src/basic/io-util.c

   1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
   2
   3 #include <errno.h>
   4 #include <limits.h>
   5 #include <stdio.h>
   6 #include <unistd.h>
   7
   8 #include "io-util.h"
   9 #include "string-util.h"
  10 #include "time-util.h"
  11
  12 int flush_fd(int fd) {
  13         int count = 0;
  14
  15         /* Read from the specified file descriptor, until POLLIN is not set anymore, throwing away everything
  16          * read. Note that some file descriptors (notable IP sockets) will trigger POLLIN even when no data can be read
  17          * (due to IP packet checksum mismatches), hence this function is only safe to be non-blocking if the fd used
  18          * was set to non-blocking too. */
  19
  20         for (;;) {
  21                 char buf[LINE_MAX];
  22                 ssize_t l;
  23                 int r;
  24
  25                 r = fd_wait_for_event(fd, POLLIN, 0);
  26                 if (r < 0) {
  27                         if (r == -EINTR)
  28                                 continue;
  29
  30                         return r;
  31                 }
  32                 if (r == 0)
  33                         return count;
  34
  35                 l = read(fd, buf, sizeof(buf));
  36                 if (l < 0) {
  37                         if (errno == EINTR)
  38                                 continue;
  39
  40                         if (errno == EAGAIN)
  41                                 return count;
  42
  43                         return -errno;
  44                 } else if (l == 0)
  45                         return count;
  46
  47                 count += (int) l;
  48         }
  49 }
  50
  51 ssize_t loop_read(int fd, void *buf, size_t nbytes, bool do_poll) {
  52         uint8_t *p = ASSERT_PTR(buf);
  53         ssize_t n = 0;
  54
  55         assert(fd >= 0);
  56
  57         /* If called with nbytes == 0, let's call read() at least once, to validate the operation */
  58
  59         if (nbytes > (size_t) SSIZE_MAX)
  60                 return -EINVAL;
  61
  62         do {
  63                 ssize_t k;
  64
  65                 k = read(fd, p, nbytes);
  66                 if (k < 0) {
  67                         if (errno == EINTR)
  68                                 continue;
  69
  70                         if (errno == EAGAIN && do_poll) {
  71
  72                                 /* We knowingly ignore any return value here,
  73                                  * and expect that any error/EOF is reported
  74                                  * via read() */
  75
  76                                 (void) fd_wait_for_event(fd, POLLIN, USEC_INFINITY);
  77                                 continue;
  78                         }
  79
  80                         return n > 0 ? n : -errno;
  81                 }
  82
  83                 if (k == 0)
  84                         return n;
  85
  86                 assert((size_t) k <= nbytes);
  87
  88                 p += k;
  89                 nbytes -= k;
  90                 n += k;
  91         } while (nbytes > 0);
  92
  93         return n;
  94 }
  95
  96 int loop_read_exact(int fd, void *buf, size_t nbytes, bool do_poll) {
  97         ssize_t n;
  98
  99         n = loop_read(fd, buf, nbytes, do_poll);
 100         if (n < 0)
 101                 return (int) n;
 102         if ((size_t) n != nbytes)
 103                 return -EIO;
 104
 105         return 0;
 106 }
 107
 108 int loop_write(int fd, const void *buf, size_t nbytes, bool do_poll) {
 109         const uint8_t *p = ASSERT_PTR(buf);
 110
 111         assert(fd >= 0);
 112
 113         if (_unlikely_(nbytes > (size_t) SSIZE_MAX))
 114                 return -EINVAL;
 115
 116         do {
 117                 ssize_t k;
 118
 119                 k = write(fd, p, nbytes);
 120                 if (k < 0) {
 121                         if (errno == EINTR)
 122                                 continue;
 123
 124                         if (errno == EAGAIN && do_poll) {
 125                                 /* We knowingly ignore any return value here,
 126                                  * and expect that any error/EOF is reported
 127                                  * via write() */
 128
 129                                 (void) fd_wait_for_event(fd, POLLOUT, USEC_INFINITY);
 130                                 continue;
 131                         }
 132
 133                         return -errno;
 134                 }
 135
 136                 if (_unlikely_(nbytes > 0 && k == 0)) /* Can't really happen */
 137                         return -EIO;
 138
 139                 assert((size_t) k <= nbytes);
 140
 141                 p += k;
 142                 nbytes -= k;
 143         } while (nbytes > 0);
 144
 145         return 0;
 146 }
 147
 148 int pipe_eof(int fd) {
 149         int r;
 150
 151         r = fd_wait_for_event(fd, POLLIN, 0);
 152         if (r <= 0)
 153                 return r;
 154
 155         return !!(r & POLLHUP);
 156 }
 157
 158 int ppoll_usec(struct pollfd *fds, size_t nfds, usec_t timeout) {
 159         int r;
 160
 161         assert(fds || nfds == 0);
 162
 163         /* This is a wrapper around ppoll() that does primarily two things:
 164          *
 165          *  ✅ Takes a usec_t instead of a struct timespec
 166          *
 167          *  ✅ Guarantees that if an invalid fd is specified we return EBADF (i.e. converts POLLNVAL to
 168          *     EBADF). This is done because EBADF is a programming error usually, and hence should bubble up
 169          *     as error, and not be eaten up as non-error POLLNVAL event.
 170          *
 171          *  ⚠️ ⚠️ ⚠️ Note that this function does not add any special handling for EINTR. Don't forget
 172          *  poll()/ppoll() will return with EINTR on any received signal always, there is no automatic
 173          *  restarting via SA_RESTART available. Thus, typically you want to handle EINTR not as an error,
 174          *  but just as reason to restart things, under the assumption you use a more appropriate mechanism
 175          *  to handle signals, such as signalfd() or signal handlers. ⚠️ ⚠️ ⚠️
 176          */
 177
 178         if (nfds == 0)
 179                 return 0;
 180
 181         r = ppoll(fds, nfds, timeout == USEC_INFINITY ? NULL : TIMESPEC_STORE(timeout), NULL);
 182         if (r < 0)
 183                 return -errno;
 184         if (r == 0)
 185                 return 0;
 186
 187         for (size_t i = 0, n = r; i < nfds && n > 0; i++) {
 188                 if (fds[i].revents == 0)
 189                         continue;
 190                 if (fds[i].revents & POLLNVAL)
 191                         return -EBADF;
 192                 n--;
 193         }
 194
 195         return r;
 196 }
 197
 198 int fd_wait_for_event(int fd, int event, usec_t timeout) {
 199         struct pollfd pollfd = {
 200                 .fd = fd,
 201                 .events = event,
 202         };
 203         int r;
 204
 205         /* ⚠️ ⚠️ ⚠️ Keep in mind you almost certainly want to handle -EINTR gracefully in the caller, see
 206          * ppoll_usec() above! ⚠️ ⚠️ ⚠️ */
 207
 208         r = ppoll_usec(&pollfd, 1, timeout);
 209         if (r <= 0)
 210                 return r;
 211
 212         return pollfd.revents;
 213 }
 214
 215 static size_t nul_length(const uint8_t *p, size_t sz) {
 216         size_t n = 0;
 217
 218         while (sz > 0) {
 219                 if (*p != 0)
 220                         break;
 221
 222                 n++;
 223                 p++;
 224                 sz--;
 225         }
 226
 227         return n;
 228 }
 229
 230 ssize_t sparse_write(int fd, const void *p, size_t sz, size_t run_length) {
 231         const uint8_t *q, *w, *e;
 232         ssize_t l;
 233
 234         q = w = p;
 235         e = q + sz;
 236         while (q < e) {
 237                 size_t n;
 238
 239                 n = nul_length(q, e - q);
 240
 241                 /* If there are more than the specified run length of
 242                  * NUL bytes, or if this is the beginning or the end
 243                  * of the buffer, then seek instead of write */
 244                 if ((n > run_length) ||
 245                     (n > 0 && q == p) ||
 246                     (n > 0 && q + n >= e)) {
 247                         if (q > w) {
 248                                 l = write(fd, w, q - w);
 249                                 if (l < 0)
 250                                         return -errno;
 251                                 if (l != q -w)
 252                                         return -EIO;
 253                         }
 254
 255                         if (lseek(fd, n, SEEK_CUR) == (off_t) -1)
 256                                 return -errno;
 257
 258                         q += n;
 259                         w = q;
 260                 } else if (n > 0)
 261                         q += n;
 262                 else
 263                         q++;
 264         }
 265
 266         if (q > w) {
 267                 l = write(fd, w, q - w);
 268                 if (l < 0)
 269                         return -errno;
 270                 if (l != q - w)
 271                         return -EIO;
 272         }
 273
 274         return q - (const uint8_t*) p;
 275 }
 276
 277 char* set_iovec_string_field(struct iovec *iovec, size_t *n_iovec, const char *field, const char *value) {
 278         char *x;
 279
 280         x = strjoin(field, value);
 281         if (x)
 282                 iovec[(*n_iovec)++] = IOVEC_MAKE_STRING(x);
 283         return x;
 284 }
 285
 286 char* set_iovec_string_field_free(struct iovec *iovec, size_t *n_iovec, const char *field, char *value) {
 287         char *x;
 288
 289         x = set_iovec_string_field(iovec, n_iovec, field, value);
 290         free(value);
 291         return x;
 292 }
 293
 294 struct iovec_wrapper *iovw_new(void) {
 295         return malloc0(sizeof(struct iovec_wrapper));
 296 }
 297
 298 void iovw_free_contents(struct iovec_wrapper *iovw, bool free_vectors) {
 299         if (free_vectors)
 300                 for (size_t i = 0; i < iovw->count; i++)
 301                         free(iovw->iovec[i].iov_base);
 302
 303         iovw->iovec = mfree(iovw->iovec);
 304         iovw->count = 0;
 305 }
 306
 307 struct iovec_wrapper *iovw_free_free(struct iovec_wrapper *iovw) {
 308         iovw_free_contents(iovw, true);
 309
 310         return mfree(iovw);
 311 }
 312
 313 struct iovec_wrapper *iovw_free(struct iovec_wrapper *iovw) {
 314         iovw_free_contents(iovw, false);
 315
 316         return mfree(iovw);
 317 }
 318
 319 int iovw_put(struct iovec_wrapper *iovw, void *data, size_t len) {
 320         if (iovw->count >= IOV_MAX)
 321                 return -E2BIG;
 322
 323         if (!GREEDY_REALLOC(iovw->iovec, iovw->count + 1))
 324                 return -ENOMEM;
 325
 326         iovw->iovec[iovw->count++] = IOVEC_MAKE(data, len);
 327         return 0;
 328 }
 329
 330 int iovw_put_string_field(struct iovec_wrapper *iovw, const char *field, const char *value) {
 331         _cleanup_free_ char *x = NULL;
 332         int r;
 333
 334         x = strjoin(field, value);
 335         if (!x)
 336                 return -ENOMEM;
 337
 338         r = iovw_put(iovw, x, strlen(x));
 339         if (r >= 0)
 340                 TAKE_PTR(x);
 341
 342         return r;
 343 }
 344
 345 int iovw_put_string_field_free(struct iovec_wrapper *iovw, const char *field, char *value) {
 346         _cleanup_free_ _unused_ char *free_ptr = value;
 347
 348         return iovw_put_string_field(iovw, field, value);
 349 }
 350
 351 void iovw_rebase(struct iovec_wrapper *iovw, char *old, char *new) {
 352         for (size_t i = 0; i < iovw->count; i++)
 353                 iovw->iovec[i].iov_base = (char *)iovw->iovec[i].iov_base - old + new;
 354 }
 355
 356 size_t iovw_size(struct iovec_wrapper *iovw) {
 357         size_t n = 0;
 358
 359         for (size_t i = 0; i < iovw->count; i++)
 360                 n += iovw->iovec[i].iov_len;
 361
 362         return n;
 363 }
 364
 365 void iovec_array_free(struct iovec *iov, size_t n) {
 366         if (!iov)
 367                 return;
 368
 369         for (size_t i = 0; i < n; i++)
 370                 free(iov[i].iov_base);
 371
 372         free(iov);
 373 }