Code

b4cbcd90118c47d47e8940322c8ec2161b33a92a
[git.git] / fast-import.c
1 /*
2 Format of STDIN stream:
4   stream ::= cmd*;
6   cmd ::= new_blob
7         | new_commit
8         | new_tag
9         | reset_branch
10         | checkpoint
11         ;
13   new_blob ::= 'blob' lf
14         mark?
15     file_content;
16   file_content ::= data;
18   new_commit ::= 'commit' sp ref_str lf
19     mark?
20     ('author' sp name '<' email '>' when lf)?
21     'committer' sp name '<' email '>' when lf
22     commit_msg
23     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
24     ('merge' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)*
25     file_change*
26     lf;
27   commit_msg ::= data;
29   file_change ::= file_clr | file_del | file_obm | file_inm;
30   file_clr ::= 'deleteall' lf;
31   file_del ::= 'D' sp path_str lf;
32   file_obm ::= 'M' sp mode sp (hexsha1 | idnum) sp path_str lf;
33   file_inm ::= 'M' sp mode sp 'inline' sp path_str lf
34     data;
36   new_tag ::= 'tag' sp tag_str lf
37     'from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf
38         'tagger' sp name '<' email '>' when lf
39     tag_msg;
40   tag_msg ::= data;
42   reset_branch ::= 'reset' sp ref_str lf
43     ('from' sp (ref_str | hexsha1 | sha1exp_str | idnum) lf)?
44     lf;
46   checkpoint ::= 'checkpoint' lf
47     lf;
49      # note: the first idnum in a stream should be 1 and subsequent
50      # idnums should not have gaps between values as this will cause
51      # the stream parser to reserve space for the gapped values.  An
52          # idnum can be updated in the future to a new object by issuing
53      # a new mark directive with the old idnum.
54          #
55   mark ::= 'mark' sp idnum lf;
56   data ::= (delimited_data | exact_data)
57     lf;
59     # note: delim may be any string but must not contain lf.
60     # data_line may contain any data but must not be exactly
61     # delim.
62   delimited_data ::= 'data' sp '<<' delim lf
63     (data_line lf)*
64         delim lf;
66      # note: declen indicates the length of binary_data in bytes.
67      # declen does not include the lf preceeding the binary data.
68      #
69   exact_data ::= 'data' sp declen lf
70     binary_data;
72      # note: quoted strings are C-style quoting supporting \c for
73      # common escapes of 'c' (e..g \n, \t, \\, \") or \nnn where nnn
74          # is the signed byte value in octal.  Note that the only
75      # characters which must actually be escaped to protect the
76      # stream formatting is: \, " and LF.  Otherwise these values
77          # are UTF8.
78      #
79   ref_str     ::= ref;
80   sha1exp_str ::= sha1exp;
81   tag_str     ::= tag;
82   path_str    ::= path    | '"' quoted(path)    '"' ;
83   mode        ::= '100644' | '644'
84                 | '100755' | '755'
85                 | '120000'
86                 ;
88   declen ::= # unsigned 32 bit value, ascii base10 notation;
89   bigint ::= # unsigned integer value, ascii base10 notation;
90   binary_data ::= # file content, not interpreted;
92   when         ::= raw_when | rfc2822_when;
93   raw_when     ::= ts sp tz;
94   rfc2822_when ::= # Valid RFC 2822 date and time;
96   sp ::= # ASCII space character;
97   lf ::= # ASCII newline (LF) character;
99      # note: a colon (':') must precede the numerical value assigned to
100          # an idnum.  This is to distinguish it from a ref or tag name as
101      # GIT does not permit ':' in ref or tag strings.
102          #
103   idnum   ::= ':' bigint;
104   path    ::= # GIT style file path, e.g. "a/b/c";
105   ref     ::= # GIT ref name, e.g. "refs/heads/MOZ_GECKO_EXPERIMENT";
106   tag     ::= # GIT tag name, e.g. "FIREFOX_1_5";
107   sha1exp ::= # Any valid GIT SHA1 expression;
108   hexsha1 ::= # SHA1 in hexadecimal format;
110      # note: name and email are UTF8 strings, however name must not
111          # contain '<' or lf and email must not contain any of the
112      # following: '<', '>', lf.
113          #
114   name  ::= # valid GIT author/committer name;
115   email ::= # valid GIT author/committer email;
116   ts    ::= # time since the epoch in seconds, ascii base10 notation;
117   tz    ::= # GIT style timezone;
118 */
120 #include "builtin.h"
121 #include "cache.h"
122 #include "object.h"
123 #include "blob.h"
124 #include "tree.h"
125 #include "commit.h"
126 #include "delta.h"
127 #include "pack.h"
128 #include "refs.h"
129 #include "csum-file.h"
130 #include "strbuf.h"
131 #include "quote.h"
133 #define PACK_ID_BITS 16
134 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
136 struct object_entry
138         struct object_entry *next;
139         uint32_t offset;
140         unsigned type : TYPE_BITS;
141         unsigned pack_id : PACK_ID_BITS;
142         unsigned char sha1[20];
143 };
145 struct object_entry_pool
147         struct object_entry_pool *next_pool;
148         struct object_entry *next_free;
149         struct object_entry *end;
150         struct object_entry entries[FLEX_ARRAY]; /* more */
151 };
153 struct mark_set
155         union {
156                 struct object_entry *marked[1024];
157                 struct mark_set *sets[1024];
158         } data;
159         unsigned int shift;
160 };
162 struct last_object
164         void *data;
165         unsigned long len;
166         uint32_t offset;
167         unsigned int depth;
168         unsigned no_free:1;
169 };
171 struct mem_pool
173         struct mem_pool *next_pool;
174         char *next_free;
175         char *end;
176         char space[FLEX_ARRAY]; /* more */
177 };
179 struct atom_str
181         struct atom_str *next_atom;
182         unsigned short str_len;
183         char str_dat[FLEX_ARRAY]; /* more */
184 };
186 struct tree_content;
187 struct tree_entry
189         struct tree_content *tree;
190         struct atom_str* name;
191         struct tree_entry_ms
192         {
193                 uint16_t mode;
194                 unsigned char sha1[20];
195         } versions[2];
196 };
198 struct tree_content
200         unsigned int entry_capacity; /* must match avail_tree_content */
201         unsigned int entry_count;
202         unsigned int delta_depth;
203         struct tree_entry *entries[FLEX_ARRAY]; /* more */
204 };
206 struct avail_tree_content
208         unsigned int entry_capacity; /* must match tree_content */
209         struct avail_tree_content *next_avail;
210 };
212 struct branch
214         struct branch *table_next_branch;
215         struct branch *active_next_branch;
216         const char *name;
217         struct tree_entry branch_tree;
218         uintmax_t last_commit;
219         unsigned active : 1;
220         unsigned pack_id : PACK_ID_BITS;
221         unsigned char sha1[20];
222 };
224 struct tag
226         struct tag *next_tag;
227         const char *name;
228         unsigned int pack_id;
229         unsigned char sha1[20];
230 };
232 struct dbuf
234         void *buffer;
235         size_t capacity;
236 };
238 struct hash_list
240         struct hash_list *next;
241         unsigned char sha1[20];
242 };
244 typedef enum {
245         WHENSPEC_RAW = 1,
246         WHENSPEC_RFC2822,
247         WHENSPEC_NOW,
248 } whenspec_type;
250 /* Configured limits on output */
251 static unsigned long max_depth = 10;
252 static off_t max_packsize = (1LL << 32) - 1;
253 static int force_update;
255 /* Stats and misc. counters */
256 static uintmax_t alloc_count;
257 static uintmax_t marks_set_count;
258 static uintmax_t object_count_by_type[1 << TYPE_BITS];
259 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
260 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
261 static unsigned long object_count;
262 static unsigned long branch_count;
263 static unsigned long branch_load_count;
264 static int failure;
265 static FILE *pack_edges;
267 /* Memory pools */
268 static size_t mem_pool_alloc = 2*1024*1024 - sizeof(struct mem_pool);
269 static size_t total_allocd;
270 static struct mem_pool *mem_pool;
272 /* Atom management */
273 static unsigned int atom_table_sz = 4451;
274 static unsigned int atom_cnt;
275 static struct atom_str **atom_table;
277 /* The .pack file being generated */
278 static unsigned int pack_id;
279 static struct packed_git *pack_data;
280 static struct packed_git **all_packs;
281 static unsigned long pack_size;
283 /* Table of objects we've written. */
284 static unsigned int object_entry_alloc = 5000;
285 static struct object_entry_pool *blocks;
286 static struct object_entry *object_table[1 << 16];
287 static struct mark_set *marks;
288 static const char* mark_file;
290 /* Our last blob */
291 static struct last_object last_blob;
293 /* Tree management */
294 static unsigned int tree_entry_alloc = 1000;
295 static void *avail_tree_entry;
296 static unsigned int avail_tree_table_sz = 100;
297 static struct avail_tree_content **avail_tree_table;
298 static struct dbuf old_tree;
299 static struct dbuf new_tree;
301 /* Branch data */
302 static unsigned long max_active_branches = 5;
303 static unsigned long cur_active_branches;
304 static unsigned long branch_table_sz = 1039;
305 static struct branch **branch_table;
306 static struct branch *active_branches;
308 /* Tag data */
309 static struct tag *first_tag;
310 static struct tag *last_tag;
312 /* Input stream parsing */
313 static whenspec_type whenspec = WHENSPEC_RAW;
314 static struct strbuf command_buf;
315 static uintmax_t next_mark;
316 static struct dbuf new_data;
319 static void alloc_objects(unsigned int cnt)
321         struct object_entry_pool *b;
323         b = xmalloc(sizeof(struct object_entry_pool)
324                 + cnt * sizeof(struct object_entry));
325         b->next_pool = blocks;
326         b->next_free = b->entries;
327         b->end = b->entries + cnt;
328         blocks = b;
329         alloc_count += cnt;
332 static struct object_entry *new_object(unsigned char *sha1)
334         struct object_entry *e;
336         if (blocks->next_free == blocks->end)
337                 alloc_objects(object_entry_alloc);
339         e = blocks->next_free++;
340         hashcpy(e->sha1, sha1);
341         return e;
344 static struct object_entry *find_object(unsigned char *sha1)
346         unsigned int h = sha1[0] << 8 | sha1[1];
347         struct object_entry *e;
348         for (e = object_table[h]; e; e = e->next)
349                 if (!hashcmp(sha1, e->sha1))
350                         return e;
351         return NULL;
354 static struct object_entry *insert_object(unsigned char *sha1)
356         unsigned int h = sha1[0] << 8 | sha1[1];
357         struct object_entry *e = object_table[h];
358         struct object_entry *p = NULL;
360         while (e) {
361                 if (!hashcmp(sha1, e->sha1))
362                         return e;
363                 p = e;
364                 e = e->next;
365         }
367         e = new_object(sha1);
368         e->next = NULL;
369         e->offset = 0;
370         if (p)
371                 p->next = e;
372         else
373                 object_table[h] = e;
374         return e;
377 static unsigned int hc_str(const char *s, size_t len)
379         unsigned int r = 0;
380         while (len-- > 0)
381                 r = r * 31 + *s++;
382         return r;
385 static void *pool_alloc(size_t len)
387         struct mem_pool *p;
388         void *r;
390         for (p = mem_pool; p; p = p->next_pool)
391                 if ((p->end - p->next_free >= len))
392                         break;
394         if (!p) {
395                 if (len >= (mem_pool_alloc/2)) {
396                         total_allocd += len;
397                         return xmalloc(len);
398                 }
399                 total_allocd += sizeof(struct mem_pool) + mem_pool_alloc;
400                 p = xmalloc(sizeof(struct mem_pool) + mem_pool_alloc);
401                 p->next_pool = mem_pool;
402                 p->next_free = p->space;
403                 p->end = p->next_free + mem_pool_alloc;
404                 mem_pool = p;
405         }
407         r = p->next_free;
408         /* round out to a pointer alignment */
409         if (len & (sizeof(void*) - 1))
410                 len += sizeof(void*) - (len & (sizeof(void*) - 1));
411         p->next_free += len;
412         return r;
415 static void *pool_calloc(size_t count, size_t size)
417         size_t len = count * size;
418         void *r = pool_alloc(len);
419         memset(r, 0, len);
420         return r;
423 static char *pool_strdup(const char *s)
425         char *r = pool_alloc(strlen(s) + 1);
426         strcpy(r, s);
427         return r;
430 static void size_dbuf(struct dbuf *b, size_t maxlen)
432         if (b->buffer) {
433                 if (b->capacity >= maxlen)
434                         return;
435                 free(b->buffer);
436         }
437         b->capacity = ((maxlen / 1024) + 1) * 1024;
438         b->buffer = xmalloc(b->capacity);
441 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
443         struct mark_set *s = marks;
444         while ((idnum >> s->shift) >= 1024) {
445                 s = pool_calloc(1, sizeof(struct mark_set));
446                 s->shift = marks->shift + 10;
447                 s->data.sets[0] = marks;
448                 marks = s;
449         }
450         while (s->shift) {
451                 uintmax_t i = idnum >> s->shift;
452                 idnum -= i << s->shift;
453                 if (!s->data.sets[i]) {
454                         s->data.sets[i] = pool_calloc(1, sizeof(struct mark_set));
455                         s->data.sets[i]->shift = s->shift - 10;
456                 }
457                 s = s->data.sets[i];
458         }
459         if (!s->data.marked[idnum])
460                 marks_set_count++;
461         s->data.marked[idnum] = oe;
464 static struct object_entry *find_mark(uintmax_t idnum)
466         uintmax_t orig_idnum = idnum;
467         struct mark_set *s = marks;
468         struct object_entry *oe = NULL;
469         if ((idnum >> s->shift) < 1024) {
470                 while (s && s->shift) {
471                         uintmax_t i = idnum >> s->shift;
472                         idnum -= i << s->shift;
473                         s = s->data.sets[i];
474                 }
475                 if (s)
476                         oe = s->data.marked[idnum];
477         }
478         if (!oe)
479                 die("mark :%" PRIuMAX " not declared", orig_idnum);
480         return oe;
483 static struct atom_str *to_atom(const char *s, unsigned short len)
485         unsigned int hc = hc_str(s, len) % atom_table_sz;
486         struct atom_str *c;
488         for (c = atom_table[hc]; c; c = c->next_atom)
489                 if (c->str_len == len && !strncmp(s, c->str_dat, len))
490                         return c;
492         c = pool_alloc(sizeof(struct atom_str) + len + 1);
493         c->str_len = len;
494         strncpy(c->str_dat, s, len);
495         c->str_dat[len] = 0;
496         c->next_atom = atom_table[hc];
497         atom_table[hc] = c;
498         atom_cnt++;
499         return c;
502 static struct branch *lookup_branch(const char *name)
504         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
505         struct branch *b;
507         for (b = branch_table[hc]; b; b = b->table_next_branch)
508                 if (!strcmp(name, b->name))
509                         return b;
510         return NULL;
513 static struct branch *new_branch(const char *name)
515         unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
516         struct branch* b = lookup_branch(name);
518         if (b)
519                 die("Invalid attempt to create duplicate branch: %s", name);
520         if (check_ref_format(name))
521                 die("Branch name doesn't conform to GIT standards: %s", name);
523         b = pool_calloc(1, sizeof(struct branch));
524         b->name = pool_strdup(name);
525         b->table_next_branch = branch_table[hc];
526         b->branch_tree.versions[0].mode = S_IFDIR;
527         b->branch_tree.versions[1].mode = S_IFDIR;
528         b->active = 0;
529         b->pack_id = MAX_PACK_ID;
530         branch_table[hc] = b;
531         branch_count++;
532         return b;
535 static unsigned int hc_entries(unsigned int cnt)
537         cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
538         return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
541 static struct tree_content *new_tree_content(unsigned int cnt)
543         struct avail_tree_content *f, *l = NULL;
544         struct tree_content *t;
545         unsigned int hc = hc_entries(cnt);
547         for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
548                 if (f->entry_capacity >= cnt)
549                         break;
551         if (f) {
552                 if (l)
553                         l->next_avail = f->next_avail;
554                 else
555                         avail_tree_table[hc] = f->next_avail;
556         } else {
557                 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
558                 f = pool_alloc(sizeof(*t) + sizeof(t->entries[0]) * cnt);
559                 f->entry_capacity = cnt;
560         }
562         t = (struct tree_content*)f;
563         t->entry_count = 0;
564         t->delta_depth = 0;
565         return t;
568 static void release_tree_entry(struct tree_entry *e);
569 static void release_tree_content(struct tree_content *t)
571         struct avail_tree_content *f = (struct avail_tree_content*)t;
572         unsigned int hc = hc_entries(f->entry_capacity);
573         f->next_avail = avail_tree_table[hc];
574         avail_tree_table[hc] = f;
577 static void release_tree_content_recursive(struct tree_content *t)
579         unsigned int i;
580         for (i = 0; i < t->entry_count; i++)
581                 release_tree_entry(t->entries[i]);
582         release_tree_content(t);
585 static struct tree_content *grow_tree_content(
586         struct tree_content *t,
587         int amt)
589         struct tree_content *r = new_tree_content(t->entry_count + amt);
590         r->entry_count = t->entry_count;
591         r->delta_depth = t->delta_depth;
592         memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
593         release_tree_content(t);
594         return r;
597 static struct tree_entry *new_tree_entry(void)
599         struct tree_entry *e;
601         if (!avail_tree_entry) {
602                 unsigned int n = tree_entry_alloc;
603                 total_allocd += n * sizeof(struct tree_entry);
604                 avail_tree_entry = e = xmalloc(n * sizeof(struct tree_entry));
605                 while (n-- > 1) {
606                         *((void**)e) = e + 1;
607                         e++;
608                 }
609                 *((void**)e) = NULL;
610         }
612         e = avail_tree_entry;
613         avail_tree_entry = *((void**)e);
614         return e;
617 static void release_tree_entry(struct tree_entry *e)
619         if (e->tree)
620                 release_tree_content_recursive(e->tree);
621         *((void**)e) = avail_tree_entry;
622         avail_tree_entry = e;
625 static void start_packfile(void)
627         static char tmpfile[PATH_MAX];
628         struct packed_git *p;
629         struct pack_header hdr;
630         int pack_fd;
632         snprintf(tmpfile, sizeof(tmpfile),
633                 "%s/tmp_pack_XXXXXX", get_object_directory());
634         pack_fd = mkstemp(tmpfile);
635         if (pack_fd < 0)
636                 die("Can't create %s: %s", tmpfile, strerror(errno));
637         p = xcalloc(1, sizeof(*p) + strlen(tmpfile) + 2);
638         strcpy(p->pack_name, tmpfile);
639         p->pack_fd = pack_fd;
641         hdr.hdr_signature = htonl(PACK_SIGNATURE);
642         hdr.hdr_version = htonl(2);
643         hdr.hdr_entries = 0;
644         write_or_die(p->pack_fd, &hdr, sizeof(hdr));
646         pack_data = p;
647         pack_size = sizeof(hdr);
648         object_count = 0;
650         all_packs = xrealloc(all_packs, sizeof(*all_packs) * (pack_id + 1));
651         all_packs[pack_id] = p;
654 static void fixup_header_footer(void)
656         static const int buf_sz = 128 * 1024;
657         int pack_fd = pack_data->pack_fd;
658         SHA_CTX c;
659         struct pack_header hdr;
660         char *buf;
662         if (lseek(pack_fd, 0, SEEK_SET) != 0)
663                 die("Failed seeking to start: %s", strerror(errno));
664         if (read_in_full(pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
665                 die("Unable to reread header of %s", pack_data->pack_name);
666         if (lseek(pack_fd, 0, SEEK_SET) != 0)
667                 die("Failed seeking to start: %s", strerror(errno));
668         hdr.hdr_entries = htonl(object_count);
669         write_or_die(pack_fd, &hdr, sizeof(hdr));
671         SHA1_Init(&c);
672         SHA1_Update(&c, &hdr, sizeof(hdr));
674         buf = xmalloc(buf_sz);
675         for (;;) {
676                 ssize_t n = xread(pack_fd, buf, buf_sz);
677                 if (!n)
678                         break;
679                 if (n < 0)
680                         die("Failed to checksum %s", pack_data->pack_name);
681                 SHA1_Update(&c, buf, n);
682         }
683         free(buf);
685         SHA1_Final(pack_data->sha1, &c);
686         write_or_die(pack_fd, pack_data->sha1, sizeof(pack_data->sha1));
687         close(pack_fd);
690 static int oecmp (const void *a_, const void *b_)
692         struct object_entry *a = *((struct object_entry**)a_);
693         struct object_entry *b = *((struct object_entry**)b_);
694         return hashcmp(a->sha1, b->sha1);
697 static char *create_index(void)
699         static char tmpfile[PATH_MAX];
700         SHA_CTX ctx;
701         struct sha1file *f;
702         struct object_entry **idx, **c, **last, *e;
703         struct object_entry_pool *o;
704         uint32_t array[256];
705         int i, idx_fd;
707         /* Build the sorted table of object IDs. */
708         idx = xmalloc(object_count * sizeof(struct object_entry*));
709         c = idx;
710         for (o = blocks; o; o = o->next_pool)
711                 for (e = o->next_free; e-- != o->entries;)
712                         if (pack_id == e->pack_id)
713                                 *c++ = e;
714         last = idx + object_count;
715         if (c != last)
716                 die("internal consistency error creating the index");
717         qsort(idx, object_count, sizeof(struct object_entry*), oecmp);
719         /* Generate the fan-out array. */
720         c = idx;
721         for (i = 0; i < 256; i++) {
722                 struct object_entry **next = c;;
723                 while (next < last) {
724                         if ((*next)->sha1[0] != i)
725                                 break;
726                         next++;
727                 }
728                 array[i] = htonl(next - idx);
729                 c = next;
730         }
732         snprintf(tmpfile, sizeof(tmpfile),
733                 "%s/tmp_idx_XXXXXX", get_object_directory());
734         idx_fd = mkstemp(tmpfile);
735         if (idx_fd < 0)
736                 die("Can't create %s: %s", tmpfile, strerror(errno));
737         f = sha1fd(idx_fd, tmpfile);
738         sha1write(f, array, 256 * sizeof(int));
739         SHA1_Init(&ctx);
740         for (c = idx; c != last; c++) {
741                 uint32_t offset = htonl((*c)->offset);
742                 sha1write(f, &offset, 4);
743                 sha1write(f, (*c)->sha1, sizeof((*c)->sha1));
744                 SHA1_Update(&ctx, (*c)->sha1, 20);
745         }
746         sha1write(f, pack_data->sha1, sizeof(pack_data->sha1));
747         sha1close(f, NULL, 1);
748         free(idx);
749         SHA1_Final(pack_data->sha1, &ctx);
750         return tmpfile;
753 static char *keep_pack(char *curr_index_name)
755         static char name[PATH_MAX];
756         static const char *keep_msg = "fast-import";
757         int keep_fd;
759         chmod(pack_data->pack_name, 0444);
760         chmod(curr_index_name, 0444);
762         snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
763                  get_object_directory(), sha1_to_hex(pack_data->sha1));
764         keep_fd = open(name, O_RDWR|O_CREAT|O_EXCL, 0600);
765         if (keep_fd < 0)
766                 die("cannot create keep file");
767         write(keep_fd, keep_msg, strlen(keep_msg));
768         close(keep_fd);
770         snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
771                  get_object_directory(), sha1_to_hex(pack_data->sha1));
772         if (move_temp_to_file(pack_data->pack_name, name))
773                 die("cannot store pack file");
775         snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
776                  get_object_directory(), sha1_to_hex(pack_data->sha1));
777         if (move_temp_to_file(curr_index_name, name))
778                 die("cannot store index file");
779         return name;
782 static void unkeep_all_packs(void)
784         static char name[PATH_MAX];
785         int k;
787         for (k = 0; k < pack_id; k++) {
788                 struct packed_git *p = all_packs[k];
789                 snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
790                          get_object_directory(), sha1_to_hex(p->sha1));
791                 unlink(name);
792         }
795 static void end_packfile(void)
797         struct packed_git *old_p = pack_data, *new_p;
799         if (object_count) {
800                 char *idx_name;
801                 int i;
802                 struct branch *b;
803                 struct tag *t;
805                 fixup_header_footer();
806                 idx_name = keep_pack(create_index());
808                 /* Register the packfile with core git's machinary. */
809                 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
810                 if (!new_p)
811                         die("core git rejected index %s", idx_name);
812                 new_p->windows = old_p->windows;
813                 all_packs[pack_id] = new_p;
814                 install_packed_git(new_p);
816                 /* Print the boundary */
817                 if (pack_edges) {
818                         fprintf(pack_edges, "%s:", new_p->pack_name);
819                         for (i = 0; i < branch_table_sz; i++) {
820                                 for (b = branch_table[i]; b; b = b->table_next_branch) {
821                                         if (b->pack_id == pack_id)
822                                                 fprintf(pack_edges, " %s", sha1_to_hex(b->sha1));
823                                 }
824                         }
825                         for (t = first_tag; t; t = t->next_tag) {
826                                 if (t->pack_id == pack_id)
827                                         fprintf(pack_edges, " %s", sha1_to_hex(t->sha1));
828                         }
829                         fputc('\n', pack_edges);
830                         fflush(pack_edges);
831                 }
833                 pack_id++;
834         }
835         else
836                 unlink(old_p->pack_name);
837         free(old_p);
839         /* We can't carry a delta across packfiles. */
840         free(last_blob.data);
841         last_blob.data = NULL;
842         last_blob.len = 0;
843         last_blob.offset = 0;
844         last_blob.depth = 0;
847 static void cycle_packfile(void)
849         end_packfile();
850         start_packfile();
853 static size_t encode_header(
854         enum object_type type,
855         size_t size,
856         unsigned char *hdr)
858         int n = 1;
859         unsigned char c;
861         if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
862                 die("bad type %d", type);
864         c = (type << 4) | (size & 15);
865         size >>= 4;
866         while (size) {
867                 *hdr++ = c | 0x80;
868                 c = size & 0x7f;
869                 size >>= 7;
870                 n++;
871         }
872         *hdr = c;
873         return n;
876 static int store_object(
877         enum object_type type,
878         void *dat,
879         size_t datlen,
880         struct last_object *last,
881         unsigned char *sha1out,
882         uintmax_t mark)
884         void *out, *delta;
885         struct object_entry *e;
886         unsigned char hdr[96];
887         unsigned char sha1[20];
888         unsigned long hdrlen, deltalen;
889         SHA_CTX c;
890         z_stream s;
892         hdrlen = sprintf((char*)hdr,"%s %lu", typename(type),
893                 (unsigned long)datlen) + 1;
894         SHA1_Init(&c);
895         SHA1_Update(&c, hdr, hdrlen);
896         SHA1_Update(&c, dat, datlen);
897         SHA1_Final(sha1, &c);
898         if (sha1out)
899                 hashcpy(sha1out, sha1);
901         e = insert_object(sha1);
902         if (mark)
903                 insert_mark(mark, e);
904         if (e->offset) {
905                 duplicate_count_by_type[type]++;
906                 return 1;
907         } else if (find_sha1_pack(sha1, packed_git)) {
908                 e->type = type;
909                 e->pack_id = MAX_PACK_ID;
910                 e->offset = 1; /* just not zero! */
911                 duplicate_count_by_type[type]++;
912                 return 1;
913         }
915         if (last && last->data && last->depth < max_depth) {
916                 delta = diff_delta(last->data, last->len,
917                         dat, datlen,
918                         &deltalen, 0);
919                 if (delta && deltalen >= datlen) {
920                         free(delta);
921                         delta = NULL;
922                 }
923         } else
924                 delta = NULL;
926         memset(&s, 0, sizeof(s));
927         deflateInit(&s, zlib_compression_level);
928         if (delta) {
929                 s.next_in = delta;
930                 s.avail_in = deltalen;
931         } else {
932                 s.next_in = dat;
933                 s.avail_in = datlen;
934         }
935         s.avail_out = deflateBound(&s, s.avail_in);
936         s.next_out = out = xmalloc(s.avail_out);
937         while (deflate(&s, Z_FINISH) == Z_OK)
938                 /* nothing */;
939         deflateEnd(&s);
941         /* Determine if we should auto-checkpoint. */
942         if ((pack_size + 60 + s.total_out) > max_packsize
943                 || (pack_size + 60 + s.total_out) < pack_size) {
945                 /* This new object needs to *not* have the current pack_id. */
946                 e->pack_id = pack_id + 1;
947                 cycle_packfile();
949                 /* We cannot carry a delta into the new pack. */
950                 if (delta) {
951                         free(delta);
952                         delta = NULL;
954                         memset(&s, 0, sizeof(s));
955                         deflateInit(&s, zlib_compression_level);
956                         s.next_in = dat;
957                         s.avail_in = datlen;
958                         s.avail_out = deflateBound(&s, s.avail_in);
959                         s.next_out = out = xrealloc(out, s.avail_out);
960                         while (deflate(&s, Z_FINISH) == Z_OK)
961                                 /* nothing */;
962                         deflateEnd(&s);
963                 }
964         }
966         e->type = type;
967         e->pack_id = pack_id;
968         e->offset = pack_size;
969         object_count++;
970         object_count_by_type[type]++;
972         if (delta) {
973                 unsigned long ofs = e->offset - last->offset;
974                 unsigned pos = sizeof(hdr) - 1;
976                 delta_count_by_type[type]++;
977                 last->depth++;
979                 hdrlen = encode_header(OBJ_OFS_DELTA, deltalen, hdr);
980                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
981                 pack_size += hdrlen;
983                 hdr[pos] = ofs & 127;
984                 while (ofs >>= 7)
985                         hdr[--pos] = 128 | (--ofs & 127);
986                 write_or_die(pack_data->pack_fd, hdr + pos, sizeof(hdr) - pos);
987                 pack_size += sizeof(hdr) - pos;
988         } else {
989                 if (last)
990                         last->depth = 0;
991                 hdrlen = encode_header(type, datlen, hdr);
992                 write_or_die(pack_data->pack_fd, hdr, hdrlen);
993                 pack_size += hdrlen;
994         }
996         write_or_die(pack_data->pack_fd, out, s.total_out);
997         pack_size += s.total_out;
999         free(out);
1000         free(delta);
1001         if (last) {
1002                 if (!last->no_free)
1003                         free(last->data);
1004                 last->data = dat;
1005                 last->offset = e->offset;
1006                 last->len = datlen;
1007         }
1008         return 0;
1011 static void *gfi_unpack_entry(
1012         struct object_entry *oe,
1013         unsigned long *sizep)
1015         enum object_type type;
1016         struct packed_git *p = all_packs[oe->pack_id];
1017         if (p == pack_data)
1018                 p->pack_size = pack_size + 20;
1019         return unpack_entry(p, oe->offset, &type, sizep);
1022 static const char *get_mode(const char *str, uint16_t *modep)
1024         unsigned char c;
1025         uint16_t mode = 0;
1027         while ((c = *str++) != ' ') {
1028                 if (c < '0' || c > '7')
1029                         return NULL;
1030                 mode = (mode << 3) + (c - '0');
1031         }
1032         *modep = mode;
1033         return str;
1036 static void load_tree(struct tree_entry *root)
1038         unsigned char* sha1 = root->versions[1].sha1;
1039         struct object_entry *myoe;
1040         struct tree_content *t;
1041         unsigned long size;
1042         char *buf;
1043         const char *c;
1045         root->tree = t = new_tree_content(8);
1046         if (is_null_sha1(sha1))
1047                 return;
1049         myoe = find_object(sha1);
1050         if (myoe) {
1051                 if (myoe->type != OBJ_TREE)
1052                         die("Not a tree: %s", sha1_to_hex(sha1));
1053                 t->delta_depth = 0;
1054                 buf = gfi_unpack_entry(myoe, &size);
1055         } else {
1056                 enum object_type type;
1057                 buf = read_sha1_file(sha1, &type, &size);
1058                 if (!buf || type != OBJ_TREE)
1059                         die("Can't load tree %s", sha1_to_hex(sha1));
1060         }
1062         c = buf;
1063         while (c != (buf + size)) {
1064                 struct tree_entry *e = new_tree_entry();
1066                 if (t->entry_count == t->entry_capacity)
1067                         root->tree = t = grow_tree_content(t, t->entry_count);
1068                 t->entries[t->entry_count++] = e;
1070                 e->tree = NULL;
1071                 c = get_mode(c, &e->versions[1].mode);
1072                 if (!c)
1073                         die("Corrupt mode in %s", sha1_to_hex(sha1));
1074                 e->versions[0].mode = e->versions[1].mode;
1075                 e->name = to_atom(c, strlen(c));
1076                 c += e->name->str_len + 1;
1077                 hashcpy(e->versions[0].sha1, (unsigned char*)c);
1078                 hashcpy(e->versions[1].sha1, (unsigned char*)c);
1079                 c += 20;
1080         }
1081         free(buf);
1084 static int tecmp0 (const void *_a, const void *_b)
1086         struct tree_entry *a = *((struct tree_entry**)_a);
1087         struct tree_entry *b = *((struct tree_entry**)_b);
1088         return base_name_compare(
1089                 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1090                 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1093 static int tecmp1 (const void *_a, const void *_b)
1095         struct tree_entry *a = *((struct tree_entry**)_a);
1096         struct tree_entry *b = *((struct tree_entry**)_b);
1097         return base_name_compare(
1098                 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1099                 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1102 static void mktree(struct tree_content *t,
1103         int v,
1104         unsigned long *szp,
1105         struct dbuf *b)
1107         size_t maxlen = 0;
1108         unsigned int i;
1109         char *c;
1111         if (!v)
1112                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp0);
1113         else
1114                 qsort(t->entries,t->entry_count,sizeof(t->entries[0]),tecmp1);
1116         for (i = 0; i < t->entry_count; i++) {
1117                 if (t->entries[i]->versions[v].mode)
1118                         maxlen += t->entries[i]->name->str_len + 34;
1119         }
1121         size_dbuf(b, maxlen);
1122         c = b->buffer;
1123         for (i = 0; i < t->entry_count; i++) {
1124                 struct tree_entry *e = t->entries[i];
1125                 if (!e->versions[v].mode)
1126                         continue;
1127                 c += sprintf(c, "%o", (unsigned int)e->versions[v].mode);
1128                 *c++ = ' ';
1129                 strcpy(c, e->name->str_dat);
1130                 c += e->name->str_len + 1;
1131                 hashcpy((unsigned char*)c, e->versions[v].sha1);
1132                 c += 20;
1133         }
1134         *szp = c - (char*)b->buffer;
1137 static void store_tree(struct tree_entry *root)
1139         struct tree_content *t = root->tree;
1140         unsigned int i, j, del;
1141         unsigned long new_len;
1142         struct last_object lo;
1143         struct object_entry *le;
1145         if (!is_null_sha1(root->versions[1].sha1))
1146                 return;
1148         for (i = 0; i < t->entry_count; i++) {
1149                 if (t->entries[i]->tree)
1150                         store_tree(t->entries[i]);
1151         }
1153         le = find_object(root->versions[0].sha1);
1154         if (!S_ISDIR(root->versions[0].mode)
1155                 || !le
1156                 || le->pack_id != pack_id) {
1157                 lo.data = NULL;
1158                 lo.depth = 0;
1159         } else {
1160                 mktree(t, 0, &lo.len, &old_tree);
1161                 lo.data = old_tree.buffer;
1162                 lo.offset = le->offset;
1163                 lo.depth = t->delta_depth;
1164                 lo.no_free = 1;
1165         }
1167         mktree(t, 1, &new_len, &new_tree);
1168         store_object(OBJ_TREE, new_tree.buffer, new_len,
1169                 &lo, root->versions[1].sha1, 0);
1171         t->delta_depth = lo.depth;
1172         for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1173                 struct tree_entry *e = t->entries[i];
1174                 if (e->versions[1].mode) {
1175                         e->versions[0].mode = e->versions[1].mode;
1176                         hashcpy(e->versions[0].sha1, e->versions[1].sha1);
1177                         t->entries[j++] = e;
1178                 } else {
1179                         release_tree_entry(e);
1180                         del++;
1181                 }
1182         }
1183         t->entry_count -= del;
1186 static int tree_content_set(
1187         struct tree_entry *root,
1188         const char *p,
1189         const unsigned char *sha1,
1190         const uint16_t mode)
1192         struct tree_content *t = root->tree;
1193         const char *slash1;
1194         unsigned int i, n;
1195         struct tree_entry *e;
1197         slash1 = strchr(p, '/');
1198         if (slash1)
1199                 n = slash1 - p;
1200         else
1201                 n = strlen(p);
1202         if (!n)
1203                 die("Empty path component found in input");
1205         for (i = 0; i < t->entry_count; i++) {
1206                 e = t->entries[i];
1207                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1208                         if (!slash1) {
1209                                 if (e->versions[1].mode == mode
1210                                                 && !hashcmp(e->versions[1].sha1, sha1))
1211                                         return 0;
1212                                 e->versions[1].mode = mode;
1213                                 hashcpy(e->versions[1].sha1, sha1);
1214                                 if (e->tree) {
1215                                         release_tree_content_recursive(e->tree);
1216                                         e->tree = NULL;
1217                                 }
1218                                 hashclr(root->versions[1].sha1);
1219                                 return 1;
1220                         }
1221                         if (!S_ISDIR(e->versions[1].mode)) {
1222                                 e->tree = new_tree_content(8);
1223                                 e->versions[1].mode = S_IFDIR;
1224                         }
1225                         if (!e->tree)
1226                                 load_tree(e);
1227                         if (tree_content_set(e, slash1 + 1, sha1, mode)) {
1228                                 hashclr(root->versions[1].sha1);
1229                                 return 1;
1230                         }
1231                         return 0;
1232                 }
1233         }
1235         if (t->entry_count == t->entry_capacity)
1236                 root->tree = t = grow_tree_content(t, t->entry_count);
1237         e = new_tree_entry();
1238         e->name = to_atom(p, n);
1239         e->versions[0].mode = 0;
1240         hashclr(e->versions[0].sha1);
1241         t->entries[t->entry_count++] = e;
1242         if (slash1) {
1243                 e->tree = new_tree_content(8);
1244                 e->versions[1].mode = S_IFDIR;
1245                 tree_content_set(e, slash1 + 1, sha1, mode);
1246         } else {
1247                 e->tree = NULL;
1248                 e->versions[1].mode = mode;
1249                 hashcpy(e->versions[1].sha1, sha1);
1250         }
1251         hashclr(root->versions[1].sha1);
1252         return 1;
1255 static int tree_content_remove(struct tree_entry *root, const char *p)
1257         struct tree_content *t = root->tree;
1258         const char *slash1;
1259         unsigned int i, n;
1260         struct tree_entry *e;
1262         slash1 = strchr(p, '/');
1263         if (slash1)
1264                 n = slash1 - p;
1265         else
1266                 n = strlen(p);
1268         for (i = 0; i < t->entry_count; i++) {
1269                 e = t->entries[i];
1270                 if (e->name->str_len == n && !strncmp(p, e->name->str_dat, n)) {
1271                         if (!slash1 || !S_ISDIR(e->versions[1].mode))
1272                                 goto del_entry;
1273                         if (!e->tree)
1274                                 load_tree(e);
1275                         if (tree_content_remove(e, slash1 + 1)) {
1276                                 for (n = 0; n < e->tree->entry_count; n++) {
1277                                         if (e->tree->entries[n]->versions[1].mode) {
1278                                                 hashclr(root->versions[1].sha1);
1279                                                 return 1;
1280                                         }
1281                                 }
1282                                 goto del_entry;
1283                         }
1284                         return 0;
1285                 }
1286         }
1287         return 0;
1289 del_entry:
1290         if (e->tree) {
1291                 release_tree_content_recursive(e->tree);
1292                 e->tree = NULL;
1293         }
1294         e->versions[1].mode = 0;
1295         hashclr(e->versions[1].sha1);
1296         hashclr(root->versions[1].sha1);
1297         return 1;
1300 static int update_branch(struct branch *b)
1302         static const char *msg = "fast-import";
1303         struct ref_lock *lock;
1304         unsigned char old_sha1[20];
1306         if (read_ref(b->name, old_sha1))
1307                 hashclr(old_sha1);
1308         lock = lock_any_ref_for_update(b->name, old_sha1);
1309         if (!lock)
1310                 return error("Unable to lock %s", b->name);
1311         if (!force_update && !is_null_sha1(old_sha1)) {
1312                 struct commit *old_cmit, *new_cmit;
1314                 old_cmit = lookup_commit_reference_gently(old_sha1, 0);
1315                 new_cmit = lookup_commit_reference_gently(b->sha1, 0);
1316                 if (!old_cmit || !new_cmit) {
1317                         unlock_ref(lock);
1318                         return error("Branch %s is missing commits.", b->name);
1319                 }
1321                 if (!in_merge_bases(old_cmit, &new_cmit, 1)) {
1322                         unlock_ref(lock);
1323                         warning("Not updating %s"
1324                                 " (new tip %s does not contain %s)",
1325                                 b->name, sha1_to_hex(b->sha1), sha1_to_hex(old_sha1));
1326                         return -1;
1327                 }
1328         }
1329         if (write_ref_sha1(lock, b->sha1, msg) < 0)
1330                 return error("Unable to update %s", b->name);
1331         return 0;
1334 static void dump_branches(void)
1336         unsigned int i;
1337         struct branch *b;
1339         for (i = 0; i < branch_table_sz; i++) {
1340                 for (b = branch_table[i]; b; b = b->table_next_branch)
1341                         failure |= update_branch(b);
1342         }
1345 static void dump_tags(void)
1347         static const char *msg = "fast-import";
1348         struct tag *t;
1349         struct ref_lock *lock;
1350         char ref_name[PATH_MAX];
1352         for (t = first_tag; t; t = t->next_tag) {
1353                 sprintf(ref_name, "tags/%s", t->name);
1354                 lock = lock_ref_sha1(ref_name, NULL);
1355                 if (!lock || write_ref_sha1(lock, t->sha1, msg) < 0)
1356                         failure |= error("Unable to update %s", ref_name);
1357         }
1360 static void dump_marks_helper(FILE *f,
1361         uintmax_t base,
1362         struct mark_set *m)
1364         uintmax_t k;
1365         if (m->shift) {
1366                 for (k = 0; k < 1024; k++) {
1367                         if (m->data.sets[k])
1368                                 dump_marks_helper(f, (base + k) << m->shift,
1369                                         m->data.sets[k]);
1370                 }
1371         } else {
1372                 for (k = 0; k < 1024; k++) {
1373                         if (m->data.marked[k])
1374                                 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1375                                         sha1_to_hex(m->data.marked[k]->sha1));
1376                 }
1377         }
1380 static void dump_marks(void)
1382         static struct lock_file mark_lock;
1383         int mark_fd;
1384         FILE *f;
1386         if (!mark_file)
1387                 return;
1389         mark_fd = hold_lock_file_for_update(&mark_lock, mark_file, 0);
1390         if (mark_fd < 0) {
1391                 failure |= error("Unable to write marks file %s: %s",
1392                         mark_file, strerror(errno));
1393                 return;
1394         }
1396         f = fdopen(mark_fd, "w");
1397         if (!f) {
1398                 rollback_lock_file(&mark_lock);
1399                 failure |= error("Unable to write marks file %s: %s",
1400                         mark_file, strerror(errno));
1401                 return;
1402         }
1404         dump_marks_helper(f, 0, marks);
1405         fclose(f);
1406         if (commit_lock_file(&mark_lock))
1407                 failure |= error("Unable to write marks file %s: %s",
1408                         mark_file, strerror(errno));
1411 static void read_next_command(void)
1413         read_line(&command_buf, stdin, '\n');
1416 static void cmd_mark(void)
1418         if (!prefixcmp(command_buf.buf, "mark :")) {
1419                 next_mark = strtoumax(command_buf.buf + 6, NULL, 10);
1420                 read_next_command();
1421         }
1422         else
1423                 next_mark = 0;
1426 static void *cmd_data (size_t *size)
1428         size_t length;
1429         char *buffer;
1431         if (prefixcmp(command_buf.buf, "data "))
1432                 die("Expected 'data n' command, found: %s", command_buf.buf);
1434         if (!prefixcmp(command_buf.buf + 5, "<<")) {
1435                 char *term = xstrdup(command_buf.buf + 5 + 2);
1436                 size_t sz = 8192, term_len = command_buf.len - 5 - 2;
1437                 length = 0;
1438                 buffer = xmalloc(sz);
1439                 for (;;) {
1440                         read_next_command();
1441                         if (command_buf.eof)
1442                                 die("EOF in data (terminator '%s' not found)", term);
1443                         if (term_len == command_buf.len
1444                                 && !strcmp(term, command_buf.buf))
1445                                 break;
1446                         if (sz < (length + command_buf.len)) {
1447                                 sz = sz * 3 / 2 + 16;
1448                                 if (sz < (length + command_buf.len))
1449                                         sz = length + command_buf.len;
1450                                 buffer = xrealloc(buffer, sz);
1451                         }
1452                         memcpy(buffer + length,
1453                                 command_buf.buf,
1454                                 command_buf.len - 1);
1455                         length += command_buf.len - 1;
1456                         buffer[length++] = '\n';
1457                 }
1458                 free(term);
1459         }
1460         else {
1461                 size_t n = 0;
1462                 length = strtoul(command_buf.buf + 5, NULL, 10);
1463                 buffer = xmalloc(length);
1464                 while (n < length) {
1465                         size_t s = fread(buffer + n, 1, length - n, stdin);
1466                         if (!s && feof(stdin))
1467                                 die("EOF in data (%lu bytes remaining)",
1468                                         (unsigned long)(length - n));
1469                         n += s;
1470                 }
1471         }
1473         if (fgetc(stdin) != '\n')
1474                 die("An lf did not trail the binary data as expected.");
1476         *size = length;
1477         return buffer;
1480 static int validate_raw_date(const char *src, char *result, int maxlen)
1482         const char *orig_src = src;
1483         char *endp, sign;
1485         strtoul(src, &endp, 10);
1486         if (endp == src || *endp != ' ')
1487                 return -1;
1489         src = endp + 1;
1490         if (*src != '-' && *src != '+')
1491                 return -1;
1492         sign = *src;
1494         strtoul(src + 1, &endp, 10);
1495         if (endp == src || *endp || (endp - orig_src) >= maxlen)
1496                 return -1;
1498         strcpy(result, orig_src);
1499         return 0;
1502 static char *parse_ident(const char *buf)
1504         const char *gt;
1505         size_t name_len;
1506         char *ident;
1508         gt = strrchr(buf, '>');
1509         if (!gt)
1510                 die("Missing > in ident string: %s", buf);
1511         gt++;
1512         if (*gt != ' ')
1513                 die("Missing space after > in ident string: %s", buf);
1514         gt++;
1515         name_len = gt - buf;
1516         ident = xmalloc(name_len + 24);
1517         strncpy(ident, buf, name_len);
1519         switch (whenspec) {
1520         case WHENSPEC_RAW:
1521                 if (validate_raw_date(gt, ident + name_len, 24) < 0)
1522                         die("Invalid raw date \"%s\" in ident: %s", gt, buf);
1523                 break;
1524         case WHENSPEC_RFC2822:
1525                 if (parse_date(gt, ident + name_len, 24) < 0)
1526                         die("Invalid rfc2822 date \"%s\" in ident: %s", gt, buf);
1527                 break;
1528         case WHENSPEC_NOW:
1529                 if (strcmp("now", gt))
1530                         die("Date in ident must be 'now': %s", buf);
1531                 datestamp(ident + name_len, 24);
1532                 break;
1533         }
1535         return ident;
1538 static void cmd_new_blob(void)
1540         size_t l;
1541         void *d;
1543         read_next_command();
1544         cmd_mark();
1545         d = cmd_data(&l);
1547         if (store_object(OBJ_BLOB, d, l, &last_blob, NULL, next_mark))
1548                 free(d);
1551 static void unload_one_branch(void)
1553         while (cur_active_branches
1554                 && cur_active_branches >= max_active_branches) {
1555                 uintmax_t min_commit = ULONG_MAX;
1556                 struct branch *e, *l = NULL, *p = NULL;
1558                 for (e = active_branches; e; e = e->active_next_branch) {
1559                         if (e->last_commit < min_commit) {
1560                                 p = l;
1561                                 min_commit = e->last_commit;
1562                         }
1563                         l = e;
1564                 }
1566                 if (p) {
1567                         e = p->active_next_branch;
1568                         p->active_next_branch = e->active_next_branch;
1569                 } else {
1570                         e = active_branches;
1571                         active_branches = e->active_next_branch;
1572                 }
1573                 e->active = 0;
1574                 e->active_next_branch = NULL;
1575                 if (e->branch_tree.tree) {
1576                         release_tree_content_recursive(e->branch_tree.tree);
1577                         e->branch_tree.tree = NULL;
1578                 }
1579                 cur_active_branches--;
1580         }
1583 static void load_branch(struct branch *b)
1585         load_tree(&b->branch_tree);
1586         if (!b->active) {
1587                 b->active = 1;
1588                 b->active_next_branch = active_branches;
1589                 active_branches = b;
1590                 cur_active_branches++;
1591                 branch_load_count++;
1592         }
1595 static void file_change_m(struct branch *b)
1597         const char *p = command_buf.buf + 2;
1598         char *p_uq;
1599         const char *endp;
1600         struct object_entry *oe = oe;
1601         unsigned char sha1[20];
1602         uint16_t mode, inline_data = 0;
1604         p = get_mode(p, &mode);
1605         if (!p)
1606                 die("Corrupt mode: %s", command_buf.buf);
1607         switch (mode) {
1608         case S_IFREG | 0644:
1609         case S_IFREG | 0755:
1610         case S_IFLNK:
1611         case 0644:
1612         case 0755:
1613                 /* ok */
1614                 break;
1615         default:
1616                 die("Corrupt mode: %s", command_buf.buf);
1617         }
1619         if (*p == ':') {
1620                 char *x;
1621                 oe = find_mark(strtoumax(p + 1, &x, 10));
1622                 hashcpy(sha1, oe->sha1);
1623                 p = x;
1624         } else if (!prefixcmp(p, "inline")) {
1625                 inline_data = 1;
1626                 p += 6;
1627         } else {
1628                 if (get_sha1_hex(p, sha1))
1629                         die("Invalid SHA1: %s", command_buf.buf);
1630                 oe = find_object(sha1);
1631                 p += 40;
1632         }
1633         if (*p++ != ' ')
1634                 die("Missing space after SHA1: %s", command_buf.buf);
1636         p_uq = unquote_c_style(p, &endp);
1637         if (p_uq) {
1638                 if (*endp)
1639                         die("Garbage after path in: %s", command_buf.buf);
1640                 p = p_uq;
1641         }
1643         if (inline_data) {
1644                 size_t l;
1645                 void *d;
1646                 if (!p_uq)
1647                         p = p_uq = xstrdup(p);
1648                 read_next_command();
1649                 d = cmd_data(&l);
1650                 if (store_object(OBJ_BLOB, d, l, &last_blob, sha1, 0))
1651                         free(d);
1652         } else if (oe) {
1653                 if (oe->type != OBJ_BLOB)
1654                         die("Not a blob (actually a %s): %s",
1655                                 command_buf.buf, typename(oe->type));
1656         } else {
1657                 enum object_type type = sha1_object_info(sha1, NULL);
1658                 if (type < 0)
1659                         die("Blob not found: %s", command_buf.buf);
1660                 if (type != OBJ_BLOB)
1661                         die("Not a blob (actually a %s): %s",
1662                             typename(type), command_buf.buf);
1663         }
1665         tree_content_set(&b->branch_tree, p, sha1, S_IFREG | mode);
1666         free(p_uq);
1669 static void file_change_d(struct branch *b)
1671         const char *p = command_buf.buf + 2;
1672         char *p_uq;
1673         const char *endp;
1675         p_uq = unquote_c_style(p, &endp);
1676         if (p_uq) {
1677                 if (*endp)
1678                         die("Garbage after path in: %s", command_buf.buf);
1679                 p = p_uq;
1680         }
1681         tree_content_remove(&b->branch_tree, p);
1682         free(p_uq);
1685 static void file_change_deleteall(struct branch *b)
1687         release_tree_content_recursive(b->branch_tree.tree);
1688         hashclr(b->branch_tree.versions[0].sha1);
1689         hashclr(b->branch_tree.versions[1].sha1);
1690         load_tree(&b->branch_tree);
1693 static void cmd_from(struct branch *b)
1695         const char *from;
1696         struct branch *s;
1698         if (prefixcmp(command_buf.buf, "from "))
1699                 return;
1701         if (b->branch_tree.tree) {
1702                 release_tree_content_recursive(b->branch_tree.tree);
1703                 b->branch_tree.tree = NULL;
1704         }
1706         from = strchr(command_buf.buf, ' ') + 1;
1707         s = lookup_branch(from);
1708         if (b == s)
1709                 die("Can't create a branch from itself: %s", b->name);
1710         else if (s) {
1711                 unsigned char *t = s->branch_tree.versions[1].sha1;
1712                 hashcpy(b->sha1, s->sha1);
1713                 hashcpy(b->branch_tree.versions[0].sha1, t);
1714                 hashcpy(b->branch_tree.versions[1].sha1, t);
1715         } else if (*from == ':') {
1716                 uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1717                 struct object_entry *oe = find_mark(idnum);
1718                 unsigned long size;
1719                 char *buf;
1720                 if (oe->type != OBJ_COMMIT)
1721                         die("Mark :%" PRIuMAX " not a commit", idnum);
1722                 hashcpy(b->sha1, oe->sha1);
1723                 buf = gfi_unpack_entry(oe, &size);
1724                 if (!buf || size < 46)
1725                         die("Not a valid commit: %s", from);
1726                 if (memcmp("tree ", buf, 5)
1727                         || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1728                         die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1729                 free(buf);
1730                 hashcpy(b->branch_tree.versions[0].sha1,
1731                         b->branch_tree.versions[1].sha1);
1732         } else if (!get_sha1(from, b->sha1)) {
1733                 if (is_null_sha1(b->sha1)) {
1734                         hashclr(b->branch_tree.versions[0].sha1);
1735                         hashclr(b->branch_tree.versions[1].sha1);
1736                 } else {
1737                         unsigned long size;
1738                         char *buf;
1740                         buf = read_object_with_reference(b->sha1,
1741                                 commit_type, &size, b->sha1);
1742                         if (!buf || size < 46)
1743                                 die("Not a valid commit: %s", from);
1744                         if (memcmp("tree ", buf, 5)
1745                                 || get_sha1_hex(buf + 5, b->branch_tree.versions[1].sha1))
1746                                 die("The commit %s is corrupt", sha1_to_hex(b->sha1));
1747                         free(buf);
1748                         hashcpy(b->branch_tree.versions[0].sha1,
1749                                 b->branch_tree.versions[1].sha1);
1750                 }
1751         } else
1752                 die("Invalid ref name or SHA1 expression: %s", from);
1754         read_next_command();
1757 static struct hash_list *cmd_merge(unsigned int *count)
1759         struct hash_list *list = NULL, *n, *e = e;
1760         const char *from;
1761         struct branch *s;
1763         *count = 0;
1764         while (!prefixcmp(command_buf.buf, "merge ")) {
1765                 from = strchr(command_buf.buf, ' ') + 1;
1766                 n = xmalloc(sizeof(*n));
1767                 s = lookup_branch(from);
1768                 if (s)
1769                         hashcpy(n->sha1, s->sha1);
1770                 else if (*from == ':') {
1771                         uintmax_t idnum = strtoumax(from + 1, NULL, 10);
1772                         struct object_entry *oe = find_mark(idnum);
1773                         if (oe->type != OBJ_COMMIT)
1774                                 die("Mark :%" PRIuMAX " not a commit", idnum);
1775                         hashcpy(n->sha1, oe->sha1);
1776                 } else if (!get_sha1(from, n->sha1)) {
1777                         unsigned long size;
1778                         char *buf = read_object_with_reference(n->sha1,
1779                                 commit_type, &size, n->sha1);
1780                         if (!buf || size < 46)
1781                                 die("Not a valid commit: %s", from);
1782                         free(buf);
1783                 } else
1784                         die("Invalid ref name or SHA1 expression: %s", from);
1786                 n->next = NULL;
1787                 if (list)
1788                         e->next = n;
1789                 else
1790                         list = n;
1791                 e = n;
1792                 (*count)++;
1793                 read_next_command();
1794         }
1795         return list;
1798 static void cmd_new_commit(void)
1800         struct branch *b;
1801         void *msg;
1802         size_t msglen;
1803         char *sp;
1804         char *author = NULL;
1805         char *committer = NULL;
1806         struct hash_list *merge_list = NULL;
1807         unsigned int merge_count;
1809         /* Obtain the branch name from the rest of our command */
1810         sp = strchr(command_buf.buf, ' ') + 1;
1811         b = lookup_branch(sp);
1812         if (!b)
1813                 b = new_branch(sp);
1815         read_next_command();
1816         cmd_mark();
1817         if (!prefixcmp(command_buf.buf, "author ")) {
1818                 author = parse_ident(command_buf.buf + 7);
1819                 read_next_command();
1820         }
1821         if (!prefixcmp(command_buf.buf, "committer ")) {
1822                 committer = parse_ident(command_buf.buf + 10);
1823                 read_next_command();
1824         }
1825         if (!committer)
1826                 die("Expected committer but didn't get one");
1827         msg = cmd_data(&msglen);
1828         read_next_command();
1829         cmd_from(b);
1830         merge_list = cmd_merge(&merge_count);
1832         /* ensure the branch is active/loaded */
1833         if (!b->branch_tree.tree || !max_active_branches) {
1834                 unload_one_branch();
1835                 load_branch(b);
1836         }
1838         /* file_change* */
1839         for (;;) {
1840                 if (1 == command_buf.len)
1841                         break;
1842                 else if (!prefixcmp(command_buf.buf, "M "))
1843                         file_change_m(b);
1844                 else if (!prefixcmp(command_buf.buf, "D "))
1845                         file_change_d(b);
1846                 else if (!strcmp("deleteall", command_buf.buf))
1847                         file_change_deleteall(b);
1848                 else
1849                         die("Unsupported file_change: %s", command_buf.buf);
1850                 read_next_command();
1851         }
1853         /* build the tree and the commit */
1854         store_tree(&b->branch_tree);
1855         hashcpy(b->branch_tree.versions[0].sha1,
1856                 b->branch_tree.versions[1].sha1);
1857         size_dbuf(&new_data, 114 + msglen
1858                 + merge_count * 49
1859                 + (author
1860                         ? strlen(author) + strlen(committer)
1861                         : 2 * strlen(committer)));
1862         sp = new_data.buffer;
1863         sp += sprintf(sp, "tree %s\n",
1864                 sha1_to_hex(b->branch_tree.versions[1].sha1));
1865         if (!is_null_sha1(b->sha1))
1866                 sp += sprintf(sp, "parent %s\n", sha1_to_hex(b->sha1));
1867         while (merge_list) {
1868                 struct hash_list *next = merge_list->next;
1869                 sp += sprintf(sp, "parent %s\n", sha1_to_hex(merge_list->sha1));
1870                 free(merge_list);
1871                 merge_list = next;
1872         }
1873         sp += sprintf(sp, "author %s\n", author ? author : committer);
1874         sp += sprintf(sp, "committer %s\n", committer);
1875         *sp++ = '\n';
1876         memcpy(sp, msg, msglen);
1877         sp += msglen;
1878         free(author);
1879         free(committer);
1880         free(msg);
1882         if (!store_object(OBJ_COMMIT,
1883                 new_data.buffer, sp - (char*)new_data.buffer,
1884                 NULL, b->sha1, next_mark))
1885                 b->pack_id = pack_id;
1886         b->last_commit = object_count_by_type[OBJ_COMMIT];
1889 static void cmd_new_tag(void)
1891         char *sp;
1892         const char *from;
1893         char *tagger;
1894         struct branch *s;
1895         void *msg;
1896         size_t msglen;
1897         struct tag *t;
1898         uintmax_t from_mark = 0;
1899         unsigned char sha1[20];
1901         /* Obtain the new tag name from the rest of our command */
1902         sp = strchr(command_buf.buf, ' ') + 1;
1903         t = pool_alloc(sizeof(struct tag));
1904         t->next_tag = NULL;
1905         t->name = pool_strdup(sp);
1906         if (last_tag)
1907                 last_tag->next_tag = t;
1908         else
1909                 first_tag = t;
1910         last_tag = t;
1911         read_next_command();
1913         /* from ... */
1914         if (prefixcmp(command_buf.buf, "from "))
1915                 die("Expected from command, got %s", command_buf.buf);
1916         from = strchr(command_buf.buf, ' ') + 1;
1917         s = lookup_branch(from);
1918         if (s) {
1919                 hashcpy(sha1, s->sha1);
1920         } else if (*from == ':') {
1921                 struct object_entry *oe;
1922                 from_mark = strtoumax(from + 1, NULL, 10);
1923                 oe = find_mark(from_mark);
1924                 if (oe->type != OBJ_COMMIT)
1925                         die("Mark :%" PRIuMAX " not a commit", from_mark);
1926                 hashcpy(sha1, oe->sha1);
1927         } else if (!get_sha1(from, sha1)) {
1928                 unsigned long size;
1929                 char *buf;
1931                 buf = read_object_with_reference(sha1,
1932                         commit_type, &size, sha1);
1933                 if (!buf || size < 46)
1934                         die("Not a valid commit: %s", from);
1935                 free(buf);
1936         } else
1937                 die("Invalid ref name or SHA1 expression: %s", from);
1938         read_next_command();
1940         /* tagger ... */
1941         if (prefixcmp(command_buf.buf, "tagger "))
1942                 die("Expected tagger command, got %s", command_buf.buf);
1943         tagger = parse_ident(command_buf.buf + 7);
1945         /* tag payload/message */
1946         read_next_command();
1947         msg = cmd_data(&msglen);
1949         /* build the tag object */
1950         size_dbuf(&new_data, 67+strlen(t->name)+strlen(tagger)+msglen);
1951         sp = new_data.buffer;
1952         sp += sprintf(sp, "object %s\n", sha1_to_hex(sha1));
1953         sp += sprintf(sp, "type %s\n", commit_type);
1954         sp += sprintf(sp, "tag %s\n", t->name);
1955         sp += sprintf(sp, "tagger %s\n", tagger);
1956         *sp++ = '\n';
1957         memcpy(sp, msg, msglen);
1958         sp += msglen;
1959         free(tagger);
1960         free(msg);
1962         if (store_object(OBJ_TAG, new_data.buffer,
1963                 sp - (char*)new_data.buffer,
1964                 NULL, t->sha1, 0))
1965                 t->pack_id = MAX_PACK_ID;
1966         else
1967                 t->pack_id = pack_id;
1970 static void cmd_reset_branch(void)
1972         struct branch *b;
1973         char *sp;
1975         /* Obtain the branch name from the rest of our command */
1976         sp = strchr(command_buf.buf, ' ') + 1;
1977         b = lookup_branch(sp);
1978         if (b) {
1979                 hashclr(b->sha1);
1980                 hashclr(b->branch_tree.versions[0].sha1);
1981                 hashclr(b->branch_tree.versions[1].sha1);
1982                 if (b->branch_tree.tree) {
1983                         release_tree_content_recursive(b->branch_tree.tree);
1984                         b->branch_tree.tree = NULL;
1985                 }
1986         }
1987         else
1988                 b = new_branch(sp);
1989         read_next_command();
1990         cmd_from(b);
1993 static void cmd_checkpoint(void)
1995         if (object_count) {
1996                 cycle_packfile();
1997                 dump_branches();
1998                 dump_tags();
1999                 dump_marks();
2000         }
2001         read_next_command();
2004 static void import_marks(const char *input_file)
2006         char line[512];
2007         FILE *f = fopen(input_file, "r");
2008         if (!f)
2009                 die("cannot read %s: %s", input_file, strerror(errno));
2010         while (fgets(line, sizeof(line), f)) {
2011                 uintmax_t mark;
2012                 char *end;
2013                 unsigned char sha1[20];
2014                 struct object_entry *e;
2016                 end = strchr(line, '\n');
2017                 if (line[0] != ':' || !end)
2018                         die("corrupt mark line: %s", line);
2019                 *end = 0;
2020                 mark = strtoumax(line + 1, &end, 10);
2021                 if (!mark || end == line + 1
2022                         || *end != ' ' || get_sha1(end + 1, sha1))
2023                         die("corrupt mark line: %s", line);
2024                 e = find_object(sha1);
2025                 if (!e) {
2026                         enum object_type type = sha1_object_info(sha1, NULL);
2027                         if (type < 0)
2028                                 die("object not found: %s", sha1_to_hex(sha1));
2029                         e = insert_object(sha1);
2030                         e->type = type;
2031                         e->pack_id = MAX_PACK_ID;
2032                         e->offset = 1; /* just not zero! */
2033                 }
2034                 insert_mark(mark, e);
2035         }
2036         fclose(f);
2039 static const char fast_import_usage[] =
2040 "git-fast-import [--date-format=f] [--max-pack-size=n] [--depth=n] [--active-branches=n] [--export-marks=marks.file]";
2042 int main(int argc, const char **argv)
2044         int i, show_stats = 1;
2046         git_config(git_default_config);
2047         alloc_objects(object_entry_alloc);
2048         strbuf_init(&command_buf);
2049         atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
2050         branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
2051         avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
2052         marks = pool_calloc(1, sizeof(struct mark_set));
2054         for (i = 1; i < argc; i++) {
2055                 const char *a = argv[i];
2057                 if (*a != '-' || !strcmp(a, "--"))
2058                         break;
2059                 else if (!prefixcmp(a, "--date-format=")) {
2060                         const char *fmt = a + 14;
2061                         if (!strcmp(fmt, "raw"))
2062                                 whenspec = WHENSPEC_RAW;
2063                         else if (!strcmp(fmt, "rfc2822"))
2064                                 whenspec = WHENSPEC_RFC2822;
2065                         else if (!strcmp(fmt, "now"))
2066                                 whenspec = WHENSPEC_NOW;
2067                         else
2068                                 die("unknown --date-format argument %s", fmt);
2069                 }
2070                 else if (!prefixcmp(a, "--max-pack-size="))
2071                         max_packsize = strtoumax(a + 16, NULL, 0) * 1024 * 1024;
2072                 else if (!prefixcmp(a, "--depth="))
2073                         max_depth = strtoul(a + 8, NULL, 0);
2074                 else if (!prefixcmp(a, "--active-branches="))
2075                         max_active_branches = strtoul(a + 18, NULL, 0);
2076                 else if (!prefixcmp(a, "--import-marks="))
2077                         import_marks(a + 15);
2078                 else if (!prefixcmp(a, "--export-marks="))
2079                         mark_file = a + 15;
2080                 else if (!prefixcmp(a, "--export-pack-edges=")) {
2081                         if (pack_edges)
2082                                 fclose(pack_edges);
2083                         pack_edges = fopen(a + 20, "a");
2084                         if (!pack_edges)
2085                                 die("Cannot open %s: %s", a + 20, strerror(errno));
2086                 } else if (!strcmp(a, "--force"))
2087                         force_update = 1;
2088                 else if (!strcmp(a, "--quiet"))
2089                         show_stats = 0;
2090                 else if (!strcmp(a, "--stats"))
2091                         show_stats = 1;
2092                 else
2093                         die("unknown option %s", a);
2094         }
2095         if (i != argc)
2096                 usage(fast_import_usage);
2098         prepare_packed_git();
2099         start_packfile();
2100         for (;;) {
2101                 read_next_command();
2102                 if (command_buf.eof)
2103                         break;
2104                 else if (!strcmp("blob", command_buf.buf))
2105                         cmd_new_blob();
2106                 else if (!prefixcmp(command_buf.buf, "commit "))
2107                         cmd_new_commit();
2108                 else if (!prefixcmp(command_buf.buf, "tag "))
2109                         cmd_new_tag();
2110                 else if (!prefixcmp(command_buf.buf, "reset "))
2111                         cmd_reset_branch();
2112                 else if (!strcmp("checkpoint", command_buf.buf))
2113                         cmd_checkpoint();
2114                 else
2115                         die("Unsupported command: %s", command_buf.buf);
2116         }
2117         end_packfile();
2119         dump_branches();
2120         dump_tags();
2121         unkeep_all_packs();
2122         dump_marks();
2124         if (pack_edges)
2125                 fclose(pack_edges);
2127         if (show_stats) {
2128                 uintmax_t total_count = 0, duplicate_count = 0;
2129                 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
2130                         total_count += object_count_by_type[i];
2131                 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
2132                         duplicate_count += duplicate_count_by_type[i];
2134                 fprintf(stderr, "%s statistics:\n", argv[0]);
2135                 fprintf(stderr, "---------------------------------------------------------------------\n");
2136                 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
2137                 fprintf(stderr, "Total objects:   %10" PRIuMAX " (%10" PRIuMAX " duplicates                  )\n", total_count, duplicate_count);
2138                 fprintf(stderr, "      blobs  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB]);
2139                 fprintf(stderr, "      trees  :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE]);
2140                 fprintf(stderr, "      commits:   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT]);
2141                 fprintf(stderr, "      tags   :   %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG]);
2142                 fprintf(stderr, "Total branches:  %10lu (%10lu loads     )\n", branch_count, branch_load_count);
2143                 fprintf(stderr, "      marks:     %10" PRIuMAX " (%10" PRIuMAX " unique    )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
2144                 fprintf(stderr, "      atoms:     %10u\n", atom_cnt);
2145                 fprintf(stderr, "Memory total:    %10" PRIuMAX " KiB\n", (total_allocd + alloc_count*sizeof(struct object_entry))/1024);
2146                 fprintf(stderr, "       pools:    %10lu KiB\n", (unsigned long)(total_allocd/1024));
2147                 fprintf(stderr, "     objects:    %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
2148                 fprintf(stderr, "---------------------------------------------------------------------\n");
2149                 pack_report();
2150                 fprintf(stderr, "---------------------------------------------------------------------\n");
2151                 fprintf(stderr, "\n");
2152         }
2154         return failure ? 1 : 0;