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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_internal.h
38 * Shared definitions and declarations for osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
43 #ifndef _OSD_INTERNAL_H
44 #define _OSD_INTERNAL_H
46 #if defined(__KERNEL__)
48 /* struct rw_semaphore */
49 #include <linux/rwsem.h>
51 #include <linux/dcache.h>
53 #include <linux/dirent.h>
55 #include <ldiskfs/ldiskfs_config.h>
56 #include <ldiskfs/ldiskfs.h>
57 #include <ldiskfs/ldiskfs_jbd2.h>
59 /* fsfilt_{get|put}_ops */
60 #include <lustre_fsfilt.h>
64 /* class_register_type(), class_unregister_type(), class_get_type() */
65 #include <obd_class.h>
66 #include <lustre_disk.h>
67 #include <dt_object.h>
68 #include <lustre_quota.h>
72 #include "osd_scrub.h"
73 #include "osd_quota_fmt.h"
77 #define OSD_COUNTERS (0)
79 /* ldiskfs special inode::i_state_flags need to be accessed with
80 * ldiskfs_{set,clear,test}_inode_state() only */
82 /* OI scrub should skip this inode. */
83 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
85 /* Do not add OI mapping for this inode. */
86 #define LDISKFS_STATE_LUSTRE_NO_OI 30
88 /** Enable thandle usage statistics */
89 #define OSD_THANDLE_STATS (0)
91 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
93 #define OBJECTS "OBJECTS"
94 #define ADMIN_USR "admin_quotafile_v2.usr"
95 #define ADMIN_GRP "admin_quotafile_v2.grp"
97 struct osd_directory {
98 struct iam_container od_container;
99 struct iam_descr od_descr;
103 * Object Index (oi) instance.
107 * underlying index object, where fid->id mapping in stored.
109 struct inode *oi_inode;
110 struct osd_directory oi_dir;
113 extern const int osd_dto_credits_noquota[];
116 struct dt_object oo_dt;
118 * Inode for file system object represented by this osd_object. This
119 * inode is pinned for the whole duration of lu_object life.
121 * Not modified concurrently (either setup early during object
122 * creation, or assigned by osd_object_create() under write lock).
124 struct inode *oo_inode;
126 * to protect index ops.
128 struct htree_lock_head *oo_hl_head;
129 struct rw_semaphore oo_ext_idx_sem;
130 struct rw_semaphore oo_sem;
131 struct osd_directory *oo_dir;
132 /** protects inode attributes. */
135 * Following two members are used to indicate the presence of dot and
136 * dotdot in the given directory. This is required for interop mode
139 int oo_compat_dot_created;
140 int oo_compat_dotdot_created;
142 const struct lu_env *oo_owner;
143 #ifdef CONFIG_LOCKDEP
144 struct lockdep_map oo_dep_map;
149 /* protects on-fly initialization */
150 int oos_subdir_count; /* subdir count for each seq */
151 struct dentry *oos_root; /* O/<seq> */
152 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
153 obd_seq oos_seq; /* seq number */
154 cfs_list_t oos_seq_list; /* list to seq_list */
158 struct dentry *om_root; /* dentry for /O */
159 rwlock_t om_seq_list_lock; /* lock for seq_list */
160 cfs_list_t om_seq_list; /* list head for seq */
162 struct semaphore om_dir_init_sem;
166 struct dentry *om_root; /* AGENT/<index> */
167 obd_seq om_index; /* mdt index */
168 cfs_list_t om_list; /* list to omm_list */
171 struct osd_mdobj_map {
172 struct dentry *omm_remote_parent;
175 #define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
176 ll_ldiskfs_find_entry(dir, dentry, de, lock)
177 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
178 ldiskfs_add_entry(handle, child, cinode, hlock)
180 #define OSD_OTABLE_IT_CACHE_SIZE 64
181 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
183 struct osd_inconsistent_item {
184 /* link into osd_scrub::os_inconsistent_items,
185 * protected by osd_scrub::os_lock. */
188 /* The right FID <=> ino#/gen mapping. */
189 struct osd_idmap_cache oii_cache;
191 unsigned int oii_insert:1; /* insert or update mapping. */
194 struct osd_otable_cache {
195 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
197 /* Index for next cache slot to be filled. */
198 int ooc_producer_idx;
200 /* Index for next cache slot to be returned by it::next(). */
201 int ooc_consumer_idx;
203 /* How many items in ooc_cache. */
204 int ooc_cached_items;
206 /* Position for up layer LFSCK iteration pre-loading. */
207 __u32 ooc_pos_preload;
210 struct osd_otable_it {
211 struct osd_device *ooi_dev;
212 struct osd_otable_cache ooi_cache;
214 /* The following bits can be updated/checked w/o lock protection.
215 * If more bits will be introduced in the future and need lock to
216 * protect, please add comment. */
217 unsigned long ooi_used_outside:1, /* Some user out of OSD
218 * uses the iteration. */
219 ooi_all_cached:1, /* No more entries can be
220 * filled into cache. */
221 ooi_user_ready:1, /* The user out of OSD is
222 * ready to iterate. */
223 ooi_waiting:1; /* it::next is waiting. */
231 struct dt_device od_dt_dev;
232 /* information about underlying file system */
233 struct vfsmount *od_mnt;
235 struct osd_oi **od_oi_table;
236 /* total number of OI containers */
241 unsigned int od_fl_capa:1,
248 unsigned long od_capa_timeout;
250 struct lustre_capa_key *od_capa_keys;
251 cfs_hlist_head_t *od_capa_hash;
253 cfs_proc_dir_entry_t *od_proc_entry;
254 struct lprocfs_stats *od_stats;
256 * statfs optimization: we cache a bit.
258 cfs_time_t od_osfs_age;
259 struct obd_statfs od_statfs;
260 spinlock_t od_osfs_lock;
262 struct fsfilt_operations *od_fsops;
264 struct lu_site od_site;
266 struct osd_obj_map *od_ost_map;
267 struct osd_mdobj_map *od_mdt_map;
269 unsigned long long od_readcache_max_filesize;
271 int od_writethrough_cache;
273 struct brw_stats od_brw_stats;
274 cfs_atomic_t od_r_in_flight;
275 cfs_atomic_t od_w_in_flight;
277 struct mutex od_otable_mutex;
278 struct osd_otable_it *od_otable_it;
279 struct osd_scrub od_scrub;
280 cfs_list_t od_ios_list;
282 /* service name associated with the osd device */
283 char od_svname[MAX_OBD_NAME];
284 char od_mntdev[MAX_OBD_NAME];
286 /* quota slave instance */
287 struct qsd_instance *od_quota_slave;
290 /* There are at most 10 uid/gids are affected in a transaction, and
291 * that's rename case:
292 * - 2 for source parent uid & gid;
293 * - 2 for source child uid & gid ('..' entry update when child is directory);
294 * - 2 for target parent uid & gid;
295 * - 2 for target child uid & gid (if the target child exists);
296 * - 2 for root uid & gid (last_rcvd, llog, etc);
298 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
299 * the id type of each id in the ot_id_array.
301 #define OSD_MAX_UGID_CNT 10
306 OSD_OT_XATTR_SET = 2,
320 struct thandle ot_super;
322 struct ldiskfs_journal_cb_entry ot_jcb;
323 cfs_list_t ot_dcb_list;
324 /* Link to the device, for debugging. */
325 struct lu_ref_link ot_dev_link;
326 unsigned short ot_credits;
327 unsigned short ot_id_cnt;
328 unsigned short ot_id_type;
329 uid_t ot_id_array[OSD_MAX_UGID_CNT];
330 struct lquota_trans *ot_quota_trans;
331 #if OSD_THANDLE_STATS
332 /** time when this handle was allocated */
333 cfs_time_t oth_alloced;
335 /** time when this thanle was started */
336 cfs_time_t oth_started;
341 * Basic transaction credit op
351 DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
365 LPROC_OSD_READ_BYTES = 0,
366 LPROC_OSD_WRITE_BYTES = 1,
367 LPROC_OSD_GET_PAGE = 2,
368 LPROC_OSD_NO_PAGE = 3,
369 LPROC_OSD_CACHE_ACCESS = 4,
370 LPROC_OSD_CACHE_HIT = 5,
371 LPROC_OSD_CACHE_MISS = 6,
373 #if OSD_THANDLE_STATS
374 LPROC_OSD_THANDLE_STARTING,
375 LPROC_OSD_THANDLE_OPEN,
376 LPROC_OSD_THANDLE_CLOSING,
383 * Storage representation for fids.
385 * Variable size, first byte contains the length of the whole record.
387 struct osd_fid_pack {
388 unsigned char fp_len;
389 char fp_area[sizeof(struct lu_fid)];
392 struct osd_it_ea_dirent {
393 struct lu_fid oied_fid;
396 unsigned short oied_namelen;
397 unsigned int oied_type;
399 } __attribute__((packed));
402 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
403 * than lu_dirent struct. osd readdir reads less number of dirent than
404 * required for mdd dir page. so buffer size need to be increased so that
405 * there would be one ext3 readdir for every mdd readdir page.
408 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
411 * This is iterator's in-memory data structure in interoperability
412 * mode (i.e. iterator over ldiskfs style directory)
415 struct osd_object *oie_obj;
416 /** used in ldiskfs iterator, to stored file pointer */
417 struct file oie_file;
418 /** how many entries have been read-cached from storage */
420 /** current entry is being iterated by caller */
422 /** current processing entry */
423 struct osd_it_ea_dirent *oie_dirent;
424 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
429 * Iterator's in-memory data structure for IAM mode.
432 struct osd_object *oi_obj;
433 struct iam_path_descr *oi_ipd;
434 struct iam_iterator oi_it;
437 struct osd_quota_leaf {
443 * Iterator's in-memory data structure for quota file.
445 struct osd_it_quota {
446 struct osd_object *oiq_obj;
447 /** tree blocks path to where the entry is stored */
448 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
449 /** on-disk offset for current key where quota record can be found */
451 /** identifier for current quota record */
453 /** the record index in the leaf/index block */
454 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
455 /** list of already processed leaf blocks */
459 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
463 cfs_atomic_t dr_numreqs; /* number of reqs being processed */
468 unsigned int dr_ignore_quota:1;
469 unsigned int dr_elapsed_valid:1; /* we really did count time */
470 unsigned int dr_rw:1;
471 struct lu_buf dr_pg_buf;
472 struct page **dr_pages;
473 struct lu_buf dr_bl_buf;
474 unsigned long *dr_blocks;
475 unsigned long dr_start_time;
476 unsigned long dr_elapsed; /* how long io took */
477 struct osd_device *dr_dev;
478 unsigned int dr_init_at; /* the line iobuf was initialized */
481 struct osd_thread_info {
482 const struct lu_env *oti_env;
484 * used for index operations.
486 struct dentry oti_obj_dentry;
487 struct dentry oti_child_dentry;
489 /** dentry for Iterator context. */
490 struct dentry oti_it_dentry;
491 struct htree_lock *oti_hlock;
493 struct lu_fid oti_fid;
494 struct lu_fid oti_fid2;
495 struct lu_fid oti_fid3;
496 struct osd_inode_id oti_id;
497 struct osd_inode_id oti_id2;
498 struct osd_inode_id oti_id3;
499 struct ost_id oti_ostid;
502 * XXX temporary: for ->i_op calls.
504 struct timespec oti_time;
506 * XXX temporary: fake struct file for osd_object_sync
508 struct file oti_file;
510 * XXX temporary: for capa operations.
512 struct lustre_capa_key oti_capa_key;
513 struct lustre_capa oti_capa;
515 /** osd_device reference, initialized in osd_trans_start() and
516 used in osd_trans_stop() */
517 struct osd_device *oti_dev;
520 * following ipd and it structures are used for osd_index_iam_lookup()
521 * these are defined separately as we might do index operation
522 * in open iterator session.
525 /** osd iterator context used for iterator session */
528 struct osd_it_iam oti_it;
529 /* ldiskfs iterator data structure,
530 * see osd_it_ea_{init, fini} */
531 struct osd_it_ea oti_it_ea;
532 struct osd_it_quota oti_it_quota;
535 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
538 struct kstatfs oti_ksfs;
540 /** IAM iterator for index operation. */
541 struct iam_iterator oti_idx_it;
543 /** union to guarantee that ->oti_ipd[] has proper alignment. */
545 char oti_it_ipd[DX_IPD_MAX_SIZE];
546 long long oti_alignment_lieutenant;
550 char oti_idx_ipd[DX_IPD_MAX_SIZE];
551 long long oti_alignment_lieutenant_colonel;
554 struct osd_idmap_cache oti_cache;
559 /** used in osd_fid_set() to put xattr */
560 struct lu_buf oti_buf;
561 /** used in osd_ea_fid_set() to set fid into common ea */
563 struct lustre_mdt_attrs oti_mdt_attrs;
564 /* old LMA for compatibility */
565 char oti_mdt_attrs_old[LMA_OLD_SIZE];
568 struct osd_iobuf oti_iobuf;
569 struct inode oti_inode;
570 #define OSD_FID_REC_SZ 32
571 char oti_ldp[OSD_FID_REC_SZ];
572 char oti_ldp2[OSD_FID_REC_SZ];
574 /* used by quota code */
576 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
577 struct fs_disk_quota oti_fdq;
579 struct if_dqblk oti_dqblk;
581 struct if_dqinfo oti_dqinfo;
583 struct lquota_id_info oti_qi;
584 struct lquota_trans oti_quota_trans;
585 union lquota_rec oti_quota_rec;
587 struct lu_seq_range oti_seq_range;
589 /* Tracking for transaction credits, to allow debugging and optimizing
590 * cases where a large number of credits are being allocated for
591 * single transaction. */
592 unsigned short oti_declare_ops[OSD_OT_MAX];
593 unsigned short oti_declare_ops_rb[OSD_OT_MAX];
594 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
599 struct filter_fid_old oti_ff;
600 struct filter_fid oti_ff_new;
604 extern int ldiskfs_pdo;
606 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
607 const char *name, void *buf, int len)
609 dentry->d_inode = inode;
610 dentry->d_sb = inode->i_sb;
611 return inode->i_op->getxattr(dentry, name, buf, len);
614 static inline int __osd_xattr_set(struct osd_thread_info *info,
615 struct inode *inode, const char *name,
616 const void *buf, int buflen, int fl)
618 struct dentry *dentry = &info->oti_child_dentry;
620 ll_vfs_dq_init(inode);
621 dentry->d_inode = inode;
622 dentry->d_sb = inode->i_sb;
623 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
628 void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars);
629 int osd_procfs_init(struct osd_device *osd, const char *name);
630 int osd_procfs_fini(struct osd_device *osd);
631 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
634 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
635 struct obd_statfs *sfs);
636 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
637 struct lustre_capa *capa, __u64 opc);
638 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
639 struct osd_inode_id *id);
640 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
641 const struct lu_fid *fid, __u32 compat, __u32 incompat);
642 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
643 struct dentry *dentry, struct lustre_mdt_attrs *lma);
644 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
645 struct osd_inode_id *id, const struct lu_fid *fid);
646 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
647 struct dentry *dentry, struct lu_fid *fid);
649 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
650 void osd_obj_map_fini(struct osd_device *dev);
651 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
652 const struct lu_fid *fid, struct osd_inode_id *id);
653 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
654 const struct lu_fid *fid, const struct osd_inode_id *id,
656 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
657 const struct lu_fid *fid, struct thandle *th);
658 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
659 const struct lu_fid *fid, const struct osd_inode_id *id,
661 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
662 struct inode *src_parent, struct dentry *src_child,
663 const struct lu_fid *fid);
664 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
665 const struct lu_fid *fid, struct osd_inode_id *id);
666 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
667 const struct lu_fid *fid, const struct osd_inode_id *id,
669 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
670 const struct lu_fid *fid, const struct osd_inode_id *id,
673 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
674 int osd_scrub_file_store(struct osd_scrub *scrub);
675 char *osd_lf_fid2name(const struct lu_fid *fid);
676 int osd_scrub_start(struct osd_device *dev);
677 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
678 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
679 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
681 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
682 struct osd_inode_id *id);
683 int osd_scrub_dump(struct osd_device *dev, char *buf, int len);
685 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
686 const struct lu_fid *fid, struct lu_seq_range *range);
688 int osd_delete_from_remote_parent(const struct lu_env *env,
689 struct osd_device *osd,
690 struct osd_object *obj,
691 struct osd_thandle *oh);
692 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
693 struct osd_object *obj, struct osd_thandle *oh);
694 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
695 struct osd_device *osd,
696 const struct lu_fid *fid,
697 struct osd_inode_id *id);
699 /* osd_quota_fmt.c */
700 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
701 int type, uint blk, int depth, uint index,
702 struct osd_it_quota *it);
703 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
704 int type, uint blk, uint index,
705 struct osd_it_quota *it);
706 loff_t find_tree_dqentry(const struct lu_env *env,
707 struct osd_object *obj, int type,
708 qid_t dqid, uint blk, int depth,
709 struct osd_it_quota *it);
711 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
712 struct lquota_id_info *qi, bool allocated, int *flags);
713 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
714 long long space, struct osd_thandle *oh,
715 bool is_blk, bool allocated, int *flags, bool force);
716 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
717 const struct dt_rec *rec,
718 union lquota_rec *quota_rec);
719 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
720 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
721 const struct dt_index_features *feat);
723 static inline bool is_quota_glb_feat(const struct dt_index_features *feat)
725 return (feat == &dt_quota_iusr_features ||
726 feat == &dt_quota_busr_features ||
727 feat == &dt_quota_igrp_features ||
728 feat == &dt_quota_bgrp_features) ? true : false;
732 * Invariants, assertions.
736 * XXX: do not enable this, until invariant checking code is made thread safe
737 * in the face of pdirops locking.
739 #define OSD_INVARIANT_CHECKS (0)
741 #if OSD_INVARIANT_CHECKS
742 static inline int osd_invariant(const struct osd_object *obj)
746 ergo(obj->oo_inode != NULL,
747 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
748 atomic_read(&obj->oo_inode->i_count) > 0) &&
749 ergo(obj->oo_dir != NULL &&
750 obj->oo_dir->od_conationer.ic_object != NULL,
751 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
754 #define osd_invariant(obj) (1)
757 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
759 extern const struct dt_index_operations osd_otable_ops;
761 static inline int osd_oi_fid2idx(struct osd_device *dev,
762 const struct lu_fid *fid)
764 return fid->f_seq & (dev->od_oi_count - 1);
767 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
768 const struct lu_fid *fid)
770 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
771 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
772 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
773 DFID"\n", PFID(fid));
774 /* It can work even od_oi_count equals to 1 although it's unexpected,
775 * the only reason we set it to 1 is for performance measurement */
776 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
779 extern const struct lu_device_operations osd_lu_ops;
781 static inline int lu_device_is_osd(const struct lu_device *d)
783 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
786 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
788 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
789 return container_of0(d, struct osd_device, od_dt_dev);
792 static inline struct osd_device *osd_dev(const struct lu_device *d)
794 LASSERT(lu_device_is_osd(d));
795 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
798 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
800 return osd_dev(o->oo_dt.do_lu.lo_dev);
803 static inline struct super_block *osd_sb(const struct osd_device *dev)
805 return dev->od_mnt->mnt_sb;
808 static inline int osd_object_is_root(const struct osd_object *obj)
810 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
813 static inline struct osd_object *osd_obj(const struct lu_object *o)
815 LASSERT(lu_device_is_osd(o->lo_dev));
816 return container_of0(o, struct osd_object, oo_dt.do_lu);
819 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
821 return osd_obj(&d->do_lu);
824 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
826 return &osd->od_dt_dev.dd_lu_dev;
829 static inline journal_t *osd_journal(const struct osd_device *dev)
831 return LDISKFS_SB(osd_sb(dev))->s_journal;
834 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
836 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
839 static inline char *osd_name(struct osd_device *osd)
841 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
845 extern const struct dt_body_operations osd_body_ops;
846 extern struct lu_context_key osd_key;
848 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
850 return lu_context_key_get(&env->le_ctx, &osd_key);
853 extern const struct dt_body_operations osd_body_ops_new;
859 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
860 const struct iam_container *bag)
862 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
863 osd_oti_get(env)->oti_it_ipd);
867 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
868 const struct iam_container *bag)
870 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
871 osd_oti_get(env)->oti_idx_ipd);
874 static inline void osd_ipd_put(const struct lu_env *env,
875 const struct iam_container *bag,
876 struct iam_path_descr *ipd)
878 bag->ic_descr->id_ops->id_ipd_free(ipd);
881 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
882 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
883 int write_NUL, loff_t *offs, handle_t *handle);
886 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
888 const char *name, const int namelen)
890 struct osd_thread_info *info = osd_oti_get(env);
891 struct dentry *child_dentry = &info->oti_child_dentry;
892 struct dentry *obj_dentry = &info->oti_obj_dentry;
894 obj_dentry->d_inode = inode;
895 obj_dentry->d_sb = inode->i_sb;
896 obj_dentry->d_name.hash = 0;
898 child_dentry->d_name.hash = 0;
899 child_dentry->d_parent = obj_dentry;
900 child_dentry->d_name.name = name;
901 child_dentry->d_name.len = namelen;
905 extern int osd_trans_declare_op2rb[];
906 extern int ldiskfs_track_declares_assert;
908 static inline void osd_trans_declare_op(const struct lu_env *env,
909 struct osd_thandle *oh,
910 unsigned int op, int credits)
912 struct osd_thread_info *oti = osd_oti_get(env);
914 LASSERT(oh->ot_handle == NULL);
915 if (unlikely(op >= OSD_OT_MAX)) {
916 if (unlikely(ldiskfs_track_declares_assert))
917 LASSERT(op < OSD_OT_MAX);
919 CWARN("%s: Invalid operation index %d\n",
920 osd_name(oti->oti_dev), op);
921 libcfs_debug_dumpstack(NULL);
924 oti->oti_declare_ops[op]++;
925 oti->oti_declare_ops_cred[op] += credits;
927 oh->ot_credits += credits;
930 static inline void osd_trans_exec_op(const struct lu_env *env,
931 struct thandle *th, unsigned int op)
933 struct osd_thread_info *oti = osd_oti_get(env);
934 struct osd_thandle *oh = container_of(th, struct osd_thandle,
938 LASSERT(oh->ot_handle != NULL);
939 if (unlikely(op >= OSD_OT_MAX)) {
940 if (unlikely(ldiskfs_track_declares_assert))
941 LASSERT(op < OSD_OT_MAX);
943 CWARN("%s: Invalid operation index %d\n",
944 osd_name(oti->oti_dev), op);
945 libcfs_debug_dumpstack(NULL);
950 if (likely(!oti->oti_rollback && oti->oti_declare_ops[op] > 0)) {
951 oti->oti_declare_ops[op]--;
952 oti->oti_declare_ops_rb[op]++;
954 /* all future updates are considered rollback */
955 oti->oti_rollback = true;
956 rb = osd_trans_declare_op2rb[op];
957 if (unlikely(rb >= OSD_OT_MAX)) {
958 if (unlikely(ldiskfs_track_declares_assert))
959 LASSERTF(rb < OSD_OT_MAX, "rb = %u\n", rb);
961 CWARN("%s: Invalid rollback index %d\n",
962 osd_name(oti->oti_dev), rb);
963 libcfs_debug_dumpstack(NULL);
967 if (unlikely(oti->oti_declare_ops_rb[rb] == 0)) {
968 if (unlikely(ldiskfs_track_declares_assert))
969 LASSERTF(oti->oti_declare_ops_rb[rb] > 0,
972 CWARN("%s: Overflow in tracking declares for "
974 osd_name(oti->oti_dev), rb);
975 libcfs_debug_dumpstack(NULL);
979 oti->oti_declare_ops_rb[rb]--;
983 static inline void osd_trans_declare_rb(const struct lu_env *env,
984 struct thandle *th, unsigned int op)
986 struct osd_thread_info *oti = osd_oti_get(env);
987 struct osd_thandle *oh = container_of(th, struct osd_thandle,
990 LASSERT(oh->ot_handle != NULL);
991 if (unlikely(op >= OSD_OT_MAX)) {
992 if (unlikely(ldiskfs_track_declares_assert))
993 LASSERT(op < OSD_OT_MAX);
995 CWARN("%s: Invalid operation index %d\n",
996 osd_name(oti->oti_dev), op);
997 libcfs_debug_dumpstack(NULL);
1001 oti->oti_declare_ops_rb[op]++;
1006 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1009 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1010 struct lu_fid *befider)
1012 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1013 memcpy(pack->fp_area, befider, sizeof(*befider));
1014 pack->fp_len = sizeof(*befider) + 1;
1018 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1023 switch (pack->fp_len) {
1024 case sizeof *fid + 1:
1025 memcpy(fid, pack->fp_area, sizeof *fid);
1026 fid_be_to_cpu(fid, fid);
1029 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1036 * Quota/Accounting handling
1038 extern const struct dt_index_operations osd_acct_index_ops;
1039 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1040 const struct lu_fid *fid, struct osd_inode_id *id);
1042 /* copy from fs/ext4/dir.c */
1043 static inline int is_32bit_api(void)
1045 #ifdef CONFIG_COMPAT
1046 return is_compat_task();
1048 return (BITS_PER_LONG == 32);
1052 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1054 if ((filp->f_mode & FMODE_32BITHASH) ||
1055 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1056 return LDISKFS_HTREE_EOF_32BIT;
1058 return LDISKFS_HTREE_EOF_64BIT;
1061 static inline int fid_is_internal(const struct lu_fid *fid)
1063 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1065 #endif /* __KERNEL__ */
1066 #endif /* _OSD_INTERNAL_H */