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, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 #define IOC_MDC_MAX_NR 50
59 #include <lustre/lustre_idl.h>
60 #ifdef HAVE_SERVER_SUPPORT
61 # include <lu_target.h>
64 #include <lustre_lib.h>
65 #include <lustre_export.h>
66 #include <lustre_quota.h>
67 #include <lustre_fld.h>
68 #include <lustre_capa.h>
70 #include <libcfs/bitmap.h>
73 #define MAX_OBD_DEVICES 8192
81 struct lov_oinfo { /* per-stripe data structure */
82 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
83 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
84 int loi_ost_gen; /* generation of this loi_ost_idx */
86 unsigned long loi_kms_valid:1;
87 __u64 loi_kms; /* known minimum size */
88 struct ost_lvb loi_lvb;
89 struct osc_async_rc loi_ar;
91 #define loi_id loi_oi.oi_id
92 #define loi_seq loi_oi.oi_seq
94 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
97 oinfo->loi_kms_valid = 1;
100 static inline void loi_init(struct lov_oinfo *loi)
104 struct lov_stripe_md {
105 cfs_atomic_t lsm_refc;
106 cfs_spinlock_t lsm_lock;
107 pid_t lsm_lock_owner; /* debugging */
109 /* maximum possible file size, might change as OSTs status changes,
110 * e.g. disconnected, deactivated */
113 /* Public members. */
114 __u64 lw_object_id; /* lov object id */
115 __u64 lw_object_seq; /* lov object seq */
117 /* LOV-private members start here -- only for use in lov/. */
119 __u32 lw_stripe_size; /* size of the stripe */
120 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
121 __u16 lw_stripe_count; /* number of objects being striped over */
122 __u16 lw_layout_gen; /* generation of the layout */
123 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
126 struct lov_oinfo *lsm_oinfo[0];
129 #define lsm_object_id lsm_wire.lw_object_id
130 #define lsm_object_seq lsm_wire.lw_object_seq
131 #define lsm_magic lsm_wire.lw_magic
132 #define lsm_layout_gen lsm_wire.lw_layout_gen
133 #define lsm_stripe_size lsm_wire.lw_stripe_size
134 #define lsm_pattern lsm_wire.lw_pattern
135 #define lsm_stripe_count lsm_wire.lw_stripe_count
136 #define lsm_pool_name lsm_wire.lw_pool_name
140 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
142 /* obd info for a particular level (lov, osc). */
144 /* Lock policy. It keeps an extent which is specific for a particular
145 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
146 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
147 ldlm_policy_data_t oi_policy;
148 /* Flags used for set request specific flags:
149 - while lock handling, the flags obtained on the enqueue
150 request are set here.
151 - while stats, the flags used for control delay/resend.
152 - while setattr, the flags used for distinguish punch operation
155 /* Lock handle specific for every OSC lock. */
156 struct lustre_handle *oi_lockh;
157 /* lsm data specific for every OSC. */
158 struct lov_stripe_md *oi_md;
159 /* obdo data specific for every OSC, if needed at all. */
161 /* statfs data specific for every OSC, if needed at all. */
162 struct obd_statfs *oi_osfs;
163 /* An update callback which is called to update some data on upper
164 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
165 * request in osc level for enqueue requests. It is also possible to
166 * update some caller data from LOV layer if needed. */
167 obd_enqueue_update_f oi_cb_up;
168 /* oss capability, its type is obd_capa in client to avoid copy.
169 * in contrary its type is lustre_capa in OSS. */
171 /* transfer jobid from ost_sync() to filter_sync()... */
175 /* compare all relevant fields. */
176 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
177 struct lov_stripe_md *m2)
180 * ->lsm_wire contains padding, but it should be zeroed out during
183 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
186 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
187 struct lov_stripe_md *lsm)
189 if (lsm->lsm_magic != lum->lmm_magic)
191 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
192 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
194 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
195 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
197 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
198 (lsm->lsm_pattern != lum->lmm_pattern))
200 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
201 (strncmp(lsm->lsm_pool_name,
202 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
203 LOV_MAXPOOLNAME) != 0))
208 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
210 struct lov_user_md *lum)
212 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
215 *lmm_magic = lumv3->lmm_magic;
217 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
218 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
219 *lmm_magic = LOV_USER_MAGIC_V1;
220 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
221 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
223 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
224 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
226 lustre_swab_lov_user_md_v3(lumv3);
227 *lmm_magic = LOV_USER_MAGIC_V3;
228 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
230 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
231 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
237 void lov_stripe_lock(struct lov_stripe_md *md);
238 void lov_stripe_unlock(struct lov_stripe_md *md);
241 cfs_list_t typ_chain;
242 struct obd_ops *typ_dt_ops;
243 struct md_ops *typ_md_ops;
244 cfs_proc_dir_entry_t *typ_procroot;
247 struct lu_device_type *typ_lu;
248 cfs_spinlock_t obd_type_lock;
258 /* Individual type definitions */
260 struct ost_server_data;
262 struct osd_properties {
263 size_t osd_max_ea_size;
266 #define OBT_MAGIC 0xBDDECEAE
267 /* hold common fields for "target" device */
268 struct obd_device_target {
271 struct super_block *obt_sb;
272 /** last_rcvd file */
273 struct file *obt_rcvd_filp;
274 #ifdef HAVE_SERVER_SUPPORT
275 struct lu_target *obt_lut;
277 __u64 obt_mount_count;
278 cfs_semaphore_t obt_quotachecking;
279 struct lustre_quota_ctxt obt_qctxt;
280 lustre_quota_version_t obt_qfmt;
281 cfs_rw_semaphore_t obt_rwsem;
282 struct vfsmount *obt_vfsmnt;
283 struct file *obt_health_check_filp;
284 struct osd_properties obt_osd_properties;
285 struct obd_job_stats obt_jobstats;
290 LLOG_CONFIG_ORIG_CTXT = 0,
291 LLOG_CONFIG_REPL_CTXT,
292 LLOG_MDS_OST_ORIG_CTXT,
293 LLOG_MDS_OST_REPL_CTXT,
300 LLOG_LOVEA_ORIG_CTXT,
301 LLOG_LOVEA_REPL_CTXT,
302 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
303 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
304 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
308 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
310 struct filter_subdirs {
311 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
321 /* NB this field MUST be first */
322 struct obd_device_target fo_obt;
323 const char *fo_fstype;
326 cfs_dentry_t *fo_dentry_O;
327 cfs_dentry_t **fo_dentry_O_groups;
328 struct filter_subdirs *fo_dentry_O_sub;
329 cfs_mutex_t fo_init_lock; /* group initialization lock */
330 int fo_committed_group;
332 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
334 unsigned long fo_destroys_in_progress;
335 cfs_mutex_t fo_create_locks[FILTER_SUBDIR_COUNT];
337 cfs_list_t fo_export_list;
340 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
341 obd_size fo_tot_granted; /* all values in bytes */
342 obd_size fo_tot_pending;
343 int fo_tot_granted_clients;
345 obd_size fo_readcache_max_filesize;
346 cfs_spinlock_t fo_flags_lock;
347 unsigned int fo_read_cache:1, /**< enable read-only cache */
348 fo_writethrough_cache:1,/**< read cache writes */
349 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
350 fo_raid_degraded:1;/**< RAID device degraded */
352 struct obd_import *fo_mdc_imp;
353 struct obd_uuid fo_mdc_uuid;
354 struct lustre_handle fo_mdc_conn;
355 struct file **fo_last_objid_files;
356 __u64 *fo_last_objids; /* last created objid for groups,
357 * protected by fo_objidlock */
359 cfs_mutex_t fo_alloc_lock;
361 cfs_atomic_t fo_r_in_flight;
362 cfs_atomic_t fo_w_in_flight;
365 * per-filter pool of kiobuf's allocated by filter_common_setup() and
366 * torn down by filter_cleanup().
368 * This pool contains kiobuf used by
369 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
372 * Locking: protected by internal lock of cfs_hash, pool can be
373 * found from this hash table by t_id of ptlrpc_thread.
375 struct cfs_hash *fo_iobuf_hash;
377 cfs_list_t fo_llog_list;
378 cfs_spinlock_t fo_llog_list_lock;
380 struct brw_stats fo_filter_stats;
381 struct lustre_quota_ctxt fo_quota_ctxt;
382 cfs_spinlock_t fo_quotacheck_lock;
383 cfs_atomic_t fo_quotachecking;
385 int fo_fmd_max_num; /* per exp filter_mod_data */
386 int fo_fmd_max_age; /* jiffies to fmd expiry */
387 unsigned long fo_syncjournal:1, /* sync journal on writes */
388 fo_sync_lock_cancel:2;/* sync on lock cancel */
392 cfs_rwlock_t fo_sptlrpc_lock;
393 struct sptlrpc_rule_set fo_sptlrpc_rset;
395 /* capability related */
396 unsigned int fo_fl_oss_capa;
397 cfs_list_t fo_capa_keys;
398 cfs_hlist_head_t *fo_capa_hash;
399 struct llog_commit_master *fo_lcm;
403 struct timeout_item {
404 enum timeout_event ti_event;
405 cfs_time_t ti_timeout;
408 cfs_list_t ti_obd_list;
412 #define OSC_MAX_RIF_DEFAULT 8
413 #define MDS_OSC_MAX_RIF_DEFAULT 50
414 #define OSC_MAX_RIF_MAX 256
415 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
416 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
417 #define OSC_DEFAULT_RESENDS 10
419 /* possible values for fo_sync_lock_cancel */
421 NEVER_SYNC_ON_CANCEL = 0,
422 BLOCKING_SYNC_ON_CANCEL = 1,
423 ALWAYS_SYNC_ON_CANCEL = 2,
424 NUM_SYNC_ON_CANCEL_STATES
427 #define MDC_MAX_RIF_DEFAULT 8
428 #define MDC_MAX_RIF_MAX 512
433 cfs_rw_semaphore_t cl_sem;
434 struct obd_uuid cl_target_uuid;
435 struct obd_import *cl_import; /* ptlrpc connection state */
437 /* max_mds_easize is purely a performance thing so we don't have to
438 * call obd_size_diskmd() all the time. */
439 int cl_default_mds_easize;
440 int cl_max_mds_easize;
441 int cl_max_mds_cookiesize;
443 enum lustre_sec_part cl_sp_me;
444 enum lustre_sec_part cl_sp_to;
445 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
447 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
448 void *cl_llcd_offset;
450 /* the grant values are protected by loi_list_lock below */
451 long cl_dirty; /* all _dirty_ in bytes */
452 long cl_dirty_max; /* allowed w/o rpc */
453 long cl_dirty_transit; /* dirty synchronous */
454 long cl_avail_grant; /* bytes of credit for ost */
455 long cl_lost_grant; /* lost credits (trunc) */
457 /* since we allocate grant by blocks, we don't know how many grant will
458 * be used to add a page into cache. As a solution, we reserve maximum
459 * grant before trying to dirty a page and unreserve the rest.
460 * See osc_{reserve|unreserve}_grant for details. */
461 long cl_reserved_grant;
462 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
463 cfs_time_t cl_next_shrink_grant; /* jiffies */
464 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
465 int cl_grant_shrink_interval; /* seconds */
467 /* A chunk is an optimal size used by osc_extent to determine
468 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
471 int cl_extent_tax; /* extent overhead, by bytes */
473 /* keep track of objects that have lois that contain pages which
474 * have been queued for async brw. this lock also protects the
475 * lists of osc_client_pages that hang off of the loi */
477 * ->cl_loi_list_lock protects consistency of
478 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
479 * ->ap_completion() call-backs are executed under this lock. As we
480 * cannot guarantee that these call-backs never block on all platforms
481 * (as a matter of fact they do block on Mac OS X), type of
482 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
483 * and blocking mutex on Mac OS X. (Alternative is to make this lock
484 * blocking everywhere, but we don't want to slow down fast-path of
485 * our main platform.)
487 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
488 * with client_obd_list_{un,}lock() and
489 * client_obd_list_lock_{init,done}() functions.
491 * NB by Jinshan: though field names are still _loi_, but actually
492 * osc_object{}s are in the list.
494 client_obd_lock_t cl_loi_list_lock;
495 cfs_list_t cl_loi_ready_list;
496 cfs_list_t cl_loi_hp_ready_list;
497 cfs_list_t cl_loi_write_list;
498 cfs_list_t cl_loi_read_list;
501 /* just a sum of the loi/lop pending numbers to be exported by /proc */
502 cfs_atomic_t cl_pending_w_pages;
503 cfs_atomic_t cl_pending_r_pages;
504 int cl_max_pages_per_rpc;
505 int cl_max_rpcs_in_flight;
506 struct obd_histogram cl_read_rpc_hist;
507 struct obd_histogram cl_write_rpc_hist;
508 struct obd_histogram cl_read_page_hist;
509 struct obd_histogram cl_write_page_hist;
510 struct obd_histogram cl_read_offset_hist;
511 struct obd_histogram cl_write_offset_hist;
513 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
514 cfs_atomic_t cl_destroy_in_flight;
515 cfs_waitq_t cl_destroy_waitq;
517 struct mdc_rpc_lock *cl_rpc_lock;
518 struct mdc_rpc_lock *cl_close_lock;
519 struct osc_creator cl_oscc;
522 cfs_semaphore_t cl_mgc_sem;
523 struct vfsmount *cl_mgc_vfsmnt;
524 struct dentry *cl_mgc_configs_dir;
525 cfs_atomic_t cl_mgc_refcount;
526 struct obd_export *cl_mgc_mgsexp;
528 /* checksumming for data sent over the network */
529 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
530 /* supported checksum types that are worked out at connect time */
531 __u32 cl_supp_cksum_types;
532 /* checksum algorithm to be used */
533 cksum_type_t cl_cksum_type;
535 /* also protected by the poorly named _loi_list_lock lock above */
536 struct osc_async_rc cl_ar;
538 /* used by quotacheck */
539 int cl_qchk_stat; /* quotacheck stat of the peer */
541 /* sequence manager */
542 struct lu_client_seq *cl_seq;
544 cfs_atomic_t cl_resends; /* resend count */
546 /* ptlrpc work for writeback in ptlrpcd context */
547 void *cl_writeback_work;
548 /* hash tables for osc_quota_info */
549 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
551 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
553 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
556 /* NB this field MUST be first */
557 struct obd_device_target mds_obt;
558 struct ptlrpc_service *mds_service;
559 struct ptlrpc_service *mds_setattr_service;
560 struct ptlrpc_service *mds_readpage_service;
562 int mds_max_cookiesize;
564 unsigned long mds_atime_diff;
565 struct ll_fid mds_rootfid;
566 cfs_dentry_t *mds_pending_dir;
567 cfs_dentry_t *mds_logs_dir;
568 cfs_dentry_t *mds_objects_dir;
569 struct llog_handle *mds_cfg_llh;
570 struct obd_device *mds_lov_obd;
571 struct obd_uuid mds_lov_uuid;
573 struct obd_export *mds_lov_exp;
574 struct lov_desc mds_lov_desc;
577 /* mark pages dirty for write. */
578 cfs_bitmap_t *mds_lov_page_dirty;
579 /* array for store pages with obd_id */
580 void **mds_lov_page_array;
581 /* file for store objid */
582 struct file *mds_lov_objid_filp;
583 __u32 mds_lov_objid_count;
584 __u32 mds_lov_objid_max_index;
585 __u32 mds_lov_objid_lastpage;
586 __u32 mds_lov_objid_lastidx;
589 struct lustre_quota_info mds_quota_info;
590 cfs_rw_semaphore_t mds_qonoff_sem;
591 unsigned long mds_fl_user_xattr:1,
593 mds_evict_ost_nids:1,
597 mds_fl_target:1; /* mds have one or
600 struct upcall_cache *mds_identity_cache;
602 /* for capability keys update */
603 struct lustre_capa_key *mds_capa_keys;
604 cfs_rw_semaphore_t mds_notify_lock;
608 extern __u32 mds_max_ost_index;
610 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
612 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
614 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
616 extern int mds_lov_init_objids(struct obd_device *obd);
617 extern void mds_lov_destroy_objids(struct obd_device *obd);
627 struct obd_device_target eo_obt;
629 cfs_spinlock_t eo_lock;
631 struct lustre_handle eo_nl_lock;
632 cfs_atomic_t eo_prep;
636 struct ptlrpc_service *ost_service;
637 struct ptlrpc_service *ost_create_service;
638 struct ptlrpc_service *ost_io_service;
639 cfs_mutex_t ost_health_mutex;
642 struct echo_client_obd {
643 struct obd_export *ec_exp; /* the local connection to osc/lov */
644 cfs_spinlock_t ec_lock;
645 cfs_list_t ec_objects;
652 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
653 cfs_list_t lqo_oss_list; /* link to lov_qos */
654 __u64 lqo_bavail; /* total bytes avail on OSS */
655 __u64 lqo_penalty; /* current penalty */
656 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
657 time_t lqo_used; /* last used time, seconds */
658 __u32 lqo_ost_count; /* number of osts on this oss */
662 struct lov_qos_oss *ltq_oss; /* oss info */
663 __u64 ltq_penalty; /* current penalty */
664 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
665 __u64 ltq_weight; /* net weighting */
666 time_t ltq_used; /* last used time, seconds */
667 unsigned int ltq_usable:1; /* usable for striping */
670 /* Generic subset of OSTs */
672 __u32 *op_array; /* array of index of
674 unsigned int op_count; /* number of OSTs in the array */
675 unsigned int op_size; /* allocated size of lp_array */
676 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
679 /* Round-robin allocator data */
681 __u32 lqr_start_idx; /* start index of new inode */
682 __u32 lqr_offset_idx; /* aliasing for start_idx */
683 int lqr_start_count; /* reseed counter */
684 struct ost_pool lqr_pool; /* round-robin optimized list */
685 unsigned long lqr_dirty:1; /* recalc round-robin list */
688 /* allow statfs data caching for 1 second */
689 #define OBD_STATFS_CACHE_SECONDS 1
691 struct lov_statfs_data {
692 struct obd_info lsd_oi;
693 struct obd_statfs lsd_statfs;
695 /* Stripe placement optimization */
697 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
698 cfs_rw_semaphore_t lq_rw_sem;
699 __u32 lq_active_oss_count;
700 unsigned int lq_prio_free; /* priority for free space */
701 unsigned int lq_threshold_rr;/* priority for rr */
702 struct lov_qos_rr lq_rr; /* round robin qos data */
703 unsigned long lq_dirty:1, /* recalc qos data */
704 lq_same_space:1,/* the ost's all have approx.
705 the same space avail */
706 lq_reset:1, /* zero current penalties */
707 lq_statfs_in_progress:1; /* statfs op in
709 /* qos statfs data */
710 struct lov_statfs_data *lq_statfs_data;
711 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
712 * requests completion */
715 struct lov_tgt_desc {
717 struct obd_uuid ltd_uuid;
718 struct obd_device *ltd_obd;
719 struct obd_export *ltd_exp;
720 struct ltd_qos ltd_qos; /* qos info per target */
722 __u32 ltd_index; /* index in lov_obd->tgts */
723 unsigned long ltd_active:1,/* is this target up for requests */
724 ltd_activate:1,/* should target be activated */
725 ltd_reap:1; /* should this target be deleted */
729 #define pool_tgt_size(_p) _p->pool_obds.op_size
730 #define pool_tgt_count(_p) _p->pool_obds.op_count
731 #define pool_tgt_array(_p) _p->pool_obds.op_array
732 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
733 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
736 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
737 struct ost_pool pool_obds; /* pool members */
738 cfs_atomic_t pool_refcount; /* pool ref. counter */
739 struct lov_qos_rr pool_rr; /* round robin qos */
740 cfs_hlist_node_t pool_hash; /* access by poolname */
741 cfs_list_t pool_list; /* serial access */
742 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
743 struct lov_obd *pool_lov; /* lov obd to which this
748 struct lov_desc desc;
749 struct lov_tgt_desc **lov_tgts; /* sparse array */
750 struct ost_pool lov_packed; /* all OSTs in a packed
752 cfs_mutex_t lov_lock;
753 struct obd_connect_data lov_ocd;
754 struct lov_qos lov_qos; /* qos info per lov */
755 cfs_atomic_t lov_refcount;
756 __u32 lov_tgt_count; /* how many OBD's */
757 __u32 lov_active_tgt_count; /* how many active */
758 __u32 lov_death_row;/* tgts scheduled to be deleted */
759 __u32 lov_tgt_size; /* size of tgts array */
762 cfs_hash_t *lov_pools_hash_body; /* used for key access */
763 cfs_list_t lov_pool_list; /* used for sequential access */
764 cfs_proc_dir_entry_t *lov_pool_proc_entry;
765 enum lustre_sec_part lov_sp_me;
768 struct lmv_tgt_desc {
769 struct obd_uuid ltd_uuid;
770 struct obd_export *ltd_exp;
771 int ltd_active; /* is this target up for requests */
773 cfs_mutex_t ltd_fid_mutex;
776 enum placement_policy {
777 PLACEMENT_CHAR_POLICY = 0,
778 PLACEMENT_NID_POLICY = 1,
779 PLACEMENT_INVAL_POLICY = 2,
783 typedef enum placement_policy placement_policy_t;
787 struct lu_client_fld lmv_fld;
788 cfs_spinlock_t lmv_lock;
789 placement_policy_t lmv_placement;
790 struct lmv_desc desc;
791 struct obd_uuid cluuid;
792 struct obd_export *exp;
799 cfs_mutex_t init_mutex;
801 struct lmv_tgt_desc *tgts;
804 struct obd_connect_data *datas;
807 struct obd_connect_data conn_data;
810 struct niobuf_local {
811 __u64 lnb_file_offset;
812 __u32 lnb_page_offset;
816 cfs_dentry_t *dentry;
821 #define LUSTRE_FLD_NAME "fld"
822 #define LUSTRE_SEQ_NAME "seq"
824 #define LUSTRE_CMM_NAME "cmm"
825 #define LUSTRE_MDD_NAME "mdd"
826 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
827 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
828 #define LUSTRE_VVP_NAME "vvp"
829 #define LUSTRE_LMV_NAME "lmv"
830 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
831 #define LUSTRE_SLP_NAME "slp"
832 #define LUSTRE_LOD_NAME "lod"
833 #define LUSTRE_OSP_NAME "osp"
835 /* obd device type names */
836 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
837 #define LUSTRE_MDS_NAME "mds"
838 #define LUSTRE_MDT_NAME "mdt"
839 #define LUSTRE_MDC_NAME "mdc"
840 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
841 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
842 #define LUSTRE_OSC_NAME "osc"
843 #define LUSTRE_LOV_NAME "lov"
844 #define LUSTRE_MGS_NAME "mgs"
845 #define LUSTRE_MGC_NAME "mgc"
847 #define LUSTRE_CACHEOBD_NAME "cobd"
848 #define LUSTRE_ECHO_NAME "obdecho"
849 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
851 /* Constant obd names (post-rename) */
852 #define LUSTRE_MDS_OBDNAME "MDS"
853 #define LUSTRE_OSS_OBDNAME "OSS"
854 #define LUSTRE_MGS_OBDNAME "MGS"
855 #define LUSTRE_MGC_OBDNAME "MGC"
857 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
858 #define N_LOCAL_TEMP_PAGE 0x10000000
860 struct obd_trans_info {
863 /* Only used on the server side for tracking acks. */
864 struct oti_req_ack_lock {
865 struct lustre_handle lock;
869 struct llog_cookie oti_onecookie;
870 struct llog_cookie *oti_logcookies;
872 /** synchronous write is needed */
873 long oti_sync_write:1;
875 /* initial thread handling transaction */
876 struct ptlrpc_thread * oti_thread;
879 __u64 oti_pre_version;
883 struct obd_uuid *oti_ost_uuid;
886 static inline void oti_init(struct obd_trans_info *oti,
887 struct ptlrpc_request *req)
891 memset(oti, 0, sizeof(*oti));
896 oti->oti_xid = req->rq_xid;
897 /** VBR: take versions from request */
898 if (req->rq_reqmsg != NULL &&
899 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
900 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
901 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
902 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
905 /** called from mds_create_objects */
906 if (req->rq_repmsg != NULL)
907 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
908 oti->oti_thread = req->rq_svc_thread;
909 if (req->rq_reqmsg != NULL)
910 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
913 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
918 if (num_cookies == 1)
919 oti->oti_logcookies = &oti->oti_onecookie;
921 OBD_ALLOC_LARGE(oti->oti_logcookies,
922 num_cookies * sizeof(oti->oti_onecookie));
924 oti->oti_numcookies = num_cookies;
927 static inline void oti_free_cookies(struct obd_trans_info *oti)
929 if (!oti || !oti->oti_logcookies)
932 if (oti->oti_logcookies == &oti->oti_onecookie)
933 LASSERT(oti->oti_numcookies == 1);
935 OBD_FREE_LARGE(oti->oti_logcookies,
936 oti->oti_numcookies*sizeof(oti->oti_onecookie));
937 oti->oti_logcookies = NULL;
938 oti->oti_numcookies = 0;
942 * Events signalled through obd_notify() upcall-chain.
944 enum obd_notify_event {
947 /* Device connect start */
949 /* Device activated */
951 /* Device deactivated */
953 /* Device disconnected */
955 /* Connect data for import were changed */
958 OBD_NOTIFY_SYNC_NONBLOCK,
960 /* Configuration event */
962 /* Administratively deactivate/activate event */
963 OBD_NOTIFY_DEACTIVATE,
967 /* bit-mask flags for config events */
969 CONFIG_LOG = 0x1, /* finished processing config log */
970 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
971 CONFIG_TARGET = 0x4 /* one target is added */
975 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
976 * and liblustre being main examples).
978 struct obd_notify_upcall {
979 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
980 enum obd_notify_event ev, void *owner, void *data);
981 /* Opaque datum supplied by upper layer listener */
985 struct target_recovery_data {
986 svc_handler_t trd_recovery_handler;
987 pid_t trd_processing_task;
988 cfs_completion_t trd_starting;
989 cfs_completion_t trd_finishing;
993 * In HEAD for CMD, the object is created in group number which is 3>=
994 * or indexing starts from 3. To test this assertions are added to disallow
995 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
996 * object in group 0 needs to be allowed.
997 * So for interop mode following changes needs to be done:
998 * 1. No need to assert on group 0 or allow group 0
999 * 2. The group number indexing starts from 0 instead of 3
1002 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
1004 static inline __u64 objseq_to_mdsno(obd_seq seq)
1006 LASSERT_SEQ_IS_MDT(seq);
1007 if (seq == FID_SEQ_OST_MDT0)
1009 return seq - FID_SEQ_OST_MDT1 + 1;
1012 static inline int mdt_to_obd_objseq(int mdtid)
1015 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
1019 return FID_SEQ_OST_MDT1 + mdtid - 1;
1023 struct obd_llog_group {
1024 cfs_list_t olg_list;
1026 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
1027 cfs_waitq_t olg_waitq;
1028 cfs_spinlock_t olg_lock;
1029 struct obd_export *olg_exp;
1030 int olg_initializing;
1031 cfs_mutex_t olg_cat_processing;
1034 /* corresponds to one of the obd's */
1035 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1036 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1039 struct obd_type *obd_type;
1042 /* common and UUID name of this device */
1043 char obd_name[MAX_OBD_NAME];
1044 struct obd_uuid obd_uuid;
1046 struct lu_device *obd_lu_dev;
1049 /* bitfield modification is protected by obd_dev_lock */
1050 unsigned long obd_attached:1, /* finished attach */
1051 obd_set_up:1, /* finished setup */
1052 obd_recovering:1, /* there are recoverable clients */
1053 obd_abort_recovery:1,/* recovery expired */
1054 obd_version_recov:1, /* obd uses version checking */
1055 obd_replayable:1, /* recovery is enabled; inform clients */
1056 obd_no_transno:1, /* no committed-transno notification */
1057 obd_no_recov:1, /* fail instead of retry messages */
1058 obd_stopping:1, /* started cleanup */
1059 obd_starting:1, /* started setup */
1060 obd_force:1, /* cleanup with > 0 obd refcount */
1061 obd_fail:1, /* cleanup with failover */
1062 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1063 obd_no_conn:1, /* deny new connections */
1064 obd_inactive:1, /* device active/inactive
1065 * (for /proc/status only!!) */
1066 obd_no_ir:1, /* no imperative recovery. */
1067 obd_process_conf:1; /* device is processing mgs config */
1068 /* use separate field as it is set in interrupt to don't mess with
1069 * protection of other bits using _bh lock */
1070 unsigned long obd_recovery_expired:1;
1071 /* uuid-export hash body */
1072 cfs_hash_t *obd_uuid_hash;
1073 /* nid-export hash body */
1074 cfs_hash_t *obd_nid_hash;
1075 /* nid stats body */
1076 cfs_hash_t *obd_nid_stats_hash;
1077 cfs_list_t obd_nid_stats;
1078 cfs_atomic_t obd_refcount;
1079 cfs_waitq_t obd_refcount_waitq;
1080 cfs_list_t obd_exports;
1081 cfs_list_t obd_unlinked_exports;
1082 cfs_list_t obd_delayed_exports;
1083 int obd_num_exports;
1084 cfs_spinlock_t obd_nid_lock;
1085 struct ldlm_namespace *obd_namespace;
1086 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1087 /* a spinlock is OK for what we do now, may need a semaphore later */
1088 cfs_spinlock_t obd_dev_lock; /* protects obd bitfield above */
1089 cfs_mutex_t obd_dev_mutex;
1090 __u64 obd_last_committed;
1091 struct fsfilt_operations *obd_fsops;
1092 cfs_spinlock_t obd_osfs_lock;
1093 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1095 struct lvfs_run_ctxt obd_lvfs_ctxt;
1096 struct obd_llog_group obd_olg; /* default llog group */
1097 struct obd_device *obd_observer;
1098 cfs_rw_semaphore_t obd_observer_link_sem;
1099 struct obd_notify_upcall obd_upcall;
1100 struct obd_export *obd_self_export;
1101 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1102 cfs_list_t obd_exports_timed;
1103 time_t obd_eviction_timer; /* for ping evictor */
1105 int obd_max_recoverable_clients;
1106 cfs_atomic_t obd_connected_clients;
1107 int obd_stale_clients;
1108 int obd_delayed_clients;
1109 /* this lock protects all recovery list_heads, timer and
1110 * obd_next_recovery_transno value */
1111 cfs_spinlock_t obd_recovery_task_lock;
1112 __u64 obd_next_recovery_transno;
1113 int obd_replayed_requests;
1114 int obd_requests_queued_for_recovery;
1115 cfs_waitq_t obd_next_transno_waitq;
1116 /* protected by obd_recovery_task_lock */
1117 cfs_timer_t obd_recovery_timer;
1118 time_t obd_recovery_start; /* seconds */
1119 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1120 int obd_recovery_time_hard;
1121 int obd_recovery_timeout;
1122 int obd_recovery_ir_factor;
1124 /* new recovery stuff from CMD2 */
1125 struct target_recovery_data obd_recovery_data;
1126 int obd_replayed_locks;
1127 cfs_atomic_t obd_req_replay_clients;
1128 cfs_atomic_t obd_lock_replay_clients;
1129 /* all lists are protected by obd_recovery_task_lock */
1130 cfs_list_t obd_req_replay_queue;
1131 cfs_list_t obd_lock_replay_queue;
1132 cfs_list_t obd_final_req_queue;
1133 int obd_recovery_stage;
1136 struct obd_device_target obt;
1137 struct filter_obd filter;
1139 struct client_obd cli;
1141 struct echo_client_obd echo_client;
1142 struct echo_obd echo;
1146 /* Fields used by LProcFS */
1147 unsigned int obd_cntr_base;
1148 struct lprocfs_stats *obd_stats;
1150 unsigned int md_cntr_base;
1151 struct lprocfs_stats *md_stats;
1153 cfs_proc_dir_entry_t *obd_proc_entry;
1154 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1155 cfs_proc_dir_entry_t *obd_svc_procroot;
1156 struct lprocfs_stats *obd_svc_stats;
1157 cfs_atomic_t obd_evict_inprogress;
1158 cfs_waitq_t obd_evict_inprogress_waitq;
1159 cfs_list_t obd_evict_list; /* protected with pet_lock */
1162 * Ldlm pool part. Save last calculated SLV and Limit.
1164 cfs_rwlock_t obd_pool_lock;
1169 * A list of outstanding class_incref()'s against this obd. For
1172 struct lu_ref obd_reference;
1174 int obd_conn_inprogress;
1177 #define OBD_LLOG_FL_SENDNOW 0x0001
1178 #define OBD_LLOG_FL_EXIT 0x0002
1180 enum obd_cleanup_stage {
1181 /* Special case hack for MDS LOVs */
1183 /* can be directly mapped to .ldto_device_fini() */
1184 OBD_CLEANUP_EXPORTS,
1187 /* get/set_info keys */
1188 #define KEY_ASYNC "async"
1189 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1190 #define KEY_BLOCKSIZE "blocksize"
1191 #define KEY_CAPA_KEY "capa_key"
1192 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1193 #define KEY_FID2PATH "fid2path"
1194 #define KEY_CHECKSUM "checksum"
1195 #define KEY_CLEAR_FS "clear_fs"
1196 #define KEY_CONN_DATA "conn_data"
1197 #define KEY_EVICT_BY_NID "evict_by_nid"
1198 #define KEY_FIEMAP "fiemap"
1199 #define KEY_FLUSH_CTX "flush_ctx"
1200 #define KEY_GRANT_SHRINK "grant_shrink"
1201 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1202 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1203 #define KEY_INIT_RECOV "initial_recov"
1204 #define KEY_INTERMDS "inter_mds"
1205 #define KEY_LAST_ID "last_id"
1206 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1207 #define KEY_LOVDESC "lovdesc"
1208 #define KEY_LOV_IDX "lov_idx"
1209 #define KEY_MAX_EASIZE "max_easize"
1210 #define KEY_MDS_CONN "mds_conn"
1211 #define KEY_MGSSEC "mgssec"
1212 #define KEY_NEXT_ID "next_id"
1213 #define KEY_READ_ONLY "read-only"
1214 #define KEY_REGISTER_TARGET "register_target"
1215 #define KEY_REVIMP_UPD "revimp_update"
1216 #define KEY_SET_FS "set_fs"
1217 #define KEY_TGT_COUNT "tgt_count"
1218 /* KEY_SET_INFO in lustre_idl.h */
1219 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1220 #define KEY_CONNECT_FLAG "connect_flags"
1221 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1225 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1226 #define IT_OPEN (1 << 0)
1227 #define IT_CREAT (1 << 1)
1228 #define IT_READDIR (1 << 2)
1229 #define IT_GETATTR (1 << 3)
1230 #define IT_LOOKUP (1 << 4)
1231 #define IT_UNLINK (1 << 5)
1232 #define IT_TRUNC (1 << 6)
1233 #define IT_GETXATTR (1 << 7)
1234 #define IT_EXEC (1 << 8)
1235 #define IT_PIN (1 << 9)
1236 #define IT_LAYOUT (1 << 10)
1237 #define IT_QUOTA_DQACQ (1 << 11)
1238 #define IT_QUOTA_CONN (1 << 12)
1240 static inline int it_to_lock_mode(struct lookup_intent *it)
1242 /* CREAT needs to be tested before open (both could be set) */
1243 if (it->it_op & IT_CREAT)
1245 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1249 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1254 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1255 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1256 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1257 struct lu_fid op_fid4; /* to the operation locks. */
1258 mdsno_t op_mds; /* what mds server open will go to */
1259 struct lustre_handle op_handle;
1260 obd_time op_mod_time;
1261 const char *op_name;
1264 struct lmv_stripe_md *op_mea1;
1265 struct lmv_stripe_md *op_mea2;
1266 __u32 op_suppgids[2];
1272 /* iattr fields and blocks. */
1273 struct iattr op_attr;
1275 unsigned int op_attr_flags;
1278 loff_t op_attr_blocks;
1280 /* Size-on-MDS epoch and flags. */
1285 struct obd_capa *op_capa1;
1286 struct obd_capa *op_capa2;
1288 /* Various operation flags. */
1291 /* Operation type */
1294 /* Used by readdir */
1299 struct md_enqueue_info;
1300 /* metadata stat-ahead */
1301 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1302 struct md_enqueue_info *minfo,
1305 struct md_enqueue_info {
1306 struct md_op_data mi_data;
1307 struct lookup_intent mi_it;
1308 struct lustre_handle mi_lockh;
1309 struct inode *mi_dir;
1310 md_enqueue_cb_t mi_cb;
1312 unsigned int mi_generation;
1316 cfs_module_t *o_owner;
1317 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1318 void *karg, void *uarg);
1319 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1320 __u32 keylen, void *key, __u32 *vallen, void *val,
1321 struct lov_stripe_md *lsm);
1322 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1323 __u32 keylen, void *key,
1324 __u32 vallen, void *val,
1325 struct ptlrpc_request_set *set);
1326 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1327 int (*o_detach)(struct obd_device *dev);
1328 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1329 int (*o_precleanup)(struct obd_device *dev,
1330 enum obd_cleanup_stage cleanup_stage);
1331 int (*o_cleanup)(struct obd_device *dev);
1332 int (*o_process_config)(struct obd_device *dev, obd_count len,
1334 int (*o_postrecov)(struct obd_device *dev);
1335 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1337 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1338 /* connect to the target device with given connection
1339 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1340 * granted by the target, which are guaranteed to be a subset of flags
1341 * asked for. If @ocd == NULL, use default parameters. */
1342 int (*o_connect)(const struct lu_env *env,
1343 struct obd_export **exp, struct obd_device *src,
1344 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1346 int (*o_reconnect)(const struct lu_env *env,
1347 struct obd_export *exp, struct obd_device *src,
1348 struct obd_uuid *cluuid,
1349 struct obd_connect_data *ocd,
1351 int (*o_disconnect)(struct obd_export *exp);
1353 /* Initialize/finalize fids infrastructure. */
1354 int (*o_fid_init)(struct obd_export *exp);
1355 int (*o_fid_fini)(struct obd_export *exp);
1357 /* Allocate new fid according to passed @hint. */
1358 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1359 struct md_op_data *op_data);
1362 * Object with @fid is getting deleted, we may want to do something
1365 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1367 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1368 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1369 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1370 __u64 max_age, struct ptlrpc_request_set *set);
1371 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1372 struct lov_stripe_md *mem_src);
1373 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1374 struct lov_mds_md *disk_src, int disk_len);
1375 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1377 /* FIXME: add fid capability support for create & destroy! */
1378 int (*o_precreate)(struct obd_export *exp);
1379 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1380 struct obdo *oa, struct lov_stripe_md **ea,
1381 struct obd_trans_info *oti);
1382 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1383 struct lov_stripe_md **ea,
1384 struct obd_trans_info *oti);
1385 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1386 struct obdo *oa, struct lov_stripe_md *ea,
1387 struct obd_trans_info *oti, struct obd_export *md_exp,
1389 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1390 struct obd_info *oinfo, struct obd_trans_info *oti);
1391 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1392 struct obd_trans_info *oti,
1393 struct ptlrpc_request_set *rqset);
1394 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1395 struct obd_info *oinfo);
1396 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1397 struct ptlrpc_request_set *set);
1398 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1399 obd_count oa_bufs, struct brw_page *pgarr,
1400 struct obd_trans_info *oti);
1401 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1402 struct ost_lvb *lvb, int kms_only);
1403 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1404 obd_off size, int shrink);
1405 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1406 struct obd_info *oinfo, struct obd_trans_info *oti,
1407 struct ptlrpc_request_set *rqset);
1408 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1409 struct obd_info *oinfo, obd_size start, obd_size end,
1410 struct ptlrpc_request_set *set);
1411 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1412 struct lov_stripe_md *src, obd_size start,
1413 obd_size end, struct obd_trans_info *oti);
1414 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1415 struct lustre_handle *srconn, struct lov_stripe_md *src,
1416 obd_size start, obd_size end, struct obd_trans_info *);
1417 int (*o_iterate)(struct lustre_handle *conn,
1418 int (*)(obd_id, obd_seq, void *),
1419 obd_id *startid, obd_seq seq, void *data);
1420 int (*o_preprw)(const struct lu_env *env, int cmd,
1421 struct obd_export *exp, struct obdo *oa, int objcount,
1422 struct obd_ioobj *obj, struct niobuf_remote *remote,
1423 int *nr_pages, struct niobuf_local *local,
1424 struct obd_trans_info *oti, struct lustre_capa *capa);
1425 int (*o_commitrw)(const struct lu_env *env, int cmd,
1426 struct obd_export *exp, struct obdo *oa,
1427 int objcount, struct obd_ioobj *obj,
1428 struct niobuf_remote *remote, int pages,
1429 struct niobuf_local *local,
1430 struct obd_trans_info *oti, int rc);
1431 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1432 struct ldlm_enqueue_info *einfo,
1433 struct ptlrpc_request_set *rqset);
1434 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1435 ldlm_iterator_t it, void *data);
1436 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1437 ldlm_iterator_t it, void *data);
1438 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1439 __u32 mode, struct lustre_handle *);
1440 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1441 ldlm_cancel_flags_t flags, void *opaque);
1442 int (*o_init_export)(struct obd_export *exp);
1443 int (*o_destroy_export)(struct obd_export *exp);
1444 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1445 int cmd, obd_off *);
1447 /* llog related obd_methods */
1448 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1449 struct obd_device *disk_obd, int *idx);
1450 int (*o_llog_finish)(struct obd_device *obd, int count);
1451 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1453 /* metadata-only methods */
1454 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1455 struct obd_capa *, struct obd_client_handle *, int flag);
1456 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1458 int (*o_import_event)(struct obd_device *, struct obd_import *,
1459 enum obd_import_event);
1461 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1462 enum obd_notify_event ev, void *data);
1464 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1465 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1468 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1469 struct obd_quotactl *);
1470 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1471 struct obd_quotactl *);
1473 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1476 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1477 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1478 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1480 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1482 void (*o_getref)(struct obd_device *obd);
1483 void (*o_putref)(struct obd_device *obd);
1485 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1486 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1487 * Also, add a wrapper function in include/linux/obd_class.h. */
1491 LUSTRE_OPC_MKDIR = (1 << 0),
1492 LUSTRE_OPC_SYMLINK = (1 << 1),
1493 LUSTRE_OPC_MKNOD = (1 << 2),
1494 LUSTRE_OPC_CREATE = (1 << 3),
1495 LUSTRE_OPC_ANY = (1 << 4)
1498 /* lmv structures */
1499 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1500 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1501 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1503 #define MAX_HASH_SIZE_32 0x7fffffffUL
1504 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1505 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1508 struct mdt_body *body;
1509 struct lov_stripe_md *lsm;
1510 struct lmv_stripe_md *mea;
1511 #ifdef CONFIG_FS_POSIX_ACL
1512 struct posix_acl *posix_acl;
1514 struct mdt_remote_perm *remote_perm;
1515 struct obd_capa *mds_capa;
1516 struct obd_capa *oss_capa;
1519 struct md_open_data {
1520 struct obd_client_handle *mod_och;
1521 struct ptlrpc_request *mod_open_req;
1522 struct ptlrpc_request *mod_close_req;
1523 cfs_atomic_t mod_refcount;
1526 struct lookup_intent;
1529 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1530 struct obd_capa **);
1531 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1532 ldlm_iterator_t, void *);
1533 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1534 ldlm_iterator_t, void *);
1535 int (*m_close)(struct obd_export *, struct md_op_data *,
1536 struct md_open_data *, struct ptlrpc_request **);
1537 int (*m_create)(struct obd_export *, struct md_op_data *,
1538 const void *, int, int, __u32, __u32, cfs_cap_t,
1539 __u64, struct ptlrpc_request **);
1540 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1541 struct md_open_data *);
1542 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1543 struct lookup_intent *, struct md_op_data *,
1544 struct lustre_handle *, void *, int,
1545 struct ptlrpc_request **, int);
1546 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1547 struct ptlrpc_request **);
1548 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1549 struct ptlrpc_request **);
1550 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1551 void *, int, struct lookup_intent *, int,
1552 struct ptlrpc_request **,
1553 ldlm_blocking_callback, int);
1554 int (*m_link)(struct obd_export *, struct md_op_data *,
1555 struct ptlrpc_request **);
1556 int (*m_rename)(struct obd_export *, struct md_op_data *,
1557 const char *, int, const char *, int,
1558 struct ptlrpc_request **);
1559 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1560 const struct lu_fid *,
1561 struct ptlrpc_request **);
1562 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1563 int , void *, int, struct ptlrpc_request **,
1564 struct md_open_data **mod);
1565 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1566 struct obd_capa *, struct ptlrpc_request **);
1567 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1568 struct page **, struct ptlrpc_request **);
1570 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1571 struct ptlrpc_request **);
1573 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1574 struct obd_capa *, obd_valid, const char *,
1575 const char *, int, int, int, __u32,
1576 struct ptlrpc_request **);
1578 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1579 struct obd_capa *, obd_valid, const char *,
1580 const char *, int, int, int,
1581 struct ptlrpc_request **);
1583 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1585 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1586 struct obd_export *, struct obd_export *,
1587 struct lustre_md *);
1589 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1591 int (*m_set_open_replay_data)(struct obd_export *,
1592 struct obd_client_handle *,
1593 struct ptlrpc_request *);
1594 int (*m_clear_open_replay_data)(struct obd_export *,
1595 struct obd_client_handle *);
1596 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1598 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1599 const struct lu_fid *, ldlm_type_t,
1600 ldlm_policy_data_t *, ldlm_mode_t,
1601 struct lustre_handle *);
1603 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1604 ldlm_policy_data_t *, ldlm_mode_t,
1605 ldlm_cancel_flags_t flags, void *opaque);
1606 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1607 renew_capa_cb_t cb);
1608 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1609 const struct req_msg_field *, struct obd_capa **);
1611 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1612 struct obd_capa *, __u32,
1613 struct ptlrpc_request **);
1615 int (*m_intent_getattr_async)(struct obd_export *,
1616 struct md_enqueue_info *,
1617 struct ldlm_enqueue_info *);
1619 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1620 struct lu_fid *, __u64 *bits);
1623 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1624 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1625 * wrapper function in include/linux/obd_class.h.
1629 struct lsm_operations {
1630 void (*lsm_free)(struct lov_stripe_md *);
1631 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1632 struct obd_export *md_exp);
1633 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1635 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1637 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1638 __u16 *stripe_count);
1639 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1640 struct lov_mds_md *lmm);
1643 extern const struct lsm_operations lsm_v1_ops;
1644 extern const struct lsm_operations lsm_v3_ops;
1645 static inline const struct lsm_operations *lsm_op_find(int magic)
1653 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1658 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1660 /* Requests for obd_extent_calc() */
1661 #define OBD_CALC_STRIPE_START 1
1662 #define OBD_CALC_STRIPE_END 2
1664 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1665 struct obd_export *exp, int error)
1668 CERROR("%s: transno "LPU64" commit error: %d\n",
1669 obd->obd_name, transno, error);
1672 if (exp && transno > exp->exp_last_committed) {
1673 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1674 obd->obd_name, transno);
1675 exp->exp_last_committed = transno;
1676 ptlrpc_commit_replies(exp);
1678 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1679 obd->obd_name, transno);
1681 if (transno > obd->obd_last_committed)
1682 obd->obd_last_committed = transno;
1685 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1687 return oinfo->oi_capa;
1690 static inline struct md_open_data *obd_mod_alloc(void)
1692 struct md_open_data *mod;
1696 cfs_atomic_set(&mod->mod_refcount, 1);
1700 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1701 #define obd_mod_put(mod) \
1703 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1704 if ((mod)->mod_open_req) \
1705 ptlrpc_req_finished((mod)->mod_open_req); \
1706 OBD_FREE_PTR(mod); \
1710 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1711 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1713 /* return 1 if client should be resend request */
1714 static inline int client_should_resend(int resend, struct client_obd *cli)
1716 return cfs_atomic_read(&cli->cl_resends) ?
1717 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1721 * Return device name for this device
1723 * XXX: lu_device is declared before obd_device, while a pointer pointing
1724 * back to obd_device in lu_device, so this helper function defines here
1725 * instead of in lu_object.h
1727 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1729 return lu_dev->ld_obd->obd_name;
1732 #endif /* __OBD_H */