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) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 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-zfs/osd_internal.h
33 * Shared definitions and declarations for zfs/dmu osd
35 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
36 * Author: Mike Pershin <tappro@whamcloud.com>
37 * Author: Johann Lombardi <johann@whamcloud.com>
40 #ifndef _OSD_INTERNAL_H
41 #define _OSD_INTERNAL_H
43 #include <dt_object.h>
44 #include <md_object.h>
45 #include <lustre_quota.h>
50 #include <sys/nvpair.h>
51 #include <sys/zfs_znode.h>
56 * By design including kmem.h overrides the Linux slab interfaces to provide
57 * the Illumos kmem cache interfaces. To override this and gain access to
58 * the Linux interfaces these preprocessor macros must be undefined.
60 #ifdef kmem_cache_destroy
61 #undef kmem_cache_destroy
64 #ifdef kmem_cache_create
65 #undef kmem_cache_create
68 #ifdef kmem_cache_alloc
69 #undef kmem_cache_alloc
72 #ifdef kmem_cache_free
73 #undef kmem_cache_free
76 #define LUSTRE_ROOT_FID_SEQ 0
77 #define DMU_OSD_SVNAME "svname"
78 #define DMU_OSD_OI_NAME_BASE "oi"
80 #define OSD_GFP_IO (GFP_NOFS | __GFP_HIGHMEM)
82 /* Statfs space reservation for grant, fragmentation, and unlink space. */
83 #define OSD_STATFS_RESERVED_SIZE (16ULL << 20) /* reserve 16MB minimum */
84 #define OSD_STATFS_RESERVED_SHIFT (7) /* reserve 0.78% of all space */
86 /* Statfs {minimum, safe estimate, and maximum} dnodes per block */
87 #define OSD_DNODE_MIN_BLKSHIFT (DNODES_PER_BLOCK_SHIFT)
88 #define OSD_DNODE_EST_BLKSHIFT (12) /* est 4KB/dnode */
89 #define OSD_DNODE_EST_COUNT 4096
91 #define OSD_GRANT_FOR_LOCAL_OIDS (2ULL << 20) /* 2MB for last_rcvd, ... */
94 * Iterator's in-memory data structure for quota file.
97 struct osd_object *oiq_obj;
98 /* DMU accounting object id */
101 zap_cursor_t *oiq_zc;
102 /** identifier for current quota record */
104 unsigned oiq_reset:1; /* 1 -- no need to advance */
108 * Iterator's in-memory data structure for ZAPs
110 * ZFS does not store . and .. on a disk, instead they are
111 * generated up on request
112 * to follow this format we do the same
115 zap_cursor_t *ozi_zc;
116 struct osd_object *ozi_obj;
117 unsigned ozi_reset:1; /* 1 -- no need to advance */
118 /* ozi_pos - position of the cursor:
119 * 0 - before any record
122 * 3 - real records */
125 char ozi_name[MAXNAMELEN]; /* file name for dir */
126 __u64 ozi_key; /* binary key for index files */
129 #define DT_IT2DT(it) (&((struct osd_zap_it *)it)->ozi_obj->oo_dt)
132 * regular ZFS direntry
134 struct zpl_direntry {
135 uint64_t zde_dnode:48,
138 } __attribute__((packed));
141 * lustre direntry adds a fid to regular ZFS direntry
143 struct luz_direntry {
144 struct zpl_direntry lzd_reg;
145 struct lu_fid lzd_fid;
146 } __attribute__((packed));
149 /* cached SA attributes */
164 #define OSD_INS_CACHE_SIZE 8
167 struct osd_idmap_cache {
168 struct osd_device *oic_dev;
169 struct lu_fid oic_fid;
170 /** max 2^48 dnodes per dataset, avoid spilling into another word */
171 uint64_t oic_dnode:DN_MAX_OBJECT_SHIFT,
172 oic_remote:1; /* FID isn't local */
175 /* max.number of regular attrubites the callers may ask for */
176 #define OSD_MAX_IN_BULK 13
178 struct osd_thread_info {
179 const struct lu_env *oti_env;
181 struct lu_fid oti_fid;
183 * XXX temporary: for ->i_op calls.
185 struct timespec oti_time;
187 struct ost_id oti_ostid;
193 char oti_key[MAXNAMELEN + 1];
194 __u64 oti_key64[(MAXNAMELEN + 1)/sizeof(__u64)];
195 sa_bulk_attr_t oti_attr_bulk[OSD_MAX_IN_BULK];
197 struct lustre_mdt_attrs oti_mdt_attrs;
199 struct lu_attr oti_la;
200 struct osa_attr oti_osa;
201 zap_attribute_t oti_za;
202 dmu_object_info_t oti_doi;
203 struct luz_direntry oti_zde;
205 struct lquota_id_info oti_qi;
206 struct lu_seq_range oti_seq_range;
208 /* dedicated OI cache for insert (which needs inum) */
209 struct osd_idmap_cache *oti_ins_cache;
210 int oti_ins_cache_size;
211 int oti_ins_cache_used;
214 extern struct lu_context_key osd_key;
216 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
218 return lu_context_key_get(&env->le_ctx, &osd_key);
222 struct thandle ot_super;
223 struct list_head ot_dcb_list;
224 struct list_head ot_stop_dcb_list;
225 struct list_head ot_unlinked_list;
226 struct list_head ot_sa_list;
227 struct semaphore ot_sa_lock;
229 struct lquota_trans ot_quota_trans;
230 __u32 ot_write_commit:1,
234 #define OSD_OI_NAME_SIZE 16
237 * Object Index (OI) instance.
240 char oi_name[OSD_OI_NAME_SIZE]; /* unused */
246 uint64_t *os_compat_dirs;
247 int os_subdir_count; /* subdir count for each seq */
248 u64 os_seq; /* seq number */
249 struct list_head os_seq_list; /* list to seq_list */
252 struct osd_seq_list {
253 rwlock_t osl_seq_list_lock; /* lock for seq_list */
254 struct list_head osl_seq_list; /* list head for seq */
255 struct semaphore osl_seq_init_sem;
258 #define OSD_OST_MAP_SIZE 32
265 struct dt_device od_dt_dev;
266 /* information about underlying file system */
267 struct objset *od_os;
268 uint64_t od_rootid; /* id of root znode */
269 uint64_t od_unlinkedid; /* id of unlinked zapobj */
270 /* SA attr mapping->id,
271 * name is the same as in ZFS to use defines SA_ZPL_...*/
272 sa_attr_type_t *z_attr_table;
274 struct proc_dir_entry *od_proc_entry;
275 struct lprocfs_stats *od_stats;
277 uint64_t od_max_blksz;
280 struct osd_oi **od_oi_table;
281 unsigned int od_oi_count;
282 struct osd_seq_list od_seq_list;
284 unsigned int od_dev_set_rdonly:1, /**< osd_ro() called */
285 od_prop_rdonly:1, /**< ZFS property readonly */
287 od_quota_iused_est:1,
295 struct lu_site od_site;
297 /* object IDs of the inode accounting indexes */
298 uint64_t od_iusr_oid;
299 uint64_t od_igrp_oid;
301 /* quota slave instance */
302 struct qsd_instance *od_quota_slave;
304 struct brw_stats od_brw_stats;
305 atomic_t od_r_in_flight;
306 atomic_t od_w_in_flight;
308 /* used to debug zerocopy logic: the fields track all
309 * allocated, loaned and referenced buffers in use.
310 * to be removed once the change is tested well. */
311 atomic_t od_zerocopy_alloc;
312 atomic_t od_zerocopy_loan;
313 atomic_t od_zerocopy_pin;
315 arc_prune_t *arc_prune_cb;
317 /* osd seq instance */
318 struct lu_client_seq *od_cl_seq;
321 enum osd_destroy_type {
322 OSD_DESTROY_NONE = 0,
323 OSD_DESTROY_SYNC = 1,
324 OSD_DESTROY_ASYNC = 2,
328 struct dt_object oo_dt;
330 * Inode for file system object represented by this osd_object. This
331 * inode is pinned for the whole duration of lu_object life.
333 * Not modified concurrently (either setup early during object
334 * creation, or assigned by osd_object_create() under write lock).
337 sa_handle_t *oo_sa_hdl;
338 nvlist_t *oo_sa_xattr;
339 struct list_head oo_sa_linkage;
341 /* used to implement osd_object_*_{lock|unlock} */
342 struct rw_semaphore oo_sem;
344 /* to serialize some updates: destroy vs. others,
345 * xattr_set, object block size change etc */
346 struct rw_semaphore oo_guard;
348 /* protected by oo_guard */
349 struct list_head oo_unlinked_linkage;
351 /* cached attributes */
352 rwlock_t oo_attr_lock;
353 struct lu_attr oo_attr;
355 /* external dnode holding large EAs, protected by oo_guard */
357 enum osd_destroy_type oo_destroy;
359 __u32 oo_destroyed:1;
361 /* the i_flags in LMA */
364 int oo_ea_in_bonus; /* EA bytes we expect */
366 /* record size for index file */
367 unsigned char oo_keysize;
368 unsigned char oo_recsize;
369 unsigned char oo_recusize; /* unit size */
376 int osd_statfs(const struct lu_env *, struct dt_device *, struct obd_statfs *);
377 extern const struct dt_index_operations osd_acct_index_ops;
378 uint64_t osd_quota_fid2dmu(const struct lu_fid *fid);
379 extern struct lu_device_operations osd_lu_ops;
380 extern struct dt_index_operations osd_dir_ops;
381 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
382 qid_t uid, qid_t gid, long long space,
383 struct osd_thandle *oh, bool is_blk, int *flags,
385 uint64_t osd_objs_count_estimate(uint64_t refdbytes, uint64_t usedobjs,
386 uint64_t nrblocks, uint64_t est_maxblockshift);
391 static inline int lu_device_is_osd(const struct lu_device *d)
393 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
396 static inline struct osd_object *osd_obj(const struct lu_object *o)
398 LASSERT(lu_device_is_osd(o->lo_dev));
399 return container_of0(o, struct osd_object, oo_dt.do_lu);
402 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
404 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
405 return container_of0(d, struct osd_device, od_dt_dev);
408 static inline struct osd_device *osd_dev(const struct lu_device *d)
410 LASSERT(lu_device_is_osd(d));
411 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
414 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
416 return osd_obj(&d->do_lu);
419 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
421 return osd_dev(o->oo_dt.do_lu.lo_dev);
424 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
426 return &osd->od_dt_dev.dd_lu_dev;
429 static inline struct objset * osd_dtobj2objset(struct dt_object *o)
431 return osd_dev(o->do_lu.lo_dev)->od_os;
434 static inline int osd_invariant(const struct osd_object *obj)
440 * Put the osd object once done with it.
442 * \param obj osd object that needs to be put
444 static inline void osd_object_put(const struct lu_env *env,
445 struct osd_object *obj)
447 dt_object_put(env, &obj->oo_dt);
450 static inline int osd_object_invariant(const struct lu_object *l)
452 return osd_invariant(osd_obj(l));
455 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
457 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
460 static inline char *osd_name(struct osd_device *osd)
462 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
465 #ifdef CONFIG_PROC_FS
467 LPROC_OSD_READ_BYTES = 0,
468 LPROC_OSD_WRITE_BYTES = 1,
469 LPROC_OSD_GET_PAGE = 2,
470 LPROC_OSD_NO_PAGE = 3,
471 LPROC_OSD_CACHE_ACCESS = 4,
472 LPROC_OSD_CACHE_HIT = 5,
473 LPROC_OSD_CACHE_MISS = 6,
474 LPROC_OSD_COPY_IO = 7,
475 LPROC_OSD_ZEROCOPY_IO = 8,
476 LPROC_OSD_TAIL_IO = 9,
480 extern struct kmem_cache *osd_zapit_cachep;
482 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
484 int osd_procfs_init(struct osd_device *osd, const char *name);
485 int osd_procfs_fini(struct osd_device *osd);
488 extern char *osd_obj_tag;
489 void osd_object_sa_dirty_rele(struct osd_thandle *oh);
490 int __osd_obj2dnode(const struct lu_env *env, objset_t *os,
491 uint64_t oid, dnode_t **dnp);
492 struct lu_object *osd_object_alloc(const struct lu_env *env,
493 const struct lu_object_header *hdr,
494 struct lu_device *d);
495 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
496 void *buf, uint32_t buflen, struct osd_thandle *oh);
497 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
498 dnode_t **zap_dnp, dmu_tx_t *tx, struct lu_attr *la,
500 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
501 dnode_t **dnp, dmu_tx_t *tx, struct lu_attr *la);
502 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
503 sa_handle_t *sa_hdl, dmu_tx_t *tx,
504 struct lu_attr *la, uint64_t parent);
507 int osd_oi_init(const struct lu_env *env, struct osd_device *o);
508 void osd_oi_fini(const struct lu_env *env, struct osd_device *o);
509 int osd_fid_lookup(const struct lu_env *env,
510 struct osd_device *, const struct lu_fid *, uint64_t *);
511 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
512 const struct lu_fid *fid, char *buf, int bufsize);
513 int osd_options_init(void);
514 int osd_ost_seq_exists(const struct lu_env *env, struct osd_device *osd,
516 int osd_idc_find_and_init(const struct lu_env *env, struct osd_device *osd,
517 struct osd_object *obj);
518 struct osd_idmap_cache *osd_idc_find_or_init(const struct lu_env *env,
519 struct osd_device *osd,
520 const struct lu_fid *fid);
521 struct osd_idmap_cache *osd_idc_find(const struct lu_env *env,
522 struct osd_device *osd,
523 const struct lu_fid *fid);
526 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
527 const struct dt_index_features *feat);
528 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
529 u64 seq, struct lu_seq_range *range);
530 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
531 uint64_t id, uint64_t dirhash);
532 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
533 uint64_t id, uint64_t dirhash);
534 void osd_zap_cursor_fini(zap_cursor_t *zc);
535 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc);
536 int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
537 const struct lu_fid *fid);
540 int __osd_xattr_load(struct osd_device *osd, uint64_t dnode,
541 nvlist_t **sa_xattr);
542 int __osd_xattr_get_large(const struct lu_env *env, struct osd_device *osd,
543 uint64_t xattr, struct lu_buf *buf,
544 const char *name, int *sizep);
545 int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
546 struct lu_buf *buf, const char *name);
547 int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
548 const struct lu_buf *buf, const char *name,
549 int fl, struct thandle *handle);
550 int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
551 const struct lu_buf *buf, const char *name, int fl,
552 struct thandle *handle);
553 int osd_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
554 const char *name, struct thandle *handle);
555 int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
556 const char *name, struct thandle *handle);
557 void osd_declare_xattrs_destroy(const struct lu_env *env,
558 struct osd_object *obj,
559 struct osd_thandle *oh);
560 int osd_xattrs_destroy(const struct lu_env *env,
561 struct osd_object *obj, struct osd_thandle *oh);
562 int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
563 const struct lu_buf *lb);
564 void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
565 int vallen, const char *name, struct osd_thandle *oh);
566 int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
567 const struct lu_buf *buf, const char *name, int fl,
568 struct osd_thandle *oh);;
569 int __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
570 const struct lu_buf *buf, const char *name, int fl,
571 struct osd_thandle *oh);
572 int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj,
573 struct osd_thandle *oh);
575 osd_xattr_set_internal(const struct lu_env *env, struct osd_object *obj,
576 const struct lu_buf *buf, const char *name, int fl,
577 struct osd_thandle *oh)
581 if (unlikely(!dt_object_exists(&obj->oo_dt) || obj->oo_destroyed))
585 if (osd_obj2dev(obj)->od_xattr_in_sa) {
586 rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
588 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
590 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
596 static inline uint64_t attrs_fs2zfs(const uint32_t flags)
598 return (flags & LUSTRE_APPEND_FL ? ZFS_APPENDONLY : 0) |
599 (flags & LUSTRE_NODUMP_FL ? ZFS_NODUMP : 0) |
600 (flags & LUSTRE_IMMUTABLE_FL ? ZFS_IMMUTABLE : 0);
603 static inline uint32_t attrs_zfs2fs(const uint64_t flags)
605 return (flags & ZFS_APPENDONLY ? LUSTRE_APPEND_FL : 0) |
606 (flags & ZFS_NODUMP ? LUSTRE_NODUMP_FL : 0) |
607 (flags & ZFS_IMMUTABLE ? LUSTRE_IMMUTABLE_FL : 0);
612 #ifndef HAVE_DSL_POOL_CONFIG
613 static inline void dsl_pool_config_enter(dsl_pool_t *dp, char *name)
617 static inline void dsl_pool_config_exit(dsl_pool_t *dp, char *name)
622 #ifdef HAVE_SPA_MAXBLOCKSIZE
623 #define osd_spa_maxblocksize(spa) spa_maxblocksize(spa)
624 #define osd_spa_maxblockshift(spa) fls64(spa_maxblocksize(spa) - 1)
626 #define osd_spa_maxblocksize(spa) SPA_MAXBLOCKSIZE
627 #define osd_spa_maxblockshift(spa) SPA_MAXBLOCKSHIFT
628 #define SPA_OLD_MAXBLOCKSIZE SPA_MAXBLOCKSIZE
631 #ifdef HAVE_SA_SPILL_ALLOC
633 osd_zio_buf_alloc(size_t size)
635 return sa_spill_alloc(KM_SLEEP);
639 osd_zio_buf_free(void *buf, size_t size)
644 #define osd_zio_buf_alloc(size) zio_buf_alloc(size)
645 #define osd_zio_buf_free(buf, size) zio_buf_free(buf, size)
648 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
649 static inline uint64_t
650 osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
651 int dnodesize, dmu_tx_t *tx)
654 dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
656 return dmu_object_alloc_dnsize(os, objtype, blocksize, DMU_OT_SA,
657 DN_BONUS_SIZE(dnodesize), dnodesize, tx);
660 static inline uint64_t
661 osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
662 dmu_object_type_t ot, int leaf_blockshift,
663 int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
666 dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
668 return zap_create_flags_dnsize(os, normflags, flags, ot,
669 leaf_blockshift, indirect_blockshift,
670 DMU_OT_SA, DN_BONUS_SIZE(dnodesize),
674 static inline uint64_t
675 osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
676 int dnodesize, dmu_tx_t *tx)
678 return dmu_object_alloc(os, objtype, blocksize, DMU_OT_SA,
679 DN_MAX_BONUSLEN, tx);
682 static inline uint64_t
683 osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
684 dmu_object_type_t ot, int leaf_blockshift,
685 int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
687 return zap_create_flags(os, normflags, flags, ot, leaf_blockshift,
688 indirect_blockshift, DMU_OT_SA,
689 DN_MAX_BONUSLEN, tx);
691 #endif /* HAVE_DMU_OBJECT_ALLOC_DNSIZE */
693 #ifdef HAVE_DMU_PREFETCH_6ARG
694 #define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
695 dmu_prefetch((os), (obj), (lvl), (off), (len), (pri))
697 #define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
698 dmu_prefetch((os), (obj), (lvl), (off))
701 #ifndef DN_MAX_BONUSLEN
702 #define DN_MAX_BONUSLEN DN_OLD_MAX_BONUSLEN
705 static inline void osd_dnode_rele(dnode_t *dn)
709 LASSERT(dn->dn_bonus);
713 dmu_buf_rele(&db->db, osd_obj_tag);
715 #endif /* _OSD_INTERNAL_H */