rds: Handle unsupported rdma request to fs dax memory

RDS doesn't support RDMA on memory apertures that require On Demand
Paging (ODP), such as FS DAX memory. User applications can try to use
RDS to perform RDMA over such memories and since it doesn't report any
failure, it can lead to unexpected issues like memory corruption because
of other processes can issue file operations like ftruncate while
memory is registered for RDMA.

The patch adds a check so that such an attempt to RDMA to/from memory
apertures requiring ODP will fail. Using 'get_user_pages_longterm'
instead of 'get_user_pages_fast' when allocating memory gives us
checking for FS DAX memory.  'get_user_pages_longterm' will fail with
errorcode -ENOTSUPP if FS DAX memory is detected.

Orabug: 30731395

Signed-off-by: Hans Westgaard Ry <[email protected]>
Reviewed-by: Santosh Shilimkar <[email protected]>
Signed-off-by: Somasundaram Krishnasamy <[email protected]>

Orabug: 30820081

UEK5 => UEK6

(cherry picked from commit 09ad94c)
cherry-pick-repo=UEK/production/linux-uek.git

Conflicts:
	net/rds/rdma.c
	  No merge conflict, but Linux-5.4 compatibility:
	  Adjusted to account for API change:
	  932f4a6 ("mm/gup: replace get_user_pages_longterm() with FOLL_LONGTERM")

Signed-off-by: Gerd Rausch <[email protected]>
Reviewed-by: Sharath Srinivasan <[email protected]>

Orabug: 33590097

UEK6 => UEK7

(cherry picked from commit 6aa3c12)
cherry-pick-repo=UEK/production/linux-uek.git

Signed-off-by: Gerd Rausch <[email protected]>
Reviewed-by: William Kucharski <[email protected]>

Author: Hans Westgaard Ry

net/rds/rdma.c

@@ -33,7 +33,7 @@
 #include <linux/pagemap.h>
 #include <linux/rbtree.h>
 #include <linux/dma-mapping.h> /* for DMA_*_DEVICE */
-
+#include <linux/sched/mm.h>
 #include "rds.h"
 
 /*
@@ -163,14 +163,20 @@ static int rds_pin_pages(unsigned long user_addr, unsigned int nr_pages,
 			struct page **pages, int write)
 {
 	int ret;
+	struct mm_struct *mm = current->mm;
+	int gup_flags = FOLL_LONGTERM | (write ? FOLL_WRITE : 0);
 
-	ret = get_user_pages_fast(user_addr, nr_pages, write, pages);
+	mmgrab(mm);
+	down_read(&mm->mmap_sem);
+	ret = get_user_pages(user_addr, nr_pages, gup_flags, pages, NULL);
 
 	if (ret >= 0 && (unsigned) ret < nr_pages) {
 		while (ret--)
 			put_page(pages[ret]);
 		ret = -EFAULT;
 	}
+	up_read(&mm->mmap_sem);
+	mmdrop(mm);
 
 	return ret;
 }