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 /* check if ldiskfs support project quota */
94 #ifndef LDISKFS_IOC_FSSETXATTR
95 #undef HAVE_PROJECT_QUOTA
98 #define OBD_BRW_MAPPED OBD_BRW_LOCAL1
100 struct osd_directory {
101 struct iam_container od_container;
102 struct iam_descr od_descr;
106 * Object Index (oi) instance.
110 * underlying index object, where fid->id mapping in stored.
112 struct inode *oi_inode;
113 struct osd_directory oi_dir;
116 extern const int osd_dto_credits_noquota[];
119 struct dt_object oo_dt;
121 * Inode for file system object represented by this osd_object. This
122 * inode is pinned for the whole duration of lu_object life.
124 * Not modified concurrently (either setup early during object
125 * creation, or assigned by osd_create() under write lock).
127 struct inode *oo_inode;
129 * to protect index ops.
131 struct htree_lock_head *oo_hl_head;
132 struct rw_semaphore oo_ext_idx_sem;
133 struct rw_semaphore oo_sem;
134 struct osd_directory *oo_dir;
135 /** protects inode attributes. */
138 __u32 oo_destroyed:1;
140 /* the i_flags in LMA */
143 * Following two members are used to indicate the presence of dot and
144 * dotdot in the given directory. This is required for interop mode
147 int oo_compat_dot_created;
148 int oo_compat_dotdot_created;
150 const struct lu_env *oo_owner;
151 #ifdef CONFIG_LOCKDEP
152 struct lockdep_map oo_dep_map;
155 struct list_head oo_xattr_list;
159 /* protects on-fly initialization */
160 int oos_subdir_count; /* subdir count for each seq */
161 struct dentry *oos_root; /* O/<seq> */
162 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
163 u64 oos_seq; /* seq number */
164 struct list_head oos_seq_list; /* list to seq_list */
168 struct dentry *om_root; /* dentry for /O */
169 rwlock_t om_seq_list_lock; /* lock for seq_list */
170 struct list_head om_seq_list; /* list head for seq */
172 struct mutex om_dir_init_mutex;
176 struct dentry *om_root; /* AGENT/<index> */
177 u64 om_index; /* mdt index */
178 struct list_head om_list; /* list to omm_list */
181 struct osd_mdobj_map {
182 struct dentry *omm_remote_parent;
184 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
185 handle_t *handle, struct dentry *child,
186 struct inode *inode, struct htree_lock *hlock);
188 #define OSD_OTABLE_IT_CACHE_SIZE 64
189 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
191 struct osd_inconsistent_item {
192 /* link into lustre_scrub::os_inconsistent_items,
193 * protected by lustre_scrub::os_lock. */
194 struct list_head oii_list;
196 /* The right FID <=> ino#/gen mapping. */
197 struct osd_idmap_cache oii_cache;
199 unsigned int oii_insert:1; /* insert or update mapping. */
202 struct osd_otable_cache {
203 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
205 /* Index for next cache slot to be filled. */
206 int ooc_producer_idx;
208 /* Index for next cache slot to be returned by it::next(). */
209 int ooc_consumer_idx;
211 /* How many items in ooc_cache. */
212 __u64 ooc_cached_items;
214 /* Position for up layer LFSCK iteration pre-loading. */
215 __u64 ooc_pos_preload;
218 struct osd_otable_it {
219 struct osd_device *ooi_dev;
220 struct osd_otable_cache ooi_cache;
221 struct osd_iit_param ooi_iit_param;
223 /* The following bits can be updated/checked w/o lock protection.
224 * If more bits will be introduced in the future and need lock to
225 * protect, please add comment. */
226 unsigned long ooi_used_outside:1, /* Some user out of OSD
227 * uses the iteration. */
228 ooi_all_cached:1, /* No more entries can be
229 * filled into cache. */
230 ooi_user_ready:1, /* The user out of OSD is
231 * ready to iterate. */
232 ooi_waiting:1; /* it::next is waiting. */
235 struct osd_obj_orphan {
236 struct list_head oor_list;
237 struct lu_env *oor_env; /* to identify "own" records */
246 struct dt_device od_dt_dev;
247 /* information about underlying file system */
248 struct vfsmount *od_mnt;
250 struct osd_oi **od_oi_table;
251 /* total number of OI containers */
256 unsigned int od_fl_capa:1,
263 __s64 od_auto_scrub_interval;
264 __u32 od_dirent_journal;
266 struct proc_dir_entry *od_proc_entry;
267 struct lprocfs_stats *od_stats;
269 spinlock_t od_osfs_lock;
272 struct lu_site od_site;
274 struct osd_obj_map *od_ost_map;
275 struct osd_mdobj_map *od_mdt_map;
277 unsigned long long od_readcache_max_filesize;
279 int od_writethrough_cache;
281 struct brw_stats od_brw_stats;
282 atomic_t od_r_in_flight;
283 atomic_t od_w_in_flight;
285 struct mutex od_otable_mutex;
286 struct osd_otable_it *od_otable_it;
287 struct osd_scrub od_scrub;
288 struct list_head od_ios_list;
290 /* service name associated with the osd device */
291 char od_svname[MAX_OBD_NAME];
292 char od_mntdev[MAX_OBD_NAME];
294 /* quota slave instance */
295 struct qsd_instance *od_quota_slave;
297 /* osd seq instance */
298 struct lu_client_seq *od_cl_seq;
299 /* If the ratio of "the total OI mappings count" vs
300 * "the bad OI mappings count" is lower than the
301 * osd_device::od_full_scrub_ratio, then trigger
302 * OI scrub to scan the whole the device. */
303 __u64 od_full_scrub_ratio;
304 /* If the speed of found bad OI mappings (per minute)
305 * exceeds the osd_device::od_full_scrub_threshold_rate,
306 * then trigger OI scrub to scan the whole device. */
307 __u64 od_full_scrub_threshold_rate;
309 /* a list of orphaned agent inodes, protected with od_osfs_lock */
310 struct list_head od_orphan_list;
313 enum osd_full_scrub_ratio {
314 /* Trigger OI scrub to scan the whole device directly. */
317 /* Because the bad OI mappings count cannot be larger than
318 * the total OI mappints count, then setting OFSR_NEVER means
319 * that the whole device scanning cannot be triggered by auto
320 * detected bad OI mappings during the RPC services. */
322 OFSR_DEFAULT = 10000,
325 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
327 /* There are at most 15 uid/gid/projids are affected in a transaction, and
328 * that's rename case:
329 * - 3 for source parent uid & gid & projid;
330 * - 3 for source child uid & gid & projid ('..' entry update when
331 * child is directory);
332 * - 3 for target parent uid & gid & projid;
333 * - 3 for target child uid & gid & projid(if the target child exists);
334 * - 3 for root uid & gid(last_rcvd, llog, etc);
337 #define OSD_MAX_UGID_CNT 15
342 OSD_OT_XATTR_SET = 2,
355 struct thandle ot_super;
357 struct ldiskfs_journal_cb_entry ot_jcb;
358 struct list_head ot_commit_dcb_list;
359 struct list_head ot_stop_dcb_list;
360 /* Link to the device, for debugging. */
361 struct lu_ref_link ot_dev_link;
362 unsigned int ot_credits;
363 unsigned short ot_id_cnt;
364 __u8 ot_id_types[OSD_MAX_UGID_CNT];
365 unsigned int ot_remove_agents:1;
366 uid_t ot_id_array[OSD_MAX_UGID_CNT];
367 struct lquota_trans *ot_quota_trans;
368 #if OSD_THANDLE_STATS
369 /** time when this handle was allocated */
372 /** time when this thanle was started */
378 * Basic transaction credit op
399 #ifdef CONFIG_PROC_FS
401 LPROC_OSD_READ_BYTES = 0,
402 LPROC_OSD_WRITE_BYTES = 1,
403 LPROC_OSD_GET_PAGE = 2,
404 LPROC_OSD_NO_PAGE = 3,
405 LPROC_OSD_CACHE_ACCESS = 4,
406 LPROC_OSD_CACHE_HIT = 5,
407 LPROC_OSD_CACHE_MISS = 6,
409 #if OSD_THANDLE_STATS
410 LPROC_OSD_THANDLE_STARTING,
411 LPROC_OSD_THANDLE_OPEN,
412 LPROC_OSD_THANDLE_CLOSING,
419 * Storage representation for fids.
421 * Variable size, first byte contains the length of the whole record.
423 struct osd_fid_pack {
424 unsigned char fp_len;
425 char fp_area[sizeof(struct lu_fid)];
428 struct osd_it_ea_dirent {
429 struct lu_fid oied_fid;
432 unsigned short oied_namelen;
433 unsigned int oied_type;
435 } __attribute__((packed));
438 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
439 * than lu_dirent struct. osd readdir reads less number of dirent than
440 * required for mdd dir page. so buffer size need to be increased so that
441 * there would be one ext3 readdir for every mdd readdir page.
444 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
447 * This is iterator's in-memory data structure in interoperability
448 * mode (i.e. iterator over ldiskfs style directory)
451 struct osd_object *oie_obj;
452 /** used in ldiskfs iterator, to stored file pointer */
453 struct file oie_file;
454 /** how many entries have been read-cached from storage */
456 /** current entry is being iterated by caller */
458 /** current processing entry */
459 struct osd_it_ea_dirent *oie_dirent;
460 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
462 struct dentry oie_dentry;
466 * Iterator's in-memory data structure for IAM mode.
469 struct osd_object *oi_obj;
470 struct iam_path_descr *oi_ipd;
471 struct iam_iterator oi_it;
474 struct osd_quota_leaf {
475 struct list_head oql_link;
480 * Iterator's in-memory data structure for quota file.
482 struct osd_it_quota {
483 struct osd_object *oiq_obj;
484 /** tree blocks path to where the entry is stored */
485 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
486 /** on-disk offset for current key where quota record can be found */
488 /** identifier for current quota record */
490 /** the record index in the leaf/index block */
491 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
492 /** list of already processed leaf blocks */
493 struct list_head oiq_list;
496 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
499 wait_queue_head_t dr_wait;
500 atomic_t dr_numreqs; /* number of reqs being processed */
505 unsigned int dr_ignore_quota:1;
506 unsigned int dr_elapsed_valid:1; /* we really did count time */
507 unsigned int dr_rw:1;
508 struct lu_buf dr_pg_buf;
509 struct page **dr_pages;
510 struct lu_buf dr_bl_buf;
512 unsigned long dr_start_time;
513 unsigned long dr_elapsed; /* how long io took */
514 struct osd_device *dr_dev;
515 unsigned int dr_init_at; /* the line iobuf was initialized */
518 #define OSD_INS_CACHE_SIZE 8
520 struct osd_thread_info {
521 const struct lu_env *oti_env;
523 * used for index operations.
525 struct dentry oti_obj_dentry;
526 struct dentry oti_child_dentry;
528 /** dentry for Iterator context. */
529 struct dentry oti_it_dentry;
532 /* fake struct file for osd_object_sync */
533 struct file oti_file;
535 struct kstatfs oti_ksfs;
538 struct htree_lock *oti_hlock;
540 struct lu_fid oti_fid;
541 struct lu_fid oti_fid2;
542 struct lu_fid oti_fid3;
543 struct osd_inode_id oti_id;
544 struct osd_inode_id oti_id2;
545 struct osd_inode_id oti_id3;
546 struct ost_id oti_ostid;
549 * XXX temporary: for ->i_op calls.
551 struct timespec oti_time;
554 * following ipd and it structures are used for osd_index_iam_lookup()
555 * these are defined separately as we might do index operation
556 * in open iterator session.
559 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
561 unsigned int oti_it_ea_buf_used:1;
563 /* IAM iterator for index operation. */
564 struct iam_iterator oti_idx_it;
566 /** union to guarantee that ->oti_ipd[] has proper alignment. */
569 char oti_it_ipd[DX_IPD_MAX_SIZE];
570 long long oti_alignment_lieutenant;
574 char oti_idx_ipd[DX_IPD_MAX_SIZE];
575 long long oti_alignment_lieutenant_colonel;
578 struct osd_idmap_cache oti_cache;
580 /* dedicated OI cache for insert (which needs inum) */
581 struct osd_idmap_cache *oti_ins_cache;
582 int oti_ins_cache_size;
583 int oti_ins_cache_used;
588 /** used in osd_fid_set() to put xattr */
589 struct lu_buf oti_buf;
590 struct lu_buf oti_big_buf;
591 /** used in osd_ea_fid_set() to set fid into common ea */
593 struct lustre_ost_attrs oti_ost_attrs;
594 struct filter_fid_old oti_ff;
595 struct filter_fid oti_ff_new;
598 struct osd_iobuf oti_iobuf;
599 /* used to access objects in /O */
600 struct inode *oti_inode;
601 #define OSD_FID_REC_SZ 32
602 char oti_ldp[OSD_FID_REC_SZ];
603 char oti_ldp2[OSD_FID_REC_SZ];
605 /* used by quota code */
607 #if defined(HAVE_DQUOT_QC_DQBLK)
608 struct qc_dqblk oti_qdq;
609 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
610 struct fs_disk_quota oti_fdq;
612 struct if_dqblk oti_dqblk;
614 struct if_dqinfo oti_dqinfo;
616 struct lquota_id_info oti_qi;
617 struct lquota_trans oti_quota_trans;
618 union lquota_rec oti_quota_rec;
620 struct lu_seq_range oti_seq_range;
622 /* Tracking for transaction credits, to allow debugging and optimizing
623 * cases where a large number of credits are being allocated for
624 * single transaction. */
625 unsigned int oti_credits_before;
626 unsigned int oti_declare_ops[OSD_OT_MAX];
627 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
628 unsigned int oti_declare_ops_used[OSD_OT_MAX];
631 extern int ldiskfs_pdo;
633 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
634 const char *name, void *buf, int len)
639 dentry->d_inode = inode;
640 dentry->d_sb = inode->i_sb;
641 return inode->i_op->getxattr(dentry, name, buf, len);
644 static inline int __osd_xattr_set(struct osd_thread_info *info,
645 struct inode *inode, const char *name,
646 const void *buf, int buflen, int fl)
648 struct dentry *dentry = &info->oti_child_dentry;
650 ll_vfs_dq_init(inode);
651 dentry->d_inode = inode;
652 dentry->d_sb = inode->i_sb;
653 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
656 #ifdef CONFIG_PROC_FS
658 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
659 extern struct lprocfs_vars lprocfs_osd_module_vars[];
660 int osd_procfs_init(struct osd_device *osd, const char *name);
661 int osd_procfs_fini(struct osd_device *osd);
662 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
663 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
664 int osd_register_proc_index_in_idif(struct osd_device *osd);
668 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
669 struct obd_statfs *sfs);
670 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
671 struct osd_inode_id *id);
672 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
673 const struct lu_fid *fid, __u32 compat, __u32 incompat);
674 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
675 struct dentry *dentry, struct lustre_ost_attrs *loa);
676 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
677 struct osd_inode_id *id, const struct lu_fid *fid);
678 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
679 struct dentry *dentry, struct lu_fid *fid);
681 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
682 void osd_obj_map_fini(struct osd_device *dev);
683 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
684 const struct lu_fid *fid, struct osd_inode_id *id);
685 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
686 const struct lu_fid *fid, const struct osd_inode_id *id,
688 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
689 const struct lu_fid *fid, handle_t *th);
690 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
691 const struct lu_fid *fid, const struct osd_inode_id *id,
693 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
694 struct inode *src_parent, struct dentry *src_child,
695 const struct lu_fid *fid);
696 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
697 const struct lu_fid *fid, struct osd_inode_id *id);
698 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
699 const struct lu_fid *fid, const struct osd_inode_id *id,
701 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
702 const struct lu_fid *fid, const struct osd_inode_id *id,
705 char *osd_lf_fid2name(const struct lu_fid *fid);
706 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
708 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
709 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
710 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
712 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
713 struct osd_inode_id *id);
714 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
716 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
717 u64 seq, struct lu_seq_range *range);
719 int osd_delete_from_remote_parent(const struct lu_env *env,
720 struct osd_device *osd,
721 struct osd_object *obj,
722 struct osd_thandle *oh);
723 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
724 struct osd_object *obj, struct osd_thandle *oh);
725 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
726 struct osd_device *osd,
727 const struct lu_fid *fid,
728 struct osd_inode_id *id);
730 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
732 /* osd_quota_fmt.c */
733 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
734 int type, uint blk, int depth, uint index,
735 struct osd_it_quota *it);
736 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
737 int type, uint blk, uint index,
738 struct osd_it_quota *it);
739 loff_t find_tree_dqentry(const struct lu_env *env,
740 struct osd_object *obj, int type,
741 qid_t dqid, uint blk, int depth,
742 struct osd_it_quota *it);
744 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
745 struct lquota_id_info *qi, struct osd_object *obj,
746 bool enforce, int *flags);
747 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
748 __u32 projid, long long space, struct osd_thandle *oh,
749 struct osd_object *obj, int *flags,
750 enum osd_qid_declare_flags);
751 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
752 const struct dt_rec *rec,
753 union lquota_rec *quota_rec);
754 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
756 #ifndef HAVE_I_UID_READ
757 static inline uid_t i_uid_read(const struct inode *inode)
762 static inline gid_t i_gid_read(const struct inode *inode)
767 static inline void i_uid_write(struct inode *inode, uid_t uid)
772 static inline void i_gid_write(struct inode *inode, gid_t gid)
778 #ifdef HAVE_PROJECT_QUOTA
779 static inline __u32 i_projid_read(struct inode *inode)
781 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
784 static inline void i_projid_write(struct inode *inode, __u32 projid)
787 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
788 LDISKFS_I(inode)->i_projid = kprojid;
791 static inline uid_t i_projid_read(struct inode *inode)
795 static inline void i_projid_write(struct inode *inode, __u32 projid)
801 #ifdef HAVE_LDISKFS_INFO_JINODE
802 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
803 #else /* HAVE_LDISKFS_INFO_JINODE */
804 # define osd_attach_jinode(inode) 0
805 #endif /* HAVE_LDISKFS_INFO_JINODE */
807 #ifdef LDISKFS_HT_MISC
808 # define osd_journal_start_sb(sb, type, nblock) \
809 ldiskfs_journal_start_sb(sb, type, nblock)
810 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
812 ldiskfs_lblk_t *nblock)
816 rc = osd_attach_jinode(inode);
819 return ldiskfs_append(handle, inode, nblock);
821 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
822 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
824 # define osd_journal_start(inode, type, nblocks) \
825 ldiskfs_journal_start(inode, type, nblocks)
826 # define osd_transaction_size(dev) \
827 (osd_journal(dev)->j_max_transaction_buffers / 2)
829 # define LDISKFS_HT_MISC 0
830 # define osd_journal_start_sb(sb, type, nblock) \
831 ldiskfs_journal_start_sb(sb, nblock)
833 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
835 ldiskfs_lblk_t *nblock)
837 struct buffer_head *bh;
840 bh = ldiskfs_append(handle, inode, nblock, &err);
847 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
848 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
850 # define osd_journal_start(inode, type, nblocks) \
851 ldiskfs_journal_start(inode, nblocks)
852 # define osd_transaction_size(dev) \
853 (osd_journal(dev)->j_max_transaction_buffers)
857 * Invariants, assertions.
861 * XXX: do not enable this, until invariant checking code is made thread safe
862 * in the face of pdirops locking.
864 #define OSD_INVARIANT_CHECKS (0)
866 #if OSD_INVARIANT_CHECKS
867 static inline int osd_invariant(const struct osd_object *obj)
871 ergo(obj->oo_inode != NULL,
872 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
873 atomic_read(&obj->oo_inode->i_count) > 0) &&
874 ergo(obj->oo_dir != NULL &&
875 obj->oo_dir->od_conationer.ic_object != NULL,
876 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
879 #define osd_invariant(obj) (1)
882 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
884 extern const struct dt_index_operations osd_otable_ops;
886 static inline int osd_oi_fid2idx(struct osd_device *dev,
887 const struct lu_fid *fid)
889 return fid->f_seq & (dev->od_oi_count - 1);
892 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
893 const struct lu_fid *fid)
895 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
896 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
897 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
898 DFID"\n", PFID(fid));
899 /* It can work even od_oi_count equals to 1 although it's unexpected,
900 * the only reason we set it to 1 is for performance measurement */
901 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
904 extern const struct lu_device_operations osd_lu_ops;
906 static inline int lu_device_is_osd(const struct lu_device *d)
908 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
911 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
913 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
914 return container_of0(d, struct osd_device, od_dt_dev);
917 static inline struct osd_device *osd_dev(const struct lu_device *d)
919 LASSERT(lu_device_is_osd(d));
920 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
923 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
925 return osd_dev(o->oo_dt.do_lu.lo_dev);
928 static inline struct super_block *osd_sb(const struct osd_device *dev)
930 return dev->od_mnt->mnt_sb;
933 static inline const char *osd_dev2name(const struct osd_device *dev)
935 return osd_sb(dev)->s_id;
938 static inline const char *osd_ino2name(const struct inode *inode)
940 return inode->i_sb->s_id;
944 * Put the osd object once done with it.
946 * \param obj osd object that needs to be put
948 static inline void osd_object_put(const struct lu_env *env,
949 struct osd_object *obj)
951 dt_object_put(env, &obj->oo_dt);
954 static inline int osd_object_is_root(const struct osd_object *obj)
956 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
959 static inline struct osd_object *osd_obj(const struct lu_object *o)
961 LASSERT(lu_device_is_osd(o->lo_dev));
962 return container_of0(o, struct osd_object, oo_dt.do_lu);
965 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
967 return osd_obj(&d->do_lu);
970 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
972 return &osd->od_dt_dev.dd_lu_dev;
975 static inline journal_t *osd_journal(const struct osd_device *dev)
977 return LDISKFS_SB(osd_sb(dev))->s_journal;
980 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
982 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
985 static inline char *osd_name(struct osd_device *osd)
987 return osd->od_svname;
990 static inline bool osd_is_ea_inode(struct inode *inode)
992 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
995 extern const struct dt_body_operations osd_body_ops;
996 extern struct lu_context_key osd_key;
998 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1000 return lu_context_key_get(&env->le_ctx, &osd_key);
1003 extern const struct dt_body_operations osd_body_ops_new;
1009 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1010 const struct iam_container *bag)
1012 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1013 osd_oti_get(env)->oti_it_ipd);
1017 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1018 const struct iam_container *bag)
1020 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1021 osd_oti_get(env)->oti_idx_ipd);
1024 static inline void osd_ipd_put(const struct lu_env *env,
1025 const struct iam_container *bag,
1026 struct iam_path_descr *ipd)
1028 bag->ic_descr->id_ops->id_ipd_free(ipd);
1031 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1032 const loff_t size, const loff_t pos,
1035 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1036 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1037 int write_NUL, loff_t *offs, handle_t *handle);
1040 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1041 struct inode *inode,
1042 const char *name, const int namelen)
1044 struct osd_thread_info *info = osd_oti_get(env);
1045 struct dentry *child_dentry = &info->oti_child_dentry;
1046 struct dentry *obj_dentry = &info->oti_obj_dentry;
1048 obj_dentry->d_inode = inode;
1049 obj_dentry->d_sb = inode->i_sb;
1050 obj_dentry->d_name.hash = 0;
1052 child_dentry->d_name.hash = 0;
1053 child_dentry->d_parent = obj_dentry;
1054 child_dentry->d_name.name = name;
1055 child_dentry->d_name.len = namelen;
1056 return child_dentry;
1059 extern int osd_trans_declare_op2rb[];
1060 extern int ldiskfs_track_declares_assert;
1061 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1063 static inline void osd_trans_declare_op(const struct lu_env *env,
1064 struct osd_thandle *oh,
1065 enum osd_op_type op, int credits)
1067 struct osd_thread_info *oti = osd_oti_get(env);
1069 LASSERT(oh->ot_handle == NULL);
1070 if (unlikely(op >= OSD_OT_MAX)) {
1071 if (unlikely(ldiskfs_track_declares_assert)) {
1072 LASSERT(op < OSD_OT_MAX);
1074 CWARN("%s: Invalid operation index %d\n",
1075 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1076 libcfs_debug_dumpstack(NULL);
1079 oti->oti_declare_ops[op]++;
1080 oti->oti_declare_ops_cred[op] += credits;
1082 oh->ot_credits += credits;
1085 static inline void osd_trans_exec_op(const struct lu_env *env,
1087 enum osd_op_type op)
1089 struct osd_thread_info *oti = osd_oti_get(env);
1090 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1092 unsigned int rb, left;
1094 LASSERT(oh->ot_handle != NULL);
1095 if (unlikely(op >= OSD_OT_MAX)) {
1096 if (unlikely(ldiskfs_track_declares_assert))
1097 LASSERT(op < OSD_OT_MAX);
1099 CWARN("%s: opcode %u: invalid value >= %u\n",
1100 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1102 libcfs_debug_dumpstack(NULL);
1107 /* find rollback (or reverse) operation for the given one
1108 * such an operation doesn't require additional credits
1109 * as the same set of blocks are modified */
1110 rb = osd_trans_declare_op2rb[op];
1112 /* check whether credits for this operation were reserved at all */
1113 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1114 oti->oti_declare_ops_cred[rb] == 0)) {
1115 /* the API is not perfect yet: CREATE does REF_ADD internally
1116 * while DESTROY does not. To rollback CREATE the callers
1117 * needs to call REF_DEL+DESTROY which is hard to detect using
1118 * a simple table of rollback operations */
1119 if (op == OSD_OT_REF_DEL &&
1120 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1122 if (op == OSD_OT_REF_ADD &&
1123 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1125 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1126 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1127 osd_trans_dump_creds(env, th);
1128 LASSERT(!ldiskfs_track_declares_assert);
1132 /* remember how many credits we have unused before the operation */
1133 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1134 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1135 if (unlikely(oti->oti_credits_before < left)) {
1136 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1137 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1138 oti->oti_credits_before, left, rb);
1139 osd_trans_dump_creds(env, th);
1140 /* on a very small fs (testing?) it's possible that
1141 * the transaction can't fit 1/4 of journal, so we
1142 * just request less credits (see osd_trans_start()).
1143 * ignore the same case here */
1144 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1145 if (unlikely(oh->ot_credits < rb))
1146 LASSERT(!ldiskfs_track_declares_assert);
1150 static inline void osd_trans_exec_check(const struct lu_env *env,
1152 enum osd_op_type op)
1154 struct osd_thread_info *oti = osd_oti_get(env);
1155 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1157 int used, over, quota;
1159 /* how many credits have been used by the operation */
1160 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1162 if (unlikely(used < 0)) {
1163 /* if some block was allocated and released in the same
1164 * transaction, then it won't be a part of the transaction
1165 * and delta can be negative */
1170 /* rollback operations (e.g. when we destroy just created
1171 * object) should not consume any credits. there is no point
1172 * to confuse the checks below */
1176 oti->oti_declare_ops_used[op] += used;
1177 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1180 /* we account quota for a whole transaction and any operation can
1181 * consume corresponding credits */
1182 over = oti->oti_declare_ops_used[op] -
1183 oti->oti_declare_ops_cred[op];
1184 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1185 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1186 if (over <= quota) {
1187 /* probably that credits were consumed by
1188 * quota indirectly (in the depths of ldiskfs) */
1189 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1190 oti->oti_declare_ops_used[op] -= over;
1192 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1193 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1194 oti->oti_declare_ops_used[op], used,
1195 oti->oti_declare_ops_cred[op]);
1196 osd_trans_dump_creds(env, th);
1197 if (unlikely(ldiskfs_track_declares_assert))
1203 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1206 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1207 struct lu_fid *befider)
1209 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1210 memcpy(pack->fp_area, befider, sizeof(*befider));
1211 pack->fp_len = sizeof(*befider) + 1;
1215 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1220 switch (pack->fp_len) {
1221 case sizeof *fid + 1:
1222 memcpy(fid, pack->fp_area, sizeof *fid);
1223 fid_be_to_cpu(fid, fid);
1226 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1233 * Quota/Accounting handling
1235 extern const struct dt_index_operations osd_acct_index_ops;
1236 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1237 const struct lu_fid *fid, struct osd_inode_id *id);
1239 /* copy from fs/ext4/dir.c */
1240 static inline int is_32bit_api(void)
1242 #ifdef CONFIG_COMPAT
1243 return in_compat_syscall();
1245 return (BITS_PER_LONG == 32);
1249 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1251 if ((filp->f_mode & FMODE_32BITHASH) ||
1252 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1253 return LDISKFS_HTREE_EOF_32BIT;
1255 return LDISKFS_HTREE_EOF_64BIT;
1258 static inline int fid_is_internal(const struct lu_fid *fid)
1260 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1263 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1265 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1269 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1270 * return code has been removed and integrated into the pointer in the
1273 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1274 struct inode *inode,
1275 ldiskfs_lblk_t block,
1279 struct buffer_head *bh;
1282 rc = osd_attach_jinode(inode);
1286 #ifdef HAVE_EXT4_BREAD_4ARGS
1287 bh = ldiskfs_bread(handle, inode, block, create);
1290 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1291 if (bh == NULL && rc != 0)
1297 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1298 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1300 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1303 #endif /* _OSD_INTERNAL_H */