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 (DNODE_BLOCK_SHIFT - 12) /* est 4KB/dnode */
89 #define OSD_DNODE_EST_COUNT 1024
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 */
163 /* max.number of regular attrubites the callers may ask for */
164 #define OSD_MAX_IN_BULK 13
166 struct osd_thread_info {
167 const struct lu_env *oti_env;
169 struct lu_fid oti_fid;
171 * XXX temporary: for ->i_op calls.
173 struct timespec oti_time;
175 struct ost_id oti_ostid;
181 char oti_key[MAXNAMELEN + 1];
182 __u64 oti_key64[(MAXNAMELEN + 1)/sizeof(__u64)];
183 sa_bulk_attr_t oti_attr_bulk[OSD_MAX_IN_BULK];
185 struct lustre_mdt_attrs oti_mdt_attrs;
187 struct lu_attr oti_la;
188 struct osa_attr oti_osa;
189 zap_attribute_t oti_za;
190 dmu_object_info_t oti_doi;
191 struct luz_direntry oti_zde;
193 struct lquota_id_info oti_qi;
194 struct lu_seq_range oti_seq_range;
197 extern struct lu_context_key osd_key;
199 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
201 return lu_context_key_get(&env->le_ctx, &osd_key);
205 struct thandle ot_super;
206 struct list_head ot_dcb_list;
207 struct list_head ot_stop_dcb_list;
208 struct list_head ot_unlinked_list;
209 struct list_head ot_sa_list;
210 struct semaphore ot_sa_lock;
212 struct lquota_trans ot_quota_trans;
213 __u32 ot_write_commit:1,
217 #define OSD_OI_NAME_SIZE 16
220 * Object Index (OI) instance.
223 char oi_name[OSD_OI_NAME_SIZE]; /* unused */
229 uint64_t *os_compat_dirs;
230 int os_subdir_count; /* subdir count for each seq */
231 u64 os_seq; /* seq number */
232 struct list_head os_seq_list; /* list to seq_list */
235 struct osd_seq_list {
236 rwlock_t osl_seq_list_lock; /* lock for seq_list */
237 struct list_head osl_seq_list; /* list head for seq */
238 struct semaphore osl_seq_init_sem;
241 #define OSD_OST_MAP_SIZE 32
248 struct dt_device od_dt_dev;
249 /* information about underlying file system */
250 struct objset *od_os;
251 uint64_t od_rootid; /* id of root znode */
252 uint64_t od_unlinkedid; /* id of unlinked zapobj */
253 /* SA attr mapping->id,
254 * name is the same as in ZFS to use defines SA_ZPL_...*/
255 sa_attr_type_t *z_attr_table;
257 struct proc_dir_entry *od_proc_entry;
258 struct lprocfs_stats *od_stats;
260 uint64_t od_max_blksz;
263 struct osd_oi **od_oi_table;
264 unsigned int od_oi_count;
265 struct osd_seq_list od_seq_list;
267 unsigned int od_dev_set_rdonly:1, /**< osd_ro() called */
268 od_prop_rdonly:1, /**< ZFS property readonly */
270 od_quota_iused_est:1,
278 struct lu_site od_site;
280 /* object IDs of the inode accounting indexes */
281 uint64_t od_iusr_oid;
282 uint64_t od_igrp_oid;
284 /* quota slave instance */
285 struct qsd_instance *od_quota_slave;
287 struct brw_stats od_brw_stats;
288 atomic_t od_r_in_flight;
289 atomic_t od_w_in_flight;
291 /* used to debug zerocopy logic: the fields track all
292 * allocated, loaned and referenced buffers in use.
293 * to be removed once the change is tested well. */
294 atomic_t od_zerocopy_alloc;
295 atomic_t od_zerocopy_loan;
296 atomic_t od_zerocopy_pin;
298 arc_prune_t *arc_prune_cb;
300 /* osd seq instance */
301 struct lu_client_seq *od_cl_seq;
304 enum osd_destroy_type {
305 OSD_DESTROY_NONE = 0,
306 OSD_DESTROY_SYNC = 1,
307 OSD_DESTROY_ASYNC = 2,
311 struct dt_object oo_dt;
313 * Inode for file system object represented by this osd_object. This
314 * inode is pinned for the whole duration of lu_object life.
316 * Not modified concurrently (either setup early during object
317 * creation, or assigned by osd_object_create() under write lock).
320 sa_handle_t *oo_sa_hdl;
321 nvlist_t *oo_sa_xattr;
322 struct list_head oo_sa_linkage;
324 /* used to implement osd_object_*_{lock|unlock} */
325 struct rw_semaphore oo_sem;
327 /* to serialize some updates: destroy vs. others,
328 * xattr_set, object block size change etc */
329 struct rw_semaphore oo_guard;
331 /* protected by oo_guard */
332 struct list_head oo_unlinked_linkage;
334 /* cached attributes */
335 rwlock_t oo_attr_lock;
336 struct lu_attr oo_attr;
338 /* external dnode holding large EAs, protected by oo_guard */
340 enum osd_destroy_type oo_destroy;
342 __u32 oo_destroyed:1;
344 /* the i_flags in LMA */
347 int oo_ea_in_bonus; /* EA bytes we expect */
349 /* record size for index file */
350 unsigned char oo_keysize;
351 unsigned char oo_recsize;
352 unsigned char oo_recusize; /* unit size */
359 int osd_statfs(const struct lu_env *, struct dt_device *, struct obd_statfs *);
360 extern const struct dt_index_operations osd_acct_index_ops;
361 uint64_t osd_quota_fid2dmu(const struct lu_fid *fid);
362 extern struct lu_device_operations osd_lu_ops;
363 extern struct dt_index_operations osd_dir_ops;
364 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
365 qid_t uid, qid_t gid, long long space,
366 struct osd_thandle *oh, bool is_blk, int *flags,
368 uint64_t osd_objs_count_estimate(uint64_t refdbytes, uint64_t usedobjs,
369 uint64_t nrblocks, uint64_t est_maxblockshift);
374 static inline int lu_device_is_osd(const struct lu_device *d)
376 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
379 static inline struct osd_object *osd_obj(const struct lu_object *o)
381 LASSERT(lu_device_is_osd(o->lo_dev));
382 return container_of0(o, struct osd_object, oo_dt.do_lu);
385 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
387 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
388 return container_of0(d, struct osd_device, od_dt_dev);
391 static inline struct osd_device *osd_dev(const struct lu_device *d)
393 LASSERT(lu_device_is_osd(d));
394 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
397 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
399 return osd_obj(&d->do_lu);
402 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
404 return osd_dev(o->oo_dt.do_lu.lo_dev);
407 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
409 return &osd->od_dt_dev.dd_lu_dev;
412 static inline struct objset * osd_dtobj2objset(struct dt_object *o)
414 return osd_dev(o->do_lu.lo_dev)->od_os;
417 static inline int osd_invariant(const struct osd_object *obj)
422 static inline int osd_object_invariant(const struct lu_object *l)
424 return osd_invariant(osd_obj(l));
427 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
429 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
432 static inline char *osd_name(struct osd_device *osd)
434 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
437 #ifdef CONFIG_PROC_FS
439 LPROC_OSD_READ_BYTES = 0,
440 LPROC_OSD_WRITE_BYTES = 1,
441 LPROC_OSD_GET_PAGE = 2,
442 LPROC_OSD_NO_PAGE = 3,
443 LPROC_OSD_CACHE_ACCESS = 4,
444 LPROC_OSD_CACHE_HIT = 5,
445 LPROC_OSD_CACHE_MISS = 6,
446 LPROC_OSD_COPY_IO = 7,
447 LPROC_OSD_ZEROCOPY_IO = 8,
448 LPROC_OSD_TAIL_IO = 9,
452 extern struct kmem_cache *osd_zapit_cachep;
454 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
456 int osd_procfs_init(struct osd_device *osd, const char *name);
457 int osd_procfs_fini(struct osd_device *osd);
460 extern char *osd_obj_tag;
461 void osd_object_sa_dirty_rele(struct osd_thandle *oh);
462 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
463 uint64_t oid, dmu_buf_t **dbp);
464 struct lu_object *osd_object_alloc(const struct lu_env *env,
465 const struct lu_object_header *hdr,
466 struct lu_device *d);
467 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
468 void *buf, uint32_t buflen, struct osd_thandle *oh);
469 int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
470 dmu_buf_t **zap_dbp, dmu_tx_t *tx, struct lu_attr *la,
472 int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
473 dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la);
474 int __osd_attr_init(const struct lu_env *env, struct osd_device *osd,
475 sa_handle_t *sa_hdl, dmu_tx_t *tx,
476 struct lu_attr *la, uint64_t parent);
479 int osd_oi_init(const struct lu_env *env, struct osd_device *o);
480 void osd_oi_fini(const struct lu_env *env, struct osd_device *o);
481 int osd_fid_lookup(const struct lu_env *env,
482 struct osd_device *, const struct lu_fid *, uint64_t *);
483 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
484 const struct lu_fid *fid, char *buf, int bufsize);
485 int osd_options_init(void);
486 int osd_ost_seq_exists(const struct lu_env *env, struct osd_device *osd,
489 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
490 const struct dt_index_features *feat);
491 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
492 u64 seq, struct lu_seq_range *range);
493 void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
494 uint64_t id, uint64_t dirhash);
495 int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
496 uint64_t id, uint64_t dirhash);
497 void osd_zap_cursor_fini(zap_cursor_t *zc);
498 uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc);
501 int __osd_xattr_load(struct osd_device *osd, uint64_t dnode,
502 nvlist_t **sa_xattr);
503 int __osd_xattr_get_large(const struct lu_env *env, struct osd_device *osd,
504 uint64_t xattr, struct lu_buf *buf,
505 const char *name, int *sizep);
506 int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
507 struct lu_buf *buf, const char *name);
508 int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
509 const struct lu_buf *buf, const char *name,
510 int fl, struct thandle *handle);
511 int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
512 const struct lu_buf *buf, const char *name, int fl,
513 struct thandle *handle);
514 int osd_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
515 const char *name, struct thandle *handle);
516 int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
517 const char *name, struct thandle *handle);
518 void osd_declare_xattrs_destroy(const struct lu_env *env,
519 struct osd_object *obj,
520 struct osd_thandle *oh);
521 int osd_xattrs_destroy(const struct lu_env *env,
522 struct osd_object *obj, struct osd_thandle *oh);
523 int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
524 const struct lu_buf *lb);
525 void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
526 int vallen, const char *name, struct osd_thandle *oh);
527 int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
528 const struct lu_buf *buf, const char *name, int fl,
529 struct osd_thandle *oh);;
530 int __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
531 const struct lu_buf *buf, const char *name, int fl,
532 struct osd_thandle *oh);
533 int __osd_sa_xattr_update(const struct lu_env *env, struct osd_object *obj,
534 struct osd_thandle *oh);
536 osd_xattr_set_internal(const struct lu_env *env, struct osd_object *obj,
537 const struct lu_buf *buf, const char *name, int fl,
538 struct osd_thandle *oh)
542 if (unlikely(!dt_object_exists(&obj->oo_dt) || obj->oo_destroyed))
546 if (osd_obj2dev(obj)->od_xattr_in_sa) {
547 rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
549 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
551 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
557 static inline uint64_t attrs_fs2zfs(const uint32_t flags)
559 return (flags & LUSTRE_APPEND_FL ? ZFS_APPENDONLY : 0) |
560 (flags & LUSTRE_NODUMP_FL ? ZFS_NODUMP : 0) |
561 (flags & LUSTRE_IMMUTABLE_FL ? ZFS_IMMUTABLE : 0);
564 static inline uint32_t attrs_zfs2fs(const uint64_t flags)
566 return (flags & ZFS_APPENDONLY ? LUSTRE_APPEND_FL : 0) |
567 (flags & ZFS_NODUMP ? LUSTRE_NODUMP_FL : 0) |
568 (flags & ZFS_IMMUTABLE ? LUSTRE_IMMUTABLE_FL : 0);
573 #ifndef HAVE_DSL_POOL_CONFIG
574 static inline void dsl_pool_config_enter(dsl_pool_t *dp, char *name)
578 static inline void dsl_pool_config_exit(dsl_pool_t *dp, char *name)
583 #ifdef HAVE_SPA_MAXBLOCKSIZE
584 #define osd_spa_maxblocksize(spa) spa_maxblocksize(spa)
585 #define osd_spa_maxblockshift(spa) fls64(spa_maxblocksize(spa) - 1)
587 #define osd_spa_maxblocksize(spa) SPA_MAXBLOCKSIZE
588 #define osd_spa_maxblockshift(spa) SPA_MAXBLOCKSHIFT
589 #define SPA_OLD_MAXBLOCKSIZE SPA_MAXBLOCKSIZE
592 #ifdef HAVE_SA_SPILL_ALLOC
594 osd_zio_buf_alloc(size_t size)
596 return sa_spill_alloc(KM_SLEEP);
600 osd_zio_buf_free(void *buf, size_t size)
605 #define osd_zio_buf_alloc(size) zio_buf_alloc(size)
606 #define osd_zio_buf_free(buf, size) zio_buf_free(buf, size)
609 #ifdef HAVE_DMU_OBJECT_ALLOC_DNSIZE
610 static inline uint64_t
611 osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
612 int dnodesize, dmu_tx_t *tx)
615 dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
617 return dmu_object_alloc_dnsize(os, objtype, blocksize, DMU_OT_SA,
618 DN_BONUS_SIZE(dnodesize), dnodesize, tx);
621 static inline uint64_t
622 osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
623 dmu_object_type_t ot, int leaf_blockshift,
624 int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
627 dnodesize = MAX(dmu_objset_dnodesize(os), DNODE_MIN_SIZE);
629 return zap_create_flags_dnsize(os, normflags, flags, ot,
630 leaf_blockshift, indirect_blockshift,
631 DMU_OT_SA, DN_BONUS_SIZE(dnodesize),
635 static inline uint64_t
636 osd_dmu_object_alloc(objset_t *os, dmu_object_type_t objtype, int blocksize,
637 int dnodesize, dmu_tx_t *tx)
639 return dmu_object_alloc(os, objtype, blocksize, DMU_OT_SA,
640 DN_MAX_BONUSLEN, tx);
643 static inline uint64_t
644 osd_zap_create_flags(objset_t *os, int normflags, zap_flags_t flags,
645 dmu_object_type_t ot, int leaf_blockshift,
646 int indirect_blockshift, int dnodesize, dmu_tx_t *tx)
648 return zap_create_flags(os, normflags, flags, ot, leaf_blockshift,
649 indirect_blockshift, DMU_OT_SA,
650 DN_MAX_BONUSLEN, tx);
652 #endif /* HAVE_DMU_OBJECT_ALLOC_DNSIZE */
654 #ifdef HAVE_DMU_PREFETCH_6ARG
655 #define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
656 dmu_prefetch((os), (obj), (lvl), (off), (len), (pri))
658 #define osd_dmu_prefetch(os, obj, lvl, off, len, pri) \
659 dmu_prefetch((os), (obj), (lvl), (off))
662 #ifndef DN_MAX_BONUSLEN
663 #define DN_MAX_BONUSLEN DN_OLD_MAX_BONUSLEN
666 #endif /* _OSD_INTERNAL_H */