Annotation of src/usr.bin/rsync/blocks.c, Revision 1.14
1.14 ! deraadt 1: /* $Id: blocks.c,v 1.13 2019/02/18 22:47:34 benno Exp $ */
1.1 benno 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/stat.h>
18:
19: #include <assert.h>
20: #include <endian.h>
21: #include <errno.h>
22: #include <inttypes.h>
23: #include <stdio.h>
24: #include <stdlib.h>
25: #include <string.h>
26: #include <unistd.h>
27:
1.6 tb 28: #include <openssl/md4.h>
29:
1.1 benno 30: #include "extern.h"
31:
32: /*
33: * From our current position of "offs" in buffer "buf" of total size
34: * "size", see if we can find a matching block in our list of blocks.
35: * The "hint" refers to the block that *might* work.
36: * Returns the blk or NULL if no matching block was found.
37: */
38: static struct blk *
39: blk_find(struct sess *sess, const void *buf, off_t size, off_t offs,
40: const struct blkset *blks, const char *path, size_t hint)
41: {
42: unsigned char md[MD4_DIGEST_LENGTH];
43: uint32_t fhash;
44: off_t remain, osz;
45: size_t i;
46: int have_md = 0;
47:
48: /*
49: * First, compute our fast hash.
50: * FIXME: yes, this can be a rolling computation, but I'm
51: * deliberately making it simple first.
52: */
53:
54: remain = size - offs;
55: assert(remain);
56: osz = remain < (off_t)blks->len ? remain : (off_t)blks->len;
57: fhash = hash_fast(buf + offs, (size_t)osz);
58: have_md = 0;
59:
60: /*
61: * Start with our match hint.
62: * This just runs the fast and slow check with the hint.
63: */
64:
65: if (hint < blks->blksz &&
66: fhash == blks->blks[hint].chksum_short &&
67: (size_t)osz == blks->blks[hint].len) {
68: hash_slow(buf + offs, (size_t)osz, md, sess);
69: have_md = 1;
1.4 deraadt 70: if (memcmp(md, blks->blks[hint].chksum_long, blks->csum) == 0) {
1.1 benno 71: LOG4(sess, "%s: found matching hinted match: "
1.14 ! deraadt 72: "position %jd, block %zu (position %jd, size %zu)",
! 73: path,
! 74: (intmax_t)offs, blks->blks[hint].idx,
! 75: (intmax_t)blks->blks[hint].offs,
! 76: blks->blks[hint].len);
1.1 benno 77: return &blks->blks[hint];
78: }
79: }
80:
81: /*
82: * Now loop and look for the fast hash.
83: * If it's found, move on to the slow hash.
84: */
85:
86: for (i = 0; i < blks->blksz; i++) {
87: if (fhash != blks->blks[i].chksum_short)
88: continue;
89: if ((size_t)osz != blks->blks[i].len)
90: continue;
91:
92: LOG4(sess, "%s: found matching fast match: "
1.14 ! deraadt 93: "position %jd, block %zu (position %jd, size %zu)",
! 94: path,
! 95: (intmax_t)offs, blks->blks[i].idx,
! 96: (intmax_t)blks->blks[i].offs,
! 97: blks->blks[i].len);
1.1 benno 98:
99: /* Compute slow hash on demand. */
100:
1.4 deraadt 101: if (have_md == 0) {
1.1 benno 102: hash_slow(buf + offs, (size_t)osz, md, sess);
103: have_md = 1;
104: }
105:
106: if (memcmp(md, blks->blks[i].chksum_long, blks->csum))
107: continue;
108:
109: LOG4(sess, "%s: sender verifies slow match", path);
110: return &blks->blks[i];
111: }
112:
113: return NULL;
114: }
115:
116: /*
117: * Given a local file "path" and the blocks created by a remote machine,
118: * find out which blocks of our file they don't have and send them.
1.7 florian 119: * This function is reentrant: it must be called while there's still
120: * data to send.
1.1 benno 121: */
1.9 florian 122: void
123: blk_match(struct sess *sess, const struct blkset *blks,
1.7 florian 124: const char *path, struct blkstat *st)
1.1 benno 125: {
1.9 florian 126: off_t last, end, sz;
127: int32_t tok;
128: struct blk *blk;
1.1 benno 129:
130: /*
131: * If the file's empty or we don't have any blocks from the
132: * sender, then simply send the whole file.
133: * Otherwise, run the hash matching routine and send raw chunks
134: * and subsequent matching tokens.
135: */
136:
1.7 florian 137: if (st->mapsz && blks->blksz) {
1.9 florian 138: /*
139: * Stop searching at the length of the file minus the
140: * size of the last block.
141: * The reason for this being that we don't need to do an
142: * incremental hash within the last block---if it
143: * doesn't match, it doesn't match.
144: */
145:
146: end = st->mapsz + 1 - blks->blks[blks->blksz - 1].len;
147: last = st->offs;
148:
149: for ( ; st->offs < end; st->offs++) {
150: blk = blk_find(sess, st->map, st->mapsz,
151: st->offs, blks, path, st->hint);
152: if (blk == NULL)
153: continue;
154:
155: sz = st->offs - last;
156: st->dirty += sz;
157: st->total += sz;
1.14 ! deraadt 158: LOG4(sess,
! 159: "%s: flushing %jd B before %zu B block %zu",
! 160: path, (intmax_t)sz,
! 161: blk->len, blk->idx);
1.9 florian 162: tok = -(blk->idx + 1);
163:
164: /*
165: * Write the data we have, then follow it with
166: * the tag of the block that matches.
167: */
168:
169: st->curpos = last;
170: st->curlen = st->curpos + sz;
171: st->curtok = tok;
1.10 deraadt 172: assert(st->curtok != 0);
1.9 florian 173: st->curst = sz ? BLKSTAT_DATA : BLKSTAT_TOK;
174: st->total += blk->len;
175: st->offs += blk->len;
176: st->hint = blk->idx + 1;
177: return;
1.1 benno 178: }
179:
1.9 florian 180: /* Emit remaining data and send terminator token. */
181:
182: sz = st->mapsz - last;
183: LOG4(sess, "%s: flushing remaining %jd B",
1.14 ! deraadt 184: path, (intmax_t)sz);
1.1 benno 185:
1.9 florian 186: st->total += sz;
187: st->dirty += sz;
188: st->curpos = last;
189: st->curlen = st->curpos + sz;
190: st->curtok = 0;
191: st->curst = sz ? BLKSTAT_DATA : BLKSTAT_TOK;
192: } else {
193: st->curpos = 0;
194: st->curlen = st->mapsz;
195: st->curtok = 0;
196: st->curst = st->mapsz ? BLKSTAT_DATA : BLKSTAT_TOK;
197: st->dirty = st->total = st->mapsz;
1.1 benno 198:
1.9 florian 199: LOG4(sess, "%s: flushing whole file %zu B",
1.14 ! deraadt 200: path, st->mapsz);
1.1 benno 201: }
202: }
203:
204: /*
1.13 benno 205: * Buffer the message from sender to the receiver indicating that the
206: * block set has been received.
1.1 benno 207: * Symmetrises blk_send_ack().
208: */
1.13 benno 209: void
210: blk_recv_ack(struct sess *sess, char buf[20],
211: const struct blkset *blocks, int32_t idx)
1.1 benno 212: {
1.8 florian 213: size_t pos = 0, sz;
1.1 benno 214:
1.8 florian 215: sz = sizeof(int32_t) + /* index */
216: sizeof(int32_t) + /* block count */
217: sizeof(int32_t) + /* block length */
218: sizeof(int32_t) + /* checksum length */
219: sizeof(int32_t); /* block remainder */
1.13 benno 220: assert(sz == 20);
1.8 florian 221:
222: io_buffer_int(sess, buf, &pos, sz, idx);
223: io_buffer_int(sess, buf, &pos, sz, blocks->blksz);
224: io_buffer_int(sess, buf, &pos, sz, blocks->len);
225: io_buffer_int(sess, buf, &pos, sz, blocks->csum);
226: io_buffer_int(sess, buf, &pos, sz, blocks->rem);
227: assert(pos == sz);
1.1 benno 228: }
229:
230: /*
231: * Read all of the checksums for a file's blocks.
232: * Returns the set of blocks or NULL on failure.
233: */
234: struct blkset *
235: blk_recv(struct sess *sess, int fd, const char *path)
236: {
237: struct blkset *s;
238: int32_t i;
239: size_t j;
240: struct blk *b;
241: off_t offs = 0;
242:
1.4 deraadt 243: if ((s = calloc(1, sizeof(struct blkset))) == NULL) {
1.1 benno 244: ERR(sess, "calloc");
245: return NULL;
246: }
247:
248: /*
249: * The block prologue consists of a few values that we'll need
250: * in reading the individual blocks for this file.
251: * FIXME: read into buffer and unbuffer.
252: */
253:
1.3 deraadt 254: if (!io_read_size(sess, fd, &s->blksz)) {
1.1 benno 255: ERRX1(sess, "io_read_size");
256: goto out;
1.3 deraadt 257: } else if (!io_read_size(sess, fd, &s->len)) {
1.1 benno 258: ERRX1(sess, "io_read_size");
259: goto out;
1.3 deraadt 260: } else if (!io_read_size(sess, fd, &s->csum)) {
1.1 benno 261: ERRX1(sess, "io_read_int");
262: goto out;
1.3 deraadt 263: } else if (!io_read_size(sess, fd, &s->rem)) {
1.1 benno 264: ERRX1(sess, "io_read_int");
265: goto out;
266: } else if (s->rem && s->rem >= s->len) {
267: ERRX(sess, "block remainder is "
268: "greater than block size");
269: goto out;
270: }
271:
272: LOG3(sess, "%s: read block prologue: %zu blocks of "
1.14 ! deraadt 273: "%zu B, %zu B remainder, %zu B checksum", path,
! 274: s->blksz, s->len, s->rem, s->csum);
1.1 benno 275:
276: if (s->blksz) {
277: s->blks = calloc(s->blksz, sizeof(struct blk));
1.4 deraadt 278: if (s->blks == NULL) {
1.1 benno 279: ERR(sess, "calloc");
280: goto out;
281: }
282: }
283:
1.2 benno 284: /*
1.1 benno 285: * Read each block individually.
286: * FIXME: read buffer and unbuffer.
287: */
288:
289: for (j = 0; j < s->blksz; j++) {
290: b = &s->blks[j];
1.3 deraadt 291: if (!io_read_int(sess, fd, &i)) {
1.1 benno 292: ERRX1(sess, "io_read_int");
293: goto out;
294: }
295: b->chksum_short = i;
296:
297: assert(s->csum <= sizeof(b->chksum_long));
1.3 deraadt 298: if (!io_read_buf(sess,
1.1 benno 299: fd, b->chksum_long, s->csum)) {
300: ERRX1(sess, "io_read_buf");
301: goto out;
302: }
303:
304: /*
305: * If we're the last block, then we're assigned the
306: * remainder of the data.
307: */
308:
309: b->offs = offs;
310: b->idx = j;
311: b->len = (j == (s->blksz - 1) && s->rem) ?
312: s->rem : s->len;
313: offs += b->len;
314:
1.14 ! deraadt 315: LOG4(sess, "%s: read block %zu, length %zu B",
! 316: path, b->idx, b->len);
1.1 benno 317: }
318:
319: s->size = offs;
1.14 ! deraadt 320: LOG3(sess, "%s: read blocks: %zu blocks, %jd B total blocked data",
! 321: path, s->blksz, (intmax_t)s->size);
1.1 benno 322: return s;
323: out:
1.7 florian 324: free(s->blks);
325: free(s);
1.1 benno 326: return NULL;
327: }
328:
329: /*
330: * Symmetrise blk_recv_ack(), except w/o the leading identifier.
331: * Return zero on failure, non-zero on success.
332: */
333: int
334: blk_send_ack(struct sess *sess, int fd, struct blkset *p)
335: {
336: char buf[16];
337: size_t pos = 0, sz;
338:
339: /* Put the entire send routine into a buffer. */
340:
341: sz = sizeof(int32_t) + /* block count */
342: sizeof(int32_t) + /* block length */
343: sizeof(int32_t) + /* checksum length */
344: sizeof(int32_t); /* block remainder */
345: assert(sz <= sizeof(buf));
346:
1.3 deraadt 347: if (!io_read_buf(sess, fd, buf, sz)) {
1.1 benno 348: ERRX1(sess, "io_read_buf");
349: return 0;
350: }
351:
1.3 deraadt 352: if (!io_unbuffer_size(sess, buf, &pos, sz, &p->blksz))
1.1 benno 353: ERRX1(sess, "io_unbuffer_size");
1.3 deraadt 354: else if (!io_unbuffer_size(sess, buf, &pos, sz, &p->len))
1.1 benno 355: ERRX1(sess, "io_unbuffer_size");
1.3 deraadt 356: else if (!io_unbuffer_size(sess, buf, &pos, sz, &p->csum))
1.1 benno 357: ERRX1(sess, "io_unbuffer_size");
1.3 deraadt 358: else if (!io_unbuffer_size(sess, buf, &pos, sz, &p->rem))
1.1 benno 359: ERRX1(sess, "io_unbuffer_size");
360: else if (p->len && p->rem >= p->len)
361: ERRX1(sess, "non-zero length is less than remainder");
1.4 deraadt 362: else if (p->csum == 0 || p->csum > 16)
1.1 benno 363: ERRX1(sess, "inappropriate checksum length");
364: else
365: return 1;
366:
367: return 0;
368: }
369:
370: /*
371: * Transmit the metadata for set and blocks.
372: * Return zero on failure, non-zero on success.
373: */
374: int
375: blk_send(struct sess *sess, int fd, size_t idx,
376: const struct blkset *p, const char *path)
377: {
378: char *buf;
379: size_t i, pos = 0, sz;
380: int rc = 0;
381:
382: /* Put the entire send routine into a buffer. */
383:
384: sz = sizeof(int32_t) + /* identifier */
1.2 benno 385: sizeof(int32_t) + /* block count */
386: sizeof(int32_t) + /* block length */
387: sizeof(int32_t) + /* checksum length */
388: sizeof(int32_t) + /* block remainder */
389: p->blksz *
390: (sizeof(int32_t) + /* short checksum */
391: p->csum); /* long checksum */
1.1 benno 392:
1.4 deraadt 393: if ((buf = malloc(sz)) == NULL) {
1.1 benno 394: ERR(sess, "malloc");
395: return 0;
396: }
397:
398: io_buffer_int(sess, buf, &pos, sz, idx);
399: io_buffer_int(sess, buf, &pos, sz, p->blksz);
400: io_buffer_int(sess, buf, &pos, sz, p->len);
401: io_buffer_int(sess, buf, &pos, sz, p->csum);
402: io_buffer_int(sess, buf, &pos, sz, p->rem);
403:
404: for (i = 0; i < p->blksz; i++) {
1.2 benno 405: io_buffer_int(sess, buf, &pos,
1.1 benno 406: sz, p->blks[i].chksum_short);
1.2 benno 407: io_buffer_buf(sess, buf, &pos, sz,
1.1 benno 408: p->blks[i].chksum_long, p->csum);
409: }
410:
411: assert(pos == sz);
412:
1.3 deraadt 413: if (!io_write_buf(sess, fd, buf, sz)) {
1.1 benno 414: ERRX1(sess, "io_write_buf");
415: goto out;
416: }
417:
418: LOG3(sess, "%s: sent block prologue: %zu blocks of %zu B, "
1.14 ! deraadt 419: "%zu B remainder, %zu B checksum",
! 420: path, p->blksz, p->len, p->rem, p->csum);
1.1 benno 421: rc = 1;
422: out:
423: free(buf);
424: return rc;
425: }
![[BACK]](/icons/back.gif){kind=icon}
Return to blocks.c CVS log ![[TXT]](/icons/text.gif){kind=icon}
![[DIR]](/icons/dir.gif){kind=icon}
Up to [local] / src / usr.bin / rsync