Fix offset retrieval problem for read and write having different bs ranges

[fio.git] / io_u.c

diff --git a/io_u.c b/io_u.c
index 4750ac600a4272142bc85ccf9333382fcb1a4d28..8f10bfdc0ce3552d1a1f7ccd6f1cdc613bf4ad73 100644 (file)
--- a/io_u.c
+++ b/io_u.c
@@ -72,11 +72,24 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
                io_u->buflen = blocks * min_bs;
 }
 
+static inline unsigned long long last_block(struct thread_data *td,
+                                           struct fio_file *f,
+                                           enum fio_ddir ddir)
+{
+       unsigned long long max_blocks;
+
+       max_blocks = f->io_size / td->o.min_bs[ddir];
+       if (!max_blocks)
+               return 0;
+
+       return max_blocks - 1;
+}
+
 /*
  * Return the next free block in the map.
  */
 static int get_next_free_block(struct thread_data *td, struct fio_file *f,
-                              unsigned long long *b)
+                              enum fio_ddir ddir, unsigned long long *b)
 {
        int i;
 
@@ -85,6 +98,8 @@ static int get_next_free_block(struct thread_data *td, struct fio_file *f,
        while ((*b) * td->o.rw_min_bs < f->real_file_size) {
                if (f->file_map[i] != -1UL) {
                        *b += fio_ffz(f->file_map[i]);
+                       if (*b > last_block(td, f, ddir))
+                               break;
                        f->last_free_lookup = i;
                        return 0;
                }
@@ -93,22 +108,20 @@ static int get_next_free_block(struct thread_data *td, struct fio_file *f,
                i++;
        }
 
+       dprint(FD_IO, "failed finding a free block\n");
        return 1;
 }
 
 static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
-                               int ddir, unsigned long long *b)
+                               enum fio_ddir ddir, unsigned long long *b)
 {
-       unsigned long long max_blocks = f->io_size / td->o.min_bs[ddir];
        unsigned long long r, rb;
        int loops = 5;
 
        do {
                r = os_random_long(&td->random_state);
-               if (!max_blocks)
-                       *b = 0;
-               else
-                       *b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
+               *b = last_block(td, f, ddir);
+
                /*
                 * if we are not maintaining a random map, we are done.
                 */
@@ -118,7 +131,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
                /*
                 * calculate map offset and chec if it's free
                 */
-               rb = *b + (f->file_offset / td->o.min_bs[ddir]);
+               rb = *b;
                if (random_map_free(td, f, rb))
                        return 0;
 
@@ -132,7 +145,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
        loops = 10;
        do {
                f->last_free_lookup = (f->num_maps - 1) * (r / (RAND_MAX+1.0));
-               if (!get_next_free_block(td, f, b))
+               if (!get_next_free_block(td, f, ddir, b))
                        return 0;
 
                r = os_random_long(&td->random_state);
@@ -142,7 +155,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
         * that didn't work either, try exhaustive search from the start
         */
        f->last_free_lookup = 0;
-       return get_next_free_block(td, f, b);
+       return get_next_free_block(td, f, ddir, b);
 }
 
 /*
@@ -153,8 +166,8 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
 static int get_next_offset(struct thread_data *td, struct io_u *io_u)
 {
        struct fio_file *f = io_u->file;
-       const int ddir = io_u->ddir;
        unsigned long long b;
+       enum fio_ddir ddir = io_u->ddir;
 
        if (td_random(td) && (td->o.ddir_nr && !--td->ddir_nr)) {
                td->ddir_nr = td->o.ddir_nr;
@@ -163,15 +176,19 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u)
                        return 1;
        } else {
                if (f->last_pos >= f->real_file_size) {
-                       if (!td_random(td) || get_next_rand_offset(td, f, ddir, &b))
+                       if (!td_random(td) ||
+                            get_next_rand_offset(td, f, ddir, &b))
                                return 1;
                } else
                        b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir];
        }
 
        io_u->offset = (b * td->o.min_bs[ddir]) + f->file_offset;
-       if (io_u->offset >= f->real_file_size)
+       if (io_u->offset >= f->real_file_size) {
+               dprint(FD_IO, "get_next_offset: offset %llu >= size %llu\n",
+                                       io_u->offset, f->real_file_size);
                return 1;
+       }
 
        return 0;
 }
@@ -186,13 +203,30 @@ static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
                buflen = td->o.min_bs[ddir];
        else {
                r = os_random_long(&td->bsrange_state);
-               buflen = (unsigned int) (1 + (double) (td->o.max_bs[ddir] - 1) * r / (RAND_MAX + 1.0));
+               if (!td->o.bssplit_nr)
+                       buflen = (unsigned int) (1 + (double) (td->o.max_bs[ddir] - 1) * r / (RAND_MAX + 1.0));
+               else {
+                       long perc = 0;
+                       unsigned int i;
+
+                       for (i = 0; i < td->o.bssplit_nr; i++) {
+                               struct bssplit *bsp = &td->o.bssplit[i];
+
+                               buflen = bsp->bs;
+                               perc += bsp->perc;
+                               if (r <= ((LONG_MAX / 100L) * perc))
+                                       break;
+                       }
+               }
                if (!td->o.bs_unaligned)
                        buflen = (buflen + td->o.min_bs[ddir] - 1) & ~(td->o.min_bs[ddir] - 1);
        }
 
-       if (io_u->offset + buflen > io_u->file->real_file_size)
+       if (io_u->offset + buflen > io_u->file->real_file_size) {
+               dprint(FD_IO, "lower buflen %u -> %u (ddir=%d)\n", buflen,
+                                               td->o.min_bs[ddir], ddir);
                buflen = td->o.min_bs[ddir];
+       }
 
        return buflen;
 }
@@ -303,6 +337,9 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
        struct io_u *__io_u = *io_u;
 
        __io_u->flags |= IO_U_F_FREE;
+       if ((__io_u->flags & IO_U_F_FLIGHT) && (__io_u->ddir != DDIR_SYNC))
+               td->io_issues[__io_u->ddir]--;
+               
        __io_u->flags &= ~IO_U_F_FLIGHT;
 
        list_del(&__io_u->list);
@@ -313,6 +350,9 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 
 static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 {
+       if (td->io_ops->flags & FIO_NOIO)
+               goto out;
+
        /*
         * see if it's time to sync
         */
@@ -338,15 +378,23 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
         * No log, let the seq/rand engine retrieve the next buflen and
         * position.
         */
-       if (get_next_offset(td, io_u))
+       if (get_next_offset(td, io_u)) {
+               dprint(FD_IO, "io_u %p, failed getting offset\n", io_u);
                return 1;
+       }
 
        io_u->buflen = get_next_buflen(td, io_u);
-       if (!io_u->buflen)
+       if (!io_u->buflen) {
+               dprint(FD_IO, "io_u %p, failed getting buflen\n", io_u);
                return 1;
+       }
 
-       if (io_u->offset + io_u->buflen > io_u->file->real_file_size)
+       if (io_u->offset + io_u->buflen > io_u->file->real_file_size) {
+               dprint(FD_IO, "io_u %p, offset too large\n", io_u);
+               dprint(FD_IO, "  off=%llu/%lu > %llu\n", io_u->offset,
+                               io_u->buflen, io_u->file->real_file_size);
                return 1;
+       }
 
        /*
         * mark entry before potentially trimming io_u
@@ -358,6 +406,8 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
         * If using a write iolog, store this entry.
         */
 out:
+       dprint_io_u(io_u, "fill_io_u");
+       td->zone_bytes += io_u->buflen;
        log_io_u(td, io_u);
        return 0;
 }
@@ -506,8 +556,10 @@ static struct fio_file *get_next_file_rand(struct thread_data *td, int goodf,
                if (f->flags & FIO_FILE_DONE)
                        continue;
 
-               if ((!goodf || (f->flags & goodf)) && !(f->flags & badf))
+               if ((!goodf || (f->flags & goodf)) && !(f->flags & badf)) {
+                       dprint(FD_FILE, "get_next_file_rand: %p\n", f);
                        return f;
+               }
        } while (1);
 }
 
@@ -538,6 +590,7 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
                f = NULL;
        } while (td->next_file != old_next_file);
 
+       dprint(FD_FILE, "get_next_file_rr: %p\n", f);
        return f;
 }
 
@@ -547,12 +600,14 @@ static struct fio_file *get_next_file(struct thread_data *td)
 
        assert(td->o.nr_files <= td->files_index);
 
-       if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files)
+       if (!td->nr_open_files || td->nr_done_files >= td->o.nr_files) {
+               dprint(FD_FILE, "get_next_file: nr_open=%d, nr_done=%d, nr_files=%d\n", td->nr_open_files, td->nr_done_files, td->o.nr_files);
                return NULL;
+       }
 
        f = td->file_service_file;
        if (f && (f->flags & FIO_FILE_OPEN) && td->file_service_left--)
-               return f;
+               goto out;
 
        if (td->o.file_service_type == FIO_FSERVICE_RR)
                f = get_next_file_rr(td, FIO_FILE_OPEN, FIO_FILE_CLOSING);
@@ -561,6 +616,8 @@ static struct fio_file *get_next_file(struct thread_data *td)
 
        td->file_service_file = f;
        td->file_service_left = td->file_service_nr - 1;
+out:
+       dprint(FD_FILE, "get_next_file: %p\n", f);
        return f;
 }
 
@@ -661,8 +718,10 @@ struct io_u *get_io_u(struct thread_data *td)
        struct io_u *io_u;
 
        io_u = __get_io_u(td);
-       if (!io_u)
+       if (!io_u) {
+               dprint(FD_IO, "__get_io_u failed\n");
                return NULL;
+       }
 
        /*
         * from a requeue, io_u already setup
@@ -676,15 +735,19 @@ struct io_u *get_io_u(struct thread_data *td)
        if (td->o.read_iolog_file) {
                if (read_iolog_get(td, io_u))
                        goto err_put;
-       } else if (set_io_u_file(td, io_u))
+       } else if (set_io_u_file(td, io_u)) {
+               dprint(FD_IO, "io_u %p, setting file failed\n", io_u);
                goto err_put;
+       }
        
        f = io_u->file;
        assert(f->flags & FIO_FILE_OPEN);
 
        if (io_u->ddir != DDIR_SYNC) {
-               if (!io_u->buflen)
+               if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
+                       dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
                        goto err_put;
+               }
 
                f->last_pos = io_u->offset + io_u->buflen;
 
@@ -705,6 +768,7 @@ out:
                return io_u;
        }
 err_put:
+       dprint(FD_IO, "get_io_u failed\n");
        put_io_u(td, io_u);
        return NULL;
 }
@@ -731,6 +795,8 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 {
        unsigned long usec;
 
+       dprint_io_u(io_u, "io complete");
+
        assert(io_u->flags & IO_U_F_FLIGHT);
        io_u->flags &= ~IO_U_F_FLIGHT;
 
@@ -748,7 +814,6 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
 
                td->io_blocks[idx]++;
                td->io_bytes[idx] += bytes;
-               td->zone_bytes += bytes;
                td->this_io_bytes[idx] += bytes;
 
                io_u->file->last_completed_pos = io_u->endpos;
@@ -760,6 +825,7 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
                io_u_mark_latency(td, usec);
 
                if (td_write(td) && idx == DDIR_WRITE &&
+                   td->o.do_verify &&
                    td->o.verify != VERIFY_NONE)
                        log_io_piece(td, io_u);
 
@@ -828,6 +894,8 @@ long io_u_queued_complete(struct thread_data *td, int min_events)
        int ret;
        struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
+       dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_events);
+
        if (!min_events)
                tvp = &ts;