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 struct bulk_checkin_state {
  14         unsigned plugged:1;
  15
  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 } state;
  25
  26 static void finish_bulk_checkin(struct bulk_checkin_state *state)
  27 {
  28         struct object_id oid;
  29         struct strbuf packname = STRBUF_INIT;
  30         int i;
  31
  32         if (!state->f)
  33                 return;
  34
  35         if (state->nr_written == 0) {
  36                 close(state->f->fd);
  37                 unlink(state->pack_tmp_name);
  38                 goto clear_exit;
  39         } else if (state->nr_written == 1) {
  40                 finalize_hashfile(state->f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
  41         } else {
  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,
  45                                          state->offset);
  46                 close(fd);
  47         }
  48
  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]);
  55
  56 clear_exit:
  57         free(state->written);
  58         memset(state, 0, sizeof(*state));
  59
  60         strbuf_release(&packname);
  61         /* Make objects we just wrote available to ourselves */
  62         reprepare_packed_git(the_repository);
  63 }
  64
  65 static int already_written(struct bulk_checkin_state *state, struct object_id *oid)
  66 {
  67         int i;
  68
  69         /* The object may already exist in the repository */
  70         if (has_object_file(oid))
  71                 return 1;
  72
  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))
  76                         return 1;
  77
  78         /* This is a new object we need to keep */
  79         return 0;
  80 }
  81
  82 /*
  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.
  90  *
  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
  95  * with a new pack.
  96  */
  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)
 101 {
 102         git_zstream s;
 103         unsigned char obuf[16384];
 104         unsigned hdrlen;
 105         int status = Z_OK;
 106         int write_object = (flags & HASH_WRITE_OBJECT);
 107         off_t offset = 0;
 108
 109         git_deflate_init(&s, pack_compression_level);
 110
 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;
 114
 115         while (status != Z_STREAM_END) {
 116                 unsigned char ibuf[16384];
 117
 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);
 121                         if (read_result < 0)
 122                                 die_errno("failed to read from '%s'", path);
 123                         if (read_result != rsize)
 124                                 die("failed to read %d bytes from '%s'",
 125                                     (int)rsize, path);
 126                         offset += rsize;
 127                         if (*already_hashed_to < offset) {
 128                                 size_t hsize = offset - *already_hashed_to;
 129                                 if (rsize < hsize)
 130                                         hsize = rsize;
 131                                 if (hsize)
 132                                         the_hash_algo->update_fn(ctx, ibuf, hsize);
 133                                 *already_hashed_to = offset;
 134                         }
 135                         s.next_in = ibuf;
 136                         s.avail_in = rsize;
 137                         size -= rsize;
 138                 }
 139
 140                 status = git_deflate(&s, size ? 0 : Z_FINISH);
 141
 142                 if (!s.avail_out || status == Z_STREAM_END) {
 143                         if (write_object) {
 144                                 size_t written = s.next_out - obuf;
 145
 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);
 151                                         return -1;
 152                                 }
 153
 154                                 hashwrite(state->f, obuf, written);
 155                                 state->offset += written;
 156                         }
 157                         s.next_out = obuf;
 158                         s.avail_out = sizeof(obuf);
 159                 }
 160
 161                 switch (status) {
 162                 case Z_OK:
 163                 case Z_BUF_ERROR:
 164                 case Z_STREAM_END:
 165                         continue;
 166                 default:
 167                         die("unexpected deflate failure: %d", status);
 168                 }
 169         }
 170         git_deflate_end(&s);
 171         return 0;
 172 }
 173
 174 /* Lazily create backing packfile for the state */
 175 static void prepare_to_stream(struct bulk_checkin_state *state,
 176                               unsigned flags)
 177 {
 178         if (!(flags & HASH_WRITE_OBJECT) || state->f)
 179                 return;
 180
 181         state->f = create_tmp_packfile(&state->pack_tmp_name);
 182         reset_pack_idx_option(&state->pack_idx_opts);
 183
 184         /* Pretend we are going to write only one object */
 185         state->offset = write_pack_header(state->f, 1);
 186         if (!state->offset)
 187                 die_errno("unable to write pack header");
 188 }
 189
 190 static int deflate_to_pack(struct bulk_checkin_state *state,
 191                            struct object_id *result_oid,
 192                            int fd, size_t size,
 193                            enum object_type type, const char *path,
 194                            unsigned flags)
 195 {
 196         off_t seekback, already_hashed_to;
 197         git_hash_ctx ctx;
 198         unsigned char obuf[16384];
 199         unsigned header_len;
 200         struct hashfile_checkpoint checkpoint = {0};
 201         struct pack_idx_entry *idx = NULL;
 202
 203         seekback = lseek(fd, 0, SEEK_CUR);
 204         if (seekback == (off_t) -1)
 205                 return error("cannot find the current offset");
 206
 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);
 211
 212         /* Note: idx is non-NULL when we are writing */
 213         if ((flags & HASH_WRITE_OBJECT) != 0)
 214                 idx = xcalloc(1, sizeof(*idx));
 215
 216         already_hashed_to = 0;
 217
 218         while (1) {
 219                 prepare_to_stream(state, flags);
 220                 if (idx) {
 221                         hashfile_checkpoint(state->f, &checkpoint);
 222                         idx->offset = state->offset;
 223                         crc32_begin(state->f);
 224                 }
 225                 if (!stream_to_pack(state, &ctx, &already_hashed_to,
 226                                     fd, size, type, path, flags))
 227                         break;
 228                 /*
 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.
 232                  */
 233                 if (!idx)
 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");
 240         }
 241         the_hash_algo->final_fn(result_oid->hash, &ctx);
 242         if (!idx)
 243                 return 0;
 244
 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;
 249                 free(idx);
 250         } else {
 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;
 256         }
 257         return 0;
 258 }
 259
 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)
 263 {
 264         int status = deflate_to_pack(&state, oid, fd, size, type,
 265                                      path, flags);
 266         if (!state.plugged)
 267                 finish_bulk_checkin(&state);
 268         return status;
 269 }
 270
 271 void plug_bulk_checkin(void)
 272 {
 273         state.plugged = 1;
 274 }
 275
 276 void unplug_bulk_checkin(void)
 277 {
 278         state.plugged = 0;
 279         if (state.f)
 280                 finish_bulk_checkin(&state);
 281 }