Index: head/sys/net/bpf.c =================================================================== --- head/sys/net/bpf.c (revision 292602) +++ head/sys/net/bpf.c (revision 292603) @@ -1,2947 +1,2947 @@ /*- * Copyright (c) 1990, 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from the Stanford/CMU enet packet filter, * (net/enet.c) distributed as part of 4.3BSD, and code contributed * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence * Berkeley Laboratory. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)bpf.c 8.4 (Berkeley) 1/9/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_bpf.h" #include "opt_compat.h" #include "opt_netgraph.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef BPF_JITTER #include #endif #include #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_BPF, "BPF", "BPF data"); struct bpf_if { #define bif_next bif_ext.bif_next #define bif_dlist bif_ext.bif_dlist struct bpf_if_ext bif_ext; /* public members */ u_int bif_dlt; /* link layer type */ u_int bif_hdrlen; /* length of link header */ struct ifnet *bif_ifp; /* corresponding interface */ struct rwlock bif_lock; /* interface lock */ LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */ int bif_flags; /* Interface flags */ }; CTASSERT(offsetof(struct bpf_if, bif_ext) == 0); #if defined(DEV_BPF) || defined(NETGRAPH_BPF) #define PRINET 26 /* interruptible */ #define SIZEOF_BPF_HDR(type) \ (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) #ifdef COMPAT_FREEBSD32 #include #include #define BPF_ALIGNMENT32 sizeof(int32_t) #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1)) #ifndef BURN_BRIDGES /* * 32-bit version of structure prepended to each packet. We use this header * instead of the standard one for 32-bit streams. We mark the a stream as * 32-bit the first time we see a 32-bit compat ioctl request. */ struct bpf_hdr32 { struct timeval32 bh_tstamp; /* time stamp */ uint32_t bh_caplen; /* length of captured portion */ uint32_t bh_datalen; /* original length of packet */ uint16_t bh_hdrlen; /* length of bpf header (this struct plus alignment padding) */ }; #endif struct bpf_program32 { u_int bf_len; uint32_t bf_insns; }; struct bpf_dltlist32 { u_int bfl_len; u_int bfl_list; }; #define BIOCSETF32 _IOW('B', 103, struct bpf_program32) #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) #endif /* * bpf_iflist is a list of BPF interface structures, each corresponding to a * specific DLT. The same network interface might have several BPF interface * structures registered by different layers in the stack (i.e., 802.11 * frames, ethernet frames, etc). */ static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist; static struct mtx bpf_mtx; /* bpf global lock */ static int bpf_bpfd_cnt; static void bpf_attachd(struct bpf_d *, struct bpf_if *); static void bpf_detachd(struct bpf_d *); static void bpf_detachd_locked(struct bpf_d *); static void bpf_freed(struct bpf_d *); static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, struct sockaddr *, int *, struct bpf_insn *); static int bpf_setif(struct bpf_d *, struct ifreq *); static void bpf_timed_out(void *); static __inline void bpf_wakeup(struct bpf_d *); static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), struct bintime *); static void reset_d(struct bpf_d *); static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); static int bpf_setdlt(struct bpf_d *, u_int); static void filt_bpfdetach(struct knote *); static int filt_bpfread(struct knote *, long); static void bpf_drvinit(void *); static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); int bpf_maxinsns = BPF_MAXINSNS; SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, &bpf_maxinsns, 0, "Maximum bpf program instructions"); static int bpf_zerocopy_enable = 0; SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, bpf_stats_sysctl, "bpf statistics portal"); static VNET_DEFINE(int, bpf_optimize_writers) = 0; #define V_bpf_optimize_writers VNET(bpf_optimize_writers) SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0, "Do not send packets until BPF program is set"); static d_open_t bpfopen; static d_read_t bpfread; static d_write_t bpfwrite; static d_ioctl_t bpfioctl; static d_poll_t bpfpoll; static d_kqfilter_t bpfkqfilter; static struct cdevsw bpf_cdevsw = { .d_version = D_VERSION, .d_open = bpfopen, .d_read = bpfread, .d_write = bpfwrite, .d_ioctl = bpfioctl, .d_poll = bpfpoll, .d_name = "bpf", .d_kqfilter = bpfkqfilter, }; static struct filterops bpfread_filtops = { .f_isfd = 1, .f_detach = filt_bpfdetach, .f_event = filt_bpfread, }; eventhandler_tag bpf_ifdetach_cookie = NULL; /* * LOCKING MODEL USED BY BPF: * Locks: * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal, * some global counters and every bpf_if reference. * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters. * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields * used by bpf_mtap code. * * Lock order: * * Global lock, interface lock, descriptor lock * * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2] * working model. In many places (like bpf_detachd) we start with BPF descriptor * (and we need to at least rlock it to get reliable interface pointer). This * gives us potential LOR. As a result, we use global lock to protect from bpf_if * change in every such place. * * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and * 3) descriptor main wlock. * Reading bd_bif can be protected by any of these locks, typically global lock. * * Changing read/write BPF filter is protected by the same three locks, * the same applies for reading. * * Sleeping in global lock is not allowed due to bpfdetach() using it. */ /* * Wrapper functions for various buffering methods. If the set of buffer * modes expands, we will probably want to introduce a switch data structure * similar to protosw, et. */ static void bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, u_int len) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_append_bytes(d, buf, offset, src, len)); case BPF_BUFMODE_ZBUF: d->bd_zcopy++; return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); default: panic("bpf_buf_append_bytes"); } } static void bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, u_int len) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); case BPF_BUFMODE_ZBUF: d->bd_zcopy++; return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); default: panic("bpf_buf_append_mbuf"); } } /* * This function gets called when the free buffer is re-assigned. */ static void bpf_buf_reclaimed(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return; case BPF_BUFMODE_ZBUF: bpf_zerocopy_buf_reclaimed(d); return; default: panic("bpf_buf_reclaimed"); } } /* * If the buffer mechanism has a way to decide that a held buffer can be made * free, then it is exposed via the bpf_canfreebuf() interface. (1) is * returned if the buffer can be discarded, (0) is returned if it cannot. */ static int bpf_canfreebuf(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_canfreebuf(d)); } return (0); } /* * Allow the buffer model to indicate that the current store buffer is * immutable, regardless of the appearance of space. Return (1) if the * buffer is writable, and (0) if not. */ static int bpf_canwritebuf(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_canwritebuf(d)); } return (1); } /* * Notify buffer model that an attempt to write to the store buffer has * resulted in a dropped packet, in which case the buffer may be considered * full. */ static void bpf_buffull(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: bpf_zerocopy_buffull(d); break; } } /* * Notify the buffer model that a buffer has moved into the hold position. */ void bpf_bufheld(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); switch (d->bd_bufmode) { case BPF_BUFMODE_ZBUF: bpf_zerocopy_bufheld(d); break; } } static void bpf_free(struct bpf_d *d) { switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: return (bpf_buffer_free(d)); case BPF_BUFMODE_ZBUF: return (bpf_zerocopy_free(d)); default: panic("bpf_buf_free"); } } static int bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) { if (d->bd_bufmode != BPF_BUFMODE_BUFFER) return (EOPNOTSUPP); return (bpf_buffer_uiomove(d, buf, len, uio)); } static int bpf_ioctl_sblen(struct bpf_d *d, u_int *i) { if (d->bd_bufmode != BPF_BUFMODE_BUFFER) return (EOPNOTSUPP); return (bpf_buffer_ioctl_sblen(d, i)); } static int bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_getzmax(td, d, i)); } static int bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); } static int bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) { if (d->bd_bufmode != BPF_BUFMODE_ZBUF) return (EOPNOTSUPP); return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); } /* * General BPF functions. */ static int bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter) { const struct ieee80211_bpf_params *p; struct ether_header *eh; struct mbuf *m; int error; int len; int hlen; int slen; /* * Build a sockaddr based on the data link layer type. * We do this at this level because the ethernet header * is copied directly into the data field of the sockaddr. * In the case of SLIP, there is no header and the packet * is forwarded as is. * Also, we are careful to leave room at the front of the mbuf * for the link level header. */ switch (linktype) { case DLT_SLIP: sockp->sa_family = AF_INET; hlen = 0; break; case DLT_EN10MB: sockp->sa_family = AF_UNSPEC; /* XXX Would MAXLINKHDR be better? */ hlen = ETHER_HDR_LEN; break; case DLT_FDDI: sockp->sa_family = AF_IMPLINK; hlen = 0; break; case DLT_RAW: sockp->sa_family = AF_UNSPEC; hlen = 0; break; case DLT_NULL: /* * null interface types require a 4 byte pseudo header which * corresponds to the address family of the packet. */ sockp->sa_family = AF_UNSPEC; hlen = 4; break; case DLT_ATM_RFC1483: /* * en atm driver requires 4-byte atm pseudo header. * though it isn't standard, vpi:vci needs to be * specified anyway. */ sockp->sa_family = AF_UNSPEC; hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ break; case DLT_PPP: sockp->sa_family = AF_UNSPEC; hlen = 4; /* This should match PPP_HDRLEN */ break; case DLT_IEEE802_11: /* IEEE 802.11 wireless */ sockp->sa_family = AF_IEEE80211; hlen = 0; break; case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ sockp->sa_family = AF_IEEE80211; sockp->sa_len = 12; /* XXX != 0 */ hlen = sizeof(struct ieee80211_bpf_params); break; default: return (EIO); } len = uio->uio_resid; if (len < hlen || len - hlen > ifp->if_mtu) return (EMSGSIZE); m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); if (m == NULL) return (EIO); m->m_pkthdr.len = m->m_len = len; *mp = m; error = uiomove(mtod(m, u_char *), len, uio); if (error) goto bad; slen = bpf_filter(wfilter, mtod(m, u_char *), len, len); if (slen == 0) { error = EPERM; goto bad; } /* Check for multicast destination */ switch (linktype) { case DLT_EN10MB: eh = mtod(m, struct ether_header *); if (ETHER_IS_MULTICAST(eh->ether_dhost)) { if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, ETHER_ADDR_LEN) == 0) m->m_flags |= M_BCAST; else m->m_flags |= M_MCAST; } break; } /* * Make room for link header, and copy it to sockaddr */ if (hlen != 0) { if (sockp->sa_family == AF_IEEE80211) { /* * Collect true length from the parameter header * NB: sockp is known to be zero'd so if we do a * short copy unspecified parameters will be * zero. * NB: packet may not be aligned after stripping * bpf params * XXX check ibp_vers */ p = mtod(m, const struct ieee80211_bpf_params *); hlen = p->ibp_len; if (hlen > sizeof(sockp->sa_data)) { error = EINVAL; goto bad; } } bcopy(mtod(m, const void *), sockp->sa_data, hlen); } *hdrlen = hlen; return (0); bad: m_freem(m); return (error); } /* * Attach file to the bpf interface, i.e. make d listen on bp. */ static void bpf_attachd(struct bpf_d *d, struct bpf_if *bp) { int op_w; BPF_LOCK_ASSERT(); /* * Save sysctl value to protect from sysctl change * between reads */ op_w = V_bpf_optimize_writers || d->bd_writer; if (d->bd_bif != NULL) bpf_detachd_locked(d); /* * Point d at bp, and add d to the interface's list. * Since there are many applications using BPF for * sending raw packets only (dhcpd, cdpd are good examples) * we can delay adding d to the list of active listeners until * some filter is configured. */ BPFIF_WLOCK(bp); BPFD_LOCK(d); d->bd_bif = bp; if (op_w != 0) { /* Add to writers-only list */ LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); /* * We decrement bd_writer on every filter set operation. * First BIOCSETF is done by pcap_open_live() to set up * snap length. After that appliation usually sets its own filter */ d->bd_writer = 2; } else LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); bpf_bpfd_cnt++; CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); if (op_w == 0) EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); } /* * Check if we need to upgrade our descriptor @d from write-only mode. */ static int bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen) { int is_snap, need_upgrade; /* * Check if we've already upgraded or new filter is empty. */ if (d->bd_writer == 0 || fcode == NULL) return (0); need_upgrade = 0; /* * Check if cmd looks like snaplen setting from * pcap_bpf.c:pcap_open_live(). * Note we're not checking .k value here: * while pcap_open_live() definitely sets to to non-zero value, * we'd prefer to treat k=0 (deny ALL) case the same way: e.g. * do not consider upgrading immediately */ if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K)) is_snap = 1; else is_snap = 0; if (is_snap == 0) { /* * We're setting first filter and it doesn't look like * setting snaplen. We're probably using bpf directly. * Upgrade immediately. */ need_upgrade = 1; } else { /* * Do not require upgrade by first BIOCSETF * (used to set snaplen) by pcap_open_live(). */ if (--d->bd_writer == 0) { /* * First snaplen filter has already * been set. This is probably catch-all * filter */ need_upgrade = 1; } } CTR5(KTR_NET, "%s: filter function set by pid %d, " "bd_writer counter %d, snap %d upgrade %d", __func__, d->bd_pid, d->bd_writer, is_snap, need_upgrade); return (need_upgrade); } /* * Add d to the list of active bp filters. * Requires bpf_attachd() to be called before. */ static void bpf_upgraded(struct bpf_d *d) { struct bpf_if *bp; BPF_LOCK_ASSERT(); bp = d->bd_bif; /* * Filter can be set several times without specifying interface. * Mark d as reader and exit. */ if (bp == NULL) { BPFD_LOCK(d); d->bd_writer = 0; BPFD_UNLOCK(d); return; } BPFIF_WLOCK(bp); BPFD_LOCK(d); /* Remove from writers-only list */ LIST_REMOVE(d, bd_next); LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); /* Mark d as reader */ d->bd_writer = 0; BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); } /* * Detach a file from its interface. */ static void bpf_detachd(struct bpf_d *d) { BPF_LOCK(); bpf_detachd_locked(d); BPF_UNLOCK(); } static void bpf_detachd_locked(struct bpf_d *d) { int error; struct bpf_if *bp; struct ifnet *ifp; CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); BPF_LOCK_ASSERT(); /* Check if descriptor is attached */ if ((bp = d->bd_bif) == NULL) return; BPFIF_WLOCK(bp); BPFD_LOCK(d); /* Save bd_writer value */ error = d->bd_writer; /* * Remove d from the interface's descriptor list. */ LIST_REMOVE(d, bd_next); ifp = bp->bif_ifp; d->bd_bif = NULL; BPFD_UNLOCK(d); BPFIF_WUNLOCK(bp); bpf_bpfd_cnt--; /* Call event handler iff d is attached */ if (error == 0) EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); /* * Check if this descriptor had requested promiscuous mode. * If so, turn it off. */ if (d->bd_promisc) { d->bd_promisc = 0; CURVNET_SET(ifp->if_vnet); error = ifpromisc(ifp, 0); CURVNET_RESTORE(); if (error != 0 && error != ENXIO) { /* * ENXIO can happen if a pccard is unplugged * Something is really wrong if we were able to put * the driver into promiscuous mode, but can't * take it out. */ if_printf(bp->bif_ifp, "bpf_detach: ifpromisc failed (%d)\n", error); } } } /* * Close the descriptor by detaching it from its interface, * deallocating its buffers, and marking it free. */ static void bpf_dtor(void *data) { struct bpf_d *d = data; BPFD_LOCK(d); if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; BPFD_UNLOCK(d); funsetown(&d->bd_sigio); bpf_detachd(d); #ifdef MAC mac_bpfdesc_destroy(d); #endif /* MAC */ seldrain(&d->bd_sel); knlist_destroy(&d->bd_sel.si_note); callout_drain(&d->bd_callout); bpf_freed(d); free(d, M_BPF); } /* * Open ethernet device. Returns ENXIO for illegal minor device number, * EBUSY if file is open by another process. */ /* ARGSUSED */ static int bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) { struct bpf_d *d; int error; d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); error = devfs_set_cdevpriv(d, bpf_dtor); if (error != 0) { free(d, M_BPF); return (error); } /* * For historical reasons, perform a one-time initialization call to * the buffer routines, even though we're not yet committed to a * particular buffer method. */ bpf_buffer_init(d); if ((flags & FREAD) == 0) d->bd_writer = 2; d->bd_hbuf_in_use = 0; d->bd_bufmode = BPF_BUFMODE_BUFFER; d->bd_sig = SIGIO; d->bd_direction = BPF_D_INOUT; BPF_PID_REFRESH(d, td); #ifdef MAC mac_bpfdesc_init(d); mac_bpfdesc_create(td->td_ucred, d); #endif mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); return (0); } /* * bpfread - read next chunk of packets from buffers */ static int bpfread(struct cdev *dev, struct uio *uio, int ioflag) { struct bpf_d *d; int error; int non_block; int timed_out; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); /* * Restrict application to use a buffer the same size as * as kernel buffers. */ if (uio->uio_resid != d->bd_bufsize) return (EINVAL); non_block = ((ioflag & O_NONBLOCK) != 0); BPFD_LOCK(d); BPF_PID_REFRESH_CUR(d); if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { BPFD_UNLOCK(d); return (EOPNOTSUPP); } if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); timed_out = (d->bd_state == BPF_TIMED_OUT); d->bd_state = BPF_IDLE; while (d->bd_hbuf_in_use) { error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET|PCATCH, "bd_hbuf", 0); if (error != 0) { BPFD_UNLOCK(d); return (error); } } /* * If the hold buffer is empty, then do a timed sleep, which * ends when the timeout expires or when enough packets * have arrived to fill the store buffer. */ while (d->bd_hbuf == NULL) { if (d->bd_slen != 0) { /* * A packet(s) either arrived since the previous * read or arrived while we were asleep. */ if (d->bd_immediate || non_block || timed_out) { /* * Rotate the buffers and return what's here * if we are in immediate mode, non-blocking * flag is set, or this descriptor timed out. */ ROTATE_BUFFERS(d); break; } } /* * No data is available, check to see if the bpf device * is still pointed at a real interface. If not, return * ENXIO so that the userland process knows to rebind * it before using it again. */ if (d->bd_bif == NULL) { BPFD_UNLOCK(d); return (ENXIO); } if (non_block) { BPFD_UNLOCK(d); return (EWOULDBLOCK); } error = msleep(d, &d->bd_lock, PRINET|PCATCH, "bpf", d->bd_rtout); if (error == EINTR || error == ERESTART) { BPFD_UNLOCK(d); return (error); } if (error == EWOULDBLOCK) { /* * On a timeout, return what's in the buffer, * which may be nothing. If there is something * in the store buffer, we can rotate the buffers. */ if (d->bd_hbuf) /* * We filled up the buffer in between * getting the timeout and arriving * here, so we don't need to rotate. */ break; if (d->bd_slen == 0) { BPFD_UNLOCK(d); return (0); } ROTATE_BUFFERS(d); break; } } /* * At this point, we know we have something in the hold slot. */ d->bd_hbuf_in_use = 1; BPFD_UNLOCK(d); /* * 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. * * We do not have to worry about simultaneous reads because * we waited for sole access to the hold buffer above. */ error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); BPFD_LOCK(d); KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); d->bd_hbuf_in_use = 0; wakeup(&d->bd_hbuf_in_use); BPFD_UNLOCK(d); return (error); } /* * If there are processes sleeping on this descriptor, wake them up. */ static __inline void bpf_wakeup(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); if (d->bd_state == BPF_WAITING) { callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; } wakeup(d); if (d->bd_async && d->bd_sig && d->bd_sigio) pgsigio(&d->bd_sigio, d->bd_sig, 0); selwakeuppri(&d->bd_sel, PRINET); KNOTE_LOCKED(&d->bd_sel.si_note, 0); } static void bpf_timed_out(void *arg) { struct bpf_d *d = (struct bpf_d *)arg; BPFD_LOCK_ASSERT(d); if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) return; if (d->bd_state == BPF_WAITING) { d->bd_state = BPF_TIMED_OUT; if (d->bd_slen != 0) bpf_wakeup(d); } } static int bpf_ready(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); if (!bpf_canfreebuf(d) && d->bd_hlen != 0) return (1); if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && d->bd_slen != 0) return (1); return (0); } static int bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) { struct bpf_d *d; struct ifnet *ifp; struct mbuf *m, *mc; struct sockaddr dst; int error, hlen; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); BPF_PID_REFRESH_CUR(d); d->bd_wcount++; /* XXX: locking required */ if (d->bd_bif == NULL) { d->bd_wdcount++; return (ENXIO); } ifp = d->bd_bif->bif_ifp; if ((ifp->if_flags & IFF_UP) == 0) { d->bd_wdcount++; return (ENETDOWN); } if (uio->uio_resid == 0) { d->bd_wdcount++; return (0); } bzero(&dst, sizeof(dst)); m = NULL; hlen = 0; /* XXX: bpf_movein() can sleep */ error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, &m, &dst, &hlen, d->bd_wfilter); if (error) { d->bd_wdcount++; return (error); } d->bd_wfcount++; if (d->bd_hdrcmplt) dst.sa_family = pseudo_AF_HDRCMPLT; if (d->bd_feedback) { mc = m_dup(m, M_NOWAIT); if (mc != NULL) mc->m_pkthdr.rcvif = ifp; /* Set M_PROMISC for outgoing packets to be discarded. */ if (d->bd_direction == BPF_D_INOUT) m->m_flags |= M_PROMISC; } else mc = NULL; m->m_pkthdr.len -= hlen; m->m_len -= hlen; m->m_data += hlen; /* XXX */ CURVNET_SET(ifp->if_vnet); #ifdef MAC BPFD_LOCK(d); mac_bpfdesc_create_mbuf(d, m); if (mc != NULL) mac_bpfdesc_create_mbuf(d, mc); BPFD_UNLOCK(d); #endif error = (*ifp->if_output)(ifp, m, &dst, NULL); if (error) d->bd_wdcount++; if (mc != NULL) { if (error == 0) (*ifp->if_input)(ifp, mc); else m_freem(mc); } CURVNET_RESTORE(); return (error); } /* * Reset a descriptor by flushing its packet buffer and clearing the receive * and drop counts. This is doable for kernel-only buffers, but with * zero-copy buffers, we can't write to (or rotate) buffers that are * currently owned by userspace. It would be nice if we could encapsulate * this logic in the buffer code rather than here. */ static void reset_d(struct bpf_d *d) { BPFD_LOCK_ASSERT(d); while (d->bd_hbuf_in_use) mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, "bd_hbuf", 0); if ((d->bd_hbuf != NULL) && (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { /* Free the hold buffer. */ d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); } if (bpf_canwritebuf(d)) d->bd_slen = 0; d->bd_rcount = 0; d->bd_dcount = 0; d->bd_fcount = 0; d->bd_wcount = 0; d->bd_wfcount = 0; d->bd_wdcount = 0; d->bd_zcopy = 0; } /* * FIONREAD Check for read packet available. * BIOCGBLEN Get buffer len [for read()]. * BIOCSETF Set read filter. * BIOCSETFNR Set read filter without resetting descriptor. * BIOCSETWF Set write filter. * BIOCFLUSH Flush read packet buffer. * BIOCPROMISC Put interface into promiscuous mode. * BIOCGDLT Get link layer type. * BIOCGETIF Get interface name. * BIOCSETIF Set interface. * BIOCSRTIMEOUT Set read timeout. * BIOCGRTIMEOUT Get read timeout. * BIOCGSTATS Get packet stats. * BIOCIMMEDIATE Set immediate mode. * BIOCVERSION Get filter language version. * BIOCGHDRCMPLT Get "header already complete" flag * BIOCSHDRCMPLT Set "header already complete" flag * BIOCGDIRECTION Get packet direction flag * BIOCSDIRECTION Set packet direction flag * BIOCGTSTAMP Get time stamp format and resolution. * BIOCSTSTAMP Set time stamp format and resolution. * BIOCLOCK Set "locked" flag * BIOCFEEDBACK Set packet feedback mode. * BIOCSETZBUF Set current zero-copy buffer locations. * BIOCGETZMAX Get maximum zero-copy buffer size. * BIOCROTZBUF Force rotation of zero-copy buffer * BIOCSETBUFMODE Set buffer mode. * BIOCGETBUFMODE Get current buffer mode. */ /* ARGSUSED */ static int bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) { struct bpf_d *d; int error; error = devfs_get_cdevpriv((void **)&d); if (error != 0) return (error); /* * Refresh PID associated with this descriptor. */ BPFD_LOCK(d); BPF_PID_REFRESH(d, td); if (d->bd_state == BPF_WAITING) callout_stop(&d->bd_callout); d->bd_state = BPF_IDLE; BPFD_UNLOCK(d); if (d->bd_locked == 1) { switch (cmd) { case BIOCGBLEN: case BIOCFLUSH: case BIOCGDLT: case BIOCGDLTLIST: #ifdef COMPAT_FREEBSD32 case BIOCGDLTLIST32: #endif case BIOCGETIF: case BIOCGRTIMEOUT: #if defined(COMPAT_FREEBSD32) && !defined(__mips__) case BIOCGRTIMEOUT32: #endif case BIOCGSTATS: case BIOCVERSION: case BIOCGRSIG: case BIOCGHDRCMPLT: case BIOCSTSTAMP: case BIOCFEEDBACK: case FIONREAD: case BIOCLOCK: case BIOCSRTIMEOUT: #if defined(COMPAT_FREEBSD32) && !defined(__mips__) case BIOCSRTIMEOUT32: #endif case BIOCIMMEDIATE: case TIOCGPGRP: case BIOCROTZBUF: break; default: return (EPERM); } } #ifdef COMPAT_FREEBSD32 /* * If we see a 32-bit compat ioctl, mark the stream as 32-bit so * that it will get 32-bit packet headers. */ switch (cmd) { case BIOCSETF32: case BIOCSETFNR32: case BIOCSETWF32: case BIOCGDLTLIST32: case BIOCGRTIMEOUT32: case BIOCSRTIMEOUT32: BPFD_LOCK(d); d->bd_compat32 = 1; BPFD_UNLOCK(d); } #endif CURVNET_SET(TD_TO_VNET(td)); switch (cmd) { default: error = EINVAL; break; /* * Check for read packet available. */ case FIONREAD: { int n; BPFD_LOCK(d); n = d->bd_slen; while (d->bd_hbuf_in_use) mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, "bd_hbuf", 0); if (d->bd_hbuf) n += d->bd_hlen; BPFD_UNLOCK(d); *(int *)addr = n; break; } /* * Get buffer len [for read()]. */ case BIOCGBLEN: BPFD_LOCK(d); *(u_int *)addr = d->bd_bufsize; BPFD_UNLOCK(d); break; /* * Set buffer length. */ case BIOCSBLEN: error = bpf_ioctl_sblen(d, (u_int *)addr); break; /* * Set link layer read filter. */ case BIOCSETF: case BIOCSETFNR: case BIOCSETWF: #ifdef COMPAT_FREEBSD32 case BIOCSETF32: case BIOCSETFNR32: case BIOCSETWF32: #endif error = bpf_setf(d, (struct bpf_program *)addr, cmd); break; /* * Flush read packet buffer. */ case BIOCFLUSH: BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); break; /* * Put interface into promiscuous mode. */ case BIOCPROMISC: if (d->bd_bif == NULL) { /* * No interface attached yet. */ error = EINVAL; break; } if (d->bd_promisc == 0) { error = ifpromisc(d->bd_bif->bif_ifp, 1); if (error == 0) d->bd_promisc = 1; } break; /* * Get current data link type. */ case BIOCGDLT: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else *(u_int *)addr = d->bd_bif->bif_dlt; BPF_UNLOCK(); break; /* * Get a list of supported data link types. */ #ifdef COMPAT_FREEBSD32 case BIOCGDLTLIST32: { struct bpf_dltlist32 *list32; struct bpf_dltlist dltlist; list32 = (struct bpf_dltlist32 *)addr; dltlist.bfl_len = list32->bfl_len; dltlist.bfl_list = PTRIN(list32->bfl_list); BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else { error = bpf_getdltlist(d, &dltlist); if (error == 0) list32->bfl_len = dltlist.bfl_len; } BPF_UNLOCK(); break; } #endif case BIOCGDLTLIST: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); BPF_UNLOCK(); break; /* * Set data link type. */ case BIOCSDLT: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else error = bpf_setdlt(d, *(u_int *)addr); BPF_UNLOCK(); break; /* * Get interface name. */ case BIOCGETIF: BPF_LOCK(); if (d->bd_bif == NULL) error = EINVAL; else { struct ifnet *const ifp = d->bd_bif->bif_ifp; struct ifreq *const ifr = (struct ifreq *)addr; strlcpy(ifr->ifr_name, ifp->if_xname, sizeof(ifr->ifr_name)); } BPF_UNLOCK(); break; /* * Set interface. */ case BIOCSETIF: { int alloc_buf, size; /* * Behavior here depends on the buffering model. If * we're using kernel memory buffers, then we can * allocate them here. If we're using zero-copy, * then the user process must have registered buffers * by the time we get here. */ alloc_buf = 0; BPFD_LOCK(d); if (d->bd_bufmode == BPF_BUFMODE_BUFFER && d->bd_sbuf == NULL) alloc_buf = 1; BPFD_UNLOCK(d); if (alloc_buf) { size = d->bd_bufsize; error = bpf_buffer_ioctl_sblen(d, &size); if (error != 0) break; } BPF_LOCK(); error = bpf_setif(d, (struct ifreq *)addr); BPF_UNLOCK(); break; } /* * Set read timeout. */ case BIOCSRTIMEOUT: #if defined(COMPAT_FREEBSD32) && !defined(__mips__) case BIOCSRTIMEOUT32: #endif { struct timeval *tv = (struct timeval *)addr; #if defined(COMPAT_FREEBSD32) && !defined(__mips__) struct timeval32 *tv32; struct timeval tv64; if (cmd == BIOCSRTIMEOUT32) { tv32 = (struct timeval32 *)addr; tv = &tv64; tv->tv_sec = tv32->tv_sec; tv->tv_usec = tv32->tv_usec; } else #endif tv = (struct timeval *)addr; /* * Subtract 1 tick from tvtohz() since this isn't * a one-shot timer. */ if ((error = itimerfix(tv)) == 0) d->bd_rtout = tvtohz(tv) - 1; break; } /* * Get read timeout. */ case BIOCGRTIMEOUT: #if defined(COMPAT_FREEBSD32) && !defined(__mips__) case BIOCGRTIMEOUT32: #endif { struct timeval *tv; #if defined(COMPAT_FREEBSD32) && !defined(__mips__) struct timeval32 *tv32; struct timeval tv64; if (cmd == BIOCGRTIMEOUT32) tv = &tv64; else #endif tv = (struct timeval *)addr; tv->tv_sec = d->bd_rtout / hz; tv->tv_usec = (d->bd_rtout % hz) * tick; #if defined(COMPAT_FREEBSD32) && !defined(__mips__) if (cmd == BIOCGRTIMEOUT32) { tv32 = (struct timeval32 *)addr; tv32->tv_sec = tv->tv_sec; tv32->tv_usec = tv->tv_usec; } #endif break; } /* * Get packet stats. */ case BIOCGSTATS: { struct bpf_stat *bs = (struct bpf_stat *)addr; /* XXXCSJP overflow */ bs->bs_recv = d->bd_rcount; bs->bs_drop = d->bd_dcount; break; } /* * Set immediate mode. */ case BIOCIMMEDIATE: BPFD_LOCK(d); d->bd_immediate = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCVERSION: { struct bpf_version *bv = (struct bpf_version *)addr; bv->bv_major = BPF_MAJOR_VERSION; bv->bv_minor = BPF_MINOR_VERSION; break; } /* * Get "header already complete" flag */ case BIOCGHDRCMPLT: BPFD_LOCK(d); *(u_int *)addr = d->bd_hdrcmplt; BPFD_UNLOCK(d); break; /* * Set "header already complete" flag */ case BIOCSHDRCMPLT: BPFD_LOCK(d); d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; BPFD_UNLOCK(d); break; /* * Get packet direction flag */ case BIOCGDIRECTION: BPFD_LOCK(d); *(u_int *)addr = d->bd_direction; BPFD_UNLOCK(d); break; /* * Set packet direction flag */ case BIOCSDIRECTION: { u_int direction; direction = *(u_int *)addr; switch (direction) { case BPF_D_IN: case BPF_D_INOUT: case BPF_D_OUT: BPFD_LOCK(d); d->bd_direction = direction; BPFD_UNLOCK(d); break; default: error = EINVAL; } } break; /* * Get packet timestamp format and resolution. */ case BIOCGTSTAMP: BPFD_LOCK(d); *(u_int *)addr = d->bd_tstamp; BPFD_UNLOCK(d); break; /* * Set packet timestamp format and resolution. */ case BIOCSTSTAMP: { u_int func; func = *(u_int *)addr; if (BPF_T_VALID(func)) d->bd_tstamp = func; else error = EINVAL; } break; case BIOCFEEDBACK: BPFD_LOCK(d); d->bd_feedback = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCLOCK: BPFD_LOCK(d); d->bd_locked = 1; BPFD_UNLOCK(d); break; case FIONBIO: /* Non-blocking I/O */ break; case FIOASYNC: /* Send signal on receive packets */ BPFD_LOCK(d); d->bd_async = *(int *)addr; BPFD_UNLOCK(d); break; case FIOSETOWN: /* * XXX: Add some sort of locking here? * fsetown() can sleep. */ error = fsetown(*(int *)addr, &d->bd_sigio); break; case FIOGETOWN: BPFD_LOCK(d); *(int *)addr = fgetown(&d->bd_sigio); BPFD_UNLOCK(d); break; /* This is deprecated, FIOSETOWN should be used instead. */ case TIOCSPGRP: error = fsetown(-(*(int *)addr), &d->bd_sigio); break; /* This is deprecated, FIOGETOWN should be used instead. */ case TIOCGPGRP: *(int *)addr = -fgetown(&d->bd_sigio); break; case BIOCSRSIG: /* Set receive signal */ { u_int sig; sig = *(u_int *)addr; if (sig >= NSIG) error = EINVAL; else { BPFD_LOCK(d); d->bd_sig = sig; BPFD_UNLOCK(d); } break; } case BIOCGRSIG: BPFD_LOCK(d); *(u_int *)addr = d->bd_sig; BPFD_UNLOCK(d); break; case BIOCGETBUFMODE: BPFD_LOCK(d); *(u_int *)addr = d->bd_bufmode; BPFD_UNLOCK(d); break; case BIOCSETBUFMODE: /* * Allow the buffering mode to be changed as long as we * haven't yet committed to a particular mode. Our * definition of commitment, for now, is whether or not a * buffer has been allocated or an interface attached, since * that's the point where things get tricky. */ switch (*(u_int *)addr) { case BPF_BUFMODE_BUFFER: break; case BPF_BUFMODE_ZBUF: if (bpf_zerocopy_enable) break; /* FALLSTHROUGH */ default: CURVNET_RESTORE(); return (EINVAL); } BPFD_LOCK(d); if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || d->bd_fbuf != NULL || d->bd_bif != NULL) { BPFD_UNLOCK(d); CURVNET_RESTORE(); return (EBUSY); } d->bd_bufmode = *(u_int *)addr; BPFD_UNLOCK(d); break; case BIOCGETZMAX: error = bpf_ioctl_getzmax(td, d, (size_t *)addr); break; case BIOCSETZBUF: error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); break; case BIOCROTZBUF: error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); break; } CURVNET_RESTORE(); return (error); } /* * Set d's packet filter program to fp. If this file already has a filter, * free it and replace it. Returns EINVAL for bogus requests. * * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls * since reading d->bd_bif can't be protected by d or interface lock due to * lock order. * * Additionally, we have to acquire interface write lock due to bpf_mtap() uses * interface read lock to read all filers. * */ static int bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) { #ifdef COMPAT_FREEBSD32 struct bpf_program fp_swab; struct bpf_program32 *fp32; #endif struct bpf_insn *fcode, *old; #ifdef BPF_JITTER bpf_jit_filter *jfunc, *ofunc; #endif size_t size; u_int flen; int need_upgrade; #ifdef COMPAT_FREEBSD32 switch (cmd) { case BIOCSETF32: case BIOCSETWF32: case BIOCSETFNR32: fp32 = (struct bpf_program32 *)fp; fp_swab.bf_len = fp32->bf_len; fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; fp = &fp_swab; switch (cmd) { case BIOCSETF32: cmd = BIOCSETF; break; case BIOCSETWF32: cmd = BIOCSETWF; break; } break; } #endif fcode = NULL; #ifdef BPF_JITTER jfunc = ofunc = NULL; #endif need_upgrade = 0; /* * Check new filter validness before acquiring any locks. * Allocate memory for new filter, if needed. */ flen = fp->bf_len; if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) return (EINVAL); size = flen * sizeof(*fp->bf_insns); if (size > 0) { /* We're setting up new filter. Copy and check actual data. */ fcode = malloc(size, M_BPF, M_WAITOK); if (copyin(fp->bf_insns, fcode, size) != 0 || !bpf_validate(fcode, flen)) { free(fcode, M_BPF); return (EINVAL); } #ifdef BPF_JITTER /* Filter is copied inside fcode and is perfectly valid. */ jfunc = bpf_jitter(fcode, flen); #endif } BPF_LOCK(); /* * Set up new filter. * Protect filter change by interface lock. * Additionally, we are protected by global lock here. */ if (d->bd_bif != NULL) BPFIF_WLOCK(d->bd_bif); BPFD_LOCK(d); if (cmd == BIOCSETWF) { old = d->bd_wfilter; d->bd_wfilter = fcode; } else { old = d->bd_rfilter; d->bd_rfilter = fcode; #ifdef BPF_JITTER ofunc = d->bd_bfilter; d->bd_bfilter = jfunc; #endif if (cmd == BIOCSETF) reset_d(d); need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen); } BPFD_UNLOCK(d); if (d->bd_bif != NULL) BPFIF_WUNLOCK(d->bd_bif); if (old != NULL) free(old, M_BPF); #ifdef BPF_JITTER if (ofunc != NULL) bpf_destroy_jit_filter(ofunc); #endif /* Move d to active readers list. */ if (need_upgrade != 0) bpf_upgraded(d); BPF_UNLOCK(); return (0); } /* * Detach a file from its current interface (if attached at all) and attach * to the interface indicated by the name stored in ifr. * Return an errno or 0. */ static int bpf_setif(struct bpf_d *d, struct ifreq *ifr) { struct bpf_if *bp; struct ifnet *theywant; BPF_LOCK_ASSERT(); theywant = ifunit(ifr->ifr_name); if (theywant == NULL || theywant->if_bpf == NULL) return (ENXIO); bp = theywant->if_bpf; /* Check if interface is not being detached from BPF */ BPFIF_RLOCK(bp); if (bp->bif_flags & BPFIF_FLAG_DYING) { BPFIF_RUNLOCK(bp); return (ENXIO); } BPFIF_RUNLOCK(bp); /* * At this point, we expect the buffer is already allocated. If not, * return an error. */ switch (d->bd_bufmode) { case BPF_BUFMODE_BUFFER: case BPF_BUFMODE_ZBUF: if (d->bd_sbuf == NULL) return (EINVAL); break; default: panic("bpf_setif: bufmode %d", d->bd_bufmode); } if (bp != d->bd_bif) bpf_attachd(d, bp); BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); return (0); } /* * Support for select() and poll() system calls * * Return true iff the specific operation will not block indefinitely. * Otherwise, return false but make a note that a selwakeup() must be done. */ static int bpfpoll(struct cdev *dev, int events, struct thread *td) { struct bpf_d *d; int revents; if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) return (events & (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); /* * Refresh PID associated with this descriptor. */ revents = events & (POLLOUT | POLLWRNORM); BPFD_LOCK(d); BPF_PID_REFRESH(d, td); if (events & (POLLIN | POLLRDNORM)) { if (bpf_ready(d)) revents |= events & (POLLIN | POLLRDNORM); else { selrecord(td, &d->bd_sel); /* Start the read timeout if necessary. */ if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { callout_reset(&d->bd_callout, d->bd_rtout, bpf_timed_out, d); d->bd_state = BPF_WAITING; } } } BPFD_UNLOCK(d); return (revents); } /* * Support for kevent() system call. Register EVFILT_READ filters and * reject all others. */ int bpfkqfilter(struct cdev *dev, struct knote *kn) { struct bpf_d *d; if (devfs_get_cdevpriv((void **)&d) != 0 || kn->kn_filter != EVFILT_READ) return (1); /* * Refresh PID associated with this descriptor. */ BPFD_LOCK(d); BPF_PID_REFRESH_CUR(d); kn->kn_fop = &bpfread_filtops; kn->kn_hook = d; knlist_add(&d->bd_sel.si_note, kn, 1); BPFD_UNLOCK(d); return (0); } static void filt_bpfdetach(struct knote *kn) { struct bpf_d *d = (struct bpf_d *)kn->kn_hook; knlist_remove(&d->bd_sel.si_note, kn, 0); } static int filt_bpfread(struct knote *kn, long hint) { struct bpf_d *d = (struct bpf_d *)kn->kn_hook; int ready; BPFD_LOCK_ASSERT(d); ready = bpf_ready(d); if (ready) { kn->kn_data = d->bd_slen; /* * Ignore the hold buffer if it is being copied to user space. */ if (!d->bd_hbuf_in_use && d->bd_hbuf) kn->kn_data += d->bd_hlen; } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { callout_reset(&d->bd_callout, d->bd_rtout, bpf_timed_out, d); d->bd_state = BPF_WAITING; } return (ready); } #define BPF_TSTAMP_NONE 0 #define BPF_TSTAMP_FAST 1 #define BPF_TSTAMP_NORMAL 2 #define BPF_TSTAMP_EXTERN 3 static int bpf_ts_quality(int tstype) { if (tstype == BPF_T_NONE) return (BPF_TSTAMP_NONE); if ((tstype & BPF_T_FAST) != 0) return (BPF_TSTAMP_FAST); return (BPF_TSTAMP_NORMAL); } static int bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) { struct m_tag *tag; int quality; quality = bpf_ts_quality(tstype); if (quality == BPF_TSTAMP_NONE) return (quality); if (m != NULL) { tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); if (tag != NULL) { *bt = *(struct bintime *)(tag + 1); return (BPF_TSTAMP_EXTERN); } } if (quality == BPF_TSTAMP_NORMAL) binuptime(bt); else getbinuptime(bt); return (quality); } /* * Incoming linkage from device drivers. Process the packet pkt, of length * pktlen, which is stored in a contiguous buffer. The packet is parsed * by each process' filter, and if accepted, stashed into the corresponding * buffer. */ void bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) { struct bintime bt; struct bpf_d *d; #ifdef BPF_JITTER bpf_jit_filter *bf; #endif u_int slen; int gottime; gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { /* * We are not using any locks for d here because: * 1) any filter change is protected by interface * write lock * 2) destroying/detaching d is protected by interface * write lock, too */ /* XXX: Do not protect counter for the sake of performance. */ ++d->bd_rcount; /* * NB: We dont call BPF_CHECK_DIRECTION() here since there is no * way for the caller to indiciate to us whether this packet * is inbound or outbound. In the bpf_mtap() routines, we use * the interface pointers on the mbuf to figure it out. */ #ifdef BPF_JITTER bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; if (bf != NULL) slen = (*(bf->func))(pkt, pktlen, pktlen); else #endif slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); if (slen != 0) { /* * Filter matches. Let's to acquire write lock. */ BPFD_LOCK(d); d->bd_fcount++; if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, pkt, pktlen, slen, bpf_append_bytes, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } #define BPF_CHECK_DIRECTION(d, r, i) \ (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ ((d)->bd_direction == BPF_D_OUT && (r) == (i))) /* * Incoming linkage from device drivers, when packet is in an mbuf chain. * Locking model is explained in bpf_tap(). */ void bpf_mtap(struct bpf_if *bp, struct mbuf *m) { struct bintime bt; struct bpf_d *d; #ifdef BPF_JITTER bpf_jit_filter *bf; #endif u_int pktlen, slen; int gottime; /* Skip outgoing duplicate packets. */ if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { m->m_flags &= ~M_PROMISC; return; } pktlen = m_length(m, NULL); gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) continue; ++d->bd_rcount; #ifdef BPF_JITTER bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; /* XXX We cannot handle multiple mbufs. */ if (bf != NULL && m->m_next == NULL) slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); else #endif slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); if (slen != 0) { BPFD_LOCK(d); d->bd_fcount++; if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, m); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, (u_char *)m, pktlen, slen, bpf_append_mbuf, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } /* * Incoming linkage from device drivers, when packet is in * an mbuf chain and to be prepended by a contiguous header. */ void bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) { struct bintime bt; struct mbuf mb; struct bpf_d *d; u_int pktlen, slen; int gottime; /* Skip outgoing duplicate packets. */ if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { m->m_flags &= ~M_PROMISC; return; } pktlen = m_length(m, NULL); /* * Craft on-stack mbuf suitable for passing to bpf_filter. * Note that we cut corners here; we only setup what's * absolutely needed--this mbuf should never go anywhere else. */ mb.m_next = m; mb.m_data = data; mb.m_len = dlen; pktlen += dlen; gottime = BPF_TSTAMP_NONE; BPFIF_RLOCK(bp); LIST_FOREACH(d, &bp->bif_dlist, bd_next) { if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) continue; ++d->bd_rcount; slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); if (slen != 0) { BPFD_LOCK(d); d->bd_fcount++; if (gottime < bpf_ts_quality(d->bd_tstamp)) gottime = bpf_gettime(&bt, d->bd_tstamp, m); #ifdef MAC if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) #endif catchpacket(d, (u_char *)&mb, pktlen, slen, bpf_append_mbuf, &bt); BPFD_UNLOCK(d); } } BPFIF_RUNLOCK(bp); } #undef BPF_CHECK_DIRECTION #undef BPF_TSTAMP_NONE #undef BPF_TSTAMP_FAST #undef BPF_TSTAMP_NORMAL #undef BPF_TSTAMP_EXTERN static int bpf_hdrlen(struct bpf_d *d) { int hdrlen; hdrlen = d->bd_bif->bif_hdrlen; #ifndef BURN_BRIDGES if (d->bd_tstamp == BPF_T_NONE || BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); else #endif hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); else #endif hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) hdrlen = BPF_WORDALIGN32(hdrlen); else #endif hdrlen = BPF_WORDALIGN(hdrlen); return (hdrlen - d->bd_bif->bif_hdrlen); } static void bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) { struct bintime bt2; struct timeval tsm; struct timespec tsn; if ((tstype & BPF_T_MONOTONIC) == 0) { bt2 = *bt; bintime_add(&bt2, &boottimebin); bt = &bt2; } switch (BPF_T_FORMAT(tstype)) { case BPF_T_MICROTIME: bintime2timeval(bt, &tsm); ts->bt_sec = tsm.tv_sec; ts->bt_frac = tsm.tv_usec; break; case BPF_T_NANOTIME: bintime2timespec(bt, &tsn); ts->bt_sec = tsn.tv_sec; ts->bt_frac = tsn.tv_nsec; break; case BPF_T_BINTIME: ts->bt_sec = bt->sec; ts->bt_frac = bt->frac; break; } } /* * Move the packet data from interface memory (pkt) into the * store buffer. "cpfn" is the routine called to do the actual data * transfer. bcopy is passed in to copy contiguous chunks, while * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, * pkt is really an mbuf. */ static void catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), struct bintime *bt) { struct bpf_xhdr hdr; #ifndef BURN_BRIDGES struct bpf_hdr hdr_old; #ifdef COMPAT_FREEBSD32 struct bpf_hdr32 hdr32_old; #endif #endif int caplen, curlen, hdrlen, totlen; int do_wakeup = 0; int do_timestamp; int tstype; BPFD_LOCK_ASSERT(d); /* * Detect whether user space has released a buffer back to us, and if * so, move it from being a hold buffer to a free buffer. This may * not be the best place to do it (for example, we might only want to * run this check if we need the space), but for now it's a reliable * spot to do it. */ if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; bpf_buf_reclaimed(d); } /* * Figure out how many bytes to move. If the packet is * greater or equal to the snapshot length, transfer that * much. Otherwise, transfer the whole packet (unless * we hit the buffer size limit). */ hdrlen = bpf_hdrlen(d); totlen = hdrlen + min(snaplen, pktlen); if (totlen > d->bd_bufsize) totlen = d->bd_bufsize; /* * Round up the end of the previous packet to the next longword. * * Drop the packet if there's no room and no hope of room * If the packet would overflow the storage buffer or the storage * buffer is considered immutable by the buffer model, try to rotate * the buffer and wakeup pending processes. */ #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) curlen = BPF_WORDALIGN32(d->bd_slen); else #endif curlen = BPF_WORDALIGN(d->bd_slen); if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { if (d->bd_fbuf == NULL) { /* * There's no room in the store buffer, and no * prospect of room, so drop the packet. Notify the * buffer model. */ bpf_buffull(d); ++d->bd_dcount; return; } KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); ROTATE_BUFFERS(d); do_wakeup = 1; curlen = 0; } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) /* * Immediate mode is set, or the read timeout has already * expired during a select call. A packet arrived, so the * reader should be woken up. */ do_wakeup = 1; caplen = totlen - hdrlen; tstype = d->bd_tstamp; do_timestamp = tstype != BPF_T_NONE; #ifndef BURN_BRIDGES if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { struct bpf_ts ts; if (do_timestamp) bpf_bintime2ts(bt, &ts, tstype); #ifdef COMPAT_FREEBSD32 if (d->bd_compat32) { bzero(&hdr32_old, sizeof(hdr32_old)); if (do_timestamp) { hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; } hdr32_old.bh_datalen = pktlen; hdr32_old.bh_hdrlen = hdrlen; hdr32_old.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, sizeof(hdr32_old)); goto copy; } #endif bzero(&hdr_old, sizeof(hdr_old)); if (do_timestamp) { hdr_old.bh_tstamp.tv_sec = ts.bt_sec; hdr_old.bh_tstamp.tv_usec = ts.bt_frac; } hdr_old.bh_datalen = pktlen; hdr_old.bh_hdrlen = hdrlen; hdr_old.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, sizeof(hdr_old)); goto copy; } #endif /* * Append the bpf header. Note we append the actual header size, but * move forward the length of the header plus padding. */ bzero(&hdr, sizeof(hdr)); if (do_timestamp) bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); hdr.bh_datalen = pktlen; hdr.bh_hdrlen = hdrlen; hdr.bh_caplen = caplen; bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); /* * Copy the packet data into the store buffer and update its length. */ #ifndef BURN_BRIDGES copy: #endif (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); d->bd_slen = curlen + totlen; if (do_wakeup) bpf_wakeup(d); } /* * Free buffers currently in use by a descriptor. * Called on close. */ static void bpf_freed(struct bpf_d *d) { /* * We don't need to lock out interrupts since this descriptor has * been detached from its interface and it yet hasn't been marked * free. */ bpf_free(d); if (d->bd_rfilter != NULL) { free((caddr_t)d->bd_rfilter, M_BPF); #ifdef BPF_JITTER if (d->bd_bfilter != NULL) bpf_destroy_jit_filter(d->bd_bfilter); #endif } if (d->bd_wfilter != NULL) free((caddr_t)d->bd_wfilter, M_BPF); mtx_destroy(&d->bd_lock); } /* * Attach an interface to bpf. dlt is the link layer type; hdrlen is the * fixed size of the link header (variable length headers not yet supported). */ void bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) { bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); } /* * Attach an interface to bpf. ifp is a pointer to the structure * defining the interface to be attached, dlt is the link layer type, * and hdrlen is the fixed size of the link header (variable length * headers are not yet supporrted). */ void bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) { struct bpf_if *bp; bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); if (bp == NULL) panic("bpfattach"); LIST_INIT(&bp->bif_dlist); LIST_INIT(&bp->bif_wlist); bp->bif_ifp = ifp; bp->bif_dlt = dlt; rw_init(&bp->bif_lock, "bpf interface lock"); KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); *driverp = bp; BPF_LOCK(); LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); BPF_UNLOCK(); bp->bif_hdrlen = hdrlen; - if (bootverbose) + if (bootverbose && IS_DEFAULT_VNET(curvnet)) if_printf(ifp, "bpf attached\n"); } /* * Detach bpf from an interface. This involves detaching each descriptor * associated with the interface. Notify each descriptor as it's detached * so that any sleepers wake up and get ENXIO. */ void bpfdetach(struct ifnet *ifp) { struct bpf_if *bp, *bp_temp; struct bpf_d *d; int ndetached; ndetached = 0; BPF_LOCK(); /* Find all bpf_if struct's which reference ifp and detach them. */ LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { if (ifp != bp->bif_ifp) continue; LIST_REMOVE(bp, bif_next); /* Add to to-be-freed list */ LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); ndetached++; /* * Delay freeing bp till interface is detached * and all routes through this interface are removed. * Mark bp as detached to restrict new consumers. */ BPFIF_WLOCK(bp); bp->bif_flags |= BPFIF_FLAG_DYING; BPFIF_WUNLOCK(bp); CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", __func__, bp->bif_dlt, bp, ifp); /* Free common descriptors */ while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { bpf_detachd_locked(d); BPFD_LOCK(d); bpf_wakeup(d); BPFD_UNLOCK(d); } /* Free writer-only descriptors */ while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { bpf_detachd_locked(d); BPFD_LOCK(d); bpf_wakeup(d); BPFD_UNLOCK(d); } } BPF_UNLOCK(); #ifdef INVARIANTS if (ndetached == 0) printf("bpfdetach: %s was not attached\n", ifp->if_xname); #endif } /* * Interface departure handler. * Note departure event does not guarantee interface is going down. * Interface renaming is currently done via departure/arrival event set. * * Departure handled is called after all routes pointing to * given interface are removed and interface is in down state * restricting any packets to be sent/received. We assume it is now safe * to free data allocated by BPF. */ static void bpf_ifdetach(void *arg __unused, struct ifnet *ifp) { struct bpf_if *bp, *bp_temp; int nmatched = 0; BPF_LOCK(); /* * Find matching entries in free list. * Nothing should be found if bpfdetach() was not called. */ LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { if (ifp != bp->bif_ifp) continue; CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", __func__, bp, ifp); LIST_REMOVE(bp, bif_next); rw_destroy(&bp->bif_lock); free(bp, M_BPF); nmatched++; } BPF_UNLOCK(); /* * Note that we cannot zero other pointers to * custom DLTs possibly used by given interface. */ if (nmatched != 0) ifp->if_bpf = NULL; } /* * Get a list of available data link type of the interface. */ static int bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) { int n, error; struct ifnet *ifp; struct bpf_if *bp; BPF_LOCK_ASSERT(); ifp = d->bd_bif->bif_ifp; n = 0; error = 0; LIST_FOREACH(bp, &bpf_iflist, bif_next) { if (bp->bif_ifp != ifp) continue; if (bfl->bfl_list != NULL) { if (n >= bfl->bfl_len) return (ENOMEM); error = copyout(&bp->bif_dlt, bfl->bfl_list + n, sizeof(u_int)); } n++; } bfl->bfl_len = n; return (error); } /* * Set the data link type of a BPF instance. */ static int bpf_setdlt(struct bpf_d *d, u_int dlt) { int error, opromisc; struct ifnet *ifp; struct bpf_if *bp; BPF_LOCK_ASSERT(); if (d->bd_bif->bif_dlt == dlt) return (0); ifp = d->bd_bif->bif_ifp; LIST_FOREACH(bp, &bpf_iflist, bif_next) { if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) break; } if (bp != NULL) { opromisc = d->bd_promisc; bpf_attachd(d, bp); BPFD_LOCK(d); reset_d(d); BPFD_UNLOCK(d); if (opromisc) { error = ifpromisc(bp->bif_ifp, 1); if (error) if_printf(bp->bif_ifp, "bpf_setdlt: ifpromisc failed (%d)\n", error); else d->bd_promisc = 1; } } return (bp == NULL ? EINVAL : 0); } static void bpf_drvinit(void *unused) { struct cdev *dev; mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF); LIST_INIT(&bpf_iflist); LIST_INIT(&bpf_freelist); dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); /* For compatibility */ make_dev_alias(dev, "bpf0"); /* Register interface departure handler */ bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( ifnet_departure_event, bpf_ifdetach, NULL, EVENTHANDLER_PRI_ANY); } /* * Zero out the various packet counters associated with all of the bpf * descriptors. At some point, we will probably want to get a bit more * granular and allow the user to specify descriptors to be zeroed. */ static void bpf_zero_counters(void) { struct bpf_if *bp; struct bpf_d *bd; BPF_LOCK(); LIST_FOREACH(bp, &bpf_iflist, bif_next) { BPFIF_RLOCK(bp); LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { BPFD_LOCK(bd); bd->bd_rcount = 0; bd->bd_dcount = 0; bd->bd_fcount = 0; bd->bd_wcount = 0; bd->bd_wfcount = 0; bd->bd_zcopy = 0; BPFD_UNLOCK(bd); } BPFIF_RUNLOCK(bp); } BPF_UNLOCK(); } /* * Fill filter statistics */ static void bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) { bzero(d, sizeof(*d)); BPFD_LOCK_ASSERT(bd); d->bd_structsize = sizeof(*d); /* XXX: reading should be protected by global lock */ d->bd_immediate = bd->bd_immediate; d->bd_promisc = bd->bd_promisc; d->bd_hdrcmplt = bd->bd_hdrcmplt; d->bd_direction = bd->bd_direction; d->bd_feedback = bd->bd_feedback; d->bd_async = bd->bd_async; d->bd_rcount = bd->bd_rcount; d->bd_dcount = bd->bd_dcount; d->bd_fcount = bd->bd_fcount; d->bd_sig = bd->bd_sig; d->bd_slen = bd->bd_slen; d->bd_hlen = bd->bd_hlen; d->bd_bufsize = bd->bd_bufsize; d->bd_pid = bd->bd_pid; strlcpy(d->bd_ifname, bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); d->bd_locked = bd->bd_locked; d->bd_wcount = bd->bd_wcount; d->bd_wdcount = bd->bd_wdcount; d->bd_wfcount = bd->bd_wfcount; d->bd_zcopy = bd->bd_zcopy; d->bd_bufmode = bd->bd_bufmode; } /* * Handle `netstat -B' stats request */ static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) { static const struct xbpf_d zerostats; struct xbpf_d *xbdbuf, *xbd, tempstats; int index, error; struct bpf_if *bp; struct bpf_d *bd; /* * XXX This is not technically correct. It is possible for non * privileged users to open bpf devices. It would make sense * if the users who opened the devices were able to retrieve * the statistics for them, too. */ error = priv_check(req->td, PRIV_NET_BPF); if (error) return (error); /* * Check to see if the user is requesting that the counters be * zeroed out. Explicitly check that the supplied data is zeroed, * as we aren't allowing the user to set the counters currently. */ if (req->newptr != NULL) { if (req->newlen != sizeof(tempstats)) return (EINVAL); memset(&tempstats, 0, sizeof(tempstats)); error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); if (error) return (error); if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) return (EINVAL); bpf_zero_counters(); return (0); } if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); if (bpf_bpfd_cnt == 0) return (SYSCTL_OUT(req, 0, 0)); xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); BPF_LOCK(); if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { BPF_UNLOCK(); free(xbdbuf, M_BPF); return (ENOMEM); } index = 0; LIST_FOREACH(bp, &bpf_iflist, bif_next) { BPFIF_RLOCK(bp); /* Send writers-only first */ LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { xbd = &xbdbuf[index++]; BPFD_LOCK(bd); bpfstats_fill_xbpf(xbd, bd); BPFD_UNLOCK(bd); } LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { xbd = &xbdbuf[index++]; BPFD_LOCK(bd); bpfstats_fill_xbpf(xbd, bd); BPFD_UNLOCK(bd); } BPFIF_RUNLOCK(bp); } BPF_UNLOCK(); error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); free(xbdbuf, M_BPF); return (error); } SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); #else /* !DEV_BPF && !NETGRAPH_BPF */ /* * NOP stubs to allow bpf-using drivers to load and function. * * A 'better' implementation would allow the core bpf functionality * to be loaded at runtime. */ static struct bpf_if bp_null; void bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) { } void bpf_mtap(struct bpf_if *bp, struct mbuf *m) { } void bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) { } void bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) { bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); } void bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) { *driverp = &bp_null; } void bpfdetach(struct ifnet *ifp) { } u_int bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) { return -1; /* "no filter" behaviour */ } int bpf_validate(const struct bpf_insn *f, int len) { return 0; /* false */ } #endif /* !DEV_BPF && !NETGRAPH_BPF */ Index: head/sys/netinet/tcp_subr.c =================================================================== --- head/sys/netinet/tcp_subr.c (revision 292602) +++ head/sys/netinet/tcp_subr.c (revision 292603) @@ -1,2846 +1,2846 @@ /*- * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_tcpdebug.h" #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #include #include #endif #include #include #include #include #include #ifdef INET6 #include #endif #include #ifdef TCPPCAP #include #endif #ifdef TCPDEBUG #include #endif #ifdef INET6 #include #endif #ifdef TCP_OFFLOAD #include #endif #ifdef IPSEC #include #include #ifdef INET6 #include #endif #include #include #endif /*IPSEC*/ #include #include #include VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; #ifdef INET6 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; #endif struct rwlock tcp_function_lock; static int sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", "Default TCP Maximum Segment Size"); #ifdef INET6 static int sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) { int error, new; new = V_tcp_v6mssdflt; error = sysctl_handle_int(oidp, &new, 0, req); if (error == 0 && req->newptr) { if (new < TCP_MINMSS) error = EINVAL; else V_tcp_v6mssdflt = new; } return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", "Default TCP Maximum Segment Size for IPv6"); #endif /* INET6 */ /* * Minimum MSS we accept and use. This prevents DoS attacks where * we are forced to a ridiculous low MSS like 20 and send hundreds * of packets instead of one. The effect scales with the available * bandwidth and quickly saturates the CPU and network interface * with packet generation and sending. Set to zero to disable MINMSS * checking. This setting prevents us from sending too small packets. */ VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_minmss), 0, "Minimum TCP Maximum Segment Size"); VNET_DEFINE(int, tcp_do_rfc1323) = 1; SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc1323), 0, "Enable rfc1323 (high performance TCP) extensions"); static int tcp_log_debug = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); static int tcp_tcbhashsize; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); static int do_tcpdrain = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, "Enable tcp_drain routine for extra help when low on mbufs"); SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); static VNET_DEFINE(int, icmp_may_rst) = 1; #define V_icmp_may_rst VNET(icmp_may_rst) SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(icmp_may_rst), 0, "Certain ICMP unreachable messages may abort connections in SYN_SENT"); static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_isn_reseed_interval), 0, "Seconds between reseeding of ISN secret"); static int tcp_soreceive_stream; SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); #ifdef TCP_SIGNATURE static int tcp_sig_checksigs = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); #endif VNET_DEFINE(uma_zone_t, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); static struct inpcb *tcp_notify(struct inpcb *, int); static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); static void tcp_mtudisc(struct inpcb *, int); static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr); static void tcp_timer_discard(struct tcpcb *, uint32_t); static struct tcp_function_block tcp_def_funcblk = { "default", tcp_output, tcp_do_segment, tcp_default_ctloutput, NULL, NULL, NULL, NULL, NULL, NULL, NULL, 0, 0 }; struct tcp_funchead t_functions; static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; static struct tcp_function_block * find_tcp_functions_locked(struct tcp_function_set *fs) { struct tcp_function *f; struct tcp_function_block *blk=NULL; TAILQ_FOREACH(f, &t_functions, tf_next) { if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) { blk = f->tf_fb; break; } } return(blk); } static struct tcp_function_block * find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) { struct tcp_function_block *rblk=NULL; struct tcp_function *f; TAILQ_FOREACH(f, &t_functions, tf_next) { if (f->tf_fb == blk) { rblk = blk; if (s) { *s = f; } break; } } return (rblk); } struct tcp_function_block * find_and_ref_tcp_functions(struct tcp_function_set *fs) { struct tcp_function_block *blk; rw_rlock(&tcp_function_lock); blk = find_tcp_functions_locked(fs); if (blk) refcount_acquire(&blk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(blk); } struct tcp_function_block * find_and_ref_tcp_fb(struct tcp_function_block *blk) { struct tcp_function_block *rblk; rw_rlock(&tcp_function_lock); rblk = find_tcp_fb_locked(blk, NULL); if (rblk) refcount_acquire(&rblk->tfb_refcnt); rw_runlock(&tcp_function_lock); return(rblk); } static int sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) { int error=ENOENT; struct tcp_function_set fs; struct tcp_function_block *blk; memset(&fs, 0, sizeof(fs)); rw_rlock(&tcp_function_lock); blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); if (blk) { /* Found him */ strcpy(fs.function_set_name, blk->tfb_tcp_block_name); fs.pcbcnt = blk->tfb_refcnt; } rw_runlock(&tcp_function_lock); error = sysctl_handle_string(oidp, fs.function_set_name, sizeof(fs.function_set_name), req); /* Check for error or no change */ if (error != 0 || req->newptr == NULL) return(error); rw_wlock(&tcp_function_lock); blk = find_tcp_functions_locked(&fs); if ((blk == NULL) || (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { error = ENOENT; goto done; } tcp_func_set_ptr = blk; done: rw_wunlock(&tcp_function_lock); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, CTLTYPE_STRING | CTLFLAG_RW, NULL, 0, sysctl_net_inet_default_tcp_functions, "A", "Set/get the default TCP functions"); static int sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) { int error, cnt, linesz; struct tcp_function *f; char *buffer, *cp; size_t bufsz, outsz; cnt = 0; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { cnt++; } rw_runlock(&tcp_function_lock); bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1; buffer = malloc(bufsz, M_TEMP, M_WAITOK); error = 0; cp = buffer; linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count"); cp += linesz; bufsz -= linesz; outsz = linesz; rw_rlock(&tcp_function_lock); TAILQ_FOREACH(f, &t_functions, tf_next) { linesz = snprintf(cp, bufsz, "%-32s%c %u\n", f->tf_fb->tfb_tcp_block_name, (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', f->tf_fb->tfb_refcnt); if (linesz >= bufsz) { error = EOVERFLOW; break; } cp += linesz; bufsz -= linesz; outsz += linesz; } rw_runlock(&tcp_function_lock); if (error == 0) error = sysctl_handle_string(oidp, buffer, outsz + 1, req); free(buffer, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, CTLTYPE_STRING|CTLFLAG_RD, NULL, 0, sysctl_net_inet_list_available, "A", "list available TCP Function sets"); /* * Target size of TCP PCB hash tables. Must be a power of two. * * Note that this can be overridden by the kernel environment * variable net.inet.tcp.tcbhashsize */ #ifndef TCBHASHSIZE #define TCBHASHSIZE 0 #endif /* * XXX * Callouts should be moved into struct tcp directly. They are currently * separate because the tcpcb structure is exported to userland for sysctl * parsing purposes, which do not know about callouts. */ struct tcpcb_mem { struct tcpcb tcb; struct tcp_timer tt; struct cc_var ccv; struct osd osd; }; static VNET_DEFINE(uma_zone_t, tcpcb_zone); #define V_tcpcb_zone VNET(tcpcb_zone) MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); static struct mtx isn_mtx; #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) #define ISN_LOCK() mtx_lock(&isn_mtx) #define ISN_UNLOCK() mtx_unlock(&isn_mtx) /* * TCP initialization. */ static void tcp_zone_change(void *tag) { uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); uma_zone_set_max(V_tcpcb_zone, maxsockets); tcp_tw_zone_change(); } static int tcp_inpcb_init(void *mem, int size, int flags) { struct inpcb *inp = mem; INP_LOCK_INIT(inp, "inp", "tcpinp"); return (0); } /* * Take a value and get the next power of 2 that doesn't overflow. * Used to size the tcp_inpcb hash buckets. */ static int maketcp_hashsize(int size) { int hashsize; /* * auto tune. * get the next power of 2 higher than maxsockets. */ hashsize = 1 << fls(size); /* catch overflow, and just go one power of 2 smaller */ if (hashsize < size) { hashsize = 1 << (fls(size) - 1); } return (hashsize); } int register_tcp_functions(struct tcp_function_block *blk, int wait) { struct tcp_function_block *lblk; struct tcp_function *n; struct tcp_function_set fs; if ((blk->tfb_tcp_output == NULL) || (blk->tfb_tcp_do_segment == NULL) || (blk->tfb_tcp_ctloutput == NULL) || (strlen(blk->tfb_tcp_block_name) == 0)) { /* * These functions are required and you * need a name. */ return (EINVAL); } if (blk->tfb_tcp_timer_stop_all || blk->tfb_tcp_timers_left || blk->tfb_tcp_timer_activate || blk->tfb_tcp_timer_active || blk->tfb_tcp_timer_stop) { /* * If you define one timer function you * must have them all. */ if ((blk->tfb_tcp_timer_stop_all == NULL) || (blk->tfb_tcp_timers_left == NULL) || (blk->tfb_tcp_timer_activate == NULL) || (blk->tfb_tcp_timer_active == NULL) || (blk->tfb_tcp_timer_stop == NULL)) { return (EINVAL); } } n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); if (n == NULL) { return (ENOMEM); } n->tf_fb = blk; strcpy(fs.function_set_name, blk->tfb_tcp_block_name); rw_wlock(&tcp_function_lock); lblk = find_tcp_functions_locked(&fs); if (lblk) { /* Duplicate name space not allowed */ rw_wunlock(&tcp_function_lock); free(n, M_TCPFUNCTIONS); return (EALREADY); } refcount_init(&blk->tfb_refcnt, 0); blk->tfb_flags = 0; TAILQ_INSERT_TAIL(&t_functions, n, tf_next); rw_wunlock(&tcp_function_lock); return(0); } int deregister_tcp_functions(struct tcp_function_block *blk) { struct tcp_function_block *lblk; struct tcp_function *f; int error=ENOENT; if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { /* You can't un-register the default */ return (EPERM); } rw_wlock(&tcp_function_lock); if (blk == tcp_func_set_ptr) { /* You can't free the current default */ rw_wunlock(&tcp_function_lock); return (EBUSY); } if (blk->tfb_refcnt) { /* Still tcb attached, mark it. */ blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; rw_wunlock(&tcp_function_lock); return (EBUSY); } lblk = find_tcp_fb_locked(blk, &f); if (lblk) { /* Found */ TAILQ_REMOVE(&t_functions, f, tf_next); f->tf_fb = NULL; free(f, M_TCPFUNCTIONS); error = 0; } rw_wunlock(&tcp_function_lock); return (error); } void tcp_init(void) { const char *tcbhash_tuneable; int hashsize; tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) printf("%s: WARNING: unable to register helper hook\n", __func__); hashsize = TCBHASHSIZE; TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); if (hashsize == 0) { /* * Auto tune the hash size based on maxsockets. * A perfect hash would have a 1:1 mapping * (hashsize = maxsockets) however it's been * suggested that O(2) average is better. */ hashsize = maketcp_hashsize(maxsockets / 4); /* * Our historical default is 512, * do not autotune lower than this. */ if (hashsize < 512) hashsize = 512; - if (bootverbose) + if (bootverbose && IS_DEFAULT_VNET(curvnet)) printf("%s: %s auto tuned to %d\n", __func__, tcbhash_tuneable, hashsize); } /* * We require a hashsize to be a power of two. * Previously if it was not a power of two we would just reset it * back to 512, which could be a nasty surprise if you did not notice * the error message. * Instead what we do is clip it to the closest power of two lower * than the specified hash value. */ if (!powerof2(hashsize)) { int oldhashsize = hashsize; hashsize = maketcp_hashsize(hashsize); /* prevent absurdly low value */ if (hashsize < 16) hashsize = 16; printf("%s: WARNING: TCB hash size not a power of 2, " "clipped from %d to %d.\n", __func__, oldhashsize, hashsize); } in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE, IPI_HASHFIELDS_4TUPLE); /* * These have to be type stable for the benefit of the timers. */ V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); uma_zone_set_max(V_tcpcb_zone, maxsockets); uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); tcp_tw_init(); syncache_init(); tcp_hc_init(); TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); /* Skip initialization of globals for non-default instances. */ if (!IS_DEFAULT_VNET(curvnet)) return; tcp_reass_global_init(); /* XXX virtualize those bellow? */ tcp_delacktime = TCPTV_DELACK; tcp_keepinit = TCPTV_KEEP_INIT; tcp_keepidle = TCPTV_KEEP_IDLE; tcp_keepintvl = TCPTV_KEEPINTVL; tcp_maxpersistidle = TCPTV_KEEP_IDLE; tcp_msl = TCPTV_MSL; tcp_rexmit_min = TCPTV_MIN; if (tcp_rexmit_min < 1) tcp_rexmit_min = 1; tcp_rexmit_slop = TCPTV_CPU_VAR; tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; tcp_tcbhashsize = hashsize; /* Setup the tcp function block list */ TAILQ_INIT(&t_functions); rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); register_tcp_functions(&tcp_def_funcblk, M_WAITOK); if (tcp_soreceive_stream) { #ifdef INET tcp_usrreqs.pru_soreceive = soreceive_stream; #endif #ifdef INET6 tcp6_usrreqs.pru_soreceive = soreceive_stream; #endif /* INET6 */ } #ifdef INET6 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) #else /* INET6 */ #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) #endif /* INET6 */ if (max_protohdr < TCP_MINPROTOHDR) max_protohdr = TCP_MINPROTOHDR; if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) panic("tcp_init"); #undef TCP_MINPROTOHDR ISN_LOCK_INIT(); EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, SHUTDOWN_PRI_DEFAULT); EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, EVENTHANDLER_PRI_ANY); #ifdef TCPPCAP tcp_pcap_init(); #endif } #ifdef VIMAGE void tcp_destroy(void) { int error; tcp_hc_destroy(); syncache_destroy(); tcp_tw_destroy(); in_pcbinfo_destroy(&V_tcbinfo); uma_zdestroy(V_sack_hole_zone); uma_zdestroy(V_tcpcb_zone); error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); } error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); if (error != 0) { printf("%s: WARNING: unable to deregister helper hook " "type=%d, id=%d: error %d returned\n", __func__, HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); } } #endif void tcp_fini(void *xtp) { } /* * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. * tcp_template used to store this data in mbufs, but we now recopy it out * of the tcpcb each time to conserve mbufs. */ void tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) { struct tcphdr *th = (struct tcphdr *)tcp_ptr; INP_WLOCK_ASSERT(inp); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { struct ip6_hdr *ip6; ip6 = (struct ip6_hdr *)ip_ptr; ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | (inp->inp_flow & IPV6_FLOWINFO_MASK); ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | (IPV6_VERSION & IPV6_VERSION_MASK); ip6->ip6_nxt = IPPROTO_TCP; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { struct ip *ip; ip = (struct ip *)ip_ptr; ip->ip_v = IPVERSION; ip->ip_hl = 5; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = 0; ip->ip_id = 0; ip->ip_off = 0; ip->ip_ttl = inp->inp_ip_ttl; ip->ip_sum = 0; ip->ip_p = IPPROTO_TCP; ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } #endif /* INET */ th->th_sport = inp->inp_lport; th->th_dport = inp->inp_fport; th->th_seq = 0; th->th_ack = 0; th->th_x2 = 0; th->th_off = 5; th->th_flags = 0; th->th_win = 0; th->th_urp = 0; th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ } /* * Create template to be used to send tcp packets on a connection. * Allocates an mbuf and fills in a skeletal tcp/ip header. The only * use for this function is in keepalives, which use tcp_respond. */ struct tcptemp * tcpip_maketemplate(struct inpcb *inp) { struct tcptemp *t; t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); if (t == NULL) return (NULL); tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); return (t); } /* * Send a single message to the TCP at address specified by * the given TCP/IP header. If m == NULL, then we make a copy * of the tcpiphdr at th and send directly to the addressed host. * This is used to force keep alive messages out using the TCP * template for a connection. If flags are given then we send * a message back to the TCP which originated the segment th, * and discard the mbuf containing it and any other attached mbufs. * * In any case the ack and sequence number of the transmitted * segment are as specified by the parameters. * * NOTE: If m != NULL, then th must point to *inside* the mbuf. */ void tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, tcp_seq ack, tcp_seq seq, int flags) { int tlen; int win = 0; struct ip *ip; struct tcphdr *nth; #ifdef INET6 struct ip6_hdr *ip6; int isipv6; #endif /* INET6 */ int ipflags = 0; struct inpcb *inp; KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); #ifdef INET6 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); ip6 = ipgen; #endif /* INET6 */ ip = ipgen; if (tp != NULL) { inp = tp->t_inpcb; KASSERT(inp != NULL, ("tcp control block w/o inpcb")); INP_WLOCK_ASSERT(inp); } else inp = NULL; if (tp != NULL) { if (!(flags & TH_RST)) { win = sbspace(&inp->inp_socket->so_rcv); if (win > (long)TCP_MAXWIN << tp->rcv_scale) win = (long)TCP_MAXWIN << tp->rcv_scale; } } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return; tlen = 0; m->m_data += max_linkhdr; #ifdef INET6 if (isipv6) { bcopy((caddr_t)ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr)); ip6 = mtod(m, struct ip6_hdr *); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); ip = mtod(m, struct ip *); nth = (struct tcphdr *)(ip + 1); } bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); flags = TH_ACK; } else { /* * reuse the mbuf. * XXX MRT We inherrit the FIB, which is lucky. */ m_freem(m->m_next); m->m_next = NULL; m->m_data = (caddr_t)ipgen; /* m_len is set later */ tlen = 0; #define xchg(a,b,type) { type t; t=a; a=b; b=t; } #ifdef INET6 if (isipv6) { xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); nth = (struct tcphdr *)(ip6 + 1); } else #endif /* INET6 */ { xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); nth = (struct tcphdr *)(ip + 1); } if (th != nth) { /* * this is usually a case when an extension header * exists between the IPv6 header and the * TCP header. */ nth->th_sport = th->th_sport; nth->th_dport = th->th_dport; } xchg(nth->th_dport, nth->th_sport, uint16_t); #undef xchg } #ifdef INET6 if (isipv6) { ip6->ip6_flow = 0; ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_nxt = IPPROTO_TCP; tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); ip6->ip6_plen = htons(tlen - sizeof(*ip6)); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { tlen += sizeof (struct tcpiphdr); ip->ip_len = htons(tlen); ip->ip_ttl = V_ip_defttl; if (V_path_mtu_discovery) ip->ip_off |= htons(IP_DF); } #endif m->m_len = tlen; m->m_pkthdr.len = tlen; m->m_pkthdr.rcvif = NULL; #ifdef MAC if (inp != NULL) { /* * Packet is associated with a socket, so allow the * label of the response to reflect the socket label. */ INP_WLOCK_ASSERT(inp); mac_inpcb_create_mbuf(inp, m); } else { /* * Packet is not associated with a socket, so possibly * update the label in place. */ mac_netinet_tcp_reply(m); } #endif nth->th_seq = htonl(seq); nth->th_ack = htonl(ack); nth->th_x2 = 0; nth->th_off = sizeof (struct tcphdr) >> 2; nth->th_flags = flags; if (tp != NULL) nth->th_win = htons((u_short) (win >> tp->rcv_scale)); else nth->th_win = htons((u_short)win); nth->th_urp = 0; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); #ifdef INET6 if (isipv6) { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; nth->th_sum = in6_cksum_pseudo(ip6, tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : NULL, NULL); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET { m->m_pkthdr.csum_flags = CSUM_TCP; nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); } #endif /* INET */ #ifdef TCPDEBUG if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); #endif TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); if (flags & TH_RST) TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), tp, nth); TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); #ifdef INET6 if (isipv6) (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); #endif } /* * Create a new TCP control block, making an * empty reassembly queue and hooking it to the argument * protocol control block. The `inp' parameter must have * come from the zone allocator set up in tcp_init(). */ struct tcpcb * tcp_newtcpcb(struct inpcb *inp) { struct tcpcb_mem *tm; struct tcpcb *tp; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); if (tm == NULL) return (NULL); tp = &tm->tcb; /* Initialise cc_var struct for this tcpcb. */ tp->ccv = &tm->ccv; tp->ccv->type = IPPROTO_TCP; tp->ccv->ccvc.tcp = tp; rw_rlock(&tcp_function_lock); tp->t_fb = tcp_func_set_ptr; refcount_acquire(&tp->t_fb->tfb_refcnt); rw_runlock(&tcp_function_lock); if (tp->t_fb->tfb_tcp_fb_init) { (*tp->t_fb->tfb_tcp_fb_init)(tp); } /* * Use the current system default CC algorithm. */ CC_LIST_RLOCK(); KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); CC_ALGO(tp) = CC_DEFAULT(); CC_LIST_RUNLOCK(); if (CC_ALGO(tp)->cb_init != NULL) if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } tp->osd = &tm->osd; if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp); refcount_release(&tp->t_fb->tfb_refcnt); uma_zfree(V_tcpcb_zone, tm); return (NULL); } #ifdef VIMAGE tp->t_vnet = inp->inp_vnet; #endif tp->t_timers = &tm->tt; /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ tp->t_maxseg = tp->t_maxopd = #ifdef INET6 isipv6 ? V_tcp_v6mssdflt : #endif /* INET6 */ V_tcp_mssdflt; /* Set up our timeouts. */ callout_init(&tp->t_timers->tt_rexmt, 1); callout_init(&tp->t_timers->tt_persist, 1); callout_init(&tp->t_timers->tt_keep, 1); callout_init(&tp->t_timers->tt_2msl, 1); callout_init(&tp->t_timers->tt_delack, 1); if (V_tcp_do_rfc1323) tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); if (V_tcp_do_sack) tp->t_flags |= TF_SACK_PERMIT; TAILQ_INIT(&tp->snd_holes); /* * The tcpcb will hold a reference on its inpcb until tcp_discardcb() * is called. */ in_pcbref(inp); /* Reference for tcpcb */ tp->t_inpcb = inp; /* * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives * reasonable initial retransmit time. */ tp->t_srtt = TCPTV_SRTTBASE; tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; tp->t_rttmin = tcp_rexmit_min; tp->t_rxtcur = TCPTV_RTOBASE; tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; tp->t_rcvtime = ticks; /* * IPv4 TTL initialization is necessary for an IPv6 socket as well, * because the socket may be bound to an IPv6 wildcard address, * which may match an IPv4-mapped IPv6 address. */ inp->inp_ip_ttl = V_ip_defttl; inp->inp_ppcb = tp; #ifdef TCPPCAP /* * Init the TCP PCAP queues. */ tcp_pcap_tcpcb_init(tp); #endif return (tp); /* XXX */ } /* * Switch the congestion control algorithm back to NewReno for any active * control blocks using an algorithm which is about to go away. * This ensures the CC framework can allow the unload to proceed without leaving * any dangling pointers which would trigger a panic. * Returning non-zero would inform the CC framework that something went wrong * and it would be unsafe to allow the unload to proceed. However, there is no * way for this to occur with this implementation so we always return zero. */ int tcp_ccalgounload(struct cc_algo *unload_algo) { struct cc_algo *tmpalgo; struct inpcb *inp; struct tcpcb *tp; VNET_ITERATOR_DECL(vnet_iter); /* * Check all active control blocks across all network stacks and change * any that are using "unload_algo" back to NewReno. If "unload_algo" * requires cleanup code to be run, call it. */ VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); INP_INFO_WLOCK(&V_tcbinfo); /* * New connections already part way through being initialised * with the CC algo we're removing will not race with this code * because the INP_INFO_WLOCK is held during initialisation. We * therefore don't enter the loop below until the connection * list has stabilised. */ LIST_FOREACH(inp, &V_tcb, inp_list) { INP_WLOCK(inp); /* Important to skip tcptw structs. */ if (!(inp->inp_flags & INP_TIMEWAIT) && (tp = intotcpcb(inp)) != NULL) { /* * By holding INP_WLOCK here, we are assured * that the connection is not currently * executing inside the CC module's functions * i.e. it is safe to make the switch back to * NewReno. */ if (CC_ALGO(tp) == unload_algo) { tmpalgo = CC_ALGO(tp); /* NewReno does not require any init. */ CC_ALGO(tp) = &newreno_cc_algo; if (tmpalgo->cb_destroy != NULL) tmpalgo->cb_destroy(tp->ccv); } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); return (0); } /* * Drop a TCP connection, reporting * the specified error. If connection is synchronized, * then send a RST to peer. */ struct tcpcb * tcp_drop(struct tcpcb *tp, int errno) { struct socket *so = tp->t_inpcb->inp_socket; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(tp->t_inpcb); if (TCPS_HAVERCVDSYN(tp->t_state)) { tcp_state_change(tp, TCPS_CLOSED); (void) tp->t_fb->tfb_tcp_output(tp); TCPSTAT_INC(tcps_drops); } else TCPSTAT_INC(tcps_conndrops); if (errno == ETIMEDOUT && tp->t_softerror) errno = tp->t_softerror; so->so_error = errno; return (tcp_close(tp)); } void tcp_discardcb(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so = inp->inp_socket; #ifdef INET6 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif /* INET6 */ int released; INP_WLOCK_ASSERT(inp); /* * Make sure that all of our timers are stopped before we delete the * PCB. * * If stopping a timer fails, we schedule a discard function in same * callout, and the last discard function called will take care of * deleting the tcpcb. */ tcp_timer_stop(tp, TT_REXMT); tcp_timer_stop(tp, TT_PERSIST); tcp_timer_stop(tp, TT_KEEP); tcp_timer_stop(tp, TT_2MSL); tcp_timer_stop(tp, TT_DELACK); if (tp->t_fb->tfb_tcp_timer_stop_all) { /* Call the stop-all function of the methods */ tp->t_fb->tfb_tcp_timer_stop_all(tp); } /* * If we got enough samples through the srtt filter, * save the rtt and rttvar in the routing entry. * 'Enough' is arbitrarily defined as 4 rtt samples. * 4 samples is enough for the srtt filter to converge * to within enough % of the correct value; fewer samples * and we could save a bogus rtt. The danger is not high * as tcp quickly recovers from everything. * XXX: Works very well but needs some more statistics! */ if (tp->t_rttupdated >= 4) { struct hc_metrics_lite metrics; u_long ssthresh; bzero(&metrics, sizeof(metrics)); /* * Update the ssthresh always when the conditions below * are satisfied. This gives us better new start value * for the congestion avoidance for new connections. * ssthresh is only set if packet loss occured on a session. * * XXXRW: 'so' may be NULL here, and/or socket buffer may be * being torn down. Ideally this code would not use 'so'. */ ssthresh = tp->snd_ssthresh; if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { /* * convert the limit from user data bytes to * packets then to packet data bytes. */ ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; if (ssthresh < 2) ssthresh = 2; ssthresh *= (u_long)(tp->t_maxseg + #ifdef INET6 (isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : #endif sizeof (struct tcpiphdr) #ifdef INET6 ) #endif ); } else ssthresh = 0; metrics.rmx_ssthresh = ssthresh; metrics.rmx_rtt = tp->t_srtt; metrics.rmx_rttvar = tp->t_rttvar; metrics.rmx_cwnd = tp->snd_cwnd; metrics.rmx_sendpipe = 0; metrics.rmx_recvpipe = 0; tcp_hc_update(&inp->inp_inc, &metrics); } /* free the reassembly queue, if any */ tcp_reass_flush(tp); #ifdef TCP_OFFLOAD /* Disconnect offload device, if any. */ if (tp->t_flags & TF_TOE) tcp_offload_detach(tp); #endif tcp_free_sackholes(tp); #ifdef TCPPCAP /* Free the TCP PCAP queues. */ tcp_pcap_drain(&(tp->t_inpkts)); tcp_pcap_drain(&(tp->t_outpkts)); #endif /* Allow the CC algorithm to clean up after itself. */ if (CC_ALGO(tp)->cb_destroy != NULL) CC_ALGO(tp)->cb_destroy(tp->ccv); khelp_destroy_osd(tp->osd); CC_ALGO(tp) = NULL; inp->inp_ppcb = NULL; if ((tp->t_timers->tt_flags & TT_MASK) == 0) { /* We own the last reference on tcpcb, let's free it. */ if ((tp->t_fb->tfb_tcp_timers_left) && (tp->t_fb->tfb_tcp_timers_left(tp))) { /* Some fb timers left running! */ return; } if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); released = in_pcbrele_wlocked(inp); KASSERT(!released, ("%s: inp %p should not have been released " "here", __func__, inp)); } } void tcp_timer_2msl_discard(void *xtp) { tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL); } void tcp_timer_keep_discard(void *xtp) { tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP); } void tcp_timer_persist_discard(void *xtp) { tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST); } void tcp_timer_rexmt_discard(void *xtp) { tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT); } void tcp_timer_delack_discard(void *xtp) { tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK); } void tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type) { struct inpcb *inp; CURVNET_SET(tp->t_vnet); INP_INFO_RLOCK(&V_tcbinfo); inp = tp->t_inpcb; KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", __func__, tp)); INP_WLOCK(inp); KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, ("%s: tcpcb has to be stopped here", __func__)); KASSERT((tp->t_timers->tt_flags & timer_type) != 0, ("%s: discard callout should be running", __func__)); tp->t_timers->tt_flags &= ~timer_type; if ((tp->t_timers->tt_flags & TT_MASK) == 0) { /* We own the last reference on this tcpcb, let's free it. */ if ((tp->t_fb->tfb_tcp_timers_left) && (tp->t_fb->tfb_tcp_timers_left(tp))) { /* Some fb timers left running! */ goto leave; } if (tp->t_fb->tfb_tcp_fb_fini) (*tp->t_fb->tfb_tcp_fb_fini)(tp); refcount_release(&tp->t_fb->tfb_refcnt); tp->t_inpcb = NULL; uma_zfree(V_tcpcb_zone, tp); if (in_pcbrele_wlocked(inp)) { INP_INFO_RUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); return; } } leave: INP_WUNLOCK(inp); INP_INFO_RUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } /* * Attempt to close a TCP control block, marking it as dropped, and freeing * the socket if we hold the only reference. */ struct tcpcb * tcp_close(struct tcpcb *tp) { struct inpcb *inp = tp->t_inpcb; struct socket *so; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_state == TCPS_LISTEN) tcp_offload_listen_stop(tp); #endif in_pcbdrop(inp); TCPSTAT_INC(tcps_closed); KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); so = inp->inp_socket; soisdisconnected(so); if (inp->inp_flags & INP_SOCKREF) { KASSERT(so->so_state & SS_PROTOREF, ("tcp_close: !SS_PROTOREF")); inp->inp_flags &= ~INP_SOCKREF; INP_WUNLOCK(inp); ACCEPT_LOCK(); SOCK_LOCK(so); so->so_state &= ~SS_PROTOREF; sofree(so); return (NULL); } return (tp); } void tcp_drain(void) { VNET_ITERATOR_DECL(vnet_iter); if (!do_tcpdrain) return; VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); struct inpcb *inpb; struct tcpcb *tcpb; /* * Walk the tcpbs, if existing, and flush the reassembly queue, * if there is one... * XXX: The "Net/3" implementation doesn't imply that the TCP * reassembly queue should be flushed, but in a situation * where we're really low on mbufs, this is potentially * useful. */ INP_INFO_WLOCK(&V_tcbinfo); LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { if (inpb->inp_flags & INP_TIMEWAIT) continue; INP_WLOCK(inpb); if ((tcpb = intotcpcb(inpb)) != NULL) { tcp_reass_flush(tcpb); tcp_clean_sackreport(tcpb); } INP_WUNLOCK(inpb); } INP_INFO_WUNLOCK(&V_tcbinfo); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Notify a tcp user of an asynchronous error; * store error as soft error, but wake up user * (for now, won't do anything until can select for soft error). * * Do not wake up user since there currently is no mechanism for * reporting soft errors (yet - a kqueue filter may be added). */ static struct inpcb * tcp_notify(struct inpcb *inp, int error) { struct tcpcb *tp; INP_INFO_LOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); /* * Ignore some errors if we are hooked up. * If connection hasn't completed, has retransmitted several times, * and receives a second error, give up now. This is better * than waiting a long time to establish a connection that * can never complete. */ if (tp->t_state == TCPS_ESTABLISHED && (error == EHOSTUNREACH || error == ENETUNREACH || error == EHOSTDOWN)) { return (inp); } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && tp->t_softerror) { tp = tcp_drop(tp, error); if (tp != NULL) return (inp); else return (NULL); } else { tp->t_softerror = error; return (inp); } #if 0 wakeup( &so->so_timeo); sorwakeup(so); sowwakeup(so); #endif } static int tcp_pcblist(SYSCTL_HANDLER_ARGS) { int error, i, m, n, pcb_count; struct inpcb *inp, **inp_list; inp_gen_t gencnt; struct xinpgen xig; /* * The process of preparing the TCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = V_tcbinfo.ipi_count + syncache_pcbcount(); n += imax(n / 8, 10); req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ INP_LIST_RLOCK(&V_tcbinfo); gencnt = V_tcbinfo.ipi_gencnt; n = V_tcbinfo.ipi_count; INP_LIST_RUNLOCK(&V_tcbinfo); m = syncache_pcbcount(); error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) + (n + m) * sizeof(struct xtcpcb)); if (error != 0) return (error); xig.xig_len = sizeof xig; xig.xig_count = n + m; xig.xig_gen = gencnt; xig.xig_sogen = so_gencnt; error = SYSCTL_OUT(req, &xig, sizeof xig); if (error) return (error); error = syncache_pcblist(req, m, &pcb_count); if (error) return (error); inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); if (inp_list == NULL) return (ENOMEM); INP_INFO_WLOCK(&V_tcbinfo); for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { INP_WLOCK(inp); if (inp->inp_gencnt <= gencnt) { /* * XXX: This use of cr_cansee(), introduced with * TCP state changes, is not quite right, but for * now, better than nothing. */ if (inp->inp_flags & INP_TIMEWAIT) { if (intotw(inp) != NULL) error = cr_cansee(req->td->td_ucred, intotw(inp)->tw_cred); else error = EINVAL; /* Skip this inp. */ } else error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) { in_pcbref(inp); inp_list[i++] = inp; } } INP_WUNLOCK(inp); } INP_INFO_WUNLOCK(&V_tcbinfo); n = i; error = 0; for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (inp->inp_gencnt <= gencnt) { struct xtcpcb xt; void *inp_ppcb; bzero(&xt, sizeof(xt)); xt.xt_len = sizeof xt; /* XXX should avoid extra copy */ bcopy(inp, &xt.xt_inp, sizeof *inp); inp_ppcb = inp->inp_ppcb; if (inp_ppcb == NULL) bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); else if (inp->inp_flags & INP_TIMEWAIT) { bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); xt.xt_tp.t_state = TCPS_TIME_WAIT; } else { bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); if (xt.xt_tp.t_timers) tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); } if (inp->inp_socket != NULL) sotoxsocket(inp->inp_socket, &xt.xt_socket); else { bzero(&xt.xt_socket, sizeof xt.xt_socket); xt.xt_socket.xso_protocol = IPPROTO_TCP; } xt.xt_inp.inp_gencnt = inp->inp_gencnt; INP_RUNLOCK(inp); error = SYSCTL_OUT(req, &xt, sizeof xt); } else INP_RUNLOCK(inp); } INP_INFO_RLOCK(&V_tcbinfo); for (i = 0; i < n; i++) { inp = inp_list[i]; INP_RLOCK(inp); if (!in_pcbrele_rlocked(inp)) INP_RUNLOCK(inp); } INP_INFO_RUNLOCK(&V_tcbinfo); if (!error) { /* * Give the user an updated idea of our state. * If the generation differs from what we told * her before, she knows that something happened * while we were processing this request, and it * might be necessary to retry. */ INP_LIST_RLOCK(&V_tcbinfo); xig.xig_gen = V_tcbinfo.ipi_gencnt; xig.xig_sogen = so_gencnt; xig.xig_count = V_tcbinfo.ipi_count + pcb_count; INP_LIST_RUNLOCK(&V_tcbinfo); error = SYSCTL_OUT(req, &xig, sizeof xig); } free(inp_list, M_TEMP); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); #ifdef INET static int tcp_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in addrs[2]; struct inpcb *inp; int error; error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); #endif /* INET */ #ifdef INET6 static int tcp6_getcred(SYSCTL_HANDLER_ARGS) { struct xucred xuc; struct sockaddr_in6 addrs[2]; struct inpcb *inp; int error; #ifdef INET int mapped = 0; #endif error = priv_check(req->td, PRIV_NETINET_GETCRED); if (error) return (error); error = SYSCTL_IN(req, addrs, sizeof(addrs)); if (error) return (error); if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { return (error); } if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { #ifdef INET if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) mapped = 1; else #endif return (EINVAL); } #ifdef INET if (mapped == 1) inp = in_pcblookup(&V_tcbinfo, *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], addrs[1].sin6_port, *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); else #endif inp = in6_pcblookup(&V_tcbinfo, &addrs[1].sin6_addr, addrs[1].sin6_port, &addrs[0].sin6_addr, addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); if (inp != NULL) { if (inp->inp_socket == NULL) error = ENOENT; if (error == 0) error = cr_canseeinpcb(req->td->td_ucred, inp); if (error == 0) cru2x(inp->inp_cred, &xuc); INP_RUNLOCK(inp); } else error = ENOENT; if (error == 0) error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); return (error); } SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); #endif /* INET6 */ #ifdef INET void tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) { struct ip *ip = vip; struct tcphdr *th; struct in_addr faddr; struct inpcb *inp; struct tcpcb *tp; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct icmp *icp; struct in_conninfo inc; tcp_seq icmp_tcp_seq; int mtu; faddr = ((struct sockaddr_in *)sa)->sin_addr; if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) notify = tcp_drop_syn_sent; /* * Redirects don't need to be handled up here. */ else if (PRC_IS_REDIRECT(cmd)) return; /* * Hostdead is ugly because it goes linearly through all PCBs. * XXX: We never get this from ICMP, otherwise it makes an * excellent DoS attack on machines with many connections. */ else if (cmd == PRC_HOSTDEAD) ip = NULL; else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) return; if (ip == NULL) { in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); return; } icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); INP_INFO_RLOCK(&V_tcbinfo); inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); if (inp != NULL) { if (!(inp->inp_flags & INP_TIMEWAIT) && !(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket == NULL)) { icmp_tcp_seq = ntohl(th->th_seq); tp = intotcpcb(inp); if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && SEQ_LT(icmp_tcp_seq, tp->snd_max)) { if (cmd == PRC_MSGSIZE) { /* * MTU discovery: * If we got a needfrag set the MTU * in the route to the suggested new * value (if given) and then notify. */ mtu = ntohs(icp->icmp_nextmtu); /* * If no alternative MTU was * proposed, try the next smaller * one. */ if (!mtu) mtu = ip_next_mtu( ntohs(ip->ip_len), 1); if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) mtu = V_tcp_minmss + sizeof(struct tcpiphdr); /* * Only process the offered MTU if it * is smaller than the current one. */ if (mtu < tp->t_maxopd + sizeof(struct tcpiphdr)) { bzero(&inc, sizeof(inc)); inc.inc_faddr = faddr; inc.inc_fibnum = inp->inp_inc.inc_fibnum; tcp_hc_updatemtu(&inc, mtu); tcp_mtudisc(inp, mtu); } } else inp = (*notify)(inp, inetctlerrmap[cmd]); } } if (inp != NULL) INP_WUNLOCK(inp); } else { bzero(&inc, sizeof(inc)); inc.inc_fport = th->th_dport; inc.inc_lport = th->th_sport; inc.inc_faddr = faddr; inc.inc_laddr = ip->ip_src; syncache_unreach(&inc, th); } INP_INFO_RUNLOCK(&V_tcbinfo); } #endif /* INET */ #ifdef INET6 void tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) { struct tcphdr th; struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; struct ip6_hdr *ip6; struct mbuf *m; struct ip6ctlparam *ip6cp = NULL; const struct sockaddr_in6 *sa6_src = NULL; int off; struct tcp_portonly { u_int16_t th_sport; u_int16_t th_dport; } *thp; if (sa->sa_family != AF_INET6 || sa->sa_len != sizeof(struct sockaddr_in6)) return; if (cmd == PRC_MSGSIZE) notify = tcp_mtudisc_notify; else if (!PRC_IS_REDIRECT(cmd) && ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) return; /* if the parameter is from icmp6, decode it. */ if (d != NULL) { ip6cp = (struct ip6ctlparam *)d; m = ip6cp->ip6c_m; ip6 = ip6cp->ip6c_ip6; off = ip6cp->ip6c_off; sa6_src = ip6cp->ip6c_src; } else { m = NULL; ip6 = NULL; off = 0; /* fool gcc */ sa6_src = &sa6_any; } if (ip6 != NULL) { struct in_conninfo inc; /* * XXX: We assume that when IPV6 is non NULL, * M and OFF are valid. */ /* check if we can safely examine src and dst ports */ if (m->m_pkthdr.len < off + sizeof(*thp)) return; bzero(&th, sizeof(th)); m_copydata(m, off, sizeof(*thp), (caddr_t)&th); in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, (struct sockaddr *)ip6cp->ip6c_src, th.th_sport, cmd, NULL, notify); bzero(&inc, sizeof(inc)); inc.inc_fport = th.th_dport; inc.inc_lport = th.th_sport; inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; inc.inc_flags |= INC_ISIPV6; INP_INFO_RLOCK(&V_tcbinfo); syncache_unreach(&inc, &th); INP_INFO_RUNLOCK(&V_tcbinfo); } else in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify); } #endif /* INET6 */ /* * Following is where TCP initial sequence number generation occurs. * * There are two places where we must use initial sequence numbers: * 1. In SYN-ACK packets. * 2. In SYN packets. * * All ISNs for SYN-ACK packets are generated by the syncache. See * tcp_syncache.c for details. * * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling * depends on this property. In addition, these ISNs should be * unguessable so as to prevent connection hijacking. To satisfy * the requirements of this situation, the algorithm outlined in * RFC 1948 is used, with only small modifications. * * Implementation details: * * Time is based off the system timer, and is corrected so that it * increases by one megabyte per second. This allows for proper * recycling on high speed LANs while still leaving over an hour * before rollover. * * As reading the *exact* system time is too expensive to be done * whenever setting up a TCP connection, we increment the time * offset in two ways. First, a small random positive increment * is added to isn_offset for each connection that is set up. * Second, the function tcp_isn_tick fires once per clock tick * and increments isn_offset as necessary so that sequence numbers * are incremented at approximately ISN_BYTES_PER_SECOND. The * random positive increments serve only to ensure that the same * exact sequence number is never sent out twice (as could otherwise * happen when a port is recycled in less than the system tick * interval.) * * net.inet.tcp.isn_reseed_interval controls the number of seconds * between seeding of isn_secret. This is normally set to zero, * as reseeding should not be necessary. * * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In * general, this means holding an exclusive (write) lock. */ #define ISN_BYTES_PER_SECOND 1048576 #define ISN_STATIC_INCREMENT 4096 #define ISN_RANDOM_INCREMENT (4096 - 1) static VNET_DEFINE(u_char, isn_secret[32]); static VNET_DEFINE(int, isn_last); static VNET_DEFINE(int, isn_last_reseed); static VNET_DEFINE(u_int32_t, isn_offset); static VNET_DEFINE(u_int32_t, isn_offset_old); #define V_isn_secret VNET(isn_secret) #define V_isn_last VNET(isn_last) #define V_isn_last_reseed VNET(isn_last_reseed) #define V_isn_offset VNET(isn_offset) #define V_isn_offset_old VNET(isn_offset_old) tcp_seq tcp_new_isn(struct tcpcb *tp) { MD5_CTX isn_ctx; u_int32_t md5_buffer[4]; tcp_seq new_isn; u_int32_t projected_offset; INP_WLOCK_ASSERT(tp->t_inpcb); ISN_LOCK(); /* Seed if this is the first use, reseed if requested. */ if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) < (u_int)ticks))) { read_random(&V_isn_secret, sizeof(V_isn_secret)); V_isn_last_reseed = ticks; } /* Compute the md5 hash and return the ISN. */ MD5Init(&isn_ctx); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); #ifdef INET6 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, sizeof(struct in6_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, sizeof(struct in6_addr)); } else #endif { MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, sizeof(struct in_addr)); MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, sizeof(struct in_addr)); } MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); MD5Final((u_char *) &md5_buffer, &isn_ctx); new_isn = (tcp_seq) md5_buffer[0]; V_isn_offset += ISN_STATIC_INCREMENT + (arc4random() & ISN_RANDOM_INCREMENT); if (ticks != V_isn_last) { projected_offset = V_isn_offset_old + ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); if (SEQ_GT(projected_offset, V_isn_offset)) V_isn_offset = projected_offset; V_isn_offset_old = V_isn_offset; V_isn_last = ticks; } new_isn += V_isn_offset; ISN_UNLOCK(); return (new_isn); } /* * When a specific ICMP unreachable message is received and the * connection state is SYN-SENT, drop the connection. This behavior * is controlled by the icmp_may_rst sysctl. */ struct inpcb * tcp_drop_syn_sent(struct inpcb *inp, int errno) { struct tcpcb *tp; INP_INFO_RLOCK_ASSERT(&V_tcbinfo); INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return (inp); tp = intotcpcb(inp); if (tp->t_state != TCPS_SYN_SENT) return (inp); tp = tcp_drop(tp, errno); if (tp != NULL) return (inp); else return (NULL); } /* * When `need fragmentation' ICMP is received, update our idea of the MSS * based on the new value. Also nudge TCP to send something, since we * know the packet we just sent was dropped. * This duplicates some code in the tcp_mss() function in tcp_input.c. */ static struct inpcb * tcp_mtudisc_notify(struct inpcb *inp, int error) { tcp_mtudisc(inp, -1); return (inp); } static void tcp_mtudisc(struct inpcb *inp, int mtuoffer) { struct tcpcb *tp; struct socket *so; INP_WLOCK_ASSERT(inp); if ((inp->inp_flags & INP_TIMEWAIT) || (inp->inp_flags & INP_DROPPED)) return; tp = intotcpcb(inp); KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); /* If the mss is larger than the socket buffer, decrease the mss. */ if (so->so_snd.sb_hiwat < tp->t_maxseg) tp->t_maxseg = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); TCPSTAT_INC(tcps_mturesent); tp->t_rtttime = 0; tp->snd_nxt = tp->snd_una; tcp_free_sackholes(tp); tp->snd_recover = tp->snd_max; if (tp->t_flags & TF_SACK_PERMIT) EXIT_FASTRECOVERY(tp->t_flags); tp->t_fb->tfb_tcp_output(tp); } #ifdef INET /* * Look-up the routing entry to the peer of this inpcb. If no route * is found and it cannot be allocated, then return 0. This routine * is called by TCP routines that access the rmx structure and by * tcp_mss_update to get the peer/interface MTU. */ u_long tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct route sro; struct sockaddr_in *dst; struct ifnet *ifp; u_long maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); bzero(&sro, sizeof(sro)); if (inc->inc_faddr.s_addr != INADDR_ANY) { dst = (struct sockaddr_in *)&sro.ro_dst; dst->sin_family = AF_INET; dst->sin_len = sizeof(*dst); dst->sin_addr = inc->inc_faddr; in_rtalloc_ign(&sro, 0, inc->inc_fibnum); } if (sro.ro_rt != NULL) { ifp = sro.ro_rt->rt_ifp; if (sro.ro_rt->rt_mtu == 0) maxmtu = ifp->if_mtu; else maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu); /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO4 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } RTFREE(sro.ro_rt); } return (maxmtu); } #endif /* INET */ #ifdef INET6 u_long tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) { struct route_in6 sro6; struct ifnet *ifp; u_long maxmtu = 0; KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); bzero(&sro6, sizeof(sro6)); if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { sro6.ro_dst.sin6_family = AF_INET6; sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); sro6.ro_dst.sin6_addr = inc->inc6_faddr; in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum); } if (sro6.ro_rt != NULL) { ifp = sro6.ro_rt->rt_ifp; if (sro6.ro_rt->rt_mtu == 0) maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); else maxmtu = min(sro6.ro_rt->rt_mtu, IN6_LINKMTU(sro6.ro_rt->rt_ifp)); /* Report additional interface capabilities. */ if (cap != NULL) { if (ifp->if_capenable & IFCAP_TSO6 && ifp->if_hwassist & CSUM_TSO) { cap->ifcap |= CSUM_TSO; cap->tsomax = ifp->if_hw_tsomax; cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; } } RTFREE(sro6.ro_rt); } return (maxmtu); } #endif /* INET6 */ #ifdef IPSEC /* compute ESP/AH header size for TCP, including outer IP header. */ size_t ipsec_hdrsiz_tcp(struct tcpcb *tp) { struct inpcb *inp; struct mbuf *m; size_t hdrsiz; struct ip *ip; #ifdef INET6 struct ip6_hdr *ip6; #endif struct tcphdr *th; if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || (!key_havesp(IPSEC_DIR_OUTBOUND))) return (0); m = m_gethdr(M_NOWAIT, MT_DATA); if (!m) return (0); #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) { ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)(ip6 + 1); m->m_pkthdr.len = m->m_len = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); tcpip_fillheaders(inp, ip6, th); hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); } else #endif /* INET6 */ { ip = mtod(m, struct ip *); th = (struct tcphdr *)(ip + 1); m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); tcpip_fillheaders(inp, ip, th); hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); } m_free(m); return (hdrsiz); } #endif /* IPSEC */ #ifdef TCP_SIGNATURE /* * Callback function invoked by m_apply() to digest TCP segment data * contained within an mbuf chain. */ static int tcp_signature_apply(void *fstate, void *data, u_int len) { MD5Update(fstate, (u_char *)data, len); return (0); } /* * XXX The key is retrieved from the system's PF_KEY SADB, by keying a * search with the destination IP address, and a 'magic SPI' to be * determined by the application. This is hardcoded elsewhere to 1179 */ struct secasvar * tcp_get_sav(struct mbuf *m, u_int direction) { union sockaddr_union dst; struct secasvar *sav; struct ip *ip; #ifdef INET6 struct ip6_hdr *ip6; char ip6buf[INET6_ADDRSTRLEN]; #endif /* Extract the destination from the IP header in the mbuf. */ bzero(&dst, sizeof(union sockaddr_union)); ip = mtod(m, struct ip *); #ifdef INET6 ip6 = NULL; /* Make the compiler happy. */ #endif switch (ip->ip_v) { #ifdef INET case IPVERSION: dst.sa.sa_len = sizeof(struct sockaddr_in); dst.sa.sa_family = AF_INET; dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? ip->ip_src : ip->ip_dst; break; #endif #ifdef INET6 case (IPV6_VERSION >> 4): ip6 = mtod(m, struct ip6_hdr *); dst.sa.sa_len = sizeof(struct sockaddr_in6); dst.sa.sa_family = AF_INET6; dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? ip6->ip6_src : ip6->ip6_dst; break; #endif default: return (NULL); /* NOTREACHED */ break; } /* Look up an SADB entry which matches the address of the peer. */ sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); if (sav == NULL) { ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : #ifdef INET6 (ip->ip_v == (IPV6_VERSION >> 4)) ? ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : #endif "(unsupported)")); } return (sav); } /* * Compute TCP-MD5 hash of a TCP segment. (RFC2385) * * Parameters: * m pointer to head of mbuf chain * len length of TCP segment data, excluding options * optlen length of TCP segment options * buf pointer to storage for computed MD5 digest * sav pointer to security assosiation * * We do this over ip, tcphdr, segment data, and the key in the SADB. * When called from tcp_input(), we can be sure that th_sum has been * zeroed out and verified already. * * Releases reference to SADB key before return. * * Return 0 if successful, otherwise return -1. * */ int tcp_signature_do_compute(struct mbuf *m, int len, int optlen, u_char *buf, struct secasvar *sav) { #ifdef INET struct ippseudo ippseudo; #endif MD5_CTX ctx; int doff; struct ip *ip; #ifdef INET struct ipovly *ipovly; #endif struct tcphdr *th; #ifdef INET6 struct ip6_hdr *ip6; struct in6_addr in6; uint32_t plen; uint16_t nhdr; #endif u_short savecsum; KASSERT(m != NULL, ("NULL mbuf chain")); KASSERT(buf != NULL, ("NULL signature pointer")); /* Extract the destination from the IP header in the mbuf. */ ip = mtod(m, struct ip *); #ifdef INET6 ip6 = NULL; /* Make the compiler happy. */ #endif MD5Init(&ctx); /* * Step 1: Update MD5 hash with IP(v6) pseudo-header. * * XXX The ippseudo header MUST be digested in network byte order, * or else we'll fail the regression test. Assume all fields we've * been doing arithmetic on have been in host byte order. * XXX One cannot depend on ipovly->ih_len here. When called from * tcp_output(), the underlying ip_len member has not yet been set. */ switch (ip->ip_v) { #ifdef INET case IPVERSION: ipovly = (struct ipovly *)ip; ippseudo.ippseudo_src = ipovly->ih_src; ippseudo.ippseudo_dst = ipovly->ih_dst; ippseudo.ippseudo_pad = 0; ippseudo.ippseudo_p = IPPROTO_TCP; ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen); MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; break; #endif #ifdef INET6 /* * RFC 2385, 2.0 Proposal * For IPv6, the pseudo-header is as described in RFC 2460, namely the * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- * extended next header value (to form 32 bits), and 32-bit segment * length. * Note: Upper-Layer Packet Length comes before Next Header. */ case (IPV6_VERSION >> 4): in6 = ip6->ip6_src; in6_clearscope(&in6); MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); in6 = ip6->ip6_dst; in6_clearscope(&in6); MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); plen = htonl(len + sizeof(struct tcphdr) + optlen); MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); nhdr = 0; MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); nhdr = IPPROTO_TCP; MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; break; #endif default: KEY_FREESAV(&sav); return (-1); /* NOTREACHED */ break; } /* * Step 2: Update MD5 hash with TCP header, excluding options. * The TCP checksum must be set to zero. */ savecsum = th->th_sum; th->th_sum = 0; MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); th->th_sum = savecsum; /* * Step 3: Update MD5 hash with TCP segment data. * Use m_apply() to avoid an early m_pullup(). */ if (len > 0) m_apply(m, doff, len, tcp_signature_apply, &ctx); /* * Step 4: Update MD5 hash with shared secret. */ MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); MD5Final(buf, &ctx); key_sa_recordxfer(sav, m); KEY_FREESAV(&sav); return (0); } /* * Compute TCP-MD5 hash of a TCP segment. (RFC2385) * * Return 0 if successful, otherwise return -1. */ int tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, u_char *buf, u_int direction) { struct secasvar *sav; if ((sav = tcp_get_sav(m, direction)) == NULL) return (-1); return (tcp_signature_do_compute(m, len, optlen, buf, sav)); } /* * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) * * Parameters: * m pointer to head of mbuf chain * len length of TCP segment data, excluding options * optlen length of TCP segment options * buf pointer to storage for computed MD5 digest * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) * * Return 1 if successful, otherwise return 0. */ int tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) { char tmpdigest[TCP_SIGLEN]; if (tcp_sig_checksigs == 0) return (1); if ((tcpbflag & TF_SIGNATURE) == 0) { if ((to->to_flags & TOF_SIGNATURE) != 0) { /* * If this socket is not expecting signature but * the segment contains signature just fail. */ TCPSTAT_INC(tcps_sig_err_sigopt); TCPSTAT_INC(tcps_sig_rcvbadsig); return (0); } /* Signature is not expected, and not present in segment. */ return (1); } /* * If this socket is expecting signature but the segment does not * contain any just fail. */ if ((to->to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); TCPSTAT_INC(tcps_sig_rcvbadsig); return (0); } if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], IPSEC_DIR_INBOUND) == -1) { TCPSTAT_INC(tcps_sig_err_buildsig); TCPSTAT_INC(tcps_sig_rcvbadsig); return (0); } if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { TCPSTAT_INC(tcps_sig_rcvbadsig); return (0); } TCPSTAT_INC(tcps_sig_rcvgoodsig); return (1); } #endif /* TCP_SIGNATURE */ static int sysctl_drop(SYSCTL_HANDLER_ARGS) { /* addrs[0] is a foreign socket, addrs[1] is a local one. */ struct sockaddr_storage addrs[2]; struct inpcb *inp; struct tcpcb *tp; struct tcptw *tw; struct sockaddr_in *fin, *lin; #ifdef INET6 struct sockaddr_in6 *fin6, *lin6; #endif int error; inp = NULL; fin = lin = NULL; #ifdef INET6 fin6 = lin6 = NULL; #endif error = 0; if (req->oldptr != NULL || req->oldlen != 0) return (EINVAL); if (req->newptr == NULL) return (EPERM); if (req->newlen < sizeof(addrs)) return (ENOMEM); error = SYSCTL_IN(req, &addrs, sizeof(addrs)); if (error) return (error); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: fin6 = (struct sockaddr_in6 *)&addrs[0]; lin6 = (struct sockaddr_in6 *)&addrs[1]; if (fin6->sin6_len != sizeof(struct sockaddr_in6) || lin6->sin6_len != sizeof(struct sockaddr_in6)) return (EINVAL); if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) return (EINVAL); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; break; } error = sa6_embedscope(fin6, V_ip6_use_defzone); if (error) return (error); error = sa6_embedscope(lin6, V_ip6_use_defzone); if (error) return (error); break; #endif #ifdef INET case AF_INET: fin = (struct sockaddr_in *)&addrs[0]; lin = (struct sockaddr_in *)&addrs[1]; if (fin->sin_len != sizeof(struct sockaddr_in) || lin->sin_len != sizeof(struct sockaddr_in)) return (EINVAL); break; #endif default: return (EINVAL); } INP_INFO_RLOCK(&V_tcbinfo); switch (addrs[0].ss_family) { #ifdef INET6 case AF_INET6: inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif #ifdef INET case AF_INET: inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); break; #endif } if (inp != NULL) { if (inp->inp_flags & INP_TIMEWAIT) { /* * XXXRW: There currently exists a state where an * inpcb is present, but its timewait state has been * discarded. For now, don't allow dropping of this * type of inpcb. */ tw = intotw(inp); if (tw != NULL) tcp_twclose(tw, 0); else INP_WUNLOCK(inp); } else if (!(inp->inp_flags & INP_DROPPED) && !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { tp = intotcpcb(inp); tp = tcp_drop(tp, ECONNABORTED); if (tp != NULL) INP_WUNLOCK(inp); } else INP_WUNLOCK(inp); } else error = ESRCH; INP_INFO_RUNLOCK(&V_tcbinfo); return (error); } SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 0, sysctl_drop, "", "Drop TCP connection"); /* * Generate a standardized TCP log line for use throughout the * tcp subsystem. Memory allocation is done with M_NOWAIT to * allow use in the interrupt context. * * NB: The caller MUST free(s, M_TCPLOG) the returned string. * NB: The function may return NULL if memory allocation failed. * * Due to header inclusion and ordering limitations the struct ip * and ip6_hdr pointers have to be passed as void pointers. */ char * tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_in_vain == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } char * tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { /* Is logging enabled? */ if (tcp_log_debug == 0) return (NULL); return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); } static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, const void *ip6hdr) { char *s, *sp; size_t size; struct ip *ip; #ifdef INET6 const struct ip6_hdr *ip6; ip6 = (const struct ip6_hdr *)ip6hdr; #endif /* INET6 */ ip = (struct ip *)ip4hdr; /* * The log line looks like this: * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2" */ size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + sizeof(PRINT_TH_FLAGS) + 1 + #ifdef INET6 2 * INET6_ADDRSTRLEN; #else 2 * INET_ADDRSTRLEN; #endif /* INET6 */ s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); if (s == NULL) return (NULL); strcat(s, "TCP: ["); sp = s + strlen(s); if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { inet_ntoa_r(inc->inc_faddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); inet_ntoa_r(inc->inc_laddr, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); #ifdef INET6 } else if (inc) { ip6_sprintf(sp, &inc->inc6_faddr); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); sp = s + strlen(s); ip6_sprintf(sp, &inc->inc6_laddr); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(inc->inc_lport)); } else if (ip6 && th) { ip6_sprintf(sp, &ip6->ip6_src); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); ip6_sprintf(sp, &ip6->ip6_dst); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET6 */ #ifdef INET } else if (ip && th) { inet_ntoa_r(ip->ip_src, sp); sp = s + strlen(s); sprintf(sp, "]:%i to [", ntohs(th->th_sport)); sp = s + strlen(s); inet_ntoa_r(ip->ip_dst, sp); sp = s + strlen(s); sprintf(sp, "]:%i", ntohs(th->th_dport)); #endif /* INET */ } else { free(s, M_TCPLOG); return (NULL); } sp = s + strlen(s); if (th) sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); if (*(s + size - 1) != '\0') panic("%s: string too long", __func__); return (s); } /* * A subroutine which makes it easy to track TCP state changes with DTrace. * This function shouldn't be called for t_state initializations that don't * correspond to actual TCP state transitions. */ void tcp_state_change(struct tcpcb *tp, int newstate) { #if defined(KDTRACE_HOOKS) int pstate = tp->t_state; #endif tp->t_state = newstate; TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); }