X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=server.c;h=c520b6bb5dbb8dd734de7fefd2955dc6cd9a132e;hp=0942e631730388bf0e225f4d20673bb3adc50df6;hb=90eff1c9f01d2f8b4ff8bc75a2bf120a1789b37c;hpb=f77d4b73613adbce2177576afd591a3322774329 diff --git a/server.c b/server.c index 0942e631..c520b6bb 100644 --- a/server.c +++ b/server.c @@ -36,22 +36,29 @@ enum { SK_F_FREE = 1, SK_F_COPY = 2, SK_F_SIMPLE = 4, + SK_F_VEC = 8, + SK_F_INLINE = 16, }; struct sk_entry { - struct flist_head list; - int opcode; + struct flist_head list; /* link on sk_out->list */ + int flags; /* SK_F_* */ + int opcode; /* Actual command fields */ void *buf; off_t size; - uint64_t *tagptr; - int flags; + uint64_t tag; + struct flist_head next; /* Other sk_entry's, if linked command */ }; struct sk_out { - int sk; - struct fio_mutex *lock; - struct flist_head list; - struct fio_mutex *wait; + unsigned int refs; /* frees sk_out when it drops to zero. + * protected by below ->lock */ + + int sk; /* socket fd to talk to client */ + struct fio_mutex lock; /* protects ref and below list */ + struct flist_head list; /* list of pending transmit work */ + struct fio_mutex wait; /* wake backend when items added to list */ + struct fio_mutex xmit; /* held while sending data */ }; static char *fio_server_arg; @@ -107,16 +114,85 @@ static const char *fio_server_ops[FIO_NET_CMD_NR] = { "LOAD_FILE", "VTRIGGER", "SENDFILE", + "JOB_OPT", }; static void sk_lock(struct sk_out *sk_out) { - fio_mutex_down(sk_out->lock); + fio_mutex_down(&sk_out->lock); } static void sk_unlock(struct sk_out *sk_out) { - fio_mutex_up(sk_out->lock); + fio_mutex_up(&sk_out->lock); +} + +void sk_out_assign(struct sk_out *sk_out) +{ + if (!sk_out) + return; + + sk_lock(sk_out); + sk_out->refs++; + sk_unlock(sk_out); + pthread_setspecific(sk_out_key, sk_out); +} + +static void sk_out_free(struct sk_out *sk_out) +{ + __fio_mutex_remove(&sk_out->lock); + __fio_mutex_remove(&sk_out->wait); + __fio_mutex_remove(&sk_out->xmit); + sfree(sk_out); +} + +static int __sk_out_drop(struct sk_out *sk_out) +{ + if (sk_out) { + int refs; + + sk_lock(sk_out); + assert(sk_out->refs != 0); + refs = --sk_out->refs; + sk_unlock(sk_out); + + if (!refs) { + sk_out_free(sk_out); + pthread_setspecific(sk_out_key, NULL); + return 0; + } + } + + return 1; +} + +void sk_out_drop(void) +{ + struct sk_out *sk_out; + + sk_out = pthread_getspecific(sk_out_key); + __sk_out_drop(sk_out); +} + +static void __fio_init_net_cmd(struct fio_net_cmd *cmd, uint16_t opcode, + uint32_t pdu_len, uint64_t tag) +{ + memset(cmd, 0, sizeof(*cmd)); + + cmd->version = __cpu_to_le16(FIO_SERVER_VER); + cmd->opcode = cpu_to_le16(opcode); + cmd->tag = cpu_to_le64(tag); + cmd->pdu_len = cpu_to_le32(pdu_len); +} + + +static void fio_init_net_cmd(struct fio_net_cmd *cmd, uint16_t opcode, + const void *pdu, uint32_t pdu_len, uint64_t tag) +{ + __fio_init_net_cmd(cmd, opcode, pdu_len, tag); + + if (pdu) + memcpy(&cmd->payload, pdu, pdu_len); } const char *fio_server_op(unsigned int op) @@ -175,13 +251,10 @@ static int fio_sendv_data(int sk, struct iovec *iov, int count) if (!total_len) return 0; - if (errno) - return -errno; - return 1; } -int fio_send_data(int sk, const void *p, unsigned int len) +static int fio_send_data(int sk, const void *p, unsigned int len) { struct iovec iov = { .iov_base = (void *) p, .iov_len = len }; @@ -190,10 +263,17 @@ int fio_send_data(int sk, const void *p, unsigned int len) return fio_sendv_data(sk, &iov, 1); } -int fio_recv_data(int sk, void *p, unsigned int len) +static int fio_recv_data(int sk, void *p, unsigned int len, bool wait) { + int flags; + + if (wait) + flags = MSG_WAITALL; + else + flags = OS_MSG_DONTWAIT; + do { - int ret = recv(sk, p, len, MSG_WAITALL); + int ret = recv(sk, p, len, flags); if (ret > 0) { len -= ret; @@ -203,9 +283,11 @@ int fio_recv_data(int sk, void *p, unsigned int len) continue; } else if (!ret) break; - else if (errno == EAGAIN || errno == EINTR) - continue; - else + else if (errno == EAGAIN || errno == EINTR) { + if (wait) + continue; + break; + } else break; } while (!exit_backend); @@ -254,7 +336,7 @@ static int verify_convert_cmd(struct fio_net_cmd *cmd) /* * Read (and defragment, if necessary) incoming commands */ -struct fio_net_cmd *fio_net_recv_cmd(int sk) +struct fio_net_cmd *fio_net_recv_cmd(int sk, bool wait) { struct fio_net_cmd cmd, *tmp, *cmdret = NULL; size_t cmd_size = 0, pdu_offset = 0; @@ -263,7 +345,7 @@ struct fio_net_cmd *fio_net_recv_cmd(int sk) void *pdu = NULL; do { - ret = fio_recv_data(sk, &cmd, sizeof(cmd)); + ret = fio_recv_data(sk, &cmd, sizeof(cmd), wait); if (ret) break; @@ -307,7 +389,7 @@ struct fio_net_cmd *fio_net_recv_cmd(int sk) /* There's payload, get it */ pdu = (void *) cmdret->payload + pdu_offset; - ret = fio_recv_data(sk, pdu, cmd.pdu_len); + ret = fio_recv_data(sk, pdu, cmd.pdu_len, wait); if (ret) break; @@ -382,7 +464,7 @@ static void free_reply(uint64_t tag) free(reply); } -void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu) +static void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu) { uint32_t pdu_len; @@ -392,7 +474,7 @@ void fio_net_cmd_crc_pdu(struct fio_net_cmd *cmd, const void *pdu) cmd->pdu_crc16 = __cpu_to_le16(fio_crc16(pdu, pdu_len)); } -void fio_net_cmd_crc(struct fio_net_cmd *cmd) +static void fio_net_cmd_crc(struct fio_net_cmd *cmd) { fio_net_cmd_crc_pdu(cmd, cmd->payload); } @@ -449,30 +531,59 @@ int fio_net_send_cmd(int fd, uint16_t opcode, const void *buf, off_t size, return ret; } -static int fio_net_queue_cmd(uint16_t opcode, void *buf, off_t size, - uint64_t *tagptr, int flags) +static struct sk_entry *fio_net_prep_cmd(uint16_t opcode, void *buf, + size_t size, uint64_t *tagptr, + int flags) { - struct sk_out *sk_out = pthread_getspecific(sk_out_key); struct sk_entry *entry; entry = smalloc(sizeof(*entry)); + INIT_FLIST_HEAD(&entry->next); entry->opcode = opcode; if (flags & SK_F_COPY) { entry->buf = smalloc(size); memcpy(entry->buf, buf, size); } else entry->buf = buf; + entry->size = size; - entry->tagptr = tagptr; + if (tagptr) + entry->tag = *tagptr; + else + entry->tag = 0; entry->flags = flags; + return entry; +} - sk_lock(sk_out); - flist_add_tail(&entry->list, &sk_out->list); - sk_unlock(sk_out); +static int handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry); - fio_mutex_up(sk_out->wait); +static void fio_net_queue_entry(struct sk_entry *entry) +{ + struct sk_out *sk_out = pthread_getspecific(sk_out_key); - return 0; + if (entry->flags & SK_F_INLINE) + handle_sk_entry(sk_out, entry); + else { + sk_lock(sk_out); + flist_add_tail(&entry->list, &sk_out->list); + sk_unlock(sk_out); + + fio_mutex_up(&sk_out->wait); + } +} + +static int fio_net_queue_cmd(uint16_t opcode, void *buf, off_t size, + uint64_t *tagptr, int flags) +{ + struct sk_entry *entry; + + entry = fio_net_prep_cmd(opcode, buf, size, tagptr, flags); + if (entry) { + fio_net_queue_entry(entry); + return 0; + } + + return 1; } static int fio_net_send_simple_stack_cmd(int sk, uint16_t opcode, uint64_t tag) @@ -515,7 +626,7 @@ static int fio_net_queue_quit(void) { dprint(FD_NET, "server: sending quit\n"); - return fio_net_queue_cmd(FIO_NET_CMD_QUIT, NULL, 0, 0, SK_F_SIMPLE); + return fio_net_queue_cmd(FIO_NET_CMD_QUIT, NULL, 0, NULL, SK_F_SIMPLE); } int fio_net_send_quit(int sk) @@ -658,10 +769,11 @@ static int handle_load_file_cmd(struct fio_net_cmd *cmd) static int handle_run_cmd(struct sk_out *sk_out, struct flist_head *job_list, struct fio_net_cmd *cmd) { - struct backend_data data; pid_t pid; int ret; + sk_out_assign(sk_out); + fio_time_init(); set_genesis_time(); @@ -671,11 +783,9 @@ static int handle_run_cmd(struct sk_out *sk_out, struct flist_head *job_list, return 0; } - data.key = sk_out_key; - data.ptr = sk_out; - //pthread_setspecific(sk_out_key, sk_out); - ret = fio_backend(&data); + ret = fio_backend(sk_out); free_threads_shm(); + sk_out_drop(); _exit(ret); } @@ -802,11 +912,11 @@ static int handle_send_eta_cmd(struct fio_net_cmd *cmd) je->files_open = cpu_to_le32(je->files_open); for (i = 0; i < DDIR_RWDIR_CNT; i++) { - je->m_rate[i] = cpu_to_le32(je->m_rate[i]); - je->t_rate[i] = cpu_to_le32(je->t_rate[i]); + je->m_rate[i] = cpu_to_le64(je->m_rate[i]); + je->t_rate[i] = cpu_to_le64(je->t_rate[i]); je->m_iops[i] = cpu_to_le32(je->m_iops[i]); je->t_iops[i] = cpu_to_le32(je->t_iops[i]); - je->rate[i] = cpu_to_le32(je->rate[i]); + je->rate[i] = cpu_to_le64(je->rate[i]); je->iops[i] = cpu_to_le32(je->iops[i]); } @@ -866,9 +976,9 @@ static int handle_trigger_cmd(struct fio_net_cmd *cmd) struct all_io_list state; state.threads = cpu_to_le64((uint64_t) 0); - fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, SK_F_COPY); + fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, &state, sizeof(state), NULL, SK_F_COPY | SK_F_INLINE); } else - fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, rep, sz, NULL, SK_F_FREE); + fio_net_queue_cmd(FIO_NET_CMD_VTRIGGER, rep, sz, NULL, SK_F_FREE | SK_F_INLINE); exec_trigger(buf); return 0; @@ -949,18 +1059,47 @@ static int handle_command(struct sk_out *sk_out, struct flist_head *job_list, return ret; } -static void handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry) +/* + * Send a command with a separate PDU, not inlined in the command + */ +static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf, + off_t size, uint64_t tag, uint32_t flags) { - if (entry->flags & SK_F_SIMPLE) { - uint64_t tag = 0; + struct fio_net_cmd cmd; + struct iovec iov[2]; + size_t this_len; + int ret; + + iov[0].iov_base = (void *) &cmd; + iov[0].iov_len = sizeof(cmd); - if (entry->tagptr) - tag = *entry->tagptr; + do { + uint32_t this_flags = flags; - fio_net_send_simple_cmd(sk_out->sk, entry->opcode, tag, NULL); - } else - fio_net_send_cmd(sk_out->sk, entry->opcode, entry->buf, entry->size, entry->tagptr, NULL); + this_len = size; + if (this_len > FIO_SERVER_MAX_FRAGMENT_PDU) + this_len = FIO_SERVER_MAX_FRAGMENT_PDU; + + if (this_len < size) + this_flags |= FIO_NET_CMD_F_MORE; + + __fio_init_net_cmd(&cmd, opcode, this_len, tag); + cmd.flags = __cpu_to_le32(this_flags); + fio_net_cmd_crc_pdu(&cmd, buf); + iov[1].iov_base = (void *) buf; + iov[1].iov_len = this_len; + + ret = fio_sendv_data(sk, iov, 2); + size -= this_len; + buf += this_len; + } while (!ret && size); + + return ret; +} + +static void finish_entry(struct sk_entry *entry) +{ if (entry->flags & SK_F_FREE) free(entry->buf); else if (entry->flags & SK_F_COPY) @@ -969,15 +1108,76 @@ static void handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry) sfree(entry); } -static void handle_xmits(struct sk_out *sk_out) +static void entry_set_flags(struct sk_entry *entry, struct flist_head *list, + unsigned int *flags) +{ + if (!flist_empty(list)) + *flags = FIO_NET_CMD_F_MORE; + else + *flags = 0; +} + +static int send_vec_entry(struct sk_out *sk_out, struct sk_entry *first) +{ + unsigned int flags; + int ret; + + entry_set_flags(first, &first->next, &flags); + + ret = fio_send_cmd_ext_pdu(sk_out->sk, first->opcode, first->buf, + first->size, first->tag, flags); + + while (!flist_empty(&first->next)) { + struct sk_entry *next; + + next = flist_first_entry(&first->next, struct sk_entry, list); + flist_del_init(&next->list); + + entry_set_flags(next, &first->next, &flags); + + ret += fio_send_cmd_ext_pdu(sk_out->sk, next->opcode, next->buf, + next->size, next->tag, flags); + finish_entry(next); + } + + return ret; +} + +static int handle_sk_entry(struct sk_out *sk_out, struct sk_entry *entry) +{ + int ret; + + fio_mutex_down(&sk_out->xmit); + + if (entry->flags & SK_F_VEC) + ret = send_vec_entry(sk_out, entry); + else if (entry->flags & SK_F_SIMPLE) { + ret = fio_net_send_simple_cmd(sk_out->sk, entry->opcode, + entry->tag, NULL); + } else { + ret = fio_net_send_cmd(sk_out->sk, entry->opcode, entry->buf, + entry->size, &entry->tag, NULL); + } + + fio_mutex_up(&sk_out->xmit); + + if (ret) + log_err("fio: failed handling cmd %s\n", fio_server_op(entry->opcode)); + + finish_entry(entry); + return ret; +} + +static int handle_xmits(struct sk_out *sk_out) { struct sk_entry *entry; FLIST_HEAD(list); + int ret = 0; sk_lock(sk_out); if (flist_empty(&sk_out->list)) { sk_unlock(sk_out); - return; + return 0; } flist_splice_init(&sk_out->list, &list); @@ -986,8 +1186,10 @@ static void handle_xmits(struct sk_out *sk_out) while (!flist_empty(&list)) { entry = flist_entry(list.next, struct sk_entry, list); flist_del(&entry->list); - handle_sk_entry(sk_out, entry); + ret += handle_sk_entry(sk_out, entry); } + + return ret; } static int handle_connection(struct sk_out *sk_out) @@ -1022,7 +1224,7 @@ static int handle_connection(struct sk_out *sk_out) break; } else if (!ret) { fio_server_check_jobs(&job_list); - fio_mutex_down_timeout(sk_out->wait, timeout); + fio_mutex_down_timeout(&sk_out->wait, timeout); continue; } @@ -1039,7 +1241,7 @@ static int handle_connection(struct sk_out *sk_out) if (ret < 0) break; - cmd = fio_net_recv_cmd(sk_out->sk); + cmd = fio_net_recv_cmd(sk_out->sk, true); if (!cmd) { ret = -1; break; @@ -1059,13 +1261,15 @@ static int handle_connection(struct sk_out *sk_out) handle_xmits(sk_out); close(sk_out->sk); + sk_out->sk = -1; + __sk_out_drop(sk_out); _exit(ret); } /* get the address on this host bound by the input socket, * whether it is ipv6 or ipv4 */ -int get_my_addr_str(int sk) +static int get_my_addr_str(int sk) { struct sockaddr_in6 myaddr6 = { 0, }; struct sockaddr_in myaddr4 = { 0, }; @@ -1106,21 +1310,14 @@ static int accept_loop(int listen_sk) socklen_t len = use_ipv6 ? sizeof(addr6) : sizeof(addr); struct pollfd pfd; int ret = 0, sk, exitval = 0; - struct sk_out *sk_out; FLIST_HEAD(conn_list); dprint(FD_NET, "server enter accept loop\n"); fio_set_fd_nonblocking(listen_sk, "server"); - sk_out = smalloc(sizeof(*sk_out)); - INIT_FLIST_HEAD(&sk_out->list); - sk_out->lock = fio_mutex_init(FIO_MUTEX_UNLOCKED); - sk_out->wait = fio_mutex_init(FIO_MUTEX_LOCKED); - - pthread_setspecific(sk_out_key, sk_out); - while (!exit_backend) { + struct sk_out *sk_out; const char *from; char buf[64]; pid_t pid; @@ -1170,28 +1367,31 @@ static int accept_loop(int listen_sk) dprint(FD_NET, "server: connect from %s\n", from); + sk_out = smalloc(sizeof(*sk_out)); sk_out->sk = sk; + INIT_FLIST_HEAD(&sk_out->list); + __fio_mutex_init(&sk_out->lock, FIO_MUTEX_UNLOCKED); + __fio_mutex_init(&sk_out->wait, FIO_MUTEX_LOCKED); + __fio_mutex_init(&sk_out->xmit, FIO_MUTEX_UNLOCKED); pid = fork(); if (pid) { close(sk); fio_server_add_conn_pid(&conn_list, pid); - pthread_setspecific(sk_out_key, sk_out); continue; } - /* exits */ - get_my_addr_str(sk); /* if error, it's already logged, non-fatal */ + /* if error, it's already logged, non-fatal */ + get_my_addr_str(sk); + + /* + * Assign sk_out here, it'll be dropped in handle_connection() + * since that function calls _exit() when done + */ + sk_out_assign(sk_out); handle_connection(sk_out); } -#if 0 - fio_mutex_remove(sk_out->lock); - fio_mutex_remove(sk_out->wait); - sfree(sk_out); - pthread_setspecific(sk_out_key, NULL); -#endif - return exitval; } @@ -1360,6 +1560,47 @@ void fio_server_send_gs(struct group_run_stats *rs) fio_net_queue_cmd(FIO_NET_CMD_GS, &gs, sizeof(gs), NULL, SK_F_COPY); } +void fio_server_send_job_options(struct flist_head *opt_list, + unsigned int groupid) +{ + struct cmd_job_option pdu; + struct flist_head *entry; + + if (flist_empty(opt_list)) + return; + + flist_for_each(entry, opt_list) { + struct print_option *p; + size_t len; + + p = flist_entry(entry, struct print_option, list); + memset(&pdu, 0, sizeof(pdu)); + + if (groupid == -1U) { + pdu.global = __cpu_to_le16(1); + pdu.groupid = 0; + } else { + pdu.global = 0; + pdu.groupid = cpu_to_le32(groupid); + } + len = strlen(p->name); + if (len >= sizeof(pdu.name)) { + len = sizeof(pdu.name) - 1; + pdu.truncated = __cpu_to_le16(1); + } + memcpy(pdu.name, p->name, len); + if (p->value) { + len = strlen(p->value); + if (len >= sizeof(pdu.value)) { + len = sizeof(pdu.value) - 1; + pdu.truncated = __cpu_to_le16(1); + } + memcpy(pdu.value, p->value, len); + } + fio_net_queue_cmd(FIO_NET_CMD_JOB_OPT, &pdu, sizeof(pdu), NULL, SK_F_COPY); + } +} + static void convert_agg(struct disk_util_agg *dst, struct disk_util_agg *src) { int i; @@ -1416,131 +1657,314 @@ void fio_server_send_du(void) } } +#ifdef CONFIG_ZLIB + +static inline void __fio_net_prep_tail(z_stream *stream, void *out_pdu, + struct sk_entry **last_entry, + struct sk_entry *first) +{ + unsigned int this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream->avail_out; + + *last_entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len, + NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE); + flist_add_tail(&(*last_entry)->list, &first->next); + +} + /* - * Send a command with a separate PDU, not inlined in the command + * Deflates the next input given, creating as many new packets in the + * linked list as necessary. */ -static int fio_send_cmd_ext_pdu(int sk, uint16_t opcode, const void *buf, - off_t size, uint64_t tag, uint32_t flags) +static int __deflate_pdu_buffer(void *next_in, unsigned int next_sz, void **out_pdu, + struct sk_entry **last_entry, z_stream *stream, + struct sk_entry *first) { - struct fio_net_cmd cmd; - struct iovec iov[2]; + int ret; - iov[0].iov_base = (void *) &cmd; - iov[0].iov_len = sizeof(cmd); - iov[1].iov_base = (void *) buf; - iov[1].iov_len = size; + stream->next_in = next_in; + stream->avail_in = next_sz; + do { + if (! stream->avail_out) { + + __fio_net_prep_tail(stream, *out_pdu, last_entry, first); + + *out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU); + + stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU; + stream->next_out = *out_pdu; + } - __fio_init_net_cmd(&cmd, opcode, size, tag); - cmd.flags = __cpu_to_le32(flags); - fio_net_cmd_crc_pdu(&cmd, buf); + ret = deflate(stream, Z_BLOCK); - return fio_sendv_data(sk, iov, 2); + if (ret < 0) { + free(*out_pdu); + return 1; + } + } while (stream->avail_in); + + return 0; } -static int fio_send_iolog_gz(struct sk_out *sk_out, struct cmd_iolog_pdu *pdu, - struct io_log *log) +static int __fio_append_iolog_gz_hist(struct sk_entry *first, struct io_log *log, + struct io_logs *cur_log, z_stream *stream) { - int ret = 0; -#ifdef CONFIG_ZLIB - z_stream stream; + struct sk_entry *entry; void *out_pdu; + int ret, i, j; + int sample_sz = log_entry_sz(log); - /* - * Dirty - since the log is potentially huge, compress it into - * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving - * side defragment it. - */ out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU); + stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU; + stream->next_out = out_pdu; + + for (i = 0; i < cur_log->nr_samples; i++) { + struct io_sample *s; + struct io_u_plat_entry *cur_plat_entry, *prev_plat_entry; + unsigned int *cur_plat, *prev_plat; + + s = get_sample(log, cur_log, i); + ret = __deflate_pdu_buffer(s, sample_sz, &out_pdu, &entry, stream, first); + if (ret) + return ret; + + /* Do the subtraction on server side so that client doesn't have to + * reconstruct our linked list from packets. + */ + cur_plat_entry = s->plat_entry; + prev_plat_entry = flist_first_entry(&cur_plat_entry->list, struct io_u_plat_entry, list); + cur_plat = cur_plat_entry->io_u_plat; + prev_plat = prev_plat_entry->io_u_plat; + + for (j = 0; j < FIO_IO_U_PLAT_NR; j++) { + cur_plat[j] -= prev_plat[j]; + } + + flist_del(&prev_plat_entry->list); + free(prev_plat_entry); + + ret = __deflate_pdu_buffer(cur_plat_entry, sizeof(*cur_plat_entry), + &out_pdu, &entry, stream, first); + + if (ret) + return ret; + } + + __fio_net_prep_tail(stream, out_pdu, &entry, first); + + return 0; +} + +static int __fio_append_iolog_gz(struct sk_entry *first, struct io_log *log, + struct io_logs *cur_log, z_stream *stream) +{ + unsigned int this_len; + void *out_pdu; + int ret; + + if (log->log_type == IO_LOG_TYPE_HIST) + return __fio_append_iolog_gz_hist(first, log, cur_log, stream); + + stream->next_in = (void *) cur_log->log; + stream->avail_in = cur_log->nr_samples * log_entry_sz(log); + do { + struct sk_entry *entry; + + /* + * Dirty - since the log is potentially huge, compress it into + * FIO_SERVER_MAX_FRAGMENT_PDU chunks and let the receiving + * side defragment it. + */ + out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU); + + stream->avail_out = FIO_SERVER_MAX_FRAGMENT_PDU; + stream->next_out = out_pdu; + ret = deflate(stream, Z_BLOCK); + /* may be Z_OK, or Z_STREAM_END */ + if (ret < 0) { + free(out_pdu); + return 1; + } + + this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream->avail_out; + + entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len, + NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE); + flist_add_tail(&entry->list, &first->next); + } while (stream->avail_in); + + return 0; +} + +static int fio_append_iolog_gz(struct sk_entry *first, struct io_log *log) +{ + int ret = 0; + z_stream stream; + + memset(&stream, 0, sizeof(stream)); stream.zalloc = Z_NULL; stream.zfree = Z_NULL; stream.opaque = Z_NULL; - if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK) { - ret = 1; - goto err; + if (deflateInit(&stream, Z_DEFAULT_COMPRESSION) != Z_OK) + return 1; + + while (!flist_empty(&log->io_logs)) { + struct io_logs *cur_log; + + cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); + flist_del_init(&cur_log->list); + + ret = __fio_append_iolog_gz(first, log, cur_log, &stream); + if (ret) + break; } - stream.next_in = (void *) log->log; - stream.avail_in = log->nr_samples * log_entry_sz(log); + ret = deflate(&stream, Z_FINISH); - do { - unsigned int this_len, flags = 0; + while (ret != Z_STREAM_END) { + struct sk_entry *entry; + unsigned int this_len; + void *out_pdu; + out_pdu = malloc(FIO_SERVER_MAX_FRAGMENT_PDU); stream.avail_out = FIO_SERVER_MAX_FRAGMENT_PDU; stream.next_out = out_pdu; + ret = deflate(&stream, Z_FINISH); /* may be Z_OK, or Z_STREAM_END */ - if (ret < 0) - goto err_zlib; + if (ret < 0) { + free(out_pdu); + break; + } this_len = FIO_SERVER_MAX_FRAGMENT_PDU - stream.avail_out; - if (stream.avail_in) - flags = FIO_NET_CMD_F_MORE; + entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, out_pdu, this_len, + NULL, SK_F_VEC | SK_F_INLINE | SK_F_FREE); + flist_add_tail(&entry->list, &first->next); + } while (ret != Z_STREAM_END); - ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG, - out_pdu, this_len, 0, flags); - if (ret) - goto err_zlib; - } while (stream.avail_in); + ret = deflateEnd(&stream); + if (ret == Z_OK) + return 0; -err_zlib: - deflateEnd(&stream); -err: - free(out_pdu); + return 1; +} +#else +static int fio_append_iolog_gz(struct sk_entry *first, struct io_log *log) +{ + return 1; +} #endif - return ret; + +static int fio_append_gz_chunks(struct sk_entry *first, struct io_log *log) +{ + struct sk_entry *entry; + struct flist_head *node; + + pthread_mutex_lock(&log->chunk_lock); + flist_for_each(node, &log->chunk_list) { + struct iolog_compress *c; + + c = flist_entry(node, struct iolog_compress, list); + entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, c->buf, c->len, + NULL, SK_F_VEC | SK_F_INLINE); + flist_add_tail(&entry->list, &first->next); + } + pthread_mutex_unlock(&log->chunk_lock); + + return 0; +} + +static int fio_append_text_log(struct sk_entry *first, struct io_log *log) +{ + struct sk_entry *entry; + + while (!flist_empty(&log->io_logs)) { + struct io_logs *cur_log; + size_t size; + + cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); + flist_del_init(&cur_log->list); + + size = cur_log->nr_samples * log_entry_sz(log); + + entry = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, cur_log->log, size, + NULL, SK_F_VEC | SK_F_INLINE); + flist_add_tail(&entry->list, &first->next); + } + + return 0; } int fio_send_iolog(struct thread_data *td, struct io_log *log, const char *name) { - struct sk_out *sk_out = pthread_getspecific(sk_out_key); struct cmd_iolog_pdu pdu; - int i, ret = 0; + struct sk_entry *first; + struct flist_head *entry; + int ret = 0; - pdu.nr_samples = cpu_to_le64(log->nr_samples); + pdu.nr_samples = cpu_to_le64(iolog_nr_samples(log)); pdu.thread_number = cpu_to_le32(td->thread_number); pdu.log_type = cpu_to_le32(log->log_type); - pdu.compressed = cpu_to_le32(use_zlib); + pdu.log_hist_coarseness = cpu_to_le32(log->hist_coarseness); + + if (!flist_empty(&log->chunk_list)) + pdu.compressed = __cpu_to_le32(STORE_COMPRESSED); + else if (use_zlib) + pdu.compressed = __cpu_to_le32(XMIT_COMPRESSED); + else + pdu.compressed = 0; strncpy((char *) pdu.name, name, FIO_NET_NAME_MAX); pdu.name[FIO_NET_NAME_MAX - 1] = '\0'; - for (i = 0; i < log->nr_samples; i++) { - struct io_sample *s = get_sample(log, i); + /* + * We can't do this for a pre-compressed log, but for that case, + * log->nr_samples is zero anyway. + */ + flist_for_each(entry, &log->io_logs) { + struct io_logs *cur_log; + int i; + + cur_log = flist_entry(entry, struct io_logs, list); + + for (i = 0; i < cur_log->nr_samples; i++) { + struct io_sample *s = get_sample(log, cur_log, i); - s->time = cpu_to_le64(s->time); - s->val = cpu_to_le64(s->val); - s->__ddir = cpu_to_le32(s->__ddir); - s->bs = cpu_to_le32(s->bs); + s->time = cpu_to_le64(s->time); + s->val = cpu_to_le64(s->val); + s->__ddir = cpu_to_le32(s->__ddir); + s->bs = cpu_to_le32(s->bs); - if (log->log_offset) { - struct io_sample_offset *so = (void *) s; + if (log->log_offset) { + struct io_sample_offset *so = (void *) s; - so->offset = cpu_to_le64(so->offset); + so->offset = cpu_to_le64(so->offset); + } } } /* - * Send header first, it's not compressed. + * Assemble header entry first */ - ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG, &pdu, - sizeof(pdu), 0, FIO_NET_CMD_F_MORE); - if (ret) - return ret; + first = fio_net_prep_cmd(FIO_NET_CMD_IOLOG, &pdu, sizeof(pdu), NULL, SK_F_VEC | SK_F_INLINE | SK_F_COPY); /* - * Now send actual log, compress if we can, otherwise just plain + * Now append actual log entries. If log compression was enabled on + * the job, just send out the compressed chunks directly. If we + * have a plain log, compress if we can, then send. Otherwise, send + * the plain text output. */ - if (use_zlib) - ret = fio_send_iolog_gz(sk_out, &pdu, log); - else { - ret = fio_send_cmd_ext_pdu(sk_out->sk, FIO_NET_CMD_IOLOG, - log->log, log->nr_samples * log_entry_sz(log), - 0, 0); - } + if (!flist_empty(&log->chunk_list)) + ret = fio_append_gz_chunks(first, log); + else if (use_zlib) + ret = fio_append_iolog_gz(first, log); + else + ret = fio_append_text_log(first, log); + fio_net_queue_entry(first); return ret; } @@ -1553,7 +1977,8 @@ void fio_server_send_add_job(struct thread_data *td) pdu.groupid = cpu_to_le32(td->groupid); convert_thread_options_to_net(&pdu.top, &td->o); - fio_net_queue_cmd(FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL, SK_F_COPY); + fio_net_queue_cmd(FIO_NET_CMD_ADD_JOB, &pdu, sizeof(pdu), NULL, + SK_F_COPY); } void fio_server_send_start(struct thread_data *td) @@ -1562,25 +1987,24 @@ void fio_server_send_start(struct thread_data *td) assert(sk_out->sk != -1); - fio_net_queue_cmd(FIO_NET_CMD_SERVER_START, NULL, 0, 0, SK_F_SIMPLE); + fio_net_queue_cmd(FIO_NET_CMD_SERVER_START, NULL, 0, NULL, SK_F_SIMPLE); } int fio_server_get_verify_state(const char *name, int threadnumber, - void **datap, int *version) + void **datap) { struct thread_io_list *s; struct cmd_sendfile out; struct cmd_reply *rep; uint64_t tag; void *data; + int ret; dprint(FD_NET, "server: request verify state\n"); rep = smalloc(sizeof(*rep)); - if (!rep) { - log_err("fio: smalloc pool too small\n"); - return 1; - } + if (!rep) + return ENOMEM; __fio_mutex_init(&rep->lock, FIO_MUTEX_LOCKED); rep->data = NULL; @@ -1589,23 +2013,27 @@ int fio_server_get_verify_state(const char *name, int threadnumber, verify_state_gen_name((char *) out.path, sizeof(out.path), name, me, threadnumber); tag = (uint64_t) (uintptr_t) rep; - fio_net_queue_cmd(FIO_NET_CMD_SENDFILE, &out, sizeof(out), &tag, SK_F_COPY); + fio_net_queue_cmd(FIO_NET_CMD_SENDFILE, &out, sizeof(out), &tag, + SK_F_COPY); /* * Wait for the backend to receive the reply */ if (fio_mutex_down_timeout(&rep->lock, 10000)) { log_err("fio: timed out waiting for reply\n"); + ret = ETIMEDOUT; goto fail; } if (rep->error) { - log_err("fio: failure on receiving state file: %s\n", strerror(rep->error)); + log_err("fio: failure on receiving state file %s: %s\n", + out.path, strerror(rep->error)); + ret = rep->error; fail: *datap = NULL; sfree(rep); fio_net_queue_quit(); - return 1; + return ret; } /* @@ -1613,12 +2041,15 @@ fail: * the header, and the thread_io_list checksum */ s = rep->data + sizeof(struct verify_state_hdr); - if (verify_state_hdr(rep->data, s, version)) + if (verify_state_hdr(rep->data, s)) { + ret = EILSEQ; goto fail; + } /* * Don't need the header from now, copy just the thread_io_list */ + ret = 0; rep->size -= sizeof(struct verify_state_hdr); data = malloc(rep->size); memcpy(data, s, rep->size); @@ -1627,7 +2058,7 @@ fail: sfree(rep->data); __fio_mutex_remove(&rep->lock); sfree(rep); - return 0; + return ret; } static int fio_init_server_ip(void) @@ -1655,11 +2086,10 @@ static int fio_init_server_ip(void) return -1; } #ifdef SO_REUSEPORT - if (setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) { - log_err("fio: setsockopt(REUSEPORT): %s\n", strerror(errno)); - close(sk); - return -1; - } + /* + * Not fatal if fails, so just ignore it if that happens + */ + setsockopt(sk, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)); #endif if (use_ipv6) { @@ -1761,7 +2191,7 @@ static int fio_init_server_connection(void) log_info("fio: server listening on %s\n", bind_str); - if (listen(sk, 0) < 0) { + if (listen(sk, 4) < 0) { log_err("fio: listen: %s\n", strerror(errno)); close(sk); return -1; @@ -1952,6 +2382,22 @@ static void set_sig_handlers(void) sigaction(SIGINT, &act, NULL); } +void fio_server_destroy_sk_key(void) +{ + pthread_key_delete(sk_out_key); +} + +int fio_server_create_sk_key(void) +{ + if (pthread_key_create(&sk_out_key, NULL)) { + log_err("fio: can't create sk_out backend key\n"); + return 1; + } + + pthread_setspecific(sk_out_key, NULL); + return 0; +} + static int fio_server(void) { int sk, ret; @@ -1967,9 +2413,6 @@ static int fio_server(void) set_sig_handlers(); - if (pthread_key_create(&sk_out_key, NULL)) - log_err("fio: can't create sk_out backend key\n"); - ret = accept_loop(sk); close(sk);