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