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.h>
56 #include <ldiskfs/ldiskfs_jbd2.h>
60 /* class_register_type(), class_unregister_type(), class_get_type() */
61 #include <obd_class.h>
62 #include <lustre_disk.h>
63 #include <dt_object.h>
64 #include <lustre_quota.h>
68 #include "osd_scrub.h"
69 #include "osd_quota_fmt.h"
73 #define OSD_COUNTERS (0)
75 /* ldiskfs special inode::i_state_flags need to be accessed with
76 * ldiskfs_{set,clear,test}_inode_state() only */
78 /* OI scrub should skip this inode. */
79 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
81 /* Do not add OI mapping for this inode. */
82 #define LDISKFS_STATE_LUSTRE_NO_OI 30
84 /** Enable thandle usage statistics */
85 #define OSD_THANDLE_STATS (0)
87 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
89 #define OBJECTS "OBJECTS"
90 #define ADMIN_USR "admin_quotafile_v2.usr"
91 #define ADMIN_GRP "admin_quotafile_v2.grp"
93 struct osd_directory {
94 struct iam_container od_container;
95 struct iam_descr od_descr;
99 * Object Index (oi) instance.
103 * underlying index object, where fid->id mapping in stored.
105 struct inode *oi_inode;
106 struct osd_directory oi_dir;
109 extern const int osd_dto_credits_noquota[];
112 struct dt_object oo_dt;
114 * Inode for file system object represented by this osd_object. This
115 * inode is pinned for the whole duration of lu_object life.
117 * Not modified concurrently (either setup early during object
118 * creation, or assigned by osd_object_create() under write lock).
120 struct inode *oo_inode;
122 * to protect index ops.
124 struct htree_lock_head *oo_hl_head;
125 struct rw_semaphore oo_ext_idx_sem;
126 struct rw_semaphore oo_sem;
127 struct osd_directory *oo_dir;
128 /** protects inode attributes. */
131 * Following two members are used to indicate the presence of dot and
132 * dotdot in the given directory. This is required for interop mode
135 int oo_compat_dot_created;
136 int oo_compat_dotdot_created;
138 const struct lu_env *oo_owner;
139 #ifdef CONFIG_LOCKDEP
140 struct lockdep_map oo_dep_map;
145 /* protects on-fly initialization */
146 int oos_subdir_count; /* subdir count for each seq */
147 struct dentry *oos_root; /* O/<seq> */
148 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
149 obd_seq oos_seq; /* seq number */
150 cfs_list_t oos_seq_list; /* list to seq_list */
154 struct dentry *om_root; /* dentry for /O */
155 rwlock_t om_seq_list_lock; /* lock for seq_list */
156 cfs_list_t om_seq_list; /* list head for seq */
158 struct semaphore om_dir_init_sem;
162 struct dentry *om_root; /* AGENT/<index> */
163 obd_seq om_index; /* mdt index */
164 cfs_list_t om_list; /* list to omm_list */
167 struct osd_mdobj_map {
168 struct dentry *omm_remote_parent;
171 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
172 ldiskfs_add_entry(handle, child, cinode, hlock)
174 #define OSD_OTABLE_IT_CACHE_SIZE 64
175 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
177 struct osd_inconsistent_item {
178 /* link into osd_scrub::os_inconsistent_items,
179 * protected by osd_scrub::os_lock. */
182 /* The right FID <=> ino#/gen mapping. */
183 struct osd_idmap_cache oii_cache;
185 unsigned int oii_insert:1; /* insert or update mapping. */
188 struct osd_otable_cache {
189 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
191 /* Index for next cache slot to be filled. */
192 int ooc_producer_idx;
194 /* Index for next cache slot to be returned by it::next(). */
195 int ooc_consumer_idx;
197 /* How many items in ooc_cache. */
198 int ooc_cached_items;
200 /* Position for up layer LFSCK iteration pre-loading. */
201 __u32 ooc_pos_preload;
204 struct osd_otable_it {
205 struct osd_device *ooi_dev;
206 struct osd_otable_cache ooi_cache;
208 /* The following bits can be updated/checked w/o lock protection.
209 * If more bits will be introduced in the future and need lock to
210 * protect, please add comment. */
211 unsigned long ooi_used_outside:1, /* Some user out of OSD
212 * uses the iteration. */
213 ooi_all_cached:1, /* No more entries can be
214 * filled into cache. */
215 ooi_user_ready:1, /* The user out of OSD is
216 * ready to iterate. */
217 ooi_waiting:1; /* it::next is waiting. */
225 struct dt_device od_dt_dev;
226 /* information about underlying file system */
227 struct vfsmount *od_mnt;
229 struct osd_oi **od_oi_table;
230 /* total number of OI containers */
235 unsigned int od_fl_capa:1,
242 od_lma_self_repair:1;
244 unsigned long od_capa_timeout;
246 struct lustre_capa_key *od_capa_keys;
247 cfs_hlist_head_t *od_capa_hash;
249 cfs_proc_dir_entry_t *od_proc_entry;
250 struct lprocfs_stats *od_stats;
252 * statfs optimization: we cache a bit.
254 cfs_time_t od_osfs_age;
255 struct obd_statfs od_statfs;
256 spinlock_t od_osfs_lock;
259 struct lu_site od_site;
261 struct osd_obj_map *od_ost_map;
262 struct osd_mdobj_map *od_mdt_map;
264 unsigned long long od_readcache_max_filesize;
266 int od_writethrough_cache;
268 struct brw_stats od_brw_stats;
269 cfs_atomic_t od_r_in_flight;
270 cfs_atomic_t od_w_in_flight;
272 struct mutex od_otable_mutex;
273 struct osd_otable_it *od_otable_it;
274 struct osd_scrub od_scrub;
275 cfs_list_t od_ios_list;
277 /* service name associated with the osd device */
278 char od_svname[MAX_OBD_NAME];
279 char od_mntdev[MAX_OBD_NAME];
281 /* quota slave instance */
282 struct qsd_instance *od_quota_slave;
285 /* There are at most 10 uid/gids are affected in a transaction, and
286 * that's rename case:
287 * - 2 for source parent uid & gid;
288 * - 2 for source child uid & gid ('..' entry update when child is directory);
289 * - 2 for target parent uid & gid;
290 * - 2 for target child uid & gid (if the target child exists);
291 * - 2 for root uid & gid (last_rcvd, llog, etc);
293 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
294 * the id type of each id in the ot_id_array.
296 #define OSD_MAX_UGID_CNT 10
301 OSD_OT_XATTR_SET = 2,
315 struct thandle ot_super;
317 struct ldiskfs_journal_cb_entry ot_jcb;
318 cfs_list_t ot_dcb_list;
319 /* Link to the device, for debugging. */
320 struct lu_ref_link ot_dev_link;
321 unsigned short ot_credits;
322 unsigned short ot_id_cnt;
323 unsigned short ot_id_type;
324 uid_t ot_id_array[OSD_MAX_UGID_CNT];
325 struct lquota_trans *ot_quota_trans;
326 #if OSD_THANDLE_STATS
327 /** time when this handle was allocated */
328 cfs_time_t oth_alloced;
330 /** time when this thanle was started */
331 cfs_time_t oth_started;
336 * Basic transaction credit op
346 DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
360 LPROC_OSD_READ_BYTES = 0,
361 LPROC_OSD_WRITE_BYTES = 1,
362 LPROC_OSD_GET_PAGE = 2,
363 LPROC_OSD_NO_PAGE = 3,
364 LPROC_OSD_CACHE_ACCESS = 4,
365 LPROC_OSD_CACHE_HIT = 5,
366 LPROC_OSD_CACHE_MISS = 6,
368 #if OSD_THANDLE_STATS
369 LPROC_OSD_THANDLE_STARTING,
370 LPROC_OSD_THANDLE_OPEN,
371 LPROC_OSD_THANDLE_CLOSING,
378 * Storage representation for fids.
380 * Variable size, first byte contains the length of the whole record.
382 struct osd_fid_pack {
383 unsigned char fp_len;
384 char fp_area[sizeof(struct lu_fid)];
387 struct osd_it_ea_dirent {
388 struct lu_fid oied_fid;
391 unsigned short oied_namelen;
392 unsigned int oied_type;
394 } __attribute__((packed));
397 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
398 * than lu_dirent struct. osd readdir reads less number of dirent than
399 * required for mdd dir page. so buffer size need to be increased so that
400 * there would be one ext3 readdir for every mdd readdir page.
403 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
406 * This is iterator's in-memory data structure in interoperability
407 * mode (i.e. iterator over ldiskfs style directory)
410 struct osd_object *oie_obj;
411 /** used in ldiskfs iterator, to stored file pointer */
412 struct file oie_file;
413 /** how many entries have been read-cached from storage */
415 /** current entry is being iterated by caller */
417 /** current processing entry */
418 struct osd_it_ea_dirent *oie_dirent;
419 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
424 * Iterator's in-memory data structure for IAM mode.
427 struct osd_object *oi_obj;
428 struct iam_path_descr *oi_ipd;
429 struct iam_iterator oi_it;
432 struct osd_quota_leaf {
438 * Iterator's in-memory data structure for quota file.
440 struct osd_it_quota {
441 struct osd_object *oiq_obj;
442 /** tree blocks path to where the entry is stored */
443 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
444 /** on-disk offset for current key where quota record can be found */
446 /** identifier for current quota record */
448 /** the record index in the leaf/index block */
449 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
450 /** list of already processed leaf blocks */
454 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
457 wait_queue_head_t dr_wait;
458 cfs_atomic_t dr_numreqs; /* number of reqs being processed */
463 unsigned int dr_ignore_quota:1;
464 unsigned int dr_elapsed_valid:1; /* we really did count time */
465 unsigned int dr_rw:1;
466 struct lu_buf dr_pg_buf;
467 struct page **dr_pages;
468 struct lu_buf dr_bl_buf;
469 unsigned long *dr_blocks;
470 unsigned long dr_start_time;
471 unsigned long dr_elapsed; /* how long io took */
472 struct osd_device *dr_dev;
473 unsigned int dr_init_at; /* the line iobuf was initialized */
476 struct osd_thread_info {
477 const struct lu_env *oti_env;
479 * used for index operations.
481 struct dentry oti_obj_dentry;
482 struct dentry oti_child_dentry;
484 /** dentry for Iterator context. */
485 struct dentry oti_it_dentry;
486 struct htree_lock *oti_hlock;
488 struct lu_fid oti_fid;
489 struct lu_fid oti_fid2;
490 struct lu_fid oti_fid3;
491 struct osd_inode_id oti_id;
492 struct osd_inode_id oti_id2;
493 struct osd_inode_id oti_id3;
494 struct ost_id oti_ostid;
497 * XXX temporary: for ->i_op calls.
499 struct timespec oti_time;
501 * XXX temporary: fake struct file for osd_object_sync
503 struct file oti_file;
505 * XXX temporary: for capa operations.
507 struct lustre_capa_key oti_capa_key;
508 struct lustre_capa oti_capa;
510 /** osd_device reference, initialized in osd_trans_start() and
511 used in osd_trans_stop() */
512 struct osd_device *oti_dev;
515 * following ipd and it structures are used for osd_index_iam_lookup()
516 * these are defined separately as we might do index operation
517 * in open iterator session.
520 /** osd iterator context used for iterator session */
523 struct osd_it_iam oti_it;
524 /* ldiskfs iterator data structure,
525 * see osd_it_ea_{init, fini} */
526 struct osd_it_ea oti_it_ea;
527 struct osd_it_quota oti_it_quota;
530 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
533 struct kstatfs oti_ksfs;
535 /** IAM iterator for index operation. */
536 struct iam_iterator oti_idx_it;
538 /** union to guarantee that ->oti_ipd[] has proper alignment. */
540 char oti_it_ipd[DX_IPD_MAX_SIZE];
541 long long oti_alignment_lieutenant;
545 char oti_idx_ipd[DX_IPD_MAX_SIZE];
546 long long oti_alignment_lieutenant_colonel;
549 struct osd_idmap_cache oti_cache;
554 /** used in osd_fid_set() to put xattr */
555 struct lu_buf oti_buf;
556 /** used in osd_ea_fid_set() to set fid into common ea */
558 struct lustre_mdt_attrs oti_mdt_attrs;
559 /* old LMA for compatibility */
560 char oti_mdt_attrs_old[LMA_OLD_SIZE];
563 struct osd_iobuf oti_iobuf;
564 struct inode oti_inode;
565 #define OSD_FID_REC_SZ 32
566 char oti_ldp[OSD_FID_REC_SZ];
567 char oti_ldp2[OSD_FID_REC_SZ];
569 /* used by quota code */
571 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
572 struct fs_disk_quota oti_fdq;
574 struct if_dqblk oti_dqblk;
576 struct if_dqinfo oti_dqinfo;
578 struct lquota_id_info oti_qi;
579 struct lquota_trans oti_quota_trans;
580 union lquota_rec oti_quota_rec;
582 struct lu_seq_range oti_seq_range;
584 /* Tracking for transaction credits, to allow debugging and optimizing
585 * cases where a large number of credits are being allocated for
586 * single transaction. */
587 unsigned short oti_declare_ops[OSD_OT_MAX];
588 unsigned short oti_declare_ops_rb[OSD_OT_MAX];
589 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
594 struct filter_fid_old oti_ff;
595 struct filter_fid oti_ff_new;
599 extern int ldiskfs_pdo;
601 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
602 const char *name, void *buf, int len)
607 dentry->d_inode = inode;
608 dentry->d_sb = inode->i_sb;
609 return inode->i_op->getxattr(dentry, name, buf, len);
612 static inline int __osd_xattr_set(struct osd_thread_info *info,
613 struct inode *inode, const char *name,
614 const void *buf, int buflen, int fl)
616 struct dentry *dentry = &info->oti_child_dentry;
618 ll_vfs_dq_init(inode);
619 dentry->d_inode = inode;
620 dentry->d_sb = inode->i_sb;
621 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
626 void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars);
627 int osd_procfs_init(struct osd_device *osd, const char *name);
628 int osd_procfs_fini(struct osd_device *osd);
629 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
632 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
633 struct obd_statfs *sfs);
634 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
635 struct lustre_capa *capa, __u64 opc);
636 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
637 struct osd_inode_id *id);
638 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
639 const struct lu_fid *fid, __u32 compat, __u32 incompat);
640 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
641 struct dentry *dentry, struct lustre_mdt_attrs *lma);
642 int osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
643 struct osd_inode_id *id, const struct lu_fid *fid);
644 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
645 struct dentry *dentry, struct lu_fid *fid);
647 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
648 void osd_obj_map_fini(struct osd_device *dev);
649 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
650 const struct lu_fid *fid, struct osd_inode_id *id);
651 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
652 const struct lu_fid *fid, const struct osd_inode_id *id,
654 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
655 const struct lu_fid *fid, struct thandle *th);
656 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
657 const struct lu_fid *fid, const struct osd_inode_id *id,
659 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
660 struct inode *src_parent, struct dentry *src_child,
661 const struct lu_fid *fid);
662 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
663 const struct lu_fid *fid, struct osd_inode_id *id);
664 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
665 const struct lu_fid *fid, const struct osd_inode_id *id,
667 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
668 const struct lu_fid *fid, const struct osd_inode_id *id,
671 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
672 int osd_scrub_file_store(struct osd_scrub *scrub);
673 char *osd_lf_fid2name(const struct lu_fid *fid);
674 int osd_scrub_start(struct osd_device *dev);
675 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
676 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
677 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
679 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
680 struct osd_inode_id *id);
681 int osd_scrub_dump(struct osd_device *dev, char *buf, int len);
683 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
684 obd_seq seq, struct lu_seq_range *range);
686 int osd_delete_from_remote_parent(const struct lu_env *env,
687 struct osd_device *osd,
688 struct osd_object *obj,
689 struct osd_thandle *oh);
690 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
691 struct osd_object *obj, struct osd_thandle *oh);
692 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
693 struct osd_device *osd,
694 const struct lu_fid *fid,
695 struct osd_inode_id *id);
697 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
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;
844 extern const struct dt_body_operations osd_body_ops;
845 extern struct lu_context_key osd_key;
847 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
849 return lu_context_key_get(&env->le_ctx, &osd_key);
852 extern const struct dt_body_operations osd_body_ops_new;
858 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
859 const struct iam_container *bag)
861 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
862 osd_oti_get(env)->oti_it_ipd);
866 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
867 const struct iam_container *bag)
869 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
870 osd_oti_get(env)->oti_idx_ipd);
873 static inline void osd_ipd_put(const struct lu_env *env,
874 const struct iam_container *bag,
875 struct iam_path_descr *ipd)
877 bag->ic_descr->id_ops->id_ipd_free(ipd);
880 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
881 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
882 int write_NUL, loff_t *offs, handle_t *handle);
885 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
887 const char *name, const int namelen)
889 struct osd_thread_info *info = osd_oti_get(env);
890 struct dentry *child_dentry = &info->oti_child_dentry;
891 struct dentry *obj_dentry = &info->oti_obj_dentry;
893 obj_dentry->d_inode = inode;
894 obj_dentry->d_sb = inode->i_sb;
895 obj_dentry->d_name.hash = 0;
897 child_dentry->d_name.hash = 0;
898 child_dentry->d_parent = obj_dentry;
899 child_dentry->d_name.name = name;
900 child_dentry->d_name.len = namelen;
904 extern int osd_trans_declare_op2rb[];
905 extern int ldiskfs_track_declares_assert;
907 static inline void osd_trans_declare_op(const struct lu_env *env,
908 struct osd_thandle *oh,
909 unsigned int op, int credits)
911 struct osd_thread_info *oti = osd_oti_get(env);
913 LASSERT(oh->ot_handle == NULL);
914 if (unlikely(op >= OSD_OT_MAX)) {
915 if (unlikely(ldiskfs_track_declares_assert))
916 LASSERT(op < OSD_OT_MAX);
918 CWARN("%s: Invalid operation index %d\n",
919 osd_name(oti->oti_dev), op);
920 libcfs_debug_dumpstack(NULL);
923 oti->oti_declare_ops[op]++;
924 oti->oti_declare_ops_cred[op] += credits;
926 oh->ot_credits += credits;
929 static inline void osd_trans_exec_op(const struct lu_env *env,
930 struct thandle *th, unsigned int op)
932 struct osd_thread_info *oti = osd_oti_get(env);
933 struct osd_thandle *oh = container_of(th, struct osd_thandle,
937 LASSERT(oh->ot_handle != NULL);
938 if (unlikely(op >= OSD_OT_MAX)) {
939 if (unlikely(ldiskfs_track_declares_assert))
940 LASSERT(op < OSD_OT_MAX);
942 CWARN("%s: Invalid operation index %d\n",
943 osd_name(oti->oti_dev), op);
944 libcfs_debug_dumpstack(NULL);
949 if (likely(!oti->oti_rollback && oti->oti_declare_ops[op] > 0)) {
950 oti->oti_declare_ops[op]--;
951 oti->oti_declare_ops_rb[op]++;
953 /* all future updates are considered rollback */
954 oti->oti_rollback = true;
955 rb = osd_trans_declare_op2rb[op];
956 if (unlikely(rb >= OSD_OT_MAX)) {
957 if (unlikely(ldiskfs_track_declares_assert))
958 LASSERTF(rb < OSD_OT_MAX, "rb = %u\n", rb);
960 CWARN("%s: Invalid rollback index %d\n",
961 osd_name(oti->oti_dev), rb);
962 libcfs_debug_dumpstack(NULL);
966 if (unlikely(oti->oti_declare_ops_rb[rb] == 0)) {
967 if (unlikely(ldiskfs_track_declares_assert))
968 LASSERTF(oti->oti_declare_ops_rb[rb] > 0,
971 CWARN("%s: Overflow in tracking declares for "
973 osd_name(oti->oti_dev), rb);
974 libcfs_debug_dumpstack(NULL);
978 oti->oti_declare_ops_rb[rb]--;
982 static inline void osd_trans_declare_rb(const struct lu_env *env,
983 struct thandle *th, unsigned int op)
985 struct osd_thread_info *oti = osd_oti_get(env);
986 struct osd_thandle *oh = container_of(th, struct osd_thandle,
989 LASSERT(oh->ot_handle != NULL);
990 if (unlikely(op >= OSD_OT_MAX)) {
991 if (unlikely(ldiskfs_track_declares_assert))
992 LASSERT(op < OSD_OT_MAX);
994 CWARN("%s: Invalid operation index %d\n",
995 osd_name(oti->oti_dev), op);
996 libcfs_debug_dumpstack(NULL);
1000 oti->oti_declare_ops_rb[op]++;
1005 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1008 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1009 struct lu_fid *befider)
1011 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1012 memcpy(pack->fp_area, befider, sizeof(*befider));
1013 pack->fp_len = sizeof(*befider) + 1;
1017 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1022 switch (pack->fp_len) {
1023 case sizeof *fid + 1:
1024 memcpy(fid, pack->fp_area, sizeof *fid);
1025 fid_be_to_cpu(fid, fid);
1028 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1035 * Quota/Accounting handling
1037 extern const struct dt_index_operations osd_acct_index_ops;
1038 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1039 const struct lu_fid *fid, struct osd_inode_id *id);
1041 /* copy from fs/ext4/dir.c */
1042 static inline int is_32bit_api(void)
1044 #ifdef CONFIG_COMPAT
1045 return is_compat_task();
1047 return (BITS_PER_LONG == 32);
1051 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1053 if ((filp->f_mode & FMODE_32BITHASH) ||
1054 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1055 return LDISKFS_HTREE_EOF_32BIT;
1057 return LDISKFS_HTREE_EOF_64BIT;
1060 static inline int fid_is_internal(const struct lu_fid *fid)
1062 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1065 #ifdef JOURNAL_START_HAS_3ARGS
1066 # define osd_journal_start_sb(sb, type, nblock) \
1067 ldiskfs_journal_start_sb(sb, type, nblock)
1068 # define osd_ldiskfs_append(handle, inode, nblock, err) \
1069 ldiskfs_append(handle, inode, nblock)
1070 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1071 ldiskfs_find_entry(dir, name, de, inlined, lock)
1072 # define osd_journal_start(inode, type, nblocks) \
1073 ldiskfs_journal_start(inode, type, nblocks);
1075 # define LDISKFS_HT_MISC 0
1076 # define osd_journal_start_sb(sb, type, nblock) \
1077 ldiskfs_journal_start_sb(sb, nblock)
1078 # define osd_ldiskfs_append(handle, inode, nblock, err) \
1079 ldiskfs_append(handle, inode, nblock, err)
1080 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1081 ldiskfs_find_entry(dir, name, de, lock)
1082 # define osd_journal_start(inode, type, nblocks) \
1083 ldiskfs_journal_start(inode, nblocks);
1086 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1087 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1089 #endif /* __KERNEL__ */
1090 #endif /* _OSD_INTERNAL_H */