X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=blktrace.c;h=00e5f9a9b7d2b621b80bc50008b1b8f391f6ee76;hp=bb0bcbe60ac52e2f7f6c1fda34983f704e94ada0;hb=HEAD;hpb=76a00ec6684317902e536365b89531431e7f8867 diff --git a/blktrace.c b/blktrace.c index bb0bcbe6..ef9ce6bf 100644 --- a/blktrace.c +++ b/blktrace.c @@ -3,98 +3,65 @@ */ #include #include -#include -#include +#include +#include +#include #include "flist.h" #include "fio.h" +#include "iolog.h" +#include "blktrace.h" #include "blktrace_api.h" -#include "lib/linux-dev-lookup.h" +#include "oslib/linux-dev-lookup.h" -#define TRACE_FIFO_SIZE 8192 - -/* - * fifo refill frontend, to avoid reading data in trace sized bites - */ -static int refill_fifo(struct thread_data *td, struct fifo *fifo, int fd) -{ - char buf[TRACE_FIFO_SIZE]; - unsigned int total; - int ret; - - total = sizeof(buf); - if (total > fifo_room(fifo)) - total = fifo_room(fifo); - - ret = read(fd, buf, total); - if (ret < 0) { - td_verror(td, errno, "read blktrace file"); - return -1; - } - - if (ret > 0) - ret = fifo_put(fifo, buf, ret); - - dprint(FD_BLKTRACE, "refill: filled %d bytes\n", ret); - return ret; -} - -/* - * Retrieve 'len' bytes from the fifo, refilling if necessary. - */ -static int trace_fifo_get(struct thread_data *td, struct fifo *fifo, int fd, - void *buf, unsigned int len) -{ - if (fifo_len(fifo) < len) { - int ret = refill_fifo(td, fifo, fd); - - if (ret < 0) - return ret; - } - - return fifo_get(fifo, buf, len); -} +struct file_cache { + unsigned int maj; + unsigned int min; + unsigned int fileno; +}; /* * Just discard the pdu by seeking past it. */ -static int discard_pdu(struct thread_data *td, struct fifo *fifo, int fd, - struct blk_io_trace *t) +static int discard_pdu(FILE* f, struct blk_io_trace *t) { if (t->pdu_len == 0) return 0; dprint(FD_BLKTRACE, "discard pdu len %u\n", t->pdu_len); - return trace_fifo_get(td, fifo, fd, NULL, t->pdu_len); + if (fseek(f, t->pdu_len, SEEK_CUR) < 0) + return -errno; + + return t->pdu_len; } /* * Check if this is a blktrace binary data file. We read a single trace * into memory and check for the magic signature. */ -int is_blktrace(const char *filename, int *need_swap) +bool is_blktrace(const char *filename, int *need_swap) { struct blk_io_trace t; int fd, ret; fd = open(filename, O_RDONLY); if (fd < 0) - return 0; + return false; ret = read(fd, &t, sizeof(t)); close(fd); if (ret < 0) { perror("read blktrace"); - return 0; + return false; } else if (ret != sizeof(t)) { log_err("fio: short read on blktrace file\n"); - return 0; + return false; } if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) { *need_swap = 0; - return 1; + return true; } /* @@ -103,10 +70,10 @@ int is_blktrace(const char *filename, int *need_swap) t.magic = fio_swap32(t.magic); if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) { *need_swap = 1; - return 1; + return true; } - return 0; + return false; } #define FMINORBITS 20 @@ -127,28 +94,28 @@ static void trace_add_open_close_event(struct thread_data *td, int fileno, enum flist_add_tail(&ipo->list, &td->io_log_list); } -static int trace_add_file(struct thread_data *td, __u32 device) +static int trace_add_file(struct thread_data *td, __u32 device, + struct file_cache *cache) { - 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; + if (cache->maj == maj && cache->min == min) + return cache->fileno; - last_maj = maj; - last_min = min; + cache->maj = maj; + cache->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; + cache->fileno = f->fileno; + return cache->fileno; } strcpy(dev, "/dev"); @@ -168,10 +135,18 @@ static int trace_add_file(struct thread_data *td, __u32 device) td->files[fileno]->major = maj; td->files[fileno]->minor = min; trace_add_open_close_event(td, fileno, FIO_LOG_OPEN_FILE); - last_fileno = fileno; + cache->fileno = fileno; } - return last_fileno; + return cache->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); } /* @@ -181,14 +156,15 @@ 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 = malloc(sizeof(*ipo)); + struct io_piece *ipo; + ipo = calloc(1, sizeof(*ipo)); init_ipo(ipo); - /* - * the 512 is wrong here, it should be the hardware sector size... - */ 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) @@ -203,7 +179,7 @@ static void store_ipo(struct thread_data *td, unsigned long long offset, queue_io_piece(td, ipo); } -static void handle_trace_notify(struct blk_io_trace *t) +static bool handle_trace_notify(struct blk_io_trace *t) { switch (t->action) { case BLK_TN_PROCESS: @@ -220,18 +196,24 @@ static void handle_trace_notify(struct blk_io_trace *t) dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action); break; } + return false; } -static void handle_trace_discard(struct thread_data *td, +static bool handle_trace_discard(struct thread_data *td, struct blk_io_trace *t, unsigned long long ttime, - unsigned long *ios, unsigned int *bs) + unsigned long *ios, unsigned long long *bs, + struct file_cache *cache) { - struct io_piece *ipo = malloc(sizeof(*ipo)); + struct io_piece *ipo; int fileno; + if (td->o.replay_skip & (1u << DDIR_TRIM)) + return false; + + ipo = calloc(1, sizeof(*ipo)); init_ipo(ipo); - fileno = trace_add_file(td, t->device); + fileno = trace_add_file(td, t->device, cache); ios[DDIR_TRIM]++; if (t->bytes > bs[DDIR_TRIM]) @@ -239,13 +221,12 @@ static void handle_trace_discard(struct thread_data *td, td->o.size += t->bytes; - memset(ipo, 0, sizeof(*ipo)); INIT_FLIST_HEAD(&ipo->list); - /* - * the 512 is wrong here, it should be the hardware sector size... - */ 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; @@ -255,56 +236,105 @@ static void handle_trace_discard(struct thread_data *td, ipo->offset, ipo->len, ipo->delay); queue_io_piece(td, ipo); + return true; } -static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t, +static void dump_trace(struct blk_io_trace *t) +{ + log_err("blktrace: ignoring zero byte trace: action=%x\n", t->action); +} + +static bool handle_trace_fs(struct thread_data *td, struct blk_io_trace *t, unsigned long long ttime, unsigned long *ios, - unsigned int *bs) + unsigned long long *bs, struct file_cache *cache) { int rw; int fileno; - fileno = trace_add_file(td, t->device); + fileno = trace_add_file(td, t->device, cache); rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; + if (rw) { + if (td->o.replay_skip & (1u << DDIR_WRITE)) + return false; + } else { + if (td->o.replay_skip & (1u << DDIR_READ)) + return false; + } + + if (!t->bytes) { + if (!fio_did_warn(FIO_WARN_BTRACE_ZERO)) + dump_trace(t); + return false; + } + 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); + return true; +} + +static bool handle_trace_flush(struct thread_data *td, struct blk_io_trace *t, + unsigned long long ttime, unsigned long *ios, + struct file_cache *cache) +{ + struct io_piece *ipo; + int fileno; + + if (td->o.replay_skip & (1u << DDIR_SYNC)) + return false; + + ipo = calloc(1, sizeof(*ipo)); + init_ipo(ipo); + fileno = trace_add_file(td, t->device, cache); + + ipo->delay = ttime / 1000; + ipo->ddir = DDIR_SYNC; + ipo->fileno = fileno; + + ios[DDIR_SYNC]++; + dprint(FD_BLKTRACE, "store flush delay=%lu\n", ipo->delay); + + if (!(td->flags & TD_F_SYNCS)) + td->flags |= TD_F_SYNCS; + + queue_io_piece(td, ipo); + return true; } /* * 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 bool queue_trace(struct thread_data *td, struct blk_io_trace *t, + unsigned long *ios, unsigned long long *bs, + struct file_cache *cache) { - static unsigned long long last_ttime; - unsigned long long delay; + unsigned long long *last_ttime = &td->io_log_last_ttime; + unsigned long long delay = 0; if ((t->action & 0xffff) != __BLK_TA_QUEUE) - return; + return false; 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; - } + delay = delay_since_ttime(td, t->time); + *last_ttime = t->time; } + t_bytes_align(&td->o, t); + if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY)) - handle_trace_notify(t); + return handle_trace_notify(t); else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD)) - handle_trace_discard(td, t, delay, ios, bs); + return handle_trace_discard(td, t, delay, ios, bs, cache); + else if (t->action & BLK_TC_ACT(BLK_TC_FLUSH)) + return handle_trace_flush(td, t, delay, ios, cache); else - handle_trace_fs(td, t, delay, ios, bs); + return handle_trace_fs(td, t, delay, ios, bs, cache); } static void byteswap_trace(struct blk_io_trace *t) @@ -322,7 +352,7 @@ static void byteswap_trace(struct blk_io_trace *t) t->pdu_len = fio_swap16(t->pdu_len); } -static int t_is_write(struct blk_io_trace *t) +static bool t_is_write(struct blk_io_trace *t) { return (t->action & BLK_TC_ACT(BLK_TC_WRITE | BLK_TC_DISCARD)) != 0; } @@ -372,49 +402,82 @@ static void depth_end(struct blk_io_trace *t, int *this_depth, int *depth) * 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) +bool init_blktrace_read(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; + int old_state; - fd = open(filename, O_RDONLY); - if (fd < 0) { + td->io_log_rfile = fopen(filename, "rb"); + if (!td->io_log_rfile) { td_verror(td, errno, "open blktrace file"); - return 1; + goto err; } + td->io_log_blktrace_swap = need_swap; + td->io_log_last_ttime = 0; + td->o.size = 0; - fifo = fifo_alloc(TRACE_FIFO_SIZE); + free_release_files(td); old_state = td_bump_runstate(td, TD_SETTING_UP); - td->o.size = 0; + if (!read_blktrace(td)) { + goto err; + } - for (i = 0; i < DDIR_RWDIR_CNT; i++) { - ios[i] = 0; - rw_bs[i] = 0; - this_depth[i] = 0; - depth[i] = 0; + td_restore_runstate(td, old_state); + + if (!td->files_index) { + log_err("fio: did not find replay device(s)\n"); + return false; + } + + return true; + +err: + if (td->io_log_rfile) { + fclose(td->io_log_rfile); + td->io_log_rfile = NULL; + } + return false; +} + +bool read_blktrace(struct thread_data* td) +{ + struct blk_io_trace t; + struct file_cache cache = { + .maj = ~0U, + .min = ~0U, + }; + unsigned long ios[DDIR_RWDIR_SYNC_CNT] = { }; + unsigned long long rw_bs[DDIR_RWDIR_CNT] = { }; + unsigned long skipped_writes; + FILE *f = td->io_log_rfile; + int i, max_depth; + struct fio_file *fiof; + int this_depth[DDIR_RWDIR_CNT] = { }; + int depth[DDIR_RWDIR_CNT] = { }; + int64_t items_to_fetch = 0; + + if (td->o.read_iolog_chunked) { + items_to_fetch = iolog_items_to_fetch(td); + if (!items_to_fetch) + return true; } skipped_writes = 0; do { - int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t)); + int ret = fread(&t, 1, sizeof(t), f); - if (ret < 0) + if (ferror(f)) { + td_verror(td, errno, "read blktrace file"); goto err; - else if (!ret) + } else if (feof(f)) { break; - else if (ret < (int) sizeof(t)) { - log_err("fio: short fifo get\n"); + } else if (ret < (int) sizeof(t)) { + log_err("fio: iolog short read\n"); break; } - if (need_swap) + if (td->io_log_blktrace_swap) byteswap_trace(&t); if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { @@ -427,12 +490,9 @@ int load_blktrace(struct thread_data *td, const char *filename, int need_swap) t.magic & 0xff); goto err; } - ret = discard_pdu(td, fifo, fd, &t); + ret = discard_pdu(f, &t); if (ret < 0) { - td_verror(td, ret, "blktrace lseek"); - goto err; - } else if (t.pdu_len != ret) { - log_err("fio: discarded %d of %d\n", ret, t.pdu_len); + td_verror(td, -ret, "blktrace lseek"); goto err; } if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) { @@ -450,24 +510,54 @@ int load_blktrace(struct thread_data *td, const char *filename, int need_swap) } } - handle_trace(td, &t, ios, rw_bs); + if (!queue_trace(td, &t, ios, rw_bs, &cache)) + continue; + + if (td->o.read_iolog_chunked) { + td->io_log_current++; + items_to_fetch--; + if (items_to_fetch == 0) + break; + } } while (1); - for (i = 0; i < td->files_index; i++) { - f = td->files[i]; - trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE); + if (td->o.read_iolog_chunked) { + td->io_log_highmark = td->io_log_current; + td->io_log_checkmark = (td->io_log_highmark + 1) / 2; + fio_gettime(&td->io_log_highmark_time, NULL); } - fifo_free(fifo); - close(fd); - - td_restore_runstate(td, old_state); + if (skipped_writes) + log_err("fio: %s skips replay of %lu writes due to read-only\n", + td->o.name, skipped_writes); - if (!td->files_index) { - log_err("fio: did not find replay device(s)\n"); - return 1; + if (td->o.read_iolog_chunked) { + if (td->io_log_current == 0) { + return false; + } + td->o.td_ddir = TD_DDIR_RW; + if ((rw_bs[DDIR_READ] > td->o.max_bs[DDIR_READ] || + rw_bs[DDIR_WRITE] > td->o.max_bs[DDIR_WRITE] || + rw_bs[DDIR_TRIM] > td->o.max_bs[DDIR_TRIM]) && + td->orig_buffer) + { + td->o.max_bs[DDIR_READ] = max(td->o.max_bs[DDIR_READ], rw_bs[DDIR_READ]); + td->o.max_bs[DDIR_WRITE] = max(td->o.max_bs[DDIR_WRITE], rw_bs[DDIR_WRITE]); + td->o.max_bs[DDIR_TRIM] = max(td->o.max_bs[DDIR_TRIM], rw_bs[DDIR_TRIM]); + io_u_quiesce(td); + free_io_mem(td); + if (init_io_u_buffers(td)) + return false; + } + return true; } + for_each_file(td, fiof, i) + trace_add_open_close_event(td, fiof->fileno, FIO_LOG_CLOSE_FILE); + + fclose(td->io_log_rfile); + td->io_log_rfile = NULL; + /* * For stacked devices, we don't always get a COMPLETE event so * the depth grows to insane values. Limit it to something sane(r). @@ -481,43 +571,240 @@ int load_blktrace(struct thread_data *td, const char *filename, int need_swap) max_depth = max(depth[i], max_depth); } - if (skipped_writes) - log_err("fio: %s skips replay of %lu writes due to read-only\n", - td->o.name, skipped_writes); - - if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) { + if (!ios[DDIR_READ] && !ios[DDIR_WRITE] && !ios[DDIR_TRIM] && + !ios[DDIR_SYNC]) { log_err("fio: found no ios in blktrace data\n"); - return 1; - } else if (ios[DDIR_READ] && !ios[DDIR_WRITE]) { - td->o.td_ddir = TD_DDIR_READ; - td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ]; - } else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) { - td->o.td_ddir = TD_DDIR_WRITE; - td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE]; - } else { - td->o.td_ddir = TD_DDIR_RW; + return false; + } + + td->o.td_ddir = 0; + if (ios[DDIR_READ]) { + td->o.td_ddir |= TD_DDIR_READ; td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ]; + } + if (ios[DDIR_WRITE]) { + td->o.td_ddir |= TD_DDIR_WRITE; td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE]; + } + if (ios[DDIR_TRIM]) { + td->o.td_ddir |= TD_DDIR_TRIM; td->o.max_bs[DDIR_TRIM] = rw_bs[DDIR_TRIM]; } - /* - * We need to do direct/raw ios to the device, to avoid getting - * read-ahead in our way. But only do so if the minimum block size - * is a multiple of 4k, otherwise we don't know if it's safe to do so. - */ - if (!fio_option_is_set(&td->o, odirect) && !(td_min_bs(td) & 4095)) - td->o.odirect = 1; - /* * If depth wasn't manually set, use probed depth */ if (!fio_option_is_set(&td->o, iodepth)) td->o.iodepth = td->o.iodepth_low = max_depth; - return 0; + return true; err: - close(fd); - fifo_free(fifo); - return 1; + fclose(f); + return false; +} + +static int init_merge_param_list(fio_fp64_t *vals, struct blktrace_cursor *bcs, + int nr_logs, int def, size_t off) +{ + int i = 0, len = 0; + + while (len < FIO_IO_U_LIST_MAX_LEN && vals[len].u.f != 0.0) + len++; + + if (len && len != nr_logs) + return len; + + for (i = 0; i < nr_logs; i++) { + int *val = (int *)((char *)&bcs[i] + off); + *val = def; + if (len) + *val = (int)vals[i].u.f; + } + + return 0; + +} + +static int find_earliest_io(struct blktrace_cursor *bcs, int nr_logs) +{ + __u64 time = ~(__u64)0; + int idx = 0, i; + + for (i = 0; i < nr_logs; i++) { + if (bcs[i].t.time < time) { + time = bcs[i].t.time; + idx = i; + } + } + + return idx; +} + +static void merge_finish_file(struct blktrace_cursor *bcs, int i, int *nr_logs) +{ + bcs[i].iter++; + if (bcs[i].iter < bcs[i].nr_iter) { + fseek(bcs[i].f, 0, SEEK_SET); + return; + } + + *nr_logs -= 1; + + /* close file */ + fclose(bcs[i].f); + + /* keep active files contiguous */ + memmove(&bcs[i], &bcs[*nr_logs], sizeof(bcs[i])); +} + +static int read_trace(struct thread_data *td, struct blktrace_cursor *bc) +{ + int ret = 0; + struct blk_io_trace *t = &bc->t; + +read_skip: + /* read an io trace */ + ret = fread(&t, 1, sizeof(t), bc->f); + if (ferror(bc->f)) { + td_verror(td, errno, "read blktrace file"); + return ret; + } else if (feof(bc->f)) { + if (!bc->length) + bc->length = bc->t.time; + return ret; + } else if (ret < (int) sizeof(*t)) { + log_err("fio: iolog short read\n"); + return -1; + } + + if (bc->swap) + byteswap_trace(t); + + /* skip over actions that fio does not care about */ + if ((t->action & 0xffff) != __BLK_TA_QUEUE || + t_get_ddir(t) == DDIR_INVAL) { + ret = discard_pdu(bc->f, t); + if (ret < 0) { + td_verror(td, -ret, "blktrace lseek"); + return ret; + } + goto read_skip; + } + + t->time = (t->time + bc->iter * bc->length) * bc->scalar / 100; + + return ret; +} + +static int write_trace(FILE *fp, struct blk_io_trace *t) +{ + /* pdu is not used so just write out only the io trace */ + t->pdu_len = 0; + return fwrite((void *)t, sizeof(*t), 1, fp); +} + +int merge_blktrace_iologs(struct thread_data *td) +{ + int nr_logs = get_max_str_idx(td->o.read_iolog_file); + struct blktrace_cursor *bcs = malloc(sizeof(struct blktrace_cursor) * + nr_logs); + struct blktrace_cursor *bc; + FILE *merge_fp; + char *str, *ptr, *name, *merge_buf; + int i, ret; + + ret = init_merge_param_list(td->o.merge_blktrace_scalars, bcs, nr_logs, + 100, offsetof(struct blktrace_cursor, + scalar)); + if (ret) { + log_err("fio: merge_blktrace_scalars(%d) != nr_logs(%d)\n", + ret, nr_logs); + goto err_param; + } + + ret = init_merge_param_list(td->o.merge_blktrace_iters, bcs, nr_logs, + 1, offsetof(struct blktrace_cursor, + nr_iter)); + if (ret) { + log_err("fio: merge_blktrace_iters(%d) != nr_logs(%d)\n", + ret, nr_logs); + goto err_param; + } + + /* setup output file */ + merge_fp = fopen(td->o.merge_blktrace_file, "w"); + merge_buf = malloc(128 * 1024); + if (!merge_buf) + goto err_out_file; + ret = setvbuf(merge_fp, merge_buf, _IOFBF, 128 * 1024); + if (ret) + goto err_merge_buf; + + /* setup input files */ + str = ptr = strdup(td->o.read_iolog_file); + nr_logs = 0; + for (i = 0; (name = get_next_str(&ptr)) != NULL; i++) { + bcs[i].f = fopen(name, "rb"); + if (!bcs[i].f) { + log_err("fio: could not open file: %s\n", name); + ret = -errno; + free(str); + goto err_file; + } + nr_logs++; + + if (!is_blktrace(name, &bcs[i].swap)) { + log_err("fio: file is not a blktrace: %s\n", name); + free(str); + goto err_file; + } + + ret = read_trace(td, &bcs[i]); + if (ret < 0) { + free(str); + goto err_file; + } else if (!ret) { + merge_finish_file(bcs, i, &nr_logs); + i--; + } + } + free(str); + + /* merge files */ + while (nr_logs) { + i = find_earliest_io(bcs, nr_logs); + bc = &bcs[i]; + /* skip over the pdu */ + ret = discard_pdu(bc->f, &bc->t); + if (ret < 0) { + td_verror(td, -ret, "blktrace lseek"); + goto err_file; + } + + ret = write_trace(merge_fp, &bc->t); + ret = read_trace(td, bc); + if (ret < 0) + goto err_file; + else if (!ret) + merge_finish_file(bcs, i, &nr_logs); + } + + /* set iolog file to read from the newly merged file */ + td->o.read_iolog_file = td->o.merge_blktrace_file; + ret = 0; + +err_file: + /* cleanup */ + for (i = 0; i < nr_logs; i++) { + fclose(bcs[i].f); + } +err_merge_buf: + free(merge_buf); +err_out_file: + fflush(merge_fp); + fclose(merge_fp); +err_param: + free(bcs); + + return ret; }