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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
31 * Copyright (c) 2012, Intel Corporation.
32 * Use is subject to license terms.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd-zfs/osd_internal.h
39 * Shared definitions and declarations for zfs/dmu osd
41 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
42 * Author: Mike Pershin <tappro@whamcloud.com>
43 * Author: Johann Lombardi <johann@whamcloud.com>
46 #ifndef _OSD_INTERNAL_H
47 #define _OSD_INTERNAL_H
49 #include <dt_object.h>
50 #include <lustre_quota.h>
53 #include <sys/nvpair.h>
57 #define LUSTRE_ROOT_FID_SEQ 0
58 #define DMU_OSD_SVNAME "svname"
59 #define DMU_OSD_OI_NAME_BASE "oi"
61 #define OSD_GFP_IO (GFP_NOFS | __GFP_HIGHMEM)
64 * Iterator's in-memory data structure for quota file.
67 struct osd_object *oiq_obj;
68 /* DMU accounting object id */
72 /** identifier for current quota record */
74 unsigned oiq_reset:1; /* 1 -- no need to advance */
78 * Iterator's in-memory data structure for ZAPs
82 struct osd_object *ozi_obj;
83 struct lustre_capa *ozi_capa;
84 unsigned ozi_reset:1; /* 1 -- no need to advance */
86 char ozi_name[NAME_MAX + 1]; /* file name for dir */
87 __u64 ozi_key; /* binary key for index files */
90 #define DT_IT2DT(it) (&((struct osd_zap_it *)it)->ozi_obj->oo_dt)
93 * regular ZFS direntry
96 uint64_t zde_dnode:48,
99 } __attribute__((packed));
102 * lustre direntry adds a fid to regular ZFS direntry
104 struct luz_direntry {
105 struct zpl_direntry lzd_reg;
106 struct lu_fid lzd_fid;
107 } __attribute__((packed));
110 /* cached SA attributes */
124 struct osd_thread_info {
125 const struct lu_env *oti_env;
127 struct lu_fid oti_fid;
129 * XXX temporary: for ->i_op calls.
131 struct timespec oti_time;
133 * XXX temporary: for capa operations.
135 struct lustre_capa_key oti_capa_key;
136 struct lustre_capa oti_capa;
138 struct ost_id oti_ostid;
142 /** osd iterator context used for iterator session */
144 struct osd_zap_it oti_it_zap;
145 struct osd_it_quota oti_it_quota;
149 char oti_key[MAXNAMELEN + 1];
150 struct lustre_mdt_attrs oti_mdt_attrs;
152 struct lu_attr oti_la;
153 struct osa_attr oti_osa;
154 zap_attribute_t oti_za;
155 dmu_object_info_t oti_doi;
156 struct luz_direntry oti_zde;
158 struct lquota_id_info oti_qi;
159 struct lu_seq_range oti_seq_range;
162 extern struct lu_context_key osd_key;
164 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
166 return lu_context_key_get(&env->le_ctx, &osd_key);
170 struct thandle ot_super;
171 cfs_list_t ot_dcb_list;
172 cfs_list_t ot_sa_list;
173 struct semaphore ot_sa_lock;
175 struct lquota_trans ot_quota_trans;
176 __u32 ot_write_commit:1,
180 #define OSD_OI_NAME_SIZE 16
183 * Object Index (OI) instance.
186 char oi_name[OSD_OI_NAME_SIZE]; /* unused */
191 uint64_t *os_compat_dirs;
192 int os_subdir_count; /* subdir count for each seq */
193 obd_seq os_seq; /* seq number */
194 cfs_list_t os_seq_list; /* list to seq_list */
197 struct osd_seq_list {
198 rwlock_t osl_seq_list_lock; /* lock for seq_list */
199 cfs_list_t osl_seq_list; /* list head for seq */
200 struct semaphore osl_seq_init_sem;
203 #define OSD_OST_MAP_SIZE 32
210 struct dt_device od_dt_dev;
211 /* information about underlying file system */
212 udmu_objset_t od_objset;
217 unsigned int od_fl_capa:1;
218 unsigned long od_capa_timeout;
220 struct lustre_capa_key *od_capa_keys;
221 cfs_hlist_head_t *od_capa_hash;
223 cfs_proc_dir_entry_t *od_proc_entry;
224 struct lprocfs_stats *od_stats;
227 struct osd_oi **od_oi_table;
228 unsigned int od_oi_count;
229 uint64_t od_ost_compat_grp0;
230 struct osd_seq_list od_seq_list;
232 unsigned int od_rdonly:1,
234 od_quota_iused_est:1;
239 struct lu_site od_site;
241 /* object IDs of the inode accounting indexes */
242 uint64_t od_iusr_oid;
243 uint64_t od_igrp_oid;
245 /* quota slave instance */
246 struct qsd_instance *od_quota_slave;
248 /* used to debug zerocopy logic: the fields track all
249 * allocated, loaned and referenced buffers in use.
250 * to be removed once the change is tested well. */
251 cfs_atomic_t od_zerocopy_alloc;
252 cfs_atomic_t od_zerocopy_loan;
253 cfs_atomic_t od_zerocopy_pin;
255 arc_prune_t *arc_prune_cb;
259 struct dt_object oo_dt;
261 * Inode for file system object represented by this osd_object. This
262 * inode is pinned for the whole duration of lu_object life.
264 * Not modified concurrently (either setup early during object
265 * creation, or assigned by osd_object_create() under write lock).
268 sa_handle_t *oo_sa_hdl;
269 nvlist_t *oo_sa_xattr;
270 cfs_list_t oo_sa_linkage;
272 struct rw_semaphore oo_sem;
274 /* cached attributes */
275 rwlock_t oo_attr_lock;
276 struct lu_attr oo_attr;
278 /* protects extended attributes */
279 struct semaphore oo_guard;
282 /* record size for index file */
286 int osd_statfs(const struct lu_env *, struct dt_device *, struct obd_statfs *);
287 extern const struct dt_index_operations osd_acct_index_ops;
288 uint64_t osd_quota_fid2dmu(const struct lu_fid *fid);
289 extern struct lu_device_operations osd_lu_ops;
290 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
291 qid_t uid, qid_t gid, long long space,
292 struct osd_thandle *oh, bool is_blk, int *flags,
298 static inline int lu_device_is_osd(const struct lu_device *d)
300 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
303 static inline struct osd_object *osd_obj(const struct lu_object *o)
305 LASSERT(lu_device_is_osd(o->lo_dev));
306 return container_of0(o, struct osd_object, oo_dt.do_lu);
309 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
311 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
312 return container_of0(d, struct osd_device, od_dt_dev);
315 static inline struct osd_device *osd_dev(const struct lu_device *d)
317 LASSERT(lu_device_is_osd(d));
318 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
321 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
323 return osd_obj(&d->do_lu);
326 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
328 return osd_dev(o->oo_dt.do_lu.lo_dev);
331 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
333 return &osd->od_dt_dev.dd_lu_dev;
336 static inline struct objset * osd_dtobj2objset(struct dt_object *o)
338 return osd_dev(o->do_lu.lo_dev)->od_objset.os;
341 static inline int osd_invariant(const struct osd_object *obj)
346 static inline int osd_object_invariant(const struct lu_object *l)
348 return osd_invariant(osd_obj(l));
351 static inline struct seq_server_site *osd_seq_site(struct osd_device *osd)
353 return osd->od_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
356 static inline char *osd_name(struct osd_device *osd)
358 return osd->od_dt_dev.dd_lu_dev.ld_obd->obd_name;
363 LPROC_OSD_READ_BYTES = 0,
364 LPROC_OSD_WRITE_BYTES = 1,
365 LPROC_OSD_GET_PAGE = 2,
366 LPROC_OSD_NO_PAGE = 3,
367 LPROC_OSD_CACHE_ACCESS = 4,
368 LPROC_OSD_CACHE_HIT = 5,
369 LPROC_OSD_CACHE_MISS = 6,
370 LPROC_OSD_COPY_IO = 7,
371 LPROC_OSD_ZEROCOPY_IO = 8,
372 LPROC_OSD_TAIL_IO = 9,
377 extern struct lprocfs_vars lprocfs_osd_obd_vars[];
378 extern struct lprocfs_vars lprocfs_osd_module_vars[];
380 int osd_procfs_init(struct osd_device *osd, const char *name);
381 int osd_procfs_fini(struct osd_device *osd);
383 int udmu_zap_cursor_retrieve_key(const struct lu_env *env,
384 zap_cursor_t *zc, char *key, int max);
385 int udmu_zap_cursor_retrieve_value(const struct lu_env *env,
386 zap_cursor_t *zc, char *buf,
387 int buf_size, int *bytes_read);
390 void osd_object_sa_dirty_rele(struct osd_thandle *oh);
391 int __osd_obj2dbuf(const struct lu_env *env, objset_t *os,
392 uint64_t oid, dmu_buf_t **dbp, void *tag);
393 struct lu_object *osd_object_alloc(const struct lu_env *env,
394 const struct lu_object_header *hdr,
395 struct lu_device *d);
396 int osd_object_sa_update(struct osd_object *obj, sa_attr_type_t type,
397 void *buf, uint32_t buflen, struct osd_thandle *oh);
398 int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
399 dmu_buf_t **zap_dbp, dmu_tx_t *tx,
400 struct lu_attr *la, void *tag, zap_flags_t flags);
401 int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
402 dmu_buf_t **dbp, dmu_tx_t *tx,
403 struct lu_attr *la, void *tag);
404 int __osd_object_free(udmu_objset_t *uos, uint64_t oid, dmu_tx_t *tx);
407 int osd_oi_init(const struct lu_env *env, struct osd_device *o);
408 void osd_oi_fini(const struct lu_env *env, struct osd_device *o);
409 int osd_fid_lookup(const struct lu_env *env,
410 struct osd_device *, const struct lu_fid *, uint64_t *);
411 uint64_t osd_get_name_n_idx(const struct lu_env *env, struct osd_device *osd,
412 const struct lu_fid *fid, char *buf);
413 int osd_options_init(void);
416 int osd_index_try(const struct lu_env *env, struct dt_object *dt,
417 const struct dt_index_features *feat);
421 int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
422 struct lu_buf *buf, const char *name,
423 struct lustre_capa *capa);
424 int osd_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
425 const struct lu_buf *buf, const char *name,
426 int fl, struct thandle *handle);
427 int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
428 const struct lu_buf *buf, const char *name, int fl,
429 struct thandle *handle, struct lustre_capa *capa);
430 int osd_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
431 const char *name, struct thandle *handle);
432 int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
433 const char *name, struct thandle *handle,
434 struct lustre_capa *capa);
435 int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
436 struct lu_buf *lb, struct lustre_capa *capa);
437 void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
438 int vallen, const char *name, struct osd_thandle *oh);
439 int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
440 const struct lu_buf *buf, const char *name, int fl,
441 struct osd_thandle *oh);;
442 int __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
443 const struct lu_buf *buf, const char *name, int fl,
444 struct osd_thandle *oh);
446 osd_xattr_set_internal(const struct lu_env *env, struct osd_object *obj,
447 const struct lu_buf *buf, const char *name, int fl,
448 struct osd_thandle *oh, struct lustre_capa *capa)
452 if (osd_obj2dev(obj)->od_xattr_in_sa) {
453 rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
455 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
457 rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
464 #endif /* _OSD_INTERNAL_H */