From: Bobi Jam <bobijam@whamcloud.com>
Date: Fri, 26 Jan 2024 10:14:36 +0000 (+0800)
Subject: LU-17482 llite: short read could mess up next read offset
X-Git-Url: https://git.whamcloud.com/gitweb?a=commitdiff_plain;h=fbb319a60cc165c1ab14d50c81c35b5fd94f8dcb;p=fs%2Flustre-release.git

LU-17482 llite: short read could mess up next read offset

When read reaches EOF, it could read data from stale pagecache, but
we need to restore the iocb->ki_pos so that next read could continue
from the correct offset.

Lustre-change: https://review.whamcloud.com/53827
Lustre-commit: TBD (from 4bec3a277c83932cfb5ba26e31336e1f4666460a)

Fixes: 4468f6c9d9 ("LU-16025 llite: adjust read count as file got truncated")
Signed-off-by: Bobi Jam <bobijam@whamcloud.com>
Change-Id: Ib8b62c41bf65f8efec82dda53fcfbdb68ad08b38
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/53828
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Patrick Farrell <pfarrell@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
---

diff --git a/lustre/include/obd_support.h b/lustre/include/obd_support.h
index b4a0e27..6f777b3 100644
--- a/lustre/include/obd_support.h
+++ b/lustre/include/obd_support.h
@@ -638,6 +638,7 @@ extern char obd_jobid_var[];
 #define OBD_FAIL_LOV_COMP_PATTERN		    0x1427
 #define OBD_FAIL_LOV_INVALID_OSTIDX		    0x1428
 #define OBD_FAIL_LLITE_DELAY_TRUNCATE		    0x1430
+#define OBD_FAIL_LLITE_READ_PAUSE		    0x1431
 
 #define OBD_FAIL_FID_INDIR	0x1501
 #define OBD_FAIL_FID_INLMA	0x1502
diff --git a/lustre/llite/file.c b/lustre/llite/file.c
index 78074b7..e38d944 100644
--- a/lustre/llite/file.c
+++ b/lustre/llite/file.c
@@ -2114,6 +2114,7 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 	struct lu_env *env;
 	struct vvp_io_args *args;
 	struct file *file = iocb->ki_filp;
+	loff_t orig_ki_pos = iocb->ki_pos;
 	ssize_t result;
 	ssize_t rc2;
 	__u16 refcheck;
@@ -2141,6 +2142,8 @@ static ssize_t ll_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 	else if (result > 0)
 		stale_data = true;
 
+	CFS_FAIL_TIMEOUT_ORSET(OBD_FAIL_LLITE_READ_PAUSE, CFS_FAIL_ONCE,
+			       cfs_fail_val);
 	/**
 	 * Currently when PCC read failed, we do not fall back to the
 	 * normal read path, just return the error.
@@ -2181,6 +2184,7 @@ out:
 		 * we've reached EOF before the read, the data read are cached
 		 * stale data.
 		 */
+		iocb->ki_pos = orig_ki_pos;
 		iov_iter_truncate(to, 0);
 		result = 0;
 	}
diff --git a/lustre/tests/multiop.c b/lustre/tests/multiop.c
index 09ff6e1..cccd454 100644
--- a/lustre/tests/multiop.c
+++ b/lustre/tests/multiop.c
@@ -623,6 +623,9 @@ int main(int argc, char **argv)
 						     ~ALIGN_LEN);
 			}
 			while (len > 0) {
+				off_t start, off;
+
+				start = lseek(fd, 0, SEEK_CUR);
 				rc = read(fd, buf_align, len);
 				if (rc == -1) {
 					save_errno = errno;
@@ -630,8 +633,9 @@ int main(int argc, char **argv)
 					exit(save_errno);
 				}
 				if (rc < len) {
-					fprintf(stderr, "short read: %lld/%u\n",
-						rc, len);
+					off = lseek(fd, 0, SEEK_CUR);
+					fprintf(stderr, "short read: %ld ->+ %u -> %ld %lld\n",
+						start, len, off, rc);
 					if (rc == 0)
 						exit(ENODATA);
 				}
diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 23d25c7..1b2f971 100755
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -22297,7 +22297,7 @@ test_250() {
 }
 run_test 250 "Write above 16T limit"
 
-test_251() {
+test_251a() {
 	$LFS setstripe -c -1 -S 1048576 $DIR/$tfile
 
 	#define OBD_FAIL_LLITE_LOST_LAYOUT 0x1407
@@ -22312,7 +22312,29 @@ test_251() {
 
 	rm -f $DIR/$tfile
 }
-run_test 251 "Handling short read and write correctly"
+run_test 251a "Handling short read and write correctly"
+
+test_251b() {
+	dd if=/dev/zero of=$DIR/$tfile bs=1k count=4 ||
+		error "write $tfile failed"
+
+	sleep 2 && echo 12345 >> $DIR/$tfile &
+
+	#define OBD_FAIL_LLITE_READ_PAUSE 0x1431
+	$LCTL set_param fail_loc=0x1431 fail_val=5
+	# seek to 4096, 2 seconds later, file size expand to 4102, and after
+	# 5 seconds, read 10 bytes, the short read should
+	# report:
+	#                start ->+ read_len -> offset_after_read read_count
+	#     short read: 4096 ->+ 10 -> 4096 0
+	# not:
+	#     short read: 4096 ->+ 10 -> 4102 0
+	local off=$($MULTIOP $DIR/$tfile oO_RDONLY:z4096r10c 2>&1 | \
+			awk '/short read/ { print $7 }')
+	(( off == 4096 )) ||
+		error "short read should set offset at 4096, not $off"
+}
+run_test 251b "short read restore offset correctly"
 
 test_252() {
 	remote_mds_nodsh && skip "remote MDS with nodsh"