log xattr=sa create/remove/update to ZIL

As such, there are no specific synchronous semantics defined for
the xattrs. But for xattr=on, it does log to ZIL and zil_commit() is
done, if sync=always is set on dataset. This provides sync semantics
for xattr=on with sync=always set on dataset.

For the xattr=sa implementation, it doesn't log to ZIL, so, even with
sync=always, xattrs are not guaranteed to be synced before xattr call
returns to caller. So, xattr can be lost if system crash happens, before
txg carrying xattr transaction is synced.

This change adds xattr=sa logging to ZIL on xattr create/remove/update
and xattrs are synced to ZIL (zil_commit() done) for sync=always.
This makes xattr=sa behavior similar to xattr=on.

Implementation notes:
The actual logging is fairly straight-forward and does not warrant
additional explanation.
However, it has been 14 years since we last added new TX types
to the ZIL [1], hence this is the first time we do it after the
introduction of zpool features. Therefore, here is an overview of the
feature activation and deactivation workflow:

1. The feature must be enabled. Otherwise, we don't log the new
    record type. This ensures compatibility with older software.
2. The feature is activated per-dataset, since the ZIL is per-dataset.
3. If the feature is enabled and dataset is not for zvol, any append to
    the ZIL chain will activate the feature for the dataset. Likewise
    for starting a new ZIL chain.
4. A dataset that doesn't have a ZIL chain has the feature deactivated.

We ensure (3) by activating on the first zil_commit() after the feature
was enabled. Since activating the features requires waiting for txg
sync, the first zil_commit() after enabling the feature will be slower
than usual. The downside is that this is really a conservative
approximation: even if we never append a 'TX_SETSAXATTR' to the ZIL
chain, we pay the penalty for feature activation. The upside is that the
user is in control of when we pay the penalty, i.e., upon enabling the
feature.

We ensure (4) by hooking into zil_sync(), where ZIL destroy actually
happens.

One more piece on feature activation, since it's spread across
multiple functions:

zil_commit()
  zil_process_commit_list()
    if lwb == NULL // first zil_commit since zil_open
      zil_create()
        if no log block pointer in ZIL header:
          if feature enabled and not active:
	    // CASE 1
            enable, COALESCE txg wait with dmu_tx that allocated the
	    log block
         else // log block was allocated earlier than this zil_open
          if feature enabled and not active:
	    // CASE 2
            enable, EXPLICIT txg wait
    else // already have an in-DRAM LWB
      if feature enabled and not active:
        // this happens when we enable the feature after zil_create
	// CASE 3
        enable, EXPLICIT txg wait

[1] illumos/illumos-gate@da6c28a

Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Christian Schwarz <christian.schwarz@nutanix.com>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Ryan Moeller <freqlabs@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jitendra Patidar <jitendra.patidar@nutanix.com>
Closes openzfs#8768 
Closes openzfs#9078

cmd/zdb/zdb_il.c

@@ -265,6 +265,29 @@ zil_prt_rec_setattr(zilog_t *zilog, int txtype, const void *arg)
 	}
 }
 
+static void
+zil_prt_rec_setsaxattr(zilog_t *zilog, int txtype, const void *arg)
+{
+	(void) zilog, (void) txtype;
+	const lr_setsaxattr_t *lr = arg;
+
+	char *name = (char *)(lr + 1);
+	(void) printf("%sfoid %llu\n", tab_prefix,
+	    (u_longlong_t)lr->lr_foid);
+
+	(void) printf("%sXAT_NAME  %s\n", tab_prefix, name);
+	if (lr->lr_size == 0) {
+		(void) printf("%sXAT_VALUE  NULL\n", tab_prefix);
+	} else {
+		(void) printf("%sXAT_VALUE  ", tab_prefix);
+		char *val = name + (strlen(name) + 1);
+		for (int i = 0; i < lr->lr_size; i++) {
+			(void) printf("%c", *val);
+			val++;
+		}
+	}
+}
+
 static void
 zil_prt_rec_acl(zilog_t *zilog, int txtype, const void *arg)
 {
@@ -304,6 +327,8 @@ static zil_rec_info_t zil_rec_info[TX_MAX_TYPE] = {
 	{.zri_print = zil_prt_rec_create,   .zri_name = "TX_MKDIR_ATTR      "},
 	{.zri_print = zil_prt_rec_create,   .zri_name = "TX_MKDIR_ACL_ATTR  "},
 	{.zri_print = zil_prt_rec_write,    .zri_name = "TX_WRITE2          "},
+	{.zri_print = zil_prt_rec_setsaxattr,
+	    .zri_name = "TX_SETSAXATTR      "},
 };
 
 static int

cmd/ztest/ztest.c

@@ -2386,6 +2386,7 @@ zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
 	NULL,			/* TX_MKDIR_ATTR */
 	NULL,			/* TX_MKDIR_ACL_ATTR */
 	NULL,			/* TX_WRITE2 */
+	NULL,			/* TX_SETSAXATTR */
 };
 
 /*

include/sys/zfs_sa.h

@@ -138,7 +138,7 @@ void zfs_sa_symlink(struct znode *, char *link, int len, dmu_tx_t *);
 void zfs_sa_get_scanstamp(struct znode *, xvattr_t *);
 void zfs_sa_set_scanstamp(struct znode *, xvattr_t *, dmu_tx_t *);
 int zfs_sa_get_xattr(struct znode *);
-int zfs_sa_set_xattr(struct znode *);
+int zfs_sa_set_xattr(struct znode *, const char *, const void *, size_t);
 void zfs_sa_upgrade(struct sa_handle  *, dmu_tx_t *);
 void zfs_sa_upgrade_txholds(dmu_tx_t *, struct znode *);
 void zfs_sa_init(void);

include/sys/zfs_znode.h

@@ -286,6 +286,8 @@ extern void zfs_log_acl(zilog_t *zilog, dmu_tx_t *tx, znode_t *zp,
     vsecattr_t *vsecp, zfs_fuid_info_t *fuidp);
 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 *);
 

include/sys/zil.h

@@ -162,7 +162,8 @@ typedef enum zil_create {
 #define	TX_MKDIR_ATTR		18	/* mkdir with attr */
 #define	TX_MKDIR_ACL_ATTR	19	/* mkdir with ACL + attrs */
 #define	TX_WRITE2		20	/* dmu_sync EALREADY write */
-#define	TX_MAX_TYPE		21	/* Max transaction type */
+#define	TX_SETSAXATTR		21	/* Set sa xattrs on file */
+#define	TX_MAX_TYPE		22	/* Max transaction type */
 
 /*
  * The transactions for mkdir, symlink, remove, rmdir, link, and rename
@@ -182,7 +183,8 @@ typedef enum zil_create {
 	(txtype) == TX_SETATTR ||	\
 	(txtype) == TX_ACL_V0 ||	\
 	(txtype) == TX_ACL ||		\
-	(txtype) == TX_WRITE2)
+	(txtype) == TX_WRITE2 ||	\
+	(txtype) == TX_SETSAXATTR)
 
 /*
  * The number of dnode slots consumed by the object is stored in the 8
@@ -335,6 +337,13 @@ typedef struct {
 	/* optional attribute lr_attr_t may be here */
 } lr_setattr_t;
 
+typedef struct {
+	lr_t		lr_common;	/* common portion of log record */
+	uint64_t	lr_foid;	/* file object to change attributes */
+	uint64_t	lr_size;
+	/* xattr name and value follows */
+} lr_setsaxattr_t;
+
 typedef struct {
 	lr_t		lr_common;	/* common portion of log record */
 	uint64_t	lr_foid;	/* obj id of file */

include/zfeature_common.h

@@ -75,6 +75,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_DEVICE_REBUILD,
 	SPA_FEATURE_ZSTD_COMPRESS,
 	SPA_FEATURE_DRAID,
+	SPA_FEATURE_ZILSAXATTR,
 	SPA_FEATURES
 } spa_feature_t;
 

lib/libzfs/libzfs.abi

@@ -595,7 +595,7 @@
     <elf-symbol name='fletcher_4_superscalar4_ops' size='64' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='fletcher_4_superscalar_ops' size='64' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='libzfs_config_ops' size='16' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
-    <elf-symbol name='spa_feature_table' size='1904' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
+    <elf-symbol name='spa_feature_table' size='1960' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zfeature_checks_disable' size='4' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zfs_deleg_perm_tab' size='512' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
     <elf-symbol name='zfs_history_event_names' size='328' type='object-type' binding='global-binding' visibility='default-visibility' is-defined='yes'/>
@@ -1854,8 +1854,8 @@
     </function-decl>
   </abi-instr>
   <abi-instr address-size='64' path='../../module/zcommon/zfeature_common.c' language='LANG_C99'>
-    <array-type-def dimensions='1' type-id='83f29ca2' size-in-bits='15232' id='d96379d0'>
-      <subrange length='34' type-id='7359adad' id='6a6a7e00'/>
+    <array-type-def dimensions='1' type-id='83f29ca2' size-in-bits='15680' id='d96379d0'>
+      <subrange length='35' type-id='7359adad' id='6a6a7e00'/>
     </array-type-def>
     <enum-decl name='spa_feature' id='33ecb627'>
       <underlying-type type-id='9cac1fee'/>
@@ -1894,7 +1894,8 @@
       <enumerator name='SPA_FEATURE_DEVICE_REBUILD' value='31'/>
       <enumerator name='SPA_FEATURE_ZSTD_COMPRESS' value='32'/>
       <enumerator name='SPA_FEATURE_DRAID' value='33'/>
-      <enumerator name='SPA_FEATURES' value='34'/>
+      <enumerator name='SPA_FEATURE_ZILSAXATTR' value='34'/>
+      <enumerator name='SPA_FEATURES' value='35'/>
     </enum-decl>
     <typedef-decl name='spa_feature_t' type-id='33ecb627' id='d6618c78'/>
     <enum-decl name='zfeature_flags' id='6db816a4'>

man/man4/zfs.4

@@ -2099,6 +2099,16 @@ Limit SLOG write size per commit executed with synchronous priority.
 Any writes above that will be executed with lower (asynchronous) priority
 to limit potential SLOG device abuse by single active ZIL writer.
 .
+.It Sy zfs_zil_saxattr Ns = Ns Sy 1 Ns | Ns 0 Pq int
+Setting this tunable to zero disables ZIL logging of new
+.Sy xattr Ns = Ns Sy sa
+records if the
+.Sy org.openzfs:zilsaxattr
+feature is enabled on the pool.
+This would only be necessary to work around bugs in the ZIL logging or replay
+code for this record type.
+The tunable has no effect if the feature is disabled.
+.
 .It Sy zfs_embedded_slog_min_ms Ns = Ns Sy 64 Pq int
 Usually, one metaslab from each normal-class vdev is dedicated for use by
 the ZIL to log synchronous writes.

man/man7/zpool-features.7

@@ -778,6 +778,24 @@ by user and group.
 \*[instant-never]
 \*[remount-upgrade]
 .
+.feature org.openzfs zilsaxattr yes extensible_dataset
+This feature enables
+.Sy xattr Ns = Ns Sy sa
+extended attribute logging in the ZIL.
+If enabled, extended attribute changes
+.Pq both Sy xattrdir Ns = Ns Sy dir No and Sy xattr Ns = Ns Sy sa
+are guaranteed to be durable if either the dataset had
+.Sy sync Ns = Ns Sy always
+set at the time the changes were made, or
+.Xr sync 2
+is called on the dataset after the changes were made.
+.Pp
+This feature becomes
+.Sy active
+when a ZIL is created for at least one dataset and will be returned to the
+.Sy enabled
+state when it is destroyed for all datasets that use this feature.
+.
 .feature com.delphix zpool_checkpoint yes
 This feature enables the
 .Nm zpool Cm checkpoint

module/os/freebsd/zfs/zfs_vnops_os.c

@@ -5549,7 +5549,7 @@ zfs_deleteextattr_sa(struct vop_deleteextattr_args *ap, const char *attrname)
 	if (error != 0)
 		error = SET_ERROR(error);
 	else
-		error = zfs_sa_set_xattr(zp);
+		error = zfs_sa_set_xattr(zp, attrname, NULL, 0);
 	if (error != 0) {
 		zp->z_xattr_cached = NULL;
 		nvlist_free(nvl);
@@ -5706,9 +5706,9 @@ zfs_setextattr_sa(struct vop_setextattr_args *ap, const char *attrname)
 		if (error != 0)
 			error = SET_ERROR(error);
 	}
-	kmem_free(buf, entry_size);
 	if (error == 0)
-		error = zfs_sa_set_xattr(zp);
+		error = zfs_sa_set_xattr(zp, attrname, buf, entry_size);
+	kmem_free(buf, entry_size);
 	if (error != 0) {
 		zp->z_xattr_cached = NULL;
 		nvlist_free(nvl);

module/os/linux/zfs/zpl_xattr.c

@@ -578,7 +578,7 @@ zpl_xattr_set_sa(struct inode *ip, const char *name, const void *value,
 	 * will be reconstructed from the ARC when next accessed.
 	 */
 	if (error == 0)
-		error = -zfs_sa_set_xattr(zp);
+		error = -zfs_sa_set_xattr(zp, name, value, size);
 
 	if (error) {
 		nvlist_free(nvl);

module/zcommon/zfeature_common.c

@@ -695,6 +695,18 @@ zpool_feature_init(void)
 	    "org.openzfs:draid", "draid", "Support for distributed spare RAID",
 	    ZFEATURE_FLAG_MOS, ZFEATURE_TYPE_BOOLEAN, NULL, sfeatures);
 
+	{
+	static const spa_feature_t zilsaxattr_deps[] = {
+		SPA_FEATURE_EXTENSIBLE_DATASET,
+		SPA_FEATURE_NONE
+	};
+	zfeature_register(SPA_FEATURE_ZILSAXATTR,
+	    "org.openzfs:zilsaxattr", "zilsaxattr",
+	    "Support for xattr=sa extended attribute logging in ZIL.",
+	    ZFEATURE_FLAG_PER_DATASET | ZFEATURE_FLAG_READONLY_COMPAT,
+	    ZFEATURE_TYPE_BOOLEAN, zilsaxattr_deps, sfeatures);
+	}
+
 	zfs_mod_list_supported_free(sfeatures);
 }
 

module/zfs/zfs_log.c

@@ -720,6 +720,40 @@ zfs_log_setattr(zilog_t *zilog, dmu_tx_t *tx, int txtype,
 	zil_itx_assign(zilog, itx, tx);
 }
 
+/*
+ * Handles TX_SETSAXATTR transactions.
+ */
+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)
+{
+	itx_t		*itx;
+	lr_setsaxattr_t	*lr;
+	size_t		recsize = sizeof (lr_setsaxattr_t);
+	void		*xattrstart;
+	int		namelen;
+
+	if (zil_replaying(zilog, tx) || zp->z_unlinked)
+		return;
+
+	namelen = strlen(name) + 1;
+	recsize += (namelen + size);
+	itx = zil_itx_create(txtype, recsize);
+	lr = (lr_setsaxattr_t *)&itx->itx_lr;
+	lr->lr_foid = zp->z_id;
+	xattrstart = (char *)(lr + 1);
+	bcopy(name, xattrstart, namelen);
+	if (value != NULL) {
+		bcopy(value, (char *)xattrstart + namelen, size);
+		lr->lr_size = size;
+	} else {
+		lr->lr_size = 0;
+	}
+
+	itx->itx_sync = (zp->z_sync_cnt != 0);
+	zil_itx_assign(zilog, itx, tx);
+}
+
 /*
  * Handles TX_ACL transactions.
  */

module/zfs/zfs_replay.c

@@ -47,6 +47,8 @@
 #include <sys/atomic.h>
 #include <sys/cred.h>
 #include <sys/zpl.h>
+#include <sys/dmu_objset.h>
+#include <sys/zfeature.h>
 
 /*
  * NB: FreeBSD expects to be able to do vnode locking in lookup and
@@ -868,6 +870,86 @@ zfs_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
 	return (error);
 }
 
+static int
+zfs_replay_setsaxattr(void *arg1, void *arg2, boolean_t byteswap)
+{
+	zfsvfs_t *zfsvfs = arg1;
+	lr_setsaxattr_t *lr = arg2;
+	znode_t *zp;
+	nvlist_t *nvl;
+	size_t sa_size;
+	char *name;
+	char *value;
+	size_t size;
+	int error = 0;
+
+	ASSERT(spa_feature_is_active(zfsvfs->z_os->os_spa,
+	    SPA_FEATURE_ZILSAXATTR));
+	if (byteswap)
+		byteswap_uint64_array(lr, sizeof (*lr));
+
+	if ((error = zfs_zget(zfsvfs, lr->lr_foid, &zp)) != 0)
+		return (error);
+
+	rw_enter(&zp->z_xattr_lock, RW_WRITER);
+	mutex_enter(&zp->z_lock);
+	if (zp->z_xattr_cached == NULL)
+		error = zfs_sa_get_xattr(zp);
+	mutex_exit(&zp->z_lock);
+
+	if (error)
+		goto out;
+
+	ASSERT(zp->z_xattr_cached);
+	nvl = zp->z_xattr_cached;
+
+	/* Get xattr name, value and size from log record */
+	size = lr->lr_size;
+	name = (char *)(lr + 1);
+	if (size == 0) {
+		value = NULL;
+		error = nvlist_remove(nvl, name, DATA_TYPE_BYTE_ARRAY);
+	} else {
+		value = name + strlen(name) + 1;
+		/* Limited to 32k to keep nvpair memory allocations small */
+		if (size > DXATTR_MAX_ENTRY_SIZE) {
+			error = SET_ERROR(EFBIG);
+			goto out;
+		}
+
+		/* Prevent the DXATTR SA from consuming the entire SA region */
+		error = nvlist_size(nvl, &sa_size, NV_ENCODE_XDR);
+		if (error)
+			goto out;
+
+		if (sa_size > DXATTR_MAX_SA_SIZE) {
+			error = SET_ERROR(EFBIG);
+			goto out;
+		}
+
+		error = nvlist_add_byte_array(nvl, name, (uchar_t *)value,
+		    size);
+	}
+
+	/*
+	 * Update the SA for additions, modifications, and removals. On
+	 * error drop the inconsistent cached version of the nvlist, it
+	 * will be reconstructed from the ARC when next accessed.
+	 */
+	if (error == 0)
+		error = zfs_sa_set_xattr(zp, name, value, size);
+
+	if (error) {
+		nvlist_free(nvl);
+		zp->z_xattr_cached = NULL;
+	}
+
+out:
+	rw_exit(&zp->z_xattr_lock);
+	zrele(zp);
+	return (error);
+}
+
 static int
 zfs_replay_acl_v0(void *arg1, void *arg2, boolean_t byteswap)
 {
@@ -989,4 +1071,5 @@ zil_replay_func_t *const zfs_replay_vector[TX_MAX_TYPE] = {
 	zfs_replay_create,	/* TX_MKDIR_ATTR */
 	zfs_replay_create_acl,	/* TX_MKDIR_ACL_ATTR */
 	zfs_replay_write2,	/* TX_WRITE2 */
+	zfs_replay_setsaxattr,	/* TX_SETSAXATTR */
 };