*/
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
* Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ * Copyright 2014 HybridCluster. All rights reserved.
+ * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
*/
/* Portions Copyright 2010 Robert Milkowski */
* dmu_spa.h.
*/
+#include <sys/zfs_context.h>
#include <sys/inttypes.h>
-#include <sys/types.h>
-#include <sys/param.h>
#include <sys/cred.h>
-#include <sys/time.h>
+#include <sys/fs/zfs.h>
#include <sys/uio.h>
#ifdef __cplusplus
struct dnode;
struct drr_begin;
struct drr_end;
-struct zbookmark;
+struct zbookmark_phys;
struct spa;
struct nvlist;
struct arc_buf;
((ot) & DMU_OT_METADATA) : \
dmu_ot[(int)(ot)].ot_metadata)
+/*
+ * These object types use bp_fill != 1 for their L0 bp's. Therefore they can't
+ * have their data embedded (i.e. use a BP_IS_EMBEDDED() bp), because bp_fill
+ * is repurposed for embedded BPs.
+ */
+#define DMU_OT_HAS_FILL(ot) \
+ ((ot) == DMU_OT_DNODE || (ot) == DMU_OT_OBJSET)
+
#define DMU_OT_BYTESWAP(ot) (((ot) & DMU_OT_NEWTYPE) ? \
((ot) & DMU_OT_BYTESWAP_MASK) : \
dmu_ot[(int)(ot)].ot_byteswap)
DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
} dmu_object_type_t;
-typedef enum dmu_objset_type {
- DMU_OST_NONE,
- DMU_OST_META,
- DMU_OST_ZFS,
- DMU_OST_ZVOL,
- DMU_OST_OTHER, /* For testing only! */
- DMU_OST_ANY, /* Be careful! */
- DMU_OST_NUMTYPES
-} dmu_objset_type_t;
+typedef enum txg_how {
+ TXG_WAIT = 1,
+ TXG_NOWAIT,
+ TXG_WAITED,
+} txg_how_t;
void byteswap_uint64_array(void *buf, size_t size);
void byteswap_uint32_array(void *buf, size_t size);
#define DS_FIND_SNAPSHOTS (1<<0)
#define DS_FIND_CHILDREN (1<<1)
+#define DS_FIND_SERIALIZE (1<<2)
/*
* The maximum number of bytes that can be accessed as part of one
* operation, including metadata.
*/
-#define DMU_MAX_ACCESS (10<<20) /* 10MB */
+#define DMU_MAX_ACCESS (64 * 1024 * 1024) /* 64MB */
#define DMU_MAX_DELETEBLKCNT (20480) /* ~5MB of indirect blocks */
#define DMU_USERUSED_OBJECT (-1ULL)
#define DMU_GROUPUSED_OBJECT (-2ULL)
-#define DMU_DEADLIST_OBJECT (-3ULL)
/*
* artificial blkids for bonus buffer and spill blocks
void dmu_objset_disown(objset_t *os, void *tag);
int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
-int dmu_objset_evict_dbufs(objset_t *os);
+void dmu_objset_evict_dbufs(objset_t *os);
int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_clone(const char *name, struct dsl_dataset *clone_origin,
- uint64_t flags);
-int dmu_objset_destroy(const char *name, boolean_t defer);
-int dmu_snapshots_destroy_nvl(struct nvlist *snaps, boolean_t defer, char *);
-int dmu_objset_snapshot(char *fsname, char *snapname, char *tag,
- struct nvlist *props, boolean_t recursive, boolean_t temporary, int fd);
-int dmu_objset_rename(const char *name, const char *newname,
- boolean_t recursive);
+int dmu_objset_clone(const char *name, const char *origin);
+int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
+ struct nvlist *errlist);
+int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
+int dmu_objset_snapshot_tmp(const char *, const char *, int);
int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
int flags);
void dmu_objset_byteswap(void *buf, size_t size);
+int dsl_dataset_rename_snapshot(const char *fsname,
+ const char *oldsnapname, const char *newsnapname, boolean_t recursive);
typedef struct dmu_buf {
uint64_t db_object; /* object that this buffer is part of */
void *db_data; /* data in buffer */
} dmu_buf_t;
-typedef void dmu_buf_evict_func_t(struct dmu_buf *db, void *user_ptr);
-
/*
* The names of zap entries in the DIRECTORY_OBJECT of the MOS.
*/
#define DMU_POOL_FEATURES_FOR_WRITE "features_for_write"
#define DMU_POOL_FEATURES_FOR_READ "features_for_read"
#define DMU_POOL_FEATURE_DESCRIPTIONS "feature_descriptions"
+#define DMU_POOL_FEATURE_ENABLED_TXG "feature_enabled_txg"
#define DMU_POOL_ROOT_DATASET "root_dataset"
#define DMU_POOL_SYNC_BPOBJ "sync_bplist"
#define DMU_POOL_ERRLOG_SCRUB "errlog_scrub"
int dmu_object_claim(objset_t *os, uint64_t object, dmu_object_type_t ot,
int blocksize, dmu_object_type_t bonus_type, int bonus_len, dmu_tx_t *tx);
int dmu_object_reclaim(objset_t *os, uint64_t object, dmu_object_type_t ot,
- int blocksize, dmu_object_type_t bonustype, int bonuslen);
+ int blocksize, dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *txp);
/*
* Free an object from this objset.
void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
dmu_tx_t *tx);
+void
+dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset,
+ void *data, uint8_t etype, uint8_t comp, int uncompressed_size,
+ int compressed_size, int byteorder, dmu_tx_t *tx);
+
/*
* Decide how to write a block: checksum, compression, number of copies, etc.
*/
* object must be held in an assigned transaction before calling
* dmu_buf_will_dirty. You may use dmu_buf_set_user() on the bonus
* buffer as well. You must release what you hold with dmu_buf_rele().
+ *
+ * Returns ENOENT, EIO, or 0.
*/
int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **);
int dmu_bonus_max(void);
*/
int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
void *tag, dmu_buf_t **, int flags);
+
+/*
+ * Add a reference to a dmu buffer that has already been held via
+ * dmu_buf_hold() in the current context.
+ */
void dmu_buf_add_ref(dmu_buf_t *db, void* tag);
+
+/*
+ * Attempt to add a reference to a dmu buffer that is in an unknown state,
+ * using a pointer that may have been invalidated by eviction processing.
+ * The request will succeed if the passed in dbuf still represents the
+ * same os/object/blkid, is ineligible for eviction, and has at least
+ * one hold by a user other than the syncer.
+ */
+boolean_t dmu_buf_try_add_ref(dmu_buf_t *, objset_t *os, uint64_t object,
+ uint64_t blkid, void *tag);
+
void dmu_buf_rele(dmu_buf_t *db, void *tag);
uint64_t dmu_buf_refcount(dmu_buf_t *db);
uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp);
void dmu_buf_rele_array(dmu_buf_t **, int numbufs, void *tag);
+typedef void dmu_buf_evict_func_t(void *user_ptr);
+
+/*
+ * A DMU buffer user object may be associated with a dbuf for the
+ * duration of its lifetime. This allows the user of a dbuf (client)
+ * to attach private data to a dbuf (e.g. in-core only data such as a
+ * dnode_children_t, zap_t, or zap_leaf_t) and be optionally notified
+ * when that dbuf has been evicted. Clients typically respond to the
+ * eviction notification by freeing their private data, thus ensuring
+ * the same lifetime for both dbuf and private data.
+ *
+ * The mapping from a dmu_buf_user_t to any client private data is the
+ * client's responsibility. All current consumers of the API with private
+ * data embed a dmu_buf_user_t as the first member of the structure for
+ * their private data. This allows conversions between the two types
+ * with a simple cast. Since the DMU buf user API never needs access
+ * to the private data, other strategies can be employed if necessary
+ * or convenient for the client (e.g. using container_of() to do the
+ * conversion for private data that cannot have the dmu_buf_user_t as
+ * its first member).
+ *
+ * Eviction callbacks are executed without the dbuf mutex held or any
+ * other type of mechanism to guarantee that the dbuf is still available.
+ * For this reason, users must assume the dbuf has already been freed
+ * and not reference the dbuf from the callback context.
+ *
+ * Users requesting "immediate eviction" are notified as soon as the dbuf
+ * is only referenced by dirty records (dirties == holds). Otherwise the
+ * notification occurs after eviction processing for the dbuf begins.
+ */
+typedef struct dmu_buf_user {
+ /*
+ * Asynchronous user eviction callback state.
+ */
+ taskq_ent_t dbu_tqent;
+
+ /* This instance's eviction function pointer. */
+ dmu_buf_evict_func_t *dbu_evict_func;
+#ifdef ZFS_DEBUG
+ /*
+ * Pointer to user's dbuf pointer. NULL for clients that do
+ * not associate a dbuf with their user data.
+ *
+ * The dbuf pointer is cleared upon eviction so as to catch
+ * use-after-evict bugs in clients.
+ */
+ dmu_buf_t **dbu_clear_on_evict_dbufp;
+#endif
+} dmu_buf_user_t;
+
+/*
+ * Initialize the given dmu_buf_user_t instance with the eviction function
+ * evict_func, to be called when the user is evicted.
+ *
+ * NOTE: This function should only be called once on a given dmu_buf_user_t.
+ * To allow enforcement of this, dbu must already be zeroed on entry.
+ */
+#ifdef __lint
+/* Very ugly, but it beats issuing suppression directives in many Makefiles. */
+extern void
+dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
+ dmu_buf_t **clear_on_evict_dbufp);
+#else /* __lint */
+static inline void
+dmu_buf_init_user(dmu_buf_user_t *dbu, dmu_buf_evict_func_t *evict_func,
+ dmu_buf_t **clear_on_evict_dbufp)
+{
+ ASSERT(dbu->dbu_evict_func == NULL);
+ ASSERT(evict_func != NULL);
+ dbu->dbu_evict_func = evict_func;
+ taskq_init_ent(&dbu->dbu_tqent);
+#ifdef ZFS_DEBUG
+ dbu->dbu_clear_on_evict_dbufp = clear_on_evict_dbufp;
+#endif
+}
+#endif /* __lint */
+
/*
- * Returns NULL on success, or the existing user ptr if it's already
- * been set.
- *
- * user_ptr is for use by the user and can be obtained via dmu_buf_get_user().
+ * Attach user data to a dbuf and mark it for normal (when the dbuf's
+ * data is cleared or its reference count goes to zero) eviction processing.
*
- * user_data_ptr_ptr should be NULL, or a pointer to a pointer which
- * will be set to db->db_data when you are allowed to access it. Note
- * that db->db_data (the pointer) can change when you do dmu_buf_read(),
- * dmu_buf_tryupgrade(), dmu_buf_will_dirty(), or dmu_buf_will_fill().
- * *user_data_ptr_ptr will be set to the new value when it changes.
+ * Returns NULL on success, or the existing user if another user currently
+ * owns the buffer.
+ */
+void *dmu_buf_set_user(dmu_buf_t *db, dmu_buf_user_t *user);
+
+/*
+ * Attach user data to a dbuf and mark it for immediate (its dirty and
+ * reference counts are equal) eviction processing.
*
- * If non-NULL, pageout func will be called when this buffer is being
- * excised from the cache, so that you can clean up the data structure
- * pointed to by user_ptr.
+ * Returns NULL on success, or the existing user if another user currently
+ * owns the buffer.
+ */
+void *dmu_buf_set_user_ie(dmu_buf_t *db, dmu_buf_user_t *user);
+
+/*
+ * Replace the current user of a dbuf.
*
- * dmu_evict_user() will call the pageout func for all buffers in a
- * objset with a given pageout func.
+ * If given the current user of a dbuf, replaces the dbuf's user with
+ * "new_user" and returns the user data pointer that was replaced.
+ * Otherwise returns the current, and unmodified, dbuf user pointer.
*/
-void *dmu_buf_set_user(dmu_buf_t *db, void *user_ptr, void *user_data_ptr_ptr,
- dmu_buf_evict_func_t *pageout_func);
+void *dmu_buf_replace_user(dmu_buf_t *db,
+ dmu_buf_user_t *old_user, dmu_buf_user_t *new_user);
+
/*
- * set_user_ie is the same as set_user, but request immediate eviction
- * when hold count goes to zero.
+ * Remove the specified user data for a DMU buffer.
+ *
+ * Returns the user that was removed on success, or the current user if
+ * another user currently owns the buffer.
*/
-void *dmu_buf_set_user_ie(dmu_buf_t *db, void *user_ptr,
- void *user_data_ptr_ptr, dmu_buf_evict_func_t *pageout_func);
-void *dmu_buf_update_user(dmu_buf_t *db_fake, void *old_user_ptr,
- void *user_ptr, void *user_data_ptr_ptr,
- dmu_buf_evict_func_t *pageout_func);
-void dmu_evict_user(objset_t *os, dmu_buf_evict_func_t *func);
+void *dmu_buf_remove_user(dmu_buf_t *db, dmu_buf_user_t *user);
/*
- * Returns the user_ptr set with dmu_buf_set_user(), or NULL if not set.
+ * Returns the user data (dmu_buf_user_t *) associated with this dbuf.
*/
void *dmu_buf_get_user(dmu_buf_t *db);
+/* Block until any in-progress dmu buf user evictions complete. */
+void dmu_buf_user_evict_wait(void);
+
+/*
+ * Returns the blkptr associated with this dbuf, or NULL if not set.
+ */
+struct blkptr *dmu_buf_get_blkptr(dmu_buf_t *db);
+
/*
* Indicate that you are going to modify the buffer's data (db_data).
*
void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
void dmu_tx_abort(dmu_tx_t *tx);
-int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
+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);
uint64_t size, dmu_tx_t *tx);
int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset,
uint64_t size);
-int dmu_free_object(objset_t *os, uint64_t object);
+int dmu_free_long_object(objset_t *os, uint64_t object);
/*
* Convenience functions.
dmu_tx_t *tx);
#ifdef _KERNEL
#include <linux/blkdev_compat.h>
-int dmu_read_req(objset_t *os, uint64_t object, struct request *req);
-int dmu_write_req(objset_t *os, uint64_t object, struct request *req,
+int dmu_read_bio(objset_t *os, uint64_t object, struct bio *bio);
+int dmu_write_bio(objset_t *os, uint64_t object, struct bio *bio,
dmu_tx_t *tx);
int dmu_read_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size);
+int dmu_read_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size);
int dmu_write_uio(objset_t *os, uint64_t object, struct uio *uio, uint64_t size,
dmu_tx_t *tx);
int dmu_write_uio_dbuf(dmu_buf_t *zdb, struct uio *uio, uint64_t size,
void xuio_stat_wbuf_nocopy(void);
extern int zfs_prefetch_disable;
+extern int zfs_max_recordsize;
/*
* Asynchronously try to read in the data.
uint8_t doi_indirection; /* 2 = dnode->indirect->data */
uint8_t doi_checksum;
uint8_t doi_compress;
- uint8_t doi_pad[5];
+ uint8_t doi_nblkptr;
+ uint8_t doi_pad[4];
uint64_t doi_physical_blocks_512; /* data + metadata, 512b blks */
uint64_t doi_max_offset;
uint64_t doi_fill_count; /* number of non-empty blocks */
* If doi is NULL, just indicates whether the object exists.
*/
int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi);
+void __dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
+/* Like dmu_object_info, but faster if you have a held dnode in hand. */
void dmu_object_info_from_dnode(struct dnode *dn, dmu_object_info_t *doi);
+/* Like dmu_object_info, but faster if you have a held dbuf in hand. */
void dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi);
+/*
+ * Like dmu_object_info_from_db, but faster still when you only care about
+ * the size. This is specifically optimized for zfs_getattr().
+ */
void dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize,
u_longlong_t *nblk512);
extern void dmu_objset_name(objset_t *os, char *buf);
extern dmu_objset_type_t dmu_objset_type(objset_t *os);
extern uint64_t dmu_objset_id(objset_t *os);
-extern uint64_t dmu_objset_syncprop(objset_t *os);
-extern uint64_t dmu_objset_logbias(objset_t *os);
+extern zfs_sync_type_t dmu_objset_syncprop(objset_t *os);
+extern zfs_logbias_op_t dmu_objset_logbias(objset_t *os);
extern int dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
uint64_t *id, uint64_t *offp, boolean_t *case_conflict);
extern int dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *val);
void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
dmu_traverse_cb_t cb, void *arg);
-int dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
- int outfd, struct vnode *vp, offset_t *off);
-int dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorign,
- uint64_t *sizep);
-
-typedef struct dmu_recv_cookie {
- /*
- * This structure is opaque!
- *
- * If logical and real are different, we are recving the stream
- * into the "real" temporary clone, and then switching it with
- * the "logical" target.
- */
- struct dsl_dataset *drc_logical_ds;
- struct dsl_dataset *drc_real_ds;
- struct drr_begin *drc_drrb;
- char *drc_tosnap;
- char *drc_top_ds;
- boolean_t drc_newfs;
- boolean_t drc_force;
- struct avl_tree *drc_guid_to_ds_map;
-} dmu_recv_cookie_t;
-
-int dmu_recv_begin(char *tofs, char *tosnap, char *topds, struct drr_begin *,
- boolean_t force, objset_t *origin, dmu_recv_cookie_t *);
-int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp,
- int cleanup_fd, uint64_t *action_handlep);
-int dmu_recv_end(dmu_recv_cookie_t *drc);
-
-int dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp,
- offset_t *off);
+int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
+ struct vnode *vp, offset_t *offp);
/* CRC64 table */
#define ZFS_CRC64_POLY 0xC96C5795D7870F42ULL /* ECMA-182, reflected form */
projects / mirror_zfs-debian.git / blobdiff