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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_internal.h
34 * Shared definitions and declarations for osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
39 #ifndef _OSD_INTERNAL_H
40 #define _OSD_INTERNAL_H
44 #include <linux/mutex.h>
45 /* struct rw_semaphore */
46 #include <linux/rwsem.h>
48 #include <linux/dcache.h>
50 #include <linux/dirent.h>
51 #include <linux/statfs.h>
52 #include <ldiskfs/ldiskfs.h>
53 #include <ldiskfs/ldiskfs_jbd2.h>
57 /* class_register_type(), class_unregister_type(), class_get_type() */
58 #include <obd_class.h>
59 #include <lustre_disk.h>
60 #include <dt_object.h>
61 #include <lustre_quota.h>
65 #include "osd_scrub.h"
66 #include "osd_quota_fmt.h"
69 extern struct kmem_cache *dynlock_cachep;
71 #define OSD_COUNTERS (0)
73 /* ldiskfs special inode::i_state_flags need to be accessed with
74 * ldiskfs_{set,clear,test}_inode_state() only */
76 /* OI scrub should skip this inode. */
77 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
78 #define LDISKFS_STATE_LUSTRE_DESTROY 30
80 /** Enable thandle usage statistics */
81 #define OSD_THANDLE_STATS (0)
83 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
85 #define OBJECTS "OBJECTS"
86 #define ADMIN_USR "admin_quotafile_v2.usr"
87 #define ADMIN_GRP "admin_quotafile_v2.grp"
89 /* Statfs space reservation for fragmentation and local objects */
90 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
91 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
93 struct osd_directory {
94 struct iam_container od_container;
95 struct iam_descr od_descr;
99 * Object Index (oi) instance.
103 * underlying index object, where fid->id mapping in stored.
105 struct inode *oi_inode;
106 struct osd_directory oi_dir;
109 extern const int osd_dto_credits_noquota[];
112 struct dt_object oo_dt;
114 * Inode for file system object represented by this osd_object. This
115 * inode is pinned for the whole duration of lu_object life.
117 * Not modified concurrently (either setup early during object
118 * creation, or assigned by osd_object_create() under write lock).
120 struct inode *oo_inode;
122 * to protect index ops.
124 struct htree_lock_head *oo_hl_head;
125 struct rw_semaphore oo_ext_idx_sem;
126 struct rw_semaphore oo_sem;
127 struct osd_directory *oo_dir;
128 /** protects inode attributes. */
131 __u32 oo_destroyed:1;
133 /* the i_flags in LMA */
136 * Following two members are used to indicate the presence of dot and
137 * dotdot in the given directory. This is required for interop mode
140 int oo_compat_dot_created;
141 int oo_compat_dotdot_created;
143 const struct lu_env *oo_owner;
144 #ifdef CONFIG_LOCKDEP
145 struct lockdep_map oo_dep_map;
148 struct list_head oo_xattr_list;
152 /* protects on-fly initialization */
153 int oos_subdir_count; /* subdir count for each seq */
154 struct dentry *oos_root; /* O/<seq> */
155 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
156 u64 oos_seq; /* seq number */
157 struct list_head oos_seq_list; /* list to seq_list */
161 struct dentry *om_root; /* dentry for /O */
162 rwlock_t om_seq_list_lock; /* lock for seq_list */
163 struct list_head om_seq_list; /* list head for seq */
165 struct mutex om_dir_init_mutex;
169 struct dentry *om_root; /* AGENT/<index> */
170 u64 om_index; /* mdt index */
171 struct list_head om_list; /* list to omm_list */
174 struct osd_mdobj_map {
175 struct dentry *omm_remote_parent;
177 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
178 handle_t *handle, struct dentry *child,
179 struct inode *inode, struct htree_lock *hlock);
181 #define OSD_OTABLE_IT_CACHE_SIZE 64
182 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
184 struct osd_inconsistent_item {
185 /* link into osd_scrub::os_inconsistent_items,
186 * protected by osd_scrub::os_lock. */
187 struct list_head oii_list;
189 /* The right FID <=> ino#/gen mapping. */
190 struct osd_idmap_cache oii_cache;
192 unsigned int oii_insert:1; /* insert or update mapping. */
195 struct osd_otable_cache {
196 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
198 /* Index for next cache slot to be filled. */
199 int ooc_producer_idx;
201 /* Index for next cache slot to be returned by it::next(). */
202 int ooc_consumer_idx;
204 /* How many items in ooc_cache. */
205 int ooc_cached_items;
207 /* Position for up layer LFSCK iteration pre-loading. */
208 __u32 ooc_pos_preload;
211 struct osd_otable_it {
212 struct osd_device *ooi_dev;
213 struct osd_otable_cache ooi_cache;
215 /* The following bits can be updated/checked w/o lock protection.
216 * If more bits will be introduced in the future and need lock to
217 * protect, please add comment. */
218 unsigned long ooi_used_outside:1, /* Some user out of OSD
219 * uses the iteration. */
220 ooi_all_cached:1, /* No more entries can be
221 * filled into cache. */
222 ooi_user_ready:1, /* The user out of OSD is
223 * ready to iterate. */
224 ooi_waiting:1; /* it::next is waiting. */
227 struct osd_obj_orphan {
228 struct list_head oor_list;
229 struct lu_env *oor_env; /* to identify "own" records */
238 struct dt_device od_dt_dev;
239 /* information about underlying file system */
240 struct vfsmount *od_mnt;
242 struct osd_oi **od_oi_table;
243 /* total number of OI containers */
248 unsigned int od_fl_capa:1,
256 __u32 od_dirent_journal;
258 struct proc_dir_entry *od_proc_entry;
259 struct lprocfs_stats *od_stats;
261 spinlock_t od_osfs_lock;
264 struct lu_site od_site;
266 struct osd_obj_map *od_ost_map;
267 struct osd_mdobj_map *od_mdt_map;
269 unsigned long long od_readcache_max_filesize;
271 int od_writethrough_cache;
273 struct brw_stats od_brw_stats;
274 atomic_t od_r_in_flight;
275 atomic_t od_w_in_flight;
277 struct mutex od_otable_mutex;
278 struct osd_otable_it *od_otable_it;
279 struct osd_scrub od_scrub;
280 struct list_head od_ios_list;
282 /* service name associated with the osd device */
283 char od_svname[MAX_OBD_NAME];
284 char od_mntdev[MAX_OBD_NAME];
286 /* quota slave instance */
287 struct qsd_instance *od_quota_slave;
289 /* osd seq instance */
290 struct lu_client_seq *od_cl_seq;
291 /* If the ratio of "the total OI mappings count" vs
292 * "the bad OI mappings count" is lower than the
293 * osd_device::od_full_scrub_ratio, then trigger
294 * OI scrub to scan the whole the device. */
295 __u64 od_full_scrub_ratio;
296 /* If the speed of found bad OI mappings (per minute)
297 * exceeds the osd_device::od_full_scrub_threshold_rate,
298 * then trigger OI scrub to scan the whole device. */
299 __u64 od_full_scrub_threshold_rate;
301 /* a list of orphaned agent inodes, protected with od_osfs_lock */
302 struct list_head od_orphan_list;
305 enum osd_full_scrub_ratio {
306 /* Trigger OI scrub to scan the whole device directly. */
309 /* Because the bad OI mappings count cannot be larger than
310 * the total OI mappints count, then setting OFSR_NEVER means
311 * that the whole device scanning cannot be triggered by auto
312 * detected bad OI mappings during the RPC services. */
314 OFSR_DEFAULT = 10000,
317 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
319 /* There are at most 10 uid/gids are affected in a transaction, and
320 * that's rename case:
321 * - 2 for source parent uid & gid;
322 * - 2 for source child uid & gid ('..' entry update when child is directory);
323 * - 2 for target parent uid & gid;
324 * - 2 for target child uid & gid (if the target child exists);
325 * - 2 for root uid & gid (last_rcvd, llog, etc);
327 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
328 * the id type of each id in the ot_id_array.
330 #define OSD_MAX_UGID_CNT 10
335 OSD_OT_XATTR_SET = 2,
348 struct thandle ot_super;
350 struct ldiskfs_journal_cb_entry ot_jcb;
351 struct list_head ot_commit_dcb_list;
352 struct list_head ot_stop_dcb_list;
353 /* Link to the device, for debugging. */
354 struct lu_ref_link ot_dev_link;
355 unsigned int ot_credits;
356 unsigned short ot_id_cnt;
357 unsigned short ot_id_type;
358 unsigned int ot_remove_agents:1;
359 uid_t ot_id_array[OSD_MAX_UGID_CNT];
360 struct lquota_trans *ot_quota_trans;
361 #if OSD_THANDLE_STATS
362 /** time when this handle was allocated */
363 cfs_time_t oth_alloced;
365 /** time when this thanle was started */
366 cfs_time_t oth_started;
371 * Basic transaction credit op
392 #ifdef CONFIG_PROC_FS
394 LPROC_OSD_READ_BYTES = 0,
395 LPROC_OSD_WRITE_BYTES = 1,
396 LPROC_OSD_GET_PAGE = 2,
397 LPROC_OSD_NO_PAGE = 3,
398 LPROC_OSD_CACHE_ACCESS = 4,
399 LPROC_OSD_CACHE_HIT = 5,
400 LPROC_OSD_CACHE_MISS = 6,
402 #if OSD_THANDLE_STATS
403 LPROC_OSD_THANDLE_STARTING,
404 LPROC_OSD_THANDLE_OPEN,
405 LPROC_OSD_THANDLE_CLOSING,
412 * Storage representation for fids.
414 * Variable size, first byte contains the length of the whole record.
416 struct osd_fid_pack {
417 unsigned char fp_len;
418 char fp_area[sizeof(struct lu_fid)];
421 struct osd_it_ea_dirent {
422 struct lu_fid oied_fid;
425 unsigned short oied_namelen;
426 unsigned int oied_type;
428 } __attribute__((packed));
431 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
432 * than lu_dirent struct. osd readdir reads less number of dirent than
433 * required for mdd dir page. so buffer size need to be increased so that
434 * there would be one ext3 readdir for every mdd readdir page.
437 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
440 * This is iterator's in-memory data structure in interoperability
441 * mode (i.e. iterator over ldiskfs style directory)
444 struct osd_object *oie_obj;
445 /** used in ldiskfs iterator, to stored file pointer */
446 struct file oie_file;
447 /** how many entries have been read-cached from storage */
449 /** current entry is being iterated by caller */
451 /** current processing entry */
452 struct osd_it_ea_dirent *oie_dirent;
453 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
455 struct dentry oie_dentry;
459 * Iterator's in-memory data structure for IAM mode.
462 struct osd_object *oi_obj;
463 struct iam_path_descr *oi_ipd;
464 struct iam_iterator oi_it;
467 struct osd_quota_leaf {
468 struct list_head oql_link;
473 * Iterator's in-memory data structure for quota file.
475 struct osd_it_quota {
476 struct osd_object *oiq_obj;
477 /** tree blocks path to where the entry is stored */
478 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
479 /** on-disk offset for current key where quota record can be found */
481 /** identifier for current quota record */
483 /** the record index in the leaf/index block */
484 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
485 /** list of already processed leaf blocks */
486 struct list_head oiq_list;
489 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
492 wait_queue_head_t dr_wait;
493 atomic_t dr_numreqs; /* number of reqs being processed */
498 unsigned int dr_ignore_quota:1;
499 unsigned int dr_elapsed_valid:1; /* we really did count time */
500 unsigned int dr_rw:1;
501 struct lu_buf dr_pg_buf;
502 struct page **dr_pages;
503 struct lu_buf dr_bl_buf;
505 unsigned long dr_start_time;
506 unsigned long dr_elapsed; /* how long io took */
507 struct osd_device *dr_dev;
508 unsigned int dr_init_at; /* the line iobuf was initialized */
511 #define OSD_INS_CACHE_SIZE 8
513 struct osd_thread_info {
514 const struct lu_env *oti_env;
516 * used for index operations.
518 struct dentry oti_obj_dentry;
519 struct dentry oti_child_dentry;
521 /** dentry for Iterator context. */
522 struct dentry oti_it_dentry;
525 /* fake struct file for osd_object_sync */
526 struct file oti_file;
528 struct kstatfs oti_ksfs;
531 struct htree_lock *oti_hlock;
533 struct lu_fid oti_fid;
534 struct lu_fid oti_fid2;
535 struct lu_fid oti_fid3;
536 struct osd_inode_id oti_id;
537 struct osd_inode_id oti_id2;
538 struct osd_inode_id oti_id3;
539 struct ost_id oti_ostid;
542 * XXX temporary: for ->i_op calls.
544 struct timespec oti_time;
547 * following ipd and it structures are used for osd_index_iam_lookup()
548 * these are defined separately as we might do index operation
549 * in open iterator session.
552 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
554 unsigned int oti_it_ea_buf_used:1;
556 /* IAM iterator for index operation. */
557 struct iam_iterator oti_idx_it;
559 /** union to guarantee that ->oti_ipd[] has proper alignment. */
562 char oti_it_ipd[DX_IPD_MAX_SIZE];
563 long long oti_alignment_lieutenant;
567 char oti_idx_ipd[DX_IPD_MAX_SIZE];
568 long long oti_alignment_lieutenant_colonel;
571 struct osd_idmap_cache oti_cache;
573 /* dedicated OI cache for insert (which needs inum) */
574 struct osd_idmap_cache *oti_ins_cache;
575 int oti_ins_cache_size;
576 int oti_ins_cache_used;
581 /** used in osd_fid_set() to put xattr */
582 struct lu_buf oti_buf;
583 struct lu_buf oti_big_buf;
584 /** used in osd_ea_fid_set() to set fid into common ea */
586 struct lustre_mdt_attrs oti_mdt_attrs;
587 /* old LMA for compatibility */
588 char oti_mdt_attrs_old[LMA_OLD_SIZE];
589 struct filter_fid_old oti_ff;
590 struct filter_fid oti_ff_new;
593 struct osd_iobuf oti_iobuf;
594 /* used to access objects in /O */
595 struct inode *oti_inode;
596 #define OSD_FID_REC_SZ 32
597 char oti_ldp[OSD_FID_REC_SZ];
598 char oti_ldp2[OSD_FID_REC_SZ];
600 /* used by quota code */
602 #if defined(HAVE_DQUOT_QC_DQBLK)
603 struct qc_dqblk oti_qdq;
604 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
605 struct fs_disk_quota oti_fdq;
607 struct if_dqblk oti_dqblk;
609 struct if_dqinfo oti_dqinfo;
611 struct lquota_id_info oti_qi;
612 struct lquota_trans oti_quota_trans;
613 union lquota_rec oti_quota_rec;
615 struct lu_seq_range oti_seq_range;
617 /* Tracking for transaction credits, to allow debugging and optimizing
618 * cases where a large number of credits are being allocated for
619 * single transaction. */
620 unsigned int oti_credits_before;
621 unsigned int oti_declare_ops[OSD_OT_MAX];
622 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
623 unsigned int oti_declare_ops_used[OSD_OT_MAX];
626 extern int ldiskfs_pdo;
628 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
629 const char *name, void *buf, int len)
634 dentry->d_inode = inode;
635 dentry->d_sb = inode->i_sb;
636 return inode->i_op->getxattr(dentry, name, buf, len);
639 static inline int __osd_xattr_set(struct osd_thread_info *info,
640 struct inode *inode, const char *name,
641 const void *buf, int buflen, int fl)
643 struct dentry *dentry = &info->oti_child_dentry;
645 ll_vfs_dq_init(inode);
646 dentry->d_inode = inode;
647 dentry->d_sb = inode->i_sb;
648 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
651 #ifdef CONFIG_PROC_FS
653 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
654 extern struct lprocfs_vars lprocfs_osd_module_vars[];
655 int osd_procfs_init(struct osd_device *osd, const char *name);
656 int osd_procfs_fini(struct osd_device *osd);
657 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
658 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
659 int osd_register_proc_index_in_idif(struct osd_device *osd);
663 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
664 struct obd_statfs *sfs);
665 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
666 struct osd_inode_id *id);
667 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
668 const struct lu_fid *fid, __u32 compat, __u32 incompat);
669 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
670 struct dentry *dentry, struct lustre_mdt_attrs *lma);
671 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
672 struct osd_inode_id *id, const struct lu_fid *fid);
673 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
674 struct dentry *dentry, struct lu_fid *fid);
676 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
677 void osd_obj_map_fini(struct osd_device *dev);
678 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
679 const struct lu_fid *fid, struct osd_inode_id *id);
680 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
681 const struct lu_fid *fid, const struct osd_inode_id *id,
683 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
684 const struct lu_fid *fid, handle_t *th);
685 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
686 const struct lu_fid *fid, const struct osd_inode_id *id,
688 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
689 struct inode *src_parent, struct dentry *src_child,
690 const struct lu_fid *fid);
691 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
692 const struct lu_fid *fid, struct osd_inode_id *id);
693 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
694 const struct lu_fid *fid, const struct osd_inode_id *id,
696 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
697 const struct lu_fid *fid, const struct osd_inode_id *id,
700 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
701 int osd_scrub_file_store(struct osd_scrub *scrub);
702 char *osd_lf_fid2name(const struct lu_fid *fid);
703 int osd_scrub_start(struct osd_device *dev, __u32 flags);
704 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
705 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
706 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
708 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
709 struct osd_inode_id *id);
710 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
712 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
713 u64 seq, struct lu_seq_range *range);
715 int osd_delete_from_remote_parent(const struct lu_env *env,
716 struct osd_device *osd,
717 struct osd_object *obj,
718 struct osd_thandle *oh);
719 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
720 struct osd_object *obj, struct osd_thandle *oh);
721 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
722 struct osd_device *osd,
723 const struct lu_fid *fid,
724 struct osd_inode_id *id);
726 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
728 /* osd_quota_fmt.c */
729 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
730 int type, uint blk, int depth, uint index,
731 struct osd_it_quota *it);
732 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
733 int type, uint blk, uint index,
734 struct osd_it_quota *it);
735 loff_t find_tree_dqentry(const struct lu_env *env,
736 struct osd_object *obj, int type,
737 qid_t dqid, uint blk, int depth,
738 struct osd_it_quota *it);
740 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
741 struct lquota_id_info *qi, struct osd_object *obj,
742 bool enforce, int *flags);
743 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
744 long long space, struct osd_thandle *oh,
745 struct osd_object *obj, bool is_blk, int *flags,
747 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
748 const struct dt_rec *rec,
749 union lquota_rec *quota_rec);
750 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
751 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
753 #ifndef HAVE_I_UID_READ
754 static inline uid_t i_uid_read(const struct inode *inode)
759 static inline gid_t i_gid_read(const struct inode *inode)
764 static inline void i_uid_write(struct inode *inode, uid_t uid)
769 static inline void i_gid_write(struct inode *inode, gid_t gid)
775 #ifdef LDISKFS_HT_MISC
776 # define osd_journal_start_sb(sb, type, nblock) \
777 ldiskfs_journal_start_sb(sb, type, nblock)
778 # define osd_ldiskfs_append(handle, inode, nblock) \
779 ldiskfs_append(handle, inode, nblock)
780 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
781 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
783 # define osd_journal_start(inode, type, nblocks) \
784 ldiskfs_journal_start(inode, type, nblocks)
785 # define osd_transaction_size(dev) \
786 (osd_journal(dev)->j_max_transaction_buffers / 2)
788 # define LDISKFS_HT_MISC 0
789 # define osd_journal_start_sb(sb, type, nblock) \
790 ldiskfs_journal_start_sb(sb, nblock)
792 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
794 ldiskfs_lblk_t *nblock)
796 struct buffer_head *bh;
799 bh = ldiskfs_append(handle, inode, nblock, &err);
806 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
807 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
809 # define osd_journal_start(inode, type, nblocks) \
810 ldiskfs_journal_start(inode, nblocks)
811 # define osd_transaction_size(dev) \
812 (osd_journal(dev)->j_max_transaction_buffers)
816 * Invariants, assertions.
820 * XXX: do not enable this, until invariant checking code is made thread safe
821 * in the face of pdirops locking.
823 #define OSD_INVARIANT_CHECKS (0)
825 #if OSD_INVARIANT_CHECKS
826 static inline int osd_invariant(const struct osd_object *obj)
830 ergo(obj->oo_inode != NULL,
831 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
832 atomic_read(&obj->oo_inode->i_count) > 0) &&
833 ergo(obj->oo_dir != NULL &&
834 obj->oo_dir->od_conationer.ic_object != NULL,
835 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
838 #define osd_invariant(obj) (1)
841 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
843 extern const struct dt_index_operations osd_otable_ops;
845 static inline int osd_oi_fid2idx(struct osd_device *dev,
846 const struct lu_fid *fid)
848 return fid->f_seq & (dev->od_oi_count - 1);
851 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
852 const struct lu_fid *fid)
854 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
855 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
856 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
857 DFID"\n", PFID(fid));
858 /* It can work even od_oi_count equals to 1 although it's unexpected,
859 * the only reason we set it to 1 is for performance measurement */
860 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
863 extern const struct lu_device_operations osd_lu_ops;
865 static inline int lu_device_is_osd(const struct lu_device *d)
867 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
870 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
872 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
873 return container_of0(d, struct osd_device, od_dt_dev);
876 static inline struct osd_device *osd_dev(const struct lu_device *d)
878 LASSERT(lu_device_is_osd(d));
879 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
882 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
884 return osd_dev(o->oo_dt.do_lu.lo_dev);
887 static inline struct super_block *osd_sb(const struct osd_device *dev)
889 return dev->od_mnt->mnt_sb;
892 static inline int osd_object_is_root(const struct osd_object *obj)
894 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
897 static inline struct osd_object *osd_obj(const struct lu_object *o)
899 LASSERT(lu_device_is_osd(o->lo_dev));
900 return container_of0(o, struct osd_object, oo_dt.do_lu);
903 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
905 return osd_obj(&d->do_lu);
908 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
910 return &osd->od_dt_dev.dd_lu_dev;
913 static inline journal_t *osd_journal(const struct osd_device *dev)
915 return LDISKFS_SB(osd_sb(dev))->s_journal;
918 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
920 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
923 static inline char *osd_name(struct osd_device *osd)
925 return osd->od_svname;
928 static inline bool osd_is_ea_inode(struct inode *inode)
930 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
933 extern const struct dt_body_operations osd_body_ops;
934 extern struct lu_context_key osd_key;
936 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
938 return lu_context_key_get(&env->le_ctx, &osd_key);
941 extern const struct dt_body_operations osd_body_ops_new;
947 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
948 const struct iam_container *bag)
950 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
951 osd_oti_get(env)->oti_it_ipd);
955 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
956 const struct iam_container *bag)
958 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
959 osd_oti_get(env)->oti_idx_ipd);
962 static inline void osd_ipd_put(const struct lu_env *env,
963 const struct iam_container *bag,
964 struct iam_path_descr *ipd)
966 bag->ic_descr->id_ops->id_ipd_free(ipd);
969 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
970 const loff_t size, const loff_t pos,
973 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
974 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
975 int write_NUL, loff_t *offs, handle_t *handle);
978 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
980 const char *name, const int namelen)
982 struct osd_thread_info *info = osd_oti_get(env);
983 struct dentry *child_dentry = &info->oti_child_dentry;
984 struct dentry *obj_dentry = &info->oti_obj_dentry;
986 obj_dentry->d_inode = inode;
987 obj_dentry->d_sb = inode->i_sb;
988 obj_dentry->d_name.hash = 0;
990 child_dentry->d_name.hash = 0;
991 child_dentry->d_parent = obj_dentry;
992 child_dentry->d_name.name = name;
993 child_dentry->d_name.len = namelen;
997 extern int osd_trans_declare_op2rb[];
998 extern int ldiskfs_track_declares_assert;
999 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1001 static inline void osd_trans_declare_op(const struct lu_env *env,
1002 struct osd_thandle *oh,
1003 enum osd_op_type op, int credits)
1005 struct osd_thread_info *oti = osd_oti_get(env);
1007 LASSERT(oh->ot_handle == NULL);
1008 if (unlikely(op >= OSD_OT_MAX)) {
1009 if (unlikely(ldiskfs_track_declares_assert)) {
1010 LASSERT(op < OSD_OT_MAX);
1012 CWARN("%s: Invalid operation index %d\n",
1013 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1014 libcfs_debug_dumpstack(NULL);
1017 oti->oti_declare_ops[op]++;
1018 oti->oti_declare_ops_cred[op] += credits;
1020 oh->ot_credits += credits;
1023 static inline void osd_trans_exec_op(const struct lu_env *env,
1025 enum osd_op_type op)
1027 struct osd_thread_info *oti = osd_oti_get(env);
1028 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1030 unsigned int rb, left;
1032 LASSERT(oh->ot_handle != NULL);
1033 if (unlikely(op >= OSD_OT_MAX)) {
1034 if (unlikely(ldiskfs_track_declares_assert))
1035 LASSERT(op < OSD_OT_MAX);
1037 CWARN("%s: opcode %u: invalid value >= %u\n",
1038 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
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 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1064 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1065 osd_trans_dump_creds(env, th);
1066 LASSERT(!ldiskfs_track_declares_assert);
1070 /* remember how many credits we have unused before the operation */
1071 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1072 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1073 if (unlikely(oti->oti_credits_before < left)) {
1074 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1075 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1076 oti->oti_credits_before, left, rb);
1077 osd_trans_dump_creds(env, th);
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 LASSERT(!ldiskfs_track_declares_assert);
1088 static inline void osd_trans_exec_check(const struct lu_env *env,
1090 enum osd_op_type op)
1092 struct osd_thread_info *oti = osd_oti_get(env);
1093 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1095 int used, over, quota;
1097 /* how many credits have been used by the operation */
1098 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1100 if (unlikely(used < 0)) {
1101 /* if some block was allocated and released in the same
1102 * transaction, then it won't be a part of the transaction
1103 * and delta can be negative */
1108 /* rollback operations (e.g. when we destroy just created
1109 * object) should not consume any credits. there is no point
1110 * to confuse the checks below */
1114 oti->oti_declare_ops_used[op] += used;
1115 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1118 /* we account quota for a whole transaction and any operation can
1119 * consume corresponding credits */
1120 over = oti->oti_declare_ops_used[op] -
1121 oti->oti_declare_ops_cred[op];
1122 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1123 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1124 if (over <= quota) {
1125 /* probably that credits were consumed by
1126 * quota indirectly (in the depths of ldiskfs) */
1127 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1128 oti->oti_declare_ops_used[op] -= over;
1130 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1131 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1132 oti->oti_declare_ops_used[op], used,
1133 oti->oti_declare_ops_cred[op]);
1134 osd_trans_dump_creds(env, th);
1135 if (unlikely(ldiskfs_track_declares_assert))
1141 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1144 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1145 struct lu_fid *befider)
1147 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1148 memcpy(pack->fp_area, befider, sizeof(*befider));
1149 pack->fp_len = sizeof(*befider) + 1;
1153 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1158 switch (pack->fp_len) {
1159 case sizeof *fid + 1:
1160 memcpy(fid, pack->fp_area, sizeof *fid);
1161 fid_be_to_cpu(fid, fid);
1164 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1171 * Quota/Accounting handling
1173 extern const struct dt_index_operations osd_acct_index_ops;
1174 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1175 const struct lu_fid *fid, struct osd_inode_id *id);
1177 /* copy from fs/ext4/dir.c */
1178 static inline int is_32bit_api(void)
1180 #ifdef CONFIG_COMPAT
1181 return in_compat_syscall();
1183 return (BITS_PER_LONG == 32);
1187 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1189 if ((filp->f_mode & FMODE_32BITHASH) ||
1190 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1191 return LDISKFS_HTREE_EOF_32BIT;
1193 return LDISKFS_HTREE_EOF_64BIT;
1196 static inline int fid_is_internal(const struct lu_fid *fid)
1198 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1201 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1203 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1207 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1208 * return code has been removed and integrated into the pointer in the
1211 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1212 struct inode *inode,
1213 ldiskfs_lblk_t block,
1216 #ifdef HAVE_EXT4_BREAD_4ARGS
1217 return ldiskfs_bread(handle, inode, block, create);
1219 struct buffer_head *bh;
1222 bh = ldiskfs_bread(handle, inode, block, create, &error);
1223 if (bh == NULL && error != 0)
1224 bh = ERR_PTR(error);
1230 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1231 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1233 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1235 #endif /* _OSD_INTERNAL_H */