Split up unpack_sha1_file() some more
[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 "cache.h"
10 #include "delta.h"
11
12 #ifndef O_NOATIME
13 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
14 #define O_NOATIME 01000000
15 #else
16 #define O_NOATIME 0
17 #endif
18 #endif
19
20 static unsigned int sha1_file_open_flag = O_NOATIME;
21
22 static unsigned hexval(char c)
23 {
24         if (c >= '0' && c <= '9')
25                 return c - '0';
26         if (c >= 'a' && c <= 'f')
27                 return c - 'a' + 10;
28         if (c >= 'A' && c <= 'F')
29                 return c - 'A' + 10;
30         return ~0;
31 }
32
33 int get_sha1_hex(const char *hex, unsigned char *sha1)
34 {
35         int i;
36         for (i = 0; i < 20; i++) {
37                 unsigned int val = (hexval(hex[0]) << 4) | hexval(hex[1]);
38                 if (val & ~0xff)
39                         return -1;
40                 *sha1++ = val;
41                 hex += 2;
42         }
43         return 0;
44 }
45
46 static int get_sha1_file(const char *path, unsigned char *result)
47 {
48         char buffer[60];
49         int fd = open(path, O_RDONLY);
50         int len;
51
52         if (fd < 0)
53                 return -1;
54         len = read(fd, buffer, sizeof(buffer));
55         close(fd);
56         if (len < 40)
57                 return -1;
58         return get_sha1_hex(buffer, result);
59 }
60
61 static char *git_dir, *git_object_dir, *git_index_file;
62 static void setup_git_env(void)
63 {
64         git_dir = gitenv(GIT_DIR_ENVIRONMENT);
65         if (!git_dir)
66                 git_dir = DEFAULT_GIT_DIR_ENVIRONMENT;
67         git_object_dir = gitenv(DB_ENVIRONMENT);
68         if (!git_object_dir) {
69                 git_object_dir = xmalloc(strlen(git_dir) + 9);
70                 sprintf(git_object_dir, "%s/objects", git_dir);
71         }
72         git_index_file = gitenv(INDEX_ENVIRONMENT);
73         if (!git_index_file) {
74                 git_index_file = xmalloc(strlen(git_dir) + 7);
75                 sprintf(git_index_file, "%s/index", git_dir);
76         }
77 }
78
79 char *get_object_directory(void)
80 {
81         if (!git_object_dir)
82                 setup_git_env();
83         return git_object_dir;
84 }
85
86 char *get_index_file(void)
87 {
88         if (!git_index_file)
89                 setup_git_env();
90         return git_index_file;
91 }
92
93 int get_sha1(const char *str, unsigned char *sha1)
94 {
95         static char pathname[PATH_MAX];
96         static const char *prefix[] = {
97                 "",
98                 "refs",
99                 "refs/tags",
100                 "refs/heads",
101                 "refs/snap",
102                 NULL
103         };
104         const char **p;
105
106         if (!get_sha1_hex(str, sha1))
107                 return 0;
108
109         if (!git_dir)
110                 setup_git_env();
111         for (p = prefix; *p; p++) {
112                 snprintf(pathname, sizeof(pathname), "%s/%s/%s",
113                          git_dir, *p, str);
114                 if (!get_sha1_file(pathname, sha1))
115                         return 0;
116         }
117
118         return -1;
119 }
120
121 char * sha1_to_hex(const unsigned char *sha1)
122 {
123         static char buffer[50];
124         static const char hex[] = "0123456789abcdef";
125         char *buf = buffer;
126         int i;
127
128         for (i = 0; i < 20; i++) {
129                 unsigned int val = *sha1++;
130                 *buf++ = hex[val >> 4];
131                 *buf++ = hex[val & 0xf];
132         }
133         return buffer;
134 }
135
136 static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
137 {
138         int i;
139         for (i = 0; i < 20; i++) {
140                 static char hex[] = "0123456789abcdef";
141                 unsigned int val = sha1[i];
142                 char *pos = pathbuf + i*2 + (i > 0);
143                 *pos++ = hex[val >> 4];
144                 *pos = hex[val & 0xf];
145         }
146 }
147
148 /*
149  * NOTE! This returns a statically allocated buffer, so you have to be
150  * careful about using it. Do a "strdup()" if you need to save the
151  * filename.
152  *
153  * Also note that this returns the location for creating.  Reading
154  * SHA1 file can happen from any alternate directory listed in the
155  * DB_ENVIRONMENT environment variable if it is not found in
156  * the primary object database.
157  */
158 char *sha1_file_name(const unsigned char *sha1)
159 {
160         static char *name, *base;
161
162         if (!base) {
163                 const char *sha1_file_directory = get_object_directory();
164                 int len = strlen(sha1_file_directory);
165                 base = xmalloc(len + 60);
166                 memcpy(base, sha1_file_directory, len);
167                 memset(base+len, 0, 60);
168                 base[len] = '/';
169                 base[len+3] = '/';
170                 name = base + len + 1;
171         }
172         fill_sha1_path(name, sha1);
173         return base;
174 }
175
176 static struct alternate_object_database {
177         char *base;
178         char *name;
179 } *alt_odb;
180
181 /*
182  * Prepare alternate object database registry.
183  * alt_odb points at an array of struct alternate_object_database.
184  * This array is terminated with an element that has both its base
185  * and name set to NULL.  alt_odb[n] comes from n'th non-empty
186  * element from colon separated ALTERNATE_DB_ENVIRONMENT environment
187  * variable, and its base points at a statically allocated buffer
188  * that contains "/the/directory/corresponding/to/.git/objects/...",
189  * while its name points just after the slash at the end of
190  * ".git/objects/" in the example above, and has enough space to hold
191  * 40-byte hex SHA1, an extra slash for the first level indirection,
192  * and the terminating NUL.
193  * This function allocates the alt_odb array and all the strings
194  * pointed by base fields of the array elements with one xmalloc();
195  * the string pool immediately follows the array.
196  */
197 static void prepare_alt_odb(void)
198 {
199         int pass, totlen, i;
200         const char *cp, *last;
201         char *op = NULL;
202         const char *alt = gitenv(ALTERNATE_DB_ENVIRONMENT) ? : "";
203
204         /* The first pass counts how large an area to allocate to
205          * hold the entire alt_odb structure, including array of
206          * structs and path buffers for them.  The second pass fills
207          * the structure and prepares the path buffers for use by
208          * fill_sha1_path().
209          */
210         for (totlen = pass = 0; pass < 2; pass++) {
211                 last = alt;
212                 i = 0;
213                 do {
214                         cp = strchr(last, ':') ? : last + strlen(last);
215                         if (last != cp) {
216                                 /* 43 = 40-byte + 2 '/' + terminating NUL */
217                                 int pfxlen = cp - last;
218                                 int entlen = pfxlen + 43;
219                                 if (pass == 0)
220                                         totlen += entlen;
221                                 else {
222                                         alt_odb[i].base = op;
223                                         alt_odb[i].name = op + pfxlen + 1;
224                                         memcpy(op, last, pfxlen);
225                                         op[pfxlen] = op[pfxlen + 3] = '/';
226                                         op[entlen-1] = 0;
227                                         op += entlen;
228                                 }
229                                 i++;
230                         }
231                         while (*cp && *cp == ':')
232                                 cp++;
233                         last = cp;
234                 } while (*cp);
235                 if (pass)
236                         break;
237                 alt_odb = xmalloc(sizeof(*alt_odb) * (i + 1) + totlen);
238                 alt_odb[i].base = alt_odb[i].name = NULL;
239                 op = (char*)(&alt_odb[i+1]);
240         }
241 }
242
243 static char *find_sha1_file(const unsigned char *sha1, struct stat *st)
244 {
245         int i;
246         char *name = sha1_file_name(sha1);
247
248         if (!stat(name, st))
249                 return name;
250         if (!alt_odb)
251                 prepare_alt_odb();
252         for (i = 0; (name = alt_odb[i].name) != NULL; i++) {
253                 fill_sha1_path(name, sha1);
254                 if (!stat(alt_odb[i].base, st))
255                         return alt_odb[i].base;
256         }
257         return NULL;
258 }
259
260 int check_sha1_signature(unsigned char *sha1, void *map, unsigned long size, const char *type)
261 {
262         char header[100];
263         unsigned char real_sha1[20];
264         SHA_CTX c;
265
266         SHA1_Init(&c);
267         SHA1_Update(&c, header, 1+sprintf(header, "%s %lu", type, size));
268         SHA1_Update(&c, map, size);
269         SHA1_Final(real_sha1, &c);
270         return memcmp(sha1, real_sha1, 20) ? -1 : 0;
271 }
272
273 void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
274 {
275         struct stat st;
276         void *map;
277         int fd;
278         char *filename = find_sha1_file(sha1, &st);
279
280         if (!filename) {
281                 error("cannot map sha1 file %s", sha1_to_hex(sha1));
282                 return NULL;
283         }
284
285         fd = open(filename, O_RDONLY | sha1_file_open_flag);
286         if (fd < 0) {
287                 /* See if it works without O_NOATIME */
288                 switch (sha1_file_open_flag) {
289                 default:
290                         fd = open(filename, O_RDONLY);
291                         if (fd >= 0)
292                                 break;
293                 /* Fallthrough */
294                 case 0:
295                         perror(filename);
296                         return NULL;
297                 }
298
299                 /* If it failed once, it will probably fail again. Stop using O_NOATIME */
300                 sha1_file_open_flag = 0;
301         }
302         map = mmap(NULL, st.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
303         close(fd);
304         if (-1 == (int)(long)map)
305                 return NULL;
306         *size = st.st_size;
307         return map;
308 }
309
310 int unpack_sha1_header(z_stream *stream, void *map, unsigned long mapsize, void *buffer, unsigned long size)
311 {
312         /* Get the data stream */
313         memset(stream, 0, sizeof(*stream));
314         stream->next_in = map;
315         stream->avail_in = mapsize;
316         stream->next_out = buffer;
317         stream->avail_out = size;
318
319         inflateInit(stream);
320         return inflate(stream, 0);
321 }
322
323 void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size)
324 {
325         int bytes = strlen(buffer) + 1;
326         char *buf = xmalloc(1+size);
327
328         memcpy(buf, buffer + bytes, stream->total_out - bytes);
329         bytes = stream->total_out - bytes;
330         if (bytes < size) {
331                 stream->next_out = buf + bytes;
332                 stream->avail_out = size - bytes;
333                 while (inflate(stream, Z_FINISH) == Z_OK)
334                         /* nothing */;
335         }
336         buf[size] = 0;
337         inflateEnd(stream);
338         return buf;
339 }
340
341 /*
342  * We used to just use "sscanf()", but that's actually way
343  * too permissive for what we want to check. So do an anal
344  * object header parse by hand.
345  */
346 int parse_sha1_header(char *hdr, char *type, unsigned long *sizep)
347 {
348         int i;
349         unsigned long size;
350
351         /*
352          * The type can be at most ten bytes (including the 
353          * terminating '\0' that we add), and is followed by
354          * a space. 
355          */
356         i = 10;
357         for (;;) {
358                 char c = *hdr++;
359                 if (c == ' ')
360                         break;
361                 if (!--i)
362                         return -1;
363                 *type++ = c;
364         }
365         *type = 0;
366
367         /*
368          * The length must follow immediately, and be in canonical
369          * decimal format (ie "010" is not valid).
370          */
371         size = *hdr++ - '0';
372         if (size > 9)
373                 return -1;
374         if (size) {
375                 for (;;) {
376                         unsigned long c = *hdr - '0';
377                         if (c > 9)
378                                 break;
379                         hdr++;
380                         size = size * 10 + c;
381                 }
382         }
383         *sizep = size;
384
385         /*
386          * The length must be followed by a zero byte
387          */
388         return *hdr ? -1 : 0;
389 }
390
391 void * unpack_sha1_file(void *map, unsigned long mapsize, char *type, unsigned long *size)
392 {
393         int ret;
394         z_stream stream;
395         char hdr[8192];
396
397         ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
398         if (ret < Z_OK || parse_sha1_header(hdr, type, size) < 0)
399                 return NULL;
400
401         return unpack_sha1_rest(&stream, hdr, *size);
402 }
403
404 void * read_sha1_file(const unsigned char *sha1, char *type, unsigned long *size)
405 {
406         unsigned long mapsize;
407         void *map, *buf;
408
409         map = map_sha1_file(sha1, &mapsize);
410         if (map) {
411                 buf = unpack_sha1_file(map, mapsize, type, size);
412                 munmap(map, mapsize);
413                 if (buf && !strcmp(type, "delta")) {
414                         void *ref = NULL, *delta = buf;
415                         unsigned long ref_size, delta_size = *size;
416                         buf = NULL;
417                         if (delta_size > 20)
418                                 ref = read_sha1_file(delta, type, &ref_size);
419                         if (ref)
420                                 buf = patch_delta(ref, ref_size,
421                                                   delta+20, delta_size-20, 
422                                                   size);
423                         free(delta);
424                         free(ref);
425                 }
426                 return buf;
427         }
428         return NULL;
429 }
430
431 void *read_object_with_reference(const unsigned char *sha1,
432                                  const char *required_type,
433                                  unsigned long *size,
434                                  unsigned char *actual_sha1_return)
435 {
436         char type[20];
437         void *buffer;
438         unsigned long isize;
439         unsigned char actual_sha1[20];
440
441         memcpy(actual_sha1, sha1, 20);
442         while (1) {
443                 int ref_length = -1;
444                 const char *ref_type = NULL;
445
446                 buffer = read_sha1_file(actual_sha1, type, &isize);
447                 if (!buffer)
448                         return NULL;
449                 if (!strcmp(type, required_type)) {
450                         *size = isize;
451                         if (actual_sha1_return)
452                                 memcpy(actual_sha1_return, actual_sha1, 20);
453                         return buffer;
454                 }
455                 /* Handle references */
456                 else if (!strcmp(type, "commit"))
457                         ref_type = "tree ";
458                 else if (!strcmp(type, "tag"))
459                         ref_type = "object ";
460                 else {
461                         free(buffer);
462                         return NULL;
463                 }
464                 ref_length = strlen(ref_type);
465
466                 if (memcmp(buffer, ref_type, ref_length) ||
467                     get_sha1_hex(buffer + ref_length, actual_sha1)) {
468                         free(buffer);
469                         return NULL;
470                 }
471                 /* Now we have the ID of the referred-to object in
472                  * actual_sha1.  Check again. */
473         }
474 }
475
476 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
477 {
478         int size;
479         unsigned char *compressed;
480         z_stream stream;
481         unsigned char sha1[20];
482         SHA_CTX c;
483         char *filename;
484         static char tmpfile[PATH_MAX];
485         unsigned char hdr[50];
486         int fd, hdrlen, ret;
487
488         /* Generate the header */
489         hdrlen = sprintf((char *)hdr, "%s %lu", type, len)+1;
490
491         /* Sha1.. */
492         SHA1_Init(&c);
493         SHA1_Update(&c, hdr, hdrlen);
494         SHA1_Update(&c, buf, len);
495         SHA1_Final(sha1, &c);
496
497         if (returnsha1)
498                 memcpy(returnsha1, sha1, 20);
499
500         filename = sha1_file_name(sha1);
501         fd = open(filename, O_RDONLY);
502         if (fd >= 0) {
503                 /*
504                  * FIXME!!! We might do collision checking here, but we'd
505                  * need to uncompress the old file and check it. Later.
506                  */
507                 close(fd);
508                 return 0;
509         }
510
511         if (errno != ENOENT) {
512                 fprintf(stderr, "sha1 file %s: %s", filename, strerror(errno));
513                 return -1;
514         }
515
516         snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
517
518         fd = mkstemp(tmpfile);
519         if (fd < 0) {
520                 fprintf(stderr, "unable to create temporary sha1 filename %s: %s", tmpfile, strerror(errno));
521                 return -1;
522         }
523
524         /* Set it up */
525         memset(&stream, 0, sizeof(stream));
526         deflateInit(&stream, Z_BEST_COMPRESSION);
527         size = deflateBound(&stream, len+hdrlen);
528         compressed = xmalloc(size);
529
530         /* Compress it */
531         stream.next_out = compressed;
532         stream.avail_out = size;
533
534         /* First header.. */
535         stream.next_in = hdr;
536         stream.avail_in = hdrlen;
537         while (deflate(&stream, 0) == Z_OK)
538                 /* nothing */;
539
540         /* Then the data itself.. */
541         stream.next_in = buf;
542         stream.avail_in = len;
543         while (deflate(&stream, Z_FINISH) == Z_OK)
544                 /* nothing */;
545         deflateEnd(&stream);
546         size = stream.total_out;
547
548         if (write(fd, compressed, size) != size)
549                 die("unable to write file");
550         fchmod(fd, 0444);
551         close(fd);
552         free(compressed);
553
554         ret = link(tmpfile, filename);
555         if (ret < 0) {
556                 ret = errno;
557
558                 /*
559                  * Coda hack - coda doesn't like cross-directory links,
560                  * so we fall back to a rename, which will mean that it
561                  * won't be able to check collisions, but that's not a
562                  * big deal.
563                  *
564                  * When this succeeds, we just return 0. We have nothing
565                  * left to unlink.
566                  */
567                 if (ret == EXDEV && !rename(tmpfile, filename))
568                         return 0;
569         }
570         unlink(tmpfile);
571         if (ret) {
572                 if (ret != EEXIST) {
573                         fprintf(stderr, "unable to write sha1 filename %s: %s", filename, strerror(ret));
574                         return -1;
575                 }
576                 /* FIXME!!! Collision check here ? */
577         }
578
579         return 0;
580 }
581
582 int write_sha1_from_fd(const unsigned char *sha1, int fd)
583 {
584         char *filename = sha1_file_name(sha1);
585
586         int local;
587         z_stream stream;
588         unsigned char real_sha1[20];
589         unsigned char buf[4096];
590         unsigned char discard[4096];
591         int ret;
592         SHA_CTX c;
593
594         local = open(filename, O_WRONLY | O_CREAT | O_EXCL, 0666);
595
596         if (local < 0)
597                 return error("Couldn't open %s\n", filename);
598
599         memset(&stream, 0, sizeof(stream));
600
601         inflateInit(&stream);
602
603         SHA1_Init(&c);
604
605         do {
606                 ssize_t size;
607                 size = read(fd, buf, 4096);
608                 if (size <= 0) {
609                         close(local);
610                         unlink(filename);
611                         if (!size)
612                                 return error("Connection closed?");
613                         perror("Reading from connection");
614                         return -1;
615                 }
616                 write(local, buf, size);
617                 stream.avail_in = size;
618                 stream.next_in = buf;
619                 do {
620                         stream.next_out = discard;
621                         stream.avail_out = sizeof(discard);
622                         ret = inflate(&stream, Z_SYNC_FLUSH);
623                         SHA1_Update(&c, discard, sizeof(discard) -
624                                     stream.avail_out);
625                 } while (stream.avail_in && ret == Z_OK);
626                 
627         } while (ret == Z_OK);
628         inflateEnd(&stream);
629
630         close(local);
631         SHA1_Final(real_sha1, &c);
632         if (ret != Z_STREAM_END) {
633                 unlink(filename);
634                 return error("File %s corrupted", sha1_to_hex(sha1));
635         }
636         if (memcmp(sha1, real_sha1, 20)) {
637                 unlink(filename);
638                 return error("File %s has bad hash\n", sha1_to_hex(sha1));
639         }
640         
641         return 0;
642 }
643
644 int has_sha1_file(const unsigned char *sha1)
645 {
646         struct stat st;
647         return !!find_sha1_file(sha1, &st);
648 }
649
650 int index_fd(unsigned char *sha1, int fd, struct stat *st)
651 {
652         unsigned long size = st->st_size;
653         void *buf;
654         int ret;
655
656         buf = "";
657         if (size)
658                 buf = mmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
659         close(fd);
660         if ((int)(long)buf == -1)
661                 return -1;
662
663         ret = write_sha1_file(buf, size, "blob", sha1);
664         if (size)
665                 munmap(buf, size);
666         return ret;
667 }