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, 2014, 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
48 #include <linux/mutex.h>
49 /* struct rw_semaphore */
50 #include <linux/rwsem.h>
52 #include <linux/dcache.h>
54 #include <linux/dirent.h>
55 #include <linux/statfs.h>
56 #include <ldiskfs/ldiskfs.h>
57 #include <ldiskfs/ldiskfs_jbd2.h>
61 /* class_register_type(), class_unregister_type(), class_get_type() */
62 #include <obd_class.h>
63 #include <lustre_disk.h>
64 #include <dt_object.h>
65 #include <lustre_quota.h>
69 #include "osd_scrub.h"
70 #include "osd_quota_fmt.h"
73 extern struct kmem_cache *dynlock_cachep;
75 #define OSD_COUNTERS (0)
77 /* ldiskfs special inode::i_state_flags need to be accessed with
78 * ldiskfs_{set,clear,test}_inode_state() only */
80 /* OI scrub should skip this inode. */
81 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
83 /** Enable thandle usage statistics */
84 #define OSD_THANDLE_STATS (0)
86 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
88 #define OBJECTS "OBJECTS"
89 #define ADMIN_USR "admin_quotafile_v2.usr"
90 #define ADMIN_GRP "admin_quotafile_v2.grp"
92 struct osd_directory {
93 struct iam_container od_container;
94 struct iam_descr od_descr;
98 * Object Index (oi) instance.
102 * underlying index object, where fid->id mapping in stored.
104 struct inode *oi_inode;
105 struct osd_directory oi_dir;
108 extern const int osd_dto_credits_noquota[];
111 struct dt_object oo_dt;
113 * Inode for file system object represented by this osd_object. This
114 * inode is pinned for the whole duration of lu_object life.
116 * Not modified concurrently (either setup early during object
117 * creation, or assigned by osd_object_create() under write lock).
119 struct inode *oo_inode;
121 * to protect index ops.
123 struct htree_lock_head *oo_hl_head;
124 struct rw_semaphore oo_ext_idx_sem;
125 struct rw_semaphore oo_sem;
126 struct osd_directory *oo_dir;
127 /** protects inode attributes. */
130 * Following two members are used to indicate the presence of dot and
131 * dotdot in the given directory. This is required for interop mode
134 int oo_compat_dot_created;
135 int oo_compat_dotdot_created;
137 const struct lu_env *oo_owner;
138 #ifdef CONFIG_LOCKDEP
139 struct lockdep_map oo_dep_map;
144 /* protects on-fly initialization */
145 int oos_subdir_count; /* subdir count for each seq */
146 struct dentry *oos_root; /* O/<seq> */
147 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
148 u64 oos_seq; /* seq number */
149 struct list_head oos_seq_list; /* list to seq_list */
153 struct dentry *om_root; /* dentry for /O */
154 rwlock_t om_seq_list_lock; /* lock for seq_list */
155 struct list_head om_seq_list; /* list head for seq */
157 struct mutex om_dir_init_mutex;
161 struct dentry *om_root; /* AGENT/<index> */
162 u64 om_index; /* mdt index */
163 struct list_head om_list; /* list to omm_list */
166 struct osd_mdobj_map {
167 struct dentry *omm_remote_parent;
170 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
171 ldiskfs_add_entry(handle, child, cinode, hlock)
173 #define OSD_OTABLE_IT_CACHE_SIZE 64
174 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
176 struct osd_inconsistent_item {
177 /* link into osd_scrub::os_inconsistent_items,
178 * protected by osd_scrub::os_lock. */
179 struct list_head oii_list;
181 /* The right FID <=> ino#/gen mapping. */
182 struct osd_idmap_cache oii_cache;
184 unsigned int oii_insert:1; /* insert or update mapping. */
187 struct osd_otable_cache {
188 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
190 /* Index for next cache slot to be filled. */
191 int ooc_producer_idx;
193 /* Index for next cache slot to be returned by it::next(). */
194 int ooc_consumer_idx;
196 /* How many items in ooc_cache. */
197 int ooc_cached_items;
199 /* Position for up layer LFSCK iteration pre-loading. */
200 __u32 ooc_pos_preload;
203 struct osd_otable_it {
204 struct osd_device *ooi_dev;
205 struct osd_otable_cache ooi_cache;
207 /* The following bits can be updated/checked w/o lock protection.
208 * If more bits will be introduced in the future and need lock to
209 * protect, please add comment. */
210 unsigned long ooi_used_outside:1, /* Some user out of OSD
211 * uses the iteration. */
212 ooi_all_cached:1, /* No more entries can be
213 * filled into cache. */
214 ooi_user_ready:1, /* The user out of OSD is
215 * ready to iterate. */
216 ooi_waiting:1; /* it::next is waiting. */
224 struct dt_device od_dt_dev;
225 /* information about underlying file system */
226 struct vfsmount *od_mnt;
228 struct osd_oi **od_oi_table;
229 /* total number of OI containers */
234 unsigned int od_fl_capa:1,
242 unsigned long od_capa_timeout;
244 __u32 od_dirent_journal;
245 struct lustre_capa_key *od_capa_keys;
246 struct hlist_head *od_capa_hash;
248 struct proc_dir_entry *od_proc_entry;
249 struct lprocfs_stats *od_stats;
251 spinlock_t od_osfs_lock;
254 struct lu_site od_site;
256 struct osd_obj_map *od_ost_map;
257 struct osd_mdobj_map *od_mdt_map;
259 unsigned long long od_readcache_max_filesize;
261 int od_writethrough_cache;
263 struct brw_stats od_brw_stats;
264 atomic_t od_r_in_flight;
265 atomic_t od_w_in_flight;
267 struct mutex od_otable_mutex;
268 struct osd_otable_it *od_otable_it;
269 struct osd_scrub od_scrub;
270 struct list_head od_ios_list;
272 /* service name associated with the osd device */
273 char od_svname[MAX_OBD_NAME];
274 char od_mntdev[MAX_OBD_NAME];
276 /* quota slave instance */
277 struct qsd_instance *od_quota_slave;
279 /* osd seq instance */
280 struct lu_client_seq *od_cl_seq;
281 /* If the ratio of "the total OI mappings count" vs
282 * "the bad OI mappings count" is lower than the
283 * osd_device::od_full_scrub_ratio, then trigger
284 * OI scrub to scan the whole the device. */
285 __u64 od_full_scrub_ratio;
286 /* If the speed of found bad OI mappings (per minute)
287 * exceeds the osd_device::od_full_scrub_threshold_rate,
288 * then trigger OI scrub to scan the whole device. */
289 __u64 od_full_scrub_threshold_rate;
292 enum osd_full_scrub_ratio {
293 /* Trigger OI scrub to scan the whole device directly. */
296 /* Because the bad OI mappings count cannot be larger than
297 * the total OI mappints count, then setting OFSR_NEVER means
298 * that the whole device scanning cannot be triggered by auto
299 * detected bad OI mappings during the RPC services. */
301 OFSR_DEFAULT = 10000,
304 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
306 /* There are at most 10 uid/gids are affected in a transaction, and
307 * that's rename case:
308 * - 2 for source parent uid & gid;
309 * - 2 for source child uid & gid ('..' entry update when child is directory);
310 * - 2 for target parent uid & gid;
311 * - 2 for target child uid & gid (if the target child exists);
312 * - 2 for root uid & gid (last_rcvd, llog, etc);
314 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
315 * the id type of each id in the ot_id_array.
317 #define OSD_MAX_UGID_CNT 10
322 OSD_OT_XATTR_SET = 2,
336 struct thandle ot_super;
338 struct ldiskfs_journal_cb_entry ot_jcb;
339 struct list_head ot_dcb_list;
340 /* Link to the device, for debugging. */
341 struct lu_ref_link ot_dev_link;
342 unsigned short ot_credits;
343 unsigned short ot_id_cnt;
344 unsigned short ot_id_type;
345 uid_t ot_id_array[OSD_MAX_UGID_CNT];
346 struct lquota_trans *ot_quota_trans;
347 #if OSD_THANDLE_STATS
348 /** time when this handle was allocated */
349 cfs_time_t oth_alloced;
351 /** time when this thanle was started */
352 cfs_time_t oth_started;
357 * Basic transaction credit op
378 #ifdef CONFIG_PROC_FS
380 LPROC_OSD_READ_BYTES = 0,
381 LPROC_OSD_WRITE_BYTES = 1,
382 LPROC_OSD_GET_PAGE = 2,
383 LPROC_OSD_NO_PAGE = 3,
384 LPROC_OSD_CACHE_ACCESS = 4,
385 LPROC_OSD_CACHE_HIT = 5,
386 LPROC_OSD_CACHE_MISS = 6,
388 #if OSD_THANDLE_STATS
389 LPROC_OSD_THANDLE_STARTING,
390 LPROC_OSD_THANDLE_OPEN,
391 LPROC_OSD_THANDLE_CLOSING,
398 * Storage representation for fids.
400 * Variable size, first byte contains the length of the whole record.
402 struct osd_fid_pack {
403 unsigned char fp_len;
404 char fp_area[sizeof(struct lu_fid)];
407 struct osd_it_ea_dirent {
408 struct lu_fid oied_fid;
411 unsigned short oied_namelen;
412 unsigned int oied_type;
414 } __attribute__((packed));
417 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
418 * than lu_dirent struct. osd readdir reads less number of dirent than
419 * required for mdd dir page. so buffer size need to be increased so that
420 * there would be one ext3 readdir for every mdd readdir page.
423 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
426 * This is iterator's in-memory data structure in interoperability
427 * mode (i.e. iterator over ldiskfs style directory)
430 struct osd_object *oie_obj;
431 /** used in ldiskfs iterator, to stored file pointer */
432 struct file oie_file;
433 /** how many entries have been read-cached from storage */
435 /** current entry is being iterated by caller */
437 /** current processing entry */
438 struct osd_it_ea_dirent *oie_dirent;
439 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
441 struct dentry oie_dentry;
445 * Iterator's in-memory data structure for IAM mode.
448 struct osd_object *oi_obj;
449 struct iam_path_descr *oi_ipd;
450 struct iam_iterator oi_it;
453 struct osd_quota_leaf {
454 struct list_head oql_link;
459 * Iterator's in-memory data structure for quota file.
461 struct osd_it_quota {
462 struct osd_object *oiq_obj;
463 /** tree blocks path to where the entry is stored */
464 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
465 /** on-disk offset for current key where quota record can be found */
467 /** identifier for current quota record */
469 /** the record index in the leaf/index block */
470 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
471 /** list of already processed leaf blocks */
472 struct list_head oiq_list;
475 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
478 wait_queue_head_t dr_wait;
479 atomic_t dr_numreqs; /* number of reqs being processed */
484 unsigned int dr_ignore_quota:1;
485 unsigned int dr_elapsed_valid:1; /* we really did count time */
486 unsigned int dr_rw:1;
487 struct lu_buf dr_pg_buf;
488 struct page **dr_pages;
489 struct lu_buf dr_bl_buf;
490 unsigned long *dr_blocks;
491 unsigned long dr_start_time;
492 unsigned long dr_elapsed; /* how long io took */
493 struct osd_device *dr_dev;
494 unsigned int dr_init_at; /* the line iobuf was initialized */
497 struct osd_thread_info {
498 const struct lu_env *oti_env;
500 * used for index operations.
502 struct dentry oti_obj_dentry;
503 struct dentry oti_child_dentry;
505 /** dentry for Iterator context. */
506 struct dentry oti_it_dentry;
507 struct htree_lock *oti_hlock;
509 struct lu_fid oti_fid;
510 struct lu_fid oti_fid2;
511 struct lu_fid oti_fid3;
512 struct osd_inode_id oti_id;
513 struct osd_inode_id oti_id2;
514 struct osd_inode_id oti_id3;
515 struct ost_id oti_ostid;
518 * XXX temporary: for ->i_op calls.
520 struct timespec oti_time;
522 * XXX temporary: fake struct file for osd_object_sync
524 struct file oti_file;
526 * XXX temporary: for capa operations.
528 struct lustre_capa_key oti_capa_key;
529 struct lustre_capa oti_capa;
531 /** osd_device reference, initialized in osd_trans_start() and
532 used in osd_trans_stop() */
533 struct osd_device *oti_dev;
536 * following ipd and it structures are used for osd_index_iam_lookup()
537 * these are defined separately as we might do index operation
538 * in open iterator session.
541 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
543 unsigned int oti_it_ea_buf_used:1;
545 struct kstatfs oti_ksfs;
547 /** IAM iterator for index operation. */
548 struct iam_iterator oti_idx_it;
550 /** union to guarantee that ->oti_ipd[] has proper alignment. */
552 char oti_it_ipd[DX_IPD_MAX_SIZE];
553 long long oti_alignment_lieutenant;
557 char oti_idx_ipd[DX_IPD_MAX_SIZE];
558 long long oti_alignment_lieutenant_colonel;
561 struct osd_idmap_cache oti_cache;
566 /** used in osd_fid_set() to put xattr */
567 struct lu_buf oti_buf;
568 struct lu_buf oti_big_buf;
569 /** used in osd_ea_fid_set() to set fid into common ea */
571 struct lustre_mdt_attrs oti_mdt_attrs;
572 /* old LMA for compatibility */
573 char oti_mdt_attrs_old[LMA_OLD_SIZE];
576 struct osd_iobuf oti_iobuf;
577 struct inode oti_inode;
578 #define OSD_FID_REC_SZ 32
579 char oti_ldp[OSD_FID_REC_SZ];
580 char oti_ldp2[OSD_FID_REC_SZ];
582 /* used by quota code */
584 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
585 struct fs_disk_quota oti_fdq;
587 struct if_dqblk oti_dqblk;
589 struct if_dqinfo oti_dqinfo;
591 struct lquota_id_info oti_qi;
592 struct lquota_trans oti_quota_trans;
593 union lquota_rec oti_quota_rec;
595 struct lu_seq_range oti_seq_range;
597 /* Tracking for transaction credits, to allow debugging and optimizing
598 * cases where a large number of credits are being allocated for
599 * single transaction. */
600 unsigned short oti_declare_ops[OSD_OT_MAX];
601 unsigned short oti_declare_ops_rb[OSD_OT_MAX];
602 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
607 struct filter_fid_old oti_ff;
608 struct filter_fid oti_ff_new;
612 extern int ldiskfs_pdo;
614 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
615 const char *name, void *buf, int len)
620 dentry->d_inode = inode;
621 dentry->d_sb = inode->i_sb;
622 return inode->i_op->getxattr(dentry, name, buf, len);
625 static inline int __osd_xattr_set(struct osd_thread_info *info,
626 struct inode *inode, const char *name,
627 const void *buf, int buflen, int fl)
629 struct dentry *dentry = &info->oti_child_dentry;
631 ll_vfs_dq_init(inode);
632 dentry->d_inode = inode;
633 dentry->d_sb = inode->i_sb;
634 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
637 #ifdef CONFIG_PROC_FS
639 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
640 extern struct lprocfs_vars lprocfs_osd_module_vars[];
641 int osd_procfs_init(struct osd_device *osd, const char *name);
642 int osd_procfs_fini(struct osd_device *osd);
643 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
644 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
645 int osd_register_proc_index_in_idif(struct osd_device *osd);
649 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
650 struct obd_statfs *sfs);
651 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
652 struct lustre_capa *capa, __u64 opc);
653 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
654 struct osd_inode_id *id);
655 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
656 const struct lu_fid *fid, __u32 compat, __u32 incompat);
657 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
658 struct dentry *dentry, struct lustre_mdt_attrs *lma);
659 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
660 struct osd_inode_id *id, const struct lu_fid *fid);
661 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
662 struct dentry *dentry, struct lu_fid *fid);
664 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
665 void osd_obj_map_fini(struct osd_device *dev);
666 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
667 const struct lu_fid *fid, struct osd_inode_id *id);
668 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
669 const struct lu_fid *fid, const struct osd_inode_id *id,
671 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
672 const struct lu_fid *fid, handle_t *th);
673 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
674 const struct lu_fid *fid, const struct osd_inode_id *id,
676 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
677 struct inode *src_parent, struct dentry *src_child,
678 const struct lu_fid *fid);
679 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
680 const struct lu_fid *fid, struct osd_inode_id *id);
681 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
682 const struct lu_fid *fid, const struct osd_inode_id *id,
684 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
685 const struct lu_fid *fid, const struct osd_inode_id *id,
688 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
689 int osd_scrub_file_store(struct osd_scrub *scrub);
690 char *osd_lf_fid2name(const struct lu_fid *fid);
691 int osd_scrub_start(struct osd_device *dev, __u32 flags);
692 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
693 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
694 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
696 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
697 struct osd_inode_id *id);
698 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
700 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
701 u64 seq, struct lu_seq_range *range);
703 int osd_delete_from_remote_parent(const struct lu_env *env,
704 struct osd_device *osd,
705 struct osd_object *obj,
706 struct osd_thandle *oh);
707 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
708 struct osd_object *obj, struct osd_thandle *oh);
709 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
710 struct osd_device *osd,
711 const struct lu_fid *fid,
712 struct osd_inode_id *id);
714 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
716 /* osd_quota_fmt.c */
717 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
718 int type, uint blk, int depth, uint index,
719 struct osd_it_quota *it);
720 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
721 int type, uint blk, uint index,
722 struct osd_it_quota *it);
723 loff_t find_tree_dqentry(const struct lu_env *env,
724 struct osd_object *obj, int type,
725 qid_t dqid, uint blk, int depth,
726 struct osd_it_quota *it);
728 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
729 struct lquota_id_info *qi, struct osd_object *obj,
730 bool enforce, int *flags);
731 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
732 long long space, struct osd_thandle *oh,
733 struct osd_object *obj, bool is_blk, int *flags,
735 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
736 const struct dt_rec *rec,
737 union lquota_rec *quota_rec);
738 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
739 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
740 const struct dt_index_features *feat);
742 static inline bool is_quota_glb_feat(const struct dt_index_features *feat)
744 return (feat == &dt_quota_iusr_features ||
745 feat == &dt_quota_busr_features ||
746 feat == &dt_quota_igrp_features ||
747 feat == &dt_quota_bgrp_features) ? true : false;
750 #ifndef HAVE_I_UID_READ
751 static inline uid_t i_uid_read(const struct inode *inode)
756 static inline gid_t i_gid_read(const struct inode *inode)
761 static inline void i_uid_write(struct inode *inode, uid_t uid)
766 static inline void i_gid_write(struct inode *inode, gid_t gid)
773 * Invariants, assertions.
777 * XXX: do not enable this, until invariant checking code is made thread safe
778 * in the face of pdirops locking.
780 #define OSD_INVARIANT_CHECKS (0)
782 #if OSD_INVARIANT_CHECKS
783 static inline int osd_invariant(const struct osd_object *obj)
787 ergo(obj->oo_inode != NULL,
788 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
789 atomic_read(&obj->oo_inode->i_count) > 0) &&
790 ergo(obj->oo_dir != NULL &&
791 obj->oo_dir->od_conationer.ic_object != NULL,
792 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
795 #define osd_invariant(obj) (1)
798 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
800 extern const struct dt_index_operations osd_otable_ops;
802 static inline int osd_oi_fid2idx(struct osd_device *dev,
803 const struct lu_fid *fid)
805 return fid->f_seq & (dev->od_oi_count - 1);
808 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
809 const struct lu_fid *fid)
811 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
812 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
813 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
814 DFID"\n", PFID(fid));
815 /* It can work even od_oi_count equals to 1 although it's unexpected,
816 * the only reason we set it to 1 is for performance measurement */
817 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
820 extern const struct lu_device_operations osd_lu_ops;
822 static inline int lu_device_is_osd(const struct lu_device *d)
824 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
827 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
829 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
830 return container_of0(d, struct osd_device, od_dt_dev);
833 static inline struct osd_device *osd_dev(const struct lu_device *d)
835 LASSERT(lu_device_is_osd(d));
836 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
839 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
841 return osd_dev(o->oo_dt.do_lu.lo_dev);
844 static inline struct super_block *osd_sb(const struct osd_device *dev)
846 return dev->od_mnt->mnt_sb;
849 static inline int osd_object_is_root(const struct osd_object *obj)
851 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
854 static inline struct osd_object *osd_obj(const struct lu_object *o)
856 LASSERT(lu_device_is_osd(o->lo_dev));
857 return container_of0(o, struct osd_object, oo_dt.do_lu);
860 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
862 return osd_obj(&d->do_lu);
865 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
867 return &osd->od_dt_dev.dd_lu_dev;
870 static inline journal_t *osd_journal(const struct osd_device *dev)
872 return LDISKFS_SB(osd_sb(dev))->s_journal;
875 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
877 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
880 static inline char *osd_name(struct osd_device *osd)
882 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
885 extern const struct dt_body_operations osd_body_ops;
886 extern struct lu_context_key osd_key;
888 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
890 return lu_context_key_get(&env->le_ctx, &osd_key);
893 extern const struct dt_body_operations osd_body_ops_new;
899 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
900 const struct iam_container *bag)
902 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
903 osd_oti_get(env)->oti_it_ipd);
907 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
908 const struct iam_container *bag)
910 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
911 osd_oti_get(env)->oti_idx_ipd);
914 static inline void osd_ipd_put(const struct lu_env *env,
915 const struct iam_container *bag,
916 struct iam_path_descr *ipd)
918 bag->ic_descr->id_ops->id_ipd_free(ipd);
921 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
922 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
923 int write_NUL, loff_t *offs, handle_t *handle);
926 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
928 const char *name, const int namelen)
930 struct osd_thread_info *info = osd_oti_get(env);
931 struct dentry *child_dentry = &info->oti_child_dentry;
932 struct dentry *obj_dentry = &info->oti_obj_dentry;
934 obj_dentry->d_inode = inode;
935 obj_dentry->d_sb = inode->i_sb;
936 obj_dentry->d_name.hash = 0;
938 child_dentry->d_name.hash = 0;
939 child_dentry->d_parent = obj_dentry;
940 child_dentry->d_name.name = name;
941 child_dentry->d_name.len = namelen;
945 extern int osd_trans_declare_op2rb[];
946 extern int ldiskfs_track_declares_assert;
948 static inline void osd_trans_declare_op(const struct lu_env *env,
949 struct osd_thandle *oh,
950 unsigned int op, int credits)
952 struct osd_thread_info *oti = osd_oti_get(env);
954 LASSERT(oh->ot_handle == NULL);
955 if (unlikely(op >= OSD_OT_MAX)) {
956 if (unlikely(ldiskfs_track_declares_assert)) {
957 LASSERT(op < OSD_OT_MAX);
959 CWARN("%s: Invalid operation index %d\n",
960 osd_name(oti->oti_dev), op);
961 libcfs_debug_dumpstack(NULL);
964 oti->oti_declare_ops[op]++;
965 oti->oti_declare_ops_cred[op] += credits;
967 oh->ot_credits += credits;
970 static inline void osd_trans_exec_op(const struct lu_env *env,
971 struct thandle *th, unsigned int op)
973 struct osd_thread_info *oti = osd_oti_get(env);
974 struct osd_thandle *oh = container_of(th, struct osd_thandle,
978 LASSERT(oh->ot_handle != NULL);
979 if (unlikely(op >= OSD_OT_MAX)) {
980 if (unlikely(ldiskfs_track_declares_assert))
981 LASSERT(op < OSD_OT_MAX);
983 CWARN("%s: Invalid operation index %d\n",
984 osd_name(oti->oti_dev), op);
985 libcfs_debug_dumpstack(NULL);
990 if (likely(!oti->oti_rollback && oti->oti_declare_ops[op] > 0)) {
991 oti->oti_declare_ops[op]--;
992 oti->oti_declare_ops_rb[op]++;
994 /* all future updates are considered rollback */
995 oti->oti_rollback = true;
996 rb = osd_trans_declare_op2rb[op];
997 if (unlikely(rb >= OSD_OT_MAX)) {
998 if (unlikely(ldiskfs_track_declares_assert))
999 LASSERTF(rb < OSD_OT_MAX, "rb = %u\n", rb);
1001 CWARN("%s: Invalid rollback index %d\n",
1002 osd_name(oti->oti_dev), rb);
1003 libcfs_debug_dumpstack(NULL);
1007 if (unlikely(oti->oti_declare_ops_rb[rb] == 0)) {
1008 if (unlikely(ldiskfs_track_declares_assert))
1009 LASSERTF(oti->oti_declare_ops_rb[rb] > 0,
1012 CWARN("%s: Overflow in tracking declares for "
1014 osd_name(oti->oti_dev), rb);
1015 libcfs_debug_dumpstack(NULL);
1019 oti->oti_declare_ops_rb[rb]--;
1023 static inline void osd_trans_declare_rb(const struct lu_env *env,
1024 struct thandle *th, unsigned int op)
1026 struct osd_thread_info *oti = osd_oti_get(env);
1027 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1030 LASSERT(oh->ot_handle != NULL);
1031 if (unlikely(op >= OSD_OT_MAX)) {
1032 if (unlikely(ldiskfs_track_declares_assert))
1033 LASSERT(op < OSD_OT_MAX);
1035 CWARN("%s: Invalid operation index %d\n",
1036 osd_name(oti->oti_dev), op);
1037 libcfs_debug_dumpstack(NULL);
1041 oti->oti_declare_ops_rb[op]++;
1046 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1049 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1050 struct lu_fid *befider)
1052 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1053 memcpy(pack->fp_area, befider, sizeof(*befider));
1054 pack->fp_len = sizeof(*befider) + 1;
1058 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1063 switch (pack->fp_len) {
1064 case sizeof *fid + 1:
1065 memcpy(fid, pack->fp_area, sizeof *fid);
1066 fid_be_to_cpu(fid, fid);
1069 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1076 * Quota/Accounting handling
1078 extern const struct dt_index_operations osd_acct_index_ops;
1079 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1080 const struct lu_fid *fid, struct osd_inode_id *id);
1082 /* copy from fs/ext4/dir.c */
1083 static inline int is_32bit_api(void)
1085 #ifdef CONFIG_COMPAT
1086 return is_compat_task();
1088 return (BITS_PER_LONG == 32);
1092 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1094 if ((filp->f_mode & FMODE_32BITHASH) ||
1095 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1096 return LDISKFS_HTREE_EOF_32BIT;
1098 return LDISKFS_HTREE_EOF_64BIT;
1101 static inline int fid_is_internal(const struct lu_fid *fid)
1103 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1106 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1108 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1111 #ifdef JOURNAL_START_HAS_3ARGS
1112 # define osd_journal_start_sb(sb, type, nblock) \
1113 ldiskfs_journal_start_sb(sb, type, nblock)
1114 # define osd_ldiskfs_append(handle, inode, nblock, err) \
1115 ldiskfs_append(handle, inode, nblock)
1116 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1117 ldiskfs_find_entry(dir, name, de, inlined, lock)
1118 # define osd_journal_start(inode, type, nblocks) \
1119 ldiskfs_journal_start(inode, type, nblocks)
1120 # define osd_transaction_size(dev) \
1121 (osd_journal(dev)->j_max_transaction_buffers / 2)
1123 # define LDISKFS_HT_MISC 0
1124 # define osd_journal_start_sb(sb, type, nblock) \
1125 ldiskfs_journal_start_sb(sb, nblock)
1126 # define osd_ldiskfs_append(handle, inode, nblock, err) \
1127 ldiskfs_append(handle, inode, nblock, err)
1128 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1129 ldiskfs_find_entry(dir, name, de, lock)
1130 # define osd_journal_start(inode, type, nblocks) \
1131 ldiskfs_journal_start(inode, nblocks)
1132 # define osd_transaction_size(dev) \
1133 (osd_journal(dev)->j_max_transaction_buffers)
1136 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1137 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1139 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1141 #endif /* _OSD_INTERNAL_H */