csum-file.c

   1 /*
   2  * csum-file.c
   3  *
   4  * Copyright (C) 2005 Linus Torvalds
   5  *
   6  * Simple file write infrastructure for writing SHA1-summed
   7  * files. Useful when you write a file that you want to be
   8  * able to verify hasn't been messed with afterwards.
   9  */
  10
  11 #define USE_THE_REPOSITORY_VARIABLE
  12
  13 #include "git-compat-util.h"
  14 #include "progress.h"
  15 #include "csum-file.h"
  16 #include "hash.h"
  17
  18 static void verify_buffer_or_die(struct hashfile *f,
  19                                  const void *buf,
  20                                  unsigned int count)
  21 {
  22         ssize_t ret = read_in_full(f->check_fd, f->check_buffer, count);
  23
  24         if (ret < 0)
  25                 die_errno("%s: sha1 file read error", f->name);
  26         if (ret != count)
  27                 die("%s: sha1 file truncated", f->name);
  28         if (memcmp(buf, f->check_buffer, count))
  29                 die("sha1 file '%s' validation error", f->name);
  30 }
  31
  32 static void flush(struct hashfile *f, const void *buf, unsigned int count)
  33 {
  34         if (0 <= f->check_fd && count)
  35                 verify_buffer_or_die(f, buf, count);
  36
  37         if (write_in_full(f->fd, buf, count) < 0) {
  38                 if (errno == ENOSPC)
  39                         die("sha1 file '%s' write error. Out of diskspace", f->name);
  40                 die_errno("sha1 file '%s' write error", f->name);
  41         }
  42
  43         f->total += count;
  44         display_throughput(f->tp, f->total);
  45 }
  46
  47 void hashflush(struct hashfile *f)
  48 {
  49         unsigned offset = f->offset;
  50
  51         if (offset) {
  52                 if (!f->skip_hash)
  53                         the_hash_algo->update_fn(&f->ctx, f->buffer, offset);
  54                 flush(f, f->buffer, offset);
  55                 f->offset = 0;
  56         }
  57 }
  58
  59 static void free_hashfile(struct hashfile *f)
  60 {
  61         free(f->buffer);
  62         free(f->check_buffer);
  63         free(f);
  64 }
  65
  66 int finalize_hashfile(struct hashfile *f, unsigned char *result,
  67                       enum fsync_component component, unsigned int flags)
  68 {
  69         int fd;
  70
  71         hashflush(f);
  72
  73         if (f->skip_hash)
  74                 hashclr(f->buffer, the_repository->hash_algo);
  75         else
  76                 the_hash_algo->final_fn(f->buffer, &f->ctx);
  77
  78         if (result)
  79                 hashcpy(result, f->buffer, the_repository->hash_algo);
  80         if (flags & CSUM_HASH_IN_STREAM)
  81                 flush(f, f->buffer, the_hash_algo->rawsz);
  82         if (flags & CSUM_FSYNC)
  83                 fsync_component_or_die(component, f->fd, f->name);
  84         if (flags & CSUM_CLOSE) {
  85                 if (close(f->fd))
  86                         die_errno("%s: sha1 file error on close", f->name);
  87                 fd = 0;
  88         } else
  89                 fd = f->fd;
  90         if (0 <= f->check_fd) {
  91                 char discard;
  92                 int cnt = read_in_full(f->check_fd, &discard, 1);
  93                 if (cnt < 0)
  94                         die_errno("%s: error when reading the tail of sha1 file",
  95                                   f->name);
  96                 if (cnt)
  97                         die("%s: sha1 file has trailing garbage", f->name);
  98                 if (close(f->check_fd))
  99                         die_errno("%s: sha1 file error on close", f->name);
 100         }
 101         free_hashfile(f);
 102         return fd;
 103 }
 104
 105 void hashwrite(struct hashfile *f, const void *buf, unsigned int count)
 106 {
 107         while (count) {
 108                 unsigned left = f->buffer_len - f->offset;
 109                 unsigned nr = count > left ? left : count;
 110
 111                 if (f->do_crc)
 112                         f->crc32 = crc32(f->crc32, buf, nr);
 113
 114                 if (nr == f->buffer_len) {
 115                         /*
 116                          * Flush a full batch worth of data directly
 117                          * from the input, skipping the memcpy() to
 118                          * the hashfile's buffer. In this block,
 119                          * f->offset is necessarily zero.
 120                          */
 121                         if (!f->skip_hash)
 122                                 the_hash_algo->update_fn(&f->ctx, buf, nr);
 123                         flush(f, buf, nr);
 124                 } else {
 125                         /*
 126                          * Copy to the hashfile's buffer, flushing only
 127                          * if it became full.
 128                          */
 129                         memcpy(f->buffer + f->offset, buf, nr);
 130                         f->offset += nr;
 131                         left -= nr;
 132                         if (!left)
 133                                 hashflush(f);
 134                 }
 135
 136                 count -= nr;
 137                 buf = (char *) buf + nr;
 138         }
 139 }
 140
 141 struct hashfile *hashfd_check(const char *name)
 142 {
 143         int sink, check;
 144         struct hashfile *f;
 145
 146         sink = xopen("/dev/null", O_WRONLY);
 147         check = xopen(name, O_RDONLY);
 148         f = hashfd(sink, name);
 149         f->check_fd = check;
 150         f->check_buffer = xmalloc(f->buffer_len);
 151
 152         return f;
 153 }
 154
 155 static struct hashfile *hashfd_internal(int fd, const char *name,
 156                                         struct progress *tp,
 157                                         size_t buffer_len)
 158 {
 159         struct hashfile *f = xmalloc(sizeof(*f));
 160         f->fd = fd;
 161         f->check_fd = -1;
 162         f->offset = 0;
 163         f->total = 0;
 164         f->tp = tp;
 165         f->name = name;
 166         f->do_crc = 0;
 167         f->skip_hash = 0;
 168         the_hash_algo->init_fn(&f->ctx);
 169
 170         f->buffer_len = buffer_len;
 171         f->buffer = xmalloc(buffer_len);
 172         f->check_buffer = NULL;
 173
 174         return f;
 175 }
 176
 177 struct hashfile *hashfd(int fd, const char *name)
 178 {
 179         /*
 180          * Since we are not going to use a progress meter to
 181          * measure the rate of data passing through this hashfile,
 182          * use a larger buffer size to reduce fsync() calls.
 183          */
 184         return hashfd_internal(fd, name, NULL, 128 * 1024);
 185 }
 186
 187 struct hashfile *hashfd_throughput(int fd, const char *name, struct progress *tp)
 188 {
 189         /*
 190          * Since we are expecting to report progress of the
 191          * write into this hashfile, use a smaller buffer
 192          * size so the progress indicators arrive at a more
 193          * frequent rate.
 194          */
 195         return hashfd_internal(fd, name, tp, 8 * 1024);
 196 }
 197
 198 void hashfile_checkpoint(struct hashfile *f, struct hashfile_checkpoint *checkpoint)
 199 {
 200         hashflush(f);
 201         checkpoint->offset = f->total;
 202         the_hash_algo->clone_fn(&checkpoint->ctx, &f->ctx);
 203 }
 204
 205 int hashfile_truncate(struct hashfile *f, struct hashfile_checkpoint *checkpoint)
 206 {
 207         off_t offset = checkpoint->offset;
 208
 209         if (ftruncate(f->fd, offset) ||
 210             lseek(f->fd, offset, SEEK_SET) != offset)
 211                 return -1;
 212         f->total = offset;
 213         the_hash_algo->clone_fn(&f->ctx, &checkpoint->ctx);
 214         f->offset = 0; /* hashflush() was called in checkpoint */
 215         return 0;
 216 }
 217
 218 void crc32_begin(struct hashfile *f)
 219 {
 220         f->crc32 = crc32(0, NULL, 0);
 221         f->do_crc = 1;
 222 }
 223
 224 uint32_t crc32_end(struct hashfile *f)
 225 {
 226         f->do_crc = 0;
 227         return f->crc32;
 228 }
 229
 230 int hashfile_checksum_valid(const unsigned char *data, size_t total_len)
 231 {
 232         unsigned char got[GIT_MAX_RAWSZ];
 233         git_hash_ctx ctx;
 234         size_t data_len = total_len - the_hash_algo->rawsz;
 235
 236         if (total_len < the_hash_algo->rawsz)
 237                 return 0; /* say "too short"? */
 238
 239         the_hash_algo->init_fn(&ctx);
 240         the_hash_algo->update_fn(&ctx, data, data_len);
 241         the_hash_algo->final_fn(got, &ctx);
 242
 243         return hasheq(got, data + data_len, the_repository->hash_algo);
 244 }