openzfs · behlendorf · Feb 14, 2023 · Jan 30, 2023
diff --git a/config/kernel-filemap.m4 b/config/kernel-filemap.m4
@@ -0,0 +1,26 @@
+dnl #
+dnl # filemap_range_has_page was not available till 4.13
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_FILEMAP], [
+	ZFS_LINUX_TEST_SRC([filemap_range_has_page], [
+		#include <linux/fs.h>
+	],[
+		struct address_space *mapping = NULL;
+		loff_t lstart = 0;
+		loff_t lend = 0;
+		bool ret __attribute__ ((unused));
+
+		ret = filemap_range_has_page(mapping, lstart, lend);
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_FILEMAP], [
+	AC_MSG_CHECKING([whether filemap_range_has_page() is available])
+	ZFS_LINUX_TEST_RESULT([filemap_range_has_page], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_FILEMAP_RANGE_HAS_PAGE, 1,
+		[filemap_range_has_page() is available])
+	],[
+		AC_MSG_RESULT(no)
+	])
+])
diff --git a/config/kernel.m4 b/config/kernel.m4
@@ -150,6 +150,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_USER_NS_COMMON_INUM
 	ZFS_AC_KERNEL_SRC_IDMAP_MNT_API
 	ZFS_AC_KERNEL_SRC_IATTR_VFSID
+	ZFS_AC_KERNEL_SRC_FILEMAP
 
 	AC_MSG_CHECKING([for available kernel interfaces])
 	ZFS_LINUX_TEST_COMPILE_ALL([kabi])
@@ -273,6 +274,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_USER_NS_COMMON_INUM
 	ZFS_AC_KERNEL_IDMAP_MNT_API
 	ZFS_AC_KERNEL_IATTR_VFSID
+	ZFS_AC_KERNEL_FILEMAP
 ])
 
 dnl #

diff --git a/include/os/freebsd/zfs/sys/zfs_znode_impl.h b/include/os/freebsd/zfs/sys/zfs_znode_impl.h
@@ -116,7 +116,8 @@ typedef struct zfs_soft_state {
 #define	Z_ISLNK(type) ((type) == VLNK)
 #define	Z_ISDIR(type) ((type) == VDIR)
 
-#define	zn_has_cached_data(zp)		vn_has_cached_data(ZTOV(zp))
+#define	zn_has_cached_data(zp, start, end) \
+    vn_has_cached_data(ZTOV(zp))
 #define	zn_flush_cached_data(zp, sync)	vn_flush_cached_data(ZTOV(zp), sync)
 #define	zn_rlimit_fsize(zp, uio) \
     vn_rlimit_fsize(ZTOV(zp), GET_UIO_STRUCT(uio), zfs_uio_td(uio))

diff --git a/include/os/linux/zfs/sys/trace_acl.h b/include/os/linux/zfs/sys/trace_acl.h
@@ -62,7 +62,6 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __field(uint32_t,		z_async_writes_cnt)
 	    __field(mode_t,		z_mode)
 	    __field(boolean_t,		z_is_sa)
-	    __field(boolean_t,		z_is_mapped)
 	    __field(boolean_t,		z_is_ctldir)
 
 	    __field(uint32_t,		i_uid)
@@ -96,7 +95,6 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __entry->z_async_writes_cnt	= zn->z_async_writes_cnt;
 	    __entry->z_mode		= zn->z_mode;
 	    __entry->z_is_sa		= zn->z_is_sa;
-	    __entry->z_is_mapped	= zn->z_is_mapped;
 	    __entry->z_is_ctldir	= zn->z_is_ctldir;
 
 	    __entry->i_uid		= KUID_TO_SUID(ZTOI(zn)->i_uid);
@@ -119,7 +117,7 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    "zn_prefetch %u blksz %u seq %u "
 	    "mapcnt %llu size %llu pflags %llu "
 	    "sync_cnt %u sync_writes_cnt %u async_writes_cnt %u "
-	    "mode 0x%x is_sa %d is_mapped %d is_ctldir %d "
+	    "mode 0x%x is_sa %d is_ctldir %d "
 	    "inode { uid %u gid %u ino %lu nlink %u size %lli "
 	    "blkbits %u bytes %u mode 0x%x generation %x } } "
 	    "ace { type %u flags %u access_mask %u } mask_matched %u",
@@ -128,9 +126,8 @@ DECLARE_EVENT_CLASS(zfs_ace_class,
 	    __entry->z_seq, __entry->z_mapcnt, __entry->z_size,
 	    __entry->z_pflags, __entry->z_sync_cnt,
 	    __entry->z_sync_writes_cnt, __entry->z_async_writes_cnt,
-	    __entry->z_mode, __entry->z_is_sa, __entry->z_is_mapped,
-	    __entry->z_is_ctldir, __entry->i_uid,
-	    __entry->i_gid, __entry->i_ino, __entry->i_nlink,
+	    __entry->z_mode, __entry->z_is_sa, __entry->z_is_ctldir,
+	    __entry->i_uid, __entry->i_gid, __entry->i_ino, __entry->i_nlink,
 	    __entry->i_size, __entry->i_blkbits,
 	    __entry->i_bytes, __entry->i_mode, __entry->i_generation,
 	    __entry->z_type, __entry->z_flags, __entry->z_access_mask,

diff --git a/include/os/linux/zfs/sys/zfs_znode_impl.h b/include/os/linux/zfs/sys/zfs_znode_impl.h
@@ -47,9 +47,16 @@
 extern "C" {
 #endif
 
+#if defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
 #define	ZNODE_OS_FIELDS			\
 	inode_timespec_t z_btime; /* creation/birth time (cached) */ \
 	struct inode	z_inode;
+#else
+#define	ZNODE_OS_FIELDS			\
+	inode_timespec_t z_btime; /* creation/birth time (cached) */ \
+	struct inode	z_inode;                                     \
+	boolean_t	z_is_mapped;    /* we are mmap'ed */
+#endif
 
 /*
  * Convert between znode pointers and inode pointers
@@ -70,7 +77,14 @@ extern "C" {
 #define	Z_ISDEV(type)	(S_ISCHR(type) || S_ISBLK(type) || S_ISFIFO(type))
 #define	Z_ISDIR(type)	S_ISDIR(type)
 
-#define	zn_has_cached_data(zp)		((zp)->z_is_mapped)
+#if defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
+#define	zn_has_cached_data(zp, start, end) \
+	filemap_range_has_page(ZTOI(zp)->i_mapping, start, end)
+#else
+#define	zn_has_cached_data(zp, start, end) \
+	((zp)->z_is_mapped)
+#endif
+
 #define	zn_flush_cached_data(zp, sync)	write_inode_now(ZTOI(zp), sync)
 #define	zn_rlimit_fsize(zp, uio)	(0)
 

diff --git a/include/sys/zfs_znode.h b/include/sys/zfs_znode.h
@@ -188,7 +188,6 @@ typedef struct znode {
 	boolean_t	z_atime_dirty;	/* atime needs to be synced */
 	boolean_t	z_zn_prefetch;	/* Prefetch znodes? */
 	boolean_t	z_is_sa;	/* are we native sa? */
-	boolean_t	z_is_mapped;	/* are we mmap'ed */
 	boolean_t	z_is_ctldir;	/* are we .zfs entry */
 	boolean_t	z_suspended;	/* extra ref from a suspend? */
 	uint_t		z_blksz;	/* block size in bytes */

diff --git a/module/os/linux/zfs/zfs_ctldir.c b/module/os/linux/zfs/zfs_ctldir.c
@@ -485,7 +485,9 @@ zfsctl_inode_alloc(zfsvfs_t *zfsvfs, uint64_t id,
 	zp->z_atime_dirty = B_FALSE;
 	zp->z_zn_prefetch = B_FALSE;
 	zp->z_is_sa = B_FALSE;
+#if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
 	zp->z_is_mapped = B_FALSE;
+#endif
 	zp->z_is_ctldir = B_TRUE;
 	zp->z_sa_hdl = NULL;
 	zp->z_blksz = 0;

diff --git a/module/os/linux/zfs/zfs_vnops_os.c b/module/os/linux/zfs/zfs_vnops_os.c
@@ -987,7 +987,7 @@ zfs_remove(znode_t *dzp, char *name, cred_t *cr, int flags)
 
 	mutex_enter(&zp->z_lock);
 	may_delete_now = atomic_read(&ZTOI(zp)->i_count) == 1 &&
-	    !(zp->z_is_mapped);
+	    !zn_has_cached_data(zp, 0, LLONG_MAX);
 	mutex_exit(&zp->z_lock);
 
 	/*
@@ -1075,7 +1075,8 @@ zfs_remove(znode_t *dzp, char *name, cred_t *cr, int flags)
 		    &xattr_obj_unlinked, sizeof (xattr_obj_unlinked));
 		delete_now = may_delete_now && !toobig &&
 		    atomic_read(&ZTOI(zp)->i_count) == 1 &&
-		    !(zp->z_is_mapped) && xattr_obj == xattr_obj_unlinked &&
+		    !zn_has_cached_data(zp, 0, LLONG_MAX) &&
+		    xattr_obj == xattr_obj_unlinked &&
 		    zfs_external_acl(zp) == acl_obj;
 	}
 

diff --git a/module/os/linux/zfs/zfs_znode.c b/module/os/linux/zfs/zfs_znode.c
@@ -551,7 +551,9 @@ zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz,
 	ASSERT3P(zp->z_xattr_cached, ==, NULL);
 	zp->z_unlinked = B_FALSE;
 	zp->z_atime_dirty = B_FALSE;
+#if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
 	zp->z_is_mapped = B_FALSE;
+#endif
 	zp->z_is_ctldir = B_FALSE;
 	zp->z_suspended = B_FALSE;
 	zp->z_sa_hdl = NULL;
@@ -1641,7 +1643,7 @@ zfs_free_range(znode_t *zp, uint64_t off, uint64_t 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) {
+	if (zn_has_cached_data(zp, off, off + len - 1)) {
 		loff_t first_page, last_page, page_len;
 		loff_t first_page_offset, last_page_offset;
 

diff --git a/module/os/linux/zfs/zpl_file.c b/module/os/linux/zfs/zpl_file.c
@@ -625,7 +625,6 @@ static int
 zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 {
 	struct inode *ip = filp->f_mapping->host;
-	znode_t *zp = ITOZ(ip);
 	int error;
 	fstrans_cookie_t cookie;
 
@@ -640,9 +639,12 @@ zpl_mmap(struct file *filp, struct vm_area_struct *vma)
 	if (error)
 		return (error);
 
+#if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
+	znode_t *zp = ITOZ(ip);
 	mutex_enter(&zp->z_lock);
 	zp->z_is_mapped = B_TRUE;
 	mutex_exit(&zp->z_lock);
+#endif
 
 	return (error);
 }
@@ -937,7 +939,7 @@ zpl_fadvise(struct file *filp, loff_t offset, loff_t len, int advice)
 	case POSIX_FADV_SEQUENTIAL:
 	case POSIX_FADV_WILLNEED:
 #ifdef HAVE_GENERIC_FADVISE
-		if (zn_has_cached_data(zp))
+		if (zn_has_cached_data(zp, offset, offset + len - 1))
 			error = generic_fadvise(filp, offset, len, advice);
 #endif
 		/*

diff --git a/module/zfs/zfs_vnops.c b/module/zfs/zfs_vnops.c
@@ -106,7 +106,7 @@ zfs_holey_common(znode_t *zp, ulong_t cmd, loff_t *off)
 		hole = B_FALSE;
 
 	/* Flush any mmap()'d data to disk */
-	if (zn_has_cached_data(zp))
+	if (zn_has_cached_data(zp, 0, file_sz - 1))
 		zn_flush_cached_data(zp, B_FALSE);
 
 	lr = zfs_rangelock_enter(&zp->z_rangelock, 0, file_sz, RL_READER);
@@ -288,7 +288,8 @@ zfs_read(struct znode *zp, zfs_uio_t *uio, int ioflag, cred_t *cr)
 			error = mappedread_sf(zp, nbytes, uio);
 		else
 #endif
-		if (zn_has_cached_data(zp) && !(ioflag & O_DIRECT)) {
+		if (zn_has_cached_data(zp, zfs_uio_offset(uio),
+		    zfs_uio_offset(uio) + nbytes - 1) && !(ioflag & O_DIRECT)) {
 			error = mappedread(zp, nbytes, uio);
 		} else {
 			error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl),
@@ -696,7 +697,8 @@ zfs_write(znode_t *zp, zfs_uio_t *uio, int ioflag, cred_t *cr)
 			zfs_uioskip(uio, nbytes);
 			tx_bytes = nbytes;
 		}
-		if (tx_bytes && zn_has_cached_data(zp) &&
+		if (tx_bytes &&
+		    zn_has_cached_data(zp, woff, woff + tx_bytes - 1) &&
 		    !(ioflag & O_DIRECT)) {
 			update_pages(zp, woff, tx_bytes, zfsvfs->z_os);
 		}