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, 2015, 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
82 #define LDISKFS_STATE_LUSTRE_DESTROY 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 /* Statfs space reservation for fragmentation and local objects */
94 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
95 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
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 __u32 oo_destroyed:1;
137 /* the i_flags in LMA */
140 * Following two members are used to indicate the presence of dot and
141 * dotdot in the given directory. This is required for interop mode
144 int oo_compat_dot_created;
145 int oo_compat_dotdot_created;
147 const struct lu_env *oo_owner;
148 #ifdef CONFIG_LOCKDEP
149 struct lockdep_map oo_dep_map;
154 /* protects on-fly initialization */
155 int oos_subdir_count; /* subdir count for each seq */
156 struct dentry *oos_root; /* O/<seq> */
157 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
158 u64 oos_seq; /* seq number */
159 struct list_head oos_seq_list; /* list to seq_list */
163 struct dentry *om_root; /* dentry for /O */
164 rwlock_t om_seq_list_lock; /* lock for seq_list */
165 struct list_head om_seq_list; /* list head for seq */
167 struct mutex om_dir_init_mutex;
171 struct dentry *om_root; /* AGENT/<index> */
172 u64 om_index; /* mdt index */
173 struct list_head om_list; /* list to omm_list */
176 struct osd_mdobj_map {
177 struct dentry *omm_remote_parent;
179 int osd_ldiskfs_add_entry(struct osd_thread_info *info,
180 handle_t *handle, struct dentry *child,
181 struct inode *inode, struct htree_lock *hlock);
183 #define OSD_OTABLE_IT_CACHE_SIZE 64
184 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
186 struct osd_inconsistent_item {
187 /* link into osd_scrub::os_inconsistent_items,
188 * protected by osd_scrub::os_lock. */
189 struct list_head oii_list;
191 /* The right FID <=> ino#/gen mapping. */
192 struct osd_idmap_cache oii_cache;
194 unsigned int oii_insert:1; /* insert or update mapping. */
197 struct osd_otable_cache {
198 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
200 /* Index for next cache slot to be filled. */
201 int ooc_producer_idx;
203 /* Index for next cache slot to be returned by it::next(). */
204 int ooc_consumer_idx;
206 /* How many items in ooc_cache. */
207 int ooc_cached_items;
209 /* Position for up layer LFSCK iteration pre-loading. */
210 __u32 ooc_pos_preload;
213 struct osd_otable_it {
214 struct osd_device *ooi_dev;
215 struct osd_otable_cache ooi_cache;
217 /* The following bits can be updated/checked w/o lock protection.
218 * If more bits will be introduced in the future and need lock to
219 * protect, please add comment. */
220 unsigned long ooi_used_outside:1, /* Some user out of OSD
221 * uses the iteration. */
222 ooi_all_cached:1, /* No more entries can be
223 * filled into cache. */
224 ooi_user_ready:1, /* The user out of OSD is
225 * ready to iterate. */
226 ooi_waiting:1; /* it::next is waiting. */
229 struct osd_obj_orphan {
230 struct list_head oor_list;
231 struct lu_env *oor_env; /* to identify "own" records */
240 struct dt_device od_dt_dev;
241 /* information about underlying file system */
242 struct vfsmount *od_mnt;
244 struct osd_oi **od_oi_table;
245 /* total number of OI containers */
250 unsigned int od_fl_capa:1,
258 __u32 od_dirent_journal;
260 struct proc_dir_entry *od_proc_entry;
261 struct lprocfs_stats *od_stats;
263 spinlock_t od_osfs_lock;
266 struct lu_site od_site;
268 struct osd_obj_map *od_ost_map;
269 struct osd_mdobj_map *od_mdt_map;
271 unsigned long long od_readcache_max_filesize;
273 int od_writethrough_cache;
275 struct brw_stats od_brw_stats;
276 atomic_t od_r_in_flight;
277 atomic_t od_w_in_flight;
279 struct mutex od_otable_mutex;
280 struct osd_otable_it *od_otable_it;
281 struct osd_scrub od_scrub;
282 struct list_head od_ios_list;
284 /* service name associated with the osd device */
285 char od_svname[MAX_OBD_NAME];
286 char od_mntdev[MAX_OBD_NAME];
288 /* quota slave instance */
289 struct qsd_instance *od_quota_slave;
291 /* osd seq instance */
292 struct lu_client_seq *od_cl_seq;
293 /* If the ratio of "the total OI mappings count" vs
294 * "the bad OI mappings count" is lower than the
295 * osd_device::od_full_scrub_ratio, then trigger
296 * OI scrub to scan the whole the device. */
297 __u64 od_full_scrub_ratio;
298 /* If the speed of found bad OI mappings (per minute)
299 * exceeds the osd_device::od_full_scrub_threshold_rate,
300 * then trigger OI scrub to scan the whole device. */
301 __u64 od_full_scrub_threshold_rate;
303 /* a list of orphaned agent inodes, protected with od_osfs_lock */
304 struct list_head od_orphan_list;
307 enum osd_full_scrub_ratio {
308 /* Trigger OI scrub to scan the whole device directly. */
311 /* Because the bad OI mappings count cannot be larger than
312 * the total OI mappints count, then setting OFSR_NEVER means
313 * that the whole device scanning cannot be triggered by auto
314 * detected bad OI mappings during the RPC services. */
316 OFSR_DEFAULT = 10000,
319 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
321 /* There are at most 10 uid/gids are affected in a transaction, and
322 * that's rename case:
323 * - 2 for source parent uid & gid;
324 * - 2 for source child uid & gid ('..' entry update when child is directory);
325 * - 2 for target parent uid & gid;
326 * - 2 for target child uid & gid (if the target child exists);
327 * - 2 for root uid & gid (last_rcvd, llog, etc);
329 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
330 * the id type of each id in the ot_id_array.
332 #define OSD_MAX_UGID_CNT 10
337 OSD_OT_XATTR_SET = 2,
350 struct thandle ot_super;
352 struct ldiskfs_journal_cb_entry ot_jcb;
353 struct list_head ot_commit_dcb_list;
354 struct list_head ot_stop_dcb_list;
355 /* Link to the device, for debugging. */
356 struct lu_ref_link ot_dev_link;
357 unsigned short ot_credits;
358 unsigned short ot_id_cnt;
359 unsigned short ot_id_type;
360 int ot_remove_agents:1;
361 uid_t ot_id_array[OSD_MAX_UGID_CNT];
362 struct lquota_trans *ot_quota_trans;
363 #if OSD_THANDLE_STATS
364 /** time when this handle was allocated */
365 cfs_time_t oth_alloced;
367 /** time when this thanle was started */
368 cfs_time_t oth_started;
373 * Basic transaction credit op
394 #ifdef CONFIG_PROC_FS
396 LPROC_OSD_READ_BYTES = 0,
397 LPROC_OSD_WRITE_BYTES = 1,
398 LPROC_OSD_GET_PAGE = 2,
399 LPROC_OSD_NO_PAGE = 3,
400 LPROC_OSD_CACHE_ACCESS = 4,
401 LPROC_OSD_CACHE_HIT = 5,
402 LPROC_OSD_CACHE_MISS = 6,
404 #if OSD_THANDLE_STATS
405 LPROC_OSD_THANDLE_STARTING,
406 LPROC_OSD_THANDLE_OPEN,
407 LPROC_OSD_THANDLE_CLOSING,
414 * Storage representation for fids.
416 * Variable size, first byte contains the length of the whole record.
418 struct osd_fid_pack {
419 unsigned char fp_len;
420 char fp_area[sizeof(struct lu_fid)];
423 struct osd_it_ea_dirent {
424 struct lu_fid oied_fid;
427 unsigned short oied_namelen;
428 unsigned int oied_type;
430 } __attribute__((packed));
433 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
434 * than lu_dirent struct. osd readdir reads less number of dirent than
435 * required for mdd dir page. so buffer size need to be increased so that
436 * there would be one ext3 readdir for every mdd readdir page.
439 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
442 * This is iterator's in-memory data structure in interoperability
443 * mode (i.e. iterator over ldiskfs style directory)
446 struct osd_object *oie_obj;
447 /** used in ldiskfs iterator, to stored file pointer */
448 struct file oie_file;
449 /** how many entries have been read-cached from storage */
451 /** current entry is being iterated by caller */
453 /** current processing entry */
454 struct osd_it_ea_dirent *oie_dirent;
455 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
457 struct dentry oie_dentry;
461 * Iterator's in-memory data structure for IAM mode.
464 struct osd_object *oi_obj;
465 struct iam_path_descr *oi_ipd;
466 struct iam_iterator oi_it;
469 struct osd_quota_leaf {
470 struct list_head oql_link;
475 * Iterator's in-memory data structure for quota file.
477 struct osd_it_quota {
478 struct osd_object *oiq_obj;
479 /** tree blocks path to where the entry is stored */
480 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
481 /** on-disk offset for current key where quota record can be found */
483 /** identifier for current quota record */
485 /** the record index in the leaf/index block */
486 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
487 /** list of already processed leaf blocks */
488 struct list_head oiq_list;
491 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
494 wait_queue_head_t dr_wait;
495 atomic_t dr_numreqs; /* number of reqs being processed */
500 unsigned int dr_ignore_quota:1;
501 unsigned int dr_elapsed_valid:1; /* we really did count time */
502 unsigned int dr_rw:1;
503 struct lu_buf dr_pg_buf;
504 struct page **dr_pages;
505 struct lu_buf dr_bl_buf;
507 unsigned long dr_start_time;
508 unsigned long dr_elapsed; /* how long io took */
509 struct osd_device *dr_dev;
510 unsigned int dr_init_at; /* the line iobuf was initialized */
513 #define OSD_INS_CACHE_SIZE 8
515 struct osd_thread_info {
516 const struct lu_env *oti_env;
518 * used for index operations.
520 struct dentry oti_obj_dentry;
521 struct dentry oti_child_dentry;
523 /** dentry for Iterator context. */
524 struct dentry oti_it_dentry;
527 /* fake struct file for osd_object_sync */
528 struct file oti_file;
530 struct kstatfs oti_ksfs;
533 struct htree_lock *oti_hlock;
535 struct lu_fid oti_fid;
536 struct lu_fid oti_fid2;
537 struct lu_fid oti_fid3;
538 struct osd_inode_id oti_id;
539 struct osd_inode_id oti_id2;
540 struct osd_inode_id oti_id3;
541 struct ost_id oti_ostid;
544 * XXX temporary: for ->i_op calls.
546 struct timespec oti_time;
548 /** osd_device reference, initialized in osd_trans_start() and
549 used in osd_trans_stop() */
550 struct osd_device *oti_dev;
553 * following ipd and it structures are used for osd_index_iam_lookup()
554 * these are defined separately as we might do index operation
555 * in open iterator session.
558 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
560 unsigned int oti_it_ea_buf_used:1;
562 /* IAM iterator for index operation. */
563 struct iam_iterator oti_idx_it;
565 /** union to guarantee that ->oti_ipd[] has proper alignment. */
568 char oti_it_ipd[DX_IPD_MAX_SIZE];
569 long long oti_alignment_lieutenant;
573 char oti_idx_ipd[DX_IPD_MAX_SIZE];
574 long long oti_alignment_lieutenant_colonel;
577 struct osd_idmap_cache oti_cache;
579 /* dedicated OI cache for insert (which needs inum) */
580 struct osd_idmap_cache *oti_ins_cache;
581 int oti_ins_cache_size;
582 int oti_ins_cache_used;
587 /** used in osd_fid_set() to put xattr */
588 struct lu_buf oti_buf;
589 struct lu_buf oti_big_buf;
590 /** used in osd_ea_fid_set() to set fid into common ea */
592 struct lustre_mdt_attrs oti_mdt_attrs;
593 /* old LMA for compatibility */
594 char oti_mdt_attrs_old[LMA_OLD_SIZE];
595 struct filter_fid_old oti_ff;
596 struct filter_fid oti_ff_new;
599 struct osd_iobuf oti_iobuf;
600 /* used to access objects in /O */
601 struct inode *oti_inode;
602 #define OSD_FID_REC_SZ 32
603 char oti_ldp[OSD_FID_REC_SZ];
604 char oti_ldp2[OSD_FID_REC_SZ];
606 /* used by quota code */
608 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
609 struct fs_disk_quota oti_fdq;
611 struct if_dqblk oti_dqblk;
613 struct if_dqinfo oti_dqinfo;
615 struct lquota_id_info oti_qi;
616 struct lquota_trans oti_quota_trans;
617 union lquota_rec oti_quota_rec;
619 struct lu_seq_range oti_seq_range;
621 /* Tracking for transaction credits, to allow debugging and optimizing
622 * cases where a large number of credits are being allocated for
623 * single transaction. */
624 unsigned int oti_credits_before;
625 unsigned short oti_declare_ops[OSD_OT_MAX];
626 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
627 unsigned short oti_declare_ops_used[OSD_OT_MAX];
630 extern int ldiskfs_pdo;
632 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
633 const char *name, void *buf, int len)
638 dentry->d_inode = inode;
639 dentry->d_sb = inode->i_sb;
640 return inode->i_op->getxattr(dentry, name, buf, len);
643 static inline int __osd_xattr_set(struct osd_thread_info *info,
644 struct inode *inode, const char *name,
645 const void *buf, int buflen, int fl)
647 struct dentry *dentry = &info->oti_child_dentry;
649 ll_vfs_dq_init(inode);
650 dentry->d_inode = inode;
651 dentry->d_sb = inode->i_sb;
652 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
655 #ifdef CONFIG_PROC_FS
657 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
658 extern struct lprocfs_vars lprocfs_osd_module_vars[];
659 int osd_procfs_init(struct osd_device *osd, const char *name);
660 int osd_procfs_fini(struct osd_device *osd);
661 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
662 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
663 int osd_register_proc_index_in_idif(struct osd_device *osd);
667 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
668 struct obd_statfs *sfs);
669 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
670 struct osd_inode_id *id);
671 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
672 const struct lu_fid *fid, __u32 compat, __u32 incompat);
673 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
674 struct dentry *dentry, struct lustre_mdt_attrs *lma);
675 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
676 struct osd_inode_id *id, const struct lu_fid *fid);
677 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
678 struct dentry *dentry, struct lu_fid *fid);
680 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
681 void osd_obj_map_fini(struct osd_device *dev);
682 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
683 const struct lu_fid *fid, struct osd_inode_id *id);
684 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
685 const struct lu_fid *fid, const struct osd_inode_id *id,
687 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
688 const struct lu_fid *fid, handle_t *th);
689 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
690 const struct lu_fid *fid, const struct osd_inode_id *id,
692 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
693 struct inode *src_parent, struct dentry *src_child,
694 const struct lu_fid *fid);
695 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
696 const struct lu_fid *fid, struct osd_inode_id *id);
697 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
698 const struct lu_fid *fid, const struct osd_inode_id *id,
700 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
701 const struct lu_fid *fid, const struct osd_inode_id *id,
704 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
705 int osd_scrub_file_store(struct osd_scrub *scrub);
706 char *osd_lf_fid2name(const struct lu_fid *fid);
707 int osd_scrub_start(struct osd_device *dev, __u32 flags);
708 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
709 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
710 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
712 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
713 struct osd_inode_id *id);
714 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
716 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
717 u64 seq, struct lu_seq_range *range);
719 int osd_delete_from_remote_parent(const struct lu_env *env,
720 struct osd_device *osd,
721 struct osd_object *obj,
722 struct osd_thandle *oh);
723 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
724 struct osd_object *obj, struct osd_thandle *oh);
725 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
726 struct osd_device *osd,
727 const struct lu_fid *fid,
728 struct osd_inode_id *id);
730 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
732 /* osd_quota_fmt.c */
733 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
734 int type, uint blk, int depth, uint index,
735 struct osd_it_quota *it);
736 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
737 int type, uint blk, uint index,
738 struct osd_it_quota *it);
739 loff_t find_tree_dqentry(const struct lu_env *env,
740 struct osd_object *obj, int type,
741 qid_t dqid, uint blk, int depth,
742 struct osd_it_quota *it);
744 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
745 struct lquota_id_info *qi, struct osd_object *obj,
746 bool enforce, int *flags);
747 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
748 long long space, struct osd_thandle *oh,
749 struct osd_object *obj, bool is_blk, int *flags,
751 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
752 const struct dt_rec *rec,
753 union lquota_rec *quota_rec);
754 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
755 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
757 #ifndef HAVE_I_UID_READ
758 static inline uid_t i_uid_read(const struct inode *inode)
763 static inline gid_t i_gid_read(const struct inode *inode)
768 static inline void i_uid_write(struct inode *inode, uid_t uid)
773 static inline void i_gid_write(struct inode *inode, gid_t gid)
779 #ifdef LDISKFS_HT_MISC
780 # define osd_journal_start_sb(sb, type, nblock) \
781 ldiskfs_journal_start_sb(sb, type, nblock)
782 # define osd_ldiskfs_append(handle, inode, nblock) \
783 ldiskfs_append(handle, inode, nblock)
784 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
785 __ldiskfs_find_entry(dir, name, de, inlined, lock)
786 # define osd_journal_start(inode, type, nblocks) \
787 ldiskfs_journal_start(inode, type, nblocks)
788 # define osd_transaction_size(dev) \
789 (osd_journal(dev)->j_max_transaction_buffers / 2)
791 # define LDISKFS_HT_MISC 0
792 # define osd_journal_start_sb(sb, type, nblock) \
793 ldiskfs_journal_start_sb(sb, nblock)
795 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
797 ldiskfs_lblk_t *nblock)
799 struct buffer_head *bh;
802 bh = ldiskfs_append(handle, inode, nblock, &err);
809 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
810 __ldiskfs_find_entry(dir, name, de, lock)
811 # define osd_journal_start(inode, type, nblocks) \
812 ldiskfs_journal_start(inode, nblocks)
813 # define osd_transaction_size(dev) \
814 (osd_journal(dev)->j_max_transaction_buffers)
818 * Invariants, assertions.
822 * XXX: do not enable this, until invariant checking code is made thread safe
823 * in the face of pdirops locking.
825 #define OSD_INVARIANT_CHECKS (0)
827 #if OSD_INVARIANT_CHECKS
828 static inline int osd_invariant(const struct osd_object *obj)
832 ergo(obj->oo_inode != NULL,
833 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
834 atomic_read(&obj->oo_inode->i_count) > 0) &&
835 ergo(obj->oo_dir != NULL &&
836 obj->oo_dir->od_conationer.ic_object != NULL,
837 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
840 #define osd_invariant(obj) (1)
843 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
845 extern const struct dt_index_operations osd_otable_ops;
847 static inline int osd_oi_fid2idx(struct osd_device *dev,
848 const struct lu_fid *fid)
850 return fid->f_seq & (dev->od_oi_count - 1);
853 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
854 const struct lu_fid *fid)
856 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
857 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
858 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
859 DFID"\n", PFID(fid));
860 /* It can work even od_oi_count equals to 1 although it's unexpected,
861 * the only reason we set it to 1 is for performance measurement */
862 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
865 extern const struct lu_device_operations osd_lu_ops;
867 static inline int lu_device_is_osd(const struct lu_device *d)
869 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
872 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
874 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
875 return container_of0(d, struct osd_device, od_dt_dev);
878 static inline struct osd_device *osd_dev(const struct lu_device *d)
880 LASSERT(lu_device_is_osd(d));
881 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
884 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
886 return osd_dev(o->oo_dt.do_lu.lo_dev);
889 static inline struct super_block *osd_sb(const struct osd_device *dev)
891 return dev->od_mnt->mnt_sb;
894 static inline int osd_object_is_root(const struct osd_object *obj)
896 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
899 static inline struct osd_object *osd_obj(const struct lu_object *o)
901 LASSERT(lu_device_is_osd(o->lo_dev));
902 return container_of0(o, struct osd_object, oo_dt.do_lu);
905 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
907 return osd_obj(&d->do_lu);
910 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
912 return &osd->od_dt_dev.dd_lu_dev;
915 static inline journal_t *osd_journal(const struct osd_device *dev)
917 return LDISKFS_SB(osd_sb(dev))->s_journal;
920 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
922 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
925 static inline char *osd_name(struct osd_device *osd)
927 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
930 static inline bool osd_is_ea_inode(struct inode *inode)
932 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
935 extern const struct dt_body_operations osd_body_ops;
936 extern struct lu_context_key osd_key;
938 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
940 return lu_context_key_get(&env->le_ctx, &osd_key);
943 extern const struct dt_body_operations osd_body_ops_new;
949 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
950 const struct iam_container *bag)
952 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
953 osd_oti_get(env)->oti_it_ipd);
957 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
958 const struct iam_container *bag)
960 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
961 osd_oti_get(env)->oti_idx_ipd);
964 static inline void osd_ipd_put(const struct lu_env *env,
965 const struct iam_container *bag,
966 struct iam_path_descr *ipd)
968 bag->ic_descr->id_ops->id_ipd_free(ipd);
971 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
972 const loff_t size, const loff_t pos,
975 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
976 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
977 int write_NUL, loff_t *offs, handle_t *handle);
980 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
982 const char *name, const int namelen)
984 struct osd_thread_info *info = osd_oti_get(env);
985 struct dentry *child_dentry = &info->oti_child_dentry;
986 struct dentry *obj_dentry = &info->oti_obj_dentry;
988 obj_dentry->d_inode = inode;
989 obj_dentry->d_sb = inode->i_sb;
990 obj_dentry->d_name.hash = 0;
992 child_dentry->d_name.hash = 0;
993 child_dentry->d_parent = obj_dentry;
994 child_dentry->d_name.name = name;
995 child_dentry->d_name.len = namelen;
999 extern int osd_trans_declare_op2rb[];
1000 extern int ldiskfs_track_declares_assert;
1001 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1003 static inline void osd_trans_declare_op(const struct lu_env *env,
1004 struct osd_thandle *oh,
1005 unsigned int op, int credits)
1007 struct osd_thread_info *oti = osd_oti_get(env);
1009 LASSERT(oh->ot_handle == NULL);
1010 if (unlikely(op >= OSD_OT_MAX)) {
1011 if (unlikely(ldiskfs_track_declares_assert)) {
1012 LASSERT(op < OSD_OT_MAX);
1014 CWARN("%s: Invalid operation index %d\n",
1015 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1016 libcfs_debug_dumpstack(NULL);
1019 oti->oti_declare_ops[op]++;
1020 oti->oti_declare_ops_cred[op] += credits;
1022 oh->ot_credits += credits;
1025 static inline void osd_trans_exec_op(const struct lu_env *env,
1026 struct thandle *th, unsigned int op)
1028 struct osd_thread_info *oti = osd_oti_get(env);
1029 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1031 unsigned int rb, left;
1033 LASSERT(oh->ot_handle != NULL);
1034 if (unlikely(op >= OSD_OT_MAX)) {
1035 if (unlikely(ldiskfs_track_declares_assert))
1036 LASSERT(op < OSD_OT_MAX);
1038 CWARN("%s: Invalid operation index %d\n",
1039 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1040 libcfs_debug_dumpstack(NULL);
1045 /* find rollback (or reverse) operation for the given one
1046 * such an operation doesn't require additional credits
1047 * as the same set of blocks are modified */
1048 rb = osd_trans_declare_op2rb[op];
1050 /* check whether credits for this operation were reserved at all */
1051 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1052 oti->oti_declare_ops_cred[rb] == 0)) {
1053 /* the API is not perfect yet: CREATE does REF_ADD internally
1054 * while DESTROY does not. To rollback CREATE the callers
1055 * needs to call REF_DEL+DESTROY which is hard to detect using
1056 * a simple table of rollback operations */
1057 if (op == OSD_OT_REF_DEL &&
1058 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1060 if (op == OSD_OT_REF_ADD &&
1061 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1063 osd_trans_dump_creds(env, th);
1064 CERROR("%s: op = %d, rb = %d\n",
1065 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1066 if (unlikely(ldiskfs_track_declares_assert))
1071 /* remember how many credits we have unused before the operation */
1072 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1073 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1074 if (unlikely(oti->oti_credits_before < left)) {
1075 osd_trans_dump_creds(env, th);
1076 CERROR("%s: op = %d, rb = %d\n",
1077 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1078 /* on a very small fs (testing?) it's possible that
1079 * the transaction can't fit 1/4 of journal, so we
1080 * just request less credits (see osd_trans_start()).
1081 * ignore the same case here */
1082 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1083 if (unlikely(oh->ot_credits < rb)) {
1084 if (unlikely(ldiskfs_track_declares_assert))
1090 static inline void osd_trans_exec_check(const struct lu_env *env,
1094 struct osd_thread_info *oti = osd_oti_get(env);
1095 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1097 int used, over, quota;
1099 /* how many credits have been used by the operation */
1100 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1102 if (unlikely(used < 0)) {
1103 /* if some block was allocated and released in the same
1104 * transaction, then it won't be a part of the transaction
1105 * and delta can be negative */
1110 /* rollback operations (e.g. when we destroy just created
1111 * object) should not consume any credits. there is no point
1112 * to confuse the checks below */
1116 oti->oti_declare_ops_used[op] += used;
1117 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1120 /* we account quota for a whole transaction and any operation can
1121 * consume corresponding credits */
1122 over = oti->oti_declare_ops_used[op] -
1123 oti->oti_declare_ops_cred[op];
1124 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1125 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1126 if (over <= quota) {
1127 /* probably that credits were consumed by
1128 * quota indirectly (in the depths of ldiskfs) */
1129 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1130 oti->oti_declare_ops_used[op] -= over;
1132 CWARN("op %d: used %u, used now %u, reserved %u\n",
1133 op, oti->oti_declare_ops_used[op], used,
1134 oti->oti_declare_ops_cred[op]);
1135 osd_trans_dump_creds(env, th);
1136 if (unlikely(ldiskfs_track_declares_assert))
1142 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1145 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1146 struct lu_fid *befider)
1148 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1149 memcpy(pack->fp_area, befider, sizeof(*befider));
1150 pack->fp_len = sizeof(*befider) + 1;
1154 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1159 switch (pack->fp_len) {
1160 case sizeof *fid + 1:
1161 memcpy(fid, pack->fp_area, sizeof *fid);
1162 fid_be_to_cpu(fid, fid);
1165 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1172 * Quota/Accounting handling
1174 extern const struct dt_index_operations osd_acct_index_ops;
1175 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1176 const struct lu_fid *fid, struct osd_inode_id *id);
1178 /* copy from fs/ext4/dir.c */
1179 static inline int is_32bit_api(void)
1181 #ifdef CONFIG_COMPAT
1182 return is_compat_task();
1184 return (BITS_PER_LONG == 32);
1188 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1190 if ((filp->f_mode & FMODE_32BITHASH) ||
1191 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1192 return LDISKFS_HTREE_EOF_32BIT;
1194 return LDISKFS_HTREE_EOF_64BIT;
1197 static inline int fid_is_internal(const struct lu_fid *fid)
1199 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1202 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1204 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1207 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1208 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1210 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1212 #endif /* _OSD_INTERNAL_H */