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. */
226 struct dt_device od_dt_dev;
227 /* information about underlying file system */
228 struct vfsmount *od_mnt;
230 struct osd_oi **od_oi_table;
231 /* total number of OI containers */
236 unsigned int od_fl_capa:1,
244 __u32 od_dirent_journal;
246 struct proc_dir_entry *od_proc_entry;
247 struct lprocfs_stats *od_stats;
249 spinlock_t od_osfs_lock;
252 struct lu_site od_site;
254 struct osd_obj_map *od_ost_map;
255 struct osd_mdobj_map *od_mdt_map;
257 unsigned long long od_readcache_max_filesize;
259 int od_writethrough_cache;
261 struct brw_stats od_brw_stats;
262 atomic_t od_r_in_flight;
263 atomic_t od_w_in_flight;
265 struct mutex od_otable_mutex;
266 struct osd_otable_it *od_otable_it;
267 struct osd_scrub od_scrub;
268 struct list_head od_ios_list;
270 /* service name associated with the osd device */
271 char od_svname[MAX_OBD_NAME];
272 char od_mntdev[MAX_OBD_NAME];
274 /* quota slave instance */
275 struct qsd_instance *od_quota_slave;
277 /* osd seq instance */
278 struct lu_client_seq *od_cl_seq;
279 /* If the ratio of "the total OI mappings count" vs
280 * "the bad OI mappings count" is lower than the
281 * osd_device::od_full_scrub_ratio, then trigger
282 * OI scrub to scan the whole the device. */
283 __u64 od_full_scrub_ratio;
284 /* If the speed of found bad OI mappings (per minute)
285 * exceeds the osd_device::od_full_scrub_threshold_rate,
286 * then trigger OI scrub to scan the whole device. */
287 __u64 od_full_scrub_threshold_rate;
290 enum osd_full_scrub_ratio {
291 /* Trigger OI scrub to scan the whole device directly. */
294 /* Because the bad OI mappings count cannot be larger than
295 * the total OI mappints count, then setting OFSR_NEVER means
296 * that the whole device scanning cannot be triggered by auto
297 * detected bad OI mappings during the RPC services. */
299 OFSR_DEFAULT = 10000,
302 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
304 /* There are at most 10 uid/gids are affected in a transaction, and
305 * that's rename case:
306 * - 2 for source parent uid & gid;
307 * - 2 for source child uid & gid ('..' entry update when child is directory);
308 * - 2 for target parent uid & gid;
309 * - 2 for target child uid & gid (if the target child exists);
310 * - 2 for root uid & gid (last_rcvd, llog, etc);
312 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
313 * the id type of each id in the ot_id_array.
315 #define OSD_MAX_UGID_CNT 10
320 OSD_OT_XATTR_SET = 2,
333 struct thandle ot_super;
335 struct ldiskfs_journal_cb_entry ot_jcb;
336 struct list_head ot_commit_dcb_list;
337 struct list_head ot_stop_dcb_list;
338 /* Link to the device, for debugging. */
339 struct lu_ref_link ot_dev_link;
340 unsigned short ot_credits;
341 unsigned short ot_id_cnt;
342 unsigned short ot_id_type;
343 uid_t ot_id_array[OSD_MAX_UGID_CNT];
344 struct lquota_trans *ot_quota_trans;
345 #if OSD_THANDLE_STATS
346 /** time when this handle was allocated */
347 cfs_time_t oth_alloced;
349 /** time when this thanle was started */
350 cfs_time_t oth_started;
355 * Basic transaction credit op
376 #ifdef CONFIG_PROC_FS
378 LPROC_OSD_READ_BYTES = 0,
379 LPROC_OSD_WRITE_BYTES = 1,
380 LPROC_OSD_GET_PAGE = 2,
381 LPROC_OSD_NO_PAGE = 3,
382 LPROC_OSD_CACHE_ACCESS = 4,
383 LPROC_OSD_CACHE_HIT = 5,
384 LPROC_OSD_CACHE_MISS = 6,
386 #if OSD_THANDLE_STATS
387 LPROC_OSD_THANDLE_STARTING,
388 LPROC_OSD_THANDLE_OPEN,
389 LPROC_OSD_THANDLE_CLOSING,
396 * Storage representation for fids.
398 * Variable size, first byte contains the length of the whole record.
400 struct osd_fid_pack {
401 unsigned char fp_len;
402 char fp_area[sizeof(struct lu_fid)];
405 struct osd_it_ea_dirent {
406 struct lu_fid oied_fid;
409 unsigned short oied_namelen;
410 unsigned int oied_type;
412 } __attribute__((packed));
415 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
416 * than lu_dirent struct. osd readdir reads less number of dirent than
417 * required for mdd dir page. so buffer size need to be increased so that
418 * there would be one ext3 readdir for every mdd readdir page.
421 #define OSD_IT_EA_BUFSIZE (PAGE_CACHE_SIZE + PAGE_CACHE_SIZE/4)
424 * This is iterator's in-memory data structure in interoperability
425 * mode (i.e. iterator over ldiskfs style directory)
428 struct osd_object *oie_obj;
429 /** used in ldiskfs iterator, to stored file pointer */
430 struct file oie_file;
431 /** how many entries have been read-cached from storage */
433 /** current entry is being iterated by caller */
435 /** current processing entry */
436 struct osd_it_ea_dirent *oie_dirent;
437 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
439 struct dentry oie_dentry;
443 * Iterator's in-memory data structure for IAM mode.
446 struct osd_object *oi_obj;
447 struct iam_path_descr *oi_ipd;
448 struct iam_iterator oi_it;
451 struct osd_quota_leaf {
452 struct list_head oql_link;
457 * Iterator's in-memory data structure for quota file.
459 struct osd_it_quota {
460 struct osd_object *oiq_obj;
461 /** tree blocks path to where the entry is stored */
462 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
463 /** on-disk offset for current key where quota record can be found */
465 /** identifier for current quota record */
467 /** the record index in the leaf/index block */
468 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
469 /** list of already processed leaf blocks */
470 struct list_head oiq_list;
473 #define MAX_BLOCKS_PER_PAGE (PAGE_CACHE_SIZE / 512)
476 wait_queue_head_t dr_wait;
477 atomic_t dr_numreqs; /* number of reqs being processed */
482 unsigned int dr_ignore_quota:1;
483 unsigned int dr_elapsed_valid:1; /* we really did count time */
484 unsigned int dr_rw:1;
485 struct lu_buf dr_pg_buf;
486 struct page **dr_pages;
487 struct lu_buf dr_bl_buf;
489 unsigned long dr_start_time;
490 unsigned long dr_elapsed; /* how long io took */
491 struct osd_device *dr_dev;
492 unsigned int dr_init_at; /* the line iobuf was initialized */
495 struct osd_thread_info {
496 const struct lu_env *oti_env;
498 * used for index operations.
500 struct dentry oti_obj_dentry;
501 struct dentry oti_child_dentry;
503 /** dentry for Iterator context. */
504 struct dentry oti_it_dentry;
507 /* fake struct file for osd_object_sync */
508 struct file oti_file;
510 struct kstatfs oti_ksfs;
513 struct htree_lock *oti_hlock;
515 struct lu_fid oti_fid;
516 struct lu_fid oti_fid2;
517 struct lu_fid oti_fid3;
518 struct osd_inode_id oti_id;
519 struct osd_inode_id oti_id2;
520 struct osd_inode_id oti_id3;
521 struct ost_id oti_ostid;
524 * XXX temporary: for ->i_op calls.
526 struct timespec oti_time;
528 /** osd_device reference, initialized in osd_trans_start() and
529 used in osd_trans_stop() */
530 struct osd_device *oti_dev;
533 * following ipd and it structures are used for osd_index_iam_lookup()
534 * these are defined separately as we might do index operation
535 * in open iterator session.
538 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
540 unsigned int oti_it_ea_buf_used:1;
542 /* IAM iterator for index operation. */
543 struct iam_iterator oti_idx_it;
545 /** union to guarantee that ->oti_ipd[] has proper alignment. */
548 char oti_it_ipd[DX_IPD_MAX_SIZE];
549 long long oti_alignment_lieutenant;
553 char oti_idx_ipd[DX_IPD_MAX_SIZE];
554 long long oti_alignment_lieutenant_colonel;
557 struct osd_idmap_cache oti_cache;
562 /** used in osd_fid_set() to put xattr */
563 struct lu_buf oti_buf;
564 struct lu_buf oti_big_buf;
565 /** used in osd_ea_fid_set() to set fid into common ea */
567 struct lustre_mdt_attrs oti_mdt_attrs;
568 /* old LMA for compatibility */
569 char oti_mdt_attrs_old[LMA_OLD_SIZE];
570 struct filter_fid_old oti_ff;
571 struct filter_fid oti_ff_new;
574 struct osd_iobuf oti_iobuf;
575 /* used to access objects in /O */
576 struct inode *oti_inode;
577 #define OSD_FID_REC_SZ 32
578 char oti_ldp[OSD_FID_REC_SZ];
579 char oti_ldp2[OSD_FID_REC_SZ];
581 /* used by quota code */
583 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
584 struct fs_disk_quota oti_fdq;
586 struct if_dqblk oti_dqblk;
588 struct if_dqinfo oti_dqinfo;
590 struct lquota_id_info oti_qi;
591 struct lquota_trans oti_quota_trans;
592 union lquota_rec oti_quota_rec;
594 struct lu_seq_range oti_seq_range;
596 /* Tracking for transaction credits, to allow debugging and optimizing
597 * cases where a large number of credits are being allocated for
598 * single transaction. */
599 unsigned int oti_credits_before;
600 unsigned short oti_declare_ops[OSD_OT_MAX];
601 unsigned short oti_declare_ops_cred[OSD_OT_MAX];
602 unsigned short oti_declare_ops_used[OSD_OT_MAX];
605 extern int ldiskfs_pdo;
607 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
608 const char *name, void *buf, int len)
613 dentry->d_inode = inode;
614 dentry->d_sb = inode->i_sb;
615 return inode->i_op->getxattr(dentry, name, buf, len);
618 static inline int __osd_xattr_set(struct osd_thread_info *info,
619 struct inode *inode, const char *name,
620 const void *buf, int buflen, int fl)
622 struct dentry *dentry = &info->oti_child_dentry;
624 ll_vfs_dq_init(inode);
625 dentry->d_inode = inode;
626 dentry->d_sb = inode->i_sb;
627 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
630 #ifdef CONFIG_PROC_FS
632 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
633 extern struct lprocfs_vars lprocfs_osd_module_vars[];
634 int osd_procfs_init(struct osd_device *osd, const char *name);
635 int osd_procfs_fini(struct osd_device *osd);
636 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
637 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
638 int osd_register_proc_index_in_idif(struct osd_device *osd);
642 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
643 struct obd_statfs *sfs);
644 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
645 struct osd_inode_id *id);
646 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
647 const struct lu_fid *fid, __u32 compat, __u32 incompat);
648 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
649 struct dentry *dentry, struct lustre_mdt_attrs *lma);
650 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
651 struct osd_inode_id *id, const struct lu_fid *fid);
652 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
653 struct dentry *dentry, struct lu_fid *fid);
655 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
656 void osd_obj_map_fini(struct osd_device *dev);
657 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
658 const struct lu_fid *fid, struct osd_inode_id *id);
659 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
660 const struct lu_fid *fid, const struct osd_inode_id *id,
662 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
663 const struct lu_fid *fid, handle_t *th);
664 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
665 const struct lu_fid *fid, const struct osd_inode_id *id,
667 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
668 struct inode *src_parent, struct dentry *src_child,
669 const struct lu_fid *fid);
670 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
671 const struct lu_fid *fid, struct osd_inode_id *id);
672 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
673 const struct lu_fid *fid, const struct osd_inode_id *id,
675 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
676 const struct lu_fid *fid, const struct osd_inode_id *id,
679 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
680 int osd_scrub_file_store(struct osd_scrub *scrub);
681 char *osd_lf_fid2name(const struct lu_fid *fid);
682 int osd_scrub_start(struct osd_device *dev, __u32 flags);
683 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
684 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
685 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
687 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
688 struct osd_inode_id *id);
689 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
691 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
692 u64 seq, struct lu_seq_range *range);
694 int osd_delete_from_remote_parent(const struct lu_env *env,
695 struct osd_device *osd,
696 struct osd_object *obj,
697 struct osd_thandle *oh);
698 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
699 struct osd_object *obj, struct osd_thandle *oh);
700 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
701 struct osd_device *osd,
702 const struct lu_fid *fid,
703 struct osd_inode_id *id);
705 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
707 /* osd_quota_fmt.c */
708 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
709 int type, uint blk, int depth, uint index,
710 struct osd_it_quota *it);
711 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
712 int type, uint blk, uint index,
713 struct osd_it_quota *it);
714 loff_t find_tree_dqentry(const struct lu_env *env,
715 struct osd_object *obj, int type,
716 qid_t dqid, uint blk, int depth,
717 struct osd_it_quota *it);
719 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
720 struct lquota_id_info *qi, struct osd_object *obj,
721 bool enforce, int *flags);
722 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
723 long long space, struct osd_thandle *oh,
724 struct osd_object *obj, bool is_blk, int *flags,
726 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
727 const struct dt_rec *rec,
728 union lquota_rec *quota_rec);
729 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
730 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt);
732 #ifndef HAVE_I_UID_READ
733 static inline uid_t i_uid_read(const struct inode *inode)
738 static inline gid_t i_gid_read(const struct inode *inode)
743 static inline void i_uid_write(struct inode *inode, uid_t uid)
748 static inline void i_gid_write(struct inode *inode, gid_t gid)
754 #ifdef LDISKFS_HT_MISC
755 # define osd_journal_start_sb(sb, type, nblock) \
756 ldiskfs_journal_start_sb(sb, type, nblock)
757 # define osd_ldiskfs_append(handle, inode, nblock, err) \
758 ldiskfs_append(handle, inode, nblock)
759 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
760 __ldiskfs_find_entry(dir, name, de, inlined, lock)
761 # define osd_journal_start(inode, type, nblocks) \
762 ldiskfs_journal_start(inode, type, nblocks)
763 # define osd_transaction_size(dev) \
764 (osd_journal(dev)->j_max_transaction_buffers / 2)
766 # define LDISKFS_HT_MISC 0
767 # define osd_journal_start_sb(sb, type, nblock) \
768 ldiskfs_journal_start_sb(sb, nblock)
769 # define osd_ldiskfs_append(handle, inode, nblock, err) \
770 ldiskfs_append(handle, inode, nblock, err)
771 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
772 __ldiskfs_find_entry(dir, name, de, lock)
773 # define osd_journal_start(inode, type, nblocks) \
774 ldiskfs_journal_start(inode, nblocks)
775 # define osd_transaction_size(dev) \
776 (osd_journal(dev)->j_max_transaction_buffers)
780 * Invariants, assertions.
784 * XXX: do not enable this, until invariant checking code is made thread safe
785 * in the face of pdirops locking.
787 #define OSD_INVARIANT_CHECKS (0)
789 #if OSD_INVARIANT_CHECKS
790 static inline int osd_invariant(const struct osd_object *obj)
794 ergo(obj->oo_inode != NULL,
795 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
796 atomic_read(&obj->oo_inode->i_count) > 0) &&
797 ergo(obj->oo_dir != NULL &&
798 obj->oo_dir->od_conationer.ic_object != NULL,
799 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
802 #define osd_invariant(obj) (1)
805 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
807 extern const struct dt_index_operations osd_otable_ops;
809 static inline int osd_oi_fid2idx(struct osd_device *dev,
810 const struct lu_fid *fid)
812 return fid->f_seq & (dev->od_oi_count - 1);
815 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
816 const struct lu_fid *fid)
818 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
819 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
820 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
821 DFID"\n", PFID(fid));
822 /* It can work even od_oi_count equals to 1 although it's unexpected,
823 * the only reason we set it to 1 is for performance measurement */
824 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
827 extern const struct lu_device_operations osd_lu_ops;
829 static inline int lu_device_is_osd(const struct lu_device *d)
831 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
834 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
836 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
837 return container_of0(d, struct osd_device, od_dt_dev);
840 static inline struct osd_device *osd_dev(const struct lu_device *d)
842 LASSERT(lu_device_is_osd(d));
843 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
846 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
848 return osd_dev(o->oo_dt.do_lu.lo_dev);
851 static inline struct super_block *osd_sb(const struct osd_device *dev)
853 return dev->od_mnt->mnt_sb;
856 static inline int osd_object_is_root(const struct osd_object *obj)
858 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
861 static inline struct osd_object *osd_obj(const struct lu_object *o)
863 LASSERT(lu_device_is_osd(o->lo_dev));
864 return container_of0(o, struct osd_object, oo_dt.do_lu);
867 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
869 return osd_obj(&d->do_lu);
872 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
874 return &osd->od_dt_dev.dd_lu_dev;
877 static inline journal_t *osd_journal(const struct osd_device *dev)
879 return LDISKFS_SB(osd_sb(dev))->s_journal;
882 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
884 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
887 static inline char *osd_name(struct osd_device *osd)
889 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
892 extern const struct dt_body_operations osd_body_ops;
893 extern struct lu_context_key osd_key;
895 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
897 return lu_context_key_get(&env->le_ctx, &osd_key);
900 extern const struct dt_body_operations osd_body_ops_new;
906 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
907 const struct iam_container *bag)
909 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
910 osd_oti_get(env)->oti_it_ipd);
914 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
915 const struct iam_container *bag)
917 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
918 osd_oti_get(env)->oti_idx_ipd);
921 static inline void osd_ipd_put(const struct lu_env *env,
922 const struct iam_container *bag,
923 struct iam_path_descr *ipd)
925 bag->ic_descr->id_ops->id_ipd_free(ipd);
928 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
929 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
930 int write_NUL, loff_t *offs, handle_t *handle);
933 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
935 const char *name, const int namelen)
937 struct osd_thread_info *info = osd_oti_get(env);
938 struct dentry *child_dentry = &info->oti_child_dentry;
939 struct dentry *obj_dentry = &info->oti_obj_dentry;
941 obj_dentry->d_inode = inode;
942 obj_dentry->d_sb = inode->i_sb;
943 obj_dentry->d_name.hash = 0;
945 child_dentry->d_name.hash = 0;
946 child_dentry->d_parent = obj_dentry;
947 child_dentry->d_name.name = name;
948 child_dentry->d_name.len = namelen;
952 extern int osd_trans_declare_op2rb[];
953 extern int ldiskfs_track_declares_assert;
954 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
956 static inline void osd_trans_declare_op(const struct lu_env *env,
957 struct osd_thandle *oh,
958 unsigned int op, int credits)
960 struct osd_thread_info *oti = osd_oti_get(env);
962 LASSERT(oh->ot_handle == NULL);
963 if (unlikely(op >= OSD_OT_MAX)) {
964 if (unlikely(ldiskfs_track_declares_assert)) {
965 LASSERT(op < OSD_OT_MAX);
967 CWARN("%s: Invalid operation index %d\n",
968 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
969 libcfs_debug_dumpstack(NULL);
972 oti->oti_declare_ops[op]++;
973 oti->oti_declare_ops_cred[op] += credits;
975 oh->ot_credits += credits;
978 static inline void osd_trans_exec_op(const struct lu_env *env,
979 struct thandle *th, unsigned int op)
981 struct osd_thread_info *oti = osd_oti_get(env);
982 struct osd_thandle *oh = container_of(th, struct osd_thandle,
984 unsigned int rb, left;
986 LASSERT(oh->ot_handle != NULL);
987 if (unlikely(op >= OSD_OT_MAX)) {
988 if (unlikely(ldiskfs_track_declares_assert))
989 LASSERT(op < OSD_OT_MAX);
991 CWARN("%s: Invalid operation index %d\n",
992 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
993 libcfs_debug_dumpstack(NULL);
998 /* find rollback (or reverse) operation for the given one
999 * such an operation doesn't require additional credits
1000 * as the same set of blocks are modified */
1001 rb = osd_trans_declare_op2rb[op];
1003 /* check whether credits for this operation were reserved at all */
1004 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1005 oti->oti_declare_ops_cred[rb] == 0)) {
1006 /* the API is not perfect yet: CREATE does REF_ADD internally
1007 * while DESTROY does not. To rollback CREATE the callers
1008 * needs to call REF_DEL+DESTROY which is hard to detect using
1009 * a simple table of rollback operations */
1010 if (op == OSD_OT_REF_DEL &&
1011 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1013 if (op == OSD_OT_REF_ADD &&
1014 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1016 osd_trans_dump_creds(env, th);
1017 CERROR("%s: op = %d, rb = %d\n",
1018 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1019 if (unlikely(ldiskfs_track_declares_assert))
1024 /* remember how many credits we have unused before the operation */
1025 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1026 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1027 if (unlikely(oti->oti_credits_before < left)) {
1028 osd_trans_dump_creds(env, th);
1029 CERROR("%s: op = %d, rb = %d\n",
1030 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1031 /* on a very small fs (testing?) it's possible that
1032 * the transaction can't fit 1/4 of journal, so we
1033 * just request less credits (see osd_trans_start()).
1034 * ignore the same case here */
1035 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1036 if (unlikely(oh->ot_credits < rb)) {
1037 if (unlikely(ldiskfs_track_declares_assert))
1043 static inline void osd_trans_exec_check(const struct lu_env *env,
1047 struct osd_thread_info *oti = osd_oti_get(env);
1048 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1050 int used, over, quota;
1052 /* how many credits have been used by the operation */
1053 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1055 if (unlikely(used < 0)) {
1056 /* if some block was allocated and released in the same
1057 * transaction, then it won't be a part of the transaction
1058 * and delta can be negative */
1063 /* rollback operations (e.g. when we destroy just created
1064 * object) should not consume any credits. there is no point
1065 * to confuse the checks below */
1069 oti->oti_declare_ops_used[op] += used;
1070 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1073 /* we account quota for a whole transaction and any operation can
1074 * consume corresponding credits */
1075 over = oti->oti_declare_ops_used[op] -
1076 oti->oti_declare_ops_cred[op];
1077 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1078 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1079 if (over <= quota) {
1080 /* probably that credits were consumed by
1081 * quota indirectly (in the depths of ldiskfs) */
1082 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1083 oti->oti_declare_ops_used[op] -= over;
1085 CWARN("op %d: used %u, used now %u, reserved %u\n",
1086 op, oti->oti_declare_ops_used[op], used,
1087 oti->oti_declare_ops_cred[op]);
1088 osd_trans_dump_creds(env, th);
1089 if (unlikely(ldiskfs_track_declares_assert))
1095 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1098 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1099 struct lu_fid *befider)
1101 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1102 memcpy(pack->fp_area, befider, sizeof(*befider));
1103 pack->fp_len = sizeof(*befider) + 1;
1107 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1112 switch (pack->fp_len) {
1113 case sizeof *fid + 1:
1114 memcpy(fid, pack->fp_area, sizeof *fid);
1115 fid_be_to_cpu(fid, fid);
1118 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1125 * Quota/Accounting handling
1127 extern const struct dt_index_operations osd_acct_index_ops;
1128 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1129 const struct lu_fid *fid, struct osd_inode_id *id);
1131 /* copy from fs/ext4/dir.c */
1132 static inline int is_32bit_api(void)
1134 #ifdef CONFIG_COMPAT
1135 return is_compat_task();
1137 return (BITS_PER_LONG == 32);
1141 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1143 if ((filp->f_mode & FMODE_32BITHASH) ||
1144 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1145 return LDISKFS_HTREE_EOF_32BIT;
1147 return LDISKFS_HTREE_EOF_64BIT;
1150 static inline int fid_is_internal(const struct lu_fid *fid)
1152 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1155 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1157 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1160 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1161 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1163 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1165 #endif /* _OSD_INTERNAL_H */