struct dt_device_operations {
/**
* Return device-wide statistics.
+ *
+ * Return device-wide stats including block size, total and
+ * free blocks, total and free objects, etc. See struct obd_statfs
+ * for the details.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ * \param[out] osfs stats information
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*dt_statfs)(const struct lu_env *env,
- struct dt_device *dev, struct obd_statfs *osfs);
+ struct dt_device *dev,
+ struct obd_statfs *osfs);
+
/**
- * Create transaction, described by \a param.
+ * Create transaction.
+ *
+ * Create in-memory structure representing the transaction for the
+ * caller. The structure returned will be used by the calling thread
+ * to specify the transaction the updates belong to. Once created
+ * successfully ->dt_trans_stop() must be called in any case (with
+ * ->dt_trans_start() and updates or not) so that the transaction
+ * handle and other resources can be released by the layers below.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ *
+ * \retval pointer to handle if creation succeeds
+ * \retval ERR_PTR(errno) if creation fails
*/
struct thandle *(*dt_trans_create)(const struct lu_env *env,
struct dt_device *dev);
+
/**
- * Start transaction, described by \a param.
+ * Start transaction.
+ *
+ * Start the transaction. The transaction described by \a th can be
+ * started only once. Another start is considered as an error.
+ * A thread is not supposed to start a transaction while another
+ * transaction isn't closed by the thread (though multiple handles
+ * can be created). The caller should start the transaction once
+ * all possible updates are declared (see the ->do_declare_* methods
+ * below) and all the needed resources are reserved.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*dt_trans_start)(const struct lu_env *env,
- struct dt_device *dev, struct thandle *th);
+ struct dt_device *dev,
+ struct thandle *th);
+
/**
- * Finish previously started transaction.
+ * Stop transaction.
+ *
+ * Once stopped the transaction described by \a th is complete (all
+ * the needed updates are applied) and further processing such as
+ * flushing to disk, sending to another target, etc, is handled by
+ * lower layers. The caller can't access this transaction by the
+ * handle anymore (except from the commit callbacks, see below).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
- int (*dt_trans_stop)(const struct lu_env *env, struct dt_device *dev,
+ int (*dt_trans_stop)(const struct lu_env *env,
+ struct dt_device *dev,
struct thandle *th);
+
/**
* Add commit callback to the transaction.
+ *
+ * Add a commit callback to the given transaction handle. The callback
+ * will be called when the associated transaction is stored. I.e. the
+ * transaction will survive an event like power off if the callback did
+ * run. The number of callbacks isn't limited, but you should note that
+ * some disk filesystems do handle the commit callbacks in the thread
+ * handling commit/flush of all the transactions, meaning that new
+ * transactions are blocked from commit and flush until all the
+ * callbacks are done. Also, note multiple callbacks can be running
+ * concurrently using multiple CPU cores. The callbacks will be running
+ * in a special environment which can not be used to pass data around.
+ *
+ * \param[in] th transaction handle
+ * \param[in] dcb commit callback description
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*dt_trans_cb_add)(struct thandle *th,
struct dt_txn_commit_cb *dcb);
+
/**
- * Return fid of root index object.
+ * Return FID of root index object.
+ *
+ * Return the FID of the root object in the filesystem. This object
+ * is usually provided as a bootstrap point by a disk filesystem.
+ * This is up to the implementation which FID to use, though
+ * [FID_SEQ_ROOT:1:0] is reserved for this purpose.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ * \param[out] fid FID of the root object
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*dt_root_get)(const struct lu_env *env,
- struct dt_device *dev, struct lu_fid *f);
+ struct dt_device *dev,
+ struct lu_fid *f);
+
/**
* Return device configuration data.
+ *
+ * Return device (disk fs, actually) specific configuration.
+ * The configuration isn't subject to change at runtime.
+ * See struct dt_device_param for the details.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ * \param[out] param configuration parameters
*/
void (*dt_conf_get)(const struct lu_env *env,
const struct dt_device *dev,
struct dt_device_param *param);
- /**
- * handling device state, mostly for tests
- */
- int (*dt_sync)(const struct lu_env *env, struct dt_device *dev);
- int (*dt_ro)(const struct lu_env *env, struct dt_device *dev);
- /**
- * Start a transaction commit asynchronously
- *
- * \param env environment
- * \param dev dt_device to start commit on
- *
- * \return 0 success, negative value if error
- */
+
+ /**
+ * Sync the device.
+ *
+ * Sync all the cached state (dirty buffers, pages, etc) to the
+ * persistent storage. The method returns control once the sync is
+ * complete. This operation may incur significant I/O to disk and
+ * should be reserved for cases where a global sync is strictly
+ * necessary.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dt_sync)(const struct lu_env *env,
+ struct dt_device *dev);
+
+ /**
+ * Make device read-only.
+ *
+ * Prevent new modifications to the device. This is a very specific
+ * state where all the changes are accepted successfully and the
+ * commit callbacks are called, but persistent state never changes.
+ * Used only in the tests to simulate power-off scenario.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dt_ro)(const struct lu_env *env,
+ struct dt_device *dev);
+
+ /**
+ * Start transaction commit asynchronously.
+ *
+
+ * Provide a hint to the underlying filesystem that it should start
+ * committing soon. The control returns immediately. It's up to the
+ * layer implementing the method how soon to start committing. Usually
+ * this should be throttled to some extent, otherwise the number of
+ * aggregated transaction goes too high causing performance drop.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dev dt device
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*dt_commit_async)(const struct lu_env *env,
struct dt_device *dev);
- /**
- * Initialize capability context.
- */
- int (*dt_init_capa_ctxt)(const struct lu_env *env,
- struct dt_device *dev,
- int mode, unsigned long timeout,
- __u32 alg, struct lustre_capa_key *keys);
+
+ /**
+ * Not used, subject to removal.
+ */
+ int (*dt_init_capa_ctxt)(const struct lu_env *env,
+ struct dt_device *dev,
+ int mode,
+ unsigned long timeout,
+ __u32 alg,
+ struct lustre_capa_key *keys);
};
struct dt_index_features {
union ldlm_policy_data;
/**
- * Per-dt-object operations.
+ * A dt_object provides common operations to create and destroy
+ * objects and to manage regular and extended attributes.
*/
struct dt_object_operations {
- void (*do_read_lock)(const struct lu_env *env,
- struct dt_object *dt, unsigned role);
- void (*do_write_lock)(const struct lu_env *env,
- struct dt_object *dt, unsigned role);
+ /**
+ * Get read lock on object.
+ *
+ * Read lock is compatible with other read locks, so it's shared.
+ * Read lock is not compatible with write lock which is exclusive.
+ * The lock is blocking and can't be used from an interrupt context.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object to lock for reading
+ * \param[in] role a hint to debug locks (see kernel's mutexes)
+ */
+ void (*do_read_lock)(const struct lu_env *env,
+ struct dt_object *dt,
+ unsigned role);
+
+ /*
+ * Get write lock on object.
+ *
+ * Write lock is exclusive and cannot be shared. The lock is blocking
+ * and can't be used from an interrupt context.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object to lock for writing
+ * \param[in] role a hint to debug locks (see kernel's mutexes)
+ *
+ */
+ void (*do_write_lock)(const struct lu_env *env,
+ struct dt_object *dt,
+ unsigned role);
+
+ /**
+ * Release read lock.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ */
void (*do_read_unlock)(const struct lu_env *env,
struct dt_object *dt);
+
+ /**
+ * Release write lock.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ */
void (*do_write_unlock)(const struct lu_env *env,
struct dt_object *dt);
+
+ /**
+ * Check whether write lock is held.
+ *
+ * The caller can learn whether write lock is held on the object
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ *
+ * \retval 0 no write lock
+ * \retval 1 write lock is held
+ */
int (*do_write_locked)(const struct lu_env *env,
struct dt_object *dt);
- /**
- * Note: following ->do_{x,}attr_{set,get}() operations are very
- * similar to ->moo_{x,}attr_{set,get}() operations in struct
- * md_object_operations (see md_object.h). These operations are not in
- * lu_object_operations, because ->do_{x,}attr_set() versions take
- * transaction handle as an argument (this transaction is started by
- * caller). We might factor ->do_{x,}attr_get() into
- * lu_object_operations, but that would break existing symmetry.
- */
+ /**
+ * Declare intention to request reqular attributes.
+ *
+ * Notity the underlying filesystem that the caller may request regular
+ * attributes with ->do_attr_get() soon. This allows OSD to implement
+ * prefetching logic in an object-oriented manner. The implementation
+ * can be noop. This method should avoid expensive delays such as
+ * waiting on disk I/O, otherwise the goal of enabling a performance
+ * optimization would be defeated.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_declare_attr_get)(const struct lu_env *env,
struct dt_object *dt,
struct lustre_capa *capa);
- /**
- * Return standard attributes.
- *
- * precondition: lu_object_exists(&dt->do_lu);
- */
- int (*do_attr_get)(const struct lu_env *env,
- struct dt_object *dt, struct lu_attr *attr,
- struct lustre_capa *capa);
- /**
- * Set standard attributes.
- *
- * precondition: dt_object_exists(dt);
+
+ /**
+ * Return regular attributes.
+ *
+ * The object must exist. Currently all the attributes should be
+ * returned, but in the future this can be improved so that only
+ * a selected set is returned. This can improve performance as in
+ * some cases attributes are stored in different places and
+ * getting them all can be an iterative and expensive process.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] attr attributes to fill
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_attr_get)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_attr *attr,
+ struct lustre_capa *capa);
+
+ /**
+ * Declare intention to change regular object's attributes.
+ *
+ * Notify the underlying filesystem that the regular attributes may
+ * change in this transaction. This enables the layer below to prepare
+ * resources (e.g. journal credits in ext4). This method should be
+ * called between creating the transaction and starting it. Note that
+ * the la_valid field of \a attr specifies which attributes will change.
+ * The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] attr attributes to change specified in attr.la_valid
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*do_declare_attr_set)(const struct lu_env *env,
struct dt_object *dt,
const struct lu_attr *attr,
- struct thandle *handle);
+ struct thandle *th);
+
+ /**
+ * Change regular attributes.
+ *
+ * Change regular attributes in the given transaction. Note only
+ * attributes flagged by attr.la_valid change. The object must
+ * exist. If the layer implementing this method is responsible for
+ * quota, then the method should maintain object accounting for the
+ * given credentials when la_uid/la_gid changes.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] attr new attributes to apply
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_attr_set)(const struct lu_env *env,
struct dt_object *dt,
const struct lu_attr *attr,
- struct thandle *handle,
+ struct thandle *th,
struct lustre_capa *capa);
+ /**
+ * Declare intention to request extented attribute.
+ *
+ * Notify the underlying filesystem that the caller may request extended
+ * attribute with ->do_xattr_get() soon. This allows OSD to implement
+ * prefetching logic in an object-oriented manner. The implementation
+ * can be noop. This method should avoid expensive delays such as
+ * waiting on disk I/O, otherwise the goal of enabling a performance
+ * optimization would be defeated.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] buf unused, may be removed in the future
+ * \param[in] name name of the extended attribute
+ * \param[in] capa unused, may be removed in the future
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_declare_xattr_get)(const struct lu_env *env,
struct dt_object *dt,
struct lu_buf *buf,
const char *name,
struct lustre_capa *capa);
- /**
- * Return a value of an extended attribute.
- *
- * precondition: dt_object_exists(dt);
- */
- int (*do_xattr_get)(const struct lu_env *env, struct dt_object *dt,
- struct lu_buf *buf, const char *name,
- struct lustre_capa *capa);
- /**
- * Set value of an extended attribute.
- *
- * \a fl - flags from enum lu_xattr_flags
- *
- * precondition: dt_object_exists(dt);
- */
- int (*do_declare_xattr_set)(const struct lu_env *env,
- struct dt_object *dt,
- const struct lu_buf *buf,
- const char *name, int fl,
- struct thandle *handle);
- int (*do_xattr_set)(const struct lu_env *env,
- struct dt_object *dt, const struct lu_buf *buf,
- const char *name, int fl, struct thandle *handle,
- struct lustre_capa *capa);
- /**
- * Delete existing extended attribute.
- *
- * precondition: dt_object_exists(dt);
- */
- int (*do_declare_xattr_del)(const struct lu_env *env,
- struct dt_object *dt,
- const char *name, struct thandle *handle);
- int (*do_xattr_del)(const struct lu_env *env,
- struct dt_object *dt,
- const char *name, struct thandle *handle,
- struct lustre_capa *capa);
- /**
- * Place list of existing extended attributes into \a buf (which has
- * length len).
- *
- * precondition: dt_object_exists(dt);
+ /**
+ * Return a value of an extended attribute.
+ *
+ * The object must exist. If the buffer is NULL, then the method
+ * must return the size of the value.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] buf buffer in which to store the value
+ * \param[in] name name of the extended attribute
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval -ERANGE if \a buf is too small
+ * \retval negative negated errno on error
+ * \retval positive value's size if \a buf is NULL or has zero size
+ */
+ int (*do_xattr_get)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_buf *buf,
+ const char *name,
+ struct lustre_capa *capa);
+
+ /**
+ * Declare intention to change an extended attribute.
+ *
+ * Notify the underlying filesystem that the extended attribute may
+ * change in this transaction. This enables the layer below to prepare
+ * resources (e.g. journal credits in ext4). This method should be
+ * called between creating the transaction and starting it. The object
+ * need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] buf buffer storing new value of the attribute
+ * \param[in] name name of the attribute
+ * \param[in] fl LU_XATTR_CREATE - fail if EA exists
+ * LU_XATTR_REPLACE - fail if EA doesn't exist
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
- int (*do_xattr_list)(const struct lu_env *env,
- struct dt_object *dt, struct lu_buf *buf,
- struct lustre_capa *capa);
- /**
- * Init allocation hint using parent object and child mode.
- * (1) The \a parent might be NULL if this is a partial creation for
- * remote object.
- * (2) The type of child is in \a child_mode.
- * (3) The result hint is stored in \a ah;
+ int (*do_declare_xattr_set)(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name,
+ int fl,
+ struct thandle *th);
+
+ /**
+ * Set an extended attribute.
+ *
+ * Change or replace the specified extended attribute (EA).
+ * The flags passed in \a fl dictate whether the EA is to be
+ * created or replaced, as follows.
+ * LU_XATTR_CREATE - fail if EA exists
+ * LU_XATTR_REPLACE - fail if EA doesn't exist
+ * The object must exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] buf buffer storing new value of the attribute
+ * \param[in] name name of the attribute
+ * \param[in] fl flags indicating EA creation or replacement
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_xattr_set)(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ const char *name,
+ int fl,
+ struct thandle *th,
+ struct lustre_capa *capa);
+
+ /**
+ * Declare intention to delete an extended attribute.
+ *
+ * Notify the underlying filesystem that the extended attribute may
+ * be deleted in this transaction. This enables the layer below to
+ * prepare resources (e.g. journal credits in ext4). This method
+ * should be called between creating the transaction and starting it.
+ * The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] name name of the attribute
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_declare_xattr_del)(const struct lu_env *env,
+ struct dt_object *dt,
+ const char *name,
+ struct thandle *th);
+
+ /**
+ * Delete an extended attribute.
+ *
+ * This method deletes the specified extended attribute. The object
+ * must exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] name name of the attribute
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_xattr_del)(const struct lu_env *env,
+ struct dt_object *dt,
+ const char *name,
+ struct thandle *th,
+ struct lustre_capa *capa);
+
+ /**
+ * Return a list of the extended attributes.
+ *
+ * Fills the passed buffer with a list of the extended attributes
+ * found in the object. The names are separated with '\0'.
+ * The object must exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] buf buffer to put the list in
+ * \param[in] capa unused
+ *
+ * \retval positive bytes used/required in the buffer
+ * \retval negative negated errno on error
*/
+ int (*do_xattr_list)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_buf *buf,
+ struct lustre_capa *capa);
+
+ /**
+ * Prepare allocation hint for a new object.
+ *
+ * This method is used by the caller to inform OSD of the parent-child
+ * relationship between two objects and enable efficient object
+ * allocation. Filled allocation hint will be passed to ->do_create()
+ * later.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[out] ah allocation hint
+ * \param[in] parent parent object (can be NULL)
+ * \param[in] child child object
+ * \param[in] _mode type of the child object
+ */
void (*do_ah_init)(const struct lu_env *env,
struct dt_allocation_hint *ah,
struct dt_object *parent,
struct dt_object *child,
- umode_t child_mode);
- /**
- * Create new object on this device.
- *
- * precondition: !dt_object_exists(dt);
- * postcondition: ergo(result == 0, dt_object_exists(dt));
- */
+ umode_t mode);
+
+ /**
+ * Declare intention to create a new object.
+ *
+ * Notify the underlying filesystem that the object may be created
+ * in this transaction. This enables the layer below to prepare
+ * resources (e.g. journal credits in ext4). This method should be
+ * called between creating the transaction and starting it.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should reserve an object for the given credentials
+ * and return an error if quota is over. If object creation later
+ * fails for some reason, then the reservation should be released
+ * properly (usually in ->dt_trans_stop()).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] attr attributes of the new object
+ * \param[in] hint allocation hint
+ * \param[in] dof object format
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_declare_create)(const struct lu_env *env,
struct dt_object *dt,
struct lu_attr *attr,
struct dt_allocation_hint *hint,
struct dt_object_format *dof,
struct thandle *th);
- int (*do_create)(const struct lu_env *env, struct dt_object *dt,
+
+ /**
+ * Create new object.
+ *
+ * The method creates the object passed with the specified attributes
+ * and object format. Object allocation procedure can use information
+ * stored in the allocation hint. Different object formats are supported
+ * (see enum dt_format_type and struct dt_object_format) depending on
+ * the device. If creation succeeds, then LOHA_EXISTS flag must be set
+ * in the LU-object header attributes.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should maintain object accounting for the given
+ * credentials.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] attr attributes of the new object
+ * \param[in] hint allocation hint
+ * \param[in] dof object format
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_create)(const struct lu_env *env,
+ struct dt_object *dt,
struct lu_attr *attr,
struct dt_allocation_hint *hint,
struct dt_object_format *dof,
struct thandle *th);
- /**
- Destroy object on this device
- * precondition: !dt_object_exists(dt);
- * postcondition: ergo(result == 0, dt_object_exists(dt));
- */
+ /**
+ * Declare intention to destroy an object.
+ *
+ * Notify the underlying filesystem that the object may be destroyed
+ * in this transaction. This enables the layer below to prepare
+ * resources (e.g. journal credits in ext4). This method should be
+ * called between creating the transaction and starting it. The object
+ * need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_declare_destroy)(const struct lu_env *env,
struct dt_object *dt,
struct thandle *th);
- int (*do_destroy)(const struct lu_env *env, struct dt_object *dt,
- struct thandle *th);
- /**
+ /**
+ * Destroy an object.
+ *
+ * This method destroys the object and all the resources associated
+ * with the object (data, key/value pairs, extended attributes, etc).
+ * The object must exist. If destroy is successful, then flag
+ * LU_OBJECT_HEARD_BANSHEE should be set to forbid access to this
+ * instance of in-core object. Any subsequent access to the same FID
+ * should get another instance with no LOHA_EXIST flag set.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should maintain object accounting for the given
+ * credentials.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_destroy)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct thandle *th);
+
+ /**
+ * Try object as an index.
+ *
* Announce that this object is going to be used as an index. This
- * operation check that object supports indexing operations and
+ * operation checks that object supports indexing operations and
* installs appropriate dt_index_operations vector on success.
- *
* Also probes for features. Operation is successful if all required
- * features are supported.
- */
+ * features are supported. It's not possible to access the object
+ * with index methods before ->do_index_try() returns success.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] feat index features
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_index_try)(const struct lu_env *env,
struct dt_object *dt,
const struct dt_index_features *feat);
- /**
- * Add nlink of the object
- * precondition: dt_object_exists(dt);
- */
- int (*do_declare_ref_add)(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th);
+
+ /**
+ * Declare intention to increment nlink count.
+ *
+ * Notify the underlying filesystem that the nlink regular attribute
+ * be changed in this transaction. This enables the layer below to
+ * prepare resources (e.g. journal credits in ext4). This method
+ * should be called between creating the transaction and starting it.
+ * The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_declare_ref_add)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct thandle *th);
+
+ /**
+ * Increment nlink.
+ *
+ * Increment nlink (from the regular attributes set) in the given
+ * transaction. Note the absolute limit for nlink should be learnt
+ * from struct dt_device_param::ddp_max_nlink. The object must exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*do_ref_add)(const struct lu_env *env,
struct dt_object *dt, struct thandle *th);
- /**
- * Del nlink of the object
- * precondition: dt_object_exists(dt);
- */
- int (*do_declare_ref_del)(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th);
- int (*do_ref_del)(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th);
+ /**
+ * Declare intention to decrement nlink count.
+ *
+ * Notify the underlying filesystem that the nlink regular attribute
+ * be changed in this transaction. This enables the layer below to
+ * prepare resources (e.g. journal credits in ext4). This method
+ * should be called between creating the transaction and starting it.
+ * The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_declare_ref_del)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct thandle *th);
+
+ /**
+ * Decrement nlink.
+ *
+ * Decrement nlink (from the regular attributes set) in the given
+ * transaction. The object must exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_ref_del)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct thandle *th);
+
+ /**
+ * Not used, subject to removal.
+ */
struct obd_capa *(*do_capa_get)(const struct lu_env *env,
struct dt_object *dt,
struct lustre_capa *old,
__u64 opc);
- int (*do_object_sync)(const struct lu_env *env, struct dt_object *obj,
- __u64 start, __u64 end);
+
+ /**
+ * Sync obect.
+ *
+ * The method is called to sync specified range of the object to a
+ * persistent storage. The control is returned once the operation is
+ * complete. The difference from ->do_sync() is that the object can
+ * be in-sync with the persistent storage (nothing to flush), then
+ * the method returns quickly with no I/O overhead. So, this method
+ * should be preferred over ->do_sync() where possible. Also note that
+ * if the object isn't clean, then some disk filesystems will call
+ * ->do_sync() to maintain overall consistency, in which case it's
+ * still very expensive.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] start start of the range to sync
+ * \param[in] end end of the range to sync
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_object_sync)(const struct lu_env *env,
+ struct dt_object *obj,
+ __u64 start,
+ __u64 end);
+
/**
- * Get object info of next level. Currently, only get inode from osd.
- * This is only used by quota b=16542
- * precondition: dt_object_exists(dt);
+ * Not used, subject to removal.
*/
- int (*do_data_get)(const struct lu_env *env, struct dt_object *dt,
+ int (*do_data_get)(const struct lu_env *env,
+ struct dt_object *dt,
void **data);
/**
* Lock object.
+ *
+ * Lock object(s) using Distributed Lock Manager (LDLM).
+ *
+ * Get LDLM locks for the object. Currently used to lock "remote"
+ * objects in DNE configuration - a service running on MDTx needs
+ * to lock an object on MDTy.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] lh lock handle, sometimes used, sometimes not
+ * \param[in] einfo ldlm callbacks, locking type and mode
+ * \param[out] einfo private data to be passed to unlock later
+ * \param[in] policy inodebits data
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
int (*do_object_lock)(const struct lu_env *env, struct dt_object *dt,
struct lustre_handle *lh,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy);
- int (*do_object_unlock)(const struct lu_env *env, struct dt_object *dt,
+ /**
+ * Unlock object.
+ *
+ * Release LDLM lock(s) granted with ->do_object_lock().
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] einfo lock handles, from ->do_object_lock()
+ * \param[in] policy inodebits data
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*do_object_unlock)(const struct lu_env *env,
+ struct dt_object *dt,
struct ldlm_enqueue_info *einfo,
union ldlm_policy_data *policy);
};
/**
- * Per-dt-object operations on "file body".
+ * Per-dt-object operations on "file body" - unstructure raw data.
*/
struct dt_body_operations {
- /**
- * precondition: dt_object_exists(dt);
- */
- ssize_t (*dbo_read)(const struct lu_env *env, struct dt_object *dt,
- struct lu_buf *buf, loff_t *pos,
- struct lustre_capa *capa);
/**
- * precondition: dt_object_exists(dt);
+ * Read data.
+ *
+ * Read unstructured data from an existing regular object.
+ * Only data before attr.la_size is returned.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] buf buffer (including size) to copy data in
+ * \param[in] pos position in the object to start
+ * \param[out] pos original value of \a pos + bytes returned
+ * \param[in] capa unused
+ *
+ * \retval positive bytes read on success
+ * \retval negative negated errno on error
+ */
+ ssize_t (*dbo_read)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct lu_buf *buf,
+ loff_t *pos,
+ struct lustre_capa *capa);
+
+ /**
+ * Declare intention to write data to object.
+ *
+ * Notify the underlying filesystem that data may be written in
+ * this transaction. This enables the layer below to prepare resources
+ * (e.g. journal credits in ext4). This method should be called
+ * between creating the transaction and starting it. The object need
+ * not exist. If the layer implementing this method is responsible for
+ * quota, then the method should reserve space for the given credentials
+ * and return an error if quota is over. If the write later fails
+ * for some reason, then the reserve should be released properly
+ * (usually in ->dt_trans_stop()).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] buf buffer (including size) to copy data from
+ * \param[in] pos position in the object to start
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
*/
ssize_t (*dbo_declare_write)(const struct lu_env *env,
struct dt_object *dt,
- const struct lu_buf *buf, loff_t pos,
- struct thandle *handle);
- ssize_t (*dbo_write)(const struct lu_env *env, struct dt_object *dt,
- const struct lu_buf *buf, loff_t *pos,
- struct thandle *handle, struct lustre_capa *capa,
- int ignore_quota);
- /*
- * methods for zero-copy IO
- */
+ const struct lu_buf *buf,
+ loff_t pos,
+ struct thandle *th);
- /*
- * precondition: dt_object_exists(dt);
- * returns:
- * < 0 - error code
- * = 0 - illegal
- * > 0 - number of local buffers prepared
- */
- int (*dbo_bufs_get)(const struct lu_env *env, struct dt_object *dt,
- loff_t pos, ssize_t len, struct niobuf_local *lb,
- int rw, struct lustre_capa *capa);
- /*
- * precondition: dt_object_exists(dt);
- */
- int (*dbo_bufs_put)(const struct lu_env *env, struct dt_object *dt,
- struct niobuf_local *lb, int nr);
- /*
- * precondition: dt_object_exists(dt);
- */
- int (*dbo_write_prep)(const struct lu_env *env, struct dt_object *dt,
- struct niobuf_local *lb, int nr);
- /*
- * precondition: dt_object_exists(dt);
- */
- int (*dbo_declare_write_commit)(const struct lu_env *env,
- struct dt_object *dt,
- struct niobuf_local *,
- int, struct thandle *);
- /*
- * precondition: dt_object_exists(dt);
- */
- int (*dbo_write_commit)(const struct lu_env *env, struct dt_object *dt,
- struct niobuf_local *, int, struct thandle *);
- /*
- * precondition: dt_object_exists(dt);
- */
- int (*dbo_read_prep)(const struct lu_env *env, struct dt_object *dt,
- struct niobuf_local *lnb, int nr);
- int (*dbo_fiemap_get)(const struct lu_env *env, struct dt_object *dt,
- struct ll_user_fiemap *fm);
- /**
- * Punch object's content
- * precondition: regular object, not index
- */
- int (*dbo_declare_punch)(const struct lu_env *, struct dt_object *,
- __u64, __u64, struct thandle *th);
- int (*dbo_punch)(const struct lu_env *env, struct dt_object *dt,
- __u64 start, __u64 end, struct thandle *th,
- struct lustre_capa *capa);
+ /**
+ * Write unstructured data to regular existing object.
+ *
+ * The method allocates space and puts data in. Also, the method should
+ * maintain attr.la_size properly. Partial writes are possible.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should maintain space accounting for the given
+ * credentials.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] buf buffer (including size) to copy data from
+ * \param[in] pos position in the object to start
+ * \param[out] pos \a pos + bytes written
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ * \param[in] ignore unused (was used to request quota ignorance)
+ *
+ * \retval positive bytes written on success
+ * \retval negative negated errno on error
+ */
+ ssize_t (*dbo_write)(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct lu_buf *buf,
+ loff_t *pos,
+ struct thandle *th,
+ struct lustre_capa *capa,
+ int ignore);
+
+ /**
+ * Return buffers for data.
+ *
+ * This method is used to access data with no copying. It's so-called
+ * zero-copy I/O. The method returns the descriptors for the internal
+ * buffers where data are managed by the disk filesystem. For example,
+ * pagecache in case of ext4 or ARC with ZFS. Then other components
+ * (e.g. networking) can transfer data from or to the buffers with no
+ * additional copying.
+ *
+ * The method should fill an array of struct niobuf_local, where
+ * each element describes a full or partial page for data at specific
+ * offset. The caller should use page/lnb_page_offset/len to find data
+ * at object's offset lnb_file_offset.
+ *
+ * The memory referenced by the descriptors can't change its purpose
+ * until the complimentary ->dbo_bufs_put() is called. The caller should
+ * specify if the buffers are used to read or modify data so that OSD
+ * can decide how to initialize the buffers: bring all the data for
+ * reads or just bring partial buffers for write. Note: the method does
+ * not check whether output array is large enough.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] pos position in the object to start
+ * \param[in] len size of region in bytes
+ * \param[out] lb array of descriptors to fill
+ * \param[in] rw 0 if used to read, 1 if used for write
+ * \param[in] capa unused
+ *
+ * \retval positive number of descriptors on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_bufs_get)(const struct lu_env *env,
+ struct dt_object *dt,
+ loff_t pos,
+ ssize_t len,
+ struct niobuf_local *lb,
+ int rw,
+ struct lustre_capa *capa);
+
+ /**
+ * Release reference granted by ->dbo_bufs_get().
+ *
+ * Release the reference granted by the previous ->dbo_bufs_get().
+ * Note the references are counted.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] lb array of descriptors to fill
+ * \param[in] nr size of the array
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_bufs_put)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct niobuf_local *lb,
+ int nr);
+
+ /**
+ * Prepare buffers for reading.
+ *
+ * The method is called on the given buffers to fill them with data
+ * if that wasn't done in ->dbo_bufs_get(). The idea is that the
+ * caller should be able to get few buffers for discontiguous regions
+ * using few calls to ->dbo_bufs_get() and then request them all for
+ * the preparation with a single call, so that OSD can fire many I/Os
+ * to run concurrently. It's up to the specific OSD whether to implement
+ * this logic in ->dbo_read_prep() or just use ->dbo_bufs_get() to
+ * prepare data for every requested region individually.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] lnb array of buffer descriptors
+ * \param[in] nr size of the array
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_read_prep)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct niobuf_local *lnb,
+ int nr);
+
+ /**
+ * Prepare buffers for write.
+ *
+ * This method is called on the given buffers to ensure the partial
+ * buffers contain correct data. The underlying idea is the same as
+ * in ->db_read_prep().
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] lb array of buffer descriptors
+ * \param[in] nr size of the array
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_write_prep)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct niobuf_local *lb,
+ int nr);
+
+ /**
+ * Declare intention to write data stored in the buffers.
+ *
+ * Notify the underlying filesystem that data may be written in
+ * this transaction. This enables the layer below to prepare resources
+ * (e.g. journal credits in ext4). This method should be called
+ * between creating the transaction and starting it.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should be reserving a space for the given
+ * credentials and return an error if quota is exceeded. If the write
+ * later fails for some reason, then the reserve should be released
+ * properly (usually in ->dt_trans_stop()).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] lb array of descriptors
+ * \param[in] nr size of the array
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_declare_write_commit)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct niobuf_local *lb,
+ int nr,
+ struct thandle *th);
+
+ /**
+ * Write to existing object.
+ *
+ * This method is used to write data to a persistent storage using
+ * the buffers returned by ->dbo_bufs_get(). The caller puts new
+ * data into the buffers using own mechanisms (e.g. direct transfer
+ * from a NIC). The method should maintain attr.la_size. Also,
+ * attr.la_blocks should be maintained but this can be done in lazy
+ * manner, when actual allocation happens.
+ *
+ * If the layer implementing this method is responsible for quota,
+ * then the method should maintain space accounting for the given
+ * credentials.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] lb array of descriptors for the buffers
+ * \param[in] nr size of the array
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_write_commit)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct niobuf_local *lb,
+ int nr,
+ struct thandle *th);
+
+ /**
+ * Return logical to physical block mapping for a given extent
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] fm describe the region to map and the output buffer
+ * see the details in include/linux/fiemap.h
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_fiemap_get)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct ll_user_fiemap *fm);
+
+ /**
+ * Declare intention to deallocate space from an object.
+ *
+ * Notify the underlying filesystem that space may be deallocated in
+ * this transactions. This enables the layer below to prepare resources
+ * (e.g. journal credits in ext4). This method should be called between
+ * creating the transaction and starting it. The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] start the start of the region to deallocate
+ * \param[in] end the end of the region to deallocate
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_declare_punch)(const struct lu_env *env,
+ struct dt_object *dt,
+ __u64 start,
+ __u64 end,
+ struct thandle *th);
+
+ /**
+ * Deallocate specified region in an object.
+ *
+ * This method is used to deallocate (release) space possibly consumed
+ * by the given region of the object. If the layer implementing this
+ * method is responsible for quota, then the method should maintain
+ * space accounting for the given credentials.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] start the start of the region to deallocate
+ * \param[in] end the end of the region to deallocate
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dbo_punch)(const struct lu_env *env,
+ struct dt_object *dt,
+ __u64 start,
+ __u64 end,
+ struct thandle *th,
+ struct lustre_capa *capa);
};
/**
struct dt_it;
/**
- * Per-dt-object operations on object as index.
+ * Per-dt-object operations on object as index. Index is a set of key/value
+ * pairs abstracted from an on-disk representation. An index supports the
+ * number of operations including lookup by key, insert and delete. Also,
+ * an index can be iterated to find the pairs one by one, from a beginning
+ * or specified point.
*/
struct dt_index_operations {
- /**
- * precondition: dt_object_exists(dt);
- */
- int (*dio_lookup)(const struct lu_env *env, struct dt_object *dt,
- struct dt_rec *rec, const struct dt_key *key,
- struct lustre_capa *capa);
- /**
- * precondition: dt_object_exists(dt);
- */
+ /**
+ * Lookup in an index by key.
+ *
+ * The method returns a value for the given key. Key/value format
+ * and size should have been negotiated with ->do_index_try() before.
+ * Thus it's the caller's responsibility to provide the method with
+ * proper key and big enough buffer. No external locking is required,
+ * all the internal consistency should be implemented by the method
+ * or lower layers. The object should should have been created with
+ * type DFT_INDEX or DFT_DIR.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[out] rec buffer where value will be stored
+ * \param[in] key key
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval -ENOENT if key isn't found
+ * \retval negative negated errno on error
+ */
+ int (*dio_lookup)(const struct lu_env *env,
+ struct dt_object *dt,
+ struct dt_rec *rec,
+ const struct dt_key *key,
+ struct lustre_capa *capa);
+
+ /**
+ * Declare intention to insert a key/value into an index.
+ *
+ * Notify the underlying filesystem that new key/value may be inserted
+ * in this transaction. This enables the layer below to prepare
+ * resources (e.g. journal credits in ext4). This method should be
+ * called between creating the transaction and starting it. key/value
+ * format and size is subject to ->do_index_try().
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] rec buffer storing value
+ * \param[in] key key
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*dio_declare_insert)(const struct lu_env *env,
- struct dt_object *dt,
- const struct dt_rec *rec,
- const struct dt_key *key,
- struct thandle *handle);
- int (*dio_insert)(const struct lu_env *env, struct dt_object *dt,
- const struct dt_rec *rec, const struct dt_key *key,
- struct thandle *handle, struct lustre_capa *capa,
- int ignore_quota);
- /**
- * precondition: dt_object_exists(dt);
- */
+ struct dt_object *dt,
+ const struct dt_rec *rec,
+ const struct dt_key *key,
+ struct thandle *th);
+
+ /**
+ * Insert a new key/value pair into an index.
+ *
+ * The method inserts specified key/value pair into the given index
+ * object. The internal consistency is maintained by the method or
+ * the functionality below. The format and size of key/value should
+ * have been negotiated before using ->do_index_try(), no additional
+ * information can be specified to the method. The keys are unique
+ * in a given index.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] rec buffer storing value
+ * \param[in] key key
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ * \param[in] ignore unused (was used to request quota ignorance)
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dio_insert)(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_rec *rec,
+ const struct dt_key *key,
+ struct thandle *th,
+ struct lustre_capa *capa,
+ int ignore);
+
+ /**
+ * Declare intention to delete a key/value from an index.
+ *
+ * Notify the underlying filesystem that key/value may be deleted in
+ * this transaction. This enables the layer below to prepare resources
+ * (e.g. journal credits in ext4). This method should be called
+ * between creating the transaction and starting it. Key/value format
+ * and size is subject to ->do_index_try(). The object need not exist.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] key key
+ * \param[in] th transaction handle
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*dio_declare_delete)(const struct lu_env *env,
- struct dt_object *dt,
- const struct dt_key *key,
- struct thandle *handle);
- int (*dio_delete)(const struct lu_env *env, struct dt_object *dt,
- const struct dt_key *key, struct thandle *handle,
- struct lustre_capa *capa);
+ struct dt_object *dt,
+ const struct dt_key *key,
+ struct thandle *th);
+
+ /**
+ * Delete key/value pair from an index.
+ *
+ * The method deletes specified key and corresponding value from the
+ * given index object. The internal consistency is maintained by the
+ * method or the functionality below. The format and size of the key
+ * should have been negotiated before using ->do_index_try(), no
+ * additional information can be specified to the method.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] key key
+ * \param[in] th transaction handle
+ * \param[in] capa unused
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
+ int (*dio_delete)(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_key *key,
+ struct thandle *th,
+ struct lustre_capa *capa);
+
/**
- * Iterator interface
+ * Iterator interface.
+ *
+ * Methods to iterate over an existing index, list the keys stored and
+ * associated values, get key/value size, etc.
*/
struct dt_it_ops {
- /**
- * Allocate and initialize new iterator.
- *
- * precondition: dt_object_exists(dt);
+ /**
+ * Allocate and initialize new iterator.
+ *
+ * The iterator is a handler to be used in the subsequent
+ * methods to access index's content. Note the position is
+ * not defined at this point and should be initialized with
+ * ->get() or ->load() method.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] dt object
+ * \param[in] attr ask the iterator to return part of
+ the records, see LUDA_* for details
+ * \param[in] capa unused
+ *
+ * \retval pointer iterator pointer on success
+ * \retval ERR_PTR(errno) on error
*/
struct dt_it *(*init)(const struct lu_env *env,
struct dt_object *dt,
__u32 attr,
struct lustre_capa *capa);
+
+ /**
+ * Release iterator.
+ *
+ * Release the specified iterator and all the resources
+ * associated (e.g. the object, index cache, etc).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator to release
+ */
void (*fini)(const struct lu_env *env,
struct dt_it *di);
+
+ /**
+ * Move position of iterator.
+ *
+ * Move the position of the specified iterator to the specified
+ * key.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ * \param[in] key key to position to
+ *
+ * \retval 0 if exact key is found
+ * \retval 1 if at the record with least key
+ * not larger than the key
+ * \retval negative negated errno on error
+ */
int (*get)(const struct lu_env *env,
struct dt_it *di,
const struct dt_key *key);
+
+ /**
+ * Release position
+ *
+ * Complimentary method for dt_it_ops::get() above. Some
+ * implementation can increase a reference on the iterator in
+ * dt_it_ops::get(). So the caller should be able to release
+ * with dt_it_ops::put().
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ */
void (*put)(const struct lu_env *env,
struct dt_it *di);
+
+ /**
+ * Move to next record.
+ *
+ * Moves the position of the iterator to a next record
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ *
+ * \retval 1 if no more records
+ * \retval 0 on success, the next record is found
+ * \retval negative negated errno on error
+ */
int (*next)(const struct lu_env *env,
struct dt_it *di);
+
+ /**
+ * Return key.
+ *
+ * Returns a pointer to a buffer containing the key of the
+ * record at the current position. The pointer is valid and
+ * retains data until ->get(), ->load() and ->fini() methods
+ * are called.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ *
+ * \retval pointer to key on success
+ * \retval ERR_PTR(errno) on error
+ */
struct dt_key *(*key)(const struct lu_env *env,
const struct dt_it *di);
+
+ /**
+ * Return key size.
+ *
+ * Returns size of the key at the current position.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ *
+ * \retval key's size on success
+ * \retval negative negated errno on error
+ */
int (*key_size)(const struct lu_env *env,
const struct dt_it *di);
+
+ /**
+ * Return record.
+ *
+ * Stores the value of the record at the current position. The
+ * buffer must be big enough (as negotiated with
+ * ->do_index_try() or ->rec_size()). The caller can specify
+ * she is interested only in part of the record, using attr
+ * argument (see LUDA_* definitions for the details).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ * \param[out] rec buffer to store value in
+ * \param[in] attr specify part of the value to copy
+ *
+ * \retval 0 on success
+ * \retval negative negated errno on error
+ */
int (*rec)(const struct lu_env *env,
const struct dt_it *di,
struct dt_rec *rec,
__u32 attr);
+
+ /**
+ * Return record size.
+ *
+ * Returns size of the record at the current position. The
+ * \a attr can be used to specify only the parts of the record
+ * needed to be returned. (see LUDA_* definitions for the
+ * details).
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ * \param[in] attr part of the record to return
+ *
+ * \retval record's size on success
+ * \retval negative negated errno on error
+ */
int (*rec_size)(const struct lu_env *env,
const struct dt_it *di,
__u32 attr);
+
+ /**
+ * Return a cookie (hash).
+ *
+ * Returns the cookie (usually hash) of the key at the current
+ * position. This allows the caller to resume iteration at this
+ * position later. The exact value is specific to implementation
+ * and should not be interpreted by the caller.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ *
+ * \retval cookie/hash of the key
+ */
__u64 (*store)(const struct lu_env *env,
const struct dt_it *di);
+
+ /**
+ * Initialize position using cookie/hash.
+ *
+ * Initializes the current position of the iterator to one
+ * described by the cookie/hash as returned by ->store()
+ * previously.
+ *
+ * \param[in] env execution environment for this thread
+ * \param[in] di iterator
+ * \param[in] hash cookie/hash value
+ *
+ * \retval positive if current position points to
+ * record with least cookie not larger
+ * than cookie
+ * \retval 0 if current position matches cookie
+ * \retval negative negated errno on error
+ */
int (*load)(const struct lu_env *env,
- const struct dt_it *di, __u64 hash);
+ const struct dt_it *di,
+ __u64 hash);
+
+ /**
+ * Not used
+ */
int (*key_rec)(const struct lu_env *env,
- const struct dt_it *di, void* key_rec);
+ const struct dt_it *di,
+ void *key_rec);
} dio_it;
};