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