[lustre-devel] [PATCH 01/27] lustre: llite: fix the wrong beyond read end calculation

From: Qian Yingjin <[email protected]>

During the test, we found a dead loop in the read path which
retruns AOP_TRUNCATED_PAGE(0x8001) endless.
The reason is that the calculation of the ending beyond offset is
wrong: (iter->count + iocb->ki_pos).
The ending beyond offset was supposed to be not changed during
the read I/O loop for each page in buffered I/O mode.
However, @iter->count is decreased with read bytes when finished
the read of each page: @iter->count -= read_bytes.

In this patch, we store the ending beyond page index in
@lcc->lcc_end_index before call @generic_file_read_iter into a
loop for each read page and solve this bug.

Fixes: c9f68ebdc6 ("lustre: llite: check read page past requested")
WC-bug-id: https://jira.whamcloud.com/browse/LU-16579
Lustre-commit: ae356dc325877bd13 ("LU-16579 llite: fix the wrong beyond read end calculation")
Signed-off-by: Qian Yingjin <[email protected]>
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/50065
Reviewed-by: Andreas Dilger <[email protected]>
Reviewed-by: Patrick Farrell <[email protected]>
Reviewed-by: Oleg Drokin <[email protected]>
Signed-off-by: James Simmons <[email protected]>
---
 fs/lustre/llite/llite_internal.h |  3 +--
 fs/lustre/llite/rw.c             | 11 +++--------
 fs/lustre/llite/vvp_io.c         |  6 ++----
 3 files changed, 6 insertions(+), 14 deletions(-)

diff --git a/fs/lustre/llite/llite_internal.h b/fs/lustre/llite/llite_internal.h
index 72de8f7..d8eee75 100644
--- a/fs/lustre/llite/llite_internal.h
+++ b/fs/lustre/llite/llite_internal.h
@@ -1375,8 +1375,7 @@ struct ll_cl_context {
 	struct cl_io		*lcc_io;
 	struct cl_page		*lcc_page;
 	enum lcc_type		 lcc_type;
-	struct kiocb		*lcc_iocb;
-	struct iov_iter		*lcc_iter;
+	pgoff_t			 lcc_end_index;
 };
 
 struct ll_thread_info {
diff --git a/fs/lustre/llite/rw.c b/fs/lustre/llite/rw.c
index d285ae1..0c73258 100644
--- a/fs/lustre/llite/rw.c
+++ b/fs/lustre/llite/rw.c
@@ -1863,9 +1863,7 @@ int ll_readpage(struct file *file, struct page *vmpage)
 	struct cl_read_ahead ra = { 0 };
 	struct ll_cl_context *lcc;
 	struct cl_io *io = NULL;
-	struct iov_iter *iter;
 	struct cl_page *page;
-	struct kiocb *iocb;
 	int result;
 
 	if (OBD_FAIL_PRECHECK(OBD_FAIL_LLITE_READPAGE_PAUSE)) {
@@ -1974,11 +1972,8 @@ int ll_readpage(struct file *file, struct page *vmpage)
 	}
 
 	if (lcc && lcc->lcc_type != LCC_MMAP) {
-		iocb = lcc->lcc_iocb;
-		iter = lcc->lcc_iter;
-
-		CDEBUG(D_VFSTRACE, "pgno:%ld, cnt:%ld, pos:%lld\n",
-		       vmpage->index, iter->count, iocb->ki_pos);
+		CDEBUG(D_VFSTRACE, "pgno:%ld, beyond read end_index:%ld\n",
+		       vmpage->index, lcc->lcc_end_index);
 
 		/*
 		 * This handles a kernel bug introduced in kernel 5.12:
@@ -2004,7 +1999,7 @@ int ll_readpage(struct file *file, struct page *vmpage)
 		 * This should never occur except in kernels with the bug
 		 * mentioned above.
 		 */
-		if (cl_offset(clob, vmpage->index) >= iter->count + iocb->ki_pos) {
+		if (vmpage->index >= lcc->lcc_end_index) {
 			result = cl_io_read_ahead(env, io, vmpage->index, &ra);
 			if (result < 0 || vmpage->index > ra.cra_end_idx) {
 				cl_read_ahead_release(env, &ra);
diff --git a/fs/lustre/llite/vvp_io.c b/fs/lustre/llite/vvp_io.c
index 561ce66..50c2872 100644
--- a/fs/lustre/llite/vvp_io.c
+++ b/fs/lustre/llite/vvp_io.c
@@ -871,10 +871,8 @@ static int vvp_io_read_start(const struct lu_env *env,
 	iter = *vio->vui_iter;
 
 	lcc = ll_cl_find(inode);
-	lcc->lcc_iter = &iter;
-	lcc->lcc_iocb = vio->vui_iocb;
-	CDEBUG(D_VFSTRACE, "cnt:%ld,iocb pos:%lld\n", lcc->lcc_iter->count,
-	       lcc->lcc_iocb->ki_pos);
+	lcc->lcc_end_index = DIV_ROUND_UP(pos + iter.count, PAGE_SIZE);
+	CDEBUG(D_VFSTRACE, "count:%ld iocb pos:%lld\n", iter.count, pos);
 
 	result = generic_file_read_iter(vio->vui_iocb, &iter);
 out:
-- 
1.8.3.1

_______________________________________________
lustre-devel mailing list
[email protected]
http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org