+ /*
+ * Maybe it needs to be endian swapped...
+ */
+ t.magic = fio_swap32(t.magic);
+ if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) {
+ *need_swap = 1;
+ return true;
+ }
+
+ return false;
+}
+
+#define FMINORBITS 20
+#define FMINORMASK ((1U << FMINORBITS) - 1)
+#define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS))
+#define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK))
+
+static void trace_add_open_close_event(struct thread_data *td, int fileno, enum file_log_act action)
+{
+ struct io_piece *ipo;
+
+ ipo = calloc(1, sizeof(*ipo));
+ init_ipo(ipo);
+
+ ipo->ddir = DDIR_INVAL;
+ ipo->fileno = fileno;
+ ipo->file_action = action;
+ flist_add_tail(&ipo->list, &td->io_log_list);
+}
+
+static int trace_add_file(struct thread_data *td, __u32 device)
+{
+ static unsigned int last_maj, last_min, last_fileno;
+ unsigned int maj = FMAJOR(device);
+ unsigned int min = FMINOR(device);
+ struct fio_file *f;
+ char dev[256];
+ unsigned int i;
+
+ if (last_maj == maj && last_min == min)
+ 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;
+ return last_fileno;
+ }
+
+ strcpy(dev, "/dev");
+ if (blktrace_lookup_device(td->o.replay_redirect, dev, maj, min)) {
+ 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);
+ td->o.open_files++;
+ td->files[fileno]->major = maj;
+ td->files[fileno]->minor = min;
+ trace_add_open_close_event(td, fileno, FIO_LOG_OPEN_FILE);
+ last_fileno = fileno;
+ }
+
+ 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);
+}
+
+/*
+ * 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)
+{
+ struct io_piece *ipo;
+
+ ipo = calloc(1, sizeof(*ipo));
+ init_ipo(ipo);
+
+ ipo->offset = offset * 512;
+ if (td->o.replay_scale)
+ ipo->offset = ipo->offset / td->o.replay_scale;
+ ipo_bytes_align(td->o.replay_align, 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 *bs)
+{
+ struct io_piece *ipo;
+ int fileno;
+
+ if (td->o.replay_skip & (1u << DDIR_TRIM))
+ return;
+
+ ipo = calloc(1, sizeof(*ipo));
+ init_ipo(ipo);
+ fileno = trace_add_file(td, t->device);
+
+ ios[DDIR_TRIM]++;
+ if (t->bytes > bs[DDIR_TRIM])
+ bs[DDIR_TRIM] = t->bytes;
+
+ td->o.size += t->bytes;
+
+ INIT_FLIST_HEAD(&ipo->list);
+
+ ipo->offset = t->sector * 512;
+ if (td->o.replay_scale)
+ ipo->offset = ipo->offset / td->o.replay_scale;
+ ipo_bytes_align(td->o.replay_align, 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 dump_trace(struct blk_io_trace *t)
+{
+ log_err("blktrace: ignoring zero byte trace: action=%x\n", t->action);
+}
+
+static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios,
+ unsigned int *bs)
+{
+ int rw;
+ int fileno;
+
+ fileno = trace_add_file(td, t->device);
+
+ rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
+
+ if (rw) {
+ if (td->o.replay_skip & (1u << DDIR_WRITE))
+ return;
+ } else {
+ if (td->o.replay_skip & (1u << DDIR_READ))
+ return;
+ }
+
+ if (!t->bytes) {
+ if (!fio_did_warn(FIO_WARN_BTRACE_ZERO))
+ dump_trace(t);
+ return;
+ }
+
+ if (t->bytes > bs[rw])
+ bs[rw] = t->bytes;
+
+ ios[rw]++;
+ td->o.size += t->bytes;
+ store_ipo(td, t->sector, t->bytes, rw, ttime, fileno);
+}
+
+static void handle_trace_flush(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios)
+{
+ struct io_piece *ipo;
+ int fileno;
+
+ if (td->o.replay_skip & (1u << DDIR_SYNC))
+ return;
+
+ ipo = calloc(1, sizeof(*ipo));
+ init_ipo(ipo);
+ fileno = trace_add_file(td, t->device);
+
+ ipo->delay = ttime / 1000;
+ ipo->ddir = DDIR_SYNC;
+ ipo->fileno = fileno;
+
+ ios[DDIR_SYNC]++;
+ dprint(FD_BLKTRACE, "store flush delay=%lu\n", ipo->delay);
+ queue_io_piece(td, ipo);
+}
+
+/*
+ * 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)
+ delay = 0;
+ else if (td->o.replay_time_scale == 100)
+ delay = t->time - last_ttime;
+ else {
+ double tmp = t->time - last_ttime;
+ double scale;
+
+ scale = (double) 100.0 / (double) td->o.replay_time_scale;
+ tmp *= scale;
+ delay = tmp;
+ }
+ 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 if (t->action & BLK_TC_ACT(BLK_TC_FLUSH))
+ handle_trace_flush(td, t, delay, ios);
+ else
+ handle_trace_fs(td, t, delay, ios, bs);