1/* 2 * GPL HEADER START 3 * 4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 5 * 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. 9 * 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). 15 * 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 19 * 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 22 * have any questions. 23 * 24 * GPL HEADER END 25 */ 26/* 27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. 28 * Use is subject to license terms. 29 * 30 * Copyright (c) 2011, 2012, Intel Corporation. 31 */ 32/* 33 * This file is part of Lustre, http://www.lustre.org/ 34 * Lustre is a trademark of Sun Microsystems, Inc. 35 */ 36 37#ifndef __OBD_H 38#define __OBD_H 39 40#include "linux/obd.h" 41 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 46 47#define IOC_MDC_TYPE 'i' 48#define IOC_MDC_MIN_NR 20 49#define IOC_MDC_MAX_NR 50 50 51#include "lustre/lustre_idl.h" 52#include "lustre_lib.h" 53#include "lu_ref.h" 54#include "lustre_export.h" 55#include "lustre_fid.h" 56#include "lustre_fld.h" 57#include "lustre_capa.h" 58 59#define MAX_OBD_DEVICES 8192 60 61struct osc_async_rc { 62 int ar_rc; 63 int ar_force_sync; 64 __u64 ar_min_xid; 65}; 66 67struct lov_oinfo { /* per-stripe data structure */ 68 struct ost_id loi_oi; /* object ID/Sequence on the target OST */ 69 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */ 70 int loi_ost_gen; /* generation of this loi_ost_idx */ 71 72 unsigned long loi_kms_valid:1; 73 __u64 loi_kms; /* known minimum size */ 74 struct ost_lvb loi_lvb; 75 struct osc_async_rc loi_ar; 76}; 77 78static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms) 79{ 80 oinfo->loi_kms = kms; 81 oinfo->loi_kms_valid = 1; 82} 83 84static inline void loi_init(struct lov_oinfo *loi) 85{ 86} 87 88struct lov_stripe_md { 89 atomic_t lsm_refc; 90 spinlock_t lsm_lock; 91 pid_t lsm_lock_owner; /* debugging */ 92 93 /* maximum possible file size, might change as OSTs status changes, 94 * e.g. disconnected, deactivated */ 95 __u64 lsm_maxbytes; 96 struct { 97 /* Public members. */ 98 struct ost_id lw_object_oi; /* lov object id/seq */ 99 100 /* LOV-private members start here -- only for use in lov/. */ 101 __u32 lw_magic; 102 __u32 lw_stripe_size; /* size of the stripe */ 103 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */ 104 __u16 lw_stripe_count; /* number of objects being striped over */ 105 __u16 lw_layout_gen; /* generation of the layout */ 106 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */ 107 } lsm_wire; 108 109 struct lov_oinfo *lsm_oinfo[0]; 110}; 111 112#define lsm_oi lsm_wire.lw_object_oi 113#define lsm_magic lsm_wire.lw_magic 114#define lsm_layout_gen lsm_wire.lw_layout_gen 115#define lsm_stripe_size lsm_wire.lw_stripe_size 116#define lsm_pattern lsm_wire.lw_pattern 117#define lsm_stripe_count lsm_wire.lw_stripe_count 118#define lsm_pool_name lsm_wire.lw_pool_name 119 120static inline bool lsm_is_released(struct lov_stripe_md *lsm) 121{ 122 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED); 123} 124 125static inline bool lsm_has_objects(struct lov_stripe_md *lsm) 126{ 127 if (lsm == NULL) 128 return false; 129 if (lsm_is_released(lsm)) 130 return false; 131 return true; 132} 133 134static inline int lov_stripe_md_size(unsigned int stripe_count) 135{ 136 struct lov_stripe_md lsm; 137 138 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]); 139} 140 141struct obd_info; 142 143typedef int (*obd_enqueue_update_f)(void *cookie, int rc); 144 145/* obd info for a particular level (lov, osc). */ 146struct obd_info { 147 /* Lock policy. It keeps an extent which is specific for a particular 148 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy, 149 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */ 150 ldlm_policy_data_t oi_policy; 151 /* Flags used for set request specific flags: 152 - while lock handling, the flags obtained on the enqueue 153 request are set here. 154 - while stats, the flags used for control delay/resend. 155 - while setattr, the flags used for distinguish punch operation 156 */ 157 __u64 oi_flags; 158 /* Lock handle specific for every OSC lock. */ 159 struct lustre_handle *oi_lockh; 160 /* lsm data specific for every OSC. */ 161 struct lov_stripe_md *oi_md; 162 /* obdo data specific for every OSC, if needed at all. */ 163 struct obdo *oi_oa; 164 /* statfs data specific for every OSC, if needed at all. */ 165 struct obd_statfs *oi_osfs; 166 /* An update callback which is called to update some data on upper 167 * level. E.g. it is used for update lsm->lsm_oinfo at every received 168 * request in osc level for enqueue requests. It is also possible to 169 * update some caller data from LOV layer if needed. */ 170 obd_enqueue_update_f oi_cb_up; 171 /* oss capability, its type is obd_capa in client to avoid copy. 172 * in contrary its type is lustre_capa in OSS. */ 173 void *oi_capa; 174 /* transfer jobid from ost_sync() to filter_sync()... */ 175 char *oi_jobid; 176}; 177 178/* compare all relevant fields. */ 179static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1, 180 struct lov_stripe_md *m2) 181{ 182 /* 183 * ->lsm_wire contains padding, but it should be zeroed out during 184 * allocation. 185 */ 186 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof(m1->lsm_wire)); 187} 188 189static inline int lov_lum_lsm_cmp(struct lov_user_md *lum, 190 struct lov_stripe_md *lsm) 191{ 192 if (lsm->lsm_magic != lum->lmm_magic) 193 return 1; 194 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) && 195 (lsm->lsm_stripe_count != lum->lmm_stripe_count)) 196 return 2; 197 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) && 198 (lsm->lsm_stripe_size != lum->lmm_stripe_size)) 199 return 3; 200 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) && 201 (lsm->lsm_pattern != lum->lmm_pattern)) 202 return 4; 203 if ((lsm->lsm_magic == LOV_MAGIC_V3) && 204 (strncmp(lsm->lsm_pool_name, 205 ((struct lov_user_md_v3 *)lum)->lmm_pool_name, 206 LOV_MAXPOOLNAME) != 0)) 207 return 5; 208 return 0; 209} 210 211static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3, 212 int *lmm_magic, 213 struct lov_user_md *lum) 214{ 215 if (lum && copy_from_user(lumv3, lum, sizeof(struct lov_user_md_v1))) 216 return -EFAULT; 217 218 *lmm_magic = lumv3->lmm_magic; 219 220 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) { 221 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3); 222 *lmm_magic = LOV_USER_MAGIC_V1; 223 } else if (*lmm_magic == LOV_USER_MAGIC_V3) { 224 if (lum && copy_from_user(lumv3, lum, sizeof(*lumv3))) 225 return -EFAULT; 226 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) { 227 if (lum && copy_from_user(lumv3, lum, sizeof(*lumv3))) 228 return -EFAULT; 229 lustre_swab_lov_user_md_v3(lumv3); 230 *lmm_magic = LOV_USER_MAGIC_V3; 231 } else if (*lmm_magic != LOV_USER_MAGIC_V1) { 232 CDEBUG(D_IOCTL, 233 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n", 234 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3); 235 return -EINVAL; 236 } 237 return 0; 238} 239 240void lov_stripe_lock(struct lov_stripe_md *md); 241void lov_stripe_unlock(struct lov_stripe_md *md); 242 243struct obd_type { 244 struct list_head typ_chain; 245 struct obd_ops *typ_dt_ops; 246 struct md_ops *typ_md_ops; 247 struct proc_dir_entry *typ_procroot; 248 char *typ_name; 249 int typ_refcnt; 250 struct lu_device_type *typ_lu; 251 spinlock_t obd_type_lock; 252}; 253 254struct brw_page { 255 u64 off; 256 struct page *pg; 257 int count; 258 u32 flag; 259}; 260 261/* llog contexts */ 262enum llog_ctxt_id { 263 LLOG_CONFIG_ORIG_CTXT = 0, 264 LLOG_CONFIG_REPL_CTXT, 265 LLOG_MDS_OST_ORIG_CTXT, 266 LLOG_MDS_OST_REPL_CTXT, 267 LLOG_SIZE_ORIG_CTXT, 268 LLOG_SIZE_REPL_CTXT, 269 LLOG_RD1_ORIG_CTXT, 270 LLOG_RD1_REPL_CTXT, 271 LLOG_TEST_ORIG_CTXT, 272 LLOG_TEST_REPL_CTXT, 273 LLOG_LOVEA_ORIG_CTXT, 274 LLOG_LOVEA_REPL_CTXT, 275 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */ 276 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */ 277 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */ 278 LLOG_AGENT_ORIG_CTXT, /**< agent requests generation on cdt */ 279 LLOG_MAX_CTXTS 280}; 281 282struct timeout_item { 283 enum timeout_event ti_event; 284 unsigned long ti_timeout; 285 timeout_cb_t ti_cb; 286 void *ti_cb_data; 287 struct list_head ti_obd_list; 288 struct list_head ti_chain; 289}; 290 291#define OSC_MAX_RIF_DEFAULT 8 292#define MDS_OSC_MAX_RIF_DEFAULT 50 293#define OSC_MAX_RIF_MAX 256 294#define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4) 295#define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */ 296#define OSC_DEFAULT_RESENDS 10 297 298/* possible values for fo_sync_lock_cancel */ 299enum { 300 NEVER_SYNC_ON_CANCEL = 0, 301 BLOCKING_SYNC_ON_CANCEL = 1, 302 ALWAYS_SYNC_ON_CANCEL = 2, 303 NUM_SYNC_ON_CANCEL_STATES 304}; 305 306#define MDC_MAX_RIF_DEFAULT 8 307#define MDC_MAX_RIF_MAX 512 308 309struct mdc_rpc_lock; 310struct obd_import; 311struct client_obd { 312 struct rw_semaphore cl_sem; 313 struct obd_uuid cl_target_uuid; 314 struct obd_import *cl_import; /* ptlrpc connection state */ 315 int cl_conn_count; 316 /* max_mds_easize is purely a performance thing so we don't have to 317 * call obd_size_diskmd() all the time. */ 318 int cl_default_mds_easize; 319 int cl_max_mds_easize; 320 int cl_default_mds_cookiesize; 321 int cl_max_mds_cookiesize; 322 323 enum lustre_sec_part cl_sp_me; 324 enum lustre_sec_part cl_sp_to; 325 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */ 326 327 /* the grant values are protected by loi_list_lock below */ 328 long cl_dirty; /* all _dirty_ in bytes */ 329 long cl_dirty_max; /* allowed w/o rpc */ 330 long cl_dirty_transit; /* dirty synchronous */ 331 long cl_avail_grant; /* bytes of credit for ost */ 332 long cl_lost_grant; /* lost credits (trunc) */ 333 334 /* since we allocate grant by blocks, we don't know how many grant will 335 * be used to add a page into cache. As a solution, we reserve maximum 336 * grant before trying to dirty a page and unreserve the rest. 337 * See osc_{reserve|unreserve}_grant for details. */ 338 long cl_reserved_grant; 339 struct list_head cl_cache_waiters; /* waiting for cache/grant */ 340 unsigned long cl_next_shrink_grant; /* jiffies */ 341 struct list_head cl_grant_shrink_list; /* Timeout event list */ 342 int cl_grant_shrink_interval; /* seconds */ 343 344 /* A chunk is an optimal size used by osc_extent to determine 345 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */ 346 int cl_chunkbits; 347 int cl_chunk; 348 int cl_extent_tax; /* extent overhead, by bytes */ 349 350 /* keep track of objects that have lois that contain pages which 351 * have been queued for async brw. this lock also protects the 352 * lists of osc_client_pages that hang off of the loi */ 353 /* 354 * ->cl_loi_list_lock protects consistency of 355 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and 356 * ->ap_completion() call-backs are executed under this lock. As we 357 * cannot guarantee that these call-backs never block on all platforms 358 * (as a matter of fact they do block on Mac OS X), type of 359 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux 360 * and blocking mutex on Mac OS X. (Alternative is to make this lock 361 * blocking everywhere, but we don't want to slow down fast-path of 362 * our main platform.) 363 * 364 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together 365 * with client_obd_list_{un,}lock() and 366 * client_obd_list_lock_{init,done}() functions. 367 * 368 * NB by Jinshan: though field names are still _loi_, but actually 369 * osc_object{}s are in the list. 370 */ 371 client_obd_lock_t cl_loi_list_lock; 372 struct list_head cl_loi_ready_list; 373 struct list_head cl_loi_hp_ready_list; 374 struct list_head cl_loi_write_list; 375 struct list_head cl_loi_read_list; 376 int cl_r_in_flight; 377 int cl_w_in_flight; 378 /* just a sum of the loi/lop pending numbers to be exported by /proc */ 379 atomic_t cl_pending_w_pages; 380 atomic_t cl_pending_r_pages; 381 __u32 cl_max_pages_per_rpc; 382 int cl_max_rpcs_in_flight; 383 struct obd_histogram cl_read_rpc_hist; 384 struct obd_histogram cl_write_rpc_hist; 385 struct obd_histogram cl_read_page_hist; 386 struct obd_histogram cl_write_page_hist; 387 struct obd_histogram cl_read_offset_hist; 388 struct obd_histogram cl_write_offset_hist; 389 390 /* lru for osc caching pages */ 391 struct cl_client_cache *cl_cache; 392 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */ 393 atomic_t *cl_lru_left; 394 atomic_t cl_lru_busy; 395 atomic_t cl_lru_shrinkers; 396 atomic_t cl_lru_in_list; 397 struct list_head cl_lru_list; /* lru page list */ 398 client_obd_lock_t cl_lru_list_lock; /* page list protector */ 399 400 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */ 401 atomic_t cl_destroy_in_flight; 402 wait_queue_head_t cl_destroy_waitq; 403 404 struct mdc_rpc_lock *cl_rpc_lock; 405 struct mdc_rpc_lock *cl_close_lock; 406 407 /* mgc datastruct */ 408 struct mutex cl_mgc_mutex; 409 struct local_oid_storage *cl_mgc_los; 410 struct dt_object *cl_mgc_configs_dir; 411 atomic_t cl_mgc_refcount; 412 struct obd_export *cl_mgc_mgsexp; 413 414 /* checksumming for data sent over the network */ 415 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */ 416 /* supported checksum types that are worked out at connect time */ 417 __u32 cl_supp_cksum_types; 418 /* checksum algorithm to be used */ 419 cksum_type_t cl_cksum_type; 420 421 /* also protected by the poorly named _loi_list_lock lock above */ 422 struct osc_async_rc cl_ar; 423 424 /* used by quotacheck when the servers are older than 2.4 */ 425 int cl_qchk_stat; /* quotacheck stat of the peer */ 426#define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */ 427#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0) 428#warning "please consider removing quotacheck compatibility code" 429#endif 430 431 /* sequence manager */ 432 struct lu_client_seq *cl_seq; 433 434 atomic_t cl_resends; /* resend count */ 435 436 /* ptlrpc work for writeback in ptlrpcd context */ 437 void *cl_writeback_work; 438 /* hash tables for osc_quota_info */ 439 struct cfs_hash *cl_quota_hash[MAXQUOTAS]; 440}; 441#define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid) 442 443struct obd_id_info { 444 __u32 idx; 445 u64 *data; 446}; 447 448struct echo_client_obd { 449 struct obd_export *ec_exp; /* the local connection to osc/lov */ 450 spinlock_t ec_lock; 451 struct list_head ec_objects; 452 struct list_head ec_locks; 453 int ec_nstripes; 454 __u64 ec_unique; 455}; 456 457struct lov_qos_oss { 458 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */ 459 struct list_head lqo_oss_list; /* link to lov_qos */ 460 __u64 lqo_bavail; /* total bytes avail on OSS */ 461 __u64 lqo_penalty; /* current penalty */ 462 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/ 463 time_t lqo_used; /* last used time, seconds */ 464 __u32 lqo_ost_count; /* number of osts on this oss */ 465}; 466 467struct ltd_qos { 468 struct lov_qos_oss *ltq_oss; /* oss info */ 469 __u64 ltq_penalty; /* current penalty */ 470 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/ 471 __u64 ltq_weight; /* net weighting */ 472 time_t ltq_used; /* last used time, seconds */ 473 unsigned int ltq_usable:1; /* usable for striping */ 474}; 475 476/* Generic subset of OSTs */ 477struct ost_pool { 478 __u32 *op_array; /* array of index of 479 lov_obd->lov_tgts */ 480 unsigned int op_count; /* number of OSTs in the array */ 481 unsigned int op_size; /* allocated size of lp_array */ 482 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */ 483}; 484 485/* Round-robin allocator data */ 486struct lov_qos_rr { 487 __u32 lqr_start_idx; /* start index of new inode */ 488 __u32 lqr_offset_idx; /* aliasing for start_idx */ 489 int lqr_start_count; /* reseed counter */ 490 struct ost_pool lqr_pool; /* round-robin optimized list */ 491 unsigned long lqr_dirty:1; /* recalc round-robin list */ 492}; 493 494/* allow statfs data caching for 1 second */ 495#define OBD_STATFS_CACHE_SECONDS 1 496 497struct lov_statfs_data { 498 struct obd_info lsd_oi; 499 struct obd_statfs lsd_statfs; 500}; 501/* Stripe placement optimization */ 502struct lov_qos { 503 struct list_head lq_oss_list; /* list of OSSs that targets use */ 504 struct rw_semaphore lq_rw_sem; 505 __u32 lq_active_oss_count; 506 unsigned int lq_prio_free; /* priority for free space */ 507 unsigned int lq_threshold_rr;/* priority for rr */ 508 struct lov_qos_rr lq_rr; /* round robin qos data */ 509 unsigned long lq_dirty:1, /* recalc qos data */ 510 lq_same_space:1,/* the ost's all have approx. 511 the same space avail */ 512 lq_reset:1, /* zero current penalties */ 513 lq_statfs_in_progress:1; /* statfs op in 514 progress */ 515 /* qos statfs data */ 516 struct lov_statfs_data *lq_statfs_data; 517 wait_queue_head_t lq_statfs_waitq; /* waitqueue to notify statfs 518 * requests completion */ 519}; 520 521struct lov_tgt_desc { 522 struct list_head ltd_kill; 523 struct obd_uuid ltd_uuid; 524 struct obd_device *ltd_obd; 525 struct obd_export *ltd_exp; 526 struct ltd_qos ltd_qos; /* qos info per target */ 527 __u32 ltd_gen; 528 __u32 ltd_index; /* index in lov_obd->tgts */ 529 unsigned long ltd_active:1,/* is this target up for requests */ 530 ltd_activate:1,/* should target be activated */ 531 ltd_reap:1; /* should this target be deleted */ 532}; 533 534/* Pool metadata */ 535#define pool_tgt_size(_p) _p->pool_obds.op_size 536#define pool_tgt_count(_p) _p->pool_obds.op_count 537#define pool_tgt_array(_p) _p->pool_obds.op_array 538#define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem 539 540struct pool_desc { 541 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */ 542 struct ost_pool pool_obds; /* pool members */ 543 atomic_t pool_refcount; /* pool ref. counter */ 544 struct lov_qos_rr pool_rr; /* round robin qos */ 545 struct hlist_node pool_hash; /* access by poolname */ 546 struct list_head pool_list; /* serial access */ 547 struct proc_dir_entry *pool_proc_entry; /* file in /proc */ 548 struct obd_device *pool_lobd; /* obd of the lov/lod to which 549 * this pool belongs */ 550}; 551 552struct lov_obd { 553 struct lov_desc desc; 554 struct lov_tgt_desc **lov_tgts; /* sparse array */ 555 struct ost_pool lov_packed; /* all OSTs in a packed 556 array */ 557 struct mutex lov_lock; 558 struct obd_connect_data lov_ocd; 559 atomic_t lov_refcount; 560 __u32 lov_tgt_count; /* how many OBD's */ 561 __u32 lov_active_tgt_count; /* how many active */ 562 __u32 lov_death_row;/* tgts scheduled to be deleted */ 563 __u32 lov_tgt_size; /* size of tgts array */ 564 int lov_connects; 565 int lov_pool_count; 566 struct cfs_hash *lov_pools_hash_body; /* used for key access */ 567 struct list_head lov_pool_list; /* used for sequential access */ 568 struct proc_dir_entry *lov_pool_proc_entry; 569 enum lustre_sec_part lov_sp_me; 570 571 /* Cached LRU pages from upper layer */ 572 void *lov_cache; 573 574 struct rw_semaphore lov_notify_lock; 575}; 576 577struct lmv_tgt_desc { 578 struct obd_uuid ltd_uuid; 579 struct obd_export *ltd_exp; 580 int ltd_idx; 581 struct mutex ltd_fid_mutex; 582 unsigned long ltd_active:1; /* target up for requests */ 583}; 584 585enum placement_policy { 586 PLACEMENT_CHAR_POLICY = 0, 587 PLACEMENT_NID_POLICY = 1, 588 PLACEMENT_INVAL_POLICY = 2, 589 PLACEMENT_MAX_POLICY 590}; 591 592struct lmv_obd { 593 int refcount; 594 struct lu_client_fld lmv_fld; 595 spinlock_t lmv_lock; 596 enum placement_policy lmv_placement; 597 struct lmv_desc desc; 598 struct obd_uuid cluuid; 599 struct obd_export *exp; 600 601 struct mutex init_mutex; 602 int connected; 603 int max_easize; 604 int max_def_easize; 605 int max_cookiesize; 606 int max_def_cookiesize; 607 int server_timeout; 608 609 int tgts_size; /* size of tgts array */ 610 struct lmv_tgt_desc **tgts; 611 612 struct obd_connect_data conn_data; 613}; 614 615struct niobuf_local { 616 __u64 lnb_file_offset; 617 __u32 lnb_page_offset; 618 __u32 len; 619 __u32 flags; 620 struct page *page; 621 struct dentry *dentry; 622 int lnb_grant_used; 623 int rc; 624}; 625 626#define LUSTRE_FLD_NAME "fld" 627#define LUSTRE_SEQ_NAME "seq" 628 629#define LUSTRE_MDD_NAME "mdd" 630#define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs" 631#define LUSTRE_OSD_ZFS_NAME "osd-zfs" 632#define LUSTRE_VVP_NAME "vvp" 633#define LUSTRE_LMV_NAME "lmv" 634#define LUSTRE_SLP_NAME "slp" 635#define LUSTRE_LOD_NAME "lod" 636#define LUSTRE_OSP_NAME "osp" 637#define LUSTRE_LWP_NAME "lwp" 638 639/* obd device type names */ 640 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */ 641#define LUSTRE_MDS_NAME "mds" 642#define LUSTRE_MDT_NAME "mdt" 643#define LUSTRE_MDC_NAME "mdc" 644#define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */ 645#define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */ 646#define LUSTRE_OSC_NAME "osc" 647#define LUSTRE_LOV_NAME "lov" 648#define LUSTRE_MGS_NAME "mgs" 649#define LUSTRE_MGC_NAME "mgc" 650 651#define LUSTRE_ECHO_NAME "obdecho" 652#define LUSTRE_ECHO_CLIENT_NAME "echo_client" 653#define LUSTRE_QMT_NAME "qmt" 654 655/* Constant obd names (post-rename) */ 656#define LUSTRE_MDS_OBDNAME "MDS" 657#define LUSTRE_OSS_OBDNAME "OSS" 658#define LUSTRE_MGS_OBDNAME "MGS" 659#define LUSTRE_MGC_OBDNAME "MGC" 660 661/* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */ 662#define N_LOCAL_TEMP_PAGE 0x10000000 663 664struct obd_trans_info { 665 __u64 oti_transno; 666 __u64 oti_xid; 667 /* Only used on the server side for tracking acks. */ 668 struct oti_req_ack_lock { 669 struct lustre_handle lock; 670 __u32 mode; 671 } oti_ack_locks[4]; 672 void *oti_handle; 673 struct llog_cookie oti_onecookie; 674 struct llog_cookie *oti_logcookies; 675 int oti_numcookies; 676 /** synchronous write is needed */ 677 unsigned long oti_sync_write:1; 678 679 /* initial thread handling transaction */ 680 struct ptlrpc_thread *oti_thread; 681 __u32 oti_conn_cnt; 682 /** VBR: versions */ 683 __u64 oti_pre_version; 684 /** JobID */ 685 char *oti_jobid; 686 687 struct obd_uuid *oti_ost_uuid; 688}; 689 690static inline void oti_init(struct obd_trans_info *oti, 691 struct ptlrpc_request *req) 692{ 693 if (oti == NULL) 694 return; 695 memset(oti, 0, sizeof(*oti)); 696 697 if (req == NULL) 698 return; 699 700 oti->oti_xid = req->rq_xid; 701 /** VBR: take versions from request */ 702 if (req->rq_reqmsg != NULL && 703 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) { 704 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg); 705 706 oti->oti_pre_version = pre_version ? pre_version[0] : 0; 707 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg); 708 } 709 710 /** called from mds_create_objects */ 711 if (req->rq_repmsg != NULL) 712 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg); 713 oti->oti_thread = req->rq_svc_thread; 714 if (req->rq_reqmsg != NULL) 715 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg); 716} 717 718static inline void oti_alloc_cookies(struct obd_trans_info *oti, 719 int num_cookies) 720{ 721 if (!oti) 722 return; 723 724 if (num_cookies == 1) 725 oti->oti_logcookies = &oti->oti_onecookie; 726 else 727 OBD_ALLOC_LARGE(oti->oti_logcookies, 728 num_cookies * sizeof(oti->oti_onecookie)); 729 730 oti->oti_numcookies = num_cookies; 731} 732 733static inline void oti_free_cookies(struct obd_trans_info *oti) 734{ 735 if (!oti || !oti->oti_logcookies) 736 return; 737 738 if (oti->oti_logcookies == &oti->oti_onecookie) 739 LASSERT(oti->oti_numcookies == 1); 740 else 741 OBD_FREE_LARGE(oti->oti_logcookies, 742 oti->oti_numcookies*sizeof(oti->oti_onecookie)); 743 oti->oti_logcookies = NULL; 744 oti->oti_numcookies = 0; 745} 746 747/* 748 * Events signalled through obd_notify() upcall-chain. 749 */ 750enum obd_notify_event { 751 /* target added */ 752 OBD_NOTIFY_CREATE, 753 /* Device connect start */ 754 OBD_NOTIFY_CONNECT, 755 /* Device activated */ 756 OBD_NOTIFY_ACTIVE, 757 /* Device deactivated */ 758 OBD_NOTIFY_INACTIVE, 759 /* Device disconnected */ 760 OBD_NOTIFY_DISCON, 761 /* Connect data for import were changed */ 762 OBD_NOTIFY_OCD, 763 /* Sync request */ 764 OBD_NOTIFY_SYNC_NONBLOCK, 765 OBD_NOTIFY_SYNC, 766 /* Configuration event */ 767 OBD_NOTIFY_CONFIG, 768 /* Administratively deactivate/activate event */ 769 OBD_NOTIFY_DEACTIVATE, 770 OBD_NOTIFY_ACTIVATE 771}; 772 773/* 774 * Data structure used to pass obd_notify()-event to non-obd listeners (llite 775 * and liblustre being main examples). 776 */ 777struct obd_notify_upcall { 778 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched, 779 enum obd_notify_event ev, void *owner, void *data); 780 /* Opaque datum supplied by upper layer listener */ 781 void *onu_owner; 782}; 783 784struct target_recovery_data { 785 svc_handler_t trd_recovery_handler; 786 pid_t trd_processing_task; 787 struct completion trd_starting; 788 struct completion trd_finishing; 789}; 790 791struct obd_llog_group { 792 int olg_seq; 793 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS]; 794 wait_queue_head_t olg_waitq; 795 spinlock_t olg_lock; 796 struct mutex olg_cat_processing; 797}; 798 799/* corresponds to one of the obd's */ 800#define OBD_DEVICE_MAGIC 0XAB5CD6EF 801#define OBD_DEV_BY_DEVNAME 0xffffd0de 802 803struct lvfs_run_ctxt { 804 struct dt_device *dt; 805}; 806 807struct obd_device { 808 struct obd_type *obd_type; 809 __u32 obd_magic; 810 811 /* common and UUID name of this device */ 812 char obd_name[MAX_OBD_NAME]; 813 struct obd_uuid obd_uuid; 814 815 struct lu_device *obd_lu_dev; 816 817 int obd_minor; 818 /* bitfield modification is protected by obd_dev_lock */ 819 unsigned long obd_attached:1, /* finished attach */ 820 obd_set_up:1, /* finished setup */ 821 obd_recovering:1, /* there are recoverable clients */ 822 obd_abort_recovery:1,/* recovery expired */ 823 obd_version_recov:1, /* obd uses version checking */ 824 obd_replayable:1, /* recovery is enabled; inform clients */ 825 obd_no_transno:1, /* no committed-transno notification */ 826 obd_no_recov:1, /* fail instead of retry messages */ 827 obd_stopping:1, /* started cleanup */ 828 obd_starting:1, /* started setup */ 829 obd_force:1, /* cleanup with > 0 obd refcount */ 830 obd_fail:1, /* cleanup with failover */ 831 obd_async_recov:1, /* allow asynchronous orphan cleanup */ 832 obd_no_conn:1, /* deny new connections */ 833 obd_inactive:1, /* device active/inactive 834 * (for /proc/status only!!) */ 835 obd_no_ir:1, /* no imperative recovery. */ 836 obd_process_conf:1; /* device is processing mgs config */ 837 /* use separate field as it is set in interrupt to don't mess with 838 * protection of other bits using _bh lock */ 839 unsigned long obd_recovery_expired:1; 840 /* uuid-export hash body */ 841 struct cfs_hash *obd_uuid_hash; 842 /* nid-export hash body */ 843 struct cfs_hash *obd_nid_hash; 844 /* nid stats body */ 845 struct cfs_hash *obd_nid_stats_hash; 846 struct list_head obd_nid_stats; 847 atomic_t obd_refcount; 848 wait_queue_head_t obd_refcount_waitq; 849 struct list_head obd_exports; 850 struct list_head obd_unlinked_exports; 851 struct list_head obd_delayed_exports; 852 int obd_num_exports; 853 spinlock_t obd_nid_lock; 854 struct ldlm_namespace *obd_namespace; 855 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */ 856 /* a spinlock is OK for what we do now, may need a semaphore later */ 857 spinlock_t obd_dev_lock; /* protect OBD bitfield above */ 858 struct mutex obd_dev_mutex; 859 __u64 obd_last_committed; 860 spinlock_t obd_osfs_lock; 861 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */ 862 __u64 obd_osfs_age; 863 struct lvfs_run_ctxt obd_lvfs_ctxt; 864 struct obd_llog_group obd_olg; /* default llog group */ 865 struct obd_device *obd_observer; 866 struct rw_semaphore obd_observer_link_sem; 867 struct obd_notify_upcall obd_upcall; 868 struct obd_export *obd_self_export; 869 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */ 870 struct list_head obd_exports_timed; 871 time_t obd_eviction_timer; /* for ping evictor */ 872 873 int obd_max_recoverable_clients; 874 atomic_t obd_connected_clients; 875 int obd_stale_clients; 876 int obd_delayed_clients; 877 /* this lock protects all recovery list_heads, timer and 878 * obd_next_recovery_transno value */ 879 spinlock_t obd_recovery_task_lock; 880 __u64 obd_next_recovery_transno; 881 int obd_replayed_requests; 882 int obd_requests_queued_for_recovery; 883 wait_queue_head_t obd_next_transno_waitq; 884 /* protected by obd_recovery_task_lock */ 885 struct timer_list obd_recovery_timer; 886 time_t obd_recovery_start; /* seconds */ 887 time_t obd_recovery_end; /* seconds, for lprocfs_status */ 888 int obd_recovery_time_hard; 889 int obd_recovery_timeout; 890 int obd_recovery_ir_factor; 891 892 /* new recovery stuff from CMD2 */ 893 struct target_recovery_data obd_recovery_data; 894 int obd_replayed_locks; 895 atomic_t obd_req_replay_clients; 896 atomic_t obd_lock_replay_clients; 897 /* all lists are protected by obd_recovery_task_lock */ 898 struct list_head obd_req_replay_queue; 899 struct list_head obd_lock_replay_queue; 900 struct list_head obd_final_req_queue; 901 int obd_recovery_stage; 902 903 union { 904 struct client_obd cli; 905 struct echo_client_obd echo_client; 906 struct lov_obd lov; 907 struct lmv_obd lmv; 908 } u; 909 /* Fields used by LProcFS */ 910 unsigned int obd_cntr_base; 911 struct lprocfs_stats *obd_stats; 912 913 unsigned int md_cntr_base; 914 struct lprocfs_stats *md_stats; 915 916 struct proc_dir_entry *obd_proc_entry; 917 void *obd_proc_private; /* type private PDEs */ 918 struct proc_dir_entry *obd_proc_exports_entry; 919 struct proc_dir_entry *obd_svc_procroot; 920 struct lprocfs_stats *obd_svc_stats; 921 atomic_t obd_evict_inprogress; 922 wait_queue_head_t obd_evict_inprogress_waitq; 923 struct list_head obd_evict_list; /* protected with pet_lock */ 924 925 /** 926 * Ldlm pool part. Save last calculated SLV and Limit. 927 */ 928 rwlock_t obd_pool_lock; 929 int obd_pool_limit; 930 __u64 obd_pool_slv; 931 932 /** 933 * A list of outstanding class_incref()'s against this obd. For 934 * debugging. 935 */ 936 struct lu_ref obd_reference; 937 938 int obd_conn_inprogress; 939}; 940 941#define OBD_LLOG_FL_SENDNOW 0x0001 942#define OBD_LLOG_FL_EXIT 0x0002 943 944enum obd_cleanup_stage { 945/* Special case hack for MDS LOVs */ 946 OBD_CLEANUP_EARLY, 947/* can be directly mapped to .ldto_device_fini() */ 948 OBD_CLEANUP_EXPORTS, 949}; 950 951/* get/set_info keys */ 952#define KEY_ASYNC "async" 953#define KEY_BLOCKSIZE_BITS "blocksize_bits" 954#define KEY_BLOCKSIZE "blocksize" 955#define KEY_CAPA_KEY "capa_key" 956#define KEY_CHANGELOG_CLEAR "changelog_clear" 957#define KEY_FID2PATH "fid2path" 958#define KEY_CHECKSUM "checksum" 959#define KEY_CLEAR_FS "clear_fs" 960#define KEY_CONN_DATA "conn_data" 961#define KEY_EVICT_BY_NID "evict_by_nid" 962#define KEY_FIEMAP "fiemap" 963#define KEY_FLUSH_CTX "flush_ctx" 964#define KEY_GRANT_SHRINK "grant_shrink" 965#define KEY_HSM_COPYTOOL_SEND "hsm_send" 966#define KEY_INIT_RECOV_BACKUP "init_recov_bk" 967#define KEY_INIT_RECOV "initial_recov" 968#define KEY_INTERMDS "inter_mds" 969#define KEY_LAST_ID "last_id" 970#define KEY_LAST_FID "last_fid" 971#define KEY_LOCK_TO_STRIPE "lock_to_stripe" 972#define KEY_LOVDESC "lovdesc" 973#define KEY_LOV_IDX "lov_idx" 974#define KEY_MAX_EASIZE "max_easize" 975#define KEY_DEFAULT_EASIZE "default_easize" 976#define KEY_MAX_COOKIESIZE "max_cookiesize" 977#define KEY_DEFAULT_COOKIESIZE "default_cookiesize" 978#define KEY_MDS_CONN "mds_conn" 979#define KEY_MGSSEC "mgssec" 980#define KEY_NEXT_ID "next_id" 981#define KEY_READ_ONLY "read-only" 982#define KEY_REGISTER_TARGET "register_target" 983#define KEY_SET_FS "set_fs" 984#define KEY_TGT_COUNT "tgt_count" 985/* KEY_SET_INFO in lustre_idl.h */ 986#define KEY_SPTLRPC_CONF "sptlrpc_conf" 987#define KEY_CONNECT_FLAG "connect_flags" 988#define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel" 989 990#define KEY_CACHE_SET "cache_set" 991#define KEY_CACHE_LRU_SHRINK "cache_lru_shrink" 992#define KEY_CHANGELOG_INDEX "changelog_index" 993 994struct lu_context; 995 996/* /!\ must be coherent with include/linux/namei.h on patched kernel */ 997#define IT_OPEN (1 << 0) 998#define IT_CREAT (1 << 1) 999#define IT_READDIR (1 << 2) 1000#define IT_GETATTR (1 << 3) 1001#define IT_LOOKUP (1 << 4) 1002#define IT_UNLINK (1 << 5) 1003#define IT_TRUNC (1 << 6) 1004#define IT_GETXATTR (1 << 7) 1005#define IT_EXEC (1 << 8) 1006#define IT_PIN (1 << 9) 1007#define IT_LAYOUT (1 << 10) 1008#define IT_QUOTA_DQACQ (1 << 11) 1009#define IT_QUOTA_CONN (1 << 12) 1010#define IT_SETXATTR (1 << 13) 1011 1012static inline int it_to_lock_mode(struct lookup_intent *it) 1013{ 1014 /* CREAT needs to be tested before open (both could be set) */ 1015 if (it->it_op & IT_CREAT) 1016 return LCK_CW; 1017 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP | 1018 IT_LAYOUT)) 1019 return LCK_CR; 1020 else if (it->it_op & IT_GETXATTR) 1021 return LCK_PR; 1022 else if (it->it_op & IT_SETXATTR) 1023 return LCK_PW; 1024 1025 LASSERTF(0, "Invalid it_op: %d\n", it->it_op); 1026 return -EINVAL; 1027} 1028 1029struct md_op_data { 1030 struct lu_fid op_fid1; /* operation fid1 (usually parent) */ 1031 struct lu_fid op_fid2; /* operation fid2 (usually child) */ 1032 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */ 1033 struct lu_fid op_fid4; /* to the operation locks. */ 1034 u32 op_mds; /* what mds server open will go to */ 1035 struct lustre_handle op_handle; 1036 s64 op_mod_time; 1037 const char *op_name; 1038 int op_namelen; 1039 __u32 op_mode; 1040 struct lmv_stripe_md *op_mea1; 1041 struct lmv_stripe_md *op_mea2; 1042 __u32 op_suppgids[2]; 1043 __u32 op_fsuid; 1044 __u32 op_fsgid; 1045 cfs_cap_t op_cap; 1046 void *op_data; 1047 1048 /* iattr fields and blocks. */ 1049 struct iattr op_attr; 1050 unsigned int op_attr_flags; 1051 __u64 op_valid; 1052 loff_t op_attr_blocks; 1053 1054 /* Size-on-MDS epoch and flags. */ 1055 __u64 op_ioepoch; 1056 __u32 op_flags; 1057 1058 /* Capa fields */ 1059 struct obd_capa *op_capa1; 1060 struct obd_capa *op_capa2; 1061 1062 /* Various operation flags. */ 1063 enum mds_op_bias op_bias; 1064 1065 /* Operation type */ 1066 __u32 op_opc; 1067 1068 /* Used by readdir */ 1069 __u64 op_offset; 1070 1071 /* Used by readdir */ 1072 __u32 op_npages; 1073 1074 /* used to transfer info between the stacks of MD client 1075 * see enum op_cli_flags */ 1076 __u32 op_cli_flags; 1077 1078 /* File object data version for HSM release, on client */ 1079 __u64 op_data_version; 1080 struct lustre_handle op_lease_handle; 1081}; 1082 1083enum op_cli_flags { 1084 CLI_SET_MEA = 1 << 0, 1085 CLI_RM_ENTRY = 1 << 1, 1086}; 1087 1088struct md_enqueue_info; 1089/* metadata stat-ahead */ 1090 1091struct md_enqueue_info { 1092 struct md_op_data mi_data; 1093 struct lookup_intent mi_it; 1094 struct lustre_handle mi_lockh; 1095 struct inode *mi_dir; 1096 int (*mi_cb)(struct ptlrpc_request *req, 1097 struct md_enqueue_info *minfo, int rc); 1098 __u64 mi_cbdata; 1099 unsigned int mi_generation; 1100}; 1101 1102struct obd_ops { 1103 struct module *o_owner; 1104 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len, 1105 void *karg, void *uarg); 1106 int (*o_get_info)(const struct lu_env *env, struct obd_export *, 1107 __u32 keylen, void *key, __u32 *vallen, void *val, 1108 struct lov_stripe_md *lsm); 1109 int (*o_set_info_async)(const struct lu_env *, struct obd_export *, 1110 __u32 keylen, void *key, 1111 __u32 vallen, void *val, 1112 struct ptlrpc_request_set *set); 1113 int (*o_attach)(struct obd_device *dev, u32 len, void *data); 1114 int (*o_detach)(struct obd_device *dev); 1115 int (*o_setup)(struct obd_device *dev, struct lustre_cfg *cfg); 1116 int (*o_precleanup)(struct obd_device *dev, 1117 enum obd_cleanup_stage cleanup_stage); 1118 int (*o_cleanup)(struct obd_device *dev); 1119 int (*o_process_config)(struct obd_device *dev, u32 len, void *data); 1120 int (*o_postrecov)(struct obd_device *dev); 1121 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid, 1122 int priority); 1123 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid); 1124 /* connect to the target device with given connection 1125 * data. @ocd->ocd_connect_flags is modified to reflect flags actually 1126 * granted by the target, which are guaranteed to be a subset of flags 1127 * asked for. If @ocd == NULL, use default parameters. */ 1128 int (*o_connect)(const struct lu_env *env, 1129 struct obd_export **exp, struct obd_device *src, 1130 struct obd_uuid *cluuid, struct obd_connect_data *ocd, 1131 void *localdata); 1132 int (*o_reconnect)(const struct lu_env *env, 1133 struct obd_export *exp, struct obd_device *src, 1134 struct obd_uuid *cluuid, 1135 struct obd_connect_data *ocd, 1136 void *localdata); 1137 int (*o_disconnect)(struct obd_export *exp); 1138 1139 /* Initialize/finalize fids infrastructure. */ 1140 int (*o_fid_init)(struct obd_device *obd, 1141 struct obd_export *exp, enum lu_cli_type type); 1142 int (*o_fid_fini)(struct obd_device *obd); 1143 1144 /* Allocate new fid according to passed @hint. */ 1145 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid, 1146 struct md_op_data *op_data); 1147 1148 /* 1149 * Object with @fid is getting deleted, we may want to do something 1150 * about this. 1151 */ 1152 int (*o_statfs)(const struct lu_env *, struct obd_export *exp, 1153 struct obd_statfs *osfs, __u64 max_age, __u32 flags); 1154 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo, 1155 __u64 max_age, struct ptlrpc_request_set *set); 1156 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt, 1157 struct lov_stripe_md *mem_src); 1158 int (*o_unpackmd)(struct obd_export *exp, 1159 struct lov_stripe_md **mem_tgt, 1160 struct lov_mds_md *disk_src, int disk_len); 1161 int (*o_preallocate)(struct lustre_handle *, u32 *req, u64 *ids); 1162 /* FIXME: add fid capability support for create & destroy! */ 1163 int (*o_create)(const struct lu_env *env, struct obd_export *exp, 1164 struct obdo *oa, struct lov_stripe_md **ea, 1165 struct obd_trans_info *oti); 1166 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp, 1167 struct obdo *oa, struct lov_stripe_md *ea, 1168 struct obd_trans_info *oti, struct obd_export *md_exp, 1169 void *capa); 1170 int (*o_setattr)(const struct lu_env *, struct obd_export *exp, 1171 struct obd_info *oinfo, struct obd_trans_info *oti); 1172 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo, 1173 struct obd_trans_info *oti, 1174 struct ptlrpc_request_set *rqset); 1175 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp, 1176 struct obd_info *oinfo); 1177 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo, 1178 struct ptlrpc_request_set *set); 1179 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm, 1180 u64 size, int shrink); 1181 int (*o_preprw)(const struct lu_env *env, int cmd, 1182 struct obd_export *exp, struct obdo *oa, int objcount, 1183 struct obd_ioobj *obj, struct niobuf_remote *remote, 1184 int *nr_pages, struct niobuf_local *local, 1185 struct obd_trans_info *oti, struct lustre_capa *capa); 1186 int (*o_commitrw)(const struct lu_env *env, int cmd, 1187 struct obd_export *exp, struct obdo *oa, 1188 int objcount, struct obd_ioobj *obj, 1189 struct niobuf_remote *remote, int pages, 1190 struct niobuf_local *local, 1191 struct obd_trans_info *oti, int rc); 1192 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *, 1193 ldlm_iterator_t it, void *data); 1194 int (*o_init_export)(struct obd_export *exp); 1195 int (*o_destroy_export)(struct obd_export *exp); 1196 1197 /* metadata-only methods */ 1198 int (*o_import_event)(struct obd_device *, struct obd_import *, 1199 enum obd_import_event); 1200 1201 int (*o_notify)(struct obd_device *obd, struct obd_device *watched, 1202 enum obd_notify_event ev, void *data); 1203 1204 int (*o_health_check)(const struct lu_env *env, struct obd_device *); 1205 struct obd_uuid *(*o_get_uuid)(struct obd_export *exp); 1206 1207 /* quota methods */ 1208 int (*o_quotacheck)(struct obd_device *, struct obd_export *, 1209 struct obd_quotactl *); 1210 int (*o_quotactl)(struct obd_device *, struct obd_export *, 1211 struct obd_quotactl *); 1212 1213 /* pools methods */ 1214 int (*o_pool_new)(struct obd_device *obd, char *poolname); 1215 int (*o_pool_del)(struct obd_device *obd, char *poolname); 1216 int (*o_pool_add)(struct obd_device *obd, char *poolname, 1217 char *ostname); 1218 int (*o_pool_rem)(struct obd_device *obd, char *poolname, 1219 char *ostname); 1220 void (*o_getref)(struct obd_device *obd); 1221 void (*o_putref)(struct obd_device *obd); 1222 /* 1223 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line 1224 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c. 1225 * Also, add a wrapper function in include/linux/obd_class.h. */ 1226}; 1227 1228enum { 1229 LUSTRE_OPC_MKDIR = (1 << 0), 1230 LUSTRE_OPC_SYMLINK = (1 << 1), 1231 LUSTRE_OPC_MKNOD = (1 << 2), 1232 LUSTRE_OPC_CREATE = (1 << 3), 1233 LUSTRE_OPC_ANY = (1 << 4) 1234}; 1235 1236/* lmv structures */ 1237#define MEA_MAGIC_LAST_CHAR 0xb2221ca1 1238#define MEA_MAGIC_ALL_CHARS 0xb222a11c 1239#define MEA_MAGIC_HASH_SEGMENT 0xb222a11b 1240 1241#define MAX_HASH_SIZE_32 0x7fffffffUL 1242#define MAX_HASH_SIZE 0x7fffffffffffffffULL 1243#define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL 1244 1245struct lustre_md { 1246 struct mdt_body *body; 1247 struct lov_stripe_md *lsm; 1248 struct lmv_stripe_md *mea; 1249#ifdef CONFIG_FS_POSIX_ACL 1250 struct posix_acl *posix_acl; 1251#endif 1252 struct mdt_remote_perm *remote_perm; 1253 struct obd_capa *mds_capa; 1254 struct obd_capa *oss_capa; 1255}; 1256 1257struct md_open_data { 1258 struct obd_client_handle *mod_och; 1259 struct ptlrpc_request *mod_open_req; 1260 struct ptlrpc_request *mod_close_req; 1261 atomic_t mod_refcount; 1262 bool mod_is_create; 1263}; 1264 1265struct lookup_intent; 1266 1267struct md_ops { 1268 int (*m_getstatus)(struct obd_export *, struct lu_fid *, 1269 struct obd_capa **); 1270 int (*m_null_inode)(struct obd_export *, const struct lu_fid *); 1271 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *, 1272 ldlm_iterator_t, void *); 1273 int (*m_close)(struct obd_export *, struct md_op_data *, 1274 struct md_open_data *, struct ptlrpc_request **); 1275 int (*m_create)(struct obd_export *, struct md_op_data *, 1276 const void *, int, int, __u32, __u32, cfs_cap_t, 1277 __u64, struct ptlrpc_request **); 1278 int (*m_done_writing)(struct obd_export *, struct md_op_data *, 1279 struct md_open_data *); 1280 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *, 1281 struct lookup_intent *, struct md_op_data *, 1282 struct lustre_handle *, void *, int, 1283 struct ptlrpc_request **, __u64); 1284 int (*m_getattr)(struct obd_export *, struct md_op_data *, 1285 struct ptlrpc_request **); 1286 int (*m_getattr_name)(struct obd_export *, struct md_op_data *, 1287 struct ptlrpc_request **); 1288 int (*m_intent_lock)(struct obd_export *, struct md_op_data *, 1289 void *, int, struct lookup_intent *, int, 1290 struct ptlrpc_request **, 1291 ldlm_blocking_callback, __u64); 1292 int (*m_link)(struct obd_export *, struct md_op_data *, 1293 struct ptlrpc_request **); 1294 int (*m_rename)(struct obd_export *, struct md_op_data *, 1295 const char *, int, const char *, int, 1296 struct ptlrpc_request **); 1297 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *, 1298 const struct lu_fid *, 1299 struct ptlrpc_request **); 1300 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *, 1301 int , void *, int, struct ptlrpc_request **, 1302 struct md_open_data **mod); 1303 int (*m_sync)(struct obd_export *, const struct lu_fid *, 1304 struct obd_capa *, struct ptlrpc_request **); 1305 int (*m_readpage)(struct obd_export *, struct md_op_data *, 1306 struct page **, struct ptlrpc_request **); 1307 1308 int (*m_unlink)(struct obd_export *, struct md_op_data *, 1309 struct ptlrpc_request **); 1310 1311 int (*m_setxattr)(struct obd_export *, const struct lu_fid *, 1312 struct obd_capa *, u64, const char *, 1313 const char *, int, int, int, __u32, 1314 struct ptlrpc_request **); 1315 1316 int (*m_getxattr)(struct obd_export *, const struct lu_fid *, 1317 struct obd_capa *, u64, const char *, 1318 const char *, int, int, int, 1319 struct ptlrpc_request **); 1320 1321 int (*m_init_ea_size)(struct obd_export *, int, int, int, int); 1322 1323 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *, 1324 struct obd_export *, struct obd_export *, 1325 struct lustre_md *); 1326 1327 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *); 1328 1329 int (*m_set_open_replay_data)(struct obd_export *, 1330 struct obd_client_handle *, 1331 struct lookup_intent *); 1332 int (*m_clear_open_replay_data)(struct obd_export *, 1333 struct obd_client_handle *); 1334 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *); 1335 1336 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64, 1337 const struct lu_fid *, ldlm_type_t, 1338 ldlm_policy_data_t *, ldlm_mode_t, 1339 struct lustre_handle *); 1340 1341 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *, 1342 ldlm_policy_data_t *, ldlm_mode_t, 1343 ldlm_cancel_flags_t flags, void *opaque); 1344 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc, 1345 renew_capa_cb_t cb); 1346 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *, 1347 const struct req_msg_field *, struct obd_capa **); 1348 1349 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *, 1350 struct obd_capa *, __u32, 1351 struct ptlrpc_request **); 1352 1353 int (*m_intent_getattr_async)(struct obd_export *, 1354 struct md_enqueue_info *, 1355 struct ldlm_enqueue_info *); 1356 1357 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *, 1358 struct lu_fid *, __u64 *bits); 1359 1360 /* 1361 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to 1362 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a 1363 * wrapper function in include/linux/obd_class.h. 1364 */ 1365}; 1366 1367struct lsm_operations { 1368 void (*lsm_free)(struct lov_stripe_md *); 1369 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa, 1370 struct obd_export *md_exp); 1371 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, u64 *, 1372 u64 *); 1373 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, u64 *, 1374 u64 *); 1375 int (*lsm_lmm_verify)(struct lov_mds_md *lmm, int lmm_bytes, 1376 __u16 *stripe_count); 1377 int (*lsm_unpackmd)(struct lov_obd *lov, struct lov_stripe_md *lsm, 1378 struct lov_mds_md *lmm); 1379}; 1380 1381extern const struct lsm_operations lsm_v1_ops; 1382extern const struct lsm_operations lsm_v3_ops; 1383static inline const struct lsm_operations *lsm_op_find(int magic) 1384{ 1385 switch (magic) { 1386 case LOV_MAGIC_V1: 1387 return &lsm_v1_ops; 1388 case LOV_MAGIC_V3: 1389 return &lsm_v3_ops; 1390 default: 1391 CERROR("Cannot recognize lsm_magic %08x\n", magic); 1392 return NULL; 1393 } 1394} 1395 1396/* Requests for obd_extent_calc() */ 1397#define OBD_CALC_STRIPE_START 1 1398#define OBD_CALC_STRIPE_END 2 1399 1400static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo) 1401{ 1402 return oinfo->oi_capa; 1403} 1404 1405static inline struct md_open_data *obd_mod_alloc(void) 1406{ 1407 struct md_open_data *mod; 1408 1409 OBD_ALLOC_PTR(mod); 1410 if (mod == NULL) 1411 return NULL; 1412 atomic_set(&mod->mod_refcount, 1); 1413 return mod; 1414} 1415 1416#define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount) 1417#define obd_mod_put(mod) \ 1418({ \ 1419 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \ 1420 if ((mod)->mod_open_req) \ 1421 ptlrpc_req_finished((mod)->mod_open_req); \ 1422 OBD_FREE_PTR(mod); \ 1423 } \ 1424}) 1425 1426void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid); 1427void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent); 1428 1429/* return 1 if client should be resend request */ 1430static inline int client_should_resend(int resend, struct client_obd *cli) 1431{ 1432 return atomic_read(&cli->cl_resends) ? 1433 atomic_read(&cli->cl_resends) > resend : 1; 1434} 1435 1436/** 1437 * Return device name for this device 1438 * 1439 * XXX: lu_device is declared before obd_device, while a pointer pointing 1440 * back to obd_device in lu_device, so this helper function defines here 1441 * instead of in lu_object.h 1442 */ 1443static inline const char *lu_dev_name(const struct lu_device *lu_dev) 1444{ 1445 return lu_dev->ld_obd->obd_name; 1446} 1447 1448static inline bool filename_is_volatile(const char *name, int namelen, int *idx) 1449{ 1450 const char *start; 1451 char *end; 1452 1453 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0) 1454 return false; 1455 1456 /* caller does not care of idx */ 1457 if (idx == NULL) 1458 return true; 1459 1460 /* volatile file, the MDT can be set from name */ 1461 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */ 1462 /* if no MDT is specified, use std way */ 1463 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2) 1464 goto bad_format; 1465 /* test for no MDT idx case */ 1466 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') && 1467 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) { 1468 *idx = -1; 1469 return true; 1470 } 1471 /* we have an idx, read it */ 1472 start = name + LUSTRE_VOLATILE_HDR_LEN + 1; 1473 *idx = strtoul(start, &end, 0); 1474 /* error cases: 1475 * no digit, no trailing :, negative value 1476 */ 1477 if (((*idx == 0) && (end == start)) || 1478 (*end != ':') || (*idx < 0)) 1479 goto bad_format; 1480 1481 return true; 1482bad_format: 1483 /* bad format of mdt idx, we cannot return an error 1484 * to caller so we use hash algo */ 1485 CERROR("Bad volatile file name format: %s\n", 1486 name + LUSTRE_VOLATILE_HDR_LEN); 1487 return false; 1488} 1489 1490static inline int cli_brw_size(struct obd_device *obd) 1491{ 1492 LASSERT(obd != NULL); 1493 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT; 1494} 1495 1496#endif /* __OBD_H */ 1497