read-cache.c

   1 /*
   2  * GIT - The information manager from hell
   3  *
   4  * Copyright (C) Linus Torvalds, 2005
   5  */
   6 #include "cache.h"
   7
   8 struct cache_entry **active_cache = NULL;
   9 unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
  10
  11 /*
  12  * This only updates the "non-critical" parts of the directory
  13  * cache, ie the parts that aren't tracked by GIT, and only used
  14  * to validate the cache.
  15  */
  16 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  17 {
  18         ce->ce_ctime.sec = htonl(st->st_ctime);
  19         ce->ce_mtime.sec = htonl(st->st_mtime);
  20 #ifdef USE_NSEC
  21         ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
  22         ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
  23 #endif
  24         ce->ce_dev = htonl(st->st_dev);
  25         ce->ce_ino = htonl(st->st_ino);
  26         ce->ce_uid = htonl(st->st_uid);
  27         ce->ce_gid = htonl(st->st_gid);
  28         ce->ce_size = htonl(st->st_size);
  29 }
  30
  31 int ce_match_stat(struct cache_entry *ce, struct stat *st)
  32 {
  33         unsigned int changed = 0;
  34
  35         switch (ntohl(ce->ce_mode) & S_IFMT) {
  36         case S_IFREG:
  37                 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
  38                 /* We consider only the owner x bit to be relevant for "mode changes" */
  39                 if (0100 & (ntohl(ce->ce_mode) ^ st->st_mode))
  40                         changed |= MODE_CHANGED;
  41                 break;
  42         case S_IFLNK:
  43                 changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
  44                 break;
  45         default:
  46                 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
  47         }
  48         if (ce->ce_mtime.sec != htonl(st->st_mtime))
  49                 changed |= MTIME_CHANGED;
  50         if (ce->ce_ctime.sec != htonl(st->st_ctime))
  51                 changed |= CTIME_CHANGED;
  52
  53 #ifdef USE_NSEC
  54         /*
  55          * nsec seems unreliable - not all filesystems support it, so
  56          * as long as it is in the inode cache you get right nsec
  57          * but after it gets flushed, you get zero nsec.
  58          */
  59         if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
  60                 changed |= MTIME_CHANGED;
  61         if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
  62                 changed |= CTIME_CHANGED;
  63 #endif
  64
  65         if (ce->ce_uid != htonl(st->st_uid) ||
  66             ce->ce_gid != htonl(st->st_gid))
  67                 changed |= OWNER_CHANGED;
  68         if (ce->ce_ino != htonl(st->st_ino))
  69                 changed |= INODE_CHANGED;
  70
  71 #ifdef USE_STDEV
  72         /*
  73          * st_dev breaks on network filesystems where different
  74          * clients will have different views of what "device"
  75          * the filesystem is on
  76          */
  77         if (ce->ce_dev != htonl(st->st_dev))
  78                 changed |= INODE_CHANGED;
  79 #endif
  80
  81         if (ce->ce_size != htonl(st->st_size))
  82                 changed |= DATA_CHANGED;
  83         return changed;
  84 }
  85
  86 int base_name_compare(const char *name1, int len1, int mode1,
  87                       const char *name2, int len2, int mode2)
  88 {
  89         unsigned char c1, c2;
  90         int len = len1 < len2 ? len1 : len2;
  91         int cmp;
  92
  93         cmp = memcmp(name1, name2, len);
  94         if (cmp)
  95                 return cmp;
  96         c1 = name1[len];
  97         c2 = name2[len];
  98         if (!c1 && S_ISDIR(mode1))
  99                 c1 = '/';
 100         if (!c2 && S_ISDIR(mode2))
 101                 c2 = '/';
 102         return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
 103 }
 104
 105 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
 106 {
 107         int len1 = flags1 & CE_NAMEMASK;
 108         int len2 = flags2 & CE_NAMEMASK;
 109         int len = len1 < len2 ? len1 : len2;
 110         int cmp;
 111
 112         cmp = memcmp(name1, name2, len);
 113         if (cmp)
 114                 return cmp;
 115         if (len1 < len2)
 116                 return -1;
 117         if (len1 > len2)
 118                 return 1;
 119         if (flags1 < flags2)
 120                 return -1;
 121         if (flags1 > flags2)
 122                 return 1;
 123         return 0;
 124 }
 125
 126 int cache_name_pos(const char *name, int namelen)
 127 {
 128         int first, last;
 129
 130         first = 0;
 131         last = active_nr;
 132         while (last > first) {
 133                 int next = (last + first) >> 1;
 134                 struct cache_entry *ce = active_cache[next];
 135                 int cmp = cache_name_compare(name, namelen, ce->name, ntohs(ce->ce_flags));
 136                 if (!cmp)
 137                         return next;
 138                 if (cmp < 0) {
 139                         last = next;
 140                         continue;
 141                 }
 142                 first = next+1;
 143         }
 144         return -first-1;
 145 }
 146
 147 /* Remove entry, return true if there are more entries to go.. */
 148 int remove_cache_entry_at(int pos)
 149 {
 150         active_cache_changed = 1;
 151         active_nr--;
 152         if (pos >= active_nr)
 153                 return 0;
 154         memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
 155         return 1;
 156 }
 157
 158 int remove_file_from_cache(char *path)
 159 {
 160         int pos = cache_name_pos(path, strlen(path));
 161         if (pos < 0)
 162                 pos = -pos-1;
 163         while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
 164                 remove_cache_entry_at(pos);
 165         return 0;
 166 }
 167
 168 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 169 {
 170         int len = ce_namelen(a);
 171         return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 172 }
 173
 174 int ce_path_match(const struct cache_entry *ce, const char **pathspec)
 175 {
 176         const char *match, *name;
 177         int len;
 178
 179         if (!pathspec)
 180                 return 1;
 181
 182         len = ce_namelen(ce);
 183         name = ce->name;
 184         while ((match = *pathspec++) != NULL) {
 185                 int matchlen = strlen(match);
 186                 if (matchlen > len)
 187                         continue;
 188                 if (memcmp(name, match, matchlen))
 189                         continue;
 190                 if (matchlen && name[matchlen-1] == '/')
 191                         return 1;
 192                 if (name[matchlen] == '/' || !name[matchlen])
 193                         return 1;
 194         }
 195         return 0;
 196 }
 197
 198 /*
 199  * Do we have another file that has the beginning components being a
 200  * proper superset of the name we're trying to add?
 201  */
 202 static int has_file_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 203 {
 204         int retval = 0;
 205         int len = ce_namelen(ce);
 206         int stage = ce_stage(ce);
 207         const char *name = ce->name;
 208
 209         while (pos < active_nr) {
 210                 struct cache_entry *p = active_cache[pos++];
 211
 212                 if (len >= ce_namelen(p))
 213                         break;
 214                 if (memcmp(name, p->name, len))
 215                         break;
 216                 if (ce_stage(p) != stage)
 217                         continue;
 218                 if (p->name[len] != '/')
 219                         continue;
 220                 retval = -1;
 221                 if (!ok_to_replace)
 222                         break;
 223                 remove_cache_entry_at(--pos);
 224         }
 225         return retval;
 226 }
 227
 228 /*
 229  * Do we have another file with a pathname that is a proper
 230  * subset of the name we're trying to add?
 231  */
 232 static int has_dir_name(const struct cache_entry *ce, int pos, int ok_to_replace)
 233 {
 234         int retval = 0;
 235         int stage = ce_stage(ce);
 236         const char *name = ce->name;
 237         const char *slash = name + ce_namelen(ce);
 238
 239         for (;;) {
 240                 int len;
 241
 242                 for (;;) {
 243                         if (*--slash == '/')
 244                                 break;
 245                         if (slash <= ce->name)
 246                                 return retval;
 247                 }
 248                 len = slash - name;
 249
 250                 pos = cache_name_pos(name, ntohs(create_ce_flags(len, stage)));
 251                 if (pos >= 0) {
 252                         retval = -1;
 253                         if (ok_to_replace)
 254                                 break;
 255                         remove_cache_entry_at(pos);
 256                         continue;
 257                 }
 258
 259                 /*
 260                  * Trivial optimization: if we find an entry that
 261                  * already matches the sub-directory, then we know
 262                  * we're ok, and we can exit.
 263                  */
 264                 pos = -pos-1;
 265                 while (pos < active_nr) {
 266                         struct cache_entry *p = active_cache[pos];
 267                         if ((ce_namelen(p) <= len) ||
 268                             (p->name[len] != '/') ||
 269                             memcmp(p->name, name, len))
 270                                 break; /* not our subdirectory */
 271                         if (ce_stage(p) == stage)
 272                                 /* p is at the same stage as our entry, and
 273                                  * is a subdirectory of what we are looking
 274                                  * at, so we cannot have conflicts at our
 275                                  * level or anything shorter.
 276                                  */
 277                                 return retval;
 278                         pos++;
 279                 }
 280         }
 281         return retval;
 282 }
 283
 284 /* We may be in a situation where we already have path/file and path
 285  * is being added, or we already have path and path/file is being
 286  * added.  Either one would result in a nonsense tree that has path
 287  * twice when git-write-tree tries to write it out.  Prevent it.
 288  *
 289  * If ok-to-replace is specified, we remove the conflicting entries
 290  * from the cache so the caller should recompute the insert position.
 291  * When this happens, we return non-zero.
 292  */
 293 static int check_file_directory_conflict(const struct cache_entry *ce, int pos, int ok_to_replace)
 294 {
 295         /*
 296          * We check if the path is a sub-path of a subsequent pathname
 297          * first, since removing those will not change the position
 298          * in the array
 299          */
 300         int retval = has_file_name(ce, pos, ok_to_replace);
 301         /*
 302          * Then check if the path might have a clashing sub-directory
 303          * before it.
 304          */
 305         return retval + has_dir_name(ce, pos, ok_to_replace);
 306 }
 307
 308 int add_cache_entry(struct cache_entry *ce, int option)
 309 {
 310         int pos;
 311         int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 312         int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 313         int skip_df_check = option & ADD_CACHE_SKIP_DFCHECK;
 314         pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 315
 316         /* existing match? Just replace it */
 317         if (pos >= 0) {
 318                 active_cache_changed = 1;
 319                 active_cache[pos] = ce;
 320                 return 0;
 321         }
 322         pos = -pos-1;
 323
 324         /*
 325          * Inserting a merged entry ("stage 0") into the index
 326          * will always replace all non-merged entries..
 327          */
 328         if (pos < active_nr && ce_stage(ce) == 0) {
 329                 while (ce_same_name(active_cache[pos], ce)) {
 330                         ok_to_add = 1;
 331                         if (!remove_cache_entry_at(pos))
 332                                 break;
 333                 }
 334         }
 335
 336         if (!ok_to_add)
 337                 return -1;
 338
 339         if (!skip_df_check && check_file_directory_conflict(ce, pos, ok_to_replace)) {
 340                 if (!ok_to_replace)
 341                         return -1;
 342                 pos = cache_name_pos(ce->name, ntohs(ce->ce_flags));
 343                 pos = -pos-1;
 344         }
 345
 346         /* Make sure the array is big enough .. */
 347         if (active_nr == active_alloc) {
 348                 active_alloc = alloc_nr(active_alloc);
 349                 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
 350         }
 351
 352         /* Add it in.. */
 353         active_nr++;
 354         if (active_nr > pos)
 355                 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
 356         active_cache[pos] = ce;
 357         active_cache_changed = 1;
 358         return 0;
 359 }
 360
 361 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 362 {
 363         SHA_CTX c;
 364         unsigned char sha1[20];
 365
 366         if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 367                 return error("bad signature");
 368         if (hdr->hdr_version != htonl(2))
 369                 return error("bad index version");
 370         SHA1_Init(&c);
 371         SHA1_Update(&c, hdr, size - 20);
 372         SHA1_Final(sha1, &c);
 373         if (memcmp(sha1, (void *)hdr + size - 20, 20))
 374                 return error("bad index file sha1 signature");
 375         return 0;
 376 }
 377
 378 int read_cache(void)
 379 {
 380         int fd, i;
 381         struct stat st;
 382         unsigned long size, offset;
 383         void *map;
 384         struct cache_header *hdr;
 385
 386         errno = EBUSY;
 387         if (active_cache)
 388                 return error("more than one cachefile");
 389         errno = ENOENT;
 390         fd = open(get_index_file(), O_RDONLY);
 391         if (fd < 0)
 392                 return (errno == ENOENT) ? 0 : error("open failed");
 393
 394         size = 0; // avoid gcc warning
 395         map = (void *)-1;
 396         if (!fstat(fd, &st)) {
 397                 size = st.st_size;
 398                 errno = EINVAL;
 399                 if (size >= sizeof(struct cache_header) + 20)
 400                         map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
 401         }
 402         close(fd);
 403         if (-1 == (int)(long)map)
 404                 return error("mmap failed");
 405
 406         hdr = map;
 407         if (verify_hdr(hdr, size) < 0)
 408                 goto unmap;
 409
 410         active_nr = ntohl(hdr->hdr_entries);
 411         active_alloc = alloc_nr(active_nr);
 412         active_cache = calloc(active_alloc, sizeof(struct cache_entry *));
 413
 414         offset = sizeof(*hdr);
 415         for (i = 0; i < active_nr; i++) {
 416                 struct cache_entry *ce = map + offset;
 417                 offset = offset + ce_size(ce);
 418                 active_cache[i] = ce;
 419         }
 420         return active_nr;
 421
 422 unmap:
 423         munmap(map, size);
 424         errno = EINVAL;
 425         return error("verify header failed");
 426 }
 427
 428 #define WRITE_BUFFER_SIZE 8192
 429 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 430 static unsigned long write_buffer_len;
 431
 432 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 433 {
 434         while (len) {
 435                 unsigned int buffered = write_buffer_len;
 436                 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
 437                 if (partial > len)
 438                         partial = len;
 439                 memcpy(write_buffer + buffered, data, partial);
 440                 buffered += partial;
 441                 if (buffered == WRITE_BUFFER_SIZE) {
 442                         SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
 443                         if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
 444                                 return -1;
 445                         buffered = 0;
 446                 }
 447                 write_buffer_len = buffered;
 448                 len -= partial;
 449                 data += partial;
 450         }
 451         return 0;
 452 }
 453
 454 static int ce_flush(SHA_CTX *context, int fd)
 455 {
 456         unsigned int left = write_buffer_len;
 457
 458         if (left) {
 459                 write_buffer_len = 0;
 460                 SHA1_Update(context, write_buffer, left);
 461         }
 462
 463         /* Append the SHA1 signature at the end */
 464         SHA1_Final(write_buffer + left, context);
 465         left += 20;
 466         if (write(fd, write_buffer, left) != left)
 467                 return -1;
 468         return 0;
 469 }
 470
 471 int write_cache(int newfd, struct cache_entry **cache, int entries)
 472 {
 473         SHA_CTX c;
 474         struct cache_header hdr;
 475         int i, removed;
 476
 477         for (i = removed = 0; i < entries; i++)
 478                 if (!cache[i]->ce_mode)
 479                         removed++;
 480
 481         hdr.hdr_signature = htonl(CACHE_SIGNATURE);
 482         hdr.hdr_version = htonl(2);
 483         hdr.hdr_entries = htonl(entries - removed);
 484
 485         SHA1_Init(&c);
 486         if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
 487                 return -1;
 488
 489         for (i = 0; i < entries; i++) {
 490                 struct cache_entry *ce = cache[i];
 491                 if (!ce->ce_mode)
 492                         continue;
 493                 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
 494                         return -1;
 495         }
 496         return ce_flush(&c, newfd);
 497 }