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, 2012, 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_fld.h>
67 #include <lustre_capa.h>
69 #include <libcfs/bitmap.h>
72 #define MAX_OBD_DEVICES 8192
80 struct lov_oinfo { /* per-stripe data structure */
81 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
82 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
83 int loi_ost_gen; /* generation of this loi_ost_idx */
85 unsigned long loi_kms_valid:1;
86 __u64 loi_kms; /* known minimum size */
87 struct ost_lvb loi_lvb;
88 struct osc_async_rc loi_ar;
90 #define loi_id loi_oi.oi_id
91 #define loi_seq loi_oi.oi_seq
93 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
96 oinfo->loi_kms_valid = 1;
99 static inline void loi_init(struct lov_oinfo *loi)
103 struct lov_stripe_md {
104 cfs_atomic_t lsm_refc;
106 pid_t lsm_lock_owner; /* debugging */
108 /* maximum possible file size, might change as OSTs status changes,
109 * e.g. disconnected, deactivated */
112 /* Public members. */
113 __u64 lw_object_id; /* lov object id */
114 __u64 lw_object_seq; /* lov object seq */
116 /* LOV-private members start here -- only for use in lov/. */
118 __u32 lw_stripe_size; /* size of the stripe */
119 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
120 __u16 lw_stripe_count; /* number of objects being striped over */
121 __u16 lw_layout_gen; /* generation of the layout */
122 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
125 struct lov_oinfo *lsm_oinfo[0];
128 #define lsm_object_id lsm_wire.lw_object_id
129 #define lsm_object_seq lsm_wire.lw_object_seq
130 #define lsm_magic lsm_wire.lw_magic
131 #define lsm_layout_gen lsm_wire.lw_layout_gen
132 #define lsm_stripe_size lsm_wire.lw_stripe_size
133 #define lsm_pattern lsm_wire.lw_pattern
134 #define lsm_stripe_count lsm_wire.lw_stripe_count
135 #define lsm_pool_name lsm_wire.lw_pool_name
139 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
141 /* obd info for a particular level (lov, osc). */
143 /* Lock policy. It keeps an extent which is specific for a particular
144 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
145 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
146 ldlm_policy_data_t oi_policy;
147 /* Flags used for set request specific flags:
148 - while lock handling, the flags obtained on the enqueue
149 request are set here.
150 - while stats, the flags used for control delay/resend.
151 - while setattr, the flags used for distinguish punch operation
154 /* Lock handle specific for every OSC lock. */
155 struct lustre_handle *oi_lockh;
156 /* lsm data specific for every OSC. */
157 struct lov_stripe_md *oi_md;
158 /* obdo data specific for every OSC, if needed at all. */
160 /* statfs data specific for every OSC, if needed at all. */
161 struct obd_statfs *oi_osfs;
162 /* An update callback which is called to update some data on upper
163 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
164 * request in osc level for enqueue requests. It is also possible to
165 * update some caller data from LOV layer if needed. */
166 obd_enqueue_update_f oi_cb_up;
167 /* oss capability, its type is obd_capa in client to avoid copy.
168 * in contrary its type is lustre_capa in OSS. */
170 /* transfer jobid from ost_sync() to filter_sync()... */
174 /* compare all relevant fields. */
175 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
176 struct lov_stripe_md *m2)
179 * ->lsm_wire contains padding, but it should be zeroed out during
182 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
185 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
186 struct lov_stripe_md *lsm)
188 if (lsm->lsm_magic != lum->lmm_magic)
190 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
191 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
193 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
194 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
196 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
197 (lsm->lsm_pattern != lum->lmm_pattern))
199 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
200 (strncmp(lsm->lsm_pool_name,
201 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
202 LOV_MAXPOOLNAME) != 0))
207 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
209 struct lov_user_md *lum)
211 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
214 *lmm_magic = lumv3->lmm_magic;
216 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
217 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
218 *lmm_magic = LOV_USER_MAGIC_V1;
219 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
220 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
222 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
223 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
225 lustre_swab_lov_user_md_v3(lumv3);
226 *lmm_magic = LOV_USER_MAGIC_V3;
227 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
229 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
230 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
236 void lov_stripe_lock(struct lov_stripe_md *md);
237 void lov_stripe_unlock(struct lov_stripe_md *md);
240 cfs_list_t typ_chain;
241 struct obd_ops *typ_dt_ops;
242 struct md_ops *typ_md_ops;
243 cfs_proc_dir_entry_t *typ_procroot;
246 struct lu_device_type *typ_lu;
247 spinlock_t obd_type_lock;
257 /* Individual type definitions */
259 struct ost_server_data;
261 struct osd_properties {
262 size_t osd_max_ea_size;
265 #define OBT_MAGIC 0xBDDECEAE
266 /* hold common fields for "target" device */
267 struct obd_device_target {
270 struct super_block *obt_sb;
271 /** last_rcvd file */
272 struct file *obt_rcvd_filp;
273 #ifdef HAVE_SERVER_SUPPORT
274 struct lu_target *obt_lut;
276 __u64 obt_mount_count;
277 struct rw_semaphore obt_rwsem;
278 struct vfsmount *obt_vfsmnt;
279 struct file *obt_health_check_filp;
280 struct osd_properties obt_osd_properties;
281 struct obd_job_stats obt_jobstats;
286 LLOG_CONFIG_ORIG_CTXT = 0,
287 LLOG_CONFIG_REPL_CTXT,
288 LLOG_MDS_OST_ORIG_CTXT,
289 LLOG_MDS_OST_REPL_CTXT,
296 LLOG_LOVEA_ORIG_CTXT,
297 LLOG_LOVEA_REPL_CTXT,
298 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
299 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
300 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
304 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
306 struct filter_subdirs {
307 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
317 /* NB this field MUST be first */
318 struct obd_device_target fo_obt;
319 const char *fo_fstype;
322 cfs_dentry_t *fo_dentry_O;
323 cfs_dentry_t **fo_dentry_O_groups;
324 struct filter_subdirs *fo_dentry_O_sub;
325 struct mutex fo_init_lock; /* group initialization lock*/
326 int fo_committed_group;
328 spinlock_t fo_objidlock; /* protect fo_lastobjid */
330 unsigned long fo_destroys_in_progress;
331 struct mutex fo_create_locks[FILTER_SUBDIR_COUNT];
333 cfs_list_t fo_export_list;
336 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
337 obd_size fo_tot_granted; /* all values in bytes */
338 obd_size fo_tot_pending;
339 int fo_tot_granted_clients;
341 obd_size fo_readcache_max_filesize;
342 spinlock_t fo_flags_lock;
343 unsigned int fo_read_cache:1, /**< enable read-only cache */
344 fo_writethrough_cache:1,/**< read cache writes */
345 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
346 fo_raid_degraded:1;/**< RAID device degraded */
348 struct obd_import *fo_mdc_imp;
349 struct obd_uuid fo_mdc_uuid;
350 struct lustre_handle fo_mdc_conn;
351 struct file **fo_last_objid_files;
352 __u64 *fo_last_objids; /* last created objid for groups,
353 * protected by fo_objidlock */
355 struct mutex fo_alloc_lock;
357 cfs_atomic_t fo_r_in_flight;
358 cfs_atomic_t fo_w_in_flight;
361 * per-filter pool of kiobuf's allocated by filter_common_setup() and
362 * torn down by filter_cleanup().
364 * This pool contains kiobuf used by
365 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
368 * Locking: protected by internal lock of cfs_hash, pool can be
369 * found from this hash table by t_id of ptlrpc_thread.
371 struct cfs_hash *fo_iobuf_hash;
373 cfs_list_t fo_llog_list;
374 spinlock_t fo_llog_list_lock;
376 struct brw_stats fo_filter_stats;
378 int fo_fmd_max_num; /* per exp filter_mod_data */
379 int fo_fmd_max_age; /* jiffies to fmd expiry */
380 unsigned long fo_syncjournal:1, /* sync journal on writes */
381 fo_sync_lock_cancel:2;/* sync on lock cancel */
385 rwlock_t fo_sptlrpc_lock;
386 struct sptlrpc_rule_set fo_sptlrpc_rset;
388 /* capability related */
389 unsigned int fo_fl_oss_capa;
390 cfs_list_t fo_capa_keys;
391 cfs_hlist_head_t *fo_capa_hash;
392 struct llog_commit_master *fo_lcm;
396 struct timeout_item {
397 enum timeout_event ti_event;
398 cfs_time_t ti_timeout;
401 cfs_list_t ti_obd_list;
405 #define OSC_MAX_RIF_DEFAULT 8
406 #define MDS_OSC_MAX_RIF_DEFAULT 50
407 #define OSC_MAX_RIF_MAX 256
408 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
409 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
410 #define OSC_DEFAULT_RESENDS 10
412 /* possible values for fo_sync_lock_cancel */
414 NEVER_SYNC_ON_CANCEL = 0,
415 BLOCKING_SYNC_ON_CANCEL = 1,
416 ALWAYS_SYNC_ON_CANCEL = 2,
417 NUM_SYNC_ON_CANCEL_STATES
420 #define MDC_MAX_RIF_DEFAULT 8
421 #define MDC_MAX_RIF_MAX 512
426 struct rw_semaphore cl_sem;
427 struct obd_uuid cl_target_uuid;
428 struct obd_import *cl_import; /* ptlrpc connection state */
430 /* max_mds_easize is purely a performance thing so we don't have to
431 * call obd_size_diskmd() all the time. */
432 int cl_default_mds_easize;
433 int cl_max_mds_easize;
434 int cl_max_mds_cookiesize;
436 enum lustre_sec_part cl_sp_me;
437 enum lustre_sec_part cl_sp_to;
438 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
440 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
441 void *cl_llcd_offset;
443 /* the grant values are protected by loi_list_lock below */
444 long cl_dirty; /* all _dirty_ in bytes */
445 long cl_dirty_max; /* allowed w/o rpc */
446 long cl_dirty_transit; /* dirty synchronous */
447 long cl_avail_grant; /* bytes of credit for ost */
448 long cl_lost_grant; /* lost credits (trunc) */
450 /* since we allocate grant by blocks, we don't know how many grant will
451 * be used to add a page into cache. As a solution, we reserve maximum
452 * grant before trying to dirty a page and unreserve the rest.
453 * See osc_{reserve|unreserve}_grant for details. */
454 long cl_reserved_grant;
455 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
456 cfs_time_t cl_next_shrink_grant; /* jiffies */
457 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
458 int cl_grant_shrink_interval; /* seconds */
460 /* A chunk is an optimal size used by osc_extent to determine
461 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
464 int cl_extent_tax; /* extent overhead, by bytes */
466 /* keep track of objects that have lois that contain pages which
467 * have been queued for async brw. this lock also protects the
468 * lists of osc_client_pages that hang off of the loi */
470 * ->cl_loi_list_lock protects consistency of
471 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
472 * ->ap_completion() call-backs are executed under this lock. As we
473 * cannot guarantee that these call-backs never block on all platforms
474 * (as a matter of fact they do block on Mac OS X), type of
475 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
476 * and blocking mutex on Mac OS X. (Alternative is to make this lock
477 * blocking everywhere, but we don't want to slow down fast-path of
478 * our main platform.)
480 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
481 * with client_obd_list_{un,}lock() and
482 * client_obd_list_lock_{init,done}() functions.
484 * NB by Jinshan: though field names are still _loi_, but actually
485 * osc_object{}s are in the list.
487 client_obd_lock_t cl_loi_list_lock;
488 cfs_list_t cl_loi_ready_list;
489 cfs_list_t cl_loi_hp_ready_list;
490 cfs_list_t cl_loi_write_list;
491 cfs_list_t cl_loi_read_list;
494 /* just a sum of the loi/lop pending numbers to be exported by /proc */
495 cfs_atomic_t cl_pending_w_pages;
496 cfs_atomic_t cl_pending_r_pages;
497 int cl_max_pages_per_rpc;
498 int cl_max_rpcs_in_flight;
499 struct obd_histogram cl_read_rpc_hist;
500 struct obd_histogram cl_write_rpc_hist;
501 struct obd_histogram cl_read_page_hist;
502 struct obd_histogram cl_write_page_hist;
503 struct obd_histogram cl_read_offset_hist;
504 struct obd_histogram cl_write_offset_hist;
506 /* lru for osc caching pages */
507 struct cl_client_cache *cl_cache;
508 cfs_list_t cl_lru_osc; /* member of cl_cache->ccc_lru */
509 cfs_atomic_t *cl_lru_left;
510 cfs_atomic_t cl_lru_busy;
511 cfs_atomic_t cl_lru_shrinkers;
512 cfs_atomic_t cl_lru_in_list;
513 cfs_list_t cl_lru_list; /* lru page list */
514 client_obd_lock_t cl_lru_list_lock; /* page list protector */
516 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
517 cfs_atomic_t cl_destroy_in_flight;
518 cfs_waitq_t cl_destroy_waitq;
520 struct mdc_rpc_lock *cl_rpc_lock;
521 struct mdc_rpc_lock *cl_close_lock;
524 struct semaphore cl_mgc_sem;
525 struct vfsmount *cl_mgc_vfsmnt;
526 struct dentry *cl_mgc_configs_dir;
527 cfs_atomic_t cl_mgc_refcount;
528 struct obd_export *cl_mgc_mgsexp;
530 /* checksumming for data sent over the network */
531 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
532 /* supported checksum types that are worked out at connect time */
533 __u32 cl_supp_cksum_types;
534 /* checksum algorithm to be used */
535 cksum_type_t cl_cksum_type;
537 /* also protected by the poorly named _loi_list_lock lock above */
538 struct osc_async_rc cl_ar;
540 /* used by quotacheck when the servers are older than 2.4 */
541 int cl_qchk_stat; /* quotacheck stat of the peer */
542 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
543 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
544 #warning "please consider removing quotacheck compatibility code"
547 /* sequence manager */
548 struct lu_client_seq *cl_seq;
550 cfs_atomic_t cl_resends; /* resend count */
552 /* ptlrpc work for writeback in ptlrpcd context */
553 void *cl_writeback_work;
554 /* hash tables for osc_quota_info */
555 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
557 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
567 struct obd_device_target eo_obt;
571 struct lustre_handle eo_nl_lock;
572 cfs_atomic_t eo_prep;
576 struct ptlrpc_service *ost_service;
577 struct ptlrpc_service *ost_create_service;
578 struct ptlrpc_service *ost_io_service;
579 struct ptlrpc_service *ost_seq_service;
580 struct mutex ost_health_mutex;
583 struct echo_client_obd {
584 struct obd_export *ec_exp; /* the local connection to osc/lov */
586 cfs_list_t ec_objects;
593 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
594 cfs_list_t lqo_oss_list; /* link to lov_qos */
595 __u64 lqo_bavail; /* total bytes avail on OSS */
596 __u64 lqo_penalty; /* current penalty */
597 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
598 time_t lqo_used; /* last used time, seconds */
599 __u32 lqo_ost_count; /* number of osts on this oss */
603 struct lov_qos_oss *ltq_oss; /* oss info */
604 __u64 ltq_penalty; /* current penalty */
605 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
606 __u64 ltq_weight; /* net weighting */
607 time_t ltq_used; /* last used time, seconds */
608 unsigned int ltq_usable:1; /* usable for striping */
611 /* Generic subset of OSTs */
613 __u32 *op_array; /* array of index of
615 unsigned int op_count; /* number of OSTs in the array */
616 unsigned int op_size; /* allocated size of lp_array */
617 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
620 /* Round-robin allocator data */
622 __u32 lqr_start_idx; /* start index of new inode */
623 __u32 lqr_offset_idx; /* aliasing for start_idx */
624 int lqr_start_count; /* reseed counter */
625 struct ost_pool lqr_pool; /* round-robin optimized list */
626 unsigned long lqr_dirty:1; /* recalc round-robin list */
629 /* allow statfs data caching for 1 second */
630 #define OBD_STATFS_CACHE_SECONDS 1
632 struct lov_statfs_data {
633 struct obd_info lsd_oi;
634 struct obd_statfs lsd_statfs;
636 /* Stripe placement optimization */
638 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
639 struct rw_semaphore lq_rw_sem;
640 __u32 lq_active_oss_count;
641 unsigned int lq_prio_free; /* priority for free space */
642 unsigned int lq_threshold_rr;/* priority for rr */
643 struct lov_qos_rr lq_rr; /* round robin qos data */
644 unsigned long lq_dirty:1, /* recalc qos data */
645 lq_same_space:1,/* the ost's all have approx.
646 the same space avail */
647 lq_reset:1, /* zero current penalties */
648 lq_statfs_in_progress:1; /* statfs op in
650 /* qos statfs data */
651 struct lov_statfs_data *lq_statfs_data;
652 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
653 * requests completion */
656 struct lov_tgt_desc {
658 struct obd_uuid ltd_uuid;
659 struct obd_device *ltd_obd;
660 struct obd_export *ltd_exp;
661 struct ltd_qos ltd_qos; /* qos info per target */
663 __u32 ltd_index; /* index in lov_obd->tgts */
664 unsigned long ltd_active:1,/* is this target up for requests */
665 ltd_activate:1,/* should target be activated */
666 ltd_reap:1; /* should this target be deleted */
670 #define pool_tgt_size(_p) _p->pool_obds.op_size
671 #define pool_tgt_count(_p) _p->pool_obds.op_count
672 #define pool_tgt_array(_p) _p->pool_obds.op_array
673 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
676 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
677 struct ost_pool pool_obds; /* pool members */
678 cfs_atomic_t pool_refcount; /* pool ref. counter */
679 struct lov_qos_rr pool_rr; /* round robin qos */
680 cfs_hlist_node_t pool_hash; /* access by poolname */
681 cfs_list_t pool_list; /* serial access */
682 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
683 struct obd_device *pool_lobd; /* obd of the lov/lod to which
684 * this pool belongs */
688 struct lov_desc desc;
689 struct lov_tgt_desc **lov_tgts; /* sparse array */
690 struct ost_pool lov_packed; /* all OSTs in a packed
692 struct mutex lov_lock;
693 struct obd_connect_data lov_ocd;
694 cfs_atomic_t lov_refcount;
695 __u32 lov_tgt_count; /* how many OBD's */
696 __u32 lov_active_tgt_count; /* how many active */
697 __u32 lov_death_row;/* tgts scheduled to be deleted */
698 __u32 lov_tgt_size; /* size of tgts array */
701 cfs_hash_t *lov_pools_hash_body; /* used for key access */
702 cfs_list_t lov_pool_list; /* used for sequential access */
703 cfs_proc_dir_entry_t *lov_pool_proc_entry;
704 enum lustre_sec_part lov_sp_me;
706 /* Cached LRU pages from upper layer */
710 struct lmv_tgt_desc {
711 struct obd_uuid ltd_uuid;
712 struct obd_export *ltd_exp;
714 struct mutex ltd_fid_mutex;
715 unsigned long ltd_active:1; /* target up for requests */
718 enum placement_policy {
719 PLACEMENT_CHAR_POLICY = 0,
720 PLACEMENT_NID_POLICY = 1,
721 PLACEMENT_INVAL_POLICY = 2,
725 typedef enum placement_policy placement_policy_t;
729 struct lu_client_fld lmv_fld;
731 placement_policy_t lmv_placement;
732 struct lmv_desc desc;
733 struct obd_uuid cluuid;
734 struct obd_export *exp;
736 struct mutex init_mutex;
743 int tgts_size; /* size of tgts array */
744 struct lmv_tgt_desc **tgts;
746 struct obd_connect_data conn_data;
749 struct niobuf_local {
750 __u64 lnb_file_offset;
751 __u32 lnb_page_offset;
755 cfs_dentry_t *dentry;
760 #define LUSTRE_FLD_NAME "fld"
761 #define LUSTRE_SEQ_NAME "seq"
763 #define LUSTRE_CMM_NAME "cmm"
764 #define LUSTRE_MDD_NAME "mdd"
765 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
766 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
767 #define LUSTRE_VVP_NAME "vvp"
768 #define LUSTRE_LMV_NAME "lmv"
769 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
770 #define LUSTRE_SLP_NAME "slp"
771 #define LUSTRE_LOD_NAME "lod"
772 #define LUSTRE_OSP_NAME "osp"
774 /* obd device type names */
775 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
776 #define LUSTRE_MDS_NAME "mds"
777 #define LUSTRE_MDT_NAME "mdt"
778 #define LUSTRE_MDC_NAME "mdc"
779 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
780 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
781 #define LUSTRE_OSC_NAME "osc"
782 #define LUSTRE_LOV_NAME "lov"
783 #define LUSTRE_MGS_NAME "mgs"
784 #define LUSTRE_MGC_NAME "mgc"
786 #define LUSTRE_CACHEOBD_NAME "cobd"
787 #define LUSTRE_ECHO_NAME "obdecho"
788 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
789 #define LUSTRE_QMT_NAME "qmt"
791 /* Constant obd names (post-rename) */
792 #define LUSTRE_MDS_OBDNAME "MDS"
793 #define LUSTRE_OSS_OBDNAME "OSS"
794 #define LUSTRE_MGS_OBDNAME "MGS"
795 #define LUSTRE_MGC_OBDNAME "MGC"
797 static inline int is_osp_on_mdt(char *name)
801 ptr = strrchr(name, '-');
803 CERROR("%s is not a obdname\n", name);
807 /* 1.8 OSC/OSP name on MDT is fsname-OSTxxxx-osc */
808 if (strncmp(ptr + 1, "osc", 3) == 0)
811 if (strncmp(ptr + 1, "MDT", 3) != 0)
814 while (*(--ptr) != '-' && ptr != name);
819 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0 &&
820 strncmp(ptr + 1, LUSTRE_OSC_NAME, strlen(LUSTRE_OSC_NAME)) != 0)
826 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
827 #define N_LOCAL_TEMP_PAGE 0x10000000
829 /* Currently the connection osp is only for connecting MDT0, the
830 * name would either be
831 * fsname-MDT0000-osp-OSTxxxx or fsname-MDT0000-osp-MDT0000 */
832 static inline int is_osp_for_connection(char *name)
838 ptr = strrchr(name, '-');
840 CERROR("%s is not a obdname\n", name);
844 if (strncmp(ptr + 1, "OST", 3) != 0 && strncmp(ptr + 1, "MDT", 3) != 0)
847 if (strncmp(ptr + 1, "MDT", 3) == 0) {
851 index = simple_strtoul(ptr + 4, &endptr, 16);
853 index = strtoul(ptr + 4, &endptr, 16);
860 /* match the "-osp" */
861 if (ptr - name < strlen(LUSTRE_OSP_NAME) + 1)
864 ptr -= (strlen(LUSTRE_OSP_NAME) + 1);
868 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0)
873 while (*(--ptr) != '-' && ptr != name);
876 CERROR("%s is not a valid osp name\n", name);
880 if (strncmp(ptr + 1, "MDT", 3) != 0)
883 index = simple_strtoul(ptr + 4, &endptr, 16);
885 index = strtoul(ptr + 4, &endptr, 16);
893 struct obd_trans_info {
896 /* Only used on the server side for tracking acks. */
897 struct oti_req_ack_lock {
898 struct lustre_handle lock;
902 struct llog_cookie oti_onecookie;
903 struct llog_cookie *oti_logcookies;
905 /** synchronous write is needed */
906 long oti_sync_write:1;
908 /* initial thread handling transaction */
909 struct ptlrpc_thread * oti_thread;
912 __u64 oti_pre_version;
916 struct obd_uuid *oti_ost_uuid;
919 static inline void oti_init(struct obd_trans_info *oti,
920 struct ptlrpc_request *req)
924 memset(oti, 0, sizeof(*oti));
929 oti->oti_xid = req->rq_xid;
930 /** VBR: take versions from request */
931 if (req->rq_reqmsg != NULL &&
932 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
933 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
934 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
935 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
938 /** called from mds_create_objects */
939 if (req->rq_repmsg != NULL)
940 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
941 oti->oti_thread = req->rq_svc_thread;
942 if (req->rq_reqmsg != NULL)
943 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
946 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
951 if (num_cookies == 1)
952 oti->oti_logcookies = &oti->oti_onecookie;
954 OBD_ALLOC_LARGE(oti->oti_logcookies,
955 num_cookies * sizeof(oti->oti_onecookie));
957 oti->oti_numcookies = num_cookies;
960 static inline void oti_free_cookies(struct obd_trans_info *oti)
962 if (!oti || !oti->oti_logcookies)
965 if (oti->oti_logcookies == &oti->oti_onecookie)
966 LASSERT(oti->oti_numcookies == 1);
968 OBD_FREE_LARGE(oti->oti_logcookies,
969 oti->oti_numcookies*sizeof(oti->oti_onecookie));
970 oti->oti_logcookies = NULL;
971 oti->oti_numcookies = 0;
975 * Events signalled through obd_notify() upcall-chain.
977 enum obd_notify_event {
980 /* Device connect start */
982 /* Device activated */
984 /* Device deactivated */
986 /* Device disconnected */
988 /* Connect data for import were changed */
991 OBD_NOTIFY_SYNC_NONBLOCK,
993 /* Configuration event */
995 /* Administratively deactivate/activate event */
996 OBD_NOTIFY_DEACTIVATE,
1000 /* bit-mask flags for config events */
1002 CONFIG_LOG = 0x1, /* finished processing config log */
1003 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
1004 CONFIG_TARGET = 0x4 /* one target is added */
1008 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
1009 * and liblustre being main examples).
1011 struct obd_notify_upcall {
1012 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
1013 enum obd_notify_event ev, void *owner, void *data);
1014 /* Opaque datum supplied by upper layer listener */
1018 struct target_recovery_data {
1019 svc_handler_t trd_recovery_handler;
1020 pid_t trd_processing_task;
1021 struct completion trd_starting;
1022 struct completion trd_finishing;
1026 * In HEAD for CMD, the object is created in group number which is 3>=
1027 * or indexing starts from 3. To test this assertions are added to disallow
1028 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
1029 * object in group 0 needs to be allowed.
1030 * So for interop mode following changes needs to be done:
1031 * 1. No need to assert on group 0 or allow group 0
1032 * 2. The group number indexing starts from 0 instead of 3
1035 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
1037 static inline __u64 objseq_to_mdsno(obd_seq seq)
1039 LASSERT_SEQ_IS_MDT(seq);
1040 if (seq == FID_SEQ_OST_MDT0)
1042 return seq - FID_SEQ_OST_MDT1 + 1;
1045 static inline int mdt_to_obd_objseq(int mdtid)
1048 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
1052 return FID_SEQ_OST_MDT1 + mdtid - 1;
1056 struct obd_llog_group {
1057 cfs_list_t olg_list;
1059 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
1060 cfs_waitq_t olg_waitq;
1061 spinlock_t olg_lock;
1062 struct obd_export *olg_exp;
1063 int olg_initializing;
1064 struct mutex olg_cat_processing;
1067 /* corresponds to one of the obd's */
1068 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1069 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1072 struct obd_type *obd_type;
1075 /* common and UUID name of this device */
1076 char obd_name[MAX_OBD_NAME];
1077 struct obd_uuid obd_uuid;
1079 struct lu_device *obd_lu_dev;
1082 /* bitfield modification is protected by obd_dev_lock */
1083 unsigned long obd_attached:1, /* finished attach */
1084 obd_set_up:1, /* finished setup */
1085 obd_recovering:1, /* there are recoverable clients */
1086 obd_abort_recovery:1,/* recovery expired */
1087 obd_version_recov:1, /* obd uses version checking */
1088 obd_replayable:1, /* recovery is enabled; inform clients */
1089 obd_no_transno:1, /* no committed-transno notification */
1090 obd_no_recov:1, /* fail instead of retry messages */
1091 obd_stopping:1, /* started cleanup */
1092 obd_starting:1, /* started setup */
1093 obd_force:1, /* cleanup with > 0 obd refcount */
1094 obd_fail:1, /* cleanup with failover */
1095 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1096 obd_no_conn:1, /* deny new connections */
1097 obd_inactive:1, /* device active/inactive
1098 * (for /proc/status only!!) */
1099 obd_no_ir:1, /* no imperative recovery. */
1100 obd_process_conf:1; /* device is processing mgs config */
1101 /* use separate field as it is set in interrupt to don't mess with
1102 * protection of other bits using _bh lock */
1103 unsigned long obd_recovery_expired:1;
1104 /* uuid-export hash body */
1105 cfs_hash_t *obd_uuid_hash;
1106 /* nid-export hash body */
1107 cfs_hash_t *obd_nid_hash;
1108 /* nid stats body */
1109 cfs_hash_t *obd_nid_stats_hash;
1110 cfs_list_t obd_nid_stats;
1111 cfs_atomic_t obd_refcount;
1112 cfs_waitq_t obd_refcount_waitq;
1113 cfs_list_t obd_exports;
1114 cfs_list_t obd_unlinked_exports;
1115 cfs_list_t obd_delayed_exports;
1116 int obd_num_exports;
1117 spinlock_t obd_nid_lock;
1118 struct ldlm_namespace *obd_namespace;
1119 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1120 /* a spinlock is OK for what we do now, may need a semaphore later */
1121 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
1122 struct mutex obd_dev_mutex;
1123 __u64 obd_last_committed;
1124 struct fsfilt_operations *obd_fsops;
1125 spinlock_t obd_osfs_lock;
1126 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1128 struct lvfs_run_ctxt obd_lvfs_ctxt;
1129 struct obd_llog_group obd_olg; /* default llog group */
1130 struct obd_device *obd_observer;
1131 struct rw_semaphore obd_observer_link_sem;
1132 struct obd_notify_upcall obd_upcall;
1133 struct obd_export *obd_self_export;
1134 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1135 cfs_list_t obd_exports_timed;
1136 time_t obd_eviction_timer; /* for ping evictor */
1138 int obd_max_recoverable_clients;
1139 cfs_atomic_t obd_connected_clients;
1140 int obd_stale_clients;
1141 int obd_delayed_clients;
1142 /* this lock protects all recovery list_heads, timer and
1143 * obd_next_recovery_transno value */
1144 spinlock_t obd_recovery_task_lock;
1145 __u64 obd_next_recovery_transno;
1146 int obd_replayed_requests;
1147 int obd_requests_queued_for_recovery;
1148 cfs_waitq_t obd_next_transno_waitq;
1149 /* protected by obd_recovery_task_lock */
1150 cfs_timer_t obd_recovery_timer;
1151 time_t obd_recovery_start; /* seconds */
1152 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1153 int obd_recovery_time_hard;
1154 int obd_recovery_timeout;
1155 int obd_recovery_ir_factor;
1157 /* new recovery stuff from CMD2 */
1158 struct target_recovery_data obd_recovery_data;
1159 int obd_replayed_locks;
1160 cfs_atomic_t obd_req_replay_clients;
1161 cfs_atomic_t obd_lock_replay_clients;
1162 /* all lists are protected by obd_recovery_task_lock */
1163 cfs_list_t obd_req_replay_queue;
1164 cfs_list_t obd_lock_replay_queue;
1165 cfs_list_t obd_final_req_queue;
1166 int obd_recovery_stage;
1169 struct obd_device_target obt;
1170 struct filter_obd filter;
1171 struct client_obd cli;
1173 struct echo_client_obd echo_client;
1174 struct echo_obd echo;
1178 /* Fields used by LProcFS */
1179 unsigned int obd_cntr_base;
1180 struct lprocfs_stats *obd_stats;
1182 unsigned int md_cntr_base;
1183 struct lprocfs_stats *md_stats;
1185 cfs_proc_dir_entry_t *obd_proc_entry;
1186 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1187 cfs_proc_dir_entry_t *obd_svc_procroot;
1188 struct lprocfs_stats *obd_svc_stats;
1189 cfs_atomic_t obd_evict_inprogress;
1190 cfs_waitq_t obd_evict_inprogress_waitq;
1191 cfs_list_t obd_evict_list; /* protected with pet_lock */
1194 * Ldlm pool part. Save last calculated SLV and Limit.
1196 rwlock_t obd_pool_lock;
1201 * A list of outstanding class_incref()'s against this obd. For
1204 struct lu_ref obd_reference;
1206 int obd_conn_inprogress;
1209 #define OBD_LLOG_FL_SENDNOW 0x0001
1210 #define OBD_LLOG_FL_EXIT 0x0002
1212 enum obd_cleanup_stage {
1213 /* Special case hack for MDS LOVs */
1215 /* can be directly mapped to .ldto_device_fini() */
1216 OBD_CLEANUP_EXPORTS,
1219 /* get/set_info keys */
1220 #define KEY_ASYNC "async"
1221 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1222 #define KEY_BLOCKSIZE "blocksize"
1223 #define KEY_CAPA_KEY "capa_key"
1224 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1225 #define KEY_FID2PATH "fid2path"
1226 #define KEY_CHECKSUM "checksum"
1227 #define KEY_CLEAR_FS "clear_fs"
1228 #define KEY_CONN_DATA "conn_data"
1229 #define KEY_EVICT_BY_NID "evict_by_nid"
1230 #define KEY_FIEMAP "fiemap"
1231 #define KEY_FLUSH_CTX "flush_ctx"
1232 #define KEY_GRANT_SHRINK "grant_shrink"
1233 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1234 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1235 #define KEY_INIT_RECOV "initial_recov"
1236 #define KEY_INTERMDS "inter_mds"
1237 #define KEY_LAST_ID "last_id"
1238 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1239 #define KEY_LOVDESC "lovdesc"
1240 #define KEY_LOV_IDX "lov_idx"
1241 #define KEY_MAX_EASIZE "max_easize"
1242 #define KEY_MDS_CONN "mds_conn"
1243 #define KEY_MGSSEC "mgssec"
1244 #define KEY_NEXT_ID "next_id"
1245 #define KEY_READ_ONLY "read-only"
1246 #define KEY_REGISTER_TARGET "register_target"
1247 #define KEY_SET_FS "set_fs"
1248 #define KEY_TGT_COUNT "tgt_count"
1249 /* KEY_SET_INFO in lustre_idl.h */
1250 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1251 #define KEY_CONNECT_FLAG "connect_flags"
1252 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1254 #define KEY_CACHE_SET "cache_set"
1255 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
1259 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1260 #define IT_OPEN (1 << 0)
1261 #define IT_CREAT (1 << 1)
1262 #define IT_READDIR (1 << 2)
1263 #define IT_GETATTR (1 << 3)
1264 #define IT_LOOKUP (1 << 4)
1265 #define IT_UNLINK (1 << 5)
1266 #define IT_TRUNC (1 << 6)
1267 #define IT_GETXATTR (1 << 7)
1268 #define IT_EXEC (1 << 8)
1269 #define IT_PIN (1 << 9)
1270 #define IT_LAYOUT (1 << 10)
1271 #define IT_QUOTA_DQACQ (1 << 11)
1272 #define IT_QUOTA_CONN (1 << 12)
1274 static inline int it_to_lock_mode(struct lookup_intent *it)
1276 /* CREAT needs to be tested before open (both could be set) */
1277 if (it->it_op & IT_CREAT)
1279 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1283 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1288 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1289 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1290 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1291 struct lu_fid op_fid4; /* to the operation locks. */
1292 mdsno_t op_mds; /* what mds server open will go to */
1293 struct lustre_handle op_handle;
1294 obd_time op_mod_time;
1295 const char *op_name;
1298 struct lmv_stripe_md *op_mea1;
1299 struct lmv_stripe_md *op_mea2;
1300 __u32 op_suppgids[2];
1306 /* iattr fields and blocks. */
1307 struct iattr op_attr;
1309 unsigned int op_attr_flags;
1312 loff_t op_attr_blocks;
1314 /* Size-on-MDS epoch and flags. */
1319 struct obd_capa *op_capa1;
1320 struct obd_capa *op_capa2;
1322 /* Various operation flags. */
1325 /* Operation type */
1328 /* Used by readdir */
1333 struct md_enqueue_info;
1334 /* metadata stat-ahead */
1335 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1336 struct md_enqueue_info *minfo,
1339 /* seq client type */
1341 LUSTRE_SEQ_METADATA = 1,
1345 struct md_enqueue_info {
1346 struct md_op_data mi_data;
1347 struct lookup_intent mi_it;
1348 struct lustre_handle mi_lockh;
1349 struct inode *mi_dir;
1350 md_enqueue_cb_t mi_cb;
1352 unsigned int mi_generation;
1356 cfs_module_t *o_owner;
1357 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1358 void *karg, void *uarg);
1359 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1360 __u32 keylen, void *key, __u32 *vallen, void *val,
1361 struct lov_stripe_md *lsm);
1362 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1363 __u32 keylen, void *key,
1364 __u32 vallen, void *val,
1365 struct ptlrpc_request_set *set);
1366 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1367 int (*o_detach)(struct obd_device *dev);
1368 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1369 int (*o_precleanup)(struct obd_device *dev,
1370 enum obd_cleanup_stage cleanup_stage);
1371 int (*o_cleanup)(struct obd_device *dev);
1372 int (*o_process_config)(struct obd_device *dev, obd_count len,
1374 int (*o_postrecov)(struct obd_device *dev);
1375 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1377 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1378 /* connect to the target device with given connection
1379 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1380 * granted by the target, which are guaranteed to be a subset of flags
1381 * asked for. If @ocd == NULL, use default parameters. */
1382 int (*o_connect)(const struct lu_env *env,
1383 struct obd_export **exp, struct obd_device *src,
1384 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1386 int (*o_reconnect)(const struct lu_env *env,
1387 struct obd_export *exp, struct obd_device *src,
1388 struct obd_uuid *cluuid,
1389 struct obd_connect_data *ocd,
1391 int (*o_disconnect)(struct obd_export *exp);
1393 /* Initialize/finalize fids infrastructure. */
1394 int (*o_fid_init)(struct obd_device *obd,
1395 struct obd_export *exp, enum lu_cli_type type);
1396 int (*o_fid_fini)(struct obd_device *obd);
1398 /* Allocate new fid according to passed @hint. */
1399 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1400 struct md_op_data *op_data);
1403 * Object with @fid is getting deleted, we may want to do something
1406 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1407 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1408 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1409 __u64 max_age, struct ptlrpc_request_set *set);
1410 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1411 struct lov_stripe_md *mem_src);
1412 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1413 struct lov_mds_md *disk_src, int disk_len);
1414 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1416 /* FIXME: add fid capability support for create & destroy! */
1417 int (*o_precreate)(struct obd_export *exp);
1418 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1419 struct obdo *oa, struct lov_stripe_md **ea,
1420 struct obd_trans_info *oti);
1421 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1422 struct lov_stripe_md **ea,
1423 struct obd_trans_info *oti);
1424 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1425 struct obdo *oa, struct lov_stripe_md *ea,
1426 struct obd_trans_info *oti, struct obd_export *md_exp,
1428 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1429 struct obd_info *oinfo, struct obd_trans_info *oti);
1430 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1431 struct obd_trans_info *oti,
1432 struct ptlrpc_request_set *rqset);
1433 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1434 struct obd_info *oinfo);
1435 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1436 struct ptlrpc_request_set *set);
1437 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1438 obd_count oa_bufs, struct brw_page *pgarr,
1439 struct obd_trans_info *oti);
1440 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1441 struct ost_lvb *lvb, int kms_only);
1442 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1443 obd_off size, int shrink);
1444 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1445 struct obd_info *oinfo, struct obd_trans_info *oti,
1446 struct ptlrpc_request_set *rqset);
1447 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1448 struct obd_info *oinfo, obd_size start, obd_size end,
1449 struct ptlrpc_request_set *set);
1450 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1451 struct lov_stripe_md *src, obd_size start,
1452 obd_size end, struct obd_trans_info *oti);
1453 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1454 struct lustre_handle *srconn, struct lov_stripe_md *src,
1455 obd_size start, obd_size end, struct obd_trans_info *);
1456 int (*o_iterate)(struct lustre_handle *conn,
1457 int (*)(obd_id, obd_seq, void *),
1458 obd_id *startid, obd_seq seq, void *data);
1459 int (*o_preprw)(const struct lu_env *env, int cmd,
1460 struct obd_export *exp, struct obdo *oa, int objcount,
1461 struct obd_ioobj *obj, struct niobuf_remote *remote,
1462 int *nr_pages, struct niobuf_local *local,
1463 struct obd_trans_info *oti, struct lustre_capa *capa);
1464 int (*o_commitrw)(const struct lu_env *env, int cmd,
1465 struct obd_export *exp, struct obdo *oa,
1466 int objcount, struct obd_ioobj *obj,
1467 struct niobuf_remote *remote, int pages,
1468 struct niobuf_local *local,
1469 struct obd_trans_info *oti, int rc);
1470 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1471 struct ldlm_enqueue_info *einfo,
1472 struct ptlrpc_request_set *rqset);
1473 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1474 ldlm_iterator_t it, void *data);
1475 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1476 ldlm_iterator_t it, void *data);
1477 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1478 __u32 mode, struct lustre_handle *);
1479 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1480 ldlm_cancel_flags_t flags, void *opaque);
1481 int (*o_init_export)(struct obd_export *exp);
1482 int (*o_destroy_export)(struct obd_export *exp);
1483 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1484 int cmd, obd_off *);
1486 /* llog related obd_methods */
1487 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1488 struct obd_device *disk_obd, int *idx);
1489 int (*o_llog_finish)(struct obd_device *obd, int count);
1490 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1492 /* metadata-only methods */
1493 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1494 struct obd_capa *, struct obd_client_handle *, int flag);
1495 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1497 int (*o_import_event)(struct obd_device *, struct obd_import *,
1498 enum obd_import_event);
1500 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1501 enum obd_notify_event ev, void *data);
1503 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1504 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1507 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1508 struct obd_quotactl *);
1509 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1510 struct obd_quotactl *);
1512 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1515 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1516 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1517 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1519 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1521 void (*o_getref)(struct obd_device *obd);
1522 void (*o_putref)(struct obd_device *obd);
1524 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1525 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1526 * Also, add a wrapper function in include/linux/obd_class.h. */
1530 LUSTRE_OPC_MKDIR = (1 << 0),
1531 LUSTRE_OPC_SYMLINK = (1 << 1),
1532 LUSTRE_OPC_MKNOD = (1 << 2),
1533 LUSTRE_OPC_CREATE = (1 << 3),
1534 LUSTRE_OPC_ANY = (1 << 4)
1537 /* lmv structures */
1538 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1539 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1540 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1542 #define MAX_HASH_SIZE_32 0x7fffffffUL
1543 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1544 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1547 struct mdt_body *body;
1548 struct lov_stripe_md *lsm;
1549 struct lmv_stripe_md *mea;
1550 #ifdef CONFIG_FS_POSIX_ACL
1551 struct posix_acl *posix_acl;
1553 struct mdt_remote_perm *remote_perm;
1554 struct obd_capa *mds_capa;
1555 struct obd_capa *oss_capa;
1558 struct md_open_data {
1559 struct obd_client_handle *mod_och;
1560 struct ptlrpc_request *mod_open_req;
1561 struct ptlrpc_request *mod_close_req;
1562 cfs_atomic_t mod_refcount;
1565 struct lookup_intent;
1568 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1569 struct obd_capa **);
1570 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1571 ldlm_iterator_t, void *);
1572 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1573 ldlm_iterator_t, void *);
1574 int (*m_close)(struct obd_export *, struct md_op_data *,
1575 struct md_open_data *, struct ptlrpc_request **);
1576 int (*m_create)(struct obd_export *, struct md_op_data *,
1577 const void *, int, int, __u32, __u32, cfs_cap_t,
1578 __u64, struct ptlrpc_request **);
1579 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1580 struct md_open_data *);
1581 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1582 struct lookup_intent *, struct md_op_data *,
1583 struct lustre_handle *, void *, int,
1584 struct ptlrpc_request **, __u64);
1585 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1586 struct ptlrpc_request **);
1587 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1588 struct ptlrpc_request **);
1589 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1590 void *, int, struct lookup_intent *, int,
1591 struct ptlrpc_request **,
1592 ldlm_blocking_callback, __u64);
1593 int (*m_link)(struct obd_export *, struct md_op_data *,
1594 struct ptlrpc_request **);
1595 int (*m_rename)(struct obd_export *, struct md_op_data *,
1596 const char *, int, const char *, int,
1597 struct ptlrpc_request **);
1598 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1599 const struct lu_fid *,
1600 struct ptlrpc_request **);
1601 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1602 int , void *, int, struct ptlrpc_request **,
1603 struct md_open_data **mod);
1604 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1605 struct obd_capa *, struct ptlrpc_request **);
1606 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1607 struct page **, struct ptlrpc_request **);
1609 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1610 struct ptlrpc_request **);
1612 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1613 struct obd_capa *, obd_valid, const char *,
1614 const char *, int, int, int, __u32,
1615 struct ptlrpc_request **);
1617 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1618 struct obd_capa *, obd_valid, const char *,
1619 const char *, int, int, int,
1620 struct ptlrpc_request **);
1622 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1624 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1625 struct obd_export *, struct obd_export *,
1626 struct lustre_md *);
1628 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1630 int (*m_set_open_replay_data)(struct obd_export *,
1631 struct obd_client_handle *,
1632 struct ptlrpc_request *);
1633 int (*m_clear_open_replay_data)(struct obd_export *,
1634 struct obd_client_handle *);
1635 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1637 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1638 const struct lu_fid *, ldlm_type_t,
1639 ldlm_policy_data_t *, ldlm_mode_t,
1640 struct lustre_handle *);
1642 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1643 ldlm_policy_data_t *, ldlm_mode_t,
1644 ldlm_cancel_flags_t flags, void *opaque);
1645 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1646 renew_capa_cb_t cb);
1647 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1648 const struct req_msg_field *, struct obd_capa **);
1650 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1651 struct obd_capa *, __u32,
1652 struct ptlrpc_request **);
1654 int (*m_intent_getattr_async)(struct obd_export *,
1655 struct md_enqueue_info *,
1656 struct ldlm_enqueue_info *);
1658 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1659 struct lu_fid *, __u64 *bits);
1662 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1663 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1664 * wrapper function in include/linux/obd_class.h.
1668 struct lsm_operations {
1669 void (*lsm_free)(struct lov_stripe_md *);
1670 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1671 struct obd_export *md_exp);
1672 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1674 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1676 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1677 __u16 *stripe_count);
1678 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1679 struct lov_mds_md *lmm);
1682 extern const struct lsm_operations lsm_v1_ops;
1683 extern const struct lsm_operations lsm_v3_ops;
1684 static inline const struct lsm_operations *lsm_op_find(int magic)
1692 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1697 /* Requests for obd_extent_calc() */
1698 #define OBD_CALC_STRIPE_START 1
1699 #define OBD_CALC_STRIPE_END 2
1701 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1703 return oinfo->oi_capa;
1706 static inline struct md_open_data *obd_mod_alloc(void)
1708 struct md_open_data *mod;
1712 cfs_atomic_set(&mod->mod_refcount, 1);
1716 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1717 #define obd_mod_put(mod) \
1719 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1720 if ((mod)->mod_open_req) \
1721 ptlrpc_req_finished((mod)->mod_open_req); \
1722 OBD_FREE_PTR(mod); \
1726 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1727 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1729 /* return 1 if client should be resend request */
1730 static inline int client_should_resend(int resend, struct client_obd *cli)
1732 return cfs_atomic_read(&cli->cl_resends) ?
1733 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1737 * Return device name for this device
1739 * XXX: lu_device is declared before obd_device, while a pointer pointing
1740 * back to obd_device in lu_device, so this helper function defines here
1741 * instead of in lu_object.h
1743 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1745 return lu_dev->ld_obd->obd_name;
1748 #endif /* __OBD_H */