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                 if (!f->skip_hash)
  49                         the_hash_algo->update_fn(&f->ctx, f->buffer, offset);
  50                 flush(f, f->buffer, offset);
  51                 f->offset = 0;
  52         }
  53 }
  54
  55 static void free_hashfile(struct hashfile *f)
  56 {
  57         free(f->buffer);
  58         free(f->check_buffer);
  59         free(f);
  60 }
  61
  62 int finalize_hashfile(struct hashfile *f, unsigned char *result,
  63                       enum fsync_component component, unsigned int flags)
  64 {
  65         int fd;
  66
  67         hashflush(f);
  68
  69         if (f->skip_hash)
  70                 hashclr(f->buffer);
  71         else
  72                 the_hash_algo->final_fn(f->buffer, &f->ctx);
  73
  74         if (result)
  75                 hashcpy(result, f->buffer);
  76         if (flags & CSUM_HASH_IN_STREAM)
  77                 flush(f, f->buffer, the_hash_algo->rawsz);
  78         if (flags & CSUM_FSYNC)
  79                 fsync_component_or_die(component, f->fd, f->name);
  80         if (flags & CSUM_CLOSE) {
  81                 if (close(f->fd))
  82                         die_errno("%s: sha1 file error on close", f->name);
  83                 fd = 0;
  84         } else
  85                 fd = f->fd;
  86         if (0 <= f->check_fd) {
  87                 char discard;
  88                 int cnt = read_in_full(f->check_fd, &discard, 1);
  89                 if (cnt < 0)
  90                         die_errno("%s: error when reading the tail of sha1 file",
  91                                   f->name);
  92                 if (cnt)
  93                         die("%s: sha1 file has trailing garbage", f->name);
  94                 if (close(f->check_fd))
  95                         die_errno("%s: sha1 file error on close", f->name);
  96         }
  97         free_hashfile(f);
  98         return fd;
  99 }
 100
 101 void hashwrite(struct hashfile *f, const void *buf, unsigned int count)
 102 {
 103         while (count) {
 104                 unsigned left = f->buffer_len - f->offset;
 105                 unsigned nr = count > left ? left : count;
 106
 107                 if (f->do_crc)
 108                         f->crc32 = crc32(f->crc32, buf, nr);
 109
 110                 if (nr == f->buffer_len) {
 111                         /*
 112                          * Flush a full batch worth of data directly
 113                          * from the input, skipping the memcpy() to
 114                          * the hashfile's buffer. In this block,
 115                          * f->offset is necessarily zero.
 116                          */
 117                         if (!f->skip_hash)
 118                                 the_hash_algo->update_fn(&f->ctx, buf, nr);
 119                         flush(f, buf, nr);
 120                 } else {
 121                         /*
 122                          * Copy to the hashfile's buffer, flushing only
 123                          * if it became full.
 124                          */
 125                         memcpy(f->buffer + f->offset, buf, nr);
 126                         f->offset += nr;
 127                         left -= nr;
 128                         if (!left)
 129                                 hashflush(f);
 130                 }
 131
 132                 count -= nr;
 133                 buf = (char *) buf + nr;
 134         }
 135 }
 136
 137 struct hashfile *hashfd_check(const char *name)
 138 {
 139         int sink, check;
 140         struct hashfile *f;
 141
 142         sink = xopen("/dev/null", O_WRONLY);
 143         check = xopen(name, O_RDONLY);
 144         f = hashfd(sink, name);
 145         f->check_fd = check;
 146         f->check_buffer = xmalloc(f->buffer_len);
 147
 148         return f;
 149 }
 150
 151 static struct hashfile *hashfd_internal(int fd, const char *name,
 152                                         struct progress *tp,
 153                                         size_t buffer_len)
 154 {
 155         struct hashfile *f = xmalloc(sizeof(*f));
 156         f->fd = fd;
 157         f->check_fd = -1;
 158         f->offset = 0;
 159         f->total = 0;
 160         f->tp = tp;
 161         f->name = name;
 162         f->do_crc = 0;
 163         f->skip_hash = 0;
 164         the_hash_algo->init_fn(&f->ctx);
 165
 166         f->buffer_len = buffer_len;
 167         f->buffer = xmalloc(buffer_len);
 168         f->check_buffer = NULL;
 169
 170         return f;
 171 }
 172
 173 struct hashfile *hashfd(int fd, const char *name)
 174 {
 175         /*
 176          * Since we are not going to use a progress meter to
 177          * measure the rate of data passing through this hashfile,
 178          * use a larger buffer size to reduce fsync() calls.
 179          */
 180         return hashfd_internal(fd, name, NULL, 128 * 1024);
 181 }
 182
 183 struct hashfile *hashfd_throughput(int fd, const char *name, struct progress *tp)
 184 {
 185         /*
 186          * Since we are expecting to report progress of the
 187          * write into this hashfile, use a smaller buffer
 188          * size so the progress indicators arrive at a more
 189          * frequent rate.
 190          */
 191         return hashfd_internal(fd, name, tp, 8 * 1024);
 192 }
 193
 194 void hashfile_checkpoint(struct hashfile *f, struct hashfile_checkpoint *checkpoint)
 195 {
 196         hashflush(f);
 197         checkpoint->offset = f->total;
 198         the_hash_algo->clone_fn(&checkpoint->ctx, &f->ctx);
 199 }
 200
 201 int hashfile_truncate(struct hashfile *f, struct hashfile_checkpoint *checkpoint)
 202 {
 203         off_t offset = checkpoint->offset;
 204
 205         if (ftruncate(f->fd, offset) ||
 206             lseek(f->fd, offset, SEEK_SET) != offset)
 207                 return -1;
 208         f->total = offset;
 209         f->ctx = checkpoint->ctx;
 210         f->offset = 0; /* hashflush() was called in checkpoint */
 211         return 0;
 212 }
 213
 214 void crc32_begin(struct hashfile *f)
 215 {
 216         f->crc32 = crc32(0, NULL, 0);
 217         f->do_crc = 1;
 218 }
 219
 220 uint32_t crc32_end(struct hashfile *f)
 221 {
 222         f->do_crc = 0;
 223         return f->crc32;
 224 }
 225
 226 int hashfile_checksum_valid(const unsigned char *data, size_t total_len)
 227 {
 228         unsigned char got[GIT_MAX_RAWSZ];
 229         git_hash_ctx ctx;
 230         size_t data_len = total_len - the_hash_algo->rawsz;
 231
 232         if (total_len < the_hash_algo->rawsz)
 233                 return 0; /* say "too short"? */
 234
 235         the_hash_algo->init_fn(&ctx);
 236         the_hash_algo->update_fn(&ctx, data, data_len);
 237         the_hash_algo->final_fn(got, &ctx);
 238
 239         return hasheq(got, data + data_len);
 240 }