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;
152 struct list_head oo_xattr_list;
156 /* protects on-fly initialization */
157 int oos_subdir_count; /* subdir count for each seq */
158 struct dentry *oos_root; /* O/<seq> */
159 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
160 u64 oos_seq; /* seq number */
161 struct list_head oos_seq_list; /* list to seq_list */
165 struct dentry *om_root; /* dentry for /O */
166 rwlock_t om_seq_list_lock; /* lock for seq_list */
167 struct list_head om_seq_list; /* list head for seq */
169 struct mutex om_dir_init_mutex;
173 struct dentry *om_root; /* AGENT/<index> */
174 u64 om_index; /* mdt index */
175 struct list_head om_list; /* list to omm_list */
178 struct osd_mdobj_map {
179 struct dentry *omm_remote_parent;
181 int osd_ldiskfs_add_entry(struct osd_thread_info *info,
182 handle_t *handle, struct dentry *child,
183 struct inode *inode, struct htree_lock *hlock);
185 #define OSD_OTABLE_IT_CACHE_SIZE 64
186 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
188 struct osd_inconsistent_item {
189 /* link into osd_scrub::os_inconsistent_items,
190 * protected by osd_scrub::os_lock. */
191 struct list_head oii_list;
193 /* The right FID <=> ino#/gen mapping. */
194 struct osd_idmap_cache oii_cache;
196 unsigned int oii_insert:1; /* insert or update mapping. */
199 struct osd_otable_cache {
200 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
202 /* Index for next cache slot to be filled. */
203 int ooc_producer_idx;
205 /* Index for next cache slot to be returned by it::next(). */
206 int ooc_consumer_idx;
208 /* How many items in ooc_cache. */
209 int ooc_cached_items;
211 /* Position for up layer LFSCK iteration pre-loading. */
212 __u32 ooc_pos_preload;
215 struct osd_otable_it {
216 struct osd_device *ooi_dev;
217 struct osd_otable_cache ooi_cache;
219 /* The following bits can be updated/checked w/o lock protection.
220 * If more bits will be introduced in the future and need lock to
221 * protect, please add comment. */
222 unsigned long ooi_used_outside:1, /* Some user out of OSD
223 * uses the iteration. */
224 ooi_all_cached:1, /* No more entries can be
225 * filled into cache. */
226 ooi_user_ready:1, /* The user out of OSD is
227 * ready to iterate. */
228 ooi_waiting:1; /* it::next is waiting. */
231 struct osd_obj_orphan {
232 struct list_head oor_list;
233 struct lu_env *oor_env; /* to identify "own" records */
242 struct dt_device od_dt_dev;
243 /* information about underlying file system */
244 struct vfsmount *od_mnt;
246 struct osd_oi **od_oi_table;
247 /* total number of OI containers */
252 unsigned int od_fl_capa:1,
260 __u32 od_dirent_journal;
262 struct proc_dir_entry *od_proc_entry;
263 struct lprocfs_stats *od_stats;
265 spinlock_t od_osfs_lock;
268 struct lu_site od_site;
270 struct osd_obj_map *od_ost_map;
271 struct osd_mdobj_map *od_mdt_map;
273 unsigned long long od_readcache_max_filesize;
275 int od_writethrough_cache;
277 struct brw_stats od_brw_stats;
278 atomic_t od_r_in_flight;
279 atomic_t od_w_in_flight;
281 struct mutex od_otable_mutex;
282 struct osd_otable_it *od_otable_it;
283 struct osd_scrub od_scrub;
284 struct list_head od_ios_list;
286 /* service name associated with the osd device */
287 char od_svname[MAX_OBD_NAME];
288 char od_mntdev[MAX_OBD_NAME];
290 /* quota slave instance */
291 struct qsd_instance *od_quota_slave;
293 /* osd seq instance */
294 struct lu_client_seq *od_cl_seq;
295 /* If the ratio of "the total OI mappings count" vs
296 * "the bad OI mappings count" is lower than the
297 * osd_device::od_full_scrub_ratio, then trigger
298 * OI scrub to scan the whole the device. */
299 __u64 od_full_scrub_ratio;
300 /* If the speed of found bad OI mappings (per minute)
301 * exceeds the osd_device::od_full_scrub_threshold_rate,
302 * then trigger OI scrub to scan the whole device. */
303 __u64 od_full_scrub_threshold_rate;
305 /* a list of orphaned agent inodes, protected with od_osfs_lock */
306 struct list_head od_orphan_list;
309 enum osd_full_scrub_ratio {
310 /* Trigger OI scrub to scan the whole device directly. */
313 /* Because the bad OI mappings count cannot be larger than
314 * the total OI mappints count, then setting OFSR_NEVER means
315 * that the whole device scanning cannot be triggered by auto
316 * detected bad OI mappings during the RPC services. */
318 OFSR_DEFAULT = 10000,
321 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
323 /* There are at most 10 uid/gids are affected in a transaction, and
324 * that's rename case:
325 * - 2 for source parent uid & gid;
326 * - 2 for source child uid & gid ('..' entry update when child is directory);
327 * - 2 for target parent uid & gid;
328 * - 2 for target child uid & gid (if the target child exists);
329 * - 2 for root uid & gid (last_rcvd, llog, etc);
331 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
332 * the id type of each id in the ot_id_array.
334 #define OSD_MAX_UGID_CNT 10
339 OSD_OT_XATTR_SET = 2,
352 struct thandle ot_super;
354 struct ldiskfs_journal_cb_entry ot_jcb;
355 struct list_head ot_commit_dcb_list;
356 struct list_head ot_stop_dcb_list;
357 /* Link to the device, for debugging. */
358 struct lu_ref_link ot_dev_link;
359 unsigned short ot_credits;
360 unsigned short ot_id_cnt;
361 unsigned short ot_id_type;
362 unsigned int ot_remove_agents:1;
363 uid_t ot_id_array[OSD_MAX_UGID_CNT];
364 struct lquota_trans *ot_quota_trans;
365 #if OSD_THANDLE_STATS
366 /** time when this handle was allocated */
367 cfs_time_t oth_alloced;
369 /** time when this thanle was started */
370 cfs_time_t oth_started;
375 * Basic transaction credit op
396 #ifdef CONFIG_PROC_FS
398 LPROC_OSD_READ_BYTES = 0,
399 LPROC_OSD_WRITE_BYTES = 1,
400 LPROC_OSD_GET_PAGE = 2,
401 LPROC_OSD_NO_PAGE = 3,
402 LPROC_OSD_CACHE_ACCESS = 4,
403 LPROC_OSD_CACHE_HIT = 5,
404 LPROC_OSD_CACHE_MISS = 6,
406 #if OSD_THANDLE_STATS
407 LPROC_OSD_THANDLE_STARTING,
408 LPROC_OSD_THANDLE_OPEN,
409 LPROC_OSD_THANDLE_CLOSING,
416 * Storage representation for fids.
418 * Variable size, first byte contains the length of the whole record.
420 struct osd_fid_pack {
421 unsigned char fp_len;
422 char fp_area[sizeof(struct lu_fid)];
425 struct osd_it_ea_dirent {
426 struct lu_fid oied_fid;
429 unsigned short oied_namelen;
430 unsigned int oied_type;
432 } __attribute__((packed));
435 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
436 * than lu_dirent struct. osd readdir reads less number of dirent than
437 * required for mdd dir page. so buffer size need to be increased so that
438 * there would be one ext3 readdir for every mdd readdir page.
441 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
444 * This is iterator's in-memory data structure in interoperability
445 * mode (i.e. iterator over ldiskfs style directory)
448 struct osd_object *oie_obj;
449 /** used in ldiskfs iterator, to stored file pointer */
450 struct file oie_file;
451 /** how many entries have been read-cached from storage */
453 /** current entry is being iterated by caller */
455 /** current processing entry */
456 struct osd_it_ea_dirent *oie_dirent;
457 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
459 struct dentry oie_dentry;
463 * Iterator's in-memory data structure for IAM mode.
466 struct osd_object *oi_obj;
467 struct iam_path_descr *oi_ipd;
468 struct iam_iterator oi_it;
471 struct osd_quota_leaf {
472 struct list_head oql_link;
477 * Iterator's in-memory data structure for quota file.
479 struct osd_it_quota {
480 struct osd_object *oiq_obj;
481 /** tree blocks path to where the entry is stored */
482 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
483 /** on-disk offset for current key where quota record can be found */
485 /** identifier for current quota record */
487 /** the record index in the leaf/index block */
488 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
489 /** list of already processed leaf blocks */
490 struct list_head oiq_list;
493 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
496 wait_queue_head_t dr_wait;
497 atomic_t dr_numreqs; /* number of reqs being processed */
502 unsigned int dr_ignore_quota:1;
503 unsigned int dr_elapsed_valid:1; /* we really did count time */
504 unsigned int dr_rw:1;
505 struct lu_buf dr_pg_buf;
506 struct page **dr_pages;
507 struct lu_buf dr_bl_buf;
509 unsigned long dr_start_time;
510 unsigned long dr_elapsed; /* how long io took */
511 struct osd_device *dr_dev;
512 unsigned int dr_init_at; /* the line iobuf was initialized */
515 #define OSD_INS_CACHE_SIZE 8
517 struct osd_thread_info {
518 const struct lu_env *oti_env;
520 * used for index operations.
522 struct dentry oti_obj_dentry;
523 struct dentry oti_child_dentry;
525 /** dentry for Iterator context. */
526 struct dentry oti_it_dentry;
529 /* fake struct file for osd_object_sync */
530 struct file oti_file;
532 struct kstatfs oti_ksfs;
535 struct htree_lock *oti_hlock;
537 struct lu_fid oti_fid;
538 struct lu_fid oti_fid2;
539 struct lu_fid oti_fid3;
540 struct osd_inode_id oti_id;
541 struct osd_inode_id oti_id2;
542 struct osd_inode_id oti_id3;
543 struct ost_id oti_ostid;
546 * XXX temporary: for ->i_op calls.
548 struct timespec oti_time;
550 /** osd_device reference, initialized in osd_trans_start() and
551 used in osd_trans_stop() */
552 struct osd_device *oti_dev;
555 * following ipd and it structures are used for osd_index_iam_lookup()
556 * these are defined separately as we might do index operation
557 * in open iterator session.
560 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
562 unsigned int oti_it_ea_buf_used:1;
564 /* IAM iterator for index operation. */
565 struct iam_iterator oti_idx_it;
567 /** union to guarantee that ->oti_ipd[] has proper alignment. */
570 char oti_it_ipd[DX_IPD_MAX_SIZE];
571 long long oti_alignment_lieutenant;
575 char oti_idx_ipd[DX_IPD_MAX_SIZE];
576 long long oti_alignment_lieutenant_colonel;
579 struct osd_idmap_cache oti_cache;
581 /* dedicated OI cache for insert (which needs inum) */
582 struct osd_idmap_cache *oti_ins_cache;
583 int oti_ins_cache_size;
584 int oti_ins_cache_used;
589 /** used in osd_fid_set() to put xattr */
590 struct lu_buf oti_buf;
591 struct lu_buf oti_big_buf;
592 /** used in osd_ea_fid_set() to set fid into common ea */
594 struct lustre_mdt_attrs oti_mdt_attrs;
595 /* old LMA for compatibility */
596 char oti_mdt_attrs_old[LMA_OLD_SIZE];
597 struct filter_fid_old oti_ff;
598 struct filter_fid oti_ff_new;
601 struct osd_iobuf oti_iobuf;
602 /* used to access objects in /O */
603 struct inode *oti_inode;
604 #define OSD_FID_REC_SZ 32
605 char oti_ldp[OSD_FID_REC_SZ];
606 char oti_ldp2[OSD_FID_REC_SZ];
608 /* used by quota code */
610 #if defined(HAVE_DQUOT_QC_DQBLK)
611 struct qc_dqblk oti_qdq;
612 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
613 struct fs_disk_quota oti_fdq;
615 struct if_dqblk oti_dqblk;
617 struct if_dqinfo oti_dqinfo;
619 struct lquota_id_info oti_qi;
620 struct lquota_trans oti_quota_trans;
621 union lquota_rec oti_quota_rec;
623 struct lu_seq_range oti_seq_range;
625 /* Tracking for transaction credits, to allow debugging and optimizing
626 * cases where a large number of credits are being allocated for
627 * single transaction. */
628 unsigned int oti_credits_before;
629 unsigned short oti_declare_ops[OSD_OT_MAX];
630 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
631 unsigned short oti_declare_ops_used[OSD_OT_MAX];
634 extern int ldiskfs_pdo;
636 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
637 const char *name, void *buf, int len)
642 dentry->d_inode = inode;
643 dentry->d_sb = inode->i_sb;
644 return inode->i_op->getxattr(dentry, name, buf, len);
647 static inline int __osd_xattr_set(struct osd_thread_info *info,
648 struct inode *inode, const char *name,
649 const void *buf, int buflen, int fl)
651 struct dentry *dentry = &info->oti_child_dentry;
653 ll_vfs_dq_init(inode);
654 dentry->d_inode = inode;
655 dentry->d_sb = inode->i_sb;
656 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
659 #ifdef CONFIG_PROC_FS
661 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
662 extern struct lprocfs_vars lprocfs_osd_module_vars[];
663 int osd_procfs_init(struct osd_device *osd, const char *name);
664 int osd_procfs_fini(struct osd_device *osd);
665 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
666 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
667 int osd_register_proc_index_in_idif(struct osd_device *osd);
671 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
672 struct obd_statfs *sfs);
673 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
674 struct osd_inode_id *id);
675 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
676 const struct lu_fid *fid, __u32 compat, __u32 incompat);
677 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
678 struct dentry *dentry, struct lustre_mdt_attrs *lma);
679 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
680 struct osd_inode_id *id, const struct lu_fid *fid);
681 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
682 struct dentry *dentry, struct lu_fid *fid);
684 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
685 void osd_obj_map_fini(struct osd_device *dev);
686 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
687 const struct lu_fid *fid, struct osd_inode_id *id);
688 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
689 const struct lu_fid *fid, const struct osd_inode_id *id,
691 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
692 const struct lu_fid *fid, handle_t *th);
693 int osd_obj_map_update(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_map_recover(struct osd_thread_info *info, struct osd_device *osd,
697 struct inode *src_parent, struct dentry *src_child,
698 const struct lu_fid *fid);
699 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
700 const struct lu_fid *fid, struct osd_inode_id *id);
701 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
702 const struct lu_fid *fid, const struct osd_inode_id *id,
704 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
705 const struct lu_fid *fid, const struct osd_inode_id *id,
708 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
709 int osd_scrub_file_store(struct osd_scrub *scrub);
710 char *osd_lf_fid2name(const struct lu_fid *fid);
711 int osd_scrub_start(struct osd_device *dev, __u32 flags);
712 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
713 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
714 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
716 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
717 struct osd_inode_id *id);
718 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
720 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
721 u64 seq, struct lu_seq_range *range);
723 int osd_delete_from_remote_parent(const struct lu_env *env,
724 struct osd_device *osd,
725 struct osd_object *obj,
726 struct osd_thandle *oh);
727 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
728 struct osd_object *obj, struct osd_thandle *oh);
729 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
730 struct osd_device *osd,
731 const struct lu_fid *fid,
732 struct osd_inode_id *id);
734 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
736 /* osd_quota_fmt.c */
737 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
738 int type, uint blk, int depth, uint index,
739 struct osd_it_quota *it);
740 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
741 int type, uint blk, uint index,
742 struct osd_it_quota *it);
743 loff_t find_tree_dqentry(const struct lu_env *env,
744 struct osd_object *obj, int type,
745 qid_t dqid, uint blk, int depth,
746 struct osd_it_quota *it);
748 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
749 struct lquota_id_info *qi, struct osd_object *obj,
750 bool enforce, int *flags);
751 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
752 long long space, struct osd_thandle *oh,
753 struct osd_object *obj, bool is_blk, int *flags,
755 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
756 const struct dt_rec *rec,
757 union lquota_rec *quota_rec);
758 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
759 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
761 #ifndef HAVE_I_UID_READ
762 static inline uid_t i_uid_read(const struct inode *inode)
767 static inline gid_t i_gid_read(const struct inode *inode)
772 static inline void i_uid_write(struct inode *inode, uid_t uid)
777 static inline void i_gid_write(struct inode *inode, gid_t gid)
783 #ifdef LDISKFS_HT_MISC
784 # define osd_journal_start_sb(sb, type, nblock) \
785 ldiskfs_journal_start_sb(sb, type, nblock)
786 # define osd_ldiskfs_append(handle, inode, nblock) \
787 ldiskfs_append(handle, inode, nblock)
788 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
789 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
791 # define osd_journal_start(inode, type, nblocks) \
792 ldiskfs_journal_start(inode, type, nblocks)
793 # define osd_transaction_size(dev) \
794 (osd_journal(dev)->j_max_transaction_buffers / 2)
796 # define LDISKFS_HT_MISC 0
797 # define osd_journal_start_sb(sb, type, nblock) \
798 ldiskfs_journal_start_sb(sb, nblock)
800 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
802 ldiskfs_lblk_t *nblock)
804 struct buffer_head *bh;
807 bh = ldiskfs_append(handle, inode, nblock, &err);
814 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
815 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
817 # define osd_journal_start(inode, type, nblocks) \
818 ldiskfs_journal_start(inode, nblocks)
819 # define osd_transaction_size(dev) \
820 (osd_journal(dev)->j_max_transaction_buffers)
824 * Invariants, assertions.
828 * XXX: do not enable this, until invariant checking code is made thread safe
829 * in the face of pdirops locking.
831 #define OSD_INVARIANT_CHECKS (0)
833 #if OSD_INVARIANT_CHECKS
834 static inline int osd_invariant(const struct osd_object *obj)
838 ergo(obj->oo_inode != NULL,
839 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
840 atomic_read(&obj->oo_inode->i_count) > 0) &&
841 ergo(obj->oo_dir != NULL &&
842 obj->oo_dir->od_conationer.ic_object != NULL,
843 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
846 #define osd_invariant(obj) (1)
849 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
851 extern const struct dt_index_operations osd_otable_ops;
853 static inline int osd_oi_fid2idx(struct osd_device *dev,
854 const struct lu_fid *fid)
856 return fid->f_seq & (dev->od_oi_count - 1);
859 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
860 const struct lu_fid *fid)
862 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
863 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
864 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
865 DFID"\n", PFID(fid));
866 /* It can work even od_oi_count equals to 1 although it's unexpected,
867 * the only reason we set it to 1 is for performance measurement */
868 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
871 extern const struct lu_device_operations osd_lu_ops;
873 static inline int lu_device_is_osd(const struct lu_device *d)
875 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
878 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
880 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
881 return container_of0(d, struct osd_device, od_dt_dev);
884 static inline struct osd_device *osd_dev(const struct lu_device *d)
886 LASSERT(lu_device_is_osd(d));
887 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
890 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
892 return osd_dev(o->oo_dt.do_lu.lo_dev);
895 static inline struct super_block *osd_sb(const struct osd_device *dev)
897 return dev->od_mnt->mnt_sb;
900 static inline int osd_object_is_root(const struct osd_object *obj)
902 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
905 static inline struct osd_object *osd_obj(const struct lu_object *o)
907 LASSERT(lu_device_is_osd(o->lo_dev));
908 return container_of0(o, struct osd_object, oo_dt.do_lu);
911 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
913 return osd_obj(&d->do_lu);
916 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
918 return &osd->od_dt_dev.dd_lu_dev;
921 static inline journal_t *osd_journal(const struct osd_device *dev)
923 return LDISKFS_SB(osd_sb(dev))->s_journal;
926 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
928 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
931 static inline char *osd_name(struct osd_device *osd)
933 return osd->od_svname;
936 static inline bool osd_is_ea_inode(struct inode *inode)
938 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
941 extern const struct dt_body_operations osd_body_ops;
942 extern struct lu_context_key osd_key;
944 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
946 return lu_context_key_get(&env->le_ctx, &osd_key);
949 extern const struct dt_body_operations osd_body_ops_new;
955 struct iam_path_descr *osd_it_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_it_ipd);
963 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
964 const struct iam_container *bag)
966 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
967 osd_oti_get(env)->oti_idx_ipd);
970 static inline void osd_ipd_put(const struct lu_env *env,
971 const struct iam_container *bag,
972 struct iam_path_descr *ipd)
974 bag->ic_descr->id_ops->id_ipd_free(ipd);
977 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
978 const loff_t size, const loff_t pos,
981 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
982 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
983 int write_NUL, loff_t *offs, handle_t *handle);
986 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
988 const char *name, const int namelen)
990 struct osd_thread_info *info = osd_oti_get(env);
991 struct dentry *child_dentry = &info->oti_child_dentry;
992 struct dentry *obj_dentry = &info->oti_obj_dentry;
994 obj_dentry->d_inode = inode;
995 obj_dentry->d_sb = inode->i_sb;
996 obj_dentry->d_name.hash = 0;
998 child_dentry->d_name.hash = 0;
999 child_dentry->d_parent = obj_dentry;
1000 child_dentry->d_name.name = name;
1001 child_dentry->d_name.len = namelen;
1002 return child_dentry;
1005 extern int osd_trans_declare_op2rb[];
1006 extern int ldiskfs_track_declares_assert;
1007 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1009 static inline void osd_trans_declare_op(const struct lu_env *env,
1010 struct osd_thandle *oh,
1011 unsigned int op, int credits)
1013 struct osd_thread_info *oti = osd_oti_get(env);
1015 LASSERT(oh->ot_handle == NULL);
1016 if (unlikely(op >= OSD_OT_MAX)) {
1017 if (unlikely(ldiskfs_track_declares_assert)) {
1018 LASSERT(op < OSD_OT_MAX);
1020 CWARN("%s: Invalid operation index %d\n",
1021 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1022 libcfs_debug_dumpstack(NULL);
1025 oti->oti_declare_ops[op]++;
1026 oti->oti_declare_ops_cred[op] += credits;
1028 oh->ot_credits += credits;
1031 static inline void osd_trans_exec_op(const struct lu_env *env,
1032 struct thandle *th, unsigned int op)
1034 struct osd_thread_info *oti = osd_oti_get(env);
1035 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1037 unsigned int rb, left;
1039 LASSERT(oh->ot_handle != NULL);
1040 if (unlikely(op >= OSD_OT_MAX)) {
1041 if (unlikely(ldiskfs_track_declares_assert))
1042 LASSERT(op < OSD_OT_MAX);
1044 CWARN("%s: opcode %u: invalid value >= %u\n",
1045 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1047 libcfs_debug_dumpstack(NULL);
1052 /* find rollback (or reverse) operation for the given one
1053 * such an operation doesn't require additional credits
1054 * as the same set of blocks are modified */
1055 rb = osd_trans_declare_op2rb[op];
1057 /* check whether credits for this operation were reserved at all */
1058 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1059 oti->oti_declare_ops_cred[rb] == 0)) {
1060 /* the API is not perfect yet: CREATE does REF_ADD internally
1061 * while DESTROY does not. To rollback CREATE the callers
1062 * needs to call REF_DEL+DESTROY which is hard to detect using
1063 * a simple table of rollback operations */
1064 if (op == OSD_OT_REF_DEL &&
1065 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1067 if (op == OSD_OT_REF_ADD &&
1068 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1070 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1071 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1072 osd_trans_dump_creds(env, th);
1073 LASSERT(!ldiskfs_track_declares_assert);
1077 /* remember how many credits we have unused before the operation */
1078 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1079 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1080 if (unlikely(oti->oti_credits_before < left)) {
1081 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1082 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1083 oti->oti_credits_before, left, rb);
1084 osd_trans_dump_creds(env, th);
1085 /* on a very small fs (testing?) it's possible that
1086 * the transaction can't fit 1/4 of journal, so we
1087 * just request less credits (see osd_trans_start()).
1088 * ignore the same case here */
1089 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1090 if (unlikely(oh->ot_credits < rb))
1091 LASSERT(!ldiskfs_track_declares_assert);
1095 static inline void osd_trans_exec_check(const struct lu_env *env,
1099 struct osd_thread_info *oti = osd_oti_get(env);
1100 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1102 int used, over, quota;
1104 /* how many credits have been used by the operation */
1105 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1107 if (unlikely(used < 0)) {
1108 /* if some block was allocated and released in the same
1109 * transaction, then it won't be a part of the transaction
1110 * and delta can be negative */
1115 /* rollback operations (e.g. when we destroy just created
1116 * object) should not consume any credits. there is no point
1117 * to confuse the checks below */
1121 oti->oti_declare_ops_used[op] += used;
1122 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1125 /* we account quota for a whole transaction and any operation can
1126 * consume corresponding credits */
1127 over = oti->oti_declare_ops_used[op] -
1128 oti->oti_declare_ops_cred[op];
1129 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1130 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1131 if (over <= quota) {
1132 /* probably that credits were consumed by
1133 * quota indirectly (in the depths of ldiskfs) */
1134 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1135 oti->oti_declare_ops_used[op] -= over;
1137 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1138 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1139 oti->oti_declare_ops_used[op], used,
1140 oti->oti_declare_ops_cred[op]);
1141 osd_trans_dump_creds(env, th);
1142 if (unlikely(ldiskfs_track_declares_assert))
1148 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1151 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1152 struct lu_fid *befider)
1154 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1155 memcpy(pack->fp_area, befider, sizeof(*befider));
1156 pack->fp_len = sizeof(*befider) + 1;
1160 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1165 switch (pack->fp_len) {
1166 case sizeof *fid + 1:
1167 memcpy(fid, pack->fp_area, sizeof *fid);
1168 fid_be_to_cpu(fid, fid);
1171 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1178 * Quota/Accounting handling
1180 extern const struct dt_index_operations osd_acct_index_ops;
1181 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1182 const struct lu_fid *fid, struct osd_inode_id *id);
1184 /* copy from fs/ext4/dir.c */
1185 static inline int is_32bit_api(void)
1187 #ifdef CONFIG_COMPAT
1188 return in_compat_syscall();
1190 return (BITS_PER_LONG == 32);
1194 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1196 if ((filp->f_mode & FMODE_32BITHASH) ||
1197 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1198 return LDISKFS_HTREE_EOF_32BIT;
1200 return LDISKFS_HTREE_EOF_64BIT;
1203 static inline int fid_is_internal(const struct lu_fid *fid)
1205 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1208 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1210 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1214 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1215 * return code has been removed and integrated into the pointer in the
1218 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1219 struct inode *inode,
1220 ldiskfs_lblk_t block,
1223 #ifdef HAVE_EXT4_BREAD_4ARGS
1224 return ldiskfs_bread(handle, inode, block, create);
1226 struct buffer_head *bh;
1229 bh = ldiskfs_bread(handle, inode, block, create, &error);
1230 if (bh == NULL && error != 0)
1231 bh = ERR_PTR(error);
1237 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1238 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1240 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1242 #endif /* _OSD_INTERNAL_H */