fileio.c

   1 /*
   2  * File IO utilities used in rsync.
   3  *
   4  * Copyright (C) 1998 Andrew Tridgell
   5  * Copyright (C) 2002 Martin Pool
   6  * Copyright (C) 2004-2023 Wayne Davison
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License as published by
  10  * the Free Software Foundation; either version 3 of the License, or
  11  * (at your option) any later version.
  12  *
  13  * This program is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  * GNU General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with this program; if not, visit the http://fsf.org website.
  20  */
  21
  22 #include "rsync.h"
  23 #include "inums.h"
  24
  25 #ifndef ENODATA
  26 #define ENODATA EAGAIN
  27 #endif
  28
  29 /* We want all reads to be aligned on 1K boundaries. */
  30 #define ALIGN_BOUNDARY 1024
  31 /* How far past the boundary is an offset? */
  32 #define ALIGNED_OVERSHOOT(oft) ((oft) & (ALIGN_BOUNDARY-1))
  33 /* Round up a length to the next boundary */
  34 #define ALIGNED_LENGTH(len) ((((len) - 1) | (ALIGN_BOUNDARY-1)) + 1)
  35
  36 extern int sparse_files;
  37
  38 OFF_T preallocated_len = 0;
  39
  40 static OFF_T sparse_seek = 0;
  41 static OFF_T sparse_past_write = 0;
  42
  43 int sparse_end(int f, OFF_T size, int updating_basis_or_equiv)
  44 {
  45         int ret = 0;
  46
  47         if (updating_basis_or_equiv) {
  48                 if (sparse_seek && do_punch_hole(f, sparse_past_write, sparse_seek) < 0)
  49                         ret = -1;
  50 #ifdef HAVE_FTRUNCATE /* A compilation formality -- in-place requires ftruncate() */
  51                 else /* Just in case the original file was longer */
  52                         ret = do_ftruncate(f, size);
  53 #endif
  54         } else if (sparse_seek) {
  55 #ifdef HAVE_FTRUNCATE
  56                 ret = do_ftruncate(f, size);
  57 #else
  58                 if (do_lseek(f, sparse_seek-1, SEEK_CUR) != size-1)
  59                         ret = -1;
  60                 else {
  61                         do {
  62                                 ret = write(f, "", 1);
  63                         } while (ret < 0 && errno == EINTR);
  64
  65                         ret = ret <= 0 ? -1 : 0;
  66                 }
  67 #endif
  68         }
  69
  70         sparse_past_write = sparse_seek = 0;
  71
  72         return ret;
  73 }
  74
  75 /* Note that the offset is just the caller letting us know where
  76  * the current file position is in the file. The use_seek arg tells
  77  * us that we should seek over matching data instead of writing it. */
  78 static int write_sparse(int f, int use_seek, OFF_T offset, const char *buf, int len)
  79 {
  80         int l1 = 0, l2 = 0;
  81         int ret;
  82
  83         for (l1 = 0; l1 < len && buf[l1] == 0; l1++) {}
  84         for (l2 = 0; l2 < len-l1 && buf[len-(l2+1)] == 0; l2++) {}
  85
  86         sparse_seek += l1;
  87
  88         if (l1 == len)
  89                 return len;
  90
  91         if (sparse_seek) {
  92                 if (sparse_past_write >= preallocated_len) {
  93                         if (do_lseek(f, sparse_seek, SEEK_CUR) < 0)
  94                                 return -1;
  95                 } else if (do_punch_hole(f, sparse_past_write, sparse_seek) < 0) {
  96                         sparse_seek = 0;
  97                         return -1;
  98                 }
  99         }
 100         sparse_seek = l2;
 101         sparse_past_write = offset + len - l2;
 102
 103         if (use_seek) {
 104                 /* The in-place data already matches. */
 105                 if (do_lseek(f, len - (l1+l2), SEEK_CUR) < 0)
 106                         return -1;
 107                 return len;
 108         }
 109
 110         while ((ret = write(f, buf + l1, len - (l1+l2))) <= 0) {
 111                 if (ret < 0 && errno == EINTR)
 112                         continue;
 113                 sparse_seek = 0;
 114                 return ret;
 115         }
 116
 117         if (ret != (int)(len - (l1+l2))) {
 118                 sparse_seek = 0;
 119                 return l1+ret;
 120         }
 121
 122         return len;
 123 }
 124
 125 static char *wf_writeBuf;
 126 static size_t wf_writeBufSize;
 127 static size_t wf_writeBufCnt;
 128
 129 int flush_write_file(int f)
 130 {
 131         int ret = 0;
 132         char *bp = wf_writeBuf;
 133
 134         while (wf_writeBufCnt > 0) {
 135                 if ((ret = write(f, bp, wf_writeBufCnt)) < 0) {
 136                         if (errno == EINTR)
 137                                 continue;
 138                         return ret;
 139                 }
 140                 wf_writeBufCnt -= ret;
 141                 bp += ret;
 142         }
 143         return ret;
 144 }
 145
 146 /* write_file does not allow incomplete writes.  It loops internally
 147  * until len bytes are written or errno is set.  Note that use_seek and
 148  * offset are only used in sparse processing (see write_sparse()). */
 149 int write_file(int f, int use_seek, OFF_T offset, const char *buf, int len)
 150 {
 151         int ret = 0;
 152
 153         while (len > 0) {
 154                 int r1;
 155                 if (sparse_files > 0) {
 156                         int len1 = MIN(len, SPARSE_WRITE_SIZE);
 157                         r1 = write_sparse(f, use_seek, offset, buf, len1);
 158                         offset += r1;
 159                 } else {
 160                         if (!wf_writeBuf) {
 161                                 wf_writeBufSize = WRITE_SIZE * 8;
 162                                 wf_writeBufCnt  = 0;
 163                                 wf_writeBuf = new_array(char, wf_writeBufSize);
 164                         }
 165                         r1 = (int)MIN((size_t)len, wf_writeBufSize - wf_writeBufCnt);
 166                         if (r1) {
 167                                 memcpy(wf_writeBuf + wf_writeBufCnt, buf, r1);
 168                                 wf_writeBufCnt += r1;
 169                         }
 170                         if (wf_writeBufCnt == wf_writeBufSize) {
 171                                 if (flush_write_file(f) < 0)
 172                                         return -1;
 173                                 if (!r1 && len)
 174                                         continue;
 175                         }
 176                 }
 177                 if (r1 <= 0) {
 178                         if (ret > 0)
 179                                 return ret;
 180                         return r1;
 181                 }
 182                 len -= r1;
 183                 buf += r1;
 184                 ret += r1;
 185         }
 186         return ret;
 187 }
 188
 189 /* An in-place update found identical data at an identical location. We either
 190  * just seek past it, or (for an in-place sparse update), we give the data to
 191  * the sparse processor with the use_seek flag set. */
 192 int skip_matched(int fd, OFF_T offset, const char *buf, int len)
 193 {
 194         OFF_T pos;
 195
 196         if (sparse_files > 0) {
 197                 if (write_file(fd, 1, offset, buf, len) != len)
 198                         return -1;
 199                 return 0;
 200         }
 201
 202         if (flush_write_file(fd) < 0)
 203                 return -1;
 204
 205         if ((pos = do_lseek(fd, len, SEEK_CUR)) != offset + len) {
 206                 rsyserr(FERROR_XFER, errno, "lseek returned %s, not %s",
 207                         big_num(pos), big_num(offset));
 208                 return -1;
 209         }
 210
 211         return 0;
 212 }
 213
 214 /* This provides functionality somewhat similar to mmap() but using read().
 215  * It gives sliding window access to a file.  mmap() is not used because of
 216  * the possibility of another program (such as a mailer) truncating the
 217  * file thus giving us a SIGBUS. */
 218 struct map_struct *map_file(int fd, OFF_T len, int32 read_size, int32 blk_size)
 219 {
 220         struct map_struct *map;
 221
 222         map = new0(struct map_struct);
 223
 224         if (blk_size && (read_size % blk_size))
 225                 read_size += blk_size - (read_size % blk_size);
 226
 227         map->fd = fd;
 228         map->file_size = len;
 229         map->def_window_size = ALIGNED_LENGTH(read_size);
 230
 231         return map;
 232 }
 233
 234
 235 /* slide the read window in the file */
 236 char *map_ptr(struct map_struct *map, OFF_T offset, int32 len)
 237 {
 238         OFF_T window_start, read_start;
 239         int32 window_size, read_size, read_offset, align_fudge;
 240
 241         if (len == 0)
 242                 return NULL;
 243         if (len < 0) {
 244                 rprintf(FERROR, "invalid len passed to map_ptr: %ld\n",
 245                         (long)len);
 246                 exit_cleanup(RERR_FILEIO);
 247         }
 248
 249         /* in most cases the region will already be available */
 250         if (offset >= map->p_offset && offset+len <= map->p_offset+map->p_len)
 251                 return map->p + (offset - map->p_offset);
 252
 253         /* nope, we are going to have to do a read. Work out our desired window */
 254         align_fudge = (int32)ALIGNED_OVERSHOOT(offset);
 255         window_start = offset - align_fudge;
 256         window_size = map->def_window_size;
 257         if (window_start + window_size > map->file_size)
 258                 window_size = (int32)(map->file_size - window_start);
 259         if (window_size < len + align_fudge)
 260                 window_size = ALIGNED_LENGTH(len + align_fudge);
 261
 262         /* make sure we have allocated enough memory for the window */
 263         if (window_size > map->p_size) {
 264                 map->p = realloc_array(map->p, char, window_size);
 265                 map->p_size = window_size;
 266         }
 267
 268         /* Now try to avoid re-reading any bytes by reusing any bytes from the previous buffer. */
 269         if (window_start >= map->p_offset && window_start < map->p_offset + map->p_len
 270          && window_start + window_size >= map->p_offset + map->p_len) {
 271                 read_start = map->p_offset + map->p_len;
 272                 read_offset = (int32)(read_start - window_start);
 273                 read_size = window_size - read_offset;
 274                 memmove(map->p, map->p + (map->p_len - read_offset), read_offset);
 275         } else {
 276                 read_start = window_start;
 277                 read_size = window_size;
 278                 read_offset = 0;
 279         }
 280
 281         if (read_size <= 0) {
 282                 rprintf(FERROR, "invalid read_size of %ld in map_ptr\n",
 283                         (long)read_size);
 284                 exit_cleanup(RERR_FILEIO);
 285         }
 286
 287         if (map->p_fd_offset != read_start) {
 288                 OFF_T ret = do_lseek(map->fd, read_start, SEEK_SET);
 289                 if (ret != read_start) {
 290                         rsyserr(FERROR, errno, "lseek returned %s, not %s",
 291                                 big_num(ret), big_num(read_start));
 292                         exit_cleanup(RERR_FILEIO);
 293                 }
 294                 map->p_fd_offset = read_start;
 295         }
 296         map->p_offset = window_start;
 297         map->p_len = window_size;
 298
 299         while (read_size > 0) {
 300                 int32 nread = read(map->fd, map->p + read_offset, read_size);
 301                 if (nread <= 0) {
 302                         if (!map->status)
 303                                 map->status = nread ? errno : ENODATA;
 304                         /* The best we can do is zero the buffer -- the file
 305                          * has changed mid transfer! */
 306                         memset(map->p + read_offset, 0, read_size);
 307                         break;
 308                 }
 309                 map->p_fd_offset += nread;
 310                 read_offset += nread;
 311                 read_size -= nread;
 312         }
 313
 314         return map->p + align_fudge;
 315 }
 316
 317 int unmap_file(struct map_struct *map)
 318 {
 319         int     ret;
 320
 321         if (map->p) {
 322                 free(map->p);
 323                 map->p = NULL;
 324         }
 325         ret = map->status;
 326 #if 0 /* I don't think we really need this. */
 327         force_memzero(map, sizeof map[0]);
 328 #endif
 329         free(map);
 330
 331         return ret;
 332 }