1 | /* |
---|
2 | * write and writev replacements to handle partial writes. |
---|
3 | * |
---|
4 | * Usage: |
---|
5 | * |
---|
6 | * ssize_t xwrite(int fildes, const void *buf, size_t nbyte); |
---|
7 | * ssize_t xpwrite(int fildes, const void *buf, size_t nbyte, |
---|
8 | * off_t offset); |
---|
9 | * ssize_t xwritev(int fildes, const struct iovec *iov, int iovcnt); |
---|
10 | * |
---|
11 | * xwrite, xpwrite, and xwritev behave exactly like their C library |
---|
12 | * counterparts except that, if write or writev succeeds but returns a number |
---|
13 | * of bytes written less than the total bytes, the write is repeated picking |
---|
14 | * up where it left off until the full amount of the data is written. The |
---|
15 | * write is also repeated if it failed with EINTR. The write will be aborted |
---|
16 | * after 10 successive writes with no forward progress. |
---|
17 | * |
---|
18 | * Both functions return the number of bytes written on success or -1 on an |
---|
19 | * error, and will leave errno set to whatever the underlying system call set |
---|
20 | * it to. Note that it is possible for a write to fail after some data was |
---|
21 | * written, on the subsequent additional write; in that case, these functions |
---|
22 | * will return -1 and the number of bytes actually written will be lost. |
---|
23 | * |
---|
24 | * Copyright 2008 Board of Trustees, Leland Stanford Jr. University |
---|
25 | * Copyright (c) 2004, 2005, 2006 |
---|
26 | * by Internet Systems Consortium, Inc. ("ISC") |
---|
27 | * Copyright (c) 1991, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, |
---|
28 | * 2002, 2003 by The Internet Software Consortium and Rich Salz |
---|
29 | * |
---|
30 | * See LICENSE for licensing terms. |
---|
31 | */ |
---|
32 | |
---|
33 | #include <config.h> |
---|
34 | #include <portable/system.h> |
---|
35 | #include <portable/uio.h> |
---|
36 | |
---|
37 | #include <errno.h> |
---|
38 | |
---|
39 | #include <util/util.h> |
---|
40 | |
---|
41 | /* |
---|
42 | * If we're running the test suite, call testing versions of the write |
---|
43 | * functions. #undef pwrite first because large file support may define a |
---|
44 | * macro pwrite (pointing to pwrite64) on some platforms (e.g. Solaris). |
---|
45 | */ |
---|
46 | #if TESTING |
---|
47 | # undef pwrite |
---|
48 | # define pwrite fake_pwrite |
---|
49 | # define write fake_write |
---|
50 | # define writev fake_writev |
---|
51 | ssize_t fake_pwrite(int, const void *, size_t, off_t); |
---|
52 | ssize_t fake_write(int, const void *, size_t); |
---|
53 | ssize_t fake_writev(int, const struct iovec *, int); |
---|
54 | #endif |
---|
55 | |
---|
56 | |
---|
57 | ssize_t |
---|
58 | xwrite(int fd, const void *buffer, size_t size) |
---|
59 | { |
---|
60 | size_t total; |
---|
61 | ssize_t status; |
---|
62 | int count = 0; |
---|
63 | |
---|
64 | if (size == 0) |
---|
65 | return 0; |
---|
66 | |
---|
67 | /* Abort the write if we try ten times with no forward progress. */ |
---|
68 | for (total = 0; total < size; total += status) { |
---|
69 | if (++count > 10) |
---|
70 | break; |
---|
71 | status = write(fd, (const char *) buffer + total, size - total); |
---|
72 | if (status > 0) |
---|
73 | count = 0; |
---|
74 | if (status < 0) { |
---|
75 | if (errno != EINTR) |
---|
76 | break; |
---|
77 | status = 0; |
---|
78 | } |
---|
79 | } |
---|
80 | return (total < size) ? -1 : (ssize_t) total; |
---|
81 | } |
---|
82 | |
---|
83 | |
---|
84 | #ifndef _WIN32 |
---|
85 | ssize_t |
---|
86 | xpwrite(int fd, const void *buffer, size_t size, off_t offset) |
---|
87 | { |
---|
88 | size_t total; |
---|
89 | ssize_t status; |
---|
90 | int count = 0; |
---|
91 | |
---|
92 | if (size == 0) |
---|
93 | return 0; |
---|
94 | |
---|
95 | /* Abort the write if we try ten times with no forward progress. */ |
---|
96 | for (total = 0; total < size; total += status) { |
---|
97 | if (++count > 10) |
---|
98 | break; |
---|
99 | status = pwrite(fd, (const char *) buffer + total, size - total, |
---|
100 | offset + total); |
---|
101 | if (status > 0) |
---|
102 | count = 0; |
---|
103 | if (status < 0) { |
---|
104 | if (errno != EINTR) |
---|
105 | break; |
---|
106 | status = 0; |
---|
107 | } |
---|
108 | } |
---|
109 | return (total < size) ? -1 : (ssize_t) total; |
---|
110 | } |
---|
111 | #endif |
---|
112 | |
---|
113 | |
---|
114 | ssize_t |
---|
115 | xwritev(int fd, const struct iovec iov[], int iovcnt) |
---|
116 | { |
---|
117 | ssize_t total, status = 0; |
---|
118 | size_t left, offset; |
---|
119 | int iovleft, i, count; |
---|
120 | struct iovec *tmpiov; |
---|
121 | |
---|
122 | if (iovcnt == 0) |
---|
123 | return 0; |
---|
124 | |
---|
125 | /* Get a count of the total number of bytes in the iov array. */ |
---|
126 | for (total = 0, i = 0; i < iovcnt; i++) |
---|
127 | total += iov[i].iov_len; |
---|
128 | |
---|
129 | if (total == 0) |
---|
130 | return 0; |
---|
131 | |
---|
132 | /* |
---|
133 | * First, try just writing it all out. Most of the time this will succeed |
---|
134 | * and save us lots of work. Abort the write if we try ten times with no |
---|
135 | * forward progress. |
---|
136 | */ |
---|
137 | count = 0; |
---|
138 | do { |
---|
139 | if (++count > 10) |
---|
140 | break; |
---|
141 | status = writev(fd, iov, iovcnt); |
---|
142 | if (status > 0) |
---|
143 | count = 0; |
---|
144 | } while (status < 0 && errno == EINTR); |
---|
145 | if (status < 0) |
---|
146 | return -1; |
---|
147 | if (status == total) |
---|
148 | return total; |
---|
149 | |
---|
150 | /* |
---|
151 | * If we fell through to here, the first write partially succeeded. |
---|
152 | * Figure out how far through the iov array we got, and then duplicate the |
---|
153 | * rest of it so that we can modify it to reflect how much we manage to |
---|
154 | * write on successive tries. |
---|
155 | */ |
---|
156 | offset = status; |
---|
157 | left = total - offset; |
---|
158 | for (i = 0; offset >= (size_t) iov[i].iov_len; i++) |
---|
159 | offset -= iov[i].iov_len; |
---|
160 | iovleft = iovcnt - i; |
---|
161 | tmpiov = malloc(iovleft * sizeof(struct iovec)); |
---|
162 | if (tmpiov == NULL) |
---|
163 | return -1; |
---|
164 | memcpy(tmpiov, iov + i, iovleft * sizeof(struct iovec)); |
---|
165 | |
---|
166 | /* |
---|
167 | * status now contains the offset into the first iovec struct in tmpiov. |
---|
168 | * Go into the write loop, trying to write out everything remaining at |
---|
169 | * each point. At the top of the loop, status will contain a count of |
---|
170 | * bytes written out at the beginning of the set of iovec structs. |
---|
171 | */ |
---|
172 | i = 0; |
---|
173 | do { |
---|
174 | if (++count > 10) |
---|
175 | break; |
---|
176 | |
---|
177 | /* Skip any leading data that has been written out. */ |
---|
178 | for (; offset >= (size_t) tmpiov[i].iov_len && iovleft > 0; i++) { |
---|
179 | offset -= tmpiov[i].iov_len; |
---|
180 | iovleft--; |
---|
181 | } |
---|
182 | tmpiov[i].iov_base = (char *) tmpiov[i].iov_base + offset; |
---|
183 | tmpiov[i].iov_len -= offset; |
---|
184 | |
---|
185 | /* Write out what's left and return success if it's all written. */ |
---|
186 | status = writev(fd, tmpiov + i, iovleft); |
---|
187 | if (status <= 0) |
---|
188 | offset = 0; |
---|
189 | else { |
---|
190 | offset = status; |
---|
191 | left -= offset; |
---|
192 | count = 0; |
---|
193 | } |
---|
194 | } while (left > 0 && (status >= 0 || errno == EINTR)); |
---|
195 | |
---|
196 | /* We're either done or got an error; if we're done, left is now 0. */ |
---|
197 | free(tmpiov); |
---|
198 | return (left == 0) ? total : -1; |
---|
199 | } |
---|