2 * Copyright (c) 2011, Google Inc.
5 #include "bulk-checkin.h"
6 #include "repository.h"
11 #include "object-store.h"
13 static struct bulk_checkin_state {
19 struct pack_idx_option pack_idx_opts;
21 struct pack_idx_entry **written;
22 uint32_t alloc_written;
26 static void finish_bulk_checkin(struct bulk_checkin_state *state)
29 struct strbuf packname = STRBUF_INIT;
35 if (state->nr_written == 0) {
37 unlink(state->pack_tmp_name);
39 } else if (state->nr_written == 1) {
40 finalize_hashfile(state->f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
42 int fd = finalize_hashfile(state->f, oid.hash, 0);
43 fixup_pack_header_footer(fd, oid.hash, state->pack_tmp_name,
44 state->nr_written, oid.hash,
49 strbuf_addf(&packname, "%s/pack/pack-", get_object_directory());
50 finish_tmp_packfile(&packname, state->pack_tmp_name,
51 state->written, state->nr_written,
52 &state->pack_idx_opts, oid.hash);
53 for (i = 0; i < state->nr_written; i++)
54 free(state->written[i]);
58 memset(state, 0, sizeof(*state));
60 strbuf_release(&packname);
61 /* Make objects we just wrote available to ourselves */
62 reprepare_packed_git(the_repository);
65 static int already_written(struct bulk_checkin_state *state, struct object_id *oid)
69 /* The object may already exist in the repository */
70 if (has_object_file(oid))
73 /* Might want to keep the list sorted */
74 for (i = 0; i < state->nr_written; i++)
75 if (oideq(&state->written[i]->oid, oid))
78 /* This is a new object we need to keep */
83 * Read the contents from fd for size bytes, streaming it to the
84 * packfile in state while updating the hash in ctx. Signal a failure
85 * by returning a negative value when the resulting pack would exceed
86 * the pack size limit and this is not the first object in the pack,
87 * so that the caller can discard what we wrote from the current pack
88 * by truncating it and opening a new one. The caller will then call
89 * us again after rewinding the input fd.
91 * The already_hashed_to pointer is kept untouched by the caller to
92 * make sure we do not hash the same byte when we are called
93 * again. This way, the caller does not have to checkpoint its hash
94 * status before calling us just in case we ask it to call us again
97 static int stream_to_pack(struct bulk_checkin_state *state,
98 git_hash_ctx *ctx, off_t *already_hashed_to,
99 int fd, size_t size, enum object_type type,
100 const char *path, unsigned flags)
103 unsigned char obuf[16384];
106 int write_object = (flags & HASH_WRITE_OBJECT);
109 git_deflate_init(&s, pack_compression_level);
111 hdrlen = encode_in_pack_object_header(obuf, sizeof(obuf), type, size);
112 s.next_out = obuf + hdrlen;
113 s.avail_out = sizeof(obuf) - hdrlen;
115 while (status != Z_STREAM_END) {
116 unsigned char ibuf[16384];
118 if (size && !s.avail_in) {
119 ssize_t rsize = size < sizeof(ibuf) ? size : sizeof(ibuf);
120 ssize_t read_result = read_in_full(fd, ibuf, rsize);
122 die_errno("failed to read from '%s'", path);
123 if (read_result != rsize)
124 die("failed to read %d bytes from '%s'",
127 if (*already_hashed_to < offset) {
128 size_t hsize = offset - *already_hashed_to;
132 the_hash_algo->update_fn(ctx, ibuf, hsize);
133 *already_hashed_to = offset;
140 status = git_deflate(&s, size ? 0 : Z_FINISH);
142 if (!s.avail_out || status == Z_STREAM_END) {
144 size_t written = s.next_out - obuf;
146 /* would we bust the size limit? */
147 if (state->nr_written &&
148 pack_size_limit_cfg &&
149 pack_size_limit_cfg < state->offset + written) {
150 git_deflate_abort(&s);
154 hashwrite(state->f, obuf, written);
155 state->offset += written;
158 s.avail_out = sizeof(obuf);
167 die("unexpected deflate failure: %d", status);
174 /* Lazily create backing packfile for the state */
175 static void prepare_to_stream(struct bulk_checkin_state *state,
178 if (!(flags & HASH_WRITE_OBJECT) || state->f)
181 state->f = create_tmp_packfile(&state->pack_tmp_name);
182 reset_pack_idx_option(&state->pack_idx_opts);
184 /* Pretend we are going to write only one object */
185 state->offset = write_pack_header(state->f, 1);
187 die_errno("unable to write pack header");
190 static int deflate_to_pack(struct bulk_checkin_state *state,
191 struct object_id *result_oid,
193 enum object_type type, const char *path,
196 off_t seekback, already_hashed_to;
198 unsigned char obuf[16384];
200 struct hashfile_checkpoint checkpoint = {0};
201 struct pack_idx_entry *idx = NULL;
203 seekback = lseek(fd, 0, SEEK_CUR);
204 if (seekback == (off_t) -1)
205 return error("cannot find the current offset");
207 header_len = xsnprintf((char *)obuf, sizeof(obuf), "%s %" PRIuMAX,
208 type_name(type), (uintmax_t)size) + 1;
209 the_hash_algo->init_fn(&ctx);
210 the_hash_algo->update_fn(&ctx, obuf, header_len);
212 /* Note: idx is non-NULL when we are writing */
213 if ((flags & HASH_WRITE_OBJECT) != 0)
214 idx = xcalloc(1, sizeof(*idx));
216 already_hashed_to = 0;
219 prepare_to_stream(state, flags);
221 hashfile_checkpoint(state->f, &checkpoint);
222 idx->offset = state->offset;
223 crc32_begin(state->f);
225 if (!stream_to_pack(state, &ctx, &already_hashed_to,
226 fd, size, type, path, flags))
229 * Writing this object to the current pack will make
230 * it too big; we need to truncate it, start a new
231 * pack, and write into it.
234 BUG("should not happen");
235 hashfile_truncate(state->f, &checkpoint);
236 state->offset = checkpoint.offset;
237 finish_bulk_checkin(state);
238 if (lseek(fd, seekback, SEEK_SET) == (off_t) -1)
239 return error("cannot seek back");
241 the_hash_algo->final_fn(result_oid->hash, &ctx);
245 idx->crc32 = crc32_end(state->f);
246 if (already_written(state, result_oid)) {
247 hashfile_truncate(state->f, &checkpoint);
248 state->offset = checkpoint.offset;
251 oidcpy(&idx->oid, result_oid);
252 ALLOC_GROW(state->written,
253 state->nr_written + 1,
254 state->alloc_written);
255 state->written[state->nr_written++] = idx;
260 int index_bulk_checkin(struct object_id *oid,
261 int fd, size_t size, enum object_type type,
262 const char *path, unsigned flags)
264 int status = deflate_to_pack(&state, oid, fd, size, type,
267 finish_bulk_checkin(&state);
271 void plug_bulk_checkin(void)
276 void unplug_bulk_checkin(void)
280 finish_bulk_checkin(&state);