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 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;
150 /* protects on-fly initialization */
151 int oos_subdir_count; /* subdir count for each seq */
152 struct dentry *oos_root; /* O/<seq> */
153 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
154 u64 oos_seq; /* seq number */
155 struct list_head oos_seq_list; /* list to seq_list */
159 struct dentry *om_root; /* dentry for /O */
160 rwlock_t om_seq_list_lock; /* lock for seq_list */
161 struct list_head om_seq_list; /* list head for seq */
163 struct mutex om_dir_init_mutex;
167 struct dentry *om_root; /* AGENT/<index> */
168 u64 om_index; /* mdt index */
169 struct list_head om_list; /* list to omm_list */
172 struct osd_mdobj_map {
173 struct dentry *omm_remote_parent;
175 int osd_ldiskfs_add_entry(struct osd_thread_info *info,
176 handle_t *handle, struct dentry *child,
177 struct inode *inode, struct htree_lock *hlock);
179 #define OSD_OTABLE_IT_CACHE_SIZE 64
180 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
182 struct osd_inconsistent_item {
183 /* link into osd_scrub::os_inconsistent_items,
184 * protected by osd_scrub::os_lock. */
185 struct list_head oii_list;
187 /* The right FID <=> ino#/gen mapping. */
188 struct osd_idmap_cache oii_cache;
190 unsigned int oii_insert:1; /* insert or update mapping. */
193 struct osd_otable_cache {
194 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
196 /* Index for next cache slot to be filled. */
197 int ooc_producer_idx;
199 /* Index for next cache slot to be returned by it::next(). */
200 int ooc_consumer_idx;
202 /* How many items in ooc_cache. */
203 int ooc_cached_items;
205 /* Position for up layer LFSCK iteration pre-loading. */
206 __u32 ooc_pos_preload;
209 struct osd_otable_it {
210 struct osd_device *ooi_dev;
211 struct osd_otable_cache ooi_cache;
213 /* The following bits can be updated/checked w/o lock protection.
214 * If more bits will be introduced in the future and need lock to
215 * protect, please add comment. */
216 unsigned long ooi_used_outside:1, /* Some user out of OSD
217 * uses the iteration. */
218 ooi_all_cached:1, /* No more entries can be
219 * filled into cache. */
220 ooi_user_ready:1, /* The user out of OSD is
221 * ready to iterate. */
222 ooi_waiting:1; /* it::next is waiting. */
225 struct osd_obj_orphan {
226 struct list_head oor_list;
227 struct lu_env *oor_env; /* to identify "own" records */
236 struct dt_device od_dt_dev;
237 /* information about underlying file system */
238 struct vfsmount *od_mnt;
240 struct osd_oi **od_oi_table;
241 /* total number of OI containers */
246 unsigned int od_fl_capa:1,
254 __u32 od_dirent_journal;
256 struct proc_dir_entry *od_proc_entry;
257 struct lprocfs_stats *od_stats;
259 spinlock_t od_osfs_lock;
262 struct lu_site od_site;
264 struct osd_obj_map *od_ost_map;
265 struct osd_mdobj_map *od_mdt_map;
267 unsigned long long od_readcache_max_filesize;
269 int od_writethrough_cache;
271 struct brw_stats od_brw_stats;
272 atomic_t od_r_in_flight;
273 atomic_t od_w_in_flight;
275 struct mutex od_otable_mutex;
276 struct osd_otable_it *od_otable_it;
277 struct osd_scrub od_scrub;
278 struct list_head od_ios_list;
280 /* service name associated with the osd device */
281 char od_svname[MAX_OBD_NAME];
282 char od_mntdev[MAX_OBD_NAME];
284 /* quota slave instance */
285 struct qsd_instance *od_quota_slave;
287 /* osd seq instance */
288 struct lu_client_seq *od_cl_seq;
289 /* If the ratio of "the total OI mappings count" vs
290 * "the bad OI mappings count" is lower than the
291 * osd_device::od_full_scrub_ratio, then trigger
292 * OI scrub to scan the whole the device. */
293 __u64 od_full_scrub_ratio;
294 /* If the speed of found bad OI mappings (per minute)
295 * exceeds the osd_device::od_full_scrub_threshold_rate,
296 * then trigger OI scrub to scan the whole device. */
297 __u64 od_full_scrub_threshold_rate;
299 /* a list of orphaned agent inodes, protected with od_osfs_lock */
300 struct list_head od_orphan_list;
303 enum osd_full_scrub_ratio {
304 /* Trigger OI scrub to scan the whole device directly. */
307 /* Because the bad OI mappings count cannot be larger than
308 * the total OI mappints count, then setting OFSR_NEVER means
309 * that the whole device scanning cannot be triggered by auto
310 * detected bad OI mappings during the RPC services. */
312 OFSR_DEFAULT = 10000,
315 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
317 /* There are at most 10 uid/gids are affected in a transaction, and
318 * that's rename case:
319 * - 2 for source parent uid & gid;
320 * - 2 for source child uid & gid ('..' entry update when child is directory);
321 * - 2 for target parent uid & gid;
322 * - 2 for target child uid & gid (if the target child exists);
323 * - 2 for root uid & gid (last_rcvd, llog, etc);
325 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
326 * the id type of each id in the ot_id_array.
328 #define OSD_MAX_UGID_CNT 10
333 OSD_OT_XATTR_SET = 2,
346 struct thandle ot_super;
348 struct ldiskfs_journal_cb_entry ot_jcb;
349 struct list_head ot_commit_dcb_list;
350 struct list_head ot_stop_dcb_list;
351 /* Link to the device, for debugging. */
352 struct lu_ref_link ot_dev_link;
353 unsigned short ot_credits;
354 unsigned short ot_id_cnt;
355 unsigned short ot_id_type;
356 int ot_remove_agents:1;
357 uid_t ot_id_array[OSD_MAX_UGID_CNT];
358 struct lquota_trans *ot_quota_trans;
359 #if OSD_THANDLE_STATS
360 /** time when this handle was allocated */
361 cfs_time_t oth_alloced;
363 /** time when this thanle was started */
364 cfs_time_t oth_started;
369 * Basic transaction credit op
390 #ifdef CONFIG_PROC_FS
392 LPROC_OSD_READ_BYTES = 0,
393 LPROC_OSD_WRITE_BYTES = 1,
394 LPROC_OSD_GET_PAGE = 2,
395 LPROC_OSD_NO_PAGE = 3,
396 LPROC_OSD_CACHE_ACCESS = 4,
397 LPROC_OSD_CACHE_HIT = 5,
398 LPROC_OSD_CACHE_MISS = 6,
400 #if OSD_THANDLE_STATS
401 LPROC_OSD_THANDLE_STARTING,
402 LPROC_OSD_THANDLE_OPEN,
403 LPROC_OSD_THANDLE_CLOSING,
410 * Storage representation for fids.
412 * Variable size, first byte contains the length of the whole record.
414 struct osd_fid_pack {
415 unsigned char fp_len;
416 char fp_area[sizeof(struct lu_fid)];
419 struct osd_it_ea_dirent {
420 struct lu_fid oied_fid;
423 unsigned short oied_namelen;
424 unsigned int oied_type;
426 } __attribute__((packed));
429 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
430 * than lu_dirent struct. osd readdir reads less number of dirent than
431 * required for mdd dir page. so buffer size need to be increased so that
432 * there would be one ext3 readdir for every mdd readdir page.
435 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
438 * This is iterator's in-memory data structure in interoperability
439 * mode (i.e. iterator over ldiskfs style directory)
442 struct osd_object *oie_obj;
443 /** used in ldiskfs iterator, to stored file pointer */
444 struct file oie_file;
445 /** how many entries have been read-cached from storage */
447 /** current entry is being iterated by caller */
449 /** current processing entry */
450 struct osd_it_ea_dirent *oie_dirent;
451 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
453 struct dentry oie_dentry;
457 * Iterator's in-memory data structure for IAM mode.
460 struct osd_object *oi_obj;
461 struct iam_path_descr *oi_ipd;
462 struct iam_iterator oi_it;
465 struct osd_quota_leaf {
466 struct list_head oql_link;
471 * Iterator's in-memory data structure for quota file.
473 struct osd_it_quota {
474 struct osd_object *oiq_obj;
475 /** tree blocks path to where the entry is stored */
476 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
477 /** on-disk offset for current key where quota record can be found */
479 /** identifier for current quota record */
481 /** the record index in the leaf/index block */
482 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
483 /** list of already processed leaf blocks */
484 struct list_head oiq_list;
487 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
490 wait_queue_head_t dr_wait;
491 atomic_t dr_numreqs; /* number of reqs being processed */
496 unsigned int dr_ignore_quota:1;
497 unsigned int dr_elapsed_valid:1; /* we really did count time */
498 unsigned int dr_rw:1;
499 struct lu_buf dr_pg_buf;
500 struct page **dr_pages;
501 struct lu_buf dr_bl_buf;
503 unsigned long dr_start_time;
504 unsigned long dr_elapsed; /* how long io took */
505 struct osd_device *dr_dev;
506 unsigned int dr_init_at; /* the line iobuf was initialized */
509 #define OSD_INS_CACHE_SIZE 8
511 struct osd_thread_info {
512 const struct lu_env *oti_env;
514 * used for index operations.
516 struct dentry oti_obj_dentry;
517 struct dentry oti_child_dentry;
519 /** dentry for Iterator context. */
520 struct dentry oti_it_dentry;
523 /* fake struct file for osd_object_sync */
524 struct file oti_file;
526 struct kstatfs oti_ksfs;
529 struct htree_lock *oti_hlock;
531 struct lu_fid oti_fid;
532 struct lu_fid oti_fid2;
533 struct lu_fid oti_fid3;
534 struct osd_inode_id oti_id;
535 struct osd_inode_id oti_id2;
536 struct osd_inode_id oti_id3;
537 struct ost_id oti_ostid;
540 * XXX temporary: for ->i_op calls.
542 struct timespec oti_time;
544 /** osd_device reference, initialized in osd_trans_start() and
545 used in osd_trans_stop() */
546 struct osd_device *oti_dev;
549 * following ipd and it structures are used for osd_index_iam_lookup()
550 * these are defined separately as we might do index operation
551 * in open iterator session.
554 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
556 unsigned int oti_it_ea_buf_used:1;
558 /* IAM iterator for index operation. */
559 struct iam_iterator oti_idx_it;
561 /** union to guarantee that ->oti_ipd[] has proper alignment. */
564 char oti_it_ipd[DX_IPD_MAX_SIZE];
565 long long oti_alignment_lieutenant;
569 char oti_idx_ipd[DX_IPD_MAX_SIZE];
570 long long oti_alignment_lieutenant_colonel;
573 struct osd_idmap_cache oti_cache;
575 /* dedicated OI cache for insert (which needs inum) */
576 struct osd_idmap_cache *oti_ins_cache;
577 int oti_ins_cache_size;
578 int oti_ins_cache_used;
583 /** used in osd_fid_set() to put xattr */
584 struct lu_buf oti_buf;
585 struct lu_buf oti_big_buf;
586 /** used in osd_ea_fid_set() to set fid into common ea */
588 struct lustre_mdt_attrs oti_mdt_attrs;
589 /* old LMA for compatibility */
590 char oti_mdt_attrs_old[LMA_OLD_SIZE];
591 struct filter_fid_old oti_ff;
592 struct filter_fid oti_ff_new;
595 struct osd_iobuf oti_iobuf;
596 /* used to access objects in /O */
597 struct inode *oti_inode;
598 #define OSD_FID_REC_SZ 32
599 char oti_ldp[OSD_FID_REC_SZ];
600 char oti_ldp2[OSD_FID_REC_SZ];
602 /* used by quota code */
604 #ifdef 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 short oti_declare_ops[OSD_OT_MAX];
622 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
623 unsigned short 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)
782 # define osd_journal_start(inode, type, nblocks) \
783 ldiskfs_journal_start(inode, type, nblocks)
784 # define osd_transaction_size(dev) \
785 (osd_journal(dev)->j_max_transaction_buffers / 2)
787 # define LDISKFS_HT_MISC 0
788 # define osd_journal_start_sb(sb, type, nblock) \
789 ldiskfs_journal_start_sb(sb, nblock)
791 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
793 ldiskfs_lblk_t *nblock)
795 struct buffer_head *bh;
798 bh = ldiskfs_append(handle, inode, nblock, &err);
805 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
806 __ldiskfs_find_entry(dir, name, de, lock)
807 # define osd_journal_start(inode, type, nblocks) \
808 ldiskfs_journal_start(inode, nblocks)
809 # define osd_transaction_size(dev) \
810 (osd_journal(dev)->j_max_transaction_buffers)
814 * Invariants, assertions.
818 * XXX: do not enable this, until invariant checking code is made thread safe
819 * in the face of pdirops locking.
821 #define OSD_INVARIANT_CHECKS (0)
823 #if OSD_INVARIANT_CHECKS
824 static inline int osd_invariant(const struct osd_object *obj)
828 ergo(obj->oo_inode != NULL,
829 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
830 atomic_read(&obj->oo_inode->i_count) > 0) &&
831 ergo(obj->oo_dir != NULL &&
832 obj->oo_dir->od_conationer.ic_object != NULL,
833 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
836 #define osd_invariant(obj) (1)
839 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
841 extern const struct dt_index_operations osd_otable_ops;
843 static inline int osd_oi_fid2idx(struct osd_device *dev,
844 const struct lu_fid *fid)
846 return fid->f_seq & (dev->od_oi_count - 1);
849 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
850 const struct lu_fid *fid)
852 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
853 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
854 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
855 DFID"\n", PFID(fid));
856 /* It can work even od_oi_count equals to 1 although it's unexpected,
857 * the only reason we set it to 1 is for performance measurement */
858 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
861 extern const struct lu_device_operations osd_lu_ops;
863 static inline int lu_device_is_osd(const struct lu_device *d)
865 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
868 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
870 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
871 return container_of0(d, struct osd_device, od_dt_dev);
874 static inline struct osd_device *osd_dev(const struct lu_device *d)
876 LASSERT(lu_device_is_osd(d));
877 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
880 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
882 return osd_dev(o->oo_dt.do_lu.lo_dev);
885 static inline struct super_block *osd_sb(const struct osd_device *dev)
887 return dev->od_mnt->mnt_sb;
890 static inline int osd_object_is_root(const struct osd_object *obj)
892 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
895 static inline struct osd_object *osd_obj(const struct lu_object *o)
897 LASSERT(lu_device_is_osd(o->lo_dev));
898 return container_of0(o, struct osd_object, oo_dt.do_lu);
901 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
903 return osd_obj(&d->do_lu);
906 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
908 return &osd->od_dt_dev.dd_lu_dev;
911 static inline journal_t *osd_journal(const struct osd_device *dev)
913 return LDISKFS_SB(osd_sb(dev))->s_journal;
916 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
918 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
921 static inline char *osd_name(struct osd_device *osd)
923 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
926 static inline bool osd_is_ea_inode(struct inode *inode)
928 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
931 extern const struct dt_body_operations osd_body_ops;
932 extern struct lu_context_key osd_key;
934 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
936 return lu_context_key_get(&env->le_ctx, &osd_key);
939 extern const struct dt_body_operations osd_body_ops_new;
945 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
946 const struct iam_container *bag)
948 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
949 osd_oti_get(env)->oti_it_ipd);
953 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
954 const struct iam_container *bag)
956 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
957 osd_oti_get(env)->oti_idx_ipd);
960 static inline void osd_ipd_put(const struct lu_env *env,
961 const struct iam_container *bag,
962 struct iam_path_descr *ipd)
964 bag->ic_descr->id_ops->id_ipd_free(ipd);
967 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
968 const loff_t size, const loff_t pos,
971 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
972 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
973 int write_NUL, loff_t *offs, handle_t *handle);
976 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
978 const char *name, const int namelen)
980 struct osd_thread_info *info = osd_oti_get(env);
981 struct dentry *child_dentry = &info->oti_child_dentry;
982 struct dentry *obj_dentry = &info->oti_obj_dentry;
984 obj_dentry->d_inode = inode;
985 obj_dentry->d_sb = inode->i_sb;
986 obj_dentry->d_name.hash = 0;
988 child_dentry->d_name.hash = 0;
989 child_dentry->d_parent = obj_dentry;
990 child_dentry->d_name.name = name;
991 child_dentry->d_name.len = namelen;
995 extern int osd_trans_declare_op2rb[];
996 extern int ldiskfs_track_declares_assert;
997 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
999 static inline void osd_trans_declare_op(const struct lu_env *env,
1000 struct osd_thandle *oh,
1001 unsigned int op, int credits)
1003 struct osd_thread_info *oti = osd_oti_get(env);
1005 LASSERT(oh->ot_handle == NULL);
1006 if (unlikely(op >= OSD_OT_MAX)) {
1007 if (unlikely(ldiskfs_track_declares_assert)) {
1008 LASSERT(op < OSD_OT_MAX);
1010 CWARN("%s: Invalid operation index %d\n",
1011 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1012 libcfs_debug_dumpstack(NULL);
1015 oti->oti_declare_ops[op]++;
1016 oti->oti_declare_ops_cred[op] += credits;
1018 oh->ot_credits += credits;
1021 static inline void osd_trans_exec_op(const struct lu_env *env,
1022 struct thandle *th, unsigned int op)
1024 struct osd_thread_info *oti = osd_oti_get(env);
1025 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1027 unsigned int rb, left;
1029 LASSERT(oh->ot_handle != NULL);
1030 if (unlikely(op >= OSD_OT_MAX)) {
1031 if (unlikely(ldiskfs_track_declares_assert))
1032 LASSERT(op < OSD_OT_MAX);
1034 CWARN("%s: Invalid operation index %d\n",
1035 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1036 libcfs_debug_dumpstack(NULL);
1041 /* find rollback (or reverse) operation for the given one
1042 * such an operation doesn't require additional credits
1043 * as the same set of blocks are modified */
1044 rb = osd_trans_declare_op2rb[op];
1046 /* check whether credits for this operation were reserved at all */
1047 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1048 oti->oti_declare_ops_cred[rb] == 0)) {
1049 /* the API is not perfect yet: CREATE does REF_ADD internally
1050 * while DESTROY does not. To rollback CREATE the callers
1051 * needs to call REF_DEL+DESTROY which is hard to detect using
1052 * a simple table of rollback operations */
1053 if (op == OSD_OT_REF_DEL &&
1054 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1056 if (op == OSD_OT_REF_ADD &&
1057 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1059 osd_trans_dump_creds(env, th);
1060 CERROR("%s: op = %d, rb = %d\n",
1061 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1062 if (unlikely(ldiskfs_track_declares_assert))
1067 /* remember how many credits we have unused before the operation */
1068 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1069 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1070 if (unlikely(oti->oti_credits_before < left)) {
1071 osd_trans_dump_creds(env, th);
1072 CERROR("%s: op = %d, rb = %d\n",
1073 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1074 /* on a very small fs (testing?) it's possible that
1075 * the transaction can't fit 1/4 of journal, so we
1076 * just request less credits (see osd_trans_start()).
1077 * ignore the same case here */
1078 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1079 if (unlikely(oh->ot_credits < rb)) {
1080 if (unlikely(ldiskfs_track_declares_assert))
1086 static inline void osd_trans_exec_check(const struct lu_env *env,
1090 struct osd_thread_info *oti = osd_oti_get(env);
1091 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1093 int used, over, quota;
1095 /* how many credits have been used by the operation */
1096 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1098 if (unlikely(used < 0)) {
1099 /* if some block was allocated and released in the same
1100 * transaction, then it won't be a part of the transaction
1101 * and delta can be negative */
1106 /* rollback operations (e.g. when we destroy just created
1107 * object) should not consume any credits. there is no point
1108 * to confuse the checks below */
1112 oti->oti_declare_ops_used[op] += used;
1113 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1116 /* we account quota for a whole transaction and any operation can
1117 * consume corresponding credits */
1118 over = oti->oti_declare_ops_used[op] -
1119 oti->oti_declare_ops_cred[op];
1120 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1121 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1122 if (over <= quota) {
1123 /* probably that credits were consumed by
1124 * quota indirectly (in the depths of ldiskfs) */
1125 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1126 oti->oti_declare_ops_used[op] -= over;
1128 CWARN("op %d: used %u, used now %u, reserved %u\n",
1129 op, oti->oti_declare_ops_used[op], used,
1130 oti->oti_declare_ops_cred[op]);
1131 osd_trans_dump_creds(env, th);
1132 if (unlikely(ldiskfs_track_declares_assert))
1138 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1141 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1142 struct lu_fid *befider)
1144 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1145 memcpy(pack->fp_area, befider, sizeof(*befider));
1146 pack->fp_len = sizeof(*befider) + 1;
1150 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1155 switch (pack->fp_len) {
1156 case sizeof *fid + 1:
1157 memcpy(fid, pack->fp_area, sizeof *fid);
1158 fid_be_to_cpu(fid, fid);
1161 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1168 * Quota/Accounting handling
1170 extern const struct dt_index_operations osd_acct_index_ops;
1171 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1172 const struct lu_fid *fid, struct osd_inode_id *id);
1174 /* copy from fs/ext4/dir.c */
1175 static inline int is_32bit_api(void)
1177 #ifdef CONFIG_COMPAT
1178 return is_compat_task();
1180 return (BITS_PER_LONG == 32);
1184 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1186 if ((filp->f_mode & FMODE_32BITHASH) ||
1187 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1188 return LDISKFS_HTREE_EOF_32BIT;
1190 return LDISKFS_HTREE_EOF_64BIT;
1193 static inline int fid_is_internal(const struct lu_fid *fid)
1195 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1198 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1200 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1203 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1204 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1206 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1208 #endif /* _OSD_INTERNAL_H */