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 128
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;
213 struct osd_otable_cache ooi_cache;
215 /* The following bits can be updated/checked w/o lock protection.
216 * If more bits will be introduced in the future and need lock to
217 * protect, please add comment. */
218 unsigned long ooi_used_outside:1, /* Some user out of OSD
219 * uses the iteration. */
220 ooi_all_cached:1, /* No more entries can be
221 * filled into cache. */
222 ooi_user_ready:1, /* The user out of OSD is
223 * ready to iterate. */
224 ooi_waiting:1, /* it::next is waiting. */
225 ooi_stopping:1; /* Someone is stopping
229 extern const int osd_dto_credits_noquota[];
236 struct dt_device od_dt_dev;
237 /* information about underlying file system */
238 struct vfsmount *od_mnt;
240 struct osd_oi **od_oi_table;
241 /* total number of OI containers */
246 unsigned int od_fl_capa:1,
247 od_is_md:1, /* set in ->ldo_prepare */
253 unsigned long od_capa_timeout;
255 struct lustre_capa_key *od_capa_keys;
256 cfs_hlist_head_t *od_capa_hash;
258 cfs_proc_dir_entry_t *od_proc_entry;
259 struct lprocfs_stats *od_stats;
261 * statfs optimization: we cache a bit.
263 cfs_time_t od_osfs_age;
264 struct obd_statfs od_statfs;
265 spinlock_t od_osfs_lock;
267 struct fsfilt_operations *od_fsops;
269 struct lu_site od_site;
271 struct osd_obj_map *od_ost_map;
272 struct osd_mdobj_map *od_mdt_map;
274 unsigned long long od_readcache_max_filesize;
276 int od_writethrough_cache;
278 struct brw_stats od_brw_stats;
279 cfs_atomic_t od_r_in_flight;
280 cfs_atomic_t od_w_in_flight;
282 struct mutex od_otable_mutex;
283 struct osd_otable_it *od_otable_it;
284 struct osd_scrub od_scrub;
285 cfs_list_t od_ios_list;
287 /* service name associated with the osd device */
288 char od_svname[MAX_OBD_NAME];
289 char od_mntdev[MAX_OBD_NAME];
291 /* quota slave instance */
292 struct qsd_instance *od_quota_slave;
295 /* There are at most 10 uid/gids are affected in a transaction, and
296 * that's rename case:
297 * - 2 for source parent uid & gid;
298 * - 2 for source child uid & gid ('..' entry update when child is directory);
299 * - 2 for target parent uid & gid;
300 * - 2 for target child uid & gid (if the target child exists);
301 * - 2 for root uid & gid (last_rcvd, llog, etc);
303 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
304 * the id type of each id in the ot_id_array.
306 #define OSD_MAX_UGID_CNT 10
311 OSD_OT_XATTR_SET = 2,
324 struct thandle ot_super;
326 struct ldiskfs_journal_cb_entry ot_jcb;
327 cfs_list_t ot_dcb_list;
328 /* Link to the device, for debugging. */
329 struct lu_ref_link ot_dev_link;
330 unsigned short ot_credits;
331 unsigned short ot_id_cnt;
332 unsigned short ot_id_type;
333 uid_t ot_id_array[OSD_MAX_UGID_CNT];
334 struct lquota_trans *ot_quota_trans;
335 #if OSD_THANDLE_STATS
336 /** time when this handle was allocated */
337 cfs_time_t oth_alloced;
339 /** time when this thanle was started */
340 cfs_time_t oth_started;
345 * Basic transaction credit op
355 DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
369 LPROC_OSD_READ_BYTES = 0,
370 LPROC_OSD_WRITE_BYTES = 1,
371 LPROC_OSD_GET_PAGE = 2,
372 LPROC_OSD_NO_PAGE = 3,
373 LPROC_OSD_CACHE_ACCESS = 4,
374 LPROC_OSD_CACHE_HIT = 5,
375 LPROC_OSD_CACHE_MISS = 6,
377 #if OSD_THANDLE_STATS
378 LPROC_OSD_THANDLE_STARTING,
379 LPROC_OSD_THANDLE_OPEN,
380 LPROC_OSD_THANDLE_CLOSING,
387 * Storage representation for fids.
389 * Variable size, first byte contains the length of the whole record.
391 struct osd_fid_pack {
392 unsigned char fp_len;
393 char fp_area[sizeof(struct lu_fid)];
396 struct osd_it_ea_dirent {
397 struct lu_fid oied_fid;
400 unsigned short oied_namelen;
401 unsigned int oied_type;
403 } __attribute__((packed));
406 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
407 * than lu_dirent struct. osd readdir reads less number of dirent than
408 * required for mdd dir page. so buffer size need to be increased so that
409 * there would be one ext3 readdir for every mdd readdir page.
412 #define OSD_IT_EA_BUFSIZE (CFS_PAGE_SIZE + CFS_PAGE_SIZE/4)
415 * This is iterator's in-memory data structure in interoperability
416 * mode (i.e. iterator over ldiskfs style directory)
419 struct osd_object *oie_obj;
420 /** used in ldiskfs iterator, to stored file pointer */
421 struct file oie_file;
422 /** how many entries have been read-cached from storage */
424 /** current entry is being iterated by caller */
426 /** current processing entry */
427 struct osd_it_ea_dirent *oie_dirent;
428 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
433 * Iterator's in-memory data structure for IAM mode.
436 struct osd_object *oi_obj;
437 struct iam_path_descr *oi_ipd;
438 struct iam_iterator oi_it;
441 struct osd_quota_leaf {
447 * Iterator's in-memory data structure for quota file.
449 struct osd_it_quota {
450 struct osd_object *oiq_obj;
451 /** tree blocks path to where the entry is stored */
452 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
453 /** on-disk offset for current key where quota record can be found */
455 /** identifier for current quota record */
457 /** the record index in the leaf/index block */
458 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
459 /** list of already processed leaf blocks */
463 #define MAX_BLOCKS_PER_PAGE (CFS_PAGE_SIZE / 512)
467 cfs_atomic_t dr_numreqs; /* number of reqs being processed */
472 unsigned int dr_ignore_quota:1;
473 unsigned int dr_elapsed_valid:1; /* we really did count time */
474 unsigned int dr_rw:1;
475 struct lu_buf dr_pg_buf;
476 struct page **dr_pages;
477 struct lu_buf dr_bl_buf;
478 unsigned long *dr_blocks;
479 unsigned long dr_start_time;
480 unsigned long dr_elapsed; /* how long io took */
481 struct osd_device *dr_dev;
482 unsigned int dr_init_at; /* the line iobuf was initialized */
485 struct osd_thread_info {
486 const struct lu_env *oti_env;
488 * used for index operations.
490 struct dentry oti_obj_dentry;
491 struct dentry oti_child_dentry;
493 /** dentry for Iterator context. */
494 struct dentry oti_it_dentry;
495 struct htree_lock *oti_hlock;
497 struct lu_fid oti_fid;
498 struct lu_fid oti_fid2;
499 struct osd_inode_id oti_id;
500 struct osd_inode_id oti_id2;
501 struct ost_id oti_ostid;
504 * XXX temporary: for ->i_op calls.
506 struct timespec oti_time;
508 * XXX temporary: fake struct file for osd_object_sync
510 struct file oti_file;
512 * XXX temporary: for capa operations.
514 struct lustre_capa_key oti_capa_key;
515 struct lustre_capa oti_capa;
517 /** osd_device reference, initialized in osd_trans_start() and
518 used in osd_trans_stop() */
519 struct osd_device *oti_dev;
522 * following ipd and it structures are used for osd_index_iam_lookup()
523 * these are defined separately as we might do index operation
524 * in open iterator session.
527 /** osd iterator context used for iterator session */
530 struct osd_it_iam oti_it;
531 /* ldiskfs iterator data structure,
532 * see osd_it_ea_{init, fini} */
533 struct osd_it_ea oti_it_ea;
534 struct osd_it_quota oti_it_quota;
537 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
540 struct kstatfs oti_ksfs;
542 /** IAM iterator for index operation. */
543 struct iam_iterator oti_idx_it;
545 /** union to guarantee that ->oti_ipd[] has proper alignment. */
547 char oti_it_ipd[DX_IPD_MAX_SIZE];
548 long long oti_alignment_lieutenant;
552 char oti_idx_ipd[DX_IPD_MAX_SIZE];
553 long long oti_alignment_lieutenant_colonel;
556 struct osd_idmap_cache oti_cache;
561 /** used in osd_fid_set() to put xattr */
562 struct lu_buf oti_buf;
563 /** used in osd_ea_fid_set() to set fid into common ea */
565 struct lustre_mdt_attrs oti_mdt_attrs;
566 /* old LMA for compatibility */
567 char oti_mdt_attrs_old[LMA_OLD_SIZE];
570 struct osd_iobuf oti_iobuf;
571 struct inode oti_inode;
572 #define OSD_FID_REC_SZ 32
573 char oti_ldp[OSD_FID_REC_SZ];
574 char oti_ldp2[OSD_FID_REC_SZ];
576 /* used by quota code */
578 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
579 struct fs_disk_quota oti_fdq;
581 struct if_dqblk oti_dqblk;
583 struct if_dqinfo oti_dqinfo;
585 struct lquota_id_info oti_qi;
586 struct lquota_trans oti_quota_trans;
587 union lquota_rec oti_quota_rec;
589 struct lu_seq_range oti_seq_range;
591 /* Tracking for transaction credits, to allow debugging and optimizing
592 * cases where a large number of credits are being allocated for
593 * single transaction. */
594 unsigned short oti_declare_ops[OSD_OT_MAX];
595 unsigned short oti_declare_ops_rb[OSD_OT_MAX];
596 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
602 extern int ldiskfs_pdo;
604 static inline int __osd_xattr_set(struct osd_thread_info *info,
605 struct inode *inode, const char *name,
606 const void *buf, int buflen, int fl)
608 struct dentry *dentry = &info->oti_child_dentry;
610 ll_vfs_dq_init(inode);
611 dentry->d_inode = inode;
612 dentry->d_sb = inode->i_sb;
613 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
618 void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars);
619 int osd_procfs_init(struct osd_device *osd, const char *name);
620 int osd_procfs_fini(struct osd_device *osd);
621 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
624 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
625 struct obd_statfs *sfs);
626 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
627 struct lustre_capa *capa, __u64 opc);
628 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
629 struct osd_inode_id *id);
630 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
631 const struct lu_fid *fid, __u64 flags);
632 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
633 struct dentry *dentry, struct lustre_mdt_attrs *lma);
635 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
636 void osd_obj_map_fini(struct osd_device *dev);
637 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
638 const struct lu_fid *fid, struct osd_inode_id *id);
639 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
640 const struct lu_fid *fid, const struct osd_inode_id *id,
642 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
643 const struct lu_fid *fid, struct thandle *th);
644 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
645 const struct lu_fid *fid, struct osd_inode_id *id);
646 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
647 const struct lu_fid *fid, const struct osd_inode_id *id,
650 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
651 int osd_scrub_file_store(struct osd_scrub *scrub);
652 char *osd_lf_fid2name(const struct lu_fid *fid);
653 int osd_scrub_start(struct osd_device *dev);
654 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
655 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
656 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
658 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
659 struct osd_inode_id *id);
660 int osd_scrub_dump(struct osd_device *dev, char *buf, int len);
662 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
663 const struct lu_fid *fid, struct lu_seq_range *range);
665 int osd_delete_from_remote_parent(const struct lu_env *env,
666 struct osd_device *osd,
667 struct osd_object *obj,
668 struct osd_thandle *oh);
669 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
670 struct osd_object *obj, struct osd_thandle *oh);
672 /* osd_quota_fmt.c */
673 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
674 int type, uint blk, int depth, uint index,
675 struct osd_it_quota *it);
676 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
677 int type, uint blk, uint index,
678 struct osd_it_quota *it);
679 loff_t find_tree_dqentry(const struct lu_env *env,
680 struct osd_object *obj, int type,
681 qid_t dqid, uint blk, int depth,
682 struct osd_it_quota *it);
684 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
685 struct lquota_id_info *qi, bool allocated, int *flags);
686 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
687 long long space, struct osd_thandle *oh,
688 bool is_blk, bool allocated, int *flags, bool force);
689 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
690 const struct dt_rec *rec,
691 union lquota_rec *quota_rec);
692 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
693 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
694 const struct dt_index_features *feat);
697 struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq);
699 static inline bool is_quota_glb_feat(const struct dt_index_features *feat)
701 return (feat == &dt_quota_iusr_features ||
702 feat == &dt_quota_busr_features ||
703 feat == &dt_quota_igrp_features ||
704 feat == &dt_quota_bgrp_features) ? true : false;
708 * Invariants, assertions.
712 * XXX: do not enable this, until invariant checking code is made thread safe
713 * in the face of pdirops locking.
715 #define OSD_INVARIANT_CHECKS (0)
717 #if OSD_INVARIANT_CHECKS
718 static inline int osd_invariant(const struct osd_object *obj)
722 ergo(obj->oo_inode != NULL,
723 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
724 atomic_read(&obj->oo_inode->i_count) > 0) &&
725 ergo(obj->oo_dir != NULL &&
726 obj->oo_dir->od_conationer.ic_object != NULL,
727 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
730 #define osd_invariant(obj) (1)
733 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
735 extern const struct dt_index_operations osd_otable_ops;
737 static inline int osd_oi_fid2idx(struct osd_device *dev,
738 const struct lu_fid *fid)
740 return fid->f_seq & (dev->od_oi_count - 1);
743 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
744 const struct lu_fid *fid)
746 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
747 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
748 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
749 DFID"\n", PFID(fid));
750 /* It can work even od_oi_count equals to 1 although it's unexpected,
751 * the only reason we set it to 1 is for performance measurement */
752 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
755 extern const struct lu_device_operations osd_lu_ops;
757 static inline int lu_device_is_osd(const struct lu_device *d)
759 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
762 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
764 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
765 return container_of0(d, struct osd_device, od_dt_dev);
768 static inline struct osd_device *osd_dev(const struct lu_device *d)
770 LASSERT(lu_device_is_osd(d));
771 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
774 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
776 return osd_dev(o->oo_dt.do_lu.lo_dev);
779 static inline struct super_block *osd_sb(const struct osd_device *dev)
781 return dev->od_mnt->mnt_sb;
784 static inline int osd_object_is_root(const struct osd_object *obj)
786 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
789 static inline struct osd_object *osd_obj(const struct lu_object *o)
791 LASSERT(lu_device_is_osd(o->lo_dev));
792 return container_of0(o, struct osd_object, oo_dt.do_lu);
795 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
797 return osd_obj(&d->do_lu);
800 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
802 return &osd->od_dt_dev.dd_lu_dev;
805 static inline journal_t *osd_journal(const struct osd_device *dev)
807 return LDISKFS_SB(osd_sb(dev))->s_journal;
810 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
812 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
815 static inline char *osd_name(struct osd_device *osd)
817 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
821 extern const struct dt_body_operations osd_body_ops;
822 extern struct lu_context_key osd_key;
824 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
826 return lu_context_key_get(&env->le_ctx, &osd_key);
829 extern const struct dt_body_operations osd_body_ops_new;
835 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
836 const struct iam_container *bag)
838 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
839 osd_oti_get(env)->oti_it_ipd);
843 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
844 const struct iam_container *bag)
846 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
847 osd_oti_get(env)->oti_idx_ipd);
850 static inline void osd_ipd_put(const struct lu_env *env,
851 const struct iam_container *bag,
852 struct iam_path_descr *ipd)
854 bag->ic_descr->id_ops->id_ipd_free(ipd);
857 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
858 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
859 int write_NUL, loff_t *offs, handle_t *handle);
862 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
864 const char *name, const int namelen)
866 struct osd_thread_info *info = osd_oti_get(env);
867 struct dentry *child_dentry = &info->oti_child_dentry;
868 struct dentry *obj_dentry = &info->oti_obj_dentry;
870 obj_dentry->d_inode = inode;
871 obj_dentry->d_sb = inode->i_sb;
872 obj_dentry->d_name.hash = 0;
874 child_dentry->d_name.hash = 0;
875 child_dentry->d_parent = obj_dentry;
876 child_dentry->d_name.name = name;
877 child_dentry->d_name.len = namelen;
881 extern int osd_trans_declare_op2rb[];
882 extern int ldiskfs_track_declares_assert;
884 static inline void osd_trans_declare_op(const struct lu_env *env,
885 struct osd_thandle *oh,
886 unsigned int op, int credits)
888 struct osd_thread_info *oti = osd_oti_get(env);
890 LASSERT(oh->ot_handle == NULL);
891 if (unlikely(op >= OSD_OT_MAX)) {
892 if (unlikely(ldiskfs_track_declares_assert))
893 LASSERT(op < OSD_OT_MAX);
895 CWARN("%s: Invalid operation index %d\n",
896 osd_name(oti->oti_dev), op);
897 libcfs_debug_dumpstack(NULL);
900 oti->oti_declare_ops[op]++;
901 oti->oti_declare_ops_cred[op] += credits;
903 oh->ot_credits += credits;
906 static inline void osd_trans_exec_op(const struct lu_env *env,
907 struct thandle *th, unsigned int op)
909 struct osd_thread_info *oti = osd_oti_get(env);
910 struct osd_thandle *oh = container_of(th, struct osd_thandle,
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);
926 if (likely(!oti->oti_rollback && oti->oti_declare_ops[op] > 0)) {
927 oti->oti_declare_ops[op]--;
928 oti->oti_declare_ops_rb[op]++;
930 /* all future updates are considered rollback */
931 oti->oti_rollback = true;
932 rb = osd_trans_declare_op2rb[op];
933 if (unlikely(rb >= OSD_OT_MAX)) {
934 if (unlikely(ldiskfs_track_declares_assert))
935 LASSERTF(rb < OSD_OT_MAX, "rb = %u\n", rb);
937 CWARN("%s: Invalid rollback index %d\n",
938 osd_name(oti->oti_dev), rb);
939 libcfs_debug_dumpstack(NULL);
943 if (unlikely(oti->oti_declare_ops_rb[rb] == 0)) {
944 if (unlikely(ldiskfs_track_declares_assert))
945 LASSERTF(oti->oti_declare_ops_rb[rb] > 0,
948 CWARN("%s: Overflow in tracking declares for "
950 osd_name(oti->oti_dev), rb);
951 libcfs_debug_dumpstack(NULL);
955 oti->oti_declare_ops_rb[rb]--;
959 static inline void osd_trans_declare_rb(const struct lu_env *env,
960 struct thandle *th, unsigned int op)
962 struct osd_thread_info *oti = osd_oti_get(env);
963 struct osd_thandle *oh = container_of(th, struct osd_thandle,
966 LASSERT(oh->ot_handle != NULL);
967 if (unlikely(op >= OSD_OT_MAX)) {
968 if (unlikely(ldiskfs_track_declares_assert))
969 LASSERT(op < OSD_OT_MAX);
971 CWARN("%s: Invalid operation index %d\n",
972 osd_name(oti->oti_dev), op);
973 libcfs_debug_dumpstack(NULL);
977 oti->oti_declare_ops_rb[op]++;
982 * Helper function to pack the fid, ldiskfs stores fid in packed format.
985 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
986 struct lu_fid *befider)
988 fid_cpu_to_be(befider, (struct lu_fid *)fid);
989 memcpy(pack->fp_area, befider, sizeof(*befider));
990 pack->fp_len = sizeof(*befider) + 1;
994 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
999 switch (pack->fp_len) {
1000 case sizeof *fid + 1:
1001 memcpy(fid, pack->fp_area, sizeof *fid);
1002 fid_be_to_cpu(fid, fid);
1005 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1012 * Quota/Accounting handling
1014 extern const struct dt_index_operations osd_acct_index_ops;
1015 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1016 const struct lu_fid *fid, struct osd_inode_id *id);
1018 /* copy from fs/ext4/dir.c */
1019 static inline int is_32bit_api(void)
1021 #ifdef CONFIG_COMPAT
1022 return is_compat_task();
1024 return (BITS_PER_LONG == 32);
1028 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1030 if ((filp->f_mode & FMODE_32BITHASH) ||
1031 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1032 return LDISKFS_HTREE_EOF_32BIT;
1034 return LDISKFS_HTREE_EOF_64BIT;
1037 static inline int fid_is_internal(const struct lu_fid *fid)
1039 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1041 #endif /* __KERNEL__ */
1042 #endif /* _OSD_INTERNAL_H */