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