush_data {
-	struct tp_work work;
+	struct workqueue_work work;
+	pthread_mutex_t lock;
+	pthread_cond_t cv;
+	int wait;
+	volatile int done;
+	volatile int refs;
 	struct io_log *log;
 	void *samples;
 	uint64_t nr_samples;
@@ -663,7 +687,6 @@ struct iolog_compress {
 	void *buf;
 	size_t len;
 	unsigned int seq;
-	int nofree;
 };
 
 #define GZ_CHUNK	131072
@@ -677,16 +700,13 @@ static struct iolog_compress *get_new_chunk(unsigned int seq)
 	c->buf = malloc(GZ_CHUNK);
 	c->len = 0;
 	c->seq = seq;
-	c->nofree = 0;
 	return c;
 }
 
 static void free_chunk(struct iolog_compress *ic)
 {
-	if (!ic->nofree) {
-		free(ic->buf);
-		free(ic);
-	}
+	free(ic->buf);
+	free(ic);
 }
 
 static int z_stream_init(z_stream *stream, int gz_hdr)
@@ -698,6 +718,10 @@ static int z_stream_init(z_stream *stream, int gz_hdr)
 	stream->opaque = Z_NULL;
 	stream->next_in = Z_NULL;
 
+	/*
+	 * zlib magic - add 32 for auto-detection of gz header or not,
+	 * if we decide to store files in a gzip friendly format.
+	 */
 	if (gz_hdr)
 		wbits += 32;
 
@@ -707,8 +731,9 @@ static int z_stream_init(z_stream *stream, int gz_hdr)
 	return 0;
 }
 
-struct flush_chunk_iter {
+struct inflate_chunk_iter {
 	unsigned int seq;
+	int err;
 	void *buf;
 	size_t buf_size;
 	size_t buf_used;
@@ -716,7 +741,7 @@ struct flush_chunk_iter {
 };
 
 static void finish_chunk(z_stream *stream, FILE *f,
-			 struct flush_chunk_iter *iter)
+			 struct inflate_chunk_iter *iter)
 {
 	int ret;
 
@@ -730,9 +755,18 @@ static void finish_chunk(z_stream *stream, FILE *f,
 	iter->buf_size = iter->buf_used = 0;
 }
 
-static int flush_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
-		       z_stream *stream, struct flush_chunk_iter *iter)
+/*
+ * Iterative chunk inflation. Handles cases where we cross into a new
+ * sequence, doing flush finish of previous chunk if needed.
+ */
+static size_t inflate_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
+			    z_stream *stream, struct inflate_chunk_iter *iter)
 {
+	size_t ret;
+
+	dprint(FD_COMPRESS, "inflate chunk size=%lu, seq=%u",
+				(unsigned long) ic->len, ic->seq);
+
 	if (ic->seq != iter->seq) {
 		if (iter->seq)
 			finish_chunk(stream, f, iter);
@@ -759,6 +793,7 @@ static int flush_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
 		err = inflate(stream, Z_NO_FLUSH);
 		if (err < 0) {
 			log_err("fio: failed inflating log: %d\n", err);
+			iter->err = err;
 			break;
 		}
 
@@ -774,43 +809,67 @@ static int flush_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
 			break;
 	}
 
-	free_chunk(ic);
-	return 0;
+	ret = (void *) stream->next_in - ic->buf;
+
+	dprint(FD_COMPRESS, "inflated to size=%lu\n", (unsigned long) ret);
+
+	return ret;
 }
 
-static void flush_gz_chunks(struct io_log *log, FILE *f)
+/*
+ * Inflate stored compressed chunks, or write them directly to the log
+ * file if so instructed.
+ */
+static int inflate_gz_chunks(struct io_log *log, FILE *f)
 {
-	struct flush_chunk_iter iter = { .chunk_sz = log->log_gz, };
-	struct flist_head *node;
+	struct inflate_chunk_iter iter = { .chunk_sz = log->log_gz, };
 	z_stream stream;
 
 	while (!flist_empty(&log->chunk_list)) {
 		struct iolog_compress *ic;
 
-		node = log->chunk_list.next;
-		ic = flist_entry(node, struct iolog_compress, list);
+		ic = flist_first_entry(&log->chunk_list, struct iolog_compress, list);
 		flist_del(&ic->list);
 
 		if (log->log_gz_store) {
-			fwrite(ic->buf, ic->len, 1, f);
-			free_chunk(ic);
+			size_t ret;
+
+			dprint(FD_COMPRESS, "log write chunk size=%lu, "
+				"seq=%u\n", (unsigned long) ic->len, ic->seq);
+
+			ret = fwrite(ic->buf, ic->len, 1, f);
+			if (ret != 1 || ferror(f)) {
+				iter.err = errno;
+				log_err("fio: error writing compressed log\n");
+			}
 		} else
-			flush_chunk(ic, log->log_gz_store, f, &stream, &iter);
+			inflate_chunk(ic, log->log_gz_store, f, &stream, &iter);
+
+		free_chunk(ic);
 	}
 
 	if (iter.seq) {
 		finish_chunk(&stream, f, &iter);
 		free(iter.buf);
 	}
+
+	return iter.err;
 }
 
+/*
+ * Open compressed log file and decompress the stored chunks and
+ * write them to stdout. The chunks are stored sequentially in the
+ * file, so we iterate over them and do them one-by-one.
+ */
 int iolog_file_inflate(const char *file)
 {
-	struct flush_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, };
+	struct inflate_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, };
 	struct iolog_compress ic;
 	z_stream stream;
 	struct stat sb;
 	ssize_t ret;
+	size_t total;
+	void *buf;
 	FILE *f;
 
 	f = fopen(file, "r");
@@ -825,49 +884,80 @@ int iolog_file_inflate(const char *file)
 		return 1;
 	}
 
-	ic.buf = malloc(sb.st_size);
+	ic.buf = buf = malloc(sb.st_size);
 	ic.len = sb.st_size;
-	ic.nofree = 1;
 	ic.seq = 1;
 
 	ret = fread(ic.buf, ic.len, 1, f);
 	if (ret < 0) {
 		perror("fread");
 		fclose(f);
+		free(buf);
 		return 1;
 	} else if (ret != 1) {
 		log_err("fio: short read on reading log\n");
 		fclose(f);
+		free(buf);
 		return 1;
 	}
 
 	fclose(f);
 
-	flush_chunk(&ic,  1, stdout, &stream, &iter);
+	/*
+	 * Each chunk will return Z_STREAM_END. We don't know how many
+	 * chunks are in the file, so we just keep looping and incrementing
+	 * the sequence number until we have consumed the whole compressed
+	 * file.
+	 */
+	total = ic.len;
+	do {
+		size_t iret;
+
+		iret = inflate_chunk(&ic,  1, stdout, &stream, &iter);
+		total -= iret;
+		if (!total)
+			break;
+		if (iter.err)
+			break;
+
+		ic.seq++;
+		ic.len -= iret;
+		ic.buf += iret;
+	} while (1);
 
 	if (iter.seq) {
 		finish_chunk(&stream, stdout, &iter);
 		free(iter.buf);
 	}
 
-	free(ic.buf);
-	return 0;
+	free(buf);
+	return iter.err;
 }
 
 #else
 
-static void flush_gz_chunks(struct io_log *log, FILE *f)
+static int inflate_gz_chunks(struct io_log *log, FILE *f)
+{
+	return 0;
+}
+
+int iolog_file_inflate(const char *file)
 {
+	log_err("fio: log inflation not possible without zlib\n");
+	return 1;
 }
 
 #endif
 
-void flush_log(struct io_log *log)
+void flush_log(struct io_log *log, int do_append)
 {
 	void *buf;
 	FILE *f;
 
-	f = fopen(log->filename, "w");
+	if (!do_append)
+		f = fopen(log->filename, "w");
+	else
+		f = fopen(log->filename, "a");
 	if (!f) {
 		perror("fopen log");
 		return;
@@ -875,7 +965,7 @@ void flush_log(struct io_log *log)
 
 	buf = set_file_buffer(f);
 
-	flush_gz_chunks(log, f);
+	inflate_gz_chunks(log, f);
 
 	flush_samples(f, log->log, log->nr_samples * log_entry_sz(log));
 
@@ -885,7 +975,7 @@ void flush_log(struct io_log *log)
 
 static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 {
-	if (td->tp_data)
+	if (td->flags & TD_F_COMPRESS_LOG)
 		iolog_flush(log, 1);
 
 	if (trylock) {
@@ -897,7 +987,7 @@ static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 	if (td->client_type == FIO_CLIENT_TYPE_GUI)
 		fio_send_iolog(td, log, log->filename);
 	else
-		flush_log(log);
+		flush_log(log, !td->o.per_job_logs);
 
 	fio_unlock_file(log->filename);
 	free_log(log);
@@ -906,7 +996,21 @@ static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
 
 #ifdef CONFIG_ZLIB
 
-static int gz_work(struct tp_work *work)
+static void drop_data(struct iolog_flush_data *data, int refs)
+{
+	if (!refs) {
+		free(data);
+		pthread_mutex_destroy(&data->lock);
+		pthread_cond_destroy(&data->cv);
+	}
+}
+
+/*
+ * Invoked from our compress helper thread, when logging would have exceeded
+ * the specified memory limitation. Compresses the previously stored
+ * entries.
+ */
+static int gz_work(struct submit_worker *sw, struct workqueue_work *work)
 {
 	struct iolog_flush_data *data;
 	struct iolog_compress *c;
@@ -935,6 +1039,8 @@ static int gz_work(struct tp_work *work)
 	stream.next_in = (void *) data->samples;
 	stream.avail_in = data->nr_samples * log_entry_sz(data->log);
 
+	dprint(FD_COMPRESS, "deflate input size=%lu, seq=%u\n",
+				(unsigned long) stream.avail_in, seq);
 	do {
 		c = get_new_chunk(seq);
 		stream.avail_out = GZ_CHUNK;
@@ -942,7 +1048,8 @@ static int gz_work(struct tp_work *work)
 		ret = deflate(&stream, Z_NO_FLUSH);
 		if (ret < 0) {
 			log_err("fio: deflate log (%d)\n", ret);
-			break;
+			free_chunk(c);
+			goto err;
 		}
 
 		c->len = GZ_CHUNK - stream.avail_out;
@@ -963,10 +1070,13 @@ static int gz_work(struct tp_work *work)
 			stream.next_out = c->buf;
 			ret = deflate(&stream, Z_FINISH);
 			c->len = GZ_CHUNK - stream.avail_out;
+			total += c->len;
 			flist_add_tail(&c->list, &list);
 		} while (ret != Z_STREAM_END);
 	}
 
+	dprint(FD_COMPRESS, "deflated to size=%lu\n", (unsigned long) total);
+
 	ret = deflateEnd(&stream);
 	if (ret != Z_OK)
 		log_err("fio: deflateEnd %d\n", ret);
@@ -979,18 +1089,60 @@ static int gz_work(struct tp_work *work)
 		pthread_mutex_unlock(&data->log->chunk_lock);
 	}
 
-	if (work->wait) {
-		work->done = 1;
-		pthread_cond_signal(&work->cv);
+	ret = 0;
+done:
+	if (data->wait) {
+		int refs;
+
+		pthread_mutex_lock(&data->lock);
+		data->done = 1;
+		pthread_cond_signal(&data->cv);
+		refs = --data->refs;
+		pthread_mutex_unlock(&data->lock);
+		drop_data(data, refs);
 	} else
 		free(data);
+	return ret;
+err:
+	while (!flist_empty(&list)) {
+		c = flist_first_entry(list.next, struct iolog_compress, list);
+		flist_del(&c->list);
+		free_chunk(c);
+	}
+	ret = 1;
+	goto done;
+}
 
+static struct workqueue_ops log_compress_wq_ops = {
+	.fn	= gz_work,
+	.nice	= 1,
+};
+
+int iolog_compress_init(struct thread_data *td)
+{
+	if (!(td->flags & TD_F_COMPRESS_LOG))
+		return 0;
+
+	workqueue_init(td, &td->log_compress_wq, &log_compress_wq_ops, 1);
 	return 0;
 }
 
+void iolog_compress_exit(struct thread_data *td)
+{
+	if (!(td->flags & TD_F_COMPRESS_LOG))
+		return;
+
+	workqueue_exit(&td->log_compress_wq);
+}
+
+/*
+ * Queue work item to compress the existing log entries. We copy the
+ * samples, and reset the log sample count to 0 (so the logging will
+ * continue to use the memory associated with the log). If called with
+ * wait == 1, will not return until the log compression has completed.
+ */
 int iolog_flush(struct io_log *log, int wait)
 {
-	struct tp_data *tdat = log->td->tp_data;
 	struct iolog_flush_data *data;
 	size_t sample_size;
 
@@ -1009,24 +1161,27 @@ int iolog_flush(struct io_log *log, int wait)
 
 	memcpy(data->samples, log->log, sample_size);
 	data->nr_samples = log->nr_samples;
-	data->work.fn = gz_work;
 	log->nr_samples = 0;
 
-	if (wait) {
-		pthread_mutex_init(&data->work.lock, NULL);
-		pthread_cond_init(&data->work.cv, NULL);
-		data->work.wait = 1;
-	} else
-		data->work.wait = 0;
+	data->wait = wait;
+	if (data->wait) {
+		pthread_mutex_init(&data->lock, NULL);
+		pthread_cond_init(&data->cv, NULL);
+		data->done = 0;
+		data->refs = 2;
+	}
 
-	tp_queue_work(tdat, &data->work);
+	workqueue_enqueue(&log->td->log_compress_wq, &data->work);
 
 	if (wait) {
-		pthread_mutex_lock(&data->work.lock);
-		while (!data->work.done)
-			pthread_cond_wait(&data->work.cv, &data->work.lock);
-		pthread_mutex_unlock(&data->work.lock);
-		free(data);
+		int refs;
+
+		pthread_mutex_lock(&data->lock);
+		while (!data->done)
+			pthread_cond_wait(&data->cv, &data->lock);
+		refs = --data->refs;
+		pthread_mutex_unlock(&data->lock);
+		drop_data(data, refs);
 	}
 
 	return 0;
@@ -1039,6 +1194,15 @@ int iolog_flush(struct io_log *log, int wait)
 	return 1;
 }
 
+int iolog_compress_init(struct thread_data *td)
+{
+	return 0;
+}
+
+void iolog_compress_exit(struct thread_data *td)
+{
+}
+
 #endif
 
 static int write_iops_log(struct thread_data *td, int try)