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