]> git.ipfire.org Git - thirdparty/git.git/blob - fast-import.c
The twentieth batch
[thirdparty/git.git] / fast-import.c
1 #include "builtin.h"
2 #include "cache.h"
3 #include "repository.h"
4 #include "config.h"
5 #include "lockfile.h"
6 #include "object.h"
7 #include "blob.h"
8 #include "tree.h"
9 #include "commit.h"
10 #include "delta.h"
11 #include "pack.h"
12 #include "refs.h"
13 #include "csum-file.h"
14 #include "quote.h"
15 #include "dir.h"
16 #include "run-command.h"
17 #include "packfile.h"
18 #include "object-store.h"
19 #include "mem-pool.h"
20 #include "commit-reach.h"
21
22 #define PACK_ID_BITS 16
23 #define MAX_PACK_ID ((1<<PACK_ID_BITS)-1)
24 #define DEPTH_BITS 13
25 #define MAX_DEPTH ((1<<DEPTH_BITS)-1)
26
27 /*
28 * We abuse the setuid bit on directories to mean "do not delta".
29 */
30 #define NO_DELTA S_ISUID
31
32 /*
33 * The amount of additional space required in order to write an object into the
34 * current pack. This is the hash lengths at the end of the pack, plus the
35 * length of one object ID.
36 */
37 #define PACK_SIZE_THRESHOLD (the_hash_algo->rawsz * 3)
38
39 struct object_entry {
40 struct pack_idx_entry idx;
41 struct object_entry *next;
42 uint32_t type : TYPE_BITS,
43 pack_id : PACK_ID_BITS,
44 depth : DEPTH_BITS;
45 };
46
47 struct object_entry_pool {
48 struct object_entry_pool *next_pool;
49 struct object_entry *next_free;
50 struct object_entry *end;
51 struct object_entry entries[FLEX_ARRAY]; /* more */
52 };
53
54 struct mark_set {
55 union {
56 struct object_entry *marked[1024];
57 struct mark_set *sets[1024];
58 } data;
59 unsigned int shift;
60 };
61
62 struct last_object {
63 struct strbuf data;
64 off_t offset;
65 unsigned int depth;
66 unsigned no_swap : 1;
67 };
68
69 struct atom_str {
70 struct atom_str *next_atom;
71 unsigned short str_len;
72 char str_dat[FLEX_ARRAY]; /* more */
73 };
74
75 struct tree_content;
76 struct tree_entry {
77 struct tree_content *tree;
78 struct atom_str *name;
79 struct tree_entry_ms {
80 uint16_t mode;
81 struct object_id oid;
82 } versions[2];
83 };
84
85 struct tree_content {
86 unsigned int entry_capacity; /* must match avail_tree_content */
87 unsigned int entry_count;
88 unsigned int delta_depth;
89 struct tree_entry *entries[FLEX_ARRAY]; /* more */
90 };
91
92 struct avail_tree_content {
93 unsigned int entry_capacity; /* must match tree_content */
94 struct avail_tree_content *next_avail;
95 };
96
97 struct branch {
98 struct branch *table_next_branch;
99 struct branch *active_next_branch;
100 const char *name;
101 struct tree_entry branch_tree;
102 uintmax_t last_commit;
103 uintmax_t num_notes;
104 unsigned active : 1;
105 unsigned delete : 1;
106 unsigned pack_id : PACK_ID_BITS;
107 struct object_id oid;
108 };
109
110 struct tag {
111 struct tag *next_tag;
112 const char *name;
113 unsigned int pack_id;
114 struct object_id oid;
115 };
116
117 struct hash_list {
118 struct hash_list *next;
119 struct object_id oid;
120 };
121
122 typedef enum {
123 WHENSPEC_RAW = 1,
124 WHENSPEC_RFC2822,
125 WHENSPEC_NOW
126 } whenspec_type;
127
128 struct recent_command {
129 struct recent_command *prev;
130 struct recent_command *next;
131 char *buf;
132 };
133
134 /* Configured limits on output */
135 static unsigned long max_depth = 50;
136 static off_t max_packsize;
137 static int unpack_limit = 100;
138 static int force_update;
139
140 /* Stats and misc. counters */
141 static uintmax_t alloc_count;
142 static uintmax_t marks_set_count;
143 static uintmax_t object_count_by_type[1 << TYPE_BITS];
144 static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
145 static uintmax_t delta_count_by_type[1 << TYPE_BITS];
146 static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
147 static unsigned long object_count;
148 static unsigned long branch_count;
149 static unsigned long branch_load_count;
150 static int failure;
151 static FILE *pack_edges;
152 static unsigned int show_stats = 1;
153 static int global_argc;
154 static const char **global_argv;
155
156 /* Memory pools */
157 static struct mem_pool fi_mem_pool = {NULL, 2*1024*1024 -
158 sizeof(struct mp_block), 0 };
159
160 /* Atom management */
161 static unsigned int atom_table_sz = 4451;
162 static unsigned int atom_cnt;
163 static struct atom_str **atom_table;
164
165 /* The .pack file being generated */
166 static struct pack_idx_option pack_idx_opts;
167 static unsigned int pack_id;
168 static struct hashfile *pack_file;
169 static struct packed_git *pack_data;
170 static struct packed_git **all_packs;
171 static off_t pack_size;
172
173 /* Table of objects we've written. */
174 static unsigned int object_entry_alloc = 5000;
175 static struct object_entry_pool *blocks;
176 static struct object_entry *object_table[1 << 16];
177 static struct mark_set *marks;
178 static const char *export_marks_file;
179 static const char *import_marks_file;
180 static int import_marks_file_from_stream;
181 static int import_marks_file_ignore_missing;
182 static int import_marks_file_done;
183 static int relative_marks_paths;
184
185 /* Our last blob */
186 static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
187
188 /* Tree management */
189 static unsigned int tree_entry_alloc = 1000;
190 static void *avail_tree_entry;
191 static unsigned int avail_tree_table_sz = 100;
192 static struct avail_tree_content **avail_tree_table;
193 static size_t tree_entry_allocd;
194 static struct strbuf old_tree = STRBUF_INIT;
195 static struct strbuf new_tree = STRBUF_INIT;
196
197 /* Branch data */
198 static unsigned long max_active_branches = 5;
199 static unsigned long cur_active_branches;
200 static unsigned long branch_table_sz = 1039;
201 static struct branch **branch_table;
202 static struct branch *active_branches;
203
204 /* Tag data */
205 static struct tag *first_tag;
206 static struct tag *last_tag;
207
208 /* Input stream parsing */
209 static whenspec_type whenspec = WHENSPEC_RAW;
210 static struct strbuf command_buf = STRBUF_INIT;
211 static int unread_command_buf;
212 static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
213 static struct recent_command *cmd_tail = &cmd_hist;
214 static struct recent_command *rc_free;
215 static unsigned int cmd_save = 100;
216 static uintmax_t next_mark;
217 static struct strbuf new_data = STRBUF_INIT;
218 static int seen_data_command;
219 static int require_explicit_termination;
220
221 /* Signal handling */
222 static volatile sig_atomic_t checkpoint_requested;
223
224 /* Where to write output of cat-blob commands */
225 static int cat_blob_fd = STDOUT_FILENO;
226
227 static void parse_argv(void);
228 static void parse_get_mark(const char *p);
229 static void parse_cat_blob(const char *p);
230 static void parse_ls(const char *p, struct branch *b);
231
232 static void write_branch_report(FILE *rpt, struct branch *b)
233 {
234 fprintf(rpt, "%s:\n", b->name);
235
236 fprintf(rpt, " status :");
237 if (b->active)
238 fputs(" active", rpt);
239 if (b->branch_tree.tree)
240 fputs(" loaded", rpt);
241 if (is_null_oid(&b->branch_tree.versions[1].oid))
242 fputs(" dirty", rpt);
243 fputc('\n', rpt);
244
245 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
246 fprintf(rpt, " old tree : %s\n",
247 oid_to_hex(&b->branch_tree.versions[0].oid));
248 fprintf(rpt, " cur tree : %s\n",
249 oid_to_hex(&b->branch_tree.versions[1].oid));
250 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
251
252 fputs(" last pack : ", rpt);
253 if (b->pack_id < MAX_PACK_ID)
254 fprintf(rpt, "%u", b->pack_id);
255 fputc('\n', rpt);
256
257 fputc('\n', rpt);
258 }
259
260 static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
261
262 static void write_crash_report(const char *err)
263 {
264 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
265 FILE *rpt = fopen(loc, "w");
266 struct branch *b;
267 unsigned long lu;
268 struct recent_command *rc;
269
270 if (!rpt) {
271 error_errno("can't write crash report %s", loc);
272 free(loc);
273 return;
274 }
275
276 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
277
278 fprintf(rpt, "fast-import crash report:\n");
279 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
280 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
281 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
282 fputc('\n', rpt);
283
284 fputs("fatal: ", rpt);
285 fputs(err, rpt);
286 fputc('\n', rpt);
287
288 fputc('\n', rpt);
289 fputs("Most Recent Commands Before Crash\n", rpt);
290 fputs("---------------------------------\n", rpt);
291 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
292 if (rc->next == &cmd_hist)
293 fputs("* ", rpt);
294 else
295 fputs(" ", rpt);
296 fputs(rc->buf, rpt);
297 fputc('\n', rpt);
298 }
299
300 fputc('\n', rpt);
301 fputs("Active Branch LRU\n", rpt);
302 fputs("-----------------\n", rpt);
303 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
304 cur_active_branches,
305 max_active_branches);
306 fputc('\n', rpt);
307 fputs(" pos clock name\n", rpt);
308 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
309 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
310 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
311 ++lu, b->last_commit, b->name);
312
313 fputc('\n', rpt);
314 fputs("Inactive Branches\n", rpt);
315 fputs("-----------------\n", rpt);
316 for (lu = 0; lu < branch_table_sz; lu++) {
317 for (b = branch_table[lu]; b; b = b->table_next_branch)
318 write_branch_report(rpt, b);
319 }
320
321 if (first_tag) {
322 struct tag *tg;
323 fputc('\n', rpt);
324 fputs("Annotated Tags\n", rpt);
325 fputs("--------------\n", rpt);
326 for (tg = first_tag; tg; tg = tg->next_tag) {
327 fputs(oid_to_hex(&tg->oid), rpt);
328 fputc(' ', rpt);
329 fputs(tg->name, rpt);
330 fputc('\n', rpt);
331 }
332 }
333
334 fputc('\n', rpt);
335 fputs("Marks\n", rpt);
336 fputs("-----\n", rpt);
337 if (export_marks_file)
338 fprintf(rpt, " exported to %s\n", export_marks_file);
339 else
340 dump_marks_helper(rpt, 0, marks);
341
342 fputc('\n', rpt);
343 fputs("-------------------\n", rpt);
344 fputs("END OF CRASH REPORT\n", rpt);
345 fclose(rpt);
346 free(loc);
347 }
348
349 static void end_packfile(void);
350 static void unkeep_all_packs(void);
351 static void dump_marks(void);
352
353 static NORETURN void die_nicely(const char *err, va_list params)
354 {
355 static int zombie;
356 char message[2 * PATH_MAX];
357
358 vsnprintf(message, sizeof(message), err, params);
359 fputs("fatal: ", stderr);
360 fputs(message, stderr);
361 fputc('\n', stderr);
362
363 if (!zombie) {
364 zombie = 1;
365 write_crash_report(message);
366 end_packfile();
367 unkeep_all_packs();
368 dump_marks();
369 }
370 exit(128);
371 }
372
373 #ifndef SIGUSR1 /* Windows, for example */
374
375 static void set_checkpoint_signal(void)
376 {
377 }
378
379 #else
380
381 static void checkpoint_signal(int signo)
382 {
383 checkpoint_requested = 1;
384 }
385
386 static void set_checkpoint_signal(void)
387 {
388 struct sigaction sa;
389
390 memset(&sa, 0, sizeof(sa));
391 sa.sa_handler = checkpoint_signal;
392 sigemptyset(&sa.sa_mask);
393 sa.sa_flags = SA_RESTART;
394 sigaction(SIGUSR1, &sa, NULL);
395 }
396
397 #endif
398
399 static void alloc_objects(unsigned int cnt)
400 {
401 struct object_entry_pool *b;
402
403 b = xmalloc(sizeof(struct object_entry_pool)
404 + cnt * sizeof(struct object_entry));
405 b->next_pool = blocks;
406 b->next_free = b->entries;
407 b->end = b->entries + cnt;
408 blocks = b;
409 alloc_count += cnt;
410 }
411
412 static struct object_entry *new_object(struct object_id *oid)
413 {
414 struct object_entry *e;
415
416 if (blocks->next_free == blocks->end)
417 alloc_objects(object_entry_alloc);
418
419 e = blocks->next_free++;
420 oidcpy(&e->idx.oid, oid);
421 return e;
422 }
423
424 static struct object_entry *find_object(struct object_id *oid)
425 {
426 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
427 struct object_entry *e;
428 for (e = object_table[h]; e; e = e->next)
429 if (oideq(oid, &e->idx.oid))
430 return e;
431 return NULL;
432 }
433
434 static struct object_entry *insert_object(struct object_id *oid)
435 {
436 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
437 struct object_entry *e = object_table[h];
438
439 while (e) {
440 if (oideq(oid, &e->idx.oid))
441 return e;
442 e = e->next;
443 }
444
445 e = new_object(oid);
446 e->next = object_table[h];
447 e->idx.offset = 0;
448 object_table[h] = e;
449 return e;
450 }
451
452 static void invalidate_pack_id(unsigned int id)
453 {
454 unsigned int h;
455 unsigned long lu;
456 struct tag *t;
457
458 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
459 struct object_entry *e;
460
461 for (e = object_table[h]; e; e = e->next)
462 if (e->pack_id == id)
463 e->pack_id = MAX_PACK_ID;
464 }
465
466 for (lu = 0; lu < branch_table_sz; lu++) {
467 struct branch *b;
468
469 for (b = branch_table[lu]; b; b = b->table_next_branch)
470 if (b->pack_id == id)
471 b->pack_id = MAX_PACK_ID;
472 }
473
474 for (t = first_tag; t; t = t->next_tag)
475 if (t->pack_id == id)
476 t->pack_id = MAX_PACK_ID;
477 }
478
479 static unsigned int hc_str(const char *s, size_t len)
480 {
481 unsigned int r = 0;
482 while (len-- > 0)
483 r = r * 31 + *s++;
484 return r;
485 }
486
487 static char *pool_strdup(const char *s)
488 {
489 size_t len = strlen(s) + 1;
490 char *r = mem_pool_alloc(&fi_mem_pool, len);
491 memcpy(r, s, len);
492 return r;
493 }
494
495 static void insert_mark(uintmax_t idnum, struct object_entry *oe)
496 {
497 struct mark_set *s = marks;
498 while ((idnum >> s->shift) >= 1024) {
499 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
500 s->shift = marks->shift + 10;
501 s->data.sets[0] = marks;
502 marks = s;
503 }
504 while (s->shift) {
505 uintmax_t i = idnum >> s->shift;
506 idnum -= i << s->shift;
507 if (!s->data.sets[i]) {
508 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
509 s->data.sets[i]->shift = s->shift - 10;
510 }
511 s = s->data.sets[i];
512 }
513 if (!s->data.marked[idnum])
514 marks_set_count++;
515 s->data.marked[idnum] = oe;
516 }
517
518 static struct object_entry *find_mark(uintmax_t idnum)
519 {
520 uintmax_t orig_idnum = idnum;
521 struct mark_set *s = marks;
522 struct object_entry *oe = NULL;
523 if ((idnum >> s->shift) < 1024) {
524 while (s && s->shift) {
525 uintmax_t i = idnum >> s->shift;
526 idnum -= i << s->shift;
527 s = s->data.sets[i];
528 }
529 if (s)
530 oe = s->data.marked[idnum];
531 }
532 if (!oe)
533 die("mark :%" PRIuMAX " not declared", orig_idnum);
534 return oe;
535 }
536
537 static struct atom_str *to_atom(const char *s, unsigned short len)
538 {
539 unsigned int hc = hc_str(s, len) % atom_table_sz;
540 struct atom_str *c;
541
542 for (c = atom_table[hc]; c; c = c->next_atom)
543 if (c->str_len == len && !strncmp(s, c->str_dat, len))
544 return c;
545
546 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
547 c->str_len = len;
548 memcpy(c->str_dat, s, len);
549 c->str_dat[len] = 0;
550 c->next_atom = atom_table[hc];
551 atom_table[hc] = c;
552 atom_cnt++;
553 return c;
554 }
555
556 static struct branch *lookup_branch(const char *name)
557 {
558 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
559 struct branch *b;
560
561 for (b = branch_table[hc]; b; b = b->table_next_branch)
562 if (!strcmp(name, b->name))
563 return b;
564 return NULL;
565 }
566
567 static struct branch *new_branch(const char *name)
568 {
569 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
570 struct branch *b = lookup_branch(name);
571
572 if (b)
573 die("Invalid attempt to create duplicate branch: %s", name);
574 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
575 die("Branch name doesn't conform to GIT standards: %s", name);
576
577 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
578 b->name = pool_strdup(name);
579 b->table_next_branch = branch_table[hc];
580 b->branch_tree.versions[0].mode = S_IFDIR;
581 b->branch_tree.versions[1].mode = S_IFDIR;
582 b->num_notes = 0;
583 b->active = 0;
584 b->pack_id = MAX_PACK_ID;
585 branch_table[hc] = b;
586 branch_count++;
587 return b;
588 }
589
590 static unsigned int hc_entries(unsigned int cnt)
591 {
592 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
593 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
594 }
595
596 static struct tree_content *new_tree_content(unsigned int cnt)
597 {
598 struct avail_tree_content *f, *l = NULL;
599 struct tree_content *t;
600 unsigned int hc = hc_entries(cnt);
601
602 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
603 if (f->entry_capacity >= cnt)
604 break;
605
606 if (f) {
607 if (l)
608 l->next_avail = f->next_avail;
609 else
610 avail_tree_table[hc] = f->next_avail;
611 } else {
612 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
613 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
614 f->entry_capacity = cnt;
615 }
616
617 t = (struct tree_content*)f;
618 t->entry_count = 0;
619 t->delta_depth = 0;
620 return t;
621 }
622
623 static void release_tree_entry(struct tree_entry *e);
624 static void release_tree_content(struct tree_content *t)
625 {
626 struct avail_tree_content *f = (struct avail_tree_content*)t;
627 unsigned int hc = hc_entries(f->entry_capacity);
628 f->next_avail = avail_tree_table[hc];
629 avail_tree_table[hc] = f;
630 }
631
632 static void release_tree_content_recursive(struct tree_content *t)
633 {
634 unsigned int i;
635 for (i = 0; i < t->entry_count; i++)
636 release_tree_entry(t->entries[i]);
637 release_tree_content(t);
638 }
639
640 static struct tree_content *grow_tree_content(
641 struct tree_content *t,
642 int amt)
643 {
644 struct tree_content *r = new_tree_content(t->entry_count + amt);
645 r->entry_count = t->entry_count;
646 r->delta_depth = t->delta_depth;
647 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
648 release_tree_content(t);
649 return r;
650 }
651
652 static struct tree_entry *new_tree_entry(void)
653 {
654 struct tree_entry *e;
655
656 if (!avail_tree_entry) {
657 unsigned int n = tree_entry_alloc;
658 tree_entry_allocd += n * sizeof(struct tree_entry);
659 ALLOC_ARRAY(e, n);
660 avail_tree_entry = e;
661 while (n-- > 1) {
662 *((void**)e) = e + 1;
663 e++;
664 }
665 *((void**)e) = NULL;
666 }
667
668 e = avail_tree_entry;
669 avail_tree_entry = *((void**)e);
670 return e;
671 }
672
673 static void release_tree_entry(struct tree_entry *e)
674 {
675 if (e->tree)
676 release_tree_content_recursive(e->tree);
677 *((void**)e) = avail_tree_entry;
678 avail_tree_entry = e;
679 }
680
681 static struct tree_content *dup_tree_content(struct tree_content *s)
682 {
683 struct tree_content *d;
684 struct tree_entry *a, *b;
685 unsigned int i;
686
687 if (!s)
688 return NULL;
689 d = new_tree_content(s->entry_count);
690 for (i = 0; i < s->entry_count; i++) {
691 a = s->entries[i];
692 b = new_tree_entry();
693 memcpy(b, a, sizeof(*a));
694 if (a->tree && is_null_oid(&b->versions[1].oid))
695 b->tree = dup_tree_content(a->tree);
696 else
697 b->tree = NULL;
698 d->entries[i] = b;
699 }
700 d->entry_count = s->entry_count;
701 d->delta_depth = s->delta_depth;
702
703 return d;
704 }
705
706 static void start_packfile(void)
707 {
708 struct strbuf tmp_file = STRBUF_INIT;
709 struct packed_git *p;
710 struct pack_header hdr;
711 int pack_fd;
712
713 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
714 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
715 strbuf_release(&tmp_file);
716
717 p->pack_fd = pack_fd;
718 p->do_not_close = 1;
719 pack_file = hashfd(pack_fd, p->pack_name);
720
721 hdr.hdr_signature = htonl(PACK_SIGNATURE);
722 hdr.hdr_version = htonl(2);
723 hdr.hdr_entries = 0;
724 hashwrite(pack_file, &hdr, sizeof(hdr));
725
726 pack_data = p;
727 pack_size = sizeof(hdr);
728 object_count = 0;
729
730 REALLOC_ARRAY(all_packs, pack_id + 1);
731 all_packs[pack_id] = p;
732 }
733
734 static const char *create_index(void)
735 {
736 const char *tmpfile;
737 struct pack_idx_entry **idx, **c, **last;
738 struct object_entry *e;
739 struct object_entry_pool *o;
740
741 /* Build the table of object IDs. */
742 ALLOC_ARRAY(idx, object_count);
743 c = idx;
744 for (o = blocks; o; o = o->next_pool)
745 for (e = o->next_free; e-- != o->entries;)
746 if (pack_id == e->pack_id)
747 *c++ = &e->idx;
748 last = idx + object_count;
749 if (c != last)
750 die("internal consistency error creating the index");
751
752 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts,
753 pack_data->hash);
754 free(idx);
755 return tmpfile;
756 }
757
758 static char *keep_pack(const char *curr_index_name)
759 {
760 static const char *keep_msg = "fast-import";
761 struct strbuf name = STRBUF_INIT;
762 int keep_fd;
763
764 odb_pack_name(&name, pack_data->hash, "keep");
765 keep_fd = odb_pack_keep(name.buf);
766 if (keep_fd < 0)
767 die_errno("cannot create keep file");
768 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
769 if (close(keep_fd))
770 die_errno("failed to write keep file");
771
772 odb_pack_name(&name, pack_data->hash, "pack");
773 if (finalize_object_file(pack_data->pack_name, name.buf))
774 die("cannot store pack file");
775
776 odb_pack_name(&name, pack_data->hash, "idx");
777 if (finalize_object_file(curr_index_name, name.buf))
778 die("cannot store index file");
779 free((void *)curr_index_name);
780 return strbuf_detach(&name, NULL);
781 }
782
783 static void unkeep_all_packs(void)
784 {
785 struct strbuf name = STRBUF_INIT;
786 int k;
787
788 for (k = 0; k < pack_id; k++) {
789 struct packed_git *p = all_packs[k];
790 odb_pack_name(&name, p->hash, "keep");
791 unlink_or_warn(name.buf);
792 }
793 strbuf_release(&name);
794 }
795
796 static int loosen_small_pack(const struct packed_git *p)
797 {
798 struct child_process unpack = CHILD_PROCESS_INIT;
799
800 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
801 die_errno("Failed seeking to start of '%s'", p->pack_name);
802
803 unpack.in = p->pack_fd;
804 unpack.git_cmd = 1;
805 unpack.stdout_to_stderr = 1;
806 argv_array_push(&unpack.args, "unpack-objects");
807 if (!show_stats)
808 argv_array_push(&unpack.args, "-q");
809
810 return run_command(&unpack);
811 }
812
813 static void end_packfile(void)
814 {
815 static int running;
816
817 if (running || !pack_data)
818 return;
819
820 running = 1;
821 clear_delta_base_cache();
822 if (object_count) {
823 struct packed_git *new_p;
824 struct object_id cur_pack_oid;
825 char *idx_name;
826 int i;
827 struct branch *b;
828 struct tag *t;
829
830 close_pack_windows(pack_data);
831 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
832 fixup_pack_header_footer(pack_data->pack_fd, pack_data->hash,
833 pack_data->pack_name, object_count,
834 cur_pack_oid.hash, pack_size);
835
836 if (object_count <= unpack_limit) {
837 if (!loosen_small_pack(pack_data)) {
838 invalidate_pack_id(pack_id);
839 goto discard_pack;
840 }
841 }
842
843 close(pack_data->pack_fd);
844 idx_name = keep_pack(create_index());
845
846 /* Register the packfile with core git's machinery. */
847 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
848 if (!new_p)
849 die("core git rejected index %s", idx_name);
850 all_packs[pack_id] = new_p;
851 install_packed_git(the_repository, new_p);
852 free(idx_name);
853
854 /* Print the boundary */
855 if (pack_edges) {
856 fprintf(pack_edges, "%s:", new_p->pack_name);
857 for (i = 0; i < branch_table_sz; i++) {
858 for (b = branch_table[i]; b; b = b->table_next_branch) {
859 if (b->pack_id == pack_id)
860 fprintf(pack_edges, " %s",
861 oid_to_hex(&b->oid));
862 }
863 }
864 for (t = first_tag; t; t = t->next_tag) {
865 if (t->pack_id == pack_id)
866 fprintf(pack_edges, " %s",
867 oid_to_hex(&t->oid));
868 }
869 fputc('\n', pack_edges);
870 fflush(pack_edges);
871 }
872
873 pack_id++;
874 }
875 else {
876 discard_pack:
877 close(pack_data->pack_fd);
878 unlink_or_warn(pack_data->pack_name);
879 }
880 FREE_AND_NULL(pack_data);
881 running = 0;
882
883 /* We can't carry a delta across packfiles. */
884 strbuf_release(&last_blob.data);
885 last_blob.offset = 0;
886 last_blob.depth = 0;
887 }
888
889 static void cycle_packfile(void)
890 {
891 end_packfile();
892 start_packfile();
893 }
894
895 static int store_object(
896 enum object_type type,
897 struct strbuf *dat,
898 struct last_object *last,
899 struct object_id *oidout,
900 uintmax_t mark)
901 {
902 void *out, *delta;
903 struct object_entry *e;
904 unsigned char hdr[96];
905 struct object_id oid;
906 unsigned long hdrlen, deltalen;
907 git_hash_ctx c;
908 git_zstream s;
909
910 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
911 type_name(type), (unsigned long)dat->len) + 1;
912 the_hash_algo->init_fn(&c);
913 the_hash_algo->update_fn(&c, hdr, hdrlen);
914 the_hash_algo->update_fn(&c, dat->buf, dat->len);
915 the_hash_algo->final_fn(oid.hash, &c);
916 if (oidout)
917 oidcpy(oidout, &oid);
918
919 e = insert_object(&oid);
920 if (mark)
921 insert_mark(mark, e);
922 if (e->idx.offset) {
923 duplicate_count_by_type[type]++;
924 return 1;
925 } else if (find_sha1_pack(oid.hash,
926 get_all_packs(the_repository))) {
927 e->type = type;
928 e->pack_id = MAX_PACK_ID;
929 e->idx.offset = 1; /* just not zero! */
930 duplicate_count_by_type[type]++;
931 return 1;
932 }
933
934 if (last && last->data.len && last->data.buf && last->depth < max_depth
935 && dat->len > the_hash_algo->rawsz) {
936
937 delta_count_attempts_by_type[type]++;
938 delta = diff_delta(last->data.buf, last->data.len,
939 dat->buf, dat->len,
940 &deltalen, dat->len - the_hash_algo->rawsz);
941 } else
942 delta = NULL;
943
944 git_deflate_init(&s, pack_compression_level);
945 if (delta) {
946 s.next_in = delta;
947 s.avail_in = deltalen;
948 } else {
949 s.next_in = (void *)dat->buf;
950 s.avail_in = dat->len;
951 }
952 s.avail_out = git_deflate_bound(&s, s.avail_in);
953 s.next_out = out = xmalloc(s.avail_out);
954 while (git_deflate(&s, Z_FINISH) == Z_OK)
955 ; /* nothing */
956 git_deflate_end(&s);
957
958 /* Determine if we should auto-checkpoint. */
959 if ((max_packsize
960 && (pack_size + PACK_SIZE_THRESHOLD + s.total_out) > max_packsize)
961 || (pack_size + PACK_SIZE_THRESHOLD + s.total_out) < pack_size) {
962
963 /* This new object needs to *not* have the current pack_id. */
964 e->pack_id = pack_id + 1;
965 cycle_packfile();
966
967 /* We cannot carry a delta into the new pack. */
968 if (delta) {
969 FREE_AND_NULL(delta);
970
971 git_deflate_init(&s, pack_compression_level);
972 s.next_in = (void *)dat->buf;
973 s.avail_in = dat->len;
974 s.avail_out = git_deflate_bound(&s, s.avail_in);
975 s.next_out = out = xrealloc(out, s.avail_out);
976 while (git_deflate(&s, Z_FINISH) == Z_OK)
977 ; /* nothing */
978 git_deflate_end(&s);
979 }
980 }
981
982 e->type = type;
983 e->pack_id = pack_id;
984 e->idx.offset = pack_size;
985 object_count++;
986 object_count_by_type[type]++;
987
988 crc32_begin(pack_file);
989
990 if (delta) {
991 off_t ofs = e->idx.offset - last->offset;
992 unsigned pos = sizeof(hdr) - 1;
993
994 delta_count_by_type[type]++;
995 e->depth = last->depth + 1;
996
997 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
998 OBJ_OFS_DELTA, deltalen);
999 hashwrite(pack_file, hdr, hdrlen);
1000 pack_size += hdrlen;
1001
1002 hdr[pos] = ofs & 127;
1003 while (ofs >>= 7)
1004 hdr[--pos] = 128 | (--ofs & 127);
1005 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
1006 pack_size += sizeof(hdr) - pos;
1007 } else {
1008 e->depth = 0;
1009 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1010 type, dat->len);
1011 hashwrite(pack_file, hdr, hdrlen);
1012 pack_size += hdrlen;
1013 }
1014
1015 hashwrite(pack_file, out, s.total_out);
1016 pack_size += s.total_out;
1017
1018 e->idx.crc32 = crc32_end(pack_file);
1019
1020 free(out);
1021 free(delta);
1022 if (last) {
1023 if (last->no_swap) {
1024 last->data = *dat;
1025 } else {
1026 strbuf_swap(&last->data, dat);
1027 }
1028 last->offset = e->idx.offset;
1029 last->depth = e->depth;
1030 }
1031 return 0;
1032 }
1033
1034 static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1035 {
1036 if (hashfile_truncate(pack_file, checkpoint))
1037 die_errno("cannot truncate pack to skip duplicate");
1038 pack_size = checkpoint->offset;
1039 }
1040
1041 static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1042 {
1043 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1044 unsigned char *in_buf = xmalloc(in_sz);
1045 unsigned char *out_buf = xmalloc(out_sz);
1046 struct object_entry *e;
1047 struct object_id oid;
1048 unsigned long hdrlen;
1049 off_t offset;
1050 git_hash_ctx c;
1051 git_zstream s;
1052 struct hashfile_checkpoint checkpoint;
1053 int status = Z_OK;
1054
1055 /* Determine if we should auto-checkpoint. */
1056 if ((max_packsize
1057 && (pack_size + PACK_SIZE_THRESHOLD + len) > max_packsize)
1058 || (pack_size + PACK_SIZE_THRESHOLD + len) < pack_size)
1059 cycle_packfile();
1060
1061 hashfile_checkpoint(pack_file, &checkpoint);
1062 offset = checkpoint.offset;
1063
1064 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1065
1066 the_hash_algo->init_fn(&c);
1067 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1068
1069 crc32_begin(pack_file);
1070
1071 git_deflate_init(&s, pack_compression_level);
1072
1073 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1074
1075 s.next_out = out_buf + hdrlen;
1076 s.avail_out = out_sz - hdrlen;
1077
1078 while (status != Z_STREAM_END) {
1079 if (0 < len && !s.avail_in) {
1080 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1081 size_t n = fread(in_buf, 1, cnt, stdin);
1082 if (!n && feof(stdin))
1083 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1084
1085 the_hash_algo->update_fn(&c, in_buf, n);
1086 s.next_in = in_buf;
1087 s.avail_in = n;
1088 len -= n;
1089 }
1090
1091 status = git_deflate(&s, len ? 0 : Z_FINISH);
1092
1093 if (!s.avail_out || status == Z_STREAM_END) {
1094 size_t n = s.next_out - out_buf;
1095 hashwrite(pack_file, out_buf, n);
1096 pack_size += n;
1097 s.next_out = out_buf;
1098 s.avail_out = out_sz;
1099 }
1100
1101 switch (status) {
1102 case Z_OK:
1103 case Z_BUF_ERROR:
1104 case Z_STREAM_END:
1105 continue;
1106 default:
1107 die("unexpected deflate failure: %d", status);
1108 }
1109 }
1110 git_deflate_end(&s);
1111 the_hash_algo->final_fn(oid.hash, &c);
1112
1113 if (oidout)
1114 oidcpy(oidout, &oid);
1115
1116 e = insert_object(&oid);
1117
1118 if (mark)
1119 insert_mark(mark, e);
1120
1121 if (e->idx.offset) {
1122 duplicate_count_by_type[OBJ_BLOB]++;
1123 truncate_pack(&checkpoint);
1124
1125 } else if (find_sha1_pack(oid.hash,
1126 get_all_packs(the_repository))) {
1127 e->type = OBJ_BLOB;
1128 e->pack_id = MAX_PACK_ID;
1129 e->idx.offset = 1; /* just not zero! */
1130 duplicate_count_by_type[OBJ_BLOB]++;
1131 truncate_pack(&checkpoint);
1132
1133 } else {
1134 e->depth = 0;
1135 e->type = OBJ_BLOB;
1136 e->pack_id = pack_id;
1137 e->idx.offset = offset;
1138 e->idx.crc32 = crc32_end(pack_file);
1139 object_count++;
1140 object_count_by_type[OBJ_BLOB]++;
1141 }
1142
1143 free(in_buf);
1144 free(out_buf);
1145 }
1146
1147 /* All calls must be guarded by find_object() or find_mark() to
1148 * ensure the 'struct object_entry' passed was written by this
1149 * process instance. We unpack the entry by the offset, avoiding
1150 * the need for the corresponding .idx file. This unpacking rule
1151 * works because we only use OBJ_REF_DELTA within the packfiles
1152 * created by fast-import.
1153 *
1154 * oe must not be NULL. Such an oe usually comes from giving
1155 * an unknown SHA-1 to find_object() or an undefined mark to
1156 * find_mark(). Callers must test for this condition and use
1157 * the standard read_sha1_file() when it happens.
1158 *
1159 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1160 * find_mark(), where the mark was reloaded from an existing marks
1161 * file and is referencing an object that this fast-import process
1162 * instance did not write out to a packfile. Callers must test for
1163 * this condition and use read_sha1_file() instead.
1164 */
1165 static void *gfi_unpack_entry(
1166 struct object_entry *oe,
1167 unsigned long *sizep)
1168 {
1169 enum object_type type;
1170 struct packed_git *p = all_packs[oe->pack_id];
1171 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1172 /* The object is stored in the packfile we are writing to
1173 * and we have modified it since the last time we scanned
1174 * back to read a previously written object. If an old
1175 * window covered [p->pack_size, p->pack_size + rawsz) its
1176 * data is stale and is not valid. Closing all windows
1177 * and updating the packfile length ensures we can read
1178 * the newly written data.
1179 */
1180 close_pack_windows(p);
1181 hashflush(pack_file);
1182
1183 /* We have to offer rawsz bytes additional on the end of
1184 * the packfile as the core unpacker code assumes the
1185 * footer is present at the file end and must promise
1186 * at least rawsz bytes within any window it maps. But
1187 * we don't actually create the footer here.
1188 */
1189 p->pack_size = pack_size + the_hash_algo->rawsz;
1190 }
1191 return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1192 }
1193
1194 static const char *get_mode(const char *str, uint16_t *modep)
1195 {
1196 unsigned char c;
1197 uint16_t mode = 0;
1198
1199 while ((c = *str++) != ' ') {
1200 if (c < '0' || c > '7')
1201 return NULL;
1202 mode = (mode << 3) + (c - '0');
1203 }
1204 *modep = mode;
1205 return str;
1206 }
1207
1208 static void load_tree(struct tree_entry *root)
1209 {
1210 struct object_id *oid = &root->versions[1].oid;
1211 struct object_entry *myoe;
1212 struct tree_content *t;
1213 unsigned long size;
1214 char *buf;
1215 const char *c;
1216
1217 root->tree = t = new_tree_content(8);
1218 if (is_null_oid(oid))
1219 return;
1220
1221 myoe = find_object(oid);
1222 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1223 if (myoe->type != OBJ_TREE)
1224 die("Not a tree: %s", oid_to_hex(oid));
1225 t->delta_depth = myoe->depth;
1226 buf = gfi_unpack_entry(myoe, &size);
1227 if (!buf)
1228 die("Can't load tree %s", oid_to_hex(oid));
1229 } else {
1230 enum object_type type;
1231 buf = read_object_file(oid, &type, &size);
1232 if (!buf || type != OBJ_TREE)
1233 die("Can't load tree %s", oid_to_hex(oid));
1234 }
1235
1236 c = buf;
1237 while (c != (buf + size)) {
1238 struct tree_entry *e = new_tree_entry();
1239
1240 if (t->entry_count == t->entry_capacity)
1241 root->tree = t = grow_tree_content(t, t->entry_count);
1242 t->entries[t->entry_count++] = e;
1243
1244 e->tree = NULL;
1245 c = get_mode(c, &e->versions[1].mode);
1246 if (!c)
1247 die("Corrupt mode in %s", oid_to_hex(oid));
1248 e->versions[0].mode = e->versions[1].mode;
1249 e->name = to_atom(c, strlen(c));
1250 c += e->name->str_len + 1;
1251 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1252 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1253 c += the_hash_algo->rawsz;
1254 }
1255 free(buf);
1256 }
1257
1258 static int tecmp0 (const void *_a, const void *_b)
1259 {
1260 struct tree_entry *a = *((struct tree_entry**)_a);
1261 struct tree_entry *b = *((struct tree_entry**)_b);
1262 return base_name_compare(
1263 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1264 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1265 }
1266
1267 static int tecmp1 (const void *_a, const void *_b)
1268 {
1269 struct tree_entry *a = *((struct tree_entry**)_a);
1270 struct tree_entry *b = *((struct tree_entry**)_b);
1271 return base_name_compare(
1272 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1273 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1274 }
1275
1276 static void mktree(struct tree_content *t, int v, struct strbuf *b)
1277 {
1278 size_t maxlen = 0;
1279 unsigned int i;
1280
1281 if (!v)
1282 QSORT(t->entries, t->entry_count, tecmp0);
1283 else
1284 QSORT(t->entries, t->entry_count, tecmp1);
1285
1286 for (i = 0; i < t->entry_count; i++) {
1287 if (t->entries[i]->versions[v].mode)
1288 maxlen += t->entries[i]->name->str_len + 34;
1289 }
1290
1291 strbuf_reset(b);
1292 strbuf_grow(b, maxlen);
1293 for (i = 0; i < t->entry_count; i++) {
1294 struct tree_entry *e = t->entries[i];
1295 if (!e->versions[v].mode)
1296 continue;
1297 strbuf_addf(b, "%o %s%c",
1298 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1299 e->name->str_dat, '\0');
1300 strbuf_add(b, e->versions[v].oid.hash, the_hash_algo->rawsz);
1301 }
1302 }
1303
1304 static void store_tree(struct tree_entry *root)
1305 {
1306 struct tree_content *t;
1307 unsigned int i, j, del;
1308 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1309 struct object_entry *le = NULL;
1310
1311 if (!is_null_oid(&root->versions[1].oid))
1312 return;
1313
1314 if (!root->tree)
1315 load_tree(root);
1316 t = root->tree;
1317
1318 for (i = 0; i < t->entry_count; i++) {
1319 if (t->entries[i]->tree)
1320 store_tree(t->entries[i]);
1321 }
1322
1323 if (!(root->versions[0].mode & NO_DELTA))
1324 le = find_object(&root->versions[0].oid);
1325 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1326 mktree(t, 0, &old_tree);
1327 lo.data = old_tree;
1328 lo.offset = le->idx.offset;
1329 lo.depth = t->delta_depth;
1330 }
1331
1332 mktree(t, 1, &new_tree);
1333 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1334
1335 t->delta_depth = lo.depth;
1336 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1337 struct tree_entry *e = t->entries[i];
1338 if (e->versions[1].mode) {
1339 e->versions[0].mode = e->versions[1].mode;
1340 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1341 t->entries[j++] = e;
1342 } else {
1343 release_tree_entry(e);
1344 del++;
1345 }
1346 }
1347 t->entry_count -= del;
1348 }
1349
1350 static void tree_content_replace(
1351 struct tree_entry *root,
1352 const struct object_id *oid,
1353 const uint16_t mode,
1354 struct tree_content *newtree)
1355 {
1356 if (!S_ISDIR(mode))
1357 die("Root cannot be a non-directory");
1358 oidclr(&root->versions[0].oid);
1359 oidcpy(&root->versions[1].oid, oid);
1360 if (root->tree)
1361 release_tree_content_recursive(root->tree);
1362 root->tree = newtree;
1363 }
1364
1365 static int tree_content_set(
1366 struct tree_entry *root,
1367 const char *p,
1368 const struct object_id *oid,
1369 const uint16_t mode,
1370 struct tree_content *subtree)
1371 {
1372 struct tree_content *t;
1373 const char *slash1;
1374 unsigned int i, n;
1375 struct tree_entry *e;
1376
1377 slash1 = strchrnul(p, '/');
1378 n = slash1 - p;
1379 if (!n)
1380 die("Empty path component found in input");
1381 if (!*slash1 && !S_ISDIR(mode) && subtree)
1382 die("Non-directories cannot have subtrees");
1383
1384 if (!root->tree)
1385 load_tree(root);
1386 t = root->tree;
1387 for (i = 0; i < t->entry_count; i++) {
1388 e = t->entries[i];
1389 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1390 if (!*slash1) {
1391 if (!S_ISDIR(mode)
1392 && e->versions[1].mode == mode
1393 && oideq(&e->versions[1].oid, oid))
1394 return 0;
1395 e->versions[1].mode = mode;
1396 oidcpy(&e->versions[1].oid, oid);
1397 if (e->tree)
1398 release_tree_content_recursive(e->tree);
1399 e->tree = subtree;
1400
1401 /*
1402 * We need to leave e->versions[0].sha1 alone
1403 * to avoid modifying the preimage tree used
1404 * when writing out the parent directory.
1405 * But after replacing the subdir with a
1406 * completely different one, it's not a good
1407 * delta base any more, and besides, we've
1408 * thrown away the tree entries needed to
1409 * make a delta against it.
1410 *
1411 * So let's just explicitly disable deltas
1412 * for the subtree.
1413 */
1414 if (S_ISDIR(e->versions[0].mode))
1415 e->versions[0].mode |= NO_DELTA;
1416
1417 oidclr(&root->versions[1].oid);
1418 return 1;
1419 }
1420 if (!S_ISDIR(e->versions[1].mode)) {
1421 e->tree = new_tree_content(8);
1422 e->versions[1].mode = S_IFDIR;
1423 }
1424 if (!e->tree)
1425 load_tree(e);
1426 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1427 oidclr(&root->versions[1].oid);
1428 return 1;
1429 }
1430 return 0;
1431 }
1432 }
1433
1434 if (t->entry_count == t->entry_capacity)
1435 root->tree = t = grow_tree_content(t, t->entry_count);
1436 e = new_tree_entry();
1437 e->name = to_atom(p, n);
1438 e->versions[0].mode = 0;
1439 oidclr(&e->versions[0].oid);
1440 t->entries[t->entry_count++] = e;
1441 if (*slash1) {
1442 e->tree = new_tree_content(8);
1443 e->versions[1].mode = S_IFDIR;
1444 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1445 } else {
1446 e->tree = subtree;
1447 e->versions[1].mode = mode;
1448 oidcpy(&e->versions[1].oid, oid);
1449 }
1450 oidclr(&root->versions[1].oid);
1451 return 1;
1452 }
1453
1454 static int tree_content_remove(
1455 struct tree_entry *root,
1456 const char *p,
1457 struct tree_entry *backup_leaf,
1458 int allow_root)
1459 {
1460 struct tree_content *t;
1461 const char *slash1;
1462 unsigned int i, n;
1463 struct tree_entry *e;
1464
1465 slash1 = strchrnul(p, '/');
1466 n = slash1 - p;
1467
1468 if (!root->tree)
1469 load_tree(root);
1470
1471 if (!*p && allow_root) {
1472 e = root;
1473 goto del_entry;
1474 }
1475
1476 t = root->tree;
1477 for (i = 0; i < t->entry_count; i++) {
1478 e = t->entries[i];
1479 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1480 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1481 /*
1482 * If p names a file in some subdirectory, and a
1483 * file or symlink matching the name of the
1484 * parent directory of p exists, then p cannot
1485 * exist and need not be deleted.
1486 */
1487 return 1;
1488 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1489 goto del_entry;
1490 if (!e->tree)
1491 load_tree(e);
1492 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1493 for (n = 0; n < e->tree->entry_count; n++) {
1494 if (e->tree->entries[n]->versions[1].mode) {
1495 oidclr(&root->versions[1].oid);
1496 return 1;
1497 }
1498 }
1499 backup_leaf = NULL;
1500 goto del_entry;
1501 }
1502 return 0;
1503 }
1504 }
1505 return 0;
1506
1507 del_entry:
1508 if (backup_leaf)
1509 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1510 else if (e->tree)
1511 release_tree_content_recursive(e->tree);
1512 e->tree = NULL;
1513 e->versions[1].mode = 0;
1514 oidclr(&e->versions[1].oid);
1515 oidclr(&root->versions[1].oid);
1516 return 1;
1517 }
1518
1519 static int tree_content_get(
1520 struct tree_entry *root,
1521 const char *p,
1522 struct tree_entry *leaf,
1523 int allow_root)
1524 {
1525 struct tree_content *t;
1526 const char *slash1;
1527 unsigned int i, n;
1528 struct tree_entry *e;
1529
1530 slash1 = strchrnul(p, '/');
1531 n = slash1 - p;
1532 if (!n && !allow_root)
1533 die("Empty path component found in input");
1534
1535 if (!root->tree)
1536 load_tree(root);
1537
1538 if (!n) {
1539 e = root;
1540 goto found_entry;
1541 }
1542
1543 t = root->tree;
1544 for (i = 0; i < t->entry_count; i++) {
1545 e = t->entries[i];
1546 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1547 if (!*slash1)
1548 goto found_entry;
1549 if (!S_ISDIR(e->versions[1].mode))
1550 return 0;
1551 if (!e->tree)
1552 load_tree(e);
1553 return tree_content_get(e, slash1 + 1, leaf, 0);
1554 }
1555 }
1556 return 0;
1557
1558 found_entry:
1559 memcpy(leaf, e, sizeof(*leaf));
1560 if (e->tree && is_null_oid(&e->versions[1].oid))
1561 leaf->tree = dup_tree_content(e->tree);
1562 else
1563 leaf->tree = NULL;
1564 return 1;
1565 }
1566
1567 static int update_branch(struct branch *b)
1568 {
1569 static const char *msg = "fast-import";
1570 struct ref_transaction *transaction;
1571 struct object_id old_oid;
1572 struct strbuf err = STRBUF_INIT;
1573
1574 if (is_null_oid(&b->oid)) {
1575 if (b->delete)
1576 delete_ref(NULL, b->name, NULL, 0);
1577 return 0;
1578 }
1579 if (read_ref(b->name, &old_oid))
1580 oidclr(&old_oid);
1581 if (!force_update && !is_null_oid(&old_oid)) {
1582 struct commit *old_cmit, *new_cmit;
1583
1584 old_cmit = lookup_commit_reference_gently(the_repository,
1585 &old_oid, 0);
1586 new_cmit = lookup_commit_reference_gently(the_repository,
1587 &b->oid, 0);
1588 if (!old_cmit || !new_cmit)
1589 return error("Branch %s is missing commits.", b->name);
1590
1591 if (!in_merge_bases(old_cmit, new_cmit)) {
1592 warning("Not updating %s"
1593 " (new tip %s does not contain %s)",
1594 b->name, oid_to_hex(&b->oid),
1595 oid_to_hex(&old_oid));
1596 return -1;
1597 }
1598 }
1599 transaction = ref_transaction_begin(&err);
1600 if (!transaction ||
1601 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1602 0, msg, &err) ||
1603 ref_transaction_commit(transaction, &err)) {
1604 ref_transaction_free(transaction);
1605 error("%s", err.buf);
1606 strbuf_release(&err);
1607 return -1;
1608 }
1609 ref_transaction_free(transaction);
1610 strbuf_release(&err);
1611 return 0;
1612 }
1613
1614 static void dump_branches(void)
1615 {
1616 unsigned int i;
1617 struct branch *b;
1618
1619 for (i = 0; i < branch_table_sz; i++) {
1620 for (b = branch_table[i]; b; b = b->table_next_branch)
1621 failure |= update_branch(b);
1622 }
1623 }
1624
1625 static void dump_tags(void)
1626 {
1627 static const char *msg = "fast-import";
1628 struct tag *t;
1629 struct strbuf ref_name = STRBUF_INIT;
1630 struct strbuf err = STRBUF_INIT;
1631 struct ref_transaction *transaction;
1632
1633 transaction = ref_transaction_begin(&err);
1634 if (!transaction) {
1635 failure |= error("%s", err.buf);
1636 goto cleanup;
1637 }
1638 for (t = first_tag; t; t = t->next_tag) {
1639 strbuf_reset(&ref_name);
1640 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1641
1642 if (ref_transaction_update(transaction, ref_name.buf,
1643 &t->oid, NULL, 0, msg, &err)) {
1644 failure |= error("%s", err.buf);
1645 goto cleanup;
1646 }
1647 }
1648 if (ref_transaction_commit(transaction, &err))
1649 failure |= error("%s", err.buf);
1650
1651 cleanup:
1652 ref_transaction_free(transaction);
1653 strbuf_release(&ref_name);
1654 strbuf_release(&err);
1655 }
1656
1657 static void dump_marks_helper(FILE *f,
1658 uintmax_t base,
1659 struct mark_set *m)
1660 {
1661 uintmax_t k;
1662 if (m->shift) {
1663 for (k = 0; k < 1024; k++) {
1664 if (m->data.sets[k])
1665 dump_marks_helper(f, base + (k << m->shift),
1666 m->data.sets[k]);
1667 }
1668 } else {
1669 for (k = 0; k < 1024; k++) {
1670 if (m->data.marked[k])
1671 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1672 oid_to_hex(&m->data.marked[k]->idx.oid));
1673 }
1674 }
1675 }
1676
1677 static void dump_marks(void)
1678 {
1679 struct lock_file mark_lock = LOCK_INIT;
1680 FILE *f;
1681
1682 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1683 return;
1684
1685 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1686 failure |= error_errno("Unable to write marks file %s",
1687 export_marks_file);
1688 return;
1689 }
1690
1691 f = fdopen_lock_file(&mark_lock, "w");
1692 if (!f) {
1693 int saved_errno = errno;
1694 rollback_lock_file(&mark_lock);
1695 failure |= error("Unable to write marks file %s: %s",
1696 export_marks_file, strerror(saved_errno));
1697 return;
1698 }
1699
1700 dump_marks_helper(f, 0, marks);
1701 if (commit_lock_file(&mark_lock)) {
1702 failure |= error_errno("Unable to write file %s",
1703 export_marks_file);
1704 return;
1705 }
1706 }
1707
1708 static void read_marks(void)
1709 {
1710 char line[512];
1711 FILE *f = fopen(import_marks_file, "r");
1712 if (f)
1713 ;
1714 else if (import_marks_file_ignore_missing && errno == ENOENT)
1715 goto done; /* Marks file does not exist */
1716 else
1717 die_errno("cannot read '%s'", import_marks_file);
1718 while (fgets(line, sizeof(line), f)) {
1719 uintmax_t mark;
1720 char *end;
1721 struct object_id oid;
1722 struct object_entry *e;
1723
1724 end = strchr(line, '\n');
1725 if (line[0] != ':' || !end)
1726 die("corrupt mark line: %s", line);
1727 *end = 0;
1728 mark = strtoumax(line + 1, &end, 10);
1729 if (!mark || end == line + 1
1730 || *end != ' ' || get_oid_hex(end + 1, &oid))
1731 die("corrupt mark line: %s", line);
1732 e = find_object(&oid);
1733 if (!e) {
1734 enum object_type type = oid_object_info(the_repository,
1735 &oid, NULL);
1736 if (type < 0)
1737 die("object not found: %s", oid_to_hex(&oid));
1738 e = insert_object(&oid);
1739 e->type = type;
1740 e->pack_id = MAX_PACK_ID;
1741 e->idx.offset = 1; /* just not zero! */
1742 }
1743 insert_mark(mark, e);
1744 }
1745 fclose(f);
1746 done:
1747 import_marks_file_done = 1;
1748 }
1749
1750
1751 static int read_next_command(void)
1752 {
1753 static int stdin_eof = 0;
1754
1755 if (stdin_eof) {
1756 unread_command_buf = 0;
1757 return EOF;
1758 }
1759
1760 for (;;) {
1761 if (unread_command_buf) {
1762 unread_command_buf = 0;
1763 } else {
1764 struct recent_command *rc;
1765
1766 strbuf_detach(&command_buf, NULL);
1767 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1768 if (stdin_eof)
1769 return EOF;
1770
1771 if (!seen_data_command
1772 && !starts_with(command_buf.buf, "feature ")
1773 && !starts_with(command_buf.buf, "option ")) {
1774 parse_argv();
1775 }
1776
1777 rc = rc_free;
1778 if (rc)
1779 rc_free = rc->next;
1780 else {
1781 rc = cmd_hist.next;
1782 cmd_hist.next = rc->next;
1783 cmd_hist.next->prev = &cmd_hist;
1784 free(rc->buf);
1785 }
1786
1787 rc->buf = command_buf.buf;
1788 rc->prev = cmd_tail;
1789 rc->next = cmd_hist.prev;
1790 rc->prev->next = rc;
1791 cmd_tail = rc;
1792 }
1793 if (command_buf.buf[0] == '#')
1794 continue;
1795 return 0;
1796 }
1797 }
1798
1799 static void skip_optional_lf(void)
1800 {
1801 int term_char = fgetc(stdin);
1802 if (term_char != '\n' && term_char != EOF)
1803 ungetc(term_char, stdin);
1804 }
1805
1806 static void parse_mark(void)
1807 {
1808 const char *v;
1809 if (skip_prefix(command_buf.buf, "mark :", &v)) {
1810 next_mark = strtoumax(v, NULL, 10);
1811 read_next_command();
1812 }
1813 else
1814 next_mark = 0;
1815 }
1816
1817 static void parse_original_identifier(void)
1818 {
1819 const char *v;
1820 if (skip_prefix(command_buf.buf, "original-oid ", &v))
1821 read_next_command();
1822 }
1823
1824 static int parse_data(struct strbuf *sb, uintmax_t limit, uintmax_t *len_res)
1825 {
1826 const char *data;
1827 strbuf_reset(sb);
1828
1829 if (!skip_prefix(command_buf.buf, "data ", &data))
1830 die("Expected 'data n' command, found: %s", command_buf.buf);
1831
1832 if (skip_prefix(data, "<<", &data)) {
1833 char *term = xstrdup(data);
1834 size_t term_len = command_buf.len - (data - command_buf.buf);
1835
1836 strbuf_detach(&command_buf, NULL);
1837 for (;;) {
1838 if (strbuf_getline_lf(&command_buf, stdin) == EOF)
1839 die("EOF in data (terminator '%s' not found)", term);
1840 if (term_len == command_buf.len
1841 && !strcmp(term, command_buf.buf))
1842 break;
1843 strbuf_addbuf(sb, &command_buf);
1844 strbuf_addch(sb, '\n');
1845 }
1846 free(term);
1847 }
1848 else {
1849 uintmax_t len = strtoumax(data, NULL, 10);
1850 size_t n = 0, length = (size_t)len;
1851
1852 if (limit && limit < len) {
1853 *len_res = len;
1854 return 0;
1855 }
1856 if (length < len)
1857 die("data is too large to use in this context");
1858
1859 while (n < length) {
1860 size_t s = strbuf_fread(sb, length - n, stdin);
1861 if (!s && feof(stdin))
1862 die("EOF in data (%lu bytes remaining)",
1863 (unsigned long)(length - n));
1864 n += s;
1865 }
1866 }
1867
1868 skip_optional_lf();
1869 return 1;
1870 }
1871
1872 static int validate_raw_date(const char *src, struct strbuf *result)
1873 {
1874 const char *orig_src = src;
1875 char *endp;
1876 unsigned long num;
1877
1878 errno = 0;
1879
1880 num = strtoul(src, &endp, 10);
1881 /* NEEDSWORK: perhaps check for reasonable values? */
1882 if (errno || endp == src || *endp != ' ')
1883 return -1;
1884
1885 src = endp + 1;
1886 if (*src != '-' && *src != '+')
1887 return -1;
1888
1889 num = strtoul(src + 1, &endp, 10);
1890 if (errno || endp == src + 1 || *endp || 1400 < num)
1891 return -1;
1892
1893 strbuf_addstr(result, orig_src);
1894 return 0;
1895 }
1896
1897 static char *parse_ident(const char *buf)
1898 {
1899 const char *ltgt;
1900 size_t name_len;
1901 struct strbuf ident = STRBUF_INIT;
1902
1903 /* ensure there is a space delimiter even if there is no name */
1904 if (*buf == '<')
1905 --buf;
1906
1907 ltgt = buf + strcspn(buf, "<>");
1908 if (*ltgt != '<')
1909 die("Missing < in ident string: %s", buf);
1910 if (ltgt != buf && ltgt[-1] != ' ')
1911 die("Missing space before < in ident string: %s", buf);
1912 ltgt = ltgt + 1 + strcspn(ltgt + 1, "<>");
1913 if (*ltgt != '>')
1914 die("Missing > in ident string: %s", buf);
1915 ltgt++;
1916 if (*ltgt != ' ')
1917 die("Missing space after > in ident string: %s", buf);
1918 ltgt++;
1919 name_len = ltgt - buf;
1920 strbuf_add(&ident, buf, name_len);
1921
1922 switch (whenspec) {
1923 case WHENSPEC_RAW:
1924 if (validate_raw_date(ltgt, &ident) < 0)
1925 die("Invalid raw date \"%s\" in ident: %s", ltgt, buf);
1926 break;
1927 case WHENSPEC_RFC2822:
1928 if (parse_date(ltgt, &ident) < 0)
1929 die("Invalid rfc2822 date \"%s\" in ident: %s", ltgt, buf);
1930 break;
1931 case WHENSPEC_NOW:
1932 if (strcmp("now", ltgt))
1933 die("Date in ident must be 'now': %s", buf);
1934 datestamp(&ident);
1935 break;
1936 }
1937
1938 return strbuf_detach(&ident, NULL);
1939 }
1940
1941 static void parse_and_store_blob(
1942 struct last_object *last,
1943 struct object_id *oidout,
1944 uintmax_t mark)
1945 {
1946 static struct strbuf buf = STRBUF_INIT;
1947 uintmax_t len;
1948
1949 if (parse_data(&buf, big_file_threshold, &len))
1950 store_object(OBJ_BLOB, &buf, last, oidout, mark);
1951 else {
1952 if (last) {
1953 strbuf_release(&last->data);
1954 last->offset = 0;
1955 last->depth = 0;
1956 }
1957 stream_blob(len, oidout, mark);
1958 skip_optional_lf();
1959 }
1960 }
1961
1962 static void parse_new_blob(void)
1963 {
1964 read_next_command();
1965 parse_mark();
1966 parse_original_identifier();
1967 parse_and_store_blob(&last_blob, NULL, next_mark);
1968 }
1969
1970 static void unload_one_branch(void)
1971 {
1972 while (cur_active_branches
1973 && cur_active_branches >= max_active_branches) {
1974 uintmax_t min_commit = ULONG_MAX;
1975 struct branch *e, *l = NULL, *p = NULL;
1976
1977 for (e = active_branches; e; e = e->active_next_branch) {
1978 if (e->last_commit < min_commit) {
1979 p = l;
1980 min_commit = e->last_commit;
1981 }
1982 l = e;
1983 }
1984
1985 if (p) {
1986 e = p->active_next_branch;
1987 p->active_next_branch = e->active_next_branch;
1988 } else {
1989 e = active_branches;
1990 active_branches = e->active_next_branch;
1991 }
1992 e->active = 0;
1993 e->active_next_branch = NULL;
1994 if (e->branch_tree.tree) {
1995 release_tree_content_recursive(e->branch_tree.tree);
1996 e->branch_tree.tree = NULL;
1997 }
1998 cur_active_branches--;
1999 }
2000 }
2001
2002 static void load_branch(struct branch *b)
2003 {
2004 load_tree(&b->branch_tree);
2005 if (!b->active) {
2006 b->active = 1;
2007 b->active_next_branch = active_branches;
2008 active_branches = b;
2009 cur_active_branches++;
2010 branch_load_count++;
2011 }
2012 }
2013
2014 static unsigned char convert_num_notes_to_fanout(uintmax_t num_notes)
2015 {
2016 unsigned char fanout = 0;
2017 while ((num_notes >>= 8))
2018 fanout++;
2019 return fanout;
2020 }
2021
2022 static void construct_path_with_fanout(const char *hex_sha1,
2023 unsigned char fanout, char *path)
2024 {
2025 unsigned int i = 0, j = 0;
2026 if (fanout >= the_hash_algo->rawsz)
2027 die("Too large fanout (%u)", fanout);
2028 while (fanout) {
2029 path[i++] = hex_sha1[j++];
2030 path[i++] = hex_sha1[j++];
2031 path[i++] = '/';
2032 fanout--;
2033 }
2034 memcpy(path + i, hex_sha1 + j, the_hash_algo->hexsz - j);
2035 path[i + the_hash_algo->hexsz - j] = '\0';
2036 }
2037
2038 static uintmax_t do_change_note_fanout(
2039 struct tree_entry *orig_root, struct tree_entry *root,
2040 char *hex_oid, unsigned int hex_oid_len,
2041 char *fullpath, unsigned int fullpath_len,
2042 unsigned char fanout)
2043 {
2044 struct tree_content *t;
2045 struct tree_entry *e, leaf;
2046 unsigned int i, tmp_hex_oid_len, tmp_fullpath_len;
2047 uintmax_t num_notes = 0;
2048 struct object_id oid;
2049 /* hex oid + '/' between each pair of hex digits + NUL */
2050 char realpath[GIT_MAX_HEXSZ + ((GIT_MAX_HEXSZ / 2) - 1) + 1];
2051 const unsigned hexsz = the_hash_algo->hexsz;
2052
2053 if (!root->tree)
2054 load_tree(root);
2055 t = root->tree;
2056
2057 for (i = 0; t && i < t->entry_count; i++) {
2058 e = t->entries[i];
2059 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2060 tmp_fullpath_len = fullpath_len;
2061
2062 /*
2063 * We're interested in EITHER existing note entries (entries
2064 * with exactly 40 hex chars in path, not including directory
2065 * separators), OR directory entries that may contain note
2066 * entries (with < 40 hex chars in path).
2067 * Also, each path component in a note entry must be a multiple
2068 * of 2 chars.
2069 */
2070 if (!e->versions[1].mode ||
2071 tmp_hex_oid_len > hexsz ||
2072 e->name->str_len % 2)
2073 continue;
2074
2075 /* This _may_ be a note entry, or a subdir containing notes */
2076 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2077 e->name->str_len);
2078 if (tmp_fullpath_len)
2079 fullpath[tmp_fullpath_len++] = '/';
2080 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2081 e->name->str_len);
2082 tmp_fullpath_len += e->name->str_len;
2083 fullpath[tmp_fullpath_len] = '\0';
2084
2085 if (tmp_hex_oid_len == hexsz && !get_oid_hex(hex_oid, &oid)) {
2086 /* This is a note entry */
2087 if (fanout == 0xff) {
2088 /* Counting mode, no rename */
2089 num_notes++;
2090 continue;
2091 }
2092 construct_path_with_fanout(hex_oid, fanout, realpath);
2093 if (!strcmp(fullpath, realpath)) {
2094 /* Note entry is in correct location */
2095 num_notes++;
2096 continue;
2097 }
2098
2099 /* Rename fullpath to realpath */
2100 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2101 die("Failed to remove path %s", fullpath);
2102 tree_content_set(orig_root, realpath,
2103 &leaf.versions[1].oid,
2104 leaf.versions[1].mode,
2105 leaf.tree);
2106 } else if (S_ISDIR(e->versions[1].mode)) {
2107 /* This is a subdir that may contain note entries */
2108 num_notes += do_change_note_fanout(orig_root, e,
2109 hex_oid, tmp_hex_oid_len,
2110 fullpath, tmp_fullpath_len, fanout);
2111 }
2112
2113 /* The above may have reallocated the current tree_content */
2114 t = root->tree;
2115 }
2116 return num_notes;
2117 }
2118
2119 static uintmax_t change_note_fanout(struct tree_entry *root,
2120 unsigned char fanout)
2121 {
2122 /*
2123 * The size of path is due to one slash between every two hex digits,
2124 * plus the terminating NUL. Note that there is no slash at the end, so
2125 * the number of slashes is one less than half the number of hex
2126 * characters.
2127 */
2128 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2129 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2130 }
2131
2132 /*
2133 * Given a pointer into a string, parse a mark reference:
2134 *
2135 * idnum ::= ':' bigint;
2136 *
2137 * Return the first character after the value in *endptr.
2138 *
2139 * Complain if the following character is not what is expected,
2140 * either a space or end of the string.
2141 */
2142 static uintmax_t parse_mark_ref(const char *p, char **endptr)
2143 {
2144 uintmax_t mark;
2145
2146 assert(*p == ':');
2147 p++;
2148 mark = strtoumax(p, endptr, 10);
2149 if (*endptr == p)
2150 die("No value after ':' in mark: %s", command_buf.buf);
2151 return mark;
2152 }
2153
2154 /*
2155 * Parse the mark reference, and complain if this is not the end of
2156 * the string.
2157 */
2158 static uintmax_t parse_mark_ref_eol(const char *p)
2159 {
2160 char *end;
2161 uintmax_t mark;
2162
2163 mark = parse_mark_ref(p, &end);
2164 if (*end != '\0')
2165 die("Garbage after mark: %s", command_buf.buf);
2166 return mark;
2167 }
2168
2169 /*
2170 * Parse the mark reference, demanding a trailing space. Return a
2171 * pointer to the space.
2172 */
2173 static uintmax_t parse_mark_ref_space(const char **p)
2174 {
2175 uintmax_t mark;
2176 char *end;
2177
2178 mark = parse_mark_ref(*p, &end);
2179 if (*end++ != ' ')
2180 die("Missing space after mark: %s", command_buf.buf);
2181 *p = end;
2182 return mark;
2183 }
2184
2185 static void file_change_m(const char *p, struct branch *b)
2186 {
2187 static struct strbuf uq = STRBUF_INIT;
2188 const char *endp;
2189 struct object_entry *oe;
2190 struct object_id oid;
2191 uint16_t mode, inline_data = 0;
2192
2193 p = get_mode(p, &mode);
2194 if (!p)
2195 die("Corrupt mode: %s", command_buf.buf);
2196 switch (mode) {
2197 case 0644:
2198 case 0755:
2199 mode |= S_IFREG;
2200 case S_IFREG | 0644:
2201 case S_IFREG | 0755:
2202 case S_IFLNK:
2203 case S_IFDIR:
2204 case S_IFGITLINK:
2205 /* ok */
2206 break;
2207 default:
2208 die("Corrupt mode: %s", command_buf.buf);
2209 }
2210
2211 if (*p == ':') {
2212 oe = find_mark(parse_mark_ref_space(&p));
2213 oidcpy(&oid, &oe->idx.oid);
2214 } else if (skip_prefix(p, "inline ", &p)) {
2215 inline_data = 1;
2216 oe = NULL; /* not used with inline_data, but makes gcc happy */
2217 } else {
2218 if (parse_oid_hex(p, &oid, &p))
2219 die("Invalid dataref: %s", command_buf.buf);
2220 oe = find_object(&oid);
2221 if (*p++ != ' ')
2222 die("Missing space after SHA1: %s", command_buf.buf);
2223 }
2224
2225 strbuf_reset(&uq);
2226 if (!unquote_c_style(&uq, p, &endp)) {
2227 if (*endp)
2228 die("Garbage after path in: %s", command_buf.buf);
2229 p = uq.buf;
2230 }
2231
2232 /* Git does not track empty, non-toplevel directories. */
2233 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2234 tree_content_remove(&b->branch_tree, p, NULL, 0);
2235 return;
2236 }
2237
2238 if (S_ISGITLINK(mode)) {
2239 if (inline_data)
2240 die("Git links cannot be specified 'inline': %s",
2241 command_buf.buf);
2242 else if (oe) {
2243 if (oe->type != OBJ_COMMIT)
2244 die("Not a commit (actually a %s): %s",
2245 type_name(oe->type), command_buf.buf);
2246 }
2247 /*
2248 * Accept the sha1 without checking; it expected to be in
2249 * another repository.
2250 */
2251 } else if (inline_data) {
2252 if (S_ISDIR(mode))
2253 die("Directories cannot be specified 'inline': %s",
2254 command_buf.buf);
2255 if (p != uq.buf) {
2256 strbuf_addstr(&uq, p);
2257 p = uq.buf;
2258 }
2259 while (read_next_command() != EOF) {
2260 const char *v;
2261 if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2262 parse_cat_blob(v);
2263 else {
2264 parse_and_store_blob(&last_blob, &oid, 0);
2265 break;
2266 }
2267 }
2268 } else {
2269 enum object_type expected = S_ISDIR(mode) ?
2270 OBJ_TREE: OBJ_BLOB;
2271 enum object_type type = oe ? oe->type :
2272 oid_object_info(the_repository, &oid,
2273 NULL);
2274 if (type < 0)
2275 die("%s not found: %s",
2276 S_ISDIR(mode) ? "Tree" : "Blob",
2277 command_buf.buf);
2278 if (type != expected)
2279 die("Not a %s (actually a %s): %s",
2280 type_name(expected), type_name(type),
2281 command_buf.buf);
2282 }
2283
2284 if (!*p) {
2285 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2286 return;
2287 }
2288 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2289 }
2290
2291 static void file_change_d(const char *p, struct branch *b)
2292 {
2293 static struct strbuf uq = STRBUF_INIT;
2294 const char *endp;
2295
2296 strbuf_reset(&uq);
2297 if (!unquote_c_style(&uq, p, &endp)) {
2298 if (*endp)
2299 die("Garbage after path in: %s", command_buf.buf);
2300 p = uq.buf;
2301 }
2302 tree_content_remove(&b->branch_tree, p, NULL, 1);
2303 }
2304
2305 static void file_change_cr(const char *s, struct branch *b, int rename)
2306 {
2307 const char *d;
2308 static struct strbuf s_uq = STRBUF_INIT;
2309 static struct strbuf d_uq = STRBUF_INIT;
2310 const char *endp;
2311 struct tree_entry leaf;
2312
2313 strbuf_reset(&s_uq);
2314 if (!unquote_c_style(&s_uq, s, &endp)) {
2315 if (*endp != ' ')
2316 die("Missing space after source: %s", command_buf.buf);
2317 } else {
2318 endp = strchr(s, ' ');
2319 if (!endp)
2320 die("Missing space after source: %s", command_buf.buf);
2321 strbuf_add(&s_uq, s, endp - s);
2322 }
2323 s = s_uq.buf;
2324
2325 endp++;
2326 if (!*endp)
2327 die("Missing dest: %s", command_buf.buf);
2328
2329 d = endp;
2330 strbuf_reset(&d_uq);
2331 if (!unquote_c_style(&d_uq, d, &endp)) {
2332 if (*endp)
2333 die("Garbage after dest in: %s", command_buf.buf);
2334 d = d_uq.buf;
2335 }
2336
2337 memset(&leaf, 0, sizeof(leaf));
2338 if (rename)
2339 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2340 else
2341 tree_content_get(&b->branch_tree, s, &leaf, 1);
2342 if (!leaf.versions[1].mode)
2343 die("Path %s not in branch", s);
2344 if (!*d) { /* C "path/to/subdir" "" */
2345 tree_content_replace(&b->branch_tree,
2346 &leaf.versions[1].oid,
2347 leaf.versions[1].mode,
2348 leaf.tree);
2349 return;
2350 }
2351 tree_content_set(&b->branch_tree, d,
2352 &leaf.versions[1].oid,
2353 leaf.versions[1].mode,
2354 leaf.tree);
2355 }
2356
2357 static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2358 {
2359 static struct strbuf uq = STRBUF_INIT;
2360 struct object_entry *oe;
2361 struct branch *s;
2362 struct object_id oid, commit_oid;
2363 char path[GIT_MAX_RAWSZ * 3];
2364 uint16_t inline_data = 0;
2365 unsigned char new_fanout;
2366
2367 /*
2368 * When loading a branch, we don't traverse its tree to count the real
2369 * number of notes (too expensive to do this for all non-note refs).
2370 * This means that recently loaded notes refs might incorrectly have
2371 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2372 *
2373 * Fix this by traversing the tree and counting the number of notes
2374 * when b->num_notes == 0. If the notes tree is truly empty, the
2375 * calculation should not take long.
2376 */
2377 if (b->num_notes == 0 && *old_fanout == 0) {
2378 /* Invoke change_note_fanout() in "counting mode". */
2379 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2380 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2381 }
2382
2383 /* Now parse the notemodify command. */
2384 /* <dataref> or 'inline' */
2385 if (*p == ':') {
2386 oe = find_mark(parse_mark_ref_space(&p));
2387 oidcpy(&oid, &oe->idx.oid);
2388 } else if (skip_prefix(p, "inline ", &p)) {
2389 inline_data = 1;
2390 oe = NULL; /* not used with inline_data, but makes gcc happy */
2391 } else {
2392 if (parse_oid_hex(p, &oid, &p))
2393 die("Invalid dataref: %s", command_buf.buf);
2394 oe = find_object(&oid);
2395 if (*p++ != ' ')
2396 die("Missing space after SHA1: %s", command_buf.buf);
2397 }
2398
2399 /* <commit-ish> */
2400 s = lookup_branch(p);
2401 if (s) {
2402 if (is_null_oid(&s->oid))
2403 die("Can't add a note on empty branch.");
2404 oidcpy(&commit_oid, &s->oid);
2405 } else if (*p == ':') {
2406 uintmax_t commit_mark = parse_mark_ref_eol(p);
2407 struct object_entry *commit_oe = find_mark(commit_mark);
2408 if (commit_oe->type != OBJ_COMMIT)
2409 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2410 oidcpy(&commit_oid, &commit_oe->idx.oid);
2411 } else if (!get_oid(p, &commit_oid)) {
2412 unsigned long size;
2413 char *buf = read_object_with_reference(&commit_oid,
2414 commit_type, &size,
2415 &commit_oid);
2416 if (!buf || size < the_hash_algo->hexsz + 6)
2417 die("Not a valid commit: %s", p);
2418 free(buf);
2419 } else
2420 die("Invalid ref name or SHA1 expression: %s", p);
2421
2422 if (inline_data) {
2423 if (p != uq.buf) {
2424 strbuf_addstr(&uq, p);
2425 p = uq.buf;
2426 }
2427 read_next_command();
2428 parse_and_store_blob(&last_blob, &oid, 0);
2429 } else if (oe) {
2430 if (oe->type != OBJ_BLOB)
2431 die("Not a blob (actually a %s): %s",
2432 type_name(oe->type), command_buf.buf);
2433 } else if (!is_null_oid(&oid)) {
2434 enum object_type type = oid_object_info(the_repository, &oid,
2435 NULL);
2436 if (type < 0)
2437 die("Blob not found: %s", command_buf.buf);
2438 if (type != OBJ_BLOB)
2439 die("Not a blob (actually a %s): %s",
2440 type_name(type), command_buf.buf);
2441 }
2442
2443 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2444 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2445 b->num_notes--;
2446
2447 if (is_null_oid(&oid))
2448 return; /* nothing to insert */
2449
2450 b->num_notes++;
2451 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2452 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2453 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2454 }
2455
2456 static void file_change_deleteall(struct branch *b)
2457 {
2458 release_tree_content_recursive(b->branch_tree.tree);
2459 oidclr(&b->branch_tree.versions[0].oid);
2460 oidclr(&b->branch_tree.versions[1].oid);
2461 load_tree(&b->branch_tree);
2462 b->num_notes = 0;
2463 }
2464
2465 static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2466 {
2467 if (!buf || size < the_hash_algo->hexsz + 6)
2468 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2469 if (memcmp("tree ", buf, 5)
2470 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2471 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2472 oidcpy(&b->branch_tree.versions[0].oid,
2473 &b->branch_tree.versions[1].oid);
2474 }
2475
2476 static void parse_from_existing(struct branch *b)
2477 {
2478 if (is_null_oid(&b->oid)) {
2479 oidclr(&b->branch_tree.versions[0].oid);
2480 oidclr(&b->branch_tree.versions[1].oid);
2481 } else {
2482 unsigned long size;
2483 char *buf;
2484
2485 buf = read_object_with_reference(&b->oid, commit_type, &size,
2486 &b->oid);
2487 parse_from_commit(b, buf, size);
2488 free(buf);
2489 }
2490 }
2491
2492 static int parse_from(struct branch *b)
2493 {
2494 const char *from;
2495 struct branch *s;
2496 struct object_id oid;
2497
2498 if (!skip_prefix(command_buf.buf, "from ", &from))
2499 return 0;
2500
2501 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2502
2503 s = lookup_branch(from);
2504 if (b == s)
2505 die("Can't create a branch from itself: %s", b->name);
2506 else if (s) {
2507 struct object_id *t = &s->branch_tree.versions[1].oid;
2508 oidcpy(&b->oid, &s->oid);
2509 oidcpy(&b->branch_tree.versions[0].oid, t);
2510 oidcpy(&b->branch_tree.versions[1].oid, t);
2511 } else if (*from == ':') {
2512 uintmax_t idnum = parse_mark_ref_eol(from);
2513 struct object_entry *oe = find_mark(idnum);
2514 if (oe->type != OBJ_COMMIT)
2515 die("Mark :%" PRIuMAX " not a commit", idnum);
2516 if (!oideq(&b->oid, &oe->idx.oid)) {
2517 oidcpy(&b->oid, &oe->idx.oid);
2518 if (oe->pack_id != MAX_PACK_ID) {
2519 unsigned long size;
2520 char *buf = gfi_unpack_entry(oe, &size);
2521 parse_from_commit(b, buf, size);
2522 free(buf);
2523 } else
2524 parse_from_existing(b);
2525 }
2526 } else if (!get_oid(from, &b->oid)) {
2527 parse_from_existing(b);
2528 if (is_null_oid(&b->oid))
2529 b->delete = 1;
2530 }
2531 else
2532 die("Invalid ref name or SHA1 expression: %s", from);
2533
2534 if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2535 release_tree_content_recursive(b->branch_tree.tree);
2536 b->branch_tree.tree = NULL;
2537 }
2538
2539 read_next_command();
2540 return 1;
2541 }
2542
2543 static struct hash_list *parse_merge(unsigned int *count)
2544 {
2545 struct hash_list *list = NULL, **tail = &list, *n;
2546 const char *from;
2547 struct branch *s;
2548
2549 *count = 0;
2550 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2551 n = xmalloc(sizeof(*n));
2552 s = lookup_branch(from);
2553 if (s)
2554 oidcpy(&n->oid, &s->oid);
2555 else if (*from == ':') {
2556 uintmax_t idnum = parse_mark_ref_eol(from);
2557 struct object_entry *oe = find_mark(idnum);
2558 if (oe->type != OBJ_COMMIT)
2559 die("Mark :%" PRIuMAX " not a commit", idnum);
2560 oidcpy(&n->oid, &oe->idx.oid);
2561 } else if (!get_oid(from, &n->oid)) {
2562 unsigned long size;
2563 char *buf = read_object_with_reference(&n->oid,
2564 commit_type,
2565 &size, &n->oid);
2566 if (!buf || size < the_hash_algo->hexsz + 6)
2567 die("Not a valid commit: %s", from);
2568 free(buf);
2569 } else
2570 die("Invalid ref name or SHA1 expression: %s", from);
2571
2572 n->next = NULL;
2573 *tail = n;
2574 tail = &n->next;
2575
2576 (*count)++;
2577 read_next_command();
2578 }
2579 return list;
2580 }
2581
2582 static void parse_new_commit(const char *arg)
2583 {
2584 static struct strbuf msg = STRBUF_INIT;
2585 struct branch *b;
2586 char *author = NULL;
2587 char *committer = NULL;
2588 const char *encoding = NULL;
2589 struct hash_list *merge_list = NULL;
2590 unsigned int merge_count;
2591 unsigned char prev_fanout, new_fanout;
2592 const char *v;
2593
2594 b = lookup_branch(arg);
2595 if (!b)
2596 b = new_branch(arg);
2597
2598 read_next_command();
2599 parse_mark();
2600 parse_original_identifier();
2601 if (skip_prefix(command_buf.buf, "author ", &v)) {
2602 author = parse_ident(v);
2603 read_next_command();
2604 }
2605 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2606 committer = parse_ident(v);
2607 read_next_command();
2608 }
2609 if (!committer)
2610 die("Expected committer but didn't get one");
2611 if (skip_prefix(command_buf.buf, "encoding ", &encoding))
2612 read_next_command();
2613 parse_data(&msg, 0, NULL);
2614 read_next_command();
2615 parse_from(b);
2616 merge_list = parse_merge(&merge_count);
2617
2618 /* ensure the branch is active/loaded */
2619 if (!b->branch_tree.tree || !max_active_branches) {
2620 unload_one_branch();
2621 load_branch(b);
2622 }
2623
2624 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2625
2626 /* file_change* */
2627 while (command_buf.len > 0) {
2628 if (skip_prefix(command_buf.buf, "M ", &v))
2629 file_change_m(v, b);
2630 else if (skip_prefix(command_buf.buf, "D ", &v))
2631 file_change_d(v, b);
2632 else if (skip_prefix(command_buf.buf, "R ", &v))
2633 file_change_cr(v, b, 1);
2634 else if (skip_prefix(command_buf.buf, "C ", &v))
2635 file_change_cr(v, b, 0);
2636 else if (skip_prefix(command_buf.buf, "N ", &v))
2637 note_change_n(v, b, &prev_fanout);
2638 else if (!strcmp("deleteall", command_buf.buf))
2639 file_change_deleteall(b);
2640 else if (skip_prefix(command_buf.buf, "ls ", &v))
2641 parse_ls(v, b);
2642 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
2643 parse_cat_blob(v);
2644 else {
2645 unread_command_buf = 1;
2646 break;
2647 }
2648 if (read_next_command() == EOF)
2649 break;
2650 }
2651
2652 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2653 if (new_fanout != prev_fanout)
2654 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2655
2656 /* build the tree and the commit */
2657 store_tree(&b->branch_tree);
2658 oidcpy(&b->branch_tree.versions[0].oid,
2659 &b->branch_tree.versions[1].oid);
2660
2661 strbuf_reset(&new_data);
2662 strbuf_addf(&new_data, "tree %s\n",
2663 oid_to_hex(&b->branch_tree.versions[1].oid));
2664 if (!is_null_oid(&b->oid))
2665 strbuf_addf(&new_data, "parent %s\n",
2666 oid_to_hex(&b->oid));
2667 while (merge_list) {
2668 struct hash_list *next = merge_list->next;
2669 strbuf_addf(&new_data, "parent %s\n",
2670 oid_to_hex(&merge_list->oid));
2671 free(merge_list);
2672 merge_list = next;
2673 }
2674 strbuf_addf(&new_data,
2675 "author %s\n"
2676 "committer %s\n",
2677 author ? author : committer, committer);
2678 if (encoding)
2679 strbuf_addf(&new_data,
2680 "encoding %s\n",
2681 encoding);
2682 strbuf_addch(&new_data, '\n');
2683 strbuf_addbuf(&new_data, &msg);
2684 free(author);
2685 free(committer);
2686
2687 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2688 b->pack_id = pack_id;
2689 b->last_commit = object_count_by_type[OBJ_COMMIT];
2690 }
2691
2692 static void parse_new_tag(const char *arg)
2693 {
2694 static struct strbuf msg = STRBUF_INIT;
2695 const char *from;
2696 char *tagger;
2697 struct branch *s;
2698 struct tag *t;
2699 uintmax_t from_mark = 0;
2700 struct object_id oid;
2701 enum object_type type;
2702 const char *v;
2703
2704 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2705 memset(t, 0, sizeof(struct tag));
2706 t->name = pool_strdup(arg);
2707 if (last_tag)
2708 last_tag->next_tag = t;
2709 else
2710 first_tag = t;
2711 last_tag = t;
2712 read_next_command();
2713
2714 /* from ... */
2715 if (!skip_prefix(command_buf.buf, "from ", &from))
2716 die("Expected from command, got %s", command_buf.buf);
2717 s = lookup_branch(from);
2718 if (s) {
2719 if (is_null_oid(&s->oid))
2720 die("Can't tag an empty branch.");
2721 oidcpy(&oid, &s->oid);
2722 type = OBJ_COMMIT;
2723 } else if (*from == ':') {
2724 struct object_entry *oe;
2725 from_mark = parse_mark_ref_eol(from);
2726 oe = find_mark(from_mark);
2727 type = oe->type;
2728 oidcpy(&oid, &oe->idx.oid);
2729 } else if (!get_oid(from, &oid)) {
2730 struct object_entry *oe = find_object(&oid);
2731 if (!oe) {
2732 type = oid_object_info(the_repository, &oid, NULL);
2733 if (type < 0)
2734 die("Not a valid object: %s", from);
2735 } else
2736 type = oe->type;
2737 } else
2738 die("Invalid ref name or SHA1 expression: %s", from);
2739 read_next_command();
2740
2741 /* original-oid ... */
2742 parse_original_identifier();
2743
2744 /* tagger ... */
2745 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2746 tagger = parse_ident(v);
2747 read_next_command();
2748 } else
2749 tagger = NULL;
2750
2751 /* tag payload/message */
2752 parse_data(&msg, 0, NULL);
2753
2754 /* build the tag object */
2755 strbuf_reset(&new_data);
2756
2757 strbuf_addf(&new_data,
2758 "object %s\n"
2759 "type %s\n"
2760 "tag %s\n",
2761 oid_to_hex(&oid), type_name(type), t->name);
2762 if (tagger)
2763 strbuf_addf(&new_data,
2764 "tagger %s\n", tagger);
2765 strbuf_addch(&new_data, '\n');
2766 strbuf_addbuf(&new_data, &msg);
2767 free(tagger);
2768
2769 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2770 t->pack_id = MAX_PACK_ID;
2771 else
2772 t->pack_id = pack_id;
2773 }
2774
2775 static void parse_reset_branch(const char *arg)
2776 {
2777 struct branch *b;
2778
2779 b = lookup_branch(arg);
2780 if (b) {
2781 oidclr(&b->oid);
2782 oidclr(&b->branch_tree.versions[0].oid);
2783 oidclr(&b->branch_tree.versions[1].oid);
2784 if (b->branch_tree.tree) {
2785 release_tree_content_recursive(b->branch_tree.tree);
2786 b->branch_tree.tree = NULL;
2787 }
2788 }
2789 else
2790 b = new_branch(arg);
2791 read_next_command();
2792 parse_from(b);
2793 if (command_buf.len > 0)
2794 unread_command_buf = 1;
2795 }
2796
2797 static void cat_blob_write(const char *buf, unsigned long size)
2798 {
2799 if (write_in_full(cat_blob_fd, buf, size) < 0)
2800 die_errno("Write to frontend failed");
2801 }
2802
2803 static void cat_blob(struct object_entry *oe, struct object_id *oid)
2804 {
2805 struct strbuf line = STRBUF_INIT;
2806 unsigned long size;
2807 enum object_type type = 0;
2808 char *buf;
2809
2810 if (!oe || oe->pack_id == MAX_PACK_ID) {
2811 buf = read_object_file(oid, &type, &size);
2812 } else {
2813 type = oe->type;
2814 buf = gfi_unpack_entry(oe, &size);
2815 }
2816
2817 /*
2818 * Output based on batch_one_object() from cat-file.c.
2819 */
2820 if (type <= 0) {
2821 strbuf_reset(&line);
2822 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2823 cat_blob_write(line.buf, line.len);
2824 strbuf_release(&line);
2825 free(buf);
2826 return;
2827 }
2828 if (!buf)
2829 die("Can't read object %s", oid_to_hex(oid));
2830 if (type != OBJ_BLOB)
2831 die("Object %s is a %s but a blob was expected.",
2832 oid_to_hex(oid), type_name(type));
2833 strbuf_reset(&line);
2834 strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2835 type_name(type), (uintmax_t)size);
2836 cat_blob_write(line.buf, line.len);
2837 strbuf_release(&line);
2838 cat_blob_write(buf, size);
2839 cat_blob_write("\n", 1);
2840 if (oe && oe->pack_id == pack_id) {
2841 last_blob.offset = oe->idx.offset;
2842 strbuf_attach(&last_blob.data, buf, size, size);
2843 last_blob.depth = oe->depth;
2844 } else
2845 free(buf);
2846 }
2847
2848 static void parse_get_mark(const char *p)
2849 {
2850 struct object_entry *oe;
2851 char output[GIT_MAX_HEXSZ + 2];
2852
2853 /* get-mark SP <object> LF */
2854 if (*p != ':')
2855 die("Not a mark: %s", p);
2856
2857 oe = find_mark(parse_mark_ref_eol(p));
2858 if (!oe)
2859 die("Unknown mark: %s", command_buf.buf);
2860
2861 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2862 cat_blob_write(output, the_hash_algo->hexsz + 1);
2863 }
2864
2865 static void parse_cat_blob(const char *p)
2866 {
2867 struct object_entry *oe;
2868 struct object_id oid;
2869
2870 /* cat-blob SP <object> LF */
2871 if (*p == ':') {
2872 oe = find_mark(parse_mark_ref_eol(p));
2873 if (!oe)
2874 die("Unknown mark: %s", command_buf.buf);
2875 oidcpy(&oid, &oe->idx.oid);
2876 } else {
2877 if (parse_oid_hex(p, &oid, &p))
2878 die("Invalid dataref: %s", command_buf.buf);
2879 if (*p)
2880 die("Garbage after SHA1: %s", command_buf.buf);
2881 oe = find_object(&oid);
2882 }
2883
2884 cat_blob(oe, &oid);
2885 }
2886
2887 static struct object_entry *dereference(struct object_entry *oe,
2888 struct object_id *oid)
2889 {
2890 unsigned long size;
2891 char *buf = NULL;
2892 const unsigned hexsz = the_hash_algo->hexsz;
2893
2894 if (!oe) {
2895 enum object_type type = oid_object_info(the_repository, oid,
2896 NULL);
2897 if (type < 0)
2898 die("object not found: %s", oid_to_hex(oid));
2899 /* cache it! */
2900 oe = insert_object(oid);
2901 oe->type = type;
2902 oe->pack_id = MAX_PACK_ID;
2903 oe->idx.offset = 1;
2904 }
2905 switch (oe->type) {
2906 case OBJ_TREE: /* easy case. */
2907 return oe;
2908 case OBJ_COMMIT:
2909 case OBJ_TAG:
2910 break;
2911 default:
2912 die("Not a tree-ish: %s", command_buf.buf);
2913 }
2914
2915 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
2916 buf = gfi_unpack_entry(oe, &size);
2917 } else {
2918 enum object_type unused;
2919 buf = read_object_file(oid, &unused, &size);
2920 }
2921 if (!buf)
2922 die("Can't load object %s", oid_to_hex(oid));
2923
2924 /* Peel one layer. */
2925 switch (oe->type) {
2926 case OBJ_TAG:
2927 if (size < hexsz + strlen("object ") ||
2928 get_oid_hex(buf + strlen("object "), oid))
2929 die("Invalid SHA1 in tag: %s", command_buf.buf);
2930 break;
2931 case OBJ_COMMIT:
2932 if (size < hexsz + strlen("tree ") ||
2933 get_oid_hex(buf + strlen("tree "), oid))
2934 die("Invalid SHA1 in commit: %s", command_buf.buf);
2935 }
2936
2937 free(buf);
2938 return find_object(oid);
2939 }
2940
2941 static struct object_entry *parse_treeish_dataref(const char **p)
2942 {
2943 struct object_id oid;
2944 struct object_entry *e;
2945
2946 if (**p == ':') { /* <mark> */
2947 e = find_mark(parse_mark_ref_space(p));
2948 if (!e)
2949 die("Unknown mark: %s", command_buf.buf);
2950 oidcpy(&oid, &e->idx.oid);
2951 } else { /* <sha1> */
2952 if (parse_oid_hex(*p, &oid, p))
2953 die("Invalid dataref: %s", command_buf.buf);
2954 e = find_object(&oid);
2955 if (*(*p)++ != ' ')
2956 die("Missing space after tree-ish: %s", command_buf.buf);
2957 }
2958
2959 while (!e || e->type != OBJ_TREE)
2960 e = dereference(e, &oid);
2961 return e;
2962 }
2963
2964 static void print_ls(int mode, const unsigned char *hash, const char *path)
2965 {
2966 static struct strbuf line = STRBUF_INIT;
2967
2968 /* See show_tree(). */
2969 const char *type =
2970 S_ISGITLINK(mode) ? commit_type :
2971 S_ISDIR(mode) ? tree_type :
2972 blob_type;
2973
2974 if (!mode) {
2975 /* missing SP path LF */
2976 strbuf_reset(&line);
2977 strbuf_addstr(&line, "missing ");
2978 quote_c_style(path, &line, NULL, 0);
2979 strbuf_addch(&line, '\n');
2980 } else {
2981 /* mode SP type SP object_name TAB path LF */
2982 strbuf_reset(&line);
2983 strbuf_addf(&line, "%06o %s %s\t",
2984 mode & ~NO_DELTA, type, hash_to_hex(hash));
2985 quote_c_style(path, &line, NULL, 0);
2986 strbuf_addch(&line, '\n');
2987 }
2988 cat_blob_write(line.buf, line.len);
2989 }
2990
2991 static void parse_ls(const char *p, struct branch *b)
2992 {
2993 struct tree_entry *root = NULL;
2994 struct tree_entry leaf = {NULL};
2995
2996 /* ls SP (<tree-ish> SP)? <path> */
2997 if (*p == '"') {
2998 if (!b)
2999 die("Not in a commit: %s", command_buf.buf);
3000 root = &b->branch_tree;
3001 } else {
3002 struct object_entry *e = parse_treeish_dataref(&p);
3003 root = new_tree_entry();
3004 oidcpy(&root->versions[1].oid, &e->idx.oid);
3005 if (!is_null_oid(&root->versions[1].oid))
3006 root->versions[1].mode = S_IFDIR;
3007 load_tree(root);
3008 }
3009 if (*p == '"') {
3010 static struct strbuf uq = STRBUF_INIT;
3011 const char *endp;
3012 strbuf_reset(&uq);
3013 if (unquote_c_style(&uq, p, &endp))
3014 die("Invalid path: %s", command_buf.buf);
3015 if (*endp)
3016 die("Garbage after path in: %s", command_buf.buf);
3017 p = uq.buf;
3018 }
3019 tree_content_get(root, p, &leaf, 1);
3020 /*
3021 * A directory in preparation would have a sha1 of zero
3022 * until it is saved. Save, for simplicity.
3023 */
3024 if (S_ISDIR(leaf.versions[1].mode))
3025 store_tree(&leaf);
3026
3027 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3028 if (leaf.tree)
3029 release_tree_content_recursive(leaf.tree);
3030 if (!b || root != &b->branch_tree)
3031 release_tree_entry(root);
3032 }
3033
3034 static void checkpoint(void)
3035 {
3036 checkpoint_requested = 0;
3037 if (object_count) {
3038 cycle_packfile();
3039 }
3040 dump_branches();
3041 dump_tags();
3042 dump_marks();
3043 }
3044
3045 static void parse_checkpoint(void)
3046 {
3047 checkpoint_requested = 1;
3048 skip_optional_lf();
3049 }
3050
3051 static void parse_progress(void)
3052 {
3053 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3054 fputc('\n', stdout);
3055 fflush(stdout);
3056 skip_optional_lf();
3057 }
3058
3059 static char* make_fast_import_path(const char *path)
3060 {
3061 if (!relative_marks_paths || is_absolute_path(path))
3062 return xstrdup(path);
3063 return git_pathdup("info/fast-import/%s", path);
3064 }
3065
3066 static void option_import_marks(const char *marks,
3067 int from_stream, int ignore_missing)
3068 {
3069 if (import_marks_file) {
3070 if (from_stream)
3071 die("Only one import-marks command allowed per stream");
3072
3073 /* read previous mark file */
3074 if(!import_marks_file_from_stream)
3075 read_marks();
3076 }
3077
3078 import_marks_file = make_fast_import_path(marks);
3079 safe_create_leading_directories_const(import_marks_file);
3080 import_marks_file_from_stream = from_stream;
3081 import_marks_file_ignore_missing = ignore_missing;
3082 }
3083
3084 static void option_date_format(const char *fmt)
3085 {
3086 if (!strcmp(fmt, "raw"))
3087 whenspec = WHENSPEC_RAW;
3088 else if (!strcmp(fmt, "rfc2822"))
3089 whenspec = WHENSPEC_RFC2822;
3090 else if (!strcmp(fmt, "now"))
3091 whenspec = WHENSPEC_NOW;
3092 else
3093 die("unknown --date-format argument %s", fmt);
3094 }
3095
3096 static unsigned long ulong_arg(const char *option, const char *arg)
3097 {
3098 char *endptr;
3099 unsigned long rv = strtoul(arg, &endptr, 0);
3100 if (strchr(arg, '-') || endptr == arg || *endptr)
3101 die("%s: argument must be a non-negative integer", option);
3102 return rv;
3103 }
3104
3105 static void option_depth(const char *depth)
3106 {
3107 max_depth = ulong_arg("--depth", depth);
3108 if (max_depth > MAX_DEPTH)
3109 die("--depth cannot exceed %u", MAX_DEPTH);
3110 }
3111
3112 static void option_active_branches(const char *branches)
3113 {
3114 max_active_branches = ulong_arg("--active-branches", branches);
3115 }
3116
3117 static void option_export_marks(const char *marks)
3118 {
3119 export_marks_file = make_fast_import_path(marks);
3120 safe_create_leading_directories_const(export_marks_file);
3121 }
3122
3123 static void option_cat_blob_fd(const char *fd)
3124 {
3125 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3126 if (n > (unsigned long) INT_MAX)
3127 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3128 cat_blob_fd = (int) n;
3129 }
3130
3131 static void option_export_pack_edges(const char *edges)
3132 {
3133 if (pack_edges)
3134 fclose(pack_edges);
3135 pack_edges = xfopen(edges, "a");
3136 }
3137
3138 static int parse_one_option(const char *option)
3139 {
3140 if (skip_prefix(option, "max-pack-size=", &option)) {
3141 unsigned long v;
3142 if (!git_parse_ulong(option, &v))
3143 return 0;
3144 if (v < 8192) {
3145 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3146 v *= 1024 * 1024;
3147 } else if (v < 1024 * 1024) {
3148 warning("minimum max-pack-size is 1 MiB");
3149 v = 1024 * 1024;
3150 }
3151 max_packsize = v;
3152 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3153 unsigned long v;
3154 if (!git_parse_ulong(option, &v))
3155 return 0;
3156 big_file_threshold = v;
3157 } else if (skip_prefix(option, "depth=", &option)) {
3158 option_depth(option);
3159 } else if (skip_prefix(option, "active-branches=", &option)) {
3160 option_active_branches(option);
3161 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3162 option_export_pack_edges(option);
3163 } else if (starts_with(option, "quiet")) {
3164 show_stats = 0;
3165 } else if (starts_with(option, "stats")) {
3166 show_stats = 1;
3167 } else {
3168 return 0;
3169 }
3170
3171 return 1;
3172 }
3173
3174 static int parse_one_feature(const char *feature, int from_stream)
3175 {
3176 const char *arg;
3177
3178 if (skip_prefix(feature, "date-format=", &arg)) {
3179 option_date_format(arg);
3180 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3181 option_import_marks(arg, from_stream, 0);
3182 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3183 option_import_marks(arg, from_stream, 1);
3184 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3185 option_export_marks(arg);
3186 } else if (!strcmp(feature, "get-mark")) {
3187 ; /* Don't die - this feature is supported */
3188 } else if (!strcmp(feature, "cat-blob")) {
3189 ; /* Don't die - this feature is supported */
3190 } else if (!strcmp(feature, "relative-marks")) {
3191 relative_marks_paths = 1;
3192 } else if (!strcmp(feature, "no-relative-marks")) {
3193 relative_marks_paths = 0;
3194 } else if (!strcmp(feature, "done")) {
3195 require_explicit_termination = 1;
3196 } else if (!strcmp(feature, "force")) {
3197 force_update = 1;
3198 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3199 ; /* do nothing; we have the feature */
3200 } else {
3201 return 0;
3202 }
3203
3204 return 1;
3205 }
3206
3207 static void parse_feature(const char *feature)
3208 {
3209 if (seen_data_command)
3210 die("Got feature command '%s' after data command", feature);
3211
3212 if (parse_one_feature(feature, 1))
3213 return;
3214
3215 die("This version of fast-import does not support feature %s.", feature);
3216 }
3217
3218 static void parse_option(const char *option)
3219 {
3220 if (seen_data_command)
3221 die("Got option command '%s' after data command", option);
3222
3223 if (parse_one_option(option))
3224 return;
3225
3226 die("This version of fast-import does not support option: %s", option);
3227 }
3228
3229 static void git_pack_config(void)
3230 {
3231 int indexversion_value;
3232 int limit;
3233 unsigned long packsizelimit_value;
3234
3235 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3236 if (max_depth > MAX_DEPTH)
3237 max_depth = MAX_DEPTH;
3238 }
3239 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3240 pack_idx_opts.version = indexversion_value;
3241 if (pack_idx_opts.version > 2)
3242 git_die_config("pack.indexversion",
3243 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3244 }
3245 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3246 max_packsize = packsizelimit_value;
3247
3248 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3249 unpack_limit = limit;
3250 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3251 unpack_limit = limit;
3252
3253 git_config(git_default_config, NULL);
3254 }
3255
3256 static const char fast_import_usage[] =
3257 "git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3258
3259 static void parse_argv(void)
3260 {
3261 unsigned int i;
3262
3263 for (i = 1; i < global_argc; i++) {
3264 const char *a = global_argv[i];
3265
3266 if (*a != '-' || !strcmp(a, "--"))
3267 break;
3268
3269 if (!skip_prefix(a, "--", &a))
3270 die("unknown option %s", a);
3271
3272 if (parse_one_option(a))
3273 continue;
3274
3275 if (parse_one_feature(a, 0))
3276 continue;
3277
3278 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3279 option_cat_blob_fd(a);
3280 continue;
3281 }
3282
3283 die("unknown option --%s", a);
3284 }
3285 if (i != global_argc)
3286 usage(fast_import_usage);
3287
3288 seen_data_command = 1;
3289 if (import_marks_file)
3290 read_marks();
3291 }
3292
3293 int cmd_main(int argc, const char **argv)
3294 {
3295 unsigned int i;
3296
3297 if (argc == 2 && !strcmp(argv[1], "-h"))
3298 usage(fast_import_usage);
3299
3300 setup_git_directory();
3301 reset_pack_idx_option(&pack_idx_opts);
3302 git_pack_config();
3303
3304 alloc_objects(object_entry_alloc);
3305 strbuf_init(&command_buf, 0);
3306 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3307 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3308 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3309 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3310
3311 global_argc = argc;
3312 global_argv = argv;
3313
3314 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3315 for (i = 0; i < (cmd_save - 1); i++)
3316 rc_free[i].next = &rc_free[i + 1];
3317 rc_free[cmd_save - 1].next = NULL;
3318
3319 start_packfile();
3320 set_die_routine(die_nicely);
3321 set_checkpoint_signal();
3322 while (read_next_command() != EOF) {
3323 const char *v;
3324 if (!strcmp("blob", command_buf.buf))
3325 parse_new_blob();
3326 else if (skip_prefix(command_buf.buf, "commit ", &v))
3327 parse_new_commit(v);
3328 else if (skip_prefix(command_buf.buf, "tag ", &v))
3329 parse_new_tag(v);
3330 else if (skip_prefix(command_buf.buf, "reset ", &v))
3331 parse_reset_branch(v);
3332 else if (skip_prefix(command_buf.buf, "ls ", &v))
3333 parse_ls(v, NULL);
3334 else if (skip_prefix(command_buf.buf, "cat-blob ", &v))
3335 parse_cat_blob(v);
3336 else if (skip_prefix(command_buf.buf, "get-mark ", &v))
3337 parse_get_mark(v);
3338 else if (!strcmp("checkpoint", command_buf.buf))
3339 parse_checkpoint();
3340 else if (!strcmp("done", command_buf.buf))
3341 break;
3342 else if (starts_with(command_buf.buf, "progress "))
3343 parse_progress();
3344 else if (skip_prefix(command_buf.buf, "feature ", &v))
3345 parse_feature(v);
3346 else if (skip_prefix(command_buf.buf, "option git ", &v))
3347 parse_option(v);
3348 else if (starts_with(command_buf.buf, "option "))
3349 /* ignore non-git options*/;
3350 else
3351 die("Unsupported command: %s", command_buf.buf);
3352
3353 if (checkpoint_requested)
3354 checkpoint();
3355 }
3356
3357 /* argv hasn't been parsed yet, do so */
3358 if (!seen_data_command)
3359 parse_argv();
3360
3361 if (require_explicit_termination && feof(stdin))
3362 die("stream ends early");
3363
3364 end_packfile();
3365
3366 dump_branches();
3367 dump_tags();
3368 unkeep_all_packs();
3369 dump_marks();
3370
3371 if (pack_edges)
3372 fclose(pack_edges);
3373
3374 if (show_stats) {
3375 uintmax_t total_count = 0, duplicate_count = 0;
3376 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3377 total_count += object_count_by_type[i];
3378 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3379 duplicate_count += duplicate_count_by_type[i];
3380
3381 fprintf(stderr, "%s statistics:\n", argv[0]);
3382 fprintf(stderr, "---------------------------------------------------------------------\n");
3383 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3384 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3385 fprintf(stderr, " blobs : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_BLOB], duplicate_count_by_type[OBJ_BLOB], delta_count_by_type[OBJ_BLOB], delta_count_attempts_by_type[OBJ_BLOB]);
3386 fprintf(stderr, " trees : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TREE], duplicate_count_by_type[OBJ_TREE], delta_count_by_type[OBJ_TREE], delta_count_attempts_by_type[OBJ_TREE]);
3387 fprintf(stderr, " commits: %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_COMMIT], duplicate_count_by_type[OBJ_COMMIT], delta_count_by_type[OBJ_COMMIT], delta_count_attempts_by_type[OBJ_COMMIT]);
3388 fprintf(stderr, " tags : %10" PRIuMAX " (%10" PRIuMAX " duplicates %10" PRIuMAX " deltas of %10" PRIuMAX" attempts)\n", object_count_by_type[OBJ_TAG], duplicate_count_by_type[OBJ_TAG], delta_count_by_type[OBJ_TAG], delta_count_attempts_by_type[OBJ_TAG]);
3389 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3390 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3391 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3392 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3393 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3394 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3395 fprintf(stderr, "---------------------------------------------------------------------\n");
3396 pack_report();
3397 fprintf(stderr, "---------------------------------------------------------------------\n");
3398 fprintf(stderr, "\n");
3399 }
3400
3401 return failure ? 1 : 0;
3402 }