+ fd = open(dev, O_RDONLY);
+ if (fd < 0)
+ return 1;
+
+ if (ioctl(fd, BLKSSZGET, bs) < 0) {
+ close(fd);
+ return 1;
+ }
+
+ close(fd);
+ return 0;
+}
+
+static int trace_add_file(struct thread_data *td, __u32 device,
+ unsigned int *bs)
+{
+ static unsigned int last_maj, last_min, last_fileno, last_bs;
+ unsigned int maj = FMAJOR(device);
+ unsigned int min = FMINOR(device);
+ struct fio_file *f;
+ unsigned int i;
+ char dev[256];
+
+ if (last_maj == maj && last_min == min) {
+ *bs = last_bs;
+ return last_fileno;
+ }
+
+ last_maj = maj;
+ last_min = min;
+
+ /*
+ * check for this file in our list
+ */
+ for_each_file(td, f, i) {
+ if (f->major == maj && f->minor == min) {
+ last_fileno = f->fileno;
+ last_bs = f->bs;
+ goto out;
+ }
+ }
+
+ strcpy(dev, "/dev");
+ if (blktrace_lookup_device(td->o.replay_redirect, dev, maj, min)) {
+ unsigned int this_bs;
+ int fileno;
+
+ if (td->o.replay_redirect)
+ dprint(FD_BLKTRACE, "device lookup: %d/%d\n overridden"
+ " with: %s\n", maj, min,
+ td->o.replay_redirect);
+ else
+ dprint(FD_BLKTRACE, "device lookup: %d/%d\n", maj, min);
+
+ dprint(FD_BLKTRACE, "add devices %s\n", dev);
+ fileno = add_file_exclusive(td, dev);
+
+ if (get_dev_blocksize(dev, &this_bs))
+ this_bs = 512;
+
+ td->o.open_files++;
+ td->files[fileno]->major = maj;
+ td->files[fileno]->minor = min;
+ td->files[fileno]->bs = this_bs;
+ trace_add_open_close_event(td, fileno, FIO_LOG_OPEN_FILE);
+
+ last_fileno = fileno;
+ last_bs = this_bs;
+ }
+
+out:
+ *bs = last_bs;
+ return last_fileno;
+}
+
+static void t_bytes_align(struct thread_options *o, struct blk_io_trace *t)
+{
+ if (!o->replay_align)
+ return;
+
+ t->bytes = (t->bytes + o->replay_align - 1) & ~(o->replay_align - 1);
+}
+
+static void ipo_bytes_align(struct thread_options *o, struct io_piece *ipo)
+{
+ if (!o->replay_align)
+ return;
+
+ ipo->offset &= ~(o->replay_align - 1);
+}
+
+
+/*
+ * Store blk_io_trace data in an ipo for later retrieval.
+ */
+static void store_ipo(struct thread_data *td, unsigned long long offset,
+ unsigned int bytes, int rw, unsigned long long ttime,
+ int fileno, unsigned int bs)
+{
+ struct io_piece *ipo = malloc(sizeof(*ipo));
+
+ init_ipo(ipo);
+
+ ipo->offset = offset * bs;
+ if (td->o.replay_scale)
+ ipo->offset = ipo->offset / td->o.replay_scale;
+ ipo_bytes_align(&td->o, ipo);
+ ipo->len = bytes;
+ ipo->delay = ttime / 1000;
+ if (rw)
+ ipo->ddir = DDIR_WRITE;
+ else
+ ipo->ddir = DDIR_READ;
+ ipo->fileno = fileno;
+
+ dprint(FD_BLKTRACE, "store ddir=%d, off=%llu, len=%lu, delay=%lu\n",
+ ipo->ddir, ipo->offset,
+ ipo->len, ipo->delay);
+ queue_io_piece(td, ipo);
+}
+
+static void handle_trace_notify(struct blk_io_trace *t)
+{
+ switch (t->action) {
+ case BLK_TN_PROCESS:
+ dprint(FD_BLKTRACE, "got process notify: %x, %d\n",
+ t->action, t->pid);
+ break;
+ case BLK_TN_TIMESTAMP:
+ dprint(FD_BLKTRACE, "got timestamp notify: %x, %d\n",
+ t->action, t->pid);
+ break;
+ case BLK_TN_MESSAGE:
+ break;
+ default:
+ dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action);
+ break;
+ }
+}
+
+static void handle_trace_discard(struct thread_data *td,
+ struct blk_io_trace *t,
+ unsigned long long ttime,
+ unsigned long *ios, unsigned int *rw_bs)
+{
+ struct io_piece *ipo = malloc(sizeof(*ipo));
+ unsigned int bs;
+ int fileno;
+
+ init_ipo(ipo);
+ fileno = trace_add_file(td, t->device, &bs);
+
+ ios[DDIR_TRIM]++;
+ if (t->bytes > rw_bs[DDIR_TRIM])
+ rw_bs[DDIR_TRIM] = t->bytes;
+
+ td->o.size += t->bytes;
+
+ memset(ipo, 0, sizeof(*ipo));
+ INIT_FLIST_HEAD(&ipo->list);
+
+ ipo->offset = t->sector * bs;
+ if (td->o.replay_scale)
+ ipo->offset = ipo->offset / td->o.replay_scale;
+ ipo_bytes_align(&td->o, ipo);
+ ipo->len = t->bytes;
+ ipo->delay = ttime / 1000;
+ ipo->ddir = DDIR_TRIM;
+ ipo->fileno = fileno;
+
+ dprint(FD_BLKTRACE, "store discard, off=%llu, len=%lu, delay=%lu\n",
+ ipo->offset, ipo->len,
+ ipo->delay);
+ queue_io_piece(td, ipo);
+}
+
+static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios,
+ unsigned int *rw_bs)
+{
+ unsigned int bs;
+ int rw;
+ int fileno;
+
+ fileno = trace_add_file(td, t->device, &bs);
+
+ rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
+
+ if (t->bytes > rw_bs[rw])
+ rw_bs[rw] = t->bytes;
+
+ ios[rw]++;
+ td->o.size += t->bytes;
+ store_ipo(td, t->sector, t->bytes, rw, ttime, fileno, bs);
+}
+
+/*
+ * We only care for queue traces, most of the others are side effects
+ * due to internal workings of the block layer.
+ */
+static void handle_trace(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long *ios, unsigned int *bs)
+{
+ static unsigned long long last_ttime;
+ unsigned long long delay = 0;
+
+ if ((t->action & 0xffff) != __BLK_TA_QUEUE)
+ return;
+
+ if (!(t->action & BLK_TC_ACT(BLK_TC_NOTIFY))) {
+ if (!last_ttime || td->o.no_stall) {
+ last_ttime = t->time;
+ delay = 0;
+ } else {
+ delay = t->time - last_ttime;
+ last_ttime = t->time;
+ }
+ }
+
+ t_bytes_align(&td->o, t);
+
+ if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
+ handle_trace_notify(t);
+ else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
+ handle_trace_discard(td, t, delay, ios, bs);
+ else
+ handle_trace_fs(td, t, delay, ios, bs);
+}
+
+static void byteswap_trace(struct blk_io_trace *t)
+{
+ t->magic = fio_swap32(t->magic);
+ t->sequence = fio_swap32(t->sequence);
+ t->time = fio_swap64(t->time);
+ t->sector = fio_swap64(t->sector);
+ t->bytes = fio_swap32(t->bytes);
+ t->action = fio_swap32(t->action);
+ t->pid = fio_swap32(t->pid);
+ t->device = fio_swap32(t->device);
+ t->cpu = fio_swap32(t->cpu);
+ t->error = fio_swap16(t->error);
+ t->pdu_len = fio_swap16(t->pdu_len);
+}
+
+static int t_is_write(struct blk_io_trace *t)
+{
+ return (t->action & BLK_TC_ACT(BLK_TC_WRITE | BLK_TC_DISCARD)) != 0;
+}
+
+static enum fio_ddir t_get_ddir(struct blk_io_trace *t)
+{
+ if (t->action & BLK_TC_ACT(BLK_TC_READ))
+ return DDIR_READ;
+ else if (t->action & BLK_TC_ACT(BLK_TC_WRITE))
+ return DDIR_WRITE;
+ else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
+ return DDIR_TRIM;
+
+ return DDIR_INVAL;
+}
+
+static void depth_inc(struct blk_io_trace *t, int *depth)
+{
+ enum fio_ddir ddir;
+
+ ddir = t_get_ddir(t);
+ if (ddir != DDIR_INVAL)
+ depth[ddir]++;
+}
+
+static void depth_dec(struct blk_io_trace *t, int *depth)
+{
+ enum fio_ddir ddir;
+
+ ddir = t_get_ddir(t);
+ if (ddir != DDIR_INVAL)
+ depth[ddir]--;
+}
+
+static void depth_end(struct blk_io_trace *t, int *this_depth, int *depth)
+{
+ enum fio_ddir ddir = DDIR_INVAL;
+
+ ddir = t_get_ddir(t);
+ if (ddir != DDIR_INVAL) {
+ depth[ddir] = max(depth[ddir], this_depth[ddir]);
+ this_depth[ddir] = 0;
+ }
+}
+
+/*
+ * Load a blktrace file by reading all the blk_io_trace entries, and storing
+ * them as io_pieces like the fio text version would do.
+ */
+int load_blktrace(struct thread_data *td, const char *filename, int need_swap)
+{
+ struct blk_io_trace t;
+ unsigned long ios[DDIR_RWDIR_CNT], skipped_writes;
+ unsigned int rw_bs[DDIR_RWDIR_CNT];
+ struct fifo *fifo;
+ int fd, i, old_state;
+ struct fio_file *f;
+ int this_depth[DDIR_RWDIR_CNT], depth[DDIR_RWDIR_CNT], max_depth;
+