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 <stdlib.h>
  29 #include <string.h>
  30 #include <fcntl.h>
  31 #include <time.h>
  32 #include <sys/socket.h>
  33 #include <sys/user.h>
  34 #include <asm/types.h>
  35 #include <linux/netlink.h>
  36 #include <pthread.h>
  37 #include <sys/mman.h>
  38
  39 #include "memory.h"
  40 #include "debug.h"
  41 #include "uevent.h"
  42
  43 typedef int (uev_trigger)(struct uevent *, void * trigger_data);
  44
  45 pthread_t uevq_thr;
  46 struct uevent *uevqhp, *uevqtp;
  47 pthread_mutex_t uevq_lock, *uevq_lockp = &uevq_lock;
  48 pthread_mutex_t uevc_lock, *uevc_lockp = &uevc_lock;
  49 pthread_cond_t  uev_cond,  *uev_condp  = &uev_cond;
  50 uev_trigger *my_uev_trigger;
  51 void * my_trigger_data;
  52
  53 struct uevent * alloc_uevent (void)
  54 {
  55         return (struct uevent *)MALLOC(sizeof(struct uevent));
  56 }
  57
  58 void
  59 service_uevq(void)
  60 {
  61         int empty;
  62         struct uevent *uev;
  63
  64         do {
  65                 pthread_mutex_lock(uevq_lockp);
  66                 empty = (uevqhp == NULL);
  67                 if (!empty) {
  68                         uev = uevqhp;
  69                         uevqhp = uev->next;
  70                         if (uevqtp == uev)
  71                                 uevqtp = uev->next;
  72                         pthread_mutex_unlock(uevq_lockp);
  73
  74                         if (my_uev_trigger && my_uev_trigger(uev,
  75                                                         my_trigger_data))
  76                                 condlog(0, "uevent trigger error");
  77
  78                         FREE(uev);
  79                 }
  80                 else {
  81                         pthread_mutex_unlock(uevq_lockp);
  82                 }
  83         } while (empty == 0);
  84 }
  85
  86 /*
  87  * Service the uevent queue.
  88  */
  89 static void *
  90 uevq_thread(void * et)
  91 {
  92         mlockall(MCL_CURRENT | MCL_FUTURE);
  93
  94         while (1) {
  95                 pthread_mutex_lock(uevc_lockp);
  96                 pthread_cond_wait(uev_condp, uevc_lockp);
  97                 pthread_mutex_unlock(uevc_lockp);
  98
  99                 service_uevq();
 100         }
 101 }
 102
 103 int uevent_listen(int (*uev_trigger)(struct uevent *, void * trigger_data),
 104                   void * trigger_data)
 105 {
 106         int sock;
 107         struct sockaddr_nl snl;
 108         int retval;
 109         int rcvbufsz = 128*1024;
 110         int rcvsz = 0;
 111         int rcvszsz = sizeof(rcvsz);
 112         unsigned int *prcvszsz = (unsigned int *)&rcvszsz;
 113         pthread_attr_t attr;
 114
 115         my_uev_trigger = uev_trigger;
 116         my_trigger_data = trigger_data;
 117
 118         /*
 119          * Queue uevents for service by dedicated thread so that the uevent
 120          * listening thread does not block on multipathd locks (vecs->lock)
 121          * thereby not getting to empty the socket's receive buffer queue
 122          * often enough.
 123          */
 124         uevqhp = uevqtp = NULL;
 125
 126         pthread_mutex_init(uevq_lockp, NULL);
 127         pthread_mutex_init(uevc_lockp, NULL);
 128         pthread_cond_init(uev_condp, NULL);
 129
 130         pthread_attr_init(&attr);
 131         pthread_attr_setstacksize(&attr, 64 * 1024);
 132         pthread_create(&uevq_thr, &attr, uevq_thread, NULL);
 133
 134         memset(&snl, 0x00, sizeof(struct sockaddr_nl));
 135         snl.nl_family = AF_NETLINK;
 136         snl.nl_pid = getpid();
 137         snl.nl_groups = 0xffffffff;
 138
 139         sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
 140         if (sock == -1) {
 141                 condlog(0, "error getting socket, exit");
 142                 return 1;
 143         }
 144
 145         /*
 146          * try to avoid dropping uevents, even so, this is not a guarantee,
 147          * but it does help to change the netlink uevent socket's
 148          * receive buffer threshold from the default value of 106,496 to
 149          * the maximum value of 262,142.
 150          */
 151         retval = setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsz,
 152                             sizeof(rcvbufsz));
 153
 154         if (retval < 0) {
 155                 condlog(0, "error setting receive buffer size for socket, exit");
 156                 exit(1);
 157         }
 158         retval = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvsz, prcvszsz);
 159
 160         if (retval < 0) {
 161                 condlog(0, "error setting receive buffer size for socket, exit");
 162                 exit(1);
 163         }
 164         condlog(3, "receive buffer size for socket is %u.", rcvsz);
 165
 166         retval = bind(sock, (struct sockaddr *) &snl,
 167                       sizeof(struct sockaddr_nl));
 168         if (retval < 0) {
 169                 condlog(0, "bind failed, exit");
 170                 goto exit;
 171         }
 172
 173         while (1) {
 174                 static char buff[HOTPLUG_BUFFER_SIZE + OBJECT_SIZE];
 175                 int i;
 176                 char *pos;
 177                 size_t bufpos;
 178                 ssize_t buflen;
 179                 struct uevent *uev;
 180                 char *buffer;
 181
 182                 buflen = recv(sock, &buff, sizeof(buff), 0);
 183                 if (buflen <  0) {
 184                         condlog(0, "error receiving message");
 185                         continue;
 186                 }
 187
 188                 if ((size_t)buflen > sizeof(buff)-1)
 189                         buflen = sizeof(buff)-1;
 190
 191                 uev = alloc_uevent();
 192
 193                 if (!uev) {
 194                         condlog(1, "lost uevent, oom");
 195                         continue;
 196                 }
 197
 198                 /*
 199                  * Copy the shared receive buffer contents to buffer private
 200                  * to this uevent so we can immediately reuse the shared buffer.
 201                  */
 202                 memcpy(uev->buffer, buff, HOTPLUG_BUFFER_SIZE + OBJECT_SIZE);
 203                 buffer = uev->buffer;
 204                 buffer[buflen] = '\0';
 205
 206                 /* save start of payload */
 207                 bufpos = strlen(buffer) + 1;
 208
 209                 /* action string */
 210                 uev->action = buffer;
 211                 pos = strchr(buffer, '@');
 212                 if (!pos)
 213                         continue;
 214                 pos[0] = '\0';
 215
 216                 /* sysfs path */
 217                 uev->devpath = &pos[1];
 218
 219                 /* hotplug events have the environment attached - reconstruct envp[] */
 220                 for (i = 0; (bufpos < (size_t)buflen) && (i < HOTPLUG_NUM_ENVP-1); i++) {
 221                         int keylen;
 222                         char *key;
 223
 224                         key = &buffer[bufpos];
 225                         keylen = strlen(key);
 226                         uev->envp[i] = key;
 227                         bufpos += keylen + 1;
 228                 }
 229                 uev->envp[i] = NULL;
 230
 231                 condlog(3, "uevent '%s' from '%s'", uev->action, uev->devpath);
 232
 233                 /* print payload environment */
 234                 for (i = 0; uev->envp[i] != NULL; i++)
 235                         condlog(3, "%s", uev->envp[i]);
 236
 237                 /*
 238                  * Queue uevent and poke service pthread.
 239                  */
 240                 pthread_mutex_lock(uevq_lockp);
 241                 if (uevqtp)
 242                         uevqtp->next = uev;
 243                 else
 244                         uevqhp = uev;
 245                 uevqtp = uev;
 246                 uev->next = NULL;
 247                 pthread_mutex_unlock(uevq_lockp);
 248
 249                 pthread_mutex_lock(uevc_lockp);
 250                 pthread_cond_signal(uev_condp);
 251                 pthread_mutex_unlock(uevc_lockp);
 252         }
 253
 254 exit:
 255         close(sock);
 256
 257         pthread_mutex_lock(uevq_lockp);
 258         pthread_cancel(uevq_thr);
 259         pthread_mutex_unlock(uevq_lockp);
 260
 261         pthread_mutex_destroy(uevq_lockp);
 262         pthread_mutex_destroy(uevc_lockp);
 263         pthread_cond_destroy(uev_condp);
 264
 265         return 1;
 266 }