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 #include <linux/obd.h>
42 #define IOC_OSC_TYPE 'h'
43 #define IOC_OSC_MIN_NR 20
44 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
45 #define IOC_OSC_MAX_NR 50
47 #define IOC_MDC_TYPE 'i'
48 #define IOC_MDC_MIN_NR 20
49 #define IOC_MDC_MAX_NR 50
51 #include <lustre/lustre_idl.h>
53 #include <lustre_lib.h>
54 #include <lustre_export.h>
55 #include <lustre_fld.h>
56 #include <lustre_capa.h>
58 #include <linux/libcfs/bitmap.h>
61 #define MAX_OBD_DEVICES 8192
69 struct lov_oinfo { /* per-stripe data structure */
70 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
71 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
72 int loi_ost_gen; /* generation of this loi_ost_idx */
74 unsigned long loi_kms_valid:1;
75 __u64 loi_kms; /* known minimum size */
76 struct ost_lvb loi_lvb;
77 struct osc_async_rc loi_ar;
80 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
83 oinfo->loi_kms_valid = 1;
86 static inline void loi_init(struct lov_oinfo *loi)
90 struct lov_stripe_md {
93 pid_t lsm_lock_owner; /* debugging */
95 /* maximum possible file size, might change as OSTs status changes,
96 * e.g. disconnected, deactivated */
100 struct ost_id lw_object_oi; /* lov object id/seq */
102 /* LOV-private members start here -- only for use in lov/. */
104 __u32 lw_stripe_size; /* size of the stripe */
105 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
106 __u16 lw_stripe_count; /* number of objects being striped over */
107 __u16 lw_layout_gen; /* generation of the layout */
108 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
111 struct lov_oinfo *lsm_oinfo[0];
114 #define lsm_oi lsm_wire.lw_object_oi
115 #define lsm_magic lsm_wire.lw_magic
116 #define lsm_layout_gen lsm_wire.lw_layout_gen
117 #define lsm_stripe_size lsm_wire.lw_stripe_size
118 #define lsm_pattern lsm_wire.lw_pattern
119 #define lsm_stripe_count lsm_wire.lw_stripe_count
120 #define lsm_pool_name lsm_wire.lw_pool_name
124 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
126 /* obd info for a particular level (lov, osc). */
128 /* Lock policy. It keeps an extent which is specific for a particular
129 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
130 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
131 ldlm_policy_data_t oi_policy;
132 /* Flags used for set request specific flags:
133 - while lock handling, the flags obtained on the enqueue
134 request are set here.
135 - while stats, the flags used for control delay/resend.
136 - while setattr, the flags used for distinguish punch operation
139 /* Lock handle specific for every OSC lock. */
140 struct lustre_handle *oi_lockh;
141 /* lsm data specific for every OSC. */
142 struct lov_stripe_md *oi_md;
143 /* obdo data specific for every OSC, if needed at all. */
145 /* statfs data specific for every OSC, if needed at all. */
146 struct obd_statfs *oi_osfs;
147 /* An update callback which is called to update some data on upper
148 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
149 * request in osc level for enqueue requests. It is also possible to
150 * update some caller data from LOV layer if needed. */
151 obd_enqueue_update_f oi_cb_up;
152 /* oss capability, its type is obd_capa in client to avoid copy.
153 * in contrary its type is lustre_capa in OSS. */
155 /* transfer jobid from ost_sync() to filter_sync()... */
159 /* compare all relevant fields. */
160 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
161 struct lov_stripe_md *m2)
164 * ->lsm_wire contains padding, but it should be zeroed out during
167 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
170 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
171 struct lov_stripe_md *lsm)
173 if (lsm->lsm_magic != lum->lmm_magic)
175 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
176 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
178 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
179 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
181 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
182 (lsm->lsm_pattern != lum->lmm_pattern))
184 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
185 (strncmp(lsm->lsm_pool_name,
186 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
187 LOV_MAXPOOLNAME) != 0))
192 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
194 struct lov_user_md *lum)
196 if (lum && copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
199 *lmm_magic = lumv3->lmm_magic;
201 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
202 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
203 *lmm_magic = LOV_USER_MAGIC_V1;
204 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
205 if (lum && copy_from_user(lumv3, lum, sizeof(*lumv3)))
207 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
208 if (lum && copy_from_user(lumv3, lum, sizeof(*lumv3)))
210 lustre_swab_lov_user_md_v3(lumv3);
211 *lmm_magic = LOV_USER_MAGIC_V3;
212 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
214 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
215 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
221 void lov_stripe_lock(struct lov_stripe_md *md);
222 void lov_stripe_unlock(struct lov_stripe_md *md);
225 struct list_head typ_chain;
226 struct obd_ops *typ_dt_ops;
227 struct md_ops *typ_md_ops;
228 proc_dir_entry_t *typ_procroot;
231 struct lu_device_type *typ_lu;
232 spinlock_t obd_type_lock;
242 /* Individual type definitions */
244 struct ost_server_data;
246 struct osd_properties {
247 size_t osd_max_ea_size;
250 #define OBT_MAGIC 0xBDDECEAE
251 /* hold common fields for "target" device */
252 struct obd_device_target {
255 struct super_block *obt_sb;
256 /** last_rcvd file */
257 struct file *obt_rcvd_filp;
258 __u64 obt_mount_count;
259 struct rw_semaphore obt_rwsem;
260 struct vfsmount *obt_vfsmnt;
261 struct file *obt_health_check_filp;
262 struct osd_properties obt_osd_properties;
263 struct obd_job_stats obt_jobstats;
268 LLOG_CONFIG_ORIG_CTXT = 0,
269 LLOG_CONFIG_REPL_CTXT,
270 LLOG_MDS_OST_ORIG_CTXT,
271 LLOG_MDS_OST_REPL_CTXT,
278 LLOG_LOVEA_ORIG_CTXT,
279 LLOG_LOVEA_REPL_CTXT,
280 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
281 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
282 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
286 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
288 struct filter_subdirs {
289 struct dentry *dentry[FILTER_SUBDIR_COUNT];
299 /* NB this field MUST be first */
300 struct obd_device_target fo_obt;
301 const char *fo_fstype;
304 struct dentry *fo_dentry_O;
305 struct dentry **fo_dentry_O_groups;
306 struct filter_subdirs *fo_dentry_O_sub;
307 struct mutex fo_init_lock; /* group initialization lock*/
308 int fo_committed_group;
310 spinlock_t fo_objidlock; /* protect fo_lastobjid */
312 unsigned long fo_destroys_in_progress;
313 struct mutex fo_create_locks[FILTER_SUBDIR_COUNT];
315 struct list_head fo_export_list;
318 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
319 obd_size fo_tot_granted; /* all values in bytes */
320 obd_size fo_tot_pending;
321 int fo_tot_granted_clients;
323 obd_size fo_readcache_max_filesize;
324 spinlock_t fo_flags_lock;
325 unsigned int fo_read_cache:1, /**< enable read-only cache */
326 fo_writethrough_cache:1,/**< read cache writes */
327 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
328 fo_raid_degraded:1;/**< RAID device degraded */
330 struct obd_import *fo_mdc_imp;
331 struct obd_uuid fo_mdc_uuid;
332 struct lustre_handle fo_mdc_conn;
333 struct file **fo_last_objid_files;
334 __u64 *fo_last_objids; /* last created objid for groups,
335 * protected by fo_objidlock */
337 struct mutex fo_alloc_lock;
339 atomic_t fo_r_in_flight;
340 atomic_t fo_w_in_flight;
343 * per-filter pool of kiobuf's allocated by filter_common_setup() and
344 * torn down by filter_cleanup().
346 * This pool contains kiobuf used by
347 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
350 * Locking: protected by internal lock of cfs_hash, pool can be
351 * found from this hash table by t_id of ptlrpc_thread.
353 struct cfs_hash *fo_iobuf_hash;
355 struct brw_stats fo_filter_stats;
357 int fo_fmd_max_num; /* per exp filter_mod_data */
358 int fo_fmd_max_age; /* jiffies to fmd expiry */
359 unsigned long fo_syncjournal:1, /* sync journal on writes */
360 fo_sync_lock_cancel:2;/* sync on lock cancel */
364 rwlock_t fo_sptlrpc_lock;
365 struct sptlrpc_rule_set fo_sptlrpc_rset;
367 /* capability related */
368 unsigned int fo_fl_oss_capa;
369 struct list_head fo_capa_keys;
370 struct hlist_head *fo_capa_hash;
374 struct timeout_item {
375 enum timeout_event ti_event;
376 cfs_time_t ti_timeout;
379 struct list_head ti_obd_list;
380 struct list_head ti_chain;
383 #define OSC_MAX_RIF_DEFAULT 8
384 #define MDS_OSC_MAX_RIF_DEFAULT 50
385 #define OSC_MAX_RIF_MAX 256
386 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
387 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
388 #define OSC_DEFAULT_RESENDS 10
390 /* possible values for fo_sync_lock_cancel */
392 NEVER_SYNC_ON_CANCEL = 0,
393 BLOCKING_SYNC_ON_CANCEL = 1,
394 ALWAYS_SYNC_ON_CANCEL = 2,
395 NUM_SYNC_ON_CANCEL_STATES
398 #define MDC_MAX_RIF_DEFAULT 8
399 #define MDC_MAX_RIF_MAX 512
404 struct rw_semaphore cl_sem;
405 struct obd_uuid cl_target_uuid;
406 struct obd_import *cl_import; /* ptlrpc connection state */
408 /* max_mds_easize is purely a performance thing so we don't have to
409 * call obd_size_diskmd() all the time. */
410 int cl_default_mds_easize;
411 int cl_max_mds_easize;
412 int cl_max_mds_cookiesize;
414 enum lustre_sec_part cl_sp_me;
415 enum lustre_sec_part cl_sp_to;
416 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
418 /* the grant values are protected by loi_list_lock below */
419 long cl_dirty; /* all _dirty_ in bytes */
420 long cl_dirty_max; /* allowed w/o rpc */
421 long cl_dirty_transit; /* dirty synchronous */
422 long cl_avail_grant; /* bytes of credit for ost */
423 long cl_lost_grant; /* lost credits (trunc) */
425 /* since we allocate grant by blocks, we don't know how many grant will
426 * be used to add a page into cache. As a solution, we reserve maximum
427 * grant before trying to dirty a page and unreserve the rest.
428 * See osc_{reserve|unreserve}_grant for details. */
429 long cl_reserved_grant;
430 struct list_head cl_cache_waiters; /* waiting for cache/grant */
431 cfs_time_t cl_next_shrink_grant; /* jiffies */
432 struct list_head cl_grant_shrink_list; /* Timeout event list */
433 int cl_grant_shrink_interval; /* seconds */
435 /* A chunk is an optimal size used by osc_extent to determine
436 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
439 int cl_extent_tax; /* extent overhead, by bytes */
441 /* keep track of objects that have lois that contain pages which
442 * have been queued for async brw. this lock also protects the
443 * lists of osc_client_pages that hang off of the loi */
445 * ->cl_loi_list_lock protects consistency of
446 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
447 * ->ap_completion() call-backs are executed under this lock. As we
448 * cannot guarantee that these call-backs never block on all platforms
449 * (as a matter of fact they do block on Mac OS X), type of
450 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
451 * and blocking mutex on Mac OS X. (Alternative is to make this lock
452 * blocking everywhere, but we don't want to slow down fast-path of
453 * our main platform.)
455 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
456 * with client_obd_list_{un,}lock() and
457 * client_obd_list_lock_{init,done}() functions.
459 * NB by Jinshan: though field names are still _loi_, but actually
460 * osc_object{}s are in the list.
462 client_obd_lock_t cl_loi_list_lock;
463 struct list_head cl_loi_ready_list;
464 struct list_head cl_loi_hp_ready_list;
465 struct list_head cl_loi_write_list;
466 struct list_head cl_loi_read_list;
469 /* just a sum of the loi/lop pending numbers to be exported by /proc */
470 atomic_t cl_pending_w_pages;
471 atomic_t cl_pending_r_pages;
472 __u32 cl_max_pages_per_rpc;
473 int cl_max_rpcs_in_flight;
474 struct obd_histogram cl_read_rpc_hist;
475 struct obd_histogram cl_write_rpc_hist;
476 struct obd_histogram cl_read_page_hist;
477 struct obd_histogram cl_write_page_hist;
478 struct obd_histogram cl_read_offset_hist;
479 struct obd_histogram cl_write_offset_hist;
481 /* lru for osc caching pages */
482 struct cl_client_cache *cl_cache;
483 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
484 atomic_t *cl_lru_left;
485 atomic_t cl_lru_busy;
486 atomic_t cl_lru_shrinkers;
487 atomic_t cl_lru_in_list;
488 struct list_head cl_lru_list; /* lru page list */
489 client_obd_lock_t cl_lru_list_lock; /* page list protector */
491 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
492 atomic_t cl_destroy_in_flight;
493 wait_queue_head_t cl_destroy_waitq;
495 struct mdc_rpc_lock *cl_rpc_lock;
496 struct mdc_rpc_lock *cl_close_lock;
499 struct semaphore cl_mgc_sem;
500 struct vfsmount *cl_mgc_vfsmnt;
501 struct dentry *cl_mgc_configs_dir;
502 atomic_t cl_mgc_refcount;
503 struct obd_export *cl_mgc_mgsexp;
505 /* checksumming for data sent over the network */
506 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
507 /* supported checksum types that are worked out at connect time */
508 __u32 cl_supp_cksum_types;
509 /* checksum algorithm to be used */
510 cksum_type_t cl_cksum_type;
512 /* also protected by the poorly named _loi_list_lock lock above */
513 struct osc_async_rc cl_ar;
515 /* used by quotacheck when the servers are older than 2.4 */
516 int cl_qchk_stat; /* quotacheck stat of the peer */
517 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
518 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
519 #warning "please consider removing quotacheck compatibility code"
522 /* sequence manager */
523 struct lu_client_seq *cl_seq;
525 atomic_t cl_resends; /* resend count */
527 /* ptlrpc work for writeback in ptlrpcd context */
528 void *cl_writeback_work;
529 /* hash tables for osc_quota_info */
530 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
532 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
542 struct obd_device_target eo_obt;
546 struct lustre_handle eo_nl_lock;
551 struct ptlrpc_service *ost_service;
552 struct ptlrpc_service *ost_create_service;
553 struct ptlrpc_service *ost_io_service;
554 struct ptlrpc_service *ost_seq_service;
555 struct mutex ost_health_mutex;
558 struct echo_client_obd {
559 struct obd_export *ec_exp; /* the local connection to osc/lov */
561 struct list_head ec_objects;
562 struct list_head ec_locks;
568 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
569 struct list_head lqo_oss_list; /* link to lov_qos */
570 __u64 lqo_bavail; /* total bytes avail on OSS */
571 __u64 lqo_penalty; /* current penalty */
572 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
573 time_t lqo_used; /* last used time, seconds */
574 __u32 lqo_ost_count; /* number of osts on this oss */
578 struct lov_qos_oss *ltq_oss; /* oss info */
579 __u64 ltq_penalty; /* current penalty */
580 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
581 __u64 ltq_weight; /* net weighting */
582 time_t ltq_used; /* last used time, seconds */
583 unsigned int ltq_usable:1; /* usable for striping */
586 /* Generic subset of OSTs */
588 __u32 *op_array; /* array of index of
590 unsigned int op_count; /* number of OSTs in the array */
591 unsigned int op_size; /* allocated size of lp_array */
592 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
595 /* Round-robin allocator data */
597 __u32 lqr_start_idx; /* start index of new inode */
598 __u32 lqr_offset_idx; /* aliasing for start_idx */
599 int lqr_start_count; /* reseed counter */
600 struct ost_pool lqr_pool; /* round-robin optimized list */
601 unsigned long lqr_dirty:1; /* recalc round-robin list */
604 /* allow statfs data caching for 1 second */
605 #define OBD_STATFS_CACHE_SECONDS 1
607 struct lov_statfs_data {
608 struct obd_info lsd_oi;
609 struct obd_statfs lsd_statfs;
611 /* Stripe placement optimization */
613 struct list_head lq_oss_list; /* list of OSSs that targets use */
614 struct rw_semaphore lq_rw_sem;
615 __u32 lq_active_oss_count;
616 unsigned int lq_prio_free; /* priority for free space */
617 unsigned int lq_threshold_rr;/* priority for rr */
618 struct lov_qos_rr lq_rr; /* round robin qos data */
619 unsigned long lq_dirty:1, /* recalc qos data */
620 lq_same_space:1,/* the ost's all have approx.
621 the same space avail */
622 lq_reset:1, /* zero current penalties */
623 lq_statfs_in_progress:1; /* statfs op in
625 /* qos statfs data */
626 struct lov_statfs_data *lq_statfs_data;
627 wait_queue_head_t lq_statfs_waitq; /* waitqueue to notify statfs
628 * requests completion */
631 struct lov_tgt_desc {
632 struct list_head ltd_kill;
633 struct obd_uuid ltd_uuid;
634 struct obd_device *ltd_obd;
635 struct obd_export *ltd_exp;
636 struct ltd_qos ltd_qos; /* qos info per target */
638 __u32 ltd_index; /* index in lov_obd->tgts */
639 unsigned long ltd_active:1,/* is this target up for requests */
640 ltd_activate:1,/* should target be activated */
641 ltd_reap:1; /* should this target be deleted */
645 #define pool_tgt_size(_p) _p->pool_obds.op_size
646 #define pool_tgt_count(_p) _p->pool_obds.op_count
647 #define pool_tgt_array(_p) _p->pool_obds.op_array
648 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
651 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
652 struct ost_pool pool_obds; /* pool members */
653 atomic_t pool_refcount; /* pool ref. counter */
654 struct lov_qos_rr pool_rr; /* round robin qos */
655 struct hlist_node pool_hash; /* access by poolname */
656 struct list_head pool_list; /* serial access */
657 proc_dir_entry_t *pool_proc_entry; /* file in /proc */
658 struct obd_device *pool_lobd; /* obd of the lov/lod to which
659 * this pool belongs */
663 struct lov_desc desc;
664 struct lov_tgt_desc **lov_tgts; /* sparse array */
665 struct ost_pool lov_packed; /* all OSTs in a packed
667 struct mutex lov_lock;
668 struct obd_connect_data lov_ocd;
669 atomic_t lov_refcount;
670 __u32 lov_tgt_count; /* how many OBD's */
671 __u32 lov_active_tgt_count; /* how many active */
672 __u32 lov_death_row;/* tgts scheduled to be deleted */
673 __u32 lov_tgt_size; /* size of tgts array */
676 cfs_hash_t *lov_pools_hash_body; /* used for key access */
677 struct list_head lov_pool_list; /* used for sequential access */
678 proc_dir_entry_t *lov_pool_proc_entry;
679 enum lustre_sec_part lov_sp_me;
681 /* Cached LRU and unstable data from upper layer */
684 struct rw_semaphore lov_notify_lock;
687 struct lmv_tgt_desc {
688 struct obd_uuid ltd_uuid;
689 struct obd_export *ltd_exp;
691 struct mutex ltd_fid_mutex;
692 unsigned long ltd_active:1; /* target up for requests */
695 enum placement_policy {
696 PLACEMENT_CHAR_POLICY = 0,
697 PLACEMENT_NID_POLICY = 1,
698 PLACEMENT_INVAL_POLICY = 2,
702 typedef enum placement_policy placement_policy_t;
706 struct lu_client_fld lmv_fld;
708 placement_policy_t lmv_placement;
709 struct lmv_desc desc;
710 struct obd_uuid cluuid;
711 struct obd_export *exp;
713 struct mutex init_mutex;
720 int tgts_size; /* size of tgts array */
721 struct lmv_tgt_desc **tgts;
723 struct obd_connect_data conn_data;
726 struct niobuf_local {
727 __u64 lnb_file_offset;
728 __u32 lnb_page_offset;
732 struct dentry *dentry;
737 #define LUSTRE_FLD_NAME "fld"
738 #define LUSTRE_SEQ_NAME "seq"
740 #define LUSTRE_MDD_NAME "mdd"
741 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
742 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
743 #define LUSTRE_VVP_NAME "vvp"
744 #define LUSTRE_LMV_NAME "lmv"
745 #define LUSTRE_SLP_NAME "slp"
746 #define LUSTRE_LOD_NAME "lod"
747 #define LUSTRE_OSP_NAME "osp"
748 #define LUSTRE_LWP_NAME "lwp"
750 /* obd device type names */
751 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
752 #define LUSTRE_MDS_NAME "mds"
753 #define LUSTRE_MDT_NAME "mdt"
754 #define LUSTRE_MDC_NAME "mdc"
755 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
756 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
757 #define LUSTRE_OSC_NAME "osc"
758 #define LUSTRE_LOV_NAME "lov"
759 #define LUSTRE_MGS_NAME "mgs"
760 #define LUSTRE_MGC_NAME "mgc"
762 #define LUSTRE_ECHO_NAME "obdecho"
763 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
764 #define LUSTRE_QMT_NAME "qmt"
766 /* Constant obd names (post-rename) */
767 #define LUSTRE_MDS_OBDNAME "MDS"
768 #define LUSTRE_OSS_OBDNAME "OSS"
769 #define LUSTRE_MGS_OBDNAME "MGS"
770 #define LUSTRE_MGC_OBDNAME "MGC"
772 static inline int is_osp_on_mdt(char *name)
776 ptr = strrchr(name, '-');
778 CERROR("%s is not a obdname\n", name);
782 /* 1.8 OSC/OSP name on MDT is fsname-OSTxxxx-osc */
783 if (strncmp(ptr + 1, "osc", 3) == 0)
786 if (strncmp(ptr + 1, "MDT", 3) != 0)
789 while (*(--ptr) != '-' && ptr != name);
794 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0 &&
795 strncmp(ptr + 1, LUSTRE_OSC_NAME, strlen(LUSTRE_OSC_NAME)) != 0)
801 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
802 #define N_LOCAL_TEMP_PAGE 0x10000000
804 struct obd_trans_info {
807 /* Only used on the server side for tracking acks. */
808 struct oti_req_ack_lock {
809 struct lustre_handle lock;
813 struct llog_cookie oti_onecookie;
814 struct llog_cookie *oti_logcookies;
816 /** synchronous write is needed */
817 unsigned long oti_sync_write:1;
819 /* initial thread handling transaction */
820 struct ptlrpc_thread * oti_thread;
823 __u64 oti_pre_version;
827 struct obd_uuid *oti_ost_uuid;
830 static inline void oti_init(struct obd_trans_info *oti,
831 struct ptlrpc_request *req)
835 memset(oti, 0, sizeof(*oti));
840 oti->oti_xid = req->rq_xid;
841 /** VBR: take versions from request */
842 if (req->rq_reqmsg != NULL &&
843 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
844 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
845 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
846 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
849 /** called from mds_create_objects */
850 if (req->rq_repmsg != NULL)
851 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
852 oti->oti_thread = req->rq_svc_thread;
853 if (req->rq_reqmsg != NULL)
854 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
857 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
862 if (num_cookies == 1)
863 oti->oti_logcookies = &oti->oti_onecookie;
865 OBD_ALLOC_LARGE(oti->oti_logcookies,
866 num_cookies * sizeof(oti->oti_onecookie));
868 oti->oti_numcookies = num_cookies;
871 static inline void oti_free_cookies(struct obd_trans_info *oti)
873 if (!oti || !oti->oti_logcookies)
876 if (oti->oti_logcookies == &oti->oti_onecookie)
877 LASSERT(oti->oti_numcookies == 1);
879 OBD_FREE_LARGE(oti->oti_logcookies,
880 oti->oti_numcookies*sizeof(oti->oti_onecookie));
881 oti->oti_logcookies = NULL;
882 oti->oti_numcookies = 0;
886 * Events signalled through obd_notify() upcall-chain.
888 enum obd_notify_event {
891 /* Device connect start */
893 /* Device activated */
895 /* Device deactivated */
897 /* Device disconnected */
899 /* Connect data for import were changed */
902 OBD_NOTIFY_SYNC_NONBLOCK,
904 /* Configuration event */
906 /* Administratively deactivate/activate event */
907 OBD_NOTIFY_DEACTIVATE,
911 /* bit-mask flags for config events */
913 CONFIG_LOG = 0x1, /* finished processing config log */
914 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
915 CONFIG_TARGET = 0x4 /* one target is added */
919 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
920 * and liblustre being main examples).
922 struct obd_notify_upcall {
923 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
924 enum obd_notify_event ev, void *owner, void *data);
925 /* Opaque datum supplied by upper layer listener */
929 struct target_recovery_data {
930 svc_handler_t trd_recovery_handler;
931 pid_t trd_processing_task;
932 struct completion trd_starting;
933 struct completion trd_finishing;
936 struct obd_llog_group {
938 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
939 wait_queue_head_t olg_waitq;
941 struct mutex olg_cat_processing;
944 /* corresponds to one of the obd's */
945 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
946 #define OBD_DEV_BY_DEVNAME 0xffffd0de
949 struct obd_type *obd_type;
952 /* common and UUID name of this device */
953 char obd_name[MAX_OBD_NAME];
954 struct obd_uuid obd_uuid;
956 struct lu_device *obd_lu_dev;
959 /* bitfield modification is protected by obd_dev_lock */
960 unsigned long obd_attached:1, /* finished attach */
961 obd_set_up:1, /* finished setup */
962 obd_recovering:1, /* there are recoverable clients */
963 obd_abort_recovery:1,/* recovery expired */
964 obd_version_recov:1, /* obd uses version checking */
965 obd_replayable:1, /* recovery is enabled; inform clients */
966 obd_no_transno:1, /* no committed-transno notification */
967 obd_no_recov:1, /* fail instead of retry messages */
968 obd_stopping:1, /* started cleanup */
969 obd_starting:1, /* started setup */
970 obd_force:1, /* cleanup with > 0 obd refcount */
971 obd_fail:1, /* cleanup with failover */
972 obd_async_recov:1, /* allow asynchronous orphan cleanup */
973 obd_no_conn:1, /* deny new connections */
974 obd_inactive:1, /* device active/inactive
975 * (for /proc/status only!!) */
976 obd_no_ir:1, /* no imperative recovery. */
977 obd_process_conf:1; /* device is processing mgs config */
978 /* use separate field as it is set in interrupt to don't mess with
979 * protection of other bits using _bh lock */
980 unsigned long obd_recovery_expired:1;
981 /* uuid-export hash body */
982 cfs_hash_t *obd_uuid_hash;
983 /* nid-export hash body */
984 cfs_hash_t *obd_nid_hash;
986 cfs_hash_t *obd_nid_stats_hash;
987 struct list_head obd_nid_stats;
988 atomic_t obd_refcount;
989 wait_queue_head_t obd_refcount_waitq;
990 struct list_head obd_exports;
991 struct list_head obd_unlinked_exports;
992 struct list_head obd_delayed_exports;
994 spinlock_t obd_nid_lock;
995 struct ldlm_namespace *obd_namespace;
996 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
997 /* a spinlock is OK for what we do now, may need a semaphore later */
998 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
999 struct mutex obd_dev_mutex;
1000 __u64 obd_last_committed;
1001 struct fsfilt_operations *obd_fsops;
1002 spinlock_t obd_osfs_lock;
1003 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1005 struct lvfs_run_ctxt obd_lvfs_ctxt;
1006 struct obd_llog_group obd_olg; /* default llog group */
1007 struct obd_device *obd_observer;
1008 struct rw_semaphore obd_observer_link_sem;
1009 struct obd_notify_upcall obd_upcall;
1010 struct obd_export *obd_self_export;
1011 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1012 struct list_head obd_exports_timed;
1013 time_t obd_eviction_timer; /* for ping evictor */
1015 int obd_max_recoverable_clients;
1016 atomic_t obd_connected_clients;
1017 int obd_stale_clients;
1018 int obd_delayed_clients;
1019 /* this lock protects all recovery list_heads, timer and
1020 * obd_next_recovery_transno value */
1021 spinlock_t obd_recovery_task_lock;
1022 __u64 obd_next_recovery_transno;
1023 int obd_replayed_requests;
1024 int obd_requests_queued_for_recovery;
1025 wait_queue_head_t obd_next_transno_waitq;
1026 /* protected by obd_recovery_task_lock */
1027 timer_list_t obd_recovery_timer;
1028 time_t obd_recovery_start; /* seconds */
1029 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1030 int obd_recovery_time_hard;
1031 int obd_recovery_timeout;
1032 int obd_recovery_ir_factor;
1034 /* new recovery stuff from CMD2 */
1035 struct target_recovery_data obd_recovery_data;
1036 int obd_replayed_locks;
1037 atomic_t obd_req_replay_clients;
1038 atomic_t obd_lock_replay_clients;
1039 /* all lists are protected by obd_recovery_task_lock */
1040 struct list_head obd_req_replay_queue;
1041 struct list_head obd_lock_replay_queue;
1042 struct list_head obd_final_req_queue;
1043 int obd_recovery_stage;
1046 struct obd_device_target obt;
1047 struct filter_obd filter;
1048 struct client_obd cli;
1050 struct echo_client_obd echo_client;
1051 struct echo_obd echo;
1055 /* Fields used by LProcFS */
1056 unsigned int obd_cntr_base;
1057 struct lprocfs_stats *obd_stats;
1059 unsigned int md_cntr_base;
1060 struct lprocfs_stats *md_stats;
1062 proc_dir_entry_t *obd_proc_entry;
1063 proc_dir_entry_t *obd_proc_exports_entry;
1064 proc_dir_entry_t *obd_svc_procroot;
1065 struct lprocfs_stats *obd_svc_stats;
1066 atomic_t obd_evict_inprogress;
1067 wait_queue_head_t obd_evict_inprogress_waitq;
1068 struct list_head obd_evict_list; /* protected with pet_lock */
1071 * Ldlm pool part. Save last calculated SLV and Limit.
1073 rwlock_t obd_pool_lock;
1078 * A list of outstanding class_incref()'s against this obd. For
1081 struct lu_ref obd_reference;
1083 int obd_conn_inprogress;
1086 #define OBD_LLOG_FL_SENDNOW 0x0001
1087 #define OBD_LLOG_FL_EXIT 0x0002
1089 enum obd_cleanup_stage {
1090 /* Special case hack for MDS LOVs */
1092 /* can be directly mapped to .ldto_device_fini() */
1093 OBD_CLEANUP_EXPORTS,
1096 /* get/set_info keys */
1097 #define KEY_ASYNC "async"
1098 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1099 #define KEY_BLOCKSIZE "blocksize"
1100 #define KEY_CAPA_KEY "capa_key"
1101 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1102 #define KEY_FID2PATH "fid2path"
1103 #define KEY_CHECKSUM "checksum"
1104 #define KEY_CLEAR_FS "clear_fs"
1105 #define KEY_CONN_DATA "conn_data"
1106 #define KEY_EVICT_BY_NID "evict_by_nid"
1107 #define KEY_FIEMAP "fiemap"
1108 #define KEY_FLUSH_CTX "flush_ctx"
1109 #define KEY_GRANT_SHRINK "grant_shrink"
1110 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1111 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1112 #define KEY_INIT_RECOV "initial_recov"
1113 #define KEY_INTERMDS "inter_mds"
1114 #define KEY_LAST_ID "last_id"
1115 #define KEY_LAST_FID "last_fid"
1116 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1117 #define KEY_LOVDESC "lovdesc"
1118 #define KEY_LOV_IDX "lov_idx"
1119 #define KEY_MAX_EASIZE "max_easize"
1120 #define KEY_MDS_CONN "mds_conn"
1121 #define KEY_MGSSEC "mgssec"
1122 #define KEY_NEXT_ID "next_id"
1123 #define KEY_READ_ONLY "read-only"
1124 #define KEY_REGISTER_TARGET "register_target"
1125 #define KEY_SET_FS "set_fs"
1126 #define KEY_TGT_COUNT "tgt_count"
1127 /* KEY_SET_INFO in lustre_idl.h */
1128 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1129 #define KEY_CONNECT_FLAG "connect_flags"
1130 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1132 #define KEY_CACHE_SET "cache_set"
1133 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
1134 #define KEY_CHANGELOG_INDEX "changelog_index"
1138 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1139 #define IT_OPEN (1 << 0)
1140 #define IT_CREAT (1 << 1)
1141 #define IT_READDIR (1 << 2)
1142 #define IT_GETATTR (1 << 3)
1143 #define IT_LOOKUP (1 << 4)
1144 #define IT_UNLINK (1 << 5)
1145 #define IT_TRUNC (1 << 6)
1146 #define IT_GETXATTR (1 << 7)
1147 #define IT_EXEC (1 << 8)
1148 #define IT_PIN (1 << 9)
1149 #define IT_LAYOUT (1 << 10)
1150 #define IT_QUOTA_DQACQ (1 << 11)
1151 #define IT_QUOTA_CONN (1 << 12)
1153 static inline int it_to_lock_mode(struct lookup_intent *it)
1155 /* CREAT needs to be tested before open (both could be set) */
1156 if (it->it_op & IT_CREAT)
1158 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1162 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1167 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1168 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1169 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1170 struct lu_fid op_fid4; /* to the operation locks. */
1171 mdsno_t op_mds; /* what mds server open will go to */
1172 struct lustre_handle op_handle;
1173 obd_time op_mod_time;
1174 const char *op_name;
1177 struct lmv_stripe_md *op_mea1;
1178 struct lmv_stripe_md *op_mea2;
1179 __u32 op_suppgids[2];
1185 /* iattr fields and blocks. */
1186 struct iattr op_attr;
1187 unsigned int op_attr_flags;
1189 loff_t op_attr_blocks;
1191 /* Size-on-MDS epoch and flags. */
1196 struct obd_capa *op_capa1;
1197 struct obd_capa *op_capa2;
1199 /* Various operation flags. */
1202 /* Operation type */
1205 /* Used by readdir */
1208 /* Used by readdir */
1211 /* used to transfer info between the stacks of MD client
1212 * see enum op_cli_flags */
1217 CLI_SET_MEA = 1 << 0,
1218 CLI_RM_ENTRY = 1 << 1,
1221 struct md_enqueue_info;
1222 /* metadata stat-ahead */
1223 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1224 struct md_enqueue_info *minfo,
1227 /* seq client type */
1229 LUSTRE_SEQ_METADATA = 1,
1233 struct md_enqueue_info {
1234 struct md_op_data mi_data;
1235 struct lookup_intent mi_it;
1236 struct lustre_handle mi_lockh;
1237 struct inode *mi_dir;
1238 md_enqueue_cb_t mi_cb;
1240 unsigned int mi_generation;
1245 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1246 void *karg, void *uarg);
1247 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1248 __u32 keylen, void *key, __u32 *vallen, void *val,
1249 struct lov_stripe_md *lsm);
1250 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1251 __u32 keylen, void *key,
1252 __u32 vallen, void *val,
1253 struct ptlrpc_request_set *set);
1254 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1255 int (*o_detach)(struct obd_device *dev);
1256 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1257 int (*o_precleanup)(struct obd_device *dev,
1258 enum obd_cleanup_stage cleanup_stage);
1259 int (*o_cleanup)(struct obd_device *dev);
1260 int (*o_process_config)(struct obd_device *dev, obd_count len,
1262 int (*o_postrecov)(struct obd_device *dev);
1263 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1265 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1266 /* connect to the target device with given connection
1267 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1268 * granted by the target, which are guaranteed to be a subset of flags
1269 * asked for. If @ocd == NULL, use default parameters. */
1270 int (*o_connect)(const struct lu_env *env,
1271 struct obd_export **exp, struct obd_device *src,
1272 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1274 int (*o_reconnect)(const struct lu_env *env,
1275 struct obd_export *exp, struct obd_device *src,
1276 struct obd_uuid *cluuid,
1277 struct obd_connect_data *ocd,
1279 int (*o_disconnect)(struct obd_export *exp);
1281 /* Initialize/finalize fids infrastructure. */
1282 int (*o_fid_init)(struct obd_device *obd,
1283 struct obd_export *exp, enum lu_cli_type type);
1284 int (*o_fid_fini)(struct obd_device *obd);
1286 /* Allocate new fid according to passed @hint. */
1287 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1288 struct md_op_data *op_data);
1291 * Object with @fid is getting deleted, we may want to do something
1294 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1295 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1296 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1297 __u64 max_age, struct ptlrpc_request_set *set);
1298 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1299 struct lov_stripe_md *mem_src);
1300 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1301 struct lov_mds_md *disk_src, int disk_len);
1302 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1304 /* FIXME: add fid capability support for create & destroy! */
1305 int (*o_precreate)(struct obd_export *exp);
1306 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1307 struct obdo *oa, struct lov_stripe_md **ea,
1308 struct obd_trans_info *oti);
1309 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1310 struct lov_stripe_md **ea,
1311 struct obd_trans_info *oti);
1312 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1313 struct obdo *oa, struct lov_stripe_md *ea,
1314 struct obd_trans_info *oti, struct obd_export *md_exp,
1316 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1317 struct obd_info *oinfo, struct obd_trans_info *oti);
1318 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1319 struct obd_trans_info *oti,
1320 struct ptlrpc_request_set *rqset);
1321 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1322 struct obd_info *oinfo);
1323 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1324 struct ptlrpc_request_set *set);
1325 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1326 obd_count oa_bufs, struct brw_page *pgarr,
1327 struct obd_trans_info *oti);
1328 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1329 struct ost_lvb *lvb, int kms_only);
1330 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1331 obd_off size, int shrink);
1332 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1333 struct obd_info *oinfo, struct obd_trans_info *oti,
1334 struct ptlrpc_request_set *rqset);
1335 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1336 struct obd_info *oinfo, obd_size start, obd_size end,
1337 struct ptlrpc_request_set *set);
1338 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1339 struct lov_stripe_md *src, obd_size start,
1340 obd_size end, struct obd_trans_info *oti);
1341 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1342 struct lustre_handle *srconn, struct lov_stripe_md *src,
1343 obd_size start, obd_size end, struct obd_trans_info *);
1344 int (*o_iterate)(struct lustre_handle *conn,
1345 int (*)(obd_id, obd_seq, void *),
1346 obd_id *startid, obd_seq seq, void *data);
1347 int (*o_preprw)(const struct lu_env *env, int cmd,
1348 struct obd_export *exp, struct obdo *oa, int objcount,
1349 struct obd_ioobj *obj, struct niobuf_remote *remote,
1350 int *nr_pages, struct niobuf_local *local,
1351 struct obd_trans_info *oti, struct lustre_capa *capa);
1352 int (*o_commitrw)(const struct lu_env *env, int cmd,
1353 struct obd_export *exp, struct obdo *oa,
1354 int objcount, struct obd_ioobj *obj,
1355 struct niobuf_remote *remote, int pages,
1356 struct niobuf_local *local,
1357 struct obd_trans_info *oti, int rc);
1358 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1359 struct ldlm_enqueue_info *einfo,
1360 struct ptlrpc_request_set *rqset);
1361 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1362 ldlm_iterator_t it, void *data);
1363 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1364 ldlm_iterator_t it, void *data);
1365 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1366 __u32 mode, struct lustre_handle *);
1367 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1368 ldlm_cancel_flags_t flags, void *opaque);
1369 int (*o_init_export)(struct obd_export *exp);
1370 int (*o_destroy_export)(struct obd_export *exp);
1371 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1372 int cmd, obd_off *);
1374 /* llog related obd_methods */
1375 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1376 struct obd_device *disk_obd, int *idx);
1377 int (*o_llog_finish)(struct obd_device *obd, int count);
1378 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1380 /* metadata-only methods */
1381 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1382 struct obd_capa *, struct obd_client_handle *, int flag);
1383 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1385 int (*o_import_event)(struct obd_device *, struct obd_import *,
1386 enum obd_import_event);
1388 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1389 enum obd_notify_event ev, void *data);
1391 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1392 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1395 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1396 struct obd_quotactl *);
1397 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1398 struct obd_quotactl *);
1400 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1403 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1404 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1405 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1407 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1409 void (*o_getref)(struct obd_device *obd);
1410 void (*o_putref)(struct obd_device *obd);
1412 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1413 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1414 * Also, add a wrapper function in include/linux/obd_class.h. */
1418 LUSTRE_OPC_MKDIR = (1 << 0),
1419 LUSTRE_OPC_SYMLINK = (1 << 1),
1420 LUSTRE_OPC_MKNOD = (1 << 2),
1421 LUSTRE_OPC_CREATE = (1 << 3),
1422 LUSTRE_OPC_ANY = (1 << 4)
1425 /* lmv structures */
1426 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1427 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1428 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1430 #define MAX_HASH_SIZE_32 0x7fffffffUL
1431 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1432 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1435 struct mdt_body *body;
1436 struct lov_stripe_md *lsm;
1437 struct lmv_stripe_md *mea;
1438 #ifdef CONFIG_FS_POSIX_ACL
1439 struct posix_acl *posix_acl;
1441 struct mdt_remote_perm *remote_perm;
1442 struct obd_capa *mds_capa;
1443 struct obd_capa *oss_capa;
1446 struct md_open_data {
1447 struct obd_client_handle *mod_och;
1448 struct ptlrpc_request *mod_open_req;
1449 struct ptlrpc_request *mod_close_req;
1450 atomic_t mod_refcount;
1453 struct lookup_intent;
1456 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1457 struct obd_capa **);
1458 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1459 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1460 ldlm_iterator_t, void *);
1461 int (*m_close)(struct obd_export *, struct md_op_data *,
1462 struct md_open_data *, struct ptlrpc_request **);
1463 int (*m_create)(struct obd_export *, struct md_op_data *,
1464 const void *, int, int, __u32, __u32, cfs_cap_t,
1465 __u64, struct ptlrpc_request **);
1466 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1467 struct md_open_data *);
1468 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1469 struct lookup_intent *, struct md_op_data *,
1470 struct lustre_handle *, void *, int,
1471 struct ptlrpc_request **, __u64);
1472 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1473 struct ptlrpc_request **);
1474 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1475 struct ptlrpc_request **);
1476 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1477 void *, int, struct lookup_intent *, int,
1478 struct ptlrpc_request **,
1479 ldlm_blocking_callback, __u64);
1480 int (*m_link)(struct obd_export *, struct md_op_data *,
1481 struct ptlrpc_request **);
1482 int (*m_rename)(struct obd_export *, struct md_op_data *,
1483 const char *, int, const char *, int,
1484 struct ptlrpc_request **);
1485 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1486 const struct lu_fid *,
1487 struct ptlrpc_request **);
1488 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1489 int , void *, int, struct ptlrpc_request **,
1490 struct md_open_data **mod);
1491 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1492 struct obd_capa *, struct ptlrpc_request **);
1493 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1494 struct page **, struct ptlrpc_request **);
1496 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1497 struct ptlrpc_request **);
1499 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1500 struct obd_capa *, obd_valid, const char *,
1501 const char *, int, int, int, __u32,
1502 struct ptlrpc_request **);
1504 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1505 struct obd_capa *, obd_valid, const char *,
1506 const char *, int, int, int,
1507 struct ptlrpc_request **);
1509 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1511 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1512 struct obd_export *, struct obd_export *,
1513 struct lustre_md *);
1515 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1517 int (*m_set_open_replay_data)(struct obd_export *,
1518 struct obd_client_handle *,
1519 struct ptlrpc_request *);
1520 int (*m_clear_open_replay_data)(struct obd_export *,
1521 struct obd_client_handle *);
1522 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1524 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1525 const struct lu_fid *, ldlm_type_t,
1526 ldlm_policy_data_t *, ldlm_mode_t,
1527 struct lustre_handle *);
1529 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1530 ldlm_policy_data_t *, ldlm_mode_t,
1531 ldlm_cancel_flags_t flags, void *opaque);
1532 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1533 renew_capa_cb_t cb);
1534 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1535 const struct req_msg_field *, struct obd_capa **);
1537 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1538 struct obd_capa *, __u32,
1539 struct ptlrpc_request **);
1541 int (*m_intent_getattr_async)(struct obd_export *,
1542 struct md_enqueue_info *,
1543 struct ldlm_enqueue_info *);
1545 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1546 struct lu_fid *, __u64 *bits);
1549 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1550 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1551 * wrapper function in include/linux/obd_class.h.
1555 struct lsm_operations {
1556 void (*lsm_free)(struct lov_stripe_md *);
1557 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1558 struct obd_export *md_exp);
1559 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1561 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1563 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1564 __u16 *stripe_count);
1565 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1566 struct lov_mds_md *lmm);
1569 extern const struct lsm_operations lsm_v1_ops;
1570 extern const struct lsm_operations lsm_v3_ops;
1571 static inline const struct lsm_operations *lsm_op_find(int magic)
1579 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1584 /* Requests for obd_extent_calc() */
1585 #define OBD_CALC_STRIPE_START 1
1586 #define OBD_CALC_STRIPE_END 2
1588 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1590 return oinfo->oi_capa;
1593 static inline struct md_open_data *obd_mod_alloc(void)
1595 struct md_open_data *mod;
1599 atomic_set(&mod->mod_refcount, 1);
1603 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1604 #define obd_mod_put(mod) \
1606 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1607 if ((mod)->mod_open_req) \
1608 ptlrpc_req_finished((mod)->mod_open_req); \
1609 OBD_FREE_PTR(mod); \
1613 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1614 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1616 /* return 1 if client should be resend request */
1617 static inline int client_should_resend(int resend, struct client_obd *cli)
1619 return atomic_read(&cli->cl_resends) ?
1620 atomic_read(&cli->cl_resends) > resend : 1;
1624 * Return device name for this device
1626 * XXX: lu_device is declared before obd_device, while a pointer pointing
1627 * back to obd_device in lu_device, so this helper function defines here
1628 * instead of in lu_object.h
1630 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1632 return lu_dev->ld_obd->obd_name;
1635 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1640 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1643 /* caller does not care of idx */
1647 /* volatile file, the MDT can be set from name */
1648 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1649 /* if no MDT is specified, use std way */
1650 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1652 /* test for no MDT idx case */
1653 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1654 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1658 /* we have an idx, read it */
1659 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1660 *idx = strtoul(start, &end, 0);
1662 * no digit, no trailing :, negative value
1664 if (((*idx == 0) && (end == start)) ||
1665 (*end != ':') || (*idx < 0))
1670 /* bad format of mdt idx, we cannot return an error
1671 * to caller so we use hash algo */
1672 CERROR("Bad volatile file name format: %s\n",
1673 name + LUSTRE_VOLATILE_HDR_LEN);
1677 static inline int cli_brw_size(struct obd_device *obd)
1679 LASSERT(obd != NULL);
1680 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1683 #endif /* __OBD_H */