git
/
pack-write.c
594 строки · 16.0 Кб
1#define USE_THE_REPOSITORY_VARIABLE2
3#include "git-compat-util.h"4#include "environment.h"5#include "gettext.h"6#include "hex.h"7#include "pack.h"8#include "csum-file.h"9#include "remote.h"10#include "chunk-format.h"11#include "pack-mtimes.h"12#include "pack-objects.h"13#include "pack-revindex.h"14#include "path.h"15#include "strbuf.h"16
17void reset_pack_idx_option(struct pack_idx_option *opts)18{
19memset(opts, 0, sizeof(*opts));20opts->version = 2;21opts->off32_limit = 0x7fffffff;22}
23
24static int sha1_compare(const void *_a, const void *_b)25{
26struct pack_idx_entry *a = *(struct pack_idx_entry **)_a;27struct pack_idx_entry *b = *(struct pack_idx_entry **)_b;28return oidcmp(&a->oid, &b->oid);29}
30
31static int cmp_uint32(const void *a_, const void *b_)32{
33uint32_t a = *((uint32_t *)a_);34uint32_t b = *((uint32_t *)b_);35
36return (a < b) ? -1 : (a != b);37}
38
39static int need_large_offset(off_t offset, const struct pack_idx_option *opts)40{
41uint32_t ofsval;42
43if ((offset >> 31) || (opts->off32_limit < offset))44return 1;45if (!opts->anomaly_nr)46return 0;47ofsval = offset;48return !!bsearch(&ofsval, opts->anomaly, opts->anomaly_nr,49sizeof(ofsval), cmp_uint32);50}
51
52/*
53* The *sha1 contains the pack content SHA1 hash.
54* The objects array passed in will be sorted by SHA1 on exit.
55*/
56const char *write_idx_file(const char *index_name, struct pack_idx_entry **objects,57int nr_objects, const struct pack_idx_option *opts,58const unsigned char *sha1)59{
60struct hashfile *f;61struct pack_idx_entry **sorted_by_sha, **list, **last;62off_t last_obj_offset = 0;63int i, fd;64uint32_t index_version;65
66if (nr_objects) {67sorted_by_sha = objects;68list = sorted_by_sha;69last = sorted_by_sha + nr_objects;70for (i = 0; i < nr_objects; ++i) {71if (objects[i]->offset > last_obj_offset)72last_obj_offset = objects[i]->offset;73}74QSORT(sorted_by_sha, nr_objects, sha1_compare);75}76else77sorted_by_sha = list = last = NULL;78
79if (opts->flags & WRITE_IDX_VERIFY) {80assert(index_name);81f = hashfd_check(index_name);82} else {83if (!index_name) {84struct strbuf tmp_file = STRBUF_INIT;85fd = odb_mkstemp(&tmp_file, "pack/tmp_idx_XXXXXX");86index_name = strbuf_detach(&tmp_file, NULL);87} else {88unlink(index_name);89fd = xopen(index_name, O_CREAT|O_EXCL|O_WRONLY, 0600);90}91f = hashfd(fd, index_name);92}93
94/* if last object's offset is >= 2^31 we should use index V2 */95index_version = need_large_offset(last_obj_offset, opts) ? 2 : opts->version;96
97/* index versions 2 and above need a header */98if (index_version >= 2) {99struct pack_idx_header hdr;100hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);101hdr.idx_version = htonl(index_version);102hashwrite(f, &hdr, sizeof(hdr));103}104
105/*106* Write the first-level table (the list is sorted,
107* but we use a 256-entry lookup to be able to avoid
108* having to do eight extra binary search iterations).
109*/
110for (i = 0; i < 256; i++) {111struct pack_idx_entry **next = list;112while (next < last) {113struct pack_idx_entry *obj = *next;114if (obj->oid.hash[0] != i)115break;116next++;117}118hashwrite_be32(f, next - sorted_by_sha);119list = next;120}121
122/*123* Write the actual SHA1 entries..
124*/
125list = sorted_by_sha;126for (i = 0; i < nr_objects; i++) {127struct pack_idx_entry *obj = *list++;128if (index_version < 2)129hashwrite_be32(f, obj->offset);130hashwrite(f, obj->oid.hash, the_hash_algo->rawsz);131if ((opts->flags & WRITE_IDX_STRICT) &&132(i && oideq(&list[-2]->oid, &obj->oid)))133die("The same object %s appears twice in the pack",134oid_to_hex(&obj->oid));135}136
137if (index_version >= 2) {138unsigned int nr_large_offset = 0;139
140/* write the crc32 table */141list = sorted_by_sha;142for (i = 0; i < nr_objects; i++) {143struct pack_idx_entry *obj = *list++;144hashwrite_be32(f, obj->crc32);145}146
147/* write the 32-bit offset table */148list = sorted_by_sha;149for (i = 0; i < nr_objects; i++) {150struct pack_idx_entry *obj = *list++;151uint32_t offset;152
153offset = (need_large_offset(obj->offset, opts)154? (0x80000000 | nr_large_offset++)155: obj->offset);156hashwrite_be32(f, offset);157}158
159/* write the large offset table */160list = sorted_by_sha;161while (nr_large_offset) {162struct pack_idx_entry *obj = *list++;163uint64_t offset = obj->offset;164
165if (!need_large_offset(offset, opts))166continue;167hashwrite_be64(f, offset);168nr_large_offset--;169}170}171
172hashwrite(f, sha1, the_hash_algo->rawsz);173finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,174CSUM_HASH_IN_STREAM | CSUM_CLOSE |175((opts->flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));176return index_name;177}
178
179static int pack_order_cmp(const void *va, const void *vb, void *ctx)180{
181struct pack_idx_entry **objects = ctx;182
183off_t oa = objects[*(uint32_t*)va]->offset;184off_t ob = objects[*(uint32_t*)vb]->offset;185
186if (oa < ob)187return -1;188if (oa > ob)189return 1;190return 0;191}
192
193static void write_rev_header(struct hashfile *f)194{
195hashwrite_be32(f, RIDX_SIGNATURE);196hashwrite_be32(f, RIDX_VERSION);197hashwrite_be32(f, oid_version(the_hash_algo));198}
199
200static void write_rev_index_positions(struct hashfile *f,201uint32_t *pack_order,202uint32_t nr_objects)203{
204uint32_t i;205for (i = 0; i < nr_objects; i++)206hashwrite_be32(f, pack_order[i]);207}
208
209static void write_rev_trailer(struct hashfile *f, const unsigned char *hash)210{
211hashwrite(f, hash, the_hash_algo->rawsz);212}
213
214const char *write_rev_file(const char *rev_name,215struct pack_idx_entry **objects,216uint32_t nr_objects,217const unsigned char *hash,218unsigned flags)219{
220uint32_t *pack_order;221uint32_t i;222const char *ret;223
224if (!(flags & WRITE_REV) && !(flags & WRITE_REV_VERIFY))225return NULL;226
227ALLOC_ARRAY(pack_order, nr_objects);228for (i = 0; i < nr_objects; i++)229pack_order[i] = i;230QSORT_S(pack_order, nr_objects, pack_order_cmp, objects);231
232ret = write_rev_file_order(rev_name, pack_order, nr_objects, hash,233flags);234
235free(pack_order);236
237return ret;238}
239
240const char *write_rev_file_order(const char *rev_name,241uint32_t *pack_order,242uint32_t nr_objects,243const unsigned char *hash,244unsigned flags)245{
246struct hashfile *f;247int fd;248
249if ((flags & WRITE_REV) && (flags & WRITE_REV_VERIFY))250die(_("cannot both write and verify reverse index"));251
252if (flags & WRITE_REV) {253if (!rev_name) {254struct strbuf tmp_file = STRBUF_INIT;255fd = odb_mkstemp(&tmp_file, "pack/tmp_rev_XXXXXX");256rev_name = strbuf_detach(&tmp_file, NULL);257} else {258unlink(rev_name);259fd = xopen(rev_name, O_CREAT|O_EXCL|O_WRONLY, 0600);260}261f = hashfd(fd, rev_name);262} else if (flags & WRITE_REV_VERIFY) {263struct stat statbuf;264if (stat(rev_name, &statbuf)) {265if (errno == ENOENT) {266/* .rev files are optional */267return NULL;268} else269die_errno(_("could not stat: %s"), rev_name);270}271f = hashfd_check(rev_name);272} else273return NULL;274
275write_rev_header(f);276
277write_rev_index_positions(f, pack_order, nr_objects);278write_rev_trailer(f, hash);279
280if (rev_name && adjust_shared_perm(rev_name) < 0)281die(_("failed to make %s readable"), rev_name);282
283finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,284CSUM_HASH_IN_STREAM | CSUM_CLOSE |285((flags & WRITE_IDX_VERIFY) ? 0 : CSUM_FSYNC));286
287return rev_name;288}
289
290static void write_mtimes_header(struct hashfile *f)291{
292hashwrite_be32(f, MTIMES_SIGNATURE);293hashwrite_be32(f, MTIMES_VERSION);294hashwrite_be32(f, oid_version(the_hash_algo));295}
296
297/*
298* Writes the object mtimes of "objects" for use in a .mtimes file.
299* Note that objects must be in lexicographic (index) order, which is
300* the expected ordering of these values in the .mtimes file.
301*/
302static void write_mtimes_objects(struct hashfile *f,303struct packing_data *to_pack,304struct pack_idx_entry **objects,305uint32_t nr_objects)306{
307uint32_t i;308for (i = 0; i < nr_objects; i++) {309struct object_entry *e = (struct object_entry*)objects[i];310hashwrite_be32(f, oe_cruft_mtime(to_pack, e));311}312}
313
314static void write_mtimes_trailer(struct hashfile *f, const unsigned char *hash)315{
316hashwrite(f, hash, the_hash_algo->rawsz);317}
318
319static char *write_mtimes_file(struct packing_data *to_pack,320struct pack_idx_entry **objects,321uint32_t nr_objects,322const unsigned char *hash)323{
324struct strbuf tmp_file = STRBUF_INIT;325char *mtimes_name;326struct hashfile *f;327int fd;328
329if (!to_pack)330BUG("cannot call write_mtimes_file with NULL packing_data");331
332fd = odb_mkstemp(&tmp_file, "pack/tmp_mtimes_XXXXXX");333mtimes_name = strbuf_detach(&tmp_file, NULL);334f = hashfd(fd, mtimes_name);335
336write_mtimes_header(f);337write_mtimes_objects(f, to_pack, objects, nr_objects);338write_mtimes_trailer(f, hash);339
340if (adjust_shared_perm(mtimes_name) < 0)341die(_("failed to make %s readable"), mtimes_name);342
343finalize_hashfile(f, NULL, FSYNC_COMPONENT_PACK_METADATA,344CSUM_HASH_IN_STREAM | CSUM_CLOSE | CSUM_FSYNC);345
346return mtimes_name;347}
348
349off_t write_pack_header(struct hashfile *f, uint32_t nr_entries)350{
351struct pack_header hdr;352
353hdr.hdr_signature = htonl(PACK_SIGNATURE);354hdr.hdr_version = htonl(PACK_VERSION);355hdr.hdr_entries = htonl(nr_entries);356hashwrite(f, &hdr, sizeof(hdr));357return sizeof(hdr);358}
359
360/*
361* Update pack header with object_count and compute new SHA1 for pack data
362* associated to pack_fd, and write that SHA1 at the end. That new SHA1
363* is also returned in new_pack_sha1.
364*
365* If partial_pack_sha1 is non null, then the SHA1 of the existing pack
366* (without the header update) is computed and validated against the
367* one provided in partial_pack_sha1. The validation is performed at
368* partial_pack_offset bytes in the pack file. The SHA1 of the remaining
369* data (i.e. from partial_pack_offset to the end) is then computed and
370* returned in partial_pack_sha1.
371*
372* Note that new_pack_sha1 is updated last, so both new_pack_sha1 and
373* partial_pack_sha1 can refer to the same buffer if the caller is not
374* interested in the resulting SHA1 of pack data above partial_pack_offset.
375*/
376void fixup_pack_header_footer(int pack_fd,377unsigned char *new_pack_hash,378const char *pack_name,379uint32_t object_count,380unsigned char *partial_pack_hash,381off_t partial_pack_offset)382{
383int aligned_sz, buf_sz = 8 * 1024;384git_hash_ctx old_hash_ctx, new_hash_ctx;385struct pack_header hdr;386char *buf;387ssize_t read_result;388
389the_hash_algo->init_fn(&old_hash_ctx);390the_hash_algo->init_fn(&new_hash_ctx);391
392if (lseek(pack_fd, 0, SEEK_SET) != 0)393die_errno("Failed seeking to start of '%s'", pack_name);394read_result = read_in_full(pack_fd, &hdr, sizeof(hdr));395if (read_result < 0)396die_errno("Unable to reread header of '%s'", pack_name);397else if (read_result != sizeof(hdr))398die_errno("Unexpected short read for header of '%s'",399pack_name);400if (lseek(pack_fd, 0, SEEK_SET) != 0)401die_errno("Failed seeking to start of '%s'", pack_name);402the_hash_algo->update_fn(&old_hash_ctx, &hdr, sizeof(hdr));403hdr.hdr_entries = htonl(object_count);404the_hash_algo->update_fn(&new_hash_ctx, &hdr, sizeof(hdr));405write_or_die(pack_fd, &hdr, sizeof(hdr));406partial_pack_offset -= sizeof(hdr);407
408buf = xmalloc(buf_sz);409aligned_sz = buf_sz - sizeof(hdr);410for (;;) {411ssize_t m, n;412m = (partial_pack_hash && partial_pack_offset < aligned_sz) ?413partial_pack_offset : aligned_sz;414n = xread(pack_fd, buf, m);415if (!n)416break;417if (n < 0)418die_errno("Failed to checksum '%s'", pack_name);419the_hash_algo->update_fn(&new_hash_ctx, buf, n);420
421aligned_sz -= n;422if (!aligned_sz)423aligned_sz = buf_sz;424
425if (!partial_pack_hash)426continue;427
428the_hash_algo->update_fn(&old_hash_ctx, buf, n);429partial_pack_offset -= n;430if (partial_pack_offset == 0) {431unsigned char hash[GIT_MAX_RAWSZ];432the_hash_algo->final_fn(hash, &old_hash_ctx);433if (!hasheq(hash, partial_pack_hash,434the_repository->hash_algo))435die("Unexpected checksum for %s "436"(disk corruption?)", pack_name);437/*438* Now let's compute the SHA1 of the remainder of the
439* pack, which also means making partial_pack_offset
440* big enough not to matter anymore.
441*/
442the_hash_algo->init_fn(&old_hash_ctx);443partial_pack_offset = ~partial_pack_offset;444partial_pack_offset -= MSB(partial_pack_offset, 1);445}446}447free(buf);448
449if (partial_pack_hash)450the_hash_algo->final_fn(partial_pack_hash, &old_hash_ctx);451the_hash_algo->final_fn(new_pack_hash, &new_hash_ctx);452write_or_die(pack_fd, new_pack_hash, the_hash_algo->rawsz);453fsync_component_or_die(FSYNC_COMPONENT_PACK, pack_fd, pack_name);454}
455
456char *index_pack_lockfile(int ip_out, int *is_well_formed)457{
458char packname[GIT_MAX_HEXSZ + 6];459const int len = the_hash_algo->hexsz + 6;460
461/*462* The first thing we expect from index-pack's output
463* is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
464* %40s is the newly created pack SHA1 name. In the "keep"
465* case, we need it to remove the corresponding .keep file
466* later on. If we don't get that then tough luck with it.
467*/
468if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {469const char *name;470
471if (is_well_formed)472*is_well_formed = 1;473packname[len-1] = 0;474if (skip_prefix(packname, "keep\t", &name))475return xstrfmt("%s/pack/pack-%s.keep",476get_object_directory(), name);477return NULL;478}479if (is_well_formed)480*is_well_formed = 0;481return NULL;482}
483
484/*
485* The per-object header is a pretty dense thing, which is
486* - first byte: low four bits are "size", then three bits of "type",
487* and the high bit is "size continues".
488* - each byte afterwards: low seven bits are size continuation,
489* with the high bit being "size continues"
490*/
491int encode_in_pack_object_header(unsigned char *hdr, int hdr_len,492enum object_type type, uintmax_t size)493{
494int n = 1;495unsigned char c;496
497if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)498die("bad type %d", type);499
500c = (type << 4) | (size & 15);501size >>= 4;502while (size) {503if (n == hdr_len)504die("object size is too enormous to format");505*hdr++ = c | 0x80;506c = size & 0x7f;507size >>= 7;508n++;509}510*hdr = c;511return n;512}
513
514struct hashfile *create_tmp_packfile(char **pack_tmp_name)515{
516struct strbuf tmpname = STRBUF_INIT;517int fd;518
519fd = odb_mkstemp(&tmpname, "pack/tmp_pack_XXXXXX");520*pack_tmp_name = strbuf_detach(&tmpname, NULL);521return hashfd(fd, *pack_tmp_name);522}
523
524static void rename_tmp_packfile(struct strbuf *name_prefix, const char *source,525const char *ext)526{
527size_t name_prefix_len = name_prefix->len;528
529strbuf_addstr(name_prefix, ext);530if (rename(source, name_prefix->buf))531die_errno("unable to rename temporary file to '%s'",532name_prefix->buf);533strbuf_setlen(name_prefix, name_prefix_len);534}
535
536void rename_tmp_packfile_idx(struct strbuf *name_buffer,537char **idx_tmp_name)538{
539rename_tmp_packfile(name_buffer, *idx_tmp_name, "idx");540}
541
542void stage_tmp_packfiles(struct strbuf *name_buffer,543const char *pack_tmp_name,544struct pack_idx_entry **written_list,545uint32_t nr_written,546struct packing_data *to_pack,547struct pack_idx_option *pack_idx_opts,548unsigned char hash[],549char **idx_tmp_name)550{
551const char *rev_tmp_name = NULL;552char *mtimes_tmp_name = NULL;553
554if (adjust_shared_perm(pack_tmp_name))555die_errno("unable to make temporary pack file readable");556
557*idx_tmp_name = (char *)write_idx_file(NULL, written_list, nr_written,558pack_idx_opts, hash);559if (adjust_shared_perm(*idx_tmp_name))560die_errno("unable to make temporary index file readable");561
562rev_tmp_name = write_rev_file(NULL, written_list, nr_written, hash,563pack_idx_opts->flags);564
565if (pack_idx_opts->flags & WRITE_MTIMES) {566mtimes_tmp_name = write_mtimes_file(to_pack, written_list,567nr_written,568hash);569}570
571rename_tmp_packfile(name_buffer, pack_tmp_name, "pack");572if (rev_tmp_name)573rename_tmp_packfile(name_buffer, rev_tmp_name, "rev");574if (mtimes_tmp_name)575rename_tmp_packfile(name_buffer, mtimes_tmp_name, "mtimes");576
577free((char *)rev_tmp_name);578free(mtimes_tmp_name);579}
580
581void write_promisor_file(const char *promisor_name, struct ref **sought, int nr_sought)582{
583int i, err;584FILE *output = xfopen(promisor_name, "w");585
586for (i = 0; i < nr_sought; i++)587fprintf(output, "%s %s\n", oid_to_hex(&sought[i]->old_oid),588sought[i]->name);589
590err = ferror(output);591err |= fclose(output);592if (err)593die(_("could not write '%s' promisor file"), promisor_name);594}
595