[lustre-devel] [PATCH 06/19] lustre: llite: tighten condition for fault not drop mmap_sem

[James Simmons]

From: Bobi Jam <bobijam at whamcloud.com>

As __lock_page_or_retry() indicates, filemap_fault() will return
VM_FAULT_RETRY without releasing mmap_sem iff flags contains
FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT.

So ll_fault0() should pass in FAULT_FLAG_ALLOW_RETRY |
FAULT_FLAG_RETRY_NOWAIT in ll_filemap_fault() so that when it
returns VM_FAULT_RETRY, we can pass on trying normal fault
under DLM lock as mmap_sem is still being held.

While in Linux 5.1 (6b4c9f4469819) FAULT_FLAG_RETRY_NOWAIT is enough
to not drop mmap_sem when failed to lock the page.

Fixes: 22cceab961 ("lustre: llite: Avoid eternel retry loops with MAP_POPULATE")
WC-bug-id: https://jira.whamcloud.com/browse/LU-14713
Lustre-commit: 81aec05103558f57a ("LU-14713 llite: tighten condition for fault not drop mmap_sem")
Signed-off-by: Bobi Jam <bobijam at whamcloud.com>
Reviewed-on: https://review.whamcloud.com/44715
Reviewed-by: Patrick Farrell <pfarrell at whamcloud.com>
Reviewed-by: Neil Brown <neilb at suse.de>
Reviewed-by: Oleg Drokin <green at whamcloud.com>
Signed-off-by: James Simmons <jsimmons at infradead.org>
---
 fs/lustre/llite/llite_mmap.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/fs/lustre/llite/llite_mmap.c b/fs/lustre/llite/llite_mmap.c
index 8047786..0009c5f 100644
--- a/fs/lustre/llite/llite_mmap.c
+++ b/fs/lustre/llite/llite_mmap.c
@@ -285,18 +285,25 @@ static vm_fault_t __ll_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	if (ll_sbi_has_fast_read(ll_i2sbi(inode))) {
 		/* do fast fault */
+		bool allow_retry = vmf->flags & FAULT_FLAG_ALLOW_RETRY;
 		bool has_retry = vmf->flags & FAULT_FLAG_RETRY_NOWAIT;
 
 		/* To avoid loops, instruct downstream to not drop mmap_sem */
-		vmf->flags |= FAULT_FLAG_RETRY_NOWAIT;
+		/**
+		 * only need FAULT_FLAG_ALLOW_RETRY prior to Linux 5.1
+		 * (6b4c9f4469819), where FAULT_FLAG_RETRY_NOWAIT is enough
+		 * to not drop mmap_sem when failed to lock the page.
+		 */
+		vmf->flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
 		ll_cl_add(inode, env, NULL, LCC_MMAP);
 		fault_ret = filemap_fault(vmf);
 		ll_cl_remove(inode, env);
 		if (has_retry)
 			vmf->flags &= ~FAULT_FLAG_RETRY_NOWAIT;
+		if (!allow_retry)
+			vmf->flags &= ~FAULT_FLAG_ALLOW_RETRY;
 
-		/*
-		 * - If there is no error, then the page was found in cache and
+		/* - If there is no error, then the page was found in cache and
 		 *   uptodate;
 		 * - If VM_FAULT_RETRY is set, the page existed but failed to
 		 *   lock. We will try slow path to avoid loops.
-- 
1.8.3.1