4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #ifndef __LUSTRE_DT_OBJECT_H
34 #define __LUSTRE_DT_OBJECT_H
37 * Sub-class of lu_object with methods common for "data" objects in OST stack.
39 * Data objects behave like regular files: you can read/write them, get and
40 * set their attributes. Implementation of dt interface is supposed to
41 * implement some form of garbage collection, normally reference counting
44 * Examples: osd (lustre/osd) is an implementation of dt interface.
48 #include <obd_support.h>
50 * super-class definitions.
52 #include <lu_object.h>
54 #include <libcfs/libcfs.h>
57 struct proc_dir_entry;
63 struct dt_index_features;
66 struct ldlm_enqueue_info;
69 MNTOPT_USERXATTR = 0x00000001,
70 MNTOPT_ACL = 0x00000002,
73 struct dt_device_param {
74 unsigned ddp_max_name_len;
75 unsigned ddp_max_nlink;
76 unsigned ddp_symlink_max;
78 unsigned ddp_max_ea_size;
79 unsigned ddp_mount_type;
80 unsigned long long ddp_maxbytes;
81 /* per-inode space consumption */
83 /* maximum number of blocks in an extent */
84 unsigned ddp_max_extent_blks;
85 /* per-extent insertion overhead to be used by client for grant
87 unsigned ddp_extent_tax;
91 * Per-transaction commit callback function
93 struct dt_txn_commit_cb;
94 typedef void (*dt_cb_t)(struct lu_env *env, struct thandle *th,
95 struct dt_txn_commit_cb *cb, int err);
97 * Special per-transaction callback for cases when just commit callback
98 * is needed and per-device callback are not convenient to use
100 #define TRANS_COMMIT_CB_MAGIC 0xa0a00a0a
101 #define MAX_COMMIT_CB_STR_LEN 32
103 #define DCB_TRANS_STOP 0x1
104 struct dt_txn_commit_cb {
105 struct list_head dcb_linkage;
110 char dcb_name[MAX_COMMIT_CB_STR_LEN];
114 * Operations on dt device.
116 struct dt_device_operations {
118 * Return device-wide statistics.
120 * Return device-wide stats including block size, total and
121 * free blocks, total and free objects, etc. See struct obd_statfs
124 * \param[in] env execution environment for this thread
125 * \param[in] dev dt device
126 * \param[out] osfs stats information
128 * \retval 0 on success
129 * \retval negative negated errno on error
131 int (*dt_statfs)(const struct lu_env *env,
132 struct dt_device *dev,
133 struct obd_statfs *osfs);
136 * Create transaction.
138 * Create in-memory structure representing the transaction for the
139 * caller. The structure returned will be used by the calling thread
140 * to specify the transaction the updates belong to. Once created
141 * successfully ->dt_trans_stop() must be called in any case (with
142 * ->dt_trans_start() and updates or not) so that the transaction
143 * handle and other resources can be released by the layers below.
145 * \param[in] env execution environment for this thread
146 * \param[in] dev dt device
148 * \retval pointer to handle if creation succeeds
149 * \retval ERR_PTR(errno) if creation fails
151 struct thandle *(*dt_trans_create)(const struct lu_env *env,
152 struct dt_device *dev);
157 * Start the transaction. The transaction described by \a th can be
158 * started only once. Another start is considered as an error.
159 * A thread is not supposed to start a transaction while another
160 * transaction isn't closed by the thread (though multiple handles
161 * can be created). The caller should start the transaction once
162 * all possible updates are declared (see the ->do_declare_* methods
163 * below) and all the needed resources are reserved.
165 * \param[in] env execution environment for this thread
166 * \param[in] dev dt device
167 * \param[in] th transaction handle
169 * \retval 0 on success
170 * \retval negative negated errno on error
172 int (*dt_trans_start)(const struct lu_env *env,
173 struct dt_device *dev,
179 * Once stopped the transaction described by \a th is complete (all
180 * the needed updates are applied) and further processing such as
181 * flushing to disk, sending to another target, etc, is handled by
182 * lower layers. The caller can't access this transaction by the
183 * handle anymore (except from the commit callbacks, see below).
185 * \param[in] env execution environment for this thread
186 * \param[in] dev dt device
187 * \param[in] th transaction handle
189 * \retval 0 on success
190 * \retval negative negated errno on error
192 int (*dt_trans_stop)(const struct lu_env *env,
193 struct dt_device *dev,
197 * Add commit callback to the transaction.
199 * Add a commit callback to the given transaction handle. The callback
200 * will be called when the associated transaction is stored. I.e. the
201 * transaction will survive an event like power off if the callback did
202 * run. The number of callbacks isn't limited, but you should note that
203 * some disk filesystems do handle the commit callbacks in the thread
204 * handling commit/flush of all the transactions, meaning that new
205 * transactions are blocked from commit and flush until all the
206 * callbacks are done. Also, note multiple callbacks can be running
207 * concurrently using multiple CPU cores. The callbacks will be running
208 * in a special environment which can not be used to pass data around.
210 * \param[in] th transaction handle
211 * \param[in] dcb commit callback description
213 * \retval 0 on success
214 * \retval negative negated errno on error
216 int (*dt_trans_cb_add)(struct thandle *th,
217 struct dt_txn_commit_cb *dcb);
220 * Return FID of root index object.
222 * Return the FID of the root object in the filesystem. This object
223 * is usually provided as a bootstrap point by a disk filesystem.
224 * This is up to the implementation which FID to use, though
225 * [FID_SEQ_ROOT:1:0] is reserved for this purpose.
227 * \param[in] env execution environment for this thread
228 * \param[in] dev dt device
229 * \param[out] fid FID of the root object
231 * \retval 0 on success
232 * \retval negative negated errno on error
234 int (*dt_root_get)(const struct lu_env *env,
235 struct dt_device *dev,
239 * Return device configuration data.
241 * Return device (disk fs, actually) specific configuration.
242 * The configuration isn't subject to change at runtime.
243 * See struct dt_device_param for the details.
245 * \param[in] env execution environment for this thread
246 * \param[in] dev dt device
247 * \param[out] param configuration parameters
249 void (*dt_conf_get)(const struct lu_env *env,
250 const struct dt_device *dev,
251 struct dt_device_param *param);
256 * Sync all the cached state (dirty buffers, pages, etc) to the
257 * persistent storage. The method returns control once the sync is
258 * complete. This operation may incur significant I/O to disk and
259 * should be reserved for cases where a global sync is strictly
262 * \param[in] env execution environment for this thread
263 * \param[in] dev dt device
265 * \retval 0 on success
266 * \retval negative negated errno on error
268 int (*dt_sync)(const struct lu_env *env,
269 struct dt_device *dev);
272 * Make device read-only.
274 * Prevent new modifications to the device. This is a very specific
275 * state where all the changes are accepted successfully and the
276 * commit callbacks are called, but persistent state never changes.
277 * Used only in the tests to simulate power-off scenario.
279 * \param[in] env execution environment for this thread
280 * \param[in] dev dt device
282 * \retval 0 on success
283 * \retval negative negated errno on error
285 int (*dt_ro)(const struct lu_env *env,
286 struct dt_device *dev);
289 * Start transaction commit asynchronously.
292 * Provide a hint to the underlying filesystem that it should start
293 * committing soon. The control returns immediately. It's up to the
294 * layer implementing the method how soon to start committing. Usually
295 * this should be throttled to some extent, otherwise the number of
296 * aggregated transaction goes too high causing performance drop.
298 * \param[in] env execution environment for this thread
299 * \param[in] dev dt device
301 * \retval 0 on success
302 * \retval negative negated errno on error
304 int (*dt_commit_async)(const struct lu_env *env,
305 struct dt_device *dev);
308 struct dt_index_features {
309 /** required feature flags from enum dt_index_flags */
311 /** minimal required key size */
312 size_t dif_keysize_min;
313 /** maximal required key size, 0 if no limit */
314 size_t dif_keysize_max;
315 /** minimal required record size */
316 size_t dif_recsize_min;
317 /** maximal required record size, 0 if no limit */
318 size_t dif_recsize_max;
319 /** pointer size for record */
323 enum dt_index_flags {
324 /** index supports variable sized keys */
325 DT_IND_VARKEY = 1 << 0,
326 /** index supports variable sized records */
327 DT_IND_VARREC = 1 << 1,
328 /** index can be modified */
329 DT_IND_UPDATE = 1 << 2,
330 /** index supports records with non-unique (duplicate) keys */
331 DT_IND_NONUNQ = 1 << 3,
333 * index support fixed-size keys sorted with natural numerical way
334 * and is able to return left-side value if no exact value found
336 DT_IND_RANGE = 1 << 4,
340 * Features, required from index to support file system directories (mapping
343 extern const struct dt_index_features dt_directory_features;
344 extern const struct dt_index_features dt_otable_features;
345 extern const struct dt_index_features dt_lfsck_layout_orphan_features;
346 extern const struct dt_index_features dt_lfsck_layout_dangling_features;
347 extern const struct dt_index_features dt_lfsck_namespace_features;
349 /* index features supported by the accounting objects */
350 extern const struct dt_index_features dt_acct_features;
352 /* index features supported by the quota global indexes */
353 extern const struct dt_index_features dt_quota_glb_features;
355 /* index features supported by the quota slave indexes */
356 extern const struct dt_index_features dt_quota_slv_features;
358 /* index features supported by the nodemap index */
359 extern const struct dt_index_features dt_nodemap_features;
362 * This is a general purpose dt allocation hint.
363 * It now contains the parent object.
364 * It can contain any allocation hint in the future.
366 struct dt_allocation_hint {
367 struct dt_object *dah_parent;
368 const void *dah_eadata;
374 * object type specifier.
377 enum dt_format_type {
382 /** for special index */
384 /** for symbolic link */
389 * object format specifier.
391 struct dt_object_format {
392 /** type for dt object */
393 enum dt_format_type dof_type;
403 * special index need feature as parameter to create
407 const struct dt_index_features *di_feat;
412 enum dt_format_type dt_mode_to_dft(__u32 mode);
414 typedef __u64 dt_obj_version_t;
416 union ldlm_policy_data;
419 * A dt_object provides common operations to create and destroy
420 * objects and to manage regular and extended attributes.
422 struct dt_object_operations {
424 * Get read lock on object.
426 * Read lock is compatible with other read locks, so it's shared.
427 * Read lock is not compatible with write lock which is exclusive.
428 * The lock is blocking and can't be used from an interrupt context.
430 * \param[in] env execution environment for this thread
431 * \param[in] dt object to lock for reading
432 * \param[in] role a hint to debug locks (see kernel's mutexes)
434 void (*do_read_lock)(const struct lu_env *env,
435 struct dt_object *dt,
439 * Get write lock on object.
441 * Write lock is exclusive and cannot be shared. The lock is blocking
442 * and can't be used from an interrupt context.
444 * \param[in] env execution environment for this thread
445 * \param[in] dt object to lock for writing
446 * \param[in] role a hint to debug locks (see kernel's mutexes)
449 void (*do_write_lock)(const struct lu_env *env,
450 struct dt_object *dt,
456 * \param[in] env execution environment for this thread
457 * \param[in] dt object
459 void (*do_read_unlock)(const struct lu_env *env,
460 struct dt_object *dt);
463 * Release write lock.
465 * \param[in] env execution environment for this thread
466 * \param[in] dt object
468 void (*do_write_unlock)(const struct lu_env *env,
469 struct dt_object *dt);
472 * Check whether write lock is held.
474 * The caller can learn whether write lock is held on the object
476 * \param[in] env execution environment for this thread
477 * \param[in] dt object
479 * \retval 0 no write lock
480 * \retval 1 write lock is held
482 int (*do_write_locked)(const struct lu_env *env,
483 struct dt_object *dt);
486 * Declare intention to request reqular attributes.
488 * Notity the underlying filesystem that the caller may request regular
489 * attributes with ->do_attr_get() soon. This allows OSD to implement
490 * prefetching logic in an object-oriented manner. The implementation
491 * can be noop. This method should avoid expensive delays such as
492 * waiting on disk I/O, otherwise the goal of enabling a performance
493 * optimization would be defeated.
495 * \param[in] env execution environment for this thread
496 * \param[in] dt object
498 * \retval 0 on success
499 * \retval negative negated errno on error
501 int (*do_declare_attr_get)(const struct lu_env *env,
502 struct dt_object *dt);
505 * Return regular attributes.
507 * The object must exist. Currently all the attributes should be
508 * returned, but in the future this can be improved so that only
509 * a selected set is returned. This can improve performance as in
510 * some cases attributes are stored in different places and
511 * getting them all can be an iterative and expensive process.
513 * \param[in] env execution environment for this thread
514 * \param[in] dt object
515 * \param[out] attr attributes to fill
517 * \retval 0 on success
518 * \retval negative negated errno on error
520 int (*do_attr_get)(const struct lu_env *env,
521 struct dt_object *dt,
522 struct lu_attr *attr);
525 * Declare intention to change regular object's attributes.
527 * Notify the underlying filesystem that the regular attributes may
528 * change in this transaction. This enables the layer below to prepare
529 * resources (e.g. journal credits in ext4). This method should be
530 * called between creating the transaction and starting it. Note that
531 * the la_valid field of \a attr specifies which attributes will change.
532 * The object need not exist.
534 * \param[in] env execution environment for this thread
535 * \param[in] dt object
536 * \param[in] attr attributes to change specified in attr.la_valid
537 * \param[in] th transaction handle
539 * \retval 0 on success
540 * \retval negative negated errno on error
542 int (*do_declare_attr_set)(const struct lu_env *env,
543 struct dt_object *dt,
544 const struct lu_attr *attr,
548 * Change regular attributes.
550 * Change regular attributes in the given transaction. Note only
551 * attributes flagged by attr.la_valid change. The object must
552 * exist. If the layer implementing this method is responsible for
553 * quota, then the method should maintain object accounting for the
554 * given credentials when la_uid/la_gid changes.
556 * \param[in] env execution environment for this thread
557 * \param[in] dt object
558 * \param[in] attr new attributes to apply
559 * \param[in] th transaction handle
561 * \retval 0 on success
562 * \retval negative negated errno on error
564 int (*do_attr_set)(const struct lu_env *env,
565 struct dt_object *dt,
566 const struct lu_attr *attr,
570 * Declare intention to request extented attribute.
572 * Notify the underlying filesystem that the caller may request extended
573 * attribute with ->do_xattr_get() soon. This allows OSD to implement
574 * prefetching logic in an object-oriented manner. The implementation
575 * can be noop. This method should avoid expensive delays such as
576 * waiting on disk I/O, otherwise the goal of enabling a performance
577 * optimization would be defeated.
579 * \param[in] env execution environment for this thread
580 * \param[in] dt object
581 * \param[in] buf unused, may be removed in the future
582 * \param[in] name name of the extended attribute
584 * \retval 0 on success
585 * \retval negative negated errno on error
587 int (*do_declare_xattr_get)(const struct lu_env *env,
588 struct dt_object *dt,
593 * Return a value of an extended attribute.
595 * The object must exist. If the buffer is NULL, then the method
596 * must return the size of the value.
598 * \param[in] env execution environment for this thread
599 * \param[in] dt object
600 * \param[out] buf buffer in which to store the value
601 * \param[in] name name of the extended attribute
603 * \retval 0 on success
604 * \retval -ERANGE if \a buf is too small
605 * \retval negative negated errno on error
606 * \retval positive value's size if \a buf is NULL or has zero size
608 int (*do_xattr_get)(const struct lu_env *env,
609 struct dt_object *dt,
614 * Declare intention to change an extended attribute.
616 * Notify the underlying filesystem that the extended attribute may
617 * change in this transaction. This enables the layer below to prepare
618 * resources (e.g. journal credits in ext4). This method should be
619 * called between creating the transaction and starting it. The object
622 * \param[in] env execution environment for this thread
623 * \param[in] dt object
624 * \param[in] buf buffer storing new value of the attribute
625 * \param[in] name name of the attribute
626 * \param[in] fl LU_XATTR_CREATE - fail if EA exists
627 * LU_XATTR_REPLACE - fail if EA doesn't exist
628 * \param[in] th transaction handle
630 * \retval 0 on success
631 * \retval negative negated errno on error
633 int (*do_declare_xattr_set)(const struct lu_env *env,
634 struct dt_object *dt,
635 const struct lu_buf *buf,
641 * Set an extended attribute.
643 * Change or replace the specified extended attribute (EA).
644 * The flags passed in \a fl dictate whether the EA is to be
645 * created or replaced, as follows.
646 * LU_XATTR_CREATE - fail if EA exists
647 * LU_XATTR_REPLACE - fail if EA doesn't exist
648 * The object must exist.
650 * \param[in] env execution environment for this thread
651 * \param[in] dt object
652 * \param[in] buf buffer storing new value of the attribute
653 * \param[in] name name of the attribute
654 * \param[in] fl flags indicating EA creation or replacement
655 * \param[in] th transaction handle
657 * \retval 0 on success
658 * \retval negative negated errno on error
660 int (*do_xattr_set)(const struct lu_env *env,
661 struct dt_object *dt,
662 const struct lu_buf *buf,
668 * Declare intention to delete an extended attribute.
670 * Notify the underlying filesystem that the extended attribute may
671 * be deleted in this transaction. This enables the layer below to
672 * prepare resources (e.g. journal credits in ext4). This method
673 * should be called between creating the transaction and starting it.
674 * The object need not exist.
676 * \param[in] env execution environment for this thread
677 * \param[in] dt object
678 * \param[in] name name of the attribute
679 * \param[in] th transaction handle
681 * \retval 0 on success
682 * \retval negative negated errno on error
684 int (*do_declare_xattr_del)(const struct lu_env *env,
685 struct dt_object *dt,
690 * Delete an extended attribute.
692 * This method deletes the specified extended attribute. The object
695 * \param[in] env execution environment for this thread
696 * \param[in] dt object
697 * \param[in] name name of the attribute
698 * \param[in] th transaction handle
700 * \retval 0 on success
701 * \retval negative negated errno on error
703 int (*do_xattr_del)(const struct lu_env *env,
704 struct dt_object *dt,
709 * Return a list of the extended attributes.
711 * Fills the passed buffer with a list of the extended attributes
712 * found in the object. The names are separated with '\0'.
713 * The object must exist.
715 * \param[in] env execution environment for this thread
716 * \param[in] dt object
717 * \param[out] buf buffer to put the list in
719 * \retval positive bytes used/required in the buffer
720 * \retval negative negated errno on error
722 int (*do_xattr_list)(const struct lu_env *env,
723 struct dt_object *dt,
724 const struct lu_buf *buf);
727 * Prepare allocation hint for a new object.
729 * This method is used by the caller to inform OSD of the parent-child
730 * relationship between two objects and enable efficient object
731 * allocation. Filled allocation hint will be passed to ->do_create()
734 * \param[in] env execution environment for this thread
735 * \param[out] ah allocation hint
736 * \param[in] parent parent object (can be NULL)
737 * \param[in] child child object
738 * \param[in] _mode type of the child object
740 void (*do_ah_init)(const struct lu_env *env,
741 struct dt_allocation_hint *ah,
742 struct dt_object *parent,
743 struct dt_object *child,
747 * Declare intention to create a new object.
749 * Notify the underlying filesystem that the object may be created
750 * in this transaction. This enables the layer below to prepare
751 * resources (e.g. journal credits in ext4). This method should be
752 * called between creating the transaction and starting it.
754 * If the layer implementing this method is responsible for quota,
755 * then the method should reserve an object for the given credentials
756 * and return an error if quota is over. If object creation later
757 * fails for some reason, then the reservation should be released
758 * properly (usually in ->dt_trans_stop()).
760 * \param[in] env execution environment for this thread
761 * \param[in] dt object
762 * \param[in] attr attributes of the new object
763 * \param[in] hint allocation hint
764 * \param[in] dof object format
765 * \param[in] th transaction handle
767 * \retval 0 on success
768 * \retval negative negated errno on error
770 int (*do_declare_create)(const struct lu_env *env,
771 struct dt_object *dt,
772 struct lu_attr *attr,
773 struct dt_allocation_hint *hint,
774 struct dt_object_format *dof,
780 * The method creates the object passed with the specified attributes
781 * and object format. Object allocation procedure can use information
782 * stored in the allocation hint. Different object formats are supported
783 * (see enum dt_format_type and struct dt_object_format) depending on
784 * the device. If creation succeeds, then LOHA_EXISTS flag must be set
785 * in the LU-object header attributes.
787 * If the layer implementing this method is responsible for quota,
788 * then the method should maintain object accounting for the given
791 * \param[in] env execution environment for this thread
792 * \param[in] dt object
793 * \param[in] attr attributes of the new object
794 * \param[in] hint allocation hint
795 * \param[in] dof object format
796 * \param[in] th transaction handle
798 * \retval 0 on success
799 * \retval negative negated errno on error
801 int (*do_create)(const struct lu_env *env,
802 struct dt_object *dt,
803 struct lu_attr *attr,
804 struct dt_allocation_hint *hint,
805 struct dt_object_format *dof,
809 * Declare intention to destroy an object.
811 * Notify the underlying filesystem that the object may be destroyed
812 * in this transaction. This enables the layer below to prepare
813 * resources (e.g. journal credits in ext4). This method should be
814 * called between creating the transaction and starting it. The object
817 * \param[in] env execution environment for this thread
818 * \param[in] dt object
819 * \param[in] th transaction handle
821 * \retval 0 on success
822 * \retval negative negated errno on error
824 int (*do_declare_destroy)(const struct lu_env *env,
825 struct dt_object *dt,
831 * This method destroys the object and all the resources associated
832 * with the object (data, key/value pairs, extended attributes, etc).
833 * The object must exist. If destroy is successful, then flag
834 * LU_OBJECT_HEARD_BANSHEE should be set to forbid access to this
835 * instance of in-core object. Any subsequent access to the same FID
836 * should get another instance with no LOHA_EXIST flag set.
838 * If the layer implementing this method is responsible for quota,
839 * then the method should maintain object accounting for the given
842 * \param[in] env execution environment for this thread
843 * \param[in] dt object
844 * \param[in] th transaction handle
846 * \retval 0 on success
847 * \retval negative negated errno on error
849 int (*do_destroy)(const struct lu_env *env,
850 struct dt_object *dt,
854 * Try object as an index.
856 * Announce that this object is going to be used as an index. This
857 * operation checks that object supports indexing operations and
858 * installs appropriate dt_index_operations vector on success.
859 * Also probes for features. Operation is successful if all required
860 * features are supported. It's not possible to access the object
861 * with index methods before ->do_index_try() returns success.
863 * \param[in] env execution environment for this thread
864 * \param[in] dt object
865 * \param[in] feat index features
867 * \retval 0 on success
868 * \retval negative negated errno on error
870 int (*do_index_try)(const struct lu_env *env,
871 struct dt_object *dt,
872 const struct dt_index_features *feat);
875 * Declare intention to increment nlink count.
877 * Notify the underlying filesystem that the nlink regular attribute
878 * be changed in this transaction. This enables the layer below to
879 * prepare resources (e.g. journal credits in ext4). This method
880 * should be called between creating the transaction and starting it.
881 * The object need not exist.
883 * \param[in] env execution environment for this thread
884 * \param[in] dt object
885 * \param[in] th transaction handle
887 * \retval 0 on success
888 * \retval negative negated errno on error
890 int (*do_declare_ref_add)(const struct lu_env *env,
891 struct dt_object *dt,
897 * Increment nlink (from the regular attributes set) in the given
898 * transaction. Note the absolute limit for nlink should be learnt
899 * from struct dt_device_param::ddp_max_nlink. The object must exist.
901 * \param[in] env execution environment for this thread
902 * \param[in] dt object
903 * \param[in] th transaction handle
905 * \retval 0 on success
906 * \retval negative negated errno on error
908 int (*do_ref_add)(const struct lu_env *env,
909 struct dt_object *dt, struct thandle *th);
912 * Declare intention to decrement nlink count.
914 * Notify the underlying filesystem that the nlink regular attribute
915 * be changed in this transaction. This enables the layer below to
916 * prepare resources (e.g. journal credits in ext4). This method
917 * should be called between creating the transaction and starting it.
918 * The object need not exist.
920 * \param[in] env execution environment for this thread
921 * \param[in] dt object
922 * \param[in] th transaction handle
924 * \retval 0 on success
925 * \retval negative negated errno on error
927 int (*do_declare_ref_del)(const struct lu_env *env,
928 struct dt_object *dt,
934 * Decrement nlink (from the regular attributes set) in the given
935 * transaction. The object must exist.
937 * \param[in] env execution environment for this thread
938 * \param[in] dt object
939 * \param[in] th transaction handle
941 * \retval 0 on success
942 * \retval negative negated errno on error
944 int (*do_ref_del)(const struct lu_env *env,
945 struct dt_object *dt,
951 * The method is called to sync specified range of the object to a
952 * persistent storage. The control is returned once the operation is
953 * complete. The difference from ->do_sync() is that the object can
954 * be in-sync with the persistent storage (nothing to flush), then
955 * the method returns quickly with no I/O overhead. So, this method
956 * should be preferred over ->do_sync() where possible. Also note that
957 * if the object isn't clean, then some disk filesystems will call
958 * ->do_sync() to maintain overall consistency, in which case it's
959 * still very expensive.
961 * \param[in] env execution environment for this thread
962 * \param[in] dt object
963 * \param[in] start start of the range to sync
964 * \param[in] end end of the range to sync
966 * \retval 0 on success
967 * \retval negative negated errno on error
969 int (*do_object_sync)(const struct lu_env *env,
970 struct dt_object *obj,
977 * Lock object(s) using Distributed Lock Manager (LDLM).
979 * Get LDLM locks for the object. Currently used to lock "remote"
980 * objects in DNE configuration - a service running on MDTx needs
981 * to lock an object on MDTy.
983 * \param[in] env execution environment for this thread
984 * \param[in] dt object
985 * \param[out] lh lock handle, sometimes used, sometimes not
986 * \param[in] einfo ldlm callbacks, locking type and mode
987 * \param[out] einfo private data to be passed to unlock later
988 * \param[in] policy inodebits data
990 * \retval 0 on success
991 * \retval negative negated errno on error
993 int (*do_object_lock)(const struct lu_env *env, struct dt_object *dt,
994 struct lustre_handle *lh,
995 struct ldlm_enqueue_info *einfo,
996 union ldlm_policy_data *policy);
1001 * Release LDLM lock(s) granted with ->do_object_lock().
1003 * \param[in] env execution environment for this thread
1004 * \param[in] dt object
1005 * \param[in] einfo lock handles, from ->do_object_lock()
1006 * \param[in] policy inodebits data
1008 * \retval 0 on success
1009 * \retval negative negated errno on error
1011 int (*do_object_unlock)(const struct lu_env *env,
1012 struct dt_object *dt,
1013 struct ldlm_enqueue_info *einfo,
1014 union ldlm_policy_data *policy);
1017 * Invalidate attribute cache.
1019 * This method invalidate attribute cache of the object, which is on OSP
1022 * \param[in] env execution envionment for this thread
1023 * \param[in] dt object
1025 * \retval 0 on success
1026 * \retval negative negated errno on error
1028 int (*do_invalidate)(const struct lu_env *env, struct dt_object *dt);
1032 * Per-dt-object operations on "file body" - unstructure raw data.
1034 struct dt_body_operations {
1038 * Read unstructured data from an existing regular object.
1039 * Only data before attr.la_size is returned.
1041 * \param[in] env execution environment for this thread
1042 * \param[in] dt object
1043 * \param[out] buf buffer (including size) to copy data in
1044 * \param[in] pos position in the object to start
1045 * \param[out] pos original value of \a pos + bytes returned
1047 * \retval positive bytes read on success
1048 * \retval negative negated errno on error
1050 ssize_t (*dbo_read)(const struct lu_env *env,
1051 struct dt_object *dt,
1056 * Declare intention to write data to object.
1058 * Notify the underlying filesystem that data may be written in
1059 * this transaction. This enables the layer below to prepare resources
1060 * (e.g. journal credits in ext4). This method should be called
1061 * between creating the transaction and starting it. The object need
1062 * not exist. If the layer implementing this method is responsible for
1063 * quota, then the method should reserve space for the given credentials
1064 * and return an error if quota is over. If the write later fails
1065 * for some reason, then the reserve should be released properly
1066 * (usually in ->dt_trans_stop()).
1068 * \param[in] env execution environment for this thread
1069 * \param[in] dt object
1070 * \param[in] buf buffer (including size) to copy data from
1071 * \param[in] pos position in the object to start
1072 * \param[in] th transaction handle
1074 * \retval 0 on success
1075 * \retval negative negated errno on error
1077 ssize_t (*dbo_declare_write)(const struct lu_env *env,
1078 struct dt_object *dt,
1079 const struct lu_buf *buf,
1081 struct thandle *th);
1084 * Write unstructured data to regular existing object.
1086 * The method allocates space and puts data in. Also, the method should
1087 * maintain attr.la_size properly. Partial writes are possible.
1089 * If the layer implementing this method is responsible for quota,
1090 * then the method should maintain space accounting for the given
1093 * \param[in] env execution environment for this thread
1094 * \param[in] dt object
1095 * \param[in] buf buffer (including size) to copy data from
1096 * \param[in] pos position in the object to start
1097 * \param[out] pos \a pos + bytes written
1098 * \param[in] th transaction handle
1099 * \param[in] ignore unused (was used to request quota ignorance)
1101 * \retval positive bytes written on success
1102 * \retval negative negated errno on error
1104 ssize_t (*dbo_write)(const struct lu_env *env,
1105 struct dt_object *dt,
1106 const struct lu_buf *buf,
1112 * Return buffers for data.
1114 * This method is used to access data with no copying. It's so-called
1115 * zero-copy I/O. The method returns the descriptors for the internal
1116 * buffers where data are managed by the disk filesystem. For example,
1117 * pagecache in case of ext4 or ARC with ZFS. Then other components
1118 * (e.g. networking) can transfer data from or to the buffers with no
1119 * additional copying.
1121 * The method should fill an array of struct niobuf_local, where
1122 * each element describes a full or partial page for data at specific
1123 * offset. The caller should use page/lnb_page_offset/len to find data
1124 * at object's offset lnb_file_offset.
1126 * The memory referenced by the descriptors can't change its purpose
1127 * until the complementary ->dbo_bufs_put() is called. The caller should
1128 * specify if the buffers are used to read or modify data so that OSD
1129 * can decide how to initialize the buffers: bring all the data for
1130 * reads or just bring partial buffers for write. Note: the method does
1131 * not check whether output array is large enough.
1133 * \param[in] env execution environment for this thread
1134 * \param[in] dt object
1135 * \param[in] pos position in the object to start
1136 * \param[in] len size of region in bytes
1137 * \param[out] lb array of descriptors to fill
1138 * \param[in] rw 0 if used to read, 1 if used for write
1140 * \retval positive number of descriptors on success
1141 * \retval negative negated errno on error
1143 int (*dbo_bufs_get)(const struct lu_env *env,
1144 struct dt_object *dt,
1147 struct niobuf_local *lb,
1151 * Release reference granted by ->dbo_bufs_get().
1153 * Release the reference granted by the previous ->dbo_bufs_get().
1154 * Note the references are counted.
1156 * \param[in] env execution environment for this thread
1157 * \param[in] dt object
1158 * \param[out] lb array of descriptors to fill
1159 * \param[in] nr size of the array
1161 * \retval 0 on success
1162 * \retval negative negated errno on error
1164 int (*dbo_bufs_put)(const struct lu_env *env,
1165 struct dt_object *dt,
1166 struct niobuf_local *lb,
1170 * Prepare buffers for reading.
1172 * The method is called on the given buffers to fill them with data
1173 * if that wasn't done in ->dbo_bufs_get(). The idea is that the
1174 * caller should be able to get few buffers for discontiguous regions
1175 * using few calls to ->dbo_bufs_get() and then request them all for
1176 * the preparation with a single call, so that OSD can fire many I/Os
1177 * to run concurrently. It's up to the specific OSD whether to implement
1178 * this logic in ->dbo_read_prep() or just use ->dbo_bufs_get() to
1179 * prepare data for every requested region individually.
1181 * \param[in] env execution environment for this thread
1182 * \param[in] dt object
1183 * \param[in] lnb array of buffer descriptors
1184 * \param[in] nr size of the array
1186 * \retval 0 on success
1187 * \retval negative negated errno on error
1189 int (*dbo_read_prep)(const struct lu_env *env,
1190 struct dt_object *dt,
1191 struct niobuf_local *lnb,
1195 * Prepare buffers for write.
1197 * This method is called on the given buffers to ensure the partial
1198 * buffers contain correct data. The underlying idea is the same as
1199 * in ->db_read_prep().
1201 * \param[in] env execution environment for this thread
1202 * \param[in] dt object
1203 * \param[in] lb array of buffer descriptors
1204 * \param[in] nr size of the array
1206 * \retval 0 on success
1207 * \retval negative negated errno on error
1209 int (*dbo_write_prep)(const struct lu_env *env,
1210 struct dt_object *dt,
1211 struct niobuf_local *lb,
1215 * Declare intention to write data stored in the buffers.
1217 * Notify the underlying filesystem that data may be written in
1218 * this transaction. This enables the layer below to prepare resources
1219 * (e.g. journal credits in ext4). This method should be called
1220 * between creating the transaction and starting it.
1222 * If the layer implementing this method is responsible for quota,
1223 * then the method should be reserving a space for the given
1224 * credentials and return an error if quota is exceeded. If the write
1225 * later fails for some reason, then the reserve should be released
1226 * properly (usually in ->dt_trans_stop()).
1228 * \param[in] env execution environment for this thread
1229 * \param[in] dt object
1230 * \param[in] lb array of descriptors
1231 * \param[in] nr size of the array
1232 * \param[in] th transaction handle
1234 * \retval 0 on success
1235 * \retval negative negated errno on error
1237 int (*dbo_declare_write_commit)(const struct lu_env *env,
1238 struct dt_object *dt,
1239 struct niobuf_local *lb,
1241 struct thandle *th);
1244 * Write to existing object.
1246 * This method is used to write data to a persistent storage using
1247 * the buffers returned by ->dbo_bufs_get(). The caller puts new
1248 * data into the buffers using own mechanisms (e.g. direct transfer
1249 * from a NIC). The method should maintain attr.la_size. Also,
1250 * attr.la_blocks should be maintained but this can be done in lazy
1251 * manner, when actual allocation happens.
1253 * If the layer implementing this method is responsible for quota,
1254 * then the method should maintain space accounting for the given
1257 * \param[in] env execution environment for this thread
1258 * \param[in] dt object
1259 * \param[in] lb array of descriptors for the buffers
1260 * \param[in] nr size of the array
1261 * \param[in] th transaction handle
1263 * \retval 0 on success
1264 * \retval negative negated errno on error
1266 int (*dbo_write_commit)(const struct lu_env *env,
1267 struct dt_object *dt,
1268 struct niobuf_local *lb,
1270 struct thandle *th);
1273 * Return logical to physical block mapping for a given extent
1275 * \param[in] env execution environment for this thread
1276 * \param[in] dt object
1277 * \param[in] fm describe the region to map and the output buffer
1278 * see the details in include/linux/fiemap.h
1280 * \retval 0 on success
1281 * \retval negative negated errno on error
1283 int (*dbo_fiemap_get)(const struct lu_env *env,
1284 struct dt_object *dt,
1288 * Declare intention to deallocate space from an object.
1290 * Notify the underlying filesystem that space may be deallocated in
1291 * this transactions. This enables the layer below to prepare resources
1292 * (e.g. journal credits in ext4). This method should be called between
1293 * creating the transaction and starting it. The object need not exist.
1295 * \param[in] env execution environment for this thread
1296 * \param[in] dt object
1297 * \param[in] start the start of the region to deallocate
1298 * \param[in] end the end of the region to deallocate
1299 * \param[in] th transaction handle
1301 * \retval 0 on success
1302 * \retval negative negated errno on error
1304 int (*dbo_declare_punch)(const struct lu_env *env,
1305 struct dt_object *dt,
1308 struct thandle *th);
1311 * Deallocate specified region in an object.
1313 * This method is used to deallocate (release) space possibly consumed
1314 * by the given region of the object. If the layer implementing this
1315 * method is responsible for quota, then the method should maintain
1316 * space accounting for the given credentials.
1318 * \param[in] env execution environment for this thread
1319 * \param[in] dt object
1320 * \param[in] start the start of the region to deallocate
1321 * \param[in] end the end of the region to deallocate
1322 * \param[in] th transaction handle
1324 * \retval 0 on success
1325 * \retval negative negated errno on error
1327 int (*dbo_punch)(const struct lu_env *env,
1328 struct dt_object *dt,
1331 struct thandle *th);
1333 * Give advices on specified region in an object.
1335 * This method is used to give advices about access pattern on an
1336 * given region of the object. The disk filesystem understands
1337 * the advices and tunes cache/read-ahead policies.
1339 * \param[in] env execution environment for this thread
1340 * \param[in] dt object
1341 * \param[in] start the start of the region affected
1342 * \param[in] end the end of the region affected
1343 * \param[in] advice advice type
1345 * \retval 0 on success
1346 * \retval negative negated errno on error
1348 int (*dbo_ladvise)(const struct lu_env *env,
1349 struct dt_object *dt,
1352 enum lu_ladvise_type advice);
1356 * Incomplete type of index record.
1361 * Incomplete type of index key.
1366 * Incomplete type of dt iterator.
1371 * Per-dt-object operations on object as index. Index is a set of key/value
1372 * pairs abstracted from an on-disk representation. An index supports the
1373 * number of operations including lookup by key, insert and delete. Also,
1374 * an index can be iterated to find the pairs one by one, from a beginning
1375 * or specified point.
1377 struct dt_index_operations {
1379 * Lookup in an index by key.
1381 * The method returns a value for the given key. Key/value format
1382 * and size should have been negotiated with ->do_index_try() before.
1383 * Thus it's the caller's responsibility to provide the method with
1384 * proper key and big enough buffer. No external locking is required,
1385 * all the internal consistency should be implemented by the method
1386 * or lower layers. The object should should have been created with
1387 * type DFT_INDEX or DFT_DIR.
1389 * \param[in] env execution environment for this thread
1390 * \param[in] dt object
1391 * \param[out] rec buffer where value will be stored
1392 * \param[in] key key
1394 * \retval 0 on success
1395 * \retval -ENOENT if key isn't found
1396 * \retval negative negated errno on error
1398 int (*dio_lookup)(const struct lu_env *env,
1399 struct dt_object *dt,
1401 const struct dt_key *key);
1404 * Declare intention to insert a key/value into an index.
1406 * Notify the underlying filesystem that new key/value may be inserted
1407 * in this transaction. This enables the layer below to prepare
1408 * resources (e.g. journal credits in ext4). This method should be
1409 * called between creating the transaction and starting it. key/value
1410 * format and size is subject to ->do_index_try().
1412 * \param[in] env execution environment for this thread
1413 * \param[in] dt object
1414 * \param[in] rec buffer storing value
1415 * \param[in] key key
1416 * \param[in] th transaction handle
1418 * \retval 0 on success
1419 * \retval negative negated errno on error
1421 int (*dio_declare_insert)(const struct lu_env *env,
1422 struct dt_object *dt,
1423 const struct dt_rec *rec,
1424 const struct dt_key *key,
1425 struct thandle *th);
1428 * Insert a new key/value pair into an index.
1430 * The method inserts specified key/value pair into the given index
1431 * object. The internal consistency is maintained by the method or
1432 * the functionality below. The format and size of key/value should
1433 * have been negotiated before using ->do_index_try(), no additional
1434 * information can be specified to the method. The keys are unique
1437 * \param[in] env execution environment for this thread
1438 * \param[in] dt object
1439 * \param[in] rec buffer storing value
1440 * \param[in] key key
1441 * \param[in] th transaction handle
1442 * \param[in] ignore unused (was used to request quota ignorance)
1444 * \retval 0 on success
1445 * \retval negative negated errno on error
1447 int (*dio_insert)(const struct lu_env *env,
1448 struct dt_object *dt,
1449 const struct dt_rec *rec,
1450 const struct dt_key *key,
1455 * Declare intention to delete a key/value from an index.
1457 * Notify the underlying filesystem that key/value may be deleted in
1458 * this transaction. This enables the layer below to prepare resources
1459 * (e.g. journal credits in ext4). This method should be called
1460 * between creating the transaction and starting it. Key/value format
1461 * and size is subject to ->do_index_try(). The object need not exist.
1463 * \param[in] env execution environment for this thread
1464 * \param[in] dt object
1465 * \param[in] key key
1466 * \param[in] th transaction handle
1468 * \retval 0 on success
1469 * \retval negative negated errno on error
1471 int (*dio_declare_delete)(const struct lu_env *env,
1472 struct dt_object *dt,
1473 const struct dt_key *key,
1474 struct thandle *th);
1477 * Delete key/value pair from an index.
1479 * The method deletes specified key and corresponding value from the
1480 * given index object. The internal consistency is maintained by the
1481 * method or the functionality below. The format and size of the key
1482 * should have been negotiated before using ->do_index_try(), no
1483 * additional information can be specified to the method.
1485 * \param[in] env execution environment for this thread
1486 * \param[in] dt object
1487 * \param[in] key key
1488 * \param[in] th transaction handle
1490 * \retval 0 on success
1491 * \retval negative negated errno on error
1493 int (*dio_delete)(const struct lu_env *env,
1494 struct dt_object *dt,
1495 const struct dt_key *key,
1496 struct thandle *th);
1499 * Iterator interface.
1501 * Methods to iterate over an existing index, list the keys stored and
1502 * associated values, get key/value size, etc.
1506 * Allocate and initialize new iterator.
1508 * The iterator is a handler to be used in the subsequent
1509 * methods to access index's content. Note the position is
1510 * not defined at this point and should be initialized with
1511 * ->get() or ->load() method.
1513 * \param[in] env execution environment for this thread
1514 * \param[in] dt object
1515 * \param[in] attr ask the iterator to return part of
1516 the records, see LUDA_* for details
1518 * \retval pointer iterator pointer on success
1519 * \retval ERR_PTR(errno) on error
1521 struct dt_it *(*init)(const struct lu_env *env,
1522 struct dt_object *dt,
1528 * Release the specified iterator and all the resources
1529 * associated (e.g. the object, index cache, etc).
1531 * \param[in] env execution environment for this thread
1532 * \param[in] di iterator to release
1534 void (*fini)(const struct lu_env *env,
1538 * Move position of iterator.
1540 * Move the position of the specified iterator to the specified
1543 * \param[in] env execution environment for this thread
1544 * \param[in] di iterator
1545 * \param[in] key key to position to
1547 * \retval 0 if exact key is found
1548 * \retval 1 if at the record with least key
1549 * not larger than the key
1550 * \retval negative negated errno on error
1552 int (*get)(const struct lu_env *env,
1554 const struct dt_key *key);
1559 * Complimentary method for dt_it_ops::get() above. Some
1560 * implementation can increase a reference on the iterator in
1561 * dt_it_ops::get(). So the caller should be able to release
1562 * with dt_it_ops::put().
1564 * \param[in] env execution environment for this thread
1565 * \param[in] di iterator
1567 void (*put)(const struct lu_env *env,
1571 * Move to next record.
1573 * Moves the position of the iterator to a next record
1575 * \param[in] env execution environment for this thread
1576 * \param[in] di iterator
1578 * \retval 1 if no more records
1579 * \retval 0 on success, the next record is found
1580 * \retval negative negated errno on error
1582 int (*next)(const struct lu_env *env,
1588 * Returns a pointer to a buffer containing the key of the
1589 * record at the current position. The pointer is valid and
1590 * retains data until ->get(), ->load() and ->fini() methods
1593 * \param[in] env execution environment for this thread
1594 * \param[in] di iterator
1596 * \retval pointer to key on success
1597 * \retval ERR_PTR(errno) on error
1599 struct dt_key *(*key)(const struct lu_env *env,
1600 const struct dt_it *di);
1605 * Returns size of the key at the current position.
1607 * \param[in] env execution environment for this thread
1608 * \param[in] di iterator
1610 * \retval key's size on success
1611 * \retval negative negated errno on error
1613 int (*key_size)(const struct lu_env *env,
1614 const struct dt_it *di);
1619 * Stores the value of the record at the current position. The
1620 * buffer must be big enough (as negotiated with
1621 * ->do_index_try() or ->rec_size()). The caller can specify
1622 * she is interested only in part of the record, using attr
1623 * argument (see LUDA_* definitions for the details).
1625 * \param[in] env execution environment for this thread
1626 * \param[in] di iterator
1627 * \param[out] rec buffer to store value in
1628 * \param[in] attr specify part of the value to copy
1630 * \retval 0 on success
1631 * \retval negative negated errno on error
1633 int (*rec)(const struct lu_env *env,
1634 const struct dt_it *di,
1639 * Return record size.
1641 * Returns size of the record at the current position. The
1642 * \a attr can be used to specify only the parts of the record
1643 * needed to be returned. (see LUDA_* definitions for the
1646 * \param[in] env execution environment for this thread
1647 * \param[in] di iterator
1648 * \param[in] attr part of the record to return
1650 * \retval record's size on success
1651 * \retval negative negated errno on error
1653 int (*rec_size)(const struct lu_env *env,
1654 const struct dt_it *di,
1658 * Return a cookie (hash).
1660 * Returns the cookie (usually hash) of the key at the current
1661 * position. This allows the caller to resume iteration at this
1662 * position later. The exact value is specific to implementation
1663 * and should not be interpreted by the caller.
1665 * \param[in] env execution environment for this thread
1666 * \param[in] di iterator
1668 * \retval cookie/hash of the key
1670 __u64 (*store)(const struct lu_env *env,
1671 const struct dt_it *di);
1674 * Initialize position using cookie/hash.
1676 * Initializes the current position of the iterator to one
1677 * described by the cookie/hash as returned by ->store()
1680 * \param[in] env execution environment for this thread
1681 * \param[in] di iterator
1682 * \param[in] hash cookie/hash value
1684 * \retval positive if current position points to
1685 * record with least cookie not larger
1687 * \retval 0 if current position matches cookie
1688 * \retval negative negated errno on error
1690 int (*load)(const struct lu_env *env,
1691 const struct dt_it *di,
1697 int (*key_rec)(const struct lu_env *env,
1698 const struct dt_it *di,
1703 enum dt_otable_it_valid {
1704 DOIV_ERROR_HANDLE = 0x0001,
1705 DOIV_DRYRUN = 0x0002,
1708 enum dt_otable_it_flags {
1709 /* Exit when fail. */
1710 DOIF_FAILOUT = 0x0001,
1712 /* Reset iteration position to the device beginning. */
1713 DOIF_RESET = 0x0002,
1715 /* There is up layer component uses the iteration. */
1716 DOIF_OUTUSED = 0x0004,
1718 /* Check only without repairing. */
1719 DOIF_DRYRUN = 0x0008,
1722 /* otable based iteration needs to use the common DT iteration APIs.
1723 * To initialize the iteration, it needs call dio_it::init() firstly.
1724 * Here is how the otable based iteration should prepare arguments to
1725 * call dt_it_ops::init().
1727 * For otable based iteration, the 32-bits 'attr' for dt_it_ops::init()
1728 * is composed of two parts:
1729 * low 16-bits is for valid bits, high 16-bits is for flags bits. */
1730 #define DT_OTABLE_IT_FLAGS_SHIFT 16
1731 #define DT_OTABLE_IT_FLAGS_MASK 0xffff0000
1734 struct lu_device dd_lu_dev;
1735 const struct dt_device_operations *dd_ops;
1738 * List of dt_txn_callback (see below). This is not protected in any
1739 * way, because callbacks are supposed to be added/deleted only during
1740 * single-threaded start-up shut-down procedures.
1742 struct list_head dd_txn_callbacks;
1743 unsigned int dd_record_fid_accessed:1;
1746 int dt_device_init(struct dt_device *dev, struct lu_device_type *t);
1747 void dt_device_fini(struct dt_device *dev);
1749 static inline int lu_device_is_dt(const struct lu_device *d)
1751 return ergo(d != NULL, d->ld_type->ldt_tags & LU_DEVICE_DT);
1754 static inline struct dt_device * lu2dt_dev(struct lu_device *l)
1756 LASSERT(lu_device_is_dt(l));
1757 return container_of0(l, struct dt_device, dd_lu_dev);
1761 struct lu_object do_lu;
1762 const struct dt_object_operations *do_ops;
1763 const struct dt_body_operations *do_body_ops;
1764 const struct dt_index_operations *do_index_ops;
1768 * In-core representation of per-device local object OID storage
1770 struct local_oid_storage {
1771 /* all initialized llog systems on this node linked by this */
1772 struct list_head los_list;
1774 /* how many handle's reference this los has */
1775 atomic_t los_refcount;
1776 struct dt_device *los_dev;
1777 struct dt_object *los_obj;
1779 /* data used to generate new fids */
1780 struct mutex los_id_lock;
1785 static inline struct lu_device *dt2lu_dev(struct dt_device *d)
1787 return &d->dd_lu_dev;
1790 static inline struct dt_object *lu2dt(struct lu_object *l)
1792 LASSERT(l == NULL || IS_ERR(l) || lu_device_is_dt(l->lo_dev));
1793 return container_of0(l, struct dt_object, do_lu);
1796 int dt_object_init(struct dt_object *obj,
1797 struct lu_object_header *h, struct lu_device *d);
1799 void dt_object_fini(struct dt_object *obj);
1801 static inline int dt_object_exists(const struct dt_object *dt)
1803 return lu_object_exists(&dt->do_lu);
1806 static inline int dt_object_remote(const struct dt_object *dt)
1808 return lu_object_remote(&dt->do_lu);
1811 static inline struct dt_object *lu2dt_obj(struct lu_object *o)
1813 LASSERT(ergo(o != NULL, lu_device_is_dt(o->lo_dev)));
1814 return container_of0(o, struct dt_object, do_lu);
1817 static inline struct dt_object *dt_object_child(struct dt_object *o)
1819 return container_of0(lu_object_next(&(o)->do_lu),
1820 struct dt_object, do_lu);
1824 * This is the general purpose transaction handle.
1825 * 1. Transaction Life Cycle
1826 * This transaction handle is allocated upon starting a new transaction,
1827 * and deallocated after this transaction is committed.
1828 * 2. Transaction Nesting
1829 * We do _NOT_ support nested transaction. So, every thread should only
1830 * have one active transaction, and a transaction only belongs to one
1831 * thread. Due to this, transaction handle need no reference count.
1832 * 3. Transaction & dt_object locking
1833 * dt_object locks should be taken inside transaction.
1834 * 4. Transaction & RPC
1835 * No RPC request should be issued inside transaction.
1838 /** the dt device on which the transactions are executed */
1839 struct dt_device *th_dev;
1841 /* point to the top thandle, XXX this is a bit hacky right now,
1842 * but normal device trans callback triggered by the bottom
1843 * device (OSP/OSD == sub thandle layer) needs to get the
1844 * top_thandle (see dt_txn_hook_start/stop()), so we put the
1845 * top thandle here for now, will fix it when we have better
1846 * callback mechanism */
1847 struct thandle *th_top;
1848 /** context for this transaction, tag is LCT_TX_HANDLE */
1849 struct lu_context th_ctx;
1851 /** additional tags (layers can add in declare) */
1854 /** the last operation result in this transaction.
1855 * this value is used in recovery */
1858 /** whether we need sync commit */
1859 unsigned int th_sync:1,
1860 /* local transation, no need to inform other layers */
1862 /* Whether we need wait the transaction to be submitted
1863 * (send to remote target) */
1865 /* complex transaction which will track updates on all targets,
1871 * Transaction call-backs.
1873 * These are invoked by osd (or underlying transaction engine) when
1874 * transaction changes state.
1876 * Call-backs are used by upper layers to modify transaction parameters and to
1877 * perform some actions on for each transaction state transition. Typical
1878 * example is mdt registering call-back to write into last-received file
1879 * before each transaction commit.
1881 struct dt_txn_callback {
1882 int (*dtc_txn_start)(const struct lu_env *env,
1883 struct thandle *txn, void *cookie);
1884 int (*dtc_txn_stop)(const struct lu_env *env,
1885 struct thandle *txn, void *cookie);
1886 void (*dtc_txn_commit)(struct thandle *txn, void *cookie);
1889 struct list_head dtc_linkage;
1892 void dt_txn_callback_add(struct dt_device *dev, struct dt_txn_callback *cb);
1893 void dt_txn_callback_del(struct dt_device *dev, struct dt_txn_callback *cb);
1895 int dt_txn_hook_start(const struct lu_env *env,
1896 struct dt_device *dev, struct thandle *txn);
1897 int dt_txn_hook_stop(const struct lu_env *env, struct thandle *txn);
1898 void dt_txn_hook_commit(struct thandle *txn);
1900 int dt_try_as_dir(const struct lu_env *env, struct dt_object *obj);
1903 * Callback function used for parsing path.
1904 * \see llo_store_resolve
1906 typedef int (*dt_entry_func_t)(const struct lu_env *env,
1910 #define DT_MAX_PATH 1024
1912 int dt_path_parser(const struct lu_env *env,
1913 char *local, dt_entry_func_t entry_func,
1917 dt_store_resolve(const struct lu_env *env, struct dt_device *dt,
1918 const char *path, struct lu_fid *fid);
1920 struct dt_object *dt_store_open(const struct lu_env *env,
1921 struct dt_device *dt,
1922 const char *dirname,
1923 const char *filename,
1924 struct lu_fid *fid);
1926 struct dt_object *dt_find_or_create(const struct lu_env *env,
1927 struct dt_device *dt,
1928 const struct lu_fid *fid,
1929 struct dt_object_format *dof,
1930 struct lu_attr *attr);
1932 struct dt_object *dt_locate_at(const struct lu_env *env,
1933 struct dt_device *dev,
1934 const struct lu_fid *fid,
1935 struct lu_device *top_dev,
1936 const struct lu_object_conf *conf);
1938 static inline struct dt_object *
1939 dt_locate(const struct lu_env *env, struct dt_device *dev,
1940 const struct lu_fid *fid)
1942 return dt_locate_at(env, dev, fid,
1943 dev->dd_lu_dev.ld_site->ls_top_dev, NULL);
1946 static inline struct dt_object *
1947 dt_object_locate(struct dt_object *dto, struct dt_device *dt_dev)
1949 struct lu_object *lo;
1951 list_for_each_entry(lo, &dto->do_lu.lo_header->loh_layers, lo_linkage) {
1952 if (lo->lo_dev == &dt_dev->dd_lu_dev)
1953 return container_of(lo, struct dt_object, do_lu);
1958 int local_oid_storage_init(const struct lu_env *env, struct dt_device *dev,
1959 const struct lu_fid *first_fid,
1960 struct local_oid_storage **los);
1961 void local_oid_storage_fini(const struct lu_env *env,
1962 struct local_oid_storage *los);
1963 int local_object_fid_generate(const struct lu_env *env,
1964 struct local_oid_storage *los,
1965 struct lu_fid *fid);
1966 int local_object_declare_create(const struct lu_env *env,
1967 struct local_oid_storage *los,
1968 struct dt_object *o,
1969 struct lu_attr *attr,
1970 struct dt_object_format *dof,
1971 struct thandle *th);
1972 int local_object_create(const struct lu_env *env,
1973 struct local_oid_storage *los,
1974 struct dt_object *o,
1975 struct lu_attr *attr, struct dt_object_format *dof,
1976 struct thandle *th);
1977 struct dt_object *local_file_find_or_create(const struct lu_env *env,
1978 struct local_oid_storage *los,
1979 struct dt_object *parent,
1980 const char *name, __u32 mode);
1981 struct dt_object *local_file_find_or_create_with_fid(const struct lu_env *env,
1982 struct dt_device *dt,
1983 const struct lu_fid *fid,
1984 struct dt_object *parent,
1988 local_index_find_or_create(const struct lu_env *env,
1989 struct local_oid_storage *los,
1990 struct dt_object *parent,
1991 const char *name, __u32 mode,
1992 const struct dt_index_features *ft);
1994 local_index_find_or_create_with_fid(const struct lu_env *env,
1995 struct dt_device *dt,
1996 const struct lu_fid *fid,
1997 struct dt_object *parent,
1998 const char *name, __u32 mode,
1999 const struct dt_index_features *ft);
2000 int local_object_unlink(const struct lu_env *env, struct dt_device *dt,
2001 struct dt_object *parent, const char *name);
2003 static inline int dt_object_lock(const struct lu_env *env,
2004 struct dt_object *o, struct lustre_handle *lh,
2005 struct ldlm_enqueue_info *einfo,
2006 union ldlm_policy_data *policy)
2009 LASSERT(o->do_ops != NULL);
2010 LASSERT(o->do_ops->do_object_lock != NULL);
2011 return o->do_ops->do_object_lock(env, o, lh, einfo, policy);
2014 static inline int dt_object_unlock(const struct lu_env *env,
2015 struct dt_object *o,
2016 struct ldlm_enqueue_info *einfo,
2017 union ldlm_policy_data *policy)
2020 LASSERT(o->do_ops != NULL);
2021 LASSERT(o->do_ops->do_object_unlock != NULL);
2022 return o->do_ops->do_object_unlock(env, o, einfo, policy);
2025 int dt_lookup_dir(const struct lu_env *env, struct dt_object *dir,
2026 const char *name, struct lu_fid *fid);
2028 static inline int dt_object_sync(const struct lu_env *env, struct dt_object *o,
2029 __u64 start, __u64 end)
2033 LASSERT(o->do_ops->do_object_sync);
2034 return o->do_ops->do_object_sync(env, o, start, end);
2037 int dt_declare_version_set(const struct lu_env *env, struct dt_object *o,
2038 struct thandle *th);
2039 void dt_version_set(const struct lu_env *env, struct dt_object *o,
2040 dt_obj_version_t version, struct thandle *th);
2041 dt_obj_version_t dt_version_get(const struct lu_env *env, struct dt_object *o);
2044 int dt_read(const struct lu_env *env, struct dt_object *dt,
2045 struct lu_buf *buf, loff_t *pos);
2046 int dt_record_read(const struct lu_env *env, struct dt_object *dt,
2047 struct lu_buf *buf, loff_t *pos);
2048 int dt_record_write(const struct lu_env *env, struct dt_object *dt,
2049 const struct lu_buf *buf, loff_t *pos, struct thandle *th);
2050 typedef int (*dt_index_page_build_t)(const struct lu_env *env,
2051 union lu_page *lp, size_t nob,
2052 const struct dt_it_ops *iops,
2053 struct dt_it *it, __u32 attr, void *arg);
2054 int dt_index_walk(const struct lu_env *env, struct dt_object *obj,
2055 const struct lu_rdpg *rdpg, dt_index_page_build_t filler,
2057 int dt_index_read(const struct lu_env *env, struct dt_device *dev,
2058 struct idx_info *ii, const struct lu_rdpg *rdpg);
2060 static inline struct thandle *dt_trans_create(const struct lu_env *env,
2061 struct dt_device *d)
2063 LASSERT(d->dd_ops->dt_trans_create);
2064 return d->dd_ops->dt_trans_create(env, d);
2067 static inline int dt_trans_start(const struct lu_env *env,
2068 struct dt_device *d, struct thandle *th)
2070 LASSERT(d->dd_ops->dt_trans_start);
2071 return d->dd_ops->dt_trans_start(env, d, th);
2074 /* for this transaction hooks shouldn't be called */
2075 static inline int dt_trans_start_local(const struct lu_env *env,
2076 struct dt_device *d, struct thandle *th)
2078 LASSERT(d->dd_ops->dt_trans_start);
2080 return d->dd_ops->dt_trans_start(env, d, th);
2083 static inline int dt_trans_stop(const struct lu_env *env,
2084 struct dt_device *d, struct thandle *th)
2086 LASSERT(d->dd_ops->dt_trans_stop);
2087 return d->dd_ops->dt_trans_stop(env, d, th);
2090 static inline int dt_trans_cb_add(struct thandle *th,
2091 struct dt_txn_commit_cb *dcb)
2093 LASSERT(th->th_dev->dd_ops->dt_trans_cb_add);
2094 dcb->dcb_magic = TRANS_COMMIT_CB_MAGIC;
2095 return th->th_dev->dd_ops->dt_trans_cb_add(th, dcb);
2100 static inline int dt_declare_record_write(const struct lu_env *env,
2101 struct dt_object *dt,
2102 const struct lu_buf *buf,
2108 LASSERTF(dt != NULL, "dt is NULL when we want to write record\n");
2109 LASSERT(th != NULL);
2110 LASSERT(dt->do_body_ops);
2111 LASSERT(dt->do_body_ops->dbo_declare_write);
2112 rc = dt->do_body_ops->dbo_declare_write(env, dt, buf, pos, th);
2116 static inline int dt_declare_create(const struct lu_env *env,
2117 struct dt_object *dt,
2118 struct lu_attr *attr,
2119 struct dt_allocation_hint *hint,
2120 struct dt_object_format *dof,
2124 LASSERT(dt->do_ops);
2125 LASSERT(dt->do_ops->do_declare_create);
2127 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_CREATE))
2128 return cfs_fail_err;
2130 return dt->do_ops->do_declare_create(env, dt, attr, hint, dof, th);
2133 static inline int dt_create(const struct lu_env *env,
2134 struct dt_object *dt,
2135 struct lu_attr *attr,
2136 struct dt_allocation_hint *hint,
2137 struct dt_object_format *dof,
2141 LASSERT(dt->do_ops);
2142 LASSERT(dt->do_ops->do_create);
2144 if (CFS_FAULT_CHECK(OBD_FAIL_DT_CREATE))
2145 return cfs_fail_err;
2147 return dt->do_ops->do_create(env, dt, attr, hint, dof, th);
2150 static inline int dt_declare_destroy(const struct lu_env *env,
2151 struct dt_object *dt,
2155 LASSERT(dt->do_ops);
2156 LASSERT(dt->do_ops->do_declare_destroy);
2158 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_DESTROY))
2159 return cfs_fail_err;
2161 return dt->do_ops->do_declare_destroy(env, dt, th);
2164 static inline int dt_destroy(const struct lu_env *env,
2165 struct dt_object *dt,
2169 LASSERT(dt->do_ops);
2170 LASSERT(dt->do_ops->do_destroy);
2172 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DESTROY))
2173 return cfs_fail_err;
2175 return dt->do_ops->do_destroy(env, dt, th);
2178 static inline void dt_read_lock(const struct lu_env *env,
2179 struct dt_object *dt,
2183 LASSERT(dt->do_ops);
2184 LASSERT(dt->do_ops->do_read_lock);
2185 dt->do_ops->do_read_lock(env, dt, role);
2188 static inline void dt_write_lock(const struct lu_env *env,
2189 struct dt_object *dt,
2193 LASSERT(dt->do_ops);
2194 LASSERT(dt->do_ops->do_write_lock);
2195 dt->do_ops->do_write_lock(env, dt, role);
2198 static inline void dt_read_unlock(const struct lu_env *env,
2199 struct dt_object *dt)
2202 LASSERT(dt->do_ops);
2203 LASSERT(dt->do_ops->do_read_unlock);
2204 dt->do_ops->do_read_unlock(env, dt);
2207 static inline void dt_write_unlock(const struct lu_env *env,
2208 struct dt_object *dt)
2211 LASSERT(dt->do_ops);
2212 LASSERT(dt->do_ops->do_write_unlock);
2213 dt->do_ops->do_write_unlock(env, dt);
2216 static inline int dt_write_locked(const struct lu_env *env,
2217 struct dt_object *dt)
2220 LASSERT(dt->do_ops);
2221 LASSERT(dt->do_ops->do_write_locked);
2222 return dt->do_ops->do_write_locked(env, dt);
2225 static inline int dt_declare_attr_get(const struct lu_env *env,
2226 struct dt_object *dt)
2229 LASSERT(dt->do_ops);
2230 LASSERT(dt->do_ops->do_declare_attr_get);
2232 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_ATTR_GET))
2233 return cfs_fail_err;
2235 return dt->do_ops->do_declare_attr_get(env, dt);
2238 static inline int dt_attr_get(const struct lu_env *env, struct dt_object *dt,
2242 LASSERT(dt->do_ops);
2243 LASSERT(dt->do_ops->do_attr_get);
2245 if (CFS_FAULT_CHECK(OBD_FAIL_DT_ATTR_GET))
2246 return cfs_fail_err;
2248 return dt->do_ops->do_attr_get(env, dt, la);
2251 static inline int dt_declare_attr_set(const struct lu_env *env,
2252 struct dt_object *dt,
2253 const struct lu_attr *la,
2257 LASSERT(dt->do_ops);
2258 LASSERT(dt->do_ops->do_declare_attr_set);
2260 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_ATTR_SET))
2261 return cfs_fail_err;
2263 return dt->do_ops->do_declare_attr_set(env, dt, la, th);
2266 static inline int dt_attr_set(const struct lu_env *env, struct dt_object *dt,
2267 const struct lu_attr *la, struct thandle *th)
2270 LASSERT(dt->do_ops);
2271 LASSERT(dt->do_ops->do_attr_set);
2273 if (CFS_FAULT_CHECK(OBD_FAIL_DT_ATTR_SET))
2274 return cfs_fail_err;
2276 return dt->do_ops->do_attr_set(env, dt, la, th);
2279 static inline int dt_declare_ref_add(const struct lu_env *env,
2280 struct dt_object *dt, struct thandle *th)
2283 LASSERT(dt->do_ops);
2284 LASSERT(dt->do_ops->do_declare_ref_add);
2286 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_REF_ADD))
2287 return cfs_fail_err;
2289 return dt->do_ops->do_declare_ref_add(env, dt, th);
2292 static inline int dt_ref_add(const struct lu_env *env,
2293 struct dt_object *dt, struct thandle *th)
2296 LASSERT(dt->do_ops);
2297 LASSERT(dt->do_ops->do_ref_add);
2299 if (CFS_FAULT_CHECK(OBD_FAIL_DT_REF_ADD))
2300 return cfs_fail_err;
2302 return dt->do_ops->do_ref_add(env, dt, th);
2305 static inline int dt_declare_ref_del(const struct lu_env *env,
2306 struct dt_object *dt, struct thandle *th)
2309 LASSERT(dt->do_ops);
2310 LASSERT(dt->do_ops->do_declare_ref_del);
2312 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_REF_DEL))
2313 return cfs_fail_err;
2315 return dt->do_ops->do_declare_ref_del(env, dt, th);
2318 static inline int dt_ref_del(const struct lu_env *env,
2319 struct dt_object *dt, struct thandle *th)
2322 LASSERT(dt->do_ops);
2323 LASSERT(dt->do_ops->do_ref_del);
2325 if (CFS_FAULT_CHECK(OBD_FAIL_DT_REF_DEL))
2326 return cfs_fail_err;
2328 return dt->do_ops->do_ref_del(env, dt, th);
2331 static inline int dt_bufs_get(const struct lu_env *env, struct dt_object *d,
2332 struct niobuf_remote *rnb,
2333 struct niobuf_local *lnb, int rw)
2336 LASSERT(d->do_body_ops);
2337 LASSERT(d->do_body_ops->dbo_bufs_get);
2338 return d->do_body_ops->dbo_bufs_get(env, d, rnb->rnb_offset,
2339 rnb->rnb_len, lnb, rw);
2342 static inline int dt_bufs_put(const struct lu_env *env, struct dt_object *d,
2343 struct niobuf_local *lnb, int n)
2346 LASSERT(d->do_body_ops);
2347 LASSERT(d->do_body_ops->dbo_bufs_put);
2348 return d->do_body_ops->dbo_bufs_put(env, d, lnb, n);
2351 static inline int dt_write_prep(const struct lu_env *env, struct dt_object *d,
2352 struct niobuf_local *lnb, int n)
2355 LASSERT(d->do_body_ops);
2356 LASSERT(d->do_body_ops->dbo_write_prep);
2357 return d->do_body_ops->dbo_write_prep(env, d, lnb, n);
2360 static inline int dt_declare_write_commit(const struct lu_env *env,
2361 struct dt_object *d,
2362 struct niobuf_local *lnb,
2363 int n, struct thandle *th)
2365 LASSERTF(d != NULL, "dt is NULL when we want to declare write\n");
2366 LASSERT(th != NULL);
2367 return d->do_body_ops->dbo_declare_write_commit(env, d, lnb, n, th);
2371 static inline int dt_write_commit(const struct lu_env *env,
2372 struct dt_object *d, struct niobuf_local *lnb,
2373 int n, struct thandle *th)
2376 LASSERT(d->do_body_ops);
2377 LASSERT(d->do_body_ops->dbo_write_commit);
2378 return d->do_body_ops->dbo_write_commit(env, d, lnb, n, th);
2381 static inline int dt_read_prep(const struct lu_env *env, struct dt_object *d,
2382 struct niobuf_local *lnb, int n)
2385 LASSERT(d->do_body_ops);
2386 LASSERT(d->do_body_ops->dbo_read_prep);
2387 return d->do_body_ops->dbo_read_prep(env, d, lnb, n);
2390 static inline int dt_declare_write(const struct lu_env *env,
2391 struct dt_object *dt,
2392 const struct lu_buf *buf, loff_t pos,
2396 LASSERT(dt->do_body_ops);
2397 LASSERT(dt->do_body_ops->dbo_declare_write);
2398 return dt->do_body_ops->dbo_declare_write(env, dt, buf, pos, th);
2401 static inline ssize_t dt_write(const struct lu_env *env, struct dt_object *dt,
2402 const struct lu_buf *buf, loff_t *pos,
2403 struct thandle *th, int rq)
2406 LASSERT(dt->do_body_ops);
2407 LASSERT(dt->do_body_ops->dbo_write);
2408 return dt->do_body_ops->dbo_write(env, dt, buf, pos, th, rq);
2411 static inline int dt_declare_punch(const struct lu_env *env,
2412 struct dt_object *dt, __u64 start,
2413 __u64 end, struct thandle *th)
2416 LASSERT(dt->do_body_ops);
2417 LASSERT(dt->do_body_ops->dbo_declare_punch);
2418 return dt->do_body_ops->dbo_declare_punch(env, dt, start, end, th);
2421 static inline int dt_punch(const struct lu_env *env, struct dt_object *dt,
2422 __u64 start, __u64 end, struct thandle *th)
2425 LASSERT(dt->do_body_ops);
2426 LASSERT(dt->do_body_ops->dbo_punch);
2427 return dt->do_body_ops->dbo_punch(env, dt, start, end, th);
2430 static inline int dt_ladvise(const struct lu_env *env, struct dt_object *dt,
2431 __u64 start, __u64 end, int advice)
2434 LASSERT(dt->do_body_ops);
2435 LASSERT(dt->do_body_ops->dbo_ladvise);
2436 return dt->do_body_ops->dbo_ladvise(env, dt, start, end, advice);
2439 static inline int dt_fiemap_get(const struct lu_env *env, struct dt_object *d,
2443 if (d->do_body_ops == NULL)
2445 if (d->do_body_ops->dbo_fiemap_get == NULL)
2447 return d->do_body_ops->dbo_fiemap_get(env, d, fm);
2450 static inline int dt_statfs(const struct lu_env *env, struct dt_device *dev,
2451 struct obd_statfs *osfs)
2454 LASSERT(dev->dd_ops);
2455 LASSERT(dev->dd_ops->dt_statfs);
2456 return dev->dd_ops->dt_statfs(env, dev, osfs);
2459 static inline int dt_root_get(const struct lu_env *env, struct dt_device *dev,
2463 LASSERT(dev->dd_ops);
2464 LASSERT(dev->dd_ops->dt_root_get);
2465 return dev->dd_ops->dt_root_get(env, dev, f);
2468 static inline void dt_conf_get(const struct lu_env *env,
2469 const struct dt_device *dev,
2470 struct dt_device_param *param)
2473 LASSERT(dev->dd_ops);
2474 LASSERT(dev->dd_ops->dt_conf_get);
2475 return dev->dd_ops->dt_conf_get(env, dev, param);
2478 static inline int dt_sync(const struct lu_env *env, struct dt_device *dev)
2481 LASSERT(dev->dd_ops);
2482 LASSERT(dev->dd_ops->dt_sync);
2483 return dev->dd_ops->dt_sync(env, dev);
2486 static inline int dt_ro(const struct lu_env *env, struct dt_device *dev)
2489 LASSERT(dev->dd_ops);
2490 LASSERT(dev->dd_ops->dt_ro);
2491 return dev->dd_ops->dt_ro(env, dev);
2494 static inline int dt_declare_insert(const struct lu_env *env,
2495 struct dt_object *dt,
2496 const struct dt_rec *rec,
2497 const struct dt_key *key,
2501 LASSERT(dt->do_index_ops);
2502 LASSERT(dt->do_index_ops->dio_declare_insert);
2504 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_INSERT))
2505 return cfs_fail_err;
2507 return dt->do_index_ops->dio_declare_insert(env, dt, rec, key, th);
2510 static inline int dt_insert(const struct lu_env *env,
2511 struct dt_object *dt,
2512 const struct dt_rec *rec,
2513 const struct dt_key *key,
2518 LASSERT(dt->do_index_ops);
2519 LASSERT(dt->do_index_ops->dio_insert);
2521 if (CFS_FAULT_CHECK(OBD_FAIL_DT_INSERT))
2522 return cfs_fail_err;
2524 return dt->do_index_ops->dio_insert(env, dt, rec, key, th, noquota);
2527 static inline int dt_declare_xattr_del(const struct lu_env *env,
2528 struct dt_object *dt,
2533 LASSERT(dt->do_ops);
2534 LASSERT(dt->do_ops->do_declare_xattr_del);
2536 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_XATTR_DEL))
2537 return cfs_fail_err;
2539 return dt->do_ops->do_declare_xattr_del(env, dt, name, th);
2542 static inline int dt_xattr_del(const struct lu_env *env,
2543 struct dt_object *dt, const char *name,
2547 LASSERT(dt->do_ops);
2548 LASSERT(dt->do_ops->do_xattr_del);
2550 if (CFS_FAULT_CHECK(OBD_FAIL_DT_XATTR_DEL))
2551 return cfs_fail_err;
2553 return dt->do_ops->do_xattr_del(env, dt, name, th);
2556 static inline int dt_declare_xattr_set(const struct lu_env *env,
2557 struct dt_object *dt,
2558 const struct lu_buf *buf,
2559 const char *name, int fl,
2563 LASSERT(dt->do_ops);
2564 LASSERT(dt->do_ops->do_declare_xattr_set);
2566 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_XATTR_SET))
2567 return cfs_fail_err;
2569 return dt->do_ops->do_declare_xattr_set(env, dt, buf, name, fl, th);
2572 static inline int dt_xattr_set(const struct lu_env *env,
2573 struct dt_object *dt, const struct lu_buf *buf,
2574 const char *name, int fl, struct thandle *th)
2577 LASSERT(dt->do_ops);
2578 LASSERT(dt->do_ops->do_xattr_set);
2580 if (CFS_FAULT_CHECK(OBD_FAIL_DT_XATTR_SET))
2581 return cfs_fail_err;
2583 return dt->do_ops->do_xattr_set(env, dt, buf, name, fl, th);
2586 static inline int dt_declare_xattr_get(const struct lu_env *env,
2587 struct dt_object *dt,
2592 LASSERT(dt->do_ops);
2593 LASSERT(dt->do_ops->do_declare_xattr_get);
2595 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_XATTR_GET))
2596 return cfs_fail_err;
2598 return dt->do_ops->do_declare_xattr_get(env, dt, buf, name);
2601 static inline int dt_xattr_get(const struct lu_env *env,
2602 struct dt_object *dt, struct lu_buf *buf,
2606 LASSERT(dt->do_ops);
2607 LASSERT(dt->do_ops->do_xattr_get);
2609 if (CFS_FAULT_CHECK(OBD_FAIL_DT_XATTR_GET))
2610 return cfs_fail_err;
2612 return dt->do_ops->do_xattr_get(env, dt, buf, name);
2615 static inline int dt_xattr_list(const struct lu_env *env, struct dt_object *dt,
2616 const struct lu_buf *buf)
2619 LASSERT(dt->do_ops);
2620 LASSERT(dt->do_ops->do_xattr_list);
2622 if (CFS_FAULT_CHECK(OBD_FAIL_DT_XATTR_LIST))
2623 return cfs_fail_err;
2625 return dt->do_ops->do_xattr_list(env, dt, buf);
2628 static inline int dt_invalidate(const struct lu_env *env, struct dt_object *dt)
2631 LASSERT(dt->do_ops);
2632 LASSERT(dt->do_ops->do_invalidate);
2634 return dt->do_ops->do_invalidate(env, dt);
2637 static inline int dt_declare_delete(const struct lu_env *env,
2638 struct dt_object *dt,
2639 const struct dt_key *key,
2643 LASSERT(dt->do_index_ops);
2644 LASSERT(dt->do_index_ops->dio_declare_delete);
2646 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DECLARE_DELETE))
2647 return cfs_fail_err;
2649 return dt->do_index_ops->dio_declare_delete(env, dt, key, th);
2652 static inline int dt_delete(const struct lu_env *env,
2653 struct dt_object *dt,
2654 const struct dt_key *key,
2658 LASSERT(dt->do_index_ops);
2659 LASSERT(dt->do_index_ops->dio_delete);
2661 if (CFS_FAULT_CHECK(OBD_FAIL_DT_DELETE))
2662 return cfs_fail_err;
2664 return dt->do_index_ops->dio_delete(env, dt, key, th);
2667 static inline int dt_commit_async(const struct lu_env *env,
2668 struct dt_device *dev)
2671 LASSERT(dev->dd_ops);
2672 LASSERT(dev->dd_ops->dt_commit_async);
2673 return dev->dd_ops->dt_commit_async(env, dev);
2676 static inline int dt_lookup(const struct lu_env *env,
2677 struct dt_object *dt,
2679 const struct dt_key *key)
2684 LASSERT(dt->do_index_ops);
2685 LASSERT(dt->do_index_ops->dio_lookup);
2687 if (CFS_FAULT_CHECK(OBD_FAIL_DT_LOOKUP))
2688 return cfs_fail_err;
2690 ret = dt->do_index_ops->dio_lookup(env, dt, rec, key);
2698 struct dt_find_hint {
2699 struct lu_fid *dfh_fid;
2700 struct dt_device *dfh_dt;
2701 struct dt_object *dfh_o;
2704 struct dt_insert_rec {
2706 const struct lu_fid *rec_fid;
2718 struct dt_thread_info {
2719 char dti_buf[DT_MAX_PATH];
2720 struct dt_find_hint dti_dfh;
2721 struct lu_attr dti_attr;
2722 struct lu_fid dti_fid;
2723 struct dt_object_format dti_dof;
2724 struct lustre_mdt_attrs dti_lma;
2725 struct lu_buf dti_lb;
2726 struct lu_object_conf dti_conf;
2728 struct dt_insert_rec dti_dt_rec;
2731 extern struct lu_context_key dt_key;
2733 static inline struct dt_thread_info *dt_info(const struct lu_env *env)
2735 struct dt_thread_info *dti;
2737 dti = lu_context_key_get(&env->le_ctx, &dt_key);
2742 int dt_global_init(void);
2743 void dt_global_fini(void);
2745 # ifdef CONFIG_PROC_FS
2746 int lprocfs_dt_blksize_seq_show(struct seq_file *m, void *v);
2747 int lprocfs_dt_kbytestotal_seq_show(struct seq_file *m, void *v);
2748 int lprocfs_dt_kbytesfree_seq_show(struct seq_file *m, void *v);
2749 int lprocfs_dt_kbytesavail_seq_show(struct seq_file *m, void *v);
2750 int lprocfs_dt_filestotal_seq_show(struct seq_file *m, void *v);
2751 int lprocfs_dt_filesfree_seq_show(struct seq_file *m, void *v);
2752 # endif /* CONFIG_PROC_FS */
2754 #endif /* __LUSTRE_DT_OBJECT_H */