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, see <http://www.gnu.org/licenses/>.
  22  *
  23  */
  24
  25 #include <unistd.h>
  26 #include <stdio.h>
  27 #include <stdbool.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 <poll.h>
  38 #include <linux/types.h>
  39 #include <linux/netlink.h>
  40 #include <pthread.h>
  41 #include <sys/mman.h>
  42 #include <sys/time.h>
  43 #include <libudev.h>
  44 #include <errno.h>
  45
  46 #include "memory.h"
  47 #include "debug.h"
  48 #include "list.h"
  49 #include "uevent.h"
  50 #include "vector.h"
  51 #include "structs.h"
  52 #include "util.h"
  53 #include "config.h"
  54 #include "blacklist.h"
  55 #include "devmapper.h"
  56
  57 #define MAX_ACCUMULATION_COUNT 2048
  58 #define MAX_ACCUMULATION_TIME 30*1000
  59 #define MIN_BURST_SPEED 10
  60
  61 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  62
  63 LIST_HEAD(uevq);
  64 pthread_mutex_t uevq_lock = PTHREAD_MUTEX_INITIALIZER;
  65 pthread_mutex_t *uevq_lockp = &uevq_lock;
  66 pthread_cond_t uev_cond = PTHREAD_COND_INITIALIZER;
  67 pthread_cond_t *uev_condp = &uev_cond;
  68 uev_trigger *my_uev_trigger;
  69 void * my_trigger_data;
  70 int servicing_uev;
  71
  72 int is_uevent_busy(void)
  73 {
  74         int empty;
  75
  76         pthread_mutex_lock(uevq_lockp);
  77         empty = list_empty(&uevq);
  78         pthread_mutex_unlock(uevq_lockp);
  79         return (!empty || servicing_uev);
  80 }
  81
  82 struct uevent * alloc_uevent (void)
  83 {
  84         struct uevent *uev = MALLOC(sizeof(struct uevent));
  85
  86         if (uev) {
  87                 INIT_LIST_HEAD(&uev->node);
  88                 INIT_LIST_HEAD(&uev->merge_node);
  89         }
  90
  91         return uev;
  92 }
  93
  94 void
  95 uevq_cleanup(struct list_head *tmpq)
  96 {
  97         struct uevent *uev, *tmp;
  98
  99         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 100                 list_del_init(&uev->node);
 101
 102                 if (uev->udev)
 103                         udev_device_unref(uev->udev);
 104                 FREE(uev);
 105         }
 106 }
 107
 108 static const char* uevent_get_env_var(const struct uevent *uev,
 109                                       const char *attr)
 110 {
 111         int i, len;
 112         const char *p = NULL;
 113
 114         if (attr == NULL)
 115                 goto invalid;
 116
 117         len = strlen(attr);
 118         if (len == 0)
 119                 goto invalid;
 120
 121         for (i = 0; uev->envp[i] != NULL; i++) {
 122                 const char *var = uev->envp[i];
 123
 124                 if (strlen(var) > len &&
 125                     !memcmp(var, attr, len) && var[len] == '=') {
 126                         p = var + len + 1;
 127                         break;
 128                 }
 129         }
 130
 131         condlog(4, "%s: %s -> '%s'", __func__, attr, p);
 132         return p;
 133
 134 invalid:
 135         condlog(2, "%s: empty variable name", __func__);
 136         return NULL;
 137 }
 138
 139 static int uevent_get_env_positive_int(const struct uevent *uev,
 140                                        const char *attr)
 141 {
 142         const char *p = uevent_get_env_var(uev, attr);
 143         char *q;
 144         int ret;
 145
 146         if (p == NULL || *p == '\0')
 147                 return -1;
 148
 149         ret = strtoul(p, &q, 10);
 150         if (*q != '\0' || ret < 0) {
 151                 condlog(2, "%s: invalid %s: '%s'", __func__, attr, p);
 152                 return -1;
 153         }
 154         return ret;
 155 }
 156
 157 void
 158 uevent_get_wwid(struct uevent *uev)
 159 {
 160         char *uid_attribute;
 161         const char *val;
 162         struct config * conf;
 163
 164         conf = get_multipath_config();
 165         pthread_cleanup_push(put_multipath_config, conf);
 166         uid_attribute = parse_uid_attribute_by_attrs(conf->uid_attrs, uev->kernel);
 167         pthread_cleanup_pop(1);
 168
 169         val = uevent_get_env_var(uev, uid_attribute);
 170         if (val)
 171                 uev->wwid = val;
 172         FREE(uid_attribute);
 173 }
 174
 175 bool
 176 uevent_need_merge(void)
 177 {
 178         struct config * conf;
 179         bool need_merge = false;
 180
 181         conf = get_multipath_config();
 182         if (conf->uid_attrs)
 183                 need_merge = true;
 184         put_multipath_config(conf);
 185
 186         return need_merge;
 187 }
 188
 189 bool
 190 uevent_can_discard(struct uevent *uev)
 191 {
 192         int invalid = 0;
 193         struct config * conf;
 194
 195         /*
 196          * do not filter dm devices by devnode
 197          */
 198         if (!strncmp(uev->kernel, "dm-", 3))
 199                 return false;
 200         /*
 201          * filter paths devices by devnode
 202          */
 203         conf = get_multipath_config();
 204         pthread_cleanup_push(put_multipath_config, conf);
 205         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 206                            uev->kernel) > 0)
 207                 invalid = 1;
 208         pthread_cleanup_pop(1);
 209
 210         if (invalid)
 211                 return true;
 212         return false;
 213 }
 214
 215 bool
 216 uevent_can_filter(struct uevent *earlier, struct uevent *later)
 217 {
 218
 219         /*
 220          * filter earlier uvents if path has removed later. Eg:
 221          * "add path1 |chang path1 |add path2 |remove path1"
 222          * can filter as:
 223          * "add path2 |remove path1"
 224          * uevents "add path1" and "chang path1" are filtered out
 225          */
 226         if (!strcmp(earlier->kernel, later->kernel) &&
 227                 !strcmp(later->action, "remove") &&
 228                 strncmp(later->kernel, "dm-", 3)) {
 229                 return true;
 230         }
 231
 232         /*
 233          * filter change uvents if add uevents exist. Eg:
 234          * "change path1| add path1 |add path2"
 235          * can filter as:
 236          * "add path1 |add path2"
 237          * uevent "chang path1" is filtered out
 238          */
 239         if (!strcmp(earlier->kernel, later->kernel) &&
 240                 !strcmp(earlier->action, "change") &&
 241                 !strcmp(later->action, "add") &&
 242                 strncmp(later->kernel, "dm-", 3)) {
 243                 return true;
 244         }
 245
 246         return false;
 247 }
 248
 249 bool
 250 merge_need_stop(struct uevent *earlier, struct uevent *later)
 251 {
 252         /*
 253          * dm uevent do not try to merge with left uevents
 254          */
 255         if (!strncmp(later->kernel, "dm-", 3))
 256                 return true;
 257
 258         /*
 259          * we can not make a jugement without wwid,
 260          * so it is sensible to stop merging
 261          */
 262         if (!earlier->wwid || !later->wwid)
 263                 return true;
 264         /*
 265          * uevents merging stopped
 266          * when we meet an opposite action uevent from the same LUN to AVOID
 267          * "add path1 |remove path1 |add path2 |remove path2 |add path3"
 268          * to merge as "remove path1, path2" and "add path1, path2, path3"
 269          * OR
 270          * "remove path1 |add path1 |remove path2 |add path2 |remove path3"
 271          * to merge as "add path1, path2" and "remove path1, path2, path3"
 272          * SO
 273          * when we meet a non-change uevent from the same LUN
 274          * with the same wwid and different action
 275          * it would be better to stop merging.
 276          */
 277         if (!strcmp(earlier->wwid, later->wwid) &&
 278             strcmp(earlier->action, later->action) &&
 279             strcmp(earlier->action, "change") &&
 280             strcmp(later->action, "change"))
 281                 return true;
 282
 283         return false;
 284 }
 285
 286 bool
 287 uevent_can_merge(struct uevent *earlier, struct uevent *later)
 288 {
 289         /* merge paths uevents
 290          * whose wwids exsit and are same
 291          * and actions are same,
 292          * and actions are addition or deletion
 293          */
 294         if (earlier->wwid && later->wwid &&
 295             !strcmp(earlier->wwid, later->wwid) &&
 296             !strcmp(earlier->action, later->action) &&
 297             strncmp(earlier->action, "change", 6) &&
 298             strncmp(earlier->kernel, "dm-", 3)) {
 299                 return true;
 300         }
 301
 302         return false;
 303 }
 304
 305 void
 306 uevent_prepare(struct list_head *tmpq)
 307 {
 308         struct uevent *uev, *tmp;
 309
 310         list_for_each_entry_reverse_safe(uev, tmp, tmpq, node) {
 311                 if (uevent_can_discard(uev)) {
 312                         list_del_init(&uev->node);
 313                         if (uev->udev)
 314                                 udev_device_unref(uev->udev);
 315                         FREE(uev);
 316                         continue;
 317                 }
 318
 319                 if (strncmp(uev->kernel, "dm-", 3) &&
 320                     uevent_need_merge())
 321                         uevent_get_wwid(uev);
 322         }
 323 }
 324
 325 void
 326 uevent_filter(struct uevent *later, struct list_head *tmpq)
 327 {
 328         struct uevent *earlier, *tmp;
 329
 330         list_for_some_entry_reverse_safe(earlier, tmp, &later->node, tmpq, node) {
 331                 /*
 332                  * filter unnessary earlier uevents
 333                  * by the later uevent
 334                  */
 335                 if (uevent_can_filter(earlier, later)) {
 336                         condlog(3, "uevent: %s-%s has filtered by uevent: %s-%s",
 337                                 earlier->kernel, earlier->action,
 338                                 later->kernel, later->action);
 339
 340                         list_del_init(&earlier->node);
 341                         if (earlier->udev)
 342                                 udev_device_unref(earlier->udev);
 343                         FREE(earlier);
 344                 }
 345         }
 346 }
 347
 348 void
 349 uevent_merge(struct uevent *later, struct list_head *tmpq)
 350 {
 351         struct uevent *earlier, *tmp;
 352
 353         list_for_some_entry_reverse_safe(earlier, tmp, &later->node, tmpq, node) {
 354                 if (merge_need_stop(earlier, later))
 355                         break;
 356                 /*
 357                  * merge earlier uevents to the later uevent
 358                  */
 359                 if (uevent_can_merge(earlier, later)) {
 360                         condlog(3, "merged uevent: %s-%s-%s with uevent: %s-%s-%s",
 361                                 earlier->action, earlier->kernel, earlier->wwid,
 362                                 later->action, later->kernel, later->wwid);
 363
 364                         list_move(&earlier->node, &later->merge_node);
 365                 }
 366         }
 367 }
 368
 369 void
 370 merge_uevq(struct list_head *tmpq)
 371 {
 372         struct uevent *later;
 373
 374         uevent_prepare(tmpq);
 375         list_for_each_entry_reverse(later, tmpq, node) {
 376                 uevent_filter(later, tmpq);
 377                 if(uevent_need_merge())
 378                         uevent_merge(later, tmpq);
 379         }
 380 }
 381
 382 void
 383 service_uevq(struct list_head *tmpq)
 384 {
 385         struct uevent *uev, *tmp;
 386
 387         list_for_each_entry_safe(uev, tmp, tmpq, node) {
 388                 list_del_init(&uev->node);
 389
 390                 if (my_uev_trigger && my_uev_trigger(uev, my_trigger_data))
 391                         condlog(0, "uevent trigger error");
 392
 393                 uevq_cleanup(&uev->merge_node);
 394
 395                 if (uev->udev)
 396                         udev_device_unref(uev->udev);
 397                 FREE(uev);
 398         }
 399 }
 400
 401 static void uevent_cleanup(void *arg)
 402 {
 403         struct udev *udev = arg;
 404
 405         condlog(3, "Releasing uevent_listen() resources");
 406         udev_unref(udev);
 407 }
 408
 409 static void monitor_cleanup(void *arg)
 410 {
 411         struct udev_monitor *monitor = arg;
 412
 413         condlog(3, "Releasing uevent_monitor() resources");
 414         udev_monitor_unref(monitor);
 415 }
 416
 417 /*
 418  * Service the uevent queue.
 419  */
 420 int uevent_dispatch(int (*uev_trigger)(struct uevent *, void * trigger_data),
 421                     void * trigger_data)
 422 {
 423         my_uev_trigger = uev_trigger;
 424         my_trigger_data = trigger_data;
 425
 426         mlockall(MCL_CURRENT | MCL_FUTURE);
 427
 428         while (1) {
 429                 LIST_HEAD(uevq_tmp);
 430
 431                 pthread_mutex_lock(uevq_lockp);
 432                 servicing_uev = 0;
 433                 /*
 434                  * Condition signals are unreliable,
 435                  * so make sure we only wait if we have to.
 436                  */
 437                 if (list_empty(&uevq)) {
 438                         pthread_cond_wait(uev_condp, uevq_lockp);
 439                 }
 440                 servicing_uev = 1;
 441                 list_splice_init(&uevq, &uevq_tmp);
 442                 pthread_mutex_unlock(uevq_lockp);
 443                 if (!my_uev_trigger)
 444                         break;
 445                 merge_uevq(&uevq_tmp);
 446                 service_uevq(&uevq_tmp);
 447         }
 448         condlog(3, "Terminating uev service queue");
 449         uevq_cleanup(&uevq);
 450         return 0;
 451 }
 452
 453 struct uevent *uevent_from_buffer(char *buf, ssize_t buflen)
 454 {
 455         struct uevent *uev;
 456         char *buffer;
 457         size_t bufpos;
 458         int i;
 459         char *pos;
 460
 461         uev = alloc_uevent();
 462         if (!uev) {
 463                 condlog(1, "lost uevent, oom");
 464                 return NULL;
 465         }
 466
 467         if ((size_t)buflen > sizeof(buf)-1)
 468                 buflen = sizeof(buf)-1;
 469
 470         /*
 471          * Copy the shared receive buffer contents to buffer private
 472          * to this uevent so we can immediately reuse the shared buffer.
 473          */
 474         memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 475         buffer = uev->buffer;
 476         buffer[buflen] = '\0';
 477
 478         /* save start of payload */
 479         bufpos = strlen(buffer) + 1;
 480
 481         /* action string */
 482         uev->action = buffer;
 483         pos = strchr(buffer, '@');
 484         if (!pos) {
 485                 condlog(3, "bad action string '%s'", buffer);
 486                 FREE(uev);
 487                 return NULL;
 488         }
 489         pos[0] = '\0';
 490
 491         /* sysfs path */
 492         uev->devpath = &pos[1];
 493
 494         /* hotplug events have the environment attached - reconstruct envp[] */
 495         for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 496                 int keylen;
 497                 char *key;
 498
 499                 key = &buffer[bufpos];
 500                 keylen = strlen(key);
 501                 uev->envp[i] = key;
 502                 /* Filter out sequence number */
 503                 if (strncmp(key, "SEQNUM=", 7) == 0) {
 504                         char *eptr;
 505
 506                         uev->seqnum = strtoul(key + 7, &eptr, 10);
 507                         if (eptr == key + 7)
 508                                 uev->seqnum = -1;
 509                 }
 510                 bufpos += keylen + 1;
 511         }
 512         uev->envp[i] = NULL;
 513
 514         condlog(3, "uevent %ld '%s' from '%s'", uev->seqnum,
 515                 uev->action, uev->devpath);
 516         uev->kernel = strrchr(uev->devpath, '/');
 517         if (uev->kernel)
 518                 uev->kernel++;
 519
 520         /* print payload environment */
 521         for (i = 0; uev->envp[i] != NULL; i++)
 522                 condlog(5, "%s", uev->envp[i]);
 523
 524         return uev;
 525 }
 526
 527 int failback_listen(void)
 528 {
 529         int sock;
 530         struct sockaddr_nl snl;
 531         struct sockaddr_un sun;
 532         socklen_t addrlen;
 533         int retval;
 534         int rcvbufsz = 128*1024;
 535         int rcvsz = 0;
 536         int rcvszsz = sizeof(rcvsz);
 537         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 538         const int feature_on = 1;
 539         /*
 540          * First check whether we have a udev socket
 541          */
 542         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 543         sun.sun_family = AF_LOCAL;
 544         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 545         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 546
 547         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 548         if (sock >= 0) {
 549
 550                 condlog(3, "reading events from udev socket.");
 551
 552                 /* the bind takes care of ensuring only one copy running */
 553                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 554                 if (retval < 0) {
 555                         condlog(0, "bind failed, exit");
 556                         goto exit;
 557                 }
 558
 559                 /* enable receiving of the sender credentials */
 560                 retval = setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 561                                     &feature_on, sizeof(feature_on));
 562                 if (retval < 0) {
 563                         condlog(0, "failed to enable credential passing, exit");
 564                         goto exit;
 565                 }
 566
 567         } else {
 568                 /* Fallback to read kernel netlink events */
 569                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 570                 snl.nl_family = AF_NETLINK;
 571                 snl.nl_pid = getpid();
 572                 snl.nl_groups = 0x01;
 573
 574                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 575                 if (sock == -1) {
 576                         condlog(0, "error getting socket, exit");
 577                         return 1;
 578                 }
 579
 580                 condlog(3, "reading events from kernel.");
 581
 582                 /*
 583                  * try to avoid dropping uevents, even so, this is not a guarantee,
 584                  * but it does help to change the netlink uevent socket's
 585                  * receive buffer threshold from the default value of 106,496 to
 586                  * the maximum value of 262,142.
 587                  */
 588                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 589                                     sizeof(rcvbufsz));
 590
 591                 if (retval < 0) {
 592                         condlog(0, "error setting receive buffer size for socket, exit");
 593                         exit(1);
 594                 }
 595                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 596                 if (retval < 0) {
 597                         condlog(0, "error setting receive buffer size for socket, exit");
 598                         exit(1);
 599                 }
 600                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 601
 602                 /* enable receiving of the sender credentials */
 603                 if (setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 604                                &feature_on, sizeof(feature_on)) < 0) {
 605                         condlog(0, "error on enabling credential passing for socket");
 606                         exit(1);
 607                 }
 608
 609                 retval = bind(sock, (struct sockaddr *) &snl,
 610                               sizeof(struct sockaddr_nl));
 611                 if (retval < 0) {
 612                         condlog(0, "bind failed, exit");
 613                         goto exit;
 614                 }
 615         }
 616
 617         while (1) {
 618                 size_t bufpos;
 619                 ssize_t buflen;
 620                 struct uevent *uev;
 621                 struct msghdr smsg;
 622                 struct iovec iov;
 623                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 624                 struct cmsghdr *cmsg;
 625                 struct ucred *cred;
 626                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 627
 628                 memset(buf, 0x00, sizeof(buf));
 629                 iov.iov_base = &buf;
 630                 iov.iov_len = sizeof(buf);
 631                 memset (&smsg, 0x00, sizeof(struct msghdr));
 632                 smsg.msg_iov = &iov;
 633                 smsg.msg_iovlen = 1;
 634                 smsg.msg_control = cred_msg;
 635                 smsg.msg_controllen = sizeof(cred_msg);
 636
 637                 buflen = recvmsg(sock, &smsg, 0);
 638                 if (buflen < 0) {
 639                         if (errno != EINTR)
 640                                 condlog(0, "error receiving message, errno %d", errno);
 641                         continue;
 642                 }
 643
 644                 cmsg = CMSG_FIRSTHDR(&smsg);
 645                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 646                         condlog(3, "no sender credentials received, message ignored");
 647                         continue;
 648                 }
 649
 650                 cred = (struct ucred *)CMSG_DATA(cmsg);
 651                 if (cred->uid != 0) {
 652                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 653                         continue;
 654                 }
 655
 656                 /* skip header */
 657                 bufpos = strlen(buf) + 1;
 658                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 659                         condlog(3, "invalid message length");
 660                         continue;
 661                 }
 662
 663                 /* check message header */
 664                 if (strstr(buf, "@/") == NULL) {
 665                         condlog(3, "unrecognized message header");
 666                         continue;
 667                 }
 668                 if ((size_t)buflen > sizeof(buf)-1) {
 669                         condlog(2, "buffer overflow for received uevent");
 670                         buflen = sizeof(buf)-1;
 671                 }
 672
 673                 uev = uevent_from_buffer(buf, buflen);
 674                 if (!uev)
 675                         continue;
 676                 /*
 677                  * Queue uevent and poke service pthread.
 678                  */
 679                 pthread_mutex_lock(uevq_lockp);
 680                 list_add_tail(&uev->node, &uevq);
 681                 pthread_cond_signal(uev_condp);
 682                 pthread_mutex_unlock(uevq_lockp);
 683         }
 684
 685 exit:
 686         close(sock);
 687         return 1;
 688 }
 689
 690 struct uevent *uevent_from_udev_device(struct udev_device *dev)
 691 {
 692         struct uevent *uev;
 693         int i = 0;
 694         char *pos, *end;
 695         struct udev_list_entry *list_entry;
 696
 697         uev = alloc_uevent();
 698         if (!uev) {
 699                 udev_device_unref(dev);
 700                 condlog(1, "lost uevent, oom");
 701                 return NULL;
 702         }
 703         pos = uev->buffer;
 704         end = pos + HOTPLUG_BUFFER_SIZE + OBJECT_SIZE - 1;
 705         udev_list_entry_foreach(list_entry, udev_device_get_properties_list_entry(dev)) {
 706                 const char *name, *value;
 707                 int bytes;
 708
 709                 name = udev_list_entry_get_name(list_entry);
 710                 if (!name)
 711                         name = "(null)";
 712                 value = udev_list_entry_get_value(list_entry);
 713                 if (!value)
 714                         value = "(null)";
 715                 bytes = snprintf(pos, end - pos, "%s=%s", name, value);
 716                 if (pos + bytes >= end) {
 717                         condlog(2, "buffer overflow for uevent");
 718                         break;
 719                 }
 720                 uev->envp[i] = pos;
 721                 pos += bytes;
 722                 *pos = '\0';
 723                 pos++;
 724                 if (strcmp(name, "DEVPATH") == 0)
 725                         uev->devpath = uev->envp[i] + 8;
 726                 if (strcmp(name, "ACTION") == 0)
 727                         uev->action = uev->envp[i] + 7;
 728                 i++;
 729                 if (i == HOTPLUG_NUM_ENVP - 1)
 730                         break;
 731         }
 732         uev->udev = dev;
 733         uev->envp[i] = NULL;
 734
 735         condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 736         uev->kernel = strrchr(uev->devpath, '/');
 737         if (uev->kernel)
 738                 uev->kernel++;
 739
 740         /* print payload environment */
 741         for (i = 0; uev->envp[i] != NULL; i++)
 742                 condlog(5, "%s", uev->envp[i]);
 743         return uev;
 744 }
 745
 746 bool uevent_burst(struct timeval *start_time, int events)
 747 {
 748         struct timeval diff_time, end_time;
 749         unsigned long speed;
 750         unsigned long eclipse_ms;
 751
 752         if(events > MAX_ACCUMULATION_COUNT) {
 753                 condlog(2, "burst got %u uevents, too much uevents, stopped", events);
 754                 return false;
 755         }
 756
 757         gettimeofday(&end_time, NULL);
 758         timersub(&end_time, start_time, &diff_time);
 759
 760         eclipse_ms = diff_time.tv_sec * 1000 + diff_time.tv_usec / 1000;
 761
 762         if (eclipse_ms == 0)
 763                 return true;
 764
 765         if (eclipse_ms > MAX_ACCUMULATION_TIME) {
 766                 condlog(2, "burst continued %lu ms, too long time, stopped", eclipse_ms);
 767                 return false;
 768         }
 769
 770         speed = (events * 1000) / eclipse_ms;
 771         if (speed > MIN_BURST_SPEED)
 772                 return true;
 773
 774         return false;
 775 }
 776
 777 int uevent_listen(struct udev *udev)
 778 {
 779         int err = 2;
 780         struct udev_monitor *monitor = NULL;
 781         int fd, socket_flags, events;
 782         struct timeval start_time;
 783         int need_failback = 1;
 784         int timeout = 30;
 785         LIST_HEAD(uevlisten_tmp);
 786
 787         /*
 788          * Queue uevents for service by dedicated thread so that the uevent
 789          * listening thread does not block on multipathd locks (vecs->lock)
 790          * thereby not getting to empty the socket's receive buffer queue
 791          * often enough.
 792          */
 793         if (!udev) {
 794                 condlog(1, "no udev context");
 795                 return 1;
 796         }
 797         udev_ref(udev);
 798         pthread_cleanup_push(uevent_cleanup, udev);
 799
 800         monitor = udev_monitor_new_from_netlink(udev, "udev");
 801         if (!monitor) {
 802                 condlog(2, "failed to create udev monitor");
 803                 goto out;
 804         }
 805         pthread_cleanup_push(monitor_cleanup, monitor);
 806 #ifdef LIBUDEV_API_RECVBUF
 807         if (udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024))
 808                 condlog(2, "failed to increase buffer size");
 809 #endif
 810         fd = udev_monitor_get_fd(monitor);
 811         if (fd < 0) {
 812                 condlog(2, "failed to get monitor fd");
 813                 goto out;
 814         }
 815         socket_flags = fcntl(fd, F_GETFL);
 816         if (socket_flags < 0) {
 817                 condlog(2, "failed to get monitor socket flags : %s",
 818                         strerror(errno));
 819                 goto out;
 820         }
 821         if (fcntl(fd, F_SETFL, socket_flags & ~O_NONBLOCK) < 0) {
 822                 condlog(2, "failed to set monitor socket flags : %s",
 823                         strerror(errno));
 824                 goto out;
 825         }
 826         err = udev_monitor_filter_add_match_subsystem_devtype(monitor, "block",
 827                                                               "disk");
 828         if (err)
 829                 condlog(2, "failed to create filter : %s", strerror(-err));
 830         err = udev_monitor_enable_receiving(monitor);
 831         if (err) {
 832                 condlog(2, "failed to enable receiving : %s", strerror(-err));
 833                 goto out;
 834         }
 835
 836         events = 0;
 837         gettimeofday(&start_time, NULL);
 838         while (1) {
 839                 struct uevent *uev;
 840                 struct udev_device *dev;
 841                 struct pollfd ev_poll;
 842                 int poll_timeout;
 843                 int fdcount;
 844
 845                 memset(&ev_poll, 0, sizeof(struct pollfd));
 846                 ev_poll.fd = fd;
 847                 ev_poll.events = POLLIN;
 848                 poll_timeout = timeout * 1000;
 849                 errno = 0;
 850                 fdcount = poll(&ev_poll, 1, poll_timeout);
 851                 if (fdcount && ev_poll.revents & POLLIN) {
 852                         timeout = uevent_burst(&start_time, events + 1) ? 1 : 0;
 853                         dev = udev_monitor_receive_device(monitor);
 854                         if (!dev) {
 855                                 condlog(0, "failed getting udev device");
 856                                 continue;
 857                         }
 858                         uev = uevent_from_udev_device(dev);
 859                         if (!uev)
 860                                 continue;
 861                         list_add_tail(&uev->node, &uevlisten_tmp);
 862                         events++;
 863                         continue;
 864                 }
 865                 if (fdcount < 0) {
 866                         if (errno == EINTR)
 867                                 continue;
 868
 869                         condlog(0, "error receiving "
 870                                 "uevent message: %m");
 871                         err = -errno;
 872                         break;
 873                 }
 874                 if (!list_empty(&uevlisten_tmp)) {
 875                         /*
 876                          * Queue uevents and poke service pthread.
 877                          */
 878                         condlog(3, "Forwarding %d uevents", events);
 879                         pthread_mutex_lock(uevq_lockp);
 880                         list_splice_tail_init(&uevlisten_tmp, &uevq);
 881                         pthread_cond_signal(uev_condp);
 882                         pthread_mutex_unlock(uevq_lockp);
 883                         events = 0;
 884                 }
 885                 gettimeofday(&start_time, NULL);
 886                 timeout = 30;
 887         }
 888         need_failback = 0;
 889 out:
 890         if (monitor)
 891                 pthread_cleanup_pop(1);
 892         if (need_failback)
 893                 err = failback_listen();
 894         pthread_cleanup_pop(1);
 895         return err;
 896 }
 897
 898 int uevent_get_major(const struct uevent *uev)
 899 {
 900         return uevent_get_env_positive_int(uev, "MAJOR");
 901 }
 902
 903 int uevent_get_minor(const struct uevent *uev)
 904 {
 905         return uevent_get_env_positive_int(uev, "MINOR");
 906 }
 907
 908 int uevent_get_disk_ro(const struct uevent *uev)
 909 {
 910         return uevent_get_env_positive_int(uev, "DISK_RO");
 911 }
 912
 913 static char *uevent_get_dm_str(const struct uevent *uev, char *attr)
 914 {
 915         const char *tmp = uevent_get_env_var(uev, attr);
 916
 917         if (tmp == NULL)
 918                 return NULL;
 919         return strdup(tmp);
 920 }
 921
 922 char *uevent_get_dm_name(const struct uevent *uev)
 923 {
 924         return uevent_get_dm_str(uev, "DM_NAME");
 925 }
 926
 927 char *uevent_get_dm_path(const struct uevent *uev)
 928 {
 929         return uevent_get_dm_str(uev, "DM_PATH");
 930 }
 931
 932 char *uevent_get_dm_action(const struct uevent *uev)
 933 {
 934         return uevent_get_dm_str(uev, "DM_ACTION");
 935 }
 936
 937 bool uevent_is_mpath(const struct uevent *uev)
 938 {
 939         const char *uuid = uevent_get_env_var(uev, "DM_UUID");
 940
 941         if (uuid == NULL)
 942                 return false;
 943         if (strncmp(uuid, UUID_PREFIX, UUID_PREFIX_LEN))
 944                 return false;
 945         return uuid[UUID_PREFIX_LEN] != '\0';
 946 }