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/
31 * lustre/osd/osd_internal.h
33 * Shared definitions and declarations for osd module
35 * Author: Nikita Danilov <nikita@clusterfs.com>
38 #ifndef _OSD_INTERNAL_H
39 #define _OSD_INTERNAL_H
43 #include <linux/mutex.h>
44 /* struct rw_semaphore */
45 #include <linux/rwsem.h>
47 #include <linux/dcache.h>
49 #include <linux/dirent.h>
50 #include <linux/statfs.h>
51 #include <linux/bio.h>
52 #include <linux/file.h>
53 #include <ldiskfs/ldiskfs.h>
54 #include <ldiskfs/ldiskfs_jbd2.h>
55 #include <lustre_compat.h>
59 /* class_register_type(), class_unregister_type() */
60 #include <obd_class.h>
61 #include <lustre_disk.h>
62 #include <dt_object.h>
63 #include <lustre_quota.h>
67 #include "osd_scrub.h"
68 #include "osd_quota_fmt.h"
70 #if IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)
71 #ifdef HAVE_LINUX_BLK_INTEGRITY_HEADER
72 #include <linux/blk-integrity.h>
77 extern struct kmem_cache *dynlock_cachep;
79 #define OSD_COUNTERS (0)
81 /* ldiskfs special inode::i_state_flags need to be accessed with
82 * ldiskfs_{set,clear,test}_inode_state() only */
84 /* OI scrub should skip this inode. */
85 #define LDISKFS_STATE_LUSTRE_NOSCRUB 31
86 #define LDISKFS_STATE_LUSTRE_DESTROY 30
88 /** Enable thandle usage statistics */
89 #define OSD_THANDLE_STATS (0)
91 #define MAX_OBJID_GROUP (FID_SEQ_ECHO + 1)
93 #define OBJECTS "OBJECTS"
94 #define ADMIN_USR "admin_quotafile_v2.usr"
95 #define ADMIN_GRP "admin_quotafile_v2.grp"
97 /* Statfs space reservation for fragmentation and local objects */
98 #define OSD_STATFS_RESERVED (1ULL << 23) /* 8MB */
99 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
101 /* Default extent bytes when declaring write commit */
102 #define OSD_DEFAULT_EXTENT_BYTES (1U << 20)
104 /* check if ldiskfs support project quota */
105 #if LDISKFS_MAXQUOTAS < 3
106 #undef HAVE_PROJECT_QUOTA
109 #define OBD_BRW_MAPPED OBD_BRW_LOCAL1
111 struct osd_directory {
112 struct iam_container od_container;
113 struct iam_descr od_descr;
117 * Object Index (oi) instance.
121 * underlying index object, where fid->id mapping in stored.
123 struct inode *oi_inode;
124 struct osd_directory oi_dir;
127 extern const int osd_dto_credits_noquota[];
130 struct dt_object oo_dt;
132 * Inode for file system object represented by this osd_object. This
133 * inode is pinned for the whole duration of lu_object life.
135 * Not modified concurrently (either setup early during object
136 * creation, or assigned by osd_create() under write lock).
138 struct inode *oo_inode;
140 * to protect index ops.
142 struct htree_lock_head *oo_hl_head;
143 struct rw_semaphore oo_ext_idx_sem;
144 struct rw_semaphore oo_sem;
145 struct osd_directory *oo_dir;
146 /** protects inode attributes. */
150 * Following two members *compat_dot* are used to indicate
151 * the presence of dot and dotdot in the given directory.
152 * This is required for interop mode (b11826).
154 __u32 oo_destroyed:1,
156 oo_compat_dot_created:1,
157 oo_compat_dotdot_created:1;
159 /* the i_flags in LMA */
162 const struct lu_env *oo_owner;
164 struct list_head oo_xattr_list;
165 struct lu_object_header *oo_header;
166 __u64 oo_dirent_count;
170 /* protects on-fly initialization */
171 int oos_subdir_count; /* subdir count for each seq */
172 struct dentry *oos_root; /* O/<seq> */
173 struct dentry **oos_dirs; /* O/<seq>/d0-dXX */
174 u64 oos_seq; /* seq number */
175 struct list_head oos_seq_list; /* list to seq_list */
179 struct dentry *om_root; /* dentry for /O */
180 rwlock_t om_seq_list_lock; /* lock for seq_list */
181 struct list_head om_seq_list; /* list head for seq */
183 struct mutex om_dir_init_mutex;
187 struct dentry *om_root; /* AGENT/<index> */
188 u64 om_index; /* mdt index */
189 struct list_head om_list; /* list to omm_list */
192 struct osd_mdobj_map {
193 struct dentry *omm_remote_parent;
195 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
196 handle_t *handle, struct dentry *child,
197 struct inode *inode, struct htree_lock *hlock);
199 #define OSD_OTABLE_IT_CACHE_SIZE 64
200 #define OSD_OTABLE_IT_CACHE_MASK (~(OSD_OTABLE_IT_CACHE_SIZE - 1))
202 struct osd_inconsistent_item {
203 /* link into lustre_scrub::os_inconsistent_items,
204 * protected by lustre_scrub::os_lock. */
205 struct list_head oii_list;
207 /* The right FID <=> ino#/gen mapping. */
208 struct osd_idmap_cache oii_cache;
210 unsigned int oii_insert:1; /* insert or update mapping. */
213 struct osd_otable_cache {
214 struct osd_idmap_cache ooc_cache[OSD_OTABLE_IT_CACHE_SIZE];
216 /* Index for next cache slot to be filled. */
217 int ooc_producer_idx;
219 /* Index for next cache slot to be returned by it::next(). */
220 int ooc_consumer_idx;
222 /* How many items in ooc_cache. */
223 __u64 ooc_cached_items;
225 /* Position for up layer LFSCK iteration pre-loading. */
226 __u64 ooc_pos_preload;
229 struct osd_otable_it {
230 struct osd_device *ooi_dev;
231 struct osd_otable_cache ooi_cache;
232 struct osd_iit_param ooi_iit_param;
234 /* The following bits can be updated/checked w/o lock protection.
235 * If more bits will be introduced in the future and need lock to
236 * protect, please add comment. */
237 unsigned long ooi_used_outside:1, /* Some user out of OSD
238 * uses the iteration. */
239 ooi_all_cached:1, /* No more entries can be
240 * filled into cache. */
241 ooi_user_ready:1, /* The user out of OSD is
242 * ready to iterate. */
243 ooi_waiting:1; /* it::next is waiting. */
246 struct osd_obj_orphan {
247 struct list_head oor_list;
248 struct lu_env *oor_env; /* to identify "own" records */
253 OSD_T10_TYPE_UNKNOWN = 0x00,
254 OSD_T10_TYPE1 = 0x01,
255 OSD_T10_TYPE2 = 0x02,
256 OSD_T10_TYPE3 = 0x04,
257 OSD_T10_TYPE_CRC = 0x08,
258 OSD_T10_TYPE_IP = 0x10,
259 OSD_T10_TYPE1_CRC = OSD_T10_TYPE1 | OSD_T10_TYPE_CRC,
260 OSD_T10_TYPE2_CRC = OSD_T10_TYPE2 | OSD_T10_TYPE_CRC,
261 OSD_T10_TYPE3_CRC = OSD_T10_TYPE3 | OSD_T10_TYPE_CRC,
262 OSD_T10_TYPE1_IP = OSD_T10_TYPE1 | OSD_T10_TYPE_IP,
263 OSD_T10_TYPE2_IP = OSD_T10_TYPE2 | OSD_T10_TYPE_IP,
264 OSD_T10_TYPE3_IP = OSD_T10_TYPE3 | OSD_T10_TYPE_IP,
272 struct dt_device od_dt_dev;
273 /* information about underlying file system */
274 struct vfsmount *od_mnt;
276 struct osd_oi **od_oi_table;
277 /* total number of OI containers */
282 unsigned int od_fl_capa:1,
291 od_writethrough_cache:1,
293 od_enable_projid_xattr:1;
296 __u32 od_dirent_journal;
298 struct proc_dir_entry *od_proc_entry;
299 struct lprocfs_stats *od_stats;
301 spinlock_t od_osfs_lock;
303 int od_fallocate_zero_blocks;
305 struct lu_site od_site;
307 struct osd_obj_map *od_ost_map;
308 struct osd_mdobj_map *od_mdt_map;
310 /* objects with size > od_readcache_max_filesize will be
311 * served bypassing pagecache unless already cached */
312 unsigned long long od_readcache_max_filesize;
314 /* reads > od_readcache_max_iosize will be
315 * served bypassing pagecache unless already cached */
316 unsigned long od_readcache_max_iosize;
318 /* writes > od_writethough_max_iosize will be
319 * served bypassing pagecache unless already cached */
320 unsigned long od_writethrough_max_iosize;
322 struct brw_stats od_brw_stats;
323 atomic_t od_r_in_flight;
324 atomic_t od_w_in_flight;
326 struct mutex od_otable_mutex;
327 struct osd_otable_it *od_otable_it;
328 struct osd_scrub od_scrub;
329 struct list_head od_ios_list;
331 /* service name associated with the osd device */
332 char od_svname[MAX_OBD_NAME];
333 char od_mntdev[MAX_OBD_NAME];
336 /* quota slave instance for inode */
337 struct qsd_instance *od_quota_slave_md;
339 /* quota slave instance for block */
340 struct qsd_instance *od_quota_slave_dt;
342 /* osd seq instance */
343 struct lu_client_seq *od_cl_seq;
344 /* If the ratio of "the total OI mappings count" vs
345 * "the bad OI mappings count" is lower than the
346 * osd_device::od_full_scrub_ratio, then trigger
347 * OI scrub to scan the whole the device. */
348 __u64 od_full_scrub_ratio;
349 /* If the speed of found bad OI mappings (per minute)
350 * exceeds the osd_device::od_full_scrub_threshold_rate,
351 * then trigger OI scrub to scan the whole device. */
352 __u64 od_full_scrub_threshold_rate;
354 /* a list of orphaned agent inodes, protected with od_osfs_lock */
355 struct list_head od_orphan_list;
356 struct list_head od_index_backup_list;
357 struct list_head od_index_restore_list;
359 struct inode *od_index_backup_inode;
360 enum lustre_index_backup_policy od_index_backup_policy;
361 int od_index_backup_stop;
362 /* T10PI type, zero if not supported */
363 enum osd_t10_type od_t10_type;
364 atomic_t od_commit_cb_in_flight;
365 wait_queue_head_t od_commit_cb_done;
366 unsigned int __percpu *od_extent_bytes_percpu;
369 static inline struct qsd_instance *osd_def_qsd(struct osd_device *osd)
372 return osd->od_quota_slave_dt;
374 return osd->od_quota_slave_md;
377 enum osd_full_scrub_ratio {
378 /* Trigger OI scrub to scan the whole device directly. */
381 /* Because the bad OI mappings count cannot be larger than
382 * the total OI mappints count, then setting OFSR_NEVER means
383 * that the whole device scanning cannot be triggered by auto
384 * detected bad OI mappings during the RPC services. */
386 OFSR_DEFAULT = 10000,
389 #define FULL_SCRUB_THRESHOLD_RATE_DEFAULT 60
391 /* There are at most 15 uid/gid/projids are affected in a transaction, and
392 * that's rename case:
393 * - 3 for source parent uid & gid & projid;
394 * - 3 for source child uid & gid & projid ('..' entry update when
395 * child is directory);
396 * - 3 for target parent uid & gid & projid;
397 * - 3 for target child uid & gid & projid(if the target child exists);
398 * - 3 for root uid & gid(last_rcvd, llog, etc);
401 #define OSD_MAX_UGID_CNT 15
419 struct osd_access_lock {
420 struct list_head tl_list;
421 struct osd_object *tl_obj;
431 struct thandle ot_super;
433 struct ldiskfs_journal_cb_entry ot_jcb;
434 struct list_head ot_commit_dcb_list;
435 struct list_head ot_stop_dcb_list;
436 /* Link to the device, for debugging. */
437 struct lu_ref_link ot_dev_link;
438 unsigned int ot_credits;
439 unsigned int oh_declared_ext;
441 /* quota IDs related to the transaction */
442 unsigned short ot_id_cnt;
443 __u8 ot_id_res[OSD_MAX_UGID_CNT];
444 __u8 ot_id_types[OSD_MAX_UGID_CNT];
445 uid_t ot_id_array[OSD_MAX_UGID_CNT];
446 struct lquota_trans *ot_quota_trans;
448 unsigned int ot_remove_agents:1;
449 #if OSD_THANDLE_STATS
450 /** time when this handle was allocated */
453 /** time when this thanle was started */
456 struct list_head ot_trunc_locks;
458 * list of declarations, used for large transactions to check
459 * for duplicates, like llogs
461 struct list_head ot_declare_list;
465 * Basic transaction credit op
482 struct osd_obj_declare {
483 struct list_head old_list;
484 struct lu_fid old_fid;
485 enum dt_txn_op old_op;
493 #ifdef CONFIG_PROC_FS
495 LPROC_OSD_READ_BYTES = 0,
496 LPROC_OSD_WRITE_BYTES = 1,
497 LPROC_OSD_GET_PAGE = 2,
498 LPROC_OSD_NO_PAGE = 3,
499 LPROC_OSD_CACHE_ACCESS = 4,
500 LPROC_OSD_CACHE_HIT = 5,
501 LPROC_OSD_CACHE_MISS = 6,
503 #if OSD_THANDLE_STATS
504 LPROC_OSD_THANDLE_STARTING,
505 LPROC_OSD_THANDLE_OPEN,
506 LPROC_OSD_THANDLE_CLOSING,
508 LPROC_OSD_TOO_MANY_CREDITS,
514 * Storage representation for fids.
516 * Variable size, first byte contains the length of the whole record.
518 struct osd_fid_pack {
519 unsigned char fp_len;
520 char fp_area[sizeof(struct lu_fid)];
523 struct osd_it_ea_dirent {
524 struct lu_fid oied_fid;
527 unsigned short oied_namelen;
528 unsigned int oied_type;
530 } __attribute__((packed));
533 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
534 * than lu_dirent struct. osd readdir reads less number of dirent than
535 * required for mdd dir page. so buffer size need to be increased so that
536 * there would be one ext3 readdir for every mdd readdir page.
539 #define OSD_IT_EA_BUFSIZE (PAGE_SIZE + PAGE_SIZE/4)
542 * This is iterator's in-memory data structure in interoperability
543 * mode (i.e. iterator over ldiskfs style directory)
546 struct osd_object *oie_obj;
547 /** used in ldiskfs iterator, to stored file pointer */
548 struct file *oie_file;
549 /** how many entries have been read-cached from storage */
551 /** current entry is being iterated by caller */
553 /** current processing entry */
554 struct osd_it_ea_dirent *oie_dirent;
555 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
560 * Iterator's in-memory data structure for IAM mode.
563 struct osd_object *oi_obj;
564 struct iam_path_descr *oi_ipd;
565 struct iam_iterator oi_it;
568 struct osd_quota_leaf {
569 struct list_head oql_link;
574 * Iterator's in-memory data structure for quota file.
576 struct osd_it_quota {
577 struct osd_object *oiq_obj;
578 /** tree blocks path to where the entry is stored */
579 uint oiq_blk[LUSTRE_DQTREEDEPTH + 1];
580 /** on-disk offset for current key where quota record can be found */
582 /** identifier for current quota record */
584 /** the record index in the leaf/index block */
585 uint oiq_index[LUSTRE_DQTREEDEPTH + 1];
586 /** list of already processed leaf blocks */
587 struct list_head oiq_list;
590 #define MAX_BLOCKS_PER_PAGE (PAGE_SIZE / 512)
593 wait_queue_head_t dr_wait;
594 atomic_t dr_numreqs; /* number of reqs being processed */
599 unsigned int dr_elapsed_valid:1; /* we really did count time */
600 unsigned int dr_rw:1;
601 struct lu_buf dr_pg_buf;
602 struct page **dr_pages;
603 struct niobuf_local **dr_lnbs;
604 struct lu_buf dr_bl_buf;
605 struct lu_buf dr_lnb_buf;
607 ktime_t dr_start_time;
608 ktime_t dr_elapsed; /* how long io took */
609 struct osd_device *dr_dev;
610 unsigned int dr_init_at; /* the line iobuf was initialized */
613 #define osd_dirty_inode(inode, flag) (inode)->i_sb->s_op->dirty_inode((inode), flag)
614 #define osd_i_blocks(inode, size) ((size) >> (inode)->i_blkbits)
617 #ifdef HAVE_INODE_TIMESPEC64
618 # define osd_timespec timespec64
620 # define osd_timespec timespec
623 static inline struct osd_timespec osd_inode_time(struct inode *inode,
626 struct osd_timespec ts = { .tv_sec = seconds };
632 * any object can be identified unique way:
634 * - inode + generation (unique object within a given filesystem)
636 * version increments each insert/delete in a directory
638 struct osd_lookup_cache_object {
639 struct osd_device *lco_osd;
645 struct osd_lookup_cache_entry {
646 struct osd_lookup_cache_object lce_obj;
647 struct lu_fid lce_fid;
650 char lce_name[LDISKFS_NAME_LEN];
653 #define OSD_LOOKUP_CACHE_MAX 3
655 struct osd_lookup_cache {
656 struct osd_lookup_cache_entry olc_entry[OSD_LOOKUP_CACHE_MAX];
661 #define OSD_INS_CACHE_SIZE 8
663 struct osd_thread_info {
664 const struct lu_env *oti_env;
666 * used for index operations.
668 struct dentry oti_obj_dentry;
669 struct dentry oti_child_dentry;
671 /** dentry for Iterator context. */
672 struct dentry oti_it_dentry;
675 struct kstatfs oti_ksfs;
677 struct htree_lock *oti_hlock;
679 struct lu_fid oti_fid;
680 struct lu_fid oti_fid2;
681 struct lu_fid oti_fid3;
682 struct osd_inode_id oti_id;
683 struct osd_inode_id oti_id2;
684 struct osd_inode_id oti_id3;
685 struct ost_id oti_ostid;
688 * following ipd and it structures are used for osd_index_iam_lookup()
689 * these are defined separately as we might do index operation
690 * in open iterator session.
693 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
695 unsigned int oti_it_ea_buf_used:1;
697 /* IAM iterator for index operation. */
698 struct iam_iterator oti_idx_it;
700 /** union to guarantee that ->oti_ipd[] has proper alignment. */
703 char oti_it_ipd[DX_IPD_MAX_SIZE];
704 long long oti_alignment_lieutenant;
708 char oti_idx_ipd[DX_IPD_MAX_SIZE];
709 long long oti_alignment_lieutenant_colonel;
712 struct osd_idmap_cache oti_cache;
714 /* dedicated OI cache for insert (which needs inum) */
715 struct osd_idmap_cache *oti_ins_cache;
716 int oti_ins_cache_size;
717 int oti_ins_cache_used;
718 /* inc by osd_trans_create and dec by osd_trans_stop */
719 int oti_ins_cache_depth;
724 /** used in osd_fid_set() to put xattr */
725 struct lu_buf oti_buf;
726 struct lu_buf oti_big_buf;
727 /** used in osd_ea_fid_set() to set fid into common ea */
729 struct lustre_ost_attrs oti_ost_attrs;
730 struct filter_fid_18_23 oti_ff_old;
731 struct filter_fid oti_ff;
734 struct osd_iobuf oti_iobuf;
735 /* used to access objects in /O */
736 struct inode *oti_inode;
737 #define OSD_FID_REC_SZ 32
738 char oti_ldp[OSD_FID_REC_SZ];
739 char oti_ldp2[OSD_FID_REC_SZ];
741 /* used by quota code */
743 #if defined(HAVE_DQUOT_QC_DQBLK)
744 struct qc_dqblk oti_qdq;
746 struct fs_disk_quota oti_fdq;
748 struct if_dqinfo oti_dqinfo;
750 struct lquota_id_info oti_qi;
751 struct lquota_trans oti_quota_trans;
752 union lquota_rec oti_quota_rec;
754 struct lu_seq_range oti_seq_range;
756 /* Tracking for transaction credits, to allow debugging and optimizing
757 * cases where a large number of credits are being allocated for
758 * single transaction. */
759 unsigned int oti_credits_before;
760 unsigned int oti_declare_ops[OSD_OT_MAX];
761 unsigned int oti_declare_ops_cred[OSD_OT_MAX];
762 unsigned int oti_declare_ops_used[OSD_OT_MAX];
763 struct osd_directory oti_iam;
765 struct page **oti_dio_pages;
766 int oti_dio_pages_used;
768 struct osd_it_ea_dirent *oti_seq_dirent;
769 struct osd_it_ea_dirent *oti_dir_dirent;
771 struct osd_lookup_cache_object oti_cobj; /* cache object id */
772 struct osd_lookup_cache *oti_lookup_cache;
775 extern int ldiskfs_pdo;
777 /* autoconf test is in lustre-build-ldiskfs.m4 */
778 #ifdef HAVE_BVEC_ITER_ALL
779 #define DECLARE_BVEC_ITER_ALL(iter) struct bvec_iter_all iter
781 #define DECLARE_BVEC_ITER_ALL(iter) int iter
784 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
785 const char *name, void *buf, int len)
790 dentry->d_inode = inode;
791 dentry->d_sb = inode->i_sb;
792 return ll_vfs_getxattr(dentry, inode, name, buf, len);
795 static inline int __osd_xattr_set(struct osd_thread_info *info,
796 struct inode *inode, const char *name,
797 const void *buf, int buflen, int fl)
799 struct dentry *dentry = &info->oti_child_dentry;
801 dquot_initialize(inode);
802 dentry->d_inode = inode;
803 dentry->d_sb = inode->i_sb;
804 return ll_vfs_setxattr(dentry, inode, name, buf, buflen, fl);
807 static inline char *osd_oid_name(char *name, size_t name_size,
808 const struct lu_fid *fid, u64 id)
810 snprintf(name, name_size,
811 (fid_seq_is_rsvd(fid_seq(fid)) ||
812 fid_seq_is_mdt0(fid_seq(fid)) ||
813 fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
818 #ifdef CONFIG_PROC_FS
820 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
821 int osd_procfs_init(struct osd_device *osd, const char *name);
822 int osd_procfs_fini(struct osd_device *osd);
823 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf);
824 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
825 int osd_register_proc_index_in_idif(struct osd_device *osd);
829 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
830 struct obd_statfs *sfs, struct obd_statfs_info *info);
831 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
832 struct osd_inode_id *id);
834 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
835 struct osd_inode_id *id, struct lu_fid *fid);
836 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
837 const struct lu_fid *fid, __u32 compat, __u32 incompat);
838 struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
839 struct osd_device *osd, u64 seq);
840 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
841 struct dentry *dentry, struct lustre_ost_attrs *loa);
842 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
843 struct osd_inode_id *id, const struct lu_fid *fid);
844 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
845 struct dentry *dentry, struct lu_fid *fid);
846 struct osd_it_ea *osd_it_dir_init(const struct lu_env *env,
847 struct osd_device *dev, struct inode *inode,
849 void osd_it_dir_fini(const struct lu_env *env, struct osd_it_ea *oie,
850 struct inode *inode);
851 int osd_ldiskfs_it_fill(const struct lu_env *env, const struct dt_it *di);
853 int osd_obj_map_init(const struct lu_env *env, struct osd_device *osd);
854 void osd_obj_map_fini(struct osd_device *dev);
855 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *osd,
856 const struct lu_fid *fid, struct osd_inode_id *id);
857 int osd_obj_map_insert(struct osd_thread_info *info, struct osd_device *osd,
858 const struct lu_fid *fid, const struct osd_inode_id *id,
860 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
861 const struct lu_fid *fid, handle_t *th);
862 int osd_obj_map_update(struct osd_thread_info *info, struct osd_device *osd,
863 const struct lu_fid *fid, const struct osd_inode_id *id,
865 int osd_obj_map_recover(struct osd_thread_info *info, struct osd_device *osd,
866 struct inode *src_parent, struct dentry *src_child,
867 const struct lu_fid *fid);
868 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
869 const struct lu_fid *fid, struct osd_inode_id *id,
870 enum oi_check_flags flags);
871 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
872 const struct lu_fid *fid, const struct osd_inode_id *id,
874 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
875 const struct lu_fid *fid, const struct osd_inode_id *id,
877 int osd_obj_del_entry(struct osd_thread_info *info, struct osd_device *osd,
878 struct dentry *dird, char *name, int namelen,
881 char *osd_lf_fid2name(const struct lu_fid *fid);
882 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
884 void osd_scrub_stop(struct osd_device *dev);
885 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev,
887 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev);
888 int osd_scrub_oi_insert(struct osd_device *dev, const struct lu_fid *fid,
889 struct osd_inode_id *id, int insert);
890 void osd_scrub_oi_resurrect(struct lustre_scrub *scrub,
891 const struct lu_fid *fid);
892 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev);
894 struct dentry *osd_lookup_one_len_unlocked(struct osd_device *dev,
896 struct dentry *base, int len);
897 struct dentry *osd_lookup_one_len(struct osd_device *dev, const char *name,
898 struct dentry *base, int len);
900 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
901 u64 seq, struct lu_seq_range *range);
903 int osd_delete_from_remote_parent(const struct lu_env *env,
904 struct osd_device *osd,
905 struct osd_object *obj,
906 struct osd_thandle *oh, bool destroy);
907 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
908 struct osd_object *obj, struct osd_thandle *oh);
909 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
910 struct osd_device *osd,
911 const struct lu_fid *fid,
912 struct osd_inode_id *id);
914 int osd_ost_seq_exists(struct osd_thread_info *info, struct osd_device *osd,
916 int osd_scrub_refresh_mapping(struct osd_thread_info *info,
917 struct osd_device *dev,
918 const struct lu_fid *fid,
919 const struct osd_inode_id *id,
921 enum oi_check_flags flags, bool *exist);
923 /* osd_quota_fmt.c */
924 int walk_tree_dqentry(const struct lu_env *env, struct osd_object *obj,
925 int type, uint blk, int depth, uint index,
926 struct osd_it_quota *it);
927 int walk_block_dqentry(const struct lu_env *env, struct osd_object *obj,
928 int type, uint blk, uint index,
929 struct osd_it_quota *it);
930 loff_t find_tree_dqentry(const struct lu_env *env,
931 struct osd_object *obj, int type,
932 qid_t dqid, uint blk, int depth,
933 struct osd_it_quota *it);
935 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
936 struct lquota_id_info *qi, struct osd_object *obj,
937 bool enforce, enum osd_quota_local_flags *local_flags);
938 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
939 __u32 projid, long long space, struct osd_thandle *oh,
940 struct osd_object *obj,
941 enum osd_quota_local_flags *local_flags,
942 enum osd_qid_declare_flags);
943 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
944 const struct dt_rec *rec,
945 union lquota_rec *quota_rec);
946 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec);
948 #ifdef HAVE_PROJECT_QUOTA
949 static inline __u32 i_projid_read(struct inode *inode)
951 return (__u32)from_kprojid(&init_user_ns, LDISKFS_I(inode)->i_projid);
954 static inline void i_projid_write(struct inode *inode, __u32 projid)
957 kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
958 LDISKFS_I(inode)->i_projid = kprojid;
961 static inline uid_t i_projid_read(struct inode *inode)
965 static inline void i_projid_write(struct inode *inode, __u32 projid)
971 #ifdef HAVE_LDISKFS_IGET_WITH_FLAGS
972 # define osd_ldiskfs_iget(sb, ino) \
973 ldiskfs_iget((sb), (ino), \
974 LDISKFS_IGET_HANDLE | LDISKFS_IGET_SPECIAL)
976 # define osd_ldiskfs_iget(sb, ino) ldiskfs_iget((sb), (ino))
979 #ifdef HAVE_LDISKFS_INFO_JINODE
980 # define osd_attach_jinode(inode) ldiskfs_inode_attach_jinode(inode)
981 #else /* HAVE_LDISKFS_INFO_JINODE */
982 # define osd_attach_jinode(inode) 0
983 #endif /* HAVE_LDISKFS_INFO_JINODE */
985 #ifndef HAVE_JBD2_JOURNAL_GET_MAX_TXN_BUFS
986 #define jbd2_journal_get_max_txn_bufs(jrnl) \
987 (jrnl->j_max_transaction_buffers)
990 #ifdef LDISKFS_HT_MISC
991 # define osd_journal_start_sb(sb, type, nblock) \
992 ldiskfs_journal_start_sb(sb, type, nblock)
993 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
995 ldiskfs_lblk_t *nblock)
999 rc = osd_attach_jinode(inode);
1002 return ldiskfs_append(handle, inode, nblock);
1005 # ifdef HAVE___LDISKFS_FIND_ENTRY
1006 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1007 (__ldiskfs_find_entry(dir, name, de, inlined, lock) ?: \
1010 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1011 (ldiskfs_find_entry_locked(dir, name, de, inlined, lock) ?: \
1015 # define osd_journal_start(inode, type, nblocks) \
1016 ldiskfs_journal_start(inode, type, nblocks)
1017 # define osd_transaction_size(dev) \
1018 (jbd2_journal_get_max_txn_bufs(osd_journal(dev)) / 2)
1019 #else /* ! defined LDISKFS_HT_MISC */
1020 # define LDISKFS_HT_MISC 0
1021 # define osd_journal_start_sb(sb, type, nblock) \
1022 ldiskfs_journal_start_sb(sb, nblock)
1024 static inline struct buffer_head *osd_ldiskfs_append(handle_t *handle,
1025 struct inode *inode,
1026 ldiskfs_lblk_t *nblock)
1028 struct buffer_head *bh;
1031 bh = ldiskfs_append(handle, inode, nblock, &err);
1038 # ifdef HAVE___LDISKFS_FIND_ENTRY
1039 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1040 (__ldiskfs_find_entry(dir, name, de, lock) ?: \
1043 # define osd_ldiskfs_find_entry(dir, name, de, inlined, lock) \
1044 (ldiskfs_find_entry_locked(dir, name, de, lock) ?: \
1047 # define osd_journal_start(inode, type, nblocks) \
1048 ldiskfs_journal_start(inode, nblocks)
1049 # define osd_transaction_size(dev) \
1050 jbd2_journal_get_max_txn_bufs(osd_journal(dev))
1051 #endif /* LDISKFS_HT_MISC */
1053 #ifndef HAVE___LDISKFS_FIND_ENTRY
1054 # define __ldiskfs_add_entry(handle, child, inode, hlock) \
1055 ldiskfs_add_entry_locked(handle, child, inode, hlock)
1059 * Invariants, assertions.
1063 * XXX: do not enable this, until invariant checking code is made thread safe
1064 * in the face of pdirops locking.
1066 #define OSD_INVARIANT_CHECKS (0)
1068 #if OSD_INVARIANT_CHECKS
1069 static inline int osd_invariant(const struct osd_object *obj)
1073 ergo(obj->oo_inode != NULL,
1074 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
1075 atomic_read(&obj->oo_inode->i_count) > 0) &&
1076 ergo(obj->oo_dir != NULL &&
1077 obj->oo_dir->od_conationer.ic_object != NULL,
1078 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
1081 #define osd_invariant(obj) (1)
1084 #define OSD_MAX_CACHE_SIZE OBD_OBJECT_EOF
1085 #define OSD_READCACHE_MAX_IO_MB 8
1086 #define OSD_WRITECACHE_MAX_IO_MB 8
1088 extern const struct dt_index_operations osd_otable_ops;
1090 static inline int osd_oi_fid2idx(struct osd_device *dev,
1091 const struct lu_fid *fid)
1093 return fid->f_seq & (dev->od_oi_count - 1);
1096 static inline struct osd_oi *osd_fid2oi(struct osd_device *osd,
1097 const struct lu_fid *fid)
1099 LASSERTF(!fid_is_idif(fid), DFID"\n", PFID(fid));
1100 LASSERTF(!fid_is_last_id(fid), DFID"\n", PFID(fid));
1101 LASSERTF(osd->od_oi_table != NULL && osd->od_oi_count >= 1,
1102 DFID"\n", PFID(fid));
1103 /* It can work even od_oi_count equals to 1 although it's unexpected,
1104 * the only reason we set it to 1 is for performance measurement */
1105 return osd->od_oi_table[osd_oi_fid2idx(osd, fid)];
1108 extern const struct lu_device_operations osd_lu_ops;
1110 static inline int lu_device_is_osd(const struct lu_device *d)
1112 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
1115 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
1117 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
1118 return container_of(d, struct osd_device, od_dt_dev);
1121 static inline struct osd_device *osd_dev(const struct lu_device *d)
1123 LASSERT(lu_device_is_osd(d));
1124 return osd_dt_dev(container_of(d, struct dt_device, dd_lu_dev));
1127 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
1129 return osd_dev(o->oo_dt.do_lu.lo_dev);
1132 static inline struct super_block *osd_sb(const struct osd_device *dev)
1134 return dev->od_mnt->mnt_sb;
1137 static inline const char *osd_sb2name(const struct super_block *sb)
1139 /* this is LDISKFS_SB(sb), but preserves "const" */
1140 const struct ldiskfs_sb_info *sbi = sb->s_fs_info;
1142 return sbi->s_es->s_volume_name;
1145 static inline const char *osd_dev2name(const struct osd_device *dev)
1147 return osd_sb2name(osd_sb(dev));
1150 static inline const char *osd_ino2name(const struct inode *inode)
1152 return osd_sb2name(inode->i_sb);
1156 * Put the osd object once done with it.
1158 * \param obj osd object that needs to be put
1160 static inline void osd_object_put(const struct lu_env *env,
1161 struct osd_object *obj)
1163 dt_object_put(env, &obj->oo_dt);
1166 static inline int osd_object_is_root(const struct osd_object *obj)
1168 return obj->oo_inode && is_root_inode(obj->oo_inode);
1171 static inline struct osd_object *osd_obj(const struct lu_object *o)
1173 LASSERT(lu_device_is_osd(o->lo_dev));
1174 return container_of(o, struct osd_object, oo_dt.do_lu);
1177 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
1179 return osd_obj(&d->do_lu);
1182 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
1184 return &osd->od_dt_dev.dd_lu_dev;
1187 static inline journal_t *osd_journal(const struct osd_device *dev)
1189 return LDISKFS_SB(osd_sb(dev))->s_journal;
1192 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
1194 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
1197 static inline char *osd_name(struct osd_device *osd)
1199 return osd->od_svname;
1202 static inline bool osd_is_ea_inode(struct inode *inode)
1204 return !!(LDISKFS_I(inode)->i_flags & LDISKFS_EA_INODE_FL);
1207 extern const struct dt_body_operations osd_body_ops;
1208 extern struct lu_context_key osd_key;
1210 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
1212 return lu_context_key_get(&env->le_ctx, &osd_key);
1215 extern const struct dt_body_operations osd_body_ops_new;
1221 struct iam_path_descr *osd_it_ipd_get(const struct lu_env *env,
1222 const struct iam_container *bag)
1224 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1225 osd_oti_get(env)->oti_it_ipd);
1229 struct iam_path_descr *osd_idx_ipd_get(const struct lu_env *env,
1230 const struct iam_container *bag)
1232 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
1233 osd_oti_get(env)->oti_idx_ipd);
1236 static inline void osd_ipd_put(const struct lu_env *env,
1237 const struct iam_container *bag,
1238 struct iam_path_descr *ipd)
1240 bag->ic_descr->id_ops->id_ipd_free(ipd);
1243 int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode,
1244 const loff_t size, const loff_t pos,
1247 int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1250 struct dentry *osd_child_dentry_by_inode(const struct lu_env *env,
1251 struct inode *inode,
1252 const char *name, const int namelen)
1254 struct osd_thread_info *info = osd_oti_get(env);
1255 struct dentry *child_dentry = &info->oti_child_dentry;
1256 struct dentry *obj_dentry = &info->oti_obj_dentry;
1258 obj_dentry->d_inode = inode;
1259 obj_dentry->d_sb = inode->i_sb;
1260 obj_dentry->d_name.hash = 0;
1262 child_dentry->d_name.hash = 0;
1263 child_dentry->d_parent = obj_dentry;
1264 child_dentry->d_name.name = name;
1265 child_dentry->d_name.len = namelen;
1266 return child_dentry;
1269 extern int osd_trans_declare_op2rb[];
1270 extern int ldiskfs_track_declares_assert;
1271 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th);
1273 static inline void osd_trans_declare_op(const struct lu_env *env,
1274 struct osd_thandle *oh,
1275 enum osd_op_type op, int credits)
1277 struct osd_thread_info *oti = osd_oti_get(env);
1279 LASSERT(oh->ot_handle == NULL);
1280 if (unlikely(op >= OSD_OT_MAX)) {
1281 if (unlikely(ldiskfs_track_declares_assert)) {
1282 LASSERT(op < OSD_OT_MAX);
1284 CWARN("%s: Invalid operation index %d\n",
1285 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op);
1286 libcfs_debug_dumpstack(NULL);
1289 oti->oti_declare_ops[op]++;
1290 oti->oti_declare_ops_cred[op] += credits;
1292 oh->ot_credits += credits;
1295 /* linux: v5.4-rc3-21-g933f1c1e0b75 renamed h_buffer_credits */
1296 #ifdef HAVE_JOURNAL_TOTAL_CREDITS
1297 #define h_buffer_credits h_total_credits
1300 static inline void osd_trans_exec_op(const struct lu_env *env,
1302 enum osd_op_type op)
1304 struct osd_thread_info *oti = osd_oti_get(env);
1305 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1307 unsigned int rb, left;
1309 LASSERT(oh->ot_handle != NULL);
1310 if (unlikely(op >= OSD_OT_MAX)) {
1311 if (unlikely(ldiskfs_track_declares_assert))
1312 LASSERT(op < OSD_OT_MAX);
1314 CWARN("%s: opcode %u: invalid value >= %u\n",
1315 osd_name(osd_dt_dev(oh->ot_super.th_dev)),
1317 libcfs_debug_dumpstack(NULL);
1322 /* find rollback (or reverse) operation for the given one
1323 * such an operation doesn't require additional credits
1324 * as the same set of blocks are modified */
1325 rb = osd_trans_declare_op2rb[op];
1327 /* check whether credits for this operation were reserved at all */
1328 if (unlikely(oti->oti_declare_ops_cred[op] == 0 &&
1329 oti->oti_declare_ops_cred[rb] == 0)) {
1330 /* the API is not perfect yet: CREATE does REF_ADD internally
1331 * while DESTROY does not. To rollback CREATE the callers
1332 * needs to call REF_DEL+DESTROY which is hard to detect using
1333 * a simple table of rollback operations */
1334 if (op == OSD_OT_REF_DEL &&
1335 oti->oti_declare_ops_cred[OSD_OT_CREATE] > 0)
1337 if (op == OSD_OT_REF_ADD &&
1338 oti->oti_declare_ops_cred[OSD_OT_DESTROY] > 0)
1340 CWARN("%s: opcode %u: credits = 0, rollback = %u\n",
1341 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op, rb);
1342 osd_trans_dump_creds(env, th);
1343 LASSERT(!ldiskfs_track_declares_assert);
1347 /* remember how many credits we have unused before the operation */
1348 oti->oti_credits_before = oh->ot_handle->h_buffer_credits;
1349 left = oti->oti_declare_ops_cred[op] - oti->oti_declare_ops_used[op];
1350 if (unlikely(oti->oti_credits_before < left)) {
1351 CWARN("%s: opcode %u: before %u < left %u, rollback = %u\n",
1352 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1353 oti->oti_credits_before, left, rb);
1354 osd_trans_dump_creds(env, th);
1355 /* on a very small fs (testing?) it's possible that
1356 * the transaction can't fit 1/4 of journal, so we
1357 * just request less credits (see osd_trans_start()).
1358 * ignore the same case here */
1359 rb = osd_transaction_size(osd_dt_dev(th->th_dev));
1360 if (unlikely(oh->ot_credits < rb))
1361 LASSERT(!ldiskfs_track_declares_assert);
1365 static inline void osd_trans_exec_check(const struct lu_env *env,
1367 enum osd_op_type op)
1369 struct osd_thread_info *oti = osd_oti_get(env);
1370 struct osd_thandle *oh = container_of(th, struct osd_thandle,
1372 int used, over, quota;
1374 /* how many credits have been used by the operation */
1375 used = oti->oti_credits_before - oh->ot_handle->h_buffer_credits;
1377 if (unlikely(used < 0)) {
1378 /* if some block was allocated and released in the same
1379 * transaction, then it won't be a part of the transaction
1380 * and delta can be negative */
1385 /* rollback operations (e.g. when we destroy just created
1386 * object) should not consume any credits. there is no point
1387 * to confuse the checks below */
1391 oti->oti_declare_ops_used[op] += used;
1392 if (oti->oti_declare_ops_used[op] <= oti->oti_declare_ops_cred[op])
1395 /* we account quota for a whole transaction and any operation can
1396 * consume corresponding credits */
1397 over = oti->oti_declare_ops_used[op] -
1398 oti->oti_declare_ops_cred[op];
1399 quota = oti->oti_declare_ops_cred[OSD_OT_QUOTA] -
1400 oti->oti_declare_ops_used[OSD_OT_QUOTA];
1401 if (over <= quota) {
1402 /* probably that credits were consumed by
1403 * quota indirectly (in the depths of ldiskfs) */
1404 oti->oti_declare_ops_used[OSD_OT_QUOTA] += over;
1405 oti->oti_declare_ops_used[op] -= over;
1407 CWARN("%s: opcode %d: used %u, used now %u, reserved %u\n",
1408 osd_name(osd_dt_dev(oh->ot_super.th_dev)), op,
1409 oti->oti_declare_ops_used[op], used,
1410 oti->oti_declare_ops_cred[op]);
1411 osd_trans_dump_creds(env, th);
1412 if (unlikely(ldiskfs_track_declares_assert))
1418 * Helper function to pack the fid, ldiskfs stores fid in packed format.
1421 void osd_fid_pack(struct osd_fid_pack *pack, const struct dt_rec *fid,
1422 struct lu_fid *befider)
1424 fid_cpu_to_be(befider, (struct lu_fid *)fid);
1425 memcpy(pack->fp_area, befider, sizeof(*befider));
1426 pack->fp_len = sizeof(*befider) + 1;
1430 int osd_fid_unpack(struct lu_fid *fid, const struct osd_fid_pack *pack)
1435 switch (pack->fp_len) {
1436 case sizeof *fid + 1:
1437 memcpy(fid, pack->fp_area, sizeof *fid);
1438 fid_be_to_cpu(fid, fid);
1441 CERROR("Unexpected packed fid size: %d\n", pack->fp_len);
1448 * Quota/Accounting handling
1450 extern const struct dt_index_operations osd_acct_index_ops;
1451 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
1452 const struct lu_fid *fid, struct osd_inode_id *id);
1454 /* copy from fs/ext4/dir.c */
1455 static inline int is_32bit_api(void)
1457 #ifdef CONFIG_COMPAT
1458 return in_compat_syscall();
1460 return (BITS_PER_LONG == 32);
1464 static inline loff_t ldiskfs_get_htree_eof(struct file *filp)
1466 if ((filp->f_mode & FMODE_32BITHASH) ||
1467 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
1468 return LDISKFS_HTREE_EOF_32BIT;
1470 return LDISKFS_HTREE_EOF_64BIT;
1473 static inline int fid_is_internal(const struct lu_fid *fid)
1475 return (!fid_is_namespace_visible(fid) && !fid_is_idif(fid));
1478 static inline bool is_remote_parent_ino(struct osd_device *o, unsigned long ino)
1483 LASSERT(o->od_mdt_map != NULL);
1485 return ino == o->od_mdt_map->omm_remote_parent->d_inode->i_ino;
1489 * ext4_bread/ldiskfs_bread has either 5 or 4 parameters. The error
1490 * return code has been removed and integrated into the pointer in the
1493 static inline struct buffer_head *__ldiskfs_bread(handle_t *handle,
1494 struct inode *inode,
1495 ldiskfs_lblk_t block,
1499 struct buffer_head *bh;
1502 rc = osd_attach_jinode(inode);
1506 #ifdef HAVE_EXT4_BREAD_4ARGS
1507 bh = ldiskfs_bread(handle, inode, block, create);
1510 bh = ldiskfs_bread(handle, inode, block, create, &rc);
1511 if (bh == NULL && rc != 0)
1517 #ifdef HAVE_EXT4_JOURNAL_GET_WRITE_ACCESS_4ARGS
1518 # define osd_ldiskfs_journal_get_write_access(handle, sb, bh, flags) \
1519 ldiskfs_journal_get_write_access((handle), (sb), (bh), (flags))
1521 # define LDISKFS_JTR_NONE 0
1522 # define osd_ldiskfs_journal_get_write_access(handle, sb, bh, flags) \
1523 ldiskfs_journal_get_write_access((handle), (bh))
1524 #endif /* HAVE_EXT4_JOURNAL_GET_WRITE_ACCESS_4ARGS */
1526 #ifdef HAVE_EXT4_INC_DEC_COUNT_2ARGS
1527 #define osd_ldiskfs_inc_count(h, inode) ldiskfs_inc_count((h), (inode))
1528 #define osd_ldiskfs_dec_count(h, inode) ldiskfs_dec_count((h), (inode))
1530 #define osd_ldiskfs_inc_count(h, inode) ldiskfs_inc_count((inode))
1531 #define osd_ldiskfs_dec_count(h, inode) ldiskfs_dec_count((inode))
1532 #endif /* HAVE_EXT4_INC_DEC_COUNT_2ARGS */
1534 void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf);
1537 osd_index_register(struct osd_device *osd, const struct lu_fid *fid,
1538 __u32 keysize, __u32 recsize)
1540 return lustre_index_register(&osd->od_dt_dev, osd_name(osd),
1541 &osd->od_index_backup_list, &osd->od_lock,
1542 &osd->od_index_backup_stop,
1543 fid, keysize, recsize);
1547 osd_index_backup(const struct lu_env *env, struct osd_device *osd, bool backup)
1549 struct osd_thread_info *info = osd_oti_get(env);
1550 struct lu_fid *fid = &info->oti_fid3;
1551 struct osd_inode_id *id = &info->oti_id3;
1553 lu_local_obj_fid(fid, INDEX_BACKUP_OID);
1554 osd_id_gen(id, osd->od_index_backup_inode->i_ino,
1555 osd->od_index_backup_inode->i_generation);
1556 osd_add_oi_cache(info, osd, id, fid);
1558 lustre_index_backup(env, &osd->od_dt_dev, osd_name(osd),
1559 &osd->od_index_backup_list, &osd->od_lock,
1560 &osd->od_index_backup_stop, backup);
1563 #ifdef LDISKFS_HAS_INCOMPAT_FEATURE
1565 # ifdef LDISKFS_FEATURE_INCOMPAT_EXTENTS
1566 # define ldiskfs_has_feature_extents(sb) \
1567 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS)
1569 # ifdef LDISKFS_FEATURE_INCOMPAT_EA_INODE
1570 # define ldiskfs_has_feature_ea_inode(sb) \
1571 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1573 # ifdef LDISKFS_FEATURE_INCOMPAT_DIRDATA
1574 # define ldiskfs_has_feature_dirdata(sb) \
1575 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_DIRDATA)
1577 # ifdef LDISKFS_FEATURE_INCOMPAT_LARGEDIR
1578 # define ldiskfs_has_feature_largedir(sb) \
1579 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_LARGEDIR)
1580 # define ldiskfs_set_feature_largedir(sb) \
1581 LDISKFS_SET_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_LARGEDIR)
1583 # ifdef LDISKFS_FEATURE_COMPAT_HAS_JOURNAL
1584 # define ldiskfs_has_feature_journal(sb) \
1585 LDISKFS_HAS_COMPAT_FEATURE(sb, LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)
1587 # ifdef LDISKFS_FEATURE_RO_COMPAT_QUOTA
1588 # define ldiskfs_has_feature_quota(sb) \
1589 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_QUOTA)
1591 # ifdef LDISKFS_FEATURE_RO_COMPAT_PROJECT
1592 # define ldiskfs_has_feature_project(sb) \
1593 LDISKFS_HAS_RO_COMPAT_FEATURE(sb, LDISKFS_FEATURE_RO_COMPAT_PROJECT)
1595 # ifdef LDISKFS_FEATURE_COMPAT_FAST_COMMIT
1596 # define ldiskfs_has_feature_fast_commit(sb) \
1597 LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_COMPAT_FAST_COMMIT)
1601 #ifndef ldiskfs_has_feature_fast_commit
1602 #define ldiskfs_has_feature_fast_commit(sb) false
1605 int osd_trunc_lock(struct osd_object *obj, struct osd_thandle *oh,
1607 void osd_trunc_unlock_all(const struct lu_env *env, struct list_head *list);
1608 int osd_process_truncates(const struct lu_env *env, struct list_head *list);
1609 void osd_execute_truncate(struct osd_object *obj);
1611 #ifndef HAVE___BI_CNT
1612 #define __bi_cnt bi_cnt
1615 #ifndef HAVE_CLEAN_BDEV_ALIASES
1616 #define clean_bdev_aliases(bdev, block, len) \
1617 unmap_underlying_metadata((bdev), (block))
1620 #ifndef HAVE_BI_STATUS
1621 #define bi_status bi_error
1625 * Maximum size of xattr attributes for FEATURE_INCOMPAT_EA_INODE 1Mb
1626 * This limit is arbitrary, but is reasonable for the xattr API.
1628 #define LDISKFS_XATTR_MAX_LARGE_EA_SIZE (1024 * 1024)
1630 struct osd_bio_private {
1631 struct osd_iobuf *obp_iobuf;
1632 /* Start page index in the obp_iobuf for the bio */
1633 int obp_start_page_idx;
1635 extern struct kmem_cache *biop_cachep;
1637 #if IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) && defined(HAVE_BIO_INTEGRITY_PREP_FN)
1638 int osd_bio_integrity_handle(struct osd_device *osd, struct bio *bio,
1639 struct osd_iobuf *iobuf,
1640 int start_page_idx, bool fault_inject,
1641 bool integrity_enabled);
1642 #else /* !CONFIG_BLK_DEV_INTEGRITY */
1643 #define osd_bio_integrity_handle(osd, bio, iobuf, start_page_idx, \
1644 fault_inject, integrity_enabled) 0
1647 #ifdef HAVE_BIO_INTEGRITY_PREP_FN
1648 # ifdef HAVE_BLK_INTEGRITY_ITER
1649 # define integrity_gen_fn integrity_processing_fn
1650 # define integrity_vrfy_fn integrity_processing_fn
1652 int osd_get_integrity_profile(struct osd_device *osd,
1653 integrity_gen_fn **generate_fn,
1654 integrity_vrfy_fn **verify_fn);
1655 #endif /* HAVE_BIO_INTEGRITY_PREP_FN */
1657 #ifdef HAVE_BIO_BI_PHYS_SEGMENTS
1658 #define osd_bio_nr_segs(bio) ((bio)->bi_phys_segments)
1660 #define osd_bio_nr_segs(bio) bio_segments((bio))
1661 #endif /* HAVE_BIO_BI_PHYS_SEGMENTS */
1663 #ifdef HAVE_GET_INODE_USAGE
1664 #define lock_dquot_transfer(inode) down_read(&LDISKFS_I(inode)->xattr_sem)
1665 #define unlock_dquot_transfer(inode) up_read(&LDISKFS_I(inode)->xattr_sem)
1667 #define lock_dquot_transfer(inode) do {} while (0)
1668 #define unlock_dquot_transfer(inode) do {} while (0)
1671 #if IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)
1672 static inline unsigned short blk_integrity_interval(struct blk_integrity *bi)
1674 #ifdef HAVE_INTERVAL_EXP_BLK_INTEGRITY
1675 return bi->interval_exp ? 1 << bi->interval_exp : 0;
1676 #elif defined(HAVE_INTERVAL_BLK_INTEGRITY)
1677 return bi->interval;
1679 return bi->sector_size;
1680 #endif /* !HAVE_INTERVAL_EXP_BLK_INTEGRITY */
1683 static inline const char *blk_integrity_name(struct blk_integrity *bi)
1685 #ifdef HAVE_INTERVAL_EXP_BLK_INTEGRITY
1686 return bi->profile->name;
1692 static inline unsigned int bip_size(struct bio_integrity_payload *bip)
1694 #ifdef HAVE_BIP_ITER_BIO_INTEGRITY_PAYLOAD
1695 return bip->bip_iter.bi_size;
1697 return bip->bip_size;
1700 #else /* !CONFIG_BLK_DEV_INTEGRITY */
1701 static inline unsigned short blk_integrity_interval(struct blk_integrity *bi)
1705 static inline const char *blk_integrity_name(struct blk_integrity *bi)
1707 /* gcc8 dislikes when strcmp() is called against NULL */
1710 #endif /* !CONFIG_BLK_DEV_INTEGRITY */
1712 #ifndef INTEGRITY_FLAG_READ
1713 #define INTEGRITY_FLAG_READ BLK_INTEGRITY_VERIFY
1716 #ifndef INTEGRITY_FLAG_WRITE
1717 #define INTEGRITY_FLAG_WRITE BLK_INTEGRITY_GENERATE
1720 static inline bool bdev_integrity_enabled(struct block_device *bdev, int rw)
1722 #if IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY)
1723 struct blk_integrity *bi = bdev_get_integrity(bdev);
1728 #ifdef HAVE_INTERVAL_EXP_BLK_INTEGRITY
1729 if (rw == 0 && bi->profile->verify_fn != NULL &&
1730 (bi->flags & INTEGRITY_FLAG_READ))
1733 if (rw == 1 && bi->profile->generate_fn != NULL &&
1734 (bi->flags & INTEGRITY_FLAG_WRITE))
1737 if (rw == 0 && bi->verify_fn != NULL &&
1738 (bi->flags & INTEGRITY_FLAG_READ))
1741 if (rw == 1 && bi->generate_fn != NULL &&
1742 (bi->flags & INTEGRITY_FLAG_WRITE))
1744 #endif /* !HAVE_INTERVAL_EXP_BLK_INTEGRITY */
1745 #endif /* !CONFIG_BLK_DEV_INTEGRITY */
1750 bool osd_tx_was_declared(const struct lu_env *env, struct osd_thandle *oth,
1751 struct dt_object *dt, enum dt_txn_op op, loff_t p);
1753 #ifdef HAVE_DQUOT_TRANSFER_WITH_USER_NS
1754 #define osd_dquot_transfer(ns, i, a) dquot_transfer((ns), (i), (a))
1756 #define osd_dquot_transfer(ns, i, a) dquot_transfer((i), (a))
1759 #ifdef HAVE_FILLDIR_USE_CTX_RETURN_BOOL
1760 #define WRAP_FILLDIR_FN(prefix, fill_fn) \
1761 static bool fill_fn(struct dir_context *buf, const char *name, int namelen, \
1762 loff_t offset, __u64 ino, unsigned int d_type) \
1764 return !prefix##fill_fn(buf, name, namelen, offset, ino, d_type); \
1766 #elif defined(HAVE_FILLDIR_USE_CTX)
1767 #define WRAP_FILLDIR_FN(prefix, fill_fn) \
1768 static int fill_fn(struct dir_context *buf, const char *name, int namelen, \
1769 loff_t offset, __u64 ino, unsigned int d_type) \
1771 return prefix##fill_fn(buf, name, namelen, offset, ino, d_type); \
1774 #define WRAP_FILLDIR_FN(prefix, fill_fn)
1777 #endif /* _OSD_INTERNAL_H */