Module Name:    src
Committed By:   martin
Date:           Fri Sep 13 14:13:06 UTC 2024

Modified Files:
        src/sys/net [netbsd-10]: bpf.c bpfdesc.h

Log Message:
Pull up following revision(s) (requested by ozaki-r in ticket #858):

        sys/net/bpfdesc.h: revision 1.49
        sys/net/bpf.c: revision 1.256
        sys/net/bpf.c: revision 1.257
        sys/net/bpfdesc.h: revision 1.50

bpf: restore wakeup softint

This change fixes the issue that fownsignal which can take an
adaptive mutex is called inside a pserialize read section in
bpf_deliver.

Fix issue #4 (only the latter of two) in PR#58596
bpf: protect selnotify and selrecord with bd_buf_mtx

We have to make updates and checks of buffers and calls of
selnotify/selrecord atomic to satisfy constraints of sel* API.

Also, bd_state and bd_cv are protected by bd_buf_mtx now.

Fix issue #3 of PR#58596

Part of the fix is inspired by riastradh's patch.


To generate a diff of this commit:
cvs rdiff -u -r1.249.2.2 -r1.249.2.3 src/sys/net/bpf.c
cvs rdiff -u -r1.48 -r1.48.10.1 src/sys/net/bpfdesc.h

Please note that diffs are not public domain; they are subject to the
copyright notices on the relevant files.

Modified files:

Index: src/sys/net/bpf.c
diff -u src/sys/net/bpf.c:1.249.2.2 src/sys/net/bpf.c:1.249.2.3
--- src/sys/net/bpf.c:1.249.2.2	Thu Aug 22 19:31:08 2024
+++ src/sys/net/bpf.c	Fri Sep 13 14:13:05 2024
@@ -1,4 +1,4 @@
-/*	$NetBSD: bpf.c,v 1.249.2.2 2024/08/22 19:31:08 martin Exp $	*/
+/*	$NetBSD: bpf.c,v 1.249.2.3 2024/09/13 14:13:05 martin Exp $	*/
 
 /*
  * Copyright (c) 1990, 1991, 1993
@@ -39,7 +39,7 @@
  */
 
 #include <sys/cdefs.h>
-__KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.249.2.2 2024/08/22 19:31:08 martin Exp $");
+__KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.249.2.3 2024/09/13 14:13:05 martin Exp $");
 
 #if defined(_KERNEL_OPT)
 #include "opt_bpf.h"
@@ -271,6 +271,7 @@ static int	bpf_poll(struct file *, int);
 static int	bpf_stat(struct file *, struct stat *);
 static int	bpf_close(struct file *);
 static int	bpf_kqfilter(struct file *, struct knote *);
+static void	bpf_softintr(void *);
 
 static const struct fileops bpf_fileops = {
 	.fo_name = "bpf",
@@ -605,6 +606,7 @@ bpfopen(dev_t dev, int flag, int mode, s
 	d->bd_atime = d->bd_mtime = d->bd_btime;
 	callout_init(&d->bd_callout, CALLOUT_MPSAFE);
 	selinit(&d->bd_sel);
+	d->bd_sih = softint_establish(SOFTINT_CLOCK, bpf_softintr, d);
 	d->bd_jitcode = NULL;
 	d->bd_rfilter = NULL;
 	d->bd_wfilter = NULL;
@@ -644,10 +646,12 @@ bpf_close(struct file *fp)
 	 */
 	d->bd_pid = curproc->p_pid;
 
-	mutex_enter(d->bd_mtx);
+	mutex_enter(d->bd_buf_mtx);
 	if (d->bd_state == BPF_WAITING)
-		callout_halt(&d->bd_callout, d->bd_mtx);
+		callout_halt(&d->bd_callout, d->bd_buf_mtx);
 	d->bd_state = BPF_IDLE;
+	mutex_exit(d->bd_buf_mtx);
+	mutex_enter(d->bd_mtx);
 	if (d->bd_bif)
 		bpf_detachd(d);
 	mutex_exit(d->bd_mtx);
@@ -663,6 +667,7 @@ bpf_close(struct file *fp)
 	bpf_freed(d);
 	callout_destroy(&d->bd_callout);
 	seldestroy(&d->bd_sel);
+	softint_disestablish(d->bd_sih);
 	mutex_obj_free(d->bd_mtx);
 	mutex_obj_free(d->bd_buf_mtx);
 	cv_destroy(&d->bd_cv);
@@ -707,12 +712,12 @@ bpf_read(struct file *fp, off_t *offp, s
 	if (uio->uio_resid != d->bd_bufsize)
 		return (EINVAL);
 
-	mutex_enter(d->bd_mtx);
+	mutex_enter(d->bd_buf_mtx);
 	if (d->bd_state == BPF_WAITING)
-		callout_halt(&d->bd_callout, d->bd_mtx);
+		callout_halt(&d->bd_callout, d->bd_buf_mtx);
 	timed_out = (d->bd_state == BPF_TIMED_OUT);
 	d->bd_state = BPF_IDLE;
-	mutex_exit(d->bd_mtx);
+	mutex_exit(d->bd_buf_mtx);
 	/*
 	 * If the hold buffer is empty, then do a timed sleep, which
 	 * ends when the timeout expires or when enough packets
@@ -797,13 +802,23 @@ static inline void
 bpf_wakeup(struct bpf_d *d)
 {
 
-	mutex_enter(d->bd_buf_mtx);
+	KASSERT(mutex_owned(d->bd_buf_mtx));
+
 	cv_broadcast(&d->bd_cv);
-	mutex_exit(d->bd_buf_mtx);
 
 	if (d->bd_async)
+		softint_schedule(d->bd_sih);
+	selnotify(&d->bd_sel, 0, NOTE_SUBMIT);
+}
+
+static void
+bpf_softintr(void *cookie)
+{
+	struct bpf_d *d;
+
+	d = cookie;
+	if (d->bd_async)
 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
-	selnotify(&d->bd_sel, 0, 0);
 }
 
 static void
@@ -811,13 +826,13 @@ bpf_timed_out(void *arg)
 {
 	struct bpf_d *d = arg;
 
-	mutex_enter(d->bd_mtx);
+	mutex_enter(d->bd_buf_mtx);
 	if (d->bd_state == BPF_WAITING) {
 		d->bd_state = BPF_TIMED_OUT;
 		if (d->bd_slen != 0)
 			bpf_wakeup(d);
 	}
-	mutex_exit(d->bd_mtx);
+	mutex_exit(d->bd_buf_mtx);
 }
 
 
@@ -979,11 +994,11 @@ bpf_ioctl(struct file *fp, u_long cmd, v
 		d->bd_compat32 = 0;
 #endif
 
-	mutex_enter(d->bd_mtx);
+	mutex_enter(d->bd_buf_mtx);
 	if (d->bd_state == BPF_WAITING)
-		callout_halt(&d->bd_callout, d->bd_mtx);
+		callout_halt(&d->bd_callout, d->bd_buf_mtx);
 	d->bd_state = BPF_IDLE;
-	mutex_exit(d->bd_mtx);
+	mutex_exit(d->bd_buf_mtx);
 
 	if (d->bd_locked) {
 		switch (cmd) {
@@ -1560,6 +1575,7 @@ bpf_poll(struct file *fp, int events)
 		 * An imitation of the FIONREAD ioctl code.
 		 */
 		mutex_enter(d->bd_mtx);
+		mutex_enter(d->bd_buf_mtx);
 		if (d->bd_hlen != 0 ||
 		    ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
 		     d->bd_slen != 0)) {
@@ -1573,6 +1589,7 @@ bpf_poll(struct file *fp, int events)
 				d->bd_state = BPF_WAITING;
 			}
 		}
+		mutex_exit(d->bd_buf_mtx);
 		mutex_exit(d->bd_mtx);
 	}
 
@@ -1601,12 +1618,18 @@ filt_bpfread(struct knote *kn, long hint
 	 */
 	d->bd_pid = curproc->p_pid;
 
-	mutex_enter(d->bd_buf_mtx);
+	if (hint & NOTE_SUBMIT)
+		KASSERT(mutex_owned(d->bd_buf_mtx));
+	else
+		mutex_enter(d->bd_buf_mtx);
 	kn->kn_data = d->bd_hlen;
 	if (d->bd_immediate)
 		kn->kn_data += d->bd_slen;
 	rv = (kn->kn_data > 0);
-	mutex_exit(d->bd_buf_mtx);
+	if (hint & NOTE_SUBMIT)
+		KASSERT(mutex_owned(d->bd_buf_mtx));
+	else
+		mutex_exit(d->bd_buf_mtx);
 	return rv;
 }
 
@@ -2094,7 +2117,6 @@ catchpacket(struct bpf_d *d, u_char *pkt
 	 */
 	(*cpfn)(h + hdrlen, pkt, caplen);
 	d->bd_slen = curlen + totlen;
-	mutex_exit(d->bd_buf_mtx);
 
 	/*
 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
@@ -2102,6 +2124,8 @@ catchpacket(struct bpf_d *d, u_char *pkt
 	 */
 	if (do_wakeup)
 		bpf_wakeup(d);
+
+	mutex_exit(d->bd_buf_mtx);
 }
 
 /*

Index: src/sys/net/bpfdesc.h
diff -u src/sys/net/bpfdesc.h:1.48 src/sys/net/bpfdesc.h:1.48.10.1
--- src/sys/net/bpfdesc.h:1.48	Wed Jun  9 15:44:15 2021
+++ src/sys/net/bpfdesc.h	Fri Sep 13 14:13:05 2024
@@ -1,4 +1,4 @@
-/*	$NetBSD: bpfdesc.h,v 1.48 2021/06/09 15:44:15 martin Exp $	*/
+/*	$NetBSD: bpfdesc.h,v 1.48.10.1 2024/09/13 14:13:05 martin Exp $	*/
 
 /*
  * Copyright (c) 1990, 1991, 1993
@@ -113,7 +113,6 @@ struct bpf_d {
 	pid_t		bd_pid;		/* corresponding PID */
 	/* DEPRECATED. Keep it to avoid breaking kvm(3) users */
 	LIST_ENTRY(bpf_d) _bd_list;	/* list of all BPF's */
-	/* DEPRECATED. Keep it to avoid breaking kvm(3) users */
 	void		*bd_sih;	/* soft interrupt handle */
 	struct timespec bd_atime;	/* access time */
 	struct timespec bd_mtime;	/* modification time */
@@ -130,7 +129,7 @@ struct bpf_d {
 	struct pslist_entry	bd_bif_dlist_entry; /* For bpf_if */
 	struct pslist_entry	bd_bpf_dlist_entry; /* For the global list */
 	kmutex_t	*bd_mtx;
-	kmutex_t	*bd_buf_mtx;
+	kmutex_t	*bd_buf_mtx;	/* For buffers, bd_state, bd_sel and bd_cv */
 	kcondvar_t	bd_cv;
 #endif
 };

Reply via email to