IB/ipath: Don't corrupt pending mmap list when unmapped objects are freed

Fix the pending mmap code so it doesn't corrupt the list of pending
mmaps and crash the machine when pending mmaps are destroyed without
first being mapped.  Also, remove an unused variable, and use standard
kernel lists instead of our own homebrewed linked list implementation
to keep the pending mmap list.

Signed-off-by: Robert Walsh <robert.walsh@qlogic.com>
Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

6 files changed, 153 insertions, 90 deletions

diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c
index ea78e6d..4715f89 100644
--- a/drivers/infiniband/hw/ipath/ipath_cq.c
+++ b/drivers/infiniband/hw/ipath/ipath_cq.c
@@ -243,33 +243,21 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
 	 * See ipath_mmap() for details.
 	 */
 	if (udata && udata->outlen >= sizeof(__u64)) {
-		struct ipath_mmap_info *ip;
-		__u64 offset = (__u64) wc;
 		int err;
+		u32 s = sizeof *wc + sizeof(struct ib_wc) * entries;
 
-		err = ib_copy_to_udata(udata, &offset, sizeof(offset));
-		if (err) {
-			ret = ERR_PTR(err);
+		cq->ip = ipath_create_mmap_info(dev, s, context, wc);
+		if (!cq->ip) {
+			ret = ERR_PTR(-ENOMEM);
 			goto bail_wc;
 		}
 
-		/* Allocate info for ipath_mmap(). */
-		ip = kmalloc(sizeof(*ip), GFP_KERNEL);
-		if (!ip) {
-			ret = ERR_PTR(-ENOMEM);
-			goto bail_wc;
+		err = ib_copy_to_udata(udata, &cq->ip->offset,
+				       sizeof(cq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
 		}
-		cq->ip = ip;
-		ip->context = context;
-		ip->obj = wc;
-		kref_init(&ip->ref);
-		ip->mmap_cnt = 0;
-		ip->size = PAGE_ALIGN(sizeof(*wc) +
-				      sizeof(struct ib_wc) * entries);
-		spin_lock_irq(&dev->pending_lock);
-		ip->next = dev->pending_mmaps;
-		dev->pending_mmaps = ip;
-		spin_unlock_irq(&dev->pending_lock);
 	} else
 		cq->ip = NULL;
 
@@ -277,12 +265,18 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
 	if (dev->n_cqs_allocated == ib_ipath_max_cqs) {
 		spin_unlock(&dev->n_cqs_lock);
 		ret = ERR_PTR(-ENOMEM);
-		goto bail_wc;
+		goto bail_ip;
 	}
 
 	dev->n_cqs_allocated++;
 	spin_unlock(&dev->n_cqs_lock);
 
+	if (cq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&cq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
 	/*
 	 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
 	 * The number of entries should be >= the number requested or return
@@ -301,12 +295,12 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries,
 
 	goto done;
 
+bail_ip:
+	kfree(cq->ip);
 bail_wc:
 	vfree(wc);
-
 bail_cq:
 	kfree(cq);
-
 done:
 	return ret;
 }
@@ -443,13 +437,12 @@ int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
 	if (cq->ip) {
 		struct ipath_ibdev *dev = to_idev(ibcq->device);
 		struct ipath_mmap_info *ip = cq->ip;
+		u32 s = sizeof *wc + sizeof(struct ib_wc) * cqe;
 
-		ip->obj = wc;
-		ip->size = PAGE_ALIGN(sizeof(*wc) +
-				      sizeof(struct ib_wc) * cqe);
+		ipath_update_mmap_info(dev, ip, s, wc);
 		spin_lock_irq(&dev->pending_lock);
-		ip->next = dev->pending_mmaps;
-		dev->pending_mmaps = ip;
+		if (list_empty(&ip->pending_mmaps))
+			list_add(&ip->pending_mmaps, &dev->pending_mmaps);
 		spin_unlock_irq(&dev->pending_lock);
 	}
 
diff --git a/drivers/infiniband/hw/ipath/ipath_mmap.c b/drivers/infiniband/hw/ipath/ipath_mmap.c
index a82157d..937bc33 100644
--- a/drivers/infiniband/hw/ipath/ipath_mmap.c
+++ b/drivers/infiniband/hw/ipath/ipath_mmap.c
@@ -46,6 +46,11 @@ void ipath_release_mmap_info(struct kref *ref)
 {
 	struct ipath_mmap_info *ip =
 		container_of(ref, struct ipath_mmap_info, ref);
+	struct ipath_ibdev *dev = to_idev(ip->context->device);
+
+	spin_lock_irq(&dev->pending_lock);
+	list_del(&ip->pending_mmaps);
+	spin_unlock_irq(&dev->pending_lock);
 
 	vfree(ip->obj);
 	kfree(ip);
@@ -60,14 +65,12 @@ static void ipath_vma_open(struct vm_area_struct *vma)
 	struct ipath_mmap_info *ip = vma->vm_private_data;
 
 	kref_get(&ip->ref);
-	ip->mmap_cnt++;
 }
 
 static void ipath_vma_close(struct vm_area_struct *vma)
 {
 	struct ipath_mmap_info *ip = vma->vm_private_data;
 
-	ip->mmap_cnt--;
 	kref_put(&ip->ref, ipath_release_mmap_info);
 }
 
@@ -87,7 +90,7 @@ int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 	struct ipath_ibdev *dev = to_idev(context->device);
 	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 	unsigned long size = vma->vm_end - vma->vm_start;
-	struct ipath_mmap_info *ip, **pp;
+	struct ipath_mmap_info *ip, *pp;
 	int ret = -EINVAL;
 
 	/*
@@ -96,15 +99,16 @@ int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 	 * CQ, QP, or SRQ is soon followed by a call to mmap().
 	 */
 	spin_lock_irq(&dev->pending_lock);
-	for (pp = &dev->pending_mmaps; (ip = *pp); pp = &ip->next) {
+	list_for_each_entry_safe(ip, pp, &dev->pending_mmaps,
+				 pending_mmaps) {
 		/* Only the creator is allowed to mmap the object */
-		if (context != ip->context || (void *) offset != ip->obj)
+		if (context != ip->context || (__u64) offset != ip->offset)
 			continue;
 		/* Don't allow a mmap larger than the object. */
 		if (size > ip->size)
 			break;
 
-		*pp = ip->next;
+		list_del_init(&ip->pending_mmaps);
 		spin_unlock_irq(&dev->pending_lock);
 
 		ret = remap_vmalloc_range(vma, ip->obj, 0);
@@ -119,3 +123,51 @@ int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 done:
 	return ret;
 }
+
+/*
+ * Allocate information for ipath_mmap
+ */
+struct ipath_mmap_info *ipath_create_mmap_info(struct ipath_ibdev *dev,
+					       u32 size,
+					       struct ib_ucontext *context,
+					       void *obj) {
+	struct ipath_mmap_info *ip;
+
+	ip = kmalloc(sizeof *ip, GFP_KERNEL);
+	if (!ip)
+		goto bail;
+
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	INIT_LIST_HEAD(&ip->pending_mmaps);
+	ip->size = size;
+	ip->context = context;
+	ip->obj = obj;
+	kref_init(&ip->ref);
+
+bail:
+	return ip;
+}
+
+void ipath_update_mmap_info(struct ipath_ibdev *dev,
+			    struct ipath_mmap_info *ip,
+			    u32 size, void *obj) {
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	ip->size = size;
+	ip->obj = obj;
+}
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 16db9ac..bfef08e 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -844,34 +844,36 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 	 * See ipath_mmap() for details.
 	 */
 	if (udata && udata->outlen >= sizeof(__u64)) {
-		struct ipath_mmap_info *ip;
-		__u64 offset = (__u64) qp->r_rq.wq;
 		int err;
 
-		err = ib_copy_to_udata(udata, &offset, sizeof(offset));
-		if (err) {
-			ret = ERR_PTR(err);
-			goto bail_rwq;
-		}
+		if (!qp->r_rq.wq) {
+			__u64 offset = 0;
 
-		if (qp->r_rq.wq) {
-			/* Allocate info for ipath_mmap(). */
-			ip = kmalloc(sizeof(*ip), GFP_KERNEL);
-			if (!ip) {
+			err = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_rwq;
+			}
+		} else {
+			u32 s = sizeof(struct ipath_rwq) +
+				qp->r_rq.size * sz;
+
+			qp->ip =
+			    ipath_create_mmap_info(dev, s,
+						   ibpd->uobject->context,
+						   qp->r_rq.wq);
+			if (!qp->ip) {
 				ret = ERR_PTR(-ENOMEM);
 				goto bail_rwq;
 			}
-			qp->ip = ip;
-			ip->context = ibpd->uobject->context;
-			ip->obj = qp->r_rq.wq;
-			kref_init(&ip->ref);
-			ip->mmap_cnt = 0;
-			ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) +
-					      qp->r_rq.size * sz);
-			spin_lock_irq(&dev->pending_lock);
-			ip->next = dev->pending_mmaps;
-			dev->pending_mmaps = ip;
-			spin_unlock_irq(&dev->pending_lock);
+
+			err = ib_copy_to_udata(udata, &(qp->ip->offset),
+					       sizeof(qp->ip->offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_ip;
+			}
 		}
 	}
 
@@ -885,6 +887,12 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 	dev->n_qps_allocated++;
 	spin_unlock(&dev->n_qps_lock);
 
+	if (qp->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&qp->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
 	ret = &qp->ibqp;
 	goto bail;
 
diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c
index 9403350..03acae6 100644
--- a/drivers/infiniband/hw/ipath/ipath_srq.c
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c
@@ -139,33 +139,24 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
 	 * See ipath_mmap() for details.
 	 */
 	if (udata && udata->outlen >= sizeof(__u64)) {
-		struct ipath_mmap_info *ip;
-		__u64 offset = (__u64) srq->rq.wq;
 		int err;
+		u32 s = sizeof(struct ipath_rwq) + srq->rq.size * sz;
 
-		err = ib_copy_to_udata(udata, &offset, sizeof(offset));
-		if (err) {
-			ret = ERR_PTR(err);
+		srq->ip =
+		    ipath_create_mmap_info(dev, s,
+					   ibpd->uobject->context,
+					   srq->rq.wq);
+		if (!srq->ip) {
+			ret = ERR_PTR(-ENOMEM);
 			goto bail_wq;
 		}
 
-		/* Allocate info for ipath_mmap(). */
-		ip = kmalloc(sizeof(*ip), GFP_KERNEL);
-		if (!ip) {
-			ret = ERR_PTR(-ENOMEM);
-			goto bail_wq;
+		err = ib_copy_to_udata(udata, &srq->ip->offset,
+				       sizeof(srq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
 		}
-		srq->ip = ip;
-		ip->context = ibpd->uobject->context;
-		ip->obj = srq->rq.wq;
-		kref_init(&ip->ref);
-		ip->mmap_cnt = 0;
-		ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) +
-				      srq->rq.size * sz);
-		spin_lock_irq(&dev->pending_lock);
-		ip->next = dev->pending_mmaps;
-		dev->pending_mmaps = ip;
-		spin_unlock_irq(&dev->pending_lock);
 	} else
 		srq->ip = NULL;
 
@@ -181,21 +172,27 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
 	if (dev->n_srqs_allocated == ib_ipath_max_srqs) {
 		spin_unlock(&dev->n_srqs_lock);
 		ret = ERR_PTR(-ENOMEM);
-		goto bail_wq;
+		goto bail_ip;
 	}
 
  	dev->n_srqs_allocated++;
 	spin_unlock(&dev->n_srqs_lock);
 
+	if (srq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
 	ret = &srq->ibsrq;
 	goto done;
 
+bail_ip:
+	kfree(srq->ip);
 bail_wq:
 	vfree(srq->rq.wq);
-
 bail_srq:
 	kfree(srq);
-
 done:
 	return ret;
 }
@@ -312,13 +309,13 @@ int ipath_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 		if (srq->ip) {
 			struct ipath_mmap_info *ip = srq->ip;
 			struct ipath_ibdev *dev = to_idev(srq->ibsrq.device);
+			u32 s = sizeof(struct ipath_rwq) + size * sz;
 
-			ip->obj = wq;
-			ip->size = PAGE_ALIGN(sizeof(struct ipath_rwq) +
-					      size * sz);
+			ipath_update_mmap_info(dev, ip, s, wq);
 			spin_lock_irq(&dev->pending_lock);
-			ip->next = dev->pending_mmaps;
-			dev->pending_mmaps = ip;
+			if (list_empty(&ip->pending_mmaps))
+				list_add(&ip->pending_mmaps,
+					 &dev->pending_mmaps);
 			spin_unlock_irq(&dev->pending_lock);
 		}
 	} else if (attr_mask & IB_SRQ_LIMIT) {
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index 18c6df2..b676ea8 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -1476,7 +1476,10 @@ int ipath_register_ib_device(struct ipath_devdata *dd)
 		ret = -ENOMEM;
 		goto err_lk;
 	}
+	INIT_LIST_HEAD(&idev->pending_mmaps);
 	spin_lock_init(&idev->pending_lock);
+	idev->mmap_offset = PAGE_SIZE;
+	spin_lock_init(&idev->mmap_offset_lock);
 	INIT_LIST_HEAD(&idev->pending[0]);
 	INIT_LIST_HEAD(&idev->pending[1]);
 	INIT_LIST_HEAD(&idev->pending[2]);
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h
index c62f9c5..ac66c00 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h
@@ -173,12 +173,12 @@ struct ipath_ah {
  * this as its vm_private_data.
  */
 struct ipath_mmap_info {
-	struct ipath_mmap_info *next;
+	struct list_head pending_mmaps;
 	struct ib_ucontext *context;
 	void *obj;
+	__u64 offset;
 	struct kref ref;
 	unsigned size;
-	unsigned mmap_cnt;
 };
 
 /*
@@ -485,9 +485,10 @@ struct ipath_opcode_stats {
 
 struct ipath_ibdev {
 	struct ib_device ibdev;
-	struct list_head dev_list;
 	struct ipath_devdata *dd;
-	struct ipath_mmap_info *pending_mmaps;
+	struct list_head pending_mmaps;
+	spinlock_t mmap_offset_lock;
+	u32 mmap_offset;
 	int ib_unit;		/* This is the device number */
 	u16 sm_lid;		/* in host order */
 	u8 sm_sl;
@@ -768,6 +769,15 @@ int ipath_dealloc_fmr(struct ib_fmr *ibfmr);
 
 void ipath_release_mmap_info(struct kref *ref);
 
+struct ipath_mmap_info *ipath_create_mmap_info(struct ipath_ibdev *dev,
+					       u32 size,
+					       struct ib_ucontext *context,
+					       void *obj);
+
+void ipath_update_mmap_info(struct ipath_ibdev *dev,
+			    struct ipath_mmap_info *ip,
+			    u32 size, void *obj);
+
 int ipath_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
 
 void ipath_no_bufs_available(struct ipath_qp *qp, struct ipath_ibdev *dev);