libmultipath/uevent.c

   1 /*
   2  * uevent.c - trigger upon netlink uevents from the kernel
   3  *
   4  *      Only kernels from version 2.6.10* on provide the uevent netlink socket.
   5  *      Until the libc-kernel-headers are updated, you need to compile with:
   6  *
   7  *        gcc -I /lib/modules/`uname -r`/build/include -o uevent_listen uevent_listen.c
   8  *
   9  * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
  10  *
  11  *      This program is free software; you can redistribute it and/or modify it
  12  *      under the terms of the GNU General Public License as published by the
  13  *      Free Software Foundation version 2 of the License.
  14  *
  15  *      This program is distributed in the hope that it will be useful, but
  16  *      WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  *      General Public License for more details.
  19  *
  20  *      You should have received a copy of the GNU General Public License along
  21  *      with this program; if not, write to the Free Software Foundation, Inc.,
  22  *      675 Mass Ave, Cambridge, MA 02139, USA.
  23  *
  24  */
  25
  26 #include <unistd.h>
  27 #include <stdio.h>
  28 #include <errno.h>
  29 #include <stdlib.h>
  30 #include <stddef.h>
  31 #include <string.h>
  32 #include <fcntl.h>
  33 #include <time.h>
  34 #include <sys/socket.h>
  35 #include <sys/user.h>
  36 #include <sys/un.h>
  37 #include <linux/types.h>
  38 #include <linux/netlink.h>
  39 #include <pthread.h>
  40 #include <limits.h>
  41 #include <sys/mman.h>
  42 #include <errno.h>
  43
  44 #include "memory.h"
  45 #include "debug.h"
  46 #include "list.h"
  47 #include "uevent.h"
  48
  49 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  50
  51 pthread_t uevq_thr;
  52 LIST_HEAD(uevq);
  53 pthread_mutex_t uevq_lock, *uevq_lockp = &uevq_lock;
  54 pthread_cond_t  uev_cond,  *uev_condp  = &uev_cond;
  55 uev_trigger *my_uev_trigger;
  56 void * my_trigger_data;
  57 int servicing_uev;
  58
  59 int is_uevent_busy(void)
  60 {
  61         int empty;
  62
  63         pthread_mutex_lock(uevq_lockp);
  64         empty = list_empty(&uevq);
  65         pthread_mutex_unlock(uevq_lockp);
  66         return (!empty || servicing_uev);
  67 }
  68
  69 struct uevent * alloc_uevent (void)
  70 {
  71         struct uevent *uev = MALLOC(sizeof(struct uevent));
  72
  73         if (uev)
  74                 INIT_LIST_HEAD(&uev->node);
  75
  76         return uev;
  77 }
  78
  79 void
  80 setup_thread_attr(pthread_attr_t *attr, size_t stacksize, int detached)
  81 {
  82         if (pthread_attr_init(attr)) {
  83                 fprintf(stderr, "can't initialize thread attr: %s\n",
  84                         strerror(errno));
  85                 exit(1);
  86         }
  87         if (stacksize < PTHREAD_STACK_MIN)
  88                 stacksize = PTHREAD_STACK_MIN;
  89
  90         if (pthread_attr_setstacksize(attr, stacksize)) {
  91                 fprintf(stderr, "can't set thread stack size to %lu: %s\n",
  92                         (unsigned long)stacksize, strerror(errno));
  93                 exit(1);
  94         }
  95         if (detached && pthread_attr_setdetachstate(attr,
  96                                                     PTHREAD_CREATE_DETACHED)) {
  97                 fprintf(stderr, "can't set thread to detached: %s\n",
  98                         strerror(errno));
  99                 exit(1);
 100         }
 101 }
 102
 103 /*
 104  * Called with uevq_lockp held
 105  */
 106 void
 107 service_uevq(struct list_head *tmpq)
 108 {
 109         struct uevent *uev, *tmp;
 110
 111         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 112                 list_del_init(&uev->node);
 113
 114                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
 115                         condlog(0, "uevent trigger error");
 116
 117                 FREE(uev);
 118         }
 119 }
 120
 121 static void uevq_stop(void *arg)
 122 {
 123         condlog(3, "Stopping uev queue");
 124         pthread_mutex_lock(uevq_lockp);
 125         my_uev_trigger = NULL;
 126         pthread_cond_signal(uev_condp);
 127         pthread_mutex_unlock(uevq_lockp);
 128 }
 129
 130 /*
 131  * Service the uevent queue.
 132  */
 133 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 134                     void * trigger_data)
 135 {
 136         my_uev_trigger = uev_trigger;
 137         my_trigger_data = trigger_data;
 138
 139         mlockall(MCL_CURRENT | MCL_FUTURE);
 140
 141         while (1) {
 142                 LIST_HEAD(uevq_tmp);
 143
 144                 pthread_mutex_lock(uevq_lockp);
 145                 servicing_uev = 0;
 146                 /*
 147                  * Condition signals are unreliable,
 148                  * so make sure we only wait if we have to.
 149                  */
 150                 if (list_empty(&uevq)) {
 151                         pthread_cond_wait(uev_condp, uevq_lockp);
 152                 }
 153                 servicing_uev = 1;
 154                 list_splice_init(&uevq, &uevq_tmp);
 155                 pthread_mutex_unlock(uevq_lockp);
 156                 if (!my_uev_trigger)
 157                         break;
 158                 service_uevq(&uevq_tmp);
 159         }
 160         condlog(3, "Terminating uev service queue");
 161         return 0;
 162 }
 163
 164 int uevent_listen(void)
 165 {
 166         int sock;
 167         struct sockaddr_nl snl;
 168         struct sockaddr_un sun;
 169         socklen_t addrlen;
 170         int retval;
 171         int rcvbufsz = 128*1024;
 172         int rcvsz = 0;
 173         int rcvszsz = sizeof(rcvsz);
 174         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 175         const int feature_on = 1;
 176
 177         /*
 178          * Queue uevents for service by dedicated thread so that the uevent
 179          * listening thread does not block on multipathd locks (vecs->lock)
 180          * thereby not getting to empty the socket's receive buffer queue
 181          * often enough.
 182          */
 183         INIT_LIST_HEAD(&uevq);
 184
 185         pthread_mutex_init(uevq_lockp, NULL);
 186         pthread_cond_init(uev_condp, NULL);
 187
 188         pthread_cleanup_push(uevq_stop, NULL);
 189
 190         /*
 191          * First check whether we have a udev socket
 192          */
 193         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 194         sun.sun_family = AF_LOCAL;
 195         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 196         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 197
 198         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 199         if (sock >= 0) {
 200
 201                 condlog(3, "reading events from udev socket.");
 202
 203                 /* the bind takes care of ensuring only one copy running */
 204                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 205                 if (retval < 0) {
 206                         condlog(0, "bind failed, exit");
 207                         goto exit;
 208                 }
 209
 210                 /* enable receiving of the sender credentials */
 211                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 212                            &feature_on, sizeof(feature_on));
 213
 214         } else {
 215                 /* Fallback to read kernel netlink events */
 216                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 217                 snl.nl_family = AF_NETLINK;
 218                 snl.nl_pid = getpid();
 219                 snl.nl_groups = 0x01;
 220
 221                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 222                 if (sock == -1) {
 223                         condlog(0, "error getting socket, exit");
 224                         return 1;
 225                 }
 226
 227                 condlog(3, "reading events from kernel.");
 228
 229                 /*
 230                  * try to avoid dropping uevents, even so, this is not a guarantee,
 231                  * but it does help to change the netlink uevent socket's
 232                  * receive buffer threshold from the default value of 106,496 to
 233                  * the maximum value of 262,142.
 234                  */
 235                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 236                                     sizeof(rcvbufsz));
 237
 238                 if (retval < 0) {
 239                         condlog(0, "error setting receive buffer size for socket, exit");
 240                         exit(1);
 241                 }
 242                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 243                 if (retval < 0) {
 244                         condlog(0, "error setting receive buffer size for socket, exit");
 245                         exit(1);
 246                 }
 247                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 248
 249                 /* enable receiving of the sender credentials */
 250                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 251                            &feature_on, sizeof(feature_on));
 252
 253                 retval = bind(sock, (struct sockaddr *) &snl,
 254                               sizeof(struct sockaddr_nl));
 255                 if (retval < 0) {
 256                         condlog(0, "bind failed, exit");
 257                         goto exit;
 258                 }
 259         }
 260
 261         while (1) {
 262                 int i;
 263                 char *pos;
 264                 size_t bufpos;
 265                 ssize_t buflen;
 266                 struct uevent *uev;
 267                 char *buffer;
 268                 struct msghdr smsg;
 269                 struct iovec iov;
 270                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 271                 struct cmsghdr *cmsg;
 272                 struct ucred *cred;
 273                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 274
 275                 memset(buf, 0x00, sizeof(buf));
 276                 iov.iov_base = &buf;
 277                 iov.iov_len = sizeof(buf);
 278                 memset (&smsg, 0x00, sizeof(struct msghdr));
 279                 smsg.msg_iov = &iov;
 280                 smsg.msg_iovlen = 1;
 281                 smsg.msg_control = cred_msg;
 282                 smsg.msg_controllen = sizeof(cred_msg);
 283
 284                 buflen = recvmsg(sock, &smsg, 0);
 285                 if (buflen < 0) {
 286                         if (errno != EINTR)
 287                                 condlog(0, "error receiving message, errno %d", errno);
 288                         continue;
 289                 }
 290
 291                 cmsg = CMSG_FIRSTHDR(&smsg);
 292                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 293                         condlog(3, "no sender credentials received, message ignored");
 294                         continue;
 295                 }
 296
 297                 cred = (struct ucred *)CMSG_DATA(cmsg);
 298                 if (cred->uid != 0) {
 299                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 300                         continue;
 301                 }
 302
 303                 /* skip header */
 304                 bufpos = strlen(buf) + 1;
 305                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 306                         condlog(3, "invalid message length");
 307                         continue;
 308                 }
 309
 310                 /* check message header */
 311                 if (strstr(buf, "@/") == NULL) {
 312                         condlog(3, "unrecognized message header");
 313                         continue;
 314                 }
 315                 if ((size_t)buflen > sizeof(buf)-1) {
 316                         condlog(2, "buffer overflow for received uevent");
 317                         buflen = sizeof(buf)-1;
 318                 }
 319
 320                 uev = alloc_uevent();
 321
 322                 if (!uev) {
 323                         condlog(1, "lost uevent, oom");
 324                         continue;
 325                 }
 326
 327                 if ((size_t)buflen > sizeof(buf)-1)
 328                         buflen = sizeof(buf)-1;
 329
 330                 /*
 331                  * Copy the shared receive buffer contents to buffer private
 332                  * to this uevent so we can immediately reuse the shared buffer.
 333                  */
 334                 memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 335                 buffer = uev->buffer;
 336                 buffer[buflen] = '\0';
 337
 338                 /* save start of payload */
 339                 bufpos = strlen(buffer) + 1;
 340
 341                 /* action string */
 342                 uev->action = buffer;
 343                 pos = strchr(buffer, '@');
 344                 if (!pos) {
 345                         condlog(3, "bad action string '%s'", buffer);
 346                         continue;
 347                 }
 348                 pos[0] = '\0';
 349
 350                 /* sysfs path */
 351                 uev->devpath = &pos[1];
 352
 353                 /* hotplug events have the environment attached - reconstruct envp[] */
 354                 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 355                         int keylen;
 356                         char *key;
 357
 358                         key = &buffer[bufpos];
 359                         keylen = strlen(key);
 360                         uev->envp[i] = key;
 361                         bufpos += keylen + 1;
 362                 }
 363                 uev->envp[i] = NULL;
 364
 365                 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 366                 uev->kernel = strrchr(uev->devpath, '/');
 367                 if (uev->kernel)
 368                         uev->kernel++;
 369
 370                 /* print payload environment */
 371                 for (i = 0; uev->envp[i] != NULL; i++)
 372                         condlog(5, "%s", uev->envp[i]);
 373
 374                 /*
 375                  * Queue uevent and poke service pthread.
 376                  */
 377                 pthread_mutex_lock(uevq_lockp);
 378                 list_add_tail(&uev->node, &uevq);
 379                 pthread_cond_signal(uev_condp);
 380                 pthread_mutex_unlock(uevq_lockp);
 381         }
 382
 383 exit:
 384         close(sock);
 385
 386         pthread_cleanup_pop(1);
 387
 388         pthread_mutex_destroy(uevq_lockp);
 389         pthread_cond_destroy(uev_condp);
 390
 391         return 1;
 392 }
 393
 394 extern int
 395 uevent_get_major(struct uevent *uev)
 396 {
 397         char *p, *q;
 398         int i, major = -1;
 399
 400         for (i = 0; uev->envp[i] != NULL; i++) {
 401                 if (!strncmp(uev->envp[i], "MAJOR", 5) && strlen(uev->envp[i]) > 6) {
 402                         p = uev->envp[i] + 6;
 403                         major = strtoul(p, &q, 10);
 404                         if (p == q) {
 405                                 condlog(2, "invalid major '%s'", p);
 406                                 major = -1;
 407                         }
 408                         break;
 409                 }
 410         }
 411         return major;
 412 }
 413
 414 extern int
 415 uevent_get_minor(struct uevent *uev)
 416 {
 417         char *p, *q;
 418         int i, minor = -1;
 419
 420         for (i = 0; uev->envp[i] != NULL; i++) {
 421                 if (!strncmp(uev->envp[i], "MINOR", 5) && strlen(uev->envp[i]) > 6) {
 422                         p = uev->envp[i] + 6;
 423                         minor = strtoul(p, &q, 10);
 424                         if (p == q) {
 425                                 condlog(2, "invalid minor '%s'", p);
 426                                 minor = -1;
 427                         }
 428                         break;
 429                 }
 430         }
 431         return minor;
 432 }
 433
 434 extern int
 435 uevent_get_disk_ro(struct uevent *uev)
 436 {
 437         char *p, *q;
 438         int i, ro = -1;
 439
 440         for (i = 0; uev->envp[i] != NULL; i++) {
 441                 if (!strncmp(uev->envp[i], "DISK_RO", 6) && strlen(uev->envp[i]) > 7) {
 442                         p = uev->envp[i] + 8;
 443                         ro = strtoul(p, &q, 10);
 444                         if (p == q) {
 445                                 condlog(2, "invalid read_only setting '%s'", p);
 446                                 ro = -1;
 447                         }
 448                         break;
 449                 }
 450         }
 451         return ro;
 452 }
 453
 454 extern char *
 455 uevent_get_dm_name(struct uevent *uev)
 456 {
 457         char *p = NULL;
 458         int i;
 459
 460         for (i = 0; uev->envp[i] != NULL; i++) {
 461                 if (!strncmp(uev->envp[i], "DM_NAME", 6) &&
 462                     strlen(uev->envp[i]) > 7) {
 463                         p = MALLOC(strlen(uev->envp[i] + 8) + 1);
 464                         strcpy(p, uev->envp[i] + 8);
 465                         break;
 466                 }
 467         }
 468         return p;
 469 }