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