1 /* SPDX-License-Identifier: LGPL-2.1+ */
4 #include <linux/filter.h>
5 #include <linux/netlink.h>
11 #include "MurmurHash2.h"
12 #include "alloc-util.h"
13 #include "device-monitor-private.h"
14 #include "device-private.h"
15 #include "device-util.h"
17 #include "format-util.h"
20 #include "mountpoint-util.h"
22 #include "socket-util.h"
23 #include "string-util.h"
26 struct sd_device_monitor
{
30 union sockaddr_union snl
;
31 union sockaddr_union snl_trusted_sender
;
34 Hashmap
*subsystem_filter
;
39 sd_event_source
*event_source
;
40 sd_device_monitor_handler_t callback
;
44 #define UDEV_MONITOR_MAGIC 0xfeedcafe
46 typedef struct monitor_netlink_header
{
47 /* "libudev" prefix to distinguish libudev and kernel messages */
49 /* Magic to protect against daemon <-> Library message format mismatch
50 * Used in the kernel from socket filter rules; needs to be stored in network order */
52 /* Total length of header structure known to the sender */
54 /* Properties string buffer */
55 unsigned properties_off
;
56 unsigned properties_len
;
57 /* Hashes of primary device properties strings, to let libudev subscribers
58 * use in-kernel socket filters; values need to be stored in network order */
59 unsigned filter_subsystem_hash
;
60 unsigned filter_devtype_hash
;
61 unsigned filter_tag_bloom_hi
;
62 unsigned filter_tag_bloom_lo
;
63 } monitor_netlink_header
;
65 static int monitor_set_nl_address(sd_device_monitor
*m
) {
66 union sockaddr_union snl
;
71 /* Get the address the kernel has assigned us.
72 * It is usually, but not necessarily the pid. */
73 addrlen
= sizeof(struct sockaddr_nl
);
74 if (getsockname(m
->sock
, &snl
.sa
, &addrlen
) < 0)
77 m
->snl
.nl
.nl_pid
= snl
.nl
.nl_pid
;
81 int device_monitor_allow_unicast_sender(sd_device_monitor
*m
, sd_device_monitor
*sender
) {
82 assert_return(m
, -EINVAL
);
83 assert_return(sender
, -EINVAL
);
85 m
->snl_trusted_sender
.nl
.nl_pid
= sender
->snl
.nl
.nl_pid
;
89 _public_
int sd_device_monitor_set_receive_buffer_size(sd_device_monitor
*m
, size_t size
) {
90 int r
, n
= (int) size
;
92 assert_return(m
, -EINVAL
);
93 assert_return((size_t) n
== size
, -EINVAL
);
95 if (setsockopt_int(m
->sock
, SOL_SOCKET
, SO_RCVBUFFORCE
, n
) < 0) {
96 r
= setsockopt_int(m
->sock
, SOL_SOCKET
, SO_RCVBUF
, n
);
104 int device_monitor_disconnect(sd_device_monitor
*m
) {
107 m
->sock
= safe_close(m
->sock
);
111 int device_monitor_get_fd(sd_device_monitor
*m
) {
112 assert_return(m
, -EINVAL
);
117 int device_monitor_new_full(sd_device_monitor
**ret
, MonitorNetlinkGroup group
, int fd
) {
118 _cleanup_(sd_device_monitor_unrefp
) sd_device_monitor
*m
= NULL
;
119 _cleanup_close_
int sock
= -1;
122 assert_return(ret
, -EINVAL
);
123 assert_return(group
>= 0 && group
< _MONITOR_NETLINK_GROUP_MAX
, -EINVAL
);
125 if (group
== MONITOR_GROUP_UDEV
&&
126 access("/run/udev/control", F_OK
) < 0 &&
127 dev_is_devtmpfs() <= 0) {
130 * We do not support subscribing to uevents if no instance of
131 * udev is running. Uevents would otherwise broadcast the
132 * processing data of the host into containers, which is not
135 * Containers will currently not get any udev uevents, until
136 * a supporting infrastructure is available.
138 * We do not set a netlink multicast group here, so the socket
139 * will not receive any messages.
142 log_debug("sd-device-monitor: The udev service seems not to be active, disabling the monitor");
143 group
= MONITOR_GROUP_NONE
;
147 sock
= socket(PF_NETLINK
, SOCK_RAW
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, NETLINK_KOBJECT_UEVENT
);
149 return log_debug_errno(errno
, "sd-device-monitor: Failed to create socket: %m");
152 m
= new(sd_device_monitor
, 1);
156 *m
= (sd_device_monitor
) {
158 .sock
= fd
>= 0 ? fd
: TAKE_FD(sock
),
160 .snl
.nl
.nl_family
= AF_NETLINK
,
161 .snl
.nl
.nl_groups
= group
,
165 r
= monitor_set_nl_address(m
);
167 return log_debug_errno(r
, "sd-device-monitor: Failed to set netlink address: %m");
174 _public_
int sd_device_monitor_new(sd_device_monitor
**ret
) {
175 return device_monitor_new_full(ret
, MONITOR_GROUP_UDEV
, -1);
178 _public_
int sd_device_monitor_stop(sd_device_monitor
*m
) {
179 assert_return(m
, -EINVAL
);
181 m
->event_source
= sd_event_source_unref(m
->event_source
);
182 (void) device_monitor_disconnect(m
);
187 static int device_monitor_event_handler(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
188 _cleanup_(sd_device_unrefp
) sd_device
*device
= NULL
;
189 sd_device_monitor
*m
= userdata
;
193 if (device_monitor_receive_device(m
, &device
) <= 0)
197 return m
->callback(m
, device
, m
->userdata
);
202 _public_
int sd_device_monitor_start(sd_device_monitor
*m
, sd_device_monitor_handler_t callback
, void *userdata
) {
205 assert_return(m
, -EINVAL
);
208 r
= sd_device_monitor_attach_event(m
, NULL
);
213 r
= device_monitor_enable_receiving(m
);
217 m
->callback
= callback
;
218 m
->userdata
= userdata
;
220 r
= sd_event_add_io(m
->event
, &m
->event_source
, m
->sock
, EPOLLIN
, device_monitor_event_handler
, m
);
224 (void) sd_event_source_set_description(m
->event_source
, "sd-device-monitor");
229 _public_
int sd_device_monitor_detach_event(sd_device_monitor
*m
) {
230 assert_return(m
, -EINVAL
);
232 (void) sd_device_monitor_stop(m
);
233 m
->event
= sd_event_unref(m
->event
);
238 _public_
int sd_device_monitor_attach_event(sd_device_monitor
*m
, sd_event
*event
) {
241 assert_return(m
, -EINVAL
);
242 assert_return(!m
->event
, -EBUSY
);
245 m
->event
= sd_event_ref(event
);
247 r
= sd_event_default(&m
->event
);
255 _public_ sd_event
*sd_device_monitor_get_event(sd_device_monitor
*m
) {
256 assert_return(m
, NULL
);
261 _public_ sd_event_source
*sd_device_monitor_get_event_source(sd_device_monitor
*m
) {
262 assert_return(m
, NULL
);
264 return m
->event_source
;
267 int device_monitor_enable_receiving(sd_device_monitor
*m
) {
270 assert_return(m
, -EINVAL
);
272 r
= sd_device_monitor_filter_update(m
);
274 return log_debug_errno(r
, "sd-device-monitor: Failed to update filter: %m");
277 /* enable receiving of sender credentials */
278 r
= setsockopt_int(m
->sock
, SOL_SOCKET
, SO_PASSCRED
, true);
280 return log_debug_errno(r
, "sd-device-monitor: Failed to set socket option SO_PASSCRED: %m");
282 if (bind(m
->sock
, &m
->snl
.sa
, sizeof(struct sockaddr_nl
)) < 0)
283 return log_debug_errno(errno
, "sd-device-monitor: Failed to bind monitoring socket: %m");
287 r
= monitor_set_nl_address(m
);
289 return log_debug_errno(r
, "sd-device-monitor: Failed to set address: %m");
295 static sd_device_monitor
*device_monitor_free(sd_device_monitor
*m
) {
298 (void) sd_device_monitor_detach_event(m
);
300 hashmap_free_free_free(m
->subsystem_filter
);
301 set_free_free(m
->tag_filter
);
306 DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_device_monitor
, sd_device_monitor
, device_monitor_free
);
308 static int passes_filter(sd_device_monitor
*m
, sd_device
*device
) {
309 const char *tag
, *subsystem
, *devtype
, *s
, *d
= NULL
;
313 assert_return(m
, -EINVAL
);
314 assert_return(device
, -EINVAL
);
316 if (hashmap_isempty(m
->subsystem_filter
))
319 r
= sd_device_get_subsystem(device
, &s
);
323 r
= sd_device_get_devtype(device
, &d
);
324 if (r
< 0 && r
!= -ENOENT
)
327 HASHMAP_FOREACH_KEY(devtype
, subsystem
, m
->subsystem_filter
, i
) {
328 if (!streq(s
, subsystem
))
337 if (streq(d
, devtype
))
344 if (set_isempty(m
->tag_filter
))
347 SET_FOREACH(tag
, m
->tag_filter
, i
)
348 if (sd_device_has_tag(device
, tag
) > 0)
354 int device_monitor_receive_device(sd_device_monitor
*m
, sd_device
**ret
) {
355 _cleanup_(sd_device_unrefp
) sd_device
*device
= NULL
;
357 monitor_netlink_header nlh
;
362 .iov_len
= sizeof(buf
)
364 char cred_msg
[CMSG_SPACE(sizeof(struct ucred
))];
365 union sockaddr_union snl
;
366 struct msghdr smsg
= {
369 .msg_control
= cred_msg
,
370 .msg_controllen
= sizeof(cred_msg
),
372 .msg_namelen
= sizeof(snl
),
374 struct cmsghdr
*cmsg
;
376 ssize_t buflen
, bufpos
;
377 bool is_initialized
= false;
382 buflen
= recvmsg(m
->sock
, &smsg
, 0);
385 log_debug_errno(errno
, "sd-device-monitor: Failed to receive message: %m");
389 if (buflen
< 32 || (smsg
.msg_flags
& MSG_TRUNC
))
390 return log_debug_errno(SYNTHETIC_ERRNO(EINVAL
),
391 "sd-device-monitor: Invalid message length.");
393 if (snl
.nl
.nl_groups
== MONITOR_GROUP_NONE
) {
394 /* unicast message, check if we trust the sender */
395 if (m
->snl_trusted_sender
.nl
.nl_pid
== 0 ||
396 snl
.nl
.nl_pid
!= m
->snl_trusted_sender
.nl
.nl_pid
)
397 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
398 "sd-device-monitor: Unicast netlink message ignored.");
400 } else if (snl
.nl
.nl_groups
== MONITOR_GROUP_KERNEL
) {
401 if (snl
.nl
.nl_pid
> 0)
402 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
403 "sd-device-monitor: Multicast kernel netlink message from PID %"PRIu32
" ignored.", snl
.nl
.nl_pid
);
406 cmsg
= CMSG_FIRSTHDR(&smsg
);
407 if (!cmsg
|| cmsg
->cmsg_type
!= SCM_CREDENTIALS
)
408 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
409 "sd-device-monitor: No sender credentials received, message ignored.");
411 cred
= (struct ucred
*) CMSG_DATA(cmsg
);
413 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
414 "sd-device-monitor: Sender uid="UID_FMT
", message ignored.", cred
->uid
);
416 if (streq(buf
.raw
, "libudev")) {
417 /* udev message needs proper version magic */
418 if (buf
.nlh
.magic
!= htobe32(UDEV_MONITOR_MAGIC
))
419 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
420 "sd-device-monitor: Invalid message signature (%x != %x)",
421 buf
.nlh
.magic
, htobe32(UDEV_MONITOR_MAGIC
));
423 if (buf
.nlh
.properties_off
+32 > (size_t) buflen
)
424 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
425 "sd-device-monitor: Invalid message length (%u > %zd)",
426 buf
.nlh
.properties_off
+32, buflen
);
428 bufpos
= buf
.nlh
.properties_off
;
430 /* devices received from udev are always initialized */
431 is_initialized
= true;
434 /* kernel message with header */
435 bufpos
= strlen(buf
.raw
) + 1;
436 if ((size_t) bufpos
< sizeof("a@/d") || bufpos
>= buflen
)
437 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
438 "sd-device-monitor: Invalid message length");
440 /* check message header */
441 if (!strstr(buf
.raw
, "@/"))
442 return log_debug_errno(SYNTHETIC_ERRNO(EAGAIN
),
443 "sd-device-monitor: Invalid message header");
446 r
= device_new_from_nulstr(&device
, (uint8_t*) &buf
.raw
[bufpos
], buflen
- bufpos
);
448 return log_debug_errno(r
, "sd-device-monitor: Failed to create device from received message: %m");
451 device_set_is_initialized(device
);
453 /* Skip device, if it does not pass the current filter */
454 r
= passes_filter(m
, device
);
456 return log_device_debug_errno(device
, r
, "sd-device-monitor: Failed to check received device passing filter: %m");
458 log_device_debug(device
, "sd-device-monitor: Received device does not pass filter, ignoring");
460 *ret
= TAKE_PTR(device
);
465 static uint32_t string_hash32(const char *str
) {
466 return MurmurHash2(str
, strlen(str
), 0);
469 /* Get a bunch of bit numbers out of the hash, and set the bits in our bit field */
470 static uint64_t string_bloom64(const char *str
) {
472 uint32_t hash
= string_hash32(str
);
474 bits
|= 1LLU << (hash
& 63);
475 bits
|= 1LLU << ((hash
>> 6) & 63);
476 bits
|= 1LLU << ((hash
>> 12) & 63);
477 bits
|= 1LLU << ((hash
>> 18) & 63);
481 int device_monitor_send_device(
482 sd_device_monitor
*m
,
483 sd_device_monitor
*destination
,
486 monitor_netlink_header nlh
= {
488 .magic
= htobe32(UDEV_MONITOR_MAGIC
),
489 .header_size
= sizeof nlh
,
491 struct iovec iov
[2] = {
492 { .iov_base
= &nlh
, .iov_len
= sizeof nlh
},
494 struct msghdr smsg
= {
498 /* default destination for sending */
499 union sockaddr_union default_destination
= {
500 .nl
.nl_family
= AF_NETLINK
,
501 .nl
.nl_groups
= MONITOR_GROUP_UDEV
,
503 uint64_t tag_bloom_bits
;
504 const char *buf
, *val
;
512 r
= device_get_properties_nulstr(device
, (const uint8_t **) &buf
, &blen
);
514 return log_device_debug_errno(device
, r
, "sd-device-monitor: Failed to get device properties: %m");
516 log_device_debug(device
, "sd-device-monitor: Length of device property nulstr is too small to contain valid device information");
520 /* fill in versioned header */
521 r
= sd_device_get_subsystem(device
, &val
);
523 return log_device_debug_errno(device
, r
, "sd-device-monitor: Failed to get device subsystem: %m");
524 nlh
.filter_subsystem_hash
= htobe32(string_hash32(val
));
526 if (sd_device_get_devtype(device
, &val
) >= 0)
527 nlh
.filter_devtype_hash
= htobe32(string_hash32(val
));
529 /* add tag bloom filter */
531 FOREACH_DEVICE_TAG(device
, val
)
532 tag_bloom_bits
|= string_bloom64(val
);
534 if (tag_bloom_bits
> 0) {
535 nlh
.filter_tag_bloom_hi
= htobe32(tag_bloom_bits
>> 32);
536 nlh
.filter_tag_bloom_lo
= htobe32(tag_bloom_bits
& 0xffffffff);
539 /* add properties list */
540 nlh
.properties_off
= iov
[0].iov_len
;
541 nlh
.properties_len
= blen
;
542 iov
[1] = IOVEC_MAKE((char*) buf
, blen
);
545 * Use custom address for target, or the default one.
547 * If we send to a multicast group, we will get
548 * ECONNREFUSED, which is expected.
550 smsg
.msg_name
= destination
? &destination
->snl
: &default_destination
;
551 smsg
.msg_namelen
= sizeof(struct sockaddr_nl
);
552 count
= sendmsg(m
->sock
, &smsg
, 0);
554 if (!destination
&& errno
== ECONNREFUSED
) {
555 log_device_debug(device
, "sd-device-monitor: Passed to netlink monitor");
558 return log_device_debug_errno(device
, errno
, "sd-device-monitor: Failed to send device to netlink monitor: %m");
561 log_device_debug(device
, "sd-device-monitor: Passed %zi byte to netlink monitor", count
);
565 static void bpf_stmt(struct sock_filter
*ins
, unsigned *i
,
566 unsigned short code
, unsigned data
) {
567 ins
[(*i
)++] = (struct sock_filter
) {
573 static void bpf_jmp(struct sock_filter
*ins
, unsigned *i
,
574 unsigned short code
, unsigned data
,
575 unsigned short jt
, unsigned short jf
) {
576 ins
[(*i
)++] = (struct sock_filter
) {
584 _public_
int sd_device_monitor_filter_update(sd_device_monitor
*m
) {
585 struct sock_filter ins
[512] = {};
586 struct sock_fprog filter
;
587 const char *subsystem
, *devtype
, *tag
;
591 assert_return(m
, -EINVAL
);
593 if (m
->filter_uptodate
)
596 if (hashmap_isempty(m
->subsystem_filter
) &&
597 set_isempty(m
->tag_filter
)) {
598 m
->filter_uptodate
= true;
602 /* load magic in A */
603 bpf_stmt(ins
, &i
, BPF_LD
|BPF_W
|BPF_ABS
, offsetof(monitor_netlink_header
, magic
));
604 /* jump if magic matches */
605 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, UDEV_MONITOR_MAGIC
, 1, 0);
606 /* wrong magic, pass packet */
607 bpf_stmt(ins
, &i
, BPF_RET
|BPF_K
, 0xffffffff);
609 if (!set_isempty(m
->tag_filter
)) {
610 int tag_matches
= set_size(m
->tag_filter
);
612 /* add all tags matches */
613 SET_FOREACH(tag
, m
->tag_filter
, it
) {
614 uint64_t tag_bloom_bits
= string_bloom64(tag
);
615 uint32_t tag_bloom_hi
= tag_bloom_bits
>> 32;
616 uint32_t tag_bloom_lo
= tag_bloom_bits
& 0xffffffff;
618 /* load device bloom bits in A */
619 bpf_stmt(ins
, &i
, BPF_LD
|BPF_W
|BPF_ABS
, offsetof(monitor_netlink_header
, filter_tag_bloom_hi
));
620 /* clear bits (tag bits & bloom bits) */
621 bpf_stmt(ins
, &i
, BPF_ALU
|BPF_AND
|BPF_K
, tag_bloom_hi
);
622 /* jump to next tag if it does not match */
623 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, tag_bloom_hi
, 0, 3);
625 /* load device bloom bits in A */
626 bpf_stmt(ins
, &i
, BPF_LD
|BPF_W
|BPF_ABS
, offsetof(monitor_netlink_header
, filter_tag_bloom_lo
));
627 /* clear bits (tag bits & bloom bits) */
628 bpf_stmt(ins
, &i
, BPF_ALU
|BPF_AND
|BPF_K
, tag_bloom_lo
);
629 /* jump behind end of tag match block if tag matches */
631 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, tag_bloom_lo
, 1 + (tag_matches
* 6), 0);
634 /* nothing matched, drop packet */
635 bpf_stmt(ins
, &i
, BPF_RET
|BPF_K
, 0);
638 /* add all subsystem matches */
639 if (!hashmap_isempty(m
->subsystem_filter
)) {
640 HASHMAP_FOREACH_KEY(devtype
, subsystem
, m
->subsystem_filter
, it
) {
641 uint32_t hash
= string_hash32(subsystem
);
643 /* load device subsystem value in A */
644 bpf_stmt(ins
, &i
, BPF_LD
|BPF_W
|BPF_ABS
, offsetof(monitor_netlink_header
, filter_subsystem_hash
));
646 /* jump if subsystem does not match */
647 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, hash
, 0, 1);
649 /* jump if subsystem does not match */
650 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, hash
, 0, 3);
651 /* load device devtype value in A */
652 bpf_stmt(ins
, &i
, BPF_LD
|BPF_W
|BPF_ABS
, offsetof(monitor_netlink_header
, filter_devtype_hash
));
653 /* jump if value does not match */
654 hash
= string_hash32(devtype
);
655 bpf_jmp(ins
, &i
, BPF_JMP
|BPF_JEQ
|BPF_K
, hash
, 0, 1);
658 /* matched, pass packet */
659 bpf_stmt(ins
, &i
, BPF_RET
|BPF_K
, 0xffffffff);
661 if (i
+1 >= ELEMENTSOF(ins
))
665 /* nothing matched, drop packet */
666 bpf_stmt(ins
, &i
, BPF_RET
|BPF_K
, 0);
669 /* matched, pass packet */
670 bpf_stmt(ins
, &i
, BPF_RET
|BPF_K
, 0xffffffff);
673 filter
= (struct sock_fprog
) {
677 if (setsockopt(m
->sock
, SOL_SOCKET
, SO_ATTACH_FILTER
, &filter
, sizeof(filter
)) < 0)
680 m
->filter_uptodate
= true;
684 _public_
int sd_device_monitor_filter_add_match_subsystem_devtype(sd_device_monitor
*m
, const char *subsystem
, const char *devtype
) {
685 _cleanup_free_
char *s
= NULL
, *d
= NULL
;
688 assert_return(m
, -EINVAL
);
689 assert_return(subsystem
, -EINVAL
);
691 s
= strdup(subsystem
);
701 r
= hashmap_ensure_allocated(&m
->subsystem_filter
, NULL
);
705 r
= hashmap_put(m
->subsystem_filter
, s
, d
);
710 m
->filter_uptodate
= false;
715 _public_
int sd_device_monitor_filter_add_match_tag(sd_device_monitor
*m
, const char *tag
) {
716 _cleanup_free_
char *t
= NULL
;
719 assert_return(m
, -EINVAL
);
720 assert_return(tag
, -EINVAL
);
726 r
= set_ensure_allocated(&m
->tag_filter
, &string_hash_ops
);
730 r
= set_put(m
->tag_filter
, t
);
737 m
->filter_uptodate
= false;
742 _public_
int sd_device_monitor_filter_remove(sd_device_monitor
*m
) {
743 static const struct sock_fprog filter
= { 0, NULL
};
745 assert_return(m
, -EINVAL
);
747 m
->subsystem_filter
= hashmap_free_free_free(m
->subsystem_filter
);
748 m
->tag_filter
= set_free_free(m
->tag_filter
);
750 if (setsockopt(m
->sock
, SOL_SOCKET
, SO_DETACH_FILTER
, &filter
, sizeof(filter
)) < 0)
753 m
->filter_uptodate
= true;