2 * Copyright (c) 2011, Google Inc.
6 #include "repository.h"
7 #include "object-store.h"
8 #include "replace-object.h"
18 typedef int (*open_istream_fn)(struct git_istream *,
21 const struct object_id *,
23 typedef int (*close_istream_fn)(struct git_istream *);
24 typedef ssize_t (*read_istream_fn)(struct git_istream *, char *, size_t);
27 close_istream_fn close;
31 #define open_method_decl(name) \
32 int open_istream_ ##name \
33 (struct git_istream *st, struct repository *r, \
34 struct object_info *oi, const struct object_id *oid, \
35 enum object_type *type)
37 #define close_method_decl(name) \
38 int close_istream_ ##name \
39 (struct git_istream *st)
41 #define read_method_decl(name) \
42 ssize_t read_istream_ ##name \
43 (struct git_istream *st, char *buf, size_t sz)
45 /* forward declaration */
46 static open_method_decl(incore);
47 static open_method_decl(loose);
48 static open_method_decl(pack_non_delta);
49 static struct git_istream *attach_stream_filter(struct git_istream *st,
50 struct stream_filter *filter);
53 static open_istream_fn open_istream_tbl[] = {
56 open_istream_pack_non_delta,
59 #define FILTER_BUFFER (1024*16)
61 struct filtered_istream {
62 struct git_istream *upstream;
63 struct stream_filter *filter;
64 char ibuf[FILTER_BUFFER];
65 char obuf[FILTER_BUFFER];
72 const struct stream_vtbl *vtbl;
73 unsigned long size; /* inflated size of full object */
75 enum { z_unused, z_used, z_done, z_error } z_state;
79 char *buf; /* from read_object() */
80 unsigned long read_ptr;
85 unsigned long mapsize;
92 struct packed_git *pack;
96 struct filtered_istream filtered;
100 int close_istream(struct git_istream *st)
102 int r = st->vtbl->close(st);
107 ssize_t read_istream(struct git_istream *st, void *buf, size_t sz)
109 return st->vtbl->read(st, buf, sz);
112 static enum input_source istream_source(struct repository *r,
113 const struct object_id *oid,
114 enum object_type *type,
115 struct object_info *oi)
122 status = oid_object_info_extended(r, oid, oi, 0);
126 switch (oi->whence) {
130 if (!oi->u.packed.is_delta && big_file_threshold < size)
131 return pack_non_delta;
138 struct git_istream *open_istream(struct repository *r,
139 const struct object_id *oid,
140 enum object_type *type,
142 struct stream_filter *filter)
144 struct git_istream *st;
145 struct object_info oi = OBJECT_INFO_INIT;
146 const struct object_id *real = lookup_replace_object(r, oid);
147 enum input_source src = istream_source(r, real, type, &oi);
152 st = xmalloc(sizeof(*st));
153 if (open_istream_tbl[src](st, r, &oi, real, type)) {
154 if (open_istream_incore(st, r, &oi, real, type)) {
160 /* Add "&& !is_null_stream_filter(filter)" for performance */
161 struct git_istream *nst = attach_stream_filter(st, filter);
174 /*****************************************************************
178 *****************************************************************/
180 static void close_deflated_stream(struct git_istream *st)
182 if (st->z_state == z_used)
183 git_inflate_end(&st->z);
187 /*****************************************************************
191 *****************************************************************/
193 static close_method_decl(filtered)
195 free_stream_filter(st->u.filtered.filter);
196 return close_istream(st->u.filtered.upstream);
199 static read_method_decl(filtered)
201 struct filtered_istream *fs = &(st->u.filtered);
205 /* do we already have filtered output? */
206 if (fs->o_ptr < fs->o_end) {
207 size_t to_move = fs->o_end - fs->o_ptr;
210 memcpy(buf + filled, fs->obuf + fs->o_ptr, to_move);
211 fs->o_ptr += to_move;
216 fs->o_end = fs->o_ptr = 0;
218 /* do we have anything to feed the filter with? */
219 if (fs->i_ptr < fs->i_end) {
220 size_t to_feed = fs->i_end - fs->i_ptr;
221 size_t to_receive = FILTER_BUFFER;
222 if (stream_filter(fs->filter,
223 fs->ibuf + fs->i_ptr, &to_feed,
224 fs->obuf, &to_receive))
226 fs->i_ptr = fs->i_end - to_feed;
227 fs->o_end = FILTER_BUFFER - to_receive;
231 /* tell the filter to drain upon no more input */
232 if (fs->input_finished) {
233 size_t to_receive = FILTER_BUFFER;
234 if (stream_filter(fs->filter,
236 fs->obuf, &to_receive))
238 fs->o_end = FILTER_BUFFER - to_receive;
243 fs->i_end = fs->i_ptr = 0;
245 /* refill the input from the upstream */
246 if (!fs->input_finished) {
247 fs->i_end = read_istream(fs->upstream, fs->ibuf, FILTER_BUFFER);
253 fs->input_finished = 1;
258 static struct stream_vtbl filtered_vtbl = {
259 close_istream_filtered,
260 read_istream_filtered,
263 static struct git_istream *attach_stream_filter(struct git_istream *st,
264 struct stream_filter *filter)
266 struct git_istream *ifs = xmalloc(sizeof(*ifs));
267 struct filtered_istream *fs = &(ifs->u.filtered);
269 ifs->vtbl = &filtered_vtbl;
272 fs->i_end = fs->i_ptr = 0;
273 fs->o_end = fs->o_ptr = 0;
274 fs->input_finished = 0;
275 ifs->size = -1; /* unknown */
279 /*****************************************************************
281 * Loose object stream
283 *****************************************************************/
285 static read_method_decl(loose)
287 size_t total_read = 0;
289 switch (st->z_state) {
298 if (st->u.loose.hdr_used < st->u.loose.hdr_avail) {
299 size_t to_copy = st->u.loose.hdr_avail - st->u.loose.hdr_used;
302 memcpy(buf, st->u.loose.hdr + st->u.loose.hdr_used, to_copy);
303 st->u.loose.hdr_used += to_copy;
304 total_read += to_copy;
307 while (total_read < sz) {
310 st->z.next_out = (unsigned char *)buf + total_read;
311 st->z.avail_out = sz - total_read;
312 status = git_inflate(&st->z, Z_FINISH);
314 total_read = st->z.next_out - (unsigned char *)buf;
316 if (status == Z_STREAM_END) {
317 git_inflate_end(&st->z);
318 st->z_state = z_done;
321 if (status != Z_OK && (status != Z_BUF_ERROR || total_read < sz)) {
322 git_inflate_end(&st->z);
323 st->z_state = z_error;
330 static close_method_decl(loose)
332 close_deflated_stream(st);
333 munmap(st->u.loose.mapped, st->u.loose.mapsize);
337 static struct stream_vtbl loose_vtbl = {
342 static open_method_decl(loose)
344 st->u.loose.mapped = map_loose_object(r, oid, &st->u.loose.mapsize);
345 if (!st->u.loose.mapped)
347 if ((unpack_loose_header(&st->z,
351 sizeof(st->u.loose.hdr)) < 0) ||
352 (parse_loose_header(st->u.loose.hdr, &st->size) < 0)) {
353 git_inflate_end(&st->z);
354 munmap(st->u.loose.mapped, st->u.loose.mapsize);
358 st->u.loose.hdr_used = strlen(st->u.loose.hdr) + 1;
359 st->u.loose.hdr_avail = st->z.total_out;
360 st->z_state = z_used;
362 st->vtbl = &loose_vtbl;
367 /*****************************************************************
369 * Non-delta packed object stream
371 *****************************************************************/
373 static read_method_decl(pack_non_delta)
375 size_t total_read = 0;
377 switch (st->z_state) {
379 memset(&st->z, 0, sizeof(st->z));
380 git_inflate_init(&st->z);
381 st->z_state = z_used;
391 while (total_read < sz) {
393 struct pack_window *window = NULL;
394 unsigned char *mapped;
396 mapped = use_pack(st->u.in_pack.pack, &window,
397 st->u.in_pack.pos, &st->z.avail_in);
399 st->z.next_out = (unsigned char *)buf + total_read;
400 st->z.avail_out = sz - total_read;
401 st->z.next_in = mapped;
402 status = git_inflate(&st->z, Z_FINISH);
404 st->u.in_pack.pos += st->z.next_in - mapped;
405 total_read = st->z.next_out - (unsigned char *)buf;
408 if (status == Z_STREAM_END) {
409 git_inflate_end(&st->z);
410 st->z_state = z_done;
415 * Unlike the loose object case, we do not have to worry here
416 * about running out of input bytes and spinning infinitely. If
417 * we get Z_BUF_ERROR due to too few input bytes, then we'll
418 * replenish them in the next use_pack() call when we loop. If
419 * we truly hit the end of the pack (i.e., because it's corrupt
420 * or truncated), then use_pack() catches that and will die().
422 if (status != Z_OK && status != Z_BUF_ERROR) {
423 git_inflate_end(&st->z);
424 st->z_state = z_error;
431 static close_method_decl(pack_non_delta)
433 close_deflated_stream(st);
437 static struct stream_vtbl pack_non_delta_vtbl = {
438 close_istream_pack_non_delta,
439 read_istream_pack_non_delta,
442 static open_method_decl(pack_non_delta)
444 struct pack_window *window;
445 enum object_type in_pack_type;
447 st->u.in_pack.pack = oi->u.packed.pack;
448 st->u.in_pack.pos = oi->u.packed.offset;
451 in_pack_type = unpack_object_header(st->u.in_pack.pack,
456 switch (in_pack_type) {
458 return -1; /* we do not do deltas for now */
465 st->z_state = z_unused;
466 st->vtbl = &pack_non_delta_vtbl;
471 /*****************************************************************
475 *****************************************************************/
477 static close_method_decl(incore)
479 free(st->u.incore.buf);
483 static read_method_decl(incore)
485 size_t read_size = sz;
486 size_t remainder = st->size - st->u.incore.read_ptr;
488 if (remainder <= read_size)
489 read_size = remainder;
491 memcpy(buf, st->u.incore.buf + st->u.incore.read_ptr, read_size);
492 st->u.incore.read_ptr += read_size;
497 static struct stream_vtbl incore_vtbl = {
498 close_istream_incore,
502 static open_method_decl(incore)
504 st->u.incore.buf = read_object_file_extended(r, oid, type, &st->size, 0);
505 st->u.incore.read_ptr = 0;
506 st->vtbl = &incore_vtbl;
508 return st->u.incore.buf ? 0 : -1;
512 /****************************************************************
513 * Users of streaming interface
514 ****************************************************************/
516 int stream_blob_to_fd(int fd, const struct object_id *oid, struct stream_filter *filter,
519 struct git_istream *st;
520 enum object_type type;
525 st = open_istream(the_repository, oid, &type, &sz, filter);
528 free_stream_filter(filter);
531 if (type != OBJ_BLOB)
535 ssize_t wrote, holeto;
536 ssize_t readlen = read_istream(st, buf, sizeof(buf));
542 if (can_seek && sizeof(buf) == readlen) {
543 for (holeto = 0; holeto < readlen; holeto++)
546 if (readlen == holeto) {
552 if (kept && lseek(fd, kept, SEEK_CUR) == (off_t) -1)
556 wrote = write_in_full(fd, buf, readlen);
561 if (kept && (lseek(fd, kept - 1, SEEK_CUR) == (off_t) -1 ||
562 xwrite(fd, "", 1) != 1))