1 /*- 2 * Copyright (c) 2003-2010 Tim Kientzle 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR(S) ``AS IS'' AND ANY EXPRESS OR 15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 17 * IN NO EVENT SHALL THE AUTHOR(S) BE LIABLE FOR ANY DIRECT, INDIRECT, 18 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 19 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 20 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 21 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 22 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 23 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 */ 25 26 #include "archive_platform.h" 27 __FBSDID("$FreeBSD: head/lib/libarchive/archive_read_open_filename.c 201093 2009-12-28 02:28:44Z kientzle $"); 28 29 #ifdef HAVE_SYS_IOCTL_H 30 #include <sys/ioctl.h> 31 #endif 32 #ifdef HAVE_SYS_STAT_H 33 #include <sys/stat.h> 34 #endif 35 #ifdef HAVE_ERRNO_H 36 #include <errno.h> 37 #endif 38 #ifdef HAVE_FCNTL_H 39 #include <fcntl.h> 40 #endif 41 #ifdef HAVE_IO_H 42 #include <io.h> 43 #endif 44 #ifdef HAVE_STDLIB_H 45 #include <stdlib.h> 46 #endif 47 #ifdef HAVE_STRING_H 48 #include <string.h> 49 #endif 50 #ifdef HAVE_UNISTD_H 51 #include <unistd.h> 52 #endif 53 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 54 #include <sys/disk.h> 55 #elif defined(__NetBSD__) || defined(__OpenBSD__) 56 #include <sys/disklabel.h> 57 #include <sys/dkio.h> 58 #elif defined(__DragonFly__) 59 #include <sys/diskslice.h> 60 #endif 61 62 #include "archive.h" 63 #include "archive_string.h" 64 65 #ifndef O_BINARY 66 #define O_BINARY 0 67 #endif 68 69 struct read_file_data { 70 int fd; 71 size_t block_size; 72 void *buffer; 73 mode_t st_mode; /* Mode bits for opened file. */ 74 char use_lseek; 75 enum fnt_e { FNT_STDIN, FNT_MBS, FNT_WCS } filename_type; 76 union { 77 char m[1];/* MBS filename. */ 78 wchar_t w[1];/* WCS filename. */ 79 } filename; /* Must be last! */ 80 }; 81 82 static int file_close(struct archive *, void *); 83 static int file_open_filename(struct archive *, enum fnt_e, const void *, 84 size_t); 85 static ssize_t file_read(struct archive *, void *, const void **buff); 86 static int64_t file_seek(struct archive *, void *, int64_t request, int); 87 static int64_t file_skip(struct archive *, void *, int64_t request); 88 static int64_t file_skip_lseek(struct archive *, void *, int64_t request); 89 90 int 91 archive_read_open_file(struct archive *a, const char *filename, 92 size_t block_size) 93 { 94 return (archive_read_open_filename(a, filename, block_size)); 95 } 96 97 int 98 archive_read_open_filename(struct archive *a, const char *filename, 99 size_t block_size) 100 { 101 enum fnt_e filename_type; 102 103 if (filename == NULL || filename[0] == '\0') { 104 filename_type = FNT_STDIN; 105 } else 106 filename_type = FNT_MBS; 107 return (file_open_filename(a, filename_type, filename, block_size)); 108 } 109 110 int 111 archive_read_open_filename_w(struct archive *a, const wchar_t *wfilename, 112 size_t block_size) 113 { 114 enum fnt_e filename_type; 115 116 if (wfilename == NULL || wfilename[0] == L'\0') { 117 filename_type = FNT_STDIN; 118 } else { 119 #if defined(_WIN32) && !defined(__CYGWIN__) 120 filename_type = FNT_WCS; 121 #else 122 /* 123 * POSIX system does not support a wchar_t interface for 124 * open() system call, so we have to translate a whcar_t 125 * filename to multi-byte one and use it. 126 */ 127 struct archive_string fn; 128 int r; 129 130 archive_string_init(&fn); 131 if (archive_string_append_from_wcs(&fn, wfilename, 132 wcslen(wfilename)) != 0) { 133 archive_set_error(a, EINVAL, 134 "Failed to convert a wide-character filename to" 135 " a multi-byte filename"); 136 archive_string_free(&fn); 137 return (ARCHIVE_FATAL); 138 } 139 r = file_open_filename(a, FNT_MBS, fn.s, block_size); 140 archive_string_free(&fn); 141 return (r); 142 #endif 143 } 144 return (file_open_filename(a, filename_type, wfilename, block_size)); 145 } 146 147 static int 148 file_open_filename(struct archive *a, enum fnt_e filename_type, 149 const void *_filename, size_t block_size) 150 { 151 struct stat st; 152 struct read_file_data *mine; 153 void *buffer; 154 const char *filename = NULL; 155 const wchar_t *wfilename = NULL; 156 int fd; 157 int is_disk_like = 0; 158 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 159 off_t mediasize = 0; /* FreeBSD-specific, so off_t okay here. */ 160 #elif defined(__NetBSD__) || defined(__OpenBSD__) 161 struct disklabel dl; 162 #elif defined(__DragonFly__) 163 struct partinfo pi; 164 #endif 165 166 archive_clear_error(a); 167 if (filename_type == FNT_STDIN) { 168 /* We used to delegate stdin support by 169 * directly calling archive_read_open_fd(a,0,block_size) 170 * here, but that doesn't (and shouldn't) handle the 171 * end-of-file flush when reading stdout from a pipe. 172 * Basically, read_open_fd() is intended for folks who 173 * are willing to handle such details themselves. This 174 * API is intended to be a little smarter for folks who 175 * want easy handling of the common case. 176 */ 177 fd = 0; 178 #if defined(__CYGWIN__) || defined(_WIN32) 179 setmode(0, O_BINARY); 180 #endif 181 filename = ""; 182 } else if (filename_type == FNT_MBS) { 183 filename = (const char *)_filename; 184 fd = open(filename, O_RDONLY | O_BINARY); 185 if (fd < 0) { 186 archive_set_error(a, errno, 187 "Failed to open '%s'", filename); 188 return (ARCHIVE_FATAL); 189 } 190 } else { 191 #if defined(_WIN32) && !defined(__CYGWIN__) 192 wfilename = (const wchar_t *)_filename; 193 fd = _wopen(wfilename, O_RDONLY | O_BINARY); 194 if (fd < 0 && errno == ENOENT) { 195 wchar_t *fullpath; 196 fullpath = __la_win_permissive_name_w(wfilename); 197 if (fullpath != NULL) { 198 fd = _wopen(fullpath, O_RDONLY | O_BINARY); 199 free(fullpath); 200 } 201 } 202 if (fd < 0) { 203 archive_set_error(a, errno, 204 "Failed to open '%S'", wfilename); 205 return (ARCHIVE_FATAL); 206 } 207 #else 208 archive_set_error(a, ARCHIVE_ERRNO_MISC, 209 "Unexpedted operation in archive_read_open_filename"); 210 return (ARCHIVE_FATAL); 211 #endif 212 } 213 if (fstat(fd, &st) != 0) { 214 if (filename_type == FNT_WCS) 215 archive_set_error(a, errno, "Can't stat '%S'", 216 wfilename); 217 else 218 archive_set_error(a, errno, "Can't stat '%s'", 219 filename); 220 return (ARCHIVE_FATAL); 221 } 222 223 /* 224 * Determine whether the input looks like a disk device or a 225 * tape device. The results are used below to select an I/O 226 * strategy: 227 * = "disk-like" devices support arbitrary lseek() and will 228 * support I/O requests of any size. So we get easy skipping 229 * and can cheat on block sizes to get better performance. 230 * = "tape-like" devices require strict blocking and use 231 * specialized ioctls for seeking. 232 * = "socket-like" devices cannot seek at all but can improve 233 * performance by using nonblocking I/O to read "whatever is 234 * available right now". 235 * 236 * Right now, we only specially recognize disk-like devices, 237 * but it should be straightforward to add probes and strategy 238 * here for tape-like and socket-like devices. 239 */ 240 if (S_ISREG(st.st_mode)) { 241 /* Safety: Tell the extractor not to overwrite the input. */ 242 archive_read_extract_set_skip_file(a, st.st_dev, st.st_ino); 243 /* Regular files act like disks. */ 244 is_disk_like = 1; 245 } 246 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 247 /* FreeBSD: if it supports DIOCGMEDIASIZE ioctl, it's disk-like. */ 248 else if (S_ISCHR(st.st_mode) && 249 ioctl(fd, DIOCGMEDIASIZE, &mediasize) == 0 && 250 mediasize > 0) { 251 is_disk_like = 1; 252 } 253 #elif defined(__NetBSD__) || defined(__OpenBSD__) 254 /* Net/OpenBSD: if it supports DIOCGDINFO ioctl, it's disk-like. */ 255 else if ((S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) && 256 ioctl(fd, DIOCGDINFO, &dl) == 0 && 257 dl.d_partitions[DISKPART(st.st_rdev)].p_size > 0) { 258 is_disk_like = 1; 259 } 260 #elif defined(__DragonFly__) 261 /* DragonFly BSD: if it supports DIOCGPART ioctl, it's disk-like. */ 262 else if (S_ISCHR(st.st_mode) && 263 ioctl(fd, DIOCGPART, &pi) == 0 && 264 pi.media_size > 0) { 265 is_disk_like = 1; 266 } 267 #elif defined(__linux__) 268 /* Linux: All block devices are disk-like. */ 269 else if (S_ISBLK(st.st_mode) && 270 lseek(fd, 0, SEEK_CUR) == 0 && 271 lseek(fd, 0, SEEK_SET) == 0 && 272 lseek(fd, 0, SEEK_END) > 0 && 273 lseek(fd, 0, SEEK_SET) == 0) { 274 is_disk_like = 1; 275 } 276 #endif 277 /* TODO: Add an "is_tape_like" variable and appropriate tests. */ 278 279 if (filename_type == FNT_WCS) 280 mine = (struct read_file_data *)calloc(1, 281 sizeof(*mine) + wcslen(wfilename) * sizeof(wchar_t)); 282 else 283 mine = (struct read_file_data *)calloc(1, 284 sizeof(*mine) + strlen(filename)); 285 /* Disk-like devices prefer power-of-two block sizes. */ 286 /* Use provided block_size as a guide so users have some control. */ 287 if (is_disk_like) { 288 size_t new_block_size = 64 * 1024; 289 while (new_block_size < block_size 290 && new_block_size < 64 * 1024 * 1024) 291 new_block_size *= 2; 292 block_size = new_block_size; 293 } 294 buffer = malloc(block_size); 295 if (mine == NULL || buffer == NULL) { 296 archive_set_error(a, ENOMEM, "No memory"); 297 free(mine); 298 free(buffer); 299 return (ARCHIVE_FATAL); 300 } 301 if (filename_type == FNT_WCS) 302 wcscpy(mine->filename.w, wfilename); 303 else 304 strcpy(mine->filename.m, filename); 305 mine->filename_type = filename_type; 306 mine->block_size = block_size; 307 mine->buffer = buffer; 308 mine->fd = fd; 309 /* Remember mode so close can decide whether to flush. */ 310 mine->st_mode = st.st_mode; 311 312 /* Disk-like inputs can use lseek(). */ 313 if (is_disk_like) { 314 archive_read_set_seek_callback(a, file_seek); 315 mine->use_lseek = 1; 316 } 317 318 archive_read_set_read_callback(a, file_read); 319 archive_read_set_skip_callback(a, file_skip); 320 archive_read_set_close_callback(a, file_close); 321 archive_read_set_callback_data(a, mine); 322 return (archive_read_open1(a)); 323 } 324 325 static ssize_t 326 file_read(struct archive *a, void *client_data, const void **buff) 327 { 328 struct read_file_data *mine = (struct read_file_data *)client_data; 329 ssize_t bytes_read; 330 331 /* TODO: If a recent lseek() operation has left us 332 * mis-aligned, read and return a short block to try to get 333 * us back in alignment. */ 334 335 /* TODO: Someday, try mmap() here; if that succeeds, give 336 * the entire file to libarchive as a single block. That 337 * could be a lot faster than block-by-block manual I/O. */ 338 339 /* TODO: We might be able to improve performance on pipes and 340 * sockets by setting non-blocking I/O and just accepting 341 * whatever we get here instead of waiting for a full block 342 * worth of data. */ 343 344 *buff = mine->buffer; 345 for (;;) { 346 bytes_read = read(mine->fd, mine->buffer, mine->block_size); 347 if (bytes_read < 0) { 348 if (errno == EINTR) 349 continue; 350 else if (mine->filename_type == FNT_STDIN) 351 archive_set_error(a, errno, 352 "Error reading stdin"); 353 else if (mine->filename_type == FNT_MBS) 354 archive_set_error(a, errno, 355 "Error reading '%s'", mine->filename.m); 356 else 357 archive_set_error(a, errno, 358 "Error reading '%S'", mine->filename.w); 359 } 360 return (bytes_read); 361 } 362 } 363 364 /* 365 * Regular files and disk-like block devices can use simple lseek 366 * without needing to round the request to the block size. 367 * 368 * TODO: This can leave future reads mis-aligned. Since we know the 369 * offset here, we should store it and use it in file_read() above 370 * to determine whether we should perform a short read to get back 371 * into alignment. Long series of mis-aligned reads can negatively 372 * impact disk throughput. (Of course, the performance impact should 373 * be carefully tested; extra code complexity is only worthwhile if 374 * it does provide measurable improvement.) 375 * 376 * TODO: Be lazy about the actual seek. There are a few pathological 377 * cases where libarchive makes a bunch of seek requests in a row 378 * without any intervening reads. This isn't a huge performance 379 * problem, since the kernel handles seeks lazily already, but 380 * it would be very slightly faster if we simply remembered the 381 * seek request here and then actually performed the seek at the 382 * top of the read callback above. 383 */ 384 static int64_t 385 file_skip_lseek(struct archive *a, void *client_data, int64_t request) 386 { 387 struct read_file_data *mine = (struct read_file_data *)client_data; 388 #if defined(_WIN32) && !defined(__CYGWIN__) 389 /* We use _lseeki64() on Windows. */ 390 int64_t old_offset, new_offset; 391 #else 392 off_t old_offset, new_offset; 393 #endif 394 395 /* We use off_t here because lseek() is declared that way. */ 396 397 /* TODO: Deal with case where off_t isn't 64 bits. 398 * This shouldn't be a problem on Linux or other POSIX 399 * systems, since the configuration logic for libarchive 400 * tries to obtain a 64-bit off_t. It's still an issue 401 * on Windows, though, so it might suffice to just use 402 * _lseeki64() on Windows. 403 */ 404 if ((old_offset = lseek(mine->fd, 0, SEEK_CUR)) >= 0 && 405 (new_offset = lseek(mine->fd, request, SEEK_CUR)) >= 0) 406 return (new_offset - old_offset); 407 408 /* If lseek() fails, don't bother trying again. */ 409 mine->use_lseek = 0; 410 411 /* Let libarchive recover with read+discard */ 412 if (errno == ESPIPE) 413 return (0); 414 415 /* If the input is corrupted or truncated, fail. */ 416 if (mine->filename_type == FNT_STDIN) 417 archive_set_error(a, errno, "Error seeking in stdin"); 418 else if (mine->filename_type == FNT_MBS) 419 archive_set_error(a, errno, "Error seeking in '%s'", 420 mine->filename.m); 421 else 422 archive_set_error(a, errno, "Error seeking in '%S'", 423 mine->filename.w); 424 return (-1); 425 } 426 427 428 /* 429 * TODO: Implement another file_skip_XXXX that uses MTIO ioctls to 430 * accelerate operation on tape drives. 431 */ 432 433 static int64_t 434 file_skip(struct archive *a, void *client_data, int64_t request) 435 { 436 struct read_file_data *mine = (struct read_file_data *)client_data; 437 438 /* Delegate skip requests. */ 439 if (mine->use_lseek) 440 return (file_skip_lseek(a, client_data, request)); 441 442 /* If we can't skip, return 0; libarchive will read+discard instead. */ 443 return (0); 444 } 445 446 /* 447 * TODO: Store the offset and use it in the read callback. 448 */ 449 static int64_t 450 file_seek(struct archive *a, void *client_data, int64_t request, int whence) 451 { 452 struct read_file_data *mine = (struct read_file_data *)client_data; 453 off_t r; 454 455 /* We use off_t here because lseek() is declared that way. */ 456 /* See above for notes about when off_t is less than 64 bits. */ 457 r = lseek(mine->fd, request, whence); 458 if (r >= 0) 459 return r; 460 461 /* If the input is corrupted or truncated, fail. */ 462 if (mine->filename_type == FNT_STDIN) 463 archive_set_error(a, errno, "Error seeking in stdin"); 464 else if (mine->filename_type == FNT_MBS) 465 archive_set_error(a, errno, "Error seeking in '%s'", 466 mine->filename.m); 467 else 468 archive_set_error(a, errno, "Error seeking in '%S'", 469 mine->filename.w); 470 return (ARCHIVE_FATAL); 471 } 472 473 static int 474 file_close(struct archive *a, void *client_data) 475 { 476 struct read_file_data *mine = (struct read_file_data *)client_data; 477 478 (void)a; /* UNUSED */ 479 480 /* Only flush and close if open succeeded. */ 481 if (mine->fd >= 0) { 482 /* 483 * Sometimes, we should flush the input before closing. 484 * Regular files: faster to just close without flush. 485 * Disk-like devices: Ditto. 486 * Tapes: must not flush (user might need to 487 * read the "next" item on a non-rewind device). 488 * Pipes and sockets: must flush (otherwise, the 489 * program feeding the pipe or socket may complain). 490 * Here, I flush everything except for regular files and 491 * device nodes. 492 */ 493 if (!S_ISREG(mine->st_mode) 494 && !S_ISCHR(mine->st_mode) 495 && !S_ISBLK(mine->st_mode)) { 496 ssize_t bytesRead; 497 do { 498 bytesRead = read(mine->fd, mine->buffer, 499 mine->block_size); 500 } while (bytesRead > 0); 501 } 502 /* If a named file was opened, then it needs to be closed. */ 503 if (mine->filename_type != FNT_STDIN) 504 close(mine->fd); 505 } 506 free(mine->buffer); 507 free(mine); 508 return (ARCHIVE_OK); 509 } 510