From: Andre Noll Date: Mon, 28 Jun 2010 11:17:48 +0000 (+0200) Subject: Merge branch 't/sched_debug' X-Git-Tag: v0.4.3~9 X-Git-Url: http://git.tuebingen.mpg.de/?a=commitdiff_plain;h=6b44f1fdf5fdfd18b9f77a87428a971eef0c547e;hp=e67116354084dcf587c03dccaeb18d8cee886c1b;p=paraslash.git Merge branch 't/sched_debug' --- diff --git a/Makefile.in b/Makefile.in index 34992de8..d46da62a 100644 --- a/Makefile.in +++ b/Makefile.in @@ -69,7 +69,7 @@ object_dir := objects man_dir := man/man1 m4_ggos := afh audioc audiod client filter gui recv server write -all_ggos := $(m4_ggos) dccp_recv oggdec_filter alsa_write oss_write fade http_recv \ +all_ggos := $(m4_ggos) dccp_recv alsa_write oss_write fade http_recv \ osx_write udp_recv amp_filter compress_filter file_write \ mp3dec_filter prebuffer_filter ggo_generated := $(addsuffix .ggo, $(addprefix $(ggo_dir)/,$(m4_ggos))) diff --git a/NEWS b/NEWS index c17a543b..a488aaff 100644 --- a/NEWS +++ b/NEWS @@ -1,6 +1,3 @@ -NEWS -==== - --------------------------------------------- 0.4.3 (to be announced) "imaginary radiation" --------------------------------------------- @@ -8,6 +5,8 @@ NEWS - Fix an end-of-file detection bug in the oggdec filter. - New user manual - The new nonblock API + - FEC support for the DCCP transport (Gerrit Renker). + - Both options of the oggdec filter have been removed. ------------------------------------------ 0.4.2 (2010-04-23) "associative expansion" diff --git a/aacdec_filter.c b/aacdec_filter.c index 209c9a46..f33f21eb 100644 --- a/aacdec_filter.c +++ b/aacdec_filter.c @@ -112,6 +112,7 @@ next_buffer: return; btr_merge(btrn, fn->min_iqs); len = btr_next_buffer(btrn, (char **)&inbuf); + len = PARA_MAX(len, (size_t)8192); consumed = 0; iqs = btr_get_input_queue_size(btrn); if (!padd->initialized) { diff --git a/aft.c b/aft.c index 25058e7c..eb0f7b57 100644 --- a/aft.c +++ b/aft.c @@ -21,7 +21,6 @@ #include "afh.h" #include "afs.h" #include "net.h" -#include "vss.h" #include "fd.h" #include "ipc.h" #include "portable_io.h" diff --git a/audiod.c b/audiod.c index 778318ce..5c17e6ea 100644 --- a/audiod.c +++ b/audiod.c @@ -863,7 +863,8 @@ static int init_default_filters(void) * If udp is used to receive this audiod format, add fecdec as * the first filter. */ - if (strcmp(afi[i].receiver->name, "udp") == 0) { + if (strcmp(afi[i].receiver->name, "udp") == 0 || + strcmp(afi[i].receiver->name, "dccp") == 0) { tmp = para_strdup("fecdec"); add_filter(i, tmp); free(tmp); diff --git a/buffer_tree.c b/buffer_tree.c index 1b859f52..3a59f8b5 100644 --- a/buffer_tree.c +++ b/buffer_tree.c @@ -1067,7 +1067,7 @@ struct btr_node *btr_search_node(const char *name, struct btr_node *root) } /** 640K ought to be enough for everybody ;) */ -#define BTRN_MAX_PENDING (640 * 1024) +#define BTRN_MAX_PENDING (96 * 1024) /** * Return the current state of a buffer tree node. diff --git a/chunk_queue.c b/chunk_queue.c index 3f5ac1d9..468b1dc2 100644 --- a/chunk_queue.c +++ b/chunk_queue.c @@ -11,7 +11,6 @@ #include "para.h" #include "list.h" #include "afh.h" -#include "vss.h" #include "string.h" #include "error.h" diff --git a/command.c b/command.c index b7795a00..42fa3e75 100644 --- a/command.c +++ b/command.c @@ -23,9 +23,9 @@ #include "afh.h" #include "afs.h" #include "server.h" -#include "vss.h" #include "list.h" #include "send.h" +#include "vss.h" #include "rc4.h" #include "net.h" #include "daemon.h" diff --git a/configure.ac b/configure.ac index 80eaf095..d8f55ce2 100644 --- a/configure.ac +++ b/configure.ac @@ -456,9 +456,6 @@ if test "$have_ogg" = "yes"; then audiod_ldflags="$audiod_ldflags $oggvorbis_libs -lvorbis -lvorbisfile" afh_ldflags="$afh_ldflags $oggvorbis_libs -logg -lvorbis -lvorbisfile" - filter_cmdline_objs="$filter_cmdline_objs add_cmdline(oggdec_filter)" - audiod_cmdline_objs="$audiod_cmdline_objs add_cmdline(oggdec_filter)" - server_errlist_objs="$server_errlist_objs ogg_afh" filter_errlist_objs="$filter_errlist_objs oggdec_filter" audiod_errlist_objs="$audiod_errlist_objs oggdec_filter" diff --git a/dccp_send.c b/dccp_send.c index 6248ae80..6cfdfbbe 100644 --- a/dccp_send.c +++ b/dccp_send.c @@ -24,19 +24,21 @@ #include "server.h" #include "net.h" #include "list.h" -#include "vss.h" #include "send.h" +#include "vss.h" #include "fd.h" #include "close_on_fork.h" #include "chunk_queue.h" #include "server.cmdline.h" #include "acl.h" -/** Do not write more than that many bytes at once. */ -#define DCCP_MAX_BYTES_PER_WRITE 1024 - static struct sender_status dccp_sender_status, *dss = &dccp_sender_status; +struct dccp_fec_client { + struct fec_client_parms fcp; + struct fec_client *fc; +}; + static void dccp_pre_select(int *max_fileno, fd_set *rfds, __a_unused fd_set *wfds) { @@ -65,9 +67,54 @@ static int dccp_get_tx_ccid(int sockfd) return tx_ccid; } +static void dccp_shutdown_client(struct sender_client *sc) +{ + struct dccp_fec_client *dfc = sc->private_data; + + vss_del_fec_client(dfc->fc); + shutdown_client(sc, dss); +} + +static void dccp_shutdown_clients(void) +{ + struct sender_client *sc, *tmp; + + list_for_each_entry_safe(sc, tmp, &dss->client_list, node) + dccp_shutdown_client(sc); +} + +/** * Obtain current MPS according to RFC 4340, sec. 14. */ +static int dccp_init_fec(struct sender_client *sc) +{ + int mps, ret; + socklen_t ml = sizeof(mps); + + /* If call fails, return some sensible minimum value */ + ret = getsockopt(sc->fd, SOL_DCCP, DCCP_SOCKOPT_GET_CUR_MPS, &mps, &ml); + if (ret < 0) { + PARA_NOTICE_LOG("can not determine MPS: %s\n", strerror(errno)); + mps = generic_max_transport_msg_size(sc->fd) - DCCP_MAX_HEADER; + } + PARA_INFO_LOG("current MPS = %d bytes\n", mps); + assert(mps > 0); + if (conf.dccp_max_slice_size_arg > 0) + mps = PARA_MIN(mps, conf.dccp_max_slice_size_arg); + return mps; +} + +static int dccp_send_fec(struct sender_client *sc, char *buf, size_t len) +{ + int ret = write_nonblock(sc->fd, buf, len); + + if (ret < 0) + dccp_shutdown_client(sc); + return ret; +} + static void dccp_post_select(fd_set *rfds, __a_unused fd_set *wfds) { struct sender_client *sc; + struct dccp_fec_client *dfc; int tx_ccid; sc = accept_sender_client(dss, rfds); @@ -87,26 +134,18 @@ static void dccp_post_select(fd_set *rfds, __a_unused fd_set *wfds) * from the client; by shutting down this unused communication path we can * reduce processing costs a bit. See analogous comment in dccp_recv.c. */ - if (shutdown(sc->fd, SHUT_RD) >= 0) + if (shutdown(sc->fd, SHUT_RD) < 0) { + PARA_WARNING_LOG("%s\n", strerror(errno)); + shutdown_client(sc, dss); return; - PARA_WARNING_LOG("%s\n", strerror(errno)); - shutdown_client(sc, dss); -} - -static void dccp_send(long unsigned current_chunk, - __a_unused long unsigned chunks_sent, const char *buf, - size_t len, const char *header_buf, size_t header_len) -{ - struct sender_client *sc, *tmp; - - list_for_each_entry_safe(sc, tmp, &dss->client_list, node) - send_chunk(sc, dss, DCCP_MAX_BYTES_PER_WRITE, current_chunk, buf, - len, header_buf, header_len); -} - -static void dccp_shutdown_clients(void) -{ - shutdown_clients(dss); + } + dfc = para_calloc(sizeof(*dfc)); + sc->private_data = dfc; + dfc->fcp.data_slices_per_group = conf.dccp_data_slices_per_group_arg; + dfc->fcp.slices_per_group = conf.dccp_slices_per_group_arg; + dfc->fcp.init_fec = dccp_init_fec; + dfc->fcp.send_fec = dccp_send_fec; + dfc->fc = vss_add_fec_client(sc, &dfc->fcp); } static int dccp_com_on(__a_unused struct sender_command_data *scd) @@ -173,10 +212,10 @@ static char *dccp_info(void) */ void dccp_send_init(struct sender *s) { - int ret; + int ret, k, n; s->info = dccp_info; - s->send = dccp_send; + s->send = NULL; s->pre_select = dccp_pre_select; s->post_select = dccp_post_select; s->shutdown_clients = dccp_shutdown_clients; @@ -188,6 +227,15 @@ void dccp_send_init(struct sender *s) s->client_cmds[SENDER_ADD] = NULL; s->client_cmds[SENDER_DELETE] = NULL; + k = conf.dccp_data_slices_per_group_arg; + n = conf.dccp_slices_per_group_arg; + + if (k <= 0 || n <= 0 || k >= n) { + PARA_WARNING_LOG("invalid FEC parameters, using defaults\n"); + conf.dccp_data_slices_per_group_arg = 3; + conf.dccp_slices_per_group_arg = 4; + } + init_sender_status(dss, conf.dccp_access_arg, conf.dccp_access_given, conf.dccp_port_arg, conf.dccp_max_clients_arg, conf.dccp_default_deny_given); diff --git a/fd.c b/fd.c index 7336bd51..978ada45 100644 --- a/fd.c +++ b/fd.c @@ -49,10 +49,6 @@ int write_all(int fd, const char *buf, size_t *len) * \param fd The file descriptor. * \param buf the buffer to write. * \param len the number of bytes of \a buf. - * \param max_bytes_per_write Do not write more than that many bytes at once. - * - * If \a max_bytes_per_write is non-zero, do not send more than that many bytes - * per write(). * * EAGAIN is not considered an error condition. For example CCID3 has a * sending wait queue which fills up and is emptied asynchronously. The EAGAIN @@ -61,8 +57,7 @@ int write_all(int fd, const char *buf, size_t *len) * * \return Negative on errors, number of bytes written else. */ -int write_nonblock(int fd, const char *buf, size_t len, - size_t max_bytes_per_write) +int write_nonblock(int fd, const char *buf, size_t len) { size_t written = 0; int ret = 0; @@ -70,8 +65,6 @@ int write_nonblock(int fd, const char *buf, size_t len, while (written < len) { size_t num = len - written; - if (max_bytes_per_write && max_bytes_per_write < num) - num = max_bytes_per_write; ret = write(fd, buf + written, num); if (ret < 0 && errno == EAGAIN) return written; diff --git a/fd.h b/fd.h index c21a7d14..00b02bef 100644 --- a/fd.h +++ b/fd.h @@ -30,7 +30,6 @@ int readv_nonblock(int fd, struct iovec *iov, int iovcnt, fd_set *rfds, size_t *num_bytes); int read_nonblock(int fd, void *buf, size_t sz, fd_set *rfds, size_t *num_bytes); int read_pattern(int fd, const char *pattern, size_t bufsize, fd_set *rfds); -int write_nonblock(int fd, const char *buf, size_t len, - size_t max_bytes_per_write); +int write_nonblock(int fd, const char *buf, size_t len); int for_each_file_in_dir(const char *dirname, int (*func)(const char *, void *), void *private_data); diff --git a/ggo/oggdec_filter.ggo b/ggo/oggdec_filter.ggo deleted file mode 100644 index e2d653d7..00000000 --- a/ggo/oggdec_filter.ggo +++ /dev/null @@ -1,21 +0,0 @@ -option "bufsize" b -#~~~~~~~~~~~~~~~~~ -"size of output buffer" -int typestr="kilobyte" -default="128" -optional -details=" - Increase this if you encounter output buffer overrun errors. Smaller - values make the oggdec filter use less memory. -" - -option "initial_buffer" i -#~~~~~~~~~~~~~~~~~~~~~~~~ -"size of initial input buffer" -int typestr="kilobyte" -default="16" -optional -details=" - On startup, defer decoding until that many kilobytes are - available in the input buffer. -" diff --git a/ggo/server.m4 b/ggo/server.m4 index 4be4e2c9..7f2d9759 100644 --- a/ggo/server.m4 +++ b/ggo/server.m4 @@ -238,6 +238,56 @@ details=" See http_max_clients for details. " +option "dccp_max_slice_size" - +#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +"Upper bound for the FEC slice size" +int typestr = "size" +optional +default = "0" +details = " + If this value is non-positive (the default) the dccp sender + uses the maximum packet size (MPS) of the connection as the + slice size. The MPS is a network parameter and depends on + the path maximum transmission unit (path MTU) of an incoming + connection, i.e. on the largest packet size that can be + transmitted without causing fragmentation. + + This option allows to use a value less than the MPS in order + to fine-tune application performance. Values greater than + the MPS of an incoming connection can not be set. +" + +option "dccp_data_slices_per_group" - +#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +"The number of non-redundant slices per FEC group" +int typestr = "num" +optional +default = "3" +details = " + This determines the number of slices in each FEC group that are + necessary to decode the group. The given number must be smaller + than the value of the dccp_slices_per_group option below. + + Note that the duration of a FEC group is proportional to the + product dccp_max_slice_size * dccp_data_slices_per_group. +" + +option "dccp_slices_per_group" - +#~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +"The total number of slices per FEC group" +int typestr = "num" +optional +default = "4" +details = " + This value must be larger than the value given for above + dccp_data_slices_per_group above. The difference being the + number of redundant slices per group, i.e. the number of + data packets that may be lost without causing interruptions + of the resulting audio stream. + + Increase this value if for lossy networks. +" + #################### section "udp sender" #################### diff --git a/grab_client.c b/grab_client.c index 9b475316..5971f115 100644 --- a/grab_client.c +++ b/grab_client.c @@ -78,7 +78,7 @@ static int gc_write(struct grab_client *gc, char *buf, size_t len) if (gc->mode == GM_SLOPPY) return len; } - ret = write_nonblock(gc->fd, buf, len, 0); + ret = write_nonblock(gc->fd, buf, len); if (ret < 0) goto err; if (ret > 0) diff --git a/http_send.c b/http_send.c index 424d63b2..e8b22cf9 100644 --- a/http_send.c +++ b/http_send.c @@ -19,9 +19,9 @@ #include "afs.h" #include "server.h" #include "http.h" -#include "vss.h" #include "list.h" #include "send.h" +#include "vss.h" #include "close_on_fork.h" #include "net.h" #include "fd.h" @@ -77,6 +77,66 @@ static void http_shutdown_clients(void) shutdown_clients(hss); } +static int queue_chunk_or_shutdown(struct sender_client *sc, + struct sender_status *ss, const char *buf, size_t num_bytes) +{ + int ret = cq_enqueue(sc->cq, buf, num_bytes); + if (ret < 0) + shutdown_client(sc, ss); + return ret; +} + +/** + * Send one chunk of audio data to a connected client. + * + * \param sc The client. + * \param ss The sender. + * \param current_chunk The number of the chunk to write. + * \param buf The data to write. + * \param len The number of bytes of \a buf. + * \param header_buf The audio file header. + * \param header_len The number of bytes of \a header_buf. + * + * On errors, the client is shut down. If only a part of the buffer could be + * written, the remainder is put into the chunk queue for that client. + */ +static void http_send_chunk(struct sender_client *sc, struct sender_status *ss, + long unsigned current_chunk, const char *buf, size_t len, + const char *header_buf, size_t header_len) +{ + int ret; + + if (!sc->header_sent && current_chunk) { + if (header_buf && header_len > 0) { + ret = queue_chunk_or_shutdown(sc, ss, header_buf, header_len); + if (ret < 0) + goto out; + } + } + sc->header_sent = 1; + ret = send_queued_chunks(sc->fd, sc->cq); + if (ret < 0) { + shutdown_client(sc, ss); + goto out; + } + if (!len) + goto out; + if (!ret) { /* still data left in the queue */ + ret = queue_chunk_or_shutdown(sc, ss, buf, len); + goto out; + } + ret = write_nonblock(sc->fd, buf, len); + if (ret < 0) { + shutdown_client(sc, ss); + goto out; + } + if (ret != len) + ret = queue_chunk_or_shutdown(sc, ss, buf + ret, len - ret); +out: + if (ret < 0) + PARA_NOTICE_LOG("%s\n", para_strerror(-ret)); +} + static void http_send(long unsigned current_chunk, __a_unused long unsigned chunks_sent, const char *buf, size_t len, const char *header_buf, size_t header_len) @@ -85,10 +145,10 @@ static void http_send(long unsigned current_chunk, list_for_each_entry_safe(sc, tmp, &hss->client_list, node) { struct private_http_sender_data *phsd = sc->private_data; - if (phsd->status != HTTP_STREAMING) - continue; - send_chunk(sc, hss, 0, current_chunk, buf, len, header_buf, - header_len); + + if (phsd->status == HTTP_STREAMING) + http_send_chunk(sc, hss, current_chunk, buf, len, + header_buf, header_len); } } diff --git a/mp3dec_filter.c b/mp3dec_filter.c index 3344ea41..35448da7 100644 --- a/mp3dec_filter.c +++ b/mp3dec_filter.c @@ -140,6 +140,12 @@ next_buffer: return; btr_merge(btrn, fn->min_iqs); len = btr_next_buffer(btrn, &inbuffer); + /* + * Decode at most 8K in one go to give the post_select() functions of + * other buffer tree nodes a chance to run. This is necessary to avoid + * buffer underruns on slow machines. + */ + len = PARA_MIN(len, (size_t)8192); mad_stream_buffer(&pmd->stream, (unsigned char *)inbuffer, len); next_frame: ret = mad_header_decode(&pmd->frame.header, &pmd->stream); diff --git a/net.c b/net.c index 59b7f367..fa9575a6 100644 --- a/net.c +++ b/net.c @@ -521,6 +521,14 @@ static socklen_t salen(const struct sockaddr *sa) : sizeof(struct sockaddr_in); } +/** True if @ss holds a v6-mapped-v4 address (RFC 4291, 2.5.5.2) */ +static bool SS_IS_ADDR_V4MAPPED(const struct sockaddr_storage *ss) +{ + const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss; + + return ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr); +} + /** * Process IPv4/v6 address, turn v6-mapped-v4 address into normal IPv4 address. * \param ss Container of IPv4/6 address. @@ -531,11 +539,10 @@ static socklen_t salen(const struct sockaddr *sa) static const struct sockaddr * normalize_ip_address(const struct sockaddr_storage *ss) { - const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss; - assert(ss->ss_family == AF_INET || ss->ss_family == AF_INET6); - if (ss->ss_family == AF_INET6 && IN6_IS_ADDR_V4MAPPED(&ia6->sin6_addr)) { + if (SS_IS_ADDR_V4MAPPED(ss)) { + const struct sockaddr_in6 *ia6 = (const struct sockaddr_in6 *)ss; static struct sockaddr_in ia; ia.sin_family = AF_INET; @@ -546,6 +553,46 @@ normalize_ip_address(const struct sockaddr_storage *ss) return (const struct sockaddr *)ss; } +/** + * Generic/fallback MTU values + * + * These are taken from RFC 1122, RFC 2460, and RFC 5405. + * - RFC 1122, 3.3.3 defines EMTU_S ("Effective MTU for sending") and recommends + * to use an EMTU_S size of of 576 bytes if the IPv4 path MTU is unknown; + * - RFC 2460, 5. requires a minimum IPv6 MTU of 1280 bytes; + * - RFC 5405, 3.2 recommends that if path MTU discovery is not done, + * UDP senders should use the respective minimum values of EMTU_S. + */ +static inline int generic_mtu(const int af_type) +{ + return af_type == AF_INET6 ? 1280 : 576; +} + +/** Crude approximation of IP header overhead - neglecting options. */ +static inline int estimated_header_overhead(const int af_type) +{ + return af_type == AF_INET6 ? 40 : 20; +} + +/** + * Maximum transport-layer message size (MMS_S) as per RFC 1122, 3.3.3 + * Socket must be connected. + */ +int generic_max_transport_msg_size(int sockfd) +{ + struct sockaddr_storage ss; + socklen_t sslen = sizeof(ss); + int af_type = AF_INET; + + if (getpeername(sockfd, (struct sockaddr *)&ss, &sslen) < 0) { + PARA_ERROR_LOG("can not determine remote address type: %s\n", + strerror(errno)); + } else if (!SS_IS_ADDR_V4MAPPED(&ss)) { + af_type = ss.ss_family; + } + return generic_mtu(af_type) - estimated_header_overhead(af_type); +} + /** * Print numeric host and port number (beware - uses static char). * diff --git a/net.h b/net.h index 457c24dc..9da63ba1 100644 --- a/net.h +++ b/net.h @@ -36,6 +36,10 @@ #define SOL_DCCP 269 /**< Linux socket level. */ #endif +#ifndef DCCP_SOCKOPT_GET_CUR_MPS +#define DCCP_SOCKOPT_GET_CUR_MPS 5 /**< Max packet size, RFC 4340, 14. */ +#endif + #ifndef DCCP_SOCKOPT_AVAILABLE_CCIDS #define DCCP_SOCKOPT_AVAILABLE_CCIDS 12 /**< List of supported CCIDs. */ #endif @@ -132,6 +136,11 @@ static inline int para_listen_simple(unsigned l4type, uint16_t port) extern char *local_name(int sockfd); extern char *remote_name(int sockfd); +/** + * Determining maximum payload (packet) size + */ +extern int generic_max_transport_msg_size(int sockfd); + int send_bin_buffer(int, const char *, size_t); int send_buffer(int, const char *); __printf_2_3 int send_va_buffer(int fd, const char *fmt, ...); @@ -149,6 +158,8 @@ ssize_t send_cred_buffer(int, char*); /** * Functions and definitions to support \p IPPROTO_DCCP */ -/** Hardcoded maximum number of separate CCID modules compiled into a host */ +/** Estimated worst-case length of a DCCP header including options. */ +#define DCCP_MAX_HEADER 128 +/** Hardcoded maximum number of separate CCID modules compiled into a host. */ #define DCCP_MAX_HOST_CCIDS 20 extern int dccp_available_ccids(uint8_t **ccid_array); diff --git a/oggdec_filter.c b/oggdec_filter.c index 2f1fb787..0130afd0 100644 --- a/oggdec_filter.c +++ b/oggdec_filter.c @@ -11,7 +11,6 @@ #include #include "para.h" -#include "oggdec_filter.cmdline.h" #include "list.h" #include "sched.h" #include "ggo.h" @@ -227,7 +226,7 @@ static void ogg_post_select(__a_unused struct sched *s, struct task *t) if (read_ret < 0) goto out; btr_add_output(out, read_ret, btrn); - if (btr_get_output_queue_size(btrn) > 640 * 1024) + if (btr_get_output_queue_size(btrn) > 128 * 1024) return; /* enough data for the moment */ } out: @@ -237,36 +236,6 @@ out: } } -static int oggdec_parse_config(int argc, char **argv, void **config) -{ - int ret; - struct oggdec_filter_args_info *ogg_conf; - - ogg_conf = para_calloc(sizeof(*ogg_conf)); - ret = -E_OGGDEC_SYNTAX; - if (oggdec_cmdline_parser(argc, argv, ogg_conf)) - goto err; - ret = -ERRNO_TO_PARA_ERROR(EINVAL); - if (ogg_conf->bufsize_arg < 0) - goto err; - if (ogg_conf->bufsize_arg >= INT_MAX / 1024) - goto err; - if (ogg_conf->initial_buffer_arg < 0) - goto err; - if (ogg_conf->initial_buffer_arg >= INT_MAX / 1024) - goto err; - *config = ogg_conf; - return 1; -err: - free(ogg_conf); - return ret; -} - -static void oggdec_free_config(void *conf) -{ - oggdec_cmdline_parser_free(conf); -} - /** * The init function of the ogg vorbis decoder. * @@ -274,18 +243,9 @@ static void oggdec_free_config(void *conf) */ void oggdec_filter_init(struct filter *f) { - struct oggdec_filter_args_info dummy; - - oggdec_cmdline_parser_init(&dummy); f->open = ogg_open; f->close = ogg_close; f->pre_select = generic_filter_pre_select; f->post_select = ogg_post_select; - f->parse_config = oggdec_parse_config; - f->free_config = oggdec_free_config; f->execute = oggdec_execute; - f->help = (struct ggo_help) { - .short_help = oggdec_filter_args_info_help, - .detailed_help = oggdec_filter_args_info_detailed_help - }; } diff --git a/oss_write.c b/oss_write.c index d0cff015..6525afa4 100644 --- a/oss_write.c +++ b/oss_write.c @@ -174,6 +174,7 @@ static void oss_post_select(__a_unused struct sched *s, goto out; return; } + btr_merge(btrn, wn->min_iqs); bytes = btr_next_buffer(btrn, &data); frames = bytes / powd->bytes_per_frame; if (!frames) { /* eof and less than a single frame available */ @@ -183,7 +184,7 @@ static void oss_post_select(__a_unused struct sched *s, ret = 0; if (!FD_ISSET(powd->fd, &s->wfds)) goto out; - ret = write_nonblock(powd->fd, data, frames * powd->bytes_per_frame, 0); + ret = write_nonblock(powd->fd, data, frames * powd->bytes_per_frame); if (ret < 0) goto out; btr_consume(btrn, ret); diff --git a/send.h b/send.h index acf62db4..df0cd1cb 100644 --- a/send.h +++ b/send.h @@ -47,7 +47,7 @@ struct sender { * only differ if the stream was repositioned by the \a ff or \a jmp * command. Of course, \a buf is a pointer to the chunk of data which * should be sent, and \a len is the length of this buffer. - */ + */ void (*send)(long unsigned current_chunk, long unsigned chunks_sent, const char *buf, size_t len, const char *header_buf, size_t header_len); @@ -105,6 +105,41 @@ struct sender_client { void *private_data; }; +/** + * Each paraslash sender may register arbitrary many clients to the virtual + * streaming system, possibly with varying fec parameters. In order to do so, + * it must allocate a \a fec_client_parms structure and pass it to \ref + * vss_add_fec_client. + * + * Clients are automatically removed from that list by the vss if an error + * occurs, or if the sender requests deletion of a client by calling \ref + * vss_del_fec_client(). + */ + +/** FEC parameters requested by FEC clients. */ +struct fec_client_parms { + /** Number of data slices plus redundant slices. */ + uint8_t slices_per_group; + /** Number of slices minus number of redundant slices. */ + uint8_t data_slices_per_group; + /** Maximal number of bytes per slice, initially zero. */ + uint16_t max_slice_bytes; + /** + * Transport-layer initialisation for FEC support. + * + * This optional function serves (a) to make the transport layer + * ready to send FEC slices and (b) to determine the Maximum + * Packet Size (MPS) supported by the connection. The MPS value + * determines the largest payload size. This value is used to + * send FEC slices that are not larger than the path MTU, to avoid + * fragmentation and to maximize packet utilization. The user can + * alternatively specify a slice size of up to this value. + */ + int (*init_fec)(struct sender_client *sc); + /** Push out FEC-encoded packets */ + int (*send_fec)(struct sender_client *sc, char *buf, size_t len); +}; + /** Describes the current status of one paraslash sender. */ struct sender_status { /** The file descriptor of the socket this sender is listening on. */ @@ -125,10 +160,6 @@ struct sender_status { void shutdown_client(struct sender_client *sc, struct sender_status *ss); void shutdown_clients(struct sender_status *ss); -void send_chunk(struct sender_client *sc, struct sender_status *ss, - size_t max_bytes_per_write, long unsigned current_chunk, - const char *buf, size_t len, const char *header_buf, - size_t header_len); void init_sender_status(struct sender_status *ss, char **access_arg, int num_access_args, int port, int max_clients, int default_deny); char *get_sender_info(struct sender_status *ss, const char *name); @@ -141,6 +172,5 @@ int generic_com_on(struct sender_status *ss, unsigned protocol); void generic_com_off(struct sender_status *ss); char *generic_sender_help(void); struct sender_client *accept_sender_client(struct sender_status *ss, fd_set *rfds); -int send_queued_chunks(int fd, struct chunk_queue *cq, - size_t max_bytes_per_write); +int send_queued_chunks(int fd, struct chunk_queue *cq); int parse_fec_url(const char *arg, struct sender_command_data *scd); diff --git a/send_common.c b/send_common.c index b44c8133..dea96081 100644 --- a/send_common.c +++ b/send_common.c @@ -96,34 +96,24 @@ void shutdown_clients(struct sender_status *ss) shutdown_client(sc, ss); } -static int queue_chunk_or_shutdown(struct sender_client *sc, - struct sender_status *ss, const char *buf, size_t num_bytes) -{ - int ret = cq_enqueue(sc->cq, buf, num_bytes); - if (ret < 0) - shutdown_client(sc, ss); - return ret; -} - /** * Try to empty the chunk queue for this fd. * * \param fd The file descriptor. * \param cq The list of queued chunks. - * \param max_bytes_per_write Do not send more than this in one go. * * \return Negative on errors, zero if not everything was sent, one otherwise. */ -int send_queued_chunks(int fd, struct chunk_queue *cq, - size_t max_bytes_per_write) +int send_queued_chunks(int fd, struct chunk_queue *cq) { struct queued_chunk *qc; while ((qc = cq_peek(cq))) { const char *buf; size_t len; int ret; + cq_get(qc, &buf, &len); - ret = write_nonblock(fd, buf, len, max_bytes_per_write); + ret = write_nonblock(fd, buf, len); if (ret < 0) return ret; cq_update(cq, ret); @@ -134,59 +124,6 @@ int send_queued_chunks(int fd, struct chunk_queue *cq, return 1; } -/** - * Send one chunk of audio data to a connected client. - * - * \param sc The client. - * \param ss The sender. - * \param max_bytes_per_write Split writes to chunks of at most that many bytes. - * \param current_chunk The number of the chunk to write. - * \param buf The data to write. - * \param len The number of bytes of \a buf. - * \param header_buf The audio file header. - * \param header_len The number of bytes of \a header_buf. - * - * On errors, the client is shut down. If only a part of the buffer could be - * written, the remainder is put into the chunk queue for that client. - */ -void send_chunk(struct sender_client *sc, struct sender_status *ss, - size_t max_bytes_per_write, long unsigned current_chunk, - const char *buf, size_t len, const char *header_buf, - size_t header_len) -{ - int ret; - - if (!sc->header_sent && current_chunk) { - if (header_buf && header_len > 0) { - ret = queue_chunk_or_shutdown(sc, ss, header_buf, header_len); - if (ret < 0) - goto out; - } - } - sc->header_sent = 1; - ret = send_queued_chunks(sc->fd, sc->cq, max_bytes_per_write); - if (ret < 0) { - shutdown_client(sc, ss); - goto out; - } - if (!len) - goto out; - if (!ret) { /* still data left in the queue */ - ret = queue_chunk_or_shutdown(sc, ss, buf, len); - goto out; - } - ret = write_nonblock(sc->fd, buf, len, max_bytes_per_write); - if (ret < 0) { - shutdown_client(sc, ss); - goto out; - } - if (ret != len) - ret = queue_chunk_or_shutdown(sc, ss, buf + ret, len - ret); -out: - if (ret < 0) - PARA_NOTICE_LOG("%s\n", para_strerror(-ret)); -} - /** * Initialize a struct sender status. * @@ -480,7 +417,7 @@ int parse_fec_url(const char *arg, struct sender_command_data *scd) goto out; } /* use default fec parameters. */ - scd->max_slice_bytes = 1472; + scd->max_slice_bytes = 0; scd->slices_per_group = 16; scd->data_slices_per_group = 14; ret = 0; diff --git a/server.c b/server.c index bc143039..0e256053 100644 --- a/server.c +++ b/server.c @@ -77,11 +77,11 @@ #include "string.h" #include "afs.h" #include "server.h" +#include "list.h" +#include "send.h" #include "vss.h" #include "config.h" #include "close_on_fork.h" -#include "list.h" -#include "send.h" #include "net.h" #include "daemon.h" #include "ipc.h" diff --git a/stdout.c b/stdout.c index eaf0f99f..775f0649 100644 --- a/stdout.c +++ b/stdout.c @@ -70,7 +70,7 @@ static void stdout_post_select(struct sched *s, struct task *t) sz = btr_next_buffer(btrn, &buf); if (sz == 0) break; - ret = write_nonblock(STDOUT_FILENO, buf, sz, 0); + ret = write_nonblock(STDOUT_FILENO, buf, sz); if (ret <= 0) break; btr_consume(btrn, ret); diff --git a/udp_send.c b/udp_send.c index e80670d0..24ebf584 100644 --- a/udp_send.c +++ b/udp_send.c @@ -11,6 +11,7 @@ #include #include #include +#include #include #include @@ -21,9 +22,9 @@ #include "afh.h" #include "afs.h" #include "server.h" -#include "vss.h" #include "list.h" #include "send.h" +#include "vss.h" #include "portable_io.h" #include "net.h" #include "fd.h" @@ -31,18 +32,23 @@ #include "close_on_fork.h" #include "chunk_queue.h" +/** + * Time window during which ICMP Destination/Port Unreachable messages are + * ignored, covering transient receiver problems such as restarting the + * client, rebooting, reconfiguration, or handover. + */ +#define UDP_MAX_UNREACHABLE_TIME 30 + /** Describes one entry in the list of targets for the udp sender. */ struct udp_target { - /** The position of this target in the list of targets. */ - struct list_head node; /** The hostname (DNS name or IPv4/v6 address string). */ char host[MAX_HOSTLEN]; /** The UDP port. */ int port; - /** The socket fd. */ - int fd; - /** The list of queued chunks for this fd. */ - struct chunk_queue *cq; + /** Track time (seconds) of last ICMP Port Unreachable error */ + time_t last_unreachable; + /** Common sender client data */ + struct sender_client *sc; /** The opaque structure returned by vss_add_fec_client(). */ struct fec_client *fc; /** The FEC parameters for this target. */ @@ -52,50 +58,55 @@ struct udp_target { static struct list_head targets; static int sender_status; -static void udp_close_target(struct udp_target *ut) +static void udp_close_target(struct sender_client *sc) { - if (ut->fd < 0) - return; - close(ut->fd); - del_close_on_fork_list(ut->fd); - cq_destroy(ut->cq); - ut->cq = NULL; - ut->fd = -1; + if (sc->cq != NULL) { + del_close_on_fork_list(sc->fd); + cq_destroy(sc->cq); + free(sc->name); + sc->name = NULL; + sc->cq = NULL; + + } } static void udp_delete_target(struct udp_target *ut, const char *msg) { + PARA_NOTICE_LOG("deleting %s#%d (%s) from list\n", ut->host, ut->port, msg); - udp_close_target(ut); + udp_close_target(ut->sc); vss_del_fec_client(ut->fc); - list_del(&ut->node); + list_del(&ut->sc->node); + free(ut->sc); free(ut); } /** * Perform AF-independent multicast sender setup. * - * \param fd The connected socket descriptor. - * \param ttl UDPv4 multicast TTL or UDPv6 multicast number of hops. - * Use -1 to mean default, 0..255 otherwise. - * \param iface The outgoing multicast interface, or NULL for the default. + * \param sc The connected sender client. * * \return Zero if okay, negative on error. */ -static int mcast_sender_setup(struct udp_target *ut, int ttl, char *iface) +static int mcast_sender_setup(struct sender_client *sc) { struct sockaddr_storage ss; socklen_t sslen = sizeof(ss); - + int ttl = conf.udp_ttl_arg, id = 0; const int on = 1; - int id = iface == NULL ? 0 : if_nametoindex(iface); - if (getpeername(ut->fd, (struct sockaddr *)&ss, &sslen) < 0) - goto err; + if (conf.udp_mcast_iface_given) { + char *iface = conf.udp_mcast_iface_arg; + + id = if_nametoindex(iface); + if (id == 0) + PARA_WARNING_LOG("could not resolve interface '%s', " + "using default", iface); + } - if (iface != NULL && id == 0) - PARA_WARNING_LOG("could not resolve interface %s, using default", iface); + if (getpeername(sc->fd, (struct sockaddr *)&ss, &sslen) < 0) + goto err; /* RFC 3493, 5.2: -1 means 'use kernel default' */ if (ttl < 0 || ttl > 255) @@ -111,7 +122,7 @@ static int mcast_sender_setup(struct udp_target *ut, int ttl, char *iface) memset(&mn, 0, sizeof(mn)); mn.imr_ifindex = id; - if (setsockopt(ut->fd, IPPROTO_IP, IP_MULTICAST_IF, &mn, sizeof(mn)) < 0) + if (setsockopt(sc->fd, IPPROTO_IP, IP_MULTICAST_IF, &mn, sizeof(mn)) < 0) goto err; #else PARA_ERROR_LOG("No support for setting outgoing IPv4 mcast interface."); @@ -121,24 +132,24 @@ static int mcast_sender_setup(struct udp_target *ut, int ttl, char *iface) * Enable receiving multicast messages generated on the local host * At least on Linux, this is enabled by default. */ - if (setsockopt(ut->fd, IPPROTO_IP, IP_MULTICAST_LOOP, &on, sizeof(on)) < 0) + if (setsockopt(sc->fd, IPPROTO_IP, IP_MULTICAST_LOOP, &on, sizeof(on)) < 0) break; /* Default: use local subnet (do not flood out into the WAN) */ if (ttl == -1) ttl = 1; - if (setsockopt(ut->fd, IPPROTO_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)) < 0) + if (setsockopt(sc->fd, IPPROTO_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)) < 0) break; return 0; case AF_INET6: if (!IN6_IS_ADDR_MULTICAST(&((struct sockaddr_in6 *)&ss)->sin6_addr)) return 0; if (id != 0 && - setsockopt(ut->fd, IPPROTO_IPV6, IPV6_MULTICAST_IF, &id, sizeof(id)) < 0) + setsockopt(sc->fd, IPPROTO_IPV6, IPV6_MULTICAST_IF, &id, sizeof(id)) < 0) break; - if (setsockopt(ut->fd, IPPROTO_IPV6, IPV6_MULTICAST_LOOP, &on, sizeof(on)) < 0) + if (setsockopt(sc->fd, IPPROTO_IPV6, IPV6_MULTICAST_LOOP, &on, sizeof(on)) < 0) break; - if (setsockopt(ut->fd, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &ttl, sizeof(ttl)) < 0) + if (setsockopt(sc->fd, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &ttl, sizeof(ttl)) < 0) break; return 0; default: @@ -152,55 +163,28 @@ err: /** The maximal size of the per-target chunk queue. */ #define UDP_CQ_BYTES 40000 -static int udp_init_session(struct udp_target *ut) +static void udp_init_session(struct sender_client *sc) { - int ret; - char *iface = NULL; - - if (ut->fd >= 0) /* nothing to do */ - return 0; - - ret = para_connect_simple(IPPROTO_UDP, ut->host, ut->port); - if (ret < 0) - return ret; - ut->fd = ret; - - if (conf.udp_mcast_iface_given) - iface = conf.udp_mcast_iface_arg; - - ret = mcast_sender_setup(ut, conf.udp_ttl_arg, iface); - if (ret < 0) { - close(ut->fd); - return ret; + if (sc->cq == NULL) { + add_close_on_fork_list(sc->fd); + sc->cq = cq_new(UDP_CQ_BYTES); + sc->name = para_strdup(remote_name(sc->fd)); + PARA_NOTICE_LOG("sending to udp %s\n", sc->name); } - - ret = mark_fd_nonblocking(ut->fd); - if (ret < 0) { - close(ut->fd); - return ret; - } - add_close_on_fork_list(ut->fd); - ut->cq = cq_new(UDP_CQ_BYTES); - PARA_NOTICE_LOG("sending to udp %s#%d\n", ut->host, ut->port); - return 1; } static void udp_shutdown_targets(void) { - struct udp_target *ut, *tmp; - const char *buf = NULL; - size_t len = 0; /* STFU, gcc */ - - list_for_each_entry_safe(ut, tmp, &targets, node) { - int ubuntu_glibc_headers_suck; - if (ut->fd < 0) - continue; - if (!buf) - len = vss_get_fec_eof_packet(&buf); - /* ignore return value, we're closing the target anyway. */ - ubuntu_glibc_headers_suck = write(ut->fd, buf, len); /* STFU */ - udp_close_target(ut); - } + struct sender_client *sc, *tmp; + const char *buf; + size_t len = vss_get_fec_eof_packet(&buf); + + list_for_each_entry_safe(sc, tmp, &targets, node) + if (sc->cq != NULL) { + /* ignore return value, closing the target anyway. */ + (void)write(sc->fd, buf, len); + udp_close_target(sc); + } } static int udp_com_on(__a_unused struct sender_command_data *scd) @@ -218,9 +202,11 @@ static int udp_com_off(__a_unused struct sender_command_data *scd) static int udp_com_delete(struct sender_command_data *scd) { - struct udp_target *ut, *tmp; + struct sender_client *sc, *tmp; + + list_for_each_entry_safe(sc, tmp, &targets, node) { + struct udp_target *ut = sc->private_data; - list_for_each_entry_safe(ut, tmp, &targets, node) { /* Unspecified port means wildcard port match */ if (scd->port > 0 && scd->port != ut->port) continue; @@ -231,34 +217,81 @@ static int udp_com_delete(struct sender_command_data *scd) return 1; } -static int udp_send_fec(char *buf, size_t len, void *private_data) +/** Initialize UDP session and set maximum payload size. */ +static int udp_init_fec(struct sender_client *sc) +{ + int mps; + + udp_init_session(sc); + mps = generic_max_transport_msg_size(sc->fd) - sizeof(struct udphdr); + PARA_INFO_LOG("current MPS = %d bytes\n", mps); + return mps; +} + +/** Check and clear socket error if any. */ +static int udp_check_socket_state(struct udp_target *ut) { - struct udp_target *ut = private_data; + int ret; + socklen_t errlen = sizeof(ret); + + if (getsockopt(ut->sc->fd, SOL_SOCKET, SO_ERROR, &ret, &errlen) < 0) { + PARA_ERROR_LOG("SO_ERROR failed: %s\n", strerror(ret)); + return 0; + } else if (ret == 0) { + return 0; + } else if (ret == ECONNREFUSED) { + time_t dist = now->tv_sec - ut->last_unreachable; + + if (dist <= UDP_MAX_UNREACHABLE_TIME) { + return 0; + } else if (dist > 2 * UDP_MAX_UNREACHABLE_TIME) { + ut->last_unreachable = now->tv_sec; + return 0; + } else { + /* + * unreachable_time < dist <= 2 * unreachable_time + * No errors are allowed during this time window. + */ + PARA_NOTICE_LOG("Evicting %s#%d after %d seconds " + "of connection errors.\n", + ut->host, ut->port, (int)dist); + } + } + return -ERRNO_TO_PARA_ERROR(ret); +} + +static int udp_send_fec(struct sender_client *sc, char *buf, size_t len) +{ + struct udp_target *ut = sc->private_data; int ret; if (sender_status == SENDER_OFF) return 0; - ret = udp_init_session(ut); + if (len == 0 && !cq_peek(ut->sc->cq)) + return 0; + ret = udp_check_socket_state(ut); if (ret < 0) goto fail; - ret = send_queued_chunks(ut->fd, ut->cq, 0); - if (ret == -ERRNO_TO_PARA_ERROR(ECONNREFUSED)) - ret = 0; + ret = send_queued_chunks(sc->fd, sc->cq); if (ret < 0) goto fail; - if (!len) - return 0; if (!ret) { /* still data left in the queue */ - ret = cq_force_enqueue(ut->cq, buf, len); + ret = cq_force_enqueue(sc->cq, buf, len); assert(ret >= 0); + return 0; } - ret = write_nonblock(ut->fd, buf, len, 0); - if (ret == -ERRNO_TO_PARA_ERROR(ECONNREFUSED)) + ret = write_nonblock(sc->fd, buf, len); + if (ret == -ERRNO_TO_PARA_ERROR(ECONNREFUSED)) { + /* + * Happens if meanwhile an ICMP Destination / Port Unreachable + * has arrived. Ignore, persistent errors will be caught above. + */ ret = 0; + } if (ret < 0) goto fail; if (ret != len) { - ret = cq_force_enqueue(ut->cq, buf + ret, len - ret); + ret = cq_force_enqueue(sc->cq, buf + ret, len - ret); assert(ret >= 0); } return 1; @@ -269,19 +302,43 @@ fail: static void udp_add_target(struct sender_command_data *scd) { - struct udp_target *ut = para_calloc(sizeof(struct udp_target)); + int ret, port = scd->port > 0 ? scd->port : conf.udp_default_port_arg; + struct udp_target *ut = para_calloc(sizeof(*ut)); strncpy(ut->host, scd->host, sizeof(ut->host)); ut->port = scd->port > 0 ? scd->port : conf.udp_default_port_arg; - ut->fd = -1; /* not yet connected */ - PARA_INFO_LOG("adding to target list (%s#%d)\n", ut->host, ut->port); - para_list_add(&ut->node, &targets); - ut->fcp.slices_per_group = scd->slices_per_group; + + ut->fcp.slices_per_group = scd->slices_per_group; ut->fcp.data_slices_per_group = scd->data_slices_per_group; - ut->fcp.max_slice_bytes = scd->max_slice_bytes; - ut->fcp.send = udp_send_fec; - ut->fcp.private_data = ut; - vss_add_fec_client(&ut->fcp, &ut->fc); + ut->fcp.max_slice_bytes = scd->max_slice_bytes; + ut->fcp.init_fec = udp_init_fec; + ut->fcp.send_fec = udp_send_fec; + + ut->sc = para_calloc(sizeof(*ut->sc)); + ut->sc->private_data = ut; + ut->sc->fd = -1; + ret = para_connect_simple(IPPROTO_UDP, scd->host, port); + if (ret < 0) + goto err; + ut->sc->fd = ret; + + ret = mcast_sender_setup(ut->sc); + if (ret < 0) + goto err; + ret = mark_fd_nonblocking(ut->sc->fd); + if (ret < 0) + goto err; + PARA_INFO_LOG("adding to target list (%s#%d)\n", ut->host, ut->port); + ut->fc = vss_add_fec_client(ut->sc, &ut->fcp); + para_list_add(&ut->sc->node, &targets); + return; +err: + if (ut->sc->fd >= 0) + close(ut->sc->fd); + PARA_NOTICE_LOG("failed to set up %s#%d (%s)- not adding it\n", + scd->host, port, para_strerror(-ret)); + free(ut->sc); + free(ut); } static int udp_com_add(struct sender_command_data *scd) @@ -292,10 +349,11 @@ static int udp_com_add(struct sender_command_data *scd) static char *udp_info(void) { - struct udp_target *ut; + struct sender_client *sc; char *ret, *tgts = NULL; - list_for_each_entry(ut, &targets, node) { + list_for_each_entry(sc, &targets, node) { + struct udp_target *ut = sc->private_data; bool is_v6 = strchr(ut->host, ':') != NULL; char *tmp = make_message("%s%s%s%s:%d/%u:%u:%u ", tgts ? : "", is_v6 ? "[" : "", ut->host, diff --git a/vss.c b/vss.c index 0eb235c7..adc0cb63 100644 --- a/vss.c +++ b/vss.c @@ -26,8 +26,8 @@ #include "net.h" #include "server.cmdline.h" #include "list.h" -#include "vss.h" #include "send.h" +#include "vss.h" #include "ipc.h" #include "fd.h" #include "sched.h" @@ -131,12 +131,20 @@ struct fec_group { uint8_t num_header_slices; }; +enum fec_client_state { + FEC_STATE_NONE = 0, /**< not initialized and not enabled */ + FEC_STATE_DISABLED, /**< temporarily disabled */ + FEC_STATE_READY_TO_RUN /**< initialized and enabled */ +}; + /** * Describes one connected FEC client. */ struct fec_client { - /** If negative, this client is temporarily disabled. */ - int error; + /** Current state of the client */ + enum fec_client_state state; + /** The connected sender client (transport layer). */ + struct sender_client *sc; /** Parameters requested by the client. */ struct fec_client_parms *fcp; /** Used by the core FEC code. */ @@ -369,7 +377,6 @@ static int setup_next_fec_group(struct fec_client *fc, struct vss_task *vsst) static int compute_next_fec_slice(struct fec_client *fc, struct vss_task *vsst) { - assert(fc->error >= 0); if (fc->first_stream_chunk < 0 || fc->current_slice_num == fc->fcp->slices_per_group + fc->num_extra_slices) { int ret = setup_next_fec_group(fc, vsst); @@ -378,8 +385,8 @@ static int compute_next_fec_slice(struct fec_client *fc, struct vss_task *vsst) if (ret < 0) { PARA_ERROR_LOG("%s\n", para_strerror(-ret)); PARA_ERROR_LOG("FEC client temporarily disabled\n"); - fc->error = ret; - return fc->error; + fc->state = FEC_STATE_DISABLED; + return ret; } } write_fec_header(fc, vsst); @@ -408,38 +415,20 @@ size_t vss_get_fec_eof_packet(const char **buf) /** * Add one entry to the list of active fec clients. * - * \param fcp Describes the fec parameters to be used for this client. - * \param result An opaque pointer that must be used by remove the client later. + * \param sc Generic sender_client data of the transport layer. + * \param fcp FEC parameters as supplied by the transport layer. * - * \return Standard. + * \return Newly allocated fec_client struct. */ -int vss_add_fec_client(struct fec_client_parms *fcp, struct fec_client **result) +struct fec_client *vss_add_fec_client(struct sender_client *sc, + struct fec_client_parms *fcp) { - int ret; - struct fec_client *fc; + struct fec_client *fc = para_calloc(sizeof(*fc)); - if (fcp->max_slice_bytes < FEC_HEADER_SIZE + fcp->data_slices_per_group) - return -ERRNO_TO_PARA_ERROR(EINVAL); - fc = para_calloc(sizeof(*fc)); + fc->sc = sc; fc->fcp = fcp; - ret = fec_new(fcp->data_slices_per_group, fcp->slices_per_group, - &fc->parms); - if (ret < 0) - goto err; - fc->first_stream_chunk = -1; /* stream not yet started */ - fc->src_data = para_malloc(fc->fcp->slices_per_group * sizeof(char *)); - fc->enc_buf = para_calloc(fc->fcp->max_slice_bytes); - fc->num_extra_slices = 0; - fc->extra_src_buf = para_calloc(fc->fcp->max_slice_bytes); - fc->next_header_time.tv_sec = 0; para_list_add(&fc->node, &fec_client_list); - *result = fc; - return 1; -err: - fec_free(fc->parms); - free(fc); - *result = NULL; - return ret; + return fc; } /** @@ -481,11 +470,10 @@ static void compute_slice_timeout(struct timeval *timeout) { struct fec_client *fc; - assert(vss_playing()); list_for_each_entry(fc, &fec_client_list, node) { struct timeval diff; - if (fc->error < 0) + if (fc->state != FEC_STATE_READY_TO_RUN) continue; if (next_slice_is_due(fc, &diff)) { timeout->tv_sec = 0; @@ -508,7 +496,7 @@ static void set_eof_barrier(struct vss_task *vsst) list_for_each_entry(fc, &fec_client_list, node) { struct timeval group_duration; - if (fc->error < 0) + if (fc->state != FEC_STATE_READY_TO_RUN) continue; tv_scale(fc->group.num_chunks, chunk_tv, &group_duration); if (tv_diff(&timeout, &group_duration, NULL) < 0) @@ -712,7 +700,7 @@ static void vss_pre_select(struct sched *s, struct task *t) senders[i].shutdown_clients(); list_for_each_entry_safe(fc, tmp, &fec_client_list, node) { fc->first_stream_chunk = -1; - fc->error = 0; + fc->state = FEC_STATE_NONE; } mmd->stream_start.tv_sec = 0; mmd->stream_start.tv_usec = 0; @@ -837,6 +825,42 @@ err: mmd->new_vss_status_flags = VSS_NEXT; } +static int initialize_fec_client(struct fec_client *fc) +{ + int ret; + struct fec_client_parms *fcp = fc->fcp; + + if (fcp->init_fec) { + /* + * Set the maximum slice size to the Maximum Packet Size if the + * transport protocol allows to determine this value. The user + * can specify a slice size up to this value. + */ + ret = fcp->init_fec(fc->sc); + if (ret < 0) + return ret; + if (!fcp->max_slice_bytes || fcp->max_slice_bytes > ret) + fcp->max_slice_bytes = ret; + } + if (fcp->max_slice_bytes < FEC_HEADER_SIZE + fcp->data_slices_per_group) + return -ERRNO_TO_PARA_ERROR(EINVAL); + ret = fec_new(fcp->data_slices_per_group, fcp->slices_per_group, + &fc->parms); + if (ret < 0) + goto err; + fc->first_stream_chunk = -1; /* stream not yet started */ + fc->src_data = para_malloc(fc->fcp->slices_per_group * sizeof(char *)); + fc->enc_buf = para_calloc(fc->fcp->max_slice_bytes); + fc->num_extra_slices = 0; + fc->extra_src_buf = para_calloc(fc->fcp->max_slice_bytes); + fc->next_header_time.tv_sec = 0; + fc->state = FEC_STATE_READY_TO_RUN; + return 1; +err: + fec_free(fc->parms); + return ret; +} + /** * Main sending function. * @@ -848,7 +872,7 @@ err: */ static void vss_send(struct vss_task *vsst) { - int i, fec_active = 0; + int ret, i, fec_active = 0; struct timeval due; struct fec_client *fc, *tmp_fc; @@ -860,8 +884,19 @@ static void vss_send(struct vss_task *vsst) &due, 1) < 0) return; list_for_each_entry_safe(fc, tmp_fc, &fec_client_list, node) { - if (fc->error < 0) + switch (fc->state) { + case FEC_STATE_DISABLED: continue; + case FEC_STATE_NONE: + ret = initialize_fec_client(fc); + if (ret < 0) { + PARA_ERROR_LOG("%s\n", para_strerror(-ret)); + continue; + } + /* fall through */ + case FEC_STATE_READY_TO_RUN: + break; + } if (!next_slice_is_due(fc, NULL)) { fec_active = 1; continue; @@ -870,9 +905,8 @@ static void vss_send(struct vss_task *vsst) continue; PARA_DEBUG_LOG("sending %d:%d (%u bytes)\n", fc->group.num, fc->current_slice_num, fc->fcp->max_slice_bytes); - fc->fcp->send((char *)fc->enc_buf, - fc->fcp->max_slice_bytes, - fc->fcp->private_data); + fc->fcp->send_fec(fc->sc, (char *)fc->enc_buf, + fc->fcp->max_slice_bytes); fc->current_slice_num++; fec_active = 1; } diff --git a/vss.h b/vss.h index ed2cd785..9412da81 100644 --- a/vss.h +++ b/vss.h @@ -24,32 +24,7 @@ const char *supported_audio_formats(void); /** Currently playing. */ #define VSS_PLAYING 8 -/** - * Each paraslash sender may register arbitrary many clients to the virtual - * streaming system, possibly with varying fec parameters. In order to do so, - * it must allocate a \a fec_client_parms structure and pass it to \ref - * add_fec_client. - * - * Clients are automatically removed from that list by the vss if an error - * occurs, or if the sender requests deletion of a client by calling \ref - * vss_del_fec_client(). - */ -struct fec_client; - -/** FEC parameters requested by FEC clients. */ -struct fec_client_parms { - /** Number of data slices plus redundant slices. */ - uint8_t slices_per_group; - /** Number of slices minus number of redundant slices. */ - uint8_t data_slices_per_group; - /** Maximal number of bytes per slice. */ - uint16_t max_slice_bytes; - /** Called by vss.c when the next slice should be sent. */ - int (*send)(char *buf, size_t num_bytes, void *private_data); - /** Passed verbatim to \a send(). */ - void *private_data; -}; - -int vss_add_fec_client(struct fec_client_parms *fcp, struct fec_client **result); +struct fec_client *vss_add_fec_client(struct sender_client *sc, + struct fec_client_parms *fcp); void vss_del_fec_client(struct fec_client *fc); size_t vss_get_fec_eof_packet(const char **buf); diff --git a/web/index.in.html b/web/index.in.html index 5a5f4aff..fa67f2e9 100644 --- a/web/index.in.html +++ b/web/index.in.html @@ -1,6 +1,7 @@

Events

    +
  • 2010-06-23: What's cooking page online
  • 2010-04-23: paraslash-0.4.2 (sig) "associative expansion" diff --git a/wmadec_filter.c b/wmadec_filter.c index 96fb5008..45fda79e 100644 --- a/wmadec_filter.c +++ b/wmadec_filter.c @@ -1279,7 +1279,7 @@ next_buffer: } success: btr_consume(btrn, converted); - goto next_buffer; + return; err: assert(ret < 0); t->error = ret;