| 1 | /* -*-c-*- |
| 2 | * |
| 3 | * Make an unscrambled copy of a DVD. |
| 4 | * |
| 5 | * (c) 2022 Mark Wooding |
| 6 | */ |
| 7 | |
| 8 | /*----- Licensing notice --------------------------------------------------* |
| 9 | * |
| 10 | * This file is part of the DVD ripping toolset. |
| 11 | * |
| 12 | * DVDrip is free software: you can redistribute it and/or modify it |
| 13 | * under the terms of the GNU General Public License as published by the |
| 14 | * Free Software Foundation; either version 3 of the License, or (at your |
| 15 | * option) any later version. |
| 16 | * |
| 17 | * DVDrip is distributed in the hope that it will be useful, but WITHOUT |
| 18 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 19 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 20 | * for more details. |
| 21 | * |
| 22 | * You should have received a copy of the GNU General Public License |
| 23 | * along with DVDrip. If not, see <https://www.gnu.org/licenses/>. |
| 24 | */ |
| 25 | |
| 26 | /*----- Header files ------------------------------------------------------*/ |
| 27 | |
| 28 | #include "lib.h" |
| 29 | |
| 30 | #ifdef __linux__ |
| 31 | # include <linux/cdrom.h> |
| 32 | #endif |
| 33 | |
| 34 | /*----- Program usage summary ---------------------------------------------*/ |
| 35 | |
| 36 | static void usage(FILE *fp) |
| 37 | { |
| 38 | fprintf(fp, |
| 39 | "usage: %s [-ci] [-B PARAM=VALUE,...] [-R MAP]\n" |
| 40 | "\t[-b OUTMAP] [-r [START]-[END]] DEVICE OUTFILE\n", |
| 41 | prog); |
| 42 | } |
| 43 | |
| 44 | /*----- Random utilities --------------------------------------------------*/ |
| 45 | |
| 46 | #define PRF_HYPHEN 1u |
| 47 | static int parse_range(const char *p, unsigned f, |
| 48 | secaddr *start_out, secaddr *end_out) |
| 49 | /* Parse a range of sectors from the string P. If successful, store |
| 50 | * the specified start sector address in *START_OUT and the end |
| 51 | * address in *END_OUT, and return zero. On failure, return -1; |
| 52 | * *START_OUT and/or *END_OUT are clobbered. |
| 53 | * |
| 54 | * The acceptable syntax depends on the flags. |
| 55 | * |
| 56 | * * The `PRF_HYPHEN' syntax is intended for use on the |
| 57 | * command-line. It accepts `[START]-[END]'; if the start and/or |
| 58 | * end addresses are omitted then *START_OUT and/or *END_OUT are |
| 59 | * left unchanged. |
| 60 | * |
| 61 | * * The default syntax matches what's written to the bad-sector |
| 62 | * output files. It accepts `START END [# COMMENT]'. |
| 63 | */ |
| 64 | { |
| 65 | char *q; |
| 66 | int err, rc; |
| 67 | unsigned long start, end; |
| 68 | |
| 69 | /* Save any existing error code. */ |
| 70 | err = errno; |
| 71 | |
| 72 | /* Parse the start address. */ |
| 73 | if (ISDIGIT(*p)) { |
| 74 | /* We found a digit: this is a good start. Convert the integer, check |
| 75 | * that it's in range, save it. |
| 76 | */ |
| 77 | |
| 78 | start = strtoul(p, &q, 0); |
| 79 | if (errno || start >= SECLIMIT) { rc = -1; goto end; } |
| 80 | *start_out = start; p = q; |
| 81 | } else if (!(f&PRF_HYPHEN)) { |
| 82 | /* No digit. We're parsing the map-file syntax, so this is an error. */ |
| 83 | |
| 84 | rc = -1; goto end; |
| 85 | } else { |
| 86 | /* We're parsing the command-line syntax, so this is OK. Set our |
| 87 | * internal idea of the position for the range check later, but don't |
| 88 | * alter the caller's variables. |
| 89 | */ |
| 90 | |
| 91 | start = 0; |
| 92 | } |
| 93 | |
| 94 | /* Parse the delimiter. */ |
| 95 | if (f&PRF_HYPHEN) { |
| 96 | if (*p != '-') { rc = -1; goto end; } |
| 97 | p++; |
| 98 | } else { |
| 99 | if (!ISSPACE(*p)) { rc = -1; goto end; } |
| 100 | do p++; while (ISSPACE(*p)); |
| 101 | } |
| 102 | |
| 103 | /* Parse the end address. */ |
| 104 | if (ISDIGIT(*p)) { |
| 105 | /* We found a digit. Parse the integer and check that it's strictly |
| 106 | * larger than the start address. |
| 107 | */ |
| 108 | |
| 109 | end = strtoul(p, &q, 0); |
| 110 | if (errno || end > SECLIMIT || end < start) { rc = -1; goto end; } |
| 111 | *end_out = end; p = q; |
| 112 | } else if (!(f&PRF_HYPHEN)) { |
| 113 | /* No digit. We're parsing the file syntax, so this is an error. */ |
| 114 | |
| 115 | rc = -1; goto end; |
| 116 | } |
| 117 | |
| 118 | /* In the file syntax, we're now allowed whitespace, so skip past that. */ |
| 119 | if (!(f&PRF_HYPHEN)) while (ISSPACE(*p)) p++; |
| 120 | |
| 121 | /* Check that there's nothing else. The file syntax allows a trailing |
| 122 | * comment here, but the command-line syntax doesn't. |
| 123 | */ |
| 124 | if (*p && ((f&PRF_HYPHEN) || *p != '#')) { rc = -1; goto end; } |
| 125 | |
| 126 | /* All done! */ |
| 127 | rc = 0; |
| 128 | end: |
| 129 | errno = err; |
| 130 | return (rc); |
| 131 | } |
| 132 | |
| 133 | /*----- A few words about the overall approach ----------------------------* |
| 134 | * |
| 135 | * The objective is to produce a working copy of the input (commercial, |
| 136 | * pressed) DVD disc, only with all of the scrambled video data unscrambled |
| 137 | * so that it can be read without the need for cracking CSS keys, which, in |
| 138 | * the absence of a cooperative drive with access to the key tables in the |
| 139 | * disc lead-in data -- which we /don't/ copy -- is often slow and prone to |
| 140 | * failure. Producing a sector-by-sector image preserves all of the menus |
| 141 | * and special features, and also any other bonus data stored in the |
| 142 | * filesystem for use by computers, such as PDF scripts. DVD images are |
| 143 | * large because DVD video is inefficiently compressed by modern standards, |
| 144 | * but disk space is cheap and the tradeoff seems worthwhile to me. |
| 145 | * |
| 146 | * The approach is, in essence, simple: start at the beginning of the disc, |
| 147 | * reading sectors into a buffer and writing them to the output file, and |
| 148 | * continue until we reach the end. But we must cope with scrambled video |
| 149 | * files. Fortunately, `libdvdread' knows how to deal with these, and will |
| 150 | * tell us where they are on the disc. |
| 151 | * |
| 152 | * Given this information, we build a table of `events', with the sector |
| 153 | * numbers at which they occur. An `event' might be something like `such- |
| 154 | * and-such a video file began' or `such-and-such a file ended'. Chunks of |
| 155 | * disc between events can be read using the same strategy -- either reading |
| 156 | * unscrambled sectors directly from the block device, or decrypting |
| 157 | * scrambled sectors through `libdvdread' -- while at sector boundaries we |
| 158 | * might need to change strategy. |
| 159 | * |
| 160 | * Note that files can /overlap/. The DVD spec says that this can't happen, |
| 161 | * and that the data for video titles is laid out with higher-numbered |
| 162 | * titlesets occupying higher-numbered sectors, but it does anyway. I think |
| 163 | * this is intended to frustrate copiers like `dvdbackup' which try to copy |
| 164 | * the DVD files into a directory on the filesystem. The result is that they |
| 165 | * copy the same sectors into multiple, very large files, and turn an 8 GB |
| 166 | * DVD image into a 60 GB directory. (The reused regions often also contain |
| 167 | * intentionally bad sectors, so you have to wait for the drive to fail the |
| 168 | * same sectors over and over again. This is no fun.) As far as I know, |
| 169 | * files are either disjoint or coincident, but more complex arrangements are |
| 170 | * possible in principle. Also, I guess it's possible that the same sector |
| 171 | * should be decrypted with different keys depending on which titleset we're |
| 172 | * considering it being part of, but (a) DVD CSS keys aren't long enough to |
| 173 | * do this very well, and (b) I'm not aware of this actually being a thing. |
| 174 | * (Indeed, `libdvdcss' indexes keys by start sector, so such a disc probably |
| 175 | * wouldn't play back properly through VLC or `mpv'.) |
| 176 | * |
| 177 | * There's an additional consideration. We want to be able to fill in an |
| 178 | * ouptut image file incrementally, in several runs. A run can be |
| 179 | * interrupted for lots of reasons (e.g., a faster drive might have become |
| 180 | * available; it might be beneficial to switch to a more forgiving drive; it |
| 181 | * might be necessary to stop and clean the disc; the output filesystem might |
| 182 | * have become full; ...). And discs don't always read perfectly: some discs |
| 183 | * are damaged and have areas which can't be read; some discs (I'm looking at |
| 184 | * you, Sony, Disney, Lionsgate, and E-One) have intentional bad sectors, |
| 185 | * presumably specifically to make my life annoying. So we have other events |
| 186 | * which say things like `start writing stuff to the output' or `stop writing |
| 187 | * things to the output'. And we have a rather elaborate algorithm for |
| 188 | * trying to skip past a region of bad blocks, because drives get /really/ |
| 189 | * slow when reading bad sectors. |
| 190 | */ |
| 191 | |
| 192 | /*----- The file and event tables -----------------------------------------*/ |
| 193 | |
| 194 | #define MAXFILES (1 + 2*99 + 1) |
| 195 | /* How many (interesting) files there can be. This counts the |
| 196 | * magical `raw' file which refers to direct disc access, the master |
| 197 | * menu file, and 99 possible menu and titleset pairs. (A titleset |
| 198 | * can be split into 9 parts in order to keep each file below a |
| 199 | * gigabyte in size, but the rules require that the parts together |
| 200 | * form a single contiguous chunk on the disc, in the right order, so |
| 201 | * we treat them as a single file. We check this in `put_title' |
| 202 | * below, just in case some disc somewhere tries to be awkward, but I |
| 203 | * don't have a disc like that in my collection, and I doubt it would |
| 204 | * work very well.) |
| 205 | */ |
| 206 | |
| 207 | struct file { |
| 208 | /* An interesting DVD file. It has a name, encoded as an `ident' |
| 209 | * (see `lib.h'), and start and end sectors. (The `end' here, as |
| 210 | * everywhere in this code, is /exclusive/, so that the file's length |
| 211 | * is simply end - start.) |
| 212 | */ |
| 213 | |
| 214 | ident id; /* file name */ |
| 215 | secaddr start, end; /* start (inclusive) and end |
| 216 | * (exclusive) sector numbers */ |
| 217 | }; |
| 218 | DEFVEC(file_v, struct file); /* a vector of files */ |
| 219 | static file_v filetab = VEC_INIT; /* the file table */ |
| 220 | |
| 221 | enum { |
| 222 | /* Event codes. The ordering of these is important, because we use |
| 223 | * them to tie-break comparisons of events happening at the same |
| 224 | * sector when we sort the event queue. |
| 225 | */ |
| 226 | |
| 227 | EV_STOP, /* stop copying stuff to output */ |
| 228 | EV_BEGIN, /* a (maybe scrambled) file begins */ |
| 229 | EV_END, /* a file ends */ |
| 230 | EV_WRITE /* start copying stuff to output */ |
| 231 | }; |
| 232 | |
| 233 | struct event { |
| 234 | /* An event. */ |
| 235 | |
| 236 | unsigned char ev; /* event code (`EV_...') */ |
| 237 | unsigned char file; /* the file (`EV_BEGIN', `EV_END'); |
| 238 | * index into `filetab' */ |
| 239 | secaddr pos; /* the sector at which it happens */ |
| 240 | }; |
| 241 | DEFVEC(event_v, struct event); /* a vector of events */ |
| 242 | static event_v eventq = VEC_INIT; /* the event queue */ |
| 243 | |
| 244 | static int compare_event(const void *a, const void *b) |
| 245 | /* A `qsort' comparison function for events. Event A sorts earlier |
| 246 | * than event B iff A's sector number is smaller than B's, or A's |
| 247 | * event code is less than B's. |
| 248 | */ |
| 249 | { |
| 250 | const struct event *eva = a, *evb = b; |
| 251 | |
| 252 | /* Primary ordering by position. */ |
| 253 | if (eva->pos < evb->pos) return (-1); |
| 254 | else if (eva->pos > evb->pos) return (+1); |
| 255 | |
| 256 | /* Secondary ordering by event code. */ |
| 257 | if (eva->ev < evb->ev) return (-1); |
| 258 | else if (eva->ev > evb->ev) return (+1); |
| 259 | |
| 260 | /* We currently have a final tie-break on file numbers so that the ordering |
| 261 | * is deterministic, but this is an arbitrary choice that shouldn't be |
| 262 | * relied upon. |
| 263 | */ |
| 264 | if (eva->file < evb->file) return (-1); |
| 265 | else if (eva->file > evb->file) return (+1); |
| 266 | |
| 267 | /* These events are equal. */ |
| 268 | return (0); |
| 269 | } |
| 270 | |
| 271 | #ifdef DEBUG |
| 272 | static void dump_eventq(const char *what) |
| 273 | /* Dump the event queue, labelling the output with WHAT. */ |
| 274 | { |
| 275 | unsigned i; |
| 276 | const struct event *ev; |
| 277 | char fn[MAXFNSZ]; |
| 278 | |
| 279 | printf("\n;; event dump (%s):\n", what); |
| 280 | for (i = 0; i < eventq.n; i++) { |
| 281 | ev = &eventq.v[i]; |
| 282 | switch (ev->ev) { |
| 283 | case EV_BEGIN: |
| 284 | store_filename(fn, filetab.v[ev->file].id); |
| 285 | printf(";; %8"PRIuSEC": begin %s\n", ev->pos, fn); |
| 286 | break; |
| 287 | case EV_END: |
| 288 | store_filename(fn, filetab.v[ev->file].id); |
| 289 | printf(";; %8"PRIuSEC": end %s\n", ev->pos, fn); |
| 290 | break; |
| 291 | case EV_WRITE: |
| 292 | printf(";; %8"PRIuSEC": write\n", ev->pos); |
| 293 | break; |
| 294 | case EV_STOP: |
| 295 | printf(";; %8"PRIuSEC": stop\n", ev->pos); |
| 296 | break; |
| 297 | default: |
| 298 | printf(";; %8"PRIuSEC": ?%u\n", ev->pos, ev->ev); |
| 299 | break; |
| 300 | } |
| 301 | } |
| 302 | } |
| 303 | #endif |
| 304 | |
| 305 | typedef uint_least32_t bits; |
| 306 | static bits live[(MAXFILES + 31)/32]; |
| 307 | /* A bitmap which keeps track of which files are currently `active', |
| 308 | * i.e., that contain the sector we're currently thinking about. We |
| 309 | * set and clear these bits as we encounter `EV_BEGIN' and `EV_END' |
| 310 | * events. |
| 311 | */ |
| 312 | |
| 313 | static inline int livep(unsigned i) |
| 314 | /* Return whether file I is active. */ |
| 315 | { return (live[i/32]&((bits)1 << (i%32))); } |
| 316 | |
| 317 | static inline void set_live(unsigned i) |
| 318 | /* Note that we've seen the start of file I. */ |
| 319 | { live[i/32] |= (bits)1 << (i%32); } |
| 320 | |
| 321 | static inline void clear_live(unsigned i) |
| 322 | /* Note that we've seen the end of file I. */ |
| 323 | { live[i/32] &= ~((bits)1 << (i%32)); } |
| 324 | |
| 325 | static inline int least_live(void) |
| 326 | /* Return the smallest index for any active file. This is going to |
| 327 | * be the file that we ask `libdvdread' to unscramble for us. This |
| 328 | * is important: the imaginary `raw' file that represents the entire |
| 329 | * block device has the highest index, and we want any actual video |
| 330 | * file to be used in preference so that we unscramble the data. |
| 331 | */ |
| 332 | { |
| 333 | unsigned i, n = (filetab.n + 32)/32; |
| 334 | bits b; |
| 335 | |
| 336 | /* First part: find the first nonzero word in the `live' table. */ |
| 337 | for (i = 0; i < n; i++) { b = live[i]; if (b) goto found; } |
| 338 | return (-1); |
| 339 | |
| 340 | found: |
| 341 | /* Second part: identify which bit in this word is nonzero. First, see if |
| 342 | * the bottom 16 bits are clear: if so, shift down and add 16 to the |
| 343 | * total. Now we know that the first set bit is indeed in the low 16 |
| 344 | * bits, so see whether the low 8 bits are clear, and so on. |
| 345 | */ |
| 346 | i *= 32; |
| 347 | if (!(b&0x0000ffff)) { b >>= 16; i += 16; } |
| 348 | if (!(b&0x000000ff)) { b >>= 8; i += 8; } |
| 349 | if (!(b&0x0000000f)) { b >>= 4; i += 4; } |
| 350 | if (!(b&0x00000003)) { b >>= 2; i += 2; } |
| 351 | if (!(b&0x00000001)) { b >>= 1; i += 1; } |
| 352 | assert(b&1); |
| 353 | |
| 354 | /* Done. */ |
| 355 | return (i); |
| 356 | } |
| 357 | |
| 358 | static void put_event(unsigned evtype, unsigned file, secaddr pos) |
| 359 | /* Add an event to the queue, with type EVTYPE, for the given FILE, |
| 360 | * and at sector POS. You can add events in any order because we'll |
| 361 | * sort them later. For `EV_WRITE' and `EV_STOP' events, the FILE |
| 362 | * doesn't matter: use zero for concreteness. |
| 363 | */ |
| 364 | { |
| 365 | struct event *ev; |
| 366 | |
| 367 | VEC_PUSH(ev, &eventq); |
| 368 | ev->ev = evtype; ev->file = file; ev->pos = pos; |
| 369 | } |
| 370 | |
| 371 | static void put_file(ident id, secaddr start, secaddr end) |
| 372 | /* Add a (VOB) file to the file table and event queue, with ident ID, |
| 373 | * starting at sector START and ending just before sector END. |
| 374 | */ |
| 375 | { |
| 376 | struct file *f; |
| 377 | size_t i; |
| 378 | |
| 379 | VEC_PUSH(f, &filetab); i = f - filetab.v; |
| 380 | f->id = id; f->start = start; f->end = end; |
| 381 | put_event(EV_BEGIN, i, start); |
| 382 | put_event(EV_END, i, end); |
| 383 | } |
| 384 | |
| 385 | static void put_menu(dvd_reader_t *dvd, unsigned title) |
| 386 | /* Add the menu file for the given TITLE number to the file table and |
| 387 | * event queue; use the reader DVD to find out which sectors it |
| 388 | * occupies, if it even exists. |
| 389 | */ |
| 390 | { |
| 391 | ident id = mkident(VOB, title, 0); |
| 392 | char fn[MAXFNSZ]; |
| 393 | secaddr start, len; |
| 394 | |
| 395 | /* Find out where the file is. */ |
| 396 | store_filename(fn, id); |
| 397 | start = UDFFindFile(dvd, fn, &len); if (!start) return; |
| 398 | |
| 399 | #ifdef DEBUG |
| 400 | /* Print out what we've discovered. */ |
| 401 | printf(";; %8"PRIuSEC" .. %-8"PRIuSEC": %s\n", |
| 402 | start, start + SECTORS(len), fn); |
| 403 | #endif |
| 404 | |
| 405 | /* Register the file and boundary events. */ |
| 406 | put_file(id, start, start + SECTORS(len)); |
| 407 | } |
| 408 | |
| 409 | static void put_title(dvd_reader_t *dvd, unsigned title) |
| 410 | /* Add the titleset file for the given TITLE number to the file table |
| 411 | * and event queue; use the reader DVD to find out which sectors it |
| 412 | * occupies, if it even exists. |
| 413 | */ |
| 414 | { |
| 415 | char fn[MAXFNSZ]; |
| 416 | secaddr start[9], len[9]; |
| 417 | unsigned i, npart; |
| 418 | |
| 419 | /* First step: find out where all of the parts of the titleset are. I'm |
| 420 | * assuming that there aren't gaps in the numbering. |
| 421 | */ |
| 422 | for (i = 0; i < 9; i++) { |
| 423 | store_filename(fn, mkident(VOB, title, i + 1)); |
| 424 | start[i] = UDFFindFile(dvd, fn, &len[i]); if (!start[i]) break; |
| 425 | } |
| 426 | npart = i; if (!npart) return; |
| 427 | |
| 428 | #ifdef DEBUG |
| 429 | /* Print out what we've discovered. */ |
| 430 | for (i = 0; i < npart; i++) { |
| 431 | store_filename(fn, mkident(VOB, title, i + 1)); |
| 432 | printf(";; %8"PRIuSEC" .. %-8"PRIuSEC": %s\n", |
| 433 | start[i], start[i] + SECTORS(len[i]), fn); |
| 434 | } |
| 435 | #endif |
| 436 | |
| 437 | /* Second step: check that the parts all butt up against each other in the |
| 438 | * correct order. For this to work, the lengths, which are expressed in |
| 439 | * /bytes/ by `UDFFindFile', of all but the last part must be a whole |
| 440 | * number of sectors. |
| 441 | */ |
| 442 | if (npart > 1) |
| 443 | for (i = 0; i < npart - 1; i++) { |
| 444 | if (len[i]%SECTORSZ) |
| 445 | bail("title %u part %u length = %"PRIuSEC" not a multiple of %d", |
| 446 | title, i, len[i], SECTORSZ); |
| 447 | if (start[i] + len[i]/SECTORSZ != start[i + 1]) |
| 448 | bail |
| 449 | ("title %u part %u end = %"PRIuSEC" /= part %u start = %"PRIuSEC"", |
| 450 | title, i, start[i] + len[i]/SECTORSZ, i + 1, start[i + 1]); |
| 451 | } |
| 452 | |
| 453 | /* All good: register a single file and its boundary events. */ |
| 454 | put_file(mkident(VOB, title, 1), |
| 455 | start[0], start[npart - 1] + SECTORS(len[npart - 1])); |
| 456 | } |
| 457 | |
| 458 | /*----- Moving average machinery ------------------------------------------* |
| 459 | * |
| 460 | * We're using an exponential moving average with a weighting factor of α |
| 461 | * (`alpha', above); larger values are more sensitive to recent changes. If |
| 462 | * the old average was v_1, and the measurement in the current interval is x, |
| 463 | * then the new average after this interval is |
| 464 | * |
| 465 | * v = α x + (1 − α) v_1 . |
| 466 | * |
| 467 | * Write β = 1 − α; so |
| 468 | * |
| 469 | * v = α x + β v_1 . |
| 470 | * |
| 471 | * Let x_0 = x, let x_1 be the measurement from the previous interval, and, |
| 472 | * in general, let x_i be the measurement from i intervals ago. Then another |
| 473 | * way to write the above would be |
| 474 | * |
| 475 | * v = α (x_0 + β x_1 + ⋯ + β^i x_i + ⋯) . |
| 476 | * |
| 477 | * Alas, our time intervals are not regular. Suppose that we get our next |
| 478 | * measurement after a gap of t intervals, for some integer t. We can |
| 479 | * compensate approximately by pretending that all of the missed intervals -- |
| 480 | * and our new one -- had the same mean rate. Then we'd have calculated |
| 481 | * |
| 482 | * v = α (x + β x + ⋯ + β^{t−1} x) + β^t v_1 |
| 483 | * |
| 484 | * 1 − β^t |
| 485 | * = α x ------- + β^t v_1 |
| 486 | * 1 − β |
| 487 | * |
| 488 | * = x (1 − β^t) + β^t v_1 (since α = 1 − β) |
| 489 | * |
| 490 | * = x + β^t (v_1 − x) . |
| 491 | * |
| 492 | * Does this work in general? It's clearly correct in the case t = 1. |
| 493 | * |
| 494 | * Suppose the old average was v_2, and that over a period of t intervals |
| 495 | * (where t is not necessarily an integer) we measured a mean rate of x, and |
| 496 | * then after u intervals we measured a mean rate of x /again/. Then we'd |
| 497 | * firstly determine |
| 498 | * |
| 499 | * v_1 = x + β^t (v_2 − x) |
| 500 | * |
| 501 | * and then |
| 502 | * |
| 503 | * v = x + β^u (v_1 − x) |
| 504 | * |
| 505 | * = x + β^u (x + β^t (v_2 − x) − x) |
| 506 | * |
| 507 | * = x + β^{t+u} (v_2 − x) , |
| 508 | * |
| 509 | * which is exactly what we'd have done if we'd calculated the same mean rate |
| 510 | * over the combined span of t + u intervals. |
| 511 | * |
| 512 | * One final wrinkle, in case that wasn't enough. There's a problem with the |
| 513 | * initial setup of an exponential moving average. Apparently |
| 514 | * (https://en.wikipedia.org/wiki/Moving_average#Exponential_moving_average) |
| 515 | * explains that we can do this better by calculating the average after k |
| 516 | * intervals as |
| 517 | * |
| 518 | * x_0 + β x_1 + β^2 x_2 + ⋯ + β^{k−1} x_{k−1} |
| 519 | * v′ = ------------------------------------------- . |
| 520 | * 1 + β + β^2 + ⋯ + β^{k−1} |
| 521 | * |
| 522 | * The numerator is our existing v/α; the denominator is (1 − β^k)/α; the |
| 523 | * factors of α cancel, and we find that v′ = v/(1 − β^k). This still holds |
| 524 | * in our situation, where k may not be an integer. |
| 525 | * |
| 526 | * To apply all of this: |
| 527 | * |
| 528 | * * we maintain the moving average v in `avg'; |
| 529 | * |
| 530 | * * we maintain the total β^k in `corr'; and |
| 531 | * |
| 532 | * * we compute v′ = v/(1 − β^k) on demand up in `render_perfstats'. |
| 533 | */ |
| 534 | |
| 535 | struct avg { |
| 536 | double avg, corr; |
| 537 | }; |
| 538 | #define AVG_INIT { 0.0, 1.0 } |
| 539 | |
| 540 | static double alpha = 0.1; /* weighting factor for average */ |
| 541 | |
| 542 | static void update_avg(struct avg *a, double t, double n) |
| 543 | { |
| 544 | double rate = n/t, beta_t = pow(1 - alpha, t); |
| 545 | |
| 546 | a->avg = rate + beta_t*(a->avg - rate); |
| 547 | a->corr *= beta_t; |
| 548 | } |
| 549 | |
| 550 | static inline double current_avg(const struct avg *a) |
| 551 | { return (a->avg/(1 - a->corr)); } |
| 552 | |
| 553 | /*----- The nonlinear progress model --------------------------------------*/ |
| 554 | |
| 555 | /* The recorded portion of a single-layer DVD (i.e., DVD-5) can hold 2298496 |
| 556 | * sectors of user data. This is preceded by 0x30000 = 196608 sectors of |
| 557 | * lead-in information, for a totoal of 2495104 sectors. |
| 558 | * |
| 559 | * The readable portion of a disc is an annulus with respective internal and |
| 560 | * external diameters of 44 mm and 117 mm. This annulus has an area of |
| 561 | * 9230.8 mm^2, so DVD has a storage density of about 270.3 sectors/mm^2. If |
| 562 | * the interior of the annulus were used for data storage rather than leaving |
| 563 | * a hole for a spindle and a clamping area, then it would be 10751 mm^2 and |
| 564 | * could store 2906107 sectors. (That means that the portion of the disc |
| 565 | * that's actually used to make it spin could have stored an additional |
| 566 | * 411003 sectors.) |
| 567 | * |
| 568 | * Sectors aren't stored on the surface willy-nilly, but arranged into a |
| 569 | * single archimedean spiral; bits are stored along this spiral at a |
| 570 | * more-or-less constant pitch. We are therefore led into an investigation |
| 571 | * of the arc-length of archimedean spirals. |
| 572 | * |
| 573 | * It's best to start with the polar equation of the spiral, which is simply |
| 574 | * |
| 575 | * r = k θ |
| 576 | * |
| 577 | * for a given constant k. The arc length of a curve expressed using polar |
| 578 | * coordinates is given by |
| 579 | * |
| 580 | * s = ∫ √(r^2 + (dr/dθ)^2) dθ |
| 581 | * |
| 582 | * = ∫ √(k^2 θ^2 + k^2) dθ |
| 583 | * |
| 584 | * = k ∫ √(1 + θ^2) dθ |
| 585 | * |
| 586 | * k |
| 587 | * = - [ θ √(1 + θ^2) + log(θ + √(1 + θ^2)) ] - s_0. |
| 588 | * 2 |
| 589 | * |
| 590 | * We're assuming that the sectors are spaced out at a constant linear |
| 591 | * density along the spiral. We don't know the units for s, but there's some |
| 592 | * constant L such that A = s/L; so |
| 593 | * |
| 594 | * k |
| 595 | * A = --- [ θ √(1 + θ^2) + log(θ + √(1 + θ^2)) ] - A_0 |
| 596 | * 2 L |
| 597 | * |
| 598 | * for some suitable constant A_0. |
| 599 | * |
| 600 | * Finally, we're assuming that the disc is spinning with some approximately |
| 601 | * constant angular velocity ω, so θ = ω t, giving |
| 602 | * |
| 603 | * k |
| 604 | * A = --- [ ω t √(1 + ω^2 t^2) + log(ω + √(1 + ω^2 t^2)) ] + A_0 . |
| 605 | * 2 L |
| 606 | * |
| 607 | * We can calculate approximate values for k/(2 L) and A_0. As stated above, |
| 608 | * the track pitch is about 0.75 µm; our inside and outside diameters of |
| 609 | * 44 mm and 117 mm correspond to angles of 184306 and 490088 radians |
| 610 | * respectively. Feeding those into the above equation for s gives arc |
| 611 | * lengths of 16984492558 and 120093346360 respectively, in unknown units. |
| 612 | * The difference is 103108853802, which should correspond to 2495104 |
| 613 | * sectors, giving us 41324 arc-length units per sector. As a cross-check, |
| 614 | * the arc length corresponding to the inside diameter yields 411003 sectors, |
| 615 | * which is the same as we calculated above. This will be our A_0 |
| 616 | */ |
| 617 | |
| 618 | #ifdef unusef |
| 619 | static double archimedes_arclen(double t) |
| 620 | /* Given an angle T, return the arc length of the canonical |
| 621 | * archimedean spiral r = θ, from θ = 0 up to θ = T. |
| 622 | */ |
| 623 | { |
| 624 | double u; |
| 625 | |
| 626 | u = sqrt(1 + t*t); |
| 627 | return (t*u + log(t + u))/2; |
| 628 | } |
| 629 | #endif |
| 630 | |
| 631 | static double inv_archimedes_arclen(double s) |
| 632 | /* Given an arc length S, return the angle T such that the arc length |
| 633 | * of the canonical archimedean spiral r = θ, from θ = 0 up to θ = T, |
| 634 | * is equal to S. |
| 635 | */ |
| 636 | { |
| 637 | /* There is no closed-form solution, so we're going to invert the arc- |
| 638 | * length formula above numerically, using the Newton--Raphson method. |
| 639 | * |
| 640 | * Given an incorrect guess x_0 of a zero of some function f, we refine the |
| 641 | * guess by approximating f by its tangent at the point (x_0, f(x_0)). |
| 642 | * This will be a line with an equation like |
| 643 | * |
| 644 | * y = f'(x_0) x + c . |
| 645 | * |
| 646 | * We know that y = f(x_0) when x = x_0, so we can calculate |
| 647 | * |
| 648 | * c = f(x_0) - f'(x_0) x_0 . |
| 649 | * |
| 650 | * This will be zero when |
| 651 | * |
| 652 | * y = f'(x_0) x + f(x_0) - f'(x_0) x_0 = 0 |
| 653 | * |
| 654 | * hwnce |
| 655 | * |
| 656 | * f'(x_0) x_0) - f(x_0) f(x_0) |
| 657 | * x = --------------------- = x_0 - ------- . |
| 658 | * f'(x_0) f'(x_0) |
| 659 | */ |
| 660 | |
| 661 | double t, ss, u, e; |
| 662 | |
| 663 | /* We need to choose an initial estimate. This seems to work well in |
| 664 | * practice. |
| 665 | */ |
| 666 | t = 1.5*sqrt(s); |
| 667 | |
| 668 | for (;;) { |
| 669 | /* Compute s' = f(t). We open-code this calculation because the |
| 670 | * intermediate value √(1 + t^2) is also the gradient. |
| 671 | */ |
| 672 | u = sqrt(1 + t*t); |
| 673 | ss = (t*u + log(t + u))/2; |
| 674 | |
| 675 | /* Determine the error in f(t). We don't actually need much precision |
| 676 | * here, but 2 ulp seems achievable in practice with minimal cost: the |
| 677 | * usually sequence converges after only five iterations. |
| 678 | */ |
| 679 | e = fabs(s/ss - 1); |
| 680 | if (e <= 2*DBL_EPSILON) return (t); |
| 681 | |
| 682 | /* Not good enough. Refine the guess and go around again. */ |
| 683 | t -= (ss - s)/u; |
| 684 | } |
| 685 | } |
| 686 | |
| 687 | static double sectors_to_angle(secaddr base, secaddr low, secaddr high) |
| 688 | /* Return the angle, in radians, subtended by the range LOW up to |
| 689 | * HIGH of user sector addresses, given the physical sector address |
| 690 | * BASE of the first user-data sectors. |
| 691 | */ |
| 692 | { |
| 693 | #define A0 411003.262489 |
| 694 | #define K 41324.4713654 |
| 695 | |
| 696 | return (inv_archimedes_arclen(K*(A0 + base + high)) - |
| 697 | inv_archimedes_arclen(K*(A0 + base + low))); |
| 698 | |
| 699 | #undef A0 |
| 700 | #undef K |
| 701 | } |
| 702 | |
| 703 | enum { |
| 704 | FLAT, /* not actually a real DVD */ |
| 705 | SINGLE, /* disc with only one layer */ |
| 706 | PTP, /* two layers, parallel track path */ |
| 707 | OTP /* two layers, opposite track path */ |
| 708 | }; |
| 709 | |
| 710 | struct geometry { |
| 711 | unsigned shape; /* one of the four codes above */ |
| 712 | secaddr start0, start1; /* initial physical sector */ |
| 713 | secaddr midpoint; /* sector address of layer switch */ |
| 714 | }; |
| 715 | |
| 716 | #define GF_BLKDEV 1u |
| 717 | static void setup_geometry(struct geometry *g, int fd, unsigned f, |
| 718 | secaddr sz) |
| 719 | /* Initialize G with information about the disc structure. FD is a |
| 720 | * file descriptor for the device; SZ is the size of the disc in |
| 721 | * sectors. If `GF_BLKDEV' is clear in F then assume that FD refers |
| 722 | * to a regular file; G is populated with a `FLAT' performance model. |
| 723 | * If `GF_BLKDEV' is set, then FD refers to a block device, so try to |
| 724 | * retreive detailed structure information from the drive. |
| 725 | */ |
| 726 | { |
| 727 | #ifdef __linux__ |
| 728 | dvd_struct ds; |
| 729 | const struct dvd_layer *ly; |
| 730 | #endif |
| 731 | secaddr t; |
| 732 | |
| 733 | #define LAYER_LIMIT 2298496 /* maximum (user) sectors on layer */ |
| 734 | #define DVDROM_OFFSET 0x30000 /* usual initial physical sector */ |
| 735 | |
| 736 | if (!(f&GF_BLKDEV)) { |
| 737 | /* We're reading from a regular file. Assume that progress will be |
| 738 | * linear. |
| 739 | */ |
| 740 | |
| 741 | g->shape = FLAT; |
| 742 | g->midpoint = SECLIMIT; |
| 743 | return; |
| 744 | } |
| 745 | |
| 746 | #ifdef __linux__ |
| 747 | /* We have Linux and its DVD ioctl(2) calls. Interrogate the disc to |
| 748 | * discover its structure. |
| 749 | */ |
| 750 | |
| 751 | ds.type = DVD_STRUCT_PHYSICAL; |
| 752 | ds.physical.layer_num = 0; |
| 753 | if (ioctl(fd, DVD_READ_STRUCT, &ds)) { |
| 754 | moan_syserr(errno, "failed to read physical disc structure"); |
| 755 | goto guess_structure; |
| 756 | } |
| 757 | ly = &ds.physical.layer[0]; |
| 758 | switch (ly->nlayers) { |
| 759 | case 0: |
| 760 | g->shape = SINGLE; |
| 761 | g->start0 = g->start1 = 0; |
| 762 | g->midpoint = SECLIMIT; |
| 763 | break; |
| 764 | case 1: |
| 765 | g->start0 = ly->start_sector; |
| 766 | if (ly->track_path) { |
| 767 | g->shape = OTP; |
| 768 | g->start1 = 0; |
| 769 | g->midpoint = ly->end_sector_l0 - ly->start_sector + 1; |
| 770 | } else { |
| 771 | g->shape = PTP; |
| 772 | g->midpoint = ly->end_sector - ly->start_sector + 1; |
| 773 | ds.physical.layer_num = 1; |
| 774 | if (ioctl(fd, DVD_READ_STRUCT, &ds)) { |
| 775 | moan_syserr(errno, "failed to read layer 1 physical structure"); |
| 776 | goto guess_structure; |
| 777 | } |
| 778 | g->start1 = ly->start_sector; |
| 779 | } |
| 780 | break; |
| 781 | default: |
| 782 | moan("unexpected layer count %d", ly->nlayers + 1); |
| 783 | goto guess_structure; |
| 784 | } |
| 785 | return; |
| 786 | guess_structure: |
| 787 | #endif |
| 788 | |
| 789 | /* Either we don't have Linux, or we found something confusing. Let's try |
| 790 | * to guess at what's going on. |
| 791 | * |
| 792 | * If the volume size is small enough to fit on a single layer then assume |
| 793 | * that's what's happened; otherwise assume opposite track path with a cut |
| 794 | * at the midpoint, rounded up to an ECC block (16 sectors). |
| 795 | */ |
| 796 | g->start0 = DVDROM_OFFSET; g->start1 = 0; |
| 797 | if (sz <= LAYER_LIMIT) { |
| 798 | g->shape = SINGLE; |
| 799 | g->midpoint = SECLIMIT; |
| 800 | } else { |
| 801 | g->shape = OTP; |
| 802 | t = (sz + DVDROM_OFFSET)/2; |
| 803 | t += 15; t &= -16; |
| 804 | t -= DVDROM_OFFSET; |
| 805 | g->midpoint = t; |
| 806 | } |
| 807 | |
| 808 | #undef LAYER_LIMIT |
| 809 | #undef DVDROM_OFFSET |
| 810 | } |
| 811 | |
| 812 | static double linear_progress(const struct geometry *g, |
| 813 | secaddr a0, secaddr a1) |
| 814 | /* Convert the sector range from A0 to A1 into a progress measurement |
| 815 | * which is, by intention, approximately linearly related to time, |
| 816 | * given a geometry description G. |
| 817 | */ |
| 818 | { |
| 819 | double theta = 0.0; |
| 820 | |
| 821 | switch (g->shape) { |
| 822 | case FLAT: |
| 823 | theta = a1 - a0; |
| 824 | break; |
| 825 | case SINGLE: |
| 826 | theta = sectors_to_angle(g->start0, a0, a1); |
| 827 | break; |
| 828 | case PTP: |
| 829 | if (a0 < g->midpoint) |
| 830 | theta += sectors_to_angle(g->start0, |
| 831 | a0, a1 < g->midpoint ? a1 : g->midpoint); |
| 832 | if (a1 > g->midpoint) |
| 833 | theta += sectors_to_angle(g->start1, |
| 834 | a0 > g->midpoint ? a0 : g->midpoint, a1); |
| 835 | break; |
| 836 | case OTP: |
| 837 | if (a0 < g->midpoint) |
| 838 | theta += sectors_to_angle(g->start0, |
| 839 | a0, a1 < g->midpoint ? a1 : g->midpoint); |
| 840 | if (a1 > g->midpoint) |
| 841 | theta += sectors_to_angle(g->start0, |
| 842 | 2*g->midpoint - a1, |
| 843 | a0 > g->midpoint ? |
| 844 | 2*g->midpoint - a0 : g->midpoint); |
| 845 | break; |
| 846 | default: |
| 847 | abort(); |
| 848 | } |
| 849 | return (theta); |
| 850 | } |
| 851 | |
| 852 | /*----- Common variables used by the copying machinery --------------------*/ |
| 853 | |
| 854 | /* General reading state. */ |
| 855 | static dvd_reader_t *dvd; /* `libdvdread' state for device */ |
| 856 | static int dvdfd = -1, outfd = -1; /* input device and output image */ |
| 857 | static struct file *file; /* currently active file */ |
| 858 | static dvd_file_t *vob; /* current `.VOB' file, or null */ |
| 859 | static const char *mapfile; static FILE *mapfp; /* skipped regions map */ |
| 860 | static const char *errfile; static FILE *errfp; /* bad-sector log */ |
| 861 | static secaddr limit; /* upper bound on sectors */ |
| 862 | |
| 863 | static secaddr bad_start; /* start of current bad region */ |
| 864 | static unsigned retry, max_retries = 4; /* retry state */ |
| 865 | |
| 866 | /*----- Progress reporting ------------------------------------------------*/ |
| 867 | |
| 868 | static secaddr nsectors, ndone; /* number of sectors done/to do */ |
| 869 | static double total_linear, done_linear; /* linear progress tracking */ |
| 870 | static secaddr last_pos; /* position last time we updated */ |
| 871 | static struct timeval last_time; /* time last time we updated */ |
| 872 | static struct geometry geom; /* disc geometry for progress */ |
| 873 | static struct avg avg_rate = AVG_INIT, avg_linear = AVG_INIT; |
| 874 | static int bad_err; /* most recent error code */ |
| 875 | static FILE *progressfp; /* file on which to trace progress data */ |
| 876 | |
| 877 | static const char throbber[] = "|<-<|>->"; /* throbber pattern */ |
| 878 | static unsigned throbix = 0; /* current throbber index */ |
| 879 | |
| 880 | static struct progress_item /* stock progress items */ |
| 881 | copy_progress, disc_progress, |
| 882 | file_progress, badblock_progress; |
| 883 | |
| 884 | static double scale_bytes(double n, const char **unit_out) |
| 885 | /* Determine a human-readable representation for N bytes. Divide N |
| 886 | * by some power of 1024, and store in *UNIT_OUT a string |
| 887 | * representing the conventional unit-prefix for that power of 1024. |
| 888 | */ |
| 889 | { |
| 890 | const char *unit = ""; |
| 891 | |
| 892 | if (n > 1600) { n /= 1024; unit = "k"; } |
| 893 | if (n > 1600) { n /= 1024; unit = "M"; } |
| 894 | if (n > 1600) { n /= 1024; unit = "G"; } |
| 895 | if (n > 1600) { n /= 1024; unit = "T"; } |
| 896 | *unit_out = unit; return (n); |
| 897 | } |
| 898 | |
| 899 | #define TIMESTRMAX 16 /* maximum length of a duration string */ |
| 900 | static char *fmttime(unsigned long t, char *buf) |
| 901 | /* Format a count T of seconds. Write a suitable string to BUF, |
| 902 | * which will be no longer than `TIMESTRMAX' bytes including the |
| 903 | * terminating zero. Return BUF. |
| 904 | */ |
| 905 | { |
| 906 | if (t < 60) sprintf(buf, "%ld s", t); |
| 907 | else if (t < 3600) sprintf(buf, "%ld:%02ld", t/60, t%60); |
| 908 | else sprintf(buf, "%ld:%02ld:%02ld", t/3600, (t/60)%60, t%60); |
| 909 | return (buf); |
| 910 | } |
| 911 | |
| 912 | static void render_perfstats(struct progress_render_state *render) |
| 913 | /* Add performance statistics to RENDER. |
| 914 | * |
| 915 | * Specifically: the average transfer rate, and the estimated time to |
| 916 | * completion. (See `update_progress' for how the average |
| 917 | * computation works.) |
| 918 | */ |
| 919 | { |
| 920 | int eta; |
| 921 | char timebuf[TIMESTRMAX]; |
| 922 | double rate, linrate; |
| 923 | const char *unit; |
| 924 | |
| 925 | /* If there's no average computed yet, then use some placeholder values. */ |
| 926 | rate = current_avg(&avg_rate); |
| 927 | linrate = current_avg(&avg_linear); |
| 928 | eta = (int)((total_linear - done_linear)/linrate + 0.5); |
| 929 | |
| 930 | /* Write out the statistics. */ |
| 931 | rate = scale_bytes(rate*SECTORSZ, &unit); |
| 932 | progress_putright(render, "ETA %s ", |
| 933 | avg_linear.avg ? fmttime(eta, timebuf) : "???"); |
| 934 | progress_putright(render, "%.1f %sB/s, ", rate, unit); |
| 935 | } |
| 936 | |
| 937 | static void render_copy_progress(struct progress_item *item, |
| 938 | struct progress_render_state *render) |
| 939 | /* Render the progress for the copy, i.e., the number of sectors |
| 940 | * copied against the total number to be copied. |
| 941 | */ |
| 942 | { |
| 943 | double frac = (double)ndone/nsectors; |
| 944 | |
| 945 | progress_putleft(render, " %c copied %.1f%%", |
| 946 | throbber[throbix], 100.0*frac); |
| 947 | render_perfstats(render); |
| 948 | progress_putleft(render, " (%"PRIuSEC" of %"PRIuSEC")", ndone, nsectors); |
| 949 | |
| 950 | progress_showbar(render, frac); |
| 951 | } |
| 952 | |
| 953 | static void render_disc_progress(struct progress_item *item, |
| 954 | struct progress_render_state *render) |
| 955 | /* Render the progress for the disc, i.e., the current position |
| 956 | * against the total number of sectors on the disc. |
| 957 | */ |
| 958 | { |
| 959 | double frac = (double)last_pos/limit; |
| 960 | |
| 961 | progress_putleft(render, " disc %.1f%% (%"PRIuSEC" of %"PRIuSEC")", |
| 962 | 100.0*frac, last_pos, limit); |
| 963 | progress_showbar(render, frac); |
| 964 | } |
| 965 | |
| 966 | static void render_file_progress(struct progress_item *item, |
| 967 | struct progress_render_state *render) |
| 968 | /* Render the progress for the current file, i.e., the current |
| 969 | * position within the file against the file size. |
| 970 | */ |
| 971 | { |
| 972 | secaddr off = last_pos - file->start, len = file->end - file->start; |
| 973 | char fn[MAXFNSZ]; |
| 974 | double frac; |
| 975 | |
| 976 | store_filename(fn, file->id); |
| 977 | frac = (double)off/len; |
| 978 | progress_putleft(render, " `%s' %.1f%% (%"PRIuSEC" of %"PRIuSEC")", |
| 979 | fn, 100.0*frac, off, len); |
| 980 | progress_showbar(render, frac); |
| 981 | } |
| 982 | |
| 983 | static void render_badblock_progress(struct progress_item *item, |
| 984 | struct progress_render_state *render) |
| 985 | /* Render a notice about the progress through the current bad block |
| 986 | * region. |
| 987 | */ |
| 988 | { |
| 989 | secaddr n = last_pos - bad_start; |
| 990 | int bg; |
| 991 | |
| 992 | if (!n) { |
| 993 | progress_putleft(render, " Retrying bad sector %"PRIuSEC"", bad_start); |
| 994 | progress_putright(render, "attempt %u/%u ", retry + 1, max_retries); |
| 995 | bg = 4; |
| 996 | } else { |
| 997 | progress_putleft(render, " Found %"PRIuSEC" bad %s", |
| 998 | n, n == 1 ? "sector" : "sectors"); |
| 999 | progress_putright(render, "%"PRIuSEC" .. %"PRIuSEC" ", |
| 1000 | bad_start, last_pos); |
| 1001 | bg = 1; |
| 1002 | } |
| 1003 | if (bad_err && bad_err != EIO) |
| 1004 | progress_putleft(render, " (%s)", strerror(bad_err)); |
| 1005 | progress_shownotice(render, bg, 7); |
| 1006 | } |
| 1007 | |
| 1008 | static void update_progress(secaddr pos) |
| 1009 | /* Recompute the data displayed by the progress renderer functions |
| 1010 | * above, based on the new current sector POS. |
| 1011 | */ |
| 1012 | { |
| 1013 | struct timeval now; |
| 1014 | double t, delta_r; |
| 1015 | |
| 1016 | /* Find the current time and the delta since the last time we updated. |
| 1017 | * This will be the length of the current interval. |
| 1018 | */ |
| 1019 | gettimeofday(&now, 0); t = tvdiff(&last_time, &now); |
| 1020 | |
| 1021 | /* If no time at all has passed (unlikely!) then skip the rate |
| 1022 | * calculation. (The moving average wouldn't be affected anyway.) |
| 1023 | */ |
| 1024 | if (t) { |
| 1025 | /* Update the moving average and the correction term, and start the next |
| 1026 | * interval. |
| 1027 | */ |
| 1028 | |
| 1029 | delta_r = linear_progress(&geom, last_pos, pos); |
| 1030 | update_avg(&avg_rate, t, pos - last_pos); |
| 1031 | update_avg(&avg_linear, t, delta_r); |
| 1032 | ndone += pos - last_pos; done_linear += delta_r; |
| 1033 | last_time = now; last_pos = pos; |
| 1034 | } |
| 1035 | |
| 1036 | /* Trace progress state if requested. */ |
| 1037 | if (progressfp) { |
| 1038 | fprintf(progressfp, "%10ju.%06ld %"PRIuSEC" %f %f\n", |
| 1039 | (uintmax_t)now.tv_sec, now.tv_usec, |
| 1040 | ndone, done_linear, |
| 1041 | (total_linear - done_linear)/current_avg(&avg_linear)); |
| 1042 | check_write(progressfp, "progress trace file"); |
| 1043 | } |
| 1044 | |
| 1045 | /* Advance the throbber character. */ |
| 1046 | throbix++; if (!throbber[throbix]) throbix = 0; |
| 1047 | } |
| 1048 | |
| 1049 | static void report_progress(secaddr pos) |
| 1050 | /* Update the progress variables (as `update_progress') and redraw |
| 1051 | * the progress display. |
| 1052 | */ |
| 1053 | { update_progress(pos); progress_update(&progress); } |
| 1054 | |
| 1055 | /*----- Basic disc I/O ----------------------------------------------------*/ |
| 1056 | |
| 1057 | struct badblock { secaddr start, end; }; |
| 1058 | DEFVEC(badblock_v, struct badblock); |
| 1059 | static badblock_v badblocks = VEC_INIT; |
| 1060 | /* This is a list of /fake/ bad-block ranges, used to test the |
| 1061 | * recovery algorithm. It's a rule that the ranges in this table |
| 1062 | * mustn't overlap -- though it's OK if they abut. |
| 1063 | */ |
| 1064 | |
| 1065 | static int compare_badblock(const void *a, const void *b) |
| 1066 | /* A `qsort' comparison function for the fake bad-blocks list. |
| 1067 | * Ranges which start earlier are sorted before rangers which start |
| 1068 | * later. |
| 1069 | */ |
| 1070 | { |
| 1071 | const struct badblock *ba = a, *bb = b; |
| 1072 | |
| 1073 | /* Order by start sector. */ |
| 1074 | if (ba->start < bb->start) return (-1); |
| 1075 | else if (ba->start > bb->start) return (+1); |
| 1076 | |
| 1077 | /* Order by end sector as a tiebreak. This shouldn't be possible. */ |
| 1078 | if (ba->end < bb->end) return (-1); |
| 1079 | else if (ba->end > bb->end) return (+1); |
| 1080 | |
| 1081 | /* They're equal. This shouldn't be possible either. */ |
| 1082 | return (0); |
| 1083 | } |
| 1084 | |
| 1085 | static double bad_block_delay = 0.0, good_block_delay = 0.0; |
| 1086 | /* delay parameters for performance testing */ |
| 1087 | |
| 1088 | static ssize_t read_sectors(secaddr pos, void *buf, secaddr want) |
| 1089 | /* Try to read WANT sectors from the input, starting with sector POS, |
| 1090 | * and write the contents to BUF. Return the number of /whole |
| 1091 | * sectors/ read; this will be 0 at end-of-file (though that |
| 1092 | * shouldn't happen). The returned length will be smaller than WANT |
| 1093 | * only if end-of-file or a system error prevents reading further. |
| 1094 | * Returns -1 on a system error if that prevented us from reading |
| 1095 | * anything at all. |
| 1096 | * |
| 1097 | * This function is where the fake bad-blocks list is handled. |
| 1098 | */ |
| 1099 | { |
| 1100 | ssize_t n, done; |
| 1101 | size_t lo, mid, hi; |
| 1102 | int fakeerr = 0; |
| 1103 | struct badblock *bad, *best; |
| 1104 | unsigned char *p = buf; |
| 1105 | |
| 1106 | /* See whether the requested range intersects a bad-blocks range. */ |
| 1107 | if (badblocks.n) { |
| 1108 | /* Since the list is sorted, we use a binary search. We're looking for |
| 1109 | * the earliest-starting range which /ends after/ POS. If this starts |
| 1110 | * /at or before/ POS, then POS itself is a bad sector, and we should |
| 1111 | * pretend an I/O error; otherwise, if the bad range /starts/ somewhere |
| 1112 | * in the range we're trying to read then we must pretend a short read; |
| 1113 | * and otherwise there's nothing to do. |
| 1114 | */ |
| 1115 | |
| 1116 | /* Throughout, `best' points to the earliest-starting range we've found |
| 1117 | * which (starts and) finishes after POS. Ranges with indices below LO |
| 1118 | * end too early to be interesting; similarly, ranges with indices HI or |
| 1119 | * above start later than POS. If we find a range which actually covers |
| 1120 | * POS exactly then we'll stop early. |
| 1121 | */ |
| 1122 | best = 0; lo = 0; hi = badblocks.n; |
| 1123 | #ifdef DEBUG |
| 1124 | progress_clear(&progress); |
| 1125 | printf(";; searching badblocks for %"PRIuSEC" .. %"PRIuSEC"\n", |
| 1126 | pos, pos + want); |
| 1127 | #endif |
| 1128 | while (lo < hi) { |
| 1129 | /* Standard binary-search loop: we continue until the pointers |
| 1130 | * converge. |
| 1131 | */ |
| 1132 | |
| 1133 | /* Try the midpoint between the two bounds. */ |
| 1134 | mid = lo + (hi - lo)/2; bad = &badblocks.v[mid]; |
| 1135 | #ifdef DEBUG |
| 1136 | printf(";; try %zu (%"PRIuSEC" .. %"PRIuSEC")... ", |
| 1137 | mid, bad->start, bad->end); |
| 1138 | #endif |
| 1139 | |
| 1140 | /* Follow our invariant. If the range starts strictly after POS, then |
| 1141 | * it's too late to overlap, so bring down HI to cover it; but it must |
| 1142 | * be closer than any previous block we've found, so remember it in |
| 1143 | * `best'. Similarly, if the range ends /at or before/ POS then it |
| 1144 | * stops too early, so bring up LO to cover it (but otherwise forget |
| 1145 | * about it because it can't affect what we're doing). |
| 1146 | * |
| 1147 | * If we get a match then we stop immediately and fake a bad block. |
| 1148 | */ |
| 1149 | if (pos < bad->start) { D( printf("high\n"); ) best = bad; hi = mid; } |
| 1150 | else if (pos >= bad->end) { D( printf("low\n"); ) lo = mid + 1; } |
| 1151 | else { |
| 1152 | D( printf("match!\n"); ) |
| 1153 | errno = EIO; sit(bad_block_delay); return (-1); |
| 1154 | } |
| 1155 | } |
| 1156 | |
| 1157 | /* We're done. Check to see whether the bad range starts early enough. |
| 1158 | * If so, remember that we're simulating an error, apply the delay, and |
| 1159 | * bamboozle the rest of the code into performing a short read. |
| 1160 | */ |
| 1161 | #ifdef DEBUG |
| 1162 | if (best) |
| 1163 | printf(";; next is %"PRIuSEC" .. %"PRIuSEC"\n", |
| 1164 | best->start, best->end); |
| 1165 | #endif |
| 1166 | if (best && pos + want > best->start) |
| 1167 | { want = best->start - pos; fakeerr = EIO; sit(bad_block_delay); } |
| 1168 | } |
| 1169 | |
| 1170 | /* Try to read stuff into the buffer until we find a reason why we can't |
| 1171 | * continue. Obviously we need to keep track of how much stuff we've read |
| 1172 | * on previous iterations. |
| 1173 | */ |
| 1174 | done = 0; errno = 0; |
| 1175 | while (want) { |
| 1176 | |
| 1177 | /* Read from the current file's input source. If that's a scrambled |
| 1178 | * video file, then use `libdvdread'; if it's the `raw' file, then go to |
| 1179 | * the block device; if it's nothing at all, then fill with zeros. |
| 1180 | * Always force a seek to the right place, in case things got messed up |
| 1181 | * by some previous error. |
| 1182 | */ |
| 1183 | if (vob) |
| 1184 | { errno = 0; n = DVDReadBlocks(vob, pos - file->start, want, p); } |
| 1185 | else if (file) { |
| 1186 | if (lseek(dvdfd, (off_t)pos*SECTORSZ, SEEK_SET) < 0) |
| 1187 | bail_syserr(errno, "failed to seek to sector %"PRIuSEC"", pos); |
| 1188 | errno = 0; n = read(dvdfd, p, want*SECTORSZ); |
| 1189 | if (n >= 0) n /= SECTORSZ; |
| 1190 | } else { |
| 1191 | memset(p, 0, want*SECTORSZ); |
| 1192 | n = want; |
| 1193 | } |
| 1194 | |
| 1195 | /* If we read some stuff then update the buffer pointer and lengths. If |
| 1196 | * we hit end-of-file then stop. If we hit a bad sector then maybe make |
| 1197 | * a note of it in the bad-sector log. On any other kind of error, just |
| 1198 | * stop. |
| 1199 | */ |
| 1200 | if (n > 0) { done += n; pos += n; p += n*SECTORSZ; want -= n; } |
| 1201 | else if (!n) break; |
| 1202 | else if (errno == EIO && errfile) { |
| 1203 | open_file_on_demand(errfile, &errfp, "bad-sector error log"); |
| 1204 | fprintf(errfp, "%"PRIuSEC" %"PRIuSEC"\n", pos, pos + 1); |
| 1205 | check_write(errfp, "bad-sector error log"); |
| 1206 | break; |
| 1207 | } else if (errno != EINTR) break; |
| 1208 | } |
| 1209 | |
| 1210 | /* We made it. If we saved up a fake error, and there wasn't a real error |
| 1211 | * (which should obviously take priority) then present the fake error to |
| 1212 | * the caller. If there wasn't an error, then everything must have been |
| 1213 | * good so impose the good-block delay -- note that a bad-block delay will |
| 1214 | * already have been imposed above. Finally, return the accumulated count |
| 1215 | * of sectors successfully read, or report the end-of-file or error |
| 1216 | * condition as applicable. |
| 1217 | */ |
| 1218 | if (fakeerr && !errno) errno = fakeerr; |
| 1219 | else if (done > 0 && good_block_delay) sit(done*good_block_delay); |
| 1220 | return (!done && errno ? -1 : done); |
| 1221 | } |
| 1222 | |
| 1223 | /*----- Tracking machinery for the bad-sector algorithm -------------------* |
| 1224 | * |
| 1225 | * While we're probing around trying to find the end of the bad region, we'll |
| 1226 | * have read some good data. We want to try to keep as much good data as we |
| 1227 | * can, and avoid re-reading it because (a) it's pointless I/O work, but more |
| 1228 | * importantly (b) it might not work the second time. The machinery here |
| 1229 | * is for making this work properly. |
| 1230 | * |
| 1231 | * There are two parts to this which don't really intersect, but for |
| 1232 | * convenience the tracking information for them is kept in the same |
| 1233 | * `recoverybuf' structure. |
| 1234 | * |
| 1235 | * * The `short-range' machinery keeps track of a contiguous region of good |
| 1236 | * data stored in the caller's buffer. |
| 1237 | * |
| 1238 | * * The `long-range' machinery keeps track of a contiguous region of good |
| 1239 | * data that's beyond the range of the buffer. |
| 1240 | */ |
| 1241 | |
| 1242 | struct recoverybuf { |
| 1243 | /* Information used to keep track of where good and bad sectors are |
| 1244 | * while we're trying to find the end of a region of bad sectors. |
| 1245 | */ |
| 1246 | |
| 1247 | /* Short-range buffer tracking. */ |
| 1248 | unsigned char *buf; /* pointer to the actual buffer */ |
| 1249 | secaddr sz; /* size of the buffer in sectors */ |
| 1250 | secaddr pos; /* sector address corresponding to |
| 1251 | * the start of the buffer */ |
| 1252 | secaddr start, end; /* bounds of the live region within |
| 1253 | * the buffer, as offsets in |
| 1254 | * sectors from the buffer start */ |
| 1255 | |
| 1256 | /* Long-range tracking. */ |
| 1257 | secaddr good_lo, good_hi; /* known-good region, as absolute |
| 1258 | * sector addresses */ |
| 1259 | }; |
| 1260 | |
| 1261 | static void rearrange_sectors(struct recoverybuf *r, |
| 1262 | secaddr dest, secaddr src, secaddr len) |
| 1263 | /* Shuffle data about in R's buffer. Specifically, move LEN sectors |
| 1264 | * starting SRC sectors from the start of the buffer to a new |
| 1265 | * position DEST sectors from the start. |
| 1266 | * |
| 1267 | * Unsurprisingly, this is a trivial wrapper around `memmove', with |
| 1268 | * some range checking thrown in; it's only used by `recovery_read_- |
| 1269 | * buffer' and `find_good_sector' below. |
| 1270 | */ |
| 1271 | { |
| 1272 | assert(dest + len <= r->sz); assert(src + len <= r->sz); |
| 1273 | memmove(r->buf + dest*SECTORSZ, r->buf + src*SECTORSZ, len*SECTORSZ); |
| 1274 | } |
| 1275 | |
| 1276 | #ifdef DEBUG |
| 1277 | static PRINTF_LIKE(2, 3) |
| 1278 | void show_recovery_buffer_map(const struct recoverybuf *r, |
| 1279 | const char *what, ...) |
| 1280 | /* Dump a simple visualization of the short-range tracking state. */ |
| 1281 | { |
| 1282 | va_list ap; |
| 1283 | |
| 1284 | va_start(ap, what); |
| 1285 | progress_clear(&progress); |
| 1286 | printf(";; recovery buffer ("); |
| 1287 | vprintf(what, ap); |
| 1288 | printf("): " |
| 1289 | "(%"PRIuSEC") ..%"PRIuSEC".. " |
| 1290 | "[%"PRIuSEC" ..%"PRIuSEC".. %"PRIuSEC"] " |
| 1291 | "..%"PRIuSEC".. (%"PRIuSEC")\n", |
| 1292 | r->pos, r->start, |
| 1293 | r->pos + r->start, r->end - r->start, r->pos + r->end, |
| 1294 | r->sz - r->end, r->pos + r->sz); |
| 1295 | va_end(ap); |
| 1296 | assert(r->start <= r->end); |
| 1297 | assert(r->end <= r->sz); |
| 1298 | } |
| 1299 | #endif |
| 1300 | |
| 1301 | static ssize_t recovery_read_sectors(struct recoverybuf *r, |
| 1302 | secaddr pos, secaddr off, secaddr want) |
| 1303 | /* Try to read WANT sectors starting at sector address POS from the |
| 1304 | * current file into R's buffer, at offset OFF sectors from the start |
| 1305 | * of the buffer. Return the number of sectors read, zero if at end |
| 1306 | * of file, or -1 in the event of a system error. |
| 1307 | * |
| 1308 | * This is a trivial wrapper around `read_sectors' with some |
| 1309 | * additional range checking, used only by `recovery_read_buffer' |
| 1310 | * below. |
| 1311 | */ |
| 1312 | { |
| 1313 | ssize_t n; |
| 1314 | |
| 1315 | assert(off <= r->sz); assert(want <= r->sz - off); |
| 1316 | assert(pos == r->pos + off); |
| 1317 | n = read_sectors(pos, r->buf + off*SECTORSZ, want); |
| 1318 | return (n); |
| 1319 | } |
| 1320 | |
| 1321 | static ssize_t recovery_read_buffer(struct recoverybuf *r, |
| 1322 | secaddr pos, secaddr want) |
| 1323 | /* Try to read WANT sectors, starting at sector address POS, from the |
| 1324 | * current file into the buffer R, returning a count of the number of |
| 1325 | * sectors read, or 0 if at end of file, or -1 in the case of a |
| 1326 | * system error, as for `read_sectors'. The data will end up |
| 1327 | * /somewhere/ in the buffer, but not necessarily at the start. |
| 1328 | */ |
| 1329 | { |
| 1330 | secaddr diff, pp, nn; |
| 1331 | ssize_t n; |
| 1332 | |
| 1333 | /* This is the main piece of the short-range tracking machinery. It's |
| 1334 | * rather complicated, so hold on tight. (It's much simpler -- and less |
| 1335 | * broken -- than earlier versions were, though.) |
| 1336 | */ |
| 1337 | |
| 1338 | #ifdef DEBUG |
| 1339 | progress_clear(&progress); |
| 1340 | show_recovery_buffer_map(r, "begin(%"PRIuSEC", %"PRIuSEC")", pos, want); |
| 1341 | #endif |
| 1342 | |
| 1343 | /* The first order of business is to make space in the buffer for this new |
| 1344 | * data. We therefore start with a case analysis. |
| 1345 | */ |
| 1346 | if (pos < r->pos) { |
| 1347 | /* The new position is before the current start of the buffer, so we have |
| 1348 | * no choice but to decrease the buffer position, which will involve |
| 1349 | * shifting the existing material upwards. |
| 1350 | */ |
| 1351 | |
| 1352 | /* Determine how far up we'll need to shift. */ |
| 1353 | diff = r->pos - pos; |
| 1354 | |
| 1355 | if (r->start + diff >= r->sz) { |
| 1356 | /* The material that's currently in the buffer would be completely |
| 1357 | * shifted off the end, so we have no choice but to discard it |
| 1358 | * completely. |
| 1359 | */ |
| 1360 | |
| 1361 | r->pos = pos; r->start = r->end = 0; |
| 1362 | #ifdef DEBUG |
| 1363 | show_recovery_buffer_map(r, "cleared; shift up by %"PRIuSEC"", diff); |
| 1364 | #endif |
| 1365 | } else { |
| 1366 | /* Some of the material in the buffer will still be there. We might |
| 1367 | * lose some stuff off the end: start by throwing that away, and then |
| 1368 | * whatever's left can be moved easily. |
| 1369 | */ |
| 1370 | |
| 1371 | if (r->end + diff > r->sz) r->end = r->sz - diff; |
| 1372 | rearrange_sectors(r, r->start + diff, r->start, r->end - r->start); |
| 1373 | r->pos -= diff; r->start += diff; r->end += diff; |
| 1374 | #ifdef DEBUG |
| 1375 | show_recovery_buffer_map(r, "shifted up by %"PRIuSEC"", diff); |
| 1376 | #endif |
| 1377 | } |
| 1378 | } else if (pos > r->pos + r->end) { |
| 1379 | /* The new position is strictly beyond the old region. We /could/ maybe |
| 1380 | * keep this material, but it turns out to be better not to. To keep it, |
| 1381 | * we'd have to also read the stuff that's in between the end of the old |
| 1382 | * region and the start of the new one, and that might contain bad |
| 1383 | * sectors which the caller is specifically trying to skip. We just |
| 1384 | * discard the entire region here so as not to subvert the caller's |
| 1385 | * optimizations. |
| 1386 | */ |
| 1387 | |
| 1388 | r->pos = pos; r->start = r->end = 0; |
| 1389 | #ifdef DEBUG |
| 1390 | show_recovery_buffer_map(r, "cleared; beyond previous region"); |
| 1391 | #endif |
| 1392 | } else if (pos + want > r->pos + r->sz) { |
| 1393 | /* The requested range of sectors extends beyond the region currently |
| 1394 | * covered by the buffer. We must therefore increase the buffer position |
| 1395 | * which will involve shifting the existing material downwards. |
| 1396 | */ |
| 1397 | |
| 1398 | /* Determine how far down we'll need to shift. */ |
| 1399 | diff = (pos + want) - (r->pos + r->sz); |
| 1400 | |
| 1401 | if (r->end <= diff) { |
| 1402 | /* The material that's currently in the buffer would be completely |
| 1403 | * shifted off the beginning, so we have no choice but to discard it |
| 1404 | * completely. |
| 1405 | */ |
| 1406 | |
| 1407 | r->pos = pos; r->start = r->end = 0; |
| 1408 | #ifdef DEBUG |
| 1409 | show_recovery_buffer_map(r, "cleared; shift down by %"PRIuSEC"", diff); |
| 1410 | #endif |
| 1411 | } else { |
| 1412 | /* Some of the material in the buffer will still be there. We might |
| 1413 | * lose some stuff off the beginning: start by throwing that away, and |
| 1414 | * then whatever's left can be moved easily. |
| 1415 | */ |
| 1416 | |
| 1417 | if (r->start < diff) r->start = diff; |
| 1418 | rearrange_sectors(r, r->start - diff, r->start, r->end - r->start); |
| 1419 | r->pos += diff; r->start -= diff; r->end -= diff; |
| 1420 | #ifdef DEBUG |
| 1421 | show_recovery_buffer_map(r, "shifted down by %"PRIuSEC"", diff); |
| 1422 | #endif |
| 1423 | } |
| 1424 | } |
| 1425 | |
| 1426 | /* We now have space in the buffer in which to put the new material. |
| 1427 | * However, the buffer already contains some stuff. We may need to read |
| 1428 | * some data from the input file into an area before the existing |
| 1429 | * material, or into an area following the existing stuff, or both, or |
| 1430 | * (possibly) neither. |
| 1431 | */ |
| 1432 | |
| 1433 | if (pos < r->pos + r->start) { |
| 1434 | /* The requested position is before the current good material, so we'll |
| 1435 | * need to read some stuff there. |
| 1436 | */ |
| 1437 | |
| 1438 | /* Determine the place in the buffer where this data will be placed, and |
| 1439 | * how long it will need to be. Try to extend it all the way to the |
| 1440 | * existing region even if this is more than the caller wants, because it |
| 1441 | * will mean that we can join it onto the existing region rather than |
| 1442 | * having to decide which of two disconnected parts to throw away. |
| 1443 | */ |
| 1444 | pp = pos - r->pos; nn = r->start - pp; |
| 1445 | |
| 1446 | /* Read the data. */ |
| 1447 | #ifdef DEBUG |
| 1448 | printf(";; read low (%"PRIuSEC"@%"PRIuSEC", %"PRIuSEC")", pos, pp, nn); |
| 1449 | fflush(stdout); |
| 1450 | #endif |
| 1451 | n = recovery_read_sectors(r, pos, pp, nn); |
| 1452 | #ifdef DEBUG |
| 1453 | printf(" -> %zd\n", n); |
| 1454 | #endif |
| 1455 | |
| 1456 | /* See whether it worked. */ |
| 1457 | if (n != nn) { |
| 1458 | /* We didn't get everything we wanted. */ |
| 1459 | |
| 1460 | /* If we got more than the caller asked for then technically this is |
| 1461 | * good; but there must be some problem lurking up ahead, and the |
| 1462 | * caller will want to skip past that. So we don't update the tracking |
| 1463 | * information to reflect our new data; even though this /looks/ like a |
| 1464 | * success, it isn't really. |
| 1465 | */ |
| 1466 | if (n >= 0 && n > want) n = want; |
| 1467 | |
| 1468 | /* We're done. */ |
| 1469 | goto end; |
| 1470 | } |
| 1471 | |
| 1472 | /* Extend the region to include the new piece. */ |
| 1473 | r->start = pp; |
| 1474 | #ifdef DEBUG |
| 1475 | show_recovery_buffer_map(r, "joined new region"); |
| 1476 | #endif |
| 1477 | } |
| 1478 | |
| 1479 | if (pos + want > r->pos + r->end) { |
| 1480 | /* The requested region extends beyond the current region, so we'll need |
| 1481 | * to read some stuff there. |
| 1482 | */ |
| 1483 | |
| 1484 | /* Determine the place in the buffer where this data will be placed, and |
| 1485 | * how long it will need to be. Note that pos <= r->pos + r->end, so |
| 1486 | * there won't be a gap between the old good region and the material |
| 1487 | * we're trying to read. |
| 1488 | */ |
| 1489 | pp = r->end; nn = (pos + want) - (r->pos + r->end); |
| 1490 | |
| 1491 | /* Read the data. */ |
| 1492 | #ifdef DEBUG |
| 1493 | printf(";; read high (%"PRIuSEC"@%"PRIuSEC", %"PRIuSEC")", |
| 1494 | r->pos + pp, pp, nn); |
| 1495 | fflush(stdout); |
| 1496 | #endif |
| 1497 | n = recovery_read_sectors(r, r->pos + pp, pp, nn); |
| 1498 | #ifdef DEBUG |
| 1499 | printf(" -> %zd\n", n); |
| 1500 | #endif |
| 1501 | |
| 1502 | /* See whether it worked. */ |
| 1503 | if (n > 0) { |
| 1504 | /* We read something, so add it onto the existing region. */ |
| 1505 | |
| 1506 | r->end += n; |
| 1507 | #ifdef DEBUG |
| 1508 | show_recovery_buffer_map(r, "joined new region"); |
| 1509 | #endif |
| 1510 | } |
| 1511 | } |
| 1512 | |
| 1513 | /* Work out the return value to pass back to the caller. The newly read |
| 1514 | * material has been merged with the existing region (the case where we |
| 1515 | * didn't manage to join the two together has been handled already), so we |
| 1516 | * can easily work out how much stuff is available by looking at the |
| 1517 | * tracking information. It only remains to bound the region size by the |
| 1518 | * requested length. |
| 1519 | */ |
| 1520 | n = r->pos + r->end - pos; |
| 1521 | if (!n && want) n = -1; |
| 1522 | else if (n > want) n = want; |
| 1523 | |
| 1524 | end: |
| 1525 | /* Done. */ |
| 1526 | #ifdef DEBUG |
| 1527 | show_recovery_buffer_map(r, "done; return %zd", n); |
| 1528 | #endif |
| 1529 | return (n); |
| 1530 | } |
| 1531 | |
| 1532 | static ssize_t recovery_read_multiple(struct recoverybuf *r, |
| 1533 | secaddr pos, secaddr want) |
| 1534 | /* Try to read WANT sectors, starting at sector address POS, from the |
| 1535 | * current file, returning a count of the number of sectors read, or |
| 1536 | * 0 if at end of file, or -1 in the case of a system error, as for |
| 1537 | * `read_sectors'. Some data might end up in R's buffer, but if WANT |
| 1538 | * is larger than R->sz then a lot will be just thrown away. |
| 1539 | * |
| 1540 | * This is only used by `recovery_read' below. |
| 1541 | */ |
| 1542 | { |
| 1543 | ssize_t n; |
| 1544 | secaddr skip, want0 = want; |
| 1545 | |
| 1546 | /* If the request is larger than the buffer, then we start at the /end/ and |
| 1547 | * work backwards. If we encounter a bad sector while we're doing this, |
| 1548 | * then we report a short read as far as the bad sector: the idea is to |
| 1549 | * find the /latest/ bad sector we can. The caller will want to skip past |
| 1550 | * the bad sector, so the fact that we implicitly lied about the earlier |
| 1551 | * data as being `good' won't matter. |
| 1552 | */ |
| 1553 | |
| 1554 | while (want > r->sz) { |
| 1555 | /* There's (strictly!) more than a buffer's worth. Fill the buffer with |
| 1556 | * stuff and reduce the requested size. |
| 1557 | */ |
| 1558 | |
| 1559 | skip = want - r->sz; |
| 1560 | n = recovery_read_buffer(r, pos + skip, r->sz); |
| 1561 | |
| 1562 | /* If it failed, then we always return a positive result, because we're |
| 1563 | * pretending we managed to read all of the (nonempty) preceding |
| 1564 | * material. |
| 1565 | */ |
| 1566 | if (n < r->sz) return (skip + (n >= 0 ? n : 0)); |
| 1567 | |
| 1568 | /* Cross off a buffer's worth and go around again. */ |
| 1569 | want -= r->sz; |
| 1570 | } |
| 1571 | |
| 1572 | /* Read the last piece. If it fails or comes up short, then we don't need |
| 1573 | * to mess with the return code this time. |
| 1574 | */ |
| 1575 | n = recovery_read_buffer(r, pos, want); |
| 1576 | if (n < 0 || n < want) return (n); |
| 1577 | |
| 1578 | /* It all worked. Return the full original amount requested. */ |
| 1579 | return (want0); |
| 1580 | } |
| 1581 | |
| 1582 | static ssize_t recovery_read(struct recoverybuf *r, |
| 1583 | secaddr pos, secaddr want) |
| 1584 | /* Try to read WANT sectors, starting at sector address POS, from the |
| 1585 | * current file, returning a count of the number of |
| 1586 | * sectors read, or 0 if at end of file, or -1 in the case of a |
| 1587 | * system error, as for `read_sectors'. Some data might end up in |
| 1588 | * R's buffer, but if WANT is larger than R->sz then a lot will be |
| 1589 | * just thrown away. |
| 1590 | */ |
| 1591 | { |
| 1592 | secaddr lo = pos, hi = pos + want, span; /* calculate the request bounds */ |
| 1593 | ssize_t n; |
| 1594 | |
| 1595 | /* This is the main piece of the long-range tracking machinery. |
| 1596 | * Fortunately, it's much simpler than the short-range stuff that we've |
| 1597 | * just dealt with. |
| 1598 | */ |
| 1599 | |
| 1600 | if (hi < r->good_lo || lo > r->good_hi) { |
| 1601 | /* The requested region doesn't abut or overlap with the existing good |
| 1602 | * region, so it's no good to us. Just read the requested region; if it |
| 1603 | * worked at all, then replace the current known-good region with the |
| 1604 | * region that was successfully read. |
| 1605 | */ |
| 1606 | |
| 1607 | n = recovery_read_multiple(r, lo, hi - lo); |
| 1608 | if (n > 0) { r->good_lo = lo; r->good_hi = lo + n; } |
| 1609 | return (n); |
| 1610 | } |
| 1611 | |
| 1612 | if (hi > r->good_hi) { |
| 1613 | /* The requested region ends later than the current known-good region. |
| 1614 | * Read the missing piece. We're doing this first so that we find later |
| 1615 | * bad sectors. |
| 1616 | */ |
| 1617 | |
| 1618 | span = hi - r->good_hi; |
| 1619 | n = recovery_read_multiple(r, r->good_hi, span); |
| 1620 | |
| 1621 | /* If we read anything at all, then extend the known-good region. */ |
| 1622 | if (n > 0) r->good_hi += n; |
| 1623 | |
| 1624 | /* If we didn't read everything we wanted, then report this as a short |
| 1625 | * read (so including some nonempty portion of the known-good region). |
| 1626 | */ |
| 1627 | if (n < 0 || n < span) return (r->good_hi - lo); |
| 1628 | } |
| 1629 | |
| 1630 | if (lo < r->good_lo) { |
| 1631 | /* The requested region begins earlier than the known-good region. */ |
| 1632 | |
| 1633 | span = r->good_lo - lo; |
| 1634 | n = recovery_read_multiple(r, lo, span); |
| 1635 | |
| 1636 | /* If we read everything we wanted, then extend the known-good region. |
| 1637 | * Otherwise, we're better off keeping the stuff after the bad block. |
| 1638 | */ |
| 1639 | if (n == span) r->good_lo = lo; |
| 1640 | else return (n); |
| 1641 | } |
| 1642 | |
| 1643 | /* Everything read OK, and we've extended the known-good region to cover |
| 1644 | * the requested region. So return an appropriate code by consulting the |
| 1645 | * new known-good region. |
| 1646 | */ |
| 1647 | n = r->good_hi - pos; if (n > want) n = want; |
| 1648 | if (!n) { errno = EIO; n = -1; } |
| 1649 | return (n); |
| 1650 | } |
| 1651 | |
| 1652 | /*----- Skipping past regions of bad sectors ------------------------------*/ |
| 1653 | |
| 1654 | static double clear_factor = 0.5; /* proportion of clear sectors needed */ |
| 1655 | static secaddr clear_min = 1, clear_max = SECLIMIT; /* absolute bounds */ |
| 1656 | static double step_factor = 2.0; /* factor for how far to look ahead */ |
| 1657 | static secaddr step_min = 1, step_max = 0; /* and absolute bounds */ |
| 1658 | |
| 1659 | static void recovered(secaddr bad_lo, secaddr bad_hi) |
| 1660 | /* Do all of the things that are necessary when a region of bad |
| 1661 | * sectors has been found between BAD_LO (inclusive) and BAD_HI |
| 1662 | * (exclusive). |
| 1663 | */ |
| 1664 | { |
| 1665 | char fn[MAXFNSZ]; |
| 1666 | |
| 1667 | /* Remove the progress display temporarily. */ |
| 1668 | progress_clear(&progress); |
| 1669 | |
| 1670 | /* Print a message into the permanent output log. */ |
| 1671 | if (!file || id_kind(file->id) == RAW) |
| 1672 | moan("skipping %"PRIuSEC" bad sectors (%"PRIuSEC" .. %"PRIuSEC")", |
| 1673 | bad_hi - bad_lo, bad_lo, bad_hi); |
| 1674 | else { |
| 1675 | store_filename(fn, file->id); |
| 1676 | moan("skipping %"PRIuSEC" bad sectors (%"PRIuSEC" .. %"PRIuSEC"; " |
| 1677 | "`%s' %"PRIuSEC" .. %"PRIuSEC" of %"PRIuSEC")", |
| 1678 | bad_hi - bad_lo, bad_lo, bad_hi, |
| 1679 | fn, bad_lo - file->start, bad_hi - file->start, |
| 1680 | file->end - file->start); |
| 1681 | } |
| 1682 | |
| 1683 | if (mapfile) { |
| 1684 | /* The user requested a map of the skipped regions, so write an entry. */ |
| 1685 | |
| 1686 | /* Open the file, if it's not open already. */ |
| 1687 | open_file_on_demand(mapfile, &mapfp, "bad-sector region map"); |
| 1688 | |
| 1689 | /* Write the sector range. */ |
| 1690 | fprintf(mapfp, "%"PRIuSEC" %"PRIuSEC" # %"PRIuSEC" sectors", |
| 1691 | bad_lo, bad_hi, bad_hi - bad_lo); |
| 1692 | |
| 1693 | /* If we're currently reading from a file then note down the position in |
| 1694 | * the file in the comment. (Intentional bad sectors are frequently at |
| 1695 | * the start and end of titles, so this helps a reader to decide how |
| 1696 | * concerned to be.) |
| 1697 | */ |
| 1698 | if (file && id_kind(file->id) != RAW) |
| 1699 | fprintf(mapfp, "; `%s' %"PRIuSEC" .. %"PRIuSEC" of %"PRIuSEC"", |
| 1700 | fn, bad_lo - file->start, bad_hi - file->start, |
| 1701 | file->end - file->start); |
| 1702 | |
| 1703 | /* Done. Flush the output to the file so that we don't lose it if we |
| 1704 | * crash! |
| 1705 | */ |
| 1706 | fputc('\n', mapfp); |
| 1707 | check_write(mapfp, "bad-sector region map"); |
| 1708 | } |
| 1709 | |
| 1710 | /* Adjust the position in our output file to skip past the bad region. |
| 1711 | * (This avoids overwriting anything that was there already, which is |
| 1712 | * almost certainly less wrong than anything we could come up with here.) |
| 1713 | */ |
| 1714 | if (lseek(outfd, (off_t)(bad_hi - bad_lo)*SECTORSZ, SEEK_CUR) < 0) |
| 1715 | bail_syserr(errno, "failed to seek past bad sectors"); |
| 1716 | |
| 1717 | /* Remove our notice now that we're no longer messing about with bad |
| 1718 | * sectors, and reinstate the progress display. |
| 1719 | */ |
| 1720 | progress_removeitem(&progress, &badblock_progress); |
| 1721 | progress_update(&progress); |
| 1722 | } |
| 1723 | |
| 1724 | static secaddr run_length_wanted(secaddr pos, secaddr badlen, secaddr end) |
| 1725 | /* Return the number of good sectors that we want to see before |
| 1726 | * we're happy, given that we're about to try to read sector POS, |
| 1727 | * which is BADLEN sectors beyond where we found the first bad |
| 1728 | * sector, and the current region ends at sector END (i.e., this is |
| 1729 | * where the next event occurs). |
| 1730 | */ |
| 1731 | { |
| 1732 | secaddr want; |
| 1733 | |
| 1734 | /* Apply the factor to BADLEN to get an initial length. */ |
| 1735 | want = ceil(clear_factor*badlen); |
| 1736 | |
| 1737 | /* Apply the user-configurable lower bound. */ |
| 1738 | if (want < clear_min) want = clear_min; |
| 1739 | |
| 1740 | /* Cap this with the end of the region. */ |
| 1741 | if (want > end - pos) want = end - pos; |
| 1742 | |
| 1743 | /* And apply the user-configurable upper bound. */ |
| 1744 | if (clear_max && want > clear_max) want = clear_max; |
| 1745 | |
| 1746 | /* We're done. */ |
| 1747 | return (want); |
| 1748 | } |
| 1749 | |
| 1750 | static void report_bad_blocks_progress(secaddr bad_hi, int err) |
| 1751 | /* Report progress while we're trying to work past a region of bad |
| 1752 | * sectors. We're about to investigate BAD_HI, and the most recent |
| 1753 | * error was ERR. |
| 1754 | */ |
| 1755 | { bad_err = err; report_progress(bad_hi); } |
| 1756 | |
| 1757 | static ssize_t find_good_sector(secaddr *pos_inout, secaddr end, |
| 1758 | unsigned char *buf, secaddr sz) |
| 1759 | /* Work out a place to resume after finding a bad sector. The |
| 1760 | * current position, where we found a problem, is in *POS_INOUT. The |
| 1761 | * current input region goes up up sector END (i.e., this is where |
| 1762 | * the next event occurs). The caller's buffer is at BUF, and can |
| 1763 | * hold SZ sectors. On exit, update *POS_INOUT to be the start of a |
| 1764 | * region of /good/ sector that we decided was worth exploring, and |
| 1765 | * return the number of sectors we've already read at that position |
| 1766 | * and left at the start of the buffer. (This number may be zero, |
| 1767 | * depending on how things work out. That doesn't mean that we hit |
| 1768 | * end-of-file.) |
| 1769 | * |
| 1770 | * Altough the return value is `ssize_t', this is only to fit in with |
| 1771 | * other read functions; a negative return is not actually possible. |
| 1772 | */ |
| 1773 | { |
| 1774 | secaddr pos = *pos_inout, bad_lo, bad_hi, good, step, want; |
| 1775 | struct recoverybuf r; |
| 1776 | ssize_t n; |
| 1777 | |
| 1778 | /* Initial setup. Save the initial state and establish the bad-blocks |
| 1779 | * progress notice. |
| 1780 | */ |
| 1781 | bad_start = pos; bad_err = errno; |
| 1782 | badblock_progress.render = render_badblock_progress; |
| 1783 | progress_additem(&progress, &badblock_progress); |
| 1784 | |
| 1785 | /* First, retry the `bad' sector a few times. Sometimes, with damaged |
| 1786 | * discs, this actually works. We'll try to read a full buffer, but we're |
| 1787 | * not expecting much. |
| 1788 | */ |
| 1789 | want = sz; if (want > end - pos) want = end - pos; |
| 1790 | for (retry = 0; retry < max_retries; retry++) { |
| 1791 | |
| 1792 | /* Show the progress report. */ |
| 1793 | report_bad_blocks_progress(pos, errno); |
| 1794 | |
| 1795 | /* Try reading stuff. */ |
| 1796 | n = read_sectors(pos, buf, want); |
| 1797 | #ifdef DEBUG |
| 1798 | progress_clear(&progress); |
| 1799 | printf(";; [retry] try reading %"PRIuSEC" .. %"PRIuSEC" -> %zd\n", |
| 1800 | pos, pos + want, n); |
| 1801 | #endif |
| 1802 | |
| 1803 | if (n > 0) { |
| 1804 | /* We won! Remove the progress display, and leave a permanent message |
| 1805 | * to inform the user what happened. |
| 1806 | */ |
| 1807 | progress_clear(&progress); |
| 1808 | moan("sector %"PRIuSEC" read ok after retry", pos); |
| 1809 | progress_removeitem(&progress, &badblock_progress); |
| 1810 | progress_update(&progress); |
| 1811 | return (n); |
| 1812 | } |
| 1813 | } |
| 1814 | |
| 1815 | /* We're going to have to be more creative. Set up the tracking state. */ |
| 1816 | r.buf = buf; r.sz = sz; r.pos = r.start = r.end = 0; |
| 1817 | r.good_lo = r.good_hi = 0; |
| 1818 | |
| 1819 | /* Set up the region bound. We know the bad area starts at POS, and that |
| 1820 | * it covers at least one sector. |
| 1821 | */ |
| 1822 | bad_lo = pos; bad_hi = pos + 1; |
| 1823 | |
| 1824 | /* Second major step: try to find somewhere on the other side of the bad |
| 1825 | * region. |
| 1826 | */ |
| 1827 | for (;;) { |
| 1828 | #ifdef DEBUG |
| 1829 | progress_clear(&progress); |
| 1830 | printf(";; bounding bad-block region: " |
| 1831 | "%"PRIuSEC" ..%"PRIuSEC".. %"PRIuSEC"\n", |
| 1832 | bad_lo, bad_hi - bad_lo, bad_hi); |
| 1833 | #endif |
| 1834 | |
| 1835 | /* If our upper bound has reached all the way to the end of the input |
| 1836 | * region then there's nowhere to recover to. Set the next position to |
| 1837 | * the end of the region and return. |
| 1838 | */ |
| 1839 | if (bad_hi >= end) { |
| 1840 | progress_clear(&progress); |
| 1841 | moan("giving up on this extent"); |
| 1842 | recovered(bad_lo, end); *pos_inout = end; |
| 1843 | return (0); |
| 1844 | } |
| 1845 | |
| 1846 | /* Give a progress update. */ |
| 1847 | report_bad_blocks_progress(bad_hi, errno); |
| 1848 | |
| 1849 | /* Choose a new place to look. Apply the step factor to the size of the |
| 1850 | * current gap between the start and end of the bad region, and then |
| 1851 | * bound by the user bounds and the input-region end. |
| 1852 | * |
| 1853 | * We make progress because `step' is at least 1: `step_min' is at least |
| 1854 | * 1, and bad_hi < end or we'd have already bailed. |
| 1855 | */ |
| 1856 | step = (step_factor - 1)*(bad_hi - bad_lo); |
| 1857 | if (step < step_min) step = step_min; |
| 1858 | if (step_max && step > step_max) step = step_max; |
| 1859 | step += bad_hi - bad_lo; |
| 1860 | if (step > end - bad_lo) step = end - bad_lo; |
| 1861 | |
| 1862 | /* Now we look at the last sector of the new interval we've just marked |
| 1863 | * out. |
| 1864 | */ |
| 1865 | pos = bad_lo + step - 1; |
| 1866 | want = run_length_wanted(pos, step, end); |
| 1867 | n = recovery_read(&r, pos, want); |
| 1868 | #ifdef DEBUG |
| 1869 | printf(";; [bound] try reading %"PRIuSEC" .. %"PRIuSEC" -> %zd\n", |
| 1870 | pos, pos + want, n); |
| 1871 | #endif |
| 1872 | |
| 1873 | /* If everything went OK then we're done with this phase. */ |
| 1874 | if (n == want) break; |
| 1875 | |
| 1876 | /* If it failed then extend the bad region to cover (the end of) the bad |
| 1877 | * sector which terminated the run, and go around again. |
| 1878 | */ |
| 1879 | if (n < 0) n = 0; |
| 1880 | bad_hi = pos + n + 1; |
| 1881 | } |
| 1882 | |
| 1883 | /* Third major step: identify exactly where the bad region ends. This is |
| 1884 | * a binary search. |
| 1885 | */ |
| 1886 | good = pos; |
| 1887 | while (good > bad_hi) { |
| 1888 | #ifdef DEBUG |
| 1889 | progress_clear(&progress); |
| 1890 | printf(";; limiting bad-block region: " |
| 1891 | "%"PRIuSEC" ..%"PRIuSEC".. %"PRIuSEC" ..%"PRIuSEC".. %"PRIuSEC"\n", |
| 1892 | bad_lo, bad_hi - bad_lo, bad_hi, good - bad_hi, good); |
| 1893 | #endif |
| 1894 | |
| 1895 | /* Update the progress report. */ |
| 1896 | report_bad_blocks_progress(bad_hi, errno); |
| 1897 | |
| 1898 | /* Pick a new place to try. */ |
| 1899 | pos = bad_hi + (good - bad_hi)/2; step = pos - bad_lo; |
| 1900 | want = run_length_wanted(pos, step, end); |
| 1901 | |
| 1902 | /* Try reading. */ |
| 1903 | n = recovery_read(&r, pos, want); |
| 1904 | #ifdef DEBUG |
| 1905 | printf(";; [limit] try reading %"PRIuSEC" .. %"PRIuSEC" -> %zd\n", |
| 1906 | pos, pos + want, n); |
| 1907 | #endif |
| 1908 | |
| 1909 | /* If that worked -- i.e., we got all the data we wanted -- then bring |
| 1910 | * down the `good' bound. If it failed, then bring up `bad_hi' to cover |
| 1911 | * the bad sector which terminated our read attempt. |
| 1912 | */ |
| 1913 | if (n < 0) n = 0; |
| 1914 | if (n == want) good = pos; |
| 1915 | else bad_hi = pos + n + 1; |
| 1916 | } |
| 1917 | |
| 1918 | /* We're done. It's time to tidy up. |
| 1919 | * |
| 1920 | * One subtle point: it's possible that, as a result of retrying previous |
| 1921 | * bad blocks, that we ended up with bad_hi > good, so it's important that |
| 1922 | * we make a consistent choice between the two. I've gone with `good' |
| 1923 | * because (a) this gives us more of the original data from the disc and |
| 1924 | * (b) hopefully any marginal sectors are now in our buffer |
| 1925 | */ |
| 1926 | recovered(bad_lo, good); *pos_inout = good; |
| 1927 | |
| 1928 | /* Figure out how much data we can return to the caller from our buffer. */ |
| 1929 | if (good < r.pos + r.start || r.pos + r.end <= good) { |
| 1930 | /* Our new position is outside of the region covered by the short-range |
| 1931 | * tracking, so there's nothing to return. |
| 1932 | */ |
| 1933 | |
| 1934 | n = 0; |
| 1935 | } else { |
| 1936 | /* The new position is covered, so shuffle the data to the start of the |
| 1937 | * buffer and return as much as we can. |
| 1938 | */ |
| 1939 | |
| 1940 | n = r.pos + r.end - good; |
| 1941 | rearrange_sectors(&r, 0, good - r.pos, n); |
| 1942 | } |
| 1943 | |
| 1944 | /* We're done. */ |
| 1945 | #ifdef DEBUG |
| 1946 | show_recovery_buffer_map(&r, "returning %zd good sectors at %"PRIuSEC"", |
| 1947 | n, good); |
| 1948 | #endif |
| 1949 | return (n); |
| 1950 | } |
| 1951 | |
| 1952 | /*----- Copying data from a single input file -----------------------------*/ |
| 1953 | |
| 1954 | static void emit(secaddr start, secaddr end) |
| 1955 | /* Copy sectors with absolute addresses from START (inclusive) to END |
| 1956 | * (exclusive) to the output. The entire input region comes from the |
| 1957 | * same source, already established as `file'. |
| 1958 | */ |
| 1959 | { |
| 1960 | #define BUFSECTORS 512 /* this is a megabyte */ |
| 1961 | |
| 1962 | int least; |
| 1963 | unsigned char buf[BUFSECTORS*SECTORSZ]; |
| 1964 | secaddr pos; |
| 1965 | size_t want; |
| 1966 | ssize_t n; |
| 1967 | static int first_time = 1; |
| 1968 | #ifdef DEBUG |
| 1969 | struct file *f; |
| 1970 | char fn[MAXFNSZ]; |
| 1971 | int act = -1; |
| 1972 | int i; |
| 1973 | #endif |
| 1974 | |
| 1975 | /* Choose an active file through which to read the source contents. We're |
| 1976 | * guaranteed that this file will do for the entire input region. We |
| 1977 | * choose the active file with the smallest index. The virtual `raw' file |
| 1978 | * which represents the underlying block device has the largest index, so |
| 1979 | * we'll always use a `.VOB' file if one is available. Looking at the |
| 1980 | * protocol suggests that the host and drive identify the per-title CSS key |
| 1981 | * by the start sector address of the `.VOB' file, so coincident files must |
| 1982 | * all use the same key. I've not encountered properly overlapping files |
| 1983 | * in the wild. |
| 1984 | */ |
| 1985 | least = least_live(); |
| 1986 | #ifdef DEBUG |
| 1987 | printf(";; %8"PRIuSEC" .. %"PRIuSEC"\n", start, end); |
| 1988 | for (i = 0; i < filetab.n; i++) { |
| 1989 | if (!livep(i)) continue; |
| 1990 | if (act == -1) act = i; |
| 1991 | f = &filetab.v[i]; store_filename(fn, f->id); |
| 1992 | printf(";;\t\t%8"PRIuSEC" .. %-8"PRIuSEC" %s\n", |
| 1993 | start - f->start, end - f->start, fn); |
| 1994 | } |
| 1995 | if (act == -1) printf(";;\t\t#<no live source>\n"); |
| 1996 | assert(act == least); |
| 1997 | #endif |
| 1998 | |
| 1999 | /* Set the global variables up for reading from the file we decided on. |
| 2000 | * These will be primarily used by `read_sectors' and `update_progress'. |
| 2001 | */ |
| 2002 | if (least == -1) { |
| 2003 | /* There's nothing at all. This can happen because the kernel reported |
| 2004 | * the wrong block-device size for some reason but the filesystem has |
| 2005 | * identified files which start beyond the reported size, leaving a gap. |
| 2006 | */ |
| 2007 | |
| 2008 | file = 0; vob = 0; |
| 2009 | } else { |
| 2010 | /* There's a (possibly) virtual file. */ |
| 2011 | |
| 2012 | file = &filetab.v[least]; |
| 2013 | switch (id_kind(file->id)) { |
| 2014 | |
| 2015 | case RAW: |
| 2016 | /* It's the raw device. Clear `vob' to prompt `read_sectors' to read |
| 2017 | * directly from `dvdfd'. |
| 2018 | */ |
| 2019 | |
| 2020 | vob = 0; |
| 2021 | break; |
| 2022 | |
| 2023 | case VOB: |
| 2024 | /* It's a `.VOB' file. We read these through `libdvdread', which |
| 2025 | * handles CSS unscrambling for us. |
| 2026 | */ |
| 2027 | |
| 2028 | /* The first time we open a `.VOB' file, `libdvdread' wants to spray |
| 2029 | * a bunch of information about how it's getting on cracking the |
| 2030 | * title keys. This will interfere with the progress display, so |
| 2031 | * preemptively hide the display. |
| 2032 | */ |
| 2033 | if (first_time) { progress_clear(&progress); first_time = 0; } |
| 2034 | |
| 2035 | /* Open the `.VOB' file. */ |
| 2036 | vob = DVDOpenFile(dvd, id_title(file->id), |
| 2037 | id_part(file->id) |
| 2038 | ? DVD_READ_TITLE_VOBS |
| 2039 | : DVD_READ_MENU_VOBS); |
| 2040 | if (!vob) |
| 2041 | bail("failed to open %s %u", |
| 2042 | id_part(file->id) ? "title" : "menu", |
| 2043 | id_title(file->id)); |
| 2044 | break; |
| 2045 | |
| 2046 | default: |
| 2047 | /* Some other kind of thing; but there shouldn't be anything else in |
| 2048 | * the file table, so there's a bug. |
| 2049 | */ |
| 2050 | abort(); |
| 2051 | |
| 2052 | } |
| 2053 | } |
| 2054 | |
| 2055 | /* If we're not reading from the raw device then add an additional progress |
| 2056 | * bar for the current file. This isn't completely pointless: having a |
| 2057 | * ready visualization for whereabouts we are in a file is valuable when we |
| 2058 | * encounter bad blocks, because regions of intentional bad blocks near the |
| 2059 | * starts and and ends of VOBs are common on discs from annoying studios. |
| 2060 | */ |
| 2061 | if (file && id_kind(file->id) != RAW) { |
| 2062 | file_progress.render = render_file_progress; |
| 2063 | progress_additem(&progress, &file_progress); |
| 2064 | } |
| 2065 | |
| 2066 | /* Put the progress display back, if we took it away, and show the file |
| 2067 | * progress bar if we added one. |
| 2068 | */ |
| 2069 | update_progress(start); |
| 2070 | |
| 2071 | /* Read the input region and copy it to the disc. */ |
| 2072 | pos = start; |
| 2073 | while (pos < end) { |
| 2074 | |
| 2075 | /* Decide how much we want. Fill the buffer, unless there's not enough |
| 2076 | * input left. |
| 2077 | */ |
| 2078 | want = end - pos; if (want > BUFSECTORS) want = BUFSECTORS; |
| 2079 | |
| 2080 | /* Try to read the input. */ |
| 2081 | n = read_sectors(pos, buf, want); |
| 2082 | |
| 2083 | if (n <= 0) { |
| 2084 | /* It didn't work. Time to deploy the skipping-past-bad-blocks |
| 2085 | * machinery we worked so hard on. This will fill the buffer with |
| 2086 | * stuff and return a new count of how much it read. |
| 2087 | */ |
| 2088 | |
| 2089 | n = find_good_sector(&pos, end, buf, BUFSECTORS); |
| 2090 | } |
| 2091 | if (n > 0) { |
| 2092 | /* We made some progress. Write the stuff that we read to the output |
| 2093 | * file and update the position. |
| 2094 | */ |
| 2095 | |
| 2096 | carefully_write(outfd, buf, n*SECTORSZ); pos += n; |
| 2097 | } |
| 2098 | |
| 2099 | /* Report our new progress. */ |
| 2100 | report_progress(pos); |
| 2101 | } |
| 2102 | |
| 2103 | /* Close the `libdvdread' file, if we opened one. */ |
| 2104 | if (vob) { DVDCloseFile(vob); vob = 0; } |
| 2105 | |
| 2106 | /* If we added a per-file progress bar, then take it away again. */ |
| 2107 | if (file && id_kind(file->id) != RAW) |
| 2108 | progress_removeitem(&progress, &file_progress); |
| 2109 | |
| 2110 | /* Update the progress display to report our glorious success. */ |
| 2111 | progress_update(&progress); |
| 2112 | |
| 2113 | #undef BUFSECTORS |
| 2114 | } |
| 2115 | |
| 2116 | /*----- Main program ------------------------------------------------------*/ |
| 2117 | |
| 2118 | int main(int argc, char *argv[]) |
| 2119 | { |
| 2120 | unsigned f = 0; |
| 2121 | const char *p; |
| 2122 | off_t volsz; |
| 2123 | secaddr pos; |
| 2124 | off_t off; |
| 2125 | secaddr start, end, last; |
| 2126 | const struct event *ev; |
| 2127 | const char *device, *outfile; |
| 2128 | struct badblock *bad; |
| 2129 | int opt, blksz; |
| 2130 | size_t i; |
| 2131 | FILE *fp; |
| 2132 | struct buf buf = BUF_INIT; |
| 2133 | struct timeval tv0, tv1; |
| 2134 | double t, rate, tot; |
| 2135 | const char *rateunit, *totunit; |
| 2136 | char timebuf[TIMESTRMAX], id_in[MAXIDSZ], id_out[MAXIDSZ]; |
| 2137 | dvd_reader_t *dvd_out; |
| 2138 | #ifdef DEBUG |
| 2139 | const struct file *file; |
| 2140 | char fn[MAXFNSZ]; |
| 2141 | #endif |
| 2142 | |
| 2143 | #define f_bogus 1u |
| 2144 | #define f_continue 2u |
| 2145 | #define f_fixup 4u |
| 2146 | #define f_stats 8u |
| 2147 | #define f_checkid 16u |
| 2148 | #define f_retry 32u |
| 2149 | #define f_write 256u |
| 2150 | #define f_file 512u |
| 2151 | |
| 2152 | set_prog(argv[0]); |
| 2153 | |
| 2154 | /* First up, handle the command-line options. */ |
| 2155 | for (;;) { |
| 2156 | opt = getopt(argc, argv, "hB:E:FP:R:X:b:cir:s"); if (opt < 0) break; |
| 2157 | switch (opt) { |
| 2158 | |
| 2159 | /* `-h': Help. */ |
| 2160 | case 'h': usage(stderr); exit(0); |
| 2161 | |
| 2162 | /* `-B PARAM=VALUE[,...]': Setting internal parameters. */ |
| 2163 | case 'B': |
| 2164 | |
| 2165 | /* Set up a cursor into the parameter string. */ |
| 2166 | p = optarg; |
| 2167 | |
| 2168 | #define SKIP_PREFIX(s) \ |
| 2169 | (STRNCMP(p, ==, s "=", sizeof(s)) && (p += sizeof(s), 1)) |
| 2170 | /* If the text at P matches `S=' then advance P past that and |
| 2171 | * evaluate nonzero; otherwise evaluate zero. |
| 2172 | */ |
| 2173 | |
| 2174 | for (;;) { |
| 2175 | |
| 2176 | if (SKIP_PREFIX("cf")) |
| 2177 | clear_factor = parse_float(&p, PNF_JUNK, 0, DBL_MAX, |
| 2178 | "clear factor"); |
| 2179 | |
| 2180 | else if (SKIP_PREFIX("cmin")) |
| 2181 | clear_min = parse_int(&p, PNF_JUNK, 1, SECLIMIT, |
| 2182 | "clear minimum"); |
| 2183 | |
| 2184 | else if (SKIP_PREFIX("cmax")) |
| 2185 | clear_max = parse_int(&p, PNF_JUNK, 1, SECLIMIT, |
| 2186 | "clear maximum"); |
| 2187 | |
| 2188 | else if (SKIP_PREFIX("sf")) |
| 2189 | step_factor = parse_float(&p, PNF_JUNK, 0, DBL_MAX, |
| 2190 | "step factor"); |
| 2191 | |
| 2192 | else if (SKIP_PREFIX("smin")) |
| 2193 | step_min = parse_int(&p, PNF_JUNK, 1, SECLIMIT - 1, |
| 2194 | "step minimum"); |
| 2195 | |
| 2196 | else if (SKIP_PREFIX("smax")) |
| 2197 | step_max = parse_int(&p, PNF_JUNK, 1, SECLIMIT - 1, |
| 2198 | "step maximum"); |
| 2199 | |
| 2200 | else if (SKIP_PREFIX("retry")) |
| 2201 | max_retries = parse_int(&p, PNF_JUNK, 0, INT_MAX, "retries"); |
| 2202 | |
| 2203 | else if (SKIP_PREFIX("alpha")) |
| 2204 | alpha = parse_float(&p, PNF_JUNK, 0, 1, "average decay factor"); |
| 2205 | |
| 2206 | else if (SKIP_PREFIX("_badwait")) |
| 2207 | bad_block_delay = parse_float(&p, PNF_JUNK, 0, DBL_MAX, |
| 2208 | "bad-block delay"); |
| 2209 | |
| 2210 | else if (SKIP_PREFIX("_blkwait")) |
| 2211 | good_block_delay = parse_float(&p, PNF_JUNK, 0, DBL_MAX, |
| 2212 | "good block delay"); |
| 2213 | |
| 2214 | else |
| 2215 | bail("unknown bad blocks parameter `%s'", p); |
| 2216 | |
| 2217 | /* If we're now at the end of the string then we're done. */ |
| 2218 | if (!*p) break; |
| 2219 | |
| 2220 | /* We're not done yet, so there should now be a comma and another |
| 2221 | * parameter setting. |
| 2222 | */ |
| 2223 | if (*p != ',') bail("unexpected junk in parameters"); |
| 2224 | p++; |
| 2225 | } |
| 2226 | |
| 2227 | #undef SKIP_PREFIX |
| 2228 | break; |
| 2229 | |
| 2230 | /* `-E FILE' (undocumented): Log the bad sectors we encountered to |
| 2231 | * FILE. |
| 2232 | */ |
| 2233 | case 'E': errfile = optarg; break; |
| 2234 | |
| 2235 | /* `-F' (undocumented): Hack for fixing up images that were broken by |
| 2236 | * an old early-stop bug. |
| 2237 | */ |
| 2238 | case 'F': f |= f_fixup; break; |
| 2239 | |
| 2240 | /* `-P FILE' (undocumented): trace progress state to FILE. */ |
| 2241 | case 'P': |
| 2242 | if (progressfp) bail("progress trace file already set"); |
| 2243 | progressfp = fopen(optarg, "w"); |
| 2244 | if (!progressfp) |
| 2245 | bail_syserr(errno, "failed to open progress trace file `%s'", |
| 2246 | optarg); |
| 2247 | break; |
| 2248 | |
| 2249 | /* `-R FILE': Read ranges to retry from FILE. Retry ranges are |
| 2250 | * converted into `EV_WRITE' and `EV_STOP' events. |
| 2251 | */ |
| 2252 | case 'R': |
| 2253 | fp = fopen(optarg, "r"); |
| 2254 | if (!fp) |
| 2255 | bail_syserr(errno, "failed to open ranges file `%s'", optarg); |
| 2256 | |
| 2257 | /* We're going to try to coalesce adjacent ranges from the file. |
| 2258 | * When we found a region to skip, we'd have stopped at the a file |
| 2259 | * boundary, and possibly restarted again immediately afterwards, |
| 2260 | * resulting in two adjacent regions in the file. To do that, and |
| 2261 | * also to police the restriction that ranges occur in ascending |
| 2262 | * order, we keep track of the upper bound for the most recent range |
| 2263 | * -- but there isn't one yet, so we use a sentinel value. |
| 2264 | */ |
| 2265 | i = 0; last = -1; |
| 2266 | for (;;) { |
| 2267 | |
| 2268 | /* Read a line from the buffer. If there's nothing left then we're |
| 2269 | * done. |
| 2270 | */ |
| 2271 | buf_rewind(&buf); if (read_line(fp, &buf)) break; |
| 2272 | |
| 2273 | /* Increment the line counter and establish a cursor. */ |
| 2274 | i++; p = buf.p; |
| 2275 | |
| 2276 | /* Skip initial whitespace. */ |
| 2277 | while (ISSPACE(*p)) p++; |
| 2278 | |
| 2279 | /* If this is a comment then ignore it and go round again. */ |
| 2280 | if (!*p || *p == '#') continue; |
| 2281 | |
| 2282 | /* Parse the range. Check that the ranges are coming out in |
| 2283 | * ascending order. |
| 2284 | */ |
| 2285 | if (parse_range(p, 0, &start, &end) || |
| 2286 | (last <= SECLIMIT && start < last)) |
| 2287 | bail("bad range `%s' at `%s' line %zu", buf.p, optarg, i); |
| 2288 | |
| 2289 | /* Ignore empty ranges: this is important (see below where we sort |
| 2290 | * the event queue). If this abuts the previous range then just |
| 2291 | * overwrite the previous end position. Otherwise, write a new |
| 2292 | * pair of events. |
| 2293 | */ |
| 2294 | if (start < end) { |
| 2295 | if (start == last) |
| 2296 | eventq.v[eventq.n - 1].pos = end; |
| 2297 | else { |
| 2298 | put_event(EV_WRITE, 0, start); |
| 2299 | put_event(EV_STOP, 0, end); |
| 2300 | } |
| 2301 | last = end; |
| 2302 | } |
| 2303 | } |
| 2304 | |
| 2305 | /* Check for read errors. */ |
| 2306 | if (ferror(fp)) |
| 2307 | bail_syserr(errno, "failed to read ranges file `%s'", optarg); |
| 2308 | f |= f_retry; |
| 2309 | break; |
| 2310 | |
| 2311 | /* `-X FILE' (undocumented): Read ranges of bad-blocks from FILE to |
| 2312 | * establish fake bad blocks: see `read_sectors' above for the details. |
| 2313 | * |
| 2314 | * This is very similar to the `-R' option above, except that it |
| 2315 | * doesn't do the range coalescing thing. |
| 2316 | */ |
| 2317 | case 'X': |
| 2318 | fp = fopen(optarg, "r"); |
| 2319 | if (!fp) |
| 2320 | bail_syserr(errno, "failed to open bad-blocks file `%s'", optarg); |
| 2321 | i = 0; last = -1; |
| 2322 | for (;;) { |
| 2323 | buf_rewind(&buf); if (read_line(fp, &buf)) break; |
| 2324 | p = buf.p; i++; |
| 2325 | while (ISSPACE(*p)) p++; |
| 2326 | if (!*p || *p == '#') continue; |
| 2327 | if (parse_range(p, 0, &start, &end) || |
| 2328 | (last <= SECLIMIT && start < last)) |
| 2329 | bail("bad range `%s' at `%s' line %zu", buf.p, optarg, i); |
| 2330 | if (start < end) { |
| 2331 | VEC_PUSH(bad, &badblocks); |
| 2332 | bad->start = start; bad->end = end; |
| 2333 | } |
| 2334 | } |
| 2335 | if (ferror(fp)) |
| 2336 | bail_syserr(errno, "failed to read bad-blocks file `%s'", optarg); |
| 2337 | break; |
| 2338 | |
| 2339 | /* Log regions skipped because of bad blocks to a file. */ |
| 2340 | case 'b': |
| 2341 | if (mapfile) bail("can't have multiple map files"); |
| 2342 | mapfile = optarg; |
| 2343 | break; |
| 2344 | |
| 2345 | /* `-c': Continue copying where we left off last time. */ |
| 2346 | case 'c': f |= f_continue; break; |
| 2347 | |
| 2348 | /* `-i': Check that we're copying from the right disc. */ |
| 2349 | case 'i': f |= f_checkid; break; |
| 2350 | |
| 2351 | /* `-r [START]-[END]': Manually provide a range of sectors to retry. */ |
| 2352 | case 'r': |
| 2353 | start = 0; end = -1; f |= f_retry; |
| 2354 | if (parse_range(optarg, PRF_HYPHEN, &start, &end)) |
| 2355 | bail("bad range `%s'", optarg); |
| 2356 | if (start < end) { |
| 2357 | /* Again, ignore empty ranges. */ |
| 2358 | put_event(EV_WRITE, 0, start); |
| 2359 | if (end <= SECLIMIT) put_event(EV_STOP, 0, end); |
| 2360 | } |
| 2361 | break; |
| 2362 | |
| 2363 | /* `-s': Print statistics at the end. */ |
| 2364 | case 's': f |= f_stats; break; |
| 2365 | |
| 2366 | /* Anything else is an error. */ |
| 2367 | default: f |= f_bogus; break; |
| 2368 | } |
| 2369 | } |
| 2370 | |
| 2371 | /* We expect two arguments. Check this. Complain about bad usage if we |
| 2372 | * have bad arguments or options. |
| 2373 | */ |
| 2374 | if (argc - optind != 2) f |= f_bogus; |
| 2375 | if (f&f_bogus) { usage(stderr); exit(2); } |
| 2376 | device = argv[optind]; outfile = argv[optind + 1]; |
| 2377 | |
| 2378 | /* If there are fake bad blocks (the `-X' option) then sort the list |
| 2379 | * because `read_sectors' wants to use a binary search. |
| 2380 | */ |
| 2381 | if (badblocks.n) { |
| 2382 | qsort(badblocks.v, badblocks.n, sizeof(struct badblock), |
| 2383 | compare_badblock); |
| 2384 | #ifdef DEBUG |
| 2385 | printf(";; fake bad blocks:\n"); |
| 2386 | for (i = 0; i < badblocks.n; i++) |
| 2387 | printf(";;\t%8"PRIuSEC" .. %"PRIuSEC"\n", |
| 2388 | badblocks.v[i].start, badblocks.v[i].end); |
| 2389 | #endif |
| 2390 | } |
| 2391 | |
| 2392 | /* Prepare to display progress information. */ |
| 2393 | setlocale(LC_ALL, ""); |
| 2394 | progress_init(&progress); |
| 2395 | |
| 2396 | /* Open the input device. (This may pop up a notice if there's nothing in |
| 2397 | * the drive.) |
| 2398 | */ |
| 2399 | if (open_dvd(device, O_RDONLY, &dvdfd, &dvd)) exit(2); |
| 2400 | |
| 2401 | /* Determine the size of the input device and check the sector size. */ |
| 2402 | blksz = -1; volsz = device_size(dvdfd, device, &blksz); |
| 2403 | if (blksz == -1) |
| 2404 | { blksz = SECTORSZ; f |= f_file; } |
| 2405 | else if (blksz != SECTORSZ) |
| 2406 | bail("device `%s' block size %d /= %d", device, blksz, SECTORSZ); |
| 2407 | if (volsz%SECTORSZ) |
| 2408 | bail("device `%s' volume size %"PRIu64" not a multiple of %d", |
| 2409 | device, volsz, SECTORSZ); |
| 2410 | |
| 2411 | setup_geometry(&geom, dvdfd, f&f_file ? 0 : GF_BLKDEV, volsz/blksz); |
| 2412 | |
| 2413 | if (progressfp) { |
| 2414 | switch (geom.shape) { |
| 2415 | case FLAT: |
| 2416 | fprintf(progressfp, ":model flat-model\n"); |
| 2417 | break; |
| 2418 | case SINGLE: |
| 2419 | fprintf(progressfp, ":model single-layer-model :start %"PRIuSEC"\n", |
| 2420 | geom.start0); |
| 2421 | break; |
| 2422 | case PTP: |
| 2423 | fprintf(progressfp, |
| 2424 | ":model parallel-track-path-model " |
| 2425 | ":start0 %"PRIuSEC" :start1 %"PRIuSEC" " |
| 2426 | ":midpoint %"PRIuSEC"\n", |
| 2427 | geom.start0, geom.start1, geom.midpoint); |
| 2428 | break; |
| 2429 | case OTP: |
| 2430 | fprintf(progressfp, |
| 2431 | ":model opposite-track-path-model " |
| 2432 | ":start %"PRIuSEC" :midpoint %"PRIuSEC"\n", |
| 2433 | geom.start0, geom.midpoint); |
| 2434 | break; |
| 2435 | default: |
| 2436 | abort(); |
| 2437 | } |
| 2438 | } |
| 2439 | |
| 2440 | /* Maybe check that we're copying from the right disc. This is intended to |
| 2441 | * help avoid image corruption by from the wrong disc, but it obviously |
| 2442 | * only works if the output file is mostly there. |
| 2443 | */ |
| 2444 | if (f&f_checkid) { |
| 2445 | if (open_dvd(outfile, O_RDONLY, 0, &dvd_out)) exit(2); |
| 2446 | if (dvd_id(id_in, dvd, DIF_MUSTIFOHASH, device) || |
| 2447 | dvd_id(id_out, dvd_out, DIF_MUSTIFOHASH, device)) |
| 2448 | exit(2); |
| 2449 | if (STRCMP(id_in, !=, id_out)) |
| 2450 | bail("DVD id mismatch: input `%s' is `%s'; output `%s' is `%s'", |
| 2451 | device, id_in, outfile, id_out); |
| 2452 | } |
| 2453 | |
| 2454 | /* Open the output file. */ |
| 2455 | outfd = open(outfile, O_WRONLY | O_CREAT, 0666); |
| 2456 | if (outfd < 0) |
| 2457 | bail_syserr(errno, "failed to create output file `%s'", outfile); |
| 2458 | |
| 2459 | if (f&f_continue) { |
| 2460 | /* If we're continuing from where we left off, then find out where that |
| 2461 | * was and make a note to copy from there to the end of the disc. Note |
| 2462 | * that we're not relying on this position: in particular, it might not |
| 2463 | * even be sector-aligned (in which case we'll ignore the final partial |
| 2464 | * sector). We'll seek to the right place again when we start writing. |
| 2465 | */ |
| 2466 | |
| 2467 | off = lseek(outfd, 0, SEEK_END); |
| 2468 | if (off < 0) |
| 2469 | bail_syserr(errno, "failed to seek to end of output file `%s'", |
| 2470 | outfile); |
| 2471 | put_event(EV_WRITE, 0, off/SECTORSZ); f |= f_retry; |
| 2472 | } |
| 2473 | |
| 2474 | if (!(f&(f_retry | f_fixup))) { |
| 2475 | /* If there are no ranges to retry and we're not fixing an ancient early- |
| 2476 | * stop bug, then there's no range to retry and we should just copy |
| 2477 | * everything. |
| 2478 | */ |
| 2479 | |
| 2480 | put_event(EV_WRITE, 0, 0); |
| 2481 | } |
| 2482 | |
| 2483 | /* Now it's time to figure out what the input looks like. Work through the |
| 2484 | * titlesets in order, mapping out where the video-object files are. We |
| 2485 | * could figure out how many there are properly, but it's fast enough just |
| 2486 | * to try everything. That's the menu only for the special titleset 0, and |
| 2487 | * menu and titles for the remaining titlesets 1 up to 99. |
| 2488 | */ |
| 2489 | put_menu(dvd, 0); |
| 2490 | for (i = 1; i < 100; i++) { |
| 2491 | put_menu(dvd, i); |
| 2492 | put_title(dvd, i); |
| 2493 | } |
| 2494 | |
| 2495 | /* Make a final virtual file for the raw device. (See `emit', which |
| 2496 | * assumes that this is the last entry in the file table.) Check that we |
| 2497 | * don't have more files than we expect, because the bitmap table has fixed |
| 2498 | * size. |
| 2499 | */ |
| 2500 | put_file(mkident(RAW, 0, 0), 0, volsz/SECTORSZ); |
| 2501 | assert(filetab.n <= MAXFILES); |
| 2502 | |
| 2503 | /* Find an upper limit for what we're supposed to copy. Since the `RAW' |
| 2504 | * entry covers the reported size of the input device, this ought to cover |
| 2505 | * all of our bases. |
| 2506 | */ |
| 2507 | for (i = 0, limit = 0; i < filetab.n; i++) |
| 2508 | if (filetab.v[i].end > limit) limit = filetab.v[i].end; |
| 2509 | #ifdef DEBUG |
| 2510 | printf("\n;; files:\n"); |
| 2511 | for (i = 0; i < filetab.n; i++) { |
| 2512 | file = &filetab.v[i]; |
| 2513 | store_filename(fn, file->id); |
| 2514 | printf(";;\t%8"PRIuSEC" .. %-8"PRIuSEC" %s\n", |
| 2515 | file->start, file->end, fn); |
| 2516 | } |
| 2517 | #endif |
| 2518 | |
| 2519 | /* Sort the event list. |
| 2520 | * |
| 2521 | * The event-code ordering is important here. |
| 2522 | * |
| 2523 | * * `EV_STOP' sorts /before/ `EV_WRITE'. If we have two abutting ranges |
| 2524 | * to retry, then we should stop at the end of the first, and then |
| 2525 | * immediately start again. If empty ranges were permitted then we'd |
| 2526 | * stop writing and /then/ start, continuing forever, which is clearly |
| 2527 | * wrong. |
| 2528 | * |
| 2529 | * * `EV_BEGIN' sorts before `EV_END'. If we have empty files then we |
| 2530 | * should set the bit that indicates that it's started, and then clear |
| 2531 | * it, in that order. If we have abutting files, then we'll just both |
| 2532 | * bits for an instant, but that's not a problem. |
| 2533 | */ |
| 2534 | qsort(eventq.v, eventq.n, sizeof(struct event), compare_event); |
| 2535 | |
| 2536 | /* Check that the event list is well-formed. We start out at the |
| 2537 | * beginning, not writing anything. |
| 2538 | */ |
| 2539 | for (i = 0, f &= ~f_write, start = 0; i < eventq.n; i++) { |
| 2540 | ev = &eventq.v[i]; |
| 2541 | switch (ev->ev) { |
| 2542 | |
| 2543 | case EV_WRITE: |
| 2544 | /* Start writing. We shouldn't be writing yet! */ |
| 2545 | |
| 2546 | if (f&f_write) |
| 2547 | bail("overlapping ranges: range from %"PRIuSEC" " |
| 2548 | "still open at %"PRIuSEC"", |
| 2549 | start, ev->pos); |
| 2550 | f |= f_write; start = ev->pos; |
| 2551 | break; |
| 2552 | |
| 2553 | case EV_STOP: |
| 2554 | /* Stop writing. Make a note that we've done this. */ |
| 2555 | |
| 2556 | f &= ~f_write; |
| 2557 | break; |
| 2558 | } |
| 2559 | } |
| 2560 | #ifdef DEBUG |
| 2561 | dump_eventq("initial"); |
| 2562 | #endif |
| 2563 | |
| 2564 | /* Now we make a second pass over the event queue to fix it up. Also |
| 2565 | * count up how much work we'll be doing so that we can report progress. |
| 2566 | */ |
| 2567 | for (i = 0, f &= ~f_write, start = last = 0; i < eventq.n; i++) { |
| 2568 | ev = &eventq.v[i]; |
| 2569 | |
| 2570 | /* If we're supposed to start writing then make a note of the start |
| 2571 | * position. We'll want this to count up how much work we're doing. The |
| 2572 | * start position of the final range is also used by the logic below that |
| 2573 | * determines the progress display. |
| 2574 | */ |
| 2575 | if (ev->ev == EV_WRITE) { start = ev->pos; f |= f_write; } |
| 2576 | |
| 2577 | /* If this event position is past our final limit then stop. Nothing |
| 2578 | * beyond here can possibly be interesting. (Since `EV_WRITE' sorts |
| 2579 | * before other events, we will notice an `EV_WRITE' exactly at the limit |
| 2580 | * sector, but not any other kind of event.) |
| 2581 | */ |
| 2582 | if (ev->pos >= limit) break; |
| 2583 | |
| 2584 | /* If we're supposed to stop writing here, then add the size of the |
| 2585 | * most recent range onto our running total. |
| 2586 | */ |
| 2587 | if (ev->ev == EV_STOP) { |
| 2588 | nsectors += ev->pos - start; |
| 2589 | total_linear += linear_progress(&geom, start, ev->pos); |
| 2590 | f &= ~f_write; |
| 2591 | } |
| 2592 | |
| 2593 | /* If we're fixing up images affected by the old early-stop bug, then |
| 2594 | * remember this position. |
| 2595 | */ |
| 2596 | if (f&f_fixup) last = ev->pos; |
| 2597 | } |
| 2598 | |
| 2599 | /* Truncate the event queue at the point we reached the sector limit. */ |
| 2600 | eventq.n = i; |
| 2601 | #ifdef DEBUG |
| 2602 | dump_eventq("trimmed"); |
| 2603 | #endif |
| 2604 | |
| 2605 | /* Finally, the early-stop bug fix. |
| 2606 | * |
| 2607 | * The bug was caused by a broken version of the event-queue truncation |
| 2608 | * logic: it trimmed the event queue, but didn't add a final event at the |
| 2609 | * file limit. The effect was that the interval between the last event -- |
| 2610 | * likely `EV_END' for a VOB file -- and the overall end of the disc didn't |
| 2611 | * get copied. We address this by starting to write at the position of |
| 2612 | * this last event. |
| 2613 | */ |
| 2614 | if (f&f_fixup) { |
| 2615 | put_event(EV_WRITE, 0, last); |
| 2616 | f |= f_write; |
| 2617 | } |
| 2618 | |
| 2619 | /* If we're still writing then avoid the early-end bug by adding an |
| 2620 | * `EV_STOP' event at the limit position. Include this range in the sector |
| 2621 | * count. |
| 2622 | */ |
| 2623 | if (f&f_write) { |
| 2624 | nsectors += limit - start; |
| 2625 | total_linear += linear_progress(&geom, start, limit); |
| 2626 | put_event(EV_STOP, 0, limit); |
| 2627 | } |
| 2628 | #ifdef DEBUG |
| 2629 | dump_eventq("final"); |
| 2630 | #endif |
| 2631 | |
| 2632 | /* Set up the main progress display. |
| 2633 | * |
| 2634 | * If we're copying a single region from somewhere to the end of the disc |
| 2635 | * then it seems more sensible to use a single progress bar for both. If |
| 2636 | * we're reading multiple ranges, maybe because we're retrying bad blocks, |
| 2637 | * then it's better to have separate bars for how much actual copying we've |
| 2638 | * done, and which part of the disc we're currently working on. |
| 2639 | */ |
| 2640 | copy_progress.render = render_copy_progress; |
| 2641 | progress_additem(&progress, ©_progress); |
| 2642 | if (nsectors == limit - start) { |
| 2643 | ndone = start; nsectors = limit; |
| 2644 | done_linear = linear_progress(&geom, 0, start); |
| 2645 | total_linear += done_linear; |
| 2646 | } |
| 2647 | else { |
| 2648 | disc_progress.render = render_disc_progress; |
| 2649 | progress_additem(&progress, &disc_progress); |
| 2650 | } |
| 2651 | |
| 2652 | /* If we're producing overall statistics then make a note of the current |
| 2653 | * time. |
| 2654 | */ |
| 2655 | if (f&f_stats) gettimeofday(&tv0, 0); |
| 2656 | |
| 2657 | /* We're now ready to start our sweep through the disc. */ |
| 2658 | #ifdef DEBUG |
| 2659 | printf("\n;; event sweep:\n"); |
| 2660 | #endif |
| 2661 | |
| 2662 | /* We start at the beginning of the disc, and the start of the event queue, |
| 2663 | * not writing. We'll advance through the events one by one. |
| 2664 | */ |
| 2665 | for (pos = 0, i = 0, f &= ~f_write; i < eventq.n; i++) { |
| 2666 | |
| 2667 | /* Get the next event. */ |
| 2668 | ev = &eventq.v[i]; |
| 2669 | |
| 2670 | /* If there's a nonempty range between here and the previous event then |
| 2671 | * we need to process this. |
| 2672 | */ |
| 2673 | if (ev->pos > pos) { |
| 2674 | |
| 2675 | /* If we're writing then copy the interval from the previous event to |
| 2676 | * here to the output. |
| 2677 | */ |
| 2678 | if (f&f_write) emit(pos, ev->pos); |
| 2679 | |
| 2680 | /* Advance the current position now that the output is up-to-date. */ |
| 2681 | pos = ev->pos; |
| 2682 | |
| 2683 | #ifdef DEBUG |
| 2684 | progress_clear(&progress); |
| 2685 | printf(";;\n"); |
| 2686 | #endif |
| 2687 | } |
| 2688 | |
| 2689 | /* Decide what to action to take in response to the event. */ |
| 2690 | switch (ev->ev) { |
| 2691 | |
| 2692 | case EV_BEGIN: |
| 2693 | /* A file has started. Set the appropriate bit in the active-files |
| 2694 | * map. |
| 2695 | */ |
| 2696 | set_live(ev->file); |
| 2697 | #ifdef DEBUG |
| 2698 | store_filename(fn, filetab.v[ev->file].id); |
| 2699 | progress_clear(&progress); |
| 2700 | printf(";; %8"PRIuSEC": begin `%s'\n", pos, fn); |
| 2701 | #endif |
| 2702 | break; |
| 2703 | |
| 2704 | case EV_WRITE: |
| 2705 | /* We're supposed to start writing. */ |
| 2706 | |
| 2707 | /* Note the current time and position for the progress display. */ |
| 2708 | gettimeofday(&last_time, 0); last_pos = pos; |
| 2709 | |
| 2710 | /* Seek to the right place in the output file. */ |
| 2711 | if (lseek(outfd, (off_t)ev->pos*SECTORSZ, SEEK_SET) < 0) |
| 2712 | bail_syserr(errno, |
| 2713 | "failed to seek to resume position " |
| 2714 | "(sector %"PRIuSEC") in output file `%s'", |
| 2715 | ev->pos, outfile); |
| 2716 | |
| 2717 | /* Engage the write head. */ |
| 2718 | f |= f_write; |
| 2719 | |
| 2720 | #ifdef DEBUG |
| 2721 | progress_clear(&progress); |
| 2722 | printf(";; %8"PRIuSEC": begin write\n", pos); |
| 2723 | #endif |
| 2724 | break; |
| 2725 | |
| 2726 | case EV_STOP: |
| 2727 | /* We're supposed to stop writing. Disengage the write head. */ |
| 2728 | |
| 2729 | f &= ~f_write; |
| 2730 | #ifdef DEBUG |
| 2731 | progress_clear(&progress); |
| 2732 | printf(";; %8"PRIuSEC": end write\n", pos); |
| 2733 | #endif |
| 2734 | break; |
| 2735 | |
| 2736 | case EV_END: |
| 2737 | /* We've found the end of a file. Clear its bit in the table. */ |
| 2738 | |
| 2739 | clear_live(ev->file); |
| 2740 | #ifdef DEBUG |
| 2741 | store_filename(fn, filetab.v[ev->file].id); |
| 2742 | progress_clear(&progress); |
| 2743 | printf(";; %8"PRIuSEC": end `%s'\n", pos, fn); |
| 2744 | #endif |
| 2745 | break; |
| 2746 | |
| 2747 | /* Something else. Clearly a bug. */ |
| 2748 | default: abort(); |
| 2749 | } |
| 2750 | } |
| 2751 | |
| 2752 | /* Take down the progress display because we're done. */ |
| 2753 | progress_clear(&progress); |
| 2754 | |
| 2755 | /* Set the output file length correctly. */ |
| 2756 | if (ftruncate(outfd, (off_t)limit*SECTORSZ) < 0) |
| 2757 | bail_syserr(errno, "failed to set output file `%s' length", outfile); |
| 2758 | |
| 2759 | /* Report overall statistics. */ |
| 2760 | if (f&f_stats) { |
| 2761 | gettimeofday(&tv1, 0); t = tvdiff(&tv0, &tv1); |
| 2762 | if (nsectors == limit) { ndone -= start; nsectors -= start; } |
| 2763 | tot = scale_bytes((double)nsectors*SECTORSZ, &totunit); |
| 2764 | rate = scale_bytes((double)nsectors*SECTORSZ/t, &rateunit); |
| 2765 | moan("all done: %.1f %sB in %s -- %.1f %sB/s", |
| 2766 | tot, totunit, fmttime(t, timebuf), rate, rateunit); |
| 2767 | } |
| 2768 | |
| 2769 | /* Close files. */ |
| 2770 | if (dvd) DVDClose(dvd); |
| 2771 | if (dvdfd >= 0) close(dvdfd); |
| 2772 | if (outfd >= 0) close(outfd); |
| 2773 | carefully_fclose(mapfp, "bad-sector region map"); |
| 2774 | carefully_fclose(errfp, "bad-sector error log"); |
| 2775 | carefully_fclose(progressfp, "progress trace file"); |
| 2776 | progress_free(&progress); |
| 2777 | |
| 2778 | /* We're done! */ |
| 2779 | return (0); |
| 2780 | |
| 2781 | #undef f_bogus |
| 2782 | #undef f_continue |
| 2783 | #undef f_fixup |
| 2784 | #undef f_stats |
| 2785 | #undef f_write |
| 2786 | } |
| 2787 | |
| 2788 | /*----- That's all, folks -------------------------------------------------*/ |