diff-tree: fix up comparison of "interesting" sub-trees
[git.git] / sha1_file.c
1 /*
2  * GIT - The information manager from hell
3  *
4  * Copyright (C) Linus Torvalds, 2005
5  *
6  * This handles basic git sha1 object files - packing, unpacking,
7  * creation etc.
8  */
9 #include <stdarg.h>
10 #include <limits.h>
11 #include "cache.h"
12
13 #ifndef O_NOATIME
14 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
15 #define O_NOATIME 01000000
16 #else
17 #define O_NOATIME 0
18 #endif
19 #endif
20
21 static unsigned int sha1_file_open_flag = O_NOATIME;
22
23 static unsigned hexval(char c)
24 {
25         if (c >= '0' && c <= '9')
26                 return c - '0';
27         if (c >= 'a' && c <= 'f')
28                 return c - 'a' + 10;
29         if (c >= 'A' && c <= 'F')
30                 return c - 'A' + 10;
31         return ~0;
32 }
33
34 int get_sha1_hex(const char *hex, unsigned char *sha1)
35 {
36         int i;
37         for (i = 0; i < 20; i++) {
38                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
39                 if (val & ~0xff)
40                         return -1;
41                 *sha1++ = val;
42                 hex += 2;
43         }
44         return 0;
45 }
46
47 int get_sha1_file(const char *path, unsigned char *result)
48 {
49         char buffer[60];
50         int fd = open(path, O_RDONLY);
51         int len;
52
53         if (fd < 0)
54                 return -1;
55         len = read(fd, buffer, sizeof(buffer));
56         close(fd);
57         if (len < 40)
58                 return -1;
59         return get_sha1_hex(buffer, result);
60 }
61
62 static char *git_dir, *git_object_dir, *git_index_file;
63 static void setup_git_env(void)
64 {
65         git_dir = gitenv(GIT_DIR_ENVIRONMENT);
66         if (!git_dir)
67                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
68         git_object_dir = gitenv(DB_ENVIRONMENT);
69         if (!git_object_dir) {
70                 git_object_dir = xmalloc(strlen(git_dir) + 9);
71                 sprintf(git_object_dir, "%s/objects", git_dir);
72         }
73         git_index_file = gitenv(INDEX_ENVIRONMENT);
74         if (!git_index_file) {
75                 git_index_file = xmalloc(strlen(git_dir) + 7);
76                 sprintf(git_index_file, "%s/index", git_dir);
77         }
78 }
79
80 char *get_object_directory(void)
81 {
82         if (!git_object_dir)
83                 setup_git_env();
84         return git_object_dir;
85 }
86
87 char *get_index_file(void)
88 {
89         if (!git_index_file)
90                 setup_git_env();
91         return git_index_file;
92 }
93
94 int get_sha1(const char *str, unsigned char *sha1)
95 {
96         static char pathname[PATH_MAX];
97         static const char *prefix[] = {
98                 "",
99                 "refs",
100                 "refs/tags",
101                 "refs/heads",
102                 "refs/snap",
103                 NULL
104         };
105         const char **p;
106
107         if (!get_sha1_hex(str, sha1))
108                 return 0;
109
110         if (!git_dir)
111                 setup_git_env();
112         for (p = prefix; *p; p++) {
113                 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
114                          git_dir, *p, str);
115                 if (!get_sha1_file(pathname, sha1))
116                         return 0;
117         }
118
119         return -1;
120 }
121
122 char * sha1_to_hex(const unsigned char *sha1)
123 {
124         static char buffer[50];
125         static const char hex[] = "0123456789abcdef";
126         char *buf = buffer;
127         int i;
128
129         for (i = 0; i < 20; i++) {
130                 unsigned int val = *sha1++;
131                 *buf++ = hex[val >> 4];
132                 *buf++ = hex[val & 0xf];
133         }
134         return buffer;
135 }
136
137 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
138 {
139         int i;
140         for (i = 0; i < 20; i++) {
141                 static char hex[] = "0123456789abcdef";
142                 unsigned int val = sha1[i];
143                 char *pos = pathbuf + i*2 + (i > 0);
144                 *pos++ = hex[val >> 4];
145                 *pos = hex[val & 0xf];
146         }
147 }
148
149 /*
150  * NOTE! This returns a statically allocated buffer, so you have to be
151  * careful about using it. Do a "strdup()" if you need to save the
152  * filename.
153  *
154  * Also note that this returns the location for creating.  Reading
155  * SHA1 file can happen from any alternate directory listed in the
156  * DB_ENVIRONMENT environment variable if it is not found in
157  * the primary object database.
158  */
159 char *sha1_file_name(const unsigned char *sha1)
160 {
161         static char *name, *base;
162
163         if (!base) {
164                 const char *sha1_file_directory = get_object_directory();
165                 int len = strlen(sha1_file_directory);
166                 base = xmalloc(len + 60);
167                 memcpy(base, sha1_file_directory, len);
168                 memset(base+len, 0, 60);
169                 base[len] = '/';
170                 base[len+3] = '/';
171                 name = base + len + 1;
172         }
173         fill_sha1_path(name, sha1);
174         return base;
175 }
176
177 static struct alternate_object_database {
178         char *base;
179         char *name;
180 } *alt_odb;
181
182 /*
183  * Prepare alternate object database registry.
184  * alt_odb points at an array of struct alternate_object_database.
185  * This array is terminated with an element that has both its base
186  * and name set to NULL.  alt_odb[n] comes from n'th non-empty
187  * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
188  * variable, and its base points at a statically allocated buffer
189  * that contains "/the/directory/corresponding/to/.git/objects/...",
190  * while its name points just after the slash at the end of
191  * ".git/objects/" in the example above, and has enough space to hold
192  * 40-byte hex SHA1, an extra slash for the first level indirection,
193  * and the terminating NUL.
194  * This function allocates the alt_odb array and all the strings
195  * pointed by base fields of the array elements with one xmalloc();
196  * the string pool immediately follows the array.
197  */
198 static void prepare_alt_odb(void)
199 {
200         int pass, totlen, i;
201         const char *cp, *last;
202         char *op = 0;
203         const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
204
205         /* The first pass counts how large an area to allocate to
206          * hold the entire alt_odb structure, including array of
207          * structs and path buffers for them.  The second pass fills
208          * the structure and prepares the path buffers for use by
209          * fill_sha1_path().
210          */
211         for (totlen = pass = 0; pass < 2; pass++) {
212                 last = alt;
213                 i = 0;
214                 do {
215                         cp = strchr(last, ':') ? : last + strlen(last);
216                         if (last != cp) {
217                                 /* 43 = 40-byte + 2 '/' + terminating NUL */
218                                 int pfxlen = cp - last;
219                                 int entlen = pfxlen + 43;
220                                 if (pass == 0)
221                                         totlen += entlen;
222                                 else {
223                                         alt_odb[i].base = op;
224                                         alt_odb[i].name = op + pfxlen + 1;
225                                         memcpy(op, last, pfxlen);
226                                         op[pfxlen] = op[pfxlen + 3] = '/';
227                                         op[entlen-1] = 0;
228                                         op += entlen;
229                                 }
230                                 i++;
231                         }
232                         while (*cp && *cp == ':')
233                                 cp++;
234                         last = cp;
235                 } while (*cp);
236                 if (pass)
237                         break;
238                 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
239                 alt_odb[i].base = alt_odb[i].name = 0;
240                 op = (char*)(&alt_odb[i+1]);
241         }
242 }
243
244 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
245 {
246         int i;
247         char *name = sha1_file_name(sha1);
248
249         if (!stat(name, st))
250                 return name;
251         if (!alt_odb)
252                 prepare_alt_odb();
253         for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
254                 fill_sha1_path(name, sha1);
255                 if (!stat(alt_odb[i].base, st))
256                         return alt_odb[i].base;
257         }
258         return NULL;
259 }
260
261 int check_sha1_signature(unsigned char *sha1, void *map, unsigned long size, const char *type)
262 {
263         char header[100];
264         unsigned char real_sha1[20];
265         SHA_CTX c;
266
267         SHA1_Init(&c);
268         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
269         SHA1_Update(&c, map, size);
270         SHA1_Final(real_sha1, &c);
271         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
272 }
273
274 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
275 {
276         struct stat st;
277         void *map;
278         int fd;
279         char *filename = find_sha1_file(sha1, &st);
280
281         if (!filename) {
282                 error("cannot map sha1 file %s", sha1_to_hex(sha1));
283                 return NULL;
284         }
285
286         fd = open(filename, O_RDONLY | sha1_file_open_flag);
287         if (fd < 0) {
288                 /* See if it works without O_NOATIME */
289                 switch (sha1_file_open_flag) {
290                 default:
291                         fd = open(filename, O_RDONLY);
292                         if (fd >= 0)
293                                 break;
294                 /* Fallthrough */
295                 case 0:
296                         perror(filename);
297                         return NULL;
298                 }
299
300                 /* If it failed once, it will probably fail again. Stop using O_NOATIME */
301                 sha1_file_open_flag = 0;
302         }
303         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
304         close(fd);
305         if (-1 == (int)(long)map)
306                 return NULL;
307         *size = st.st_size;
308         return map;
309 }
310
311 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
312 {
313         int ret, bytes;
314         z_stream stream;
315         char buffer[8192];
316         unsigned char *buf;
317
318         /* Get the data stream */
319         memset(&stream, 0, sizeof(stream));
320         stream.next_in = map;
321         stream.avail_in = mapsize;
322         stream.next_out = (unsigned char *)buffer;
323         stream.avail_out = sizeof(buffer);
324
325         inflateInit(&stream);
326         ret = inflate(&stream, 0);
327         if (ret < Z_OK)
328                 return NULL;
329         if (sscanf(buffer, "%10s %lu", type, size) != 2)
330                 return NULL;
331
332         bytes = strlen(buffer) + 1;
333         buf = xmalloc(*size);
334
335         memcpy(buf, buffer + bytes, stream.total_out - bytes);
336         bytes = stream.total_out - bytes;
337         if (bytes < *size && ret == Z_OK) {
338                 stream.next_out = buf + bytes;
339                 stream.avail_out = *size - bytes;
340                 while (inflate(&stream, Z_FINISH) == Z_OK)
341                         /* nothing */;
342         }
343         inflateEnd(&stream);
344         return buf;
345 }
346
347 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
348 {
349         unsigned long mapsize;
350         void *map, *buf;
351
352         map = map_sha1_file(sha1, &mapsize);
353         if (map) {
354                 buf = unpack_sha1_file(map, mapsize, type, size);
355                 munmap(map, mapsize);
356                 return buf;
357         }
358         return NULL;
359 }
360
361 void *read_object_with_reference(const unsigned char *sha1,
362                                  const char *required_type,
363                                  unsigned long *size,
364                                  unsigned char *actual_sha1_return)
365 {
366         char type[20];
367         void *buffer;
368         unsigned long isize;
369         unsigned char actual_sha1[20];
370
371         memcpy(actual_sha1, sha1, 20);
372         while (1) {
373                 int ref_length = -1;
374                 const char *ref_type = NULL;
375
376                 buffer = read_sha1_file(actual_sha1, type, &isize);
377                 if (!buffer)
378                         return NULL;
379                 if (!strcmp(type, required_type)) {
380                         *size = isize;
381                         if (actual_sha1_return)
382                                 memcpy(actual_sha1_return, actual_sha1, 20);
383                         return buffer;
384                 }
385                 /* Handle references */
386                 else if (!strcmp(type, "commit"))
387                         ref_type = "tree ";
388                 else if (!strcmp(type, "tag"))
389                         ref_type = "object ";
390                 else {
391                         free(buffer);
392                         return NULL;
393                 }
394                 ref_length = strlen(ref_type);
395
396                 if (memcmp(buffer, ref_type, ref_length) ||
397                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
398                         free(buffer);
399                         return NULL;
400                 }
401                 /* Now we have the ID of the referred-to object in
402                  * actual_sha1.  Check again. */
403         }
404 }
405
406 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
407 {
408         int size;
409         unsigned char *compressed;
410         z_stream stream;
411         unsigned char sha1[20];
412         SHA_CTX c;
413         char *filename;
414         static char tmpfile[PATH_MAX];
415         unsigned char hdr[50];
416         int fd, hdrlen, ret;
417
418         /* Generate the header */
419         hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
420
421         /* Sha1.. */
422         SHA1_Init(&c);
423         SHA1_Update(&c, hdr, hdrlen);
424         SHA1_Update(&c, buf, len);
425         SHA1_Final(sha1, &c);
426
427         if (returnsha1)
428                 memcpy(returnsha1, sha1, 20);
429
430         filename = sha1_file_name(sha1);
431         fd = open(filename, O_RDONLY);
432         if (fd >= 0) {
433                 /*
434                  * FIXME!!! We might do collision checking here, but we'd
435                  * need to uncompress the old file and check it. Later.
436                  */
437                 close(fd);
438                 return 0;
439         }
440
441         if (errno != ENOENT) {
442                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
443                 return -1;
444         }
445
446         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
447
448         fd = mkstemp(tmpfile);
449         if (fd < 0) {
450                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
451                 return -1;
452         }
453
454         /* Set it up */
455         memset(&stream, 0, sizeof(stream));
456         deflateInit(&stream, Z_BEST_COMPRESSION);
457         size = deflateBound(&stream, len+hdrlen);
458         compressed = xmalloc(size);
459
460         /* Compress it */
461         stream.next_out = compressed;
462         stream.avail_out = size;
463
464         /* First header.. */
465         stream.next_in = hdr;
466         stream.avail_in = hdrlen;
467         while (deflate(&stream, 0) == Z_OK)
468                 /* nothing */;
469
470         /* Then the data itself.. */
471         stream.next_in = buf;
472         stream.avail_in = len;
473         while (deflate(&stream, Z_FINISH) == Z_OK)
474                 /* nothing */;
475         deflateEnd(&stream);
476         size = stream.total_out;
477
478         if (write(fd, compressed, size) != size)
479                 die("unable to write file");
480         fchmod(fd, 0444);
481         close(fd);
482         free(compressed);
483
484         ret = link(tmpfile, filename);
485         if (ret < 0) {
486                 ret = errno;
487
488                 /*
489                  * Coda hack - coda doesn't like cross-directory links,
490                  * so we fall back to a rename, which will mean that it
491                  * won't be able to check collisions, but that's not a
492                  * big deal.
493                  *
494                  * When this succeeds, we just return 0. We have nothing
495                  * left to unlink.
496                  */
497                 if (ret == EXDEV && !rename(tmpfile, filename))
498                         return 0;
499         }
500         unlink(tmpfile);
501         if (ret) {
502                 if (ret != EEXIST) {
503                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
504                         return -1;
505                 }
506                 /* FIXME!!! Collision check here ? */
507         }
508
509         return 0;
510 }
511
512 int write_sha1_from_fd(const unsigned char *sha1, int fd)
513 {
514         char *filename = sha1_file_name(sha1);
515
516         int local;
517         z_stream stream;
518         unsigned char real_sha1[20];
519         unsigned char buf[4096];
520         unsigned char discard[4096];
521         int ret;
522         SHA_CTX c;
523
524         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
525
526         if (local < 0)
527                 return error("Couldn't open %s\n", filename);
528
529         memset(&stream, 0, sizeof(stream));
530
531         inflateInit(&stream);
532
533         SHA1_Init(&c);
534
535         do {
536                 ssize_t size;
537                 size = read(fd, buf, 4096);
538                 if (size <= 0) {
539                         close(local);
540                         unlink(filename);
541                         if (!size)
542                                 return error("Connection closed?");
543                         perror("Reading from connection");
544                         return -1;
545                 }
546                 write(local, buf, size);
547                 stream.avail_in = size;
548                 stream.next_in = buf;
549                 do {
550                         stream.next_out = discard;
551                         stream.avail_out = sizeof(discard);
552                         ret = inflate(&stream, Z_SYNC_FLUSH);
553                         SHA1_Update(&c, discard, sizeof(discard) -
554                                     stream.avail_out);
555                 } while (stream.avail_in && ret == Z_OK);
556                 
557         } while (ret == Z_OK);
558         inflateEnd(&stream);
559
560         close(local);
561         SHA1_Final(real_sha1, &c);
562         if (ret != Z_STREAM_END) {
563                 unlink(filename);
564                 return error("File %s corrupted", sha1_to_hex(sha1));
565         }
566         if (memcmp(sha1, real_sha1, 20)) {
567                 unlink(filename);
568                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
569         }
570         
571         return 0;
572 }
573
574 int has_sha1_file(const unsigned char *sha1)
575 {
576         struct stat st;
577         return !!find_sha1_file(sha1, &st);
578 }
579
580 int index_fd(unsigned char *sha1, int fd, struct stat *st)
581 {
582         unsigned long size = st->st_size;
583         void *buf;
584         int ret;
585
586         buf = "";
587         if (size)
588                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
589         close(fd);
590         if ((int)(long)buf == -1)
591                 return -1;
592
593         ret = write_sha1_file(buf, size, "blob", sha1);
594         if (size)
595                 munmap(buf, size);
596         return ret;
597 }