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 <linux/bio.h>
53 #include <ldiskfs/ldiskfs.h>
54 #include <ldiskfs/ldiskfs_jbd2.h>
58 /* class_register_type(), class_unregister_type(), class_get_type() */
59 #include <obd_class.h>
60 #include <lustre_disk.h>
61 #include <dt_object.h>
62 #include <lustre_quota.h>
66 #include "osd_scrub.h"
67 #include "osd_quota_fmt.h"
70 extern struct kmem_cache *dynlock_cachep;
72 #define OSD_COUNTERS (0)
74 /* ldiskfs special inode::i_state_flags need to be accessed with
75 * ldiskfs_{set,clear,test}_inode_state() only */
77 /* OI scrub should skip this inode. */
78 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
79 #define LDISKFS_STATE_LUSTRE_DESTROY 30
81 /** Enable thandle usage statistics */
82 #define OSD_THANDLE_STATS (0)
84 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
86 #define OBJECTS "OBJECTS"
87 #define ADMIN_USR "admin_quotafile_v2.usr"
88 #define ADMIN_GRP "admin_quotafile_v2.grp"
90 /* Statfs space reservation for fragmentation and local objects */
91 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
92 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
94 /* check if ldiskfs support project quota */
95 #ifndef LDISKFS_IOC_FSSETXATTR
96 #undef HAVE_PROJECT_QUOTA
99 #define OBD_BRW_MAPPED OBD_BRW_LOCAL1
101 struct osd_directory {
102 struct iam_container od_container;
103 struct iam_descr od_descr;
107 * Object Index (oi) instance.
111 * underlying index object, where fid->id mapping in stored.
113 struct inode *oi_inode;
114 struct osd_directory oi_dir;
117 extern const int osd_dto_credits_noquota[];
120 struct dt_object oo_dt;
122 * Inode for file system object represented by this osd_object. This
123 * inode is pinned for the whole duration of lu_object life.
125 * Not modified concurrently (either setup early during object
126 * creation, or assigned by osd_create() under write lock).
128 struct inode *oo_inode;
130 * to protect index ops.
132 struct htree_lock_head *oo_hl_head;
133 struct rw_semaphore oo_ext_idx_sem;
134 struct rw_semaphore oo_sem;
135 struct osd_directory *oo_dir;
136 /** protects inode attributes. */
140 * Following two members *compat_dot* are used to indicate
141 * the presence of dot and dotdot in the given directory.
142 * This is required for interop mode (b11826).
144 __u32 oo_destroyed:1,
146 oo_compat_dot_created:1,
147 oo_compat_dotdot_created:1;
149 /* the i_flags in LMA */
152 const struct lu_env *oo_owner;
154 struct list_head oo_xattr_list;
155 struct lu_object_header *oo_header;
159 /* protects on-fly initialization */
160 int oos_subdir_count; /* subdir count for each seq */
161 struct dentry *oos_root; /* O/<seq> */
162 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
163 u64 oos_seq; /* seq number */
164 struct list_head oos_seq_list; /* list to seq_list */
168 struct dentry *om_root; /* dentry for /O */
169 rwlock_t om_seq_list_lock; /* lock for seq_list */
170 struct list_head om_seq_list; /* list head for seq */
172 struct mutex om_dir_init_mutex;
176 struct dentry *om_root; /* AGENT/<index> */
177 u64 om_index; /* mdt index */
178 struct list_head om_list; /* list to omm_list */
181 struct osd_mdobj_map {
182 struct dentry *omm_remote_parent;
184 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
185 handle_t *handle, struct dentry *child,
186 struct inode *inode, struct htree_lock *hlock);
188 #define OSD_OTABLE_IT_CACHE_SIZE 64
189 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
191 struct osd_inconsistent_item {
192 /* link into lustre_scrub::os_inconsistent_items,
193 * protected by lustre_scrub::os_lock. */
194 struct list_head oii_list;
196 /* The right FID <=> ino#/gen mapping. */
197 struct osd_idmap_cache oii_cache;
199 unsigned int oii_insert:1; /* insert or update mapping. */
202 struct osd_otable_cache {
203 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
205 /* Index for next cache slot to be filled. */
206 int ooc_producer_idx;
208 /* Index for next cache slot to be returned by it::next(). */
209 int ooc_consumer_idx;
211 /* How many items in ooc_cache. */
212 __u64 ooc_cached_items;
214 /* Position for up layer LFSCK iteration pre-loading. */
215 __u64 ooc_pos_preload;
218 struct osd_otable_it {
219 struct osd_device *ooi_dev;
220 struct osd_otable_cache ooi_cache;
221 struct osd_iit_param ooi_iit_param;
223 /* The following bits can be updated/checked w/o lock protection.
224 * If more bits will be introduced in the future and need lock to
225 * protect, please add comment. */
226 unsigned long ooi_used_outside:1, /* Some user out of OSD
227 * uses the iteration. */
228 ooi_all_cached:1, /* No more entries can be
229 * filled into cache. */
230 ooi_user_ready:1, /* The user out of OSD is
231 * ready to iterate. */
232 ooi_waiting:1; /* it::next is waiting. */
235 struct osd_obj_orphan {
236 struct list_head oor_list;
237 struct lu_env *oor_env; /* to identify "own" records */
246 struct dt_device od_dt_dev;
247 /* information about underlying file system */
248 struct vfsmount *od_mnt;
250 struct osd_oi **od_oi_table;
251 /* total number of OI containers */
256 unsigned int od_fl_capa:1,
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;
312 struct list_head od_index_backup_list;
313 struct list_head od_index_restore_list;
315 struct inode *od_index_backup_inode;
316 enum lustre_index_backup_policy od_index_backup_policy;
317 int od_index_backup_stop;
320 enum osd_full_scrub_ratio {
321 /* Trigger OI scrub to scan the whole device directly. */
324 /* Because the bad OI mappings count cannot be larger than
325 * the total OI mappints count, then setting OFSR_NEVER means
326 * that the whole device scanning cannot be triggered by auto
327 * detected bad OI mappings during the RPC services. */
329 OFSR_DEFAULT = 10000,
332 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
334 /* There are at most 15 uid/gid/projids are affected in a transaction, and
335 * that's rename case:
336 * - 3 for source parent uid & gid & projid;
337 * - 3 for source child uid & gid & projid ('..' entry update when
338 * child is directory);
339 * - 3 for target parent uid & gid & projid;
340 * - 3 for target child uid & gid & projid(if the target child exists);
341 * - 3 for root uid & gid(last_rcvd, llog, etc);
344 #define OSD_MAX_UGID_CNT 15
349 OSD_OT_XATTR_SET = 2,
361 struct osd_access_lock {
362 struct list_head tl_list;
363 struct osd_object *tl_obj;
369 struct thandle ot_super;
371 struct ldiskfs_journal_cb_entry ot_jcb;
372 struct list_head ot_commit_dcb_list;
373 struct list_head ot_stop_dcb_list;
374 /* Link to the device, for debugging. */
375 struct lu_ref_link ot_dev_link;
376 unsigned int ot_credits;
377 unsigned short ot_id_cnt;
378 __u8 ot_id_types[OSD_MAX_UGID_CNT];
379 unsigned int ot_remove_agents:1;
380 uid_t ot_id_array[OSD_MAX_UGID_CNT];
381 struct lquota_trans *ot_quota_trans;
382 #if OSD_THANDLE_STATS
383 /** time when this handle was allocated */
386 /** time when this thanle was started */
389 struct list_head ot_trunc_locks;
393 * Basic transaction credit op
414 #ifdef CONFIG_PROC_FS
416 LPROC_OSD_READ_BYTES = 0,
417 LPROC_OSD_WRITE_BYTES = 1,
418 LPROC_OSD_GET_PAGE = 2,
419 LPROC_OSD_NO_PAGE = 3,
420 LPROC_OSD_CACHE_ACCESS = 4,
421 LPROC_OSD_CACHE_HIT = 5,
422 LPROC_OSD_CACHE_MISS = 6,
424 #if OSD_THANDLE_STATS
425 LPROC_OSD_THANDLE_STARTING,
426 LPROC_OSD_THANDLE_OPEN,
427 LPROC_OSD_THANDLE_CLOSING,
434 * Storage representation for fids.
436 * Variable size, first byte contains the length of the whole record.
438 struct osd_fid_pack {
439 unsigned char fp_len;
440 char fp_area[sizeof(struct lu_fid)];
443 struct osd_it_ea_dirent {
444 struct lu_fid oied_fid;
447 unsigned short oied_namelen;
448 unsigned int oied_type;
450 } __attribute__((packed));
453 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
454 * than lu_dirent struct. osd readdir reads less number of dirent than
455 * required for mdd dir page. so buffer size need to be increased so that
456 * there would be one ext3 readdir for every mdd readdir page.
459 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
462 * This is iterator's in-memory data structure in interoperability
463 * mode (i.e. iterator over ldiskfs style directory)
466 struct osd_object *oie_obj;
467 /** used in ldiskfs iterator, to stored file pointer */
468 struct file oie_file;
469 /** how many entries have been read-cached from storage */
471 /** current entry is being iterated by caller */
473 /** current processing entry */
474 struct osd_it_ea_dirent *oie_dirent;
475 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
477 struct dentry oie_dentry;
481 * Iterator's in-memory data structure for IAM mode.
484 struct osd_object *oi_obj;
485 struct iam_path_descr *oi_ipd;
486 struct iam_iterator oi_it;
489 struct osd_quota_leaf {
490 struct list_head oql_link;
495 * Iterator's in-memory data structure for quota file.
497 struct osd_it_quota {
498 struct osd_object *oiq_obj;
499 /** tree blocks path to where the entry is stored */
500 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
501 /** on-disk offset for current key where quota record can be found */
503 /** identifier for current quota record */
505 /** the record index in the leaf/index block */
506 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
507 /** list of already processed leaf blocks */
508 struct list_head oiq_list;
511 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
514 wait_queue_head_t dr_wait;
515 atomic_t dr_numreqs; /* number of reqs being processed */
520 unsigned int dr_elapsed_valid:1; /* we really did count time */
521 unsigned int dr_rw:1;
522 struct lu_buf dr_pg_buf;
523 struct page **dr_pages;
524 struct lu_buf dr_bl_buf;
526 ktime_t dr_start_time;
527 ktime_t dr_elapsed; /* how long io took */
528 struct osd_device *dr_dev;
529 unsigned int dr_init_at; /* the line iobuf was initialized */
532 #define OSD_INS_CACHE_SIZE 8
534 struct osd_thread_info {
535 const struct lu_env *oti_env;
537 * used for index operations.
539 struct dentry oti_obj_dentry;
540 struct dentry oti_child_dentry;
542 /** dentry for Iterator context. */
543 struct dentry oti_it_dentry;
546 /* fake struct file for osd_object_sync */
547 struct file oti_file;
549 struct kstatfs oti_ksfs;
552 struct htree_lock *oti_hlock;
554 struct lu_fid oti_fid;
555 struct lu_fid oti_fid2;
556 struct lu_fid oti_fid3;
557 struct osd_inode_id oti_id;
558 struct osd_inode_id oti_id2;
559 struct osd_inode_id oti_id3;
560 struct ost_id oti_ostid;
563 * XXX temporary: for ->i_op calls.
565 struct timespec oti_time;
568 * following ipd and it structures are used for osd_index_iam_lookup()
569 * these are defined separately as we might do index operation
570 * in open iterator session.
573 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
575 unsigned int oti_it_ea_buf_used:1;
577 /* IAM iterator for index operation. */
578 struct iam_iterator oti_idx_it;
580 /** union to guarantee that ->oti_ipd[] has proper alignment. */
583 char oti_it_ipd[DX_IPD_MAX_SIZE];
584 long long oti_alignment_lieutenant;
588 char oti_idx_ipd[DX_IPD_MAX_SIZE];
589 long long oti_alignment_lieutenant_colonel;
592 struct osd_idmap_cache oti_cache;
594 /* dedicated OI cache for insert (which needs inum) */
595 struct osd_idmap_cache *oti_ins_cache;
596 int oti_ins_cache_size;
597 int oti_ins_cache_used;
598 /* inc by osd_trans_create and dec by osd_trans_stop */
599 int oti_ins_cache_depth;
604 /** used in osd_fid_set() to put xattr */
605 struct lu_buf oti_buf;
606 struct lu_buf oti_big_buf;
607 /** used in osd_ea_fid_set() to set fid into common ea */
609 struct lustre_ost_attrs oti_ost_attrs;
610 struct filter_fid_old oti_ff;
611 struct filter_fid oti_ff_new;
614 struct osd_iobuf oti_iobuf;
615 /* used to access objects in /O */
616 struct inode *oti_inode;
617 #define OSD_FID_REC_SZ 32
618 char oti_ldp[OSD_FID_REC_SZ];
619 char oti_ldp2[OSD_FID_REC_SZ];
621 /* used by quota code */
623 #if defined(HAVE_DQUOT_QC_DQBLK)
624 struct qc_dqblk oti_qdq;
625 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
626 struct fs_disk_quota oti_fdq;
628 struct if_dqblk oti_dqblk;
630 struct if_dqinfo oti_dqinfo;
632 struct lquota_id_info oti_qi;
633 struct lquota_trans oti_quota_trans;
634 union lquota_rec oti_quota_rec;
636 struct lu_seq_range oti_seq_range;
638 /* Tracking for transaction credits, to allow debugging and optimizing
639 * cases where a large number of credits are being allocated for
640 * single transaction. */
641 unsigned int oti_credits_before;
642 unsigned int oti_declare_ops[OSD_OT_MAX];
643 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
644 unsigned int oti_declare_ops_used[OSD_OT_MAX];
645 struct osd_directory oti_iam;
648 extern int ldiskfs_pdo;
650 #ifndef HAVE_VFS_SETXATTR
651 #define osd_setxattr(dentry, inode, name, buf, len, flag) \
652 ((inode)->i_op->setxattr(dentry, name, buf, len, flag))
653 #define osd_getxattr(dentry, inode, name, buf, len) \
654 ((inode)->i_op->getxattr(dentry, name, buf, len))
655 #define osd_removexattr(dentry, inode, name) \
656 ((inode)->i_op->removexattr(dentry, name))
657 #else /* HAVE_VFS_SETXATTR */
658 #define osd_setxattr(dentry, inode, name, buf, len, flag) \
659 __vfs_setxattr(dentry, inode, name, buf, len, flag)
660 #define osd_getxattr(dentry, inode, name, buf, len) \
661 __vfs_getxattr(dentry, inode, name, buf, len)
662 #define osd_removexattr(dentry, inode, name) \
663 __vfs_removexattr(dentry, name)
664 #endif /* !HAVE_VFS_SETXATTR */
666 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
667 const char *name, void *buf, int len)
672 dentry->d_inode = inode;
673 dentry->d_sb = inode->i_sb;
674 return osd_getxattr(dentry, inode, name, buf, len);
677 static inline int __osd_xattr_set(struct osd_thread_info *info,
678 struct inode *inode, const char *name,
679 const void *buf, int buflen, int fl)
681 struct dentry *dentry = &info->oti_child_dentry;
683 ll_vfs_dq_init(inode);
684 dentry->d_inode = inode;
685 dentry->d_sb = inode->i_sb;
686 return osd_setxattr(dentry, inode, name, buf, buflen, fl);
689 #ifdef CONFIG_PROC_FS
691 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
692 extern struct lprocfs_vars lprocfs_osd_module_vars[];
693 int osd_procfs_init(struct osd_device *osd, const char *name);
694 int osd_procfs_fini(struct osd_device *osd);
695 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
696 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
697 int osd_register_proc_index_in_idif(struct osd_device *osd);
701 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
702 struct obd_statfs *sfs);
703 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
704 struct osd_inode_id *id);
706 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
707 struct osd_inode_id *id, struct lu_fid *fid);
708 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
709 const struct lu_fid *fid, __u32 compat, __u32 incompat);
710 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
711 struct dentry *dentry, struct lustre_ost_attrs *loa);
712 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
713 struct osd_inode_id *id, const struct lu_fid *fid);
714 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
715 struct dentry *dentry, struct lu_fid *fid);
717 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
718 void osd_obj_map_fini(struct osd_device *dev);
719 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
720 const struct lu_fid *fid, struct osd_inode_id *id);
721 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
722 const struct lu_fid *fid, const struct osd_inode_id *id,
724 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
725 const struct lu_fid *fid, handle_t *th);
726 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
727 const struct lu_fid *fid, const struct osd_inode_id *id,
729 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
730 struct inode *src_parent, struct dentry *src_child,
731 const struct lu_fid *fid);
732 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
733 const struct lu_fid *fid, struct osd_inode_id *id);
734 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
735 const struct lu_fid *fid, const struct osd_inode_id *id,
737 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
738 const struct lu_fid *fid, const struct osd_inode_id *id,
741 char *osd_lf_fid2name(const struct lu_fid *fid);
742 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
744 void osd_scrub_stop(struct osd_device *dev);
745 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
746 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
747 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
749 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
750 struct osd_inode_id *id);
751 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
753 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
754 u64 seq, struct lu_seq_range *range);
756 int osd_delete_from_remote_parent(const struct lu_env *env,
757 struct osd_device *osd,
758 struct osd_object *obj,
759 struct osd_thandle *oh, bool destroy);
760 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
761 struct osd_object *obj, struct osd_thandle *oh);
762 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
763 struct osd_device *osd,
764 const struct lu_fid *fid,
765 struct osd_inode_id *id);
767 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
769 /* osd_quota_fmt.c */
770 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
771 int type, uint blk, int depth, uint index,
772 struct osd_it_quota *it);
773 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
774 int type, uint blk, uint index,
775 struct osd_it_quota *it);
776 loff_t find_tree_dqentry(const struct lu_env *env,
777 struct osd_object *obj, int type,
778 qid_t dqid, uint blk, int depth,
779 struct osd_it_quota *it);
781 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
782 struct lquota_id_info *qi, struct osd_object *obj,
783 bool enforce, int *flags);
784 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
785 __u32 projid, long long space, struct osd_thandle *oh,
786 struct osd_object *obj, int *flags,
787 enum osd_qid_declare_flags);
788 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
789 const struct dt_rec *rec,
790 union lquota_rec *quota_rec);
791 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
793 #ifndef HAVE_I_UID_READ
794 static inline uid_t i_uid_read(const struct inode *inode)
799 static inline gid_t i_gid_read(const struct inode *inode)
804 static inline void i_uid_write(struct inode *inode, uid_t uid)
809 static inline void i_gid_write(struct inode *inode, gid_t gid)
815 #ifdef HAVE_PROJECT_QUOTA
816 static inline __u32 i_projid_read(struct inode *inode)
818 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
821 static inline void i_projid_write(struct inode *inode, __u32 projid)
824 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
825 LDISKFS_I(inode)->i_projid = kprojid;
828 static inline uid_t i_projid_read(struct inode *inode)
832 static inline void i_projid_write(struct inode *inode, __u32 projid)
838 #ifdef HAVE_LDISKFS_INFO_JINODE
839 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
840 #else /* HAVE_LDISKFS_INFO_JINODE */
841 # define osd_attach_jinode(inode) 0
842 #endif /* HAVE_LDISKFS_INFO_JINODE */
844 #ifdef LDISKFS_HT_MISC
845 # define osd_journal_start_sb(sb, type, nblock) \
846 ldiskfs_journal_start_sb(sb, type, nblock)
847 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
849 ldiskfs_lblk_t *nblock)
853 rc = osd_attach_jinode(inode);
856 return ldiskfs_append(handle, inode, nblock);
858 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
859 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
861 # define osd_journal_start(inode, type, nblocks) \
862 ldiskfs_journal_start(inode, type, nblocks)
863 # define osd_transaction_size(dev) \
864 (osd_journal(dev)->j_max_transaction_buffers / 2)
866 # define LDISKFS_HT_MISC 0
867 # define osd_journal_start_sb(sb, type, nblock) \
868 ldiskfs_journal_start_sb(sb, nblock)
870 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
872 ldiskfs_lblk_t *nblock)
874 struct buffer_head *bh;
877 bh = ldiskfs_append(handle, inode, nblock, &err);
884 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
885 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
887 # define osd_journal_start(inode, type, nblocks) \
888 ldiskfs_journal_start(inode, nblocks)
889 # define osd_transaction_size(dev) \
890 (osd_journal(dev)->j_max_transaction_buffers)
894 * Invariants, assertions.
898 * XXX: do not enable this, until invariant checking code is made thread safe
899 * in the face of pdirops locking.
901 #define OSD_INVARIANT_CHECKS (0)
903 #if OSD_INVARIANT_CHECKS
904 static inline int osd_invariant(const struct osd_object *obj)
908 ergo(obj->oo_inode != NULL,
909 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
910 atomic_read(&obj->oo_inode->i_count) > 0) &&
911 ergo(obj->oo_dir != NULL &&
912 obj->oo_dir->od_conationer.ic_object != NULL,
913 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
916 #define osd_invariant(obj) (1)
919 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
921 extern const struct dt_index_operations osd_otable_ops;
923 static inline int osd_oi_fid2idx(struct osd_device *dev,
924 const struct lu_fid *fid)
926 return fid->f_seq & (dev->od_oi_count - 1);
929 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
930 const struct lu_fid *fid)
932 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
933 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
934 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
935 DFID"\n", PFID(fid));
936 /* It can work even od_oi_count equals to 1 although it's unexpected,
937 * the only reason we set it to 1 is for performance measurement */
938 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
941 extern const struct lu_device_operations osd_lu_ops;
943 static inline int lu_device_is_osd(const struct lu_device *d)
945 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
948 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
950 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
951 return container_of0(d, struct osd_device, od_dt_dev);
954 static inline struct osd_device *osd_dev(const struct lu_device *d)
956 LASSERT(lu_device_is_osd(d));
957 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
960 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
962 return osd_dev(o->oo_dt.do_lu.lo_dev);
965 static inline struct super_block *osd_sb(const struct osd_device *dev)
967 return dev->od_mnt->mnt_sb;
970 static inline const char *osd_dev2name(const struct osd_device *dev)
972 return osd_sb(dev)->s_id;
975 static inline const char *osd_ino2name(const struct inode *inode)
977 return inode->i_sb->s_id;
981 * Put the osd object once done with it.
983 * \param obj osd object that needs to be put
985 static inline void osd_object_put(const struct lu_env *env,
986 struct osd_object *obj)
988 dt_object_put(env, &obj->oo_dt);
991 static inline int osd_object_is_root(const struct osd_object *obj)
993 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
996 static inline struct osd_object *osd_obj(const struct lu_object *o)
998 LASSERT(lu_device_is_osd(o->lo_dev));
999 return container_of0(o, struct osd_object, oo_dt.do_lu);
1002 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
1004 return osd_obj(&d->do_lu);
1007 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
1009 return &osd->od_dt_dev.dd_lu_dev;
1012 static inline journal_t *osd_journal(const struct osd_device *dev)
1014 return LDISKFS_SB(osd_sb(dev))->s_journal;
1017 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
1019 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
1022 static inline char *osd_name(struct osd_device *osd)
1024 return osd->od_svname;
1027 static inline bool osd_is_ea_inode(struct inode *inode)
1029 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
1032 extern const struct dt_body_operations osd_body_ops;
1033 extern struct lu_context_key osd_key;
1035 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1037 return lu_context_key_get(&env->le_ctx, &osd_key);
1040 extern const struct dt_body_operations osd_body_ops_new;
1046 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1047 const struct iam_container *bag)
1049 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1050 osd_oti_get(env)->oti_it_ipd);
1054 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1055 const struct iam_container *bag)
1057 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1058 osd_oti_get(env)->oti_idx_ipd);
1061 static inline void osd_ipd_put(const struct lu_env *env,
1062 const struct iam_container *bag,
1063 struct iam_path_descr *ipd)
1065 bag->ic_descr->id_ops->id_ipd_free(ipd);
1068 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1069 const loff_t size, const loff_t pos,
1072 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1073 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1074 int write_NUL, loff_t *offs, handle_t *handle);
1077 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1078 struct inode *inode,
1079 const char *name, const int namelen)
1081 struct osd_thread_info *info = osd_oti_get(env);
1082 struct dentry *child_dentry = &info->oti_child_dentry;
1083 struct dentry *obj_dentry = &info->oti_obj_dentry;
1085 obj_dentry->d_inode = inode;
1086 obj_dentry->d_sb = inode->i_sb;
1087 obj_dentry->d_name.hash = 0;
1089 child_dentry->d_name.hash = 0;
1090 child_dentry->d_parent = obj_dentry;
1091 child_dentry->d_name.name = name;
1092 child_dentry->d_name.len = namelen;
1093 return child_dentry;
1096 extern int osd_trans_declare_op2rb[];
1097 extern int ldiskfs_track_declares_assert;
1098 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1100 static inline void osd_trans_declare_op(const struct lu_env *env,
1101 struct osd_thandle *oh,
1102 enum osd_op_type op, int credits)
1104 struct osd_thread_info *oti = osd_oti_get(env);
1106 LASSERT(oh->ot_handle == NULL);
1107 if (unlikely(op >= OSD_OT_MAX)) {
1108 if (unlikely(ldiskfs_track_declares_assert)) {
1109 LASSERT(op < OSD_OT_MAX);
1111 CWARN("%s: Invalid operation index %d\n",
1112 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1113 libcfs_debug_dumpstack(NULL);
1116 oti->oti_declare_ops[op]++;
1117 oti->oti_declare_ops_cred[op] += credits;
1119 oh->ot_credits += credits;
1122 static inline void osd_trans_exec_op(const struct lu_env *env,
1124 enum osd_op_type op)
1126 struct osd_thread_info *oti = osd_oti_get(env);
1127 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1129 unsigned int rb, left;
1131 LASSERT(oh->ot_handle != NULL);
1132 if (unlikely(op >= OSD_OT_MAX)) {
1133 if (unlikely(ldiskfs_track_declares_assert))
1134 LASSERT(op < OSD_OT_MAX);
1136 CWARN("%s: opcode %u: invalid value >= %u\n",
1137 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1139 libcfs_debug_dumpstack(NULL);
1144 /* find rollback (or reverse) operation for the given one
1145 * such an operation doesn't require additional credits
1146 * as the same set of blocks are modified */
1147 rb = osd_trans_declare_op2rb[op];
1149 /* check whether credits for this operation were reserved at all */
1150 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1151 oti->oti_declare_ops_cred[rb] == 0)) {
1152 /* the API is not perfect yet: CREATE does REF_ADD internally
1153 * while DESTROY does not. To rollback CREATE the callers
1154 * needs to call REF_DEL+DESTROY which is hard to detect using
1155 * a simple table of rollback operations */
1156 if (op == OSD_OT_REF_DEL &&
1157 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1159 if (op == OSD_OT_REF_ADD &&
1160 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1162 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1163 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1164 osd_trans_dump_creds(env, th);
1165 LASSERT(!ldiskfs_track_declares_assert);
1169 /* remember how many credits we have unused before the operation */
1170 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1171 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1172 if (unlikely(oti->oti_credits_before < left)) {
1173 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1174 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1175 oti->oti_credits_before, left, rb);
1176 osd_trans_dump_creds(env, th);
1177 /* on a very small fs (testing?) it's possible that
1178 * the transaction can't fit 1/4 of journal, so we
1179 * just request less credits (see osd_trans_start()).
1180 * ignore the same case here */
1181 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1182 if (unlikely(oh->ot_credits < rb))
1183 LASSERT(!ldiskfs_track_declares_assert);
1187 static inline void osd_trans_exec_check(const struct lu_env *env,
1189 enum osd_op_type op)
1191 struct osd_thread_info *oti = osd_oti_get(env);
1192 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1194 int used, over, quota;
1196 /* how many credits have been used by the operation */
1197 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1199 if (unlikely(used < 0)) {
1200 /* if some block was allocated and released in the same
1201 * transaction, then it won't be a part of the transaction
1202 * and delta can be negative */
1207 /* rollback operations (e.g. when we destroy just created
1208 * object) should not consume any credits. there is no point
1209 * to confuse the checks below */
1213 oti->oti_declare_ops_used[op] += used;
1214 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1217 /* we account quota for a whole transaction and any operation can
1218 * consume corresponding credits */
1219 over = oti->oti_declare_ops_used[op] -
1220 oti->oti_declare_ops_cred[op];
1221 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1222 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1223 if (over <= quota) {
1224 /* probably that credits were consumed by
1225 * quota indirectly (in the depths of ldiskfs) */
1226 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1227 oti->oti_declare_ops_used[op] -= over;
1229 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1230 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1231 oti->oti_declare_ops_used[op], used,
1232 oti->oti_declare_ops_cred[op]);
1233 osd_trans_dump_creds(env, th);
1234 if (unlikely(ldiskfs_track_declares_assert))
1240 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1243 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1244 struct lu_fid *befider)
1246 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1247 memcpy(pack->fp_area, befider, sizeof(*befider));
1248 pack->fp_len = sizeof(*befider) + 1;
1252 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1257 switch (pack->fp_len) {
1258 case sizeof *fid + 1:
1259 memcpy(fid, pack->fp_area, sizeof *fid);
1260 fid_be_to_cpu(fid, fid);
1263 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1270 * Quota/Accounting handling
1272 extern const struct dt_index_operations osd_acct_index_ops;
1273 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1274 const struct lu_fid *fid, struct osd_inode_id *id);
1276 /* copy from fs/ext4/dir.c */
1277 static inline int is_32bit_api(void)
1279 #ifdef CONFIG_COMPAT
1280 return in_compat_syscall();
1282 return (BITS_PER_LONG == 32);
1286 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1288 if ((filp->f_mode & FMODE_32BITHASH) ||
1289 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1290 return LDISKFS_HTREE_EOF_32BIT;
1292 return LDISKFS_HTREE_EOF_64BIT;
1295 static inline int fid_is_internal(const struct lu_fid *fid)
1297 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1300 static inline bool is_remote_parent_ino(struct osd_device *o, unsigned long ino)
1305 LASSERT(o->od_mdt_map != NULL);
1307 return ino == o->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1311 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1312 * return code has been removed and integrated into the pointer in the
1315 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1316 struct inode *inode,
1317 ldiskfs_lblk_t block,
1321 struct buffer_head *bh;
1324 rc = osd_attach_jinode(inode);
1328 #ifdef HAVE_EXT4_BREAD_4ARGS
1329 bh = ldiskfs_bread(handle, inode, block, create);
1332 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1333 if (bh == NULL && rc != 0)
1339 #ifndef HAVE_BIO_INTEGRITY_ENABLED
1340 bool bio_integrity_enabled(struct bio *bio);
1344 # define bio_set_dev(bio, bdev) ((bio)->bi_bdev = bdev)
1345 # define bio_get_queue(bio) bdev_get_queue((bio)->bi_bdev)
1347 # define bio_get_queue(bio) ((bio)->bi_disk->queue)
1350 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1351 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1353 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1356 osd_index_register(struct osd_device *osd, const struct lu_fid *fid,
1357 __u32 keysize, __u32 recsize)
1359 return lustre_index_register(&osd->od_dt_dev, osd_name(osd),
1360 &osd->od_index_backup_list, &osd->od_lock,
1361 &osd->od_index_backup_stop,
1362 fid, keysize, recsize);
1366 osd_index_backup(const struct lu_env *env, struct osd_device *osd, bool backup)
1368 struct osd_thread_info *info = osd_oti_get(env);
1369 struct lu_fid *fid = &info->oti_fid3;
1370 struct osd_inode_id *id = &info->oti_id3;
1372 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
1373 osd_id_gen(id, osd->od_index_backup_inode->i_ino,
1374 osd->od_index_backup_inode->i_generation);
1375 osd_add_oi_cache(info, osd, id, fid);
1377 lustre_index_backup(env, &osd->od_dt_dev, osd_name(osd),
1378 &osd->od_index_backup_list, &osd->od_lock,
1379 &osd->od_index_backup_stop, backup);
1382 #ifdef LDISKFS_HAS_INCOMPAT_FEATURE
1384 # ifdef LDISKFS_FEATURE_INCOMPAT_EXTENTS
1385 # define ldiskfs_has_feature_extents(sb) \
1386 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)
1388 # ifdef LDISKFS_FEATURE_INCOMPAT_EA_INODE
1389 # define ldiskfs_has_feature_ea_inode(sb) \
1390 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1392 # ifdef LDISKFS_FEATURE_INCOMPAT_DIRDATA
1393 # define ldiskfs_has_feature_dirdata(sb) \
1394 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_DIRDATA)
1396 # ifdef LDISKFS_FEATURE_COMPAT_HAS_JOURNAL
1397 # define ldiskfs_has_feature_journal(sb) \
1398 LDISKFS_HAS_COMPAT_FEATURE(sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)
1400 # ifdef LDISKFS_FEATURE_RO_COMPAT_QUOTA
1401 # define ldiskfs_has_feature_quota(sb) \
1402 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_QUOTA)
1404 # ifdef LDISKFS_FEATURE_RO_COMPAT_PROJECT
1405 # define ldiskfs_has_feature_project(sb) \
1406 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_PROJECT)
1411 int osd_trunc_lock(struct osd_object *obj, struct osd_thandle *oh,
1413 void osd_trunc_unlock_all(struct list_head *list);
1414 void osd_process_truncates(struct list_head *list);
1415 void osd_execute_truncate(struct osd_object *obj);
1418 * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb
1419 * This limit is arbitrary, but is reasonable for the xattr API.
1421 #define LDISKFS_XATTR_MAX_LARGE_EA_SIZE (1024 * 1024)
1423 #endif /* _OSD_INTERNAL_H */