Initial suppor for sync_file_range()
[fio.git] / log.c
diff --git a/log.c b/log.c
index 63c0a332d6bb6680998cbfd81a186d0ff6f24a3e..99f20b53083b29ae4399827ff677b89e1d16781f 100644 (file)
--- a/log.c
+++ b/log.c
@@ -4,23 +4,26 @@
  */
 #include <stdio.h>
 #include <stdlib.h>
+#include <libgen.h>
 #include <assert.h>
-#include "list.h"
+#include "flist.h"
 #include "fio.h"
+#include "verify.h"
 
 static const char iolog_ver2[] = "fio version 2 iolog";
 
 void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
 {
-       list_add_tail(&ipo->list, &td->io_log_list);
+       flist_add_tail(&ipo->list, &td->io_log_list);
        td->total_io_size += ipo->len;
 }
 
 void log_io_u(struct thread_data *td, struct io_u *io_u)
 {
-       const char *act[] = { "read", "write", "sync" };
+       const char *act[] = { "read", "write", "sync", "datasync",
+                               "sync_file_range" };
 
-       assert(io_u->ddir < 3);
+       assert(io_u->ddir <= 4);
 
        if (!td->o.write_iolog_file)
                return;
@@ -40,6 +43,13 @@ void log_file(struct thread_data *td, struct fio_file *f,
        if (!td->o.write_iolog_file)
                return;
 
+
+       /*
+        * this happens on the pre-open/close done before the job starts
+        */
+       if (!td->iolog_f)
+               return;
+
        fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
 }
 
@@ -77,10 +87,8 @@ static int ipo_special(struct thread_data *td, struct io_piece *ipo)
        switch (ipo->file_action) {
        case FIO_LOG_OPEN_FILE:
                ret = td_io_open_file(td, f);
-               if (!ret) {
-                       free(ipo);
+               if (!ret)
                        break;
-               }
                td_verror(td, ret, "iolog open file");
                return -1;
        case FIO_LOG_CLOSE_FILE:
@@ -100,12 +108,13 @@ static int ipo_special(struct thread_data *td, struct io_piece *ipo)
 int read_iolog_get(struct thread_data *td, struct io_u *io_u)
 {
        struct io_piece *ipo;
-
-       while (!list_empty(&td->io_log_list)) {
+       unsigned long elapsed;
+       
+       while (!flist_empty(&td->io_log_list)) {
                int ret;
 
-               ipo = list_entry(td->io_log_list.next, struct io_piece, list);
-               list_del(&ipo->list);
+               ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
+               flist_del(&ipo->list);
 
                ret = ipo_special(td, ipo);
                if (ret < 0) {
@@ -116,20 +125,27 @@ int read_iolog_get(struct thread_data *td, struct io_u *io_u)
                        continue;
                }
 
-               io_u->offset = ipo->offset;
-               io_u->buflen = ipo->len;
                io_u->ddir = ipo->ddir;
-               io_u->file = td->files[ipo->fileno];
-               get_file(io_u->file);
-
-               dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
-                                       io_u->buflen, io_u->file->file_name);
-
-               if (ipo->delay)
-                       iolog_delay(td, ipo->delay);
+               if (ipo->ddir != DDIR_WAIT) {
+                       io_u->offset = ipo->offset;
+                       io_u->buflen = ipo->len;
+                       io_u->file = td->files[ipo->fileno];
+                       get_file(io_u->file);
+
+                       dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
+                                               io_u->buflen, io_u->file->file_name);
+                       if (ipo->delay) iolog_delay(td, ipo->delay);
+               } else {
+                       elapsed = mtime_since_genesis();
+                       if (ipo->delay > elapsed)
+                               usec_sleep(td, (ipo->delay - elapsed) * 1000);
+                               
+               }
 
                free(ipo);
-               return 0;
+               
+               if (ipo->ddir != DDIR_WAIT)
+                       return 0;
        }
 
        td->done = 1;
@@ -147,9 +163,9 @@ void prune_io_piece_log(struct thread_data *td)
                free(ipo);
        }
 
-       while (!list_empty(&td->io_hist_list)) {
-               ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
-               list_del(&ipo->list);
+       while (!flist_empty(&td->io_hist_list)) {
+               ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
+               flist_del(&ipo->list);
                free(ipo);
        }
 }
@@ -175,28 +191,40 @@ void log_io_piece(struct thread_data *td, struct io_u *io_u)
         *
         * For both these cases, just reading back data in the order we
         * wrote it out is the fastest.
+        *
+        * One exception is if we don't have a random map AND we are doing
+        * verifies, in that case we need to check for duplicate blocks and
+        * drop the old one, which we rely on the rb insert/lookup for
+        * handling.
         */
-       if (!td_random(td) || !td->o.overwrite) {
-               INIT_LIST_HEAD(&ipo->list);
-               list_add_tail(&ipo->list, &td->io_hist_list);
+       if ((!td_random(td) || !td->o.overwrite) &&
+             (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
+               INIT_FLIST_HEAD(&ipo->list);
+               flist_add_tail(&ipo->list, &td->io_hist_list);
                return;
        }
 
        RB_CLEAR_NODE(&ipo->rb_node);
-       p = &td->io_hist_tree.rb_node;
-       parent = NULL;
 
        /*
         * Sort the entry into the verification list
         */
+restart:
+       p = &td->io_hist_tree.rb_node;
+       parent = NULL;
        while (*p) {
                parent = *p;
 
                __ipo = rb_entry(parent, struct io_piece, rb_node);
-               if (ipo->offset <= __ipo->offset)
+               if (ipo->offset < __ipo->offset)
                        p = &(*p)->rb_left;
-               else
+               else if (ipo->offset > __ipo->offset)
                        p = &(*p)->rb_right;
+               else {
+                       assert(ipo->len == __ipo->len);
+                       rb_erase(parent, &td->io_hist_tree);
+                       goto restart;
+               }
        }
 
        rb_link_node(&ipo->rb_node, parent, p);
@@ -220,7 +248,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
 {
        unsigned long long offset;
        unsigned int bytes;
-       int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
+       int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
        char *fname, *act;
        char *str, *p;
        enum fio_ddir rw;
@@ -235,7 +263,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
        fname = malloc(256+16);
        act = malloc(256+16);
 
-       reads = writes = 0;
+       reads = writes = waits = 0;
        while ((p = fgets(str, 4096, f)) != NULL) {
                struct io_piece *ipo;
                int r;
@@ -246,12 +274,16 @@ static int read_iolog2(struct thread_data *td, FILE *f)
                        /*
                         * Check action first
                         */
-                       if (!strcmp(act, "read"))
+                       if (!strcmp(act, "wait"))
+                               rw = DDIR_WAIT;
+                       else if (!strcmp(act, "read"))
                                rw = DDIR_READ;
                        else if (!strcmp(act, "write"))
                                rw = DDIR_WRITE;
                        else if (!strcmp(act, "sync"))
                                rw = DDIR_SYNC;
+                       else if (!strcmp(act, "datasync"))
+                               rw = DDIR_DATASYNC;
                        else {
                                log_err("fio: bad iolog file action: %s\n",
                                                                        act);
@@ -289,7 +321,10 @@ static int read_iolog2(struct thread_data *td, FILE *f)
                        if (read_only)
                                continue;
                        writes++;
-               } else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
+               } else if (rw == DDIR_WAIT) {
+                       waits++;
+               } else if (rw == DDIR_INVAL) {
+               } else if (!ddir_sync(rw)) {
                        log_err("bad ddir: %d\n", rw);
                        continue;
                }
@@ -299,16 +334,19 @@ static int read_iolog2(struct thread_data *td, FILE *f)
                 */
                ipo = malloc(sizeof(*ipo));
                memset(ipo, 0, sizeof(*ipo));
-               INIT_LIST_HEAD(&ipo->list);
-               ipo->offset = offset;
-               ipo->len = bytes;
+               INIT_FLIST_HEAD(&ipo->list);
                ipo->ddir = rw;
-               if (bytes > td->o.max_bs[rw])
-                       td->o.max_bs[rw] = bytes;
-               if (rw == DDIR_INVAL) {
+               if (rw == DDIR_WAIT) {
+                       ipo->delay = offset;
+               } else {
+                       ipo->offset = offset;
+                       ipo->len = bytes;
+                       if (bytes > td->o.max_bs[rw])
+                               td->o.max_bs[rw] = bytes;
                        ipo->fileno = fileno;
                        ipo->file_action = file_action;
                }
+                       
                queue_io_piece(td, ipo);
        }
 
@@ -322,7 +360,7 @@ static int read_iolog2(struct thread_data *td, FILE *f)
                writes = 0;
        }
 
-       if (!reads && !writes)
+       if (!reads && !writes && !waits)
                return 1;
        else if (reads && !writes)
                td->o.td_ddir = TD_DDIR_READ;
@@ -334,15 +372,6 @@ static int read_iolog2(struct thread_data *td, FILE *f)
        return 0;
 }
 
-/*
- * Read version 1 iolog data.
- */
-static int read_iolog(struct thread_data *td, FILE *f)
-{
-       log_err("fio: iolog version 1 is no longer supported\n");
-       return 1;
-}
-
 /*
  * open iolog, check version, and call appropriate parser
  */
@@ -372,16 +401,8 @@ static int init_iolog_read(struct thread_data *td)
        if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
                ret = read_iolog2(td, f);
        else {
-               /*
-                * seek back to the beginning
-                */
-               if (fseek(f, 0, SEEK_SET) < 0) {
-                       td_verror(td, errno, "iolog read");
-                       log_err("fio: unable to read iolog\n");
-                       return 1;
-               }
-
-               ret = read_iolog(td, f);
+               log_err("fio: iolog version 1 is no longer supported\n");
+               ret = 1;
        }
 
        fclose(f);
@@ -397,7 +418,7 @@ static int init_iolog_write(struct thread_data *td)
        FILE *f;
        unsigned int i;
 
-       f = fopen(td->o.write_iolog_file, "w+");
+       f = fopen(td->o.write_iolog_file, "a");
        if (!f) {
                perror("fopen write iolog");
                return 1;
@@ -431,9 +452,6 @@ int init_iolog(struct thread_data *td)
 {
        int ret = 0;
 
-       if (td->io_ops->flags & FIO_DISKLESSIO)
-               return 0;
-
        if (td->o.read_iolog_file) {
                /*
                 * Check if it's a blktrace file and load that if possible.
@@ -464,15 +482,17 @@ void __finish_log(struct io_log *log, const char *name)
        unsigned int i;
        FILE *f;
 
-       f = fopen(name, "w");
+       f = fopen(name, "a");
        if (!f) {
                perror("fopen log");
                return;
        }
 
        for (i = 0; i < log->nr_samples; i++) {
-               fprintf(f, "%lu, %lu, %u\n", log->log[i].time, log->log[i].val,
-                                               log->log[i].ddir);
+               fprintf(f, "%lu, %lu, %u, %u\n", log->log[i].time,
+                                               log->log[i].val,
+                                               log->log[i].ddir,
+                                               log->log[i].bs);
        }
 
        fclose(f);
@@ -480,10 +500,17 @@ void __finish_log(struct io_log *log, const char *name)
        free(log);
 }
 
-void finish_log(struct thread_data *td, struct io_log *log, const char *name)
+void finish_log_named(struct thread_data *td, struct io_log *log,
+                      const char *prefix, const char *postfix)
 {
-       char file_name[256];
+       char file_name[256], *p;
 
-       snprintf(file_name, 200, "client%d_%s.log", td->thread_number, name);
-       __finish_log(log, file_name);
+       snprintf(file_name, 200, "%s_%s.log", prefix, postfix);
+       p = basename(file_name);
+       __finish_log(log, p);
+}
+
+void finish_log(struct thread_data *td, struct io_log *log, const char *name)
+{
+       finish_log_named(td, log, td->o.name, name);
 }