dmu: Allow buffer fills to fail

When ZFS overwrites a whole block, it does not bother to read the
old content from disk. It is a good optimization, but if the buffer
fill fails due to page fault or something else, the buffer ends up
corrupted, neither keeping old content, nor getting the new one.

On FreeBSD this is additionally complicated by page faults being
blocked by VFS layer, always returning EFAULT on attempt to write
from mmap()'ed but not yet cached address range.  Normally it is
not a big problem, since after original failure VFS will retry the
write after reading the required data.  The problem becomes worse
in specific case when somebody tries to write into a file its own
mmap()'ed content from the same location.  In that situation the
only copy of the data is getting corrupted on the page fault and
the following retries only fixate the status quo.  Block cloning
makes this issue easier to reproduce, since it does not read the
old data, unlike traditional file copy, that may work by chance.

This patch provides the fill status to dmu_buf_fill_done(), that
in case of error can destroy the corrupted buffer as if no write
happened.  One more complication in case of block cloning is that
if error is possible during fill, dmu_buf_will_fill() must read
the data via fall-back to dmu_buf_will_dirty().  It is required
to allow in case of error restoring the buffer to a state after
the cloning, not not before it, that would happen if we just call
dbuf_undirty().

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes openzfs#15665

include/os/freebsd/spl/sys/uio.h

@@ -62,7 +62,7 @@ zfs_uio_setoffset(zfs_uio_t *uio, offset_t off)
 }
 
 static inline void
-zfs_uio_advance(zfs_uio_t *uio, size_t size)
+zfs_uio_advance(zfs_uio_t *uio, ssize_t size)
 {
 	zfs_uio_resid(uio) -= size;
 	zfs_uio_offset(uio) += size;

include/os/linux/spl/sys/uio.h

@@ -95,7 +95,7 @@ zfs_uio_setoffset(zfs_uio_t *uio, offset_t off)
 }
 
 static inline void
-zfs_uio_advance(zfs_uio_t *uio, size_t size)
+zfs_uio_advance(zfs_uio_t *uio, ssize_t size)
 {
 	uio->uio_resid -= size;
 	uio->uio_loffset += size;

include/sys/dbuf.h

@@ -379,8 +379,8 @@ dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
 void dmu_buf_will_clone(dmu_buf_t *db, dmu_tx_t *tx);
 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
-void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
-void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
+void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx, boolean_t canfail);
+boolean_t dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx, boolean_t failed);
 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid,

lib/libspl/include/sys/uio.h

@@ -90,7 +90,7 @@ zfs_uio_iov_at_index(zfs_uio_t *uio, uint_t idx, void **base, uint64_t *len)
 }
 
 static inline void
-zfs_uio_advance(zfs_uio_t *uio, size_t size)
+zfs_uio_advance(zfs_uio_t *uio, ssize_t size)
 {
 	uio->uio_resid -= size;
 	uio->uio_loffset += size;

module/os/freebsd/zfs/dmu_os.c

@@ -110,7 +110,7 @@ dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
 
 		if (tocpy == db->db_size)
-			dmu_buf_will_fill(db, tx);
+			dmu_buf_will_fill(db, tx, B_FALSE);
 		else
 			dmu_buf_will_dirty(db, tx);
 
@@ -126,7 +126,7 @@ dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 		}
 
 		if (tocpy == db->db_size)
-			dmu_buf_fill_done(db, tx);
+			dmu_buf_fill_done(db, tx, B_FALSE);
 
 		offset += tocpy;
 		size -= tocpy;

module/zfs/dbuf.c

@@ -2734,7 +2734,7 @@ dmu_buf_will_not_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
 }
 
 void
-dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
+dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx, boolean_t canfail)
 {
 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake;
 
@@ -2752,8 +2752,14 @@ dmu_buf_will_fill(dmu_buf_t *db_fake, dmu_tx_t *tx)
 		 * Block cloning: We will be completely overwriting a block
 		 * cloned in this transaction group, so let's undirty the
 		 * pending clone and mark the block as uncached. This will be
-		 * as if the clone was never done.
+		 * as if the clone was never done.  But if the fill can fail
+		 * we should have a way to return back to the cloned data.
 		 */
+		if (canfail && dbuf_find_dirty_eq(db, tx->tx_txg) != NULL) {
+			mutex_exit(&db->db_mtx);
+			dmu_buf_will_dirty(db_fake, tx);
+			return;
+		}
 		VERIFY(!dbuf_undirty(db, tx));
 		db->db_state = DB_UNCACHED;
 	}
@@ -2814,32 +2820,41 @@ dbuf_override_impl(dmu_buf_impl_t *db, const blkptr_t *bp, dmu_tx_t *tx)
 	dl->dr_overridden_by.blk_birth = dr->dr_txg;
 }
 
-void
-dmu_buf_fill_done(dmu_buf_t *dbuf, dmu_tx_t *tx)
+boolean_t
+dmu_buf_fill_done(dmu_buf_t *dbuf, dmu_tx_t *tx, boolean_t failed)
 {
 	(void) tx;
 	dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbuf;
-	dbuf_states_t old_state;
 	mutex_enter(&db->db_mtx);
 	DBUF_VERIFY(db);
 
-	old_state = db->db_state;
-	db->db_state = DB_CACHED;
-	if (old_state == DB_FILL) {
+	if (db->db_state == DB_FILL) {
 		if (db->db_level == 0 && db->db_freed_in_flight) {
 			ASSERT(db->db_blkid != DMU_BONUS_BLKID);
 			/* we were freed while filling */
 			/* XXX dbuf_undirty? */
 			memset(db->db.db_data, 0, db->db.db_size);
 			db->db_freed_in_flight = FALSE;
+			db->db_state = DB_CACHED;
 			DTRACE_SET_STATE(db,
 			    "fill done handling freed in flight");
+			failed = B_FALSE;
+		} else if (failed) {
+			VERIFY(!dbuf_undirty(db, tx));
+			db->db_buf = NULL;
+			dbuf_clear_data(db);
+			DTRACE_SET_STATE(db, "fill failed");
 		} else {
+			db->db_state = DB_CACHED;
 			DTRACE_SET_STATE(db, "fill done");
 		}
 		cv_broadcast(&db->db_changed);
+	} else {
+		db->db_state = DB_CACHED;
+		failed = B_FALSE;
 	}
 	mutex_exit(&db->db_mtx);
+	return (failed);
 }
 
 void
@@ -2984,7 +2999,7 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
 	DTRACE_SET_STATE(db, "filling assigned arcbuf");
 	mutex_exit(&db->db_mtx);
 	(void) dbuf_dirty(db, tx);
-	dmu_buf_fill_done(&db->db, tx);
+	dmu_buf_fill_done(&db->db, tx, B_FALSE);
 }
 
 void

module/zfs/dmu.c

@@ -1115,14 +1115,14 @@ dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size,
 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
 
 		if (tocpy == db->db_size)
-			dmu_buf_will_fill(db, tx);
+			dmu_buf_will_fill(db, tx, B_FALSE);
 		else
 			dmu_buf_will_dirty(db, tx);
 
 		(void) memcpy((char *)db->db_data + bufoff, buf, tocpy);
 
 		if (tocpy == db->db_size)
-			dmu_buf_fill_done(db, tx);
+			dmu_buf_fill_done(db, tx, B_FALSE);
 
 		offset += tocpy;
 		size -= tocpy;
@@ -1330,27 +1330,24 @@ dmu_write_uio_dnode(dnode_t *dn, zfs_uio_t *uio, uint64_t size, dmu_tx_t *tx)
 
 		ASSERT(size > 0);
 
-		bufoff = zfs_uio_offset(uio) - db->db_offset;
+		offset_t off = zfs_uio_offset(uio);
+		bufoff = off - db->db_offset;
 		tocpy = MIN(db->db_size - bufoff, size);
 
 		ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
 
 		if (tocpy == db->db_size)
-			dmu_buf_will_fill(db, tx);
+			dmu_buf_will_fill(db, tx, B_TRUE);
 		else
 			dmu_buf_will_dirty(db, tx);
 
-		/*
-		 * XXX zfs_uiomove could block forever (eg.nfs-backed
-		 * pages).  There needs to be a uiolockdown() function
-		 * to lock the pages in memory, so that zfs_uiomove won't
-		 * block.
-		 */
 		err = zfs_uio_fault_move((char *)db->db_data + bufoff,
 		    tocpy, UIO_WRITE, uio);
 
-		if (tocpy == db->db_size)
-			dmu_buf_fill_done(db, tx);
+		if (tocpy == db->db_size && dmu_buf_fill_done(db, tx, err)) {
+			/* The fill was reverted.  Undo any uio progress. */
+			zfs_uio_advance(uio, off - zfs_uio_offset(uio));
+		}
 
 		if (err)
 			break;

module/zfs/dmu_recv.c

@@ -2532,7 +2532,7 @@ receive_spill(struct receive_writer_arg *rwa, struct drr_spill *drrs,
 	 * size of the provided arc_buf_t.
 	 */
 	if (db_spill->db_size != drrs->drr_length) {
-		dmu_buf_will_fill(db_spill, tx);
+		dmu_buf_will_fill(db_spill, tx, B_FALSE);
 		VERIFY0(dbuf_spill_set_blksz(db_spill,
 		    drrs->drr_length, tx));
 	}