Implement pageoutv2 feature

Remodel the pageout to use V2 style calling so we control when the allocate
UPL, and model it after HFS vnop_pageout.

include/sys/hfs_internal.h

@@ -99,6 +99,10 @@ struct user64_ext_access_t {
 
 #define HFSIOC_GETPATH  _IOWR('h', 13, pathname_t)
 #define HFS_GETPATH  IOCBASECMD(HFSIOC_GETPATH)
+#define HFS_GETPATH_VOLUME_RELATIVE	0x1
+
+/* This define is deemed secret by Apple */
+#define BUILDPATH_VOLUME_RELATIVE 0x8 /* Return path relative to the nearest mount point */
 
 /* Enable/disable extent-based extended attributes */
 #define HFSIOC_SET_XATTREXTENTS_STATE  _IOW('h', 14, u_int32_t)
@@ -146,6 +150,16 @@ struct user64_ext_access_t {
 #define	HFSIOC_GET_DESIRED_DISK	_IOR('h', 29, u_int32_t)
 #define	HFS_FSCTL_GET_DESIRED_DISK	IOCBASECMD(HFSIOC_GET_DESIRED_DISK)
 
+/* revisiond only uses this when something transforms in a way the kernel can't track
+such as "foo.rtf" -> "foo.rtfd" */
+#define HFSIOC_TRANSFER_DOCUMENT_ID  _IOW('h', 32, u_int32_t)
+#define HFS_TRANSFER_DOCUMENT_ID  IOCBASECMD(HFSIOC_TRANSFER_DOCUMENT_ID)
+
+
+/* fcntl.h */
+#define F_MAKECOMPRESSED                80 /* Make the file compressed; truncate & toggle BSD bits */
+
+
 // END of definitions
 
 #endif

include/sys/zfs_vfsops.h

@@ -46,6 +46,7 @@ struct znode;
 /* #define WITH_SEARCHFS */
 /* #define WITH_READDIRATTR */
 #define	HAVE_NAMED_STREAMS 1
+#define	HAVE_PAGEOUT_V2 1
 #endif
 
 

include/sys/zfs_znode.h

@@ -66,6 +66,14 @@ extern "C" {
 #define	ZFS_OFFLINE		0x0000100000000000ull
 #define	ZFS_SPARSE		0x0000200000000000ull
 
+#ifdef __APPLE__
+	/* Unsure how we officially register new flags bits, but
+	 * I guess we will claim the whole nibble for OSX
+	 * 0x00n0000000000000ull : n = 1 2 4 8
+	 */
+#define	ZFS_TRACKED		0x0010000000000000ull
+#endif
+
 #define	ZFS_ATTR_SET(zp, attr, value, pflags, tx) \
 { \
 	if (value) \
@@ -237,6 +245,7 @@ typedef struct znode {
 #ifdef __APPLE__
 	list_node_t	z_link_reclaim_node;	/* all reclaim znodes in fs link */
     uint32_t    z_vid;  /* OSX vnode_vid */
+	uint32_t    z_document_id;
     /* Track vnop_lookup name for Finder - not for anything else */
     char        z_finder_hardlink_name[MAXPATHLEN];
 	boolean_t   z_fastpath;
@@ -290,7 +299,11 @@ typedef struct znode {
  */
 #define ZFS_ENTER(zfsvfs)                       \
     {                                                                   \
-		rrm_enter_read(&(zfsvfs)->z_teardown_lock, FTAG); \
+		if (!POINTER_IS_VALID(zfsvfs)) {								\
+			printf("ZFS: ZFS_ENTER on released %s:%d",					\
+				   __FILE__, __LINE__);									\
+			return EIO; }												\
+		rrm_enter_read(&(zfsvfs)->z_teardown_lock, FTAG);				\
         if ((zfsvfs)->z_unmounted) {                                    \
             ZFS_EXIT(zfsvfs);                                           \
             return (EIO);                                               \
@@ -445,6 +458,13 @@ int zfs_attach_vnode(znode_t *zp);
 uint32_t zfs_getbsdflags(znode_t *zp);
 void zfs_setbsdflags(znode_t *zp, uint32_t bsdflags);
 void zfs_time_stamper_locked(znode_t *zp, uint_t flag, dmu_tx_t *tx);
+int zfs_setattr_set_documentid(znode_t *zp, boolean_t update_flags);
+void zfs_setattr_generate_id(znode_t *zp, uint64_t val, char *name);
+
+#define FNV1_32A_INIT ((uint32_t)0x811c9dc5)
+uint32_t fnv_32a_str(const char *str, uint32_t hval);
+uint32_t fnv_32a_buf(void *buf, size_t len, uint32_t hval);
+
 
 #ifdef ZFS_DEBUG
 typedef enum whereami {

module/zfs/dmu_tx.c

@@ -1297,6 +1297,8 @@ dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how)
 	    txg_how == TXG_WAITED);
 	ASSERT(!dsl_pool_sync_context(tx->tx_pool));
 
+	VERIFY0(dsl_pool_sync_context(tx->tx_pool));
+
 	if (txg_how == TXG_WAITED)
 		tx->tx_waited = B_TRUE;
 

module/zfs/zfs_osx.cpp

@@ -143,93 +143,10 @@ zfs_vfs_sysctl(int *name, __unused u_int namelen, user_addr_t oldp, size_t *oldl
 #endif /* __APPLE__ */
 
 
-
-
-
 #include <sys/utsname.h>
 #include <string.h>
 
 
-/*
- * fnv_32a_str - perform a 32 bit Fowler/Noll/Vo FNV-1a hash on a string
- *
- * input:
- *	str	- string to hash
- *	hval	- previous hash value or 0 if first call
- *
- * returns:
- *	32 bit hash as a static hash type
- *
- * NOTE: To use the recommended 32 bit FNV-1a hash, use FNV1_32A_INIT as the
- *  	 hval arg on the first call to either fnv_32a_buf() or fnv_32a_str().
- */
-#define FNV1_32A_INIT ((uint32_t)0x811c9dc5)
-uint32_t
-fnv_32a_str(const char *str, uint32_t hval)
-{
-    unsigned char *s = (unsigned char *)str;	/* unsigned string */
-
-    /*
-     * FNV-1a hash each octet in the buffer
-     */
-    while (*s) {
-
-	/* xor the bottom with the current octet */
-	hval ^= (uint32_t)*s++;
-
-	/* multiply by the 32 bit FNV magic prime mod 2^32 */
-#if defined(NO_FNV_GCC_OPTIMIZATION)
-	hval *= FNV_32_PRIME;
-#else
-	hval += (hval<<1) + (hval<<4) + (hval<<7) + (hval<<8) + (hval<<24);
-#endif
-    }
-
-    /* return our new hash value */
-    return hval;
-}
-
-/*
- * fnv_32a_buf - perform a 32 bit Fowler/Noll/Vo FNV-1a hash on a buffer
- *
- * input:
- *buf- start of buffer to hash
- *len- length of buffer in octets
- *hval- previous hash value or 0 if first call
- *
- * returns:
- *32 bit hash as a static hash type
- *
- * NOTE: To use the recommended 32 bit FNV-1a hash, use FNV1_32A_INIT as the
- *  hval arg on the first call to either fnv_32a_buf() or fnv_32a_str().
- */
-uint32_t
-fnv_32a_buf(void *buf, size_t len, uint32_t hval)
-{
-    unsigned char *bp = (unsigned char *)buf;/* start of buffer */
-    unsigned char *be = bp + len;/* beyond end of buffer */
-
-    /*
-     * FNV-1a hash each octet in the buffer
-     */
-    while (bp < be) {
-
-		/* xor the bottom with the current octet */
-		hval ^= (uint32_t)*bp++;
-
-		/* multiply by the 32 bit FNV magic prime mod 2^32 */
-#if defined(NO_FNV_GCC_OPTIMIZATION)
-		hval *= FNV_32_PRIME;
-#else
-		hval += (hval<<1) + (hval<<4) + (hval<<7) + (hval<<8) + (hval<<24);
-#endif
-    }
-
-    /* return our new hash value */
-    return hval;
-}
-
-
 } // Extern "C"
 
 

module/zfs/zfs_vnops.c

@@ -338,7 +338,7 @@ update_pages(vnode_t *vp, int64_t nbytes, struct uio *uio,
              dmu_tx_t *tx)
 {
     znode_t *zp = VTOZ(vp);
-    zfsvfs_t *zfsvfs = zp->z_zfsvfs;
+    //zfsvfs_t *zfsvfs = zp->z_zfsvfs;
     int len = nbytes;
     int error = 0;
     vm_offset_t vaddr = 0;
@@ -360,15 +360,22 @@ update_pages(vnode_t *vp, int64_t nbytes, struct uio *uio,
      * Create a UPL for the current range and map its
      * page list into the kernel virtual address space.
      */
-    if ( ubc_create_upl(vp, upl_start, upl_size, &upl, NULL,
-                        UPL_FILE_IO | UPL_SET_LITE) == KERN_SUCCESS ) {
-        pl = ubc_upl_pageinfo(upl);
-        ubc_upl_map(upl, &vaddr);
-    }
+    error = ubc_create_upl(vp, upl_start, upl_size, &upl, &pl,
+						   UPL_FILE_IO | UPL_SET_LITE);
+	if ((error != KERN_SUCCESS) || !upl) {
+		printf("ZFS: update_pages failed to ubc_create_upl: %d\n", error);
+		return;
+	}
+
+	if (ubc_upl_map(upl, &vaddr) != KERN_SUCCESS) {
+		printf("ZFS: update_pages failed to ubc_upl_map: %d\n", error);
+		(void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
+		return;
+	}
 
     for (upl_page = 0; len > 0; ++upl_page) {
         uint64_t bytes = MIN(PAGESIZE - off, len);
-        uint64_t woff = uio_offset(uio);
+        //uint64_t woff = uio_offset(uio);
         /*
          * We don't want a new page to "appear" in the middle of
          * the file update (because it may not get the write
@@ -417,14 +424,12 @@ update_pages(vnode_t *vp, int64_t nbytes, struct uio *uio,
     /*
      * Unmap the page list and free the UPL.
      */
-    if (pl) {
-        (void) ubc_upl_unmap(upl);
-        /*
-         * We want to abort here since due to dmu_write()
-         * we effectively didn't dirty any pages.
-         */
-        (void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
-    }
+	(void) ubc_upl_unmap(upl);
+	/*
+	 * We want to abort here since due to dmu_write()
+	 * we effectively didn't dirty any pages.
+	 */
+	(void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
 
 }
 #endif
@@ -589,11 +594,18 @@ mappedread(vnode_t *vp, int nbytes, struct uio *uio)
      * Create a UPL for the current range and map its
      * page list into the kernel virtual address space.
      */
-    if ( ubc_create_upl(vp, upl_start, upl_size, &upl, NULL,
-                        UPL_FILE_IO | UPL_SET_LITE) == KERN_SUCCESS ) {
-        pl = ubc_upl_pageinfo(upl);
-        ubc_upl_map(upl, &vaddr);
-    }
+    error = ubc_create_upl(vp, upl_start, upl_size, &upl, &pl,
+						   UPL_FILE_IO | UPL_SET_LITE);
+	if ((error != KERN_SUCCESS) || !upl) {
+		printf("ZFS: mappedread failed to ubc_create_upl: %d\n", error);
+		return EIO;
+	}
+
+	if (ubc_upl_map(upl, &vaddr) != KERN_SUCCESS) {
+		printf("ZFS: mappedread failed to ubc_upl_map: %d\n", error);
+		(void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
+		return ENOMEM;
+	}
 
     for (upl_page = 0; len > 0; ++upl_page) {
         uint64_t bytes = MIN(PAGE_SIZE - off, len);
@@ -620,10 +632,8 @@ mappedread(vnode_t *vp, int nbytes, struct uio *uio)
     /*
      * Unmap the page list and free the UPL.
      */
-    if (pl) {
-        (void) ubc_upl_unmap(upl);
-        (void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
-    }
+	(void) ubc_upl_unmap(upl);
+	(void) ubc_upl_abort(upl, UPL_ABORT_FREE_ON_EMPTY);
 
     return (error);
 }