freshlybakedca.ke
Git fork
1/*
2 * git gc builtin command
3 *
4 * Cleanup unreachable files and optimize the repository.
5 *
6 * Copyright (c) 2007 James Bowes
7 *
8 * Based on git-gc.sh, which is
9 *
10 * Copyright (c) 2006 Shawn O. Pearce
11 */
12
13#define USE_THE_REPOSITORY_VARIABLE
14#define DISABLE_SIGN_COMPARE_WARNINGS
15
16#include "builtin.h"
17#include "abspath.h"
18#include "date.h"
19#include "dir.h"
20#include "environment.h"
21#include "hex.h"
22#include "config.h"
23#include "tempfile.h"
24#include "lockfile.h"
25#include "parse-options.h"
26#include "run-command.h"
27#include "sigchain.h"
28#include "strvec.h"
29#include "commit.h"
30#include "commit-graph.h"
31#include "packfile.h"
32#include "object-file.h"
33#include "pack.h"
34#include "pack-objects.h"
35#include "path.h"
36#include "reflog.h"
37#include "rerere.h"
38#include "blob.h"
39#include "tree.h"
40#include "promisor-remote.h"
41#include "refs.h"
42#include "remote.h"
43#include "exec-cmd.h"
44#include "gettext.h"
45#include "hook.h"
46#include "setup.h"
47#include "trace2.h"
48#include "worktree.h"
49
50#define FAILED_RUN "failed to run %s"
51
52static const char * const builtin_gc_usage[] = {
53 N_("git gc [<options>]"),
54 NULL
55};
56
57static timestamp_t gc_log_expire_time;
58static struct strvec repack = STRVEC_INIT;
59static struct tempfile *pidfile;
60static struct lock_file log_lock;
61static struct string_list pack_garbage = STRING_LIST_INIT_DUP;
62
63static void clean_pack_garbage(void)
64{
65 int i;
66 for (i = 0; i < pack_garbage.nr; i++)
67 unlink_or_warn(pack_garbage.items[i].string);
68 string_list_clear(&pack_garbage, 0);
69}
70
71static void report_pack_garbage(unsigned seen_bits, const char *path)
72{
73 if (seen_bits == PACKDIR_FILE_IDX)
74 string_list_append(&pack_garbage, path);
75}
76
77static void process_log_file(void)
78{
79 struct stat st;
80 if (fstat(get_lock_file_fd(&log_lock), &st)) {
81 /*
82 * Perhaps there was an i/o error or another
83 * unlikely situation. Try to make a note of
84 * this in gc.log along with any existing
85 * messages.
86 */
87 int saved_errno = errno;
88 fprintf(stderr, _("Failed to fstat %s: %s"),
89 get_lock_file_path(&log_lock),
90 strerror(saved_errno));
91 fflush(stderr);
92 commit_lock_file(&log_lock);
93 errno = saved_errno;
94 } else if (st.st_size) {
95 /* There was some error recorded in the lock file */
96 commit_lock_file(&log_lock);
97 } else {
98 char *path = repo_git_path(the_repository, "gc.log");
99 /* No error, clean up any old gc.log */
100 unlink(path);
101 rollback_lock_file(&log_lock);
102 free(path);
103 }
104}
105
106static void process_log_file_at_exit(void)
107{
108 fflush(stderr);
109 process_log_file();
110}
111
112static int gc_config_is_timestamp_never(const char *var)
113{
114 const char *value;
115 timestamp_t expire;
116
117 if (!repo_config_get_value(the_repository, var, &value) && value) {
118 if (parse_expiry_date(value, &expire))
119 die(_("failed to parse '%s' value '%s'"), var, value);
120 return expire == 0;
121 }
122 return 0;
123}
124
125struct gc_config {
126 int pack_refs;
127 int prune_reflogs;
128 int cruft_packs;
129 unsigned long max_cruft_size;
130 int aggressive_depth;
131 int aggressive_window;
132 int gc_auto_threshold;
133 int gc_auto_pack_limit;
134 int detach_auto;
135 char *gc_log_expire;
136 char *prune_expire;
137 char *prune_worktrees_expire;
138 char *repack_filter;
139 char *repack_filter_to;
140 char *repack_expire_to;
141 unsigned long big_pack_threshold;
142 unsigned long max_delta_cache_size;
143 /*
144 * Remove this member from gc_config once repo_settings is passed
145 * through the callchain.
146 */
147 size_t delta_base_cache_limit;
148};
149
150#define GC_CONFIG_INIT { \
151 .pack_refs = 1, \
152 .prune_reflogs = 1, \
153 .cruft_packs = 1, \
154 .aggressive_depth = 50, \
155 .aggressive_window = 250, \
156 .gc_auto_threshold = 6700, \
157 .gc_auto_pack_limit = 50, \
158 .detach_auto = 1, \
159 .gc_log_expire = xstrdup("1.day.ago"), \
160 .prune_expire = xstrdup("2.weeks.ago"), \
161 .prune_worktrees_expire = xstrdup("3.months.ago"), \
162 .max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE, \
163 .delta_base_cache_limit = DEFAULT_DELTA_BASE_CACHE_LIMIT, \
164}
165
166static void gc_config_release(struct gc_config *cfg)
167{
168 free(cfg->gc_log_expire);
169 free(cfg->prune_expire);
170 free(cfg->prune_worktrees_expire);
171 free(cfg->repack_filter);
172 free(cfg->repack_filter_to);
173}
174
175static void gc_config(struct gc_config *cfg)
176{
177 const char *value;
178 char *owned = NULL;
179 unsigned long ulongval;
180
181 if (!repo_config_get_value(the_repository, "gc.packrefs", &value)) {
182 if (value && !strcmp(value, "notbare"))
183 cfg->pack_refs = -1;
184 else
185 cfg->pack_refs = git_config_bool("gc.packrefs", value);
186 }
187
188 if (gc_config_is_timestamp_never("gc.reflogexpire") &&
189 gc_config_is_timestamp_never("gc.reflogexpireunreachable"))
190 cfg->prune_reflogs = 0;
191
192 repo_config_get_int(the_repository, "gc.aggressivewindow", &cfg->aggressive_window);
193 repo_config_get_int(the_repository, "gc.aggressivedepth", &cfg->aggressive_depth);
194 repo_config_get_int(the_repository, "gc.auto", &cfg->gc_auto_threshold);
195 repo_config_get_int(the_repository, "gc.autopacklimit", &cfg->gc_auto_pack_limit);
196 repo_config_get_bool(the_repository, "gc.autodetach", &cfg->detach_auto);
197 repo_config_get_bool(the_repository, "gc.cruftpacks", &cfg->cruft_packs);
198 repo_config_get_ulong(the_repository, "gc.maxcruftsize", &cfg->max_cruft_size);
199
200 if (!repo_config_get_expiry(the_repository, "gc.pruneexpire", &owned)) {
201 free(cfg->prune_expire);
202 cfg->prune_expire = owned;
203 }
204
205 if (!repo_config_get_expiry(the_repository, "gc.worktreepruneexpire", &owned)) {
206 free(cfg->prune_worktrees_expire);
207 cfg->prune_worktrees_expire = owned;
208 }
209
210 if (!repo_config_get_expiry(the_repository, "gc.logexpiry", &owned)) {
211 free(cfg->gc_log_expire);
212 cfg->gc_log_expire = owned;
213 }
214
215 repo_config_get_ulong(the_repository, "gc.bigpackthreshold", &cfg->big_pack_threshold);
216 repo_config_get_ulong(the_repository, "pack.deltacachesize", &cfg->max_delta_cache_size);
217
218 if (!repo_config_get_ulong(the_repository, "core.deltabasecachelimit", &ulongval))
219 cfg->delta_base_cache_limit = ulongval;
220
221 if (!repo_config_get_string(the_repository, "gc.repackfilter", &owned)) {
222 free(cfg->repack_filter);
223 cfg->repack_filter = owned;
224 }
225
226 if (!repo_config_get_string(the_repository, "gc.repackfilterto", &owned)) {
227 free(cfg->repack_filter_to);
228 cfg->repack_filter_to = owned;
229 }
230
231 repo_config(the_repository, git_default_config, NULL);
232}
233
234enum schedule_priority {
235 SCHEDULE_NONE = 0,
236 SCHEDULE_WEEKLY = 1,
237 SCHEDULE_DAILY = 2,
238 SCHEDULE_HOURLY = 3,
239};
240
241static enum schedule_priority parse_schedule(const char *value)
242{
243 if (!value)
244 return SCHEDULE_NONE;
245 if (!strcasecmp(value, "hourly"))
246 return SCHEDULE_HOURLY;
247 if (!strcasecmp(value, "daily"))
248 return SCHEDULE_DAILY;
249 if (!strcasecmp(value, "weekly"))
250 return SCHEDULE_WEEKLY;
251 return SCHEDULE_NONE;
252}
253
254enum maintenance_task_label {
255 TASK_PREFETCH,
256 TASK_LOOSE_OBJECTS,
257 TASK_INCREMENTAL_REPACK,
258 TASK_GC,
259 TASK_COMMIT_GRAPH,
260 TASK_PACK_REFS,
261 TASK_REFLOG_EXPIRE,
262 TASK_WORKTREE_PRUNE,
263 TASK_RERERE_GC,
264
265 /* Leave as final value */
266 TASK__COUNT
267};
268
269struct maintenance_run_opts {
270 enum maintenance_task_label *tasks;
271 size_t tasks_nr, tasks_alloc;
272 int auto_flag;
273 int detach;
274 int quiet;
275 enum schedule_priority schedule;
276};
277#define MAINTENANCE_RUN_OPTS_INIT { \
278 .detach = -1, \
279}
280
281static void maintenance_run_opts_release(struct maintenance_run_opts *opts)
282{
283 free(opts->tasks);
284}
285
286static int pack_refs_condition(UNUSED struct gc_config *cfg)
287{
288 /*
289 * The auto-repacking logic for refs is handled by the ref backends and
290 * exposed via `git pack-refs --auto`. We thus always return truish
291 * here and let the backend decide for us.
292 */
293 return 1;
294}
295
296static int maintenance_task_pack_refs(struct maintenance_run_opts *opts,
297 UNUSED struct gc_config *cfg)
298{
299 struct child_process cmd = CHILD_PROCESS_INIT;
300
301 cmd.git_cmd = 1;
302 strvec_pushl(&cmd.args, "pack-refs", "--all", "--prune", NULL);
303 if (opts->auto_flag)
304 strvec_push(&cmd.args, "--auto");
305
306 return run_command(&cmd);
307}
308
309struct count_reflog_entries_data {
310 struct expire_reflog_policy_cb policy;
311 size_t count;
312 size_t limit;
313};
314
315static int count_reflog_entries(const char *refname UNUSED,
316 struct object_id *old_oid, struct object_id *new_oid,
317 const char *committer, timestamp_t timestamp,
318 int tz, const char *msg, void *cb_data)
319{
320 struct count_reflog_entries_data *data = cb_data;
321 if (should_expire_reflog_ent(old_oid, new_oid, committer, timestamp, tz, msg, &data->policy))
322 data->count++;
323 return data->count >= data->limit;
324}
325
326static int reflog_expire_condition(struct gc_config *cfg UNUSED)
327{
328 timestamp_t now = time(NULL);
329 struct count_reflog_entries_data data = {
330 .policy = {
331 .opts = REFLOG_EXPIRE_OPTIONS_INIT(now),
332 },
333 };
334 int limit = 100;
335
336 repo_config_get_int(the_repository, "maintenance.reflog-expire.auto", &limit);
337 if (!limit)
338 return 0;
339 if (limit < 0)
340 return 1;
341 data.limit = limit;
342
343 repo_config(the_repository, reflog_expire_config, &data.policy.opts);
344
345 reflog_expire_options_set_refname(&data.policy.opts, "HEAD");
346 refs_for_each_reflog_ent(get_main_ref_store(the_repository), "HEAD",
347 count_reflog_entries, &data);
348
349 reflog_expiry_cleanup(&data.policy);
350 reflog_clear_expire_config(&data.policy.opts);
351 return data.count >= data.limit;
352}
353
354static int maintenance_task_reflog_expire(struct maintenance_run_opts *opts UNUSED,
355 struct gc_config *cfg UNUSED)
356{
357 struct child_process cmd = CHILD_PROCESS_INIT;
358 cmd.git_cmd = 1;
359 strvec_pushl(&cmd.args, "reflog", "expire", "--all", NULL);
360 return run_command(&cmd);
361}
362
363static int maintenance_task_worktree_prune(struct maintenance_run_opts *opts UNUSED,
364 struct gc_config *cfg)
365{
366 struct child_process prune_worktrees_cmd = CHILD_PROCESS_INIT;
367
368 prune_worktrees_cmd.git_cmd = 1;
369 strvec_pushl(&prune_worktrees_cmd.args, "worktree", "prune", "--expire", NULL);
370 strvec_push(&prune_worktrees_cmd.args, cfg->prune_worktrees_expire);
371
372 return run_command(&prune_worktrees_cmd);
373}
374
375static int worktree_prune_condition(struct gc_config *cfg)
376{
377 struct strbuf buf = STRBUF_INIT;
378 int should_prune = 0, limit = 1;
379 timestamp_t expiry_date;
380 struct dirent *d;
381 DIR *dir = NULL;
382
383 repo_config_get_int(the_repository, "maintenance.worktree-prune.auto", &limit);
384 if (limit <= 0) {
385 should_prune = limit < 0;
386 goto out;
387 }
388
389 if (parse_expiry_date(cfg->prune_worktrees_expire, &expiry_date))
390 goto out;
391
392 dir = opendir(repo_git_path_replace(the_repository, &buf, "worktrees"));
393 if (!dir)
394 goto out;
395
396 while (limit && (d = readdir_skip_dot_and_dotdot(dir))) {
397 char *wtpath;
398 strbuf_reset(&buf);
399 if (should_prune_worktree(d->d_name, &buf, &wtpath, expiry_date))
400 limit--;
401 free(wtpath);
402 }
403
404 should_prune = !limit;
405
406out:
407 if (dir)
408 closedir(dir);
409 strbuf_release(&buf);
410 return should_prune;
411}
412
413static int maintenance_task_rerere_gc(struct maintenance_run_opts *opts UNUSED,
414 struct gc_config *cfg UNUSED)
415{
416 struct child_process rerere_cmd = CHILD_PROCESS_INIT;
417 rerere_cmd.git_cmd = 1;
418 strvec_pushl(&rerere_cmd.args, "rerere", "gc", NULL);
419 return run_command(&rerere_cmd);
420}
421
422static int rerere_gc_condition(struct gc_config *cfg UNUSED)
423{
424 struct strbuf path = STRBUF_INIT;
425 int should_gc = 0, limit = 1;
426 DIR *dir = NULL;
427
428 repo_config_get_int(the_repository, "maintenance.rerere-gc.auto", &limit);
429 if (limit <= 0) {
430 should_gc = limit < 0;
431 goto out;
432 }
433
434 /*
435 * We skip garbage collection in case we either have no "rr-cache"
436 * directory or when it doesn't contain at least one entry.
437 */
438 repo_git_path_replace(the_repository, &path, "rr-cache");
439 dir = opendir(path.buf);
440 if (!dir)
441 goto out;
442 should_gc = !!readdir_skip_dot_and_dotdot(dir);
443
444out:
445 strbuf_release(&path);
446 if (dir)
447 closedir(dir);
448 return should_gc;
449}
450
451static int too_many_loose_objects(struct gc_config *cfg)
452{
453 /*
454 * Quickly check if a "gc" is needed, by estimating how
455 * many loose objects there are. Because SHA-1 is evenly
456 * distributed, we can check only one and get a reasonable
457 * estimate.
458 */
459 DIR *dir;
460 struct dirent *ent;
461 int auto_threshold;
462 int num_loose = 0;
463 int needed = 0;
464 const unsigned hexsz_loose = the_hash_algo->hexsz - 2;
465 char *path;
466
467 path = repo_git_path(the_repository, "objects/17");
468 dir = opendir(path);
469 free(path);
470 if (!dir)
471 return 0;
472
473 auto_threshold = DIV_ROUND_UP(cfg->gc_auto_threshold, 256);
474 while ((ent = readdir(dir)) != NULL) {
475 if (strspn(ent->d_name, "0123456789abcdef") != hexsz_loose ||
476 ent->d_name[hexsz_loose] != '\0')
477 continue;
478 if (++num_loose > auto_threshold) {
479 needed = 1;
480 break;
481 }
482 }
483 closedir(dir);
484 return needed;
485}
486
487static struct packed_git *find_base_packs(struct string_list *packs,
488 unsigned long limit)
489{
490 struct packfile_store *packfiles = the_repository->objects->packfiles;
491 struct packed_git *p, *base = NULL;
492
493 for (p = packfile_store_get_all_packs(packfiles); p; p = p->next) {
494 if (!p->pack_local || p->is_cruft)
495 continue;
496 if (limit) {
497 if (p->pack_size >= limit)
498 string_list_append(packs, p->pack_name);
499 } else if (!base || base->pack_size < p->pack_size) {
500 base = p;
501 }
502 }
503
504 if (base)
505 string_list_append(packs, base->pack_name);
506
507 return base;
508}
509
510static int too_many_packs(struct gc_config *cfg)
511{
512 struct packfile_store *packs = the_repository->objects->packfiles;
513 struct packed_git *p;
514 int cnt;
515
516 if (cfg->gc_auto_pack_limit <= 0)
517 return 0;
518
519 for (cnt = 0, p = packfile_store_get_all_packs(packs); p; p = p->next) {
520 if (!p->pack_local)
521 continue;
522 if (p->pack_keep)
523 continue;
524 /*
525 * Perhaps check the size of the pack and count only
526 * very small ones here?
527 */
528 cnt++;
529 }
530 return cfg->gc_auto_pack_limit < cnt;
531}
532
533static uint64_t total_ram(void)
534{
535#if defined(HAVE_SYSINFO)
536 struct sysinfo si;
537
538 if (!sysinfo(&si)) {
539 uint64_t total = si.totalram;
540
541 if (si.mem_unit > 1)
542 total *= (uint64_t)si.mem_unit;
543 return total;
544 }
545#elif defined(HAVE_BSD_SYSCTL) && (defined(HW_MEMSIZE) || defined(HW_PHYSMEM) || defined(HW_PHYSMEM64))
546 uint64_t physical_memory;
547 int mib[2];
548 size_t length;
549
550 mib[0] = CTL_HW;
551# if defined(HW_MEMSIZE)
552 mib[1] = HW_MEMSIZE;
553# elif defined(HW_PHYSMEM64)
554 mib[1] = HW_PHYSMEM64;
555# else
556 mib[1] = HW_PHYSMEM;
557# endif
558 length = sizeof(physical_memory);
559 if (!sysctl(mib, 2, &physical_memory, &length, NULL, 0)) {
560 if (length == 4) {
561 uint32_t mem;
562
563 if (!sysctl(mib, 2, &mem, &length, NULL, 0))
564 physical_memory = mem;
565 }
566 return physical_memory;
567 }
568#elif defined(GIT_WINDOWS_NATIVE)
569 MEMORYSTATUSEX memInfo;
570
571 memInfo.dwLength = sizeof(MEMORYSTATUSEX);
572 if (GlobalMemoryStatusEx(&memInfo))
573 return memInfo.ullTotalPhys;
574#endif
575 return 0;
576}
577
578static uint64_t estimate_repack_memory(struct gc_config *cfg,
579 struct packed_git *pack)
580{
581 unsigned long nr_objects = repo_approximate_object_count(the_repository);
582 size_t os_cache, heap;
583
584 if (!pack || !nr_objects)
585 return 0;
586
587 /*
588 * First we have to scan through at least one pack.
589 * Assume enough room in OS file cache to keep the entire pack
590 * or we may accidentally evict data of other processes from
591 * the cache.
592 */
593 os_cache = pack->pack_size + pack->index_size;
594 /* then pack-objects needs lots more for book keeping */
595 heap = sizeof(struct object_entry) * nr_objects;
596 /*
597 * internal rev-list --all --objects takes up some memory too,
598 * let's say half of it is for blobs
599 */
600 heap += sizeof(struct blob) * nr_objects / 2;
601 /*
602 * and the other half is for trees (commits and tags are
603 * usually insignificant)
604 */
605 heap += sizeof(struct tree) * nr_objects / 2;
606 /* and then obj_hash[], underestimated in fact */
607 heap += sizeof(struct object *) * nr_objects;
608 /* revindex is used also */
609 heap += (sizeof(off_t) + sizeof(uint32_t)) * nr_objects;
610 /*
611 * read_sha1_file() (either at delta calculation phase, or
612 * writing phase) also fills up the delta base cache
613 */
614 heap += cfg->delta_base_cache_limit;
615 /* and of course pack-objects has its own delta cache */
616 heap += cfg->max_delta_cache_size;
617
618 return os_cache + heap;
619}
620
621static int keep_one_pack(struct string_list_item *item, void *data UNUSED)
622{
623 strvec_pushf(&repack, "--keep-pack=%s", basename(item->string));
624 return 0;
625}
626
627static void add_repack_all_option(struct gc_config *cfg,
628 struct string_list *keep_pack)
629{
630 if (cfg->prune_expire && !strcmp(cfg->prune_expire, "now")
631 && !(cfg->cruft_packs && cfg->repack_expire_to))
632 strvec_push(&repack, "-a");
633 else if (cfg->cruft_packs) {
634 strvec_push(&repack, "--cruft");
635 if (cfg->prune_expire)
636 strvec_pushf(&repack, "--cruft-expiration=%s", cfg->prune_expire);
637 if (cfg->max_cruft_size)
638 strvec_pushf(&repack, "--max-cruft-size=%lu",
639 cfg->max_cruft_size);
640 if (cfg->repack_expire_to)
641 strvec_pushf(&repack, "--expire-to=%s", cfg->repack_expire_to);
642 } else {
643 strvec_push(&repack, "-A");
644 if (cfg->prune_expire)
645 strvec_pushf(&repack, "--unpack-unreachable=%s", cfg->prune_expire);
646 }
647
648 if (keep_pack)
649 for_each_string_list(keep_pack, keep_one_pack, NULL);
650
651 if (cfg->repack_filter && *cfg->repack_filter)
652 strvec_pushf(&repack, "--filter=%s", cfg->repack_filter);
653 if (cfg->repack_filter_to && *cfg->repack_filter_to)
654 strvec_pushf(&repack, "--filter-to=%s", cfg->repack_filter_to);
655}
656
657static void add_repack_incremental_option(void)
658{
659 strvec_push(&repack, "--no-write-bitmap-index");
660}
661
662static int need_to_gc(struct gc_config *cfg)
663{
664 /*
665 * Setting gc.auto to 0 or negative can disable the
666 * automatic gc.
667 */
668 if (cfg->gc_auto_threshold <= 0)
669 return 0;
670
671 /*
672 * If there are too many loose objects, but not too many
673 * packs, we run "repack -d -l". If there are too many packs,
674 * we run "repack -A -d -l". Otherwise we tell the caller
675 * there is no need.
676 */
677 if (too_many_packs(cfg)) {
678 struct string_list keep_pack = STRING_LIST_INIT_NODUP;
679
680 if (cfg->big_pack_threshold) {
681 find_base_packs(&keep_pack, cfg->big_pack_threshold);
682 if (keep_pack.nr >= cfg->gc_auto_pack_limit) {
683 cfg->big_pack_threshold = 0;
684 string_list_clear(&keep_pack, 0);
685 find_base_packs(&keep_pack, 0);
686 }
687 } else {
688 struct packed_git *p = find_base_packs(&keep_pack, 0);
689 uint64_t mem_have, mem_want;
690
691 mem_have = total_ram();
692 mem_want = estimate_repack_memory(cfg, p);
693
694 /*
695 * Only allow 1/2 of memory for pack-objects, leave
696 * the rest for the OS and other processes in the
697 * system.
698 */
699 if (!mem_have || mem_want < mem_have / 2)
700 string_list_clear(&keep_pack, 0);
701 }
702
703 add_repack_all_option(cfg, &keep_pack);
704 string_list_clear(&keep_pack, 0);
705 } else if (too_many_loose_objects(cfg))
706 add_repack_incremental_option();
707 else
708 return 0;
709
710 if (run_hooks(the_repository, "pre-auto-gc"))
711 return 0;
712 return 1;
713}
714
715/* return NULL on success, else hostname running the gc */
716static const char *lock_repo_for_gc(int force, pid_t* ret_pid)
717{
718 struct lock_file lock = LOCK_INIT;
719 char my_host[HOST_NAME_MAX + 1];
720 struct strbuf sb = STRBUF_INIT;
721 struct stat st;
722 uintmax_t pid;
723 FILE *fp;
724 int fd;
725 char *pidfile_path;
726
727 if (is_tempfile_active(pidfile))
728 /* already locked */
729 return NULL;
730
731 if (xgethostname(my_host, sizeof(my_host)))
732 xsnprintf(my_host, sizeof(my_host), "unknown");
733
734 pidfile_path = repo_git_path(the_repository, "gc.pid");
735 fd = hold_lock_file_for_update(&lock, pidfile_path,
736 LOCK_DIE_ON_ERROR);
737 if (!force) {
738 static char locking_host[HOST_NAME_MAX + 1];
739 static char *scan_fmt;
740 int should_exit;
741
742 if (!scan_fmt)
743 scan_fmt = xstrfmt("%s %%%ds", "%"SCNuMAX, HOST_NAME_MAX);
744 fp = fopen(pidfile_path, "r");
745 memset(locking_host, 0, sizeof(locking_host));
746 should_exit =
747 fp != NULL &&
748 !fstat(fileno(fp), &st) &&
749 /*
750 * 12 hour limit is very generous as gc should
751 * never take that long. On the other hand we
752 * don't really need a strict limit here,
753 * running gc --auto one day late is not a big
754 * problem. --force can be used in manual gc
755 * after the user verifies that no gc is
756 * running.
757 */
758 time(NULL) - st.st_mtime <= 12 * 3600 &&
759 fscanf(fp, scan_fmt, &pid, locking_host) == 2 &&
760 /* be gentle to concurrent "gc" on remote hosts */
761 (strcmp(locking_host, my_host) || !kill(pid, 0) || errno == EPERM);
762 if (fp)
763 fclose(fp);
764 if (should_exit) {
765 if (fd >= 0)
766 rollback_lock_file(&lock);
767 *ret_pid = pid;
768 free(pidfile_path);
769 return locking_host;
770 }
771 }
772
773 strbuf_addf(&sb, "%"PRIuMAX" %s",
774 (uintmax_t) getpid(), my_host);
775 write_in_full(fd, sb.buf, sb.len);
776 strbuf_release(&sb);
777 commit_lock_file(&lock);
778 pidfile = register_tempfile(pidfile_path);
779 free(pidfile_path);
780 return NULL;
781}
782
783/*
784 * Returns 0 if there was no previous error and gc can proceed, 1 if
785 * gc should not proceed due to an error in the last run. Prints a
786 * message and returns with a non-[01] status code if an error occurred
787 * while reading gc.log
788 */
789static int report_last_gc_error(void)
790{
791 struct strbuf sb = STRBUF_INIT;
792 int ret = 0;
793 ssize_t len;
794 struct stat st;
795 char *gc_log_path = repo_git_path(the_repository, "gc.log");
796
797 if (stat(gc_log_path, &st)) {
798 if (errno == ENOENT)
799 goto done;
800
801 ret = die_message_errno(_("cannot stat '%s'"), gc_log_path);
802 goto done;
803 }
804
805 if (st.st_mtime < gc_log_expire_time)
806 goto done;
807
808 len = strbuf_read_file(&sb, gc_log_path, 0);
809 if (len < 0)
810 ret = die_message_errno(_("cannot read '%s'"), gc_log_path);
811 else if (len > 0) {
812 /*
813 * A previous gc failed. Report the error, and don't
814 * bother with an automatic gc run since it is likely
815 * to fail in the same way.
816 */
817 warning(_("The last gc run reported the following. "
818 "Please correct the root cause\n"
819 "and remove %s\n"
820 "Automatic cleanup will not be performed "
821 "until the file is removed.\n\n"
822 "%s"),
823 gc_log_path, sb.buf);
824 ret = 1;
825 }
826 strbuf_release(&sb);
827done:
828 free(gc_log_path);
829 return ret;
830}
831
832static int gc_foreground_tasks(struct maintenance_run_opts *opts,
833 struct gc_config *cfg)
834{
835 if (cfg->pack_refs && maintenance_task_pack_refs(opts, cfg))
836 return error(FAILED_RUN, "pack-refs");
837 if (cfg->prune_reflogs && maintenance_task_reflog_expire(opts, cfg))
838 return error(FAILED_RUN, "reflog");
839 return 0;
840}
841
842int cmd_gc(int argc,
843 const char **argv,
844 const char *prefix,
845 struct repository *repo UNUSED)
846{
847 int aggressive = 0;
848 int force = 0;
849 const char *name;
850 pid_t pid;
851 int daemonized = 0;
852 int keep_largest_pack = -1;
853 int skip_foreground_tasks = 0;
854 timestamp_t dummy;
855 struct maintenance_run_opts opts = MAINTENANCE_RUN_OPTS_INIT;
856 struct gc_config cfg = GC_CONFIG_INIT;
857 const char *prune_expire_sentinel = "sentinel";
858 const char *prune_expire_arg = prune_expire_sentinel;
859 int ret;
860 struct option builtin_gc_options[] = {
861 OPT__QUIET(&opts.quiet, N_("suppress progress reporting")),
862 {
863 .type = OPTION_STRING,
864 .long_name = "prune",
865 .value = &prune_expire_arg,
866 .argh = N_("date"),
867 .help = N_("prune unreferenced objects"),
868 .flags = PARSE_OPT_OPTARG,
869 .defval = (intptr_t)prune_expire_arg,
870 },
871 OPT_BOOL(0, "cruft", &cfg.cruft_packs, N_("pack unreferenced objects separately")),
872 OPT_UNSIGNED(0, "max-cruft-size", &cfg.max_cruft_size,
873 N_("with --cruft, limit the size of new cruft packs")),
874 OPT_BOOL(0, "aggressive", &aggressive, N_("be more thorough (increased runtime)")),
875 OPT_BOOL_F(0, "auto", &opts.auto_flag, N_("enable auto-gc mode"),
876 PARSE_OPT_NOCOMPLETE),
877 OPT_BOOL(0, "detach", &opts.detach,
878 N_("perform garbage collection in the background")),
879 OPT_BOOL_F(0, "force", &force,
880 N_("force running gc even if there may be another gc running"),
881 PARSE_OPT_NOCOMPLETE),
882 OPT_BOOL(0, "keep-largest-pack", &keep_largest_pack,
883 N_("repack all other packs except the largest pack")),
884 OPT_STRING(0, "expire-to", &cfg.repack_expire_to, N_("dir"),
885 N_("pack prefix to store a pack containing pruned objects")),
886 OPT_HIDDEN_BOOL(0, "skip-foreground-tasks", &skip_foreground_tasks,
887 N_("skip maintenance tasks typically done in the foreground")),
888 OPT_END()
889 };
890
891 show_usage_with_options_if_asked(argc, argv,
892 builtin_gc_usage, builtin_gc_options);
893
894 strvec_pushl(&repack, "repack", "-d", "-l", NULL);
895
896 gc_config(&cfg);
897
898 if (parse_expiry_date(cfg.gc_log_expire, &gc_log_expire_time))
899 die(_("failed to parse gc.logExpiry value %s"), cfg.gc_log_expire);
900
901 if (cfg.pack_refs < 0)
902 cfg.pack_refs = !is_bare_repository();
903
904 argc = parse_options(argc, argv, prefix, builtin_gc_options,
905 builtin_gc_usage, 0);
906 if (argc > 0)
907 usage_with_options(builtin_gc_usage, builtin_gc_options);
908
909 if (prune_expire_arg != prune_expire_sentinel) {
910 free(cfg.prune_expire);
911 cfg.prune_expire = xstrdup_or_null(prune_expire_arg);
912 }
913 if (cfg.prune_expire && parse_expiry_date(cfg.prune_expire, &dummy))
914 die(_("failed to parse prune expiry value %s"), cfg.prune_expire);
915
916 if (aggressive) {
917 strvec_push(&repack, "-f");
918 if (cfg.aggressive_depth > 0)
919 strvec_pushf(&repack, "--depth=%d", cfg.aggressive_depth);
920 if (cfg.aggressive_window > 0)
921 strvec_pushf(&repack, "--window=%d", cfg.aggressive_window);
922 }
923 if (opts.quiet)
924 strvec_push(&repack, "-q");
925
926 if (opts.auto_flag) {
927 if (cfg.detach_auto && opts.detach < 0)
928 opts.detach = 1;
929
930 /*
931 * Auto-gc should be least intrusive as possible.
932 */
933 if (!need_to_gc(&cfg)) {
934 ret = 0;
935 goto out;
936 }
937
938 if (!opts.quiet) {
939 if (opts.detach > 0)
940 fprintf(stderr, _("Auto packing the repository in background for optimum performance.\n"));
941 else
942 fprintf(stderr, _("Auto packing the repository for optimum performance.\n"));
943 fprintf(stderr, _("See \"git help gc\" for manual housekeeping.\n"));
944 }
945 } else {
946 struct string_list keep_pack = STRING_LIST_INIT_NODUP;
947
948 if (keep_largest_pack != -1) {
949 if (keep_largest_pack)
950 find_base_packs(&keep_pack, 0);
951 } else if (cfg.big_pack_threshold) {
952 find_base_packs(&keep_pack, cfg.big_pack_threshold);
953 }
954
955 add_repack_all_option(&cfg, &keep_pack);
956 string_list_clear(&keep_pack, 0);
957 }
958
959 if (opts.detach > 0) {
960 ret = report_last_gc_error();
961 if (ret == 1) {
962 /* Last gc --auto failed. Skip this one. */
963 ret = 0;
964 goto out;
965
966 } else if (ret) {
967 /* an I/O error occurred, already reported */
968 goto out;
969 }
970
971 if (!skip_foreground_tasks) {
972 if (lock_repo_for_gc(force, &pid)) {
973 ret = 0;
974 goto out;
975 }
976
977 if (gc_foreground_tasks(&opts, &cfg) < 0)
978 die(NULL);
979 delete_tempfile(&pidfile);
980 }
981
982 /*
983 * failure to daemonize is ok, we'll continue
984 * in foreground
985 */
986 daemonized = !daemonize();
987 }
988
989 name = lock_repo_for_gc(force, &pid);
990 if (name) {
991 if (opts.auto_flag) {
992 ret = 0;
993 goto out; /* be quiet on --auto */
994 }
995
996 die(_("gc is already running on machine '%s' pid %"PRIuMAX" (use --force if not)"),
997 name, (uintmax_t)pid);
998 }
999
1000 if (daemonized) {
1001 char *path = repo_git_path(the_repository, "gc.log");
1002 hold_lock_file_for_update(&log_lock, path,
1003 LOCK_DIE_ON_ERROR);
1004 dup2(get_lock_file_fd(&log_lock), 2);
1005 atexit(process_log_file_at_exit);
1006 free(path);
1007 }
1008
1009 if (opts.detach <= 0 && !skip_foreground_tasks)
1010 gc_foreground_tasks(&opts, &cfg);
1011
1012 if (!the_repository->repository_format_precious_objects) {
1013 struct child_process repack_cmd = CHILD_PROCESS_INIT;
1014
1015 repack_cmd.git_cmd = 1;
1016 repack_cmd.close_object_store = 1;
1017 strvec_pushv(&repack_cmd.args, repack.v);
1018 if (run_command(&repack_cmd))
1019 die(FAILED_RUN, repack.v[0]);
1020
1021 if (cfg.prune_expire) {
1022 struct child_process prune_cmd = CHILD_PROCESS_INIT;
1023
1024 strvec_pushl(&prune_cmd.args, "prune", "--expire", NULL);
1025 /* run `git prune` even if using cruft packs */
1026 strvec_push(&prune_cmd.args, cfg.prune_expire);
1027 if (opts.quiet)
1028 strvec_push(&prune_cmd.args, "--no-progress");
1029 if (repo_has_promisor_remote(the_repository))
1030 strvec_push(&prune_cmd.args,
1031 "--exclude-promisor-objects");
1032 prune_cmd.git_cmd = 1;
1033
1034 if (run_command(&prune_cmd))
1035 die(FAILED_RUN, prune_cmd.args.v[0]);
1036 }
1037 }
1038
1039 if (cfg.prune_worktrees_expire &&
1040 maintenance_task_worktree_prune(&opts, &cfg))
1041 die(FAILED_RUN, "worktree");
1042
1043 if (maintenance_task_rerere_gc(&opts, &cfg))
1044 die(FAILED_RUN, "rerere");
1045
1046 report_garbage = report_pack_garbage;
1047 odb_reprepare(the_repository->objects);
1048 if (pack_garbage.nr > 0) {
1049 close_object_store(the_repository->objects);
1050 clean_pack_garbage();
1051 }
1052
1053 if (the_repository->settings.gc_write_commit_graph == 1)
1054 write_commit_graph_reachable(the_repository->objects->sources,
1055 !opts.quiet && !daemonized ? COMMIT_GRAPH_WRITE_PROGRESS : 0,
1056 NULL);
1057
1058 if (opts.auto_flag && too_many_loose_objects(&cfg))
1059 warning(_("There are too many unreachable loose objects; "
1060 "run 'git prune' to remove them."));
1061
1062 if (!daemonized) {
1063 char *path = repo_git_path(the_repository, "gc.log");
1064 unlink(path);
1065 free(path);
1066 }
1067
1068out:
1069 maintenance_run_opts_release(&opts);
1070 gc_config_release(&cfg);
1071 return 0;
1072}
1073
1074static const char *const builtin_maintenance_run_usage[] = {
1075 N_("git maintenance run [--auto] [--[no-]quiet] [--task=<task>] [--schedule]"),
1076 NULL
1077};
1078
1079static int maintenance_opt_schedule(const struct option *opt, const char *arg,
1080 int unset)
1081{
1082 enum schedule_priority *priority = opt->value;
1083
1084 if (unset)
1085 die(_("--no-schedule is not allowed"));
1086
1087 *priority = parse_schedule(arg);
1088
1089 if (!*priority)
1090 die(_("unrecognized --schedule argument '%s'"), arg);
1091
1092 return 0;
1093}
1094
1095/* Remember to update object flag allocation in object.h */
1096#define SEEN (1u<<0)
1097
1098struct cg_auto_data {
1099 int num_not_in_graph;
1100 int limit;
1101};
1102
1103static int dfs_on_ref(const char *refname UNUSED,
1104 const char *referent UNUSED,
1105 const struct object_id *oid,
1106 int flags UNUSED,
1107 void *cb_data)
1108{
1109 struct cg_auto_data *data = (struct cg_auto_data *)cb_data;
1110 int result = 0;
1111 struct object_id peeled;
1112 struct commit_list *stack = NULL;
1113 struct commit *commit;
1114
1115 if (!peel_iterated_oid(the_repository, oid, &peeled))
1116 oid = &peeled;
1117 if (odb_read_object_info(the_repository->objects, oid, NULL) != OBJ_COMMIT)
1118 return 0;
1119
1120 commit = lookup_commit(the_repository, oid);
1121 if (!commit)
1122 return 0;
1123 if (repo_parse_commit(the_repository, commit) ||
1124 commit_graph_position(commit) != COMMIT_NOT_FROM_GRAPH)
1125 return 0;
1126
1127 data->num_not_in_graph++;
1128
1129 if (data->num_not_in_graph >= data->limit)
1130 return 1;
1131
1132 commit_list_append(commit, &stack);
1133
1134 while (!result && stack) {
1135 struct commit_list *parent;
1136
1137 commit = pop_commit(&stack);
1138
1139 for (parent = commit->parents; parent; parent = parent->next) {
1140 if (repo_parse_commit(the_repository, parent->item) ||
1141 commit_graph_position(parent->item) != COMMIT_NOT_FROM_GRAPH ||
1142 parent->item->object.flags & SEEN)
1143 continue;
1144
1145 parent->item->object.flags |= SEEN;
1146 data->num_not_in_graph++;
1147
1148 if (data->num_not_in_graph >= data->limit) {
1149 result = 1;
1150 break;
1151 }
1152
1153 commit_list_append(parent->item, &stack);
1154 }
1155 }
1156
1157 free_commit_list(stack);
1158 return result;
1159}
1160
1161static int should_write_commit_graph(struct gc_config *cfg UNUSED)
1162{
1163 int result;
1164 struct cg_auto_data data;
1165
1166 data.num_not_in_graph = 0;
1167 data.limit = 100;
1168 repo_config_get_int(the_repository, "maintenance.commit-graph.auto",
1169 &data.limit);
1170
1171 if (!data.limit)
1172 return 0;
1173 if (data.limit < 0)
1174 return 1;
1175
1176 result = refs_for_each_ref(get_main_ref_store(the_repository),
1177 dfs_on_ref, &data);
1178
1179 repo_clear_commit_marks(the_repository, SEEN);
1180
1181 return result;
1182}
1183
1184static int run_write_commit_graph(struct maintenance_run_opts *opts)
1185{
1186 struct child_process child = CHILD_PROCESS_INIT;
1187
1188 child.git_cmd = child.close_object_store = 1;
1189 strvec_pushl(&child.args, "commit-graph", "write",
1190 "--split", "--reachable", NULL);
1191
1192 if (opts->quiet)
1193 strvec_push(&child.args, "--no-progress");
1194 else
1195 strvec_push(&child.args, "--progress");
1196
1197 return !!run_command(&child);
1198}
1199
1200static int maintenance_task_commit_graph(struct maintenance_run_opts *opts,
1201 struct gc_config *cfg UNUSED)
1202{
1203 prepare_repo_settings(the_repository);
1204 if (!the_repository->settings.core_commit_graph)
1205 return 0;
1206
1207 if (run_write_commit_graph(opts)) {
1208 error(_("failed to write commit-graph"));
1209 return 1;
1210 }
1211
1212 return 0;
1213}
1214
1215static int fetch_remote(struct remote *remote, void *cbdata)
1216{
1217 struct maintenance_run_opts *opts = cbdata;
1218 struct child_process child = CHILD_PROCESS_INIT;
1219
1220 if (remote->skip_default_update)
1221 return 0;
1222
1223 child.git_cmd = 1;
1224 strvec_pushl(&child.args, "fetch", remote->name,
1225 "--prefetch", "--prune", "--no-tags",
1226 "--no-write-fetch-head", "--recurse-submodules=no",
1227 NULL);
1228
1229 if (opts->quiet)
1230 strvec_push(&child.args, "--quiet");
1231
1232 return !!run_command(&child);
1233}
1234
1235static int maintenance_task_prefetch(struct maintenance_run_opts *opts,
1236 struct gc_config *cfg UNUSED)
1237{
1238 if (for_each_remote(fetch_remote, opts)) {
1239 error(_("failed to prefetch remotes"));
1240 return 1;
1241 }
1242
1243 return 0;
1244}
1245
1246static int maintenance_task_gc_foreground(struct maintenance_run_opts *opts,
1247 struct gc_config *cfg)
1248{
1249 return gc_foreground_tasks(opts, cfg);
1250}
1251
1252static int maintenance_task_gc_background(struct maintenance_run_opts *opts,
1253 struct gc_config *cfg UNUSED)
1254{
1255 struct child_process child = CHILD_PROCESS_INIT;
1256
1257 child.git_cmd = child.close_object_store = 1;
1258 strvec_push(&child.args, "gc");
1259
1260 if (opts->auto_flag)
1261 strvec_push(&child.args, "--auto");
1262 if (opts->quiet)
1263 strvec_push(&child.args, "--quiet");
1264 else
1265 strvec_push(&child.args, "--no-quiet");
1266 strvec_push(&child.args, "--no-detach");
1267 strvec_push(&child.args, "--skip-foreground-tasks");
1268
1269 return run_command(&child);
1270}
1271
1272static int prune_packed(struct maintenance_run_opts *opts)
1273{
1274 struct child_process child = CHILD_PROCESS_INIT;
1275
1276 child.git_cmd = 1;
1277 strvec_push(&child.args, "prune-packed");
1278
1279 if (opts->quiet)
1280 strvec_push(&child.args, "--quiet");
1281
1282 return !!run_command(&child);
1283}
1284
1285struct write_loose_object_data {
1286 FILE *in;
1287 int count;
1288 int batch_size;
1289};
1290
1291static int loose_object_auto_limit = 100;
1292
1293static int loose_object_count(const struct object_id *oid UNUSED,
1294 const char *path UNUSED,
1295 void *data)
1296{
1297 int *count = (int*)data;
1298 if (++(*count) >= loose_object_auto_limit)
1299 return 1;
1300 return 0;
1301}
1302
1303static int loose_object_auto_condition(struct gc_config *cfg UNUSED)
1304{
1305 int count = 0;
1306
1307 repo_config_get_int(the_repository, "maintenance.loose-objects.auto",
1308 &loose_object_auto_limit);
1309
1310 if (!loose_object_auto_limit)
1311 return 0;
1312 if (loose_object_auto_limit < 0)
1313 return 1;
1314
1315 return for_each_loose_file_in_source(the_repository->objects->sources,
1316 loose_object_count,
1317 NULL, NULL, &count);
1318}
1319
1320static int bail_on_loose(const struct object_id *oid UNUSED,
1321 const char *path UNUSED,
1322 void *data UNUSED)
1323{
1324 return 1;
1325}
1326
1327static int write_loose_object_to_stdin(const struct object_id *oid,
1328 const char *path UNUSED,
1329 void *data)
1330{
1331 struct write_loose_object_data *d = (struct write_loose_object_data *)data;
1332
1333 fprintf(d->in, "%s\n", oid_to_hex(oid));
1334
1335 /* If batch_size is INT_MAX, then this will return 0 always. */
1336 return ++(d->count) > d->batch_size;
1337}
1338
1339static int pack_loose(struct maintenance_run_opts *opts)
1340{
1341 struct repository *r = the_repository;
1342 int result = 0;
1343 struct write_loose_object_data data;
1344 struct child_process pack_proc = CHILD_PROCESS_INIT;
1345
1346 /*
1347 * Do not start pack-objects process
1348 * if there are no loose objects.
1349 */
1350 if (!for_each_loose_file_in_source(r->objects->sources,
1351 bail_on_loose,
1352 NULL, NULL, NULL))
1353 return 0;
1354
1355 pack_proc.git_cmd = 1;
1356
1357 strvec_push(&pack_proc.args, "pack-objects");
1358 if (opts->quiet)
1359 strvec_push(&pack_proc.args, "--quiet");
1360 else
1361 strvec_push(&pack_proc.args, "--no-quiet");
1362 strvec_pushf(&pack_proc.args, "%s/pack/loose", r->objects->sources->path);
1363
1364 pack_proc.in = -1;
1365
1366 /*
1367 * git-pack-objects(1) ends up writing the pack hash to stdout, which
1368 * we do not care for.
1369 */
1370 pack_proc.out = -1;
1371
1372 if (start_command(&pack_proc)) {
1373 error(_("failed to start 'git pack-objects' process"));
1374 return 1;
1375 }
1376
1377 data.in = xfdopen(pack_proc.in, "w");
1378 data.count = 0;
1379 data.batch_size = 50000;
1380
1381 repo_config_get_int(r, "maintenance.loose-objects.batchSize",
1382 &data.batch_size);
1383
1384 /* If configured as 0, then remove limit. */
1385 if (!data.batch_size)
1386 data.batch_size = INT_MAX;
1387 else if (data.batch_size > 0)
1388 data.batch_size--; /* Decrease for equality on limit. */
1389
1390 for_each_loose_file_in_source(r->objects->sources,
1391 write_loose_object_to_stdin,
1392 NULL, NULL, &data);
1393
1394 fclose(data.in);
1395
1396 if (finish_command(&pack_proc)) {
1397 error(_("failed to finish 'git pack-objects' process"));
1398 result = 1;
1399 }
1400
1401 return result;
1402}
1403
1404static int maintenance_task_loose_objects(struct maintenance_run_opts *opts,
1405 struct gc_config *cfg UNUSED)
1406{
1407 return prune_packed(opts) || pack_loose(opts);
1408}
1409
1410static int incremental_repack_auto_condition(struct gc_config *cfg UNUSED)
1411{
1412 struct packed_git *p;
1413 int incremental_repack_auto_limit = 10;
1414 int count = 0;
1415
1416 prepare_repo_settings(the_repository);
1417 if (!the_repository->settings.core_multi_pack_index)
1418 return 0;
1419
1420 repo_config_get_int(the_repository, "maintenance.incremental-repack.auto",
1421 &incremental_repack_auto_limit);
1422
1423 if (!incremental_repack_auto_limit)
1424 return 0;
1425 if (incremental_repack_auto_limit < 0)
1426 return 1;
1427
1428 for (p = packfile_store_get_packs(the_repository->objects->packfiles);
1429 count < incremental_repack_auto_limit && p;
1430 p = p->next) {
1431 if (!p->multi_pack_index)
1432 count++;
1433 }
1434
1435 return count >= incremental_repack_auto_limit;
1436}
1437
1438static int multi_pack_index_write(struct maintenance_run_opts *opts)
1439{
1440 struct child_process child = CHILD_PROCESS_INIT;
1441
1442 child.git_cmd = 1;
1443 strvec_pushl(&child.args, "multi-pack-index", "write", NULL);
1444
1445 if (opts->quiet)
1446 strvec_push(&child.args, "--no-progress");
1447 else
1448 strvec_push(&child.args, "--progress");
1449
1450 if (run_command(&child))
1451 return error(_("failed to write multi-pack-index"));
1452
1453 return 0;
1454}
1455
1456static int multi_pack_index_expire(struct maintenance_run_opts *opts)
1457{
1458 struct child_process child = CHILD_PROCESS_INIT;
1459
1460 child.git_cmd = child.close_object_store = 1;
1461 strvec_pushl(&child.args, "multi-pack-index", "expire", NULL);
1462
1463 if (opts->quiet)
1464 strvec_push(&child.args, "--no-progress");
1465 else
1466 strvec_push(&child.args, "--progress");
1467
1468 if (run_command(&child))
1469 return error(_("'git multi-pack-index expire' failed"));
1470
1471 return 0;
1472}
1473
1474#define TWO_GIGABYTES (INT32_MAX)
1475
1476static off_t get_auto_pack_size(void)
1477{
1478 /*
1479 * The "auto" value is special: we optimize for
1480 * one large pack-file (i.e. from a clone) and
1481 * expect the rest to be small and they can be
1482 * repacked quickly.
1483 *
1484 * The strategy we select here is to select a
1485 * size that is one more than the second largest
1486 * pack-file. This ensures that we will repack
1487 * at least two packs if there are three or more
1488 * packs.
1489 */
1490 off_t max_size = 0;
1491 off_t second_largest_size = 0;
1492 off_t result_size;
1493 struct packed_git *p;
1494 struct repository *r = the_repository;
1495
1496 odb_reprepare(r->objects);
1497 for (p = packfile_store_get_all_packs(r->objects->packfiles); p; p = p->next) {
1498 if (p->pack_size > max_size) {
1499 second_largest_size = max_size;
1500 max_size = p->pack_size;
1501 } else if (p->pack_size > second_largest_size)
1502 second_largest_size = p->pack_size;
1503 }
1504
1505 result_size = second_largest_size + 1;
1506
1507 /* But limit ourselves to a batch size of 2g */
1508 if (result_size > TWO_GIGABYTES)
1509 result_size = TWO_GIGABYTES;
1510
1511 return result_size;
1512}
1513
1514static int multi_pack_index_repack(struct maintenance_run_opts *opts)
1515{
1516 struct child_process child = CHILD_PROCESS_INIT;
1517
1518 child.git_cmd = child.close_object_store = 1;
1519 strvec_pushl(&child.args, "multi-pack-index", "repack", NULL);
1520
1521 if (opts->quiet)
1522 strvec_push(&child.args, "--no-progress");
1523 else
1524 strvec_push(&child.args, "--progress");
1525
1526 strvec_pushf(&child.args, "--batch-size=%"PRIuMAX,
1527 (uintmax_t)get_auto_pack_size());
1528
1529 if (run_command(&child))
1530 return error(_("'git multi-pack-index repack' failed"));
1531
1532 return 0;
1533}
1534
1535static int maintenance_task_incremental_repack(struct maintenance_run_opts *opts,
1536 struct gc_config *cfg UNUSED)
1537{
1538 prepare_repo_settings(the_repository);
1539 if (!the_repository->settings.core_multi_pack_index) {
1540 warning(_("skipping incremental-repack task because core.multiPackIndex is disabled"));
1541 return 0;
1542 }
1543
1544 if (multi_pack_index_write(opts))
1545 return 1;
1546 if (multi_pack_index_expire(opts))
1547 return 1;
1548 if (multi_pack_index_repack(opts))
1549 return 1;
1550 return 0;
1551}
1552
1553typedef int (*maintenance_task_fn)(struct maintenance_run_opts *opts,
1554 struct gc_config *cfg);
1555typedef int (*maintenance_auto_fn)(struct gc_config *cfg);
1556
1557struct maintenance_task {
1558 const char *name;
1559
1560 /*
1561 * Work that will be executed before detaching. This should not include
1562 * tasks that may run for an extended amount of time as it does cause
1563 * auto-maintenance to block until foreground tasks have been run.
1564 */
1565 maintenance_task_fn foreground;
1566
1567 /*
1568 * Work that will be executed after detaching. When not detaching the
1569 * work will be run in the foreground, as well.
1570 */
1571 maintenance_task_fn background;
1572
1573 /*
1574 * An auto condition function returns 1 if the task should run and 0 if
1575 * the task should NOT run. See needs_to_gc() for an example.
1576 */
1577 maintenance_auto_fn auto_condition;
1578};
1579
1580static const struct maintenance_task tasks[] = {
1581 [TASK_PREFETCH] = {
1582 .name = "prefetch",
1583 .background = maintenance_task_prefetch,
1584 },
1585 [TASK_LOOSE_OBJECTS] = {
1586 .name = "loose-objects",
1587 .background = maintenance_task_loose_objects,
1588 .auto_condition = loose_object_auto_condition,
1589 },
1590 [TASK_INCREMENTAL_REPACK] = {
1591 .name = "incremental-repack",
1592 .background = maintenance_task_incremental_repack,
1593 .auto_condition = incremental_repack_auto_condition,
1594 },
1595 [TASK_GC] = {
1596 .name = "gc",
1597 .foreground = maintenance_task_gc_foreground,
1598 .background = maintenance_task_gc_background,
1599 .auto_condition = need_to_gc,
1600 },
1601 [TASK_COMMIT_GRAPH] = {
1602 .name = "commit-graph",
1603 .background = maintenance_task_commit_graph,
1604 .auto_condition = should_write_commit_graph,
1605 },
1606 [TASK_PACK_REFS] = {
1607 .name = "pack-refs",
1608 .foreground = maintenance_task_pack_refs,
1609 .auto_condition = pack_refs_condition,
1610 },
1611 [TASK_REFLOG_EXPIRE] = {
1612 .name = "reflog-expire",
1613 .foreground = maintenance_task_reflog_expire,
1614 .auto_condition = reflog_expire_condition,
1615 },
1616 [TASK_WORKTREE_PRUNE] = {
1617 .name = "worktree-prune",
1618 .background = maintenance_task_worktree_prune,
1619 .auto_condition = worktree_prune_condition,
1620 },
1621 [TASK_RERERE_GC] = {
1622 .name = "rerere-gc",
1623 .background = maintenance_task_rerere_gc,
1624 .auto_condition = rerere_gc_condition,
1625 },
1626};
1627
1628enum task_phase {
1629 TASK_PHASE_FOREGROUND,
1630 TASK_PHASE_BACKGROUND,
1631};
1632
1633static int maybe_run_task(const struct maintenance_task *task,
1634 struct repository *repo,
1635 struct maintenance_run_opts *opts,
1636 struct gc_config *cfg,
1637 enum task_phase phase)
1638{
1639 int foreground = (phase == TASK_PHASE_FOREGROUND);
1640 maintenance_task_fn fn = foreground ? task->foreground : task->background;
1641 const char *region = foreground ? "maintenance foreground" : "maintenance";
1642 int ret = 0;
1643
1644 if (!fn)
1645 return 0;
1646 if (opts->auto_flag &&
1647 (!task->auto_condition || !task->auto_condition(cfg)))
1648 return 0;
1649
1650 trace2_region_enter(region, task->name, repo);
1651 if (fn(opts, cfg)) {
1652 error(_("task '%s' failed"), task->name);
1653 ret = 1;
1654 }
1655 trace2_region_leave(region, task->name, repo);
1656
1657 return ret;
1658}
1659
1660static int maintenance_run_tasks(struct maintenance_run_opts *opts,
1661 struct gc_config *cfg)
1662{
1663 int result = 0;
1664 struct lock_file lk;
1665 struct repository *r = the_repository;
1666 char *lock_path = xstrfmt("%s/maintenance", r->objects->sources->path);
1667
1668 if (hold_lock_file_for_update(&lk, lock_path, LOCK_NO_DEREF) < 0) {
1669 /*
1670 * Another maintenance command is running.
1671 *
1672 * If --auto was provided, then it is likely due to a
1673 * recursive process stack. Do not report an error in
1674 * that case.
1675 */
1676 if (!opts->auto_flag && !opts->quiet)
1677 warning(_("lock file '%s' exists, skipping maintenance"),
1678 lock_path);
1679 free(lock_path);
1680 return 0;
1681 }
1682 free(lock_path);
1683
1684 for (size_t i = 0; i < opts->tasks_nr; i++)
1685 if (maybe_run_task(&tasks[opts->tasks[i]], r, opts, cfg,
1686 TASK_PHASE_FOREGROUND))
1687 result = 1;
1688
1689 /* Failure to daemonize is ok, we'll continue in foreground. */
1690 if (opts->detach > 0) {
1691 trace2_region_enter("maintenance", "detach", the_repository);
1692 daemonize();
1693 trace2_region_leave("maintenance", "detach", the_repository);
1694 }
1695
1696 for (size_t i = 0; i < opts->tasks_nr; i++)
1697 if (maybe_run_task(&tasks[opts->tasks[i]], r, opts, cfg,
1698 TASK_PHASE_BACKGROUND))
1699 result = 1;
1700
1701 rollback_lock_file(&lk);
1702 return result;
1703}
1704
1705struct maintenance_strategy {
1706 struct {
1707 int enabled;
1708 enum schedule_priority schedule;
1709 } tasks[TASK__COUNT];
1710};
1711
1712static const struct maintenance_strategy none_strategy = { 0 };
1713static const struct maintenance_strategy default_strategy = {
1714 .tasks = {
1715 [TASK_GC].enabled = 1,
1716 },
1717};
1718static const struct maintenance_strategy incremental_strategy = {
1719 .tasks = {
1720 [TASK_COMMIT_GRAPH].enabled = 1,
1721 [TASK_COMMIT_GRAPH].schedule = SCHEDULE_HOURLY,
1722 [TASK_PREFETCH].enabled = 1,
1723 [TASK_PREFETCH].schedule = SCHEDULE_HOURLY,
1724 [TASK_INCREMENTAL_REPACK].enabled = 1,
1725 [TASK_INCREMENTAL_REPACK].schedule = SCHEDULE_DAILY,
1726 [TASK_LOOSE_OBJECTS].enabled = 1,
1727 [TASK_LOOSE_OBJECTS].schedule = SCHEDULE_DAILY,
1728 [TASK_PACK_REFS].enabled = 1,
1729 [TASK_PACK_REFS].schedule = SCHEDULE_WEEKLY,
1730 },
1731};
1732
1733static void initialize_task_config(struct maintenance_run_opts *opts,
1734 const struct string_list *selected_tasks)
1735{
1736 struct strbuf config_name = STRBUF_INIT;
1737 struct maintenance_strategy strategy;
1738 const char *config_str;
1739
1740 /*
1741 * In case the user has asked us to run tasks explicitly we only use
1742 * those specified tasks. Specifically, we do _not_ want to consult the
1743 * config or maintenance strategy.
1744 */
1745 if (selected_tasks->nr) {
1746 for (size_t i = 0; i < selected_tasks->nr; i++) {
1747 enum maintenance_task_label label = (intptr_t)selected_tasks->items[i].util;;
1748 ALLOC_GROW(opts->tasks, opts->tasks_nr + 1, opts->tasks_alloc);
1749 opts->tasks[opts->tasks_nr++] = label;
1750 }
1751
1752 return;
1753 }
1754
1755 /*
1756 * Otherwise, the strategy depends on whether we run as part of a
1757 * scheduled job or not:
1758 *
1759 * - Scheduled maintenance does not perform any housekeeping by
1760 * default, but requires the user to pick a maintenance strategy.
1761 *
1762 * - Unscheduled maintenance uses our default strategy.
1763 *
1764 * Both of these are affected by the gitconfig though, which may
1765 * override specific aspects of our strategy.
1766 */
1767 if (opts->schedule) {
1768 strategy = none_strategy;
1769
1770 if (!repo_config_get_string_tmp(the_repository, "maintenance.strategy", &config_str)) {
1771 if (!strcasecmp(config_str, "incremental"))
1772 strategy = incremental_strategy;
1773 }
1774 } else {
1775 strategy = default_strategy;
1776 }
1777
1778 for (size_t i = 0; i < TASK__COUNT; i++) {
1779 int config_value;
1780
1781 strbuf_reset(&config_name);
1782 strbuf_addf(&config_name, "maintenance.%s.enabled",
1783 tasks[i].name);
1784 if (!repo_config_get_bool(the_repository, config_name.buf, &config_value))
1785 strategy.tasks[i].enabled = config_value;
1786 if (!strategy.tasks[i].enabled)
1787 continue;
1788
1789 if (opts->schedule) {
1790 strbuf_reset(&config_name);
1791 strbuf_addf(&config_name, "maintenance.%s.schedule",
1792 tasks[i].name);
1793 if (!repo_config_get_string_tmp(the_repository, config_name.buf, &config_str))
1794 strategy.tasks[i].schedule = parse_schedule(config_str);
1795 if (strategy.tasks[i].schedule < opts->schedule)
1796 continue;
1797 }
1798
1799 ALLOC_GROW(opts->tasks, opts->tasks_nr + 1, opts->tasks_alloc);
1800 opts->tasks[opts->tasks_nr++] = i;
1801 }
1802
1803 strbuf_release(&config_name);
1804}
1805
1806static int task_option_parse(const struct option *opt,
1807 const char *arg, int unset)
1808{
1809 struct string_list *selected_tasks = opt->value;
1810 size_t i;
1811
1812 BUG_ON_OPT_NEG(unset);
1813
1814 for (i = 0; i < TASK__COUNT; i++)
1815 if (!strcasecmp(tasks[i].name, arg))
1816 break;
1817 if (i >= TASK__COUNT) {
1818 error(_("'%s' is not a valid task"), arg);
1819 return 1;
1820 }
1821
1822 if (unsorted_string_list_has_string(selected_tasks, arg)) {
1823 error(_("task '%s' cannot be selected multiple times"), arg);
1824 return 1;
1825 }
1826
1827 string_list_append(selected_tasks, arg)->util = (void *)(intptr_t)i;
1828
1829 return 0;
1830}
1831
1832static int maintenance_run(int argc, const char **argv, const char *prefix,
1833 struct repository *repo UNUSED)
1834{
1835 struct maintenance_run_opts opts = MAINTENANCE_RUN_OPTS_INIT;
1836 struct string_list selected_tasks = STRING_LIST_INIT_DUP;
1837 struct gc_config cfg = GC_CONFIG_INIT;
1838 struct option builtin_maintenance_run_options[] = {
1839 OPT_BOOL(0, "auto", &opts.auto_flag,
1840 N_("run tasks based on the state of the repository")),
1841 OPT_BOOL(0, "detach", &opts.detach,
1842 N_("perform maintenance in the background")),
1843 OPT_CALLBACK(0, "schedule", &opts.schedule, N_("frequency"),
1844 N_("run tasks based on frequency"),
1845 maintenance_opt_schedule),
1846 OPT_BOOL(0, "quiet", &opts.quiet,
1847 N_("do not report progress or other information over stderr")),
1848 OPT_CALLBACK_F(0, "task", &selected_tasks, N_("task"),
1849 N_("run a specific task"),
1850 PARSE_OPT_NONEG, task_option_parse),
1851 OPT_END()
1852 };
1853 int ret;
1854
1855 opts.quiet = !isatty(2);
1856
1857 argc = parse_options(argc, argv, prefix,
1858 builtin_maintenance_run_options,
1859 builtin_maintenance_run_usage,
1860 PARSE_OPT_STOP_AT_NON_OPTION);
1861
1862 die_for_incompatible_opt2(opts.auto_flag, "--auto",
1863 opts.schedule, "--schedule=");
1864 die_for_incompatible_opt2(selected_tasks.nr, "--task=",
1865 opts.schedule, "--schedule=");
1866
1867 gc_config(&cfg);
1868 initialize_task_config(&opts, &selected_tasks);
1869
1870 if (argc != 0)
1871 usage_with_options(builtin_maintenance_run_usage,
1872 builtin_maintenance_run_options);
1873
1874 ret = maintenance_run_tasks(&opts, &cfg);
1875
1876 string_list_clear(&selected_tasks, 0);
1877 maintenance_run_opts_release(&opts);
1878 gc_config_release(&cfg);
1879 return ret;
1880}
1881
1882static char *get_maintpath(void)
1883{
1884 struct strbuf sb = STRBUF_INIT;
1885 const char *p = the_repository->worktree ?
1886 the_repository->worktree : the_repository->gitdir;
1887
1888 strbuf_realpath(&sb, p, 1);
1889 return strbuf_detach(&sb, NULL);
1890}
1891
1892static char const * const builtin_maintenance_register_usage[] = {
1893 "git maintenance register [--config-file <path>]",
1894 NULL
1895};
1896
1897static int maintenance_register(int argc, const char **argv, const char *prefix,
1898 struct repository *repo UNUSED)
1899{
1900 char *config_file = NULL;
1901 struct option options[] = {
1902 OPT_STRING(0, "config-file", &config_file, N_("file"), N_("use given config file")),
1903 OPT_END(),
1904 };
1905 int found = 0;
1906 const char *key = "maintenance.repo";
1907 char *maintpath = get_maintpath();
1908 struct string_list_item *item;
1909 const struct string_list *list;
1910
1911 argc = parse_options(argc, argv, prefix, options,
1912 builtin_maintenance_register_usage, 0);
1913 if (argc)
1914 usage_with_options(builtin_maintenance_register_usage,
1915 options);
1916
1917 /* Disable foreground maintenance */
1918 repo_config_set(the_repository, "maintenance.auto", "false");
1919
1920 /* Set maintenance strategy, if unset */
1921 if (repo_config_get(the_repository, "maintenance.strategy"))
1922 repo_config_set(the_repository, "maintenance.strategy", "incremental");
1923
1924 if (!repo_config_get_string_multi(the_repository, key, &list)) {
1925 for_each_string_list_item(item, list) {
1926 if (!strcmp(maintpath, item->string)) {
1927 found = 1;
1928 break;
1929 }
1930 }
1931 }
1932
1933 if (!found) {
1934 int rc;
1935 char *global_config_file = NULL;
1936
1937 if (!config_file) {
1938 global_config_file = git_global_config();
1939 config_file = global_config_file;
1940 }
1941 if (!config_file)
1942 die(_("$HOME not set"));
1943 rc = repo_config_set_multivar_in_file_gently(the_repository,
1944 config_file, "maintenance.repo", maintpath,
1945 CONFIG_REGEX_NONE, NULL, 0);
1946 free(global_config_file);
1947
1948 if (rc)
1949 die(_("unable to add '%s' value of '%s'"),
1950 key, maintpath);
1951 }
1952
1953 free(maintpath);
1954 return 0;
1955}
1956
1957static char const * const builtin_maintenance_unregister_usage[] = {
1958 "git maintenance unregister [--config-file <path>] [--force]",
1959 NULL
1960};
1961
1962static int maintenance_unregister(int argc, const char **argv, const char *prefix,
1963 struct repository *repo UNUSED)
1964{
1965 int force = 0;
1966 char *config_file = NULL;
1967 struct option options[] = {
1968 OPT_STRING(0, "config-file", &config_file, N_("file"), N_("use given config file")),
1969 OPT__FORCE(&force,
1970 N_("return success even if repository was not registered"),
1971 PARSE_OPT_NOCOMPLETE),
1972 OPT_END(),
1973 };
1974 const char *key = "maintenance.repo";
1975 char *maintpath = get_maintpath();
1976 int found = 0;
1977 struct string_list_item *item;
1978 const struct string_list *list;
1979 struct config_set cs = { { 0 } };
1980
1981 argc = parse_options(argc, argv, prefix, options,
1982 builtin_maintenance_unregister_usage, 0);
1983 if (argc)
1984 usage_with_options(builtin_maintenance_unregister_usage,
1985 options);
1986
1987 if (config_file) {
1988 git_configset_init(&cs);
1989 git_configset_add_file(&cs, config_file);
1990 }
1991 if (!(config_file
1992 ? git_configset_get_string_multi(&cs, key, &list)
1993 : repo_config_get_string_multi(the_repository, key, &list))) {
1994 for_each_string_list_item(item, list) {
1995 if (!strcmp(maintpath, item->string)) {
1996 found = 1;
1997 break;
1998 }
1999 }
2000 }
2001
2002 if (found) {
2003 int rc;
2004 char *global_config_file = NULL;
2005
2006 if (!config_file) {
2007 global_config_file = git_global_config();
2008 config_file = global_config_file;
2009 }
2010 if (!config_file)
2011 die(_("$HOME not set"));
2012 rc = repo_config_set_multivar_in_file_gently(the_repository,
2013 config_file, key, NULL, maintpath, NULL,
2014 CONFIG_FLAGS_MULTI_REPLACE | CONFIG_FLAGS_FIXED_VALUE);
2015 free(global_config_file);
2016
2017 if (rc &&
2018 (!force || rc == CONFIG_NOTHING_SET))
2019 die(_("unable to unset '%s' value of '%s'"),
2020 key, maintpath);
2021 } else if (!force) {
2022 die(_("repository '%s' is not registered"), maintpath);
2023 }
2024
2025 git_configset_clear(&cs);
2026 free(maintpath);
2027 return 0;
2028}
2029
2030static const char *get_frequency(enum schedule_priority schedule)
2031{
2032 switch (schedule) {
2033 case SCHEDULE_HOURLY:
2034 return "hourly";
2035 case SCHEDULE_DAILY:
2036 return "daily";
2037 case SCHEDULE_WEEKLY:
2038 return "weekly";
2039 default:
2040 BUG("invalid schedule %d", schedule);
2041 }
2042}
2043
2044static const char *extraconfig[] = {
2045 "credential.interactive=false",
2046 "core.askPass=true", /* 'true' returns success, but no output. */
2047 NULL
2048};
2049
2050static const char *get_extra_config_parameters(void) {
2051 static const char *result = NULL;
2052 struct strbuf builder = STRBUF_INIT;
2053
2054 if (result)
2055 return result;
2056
2057 for (const char **s = extraconfig; s && *s; s++)
2058 strbuf_addf(&builder, "-c %s ", *s);
2059
2060 result = strbuf_detach(&builder, NULL);
2061 return result;
2062}
2063
2064static const char *get_extra_launchctl_strings(void) {
2065 static const char *result = NULL;
2066 struct strbuf builder = STRBUF_INIT;
2067
2068 if (result)
2069 return result;
2070
2071 for (const char **s = extraconfig; s && *s; s++) {
2072 strbuf_addstr(&builder, "<string>-c</string>\n");
2073 strbuf_addf(&builder, "<string>%s</string>\n", *s);
2074 }
2075
2076 result = strbuf_detach(&builder, NULL);
2077 return result;
2078}
2079
2080/*
2081 * get_schedule_cmd` reads the GIT_TEST_MAINT_SCHEDULER environment variable
2082 * to mock the schedulers that `git maintenance start` rely on.
2083 *
2084 * For test purpose, GIT_TEST_MAINT_SCHEDULER can be set to a comma-separated
2085 * list of colon-separated key/value pairs where each pair contains a scheduler
2086 * and its corresponding mock.
2087 *
2088 * * If $GIT_TEST_MAINT_SCHEDULER is not set, return false and leave the
2089 * arguments unmodified.
2090 *
2091 * * If $GIT_TEST_MAINT_SCHEDULER is set, return true.
2092 * In this case, the *cmd value is read as input.
2093 *
2094 * * if the input value cmd is the key of one of the comma-separated list
2095 * item, then *is_available is set to true and *out is set to
2096 * the mock command.
2097 *
2098 * * if the input value *cmd isn’t the key of any of the comma-separated list
2099 * item, then *is_available is set to false and *out is set to the original
2100 * command.
2101 *
2102 * Ex.:
2103 * GIT_TEST_MAINT_SCHEDULER not set
2104 * +-------+-------------------------------------------------+
2105 * | Input | Output |
2106 * | *cmd | return code | *out | *is_available |
2107 * +-------+-------------+-------------------+---------------+
2108 * | "foo" | false | "foo" (allocated) | (unchanged) |
2109 * +-------+-------------+-------------------+---------------+
2110 *
2111 * GIT_TEST_MAINT_SCHEDULER set to “foo:./mock_foo.sh,bar:./mock_bar.sh”
2112 * +-------+-------------------------------------------------+
2113 * | Input | Output |
2114 * | *cmd | return code | *out | *is_available |
2115 * +-------+-------------+-------------------+---------------+
2116 * | "foo" | true | "./mock.foo.sh" | true |
2117 * | "qux" | true | "qux" (allocated) | false |
2118 * +-------+-------------+-------------------+---------------+
2119 */
2120static int get_schedule_cmd(const char *cmd, int *is_available, char **out)
2121{
2122 char *testing = xstrdup_or_null(getenv("GIT_TEST_MAINT_SCHEDULER"));
2123 struct string_list_item *item;
2124 struct string_list list = STRING_LIST_INIT_NODUP;
2125
2126 if (!testing) {
2127 if (out)
2128 *out = xstrdup(cmd);
2129 return 0;
2130 }
2131
2132 if (is_available)
2133 *is_available = 0;
2134
2135 string_list_split_in_place(&list, testing, ",", -1);
2136 for_each_string_list_item(item, &list) {
2137 struct string_list pair = STRING_LIST_INIT_NODUP;
2138
2139 if (string_list_split_in_place(&pair, item->string, ":", 2) != 2)
2140 continue;
2141
2142 if (!strcmp(cmd, pair.items[0].string)) {
2143 if (out)
2144 *out = xstrdup(pair.items[1].string);
2145 if (is_available)
2146 *is_available = 1;
2147 string_list_clear(&pair, 0);
2148 goto out;
2149 }
2150
2151 string_list_clear(&pair, 0);
2152 }
2153
2154 if (out)
2155 *out = xstrdup(cmd);
2156
2157out:
2158 string_list_clear(&list, 0);
2159 free(testing);
2160 return 1;
2161}
2162
2163static int get_random_minute(void)
2164{
2165 /* Use a static value when under tests. */
2166 if (getenv("GIT_TEST_MAINT_SCHEDULER"))
2167 return 13;
2168
2169 return git_rand(0) % 60;
2170}
2171
2172static int is_launchctl_available(void)
2173{
2174 int is_available;
2175 if (get_schedule_cmd("launchctl", &is_available, NULL))
2176 return is_available;
2177
2178#ifdef __APPLE__
2179 return 1;
2180#else
2181 return 0;
2182#endif
2183}
2184
2185static char *launchctl_service_name(const char *frequency)
2186{
2187 struct strbuf label = STRBUF_INIT;
2188 strbuf_addf(&label, "org.git-scm.git.%s", frequency);
2189 return strbuf_detach(&label, NULL);
2190}
2191
2192static char *launchctl_service_filename(const char *name)
2193{
2194 char *expanded;
2195 struct strbuf filename = STRBUF_INIT;
2196 strbuf_addf(&filename, "~/Library/LaunchAgents/%s.plist", name);
2197
2198 expanded = interpolate_path(filename.buf, 1);
2199 if (!expanded)
2200 die(_("failed to expand path '%s'"), filename.buf);
2201
2202 strbuf_release(&filename);
2203 return expanded;
2204}
2205
2206static char *launchctl_get_uid(void)
2207{
2208 return xstrfmt("gui/%d", getuid());
2209}
2210
2211static int launchctl_boot_plist(int enable, const char *filename)
2212{
2213 char *cmd;
2214 int result;
2215 struct child_process child = CHILD_PROCESS_INIT;
2216 char *uid = launchctl_get_uid();
2217
2218 get_schedule_cmd("launchctl", NULL, &cmd);
2219 strvec_split(&child.args, cmd);
2220 strvec_pushl(&child.args, enable ? "bootstrap" : "bootout", uid,
2221 filename, NULL);
2222
2223 child.no_stderr = 1;
2224 child.no_stdout = 1;
2225
2226 if (start_command(&child))
2227 die(_("failed to start launchctl"));
2228
2229 result = finish_command(&child);
2230
2231 free(cmd);
2232 free(uid);
2233 return result;
2234}
2235
2236static int launchctl_remove_plist(enum schedule_priority schedule)
2237{
2238 const char *frequency = get_frequency(schedule);
2239 char *name = launchctl_service_name(frequency);
2240 char *filename = launchctl_service_filename(name);
2241 int result = launchctl_boot_plist(0, filename);
2242 unlink(filename);
2243 free(filename);
2244 free(name);
2245 return result;
2246}
2247
2248static int launchctl_remove_plists(void)
2249{
2250 return launchctl_remove_plist(SCHEDULE_HOURLY) ||
2251 launchctl_remove_plist(SCHEDULE_DAILY) ||
2252 launchctl_remove_plist(SCHEDULE_WEEKLY);
2253}
2254
2255static int launchctl_list_contains_plist(const char *name, const char *cmd)
2256{
2257 struct child_process child = CHILD_PROCESS_INIT;
2258
2259 strvec_split(&child.args, cmd);
2260 strvec_pushl(&child.args, "list", name, NULL);
2261
2262 child.no_stderr = 1;
2263 child.no_stdout = 1;
2264
2265 if (start_command(&child))
2266 die(_("failed to start launchctl"));
2267
2268 /* Returns failure if 'name' doesn't exist. */
2269 return !finish_command(&child);
2270}
2271
2272static int launchctl_schedule_plist(const char *exec_path, enum schedule_priority schedule)
2273{
2274 int i, fd;
2275 const char *preamble, *repeat;
2276 const char *frequency = get_frequency(schedule);
2277 char *name = launchctl_service_name(frequency);
2278 char *filename = launchctl_service_filename(name);
2279 struct lock_file lk = LOCK_INIT;
2280 static unsigned long lock_file_timeout_ms = ULONG_MAX;
2281 struct strbuf plist = STRBUF_INIT, plist2 = STRBUF_INIT;
2282 struct stat st;
2283 char *cmd;
2284 int minute = get_random_minute();
2285
2286 get_schedule_cmd("launchctl", NULL, &cmd);
2287 preamble = "<?xml version=\"1.0\"?>\n"
2288 "<!DOCTYPE plist PUBLIC \"-//Apple//DTD PLIST 1.0//EN\" \"http://www.apple.com/DTDs/PropertyList-1.0.dtd\">\n"
2289 "<plist version=\"1.0\">"
2290 "<dict>\n"
2291 "<key>Label</key><string>%s</string>\n"
2292 "<key>ProgramArguments</key>\n"
2293 "<array>\n"
2294 "<string>%s/git</string>\n"
2295 "<string>--exec-path=%s</string>\n"
2296 "%s" /* For extra config parameters. */
2297 "<string>for-each-repo</string>\n"
2298 "<string>--keep-going</string>\n"
2299 "<string>--config=maintenance.repo</string>\n"
2300 "<string>maintenance</string>\n"
2301 "<string>run</string>\n"
2302 "<string>--schedule=%s</string>\n"
2303 "</array>\n"
2304 "<key>StartCalendarInterval</key>\n"
2305 "<array>\n";
2306 strbuf_addf(&plist, preamble, name, exec_path, exec_path,
2307 get_extra_launchctl_strings(), frequency);
2308
2309 switch (schedule) {
2310 case SCHEDULE_HOURLY:
2311 repeat = "<dict>\n"
2312 "<key>Hour</key><integer>%d</integer>\n"
2313 "<key>Minute</key><integer>%d</integer>\n"
2314 "</dict>\n";
2315 for (i = 1; i <= 23; i++)
2316 strbuf_addf(&plist, repeat, i, minute);
2317 break;
2318
2319 case SCHEDULE_DAILY:
2320 repeat = "<dict>\n"
2321 "<key>Weekday</key><integer>%d</integer>\n"
2322 "<key>Hour</key><integer>0</integer>\n"
2323 "<key>Minute</key><integer>%d</integer>\n"
2324 "</dict>\n";
2325 for (i = 1; i <= 6; i++)
2326 strbuf_addf(&plist, repeat, i, minute);
2327 break;
2328
2329 case SCHEDULE_WEEKLY:
2330 strbuf_addf(&plist,
2331 "<dict>\n"
2332 "<key>Weekday</key><integer>0</integer>\n"
2333 "<key>Hour</key><integer>0</integer>\n"
2334 "<key>Minute</key><integer>%d</integer>\n"
2335 "</dict>\n",
2336 minute);
2337 break;
2338
2339 default:
2340 /* unreachable */
2341 break;
2342 }
2343 strbuf_addstr(&plist, "</array>\n</dict>\n</plist>\n");
2344
2345 if (safe_create_leading_directories(the_repository, filename))
2346 die(_("failed to create directories for '%s'"), filename);
2347
2348 if ((long)lock_file_timeout_ms < 0 &&
2349 repo_config_get_ulong(the_repository, "gc.launchctlplistlocktimeoutms",
2350 &lock_file_timeout_ms))
2351 lock_file_timeout_ms = 150;
2352
2353 fd = hold_lock_file_for_update_timeout(&lk, filename, LOCK_DIE_ON_ERROR,
2354 lock_file_timeout_ms);
2355
2356 /*
2357 * Does this file already exist? With the intended contents? Is it
2358 * registered already? Then it does not need to be re-registered.
2359 */
2360 if (!stat(filename, &st) && st.st_size == plist.len &&
2361 strbuf_read_file(&plist2, filename, plist.len) == plist.len &&
2362 !strbuf_cmp(&plist, &plist2) &&
2363 launchctl_list_contains_plist(name, cmd))
2364 rollback_lock_file(&lk);
2365 else {
2366 if (write_in_full(fd, plist.buf, plist.len) < 0 ||
2367 commit_lock_file(&lk))
2368 die_errno(_("could not write '%s'"), filename);
2369
2370 /* bootout might fail if not already running, so ignore */
2371 launchctl_boot_plist(0, filename);
2372 if (launchctl_boot_plist(1, filename))
2373 die(_("failed to bootstrap service %s"), filename);
2374 }
2375
2376 free(filename);
2377 free(name);
2378 free(cmd);
2379 strbuf_release(&plist);
2380 strbuf_release(&plist2);
2381 return 0;
2382}
2383
2384static int launchctl_add_plists(void)
2385{
2386 const char *exec_path = git_exec_path();
2387
2388 return launchctl_schedule_plist(exec_path, SCHEDULE_HOURLY) ||
2389 launchctl_schedule_plist(exec_path, SCHEDULE_DAILY) ||
2390 launchctl_schedule_plist(exec_path, SCHEDULE_WEEKLY);
2391}
2392
2393static int launchctl_update_schedule(int run_maintenance, int fd UNUSED)
2394{
2395 if (run_maintenance)
2396 return launchctl_add_plists();
2397 else
2398 return launchctl_remove_plists();
2399}
2400
2401static int is_schtasks_available(void)
2402{
2403 int is_available;
2404 if (get_schedule_cmd("schtasks", &is_available, NULL))
2405 return is_available;
2406
2407#ifdef GIT_WINDOWS_NATIVE
2408 return 1;
2409#else
2410 return 0;
2411#endif
2412}
2413
2414static char *schtasks_task_name(const char *frequency)
2415{
2416 struct strbuf label = STRBUF_INIT;
2417 strbuf_addf(&label, "Git Maintenance (%s)", frequency);
2418 return strbuf_detach(&label, NULL);
2419}
2420
2421static int schtasks_remove_task(enum schedule_priority schedule)
2422{
2423 char *cmd;
2424 struct child_process child = CHILD_PROCESS_INIT;
2425 const char *frequency = get_frequency(schedule);
2426 char *name = schtasks_task_name(frequency);
2427
2428 get_schedule_cmd("schtasks", NULL, &cmd);
2429 strvec_split(&child.args, cmd);
2430 strvec_pushl(&child.args, "/delete", "/tn", name, "/f", NULL);
2431 free(name);
2432 free(cmd);
2433
2434 return run_command(&child);
2435}
2436
2437static int schtasks_remove_tasks(void)
2438{
2439 return schtasks_remove_task(SCHEDULE_HOURLY) ||
2440 schtasks_remove_task(SCHEDULE_DAILY) ||
2441 schtasks_remove_task(SCHEDULE_WEEKLY);
2442}
2443
2444static int schtasks_schedule_task(const char *exec_path, enum schedule_priority schedule)
2445{
2446 char *cmd;
2447 int result;
2448 struct child_process child = CHILD_PROCESS_INIT;
2449 const char *xml;
2450 struct tempfile *tfile;
2451 const char *frequency = get_frequency(schedule);
2452 char *name = schtasks_task_name(frequency);
2453 struct strbuf tfilename = STRBUF_INIT;
2454 int minute = get_random_minute();
2455
2456 get_schedule_cmd("schtasks", NULL, &cmd);
2457
2458 strbuf_addf(&tfilename, "%s/schedule_%s_XXXXXX",
2459 repo_get_common_dir(the_repository), frequency);
2460 tfile = xmks_tempfile(tfilename.buf);
2461 strbuf_release(&tfilename);
2462
2463 if (!fdopen_tempfile(tfile, "w"))
2464 die(_("failed to create temp xml file"));
2465
2466 xml = "<?xml version=\"1.0\" ?>\n"
2467 "<Task version=\"1.4\" xmlns=\"http://schemas.microsoft.com/windows/2004/02/mit/task\">\n"
2468 "<Triggers>\n"
2469 "<CalendarTrigger>\n";
2470 fputs(xml, tfile->fp);
2471
2472 switch (schedule) {
2473 case SCHEDULE_HOURLY:
2474 fprintf(tfile->fp,
2475 "<StartBoundary>2020-01-01T01:%02d:00</StartBoundary>\n"
2476 "<Enabled>true</Enabled>\n"
2477 "<ScheduleByDay>\n"
2478 "<DaysInterval>1</DaysInterval>\n"
2479 "</ScheduleByDay>\n"
2480 "<Repetition>\n"
2481 "<Interval>PT1H</Interval>\n"
2482 "<Duration>PT23H</Duration>\n"
2483 "<StopAtDurationEnd>false</StopAtDurationEnd>\n"
2484 "</Repetition>\n",
2485 minute);
2486 break;
2487
2488 case SCHEDULE_DAILY:
2489 fprintf(tfile->fp,
2490 "<StartBoundary>2020-01-01T00:%02d:00</StartBoundary>\n"
2491 "<Enabled>true</Enabled>\n"
2492 "<ScheduleByWeek>\n"
2493 "<DaysOfWeek>\n"
2494 "<Monday />\n"
2495 "<Tuesday />\n"
2496 "<Wednesday />\n"
2497 "<Thursday />\n"
2498 "<Friday />\n"
2499 "<Saturday />\n"
2500 "</DaysOfWeek>\n"
2501 "<WeeksInterval>1</WeeksInterval>\n"
2502 "</ScheduleByWeek>\n",
2503 minute);
2504 break;
2505
2506 case SCHEDULE_WEEKLY:
2507 fprintf(tfile->fp,
2508 "<StartBoundary>2020-01-01T00:%02d:00</StartBoundary>\n"
2509 "<Enabled>true</Enabled>\n"
2510 "<ScheduleByWeek>\n"
2511 "<DaysOfWeek>\n"
2512 "<Sunday />\n"
2513 "</DaysOfWeek>\n"
2514 "<WeeksInterval>1</WeeksInterval>\n"
2515 "</ScheduleByWeek>\n",
2516 minute);
2517 break;
2518
2519 default:
2520 break;
2521 }
2522
2523 xml = "</CalendarTrigger>\n"
2524 "</Triggers>\n"
2525 "<Principals>\n"
2526 "<Principal id=\"Author\">\n"
2527 "<LogonType>InteractiveToken</LogonType>\n"
2528 "<RunLevel>LeastPrivilege</RunLevel>\n"
2529 "</Principal>\n"
2530 "</Principals>\n"
2531 "<Settings>\n"
2532 "<MultipleInstancesPolicy>IgnoreNew</MultipleInstancesPolicy>\n"
2533 "<Enabled>true</Enabled>\n"
2534 "<Hidden>true</Hidden>\n"
2535 "<UseUnifiedSchedulingEngine>true</UseUnifiedSchedulingEngine>\n"
2536 "<WakeToRun>false</WakeToRun>\n"
2537 "<ExecutionTimeLimit>PT72H</ExecutionTimeLimit>\n"
2538 "<Priority>7</Priority>\n"
2539 "</Settings>\n"
2540 "<Actions Context=\"Author\">\n"
2541 "<Exec>\n"
2542 "<Command>\"%s\\headless-git.exe\"</Command>\n"
2543 "<Arguments>--exec-path=\"%s\" %s for-each-repo --keep-going --config=maintenance.repo maintenance run --schedule=%s</Arguments>\n"
2544 "</Exec>\n"
2545 "</Actions>\n"
2546 "</Task>\n";
2547 fprintf(tfile->fp, xml, exec_path, exec_path,
2548 get_extra_config_parameters(), frequency);
2549 strvec_split(&child.args, cmd);
2550 strvec_pushl(&child.args, "/create", "/tn", name, "/f", "/xml",
2551 get_tempfile_path(tfile), NULL);
2552 close_tempfile_gently(tfile);
2553
2554 child.no_stdout = 1;
2555 child.no_stderr = 1;
2556
2557 if (start_command(&child))
2558 die(_("failed to start schtasks"));
2559 result = finish_command(&child);
2560
2561 delete_tempfile(&tfile);
2562 free(name);
2563 free(cmd);
2564 return result;
2565}
2566
2567static int schtasks_schedule_tasks(void)
2568{
2569 const char *exec_path = git_exec_path();
2570
2571 return schtasks_schedule_task(exec_path, SCHEDULE_HOURLY) ||
2572 schtasks_schedule_task(exec_path, SCHEDULE_DAILY) ||
2573 schtasks_schedule_task(exec_path, SCHEDULE_WEEKLY);
2574}
2575
2576static int schtasks_update_schedule(int run_maintenance, int fd UNUSED)
2577{
2578 if (run_maintenance)
2579 return schtasks_schedule_tasks();
2580 else
2581 return schtasks_remove_tasks();
2582}
2583
2584MAYBE_UNUSED
2585static int check_crontab_process(const char *cmd)
2586{
2587 struct child_process child = CHILD_PROCESS_INIT;
2588
2589 strvec_split(&child.args, cmd);
2590 strvec_push(&child.args, "-l");
2591 child.no_stdin = 1;
2592 child.no_stdout = 1;
2593 child.no_stderr = 1;
2594 child.silent_exec_failure = 1;
2595
2596 if (start_command(&child))
2597 return 0;
2598 /* Ignore exit code, as an empty crontab will return error. */
2599 finish_command(&child);
2600 return 1;
2601}
2602
2603static int is_crontab_available(void)
2604{
2605 char *cmd;
2606 int is_available;
2607 int ret;
2608
2609 if (get_schedule_cmd("crontab", &is_available, &cmd)) {
2610 ret = is_available;
2611 goto out;
2612 }
2613
2614#ifdef __APPLE__
2615 /*
2616 * macOS has cron, but it requires special permissions and will
2617 * create a UI alert when attempting to run this command.
2618 */
2619 ret = 0;
2620#else
2621 ret = check_crontab_process(cmd);
2622#endif
2623
2624out:
2625 free(cmd);
2626 return ret;
2627}
2628
2629#define BEGIN_LINE "# BEGIN GIT MAINTENANCE SCHEDULE"
2630#define END_LINE "# END GIT MAINTENANCE SCHEDULE"
2631
2632static int crontab_update_schedule(int run_maintenance, int fd)
2633{
2634 char *cmd;
2635 int result = 0;
2636 int in_old_region = 0;
2637 struct child_process crontab_list = CHILD_PROCESS_INIT;
2638 struct child_process crontab_edit = CHILD_PROCESS_INIT;
2639 FILE *cron_list, *cron_in;
2640 struct strbuf line = STRBUF_INIT;
2641 struct tempfile *tmpedit = NULL;
2642 int minute = get_random_minute();
2643
2644 get_schedule_cmd("crontab", NULL, &cmd);
2645 strvec_split(&crontab_list.args, cmd);
2646 strvec_push(&crontab_list.args, "-l");
2647 crontab_list.in = -1;
2648 crontab_list.out = dup(fd);
2649 crontab_list.git_cmd = 0;
2650
2651 if (start_command(&crontab_list)) {
2652 result = error(_("failed to run 'crontab -l'; your system might not support 'cron'"));
2653 goto out;
2654 }
2655
2656 /* Ignore exit code, as an empty crontab will return error. */
2657 finish_command(&crontab_list);
2658
2659 tmpedit = mks_tempfile_t(".git_cron_edit_tmpXXXXXX");
2660 if (!tmpedit) {
2661 result = error(_("failed to create crontab temporary file"));
2662 goto out;
2663 }
2664 cron_in = fdopen_tempfile(tmpedit, "w");
2665 if (!cron_in) {
2666 result = error(_("failed to open temporary file"));
2667 goto out;
2668 }
2669
2670 /*
2671 * Read from the .lock file, filtering out the old
2672 * schedule while appending the new schedule.
2673 */
2674 cron_list = fdopen(fd, "r");
2675 rewind(cron_list);
2676
2677 while (!strbuf_getline_lf(&line, cron_list)) {
2678 if (!in_old_region && !strcmp(line.buf, BEGIN_LINE))
2679 in_old_region = 1;
2680 else if (in_old_region && !strcmp(line.buf, END_LINE))
2681 in_old_region = 0;
2682 else if (!in_old_region)
2683 fprintf(cron_in, "%s\n", line.buf);
2684 }
2685 strbuf_release(&line);
2686
2687 if (run_maintenance) {
2688 struct strbuf line_format = STRBUF_INIT;
2689 const char *exec_path = git_exec_path();
2690
2691 fprintf(cron_in, "%s\n", BEGIN_LINE);
2692 fprintf(cron_in,
2693 "# The following schedule was created by Git\n");
2694 fprintf(cron_in, "# Any edits made in this region might be\n");
2695 fprintf(cron_in,
2696 "# replaced in the future by a Git command.\n\n");
2697
2698 strbuf_addf(&line_format,
2699 "%%d %%s * * %%s \"%s/git\" --exec-path=\"%s\" %s for-each-repo --keep-going --config=maintenance.repo maintenance run --schedule=%%s\n",
2700 exec_path, exec_path, get_extra_config_parameters());
2701 fprintf(cron_in, line_format.buf, minute, "1-23", "*", "hourly");
2702 fprintf(cron_in, line_format.buf, minute, "0", "1-6", "daily");
2703 fprintf(cron_in, line_format.buf, minute, "0", "0", "weekly");
2704 strbuf_release(&line_format);
2705
2706 fprintf(cron_in, "\n%s\n", END_LINE);
2707 }
2708
2709 fflush(cron_in);
2710
2711 strvec_split(&crontab_edit.args, cmd);
2712 strvec_push(&crontab_edit.args, get_tempfile_path(tmpedit));
2713 crontab_edit.git_cmd = 0;
2714
2715 if (start_command(&crontab_edit)) {
2716 result = error(_("failed to run 'crontab'; your system might not support 'cron'"));
2717 goto out;
2718 }
2719
2720 if (finish_command(&crontab_edit))
2721 result = error(_("'crontab' died"));
2722 else
2723 fclose(cron_list);
2724
2725out:
2726 delete_tempfile(&tmpedit);
2727 free(cmd);
2728 return result;
2729}
2730
2731static int real_is_systemd_timer_available(void)
2732{
2733 struct child_process child = CHILD_PROCESS_INIT;
2734
2735 strvec_pushl(&child.args, "systemctl", "--user", "list-timers", NULL);
2736 child.no_stdin = 1;
2737 child.no_stdout = 1;
2738 child.no_stderr = 1;
2739 child.silent_exec_failure = 1;
2740
2741 if (start_command(&child))
2742 return 0;
2743 if (finish_command(&child))
2744 return 0;
2745 return 1;
2746}
2747
2748static int is_systemd_timer_available(void)
2749{
2750 int is_available;
2751
2752 if (get_schedule_cmd("systemctl", &is_available, NULL))
2753 return is_available;
2754
2755 return real_is_systemd_timer_available();
2756}
2757
2758static char *xdg_config_home_systemd(const char *filename)
2759{
2760 return xdg_config_home_for("systemd/user", filename);
2761}
2762
2763#define SYSTEMD_UNIT_FORMAT "git-maintenance@%s.%s"
2764
2765static int systemd_timer_delete_timer_file(enum schedule_priority priority)
2766{
2767 int ret = 0;
2768 const char *frequency = get_frequency(priority);
2769 char *local_timer_name = xstrfmt(SYSTEMD_UNIT_FORMAT, frequency, "timer");
2770 char *filename = xdg_config_home_systemd(local_timer_name);
2771
2772 if (unlink(filename) && !is_missing_file_error(errno))
2773 ret = error_errno(_("failed to delete '%s'"), filename);
2774
2775 free(filename);
2776 free(local_timer_name);
2777 return ret;
2778}
2779
2780static int systemd_timer_delete_service_template(void)
2781{
2782 int ret = 0;
2783 char *local_service_name = xstrfmt(SYSTEMD_UNIT_FORMAT, "", "service");
2784 char *filename = xdg_config_home_systemd(local_service_name);
2785 if (unlink(filename) && !is_missing_file_error(errno))
2786 ret = error_errno(_("failed to delete '%s'"), filename);
2787
2788 free(filename);
2789 free(local_service_name);
2790 return ret;
2791}
2792
2793/*
2794 * Write the schedule information into a git-maintenance@<schedule>.timer
2795 * file using a custom minute. This timer file cannot use the templating
2796 * system, so we generate a specific file for each.
2797 */
2798static int systemd_timer_write_timer_file(enum schedule_priority schedule,
2799 int minute)
2800{
2801 int res = -1;
2802 char *filename;
2803 FILE *file;
2804 const char *unit;
2805 char *schedule_pattern = NULL;
2806 const char *frequency = get_frequency(schedule);
2807 char *local_timer_name = xstrfmt(SYSTEMD_UNIT_FORMAT, frequency, "timer");
2808
2809 filename = xdg_config_home_systemd(local_timer_name);
2810
2811 if (safe_create_leading_directories(the_repository, filename)) {
2812 error(_("failed to create directories for '%s'"), filename);
2813 goto error;
2814 }
2815 file = fopen_or_warn(filename, "w");
2816 if (!file)
2817 goto error;
2818
2819 switch (schedule) {
2820 case SCHEDULE_HOURLY:
2821 schedule_pattern = xstrfmt("*-*-* 1..23:%02d:00", minute);
2822 break;
2823
2824 case SCHEDULE_DAILY:
2825 schedule_pattern = xstrfmt("Tue..Sun *-*-* 0:%02d:00", minute);
2826 break;
2827
2828 case SCHEDULE_WEEKLY:
2829 schedule_pattern = xstrfmt("Mon 0:%02d:00", minute);
2830 break;
2831
2832 default:
2833 BUG("Unhandled schedule_priority");
2834 }
2835
2836 unit = "# This file was created and is maintained by Git.\n"
2837 "# Any edits made in this file might be replaced in the future\n"
2838 "# by a Git command.\n"
2839 "\n"
2840 "[Unit]\n"
2841 "Description=Optimize Git repositories data\n"
2842 "\n"
2843 "[Timer]\n"
2844 "OnCalendar=%s\n"
2845 "Persistent=true\n"
2846 "\n"
2847 "[Install]\n"
2848 "WantedBy=timers.target\n";
2849 if (fprintf(file, unit, schedule_pattern) < 0) {
2850 error(_("failed to write to '%s'"), filename);
2851 fclose(file);
2852 goto error;
2853 }
2854 if (fclose(file) == EOF) {
2855 error_errno(_("failed to flush '%s'"), filename);
2856 goto error;
2857 }
2858
2859 res = 0;
2860
2861error:
2862 free(schedule_pattern);
2863 free(local_timer_name);
2864 free(filename);
2865 return res;
2866}
2867
2868/*
2869 * No matter the schedule, we use the same service and can make use of the
2870 * templating system. When installing git-maintenance@<schedule>.timer,
2871 * systemd will notice that git-maintenance@.service exists as a template
2872 * and will use this file and insert the <schedule> into the template at
2873 * the position of "%i".
2874 */
2875static int systemd_timer_write_service_template(const char *exec_path)
2876{
2877 int res = -1;
2878 char *filename;
2879 FILE *file;
2880 const char *unit;
2881 char *local_service_name = xstrfmt(SYSTEMD_UNIT_FORMAT, "", "service");
2882
2883 filename = xdg_config_home_systemd(local_service_name);
2884 if (safe_create_leading_directories(the_repository, filename)) {
2885 error(_("failed to create directories for '%s'"), filename);
2886 goto error;
2887 }
2888 file = fopen_or_warn(filename, "w");
2889 if (!file)
2890 goto error;
2891
2892 unit = "# This file was created and is maintained by Git.\n"
2893 "# Any edits made in this file might be replaced in the future\n"
2894 "# by a Git command.\n"
2895 "\n"
2896 "[Unit]\n"
2897 "Description=Optimize Git repositories data\n"
2898 "\n"
2899 "[Service]\n"
2900 "Type=oneshot\n"
2901 "ExecStart=\"%s/git\" --exec-path=\"%s\" %s for-each-repo --keep-going --config=maintenance.repo maintenance run --schedule=%%i\n"
2902 "LockPersonality=yes\n"
2903 "MemoryDenyWriteExecute=yes\n"
2904 "NoNewPrivileges=yes\n"
2905 "RestrictAddressFamilies=AF_UNIX AF_INET AF_INET6 AF_VSOCK\n"
2906 "RestrictNamespaces=yes\n"
2907 "RestrictRealtime=yes\n"
2908 "RestrictSUIDSGID=yes\n"
2909 "SystemCallArchitectures=native\n"
2910 "SystemCallFilter=@system-service\n";
2911 if (fprintf(file, unit, exec_path, exec_path, get_extra_config_parameters()) < 0) {
2912 error(_("failed to write to '%s'"), filename);
2913 fclose(file);
2914 goto error;
2915 }
2916 if (fclose(file) == EOF) {
2917 error_errno(_("failed to flush '%s'"), filename);
2918 goto error;
2919 }
2920
2921 res = 0;
2922
2923error:
2924 free(local_service_name);
2925 free(filename);
2926 return res;
2927}
2928
2929static int systemd_timer_enable_unit(int enable,
2930 enum schedule_priority schedule,
2931 int minute)
2932{
2933 char *cmd = NULL;
2934 struct child_process child = CHILD_PROCESS_INIT;
2935 const char *frequency = get_frequency(schedule);
2936 int ret;
2937
2938 /*
2939 * Disabling the systemd unit while it is already disabled makes
2940 * systemctl print an error.
2941 * Let's ignore it since it means we already are in the expected state:
2942 * the unit is disabled.
2943 *
2944 * On the other hand, enabling a systemd unit which is already enabled
2945 * produces no error.
2946 */
2947 if (!enable) {
2948 child.no_stderr = 1;
2949 } else if (systemd_timer_write_timer_file(schedule, minute)) {
2950 ret = -1;
2951 goto out;
2952 }
2953
2954 get_schedule_cmd("systemctl", NULL, &cmd);
2955 strvec_split(&child.args, cmd);
2956 strvec_pushl(&child.args, "--user", enable ? "enable" : "disable",
2957 "--now", NULL);
2958 strvec_pushf(&child.args, SYSTEMD_UNIT_FORMAT, frequency, "timer");
2959
2960 if (start_command(&child)) {
2961 ret = error(_("failed to start systemctl"));
2962 goto out;
2963 }
2964
2965 if (finish_command(&child)) {
2966 /*
2967 * Disabling an already disabled systemd unit makes
2968 * systemctl fail.
2969 * Let's ignore this failure.
2970 *
2971 * Enabling an enabled systemd unit doesn't fail.
2972 */
2973 if (enable) {
2974 ret = error(_("failed to run systemctl"));
2975 goto out;
2976 }
2977 }
2978
2979 ret = 0;
2980
2981out:
2982 free(cmd);
2983 return ret;
2984}
2985
2986/*
2987 * A previous version of Git wrote the timer units as template files.
2988 * Clean these up, if they exist.
2989 */
2990static void systemd_timer_delete_stale_timer_templates(void)
2991{
2992 char *timer_template_name = xstrfmt(SYSTEMD_UNIT_FORMAT, "", "timer");
2993 char *filename = xdg_config_home_systemd(timer_template_name);
2994
2995 if (unlink(filename) && !is_missing_file_error(errno))
2996 warning(_("failed to delete '%s'"), filename);
2997
2998 free(filename);
2999 free(timer_template_name);
3000}
3001
3002static int systemd_timer_delete_unit_files(void)
3003{
3004 systemd_timer_delete_stale_timer_templates();
3005
3006 /* Purposefully not short-circuited to make sure all are called. */
3007 return systemd_timer_delete_timer_file(SCHEDULE_HOURLY) |
3008 systemd_timer_delete_timer_file(SCHEDULE_DAILY) |
3009 systemd_timer_delete_timer_file(SCHEDULE_WEEKLY) |
3010 systemd_timer_delete_service_template();
3011}
3012
3013static int systemd_timer_delete_units(void)
3014{
3015 int minute = get_random_minute();
3016 /* Purposefully not short-circuited to make sure all are called. */
3017 return systemd_timer_enable_unit(0, SCHEDULE_HOURLY, minute) |
3018 systemd_timer_enable_unit(0, SCHEDULE_DAILY, minute) |
3019 systemd_timer_enable_unit(0, SCHEDULE_WEEKLY, minute) |
3020 systemd_timer_delete_unit_files();
3021}
3022
3023static int systemd_timer_setup_units(void)
3024{
3025 int minute = get_random_minute();
3026 const char *exec_path = git_exec_path();
3027
3028 int ret = systemd_timer_write_service_template(exec_path) ||
3029 systemd_timer_enable_unit(1, SCHEDULE_HOURLY, minute) ||
3030 systemd_timer_enable_unit(1, SCHEDULE_DAILY, minute) ||
3031 systemd_timer_enable_unit(1, SCHEDULE_WEEKLY, minute);
3032
3033 if (ret)
3034 systemd_timer_delete_units();
3035 else
3036 systemd_timer_delete_stale_timer_templates();
3037
3038 return ret;
3039}
3040
3041static int systemd_timer_update_schedule(int run_maintenance, int fd UNUSED)
3042{
3043 if (run_maintenance)
3044 return systemd_timer_setup_units();
3045 else
3046 return systemd_timer_delete_units();
3047}
3048
3049enum scheduler {
3050 SCHEDULER_INVALID = -1,
3051 SCHEDULER_AUTO,
3052 SCHEDULER_CRON,
3053 SCHEDULER_SYSTEMD,
3054 SCHEDULER_LAUNCHCTL,
3055 SCHEDULER_SCHTASKS,
3056};
3057
3058static const struct {
3059 const char *name;
3060 int (*is_available)(void);
3061 int (*update_schedule)(int run_maintenance, int fd);
3062} scheduler_fn[] = {
3063 [SCHEDULER_CRON] = {
3064 .name = "crontab",
3065 .is_available = is_crontab_available,
3066 .update_schedule = crontab_update_schedule,
3067 },
3068 [SCHEDULER_SYSTEMD] = {
3069 .name = "systemctl",
3070 .is_available = is_systemd_timer_available,
3071 .update_schedule = systemd_timer_update_schedule,
3072 },
3073 [SCHEDULER_LAUNCHCTL] = {
3074 .name = "launchctl",
3075 .is_available = is_launchctl_available,
3076 .update_schedule = launchctl_update_schedule,
3077 },
3078 [SCHEDULER_SCHTASKS] = {
3079 .name = "schtasks",
3080 .is_available = is_schtasks_available,
3081 .update_schedule = schtasks_update_schedule,
3082 },
3083};
3084
3085static enum scheduler parse_scheduler(const char *value)
3086{
3087 if (!value)
3088 return SCHEDULER_INVALID;
3089 else if (!strcasecmp(value, "auto"))
3090 return SCHEDULER_AUTO;
3091 else if (!strcasecmp(value, "cron") || !strcasecmp(value, "crontab"))
3092 return SCHEDULER_CRON;
3093 else if (!strcasecmp(value, "systemd") ||
3094 !strcasecmp(value, "systemd-timer"))
3095 return SCHEDULER_SYSTEMD;
3096 else if (!strcasecmp(value, "launchctl"))
3097 return SCHEDULER_LAUNCHCTL;
3098 else if (!strcasecmp(value, "schtasks"))
3099 return SCHEDULER_SCHTASKS;
3100 else
3101 return SCHEDULER_INVALID;
3102}
3103
3104static int maintenance_opt_scheduler(const struct option *opt, const char *arg,
3105 int unset)
3106{
3107 enum scheduler *scheduler = opt->value;
3108
3109 BUG_ON_OPT_NEG(unset);
3110
3111 *scheduler = parse_scheduler(arg);
3112 if (*scheduler == SCHEDULER_INVALID)
3113 return error(_("unrecognized --scheduler argument '%s'"), arg);
3114 return 0;
3115}
3116
3117struct maintenance_start_opts {
3118 enum scheduler scheduler;
3119};
3120
3121static enum scheduler resolve_scheduler(enum scheduler scheduler)
3122{
3123 if (scheduler != SCHEDULER_AUTO)
3124 return scheduler;
3125
3126#if defined(__APPLE__)
3127 return SCHEDULER_LAUNCHCTL;
3128
3129#elif defined(GIT_WINDOWS_NATIVE)
3130 return SCHEDULER_SCHTASKS;
3131
3132#elif defined(__linux__)
3133 if (is_systemd_timer_available())
3134 return SCHEDULER_SYSTEMD;
3135 else if (is_crontab_available())
3136 return SCHEDULER_CRON;
3137 else
3138 die(_("neither systemd timers nor crontab are available"));
3139
3140#else
3141 return SCHEDULER_CRON;
3142#endif
3143}
3144
3145static void validate_scheduler(enum scheduler scheduler)
3146{
3147 if (scheduler == SCHEDULER_INVALID)
3148 BUG("invalid scheduler");
3149 if (scheduler == SCHEDULER_AUTO)
3150 BUG("resolve_scheduler should have been called before");
3151
3152 if (!scheduler_fn[scheduler].is_available())
3153 die(_("%s scheduler is not available"),
3154 scheduler_fn[scheduler].name);
3155}
3156
3157static int update_background_schedule(const struct maintenance_start_opts *opts,
3158 int enable)
3159{
3160 unsigned int i;
3161 int result = 0;
3162 struct lock_file lk;
3163 char *lock_path = xstrfmt("%s/schedule", the_repository->objects->sources->path);
3164
3165 if (hold_lock_file_for_update(&lk, lock_path, LOCK_NO_DEREF) < 0) {
3166 if (errno == EEXIST)
3167 error(_("unable to create '%s.lock': %s.\n\n"
3168 "Another scheduled git-maintenance(1) process seems to be running in this\n"
3169 "repository. Please make sure no other maintenance processes are running and\n"
3170 "then try again. If it still fails, a git-maintenance(1) process may have\n"
3171 "crashed in this repository earlier: remove the file manually to continue."),
3172 absolute_path(lock_path), strerror(errno));
3173 else
3174 error_errno(_("cannot acquire lock for scheduled background maintenance"));
3175 free(lock_path);
3176 return -1;
3177 }
3178
3179 for (i = 1; i < ARRAY_SIZE(scheduler_fn); i++) {
3180 if (enable && opts->scheduler == i)
3181 continue;
3182 if (!scheduler_fn[i].is_available())
3183 continue;
3184 scheduler_fn[i].update_schedule(0, get_lock_file_fd(&lk));
3185 }
3186
3187 if (enable)
3188 result = scheduler_fn[opts->scheduler].update_schedule(
3189 1, get_lock_file_fd(&lk));
3190
3191 rollback_lock_file(&lk);
3192
3193 free(lock_path);
3194 return result;
3195}
3196
3197static const char *const builtin_maintenance_start_usage[] = {
3198 N_("git maintenance start [--scheduler=<scheduler>]"),
3199 NULL
3200};
3201
3202static int maintenance_start(int argc, const char **argv, const char *prefix,
3203 struct repository *repo)
3204{
3205 struct maintenance_start_opts opts = { 0 };
3206 struct option options[] = {
3207 OPT_CALLBACK_F(
3208 0, "scheduler", &opts.scheduler, N_("scheduler"),
3209 N_("scheduler to trigger git maintenance run"),
3210 PARSE_OPT_NONEG, maintenance_opt_scheduler),
3211 OPT_END()
3212 };
3213 const char *register_args[] = { "register", NULL };
3214
3215 argc = parse_options(argc, argv, prefix, options,
3216 builtin_maintenance_start_usage, 0);
3217 if (argc)
3218 usage_with_options(builtin_maintenance_start_usage, options);
3219
3220 opts.scheduler = resolve_scheduler(opts.scheduler);
3221 validate_scheduler(opts.scheduler);
3222
3223 if (update_background_schedule(&opts, 1))
3224 die(_("failed to set up maintenance schedule"));
3225
3226 if (maintenance_register(ARRAY_SIZE(register_args)-1, register_args, NULL, repo))
3227 warning(_("failed to add repo to global config"));
3228 return 0;
3229}
3230
3231static const char *const builtin_maintenance_stop_usage[] = {
3232 "git maintenance stop",
3233 NULL
3234};
3235
3236static int maintenance_stop(int argc, const char **argv, const char *prefix,
3237 struct repository *repo UNUSED)
3238{
3239 struct option options[] = {
3240 OPT_END()
3241 };
3242 argc = parse_options(argc, argv, prefix, options,
3243 builtin_maintenance_stop_usage, 0);
3244 if (argc)
3245 usage_with_options(builtin_maintenance_stop_usage, options);
3246 return update_background_schedule(NULL, 0);
3247}
3248
3249static const char * const builtin_maintenance_usage[] = {
3250 N_("git maintenance <subcommand> [<options>]"),
3251 NULL,
3252};
3253
3254int cmd_maintenance(int argc,
3255 const char **argv,
3256 const char *prefix,
3257 struct repository *repo)
3258{
3259 parse_opt_subcommand_fn *fn = NULL;
3260 struct option builtin_maintenance_options[] = {
3261 OPT_SUBCOMMAND("run", &fn, maintenance_run),
3262 OPT_SUBCOMMAND("start", &fn, maintenance_start),
3263 OPT_SUBCOMMAND("stop", &fn, maintenance_stop),
3264 OPT_SUBCOMMAND("register", &fn, maintenance_register),
3265 OPT_SUBCOMMAND("unregister", &fn, maintenance_unregister),
3266 OPT_END(),
3267 };
3268
3269 argc = parse_options(argc, argv, prefix, builtin_maintenance_options,
3270 builtin_maintenance_usage, 0);
3271 return fn(argc, argv, prefix, repo);
3272}