src/lxc/utils.h

   1 /*
   2  * lxc: linux Container library
   3  *
   4  * (C) Copyright IBM Corp. 2007, 2008
   5  *
   6  * Authors:
   7  * Daniel Lezcano <daniel.lezcano at free.fr>
   8  *
   9  * This library is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public
  11  * License as published by the Free Software Foundation; either
  12  * version 2.1 of the License, or (at your option) any later version.
  13  *
  14  * This library is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with this library; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  22  */
  23 #ifndef __LXC_UTILS_H
  24 #define __LXC_UTILS_H
  25
  26 /* Properly support loop devices on 32bit systems. */
  27 #define _FILE_OFFSET_BITS 64
  28
  29 #include "config.h"
  30
  31 #include <errno.h>
  32 #include <stdarg.h>
  33 #include <stdio.h>
  34 #include <stdbool.h>
  35 #include <unistd.h>
  36 #include <linux/loop.h>
  37 #include <linux/magic.h>
  38 #include <sys/syscall.h>
  39 #include <sys/types.h>
  40 #include <sys/vfs.h>
  41
  42 #ifdef HAVE_LINUX_MEMFD_H
  43 #include <linux/memfd.h>
  44 #endif
  45
  46 #include "initutils.h"
  47
  48 /* Define __S_ISTYPE if missing from the C library. */
  49 #ifndef __S_ISTYPE
  50 #define __S_ISTYPE(mode, mask) (((mode)&S_IFMT) == (mask))
  51 #endif
  52
  53 #if HAVE_LIBCAP
  54 #ifndef CAP_SETFCAP
  55 #define CAP_SETFCAP 31
  56 #endif
  57
  58 #ifndef CAP_MAC_OVERRIDE
  59 #define CAP_MAC_OVERRIDE 32
  60 #endif
  61
  62 #ifndef CAP_MAC_ADMIN
  63 #define CAP_MAC_ADMIN 33
  64 #endif
  65 #endif
  66
  67 #ifndef PR_CAPBSET_DROP
  68 #define PR_CAPBSET_DROP 24
  69 #endif
  70
  71 #ifndef LO_FLAGS_AUTOCLEAR
  72 #define LO_FLAGS_AUTOCLEAR 4
  73 #endif
  74
  75 #ifndef CAP_SETUID
  76 #define CAP_SETUID 7
  77 #endif
  78
  79 #ifndef CAP_SETGID
  80 #define CAP_SETGID 6
  81 #endif
  82
  83 /* needed for cgroup automount checks, regardless of whether we
  84  * have included linux/capability.h or not */
  85 #ifndef CAP_SYS_ADMIN
  86 #define CAP_SYS_ADMIN 21
  87 #endif
  88
  89 /* Useful macros */
  90 /* Maximum number for 64 bit integer is a string with 21 digits: 2^64 - 1 = 21 */
  91 #define LXC_NUMSTRLEN64 21
  92 #define LXC_LINELEN 4096
  93 #define LXC_IDMAPLEN 4096
  94
  95 /* returns 1 on success, 0 if there were any failures */
  96 extern int lxc_rmdir_onedev(const char *path, const char *exclude);
  97 extern int get_u16(unsigned short *val, const char *arg, int base);
  98 extern int mkdir_p(const char *dir, mode_t mode);
  99 extern char *get_rundir(void);
 100
 101 /* Define getline() if missing from the C library */
 102 #ifndef HAVE_GETLINE
 103 #ifdef HAVE_FGETLN
 104 #include <../include/getline.h>
 105 #endif
 106 #endif
 107
 108 /* Define setns() if missing from the C library */
 109 #ifndef HAVE_SETNS
 110 static inline int setns(int fd, int nstype)
 111 {
 112 #ifdef __NR_setns
 113         return syscall(__NR_setns, fd, nstype);
 114 #elif defined(__NR_set_ns)
 115         return syscall(__NR_set_ns, fd, nstype);
 116 #else
 117         errno = ENOSYS;
 118         return -1;
 119 #endif
 120 }
 121 #endif
 122
 123 /* Define sethostname() if missing from the C library */
 124 #ifndef HAVE_SETHOSTNAME
 125 static inline int sethostname(const char * name, size_t len)
 126 {
 127 #ifdef __NR_sethostname
 128 return syscall(__NR_sethostname, name, len);
 129 #else
 130 errno = ENOSYS;
 131 return -1;
 132 #endif
 133 }
 134 #endif
 135
 136 /* Define unshare() if missing from the C library */
 137 #ifndef HAVE_UNSHARE
 138 static inline int unshare(int flags)
 139 {
 140 #ifdef __NR_unshare
 141         return syscall(__NR_unshare, flags);
 142 #else
 143         errno = ENOSYS;
 144         return -1;
 145 #endif
 146 }
 147 #else
 148 extern int unshare(int);
 149 #endif
 150
 151 /* Define signalfd() if missing from the C library */
 152 #ifdef HAVE_SYS_SIGNALFD_H
 153 #  include <sys/signalfd.h>
 154 #else
 155 /* assume kernel headers are too old */
 156 #include <stdint.h>
 157 struct signalfd_siginfo
 158 {
 159         uint32_t ssi_signo;
 160         int32_t ssi_errno;
 161         int32_t ssi_code;
 162         uint32_t ssi_pid;
 163         uint32_t ssi_uid;
 164         int32_t ssi_fd;
 165         uint32_t ssi_tid;
 166         uint32_t ssi_band;
 167         uint32_t ssi_overrun;
 168         uint32_t ssi_trapno;
 169         int32_t ssi_status;
 170         int32_t ssi_int;
 171         uint64_t ssi_ptr;
 172         uint64_t ssi_utime;
 173         uint64_t ssi_stime;
 174         uint64_t ssi_addr;
 175         uint8_t __pad[48];
 176 };
 177
 178 #  ifndef __NR_signalfd4
 179 /* assume kernel headers are too old */
 180 #    if __i386__
 181 #      define __NR_signalfd4 327
 182 #    elif __x86_64__
 183 #      define __NR_signalfd4 289
 184 #    elif __powerpc__
 185 #      define __NR_signalfd4 313
 186 #    elif __s390x__
 187 #      define __NR_signalfd4 322
 188 #    elif __arm__
 189 #      define __NR_signalfd4 355
 190 #    elif __mips__ && _MIPS_SIM == _ABIO32
 191 #      define __NR_signalfd4 4324
 192 #    elif __mips__ && _MIPS_SIM == _ABI64
 193 #      define __NR_signalfd4 5283
 194 #    elif __mips__ && _MIPS_SIM == _ABIN32
 195 #      define __NR_signalfd4 6287
 196 #    endif
 197 #endif
 198
 199 #  ifndef __NR_signalfd
 200 /* assume kernel headers are too old */
 201 #    if __i386__
 202 #      define __NR_signalfd 321
 203 #    elif __x86_64__
 204 #      define __NR_signalfd 282
 205 #    elif __powerpc__
 206 #      define __NR_signalfd 305
 207 #    elif __s390x__
 208 #      define __NR_signalfd 316
 209 #    elif __arm__
 210 #      define __NR_signalfd 349
 211 #    elif __mips__ && _MIPS_SIM == _ABIO32
 212 #      define __NR_signalfd 4317
 213 #    elif __mips__ && _MIPS_SIM == _ABI64
 214 #      define __NR_signalfd 5276
 215 #    elif __mips__ && _MIPS_SIM == _ABIN32
 216 #      define __NR_signalfd 6280
 217 #    endif
 218 #endif
 219
 220 static inline int signalfd(int fd, const sigset_t *mask, int flags)
 221 {
 222         int retval;
 223
 224         retval = syscall (__NR_signalfd4, fd, mask, _NSIG / 8, flags);
 225         if (errno == ENOSYS && flags == 0)
 226                 retval = syscall (__NR_signalfd, fd, mask, _NSIG / 8);
 227         return retval;
 228 }
 229 #endif
 230
 231 /* loop devices */
 232 #ifndef LO_FLAGS_AUTOCLEAR
 233 #define LO_FLAGS_AUTOCLEAR 4
 234 #endif
 235
 236 #ifndef LOOP_CTL_GET_FREE
 237 #define LOOP_CTL_GET_FREE 0x4C82
 238 #endif
 239
 240 /* memfd_create() */
 241 #ifndef MFD_CLOEXEC
 242 #define MFD_CLOEXEC 0x0001U
 243 #endif
 244
 245 #ifndef MFD_ALLOW_SEALING
 246 #define MFD_ALLOW_SEALING 0x0002U
 247 #endif
 248
 249 #ifndef HAVE_MEMFD_CREATE
 250 static inline int memfd_create(const char *name, unsigned int flags) {
 251         #ifndef __NR_memfd_create
 252                 #if defined __i386__
 253                         #define __NR_memfd_create 356
 254                 #elif defined __x86_64__
 255                         #define __NR_memfd_create 319
 256                 #elif defined __arm__
 257                         #define __NR_memfd_create 385
 258                 #elif defined __aarch64__
 259                         #define __NR_memfd_create 279
 260                 #elif defined __s390__
 261                         #define __NR_memfd_create 350
 262                 #elif defined __powerpc__
 263                         #define __NR_memfd_create 360
 264                 #elif defined __sparc__
 265                         #define __NR_memfd_create 348
 266                 #elif defined __blackfin__
 267                         #define __NR_memfd_create 390
 268                 #elif defined __ia64__
 269                         #define __NR_memfd_create 1340
 270                 #elif defined _MIPS_SIM
 271                         #if _MIPS_SIM == _MIPS_SIM_ABI32
 272                                 #define __NR_memfd_create 4354
 273                         #endif
 274                         #if _MIPS_SIM == _MIPS_SIM_NABI32
 275                                 #define __NR_memfd_create 6318
 276                         #endif
 277                         #if _MIPS_SIM == _MIPS_SIM_ABI64
 278                                 #define __NR_memfd_create 5314
 279                         #endif
 280                 #endif
 281         #endif
 282         #ifdef __NR_memfd_create
 283         return syscall(__NR_memfd_create, name, flags);
 284         #else
 285         errno = ENOSYS;
 286         return -1;
 287         #endif
 288 }
 289 #else
 290 extern int memfd_create(const char *name, unsigned int flags);
 291 #endif
 292
 293 static inline int lxc_set_cloexec(int fd)
 294 {
 295         return fcntl(fd, F_SETFD, FD_CLOEXEC);
 296 }
 297
 298 /* Struct to carry child pid from lxc_popen() to lxc_pclose().
 299  * Not an opaque struct to allow direct access to the underlying FILE *
 300  * (i.e., struct lxc_popen_FILE *file; fgets(buf, sizeof(buf), file->f))
 301  * without additional wrappers.
 302  */
 303 struct lxc_popen_FILE {
 304         int pipe;
 305         FILE *f;
 306         pid_t child_pid;
 307 };
 308
 309 /* popen(command, "re") replacement that restores default signal mask
 310  * via sigprocmask(2) (unblocks all signals) after fork(2) but prior to calling exec(3).
 311  * In short, popen(command, "re") does pipe() + fork()                 + exec()
 312  * while lxc_popen(command)       does pipe() + fork() + sigprocmask() + exec().
 313  * Returns pointer to struct lxc_popen_FILE, that should be freed with lxc_pclose().
 314  * On error returns NULL.
 315  */
 316 extern struct lxc_popen_FILE *lxc_popen(const char *command);
 317
 318 /* pclose() replacement to be used on struct lxc_popen_FILE *,
 319  * returned by lxc_popen().
 320  * Waits for associated process to terminate, returns its exit status and
 321  * frees resources, pointed to by struct lxc_popen_FILE *.
 322  */
 323 extern int lxc_pclose(struct lxc_popen_FILE *fp);
 324
 325 /**
 326  * BUILD_BUG_ON - break compile if a condition is true.
 327  * @condition: the condition which the compiler should know is false.
 328  *
 329  * If you have some code which relies on certain constants being equal, or
 330  * other compile-time-evaluated condition, you should use BUILD_BUG_ON to
 331  * detect if someone changes it.
 332  *
 333  * The implementation uses gcc's reluctance to create a negative array, but
 334  * gcc (as of 4.4) only emits that error for obvious cases (eg. not arguments
 335  * to inline functions).  So as a fallback we use the optimizer; if it can't
 336  * prove the condition is false, it will cause a link error on the undefined
 337  * "__build_bug_on_failed".  This error message can be harder to track down
 338  * though, hence the two different methods.
 339  */
 340 #ifndef __OPTIMIZE__
 341 #define BUILD_BUG_ON(condition) ((void)sizeof(char[1 - 2*!!(condition)]))
 342 #else
 343 extern int __build_bug_on_failed;
 344 #define BUILD_BUG_ON(condition)                                 \
 345         do {                                                    \
 346                 ((void)sizeof(char[1 - 2*!!(condition)]));      \
 347                 if (condition) __build_bug_on_failed = 1;       \
 348         } while(0)
 349 #endif
 350
 351 /*
 352  * wait on a child we forked
 353  */
 354 extern int wait_for_pid(pid_t pid);
 355 extern int lxc_wait_for_pid_status(pid_t pid);
 356
 357 /* send and receive buffers completely */
 358 extern ssize_t lxc_write_nointr(int fd, const void* buf, size_t count);
 359 extern ssize_t lxc_read_nointr(int fd, void* buf, size_t count);
 360 extern ssize_t lxc_read_nointr_expect(int fd, void *buf, size_t count,
 361                                       const void *expected_buf);
 362 #if HAVE_LIBGNUTLS
 363 #define SHA_DIGEST_LENGTH 20
 364 extern int sha1sum_file(char *fnam, unsigned char *md_value);
 365 #endif
 366
 367 /* read and write whole files */
 368 extern int lxc_write_to_file(const char *filename, const void *buf,
 369                              size_t count, bool add_newline);
 370 extern int lxc_read_from_file(const char *filename, void* buf, size_t count);
 371
 372 /* convert variadic argument lists to arrays (for execl type argument lists) */
 373 extern char** lxc_va_arg_list_to_argv(va_list ap, size_t skip, int do_strdup);
 374 extern const char** lxc_va_arg_list_to_argv_const(va_list ap, size_t skip);
 375
 376 /* Some simple string functions; if they return pointers, they are allocated
 377  * buffers.
 378  */
 379 extern char *lxc_string_replace(const char *needle, const char *replacement,
 380                                 const char *haystack);
 381 extern bool lxc_string_in_array(const char *needle, const char **haystack);
 382 extern char *lxc_string_join(const char *sep, const char **parts,
 383                              bool use_as_prefix);
 384 /* Normalize and split path: Leading and trailing / are removed, multiple
 385  * / are compactified, .. and . are resolved (.. on the top level is considered
 386  * identical to .).
 387  * Examples:
 388  *     /            ->   { NULL }
 389  *     foo/../bar   ->   { bar, NULL }
 390  *     ../../       ->   { NULL }
 391  *     ./bar/baz/.. ->   { bar, NULL }
 392  *     foo//bar     ->   { foo, bar, NULL }
 393  */
 394 extern char **lxc_normalize_path(const char *path);
 395 /* remove multiple slashes from the path, e.g. ///foo//bar -> /foo/bar */
 396 extern char *lxc_deslashify(const char *path);
 397 extern char *lxc_append_paths(const char *first, const char *second);
 398 /* Note: the following two functions use strtok(), so they will never
 399  *       consider an empty element, even if two delimiters are next to
 400  *       each other.
 401  */
 402 extern bool lxc_string_in_list(const char *needle, const char *haystack,
 403                                char sep);
 404 extern char **lxc_string_split(const char *string, char sep);
 405 extern char **lxc_string_split_and_trim(const char *string, char sep);
 406 extern char **lxc_string_split_quoted(char *string);
 407 /* Append string to NULL-terminated string array. */
 408 extern int lxc_append_string(char ***list, char *entry);
 409
 410 /* some simple array manipulation utilities */
 411 typedef void (*lxc_free_fn)(void *);
 412 typedef void *(*lxc_dup_fn)(void *);
 413 extern int lxc_grow_array(void ***array, size_t *capacity, size_t new_size,
 414                           size_t capacity_increment);
 415 extern void lxc_free_array(void **array, lxc_free_fn element_free_fn);
 416 extern size_t lxc_array_len(void **array);
 417
 418 extern void **lxc_append_null_to_array(void **array, size_t count);
 419
 420 /* mmap() wrapper. lxc_strmmap() will take care to \0-terminate files so that
 421  * normal string-handling functions can be used on the buffer. */
 422 extern void *lxc_strmmap(void *addr, size_t length, int prot, int flags, int fd,
 423                          off_t offset);
 424 /* munmap() wrapper. Use it to free memory mmap()ed with lxc_strmmap(). */
 425 extern int lxc_strmunmap(void *addr, size_t length);
 426
 427 /* initialize rand with urandom */
 428 extern int randseed(bool);
 429
 430 inline static bool am_unpriv(void) {
 431         return geteuid() != 0;
 432 }
 433
 434 /*
 435  * parse /proc/self/uid_map to find what @orig maps to
 436  */
 437 extern uid_t get_ns_uid(uid_t orig);
 438
 439 extern bool dir_exists(const char *path);
 440
 441 #define FNV1A_64_INIT ((uint64_t)0xcbf29ce484222325ULL)
 442 extern uint64_t fnv_64a_buf(void *buf, size_t len, uint64_t hval);
 443
 444 extern int detect_shared_rootfs(void);
 445 extern bool detect_ramfs_rootfs(void);
 446 extern char *on_path(const char *cmd, const char *rootfs);
 447 extern bool file_exists(const char *f);
 448 extern bool cgns_supported(void);
 449 extern char *choose_init(const char *rootfs);
 450 extern int print_to_file(const char *file, const char *content);
 451 extern bool switch_to_ns(pid_t pid, const char *ns);
 452 extern int is_dir(const char *path);
 453 extern char *get_template_path(const char *t);
 454 extern int safe_mount(const char *src, const char *dest, const char *fstype,
 455                       unsigned long flags, const void *data,
 456                       const char *rootfs);
 457 extern int lxc_mount_proc_if_needed(const char *rootfs);
 458 extern int open_devnull(void);
 459 extern int set_stdfds(int fd);
 460 extern int null_stdfds(void);
 461 extern int lxc_count_file_lines(const char *fn);
 462 extern int lxc_preserve_ns(const int pid, const char *ns);
 463
 464 /* Check whether a signal is blocked by a process. */
 465 extern bool task_blocking_signal(pid_t pid, int signal);
 466
 467 /* Helper functions to parse numbers. */
 468 extern int lxc_safe_uint(const char *numstr, unsigned int *converted);
 469 extern int lxc_safe_int(const char *numstr, int *converted);
 470 extern int lxc_safe_long(const char *numstr, long int *converted);
 471 extern int lxc_safe_long_long(const char *numstr, long long int *converted);
 472 extern int lxc_safe_ulong(const char *numstr, unsigned long *converted);
 473 /* Handles B, kb, MB, GB. Detects overflows and reports -ERANGE. */
 474 extern int parse_byte_size_string(const char *s, int64_t *converted);
 475
 476 /* Switch to a new uid and gid. */
 477 extern int lxc_switch_uid_gid(uid_t uid, gid_t gid);
 478 extern int lxc_setgroups(int size, gid_t list[]);
 479
 480 /* Find an unused loop device and associate it with source. */
 481 extern int lxc_prepare_loop_dev(const char *source, char *loop_dev, int flags);
 482
 483 /* Clear all mounts on a given node.
 484  * >= 0 successfully cleared. The number returned is the number of umounts
 485  *      performed.
 486  * < 0  error umounting. Return -errno.
 487  */
 488 extern int lxc_unstack_mountpoint(const char *path, bool lazy);
 489
 490 /*
 491  * run_command runs a command and collect it's std{err,out} output in buf.
 492  *
 493  * @param[out] buf     The buffer where the commands std{err,out] output will be
 494  *                     read into. If no output was produced, buf will be memset
 495  *                     to 0.
 496  * @param[in] buf_size The size of buf. This function will reserve one byte for
 497  *                     \0-termination.
 498  * @param[in] child_fn The function to be run in the child process. This
 499  *                     function must exec.
 500  * @param[in] args     Arguments to be passed to child_fn.
 501  */
 502 extern int run_command(char *buf, size_t buf_size, int (*child_fn)(void *),
 503                        void *args);
 504
 505 /* Concatenate all passed-in strings into one path. Do not fail. If any piece
 506  * is not prefixed with '/', add a '/'.
 507  */
 508 extern char *must_make_path(const char *first, ...) __attribute__((sentinel));
 509
 510 /* return copy of string @entry;  do not fail. */
 511 extern char *must_copy_string(const char *entry);
 512
 513 /* Re-alllocate a pointer, do not fail */
 514 extern void *must_realloc(void *orig, size_t sz);
 515
 516 /* __typeof__ should be safe to use with all compilers. */
 517 typedef __typeof__(((struct statfs *)NULL)->f_type) fs_type_magic;
 518 extern bool has_fs_type(const char *path, fs_type_magic magic_val);
 519 extern bool is_fs_type(const struct statfs *fs, fs_type_magic magic_val);
 520 extern bool lxc_nic_exists(char *nic);
 521 extern int lxc_make_tmpfile(char *template, bool rm);
 522
 523 static inline uint64_t lxc_getpagesize(void)
 524 {
 525         int64_t pgsz;
 526
 527         pgsz = sysconf(_SC_PAGESIZE);
 528         if (pgsz <= 0)
 529                 pgsz = 1 << 12;
 530
 531         return pgsz;
 532 }
 533
 534 /* If n is not a power of 2 this function will return the next power of 2
 535  * greater than that number. Note that this function always returns the *next*
 536  * power of 2 *greater* that number not the *nearest*. For example, passing 1025
 537  * as argument this function will return 2048 although the closest power of 2
 538  * would be 1024.
 539  * If the caller passes in 0 they will receive 0 in return since this is invalid
 540  * input and 0 is not a power of 2.
 541  */
 542 extern uint64_t lxc_find_next_power2(uint64_t n);
 543
 544 #endif /* __LXC_UTILS_H */