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 struct osd_directory {
99 struct iam_container od_container;
100 struct iam_descr od_descr;
104 * Object Index (oi) instance.
108 * underlying index object, where fid->id mapping in stored.
110 struct inode *oi_inode;
111 struct osd_directory oi_dir;
114 extern const int osd_dto_credits_noquota[];
117 struct dt_object oo_dt;
119 * Inode for file system object represented by this osd_object. This
120 * inode is pinned for the whole duration of lu_object life.
122 * Not modified concurrently (either setup early during object
123 * creation, or assigned by osd_object_create() under write lock).
125 struct inode *oo_inode;
127 * to protect index ops.
129 struct htree_lock_head *oo_hl_head;
130 struct rw_semaphore oo_ext_idx_sem;
131 struct rw_semaphore oo_sem;
132 struct osd_directory *oo_dir;
133 /** protects inode attributes. */
136 __u32 oo_destroyed:1;
138 /* the i_flags in LMA */
141 * Following two members are used to indicate the presence of dot and
142 * dotdot in the given directory. This is required for interop mode
145 int oo_compat_dot_created;
146 int oo_compat_dotdot_created;
148 const struct lu_env *oo_owner;
149 #ifdef CONFIG_LOCKDEP
150 struct lockdep_map oo_dep_map;
153 struct list_head oo_xattr_list;
157 /* protects on-fly initialization */
158 int oos_subdir_count; /* subdir count for each seq */
159 struct dentry *oos_root; /* O/<seq> */
160 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
161 u64 oos_seq; /* seq number */
162 struct list_head oos_seq_list; /* list to seq_list */
166 struct dentry *om_root; /* dentry for /O */
167 rwlock_t om_seq_list_lock; /* lock for seq_list */
168 struct list_head om_seq_list; /* list head for seq */
170 struct mutex om_dir_init_mutex;
174 struct dentry *om_root; /* AGENT/<index> */
175 u64 om_index; /* mdt index */
176 struct list_head om_list; /* list to omm_list */
179 struct osd_mdobj_map {
180 struct dentry *omm_remote_parent;
182 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
183 handle_t *handle, struct dentry *child,
184 struct inode *inode, struct htree_lock *hlock);
186 #define OSD_OTABLE_IT_CACHE_SIZE 64
187 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
189 struct osd_inconsistent_item {
190 /* link into osd_scrub::os_inconsistent_items,
191 * protected by osd_scrub::os_lock. */
192 struct list_head oii_list;
194 /* The right FID <=> ino#/gen mapping. */
195 struct osd_idmap_cache oii_cache;
197 unsigned int oii_insert:1; /* insert or update mapping. */
200 struct osd_otable_cache {
201 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
203 /* Index for next cache slot to be filled. */
204 int ooc_producer_idx;
206 /* Index for next cache slot to be returned by it::next(). */
207 int ooc_consumer_idx;
209 /* How many items in ooc_cache. */
210 int ooc_cached_items;
212 /* Position for up layer LFSCK iteration pre-loading. */
213 __u32 ooc_pos_preload;
216 struct osd_otable_it {
217 struct osd_device *ooi_dev;
218 struct osd_otable_cache ooi_cache;
220 /* The following bits can be updated/checked w/o lock protection.
221 * If more bits will be introduced in the future and need lock to
222 * protect, please add comment. */
223 unsigned long ooi_used_outside:1, /* Some user out of OSD
224 * uses the iteration. */
225 ooi_all_cached:1, /* No more entries can be
226 * filled into cache. */
227 ooi_user_ready:1, /* The user out of OSD is
228 * ready to iterate. */
229 ooi_waiting:1; /* it::next is waiting. */
232 struct osd_obj_orphan {
233 struct list_head oor_list;
234 struct lu_env *oor_env; /* to identify "own" records */
243 struct dt_device od_dt_dev;
244 /* information about underlying file system */
245 struct vfsmount *od_mnt;
247 struct osd_oi **od_oi_table;
248 /* total number of OI containers */
253 unsigned int od_fl_capa:1,
261 __u32 od_dirent_journal;
263 struct proc_dir_entry *od_proc_entry;
264 struct lprocfs_stats *od_stats;
266 spinlock_t od_osfs_lock;
269 struct lu_site od_site;
271 struct osd_obj_map *od_ost_map;
272 struct osd_mdobj_map *od_mdt_map;
274 unsigned long long od_readcache_max_filesize;
276 int od_writethrough_cache;
278 struct brw_stats od_brw_stats;
279 atomic_t od_r_in_flight;
280 atomic_t od_w_in_flight;
282 struct mutex od_otable_mutex;
283 struct osd_otable_it *od_otable_it;
284 struct osd_scrub od_scrub;
285 struct list_head od_ios_list;
287 /* service name associated with the osd device */
288 char od_svname[MAX_OBD_NAME];
289 char od_mntdev[MAX_OBD_NAME];
291 /* quota slave instance */
292 struct qsd_instance *od_quota_slave;
294 /* osd seq instance */
295 struct lu_client_seq *od_cl_seq;
296 /* If the ratio of "the total OI mappings count" vs
297 * "the bad OI mappings count" is lower than the
298 * osd_device::od_full_scrub_ratio, then trigger
299 * OI scrub to scan the whole the device. */
300 __u64 od_full_scrub_ratio;
301 /* If the speed of found bad OI mappings (per minute)
302 * exceeds the osd_device::od_full_scrub_threshold_rate,
303 * then trigger OI scrub to scan the whole device. */
304 __u64 od_full_scrub_threshold_rate;
306 /* a list of orphaned agent inodes, protected with od_osfs_lock */
307 struct list_head od_orphan_list;
310 enum osd_full_scrub_ratio {
311 /* Trigger OI scrub to scan the whole device directly. */
314 /* Because the bad OI mappings count cannot be larger than
315 * the total OI mappints count, then setting OFSR_NEVER means
316 * that the whole device scanning cannot be triggered by auto
317 * detected bad OI mappings during the RPC services. */
319 OFSR_DEFAULT = 10000,
322 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
324 /* There are at most 15 uid/gid/projids are affected in a transaction, and
325 * that's rename case:
326 * - 3 for source parent uid & gid & projid;
327 * - 3 for source child uid & gid & projid ('..' entry update when
328 * child is directory);
329 * - 3 for target parent uid & gid & projid;
330 * - 3 for target child uid & gid & projid(if the target child exists);
331 * - 3 for root uid & gid(last_rcvd, llog, etc);
334 #define OSD_MAX_UGID_CNT 15
339 OSD_OT_XATTR_SET = 2,
352 struct thandle ot_super;
354 struct ldiskfs_journal_cb_entry ot_jcb;
355 struct list_head ot_commit_dcb_list;
356 struct list_head ot_stop_dcb_list;
357 /* Link to the device, for debugging. */
358 struct lu_ref_link ot_dev_link;
359 unsigned int ot_credits;
360 unsigned short ot_id_cnt;
361 __u8 ot_id_types[OSD_MAX_UGID_CNT];
362 unsigned int ot_remove_agents:1;
363 uid_t ot_id_array[OSD_MAX_UGID_CNT];
364 struct lquota_trans *ot_quota_trans;
365 #if OSD_THANDLE_STATS
366 /** time when this handle was allocated */
369 /** time when this thanle was started */
375 * Basic transaction credit op
396 #ifdef CONFIG_PROC_FS
398 LPROC_OSD_READ_BYTES = 0,
399 LPROC_OSD_WRITE_BYTES = 1,
400 LPROC_OSD_GET_PAGE = 2,
401 LPROC_OSD_NO_PAGE = 3,
402 LPROC_OSD_CACHE_ACCESS = 4,
403 LPROC_OSD_CACHE_HIT = 5,
404 LPROC_OSD_CACHE_MISS = 6,
406 #if OSD_THANDLE_STATS
407 LPROC_OSD_THANDLE_STARTING,
408 LPROC_OSD_THANDLE_OPEN,
409 LPROC_OSD_THANDLE_CLOSING,
416 * Storage representation for fids.
418 * Variable size, first byte contains the length of the whole record.
420 struct osd_fid_pack {
421 unsigned char fp_len;
422 char fp_area[sizeof(struct lu_fid)];
425 struct osd_it_ea_dirent {
426 struct lu_fid oied_fid;
429 unsigned short oied_namelen;
430 unsigned int oied_type;
432 } __attribute__((packed));
435 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
436 * than lu_dirent struct. osd readdir reads less number of dirent than
437 * required for mdd dir page. so buffer size need to be increased so that
438 * there would be one ext3 readdir for every mdd readdir page.
441 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
444 * This is iterator's in-memory data structure in interoperability
445 * mode (i.e. iterator over ldiskfs style directory)
448 struct osd_object *oie_obj;
449 /** used in ldiskfs iterator, to stored file pointer */
450 struct file oie_file;
451 /** how many entries have been read-cached from storage */
453 /** current entry is being iterated by caller */
455 /** current processing entry */
456 struct osd_it_ea_dirent *oie_dirent;
457 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
459 struct dentry oie_dentry;
463 * Iterator's in-memory data structure for IAM mode.
466 struct osd_object *oi_obj;
467 struct iam_path_descr *oi_ipd;
468 struct iam_iterator oi_it;
471 struct osd_quota_leaf {
472 struct list_head oql_link;
477 * Iterator's in-memory data structure for quota file.
479 struct osd_it_quota {
480 struct osd_object *oiq_obj;
481 /** tree blocks path to where the entry is stored */
482 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
483 /** on-disk offset for current key where quota record can be found */
485 /** identifier for current quota record */
487 /** the record index in the leaf/index block */
488 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
489 /** list of already processed leaf blocks */
490 struct list_head oiq_list;
493 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
496 wait_queue_head_t dr_wait;
497 atomic_t dr_numreqs; /* number of reqs being processed */
502 unsigned int dr_ignore_quota:1;
503 unsigned int dr_elapsed_valid:1; /* we really did count time */
504 unsigned int dr_rw:1;
505 struct lu_buf dr_pg_buf;
506 struct page **dr_pages;
507 struct lu_buf dr_bl_buf;
509 unsigned long dr_start_time;
510 unsigned long dr_elapsed; /* how long io took */
511 struct osd_device *dr_dev;
512 unsigned int dr_init_at; /* the line iobuf was initialized */
515 #define OSD_INS_CACHE_SIZE 8
517 struct osd_thread_info {
518 const struct lu_env *oti_env;
520 * used for index operations.
522 struct dentry oti_obj_dentry;
523 struct dentry oti_child_dentry;
525 /** dentry for Iterator context. */
526 struct dentry oti_it_dentry;
529 /* fake struct file for osd_object_sync */
530 struct file oti_file;
532 struct kstatfs oti_ksfs;
535 struct htree_lock *oti_hlock;
537 struct lu_fid oti_fid;
538 struct lu_fid oti_fid2;
539 struct lu_fid oti_fid3;
540 struct osd_inode_id oti_id;
541 struct osd_inode_id oti_id2;
542 struct osd_inode_id oti_id3;
543 struct ost_id oti_ostid;
546 * XXX temporary: for ->i_op calls.
548 struct timespec oti_time;
551 * following ipd and it structures are used for osd_index_iam_lookup()
552 * these are defined separately as we might do index operation
553 * in open iterator session.
556 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
558 unsigned int oti_it_ea_buf_used:1;
560 /* IAM iterator for index operation. */
561 struct iam_iterator oti_idx_it;
563 /** union to guarantee that ->oti_ipd[] has proper alignment. */
566 char oti_it_ipd[DX_IPD_MAX_SIZE];
567 long long oti_alignment_lieutenant;
571 char oti_idx_ipd[DX_IPD_MAX_SIZE];
572 long long oti_alignment_lieutenant_colonel;
575 struct osd_idmap_cache oti_cache;
577 /* dedicated OI cache for insert (which needs inum) */
578 struct osd_idmap_cache *oti_ins_cache;
579 int oti_ins_cache_size;
580 int oti_ins_cache_used;
585 /** used in osd_fid_set() to put xattr */
586 struct lu_buf oti_buf;
587 struct lu_buf oti_big_buf;
588 /** used in osd_ea_fid_set() to set fid into common ea */
590 struct lustre_ost_attrs oti_ost_attrs;
591 struct filter_fid_old oti_ff;
592 struct filter_fid oti_ff_new;
595 struct osd_iobuf oti_iobuf;
596 /* used to access objects in /O */
597 struct inode *oti_inode;
598 #define OSD_FID_REC_SZ 32
599 char oti_ldp[OSD_FID_REC_SZ];
600 char oti_ldp2[OSD_FID_REC_SZ];
602 /* used by quota code */
604 #if defined(HAVE_DQUOT_QC_DQBLK)
605 struct qc_dqblk oti_qdq;
606 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
607 struct fs_disk_quota oti_fdq;
609 struct if_dqblk oti_dqblk;
611 struct if_dqinfo oti_dqinfo;
613 struct lquota_id_info oti_qi;
614 struct lquota_trans oti_quota_trans;
615 union lquota_rec oti_quota_rec;
617 struct lu_seq_range oti_seq_range;
619 /* Tracking for transaction credits, to allow debugging and optimizing
620 * cases where a large number of credits are being allocated for
621 * single transaction. */
622 unsigned int oti_credits_before;
623 unsigned int oti_declare_ops[OSD_OT_MAX];
624 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
625 unsigned int oti_declare_ops_used[OSD_OT_MAX];
628 /* flags for inode/block quota accounting */
629 enum osd_qid_declare_flags {
630 OSD_QID_INODE = 1 << 0,
631 OSD_QID_BLK = 1 << 1,
632 OSD_QID_FORCE = 1 << 2,
635 extern int ldiskfs_pdo;
637 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
638 const char *name, void *buf, int len)
643 dentry->d_inode = inode;
644 dentry->d_sb = inode->i_sb;
645 return inode->i_op->getxattr(dentry, name, buf, len);
648 static inline int __osd_xattr_set(struct osd_thread_info *info,
649 struct inode *inode, const char *name,
650 const void *buf, int buflen, int fl)
652 struct dentry *dentry = &info->oti_child_dentry;
654 ll_vfs_dq_init(inode);
655 dentry->d_inode = inode;
656 dentry->d_sb = inode->i_sb;
657 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
660 #ifdef CONFIG_PROC_FS
662 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
663 extern struct lprocfs_vars lprocfs_osd_module_vars[];
664 int osd_procfs_init(struct osd_device *osd, const char *name);
665 int osd_procfs_fini(struct osd_device *osd);
666 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
667 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
668 int osd_register_proc_index_in_idif(struct osd_device *osd);
672 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
673 struct obd_statfs *sfs);
674 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
675 struct osd_inode_id *id);
676 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
677 const struct lu_fid *fid, __u32 compat, __u32 incompat);
678 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
679 struct dentry *dentry, struct lustre_ost_attrs *loa);
680 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
681 struct osd_inode_id *id, const struct lu_fid *fid);
682 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
683 struct dentry *dentry, struct lu_fid *fid);
685 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
686 void osd_obj_map_fini(struct osd_device *dev);
687 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
688 const struct lu_fid *fid, struct osd_inode_id *id);
689 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
690 const struct lu_fid *fid, const struct osd_inode_id *id,
692 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
693 const struct lu_fid *fid, handle_t *th);
694 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
695 const struct lu_fid *fid, const struct osd_inode_id *id,
697 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
698 struct inode *src_parent, struct dentry *src_child,
699 const struct lu_fid *fid);
700 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
701 const struct lu_fid *fid, struct osd_inode_id *id);
702 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
703 const struct lu_fid *fid, const struct osd_inode_id *id,
705 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
706 const struct lu_fid *fid, const struct osd_inode_id *id,
709 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
710 int osd_scrub_file_store(struct osd_scrub *scrub);
711 char *osd_lf_fid2name(const struct lu_fid *fid);
712 int osd_scrub_start(struct osd_device *dev, __u32 flags);
713 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
714 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
715 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
717 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
718 struct osd_inode_id *id);
719 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
721 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
722 u64 seq, struct lu_seq_range *range);
724 int osd_delete_from_remote_parent(const struct lu_env *env,
725 struct osd_device *osd,
726 struct osd_object *obj,
727 struct osd_thandle *oh);
728 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
729 struct osd_object *obj, struct osd_thandle *oh);
730 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
731 struct osd_device *osd,
732 const struct lu_fid *fid,
733 struct osd_inode_id *id);
735 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
737 /* osd_quota_fmt.c */
738 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
739 int type, uint blk, int depth, uint index,
740 struct osd_it_quota *it);
741 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
742 int type, uint blk, uint index,
743 struct osd_it_quota *it);
744 loff_t find_tree_dqentry(const struct lu_env *env,
745 struct osd_object *obj, int type,
746 qid_t dqid, uint blk, int depth,
747 struct osd_it_quota *it);
749 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
750 struct lquota_id_info *qi, struct osd_object *obj,
751 bool enforce, int *flags);
752 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
753 __u32 projid, long long space, struct osd_thandle *oh,
754 struct osd_object *obj, int *flags,
755 enum osd_qid_declare_flags);
756 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
757 const struct dt_rec *rec,
758 union lquota_rec *quota_rec);
759 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
761 #ifndef HAVE_I_UID_READ
762 static inline uid_t i_uid_read(const struct inode *inode)
767 static inline gid_t i_gid_read(const struct inode *inode)
772 static inline void i_uid_write(struct inode *inode, uid_t uid)
777 static inline void i_gid_write(struct inode *inode, gid_t gid)
783 #ifdef HAVE_PROJECT_QUOTA
784 static inline __u32 i_projid_read(struct inode *inode)
786 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
789 static inline void i_projid_write(struct inode *inode, __u32 projid)
792 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
793 LDISKFS_I(inode)->i_projid = kprojid;
796 static inline uid_t i_projid_read(struct inode *inode)
800 static inline void i_projid_write(struct inode *inode, __u32 projid)
806 #ifdef HAVE_LDISKFS_INFO_JINODE
807 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
808 #else /* HAVE_LDISKFS_INFO_JINODE */
809 # define osd_attach_jinode(inode) 0
810 #endif /* HAVE_LDISKFS_INFO_JINODE */
812 #ifdef LDISKFS_HT_MISC
813 # define osd_journal_start_sb(sb, type, nblock) \
814 ldiskfs_journal_start_sb(sb, type, nblock)
815 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
817 ldiskfs_lblk_t *nblock)
821 rc = osd_attach_jinode(inode);
824 return ldiskfs_append(handle, inode, nblock);
826 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
827 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
829 # define osd_journal_start(inode, type, nblocks) \
830 ldiskfs_journal_start(inode, type, nblocks)
831 # define osd_transaction_size(dev) \
832 (osd_journal(dev)->j_max_transaction_buffers / 2)
834 # define LDISKFS_HT_MISC 0
835 # define osd_journal_start_sb(sb, type, nblock) \
836 ldiskfs_journal_start_sb(sb, nblock)
838 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
840 ldiskfs_lblk_t *nblock)
842 struct buffer_head *bh;
845 bh = ldiskfs_append(handle, inode, nblock, &err);
852 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
853 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
855 # define osd_journal_start(inode, type, nblocks) \
856 ldiskfs_journal_start(inode, nblocks)
857 # define osd_transaction_size(dev) \
858 (osd_journal(dev)->j_max_transaction_buffers)
862 * Invariants, assertions.
866 * XXX: do not enable this, until invariant checking code is made thread safe
867 * in the face of pdirops locking.
869 #define OSD_INVARIANT_CHECKS (0)
871 #if OSD_INVARIANT_CHECKS
872 static inline int osd_invariant(const struct osd_object *obj)
876 ergo(obj->oo_inode != NULL,
877 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
878 atomic_read(&obj->oo_inode->i_count) > 0) &&
879 ergo(obj->oo_dir != NULL &&
880 obj->oo_dir->od_conationer.ic_object != NULL,
881 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
884 #define osd_invariant(obj) (1)
887 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
889 extern const struct dt_index_operations osd_otable_ops;
891 static inline int osd_oi_fid2idx(struct osd_device *dev,
892 const struct lu_fid *fid)
894 return fid->f_seq & (dev->od_oi_count - 1);
897 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
898 const struct lu_fid *fid)
900 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
901 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
902 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
903 DFID"\n", PFID(fid));
904 /* It can work even od_oi_count equals to 1 although it's unexpected,
905 * the only reason we set it to 1 is for performance measurement */
906 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
909 extern const struct lu_device_operations osd_lu_ops;
911 static inline int lu_device_is_osd(const struct lu_device *d)
913 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
916 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
918 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
919 return container_of0(d, struct osd_device, od_dt_dev);
922 static inline struct osd_device *osd_dev(const struct lu_device *d)
924 LASSERT(lu_device_is_osd(d));
925 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
928 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
930 return osd_dev(o->oo_dt.do_lu.lo_dev);
933 static inline struct super_block *osd_sb(const struct osd_device *dev)
935 return dev->od_mnt->mnt_sb;
939 * Put the osd object once done with it.
941 * \param obj osd object that needs to be put
943 static inline void osd_object_put(const struct lu_env *env,
944 struct osd_object *obj)
946 dt_object_put(env, &obj->oo_dt);
949 static inline int osd_object_is_root(const struct osd_object *obj)
951 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
954 static inline struct osd_object *osd_obj(const struct lu_object *o)
956 LASSERT(lu_device_is_osd(o->lo_dev));
957 return container_of0(o, struct osd_object, oo_dt.do_lu);
960 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
962 return osd_obj(&d->do_lu);
965 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
967 return &osd->od_dt_dev.dd_lu_dev;
970 static inline journal_t *osd_journal(const struct osd_device *dev)
972 return LDISKFS_SB(osd_sb(dev))->s_journal;
975 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
977 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
980 static inline char *osd_name(struct osd_device *osd)
982 return osd->od_svname;
985 static inline bool osd_is_ea_inode(struct inode *inode)
987 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
990 extern const struct dt_body_operations osd_body_ops;
991 extern struct lu_context_key osd_key;
993 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
995 return lu_context_key_get(&env->le_ctx, &osd_key);
998 extern const struct dt_body_operations osd_body_ops_new;
1004 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1005 const struct iam_container *bag)
1007 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1008 osd_oti_get(env)->oti_it_ipd);
1012 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1013 const struct iam_container *bag)
1015 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1016 osd_oti_get(env)->oti_idx_ipd);
1019 static inline void osd_ipd_put(const struct lu_env *env,
1020 const struct iam_container *bag,
1021 struct iam_path_descr *ipd)
1023 bag->ic_descr->id_ops->id_ipd_free(ipd);
1026 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1027 const loff_t size, const loff_t pos,
1030 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1031 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1032 int write_NUL, loff_t *offs, handle_t *handle);
1035 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1036 struct inode *inode,
1037 const char *name, const int namelen)
1039 struct osd_thread_info *info = osd_oti_get(env);
1040 struct dentry *child_dentry = &info->oti_child_dentry;
1041 struct dentry *obj_dentry = &info->oti_obj_dentry;
1043 obj_dentry->d_inode = inode;
1044 obj_dentry->d_sb = inode->i_sb;
1045 obj_dentry->d_name.hash = 0;
1047 child_dentry->d_name.hash = 0;
1048 child_dentry->d_parent = obj_dentry;
1049 child_dentry->d_name.name = name;
1050 child_dentry->d_name.len = namelen;
1051 return child_dentry;
1054 extern int osd_trans_declare_op2rb[];
1055 extern int ldiskfs_track_declares_assert;
1056 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1058 static inline void osd_trans_declare_op(const struct lu_env *env,
1059 struct osd_thandle *oh,
1060 enum osd_op_type op, int credits)
1062 struct osd_thread_info *oti = osd_oti_get(env);
1064 LASSERT(oh->ot_handle == NULL);
1065 if (unlikely(op >= OSD_OT_MAX)) {
1066 if (unlikely(ldiskfs_track_declares_assert)) {
1067 LASSERT(op < OSD_OT_MAX);
1069 CWARN("%s: Invalid operation index %d\n",
1070 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1071 libcfs_debug_dumpstack(NULL);
1074 oti->oti_declare_ops[op]++;
1075 oti->oti_declare_ops_cred[op] += credits;
1077 oh->ot_credits += credits;
1080 static inline void osd_trans_exec_op(const struct lu_env *env,
1082 enum osd_op_type op)
1084 struct osd_thread_info *oti = osd_oti_get(env);
1085 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1087 unsigned int rb, left;
1089 LASSERT(oh->ot_handle != NULL);
1090 if (unlikely(op >= OSD_OT_MAX)) {
1091 if (unlikely(ldiskfs_track_declares_assert))
1092 LASSERT(op < OSD_OT_MAX);
1094 CWARN("%s: opcode %u: invalid value >= %u\n",
1095 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1097 libcfs_debug_dumpstack(NULL);
1102 /* find rollback (or reverse) operation for the given one
1103 * such an operation doesn't require additional credits
1104 * as the same set of blocks are modified */
1105 rb = osd_trans_declare_op2rb[op];
1107 /* check whether credits for this operation were reserved at all */
1108 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1109 oti->oti_declare_ops_cred[rb] == 0)) {
1110 /* the API is not perfect yet: CREATE does REF_ADD internally
1111 * while DESTROY does not. To rollback CREATE the callers
1112 * needs to call REF_DEL+DESTROY which is hard to detect using
1113 * a simple table of rollback operations */
1114 if (op == OSD_OT_REF_DEL &&
1115 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1117 if (op == OSD_OT_REF_ADD &&
1118 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1120 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1121 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1122 osd_trans_dump_creds(env, th);
1123 LASSERT(!ldiskfs_track_declares_assert);
1127 /* remember how many credits we have unused before the operation */
1128 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1129 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1130 if (unlikely(oti->oti_credits_before < left)) {
1131 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1132 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1133 oti->oti_credits_before, left, rb);
1134 osd_trans_dump_creds(env, th);
1135 /* on a very small fs (testing?) it's possible that
1136 * the transaction can't fit 1/4 of journal, so we
1137 * just request less credits (see osd_trans_start()).
1138 * ignore the same case here */
1139 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1140 if (unlikely(oh->ot_credits < rb))
1141 LASSERT(!ldiskfs_track_declares_assert);
1145 static inline void osd_trans_exec_check(const struct lu_env *env,
1147 enum osd_op_type op)
1149 struct osd_thread_info *oti = osd_oti_get(env);
1150 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1152 int used, over, quota;
1154 /* how many credits have been used by the operation */
1155 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1157 if (unlikely(used < 0)) {
1158 /* if some block was allocated and released in the same
1159 * transaction, then it won't be a part of the transaction
1160 * and delta can be negative */
1165 /* rollback operations (e.g. when we destroy just created
1166 * object) should not consume any credits. there is no point
1167 * to confuse the checks below */
1171 oti->oti_declare_ops_used[op] += used;
1172 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1175 /* we account quota for a whole transaction and any operation can
1176 * consume corresponding credits */
1177 over = oti->oti_declare_ops_used[op] -
1178 oti->oti_declare_ops_cred[op];
1179 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1180 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1181 if (over <= quota) {
1182 /* probably that credits were consumed by
1183 * quota indirectly (in the depths of ldiskfs) */
1184 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1185 oti->oti_declare_ops_used[op] -= over;
1187 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1188 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1189 oti->oti_declare_ops_used[op], used,
1190 oti->oti_declare_ops_cred[op]);
1191 osd_trans_dump_creds(env, th);
1192 if (unlikely(ldiskfs_track_declares_assert))
1198 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1201 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1202 struct lu_fid *befider)
1204 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1205 memcpy(pack->fp_area, befider, sizeof(*befider));
1206 pack->fp_len = sizeof(*befider) + 1;
1210 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1215 switch (pack->fp_len) {
1216 case sizeof *fid + 1:
1217 memcpy(fid, pack->fp_area, sizeof *fid);
1218 fid_be_to_cpu(fid, fid);
1221 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1228 * Quota/Accounting handling
1230 extern const struct dt_index_operations osd_acct_index_ops;
1231 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1232 const struct lu_fid *fid, struct osd_inode_id *id);
1234 /* copy from fs/ext4/dir.c */
1235 static inline int is_32bit_api(void)
1237 #ifdef CONFIG_COMPAT
1238 return in_compat_syscall();
1240 return (BITS_PER_LONG == 32);
1244 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1246 if ((filp->f_mode & FMODE_32BITHASH) ||
1247 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1248 return LDISKFS_HTREE_EOF_32BIT;
1250 return LDISKFS_HTREE_EOF_64BIT;
1253 static inline int fid_is_internal(const struct lu_fid *fid)
1255 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1258 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1260 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1264 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1265 * return code has been removed and integrated into the pointer in the
1268 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1269 struct inode *inode,
1270 ldiskfs_lblk_t block,
1274 struct buffer_head *bh;
1277 rc = osd_attach_jinode(inode);
1281 #ifdef HAVE_EXT4_BREAD_4ARGS
1282 bh = ldiskfs_bread(handle, inode, block, create);
1285 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1286 if (bh == NULL && rc != 0)
1292 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1293 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1295 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1298 #endif /* _OSD_INTERNAL_H */