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 # error "userspace should not include <obd.h>"
43 #if defined(__linux__)
44 #include <linux/obd.h>
45 #elif defined(__APPLE__)
46 #include <darwin/obd.h>
47 #elif defined(__WINNT__)
48 #include <winnt/obd.h>
50 #error Unsupported operating system.
53 #include <lustre/lustre_idl.h>
54 #include <lustre_lib.h>
55 #include <libcfs/bitmap.h>
56 #ifdef HAVE_SERVER_SUPPORT
57 # include <lu_target.h>
58 # include <obd_target.h>
61 #include <lustre_export.h>
62 #include <lustre_fid.h>
63 #include <lustre_fld.h>
64 #include <lustre_capa.h>
67 #define MAX_OBD_DEVICES 8192
75 struct lov_oinfo { /* per-stripe data structure */
76 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
77 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
78 int loi_ost_gen; /* generation of this loi_ost_idx */
80 unsigned long loi_kms_valid:1;
81 __u64 loi_kms; /* known minimum size */
82 struct ost_lvb loi_lvb;
83 struct osc_async_rc loi_ar;
86 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
89 oinfo->loi_kms_valid = 1;
92 static inline void loi_init(struct lov_oinfo *loi)
96 struct lov_stripe_md {
99 pid_t lsm_lock_owner; /* debugging */
101 /* maximum possible file size, might change as OSTs status changes,
102 * e.g. disconnected, deactivated */
105 /* Public members. */
106 struct ost_id lw_object_oi; /* lov object id/seq */
108 /* LOV-private members start here -- only for use in lov/. */
110 __u32 lw_stripe_size; /* size of the stripe */
111 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
112 __u16 lw_stripe_count; /* number of objects being striped over */
113 __u16 lw_layout_gen; /* generation of the layout */
114 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
117 struct lov_oinfo *lsm_oinfo[0];
120 #define lsm_oi lsm_wire.lw_object_oi
121 #define lsm_magic lsm_wire.lw_magic
122 #define lsm_layout_gen lsm_wire.lw_layout_gen
123 #define lsm_stripe_size lsm_wire.lw_stripe_size
124 #define lsm_pattern lsm_wire.lw_pattern
125 #define lsm_stripe_count lsm_wire.lw_stripe_count
126 #define lsm_pool_name lsm_wire.lw_pool_name
128 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
130 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
133 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
137 if (lsm_is_released(lsm))
142 static inline int lov_stripe_md_size(unsigned int stripe_count)
144 struct lov_stripe_md lsm;
146 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
151 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
153 /* obd info for a particular level (lov, osc). */
155 /* Lock policy. It keeps an extent which is specific for a particular
156 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
157 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
158 ldlm_policy_data_t oi_policy;
159 /* Flags used for set request specific flags:
160 - while lock handling, the flags obtained on the enqueue
161 request are set here.
162 - while stats, the flags used for control delay/resend.
163 - while setattr, the flags used for distinguish punch operation
166 /* lsm data specific for every OSC. */
167 struct lov_stripe_md *oi_md;
168 /* obdo data specific for every OSC, if needed at all. */
170 /* statfs data specific for every OSC, if needed at all. */
171 struct obd_statfs *oi_osfs;
172 /* An update callback which is called to update some data on upper
173 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
174 * request in osc level for enqueue requests. It is also possible to
175 * update some caller data from LOV layer if needed. */
176 obd_enqueue_update_f oi_cb_up;
177 /* oss capability, its type is obd_capa in client to avoid copy.
178 * in contrary its type is lustre_capa in OSS. */
183 struct list_head typ_chain;
184 struct obd_ops *typ_dt_ops;
185 struct md_ops *typ_md_ops;
186 struct proc_dir_entry *typ_procroot;
187 struct proc_dir_entry *typ_procsym;
188 __u32 typ_sym_filter;
191 struct lu_device_type *typ_lu;
192 spinlock_t obd_type_lock;
204 LLOG_CONFIG_ORIG_CTXT = 0,
205 LLOG_CONFIG_REPL_CTXT,
206 LLOG_MDS_OST_ORIG_CTXT,
210 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
211 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
212 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
213 LLOG_AGENT_ORIG_CTXT, /**< agent requests generation on cdt */
217 struct timeout_item {
218 enum timeout_event ti_event;
219 cfs_time_t ti_timeout;
222 cfs_list_t ti_obd_list;
226 #define OBD_MAX_RIF_DEFAULT 8
227 #define OBD_MAX_RIF_MAX 512
228 #define OSC_MAX_RIF_MAX 256
229 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
230 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
231 #define OSC_DEFAULT_RESENDS 10
233 /* possible values for fo_sync_lock_cancel */
235 NEVER_SYNC_ON_CANCEL = 0,
236 BLOCKING_SYNC_ON_CANCEL = 1,
237 ALWAYS_SYNC_ON_CANCEL = 2,
238 NUM_SYNC_ON_CANCEL_STATES
244 struct rw_semaphore cl_sem;
245 struct obd_uuid cl_target_uuid;
246 struct obd_import *cl_import; /* ptlrpc connection state */
248 /* max_mds_easize is purely a performance thing so we don't have to
249 * call obd_size_diskmd() all the time. */
250 int cl_default_mds_easize;
251 int cl_max_mds_easize;
252 int cl_default_mds_cookiesize;
253 int cl_max_mds_cookiesize;
255 enum lustre_sec_part cl_sp_me;
256 enum lustre_sec_part cl_sp_to;
257 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
259 /* the grant values are protected by loi_list_lock below */
260 long cl_dirty; /* all _dirty_ in bytes */
261 long cl_dirty_max; /* allowed w/o rpc */
262 long cl_dirty_transit; /* dirty synchronous */
263 long cl_avail_grant; /* bytes of credit for ost */
264 long cl_lost_grant; /* lost credits (trunc) */
266 /* since we allocate grant by blocks, we don't know how many grant will
267 * be used to add a page into cache. As a solution, we reserve maximum
268 * grant before trying to dirty a page and unreserve the rest.
269 * See osc_{reserve|unreserve}_grant for details. */
270 long cl_reserved_grant;
271 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
272 cfs_time_t cl_next_shrink_grant; /* jiffies */
273 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
274 int cl_grant_shrink_interval; /* seconds */
276 /* A chunk is an optimal size used by osc_extent to determine
277 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
279 int cl_extent_tax; /* extent overhead, by bytes */
281 /* keep track of objects that have lois that contain pages which
282 * have been queued for async brw. this lock also protects the
283 * lists of osc_client_pages that hang off of the loi */
285 * ->cl_loi_list_lock protects consistency of
286 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
287 * ->ap_completion() call-backs are executed under this lock. As we
288 * cannot guarantee that these call-backs never block on all platforms
289 * (as a matter of fact they do block on Mac OS X), type of
290 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
291 * and blocking mutex on Mac OS X. (Alternative is to make this lock
292 * blocking everywhere, but we don't want to slow down fast-path of
293 * our main platform.)
295 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
296 * with client_obd_list_{un,}lock() and
297 * client_obd_list_lock_{init,done}() functions.
299 * NB by Jinshan: though field names are still _loi_, but actually
300 * osc_object{}s are in the list.
302 client_obd_lock_t cl_loi_list_lock;
303 cfs_list_t cl_loi_ready_list;
304 cfs_list_t cl_loi_hp_ready_list;
305 cfs_list_t cl_loi_write_list;
306 cfs_list_t cl_loi_read_list;
309 /* just a sum of the loi/lop pending numbers to be exported by /proc */
310 atomic_t cl_pending_w_pages;
311 atomic_t cl_pending_r_pages;
312 __u32 cl_max_pages_per_rpc;
313 int cl_max_rpcs_in_flight;
314 struct obd_histogram cl_read_rpc_hist;
315 struct obd_histogram cl_write_rpc_hist;
316 struct obd_histogram cl_read_page_hist;
317 struct obd_histogram cl_write_page_hist;
318 struct obd_histogram cl_read_offset_hist;
319 struct obd_histogram cl_write_offset_hist;
321 /* lru for osc caching pages */
322 struct cl_client_cache *cl_cache;
323 cfs_list_t cl_lru_osc; /* member of cl_cache->ccc_lru */
324 atomic_t *cl_lru_left;
325 atomic_t cl_lru_busy;
326 atomic_t cl_lru_shrinkers;
327 atomic_t cl_lru_in_list;
328 cfs_list_t cl_lru_list; /* lru page list */
329 client_obd_lock_t cl_lru_list_lock; /* page list protector */
330 atomic_t cl_unstable_count;
332 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
333 atomic_t cl_destroy_in_flight;
334 wait_queue_head_t cl_destroy_waitq;
336 struct mdc_rpc_lock *cl_rpc_lock;
337 struct mdc_rpc_lock *cl_close_lock;
340 struct mutex cl_mgc_mutex;
341 struct local_oid_storage *cl_mgc_los;
342 struct dt_object *cl_mgc_configs_dir;
343 atomic_t cl_mgc_refcount;
344 struct obd_export *cl_mgc_mgsexp;
346 /* checksumming for data sent over the network */
347 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
348 /* supported checksum types that are worked out at connect time */
349 __u32 cl_supp_cksum_types;
350 /* checksum algorithm to be used */
351 cksum_type_t cl_cksum_type;
353 /* also protected by the poorly named _loi_list_lock lock above */
354 struct osc_async_rc cl_ar;
356 /* used by quotacheck when the servers are older than 2.4 */
357 int cl_qchk_stat; /* quotacheck stat of the peer */
358 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
359 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
360 #warning "please consider removing quotacheck compatibility code"
363 /* sequence manager */
364 struct lu_client_seq *cl_seq;
366 atomic_t cl_resends; /* resend count */
368 /* ptlrpc work for writeback in ptlrpcd context */
369 void *cl_writeback_work;
371 /* hash tables for osc_quota_info */
372 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
374 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
381 struct echo_client_obd {
382 struct obd_export *ec_exp; /* the local connection to osc/lov */
384 cfs_list_t ec_objects;
390 /* Generic subset of OSTs */
392 __u32 *op_array; /* array of index of
394 unsigned int op_count; /* number of OSTs in the array */
395 unsigned int op_size; /* allocated size of lp_array */
396 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
399 /* allow statfs data caching for 1 second */
400 #define OBD_STATFS_CACHE_SECONDS 1
402 struct lov_tgt_desc {
404 struct obd_uuid ltd_uuid;
405 struct obd_device *ltd_obd;
406 struct obd_export *ltd_exp;
408 __u32 ltd_index; /* index in lov_obd->tgts */
409 unsigned long ltd_active:1,/* is this target up for requests */
410 ltd_activate:1,/* should target be activated */
411 ltd_reap:1; /* should this target be deleted */
415 struct lov_desc desc;
416 struct lov_tgt_desc **lov_tgts; /* sparse array */
417 struct ost_pool lov_packed; /* all OSTs in a packed
419 struct mutex lov_lock;
420 struct obd_connect_data lov_ocd;
421 struct proc_dir_entry *targets_proc_entry;
422 atomic_t lov_refcount;
423 __u32 lov_death_row; /* tgts scheduled to be deleted */
424 __u32 lov_tgt_size; /* size of tgts array */
427 cfs_hash_t *lov_pools_hash_body; /* used for key access */
428 struct list_head lov_pool_list; /* used for sequential access */
429 struct proc_dir_entry *lov_pool_proc_entry;
430 enum lustre_sec_part lov_sp_me;
432 /* Cached LRU and unstable data from upper layer */
435 struct rw_semaphore lov_notify_lock;
438 struct lmv_tgt_desc {
439 struct obd_uuid ltd_uuid;
440 struct obd_export *ltd_exp;
442 struct mutex ltd_fid_mutex;
443 unsigned long ltd_active:1; /* target up for requests */
446 enum placement_policy {
447 PLACEMENT_CHAR_POLICY = 0,
448 PLACEMENT_NID_POLICY = 1,
449 PLACEMENT_INVAL_POLICY = 2,
453 typedef enum placement_policy placement_policy_t;
457 struct lu_client_fld lmv_fld;
459 placement_policy_t lmv_placement;
460 struct lmv_desc desc;
461 struct obd_uuid cluuid;
462 struct obd_export *exp;
463 struct proc_dir_entry *targets_proc_entry;
465 struct mutex init_mutex;
470 int max_def_cookiesize;
473 int tgts_size; /* size of tgts array */
474 struct lmv_tgt_desc **tgts;
476 struct obd_connect_data conn_data;
479 struct niobuf_local {
480 __u64 lnb_file_offset;
481 __u32 lnb_page_offset;
485 struct dentry *dentry;
489 #define LUSTRE_FLD_NAME "fld"
490 #define LUSTRE_SEQ_NAME "seq"
492 #define LUSTRE_MDD_NAME "mdd"
493 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
494 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
495 #define LUSTRE_VVP_NAME "vvp"
496 #define LUSTRE_LMV_NAME "lmv"
497 #define LUSTRE_SLP_NAME "slp"
498 #define LUSTRE_LOD_NAME "lod"
499 #define LUSTRE_OSP_NAME "osp"
500 #define LUSTRE_LWP_NAME "lwp"
502 /* obd device type names */
503 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
504 #define LUSTRE_MDS_NAME "mds"
505 #define LUSTRE_MDT_NAME "mdt"
506 #define LUSTRE_MDC_NAME "mdc"
507 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
508 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
509 #define LUSTRE_OSC_NAME "osc"
510 #define LUSTRE_LOV_NAME "lov"
511 #define LUSTRE_MGS_NAME "mgs"
512 #define LUSTRE_MGC_NAME "mgc"
514 #define LUSTRE_ECHO_NAME "obdecho"
515 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
516 #define LUSTRE_QMT_NAME "qmt"
518 /* Constant obd names (post-rename) */
519 #define LUSTRE_MDS_OBDNAME "MDS"
520 #define LUSTRE_OSS_OBDNAME "OSS"
521 #define LUSTRE_MGS_OBDNAME "MGS"
522 #define LUSTRE_MGC_OBDNAME "MGC"
524 struct obd_trans_info {
526 /* Only used on the server side for tracking acks. */
527 struct oti_req_ack_lock {
528 struct lustre_handle lock;
532 struct llog_cookie oti_onecookie;
533 struct llog_cookie *oti_logcookies;
536 __u64 oti_pre_version;
540 * Events signalled through obd_notify() upcall-chain.
542 enum obd_notify_event {
545 /* Device connect start */
547 /* Device activated */
549 /* Device deactivated */
551 /* Device disconnected */
553 /* Connect data for import were changed */
556 OBD_NOTIFY_SYNC_NONBLOCK,
558 /* Configuration event */
560 /* Administratively deactivate/activate event */
561 OBD_NOTIFY_DEACTIVATE,
565 /* bit-mask flags for config events */
567 CONFIG_LOG = 0x1, /* finished processing config log */
568 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
569 CONFIG_TARGET = 0x4 /* one target is added */
573 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
574 * and liblustre being main examples).
576 struct obd_notify_upcall {
577 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
578 enum obd_notify_event ev, void *owner, void *data);
579 /* Opaque datum supplied by upper layer listener */
583 struct target_recovery_data {
584 svc_handler_t trd_recovery_handler;
585 pid_t trd_processing_task;
586 struct completion trd_starting;
587 struct completion trd_finishing;
590 struct obd_llog_group {
591 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
592 wait_queue_head_t olg_waitq;
594 struct mutex olg_cat_processing;
597 /* corresponds to one of the obd's */
598 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
601 struct obd_type *obd_type;
604 /* common and UUID name of this device */
605 char obd_name[MAX_OBD_NAME];
606 struct obd_uuid obd_uuid;
608 struct lu_device *obd_lu_dev;
610 /* bitfield modification is protected by obd_dev_lock */
612 obd_attached:1, /* finished attach */
613 obd_set_up:1, /* finished setup */
614 obd_recovering:1, /* there are recoverable clients */
615 obd_abort_recovery:1, /* recovery expired */
616 obd_version_recov:1, /* obd uses version checking */
617 obd_replayable:1, /* recovery is enabled;
619 obd_no_transno:1, /* no committed-transno notification */
620 obd_no_recov:1, /* fail instead of retry messages */
621 obd_stopping:1, /* started cleanup */
622 obd_starting:1, /* started setup */
623 obd_force:1, /* cleanup with > 0 obd refcount */
624 obd_fail:1, /* cleanup with failover */
625 obd_no_conn:1, /* deny new connections */
626 obd_inactive:1, /* device active/inactive
627 * (for /proc/status only!!) */
628 obd_no_ir:1, /* no imperative recovery. */
629 obd_process_conf:1, /* device is processing mgs config */
630 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
632 /* use separate field as it is set in interrupt to don't mess with
633 * protection of other bits using _bh lock */
634 unsigned long obd_recovery_expired:1;
635 /* uuid-export hash body */
636 cfs_hash_t *obd_uuid_hash;
637 /* nid-export hash body */
638 cfs_hash_t *obd_nid_hash;
640 cfs_hash_t *obd_nid_stats_hash;
641 cfs_list_t obd_nid_stats;
642 atomic_t obd_refcount;
643 cfs_list_t obd_exports;
644 cfs_list_t obd_unlinked_exports;
645 cfs_list_t obd_delayed_exports;
646 struct list_head obd_lwp_list;
648 spinlock_t obd_nid_lock;
649 struct ldlm_namespace *obd_namespace;
650 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
651 /* a spinlock is OK for what we do now, may need a semaphore later */
652 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
653 struct mutex obd_dev_mutex;
654 __u64 obd_last_committed;
655 spinlock_t obd_osfs_lock;
656 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
658 struct lvfs_run_ctxt obd_lvfs_ctxt;
659 struct obd_llog_group obd_olg; /* default llog group */
660 struct obd_device *obd_observer;
661 struct rw_semaphore obd_observer_link_sem;
662 struct obd_notify_upcall obd_upcall;
663 struct obd_export *obd_self_export;
664 struct obd_export *obd_lwp_export;
665 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
666 cfs_list_t obd_exports_timed;
667 time_t obd_eviction_timer; /* for ping evictor */
669 int obd_max_recoverable_clients;
670 atomic_t obd_connected_clients;
671 int obd_stale_clients;
672 /* this lock protects all recovery list_heads, timer and
673 * obd_next_recovery_transno value */
674 spinlock_t obd_recovery_task_lock;
675 __u64 obd_next_recovery_transno;
676 int obd_replayed_requests;
677 int obd_requests_queued_for_recovery;
678 wait_queue_head_t obd_next_transno_waitq;
679 /* protected by obd_recovery_task_lock */
680 struct timer_list obd_recovery_timer;
681 time_t obd_recovery_start; /* seconds */
682 time_t obd_recovery_end; /* seconds, for lprocfs_status */
683 int obd_recovery_time_hard;
684 int obd_recovery_timeout;
685 int obd_recovery_ir_factor;
687 /* new recovery stuff from CMD2 */
688 struct target_recovery_data obd_recovery_data;
689 int obd_replayed_locks;
690 atomic_t obd_req_replay_clients;
691 atomic_t obd_lock_replay_clients;
692 /* all lists are protected by obd_recovery_task_lock */
693 cfs_list_t obd_req_replay_queue;
694 cfs_list_t obd_lock_replay_queue;
695 cfs_list_t obd_final_req_queue;
698 #ifdef HAVE_SERVER_SUPPORT
699 struct obd_device_target obt;
700 struct filter_obd filter;
702 struct echo_obd echo;
704 struct client_obd cli;
705 struct echo_client_obd echo_client;
709 /* Fields used by LProcFS */
710 unsigned int obd_cntr_base;
711 struct lprocfs_stats *obd_stats;
713 unsigned int obd_md_cntr_base;
714 struct lprocfs_stats *obd_md_stats;
716 struct proc_dir_entry *obd_proc_entry;
717 struct proc_dir_entry *obd_proc_exports_entry;
718 struct proc_dir_entry *obd_svc_procroot;
719 struct lprocfs_stats *obd_svc_stats;
720 struct lprocfs_seq_vars *obd_vars;
721 atomic_t obd_evict_inprogress;
722 wait_queue_head_t obd_evict_inprogress_waitq;
723 cfs_list_t obd_evict_list; /* protected with pet_lock */
726 * Ldlm pool part. Save last calculated SLV and Limit.
728 rwlock_t obd_pool_lock;
733 * A list of outstanding class_incref()'s against this obd. For
736 struct lu_ref obd_reference;
738 int obd_conn_inprogress;
741 enum obd_cleanup_stage {
742 /* Special case hack for MDS LOVs */
744 /* can be directly mapped to .ldto_device_fini() */
748 /* get/set_info keys */
749 #define KEY_ASYNC "async"
750 #define KEY_CAPA_KEY "capa_key"
751 #define KEY_CHANGELOG_CLEAR "changelog_clear"
752 #define KEY_FID2PATH "fid2path"
753 #define KEY_CHECKSUM "checksum"
754 #define KEY_CLEAR_FS "clear_fs"
755 #define KEY_CONN_DATA "conn_data"
756 #define KEY_EVICT_BY_NID "evict_by_nid"
757 #define KEY_FIEMAP "fiemap"
758 #define KEY_FLUSH_CTX "flush_ctx"
759 #define KEY_GRANT_SHRINK "grant_shrink"
760 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
761 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
762 #define KEY_INTERMDS "inter_mds"
763 #define KEY_LAST_ID "last_id"
764 #define KEY_LAST_FID "last_fid"
765 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
766 #define KEY_LOVDESC "lovdesc"
767 #define KEY_MAX_EASIZE "max_easize"
768 #define KEY_DEFAULT_EASIZE "default_easize"
769 #define KEY_MAX_COOKIESIZE "max_cookiesize"
770 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
771 #define KEY_MDS_CONN "mds_conn"
772 #define KEY_MGSSEC "mgssec"
773 #define KEY_NEXT_ID "next_id"
774 #define KEY_READ_ONLY "read-only"
775 #define KEY_REGISTER_TARGET "register_target"
776 #define KEY_SET_FS "set_fs"
777 #define KEY_TGT_COUNT "tgt_count"
778 /* KEY_SET_INFO in lustre_idl.h */
779 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
780 #define KEY_CONNECT_FLAG "connect_flags"
782 #define KEY_CACHE_SET "cache_set"
783 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
784 #define KEY_OSP_CONNECTED "osp_connected"
788 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
789 #define IT_OPEN (1 << 0)
790 #define IT_CREAT (1 << 1)
791 #define IT_READDIR (1 << 2)
792 #define IT_GETATTR (1 << 3)
793 #define IT_LOOKUP (1 << 4)
794 #define IT_UNLINK (1 << 5)
795 #define IT_TRUNC (1 << 6)
796 #define IT_GETXATTR (1 << 7)
797 #define IT_EXEC (1 << 8)
798 #define IT_PIN (1 << 9)
799 #define IT_LAYOUT (1 << 10)
800 #define IT_QUOTA_DQACQ (1 << 11)
801 #define IT_QUOTA_CONN (1 << 12)
802 #define IT_SETXATTR (1 << 13)
804 static inline int it_to_lock_mode(struct lookup_intent *it)
806 /* CREAT needs to be tested before open (both could be set) */
807 if (it->it_op & IT_CREAT)
809 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
812 else if (it->it_op & IT_READDIR)
814 else if (it->it_op & IT_GETXATTR)
816 else if (it->it_op & IT_SETXATTR)
819 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
824 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
825 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
826 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
827 struct lu_fid op_fid4; /* to the operation locks. */
828 mdsno_t op_mds; /* what mds server open will go to */
829 struct lustre_handle op_handle;
830 obd_time op_mod_time;
834 struct lmv_stripe_md *op_mea1;
835 struct lmv_stripe_md *op_mea2;
836 __u32 op_suppgids[2];
843 /* iattr fields and blocks. */
844 struct iattr op_attr;
846 loff_t op_attr_blocks;
848 /* Size-on-MDS epoch and flags. */
853 struct obd_capa *op_capa1;
854 struct obd_capa *op_capa2;
856 /* Various operation flags. */
857 enum mds_op_bias op_bias;
859 /* Used by readdir */
860 __u64 op_hash_offset;
862 /* The offset of dir entry among the same hash entries, which
863 * is used to resolve the hash conflict. */
864 __u32 op_same_hash_offset;
866 /* Used by readdir */
869 /* used to transfer info between the stacks of MD client
870 * see enum op_cli_flags */
873 /* File object data version for HSM release, on client */
874 __u64 op_data_version;
875 struct lustre_handle op_lease_handle;
878 #define op_stripe_offset op_ioepoch
879 #define op_max_pages op_valid
882 int (*md_blocking_ast)(struct ldlm_lock *lock,
883 struct ldlm_lock_desc *desc,
884 void *data, int flag);
887 struct md_enqueue_info;
888 /* metadata stat-ahead */
889 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
890 struct md_enqueue_info *minfo,
893 struct md_enqueue_info {
894 struct md_op_data mi_data;
895 struct lookup_intent mi_it;
896 struct lustre_handle mi_lockh;
897 struct inode *mi_dir;
898 md_enqueue_cb_t mi_cb;
900 unsigned int mi_generation;
904 struct module *o_owner;
905 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
906 void *karg, void *uarg);
907 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
908 __u32 keylen, void *key, __u32 *vallen, void *val,
909 struct lov_stripe_md *lsm);
910 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
911 __u32 keylen, void *key,
912 __u32 vallen, void *val,
913 struct ptlrpc_request_set *set);
914 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
915 int (*o_precleanup)(struct obd_device *dev,
916 enum obd_cleanup_stage cleanup_stage);
917 int (*o_cleanup)(struct obd_device *dev);
918 int (*o_process_config)(struct obd_device *dev, obd_count len,
920 int (*o_postrecov)(struct obd_device *dev);
921 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
923 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
924 /* connect to the target device with given connection
925 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
926 * granted by the target, which are guaranteed to be a subset of flags
927 * asked for. If @ocd == NULL, use default parameters. */
928 int (*o_connect)(const struct lu_env *env,
929 struct obd_export **exp, struct obd_device *src,
930 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
932 int (*o_reconnect)(const struct lu_env *env,
933 struct obd_export *exp, struct obd_device *src,
934 struct obd_uuid *cluuid,
935 struct obd_connect_data *ocd,
937 int (*o_disconnect)(struct obd_export *exp);
939 /* Initialize/finalize fids infrastructure. */
940 int (*o_fid_init)(struct obd_device *obd,
941 struct obd_export *exp, enum lu_cli_type type);
942 int (*o_fid_fini)(struct obd_device *obd);
944 /* Allocate new fid according to passed @hint. */
945 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
946 struct lu_fid *fid, struct md_op_data *op_data);
949 * Object with @fid is getting deleted, we may want to do something
952 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
953 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
954 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
955 __u64 max_age, struct ptlrpc_request_set *set);
956 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
957 struct lov_stripe_md *mem_src);
958 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
959 struct lov_mds_md *disk_src, int disk_len);
960 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
961 struct obdo *oa, struct lov_stripe_md **ea,
962 struct obd_trans_info *oti);
963 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
964 struct obdo *oa, struct lov_stripe_md *ea,
965 struct obd_trans_info *oti, struct obd_export *md_exp,
967 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
968 struct obd_info *oinfo, struct obd_trans_info *oti);
969 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
970 struct obd_trans_info *oti,
971 struct ptlrpc_request_set *rqset);
972 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
973 struct obd_info *oinfo);
974 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
975 struct ptlrpc_request_set *set);
976 int (*o_preprw)(const struct lu_env *env, int cmd,
977 struct obd_export *exp, struct obdo *oa, int objcount,
978 struct obd_ioobj *obj, struct niobuf_remote *remote,
979 int *nr_pages, struct niobuf_local *local,
980 struct obd_trans_info *oti, struct lustre_capa *capa);
981 int (*o_commitrw)(const struct lu_env *env, int cmd,
982 struct obd_export *exp, struct obdo *oa,
983 int objcount, struct obd_ioobj *obj,
984 struct niobuf_remote *remote, int pages,
985 struct niobuf_local *local,
986 struct obd_trans_info *oti, int rc);
987 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
988 ldlm_iterator_t it, void *data);
989 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
990 ldlm_iterator_t it, void *data);
991 int (*o_init_export)(struct obd_export *exp);
992 int (*o_destroy_export)(struct obd_export *exp);
994 /* llog related obd_methods */
995 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
996 struct obd_device *disk_obd, int *idx);
997 int (*o_llog_finish)(struct obd_device *obd, int count);
999 int (*o_import_event)(struct obd_device *, struct obd_import *,
1000 enum obd_import_event);
1002 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1003 enum obd_notify_event ev, void *data);
1005 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1006 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1009 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1010 struct obd_quotactl *);
1011 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1012 struct obd_quotactl *);
1014 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1017 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1018 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1019 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1021 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1023 void (*o_getref)(struct obd_device *obd);
1024 void (*o_putref)(struct obd_device *obd);
1026 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1027 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1028 * Also, add a wrapper function in include/linux/obd_class.h. */
1031 /* lmv structures */
1033 struct mdt_body *body;
1034 struct lov_stripe_md *lsm;
1035 struct lmv_stripe_md *lmv;
1036 #ifdef CONFIG_FS_POSIX_ACL
1037 struct posix_acl *posix_acl;
1039 struct mdt_remote_perm *remote_perm;
1040 struct obd_capa *mds_capa;
1041 struct obd_capa *oss_capa;
1044 struct md_open_data {
1045 struct obd_client_handle *mod_och;
1046 struct ptlrpc_request *mod_open_req;
1047 struct ptlrpc_request *mod_close_req;
1048 atomic_t mod_refcount;
1052 struct lookup_intent;
1056 /* Every operation from MD_STATS_FIRST_OP up to and including
1057 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1058 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1059 * Operations after MD_STATS_LAST_OP are excluded from stats.
1060 * There are a few reasons for doing this: we prune the 17
1061 * counters which will be of minimal use in understanding
1062 * metadata utilization, we save memory by allocating 15
1063 * instead of 32 counters, we save cycles by not counting.
1065 * MD_STATS_FIRST_OP must be the first member of md_ops.
1067 #define MD_STATS_FIRST_OP m_close
1068 int (*m_close)(struct obd_export *, struct md_op_data *,
1069 struct md_open_data *, struct ptlrpc_request **);
1071 int (*m_create)(struct obd_export *, struct md_op_data *,
1072 const void *, int, int, __u32, __u32, cfs_cap_t,
1073 __u64, struct ptlrpc_request **);
1075 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1076 const union ldlm_policy_data *,
1077 struct lookup_intent *, struct md_op_data *,
1078 struct lustre_handle *, __u64);
1080 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1081 struct ptlrpc_request **);
1083 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1084 struct lookup_intent *,
1085 struct ptlrpc_request **,
1086 ldlm_blocking_callback, __u64);
1088 int (*m_link)(struct obd_export *, struct md_op_data *,
1089 struct ptlrpc_request **);
1091 int (*m_rename)(struct obd_export *, struct md_op_data *,
1092 const char *, int, const char *, int,
1093 struct ptlrpc_request **);
1095 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1096 int , void *, int, struct ptlrpc_request **,
1097 struct md_open_data **mod);
1099 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1100 struct obd_capa *, struct ptlrpc_request **);
1102 int (*m_read_entry)(struct obd_export *, struct md_op_data *,
1103 struct md_callback *cb_op, struct lu_dirent **ld,
1104 struct page **ppage);
1106 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1107 struct ptlrpc_request **);
1109 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1110 struct obd_capa *, obd_valid, const char *,
1111 const char *, int, int, int, __u32,
1112 struct ptlrpc_request **);
1114 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1115 struct obd_capa *, obd_valid, const char *,
1116 const char *, int, int, int,
1117 struct ptlrpc_request **);
1119 int (*m_intent_getattr_async)(struct obd_export *,
1120 struct md_enqueue_info *,
1121 struct ldlm_enqueue_info *);
1123 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1124 struct lu_fid *, __u64 *bits);
1126 #define MD_STATS_LAST_OP m_revalidate_lock
1128 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1129 struct obd_capa **);
1131 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1133 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1134 ldlm_iterator_t, void *);
1136 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1137 struct md_open_data *);
1139 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1140 struct ptlrpc_request **);
1142 int (*m_init_ea_size)(struct obd_export *, int, int, int, int);
1144 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1145 struct obd_export *, struct obd_export *,
1146 struct lustre_md *);
1148 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1150 int (*m_merge_attr)(struct obd_export *,
1151 const struct lmv_stripe_md *lsm,
1152 struct cl_attr *attr);
1154 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1155 struct mdt_body *, ldlm_blocking_callback);
1157 int (*m_set_open_replay_data)(struct obd_export *,
1158 struct obd_client_handle *,
1159 struct lookup_intent *);
1161 int (*m_clear_open_replay_data)(struct obd_export *,
1162 struct obd_client_handle *);
1164 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1166 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1167 const struct lu_fid *, ldlm_type_t,
1168 ldlm_policy_data_t *, ldlm_mode_t,
1169 struct lustre_handle *);
1171 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1172 ldlm_policy_data_t *, ldlm_mode_t,
1173 ldlm_cancel_flags_t flags, void *opaque);
1175 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1176 renew_capa_cb_t cb);
1178 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1179 const struct req_msg_field *, struct obd_capa **);
1181 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1182 struct obd_capa *, __u32,
1183 struct ptlrpc_request **);
1185 int (*m_get_fid_from_lsm)(struct obd_export *,
1186 const struct lmv_stripe_md *,
1187 const char *name, int namelen,
1188 struct lu_fid *fid);
1191 struct lsm_operations {
1192 void (*lsm_free)(struct lov_stripe_md *);
1193 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1194 struct obd_export *md_exp);
1195 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1197 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1199 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1200 __u16 *stripe_count);
1201 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1202 struct lov_mds_md *lmm);
1205 extern const struct lsm_operations lsm_v1_ops;
1206 extern const struct lsm_operations lsm_v3_ops;
1207 static inline const struct lsm_operations *lsm_op_find(int magic)
1215 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1220 static inline struct md_open_data *obd_mod_alloc(void)
1222 struct md_open_data *mod;
1226 atomic_set(&mod->mod_refcount, 1);
1230 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1231 #define obd_mod_put(mod) \
1233 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1234 if ((mod)->mod_open_req) \
1235 ptlrpc_req_finished((mod)->mod_open_req); \
1236 OBD_FREE_PTR(mod); \
1240 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1241 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1243 /* return 1 if client should be resend request */
1244 static inline int client_should_resend(int resend, struct client_obd *cli)
1246 return atomic_read(&cli->cl_resends) ?
1247 atomic_read(&cli->cl_resends) > resend : 1;
1251 * Return device name for this device
1253 * XXX: lu_device is declared before obd_device, while a pointer pointing
1254 * back to obd_device in lu_device, so this helper function defines here
1255 * instead of in lu_object.h
1257 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1259 return lu_dev->ld_obd->obd_name;
1262 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1267 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1270 /* caller does not care of idx */
1274 /* volatile file, the MDT can be set from name */
1275 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1276 /* if no MDT is specified, use std way */
1277 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1279 /* test for no MDT idx case */
1280 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1281 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1285 /* we have an idx, read it */
1286 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1287 *idx = strtoul(start, &end, 16);
1289 * no digit, no trailing :, negative value
1291 if (((*idx == 0) && (end == start)) ||
1292 (*end != ':') || (*idx < 0))
1297 /* bad format of mdt idx, we cannot return an error
1298 * to caller so we use hash algo */
1299 CERROR("Bad volatile file name format: %s\n",
1300 name + LUSTRE_VOLATILE_HDR_LEN);
1304 static inline int cli_brw_size(struct obd_device *obd)
1306 LASSERT(obd != NULL);
1307 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1310 #endif /* __OBD_H */