4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 #define IOC_MDC_MAX_NR 50
59 #include <lustre/lustre_idl.h>
60 #include <lustre_lib.h>
61 #include <libcfs/bitmap.h>
62 #ifdef HAVE_SERVER_SUPPORT
63 # include <lu_target.h>
64 # include <obd_target.h>
67 #include <lustre_export.h>
68 #include <lustre_fid.h>
69 #include <lustre_fld.h>
70 #include <lustre_capa.h>
73 #define MAX_OBD_DEVICES 8192
81 struct lov_oinfo { /* per-stripe data structure */
82 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
83 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
84 int loi_ost_gen; /* generation of this loi_ost_idx */
86 unsigned long loi_kms_valid:1;
87 __u64 loi_kms; /* known minimum size */
88 struct ost_lvb loi_lvb;
89 struct osc_async_rc loi_ar;
92 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
95 oinfo->loi_kms_valid = 1;
98 static inline void loi_init(struct lov_oinfo *loi)
102 struct lov_stripe_md {
105 pid_t lsm_lock_owner; /* debugging */
107 /* maximum possible file size, might change as OSTs status changes,
108 * e.g. disconnected, deactivated */
111 /* Public members. */
112 struct ost_id lw_object_oi; /* lov object id/seq */
114 /* LOV-private members start here -- only for use in lov/. */
116 __u32 lw_stripe_size; /* size of the stripe */
117 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
118 __u16 lw_stripe_count; /* number of objects being striped over */
119 __u16 lw_layout_gen; /* generation of the layout */
120 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
123 struct lov_oinfo *lsm_oinfo[0];
126 #define lsm_oi lsm_wire.lw_object_oi
127 #define lsm_magic lsm_wire.lw_magic
128 #define lsm_layout_gen lsm_wire.lw_layout_gen
129 #define lsm_stripe_size lsm_wire.lw_stripe_size
130 #define lsm_pattern lsm_wire.lw_pattern
131 #define lsm_stripe_count lsm_wire.lw_stripe_count
132 #define lsm_pool_name lsm_wire.lw_pool_name
134 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
136 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
139 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
143 if (lsm_is_released(lsm))
150 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
152 /* obd info for a particular level (lov, osc). */
154 /* Lock policy. It keeps an extent which is specific for a particular
155 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
156 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
157 ldlm_policy_data_t oi_policy;
158 /* Flags used for set request specific flags:
159 - while lock handling, the flags obtained on the enqueue
160 request are set here.
161 - while stats, the flags used for control delay/resend.
162 - while setattr, the flags used for distinguish punch operation
165 /* Lock handle specific for every OSC lock. */
166 struct lustre_handle *oi_lockh;
167 /* lsm data specific for every OSC. */
168 struct lov_stripe_md *oi_md;
169 /* obdo data specific for every OSC, if needed at all. */
171 /* statfs data specific for every OSC, if needed at all. */
172 struct obd_statfs *oi_osfs;
173 /* An update callback which is called to update some data on upper
174 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
175 * request in osc level for enqueue requests. It is also possible to
176 * update some caller data from LOV layer if needed. */
177 obd_enqueue_update_f oi_cb_up;
178 /* oss capability, its type is obd_capa in client to avoid copy.
179 * in contrary its type is lustre_capa in OSS. */
181 /* transfer jobid from ost_sync() to filter_sync()... */
186 cfs_list_t typ_chain;
187 struct obd_ops *typ_dt_ops;
188 struct md_ops *typ_md_ops;
189 cfs_proc_dir_entry_t *typ_procroot;
192 struct lu_device_type *typ_lu;
193 spinlock_t obd_type_lock;
205 LLOG_CONFIG_ORIG_CTXT = 0,
206 LLOG_CONFIG_REPL_CTXT,
207 LLOG_MDS_OST_ORIG_CTXT,
208 LLOG_MDS_OST_REPL_CTXT,
215 LLOG_LOVEA_ORIG_CTXT,
216 LLOG_LOVEA_REPL_CTXT,
217 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
218 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
219 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
220 LLOG_AGENT_ORIG_CTXT, /**< agent requests generation on cdt */
224 struct timeout_item {
225 enum timeout_event ti_event;
226 cfs_time_t ti_timeout;
229 cfs_list_t ti_obd_list;
233 #define OSC_MAX_RIF_DEFAULT 8
234 #define MDS_OSC_MAX_RIF_DEFAULT 50
235 #define OSC_MAX_RIF_MAX 256
236 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
237 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
238 #define OSC_DEFAULT_RESENDS 10
240 /* possible values for fo_sync_lock_cancel */
242 NEVER_SYNC_ON_CANCEL = 0,
243 BLOCKING_SYNC_ON_CANCEL = 1,
244 ALWAYS_SYNC_ON_CANCEL = 2,
245 NUM_SYNC_ON_CANCEL_STATES
248 #define MDC_MAX_RIF_DEFAULT 8
249 #define MDC_MAX_RIF_MAX 512
254 struct rw_semaphore cl_sem;
255 struct obd_uuid cl_target_uuid;
256 struct obd_import *cl_import; /* ptlrpc connection state */
258 /* max_mds_easize is purely a performance thing so we don't have to
259 * call obd_size_diskmd() all the time. */
260 int cl_default_mds_easize;
261 int cl_max_mds_easize;
262 int cl_max_mds_cookiesize;
264 enum lustre_sec_part cl_sp_me;
265 enum lustre_sec_part cl_sp_to;
266 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
268 /* the grant values are protected by loi_list_lock below */
269 long cl_dirty; /* all _dirty_ in bytes */
270 long cl_dirty_max; /* allowed w/o rpc */
271 long cl_dirty_transit; /* dirty synchronous */
272 long cl_avail_grant; /* bytes of credit for ost */
273 long cl_lost_grant; /* lost credits (trunc) */
275 /* since we allocate grant by blocks, we don't know how many grant will
276 * be used to add a page into cache. As a solution, we reserve maximum
277 * grant before trying to dirty a page and unreserve the rest.
278 * See osc_{reserve|unreserve}_grant for details. */
279 long cl_reserved_grant;
280 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
281 cfs_time_t cl_next_shrink_grant; /* jiffies */
282 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
283 int cl_grant_shrink_interval; /* seconds */
285 /* A chunk is an optimal size used by osc_extent to determine
286 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
289 int cl_extent_tax; /* extent overhead, by bytes */
291 /* keep track of objects that have lois that contain pages which
292 * have been queued for async brw. this lock also protects the
293 * lists of osc_client_pages that hang off of the loi */
295 * ->cl_loi_list_lock protects consistency of
296 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
297 * ->ap_completion() call-backs are executed under this lock. As we
298 * cannot guarantee that these call-backs never block on all platforms
299 * (as a matter of fact they do block on Mac OS X), type of
300 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
301 * and blocking mutex on Mac OS X. (Alternative is to make this lock
302 * blocking everywhere, but we don't want to slow down fast-path of
303 * our main platform.)
305 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
306 * with client_obd_list_{un,}lock() and
307 * client_obd_list_lock_{init,done}() functions.
309 * NB by Jinshan: though field names are still _loi_, but actually
310 * osc_object{}s are in the list.
312 client_obd_lock_t cl_loi_list_lock;
313 cfs_list_t cl_loi_ready_list;
314 cfs_list_t cl_loi_hp_ready_list;
315 cfs_list_t cl_loi_write_list;
316 cfs_list_t cl_loi_read_list;
319 /* just a sum of the loi/lop pending numbers to be exported by /proc */
320 atomic_t cl_pending_w_pages;
321 atomic_t cl_pending_r_pages;
322 __u32 cl_max_pages_per_rpc;
323 int cl_max_rpcs_in_flight;
324 struct obd_histogram cl_read_rpc_hist;
325 struct obd_histogram cl_write_rpc_hist;
326 struct obd_histogram cl_read_page_hist;
327 struct obd_histogram cl_write_page_hist;
328 struct obd_histogram cl_read_offset_hist;
329 struct obd_histogram cl_write_offset_hist;
331 /* lru for osc caching pages */
332 struct cl_client_cache *cl_cache;
333 cfs_list_t cl_lru_osc; /* member of cl_cache->ccc_lru */
334 atomic_t *cl_lru_left;
335 atomic_t cl_lru_busy;
336 atomic_t cl_lru_shrinkers;
337 atomic_t cl_lru_in_list;
338 cfs_list_t cl_lru_list; /* lru page list */
339 client_obd_lock_t cl_lru_list_lock; /* page list protector */
340 atomic_t cl_unstable_count;
342 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
343 atomic_t cl_destroy_in_flight;
344 wait_queue_head_t cl_destroy_waitq;
346 struct mdc_rpc_lock *cl_rpc_lock;
347 struct mdc_rpc_lock *cl_close_lock;
350 struct semaphore cl_mgc_sem;
351 struct local_oid_storage *cl_mgc_los;
352 struct dt_object *cl_mgc_configs_dir;
353 atomic_t cl_mgc_refcount;
354 struct obd_export *cl_mgc_mgsexp;
356 /* checksumming for data sent over the network */
357 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
358 /* supported checksum types that are worked out at connect time */
359 __u32 cl_supp_cksum_types;
360 /* checksum algorithm to be used */
361 cksum_type_t cl_cksum_type;
363 /* also protected by the poorly named _loi_list_lock lock above */
364 struct osc_async_rc cl_ar;
366 /* used by quotacheck when the servers are older than 2.4 */
367 int cl_qchk_stat; /* quotacheck stat of the peer */
368 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
369 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
370 #warning "please consider removing quotacheck compatibility code"
373 /* sequence manager */
374 struct lu_client_seq *cl_seq;
376 atomic_t cl_resends; /* resend count */
378 /* ptlrpc work for writeback in ptlrpcd context */
379 void *cl_writeback_work;
381 /* hash tables for osc_quota_info */
382 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
384 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
391 struct echo_client_obd {
392 struct obd_export *ec_exp; /* the local connection to osc/lov */
394 cfs_list_t ec_objects;
400 /* Generic subset of OSTs */
402 __u32 *op_array; /* array of index of
404 unsigned int op_count; /* number of OSTs in the array */
405 unsigned int op_size; /* allocated size of lp_array */
406 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
409 /* allow statfs data caching for 1 second */
410 #define OBD_STATFS_CACHE_SECONDS 1
412 struct lov_tgt_desc {
414 struct obd_uuid ltd_uuid;
415 struct obd_device *ltd_obd;
416 struct obd_export *ltd_exp;
418 __u32 ltd_index; /* index in lov_obd->tgts */
419 unsigned long ltd_active:1,/* is this target up for requests */
420 ltd_activate:1,/* should target be activated */
421 ltd_reap:1; /* should this target be deleted */
425 struct lov_desc desc;
426 struct lov_tgt_desc **lov_tgts; /* sparse array */
427 struct ost_pool lov_packed; /* all OSTs in a packed
429 struct mutex lov_lock;
430 struct obd_connect_data lov_ocd;
431 atomic_t lov_refcount;
432 __u32 lov_death_row; /* tgts scheduled to be deleted */
433 __u32 lov_tgt_size; /* size of tgts array */
436 cfs_hash_t *lov_pools_hash_body; /* used for key access */
437 cfs_list_t lov_pool_list; /* used for sequential access */
438 cfs_proc_dir_entry_t *lov_pool_proc_entry;
439 enum lustre_sec_part lov_sp_me;
441 /* Cached LRU and unstable data from upper layer */
444 struct rw_semaphore lov_notify_lock;
447 struct lmv_tgt_desc {
448 struct obd_uuid ltd_uuid;
449 struct obd_export *ltd_exp;
451 struct mutex ltd_fid_mutex;
452 unsigned long ltd_active:1; /* target up for requests */
455 enum placement_policy {
456 PLACEMENT_CHAR_POLICY = 0,
457 PLACEMENT_NID_POLICY = 1,
458 PLACEMENT_INVAL_POLICY = 2,
462 typedef enum placement_policy placement_policy_t;
466 struct lu_client_fld lmv_fld;
468 placement_policy_t lmv_placement;
469 struct lmv_desc desc;
470 struct obd_uuid cluuid;
471 struct obd_export *exp;
473 struct mutex init_mutex;
480 int tgts_size; /* size of tgts array */
481 struct lmv_tgt_desc **tgts;
483 struct obd_connect_data conn_data;
486 struct niobuf_local {
487 __u64 lnb_file_offset;
488 __u32 lnb_page_offset;
492 struct dentry *dentry;
497 #define LUSTRE_FLD_NAME "fld"
498 #define LUSTRE_SEQ_NAME "seq"
500 #define LUSTRE_MDD_NAME "mdd"
501 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
502 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
503 #define LUSTRE_VVP_NAME "vvp"
504 #define LUSTRE_LMV_NAME "lmv"
505 #define LUSTRE_SLP_NAME "slp"
506 #define LUSTRE_LOD_NAME "lod"
507 #define LUSTRE_OSP_NAME "osp"
508 #define LUSTRE_LWP_NAME "lwp"
510 /* obd device type names */
511 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
512 #define LUSTRE_MDS_NAME "mds"
513 #define LUSTRE_MDT_NAME "mdt"
514 #define LUSTRE_MDC_NAME "mdc"
515 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
516 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
517 #define LUSTRE_OSC_NAME "osc"
518 #define LUSTRE_LOV_NAME "lov"
519 #define LUSTRE_MGS_NAME "mgs"
520 #define LUSTRE_MGC_NAME "mgc"
522 #define LUSTRE_ECHO_NAME "obdecho"
523 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
524 #define LUSTRE_QMT_NAME "qmt"
526 /* Constant obd names (post-rename) */
527 #define LUSTRE_MDS_OBDNAME "MDS"
528 #define LUSTRE_OSS_OBDNAME "OSS"
529 #define LUSTRE_MGS_OBDNAME "MGS"
530 #define LUSTRE_MGC_OBDNAME "MGC"
532 static inline int is_osp_on_mdt(char *name)
536 ptr = strrchr(name, '-');
538 CERROR("%s is not a obdname\n", name);
542 /* 1.8 OSC/OSP name on MDT is fsname-OSTxxxx-osc */
543 if (strncmp(ptr + 1, "osc", 3) == 0)
546 if (strncmp(ptr + 1, "MDT", 3) != 0)
549 while (*(--ptr) != '-' && ptr != name);
554 if (strncmp(ptr + 1, LUSTRE_OSP_NAME, strlen(LUSTRE_OSP_NAME)) != 0 &&
555 strncmp(ptr + 1, LUSTRE_OSC_NAME, strlen(LUSTRE_OSC_NAME)) != 0)
561 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
562 #define N_LOCAL_TEMP_PAGE 0x10000000
564 struct obd_trans_info {
567 /* Only used on the server side for tracking acks. */
568 struct oti_req_ack_lock {
569 struct lustre_handle lock;
573 struct llog_cookie oti_onecookie;
574 struct llog_cookie *oti_logcookies;
576 /** synchronous write is needed */
577 unsigned long oti_sync_write:1;
579 /* initial thread handling transaction */
580 struct ptlrpc_thread * oti_thread;
583 __u64 oti_pre_version;
587 struct obd_uuid *oti_ost_uuid;
590 static inline void oti_init(struct obd_trans_info *oti,
591 struct ptlrpc_request *req)
595 memset(oti, 0, sizeof(*oti));
600 oti->oti_xid = req->rq_xid;
601 /** VBR: take versions from request */
602 if (req->rq_reqmsg != NULL &&
603 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
604 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
605 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
606 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
609 /** called from mds_create_objects */
610 if (req->rq_repmsg != NULL)
611 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
612 oti->oti_thread = req->rq_svc_thread;
613 if (req->rq_reqmsg != NULL)
614 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
617 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
622 if (num_cookies == 1)
623 oti->oti_logcookies = &oti->oti_onecookie;
625 OBD_ALLOC_LARGE(oti->oti_logcookies,
626 num_cookies * sizeof(oti->oti_onecookie));
628 oti->oti_numcookies = num_cookies;
631 static inline void oti_free_cookies(struct obd_trans_info *oti)
633 if (!oti || !oti->oti_logcookies)
636 if (oti->oti_logcookies == &oti->oti_onecookie)
637 LASSERT(oti->oti_numcookies == 1);
639 OBD_FREE_LARGE(oti->oti_logcookies,
640 oti->oti_numcookies*sizeof(oti->oti_onecookie));
641 oti->oti_logcookies = NULL;
642 oti->oti_numcookies = 0;
646 * Events signalled through obd_notify() upcall-chain.
648 enum obd_notify_event {
651 /* Device connect start */
653 /* Device activated */
655 /* Device deactivated */
657 /* Device disconnected */
659 /* Connect data for import were changed */
662 OBD_NOTIFY_SYNC_NONBLOCK,
664 /* Configuration event */
666 /* Administratively deactivate/activate event */
667 OBD_NOTIFY_DEACTIVATE,
671 /* bit-mask flags for config events */
673 CONFIG_LOG = 0x1, /* finished processing config log */
674 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
675 CONFIG_TARGET = 0x4 /* one target is added */
679 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
680 * and liblustre being main examples).
682 struct obd_notify_upcall {
683 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
684 enum obd_notify_event ev, void *owner, void *data);
685 /* Opaque datum supplied by upper layer listener */
689 struct target_recovery_data {
690 svc_handler_t trd_recovery_handler;
691 pid_t trd_processing_task;
692 struct completion trd_starting;
693 struct completion trd_finishing;
696 struct obd_llog_group {
698 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
699 wait_queue_head_t olg_waitq;
701 struct mutex olg_cat_processing;
704 /* corresponds to one of the obd's */
705 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
706 #define OBD_DEV_BY_DEVNAME 0xffffd0de
709 struct obd_type *obd_type;
712 /* common and UUID name of this device */
713 char obd_name[MAX_OBD_NAME];
714 struct obd_uuid obd_uuid;
716 struct lu_device *obd_lu_dev;
718 /* bitfield modification is protected by obd_dev_lock */
720 obd_attached:1, /* finished attach */
721 obd_set_up:1, /* finished setup */
722 obd_recovering:1, /* there are recoverable clients */
723 obd_abort_recovery:1, /* recovery expired */
724 obd_version_recov:1, /* obd uses version checking */
725 obd_replayable:1, /* recovery is enabled;
727 obd_no_transno:1, /* no committed-transno notification */
728 obd_no_recov:1, /* fail instead of retry messages */
729 obd_stopping:1, /* started cleanup */
730 obd_starting:1, /* started setup */
731 obd_force:1, /* cleanup with > 0 obd refcount */
732 obd_fail:1, /* cleanup with failover */
733 obd_async_recov:1, /* allow asynchronous orphan cleanup */
734 obd_no_conn:1, /* deny new connections */
735 obd_inactive:1, /* device active/inactive
736 * (for /proc/status only!!) */
737 obd_no_ir:1, /* no imperative recovery. */
738 obd_process_conf:1, /* device is processing mgs config */
739 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
741 /* use separate field as it is set in interrupt to don't mess with
742 * protection of other bits using _bh lock */
743 unsigned long obd_recovery_expired:1;
744 /* uuid-export hash body */
745 cfs_hash_t *obd_uuid_hash;
746 /* nid-export hash body */
747 cfs_hash_t *obd_nid_hash;
749 cfs_hash_t *obd_nid_stats_hash;
750 cfs_list_t obd_nid_stats;
751 atomic_t obd_refcount;
752 cfs_list_t obd_exports;
753 cfs_list_t obd_unlinked_exports;
754 cfs_list_t obd_delayed_exports;
755 struct list_head obd_lwp_list;
757 spinlock_t obd_nid_lock;
758 struct ldlm_namespace *obd_namespace;
759 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
760 /* a spinlock is OK for what we do now, may need a semaphore later */
761 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
762 struct mutex obd_dev_mutex;
763 __u64 obd_last_committed;
764 spinlock_t obd_osfs_lock;
765 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
767 struct lvfs_run_ctxt obd_lvfs_ctxt;
768 struct obd_llog_group obd_olg; /* default llog group */
769 struct obd_device *obd_observer;
770 struct rw_semaphore obd_observer_link_sem;
771 struct obd_notify_upcall obd_upcall;
772 struct obd_export *obd_self_export;
773 struct obd_export *obd_lwp_export;
774 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
775 cfs_list_t obd_exports_timed;
776 time_t obd_eviction_timer; /* for ping evictor */
778 int obd_max_recoverable_clients;
779 atomic_t obd_connected_clients;
780 int obd_stale_clients;
781 int obd_delayed_clients;
782 /* this lock protects all recovery list_heads, timer and
783 * obd_next_recovery_transno value */
784 spinlock_t obd_recovery_task_lock;
785 __u64 obd_next_recovery_transno;
786 int obd_replayed_requests;
787 int obd_requests_queued_for_recovery;
788 wait_queue_head_t obd_next_transno_waitq;
789 /* protected by obd_recovery_task_lock */
790 struct timer_list obd_recovery_timer;
791 time_t obd_recovery_start; /* seconds */
792 time_t obd_recovery_end; /* seconds, for lprocfs_status */
793 int obd_recovery_time_hard;
794 int obd_recovery_timeout;
795 int obd_recovery_ir_factor;
797 /* new recovery stuff from CMD2 */
798 struct target_recovery_data obd_recovery_data;
799 int obd_replayed_locks;
800 atomic_t obd_req_replay_clients;
801 atomic_t obd_lock_replay_clients;
802 /* all lists are protected by obd_recovery_task_lock */
803 cfs_list_t obd_req_replay_queue;
804 cfs_list_t obd_lock_replay_queue;
805 cfs_list_t obd_final_req_queue;
808 #ifdef HAVE_SERVER_SUPPORT
809 struct obd_device_target obt;
810 struct filter_obd filter;
812 struct echo_obd echo;
814 struct client_obd cli;
815 struct echo_client_obd echo_client;
819 /* Fields used by LProcFS */
820 unsigned int obd_cntr_base;
821 struct lprocfs_stats *obd_stats;
823 unsigned int obd_md_cntr_base;
824 struct lprocfs_stats *obd_md_stats;
826 struct proc_dir_entry *obd_proc_entry;
827 struct proc_dir_entry *obd_proc_exports_entry;
828 void *obd_proc_private; /* type private PDEs */
829 struct proc_dir_entry *obd_svc_procroot;
830 struct lprocfs_stats *obd_svc_stats;
831 struct lprocfs_seq_vars *obd_vars;
832 atomic_t obd_evict_inprogress;
833 wait_queue_head_t obd_evict_inprogress_waitq;
834 cfs_list_t obd_evict_list; /* protected with pet_lock */
837 * Ldlm pool part. Save last calculated SLV and Limit.
839 rwlock_t obd_pool_lock;
844 * A list of outstanding class_incref()'s against this obd. For
847 struct lu_ref obd_reference;
849 int obd_conn_inprogress;
852 #define OBD_LLOG_FL_SENDNOW 0x0001
853 #define OBD_LLOG_FL_EXIT 0x0002
855 enum obd_cleanup_stage {
856 /* Special case hack for MDS LOVs */
858 /* can be directly mapped to .ldto_device_fini() */
862 /* get/set_info keys */
863 #define KEY_ASYNC "async"
864 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
865 #define KEY_BLOCKSIZE "blocksize"
866 #define KEY_CAPA_KEY "capa_key"
867 #define KEY_CHANGELOG_CLEAR "changelog_clear"
868 #define KEY_FID2PATH "fid2path"
869 #define KEY_CHECKSUM "checksum"
870 #define KEY_CLEAR_FS "clear_fs"
871 #define KEY_CONN_DATA "conn_data"
872 #define KEY_EVICT_BY_NID "evict_by_nid"
873 #define KEY_FIEMAP "fiemap"
874 #define KEY_FLUSH_CTX "flush_ctx"
875 #define KEY_GRANT_SHRINK "grant_shrink"
876 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
877 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
878 #define KEY_INIT_RECOV "initial_recov"
879 #define KEY_INTERMDS "inter_mds"
880 #define KEY_LAST_ID "last_id"
881 #define KEY_LAST_FID "last_fid"
882 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
883 #define KEY_LOVDESC "lovdesc"
884 #define KEY_LOV_IDX "lov_idx"
885 #define KEY_MAX_EASIZE "max_easize"
886 #define KEY_MDS_CONN "mds_conn"
887 #define KEY_MGSSEC "mgssec"
888 #define KEY_NEXT_ID "next_id"
889 #define KEY_READ_ONLY "read-only"
890 #define KEY_REGISTER_TARGET "register_target"
891 #define KEY_SET_FS "set_fs"
892 #define KEY_TGT_COUNT "tgt_count"
893 /* KEY_SET_INFO in lustre_idl.h */
894 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
895 #define KEY_CONNECT_FLAG "connect_flags"
896 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
898 #define KEY_CACHE_SET "cache_set"
899 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
900 #define KEY_CHANGELOG_INDEX "changelog_index"
901 #define KEY_OSP_CONNECTED "osp_connected"
905 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
906 #define IT_OPEN (1 << 0)
907 #define IT_CREAT (1 << 1)
908 #define IT_READDIR (1 << 2)
909 #define IT_GETATTR (1 << 3)
910 #define IT_LOOKUP (1 << 4)
911 #define IT_UNLINK (1 << 5)
912 #define IT_TRUNC (1 << 6)
913 #define IT_GETXATTR (1 << 7)
914 #define IT_EXEC (1 << 8)
915 #define IT_PIN (1 << 9)
916 #define IT_LAYOUT (1 << 10)
917 #define IT_QUOTA_DQACQ (1 << 11)
918 #define IT_QUOTA_CONN (1 << 12)
919 #define IT_SETXATTR (1 << 13)
921 static inline int it_to_lock_mode(struct lookup_intent *it)
923 /* CREAT needs to be tested before open (both could be set) */
924 if (it->it_op & IT_CREAT)
926 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
929 else if (it->it_op & IT_READDIR)
931 else if (it->it_op & IT_GETXATTR)
933 else if (it->it_op & IT_SETXATTR)
936 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
941 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
942 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
943 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
944 struct lu_fid op_fid4; /* to the operation locks. */
945 mdsno_t op_mds; /* what mds server open will go to */
946 struct lustre_handle op_handle;
947 obd_time op_mod_time;
951 struct lmv_stripe_md *op_mea1;
952 struct lmv_stripe_md *op_mea2;
953 __u32 op_suppgids[2];
959 /* iattr fields and blocks. */
960 struct iattr op_attr;
962 unsigned int op_attr_flags;
965 loff_t op_attr_blocks;
967 /* Size-on-MDS epoch and flags. */
972 struct obd_capa *op_capa1;
973 struct obd_capa *op_capa2;
975 /* Various operation flags. */
976 enum mds_op_bias op_bias;
981 /* Used by readdir */
982 __u64 op_hash_offset;
984 /* Used by readdir */
987 /* used to transfer info between the stacks of MD client
988 * see enum op_cli_flags */
991 /* File object data version for HSM release, on client */
992 __u64 op_data_version;
993 struct lustre_handle op_lease_handle;
996 #define op_stripe_offset op_ioepoch
997 #define op_max_pages op_valid
1000 int (*md_blocking_ast)(struct ldlm_lock *lock,
1001 struct ldlm_lock_desc *desc,
1002 void *data, int flag);
1005 struct md_enqueue_info;
1006 /* metadata stat-ahead */
1007 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1008 struct md_enqueue_info *minfo,
1011 struct md_enqueue_info {
1012 struct md_op_data mi_data;
1013 struct lookup_intent mi_it;
1014 struct lustre_handle mi_lockh;
1015 struct inode *mi_dir;
1016 md_enqueue_cb_t mi_cb;
1018 unsigned int mi_generation;
1022 struct module *o_owner;
1023 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1024 void *karg, void *uarg);
1025 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1026 __u32 keylen, void *key, __u32 *vallen, void *val,
1027 struct lov_stripe_md *lsm);
1028 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1029 __u32 keylen, void *key,
1030 __u32 vallen, void *val,
1031 struct ptlrpc_request_set *set);
1032 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1033 int (*o_detach)(struct obd_device *dev);
1034 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1035 int (*o_precleanup)(struct obd_device *dev,
1036 enum obd_cleanup_stage cleanup_stage);
1037 int (*o_cleanup)(struct obd_device *dev);
1038 int (*o_process_config)(struct obd_device *dev, obd_count len,
1040 int (*o_postrecov)(struct obd_device *dev);
1041 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1043 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1044 /* connect to the target device with given connection
1045 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1046 * granted by the target, which are guaranteed to be a subset of flags
1047 * asked for. If @ocd == NULL, use default parameters. */
1048 int (*o_connect)(const struct lu_env *env,
1049 struct obd_export **exp, struct obd_device *src,
1050 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1052 int (*o_reconnect)(const struct lu_env *env,
1053 struct obd_export *exp, struct obd_device *src,
1054 struct obd_uuid *cluuid,
1055 struct obd_connect_data *ocd,
1057 int (*o_disconnect)(struct obd_export *exp);
1059 /* Initialize/finalize fids infrastructure. */
1060 int (*o_fid_init)(struct obd_device *obd,
1061 struct obd_export *exp, enum lu_cli_type type);
1062 int (*o_fid_fini)(struct obd_device *obd);
1064 /* Allocate new fid according to passed @hint. */
1065 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1066 struct md_op_data *op_data);
1069 * Object with @fid is getting deleted, we may want to do something
1072 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1073 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1074 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1075 __u64 max_age, struct ptlrpc_request_set *set);
1076 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1077 struct lov_stripe_md *mem_src);
1078 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1079 struct lov_mds_md *disk_src, int disk_len);
1080 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1082 /* FIXME: add fid capability support for create & destroy! */
1083 int (*o_precreate)(struct obd_export *exp);
1084 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1085 struct obdo *oa, struct lov_stripe_md **ea,
1086 struct obd_trans_info *oti);
1087 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1088 struct lov_stripe_md **ea,
1089 struct obd_trans_info *oti);
1090 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1091 struct obdo *oa, struct lov_stripe_md *ea,
1092 struct obd_trans_info *oti, struct obd_export *md_exp,
1094 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1095 struct obd_info *oinfo, struct obd_trans_info *oti);
1096 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1097 struct obd_trans_info *oti,
1098 struct ptlrpc_request_set *rqset);
1099 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1100 struct obd_info *oinfo);
1101 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1102 struct ptlrpc_request_set *set);
1103 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1104 obd_count oa_bufs, struct brw_page *pgarr,
1105 struct obd_trans_info *oti);
1106 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1107 struct ost_lvb *lvb, int kms_only);
1108 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1109 obd_off size, int shrink);
1110 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1111 struct obd_info *oinfo, struct obd_trans_info *oti,
1112 struct ptlrpc_request_set *rqset);
1113 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1114 struct obd_info *oinfo, obd_size start, obd_size end,
1115 struct ptlrpc_request_set *set);
1116 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1117 struct lov_stripe_md *src, obd_size start,
1118 obd_size end, struct obd_trans_info *oti);
1119 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1120 struct lustre_handle *srconn, struct lov_stripe_md *src,
1121 obd_size start, obd_size end, struct obd_trans_info *);
1122 int (*o_iterate)(struct lustre_handle *conn,
1123 int (*)(obd_id, obd_seq, void *),
1124 obd_id *startid, obd_seq seq, void *data);
1125 int (*o_preprw)(const struct lu_env *env, int cmd,
1126 struct obd_export *exp, struct obdo *oa, int objcount,
1127 struct obd_ioobj *obj, struct niobuf_remote *remote,
1128 int *nr_pages, struct niobuf_local *local,
1129 struct obd_trans_info *oti, struct lustre_capa *capa);
1130 int (*o_commitrw)(const struct lu_env *env, int cmd,
1131 struct obd_export *exp, struct obdo *oa,
1132 int objcount, struct obd_ioobj *obj,
1133 struct niobuf_remote *remote, int pages,
1134 struct niobuf_local *local,
1135 struct obd_trans_info *oti, int rc);
1136 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1137 struct ldlm_enqueue_info *einfo,
1138 struct ptlrpc_request_set *rqset);
1139 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1140 ldlm_iterator_t it, void *data);
1141 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1142 ldlm_iterator_t it, void *data);
1143 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1144 __u32 mode, struct lustre_handle *);
1145 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1146 ldlm_cancel_flags_t flags, void *opaque);
1147 int (*o_init_export)(struct obd_export *exp);
1148 int (*o_destroy_export)(struct obd_export *exp);
1149 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1150 int cmd, obd_off *);
1152 /* llog related obd_methods */
1153 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1154 struct obd_device *disk_obd, int *idx);
1155 int (*o_llog_finish)(struct obd_device *obd, int count);
1156 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1158 /* metadata-only methods */
1159 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1160 struct obd_capa *, struct obd_client_handle *, int flag);
1161 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1163 int (*o_import_event)(struct obd_device *, struct obd_import *,
1164 enum obd_import_event);
1166 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1167 enum obd_notify_event ev, void *data);
1169 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1170 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1173 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1174 struct obd_quotactl *);
1175 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1176 struct obd_quotactl *);
1178 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1181 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1182 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1183 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1185 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1187 void (*o_getref)(struct obd_device *obd);
1188 void (*o_putref)(struct obd_device *obd);
1190 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1191 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1192 * Also, add a wrapper function in include/linux/obd_class.h. */
1196 LUSTRE_OPC_MKDIR = (1 << 0),
1197 LUSTRE_OPC_SYMLINK = (1 << 1),
1198 LUSTRE_OPC_MKNOD = (1 << 2),
1199 LUSTRE_OPC_CREATE = (1 << 3),
1200 LUSTRE_OPC_ANY = (1 << 4)
1203 /* lmv structures */
1205 struct mdt_body *body;
1206 struct lov_stripe_md *lsm;
1207 struct lmv_stripe_md *lmv;
1208 #ifdef CONFIG_FS_POSIX_ACL
1209 struct posix_acl *posix_acl;
1211 struct mdt_remote_perm *remote_perm;
1212 struct obd_capa *mds_capa;
1213 struct obd_capa *oss_capa;
1217 struct md_open_data {
1218 struct obd_client_handle *mod_och;
1219 struct ptlrpc_request *mod_open_req;
1220 struct ptlrpc_request *mod_close_req;
1221 atomic_t mod_refcount;
1225 struct lookup_intent;
1229 /* Every operation from MD_STATS_FIRST_OP up to and including
1230 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1231 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1232 * Operations after MD_STATS_LAST_OP are excluded from stats.
1233 * There are a few reasons for doing this: we prune the 17
1234 * counters which will be of minimal use in understanding
1235 * metadata utilization, we save memory by allocating 15
1236 * instead of 32 counters, we save cycles by not counting.
1238 * MD_STATS_FIRST_OP must be the first member of md_ops.
1240 #define MD_STATS_FIRST_OP m_close
1241 int (*m_close)(struct obd_export *, struct md_op_data *,
1242 struct md_open_data *, struct ptlrpc_request **);
1244 int (*m_create)(struct obd_export *, struct md_op_data *,
1245 const void *, int, int, __u32, __u32, cfs_cap_t,
1246 __u64, struct ptlrpc_request **);
1248 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1249 struct lookup_intent *, struct md_op_data *,
1250 struct lustre_handle *, void *, int,
1251 struct ptlrpc_request **, __u64);
1253 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1254 struct ptlrpc_request **);
1256 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1257 void *, int, struct lookup_intent *, int,
1258 struct ptlrpc_request **,
1259 ldlm_blocking_callback, __u64);
1261 int (*m_link)(struct obd_export *, struct md_op_data *,
1262 struct ptlrpc_request **);
1264 int (*m_rename)(struct obd_export *, struct md_op_data *,
1265 const char *, int, const char *, int,
1266 struct ptlrpc_request **);
1268 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1269 int , void *, int, struct ptlrpc_request **,
1270 struct md_open_data **mod);
1272 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1273 struct obd_capa *, struct ptlrpc_request **);
1275 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1276 struct page **, struct ptlrpc_request **);
1278 int (*m_read_entry)(struct obd_export *, struct md_op_data *,
1279 struct md_callback *cb_op, struct lu_dirent **ld,
1280 struct page **ppage);
1282 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1283 struct ptlrpc_request **);
1285 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1286 struct obd_capa *, obd_valid, const char *,
1287 const char *, int, int, int, __u32,
1288 struct ptlrpc_request **);
1290 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1291 struct obd_capa *, obd_valid, const char *,
1292 const char *, int, int, int,
1293 struct ptlrpc_request **);
1295 int (*m_intent_getattr_async)(struct obd_export *,
1296 struct md_enqueue_info *,
1297 struct ldlm_enqueue_info *);
1299 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1300 struct lu_fid *, __u64 *bits);
1301 #define MD_STATS_LAST_OP m_revalidate_lock
1303 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1304 struct obd_capa **);
1306 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1308 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1309 ldlm_iterator_t, void *);
1311 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1312 struct md_open_data *);
1314 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1315 struct ptlrpc_request **);
1317 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1318 const struct lu_fid *,
1319 struct ptlrpc_request **);
1321 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1323 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1324 struct obd_export *, struct obd_export *,
1325 struct lustre_md *);
1327 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1329 int (*m_merge_attr)(struct obd_export *,
1330 const struct lmv_stripe_md *lsm,
1331 struct cl_attr *attr);
1333 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1334 struct mdt_body *, ldlm_blocking_callback);
1336 int (*m_set_open_replay_data)(struct obd_export *,
1337 struct obd_client_handle *,
1338 struct lookup_intent *);
1340 int (*m_clear_open_replay_data)(struct obd_export *,
1341 struct obd_client_handle *);
1343 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1345 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1346 const struct lu_fid *, ldlm_type_t,
1347 ldlm_policy_data_t *, ldlm_mode_t,
1348 struct lustre_handle *);
1350 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1351 ldlm_policy_data_t *, ldlm_mode_t,
1352 ldlm_cancel_flags_t flags, void *opaque);
1354 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1355 renew_capa_cb_t cb);
1357 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1358 const struct req_msg_field *, struct obd_capa **);
1360 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1361 struct obd_capa *, __u32,
1362 struct ptlrpc_request **);
1365 struct lsm_operations {
1366 void (*lsm_free)(struct lov_stripe_md *);
1367 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1368 struct obd_export *md_exp);
1369 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1371 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1373 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1374 __u16 *stripe_count);
1375 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1376 struct lov_mds_md *lmm);
1379 extern const struct lsm_operations lsm_v1_ops;
1380 extern const struct lsm_operations lsm_v3_ops;
1381 static inline const struct lsm_operations *lsm_op_find(int magic)
1389 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1394 /* Requests for obd_extent_calc() */
1395 #define OBD_CALC_STRIPE_START 1
1396 #define OBD_CALC_STRIPE_END 2
1398 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1400 return oinfo->oi_capa;
1403 static inline struct md_open_data *obd_mod_alloc(void)
1405 struct md_open_data *mod;
1409 atomic_set(&mod->mod_refcount, 1);
1413 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1414 #define obd_mod_put(mod) \
1416 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1417 if ((mod)->mod_open_req) \
1418 ptlrpc_req_finished((mod)->mod_open_req); \
1419 OBD_FREE_PTR(mod); \
1423 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1424 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1426 /* return 1 if client should be resend request */
1427 static inline int client_should_resend(int resend, struct client_obd *cli)
1429 return atomic_read(&cli->cl_resends) ?
1430 atomic_read(&cli->cl_resends) > resend : 1;
1434 * Return device name for this device
1436 * XXX: lu_device is declared before obd_device, while a pointer pointing
1437 * back to obd_device in lu_device, so this helper function defines here
1438 * instead of in lu_object.h
1440 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1442 return lu_dev->ld_obd->obd_name;
1445 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1450 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1453 /* caller does not care of idx */
1457 /* volatile file, the MDT can be set from name */
1458 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1459 /* if no MDT is specified, use std way */
1460 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1462 /* test for no MDT idx case */
1463 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1464 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1468 /* we have an idx, read it */
1469 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1470 *idx = strtoul(start, &end, 16);
1472 * no digit, no trailing :, negative value
1474 if (((*idx == 0) && (end == start)) ||
1475 (*end != ':') || (*idx < 0))
1480 /* bad format of mdt idx, we cannot return an error
1481 * to caller so we use hash algo */
1482 CERROR("Bad volatile file name format: %s\n",
1483 name + LUSTRE_VOLATILE_HDR_LEN);
1487 static inline int cli_brw_size(struct obd_device *obd)
1489 LASSERT(obd != NULL);
1490 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1493 #endif /* __OBD_H */