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. */
139 * Following two members *compat_dot* are used to indicate
140 * the presence of dot and dotdot in the given directory.
141 * This is required for interop mode (b11826).
143 __u32 oo_destroyed:1,
145 oo_compat_dot_created:1,
146 oo_compat_dotdot_created:1;
148 /* the i_flags in LMA */
151 const struct lu_env *oo_owner;
153 struct list_head oo_xattr_list;
154 struct lu_object_header *oo_header;
158 /* protects on-fly initialization */
159 int oos_subdir_count; /* subdir count for each seq */
160 struct dentry *oos_root; /* O/<seq> */
161 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
162 u64 oos_seq; /* seq number */
163 struct list_head oos_seq_list; /* list to seq_list */
167 struct dentry *om_root; /* dentry for /O */
168 rwlock_t om_seq_list_lock; /* lock for seq_list */
169 struct list_head om_seq_list; /* list head for seq */
171 struct mutex om_dir_init_mutex;
175 struct dentry *om_root; /* AGENT/<index> */
176 u64 om_index; /* mdt index */
177 struct list_head om_list; /* list to omm_list */
180 struct osd_mdobj_map {
181 struct dentry *omm_remote_parent;
183 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
184 handle_t *handle, struct dentry *child,
185 struct inode *inode, struct htree_lock *hlock);
187 #define OSD_OTABLE_IT_CACHE_SIZE 64
188 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
190 struct osd_inconsistent_item {
191 /* link into lustre_scrub::os_inconsistent_items,
192 * protected by lustre_scrub::os_lock. */
193 struct list_head oii_list;
195 /* The right FID <=> ino#/gen mapping. */
196 struct osd_idmap_cache oii_cache;
198 unsigned int oii_insert:1; /* insert or update mapping. */
201 struct osd_otable_cache {
202 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
204 /* Index for next cache slot to be filled. */
205 int ooc_producer_idx;
207 /* Index for next cache slot to be returned by it::next(). */
208 int ooc_consumer_idx;
210 /* How many items in ooc_cache. */
211 __u64 ooc_cached_items;
213 /* Position for up layer LFSCK iteration pre-loading. */
214 __u64 ooc_pos_preload;
217 struct osd_otable_it {
218 struct osd_device *ooi_dev;
219 struct osd_otable_cache ooi_cache;
220 struct osd_iit_param ooi_iit_param;
222 /* The following bits can be updated/checked w/o lock protection.
223 * If more bits will be introduced in the future and need lock to
224 * protect, please add comment. */
225 unsigned long ooi_used_outside:1, /* Some user out of OSD
226 * uses the iteration. */
227 ooi_all_cached:1, /* No more entries can be
228 * filled into cache. */
229 ooi_user_ready:1, /* The user out of OSD is
230 * ready to iterate. */
231 ooi_waiting:1; /* it::next is waiting. */
234 struct osd_obj_orphan {
235 struct list_head oor_list;
236 struct lu_env *oor_env; /* to identify "own" records */
245 struct dt_device od_dt_dev;
246 /* information about underlying file system */
247 struct vfsmount *od_mnt;
249 struct osd_oi **od_oi_table;
250 /* total number of OI containers */
255 unsigned int od_fl_capa:1,
263 __s64 od_auto_scrub_interval;
264 __u32 od_dirent_journal;
266 struct proc_dir_entry *od_proc_entry;
267 struct lprocfs_stats *od_stats;
269 spinlock_t od_osfs_lock;
272 struct lu_site od_site;
274 struct osd_obj_map *od_ost_map;
275 struct osd_mdobj_map *od_mdt_map;
277 unsigned long long od_readcache_max_filesize;
279 int od_writethrough_cache;
281 struct brw_stats od_brw_stats;
282 atomic_t od_r_in_flight;
283 atomic_t od_w_in_flight;
285 struct mutex od_otable_mutex;
286 struct osd_otable_it *od_otable_it;
287 struct osd_scrub od_scrub;
288 struct list_head od_ios_list;
290 /* service name associated with the osd device */
291 char od_svname[MAX_OBD_NAME];
292 char od_mntdev[MAX_OBD_NAME];
294 /* quota slave instance */
295 struct qsd_instance *od_quota_slave;
297 /* osd seq instance */
298 struct lu_client_seq *od_cl_seq;
299 /* If the ratio of "the total OI mappings count" vs
300 * "the bad OI mappings count" is lower than the
301 * osd_device::od_full_scrub_ratio, then trigger
302 * OI scrub to scan the whole the device. */
303 __u64 od_full_scrub_ratio;
304 /* If the speed of found bad OI mappings (per minute)
305 * exceeds the osd_device::od_full_scrub_threshold_rate,
306 * then trigger OI scrub to scan the whole device. */
307 __u64 od_full_scrub_threshold_rate;
309 /* a list of orphaned agent inodes, protected with od_osfs_lock */
310 struct list_head od_orphan_list;
311 struct list_head od_index_backup_list;
312 struct list_head od_index_restore_list;
314 struct inode *od_index_backup_inode;
315 enum lustre_index_backup_policy od_index_backup_policy;
316 int od_index_backup_stop;
319 enum osd_full_scrub_ratio {
320 /* Trigger OI scrub to scan the whole device directly. */
323 /* Because the bad OI mappings count cannot be larger than
324 * the total OI mappints count, then setting OFSR_NEVER means
325 * that the whole device scanning cannot be triggered by auto
326 * detected bad OI mappings during the RPC services. */
328 OFSR_DEFAULT = 10000,
331 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
333 /* There are at most 15 uid/gid/projids are affected in a transaction, and
334 * that's rename case:
335 * - 3 for source parent uid & gid & projid;
336 * - 3 for source child uid & gid & projid ('..' entry update when
337 * child is directory);
338 * - 3 for target parent uid & gid & projid;
339 * - 3 for target child uid & gid & projid(if the target child exists);
340 * - 3 for root uid & gid(last_rcvd, llog, etc);
343 #define OSD_MAX_UGID_CNT 15
348 OSD_OT_XATTR_SET = 2,
361 struct thandle ot_super;
363 struct ldiskfs_journal_cb_entry ot_jcb;
364 struct list_head ot_commit_dcb_list;
365 struct list_head ot_stop_dcb_list;
366 /* Link to the device, for debugging. */
367 struct lu_ref_link ot_dev_link;
368 unsigned int ot_credits;
369 unsigned short ot_id_cnt;
370 __u8 ot_id_types[OSD_MAX_UGID_CNT];
371 unsigned int ot_remove_agents:1;
372 uid_t ot_id_array[OSD_MAX_UGID_CNT];
373 struct lquota_trans *ot_quota_trans;
374 #if OSD_THANDLE_STATS
375 /** time when this handle was allocated */
378 /** time when this thanle was started */
384 * Basic transaction credit op
405 #ifdef CONFIG_PROC_FS
407 LPROC_OSD_READ_BYTES = 0,
408 LPROC_OSD_WRITE_BYTES = 1,
409 LPROC_OSD_GET_PAGE = 2,
410 LPROC_OSD_NO_PAGE = 3,
411 LPROC_OSD_CACHE_ACCESS = 4,
412 LPROC_OSD_CACHE_HIT = 5,
413 LPROC_OSD_CACHE_MISS = 6,
415 #if OSD_THANDLE_STATS
416 LPROC_OSD_THANDLE_STARTING,
417 LPROC_OSD_THANDLE_OPEN,
418 LPROC_OSD_THANDLE_CLOSING,
425 * Storage representation for fids.
427 * Variable size, first byte contains the length of the whole record.
429 struct osd_fid_pack {
430 unsigned char fp_len;
431 char fp_area[sizeof(struct lu_fid)];
434 struct osd_it_ea_dirent {
435 struct lu_fid oied_fid;
438 unsigned short oied_namelen;
439 unsigned int oied_type;
441 } __attribute__((packed));
444 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
445 * than lu_dirent struct. osd readdir reads less number of dirent than
446 * required for mdd dir page. so buffer size need to be increased so that
447 * there would be one ext3 readdir for every mdd readdir page.
450 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
453 * This is iterator's in-memory data structure in interoperability
454 * mode (i.e. iterator over ldiskfs style directory)
457 struct osd_object *oie_obj;
458 /** used in ldiskfs iterator, to stored file pointer */
459 struct file oie_file;
460 /** how many entries have been read-cached from storage */
462 /** current entry is being iterated by caller */
464 /** current processing entry */
465 struct osd_it_ea_dirent *oie_dirent;
466 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
468 struct dentry oie_dentry;
472 * Iterator's in-memory data structure for IAM mode.
475 struct osd_object *oi_obj;
476 struct iam_path_descr *oi_ipd;
477 struct iam_iterator oi_it;
480 struct osd_quota_leaf {
481 struct list_head oql_link;
486 * Iterator's in-memory data structure for quota file.
488 struct osd_it_quota {
489 struct osd_object *oiq_obj;
490 /** tree blocks path to where the entry is stored */
491 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
492 /** on-disk offset for current key where quota record can be found */
494 /** identifier for current quota record */
496 /** the record index in the leaf/index block */
497 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
498 /** list of already processed leaf blocks */
499 struct list_head oiq_list;
502 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
505 wait_queue_head_t dr_wait;
506 atomic_t dr_numreqs; /* number of reqs being processed */
511 unsigned int dr_ignore_quota:1;
512 unsigned int dr_elapsed_valid:1; /* we really did count time */
513 unsigned int dr_rw:1;
514 struct lu_buf dr_pg_buf;
515 struct page **dr_pages;
516 struct lu_buf dr_bl_buf;
518 ktime_t dr_start_time;
519 ktime_t dr_elapsed; /* how long io took */
520 struct osd_device *dr_dev;
521 unsigned int dr_init_at; /* the line iobuf was initialized */
524 #define OSD_INS_CACHE_SIZE 8
526 struct osd_thread_info {
527 const struct lu_env *oti_env;
529 * used for index operations.
531 struct dentry oti_obj_dentry;
532 struct dentry oti_child_dentry;
534 /** dentry for Iterator context. */
535 struct dentry oti_it_dentry;
538 /* fake struct file for osd_object_sync */
539 struct file oti_file;
541 struct kstatfs oti_ksfs;
544 struct htree_lock *oti_hlock;
546 struct lu_fid oti_fid;
547 struct lu_fid oti_fid2;
548 struct lu_fid oti_fid3;
549 struct osd_inode_id oti_id;
550 struct osd_inode_id oti_id2;
551 struct osd_inode_id oti_id3;
552 struct ost_id oti_ostid;
555 * XXX temporary: for ->i_op calls.
557 struct timespec oti_time;
560 * following ipd and it structures are used for osd_index_iam_lookup()
561 * these are defined separately as we might do index operation
562 * in open iterator session.
565 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
567 unsigned int oti_it_ea_buf_used:1;
569 /* IAM iterator for index operation. */
570 struct iam_iterator oti_idx_it;
572 /** union to guarantee that ->oti_ipd[] has proper alignment. */
575 char oti_it_ipd[DX_IPD_MAX_SIZE];
576 long long oti_alignment_lieutenant;
580 char oti_idx_ipd[DX_IPD_MAX_SIZE];
581 long long oti_alignment_lieutenant_colonel;
584 struct osd_idmap_cache oti_cache;
586 /* dedicated OI cache for insert (which needs inum) */
587 struct osd_idmap_cache *oti_ins_cache;
588 int oti_ins_cache_size;
589 int oti_ins_cache_used;
590 /* inc by osd_trans_create and dec by osd_trans_stop */
591 int oti_ins_cache_depth;
596 /** used in osd_fid_set() to put xattr */
597 struct lu_buf oti_buf;
598 struct lu_buf oti_big_buf;
599 /** used in osd_ea_fid_set() to set fid into common ea */
601 struct lustre_ost_attrs oti_ost_attrs;
602 struct filter_fid_old oti_ff;
603 struct filter_fid oti_ff_new;
606 struct osd_iobuf oti_iobuf;
607 /* used to access objects in /O */
608 struct inode *oti_inode;
609 #define OSD_FID_REC_SZ 32
610 char oti_ldp[OSD_FID_REC_SZ];
611 char oti_ldp2[OSD_FID_REC_SZ];
613 /* used by quota code */
615 #if defined(HAVE_DQUOT_QC_DQBLK)
616 struct qc_dqblk oti_qdq;
617 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
618 struct fs_disk_quota oti_fdq;
620 struct if_dqblk oti_dqblk;
622 struct if_dqinfo oti_dqinfo;
624 struct lquota_id_info oti_qi;
625 struct lquota_trans oti_quota_trans;
626 union lquota_rec oti_quota_rec;
628 struct lu_seq_range oti_seq_range;
630 /* Tracking for transaction credits, to allow debugging and optimizing
631 * cases where a large number of credits are being allocated for
632 * single transaction. */
633 unsigned int oti_credits_before;
634 unsigned int oti_declare_ops[OSD_OT_MAX];
635 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
636 unsigned int oti_declare_ops_used[OSD_OT_MAX];
637 struct osd_directory oti_iam;
640 extern int ldiskfs_pdo;
642 #ifndef HAVE_VFS_SETXATTR
643 #define osd_setxattr(dentry, inode, name, buf, len, flag) \
644 ((inode)->i_op->setxattr(dentry, name, buf, len, flag))
645 #define osd_getxattr(dentry, inode, name, buf, len) \
646 ((inode)->i_op->getxattr(dentry, name, buf, len))
647 #define osd_removexattr(dentry, inode, name) \
648 ((inode)->i_op->removexattr(dentry, name))
649 #else /* HAVE_VFS_SETXATTR */
650 #define osd_setxattr(dentry, inode, name, buf, len, flag) \
651 __vfs_setxattr(dentry, inode, name, buf, len, flag)
652 #define osd_getxattr(dentry, inode, name, buf, len) \
653 __vfs_getxattr(dentry, inode, name, buf, len)
654 #define osd_removexattr(dentry, inode, name) \
655 __vfs_removexattr(dentry, name)
656 #endif /* !HAVE_VFS_SETXATTR */
658 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
659 const char *name, void *buf, int len)
664 dentry->d_inode = inode;
665 dentry->d_sb = inode->i_sb;
666 return osd_getxattr(dentry, inode, name, buf, len);
669 static inline int __osd_xattr_set(struct osd_thread_info *info,
670 struct inode *inode, const char *name,
671 const void *buf, int buflen, int fl)
673 struct dentry *dentry = &info->oti_child_dentry;
675 ll_vfs_dq_init(inode);
676 dentry->d_inode = inode;
677 dentry->d_sb = inode->i_sb;
678 return osd_setxattr(dentry, inode, name, buf, buflen, fl);
681 #ifdef CONFIG_PROC_FS
683 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
684 extern struct lprocfs_vars lprocfs_osd_module_vars[];
685 int osd_procfs_init(struct osd_device *osd, const char *name);
686 int osd_procfs_fini(struct osd_device *osd);
687 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
688 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
689 int osd_register_proc_index_in_idif(struct osd_device *osd);
693 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
694 struct obd_statfs *sfs);
695 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
696 struct osd_inode_id *id);
698 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
699 struct osd_inode_id *id, struct lu_fid *fid);
700 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
701 const struct lu_fid *fid, __u32 compat, __u32 incompat);
702 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
703 struct dentry *dentry, struct lustre_ost_attrs *loa);
704 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
705 struct osd_inode_id *id, const struct lu_fid *fid);
706 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
707 struct dentry *dentry, struct lu_fid *fid);
709 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
710 void osd_obj_map_fini(struct osd_device *dev);
711 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
712 const struct lu_fid *fid, struct osd_inode_id *id);
713 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
714 const struct lu_fid *fid, const struct osd_inode_id *id,
716 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
717 const struct lu_fid *fid, handle_t *th);
718 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
719 const struct lu_fid *fid, const struct osd_inode_id *id,
721 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
722 struct inode *src_parent, struct dentry *src_child,
723 const struct lu_fid *fid);
724 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
725 const struct lu_fid *fid, struct osd_inode_id *id);
726 int osd_obj_spec_insert(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_spec_update(struct osd_thread_info *info, struct osd_device *osd,
730 const struct lu_fid *fid, const struct osd_inode_id *id,
733 char *osd_lf_fid2name(const struct lu_fid *fid);
734 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
736 void osd_scrub_stop(struct osd_device *dev);
737 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
738 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
739 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
741 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
742 struct osd_inode_id *id);
743 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
745 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
746 u64 seq, struct lu_seq_range *range);
748 int osd_delete_from_remote_parent(const struct lu_env *env,
749 struct osd_device *osd,
750 struct osd_object *obj,
751 struct osd_thandle *oh, bool destroy);
752 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
753 struct osd_object *obj, struct osd_thandle *oh);
754 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
755 struct osd_device *osd,
756 const struct lu_fid *fid,
757 struct osd_inode_id *id);
759 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
761 /* osd_quota_fmt.c */
762 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
763 int type, uint blk, int depth, uint index,
764 struct osd_it_quota *it);
765 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
766 int type, uint blk, uint index,
767 struct osd_it_quota *it);
768 loff_t find_tree_dqentry(const struct lu_env *env,
769 struct osd_object *obj, int type,
770 qid_t dqid, uint blk, int depth,
771 struct osd_it_quota *it);
773 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
774 struct lquota_id_info *qi, struct osd_object *obj,
775 bool enforce, int *flags);
776 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
777 __u32 projid, long long space, struct osd_thandle *oh,
778 struct osd_object *obj, int *flags,
779 enum osd_qid_declare_flags);
780 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
781 const struct dt_rec *rec,
782 union lquota_rec *quota_rec);
783 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
785 #ifndef HAVE_I_UID_READ
786 static inline uid_t i_uid_read(const struct inode *inode)
791 static inline gid_t i_gid_read(const struct inode *inode)
796 static inline void i_uid_write(struct inode *inode, uid_t uid)
801 static inline void i_gid_write(struct inode *inode, gid_t gid)
807 #ifdef HAVE_PROJECT_QUOTA
808 static inline __u32 i_projid_read(struct inode *inode)
810 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
813 static inline void i_projid_write(struct inode *inode, __u32 projid)
816 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
817 LDISKFS_I(inode)->i_projid = kprojid;
820 static inline uid_t i_projid_read(struct inode *inode)
824 static inline void i_projid_write(struct inode *inode, __u32 projid)
830 #ifdef HAVE_LDISKFS_INFO_JINODE
831 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
832 #else /* HAVE_LDISKFS_INFO_JINODE */
833 # define osd_attach_jinode(inode) 0
834 #endif /* HAVE_LDISKFS_INFO_JINODE */
836 #ifdef LDISKFS_HT_MISC
837 # define osd_journal_start_sb(sb, type, nblock) \
838 ldiskfs_journal_start_sb(sb, type, nblock)
839 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
841 ldiskfs_lblk_t *nblock)
845 rc = osd_attach_jinode(inode);
848 return ldiskfs_append(handle, inode, nblock);
850 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
851 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
853 # define osd_journal_start(inode, type, nblocks) \
854 ldiskfs_journal_start(inode, type, nblocks)
855 # define osd_transaction_size(dev) \
856 (osd_journal(dev)->j_max_transaction_buffers / 2)
858 # define LDISKFS_HT_MISC 0
859 # define osd_journal_start_sb(sb, type, nblock) \
860 ldiskfs_journal_start_sb(sb, nblock)
862 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
864 ldiskfs_lblk_t *nblock)
866 struct buffer_head *bh;
869 bh = ldiskfs_append(handle, inode, nblock, &err);
876 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
877 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
879 # define osd_journal_start(inode, type, nblocks) \
880 ldiskfs_journal_start(inode, nblocks)
881 # define osd_transaction_size(dev) \
882 (osd_journal(dev)->j_max_transaction_buffers)
886 * Invariants, assertions.
890 * XXX: do not enable this, until invariant checking code is made thread safe
891 * in the face of pdirops locking.
893 #define OSD_INVARIANT_CHECKS (0)
895 #if OSD_INVARIANT_CHECKS
896 static inline int osd_invariant(const struct osd_object *obj)
900 ergo(obj->oo_inode != NULL,
901 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
902 atomic_read(&obj->oo_inode->i_count) > 0) &&
903 ergo(obj->oo_dir != NULL &&
904 obj->oo_dir->od_conationer.ic_object != NULL,
905 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
908 #define osd_invariant(obj) (1)
911 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
913 extern const struct dt_index_operations osd_otable_ops;
915 static inline int osd_oi_fid2idx(struct osd_device *dev,
916 const struct lu_fid *fid)
918 return fid->f_seq & (dev->od_oi_count - 1);
921 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
922 const struct lu_fid *fid)
924 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
925 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
926 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
927 DFID"\n", PFID(fid));
928 /* It can work even od_oi_count equals to 1 although it's unexpected,
929 * the only reason we set it to 1 is for performance measurement */
930 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
933 extern const struct lu_device_operations osd_lu_ops;
935 static inline int lu_device_is_osd(const struct lu_device *d)
937 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
940 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
942 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
943 return container_of0(d, struct osd_device, od_dt_dev);
946 static inline struct osd_device *osd_dev(const struct lu_device *d)
948 LASSERT(lu_device_is_osd(d));
949 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
952 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
954 return osd_dev(o->oo_dt.do_lu.lo_dev);
957 static inline struct super_block *osd_sb(const struct osd_device *dev)
959 return dev->od_mnt->mnt_sb;
962 static inline const char *osd_dev2name(const struct osd_device *dev)
964 return osd_sb(dev)->s_id;
967 static inline const char *osd_ino2name(const struct inode *inode)
969 return inode->i_sb->s_id;
973 * Put the osd object once done with it.
975 * \param obj osd object that needs to be put
977 static inline void osd_object_put(const struct lu_env *env,
978 struct osd_object *obj)
980 dt_object_put(env, &obj->oo_dt);
983 static inline int osd_object_is_root(const struct osd_object *obj)
985 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
988 static inline struct osd_object *osd_obj(const struct lu_object *o)
990 LASSERT(lu_device_is_osd(o->lo_dev));
991 return container_of0(o, struct osd_object, oo_dt.do_lu);
994 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
996 return osd_obj(&d->do_lu);
999 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
1001 return &osd->od_dt_dev.dd_lu_dev;
1004 static inline journal_t *osd_journal(const struct osd_device *dev)
1006 return LDISKFS_SB(osd_sb(dev))->s_journal;
1009 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
1011 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
1014 static inline char *osd_name(struct osd_device *osd)
1016 return osd->od_svname;
1019 static inline bool osd_is_ea_inode(struct inode *inode)
1021 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
1024 extern const struct dt_body_operations osd_body_ops;
1025 extern struct lu_context_key osd_key;
1027 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1029 return lu_context_key_get(&env->le_ctx, &osd_key);
1032 extern const struct dt_body_operations osd_body_ops_new;
1038 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1039 const struct iam_container *bag)
1041 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1042 osd_oti_get(env)->oti_it_ipd);
1046 struct iam_path_descr *osd_idx_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_idx_ipd);
1053 static inline void osd_ipd_put(const struct lu_env *env,
1054 const struct iam_container *bag,
1055 struct iam_path_descr *ipd)
1057 bag->ic_descr->id_ops->id_ipd_free(ipd);
1060 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1061 const loff_t size, const loff_t pos,
1064 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1065 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1066 int write_NUL, loff_t *offs, handle_t *handle);
1069 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1070 struct inode *inode,
1071 const char *name, const int namelen)
1073 struct osd_thread_info *info = osd_oti_get(env);
1074 struct dentry *child_dentry = &info->oti_child_dentry;
1075 struct dentry *obj_dentry = &info->oti_obj_dentry;
1077 obj_dentry->d_inode = inode;
1078 obj_dentry->d_sb = inode->i_sb;
1079 obj_dentry->d_name.hash = 0;
1081 child_dentry->d_name.hash = 0;
1082 child_dentry->d_parent = obj_dentry;
1083 child_dentry->d_name.name = name;
1084 child_dentry->d_name.len = namelen;
1085 return child_dentry;
1088 extern int osd_trans_declare_op2rb[];
1089 extern int ldiskfs_track_declares_assert;
1090 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1092 static inline void osd_trans_declare_op(const struct lu_env *env,
1093 struct osd_thandle *oh,
1094 enum osd_op_type op, int credits)
1096 struct osd_thread_info *oti = osd_oti_get(env);
1098 LASSERT(oh->ot_handle == NULL);
1099 if (unlikely(op >= OSD_OT_MAX)) {
1100 if (unlikely(ldiskfs_track_declares_assert)) {
1101 LASSERT(op < OSD_OT_MAX);
1103 CWARN("%s: Invalid operation index %d\n",
1104 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1105 libcfs_debug_dumpstack(NULL);
1108 oti->oti_declare_ops[op]++;
1109 oti->oti_declare_ops_cred[op] += credits;
1111 oh->ot_credits += credits;
1114 static inline void osd_trans_exec_op(const struct lu_env *env,
1116 enum osd_op_type op)
1118 struct osd_thread_info *oti = osd_oti_get(env);
1119 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1121 unsigned int rb, left;
1123 LASSERT(oh->ot_handle != NULL);
1124 if (unlikely(op >= OSD_OT_MAX)) {
1125 if (unlikely(ldiskfs_track_declares_assert))
1126 LASSERT(op < OSD_OT_MAX);
1128 CWARN("%s: opcode %u: invalid value >= %u\n",
1129 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1131 libcfs_debug_dumpstack(NULL);
1136 /* find rollback (or reverse) operation for the given one
1137 * such an operation doesn't require additional credits
1138 * as the same set of blocks are modified */
1139 rb = osd_trans_declare_op2rb[op];
1141 /* check whether credits for this operation were reserved at all */
1142 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1143 oti->oti_declare_ops_cred[rb] == 0)) {
1144 /* the API is not perfect yet: CREATE does REF_ADD internally
1145 * while DESTROY does not. To rollback CREATE the callers
1146 * needs to call REF_DEL+DESTROY which is hard to detect using
1147 * a simple table of rollback operations */
1148 if (op == OSD_OT_REF_DEL &&
1149 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1151 if (op == OSD_OT_REF_ADD &&
1152 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1154 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1155 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1156 osd_trans_dump_creds(env, th);
1157 LASSERT(!ldiskfs_track_declares_assert);
1161 /* remember how many credits we have unused before the operation */
1162 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1163 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1164 if (unlikely(oti->oti_credits_before < left)) {
1165 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1166 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1167 oti->oti_credits_before, left, rb);
1168 osd_trans_dump_creds(env, th);
1169 /* on a very small fs (testing?) it's possible that
1170 * the transaction can't fit 1/4 of journal, so we
1171 * just request less credits (see osd_trans_start()).
1172 * ignore the same case here */
1173 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1174 if (unlikely(oh->ot_credits < rb))
1175 LASSERT(!ldiskfs_track_declares_assert);
1179 static inline void osd_trans_exec_check(const struct lu_env *env,
1181 enum osd_op_type op)
1183 struct osd_thread_info *oti = osd_oti_get(env);
1184 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1186 int used, over, quota;
1188 /* how many credits have been used by the operation */
1189 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1191 if (unlikely(used < 0)) {
1192 /* if some block was allocated and released in the same
1193 * transaction, then it won't be a part of the transaction
1194 * and delta can be negative */
1199 /* rollback operations (e.g. when we destroy just created
1200 * object) should not consume any credits. there is no point
1201 * to confuse the checks below */
1205 oti->oti_declare_ops_used[op] += used;
1206 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1209 /* we account quota for a whole transaction and any operation can
1210 * consume corresponding credits */
1211 over = oti->oti_declare_ops_used[op] -
1212 oti->oti_declare_ops_cred[op];
1213 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1214 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1215 if (over <= quota) {
1216 /* probably that credits were consumed by
1217 * quota indirectly (in the depths of ldiskfs) */
1218 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1219 oti->oti_declare_ops_used[op] -= over;
1221 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1222 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1223 oti->oti_declare_ops_used[op], used,
1224 oti->oti_declare_ops_cred[op]);
1225 osd_trans_dump_creds(env, th);
1226 if (unlikely(ldiskfs_track_declares_assert))
1232 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1235 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1236 struct lu_fid *befider)
1238 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1239 memcpy(pack->fp_area, befider, sizeof(*befider));
1240 pack->fp_len = sizeof(*befider) + 1;
1244 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1249 switch (pack->fp_len) {
1250 case sizeof *fid + 1:
1251 memcpy(fid, pack->fp_area, sizeof *fid);
1252 fid_be_to_cpu(fid, fid);
1255 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1262 * Quota/Accounting handling
1264 extern const struct dt_index_operations osd_acct_index_ops;
1265 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1266 const struct lu_fid *fid, struct osd_inode_id *id);
1268 /* copy from fs/ext4/dir.c */
1269 static inline int is_32bit_api(void)
1271 #ifdef CONFIG_COMPAT
1272 return in_compat_syscall();
1274 return (BITS_PER_LONG == 32);
1278 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1280 if ((filp->f_mode & FMODE_32BITHASH) ||
1281 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1282 return LDISKFS_HTREE_EOF_32BIT;
1284 return LDISKFS_HTREE_EOF_64BIT;
1287 static inline int fid_is_internal(const struct lu_fid *fid)
1289 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1292 static inline bool is_remote_parent_ino(struct osd_device *o, unsigned long ino)
1297 LASSERT(o->od_mdt_map != NULL);
1299 return ino == o->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1303 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1304 * return code has been removed and integrated into the pointer in the
1307 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1308 struct inode *inode,
1309 ldiskfs_lblk_t block,
1313 struct buffer_head *bh;
1316 rc = osd_attach_jinode(inode);
1320 #ifdef HAVE_EXT4_BREAD_4ARGS
1321 bh = ldiskfs_bread(handle, inode, block, create);
1324 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1325 if (bh == NULL && rc != 0)
1331 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1332 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1334 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1337 osd_index_register(struct osd_device *osd, const struct lu_fid *fid,
1338 __u32 keysize, __u32 recsize)
1340 return lustre_index_register(&osd->od_dt_dev, osd_name(osd),
1341 &osd->od_index_backup_list, &osd->od_lock,
1342 &osd->od_index_backup_stop,
1343 fid, keysize, recsize);
1347 osd_index_backup(const struct lu_env *env, struct osd_device *osd, bool backup)
1349 struct osd_thread_info *info = osd_oti_get(env);
1350 struct lu_fid *fid = &info->oti_fid3;
1351 struct osd_inode_id *id = &info->oti_id3;
1353 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
1354 osd_id_gen(id, osd->od_index_backup_inode->i_ino,
1355 osd->od_index_backup_inode->i_generation);
1356 osd_add_oi_cache(info, osd, id, fid);
1358 lustre_index_backup(env, &osd->od_dt_dev, osd_name(osd),
1359 &osd->od_index_backup_list, &osd->od_lock,
1360 &osd->od_index_backup_stop, backup);
1363 #ifdef LDISKFS_HAS_INCOMPAT_FEATURE
1365 # ifdef LDISKFS_FEATURE_INCOMPAT_EXTENTS
1366 # define ldiskfs_has_feature_extents(sb) \
1367 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)
1369 # ifdef LDISKFS_FEATURE_INCOMPAT_EA_INODE
1370 # define ldiskfs_has_feature_ea_inode(sb) \
1371 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1373 # ifdef LDISKFS_FEATURE_INCOMPAT_DIRDATA
1374 # define ldiskfs_has_feature_dirdata(sb) \
1375 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_DIRDATA)
1377 # ifdef LDISKFS_FEATURE_COMPAT_HAS_JOURNAL
1378 # define ldiskfs_has_feature_journal(sb) \
1379 LDISKFS_HAS_COMPAT_FEATURE(sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)
1381 # ifdef LDISKFS_FEATURE_RO_COMPAT_QUOTA
1382 # define ldiskfs_has_feature_quota(sb) \
1383 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_QUOTA)
1385 # ifdef LDISKFS_FEATURE_RO_COMPAT_PROJECT
1386 # define ldiskfs_has_feature_project(sb) \
1387 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_PROJECT)
1392 #endif /* _OSD_INTERNAL_H */