sender.c revision 1.17
1/* $Id: sender.c,v 1.17 2019/02/21 22:08:53 benno Exp $ */ 2/* 3 * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv> 4 * 5 * Permission to use, copy, modify, and distribute this software for any 6 * purpose with or without fee is hereby granted, provided that the above 7 * copyright notice and this permission notice appear in all copies. 8 * 9 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 10 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 11 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 12 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 13 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 14 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 15 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 16 */ 17#include <sys/mman.h> 18#include <sys/queue.h> 19#include <sys/stat.h> 20 21#include <assert.h> 22#include <fcntl.h> 23#include <inttypes.h> 24#include <poll.h> 25#include <stdlib.h> 26#include <string.h> 27#include <unistd.h> 28 29#include <openssl/md4.h> 30 31#include "extern.h" 32 33/* 34 * A request from the receiver to download updated file data. 35 */ 36struct send_dl { 37 int32_t idx; /* index in our file list */ 38 struct blkset *blks; /* the sender's block information */ 39 TAILQ_ENTRY(send_dl) entries; 40}; 41 42/* 43 * The current file being "updated": sent from sender to receiver. 44 * If there is no file being uploaded, "cur" is NULL. 45 */ 46struct send_up { 47 struct send_dl *cur; /* file being updated or NULL */ 48 struct blkstat stat; /* status of file being updated */ 49}; 50 51TAILQ_HEAD(send_dlq, send_dl); 52 53/* 54 * We have finished updating the receiver's file with sender data. 55 * Deallocate and wipe clean all resources required for that. 56 */ 57static void 58send_up_reset(struct send_up *p) 59{ 60 61 assert(p != NULL); 62 63 /* Free the download request, if applicable. */ 64 65 if (p->cur != NULL) { 66 free(p->cur->blks); 67 free(p->cur); 68 p->cur = NULL; 69 } 70 71 /* If we mapped a file for scanning, unmap it and close. */ 72 73 if (p->stat.map != MAP_FAILED) 74 munmap(p->stat.map, p->stat.mapsz); 75 76 p->stat.map = MAP_FAILED; 77 p->stat.mapsz = 0; 78 79 if (p->stat.fd != -1) 80 close(p->stat.fd); 81 82 p->stat.fd = -1; 83 84 /* Now clear the in-transfer information. */ 85 86 p->stat.offs = 0; 87 p->stat.hint = 0; 88 p->stat.curst = BLKSTAT_NONE; 89} 90 91/* 92 * This is the bulk of the sender work. 93 * Here we tend to an output buffer that responds to receiver requests 94 * for data. 95 * This does not act upon the output descriptor itself so as to avoid 96 * blocking, which otherwise would deadlock the protocol. 97 * Returns zero on failure, non-zero on success. 98 */ 99static int 100send_up_fsm(struct sess *sess, size_t *phase, 101 struct send_up *up, void **wb, size_t *wbsz, size_t *wbmax, 102 const struct flist *fl) 103{ 104 size_t pos = 0, isz = sizeof(int32_t), 105 dsz = MD4_DIGEST_LENGTH; 106 unsigned char fmd[MD4_DIGEST_LENGTH]; 107 off_t sz; 108 char buf[20]; 109 110 switch (up->stat.curst) { 111 case BLKSTAT_DATA: 112 /* 113 * A data segment to be written: buffer both the length 114 * and the data. 115 * If we've finished the transfer, move on to the token; 116 * otherwise, keep sending data. 117 */ 118 119 sz = MINIMUM(MAX_CHUNK, 120 up->stat.curlen - up->stat.curpos); 121 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, isz)) { 122 ERRX1(sess, "io_lowbuffer_alloc"); 123 return 0; 124 } 125 io_lowbuffer_int(sess, *wb, &pos, *wbsz, sz); 126 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, sz)) { 127 ERRX1(sess, "io_lowbuffer_alloc"); 128 return 0; 129 } 130 io_lowbuffer_buf(sess, *wb, &pos, *wbsz, 131 up->stat.map + up->stat.curpos, sz); 132 133 up->stat.curpos += sz; 134 if (up->stat.curpos == up->stat.curlen) 135 up->stat.curst = BLKSTAT_TOK; 136 return 1; 137 case BLKSTAT_TOK: 138 /* 139 * The data token following (maybe) a data segment. 140 * These can also come standalone if, say, the file's 141 * being fully written. 142 * It's followed by a hash or another data segment, 143 * depending on the token. 144 */ 145 146 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, isz)) { 147 ERRX1(sess, "io_lowbuffer_alloc"); 148 return 0; 149 } 150 io_lowbuffer_int(sess, *wb, 151 &pos, *wbsz, up->stat.curtok); 152 up->stat.curst = up->stat.curtok ? 153 BLKSTAT_NEXT : BLKSTAT_HASH; 154 return 1; 155 case BLKSTAT_HASH: 156 /* 157 * The hash following transmission of all file contents. 158 * This is always followed by the state that we're 159 * finished with the file. 160 */ 161 162 hash_file(up->stat.map, up->stat.mapsz, fmd, sess); 163 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, dsz)) { 164 ERRX1(sess, "io_lowbuffer_alloc"); 165 return 0; 166 } 167 io_lowbuffer_buf(sess, *wb, &pos, *wbsz, fmd, dsz); 168 up->stat.curst = BLKSTAT_DONE; 169 return 1; 170 case BLKSTAT_DONE: 171 /* 172 * The data has been written. 173 * Clear our current send file and allow the block below 174 * to find another. 175 */ 176 177 if (!sess->opts->dry_run) 178 LOG3(sess, "%s: flushed %jd KB total, %.2f%% " 179 "uploaded", fl[up->cur->idx].path, 180 (intmax_t)up->stat.total / 1024, 181 100.0 * up->stat.dirty / up->stat.total); 182 send_up_reset(up); 183 return 1; 184 case BLKSTAT_PHASE: 185 /* 186 * This is where we actually stop the algorithm: we're 187 * already at the second phase. 188 */ 189 190 send_up_reset(up); 191 (*phase)++; 192 return 1; 193 case BLKSTAT_NEXT: 194 /* 195 * Our last case: we need to find the 196 * next block (and token) to transmit to 197 * the receiver. 198 * These will drive the finite state 199 * machine in the first few conditional 200 * blocks of this set. 201 */ 202 203 assert(up->stat.fd != -1); 204 blk_match(sess, up->cur->blks, 205 fl[up->cur->idx].path, &up->stat); 206 return 1; 207 case BLKSTAT_NONE: 208 break; 209 } 210 211 assert(BLKSTAT_NONE == up->stat.curst); 212 213 /* 214 * We've either hit the phase change following the last file (or 215 * start, or prior phase change), or we need to prime the next 216 * file for transmission. 217 * We special-case dry-run mode. 218 */ 219 220 if (up->cur->idx < 0) { 221 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, isz)) { 222 ERRX1(sess, "io_lowbuffer_alloc"); 223 return 0; 224 } 225 io_lowbuffer_int(sess, *wb, &pos, *wbsz, -1); 226 227 if (sess->opts->server && sess->rver > 27) { 228 if (!io_lowbuffer_alloc(sess, 229 wb, wbsz, wbmax, isz)) { 230 ERRX1(sess, "io_lowbuffer_alloc"); 231 return 0; 232 } 233 io_lowbuffer_int(sess, *wb, &pos, *wbsz, -1); 234 } 235 up->stat.curst = BLKSTAT_PHASE; 236 } else if (sess->opts->dry_run) { 237 if (!sess->opts->server) 238 LOG1(sess, "%s", fl[up->cur->idx].wpath); 239 240 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, isz)) { 241 ERRX1(sess, "io_lowbuffer_alloc"); 242 return 0; 243 } 244 io_lowbuffer_int(sess, *wb, &pos, *wbsz, up->cur->idx); 245 up->stat.curst = BLKSTAT_DONE; 246 } else { 247 assert(up->stat.fd != -1); 248 249 /* 250 * FIXME: use the nice output of log_file() and so on in 251 * downloader.c, which means moving this into 252 * BLKSTAT_DONE instead of having it be here. 253 */ 254 255 if (!sess->opts->server) 256 LOG1(sess, "%s", fl[up->cur->idx].wpath); 257 258 if (!io_lowbuffer_alloc(sess, wb, wbsz, wbmax, 20)) { 259 ERRX1(sess, "io_lowbuffer_alloc"); 260 return 0; 261 } 262 assert(sizeof(buf) == 20); 263 blk_recv_ack(sess, buf, up->cur->blks, up->cur->idx); 264 io_lowbuffer_buf(sess, *wb, &pos, *wbsz, buf, 20); 265 266 LOG3(sess, "%s: primed for %jd B total", 267 fl[up->cur->idx].path, 268 (intmax_t)up->cur->blks->size); 269 up->stat.curst = BLKSTAT_NEXT; 270 } 271 272 return 1; 273} 274 275/* 276 * Enqueue a download request, getting it off the read channel as 277 * quickly a possible. 278 * This frees up the read channel for further incoming requests. 279 * We'll handle each element in turn, up to and including the last 280 * request (phase change), which is always a -1 idx. 281 * Returns zero on failure, non-zero on success. 282 */ 283static int 284send_dl_enqueue(struct sess *sess, struct send_dlq *q, 285 int32_t idx, const struct flist *fl, size_t flsz, int fd) 286{ 287 struct send_dl *s; 288 289 /* End-of-phase marker. */ 290 291 if (idx == -1) { 292 if ((s = calloc(1, sizeof(struct send_dl))) == NULL) { 293 ERR(sess, "calloc"); 294 return 0; 295 } 296 s->idx = -1; 297 s->blks = NULL; 298 TAILQ_INSERT_TAIL(q, s, entries); 299 return 1; 300 } 301 302 /* Validate the index. */ 303 304 if (idx < 0 || (uint32_t)idx >= flsz) { 305 ERRX(sess, "file index out of bounds: invalid %" 306 PRId32 " out of %zu", idx, flsz); 307 return 0; 308 } else if (S_ISDIR(fl[idx].st.mode)) { 309 ERRX(sess, "blocks requested for " 310 "directory: %s", fl[idx].path); 311 return 0; 312 } else if (S_ISLNK(fl[idx].st.mode)) { 313 ERRX(sess, "blocks requested for " 314 "symlink: %s", fl[idx].path); 315 return 0; 316 } else if (!S_ISREG(fl[idx].st.mode)) { 317 ERRX(sess, "blocks requested for " 318 "special: %s", fl[idx].path); 319 return 0; 320 } 321 322 if ((s = calloc(1, sizeof(struct send_dl))) == NULL) { 323 ERR(sess, "callloc"); 324 return 0; 325 } 326 s->idx = idx; 327 s->blks = NULL; 328 TAILQ_INSERT_TAIL(q, s, entries); 329 330 /* 331 * This blocks til the full blockset has been read. 332 * That's ok, because the most important thing is getting data 333 * off the wire. 334 */ 335 336 if (!sess->opts->dry_run) { 337 s->blks = blk_recv(sess, fd, fl[idx].path); 338 if (s->blks == NULL) { 339 ERRX1(sess, "blk_recv"); 340 return 0; 341 } 342 } 343 return 1; 344} 345 346/* 347 * A client sender manages the read-only source files and sends data to 348 * the receiver as requested. 349 * First it sends its list of files, then it waits for the server to 350 * request updates to individual files. 351 * It queues requests for updates as soon as it receives them. 352 * Returns zero on failure, non-zero on success. 353 * 354 * Pledges: stdio, rpath, unveil. 355 */ 356int 357rsync_sender(struct sess *sess, int fdin, 358 int fdout, size_t argc, char **argv) 359{ 360 struct flist *fl = NULL; 361 const struct flist *f; 362 size_t i, flsz = 0, phase = 0, excl; 363 int rc = 0, c; 364 int32_t idx; 365 struct pollfd pfd[3]; 366 struct send_dlq sdlq; 367 struct send_dl *dl; 368 struct send_up up; 369 struct stat st; 370 void *wbuf = NULL; 371 size_t wbufpos = 0, wbufsz = 0, wbufmax = 0; 372 ssize_t ssz; 373 374 if (pledge("stdio getpw rpath unveil", NULL) == -1) { 375 ERR(sess, "pledge"); 376 return 0; 377 } 378 379 memset(&up, 0, sizeof(struct send_up)); 380 TAILQ_INIT(&sdlq); 381 up.stat.fd = -1; 382 up.stat.map = MAP_FAILED; 383 384 /* 385 * Generate the list of files we want to send from our 386 * command-line input. 387 * This will also remove all invalid files. 388 */ 389 390 if (!flist_gen(sess, argc, argv, &fl, &flsz)) { 391 ERRX1(sess, "flist_gen"); 392 goto out; 393 } 394 395 /* Client sends zero-length exclusions if deleting. */ 396 397 if (!sess->opts->server && sess->opts->del && 398 !io_write_int(sess, fdout, 0)) { 399 ERRX1(sess, "io_write_int"); 400 goto out; 401 } 402 403 /* 404 * Then the file list in any mode. 405 * Finally, the IO error (always zero for us). 406 */ 407 408 if (!flist_send(sess, fdin, fdout, fl, flsz)) { 409 ERRX1(sess, "flist_send"); 410 goto out; 411 } else if (!io_write_int(sess, fdout, 0)) { 412 ERRX1(sess, "io_write_int"); 413 goto out; 414 } 415 416 /* Exit if we're the server with zero files. */ 417 418 if (flsz == 0 && sess->opts->server) { 419 WARNX(sess, "sender has empty file list: exiting"); 420 rc = 1; 421 goto out; 422 } else if (!sess->opts->server) 423 LOG1(sess, "Transfer starting: %zu files", flsz); 424 425 /* 426 * If we're the server, read our exclusion list. 427 * This is always 0 for now. 428 */ 429 430 if (sess->opts->server) { 431 if (!io_read_size(sess, fdin, &excl)) { 432 ERRX1(sess, "io_read_size"); 433 goto out; 434 } else if (excl != 0) { 435 ERRX1(sess, "exclusion list is non-empty"); 436 goto out; 437 } 438 } 439 440 /* 441 * Set up our poll events. 442 * We start by polling only in receiver requests, enabling other 443 * poll events on demand. 444 */ 445 446 pfd[0].fd = fdin; /* from receiver */ 447 pfd[0].events = POLLIN; 448 pfd[1].fd = -1; /* to receiver */ 449 pfd[1].events = POLLOUT; 450 pfd[2].fd = -1; /* from local file */ 451 pfd[2].events = POLLIN; 452 453 for (;;) { 454 assert(pfd[0].fd != -1); 455 if ((c = poll(pfd, 3, POLL_TIMEOUT)) == -1) { 456 ERR(sess, "poll"); 457 goto out; 458 } else if (c == 0) { 459 ERRX(sess, "poll: timeout"); 460 goto out; 461 } 462 for (i = 0; i < 3; i++) 463 if (pfd[i].revents & (POLLERR|POLLNVAL)) { 464 ERRX(sess, "poll: bad fd"); 465 goto out; 466 } else if (pfd[i].revents & POLLHUP) { 467 ERRX(sess, "poll: hangup"); 468 goto out; 469 } 470 471 /* 472 * If we have a request coming down off the wire, pull 473 * it in as quickly as possible into our buffer. 474 * This unclogs the socket buffers so the data can flow. 475 * FIXME: if we're multiplexing, we might stall here if 476 * there's only a log message and no actual data. 477 * This can be fixed by doing a conditional test. 478 */ 479 480 if (pfd[0].revents & POLLIN) 481 for (;;) { 482 if (!io_read_int(sess, fdin, &idx)) { 483 ERRX1(sess, "io_read_int"); 484 goto out; 485 } 486 if (!send_dl_enqueue(sess, 487 &sdlq, idx, fl, flsz, fdin)) { 488 ERRX1(sess, "send_dl_enqueue"); 489 goto out; 490 } 491 c = io_read_check(sess, fdin); 492 if (c < 0) { 493 ERRX1(sess, "io_read_check"); 494 goto out; 495 } else if (c == 0) 496 break; 497 } 498 499 /* 500 * One of our local files has been opened in response 501 * to a receiver request and now we can map it. 502 * We'll respond to the event by looking at the map when 503 * the writer is available. 504 * Here we also enable the poll event for output. 505 */ 506 507 if (pfd[2].revents & POLLIN) { 508 assert(up.cur != NULL); 509 assert(up.stat.fd != -1); 510 assert(up.stat.map == MAP_FAILED); 511 assert(up.stat.mapsz == 0); 512 f = &fl[up.cur->idx]; 513 514 if (fstat(up.stat.fd, &st) == -1) { 515 ERR(sess, "%s: fstat", f->path); 516 goto out; 517 } 518 519 /* 520 * If the file is zero-length, the map will 521 * fail, but either way we want to unset that 522 * we're waiting for the file to open and set 523 * that we're ready for the output channel. 524 */ 525 526 if ((up.stat.mapsz = st.st_size) > 0) { 527 up.stat.map = mmap(NULL, 528 up.stat.mapsz, PROT_READ, 529 MAP_SHARED, up.stat.fd, 0); 530 if (up.stat.map == MAP_FAILED) { 531 ERR(sess, "%s: mmap", f->path); 532 goto out; 533 } 534 } 535 536 pfd[2].fd = -1; 537 pfd[1].fd = fdout; 538 } 539 540 /* 541 * If we have buffers waiting to write, write them out 542 * as soon as we can in a non-blocking fashion. 543 * We must not be waiting for any local files. 544 * ALL WRITES MUST HAPPEN HERE. 545 * This keeps the sender deadlock-free. 546 */ 547 548 if ((pfd[1].revents & POLLOUT) && wbufsz > 0) { 549 assert(pfd[2].fd == -1); 550 assert(wbufsz - wbufpos); 551 ssz = write(fdout, 552 wbuf + wbufpos, wbufsz - wbufpos); 553 if (ssz < 0) { 554 ERR(sess, "write"); 555 goto out; 556 } 557 wbufpos += ssz; 558 if (wbufpos == wbufsz) 559 wbufpos = wbufsz = 0; 560 pfd[1].revents &= ~POLLOUT; 561 562 /* This is usually in io.c... */ 563 564 sess->total_write += ssz; 565 } 566 567 /* 568 * Engage the FSM for the current transfer. 569 * If our phase changes, stop processing. 570 */ 571 572 if (pfd[1].revents & POLLOUT && up.cur != NULL) { 573 assert(pfd[2].fd == -1); 574 assert(wbufpos == 0 && wbufsz == 0); 575 if (!send_up_fsm(sess, &phase, 576 &up, &wbuf, &wbufsz, &wbufmax, fl)) { 577 ERRX1(sess, "send_up_fsm"); 578 goto out; 579 } else if (phase > 1) 580 break; 581 } 582 583 /* 584 * Incoming queue management. 585 * If we have no queue component that we're waiting on, 586 * then pull off the receiver-request queue and start 587 * processing the request. 588 */ 589 590 if (up.cur == NULL) { 591 assert(pfd[2].fd == -1); 592 assert(up.stat.fd == -1); 593 assert(up.stat.map == MAP_FAILED); 594 assert(up.stat.mapsz == 0); 595 assert(wbufsz == 0 && wbufpos == 0); 596 pfd[1].fd = -1; 597 598 /* 599 * If there's nothing in the queue, then keep 600 * the output channel disabled and wait for 601 * whatever comes next from the reader. 602 */ 603 604 if ((up.cur = TAILQ_FIRST(&sdlq)) == NULL) 605 continue; 606 607 TAILQ_REMOVE(&sdlq, up.cur, entries); 608 609 /* 610 * End of phase: enable channel to receiver. 611 * We'll need our output buffer enabled in order 612 * to process this event. 613 */ 614 615 if (up.cur->idx == -1) { 616 pfd[1].fd = fdout; 617 continue; 618 } 619 620 /* 621 * Non-blocking open of file. 622 * This will be picked up in the state machine 623 * block of not being primed. 624 */ 625 626 up.stat.fd = open(fl[up.cur->idx].path, 627 O_RDONLY|O_NONBLOCK, 0); 628 if (up.stat.fd == -1) { 629 ERR(sess, "%s: open", fl[up.cur->idx].path); 630 goto out; 631 } 632 pfd[2].fd = up.stat.fd; 633 } 634 } 635 636 if (!TAILQ_EMPTY(&sdlq)) { 637 ERRX(sess, "phases complete with files still queued"); 638 goto out; 639 } 640 641 if (!sess_stats_send(sess, fdout)) { 642 ERRX1(sess, "sess_stats_end"); 643 goto out; 644 } 645 646 /* Final "goodbye" message. */ 647 648 if (!io_read_int(sess, fdin, &idx)) { 649 ERRX1(sess, "io_read_int"); 650 goto out; 651 } else if (idx != -1) { 652 ERRX(sess, "read incorrect update complete ack"); 653 goto out; 654 } 655 656 LOG2(sess, "sender finished updating"); 657 rc = 1; 658out: 659 send_up_reset(&up); 660 while ((dl = TAILQ_FIRST(&sdlq)) != NULL) { 661 TAILQ_REMOVE(&sdlq, dl, entries); 662 free(dl->blks); 663 free(dl); 664 } 665 flist_free(fl, flsz); 666 free(wbuf); 667 return rc; 668} 669