Linux: log tmpfile link as new file creation

WIP

Signed-off-by: Pavel Snajdr <[email protected]>

Author: snajpa

include/sys/zfs_znode.h

@@ -325,8 +325,9 @@ extern void zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx);
 extern void zfs_upgrade(zfsvfs_t *zfsvfs, dmu_tx_t *tx);
 extern void zfs_log_setsaxattr(zilog_t *zilog, dmu_tx_t *tx, int txtype,
     znode_t *zp, const char *name, const void *value, size_t size);
-
 extern void zfs_znode_update_vfs(struct znode *);
+extern void zfs_log_link_tmpfile(zilog_t *zilog, dmu_tx_t *tx,
+    znode_t *dzp, znode_t *zp, const char *name);
 
 #endif
 #ifdef	__cplusplus

module/os/linux/zfs/zfs_vnops_os.c

@@ -3645,16 +3645,11 @@ zfs_link(znode_t *tdzp, znode_t *szp, char *name, cred_t *cr,
 
 	if (error == 0) {
 		uint64_t txtype = TX_LINK;
-		/*
-		 * tmpfile is created to be in z_unlinkedobj, so remove it.
-		 * Also, we don't log in ZIL, because all previous file
-		 * operation on the tmpfile are ignored by ZIL. Instead we
-		 * always wait for txg to sync to make sure all previous
-		 * operation are sync safe.
-		 */
 		if (is_tmpfile) {
+			/* remove from unlinked set */
 			VERIFY(zap_remove_int(zfsvfs->z_os,
 			    zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0);
+			zfs_log_link_tmpfile(zilog, tx, tdzp, szp, name);
 		} else {
 			if (flags & FIGNORECASE)
 				txtype |= TX_CI;
@@ -3669,12 +3664,12 @@ zfs_link(znode_t *tdzp, znode_t *szp, char *name, cred_t *cr,
 
 	zfs_dirent_unlock(dl);
 
-	if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
+	if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)
 		zil_commit(zilog, 0);
 
-	if (is_tmpfile && zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED)
+	if (is_tmpfile && !zfsvfs->z_xattr_sa &&
+	    zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED)
 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg);
-
 	zfs_znode_update_vfs(tdzp);
 	zfs_znode_update_vfs(szp);
 	zfs_exit(zfsvfs, FTAG);

module/zfs/zfs_log.c

@@ -935,5 +935,74 @@ zfs_log_clone_range(zilog_t *zilog, dmu_tx_t *tx, int txtype, znode_t *zp,
 	}
 }
 
+void
+zfs_log_link_tmpfile(zilog_t *zilog, dmu_tx_t *tx, znode_t *dzp, znode_t *zp,
+    const char *name)
+{
+	zfsvfs_t *zfsvfs = ZTOZSB(dzp);
+
+	if (zil_replaying(zilog, tx))
+		return;
+
+	vattr_t va = { 0 };
+	va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID;
+	va.va_mode = zp->z_mode;
+	va.va_uid  = (uid_t)KUID_TO_SUID(ZTOUID(zp));
+	va.va_gid  = (gid_t)KGID_TO_SGID(ZTOGID(zp));
+
+	zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name, NULL, NULL, &va);
+
+	uint64_t size = zp->z_size;
+	uint64_t off = 0;
+	uint64_t max_chunk = (uint64_t)zfs_immediate_write_sz;
+
+	/* TODO: Reflect the file sync open flags here */
+	boolean_t commit =
+	    (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS);
+	while (off < size) {
+		uint64_t len = size - off;
+		if (len > max_chunk)
+			len = max_chunk;
+		zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, commit,
+		    B_FALSE, NULL, NULL);
+		off += len;
+	}
+
+	if (!zp->z_xattr_cached)
+		return;
+
+	/* Haven't thought about xattr=dir tbh, is there a way to log those? */
+	ASSERT3P(zfsvfs->z_xattr_sa, !=, NULL);
+
+	nvpair_t *nvp = NULL;
+	nvlist_t *nvl = zp->z_xattr_cached;
+
+	while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
+		const char *xname = nvpair_name(nvp);
+		const void *val = NULL;
+		size_t vlen = 0;
+
+		if (nvpair_type(nvp) == DATA_TYPE_STRING) {
+			const char *s;
+			VERIFY0(nvpair_value_string(nvp, &s));
+			val = s;
+			vlen = strlen(s);
+		} else if (nvpair_type(nvp) == DATA_TYPE_BYTE_ARRAY) {
+			const uchar_t *buf;
+			uint_t n;
+			VERIFY0(nvpair_value_byte_array(nvp,
+			    (uchar_t **)&buf, &n));
+			val = buf;
+			vlen = n;
+		} else
+			continue;
+
+
+		zfs_log_setsaxattr(zilog, tx, TX_SETSAXATTR,
+		    zp, xname, val, vlen);
+	}
+}
+
+
 ZFS_MODULE_PARAM(zfs, zfs_, immediate_write_sz, S64, ZMOD_RW,
 	"Largest data block to write to zil");

tests/runfiles/linux.run

@@ -216,7 +216,7 @@ tags = ['functional', 'snapshot']
 
 [tests/functional/tmpfile:Linux]
 tests = ['tmpfile_001_pos', 'tmpfile_002_pos', 'tmpfile_003_pos',
-    'tmpfile_stat_mode']
+    'tmpfile_004_pos', 'tmpfile_stat_mode']
 tags = ['functional', 'tmpfile']
 
 [tests/functional/upgrade:Linux]

tests/zfs-tests/Makefile.am

@@ -25,6 +25,7 @@ scripts_zfs_tests_functional_tmpfile_PROGRAMS = \
 	%D%/tests/functional/tmpfile/tmpfile_001_pos \
 	%D%/tests/functional/tmpfile/tmpfile_002_pos \
 	%D%/tests/functional/tmpfile/tmpfile_003_pos \
+	%D%/tests/functional/tmpfile/tmpfile_replay \
 	%D%/tests/functional/tmpfile/tmpfile_stat_mode \
 	%D%/tests/functional/tmpfile/tmpfile_test
 endif

tests/zfs-tests/tests/functional/tmpfile/.gitignore

@@ -2,4 +2,5 @@
 /tmpfile_001_pos
 /tmpfile_002_pos
 /tmpfile_003_pos
+/tmpfile_replay
 /tmpfile_stat_mode

tests/zfs-tests/tests/functional/tmpfile/tmpfile_004_pos.sh

@@ -0,0 +1,37 @@
+#!/bin/ksh -p
+# SPDX-License-Identifier: CDDL-1.0
+
+. "$STF_SUITE/include/libtest.shlib"
+
+if [ "$(uname -s)" != "Linux" ]; then
+        log_skip "O_TMPFILE is not supported on this platform."
+fi
+
+verify_runnable "global"
+
+log_assert "O_TMPFILE linking with xattr=sa should survive replay"
+
+log_must zfs set xattr=sa "$TESTPOOL/$TESTFS"
+
+typeset target_uid target_gid
+if [ "$(id -u)" -eq 0 ]; then
+        target_uid=1000
+        target_gid=1001
+else
+        target_uid=$(id -u)
+        target_gid=$(id -g)
+fi
+
+log_must test -n "$target_uid" -a -n "$target_gid"
+
+export TARGET_UID="$target_uid"
+export TARGET_GID="$target_gid"
+export TESTFILE="tmpfile_replay.$$"
+
+for sync_mode in standard always; do
+        log_note "Running O_TMPFILE crash-replay test with sync=$sync_mode"
+        log_must zfs set sync="$sync_mode" "$TESTPOOL/$TESTFS"
+        log_must "$STF_SUITE/tests/functional/tmpfile/tmpfile_replay"
+done
+
+log_pass "Crash-replay of O_TMPFILE linking (xattr=sa) verified."