1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/osd/osd_internal.h
40 * Shared definitions and declarations for osd module
42 * Author: Nikita Danilov <nikita@clusterfs.com>
45 #ifndef _OSD_INTERNAL_H
46 #define _OSD_INTERNAL_H
48 #if defined(__KERNEL__)
50 /* struct rw_semaphore */
51 #include <linux/rwsem.h>
53 #include <linux/dcache.h>
55 #include <linux/dirent.h>
57 #include <ldiskfs/ldiskfs.h>
58 #include <ldiskfs/ldiskfs_jbd2.h>
59 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
60 # define journal_callback ldiskfs_journal_cb_entry
61 # define osd_journal_callback_set(handle, func, jcb) \
62 ldiskfs_journal_callback_add(handle, func, jcb)
64 # define osd_journal_callback_set(handle, func, jcb) \
65 jbd2_journal_callback_set(handle, func, jcb)
71 /* class_register_type(), class_unregister_type(), class_get_type() */
72 #include <obd_class.h>
73 #include <lustre_disk.h>
74 #include <dt_object.h>
81 #define OSD_OII_NOGEN (0)
82 #define OSD_COUNTERS (0)
84 /** Enable thandle usage statistics */
85 #define OSD_THANDLE_STATS (0)
87 #ifdef HAVE_QUOTA_SUPPORT
91 cfs_kernel_cap_t oc_cap;
95 struct osd_directory {
96 struct iam_container od_container;
97 struct iam_descr od_descr;
101 struct dt_object oo_dt;
103 * Inode for file system object represented by this osd_object. This
104 * inode is pinned for the whole duration of lu_object life.
106 * Not modified concurrently (either setup early during object
107 * creation, or assigned by osd_object_create() under write lock).
109 struct inode *oo_inode;
111 * to protect index ops.
113 struct htree_lock_head *oo_hl_head;
114 cfs_rw_semaphore_t oo_ext_idx_sem;
115 cfs_rw_semaphore_t oo_sem;
116 struct osd_directory *oo_dir;
117 /** protects inode attributes. */
118 cfs_spinlock_t oo_guard;
120 * Following two members are used to indicate the presence of dot and
121 * dotdot in the given directory. This is required for interop mode
124 int oo_compat_dot_created;
125 int oo_compat_dotdot_created;
127 const struct lu_env *oo_owner;
128 #ifdef CONFIG_LOCKDEP
129 struct lockdep_map oo_dep_map;
133 #ifdef HAVE_LDISKFS_PDO
135 #define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
136 ll_ldiskfs_find_entry(dir, dentry, de, lock)
137 #define osd_ldiskfs_add_entry(handle, child, cinode, hlock) \
138 ldiskfs_add_entry(handle, child, cinode, hlock)
140 #else /* HAVE_LDISKFS_PDO */
146 struct htree_lock_head {
150 #define ldiskfs_htree_lock(lock, head, inode, op) do { LBUG(); } while (0)
151 #define ldiskfs_htree_unlock(lock) do { LBUG(); } while (0)
153 static inline struct htree_lock_head *ldiskfs_htree_lock_head_alloc(int dep)
159 #define ldiskfs_htree_lock_head_free(lh) do { LBUG(); } while (0)
161 #define LDISKFS_DUMMY_HTREE_LOCK 0xbabecafe
163 static inline struct htree_lock *ldiskfs_htree_lock_alloc(void)
165 return (struct htree_lock *)LDISKFS_DUMMY_HTREE_LOCK;
168 static inline void ldiskfs_htree_lock_free(struct htree_lock *lk)
170 LASSERT((unsigned long)lk == LDISKFS_DUMMY_HTREE_LOCK);
173 #define HTREE_HBITS_DEF 0
175 #define osd_ldiskfs_find_entry(dir, dentry, de, lock) \
176 ll_ldiskfs_find_entry(dir, dentry, de)
177 #define osd_ldiskfs_add_entry(handle, child, cinode, lock) \
178 ldiskfs_add_entry(handle, child, cinode)
180 #endif /* HAVE_LDISKFS_PDO */
182 extern const int osd_dto_credits_noquota[];
189 struct dt_device od_dt_dev;
190 /* information about underlying file system */
191 struct lustre_mount_info *od_mount;
193 * XXX temporary stuff for object index: directory where every object
194 * is named by its fid.
196 struct dt_object *od_obj_area;
198 struct osd_oi *od_oi_table;
199 /* total number of OI containers */
204 unsigned int od_fl_capa:1;
205 unsigned long od_capa_timeout;
207 struct lustre_capa_key *od_capa_keys;
208 cfs_hlist_head_t *od_capa_hash;
210 cfs_proc_dir_entry_t *od_proc_entry;
211 struct lprocfs_stats *od_stats;
213 * statfs optimization: we cache a bit.
215 cfs_time_t od_osfs_age;
216 cfs_kstatfs_t od_kstatfs;
217 cfs_spinlock_t od_osfs_lock;
220 * The following flag indicates, if it is interop mode or not.
221 * It will be initialized, using mount param.
226 #define OSD_TRACK_DECLARES
227 #ifdef OSD_TRACK_DECLARES
228 #define OSD_DECLARE_OP(oh, op) { \
229 LASSERT(oh->ot_handle == NULL); \
230 ((oh)->ot_declare_ ##op)++; }
231 #define OSD_EXEC_OP(handle, op) { \
232 struct osd_thandle *oh; \
233 oh = container_of0(handle, struct osd_thandle, ot_super);\
234 LASSERT((oh)->ot_declare_ ##op > 0); \
235 ((oh)->ot_declare_ ##op)--; }
237 #define OSD_DECLARE_OP(oh, op)
238 #define OSD_EXEC_OP(oh, op)
241 /* There are at most 10 uid/gids are affected in a transaction, and
242 * that's rename case:
243 * - 2 for source parent uid & gid;
244 * - 2 for source child uid & gid ('..' entry update when the child
246 * - 2 for target parent uid & gid;
247 * - 2 for target child uid & gid (if the target child exists);
248 * - 2 for root uid & gid (last_rcvd, llog, etc);
250 * The 0 to (OSD_MAX_UGID_CNT - 1) bits of ot_id_type is for indicating
251 * the id type of each id in the ot_id_array.
253 #define OSD_MAX_UGID_CNT 10
256 struct thandle ot_super;
258 struct journal_callback ot_jcb;
259 cfs_list_t ot_dcb_list;
260 /* Link to the device, for debugging. */
261 struct lu_ref_link *ot_dev_link;
262 unsigned short ot_credits;
263 unsigned short ot_id_cnt;
264 unsigned short ot_id_type;
265 uid_t ot_id_array[OSD_MAX_UGID_CNT];
267 #ifdef OSD_TRACK_DECLARES
268 unsigned char ot_declare_attr_set;
269 unsigned char ot_declare_punch;
270 unsigned char ot_declare_xattr_set;
271 unsigned char ot_declare_create;
272 unsigned char ot_declare_destroy;
273 unsigned char ot_declare_ref_add;
274 unsigned char ot_declare_ref_del;
275 unsigned char ot_declare_write;
276 unsigned char ot_declare_insert;
277 unsigned char ot_declare_delete;
280 #if OSD_THANDLE_STATS
281 /** time when this handle was allocated */
282 cfs_time_t oth_alloced;
284 /** time when this thanle was started */
285 cfs_time_t oth_started;
290 * Basic transaction credit op
300 DTO_LOG_REC, /**< XXX temporary: dt layer knows nothing about llog. */
314 #if OSD_THANDLE_STATS
315 LPROC_OSD_THANDLE_STARTING,
316 LPROC_OSD_THANDLE_OPEN,
317 LPROC_OSD_THANDLE_CLOSING,
324 * Storage representation for fids.
326 * Variable size, first byte contains the length of the whole record.
328 struct osd_fid_pack {
329 unsigned char fp_len;
330 char fp_area[sizeof(struct lu_fid)];
333 struct osd_it_ea_dirent {
334 struct lu_fid oied_fid;
337 unsigned short oied_namelen;
338 unsigned int oied_type;
340 } __attribute__((packed));
343 * as osd_it_ea_dirent (in memory dirent struct for osd) is greater
344 * than lu_dirent struct. osd readdir reads less number of dirent than
345 * required for mdd dir page. so buffer size need to be increased so that
346 * there would be one ext3 readdir for every mdd readdir page.
349 #define OSD_IT_EA_BUFSIZE (CFS_PAGE_SIZE + CFS_PAGE_SIZE/4)
352 * This is iterator's in-memory data structure in interoperability
353 * mode (i.e. iterator over ldiskfs style directory)
356 struct osd_object *oie_obj;
357 /** used in ldiskfs iterator, to stored file pointer */
358 struct file oie_file;
359 /** how many entries have been read-cached from storage */
361 /** current entry is being iterated by caller */
363 /** current processing entry */
364 struct osd_it_ea_dirent *oie_dirent;
365 /** buffer to hold entries, size == OSD_IT_EA_BUFSIZE */
370 * Iterator's in-memory data structure for IAM mode.
373 struct osd_object *oi_obj;
374 struct iam_path_descr *oi_ipd;
375 struct iam_iterator oi_it;
378 struct osd_thread_info {
379 const struct lu_env *oti_env;
381 * used for index operations.
383 struct dentry oti_obj_dentry;
384 struct dentry oti_child_dentry;
386 /** dentry for Iterator context. */
387 struct dentry oti_it_dentry;
388 struct htree_lock *oti_hlock;
390 struct lu_fid oti_fid;
391 struct osd_inode_id oti_id;
393 * XXX temporary: for ->i_op calls.
395 struct timespec oti_time;
397 * XXX temporary: fake struct file for osd_object_sync
399 struct file oti_file;
401 * XXX temporary: for capa operations.
403 struct lustre_capa_key oti_capa_key;
404 struct lustre_capa oti_capa;
406 /** osd_device reference, initialized in osd_trans_start() and
407 used in osd_trans_stop() */
408 struct osd_device *oti_dev;
411 * following ipd and it structures are used for osd_index_iam_lookup()
412 * these are defined separately as we might do index operation
413 * in open iterator session.
416 /** osd iterator context used for iterator session */
419 struct osd_it_iam oti_it;
420 /** ldiskfs iterator data structure, see osd_it_ea_{init, fini} */
421 struct osd_it_ea oti_it_ea;
424 /** pre-allocated buffer used by oti_it_ea, size OSD_IT_EA_BUFSIZE */
427 /** IAM iterator for index operation. */
428 struct iam_iterator oti_idx_it;
430 /** union to guarantee that ->oti_ipd[] has proper alignment. */
432 char oti_it_ipd[DX_IPD_MAX_SIZE];
433 long long oti_alignment_lieutenant;
437 char oti_idx_ipd[DX_IPD_MAX_SIZE];
438 long long oti_alignment_lieutenant_colonel;
445 /** used in osd_fid_set() to put xattr */
446 struct lu_buf oti_buf;
447 /** used in osd_ea_fid_set() to set fid into common ea */
448 struct lustre_mdt_attrs oti_mdt_attrs;
449 #ifdef HAVE_QUOTA_SUPPORT
450 struct osd_ctxt oti_ctxt;
452 struct lu_env oti_obj_delete_tx_env;
453 #define OSD_FID_REC_SZ 32
454 char oti_ldp[OSD_FID_REC_SZ];
455 char oti_ldp2[OSD_FID_REC_SZ];
458 extern int ldiskfs_pdo;
462 void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars);
463 int osd_procfs_init(struct osd_device *osd, const char *name);
464 int osd_procfs_fini(struct osd_device *osd);
465 void osd_lprocfs_time_start(const struct lu_env *env);
466 void osd_lprocfs_time_end(const struct lu_env *env,
467 struct osd_device *osd, int op);
469 int osd_statfs(const struct lu_env *env, struct dt_device *dev,
471 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
472 struct lustre_capa *capa, __u64 opc);
475 * Invariants, assertions.
479 * XXX: do not enable this, until invariant checking code is made thread safe
480 * in the face of pdirops locking.
482 #define OSD_INVARIANT_CHECKS (0)
484 #if OSD_INVARIANT_CHECKS
485 static inline int osd_invariant(const struct osd_object *obj)
489 ergo(obj->oo_inode != NULL,
490 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
491 atomic_read(&obj->oo_inode->i_count) > 0) &&
492 ergo(obj->oo_dir != NULL &&
493 obj->oo_dir->od_conationer.ic_object != NULL,
494 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
497 #define osd_invariant(obj) (1)
500 /* The on-disk extN format reserves inodes 0-11 for internal filesystem
501 * use, and these inodes will be invisible on client side, so the valid
502 * sequence for IGIF fid is 12-0xffffffff. But root inode (2#) will be seen
503 * on server side (osd), and it should be valid too here.
505 #define OSD_ROOT_SEQ 2
506 static inline int osd_fid_is_root(const struct lu_fid *fid)
508 return fid_seq(fid) == OSD_ROOT_SEQ;
511 static inline int osd_fid_is_igif(const struct lu_fid *fid)
513 return fid_is_igif(fid) || osd_fid_is_root(fid);
516 static inline struct osd_oi *
517 osd_fid2oi(struct osd_device *osd, const struct lu_fid *fid)
519 if (!fid_is_norm(fid))
522 LASSERT(osd->od_oi_table != NULL && osd->od_oi_count >= 1);
523 /* It can work even od_oi_count equals to 1 although it's unexpected,
524 * the only reason we set it to 1 is for performance measurement */
525 return &osd->od_oi_table[fid->f_seq & (osd->od_oi_count - 1)];
531 extern const struct lu_device_operations osd_lu_ops;
533 static inline int lu_device_is_osd(const struct lu_device *d)
535 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
538 static inline struct osd_device *osd_dt_dev(const struct dt_device *d)
540 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
541 return container_of0(d, struct osd_device, od_dt_dev);
544 static inline struct osd_device *osd_dev(const struct lu_device *d)
546 LASSERT(lu_device_is_osd(d));
547 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
550 static inline struct osd_device *osd_obj2dev(const struct osd_object *o)
552 return osd_dev(o->oo_dt.do_lu.lo_dev);
555 static inline struct super_block *osd_sb(const struct osd_device *dev)
557 return dev->od_mount->lmi_mnt->mnt_sb;
560 static inline int osd_object_is_root(const struct osd_object *obj)
562 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
565 static inline struct osd_object *osd_obj(const struct lu_object *o)
567 LASSERT(lu_device_is_osd(o->lo_dev));
568 return container_of0(o, struct osd_object, oo_dt.do_lu);
571 static inline struct osd_object *osd_dt_obj(const struct dt_object *d)
573 return osd_obj(&d->do_lu);
576 static inline struct lu_device *osd2lu_dev(struct osd_device *osd)
578 return &osd->od_dt_dev.dd_lu_dev;
581 static inline journal_t *osd_journal(const struct osd_device *dev)
583 return LDISKFS_SB(osd_sb(dev))->s_journal;
586 extern const struct dt_body_operations osd_body_ops;
587 extern struct lu_context_key osd_key;
589 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
591 return lu_context_key_get(&env->le_ctx, &osd_key);
594 #endif /* __KERNEL__ */
595 #endif /* _OSD_INTERNAL_H */