/*
- * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
+ * Copyright (c) 2000-2014 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
#include <miscfs/devfs/devfs.h>
#include <net/dlil.h>
+#include <net/pktap.h>
#include <kern/locks.h>
#include <kern/thread_call.h>
static unsigned int bpf_maxdevices = 256;
SYSCTL_UINT(_debug, OID_AUTO, bpf_maxdevices, CTLFLAG_RW | CTLFLAG_LOCKED,
&bpf_maxdevices, 0, "");
+/*
+ * bpf_wantpktap controls the defaul visibility of DLT_PKTAP
+ * For OS X is off by default so process need to use the ioctl BPF_WANT_PKTAP
+ * explicitly to be able to use DLT_PKTAP.
+ */
+static unsigned int bpf_wantpktap = 0;
+SYSCTL_UINT(_debug, OID_AUTO, bpf_wantpktap, CTLFLAG_RW | CTLFLAG_LOCKED,
+ &bpf_wantpktap, 0, "");
/*
* bpf_iflist is the list of interfaces; each corresponds to an ifnet
static lck_grp_attr_t *bpf_mlock_grp_attr;
static lck_attr_t *bpf_mlock_attr;
-/*
- * Mark a descriptor free by making it point to itself.
- * This is probably cheaper than marking with a constant since
- * the address should be in a register anyway.
- */
+static mbuf_tag_id_t bpf_mtag_id;
#endif /* __APPLE__ */
static int bpf_allocbufs(struct bpf_d *);
static void bpf_mcopy(const void *, void *, size_t);
static int bpf_movein(struct uio *, int,
struct mbuf **, struct sockaddr *, int *);
-static int bpf_setif(struct bpf_d *, ifnet_t ifp, u_int32_t dlt);
-static void bpf_timed_out(void *, void *);
-static void bpf_wakeup(struct bpf_d *);
+static int bpf_setif(struct bpf_d *, ifnet_t ifp, u_int32_t dlt, dev_t);
+static void bpf_timed_out(void *, void *);
+static void bpf_wakeup(struct bpf_d *);
static void catchpacket(struct bpf_d *, u_char *, struct mbuf *, u_int,
u_int, int, void (*)(const void *, void *, size_t));
static void reset_d(struct bpf_d *);
-static int bpf_setf(struct bpf_d *, u_int bf_len, user_addr_t bf_insns);
+static int bpf_setf(struct bpf_d *, u_int , user_addr_t , dev_t, u_long);
static int bpf_getdltlist(struct bpf_d *, caddr_t, struct proc *);
-static int bpf_setdlt(struct bpf_d *, u_int);
+static int bpf_setdlt(struct bpf_d *, u_int, dev_t);
static int bpf_set_traffic_class(struct bpf_d *, int);
static void bpf_set_packet_service_class(struct mbuf *, int);
#ifdef __APPLE__
/*
- * The dynamic addition of a new device node must block all processes that are opening
- * the last device so that no process will get an unexpected ENOENT
+ * The dynamic addition of a new device node must block all processes that
+ * are opening the last device so that no process will get an unexpected
+ * ENOENT
*/
static void
bpf_make_dev_t(int maj)
if (first) {
/* Find the default bpf entry for this ifp */
if (bp->bif_ifp->if_bpf == NULL) {
- struct bpf_if *primary;
+ struct bpf_if *tmp, *primary = NULL;
- for (primary = bpf_iflist; primary && primary->bif_ifp != bp->bif_ifp;
- primary = primary->bif_next)
- ;
-
+ for (tmp = bpf_iflist; tmp; tmp = tmp->bif_next) {
+ if (tmp->bif_ifp != bp->bif_ifp)
+ continue;
+ primary = tmp;
+ /*
+ * Make DLT_PKTAP only if process knows how
+ * to deal with it, otherwise find another one
+ */
+ if (tmp->bif_dlt == DLT_PKTAP &&
+ !(d->bd_flags & BPF_WANT_PKTAP))
+ continue;
+ break;
+ }
bp->bif_ifp->if_bpf = primary;
}
error = bp->bif_tap(bp->bif_ifp, bp->bif_dlt, BPF_TAP_INPUT_OUTPUT);
}
+ if (bp->bif_ifp->if_bpf != NULL &&
+ bp->bif_ifp->if_bpf->bif_dlt == DLT_PKTAP)
+ d->bd_flags |= BPF_FINALIZE_PKTAP;
+ else
+ d->bd_flags &= ~BPF_FINALIZE_PKTAP;
+
return error;
}
tv.tv_sec = d->bd_rtout / hz;
tv.tv_usec = (d->bd_rtout % hz) * tick;
- clock_interval_to_deadline((uint64_t)tv.tv_sec * USEC_PER_SEC + tv.tv_usec,
- NSEC_PER_USEC,
- &deadline);
+ clock_interval_to_deadline(
+ (uint64_t)tv.tv_sec * USEC_PER_SEC + tv.tv_usec,
+ NSEC_PER_USEC, &deadline);
/*
* The state is BPF_IDLE, so the timer hasn't
* been started yet, and hasn't gone off yet;
d->bd_state = BPF_IDLE;
d->bd_thread_call = thread_call_allocate(bpf_timed_out, d);
d->bd_traffic_class = SO_TC_BE;
+ if (bpf_wantpktap)
+ d->bd_flags |= BPF_WANT_PKTAP;
+ else
+ d->bd_flags &= ~BPF_WANT_PKTAP;
if (d->bd_thread_call == NULL) {
printf("bpfopen: malloc thread call failed\n");
mac_bpfdesc_label_destroy(d);
#endif
thread_call_free(d->bd_thread_call);
+
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
bpf_freed(d);
/* Mark free in same context as bpfopen comes to check */
return msleep1((caddr_t)d, bpf_mlock, pri, wmesg, abstime);
}
-static struct inpcb *
-bpf_findinpcb(struct inpcbinfo *pcbinfo, uint32_t flowhash)
-{
- struct inpcb *inp = NULL;
-
- if (!flowhash) return (NULL);
-
- lck_rw_lock_shared(pcbinfo->mtx);
- LIST_FOREACH(inp, pcbinfo->listhead, inp_list) {
- if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != WNT_STOPUSING) {
- if (inp->inp_flowhash == flowhash)
- break;
- in_pcb_checkstate(inp, WNT_RELEASE, 0);
- }
- }
- lck_rw_done(pcbinfo->mtx);
-
- return (inp);
-}
-
/*
* Rotate the packet buffers in descriptor d. Move the store buffer
* into the hold slot, and the free buffer into the store slot.
* Zero the length of the new store buffer.
*/
#define ROTATE_BUFFERS(d) \
+ if (d->bd_hbuf_read) \
+ panic("rotating bpf buffers during read"); \
(d)->bd_hbuf = (d)->bd_sbuf; \
(d)->bd_hlen = (d)->bd_slen; \
(d)->bd_sbuf = (d)->bd_fbuf; \
bpfread(dev_t dev, struct uio *uio, int ioflag)
{
struct bpf_d *d;
- int timed_out;
+ caddr_t hbuf;
+ int timed_out, hbuf_len;
int error;
+ int flags;
lck_mtx_lock(bpf_mlock);
timed_out = (d->bd_state == BPF_TIMED_OUT);
d->bd_state = BPF_IDLE;
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1) {
+ lck_mtx_unlock(bpf_mlock);
+ return (ENXIO);
+ }
/*
* If the hold buffer is empty, then do a timed sleep, which
* ends when the timeout expires or when enough packets
lck_mtx_unlock(bpf_mlock);
return (ENXIO);
}
+
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1) {
+ lck_mtx_unlock(bpf_mlock);
+ return (ENXIO);
+ }
+
if (error == EINTR || error == ERESTART) {
+ if (d->bd_hbuf) {
+ /*
+ * Because we msleep, the hold buffer might
+ * be filled when we wake up. Avoid rotating
+ * in this case.
+ */
+ break;
+ }
+ if (d->bd_slen) {
+ /*
+ * Sometimes we may be interrupted often and
+ * the sleep above will not timeout.
+ * Regardless, we should rotate the buffers
+ * if there's any new data pending and
+ * return it.
+ */
+ ROTATE_BUFFERS(d);
+ break;
+ }
lck_mtx_unlock(bpf_mlock);
return (error);
}
* At this point, we know we have something in the hold slot.
*/
+ /*
+ * Set the hold buffer read. So we do not
+ * rotate the buffers until the hold buffer
+ * read is complete. Also to avoid issues resulting
+ * from page faults during disk sleep (<rdar://problem/13436396>).
+ */
+ d->bd_hbuf_read = 1;
+ hbuf = d->bd_hbuf;
+ hbuf_len = d->bd_hlen;
+ flags = d->bd_flags;
+ lck_mtx_unlock(bpf_mlock);
+
+#ifdef __APPLE__
/*
* Before we move data to userland, we fill out the extended
* header fields.
*/
- if (d->bd_extendedhdr) {
+ if (flags & BPF_EXTENDED_HDR) {
char *p;
- p = d->bd_hbuf;
- while (p < d->bd_hbuf + d->bd_hlen) {
+ p = hbuf;
+ while (p < hbuf + hbuf_len) {
struct bpf_hdr_ext *ehp;
- struct inpcb *inp;
- uint32_t flowhash;
- pid_t pid;
+ uint32_t flowid;
+ struct so_procinfo soprocinfo;
+ int found = 0;
ehp = (struct bpf_hdr_ext *)(void *)p;
- if ((flowhash = ehp->bh_flowhash)) {
- if (ehp->bh_flags & BPF_HDR_EXT_FLAGS_TCP)
- inp = bpf_findinpcb(&tcbinfo, flowhash);
- else
- inp = bpf_findinpcb(&udbinfo, flowhash);
- if (inp) {
- socket_lock(inp->inp_socket, 0);
- pid = inp->inp_socket->last_pid;
- in_pcb_checkstate(inp, WNT_RELEASE, 1);
- socket_unlock(inp->inp_socket, 0);
- ehp->bh_pid = pid;
- proc_name(pid, ehp->bh_comm, MAXCOMLEN);
+ if ((flowid = ehp->bh_flowid)) {
+ if (ehp->bh_proto == IPPROTO_TCP)
+ found = inp_findinpcb_procinfo(&tcbinfo,
+ flowid, &soprocinfo);
+ else if (ehp->bh_proto == IPPROTO_UDP)
+ found = inp_findinpcb_procinfo(&udbinfo,
+ flowid, &soprocinfo);
+ if (found == 1) {
+ ehp->bh_pid = soprocinfo.spi_pid;
+ proc_name(ehp->bh_pid, ehp->bh_comm, MAXCOMLEN);
+ }
+ ehp->bh_flowid = 0;
+ }
+ if (flags & BPF_FINALIZE_PKTAP) {
+ struct pktap_header *pktaphdr;
+
+ pktaphdr = (struct pktap_header *)(void *)
+ (p + BPF_WORDALIGN(ehp->bh_hdrlen));
+
+ if (pktaphdr->pth_flags & PTH_FLAG_DELAY_PKTAP)
+ pktap_finalize_proc_info(pktaphdr);
+
+ if (pktaphdr->pth_flags & PTH_FLAG_TSTAMP) {
+ ehp->bh_tstamp.tv_sec =
+ pktaphdr->pth_tstamp.tv_sec;
+ ehp->bh_tstamp.tv_usec =
+ pktaphdr->pth_tstamp.tv_usec;
}
- ehp->bh_flowhash = 0;
}
p += BPF_WORDALIGN(ehp->bh_hdrlen + ehp->bh_caplen);
}
+ } else if (flags & BPF_FINALIZE_PKTAP) {
+ char *p;
+
+ p = hbuf;
+ while (p < hbuf + hbuf_len) {
+ struct bpf_hdr *hp;
+ struct pktap_header *pktaphdr;
+
+ hp = (struct bpf_hdr *)(void *)p;
+ pktaphdr = (struct pktap_header *)(void *)
+ (p + BPF_WORDALIGN(hp->bh_hdrlen));
+
+ if (pktaphdr->pth_flags & PTH_FLAG_DELAY_PKTAP)
+ pktap_finalize_proc_info(pktaphdr);
+
+ if (pktaphdr->pth_flags & PTH_FLAG_TSTAMP) {
+ hp->bh_tstamp.tv_sec =
+ pktaphdr->pth_tstamp.tv_sec;
+ hp->bh_tstamp.tv_usec =
+ pktaphdr->pth_tstamp.tv_usec;
+ }
+
+ p += BPF_WORDALIGN(hp->bh_hdrlen + hp->bh_caplen);
+ }
}
+#endif
+
/*
* Move data from hold buffer into user space.
* We know the entire buffer is transferred since
* we checked above that the read buffer is bpf_bufsize bytes.
*/
- error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
-
+ error = UIOMOVE(hbuf, hbuf_len, UIO_READ, uio);
+
+ lck_mtx_lock(bpf_mlock);
+ /*
+ * Make sure device is still opened
+ */
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1) {
+ lck_mtx_unlock(bpf_mlock);
+ return (ENXIO);
+ }
+
+ d->bd_hbuf_read = 0;
d->bd_fbuf = d->bd_hbuf;
d->bd_hbuf = NULL;
d->bd_hlen = 0;
+ wakeup((caddr_t)d);
lck_mtx_unlock(bpf_mlock);
return (error);
+
}
int error;
char dst_buf[SOCKADDR_HDR_LEN + MAX_DATALINK_HDR_LEN];
int datlen = 0;
- int bif_dlt;
- int bd_hdrcmplt;
+ int bif_dlt;
+ int bd_hdrcmplt;
lck_mtx_lock(bpf_mlock);
static void
reset_d(struct bpf_d *d)
{
+ if (d->bd_hbuf_read)
+ panic("resetting buffers during read");
+
if (d->bd_hbuf) {
/* Free the hold buffer. */
d->bd_fbuf = d->bd_hbuf;
struct proc *p)
{
struct bpf_d *d;
- int error = 0, int_arg;
+ int error = 0;
+ u_int int_arg;
struct ifreq ifr;
lck_mtx_lock(bpf_mlock);
int n;
n = d->bd_slen;
- if (d->bd_hbuf)
+ if (d->bd_hbuf && d->bd_hbuf_read == 0)
n += d->bd_hlen;
bcopy(&n, addr, sizeof (n));
/*
* Set link layer read filter.
*/
- case BIOCSETF32: { /* struct bpf_program32 */
+ case BIOCSETF32:
+ case BIOCSETFNR32: { /* struct bpf_program32 */
struct bpf_program32 prg32;
bcopy(addr, &prg32, sizeof (prg32));
error = bpf_setf(d, prg32.bf_len,
- CAST_USER_ADDR_T(prg32.bf_insns));
+ CAST_USER_ADDR_T(prg32.bf_insns), dev, cmd);
break;
}
- case BIOCSETF64: { /* struct bpf_program64 */
+ case BIOCSETF64:
+ case BIOCSETFNR64: { /* struct bpf_program64 */
struct bpf_program64 prg64;
bcopy(addr, &prg64, sizeof (prg64));
- error = bpf_setf(d, prg64.bf_len, prg64.bf_insns);
+ error = bpf_setf(d, prg64.bf_len, prg64.bf_insns, dev, cmd);
break;
}
* Flush read packet buffer.
*/
case BIOCFLUSH:
+ while (d->bd_hbuf_read) {
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+ }
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1)
+ return (ENXIO);
+
reset_d(d);
break;
u_int dlt;
bcopy(addr, &dlt, sizeof (dlt));
- error = bpf_setdlt(d, dlt);
+ error = bpf_setdlt(d, dlt, dev);
}
break;
struct ifnet *const ifp = d->bd_bif->bif_ifp;
snprintf(((struct ifreq *)(void *)addr)->ifr_name,
- sizeof (ifr.ifr_name), "%s%d", ifp->if_name,
- ifp->if_unit);
+ sizeof (ifr.ifr_name), "%s", if_name(ifp));
}
break;
if (ifp == NULL)
error = ENXIO;
else
- error = bpf_setif(d, ifp, 0);
+ error = bpf_setif(d, ifp, 0, dev);
break;
}
/*
* Set read timeout.
*/
- case BIOCSRTIMEOUT32: { /* struct user32_timeval */
+ case BIOCSRTIMEOUT32: { /* struct user32_timeval */
struct user32_timeval _tv;
struct timeval tv;
break;
}
- case BIOCSRTIMEOUT64: { /* struct user64_timeval */
+ case BIOCSRTIMEOUT64: { /* struct user64_timeval */
struct user64_timeval _tv;
struct timeval tv;
break;
}
- /*
+ /*
* Get read timeout.
*/
case BIOCGRTIMEOUT32: { /* struct user32_timeval */
case BIOCGRSIG: /* u_int */
bcopy(&d->bd_sig, addr, sizeof (u_int));
break;
- case BIOCSEXTHDR:
- bcopy(addr, &d->bd_extendedhdr, sizeof (u_int));
+#ifdef __APPLE__
+ case BIOCSEXTHDR: /* u_int */
+ bcopy(addr, &int_arg, sizeof (int_arg));
+ if (int_arg)
+ d->bd_flags |= BPF_EXTENDED_HDR;
+ else
+ d->bd_flags &= ~BPF_EXTENDED_HDR;
+ break;
+
+ case BIOCGIFATTACHCOUNT: { /* struct ifreq */
+ ifnet_t ifp;
+ struct bpf_if *bp;
+
+ bcopy(addr, &ifr, sizeof (ifr));
+ ifr.ifr_name[IFNAMSIZ - 1] = '\0';
+ ifp = ifunit(ifr.ifr_name);
+ if (ifp == NULL) {
+ error = ENXIO;
+ break;
+ }
+ ifr.ifr_intval = 0;
+ for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
+ struct bpf_d *bpf_d;
+
+ if (bp->bif_ifp == NULL || bp->bif_ifp != ifp)
+ continue;
+ for (bpf_d = bp->bif_dlist; bpf_d; bpf_d = bpf_d->bd_next) {
+ ifr.ifr_intval += 1;
+ }
+ }
+ bcopy(&ifr, addr, sizeof (ifr));
break;
}
+ case BIOCGWANTPKTAP: /* u_int */
+ int_arg = d->bd_flags & BPF_WANT_PKTAP ? 1 : 0;
+ bcopy(&int_arg, addr, sizeof (int_arg));
+ break;
+
+ case BIOCSWANTPKTAP: /* u_int */
+ bcopy(addr, &int_arg, sizeof (int_arg));
+ if (int_arg)
+ d->bd_flags |= BPF_WANT_PKTAP;
+ else
+ d->bd_flags &= ~BPF_WANT_PKTAP;
+ break;
+#endif
+ }
lck_mtx_unlock(bpf_mlock);
* free it and replace it. Returns EINVAL for bogus requests.
*/
static int
-bpf_setf(struct bpf_d *d, u_int bf_len, user_addr_t bf_insns)
+bpf_setf(struct bpf_d *d, u_int bf_len, user_addr_t bf_insns, dev_t dev, u_long cmd)
{
struct bpf_insn *fcode, *old;
u_int flen, size;
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1)
+ return (ENXIO);
+
old = d->bd_filter;
if (bf_insns == USER_ADDR_NULL) {
if (bf_len != 0)
if (copyin(bf_insns, (caddr_t)fcode, size) == 0 &&
bpf_validate(fcode, (int)flen)) {
d->bd_filter = fcode;
- reset_d(d);
+
+ if (cmd == BIOCSETF32 || cmd == BIOCSETF64)
+ reset_d(d);
+
if (old != 0)
FREE((caddr_t)old, M_DEVBUF);
* Return an errno or 0.
*/
static int
-bpf_setif(struct bpf_d *d, ifnet_t theywant, u_int32_t dlt)
+bpf_setif(struct bpf_d *d, ifnet_t theywant, u_int32_t dlt, dev_t dev)
{
struct bpf_if *bp;
int error;
-
+
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1)
+ return (ENXIO);
+
/*
* Look through attached interfaces for the named one.
*/
if (ifp == 0 || ifp != theywant || (dlt != 0 && dlt != bp->bif_dlt))
continue;
+ /*
+ * If the process knows how to deal with DLT_PKTAP, use it
+ * by default
+ */
+ if (dlt == 0 && bp->bif_dlt == DLT_PKTAP &&
+ !(d->bd_flags & BPF_WANT_PKTAP))
+ continue;
/*
* We found the requested interface.
* Allocate the packet buffers if we need to.
ifp = d->bd_bif->bif_ifp;
n = 0;
error = 0;
+
for (bp = bpf_iflist; bp; bp = bp->bif_next) {
if (bp->bif_ifp != ifp)
continue;
+ /*
+ * Return DLT_PKTAP only to processes that know how to handle it
+ */
+ if (bp->bif_dlt == DLT_PKTAP && !(d->bd_flags & BPF_WANT_PKTAP))
+ continue;
if (dlist != USER_ADDR_NULL) {
if (n >= bfl.bfl_len) {
return (ENOMEM);
* Set the data link type of a BPF instance.
*/
static int
-bpf_setdlt(struct bpf_d *d, uint32_t dlt)
-
-
+bpf_setdlt(struct bpf_d *d, uint32_t dlt, dev_t dev)
{
int error, opromisc;
struct ifnet *ifp;
if (d->bd_bif->bif_dlt == dlt)
return (0);
+
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1)
+ return (ENXIO);
+
ifp = d->bd_bif->bif_ifp;
for (bp = bpf_iflist; bp; bp = bp->bif_next) {
if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
return (ENXIO);
}
+ while (d->bd_hbuf_read)
+ msleep((caddr_t)d, bpf_mlock, PRINET, "bpf_reading", NULL);
+
+ d = bpf_dtab[minor(dev)];
+ if (d == 0 || d == (void *)1) {
+ lck_mtx_unlock(bpf_mlock);
+ return (ENXIO);
+ }
+
switch (which) {
case FREAD:
if (d->bd_hlen != 0 ||
if (hint == 0)
lck_mtx_lock(bpf_mlock);
-
+
if (d->bd_immediate) {
/*
* If there's data in the hold buffer, it's the
* If there's no data in either buffer, we're not
* ready to read.
*/
- kn->kn_data = (d->bd_hlen == 0 ? d->bd_slen : d->bd_hlen);
+ kn->kn_data = ((d->bd_hlen == 0 || d->bd_hbuf_read)
+ ? d->bd_slen : d->bd_hlen);
int64_t lowwat = 1;
if (kn->kn_sfflags & NOTE_LOWAT)
{
* no data in the hold buffer and the timer hasn't
* expired, we're not ready to read.
*/
- kn->kn_data = (d->bd_hlen == 0 && d->bd_state == BPF_TIMED_OUT ?
+ kn->kn_data = ((d->bd_hlen == 0 || d->bd_hbuf_read) && d->bd_state == BPF_TIMED_OUT ?
d->bd_slen : d->bd_hlen);
ready = (kn->kn_data > 0);
}
int hdrlen, caplen;
int do_wakeup = 0;
u_char *payload;
+ struct timeval tv;
+ struct m_tag *mt = NULL;
+ struct bpf_mtag *bt = NULL;
- hdrlen = d->bd_extendedhdr ? d->bd_bif->bif_exthdrlen :
+ hdrlen = (d->bd_flags & BPF_EXTENDED_HDR) ? d->bd_bif->bif_exthdrlen :
d->bd_bif->bif_hdrlen;
/*
* Figure out how many bytes to move. If the packet is
/*
* Append the bpf header.
*/
- struct timeval tv;
microtime(&tv);
- if (d->bd_extendedhdr) {
+ if (d->bd_flags & BPF_EXTENDED_HDR) {
ehp = (struct bpf_hdr_ext *)(void *)(d->bd_sbuf + curlen);
memset(ehp, 0, sizeof(*ehp));
ehp->bh_tstamp.tv_sec = tv.tv_sec;
ehp->bh_datalen = pktlen;
ehp->bh_hdrlen = hdrlen;
ehp->bh_caplen = totlen - hdrlen;
- if (outbound) {
- if (m->m_pkthdr.m_fhflags & PF_TAG_FLOWHASH)
- ehp->bh_flowhash = m->m_pkthdr.m_flowhash;
- ehp->bh_svc = so_svc2tc(m->m_pkthdr.svc);
+ mt = m_tag_locate(m, bpf_mtag_id, 0, NULL);
+ if (mt && mt->m_tag_len >= sizeof(*bt)) {
+ bt = (struct bpf_mtag *)(mt + 1);
+ ehp->bh_pid = bt->bt_pid;
+ strlcpy(ehp->bh_comm, bt->bt_comm,
+ sizeof(ehp->bh_comm));
+ ehp->bh_svc = so_svc2tc(bt->bt_svc);
+ if (bt->bt_direction == BPF_MTAG_DIR_OUT)
+ ehp->bh_flags |= BPF_HDR_EXT_FLAGS_DIR_OUT;
+ else
+ ehp->bh_flags |= BPF_HDR_EXT_FLAGS_DIR_IN;
+ m_tag_delete(m, mt);
+ } else if (outbound) {
+ /* only do lookups on non-raw INPCB */
+ if ((m->m_pkthdr.pkt_flags & (PKTF_FLOW_ID|
+ PKTF_FLOW_LOCALSRC|PKTF_FLOW_RAWSOCK)) ==
+ (PKTF_FLOW_ID|PKTF_FLOW_LOCALSRC) &&
+ m->m_pkthdr.pkt_flowsrc == FLOWSRC_INPCB) {
+ ehp->bh_flowid = m->m_pkthdr.pkt_flowid;
+ ehp->bh_proto = m->m_pkthdr.pkt_proto;
+ }
+ ehp->bh_svc = so_svc2tc(m->m_pkthdr.pkt_svc);
ehp->bh_flags |= BPF_HDR_EXT_FLAGS_DIR_OUT;
- if (m->m_pkthdr.m_fhflags & PF_TAG_TCP)
- ehp->bh_flags |= BPF_HDR_EXT_FLAGS_TCP;
- } else
+ } else
ehp->bh_flags |= BPF_HDR_EXT_FLAGS_DIR_IN;
payload = (u_char *)ehp + hdrlen;
caplen = ehp->bh_caplen;
* been detached from its interface and it yet hasn't been marked
* free.
*/
+ if (d->bd_hbuf_read)
+ panic("bpf buffer freed during read");
+
if (d->bd_sbuf != 0) {
FREE(d->bd_sbuf, M_DEVBUF);
- if (d->bd_hbuf != 0)
+ if (d->bd_hbuf != 0)
FREE(d->bd_hbuf, M_DEVBUF);
if (d->bd_fbuf != 0)
FREE(d->bd_fbuf, M_DEVBUF);
}
if (bp_temp != NULL) {
- printf("bpfattach - %s%d with dlt %d is already attached\n",
- ifp->if_name, ifp->if_unit, dlt);
+ printf("bpfattach - %s with dlt %d is already attached\n",
+ if_name(ifp), dlt);
FREE(bp_new, M_DEVBUF);
lck_mtx_unlock(bpf_mlock);
return EEXIST;
#ifndef __APPLE__
if (bootverbose)
- printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
+ printf("bpf: %s attached\n", if_name(ifp));
#endif
return 0;
bpfdetach(struct ifnet *ifp)
{
struct bpf_if *bp, *bp_prev, *bp_next;
- struct bpf_if *bp_free = NULL;
+ struct bpf_if *bp_free_list = NULL;
struct bpf_d *d;
-
lck_mtx_lock(bpf_mlock);
- /* Locate BPF interface information */
+ /*
+ * Build the list of devices attached to that interface
+ * that we need to free while keeping the lock to maintain
+ * the integrity of the interface list
+ */
bp_prev = NULL;
for (bp = bpf_iflist; bp != NULL; bp = bp_next) {
bp_next = bp->bif_next;
+
if (ifp != bp->bif_ifp) {
bp_prev = bp;
continue;
}
-
+ /* Unlink from the interface list */
+ if (bp_prev)
+ bp_prev->bif_next = bp->bif_next;
+ else
+ bpf_iflist = bp->bif_next;
+
+ /* Add to the list to be freed */
+ bp->bif_next = bp_free_list;
+ bp_free_list = bp;
+ }
+
+ /*
+ * Detach the bpf devices attached to the interface
+ * Now we do not care if we lose the bpf_mlock in bpf_detachd
+ */
+ for (bp = bp_free_list; bp != NULL; bp = bp->bif_next) {
while ((d = bp->bif_dlist) != NULL) {
bpf_detachd(d);
bpf_wakeup(d);
}
-
- if (bp_prev) {
- bp_prev->bif_next = bp->bif_next;
- } else {
- bpf_iflist = bp->bif_next;
- }
-
- bp->bif_next = bp_free;
- bp_free = bp;
-
ifnet_release(ifp);
}
lck_mtx_unlock(bpf_mlock);
- FREE(bp, M_DEVBUF);
-
+ /*
+ * Free the list
+ */
+ while ((bp = bp_free_list) != NULL) {
+ bp_free_list = bp->bif_next;
+ FREE(bp, M_DEVBUF);
+ }
}
void
if (bpf_devsw_installed == 0) {
bpf_devsw_installed = 1;
-
- bpf_mlock_grp_attr = lck_grp_attr_alloc_init();
-
- bpf_mlock_grp = lck_grp_alloc_init("bpf", bpf_mlock_grp_attr);
-
- bpf_mlock_attr = lck_attr_alloc_init();
-
- lck_mtx_init(bpf_mlock, bpf_mlock_grp, bpf_mlock_attr);
-
+ bpf_mlock_grp_attr = lck_grp_attr_alloc_init();
+ bpf_mlock_grp = lck_grp_alloc_init("bpf", bpf_mlock_grp_attr);
+ bpf_mlock_attr = lck_attr_alloc_init();
+ lck_mtx_init(bpf_mlock, bpf_mlock_grp, bpf_mlock_attr);
maj = cdevsw_add(CDEV_MAJOR, &bpf_cdevsw);
if (maj == -1) {
if (bpf_mlock_attr)
for (i = 0 ; i < NBPFILTER; i++)
bpf_make_dev_t(maj);
+
+ VERIFY(mbuf_tag_id_find(BPF_CONTROL_NAME, &bpf_mtag_id) == 0);
}
#else
cdevsw_add(&bpf_cdevsw);
projects / apple / xnu.git / blobdiff