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, 2017, 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,
141 /* the i_flags in LMA */
144 * Following two members are used to indicate the presence of dot and
145 * dotdot in the given directory. This is required for interop mode
148 int oo_compat_dot_created;
149 int oo_compat_dotdot_created;
151 const struct lu_env *oo_owner;
152 #ifdef CONFIG_LOCKDEP
153 struct lockdep_map oo_dep_map;
156 struct list_head oo_xattr_list;
160 /* protects on-fly initialization */
161 int oos_subdir_count; /* subdir count for each seq */
162 struct dentry *oos_root; /* O/<seq> */
163 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
164 u64 oos_seq; /* seq number */
165 struct list_head oos_seq_list; /* list to seq_list */
169 struct dentry *om_root; /* dentry for /O */
170 rwlock_t om_seq_list_lock; /* lock for seq_list */
171 struct list_head om_seq_list; /* list head for seq */
173 struct mutex om_dir_init_mutex;
177 struct dentry *om_root; /* AGENT/<index> */
178 u64 om_index; /* mdt index */
179 struct list_head om_list; /* list to omm_list */
182 struct osd_mdobj_map {
183 struct dentry *omm_remote_parent;
185 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
186 handle_t *handle, struct dentry *child,
187 struct inode *inode, struct htree_lock *hlock);
189 #define OSD_OTABLE_IT_CACHE_SIZE 64
190 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
192 struct osd_inconsistent_item {
193 /* link into lustre_scrub::os_inconsistent_items,
194 * protected by lustre_scrub::os_lock. */
195 struct list_head oii_list;
197 /* The right FID <=> ino#/gen mapping. */
198 struct osd_idmap_cache oii_cache;
200 unsigned int oii_insert:1; /* insert or update mapping. */
203 struct osd_otable_cache {
204 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
206 /* Index for next cache slot to be filled. */
207 int ooc_producer_idx;
209 /* Index for next cache slot to be returned by it::next(). */
210 int ooc_consumer_idx;
212 /* How many items in ooc_cache. */
213 __u64 ooc_cached_items;
215 /* Position for up layer LFSCK iteration pre-loading. */
216 __u64 ooc_pos_preload;
219 struct osd_otable_it {
220 struct osd_device *ooi_dev;
221 struct osd_otable_cache ooi_cache;
222 struct osd_iit_param ooi_iit_param;
224 /* The following bits can be updated/checked w/o lock protection.
225 * If more bits will be introduced in the future and need lock to
226 * protect, please add comment. */
227 unsigned long ooi_used_outside:1, /* Some user out of OSD
228 * uses the iteration. */
229 ooi_all_cached:1, /* No more entries can be
230 * filled into cache. */
231 ooi_user_ready:1, /* The user out of OSD is
232 * ready to iterate. */
233 ooi_waiting:1; /* it::next is waiting. */
236 struct osd_obj_orphan {
237 struct list_head oor_list;
238 struct lu_env *oor_env; /* to identify "own" records */
247 struct dt_device od_dt_dev;
248 /* information about underlying file system */
249 struct vfsmount *od_mnt;
251 struct osd_oi **od_oi_table;
252 /* total number of OI containers */
257 unsigned int od_fl_capa:1,
264 __s64 od_auto_scrub_interval;
265 __u32 od_dirent_journal;
267 struct proc_dir_entry *od_proc_entry;
268 struct lprocfs_stats *od_stats;
270 spinlock_t od_osfs_lock;
273 struct lu_site od_site;
275 struct osd_obj_map *od_ost_map;
276 struct osd_mdobj_map *od_mdt_map;
278 unsigned long long od_readcache_max_filesize;
280 int od_writethrough_cache;
282 struct brw_stats od_brw_stats;
283 atomic_t od_r_in_flight;
284 atomic_t od_w_in_flight;
286 struct mutex od_otable_mutex;
287 struct osd_otable_it *od_otable_it;
288 struct osd_scrub od_scrub;
289 struct list_head od_ios_list;
291 /* service name associated with the osd device */
292 char od_svname[MAX_OBD_NAME];
293 char od_mntdev[MAX_OBD_NAME];
295 /* quota slave instance */
296 struct qsd_instance *od_quota_slave;
298 /* osd seq instance */
299 struct lu_client_seq *od_cl_seq;
300 /* If the ratio of "the total OI mappings count" vs
301 * "the bad OI mappings count" is lower than the
302 * osd_device::od_full_scrub_ratio, then trigger
303 * OI scrub to scan the whole the device. */
304 __u64 od_full_scrub_ratio;
305 /* If the speed of found bad OI mappings (per minute)
306 * exceeds the osd_device::od_full_scrub_threshold_rate,
307 * then trigger OI scrub to scan the whole device. */
308 __u64 od_full_scrub_threshold_rate;
310 /* a list of orphaned agent inodes, protected with od_osfs_lock */
311 struct list_head od_orphan_list;
314 enum osd_full_scrub_ratio {
315 /* Trigger OI scrub to scan the whole device directly. */
318 /* Because the bad OI mappings count cannot be larger than
319 * the total OI mappints count, then setting OFSR_NEVER means
320 * that the whole device scanning cannot be triggered by auto
321 * detected bad OI mappings during the RPC services. */
323 OFSR_DEFAULT = 10000,
326 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
328 /* There are at most 15 uid/gid/projids are affected in a transaction, and
329 * that's rename case:
330 * - 3 for source parent uid & gid & projid;
331 * - 3 for source child uid & gid & projid ('..' entry update when
332 * child is directory);
333 * - 3 for target parent uid & gid & projid;
334 * - 3 for target child uid & gid & projid(if the target child exists);
335 * - 3 for root uid & gid(last_rcvd, llog, etc);
338 #define OSD_MAX_UGID_CNT 15
343 OSD_OT_XATTR_SET = 2,
356 struct thandle ot_super;
358 struct ldiskfs_journal_cb_entry ot_jcb;
359 struct list_head ot_commit_dcb_list;
360 struct list_head ot_stop_dcb_list;
361 /* Link to the device, for debugging. */
362 struct lu_ref_link ot_dev_link;
363 unsigned int ot_credits;
364 unsigned short ot_id_cnt;
365 __u8 ot_id_types[OSD_MAX_UGID_CNT];
366 unsigned int ot_remove_agents:1;
367 uid_t ot_id_array[OSD_MAX_UGID_CNT];
368 struct lquota_trans *ot_quota_trans;
369 #if OSD_THANDLE_STATS
370 /** time when this handle was allocated */
373 /** time when this thanle was started */
379 * Basic transaction credit op
400 #ifdef CONFIG_PROC_FS
402 LPROC_OSD_READ_BYTES = 0,
403 LPROC_OSD_WRITE_BYTES = 1,
404 LPROC_OSD_GET_PAGE = 2,
405 LPROC_OSD_NO_PAGE = 3,
406 LPROC_OSD_CACHE_ACCESS = 4,
407 LPROC_OSD_CACHE_HIT = 5,
408 LPROC_OSD_CACHE_MISS = 6,
410 #if OSD_THANDLE_STATS
411 LPROC_OSD_THANDLE_STARTING,
412 LPROC_OSD_THANDLE_OPEN,
413 LPROC_OSD_THANDLE_CLOSING,
420 * Storage representation for fids.
422 * Variable size, first byte contains the length of the whole record.
424 struct osd_fid_pack {
425 unsigned char fp_len;
426 char fp_area[sizeof(struct lu_fid)];
429 struct osd_it_ea_dirent {
430 struct lu_fid oied_fid;
433 unsigned short oied_namelen;
434 unsigned int oied_type;
436 } __attribute__((packed));
439 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
440 * than lu_dirent struct. osd readdir reads less number of dirent than
441 * required for mdd dir page. so buffer size need to be increased so that
442 * there would be one ext3 readdir for every mdd readdir page.
445 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
448 * This is iterator's in-memory data structure in interoperability
449 * mode (i.e. iterator over ldiskfs style directory)
452 struct osd_object *oie_obj;
453 /** used in ldiskfs iterator, to stored file pointer */
454 struct file oie_file;
455 /** how many entries have been read-cached from storage */
457 /** current entry is being iterated by caller */
459 /** current processing entry */
460 struct osd_it_ea_dirent *oie_dirent;
461 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
463 struct dentry oie_dentry;
467 * Iterator's in-memory data structure for IAM mode.
470 struct osd_object *oi_obj;
471 struct iam_path_descr *oi_ipd;
472 struct iam_iterator oi_it;
475 struct osd_quota_leaf {
476 struct list_head oql_link;
481 * Iterator's in-memory data structure for quota file.
483 struct osd_it_quota {
484 struct osd_object *oiq_obj;
485 /** tree blocks path to where the entry is stored */
486 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
487 /** on-disk offset for current key where quota record can be found */
489 /** identifier for current quota record */
491 /** the record index in the leaf/index block */
492 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
493 /** list of already processed leaf blocks */
494 struct list_head oiq_list;
497 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
500 wait_queue_head_t dr_wait;
501 atomic_t dr_numreqs; /* number of reqs being processed */
506 unsigned int dr_ignore_quota:1;
507 unsigned int dr_elapsed_valid:1; /* we really did count time */
508 unsigned int dr_rw:1;
509 struct lu_buf dr_pg_buf;
510 struct page **dr_pages;
511 struct lu_buf dr_bl_buf;
513 unsigned long dr_start_time;
514 unsigned long dr_elapsed; /* how long io took */
515 struct osd_device *dr_dev;
516 unsigned int dr_init_at; /* the line iobuf was initialized */
519 #define OSD_INS_CACHE_SIZE 8
521 struct osd_thread_info {
522 const struct lu_env *oti_env;
524 * used for index operations.
526 struct dentry oti_obj_dentry;
527 struct dentry oti_child_dentry;
529 /** dentry for Iterator context. */
530 struct dentry oti_it_dentry;
533 /* fake struct file for osd_object_sync */
534 struct file oti_file;
536 struct kstatfs oti_ksfs;
539 struct htree_lock *oti_hlock;
541 struct lu_fid oti_fid;
542 struct lu_fid oti_fid2;
543 struct lu_fid oti_fid3;
544 struct osd_inode_id oti_id;
545 struct osd_inode_id oti_id2;
546 struct osd_inode_id oti_id3;
547 struct ost_id oti_ostid;
550 * XXX temporary: for ->i_op calls.
552 struct timespec oti_time;
555 * following ipd and it structures are used for osd_index_iam_lookup()
556 * these are defined separately as we might do index operation
557 * in open iterator session.
560 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
562 unsigned int oti_it_ea_buf_used:1;
564 /* IAM iterator for index operation. */
565 struct iam_iterator oti_idx_it;
567 /** union to guarantee that ->oti_ipd[] has proper alignment. */
570 char oti_it_ipd[DX_IPD_MAX_SIZE];
571 long long oti_alignment_lieutenant;
575 char oti_idx_ipd[DX_IPD_MAX_SIZE];
576 long long oti_alignment_lieutenant_colonel;
579 struct osd_idmap_cache oti_cache;
581 /* dedicated OI cache for insert (which needs inum) */
582 struct osd_idmap_cache *oti_ins_cache;
583 int oti_ins_cache_size;
584 int oti_ins_cache_used;
589 /** used in osd_fid_set() to put xattr */
590 struct lu_buf oti_buf;
591 struct lu_buf oti_big_buf;
592 /** used in osd_ea_fid_set() to set fid into common ea */
594 struct lustre_ost_attrs oti_ost_attrs;
595 struct filter_fid_old oti_ff;
596 struct filter_fid oti_ff_new;
599 struct osd_iobuf oti_iobuf;
600 /* used to access objects in /O */
601 struct inode *oti_inode;
602 #define OSD_FID_REC_SZ 32
603 char oti_ldp[OSD_FID_REC_SZ];
604 char oti_ldp2[OSD_FID_REC_SZ];
606 /* used by quota code */
608 #if defined(HAVE_DQUOT_QC_DQBLK)
609 struct qc_dqblk oti_qdq;
610 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
611 struct fs_disk_quota oti_fdq;
613 struct if_dqblk oti_dqblk;
615 struct if_dqinfo oti_dqinfo;
617 struct lquota_id_info oti_qi;
618 struct lquota_trans oti_quota_trans;
619 union lquota_rec oti_quota_rec;
621 struct lu_seq_range oti_seq_range;
623 /* Tracking for transaction credits, to allow debugging and optimizing
624 * cases where a large number of credits are being allocated for
625 * single transaction. */
626 unsigned int oti_credits_before;
627 unsigned int oti_declare_ops[OSD_OT_MAX];
628 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
629 unsigned int oti_declare_ops_used[OSD_OT_MAX];
632 extern int ldiskfs_pdo;
634 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
635 const char *name, void *buf, int len)
640 dentry->d_inode = inode;
641 dentry->d_sb = inode->i_sb;
642 return inode->i_op->getxattr(dentry, name, buf, len);
645 static inline int __osd_xattr_set(struct osd_thread_info *info,
646 struct inode *inode, const char *name,
647 const void *buf, int buflen, int fl)
649 struct dentry *dentry = &info->oti_child_dentry;
651 ll_vfs_dq_init(inode);
652 dentry->d_inode = inode;
653 dentry->d_sb = inode->i_sb;
654 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
657 #ifdef CONFIG_PROC_FS
659 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
660 extern struct lprocfs_vars lprocfs_osd_module_vars[];
661 int osd_procfs_init(struct osd_device *osd, const char *name);
662 int osd_procfs_fini(struct osd_device *osd);
663 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
664 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
665 int osd_register_proc_index_in_idif(struct osd_device *osd);
669 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
670 struct obd_statfs *sfs);
671 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
672 struct osd_inode_id *id);
673 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
674 const struct lu_fid *fid, __u32 compat, __u32 incompat);
675 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
676 struct dentry *dentry, struct lustre_ost_attrs *loa);
677 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
678 struct osd_inode_id *id, const struct lu_fid *fid);
679 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
680 struct dentry *dentry, struct lu_fid *fid);
682 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
683 void osd_obj_map_fini(struct osd_device *dev);
684 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
685 const struct lu_fid *fid, struct osd_inode_id *id);
686 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
687 const struct lu_fid *fid, const struct osd_inode_id *id,
689 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
690 const struct lu_fid *fid, handle_t *th);
691 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
692 const struct lu_fid *fid, const struct osd_inode_id *id,
694 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
695 struct inode *src_parent, struct dentry *src_child,
696 const struct lu_fid *fid);
697 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
698 const struct lu_fid *fid, struct osd_inode_id *id);
699 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
700 const struct lu_fid *fid, const struct osd_inode_id *id,
702 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
703 const struct lu_fid *fid, const struct osd_inode_id *id,
706 char *osd_lf_fid2name(const struct lu_fid *fid);
707 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
709 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
710 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
711 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
713 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
714 struct osd_inode_id *id);
715 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
717 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
718 u64 seq, struct lu_seq_range *range);
720 int osd_delete_from_remote_parent(const struct lu_env *env,
721 struct osd_device *osd,
722 struct osd_object *obj,
723 struct osd_thandle *oh, bool destroy);
724 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
725 struct osd_object *obj, struct osd_thandle *oh);
726 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
727 struct osd_device *osd,
728 const struct lu_fid *fid,
729 struct osd_inode_id *id);
731 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
733 /* osd_quota_fmt.c */
734 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
735 int type, uint blk, int depth, uint index,
736 struct osd_it_quota *it);
737 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
738 int type, uint blk, uint index,
739 struct osd_it_quota *it);
740 loff_t find_tree_dqentry(const struct lu_env *env,
741 struct osd_object *obj, int type,
742 qid_t dqid, uint blk, int depth,
743 struct osd_it_quota *it);
745 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
746 struct lquota_id_info *qi, struct osd_object *obj,
747 bool enforce, int *flags);
748 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
749 __u32 projid, long long space, struct osd_thandle *oh,
750 struct osd_object *obj, int *flags,
751 enum osd_qid_declare_flags);
752 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
753 const struct dt_rec *rec,
754 union lquota_rec *quota_rec);
755 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
757 #ifndef HAVE_I_UID_READ
758 static inline uid_t i_uid_read(const struct inode *inode)
763 static inline gid_t i_gid_read(const struct inode *inode)
768 static inline void i_uid_write(struct inode *inode, uid_t uid)
773 static inline void i_gid_write(struct inode *inode, gid_t gid)
779 #ifdef HAVE_PROJECT_QUOTA
780 static inline __u32 i_projid_read(struct inode *inode)
782 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
785 static inline void i_projid_write(struct inode *inode, __u32 projid)
788 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
789 LDISKFS_I(inode)->i_projid = kprojid;
792 static inline uid_t i_projid_read(struct inode *inode)
796 static inline void i_projid_write(struct inode *inode, __u32 projid)
802 #ifdef HAVE_LDISKFS_INFO_JINODE
803 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
804 #else /* HAVE_LDISKFS_INFO_JINODE */
805 # define osd_attach_jinode(inode) 0
806 #endif /* HAVE_LDISKFS_INFO_JINODE */
808 #ifdef LDISKFS_HT_MISC
809 # define osd_journal_start_sb(sb, type, nblock) \
810 ldiskfs_journal_start_sb(sb, type, nblock)
811 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
813 ldiskfs_lblk_t *nblock)
817 rc = osd_attach_jinode(inode);
820 return ldiskfs_append(handle, inode, nblock);
822 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
823 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
825 # define osd_journal_start(inode, type, nblocks) \
826 ldiskfs_journal_start(inode, type, nblocks)
827 # define osd_transaction_size(dev) \
828 (osd_journal(dev)->j_max_transaction_buffers / 2)
830 # define LDISKFS_HT_MISC 0
831 # define osd_journal_start_sb(sb, type, nblock) \
832 ldiskfs_journal_start_sb(sb, nblock)
834 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
836 ldiskfs_lblk_t *nblock)
838 struct buffer_head *bh;
841 bh = ldiskfs_append(handle, inode, nblock, &err);
848 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
849 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
851 # define osd_journal_start(inode, type, nblocks) \
852 ldiskfs_journal_start(inode, nblocks)
853 # define osd_transaction_size(dev) \
854 (osd_journal(dev)->j_max_transaction_buffers)
858 * Invariants, assertions.
862 * XXX: do not enable this, until invariant checking code is made thread safe
863 * in the face of pdirops locking.
865 #define OSD_INVARIANT_CHECKS (0)
867 #if OSD_INVARIANT_CHECKS
868 static inline int osd_invariant(const struct osd_object *obj)
872 ergo(obj->oo_inode != NULL,
873 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
874 atomic_read(&obj->oo_inode->i_count) > 0) &&
875 ergo(obj->oo_dir != NULL &&
876 obj->oo_dir->od_conationer.ic_object != NULL,
877 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
880 #define osd_invariant(obj) (1)
883 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
885 extern const struct dt_index_operations osd_otable_ops;
887 static inline int osd_oi_fid2idx(struct osd_device *dev,
888 const struct lu_fid *fid)
890 return fid->f_seq & (dev->od_oi_count - 1);
893 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
894 const struct lu_fid *fid)
896 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
897 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
898 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
899 DFID"\n", PFID(fid));
900 /* It can work even od_oi_count equals to 1 although it's unexpected,
901 * the only reason we set it to 1 is for performance measurement */
902 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
905 extern const struct lu_device_operations osd_lu_ops;
907 static inline int lu_device_is_osd(const struct lu_device *d)
909 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
912 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
914 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
915 return container_of0(d, struct osd_device, od_dt_dev);
918 static inline struct osd_device *osd_dev(const struct lu_device *d)
920 LASSERT(lu_device_is_osd(d));
921 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
924 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
926 return osd_dev(o->oo_dt.do_lu.lo_dev);
929 static inline struct super_block *osd_sb(const struct osd_device *dev)
931 return dev->od_mnt->mnt_sb;
934 static inline const char *osd_dev2name(const struct osd_device *dev)
936 return osd_sb(dev)->s_id;
939 static inline const char *osd_ino2name(const struct inode *inode)
941 return inode->i_sb->s_id;
945 * Put the osd object once done with it.
947 * \param obj osd object that needs to be put
949 static inline void osd_object_put(const struct lu_env *env,
950 struct osd_object *obj)
952 dt_object_put(env, &obj->oo_dt);
955 static inline int osd_object_is_root(const struct osd_object *obj)
957 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
960 static inline struct osd_object *osd_obj(const struct lu_object *o)
962 LASSERT(lu_device_is_osd(o->lo_dev));
963 return container_of0(o, struct osd_object, oo_dt.do_lu);
966 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
968 return osd_obj(&d->do_lu);
971 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
973 return &osd->od_dt_dev.dd_lu_dev;
976 static inline journal_t *osd_journal(const struct osd_device *dev)
978 return LDISKFS_SB(osd_sb(dev))->s_journal;
981 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
983 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
986 static inline char *osd_name(struct osd_device *osd)
988 return osd->od_svname;
991 static inline bool osd_is_ea_inode(struct inode *inode)
993 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
996 extern const struct dt_body_operations osd_body_ops;
997 extern struct lu_context_key osd_key;
999 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1001 return lu_context_key_get(&env->le_ctx, &osd_key);
1004 extern const struct dt_body_operations osd_body_ops_new;
1010 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1011 const struct iam_container *bag)
1013 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1014 osd_oti_get(env)->oti_it_ipd);
1018 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1019 const struct iam_container *bag)
1021 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1022 osd_oti_get(env)->oti_idx_ipd);
1025 static inline void osd_ipd_put(const struct lu_env *env,
1026 const struct iam_container *bag,
1027 struct iam_path_descr *ipd)
1029 bag->ic_descr->id_ops->id_ipd_free(ipd);
1032 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1033 const loff_t size, const loff_t pos,
1036 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1037 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1038 int write_NUL, loff_t *offs, handle_t *handle);
1041 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1042 struct inode *inode,
1043 const char *name, const int namelen)
1045 struct osd_thread_info *info = osd_oti_get(env);
1046 struct dentry *child_dentry = &info->oti_child_dentry;
1047 struct dentry *obj_dentry = &info->oti_obj_dentry;
1049 obj_dentry->d_inode = inode;
1050 obj_dentry->d_sb = inode->i_sb;
1051 obj_dentry->d_name.hash = 0;
1053 child_dentry->d_name.hash = 0;
1054 child_dentry->d_parent = obj_dentry;
1055 child_dentry->d_name.name = name;
1056 child_dentry->d_name.len = namelen;
1057 return child_dentry;
1060 extern int osd_trans_declare_op2rb[];
1061 extern int ldiskfs_track_declares_assert;
1062 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1064 static inline void osd_trans_declare_op(const struct lu_env *env,
1065 struct osd_thandle *oh,
1066 enum osd_op_type op, int credits)
1068 struct osd_thread_info *oti = osd_oti_get(env);
1070 LASSERT(oh->ot_handle == NULL);
1071 if (unlikely(op >= OSD_OT_MAX)) {
1072 if (unlikely(ldiskfs_track_declares_assert)) {
1073 LASSERT(op < OSD_OT_MAX);
1075 CWARN("%s: Invalid operation index %d\n",
1076 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1077 libcfs_debug_dumpstack(NULL);
1080 oti->oti_declare_ops[op]++;
1081 oti->oti_declare_ops_cred[op] += credits;
1083 oh->ot_credits += credits;
1086 static inline void osd_trans_exec_op(const struct lu_env *env,
1088 enum osd_op_type op)
1090 struct osd_thread_info *oti = osd_oti_get(env);
1091 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1093 unsigned int rb, left;
1095 LASSERT(oh->ot_handle != NULL);
1096 if (unlikely(op >= OSD_OT_MAX)) {
1097 if (unlikely(ldiskfs_track_declares_assert))
1098 LASSERT(op < OSD_OT_MAX);
1100 CWARN("%s: opcode %u: invalid value >= %u\n",
1101 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1103 libcfs_debug_dumpstack(NULL);
1108 /* find rollback (or reverse) operation for the given one
1109 * such an operation doesn't require additional credits
1110 * as the same set of blocks are modified */
1111 rb = osd_trans_declare_op2rb[op];
1113 /* check whether credits for this operation were reserved at all */
1114 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1115 oti->oti_declare_ops_cred[rb] == 0)) {
1116 /* the API is not perfect yet: CREATE does REF_ADD internally
1117 * while DESTROY does not. To rollback CREATE the callers
1118 * needs to call REF_DEL+DESTROY which is hard to detect using
1119 * a simple table of rollback operations */
1120 if (op == OSD_OT_REF_DEL &&
1121 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1123 if (op == OSD_OT_REF_ADD &&
1124 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1126 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1127 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1128 osd_trans_dump_creds(env, th);
1129 LASSERT(!ldiskfs_track_declares_assert);
1133 /* remember how many credits we have unused before the operation */
1134 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1135 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1136 if (unlikely(oti->oti_credits_before < left)) {
1137 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1138 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1139 oti->oti_credits_before, left, rb);
1140 osd_trans_dump_creds(env, th);
1141 /* on a very small fs (testing?) it's possible that
1142 * the transaction can't fit 1/4 of journal, so we
1143 * just request less credits (see osd_trans_start()).
1144 * ignore the same case here */
1145 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1146 if (unlikely(oh->ot_credits < rb))
1147 LASSERT(!ldiskfs_track_declares_assert);
1151 static inline void osd_trans_exec_check(const struct lu_env *env,
1153 enum osd_op_type op)
1155 struct osd_thread_info *oti = osd_oti_get(env);
1156 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1158 int used, over, quota;
1160 /* how many credits have been used by the operation */
1161 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1163 if (unlikely(used < 0)) {
1164 /* if some block was allocated and released in the same
1165 * transaction, then it won't be a part of the transaction
1166 * and delta can be negative */
1171 /* rollback operations (e.g. when we destroy just created
1172 * object) should not consume any credits. there is no point
1173 * to confuse the checks below */
1177 oti->oti_declare_ops_used[op] += used;
1178 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1181 /* we account quota for a whole transaction and any operation can
1182 * consume corresponding credits */
1183 over = oti->oti_declare_ops_used[op] -
1184 oti->oti_declare_ops_cred[op];
1185 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1186 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1187 if (over <= quota) {
1188 /* probably that credits were consumed by
1189 * quota indirectly (in the depths of ldiskfs) */
1190 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1191 oti->oti_declare_ops_used[op] -= over;
1193 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1194 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1195 oti->oti_declare_ops_used[op], used,
1196 oti->oti_declare_ops_cred[op]);
1197 osd_trans_dump_creds(env, th);
1198 if (unlikely(ldiskfs_track_declares_assert))
1204 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1207 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1208 struct lu_fid *befider)
1210 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1211 memcpy(pack->fp_area, befider, sizeof(*befider));
1212 pack->fp_len = sizeof(*befider) + 1;
1216 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1221 switch (pack->fp_len) {
1222 case sizeof *fid + 1:
1223 memcpy(fid, pack->fp_area, sizeof *fid);
1224 fid_be_to_cpu(fid, fid);
1227 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1234 * Quota/Accounting handling
1236 extern const struct dt_index_operations osd_acct_index_ops;
1237 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1238 const struct lu_fid *fid, struct osd_inode_id *id);
1240 /* copy from fs/ext4/dir.c */
1241 static inline int is_32bit_api(void)
1243 #ifdef CONFIG_COMPAT
1244 return in_compat_syscall();
1246 return (BITS_PER_LONG == 32);
1250 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1252 if ((filp->f_mode & FMODE_32BITHASH) ||
1253 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1254 return LDISKFS_HTREE_EOF_32BIT;
1256 return LDISKFS_HTREE_EOF_64BIT;
1259 static inline int fid_is_internal(const struct lu_fid *fid)
1261 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1264 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1266 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1270 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1271 * return code has been removed and integrated into the pointer in the
1274 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1275 struct inode *inode,
1276 ldiskfs_lblk_t block,
1280 struct buffer_head *bh;
1283 rc = osd_attach_jinode(inode);
1287 #ifdef HAVE_EXT4_BREAD_4ARGS
1288 bh = ldiskfs_bread(handle, inode, block, create);
1291 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1292 if (bh == NULL && rc != 0)
1298 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1299 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1301 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1304 #endif /* _OSD_INTERNAL_H */