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 * Following two members are used to indicate the presence of dot and
134 * dotdot in the given directory. This is required for interop mode
137 int oo_compat_dot_created;
138 int oo_compat_dotdot_created;
140 const struct lu_env *oo_owner;
141 #ifdef CONFIG_LOCKDEP
142 struct lockdep_map oo_dep_map;
147 /* protects on-fly initialization */
148 int oos_subdir_count; /* subdir count for each seq */
149 struct dentry *oos_root; /* O/<seq> */
150 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
151 u64 oos_seq; /* seq number */
152 struct list_head oos_seq_list; /* list to seq_list */
156 struct dentry *om_root; /* dentry for /O */
157 rwlock_t om_seq_list_lock; /* lock for seq_list */
158 struct list_head om_seq_list; /* list head for seq */
160 struct mutex om_dir_init_mutex;
164 struct dentry *om_root; /* AGENT/<index> */
165 u64 om_index; /* mdt index */
166 struct list_head om_list; /* list to omm_list */
169 struct osd_mdobj_map {
170 struct dentry *omm_remote_parent;
173 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
174 __ldiskfs_add_entry(handle, child, cinode, hlock)
176 #define OSD_OTABLE_IT_CACHE_SIZE 64
177 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
179 struct osd_inconsistent_item {
180 /* link into osd_scrub::os_inconsistent_items,
181 * protected by osd_scrub::os_lock. */
182 struct list_head oii_list;
184 /* The right FID <=> ino#/gen mapping. */
185 struct osd_idmap_cache oii_cache;
187 unsigned int oii_insert:1; /* insert or update mapping. */
190 struct osd_otable_cache {
191 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
193 /* Index for next cache slot to be filled. */
194 int ooc_producer_idx;
196 /* Index for next cache slot to be returned by it::next(). */
197 int ooc_consumer_idx;
199 /* How many items in ooc_cache. */
200 int ooc_cached_items;
202 /* Position for up layer LFSCK iteration pre-loading. */
203 __u32 ooc_pos_preload;
206 struct osd_otable_it {
207 struct osd_device *ooi_dev;
208 struct osd_otable_cache ooi_cache;
210 /* The following bits can be updated/checked w/o lock protection.
211 * If more bits will be introduced in the future and need lock to
212 * protect, please add comment. */
213 unsigned long ooi_used_outside:1, /* Some user out of OSD
214 * uses the iteration. */
215 ooi_all_cached:1, /* No more entries can be
216 * filled into cache. */
217 ooi_user_ready:1, /* The user out of OSD is
218 * ready to iterate. */
219 ooi_waiting:1; /* it::next is waiting. */
222 struct osd_obj_orphan {
223 struct list_head oor_list;
224 struct lu_env *oor_env; /* to identify "own" records */
233 struct dt_device od_dt_dev;
234 /* information about underlying file system */
235 struct vfsmount *od_mnt;
237 struct osd_oi **od_oi_table;
238 /* total number of OI containers */
243 unsigned int od_fl_capa:1,
251 __u32 od_dirent_journal;
253 struct proc_dir_entry *od_proc_entry;
254 struct lprocfs_stats *od_stats;
256 spinlock_t od_osfs_lock;
259 struct lu_site od_site;
261 struct osd_obj_map *od_ost_map;
262 struct osd_mdobj_map *od_mdt_map;
264 unsigned long long od_readcache_max_filesize;
266 int od_writethrough_cache;
268 struct brw_stats od_brw_stats;
269 atomic_t od_r_in_flight;
270 atomic_t od_w_in_flight;
272 struct mutex od_otable_mutex;
273 struct osd_otable_it *od_otable_it;
274 struct osd_scrub od_scrub;
275 struct list_head od_ios_list;
277 /* service name associated with the osd device */
278 char od_svname[MAX_OBD_NAME];
279 char od_mntdev[MAX_OBD_NAME];
281 /* quota slave instance */
282 struct qsd_instance *od_quota_slave;
284 /* osd seq instance */
285 struct lu_client_seq *od_cl_seq;
286 /* If the ratio of "the total OI mappings count" vs
287 * "the bad OI mappings count" is lower than the
288 * osd_device::od_full_scrub_ratio, then trigger
289 * OI scrub to scan the whole the device. */
290 __u64 od_full_scrub_ratio;
291 /* If the speed of found bad OI mappings (per minute)
292 * exceeds the osd_device::od_full_scrub_threshold_rate,
293 * then trigger OI scrub to scan the whole device. */
294 __u64 od_full_scrub_threshold_rate;
296 /* a list of orphaned agent inodes, protected with od_osfs_lock */
297 struct list_head od_orphan_list;
300 enum osd_full_scrub_ratio {
301 /* Trigger OI scrub to scan the whole device directly. */
304 /* Because the bad OI mappings count cannot be larger than
305 * the total OI mappints count, then setting OFSR_NEVER means
306 * that the whole device scanning cannot be triggered by auto
307 * detected bad OI mappings during the RPC services. */
309 OFSR_DEFAULT = 10000,
312 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
314 /* There are at most 10 uid/gids are affected in a transaction, and
315 * that's rename case:
316 * - 2 for source parent uid & gid;
317 * - 2 for source child uid & gid ('..' entry update when child is directory);
318 * - 2 for target parent uid & gid;
319 * - 2 for target child uid & gid (if the target child exists);
320 * - 2 for root uid & gid (last_rcvd, llog, etc);
322 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
323 * the id type of each id in the ot_id_array.
325 #define OSD_MAX_UGID_CNT 10
330 OSD_OT_XATTR_SET = 2,
343 struct thandle ot_super;
345 struct ldiskfs_journal_cb_entry ot_jcb;
346 struct list_head ot_commit_dcb_list;
347 struct list_head ot_stop_dcb_list;
348 /* Link to the device, for debugging. */
349 struct lu_ref_link ot_dev_link;
350 unsigned short ot_credits;
351 unsigned short ot_id_cnt;
352 unsigned short ot_id_type;
353 int ot_remove_agents:1;
354 uid_t ot_id_array[OSD_MAX_UGID_CNT];
355 struct lquota_trans *ot_quota_trans;
356 #if OSD_THANDLE_STATS
357 /** time when this handle was allocated */
358 cfs_time_t oth_alloced;
360 /** time when this thanle was started */
361 cfs_time_t oth_started;
366 * Basic transaction credit op
387 #ifdef CONFIG_PROC_FS
389 LPROC_OSD_READ_BYTES = 0,
390 LPROC_OSD_WRITE_BYTES = 1,
391 LPROC_OSD_GET_PAGE = 2,
392 LPROC_OSD_NO_PAGE = 3,
393 LPROC_OSD_CACHE_ACCESS = 4,
394 LPROC_OSD_CACHE_HIT = 5,
395 LPROC_OSD_CACHE_MISS = 6,
397 #if OSD_THANDLE_STATS
398 LPROC_OSD_THANDLE_STARTING,
399 LPROC_OSD_THANDLE_OPEN,
400 LPROC_OSD_THANDLE_CLOSING,
407 * Storage representation for fids.
409 * Variable size, first byte contains the length of the whole record.
411 struct osd_fid_pack {
412 unsigned char fp_len;
413 char fp_area[sizeof(struct lu_fid)];
416 struct osd_it_ea_dirent {
417 struct lu_fid oied_fid;
420 unsigned short oied_namelen;
421 unsigned int oied_type;
423 } __attribute__((packed));
426 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
427 * than lu_dirent struct. osd readdir reads less number of dirent than
428 * required for mdd dir page. so buffer size need to be increased so that
429 * there would be one ext3 readdir for every mdd readdir page.
432 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
435 * This is iterator's in-memory data structure in interoperability
436 * mode (i.e. iterator over ldiskfs style directory)
439 struct osd_object *oie_obj;
440 /** used in ldiskfs iterator, to stored file pointer */
441 struct file oie_file;
442 /** how many entries have been read-cached from storage */
444 /** current entry is being iterated by caller */
446 /** current processing entry */
447 struct osd_it_ea_dirent *oie_dirent;
448 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
450 struct dentry oie_dentry;
454 * Iterator's in-memory data structure for IAM mode.
457 struct osd_object *oi_obj;
458 struct iam_path_descr *oi_ipd;
459 struct iam_iterator oi_it;
462 struct osd_quota_leaf {
463 struct list_head oql_link;
468 * Iterator's in-memory data structure for quota file.
470 struct osd_it_quota {
471 struct osd_object *oiq_obj;
472 /** tree blocks path to where the entry is stored */
473 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
474 /** on-disk offset for current key where quota record can be found */
476 /** identifier for current quota record */
478 /** the record index in the leaf/index block */
479 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
480 /** list of already processed leaf blocks */
481 struct list_head oiq_list;
484 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
487 wait_queue_head_t dr_wait;
488 atomic_t dr_numreqs; /* number of reqs being processed */
493 unsigned int dr_ignore_quota:1;
494 unsigned int dr_elapsed_valid:1; /* we really did count time */
495 unsigned int dr_rw:1;
496 struct lu_buf dr_pg_buf;
497 struct page **dr_pages;
498 struct lu_buf dr_bl_buf;
500 unsigned long dr_start_time;
501 unsigned long dr_elapsed; /* how long io took */
502 struct osd_device *dr_dev;
503 unsigned int dr_init_at; /* the line iobuf was initialized */
506 struct osd_thread_info {
507 const struct lu_env *oti_env;
509 * used for index operations.
511 struct dentry oti_obj_dentry;
512 struct dentry oti_child_dentry;
514 /** dentry for Iterator context. */
515 struct dentry oti_it_dentry;
518 /* fake struct file for osd_object_sync */
519 struct file oti_file;
521 struct kstatfs oti_ksfs;
524 struct htree_lock *oti_hlock;
526 struct lu_fid oti_fid;
527 struct lu_fid oti_fid2;
528 struct lu_fid oti_fid3;
529 struct osd_inode_id oti_id;
530 struct osd_inode_id oti_id2;
531 struct osd_inode_id oti_id3;
532 struct ost_id oti_ostid;
535 * XXX temporary: for ->i_op calls.
537 struct timespec oti_time;
539 /** osd_device reference, initialized in osd_trans_start() and
540 used in osd_trans_stop() */
541 struct osd_device *oti_dev;
544 * following ipd and it structures are used for osd_index_iam_lookup()
545 * these are defined separately as we might do index operation
546 * in open iterator session.
549 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
551 unsigned int oti_it_ea_buf_used:1;
553 /* IAM iterator for index operation. */
554 struct iam_iterator oti_idx_it;
556 /** union to guarantee that ->oti_ipd[] has proper alignment. */
559 char oti_it_ipd[DX_IPD_MAX_SIZE];
560 long long oti_alignment_lieutenant;
564 char oti_idx_ipd[DX_IPD_MAX_SIZE];
565 long long oti_alignment_lieutenant_colonel;
568 struct osd_idmap_cache oti_cache;
573 /** used in osd_fid_set() to put xattr */
574 struct lu_buf oti_buf;
575 struct lu_buf oti_big_buf;
576 /** used in osd_ea_fid_set() to set fid into common ea */
578 struct lustre_mdt_attrs oti_mdt_attrs;
579 /* old LMA for compatibility */
580 char oti_mdt_attrs_old[LMA_OLD_SIZE];
581 struct filter_fid_old oti_ff;
582 struct filter_fid oti_ff_new;
585 struct osd_iobuf oti_iobuf;
586 /* used to access objects in /O */
587 struct inode *oti_inode;
588 #define OSD_FID_REC_SZ 32
589 char oti_ldp[OSD_FID_REC_SZ];
590 char oti_ldp2[OSD_FID_REC_SZ];
592 /* used by quota code */
594 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
595 struct fs_disk_quota oti_fdq;
597 struct if_dqblk oti_dqblk;
599 struct if_dqinfo oti_dqinfo;
601 struct lquota_id_info oti_qi;
602 struct lquota_trans oti_quota_trans;
603 union lquota_rec oti_quota_rec;
605 struct lu_seq_range oti_seq_range;
607 /* Tracking for transaction credits, to allow debugging and optimizing
608 * cases where a large number of credits are being allocated for
609 * single transaction. */
610 unsigned int oti_credits_before;
611 unsigned short oti_declare_ops[OSD_OT_MAX];
612 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
613 unsigned short oti_declare_ops_used[OSD_OT_MAX];
616 extern int ldiskfs_pdo;
618 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
619 const char *name, void *buf, int len)
624 dentry->d_inode = inode;
625 dentry->d_sb = inode->i_sb;
626 return inode->i_op->getxattr(dentry, name, buf, len);
629 static inline int __osd_xattr_set(struct osd_thread_info *info,
630 struct inode *inode, const char *name,
631 const void *buf, int buflen, int fl)
633 struct dentry *dentry = &info->oti_child_dentry;
635 ll_vfs_dq_init(inode);
636 dentry->d_inode = inode;
637 dentry->d_sb = inode->i_sb;
638 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
641 #ifdef CONFIG_PROC_FS
643 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
644 extern struct lprocfs_vars lprocfs_osd_module_vars[];
645 int osd_procfs_init(struct osd_device *osd, const char *name);
646 int osd_procfs_fini(struct osd_device *osd);
647 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
648 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
649 int osd_register_proc_index_in_idif(struct osd_device *osd);
653 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
654 struct obd_statfs *sfs);
655 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
656 struct osd_inode_id *id);
657 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
658 const struct lu_fid *fid, __u32 compat, __u32 incompat);
659 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
660 struct dentry *dentry, struct lustre_mdt_attrs *lma);
661 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
662 struct osd_inode_id *id, const struct lu_fid *fid);
663 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
664 struct dentry *dentry, struct lu_fid *fid);
666 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
667 void osd_obj_map_fini(struct osd_device *dev);
668 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
669 const struct lu_fid *fid, struct osd_inode_id *id);
670 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
671 const struct lu_fid *fid, const struct osd_inode_id *id,
673 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
674 const struct lu_fid *fid, handle_t *th);
675 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
676 const struct lu_fid *fid, const struct osd_inode_id *id,
678 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
679 struct inode *src_parent, struct dentry *src_child,
680 const struct lu_fid *fid);
681 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
682 const struct lu_fid *fid, struct osd_inode_id *id);
683 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
684 const struct lu_fid *fid, const struct osd_inode_id *id,
686 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
687 const struct lu_fid *fid, const struct osd_inode_id *id,
690 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
691 int osd_scrub_file_store(struct osd_scrub *scrub);
692 char *osd_lf_fid2name(const struct lu_fid *fid);
693 int osd_scrub_start(struct osd_device *dev, __u32 flags);
694 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
695 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
696 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
698 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
699 struct osd_inode_id *id);
700 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
702 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
703 u64 seq, struct lu_seq_range *range);
705 int osd_delete_from_remote_parent(const struct lu_env *env,
706 struct osd_device *osd,
707 struct osd_object *obj,
708 struct osd_thandle *oh);
709 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
710 struct osd_object *obj, struct osd_thandle *oh);
711 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
712 struct osd_device *osd,
713 const struct lu_fid *fid,
714 struct osd_inode_id *id);
716 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
718 /* osd_quota_fmt.c */
719 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
720 int type, uint blk, int depth, uint index,
721 struct osd_it_quota *it);
722 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
723 int type, uint blk, uint index,
724 struct osd_it_quota *it);
725 loff_t find_tree_dqentry(const struct lu_env *env,
726 struct osd_object *obj, int type,
727 qid_t dqid, uint blk, int depth,
728 struct osd_it_quota *it);
730 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
731 struct lquota_id_info *qi, struct osd_object *obj,
732 bool enforce, int *flags);
733 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
734 long long space, struct osd_thandle *oh,
735 struct osd_object *obj, bool is_blk, int *flags,
737 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
738 const struct dt_rec *rec,
739 union lquota_rec *quota_rec);
740 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
741 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
743 #ifndef HAVE_I_UID_READ
744 static inline uid_t i_uid_read(const struct inode *inode)
749 static inline gid_t i_gid_read(const struct inode *inode)
754 static inline void i_uid_write(struct inode *inode, uid_t uid)
759 static inline void i_gid_write(struct inode *inode, gid_t gid)
765 #ifdef LDISKFS_HT_MISC
766 # define osd_journal_start_sb(sb, type, nblock) \
767 ldiskfs_journal_start_sb(sb, type, nblock)
768 # define osd_ldiskfs_append(handle, inode, nblock, err) \
769 ldiskfs_append(handle, inode, nblock)
770 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
771 __ldiskfs_find_entry(dir, name, de, inlined, lock)
772 # define osd_journal_start(inode, type, nblocks) \
773 ldiskfs_journal_start(inode, type, nblocks)
774 # define osd_transaction_size(dev) \
775 (osd_journal(dev)->j_max_transaction_buffers / 2)
777 # define LDISKFS_HT_MISC 0
778 # define osd_journal_start_sb(sb, type, nblock) \
779 ldiskfs_journal_start_sb(sb, nblock)
780 # define osd_ldiskfs_append(handle, inode, nblock, err) \
781 ldiskfs_append(handle, inode, nblock, err)
782 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
783 __ldiskfs_find_entry(dir, name, de, lock)
784 # define osd_journal_start(inode, type, nblocks) \
785 ldiskfs_journal_start(inode, nblocks)
786 # define osd_transaction_size(dev) \
787 (osd_journal(dev)->j_max_transaction_buffers)
791 * Invariants, assertions.
795 * XXX: do not enable this, until invariant checking code is made thread safe
796 * in the face of pdirops locking.
798 #define OSD_INVARIANT_CHECKS (0)
800 #if OSD_INVARIANT_CHECKS
801 static inline int osd_invariant(const struct osd_object *obj)
805 ergo(obj->oo_inode != NULL,
806 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
807 atomic_read(&obj->oo_inode->i_count) > 0) &&
808 ergo(obj->oo_dir != NULL &&
809 obj->oo_dir->od_conationer.ic_object != NULL,
810 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
813 #define osd_invariant(obj) (1)
816 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
818 extern const struct dt_index_operations osd_otable_ops;
820 static inline int osd_oi_fid2idx(struct osd_device *dev,
821 const struct lu_fid *fid)
823 return fid->f_seq & (dev->od_oi_count - 1);
826 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
827 const struct lu_fid *fid)
829 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
830 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
831 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
832 DFID"\n", PFID(fid));
833 /* It can work even od_oi_count equals to 1 although it's unexpected,
834 * the only reason we set it to 1 is for performance measurement */
835 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
838 extern const struct lu_device_operations osd_lu_ops;
840 static inline int lu_device_is_osd(const struct lu_device *d)
842 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
845 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
847 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
848 return container_of0(d, struct osd_device, od_dt_dev);
851 static inline struct osd_device *osd_dev(const struct lu_device *d)
853 LASSERT(lu_device_is_osd(d));
854 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
857 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
859 return osd_dev(o->oo_dt.do_lu.lo_dev);
862 static inline struct super_block *osd_sb(const struct osd_device *dev)
864 return dev->od_mnt->mnt_sb;
867 static inline int osd_object_is_root(const struct osd_object *obj)
869 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
872 static inline struct osd_object *osd_obj(const struct lu_object *o)
874 LASSERT(lu_device_is_osd(o->lo_dev));
875 return container_of0(o, struct osd_object, oo_dt.do_lu);
878 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
880 return osd_obj(&d->do_lu);
883 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
885 return &osd->od_dt_dev.dd_lu_dev;
888 static inline journal_t *osd_journal(const struct osd_device *dev)
890 return LDISKFS_SB(osd_sb(dev))->s_journal;
893 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
895 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
898 static inline char *osd_name(struct osd_device *osd)
900 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
903 extern const struct dt_body_operations osd_body_ops;
904 extern struct lu_context_key osd_key;
906 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
908 return lu_context_key_get(&env->le_ctx, &osd_key);
911 extern const struct dt_body_operations osd_body_ops_new;
917 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
918 const struct iam_container *bag)
920 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
921 osd_oti_get(env)->oti_it_ipd);
925 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
926 const struct iam_container *bag)
928 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
929 osd_oti_get(env)->oti_idx_ipd);
932 static inline void osd_ipd_put(const struct lu_env *env,
933 const struct iam_container *bag,
934 struct iam_path_descr *ipd)
936 bag->ic_descr->id_ops->id_ipd_free(ipd);
939 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
940 const loff_t size, const loff_t pos,
943 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
944 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
945 int write_NUL, loff_t *offs, handle_t *handle);
948 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
950 const char *name, const int namelen)
952 struct osd_thread_info *info = osd_oti_get(env);
953 struct dentry *child_dentry = &info->oti_child_dentry;
954 struct dentry *obj_dentry = &info->oti_obj_dentry;
956 obj_dentry->d_inode = inode;
957 obj_dentry->d_sb = inode->i_sb;
958 obj_dentry->d_name.hash = 0;
960 child_dentry->d_name.hash = 0;
961 child_dentry->d_parent = obj_dentry;
962 child_dentry->d_name.name = name;
963 child_dentry->d_name.len = namelen;
967 extern int osd_trans_declare_op2rb[];
968 extern int ldiskfs_track_declares_assert;
969 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
971 static inline void osd_trans_declare_op(const struct lu_env *env,
972 struct osd_thandle *oh,
973 unsigned int op, int credits)
975 struct osd_thread_info *oti = osd_oti_get(env);
977 LASSERT(oh->ot_handle == NULL);
978 if (unlikely(op >= OSD_OT_MAX)) {
979 if (unlikely(ldiskfs_track_declares_assert)) {
980 LASSERT(op < OSD_OT_MAX);
982 CWARN("%s: Invalid operation index %d\n",
983 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
984 libcfs_debug_dumpstack(NULL);
987 oti->oti_declare_ops[op]++;
988 oti->oti_declare_ops_cred[op] += credits;
990 oh->ot_credits += credits;
993 static inline void osd_trans_exec_op(const struct lu_env *env,
994 struct thandle *th, unsigned int op)
996 struct osd_thread_info *oti = osd_oti_get(env);
997 struct osd_thandle *oh = container_of(th, struct osd_thandle,
999 unsigned int rb, left;
1001 LASSERT(oh->ot_handle != NULL);
1002 if (unlikely(op >= OSD_OT_MAX)) {
1003 if (unlikely(ldiskfs_track_declares_assert))
1004 LASSERT(op < OSD_OT_MAX);
1006 CWARN("%s: Invalid operation index %d\n",
1007 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1008 libcfs_debug_dumpstack(NULL);
1013 /* find rollback (or reverse) operation for the given one
1014 * such an operation doesn't require additional credits
1015 * as the same set of blocks are modified */
1016 rb = osd_trans_declare_op2rb[op];
1018 /* check whether credits for this operation were reserved at all */
1019 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1020 oti->oti_declare_ops_cred[rb] == 0)) {
1021 /* the API is not perfect yet: CREATE does REF_ADD internally
1022 * while DESTROY does not. To rollback CREATE the callers
1023 * needs to call REF_DEL+DESTROY which is hard to detect using
1024 * a simple table of rollback operations */
1025 if (op == OSD_OT_REF_DEL &&
1026 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1028 if (op == OSD_OT_REF_ADD &&
1029 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1031 osd_trans_dump_creds(env, th);
1032 CERROR("%s: op = %d, rb = %d\n",
1033 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1034 if (unlikely(ldiskfs_track_declares_assert))
1039 /* remember how many credits we have unused before the operation */
1040 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1041 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1042 if (unlikely(oti->oti_credits_before < left)) {
1043 osd_trans_dump_creds(env, th);
1044 CERROR("%s: op = %d, rb = %d\n",
1045 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1046 /* on a very small fs (testing?) it's possible that
1047 * the transaction can't fit 1/4 of journal, so we
1048 * just request less credits (see osd_trans_start()).
1049 * ignore the same case here */
1050 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1051 if (unlikely(oh->ot_credits < rb)) {
1052 if (unlikely(ldiskfs_track_declares_assert))
1058 static inline void osd_trans_exec_check(const struct lu_env *env,
1062 struct osd_thread_info *oti = osd_oti_get(env);
1063 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1065 int used, over, quota;
1067 /* how many credits have been used by the operation */
1068 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1070 if (unlikely(used < 0)) {
1071 /* if some block was allocated and released in the same
1072 * transaction, then it won't be a part of the transaction
1073 * and delta can be negative */
1078 /* rollback operations (e.g. when we destroy just created
1079 * object) should not consume any credits. there is no point
1080 * to confuse the checks below */
1084 oti->oti_declare_ops_used[op] += used;
1085 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1088 /* we account quota for a whole transaction and any operation can
1089 * consume corresponding credits */
1090 over = oti->oti_declare_ops_used[op] -
1091 oti->oti_declare_ops_cred[op];
1092 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1093 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1094 if (over <= quota) {
1095 /* probably that credits were consumed by
1096 * quota indirectly (in the depths of ldiskfs) */
1097 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1098 oti->oti_declare_ops_used[op] -= over;
1100 CWARN("op %d: used %u, used now %u, reserved %u\n",
1101 op, oti->oti_declare_ops_used[op], used,
1102 oti->oti_declare_ops_cred[op]);
1103 osd_trans_dump_creds(env, th);
1104 if (unlikely(ldiskfs_track_declares_assert))
1110 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1113 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1114 struct lu_fid *befider)
1116 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1117 memcpy(pack->fp_area, befider, sizeof(*befider));
1118 pack->fp_len = sizeof(*befider) + 1;
1122 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1127 switch (pack->fp_len) {
1128 case sizeof *fid + 1:
1129 memcpy(fid, pack->fp_area, sizeof *fid);
1130 fid_be_to_cpu(fid, fid);
1133 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1140 * Quota/Accounting handling
1142 extern const struct dt_index_operations osd_acct_index_ops;
1143 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1144 const struct lu_fid *fid, struct osd_inode_id *id);
1146 /* copy from fs/ext4/dir.c */
1147 static inline int is_32bit_api(void)
1149 #ifdef CONFIG_COMPAT
1150 return is_compat_task();
1152 return (BITS_PER_LONG == 32);
1156 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1158 if ((filp->f_mode & FMODE_32BITHASH) ||
1159 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1160 return LDISKFS_HTREE_EOF_32BIT;
1162 return LDISKFS_HTREE_EOF_64BIT;
1165 static inline int fid_is_internal(const struct lu_fid *fid)
1167 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1170 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1172 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1175 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1176 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1178 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1180 #endif /* _OSD_INTERNAL_H */