whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2000 Silicon Graphics, Inc. All Rights Reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify it |
| 5 | * under the terms of version 2 of the GNU General Public License as |
| 6 | * published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it would be useful, but |
| 9 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. |
| 11 | * |
| 12 | * Further, this software is distributed without any warranty that it is |
| 13 | * free of the rightful claim of any third person regarding infringement |
| 14 | * or the like. Any license provided herein, whether implied or |
| 15 | * otherwise, applies only to this software file. Patent licenses, if |
| 16 | * any, provided herein do not apply to combinations of this program with |
| 17 | * other software, or any other product whatsoever. |
| 18 | * |
| 19 | * You should have received a copy of the GNU General Public License along |
| 20 | * with this program; if not, write the Free Software Foundation, Inc., 59 |
| 21 | * Temple Place - Suite 330, Boston MA 02111-1307, USA. |
| 22 | * |
| 23 | * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy, |
| 24 | * Mountain View, CA 94043, or: |
| 25 | * |
| 26 | * http://www.sgi.com |
| 27 | * |
| 28 | * For further information regarding this notice, see: |
| 29 | * |
| 30 | * http://oss.sgi.com/projects/GenInfo/NoticeExplan/ |
| 31 | */ |
| 32 | /* |
| 33 | * This module contains code for logging writes to files, and for |
| 34 | * perusing the resultant logfile. The main intent of all this is |
| 35 | * to provide a 'write history' of a file which can be examined to |
| 36 | * judge the state of a file (ie. whether it is corrupted or not) based |
| 37 | * on the write activity. |
| 38 | * |
| 39 | * The main abstractions available to the user are the wlog_file, and |
| 40 | * the wlog_rec. A wlog_file is a handle encapsulating a write logfile. |
| 41 | * It is initialized with the wlog_open() function. This handle is |
| 42 | * then passed to the various wlog_xxx() functions to provide transparent |
| 43 | * access to the write logfile. |
| 44 | * |
| 45 | * The wlog_rec datatype is a structure which contains all the information |
| 46 | * about a file write. Examples include the file name, offset, length, |
| 47 | * pattern, etc. In addition there is a bit which is cleared/set based |
| 48 | * on whether or not the write has been confirmed as complete. This |
| 49 | * allows the write logfile to contain information on writes which have |
| 50 | * been initiated, but not yet completed (as in async io). |
| 51 | * |
| 52 | * There is also a function to scan a write logfile in reverse order. |
| 53 | * |
| 54 | * NOTE: For target file analysis based on a write logfile, the |
| 55 | * assumption is made that the file being written to is |
| 56 | * locked from simultaneous access, so that the order of |
| 57 | * write completion is predictable. This is an issue when |
| 58 | * more than 1 process is trying to write data to the same |
| 59 | * target file simultaneously. |
| 60 | * |
| 61 | * The history file created is a collection of variable length records |
| 62 | * described by scruct wlog_rec_disk in write_log.h. See that module for |
| 63 | * the layout of the data on disk. |
| 64 | */ |
| 65 | |
| 66 | #include <stdio.h> |
| 67 | #include <unistd.h> |
| 68 | #include <fcntl.h> |
| 69 | #include <errno.h> |
| 70 | #include <string.h> |
| 71 | #include <sys/param.h> |
| 72 | #include <sys/stat.h> |
| 73 | #include <sys/types.h> |
| 74 | #include "write_log.h" |
| 75 | |
| 76 | #ifndef BSIZE |
| 77 | #ifdef linux |
| 78 | #define BSIZE DEV_BSIZE |
| 79 | #else |
| 80 | #define BSIZE BBSIZE |
| 81 | #endif |
| 82 | #endif |
| 83 | |
| 84 | #ifndef PATH_MAX |
| 85 | #define PATH_MAX 255 |
| 86 | /*#define PATH_MAX pathconf("/", _PC_PATH_MAX)*/ |
| 87 | #endif |
| 88 | |
whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 89 | char Wlog_Error_String[256]; |
| 90 | |
| 91 | #if __STDC__ |
| 92 | static int wlog_rec_pack(struct wlog_rec *wrec, char *buf, int flag); |
| 93 | static int wlog_rec_unpack(struct wlog_rec *wrec, char *buf); |
| 94 | #else |
| 95 | static int wlog_rec_pack(); |
| 96 | static int wlog_rec_unpack(); |
| 97 | #endif |
| 98 | |
| 99 | /* |
| 100 | * Initialize a write logfile. wfile is a wlog_file structure that has |
| 101 | * the w_file field filled in. The rest of the information in the |
| 102 | * structure is initialized by the routine. |
| 103 | * |
| 104 | * The trunc flag is used to indicate whether or not the logfile should |
| 105 | * be truncated if it currently exists. If it is non-zero, the file will |
| 106 | * be truncated, otherwise it will be appended to. |
| 107 | * |
| 108 | * The mode argument is the [absolute] mode which the file will be |
| 109 | * given if it does not exist. This mode is not affected by your process |
| 110 | * umask. |
| 111 | */ |
| 112 | |
| 113 | int |
| 114 | wlog_open(wfile, trunc, mode) |
| 115 | struct wlog_file *wfile; |
| 116 | int trunc; |
| 117 | int mode; |
| 118 | { |
| 119 | int omask, oflags; |
| 120 | |
| 121 | if (trunc) |
| 122 | trunc = O_TRUNC; |
| 123 | |
| 124 | omask = umask(0); |
| 125 | |
| 126 | /* |
| 127 | * Open 1 file descriptor as O_APPEND |
| 128 | */ |
| 129 | |
| 130 | oflags = O_WRONLY | O_APPEND | O_CREAT | trunc; |
| 131 | wfile->w_afd = |
| 132 | open(wfile->w_file, oflags, mode); |
| 133 | umask(omask); |
| 134 | |
| 135 | if (wfile->w_afd == -1) { |
| 136 | sprintf(Wlog_Error_String, |
| 137 | "Could not open write_log - open(%s, %#o, %#o) failed: %s\n", |
nstraz | 9418108 | 2000-08-30 18:43:38 +0000 | [diff] [blame] | 138 | wfile->w_file, oflags, mode, strerror(errno)); |
whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 139 | return -1; |
| 140 | } |
| 141 | |
| 142 | /* |
| 143 | * Open the next fd as a random access descriptor |
| 144 | */ |
| 145 | |
| 146 | oflags = O_RDWR; |
| 147 | if ((wfile->w_rfd = open(wfile->w_file, oflags)) == -1) { |
| 148 | sprintf(Wlog_Error_String, |
| 149 | "Could not open write log - open(%s, %#o) failed: %s\n", |
nstraz | 9418108 | 2000-08-30 18:43:38 +0000 | [diff] [blame] | 150 | wfile->w_file, oflags, strerror(errno)); |
whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 151 | close(wfile->w_afd); |
| 152 | wfile->w_afd = -1; |
| 153 | return -1; |
| 154 | } |
| 155 | |
| 156 | return 0; |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * Release all resources associated with a wlog_file structure allocated |
| 161 | * with the wlog_open() call. |
| 162 | */ |
| 163 | |
| 164 | int |
| 165 | wlog_close(wfile) |
| 166 | struct wlog_file *wfile; |
| 167 | { |
| 168 | close(wfile->w_afd); |
| 169 | close(wfile->w_rfd); |
| 170 | return 0; |
| 171 | } |
| 172 | |
| 173 | /* |
| 174 | * Write a wlog_rec structure to a write logfile. Offset is used to |
| 175 | * control where the record will be written. If offset is < 0, the |
| 176 | * record will be appended to the end of the logfile. Otherwise, the |
| 177 | * record which exists at the indicated offset will be overlayed. This |
| 178 | * is so that we can record writes which are outstanding (with the w_done |
| 179 | * bit in wrec cleared), but not completed, and then later update the |
| 180 | * logfile when the write request completes (as with async io). When |
| 181 | * offset is >= 0, only the fixed length portion of the record is |
| 182 | * rewritten. See text in write_log.h for details on the format of an |
| 183 | * on-disk record. |
| 184 | * |
| 185 | * The return value of the function is the byte offset in the logfile |
| 186 | * where the record begins. |
| 187 | * |
| 188 | * Note: It is the callers responsibility to make sure that the offset |
| 189 | * parameter 'points' to a valid record location when a record is to be |
| 190 | * overlayed. This is guarenteed by saving the return value of a previous |
| 191 | * call to wlog_record_write() which wrote the record to be overlayed. |
| 192 | * |
| 193 | * Note2: The on-disk version of the wlog_rec is MUCH different than |
| 194 | * the user version. Don't expect to od the logfile and see data formatted |
| 195 | * as it is in the wlog_rec structure. Considerable data packing takes |
| 196 | * place before the record is written. |
| 197 | */ |
| 198 | |
| 199 | int |
| 200 | wlog_record_write(wfile, wrec, offset) |
| 201 | struct wlog_file *wfile; |
| 202 | struct wlog_rec *wrec; |
| 203 | long offset; |
| 204 | { |
| 205 | int reclen; |
| 206 | char wbuf[WLOG_REC_MAX_SIZE + 2]; |
| 207 | |
| 208 | /* |
| 209 | * If offset is -1, we append the record at the end of file |
| 210 | * |
| 211 | * Otherwise, we overlay wrec at the file offset indicated and assume |
| 212 | * that the caller passed us the correct offset. We do not record the |
| 213 | * fname in this case. |
| 214 | */ |
| 215 | |
| 216 | reclen = wlog_rec_pack(wrec, wbuf, (offset < 0)); |
| 217 | |
| 218 | if (offset < 0) { |
| 219 | /* |
| 220 | * Since we're writing a complete new record, we must also tack |
| 221 | * its length onto the end so that wlog_scan_backward() will work. |
| 222 | * Length is asumed to fit into 2 bytes. |
| 223 | */ |
| 224 | |
| 225 | wbuf[reclen] = reclen / 256; |
| 226 | wbuf[reclen+1] = reclen % 256; |
| 227 | reclen += 2; |
| 228 | |
| 229 | write(wfile->w_afd, wbuf, reclen); |
| 230 | offset = lseek(wfile->w_afd, 0, SEEK_CUR) - reclen; |
| 231 | } else { |
| 232 | lseek(wfile->w_rfd, offset, SEEK_SET); |
| 233 | write(wfile->w_rfd, wbuf, reclen); |
| 234 | } |
| 235 | |
| 236 | return offset; |
| 237 | } |
| 238 | |
| 239 | /* |
| 240 | * Function to scan a logfile in reverse order. Wfile is a valid |
| 241 | * wlog_file structure initialized by wlog_open(). nrecs is the number |
| 242 | * of records to scan (all records are scanned if nrecs is 0). func is |
| 243 | * a user-supplied function to call for each record found. The function |
| 244 | * will be passed a single parameter - a wlog_rec structure . |
| 245 | */ |
| 246 | |
| 247 | int |
| 248 | wlog_scan_backward(wfile, nrecs, func, data) |
| 249 | struct wlog_file *wfile; |
| 250 | int nrecs; |
| 251 | int (*func)(); |
| 252 | long data; |
| 253 | { |
| 254 | int fd, leftover, nbytes, offset, recnum, reclen, rval; |
| 255 | char buf[BSIZE*32], *bufend, *cp, *bufstart; |
nstraz | 9418108 | 2000-08-30 18:43:38 +0000 | [diff] [blame] | 256 | char albuf[WLOG_REC_MAX_SIZE]; |
whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 257 | struct wlog_rec wrec; |
| 258 | |
| 259 | fd = wfile->w_rfd; |
| 260 | |
| 261 | /* |
| 262 | * Move to EOF. offset will always hold the current file offset |
| 263 | */ |
| 264 | |
| 265 | lseek(fd, 0, SEEK_END); |
| 266 | offset = lseek(fd, 0, SEEK_CUR); |
| 267 | |
| 268 | bufend = buf + sizeof(buf); |
| 269 | bufstart = buf; |
| 270 | |
| 271 | recnum = 0; |
| 272 | leftover = 0; |
| 273 | while ((!nrecs || recnum < nrecs) && offset > 0) { |
| 274 | /* |
| 275 | * Check for beginning of file - if there aren't enough bytes |
| 276 | * remaining to fill buf, adjust bufstart. |
| 277 | */ |
| 278 | |
| 279 | if (offset + leftover < sizeof(buf)) { |
| 280 | bufstart = bufend - (offset + leftover); |
| 281 | offset = 0; |
| 282 | } else { |
| 283 | offset -= sizeof(buf) - leftover; |
| 284 | } |
| 285 | |
| 286 | /* |
| 287 | * Move to the proper file offset, and read into buf |
| 288 | */ |
| 289 | |
| 290 | lseek(fd, offset, SEEK_SET); |
| 291 | nbytes = read(fd, bufstart, bufend - bufstart - leftover); |
| 292 | |
| 293 | if (nbytes == -1) { |
| 294 | sprintf(Wlog_Error_String, |
nstraz | 9418108 | 2000-08-30 18:43:38 +0000 | [diff] [blame] | 295 | "Could not read history file at offset %d - read(%d, %#o, %d) failed: %s\n", |
| 296 | offset, fd, (int)bufstart, |
| 297 | bufend - bufstart - leftover, strerror(errno)); |
whr | b973f2b | 2000-05-05 19:34:50 +0000 | [diff] [blame] | 298 | return -1; |
| 299 | } |
| 300 | |
| 301 | cp = bufend; |
| 302 | leftover = 0; |
| 303 | |
| 304 | while (cp >= bufstart) { |
| 305 | |
| 306 | /* |
| 307 | * If cp-bufstart is not large enough to hold a piece |
| 308 | * of record length information, copy remainder to end |
| 309 | * of buf and continue reading the file. |
| 310 | */ |
| 311 | |
| 312 | if (cp - bufstart < 2) { |
| 313 | leftover = cp - bufstart; |
| 314 | memcpy(bufend - leftover, bufstart, leftover); |
| 315 | break; |
| 316 | } |
| 317 | |
| 318 | /* |
| 319 | * Extract the record length. We must do it this way |
| 320 | * instead of casting cp to an int because cp might |
| 321 | * not be word aligned. |
| 322 | */ |
| 323 | |
| 324 | reclen = (*(cp-2) * 256) + *(cp -1); |
| 325 | |
| 326 | /* |
| 327 | * If cp-bufstart isn't large enough to hold a |
| 328 | * complete record, plus the length information, copy |
| 329 | * the leftover bytes to the end of buf and continue |
| 330 | * reading. |
| 331 | */ |
| 332 | |
| 333 | if (cp - bufstart < reclen + 2) { |
| 334 | leftover = cp - bufstart; |
| 335 | memcpy(bufend - leftover, bufstart, leftover); |
| 336 | break; |
| 337 | } |
| 338 | |
| 339 | /* |
| 340 | * Adjust cp to point at the start of the record. |
| 341 | * Copy the record into wbuf so that it is word |
| 342 | * aligned and pass the record to the user supplied |
| 343 | * function. |
| 344 | */ |
| 345 | |
| 346 | cp -= reclen + 2; |
| 347 | memcpy(albuf, cp, reclen); |
| 348 | |
| 349 | wlog_rec_unpack(&wrec, albuf); |
| 350 | |
| 351 | /* |
| 352 | * Call the user supplied function - |
| 353 | * stop if instructed to. |
| 354 | */ |
| 355 | |
| 356 | if ((rval = (*func)(&wrec, data)) == WLOG_STOP_SCAN) { |
| 357 | break; |
| 358 | } |
| 359 | |
| 360 | recnum++; |
| 361 | |
| 362 | if (nrecs && recnum >= nrecs) |
| 363 | break; |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | return 0; |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * The following 2 routines are used to pack and unpack the user |
| 372 | * visible wlog_rec structure to/from a character buffer which is |
| 373 | * stored or read from the write logfile. Any changes to either of |
| 374 | * these routines must be reflected in the other. |
| 375 | */ |
| 376 | |
| 377 | static int |
| 378 | wlog_rec_pack(wrec, buf, flag) |
| 379 | struct wlog_rec *wrec; |
| 380 | char *buf; |
| 381 | int flag; |
| 382 | { |
| 383 | char *file, *host, *pattern; |
| 384 | struct wlog_rec_disk *wrecd; |
| 385 | |
| 386 | wrecd = (struct wlog_rec_disk *)buf; |
| 387 | |
| 388 | wrecd->w_pid = (uint)wrec->w_pid; |
| 389 | wrecd->w_offset = (uint)wrec->w_offset; |
| 390 | wrecd->w_nbytes = (uint)wrec->w_nbytes; |
| 391 | wrecd->w_oflags = (uint)wrec->w_oflags; |
| 392 | wrecd->w_done = (uint)wrec->w_done; |
| 393 | wrecd->w_async = (uint)wrec->w_async; |
| 394 | |
| 395 | wrecd->w_pathlen = (wrec->w_pathlen > 0) ? (uint)wrec->w_pathlen : 0; |
| 396 | wrecd->w_hostlen = (wrec->w_hostlen > 0) ? (uint)wrec->w_hostlen : 0; |
| 397 | wrecd->w_patternlen = (wrec->w_patternlen > 0) ? (uint)wrec->w_patternlen : 0; |
| 398 | |
| 399 | /* |
| 400 | * If flag is true, we should also pack the variable length parts |
| 401 | * of the wlog_rec. By default, we only pack the fixed length |
| 402 | * parts. |
| 403 | */ |
| 404 | |
| 405 | if (flag) { |
| 406 | file = buf + sizeof(struct wlog_rec_disk); |
| 407 | host = file + wrecd->w_pathlen; |
| 408 | pattern = host + wrecd->w_hostlen; |
| 409 | |
| 410 | if (wrecd->w_pathlen > 0) |
| 411 | memcpy(file, wrec->w_path, wrecd->w_pathlen); |
| 412 | |
| 413 | if (wrecd->w_hostlen > 0) |
| 414 | memcpy(host, wrec->w_host, wrecd->w_hostlen); |
| 415 | |
| 416 | if (wrecd->w_patternlen > 0) |
| 417 | memcpy(pattern, wrec->w_pattern, wrecd->w_patternlen); |
| 418 | |
| 419 | return (sizeof(struct wlog_rec_disk) + |
| 420 | wrecd->w_pathlen + wrecd->w_hostlen + wrecd->w_patternlen); |
| 421 | } else { |
| 422 | return sizeof(struct wlog_rec_disk); |
| 423 | } |
| 424 | } |
| 425 | |
| 426 | static int |
| 427 | wlog_rec_unpack(wrec, buf) |
| 428 | struct wlog_rec *wrec; |
| 429 | char *buf; |
| 430 | { |
| 431 | char *file, *host, *pattern; |
| 432 | struct wlog_rec_disk *wrecd; |
| 433 | |
| 434 | bzero((char *)wrec, sizeof(struct wlog_rec)); |
| 435 | wrecd = (struct wlog_rec_disk *)buf; |
| 436 | |
| 437 | wrec->w_pid = wrecd->w_pid; |
| 438 | wrec->w_offset = wrecd->w_offset; |
| 439 | wrec->w_nbytes = wrecd->w_nbytes; |
| 440 | wrec->w_oflags = wrecd->w_oflags; |
| 441 | wrec->w_hostlen = wrecd->w_hostlen; |
| 442 | wrec->w_pathlen = wrecd->w_pathlen; |
| 443 | wrec->w_patternlen = wrecd->w_patternlen; |
| 444 | wrec->w_done = wrecd->w_done; |
| 445 | wrec->w_async = wrecd->w_async; |
| 446 | |
| 447 | if (wrec->w_pathlen > 0) { |
| 448 | file = buf + sizeof(struct wlog_rec_disk); |
| 449 | memcpy(wrec->w_path, file, wrec->w_pathlen); |
| 450 | } |
| 451 | |
| 452 | if (wrec->w_hostlen > 0) { |
| 453 | host = buf + sizeof(struct wlog_rec_disk) + wrec->w_pathlen; |
| 454 | memcpy(wrec->w_host, host, wrec->w_hostlen); |
| 455 | } |
| 456 | |
| 457 | if (wrec->w_patternlen > 0) { |
| 458 | pattern = buf + sizeof(struct wlog_rec_disk) + |
| 459 | wrec->w_pathlen + wrec->w_hostlen; |
| 460 | memcpy(wrec->w_pattern, pattern, wrec->w_patternlen); |
| 461 | } |
| 462 | |
| 463 | return 0; |
| 464 | } |