4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd/osd_internal.h
34 * Shared definitions and declarations for osd module
36 * Author: Nikita Danilov <nikita@clusterfs.com>
39 #ifndef _OSD_INTERNAL_H
40 #define _OSD_INTERNAL_H
44 #include <linux/mutex.h>
45 /* struct rw_semaphore */
46 #include <linux/rwsem.h>
48 #include <linux/dcache.h>
50 #include <linux/dirent.h>
51 #include <linux/statfs.h>
52 #include <ldiskfs/ldiskfs.h>
53 #include <ldiskfs/ldiskfs_jbd2.h>
57 /* class_register_type(), class_unregister_type(), class_get_type() */
58 #include <obd_class.h>
59 #include <lustre_disk.h>
60 #include <dt_object.h>
61 #include <lustre_quota.h>
65 #include "osd_scrub.h"
66 #include "osd_quota_fmt.h"
69 extern struct kmem_cache *dynlock_cachep;
71 #define OSD_COUNTERS (0)
73 /* ldiskfs special inode::i_state_flags need to be accessed with
74 * ldiskfs_{set,clear,test}_inode_state() only */
76 /* OI scrub should skip this inode. */
77 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
78 #define LDISKFS_STATE_LUSTRE_DESTROY 30
80 /** Enable thandle usage statistics */
81 #define OSD_THANDLE_STATS (0)
83 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
85 #define OBJECTS "OBJECTS"
86 #define ADMIN_USR "admin_quotafile_v2.usr"
87 #define ADMIN_GRP "admin_quotafile_v2.grp"
89 /* Statfs space reservation for fragmentation and local objects */
90 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
91 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
93 /* check if ldiskfs support project quota */
94 #ifndef LDISKFS_IOC_FSSETXATTR
95 #undef HAVE_PROJECT_QUOTA
98 struct osd_directory {
99 struct iam_container od_container;
100 struct iam_descr od_descr;
104 * Object Index (oi) instance.
108 * underlying index object, where fid->id mapping in stored.
110 struct inode *oi_inode;
111 struct osd_directory oi_dir;
114 extern const int osd_dto_credits_noquota[];
117 struct dt_object oo_dt;
119 * Inode for file system object represented by this osd_object. This
120 * inode is pinned for the whole duration of lu_object life.
122 * Not modified concurrently (either setup early during object
123 * creation, or assigned by osd_create() under write lock).
125 struct inode *oo_inode;
127 * to protect index ops.
129 struct htree_lock_head *oo_hl_head;
130 struct rw_semaphore oo_ext_idx_sem;
131 struct rw_semaphore oo_sem;
132 struct osd_directory *oo_dir;
133 /** protects inode attributes. */
136 __u32 oo_destroyed:1;
138 /* the i_flags in LMA */
141 * Following two members are used to indicate the presence of dot and
142 * dotdot in the given directory. This is required for interop mode
145 int oo_compat_dot_created;
146 int oo_compat_dotdot_created;
148 const struct lu_env *oo_owner;
149 #ifdef CONFIG_LOCKDEP
150 struct lockdep_map oo_dep_map;
153 struct list_head oo_xattr_list;
157 /* protects on-fly initialization */
158 int oos_subdir_count; /* subdir count for each seq */
159 struct dentry *oos_root; /* O/<seq> */
160 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
161 u64 oos_seq; /* seq number */
162 struct list_head oos_seq_list; /* list to seq_list */
166 struct dentry *om_root; /* dentry for /O */
167 rwlock_t om_seq_list_lock; /* lock for seq_list */
168 struct list_head om_seq_list; /* list head for seq */
170 struct mutex om_dir_init_mutex;
174 struct dentry *om_root; /* AGENT/<index> */
175 u64 om_index; /* mdt index */
176 struct list_head om_list; /* list to omm_list */
179 struct osd_mdobj_map {
180 struct dentry *omm_remote_parent;
182 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
183 handle_t *handle, struct dentry *child,
184 struct inode *inode, struct htree_lock *hlock);
186 #define OSD_OTABLE_IT_CACHE_SIZE 64
187 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
189 struct osd_inconsistent_item {
190 /* link into osd_scrub::os_inconsistent_items,
191 * protected by osd_scrub::os_lock. */
192 struct list_head oii_list;
194 /* The right FID <=> ino#/gen mapping. */
195 struct osd_idmap_cache oii_cache;
197 unsigned int oii_insert:1; /* insert or update mapping. */
200 struct osd_otable_cache {
201 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
203 /* Index for next cache slot to be filled. */
204 int ooc_producer_idx;
206 /* Index for next cache slot to be returned by it::next(). */
207 int ooc_consumer_idx;
209 /* How many items in ooc_cache. */
210 int ooc_cached_items;
212 /* Position for up layer LFSCK iteration pre-loading. */
213 __u32 ooc_pos_preload;
216 struct osd_otable_it {
217 struct osd_device *ooi_dev;
218 struct osd_otable_cache ooi_cache;
219 struct osd_iit_param ooi_iit_param;
221 /* The following bits can be updated/checked w/o lock protection.
222 * If more bits will be introduced in the future and need lock to
223 * protect, please add comment. */
224 unsigned long ooi_used_outside:1, /* Some user out of OSD
225 * uses the iteration. */
226 ooi_all_cached:1, /* No more entries can be
227 * filled into cache. */
228 ooi_user_ready:1, /* The user out of OSD is
229 * ready to iterate. */
230 ooi_waiting:1; /* it::next is waiting. */
233 struct osd_obj_orphan {
234 struct list_head oor_list;
235 struct lu_env *oor_env; /* to identify "own" records */
244 struct dt_device od_dt_dev;
245 /* information about underlying file system */
246 struct vfsmount *od_mnt;
248 struct osd_oi **od_oi_table;
249 /* total number of OI containers */
254 unsigned int od_fl_capa:1,
262 __u32 od_dirent_journal;
264 struct proc_dir_entry *od_proc_entry;
265 struct lprocfs_stats *od_stats;
267 spinlock_t od_osfs_lock;
270 struct lu_site od_site;
272 struct osd_obj_map *od_ost_map;
273 struct osd_mdobj_map *od_mdt_map;
275 unsigned long long od_readcache_max_filesize;
277 int od_writethrough_cache;
279 struct brw_stats od_brw_stats;
280 atomic_t od_r_in_flight;
281 atomic_t od_w_in_flight;
283 struct mutex od_otable_mutex;
284 struct osd_otable_it *od_otable_it;
285 struct osd_scrub od_scrub;
286 struct list_head od_ios_list;
288 /* service name associated with the osd device */
289 char od_svname[MAX_OBD_NAME];
290 char od_mntdev[MAX_OBD_NAME];
292 /* quota slave instance */
293 struct qsd_instance *od_quota_slave;
295 /* osd seq instance */
296 struct lu_client_seq *od_cl_seq;
297 /* If the ratio of "the total OI mappings count" vs
298 * "the bad OI mappings count" is lower than the
299 * osd_device::od_full_scrub_ratio, then trigger
300 * OI scrub to scan the whole the device. */
301 __u64 od_full_scrub_ratio;
302 /* If the speed of found bad OI mappings (per minute)
303 * exceeds the osd_device::od_full_scrub_threshold_rate,
304 * then trigger OI scrub to scan the whole device. */
305 __u64 od_full_scrub_threshold_rate;
307 /* a list of orphaned agent inodes, protected with od_osfs_lock */
308 struct list_head od_orphan_list;
311 enum osd_full_scrub_ratio {
312 /* Trigger OI scrub to scan the whole device directly. */
315 /* Because the bad OI mappings count cannot be larger than
316 * the total OI mappints count, then setting OFSR_NEVER means
317 * that the whole device scanning cannot be triggered by auto
318 * detected bad OI mappings during the RPC services. */
320 OFSR_DEFAULT = 10000,
323 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
325 /* There are at most 15 uid/gid/projids are affected in a transaction, and
326 * that's rename case:
327 * - 3 for source parent uid & gid & projid;
328 * - 3 for source child uid & gid & projid ('..' entry update when
329 * child is directory);
330 * - 3 for target parent uid & gid & projid;
331 * - 3 for target child uid & gid & projid(if the target child exists);
332 * - 3 for root uid & gid(last_rcvd, llog, etc);
335 #define OSD_MAX_UGID_CNT 15
340 OSD_OT_XATTR_SET = 2,
353 struct thandle ot_super;
355 struct ldiskfs_journal_cb_entry ot_jcb;
356 struct list_head ot_commit_dcb_list;
357 struct list_head ot_stop_dcb_list;
358 /* Link to the device, for debugging. */
359 struct lu_ref_link ot_dev_link;
360 unsigned int ot_credits;
361 unsigned short ot_id_cnt;
362 __u8 ot_id_types[OSD_MAX_UGID_CNT];
363 unsigned int ot_remove_agents:1;
364 uid_t ot_id_array[OSD_MAX_UGID_CNT];
365 struct lquota_trans *ot_quota_trans;
366 #if OSD_THANDLE_STATS
367 /** time when this handle was allocated */
370 /** time when this thanle was started */
376 * Basic transaction credit op
397 #ifdef CONFIG_PROC_FS
399 LPROC_OSD_READ_BYTES = 0,
400 LPROC_OSD_WRITE_BYTES = 1,
401 LPROC_OSD_GET_PAGE = 2,
402 LPROC_OSD_NO_PAGE = 3,
403 LPROC_OSD_CACHE_ACCESS = 4,
404 LPROC_OSD_CACHE_HIT = 5,
405 LPROC_OSD_CACHE_MISS = 6,
407 #if OSD_THANDLE_STATS
408 LPROC_OSD_THANDLE_STARTING,
409 LPROC_OSD_THANDLE_OPEN,
410 LPROC_OSD_THANDLE_CLOSING,
417 * Storage representation for fids.
419 * Variable size, first byte contains the length of the whole record.
421 struct osd_fid_pack {
422 unsigned char fp_len;
423 char fp_area[sizeof(struct lu_fid)];
426 struct osd_it_ea_dirent {
427 struct lu_fid oied_fid;
430 unsigned short oied_namelen;
431 unsigned int oied_type;
433 } __attribute__((packed));
436 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
437 * than lu_dirent struct. osd readdir reads less number of dirent than
438 * required for mdd dir page. so buffer size need to be increased so that
439 * there would be one ext3 readdir for every mdd readdir page.
442 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
445 * This is iterator's in-memory data structure in interoperability
446 * mode (i.e. iterator over ldiskfs style directory)
449 struct osd_object *oie_obj;
450 /** used in ldiskfs iterator, to stored file pointer */
451 struct file oie_file;
452 /** how many entries have been read-cached from storage */
454 /** current entry is being iterated by caller */
456 /** current processing entry */
457 struct osd_it_ea_dirent *oie_dirent;
458 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
460 struct dentry oie_dentry;
464 * Iterator's in-memory data structure for IAM mode.
467 struct osd_object *oi_obj;
468 struct iam_path_descr *oi_ipd;
469 struct iam_iterator oi_it;
472 struct osd_quota_leaf {
473 struct list_head oql_link;
478 * Iterator's in-memory data structure for quota file.
480 struct osd_it_quota {
481 struct osd_object *oiq_obj;
482 /** tree blocks path to where the entry is stored */
483 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
484 /** on-disk offset for current key where quota record can be found */
486 /** identifier for current quota record */
488 /** the record index in the leaf/index block */
489 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
490 /** list of already processed leaf blocks */
491 struct list_head oiq_list;
494 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
497 wait_queue_head_t dr_wait;
498 atomic_t dr_numreqs; /* number of reqs being processed */
503 unsigned int dr_ignore_quota:1;
504 unsigned int dr_elapsed_valid:1; /* we really did count time */
505 unsigned int dr_rw:1;
506 struct lu_buf dr_pg_buf;
507 struct page **dr_pages;
508 struct lu_buf dr_bl_buf;
510 unsigned long dr_start_time;
511 unsigned long dr_elapsed; /* how long io took */
512 struct osd_device *dr_dev;
513 unsigned int dr_init_at; /* the line iobuf was initialized */
516 #define OSD_INS_CACHE_SIZE 8
518 struct osd_thread_info {
519 const struct lu_env *oti_env;
521 * used for index operations.
523 struct dentry oti_obj_dentry;
524 struct dentry oti_child_dentry;
526 /** dentry for Iterator context. */
527 struct dentry oti_it_dentry;
530 /* fake struct file for osd_object_sync */
531 struct file oti_file;
533 struct kstatfs oti_ksfs;
536 struct htree_lock *oti_hlock;
538 struct lu_fid oti_fid;
539 struct lu_fid oti_fid2;
540 struct lu_fid oti_fid3;
541 struct osd_inode_id oti_id;
542 struct osd_inode_id oti_id2;
543 struct osd_inode_id oti_id3;
544 struct ost_id oti_ostid;
547 * XXX temporary: for ->i_op calls.
549 struct timespec oti_time;
552 * following ipd and it structures are used for osd_index_iam_lookup()
553 * these are defined separately as we might do index operation
554 * in open iterator session.
557 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
559 unsigned int oti_it_ea_buf_used:1;
561 /* IAM iterator for index operation. */
562 struct iam_iterator oti_idx_it;
564 /** union to guarantee that ->oti_ipd[] has proper alignment. */
567 char oti_it_ipd[DX_IPD_MAX_SIZE];
568 long long oti_alignment_lieutenant;
572 char oti_idx_ipd[DX_IPD_MAX_SIZE];
573 long long oti_alignment_lieutenant_colonel;
576 struct osd_idmap_cache oti_cache;
578 /* dedicated OI cache for insert (which needs inum) */
579 struct osd_idmap_cache *oti_ins_cache;
580 int oti_ins_cache_size;
581 int oti_ins_cache_used;
586 /** used in osd_fid_set() to put xattr */
587 struct lu_buf oti_buf;
588 struct lu_buf oti_big_buf;
589 /** used in osd_ea_fid_set() to set fid into common ea */
591 struct lustre_ost_attrs oti_ost_attrs;
592 struct filter_fid_old oti_ff;
593 struct filter_fid oti_ff_new;
596 struct osd_iobuf oti_iobuf;
597 /* used to access objects in /O */
598 struct inode *oti_inode;
599 #define OSD_FID_REC_SZ 32
600 char oti_ldp[OSD_FID_REC_SZ];
601 char oti_ldp2[OSD_FID_REC_SZ];
603 /* used by quota code */
605 #if defined(HAVE_DQUOT_QC_DQBLK)
606 struct qc_dqblk oti_qdq;
607 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
608 struct fs_disk_quota oti_fdq;
610 struct if_dqblk oti_dqblk;
612 struct if_dqinfo oti_dqinfo;
614 struct lquota_id_info oti_qi;
615 struct lquota_trans oti_quota_trans;
616 union lquota_rec oti_quota_rec;
618 struct lu_seq_range oti_seq_range;
620 /* Tracking for transaction credits, to allow debugging and optimizing
621 * cases where a large number of credits are being allocated for
622 * single transaction. */
623 unsigned int oti_credits_before;
624 unsigned int oti_declare_ops[OSD_OT_MAX];
625 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
626 unsigned int oti_declare_ops_used[OSD_OT_MAX];
629 /* flags for inode/block quota accounting */
630 enum osd_qid_declare_flags {
631 OSD_QID_INODE = 1 << 0,
632 OSD_QID_BLK = 1 << 1,
633 OSD_QID_FORCE = 1 << 2,
636 extern int ldiskfs_pdo;
638 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
639 const char *name, void *buf, int len)
644 dentry->d_inode = inode;
645 dentry->d_sb = inode->i_sb;
646 return inode->i_op->getxattr(dentry, name, buf, len);
649 static inline int __osd_xattr_set(struct osd_thread_info *info,
650 struct inode *inode, const char *name,
651 const void *buf, int buflen, int fl)
653 struct dentry *dentry = &info->oti_child_dentry;
655 ll_vfs_dq_init(inode);
656 dentry->d_inode = inode;
657 dentry->d_sb = inode->i_sb;
658 return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
661 #ifdef CONFIG_PROC_FS
663 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
664 extern struct lprocfs_vars lprocfs_osd_module_vars[];
665 int osd_procfs_init(struct osd_device *osd, const char *name);
666 int osd_procfs_fini(struct osd_device *osd);
667 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
668 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
669 int osd_register_proc_index_in_idif(struct osd_device *osd);
673 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
674 struct obd_statfs *sfs);
675 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
676 struct osd_inode_id *id);
677 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
678 const struct lu_fid *fid, __u32 compat, __u32 incompat);
679 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
680 struct dentry *dentry, struct lustre_ost_attrs *loa);
681 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
682 struct osd_inode_id *id, const struct lu_fid *fid);
683 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
684 struct dentry *dentry, struct lu_fid *fid);
686 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
687 void osd_obj_map_fini(struct osd_device *dev);
688 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
689 const struct lu_fid *fid, struct osd_inode_id *id);
690 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
691 const struct lu_fid *fid, const struct osd_inode_id *id,
693 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
694 const struct lu_fid *fid, handle_t *th);
695 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
696 const struct lu_fid *fid, const struct osd_inode_id *id,
698 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
699 struct inode *src_parent, struct dentry *src_child,
700 const struct lu_fid *fid);
701 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
702 const struct lu_fid *fid, struct osd_inode_id *id);
703 int osd_obj_spec_insert(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_spec_update(struct osd_thread_info *info, struct osd_device *osd,
707 const struct lu_fid *fid, const struct osd_inode_id *id,
710 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags);
711 int osd_scrub_file_store(struct osd_scrub *scrub);
712 char *osd_lf_fid2name(const struct lu_fid *fid);
713 int osd_scrub_start(struct osd_device *dev, __u32 flags);
714 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev);
715 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
716 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
718 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
719 struct osd_inode_id *id);
720 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
722 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
723 u64 seq, struct lu_seq_range *range);
725 int osd_delete_from_remote_parent(const struct lu_env *env,
726 struct osd_device *osd,
727 struct osd_object *obj,
728 struct osd_thandle *oh);
729 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
730 struct osd_object *obj, struct osd_thandle *oh);
731 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
732 struct osd_device *osd,
733 const struct lu_fid *fid,
734 struct osd_inode_id *id);
736 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
738 /* osd_quota_fmt.c */
739 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
740 int type, uint blk, int depth, uint index,
741 struct osd_it_quota *it);
742 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
743 int type, uint blk, uint index,
744 struct osd_it_quota *it);
745 loff_t find_tree_dqentry(const struct lu_env *env,
746 struct osd_object *obj, int type,
747 qid_t dqid, uint blk, int depth,
748 struct osd_it_quota *it);
750 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
751 struct lquota_id_info *qi, struct osd_object *obj,
752 bool enforce, int *flags);
753 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
754 __u32 projid, long long space, struct osd_thandle *oh,
755 struct osd_object *obj, int *flags,
756 enum osd_qid_declare_flags);
757 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
758 const struct dt_rec *rec,
759 union lquota_rec *quota_rec);
760 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
762 #ifndef HAVE_I_UID_READ
763 static inline uid_t i_uid_read(const struct inode *inode)
768 static inline gid_t i_gid_read(const struct inode *inode)
773 static inline void i_uid_write(struct inode *inode, uid_t uid)
778 static inline void i_gid_write(struct inode *inode, gid_t gid)
784 #ifdef HAVE_PROJECT_QUOTA
785 static inline __u32 i_projid_read(struct inode *inode)
787 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
790 static inline void i_projid_write(struct inode *inode, __u32 projid)
793 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
794 LDISKFS_I(inode)->i_projid = kprojid;
797 static inline uid_t i_projid_read(struct inode *inode)
801 static inline void i_projid_write(struct inode *inode, __u32 projid)
807 #ifdef HAVE_LDISKFS_INFO_JINODE
808 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
809 #else /* HAVE_LDISKFS_INFO_JINODE */
810 # define osd_attach_jinode(inode) 0
811 #endif /* HAVE_LDISKFS_INFO_JINODE */
813 #ifdef LDISKFS_HT_MISC
814 # define osd_journal_start_sb(sb, type, nblock) \
815 ldiskfs_journal_start_sb(sb, type, nblock)
816 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
818 ldiskfs_lblk_t *nblock)
822 rc = osd_attach_jinode(inode);
825 return ldiskfs_append(handle, inode, nblock);
827 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
828 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
830 # define osd_journal_start(inode, type, nblocks) \
831 ldiskfs_journal_start(inode, type, nblocks)
832 # define osd_transaction_size(dev) \
833 (osd_journal(dev)->j_max_transaction_buffers / 2)
835 # define LDISKFS_HT_MISC 0
836 # define osd_journal_start_sb(sb, type, nblock) \
837 ldiskfs_journal_start_sb(sb, nblock)
839 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
841 ldiskfs_lblk_t *nblock)
843 struct buffer_head *bh;
846 bh = ldiskfs_append(handle, inode, nblock, &err);
853 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
854 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
856 # define osd_journal_start(inode, type, nblocks) \
857 ldiskfs_journal_start(inode, nblocks)
858 # define osd_transaction_size(dev) \
859 (osd_journal(dev)->j_max_transaction_buffers)
863 * Invariants, assertions.
867 * XXX: do not enable this, until invariant checking code is made thread safe
868 * in the face of pdirops locking.
870 #define OSD_INVARIANT_CHECKS (0)
872 #if OSD_INVARIANT_CHECKS
873 static inline int osd_invariant(const struct osd_object *obj)
877 ergo(obj->oo_inode != NULL,
878 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
879 atomic_read(&obj->oo_inode->i_count) > 0) &&
880 ergo(obj->oo_dir != NULL &&
881 obj->oo_dir->od_conationer.ic_object != NULL,
882 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
885 #define osd_invariant(obj) (1)
888 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
890 extern const struct dt_index_operations osd_otable_ops;
892 static inline int osd_oi_fid2idx(struct osd_device *dev,
893 const struct lu_fid *fid)
895 return fid->f_seq & (dev->od_oi_count - 1);
898 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
899 const struct lu_fid *fid)
901 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
902 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
903 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
904 DFID"\n", PFID(fid));
905 /* It can work even od_oi_count equals to 1 although it's unexpected,
906 * the only reason we set it to 1 is for performance measurement */
907 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
910 extern const struct lu_device_operations osd_lu_ops;
912 static inline int lu_device_is_osd(const struct lu_device *d)
914 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
917 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
919 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
920 return container_of0(d, struct osd_device, od_dt_dev);
923 static inline struct osd_device *osd_dev(const struct lu_device *d)
925 LASSERT(lu_device_is_osd(d));
926 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
929 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
931 return osd_dev(o->oo_dt.do_lu.lo_dev);
934 static inline struct super_block *osd_sb(const struct osd_device *dev)
936 return dev->od_mnt->mnt_sb;
939 static inline const char *osd_dev2name(const struct osd_device *dev)
941 return osd_sb(dev)->s_id;
944 static inline const char *osd_ino2name(const struct inode *inode)
946 return inode->i_sb->s_id;
950 * Put the osd object once done with it.
952 * \param obj osd object that needs to be put
954 static inline void osd_object_put(const struct lu_env *env,
955 struct osd_object *obj)
957 dt_object_put(env, &obj->oo_dt);
960 static inline int osd_object_is_root(const struct osd_object *obj)
962 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
965 static inline struct osd_object *osd_obj(const struct lu_object *o)
967 LASSERT(lu_device_is_osd(o->lo_dev));
968 return container_of0(o, struct osd_object, oo_dt.do_lu);
971 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
973 return osd_obj(&d->do_lu);
976 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
978 return &osd->od_dt_dev.dd_lu_dev;
981 static inline journal_t *osd_journal(const struct osd_device *dev)
983 return LDISKFS_SB(osd_sb(dev))->s_journal;
986 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
988 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
991 static inline char *osd_name(struct osd_device *osd)
993 return osd->od_svname;
996 static inline bool osd_is_ea_inode(struct inode *inode)
998 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
1001 extern const struct dt_body_operations osd_body_ops;
1002 extern struct lu_context_key osd_key;
1004 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1006 return lu_context_key_get(&env->le_ctx, &osd_key);
1009 extern const struct dt_body_operations osd_body_ops_new;
1015 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1016 const struct iam_container *bag)
1018 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1019 osd_oti_get(env)->oti_it_ipd);
1023 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1024 const struct iam_container *bag)
1026 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1027 osd_oti_get(env)->oti_idx_ipd);
1030 static inline void osd_ipd_put(const struct lu_env *env,
1031 const struct iam_container *bag,
1032 struct iam_path_descr *ipd)
1034 bag->ic_descr->id_ops->id_ipd_free(ipd);
1037 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1038 const loff_t size, const loff_t pos,
1041 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1042 int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize,
1043 int write_NUL, loff_t *offs, handle_t *handle);
1046 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1047 struct inode *inode,
1048 const char *name, const int namelen)
1050 struct osd_thread_info *info = osd_oti_get(env);
1051 struct dentry *child_dentry = &info->oti_child_dentry;
1052 struct dentry *obj_dentry = &info->oti_obj_dentry;
1054 obj_dentry->d_inode = inode;
1055 obj_dentry->d_sb = inode->i_sb;
1056 obj_dentry->d_name.hash = 0;
1058 child_dentry->d_name.hash = 0;
1059 child_dentry->d_parent = obj_dentry;
1060 child_dentry->d_name.name = name;
1061 child_dentry->d_name.len = namelen;
1062 return child_dentry;
1065 extern int osd_trans_declare_op2rb[];
1066 extern int ldiskfs_track_declares_assert;
1067 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1069 static inline void osd_trans_declare_op(const struct lu_env *env,
1070 struct osd_thandle *oh,
1071 enum osd_op_type op, int credits)
1073 struct osd_thread_info *oti = osd_oti_get(env);
1075 LASSERT(oh->ot_handle == NULL);
1076 if (unlikely(op >= OSD_OT_MAX)) {
1077 if (unlikely(ldiskfs_track_declares_assert)) {
1078 LASSERT(op < OSD_OT_MAX);
1080 CWARN("%s: Invalid operation index %d\n",
1081 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1082 libcfs_debug_dumpstack(NULL);
1085 oti->oti_declare_ops[op]++;
1086 oti->oti_declare_ops_cred[op] += credits;
1088 oh->ot_credits += credits;
1091 static inline void osd_trans_exec_op(const struct lu_env *env,
1093 enum osd_op_type op)
1095 struct osd_thread_info *oti = osd_oti_get(env);
1096 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1098 unsigned int rb, left;
1100 LASSERT(oh->ot_handle != NULL);
1101 if (unlikely(op >= OSD_OT_MAX)) {
1102 if (unlikely(ldiskfs_track_declares_assert))
1103 LASSERT(op < OSD_OT_MAX);
1105 CWARN("%s: opcode %u: invalid value >= %u\n",
1106 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1108 libcfs_debug_dumpstack(NULL);
1113 /* find rollback (or reverse) operation for the given one
1114 * such an operation doesn't require additional credits
1115 * as the same set of blocks are modified */
1116 rb = osd_trans_declare_op2rb[op];
1118 /* check whether credits for this operation were reserved at all */
1119 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1120 oti->oti_declare_ops_cred[rb] == 0)) {
1121 /* the API is not perfect yet: CREATE does REF_ADD internally
1122 * while DESTROY does not. To rollback CREATE the callers
1123 * needs to call REF_DEL+DESTROY which is hard to detect using
1124 * a simple table of rollback operations */
1125 if (op == OSD_OT_REF_DEL &&
1126 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1128 if (op == OSD_OT_REF_ADD &&
1129 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1131 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1132 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1133 osd_trans_dump_creds(env, th);
1134 LASSERT(!ldiskfs_track_declares_assert);
1138 /* remember how many credits we have unused before the operation */
1139 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1140 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1141 if (unlikely(oti->oti_credits_before < left)) {
1142 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1143 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1144 oti->oti_credits_before, left, rb);
1145 osd_trans_dump_creds(env, th);
1146 /* on a very small fs (testing?) it's possible that
1147 * the transaction can't fit 1/4 of journal, so we
1148 * just request less credits (see osd_trans_start()).
1149 * ignore the same case here */
1150 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1151 if (unlikely(oh->ot_credits < rb))
1152 LASSERT(!ldiskfs_track_declares_assert);
1156 static inline void osd_trans_exec_check(const struct lu_env *env,
1158 enum osd_op_type op)
1160 struct osd_thread_info *oti = osd_oti_get(env);
1161 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1163 int used, over, quota;
1165 /* how many credits have been used by the operation */
1166 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1168 if (unlikely(used < 0)) {
1169 /* if some block was allocated and released in the same
1170 * transaction, then it won't be a part of the transaction
1171 * and delta can be negative */
1176 /* rollback operations (e.g. when we destroy just created
1177 * object) should not consume any credits. there is no point
1178 * to confuse the checks below */
1182 oti->oti_declare_ops_used[op] += used;
1183 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1186 /* we account quota for a whole transaction and any operation can
1187 * consume corresponding credits */
1188 over = oti->oti_declare_ops_used[op] -
1189 oti->oti_declare_ops_cred[op];
1190 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1191 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1192 if (over <= quota) {
1193 /* probably that credits were consumed by
1194 * quota indirectly (in the depths of ldiskfs) */
1195 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1196 oti->oti_declare_ops_used[op] -= over;
1198 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1199 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1200 oti->oti_declare_ops_used[op], used,
1201 oti->oti_declare_ops_cred[op]);
1202 osd_trans_dump_creds(env, th);
1203 if (unlikely(ldiskfs_track_declares_assert))
1209 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1212 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1213 struct lu_fid *befider)
1215 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1216 memcpy(pack->fp_area, befider, sizeof(*befider));
1217 pack->fp_len = sizeof(*befider) + 1;
1221 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1226 switch (pack->fp_len) {
1227 case sizeof *fid + 1:
1228 memcpy(fid, pack->fp_area, sizeof *fid);
1229 fid_be_to_cpu(fid, fid);
1232 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1239 * Quota/Accounting handling
1241 extern const struct dt_index_operations osd_acct_index_ops;
1242 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1243 const struct lu_fid *fid, struct osd_inode_id *id);
1245 /* copy from fs/ext4/dir.c */
1246 static inline int is_32bit_api(void)
1248 #ifdef CONFIG_COMPAT
1249 return in_compat_syscall();
1251 return (BITS_PER_LONG == 32);
1255 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1257 if ((filp->f_mode & FMODE_32BITHASH) ||
1258 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1259 return LDISKFS_HTREE_EOF_32BIT;
1261 return LDISKFS_HTREE_EOF_64BIT;
1264 static inline int fid_is_internal(const struct lu_fid *fid)
1266 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1269 static inline unsigned long osd_remote_parent_ino(struct osd_device *dev)
1271 return dev->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1275 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1276 * return code has been removed and integrated into the pointer in the
1279 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1280 struct inode *inode,
1281 ldiskfs_lblk_t block,
1285 struct buffer_head *bh;
1288 rc = osd_attach_jinode(inode);
1292 #ifdef HAVE_EXT4_BREAD_4ARGS
1293 bh = ldiskfs_bread(handle, inode, block, create);
1296 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1297 if (bh == NULL && rc != 0)
1303 void ldiskfs_inc_count(handle_t *handle, struct inode *inode);
1304 void ldiskfs_dec_count(handle_t *handle, struct inode *inode);
1306 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1309 #endif /* _OSD_INTERNAL_H */