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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
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 #define IOC_MDC_MAX_NR 50
59 #include <lustre/lustre_idl.h>
60 #ifdef HAVE_SERVER_SUPPORT
61 # include <lu_target.h>
64 #include <lustre_lib.h>
65 #include <lustre_export.h>
66 #include <lustre_fid.h>
67 #include <lustre_fld.h>
68 #include <lustre_capa.h>
70 #include <libcfs/bitmap.h>
73 #define MAX_OBD_DEVICES 8192
81 struct lov_oinfo { /* per-stripe data structure */
82 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
83 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
84 int loi_ost_gen; /* generation of this loi_ost_idx */
86 unsigned long loi_kms_valid:1;
87 __u64 loi_kms; /* known minimum size */
88 struct ost_lvb loi_lvb;
89 struct osc_async_rc loi_ar;
92 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
95 oinfo->loi_kms_valid = 1;
98 static inline void loi_init(struct lov_oinfo *loi)
102 struct lov_stripe_md {
103 cfs_atomic_t lsm_refc;
105 pid_t lsm_lock_owner; /* debugging */
107 /* maximum possible file size, might change as OSTs status changes,
108 * e.g. disconnected, deactivated */
111 /* Public members. */
112 struct ost_id lw_object_oi; /* lov object id/seq */
114 /* LOV-private members start here -- only for use in lov/. */
116 __u32 lw_stripe_size; /* size of the stripe */
117 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
118 __u16 lw_stripe_count; /* number of objects being striped over */
119 __u16 lw_layout_gen; /* generation of the layout */
120 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
123 struct lov_oinfo *lsm_oinfo[0];
126 #define lsm_oi lsm_wire.lw_object_oi
127 #define lsm_magic lsm_wire.lw_magic
128 #define lsm_layout_gen lsm_wire.lw_layout_gen
129 #define lsm_stripe_size lsm_wire.lw_stripe_size
130 #define lsm_pattern lsm_wire.lw_pattern
131 #define lsm_stripe_count lsm_wire.lw_stripe_count
132 #define lsm_pool_name lsm_wire.lw_pool_name
134 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
136 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
139 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
143 if (lsm_is_released(lsm))
150 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
152 /* obd info for a particular level (lov, osc). */
154 /* Lock policy. It keeps an extent which is specific for a particular
155 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
156 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
157 ldlm_policy_data_t oi_policy;
158 /* Flags used for set request specific flags:
159 - while lock handling, the flags obtained on the enqueue
160 request are set here.
161 - while stats, the flags used for control delay/resend.
162 - while setattr, the flags used for distinguish punch operation
165 /* Lock handle specific for every OSC lock. */
166 struct lustre_handle *oi_lockh;
167 /* lsm data specific for every OSC. */
168 struct lov_stripe_md *oi_md;
169 /* obdo data specific for every OSC, if needed at all. */
171 /* statfs data specific for every OSC, if needed at all. */
172 struct obd_statfs *oi_osfs;
173 /* An update callback which is called to update some data on upper
174 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
175 * request in osc level for enqueue requests. It is also possible to
176 * update some caller data from LOV layer if needed. */
177 obd_enqueue_update_f oi_cb_up;
178 /* oss capability, its type is obd_capa in client to avoid copy.
179 * in contrary its type is lustre_capa in OSS. */
181 /* transfer jobid from ost_sync() to filter_sync()... */
185 /* compare all relevant fields. */
186 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
187 struct lov_stripe_md *m2)
190 * ->lsm_wire contains padding, but it should be zeroed out during
193 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
196 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
197 struct lov_stripe_md *lsm)
199 if (lsm->lsm_magic != lum->lmm_magic)
201 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
202 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
204 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
205 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
207 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
208 (lsm->lsm_pattern != lum->lmm_pattern))
210 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
211 (strncmp(lsm->lsm_pool_name,
212 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
213 LOV_MAXPOOLNAME) != 0))
218 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
220 struct lov_user_md *lum)
222 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
225 *lmm_magic = lumv3->lmm_magic;
227 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
228 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
229 *lmm_magic = LOV_USER_MAGIC_V1;
230 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
231 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
233 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
234 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
236 lustre_swab_lov_user_md_v3(lumv3);
237 *lmm_magic = LOV_USER_MAGIC_V3;
238 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
240 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
241 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
247 void lov_stripe_lock(struct lov_stripe_md *md);
248 void lov_stripe_unlock(struct lov_stripe_md *md);
251 cfs_list_t typ_chain;
252 struct obd_ops *typ_dt_ops;
253 struct md_ops *typ_md_ops;
254 cfs_proc_dir_entry_t *typ_procroot;
257 struct lu_device_type *typ_lu;
258 spinlock_t obd_type_lock;
268 /* Individual type definitions */
270 struct ost_server_data;
272 struct osd_properties {
273 size_t osd_max_ea_size;
276 #define OBT_MAGIC 0xBDDECEAE
277 /* hold common fields for "target" device */
278 struct obd_device_target {
281 struct super_block *obt_sb;
282 /** last_rcvd file */
283 struct file *obt_rcvd_filp;
284 #ifdef HAVE_SERVER_SUPPORT
285 struct lu_target *obt_lut;
287 __u64 obt_mount_count;
288 struct rw_semaphore obt_rwsem;
289 struct vfsmount *obt_vfsmnt;
290 struct file *obt_health_check_filp;
291 struct osd_properties obt_osd_properties;
292 struct obd_job_stats obt_jobstats;
297 LLOG_CONFIG_ORIG_CTXT = 0,
298 LLOG_CONFIG_REPL_CTXT,
299 LLOG_MDS_OST_ORIG_CTXT,
300 LLOG_MDS_OST_REPL_CTXT,
307 LLOG_LOVEA_ORIG_CTXT,
308 LLOG_LOVEA_REPL_CTXT,
309 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
310 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
311 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
315 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
317 struct filter_subdirs {
318 struct dentry *dentry[FILTER_SUBDIR_COUNT];
328 /* NB this field MUST be first */
329 struct obd_device_target fo_obt;
330 const char *fo_fstype;
333 struct dentry *fo_dentry_O;
334 struct dentry **fo_dentry_O_groups;
335 struct filter_subdirs *fo_dentry_O_sub;
336 struct mutex fo_init_lock; /* group initialization lock*/
337 int fo_committed_group;
339 spinlock_t fo_objidlock; /* protect fo_lastobjid */
341 unsigned long fo_destroys_in_progress;
342 struct mutex fo_create_locks[FILTER_SUBDIR_COUNT];
344 cfs_list_t fo_export_list;
347 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
348 obd_size fo_tot_granted; /* all values in bytes */
349 obd_size fo_tot_pending;
350 int fo_tot_granted_clients;
352 obd_size fo_readcache_max_filesize;
353 spinlock_t fo_flags_lock;
354 unsigned int fo_read_cache:1, /**< enable read-only cache */
355 fo_writethrough_cache:1,/**< read cache writes */
356 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
357 fo_raid_degraded:1;/**< RAID device degraded */
359 struct obd_import *fo_mdc_imp;
360 struct obd_uuid fo_mdc_uuid;
361 struct lustre_handle fo_mdc_conn;
362 struct file **fo_last_objid_files;
363 __u64 *fo_last_objids; /* last created objid for groups,
364 * protected by fo_objidlock */
366 struct mutex fo_alloc_lock;
368 cfs_atomic_t fo_r_in_flight;
369 cfs_atomic_t fo_w_in_flight;
372 * per-filter pool of kiobuf's allocated by filter_common_setup() and
373 * torn down by filter_cleanup().
375 * This pool contains kiobuf used by
376 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
379 * Locking: protected by internal lock of cfs_hash, pool can be
380 * found from this hash table by t_id of ptlrpc_thread.
382 struct cfs_hash *fo_iobuf_hash;
384 struct brw_stats fo_filter_stats;
386 int fo_fmd_max_num; /* per exp filter_mod_data */
387 int fo_fmd_max_age; /* jiffies to fmd expiry */
388 unsigned long fo_syncjournal:1, /* sync journal on writes */
389 fo_sync_lock_cancel:2;/* sync on lock cancel */
393 rwlock_t fo_sptlrpc_lock;
394 struct sptlrpc_rule_set fo_sptlrpc_rset;
396 /* capability related */
397 unsigned int fo_fl_oss_capa;
398 cfs_list_t fo_capa_keys;
399 cfs_hlist_head_t *fo_capa_hash;
403 struct timeout_item {
404 enum timeout_event ti_event;
405 cfs_time_t ti_timeout;
408 cfs_list_t ti_obd_list;
412 #define OSC_MAX_RIF_DEFAULT 8
413 #define MDS_OSC_MAX_RIF_DEFAULT 50
414 #define OSC_MAX_RIF_MAX 256
415 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
416 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
417 #define OSC_DEFAULT_RESENDS 10
419 /* possible values for fo_sync_lock_cancel */
421 NEVER_SYNC_ON_CANCEL = 0,
422 BLOCKING_SYNC_ON_CANCEL = 1,
423 ALWAYS_SYNC_ON_CANCEL = 2,
424 NUM_SYNC_ON_CANCEL_STATES
427 #define MDC_MAX_RIF_DEFAULT 8
428 #define MDC_MAX_RIF_MAX 512
433 struct rw_semaphore cl_sem;
434 struct obd_uuid cl_target_uuid;
435 struct obd_import *cl_import; /* ptlrpc connection state */
437 /* max_mds_easize is purely a performance thing so we don't have to
438 * call obd_size_diskmd() all the time. */
439 int cl_default_mds_easize;
440 int cl_max_mds_easize;
441 int cl_max_mds_cookiesize;
443 enum lustre_sec_part cl_sp_me;
444 enum lustre_sec_part cl_sp_to;
445 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
447 /* the grant values are protected by loi_list_lock below */
448 long cl_dirty; /* all _dirty_ in bytes */
449 long cl_dirty_max; /* allowed w/o rpc */
450 long cl_dirty_transit; /* dirty synchronous */
451 long cl_avail_grant; /* bytes of credit for ost */
452 long cl_lost_grant; /* lost credits (trunc) */
454 /* since we allocate grant by blocks, we don't know how many grant will
455 * be used to add a page into cache. As a solution, we reserve maximum
456 * grant before trying to dirty a page and unreserve the rest.
457 * See osc_{reserve|unreserve}_grant for details. */
458 long cl_reserved_grant;
459 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
460 cfs_time_t cl_next_shrink_grant; /* jiffies */
461 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
462 int cl_grant_shrink_interval; /* seconds */
464 /* A chunk is an optimal size used by osc_extent to determine
465 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
468 int cl_extent_tax; /* extent overhead, by bytes */
470 /* keep track of objects that have lois that contain pages which
471 * have been queued for async brw. this lock also protects the
472 * lists of osc_client_pages that hang off of the loi */
474 * ->cl_loi_list_lock protects consistency of
475 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
476 * ->ap_completion() call-backs are executed under this lock. As we
477 * cannot guarantee that these call-backs never block on all platforms
478 * (as a matter of fact they do block on Mac OS X), type of
479 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
480 * and blocking mutex on Mac OS X. (Alternative is to make this lock
481 * blocking everywhere, but we don't want to slow down fast-path of
482 * our main platform.)
484 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
485 * with client_obd_list_{un,}lock() and
486 * client_obd_list_lock_{init,done}() functions.
488 * NB by Jinshan: though field names are still _loi_, but actually
489 * osc_object{}s are in the list.
491 client_obd_lock_t cl_loi_list_lock;
492 cfs_list_t cl_loi_ready_list;
493 cfs_list_t cl_loi_hp_ready_list;
494 cfs_list_t cl_loi_write_list;
495 cfs_list_t cl_loi_read_list;
498 /* just a sum of the loi/lop pending numbers to be exported by /proc */
499 cfs_atomic_t cl_pending_w_pages;
500 cfs_atomic_t cl_pending_r_pages;
501 __u32 cl_max_pages_per_rpc;
502 int cl_max_rpcs_in_flight;
503 struct obd_histogram cl_read_rpc_hist;
504 struct obd_histogram cl_write_rpc_hist;
505 struct obd_histogram cl_read_page_hist;
506 struct obd_histogram cl_write_page_hist;
507 struct obd_histogram cl_read_offset_hist;
508 struct obd_histogram cl_write_offset_hist;
510 /* lru for osc caching pages */
511 struct cl_client_cache *cl_cache;
512 cfs_list_t cl_lru_osc; /* member of cl_cache->ccc_lru */
513 cfs_atomic_t *cl_lru_left;
514 cfs_atomic_t cl_lru_busy;
515 cfs_atomic_t cl_lru_shrinkers;
516 cfs_atomic_t cl_lru_in_list;
517 cfs_list_t cl_lru_list; /* lru page list */
518 client_obd_lock_t cl_lru_list_lock; /* page list protector */
520 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
521 cfs_atomic_t cl_destroy_in_flight;
522 cfs_waitq_t cl_destroy_waitq;
524 struct mdc_rpc_lock *cl_rpc_lock;
525 struct mdc_rpc_lock *cl_close_lock;
528 struct semaphore cl_mgc_sem;
529 struct vfsmount *cl_mgc_vfsmnt;
530 struct dentry *cl_mgc_configs_dir;
531 cfs_atomic_t cl_mgc_refcount;
532 struct obd_export *cl_mgc_mgsexp;
534 /* checksumming for data sent over the network */
535 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
536 /* supported checksum types that are worked out at connect time */
537 __u32 cl_supp_cksum_types;
538 /* checksum algorithm to be used */
539 cksum_type_t cl_cksum_type;
541 /* also protected by the poorly named _loi_list_lock lock above */
542 struct osc_async_rc cl_ar;
544 /* used by quotacheck when the servers are older than 2.4 */
545 int cl_qchk_stat; /* quotacheck stat of the peer */
546 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
547 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
548 #warning "please consider removing quotacheck compatibility code"
551 /* sequence manager */
552 struct lu_client_seq *cl_seq;
554 cfs_atomic_t cl_resends; /* resend count */
556 /* ptlrpc work for writeback in ptlrpcd context */
557 void *cl_writeback_work;
558 /* hash tables for osc_quota_info */
559 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
561 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
571 struct obd_device_target eo_obt;
575 struct lustre_handle eo_nl_lock;
576 cfs_atomic_t eo_prep;
580 struct ptlrpc_service *ost_service;
581 struct ptlrpc_service *ost_create_service;
582 struct ptlrpc_service *ost_io_service;
583 struct ptlrpc_service *ost_seq_service;
584 struct mutex ost_health_mutex;
587 struct echo_client_obd {
588 struct obd_export *ec_exp; /* the local connection to osc/lov */
590 cfs_list_t ec_objects;
597 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
598 cfs_list_t lqo_oss_list; /* link to lov_qos */
599 __u64 lqo_bavail; /* total bytes avail on OSS */
600 __u64 lqo_penalty; /* current penalty */
601 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
602 time_t lqo_used; /* last used time, seconds */
603 __u32 lqo_ost_count; /* number of osts on this oss */
607 struct lov_qos_oss *ltq_oss; /* oss info */
608 __u64 ltq_penalty; /* current penalty */
609 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
610 __u64 ltq_weight; /* net weighting */
611 time_t ltq_used; /* last used time, seconds */
612 unsigned int ltq_usable:1; /* usable for striping */
615 /* Generic subset of OSTs */
617 __u32 *op_array; /* array of index of
619 unsigned int op_count; /* number of OSTs in the array */
620 unsigned int op_size; /* allocated size of lp_array */
621 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
624 /* Round-robin allocator data */
626 __u32 lqr_start_idx; /* start index of new inode */
627 __u32 lqr_offset_idx; /* aliasing for start_idx */
628 int lqr_start_count; /* reseed counter */
629 struct ost_pool lqr_pool; /* round-robin optimized list */
630 unsigned long lqr_dirty:1; /* recalc round-robin list */
633 /* allow statfs data caching for 1 second */
634 #define OBD_STATFS_CACHE_SECONDS 1
636 struct lov_statfs_data {
637 struct obd_info lsd_oi;
638 struct obd_statfs lsd_statfs;
640 /* Stripe placement optimization */
642 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
643 struct rw_semaphore lq_rw_sem;
644 __u32 lq_active_oss_count;
645 unsigned int lq_prio_free; /* priority for free space */
646 unsigned int lq_threshold_rr;/* priority for rr */
647 struct lov_qos_rr lq_rr; /* round robin qos data */
648 unsigned long lq_dirty:1, /* recalc qos data */
649 lq_same_space:1,/* the ost's all have approx.
650 the same space avail */
651 lq_reset:1, /* zero current penalties */
652 lq_statfs_in_progress:1; /* statfs op in
654 /* qos statfs data */
655 struct lov_statfs_data *lq_statfs_data;
656 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
657 * requests completion */
660 struct lov_tgt_desc {
662 struct obd_uuid ltd_uuid;
663 struct obd_device *ltd_obd;
664 struct obd_export *ltd_exp;
665 struct ltd_qos ltd_qos; /* qos info per target */
667 __u32 ltd_index; /* index in lov_obd->tgts */
668 unsigned long ltd_active:1,/* is this target up for requests */
669 ltd_activate:1,/* should target be activated */
670 ltd_reap:1; /* should this target be deleted */
674 #define pool_tgt_size(_p) _p->pool_obds.op_size
675 #define pool_tgt_count(_p) _p->pool_obds.op_count
676 #define pool_tgt_array(_p) _p->pool_obds.op_array
677 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
680 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
681 struct ost_pool pool_obds; /* pool members */
682 cfs_atomic_t pool_refcount; /* pool ref. counter */
683 struct lov_qos_rr pool_rr; /* round robin qos */
684 cfs_hlist_node_t pool_hash; /* access by poolname */
685 cfs_list_t pool_list; /* serial access */
686 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
687 struct obd_device *pool_lobd; /* obd of the lov/lod to which
688 * this pool belongs */
692 struct lov_desc desc;
693 struct lov_tgt_desc **lov_tgts; /* sparse array */
694 struct ost_pool lov_packed; /* all OSTs in a packed
696 struct mutex lov_lock;
697 struct obd_connect_data lov_ocd;
698 cfs_atomic_t lov_refcount;
699 __u32 lov_tgt_count; /* how many OBD's */
700 __u32 lov_active_tgt_count; /* how many active */
701 __u32 lov_death_row;/* tgts scheduled to be deleted */
702 __u32 lov_tgt_size; /* size of tgts array */
705 cfs_hash_t *lov_pools_hash_body; /* used for key access */
706 cfs_list_t lov_pool_list; /* used for sequential access */
707 cfs_proc_dir_entry_t *lov_pool_proc_entry;
708 enum lustre_sec_part lov_sp_me;
710 /* Cached LRU pages from upper layer */
713 struct rw_semaphore lov_notify_lock;
716 struct lmv_tgt_desc {
717 struct obd_uuid ltd_uuid;
718 struct obd_export *ltd_exp;
720 struct mutex ltd_fid_mutex;
721 unsigned long ltd_active:1; /* target up for requests */
724 enum placement_policy {
725 PLACEMENT_CHAR_POLICY = 0,
726 PLACEMENT_NID_POLICY = 1,
727 PLACEMENT_INVAL_POLICY = 2,
731 typedef enum placement_policy placement_policy_t;
735 struct lu_client_fld lmv_fld;
737 placement_policy_t lmv_placement;
738 struct lmv_desc desc;
739 struct obd_uuid cluuid;
740 struct obd_export *exp;
742 struct mutex init_mutex;
749 int tgts_size; /* size of tgts array */
750 struct lmv_tgt_desc **tgts;
752 struct obd_connect_data conn_data;
755 struct niobuf_local {
756 __u64 lnb_file_offset;
757 __u32 lnb_page_offset;
761 struct dentry *dentry;
766 #define LUSTRE_FLD_NAME "fld"
767 #define LUSTRE_SEQ_NAME "seq"
769 #define LUSTRE_MDD_NAME "mdd"
770 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
771 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
772 #define LUSTRE_VVP_NAME "vvp"
773 #define LUSTRE_LMV_NAME "lmv"
774 #define LUSTRE_SLP_NAME "slp"
775 #define LUSTRE_LOD_NAME "lod"
776 #define LUSTRE_OSP_NAME "osp"
777 #define LUSTRE_LWP_NAME "lwp"
779 /* obd device type names */
780 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
781 #define LUSTRE_MDS_NAME "mds"
782 #define LUSTRE_MDT_NAME "mdt"
783 #define LUSTRE_MDC_NAME "mdc"
784 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
785 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
786 #define LUSTRE_OSC_NAME "osc"
787 #define LUSTRE_LOV_NAME "lov"
788 #define LUSTRE_MGS_NAME "mgs"
789 #define LUSTRE_MGC_NAME "mgc"
791 #define LUSTRE_ECHO_NAME "obdecho"
792 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
793 #define LUSTRE_QMT_NAME "qmt"
795 /* Constant obd names (post-rename) */
796 #define LUSTRE_MDS_OBDNAME "MDS"
797 #define LUSTRE_OSS_OBDNAME "OSS"
798 #define LUSTRE_MGS_OBDNAME "MGS"
799 #define LUSTRE_MGC_OBDNAME "MGC"
801 static inline int is_osp_on_mdt(char *name)
805 ptr = strrchr(name, '-');
807 CERROR("%s is not a obdname\n", name);
811 /* 1.8 OSC/OSP name on MDT is fsname-OSTxxxx-osc */
812 if (strncmp(ptr + 1, "osc", 3) == 0)
815 if (strncmp(ptr + 1, "MDT", 3) != 0)
818 while (*(--ptr) != '-' && ptr != name);
823 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0 &&
824 strncmp(ptr + 1, LUSTRE_OSC_NAME, strlen(LUSTRE_OSC_NAME)) != 0)
830 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
831 #define N_LOCAL_TEMP_PAGE 0x10000000
833 struct obd_trans_info {
836 /* Only used on the server side for tracking acks. */
837 struct oti_req_ack_lock {
838 struct lustre_handle lock;
842 struct llog_cookie oti_onecookie;
843 struct llog_cookie *oti_logcookies;
845 /** synchronous write is needed */
846 unsigned long oti_sync_write:1;
848 /* initial thread handling transaction */
849 struct ptlrpc_thread * oti_thread;
852 __u64 oti_pre_version;
856 struct obd_uuid *oti_ost_uuid;
859 static inline void oti_init(struct obd_trans_info *oti,
860 struct ptlrpc_request *req)
864 memset(oti, 0, sizeof(*oti));
869 oti->oti_xid = req->rq_xid;
870 /** VBR: take versions from request */
871 if (req->rq_reqmsg != NULL &&
872 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
873 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
874 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
875 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
878 /** called from mds_create_objects */
879 if (req->rq_repmsg != NULL)
880 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
881 oti->oti_thread = req->rq_svc_thread;
882 if (req->rq_reqmsg != NULL)
883 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
886 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
891 if (num_cookies == 1)
892 oti->oti_logcookies = &oti->oti_onecookie;
894 OBD_ALLOC_LARGE(oti->oti_logcookies,
895 num_cookies * sizeof(oti->oti_onecookie));
897 oti->oti_numcookies = num_cookies;
900 static inline void oti_free_cookies(struct obd_trans_info *oti)
902 if (!oti || !oti->oti_logcookies)
905 if (oti->oti_logcookies == &oti->oti_onecookie)
906 LASSERT(oti->oti_numcookies == 1);
908 OBD_FREE_LARGE(oti->oti_logcookies,
909 oti->oti_numcookies*sizeof(oti->oti_onecookie));
910 oti->oti_logcookies = NULL;
911 oti->oti_numcookies = 0;
915 * Events signalled through obd_notify() upcall-chain.
917 enum obd_notify_event {
920 /* Device connect start */
922 /* Device activated */
924 /* Device deactivated */
926 /* Device disconnected */
928 /* Connect data for import were changed */
931 OBD_NOTIFY_SYNC_NONBLOCK,
933 /* Configuration event */
935 /* Administratively deactivate/activate event */
936 OBD_NOTIFY_DEACTIVATE,
940 /* bit-mask flags for config events */
942 CONFIG_LOG = 0x1, /* finished processing config log */
943 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
944 CONFIG_TARGET = 0x4 /* one target is added */
948 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
949 * and liblustre being main examples).
951 struct obd_notify_upcall {
952 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
953 enum obd_notify_event ev, void *owner, void *data);
954 /* Opaque datum supplied by upper layer listener */
958 struct target_recovery_data {
959 svc_handler_t trd_recovery_handler;
960 pid_t trd_processing_task;
961 struct completion trd_starting;
962 struct completion trd_finishing;
965 struct obd_llog_group {
967 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
968 cfs_waitq_t olg_waitq;
970 struct mutex olg_cat_processing;
973 /* corresponds to one of the obd's */
974 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
975 #define OBD_DEV_BY_DEVNAME 0xffffd0de
978 struct obd_type *obd_type;
981 /* common and UUID name of this device */
982 char obd_name[MAX_OBD_NAME];
983 struct obd_uuid obd_uuid;
985 struct lu_device *obd_lu_dev;
988 /* bitfield modification is protected by obd_dev_lock */
989 unsigned long obd_attached:1, /* finished attach */
990 obd_set_up:1, /* finished setup */
991 obd_recovering:1, /* there are recoverable clients */
992 obd_abort_recovery:1,/* recovery expired */
993 obd_version_recov:1, /* obd uses version checking */
994 obd_replayable:1, /* recovery is enabled; inform clients */
995 obd_no_transno:1, /* no committed-transno notification */
996 obd_no_recov:1, /* fail instead of retry messages */
997 obd_stopping:1, /* started cleanup */
998 obd_starting:1, /* started setup */
999 obd_force:1, /* cleanup with > 0 obd refcount */
1000 obd_fail:1, /* cleanup with failover */
1001 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1002 obd_no_conn:1, /* deny new connections */
1003 obd_inactive:1, /* device active/inactive
1004 * (for /proc/status only!!) */
1005 obd_no_ir:1, /* no imperative recovery. */
1006 obd_process_conf:1; /* device is processing mgs config */
1007 /* use separate field as it is set in interrupt to don't mess with
1008 * protection of other bits using _bh lock */
1009 unsigned long obd_recovery_expired:1;
1010 /* uuid-export hash body */
1011 cfs_hash_t *obd_uuid_hash;
1012 /* nid-export hash body */
1013 cfs_hash_t *obd_nid_hash;
1014 /* nid stats body */
1015 cfs_hash_t *obd_nid_stats_hash;
1016 cfs_list_t obd_nid_stats;
1017 cfs_atomic_t obd_refcount;
1018 cfs_waitq_t obd_refcount_waitq;
1019 cfs_list_t obd_exports;
1020 cfs_list_t obd_unlinked_exports;
1021 cfs_list_t obd_delayed_exports;
1022 int obd_num_exports;
1023 spinlock_t obd_nid_lock;
1024 struct ldlm_namespace *obd_namespace;
1025 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1026 /* a spinlock is OK for what we do now, may need a semaphore later */
1027 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
1028 struct mutex obd_dev_mutex;
1029 __u64 obd_last_committed;
1030 struct fsfilt_operations *obd_fsops;
1031 spinlock_t obd_osfs_lock;
1032 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1034 struct lvfs_run_ctxt obd_lvfs_ctxt;
1035 struct obd_llog_group obd_olg; /* default llog group */
1036 struct obd_device *obd_observer;
1037 struct rw_semaphore obd_observer_link_sem;
1038 struct obd_notify_upcall obd_upcall;
1039 struct obd_export *obd_self_export;
1040 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1041 cfs_list_t obd_exports_timed;
1042 time_t obd_eviction_timer; /* for ping evictor */
1044 int obd_max_recoverable_clients;
1045 cfs_atomic_t obd_connected_clients;
1046 int obd_stale_clients;
1047 int obd_delayed_clients;
1048 /* this lock protects all recovery list_heads, timer and
1049 * obd_next_recovery_transno value */
1050 spinlock_t obd_recovery_task_lock;
1051 __u64 obd_next_recovery_transno;
1052 int obd_replayed_requests;
1053 int obd_requests_queued_for_recovery;
1054 cfs_waitq_t obd_next_transno_waitq;
1055 /* protected by obd_recovery_task_lock */
1056 cfs_timer_t obd_recovery_timer;
1057 time_t obd_recovery_start; /* seconds */
1058 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1059 int obd_recovery_time_hard;
1060 int obd_recovery_timeout;
1061 int obd_recovery_ir_factor;
1063 /* new recovery stuff from CMD2 */
1064 struct target_recovery_data obd_recovery_data;
1065 int obd_replayed_locks;
1066 cfs_atomic_t obd_req_replay_clients;
1067 cfs_atomic_t obd_lock_replay_clients;
1068 /* all lists are protected by obd_recovery_task_lock */
1069 cfs_list_t obd_req_replay_queue;
1070 cfs_list_t obd_lock_replay_queue;
1071 cfs_list_t obd_final_req_queue;
1072 int obd_recovery_stage;
1075 struct obd_device_target obt;
1076 struct filter_obd filter;
1077 struct client_obd cli;
1079 struct echo_client_obd echo_client;
1080 struct echo_obd echo;
1084 /* Fields used by LProcFS */
1085 unsigned int obd_cntr_base;
1086 struct lprocfs_stats *obd_stats;
1088 unsigned int md_cntr_base;
1089 struct lprocfs_stats *md_stats;
1091 cfs_proc_dir_entry_t *obd_proc_entry;
1092 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1093 cfs_proc_dir_entry_t *obd_svc_procroot;
1094 struct lprocfs_stats *obd_svc_stats;
1095 cfs_atomic_t obd_evict_inprogress;
1096 cfs_waitq_t obd_evict_inprogress_waitq;
1097 cfs_list_t obd_evict_list; /* protected with pet_lock */
1100 * Ldlm pool part. Save last calculated SLV and Limit.
1102 rwlock_t obd_pool_lock;
1107 * A list of outstanding class_incref()'s against this obd. For
1110 struct lu_ref obd_reference;
1112 int obd_conn_inprogress;
1115 #define OBD_LLOG_FL_SENDNOW 0x0001
1116 #define OBD_LLOG_FL_EXIT 0x0002
1118 enum obd_cleanup_stage {
1119 /* Special case hack for MDS LOVs */
1121 /* can be directly mapped to .ldto_device_fini() */
1122 OBD_CLEANUP_EXPORTS,
1125 /* get/set_info keys */
1126 #define KEY_ASYNC "async"
1127 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1128 #define KEY_BLOCKSIZE "blocksize"
1129 #define KEY_CAPA_KEY "capa_key"
1130 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1131 #define KEY_FID2PATH "fid2path"
1132 #define KEY_CHECKSUM "checksum"
1133 #define KEY_CLEAR_FS "clear_fs"
1134 #define KEY_CONN_DATA "conn_data"
1135 #define KEY_EVICT_BY_NID "evict_by_nid"
1136 #define KEY_FIEMAP "fiemap"
1137 #define KEY_FLUSH_CTX "flush_ctx"
1138 #define KEY_GRANT_SHRINK "grant_shrink"
1139 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1140 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1141 #define KEY_INIT_RECOV "initial_recov"
1142 #define KEY_INTERMDS "inter_mds"
1143 #define KEY_LAST_ID "last_id"
1144 #define KEY_LAST_FID "last_fid"
1145 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1146 #define KEY_LOVDESC "lovdesc"
1147 #define KEY_LOV_IDX "lov_idx"
1148 #define KEY_MAX_EASIZE "max_easize"
1149 #define KEY_MDS_CONN "mds_conn"
1150 #define KEY_MGSSEC "mgssec"
1151 #define KEY_NEXT_ID "next_id"
1152 #define KEY_READ_ONLY "read-only"
1153 #define KEY_REGISTER_TARGET "register_target"
1154 #define KEY_SET_FS "set_fs"
1155 #define KEY_TGT_COUNT "tgt_count"
1156 /* KEY_SET_INFO in lustre_idl.h */
1157 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1158 #define KEY_CONNECT_FLAG "connect_flags"
1159 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1161 #define KEY_CACHE_SET "cache_set"
1162 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
1163 #define KEY_CHANGELOG_INDEX "changelog_index"
1167 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1168 #define IT_OPEN (1 << 0)
1169 #define IT_CREAT (1 << 1)
1170 #define IT_READDIR (1 << 2)
1171 #define IT_GETATTR (1 << 3)
1172 #define IT_LOOKUP (1 << 4)
1173 #define IT_UNLINK (1 << 5)
1174 #define IT_TRUNC (1 << 6)
1175 #define IT_GETXATTR (1 << 7)
1176 #define IT_EXEC (1 << 8)
1177 #define IT_PIN (1 << 9)
1178 #define IT_LAYOUT (1 << 10)
1179 #define IT_QUOTA_DQACQ (1 << 11)
1180 #define IT_QUOTA_CONN (1 << 12)
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 |
1191 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1196 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1197 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1198 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1199 struct lu_fid op_fid4; /* to the operation locks. */
1200 mdsno_t op_mds; /* what mds server open will go to */
1201 struct lustre_handle op_handle;
1202 obd_time op_mod_time;
1203 const char *op_name;
1206 struct lmv_stripe_md *op_mea1;
1207 struct lmv_stripe_md *op_mea2;
1208 __u32 op_suppgids[2];
1214 /* iattr fields and blocks. */
1215 struct iattr op_attr;
1217 unsigned int op_attr_flags;
1220 loff_t op_attr_blocks;
1222 /* Size-on-MDS epoch and flags. */
1227 struct obd_capa *op_capa1;
1228 struct obd_capa *op_capa2;
1230 /* Various operation flags. */
1233 /* Operation type */
1236 /* Used by readdir */
1239 /* Used by readdir */
1242 /* used to transfer info between the stacks of MD client
1243 * see enum op_cli_flags */
1248 CLI_SET_MEA = 1 << 0,
1249 CLI_RM_ENTRY = 1 << 1,
1252 struct md_enqueue_info;
1253 /* metadata stat-ahead */
1254 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1255 struct md_enqueue_info *minfo,
1258 struct md_enqueue_info {
1259 struct md_op_data mi_data;
1260 struct lookup_intent mi_it;
1261 struct lustre_handle mi_lockh;
1262 struct inode *mi_dir;
1263 md_enqueue_cb_t mi_cb;
1265 unsigned int mi_generation;
1269 cfs_module_t *o_owner;
1270 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1271 void *karg, void *uarg);
1272 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1273 __u32 keylen, void *key, __u32 *vallen, void *val,
1274 struct lov_stripe_md *lsm);
1275 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1276 __u32 keylen, void *key,
1277 __u32 vallen, void *val,
1278 struct ptlrpc_request_set *set);
1279 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1280 int (*o_detach)(struct obd_device *dev);
1281 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1282 int (*o_precleanup)(struct obd_device *dev,
1283 enum obd_cleanup_stage cleanup_stage);
1284 int (*o_cleanup)(struct obd_device *dev);
1285 int (*o_process_config)(struct obd_device *dev, obd_count len,
1287 int (*o_postrecov)(struct obd_device *dev);
1288 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1290 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1291 /* connect to the target device with given connection
1292 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1293 * granted by the target, which are guaranteed to be a subset of flags
1294 * asked for. If @ocd == NULL, use default parameters. */
1295 int (*o_connect)(const struct lu_env *env,
1296 struct obd_export **exp, struct obd_device *src,
1297 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1299 int (*o_reconnect)(const struct lu_env *env,
1300 struct obd_export *exp, struct obd_device *src,
1301 struct obd_uuid *cluuid,
1302 struct obd_connect_data *ocd,
1304 int (*o_disconnect)(struct obd_export *exp);
1306 /* Initialize/finalize fids infrastructure. */
1307 int (*o_fid_init)(struct obd_device *obd,
1308 struct obd_export *exp, enum lu_cli_type type);
1309 int (*o_fid_fini)(struct obd_device *obd);
1311 /* Allocate new fid according to passed @hint. */
1312 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1313 struct md_op_data *op_data);
1316 * Object with @fid is getting deleted, we may want to do something
1319 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1320 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1321 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1322 __u64 max_age, struct ptlrpc_request_set *set);
1323 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1324 struct lov_stripe_md *mem_src);
1325 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1326 struct lov_mds_md *disk_src, int disk_len);
1327 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1329 /* FIXME: add fid capability support for create & destroy! */
1330 int (*o_precreate)(struct obd_export *exp);
1331 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1332 struct obdo *oa, struct lov_stripe_md **ea,
1333 struct obd_trans_info *oti);
1334 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1335 struct lov_stripe_md **ea,
1336 struct obd_trans_info *oti);
1337 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1338 struct obdo *oa, struct lov_stripe_md *ea,
1339 struct obd_trans_info *oti, struct obd_export *md_exp,
1341 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1342 struct obd_info *oinfo, struct obd_trans_info *oti);
1343 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1344 struct obd_trans_info *oti,
1345 struct ptlrpc_request_set *rqset);
1346 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1347 struct obd_info *oinfo);
1348 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1349 struct ptlrpc_request_set *set);
1350 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1351 obd_count oa_bufs, struct brw_page *pgarr,
1352 struct obd_trans_info *oti);
1353 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1354 struct ost_lvb *lvb, int kms_only);
1355 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1356 obd_off size, int shrink);
1357 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1358 struct obd_info *oinfo, struct obd_trans_info *oti,
1359 struct ptlrpc_request_set *rqset);
1360 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1361 struct obd_info *oinfo, obd_size start, obd_size end,
1362 struct ptlrpc_request_set *set);
1363 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1364 struct lov_stripe_md *src, obd_size start,
1365 obd_size end, struct obd_trans_info *oti);
1366 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1367 struct lustre_handle *srconn, struct lov_stripe_md *src,
1368 obd_size start, obd_size end, struct obd_trans_info *);
1369 int (*o_iterate)(struct lustre_handle *conn,
1370 int (*)(obd_id, obd_seq, void *),
1371 obd_id *startid, obd_seq seq, void *data);
1372 int (*o_preprw)(const struct lu_env *env, int cmd,
1373 struct obd_export *exp, struct obdo *oa, int objcount,
1374 struct obd_ioobj *obj, struct niobuf_remote *remote,
1375 int *nr_pages, struct niobuf_local *local,
1376 struct obd_trans_info *oti, struct lustre_capa *capa);
1377 int (*o_commitrw)(const struct lu_env *env, int cmd,
1378 struct obd_export *exp, struct obdo *oa,
1379 int objcount, struct obd_ioobj *obj,
1380 struct niobuf_remote *remote, int pages,
1381 struct niobuf_local *local,
1382 struct obd_trans_info *oti, int rc);
1383 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1384 struct ldlm_enqueue_info *einfo,
1385 struct ptlrpc_request_set *rqset);
1386 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1387 ldlm_iterator_t it, void *data);
1388 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1389 ldlm_iterator_t it, void *data);
1390 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1391 __u32 mode, struct lustre_handle *);
1392 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1393 ldlm_cancel_flags_t flags, void *opaque);
1394 int (*o_init_export)(struct obd_export *exp);
1395 int (*o_destroy_export)(struct obd_export *exp);
1396 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1397 int cmd, obd_off *);
1399 /* llog related obd_methods */
1400 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1401 struct obd_device *disk_obd, int *idx);
1402 int (*o_llog_finish)(struct obd_device *obd, int count);
1403 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1405 /* metadata-only methods */
1406 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1407 struct obd_capa *, struct obd_client_handle *, int flag);
1408 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1410 int (*o_import_event)(struct obd_device *, struct obd_import *,
1411 enum obd_import_event);
1413 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1414 enum obd_notify_event ev, void *data);
1416 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1417 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1420 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1421 struct obd_quotactl *);
1422 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1423 struct obd_quotactl *);
1425 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1428 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1429 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1430 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1432 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1434 void (*o_getref)(struct obd_device *obd);
1435 void (*o_putref)(struct obd_device *obd);
1437 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1438 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1439 * Also, add a wrapper function in include/linux/obd_class.h. */
1443 LUSTRE_OPC_MKDIR = (1 << 0),
1444 LUSTRE_OPC_SYMLINK = (1 << 1),
1445 LUSTRE_OPC_MKNOD = (1 << 2),
1446 LUSTRE_OPC_CREATE = (1 << 3),
1447 LUSTRE_OPC_ANY = (1 << 4)
1450 /* lmv structures */
1451 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1452 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1453 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1455 #define MAX_HASH_SIZE_32 0x7fffffffUL
1456 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1457 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1460 struct mdt_body *body;
1461 struct lov_stripe_md *lsm;
1462 struct lmv_stripe_md *mea;
1463 #ifdef CONFIG_FS_POSIX_ACL
1464 struct posix_acl *posix_acl;
1466 struct mdt_remote_perm *remote_perm;
1467 struct obd_capa *mds_capa;
1468 struct obd_capa *oss_capa;
1471 struct md_open_data {
1472 struct obd_client_handle *mod_och;
1473 struct ptlrpc_request *mod_open_req;
1474 struct ptlrpc_request *mod_close_req;
1475 cfs_atomic_t mod_refcount;
1478 struct lookup_intent;
1481 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1482 struct obd_capa **);
1483 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1484 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1485 ldlm_iterator_t, void *);
1486 int (*m_close)(struct obd_export *, struct md_op_data *,
1487 struct md_open_data *, struct ptlrpc_request **);
1488 int (*m_create)(struct obd_export *, struct md_op_data *,
1489 const void *, int, int, __u32, __u32, cfs_cap_t,
1490 __u64, struct ptlrpc_request **);
1491 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1492 struct md_open_data *);
1493 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1494 struct lookup_intent *, struct md_op_data *,
1495 struct lustre_handle *, void *, int,
1496 struct ptlrpc_request **, __u64);
1497 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1498 struct ptlrpc_request **);
1499 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1500 struct ptlrpc_request **);
1501 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1502 void *, int, struct lookup_intent *, int,
1503 struct ptlrpc_request **,
1504 ldlm_blocking_callback, __u64);
1505 int (*m_link)(struct obd_export *, struct md_op_data *,
1506 struct ptlrpc_request **);
1507 int (*m_rename)(struct obd_export *, struct md_op_data *,
1508 const char *, int, const char *, int,
1509 struct ptlrpc_request **);
1510 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1511 const struct lu_fid *,
1512 struct ptlrpc_request **);
1513 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1514 int , void *, int, struct ptlrpc_request **,
1515 struct md_open_data **mod);
1516 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1517 struct obd_capa *, struct ptlrpc_request **);
1518 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1519 struct page **, struct ptlrpc_request **);
1521 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1522 struct ptlrpc_request **);
1524 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1525 struct obd_capa *, obd_valid, const char *,
1526 const char *, int, int, int, __u32,
1527 struct ptlrpc_request **);
1529 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1530 struct obd_capa *, obd_valid, const char *,
1531 const char *, int, int, int,
1532 struct ptlrpc_request **);
1534 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1536 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1537 struct obd_export *, struct obd_export *,
1538 struct lustre_md *);
1540 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1542 int (*m_set_open_replay_data)(struct obd_export *,
1543 struct obd_client_handle *,
1544 struct ptlrpc_request *);
1545 int (*m_clear_open_replay_data)(struct obd_export *,
1546 struct obd_client_handle *);
1547 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1549 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1550 const struct lu_fid *, ldlm_type_t,
1551 ldlm_policy_data_t *, ldlm_mode_t,
1552 struct lustre_handle *);
1554 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1555 ldlm_policy_data_t *, ldlm_mode_t,
1556 ldlm_cancel_flags_t flags, void *opaque);
1557 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1558 renew_capa_cb_t cb);
1559 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1560 const struct req_msg_field *, struct obd_capa **);
1562 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1563 struct obd_capa *, __u32,
1564 struct ptlrpc_request **);
1566 int (*m_intent_getattr_async)(struct obd_export *,
1567 struct md_enqueue_info *,
1568 struct ldlm_enqueue_info *);
1570 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1571 struct lu_fid *, __u64 *bits);
1574 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1575 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1576 * wrapper function in include/linux/obd_class.h.
1580 struct lsm_operations {
1581 void (*lsm_free)(struct lov_stripe_md *);
1582 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1583 struct obd_export *md_exp);
1584 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1586 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1588 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1589 __u16 *stripe_count);
1590 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1591 struct lov_mds_md *lmm);
1594 extern const struct lsm_operations lsm_v1_ops;
1595 extern const struct lsm_operations lsm_v3_ops;
1596 static inline const struct lsm_operations *lsm_op_find(int magic)
1604 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1609 /* Requests for obd_extent_calc() */
1610 #define OBD_CALC_STRIPE_START 1
1611 #define OBD_CALC_STRIPE_END 2
1613 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1615 return oinfo->oi_capa;
1618 static inline struct md_open_data *obd_mod_alloc(void)
1620 struct md_open_data *mod;
1624 cfs_atomic_set(&mod->mod_refcount, 1);
1628 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1629 #define obd_mod_put(mod) \
1631 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1632 if ((mod)->mod_open_req) \
1633 ptlrpc_req_finished((mod)->mod_open_req); \
1634 OBD_FREE_PTR(mod); \
1638 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1639 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1641 /* return 1 if client should be resend request */
1642 static inline int client_should_resend(int resend, struct client_obd *cli)
1644 return cfs_atomic_read(&cli->cl_resends) ?
1645 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1649 * Return device name for this device
1651 * XXX: lu_device is declared before obd_device, while a pointer pointing
1652 * back to obd_device in lu_device, so this helper function defines here
1653 * instead of in lu_object.h
1655 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1657 return lu_dev->ld_obd->obd_name;
1660 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1665 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1668 /* caller does not care of idx */
1672 /* volatile file, the MDT can be set from name */
1673 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1674 /* if no MDT is specified, use std way */
1675 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1677 /* test for no MDT idx case */
1678 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1679 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1683 /* we have an idx, read it */
1684 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1685 *idx = strtoul(start, &end, 0);
1687 * no digit, no trailing :, negative value
1689 if (((*idx == 0) && (end == start)) ||
1690 (*end != ':') || (*idx < 0))
1695 /* bad format of mdt idx, we cannot return an error
1696 * to caller so we use hash algo */
1697 CERROR("Bad volatile file name format: %s\n",
1698 name + LUSTRE_VOLATILE_HDR_LEN);
1702 static inline int cli_brw_size(struct obd_device *obd)
1704 LASSERT(obd != NULL);
1705 return obd->u.cli.cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
1708 #endif /* __OBD_H */