MFC r329825:

Return correct error code to user-space when a system call receives a
signal in the LinuxKPI.

The read(), write() and mmap() system calls can return either EINTR or
ERESTART upon receiving a signal. Add code to figure out the correct
return value by temporarily storing the return code from the relevant
FreeBSD kernel APIs in the Linux task structure.

Sponsored by:	Mellanox Technologies

6 files changed, 132 insertions, 47 deletions

diff --git a/sys/compat/linuxkpi/common/include/linux/mutex.h b/sys/compat/linuxkpi/common/include/linux/mutex.h
index 8adbba6..bbf6023 100644
--- a/sys/compat/linuxkpi/common/include/linux/mutex.h
+++ b/sys/compat/linuxkpi/common/include/linux/mutex.h
@@ -63,7 +63,7 @@ typedef struct mutex {
 
 #define	mutex_lock_interruptible(_m) ({		\
 	MUTEX_SKIP() ? 0 :			\
-	(sx_xlock_sig(&(_m)->sx) ? -EINTR : 0);	\
+	linux_mutex_lock_interruptible(_m);	\
 })
 
 #define	mutex_unlock(_m) do {			\
@@ -143,4 +143,6 @@ linux_mutex_destroy(mutex_t *m)
 	sx_destroy(&m->sx);
 }
 
+extern int linux_mutex_lock_interruptible(mutex_t *m);
+
 #endif					/* _LINUX_MUTEX_H_ */
diff --git a/sys/compat/linuxkpi/common/include/linux/rwsem.h b/sys/compat/linuxkpi/common/include/linux/rwsem.h
index 3042dcf..34e51c1 100644
--- a/sys/compat/linuxkpi/common/include/linux/rwsem.h
+++ b/sys/compat/linuxkpi/common/include/linux/rwsem.h
@@ -47,7 +47,7 @@ struct rw_semaphore {
 #define	up_read(_rw)			sx_sunlock(&(_rw)->sx)
 #define	down_read_trylock(_rw)		!!sx_try_slock(&(_rw)->sx)
 #define	down_write_trylock(_rw)		!!sx_try_xlock(&(_rw)->sx)
-#define	down_write_killable(_rw)	!!sx_xlock_sig(&(_rw)->sx)
+#define	down_write_killable(_rw)	linux_down_write_killable(_rw)
 #define	downgrade_write(_rw)		sx_downgrade(&(_rw)->sx)
 #define	down_read_nested(_rw, _sc)	down_read(_rw)
 #define	init_rwsem(_rw)			linux_init_rwsem(_rw, rwsem_name("lnxrwsem"))
@@ -79,4 +79,6 @@ linux_init_rwsem(struct rw_semaphore *rw, const char *name)
 	sx_init_flags(&rw->sx, name, SX_NOWITNESS);
 }
 
+extern int linux_down_write_killable(struct rw_semaphore *);
+
 #endif					/* _LINUX_RWSEM_H_ */
diff --git a/sys/compat/linuxkpi/common/include/linux/sched.h b/sys/compat/linuxkpi/common/include/linux/sched.h
index 817e16c..b24de2c 100644
--- a/sys/compat/linuxkpi/common/include/linux/sched.h
+++ b/sys/compat/linuxkpi/common/include/linux/sched.h
@@ -2,7 +2,7 @@
  * Copyright (c) 2010 Isilon Systems, Inc.
  * Copyright (c) 2010 iX Systems, Inc.
  * Copyright (c) 2010 Panasas, Inc.
- * Copyright (c) 2013-2017 Mellanox Technologies, Ltd.
+ * Copyright (c) 2013-2018 Mellanox Technologies, Ltd.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -77,6 +77,7 @@ struct task_struct {
 	struct completion exited;
 	TAILQ_ENTRY(task_struct) rcu_entry;
 	int rcu_recurse;
+	int bsd_interrupt_value;
 };
 
 #define	current	({ \
@@ -127,12 +128,26 @@ void linux_send_sig(int signo, struct task_struct *task);
 #define	signal_pending_state(state, task)		\
 	linux_signal_pending_state(state, task)
 #define	send_sig(signo, task, priv) do {		\
-	CTASSERT(priv == 0);				\
+	CTASSERT((priv) == 0);				\
 	linux_send_sig(signo, task);			\
 } while (0)
 
 int linux_schedule_timeout(int timeout);
 
+static inline void
+linux_schedule_save_interrupt_value(struct task_struct *task, int value)
+{
+	task->bsd_interrupt_value = value;
+}
+
+static inline int
+linux_schedule_get_interrupt_value(struct task_struct *task)
+{
+	int value = task->bsd_interrupt_value;
+	task->bsd_interrupt_value = 0;
+	return (value);
+}
+
 #define	schedule()					\
 	(void)linux_schedule_timeout(MAX_SCHEDULE_TIMEOUT)
 #define	schedule_timeout(timeout)			\
diff --git a/sys/compat/linuxkpi/common/src/linux_compat.c b/sys/compat/linuxkpi/common/src/linux_compat.c
index ecb1159..b39d718 100644
--- a/sys/compat/linuxkpi/common/src/linux_compat.c
+++ b/sys/compat/linuxkpi/common/src/linux_compat.c
@@ -2,7 +2,7 @@
  * Copyright (c) 2010 Isilon Systems, Inc.
  * Copyright (c) 2010 iX Systems, Inc.
  * Copyright (c) 2010 Panasas, Inc.
- * Copyright (c) 2013-2017 Mellanox Technologies, Ltd.
+ * Copyright (c) 2013-2018 Mellanox Technologies, Ltd.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -828,10 +828,27 @@ linux_access_ok(int rw, const void *uaddr, size_t len)
 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
 }
 
+/*
+ * This function should return either EINTR or ERESTART depending on
+ * the signal type sent to this thread:
+ */
+static int
+linux_get_error(struct task_struct *task, int error)
+{
+	/* check for signal type interrupt code */
+	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
+		error = -linux_schedule_get_interrupt_value(task);
+		if (error == 0)
+			error = EINTR;
+	}
+	return (error);
+}
+
 static int
 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
     u_long cmd, caddr_t data, struct thread *td)
 {
+	struct task_struct *task = current;
 	unsigned size;
 	int error;
 
@@ -844,8 +861,8 @@ linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
 		 * Background: Linux code expects a user-space address
 		 * while FreeBSD supplies a kernel-space address.
 		 */
-		current->bsd_ioctl_data = data;
-		current->bsd_ioctl_len = size;
+		task->bsd_ioctl_data = data;
+		task->bsd_ioctl_len = size;
 		data = (void *)LINUX_IOCTL_MIN_PTR;
 	} else {
 		/* fetch user-space pointer */
@@ -869,16 +886,17 @@ linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
 	else
 		error = ENOTTY;
 	if (size > 0) {
-		current->bsd_ioctl_data = NULL;
-		current->bsd_ioctl_len = 0;
+		task->bsd_ioctl_data = NULL;
+		task->bsd_ioctl_len = 0;
 	}
 
 	if (error == EWOULDBLOCK) {
 		/* update kqfilter status, if any */
 		linux_file_kqfilter_poll(filp,
 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
-	} else if (error == ERESTARTSYS)
-		error = ERESTART;
+	} else {
+		error = linux_get_error(task, error);
+	}
 	return (error);
 }
 
@@ -1111,6 +1129,7 @@ linux_file_mmap_single(struct file *fp, vm_ooffset_t *offset,
     vm_size_t size, struct vm_object **object, int nprot,
     struct thread *td)
 {
+	struct task_struct *task;
 	struct vm_area_struct *vmap;
 	struct mm_struct *mm;
 	struct linux_file *filp;
@@ -1132,7 +1151,8 @@ linux_file_mmap_single(struct file *fp, vm_ooffset_t *offset,
 	 * The atomic reference below makes sure the mm_struct is
 	 * available as long as the vmap is in the linux_vma_head.
 	 */
-	mm = current->mm;
+	task = current;
+	mm = task->mm;
 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
 		return (EINVAL);
 
@@ -1147,11 +1167,10 @@ linux_file_mmap_single(struct file *fp, vm_ooffset_t *offset,
 	vmap->vm_mm = mm;
 
 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
-		error = EINTR;
+		error = linux_get_error(task, EINTR);
 	} else {
 		error = -OPW(fp, td, filp->f_op->mmap(filp, vmap));
-		if (error == ERESTARTSYS)
-			error = ERESTART;
+		error = linux_get_error(task, error);
 		up_write(&vmap->vm_mm->mmap_sem);
 	}
 
@@ -1290,9 +1309,7 @@ linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
 			uio->uio_iov->iov_len -= bytes;
 			uio->uio_resid -= bytes;
 		} else {
-			error = -bytes;
-			if (error == ERESTARTSYS)
-				error = ERESTART;
+			error = linux_get_error(current, -bytes);
 		}
 	} else
 		error = ENXIO;
@@ -1329,9 +1346,7 @@ linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
 			uio->uio_iov->iov_len -= bytes;
 			uio->uio_resid -= bytes;
 		} else {
-			error = -bytes;
-			if (error == ERESTARTSYS)
-				error = ERESTART;
+			error = linux_get_error(current, -bytes);
 		}
 	} else
 		error = ENXIO;
@@ -1780,6 +1795,7 @@ linux_complete_common(struct completion *c, int all)
 int
 linux_wait_for_common(struct completion *c, int flags)
 {
+	struct task_struct *task;
 	int error;
 
 	if (SCHEDULER_STOPPED())
@@ -1787,6 +1803,8 @@ linux_wait_for_common(struct completion *c, int flags)
 
 	DROP_GIANT();
 
+	task = current;
+
 	if (flags != 0)
 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
 	else
@@ -1798,7 +1816,9 @@ linux_wait_for_common(struct completion *c, int flags)
 			break;
 		sleepq_add(c, NULL, "completion", flags, 0);
 		if (flags & SLEEPQ_INTERRUPTIBLE) {
-			if (sleepq_wait_sig(c, 0) != 0) {
+			error = -sleepq_wait_sig(c, 0);
+			if (error != 0) {
+				linux_schedule_save_interrupt_value(task, error);
 				error = -ERESTARTSYS;
 				goto intr;
 			}
@@ -1820,22 +1840,22 @@ intr:
 int
 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
 {
+	struct task_struct *task;
 	int end = jiffies + timeout;
 	int error;
-	int ret;
 
 	if (SCHEDULER_STOPPED())
 		return (0);
 
 	DROP_GIANT();
 
+	task = current;
+
 	if (flags != 0)
 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
 	else
 		flags = SLEEPQ_SLEEP;
 
-	error = 0;
-	ret = 0;
 	for (;;) {
 		sleepq_lock(c);
 		if (c->done)
@@ -1843,26 +1863,30 @@ linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
 		sleepq_add(c, NULL, "completion", flags, 0);
 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
 		if (flags & SLEEPQ_INTERRUPTIBLE)
-			ret = sleepq_timedwait_sig(c, 0);
+			error = -sleepq_timedwait_sig(c, 0);
 		else
-			ret = sleepq_timedwait(c, 0);
-		if (ret != 0) {
-			/* check for timeout or signal */
-			if (ret == EWOULDBLOCK)
-				error = 0;
-			else
+			error = -sleepq_timedwait(c, 0);
+		if (error != 0) {
+			/* check for timeout */
+			if (error == -EWOULDBLOCK) {
+				error = 0;	/* timeout */
+			} else {
+				/* signal happened */
+				linux_schedule_save_interrupt_value(task, error);
 				error = -ERESTARTSYS;
-			goto intr;
+			}
+			goto done;
 		}
 	}
 	c->done--;
 	sleepq_release(c);
 
-intr:
+	/* return how many jiffies are left */
+	error = linux_timer_jiffies_until(end);
+done:
 	PICKUP_GIANT();
 
-	/* return how many jiffies are left */
-	return (ret != 0 ? error : linux_timer_jiffies_until(end));
+	return (error);
 }
 
 int
diff --git a/sys/compat/linuxkpi/common/src/linux_lock.c b/sys/compat/linuxkpi/common/src/linux_lock.c
index ff91514..f037cd3 100644
--- a/sys/compat/linuxkpi/common/src/linux_lock.c
+++ b/sys/compat/linuxkpi/common/src/linux_lock.c
@@ -28,6 +28,7 @@
 
 #include <sys/queue.h>
 
+#include <linux/sched.h>
 #include <linux/ww_mutex.h>
 
 struct ww_mutex_thread {
@@ -72,10 +73,13 @@ linux_ww_unlock(void)
 int
 linux_ww_mutex_lock_sub(struct ww_mutex *lock, int catch_signal)
 {
+	struct task_struct *task;
 	struct ww_mutex_thread entry;
 	struct ww_mutex_thread *other;
 	int retval = 0;
 
+	task = current;
+
 	linux_ww_lock();
 	if (unlikely(sx_try_xlock(&lock->base.sx) == 0)) {
 		entry.thread = curthread;
@@ -105,7 +109,9 @@ linux_ww_mutex_lock_sub(struct ww_mutex *lock, int catch_signal)
 				}
 			}
 			if (catch_signal) {
-				if (cv_wait_sig(&lock->condvar, &ww_mutex_global) != 0) {
+				retval = -cv_wait_sig(&lock->condvar, &ww_mutex_global);
+				if (retval != 0) {
+					linux_schedule_save_interrupt_value(task, retval);
 					retval = -EINTR;
 					goto done;
 				}
@@ -134,3 +140,29 @@ linux_ww_mutex_unlock_sub(struct ww_mutex *lock)
 	cv_signal(&lock->condvar);
 	linux_ww_unlock();
 }
+
+int
+linux_mutex_lock_interruptible(mutex_t *m)
+{
+	int error;
+
+	error = -sx_xlock_sig(&m->sx);
+	if (error != 0) {
+		linux_schedule_save_interrupt_value(current, error);
+		error = -EINTR;
+	}
+	return (error);
+}
+
+int
+linux_down_write_killable(struct rw_semaphore *rw)
+{
+	int error;
+
+	error = -sx_xlock_sig(&rw->sx);
+	if (error != 0) {
+		linux_schedule_save_interrupt_value(current, error);
+		error = -EINTR;
+	}
+	return (error);
+}
diff --git a/sys/compat/linuxkpi/common/src/linux_schedule.c b/sys/compat/linuxkpi/common/src/linux_schedule.c
index dc3dd91..0958b3a 100644
--- a/sys/compat/linuxkpi/common/src/linux_schedule.c
+++ b/sys/compat/linuxkpi/common/src/linux_schedule.c
@@ -41,7 +41,8 @@ __FBSDID("$FreeBSD$");
 #include <linux/wait.h>
 
 static int
-linux_add_to_sleepqueue(void *wchan, const char *wmesg, int timeout, int state)
+linux_add_to_sleepqueue(void *wchan, struct task_struct *task,
+    const char *wmesg, int timeout, int state)
 {
 	int flags, ret;
 
@@ -66,8 +67,10 @@ linux_add_to_sleepqueue(void *wchan, const char *wmesg, int timeout, int state)
 			ret = -sleepq_timedwait(wchan, 0);
 	}
 	/* filter return value */
-	if (ret != 0 && ret != -EWOULDBLOCK)
+	if (ret != 0 && ret != -EWOULDBLOCK) {
+		linux_schedule_save_interrupt_value(task, ret);
 		ret = -ERESTARTSYS;
+	}
 	return (ret);
 }
 
@@ -235,10 +238,10 @@ linux_wait_event_common(wait_queue_head_t *wqh, wait_queue_t *wq, int timeout,
 	PHOLD(task->task_thread->td_proc);
 	sleepq_lock(task);
 	if (atomic_read(&task->state) != TASK_WAKING) {
-		ret = linux_add_to_sleepqueue(task, "wevent", timeout, state);
+		ret = linux_add_to_sleepqueue(task, task, "wevent", timeout, state);
 	} else {
 		sleepq_release(task);
-		ret = linux_signal_pending_state(state, task) ? -ERESTARTSYS : 0;
+		ret = 0;
 	}
 	PRELE(task->task_thread->td_proc);
 
@@ -253,6 +256,7 @@ int
 linux_schedule_timeout(int timeout)
 {
 	struct task_struct *task;
+	int ret;
 	int state;
 	int remainder;
 
@@ -270,10 +274,12 @@ linux_schedule_timeout(int timeout)
 
 	sleepq_lock(task);
 	state = atomic_read(&task->state);
-	if (state != TASK_WAKING)
-		(void)linux_add_to_sleepqueue(task, "sched", timeout, state);
-	else
+	if (state != TASK_WAKING) {
+		ret = linux_add_to_sleepqueue(task, task, "sched", timeout, state);
+	} else {
 		sleepq_release(task);
+		ret = 0;
+	}
 	set_task_state(task, TASK_RUNNING);
 
 	PICKUP_GIANT();
@@ -283,7 +289,11 @@ linux_schedule_timeout(int timeout)
 
 	/* range check return value */
 	remainder -= ticks;
-	if (remainder < 0)
+
+	/* range check return value */
+	if (ret == -ERESTARTSYS && remainder < 1)
+		remainder = 1;
+	else if (remainder < 0)
 		remainder = 0;
 	else if (remainder > timeout)
 		remainder = timeout;
@@ -337,7 +347,7 @@ linux_wait_on_bit_timeout(unsigned long *word, int bit, unsigned int state,
 			break;
 		}
 		set_task_state(task, state);
-		ret = linux_add_to_sleepqueue(wchan, "wbit", timeout, state);
+		ret = linux_add_to_sleepqueue(wchan, task, "wbit", timeout, state);
 		if (ret != 0)
 			break;
 	}
@@ -374,7 +384,7 @@ linux_wait_on_atomic_t(atomic_t *a, unsigned int state)
 			break;
 		}
 		set_task_state(task, state);
-		ret = linux_add_to_sleepqueue(wchan, "watomic", 0, state);
+		ret = linux_add_to_sleepqueue(wchan, task, "watomic", 0, state);
 		if (ret != 0)
 			break;
 	}