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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 /* Moved to lustre_user.h
58 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_ioctl_data *)
59 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
60 #define IOC_MDC_MAX_NR 50
62 #include <lustre/lustre_idl.h>
63 #include <lu_target.h>
65 #include <lustre_lib.h>
66 #include <lustre_export.h>
67 #include <lustre_quota.h>
68 #include <lustre_fld.h>
69 #include <lustre_capa.h>
71 #include <libcfs/bitmap.h>
74 #define MAX_OBD_DEVICES 8192
76 /* this is really local to the OSC */
77 struct loi_oap_pages {
78 cfs_list_t lop_pending;
79 cfs_list_t lop_urgent;
80 cfs_list_t lop_pending_group;
90 struct lov_oinfo { /* per-stripe data structure */
91 __u64 loi_id; /* object ID on the target OST */
92 __u64 loi_gr; /* object group on the target OST */
93 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
94 int loi_ost_gen; /* generation of this loi_ost_idx */
96 /* used by the osc to keep track of what objects to build into rpcs */
97 struct loi_oap_pages loi_read_lop;
98 struct loi_oap_pages loi_write_lop;
99 cfs_list_t loi_ready_item;
100 cfs_list_t loi_hp_ready_item;
101 cfs_list_t loi_write_item;
102 cfs_list_t loi_read_item;
104 unsigned long loi_kms_valid:1;
105 __u64 loi_kms; /* known minimum size */
106 struct ost_lvb loi_lvb;
107 struct osc_async_rc loi_ar;
110 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
112 oinfo->loi_kms = kms;
113 oinfo->loi_kms_valid = 1;
116 static inline void loi_init(struct lov_oinfo *loi)
118 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
119 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
120 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
121 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
122 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
123 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
124 CFS_INIT_LIST_HEAD(&loi->loi_ready_item);
125 CFS_INIT_LIST_HEAD(&loi->loi_hp_ready_item);
126 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
127 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
130 struct lov_stripe_md {
131 cfs_spinlock_t lsm_lock;
132 pid_t lsm_lock_owner; /* debugging */
135 /* Public members. */
136 __u64 lw_object_id; /* lov object id */
137 __u64 lw_object_gr; /* lov object group */
138 __u64 lw_maxbytes; /* maximum possible file size */
140 /* LOV-private members start here -- only for use in lov/. */
142 __u32 lw_stripe_size; /* size of the stripe */
143 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
144 unsigned lw_stripe_count; /* number of objects being striped over */
145 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
148 struct lov_oinfo *lsm_oinfo[0];
151 #define lsm_object_id lsm_wire.lw_object_id
152 #define lsm_object_gr lsm_wire.lw_object_gr
153 #define lsm_maxbytes lsm_wire.lw_maxbytes
154 #define lsm_magic lsm_wire.lw_magic
155 #define lsm_stripe_size lsm_wire.lw_stripe_size
156 #define lsm_pattern lsm_wire.lw_pattern
157 #define lsm_stripe_count lsm_wire.lw_stripe_count
158 #define lsm_pool_name lsm_wire.lw_pool_name
162 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
164 /* obd info for a particular level (lov, osc). */
166 /* Lock policy. It keeps an extent which is specific for a particular
167 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
168 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
169 ldlm_policy_data_t oi_policy;
170 /* Flags used for set request specific flags:
171 - while lock handling, the flags obtained on the enqueue
172 request are set here.
173 - while stats, the flags used for control delay/resend.
176 /* Lock handle specific for every OSC lock. */
177 struct lustre_handle *oi_lockh;
178 /* lsm data specific for every OSC. */
179 struct lov_stripe_md *oi_md;
180 /* obdo data specific for every OSC, if needed at all. */
182 /* statfs data specific for every OSC, if needed at all. */
183 struct obd_statfs *oi_osfs;
184 /* An update callback which is called to update some data on upper
185 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
186 * request in osc level for enqueue requests. It is also possible to
187 * update some caller data from LOV layer if needed. */
188 obd_enqueue_update_f oi_cb_up;
189 /* oss capability, its type is obd_capa in client to avoid copy.
190 * in contrary its type is lustre_capa in OSS. */
194 /* compare all relevant fields. */
195 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
196 struct lov_stripe_md *m2)
199 * ->lsm_wire contains padding, but it should be zeroed out during
202 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
205 void lov_stripe_lock(struct lov_stripe_md *md);
206 void lov_stripe_unlock(struct lov_stripe_md *md);
209 cfs_list_t typ_chain;
210 struct obd_ops *typ_dt_ops;
211 struct md_ops *typ_md_ops;
212 cfs_proc_dir_entry_t *typ_procroot;
215 struct lu_device_type *typ_lu;
216 cfs_spinlock_t obd_type_lock;
226 /* Individual type definitions */
228 struct ost_server_data;
230 /* hold common fields for "target" device */
231 struct obd_device_target {
232 struct super_block *obt_sb;
233 /** last_rcvd file */
234 struct file *obt_rcvd_filp;
235 struct lu_target *obt_lut;
236 __u64 obt_mount_count;
237 cfs_semaphore_t obt_quotachecking;
238 struct lustre_quota_ctxt obt_qctxt;
239 lustre_quota_version_t obt_qfmt;
240 cfs_rw_semaphore_t obt_rwsem;
241 struct vfsmount *obt_vfsmnt;
242 struct file *obt_health_check_filp;
247 LLOG_CONFIG_ORIG_CTXT = 0,
248 LLOG_CONFIG_REPL_CTXT,
249 LLOG_MDS_OST_ORIG_CTXT,
250 LLOG_MDS_OST_REPL_CTXT,
257 LLOG_LOVEA_ORIG_CTXT,
258 LLOG_LOVEA_REPL_CTXT,
259 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
260 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
261 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
265 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
267 struct filter_subdirs {
268 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
278 /* NB this field MUST be first */
279 struct obd_device_target fo_obt;
280 const char *fo_fstype;
283 cfs_dentry_t *fo_dentry_O;
284 cfs_dentry_t **fo_dentry_O_groups;
285 struct filter_subdirs *fo_dentry_O_sub;
286 cfs_semaphore_t fo_init_lock; /* group initialization lock */
287 int fo_committed_group;
289 #define CLIENT_QUOTA_DEFAULT_RESENDS 10
291 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
293 unsigned long fo_destroys_in_progress;
294 cfs_semaphore_t fo_create_locks[FILTER_SUBDIR_COUNT];
296 cfs_list_t fo_export_list;
299 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
300 obd_size fo_tot_granted; /* all values in bytes */
301 obd_size fo_tot_pending;
302 int fo_tot_granted_clients;
304 obd_size fo_readcache_max_filesize;
305 int fo_read_cache:1, /**< enable read-only cache */
306 fo_writethrough_cache:1,/**< read cache writes */
307 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
308 fo_raid_degraded:1;/**< RAID device degraded */
310 struct obd_import *fo_mdc_imp;
311 struct obd_uuid fo_mdc_uuid;
312 struct lustre_handle fo_mdc_conn;
313 struct file **fo_last_objid_files;
314 __u64 *fo_last_objids; /* last created objid for groups,
315 * protected by fo_objidlock */
317 cfs_semaphore_t fo_alloc_lock;
319 cfs_atomic_t fo_r_in_flight;
320 cfs_atomic_t fo_w_in_flight;
323 * per-filter pool of kiobuf's allocated by filter_common_setup() and
324 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
325 * which ->fo_iobuf_count were allocated.
327 * This pool contains kiobuf used by
328 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
331 * Locking: none, each OST thread uses only one element, determined by
332 * its "ordinal number", ->t_id.
334 struct filter_iobuf **fo_iobuf_pool;
337 cfs_list_t fo_llog_list;
338 cfs_spinlock_t fo_llog_list_lock;
340 struct brw_stats fo_filter_stats;
341 struct lustre_quota_ctxt fo_quota_ctxt;
342 cfs_spinlock_t fo_quotacheck_lock;
343 cfs_atomic_t fo_quotachecking;
345 int fo_fmd_max_num; /* per exp filter_mod_data */
346 int fo_fmd_max_age; /* jiffies to fmd expiry */
349 cfs_rwlock_t fo_sptlrpc_lock;
350 struct sptlrpc_rule_set fo_sptlrpc_rset;
352 /* capability related */
353 unsigned int fo_fl_oss_capa;
354 cfs_list_t fo_capa_keys;
355 cfs_hlist_head_t *fo_capa_hash;
356 struct llog_commit_master *fo_lcm;
360 #define fo_translock fo_obt.obt_lut->lut_translock
361 #define fo_last_rcvd_slots fo_obt.obt_lut->lut_client_bitmap
362 #define fo_mount_count fo_obt.obt_lut->lut_mount_count
363 #define fo_rcvd_filp fo_obt.obt_rcvd_filp
364 #define fo_vfsmnt fo_obt.obt_vfsmnt
366 struct timeout_item {
367 enum timeout_event ti_event;
368 cfs_time_t ti_timeout;
371 cfs_list_t ti_obd_list;
375 #define OSC_MAX_RIF_DEFAULT 8
376 #define OSC_MAX_RIF_MAX 256
377 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
378 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
379 #define OSC_DEFAULT_RESENDS 10
381 #define MDC_MAX_RIF_DEFAULT 8
382 #define MDC_MAX_RIF_MAX 512
387 cfs_rw_semaphore_t cl_sem;
388 struct obd_uuid cl_target_uuid;
389 struct obd_import *cl_import; /* ptlrpc connection state */
391 /* max_mds_easize is purely a performance thing so we don't have to
392 * call obd_size_diskmd() all the time. */
393 int cl_default_mds_easize;
394 int cl_max_mds_easize;
395 int cl_max_mds_cookiesize;
397 enum lustre_sec_part cl_sp_me;
398 enum lustre_sec_part cl_sp_to;
399 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
401 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
402 void *cl_llcd_offset;
404 /* the grant values are protected by loi_list_lock below */
405 long cl_dirty; /* all _dirty_ in bytes */
406 long cl_dirty_max; /* allowed w/o rpc */
407 long cl_dirty_transit; /* dirty synchronous */
408 long cl_avail_grant; /* bytes of credit for ost */
409 long cl_lost_grant; /* lost credits (trunc) */
410 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
411 cfs_time_t cl_next_shrink_grant; /* jiffies */
412 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
413 cfs_semaphore_t cl_grant_sem; /*grant shrink list cfs_semaphore*/
414 int cl_grant_shrink_interval; /* seconds */
416 /* keep track of objects that have lois that contain pages which
417 * have been queued for async brw. this lock also protects the
418 * lists of osc_client_pages that hang off of the loi */
420 * ->cl_loi_list_lock protects consistency of
421 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
422 * ->ap_completion() call-backs are executed under this lock. As we
423 * cannot guarantee that these call-backs never block on all platforms
424 * (as a matter of fact they do block on Mac OS X), type of
425 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
426 * and blocking mutex on Mac OS X. (Alternative is to make this lock
427 * blocking everywhere, but we don't want to slow down fast-path of
428 * our main platform.)
430 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
431 * with client_obd_list_{un,}lock() and
432 * client_obd_list_lock_{init,done}() functions.
434 client_obd_lock_t cl_loi_list_lock;
435 cfs_list_t cl_loi_ready_list;
436 cfs_list_t cl_loi_hp_ready_list;
437 cfs_list_t cl_loi_write_list;
438 cfs_list_t cl_loi_read_list;
441 /* just a sum of the loi/lop pending numbers to be exported by /proc */
442 int cl_pending_w_pages;
443 int cl_pending_r_pages;
444 int cl_max_pages_per_rpc;
445 int cl_max_rpcs_in_flight;
446 struct obd_histogram cl_read_rpc_hist;
447 struct obd_histogram cl_write_rpc_hist;
448 struct obd_histogram cl_read_page_hist;
449 struct obd_histogram cl_write_page_hist;
450 struct obd_histogram cl_read_offset_hist;
451 struct obd_histogram cl_write_offset_hist;
453 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
454 cfs_atomic_t cl_destroy_in_flight;
455 cfs_waitq_t cl_destroy_waitq;
457 struct mdc_rpc_lock *cl_rpc_lock;
458 struct mdc_rpc_lock *cl_setattr_lock;
459 struct mdc_rpc_lock *cl_close_lock;
460 struct osc_creator cl_oscc;
463 cfs_semaphore_t cl_mgc_sem;
464 struct vfsmount *cl_mgc_vfsmnt;
465 struct dentry *cl_mgc_configs_dir;
466 cfs_atomic_t cl_mgc_refcount;
467 struct obd_export *cl_mgc_mgsexp;
469 /* checksumming for data sent over the network */
470 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
471 /* supported checksum types that are worked out at connect time */
472 __u32 cl_supp_cksum_types;
473 /* checksum algorithm to be used */
474 cksum_type_t cl_cksum_type;
476 /* also protected by the poorly named _loi_list_lock lock above */
477 struct osc_async_rc cl_ar;
479 /* used by quotacheck */
480 int cl_qchk_stat; /* quotacheck stat of the peer */
482 /* sequence manager */
483 struct lu_client_seq *cl_seq;
485 cfs_atomic_t cl_resends; /* resend count */
486 cfs_atomic_t cl_quota_resends; /* quota related resend count */
488 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
490 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
493 struct ptlrpc_service *mgs_service;
494 struct vfsmount *mgs_vfsmnt;
495 struct super_block *mgs_sb;
496 struct dentry *mgs_configs_dir;
497 struct dentry *mgs_fid_de;
498 cfs_list_t mgs_fs_db_list;
499 cfs_semaphore_t mgs_sem;
500 cfs_proc_dir_entry_t *mgs_proc_live;
504 /* NB this field MUST be first */
505 struct obd_device_target mds_obt;
506 struct ptlrpc_service *mds_service;
507 struct ptlrpc_service *mds_setattr_service;
508 struct ptlrpc_service *mds_readpage_service;
509 cfs_dentry_t *mds_fid_de;
511 int mds_max_cookiesize;
513 unsigned long mds_atime_diff;
514 cfs_semaphore_t mds_epoch_sem;
515 struct ll_fid mds_rootfid;
516 cfs_dentry_t *mds_pending_dir;
517 cfs_dentry_t *mds_logs_dir;
518 cfs_dentry_t *mds_objects_dir;
519 struct llog_handle *mds_cfg_llh;
520 struct obd_device *mds_osc_obd; /* XXX lov_obd */
521 struct obd_uuid mds_lov_uuid;
523 struct obd_export *mds_osc_exp; /* XXX lov_exp */
524 struct lov_desc mds_lov_desc;
527 /* mark pages dirty for write. */
528 cfs_bitmap_t *mds_lov_page_dirty;
529 /* array for store pages with obd_id */
530 void **mds_lov_page_array;
531 /* file for store objid */
532 struct file *mds_lov_objid_filp;
533 __u32 mds_lov_objid_count;
534 __u32 mds_lov_objid_max_index;
535 __u32 mds_lov_objid_lastpage;
536 __u32 mds_lov_objid_lastidx;
539 struct lustre_quota_info mds_quota_info;
540 cfs_semaphore_t mds_qonoff_sem;
541 cfs_semaphore_t mds_health_sem;
542 unsigned long mds_fl_user_xattr:1,
544 mds_evict_ost_nids:1,
548 mds_fl_target:1; /* mds have one or
551 struct upcall_cache *mds_identity_cache;
553 /* for capability keys update */
554 struct lustre_capa_key *mds_capa_keys;
555 cfs_rw_semaphore_t mds_notify_lock;
558 #define mds_transno_lock mds_obt.obt_translock
559 #define mds_rcvd_filp mds_obt.obt_rcvd_filp
560 #define mds_server_data mds_obt.obt_lsd
561 #define mds_client_bitmap mds_obt.obt_client_bitmap
562 #define mds_mount_count mds_obt.obt_mount_count
563 #define mds_last_transno mds_obt.obt_last_transno
564 #define mds_vfsmnt mds_obt.obt_vfsmnt
567 extern __u32 mds_max_ost_index;
569 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
571 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
573 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
575 extern int mds_lov_init_objids(struct obd_device *obd);
576 extern void mds_lov_destroy_objids(struct obd_device *obd);
587 cfs_spinlock_t eo_lock;
589 struct lustre_handle eo_nl_lock;
590 cfs_atomic_t eo_prep;
594 struct ptlrpc_service *ost_service;
595 struct ptlrpc_service *ost_create_service;
596 struct ptlrpc_service *ost_io_service;
597 cfs_semaphore_t ost_health_sem;
600 struct echo_client_obd {
601 struct obd_export *ec_exp; /* the local connection to osc/lov */
602 cfs_spinlock_t ec_lock;
603 cfs_list_t ec_objects;
610 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
611 cfs_list_t lqo_oss_list; /* link to lov_qos */
612 __u64 lqo_bavail; /* total bytes avail on OSS */
613 __u64 lqo_penalty; /* current penalty */
614 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
615 time_t lqo_used; /* last used time, seconds */
616 __u32 lqo_ost_count; /* number of osts on this oss */
620 struct lov_qos_oss *ltq_oss; /* oss info */
621 __u64 ltq_penalty; /* current penalty */
622 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
623 __u64 ltq_weight; /* net weighting */
624 time_t ltq_used; /* last used time, seconds */
625 unsigned int ltq_usable:1; /* usable for striping */
628 /* Generic subset of OSTs */
630 __u32 *op_array; /* array of index of
632 unsigned int op_count; /* number of OSTs in the array */
633 unsigned int op_size; /* allocated size of lp_array */
634 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
637 /* Round-robin allocator data */
639 __u32 lqr_start_idx; /* start index of new inode */
640 __u32 lqr_offset_idx; /* aliasing for start_idx */
641 int lqr_start_count; /* reseed counter */
642 struct ost_pool lqr_pool; /* round-robin optimized list */
643 unsigned long lqr_dirty:1; /* recalc round-robin list */
646 struct lov_statfs_data {
647 struct obd_info lsd_oi;
648 struct obd_statfs lsd_statfs;
650 /* Stripe placement optimization */
652 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
653 cfs_rw_semaphore_t lq_rw_sem;
654 __u32 lq_active_oss_count;
655 unsigned int lq_prio_free; /* priority for free space */
656 unsigned int lq_threshold_rr;/* priority for rr */
657 struct lov_qos_rr lq_rr; /* round robin qos data */
658 unsigned long lq_dirty:1, /* recalc qos data */
659 lq_same_space:1,/* the ost's all have approx.
660 the same space avail */
661 lq_reset:1, /* zero current penalties */
662 lq_statfs_in_progress:1; /* statfs op in
664 /* qos statfs data */
665 struct lov_statfs_data *lq_statfs_data;
666 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
667 * requests completion */
670 struct lov_tgt_desc {
672 struct obd_uuid ltd_uuid;
673 struct obd_device *ltd_obd;
674 struct obd_export *ltd_exp;
675 struct ltd_qos ltd_qos; /* qos info per target */
677 __u32 ltd_index; /* index in lov_obd->tgts */
678 unsigned long ltd_active:1,/* is this target up for requests */
679 ltd_activate:1,/* should target be activated */
680 ltd_reap:1; /* should this target be deleted */
684 #define pool_tgt_size(_p) _p->pool_obds.op_size
685 #define pool_tgt_count(_p) _p->pool_obds.op_count
686 #define pool_tgt_array(_p) _p->pool_obds.op_array
687 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
688 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
691 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
692 struct ost_pool pool_obds; /* pool members */
693 cfs_atomic_t pool_refcount; /* pool ref. counter */
694 struct lov_qos_rr pool_rr; /* round robin qos */
695 cfs_hlist_node_t pool_hash; /* access by poolname */
696 cfs_list_t pool_list; /* serial access */
697 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
698 struct lov_obd *pool_lov; /* lov obd to which this
703 struct lov_desc desc;
704 struct lov_tgt_desc **lov_tgts; /* sparse array */
705 struct ost_pool lov_packed; /* all OSTs in a packed
707 cfs_semaphore_t lov_lock;
708 struct obd_connect_data lov_ocd;
709 struct lov_qos lov_qos; /* qos info per lov */
710 cfs_atomic_t lov_refcount;
711 __u32 lov_tgt_count; /* how many OBD's */
712 __u32 lov_active_tgt_count; /* how many active */
713 __u32 lov_death_row;/* tgts scheduled to be deleted */
714 __u32 lov_tgt_size; /* size of tgts array */
717 cfs_hash_t *lov_pools_hash_body; /* used for key access */
718 cfs_list_t lov_pool_list; /* used for sequential access */
719 cfs_proc_dir_entry_t *lov_pool_proc_entry;
720 enum lustre_sec_part lov_sp_me;
723 struct lmv_tgt_desc {
724 struct obd_uuid ltd_uuid;
725 struct obd_export *ltd_exp;
726 int ltd_active; /* is this target up for requests */
728 cfs_semaphore_t ltd_fid_sem;
731 enum placement_policy {
732 PLACEMENT_CHAR_POLICY = 0,
733 PLACEMENT_NID_POLICY = 1,
734 PLACEMENT_INVAL_POLICY = 2,
738 typedef enum placement_policy placement_policy_t;
742 struct lu_client_fld lmv_fld;
743 cfs_spinlock_t lmv_lock;
744 placement_policy_t lmv_placement;
745 struct lmv_desc desc;
746 struct obd_uuid cluuid;
747 struct obd_export *exp;
754 cfs_semaphore_t init_sem;
756 struct lmv_tgt_desc *tgts;
759 struct obd_connect_data *datas;
762 struct obd_connect_data conn_data;
765 struct niobuf_local {
770 cfs_dentry_t *dentry;
775 #define LUSTRE_FLD_NAME "fld"
776 #define LUSTRE_SEQ_NAME "seq"
778 #define LUSTRE_CMM_NAME "cmm"
779 #define LUSTRE_MDD_NAME "mdd"
780 #define LUSTRE_OSD_NAME "osd"
781 #define LUSTRE_VVP_NAME "vvp"
782 #define LUSTRE_LMV_NAME "lmv"
783 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
784 #define LUSTRE_SLP_NAME "slp"
786 /* obd device type names */
787 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
788 #define LUSTRE_MDS_NAME "mds"
789 #define LUSTRE_MDT_NAME "mdt"
790 #define LUSTRE_MDC_NAME "mdc"
791 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
792 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
793 #define LUSTRE_OSC_NAME "osc"
794 #define LUSTRE_LOV_NAME "lov"
795 #define LUSTRE_MGS_NAME "mgs"
796 #define LUSTRE_MGC_NAME "mgc"
798 #define LUSTRE_CACHEOBD_NAME "cobd"
799 #define LUSTRE_ECHO_NAME "obdecho"
800 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
802 /* Constant obd names (post-rename) */
803 #define LUSTRE_MDS_OBDNAME "MDS"
804 #define LUSTRE_OSS_OBDNAME "OSS"
805 #define LUSTRE_MGS_OBDNAME "MGS"
806 #define LUSTRE_MGC_OBDNAME "MGC"
808 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
809 #define N_LOCAL_TEMP_PAGE 0x10000000
811 struct obd_trans_info {
814 /* Only used on the server side for tracking acks. */
815 struct oti_req_ack_lock {
816 struct lustre_handle lock;
820 struct llog_cookie oti_onecookie;
821 struct llog_cookie *oti_logcookies;
824 /* initial thread handling transaction */
825 struct ptlrpc_thread * oti_thread;
828 __u64 oti_pre_version;
830 struct obd_uuid *oti_ost_uuid;
833 static inline void oti_init(struct obd_trans_info *oti,
834 struct ptlrpc_request *req)
838 memset(oti, 0, sizeof(*oti));
843 oti->oti_xid = req->rq_xid;
844 /** VBR: take versions from request */
845 if (req->rq_reqmsg != NULL &&
846 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
847 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
848 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
849 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
852 /** called from mds_create_objects */
853 if (req->rq_repmsg != NULL)
854 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
855 oti->oti_thread = req->rq_svc_thread;
856 if (req->rq_reqmsg != NULL)
857 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
860 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
865 if (num_cookies == 1)
866 oti->oti_logcookies = &oti->oti_onecookie;
868 OBD_ALLOC(oti->oti_logcookies,
869 num_cookies * sizeof(oti->oti_onecookie));
871 oti->oti_numcookies = num_cookies;
874 static inline void oti_free_cookies(struct obd_trans_info *oti)
876 if (!oti || !oti->oti_logcookies)
879 if (oti->oti_logcookies == &oti->oti_onecookie)
880 LASSERT(oti->oti_numcookies == 1);
882 OBD_FREE(oti->oti_logcookies,
883 oti->oti_numcookies * sizeof(oti->oti_onecookie));
884 oti->oti_logcookies = NULL;
885 oti->oti_numcookies = 0;
889 * Events signalled through obd_notify() upcall-chain.
891 enum obd_notify_event {
892 /* Device connect start */
894 /* Device activated */
896 /* Device deactivated */
898 /* Device disconnected */
900 /* Connect data for import were changed */
903 OBD_NOTIFY_SYNC_NONBLOCK,
905 /* Configuration event */
907 /* Trigger quota recovery */
911 /* bit-mask flags for config events */
913 CONFIG_LOG = 0x1, /* finished processing config log */
914 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
915 CONFIG_TARGET = 0x4 /* one target is added */
919 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
920 * and liblustre being main examples).
922 struct obd_notify_upcall {
923 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
924 enum obd_notify_event ev, void *owner, void *data);
925 /* Opaque datum supplied by upper layer listener */
929 struct target_recovery_data {
930 svc_handler_t trd_recovery_handler;
931 pid_t trd_processing_task;
932 cfs_completion_t trd_starting;
933 cfs_completion_t trd_finishing;
937 FILTER_GROUP_MDS0 = 0,
938 FILTER_GROUP_LLOG = 1,
939 FILTER_GROUP_ECHO = 2 ,
940 FILTER_GROUP_MDS1_N_BASE = 3
944 * In HEAD for CMD, the object is created in group number which is 3>=
945 * or indexing starts from 3. To test this assertions are added to disallow
946 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
947 * object in group 0 needs to be allowed.
948 * So for interop mode following changes needs to be done:
949 * 1. No need to assert on group 0 or allow group 0
950 * 2. The group number indexing starts from 0 instead of 3
953 static inline int filter_group_is_mds(obd_gr group)
955 return (group == FILTER_GROUP_MDS0 ||
956 group >= FILTER_GROUP_MDS1_N_BASE);
959 #define LASSERT_MDS_GROUP(group) LASSERT(filter_group_is_mds(group))
961 static inline __u64 objgrp_to_mdsno(obd_gr group)
963 LASSERT(filter_group_is_mds(group));
964 if (group == FILTER_GROUP_MDS0)
966 return group - FILTER_GROUP_MDS1_N_BASE + 1;
969 static inline int mdt_to_obd_objgrp(int mdtid)
972 * MDS0 uses group 0 always, other MDSes will use groups from
973 * FILTER_GROUP_MDS1_N_BASE
976 return FILTER_GROUP_MDS1_N_BASE + mdtid - 1;
980 static inline __u64 obdo_mdsno(struct obdo *oa)
982 LASSERT((oa->o_valid & OBD_MD_FLGROUP));
983 return objgrp_to_mdsno(oa->o_gr);
986 static inline int obdo_is_mds(struct obdo *oa)
988 LASSERT(oa->o_valid & OBD_MD_FLGROUP);
989 return filter_group_is_mds(oa->o_gr);
992 struct obd_llog_group {
995 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
996 cfs_waitq_t olg_waitq;
997 cfs_spinlock_t olg_lock;
998 struct obd_export *olg_exp;
999 int olg_initializing;
1000 cfs_semaphore_t olg_cat_processing;
1003 /* corresponds to one of the obd's */
1004 #define MAX_OBD_NAME 128
1005 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1006 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1008 struct obd_type *obd_type;
1011 /* common and UUID name of this device */
1012 char obd_name[MAX_OBD_NAME];
1013 struct obd_uuid obd_uuid;
1015 struct lu_device *obd_lu_dev;
1018 unsigned long obd_attached:1, /* finished attach */
1019 obd_set_up:1, /* finished setup */
1020 obd_recovering:1, /* there are recoverable clients */
1021 obd_abort_recovery:1,/* recovery expired */
1022 obd_version_recov:1, /* obd uses version checking */
1023 obd_recovery_expired:1,
1024 obd_replayable:1, /* recovery is enabled; inform clients */
1025 obd_no_transno:1, /* no committed-transno notification */
1026 obd_no_recov:1, /* fail instead of retry messages */
1027 obd_req_replaying:1, /* replaying requests */
1028 obd_stopping:1, /* started cleanup */
1029 obd_starting:1, /* started setup */
1030 obd_force:1, /* cleanup with > 0 obd refcount */
1031 obd_fail:1, /* cleanup with failover */
1032 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1033 obd_no_conn:1, /* deny new connections */
1034 obd_inactive:1, /* device active/inactive
1035 * (for /proc/status only!!) */
1036 obd_process_conf:1; /* device is processing mgs config */
1037 /* uuid-export hash body */
1038 cfs_hash_t *obd_uuid_hash;
1039 /* nid-export hash body */
1040 cfs_hash_t *obd_nid_hash;
1041 /* nid stats body */
1042 cfs_hash_t *obd_nid_stats_hash;
1043 cfs_list_t obd_nid_stats;
1044 cfs_atomic_t obd_refcount;
1045 cfs_waitq_t obd_refcount_waitq;
1046 cfs_list_t obd_exports;
1047 cfs_list_t obd_unlinked_exports;
1048 cfs_list_t obd_delayed_exports;
1049 int obd_num_exports;
1050 cfs_spinlock_t obd_nid_lock;
1051 struct ldlm_namespace *obd_namespace;
1052 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1053 /* a spinlock is OK for what we do now, may need a semaphore later */
1054 cfs_spinlock_t obd_dev_lock;
1055 cfs_semaphore_t obd_dev_sem;
1056 __u64 obd_last_committed;
1057 struct fsfilt_operations *obd_fsops;
1058 cfs_spinlock_t obd_osfs_lock;
1059 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1061 struct lvfs_run_ctxt obd_lvfs_ctxt;
1062 struct obd_llog_group obd_olg; /* default llog group */
1063 struct obd_device *obd_observer;
1064 cfs_rw_semaphore_t obd_observer_link_sem;
1065 struct obd_notify_upcall obd_upcall;
1066 struct obd_export *obd_self_export;
1067 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1068 cfs_list_t obd_exports_timed;
1069 time_t obd_eviction_timer; /* for ping evictor */
1071 int obd_max_recoverable_clients;
1072 int obd_connected_clients;
1073 int obd_stale_clients;
1074 int obd_delayed_clients;
1075 cfs_spinlock_t obd_processing_task_lock; /* BH lock (timer) */
1076 __u64 obd_next_recovery_transno;
1077 int obd_replayed_requests;
1078 int obd_requests_queued_for_recovery;
1079 cfs_waitq_t obd_next_transno_waitq;
1080 cfs_timer_t obd_recovery_timer;
1081 time_t obd_recovery_start; /* seconds */
1082 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1083 time_t obd_recovery_max_time; /* seconds, bz13079 */
1084 int obd_recovery_timeout;
1086 /* new recovery stuff from CMD2 */
1087 struct target_recovery_data obd_recovery_data;
1088 int obd_replayed_locks;
1089 cfs_atomic_t obd_req_replay_clients;
1090 cfs_atomic_t obd_lock_replay_clients;
1091 cfs_list_t obd_req_replay_queue;
1092 cfs_list_t obd_lock_replay_queue;
1093 cfs_list_t obd_final_req_queue;
1094 int obd_recovery_stage;
1097 struct obd_device_target obt;
1098 struct filter_obd filter;
1100 struct client_obd cli;
1102 struct echo_client_obd echo_client;
1103 struct echo_obd echo;
1108 /* Fields used by LProcFS */
1109 unsigned int obd_cntr_base;
1110 struct lprocfs_stats *obd_stats;
1112 unsigned int md_cntr_base;
1113 struct lprocfs_stats *md_stats;
1115 cfs_proc_dir_entry_t *obd_proc_entry;
1116 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1117 cfs_proc_dir_entry_t *obd_svc_procroot;
1118 struct lprocfs_stats *obd_svc_stats;
1119 cfs_atomic_t obd_evict_inprogress;
1120 cfs_waitq_t obd_evict_inprogress_waitq;
1121 cfs_list_t obd_evict_list; /* protected with pet_lock */
1124 * Ldlm pool part. Save last calculated SLV and Limit.
1126 cfs_rwlock_t obd_pool_lock;
1131 * A list of outstanding class_incref()'s against this obd. For
1134 struct lu_ref obd_reference;
1137 #define OBD_LLOG_FL_SENDNOW 0x0001
1139 enum obd_cleanup_stage {
1140 /* Special case hack for MDS LOVs */
1142 /* can be directly mapped to .ldto_device_fini() */
1143 OBD_CLEANUP_EXPORTS,
1146 /* get/set_info keys */
1147 #define KEY_ASYNC "async"
1148 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1149 #define KEY_BLOCKSIZE "blocksize"
1150 #define KEY_CAPA_KEY "capa_key"
1151 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1152 #define KEY_FID2PATH "fid2path"
1153 #define KEY_CHECKSUM "checksum"
1154 #define KEY_CLEAR_FS "clear_fs"
1155 #define KEY_CONN_DATA "conn_data"
1156 #define KEY_EVICT_BY_NID "evict_by_nid"
1157 #define KEY_FIEMAP "fiemap"
1158 #define KEY_FLUSH_CTX "flush_ctx"
1159 #define KEY_GRANT_SHRINK "grant_shrink"
1160 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1161 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1162 #define KEY_INIT_RECOV "initial_recov"
1163 #define KEY_INTERMDS "inter_mds"
1164 #define KEY_LAST_ID "last_id"
1165 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1166 #define KEY_LOVDESC "lovdesc"
1167 #define KEY_LOV_IDX "lov_idx"
1168 #define KEY_MAX_EASIZE "max_easize"
1169 #define KEY_MDS_CONN "mds_conn"
1170 #define KEY_MGSSEC "mgssec"
1171 #define KEY_NEXT_ID "next_id"
1172 #define KEY_READ_ONLY "read-only"
1173 #define KEY_REGISTER_TARGET "register_target"
1174 #define KEY_REVIMP_UPD "revimp_update"
1175 #define KEY_SET_FS "set_fs"
1176 /* KEY_SET_INFO in lustre_idl.h */
1177 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1182 static inline int it_to_lock_mode(struct lookup_intent *it)
1184 /* CREAT needs to be tested before open (both could be set) */
1185 if (it->it_op & IT_CREAT)
1187 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
1190 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1195 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1196 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1197 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1198 struct lu_fid op_fid4; /* to the operation locks. */
1199 mdsno_t op_mds; /* what mds server open will go to */
1200 struct lustre_handle op_handle;
1202 const char *op_name;
1205 struct lmv_stripe_md *op_mea1;
1206 struct lmv_stripe_md *op_mea2;
1207 __u32 op_suppgids[2];
1213 /* iattr fields and blocks. */
1214 struct iattr op_attr;
1216 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
1217 unsigned int op_attr_flags;
1221 loff_t op_attr_blocks;
1223 /* Size-on-MDS epoch and flags. */
1228 struct obd_capa *op_capa1;
1229 struct obd_capa *op_capa2;
1231 /* Various operation flags. */
1234 /* Operation type */
1238 struct md_enqueue_info;
1239 /* metadata stat-ahead */
1240 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1241 struct md_enqueue_info *minfo,
1244 struct md_enqueue_info {
1245 struct md_op_data mi_data;
1246 struct lookup_intent mi_it;
1247 struct lustre_handle mi_lockh;
1248 struct dentry *mi_dentry;
1249 struct inode *mi_dir;
1250 md_enqueue_cb_t mi_cb;
1251 unsigned int mi_generation;
1256 cfs_module_t *o_owner;
1257 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1258 void *karg, void *uarg);
1259 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
1260 __u32 *vallen, void *val, struct lov_stripe_md *lsm);
1261 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
1262 __u32 vallen, void *val,
1263 struct ptlrpc_request_set *set);
1264 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1265 int (*o_detach)(struct obd_device *dev);
1266 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1267 int (*o_precleanup)(struct obd_device *dev,
1268 enum obd_cleanup_stage cleanup_stage);
1269 int (*o_cleanup)(struct obd_device *dev);
1270 int (*o_process_config)(struct obd_device *dev, obd_count len,
1272 int (*o_postrecov)(struct obd_device *dev);
1273 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1275 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1276 /* connect to the target device with given connection
1277 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1278 * granted by the target, which are guaranteed to be a subset of flags
1279 * asked for. If @ocd == NULL, use default parameters. */
1280 int (*o_connect)(const struct lu_env *env,
1281 struct obd_export **exp, struct obd_device *src,
1282 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1284 int (*o_reconnect)(const struct lu_env *env,
1285 struct obd_export *exp, struct obd_device *src,
1286 struct obd_uuid *cluuid,
1287 struct obd_connect_data *ocd,
1289 int (*o_disconnect)(struct obd_export *exp);
1291 /* Initialize/finalize fids infrastructure. */
1292 int (*o_fid_init)(struct obd_export *exp);
1293 int (*o_fid_fini)(struct obd_export *exp);
1295 /* Allocate new fid according to passed @hint. */
1296 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1297 struct md_op_data *op_data);
1300 * Object with @fid is getting deleted, we may want to do something
1303 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1305 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
1306 __u64 max_age, __u32 flags);
1307 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
1308 __u64 max_age, struct ptlrpc_request_set *set);
1309 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1310 struct lov_stripe_md *mem_src);
1311 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1312 struct lov_mds_md *disk_src, int disk_len);
1313 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1315 /* FIXME: add fid capability support for create & destroy! */
1316 int (*o_precreate)(struct obd_export *exp);
1317 int (*o_create)(struct obd_export *exp, struct obdo *oa,
1318 struct lov_stripe_md **ea, struct obd_trans_info *oti);
1319 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1320 struct lov_stripe_md **ea,
1321 struct obd_trans_info *oti);
1322 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
1323 struct lov_stripe_md *ea, struct obd_trans_info *oti,
1324 struct obd_export *md_exp, void *capa);
1325 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
1326 struct obd_trans_info *oti);
1327 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1328 struct obd_trans_info *oti,
1329 struct ptlrpc_request_set *rqset);
1330 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
1331 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1332 struct ptlrpc_request_set *set);
1333 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1334 obd_count oa_bufs, struct brw_page *pgarr,
1335 struct obd_trans_info *oti);
1336 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1337 struct ost_lvb *lvb, int kms_only);
1338 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1339 obd_off size, int shrink);
1340 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
1341 struct obd_trans_info *oti,
1342 struct ptlrpc_request_set *rqset);
1343 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
1344 struct lov_stripe_md *ea, obd_size start, obd_size end,
1346 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1347 struct lov_stripe_md *src, obd_size start,
1348 obd_size end, struct obd_trans_info *oti);
1349 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1350 struct lustre_handle *srconn, struct lov_stripe_md *src,
1351 obd_size start, obd_size end, struct obd_trans_info *);
1352 int (*o_iterate)(struct lustre_handle *conn,
1353 int (*)(obd_id, obd_gr, void *),
1354 obd_id *startid, obd_gr group, void *data);
1355 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
1356 int objcount, struct obd_ioobj *obj,
1357 struct niobuf_remote *remote, int *nr_pages,
1358 struct niobuf_local *local,
1359 struct obd_trans_info *oti,
1360 struct lustre_capa *capa);
1361 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
1362 int objcount, struct obd_ioobj *obj,
1363 struct niobuf_remote *remote, int pages,
1364 struct niobuf_local *local,
1365 struct obd_trans_info *oti, int rc);
1366 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1367 struct ldlm_enqueue_info *einfo,
1368 struct ptlrpc_request_set *rqset);
1369 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1370 ldlm_iterator_t it, void *data);
1371 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1372 __u32 mode, struct lustre_handle *);
1373 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1374 int flags, void *opaque);
1375 int (*o_init_export)(struct obd_export *exp);
1376 int (*o_destroy_export)(struct obd_export *exp);
1377 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1378 int cmd, obd_off *);
1380 /* llog related obd_methods */
1381 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1382 struct obd_device *disk_obd, int *idx);
1383 int (*o_llog_finish)(struct obd_device *obd, int count);
1384 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1386 /* metadata-only methods */
1387 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1388 struct obd_capa *, struct obd_client_handle *, int flag);
1389 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1391 int (*o_import_event)(struct obd_device *, struct obd_import *,
1392 enum obd_import_event);
1394 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1395 enum obd_notify_event ev, void *data);
1397 int (*o_health_check)(struct obd_device *);
1398 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1401 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1402 struct obd_quotactl *);
1403 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1404 struct obd_quotactl *);
1405 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1406 struct quota_adjust_qunit *oqaq,
1407 struct lustre_quota_ctxt *qctxt);
1410 int (*o_ping)(struct obd_export *exp);
1413 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1414 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1415 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1417 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1419 void (*o_getref)(struct obd_device *obd);
1420 void (*o_putref)(struct obd_device *obd);
1422 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1423 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1424 * Also, add a wrapper function in include/linux/obd_class.h. */
1428 LUSTRE_OPC_MKDIR = (1 << 0),
1429 LUSTRE_OPC_SYMLINK = (1 << 1),
1430 LUSTRE_OPC_MKNOD = (1 << 2),
1431 LUSTRE_OPC_CREATE = (1 << 3),
1432 LUSTRE_OPC_ANY = (1 << 4)
1435 /* lmv structures */
1436 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1437 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1438 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1440 #define MAX_HASH_SIZE_32 0x7fffffffUL
1441 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1442 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1445 struct mdt_body *body;
1446 struct lov_stripe_md *lsm;
1447 struct lmv_stripe_md *mea;
1448 #ifdef CONFIG_FS_POSIX_ACL
1449 struct posix_acl *posix_acl;
1451 struct mdt_remote_perm *remote_perm;
1452 struct obd_capa *mds_capa;
1453 struct obd_capa *oss_capa;
1456 struct md_open_data {
1457 struct obd_client_handle *mod_och;
1458 struct ptlrpc_request *mod_open_req;
1459 struct ptlrpc_request *mod_close_req;
1460 cfs_atomic_t mod_refcount;
1463 struct lookup_intent;
1466 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1467 struct obd_capa **);
1468 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1469 ldlm_iterator_t, void *);
1470 int (*m_close)(struct obd_export *, struct md_op_data *,
1471 struct md_open_data *, struct ptlrpc_request **);
1472 int (*m_create)(struct obd_export *, struct md_op_data *,
1473 const void *, int, int, __u32, __u32, cfs_cap_t,
1474 __u64, struct ptlrpc_request **);
1475 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1476 struct md_open_data *);
1477 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1478 struct lookup_intent *, struct md_op_data *,
1479 struct lustre_handle *, void *, int,
1480 struct ptlrpc_request **, int);
1481 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1482 struct ptlrpc_request **);
1483 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1484 struct ptlrpc_request **);
1485 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1486 void *, int, struct lookup_intent *, int,
1487 struct ptlrpc_request **,
1488 ldlm_blocking_callback, int);
1489 int (*m_link)(struct obd_export *, struct md_op_data *,
1490 struct ptlrpc_request **);
1491 int (*m_rename)(struct obd_export *, struct md_op_data *,
1492 const char *, int, const char *, int,
1493 struct ptlrpc_request **);
1494 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1495 const struct lu_fid *,
1496 struct ptlrpc_request **);
1497 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1498 int , void *, int, struct ptlrpc_request **,
1499 struct md_open_data **mod);
1500 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1501 struct obd_capa *, struct ptlrpc_request **);
1502 int (*m_readpage)(struct obd_export *, const struct lu_fid *,
1503 struct obd_capa *, __u64, struct page *,
1504 struct ptlrpc_request **);
1506 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1507 struct ptlrpc_request **);
1509 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1510 struct obd_capa *, obd_valid, const char *,
1511 const char *, int, int, int, __u32,
1512 struct ptlrpc_request **);
1514 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1515 struct obd_capa *, obd_valid, const char *,
1516 const char *, int, int, int,
1517 struct ptlrpc_request **);
1519 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1521 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1522 struct obd_export *, struct obd_export *,
1523 struct lustre_md *);
1525 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1527 int (*m_set_open_replay_data)(struct obd_export *,
1528 struct obd_client_handle *,
1529 struct ptlrpc_request *);
1530 int (*m_clear_open_replay_data)(struct obd_export *,
1531 struct obd_client_handle *);
1532 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u32 *);
1534 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1535 const struct lu_fid *, ldlm_type_t,
1536 ldlm_policy_data_t *, ldlm_mode_t,
1537 struct lustre_handle *);
1539 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1540 ldlm_policy_data_t *, ldlm_mode_t, int flags,
1542 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1543 renew_capa_cb_t cb);
1544 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1545 const struct req_msg_field *, struct obd_capa **);
1547 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1548 struct obd_capa *, __u32,
1549 struct ptlrpc_request **);
1551 int (*m_intent_getattr_async)(struct obd_export *,
1552 struct md_enqueue_info *,
1553 struct ldlm_enqueue_info *);
1555 int (*m_revalidate_lock)(struct obd_export *,
1556 struct lookup_intent *,
1560 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1561 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1562 * wrapper function in include/linux/obd_class.h.
1566 struct lsm_operations {
1567 void (*lsm_free)(struct lov_stripe_md *);
1568 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1569 struct obd_export *md_exp);
1570 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1572 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1574 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1576 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1577 struct lov_mds_md *lmm);
1580 extern const struct lsm_operations lsm_v1_ops;
1581 extern const struct lsm_operations lsm_v3_ops;
1582 static inline const struct lsm_operations *lsm_op_find(int magic)
1590 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1595 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1597 /* Requests for obd_extent_calc() */
1598 #define OBD_CALC_STRIPE_START 1
1599 #define OBD_CALC_STRIPE_END 2
1601 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1602 struct obd_export *exp, int error)
1605 CERROR("%s: transno "LPU64" commit error: %d\n",
1606 obd->obd_name, transno, error);
1609 if (exp && transno > exp->exp_last_committed) {
1610 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1611 obd->obd_name, transno);
1612 exp->exp_last_committed = transno;
1613 ptlrpc_commit_replies(exp);
1615 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1616 obd->obd_name, transno);
1618 if (transno > obd->obd_last_committed)
1619 obd->obd_last_committed = transno;
1622 static inline void init_obd_quota_ops(quota_interface_t *interface,
1623 struct obd_ops *obd_ops)
1629 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1630 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1631 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1634 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1636 return oinfo->oi_capa;
1639 static inline struct md_open_data *obd_mod_alloc(void)
1641 struct md_open_data *mod;
1645 cfs_atomic_set(&mod->mod_refcount, 1);
1649 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1650 #define obd_mod_put(mod) \
1652 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1653 if ((mod)->mod_open_req) \
1654 ptlrpc_req_finished((mod)->mod_open_req); \
1655 OBD_FREE_PTR(mod); \
1659 #endif /* __OBD_H */