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
32struct object_entry {
33 struct pack_idx_entry idx;
34 struct object_entry *next;
35 uint32_t type : TYPE_BITS,
36 pack_id : PACK_ID_BITS,
37 depth : DEPTH_BITS;
38};
39
40struct object_entry_pool {
41 struct object_entry_pool *next_pool;
42 struct object_entry *next_free;
43 struct object_entry *end;
44 struct object_entry entries[FLEX_ARRAY]; /* more */
45};
46
47struct mark_set {
48 union {
49 struct object_entry *marked[1024];
50 struct mark_set *sets[1024];
51 } data;
52 unsigned int shift;
53};
54
55struct last_object {
56 struct strbuf data;
57 off_t offset;
58 unsigned int depth;
59 unsigned no_swap : 1;
60};
61
62struct atom_str {
63 struct atom_str *next_atom;
64 unsigned short str_len;
65 char str_dat[FLEX_ARRAY]; /* more */
66};
67
68struct tree_content;
69struct tree_entry {
70 struct tree_content *tree;
71 struct atom_str *name;
72 struct tree_entry_ms {
73 uint16_t mode;
74 struct object_id oid;
75 } versions[2];
76};
77
78struct tree_content {
79 unsigned int entry_capacity; /* must match avail_tree_content */
80 unsigned int entry_count;
81 unsigned int delta_depth;
82 struct tree_entry *entries[FLEX_ARRAY]; /* more */
83};
84
85struct avail_tree_content {
86 unsigned int entry_capacity; /* must match tree_content */
87 struct avail_tree_content *next_avail;
88};
89
90struct branch {
91 struct branch *table_next_branch;
92 struct branch *active_next_branch;
93 const char *name;
94 struct tree_entry branch_tree;
95 uintmax_t last_commit;
96 uintmax_t num_notes;
97 unsigned active : 1;
98 unsigned delete : 1;
99 unsigned pack_id : PACK_ID_BITS;
100 struct object_id oid;
101};
102
103struct tag {
104 struct tag *next_tag;
105 const char *name;
106 unsigned int pack_id;
107 struct object_id oid;
108};
109
110struct hash_list {
111 struct hash_list *next;
112 struct object_id oid;
113};
114
115typedef enum {
116 WHENSPEC_RAW = 1,
117 WHENSPEC_RFC2822,
118 WHENSPEC_NOW
119} whenspec_type;
120
121struct recent_command {
122 struct recent_command *prev;
123 struct recent_command *next;
124 char *buf;
125};
126
127/* Configured limits on output */
128static unsigned long max_depth = 50;
129static off_t max_packsize;
130static int unpack_limit = 100;
131static int force_update;
132
133/* Stats and misc. counters */
134static uintmax_t alloc_count;
135static uintmax_t marks_set_count;
136static uintmax_t object_count_by_type[1 << TYPE_BITS];
137static uintmax_t duplicate_count_by_type[1 << TYPE_BITS];
138static uintmax_t delta_count_by_type[1 << TYPE_BITS];
139static uintmax_t delta_count_attempts_by_type[1 << TYPE_BITS];
140static unsigned long object_count;
141static unsigned long branch_count;
142static unsigned long branch_load_count;
143static int failure;
144static FILE *pack_edges;
145static unsigned int show_stats = 1;
146static int global_argc;
147static const char **global_argv;
148
149/* Memory pools */
150static struct mem_pool fi_mem_pool = {NULL, 2*1024*1024 -
151 sizeof(struct mp_block), 0 };
152
153/* Atom management */
154static unsigned int atom_table_sz = 4451;
155static unsigned int atom_cnt;
156static struct atom_str **atom_table;
157
158/* The .pack file being generated */
159static struct pack_idx_option pack_idx_opts;
160static unsigned int pack_id;
161static struct hashfile *pack_file;
162static struct packed_git *pack_data;
163static struct packed_git **all_packs;
164static off_t pack_size;
165
166/* Table of objects we've written. */
167static unsigned int object_entry_alloc = 5000;
168static struct object_entry_pool *blocks;
169static struct object_entry *object_table[1 << 16];
170static struct mark_set *marks;
171static const char *export_marks_file;
172static const char *import_marks_file;
173static int import_marks_file_from_stream;
174static int import_marks_file_ignore_missing;
175static int import_marks_file_done;
176static int relative_marks_paths;
177
178/* Our last blob */
179static struct last_object last_blob = { STRBUF_INIT, 0, 0, 0 };
180
181/* Tree management */
182static unsigned int tree_entry_alloc = 1000;
183static void *avail_tree_entry;
184static unsigned int avail_tree_table_sz = 100;
185static struct avail_tree_content **avail_tree_table;
186static size_t tree_entry_allocd;
187static struct strbuf old_tree = STRBUF_INIT;
188static struct strbuf new_tree = STRBUF_INIT;
189
190/* Branch data */
191static unsigned long max_active_branches = 5;
192static unsigned long cur_active_branches;
193static unsigned long branch_table_sz = 1039;
194static struct branch **branch_table;
195static struct branch *active_branches;
196
197/* Tag data */
198static struct tag *first_tag;
199static struct tag *last_tag;
200
201/* Input stream parsing */
202static whenspec_type whenspec = WHENSPEC_RAW;
203static struct strbuf command_buf = STRBUF_INIT;
204static int unread_command_buf;
205static struct recent_command cmd_hist = {&cmd_hist, &cmd_hist, NULL};
206static struct recent_command *cmd_tail = &cmd_hist;
207static struct recent_command *rc_free;
208static unsigned int cmd_save = 100;
209static uintmax_t next_mark;
210static struct strbuf new_data = STRBUF_INIT;
211static int seen_data_command;
212static int require_explicit_termination;
213
214/* Signal handling */
215static volatile sig_atomic_t checkpoint_requested;
216
217/* Where to write output of cat-blob commands */
218static int cat_blob_fd = STDOUT_FILENO;
219
220static void parse_argv(void);
221static void parse_get_mark(const char *p);
222static void parse_cat_blob(const char *p);
223static void parse_ls(const char *p, struct branch *b);
224
225static void write_branch_report(FILE *rpt, struct branch *b)
226{
227 fprintf(rpt, "%s:\n", b->name);
228
229 fprintf(rpt, " status :");
230 if (b->active)
231 fputs(" active", rpt);
232 if (b->branch_tree.tree)
233 fputs(" loaded", rpt);
234 if (is_null_oid(&b->branch_tree.versions[1].oid))
235 fputs(" dirty", rpt);
236 fputc('\n', rpt);
237
238 fprintf(rpt, " tip commit : %s\n", oid_to_hex(&b->oid));
239 fprintf(rpt, " old tree : %s\n",
240 oid_to_hex(&b->branch_tree.versions[0].oid));
241 fprintf(rpt, " cur tree : %s\n",
242 oid_to_hex(&b->branch_tree.versions[1].oid));
243 fprintf(rpt, " commit clock: %" PRIuMAX "\n", b->last_commit);
244
245 fputs(" last pack : ", rpt);
246 if (b->pack_id < MAX_PACK_ID)
247 fprintf(rpt, "%u", b->pack_id);
248 fputc('\n', rpt);
249
250 fputc('\n', rpt);
251}
252
253static void dump_marks_helper(FILE *, uintmax_t, struct mark_set *);
254
255static void write_crash_report(const char *err)
256{
257 char *loc = git_pathdup("fast_import_crash_%"PRIuMAX, (uintmax_t) getpid());
258 FILE *rpt = fopen(loc, "w");
259 struct branch *b;
260 unsigned long lu;
261 struct recent_command *rc;
262
263 if (!rpt) {
264 error_errno("can't write crash report %s", loc);
265 free(loc);
266 return;
267 }
268
269 fprintf(stderr, "fast-import: dumping crash report to %s\n", loc);
270
271 fprintf(rpt, "fast-import crash report:\n");
272 fprintf(rpt, " fast-import process: %"PRIuMAX"\n", (uintmax_t) getpid());
273 fprintf(rpt, " parent process : %"PRIuMAX"\n", (uintmax_t) getppid());
274 fprintf(rpt, " at %s\n", show_date(time(NULL), 0, DATE_MODE(ISO8601)));
275 fputc('\n', rpt);
276
277 fputs("fatal: ", rpt);
278 fputs(err, rpt);
279 fputc('\n', rpt);
280
281 fputc('\n', rpt);
282 fputs("Most Recent Commands Before Crash\n", rpt);
283 fputs("---------------------------------\n", rpt);
284 for (rc = cmd_hist.next; rc != &cmd_hist; rc = rc->next) {
285 if (rc->next == &cmd_hist)
286 fputs("* ", rpt);
287 else
288 fputs(" ", rpt);
289 fputs(rc->buf, rpt);
290 fputc('\n', rpt);
291 }
292
293 fputc('\n', rpt);
294 fputs("Active Branch LRU\n", rpt);
295 fputs("-----------------\n", rpt);
296 fprintf(rpt, " active_branches = %lu cur, %lu max\n",
297 cur_active_branches,
298 max_active_branches);
299 fputc('\n', rpt);
300 fputs(" pos clock name\n", rpt);
301 fputs(" ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\n", rpt);
302 for (b = active_branches, lu = 0; b; b = b->active_next_branch)
303 fprintf(rpt, " %2lu) %6" PRIuMAX" %s\n",
304 ++lu, b->last_commit, b->name);
305
306 fputc('\n', rpt);
307 fputs("Inactive Branches\n", rpt);
308 fputs("-----------------\n", rpt);
309 for (lu = 0; lu < branch_table_sz; lu++) {
310 for (b = branch_table[lu]; b; b = b->table_next_branch)
311 write_branch_report(rpt, b);
312 }
313
314 if (first_tag) {
315 struct tag *tg;
316 fputc('\n', rpt);
317 fputs("Annotated Tags\n", rpt);
318 fputs("--------------\n", rpt);
319 for (tg = first_tag; tg; tg = tg->next_tag) {
320 fputs(oid_to_hex(&tg->oid), rpt);
321 fputc(' ', rpt);
322 fputs(tg->name, rpt);
323 fputc('\n', rpt);
324 }
325 }
326
327 fputc('\n', rpt);
328 fputs("Marks\n", rpt);
329 fputs("-----\n", rpt);
330 if (export_marks_file)
331 fprintf(rpt, " exported to %s\n", export_marks_file);
332 else
333 dump_marks_helper(rpt, 0, marks);
334
335 fputc('\n', rpt);
336 fputs("-------------------\n", rpt);
337 fputs("END OF CRASH REPORT\n", rpt);
338 fclose(rpt);
339 free(loc);
340}
341
342static void end_packfile(void);
343static void unkeep_all_packs(void);
344static void dump_marks(void);
345
346static NORETURN void die_nicely(const char *err, va_list params)
347{
348 static int zombie;
349 char message[2 * PATH_MAX];
350
351 vsnprintf(message, sizeof(message), err, params);
352 fputs("fatal: ", stderr);
353 fputs(message, stderr);
354 fputc('\n', stderr);
355
356 if (!zombie) {
357 zombie = 1;
358 write_crash_report(message);
359 end_packfile();
360 unkeep_all_packs();
361 dump_marks();
362 }
363 exit(128);
364}
365
366#ifndef SIGUSR1 /* Windows, for example */
367
368static void set_checkpoint_signal(void)
369{
370}
371
372#else
373
374static void checkpoint_signal(int signo)
375{
376 checkpoint_requested = 1;
377}
378
379static void set_checkpoint_signal(void)
380{
381 struct sigaction sa;
382
383 memset(&sa, 0, sizeof(sa));
384 sa.sa_handler = checkpoint_signal;
385 sigemptyset(&sa.sa_mask);
386 sa.sa_flags = SA_RESTART;
387 sigaction(SIGUSR1, &sa, NULL);
388}
389
390#endif
391
392static void alloc_objects(unsigned int cnt)
393{
394 struct object_entry_pool *b;
395
396 b = xmalloc(sizeof(struct object_entry_pool)
397 + cnt * sizeof(struct object_entry));
398 b->next_pool = blocks;
399 b->next_free = b->entries;
400 b->end = b->entries + cnt;
401 blocks = b;
402 alloc_count += cnt;
403}
404
405static struct object_entry *new_object(struct object_id *oid)
406{
407 struct object_entry *e;
408
409 if (blocks->next_free == blocks->end)
410 alloc_objects(object_entry_alloc);
411
412 e = blocks->next_free++;
413 oidcpy(&e->idx.oid, oid);
414 return e;
415}
416
417static struct object_entry *find_object(struct object_id *oid)
418{
419 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
420 struct object_entry *e;
421 for (e = object_table[h]; e; e = e->next)
422 if (oideq(oid, &e->idx.oid))
423 return e;
424 return NULL;
425}
426
427static struct object_entry *insert_object(struct object_id *oid)
428{
429 unsigned int h = oid->hash[0] << 8 | oid->hash[1];
430 struct object_entry *e = object_table[h];
431
432 while (e) {
433 if (oideq(oid, &e->idx.oid))
434 return e;
435 e = e->next;
436 }
437
438 e = new_object(oid);
439 e->next = object_table[h];
440 e->idx.offset = 0;
441 object_table[h] = e;
442 return e;
443}
444
445static void invalidate_pack_id(unsigned int id)
446{
447 unsigned int h;
448 unsigned long lu;
449 struct tag *t;
450
451 for (h = 0; h < ARRAY_SIZE(object_table); h++) {
452 struct object_entry *e;
453
454 for (e = object_table[h]; e; e = e->next)
455 if (e->pack_id == id)
456 e->pack_id = MAX_PACK_ID;
457 }
458
459 for (lu = 0; lu < branch_table_sz; lu++) {
460 struct branch *b;
461
462 for (b = branch_table[lu]; b; b = b->table_next_branch)
463 if (b->pack_id == id)
464 b->pack_id = MAX_PACK_ID;
465 }
466
467 for (t = first_tag; t; t = t->next_tag)
468 if (t->pack_id == id)
469 t->pack_id = MAX_PACK_ID;
470}
471
472static unsigned int hc_str(const char *s, size_t len)
473{
474 unsigned int r = 0;
475 while (len-- > 0)
476 r = r * 31 + *s++;
477 return r;
478}
479
480static char *pool_strdup(const char *s)
481{
482 size_t len = strlen(s) + 1;
483 char *r = mem_pool_alloc(&fi_mem_pool, len);
484 memcpy(r, s, len);
485 return r;
486}
487
488static void insert_mark(uintmax_t idnum, struct object_entry *oe)
489{
490 struct mark_set *s = marks;
491 while ((idnum >> s->shift) >= 1024) {
492 s = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
493 s->shift = marks->shift + 10;
494 s->data.sets[0] = marks;
495 marks = s;
496 }
497 while (s->shift) {
498 uintmax_t i = idnum >> s->shift;
499 idnum -= i << s->shift;
500 if (!s->data.sets[i]) {
501 s->data.sets[i] = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
502 s->data.sets[i]->shift = s->shift - 10;
503 }
504 s = s->data.sets[i];
505 }
506 if (!s->data.marked[idnum])
507 marks_set_count++;
508 s->data.marked[idnum] = oe;
509}
510
511static struct object_entry *find_mark(uintmax_t idnum)
512{
513 uintmax_t orig_idnum = idnum;
514 struct mark_set *s = marks;
515 struct object_entry *oe = NULL;
516 if ((idnum >> s->shift) < 1024) {
517 while (s && s->shift) {
518 uintmax_t i = idnum >> s->shift;
519 idnum -= i << s->shift;
520 s = s->data.sets[i];
521 }
522 if (s)
523 oe = s->data.marked[idnum];
524 }
525 if (!oe)
526 die("mark :%" PRIuMAX " not declared", orig_idnum);
527 return oe;
528}
529
530static struct atom_str *to_atom(const char *s, unsigned short len)
531{
532 unsigned int hc = hc_str(s, len) % atom_table_sz;
533 struct atom_str *c;
534
535 for (c = atom_table[hc]; c; c = c->next_atom)
536 if (c->str_len == len && !strncmp(s, c->str_dat, len))
537 return c;
538
539 c = mem_pool_alloc(&fi_mem_pool, sizeof(struct atom_str) + len + 1);
540 c->str_len = len;
541 memcpy(c->str_dat, s, len);
542 c->str_dat[len] = 0;
543 c->next_atom = atom_table[hc];
544 atom_table[hc] = c;
545 atom_cnt++;
546 return c;
547}
548
549static struct branch *lookup_branch(const char *name)
550{
551 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
552 struct branch *b;
553
554 for (b = branch_table[hc]; b; b = b->table_next_branch)
555 if (!strcmp(name, b->name))
556 return b;
557 return NULL;
558}
559
560static struct branch *new_branch(const char *name)
561{
562 unsigned int hc = hc_str(name, strlen(name)) % branch_table_sz;
563 struct branch *b = lookup_branch(name);
564
565 if (b)
566 die("Invalid attempt to create duplicate branch: %s", name);
567 if (check_refname_format(name, REFNAME_ALLOW_ONELEVEL))
568 die("Branch name doesn't conform to GIT standards: %s", name);
569
570 b = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct branch));
571 b->name = pool_strdup(name);
572 b->table_next_branch = branch_table[hc];
573 b->branch_tree.versions[0].mode = S_IFDIR;
574 b->branch_tree.versions[1].mode = S_IFDIR;
575 b->num_notes = 0;
576 b->active = 0;
577 b->pack_id = MAX_PACK_ID;
578 branch_table[hc] = b;
579 branch_count++;
580 return b;
581}
582
583static unsigned int hc_entries(unsigned int cnt)
584{
585 cnt = cnt & 7 ? (cnt / 8) + 1 : cnt / 8;
586 return cnt < avail_tree_table_sz ? cnt : avail_tree_table_sz - 1;
587}
588
589static struct tree_content *new_tree_content(unsigned int cnt)
590{
591 struct avail_tree_content *f, *l = NULL;
592 struct tree_content *t;
593 unsigned int hc = hc_entries(cnt);
594
595 for (f = avail_tree_table[hc]; f; l = f, f = f->next_avail)
596 if (f->entry_capacity >= cnt)
597 break;
598
599 if (f) {
600 if (l)
601 l->next_avail = f->next_avail;
602 else
603 avail_tree_table[hc] = f->next_avail;
604 } else {
605 cnt = cnt & 7 ? ((cnt / 8) + 1) * 8 : cnt;
606 f = mem_pool_alloc(&fi_mem_pool, sizeof(*t) + sizeof(t->entries[0]) * cnt);
607 f->entry_capacity = cnt;
608 }
609
610 t = (struct tree_content*)f;
611 t->entry_count = 0;
612 t->delta_depth = 0;
613 return t;
614}
615
616static void release_tree_entry(struct tree_entry *e);
617static void release_tree_content(struct tree_content *t)
618{
619 struct avail_tree_content *f = (struct avail_tree_content*)t;
620 unsigned int hc = hc_entries(f->entry_capacity);
621 f->next_avail = avail_tree_table[hc];
622 avail_tree_table[hc] = f;
623}
624
625static void release_tree_content_recursive(struct tree_content *t)
626{
627 unsigned int i;
628 for (i = 0; i < t->entry_count; i++)
629 release_tree_entry(t->entries[i]);
630 release_tree_content(t);
631}
632
633static struct tree_content *grow_tree_content(
634 struct tree_content *t,
635 int amt)
636{
637 struct tree_content *r = new_tree_content(t->entry_count + amt);
638 r->entry_count = t->entry_count;
639 r->delta_depth = t->delta_depth;
640 memcpy(r->entries,t->entries,t->entry_count*sizeof(t->entries[0]));
641 release_tree_content(t);
642 return r;
643}
644
645static struct tree_entry *new_tree_entry(void)
646{
647 struct tree_entry *e;
648
649 if (!avail_tree_entry) {
650 unsigned int n = tree_entry_alloc;
651 tree_entry_allocd += n * sizeof(struct tree_entry);
652 ALLOC_ARRAY(e, n);
653 avail_tree_entry = e;
654 while (n-- > 1) {
655 *((void**)e) = e + 1;
656 e++;
657 }
658 *((void**)e) = NULL;
659 }
660
661 e = avail_tree_entry;
662 avail_tree_entry = *((void**)e);
663 return e;
664}
665
666static void release_tree_entry(struct tree_entry *e)
667{
668 if (e->tree)
669 release_tree_content_recursive(e->tree);
670 *((void**)e) = avail_tree_entry;
671 avail_tree_entry = e;
672}
673
674static struct tree_content *dup_tree_content(struct tree_content *s)
675{
676 struct tree_content *d;
677 struct tree_entry *a, *b;
678 unsigned int i;
679
680 if (!s)
681 return NULL;
682 d = new_tree_content(s->entry_count);
683 for (i = 0; i < s->entry_count; i++) {
684 a = s->entries[i];
685 b = new_tree_entry();
686 memcpy(b, a, sizeof(*a));
687 if (a->tree && is_null_oid(&b->versions[1].oid))
688 b->tree = dup_tree_content(a->tree);
689 else
690 b->tree = NULL;
691 d->entries[i] = b;
692 }
693 d->entry_count = s->entry_count;
694 d->delta_depth = s->delta_depth;
695
696 return d;
697}
698
699static void start_packfile(void)
700{
701 struct strbuf tmp_file = STRBUF_INIT;
702 struct packed_git *p;
703 struct pack_header hdr;
704 int pack_fd;
705
706 pack_fd = odb_mkstemp(&tmp_file, "pack/tmp_pack_XXXXXX");
707 FLEX_ALLOC_STR(p, pack_name, tmp_file.buf);
708 strbuf_release(&tmp_file);
709
710 p->pack_fd = pack_fd;
711 p->do_not_close = 1;
712 pack_file = hashfd(pack_fd, p->pack_name);
713
714 hdr.hdr_signature = htonl(PACK_SIGNATURE);
715 hdr.hdr_version = htonl(2);
716 hdr.hdr_entries = 0;
717 hashwrite(pack_file, &hdr, sizeof(hdr));
718
719 pack_data = p;
720 pack_size = sizeof(hdr);
721 object_count = 0;
722
723 REALLOC_ARRAY(all_packs, pack_id + 1);
724 all_packs[pack_id] = p;
725}
726
727static const char *create_index(void)
728{
729 const char *tmpfile;
730 struct pack_idx_entry **idx, **c, **last;
731 struct object_entry *e;
732 struct object_entry_pool *o;
733
734 /* Build the table of object IDs. */
735 ALLOC_ARRAY(idx, object_count);
736 c = idx;
737 for (o = blocks; o; o = o->next_pool)
738 for (e = o->next_free; e-- != o->entries;)
739 if (pack_id == e->pack_id)
740 *c++ = &e->idx;
741 last = idx + object_count;
742 if (c != last)
743 die("internal consistency error creating the index");
744
745 tmpfile = write_idx_file(NULL, idx, object_count, &pack_idx_opts, pack_data->sha1);
746 free(idx);
747 return tmpfile;
748}
749
750static char *keep_pack(const char *curr_index_name)
751{
752 static const char *keep_msg = "fast-import";
753 struct strbuf name = STRBUF_INIT;
754 int keep_fd;
755
756 odb_pack_name(&name, pack_data->sha1, "keep");
757 keep_fd = odb_pack_keep(name.buf);
758 if (keep_fd < 0)
759 die_errno("cannot create keep file");
760 write_or_die(keep_fd, keep_msg, strlen(keep_msg));
761 if (close(keep_fd))
762 die_errno("failed to write keep file");
763
764 odb_pack_name(&name, pack_data->sha1, "pack");
765 if (finalize_object_file(pack_data->pack_name, name.buf))
766 die("cannot store pack file");
767
768 odb_pack_name(&name, pack_data->sha1, "idx");
769 if (finalize_object_file(curr_index_name, name.buf))
770 die("cannot store index file");
771 free((void *)curr_index_name);
772 return strbuf_detach(&name, NULL);
773}
774
775static void unkeep_all_packs(void)
776{
777 struct strbuf name = STRBUF_INIT;
778 int k;
779
780 for (k = 0; k < pack_id; k++) {
781 struct packed_git *p = all_packs[k];
782 odb_pack_name(&name, p->sha1, "keep");
783 unlink_or_warn(name.buf);
784 }
785 strbuf_release(&name);
786}
787
788static int loosen_small_pack(const struct packed_git *p)
789{
790 struct child_process unpack = CHILD_PROCESS_INIT;
791
792 if (lseek(p->pack_fd, 0, SEEK_SET) < 0)
793 die_errno("Failed seeking to start of '%s'", p->pack_name);
794
795 unpack.in = p->pack_fd;
796 unpack.git_cmd = 1;
797 unpack.stdout_to_stderr = 1;
798 argv_array_push(&unpack.args, "unpack-objects");
799 if (!show_stats)
800 argv_array_push(&unpack.args, "-q");
801
802 return run_command(&unpack);
803}
804
805static void end_packfile(void)
806{
807 static int running;
808
809 if (running || !pack_data)
810 return;
811
812 running = 1;
813 clear_delta_base_cache();
814 if (object_count) {
815 struct packed_git *new_p;
816 struct object_id cur_pack_oid;
817 char *idx_name;
818 int i;
819 struct branch *b;
820 struct tag *t;
821
822 close_pack_windows(pack_data);
823 finalize_hashfile(pack_file, cur_pack_oid.hash, 0);
824 fixup_pack_header_footer(pack_data->pack_fd, pack_data->sha1,
825 pack_data->pack_name, object_count,
826 cur_pack_oid.hash, pack_size);
827
828 if (object_count <= unpack_limit) {
829 if (!loosen_small_pack(pack_data)) {
830 invalidate_pack_id(pack_id);
831 goto discard_pack;
832 }
833 }
834
835 close(pack_data->pack_fd);
836 idx_name = keep_pack(create_index());
837
838 /* Register the packfile with core git's machinery. */
839 new_p = add_packed_git(idx_name, strlen(idx_name), 1);
840 if (!new_p)
841 die("core git rejected index %s", idx_name);
842 all_packs[pack_id] = new_p;
843 install_packed_git(the_repository, new_p);
844 free(idx_name);
845
846 /* Print the boundary */
847 if (pack_edges) {
848 fprintf(pack_edges, "%s:", new_p->pack_name);
849 for (i = 0; i < branch_table_sz; i++) {
850 for (b = branch_table[i]; b; b = b->table_next_branch) {
851 if (b->pack_id == pack_id)
852 fprintf(pack_edges, " %s",
853 oid_to_hex(&b->oid));
854 }
855 }
856 for (t = first_tag; t; t = t->next_tag) {
857 if (t->pack_id == pack_id)
858 fprintf(pack_edges, " %s",
859 oid_to_hex(&t->oid));
860 }
861 fputc('\n', pack_edges);
862 fflush(pack_edges);
863 }
864
865 pack_id++;
866 }
867 else {
868discard_pack:
869 close(pack_data->pack_fd);
870 unlink_or_warn(pack_data->pack_name);
871 }
872 FREE_AND_NULL(pack_data);
873 running = 0;
874
875 /* We can't carry a delta across packfiles. */
876 strbuf_release(&last_blob.data);
877 last_blob.offset = 0;
878 last_blob.depth = 0;
879}
880
881static void cycle_packfile(void)
882{
883 end_packfile();
884 start_packfile();
885}
886
887static int store_object(
888 enum object_type type,
889 struct strbuf *dat,
890 struct last_object *last,
891 struct object_id *oidout,
892 uintmax_t mark)
893{
894 void *out, *delta;
895 struct object_entry *e;
896 unsigned char hdr[96];
897 struct object_id oid;
898 unsigned long hdrlen, deltalen;
899 git_hash_ctx c;
900 git_zstream s;
901
902 hdrlen = xsnprintf((char *)hdr, sizeof(hdr), "%s %lu",
903 type_name(type), (unsigned long)dat->len) + 1;
904 the_hash_algo->init_fn(&c);
905 the_hash_algo->update_fn(&c, hdr, hdrlen);
906 the_hash_algo->update_fn(&c, dat->buf, dat->len);
907 the_hash_algo->final_fn(oid.hash, &c);
908 if (oidout)
909 oidcpy(oidout, &oid);
910
911 e = insert_object(&oid);
912 if (mark)
913 insert_mark(mark, e);
914 if (e->idx.offset) {
915 duplicate_count_by_type[type]++;
916 return 1;
917 } else if (find_sha1_pack(oid.hash,
918 get_all_packs(the_repository))) {
919 e->type = type;
920 e->pack_id = MAX_PACK_ID;
921 e->idx.offset = 1; /* just not zero! */
922 duplicate_count_by_type[type]++;
923 return 1;
924 }
925
926 if (last && last->data.len && last->data.buf && last->depth < max_depth
927 && dat->len > the_hash_algo->rawsz) {
928
929 delta_count_attempts_by_type[type]++;
930 delta = diff_delta(last->data.buf, last->data.len,
931 dat->buf, dat->len,
932 &deltalen, dat->len - the_hash_algo->rawsz);
933 } else
934 delta = NULL;
935
936 git_deflate_init(&s, pack_compression_level);
937 if (delta) {
938 s.next_in = delta;
939 s.avail_in = deltalen;
940 } else {
941 s.next_in = (void *)dat->buf;
942 s.avail_in = dat->len;
943 }
944 s.avail_out = git_deflate_bound(&s, s.avail_in);
945 s.next_out = out = xmalloc(s.avail_out);
946 while (git_deflate(&s, Z_FINISH) == Z_OK)
947 ; /* nothing */
948 git_deflate_end(&s);
949
950 /* Determine if we should auto-checkpoint. */
951 if ((max_packsize && (pack_size + 60 + s.total_out) > max_packsize)
952 || (pack_size + 60 + s.total_out) < pack_size) {
953
954 /* This new object needs to *not* have the current pack_id. */
955 e->pack_id = pack_id + 1;
956 cycle_packfile();
957
958 /* We cannot carry a delta into the new pack. */
959 if (delta) {
960 FREE_AND_NULL(delta);
961
962 git_deflate_init(&s, pack_compression_level);
963 s.next_in = (void *)dat->buf;
964 s.avail_in = dat->len;
965 s.avail_out = git_deflate_bound(&s, s.avail_in);
966 s.next_out = out = xrealloc(out, s.avail_out);
967 while (git_deflate(&s, Z_FINISH) == Z_OK)
968 ; /* nothing */
969 git_deflate_end(&s);
970 }
971 }
972
973 e->type = type;
974 e->pack_id = pack_id;
975 e->idx.offset = pack_size;
976 object_count++;
977 object_count_by_type[type]++;
978
979 crc32_begin(pack_file);
980
981 if (delta) {
982 off_t ofs = e->idx.offset - last->offset;
983 unsigned pos = sizeof(hdr) - 1;
984
985 delta_count_by_type[type]++;
986 e->depth = last->depth + 1;
987
988 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
989 OBJ_OFS_DELTA, deltalen);
990 hashwrite(pack_file, hdr, hdrlen);
991 pack_size += hdrlen;
992
993 hdr[pos] = ofs & 127;
994 while (ofs >>= 7)
995 hdr[--pos] = 128 | (--ofs & 127);
996 hashwrite(pack_file, hdr + pos, sizeof(hdr) - pos);
997 pack_size += sizeof(hdr) - pos;
998 } else {
999 e->depth = 0;
1000 hdrlen = encode_in_pack_object_header(hdr, sizeof(hdr),
1001 type, dat->len);
1002 hashwrite(pack_file, hdr, hdrlen);
1003 pack_size += hdrlen;
1004 }
1005
1006 hashwrite(pack_file, out, s.total_out);
1007 pack_size += s.total_out;
1008
1009 e->idx.crc32 = crc32_end(pack_file);
1010
1011 free(out);
1012 free(delta);
1013 if (last) {
1014 if (last->no_swap) {
1015 last->data = *dat;
1016 } else {
1017 strbuf_swap(&last->data, dat);
1018 }
1019 last->offset = e->idx.offset;
1020 last->depth = e->depth;
1021 }
1022 return 0;
1023}
1024
1025static void truncate_pack(struct hashfile_checkpoint *checkpoint)
1026{
1027 if (hashfile_truncate(pack_file, checkpoint))
1028 die_errno("cannot truncate pack to skip duplicate");
1029 pack_size = checkpoint->offset;
1030}
1031
1032static void stream_blob(uintmax_t len, struct object_id *oidout, uintmax_t mark)
1033{
1034 size_t in_sz = 64 * 1024, out_sz = 64 * 1024;
1035 unsigned char *in_buf = xmalloc(in_sz);
1036 unsigned char *out_buf = xmalloc(out_sz);
1037 struct object_entry *e;
1038 struct object_id oid;
1039 unsigned long hdrlen;
1040 off_t offset;
1041 git_hash_ctx c;
1042 git_zstream s;
1043 struct hashfile_checkpoint checkpoint;
1044 int status = Z_OK;
1045
1046 /* Determine if we should auto-checkpoint. */
1047 if ((max_packsize && (pack_size + 60 + len) > max_packsize)
1048 || (pack_size + 60 + len) < pack_size)
1049 cycle_packfile();
1050
1051 hashfile_checkpoint(pack_file, &checkpoint);
1052 offset = checkpoint.offset;
1053
1054 hdrlen = xsnprintf((char *)out_buf, out_sz, "blob %" PRIuMAX, len) + 1;
1055
1056 the_hash_algo->init_fn(&c);
1057 the_hash_algo->update_fn(&c, out_buf, hdrlen);
1058
1059 crc32_begin(pack_file);
1060
1061 git_deflate_init(&s, pack_compression_level);
1062
1063 hdrlen = encode_in_pack_object_header(out_buf, out_sz, OBJ_BLOB, len);
1064
1065 s.next_out = out_buf + hdrlen;
1066 s.avail_out = out_sz - hdrlen;
1067
1068 while (status != Z_STREAM_END) {
1069 if (0 < len && !s.avail_in) {
1070 size_t cnt = in_sz < len ? in_sz : (size_t)len;
1071 size_t n = fread(in_buf, 1, cnt, stdin);
1072 if (!n && feof(stdin))
1073 die("EOF in data (%" PRIuMAX " bytes remaining)", len);
1074
1075 the_hash_algo->update_fn(&c, in_buf, n);
1076 s.next_in = in_buf;
1077 s.avail_in = n;
1078 len -= n;
1079 }
1080
1081 status = git_deflate(&s, len ? 0 : Z_FINISH);
1082
1083 if (!s.avail_out || status == Z_STREAM_END) {
1084 size_t n = s.next_out - out_buf;
1085 hashwrite(pack_file, out_buf, n);
1086 pack_size += n;
1087 s.next_out = out_buf;
1088 s.avail_out = out_sz;
1089 }
1090
1091 switch (status) {
1092 case Z_OK:
1093 case Z_BUF_ERROR:
1094 case Z_STREAM_END:
1095 continue;
1096 default:
1097 die("unexpected deflate failure: %d", status);
1098 }
1099 }
1100 git_deflate_end(&s);
1101 the_hash_algo->final_fn(oid.hash, &c);
1102
1103 if (oidout)
1104 oidcpy(oidout, &oid);
1105
1106 e = insert_object(&oid);
1107
1108 if (mark)
1109 insert_mark(mark, e);
1110
1111 if (e->idx.offset) {
1112 duplicate_count_by_type[OBJ_BLOB]++;
1113 truncate_pack(&checkpoint);
1114
1115 } else if (find_sha1_pack(oid.hash,
1116 get_all_packs(the_repository))) {
1117 e->type = OBJ_BLOB;
1118 e->pack_id = MAX_PACK_ID;
1119 e->idx.offset = 1; /* just not zero! */
1120 duplicate_count_by_type[OBJ_BLOB]++;
1121 truncate_pack(&checkpoint);
1122
1123 } else {
1124 e->depth = 0;
1125 e->type = OBJ_BLOB;
1126 e->pack_id = pack_id;
1127 e->idx.offset = offset;
1128 e->idx.crc32 = crc32_end(pack_file);
1129 object_count++;
1130 object_count_by_type[OBJ_BLOB]++;
1131 }
1132
1133 free(in_buf);
1134 free(out_buf);
1135}
1136
1137/* All calls must be guarded by find_object() or find_mark() to
1138 * ensure the 'struct object_entry' passed was written by this
1139 * process instance. We unpack the entry by the offset, avoiding
1140 * the need for the corresponding .idx file. This unpacking rule
1141 * works because we only use OBJ_REF_DELTA within the packfiles
1142 * created by fast-import.
1143 *
1144 * oe must not be NULL. Such an oe usually comes from giving
1145 * an unknown SHA-1 to find_object() or an undefined mark to
1146 * find_mark(). Callers must test for this condition and use
1147 * the standard read_sha1_file() when it happens.
1148 *
1149 * oe->pack_id must not be MAX_PACK_ID. Such an oe is usually from
1150 * find_mark(), where the mark was reloaded from an existing marks
1151 * file and is referencing an object that this fast-import process
1152 * instance did not write out to a packfile. Callers must test for
1153 * this condition and use read_sha1_file() instead.
1154 */
1155static void *gfi_unpack_entry(
1156 struct object_entry *oe,
1157 unsigned long *sizep)
1158{
1159 enum object_type type;
1160 struct packed_git *p = all_packs[oe->pack_id];
1161 if (p == pack_data && p->pack_size < (pack_size + the_hash_algo->rawsz)) {
1162 /* The object is stored in the packfile we are writing to
1163 * and we have modified it since the last time we scanned
1164 * back to read a previously written object. If an old
1165 * window covered [p->pack_size, p->pack_size + rawsz) its
1166 * data is stale and is not valid. Closing all windows
1167 * and updating the packfile length ensures we can read
1168 * the newly written data.
1169 */
1170 close_pack_windows(p);
1171 hashflush(pack_file);
1172
1173 /* We have to offer rawsz bytes additional on the end of
1174 * the packfile as the core unpacker code assumes the
1175 * footer is present at the file end and must promise
1176 * at least rawsz bytes within any window it maps. But
1177 * we don't actually create the footer here.
1178 */
1179 p->pack_size = pack_size + the_hash_algo->rawsz;
1180 }
1181 return unpack_entry(the_repository, p, oe->idx.offset, &type, sizep);
1182}
1183
1184static const char *get_mode(const char *str, uint16_t *modep)
1185{
1186 unsigned char c;
1187 uint16_t mode = 0;
1188
1189 while ((c = *str++) != ' ') {
1190 if (c < '0' || c > '7')
1191 return NULL;
1192 mode = (mode << 3) + (c - '0');
1193 }
1194 *modep = mode;
1195 return str;
1196}
1197
1198static void load_tree(struct tree_entry *root)
1199{
1200 struct object_id *oid = &root->versions[1].oid;
1201 struct object_entry *myoe;
1202 struct tree_content *t;
1203 unsigned long size;
1204 char *buf;
1205 const char *c;
1206
1207 root->tree = t = new_tree_content(8);
1208 if (is_null_oid(oid))
1209 return;
1210
1211 myoe = find_object(oid);
1212 if (myoe && myoe->pack_id != MAX_PACK_ID) {
1213 if (myoe->type != OBJ_TREE)
1214 die("Not a tree: %s", oid_to_hex(oid));
1215 t->delta_depth = myoe->depth;
1216 buf = gfi_unpack_entry(myoe, &size);
1217 if (!buf)
1218 die("Can't load tree %s", oid_to_hex(oid));
1219 } else {
1220 enum object_type type;
1221 buf = read_object_file(oid, &type, &size);
1222 if (!buf || type != OBJ_TREE)
1223 die("Can't load tree %s", oid_to_hex(oid));
1224 }
1225
1226 c = buf;
1227 while (c != (buf + size)) {
1228 struct tree_entry *e = new_tree_entry();
1229
1230 if (t->entry_count == t->entry_capacity)
1231 root->tree = t = grow_tree_content(t, t->entry_count);
1232 t->entries[t->entry_count++] = e;
1233
1234 e->tree = NULL;
1235 c = get_mode(c, &e->versions[1].mode);
1236 if (!c)
1237 die("Corrupt mode in %s", oid_to_hex(oid));
1238 e->versions[0].mode = e->versions[1].mode;
1239 e->name = to_atom(c, strlen(c));
1240 c += e->name->str_len + 1;
1241 hashcpy(e->versions[0].oid.hash, (unsigned char *)c);
1242 hashcpy(e->versions[1].oid.hash, (unsigned char *)c);
1243 c += GIT_SHA1_RAWSZ;
1244 }
1245 free(buf);
1246}
1247
1248static int tecmp0 (const void *_a, const void *_b)
1249{
1250 struct tree_entry *a = *((struct tree_entry**)_a);
1251 struct tree_entry *b = *((struct tree_entry**)_b);
1252 return base_name_compare(
1253 a->name->str_dat, a->name->str_len, a->versions[0].mode,
1254 b->name->str_dat, b->name->str_len, b->versions[0].mode);
1255}
1256
1257static int tecmp1 (const void *_a, const void *_b)
1258{
1259 struct tree_entry *a = *((struct tree_entry**)_a);
1260 struct tree_entry *b = *((struct tree_entry**)_b);
1261 return base_name_compare(
1262 a->name->str_dat, a->name->str_len, a->versions[1].mode,
1263 b->name->str_dat, b->name->str_len, b->versions[1].mode);
1264}
1265
1266static void mktree(struct tree_content *t, int v, struct strbuf *b)
1267{
1268 size_t maxlen = 0;
1269 unsigned int i;
1270
1271 if (!v)
1272 QSORT(t->entries, t->entry_count, tecmp0);
1273 else
1274 QSORT(t->entries, t->entry_count, tecmp1);
1275
1276 for (i = 0; i < t->entry_count; i++) {
1277 if (t->entries[i]->versions[v].mode)
1278 maxlen += t->entries[i]->name->str_len + 34;
1279 }
1280
1281 strbuf_reset(b);
1282 strbuf_grow(b, maxlen);
1283 for (i = 0; i < t->entry_count; i++) {
1284 struct tree_entry *e = t->entries[i];
1285 if (!e->versions[v].mode)
1286 continue;
1287 strbuf_addf(b, "%o %s%c",
1288 (unsigned int)(e->versions[v].mode & ~NO_DELTA),
1289 e->name->str_dat, '\0');
1290 strbuf_add(b, e->versions[v].oid.hash, GIT_SHA1_RAWSZ);
1291 }
1292}
1293
1294static void store_tree(struct tree_entry *root)
1295{
1296 struct tree_content *t;
1297 unsigned int i, j, del;
1298 struct last_object lo = { STRBUF_INIT, 0, 0, /* no_swap */ 1 };
1299 struct object_entry *le = NULL;
1300
1301 if (!is_null_oid(&root->versions[1].oid))
1302 return;
1303
1304 if (!root->tree)
1305 load_tree(root);
1306 t = root->tree;
1307
1308 for (i = 0; i < t->entry_count; i++) {
1309 if (t->entries[i]->tree)
1310 store_tree(t->entries[i]);
1311 }
1312
1313 if (!(root->versions[0].mode & NO_DELTA))
1314 le = find_object(&root->versions[0].oid);
1315 if (S_ISDIR(root->versions[0].mode) && le && le->pack_id == pack_id) {
1316 mktree(t, 0, &old_tree);
1317 lo.data = old_tree;
1318 lo.offset = le->idx.offset;
1319 lo.depth = t->delta_depth;
1320 }
1321
1322 mktree(t, 1, &new_tree);
1323 store_object(OBJ_TREE, &new_tree, &lo, &root->versions[1].oid, 0);
1324
1325 t->delta_depth = lo.depth;
1326 for (i = 0, j = 0, del = 0; i < t->entry_count; i++) {
1327 struct tree_entry *e = t->entries[i];
1328 if (e->versions[1].mode) {
1329 e->versions[0].mode = e->versions[1].mode;
1330 oidcpy(&e->versions[0].oid, &e->versions[1].oid);
1331 t->entries[j++] = e;
1332 } else {
1333 release_tree_entry(e);
1334 del++;
1335 }
1336 }
1337 t->entry_count -= del;
1338}
1339
1340static void tree_content_replace(
1341 struct tree_entry *root,
1342 const struct object_id *oid,
1343 const uint16_t mode,
1344 struct tree_content *newtree)
1345{
1346 if (!S_ISDIR(mode))
1347 die("Root cannot be a non-directory");
1348 oidclr(&root->versions[0].oid);
1349 oidcpy(&root->versions[1].oid, oid);
1350 if (root->tree)
1351 release_tree_content_recursive(root->tree);
1352 root->tree = newtree;
1353}
1354
1355static int tree_content_set(
1356 struct tree_entry *root,
1357 const char *p,
1358 const struct object_id *oid,
1359 const uint16_t mode,
1360 struct tree_content *subtree)
1361{
1362 struct tree_content *t;
1363 const char *slash1;
1364 unsigned int i, n;
1365 struct tree_entry *e;
1366
1367 slash1 = strchrnul(p, '/');
1368 n = slash1 - p;
1369 if (!n)
1370 die("Empty path component found in input");
1371 if (!*slash1 && !S_ISDIR(mode) && subtree)
1372 die("Non-directories cannot have subtrees");
1373
1374 if (!root->tree)
1375 load_tree(root);
1376 t = root->tree;
1377 for (i = 0; i < t->entry_count; i++) {
1378 e = t->entries[i];
1379 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1380 if (!*slash1) {
1381 if (!S_ISDIR(mode)
1382 && e->versions[1].mode == mode
1383 && oideq(&e->versions[1].oid, oid))
1384 return 0;
1385 e->versions[1].mode = mode;
1386 oidcpy(&e->versions[1].oid, oid);
1387 if (e->tree)
1388 release_tree_content_recursive(e->tree);
1389 e->tree = subtree;
1390
1391 /*
1392 * We need to leave e->versions[0].sha1 alone
1393 * to avoid modifying the preimage tree used
1394 * when writing out the parent directory.
1395 * But after replacing the subdir with a
1396 * completely different one, it's not a good
1397 * delta base any more, and besides, we've
1398 * thrown away the tree entries needed to
1399 * make a delta against it.
1400 *
1401 * So let's just explicitly disable deltas
1402 * for the subtree.
1403 */
1404 if (S_ISDIR(e->versions[0].mode))
1405 e->versions[0].mode |= NO_DELTA;
1406
1407 oidclr(&root->versions[1].oid);
1408 return 1;
1409 }
1410 if (!S_ISDIR(e->versions[1].mode)) {
1411 e->tree = new_tree_content(8);
1412 e->versions[1].mode = S_IFDIR;
1413 }
1414 if (!e->tree)
1415 load_tree(e);
1416 if (tree_content_set(e, slash1 + 1, oid, mode, subtree)) {
1417 oidclr(&root->versions[1].oid);
1418 return 1;
1419 }
1420 return 0;
1421 }
1422 }
1423
1424 if (t->entry_count == t->entry_capacity)
1425 root->tree = t = grow_tree_content(t, t->entry_count);
1426 e = new_tree_entry();
1427 e->name = to_atom(p, n);
1428 e->versions[0].mode = 0;
1429 oidclr(&e->versions[0].oid);
1430 t->entries[t->entry_count++] = e;
1431 if (*slash1) {
1432 e->tree = new_tree_content(8);
1433 e->versions[1].mode = S_IFDIR;
1434 tree_content_set(e, slash1 + 1, oid, mode, subtree);
1435 } else {
1436 e->tree = subtree;
1437 e->versions[1].mode = mode;
1438 oidcpy(&e->versions[1].oid, oid);
1439 }
1440 oidclr(&root->versions[1].oid);
1441 return 1;
1442}
1443
1444static int tree_content_remove(
1445 struct tree_entry *root,
1446 const char *p,
1447 struct tree_entry *backup_leaf,
1448 int allow_root)
1449{
1450 struct tree_content *t;
1451 const char *slash1;
1452 unsigned int i, n;
1453 struct tree_entry *e;
1454
1455 slash1 = strchrnul(p, '/');
1456 n = slash1 - p;
1457
1458 if (!root->tree)
1459 load_tree(root);
1460
1461 if (!*p && allow_root) {
1462 e = root;
1463 goto del_entry;
1464 }
1465
1466 t = root->tree;
1467 for (i = 0; i < t->entry_count; i++) {
1468 e = t->entries[i];
1469 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1470 if (*slash1 && !S_ISDIR(e->versions[1].mode))
1471 /*
1472 * If p names a file in some subdirectory, and a
1473 * file or symlink matching the name of the
1474 * parent directory of p exists, then p cannot
1475 * exist and need not be deleted.
1476 */
1477 return 1;
1478 if (!*slash1 || !S_ISDIR(e->versions[1].mode))
1479 goto del_entry;
1480 if (!e->tree)
1481 load_tree(e);
1482 if (tree_content_remove(e, slash1 + 1, backup_leaf, 0)) {
1483 for (n = 0; n < e->tree->entry_count; n++) {
1484 if (e->tree->entries[n]->versions[1].mode) {
1485 oidclr(&root->versions[1].oid);
1486 return 1;
1487 }
1488 }
1489 backup_leaf = NULL;
1490 goto del_entry;
1491 }
1492 return 0;
1493 }
1494 }
1495 return 0;
1496
1497del_entry:
1498 if (backup_leaf)
1499 memcpy(backup_leaf, e, sizeof(*backup_leaf));
1500 else if (e->tree)
1501 release_tree_content_recursive(e->tree);
1502 e->tree = NULL;
1503 e->versions[1].mode = 0;
1504 oidclr(&e->versions[1].oid);
1505 oidclr(&root->versions[1].oid);
1506 return 1;
1507}
1508
1509static int tree_content_get(
1510 struct tree_entry *root,
1511 const char *p,
1512 struct tree_entry *leaf,
1513 int allow_root)
1514{
1515 struct tree_content *t;
1516 const char *slash1;
1517 unsigned int i, n;
1518 struct tree_entry *e;
1519
1520 slash1 = strchrnul(p, '/');
1521 n = slash1 - p;
1522 if (!n && !allow_root)
1523 die("Empty path component found in input");
1524
1525 if (!root->tree)
1526 load_tree(root);
1527
1528 if (!n) {
1529 e = root;
1530 goto found_entry;
1531 }
1532
1533 t = root->tree;
1534 for (i = 0; i < t->entry_count; i++) {
1535 e = t->entries[i];
1536 if (e->name->str_len == n && !fspathncmp(p, e->name->str_dat, n)) {
1537 if (!*slash1)
1538 goto found_entry;
1539 if (!S_ISDIR(e->versions[1].mode))
1540 return 0;
1541 if (!e->tree)
1542 load_tree(e);
1543 return tree_content_get(e, slash1 + 1, leaf, 0);
1544 }
1545 }
1546 return 0;
1547
1548found_entry:
1549 memcpy(leaf, e, sizeof(*leaf));
1550 if (e->tree && is_null_oid(&e->versions[1].oid))
1551 leaf->tree = dup_tree_content(e->tree);
1552 else
1553 leaf->tree = NULL;
1554 return 1;
1555}
1556
1557static int update_branch(struct branch *b)
1558{
1559 static const char *msg = "fast-import";
1560 struct ref_transaction *transaction;
1561 struct object_id old_oid;
1562 struct strbuf err = STRBUF_INIT;
1563
1564 if (is_null_oid(&b->oid)) {
1565 if (b->delete)
1566 delete_ref(NULL, b->name, NULL, 0);
1567 return 0;
1568 }
1569 if (read_ref(b->name, &old_oid))
1570 oidclr(&old_oid);
1571 if (!force_update && !is_null_oid(&old_oid)) {
1572 struct commit *old_cmit, *new_cmit;
1573
1574 old_cmit = lookup_commit_reference_gently(the_repository,
1575 &old_oid, 0);
1576 new_cmit = lookup_commit_reference_gently(the_repository,
1577 &b->oid, 0);
1578 if (!old_cmit || !new_cmit)
1579 return error("Branch %s is missing commits.", b->name);
1580
1581 if (!in_merge_bases(old_cmit, new_cmit)) {
1582 warning("Not updating %s"
1583 " (new tip %s does not contain %s)",
1584 b->name, oid_to_hex(&b->oid),
1585 oid_to_hex(&old_oid));
1586 return -1;
1587 }
1588 }
1589 transaction = ref_transaction_begin(&err);
1590 if (!transaction ||
1591 ref_transaction_update(transaction, b->name, &b->oid, &old_oid,
1592 0, msg, &err) ||
1593 ref_transaction_commit(transaction, &err)) {
1594 ref_transaction_free(transaction);
1595 error("%s", err.buf);
1596 strbuf_release(&err);
1597 return -1;
1598 }
1599 ref_transaction_free(transaction);
1600 strbuf_release(&err);
1601 return 0;
1602}
1603
1604static void dump_branches(void)
1605{
1606 unsigned int i;
1607 struct branch *b;
1608
1609 for (i = 0; i < branch_table_sz; i++) {
1610 for (b = branch_table[i]; b; b = b->table_next_branch)
1611 failure |= update_branch(b);
1612 }
1613}
1614
1615static void dump_tags(void)
1616{
1617 static const char *msg = "fast-import";
1618 struct tag *t;
1619 struct strbuf ref_name = STRBUF_INIT;
1620 struct strbuf err = STRBUF_INIT;
1621 struct ref_transaction *transaction;
1622
1623 transaction = ref_transaction_begin(&err);
1624 if (!transaction) {
1625 failure |= error("%s", err.buf);
1626 goto cleanup;
1627 }
1628 for (t = first_tag; t; t = t->next_tag) {
1629 strbuf_reset(&ref_name);
1630 strbuf_addf(&ref_name, "refs/tags/%s", t->name);
1631
1632 if (ref_transaction_update(transaction, ref_name.buf,
1633 &t->oid, NULL, 0, msg, &err)) {
1634 failure |= error("%s", err.buf);
1635 goto cleanup;
1636 }
1637 }
1638 if (ref_transaction_commit(transaction, &err))
1639 failure |= error("%s", err.buf);
1640
1641 cleanup:
1642 ref_transaction_free(transaction);
1643 strbuf_release(&ref_name);
1644 strbuf_release(&err);
1645}
1646
1647static void dump_marks_helper(FILE *f,
1648 uintmax_t base,
1649 struct mark_set *m)
1650{
1651 uintmax_t k;
1652 if (m->shift) {
1653 for (k = 0; k < 1024; k++) {
1654 if (m->data.sets[k])
1655 dump_marks_helper(f, base + (k << m->shift),
1656 m->data.sets[k]);
1657 }
1658 } else {
1659 for (k = 0; k < 1024; k++) {
1660 if (m->data.marked[k])
1661 fprintf(f, ":%" PRIuMAX " %s\n", base + k,
1662 oid_to_hex(&m->data.marked[k]->idx.oid));
1663 }
1664 }
1665}
1666
1667static void dump_marks(void)
1668{
1669 struct lock_file mark_lock = LOCK_INIT;
1670 FILE *f;
1671
1672 if (!export_marks_file || (import_marks_file && !import_marks_file_done))
1673 return;
1674
1675 if (hold_lock_file_for_update(&mark_lock, export_marks_file, 0) < 0) {
1676 failure |= error_errno("Unable to write marks file %s",
1677 export_marks_file);
1678 return;
1679 }
1680
1681 f = fdopen_lock_file(&mark_lock, "w");
1682 if (!f) {
1683 int saved_errno = errno;
1684 rollback_lock_file(&mark_lock);
1685 failure |= error("Unable to write marks file %s: %s",
1686 export_marks_file, strerror(saved_errno));
1687 return;
1688 }
1689
1690 dump_marks_helper(f, 0, marks);
1691 if (commit_lock_file(&mark_lock)) {
1692 failure |= error_errno("Unable to write file %s",
1693 export_marks_file);
1694 return;
1695 }
1696}
1697
1698static void read_marks(void)
1699{
1700 char line[512];
1701 FILE *f = fopen(import_marks_file, "r");
1702 if (f)
1703 ;
1704 else if (import_marks_file_ignore_missing && errno == ENOENT)
1705 goto done; /* Marks file does not exist */
1706 else
1707 die_errno("cannot read '%s'", import_marks_file);
1708 while (fgets(line, sizeof(line), f)) {
1709 uintmax_t mark;
1710 char *end;
1711 struct object_id oid;
1712 struct object_entry *e;
1713
1714 end = strchr(line, '\n');
1715 if (line[0] != ':' || !end)
1716 die("corrupt mark line: %s", line);
1717 *end = 0;
1718 mark = strtoumax(line + 1, &end, 10);
1719 if (!mark || end == line + 1
1720 || *end != ' ' || get_oid_hex(end + 1, &oid))
1721 die("corrupt mark line: %s", line);
1722 e = find_object(&oid);
1723 if (!e) {
1724 enum object_type type = oid_object_info(the_repository,
1725 &oid, NULL);
1726 if (type < 0)
1727 die("object not found: %s", oid_to_hex(&oid));
1728 e = insert_object(&oid);
1729 e->type = type;
1730 e->pack_id = MAX_PACK_ID;
1731 e->idx.offset = 1; /* just not zero! */
1732 }
1733 insert_mark(mark, e);
1734 }
1735 fclose(f);
1736done:
1737 import_marks_file_done = 1;
1738}
1739
1740
1741static int read_next_command(void)
1742{
1743 static int stdin_eof = 0;
1744
1745 if (stdin_eof) {
1746 unread_command_buf = 0;
1747 return EOF;
1748 }
1749
1750 for (;;) {
1751 const char *p;
1752
1753 if (unread_command_buf) {
1754 unread_command_buf = 0;
1755 } else {
1756 struct recent_command *rc;
1757
1758 strbuf_detach(&command_buf, NULL);
1759 stdin_eof = strbuf_getline_lf(&command_buf, stdin);
1760 if (stdin_eof)
1761 return EOF;
1762
1763 if (!seen_data_command
1764 && !starts_with(command_buf.buf, "feature ")
1765 && !starts_with(command_buf.buf, "option ")) {
1766 parse_argv();
1767 }
1768
1769 rc = rc_free;
1770 if (rc)
1771 rc_free = rc->next;
1772 else {
1773 rc = cmd_hist.next;
1774 cmd_hist.next = rc->next;
1775 cmd_hist.next->prev = &cmd_hist;
1776 free(rc->buf);
1777 }
1778
1779 rc->buf = command_buf.buf;
1780 rc->prev = cmd_tail;
1781 rc->next = cmd_hist.prev;
1782 rc->prev->next = rc;
1783 cmd_tail = rc;
1784 }
1785 if (skip_prefix(command_buf.buf, "get-mark ", &p)) {
1786 parse_get_mark(p);
1787 continue;
1788 }
1789 if (skip_prefix(command_buf.buf, "cat-blob ", &p)) {
1790 parse_cat_blob(p);
1791 continue;
1792 }
1793 if (command_buf.buf[0] == '#')
1794 continue;
1795 return 0;
1796 }
1797}
1798
1799static 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
1806static 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
1817static 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
1824static 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
1872static 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
1897static 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
1941static 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
1962static 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
1970static 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
2002static 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
2014static 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
2022static 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
2038static 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 char realpath[60];
2050
2051 if (!root->tree)
2052 load_tree(root);
2053 t = root->tree;
2054
2055 for (i = 0; t && i < t->entry_count; i++) {
2056 e = t->entries[i];
2057 tmp_hex_oid_len = hex_oid_len + e->name->str_len;
2058 tmp_fullpath_len = fullpath_len;
2059
2060 /*
2061 * We're interested in EITHER existing note entries (entries
2062 * with exactly 40 hex chars in path, not including directory
2063 * separators), OR directory entries that may contain note
2064 * entries (with < 40 hex chars in path).
2065 * Also, each path component in a note entry must be a multiple
2066 * of 2 chars.
2067 */
2068 if (!e->versions[1].mode ||
2069 tmp_hex_oid_len > GIT_SHA1_HEXSZ ||
2070 e->name->str_len % 2)
2071 continue;
2072
2073 /* This _may_ be a note entry, or a subdir containing notes */
2074 memcpy(hex_oid + hex_oid_len, e->name->str_dat,
2075 e->name->str_len);
2076 if (tmp_fullpath_len)
2077 fullpath[tmp_fullpath_len++] = '/';
2078 memcpy(fullpath + tmp_fullpath_len, e->name->str_dat,
2079 e->name->str_len);
2080 tmp_fullpath_len += e->name->str_len;
2081 fullpath[tmp_fullpath_len] = '\0';
2082
2083 if (tmp_hex_oid_len == GIT_SHA1_HEXSZ && !get_oid_hex(hex_oid, &oid)) {
2084 /* This is a note entry */
2085 if (fanout == 0xff) {
2086 /* Counting mode, no rename */
2087 num_notes++;
2088 continue;
2089 }
2090 construct_path_with_fanout(hex_oid, fanout, realpath);
2091 if (!strcmp(fullpath, realpath)) {
2092 /* Note entry is in correct location */
2093 num_notes++;
2094 continue;
2095 }
2096
2097 /* Rename fullpath to realpath */
2098 if (!tree_content_remove(orig_root, fullpath, &leaf, 0))
2099 die("Failed to remove path %s", fullpath);
2100 tree_content_set(orig_root, realpath,
2101 &leaf.versions[1].oid,
2102 leaf.versions[1].mode,
2103 leaf.tree);
2104 } else if (S_ISDIR(e->versions[1].mode)) {
2105 /* This is a subdir that may contain note entries */
2106 num_notes += do_change_note_fanout(orig_root, e,
2107 hex_oid, tmp_hex_oid_len,
2108 fullpath, tmp_fullpath_len, fanout);
2109 }
2110
2111 /* The above may have reallocated the current tree_content */
2112 t = root->tree;
2113 }
2114 return num_notes;
2115}
2116
2117static uintmax_t change_note_fanout(struct tree_entry *root,
2118 unsigned char fanout)
2119{
2120 /*
2121 * The size of path is due to one slash between every two hex digits,
2122 * plus the terminating NUL. Note that there is no slash at the end, so
2123 * the number of slashes is one less than half the number of hex
2124 * characters.
2125 */
2126 char hex_oid[GIT_MAX_HEXSZ], path[GIT_MAX_HEXSZ + (GIT_MAX_HEXSZ / 2) - 1 + 1];
2127 return do_change_note_fanout(root, root, hex_oid, 0, path, 0, fanout);
2128}
2129
2130/*
2131 * Given a pointer into a string, parse a mark reference:
2132 *
2133 * idnum ::= ':' bigint;
2134 *
2135 * Return the first character after the value in *endptr.
2136 *
2137 * Complain if the following character is not what is expected,
2138 * either a space or end of the string.
2139 */
2140static uintmax_t parse_mark_ref(const char *p, char **endptr)
2141{
2142 uintmax_t mark;
2143
2144 assert(*p == ':');
2145 p++;
2146 mark = strtoumax(p, endptr, 10);
2147 if (*endptr == p)
2148 die("No value after ':' in mark: %s", command_buf.buf);
2149 return mark;
2150}
2151
2152/*
2153 * Parse the mark reference, and complain if this is not the end of
2154 * the string.
2155 */
2156static uintmax_t parse_mark_ref_eol(const char *p)
2157{
2158 char *end;
2159 uintmax_t mark;
2160
2161 mark = parse_mark_ref(p, &end);
2162 if (*end != '\0')
2163 die("Garbage after mark: %s", command_buf.buf);
2164 return mark;
2165}
2166
2167/*
2168 * Parse the mark reference, demanding a trailing space. Return a
2169 * pointer to the space.
2170 */
2171static uintmax_t parse_mark_ref_space(const char **p)
2172{
2173 uintmax_t mark;
2174 char *end;
2175
2176 mark = parse_mark_ref(*p, &end);
2177 if (*end++ != ' ')
2178 die("Missing space after mark: %s", command_buf.buf);
2179 *p = end;
2180 return mark;
2181}
2182
2183static void file_change_m(const char *p, struct branch *b)
2184{
2185 static struct strbuf uq = STRBUF_INIT;
2186 const char *endp;
2187 struct object_entry *oe;
2188 struct object_id oid;
2189 uint16_t mode, inline_data = 0;
2190
2191 p = get_mode(p, &mode);
2192 if (!p)
2193 die("Corrupt mode: %s", command_buf.buf);
2194 switch (mode) {
2195 case 0644:
2196 case 0755:
2197 mode |= S_IFREG;
2198 case S_IFREG | 0644:
2199 case S_IFREG | 0755:
2200 case S_IFLNK:
2201 case S_IFDIR:
2202 case S_IFGITLINK:
2203 /* ok */
2204 break;
2205 default:
2206 die("Corrupt mode: %s", command_buf.buf);
2207 }
2208
2209 if (*p == ':') {
2210 oe = find_mark(parse_mark_ref_space(&p));
2211 oidcpy(&oid, &oe->idx.oid);
2212 } else if (skip_prefix(p, "inline ", &p)) {
2213 inline_data = 1;
2214 oe = NULL; /* not used with inline_data, but makes gcc happy */
2215 } else {
2216 if (parse_oid_hex(p, &oid, &p))
2217 die("Invalid dataref: %s", command_buf.buf);
2218 oe = find_object(&oid);
2219 if (*p++ != ' ')
2220 die("Missing space after SHA1: %s", command_buf.buf);
2221 }
2222
2223 strbuf_reset(&uq);
2224 if (!unquote_c_style(&uq, p, &endp)) {
2225 if (*endp)
2226 die("Garbage after path in: %s", command_buf.buf);
2227 p = uq.buf;
2228 }
2229
2230 /* Git does not track empty, non-toplevel directories. */
2231 if (S_ISDIR(mode) && is_empty_tree_oid(&oid) && *p) {
2232 tree_content_remove(&b->branch_tree, p, NULL, 0);
2233 return;
2234 }
2235
2236 if (S_ISGITLINK(mode)) {
2237 if (inline_data)
2238 die("Git links cannot be specified 'inline': %s",
2239 command_buf.buf);
2240 else if (oe) {
2241 if (oe->type != OBJ_COMMIT)
2242 die("Not a commit (actually a %s): %s",
2243 type_name(oe->type), command_buf.buf);
2244 }
2245 /*
2246 * Accept the sha1 without checking; it expected to be in
2247 * another repository.
2248 */
2249 } else if (inline_data) {
2250 if (S_ISDIR(mode))
2251 die("Directories cannot be specified 'inline': %s",
2252 command_buf.buf);
2253 if (p != uq.buf) {
2254 strbuf_addstr(&uq, p);
2255 p = uq.buf;
2256 }
2257 read_next_command();
2258 parse_and_store_blob(&last_blob, &oid, 0);
2259 } else {
2260 enum object_type expected = S_ISDIR(mode) ?
2261 OBJ_TREE: OBJ_BLOB;
2262 enum object_type type = oe ? oe->type :
2263 oid_object_info(the_repository, &oid,
2264 NULL);
2265 if (type < 0)
2266 die("%s not found: %s",
2267 S_ISDIR(mode) ? "Tree" : "Blob",
2268 command_buf.buf);
2269 if (type != expected)
2270 die("Not a %s (actually a %s): %s",
2271 type_name(expected), type_name(type),
2272 command_buf.buf);
2273 }
2274
2275 if (!*p) {
2276 tree_content_replace(&b->branch_tree, &oid, mode, NULL);
2277 return;
2278 }
2279 tree_content_set(&b->branch_tree, p, &oid, mode, NULL);
2280}
2281
2282static void file_change_d(const char *p, struct branch *b)
2283{
2284 static struct strbuf uq = STRBUF_INIT;
2285 const char *endp;
2286
2287 strbuf_reset(&uq);
2288 if (!unquote_c_style(&uq, p, &endp)) {
2289 if (*endp)
2290 die("Garbage after path in: %s", command_buf.buf);
2291 p = uq.buf;
2292 }
2293 tree_content_remove(&b->branch_tree, p, NULL, 1);
2294}
2295
2296static void file_change_cr(const char *s, struct branch *b, int rename)
2297{
2298 const char *d;
2299 static struct strbuf s_uq = STRBUF_INIT;
2300 static struct strbuf d_uq = STRBUF_INIT;
2301 const char *endp;
2302 struct tree_entry leaf;
2303
2304 strbuf_reset(&s_uq);
2305 if (!unquote_c_style(&s_uq, s, &endp)) {
2306 if (*endp != ' ')
2307 die("Missing space after source: %s", command_buf.buf);
2308 } else {
2309 endp = strchr(s, ' ');
2310 if (!endp)
2311 die("Missing space after source: %s", command_buf.buf);
2312 strbuf_add(&s_uq, s, endp - s);
2313 }
2314 s = s_uq.buf;
2315
2316 endp++;
2317 if (!*endp)
2318 die("Missing dest: %s", command_buf.buf);
2319
2320 d = endp;
2321 strbuf_reset(&d_uq);
2322 if (!unquote_c_style(&d_uq, d, &endp)) {
2323 if (*endp)
2324 die("Garbage after dest in: %s", command_buf.buf);
2325 d = d_uq.buf;
2326 }
2327
2328 memset(&leaf, 0, sizeof(leaf));
2329 if (rename)
2330 tree_content_remove(&b->branch_tree, s, &leaf, 1);
2331 else
2332 tree_content_get(&b->branch_tree, s, &leaf, 1);
2333 if (!leaf.versions[1].mode)
2334 die("Path %s not in branch", s);
2335 if (!*d) { /* C "path/to/subdir" "" */
2336 tree_content_replace(&b->branch_tree,
2337 &leaf.versions[1].oid,
2338 leaf.versions[1].mode,
2339 leaf.tree);
2340 return;
2341 }
2342 tree_content_set(&b->branch_tree, d,
2343 &leaf.versions[1].oid,
2344 leaf.versions[1].mode,
2345 leaf.tree);
2346}
2347
2348static void note_change_n(const char *p, struct branch *b, unsigned char *old_fanout)
2349{
2350 static struct strbuf uq = STRBUF_INIT;
2351 struct object_entry *oe;
2352 struct branch *s;
2353 struct object_id oid, commit_oid;
2354 char path[60];
2355 uint16_t inline_data = 0;
2356 unsigned char new_fanout;
2357
2358 /*
2359 * When loading a branch, we don't traverse its tree to count the real
2360 * number of notes (too expensive to do this for all non-note refs).
2361 * This means that recently loaded notes refs might incorrectly have
2362 * b->num_notes == 0, and consequently, old_fanout might be wrong.
2363 *
2364 * Fix this by traversing the tree and counting the number of notes
2365 * when b->num_notes == 0. If the notes tree is truly empty, the
2366 * calculation should not take long.
2367 */
2368 if (b->num_notes == 0 && *old_fanout == 0) {
2369 /* Invoke change_note_fanout() in "counting mode". */
2370 b->num_notes = change_note_fanout(&b->branch_tree, 0xff);
2371 *old_fanout = convert_num_notes_to_fanout(b->num_notes);
2372 }
2373
2374 /* Now parse the notemodify command. */
2375 /* <dataref> or 'inline' */
2376 if (*p == ':') {
2377 oe = find_mark(parse_mark_ref_space(&p));
2378 oidcpy(&oid, &oe->idx.oid);
2379 } else if (skip_prefix(p, "inline ", &p)) {
2380 inline_data = 1;
2381 oe = NULL; /* not used with inline_data, but makes gcc happy */
2382 } else {
2383 if (parse_oid_hex(p, &oid, &p))
2384 die("Invalid dataref: %s", command_buf.buf);
2385 oe = find_object(&oid);
2386 if (*p++ != ' ')
2387 die("Missing space after SHA1: %s", command_buf.buf);
2388 }
2389
2390 /* <commit-ish> */
2391 s = lookup_branch(p);
2392 if (s) {
2393 if (is_null_oid(&s->oid))
2394 die("Can't add a note on empty branch.");
2395 oidcpy(&commit_oid, &s->oid);
2396 } else if (*p == ':') {
2397 uintmax_t commit_mark = parse_mark_ref_eol(p);
2398 struct object_entry *commit_oe = find_mark(commit_mark);
2399 if (commit_oe->type != OBJ_COMMIT)
2400 die("Mark :%" PRIuMAX " not a commit", commit_mark);
2401 oidcpy(&commit_oid, &commit_oe->idx.oid);
2402 } else if (!get_oid(p, &commit_oid)) {
2403 unsigned long size;
2404 char *buf = read_object_with_reference(&commit_oid,
2405 commit_type, &size,
2406 &commit_oid);
2407 if (!buf || size < 46)
2408 die("Not a valid commit: %s", p);
2409 free(buf);
2410 } else
2411 die("Invalid ref name or SHA1 expression: %s", p);
2412
2413 if (inline_data) {
2414 if (p != uq.buf) {
2415 strbuf_addstr(&uq, p);
2416 p = uq.buf;
2417 }
2418 read_next_command();
2419 parse_and_store_blob(&last_blob, &oid, 0);
2420 } else if (oe) {
2421 if (oe->type != OBJ_BLOB)
2422 die("Not a blob (actually a %s): %s",
2423 type_name(oe->type), command_buf.buf);
2424 } else if (!is_null_oid(&oid)) {
2425 enum object_type type = oid_object_info(the_repository, &oid,
2426 NULL);
2427 if (type < 0)
2428 die("Blob not found: %s", command_buf.buf);
2429 if (type != OBJ_BLOB)
2430 die("Not a blob (actually a %s): %s",
2431 type_name(type), command_buf.buf);
2432 }
2433
2434 construct_path_with_fanout(oid_to_hex(&commit_oid), *old_fanout, path);
2435 if (tree_content_remove(&b->branch_tree, path, NULL, 0))
2436 b->num_notes--;
2437
2438 if (is_null_oid(&oid))
2439 return; /* nothing to insert */
2440
2441 b->num_notes++;
2442 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2443 construct_path_with_fanout(oid_to_hex(&commit_oid), new_fanout, path);
2444 tree_content_set(&b->branch_tree, path, &oid, S_IFREG | 0644, NULL);
2445}
2446
2447static void file_change_deleteall(struct branch *b)
2448{
2449 release_tree_content_recursive(b->branch_tree.tree);
2450 oidclr(&b->branch_tree.versions[0].oid);
2451 oidclr(&b->branch_tree.versions[1].oid);
2452 load_tree(&b->branch_tree);
2453 b->num_notes = 0;
2454}
2455
2456static void parse_from_commit(struct branch *b, char *buf, unsigned long size)
2457{
2458 if (!buf || size < GIT_SHA1_HEXSZ + 6)
2459 die("Not a valid commit: %s", oid_to_hex(&b->oid));
2460 if (memcmp("tree ", buf, 5)
2461 || get_oid_hex(buf + 5, &b->branch_tree.versions[1].oid))
2462 die("The commit %s is corrupt", oid_to_hex(&b->oid));
2463 oidcpy(&b->branch_tree.versions[0].oid,
2464 &b->branch_tree.versions[1].oid);
2465}
2466
2467static void parse_from_existing(struct branch *b)
2468{
2469 if (is_null_oid(&b->oid)) {
2470 oidclr(&b->branch_tree.versions[0].oid);
2471 oidclr(&b->branch_tree.versions[1].oid);
2472 } else {
2473 unsigned long size;
2474 char *buf;
2475
2476 buf = read_object_with_reference(&b->oid, commit_type, &size,
2477 &b->oid);
2478 parse_from_commit(b, buf, size);
2479 free(buf);
2480 }
2481}
2482
2483static int parse_from(struct branch *b)
2484{
2485 const char *from;
2486 struct branch *s;
2487 struct object_id oid;
2488
2489 if (!skip_prefix(command_buf.buf, "from ", &from))
2490 return 0;
2491
2492 oidcpy(&oid, &b->branch_tree.versions[1].oid);
2493
2494 s = lookup_branch(from);
2495 if (b == s)
2496 die("Can't create a branch from itself: %s", b->name);
2497 else if (s) {
2498 struct object_id *t = &s->branch_tree.versions[1].oid;
2499 oidcpy(&b->oid, &s->oid);
2500 oidcpy(&b->branch_tree.versions[0].oid, t);
2501 oidcpy(&b->branch_tree.versions[1].oid, t);
2502 } else if (*from == ':') {
2503 uintmax_t idnum = parse_mark_ref_eol(from);
2504 struct object_entry *oe = find_mark(idnum);
2505 if (oe->type != OBJ_COMMIT)
2506 die("Mark :%" PRIuMAX " not a commit", idnum);
2507 if (!oideq(&b->oid, &oe->idx.oid)) {
2508 oidcpy(&b->oid, &oe->idx.oid);
2509 if (oe->pack_id != MAX_PACK_ID) {
2510 unsigned long size;
2511 char *buf = gfi_unpack_entry(oe, &size);
2512 parse_from_commit(b, buf, size);
2513 free(buf);
2514 } else
2515 parse_from_existing(b);
2516 }
2517 } else if (!get_oid(from, &b->oid)) {
2518 parse_from_existing(b);
2519 if (is_null_oid(&b->oid))
2520 b->delete = 1;
2521 }
2522 else
2523 die("Invalid ref name or SHA1 expression: %s", from);
2524
2525 if (b->branch_tree.tree && !oideq(&oid, &b->branch_tree.versions[1].oid)) {
2526 release_tree_content_recursive(b->branch_tree.tree);
2527 b->branch_tree.tree = NULL;
2528 }
2529
2530 read_next_command();
2531 return 1;
2532}
2533
2534static struct hash_list *parse_merge(unsigned int *count)
2535{
2536 struct hash_list *list = NULL, **tail = &list, *n;
2537 const char *from;
2538 struct branch *s;
2539
2540 *count = 0;
2541 while (skip_prefix(command_buf.buf, "merge ", &from)) {
2542 n = xmalloc(sizeof(*n));
2543 s = lookup_branch(from);
2544 if (s)
2545 oidcpy(&n->oid, &s->oid);
2546 else if (*from == ':') {
2547 uintmax_t idnum = parse_mark_ref_eol(from);
2548 struct object_entry *oe = find_mark(idnum);
2549 if (oe->type != OBJ_COMMIT)
2550 die("Mark :%" PRIuMAX " not a commit", idnum);
2551 oidcpy(&n->oid, &oe->idx.oid);
2552 } else if (!get_oid(from, &n->oid)) {
2553 unsigned long size;
2554 char *buf = read_object_with_reference(&n->oid,
2555 commit_type,
2556 &size, &n->oid);
2557 if (!buf || size < 46)
2558 die("Not a valid commit: %s", from);
2559 free(buf);
2560 } else
2561 die("Invalid ref name or SHA1 expression: %s", from);
2562
2563 n->next = NULL;
2564 *tail = n;
2565 tail = &n->next;
2566
2567 (*count)++;
2568 read_next_command();
2569 }
2570 return list;
2571}
2572
2573static void parse_new_commit(const char *arg)
2574{
2575 static struct strbuf msg = STRBUF_INIT;
2576 struct branch *b;
2577 char *author = NULL;
2578 char *committer = NULL;
2579 struct hash_list *merge_list = NULL;
2580 unsigned int merge_count;
2581 unsigned char prev_fanout, new_fanout;
2582 const char *v;
2583
2584 b = lookup_branch(arg);
2585 if (!b)
2586 b = new_branch(arg);
2587
2588 read_next_command();
2589 parse_mark();
2590 parse_original_identifier();
2591 if (skip_prefix(command_buf.buf, "author ", &v)) {
2592 author = parse_ident(v);
2593 read_next_command();
2594 }
2595 if (skip_prefix(command_buf.buf, "committer ", &v)) {
2596 committer = parse_ident(v);
2597 read_next_command();
2598 }
2599 if (!committer)
2600 die("Expected committer but didn't get one");
2601 parse_data(&msg, 0, NULL);
2602 read_next_command();
2603 parse_from(b);
2604 merge_list = parse_merge(&merge_count);
2605
2606 /* ensure the branch is active/loaded */
2607 if (!b->branch_tree.tree || !max_active_branches) {
2608 unload_one_branch();
2609 load_branch(b);
2610 }
2611
2612 prev_fanout = convert_num_notes_to_fanout(b->num_notes);
2613
2614 /* file_change* */
2615 while (command_buf.len > 0) {
2616 if (skip_prefix(command_buf.buf, "M ", &v))
2617 file_change_m(v, b);
2618 else if (skip_prefix(command_buf.buf, "D ", &v))
2619 file_change_d(v, b);
2620 else if (skip_prefix(command_buf.buf, "R ", &v))
2621 file_change_cr(v, b, 1);
2622 else if (skip_prefix(command_buf.buf, "C ", &v))
2623 file_change_cr(v, b, 0);
2624 else if (skip_prefix(command_buf.buf, "N ", &v))
2625 note_change_n(v, b, &prev_fanout);
2626 else if (!strcmp("deleteall", command_buf.buf))
2627 file_change_deleteall(b);
2628 else if (skip_prefix(command_buf.buf, "ls ", &v))
2629 parse_ls(v, b);
2630 else {
2631 unread_command_buf = 1;
2632 break;
2633 }
2634 if (read_next_command() == EOF)
2635 break;
2636 }
2637
2638 new_fanout = convert_num_notes_to_fanout(b->num_notes);
2639 if (new_fanout != prev_fanout)
2640 b->num_notes = change_note_fanout(&b->branch_tree, new_fanout);
2641
2642 /* build the tree and the commit */
2643 store_tree(&b->branch_tree);
2644 oidcpy(&b->branch_tree.versions[0].oid,
2645 &b->branch_tree.versions[1].oid);
2646
2647 strbuf_reset(&new_data);
2648 strbuf_addf(&new_data, "tree %s\n",
2649 oid_to_hex(&b->branch_tree.versions[1].oid));
2650 if (!is_null_oid(&b->oid))
2651 strbuf_addf(&new_data, "parent %s\n",
2652 oid_to_hex(&b->oid));
2653 while (merge_list) {
2654 struct hash_list *next = merge_list->next;
2655 strbuf_addf(&new_data, "parent %s\n",
2656 oid_to_hex(&merge_list->oid));
2657 free(merge_list);
2658 merge_list = next;
2659 }
2660 strbuf_addf(&new_data,
2661 "author %s\n"
2662 "committer %s\n"
2663 "\n",
2664 author ? author : committer, committer);
2665 strbuf_addbuf(&new_data, &msg);
2666 free(author);
2667 free(committer);
2668
2669 if (!store_object(OBJ_COMMIT, &new_data, NULL, &b->oid, next_mark))
2670 b->pack_id = pack_id;
2671 b->last_commit = object_count_by_type[OBJ_COMMIT];
2672}
2673
2674static void parse_new_tag(const char *arg)
2675{
2676 static struct strbuf msg = STRBUF_INIT;
2677 const char *from;
2678 char *tagger;
2679 struct branch *s;
2680 struct tag *t;
2681 uintmax_t from_mark = 0;
2682 struct object_id oid;
2683 enum object_type type;
2684 const char *v;
2685
2686 t = mem_pool_alloc(&fi_mem_pool, sizeof(struct tag));
2687 memset(t, 0, sizeof(struct tag));
2688 t->name = pool_strdup(arg);
2689 if (last_tag)
2690 last_tag->next_tag = t;
2691 else
2692 first_tag = t;
2693 last_tag = t;
2694 read_next_command();
2695
2696 /* from ... */
2697 if (!skip_prefix(command_buf.buf, "from ", &from))
2698 die("Expected from command, got %s", command_buf.buf);
2699 s = lookup_branch(from);
2700 if (s) {
2701 if (is_null_oid(&s->oid))
2702 die("Can't tag an empty branch.");
2703 oidcpy(&oid, &s->oid);
2704 type = OBJ_COMMIT;
2705 } else if (*from == ':') {
2706 struct object_entry *oe;
2707 from_mark = parse_mark_ref_eol(from);
2708 oe = find_mark(from_mark);
2709 type = oe->type;
2710 oidcpy(&oid, &oe->idx.oid);
2711 } else if (!get_oid(from, &oid)) {
2712 struct object_entry *oe = find_object(&oid);
2713 if (!oe) {
2714 type = oid_object_info(the_repository, &oid, NULL);
2715 if (type < 0)
2716 die("Not a valid object: %s", from);
2717 } else
2718 type = oe->type;
2719 } else
2720 die("Invalid ref name or SHA1 expression: %s", from);
2721 read_next_command();
2722
2723 /* original-oid ... */
2724 parse_original_identifier();
2725
2726 /* tagger ... */
2727 if (skip_prefix(command_buf.buf, "tagger ", &v)) {
2728 tagger = parse_ident(v);
2729 read_next_command();
2730 } else
2731 tagger = NULL;
2732
2733 /* tag payload/message */
2734 parse_data(&msg, 0, NULL);
2735
2736 /* build the tag object */
2737 strbuf_reset(&new_data);
2738
2739 strbuf_addf(&new_data,
2740 "object %s\n"
2741 "type %s\n"
2742 "tag %s\n",
2743 oid_to_hex(&oid), type_name(type), t->name);
2744 if (tagger)
2745 strbuf_addf(&new_data,
2746 "tagger %s\n", tagger);
2747 strbuf_addch(&new_data, '\n');
2748 strbuf_addbuf(&new_data, &msg);
2749 free(tagger);
2750
2751 if (store_object(OBJ_TAG, &new_data, NULL, &t->oid, 0))
2752 t->pack_id = MAX_PACK_ID;
2753 else
2754 t->pack_id = pack_id;
2755}
2756
2757static void parse_reset_branch(const char *arg)
2758{
2759 struct branch *b;
2760
2761 b = lookup_branch(arg);
2762 if (b) {
2763 oidclr(&b->oid);
2764 oidclr(&b->branch_tree.versions[0].oid);
2765 oidclr(&b->branch_tree.versions[1].oid);
2766 if (b->branch_tree.tree) {
2767 release_tree_content_recursive(b->branch_tree.tree);
2768 b->branch_tree.tree = NULL;
2769 }
2770 }
2771 else
2772 b = new_branch(arg);
2773 read_next_command();
2774 parse_from(b);
2775 if (command_buf.len > 0)
2776 unread_command_buf = 1;
2777}
2778
2779static void cat_blob_write(const char *buf, unsigned long size)
2780{
2781 if (write_in_full(cat_blob_fd, buf, size) < 0)
2782 die_errno("Write to frontend failed");
2783}
2784
2785static void cat_blob(struct object_entry *oe, struct object_id *oid)
2786{
2787 struct strbuf line = STRBUF_INIT;
2788 unsigned long size;
2789 enum object_type type = 0;
2790 char *buf;
2791
2792 if (!oe || oe->pack_id == MAX_PACK_ID) {
2793 buf = read_object_file(oid, &type, &size);
2794 } else {
2795 type = oe->type;
2796 buf = gfi_unpack_entry(oe, &size);
2797 }
2798
2799 /*
2800 * Output based on batch_one_object() from cat-file.c.
2801 */
2802 if (type <= 0) {
2803 strbuf_reset(&line);
2804 strbuf_addf(&line, "%s missing\n", oid_to_hex(oid));
2805 cat_blob_write(line.buf, line.len);
2806 strbuf_release(&line);
2807 free(buf);
2808 return;
2809 }
2810 if (!buf)
2811 die("Can't read object %s", oid_to_hex(oid));
2812 if (type != OBJ_BLOB)
2813 die("Object %s is a %s but a blob was expected.",
2814 oid_to_hex(oid), type_name(type));
2815 strbuf_reset(&line);
2816 strbuf_addf(&line, "%s %s %"PRIuMAX"\n", oid_to_hex(oid),
2817 type_name(type), (uintmax_t)size);
2818 cat_blob_write(line.buf, line.len);
2819 strbuf_release(&line);
2820 cat_blob_write(buf, size);
2821 cat_blob_write("\n", 1);
2822 if (oe && oe->pack_id == pack_id) {
2823 last_blob.offset = oe->idx.offset;
2824 strbuf_attach(&last_blob.data, buf, size, size);
2825 last_blob.depth = oe->depth;
2826 } else
2827 free(buf);
2828}
2829
2830static void parse_get_mark(const char *p)
2831{
2832 struct object_entry *oe;
2833 char output[GIT_MAX_HEXSZ + 2];
2834
2835 /* get-mark SP <object> LF */
2836 if (*p != ':')
2837 die("Not a mark: %s", p);
2838
2839 oe = find_mark(parse_mark_ref_eol(p));
2840 if (!oe)
2841 die("Unknown mark: %s", command_buf.buf);
2842
2843 xsnprintf(output, sizeof(output), "%s\n", oid_to_hex(&oe->idx.oid));
2844 cat_blob_write(output, GIT_SHA1_HEXSZ + 1);
2845}
2846
2847static void parse_cat_blob(const char *p)
2848{
2849 struct object_entry *oe;
2850 struct object_id oid;
2851
2852 /* cat-blob SP <object> LF */
2853 if (*p == ':') {
2854 oe = find_mark(parse_mark_ref_eol(p));
2855 if (!oe)
2856 die("Unknown mark: %s", command_buf.buf);
2857 oidcpy(&oid, &oe->idx.oid);
2858 } else {
2859 if (parse_oid_hex(p, &oid, &p))
2860 die("Invalid dataref: %s", command_buf.buf);
2861 if (*p)
2862 die("Garbage after SHA1: %s", command_buf.buf);
2863 oe = find_object(&oid);
2864 }
2865
2866 cat_blob(oe, &oid);
2867}
2868
2869static struct object_entry *dereference(struct object_entry *oe,
2870 struct object_id *oid)
2871{
2872 unsigned long size;
2873 char *buf = NULL;
2874 if (!oe) {
2875 enum object_type type = oid_object_info(the_repository, oid,
2876 NULL);
2877 if (type < 0)
2878 die("object not found: %s", oid_to_hex(oid));
2879 /* cache it! */
2880 oe = insert_object(oid);
2881 oe->type = type;
2882 oe->pack_id = MAX_PACK_ID;
2883 oe->idx.offset = 1;
2884 }
2885 switch (oe->type) {
2886 case OBJ_TREE: /* easy case. */
2887 return oe;
2888 case OBJ_COMMIT:
2889 case OBJ_TAG:
2890 break;
2891 default:
2892 die("Not a tree-ish: %s", command_buf.buf);
2893 }
2894
2895 if (oe->pack_id != MAX_PACK_ID) { /* in a pack being written */
2896 buf = gfi_unpack_entry(oe, &size);
2897 } else {
2898 enum object_type unused;
2899 buf = read_object_file(oid, &unused, &size);
2900 }
2901 if (!buf)
2902 die("Can't load object %s", oid_to_hex(oid));
2903
2904 /* Peel one layer. */
2905 switch (oe->type) {
2906 case OBJ_TAG:
2907 if (size < GIT_SHA1_HEXSZ + strlen("object ") ||
2908 get_oid_hex(buf + strlen("object "), oid))
2909 die("Invalid SHA1 in tag: %s", command_buf.buf);
2910 break;
2911 case OBJ_COMMIT:
2912 if (size < GIT_SHA1_HEXSZ + strlen("tree ") ||
2913 get_oid_hex(buf + strlen("tree "), oid))
2914 die("Invalid SHA1 in commit: %s", command_buf.buf);
2915 }
2916
2917 free(buf);
2918 return find_object(oid);
2919}
2920
2921static struct object_entry *parse_treeish_dataref(const char **p)
2922{
2923 struct object_id oid;
2924 struct object_entry *e;
2925
2926 if (**p == ':') { /* <mark> */
2927 e = find_mark(parse_mark_ref_space(p));
2928 if (!e)
2929 die("Unknown mark: %s", command_buf.buf);
2930 oidcpy(&oid, &e->idx.oid);
2931 } else { /* <sha1> */
2932 if (parse_oid_hex(*p, &oid, p))
2933 die("Invalid dataref: %s", command_buf.buf);
2934 e = find_object(&oid);
2935 if (*(*p)++ != ' ')
2936 die("Missing space after tree-ish: %s", command_buf.buf);
2937 }
2938
2939 while (!e || e->type != OBJ_TREE)
2940 e = dereference(e, &oid);
2941 return e;
2942}
2943
2944static void print_ls(int mode, const unsigned char *sha1, const char *path)
2945{
2946 static struct strbuf line = STRBUF_INIT;
2947
2948 /* See show_tree(). */
2949 const char *type =
2950 S_ISGITLINK(mode) ? commit_type :
2951 S_ISDIR(mode) ? tree_type :
2952 blob_type;
2953
2954 if (!mode) {
2955 /* missing SP path LF */
2956 strbuf_reset(&line);
2957 strbuf_addstr(&line, "missing ");
2958 quote_c_style(path, &line, NULL, 0);
2959 strbuf_addch(&line, '\n');
2960 } else {
2961 /* mode SP type SP object_name TAB path LF */
2962 strbuf_reset(&line);
2963 strbuf_addf(&line, "%06o %s %s\t",
2964 mode & ~NO_DELTA, type, sha1_to_hex(sha1));
2965 quote_c_style(path, &line, NULL, 0);
2966 strbuf_addch(&line, '\n');
2967 }
2968 cat_blob_write(line.buf, line.len);
2969}
2970
2971static void parse_ls(const char *p, struct branch *b)
2972{
2973 struct tree_entry *root = NULL;
2974 struct tree_entry leaf = {NULL};
2975
2976 /* ls SP (<tree-ish> SP)? <path> */
2977 if (*p == '"') {
2978 if (!b)
2979 die("Not in a commit: %s", command_buf.buf);
2980 root = &b->branch_tree;
2981 } else {
2982 struct object_entry *e = parse_treeish_dataref(&p);
2983 root = new_tree_entry();
2984 oidcpy(&root->versions[1].oid, &e->idx.oid);
2985 if (!is_null_oid(&root->versions[1].oid))
2986 root->versions[1].mode = S_IFDIR;
2987 load_tree(root);
2988 }
2989 if (*p == '"') {
2990 static struct strbuf uq = STRBUF_INIT;
2991 const char *endp;
2992 strbuf_reset(&uq);
2993 if (unquote_c_style(&uq, p, &endp))
2994 die("Invalid path: %s", command_buf.buf);
2995 if (*endp)
2996 die("Garbage after path in: %s", command_buf.buf);
2997 p = uq.buf;
2998 }
2999 tree_content_get(root, p, &leaf, 1);
3000 /*
3001 * A directory in preparation would have a sha1 of zero
3002 * until it is saved. Save, for simplicity.
3003 */
3004 if (S_ISDIR(leaf.versions[1].mode))
3005 store_tree(&leaf);
3006
3007 print_ls(leaf.versions[1].mode, leaf.versions[1].oid.hash, p);
3008 if (leaf.tree)
3009 release_tree_content_recursive(leaf.tree);
3010 if (!b || root != &b->branch_tree)
3011 release_tree_entry(root);
3012}
3013
3014static void checkpoint(void)
3015{
3016 checkpoint_requested = 0;
3017 if (object_count) {
3018 cycle_packfile();
3019 }
3020 dump_branches();
3021 dump_tags();
3022 dump_marks();
3023}
3024
3025static void parse_checkpoint(void)
3026{
3027 checkpoint_requested = 1;
3028 skip_optional_lf();
3029}
3030
3031static void parse_progress(void)
3032{
3033 fwrite(command_buf.buf, 1, command_buf.len, stdout);
3034 fputc('\n', stdout);
3035 fflush(stdout);
3036 skip_optional_lf();
3037}
3038
3039static char* make_fast_import_path(const char *path)
3040{
3041 if (!relative_marks_paths || is_absolute_path(path))
3042 return xstrdup(path);
3043 return git_pathdup("info/fast-import/%s", path);
3044}
3045
3046static void option_import_marks(const char *marks,
3047 int from_stream, int ignore_missing)
3048{
3049 if (import_marks_file) {
3050 if (from_stream)
3051 die("Only one import-marks command allowed per stream");
3052
3053 /* read previous mark file */
3054 if(!import_marks_file_from_stream)
3055 read_marks();
3056 }
3057
3058 import_marks_file = make_fast_import_path(marks);
3059 safe_create_leading_directories_const(import_marks_file);
3060 import_marks_file_from_stream = from_stream;
3061 import_marks_file_ignore_missing = ignore_missing;
3062}
3063
3064static void option_date_format(const char *fmt)
3065{
3066 if (!strcmp(fmt, "raw"))
3067 whenspec = WHENSPEC_RAW;
3068 else if (!strcmp(fmt, "rfc2822"))
3069 whenspec = WHENSPEC_RFC2822;
3070 else if (!strcmp(fmt, "now"))
3071 whenspec = WHENSPEC_NOW;
3072 else
3073 die("unknown --date-format argument %s", fmt);
3074}
3075
3076static unsigned long ulong_arg(const char *option, const char *arg)
3077{
3078 char *endptr;
3079 unsigned long rv = strtoul(arg, &endptr, 0);
3080 if (strchr(arg, '-') || endptr == arg || *endptr)
3081 die("%s: argument must be a non-negative integer", option);
3082 return rv;
3083}
3084
3085static void option_depth(const char *depth)
3086{
3087 max_depth = ulong_arg("--depth", depth);
3088 if (max_depth > MAX_DEPTH)
3089 die("--depth cannot exceed %u", MAX_DEPTH);
3090}
3091
3092static void option_active_branches(const char *branches)
3093{
3094 max_active_branches = ulong_arg("--active-branches", branches);
3095}
3096
3097static void option_export_marks(const char *marks)
3098{
3099 export_marks_file = make_fast_import_path(marks);
3100 safe_create_leading_directories_const(export_marks_file);
3101}
3102
3103static void option_cat_blob_fd(const char *fd)
3104{
3105 unsigned long n = ulong_arg("--cat-blob-fd", fd);
3106 if (n > (unsigned long) INT_MAX)
3107 die("--cat-blob-fd cannot exceed %d", INT_MAX);
3108 cat_blob_fd = (int) n;
3109}
3110
3111static void option_export_pack_edges(const char *edges)
3112{
3113 if (pack_edges)
3114 fclose(pack_edges);
3115 pack_edges = xfopen(edges, "a");
3116}
3117
3118static int parse_one_option(const char *option)
3119{
3120 if (skip_prefix(option, "max-pack-size=", &option)) {
3121 unsigned long v;
3122 if (!git_parse_ulong(option, &v))
3123 return 0;
3124 if (v < 8192) {
3125 warning("max-pack-size is now in bytes, assuming --max-pack-size=%lum", v);
3126 v *= 1024 * 1024;
3127 } else if (v < 1024 * 1024) {
3128 warning("minimum max-pack-size is 1 MiB");
3129 v = 1024 * 1024;
3130 }
3131 max_packsize = v;
3132 } else if (skip_prefix(option, "big-file-threshold=", &option)) {
3133 unsigned long v;
3134 if (!git_parse_ulong(option, &v))
3135 return 0;
3136 big_file_threshold = v;
3137 } else if (skip_prefix(option, "depth=", &option)) {
3138 option_depth(option);
3139 } else if (skip_prefix(option, "active-branches=", &option)) {
3140 option_active_branches(option);
3141 } else if (skip_prefix(option, "export-pack-edges=", &option)) {
3142 option_export_pack_edges(option);
3143 } else if (starts_with(option, "quiet")) {
3144 show_stats = 0;
3145 } else if (starts_with(option, "stats")) {
3146 show_stats = 1;
3147 } else {
3148 return 0;
3149 }
3150
3151 return 1;
3152}
3153
3154static int parse_one_feature(const char *feature, int from_stream)
3155{
3156 const char *arg;
3157
3158 if (skip_prefix(feature, "date-format=", &arg)) {
3159 option_date_format(arg);
3160 } else if (skip_prefix(feature, "import-marks=", &arg)) {
3161 option_import_marks(arg, from_stream, 0);
3162 } else if (skip_prefix(feature, "import-marks-if-exists=", &arg)) {
3163 option_import_marks(arg, from_stream, 1);
3164 } else if (skip_prefix(feature, "export-marks=", &arg)) {
3165 option_export_marks(arg);
3166 } else if (!strcmp(feature, "get-mark")) {
3167 ; /* Don't die - this feature is supported */
3168 } else if (!strcmp(feature, "cat-blob")) {
3169 ; /* Don't die - this feature is supported */
3170 } else if (!strcmp(feature, "relative-marks")) {
3171 relative_marks_paths = 1;
3172 } else if (!strcmp(feature, "no-relative-marks")) {
3173 relative_marks_paths = 0;
3174 } else if (!strcmp(feature, "done")) {
3175 require_explicit_termination = 1;
3176 } else if (!strcmp(feature, "force")) {
3177 force_update = 1;
3178 } else if (!strcmp(feature, "notes") || !strcmp(feature, "ls")) {
3179 ; /* do nothing; we have the feature */
3180 } else {
3181 return 0;
3182 }
3183
3184 return 1;
3185}
3186
3187static void parse_feature(const char *feature)
3188{
3189 if (seen_data_command)
3190 die("Got feature command '%s' after data command", feature);
3191
3192 if (parse_one_feature(feature, 1))
3193 return;
3194
3195 die("This version of fast-import does not support feature %s.", feature);
3196}
3197
3198static void parse_option(const char *option)
3199{
3200 if (seen_data_command)
3201 die("Got option command '%s' after data command", option);
3202
3203 if (parse_one_option(option))
3204 return;
3205
3206 die("This version of fast-import does not support option: %s", option);
3207}
3208
3209static void git_pack_config(void)
3210{
3211 int indexversion_value;
3212 int limit;
3213 unsigned long packsizelimit_value;
3214
3215 if (!git_config_get_ulong("pack.depth", &max_depth)) {
3216 if (max_depth > MAX_DEPTH)
3217 max_depth = MAX_DEPTH;
3218 }
3219 if (!git_config_get_int("pack.indexversion", &indexversion_value)) {
3220 pack_idx_opts.version = indexversion_value;
3221 if (pack_idx_opts.version > 2)
3222 git_die_config("pack.indexversion",
3223 "bad pack.indexversion=%"PRIu32, pack_idx_opts.version);
3224 }
3225 if (!git_config_get_ulong("pack.packsizelimit", &packsizelimit_value))
3226 max_packsize = packsizelimit_value;
3227
3228 if (!git_config_get_int("fastimport.unpacklimit", &limit))
3229 unpack_limit = limit;
3230 else if (!git_config_get_int("transfer.unpacklimit", &limit))
3231 unpack_limit = limit;
3232
3233 git_config(git_default_config, NULL);
3234}
3235
3236static const char fast_import_usage[] =
3237"git fast-import [--date-format=<f>] [--max-pack-size=<n>] [--big-file-threshold=<n>] [--depth=<n>] [--active-branches=<n>] [--export-marks=<marks.file>]";
3238
3239static void parse_argv(void)
3240{
3241 unsigned int i;
3242
3243 for (i = 1; i < global_argc; i++) {
3244 const char *a = global_argv[i];
3245
3246 if (*a != '-' || !strcmp(a, "--"))
3247 break;
3248
3249 if (!skip_prefix(a, "--", &a))
3250 die("unknown option %s", a);
3251
3252 if (parse_one_option(a))
3253 continue;
3254
3255 if (parse_one_feature(a, 0))
3256 continue;
3257
3258 if (skip_prefix(a, "cat-blob-fd=", &a)) {
3259 option_cat_blob_fd(a);
3260 continue;
3261 }
3262
3263 die("unknown option --%s", a);
3264 }
3265 if (i != global_argc)
3266 usage(fast_import_usage);
3267
3268 seen_data_command = 1;
3269 if (import_marks_file)
3270 read_marks();
3271}
3272
3273int cmd_main(int argc, const char **argv)
3274{
3275 unsigned int i;
3276
3277 if (argc == 2 && !strcmp(argv[1], "-h"))
3278 usage(fast_import_usage);
3279
3280 setup_git_directory();
3281 reset_pack_idx_option(&pack_idx_opts);
3282 git_pack_config();
3283
3284 alloc_objects(object_entry_alloc);
3285 strbuf_init(&command_buf, 0);
3286 atom_table = xcalloc(atom_table_sz, sizeof(struct atom_str*));
3287 branch_table = xcalloc(branch_table_sz, sizeof(struct branch*));
3288 avail_tree_table = xcalloc(avail_tree_table_sz, sizeof(struct avail_tree_content*));
3289 marks = mem_pool_calloc(&fi_mem_pool, 1, sizeof(struct mark_set));
3290
3291 global_argc = argc;
3292 global_argv = argv;
3293
3294 rc_free = mem_pool_alloc(&fi_mem_pool, cmd_save * sizeof(*rc_free));
3295 for (i = 0; i < (cmd_save - 1); i++)
3296 rc_free[i].next = &rc_free[i + 1];
3297 rc_free[cmd_save - 1].next = NULL;
3298
3299 start_packfile();
3300 set_die_routine(die_nicely);
3301 set_checkpoint_signal();
3302 while (read_next_command() != EOF) {
3303 const char *v;
3304 if (!strcmp("blob", command_buf.buf))
3305 parse_new_blob();
3306 else if (skip_prefix(command_buf.buf, "ls ", &v))
3307 parse_ls(v, NULL);
3308 else if (skip_prefix(command_buf.buf, "commit ", &v))
3309 parse_new_commit(v);
3310 else if (skip_prefix(command_buf.buf, "tag ", &v))
3311 parse_new_tag(v);
3312 else if (skip_prefix(command_buf.buf, "reset ", &v))
3313 parse_reset_branch(v);
3314 else if (!strcmp("checkpoint", command_buf.buf))
3315 parse_checkpoint();
3316 else if (!strcmp("done", command_buf.buf))
3317 break;
3318 else if (starts_with(command_buf.buf, "progress "))
3319 parse_progress();
3320 else if (skip_prefix(command_buf.buf, "feature ", &v))
3321 parse_feature(v);
3322 else if (skip_prefix(command_buf.buf, "option git ", &v))
3323 parse_option(v);
3324 else if (starts_with(command_buf.buf, "option "))
3325 /* ignore non-git options*/;
3326 else
3327 die("Unsupported command: %s", command_buf.buf);
3328
3329 if (checkpoint_requested)
3330 checkpoint();
3331 }
3332
3333 /* argv hasn't been parsed yet, do so */
3334 if (!seen_data_command)
3335 parse_argv();
3336
3337 if (require_explicit_termination && feof(stdin))
3338 die("stream ends early");
3339
3340 end_packfile();
3341
3342 dump_branches();
3343 dump_tags();
3344 unkeep_all_packs();
3345 dump_marks();
3346
3347 if (pack_edges)
3348 fclose(pack_edges);
3349
3350 if (show_stats) {
3351 uintmax_t total_count = 0, duplicate_count = 0;
3352 for (i = 0; i < ARRAY_SIZE(object_count_by_type); i++)
3353 total_count += object_count_by_type[i];
3354 for (i = 0; i < ARRAY_SIZE(duplicate_count_by_type); i++)
3355 duplicate_count += duplicate_count_by_type[i];
3356
3357 fprintf(stderr, "%s statistics:\n", argv[0]);
3358 fprintf(stderr, "---------------------------------------------------------------------\n");
3359 fprintf(stderr, "Alloc'd objects: %10" PRIuMAX "\n", alloc_count);
3360 fprintf(stderr, "Total objects: %10" PRIuMAX " (%10" PRIuMAX " duplicates )\n", total_count, duplicate_count);
3361 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]);
3362 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]);
3363 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]);
3364 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]);
3365 fprintf(stderr, "Total branches: %10lu (%10lu loads )\n", branch_count, branch_load_count);
3366 fprintf(stderr, " marks: %10" PRIuMAX " (%10" PRIuMAX " unique )\n", (((uintmax_t)1) << marks->shift) * 1024, marks_set_count);
3367 fprintf(stderr, " atoms: %10u\n", atom_cnt);
3368 fprintf(stderr, "Memory total: %10" PRIuMAX " KiB\n", (tree_entry_allocd + fi_mem_pool.pool_alloc + alloc_count*sizeof(struct object_entry))/1024);
3369 fprintf(stderr, " pools: %10lu KiB\n", (unsigned long)((tree_entry_allocd + fi_mem_pool.pool_alloc) /1024));
3370 fprintf(stderr, " objects: %10" PRIuMAX " KiB\n", (alloc_count*sizeof(struct object_entry))/1024);
3371 fprintf(stderr, "---------------------------------------------------------------------\n");
3372 pack_report();
3373 fprintf(stderr, "---------------------------------------------------------------------\n");
3374 fprintf(stderr, "\n");
3375 }
3376
3377 return failure ? 1 : 0;
3378}