MFC r283422:

Refund the proc emuldata struct for future use. For now move flags from
thread emuldata to proc emuldata as it was originally intended.

As we can have both 64 & 32 bit Linuxulator running any eventhandler
can be called twice for us. To prevent this move eventhandlers code
from linux_emul.c to the linux_common.ko module.

6 files changed, 160 insertions, 173 deletions

diff --git a/sys/compat/linux/linux_common.c b/sys/compat/linux/linux_common.c
index 5427b61..3cee92a 100644
--- a/sys/compat/linux/linux_common.c
+++ b/sys/compat/linux/linux_common.c
@@ -28,13 +28,15 @@
 __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
-#include <sys/module.h>
-#include <sys/malloc.h>
-#include <sys/module.h>
-#include <sys/types.h>
+#include <sys/systm.h>
+#include <sys/exec.h>
+#include <sys/imgact.h>
+#include <sys/imgact_elf.h>
 #include <sys/kernel.h>
-#include <sys/proc.h>
+#include <sys/malloc.h>
+#include <sys/eventhandler.h>
 
+#include <compat/linux/linux_emul.h>
 #include <compat/linux/linux_mib.h>
 #include <compat/linux/linux_util.h>
 
@@ -42,6 +44,10 @@ MODULE_VERSION(linux_common, 1);
 
 SET_DECLARE(linux_device_handler_set, struct linux_device_handler);
 
+static eventhandler_tag linux_exec_tag;
+static eventhandler_tag linux_thread_dtor_tag;
+static eventhandler_tag	linux_exit_tag;
+
 
 static int
 linux_common_modevent(module_t mod, int type, void *data)
@@ -51,6 +57,12 @@ linux_common_modevent(module_t mod, int type, void *data)
 	switch(type) {
 	case MOD_LOAD:
 		linux_osd_jail_register();
+		linux_exit_tag = EVENTHANDLER_REGISTER(process_exit,
+		    linux_proc_exit, NULL, 1000);
+		linux_exec_tag = EVENTHANDLER_REGISTER(process_exec,
+		    linux_proc_exec, NULL, 1000);
+		linux_thread_dtor_tag = EVENTHANDLER_REGISTER(thread_dtor,
+		    linux_thread_dtor, NULL, EVENTHANDLER_PRI_ANY);
 		SET_FOREACH(ldhp, linux_device_handler_set)
 			linux_device_register_handler(*ldhp);
 		break;
@@ -58,6 +70,9 @@ linux_common_modevent(module_t mod, int type, void *data)
 		linux_osd_jail_deregister();
 		SET_FOREACH(ldhp, linux_device_handler_set)
 			linux_device_unregister_handler(*ldhp);
+		EVENTHANDLER_DEREGISTER(process_exit, linux_exit_tag);
+		EVENTHANDLER_DEREGISTER(process_exec, linux_exec_tag);
+		EVENTHANDLER_DEREGISTER(thread_dtor, linux_thread_dtor_tag);
 		break;
 	default:
 		return (EOPNOTSUPP);
diff --git a/sys/compat/linux/linux_emul.c b/sys/compat/linux/linux_emul.c
index aaead8d..b9f2373 100644
--- a/sys/compat/linux/linux_emul.c
+++ b/sys/compat/linux/linux_emul.c
@@ -30,9 +30,6 @@
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
 
-#include "opt_compat.h"
-#include "opt_kdtrace.h"
-
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/imgact.h>
@@ -41,7 +38,6 @@ __FBSDID("$FreeBSD$");
 #include <sys/lock.h>
 #include <sys/malloc.h>
 #include <sys/mutex.h>
-#include <sys/sdt.h>
 #include <sys/sx.h>
 #include <sys/proc.h>
 #include <sys/syscallsubr.h>
@@ -49,60 +45,13 @@ __FBSDID("$FreeBSD$");
 #include <sys/sysproto.h>
 #include <sys/unistd.h>
 
-#ifdef COMPAT_LINUX32
-#include <machine/../linux32/linux.h>
-#include <machine/../linux32/linux32_proto.h>
-#else
-#include <machine/../linux/linux.h>
-#include <machine/../linux/linux_proto.h>
-#endif
-
-#include <compat/linux/linux_dtrace.h>
 #include <compat/linux/linux_emul.h>
-#include <compat/linux/linux_futex.h>
 #include <compat/linux/linux_misc.h>
 #include <compat/linux/linux_util.h>
 
-/**
- * Special DTrace provider for the linuxulator.
- *
- * In this file we define the provider for the entire linuxulator. All
- * modules (= files of the linuxulator) use it.
- *
- * We define a different name depending on the emulated bitsize, see
- * ../../<ARCH>/linux{,32}/linux.h, e.g.:
- *      native bitsize          = linuxulator
- *      amd64, 32bit emulation  = linuxulator32
- */
-LIN_SDT_PROVIDER_DEFINE(LINUX_DTRACE);
-
-/**
- * DTrace probes in this module.
- */
-LIN_SDT_PROBE_DEFINE1(emul, em_find, entry, "struct thread *");
-LIN_SDT_PROBE_DEFINE0(emul, em_find, return);
-LIN_SDT_PROBE_DEFINE3(emul, proc_init, entry, "struct thread *",
-    "struct thread *", "int");
-LIN_SDT_PROBE_DEFINE0(emul, proc_init, create_thread);
-LIN_SDT_PROBE_DEFINE0(emul, proc_init, fork);
-LIN_SDT_PROBE_DEFINE0(emul, proc_init, exec);
-LIN_SDT_PROBE_DEFINE0(emul, proc_init, return);
-LIN_SDT_PROBE_DEFINE1(emul, proc_exit, entry, "struct proc *");
-LIN_SDT_PROBE_DEFINE1(emul, linux_thread_detach, entry, "struct thread *");
-LIN_SDT_PROBE_DEFINE0(emul, linux_thread_detach, futex_failed);
-LIN_SDT_PROBE_DEFINE1(emul, linux_thread_detach, child_clear_tid_error, "int");
-LIN_SDT_PROBE_DEFINE0(emul, linux_thread_detach, return);
-LIN_SDT_PROBE_DEFINE2(emul, proc_exec, entry, "struct proc *",
-    "struct image_params *");
-LIN_SDT_PROBE_DEFINE0(emul, proc_exec, return);
-LIN_SDT_PROBE_DEFINE0(emul, linux_schedtail, entry);
-LIN_SDT_PROBE_DEFINE1(emul, linux_schedtail, copyout_error, "int");
-LIN_SDT_PROBE_DEFINE0(emul, linux_schedtail, return);
-LIN_SDT_PROBE_DEFINE1(emul, linux_set_tid_address, entry, "int *");
-LIN_SDT_PROBE_DEFINE0(emul, linux_set_tid_address, return);
 
 /*
- * This returns reference to the emuldata entry (if found)
+ * This returns reference to the thread emuldata entry (if found)
  *
  * Hold PROC_LOCK when referencing emuldata from other threads.
  */
@@ -111,41 +60,51 @@ em_find(struct thread *td)
 {
 	struct linux_emuldata *em;
 
-	LIN_SDT_PROBE1(emul, em_find, entry, td);
-
 	em = td->td_emuldata;
 
-	LIN_SDT_PROBE1(emul, em_find, return, em);
-
 	return (em);
 }
 
+/*
+ * This returns reference to the proc pemuldata entry (if found)
+ *
+ * Hold PROC_LOCK when referencing proc pemuldata from other threads.
+ * Hold LINUX_PEM_LOCK wher referencing pemuldata members.
+ */
+struct linux_pemuldata *
+pem_find(struct proc *p)
+{
+	struct linux_pemuldata *pem;
+
+	pem = p->p_emuldata;
+
+	return (pem);
+}
+
 void
 linux_proc_init(struct thread *td, struct thread *newtd, int flags)
 {
 	struct linux_emuldata *em;
-
-	LIN_SDT_PROBE3(emul, proc_init, entry, td, newtd, flags);
+	struct linux_pemuldata *pem;
 
 	if (newtd != NULL) {
 		/* non-exec call */
 		em = malloc(sizeof(*em), M_TEMP, M_WAITOK | M_ZERO);
 		em->pdeath_signal = 0;
-		em->flags = 0;
 		em->robust_futexes = NULL;
 		if (flags & LINUX_CLONE_THREAD) {
-			LIN_SDT_PROBE0(emul, proc_init, create_thread);
-
 			em->em_tid = newtd->td_tid;
 		} else {
-			LIN_SDT_PROBE0(emul, proc_init, fork);
 
 			em->em_tid = newtd->td_proc->p_pid;
+
+			pem = malloc(sizeof(*pem), M_TEMP, M_WAITOK | M_ZERO);
+			sx_init(&pem->pem_sx, "lpemlk");
+			newtd->td_proc->p_emuldata = pem;
 		}
 		newtd->td_emuldata = em;
 	} else {
 		/* exec */
-		LIN_SDT_PROBE0(emul, proc_init, exec);
 
 		/* lookup the old one */
 		em = em_find(td);
@@ -156,24 +115,32 @@ linux_proc_init(struct thread *td, struct thread *newtd, int flags)
 
 	em->child_clear_tid = NULL;
 	em->child_set_tid = NULL;
-
-	LIN_SDT_PROBE0(emul, proc_init, return);
 }
 
 void 
 linux_proc_exit(void *arg __unused, struct proc *p)
 {
+	struct linux_pemuldata *pem;
 	struct thread *td = curthread;
 
-	if (__predict_false(SV_CURPROC_ABI() != SV_ABI_LINUX)) {
-		LIN_SDT_PROBE1(emul, proc_exit, entry, p);
-		(p->p_sysent->sv_thread_detach)(td);
-	}
+	if (__predict_false(SV_CURPROC_ABI() != SV_ABI_LINUX))
+		return;
+
+	pem = pem_find(p);
+	if (pem == NULL)
+		return;	
+	(p->p_sysent->sv_thread_detach)(td);
+
+	p->p_emuldata = NULL;
+
+	sx_destroy(&pem->pem_sx);
+	free(pem, M_TEMP);
 }
 
 int 
 linux_common_execve(struct thread *td, struct image_args *eargs)
 {
+	struct linux_pemuldata *pem;
 	struct linux_emuldata *em;
 	struct proc *p;
 	int error;
@@ -200,16 +167,21 @@ linux_common_execve(struct thread *td, struct image_args *eargs)
 
 	/*
 	 * In a case of transition from Linux binary execing to
-	 * FreeBSD binary we destroy linux emuldata thread entry.
+	 * FreeBSD binary we destroy linux emuldata thread & proc entries.
 	 */
 	if (SV_CURPROC_ABI() != SV_ABI_LINUX) {
 		PROC_LOCK(p);
 		em = em_find(td);
-		KASSERT(em != NULL, ("proc_exec: emuldata not found.\n"));
+		KASSERT(em != NULL, ("proc_exec: thread emuldata not found.\n"));
 		td->td_emuldata = NULL;
+
+		pem = pem_find(p);
+		KASSERT(pem != NULL, ("proc_exec: proc pemuldata not found.\n"));
+		p->p_emuldata = NULL;
 		PROC_UNLOCK(p);
 
 		free(em, M_TEMP);
+		free(pem, M_TEMP);
 	}
 	return (0);
 }
@@ -225,68 +197,11 @@ linux_proc_exec(void *arg __unused, struct proc *p, struct image_params *imgp)
 	 */
 	if (__predict_false((imgp->sysent->sv_flags & SV_ABI_MASK) ==
 	    SV_ABI_LINUX)) {
-		LIN_SDT_PROBE2(emul, proc_exec, entry, p, imgp);
 		if (SV_PROC_ABI(p) == SV_ABI_LINUX)
 			linux_proc_init(td, NULL, 0);
 		else
 			linux_proc_init(td, td, 0);
-
-		LIN_SDT_PROBE0(emul, proc_exec, return);
-	}
-}
-
-void
-linux_thread_detach(struct thread *td)
-{
-	struct linux_sys_futex_args cup;
-	struct linux_emuldata *em;
-	int *child_clear_tid;
-	int null = 0;
-	int error;
-
-	LIN_SDT_PROBE1(emul, linux_thread_detach, entry, td);
-
-	em = em_find(td);
-	KASSERT(em != NULL, ("thread_detach: emuldata not found.\n"));
-
-	LINUX_CTR1(exit, "thread detach(%d)", em->em_tid);
-
-	release_futexes(td, em);
-
-	child_clear_tid = em->child_clear_tid;
-
-	if (child_clear_tid != NULL) {
-
-		LINUX_CTR2(exit, "thread detach(%d) %p",
-		    em->em_tid, child_clear_tid);
-	
-		error = copyout(&null, child_clear_tid, sizeof(null));
-		if (error) {
-			LIN_SDT_PROBE1(emul, linux_thread_detach,
-			    child_clear_tid_error, error);
-
-			LIN_SDT_PROBE0(emul, linux_thread_detach, return);
-			return;
-		}
-
-		cup.uaddr = child_clear_tid;
-		cup.op = LINUX_FUTEX_WAKE;
-		cup.val = 1;		/* wake one */
-		cup.timeout = NULL;
-		cup.uaddr2 = NULL;
-		cup.val3 = 0;
-		error = linux_sys_futex(td, &cup);
-		/*
-		 * this cannot happen at the moment and if this happens it
-		 * probably means there is a user space bug
-		 */
-		if (error) {
-			LIN_SDT_PROBE0(emul, linux_thread_detach, futex_failed);
-			printf(LMSG("futex stuff in thread_detach failed.\n"));
-		}
 	}
-
-	LIN_SDT_PROBE0(emul, linux_thread_detach, return);
 }
 
 void
@@ -312,12 +227,10 @@ linux_schedtail(struct thread *td)
 	int error = 0;
 	int *child_set_tid;
 
-	LIN_SDT_PROBE1(emul, linux_schedtail, entry, td);
-
 	p = td->td_proc;
 
 	em = em_find(td);
-	KASSERT(em != NULL, ("linux_schedtail: emuldata not found.\n"));
+	KASSERT(em != NULL, ("linux_schedtail: thread emuldata not found.\n"));
 	child_set_tid = em->child_set_tid;
 
 	if (child_set_tid != NULL) {
@@ -325,34 +238,6 @@ linux_schedtail(struct thread *td)
 		    sizeof(em->em_tid));
 		LINUX_CTR4(clone, "schedtail(%d) %p stored %d error %d",
 		    td->td_tid, child_set_tid, em->em_tid, error);
-
-		if (error != 0) {
-			LIN_SDT_PROBE1(emul, linux_schedtail, copyout_error,
-			    error);
-		}
 	} else
 		LINUX_CTR1(clone, "schedtail(%d)", em->em_tid);
-
-	LIN_SDT_PROBE0(emul, linux_schedtail, return);
-}
-
-int
-linux_set_tid_address(struct thread *td, struct linux_set_tid_address_args *args)
-{
-	struct linux_emuldata *em;
-
-	LIN_SDT_PROBE1(emul, linux_set_tid_address, entry, args->tidptr);
-
-	em = em_find(td);
-	KASSERT(em != NULL, ("set_tid_address: emuldata not found.\n"));
-
-	em->child_clear_tid = args->tidptr;
-
-	td->td_retval[0] = em->em_tid;
-
-	LINUX_CTR3(set_tid_address, "tidptr(%d) %p, returns %d",
-	    em->em_tid, args->tidptr, td->td_retval[0]);
-
-	LIN_SDT_PROBE0(emul, linux_set_tid_address, return);
-	return (0);
 }
diff --git a/sys/compat/linux/linux_emul.h b/sys/compat/linux/linux_emul.h
index ae482f9..2178701 100644
--- a/sys/compat/linux/linux_emul.h
+++ b/sys/compat/linux/linux_emul.h
@@ -41,7 +41,7 @@ struct linux_emuldata {
 	int    *child_clear_tid;/* in clone(): Child's TID to clear on exit */
 
 	int	pdeath_signal;		/* parent death signal */
-	int	flags;			/* different emuldata flags */
+	int	flags;			/* thread emuldata flags */
 	int	em_tid;			/* thread id */
 
 	struct	linux_robust_list_head	*robust_futexes;
@@ -49,10 +49,6 @@ struct linux_emuldata {
 
 struct linux_emuldata	*em_find(struct thread *);
 
-/* emuldata flags */
-#define	LINUX_XDEPR_REQUEUEOP	0x00000001	/* uses deprecated
-						   futex REQUEUE op*/
-
 void	linux_proc_init(struct thread *, struct thread *, int);
 void	linux_proc_exit(void *, struct proc *);
 void	linux_schedtail(struct thread *);
@@ -61,4 +57,19 @@ void	linux_thread_dtor(void *arg __unused, struct thread *);
 void	linux_thread_detach(struct thread *);
 int	linux_common_execve(struct thread *, struct image_args *);
 
+/* process emuldata flags */
+#define	LINUX_XDEPR_REQUEUEOP	0x00000001	/* uses deprecated
+						   futex REQUEUE op*/
+struct linux_pemuldata {
+	uint32_t	flags;		/* process emuldata flags */
+	struct sx	pem_sx;		/* lock for this struct */
+};
+
+#define	LINUX_PEM_XLOCK(p)	sx_xlock(&(p)->pem_sx)
+#define	LINUX_PEM_XUNLOCK(p)	sx_xunlock(&(p)->pem_sx)
+#define	LINUX_PEM_SLOCK(p)	sx_slock(&(p)->pem_sx)
+#define	LINUX_PEM_SUNLOCK(p)	sx_sunlock(&(p)->pem_sx)
+
+struct linux_pemuldata	*pem_find(struct proc *);
+
 #endif	/* !_LINUX_EMUL_H_ */
diff --git a/sys/compat/linux/linux_fork.c b/sys/compat/linux/linux_fork.c
index bffe468..f74c9a2 100644
--- a/sys/compat/linux/linux_fork.c
+++ b/sys/compat/linux/linux_fork.c
@@ -59,6 +59,7 @@ __FBSDID("$FreeBSD$");
 #endif
 #include <compat/linux/linux_signal.h>
 #include <compat/linux/linux_emul.h>
+#include <compat/linux/linux_futex.h>
 #include <compat/linux/linux_misc.h>
 #include <compat/linux/linux_util.h>
 
@@ -405,3 +406,63 @@ linux_exit(struct thread *td, struct linux_exit_args *args)
 	exit1(td, W_EXITCODE(args->rval, 0));
 		/* NOTREACHED */
 }
+
+int
+linux_set_tid_address(struct thread *td, struct linux_set_tid_address_args *args)
+{
+	struct linux_emuldata *em;
+
+	em = em_find(td);
+	KASSERT(em != NULL, ("set_tid_address: emuldata not found.\n"));
+
+	em->child_clear_tid = args->tidptr;
+
+	td->td_retval[0] = em->em_tid;
+
+	LINUX_CTR3(set_tid_address, "tidptr(%d) %p, returns %d",
+	    em->em_tid, args->tidptr, td->td_retval[0]);
+
+	return (0);
+}
+
+void
+linux_thread_detach(struct thread *td)
+{
+	struct linux_sys_futex_args cup;
+	struct linux_emuldata *em;
+	int *child_clear_tid;
+	int error;
+
+	em = em_find(td);
+	KASSERT(em != NULL, ("thread_detach: emuldata not found.\n"));
+
+	LINUX_CTR1(exit, "thread detach(%d)", em->em_tid);
+
+	release_futexes(td, em);
+
+	child_clear_tid = em->child_clear_tid;
+
+	if (child_clear_tid != NULL) {
+
+		LINUX_CTR2(exit, "thread detach(%d) %p",
+		    em->em_tid, child_clear_tid);
+	
+		error = suword32(child_clear_tid, 0);
+		if (error != 0)
+			return;
+
+		cup.uaddr = child_clear_tid;
+		cup.op = LINUX_FUTEX_WAKE;
+		cup.val = 1;		/* wake one */
+		cup.timeout = NULL;
+		cup.uaddr2 = NULL;
+		cup.val3 = 0;
+		error = linux_sys_futex(td, &cup);
+		/*
+		 * this cannot happen at the moment and if this happens it
+		 * probably means there is a user space bug
+		 */
+		if (error != 0)
+			linux_msg(td, "futex stuff in thread_detach failed.");
+	}
+}
diff --git a/sys/compat/linux/linux_futex.c b/sys/compat/linux/linux_futex.c
index 5260d9e..b3d175e 100644
--- a/sys/compat/linux/linux_futex.c
+++ b/sys/compat/linux/linux_futex.c
@@ -654,7 +654,7 @@ int
 linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args)
 {
 	int clockrt, nrwake, op_ret, ret;
-	struct linux_emuldata *em;
+	struct linux_pemuldata *pem;
 	struct waiting_proc *wp;
 	struct futex *f, *f2;
 	struct l_timespec timeout;
@@ -974,12 +974,12 @@ linux_sys_futex(struct thread *td, struct linux_sys_futex_args *args)
 		 * Glibc versions prior to 2.3.3 fall back to FUTEX_WAKE when
 		 * FUTEX_REQUEUE returned EINVAL.
 		 */
-		em = em_find(td);
-		if ((em->flags & LINUX_XDEPR_REQUEUEOP) == 0) {
+		pem = pem_find(td->td_proc);
+		if ((pem->flags & LINUX_XDEPR_REQUEUEOP) == 0) {
 			linux_msg(td,
 				  "linux_sys_futex: "
 				  "unsupported futex_requeue op\n");
-			em->flags |= LINUX_XDEPR_REQUEUEOP;
+			pem->flags |= LINUX_XDEPR_REQUEUEOP;
 			LIN_SDT_PROBE0(futex, linux_sys_futex,
 			    deprecated_requeue);
 		}
diff --git a/sys/compat/linux/linux_misc.c b/sys/compat/linux/linux_misc.c
index 550e36d..2106dcd 100644
--- a/sys/compat/linux/linux_misc.c
+++ b/sys/compat/linux/linux_misc.c
@@ -54,6 +54,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/racct.h>
 #include <sys/resourcevar.h>
 #include <sys/sched.h>
+#include <sys/sdt.h>
 #include <sys/signalvar.h>
 #include <sys/stat.h>
 #include <sys/syscallsubr.h>
@@ -84,6 +85,7 @@ __FBSDID("$FreeBSD$");
 #include <machine/../linux/linux_proto.h>
 #endif
 
+#include <compat/linux/linux_dtrace.h>
 #include <compat/linux/linux_file.h>
 #include <compat/linux/linux_mib.h>
 #include <compat/linux/linux_signal.h>
@@ -92,6 +94,19 @@ __FBSDID("$FreeBSD$");
 #include <compat/linux/linux_emul.h>
 #include <compat/linux/linux_misc.h>
 
+/**
+ * Special DTrace provider for the linuxulator.
+ *
+ * In this file we define the provider for the entire linuxulator. All
+ * modules (= files of the linuxulator) use it.
+ *
+ * We define a different name depending on the emulated bitsize, see
+ * ../../<ARCH>/linux{,32}/linux.h, e.g.:
+ *      native bitsize          = linuxulator
+ *      amd64, 32bit emulation  = linuxulator32
+ */
+LIN_SDT_PROVIDER_DEFINE(LINUX_DTRACE);
+
 int stclohz;				/* Statistics clock frequency */
 
 static unsigned int linux_to_bsd_resource[LINUX_RLIM_NLIMITS] = {