X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/58cadc8608fbb2cbc7b74578cd92de4337a4b887..4a19c3b254b01c298fe25d72f450a760278e9386:/io.c diff --git a/io.c b/io.c index 9fc37152..02511764 100644 --- a/io.c +++ b/io.c @@ -1,150 +1,407 @@ -/* -*- c-file-style: "linux" -*- - * - * Copyright (C) 1996-2001 by Andrew Tridgell - * Copyright (C) Paul Mackerras 1996 - * Copyright (C) 2001, 2002 by Martin Pool - * +/* + * Socket and pipe I/O utilities used in rsync. + * + * Copyright (C) 1996-2001 Andrew Tridgell + * Copyright (C) 1996 Paul Mackerras + * Copyright (C) 2001, 2002 Martin Pool + * Copyright (C) 2003, 2004, 2005, 2006 Wayne Davison + * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. - * + * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * + * You should have received a copy of the GNU General Public License along + * with this program; if not, write to the Free Software Foundation, Inc., + * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA. */ -/** - * @file io.c - * - * Socket and pipe IO utilities used in rsync. - * - * rsync provides its own multiplexing system, which is used to send - * stderr and stdout over a single socket. We need this because - * stdout normally carries the binary data stream, and stderr all our - * error messages. +/* Rsync provides its own multiplexing system, which is used to send + * stderr and stdout over a single socket. * * For historical reasons this is off during the start of the * connection, but it's switched on quite early using - * io_start_multiplex_out() and io_start_multiplex_in(). - **/ + * io_start_multiplex_out() and io_start_multiplex_in(). */ #include "rsync.h" /** If no timeout is specified then use a 60 second select timeout */ #define SELECT_TIMEOUT 60 -static int io_multiplexing_out; -static int io_multiplexing_in; -static int multiplex_in_fd; -static int multiplex_out_fd; -static time_t last_io; -static int no_flush; - extern int bwlimit; -extern int verbose; +extern size_t bwlimit_writemax; extern int io_timeout; +extern int allowed_lull; +extern int am_server; +extern int am_daemon; +extern int am_sender; +extern int am_generator; +extern int eol_nulls; +extern int read_batch; +extern int csum_length; +extern int checksum_seed; +extern int protocol_version; +extern int remove_source_files; +extern int preserve_hard_links; +extern char *filesfrom_host; extern struct stats stats; - +extern struct file_list *the_file_list; const char phase_unknown[] = "unknown"; +int ignore_timeout = 0; +int batch_fd = -1; +int batch_gen_fd = -1; -/** - * The connection might be dropped at some point; perhaps because the - * remote instance crashed. Just giving the offset on the stream is - * not very helpful. So instead we try to make io_phase_name point to - * something useful. - * - * For buffered/multiplexed IO these names will be somewhat - * approximate; perhaps for ease of support we would rather make the - * buffer always flush when a single application-level IO finishes. - * - * @todo Perhaps we want some simple stack functionality, but there's - * no need to overdo it. - **/ -const char *io_write_phase = phase_unknown; -const char *io_read_phase = phase_unknown; +/* Ignore an EOF error if non-zero. See whine_about_eof(). */ +int kluge_around_eof = 0; -/** Ignore EOF errors while reading a module listing if the remote - version is 24 or less. */ -int kludge_around_eof = False; +int msg_fd_in = -1; +int msg_fd_out = -1; +int sock_f_in = -1; +int sock_f_out = -1; +static int io_multiplexing_out; +static int io_multiplexing_in; +static time_t last_io_in; +static time_t last_io_out; +static int no_flush; + +static int write_batch_monitor_in = -1; +static int write_batch_monitor_out = -1; -static int io_error_fd = -1; static int io_filesfrom_f_in = -1; static int io_filesfrom_f_out = -1; static char io_filesfrom_buf[2048]; static char *io_filesfrom_bp; static char io_filesfrom_lastchar; static int io_filesfrom_buflen; +static int defer_forwarding_messages = 0; +static int select_timeout = SELECT_TIMEOUT; +static int active_filecnt = 0; +static OFF_T active_bytecnt = 0; static void read_loop(int fd, char *buf, size_t len); +struct flist_ndx_item { + struct flist_ndx_item *next; + int ndx; +}; + +struct flist_ndx_list { + struct flist_ndx_item *head, *tail; +}; + +static struct flist_ndx_list redo_list, hlink_list; + +struct msg_list_item { + struct msg_list_item *next; + int len; + char buf[1]; +}; + +struct msg_list { + struct msg_list_item *head, *tail; +}; + +static struct msg_list msg2genr, msg2sndr; + +static void flist_ndx_push(struct flist_ndx_list *lp, int ndx) +{ + struct flist_ndx_item *item; + + if (!(item = new(struct flist_ndx_item))) + out_of_memory("flist_ndx_push"); + item->next = NULL; + item->ndx = ndx; + if (lp->tail) + lp->tail->next = item; + else + lp->head = item; + lp->tail = item; +} + +static int flist_ndx_pop(struct flist_ndx_list *lp) +{ + struct flist_ndx_item *next; + int ndx; + + if (!lp->head) + return -1; + + ndx = lp->head->ndx; + next = lp->head->next; + free(lp->head); + lp->head = next; + if (!next) + lp->tail = NULL; + + return ndx; +} + static void check_timeout(void) { - extern int am_server, am_daemon; time_t t; - err_list_push(); - - if (!io_timeout) return; + if (!io_timeout || ignore_timeout) + return; - if (!last_io) { - last_io = time(NULL); + if (!last_io_in) { + last_io_in = time(NULL); return; } t = time(NULL); - if (last_io && io_timeout && (t-last_io) >= io_timeout) { + if (t - last_io_in >= io_timeout) { if (!am_server && !am_daemon) { - rprintf(FERROR,"io timeout after %d seconds - exiting\n", - (int)(t-last_io)); + rprintf(FERROR, "io timeout after %d seconds -- exiting\n", + (int)(t-last_io_in)); } exit_cleanup(RERR_TIMEOUT); } } -/** Setup the fd used to propagate errors */ -void io_set_error_fd(int fd) +/* Note the fds used for the main socket (which might really be a pipe + * for a local transfer, but we can ignore that). */ +void io_set_sock_fds(int f_in, int f_out) +{ + sock_f_in = f_in; + sock_f_out = f_out; +} + +void set_io_timeout(int secs) +{ + io_timeout = secs; + + if (!io_timeout || io_timeout > SELECT_TIMEOUT) + select_timeout = SELECT_TIMEOUT; + else + select_timeout = io_timeout; + + allowed_lull = read_batch ? 0 : (io_timeout + 1) / 2; +} + +/* Setup the fd used to receive MSG_* messages. Only needed during the + * early stages of being a local sender (up through the sending of the + * file list) or when we're the generator (to fetch the messages from + * the receiver). */ +void set_msg_fd_in(int fd) +{ + msg_fd_in = fd; +} + +/* Setup the fd used to send our MSG_* messages. Only needed when + * we're the receiver (to send our messages to the generator). */ +void set_msg_fd_out(int fd) { - io_error_fd = fd; + msg_fd_out = fd; + set_nonblocking(msg_fd_out); } -/** Read some data from the error fd and write it to the write log code */ -static void read_error_fd(void) +/* Add a message to the pending MSG_* list. */ +static void msg_list_add(struct msg_list *lst, int code, const char *buf, int len) { - char buf[200]; + struct msg_list_item *m; + int sz = len + 4 + sizeof m[0] - 1; + + if (!(m = (struct msg_list_item *)new_array(char, sz))) + out_of_memory("msg_list_add"); + m->next = NULL; + m->len = len + 4; + SIVAL(m->buf, 0, ((code+MPLEX_BASE)<<24) | len); + memcpy(m->buf + 4, buf, len); + if (lst->tail) + lst->tail->next = m; + else + lst->head = m; + lst->tail = m; +} + +/* Read a message from the MSG_* fd and handle it. This is called either + * during the early stages of being a local sender (up through the sending + * of the file list) or when we're the generator (to fetch the messages + * from the receiver). */ +static void read_msg_fd(void) +{ + char buf[2048]; size_t n; - int fd = io_error_fd; + int fd = msg_fd_in; int tag, len; - /* io_error_fd is temporarily disabled -- is this meant to - * prevent indefinite recursion? */ - io_error_fd = -1; + /* Temporarily disable msg_fd_in. This is needed to avoid looping back + * to this routine from writefd_unbuffered(). */ + msg_fd_in = -1; read_loop(fd, buf, 4); tag = IVAL(buf, 0); len = tag & 0xFFFFFF; - tag = tag >> 24; - tag -= MPLEX_BASE; + tag = (tag >> 24) - MPLEX_BASE; - while (len) { - n = len; - if (n > (sizeof buf - 1)) - n = sizeof buf - 1; - read_loop(fd, buf, n); - rwrite((enum logcode)tag, buf, n); - len -= n; + switch (tag) { + case MSG_DONE: + if (len != 0 || !am_generator) { + rprintf(FERROR, "invalid message %d:%d\n", tag, len); + exit_cleanup(RERR_STREAMIO); + } + flist_ndx_push(&redo_list, -1); + break; + case MSG_REDO: + if (len != 4 || !am_generator) { + rprintf(FERROR, "invalid message %d:%d\n", tag, len); + exit_cleanup(RERR_STREAMIO); + } + read_loop(fd, buf, 4); + if (remove_source_files) + decrement_active_files(IVAL(buf,0)); + flist_ndx_push(&redo_list, IVAL(buf,0)); + break; + case MSG_DELETED: + if (len >= (int)sizeof buf || !am_generator) { + rprintf(FERROR, "invalid message %d:%d\n", tag, len); + exit_cleanup(RERR_STREAMIO); + } + read_loop(fd, buf, len); + send_msg(MSG_DELETED, buf, len); + break; + case MSG_SUCCESS: + if (len != 4 || !am_generator) { + rprintf(FERROR, "invalid message %d:%d\n", tag, len); + exit_cleanup(RERR_STREAMIO); + } + read_loop(fd, buf, len); + if (remove_source_files) { + decrement_active_files(IVAL(buf,0)); + send_msg(MSG_SUCCESS, buf, len); + } + if (preserve_hard_links) + flist_ndx_push(&hlink_list, IVAL(buf,0)); + break; + case MSG_SOCKERR: + if (!am_generator) { + rprintf(FERROR, "invalid message %d:%d\n", tag, len); + exit_cleanup(RERR_STREAMIO); + } + close_multiplexing_out(); + /* FALL THROUGH */ + case MSG_INFO: + case MSG_ERROR: + case MSG_LOG: + while (len) { + n = len; + if (n >= sizeof buf) + n = sizeof buf - 1; + read_loop(fd, buf, n); + rwrite(tag, buf, n); + len -= n; + } + break; + default: + rprintf(FERROR, "unknown message %d:%d [%s]\n", + tag, len, who_am_i()); + exit_cleanup(RERR_STREAMIO); } - io_error_fd = fd; + msg_fd_in = fd; +} + +/* This is used by the generator to limit how many file transfers can + * be active at once when --remove-source-files is specified. Without + * this, sender-side deletions were mostly happening at the end. */ +void increment_active_files(int ndx, int itemizing, enum logcode code) +{ + /* TODO: tune these limits? */ + while (active_filecnt >= (active_bytecnt >= 128*1024 ? 10 : 50)) { + if (hlink_list.head) + check_for_finished_hlinks(itemizing, code); + read_msg_fd(); + } + + active_filecnt++; + active_bytecnt += the_file_list->files[ndx]->length; +} + +void decrement_active_files(int ndx) +{ + active_filecnt--; + active_bytecnt -= the_file_list->files[ndx]->length; +} + +/* Try to push messages off the list onto the wire. If we leave with more + * to do, return 0. On error, return -1. If everything flushed, return 1. + * This is only active in the receiver. */ +static int msg2genr_flush(int flush_it_all) +{ + static int written = 0; + struct timeval tv; + fd_set fds; + + if (msg_fd_out < 0) + return -1; + + while (msg2genr.head) { + struct msg_list_item *m = msg2genr.head; + int n = write(msg_fd_out, m->buf + written, m->len - written); + if (n < 0) { + if (errno == EINTR) + continue; + if (errno != EWOULDBLOCK && errno != EAGAIN) + return -1; + if (!flush_it_all) + return 0; + FD_ZERO(&fds); + FD_SET(msg_fd_out, &fds); + tv.tv_sec = select_timeout; + tv.tv_usec = 0; + if (!select(msg_fd_out+1, NULL, &fds, NULL, &tv)) + check_timeout(); + } else if ((written += n) == m->len) { + msg2genr.head = m->next; + if (!msg2genr.head) + msg2genr.tail = NULL; + free(m); + written = 0; + } + } + return 1; +} + +int send_msg(enum msgcode code, const char *buf, int len) +{ + if (msg_fd_out < 0) { + if (!defer_forwarding_messages) + return io_multiplex_write(code, buf, len); + if (!io_multiplexing_out) + return 0; + msg_list_add(&msg2sndr, code, buf, len); + return 1; + } + msg_list_add(&msg2genr, code, buf, len); + msg2genr_flush(NORMAL_FLUSH); + return 1; +} + +int get_redo_num(int itemizing, enum logcode code) +{ + while (1) { + if (hlink_list.head) + check_for_finished_hlinks(itemizing, code); + if (redo_list.head) + break; + read_msg_fd(); + } + + return flist_ndx_pop(&redo_list); +} + +int get_hlink_num(void) +{ + return flist_ndx_pop(&hlink_list); } /** @@ -166,44 +423,38 @@ void io_set_filesfrom_fds(int f_in, int f_out) io_filesfrom_buflen = 0; } -/** - * It's almost always an error to get an EOF when we're trying to read - * from the network, because the protocol is self-terminating. +/* It's almost always an error to get an EOF when we're trying to read from the + * network, because the protocol is (for the most part) self-terminating. * - * However, there is one unfortunate cases where it is not, which is - * rsync <2.4.6 sending a list of modules on a server, since the list - * is terminated by closing the socket. So, for the section of the - * program where that is a problem (start_socket_client), - * kludge_around_eof is True and we just exit. - */ -static void whine_about_eof(void) + * There is one case for the receiver when it is at the end of the transfer + * (hanging around reading any keep-alive packets that might come its way): if + * the sender dies before the generator's kill-signal comes through, we can end + * up here needing to loop until the kill-signal arrives. In this situation, + * kluge_around_eof will be < 0. + * + * There is another case for older protocol versions (< 24) where the module + * listing was not terminated, so we must ignore an EOF error in that case and + * exit. In this situation, kluge_around_eof will be > 0. */ +static void whine_about_eof(int fd) { - if (kludge_around_eof) - exit_cleanup(0); - else { - rprintf(FERROR, - "%s: connection unexpectedly closed " - "(%.0f bytes read so far)\n", - RSYNC_NAME, (double)stats.total_read); - - exit_cleanup(RERR_STREAMIO); + if (kluge_around_eof && fd == sock_f_in) { + int i; + if (kluge_around_eof > 0) + exit_cleanup(0); + /* If we're still here after 10 seconds, exit with an error. */ + for (i = 10*1000/20; i--; ) + msleep(20); } -} + rprintf(FERROR, RSYNC_NAME ": connection unexpectedly closed " + "(%.0f bytes received so far) [%s]\n", + (double)stats.total_read, who_am_i()); -static void die_from_readerr(int err) -{ - /* this prevents us trying to write errors on a dead socket */ - io_multiplexing_close(); - - rprintf(FERROR, "%s: read error: %s\n", - RSYNC_NAME, strerror(err)); exit_cleanup(RERR_STREAMIO); } - /** - * Read from a socket with IO timeout. return the number of bytes + * Read from a socket with I/O timeout. return the number of bytes * read. If no bytes can be read then exit, never return a number <= 0. * * TODO: If the remote shell connection fails, then current versions @@ -215,27 +466,29 @@ static void die_from_readerr(int err) */ static int read_timeout(int fd, char *buf, size_t len) { - int n, ret=0; + int n, cnt = 0; - io_flush(); + io_flush(NORMAL_FLUSH); - while (ret == 0) { + while (cnt == 0) { /* until we manage to read *something* */ fd_set r_fds, w_fds; struct timeval tv; - int fd_count = fd+1; + int maxfd = fd; int count; FD_ZERO(&r_fds); + FD_ZERO(&w_fds); FD_SET(fd, &r_fds); - if (io_error_fd != -1) { - FD_SET(io_error_fd, &r_fds); - if (io_error_fd >= fd_count) fd_count = io_error_fd+1; + if (msg2genr.head) { + FD_SET(msg_fd_out, &w_fds); + if (msg_fd_out > maxfd) + maxfd = msg_fd_out; } - if (io_filesfrom_f_out != -1) { + if (io_filesfrom_f_out >= 0) { int new_fd; if (io_filesfrom_buflen == 0) { - if (io_filesfrom_f_in != -1) { + if (io_filesfrom_f_in >= 0) { FD_SET(io_filesfrom_f_in, &r_fds); new_fd = io_filesfrom_f_in; } else { @@ -243,39 +496,31 @@ static int read_timeout(int fd, char *buf, size_t len) new_fd = -1; } } else { - FD_ZERO(&w_fds); FD_SET(io_filesfrom_f_out, &w_fds); new_fd = io_filesfrom_f_out; } - if (new_fd >= fd_count) fd_count = new_fd+1; + if (new_fd > maxfd) + maxfd = new_fd; } - tv.tv_sec = io_timeout?io_timeout:SELECT_TIMEOUT; + tv.tv_sec = select_timeout; tv.tv_usec = 0; errno = 0; - count = select(fd_count, &r_fds, - io_filesfrom_buflen? &w_fds : NULL, - NULL, &tv); - - if (count == 0) { - check_timeout(); - } + count = select(maxfd + 1, &r_fds, &w_fds, NULL, &tv); if (count <= 0) { - if (errno == EBADF) { + if (errno == EBADF) exit_cleanup(RERR_SOCKETIO); - } + check_timeout(); continue; } + if (msg2genr.head && FD_ISSET(msg_fd_out, &w_fds)) + msg2genr_flush(NORMAL_FLUSH); - if (io_error_fd != -1 && FD_ISSET(io_error_fd, &r_fds)) { - read_error_fd(); - } - - if (io_filesfrom_f_out != -1) { + if (io_filesfrom_f_out >= 0) { if (io_filesfrom_buflen) { if (FD_ISSET(io_filesfrom_f_out, &w_fds)) { int l = write(io_filesfrom_f_out, @@ -291,7 +536,7 @@ static int read_timeout(int fd, char *buf, size_t len) io_filesfrom_f_out = -1; } } - } else if (io_filesfrom_f_in != -1) { + } else if (io_filesfrom_f_in >= 0) { if (FD_ISSET(io_filesfrom_f_in, &r_fds)) { int l = read(io_filesfrom_f_in, io_filesfrom_buf, @@ -303,7 +548,6 @@ static int read_timeout(int fd, char *buf, size_t len) io_filesfrom_buflen = io_filesfrom_lastchar? 2 : 1; io_filesfrom_f_in = -1; } else { - extern int eol_nulls; if (!eol_nulls) { char *s = io_filesfrom_buf + l; /* Transform CR and/or LF into '\0' */ @@ -339,29 +583,36 @@ static int read_timeout(int fd, char *buf, size_t len) } } - if (!FD_ISSET(fd, &r_fds)) continue; + if (!FD_ISSET(fd, &r_fds)) + continue; n = read(fd, buf, len); - if (n > 0) { - buf += n; - len -= n; - ret += n; - if (io_timeout) - last_io = time(NULL); - continue; - } else if (n == 0) { - whine_about_eof(); - return -1; /* doesn't return */ - } else if (n == -1) { - if (errno == EINTR || errno == EWOULDBLOCK || - errno == EAGAIN) + if (n <= 0) { + if (n == 0) + whine_about_eof(fd); /* Doesn't return. */ + if (errno == EINTR || errno == EWOULDBLOCK + || errno == EAGAIN) continue; - die_from_readerr(errno); + + /* Don't write errors on a dead socket. */ + if (fd == sock_f_in) { + close_multiplexing_out(); + rsyserr(FSOCKERR, errno, "read error"); + } else + rsyserr(FERROR, errno, "read error"); + exit_cleanup(RERR_STREAMIO); } + + buf += n; + len -= n; + cnt += n; + + if (fd == sock_f_in && io_timeout) + last_io_in = time(NULL); } - return ret; + return cnt; } /** @@ -372,11 +623,7 @@ int read_filesfrom_line(int fd, char *fname) { char ch, *s, *eob = fname + MAXPATHLEN - 1; int cnt; - extern int io_timeout; - extern int eol_nulls; - extern char *remote_filesfrom_file; - extern int am_server; - int reading_remotely = remote_filesfrom_file || (am_server && fd == 0); + int reading_remotely = filesfrom_host != NULL; int nulls = eol_nulls || reading_remotely; start: @@ -386,13 +633,19 @@ int read_filesfrom_line(int fd, char *fname) if (cnt < 0 && (errno == EWOULDBLOCK || errno == EINTR || errno == EAGAIN)) { struct timeval tv; - fd_set fds; - FD_ZERO(&fds); - FD_SET(fd, &fds); - tv.tv_sec = io_timeout? io_timeout : SELECT_TIMEOUT; + fd_set r_fds, e_fds; + FD_ZERO(&r_fds); + FD_SET(fd, &r_fds); + FD_ZERO(&e_fds); + FD_SET(fd, &e_fds); + tv.tv_sec = select_timeout; tv.tv_usec = 0; - if (!select(fd+1, &fds, NULL, NULL, &tv)) + if (!select(fd+1, &r_fds, NULL, &e_fds, &tv)) check_timeout(); + if (FD_ISSET(fd, &e_fds)) { + rsyserr(FINFO, errno, + "select exception on fd %d", fd); + } continue; } if (cnt != 1) @@ -415,6 +668,58 @@ int read_filesfrom_line(int fd, char *fname) return s - fname; } +static char *iobuf_out; +static int iobuf_out_cnt; + +void io_start_buffering_out(void) +{ + if (iobuf_out) + return; + if (!(iobuf_out = new_array(char, IO_BUFFER_SIZE))) + out_of_memory("io_start_buffering_out"); + iobuf_out_cnt = 0; +} + +static char *iobuf_in; +static size_t iobuf_in_siz; + +void io_start_buffering_in(void) +{ + if (iobuf_in) + return; + iobuf_in_siz = 2 * IO_BUFFER_SIZE; + if (!(iobuf_in = new_array(char, iobuf_in_siz))) + out_of_memory("io_start_buffering_in"); +} + +void io_end_buffering(void) +{ + io_flush(NORMAL_FLUSH); + if (!io_multiplexing_out) { + free(iobuf_out); + iobuf_out = NULL; + } +} + +void maybe_flush_socket(void) +{ + if (iobuf_out && iobuf_out_cnt && time(NULL) - last_io_out >= 5) + io_flush(NORMAL_FLUSH); +} + +void maybe_send_keepalive(void) +{ + if (time(NULL) - last_io_out >= allowed_lull) { + if (!iobuf_out || !iobuf_out_cnt) { + if (protocol_version < 29) + return; /* there's nothing we can do */ + write_int(sock_f_out, the_file_list->count); + write_shortint(sock_f_out, ITEM_IS_NEW); + } + if (iobuf_out) + io_flush(NORMAL_FLUSH); + } +} /** * Continue trying to read len bytes - don't return until len has been @@ -430,64 +735,101 @@ static void read_loop(int fd, char *buf, size_t len) } } - /** * Read from the file descriptor handling multiplexing - return number * of bytes read. - * - * Never returns <= 0. + * + * Never returns <= 0. */ -static int read_unbuffered(int fd, char *buf, size_t len) +static int readfd_unbuffered(int fd, char *buf, size_t len) { static size_t remaining; - int tag, ret = 0; - char line[1024]; + static size_t iobuf_in_ndx; + size_t msg_bytes; + int tag, cnt = 0; + char line[BIGPATHBUFLEN]; - if (!io_multiplexing_in || fd != multiplex_in_fd) + if (!iobuf_in || fd != sock_f_in) return read_timeout(fd, buf, len); - while (ret == 0) { + if (!io_multiplexing_in && remaining == 0) { + remaining = read_timeout(fd, iobuf_in, iobuf_in_siz); + iobuf_in_ndx = 0; + } + + while (cnt == 0) { if (remaining) { len = MIN(len, remaining); - read_loop(fd, buf, len); + memcpy(buf, iobuf_in + iobuf_in_ndx, len); + iobuf_in_ndx += len; remaining -= len; - ret = len; - continue; + cnt = len; + break; } read_loop(fd, line, 4); tag = IVAL(line, 0); - remaining = tag & 0xFFFFFF; - tag = tag >> 24; - - if (tag == MPLEX_BASE) - continue; - - tag -= MPLEX_BASE; - - if (tag != FERROR && tag != FINFO) { - rprintf(FERROR, "unexpected tag %d\n", tag); - exit_cleanup(RERR_STREAMIO); - } + msg_bytes = tag & 0xFFFFFF; + tag = (tag >> 24) - MPLEX_BASE; - if (remaining > sizeof line - 1) { - rprintf(FERROR, "multiplexing overflow %ld\n\n", - (long)remaining); + switch (tag) { + case MSG_DATA: + if (msg_bytes > iobuf_in_siz) { + if (!(iobuf_in = realloc_array(iobuf_in, char, + msg_bytes))) + out_of_memory("readfd_unbuffered"); + iobuf_in_siz = msg_bytes; + } + read_loop(fd, iobuf_in, msg_bytes); + remaining = msg_bytes; + iobuf_in_ndx = 0; + break; + case MSG_DELETED: + if (msg_bytes >= sizeof line) + goto overflow; + read_loop(fd, line, msg_bytes); + /* A directory name was sent with the trailing null */ + if (msg_bytes > 0 && !line[msg_bytes-1]) + log_delete(line, S_IFDIR); + else { + line[msg_bytes] = '\0'; + log_delete(line, S_IFREG); + } + break; + case MSG_SUCCESS: + if (msg_bytes != 4) { + rprintf(FERROR, "invalid multi-message %d:%ld [%s]\n", + tag, (long)msg_bytes, who_am_i()); + exit_cleanup(RERR_STREAMIO); + } + read_loop(fd, line, msg_bytes); + successful_send(IVAL(line, 0)); + break; + case MSG_INFO: + case MSG_ERROR: + if (msg_bytes >= sizeof line) { + overflow: + rprintf(FERROR, + "multiplexing overflow %d:%ld [%s]\n", + tag, (long)msg_bytes, who_am_i()); + exit_cleanup(RERR_STREAMIO); + } + read_loop(fd, line, msg_bytes); + rwrite((enum logcode)tag, line, msg_bytes); + break; + default: + rprintf(FERROR, "unexpected tag %d [%s]\n", + tag, who_am_i()); exit_cleanup(RERR_STREAMIO); } - - read_loop(fd, line, remaining); - line[remaining] = 0; - - rprintf((enum logcode) tag, "%s", line); - remaining = 0; } - return ret; -} - + if (remaining == 0) + io_flush(NORMAL_FLUSH); + return cnt; +} /** * Do a buffered read from @p fd. Don't return until all @p n bytes @@ -496,70 +838,145 @@ static int read_unbuffered(int fd, char *buf, size_t len) **/ static void readfd(int fd, char *buffer, size_t N) { - int ret; - size_t total=0; + int cnt; + size_t total = 0; while (total < N) { - io_flush(); + cnt = readfd_unbuffered(fd, buffer + total, N-total); + total += cnt; + } - ret = read_unbuffered(fd, buffer + total, N-total); - total += ret; + if (fd == write_batch_monitor_in) { + if ((size_t)write(batch_fd, buffer, total) != total) + exit_cleanup(RERR_FILEIO); } - stats.total_read += total; + if (fd == sock_f_in) + stats.total_read += total; } +int read_shortint(int f) +{ + uchar b[2]; + readfd(f, (char *)b, 2); + return (b[1] << 8) + b[0]; +} int32 read_int(int f) { char b[4]; - int32 ret; + int32 num; readfd(f,b,4); - ret = IVAL(b,0); - if (ret == (int32)0xffffffff) return -1; - return ret; + num = IVAL(b,0); + if (num == (int32)0xffffffff) + return -1; + return num; } int64 read_longint(int f) { - int64 ret; + int64 num; char b[8]; - ret = read_int(f); + num = read_int(f); - if ((int32)ret != (int32)0xffffffff) { - return ret; - } + if ((int32)num != (int32)0xffffffff) + return num; -#ifdef NO_INT64 - rprintf(FERROR,"Integer overflow - attempted 64 bit offset\n"); +#if SIZEOF_INT64 < 8 + rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n"); exit_cleanup(RERR_UNSUPPORTED); #else readfd(f,b,8); - ret = IVAL(b,0) | (((int64)IVAL(b,4))<<32); + num = IVAL(b,0) | (((int64)IVAL(b,4))<<32); #endif - return ret; + return num; } -void read_buf(int f,char *buf,size_t len) +void read_buf(int f, char *buf, size_t len) { readfd(f,buf,len); } -void read_sbuf(int f,char *buf,size_t len) +void read_sbuf(int f, char *buf, size_t len) { - read_buf(f,buf,len); - buf[len] = 0; + readfd(f, buf, len); + buf[len] = '\0'; } -unsigned char read_byte(int f) +uchar read_byte(int f) { - unsigned char c; - read_buf(f, (char *)&c, 1); + uchar c; + readfd(f, (char *)&c, 1); return c; } +int read_vstring(int f, char *buf, int bufsize) +{ + int len = read_byte(f); + + if (len & 0x80) + len = (len & ~0x80) * 0x100 + read_byte(f); + + if (len >= bufsize) { + rprintf(FERROR, "over-long vstring received (%d > %d)\n", + len, bufsize - 1); + return -1; + } + + if (len) + readfd(f, buf, len); + buf[len] = '\0'; + return len; +} + +/* Populate a sum_struct with values from the socket. This is + * called by both the sender and the receiver. */ +void read_sum_head(int f, struct sum_struct *sum) +{ + sum->count = read_int(f); + if (sum->count < 0) { + rprintf(FERROR, "Invalid checksum count %ld [%s]\n", + (long)sum->count, who_am_i()); + exit_cleanup(RERR_PROTOCOL); + } + sum->blength = read_int(f); + if (sum->blength < 0 || sum->blength > MAX_BLOCK_SIZE) { + rprintf(FERROR, "Invalid block length %ld [%s]\n", + (long)sum->blength, who_am_i()); + exit_cleanup(RERR_PROTOCOL); + } + sum->s2length = protocol_version < 27 ? csum_length : (int)read_int(f); + if (sum->s2length < 0 || sum->s2length > MD4_SUM_LENGTH) { + rprintf(FERROR, "Invalid checksum length %d [%s]\n", + sum->s2length, who_am_i()); + exit_cleanup(RERR_PROTOCOL); + } + sum->remainder = read_int(f); + if (sum->remainder < 0 || sum->remainder > sum->blength) { + rprintf(FERROR, "Invalid remainder length %ld [%s]\n", + (long)sum->remainder, who_am_i()); + exit_cleanup(RERR_PROTOCOL); + } +} + +/* Send the values from a sum_struct over the socket. Set sum to + * NULL if there are no checksums to send. This is called by both + * the generator and the sender. */ +void write_sum_head(int f, struct sum_struct *sum) +{ + static struct sum_struct null_sum; + + if (sum == NULL) + sum = &null_sum; + + write_int(f, sum->count); + write_int(f, sum->blength); + if (protocol_version >= 27) + write_int(f, sum->s2length); + write_int(f, sum->remainder); +} /** * Sleep after writing to limit I/O bandwidth usage. @@ -569,207 +986,258 @@ unsigned char read_byte(int f) * use a bit less bandwidth than specified, because it doesn't make up * for slow periods. But arguably this is a feature. In addition, we * ought to take the time used to write the data into account. + * + * During some phases of big transfers (file FOO is uptodate) this is + * called with a small bytes_written every time. As the kernel has to + * round small waits up to guarantee that we actually wait at least the + * requested number of microseconds, this can become grossly inaccurate. + * We therefore keep track of the bytes we've written over time and only + * sleep when the accumulated delay is at least 1 tenth of a second. **/ static void sleep_for_bwlimit(int bytes_written) { - struct timeval tv; + static struct timeval prior_tv; + static long total_written = 0; + struct timeval tv, start_tv; + long elapsed_usec, sleep_usec; - if (!bwlimit) +#define ONE_SEC 1000000L /* # of microseconds in a second */ + + if (!bwlimit_writemax) return; - assert(bytes_written > 0); - assert(bwlimit > 0); + total_written += bytes_written; + + gettimeofday(&start_tv, NULL); + if (prior_tv.tv_sec) { + elapsed_usec = (start_tv.tv_sec - prior_tv.tv_sec) * ONE_SEC + + (start_tv.tv_usec - prior_tv.tv_usec); + total_written -= elapsed_usec * bwlimit / (ONE_SEC/1024); + if (total_written < 0) + total_written = 0; + } - tv.tv_usec = bytes_written * 1000 / bwlimit; - tv.tv_sec = tv.tv_usec / 1000000; - tv.tv_usec = tv.tv_usec % 1000000; + sleep_usec = total_written * (ONE_SEC/1024) / bwlimit; + if (sleep_usec < ONE_SEC / 10) { + prior_tv = start_tv; + return; + } + tv.tv_sec = sleep_usec / ONE_SEC; + tv.tv_usec = sleep_usec % ONE_SEC; select(0, NULL, NULL, NULL, &tv); -} + gettimeofday(&prior_tv, NULL); + elapsed_usec = (prior_tv.tv_sec - start_tv.tv_sec) * ONE_SEC + + (prior_tv.tv_usec - start_tv.tv_usec); + total_written = (sleep_usec - elapsed_usec) * bwlimit / (ONE_SEC/1024); +} -/** - * Write len bytes to the file descriptor @p fd. +/* Write len bytes to the file descriptor fd, looping as necessary to get + * the job done and also (in certain circumstances) reading any data on + * msg_fd_in to avoid deadlock. * * This function underlies the multiplexing system. The body of the - * application never calls this function directly. - **/ -static void writefd_unbuffered(int fd,char *buf,size_t len) + * application never calls this function directly. */ +static void writefd_unbuffered(int fd, const char *buf, size_t len) { - size_t total = 0; - fd_set w_fds, r_fds; - int fd_count, count; + size_t n, total = 0; + fd_set w_fds, r_fds, e_fds; + int maxfd, count, cnt, using_r_fds; + int defer_save = defer_forwarding_messages; struct timeval tv; - err_list_push(); - no_flush++; while (total < len) { FD_ZERO(&w_fds); - FD_SET(fd,&w_fds); - fd_count = fd; + FD_SET(fd, &w_fds); + FD_ZERO(&e_fds); + FD_SET(fd, &e_fds); + maxfd = fd; - if (io_error_fd != -1) { + if (msg_fd_in >= 0) { FD_ZERO(&r_fds); - FD_SET(io_error_fd,&r_fds); - if (io_error_fd > fd_count) - fd_count = io_error_fd; - } - - tv.tv_sec = io_timeout?io_timeout:SELECT_TIMEOUT; + FD_SET(msg_fd_in, &r_fds); + if (msg_fd_in > maxfd) + maxfd = msg_fd_in; + using_r_fds = 1; + } else + using_r_fds = 0; + + tv.tv_sec = select_timeout; tv.tv_usec = 0; errno = 0; - - count = select(fd_count+1, - io_error_fd != -1?&r_fds:NULL, - &w_fds,NULL, - &tv); - - if (count == 0) { - check_timeout(); - } + count = select(maxfd + 1, using_r_fds ? &r_fds : NULL, + &w_fds, &e_fds, &tv); if (count <= 0) { - if (errno == EBADF) { + if (count < 0 && errno == EBADF) exit_cleanup(RERR_SOCKETIO); - } + check_timeout(); continue; } - if (io_error_fd != -1 && FD_ISSET(io_error_fd, &r_fds)) { - read_error_fd(); + if (FD_ISSET(fd, &e_fds)) { + rsyserr(FINFO, errno, + "select exception on fd %d", fd); } - if (FD_ISSET(fd, &w_fds)) { - int ret; - size_t n = len-total; - ret = write(fd,buf+total,n); + if (using_r_fds && FD_ISSET(msg_fd_in, &r_fds)) + read_msg_fd(); - if (ret == -1 && errno == EINTR) { - continue; - } + if (!FD_ISSET(fd, &w_fds)) + continue; - if (ret == -1 && - (errno == EWOULDBLOCK || errno == EAGAIN)) { - msleep(1); - continue; + n = len - total; + if (bwlimit_writemax && n > bwlimit_writemax) + n = bwlimit_writemax; + cnt = write(fd, buf + total, n); + + if (cnt <= 0) { + if (cnt < 0) { + if (errno == EINTR) + continue; + if (errno == EWOULDBLOCK || errno == EAGAIN) { + msleep(1); + continue; + } } - if (ret <= 0) { - /* Don't try to write errors back - * across the stream */ - io_multiplexing_close(); - rprintf(FERROR, RSYNC_NAME - ": writefd_unbuffered failed to write %ld bytes: phase \"%s\": %s\n", - (long) len, io_write_phase, - strerror(errno)); - exit_cleanup(RERR_STREAMIO); + /* Don't try to write errors back across the stream. */ + if (fd == sock_f_out) + close_multiplexing_out(); + rsyserr(FERROR, errno, + "writefd_unbuffered failed to write %ld bytes [%s]", + (long)len, who_am_i()); + /* If the other side is sending us error messages, try + * to grab any messages they sent before they died. */ + while (fd == sock_f_out && io_multiplexing_in) { + set_io_timeout(30); + ignore_timeout = 0; + readfd_unbuffered(sock_f_in, io_filesfrom_buf, + sizeof io_filesfrom_buf); } + exit_cleanup(RERR_STREAMIO); + } - sleep_for_bwlimit(ret); - - total += ret; + total += cnt; + defer_forwarding_messages = 1; - if (io_timeout) - last_io = time(NULL); + if (fd == sock_f_out) { + if (io_timeout || am_generator) + last_io_out = time(NULL); + sleep_for_bwlimit(cnt); } } + defer_forwarding_messages = defer_save; no_flush--; } - -static char *io_buffer; -static int io_buffer_count; - -void io_start_buffering(int fd) +static void msg2sndr_flush(void) { - if (io_buffer) return; - multiplex_out_fd = fd; - io_buffer = new_array(char, IO_BUFFER_SIZE); - if (!io_buffer) out_of_memory("writefd"); - io_buffer_count = 0; + if (defer_forwarding_messages) + return; + + while (msg2sndr.head && io_multiplexing_out) { + struct msg_list_item *m = msg2sndr.head; + if (!(msg2sndr.head = m->next)) + msg2sndr.tail = NULL; + stats.total_written += m->len; + defer_forwarding_messages = 1; + writefd_unbuffered(sock_f_out, m->buf, m->len); + defer_forwarding_messages = 0; + free(m); + } } /** * Write an message to a multiplexed stream. If this fails then rsync * exits. **/ -static void mplex_write(int fd, enum logcode code, char *buf, size_t len) +static void mplex_write(enum msgcode code, const char *buf, size_t len) { - char buffer[4096]; + char buffer[1024]; size_t n = len; SIVAL(buffer, 0, ((MPLEX_BASE + (int)code)<<24) + len); - if (n > (sizeof buffer - 4)) { - n = sizeof buffer - 4; - } + if (n > sizeof buffer - 4) + n = 0; + else + memcpy(buffer + 4, buf, n); - memcpy(&buffer[4], buf, n); - writefd_unbuffered(fd, buffer, n+4); + writefd_unbuffered(sock_f_out, buffer, n+4); len -= n; buf += n; if (len) { - writefd_unbuffered(fd, buf, len); + defer_forwarding_messages = 1; + writefd_unbuffered(sock_f_out, buf, len); + defer_forwarding_messages = 0; + msg2sndr_flush(); } } - -void io_flush(void) +void io_flush(int flush_it_all) { - int fd = multiplex_out_fd; - - err_list_push(); + msg2genr_flush(flush_it_all); + msg2sndr_flush(); - if (!io_buffer_count || no_flush) return; + if (!iobuf_out_cnt || no_flush) + return; - if (io_multiplexing_out) { - mplex_write(fd, FNONE, io_buffer, io_buffer_count); - } else { - writefd_unbuffered(fd, io_buffer, io_buffer_count); - } - io_buffer_count = 0; + if (io_multiplexing_out) + mplex_write(MSG_DATA, iobuf_out, iobuf_out_cnt); + else + writefd_unbuffered(sock_f_out, iobuf_out, iobuf_out_cnt); + iobuf_out_cnt = 0; } - -void io_end_buffering(void) +static void writefd(int fd, const char *buf, size_t len) { - io_flush(); - if (!io_multiplexing_out) { - free(io_buffer); - io_buffer = NULL; + if (fd == msg_fd_out) { + rprintf(FERROR, "Internal error: wrong write used in receiver.\n"); + exit_cleanup(RERR_PROTOCOL); } -} -static void writefd(int fd,char *buf,size_t len) -{ - stats.total_written += len; + if (fd == sock_f_out) + stats.total_written += len; - err_list_push(); + if (fd == write_batch_monitor_out) { + if ((size_t)write(batch_fd, buf, len) != len) + exit_cleanup(RERR_FILEIO); + } - if (!io_buffer || fd != multiplex_out_fd) { + if (!iobuf_out || fd != sock_f_out) { writefd_unbuffered(fd, buf, len); return; } while (len) { - int n = MIN((int) len, IO_BUFFER_SIZE-io_buffer_count); + int n = MIN((int)len, IO_BUFFER_SIZE - iobuf_out_cnt); if (n > 0) { - memcpy(io_buffer+io_buffer_count, buf, n); + memcpy(iobuf_out+iobuf_out_cnt, buf, n); buf += n; len -= n; - io_buffer_count += n; + iobuf_out_cnt += n; } - if (io_buffer_count == IO_BUFFER_SIZE) io_flush(); + if (iobuf_out_cnt == IO_BUFFER_SIZE) + io_flush(NORMAL_FLUSH); } } +void write_shortint(int f, int x) +{ + uchar b[2]; + b[0] = x; + b[1] = x >> 8; + writefd(f, (char *)b, 2); +} void write_int(int f,int32 x) { @@ -778,15 +1246,6 @@ void write_int(int f,int32 x) writefd(f,b,4); } - -void write_int_named(int f, int32 x, const char *phase) -{ - io_write_phase = phase; - write_int(f, x); - io_write_phase = phase_unknown; -} - - /* * Note: int64 may actually be a 32-bit type if ./configure couldn't find any * 64-bit types on this platform. @@ -800,8 +1259,8 @@ void write_longint(int f, int64 x) return; } -#ifdef NO_INT64 - rprintf(FERROR,"Integer overflow - attempted 64 bit offset\n"); +#if SIZEOF_INT64 < 8 + rprintf(FERROR, "Integer overflow: attempted 64-bit offset\n"); exit_cleanup(RERR_UNSUPPORTED); #else write_int(f, (int32)0xFFFFFFFF); @@ -812,30 +1271,48 @@ void write_longint(int f, int64 x) #endif } -void write_buf(int f,char *buf,size_t len) +void write_buf(int f, const char *buf, size_t len) { writefd(f,buf,len); } /** Write a string to the connection */ -static void write_sbuf(int f,char *buf) +void write_sbuf(int f, const char *buf) { - write_buf(f, buf, strlen(buf)); + writefd(f, buf, strlen(buf)); } - -void write_byte(int f,unsigned char c) +void write_byte(int f, uchar c) { - write_buf(f,(char *)&c,1); + writefd(f, (char *)&c, 1); } +void write_vstring(int f, const char *str, int len) +{ + uchar lenbuf[3], *lb = lenbuf; + + if (len > 0x7F) { + if (len > 0x7FFF) { + rprintf(FERROR, + "attempting to send over-long vstring (%d > %d)\n", + len, 0x7FFF); + exit_cleanup(RERR_PROTOCOL); + } + *lb++ = len / 0x100 + 0x80; + } + *lb = len; + writefd(f, (char*)lenbuf, lb - lenbuf + 1); + if (len) + writefd(f, str, len); +} /** - * Read a line of up to @p maxlen characters into @p buf. Does not - * contain a trailing newline or carriage return. + * Read a line of up to @p maxlen characters into @p buf (not counting + * the trailing null). Strips the (required) trailing newline and all + * carriage returns. * - * @return 1 for success; 0 for io error or truncation. + * @return 1 for success; 0 for I/O error or truncation. **/ int read_line(int f, char *buf, size_t maxlen) { @@ -844,71 +1321,97 @@ int read_line(int f, char *buf, size_t maxlen) read_buf(f, buf, 1); if (buf[0] == 0) return 0; - if (buf[0] == '\n') { - buf[0] = 0; + if (buf[0] == '\n') break; - } if (buf[0] != '\r') { buf++; maxlen--; } } - if (maxlen == 0) { - *buf = 0; - return 0; - } - - return 1; + *buf = '\0'; + return maxlen > 0; } - void io_printf(int fd, const char *format, ...) { - va_list ap; - char buf[1024]; + va_list ap; + char buf[BIGPATHBUFLEN]; int len; va_start(ap, format); len = vsnprintf(buf, sizeof buf, format, ap); va_end(ap); - if (len < 0) exit_cleanup(RERR_STREAMIO); + if (len < 0) + exit_cleanup(RERR_STREAMIO); + + if (len > (int)sizeof buf) { + rprintf(FERROR, "io_printf() was too long for the buffer.\n"); + exit_cleanup(RERR_STREAMIO); + } write_sbuf(fd, buf); } - -/** Setup for multiplexing an error stream with the data stream */ -void io_start_multiplex_out(int fd) +/** Setup for multiplexing a MSG_* stream with the data stream. */ +void io_start_multiplex_out(void) { - multiplex_out_fd = fd; - io_flush(); - io_start_buffering(fd); + io_flush(NORMAL_FLUSH); + io_start_buffering_out(); io_multiplexing_out = 1; } -/** Setup for multiplexing an error stream with the data stream */ -void io_start_multiplex_in(int fd) +/** Setup for multiplexing a MSG_* stream with the data stream. */ +void io_start_multiplex_in(void) { - multiplex_in_fd = fd; - io_flush(); + io_flush(NORMAL_FLUSH); + io_start_buffering_in(); io_multiplexing_in = 1; } -/** Write an message to the multiplexed error stream */ -int io_multiplex_write(enum logcode code, char *buf, size_t len) +/** Write an message to the multiplexed data stream. */ +int io_multiplex_write(enum msgcode code, const char *buf, size_t len) { - if (!io_multiplexing_out) return 0; + if (!io_multiplexing_out) + return 0; - io_flush(); + io_flush(NORMAL_FLUSH); stats.total_written += (len+4); - mplex_write(multiplex_out_fd, code, buf, len); + mplex_write(code, buf, len); return 1; } -/** Stop output multiplexing */ -void io_multiplexing_close(void) +void close_multiplexing_in(void) +{ + io_multiplexing_in = 0; +} + +/** Stop output multiplexing. */ +void close_multiplexing_out(void) { io_multiplexing_out = 0; } +void start_write_batch(int fd) +{ + write_stream_flags(batch_fd); + + /* Some communication has already taken place, but we don't + * enable batch writing until here so that we can write a + * canonical record of the communication even though the + * actual communication so far depends on whether a daemon + * is involved. */ + write_int(batch_fd, protocol_version); + write_int(batch_fd, checksum_seed); + + if (am_sender) + write_batch_monitor_out = fd; + else + write_batch_monitor_in = fd; +} + +void stop_write_batch(void) +{ + write_batch_monitor_out = -1; + write_batch_monitor_in = -1; +}