Merge pull request #152 from truenas/truenas/zfs-2.1-release

Sync stable/bluefin with truenas/zfs-2.1-release

config/kernel-inode-permission.m4

@@ -1,4 +1,22 @@
 AC_DEFUN([ZFS_AC_KERNEL_SRC_PERMISSION], [
+	dnl #
+	dnl # 6.3 API change
+	dnl # iops->permission() now takes struct mnt_idmap*
+	dnl # as its first arg
+	dnl #
+	ZFS_LINUX_TEST_SRC([permission_mnt_idmap], [
+		#include <linux/fs.h>
+		#include <linux/sched.h>
+
+		int inode_permission(struct mnt_idmap *idmap,
+		    struct inode *inode, int mask) { return 0; }
+
+		static const struct inode_operations
+			iops __attribute__ ((unused)) = {
+			.permission             = inode_permission,
+		};
+	],[])
+
 	dnl #
 	dnl # 5.12 API change that added the struct user_namespace* arg
 	dnl # to the front of this function type's arg list.
@@ -18,12 +36,20 @@ AC_DEFUN([ZFS_AC_KERNEL_SRC_PERMISSION], [
 ])
 
 AC_DEFUN([ZFS_AC_KERNEL_PERMISSION], [
-	AC_MSG_CHECKING([whether iops->permission() takes struct user_namespace*])
-	ZFS_LINUX_TEST_RESULT([permission_userns], [
+	AC_MSG_CHECKING([whether iops->permission() takes struct mnt_idmap*])
+	ZFS_LINUX_TEST_RESULT([permission_mnt_idmap], [
 		AC_MSG_RESULT(yes)
-		AC_DEFINE(HAVE_IOPS_PERMISSION_USERNS, 1,
-		   [iops->permission() takes struct user_namespace*])
+		AC_DEFINE(HAVE_IOPS_PERMISSION_IDMAP, 1,
+		   [iops->permission() takes struct mnt_idmap*])
 	],[
-		AC_MSG_RESULT(no)
+		AC_MSG_CHECKING([whether iops->permission() takes struct user_namespace*])
+		ZFS_LINUX_TEST_RESULT([permission_userns], [
+			AC_MSG_RESULT(yes)
+			AC_DEFINE(HAVE_IOPS_PERMISSION_USERNS, 1,
+			   [iops->permission() takes struct user_namespace*])
+		],[
+			AC_MSG_RESULT(no)
+		])
 	])
 ])
+

include/os/linux/zfs/sys/zpl.h

@@ -105,6 +105,8 @@ zpl_chmod_acl(struct inode *ip)
 #if defined(HAVE_IOPS_PERMISSION_USERNS)
 extern int zpl_permission(struct user_namespace *userns, struct inode *ip,
     int mask);
+#elif defined(HAVE_IOPS_PERMISSION_IDMAP)
+extern int zpl_permission(struct mnt_idmap *idmap, struct inode *ip, int mask);
 #else
 extern int zpl_permission(struct inode *ip, int mask);
 #endif

include/sys/spa_impl.h

@@ -249,6 +249,7 @@ struct spa {
 	uint64_t	spa_min_ashift;		/* of vdevs in normal class */
 	uint64_t	spa_max_ashift;		/* of vdevs in normal class */
 	uint64_t	spa_min_alloc;		/* of vdevs in normal class */
+	uint64_t	spa_gcd_alloc;		/* of vdevs in normal class */
 	uint64_t	spa_config_guid;	/* config pool guid */
 	uint64_t	spa_load_guid;		/* spa_load initialized guid */
 	uint64_t	spa_last_synced_guid;	/* last synced guid */

include/sys/vdev_impl.h

@@ -438,6 +438,7 @@ struct vdev {
 	boolean_t	vdev_copy_uberblocks;  /* post expand copy uberblocks */
 	boolean_t	vdev_resilver_deferred;  /* resilver deferred */
 	boolean_t	vdev_kobj_flag; /* kobj event record */
+	boolean_t	vdev_attaching; /* vdev attach ashift handling */
 	vdev_queue_t	vdev_queue;	/* I/O deadline schedule queue	*/
 	vdev_cache_t	vdev_cache;	/* physical block cache		*/
 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/

module/os/linux/zfs/policy.c

@@ -123,7 +123,7 @@ secpolicy_vnode_access2(const cred_t *cr, struct inode *ip, uid_t owner,
 	if ((fsuid == owner) || (fsuid == 0))
 		return (0);
 
-	if (zpl_inode_owner_or_capable(kcred->user_ns, ip))
+	if (zpl_inode_owner_or_capable(zfs_init_idmap, ip))
 		return (0);
 
 #if defined(CONFIG_USER_NS)

module/os/linux/zfs/zfs_acl.c

@@ -2543,8 +2543,9 @@ zfs_zaccess_trivial(znode_t *zp, uint32_t *working_mode, cred_t *cr)
 		return (unmapped ? SET_ERROR(EPERM) : 0);
 	}
 
-#if defined(HAVE_IOPS_PERMISSION_USERNS)
-	err = generic_permission(cr->user_ns, ZTOI(zp), mask);
+#if (defined(HAVE_IOPS_PERMISSION_USERNS) || \
+	defined(HAVE_IOPS_PERMISSION_IDMAP))
+	err = generic_permission(zfs_init_idmap, ZTOI(zp), mask);
 #else
 	err = generic_permission(ZTOI(zp), mask);
 #endif

module/os/linux/zfs/zpl_file.c

@@ -1116,7 +1116,7 @@ __zpl_ioctl_setdosflags(struct inode *ip, uint64_t ioctl_flags, xvattr_t *xva)
 	    !capable(CAP_LINUX_IMMUTABLE))
 		return (-EPERM);
 
-	if (!zpl_inode_owner_or_capable(kcred->user_ns, ip))
+	if (!zpl_inode_owner_or_capable(zfs_init_idmap, ip))
 		return (-EACCES);
 
 	xva_init(xva);

module/os/linux/zfs/zpl_xattr.c

@@ -1498,6 +1498,8 @@ xattr_handler_t zpl_xattr_acl_default_handler =
 int
 #if defined(HAVE_IOPS_PERMISSION_USERNS)
 zpl_permission(struct user_namespace *userns, struct inode *ip, int mask)
+#elif defined(HAVE_IOPS_PERMISSION_IDMAP)
+zpl_permission(struct mnt_idmap *idmap, struct inode *ip, int mask)
 #else
 zpl_permission(struct inode *ip, int mask)
 #endif
@@ -1513,8 +1515,9 @@ zpl_permission(struct inode *ip, int mask)
 	 */
 	if ((ITOZSB(ip)->z_acl_type != ZFS_ACLTYPE_NFSV4) ||
 	    ((ITOZ(ip)->z_pflags & ZFS_ACL_TRIVIAL && GENERIC_MASK(mask)))) {
-#if defined(HAVE_IOPS_PERMISSION_USERNS)
-		return (generic_permission(userns, ip, mask));
+#if (defined(HAVE_IOPS_PERMISSION_USERNS) || \
+	defined(HAVE_IOPS_PERMISSION_IDMAP))
+		return (generic_permission(zfs_init_idmap, ip, mask));
 #else
 		return (generic_permission(ip, mask));
 #endif
@@ -1531,8 +1534,9 @@ zpl_permission(struct inode *ip, int mask)
 	 * NFSv4 ACE. Pass back to default kernel permissions check.
 	 */
 	if (to_check == 0) {
-#if defined(HAVE_IOPS_PERMISSION_USERNS)
-		return (generic_permission(userns, ip, mask));
+#if (defined(HAVE_IOPS_PERMISSION_USERNS) || \
+	defined(HAVE_IOPS_PERMISSION_IDMAP))
+		return (generic_permission(zfs_init_idmap, ip, mask));
 #else
 		return (generic_permission(ip, mask));
 #endif

module/zfs/spa_misc.c

@@ -773,6 +773,7 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	spa->spa_min_ashift = INT_MAX;
 	spa->spa_max_ashift = 0;
 	spa->spa_min_alloc = INT_MAX;
+	spa->spa_gcd_alloc = INT_MAX;
 
 	/* Reset cached value */
 	spa->spa_dedup_dspace = ~0ULL;

module/zfs/vdev.c

@@ -854,9 +854,15 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 	    &vd->vdev_not_present);
 
 	/*
-	 * Get the alignment requirement.
+	 * Get the alignment requirement. Ignore pool ashift for vdev
+	 * attach case.
 	 */
-	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
+	if (alloctype != VDEV_ALLOC_ATTACH) {
+		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT,
+		    &vd->vdev_ashift);
+	} else {
+		vd->vdev_attaching = B_TRUE;
+	}
 
 	/*
 	 * Retrieve the vdev creation time.
@@ -1345,6 +1351,36 @@ vdev_remove_parent(vdev_t *cvd)
 	vdev_free(mvd);
 }
 
+/*
+ * Choose GCD for spa_gcd_alloc.
+ */
+static uint64_t
+vdev_gcd(uint64_t a, uint64_t b)
+{
+	while (b != 0) {
+		uint64_t t = b;
+		b = a % b;
+		a = t;
+	}
+	return (a);
+}
+
+/*
+ * Set spa_min_alloc and spa_gcd_alloc.
+ */
+static void
+vdev_spa_set_alloc(spa_t *spa, uint64_t min_alloc)
+{
+	if (min_alloc < spa->spa_min_alloc)
+		spa->spa_min_alloc = min_alloc;
+	if (spa->spa_gcd_alloc == INT_MAX) {
+		spa->spa_gcd_alloc = min_alloc;
+	} else {
+		spa->spa_gcd_alloc = vdev_gcd(min_alloc,
+		    spa->spa_gcd_alloc);
+	}
+}
+
 void
 vdev_metaslab_group_create(vdev_t *vd)
 {
@@ -1397,8 +1433,7 @@ vdev_metaslab_group_create(vdev_t *vd)
 				spa->spa_min_ashift = vd->vdev_ashift;
 
 			uint64_t min_alloc = vdev_get_min_alloc(vd);
-			if (min_alloc < spa->spa_min_alloc)
-				spa->spa_min_alloc = min_alloc;
+			vdev_spa_set_alloc(spa, min_alloc);
 		}
 	}
 }
@@ -2093,9 +2128,9 @@ vdev_open(vdev_t *vd)
 				return (SET_ERROR(EDOM));
 			}
 
-			if (vd->vdev_top == vd) {
+			if (vd->vdev_top == vd && vd->vdev_attaching == B_FALSE)
 				vdev_ashift_optimize(vd);
-			}
+			vd->vdev_attaching = B_FALSE;
 		}
 		if (vd->vdev_ashift != 0 && (vd->vdev_ashift < ASHIFT_MIN ||
 		    vd->vdev_ashift > ASHIFT_MAX)) {
@@ -2156,8 +2191,7 @@ vdev_open(vdev_t *vd)
 	if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
 	    vd->vdev_islog == 0 && vd->vdev_aux == NULL) {
 		uint64_t min_alloc = vdev_get_min_alloc(vd);
-		if (min_alloc < spa->spa_min_alloc)
-			spa->spa_min_alloc = min_alloc;
+		vdev_spa_set_alloc(spa, min_alloc);
 	}
 
 	/*

module/zfs/zio.c

@@ -1529,6 +1529,19 @@ zio_shrink(zio_t *zio, uint64_t size)
 	}
 }
 
+/*
+ * Round provided allocation size up to a value that can be allocated
+ * by at least some vdev(s) in the pool with minimum or no additional
+ * padding and without extra space usage on others
+ */
+static uint64_t
+zio_roundup_alloc_size(spa_t *spa, uint64_t size)
+{
+	if (size > spa->spa_min_alloc)
+		return (roundup(size, spa->spa_gcd_alloc));
+	return (spa->spa_min_alloc);
+}
+
 /*
  * ==========================================================================
  * Prepare to read and write logical blocks
@@ -1735,9 +1748,8 @@ zio_write_compress(zio_t *zio)
 			 * in that we charge for the padding used to fill out
 			 * the last sector.
 			 */
-			ASSERT3U(spa->spa_min_alloc, >=, SPA_MINBLOCKSHIFT);
-			size_t rounded = (size_t)roundup(psize,
-			    spa->spa_min_alloc);
+			size_t rounded = (size_t)zio_roundup_alloc_size(spa,
+			    psize);
 			if (rounded >= lsize) {
 				compress = ZIO_COMPRESS_OFF;
 				zio_buf_free(cbuf, lsize);
@@ -1780,8 +1792,8 @@ zio_write_compress(zio_t *zio)
 		 * take this codepath because it will change the on-disk block
 		 * and decryption will fail.
 		 */
-		size_t rounded = MIN((size_t)roundup(psize,
-		    spa->spa_min_alloc), lsize);
+		size_t rounded = MIN((size_t)zio_roundup_alloc_size(spa, psize),
+		    lsize);
 
 		if (rounded != psize) {
 			abd_t *cdata = abd_alloc_linear(rounded, B_TRUE);

tests/zfs-tests/tests/functional/cli_root/zpool_attach/attach-o_ashift.ksh

@@ -66,35 +66,14 @@ log_must set_tunable64 VDEV_FILE_PHYSICAL_ASHIFT 16
 typeset ashifts=("9" "10" "11" "12" "13" "14" "15" "16")
 for ashift in ${ashifts[@]}
 do
-	for cmdval in ${ashifts[@]}
-	do
-		log_must zpool create -o ashift=$ashift $TESTPOOL1 $disk1
-		verify_ashift $disk1 $ashift
-		if [[ $? -ne 0 ]]
-		then
-			log_fail "Pool was created without setting ashift " \
-			    "value to $ashift"
-		fi
-		# ashift_of(attached_disk) <= ashift_of(existing_vdev)
-		if [[ $cmdval -le $ashift ]]
-		then
-			log_must zpool attach -o ashift=$cmdval $TESTPOOL1 \
-			    $disk1 $disk2
-			verify_ashift $disk2 $ashift
-			if [[ $? -ne 0 ]]
-			then
-				log_fail "Device was attached without " \
-				    "setting ashift value to $ashift"
-			fi
-		else
-			log_mustnot zpool attach -o ashift=$cmdval $TESTPOOL1 \
-			    $disk1 $disk2
-		fi
-		# clean things for the next run
-		log_must zpool destroy $TESTPOOL1
-		log_must zpool labelclear $disk1
-		log_must zpool labelclear $disk2
-	done
+	log_must zpool create -o ashift=$ashift $TESTPOOL1 $disk1
+	log_must verify_ashift $disk1 $ashift
+	log_must zpool attach $TESTPOOL1 $disk1 $disk2
+	log_must verify_ashift $disk2 $ashift
+	# clean things for the next run
+	log_must zpool destroy $TESTPOOL1
+	log_must zpool labelclear $disk1
+	log_must zpool labelclear $disk2
 done
 
 typeset badvals=("off" "on" "1" "8" "17" "1b" "ff" "-")

tests/zfs-tests/tests/functional/cli_root/zpool_replace/replace-o_ashift.ksh

@@ -66,36 +66,16 @@ log_must set_tunable64 VDEV_FILE_PHYSICAL_ASHIFT 16
 typeset ashifts=("9" "10" "11" "12" "13" "14" "15" "16")
 for ashift in ${ashifts[@]}
 do
-	for cmdval in ${ashifts[@]}
-	do
-		log_must zpool create -o ashift=$ashift $TESTPOOL1 $disk1
-		verify_ashift $disk1 $ashift
-		if [[ $? -ne 0 ]]
-		then
-			log_fail "Pool was created without setting ashift " \
-			    "value to $ashift"
-		fi
-		# ashift_of(replacing_disk) <= ashift_of(existing_vdev)
-		if [[ $cmdval -le $ashift ]]
-		then
-			log_must zpool replace -o ashift=$cmdval $TESTPOOL1 \
-			    $disk1 $disk2
-			verify_ashift $disk2 $ashift
-			if [[ $? -ne 0 ]]
-			then
-				log_fail "Device was replaced without " \
-				    "setting ashift value to $ashift"
-			fi
-			wait_replacing $TESTPOOL1
-		else
-			log_mustnot zpool replace -o ashift=$cmdval $TESTPOOL1 \
-			    $disk1 $disk2
-		fi
-		# clean things for the next run
-		log_must zpool destroy $TESTPOOL1
-		log_must zpool labelclear $disk1
-		log_must zpool labelclear $disk2
-	done
+	log_must zpool create -o ashift=$ashift $TESTPOOL1 $disk1
+	log_must verify_ashift $disk1 $ashift
+	# ashift_of(replacing_disk) <= ashift_of(existing_vdev)
+	log_must zpool replace $TESTPOOL1 $disk1 $disk2
+	log_must verify_ashift $disk2 $ashift
+	wait_replacing $TESTPOOL1
+	# clean things for the next run
+	log_must zpool destroy $TESTPOOL1
+	log_must zpool labelclear $disk1
+	log_must zpool labelclear $disk2
 done
 
 typeset badvals=("off" "on" "1" "8" "17" "1b" "ff" "-")