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;
157 struct lu_object_header *oo_header;
161 /* protects on-fly initialization */
162 int oos_subdir_count; /* subdir count for each seq */
163 struct dentry *oos_root; /* O/<seq> */
164 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
165 u64 oos_seq; /* seq number */
166 struct list_head oos_seq_list; /* list to seq_list */
170 struct dentry *om_root; /* dentry for /O */
171 rwlock_t om_seq_list_lock; /* lock for seq_list */
172 struct list_head om_seq_list; /* list head for seq */
174 struct mutex om_dir_init_mutex;
178 struct dentry *om_root; /* AGENT/<index> */
179 u64 om_index; /* mdt index */
180 struct list_head om_list; /* list to omm_list */
183 struct osd_mdobj_map {
184 struct dentry *omm_remote_parent;
186 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
187 handle_t *handle, struct dentry *child,
188 struct inode *inode, struct htree_lock *hlock);
190 #define OSD_OTABLE_IT_CACHE_SIZE 64
191 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
193 struct osd_inconsistent_item {
194 /* link into lustre_scrub::os_inconsistent_items,
195 * protected by lustre_scrub::os_lock. */
196 struct list_head oii_list;
198 /* The right FID <=> ino#/gen mapping. */
199 struct osd_idmap_cache oii_cache;
201 unsigned int oii_insert:1; /* insert or update mapping. */
204 struct osd_otable_cache {
205 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
207 /* Index for next cache slot to be filled. */
208 int ooc_producer_idx;
210 /* Index for next cache slot to be returned by it::next(). */
211 int ooc_consumer_idx;
213 /* How many items in ooc_cache. */
214 __u64 ooc_cached_items;
216 /* Position for up layer LFSCK iteration pre-loading. */
217 __u64 ooc_pos_preload;
220 struct osd_otable_it {
221 struct osd_device *ooi_dev;
222 struct osd_otable_cache ooi_cache;
223 struct osd_iit_param ooi_iit_param;
225 /* The following bits can be updated/checked w/o lock protection.
226 * If more bits will be introduced in the future and need lock to
227 * protect, please add comment. */
228 unsigned long ooi_used_outside:1, /* Some user out of OSD
229 * uses the iteration. */
230 ooi_all_cached:1, /* No more entries can be
231 * filled into cache. */
232 ooi_user_ready:1, /* The user out of OSD is
233 * ready to iterate. */
234 ooi_waiting:1; /* it::next is waiting. */
237 struct osd_obj_orphan {
238 struct list_head oor_list;
239 struct lu_env *oor_env; /* to identify "own" records */
248 struct dt_device od_dt_dev;
249 /* information about underlying file system */
250 struct vfsmount *od_mnt;
252 struct osd_oi **od_oi_table;
253 /* total number of OI containers */
258 unsigned int od_fl_capa:1,
266 __s64 od_auto_scrub_interval;
267 __u32 od_dirent_journal;
269 struct proc_dir_entry *od_proc_entry;
270 struct lprocfs_stats *od_stats;
272 spinlock_t od_osfs_lock;
275 struct lu_site od_site;
277 struct osd_obj_map *od_ost_map;
278 struct osd_mdobj_map *od_mdt_map;
280 unsigned long long od_readcache_max_filesize;
282 int od_writethrough_cache;
284 struct brw_stats od_brw_stats;
285 atomic_t od_r_in_flight;
286 atomic_t od_w_in_flight;
288 struct mutex od_otable_mutex;
289 struct osd_otable_it *od_otable_it;
290 struct osd_scrub od_scrub;
291 struct list_head od_ios_list;
293 /* service name associated with the osd device */
294 char od_svname[MAX_OBD_NAME];
295 char od_mntdev[MAX_OBD_NAME];
297 /* quota slave instance */
298 struct qsd_instance *od_quota_slave;
300 /* osd seq instance */
301 struct lu_client_seq *od_cl_seq;
302 /* If the ratio of "the total OI mappings count" vs
303 * "the bad OI mappings count" is lower than the
304 * osd_device::od_full_scrub_ratio, then trigger
305 * OI scrub to scan the whole the device. */
306 __u64 od_full_scrub_ratio;
307 /* If the speed of found bad OI mappings (per minute)
308 * exceeds the osd_device::od_full_scrub_threshold_rate,
309 * then trigger OI scrub to scan the whole device. */
310 __u64 od_full_scrub_threshold_rate;
312 /* a list of orphaned agent inodes, protected with od_osfs_lock */
313 struct list_head od_orphan_list;
314 struct list_head od_index_backup_list;
315 struct list_head od_index_restore_list;
317 struct inode *od_index_backup_inode;
318 enum lustre_index_backup_policy od_index_backup_policy;
319 int od_index_backup_stop;
322 enum osd_full_scrub_ratio {
323 /* Trigger OI scrub to scan the whole device directly. */
326 /* Because the bad OI mappings count cannot be larger than
327 * the total OI mappints count, then setting OFSR_NEVER means
328 * that the whole device scanning cannot be triggered by auto
329 * detected bad OI mappings during the RPC services. */
331 OFSR_DEFAULT = 10000,
334 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
336 /* There are at most 15 uid/gid/projids are affected in a transaction, and
337 * that's rename case:
338 * - 3 for source parent uid & gid & projid;
339 * - 3 for source child uid & gid & projid ('..' entry update when
340 * child is directory);
341 * - 3 for target parent uid & gid & projid;
342 * - 3 for target child uid & gid & projid(if the target child exists);
343 * - 3 for root uid & gid(last_rcvd, llog, etc);
346 #define OSD_MAX_UGID_CNT 15
351 OSD_OT_XATTR_SET = 2,
364 struct thandle ot_super;
366 struct ldiskfs_journal_cb_entry ot_jcb;
367 struct list_head ot_commit_dcb_list;
368 struct list_head ot_stop_dcb_list;
369 /* Link to the device, for debugging. */
370 struct lu_ref_link ot_dev_link;
371 unsigned int ot_credits;
372 unsigned short ot_id_cnt;
373 __u8 ot_id_types[OSD_MAX_UGID_CNT];
374 unsigned int ot_remove_agents:1;
375 uid_t ot_id_array[OSD_MAX_UGID_CNT];
376 struct lquota_trans *ot_quota_trans;
377 #if OSD_THANDLE_STATS
378 /** time when this handle was allocated */
381 /** time when this thanle was started */
387 * Basic transaction credit op
408 #ifdef CONFIG_PROC_FS
410 LPROC_OSD_READ_BYTES = 0,
411 LPROC_OSD_WRITE_BYTES = 1,
412 LPROC_OSD_GET_PAGE = 2,
413 LPROC_OSD_NO_PAGE = 3,
414 LPROC_OSD_CACHE_ACCESS = 4,
415 LPROC_OSD_CACHE_HIT = 5,
416 LPROC_OSD_CACHE_MISS = 6,
418 #if OSD_THANDLE_STATS
419 LPROC_OSD_THANDLE_STARTING,
420 LPROC_OSD_THANDLE_OPEN,
421 LPROC_OSD_THANDLE_CLOSING,
428 * Storage representation for fids.
430 * Variable size, first byte contains the length of the whole record.
432 struct osd_fid_pack {
433 unsigned char fp_len;
434 char fp_area[sizeof(struct lu_fid)];
437 struct osd_it_ea_dirent {
438 struct lu_fid oied_fid;
441 unsigned short oied_namelen;
442 unsigned int oied_type;
444 } __attribute__((packed));
447 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
448 * than lu_dirent struct. osd readdir reads less number of dirent than
449 * required for mdd dir page. so buffer size need to be increased so that
450 * there would be one ext3 readdir for every mdd readdir page.
453 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
456 * This is iterator's in-memory data structure in interoperability
457 * mode (i.e. iterator over ldiskfs style directory)
460 struct osd_object *oie_obj;
461 /** used in ldiskfs iterator, to stored file pointer */
462 struct file oie_file;
463 /** how many entries have been read-cached from storage */
465 /** current entry is being iterated by caller */
467 /** current processing entry */
468 struct osd_it_ea_dirent *oie_dirent;
469 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
471 struct dentry oie_dentry;
475 * Iterator's in-memory data structure for IAM mode.
478 struct osd_object *oi_obj;
479 struct iam_path_descr *oi_ipd;
480 struct iam_iterator oi_it;
483 struct osd_quota_leaf {
484 struct list_head oql_link;
489 * Iterator's in-memory data structure for quota file.
491 struct osd_it_quota {
492 struct osd_object *oiq_obj;
493 /** tree blocks path to where the entry is stored */
494 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
495 /** on-disk offset for current key where quota record can be found */
497 /** identifier for current quota record */
499 /** the record index in the leaf/index block */
500 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
501 /** list of already processed leaf blocks */
502 struct list_head oiq_list;
505 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
508 wait_queue_head_t dr_wait;
509 atomic_t dr_numreqs; /* number of reqs being processed */
514 unsigned int dr_ignore_quota:1;
515 unsigned int dr_elapsed_valid:1; /* we really did count time */
516 unsigned int dr_rw:1;
517 struct lu_buf dr_pg_buf;
518 struct page **dr_pages;
519 struct lu_buf dr_bl_buf;
521 unsigned long dr_start_time;
522 unsigned long dr_elapsed; /* how long io took */
523 struct osd_device *dr_dev;
524 unsigned int dr_init_at; /* the line iobuf was initialized */
527 #define OSD_INS_CACHE_SIZE 8
529 struct osd_thread_info {
530 const struct lu_env *oti_env;
532 * used for index operations.
534 struct dentry oti_obj_dentry;
535 struct dentry oti_child_dentry;
537 /** dentry for Iterator context. */
538 struct dentry oti_it_dentry;
541 /* fake struct file for osd_object_sync */
542 struct file oti_file;
544 struct kstatfs oti_ksfs;
547 struct htree_lock *oti_hlock;
549 struct lu_fid oti_fid;
550 struct lu_fid oti_fid2;
551 struct lu_fid oti_fid3;
552 struct osd_inode_id oti_id;
553 struct osd_inode_id oti_id2;
554 struct osd_inode_id oti_id3;
555 struct ost_id oti_ostid;
558 * XXX temporary: for ->i_op calls.
560 struct timespec oti_time;
563 * following ipd and it structures are used for osd_index_iam_lookup()
564 * these are defined separately as we might do index operation
565 * in open iterator session.
568 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
570 unsigned int oti_it_ea_buf_used:1;
572 /* IAM iterator for index operation. */
573 struct iam_iterator oti_idx_it;
575 /** union to guarantee that ->oti_ipd[] has proper alignment. */
578 char oti_it_ipd[DX_IPD_MAX_SIZE];
579 long long oti_alignment_lieutenant;
583 char oti_idx_ipd[DX_IPD_MAX_SIZE];
584 long long oti_alignment_lieutenant_colonel;
587 struct osd_idmap_cache oti_cache;
589 /* dedicated OI cache for insert (which needs inum) */
590 struct osd_idmap_cache *oti_ins_cache;
591 int oti_ins_cache_size;
592 int oti_ins_cache_used;
597 /** used in osd_fid_set() to put xattr */
598 struct lu_buf oti_buf;
599 struct lu_buf oti_big_buf;
600 /** used in osd_ea_fid_set() to set fid into common ea */
602 struct lustre_ost_attrs oti_ost_attrs;
603 struct filter_fid_old oti_ff;
604 struct filter_fid oti_ff_new;
607 struct osd_iobuf oti_iobuf;
608 /* used to access objects in /O */
609 struct inode *oti_inode;
610 #define OSD_FID_REC_SZ 32
611 char oti_ldp[OSD_FID_REC_SZ];
612 char oti_ldp2[OSD_FID_REC_SZ];
614 /* used by quota code */
616 #if defined(HAVE_DQUOT_QC_DQBLK)
617 struct qc_dqblk oti_qdq;
618 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
619 struct fs_disk_quota oti_fdq;
621 struct if_dqblk oti_dqblk;
623 struct if_dqinfo oti_dqinfo;
625 struct lquota_id_info oti_qi;
626 struct lquota_trans oti_quota_trans;
627 union lquota_rec oti_quota_rec;
629 struct lu_seq_range oti_seq_range;
631 /* Tracking for transaction credits, to allow debugging and optimizing
632 * cases where a large number of credits are being allocated for
633 * single transaction. */
634 unsigned int oti_credits_before;
635 unsigned int oti_declare_ops[OSD_OT_MAX];
636 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
637 unsigned int oti_declare_ops_used[OSD_OT_MAX];
638 struct osd_directory oti_iam;
641 extern int ldiskfs_pdo;
643 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
644 const char *name, void *buf, int len)
649 dentry->d_inode = inode;
650 dentry->d_sb = inode->i_sb;
651 return inode->i_op->getxattr(dentry, name, buf, len);
654 static inline int __osd_xattr_set(struct osd_thread_info *info,
655 struct inode *inode, const char *name,
656 const void *buf, int buflen, int fl)
658 struct dentry *dentry = &info->oti_child_dentry;
660 ll_vfs_dq_init(inode);
661 dentry->d_inode = inode;
662 dentry->d_sb = inode->i_sb;
663 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
666 #ifdef CONFIG_PROC_FS
668 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
669 extern struct lprocfs_vars lprocfs_osd_module_vars[];
670 int osd_procfs_init(struct osd_device *osd, const char *name);
671 int osd_procfs_fini(struct osd_device *osd);
672 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
673 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
674 int osd_register_proc_index_in_idif(struct osd_device *osd);
678 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
679 struct obd_statfs *sfs);
680 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
681 struct osd_inode_id *id);
683 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
684 struct osd_inode_id *id, struct lu_fid *fid);
685 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
686 const struct lu_fid *fid, __u32 compat, __u32 incompat);
687 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
688 struct dentry *dentry, struct lustre_ost_attrs *loa);
689 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
690 struct osd_inode_id *id, const struct lu_fid *fid);
691 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
692 struct dentry *dentry, struct lu_fid *fid);
694 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
695 void osd_obj_map_fini(struct osd_device *dev);
696 int osd_obj_map_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_map_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_map_delete(struct osd_thread_info *info, struct osd_device *osd,
702 const struct lu_fid *fid, handle_t *th);
703 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
704 const struct lu_fid *fid, const struct osd_inode_id *id,
706 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
707 struct inode *src_parent, struct dentry *src_child,
708 const struct lu_fid *fid);
709 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
710 const struct lu_fid *fid, struct osd_inode_id *id);
711 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
712 const struct lu_fid *fid, const struct osd_inode_id *id,
714 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
715 const struct lu_fid *fid, const struct osd_inode_id *id,
718 char *osd_lf_fid2name(const struct lu_fid *fid);
719 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
721 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
722 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
723 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
725 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
726 struct osd_inode_id *id);
727 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
729 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
730 u64 seq, struct lu_seq_range *range);
732 int osd_delete_from_remote_parent(const struct lu_env *env,
733 struct osd_device *osd,
734 struct osd_object *obj,
735 struct osd_thandle *oh, bool destroy);
736 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
737 struct osd_object *obj, struct osd_thandle *oh);
738 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
739 struct osd_device *osd,
740 const struct lu_fid *fid,
741 struct osd_inode_id *id);
743 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
745 /* osd_quota_fmt.c */
746 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
747 int type, uint blk, int depth, uint index,
748 struct osd_it_quota *it);
749 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
750 int type, uint blk, uint index,
751 struct osd_it_quota *it);
752 loff_t find_tree_dqentry(const struct lu_env *env,
753 struct osd_object *obj, int type,
754 qid_t dqid, uint blk, int depth,
755 struct osd_it_quota *it);
757 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
758 struct lquota_id_info *qi, struct osd_object *obj,
759 bool enforce, int *flags);
760 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
761 __u32 projid, long long space, struct osd_thandle *oh,
762 struct osd_object *obj, int *flags,
763 enum osd_qid_declare_flags);
764 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
765 const struct dt_rec *rec,
766 union lquota_rec *quota_rec);
767 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
769 #ifndef HAVE_I_UID_READ
770 static inline uid_t i_uid_read(const struct inode *inode)
775 static inline gid_t i_gid_read(const struct inode *inode)
780 static inline void i_uid_write(struct inode *inode, uid_t uid)
785 static inline void i_gid_write(struct inode *inode, gid_t gid)
791 #ifdef HAVE_PROJECT_QUOTA
792 static inline __u32 i_projid_read(struct inode *inode)
794 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
797 static inline void i_projid_write(struct inode *inode, __u32 projid)
800 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
801 LDISKFS_I(inode)->i_projid = kprojid;
804 static inline uid_t i_projid_read(struct inode *inode)
808 static inline void i_projid_write(struct inode *inode, __u32 projid)
814 #ifdef HAVE_LDISKFS_INFO_JINODE
815 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
816 #else /* HAVE_LDISKFS_INFO_JINODE */
817 # define osd_attach_jinode(inode) 0
818 #endif /* HAVE_LDISKFS_INFO_JINODE */
820 #ifdef LDISKFS_HT_MISC
821 # define osd_journal_start_sb(sb, type, nblock) \
822 ldiskfs_journal_start_sb(sb, type, nblock)
823 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
825 ldiskfs_lblk_t *nblock)
829 rc = osd_attach_jinode(inode);
832 return ldiskfs_append(handle, inode, nblock);
834 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
835 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
837 # define osd_journal_start(inode, type, nblocks) \
838 ldiskfs_journal_start(inode, type, nblocks)
839 # define osd_transaction_size(dev) \
840 (osd_journal(dev)->j_max_transaction_buffers / 2)
842 # define LDISKFS_HT_MISC 0
843 # define osd_journal_start_sb(sb, type, nblock) \
844 ldiskfs_journal_start_sb(sb, nblock)
846 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
848 ldiskfs_lblk_t *nblock)
850 struct buffer_head *bh;
853 bh = ldiskfs_append(handle, inode, nblock, &err);
860 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
861 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
863 # define osd_journal_start(inode, type, nblocks) \
864 ldiskfs_journal_start(inode, nblocks)
865 # define osd_transaction_size(dev) \
866 (osd_journal(dev)->j_max_transaction_buffers)
870 * Invariants, assertions.
874 * XXX: do not enable this, until invariant checking code is made thread safe
875 * in the face of pdirops locking.
877 #define OSD_INVARIANT_CHECKS (0)
879 #if OSD_INVARIANT_CHECKS
880 static inline int osd_invariant(const struct osd_object *obj)
884 ergo(obj->oo_inode != NULL,
885 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
886 atomic_read(&obj->oo_inode->i_count) > 0) &&
887 ergo(obj->oo_dir != NULL &&
888 obj->oo_dir->od_conationer.ic_object != NULL,
889 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
892 #define osd_invariant(obj) (1)
895 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
897 extern const struct dt_index_operations osd_otable_ops;
899 static inline int osd_oi_fid2idx(struct osd_device *dev,
900 const struct lu_fid *fid)
902 return fid->f_seq & (dev->od_oi_count - 1);
905 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
906 const struct lu_fid *fid)
908 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
909 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
910 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
911 DFID"\n", PFID(fid));
912 /* It can work even od_oi_count equals to 1 although it's unexpected,
913 * the only reason we set it to 1 is for performance measurement */
914 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
917 extern const struct lu_device_operations osd_lu_ops;
919 static inline int lu_device_is_osd(const struct lu_device *d)
921 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
924 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
926 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
927 return container_of0(d, struct osd_device, od_dt_dev);
930 static inline struct osd_device *osd_dev(const struct lu_device *d)
932 LASSERT(lu_device_is_osd(d));
933 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
936 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
938 return osd_dev(o->oo_dt.do_lu.lo_dev);
941 static inline struct super_block *osd_sb(const struct osd_device *dev)
943 return dev->od_mnt->mnt_sb;
946 static inline const char *osd_dev2name(const struct osd_device *dev)
948 return osd_sb(dev)->s_id;
951 static inline const char *osd_ino2name(const struct inode *inode)
953 return inode->i_sb->s_id;
957 * Put the osd object once done with it.
959 * \param obj osd object that needs to be put
961 static inline void osd_object_put(const struct lu_env *env,
962 struct osd_object *obj)
964 dt_object_put(env, &obj->oo_dt);
967 static inline int osd_object_is_root(const struct osd_object *obj)
969 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
972 static inline struct osd_object *osd_obj(const struct lu_object *o)
974 LASSERT(lu_device_is_osd(o->lo_dev));
975 return container_of0(o, struct osd_object, oo_dt.do_lu);
978 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
980 return osd_obj(&d->do_lu);
983 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
985 return &osd->od_dt_dev.dd_lu_dev;
988 static inline journal_t *osd_journal(const struct osd_device *dev)
990 return LDISKFS_SB(osd_sb(dev))->s_journal;
993 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
995 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
998 static inline char *osd_name(struct osd_device *osd)
1000 return osd->od_svname;
1003 static inline bool osd_is_ea_inode(struct inode *inode)
1005 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
1008 extern const struct dt_body_operations osd_body_ops;
1009 extern struct lu_context_key osd_key;
1011 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1013 return lu_context_key_get(&env->le_ctx, &osd_key);
1016 extern const struct dt_body_operations osd_body_ops_new;
1022 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1023 const struct iam_container *bag)
1025 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1026 osd_oti_get(env)->oti_it_ipd);
1030 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1031 const struct iam_container *bag)
1033 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1034 osd_oti_get(env)->oti_idx_ipd);
1037 static inline void osd_ipd_put(const struct lu_env *env,
1038 const struct iam_container *bag,
1039 struct iam_path_descr *ipd)
1041 bag->ic_descr->id_ops->id_ipd_free(ipd);
1044 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1045 const loff_t size, const loff_t pos,
1048 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1049 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1050 int write_NUL, loff_t *offs, handle_t *handle);
1053 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1054 struct inode *inode,
1055 const char *name, const int namelen)
1057 struct osd_thread_info *info = osd_oti_get(env);
1058 struct dentry *child_dentry = &info->oti_child_dentry;
1059 struct dentry *obj_dentry = &info->oti_obj_dentry;
1061 obj_dentry->d_inode = inode;
1062 obj_dentry->d_sb = inode->i_sb;
1063 obj_dentry->d_name.hash = 0;
1065 child_dentry->d_name.hash = 0;
1066 child_dentry->d_parent = obj_dentry;
1067 child_dentry->d_name.name = name;
1068 child_dentry->d_name.len = namelen;
1069 return child_dentry;
1072 extern int osd_trans_declare_op2rb[];
1073 extern int ldiskfs_track_declares_assert;
1074 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1076 static inline void osd_trans_declare_op(const struct lu_env *env,
1077 struct osd_thandle *oh,
1078 enum osd_op_type op, int credits)
1080 struct osd_thread_info *oti = osd_oti_get(env);
1082 LASSERT(oh->ot_handle == NULL);
1083 if (unlikely(op >= OSD_OT_MAX)) {
1084 if (unlikely(ldiskfs_track_declares_assert)) {
1085 LASSERT(op < OSD_OT_MAX);
1087 CWARN("%s: Invalid operation index %d\n",
1088 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1089 libcfs_debug_dumpstack(NULL);
1092 oti->oti_declare_ops[op]++;
1093 oti->oti_declare_ops_cred[op] += credits;
1095 oh->ot_credits += credits;
1098 static inline void osd_trans_exec_op(const struct lu_env *env,
1100 enum osd_op_type op)
1102 struct osd_thread_info *oti = osd_oti_get(env);
1103 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1105 unsigned int rb, left;
1107 LASSERT(oh->ot_handle != NULL);
1108 if (unlikely(op >= OSD_OT_MAX)) {
1109 if (unlikely(ldiskfs_track_declares_assert))
1110 LASSERT(op < OSD_OT_MAX);
1112 CWARN("%s: opcode %u: invalid value >= %u\n",
1113 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1115 libcfs_debug_dumpstack(NULL);
1120 /* find rollback (or reverse) operation for the given one
1121 * such an operation doesn't require additional credits
1122 * as the same set of blocks are modified */
1123 rb = osd_trans_declare_op2rb[op];
1125 /* check whether credits for this operation were reserved at all */
1126 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1127 oti->oti_declare_ops_cred[rb] == 0)) {
1128 /* the API is not perfect yet: CREATE does REF_ADD internally
1129 * while DESTROY does not. To rollback CREATE the callers
1130 * needs to call REF_DEL+DESTROY which is hard to detect using
1131 * a simple table of rollback operations */
1132 if (op == OSD_OT_REF_DEL &&
1133 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1135 if (op == OSD_OT_REF_ADD &&
1136 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1138 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1139 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1140 osd_trans_dump_creds(env, th);
1141 LASSERT(!ldiskfs_track_declares_assert);
1145 /* remember how many credits we have unused before the operation */
1146 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1147 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1148 if (unlikely(oti->oti_credits_before < left)) {
1149 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1150 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1151 oti->oti_credits_before, left, rb);
1152 osd_trans_dump_creds(env, th);
1153 /* on a very small fs (testing?) it's possible that
1154 * the transaction can't fit 1/4 of journal, so we
1155 * just request less credits (see osd_trans_start()).
1156 * ignore the same case here */
1157 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1158 if (unlikely(oh->ot_credits < rb))
1159 LASSERT(!ldiskfs_track_declares_assert);
1163 static inline void osd_trans_exec_check(const struct lu_env *env,
1165 enum osd_op_type op)
1167 struct osd_thread_info *oti = osd_oti_get(env);
1168 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1170 int used, over, quota;
1172 /* how many credits have been used by the operation */
1173 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1175 if (unlikely(used < 0)) {
1176 /* if some block was allocated and released in the same
1177 * transaction, then it won't be a part of the transaction
1178 * and delta can be negative */
1183 /* rollback operations (e.g. when we destroy just created
1184 * object) should not consume any credits. there is no point
1185 * to confuse the checks below */
1189 oti->oti_declare_ops_used[op] += used;
1190 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1193 /* we account quota for a whole transaction and any operation can
1194 * consume corresponding credits */
1195 over = oti->oti_declare_ops_used[op] -
1196 oti->oti_declare_ops_cred[op];
1197 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1198 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1199 if (over <= quota) {
1200 /* probably that credits were consumed by
1201 * quota indirectly (in the depths of ldiskfs) */
1202 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1203 oti->oti_declare_ops_used[op] -= over;
1205 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1206 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1207 oti->oti_declare_ops_used[op], used,
1208 oti->oti_declare_ops_cred[op]);
1209 osd_trans_dump_creds(env, th);
1210 if (unlikely(ldiskfs_track_declares_assert))
1216 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1219 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1220 struct lu_fid *befider)
1222 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1223 memcpy(pack->fp_area, befider, sizeof(*befider));
1224 pack->fp_len = sizeof(*befider) + 1;
1228 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1233 switch (pack->fp_len) {
1234 case sizeof *fid + 1:
1235 memcpy(fid, pack->fp_area, sizeof *fid);
1236 fid_be_to_cpu(fid, fid);
1239 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1246 * Quota/Accounting handling
1248 extern const struct dt_index_operations osd_acct_index_ops;
1249 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1250 const struct lu_fid *fid, struct osd_inode_id *id);
1252 /* copy from fs/ext4/dir.c */
1253 static inline int is_32bit_api(void)
1255 #ifdef CONFIG_COMPAT
1256 return in_compat_syscall();
1258 return (BITS_PER_LONG == 32);
1262 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1264 if ((filp->f_mode & FMODE_32BITHASH) ||
1265 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1266 return LDISKFS_HTREE_EOF_32BIT;
1268 return LDISKFS_HTREE_EOF_64BIT;
1271 static inline int fid_is_internal(const struct lu_fid *fid)
1273 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1276 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1278 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1282 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1283 * return code has been removed and integrated into the pointer in the
1286 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1287 struct inode *inode,
1288 ldiskfs_lblk_t block,
1292 struct buffer_head *bh;
1295 rc = osd_attach_jinode(inode);
1299 #ifdef HAVE_EXT4_BREAD_4ARGS
1300 bh = ldiskfs_bread(handle, inode, block, create);
1303 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1304 if (bh == NULL && rc != 0)
1310 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1311 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1313 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1316 osd_index_register(struct osd_device *osd, const struct lu_fid *fid,
1317 __u32 keysize, __u32 recsize)
1319 return lustre_index_register(&osd->od_dt_dev, osd_name(osd),
1320 &osd->od_index_backup_list, &osd->od_lock,
1321 &osd->od_index_backup_stop,
1322 fid, keysize, recsize);
1326 osd_index_backup(const struct lu_env *env, struct osd_device *osd, bool backup)
1328 struct osd_thread_info *info = osd_oti_get(env);
1329 struct lu_fid *fid = &info->oti_fid3;
1330 struct osd_inode_id *id = &info->oti_id3;
1332 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
1333 osd_id_gen(id, osd->od_index_backup_inode->i_ino,
1334 osd->od_index_backup_inode->i_generation);
1335 osd_add_oi_cache(info, osd, id, fid);
1337 lustre_index_backup(env, &osd->od_dt_dev, osd_name(osd),
1338 &osd->od_index_backup_list, &osd->od_lock,
1339 &osd->od_index_backup_stop, backup);
1342 #endif /* _OSD_INTERNAL_H */