Date: Mon, 17 Aug 2015 18:43:40 +0000 (UTC) From: Luiz Otavio O Souza <loos@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r286856 - stable/10/sys/net Message-ID: <201508171843.t7HIheAO076769@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: loos Date: Mon Aug 17 18:43:39 2015 New Revision: 286856 URL: https://svnweb.freebsd.org/changeset/base/286856 Log: MFC r286140: Remove the sleep from the buffer allocation routine. The buffer must be allocated (or even changed) before the interface is set and thus, there is no need to verify if the buffer is in use. MFC r286142: Remove two unnecessary sleeps from the hot path in bpf(4). The first one never triggers because bpf_canfreebuf() can only be true for zero-copy buffers and zero-copy buffers are not read with read(2). The second also never triggers, because we check the free buffer before calling ROTATE_BUFFERS(). If the hold buffer is in use the free buffer will be NULL and there is nothing else to do besides drop the packet. If the free buffer isn't NULL the hold buffer _is_ free and it is safe to rotate the buffers. Update the comment in ROTATE_BUFFERS macro to match the logic described here. While here fix a few typos in comments. MFC r286243: Add a KASSERT() to make sure we wont rotate the buffers twice (rotate the buffers while the hold buffer is in use). Sponsored by: Rubicon Communications (Netgate) Modified: stable/10/sys/net/bpf.c stable/10/sys/net/bpf.h stable/10/sys/net/bpf_buffer.c Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/net/bpf.c ============================================================================== --- stable/10/sys/net/bpf.c Mon Aug 17 18:42:45 2015 (r286855) +++ stable/10/sys/net/bpf.c Mon Aug 17 18:43:39 2015 (r286856) @@ -606,7 +606,7 @@ bpf_attachd(struct bpf_d *d, struct bpf_ bpf_detachd_locked(d); /* * Point d at bp, and add d to the interface's list. - * Since there are many applicaiotns using BPF for + * 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. @@ -643,7 +643,7 @@ bpf_attachd(struct bpf_d *d, struct bpf_ /* * Add d to the list of active bp filters. - * Reuqires bpf_attachd() to be called before + * Requires bpf_attachd() to be called before. */ static void bpf_upgraded(struct bpf_d *d) @@ -2319,9 +2319,6 @@ catchpacket(struct bpf_d *d, u_char *pkt * spot to do it. */ if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { - while (d->bd_hbuf_in_use) - mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, - PRINET, "bd_hbuf", 0); d->bd_fbuf = d->bd_hbuf; d->bd_hbuf = NULL; d->bd_hlen = 0; @@ -2364,9 +2361,7 @@ catchpacket(struct bpf_d *d, u_char *pkt ++d->bd_dcount; return; } - while (d->bd_hbuf_in_use) - mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, - PRINET, "bd_hbuf", 0); + KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); ROTATE_BUFFERS(d); do_wakeup = 1; curlen = 0; Modified: stable/10/sys/net/bpf.h ============================================================================== --- stable/10/sys/net/bpf.h Mon Aug 17 18:42:45 2015 (r286855) +++ stable/10/sys/net/bpf.h Mon Aug 17 18:43:39 2015 (r286856) @@ -1234,9 +1234,9 @@ SYSCTL_DECL(_net_bpf); /* * Rotate the packet buffers in descriptor d. Move the store buffer into the - * hold slot, and the free buffer ino the store slot. Zero the length of the - * new store buffer. Descriptor lock should be held. Hold buffer must - * not be marked "in use". + * hold slot, and the free buffer into the store slot. Zero the length of the + * new store buffer. Descriptor lock should be held. One must be careful to + * not rotate the buffers twice, i.e. if fbuf != NULL. */ #define ROTATE_BUFFERS(d) do { \ (d)->bd_hbuf = (d)->bd_sbuf; \ Modified: stable/10/sys/net/bpf_buffer.c ============================================================================== --- stable/10/sys/net/bpf_buffer.c Mon Aug 17 18:42:45 2015 (r286855) +++ stable/10/sys/net/bpf_buffer.c Mon Aug 17 18:43:39 2015 (r286856) @@ -79,8 +79,6 @@ __FBSDID("$FreeBSD$"); #include <net/bpf_buffer.h> #include <net/bpfdesc.h> -#define PRINET 26 /* interruptible */ - /* * Implement historical kernel memory buffering model for BPF: two malloc(9) * kernel buffers are hung off of the descriptor. The size is fixed prior to @@ -191,9 +189,6 @@ bpf_buffer_ioctl_sblen(struct bpf_d *d, return (EINVAL); } - while (d->bd_hbuf_in_use) - mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, - PRINET, "bd_hbuf", 0); /* Free old buffers if set */ if (d->bd_fbuf != NULL) free(d->bd_fbuf, M_BPF);
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201508171843.t7HIheAO076769>