Illumos #4950 files sometimes can't be removed from a full filesystem

Reviewed by: Adam Leventhal <adam.leventhal@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Sebastien Roy <sebastien.roy@delphix.com>
Reviewed by: Boris Protopopov <bprotopopov@hotmail.com>
Approved by: Dan McDonald <danmcd@omniti.com>
Ported-by: Richard Yao <richard.yao@clusterhq.com>

Porting notes:
1. ZoL currently does not log discards to zvols, so the portion of this patch that
modifies the discard logging to mark it as freeing space has been discarded.

2. may_delete_now had been removed from zfs_remove() in ZoL. It has been
reintroduced.

include/sys/dmu.h

@@ -564,6 +564,7 @@ void dmu_tx_abort(dmu_tx_t *tx);
 int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
 void dmu_tx_wait(dmu_tx_t *tx);
 void dmu_tx_commit(dmu_tx_t *tx);
+void dmu_tx_mark_netfree(dmu_tx_t *tx);
 
 /*
  * To register a commit callback, dmu_tx_callback_register() must be called.

module/zfs/dmu.c

@@ -668,6 +668,12 @@ dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
 		tx = dmu_tx_create(os);
 		dmu_tx_hold_free(tx, dn->dn_object,
 		    chunk_begin, chunk_end - chunk_begin);
+
+		/*
+		 * Mark this transaction as typically resulting in a net
+		 * reduction in space used.
+		 */
+		dmu_tx_mark_netfree(tx);
 		err = dmu_tx_assign(tx, TXG_WAIT);
 		if (err) {
 			dmu_tx_abort(tx);
@@ -719,6 +725,7 @@ dmu_free_long_object(objset_t *os, uint64_t object)
 	tx = dmu_tx_create(os);
 	dmu_tx_hold_bonus(tx, object);
 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
+	dmu_tx_mark_netfree(tx);
 	err = dmu_tx_assign(tx, TXG_WAIT);
 	if (err == 0) {
 		err = dmu_object_free(os, object, tx);

module/zfs/dmu_tx.c

@@ -21,7 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  */
 
 #include <sys/dmu.h>
@@ -599,6 +599,32 @@ dmu_tx_count_free(dmu_tx_hold_t *txh, uint64_t off, uint64_t len)
 	txh->txh_space_tounref += unref;
 }
 
+/*
+ * This function marks the transaction as being a "net free".  The end
+ * result is that refquotas will be disabled for this transaction, and
+ * this transaction will be able to use half of the pool space overhead
+ * (see dsl_pool_adjustedsize()).  Therefore this function should only
+ * be called for transactions that we expect will not cause a net increase
+ * in the amount of space used (but it's OK if that is occasionally not true).
+ */
+void
+dmu_tx_mark_netfree(dmu_tx_t *tx)
+{
+	dmu_tx_hold_t *txh;
+
+	txh = dmu_tx_hold_object_impl(tx, tx->tx_objset,
+	    DMU_NEW_OBJECT, THT_FREE, 0, 0);
+
+	/*
+	 * Pretend that this operation will free 1GB of space.  This
+	 * should be large enough to cancel out the largest write.
+	 * We don't want to use something like UINT64_MAX, because that would
+	 * cause overflows when doing math with these values (e.g. in
+	 * dmu_tx_try_assign()).
+	 */
+	txh->txh_space_tofree = txh->txh_space_tounref = 1024 * 1024 * 1024;
+}
+
 void
 dmu_tx_hold_free(dmu_tx_t *tx, uint64_t object, uint64_t off, uint64_t len)
 {

module/zfs/zfs_dir.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
  */
 
 
@@ -578,6 +578,7 @@ zfs_purgedir(znode_t *dzp)
 		dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
 		/* Is this really needed ? */
 		zfs_sa_upgrade_txholds(tx, xzp);
+		dmu_tx_mark_netfree(tx);
 		error = dmu_tx_assign(tx, TXG_WAIT);
 		if (error) {
 			dmu_tx_abort(tx);

module/zfs/zfs_vnops.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2013, 2014 by Delphix. All rights reserved.
  */
 
 /* Portions Copyright 2007 Jeremy Teo */
@@ -1525,6 +1525,9 @@ zfs_remove(struct inode *dip, char *name, cred_t *cr)
 	uint64_t	obj = 0;
 	zfs_dirlock_t	*dl;
 	dmu_tx_t	*tx;
+#if defined(_KERNEL)
+	boolean_t	may_delete_now;
+#endif
 	boolean_t	unlinked;
 	uint64_t	txtype;
 	pathname_t	*realnmp = NULL;
@@ -1584,6 +1587,11 @@ zfs_remove(struct inode *dip, char *name, cred_t *cr)
 		dnlc_remove(dvp, name);
 #endif /* HAVE_DNLC */
 
+#if defined(_KERNEL) && defined(__linux__)
+	may_delete_now = atomic_read(&ip->i_count) == 1
+	    && !(ip->i_state & I_DIRTY_PAGES);
+#endif
+
 	/*
 	 * We never delete the znode and always place it in the unlinked
 	 * set.  The dentry cache will always hold the last reference and
@@ -1609,6 +1617,14 @@ zfs_remove(struct inode *dip, char *name, cred_t *cr)
 	/* charge as an update -- would be nice not to charge at all */
 	dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
 
+	/*
+	 * Mark this transaction as typically resulting in a net free of
+	 * space, unless object removal will be delayed indefinitely
+	 * (due to active holds on the vnode due to the file being open).
+	 */
+	if (may_delete_now)
+		dmu_tx_mark_netfree(tx);
+
 	error = dmu_tx_assign(tx, waited ? TXG_WAITED : TXG_NOWAIT);
 	if (error) {
 		zfs_dirent_unlock(dl);

module/zfs/zfs_znode.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  */
 
 /* Portions Copyright 2007 Jeremy Teo */
@@ -1280,7 +1280,7 @@ zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx)
  *	IN:	zp	- znode of file to free data in.
  *		end	- new end-of-file
  *
- * 	RETURN:	0 on success, error code on failure
+ *	RETURN:	0 on success, error code on failure
  */
 static int
 zfs_extend(znode_t *zp, uint64_t end)
@@ -1395,7 +1395,7 @@ zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
  *		off	- start of section to free.
  *		len	- length of section to free.
  *
- * 	RETURN:	0 on success, error code on failure
+ *	RETURN:	0 on success, error code on failure
  */
 static int
 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
@@ -1474,7 +1474,7 @@ zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
  *	IN:	zp	- znode of file to free data in.
  *		end	- new end-of-file.
  *
- * 	RETURN:	0 on success, error code on failure
+ *	RETURN:	0 on success, error code on failure
  */
 static int
 zfs_trunc(znode_t *zp, uint64_t end)
@@ -1507,6 +1507,7 @@ zfs_trunc(znode_t *zp, uint64_t end)
 	tx = dmu_tx_create(zsb->z_os);
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
 	zfs_sa_upgrade_txholds(tx, zp);
+	dmu_tx_mark_netfree(tx);
 	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
 		dmu_tx_abort(tx);
@@ -1541,7 +1542,7 @@ zfs_trunc(znode_t *zp, uint64_t end)
  *		flag	- current file open mode flags.
  *		log	- TRUE if this action should be logged
  *
- * 	RETURN:	0 on success, error code on failure
+ *	RETURN:	0 on success, error code on failure
  */
 int
 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)

module/zfs/zvol.c

@@ -288,6 +288,7 @@ zvol_update_volsize(uint64_t volsize, objset_t *os)
 
 	tx = dmu_tx_create(os);
 	dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
+	dmu_tx_mark_netfree(tx);
 	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
 		dmu_tx_abort(tx);