libmultipath/uevent.c

   1 /*
   2  * uevent.c - trigger upon netlink uevents from the kernel
   3  *
   4  *      Only kernels from version 2.6.10* on provide the uevent netlink socket.
   5  *      Until the libc-kernel-headers are updated, you need to compile with:
   6  *
   7  *        gcc -I /lib/modules/`uname -r`/build/include -o uevent_listen uevent_listen.c
   8  *
   9  * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
  10  *
  11  *      This program is free software; you can redistribute it and/or modify it
  12  *      under the terms of the GNU General Public License as published by the
  13  *      Free Software Foundation version 2 of the License.
  14  *
  15  *      This program is distributed in the hope that it will be useful, but
  16  *      WITHOUT ANY WARRANTY; without even the implied warranty of
  17  *      MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  *      General Public License for more details.
  19  *
  20  *      You should have received a copy of the GNU General Public License along
  21  *      with this program; if not, write to the Free Software Foundation, Inc.,
  22  *      675 Mass Ave, Cambridge, MA 02139, USA.
  23  *
  24  */
  25
  26 #include <unistd.h>
  27 #include <stdio.h>
  28 #include <errno.h>
  29 #include <stdlib.h>
  30 #include <stddef.h>
  31 #include <string.h>
  32 #include <fcntl.h>
  33 #include <time.h>
  34 #include <sys/socket.h>
  35 #include <sys/user.h>
  36 #include <sys/un.h>
  37 #include <linux/types.h>
  38 #include <linux/netlink.h>
  39 #include <pthread.h>
  40 #include <sys/mman.h>
  41
  42 #include "memory.h"
  43 #include "debug.h"
  44 #include "uevent.h"
  45
  46 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  47
  48 pthread_t uevq_thr;
  49 struct uevent *uevqhp, *uevqtp;
  50 pthread_mutex_t uevq_lock, *uevq_lockp = &uevq_lock;
  51 pthread_mutex_t uevc_lock, *uevc_lockp = &uevc_lock;
  52 pthread_cond_t  uev_cond,  *uev_condp  = &uev_cond;
  53 uev_trigger *my_uev_trigger;
  54 void * my_trigger_data;
  55
  56 static struct uevent * alloc_uevent (void)
  57 {
  58         return (struct uevent *)MALLOC(sizeof(struct uevent));
  59 }
  60
  61 void
  62 service_uevq(void)
  63 {
  64         int empty;
  65         struct uevent *uev;
  66
  67         do {
  68                 pthread_mutex_lock(uevq_lockp);
  69                 empty = (uevqhp == NULL);
  70                 if (!empty) {
  71                         uev = uevqhp;
  72                         uevqhp = uev->next;
  73                         if (uevqtp == uev)
  74                                 uevqtp = uev->next;
  75                         pthread_mutex_unlock(uevq_lockp);
  76
  77                         if (my_uev_trigger && my_uev_trigger(uev,
  78                                                         my_trigger_data))
  79                                 condlog(0, "uevent trigger error");
  80
  81                         FREE(uev);
  82                 }
  83                 else {
  84                         pthread_mutex_unlock(uevq_lockp);
  85                 }
  86         } while (empty == 0);
  87 }
  88
  89 /*
  90  * Service the uevent queue.
  91  */
  92 static void *
  93 uevq_thread(void * et)
  94 {
  95         mlockall(MCL_CURRENT | MCL_FUTURE);
  96
  97         while (1) {
  98                 pthread_mutex_lock(uevc_lockp);
  99                 pthread_cond_wait(uev_condp, uevc_lockp);
 100                 pthread_mutex_unlock(uevc_lockp);
 101
 102                 service_uevq();
 103         }
 104 }
 105
 106 int uevent_listen(int (*uev_trigger)(struct uevent *, void * trigger_data),
 107                   void * trigger_data)
 108 {
 109         int sock;
 110         struct sockaddr_nl snl;
 111         struct sockaddr_un sun;
 112         socklen_t addrlen;
 113         int retval;
 114         int rcvbufsz = 128*1024;
 115         int rcvsz = 0;
 116         int rcvszsz = sizeof(rcvsz);
 117         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 118         pthread_attr_t attr;
 119         const int feature_on = 1;
 120
 121         my_uev_trigger = uev_trigger;
 122         my_trigger_data = trigger_data;
 123
 124         /*
 125          * Queue uevents for service by dedicated thread so that the uevent
 126          * listening thread does not block on multipathd locks (vecs->lock)
 127          * thereby not getting to empty the socket's receive buffer queue
 128          * often enough.
 129          */
 130         uevqhp = uevqtp = NULL;
 131
 132         pthread_mutex_init(uevq_lockp, NULL);
 133         pthread_mutex_init(uevc_lockp, NULL);
 134         pthread_cond_init(uev_condp, NULL);
 135
 136         pthread_attr_init(&attr);
 137         pthread_attr_setstacksize(&attr, 64 * 1024);
 138         pthread_create(&uevq_thr, &attr, uevq_thread, NULL);
 139
 140         /*
 141          * First check whether we have a udev socket
 142          */
 143         memset(&sun, 0x00, sizeof(struct sockaddr_un));
 144         sun.sun_family = AF_LOCAL;
 145         strcpy(&sun.sun_path[1], "/org/kernel/dm/multipath_event");
 146         addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(sun.sun_path+1) + 1;
 147
 148         sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
 149         if (sock >= 0) {
 150
 151                 condlog(3, "reading events from udev socket.");
 152
 153                 /* the bind takes care of ensuring only one copy running */
 154                 retval = bind(sock, (struct sockaddr *) &sun, addrlen);
 155                 if (retval < 0) {
 156                         condlog(0, "bind failed, exit");
 157                         goto exit;
 158                 }
 159
 160                 /* enable receiving of the sender credentials */
 161                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 162                            &feature_on, sizeof(feature_on));
 163
 164         } else {
 165                 /* Fallback to read kernel netlink events */
 166                 memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 167                 snl.nl_family = AF_NETLINK;
 168                 snl.nl_pid = getpid();
 169                 snl.nl_groups = 0x01;
 170
 171                 sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 172                 if (sock == -1) {
 173                         condlog(0, "error getting socket, exit");
 174                         return 1;
 175                 }
 176
 177                 condlog(3, "reading events from kernel.");
 178
 179                 /*
 180                  * try to avoid dropping uevents, even so, this is not a guarantee,
 181                  * but it does help to change the netlink uevent socket's
 182                  * receive buffer threshold from the default value of 106,496 to
 183                  * the maximum value of 262,142.
 184                  */
 185                 retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 186                                     sizeof(rcvbufsz));
 187
 188                 if (retval < 0) {
 189                         condlog(0, "error setting receive buffer size for socket, exit");
 190                         exit(1);
 191                 }
 192                 retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 193                 if (retval < 0) {
 194                         condlog(0, "error setting receive buffer size for socket, exit");
 195                         exit(1);
 196                 }
 197                 condlog(3, "receive buffer size for socket is %u.", rcvsz);
 198
 199                 /* enable receiving of the sender credentials */
 200                 setsockopt(sock, SOL_SOCKET, SO_PASSCRED,
 201                            &feature_on, sizeof(feature_on));
 202
 203                 retval = bind(sock, (struct sockaddr *) &snl,
 204                               sizeof(struct sockaddr_nl));
 205                 if (retval < 0) {
 206                         condlog(0, "bind failed, exit");
 207                         goto exit;
 208                 }
 209         }
 210
 211         while (1) {
 212                 int i;
 213                 char *pos;
 214                 size_t bufpos;
 215                 ssize_t buflen;
 216                 struct uevent *uev;
 217                 char *buffer;
 218                 struct msghdr smsg;
 219                 struct iovec iov;
 220                 char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
 221                 struct cmsghdr *cmsg;
 222                 struct ucred *cred;
 223                 static char buf[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 224
 225                 memset(buf, 0x00, sizeof(buf));
 226                 iov.iov_base = &buf;
 227                 iov.iov_len = sizeof(buf);
 228                 memset (&smsg, 0x00, sizeof(struct msghdr));
 229                 smsg.msg_iov = &iov;
 230                 smsg.msg_iovlen = 1;
 231                 smsg.msg_control = cred_msg;
 232                 smsg.msg_controllen = sizeof(cred_msg);
 233
 234                 buflen = recvmsg(sock, &smsg, 0);
 235                 if (buflen < 0) {
 236                         if (errno != EINTR)
 237                                 condlog(0, "error receiving message");
 238                         continue;
 239                 }
 240
 241                 cmsg = CMSG_FIRSTHDR(&smsg);
 242                 if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
 243                         condlog(3, "no sender credentials received, message ignored");
 244                         continue;
 245                 }
 246
 247                 cred = (struct ucred *)CMSG_DATA(cmsg);
 248                 if (cred->uid != 0) {
 249                         condlog(3, "sender uid=%d, message ignored", cred->uid);
 250                         continue;
 251                 }
 252
 253                 /* skip header */
 254                 bufpos = strlen(buf) + 1;
 255                 if (bufpos < sizeof("a@/d") || bufpos >= sizeof(buf)) {
 256                         condlog(3, "invalid message length");
 257                         continue;
 258                 }
 259
 260                 /* check message header */
 261                 if (strstr(buf, "@/") == NULL) {
 262                         condlog(3, "unrecognized message header");
 263                         continue;
 264                 }
 265
 266                 uev = alloc_uevent();
 267
 268                 if (!uev) {
 269                         condlog(1, "lost uevent, oom");
 270                         continue;
 271                 }
 272
 273                 if ((size_t)buflen > sizeof(buf)-1)
 274                         buflen = sizeof(buf)-1;
 275
 276                 /*
 277                  * Copy the shared receive buffer contents to buffer private
 278                  * to this uevent so we can immediately reuse the shared buffer.
 279                  */
 280                 memcpy(uev->buffer, buf, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 281                 buffer = uev->buffer;
 282                 buffer[buflen] = '\0';
 283
 284                 /* save start of payload */
 285                 bufpos = strlen(buffer) + 1;
 286
 287                 /* action string */
 288                 uev->action = buffer;
 289                 pos = strchr(buffer, '@');
 290                 if (!pos) {
 291                         condlog(3, "bad action string '%s'", buffer);
 292                         continue;
 293                 }
 294                 pos[0] = '\0';
 295
 296                 /* sysfs path */
 297                 uev->devpath = &pos[1];
 298
 299                 /* hotplug events have the environment attached - reconstruct envp[] */
 300                 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 301                         int keylen;
 302                         char *key;
 303
 304                         key = &buffer[bufpos];
 305                         keylen = strlen(key);
 306                         uev->envp[i] = key;
 307                         bufpos += keylen + 1;
 308                 }
 309                 uev->envp[i] = NULL;
 310
 311                 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 312
 313                 /* print payload environment */
 314                 for (i = 0; uev->envp[i] != NULL; i++)
 315                         condlog(3, "%s", uev->envp[i]);
 316
 317                 /*
 318                  * Queue uevent and poke service pthread.
 319                  */
 320                 pthread_mutex_lock(uevq_lockp);
 321                 if (uevqtp)
 322                         uevqtp->next = uev;
 323                 else
 324                         uevqhp = uev;
 325                 uevqtp = uev;
 326                 uev->next = NULL;
 327                 pthread_mutex_unlock(uevq_lockp);
 328
 329                 pthread_mutex_lock(uevc_lockp);
 330                 pthread_cond_signal(uev_condp);
 331                 pthread_mutex_unlock(uevc_lockp);
 332         }
 333
 334 exit:
 335         close(sock);
 336
 337         pthread_mutex_lock(uevq_lockp);
 338         pthread_cancel(uevq_thr);
 339         pthread_mutex_unlock(uevq_lockp);
 340
 341         pthread_mutex_destroy(uevq_lockp);
 342         pthread_mutex_destroy(uevc_lockp);
 343         pthread_cond_destroy(uev_condp);
 344
 345         return 1;
 346 }