read-cache.c

   1 /*
   2  * GIT - The information manager from hell
   3  *
   4  * Copyright (C) Linus Torvalds, 2005
   5  */
   6 #include <stdarg.h>
   7 #include "cache.h"
   8
   9 struct cache_entry **active_cache = NULL;
  10 unsigned int active_nr = 0, active_alloc = 0, active_cache_changed = 0;
  11
  12 /*
  13  * This only updates the "non-critical" parts of the directory
  14  * cache, ie the parts that aren't tracked by GIT, and only used
  15  * to validate the cache.
  16  */
  17 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
  18 {
  19         ce->ce_ctime.sec = htonl(st->st_ctime);
  20         ce->ce_mtime.sec = htonl(st->st_mtime);
  21 #ifdef NSEC
  22         ce->ce_ctime.nsec = htonl(st->st_ctim.tv_nsec);
  23         ce->ce_mtime.nsec = htonl(st->st_mtim.tv_nsec);
  24 #endif
  25         ce->ce_dev = htonl(st->st_dev);
  26         ce->ce_ino = htonl(st->st_ino);
  27         ce->ce_uid = htonl(st->st_uid);
  28         ce->ce_gid = htonl(st->st_gid);
  29         ce->ce_size = htonl(st->st_size);
  30 }
  31
  32 int ce_match_stat(struct cache_entry *ce, struct stat *st)
  33 {
  34         unsigned int changed = 0;
  35
  36         switch (ntohl(ce->ce_mode) & S_IFMT) {
  37         case S_IFREG:
  38                 changed |= !S_ISREG(st->st_mode) ? TYPE_CHANGED : 0;
  39                 /* We consider only the owner x bit to be relevant for "mode changes" */
  40                 if (0100 & (ntohl(ce->ce_mode) ^ st->st_mode))
  41                         changed |= MODE_CHANGED;
  42                 break;
  43         case S_IFLNK:
  44                 changed |= !S_ISLNK(st->st_mode) ? TYPE_CHANGED : 0;
  45                 break;
  46         default:
  47                 die("internal error: ce_mode is %o", ntohl(ce->ce_mode));
  48         }
  49         if (ce->ce_mtime.sec != htonl(st->st_mtime))
  50                 changed |= MTIME_CHANGED;
  51         if (ce->ce_ctime.sec != htonl(st->st_ctime))
  52                 changed |= CTIME_CHANGED;
  53
  54 #ifdef NSEC
  55         /*
  56          * nsec seems unreliable - not all filesystems support it, so
  57          * as long as it is in the inode cache you get right nsec
  58          * but after it gets flushed, you get zero nsec.
  59          */
  60         if (ce->ce_mtime.nsec != htonl(st->st_mtim.tv_nsec))
  61                 changed |= MTIME_CHANGED;
  62         if (ce->ce_ctime.nsec != htonl(st->st_ctim.tv_nsec))
  63                 changed |= CTIME_CHANGED;
  64 #endif
  65
  66         if (ce->ce_uid != htonl(st->st_uid) ||
  67             ce->ce_gid != htonl(st->st_gid))
  68                 changed |= OWNER_CHANGED;
  69         if (ce->ce_dev != htonl(st->st_dev) ||
  70             ce->ce_ino != htonl(st->st_ino))
  71                 changed |= INODE_CHANGED;
  72         if (ce->ce_size != htonl(st->st_size))
  73                 changed |= DATA_CHANGED;
  74         return changed;
  75 }
  76
  77 int base_name_compare(const char *name1, int len1, int mode1,
  78                       const char *name2, int len2, int mode2)
  79 {
  80         unsigned char c1, c2;
  81         int len = len1 < len2 ? len1 : len2;
  82         int cmp;
  83
  84         cmp = memcmp(name1, name2, len);
  85         if (cmp)
  86                 return cmp;
  87         c1 = name1[len];
  88         c2 = name2[len];
  89         if (!c1 && S_ISDIR(mode1))
  90                 c1 = '/';
  91         if (!c2 && S_ISDIR(mode2))
  92                 c2 = '/';
  93         return (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
  94 }
  95
  96 int cache_name_compare(const char *name1, int flags1, const char *name2, int flags2)
  97 {
  98         int len1 = flags1 & CE_NAMEMASK;
  99         int len2 = flags2 & CE_NAMEMASK;
 100         int len = len1 < len2 ? len1 : len2;
 101         int cmp;
 102
 103         cmp = memcmp(name1, name2, len);
 104         if (cmp)
 105                 return cmp;
 106         if (len1 < len2)
 107                 return -1;
 108         if (len1 > len2)
 109                 return 1;
 110         if (flags1 < flags2)
 111                 return -1;
 112         if (flags1 > flags2)
 113                 return 1;
 114         return 0;
 115 }
 116
 117 int cache_name_pos(const char *name, int namelen)
 118 {
 119         int first, last;
 120
 121         first = 0;
 122         last = active_nr;
 123         while (last > first) {
 124                 int next = (last + first) >> 1;
 125                 struct cache_entry *ce = active_cache[next];
 126                 int cmp = cache_name_compare(name, namelen, ce->name, htons(ce->ce_flags));
 127                 if (!cmp)
 128                         return next;
 129                 if (cmp < 0) {
 130                         last = next;
 131                         continue;
 132                 }
 133                 first = next+1;
 134         }
 135         return -first-1;
 136 }
 137
 138 /* Remove entry, return true if there are more entries to go.. */
 139 int remove_cache_entry_at(int pos)
 140 {
 141         active_cache_changed = 1;
 142         active_nr--;
 143         if (pos >= active_nr)
 144                 return 0;
 145         memmove(active_cache + pos, active_cache + pos + 1, (active_nr - pos) * sizeof(struct cache_entry *));
 146         return 1;
 147 }
 148
 149 int remove_file_from_cache(char *path)
 150 {
 151         int pos = cache_name_pos(path, strlen(path));
 152         if (pos < 0)
 153                 pos = -pos-1;
 154         while (pos < active_nr && !strcmp(active_cache[pos]->name, path))
 155                 remove_cache_entry_at(pos);
 156         return 0;
 157 }
 158
 159 int ce_same_name(struct cache_entry *a, struct cache_entry *b)
 160 {
 161         int len = ce_namelen(a);
 162         return ce_namelen(b) == len && !memcmp(a->name, b->name, len);
 163 }
 164
 165 /* We may be in a situation where we already have path/file and path
 166  * is being added, or we already have path and path/file is being
 167  * added.  Either one would result in a nonsense tree that has path
 168  * twice when git-write-tree tries to write it out.  Prevent it.
 169  *
 170  * If ok-to-replace is specified, we remove the conflicting entries
 171  * from the cache so the caller should recompute the insert position.
 172  * When this happens, we return non-zero.
 173  */
 174 static int check_file_directory_conflict(const struct cache_entry *ce,
 175                                          int ok_to_replace)
 176 {
 177         int pos, replaced = 0;
 178         const char *path = ce->name;
 179         int namelen = strlen(path);
 180         int stage = ce_stage(ce);
 181         char *pathbuf = xmalloc(namelen + 1);
 182         char *cp;
 183
 184         memcpy(pathbuf, path, namelen + 1);
 185
 186         /*
 187          * We are inserting path/file.  Do they have path registered at
 188          * the same stage?  We need to do this for all the levels of our
 189          * subpath.
 190          */
 191         cp = pathbuf;
 192         while (1) {
 193                 char *ep = strchr(cp, '/');
 194                 if (!ep)
 195                         break;
 196                 *ep = 0;    /* first cut it at slash */
 197                 pos = cache_name_pos(pathbuf,
 198                                      htons(create_ce_flags(ep-cp, stage)));
 199                 if (0 <= pos) {
 200                         /* Our leading path component is registered as a file,
 201                          * and we are trying to make it a directory.  This is
 202                          * bad.
 203                          */
 204                         if (!ok_to_replace) {
 205                                 free(pathbuf);
 206                                 return -1;
 207                         }
 208                         fprintf(stderr, "removing file '%s' to replace it with a directory to create '%s'.\n", pathbuf, path);
 209                         remove_cache_entry_at(pos);
 210                         replaced = 1;
 211                 }
 212                 *ep = '/';  /* then restore it and go downwards */
 213                 cp = ep + 1;
 214         }
 215         free(pathbuf);
 216
 217         /* Do we have an entry in the cache that makes our path a prefix
 218          * of it?  That is, are we creating a file where they already expect
 219          * a directory there?
 220          */
 221         pos = cache_name_pos(path,
 222                              htons(create_ce_flags(namelen, stage)));
 223
 224         /* (0 <= pos) cannot happen because add_cache_entry()
 225          * should have taken care of that case.
 226          */
 227         pos = -pos-1;
 228
 229         /* pos would point at an existing entry that would come immediately
 230          * after our path.  It could be the same as our path in higher stage,
 231          * or different path but in a lower stage.
 232          *
 233          * E.g. when we are inserting path at stage 2,
 234          *
 235          *        1 path
 236          * pos->  3 path
 237          *        2 path/file1
 238          *        3 path/file1
 239          *        2 path/file2
 240          *        2 patho
 241          *
 242          * We need to examine pos, ignore it because it is at different
 243          * stage, examine next to find the path/file at stage 2, and
 244          * complain.  We need to do this until we are not the leading
 245          * path of an existing entry anymore.
 246          */
 247
 248         while (pos < active_nr) {
 249                 struct cache_entry *other = active_cache[pos];
 250                 if (strncmp(other->name, path, namelen))
 251                         break; /* it is not our "subdirectory" anymore */
 252                 if ((ce_stage(other) == stage) &&
 253                     other->name[namelen] == '/') {
 254                         if (!ok_to_replace)
 255                                 return -1;
 256                         fprintf(stderr, "removing file '%s' under '%s' to be replaced with a file\n", other->name, path);
 257                         remove_cache_entry_at(pos);
 258                         replaced = 1;
 259                         continue; /* cycle without updating pos */
 260                 }
 261                 pos++;
 262         }
 263         return replaced;
 264 }
 265
 266 int add_cache_entry(struct cache_entry *ce, int option)
 267 {
 268         int pos;
 269         int ok_to_add = option & ADD_CACHE_OK_TO_ADD;
 270         int ok_to_replace = option & ADD_CACHE_OK_TO_REPLACE;
 271         pos = cache_name_pos(ce->name, htons(ce->ce_flags));
 272
 273         /* existing match? Just replace it */
 274         if (pos >= 0) {
 275                 active_cache_changed = 1;
 276                 active_cache[pos] = ce;
 277                 return 0;
 278         }
 279         pos = -pos-1;
 280
 281         /*
 282          * Inserting a merged entry ("stage 0") into the index
 283          * will always replace all non-merged entries..
 284          */
 285         if (pos < active_nr && ce_stage(ce) == 0) {
 286                 while (ce_same_name(active_cache[pos], ce)) {
 287                         ok_to_add = 1;
 288                         if (!remove_cache_entry_at(pos))
 289                                 break;
 290                 }
 291         }
 292
 293         if (!ok_to_add)
 294                 return -1;
 295
 296         if (check_file_directory_conflict(ce, ok_to_replace)) {
 297                 if (!ok_to_replace)
 298                         return -1;
 299                 pos = cache_name_pos(ce->name, htons(ce->ce_flags));
 300                 pos = -pos-1;
 301         }
 302
 303         /* Make sure the array is big enough .. */
 304         if (active_nr == active_alloc) {
 305                 active_alloc = alloc_nr(active_alloc);
 306                 active_cache = xrealloc(active_cache, active_alloc * sizeof(struct cache_entry *));
 307         }
 308
 309         /* Add it in.. */
 310         active_nr++;
 311         if (active_nr > pos)
 312                 memmove(active_cache + pos + 1, active_cache + pos, (active_nr - pos - 1) * sizeof(ce));
 313         active_cache[pos] = ce;
 314         active_cache_changed = 1;
 315         return 0;
 316 }
 317
 318 static int verify_hdr(struct cache_header *hdr, unsigned long size)
 319 {
 320         SHA_CTX c;
 321         unsigned char sha1[20];
 322
 323         if (hdr->hdr_signature != htonl(CACHE_SIGNATURE))
 324                 return error("bad signature");
 325         if (hdr->hdr_version != htonl(2))
 326                 return error("bad index version");
 327         SHA1_Init(&c);
 328         SHA1_Update(&c, hdr, size - 20);
 329         SHA1_Final(sha1, &c);
 330         if (memcmp(sha1, (void *)hdr + size - 20, 20))
 331                 return error("bad index file sha1 signature");
 332         return 0;
 333 }
 334
 335 int read_cache(void)
 336 {
 337         int fd, i;
 338         struct stat st;
 339         unsigned long size, offset;
 340         void *map;
 341         struct cache_header *hdr;
 342
 343         errno = EBUSY;
 344         if (active_cache)
 345                 return error("more than one cachefile");
 346         errno = ENOENT;
 347         fd = open(get_index_file(), O_RDONLY);
 348         if (fd < 0)
 349                 return (errno == ENOENT) ? 0 : error("open failed");
 350
 351         size = 0; // avoid gcc warning
 352         map = (void *)-1;
 353         if (!fstat(fd, &st)) {
 354                 size = st.st_size;
 355                 errno = EINVAL;
 356                 if (size >= sizeof(struct cache_header) + 20)
 357                         map = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0);
 358         }
 359         close(fd);
 360         if (-1 == (int)(long)map)
 361                 return error("mmap failed");
 362
 363         hdr = map;
 364         if (verify_hdr(hdr, size) < 0)
 365                 goto unmap;
 366
 367         active_nr = ntohl(hdr->hdr_entries);
 368         active_alloc = alloc_nr(active_nr);
 369         active_cache = calloc(active_alloc, sizeof(struct cache_entry *));
 370
 371         offset = sizeof(*hdr);
 372         for (i = 0; i < active_nr; i++) {
 373                 struct cache_entry *ce = map + offset;
 374                 offset = offset + ce_size(ce);
 375                 active_cache[i] = ce;
 376         }
 377         return active_nr;
 378
 379 unmap:
 380         munmap(map, size);
 381         errno = EINVAL;
 382         return error("verify header failed");
 383 }
 384
 385 #define WRITE_BUFFER_SIZE 8192
 386 static unsigned char write_buffer[WRITE_BUFFER_SIZE];
 387 static unsigned long write_buffer_len;
 388
 389 static int ce_write(SHA_CTX *context, int fd, void *data, unsigned int len)
 390 {
 391         while (len) {
 392                 unsigned int buffered = write_buffer_len;
 393                 unsigned int partial = WRITE_BUFFER_SIZE - buffered;
 394                 if (partial > len)
 395                         partial = len;
 396                 memcpy(write_buffer + buffered, data, partial);
 397                 buffered += partial;
 398                 if (buffered == WRITE_BUFFER_SIZE) {
 399                         SHA1_Update(context, write_buffer, WRITE_BUFFER_SIZE);
 400                         if (write(fd, write_buffer, WRITE_BUFFER_SIZE) != WRITE_BUFFER_SIZE)
 401                                 return -1;
 402                         buffered = 0;
 403                 }
 404                 write_buffer_len = buffered;
 405                 len -= partial;
 406                 data += partial;
 407         }
 408         return 0;
 409 }
 410
 411 static int ce_flush(SHA_CTX *context, int fd)
 412 {
 413         unsigned int left = write_buffer_len;
 414
 415         if (left) {
 416                 write_buffer_len = 0;
 417                 SHA1_Update(context, write_buffer, left);
 418         }
 419
 420         /* Append the SHA1 signature at the end */
 421         SHA1_Final(write_buffer + left, context);
 422         left += 20;
 423         if (write(fd, write_buffer, left) != left)
 424                 return -1;
 425         return 0;
 426 }
 427
 428 int write_cache(int newfd, struct cache_entry **cache, int entries)
 429 {
 430         SHA_CTX c;
 431         struct cache_header hdr;
 432         int i;
 433
 434         hdr.hdr_signature = htonl(CACHE_SIGNATURE);
 435         hdr.hdr_version = htonl(2);
 436         hdr.hdr_entries = htonl(entries);
 437
 438         SHA1_Init(&c);
 439         if (ce_write(&c, newfd, &hdr, sizeof(hdr)) < 0)
 440                 return -1;
 441
 442         for (i = 0; i < entries; i++) {
 443                 struct cache_entry *ce = cache[i];
 444                 if (ce_write(&c, newfd, ce, ce_size(ce)) < 0)
 445                         return -1;
 446         }
 447         return ce_flush(&c, newfd);
 448 }