4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_internal.h
38 * Shared definitions and declarations for osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
43 #ifndef _OSD_INTERNAL_H
44 #define _OSD_INTERNAL_H
48 #include <linux/mutex.h>
49 /* struct rw_semaphore */
50 #include <linux/rwsem.h>
52 #include <linux/dcache.h>
54 #include <linux/dirent.h>
55 #include <linux/statfs.h>
56 #include <ldiskfs/ldiskfs.h>
57 #include <ldiskfs/ldiskfs_jbd2.h>
61 /* class_register_type(), class_unregister_type(), class_get_type() */
62 #include <obd_class.h>
63 #include <lustre_disk.h>
64 #include <dt_object.h>
65 #include <lustre_quota.h>
69 #include "osd_scrub.h"
70 #include "osd_quota_fmt.h"
73 extern struct kmem_cache *dynlock_cachep;
75 #define OSD_COUNTERS (0)
77 /* ldiskfs special inode::i_state_flags need to be accessed with
78 * ldiskfs_{set,clear,test}_inode_state() only */
80 /* OI scrub should skip this inode. */
81 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
83 /** Enable thandle usage statistics */
84 #define OSD_THANDLE_STATS (0)
86 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
88 #define OBJECTS "OBJECTS"
89 #define ADMIN_USR "admin_quotafile_v2.usr"
90 #define ADMIN_GRP "admin_quotafile_v2.grp"
92 struct osd_directory {
93 struct iam_container od_container;
94 struct iam_descr od_descr;
98 * Object Index (oi) instance.
102 * underlying index object, where fid->id mapping in stored.
104 struct inode *oi_inode;
105 struct osd_directory oi_dir;
108 extern const int osd_dto_credits_noquota[];
111 struct dt_object oo_dt;
113 * Inode for file system object represented by this osd_object. This
114 * inode is pinned for the whole duration of lu_object life.
116 * Not modified concurrently (either setup early during object
117 * creation, or assigned by osd_object_create() under write lock).
119 struct inode *oo_inode;
121 * to protect index ops.
123 struct htree_lock_head *oo_hl_head;
124 struct rw_semaphore oo_ext_idx_sem;
125 struct rw_semaphore oo_sem;
126 struct osd_directory *oo_dir;
127 /** protects inode attributes. */
130 __u32 oo_destroyed:1;
132 * Following two members are used to indicate the presence of dot and
133 * dotdot in the given directory. This is required for interop mode
136 int oo_compat_dot_created;
137 int oo_compat_dotdot_created;
139 const struct lu_env *oo_owner;
140 #ifdef CONFIG_LOCKDEP
141 struct lockdep_map oo_dep_map;
146 /* protects on-fly initialization */
147 int oos_subdir_count; /* subdir count for each seq */
148 struct dentry *oos_root; /* O/<seq> */
149 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
150 u64 oos_seq; /* seq number */
151 struct list_head oos_seq_list; /* list to seq_list */
155 struct dentry *om_root; /* dentry for /O */
156 rwlock_t om_seq_list_lock; /* lock for seq_list */
157 struct list_head om_seq_list; /* list head for seq */
159 struct mutex om_dir_init_mutex;
163 struct dentry *om_root; /* AGENT/<index> */
164 u64 om_index; /* mdt index */
165 struct list_head om_list; /* list to omm_list */
168 struct osd_mdobj_map {
169 struct dentry *omm_remote_parent;
172 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
173 __ldiskfs_add_entry(handle, child, cinode, hlock)
175 #define OSD_OTABLE_IT_CACHE_SIZE 64
176 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
178 struct osd_inconsistent_item {
179 /* link into osd_scrub::os_inconsistent_items,
180 * protected by osd_scrub::os_lock. */
181 struct list_head oii_list;
183 /* The right FID <=> ino#/gen mapping. */
184 struct osd_idmap_cache oii_cache;
186 unsigned int oii_insert:1; /* insert or update mapping. */
189 struct osd_otable_cache {
190 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
192 /* Index for next cache slot to be filled. */
193 int ooc_producer_idx;
195 /* Index for next cache slot to be returned by it::next(). */
196 int ooc_consumer_idx;
198 /* How many items in ooc_cache. */
199 int ooc_cached_items;
201 /* Position for up layer LFSCK iteration pre-loading. */
202 __u32 ooc_pos_preload;
205 struct osd_otable_it {
206 struct osd_device *ooi_dev;
207 struct osd_otable_cache ooi_cache;
209 /* The following bits can be updated/checked w/o lock protection.
210 * If more bits will be introduced in the future and need lock to
211 * protect, please add comment. */
212 unsigned long ooi_used_outside:1, /* Some user out of OSD
213 * uses the iteration. */
214 ooi_all_cached:1, /* No more entries can be
215 * filled into cache. */
216 ooi_user_ready:1, /* The user out of OSD is
217 * ready to iterate. */
218 ooi_waiting:1; /* it::next is waiting. */
221 struct osd_obj_orphan {
222 struct list_head oor_list;
223 struct lu_env *oor_env; /* to identify "own" records */
232 struct dt_device od_dt_dev;
233 /* information about underlying file system */
234 struct vfsmount *od_mnt;
236 struct osd_oi **od_oi_table;
237 /* total number of OI containers */
242 unsigned int od_fl_capa:1,
250 __u32 od_dirent_journal;
252 struct proc_dir_entry *od_proc_entry;
253 struct lprocfs_stats *od_stats;
255 spinlock_t od_osfs_lock;
258 struct lu_site od_site;
260 struct osd_obj_map *od_ost_map;
261 struct osd_mdobj_map *od_mdt_map;
263 unsigned long long od_readcache_max_filesize;
265 int od_writethrough_cache;
267 struct brw_stats od_brw_stats;
268 atomic_t od_r_in_flight;
269 atomic_t od_w_in_flight;
271 struct mutex od_otable_mutex;
272 struct osd_otable_it *od_otable_it;
273 struct osd_scrub od_scrub;
274 struct list_head od_ios_list;
276 /* service name associated with the osd device */
277 char od_svname[MAX_OBD_NAME];
278 char od_mntdev[MAX_OBD_NAME];
280 /* quota slave instance */
281 struct qsd_instance *od_quota_slave;
283 /* osd seq instance */
284 struct lu_client_seq *od_cl_seq;
285 /* If the ratio of "the total OI mappings count" vs
286 * "the bad OI mappings count" is lower than the
287 * osd_device::od_full_scrub_ratio, then trigger
288 * OI scrub to scan the whole the device. */
289 __u64 od_full_scrub_ratio;
290 /* If the speed of found bad OI mappings (per minute)
291 * exceeds the osd_device::od_full_scrub_threshold_rate,
292 * then trigger OI scrub to scan the whole device. */
293 __u64 od_full_scrub_threshold_rate;
295 /* a list of orphaned agent inodes, protected with od_osfs_lock */
296 struct list_head od_orphan_list;
299 enum osd_full_scrub_ratio {
300 /* Trigger OI scrub to scan the whole device directly. */
303 /* Because the bad OI mappings count cannot be larger than
304 * the total OI mappints count, then setting OFSR_NEVER means
305 * that the whole device scanning cannot be triggered by auto
306 * detected bad OI mappings during the RPC services. */
308 OFSR_DEFAULT = 10000,
311 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
313 /* There are at most 10 uid/gids are affected in a transaction, and
314 * that's rename case:
315 * - 2 for source parent uid & gid;
316 * - 2 for source child uid & gid ('..' entry update when child is directory);
317 * - 2 for target parent uid & gid;
318 * - 2 for target child uid & gid (if the target child exists);
319 * - 2 for root uid & gid (last_rcvd, llog, etc);
321 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
322 * the id type of each id in the ot_id_array.
324 #define OSD_MAX_UGID_CNT 10
329 OSD_OT_XATTR_SET = 2,
342 struct thandle ot_super;
344 struct ldiskfs_journal_cb_entry ot_jcb;
345 struct list_head ot_commit_dcb_list;
346 struct list_head ot_stop_dcb_list;
347 /* Link to the device, for debugging. */
348 struct lu_ref_link ot_dev_link;
349 unsigned short ot_credits;
350 unsigned short ot_id_cnt;
351 unsigned short ot_id_type;
352 int ot_remove_agents:1;
353 uid_t ot_id_array[OSD_MAX_UGID_CNT];
354 struct lquota_trans *ot_quota_trans;
355 #if OSD_THANDLE_STATS
356 /** time when this handle was allocated */
357 cfs_time_t oth_alloced;
359 /** time when this thanle was started */
360 cfs_time_t oth_started;
365 * Basic transaction credit op
386 #ifdef CONFIG_PROC_FS
388 LPROC_OSD_READ_BYTES = 0,
389 LPROC_OSD_WRITE_BYTES = 1,
390 LPROC_OSD_GET_PAGE = 2,
391 LPROC_OSD_NO_PAGE = 3,
392 LPROC_OSD_CACHE_ACCESS = 4,
393 LPROC_OSD_CACHE_HIT = 5,
394 LPROC_OSD_CACHE_MISS = 6,
396 #if OSD_THANDLE_STATS
397 LPROC_OSD_THANDLE_STARTING,
398 LPROC_OSD_THANDLE_OPEN,
399 LPROC_OSD_THANDLE_CLOSING,
406 * Storage representation for fids.
408 * Variable size, first byte contains the length of the whole record.
410 struct osd_fid_pack {
411 unsigned char fp_len;
412 char fp_area[sizeof(struct lu_fid)];
415 struct osd_it_ea_dirent {
416 struct lu_fid oied_fid;
419 unsigned short oied_namelen;
420 unsigned int oied_type;
422 } __attribute__((packed));
425 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
426 * than lu_dirent struct. osd readdir reads less number of dirent than
427 * required for mdd dir page. so buffer size need to be increased so that
428 * there would be one ext3 readdir for every mdd readdir page.
431 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
434 * This is iterator's in-memory data structure in interoperability
435 * mode (i.e. iterator over ldiskfs style directory)
438 struct osd_object *oie_obj;
439 /** used in ldiskfs iterator, to stored file pointer */
440 struct file oie_file;
441 /** how many entries have been read-cached from storage */
443 /** current entry is being iterated by caller */
445 /** current processing entry */
446 struct osd_it_ea_dirent *oie_dirent;
447 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
449 struct dentry oie_dentry;
453 * Iterator's in-memory data structure for IAM mode.
456 struct osd_object *oi_obj;
457 struct iam_path_descr *oi_ipd;
458 struct iam_iterator oi_it;
461 struct osd_quota_leaf {
462 struct list_head oql_link;
467 * Iterator's in-memory data structure for quota file.
469 struct osd_it_quota {
470 struct osd_object *oiq_obj;
471 /** tree blocks path to where the entry is stored */
472 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
473 /** on-disk offset for current key where quota record can be found */
475 /** identifier for current quota record */
477 /** the record index in the leaf/index block */
478 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
479 /** list of already processed leaf blocks */
480 struct list_head oiq_list;
483 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
486 wait_queue_head_t dr_wait;
487 atomic_t dr_numreqs; /* number of reqs being processed */
492 unsigned int dr_ignore_quota:1;
493 unsigned int dr_elapsed_valid:1; /* we really did count time */
494 unsigned int dr_rw:1;
495 struct lu_buf dr_pg_buf;
496 struct page **dr_pages;
497 struct lu_buf dr_bl_buf;
499 unsigned long dr_start_time;
500 unsigned long dr_elapsed; /* how long io took */
501 struct osd_device *dr_dev;
502 unsigned int dr_init_at; /* the line iobuf was initialized */
505 struct osd_thread_info {
506 const struct lu_env *oti_env;
508 * used for index operations.
510 struct dentry oti_obj_dentry;
511 struct dentry oti_child_dentry;
513 /** dentry for Iterator context. */
514 struct dentry oti_it_dentry;
517 /* fake struct file for osd_object_sync */
518 struct file oti_file;
520 struct kstatfs oti_ksfs;
523 struct htree_lock *oti_hlock;
525 struct lu_fid oti_fid;
526 struct lu_fid oti_fid2;
527 struct lu_fid oti_fid3;
528 struct osd_inode_id oti_id;
529 struct osd_inode_id oti_id2;
530 struct osd_inode_id oti_id3;
531 struct ost_id oti_ostid;
534 * XXX temporary: for ->i_op calls.
536 struct timespec oti_time;
538 /** osd_device reference, initialized in osd_trans_start() and
539 used in osd_trans_stop() */
540 struct osd_device *oti_dev;
543 * following ipd and it structures are used for osd_index_iam_lookup()
544 * these are defined separately as we might do index operation
545 * in open iterator session.
548 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
550 unsigned int oti_it_ea_buf_used:1;
552 /* IAM iterator for index operation. */
553 struct iam_iterator oti_idx_it;
555 /** union to guarantee that ->oti_ipd[] has proper alignment. */
558 char oti_it_ipd[DX_IPD_MAX_SIZE];
559 long long oti_alignment_lieutenant;
563 char oti_idx_ipd[DX_IPD_MAX_SIZE];
564 long long oti_alignment_lieutenant_colonel;
567 struct osd_idmap_cache oti_cache;
572 /** used in osd_fid_set() to put xattr */
573 struct lu_buf oti_buf;
574 struct lu_buf oti_big_buf;
575 /** used in osd_ea_fid_set() to set fid into common ea */
577 struct lustre_mdt_attrs oti_mdt_attrs;
578 /* old LMA for compatibility */
579 char oti_mdt_attrs_old[LMA_OLD_SIZE];
580 struct filter_fid_old oti_ff;
581 struct filter_fid oti_ff_new;
584 struct osd_iobuf oti_iobuf;
585 /* used to access objects in /O */
586 struct inode *oti_inode;
587 #define OSD_FID_REC_SZ 32
588 char oti_ldp[OSD_FID_REC_SZ];
589 char oti_ldp2[OSD_FID_REC_SZ];
591 /* used by quota code */
593 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
594 struct fs_disk_quota oti_fdq;
596 struct if_dqblk oti_dqblk;
598 struct if_dqinfo oti_dqinfo;
600 struct lquota_id_info oti_qi;
601 struct lquota_trans oti_quota_trans;
602 union lquota_rec oti_quota_rec;
604 struct lu_seq_range oti_seq_range;
606 /* Tracking for transaction credits, to allow debugging and optimizing
607 * cases where a large number of credits are being allocated for
608 * single transaction. */
609 unsigned int oti_credits_before;
610 unsigned short oti_declare_ops[OSD_OT_MAX];
611 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
612 unsigned short oti_declare_ops_used[OSD_OT_MAX];
615 extern int ldiskfs_pdo;
617 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
618 const char *name, void *buf, int len)
623 dentry->d_inode = inode;
624 dentry->d_sb = inode->i_sb;
625 return inode->i_op->getxattr(dentry, name, buf, len);
628 static inline int __osd_xattr_set(struct osd_thread_info *info,
629 struct inode *inode, const char *name,
630 const void *buf, int buflen, int fl)
632 struct dentry *dentry = &info->oti_child_dentry;
634 ll_vfs_dq_init(inode);
635 dentry->d_inode = inode;
636 dentry->d_sb = inode->i_sb;
637 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
640 #ifdef CONFIG_PROC_FS
642 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
643 extern struct lprocfs_vars lprocfs_osd_module_vars[];
644 int osd_procfs_init(struct osd_device *osd, const char *name);
645 int osd_procfs_fini(struct osd_device *osd);
646 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
647 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
648 int osd_register_proc_index_in_idif(struct osd_device *osd);
652 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
653 struct obd_statfs *sfs);
654 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
655 struct osd_inode_id *id);
656 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
657 const struct lu_fid *fid, __u32 compat, __u32 incompat);
658 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
659 struct dentry *dentry, struct lustre_mdt_attrs *lma);
660 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
661 struct osd_inode_id *id, const struct lu_fid *fid);
662 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
663 struct dentry *dentry, struct lu_fid *fid);
665 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
666 void osd_obj_map_fini(struct osd_device *dev);
667 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
668 const struct lu_fid *fid, struct osd_inode_id *id);
669 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
670 const struct lu_fid *fid, const struct osd_inode_id *id,
672 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
673 const struct lu_fid *fid, handle_t *th);
674 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
675 const struct lu_fid *fid, const struct osd_inode_id *id,
677 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
678 struct inode *src_parent, struct dentry *src_child,
679 const struct lu_fid *fid);
680 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
681 const struct lu_fid *fid, struct osd_inode_id *id);
682 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
683 const struct lu_fid *fid, const struct osd_inode_id *id,
685 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
686 const struct lu_fid *fid, const struct osd_inode_id *id,
689 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
690 int osd_scrub_file_store(struct osd_scrub *scrub);
691 char *osd_lf_fid2name(const struct lu_fid *fid);
692 int osd_scrub_start(struct osd_device *dev, __u32 flags);
693 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
694 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
695 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
697 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
698 struct osd_inode_id *id);
699 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
701 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
702 u64 seq, struct lu_seq_range *range);
704 int osd_delete_from_remote_parent(const struct lu_env *env,
705 struct osd_device *osd,
706 struct osd_object *obj,
707 struct osd_thandle *oh);
708 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
709 struct osd_object *obj, struct osd_thandle *oh);
710 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
711 struct osd_device *osd,
712 const struct lu_fid *fid,
713 struct osd_inode_id *id);
715 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
717 /* osd_quota_fmt.c */
718 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
719 int type, uint blk, int depth, uint index,
720 struct osd_it_quota *it);
721 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
722 int type, uint blk, uint index,
723 struct osd_it_quota *it);
724 loff_t find_tree_dqentry(const struct lu_env *env,
725 struct osd_object *obj, int type,
726 qid_t dqid, uint blk, int depth,
727 struct osd_it_quota *it);
729 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
730 struct lquota_id_info *qi, struct osd_object *obj,
731 bool enforce, int *flags);
732 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
733 long long space, struct osd_thandle *oh,
734 struct osd_object *obj, bool is_blk, int *flags,
736 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
737 const struct dt_rec *rec,
738 union lquota_rec *quota_rec);
739 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
740 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
742 #ifndef HAVE_I_UID_READ
743 static inline uid_t i_uid_read(const struct inode *inode)
748 static inline gid_t i_gid_read(const struct inode *inode)
753 static inline void i_uid_write(struct inode *inode, uid_t uid)
758 static inline void i_gid_write(struct inode *inode, gid_t gid)
764 #ifdef LDISKFS_HT_MISC
765 # define osd_journal_start_sb(sb, type, nblock) \
766 ldiskfs_journal_start_sb(sb, type, nblock)
767 # define osd_ldiskfs_append(handle, inode, nblock, err) \
768 ldiskfs_append(handle, inode, nblock)
769 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
770 __ldiskfs_find_entry(dir, name, de, inlined, lock)
771 # define osd_journal_start(inode, type, nblocks) \
772 ldiskfs_journal_start(inode, type, nblocks)
773 # define osd_transaction_size(dev) \
774 (osd_journal(dev)->j_max_transaction_buffers / 2)
776 # define LDISKFS_HT_MISC 0
777 # define osd_journal_start_sb(sb, type, nblock) \
778 ldiskfs_journal_start_sb(sb, nblock)
779 # define osd_ldiskfs_append(handle, inode, nblock, err) \
780 ldiskfs_append(handle, inode, nblock, err)
781 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
782 __ldiskfs_find_entry(dir, name, de, lock)
783 # define osd_journal_start(inode, type, nblocks) \
784 ldiskfs_journal_start(inode, nblocks)
785 # define osd_transaction_size(dev) \
786 (osd_journal(dev)->j_max_transaction_buffers)
790 * Invariants, assertions.
794 * XXX: do not enable this, until invariant checking code is made thread safe
795 * in the face of pdirops locking.
797 #define OSD_INVARIANT_CHECKS (0)
799 #if OSD_INVARIANT_CHECKS
800 static inline int osd_invariant(const struct osd_object *obj)
804 ergo(obj->oo_inode != NULL,
805 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
806 atomic_read(&obj->oo_inode->i_count) > 0) &&
807 ergo(obj->oo_dir != NULL &&
808 obj->oo_dir->od_conationer.ic_object != NULL,
809 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
812 #define osd_invariant(obj) (1)
815 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
817 extern const struct dt_index_operations osd_otable_ops;
819 static inline int osd_oi_fid2idx(struct osd_device *dev,
820 const struct lu_fid *fid)
822 return fid->f_seq & (dev->od_oi_count - 1);
825 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
826 const struct lu_fid *fid)
828 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
829 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
830 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
831 DFID"\n", PFID(fid));
832 /* It can work even od_oi_count equals to 1 although it's unexpected,
833 * the only reason we set it to 1 is for performance measurement */
834 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
837 extern const struct lu_device_operations osd_lu_ops;
839 static inline int lu_device_is_osd(const struct lu_device *d)
841 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
844 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
846 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
847 return container_of0(d, struct osd_device, od_dt_dev);
850 static inline struct osd_device *osd_dev(const struct lu_device *d)
852 LASSERT(lu_device_is_osd(d));
853 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
856 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
858 return osd_dev(o->oo_dt.do_lu.lo_dev);
861 static inline struct super_block *osd_sb(const struct osd_device *dev)
863 return dev->od_mnt->mnt_sb;
866 static inline int osd_object_is_root(const struct osd_object *obj)
868 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
871 static inline struct osd_object *osd_obj(const struct lu_object *o)
873 LASSERT(lu_device_is_osd(o->lo_dev));
874 return container_of0(o, struct osd_object, oo_dt.do_lu);
877 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
879 return osd_obj(&d->do_lu);
882 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
884 return &osd->od_dt_dev.dd_lu_dev;
887 static inline journal_t *osd_journal(const struct osd_device *dev)
889 return LDISKFS_SB(osd_sb(dev))->s_journal;
892 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
894 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
897 static inline char *osd_name(struct osd_device *osd)
899 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
902 extern const struct dt_body_operations osd_body_ops;
903 extern struct lu_context_key osd_key;
905 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
907 return lu_context_key_get(&env->le_ctx, &osd_key);
910 extern const struct dt_body_operations osd_body_ops_new;
916 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
917 const struct iam_container *bag)
919 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
920 osd_oti_get(env)->oti_it_ipd);
924 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
925 const struct iam_container *bag)
927 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
928 osd_oti_get(env)->oti_idx_ipd);
931 static inline void osd_ipd_put(const struct lu_env *env,
932 const struct iam_container *bag,
933 struct iam_path_descr *ipd)
935 bag->ic_descr->id_ops->id_ipd_free(ipd);
938 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
939 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
940 int write_NUL, loff_t *offs, handle_t *handle);
943 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
945 const char *name, const int namelen)
947 struct osd_thread_info *info = osd_oti_get(env);
948 struct dentry *child_dentry = &info->oti_child_dentry;
949 struct dentry *obj_dentry = &info->oti_obj_dentry;
951 obj_dentry->d_inode = inode;
952 obj_dentry->d_sb = inode->i_sb;
953 obj_dentry->d_name.hash = 0;
955 child_dentry->d_name.hash = 0;
956 child_dentry->d_parent = obj_dentry;
957 child_dentry->d_name.name = name;
958 child_dentry->d_name.len = namelen;
962 extern int osd_trans_declare_op2rb[];
963 extern int ldiskfs_track_declares_assert;
964 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
966 static inline void osd_trans_declare_op(const struct lu_env *env,
967 struct osd_thandle *oh,
968 unsigned int op, int credits)
970 struct osd_thread_info *oti = osd_oti_get(env);
972 LASSERT(oh->ot_handle == NULL);
973 if (unlikely(op >= OSD_OT_MAX)) {
974 if (unlikely(ldiskfs_track_declares_assert)) {
975 LASSERT(op < OSD_OT_MAX);
977 CWARN("%s: Invalid operation index %d\n",
978 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
979 libcfs_debug_dumpstack(NULL);
982 oti->oti_declare_ops[op]++;
983 oti->oti_declare_ops_cred[op] += credits;
985 oh->ot_credits += credits;
988 static inline void osd_trans_exec_op(const struct lu_env *env,
989 struct thandle *th, unsigned int op)
991 struct osd_thread_info *oti = osd_oti_get(env);
992 struct osd_thandle *oh = container_of(th, struct osd_thandle,
994 unsigned int rb, left;
996 LASSERT(oh->ot_handle != NULL);
997 if (unlikely(op >= OSD_OT_MAX)) {
998 if (unlikely(ldiskfs_track_declares_assert))
999 LASSERT(op < OSD_OT_MAX);
1001 CWARN("%s: Invalid operation index %d\n",
1002 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1003 libcfs_debug_dumpstack(NULL);
1008 /* find rollback (or reverse) operation for the given one
1009 * such an operation doesn't require additional credits
1010 * as the same set of blocks are modified */
1011 rb = osd_trans_declare_op2rb[op];
1013 /* check whether credits for this operation were reserved at all */
1014 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1015 oti->oti_declare_ops_cred[rb] == 0)) {
1016 /* the API is not perfect yet: CREATE does REF_ADD internally
1017 * while DESTROY does not. To rollback CREATE the callers
1018 * needs to call REF_DEL+DESTROY which is hard to detect using
1019 * a simple table of rollback operations */
1020 if (op == OSD_OT_REF_DEL &&
1021 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1023 if (op == OSD_OT_REF_ADD &&
1024 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1026 osd_trans_dump_creds(env, th);
1027 CERROR("%s: op = %d, rb = %d\n",
1028 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1029 if (unlikely(ldiskfs_track_declares_assert))
1034 /* remember how many credits we have unused before the operation */
1035 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1036 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1037 if (unlikely(oti->oti_credits_before < left)) {
1038 osd_trans_dump_creds(env, th);
1039 CERROR("%s: op = %d, rb = %d\n",
1040 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1041 /* on a very small fs (testing?) it's possible that
1042 * the transaction can't fit 1/4 of journal, so we
1043 * just request less credits (see osd_trans_start()).
1044 * ignore the same case here */
1045 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1046 if (unlikely(oh->ot_credits < rb)) {
1047 if (unlikely(ldiskfs_track_declares_assert))
1053 static inline void osd_trans_exec_check(const struct lu_env *env,
1057 struct osd_thread_info *oti = osd_oti_get(env);
1058 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1060 int used, over, quota;
1062 /* how many credits have been used by the operation */
1063 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1065 if (unlikely(used < 0)) {
1066 /* if some block was allocated and released in the same
1067 * transaction, then it won't be a part of the transaction
1068 * and delta can be negative */
1073 /* rollback operations (e.g. when we destroy just created
1074 * object) should not consume any credits. there is no point
1075 * to confuse the checks below */
1079 oti->oti_declare_ops_used[op] += used;
1080 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1083 /* we account quota for a whole transaction and any operation can
1084 * consume corresponding credits */
1085 over = oti->oti_declare_ops_used[op] -
1086 oti->oti_declare_ops_cred[op];
1087 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1088 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1089 if (over <= quota) {
1090 /* probably that credits were consumed by
1091 * quota indirectly (in the depths of ldiskfs) */
1092 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1093 oti->oti_declare_ops_used[op] -= over;
1095 CWARN("op %d: used %u, used now %u, reserved %u\n",
1096 op, oti->oti_declare_ops_used[op], used,
1097 oti->oti_declare_ops_cred[op]);
1098 osd_trans_dump_creds(env, th);
1099 if (unlikely(ldiskfs_track_declares_assert))
1105 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1108 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1109 struct lu_fid *befider)
1111 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1112 memcpy(pack->fp_area, befider, sizeof(*befider));
1113 pack->fp_len = sizeof(*befider) + 1;
1117 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1122 switch (pack->fp_len) {
1123 case sizeof *fid + 1:
1124 memcpy(fid, pack->fp_area, sizeof *fid);
1125 fid_be_to_cpu(fid, fid);
1128 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1135 * Quota/Accounting handling
1137 extern const struct dt_index_operations osd_acct_index_ops;
1138 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1139 const struct lu_fid *fid, struct osd_inode_id *id);
1141 /* copy from fs/ext4/dir.c */
1142 static inline int is_32bit_api(void)
1144 #ifdef CONFIG_COMPAT
1145 return is_compat_task();
1147 return (BITS_PER_LONG == 32);
1151 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1153 if ((filp->f_mode & FMODE_32BITHASH) ||
1154 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1155 return LDISKFS_HTREE_EOF_32BIT;
1157 return LDISKFS_HTREE_EOF_64BIT;
1160 static inline int fid_is_internal(const struct lu_fid *fid)
1162 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1165 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1167 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1170 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1171 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1173 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1175 #endif /* _OSD_INTERNAL_H */