A few minor man page tweaks.
[rsync.git] / fileio.c
blob32dc62da9d6a833e9c69014425eec87f785fa27d
1 /*
2 * File IO utilities used in rsync.
4 * Copyright (C) 1998 Andrew Tridgell
5 * Copyright (C) 2002 Martin Pool
6 * Copyright (C) 2004-2020 Wayne Davison
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.
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.
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.
22 #include "rsync.h"
23 #include "inums.h"
25 #ifndef ENODATA
26 #define ENODATA EAGAIN
27 #endif
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)
36 extern int sparse_files;
38 OFF_T preallocated_len = 0;
40 static OFF_T sparse_seek = 0;
41 static OFF_T sparse_past_write = 0;
43 int sparse_end(int f, OFF_T size)
45 int ret;
47 sparse_past_write = 0;
49 if (!sparse_seek)
50 return 0;
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);
62 ret = ret <= 0 ? -1 : 0;
64 #endif
66 sparse_seek = 0;
68 return ret;
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. */
74 static int write_sparse(int f, int use_seek, OFF_T offset, const char *buf, int len)
76 int l1 = 0, l2 = 0;
77 int ret;
79 for (l1 = 0; l1 < len && buf[l1] == 0; l1++) {}
80 for (l2 = 0; l2 < len-l1 && buf[len-(l2+1)] == 0; l2++) {}
82 sparse_seek += l1;
84 if (l1 == len)
85 return len;
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;
96 sparse_seek = l2;
97 sparse_past_write = offset + len - l2;
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;
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;
113 if (ret != (int)(len - (l1+l2))) {
114 sparse_seek = 0;
115 return l1+ret;
118 return len;
121 static char *wf_writeBuf;
122 static size_t wf_writeBufSize;
123 static size_t wf_writeBufCnt;
125 int flush_write_file(int f)
127 int ret = 0;
128 char *bp = wf_writeBuf;
130 while (wf_writeBufCnt > 0) {
131 if ((ret = write(f, bp, wf_writeBufCnt)) < 0) {
132 if (errno == EINTR)
133 continue;
134 return ret;
136 wf_writeBufCnt -= ret;
137 bp += ret;
139 return ret;
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()). */
145 int write_file(int f, int use_seek, OFF_T offset, const char *buf, int len)
147 int ret = 0;
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 if (!wf_writeBuf)
161 out_of_memory("write_file");
163 r1 = (int)MIN((size_t)len, wf_writeBufSize - wf_writeBufCnt);
164 if (r1) {
165 memcpy(wf_writeBuf + wf_writeBufCnt, buf, r1);
166 wf_writeBufCnt += r1;
168 if (wf_writeBufCnt == wf_writeBufSize) {
169 if (flush_write_file(f) < 0)
170 return -1;
171 if (!r1 && len)
172 continue;
175 if (r1 <= 0) {
176 if (ret > 0)
177 return ret;
178 return r1;
180 len -= r1;
181 buf += r1;
182 ret += r1;
184 return ret;
187 /* An in-place update found identical data at an identical location. We either
188 * just seek past it, or (for an in-place sparse update), we give the data to
189 * the sparse processor with the use_seek flag set. */
190 int skip_matched(int fd, OFF_T offset, const char *buf, int len)
192 OFF_T pos;
194 if (sparse_files > 0) {
195 if (write_file(fd, 1, offset, buf, len) != len)
196 return -1;
197 return 0;
200 if (flush_write_file(fd) < 0)
201 return -1;
203 if ((pos = do_lseek(fd, len, SEEK_CUR)) != offset + len) {
204 rsyserr(FERROR_XFER, errno, "lseek returned %s, not %s",
205 big_num(pos), big_num(offset));
206 return -1;
209 return 0;
212 /* This provides functionality somewhat similar to mmap() but using read().
213 * It gives sliding window access to a file. mmap() is not used because of
214 * the possibility of another program (such as a mailer) truncating the
215 * file thus giving us a SIGBUS. */
216 struct map_struct *map_file(int fd, OFF_T len, int32 read_size, int32 blk_size)
218 struct map_struct *map;
220 if (!(map = new0(struct map_struct)))
221 out_of_memory("map_file");
223 if (blk_size && (read_size % blk_size))
224 read_size += blk_size - (read_size % blk_size);
226 map->fd = fd;
227 map->file_size = len;
228 map->def_window_size = ALIGNED_LENGTH(read_size);
230 return map;
234 /* slide the read window in the file */
235 char *map_ptr(struct map_struct *map, OFF_T offset, int32 len)
237 OFF_T window_start, read_start;
238 int32 window_size, read_size, read_offset, align_fudge;
240 if (len == 0)
241 return NULL;
242 if (len < 0) {
243 rprintf(FERROR, "invalid len passed to map_ptr: %ld\n",
244 (long)len);
245 exit_cleanup(RERR_FILEIO);
248 /* in most cases the region will already be available */
249 if (offset >= map->p_offset && offset+len <= map->p_offset+map->p_len)
250 return map->p + (offset - map->p_offset);
252 /* nope, we are going to have to do a read. Work out our desired window */
253 align_fudge = (int32)ALIGNED_OVERSHOOT(offset);
254 window_start = offset - align_fudge;
255 window_size = map->def_window_size;
256 if (window_start + window_size > map->file_size)
257 window_size = (int32)(map->file_size - window_start);
258 if (window_size < len + align_fudge)
259 window_size = ALIGNED_LENGTH(len + align_fudge);
261 /* make sure we have allocated enough memory for the window */
262 if (window_size > map->p_size) {
263 map->p = realloc_array(map->p, char, window_size);
264 if (!map->p)
265 out_of_memory("map_ptr");
266 map->p_size = window_size;
269 /* Now try to avoid re-reading any bytes by reusing any bytes from the previous buffer. */
270 if (window_start >= map->p_offset && window_start < map->p_offset + map->p_len
271 && window_start + window_size >= map->p_offset + map->p_len) {
272 read_start = map->p_offset + map->p_len;
273 read_offset = (int32)(read_start - window_start);
274 read_size = window_size - read_offset;
275 memmove(map->p, map->p + (map->p_len - read_offset), read_offset);
276 } else {
277 read_start = window_start;
278 read_size = window_size;
279 read_offset = 0;
282 if (read_size <= 0) {
283 rprintf(FERROR, "invalid read_size of %ld in map_ptr\n",
284 (long)read_size);
285 exit_cleanup(RERR_FILEIO);
288 if (map->p_fd_offset != read_start) {
289 OFF_T ret = do_lseek(map->fd, read_start, SEEK_SET);
290 if (ret != read_start) {
291 rsyserr(FERROR, errno, "lseek returned %s, not %s",
292 big_num(ret), big_num(read_start));
293 exit_cleanup(RERR_FILEIO);
295 map->p_fd_offset = read_start;
297 map->p_offset = window_start;
298 map->p_len = window_size;
300 while (read_size > 0) {
301 int32 nread = read(map->fd, map->p + read_offset, read_size);
302 if (nread <= 0) {
303 if (!map->status)
304 map->status = nread ? errno : ENODATA;
305 /* The best we can do is zero the buffer -- the file
306 * has changed mid transfer! */
307 memset(map->p + read_offset, 0, read_size);
308 break;
310 map->p_fd_offset += nread;
311 read_offset += nread;
312 read_size -= nread;
315 return map->p + align_fudge;
318 int unmap_file(struct map_struct *map)
320 int ret;
322 if (map->p) {
323 free(map->p);
324 map->p = NULL;
326 ret = map->status;
327 #if 0 /* I don't think we really need this. */
328 force_memzero(map, sizeof map[0]);
329 #endif
330 free(map);
332 return ret;