4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_internal.h
34 * Shared definitions and declarations for osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
39 #ifndef _OSD_INTERNAL_H
40 #define _OSD_INTERNAL_H
44 #include <linux/mutex.h>
45 /* struct rw_semaphore */
46 #include <linux/rwsem.h>
48 #include <linux/dcache.h>
50 #include <linux/dirent.h>
51 #include <linux/statfs.h>
52 #include <ldiskfs/ldiskfs.h>
53 #include <ldiskfs/ldiskfs_jbd2.h>
57 /* class_register_type(), class_unregister_type(), class_get_type() */
58 #include <obd_class.h>
59 #include <lustre_disk.h>
60 #include <dt_object.h>
61 #include <lustre_quota.h>
65 #include "osd_scrub.h"
66 #include "osd_quota_fmt.h"
69 extern struct kmem_cache *dynlock_cachep;
71 #define OSD_COUNTERS (0)
73 /* ldiskfs special inode::i_state_flags need to be accessed with
74 * ldiskfs_{set,clear,test}_inode_state() only */
76 /* OI scrub should skip this inode. */
77 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
78 #define LDISKFS_STATE_LUSTRE_DESTROY 30
80 /** Enable thandle usage statistics */
81 #define OSD_THANDLE_STATS (0)
83 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
85 #define OBJECTS "OBJECTS"
86 #define ADMIN_USR "admin_quotafile_v2.usr"
87 #define ADMIN_GRP "admin_quotafile_v2.grp"
89 /* Statfs space reservation for fragmentation and local objects */
90 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
91 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
93 struct osd_directory {
94 struct iam_container od_container;
95 struct iam_descr od_descr;
99 * Object Index (oi) instance.
103 * underlying index object, where fid->id mapping in stored.
105 struct inode *oi_inode;
106 struct osd_directory oi_dir;
109 extern const int osd_dto_credits_noquota[];
112 struct dt_object oo_dt;
114 * Inode for file system object represented by this osd_object. This
115 * inode is pinned for the whole duration of lu_object life.
117 * Not modified concurrently (either setup early during object
118 * creation, or assigned by osd_object_create() under write lock).
120 struct inode *oo_inode;
122 * to protect index ops.
124 struct htree_lock_head *oo_hl_head;
125 struct rw_semaphore oo_ext_idx_sem;
126 struct rw_semaphore oo_sem;
127 struct osd_directory *oo_dir;
128 /** protects inode attributes. */
131 __u32 oo_destroyed:1;
133 /* the i_flags in LMA */
136 * Following two members are used to indicate the presence of dot and
137 * dotdot in the given directory. This is required for interop mode
140 int oo_compat_dot_created;
141 int oo_compat_dotdot_created;
143 const struct lu_env *oo_owner;
144 #ifdef CONFIG_LOCKDEP
145 struct lockdep_map oo_dep_map;
148 struct list_head oo_xattr_list;
152 /* protects on-fly initialization */
153 int oos_subdir_count; /* subdir count for each seq */
154 struct dentry *oos_root; /* O/<seq> */
155 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
156 u64 oos_seq; /* seq number */
157 struct list_head oos_seq_list; /* list to seq_list */
161 struct dentry *om_root; /* dentry for /O */
162 rwlock_t om_seq_list_lock; /* lock for seq_list */
163 struct list_head om_seq_list; /* list head for seq */
165 struct mutex om_dir_init_mutex;
169 struct dentry *om_root; /* AGENT/<index> */
170 u64 om_index; /* mdt index */
171 struct list_head om_list; /* list to omm_list */
174 struct osd_mdobj_map {
175 struct dentry *omm_remote_parent;
177 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
178 handle_t *handle, struct dentry *child,
179 struct inode *inode, struct htree_lock *hlock);
181 #define OSD_OTABLE_IT_CACHE_SIZE 64
182 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
184 struct osd_inconsistent_item {
185 /* link into osd_scrub::os_inconsistent_items,
186 * protected by osd_scrub::os_lock. */
187 struct list_head oii_list;
189 /* The right FID <=> ino#/gen mapping. */
190 struct osd_idmap_cache oii_cache;
192 unsigned int oii_insert:1; /* insert or update mapping. */
195 struct osd_otable_cache {
196 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
198 /* Index for next cache slot to be filled. */
199 int ooc_producer_idx;
201 /* Index for next cache slot to be returned by it::next(). */
202 int ooc_consumer_idx;
204 /* How many items in ooc_cache. */
205 int ooc_cached_items;
207 /* Position for up layer LFSCK iteration pre-loading. */
208 __u32 ooc_pos_preload;
211 struct osd_otable_it {
212 struct osd_device *ooi_dev;
213 struct osd_otable_cache ooi_cache;
215 /* The following bits can be updated/checked w/o lock protection.
216 * If more bits will be introduced in the future and need lock to
217 * protect, please add comment. */
218 unsigned long ooi_used_outside:1, /* Some user out of OSD
219 * uses the iteration. */
220 ooi_all_cached:1, /* No more entries can be
221 * filled into cache. */
222 ooi_user_ready:1, /* The user out of OSD is
223 * ready to iterate. */
224 ooi_waiting:1; /* it::next is waiting. */
227 struct osd_obj_orphan {
228 struct list_head oor_list;
229 struct lu_env *oor_env; /* to identify "own" records */
238 struct dt_device od_dt_dev;
239 /* information about underlying file system */
240 struct vfsmount *od_mnt;
242 struct osd_oi **od_oi_table;
243 /* total number of OI containers */
248 unsigned int od_fl_capa:1,
256 __u32 od_dirent_journal;
258 struct proc_dir_entry *od_proc_entry;
259 struct lprocfs_stats *od_stats;
261 spinlock_t od_osfs_lock;
264 struct lu_site od_site;
266 struct osd_obj_map *od_ost_map;
267 struct osd_mdobj_map *od_mdt_map;
269 unsigned long long od_readcache_max_filesize;
271 int od_writethrough_cache;
273 struct brw_stats od_brw_stats;
274 atomic_t od_r_in_flight;
275 atomic_t od_w_in_flight;
277 struct mutex od_otable_mutex;
278 struct osd_otable_it *od_otable_it;
279 struct osd_scrub od_scrub;
280 struct list_head od_ios_list;
282 /* service name associated with the osd device */
283 char od_svname[MAX_OBD_NAME];
284 char od_mntdev[MAX_OBD_NAME];
286 /* quota slave instance */
287 struct qsd_instance *od_quota_slave;
289 /* osd seq instance */
290 struct lu_client_seq *od_cl_seq;
291 /* If the ratio of "the total OI mappings count" vs
292 * "the bad OI mappings count" is lower than the
293 * osd_device::od_full_scrub_ratio, then trigger
294 * OI scrub to scan the whole the device. */
295 __u64 od_full_scrub_ratio;
296 /* If the speed of found bad OI mappings (per minute)
297 * exceeds the osd_device::od_full_scrub_threshold_rate,
298 * then trigger OI scrub to scan the whole device. */
299 __u64 od_full_scrub_threshold_rate;
301 /* a list of orphaned agent inodes, protected with od_osfs_lock */
302 struct list_head od_orphan_list;
305 enum osd_full_scrub_ratio {
306 /* Trigger OI scrub to scan the whole device directly. */
309 /* Because the bad OI mappings count cannot be larger than
310 * the total OI mappints count, then setting OFSR_NEVER means
311 * that the whole device scanning cannot be triggered by auto
312 * detected bad OI mappings during the RPC services. */
314 OFSR_DEFAULT = 10000,
317 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
319 /* There are at most 10 uid/gids are affected in a transaction, and
320 * that's rename case:
321 * - 2 for source parent uid & gid;
322 * - 2 for source child uid & gid ('..' entry update when child is directory);
323 * - 2 for target parent uid & gid;
324 * - 2 for target child uid & gid (if the target child exists);
325 * - 2 for root uid & gid (last_rcvd, llog, etc);
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 int ot_credits;
354 unsigned short ot_id_cnt;
355 __u8 ot_id_types[OSD_MAX_UGID_CNT];
356 unsigned 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 */
363 /** time when this thanle was 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_SIZE + PAGE_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_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;
545 * following ipd and it structures are used for osd_index_iam_lookup()
546 * these are defined separately as we might do index operation
547 * in open iterator session.
550 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
552 unsigned int oti_it_ea_buf_used:1;
554 /* IAM iterator for index operation. */
555 struct iam_iterator oti_idx_it;
557 /** union to guarantee that ->oti_ipd[] has proper alignment. */
560 char oti_it_ipd[DX_IPD_MAX_SIZE];
561 long long oti_alignment_lieutenant;
565 char oti_idx_ipd[DX_IPD_MAX_SIZE];
566 long long oti_alignment_lieutenant_colonel;
569 struct osd_idmap_cache oti_cache;
571 /* dedicated OI cache for insert (which needs inum) */
572 struct osd_idmap_cache *oti_ins_cache;
573 int oti_ins_cache_size;
574 int oti_ins_cache_used;
579 /** used in osd_fid_set() to put xattr */
580 struct lu_buf oti_buf;
581 struct lu_buf oti_big_buf;
582 /** used in osd_ea_fid_set() to set fid into common ea */
584 struct lustre_ost_attrs oti_ost_attrs;
585 struct filter_fid_old oti_ff;
586 struct filter_fid oti_ff_new;
589 struct osd_iobuf oti_iobuf;
590 /* used to access objects in /O */
591 struct inode *oti_inode;
592 #define OSD_FID_REC_SZ 32
593 char oti_ldp[OSD_FID_REC_SZ];
594 char oti_ldp2[OSD_FID_REC_SZ];
596 /* used by quota code */
598 #if defined(HAVE_DQUOT_QC_DQBLK)
599 struct qc_dqblk oti_qdq;
600 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
601 struct fs_disk_quota oti_fdq;
603 struct if_dqblk oti_dqblk;
605 struct if_dqinfo oti_dqinfo;
607 struct lquota_id_info oti_qi;
608 struct lquota_trans oti_quota_trans;
609 union lquota_rec oti_quota_rec;
611 struct lu_seq_range oti_seq_range;
613 /* Tracking for transaction credits, to allow debugging and optimizing
614 * cases where a large number of credits are being allocated for
615 * single transaction. */
616 unsigned int oti_credits_before;
617 unsigned int oti_declare_ops[OSD_OT_MAX];
618 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
619 unsigned int oti_declare_ops_used[OSD_OT_MAX];
622 extern int ldiskfs_pdo;
624 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
625 const char *name, void *buf, int len)
630 dentry->d_inode = inode;
631 dentry->d_sb = inode->i_sb;
632 return inode->i_op->getxattr(dentry, name, buf, len);
635 static inline int __osd_xattr_set(struct osd_thread_info *info,
636 struct inode *inode, const char *name,
637 const void *buf, int buflen, int fl)
639 struct dentry *dentry = &info->oti_child_dentry;
641 ll_vfs_dq_init(inode);
642 dentry->d_inode = inode;
643 dentry->d_sb = inode->i_sb;
644 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
647 #ifdef CONFIG_PROC_FS
649 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
650 extern struct lprocfs_vars lprocfs_osd_module_vars[];
651 int osd_procfs_init(struct osd_device *osd, const char *name);
652 int osd_procfs_fini(struct osd_device *osd);
653 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
654 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
655 int osd_register_proc_index_in_idif(struct osd_device *osd);
659 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
660 struct obd_statfs *sfs);
661 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
662 struct osd_inode_id *id);
663 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
664 const struct lu_fid *fid, __u32 compat, __u32 incompat);
665 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
666 struct dentry *dentry, struct lustre_ost_attrs *loa);
667 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
668 struct osd_inode_id *id, const struct lu_fid *fid);
669 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
670 struct dentry *dentry, struct lu_fid *fid);
672 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
673 void osd_obj_map_fini(struct osd_device *dev);
674 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
675 const struct lu_fid *fid, struct osd_inode_id *id);
676 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
677 const struct lu_fid *fid, const struct osd_inode_id *id,
679 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
680 const struct lu_fid *fid, handle_t *th);
681 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
682 const struct lu_fid *fid, const struct osd_inode_id *id,
684 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
685 struct inode *src_parent, struct dentry *src_child,
686 const struct lu_fid *fid);
687 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
688 const struct lu_fid *fid, struct osd_inode_id *id);
689 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
690 const struct lu_fid *fid, const struct osd_inode_id *id,
692 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
693 const struct lu_fid *fid, const struct osd_inode_id *id,
696 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
697 int osd_scrub_file_store(struct osd_scrub *scrub);
698 char *osd_lf_fid2name(const struct lu_fid *fid);
699 int osd_scrub_start(struct osd_device *dev, __u32 flags);
700 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
701 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
702 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
704 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
705 struct osd_inode_id *id);
706 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
708 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
709 u64 seq, struct lu_seq_range *range);
711 int osd_delete_from_remote_parent(const struct lu_env *env,
712 struct osd_device *osd,
713 struct osd_object *obj,
714 struct osd_thandle *oh);
715 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
716 struct osd_object *obj, struct osd_thandle *oh);
717 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
718 struct osd_device *osd,
719 const struct lu_fid *fid,
720 struct osd_inode_id *id);
722 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
724 /* osd_quota_fmt.c */
725 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
726 int type, uint blk, int depth, uint index,
727 struct osd_it_quota *it);
728 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
729 int type, uint blk, uint index,
730 struct osd_it_quota *it);
731 loff_t find_tree_dqentry(const struct lu_env *env,
732 struct osd_object *obj, int type,
733 qid_t dqid, uint blk, int depth,
734 struct osd_it_quota *it);
736 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
737 struct lquota_id_info *qi, struct osd_object *obj,
738 bool enforce, int *flags);
739 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
740 long long space, struct osd_thandle *oh,
741 struct osd_object *obj, bool is_blk, int *flags,
743 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
744 const struct dt_rec *rec,
745 union lquota_rec *quota_rec);
746 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
747 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
749 #ifndef HAVE_I_UID_READ
750 static inline uid_t i_uid_read(const struct inode *inode)
755 static inline gid_t i_gid_read(const struct inode *inode)
760 static inline void i_uid_write(struct inode *inode, uid_t uid)
765 static inline void i_gid_write(struct inode *inode, gid_t gid)
771 #ifdef HAVE_LDISKFS_INFO_JINODE
772 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
773 #else /* HAVE_LDISKFS_INFO_JINODE */
774 # define osd_attach_jinode(inode) 0
775 #endif /* HAVE_LDISKFS_INFO_JINODE */
777 #ifdef LDISKFS_HT_MISC
778 # define osd_journal_start_sb(sb, type, nblock) \
779 ldiskfs_journal_start_sb(sb, type, nblock)
780 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
782 ldiskfs_lblk_t *nblock)
786 rc = osd_attach_jinode(inode);
789 return ldiskfs_append(handle, inode, nblock);
791 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
792 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
794 # define osd_journal_start(inode, type, nblocks) \
795 ldiskfs_journal_start(inode, type, nblocks)
796 # define osd_transaction_size(dev) \
797 (osd_journal(dev)->j_max_transaction_buffers / 2)
799 # define LDISKFS_HT_MISC 0
800 # define osd_journal_start_sb(sb, type, nblock) \
801 ldiskfs_journal_start_sb(sb, nblock)
803 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
805 ldiskfs_lblk_t *nblock)
807 struct buffer_head *bh;
810 bh = ldiskfs_append(handle, inode, nblock, &err);
817 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
818 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
820 # define osd_journal_start(inode, type, nblocks) \
821 ldiskfs_journal_start(inode, nblocks)
822 # define osd_transaction_size(dev) \
823 (osd_journal(dev)->j_max_transaction_buffers)
827 * Invariants, assertions.
831 * XXX: do not enable this, until invariant checking code is made thread safe
832 * in the face of pdirops locking.
834 #define OSD_INVARIANT_CHECKS (0)
836 #if OSD_INVARIANT_CHECKS
837 static inline int osd_invariant(const struct osd_object *obj)
841 ergo(obj->oo_inode != NULL,
842 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
843 atomic_read(&obj->oo_inode->i_count) > 0) &&
844 ergo(obj->oo_dir != NULL &&
845 obj->oo_dir->od_conationer.ic_object != NULL,
846 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
849 #define osd_invariant(obj) (1)
852 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
854 extern const struct dt_index_operations osd_otable_ops;
856 static inline int osd_oi_fid2idx(struct osd_device *dev,
857 const struct lu_fid *fid)
859 return fid->f_seq & (dev->od_oi_count - 1);
862 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
863 const struct lu_fid *fid)
865 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
866 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
867 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
868 DFID"\n", PFID(fid));
869 /* It can work even od_oi_count equals to 1 although it's unexpected,
870 * the only reason we set it to 1 is for performance measurement */
871 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
874 extern const struct lu_device_operations osd_lu_ops;
876 static inline int lu_device_is_osd(const struct lu_device *d)
878 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
881 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
883 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
884 return container_of0(d, struct osd_device, od_dt_dev);
887 static inline struct osd_device *osd_dev(const struct lu_device *d)
889 LASSERT(lu_device_is_osd(d));
890 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
893 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
895 return osd_dev(o->oo_dt.do_lu.lo_dev);
898 static inline struct super_block *osd_sb(const struct osd_device *dev)
900 return dev->od_mnt->mnt_sb;
904 * Put the osd object once done with it.
906 * \param obj osd object that needs to be put
908 static inline void osd_object_put(const struct lu_env *env,
909 struct osd_object *obj)
911 dt_object_put(env, &obj->oo_dt);
914 static inline int osd_object_is_root(const struct osd_object *obj)
916 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
919 static inline struct osd_object *osd_obj(const struct lu_object *o)
921 LASSERT(lu_device_is_osd(o->lo_dev));
922 return container_of0(o, struct osd_object, oo_dt.do_lu);
925 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
927 return osd_obj(&d->do_lu);
930 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
932 return &osd->od_dt_dev.dd_lu_dev;
935 static inline journal_t *osd_journal(const struct osd_device *dev)
937 return LDISKFS_SB(osd_sb(dev))->s_journal;
940 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
942 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
945 static inline char *osd_name(struct osd_device *osd)
947 return osd->od_svname;
950 static inline bool osd_is_ea_inode(struct inode *inode)
952 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
955 extern const struct dt_body_operations osd_body_ops;
956 extern struct lu_context_key osd_key;
958 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
960 return lu_context_key_get(&env->le_ctx, &osd_key);
963 extern const struct dt_body_operations osd_body_ops_new;
969 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
970 const struct iam_container *bag)
972 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
973 osd_oti_get(env)->oti_it_ipd);
977 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
978 const struct iam_container *bag)
980 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
981 osd_oti_get(env)->oti_idx_ipd);
984 static inline void osd_ipd_put(const struct lu_env *env,
985 const struct iam_container *bag,
986 struct iam_path_descr *ipd)
988 bag->ic_descr->id_ops->id_ipd_free(ipd);
991 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
992 const loff_t size, const loff_t pos,
995 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
996 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
997 int write_NUL, loff_t *offs, handle_t *handle);
1000 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1001 struct inode *inode,
1002 const char *name, const int namelen)
1004 struct osd_thread_info *info = osd_oti_get(env);
1005 struct dentry *child_dentry = &info->oti_child_dentry;
1006 struct dentry *obj_dentry = &info->oti_obj_dentry;
1008 obj_dentry->d_inode = inode;
1009 obj_dentry->d_sb = inode->i_sb;
1010 obj_dentry->d_name.hash = 0;
1012 child_dentry->d_name.hash = 0;
1013 child_dentry->d_parent = obj_dentry;
1014 child_dentry->d_name.name = name;
1015 child_dentry->d_name.len = namelen;
1016 return child_dentry;
1019 extern int osd_trans_declare_op2rb[];
1020 extern int ldiskfs_track_declares_assert;
1021 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1023 static inline void osd_trans_declare_op(const struct lu_env *env,
1024 struct osd_thandle *oh,
1025 enum osd_op_type op, int credits)
1027 struct osd_thread_info *oti = osd_oti_get(env);
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);
1039 oti->oti_declare_ops[op]++;
1040 oti->oti_declare_ops_cred[op] += credits;
1042 oh->ot_credits += credits;
1045 static inline void osd_trans_exec_op(const struct lu_env *env,
1047 enum osd_op_type op)
1049 struct osd_thread_info *oti = osd_oti_get(env);
1050 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1052 unsigned int rb, left;
1054 LASSERT(oh->ot_handle != NULL);
1055 if (unlikely(op >= OSD_OT_MAX)) {
1056 if (unlikely(ldiskfs_track_declares_assert))
1057 LASSERT(op < OSD_OT_MAX);
1059 CWARN("%s: opcode %u: invalid value >= %u\n",
1060 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1062 libcfs_debug_dumpstack(NULL);
1067 /* find rollback (or reverse) operation for the given one
1068 * such an operation doesn't require additional credits
1069 * as the same set of blocks are modified */
1070 rb = osd_trans_declare_op2rb[op];
1072 /* check whether credits for this operation were reserved at all */
1073 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1074 oti->oti_declare_ops_cred[rb] == 0)) {
1075 /* the API is not perfect yet: CREATE does REF_ADD internally
1076 * while DESTROY does not. To rollback CREATE the callers
1077 * needs to call REF_DEL+DESTROY which is hard to detect using
1078 * a simple table of rollback operations */
1079 if (op == OSD_OT_REF_DEL &&
1080 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1082 if (op == OSD_OT_REF_ADD &&
1083 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1085 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1086 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1087 osd_trans_dump_creds(env, th);
1088 LASSERT(!ldiskfs_track_declares_assert);
1092 /* remember how many credits we have unused before the operation */
1093 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1094 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1095 if (unlikely(oti->oti_credits_before < left)) {
1096 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1097 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1098 oti->oti_credits_before, left, rb);
1099 osd_trans_dump_creds(env, th);
1100 /* on a very small fs (testing?) it's possible that
1101 * the transaction can't fit 1/4 of journal, so we
1102 * just request less credits (see osd_trans_start()).
1103 * ignore the same case here */
1104 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1105 if (unlikely(oh->ot_credits < rb))
1106 LASSERT(!ldiskfs_track_declares_assert);
1110 static inline void osd_trans_exec_check(const struct lu_env *env,
1112 enum osd_op_type op)
1114 struct osd_thread_info *oti = osd_oti_get(env);
1115 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1117 int used, over, quota;
1119 /* how many credits have been used by the operation */
1120 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1122 if (unlikely(used < 0)) {
1123 /* if some block was allocated and released in the same
1124 * transaction, then it won't be a part of the transaction
1125 * and delta can be negative */
1130 /* rollback operations (e.g. when we destroy just created
1131 * object) should not consume any credits. there is no point
1132 * to confuse the checks below */
1136 oti->oti_declare_ops_used[op] += used;
1137 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1140 /* we account quota for a whole transaction and any operation can
1141 * consume corresponding credits */
1142 over = oti->oti_declare_ops_used[op] -
1143 oti->oti_declare_ops_cred[op];
1144 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1145 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1146 if (over <= quota) {
1147 /* probably that credits were consumed by
1148 * quota indirectly (in the depths of ldiskfs) */
1149 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1150 oti->oti_declare_ops_used[op] -= over;
1152 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1153 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1154 oti->oti_declare_ops_used[op], used,
1155 oti->oti_declare_ops_cred[op]);
1156 osd_trans_dump_creds(env, th);
1157 if (unlikely(ldiskfs_track_declares_assert))
1163 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1166 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1167 struct lu_fid *befider)
1169 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1170 memcpy(pack->fp_area, befider, sizeof(*befider));
1171 pack->fp_len = sizeof(*befider) + 1;
1175 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1180 switch (pack->fp_len) {
1181 case sizeof *fid + 1:
1182 memcpy(fid, pack->fp_area, sizeof *fid);
1183 fid_be_to_cpu(fid, fid);
1186 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1193 * Quota/Accounting handling
1195 extern const struct dt_index_operations osd_acct_index_ops;
1196 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1197 const struct lu_fid *fid, struct osd_inode_id *id);
1199 /* copy from fs/ext4/dir.c */
1200 static inline int is_32bit_api(void)
1202 #ifdef CONFIG_COMPAT
1203 return in_compat_syscall();
1205 return (BITS_PER_LONG == 32);
1209 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1211 if ((filp->f_mode & FMODE_32BITHASH) ||
1212 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1213 return LDISKFS_HTREE_EOF_32BIT;
1215 return LDISKFS_HTREE_EOF_64BIT;
1218 static inline int fid_is_internal(const struct lu_fid *fid)
1220 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1223 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1225 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1229 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1230 * return code has been removed and integrated into the pointer in the
1233 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1234 struct inode *inode,
1235 ldiskfs_lblk_t block,
1239 struct buffer_head *bh;
1242 rc = osd_attach_jinode(inode);
1246 #ifdef HAVE_EXT4_BREAD_4ARGS
1247 bh = ldiskfs_bread(handle, inode, block, create);
1250 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1251 if (bh == NULL && rc != 0)
1257 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1258 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1260 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1263 #endif /* _OSD_INTERNAL_H */