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 __u32 od_dirent_journal;
244 struct proc_dir_entry *od_proc_entry;
245 struct lprocfs_stats *od_stats;
247 spinlock_t od_osfs_lock;
250 struct lu_site od_site;
252 struct osd_obj_map *od_ost_map;
253 struct osd_mdobj_map *od_mdt_map;
255 unsigned long long od_readcache_max_filesize;
257 int od_writethrough_cache;
259 struct brw_stats od_brw_stats;
260 atomic_t od_r_in_flight;
261 atomic_t od_w_in_flight;
263 struct mutex od_otable_mutex;
264 struct osd_otable_it *od_otable_it;
265 struct osd_scrub od_scrub;
266 struct list_head od_ios_list;
268 /* service name associated with the osd device */
269 char od_svname[MAX_OBD_NAME];
270 char od_mntdev[MAX_OBD_NAME];
272 /* quota slave instance */
273 struct qsd_instance *od_quota_slave;
275 /* osd seq instance */
276 struct lu_client_seq *od_cl_seq;
277 /* If the ratio of "the total OI mappings count" vs
278 * "the bad OI mappings count" is lower than the
279 * osd_device::od_full_scrub_ratio, then trigger
280 * OI scrub to scan the whole the device. */
281 __u64 od_full_scrub_ratio;
282 /* If the speed of found bad OI mappings (per minute)
283 * exceeds the osd_device::od_full_scrub_threshold_rate,
284 * then trigger OI scrub to scan the whole device. */
285 __u64 od_full_scrub_threshold_rate;
288 enum osd_full_scrub_ratio {
289 /* Trigger OI scrub to scan the whole device directly. */
292 /* Because the bad OI mappings count cannot be larger than
293 * the total OI mappints count, then setting OFSR_NEVER means
294 * that the whole device scanning cannot be triggered by auto
295 * detected bad OI mappings during the RPC services. */
297 OFSR_DEFAULT = 10000,
300 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
302 /* There are at most 10 uid/gids are affected in a transaction, and
303 * that's rename case:
304 * - 2 for source parent uid & gid;
305 * - 2 for source child uid & gid ('..' entry update when child is directory);
306 * - 2 for target parent uid & gid;
307 * - 2 for target child uid & gid (if the target child exists);
308 * - 2 for root uid & gid (last_rcvd, llog, etc);
310 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
311 * the id type of each id in the ot_id_array.
313 #define OSD_MAX_UGID_CNT 10
318 OSD_OT_XATTR_SET = 2,
331 struct thandle ot_super;
333 struct ldiskfs_journal_cb_entry ot_jcb;
334 struct list_head ot_commit_dcb_list;
335 struct list_head ot_stop_dcb_list;
336 /* Link to the device, for debugging. */
337 struct lu_ref_link ot_dev_link;
338 unsigned short ot_credits;
339 unsigned short ot_id_cnt;
340 unsigned short ot_id_type;
341 uid_t ot_id_array[OSD_MAX_UGID_CNT];
342 struct lquota_trans *ot_quota_trans;
343 #if OSD_THANDLE_STATS
344 /** time when this handle was allocated */
345 cfs_time_t oth_alloced;
347 /** time when this thanle was started */
348 cfs_time_t oth_started;
353 * Basic transaction credit op
374 #ifdef CONFIG_PROC_FS
376 LPROC_OSD_READ_BYTES = 0,
377 LPROC_OSD_WRITE_BYTES = 1,
378 LPROC_OSD_GET_PAGE = 2,
379 LPROC_OSD_NO_PAGE = 3,
380 LPROC_OSD_CACHE_ACCESS = 4,
381 LPROC_OSD_CACHE_HIT = 5,
382 LPROC_OSD_CACHE_MISS = 6,
384 #if OSD_THANDLE_STATS
385 LPROC_OSD_THANDLE_STARTING,
386 LPROC_OSD_THANDLE_OPEN,
387 LPROC_OSD_THANDLE_CLOSING,
394 * Storage representation for fids.
396 * Variable size, first byte contains the length of the whole record.
398 struct osd_fid_pack {
399 unsigned char fp_len;
400 char fp_area[sizeof(struct lu_fid)];
403 struct osd_it_ea_dirent {
404 struct lu_fid oied_fid;
407 unsigned short oied_namelen;
408 unsigned int oied_type;
410 } __attribute__((packed));
413 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
414 * than lu_dirent struct. osd readdir reads less number of dirent than
415 * required for mdd dir page. so buffer size need to be increased so that
416 * there would be one ext3 readdir for every mdd readdir page.
419 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
422 * This is iterator's in-memory data structure in interoperability
423 * mode (i.e. iterator over ldiskfs style directory)
426 struct osd_object *oie_obj;
427 /** used in ldiskfs iterator, to stored file pointer */
428 struct file oie_file;
429 /** how many entries have been read-cached from storage */
431 /** current entry is being iterated by caller */
433 /** current processing entry */
434 struct osd_it_ea_dirent *oie_dirent;
435 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
437 struct dentry oie_dentry;
441 * Iterator's in-memory data structure for IAM mode.
444 struct osd_object *oi_obj;
445 struct iam_path_descr *oi_ipd;
446 struct iam_iterator oi_it;
449 struct osd_quota_leaf {
450 struct list_head oql_link;
455 * Iterator's in-memory data structure for quota file.
457 struct osd_it_quota {
458 struct osd_object *oiq_obj;
459 /** tree blocks path to where the entry is stored */
460 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
461 /** on-disk offset for current key where quota record can be found */
463 /** identifier for current quota record */
465 /** the record index in the leaf/index block */
466 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
467 /** list of already processed leaf blocks */
468 struct list_head oiq_list;
471 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
474 wait_queue_head_t dr_wait;
475 atomic_t dr_numreqs; /* number of reqs being processed */
480 unsigned int dr_ignore_quota:1;
481 unsigned int dr_elapsed_valid:1; /* we really did count time */
482 unsigned int dr_rw:1;
483 struct lu_buf dr_pg_buf;
484 struct page **dr_pages;
485 struct lu_buf dr_bl_buf;
487 unsigned long dr_start_time;
488 unsigned long dr_elapsed; /* how long io took */
489 struct osd_device *dr_dev;
490 unsigned int dr_init_at; /* the line iobuf was initialized */
493 struct osd_thread_info {
494 const struct lu_env *oti_env;
496 * used for index operations.
498 struct dentry oti_obj_dentry;
499 struct dentry oti_child_dentry;
501 /** dentry for Iterator context. */
502 struct dentry oti_it_dentry;
505 /* fake struct file for osd_object_sync */
506 struct file oti_file;
508 struct kstatfs oti_ksfs;
511 struct htree_lock *oti_hlock;
513 struct lu_fid oti_fid;
514 struct lu_fid oti_fid2;
515 struct lu_fid oti_fid3;
516 struct osd_inode_id oti_id;
517 struct osd_inode_id oti_id2;
518 struct osd_inode_id oti_id3;
519 struct ost_id oti_ostid;
522 * XXX temporary: for ->i_op calls.
524 struct timespec oti_time;
526 /** osd_device reference, initialized in osd_trans_start() and
527 used in osd_trans_stop() */
528 struct osd_device *oti_dev;
531 * following ipd and it structures are used for osd_index_iam_lookup()
532 * these are defined separately as we might do index operation
533 * in open iterator session.
536 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
538 unsigned int oti_it_ea_buf_used:1;
540 /* IAM iterator for index operation. */
541 struct iam_iterator oti_idx_it;
543 /** union to guarantee that ->oti_ipd[] has proper alignment. */
546 char oti_it_ipd[DX_IPD_MAX_SIZE];
547 long long oti_alignment_lieutenant;
551 char oti_idx_ipd[DX_IPD_MAX_SIZE];
552 long long oti_alignment_lieutenant_colonel;
555 struct osd_idmap_cache oti_cache;
560 /** used in osd_fid_set() to put xattr */
561 struct lu_buf oti_buf;
562 struct lu_buf oti_big_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];
568 struct filter_fid_old oti_ff;
569 struct filter_fid oti_ff_new;
572 struct osd_iobuf oti_iobuf;
573 /* used to access objects in /O */
574 struct inode *oti_inode;
575 #define OSD_FID_REC_SZ 32
576 char oti_ldp[OSD_FID_REC_SZ];
577 char oti_ldp2[OSD_FID_REC_SZ];
579 /* used by quota code */
581 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
582 struct fs_disk_quota oti_fdq;
584 struct if_dqblk oti_dqblk;
586 struct if_dqinfo oti_dqinfo;
588 struct lquota_id_info oti_qi;
589 struct lquota_trans oti_quota_trans;
590 union lquota_rec oti_quota_rec;
592 struct lu_seq_range oti_seq_range;
594 /* Tracking for transaction credits, to allow debugging and optimizing
595 * cases where a large number of credits are being allocated for
596 * single transaction. */
597 unsigned int oti_credits_before;
598 unsigned short oti_declare_ops[OSD_OT_MAX];
599 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
600 unsigned short oti_declare_ops_used[OSD_OT_MAX];
603 extern int ldiskfs_pdo;
605 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
606 const char *name, void *buf, int len)
611 dentry->d_inode = inode;
612 dentry->d_sb = inode->i_sb;
613 return inode->i_op->getxattr(dentry, name, buf, len);
616 static inline int __osd_xattr_set(struct osd_thread_info *info,
617 struct inode *inode, const char *name,
618 const void *buf, int buflen, int fl)
620 struct dentry *dentry = &info->oti_child_dentry;
622 ll_vfs_dq_init(inode);
623 dentry->d_inode = inode;
624 dentry->d_sb = inode->i_sb;
625 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
628 #ifdef CONFIG_PROC_FS
630 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
631 extern struct lprocfs_vars lprocfs_osd_module_vars[];
632 int osd_procfs_init(struct osd_device *osd, const char *name);
633 int osd_procfs_fini(struct osd_device *osd);
634 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
635 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
636 int osd_register_proc_index_in_idif(struct osd_device *osd);
640 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
641 struct obd_statfs *sfs);
642 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
643 struct osd_inode_id *id);
644 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
645 const struct lu_fid *fid, __u32 compat, __u32 incompat);
646 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
647 struct dentry *dentry, struct lustre_mdt_attrs *lma);
648 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
649 struct osd_inode_id *id, const struct lu_fid *fid);
650 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
651 struct dentry *dentry, struct lu_fid *fid);
653 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
654 void osd_obj_map_fini(struct osd_device *dev);
655 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
656 const struct lu_fid *fid, struct osd_inode_id *id);
657 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
658 const struct lu_fid *fid, const struct osd_inode_id *id,
660 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
661 const struct lu_fid *fid, handle_t *th);
662 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
663 const struct lu_fid *fid, const struct osd_inode_id *id,
665 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
666 struct inode *src_parent, struct dentry *src_child,
667 const struct lu_fid *fid);
668 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
669 const struct lu_fid *fid, struct osd_inode_id *id);
670 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
671 const struct lu_fid *fid, const struct osd_inode_id *id,
673 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
674 const struct lu_fid *fid, const struct osd_inode_id *id,
677 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
678 int osd_scrub_file_store(struct osd_scrub *scrub);
679 char *osd_lf_fid2name(const struct lu_fid *fid);
680 int osd_scrub_start(struct osd_device *dev, __u32 flags);
681 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
682 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
683 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
685 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
686 struct osd_inode_id *id);
687 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
689 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
690 u64 seq, struct lu_seq_range *range);
692 int osd_delete_from_remote_parent(const struct lu_env *env,
693 struct osd_device *osd,
694 struct osd_object *obj,
695 struct osd_thandle *oh);
696 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
697 struct osd_object *obj, struct osd_thandle *oh);
698 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
699 struct osd_device *osd,
700 const struct lu_fid *fid,
701 struct osd_inode_id *id);
703 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
705 /* osd_quota_fmt.c */
706 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
707 int type, uint blk, int depth, uint index,
708 struct osd_it_quota *it);
709 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
710 int type, uint blk, uint index,
711 struct osd_it_quota *it);
712 loff_t find_tree_dqentry(const struct lu_env *env,
713 struct osd_object *obj, int type,
714 qid_t dqid, uint blk, int depth,
715 struct osd_it_quota *it);
717 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
718 struct lquota_id_info *qi, struct osd_object *obj,
719 bool enforce, int *flags);
720 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
721 long long space, struct osd_thandle *oh,
722 struct osd_object *obj, bool is_blk, int *flags,
724 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
725 const struct dt_rec *rec,
726 union lquota_rec *quota_rec);
727 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
728 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
730 #ifndef HAVE_I_UID_READ
731 static inline uid_t i_uid_read(const struct inode *inode)
736 static inline gid_t i_gid_read(const struct inode *inode)
741 static inline void i_uid_write(struct inode *inode, uid_t uid)
746 static inline void i_gid_write(struct inode *inode, gid_t gid)
752 #ifdef LDISKFS_HT_MISC
753 # define osd_journal_start_sb(sb, type, nblock) \
754 ldiskfs_journal_start_sb(sb, type, nblock)
755 # define osd_ldiskfs_append(handle, inode, nblock, err) \
756 ldiskfs_append(handle, inode, nblock)
757 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
758 __ldiskfs_find_entry(dir, name, de, inlined, lock)
759 # define osd_journal_start(inode, type, nblocks) \
760 ldiskfs_journal_start(inode, type, nblocks)
761 # define osd_transaction_size(dev) \
762 (osd_journal(dev)->j_max_transaction_buffers / 2)
764 # define LDISKFS_HT_MISC 0
765 # define osd_journal_start_sb(sb, type, nblock) \
766 ldiskfs_journal_start_sb(sb, nblock)
767 # define osd_ldiskfs_append(handle, inode, nblock, err) \
768 ldiskfs_append(handle, inode, nblock, err)
769 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
770 __ldiskfs_find_entry(dir, name, de, lock)
771 # define osd_journal_start(inode, type, nblocks) \
772 ldiskfs_journal_start(inode, nblocks)
773 # define osd_transaction_size(dev) \
774 (osd_journal(dev)->j_max_transaction_buffers)
778 * Invariants, assertions.
782 * XXX: do not enable this, until invariant checking code is made thread safe
783 * in the face of pdirops locking.
785 #define OSD_INVARIANT_CHECKS (0)
787 #if OSD_INVARIANT_CHECKS
788 static inline int osd_invariant(const struct osd_object *obj)
792 ergo(obj->oo_inode != NULL,
793 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
794 atomic_read(&obj->oo_inode->i_count) > 0) &&
795 ergo(obj->oo_dir != NULL &&
796 obj->oo_dir->od_conationer.ic_object != NULL,
797 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
800 #define osd_invariant(obj) (1)
803 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
805 extern const struct dt_index_operations osd_otable_ops;
807 static inline int osd_oi_fid2idx(struct osd_device *dev,
808 const struct lu_fid *fid)
810 return fid->f_seq & (dev->od_oi_count - 1);
813 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
814 const struct lu_fid *fid)
816 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
817 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
818 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
819 DFID"\n", PFID(fid));
820 /* It can work even od_oi_count equals to 1 although it's unexpected,
821 * the only reason we set it to 1 is for performance measurement */
822 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
825 extern const struct lu_device_operations osd_lu_ops;
827 static inline int lu_device_is_osd(const struct lu_device *d)
829 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
832 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
834 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
835 return container_of0(d, struct osd_device, od_dt_dev);
838 static inline struct osd_device *osd_dev(const struct lu_device *d)
840 LASSERT(lu_device_is_osd(d));
841 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
844 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
846 return osd_dev(o->oo_dt.do_lu.lo_dev);
849 static inline struct super_block *osd_sb(const struct osd_device *dev)
851 return dev->od_mnt->mnt_sb;
854 static inline int osd_object_is_root(const struct osd_object *obj)
856 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
859 static inline struct osd_object *osd_obj(const struct lu_object *o)
861 LASSERT(lu_device_is_osd(o->lo_dev));
862 return container_of0(o, struct osd_object, oo_dt.do_lu);
865 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
867 return osd_obj(&d->do_lu);
870 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
872 return &osd->od_dt_dev.dd_lu_dev;
875 static inline journal_t *osd_journal(const struct osd_device *dev)
877 return LDISKFS_SB(osd_sb(dev))->s_journal;
880 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
882 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
885 static inline char *osd_name(struct osd_device *osd)
887 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
890 extern const struct dt_body_operations osd_body_ops;
891 extern struct lu_context_key osd_key;
893 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
895 return lu_context_key_get(&env->le_ctx, &osd_key);
898 extern const struct dt_body_operations osd_body_ops_new;
904 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
905 const struct iam_container *bag)
907 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
908 osd_oti_get(env)->oti_it_ipd);
912 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
913 const struct iam_container *bag)
915 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
916 osd_oti_get(env)->oti_idx_ipd);
919 static inline void osd_ipd_put(const struct lu_env *env,
920 const struct iam_container *bag,
921 struct iam_path_descr *ipd)
923 bag->ic_descr->id_ops->id_ipd_free(ipd);
926 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
927 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
928 int write_NUL, loff_t *offs, handle_t *handle);
931 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
933 const char *name, const int namelen)
935 struct osd_thread_info *info = osd_oti_get(env);
936 struct dentry *child_dentry = &info->oti_child_dentry;
937 struct dentry *obj_dentry = &info->oti_obj_dentry;
939 obj_dentry->d_inode = inode;
940 obj_dentry->d_sb = inode->i_sb;
941 obj_dentry->d_name.hash = 0;
943 child_dentry->d_name.hash = 0;
944 child_dentry->d_parent = obj_dentry;
945 child_dentry->d_name.name = name;
946 child_dentry->d_name.len = namelen;
950 extern int osd_trans_declare_op2rb[];
951 extern int ldiskfs_track_declares_assert;
952 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
954 static inline void osd_trans_declare_op(const struct lu_env *env,
955 struct osd_thandle *oh,
956 unsigned int op, int credits)
958 struct osd_thread_info *oti = osd_oti_get(env);
960 LASSERT(oh->ot_handle == NULL);
961 if (unlikely(op >= OSD_OT_MAX)) {
962 if (unlikely(ldiskfs_track_declares_assert)) {
963 LASSERT(op < OSD_OT_MAX);
965 CWARN("%s: Invalid operation index %d\n",
966 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
967 libcfs_debug_dumpstack(NULL);
970 oti->oti_declare_ops[op]++;
971 oti->oti_declare_ops_cred[op] += credits;
973 oh->ot_credits += credits;
976 static inline void osd_trans_exec_op(const struct lu_env *env,
977 struct thandle *th, unsigned int op)
979 struct osd_thread_info *oti = osd_oti_get(env);
980 struct osd_thandle *oh = container_of(th, struct osd_thandle,
982 unsigned int rb, left;
984 LASSERT(oh->ot_handle != NULL);
985 if (unlikely(op >= OSD_OT_MAX)) {
986 if (unlikely(ldiskfs_track_declares_assert))
987 LASSERT(op < OSD_OT_MAX);
989 CWARN("%s: Invalid operation index %d\n",
990 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
991 libcfs_debug_dumpstack(NULL);
996 /* find rollback (or reverse) operation for the given one
997 * such an operation doesn't require additional credits
998 * as the same set of blocks are modified */
999 rb = osd_trans_declare_op2rb[op];
1001 /* check whether credits for this operation were reserved at all */
1002 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1003 oti->oti_declare_ops_cred[rb] == 0)) {
1004 /* the API is not perfect yet: CREATE does REF_ADD internally
1005 * while DESTROY does not. To rollback CREATE the callers
1006 * needs to call REF_DEL+DESTROY which is hard to detect using
1007 * a simple table of rollback operations */
1008 if (op == OSD_OT_REF_DEL &&
1009 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1011 if (op == OSD_OT_REF_ADD &&
1012 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1014 osd_trans_dump_creds(env, th);
1015 CERROR("%s: op = %d, rb = %d\n",
1016 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1017 if (unlikely(ldiskfs_track_declares_assert))
1022 /* remember how many credits we have unused before the operation */
1023 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1024 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1025 if (unlikely(oti->oti_credits_before < left)) {
1026 osd_trans_dump_creds(env, th);
1027 CERROR("%s: op = %d, rb = %d\n",
1028 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1029 /* on a very small fs (testing?) it's possible that
1030 * the transaction can't fit 1/4 of journal, so we
1031 * just request less credits (see osd_trans_start()).
1032 * ignore the same case here */
1033 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1034 if (unlikely(oh->ot_credits < rb)) {
1035 if (unlikely(ldiskfs_track_declares_assert))
1041 static inline void osd_trans_exec_check(const struct lu_env *env,
1045 struct osd_thread_info *oti = osd_oti_get(env);
1046 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1048 int used, over, quota;
1050 /* how many credits have been used by the operation */
1051 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1053 if (unlikely(used < 0)) {
1054 /* if some block was allocated and released in the same
1055 * transaction, then it won't be a part of the transaction
1056 * and delta can be negative */
1061 /* rollback operations (e.g. when we destroy just created
1062 * object) should not consume any credits. there is no point
1063 * to confuse the checks below */
1067 oti->oti_declare_ops_used[op] += used;
1068 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1071 /* we account quota for a whole transaction and any operation can
1072 * consume corresponding credits */
1073 over = oti->oti_declare_ops_used[op] -
1074 oti->oti_declare_ops_cred[op];
1075 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1076 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1077 if (over <= quota) {
1078 /* probably that credits were consumed by
1079 * quota indirectly (in the depths of ldiskfs) */
1080 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1081 oti->oti_declare_ops_used[op] -= over;
1083 CWARN("op %d: used %u, used now %u, reserved %u\n",
1084 op, oti->oti_declare_ops_used[op], used,
1085 oti->oti_declare_ops_cred[op]);
1086 osd_trans_dump_creds(env, th);
1087 if (unlikely(ldiskfs_track_declares_assert))
1093 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1096 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1097 struct lu_fid *befider)
1099 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1100 memcpy(pack->fp_area, befider, sizeof(*befider));
1101 pack->fp_len = sizeof(*befider) + 1;
1105 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1110 switch (pack->fp_len) {
1111 case sizeof *fid + 1:
1112 memcpy(fid, pack->fp_area, sizeof *fid);
1113 fid_be_to_cpu(fid, fid);
1116 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1123 * Quota/Accounting handling
1125 extern const struct dt_index_operations osd_acct_index_ops;
1126 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1127 const struct lu_fid *fid, struct osd_inode_id *id);
1129 /* copy from fs/ext4/dir.c */
1130 static inline int is_32bit_api(void)
1132 #ifdef CONFIG_COMPAT
1133 return is_compat_task();
1135 return (BITS_PER_LONG == 32);
1139 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1141 if ((filp->f_mode & FMODE_32BITHASH) ||
1142 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1143 return LDISKFS_HTREE_EOF_32BIT;
1145 return LDISKFS_HTREE_EOF_64BIT;
1148 static inline int fid_is_internal(const struct lu_fid *fid)
1150 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1153 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1155 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1158 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1159 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1161 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1163 #endif /* _OSD_INTERNAL_H */