1 /*
2 * File IO utilities used in rsync.
3 *
4 * Copyright (C) 1998 Andrew Tridgell
5 * Copyright (C) 2002 Martin Pool
6 * Copyright (C) 2004-2020 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
sparse_end(int f,OFF_T size)43 int sparse_end(int f, OFF_T size)
44 {
45 int ret;
46
47 sparse_past_write = 0;
48
49 if (!sparse_seek)
50 return 0;
51
52 #ifdef HAVE_FTRUNCATE
53 ret = do_ftruncate(f, size);
54 #else
55 if (do_lseek(f, sparse_seek-1, SEEK_CUR) != size-1)
56 ret = -1;
57 else {
58 do {
59 ret = write(f, "", 1);
60 } while (ret < 0 && errno == EINTR);
61
62 ret = ret <= 0 ? -1 : 0;
63 }
64 #endif
65
66 sparse_seek = 0;
67
68 return ret;
69 }
70
71 /* Note that the offset is just the caller letting us know where
72 * the current file position is in the file. The use_seek arg tells
73 * us that we should seek over matching data instead of writing it. */
write_sparse(int f,int use_seek,OFF_T offset,const char * buf,int len)74 static int write_sparse(int f, int use_seek, OFF_T offset, const char *buf, int len)
75 {
76 int l1 = 0, l2 = 0;
77 int ret;
78
79 for (l1 = 0; l1 < len && buf[l1] == 0; l1++) {}
80 for (l2 = 0; l2 < len-l1 && buf[len-(l2+1)] == 0; l2++) {}
81
82 sparse_seek += l1;
83
84 if (l1 == len)
85 return len;
86
87 if (sparse_seek) {
88 if (sparse_past_write >= preallocated_len) {
89 if (do_lseek(f, sparse_seek, SEEK_CUR) < 0)
90 return -1;
91 } else if (do_punch_hole(f, sparse_past_write, sparse_seek) < 0) {
92 sparse_seek = 0;
93 return -1;
94 }
95 }
96 sparse_seek = l2;
97 sparse_past_write = offset + len - l2;
98
99 if (use_seek) {
100 /* The in-place data already matches. */
101 if (do_lseek(f, len - (l1+l2), SEEK_CUR) < 0)
102 return -1;
103 return len;
104 }
105
106 while ((ret = write(f, buf + l1, len - (l1+l2))) <= 0) {
107 if (ret < 0 && errno == EINTR)
108 continue;
109 sparse_seek = 0;
110 return ret;
111 }
112
113 if (ret != (int)(len - (l1+l2))) {
114 sparse_seek = 0;
115 return l1+ret;
116 }
117
118 return len;
119 }
120
121 static char *wf_writeBuf;
122 static size_t wf_writeBufSize;
123 static size_t wf_writeBufCnt;
124
flush_write_file(int f)125 int flush_write_file(int f)
126 {
127 int ret = 0;
128 char *bp = wf_writeBuf;
129
130 while (wf_writeBufCnt > 0) {
131 if ((ret = write(f, bp, wf_writeBufCnt)) < 0) {
132 if (errno == EINTR)
133 continue;
134 return ret;
135 }
136 wf_writeBufCnt -= ret;
137 bp += ret;
138 }
139 return ret;
140 }
141
142 /* write_file does not allow incomplete writes. It loops internally
143 * until len bytes are written or errno is set. Note that use_seek and
144 * offset are only used in sparse processing (see write_sparse()). */
write_file(int f,int use_seek,OFF_T offset,const char * buf,int len)145 int write_file(int f, int use_seek, OFF_T offset, const char *buf, int len)
146 {
147 int ret = 0;
148
149 while (len > 0) {
150 int r1;
151 if (sparse_files > 0) {
152 int len1 = MIN(len, SPARSE_WRITE_SIZE);
153 r1 = write_sparse(f, use_seek, offset, buf, len1);
154 offset += r1;
155 } else {
156 if (!wf_writeBuf) {
157 wf_writeBufSize = WRITE_SIZE * 8;
158 wf_writeBufCnt = 0;
159 wf_writeBuf = new_array(char, wf_writeBufSize);
160 }
161 r1 = (int)MIN((size_t)len, wf_writeBufSize - wf_writeBufCnt);
162 if (r1) {
163 memcpy(wf_writeBuf + wf_writeBufCnt, buf, r1);
164 wf_writeBufCnt += r1;
165 }
166 if (wf_writeBufCnt == wf_writeBufSize) {
167 if (flush_write_file(f) < 0)
168 return -1;
169 if (!r1 && len)
170 continue;
171 }
172 }
173 if (r1 <= 0) {
174 if (ret > 0)
175 return ret;
176 return r1;
177 }
178 len -= r1;
179 buf += r1;
180 ret += r1;
181 }
182 return ret;
183 }
184
185 /* An in-place update found identical data at an identical location. We either
186 * just seek past it, or (for an in-place sparse update), we give the data to
187 * the sparse processor with the use_seek flag set. */
skip_matched(int fd,OFF_T offset,const char * buf,int len)188 int skip_matched(int fd, OFF_T offset, const char *buf, int len)
189 {
190 OFF_T pos;
191
192 if (sparse_files > 0) {
193 if (write_file(fd, 1, offset, buf, len) != len)
194 return -1;
195 return 0;
196 }
197
198 if (flush_write_file(fd) < 0)
199 return -1;
200
201 if ((pos = do_lseek(fd, len, SEEK_CUR)) != offset + len) {
202 rsyserr(FERROR_XFER, errno, "lseek returned %s, not %s",
203 big_num(pos), big_num(offset));
204 return -1;
205 }
206
207 return 0;
208 }
209
210 /* This provides functionality somewhat similar to mmap() but using read().
211 * It gives sliding window access to a file. mmap() is not used because of
212 * the possibility of another program (such as a mailer) truncating the
213 * file thus giving us a SIGBUS. */
map_file(int fd,OFF_T len,int32 read_size,int32 blk_size)214 struct map_struct *map_file(int fd, OFF_T len, int32 read_size, int32 blk_size)
215 {
216 struct map_struct *map;
217
218 map = new0(struct map_struct);
219
220 if (blk_size && (read_size % blk_size))
221 read_size += blk_size - (read_size % blk_size);
222
223 map->fd = fd;
224 map->file_size = len;
225 map->def_window_size = ALIGNED_LENGTH(read_size);
226
227 return map;
228 }
229
230
231 /* slide the read window in the file */
map_ptr(struct map_struct * map,OFF_T offset,int32 len)232 char *map_ptr(struct map_struct *map, OFF_T offset, int32 len)
233 {
234 OFF_T window_start, read_start;
235 int32 window_size, read_size, read_offset, align_fudge;
236
237 if (len == 0)
238 return NULL;
239 if (len < 0) {
240 rprintf(FERROR, "invalid len passed to map_ptr: %ld\n",
241 (long)len);
242 exit_cleanup(RERR_FILEIO);
243 }
244
245 /* in most cases the region will already be available */
246 if (offset >= map->p_offset && offset+len <= map->p_offset+map->p_len)
247 return map->p + (offset - map->p_offset);
248
249 /* nope, we are going to have to do a read. Work out our desired window */
250 align_fudge = (int32)ALIGNED_OVERSHOOT(offset);
251 window_start = offset - align_fudge;
252 window_size = map->def_window_size;
253 if (window_start + window_size > map->file_size)
254 window_size = (int32)(map->file_size - window_start);
255 if (window_size < len + align_fudge)
256 window_size = ALIGNED_LENGTH(len + align_fudge);
257
258 /* make sure we have allocated enough memory for the window */
259 if (window_size > map->p_size) {
260 map->p = realloc_array(map->p, char, window_size);
261 map->p_size = window_size;
262 }
263
264 /* Now try to avoid re-reading any bytes by reusing any bytes from the previous buffer. */
265 if (window_start >= map->p_offset && window_start < map->p_offset + map->p_len
266 && window_start + window_size >= map->p_offset + map->p_len) {
267 read_start = map->p_offset + map->p_len;
268 read_offset = (int32)(read_start - window_start);
269 read_size = window_size - read_offset;
270 memmove(map->p, map->p + (map->p_len - read_offset), read_offset);
271 } else {
272 read_start = window_start;
273 read_size = window_size;
274 read_offset = 0;
275 }
276
277 if (read_size <= 0) {
278 rprintf(FERROR, "invalid read_size of %ld in map_ptr\n",
279 (long)read_size);
280 exit_cleanup(RERR_FILEIO);
281 }
282
283 if (map->p_fd_offset != read_start) {
284 OFF_T ret = do_lseek(map->fd, read_start, SEEK_SET);
285 if (ret != read_start) {
286 rsyserr(FERROR, errno, "lseek returned %s, not %s",
287 big_num(ret), big_num(read_start));
288 exit_cleanup(RERR_FILEIO);
289 }
290 map->p_fd_offset = read_start;
291 }
292 map->p_offset = window_start;
293 map->p_len = window_size;
294
295 while (read_size > 0) {
296 int32 nread = read(map->fd, map->p + read_offset, read_size);
297 if (nread <= 0) {
298 if (!map->status)
299 map->status = nread ? errno : ENODATA;
300 /* The best we can do is zero the buffer -- the file
301 * has changed mid transfer! */
302 memset(map->p + read_offset, 0, read_size);
303 break;
304 }
305 map->p_fd_offset += nread;
306 read_offset += nread;
307 read_size -= nread;
308 }
309
310 return map->p + align_fudge;
311 }
312
unmap_file(struct map_struct * map)313 int unmap_file(struct map_struct *map)
314 {
315 int ret;
316
317 if (map->p) {
318 free(map->p);
319 map->p = NULL;
320 }
321 ret = map->status;
322 #if 0 /* I don't think we really need this. */
323 force_memzero(map, sizeof map[0]);
324 #endif
325 free(map);
326
327 return ret;
328 }
329