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