2 * Blktrace replay utility - Play traces back
4 * Copyright (C) 2007 Alan D. Brunelle <Alan.Brunelle@hp.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 static char build_date[] = __DATE__ " at "__TIME__;
35 #include <sys/param.h>
38 #include <sys/types.h>
42 #if !defined(_GNU_SOURCE)
51 * ========================================================================
52 * ==== STRUCTURE DEFINITIONS =============================================
53 * ========================================================================
57 * Each device map has one of these:
59 * @head: Linked on to map_devs
60 * @from_dev: Device name as seen on recorded system
61 * @to_dev: Device name to be used on replay system
64 struct list_head head;
65 char *from_dev, *to_dev;
69 * Each device name specified has one of these (until threads are created)
71 * @head: Linked onto input_devs
72 * @devnm: Device name -- 'sd*'
75 struct list_head head;
80 * Per input file information
82 * @head: Used to link up on input_files
83 * @free_iocbs: List of free iocb's available for use
84 * @used_iocbs: List of iocb's currently outstanding
85 * @mutex: Mutex used with condition variable to protect volatile values
86 * @cond: Condition variable used when waiting on a volatile value change
87 * @naios_out: Current number of AIOs outstanding on this context
88 * @naios_free: Number of AIOs on the free list (short cut for list_len)
89 * @send_wait: Boolean: When true, the sub thread is waiting on free IOCBs
90 * @reap_wait: Boolean: When true, the rec thread is waiting on used IOCBs
91 * @send_done: Boolean: When true, the sub thread has completed work
92 * @reap_done: Boolean: When true, the rec thread has completed work
93 * @sub_thread: Thread used to submit IOs.
94 * @rec_thread: Thread used to reclaim IOs.
96 * @devnm: Copy of the device name being managed by this thread
97 * @file_name: Full name of the input file
98 * @cpu: CPU this thread is pinned to
99 * @ifd: Input file descriptor
100 * @ofd: Output file descriptor
101 * @iterations: Remaining iterations to process
102 * @vfp: For verbose dumping of actions performed
105 struct list_head head, free_iocbs, used_iocbs;
106 pthread_mutex_t mutex;
108 volatile long naios_out, naios_free;
109 volatile int send_wait, reap_wait, send_done, reap_done;
110 pthread_t sub_thread, rec_thread;
112 char *devnm, *file_name;
113 int cpu, ifd, ofd, iterations;
118 * Every Asynchronous IO used has one of these (naios per file/device).
120 * @iocb: IOCB sent down via io_submit
121 * @head: Linked onto file_list.free_iocbs or file_list.used_iocbs
122 * @tip: Pointer to per-thread information this IO is associated with
123 * @nbytes: Number of bytes in buffer associated with iocb
127 struct list_head head;
128 struct thr_info *tip;
133 * ========================================================================
134 * ==== GLOBAL VARIABLES ==================================================
135 * ========================================================================
138 static volatile int signal_done = 0; // Boolean: Signal'ed, need to quit
140 static char *ibase = "replay"; // Input base name
141 static char *idir = "."; // Input directory base
142 static int cpus_to_use = -1; // Number of CPUs to use
143 static int def_iterations = 1; // Default number of iterations
144 static int naios = 512; // Number of AIOs per thread
145 static int ncpus = 0; // Number of CPUs in the system
146 static int verbose = 0; // Boolean: Output some extra info
147 static int write_enabled = 0; // Boolean: Enable writing
148 static __u64 genesis = ~0; // Earliest time seen
149 static __u64 rgenesis; // Our start time
150 static size_t pgsize; // System Page size
151 static int nb_sec = 512; // Number of bytes per sector
152 static LIST_HEAD(input_devs); // List of devices to handle
153 static LIST_HEAD(input_files); // List of input files to handle
154 static LIST_HEAD(map_devs); // List of device maps
155 static int nfiles = 0; // Number of files to handle
156 static int no_stalls = 0; // Boolean: Disable pre-stalls
157 static int find_records = 0; // Boolean: Find record files auto
160 * Variables managed under control of condition variables.
162 * n_reclaims_done: Counts number of reclaim threads that have completed.
163 * n_replays_done: Counts number of replay threads that have completed.
164 * n_replays_ready: Counts number of replay threads ready to start.
165 * n_iters_done: Counts number of replay threads done one iteration.
166 * iter_start: Starts an iteration for the replay threads.
168 static volatile int n_reclaims_done = 0;
169 static pthread_mutex_t reclaim_done_mutex = PTHREAD_MUTEX_INITIALIZER;
170 static pthread_cond_t reclaim_done_cond = PTHREAD_COND_INITIALIZER;
172 static volatile int n_replays_done = 0;
173 static pthread_mutex_t replay_done_mutex = PTHREAD_MUTEX_INITIALIZER;
174 static pthread_cond_t replay_done_cond = PTHREAD_COND_INITIALIZER;
176 static volatile int n_replays_ready = 0;
177 static pthread_mutex_t replay_ready_mutex = PTHREAD_MUTEX_INITIALIZER;
178 static pthread_cond_t replay_ready_cond = PTHREAD_COND_INITIALIZER;
180 static volatile int n_iters_done = 0;
181 static pthread_mutex_t iter_done_mutex = PTHREAD_MUTEX_INITIALIZER;
182 static pthread_cond_t iter_done_cond = PTHREAD_COND_INITIALIZER;
184 static volatile int iter_start = 0;
185 static pthread_mutex_t iter_start_mutex = PTHREAD_MUTEX_INITIALIZER;
186 static pthread_cond_t iter_start_cond = PTHREAD_COND_INITIALIZER;
189 * ========================================================================
190 * ==== FORWARD REFERENECES ===============================================
191 * ========================================================================
194 static void *replay_sub(void *arg);
195 static void *replay_rec(void *arg);
196 static char usage_str[];
199 * ========================================================================
200 * ==== INLINE ROUTINES ===================================================
201 * ========================================================================
205 * The 'fatal' macro will output a perror message (if errstring is !NULL)
206 * and display a string (with variable arguments) and then exit with the
207 * specified exit value.
210 #define ERR_SYSCALL 2
211 static inline void fatal(const char *errstring, const int exitval,
212 const char *fmt, ...)
220 vfprintf(stderr, fmt, ap);
227 static inline long long unsigned du64_to_sec(__u64 du64)
229 return (long long unsigned)du64 / (1000 * 1000 * 1000);
232 static inline long long unsigned du64_to_nsec(__u64 du64)
234 return llabs((long long)du64) % (1000 * 1000 * 1000);
238 * min - Return minimum of two integers
240 static inline int min(int a, int b)
242 return a < b ? a : b;
246 * minl - Return minimum of two longs
248 static inline long minl(long a, long b)
250 return a < b ? a : b;
254 * usage - Display usage string and version
256 static inline void usage(void)
258 fprintf(stderr, "Usage: btreplay -- version %s\n%s",
259 my_btversion, usage_str);
263 * is_send_done - Returns true if sender should quit early
264 * @tip: Per-thread information
266 static inline int is_send_done(struct thr_info *tip)
268 return signal_done || tip->send_done;
272 * is_reap_done - Returns true if reaper should quit early
273 * @tip: Per-thread information
275 static inline int is_reap_done(struct thr_info *tip)
277 return tip->send_done && tip->naios_out == 0;
281 * ts2ns - Convert timespec values to a nanosecond value
283 #define NS_TICKS ((__u64)1000 * (__u64)1000 * (__u64)1000)
284 static inline __u64 ts2ns(struct timespec *ts)
286 return ((__u64)(ts->tv_sec) * NS_TICKS) + (__u64)(ts->tv_nsec);
290 * ts2ns - Convert timeval values to a nanosecond value
292 static inline __u64 tv2ns(struct timeval *tp)
294 return ((__u64)(tp->tv_sec)) + ((__u64)(tp->tv_usec) * (__u64)1000);
298 * touch_memory - Force physical memory to be allocating it
300 * For malloc()ed memory we need to /touch/ it to make it really
301 * exist. Otherwise, for write's (to storage) things may not work
302 * as planned - we see Linux just use a single area to /read/ from
303 * (as there isn't any memory that has been associated with the
304 * allocated virtual addresses yet).
306 static inline void touch_memory(char *buf, size_t bsize)
308 #if defined(PREP_BUFS)
309 memset(buf, 0, bsize);
313 for (i = 0; i < bsize; i += pgsize)
319 * buf_alloc - Returns a page-aligned buffer of the specified size
320 * @nbytes: Number of bytes to allocate
322 static inline void *buf_alloc(size_t nbytes)
326 if (posix_memalign(&buf, pgsize, nbytes)) {
327 fatal("posix_memalign", ERR_SYSCALL, "Allocation failed\n");
335 * gettime - Returns current time
337 static inline __u64 gettime(void)
339 static int use_clock_gettime = -1; // Which clock to use
341 if (use_clock_gettime < 0) {
342 use_clock_gettime = clock_getres(CLOCK_MONOTONIC, NULL) == 0;
343 if (use_clock_gettime) {
344 struct timespec ts = {
348 clock_settime(CLOCK_MONOTONIC, &ts);
352 if (use_clock_gettime) {
354 clock_gettime(CLOCK_MONOTONIC, &ts);
359 gettimeofday(&tp, NULL);
365 * setup_signal - Set up a signal handler for the specified signum
367 static inline void setup_signal(int signum, sighandler_t handler)
369 if (signal(signum, handler) == SIG_ERR) {
370 fatal("signal", ERR_SYSCALL, "Failed to set signal %d\n",
377 * ========================================================================
378 * ==== CONDITION VARIABLE ROUTINES =======================================
379 * ========================================================================
383 * __set_cv - Increments a variable under condition variable control.
384 * @pmp: Pointer to the associated mutex
385 * @pcp: Pointer to the associated condition variable
386 * @vp: Pointer to the variable being incremented
387 * @mxv: Max value for variable (Used only when ASSERTS are on)
389 static inline void __set_cv(pthread_mutex_t *pmp, pthread_cond_t *pcp,
391 __attribute__((__unused__))int mxv)
393 pthread_mutex_lock(pmp);
396 pthread_cond_signal(pcp);
397 pthread_mutex_unlock(pmp);
401 * __wait_cv - Waits for a variable under cond var control to hit a value
402 * @pmp: Pointer to the associated mutex
403 * @pcp: Pointer to the associated condition variable
404 * @vp: Pointer to the variable being incremented
405 * @mxv: Value to wait for
407 static inline void __wait_cv(pthread_mutex_t *pmp, pthread_cond_t *pcp,
408 volatile int *vp, int mxv)
410 pthread_mutex_lock(pmp);
412 pthread_cond_wait(pcp, pmp);
414 pthread_mutex_unlock(pmp);
417 static inline void set_reclaim_done(void)
419 __set_cv(&reclaim_done_mutex, &reclaim_done_cond, &n_reclaims_done,
423 static inline void wait_reclaims_done(void)
425 __wait_cv(&reclaim_done_mutex, &reclaim_done_cond, &n_reclaims_done,
429 static inline void set_replay_ready(void)
431 __set_cv(&replay_ready_mutex, &replay_ready_cond, &n_replays_ready,
435 static inline void wait_replays_ready(void)
437 __wait_cv(&replay_ready_mutex, &replay_ready_cond, &n_replays_ready,
441 static inline void set_replay_done(void)
443 __set_cv(&replay_done_mutex, &replay_done_cond, &n_replays_done,
447 static inline void wait_replays_done(void)
449 __wait_cv(&replay_done_mutex, &replay_done_cond, &n_replays_done,
453 static inline void set_iter_done(void)
455 __set_cv(&iter_done_mutex, &iter_done_cond, &n_iters_done,
459 static inline void wait_iters_done(void)
461 __wait_cv(&iter_done_mutex, &iter_done_cond, &n_iters_done,
466 * wait_iter_start - Wait for an iteration to start
468 * This is /slightly/ different: we are waiting for a value to become
469 * non-zero, and then we decrement it and go on.
471 static inline void wait_iter_start(void)
473 pthread_mutex_lock(&iter_start_mutex);
474 while (iter_start == 0)
475 pthread_cond_wait(&iter_start_cond, &iter_start_mutex);
476 assert(1 <= iter_start && iter_start <= nfiles);
478 pthread_mutex_unlock(&iter_start_mutex);
482 * start_iter - Start an iteration at the replay thread level
484 static inline void start_iter(void)
486 pthread_mutex_lock(&iter_start_mutex);
487 assert(iter_start == 0);
489 pthread_cond_broadcast(&iter_start_cond);
490 pthread_mutex_unlock(&iter_start_mutex);
494 * ========================================================================
495 * ==== CPU RELATED ROUTINES ==============================================
496 * ========================================================================
500 * get_ncpus - Sets up the global 'ncpus' value
502 static void get_ncpus(void)
506 if (sched_getaffinity(getpid(), sizeof(cpus), &cpus)) {
507 fatal("sched_getaffinity", ERR_SYSCALL, "Can't get CPU info\n");
512 * XXX This assumes (perhaps wrongly) that there are no /holes/
515 for (ncpus = 0; ncpus < CPU_SETSIZE && CPU_ISSET(ncpus, &cpus); ncpus++)
518 fatal(NULL, ERR_SYSCALL, "Insufficient number of CPUs\n");
524 * pin_to_cpu - Pin this thread to a specific CPU
525 * @tip: Thread information
527 static void pin_to_cpu(struct thr_info *tip)
531 assert(0 <= tip->cpu && tip->cpu < ncpus);
534 CPU_SET(tip->cpu, &cpus);
535 if (sched_setaffinity(getpid(), sizeof(cpus), &cpus)) {
536 fatal("sched_setaffinity", ERR_SYSCALL, "Failed to pin CPU\n");
544 (void)sched_getaffinity(getpid(), sizeof(now), &now);
545 fprintf(tip->vfp, "Pinned to CPU %02d ", tip->cpu);
546 for (i = 0; i < ncpus; i++)
547 fprintf(tip->vfp, "%1d", CPU_ISSET(i, &now));
548 fprintf(tip->vfp, "\n");
553 * ========================================================================
554 * ==== INPUT DEVICE HANDLERS =============================================
555 * ========================================================================
559 * add_input_dev - Add a device ('sd*') to the list of devices to handle
561 static void add_input_dev(char *devnm)
564 struct dev_info *dip;
566 __list_for_each(p, &input_devs) {
567 dip = list_entry(p, struct dev_info, head);
568 if (strcmp(dip->devnm, devnm) == 0)
572 dip = malloc(sizeof(*dip));
573 dip->devnm = strdup(devnm);
574 list_add_tail(&dip->head, &input_devs);
578 * rem_input_dev - Remove resources associated with this device
580 static void rem_input_dev(struct dev_info *dip)
582 list_del(&dip->head);
587 static void find_input_devs(char *idir)
590 DIR *dir = opendir(idir);
593 fatal(idir, ERR_ARGS, "Unable to open %s\n", idir);
597 while ((ent = readdir(dir)) != NULL) {
598 char *p, *dsf = malloc(256);
600 if (strstr(ent->d_name, ".replay.") == NULL)
603 dsf = strdup(ent->d_name);
615 * ========================================================================
616 * ==== MAP DEVICE INTERFACES =============================================
617 * ========================================================================
621 * read_map_devs - Read in a set of device mapping from the provided file.
622 * @file_name: File containing device maps
624 * We support the notion of multiple such files being specifed on the cmd line
626 static void read_map_devs(char *file_name)
629 char *from_dev, *to_dev;
631 fp = fopen(file_name, "r");
633 fatal(file_name, ERR_SYSCALL, "Could not open map devs file\n");
637 while (fscanf(fp, "%as %as", &from_dev, &to_dev) == 2) {
638 struct map_dev *mdp = malloc(sizeof(*mdp));
640 mdp->from_dev = from_dev;
641 mdp->to_dev = to_dev;
642 list_add_tail(&mdp->head, &map_devs);
649 * release_map_devs - Release resources associated with device mappings.
651 static void release_map_devs(void)
653 struct list_head *p, *q;
655 list_for_each_safe(p, q, &map_devs) {
656 struct map_dev *mdp = list_entry(p, struct map_dev, head);
658 list_del(&mdp->head);
667 * map_dev - Return the mapped device for that specified
668 * @from_dev: Device name as seen on recorded system
670 * Note: If there is no such mapping, we return the same name.
672 static char *map_dev(char *from_dev)
676 __list_for_each(p, &map_devs) {
677 struct map_dev *mdp = list_entry(p, struct map_dev, head);
679 if (strcmp(from_dev, mdp->from_dev) == 0)
687 * ========================================================================
688 * ==== IOCB MANAGEMENT ROUTINES ==========================================
689 * ========================================================================
693 * iocb_init - Initialize the fields of an IOCB
694 * @tip: Per-thread information
695 * iocbp: IOCB pointer to update
697 static void iocb_init(struct thr_info *tip, struct iocb_pkt *iocbp)
701 iocbp->iocb.u.c.buf = NULL;
705 * iocb_setup - Set up an iocb with this AIOs information
706 * @iocbp: IOCB pointer to update
707 * @rw: Direction (0 == write, 1 == read)
708 * @n: Number of bytes to transfer
709 * @off: Offset (in bytes)
711 static void iocb_setup(struct iocb_pkt *iocbp, int rw, int n, long long off)
714 struct iocb *iop = &iocbp->iocb;
716 assert(rw == 0 || rw == 1);
717 assert(0 < n && (n % nb_sec) == 0);
721 if (iocbp->nbytes >= n) {
726 assert(iop->u.c.buf);
735 io_prep_pread(iop, iocbp->tip->ofd, buf, n, off);
737 assert(write_enabled);
738 io_prep_pwrite(iop, iocbp->tip->ofd, buf, n, off);
739 touch_memory(buf, n);
746 * ========================================================================
747 * ==== PER-THREAD SET UP & TEAR DOWN =====================================
748 * ========================================================================
752 * tip_init - Per thread initialization function
754 static void tip_init(struct thr_info *tip)
758 INIT_LIST_HEAD(&tip->free_iocbs);
759 INIT_LIST_HEAD(&tip->used_iocbs);
761 pthread_mutex_init(&tip->mutex, NULL);
762 pthread_cond_init(&tip->cond, NULL);
764 if (io_setup(naios, &tip->ctx)) {
765 fatal("io_setup", ERR_SYSCALL, "io_setup failed\n");
771 tip->send_done = tip->reap_done = 0;
772 tip->send_wait = tip->reap_wait = 0;
774 memset(&tip->sub_thread, 0, sizeof(tip->sub_thread));
775 memset(&tip->rec_thread, 0, sizeof(tip->rec_thread));
777 for (i = 0; i < naios; i++) {
778 struct iocb_pkt *iocbp = buf_alloc(sizeof(*iocbp));
780 iocb_init(tip, iocbp);
781 list_add_tail(&iocbp->head, &tip->free_iocbs);
783 tip->naios_free = naios;
788 sprintf(fn, "%s/%s.%s.%d.rep", idir, tip->devnm, ibase,
790 tip->vfp = fopen(fn, "w");
792 fatal(fn, ERR_SYSCALL, "Failed to open report\n");
796 setlinebuf(tip->vfp);
799 if (pthread_create(&tip->sub_thread, NULL, replay_sub, tip)) {
800 fatal("pthread_create", ERR_SYSCALL,
801 "thread create failed\n");
805 if (pthread_create(&tip->rec_thread, NULL, replay_rec, tip)) {
806 fatal("pthread_create", ERR_SYSCALL,
807 "thread create failed\n");
813 * tip_release - Release resources associated with this thread
815 static void tip_release(struct thr_info *tip)
817 struct list_head *p, *q;
819 assert(tip->send_done);
820 assert(tip->reap_done);
821 assert(list_len(&tip->used_iocbs) == 0);
822 assert(tip->naios_free == naios);
824 if (pthread_join(tip->sub_thread, NULL)) {
825 fatal("pthread_join", ERR_SYSCALL, "pthread sub join failed\n");
828 if (pthread_join(tip->rec_thread, NULL)) {
829 fatal("pthread_join", ERR_SYSCALL, "pthread rec join failed\n");
833 io_destroy(tip->ctx);
835 list_splice(&tip->used_iocbs, &tip->free_iocbs);
836 list_for_each_safe(p, q, &tip->free_iocbs) {
837 struct iocb_pkt *iocbp = list_entry(p, struct iocb_pkt, head);
839 list_del(&iocbp->head);
841 free(iocbp->iocb.u.c.buf);
845 pthread_cond_destroy(&tip->cond);
846 pthread_mutex_destroy(&tip->mutex);
850 * add_input_file - Allocate and initialize per-input file structure
851 * @cpu: CPU for this file
852 * @devnm: Device name for this file
853 * @file_name: Fully qualifed input file name
855 static void add_input_file(int cpu, char *devnm, char *file_name)
858 struct io_file_hdr hdr;
859 struct thr_info *tip = buf_alloc(sizeof(*tip));
860 __u64 my_version = mk_btversion(btver_mjr, btver_mnr, btver_sub);
862 assert(0 <= cpu && cpu < ncpus);
864 memset(&hdr, 0, sizeof(hdr));
865 memset(tip, 0, sizeof(*tip));
866 tip->cpu = cpu % cpus_to_use;
867 tip->iterations = def_iterations;
869 tip->ifd = open(file_name, O_RDONLY);
871 fatal(file_name, ERR_ARGS, "Unable to open\n");
874 if (fstat(tip->ifd, &buf) < 0) {
875 fatal(file_name, ERR_SYSCALL, "fstat failed\n");
878 if (buf.st_size < (off_t)sizeof(hdr)) {
880 fprintf(stderr, "\t%s empty\n", file_name);
884 if (read(tip->ifd, &hdr, sizeof(hdr)) != sizeof(hdr)) {
885 fatal(file_name, ERR_ARGS, "Header read failed\n");
889 if (hdr.version != my_version) {
890 fprintf(stderr, "%llx %llx %llx %llx\n",
891 (long long unsigned)hdr.version,
892 (long long unsigned)hdr.genesis,
893 (long long unsigned)hdr.nbunches,
894 (long long unsigned)hdr.total_pkts);
895 fatal(NULL, ERR_ARGS,
896 "BT version mismatch: %lx versus my %lx\n",
897 (long)hdr.version, (long)my_version);
901 if (hdr.nbunches == 0) {
908 if (hdr.genesis < genesis) {
910 fprintf(stderr, "Setting genesis to %llu.%llu\n",
911 du64_to_sec(hdr.genesis),
912 du64_to_nsec(hdr.genesis));
913 genesis = hdr.genesis;
916 tip->devnm = strdup(devnm);
917 tip->file_name = strdup(file_name);
919 list_add_tail(&tip->head, &input_files);
922 fprintf(stderr, "Added %s %llu\n", file_name,
923 (long long)hdr.genesis);
927 * rem_input_file - Release resources associated with an input file
928 * @tip: Per-input file information
930 static void rem_input_file(struct thr_info *tip)
932 list_del(&tip->head);
938 free(tip->file_name);
944 * rem_input_files - Remove all input files
946 static void rem_input_files(void)
948 struct list_head *p, *q;
950 list_for_each_safe(p, q, &input_files) {
951 rem_input_file(list_entry(p, struct thr_info, head));
956 * __find_input_files - Find input files associated with this device (per cpu)
958 static void __find_input_files(struct dev_info *dip)
963 char full_name[MAXPATHLEN];
965 sprintf(full_name, "%s/%s.%s.%d", idir, dip->devnm, ibase, cpu);
966 if (access(full_name, R_OK) != 0)
969 add_input_file(cpu, dip->devnm, full_name);
974 fatal(NULL, ERR_ARGS, "No traces found for %s\n", dip->devnm);
983 * find_input_files - Find input files for all devices
985 static void find_input_files(void)
987 struct list_head *p, *q;
989 list_for_each_safe(p, q, &input_devs) {
990 __find_input_files(list_entry(p, struct dev_info, head));
995 * ========================================================================
996 * ==== RECLAIM ROUTINES ==================================================
997 * ========================================================================
1001 * reap_wait_aios - Wait for and return number of outstanding AIOs
1003 * Will return 0 if we are done
1005 static int reap_wait_aios(struct thr_info *tip)
1009 if (!is_reap_done(tip)) {
1010 pthread_mutex_lock(&tip->mutex);
1011 while (tip->naios_out == 0) {
1013 if (pthread_cond_wait(&tip->cond, &tip->mutex)) {
1014 fatal("pthread_cond_wait", ERR_SYSCALL,
1015 "nfree_current cond wait failed\n");
1019 naios = tip->naios_out;
1020 pthread_mutex_unlock(&tip->mutex);
1022 assert(is_reap_done(tip) || naios > 0);
1024 return is_reap_done(tip) ? 0 : naios;
1028 * reclaim_ios - Reclaim AIOs completed, recycle IOCBs
1029 * @tip: Per-thread information
1030 * @naios_out: Number of AIOs we have outstanding (min)
1032 static void reclaim_ios(struct thr_info *tip, long naios_out)
1035 struct io_event *evp, events[naios_out];
1040 ndone = io_getevents(tip->ctx, 1, naios_out, events, NULL);
1044 if (errno && errno != EINTR) {
1045 fatal("io_getevents", ERR_SYSCALL,
1046 "io_getevents failed\n");
1050 assert(0 < ndone && ndone <= naios_out);
1052 pthread_mutex_lock(&tip->mutex);
1053 for (i = 0, evp = events; i < ndone; i++, evp++) {
1054 struct iocb_pkt *iocbp = evp->data;
1056 if (evp->res != iocbp->iocb.u.c.nbytes) {
1057 fatal(NULL, ERR_SYSCALL,
1058 "Event failure %ld/%ld\t(%ld + %ld)\n",
1059 (long)evp->res, (long)evp->res2,
1060 (long)iocbp->iocb.u.c.offset / nb_sec,
1061 (long)iocbp->iocb.u.c.nbytes / nb_sec);
1065 list_move_tail(&iocbp->head, &tip->free_iocbs);
1068 tip->naios_free += ndone;
1069 tip->naios_out -= ndone;
1070 naios_out = minl(naios_out, tip->naios_out);
1072 if (tip->send_wait) {
1074 pthread_cond_signal(&tip->cond);
1076 pthread_mutex_unlock(&tip->mutex);
1079 * Short cut: If we /know/ there are some more AIOs, go handle them
1086 * replay_rec - Worker thread to reclaim AIOs
1087 * @arg: Pointer to thread information
1089 static void *replay_rec(void *arg)
1092 struct thr_info *tip = arg;
1094 while ((naios_out = reap_wait_aios(tip)) > 0)
1095 reclaim_ios(tip, naios_out);
1097 assert(tip->send_done);
1105 * ========================================================================
1106 * ==== REPLAY ROUTINES ===================================================
1107 * ========================================================================
1111 * next_bunch - Retrieve next bunch of AIOs to process
1112 * @tip: Per-thread information
1113 * @bunch: Bunch information
1115 * Returns TRUE if we recovered a bunch of IOs, else hit EOF
1117 static int next_bunch(struct thr_info *tip, struct io_bunch *bunch)
1119 size_t count, result;
1121 result = read(tip->ifd, &bunch->hdr, sizeof(bunch->hdr));
1122 if (result != sizeof(bunch->hdr)) {
1126 fatal(tip->file_name, ERR_SYSCALL, "Short hdr(%ld)\n",
1130 assert(bunch->hdr.npkts <= BT_MAX_PKTS);
1132 count = bunch->hdr.npkts * sizeof(struct io_pkt);
1133 result = read(tip->ifd, &bunch->pkts, count);
1134 if (result != count) {
1135 fatal(tip->file_name, ERR_SYSCALL, "Short pkts(%ld/%ld)\n",
1136 (long)result, (long)count);
1144 * nfree_current - Returns current number of AIOs that are free
1146 * Will wait for available ones...
1148 * Returns 0 if we have some condition that causes us to exit
1150 static int nfree_current(struct thr_info *tip)
1154 pthread_mutex_lock(&tip->mutex);
1155 while (!is_send_done(tip) && ((nfree = tip->naios_free) == 0)) {
1157 if (pthread_cond_wait(&tip->cond, &tip->mutex)) {
1158 fatal("pthread_cond_wait", ERR_SYSCALL,
1159 "nfree_current cond wait failed\n");
1163 pthread_mutex_unlock(&tip->mutex);
1169 * stall - Stall for the number of nanoseconds requested
1171 * We may be late, in which case we just return.
1173 static void stall(struct thr_info *tip, long long oclock)
1175 struct timespec req;
1176 long long dreal, tclock = gettime() - rgenesis;
1179 fprintf(tip->vfp, " stall(%lld.%09lld, %lld.%09lld)\n",
1180 du64_to_sec(oclock), du64_to_nsec(oclock),
1181 du64_to_sec(tclock), du64_to_nsec(tclock));
1183 while (!is_send_done(tip) && tclock < oclock) {
1184 dreal = oclock - tclock;
1185 req.tv_sec = dreal / (1000 * 1000 * 1000);
1186 req.tv_nsec = dreal % (1000 * 1000 * 1000);
1189 fprintf(tip->vfp, "++ stall(%lld.%09lld) ++\n",
1190 (long long)req.tv_sec,
1191 (long long)req.tv_nsec);
1194 if (nanosleep(&req, NULL) < 0 && signal_done)
1197 tclock = gettime() - rgenesis;
1202 * iocbs_map - Map a set of AIOs onto a set of IOCBs
1203 * @tip: Per-thread information
1204 * @list: List of AIOs created
1205 * @pkts: AIOs to map
1206 * @ntodo: Number of AIOs to map
1208 static void iocbs_map(struct thr_info *tip, struct iocb **list,
1209 struct io_pkt *pkts, int ntodo)
1214 assert(0 < ntodo && ntodo <= naios);
1216 pthread_mutex_lock(&tip->mutex);
1217 assert(ntodo <= list_len(&tip->free_iocbs));
1218 for (i = 0, pkt = pkts; i < ntodo; i++, pkt++) {
1220 struct iocb_pkt *iocbp;
1222 if (!pkt->rw && !write_enabled)
1226 fprintf(tip->vfp, "\t%10llu + %10llu %c%c\n",
1227 (unsigned long long)pkt->sector,
1228 (unsigned long long)pkt->nbytes / nb_sec,
1230 (rw == 1 && pkt->rw == 0) ? '!' : ' ');
1232 iocbp = list_entry(tip->free_iocbs.next, struct iocb_pkt, head);
1233 iocb_setup(iocbp, rw, pkt->nbytes, pkt->sector * nb_sec);
1235 list_move_tail(&iocbp->head, &tip->used_iocbs);
1236 list[i] = &iocbp->iocb;
1239 tip->naios_free -= ntodo;
1240 assert(tip->naios_free >= 0);
1241 pthread_mutex_unlock(&tip->mutex);
1245 * process_bunch - Process a bunch of requests
1246 * @tip: Per-thread information
1247 * @bunch: Bunch to process
1249 static void process_bunch(struct thr_info *tip, struct io_bunch *bunch)
1252 struct iocb *list[bunch->hdr.npkts];
1254 assert(0 < bunch->hdr.npkts && bunch->hdr.npkts <= BT_MAX_PKTS);
1255 while (!is_send_done(tip) && (i < bunch->hdr.npkts)) {
1257 int ntodo = min(nfree_current(tip), bunch->hdr.npkts - i);
1259 assert(0 < ntodo && ntodo <= naios);
1260 iocbs_map(tip, list, &bunch->pkts[i], ntodo);
1262 stall(tip, bunch->hdr.time_stamp - genesis);
1266 fprintf(tip->vfp, "submit(%d)\n", ntodo);
1267 ndone = io_submit(tip->ctx, ntodo, list);
1268 if (ndone != (long)ntodo) {
1269 fatal("io_submit", ERR_SYSCALL,
1270 "%d: io_submit(%d:%ld) failed (%s)\n",
1271 tip->cpu, ntodo, ndone,
1272 strerror(labs(ndone)));
1276 pthread_mutex_lock(&tip->mutex);
1277 tip->naios_out += ndone;
1278 assert(tip->naios_out <= naios);
1279 if (tip->reap_wait) {
1281 pthread_cond_signal(&tip->cond);
1283 pthread_mutex_unlock(&tip->mutex);
1286 assert(i <= bunch->hdr.npkts);
1292 * reset_input_file - Reset the input file for the next iteration
1293 * @tip: Thread information
1295 * We also do a dummy read of the file header to get us to the first bunch.
1297 static void reset_input_file(struct thr_info *tip)
1299 struct io_file_hdr hdr;
1301 lseek(tip->ifd, 0, 0);
1303 if (read(tip->ifd, &hdr, sizeof(hdr)) != sizeof(hdr)) {
1304 fatal(tip->file_name, ERR_ARGS, "Header reread failed\n");
1310 * replay_sub - Worker thread to submit AIOs that are being replayed
1312 static void *replay_sub(void *arg)
1314 char path[MAXPATHLEN];
1315 struct io_bunch bunch;
1316 struct thr_info *tip = arg;
1320 sprintf(path, "/dev/%s", map_dev(tip->devnm));
1321 tip->ofd = open(path, O_RDWR | O_DIRECT | O_NOATIME);
1323 fatal(path, ERR_SYSCALL, "Failed device open\n");
1328 while (!is_send_done(tip) && tip->iterations--) {
1331 fprintf(tip->vfp, "\n=== %d ===\n", tip->iterations);
1332 while (!is_send_done(tip) && next_bunch(tip, &bunch))
1333 process_bunch(tip, &bunch);
1335 reset_input_file(tip);
1344 * ========================================================================
1345 * ==== COMMAND LINE ARGUMENT HANDLING ====================================
1346 * ========================================================================
1349 static char usage_str[] = \
1351 "\t[ -c <cpus> : --cpus=<cpus> ] Default: 1\n" \
1352 "\t[ -d <dir> : --input-directory=<dir> ] Default: .\n" \
1353 "\t[ -F : --find-records ] Default: Off\n" \
1354 "\t[ -h : --help ] Default: Off\n" \
1355 "\t[ -i <base> : --input-base=<base> ] Default: replay\n" \
1356 "\t[ -I <iters>: --iterations=<iters> ] Default: 1\n" \
1357 "\t[ -M <file> : --map-devs=<file> ] Default: None\n" \
1358 "\t[ -N : --no-stalls ] Default: Off\n" \
1359 "\t[ -v : --verbose ] Default: Off\n" \
1360 "\t[ -V : --version ] Default: Off\n" \
1361 "\t[ -W : --write-enable ] Default: Off\n" \
1362 "\t<dev...> Default: None\n" \
1365 #define S_OPTS "c:d:Fhi:I:M:Nt:vVW"
1366 static struct option l_opts[] = {
1369 .has_arg = required_argument,
1374 .name = "input-directory",
1375 .has_arg = required_argument,
1380 .name = "find-records",
1381 .has_arg = no_argument,
1387 .has_arg = no_argument,
1392 .name = "input-base",
1393 .has_arg = required_argument,
1398 .name = "iterations",
1399 .has_arg = required_argument,
1405 .has_arg = required_argument,
1410 .name = "no-stalls",
1411 .has_arg = no_argument,
1417 .has_arg = no_argument,
1423 .has_arg = no_argument,
1428 .name = "write-enable",
1429 .has_arg = no_argument,
1439 * handle_args: Parse passed in argument list
1440 * @argc: Number of arguments in argv
1441 * @argv: Arguments passed in
1443 * Does rudimentary parameter verification as well.
1445 static void handle_args(int argc, char *argv[])
1449 while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
1452 cpus_to_use = atoi(optarg);
1453 if (cpus_to_use <= 0 || cpus_to_use > ncpus) {
1454 fatal(NULL, ERR_ARGS,
1455 "Invalid number of cpus %d (0<x<%d)\n",
1456 cpus_to_use, ncpus);
1463 if (access(idir, R_OK | X_OK) != 0) {
1464 fatal(idir, ERR_ARGS,
1465 "Invalid input directory specified\n");
1484 def_iterations = atoi(optarg);
1485 if (def_iterations <= 0) {
1487 "Invalid number of iterations %d\n",
1495 read_map_devs(optarg);
1503 fprintf(stderr, "btreplay -- version %s\n",
1505 fprintf(stderr, " Built on %s\n",
1520 fatal(NULL, ERR_ARGS,
1521 "Invalid command line argument %c\n", c);
1526 while (optind < argc)
1527 add_input_dev(argv[optind++]);
1530 find_input_devs(idir);
1532 if (list_len(&input_devs) == 0) {
1533 fatal(NULL, ERR_ARGS, "Missing required input dev name(s)\n");
1537 if (cpus_to_use < 0)
1538 cpus_to_use = ncpus;
1542 * ========================================================================
1543 * ==== MAIN ROUTINE ======================================================
1544 * ========================================================================
1548 * set_signal_done - Signal handler, catches signals & sets signal_done
1550 static void set_signal_done(__attribute__((__unused__))int signum)
1557 * @argc: Number of arguments
1558 * @argv: Array of arguments
1560 int main(int argc, char *argv[])
1563 struct list_head *p;
1565 pgsize = getpagesize();
1568 setup_signal(SIGINT, set_signal_done);
1569 setup_signal(SIGTERM, set_signal_done);
1572 handle_args(argc, argv);
1575 nfiles = list_len(&input_files);
1576 __list_for_each(p, &input_files) {
1577 tip_init(list_entry(p, struct thr_info, head));
1580 wait_replays_ready();
1581 for (i = 0; i < def_iterations; i++) {
1582 rgenesis = gettime();
1585 fprintf(stderr, "I");
1589 wait_replays_done();
1590 wait_reclaims_done();
1593 fprintf(stderr, "\n");