]> git.ipfire.org Git - thirdparty/git.git/blob - bulk-checkin.c
81560e87d95ae9b71fc556a9a919918f623bf0be
[thirdparty/git.git] / bulk-checkin.c
1 /*
2 * Copyright (c) 2011, Google Inc.
3 */
4 #include "cache.h"
5 #include "bulk-checkin.h"
6 #include "repository.h"
7 #include "csum-file.h"
8 #include "pack.h"
9 #include "strbuf.h"
10 #include "packfile.h"
11 #include "object-store.h"
12
13 static int bulk_checkin_plugged;
14
15 static struct bulk_checkin_packfile {
16 char *pack_tmp_name;
17 struct hashfile *f;
18 off_t offset;
19 struct pack_idx_option pack_idx_opts;
20
21 struct pack_idx_entry **written;
22 uint32_t alloc_written;
23 uint32_t nr_written;
24 } bulk_checkin_packfile;
25
26 static void finish_tmp_packfile(struct strbuf *basename,
27 const char *pack_tmp_name,
28 struct pack_idx_entry **written_list,
29 uint32_t nr_written,
30 struct pack_idx_option *pack_idx_opts,
31 unsigned char hash[])
32 {
33 char *idx_tmp_name = NULL;
34
35 stage_tmp_packfiles(basename, pack_tmp_name, written_list, nr_written,
36 pack_idx_opts, hash, &idx_tmp_name);
37 rename_tmp_packfile_idx(basename, &idx_tmp_name);
38
39 free(idx_tmp_name);
40 }
41
42 static void flush_bulk_checkin_packfile(struct bulk_checkin_packfile *state)
43 {
44 unsigned char hash[GIT_MAX_RAWSZ];
45 struct strbuf packname = STRBUF_INIT;
46 int i;
47
48 if (!state->f)
49 return;
50
51 if (state->nr_written == 0) {
52 close(state->f->fd);
53 unlink(state->pack_tmp_name);
54 goto clear_exit;
55 } else if (state->nr_written == 1) {
56 finalize_hashfile(state->f, hash, FSYNC_COMPONENT_PACK,
57 CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
58 } else {
59 int fd = finalize_hashfile(state->f, hash, FSYNC_COMPONENT_PACK, 0);
60 fixup_pack_header_footer(fd, hash, state->pack_tmp_name,
61 state->nr_written, hash,
62 state->offset);
63 close(fd);
64 }
65
66 strbuf_addf(&packname, "%s/pack/pack-%s.", get_object_directory(),
67 hash_to_hex(hash));
68 finish_tmp_packfile(&packname, state->pack_tmp_name,
69 state->written, state->nr_written,
70 &state->pack_idx_opts, hash);
71 for (i = 0; i < state->nr_written; i++)
72 free(state->written[i]);
73
74 clear_exit:
75 free(state->written);
76 memset(state, 0, sizeof(*state));
77
78 strbuf_release(&packname);
79 /* Make objects we just wrote available to ourselves */
80 reprepare_packed_git(the_repository);
81 }
82
83 static int already_written(struct bulk_checkin_packfile *state, struct object_id *oid)
84 {
85 int i;
86
87 /* The object may already exist in the repository */
88 if (has_object_file(oid))
89 return 1;
90
91 /* Might want to keep the list sorted */
92 for (i = 0; i < state->nr_written; i++)
93 if (oideq(&state->written[i]->oid, oid))
94 return 1;
95
96 /* This is a new object we need to keep */
97 return 0;
98 }
99
100 /*
101 * Read the contents from fd for size bytes, streaming it to the
102 * packfile in state while updating the hash in ctx. Signal a failure
103 * by returning a negative value when the resulting pack would exceed
104 * the pack size limit and this is not the first object in the pack,
105 * so that the caller can discard what we wrote from the current pack
106 * by truncating it and opening a new one. The caller will then call
107 * us again after rewinding the input fd.
108 *
109 * The already_hashed_to pointer is kept untouched by the caller to
110 * make sure we do not hash the same byte when we are called
111 * again. This way, the caller does not have to checkpoint its hash
112 * status before calling us just in case we ask it to call us again
113 * with a new pack.
114 */
115 static int stream_to_pack(struct bulk_checkin_packfile *state,
116 git_hash_ctx *ctx, off_t *already_hashed_to,
117 int fd, size_t size, enum object_type type,
118 const char *path, unsigned flags)
119 {
120 git_zstream s;
121 unsigned char ibuf[16384];
122 unsigned char obuf[16384];
123 unsigned hdrlen;
124 int status = Z_OK;
125 int write_object = (flags & HASH_WRITE_OBJECT);
126 off_t offset = 0;
127
128 git_deflate_init(&s, pack_compression_level);
129
130 hdrlen = encode_in_pack_object_header(obuf, sizeof(obuf), type, size);
131 s.next_out = obuf + hdrlen;
132 s.avail_out = sizeof(obuf) - hdrlen;
133
134 while (status != Z_STREAM_END) {
135 if (size && !s.avail_in) {
136 ssize_t rsize = size < sizeof(ibuf) ? size : sizeof(ibuf);
137 ssize_t read_result = read_in_full(fd, ibuf, rsize);
138 if (read_result < 0)
139 die_errno("failed to read from '%s'", path);
140 if (read_result != rsize)
141 die("failed to read %d bytes from '%s'",
142 (int)rsize, path);
143 offset += rsize;
144 if (*already_hashed_to < offset) {
145 size_t hsize = offset - *already_hashed_to;
146 if (rsize < hsize)
147 hsize = rsize;
148 if (hsize)
149 the_hash_algo->update_fn(ctx, ibuf, hsize);
150 *already_hashed_to = offset;
151 }
152 s.next_in = ibuf;
153 s.avail_in = rsize;
154 size -= rsize;
155 }
156
157 status = git_deflate(&s, size ? 0 : Z_FINISH);
158
159 if (!s.avail_out || status == Z_STREAM_END) {
160 if (write_object) {
161 size_t written = s.next_out - obuf;
162
163 /* would we bust the size limit? */
164 if (state->nr_written &&
165 pack_size_limit_cfg &&
166 pack_size_limit_cfg < state->offset + written) {
167 git_deflate_abort(&s);
168 return -1;
169 }
170
171 hashwrite(state->f, obuf, written);
172 state->offset += written;
173 }
174 s.next_out = obuf;
175 s.avail_out = sizeof(obuf);
176 }
177
178 switch (status) {
179 case Z_OK:
180 case Z_BUF_ERROR:
181 case Z_STREAM_END:
182 continue;
183 default:
184 die("unexpected deflate failure: %d", status);
185 }
186 }
187 git_deflate_end(&s);
188 return 0;
189 }
190
191 /* Lazily create backing packfile for the state */
192 static void prepare_to_stream(struct bulk_checkin_packfile *state,
193 unsigned flags)
194 {
195 if (!(flags & HASH_WRITE_OBJECT) || state->f)
196 return;
197
198 state->f = create_tmp_packfile(&state->pack_tmp_name);
199 reset_pack_idx_option(&state->pack_idx_opts);
200
201 /* Pretend we are going to write only one object */
202 state->offset = write_pack_header(state->f, 1);
203 if (!state->offset)
204 die_errno("unable to write pack header");
205 }
206
207 static int deflate_to_pack(struct bulk_checkin_packfile *state,
208 struct object_id *result_oid,
209 int fd, size_t size,
210 enum object_type type, const char *path,
211 unsigned flags)
212 {
213 off_t seekback, already_hashed_to;
214 git_hash_ctx ctx;
215 unsigned char obuf[16384];
216 unsigned header_len;
217 struct hashfile_checkpoint checkpoint = {0};
218 struct pack_idx_entry *idx = NULL;
219
220 seekback = lseek(fd, 0, SEEK_CUR);
221 if (seekback == (off_t) -1)
222 return error("cannot find the current offset");
223
224 header_len = xsnprintf((char *)obuf, sizeof(obuf), "%s %" PRIuMAX,
225 type_name(type), (uintmax_t)size) + 1;
226 the_hash_algo->init_fn(&ctx);
227 the_hash_algo->update_fn(&ctx, obuf, header_len);
228
229 /* Note: idx is non-NULL when we are writing */
230 if ((flags & HASH_WRITE_OBJECT) != 0)
231 CALLOC_ARRAY(idx, 1);
232
233 already_hashed_to = 0;
234
235 while (1) {
236 prepare_to_stream(state, flags);
237 if (idx) {
238 hashfile_checkpoint(state->f, &checkpoint);
239 idx->offset = state->offset;
240 crc32_begin(state->f);
241 }
242 if (!stream_to_pack(state, &ctx, &already_hashed_to,
243 fd, size, type, path, flags))
244 break;
245 /*
246 * Writing this object to the current pack will make
247 * it too big; we need to truncate it, start a new
248 * pack, and write into it.
249 */
250 if (!idx)
251 BUG("should not happen");
252 hashfile_truncate(state->f, &checkpoint);
253 state->offset = checkpoint.offset;
254 flush_bulk_checkin_packfile(state);
255 if (lseek(fd, seekback, SEEK_SET) == (off_t) -1)
256 return error("cannot seek back");
257 }
258 the_hash_algo->final_oid_fn(result_oid, &ctx);
259 if (!idx)
260 return 0;
261
262 idx->crc32 = crc32_end(state->f);
263 if (already_written(state, result_oid)) {
264 hashfile_truncate(state->f, &checkpoint);
265 state->offset = checkpoint.offset;
266 free(idx);
267 } else {
268 oidcpy(&idx->oid, result_oid);
269 ALLOC_GROW(state->written,
270 state->nr_written + 1,
271 state->alloc_written);
272 state->written[state->nr_written++] = idx;
273 }
274 return 0;
275 }
276
277 int index_bulk_checkin(struct object_id *oid,
278 int fd, size_t size, enum object_type type,
279 const char *path, unsigned flags)
280 {
281 int status = deflate_to_pack(&bulk_checkin_packfile, oid, fd, size, type,
282 path, flags);
283 if (!bulk_checkin_plugged)
284 flush_bulk_checkin_packfile(&bulk_checkin_packfile);
285 return status;
286 }
287
288 void plug_bulk_checkin(void)
289 {
290 assert(!bulk_checkin_plugged);
291 bulk_checkin_plugged = 1;
292 }
293
294 void unplug_bulk_checkin(void)
295 {
296 assert(bulk_checkin_plugged);
297 bulk_checkin_plugged = 0;
298 flush_bulk_checkin_packfile(&bulk_checkin_packfile);
299 }