Implement fallocate FALLOC_FL_PUNCH_HOLE

Add support for the FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE mode of
fallocate(2).  Mimic the behavior of other native file systems such as
ext4 in cases where the file might be extended. If the offset is beyond
the end of the file, return success without changing the file. If the
extent of the punched hole would extend the file, only the existing tail
of the file is punched.

Add the zfs_zero_partial_page() function, modeled after update_page(),
to handle zeroing partial pages in a hole-punching operation.  It must
be used under a range lock for the requested region in order that the
ARC and page cache stay in sync.

Move the existing page cache truncation via truncate_setsize() into
zfs_freesp() for better source structure compatibility with upstream code.

Add page cache truncation to zfs_freesp() and zfs_free_range() to handle
hole punching.

Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Tim Chase <[email protected]>
Closes #2619

3 files changed, 143 insertions, 21 deletions

diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
index 18b2564a2..ddd997fae 100644
--- a/module/zfs/zfs_vnops.c
+++ b/module/zfs/zfs_vnops.c
@@ -2565,8 +2565,6 @@ top:
 		if (err)
 			goto out3;
 
-		truncate_setsize(ip, vap->va_size);
-
 		/*
 		 * XXX - Note, we are not providing any open
 		 * mode flags here (like FNDELAY), so we may
diff --git a/module/zfs/zfs_znode.c b/module/zfs/zfs_znode.c
index 5fcb9e930..f2e305f7a 100644
--- a/module/zfs/zfs_znode.c
+++ b/module/zfs/zfs_znode.c
@@ -1345,6 +1345,50 @@ zfs_extend(znode_t *zp, uint64_t end)
 }
 
 /*
+ * zfs_zero_partial_page - Modeled after update_pages() but
+ * with different arguments and semantics for use by zfs_freesp().
+ *
+ * Zeroes a piece of a single page cache entry for zp at offset
+ * start and length len.
+ *
+ * Caller must acquire a range lock on the file for the region
+ * being zeroed in order that the ARC and page cache stay in sync.
+ */
+static void
+zfs_zero_partial_page(znode_t *zp, uint64_t start, uint64_t len)
+{
+	struct address_space *mp = ZTOI(zp)->i_mapping;
+	struct page *pp;
+	int64_t	off;
+	void *pb;
+
+	ASSERT((start & PAGE_CACHE_MASK) ==
+	    ((start + len - 1) & PAGE_CACHE_MASK));
+
+	off = start & (PAGE_CACHE_SIZE - 1);
+	start &= PAGE_CACHE_MASK;
+
+	pp = find_lock_page(mp, start >> PAGE_CACHE_SHIFT);
+	if (pp) {
+		if (mapping_writably_mapped(mp))
+			flush_dcache_page(pp);
+
+		pb = kmap(pp);
+		bzero(pb + off, len);
+		kunmap(pp);
+
+		if (mapping_writably_mapped(mp))
+			flush_dcache_page(pp);
+
+		mark_page_accessed(pp);
+		SetPageUptodate(pp);
+		ClearPageError(pp);
+		unlock_page(pp);
+		page_cache_release(pp);
+	}
+}
+
+/*
  * Free space in a file.
  *
  *	IN:	zp	- znode of file to free data in.
@@ -1378,6 +1422,40 @@ zfs_free_range(znode_t *zp, uint64_t off, uint64_t len)
 
 	error = dmu_free_long_range(zsb->z_os, zp->z_id, off, len);
 
+	/*
+	 * Zero partial page cache entries.  This must be done under a
+	 * range lock in order to keep the ARC and page cache in sync.
+	 */
+	if (zp->z_is_mapped) {
+		loff_t first_page, last_page, page_len;
+		loff_t first_page_offset, last_page_offset;
+
+		/* first possible full page in hole */
+		first_page = (off + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+		/* last page of hole */
+		last_page = (off + len) >> PAGE_CACHE_SHIFT;
+
+		/* offset of first_page */
+		first_page_offset = first_page << PAGE_CACHE_SHIFT;
+		/* offset of last_page */
+		last_page_offset = last_page << PAGE_CACHE_SHIFT;
+
+		if (first_page > last_page) {
+			/* entire punched area within a single page */
+			zfs_zero_partial_page(zp, off, len);
+		} else {
+			/* beginning of punched area at the end of a page */
+			page_len  = first_page_offset - off;
+			if (page_len > 0)
+				zfs_zero_partial_page(zp, off, page_len);
+
+			/* end of punched area at the beginning of a page */
+			page_len = off + len - last_page_offset;
+			if (page_len > 0)
+				zfs_zero_partial_page(zp, last_page_offset,
+				    page_len);
+		}
+	}
 	zfs_range_unlock(rl);
 
 	return (error);
@@ -1479,8 +1557,7 @@ zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
 		error =  zfs_extend(zp, off+len);
 		if (error == 0 && log)
 			goto log;
-		else
-			return (error);
+		goto out;
 	}
 
 	/*
@@ -1500,7 +1577,7 @@ zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log)
 			error = zfs_extend(zp, off+len);
 	}
 	if (error || !log)
-		return (error);
+		goto out;
 log:
 	tx = dmu_tx_create(zsb->z_os);
 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
@@ -1508,7 +1585,7 @@ log:
 	error = dmu_tx_assign(tx, TXG_WAIT);
 	if (error) {
 		dmu_tx_abort(tx);
-		return (error);
+		goto out;
 	}
 
 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zsb), NULL, mtime, 16);
@@ -1522,8 +1599,40 @@ log:
 	zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len);
 
 	dmu_tx_commit(tx);
+
 	zfs_inode_update(zp);
-	return (0);
+	error = 0;
+
+out:
+	/*
+	 * Truncate the page cache - for file truncate operations, use
+	 * the purpose-built API for truncations.  For punching operations,
+	 * truncate only whole pages within the region; partial pages are
+	 * zeroed under a range lock in zfs_free_range().
+	 */
+	if (len == 0)
+		truncate_setsize(ZTOI(zp), off);
+	else if (zp->z_is_mapped) {
+		loff_t first_page, last_page;
+		loff_t first_page_offset, last_page_offset;
+
+		/* first possible full page in hole */
+		first_page = (off + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+		/* last page of hole */
+		last_page = (off + len) >> PAGE_CACHE_SHIFT;
+
+		/* offset of first_page */
+		first_page_offset = first_page << PAGE_CACHE_SHIFT;
+		/* offset of last_page */
+		last_page_offset = last_page << PAGE_CACHE_SHIFT;
+
+		/* truncate whole pages */
+		if (last_page_offset > first_page_offset) {
+			truncate_inode_pages_range(ZTOI(zp)->i_mapping,
+			    first_page_offset, last_page_offset - 1);
+		}
+	}
+	return (error);
 }
 
 void
diff --git a/module/zfs/zpl_file.c b/module/zfs/zpl_file.c
index 5ea892320..c72d5c947 100644
--- a/module/zfs/zpl_file.c
+++ b/module/zfs/zpl_file.c
@@ -558,38 +558,53 @@ zpl_writepage(struct page *pp, struct writeback_control *wbc)
 
 /*
  * The only flag combination which matches the behavior of zfs_space()
- * is FALLOC_FL_PUNCH_HOLE.  This flag was introduced in the 2.6.38 kernel.
+ * is FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE.  The FALLOC_FL_PUNCH_HOLE
+ * flag was introduced in the 2.6.38 kernel.
  */
+#if defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE)
 long
 zpl_fallocate_common(struct inode *ip, int mode, loff_t offset, loff_t len)
 {
-	cred_t *cr = CRED();
 	int error = -EOPNOTSUPP;
 
-	if (mode & FALLOC_FL_KEEP_SIZE)
-		return (-EOPNOTSUPP);
+#if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
+	cred_t *cr = CRED();
+	flock64_t bf;
+	loff_t olen;
+
+	if (mode != (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
+		return (error);
 
 	crhold(cr);
 
-#ifdef FALLOC_FL_PUNCH_HOLE
-	if (mode & FALLOC_FL_PUNCH_HOLE) {
-		flock64_t bf;
+	if (offset < 0 || len <= 0)
+		return (-EINVAL);
 
-		bf.l_type = F_WRLCK;
-		bf.l_whence = 0;
-		bf.l_start = offset;
-		bf.l_len = len;
-		bf.l_pid = 0;
+	spl_inode_lock(ip);
+	olen = i_size_read(ip);
 
-		error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
+	if (offset > olen) {
+		spl_inode_unlock(ip);
+		return (0);
 	}
-#endif /* FALLOC_FL_PUNCH_HOLE */
+	if (offset + len > olen)
+		len = olen - offset;
+	bf.l_type = F_WRLCK;
+	bf.l_whence = 0;
+	bf.l_start = offset;
+	bf.l_len = len;
+	bf.l_pid = 0;
+
+	error = -zfs_space(ip, F_FREESP, &bf, FWRITE, offset, cr);
+	spl_inode_unlock(ip);
 
 	crfree(cr);
+#endif /* defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) */
 
 	ASSERT3S(error, <=, 0);
 	return (error);
 }
+#endif /* defined(HAVE_FILE_FALLOCATE) || defined(HAVE_INODE_FALLOCATE) */
 
 #ifdef HAVE_FILE_FALLOCATE
 static long