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 #include <linux/obd.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
50 #include <lustre_export.h>
51 #include <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <lustre_handles.h>
54 #include <lustre_intent.h>
55 #include <lustre_capa.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
87 /* If we are unable to get the maximum object size from the OST in
88 * ocd_maxbytes using OBD_CONNECT_MAXBYTES, then we fall back to using
89 * the old maximum object size from ext3. */
90 #define LUSTRE_EXT3_STRIPE_MAXBYTES 0x1fffffff000ULL
92 struct lov_stripe_md {
95 pid_t lsm_lock_owner; /* debugging */
97 /* maximum possible file size, might change as OSTs status changes,
98 * e.g. disconnected, deactivated */
101 /* Public members. */
102 struct ost_id lw_object_oi; /* lov object id/seq */
104 /* LOV-private members start here -- only for use in lov/. */
106 __u32 lw_stripe_size; /* size of the stripe */
107 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
108 __u16 lw_stripe_count; /* number of objects being striped over */
109 __u16 lw_layout_gen; /* generation of the layout */
110 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
113 struct lov_oinfo *lsm_oinfo[0];
116 #define lsm_oi lsm_wire.lw_object_oi
117 #define lsm_magic lsm_wire.lw_magic
118 #define lsm_layout_gen lsm_wire.lw_layout_gen
119 #define lsm_stripe_size lsm_wire.lw_stripe_size
120 #define lsm_pattern lsm_wire.lw_pattern
121 #define lsm_stripe_count lsm_wire.lw_stripe_count
122 #define lsm_pool_name lsm_wire.lw_pool_name
124 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
126 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
129 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
133 if (lsm_is_released(lsm))
138 static inline int lov_stripe_md_size(unsigned int stripe_count)
140 struct lov_stripe_md lsm;
142 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
147 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
149 /* obd info for a particular level (lov, osc). */
151 /* Lock policy. It keeps an extent which is specific for a particular
152 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
153 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
154 ldlm_policy_data_t oi_policy;
155 /* Flags used for set request specific flags:
156 - while lock handling, the flags obtained on the enqueue
157 request are set here.
158 - while stats, the flags used for control delay/resend.
159 - while setattr, the flags used for distinguish punch operation
162 /* lsm data specific for every OSC. */
163 struct lov_stripe_md *oi_md;
164 /* obdo data specific for every OSC, if needed at all. */
166 /* statfs data specific for every OSC, if needed at all. */
167 struct obd_statfs *oi_osfs;
168 /* An update callback which is called to update some data on upper
169 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
170 * request in osc level for enqueue requests. It is also possible to
171 * update some caller data from LOV layer if needed. */
172 obd_enqueue_update_f oi_cb_up;
173 /* oss capability, its type is obd_capa in client to avoid copy.
174 * in contrary its type is lustre_capa in OSS. */
179 struct list_head typ_chain;
180 struct obd_ops *typ_dt_ops;
181 struct md_ops *typ_md_ops;
182 struct proc_dir_entry *typ_procroot;
183 struct proc_dir_entry *typ_procsym;
184 __u32 typ_sym_filter;
187 struct lu_device_type *typ_lu;
188 spinlock_t obd_type_lock;
198 struct timeout_item {
199 enum timeout_event ti_event;
200 cfs_time_t ti_timeout;
203 struct list_head ti_obd_list;
204 struct list_head ti_chain;
207 #define OBD_MAX_RIF_DEFAULT 8
208 #define OBD_MAX_RIF_MAX 512
209 #define OSC_MAX_RIF_MAX 256
210 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
211 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
212 #define OSC_DEFAULT_RESENDS 10
214 /* possible values for fo_sync_lock_cancel */
216 NEVER_SYNC_ON_CANCEL = 0,
217 BLOCKING_SYNC_ON_CANCEL = 1,
218 ALWAYS_SYNC_ON_CANCEL = 2,
219 NUM_SYNC_ON_CANCEL_STATES
225 struct rw_semaphore cl_sem;
226 struct obd_uuid cl_target_uuid;
227 struct obd_import *cl_import; /* ptlrpc connection state */
229 /* max_mds_easize is purely a performance thing so we don't have to
230 * call obd_size_diskmd() all the time. */
231 int cl_default_mds_easize;
232 int cl_max_mds_easize;
233 int cl_default_mds_cookiesize;
234 int cl_max_mds_cookiesize;
236 enum lustre_sec_part cl_sp_me;
237 enum lustre_sec_part cl_sp_to;
238 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
240 /* the grant values are protected by loi_list_lock below */
241 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
242 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
243 unsigned long cl_dirty_transit; /* dirty synchronous */
244 unsigned long cl_avail_grant; /* bytes of credit for ost */
245 unsigned long cl_lost_grant; /* lost credits (trunc) */
247 /* since we allocate grant by blocks, we don't know how many grant will
248 * be used to add a page into cache. As a solution, we reserve maximum
249 * grant before trying to dirty a page and unreserve the rest.
250 * See osc_{reserve|unreserve}_grant for details. */
251 long cl_reserved_grant;
252 struct list_head cl_cache_waiters; /* waiting for cache/grant */
253 cfs_time_t cl_next_shrink_grant; /* jiffies */
254 struct list_head cl_grant_shrink_list; /* Timeout event list */
255 int cl_grant_shrink_interval; /* seconds */
257 /* A chunk is an optimal size used by osc_extent to determine
258 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
260 int cl_extent_tax; /* extent overhead, by bytes */
262 /* keep track of objects that have lois that contain pages which
263 * have been queued for async brw. this lock also protects the
264 * lists of osc_client_pages that hang off of the loi */
266 * ->cl_loi_list_lock protects consistency of
267 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
268 * ->ap_completion() call-backs are executed under this lock. As we
269 * cannot guarantee that these call-backs never block on all platforms
270 * (as a matter of fact they do block on Mac OS X), type of
271 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
272 * and blocking mutex on Mac OS X. (Alternative is to make this lock
273 * blocking everywhere, but we don't want to slow down fast-path of
274 * our main platform.)
276 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
277 * with client_obd_list_{un,}lock() and
278 * client_obd_list_lock_{init,done}() functions.
280 * NB by Jinshan: though field names are still _loi_, but actually
281 * osc_object{}s are in the list.
283 client_obd_lock_t cl_loi_list_lock;
284 struct list_head cl_loi_ready_list;
285 struct list_head cl_loi_hp_ready_list;
286 struct list_head cl_loi_write_list;
287 struct list_head cl_loi_read_list;
288 __u32 cl_r_in_flight;
289 __u32 cl_w_in_flight;
290 /* just a sum of the loi/lop pending numbers to be exported by /proc */
291 atomic_t cl_pending_w_pages;
292 atomic_t cl_pending_r_pages;
293 __u32 cl_max_pages_per_rpc;
294 __u32 cl_max_rpcs_in_flight;
295 struct obd_histogram cl_read_rpc_hist;
296 struct obd_histogram cl_write_rpc_hist;
297 struct obd_histogram cl_read_page_hist;
298 struct obd_histogram cl_write_page_hist;
299 struct obd_histogram cl_read_offset_hist;
300 struct obd_histogram cl_write_offset_hist;
302 /* lru for osc caching pages */
303 struct cl_client_cache *cl_cache;
304 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
305 atomic_t *cl_lru_left;
306 atomic_t cl_lru_busy;
307 atomic_t cl_lru_shrinkers;
308 atomic_t cl_lru_in_list;
309 struct list_head cl_lru_list; /* lru page list */
310 client_obd_lock_t cl_lru_list_lock; /* page list protector */
311 atomic_t cl_unstable_count;
313 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
314 atomic_t cl_destroy_in_flight;
315 wait_queue_head_t cl_destroy_waitq;
317 struct mdc_rpc_lock *cl_rpc_lock;
318 struct mdc_rpc_lock *cl_close_lock;
321 struct mutex cl_mgc_mutex;
322 struct local_oid_storage *cl_mgc_los;
323 struct dt_object *cl_mgc_configs_dir;
324 atomic_t cl_mgc_refcount;
325 struct obd_export *cl_mgc_mgsexp;
327 /* checksumming for data sent over the network */
328 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
329 /* supported checksum types that are worked out at connect time */
330 __u32 cl_supp_cksum_types;
331 /* checksum algorithm to be used */
332 cksum_type_t cl_cksum_type;
334 /* also protected by the poorly named _loi_list_lock lock above */
335 struct osc_async_rc cl_ar;
337 /* used by quotacheck when the servers are older than 2.4 */
338 int cl_qchk_stat; /* quotacheck stat of the peer */
339 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
340 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
341 #warning "please consider removing quotacheck compatibility code"
344 /* sequence manager */
345 struct lu_client_seq *cl_seq;
347 atomic_t cl_resends; /* resend count */
349 /* ptlrpc work for writeback in ptlrpcd context */
350 void *cl_writeback_work;
352 /* hash tables for osc_quota_info */
353 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
355 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
362 struct echo_client_obd {
363 struct obd_export *ec_exp; /* the local connection to osc/lov */
365 struct list_head ec_objects;
366 struct list_head ec_locks;
371 /* Generic subset of OSTs */
373 __u32 *op_array; /* array of index of
375 unsigned int op_count; /* number of OSTs in the array */
376 unsigned int op_size; /* allocated size of lp_array */
377 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
380 /* allow statfs data caching for 1 second */
381 #define OBD_STATFS_CACHE_SECONDS 1
383 struct lov_tgt_desc {
384 struct list_head ltd_kill;
385 struct obd_uuid ltd_uuid;
386 struct obd_device *ltd_obd;
387 struct obd_export *ltd_exp;
389 __u32 ltd_index; /* index in lov_obd->tgts */
390 unsigned long ltd_active:1,/* is this target up for requests */
391 ltd_activate:1,/* should target be activated */
392 ltd_reap:1; /* should this target be deleted */
396 struct lov_desc desc;
397 struct lov_tgt_desc **lov_tgts; /* sparse array */
398 struct ost_pool lov_packed; /* all OSTs in a packed
400 struct mutex lov_lock;
401 struct obd_connect_data lov_ocd;
402 struct proc_dir_entry *targets_proc_entry;
403 atomic_t lov_refcount;
404 __u32 lov_death_row; /* tgts scheduled to be deleted */
405 __u32 lov_tgt_size; /* size of tgts array */
408 cfs_hash_t *lov_pools_hash_body; /* used for key access */
409 struct list_head lov_pool_list; /* used for sequential access */
410 struct proc_dir_entry *lov_pool_proc_entry;
411 enum lustre_sec_part lov_sp_me;
413 /* Cached LRU and unstable data from upper layer */
416 struct rw_semaphore lov_notify_lock;
419 struct lmv_tgt_desc {
420 struct obd_uuid ltd_uuid;
421 struct obd_export *ltd_exp;
423 struct mutex ltd_fid_mutex;
424 unsigned long ltd_active:1; /* target up for requests */
427 enum placement_policy {
428 PLACEMENT_CHAR_POLICY = 0,
429 PLACEMENT_NID_POLICY = 1,
430 PLACEMENT_INVAL_POLICY = 2,
434 typedef enum placement_policy placement_policy_t;
438 struct lu_client_fld lmv_fld;
440 placement_policy_t lmv_placement;
441 struct lmv_desc desc;
442 struct obd_uuid cluuid;
443 struct obd_export *exp;
444 struct proc_dir_entry *targets_proc_entry;
446 struct mutex init_mutex;
451 int max_def_cookiesize;
454 int tgts_size; /* size of tgts array */
455 struct lmv_tgt_desc **tgts;
457 struct obd_connect_data conn_data;
460 struct niobuf_local {
461 __u64 lnb_file_offset;
462 __u32 lnb_page_offset;
465 struct page *lnb_page;
470 #define LUSTRE_FLD_NAME "fld"
471 #define LUSTRE_SEQ_NAME "seq"
473 #define LUSTRE_MDD_NAME "mdd"
474 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
475 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
476 #define LUSTRE_VVP_NAME "vvp"
477 #define LUSTRE_LMV_NAME "lmv"
478 #define LUSTRE_SLP_NAME "slp"
479 #define LUSTRE_LOD_NAME "lod"
480 #define LUSTRE_OSP_NAME "osp"
481 #define LUSTRE_LWP_NAME "lwp"
483 /* obd device type names */
484 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
485 #define LUSTRE_MDS_NAME "mds"
486 #define LUSTRE_MDT_NAME "mdt"
487 #define LUSTRE_MDC_NAME "mdc"
488 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
489 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
490 #define LUSTRE_OSC_NAME "osc"
491 #define LUSTRE_LOV_NAME "lov"
492 #define LUSTRE_MGS_NAME "mgs"
493 #define LUSTRE_MGC_NAME "mgc"
495 #define LUSTRE_ECHO_NAME "obdecho"
496 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
497 #define LUSTRE_QMT_NAME "qmt"
499 /* Constant obd names (post-rename) */
500 #define LUSTRE_MDS_OBDNAME "MDS"
501 #define LUSTRE_OSS_OBDNAME "OSS"
502 #define LUSTRE_MGS_OBDNAME "MGS"
503 #define LUSTRE_MGC_OBDNAME "MGC"
505 struct obd_trans_info {
507 /* Only used on the server side for tracking acks. */
508 struct oti_req_ack_lock {
509 struct lustre_handle lock;
513 struct llog_cookie oti_onecookie;
514 struct llog_cookie *oti_logcookies;
517 __u64 oti_pre_version;
521 * Events signalled through obd_notify() upcall-chain.
523 enum obd_notify_event {
526 /* Device connect start */
528 /* Device activated */
530 /* Device deactivated */
532 /* Device disconnected */
534 /* Connect data for import were changed */
537 OBD_NOTIFY_SYNC_NONBLOCK,
539 /* Configuration event */
541 /* Administratively deactivate/activate event */
542 OBD_NOTIFY_DEACTIVATE,
546 /* bit-mask flags for config events */
548 CONFIG_LOG = 0x1, /* finished processing config log */
549 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
550 CONFIG_TARGET = 0x4 /* one target is added */
554 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
555 * and liblustre being main examples).
557 struct obd_notify_upcall {
558 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
559 enum obd_notify_event ev, void *owner, void *data);
560 /* Opaque datum supplied by upper layer listener */
564 struct target_recovery_data {
565 svc_handler_t trd_recovery_handler;
566 pid_t trd_processing_task;
567 struct completion trd_starting;
568 struct completion trd_finishing;
571 struct obd_llog_group {
572 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
573 wait_queue_head_t olg_waitq;
575 struct mutex olg_cat_processing;
578 /* corresponds to one of the obd's */
579 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
582 struct obd_type *obd_type;
585 /* common and UUID name of this device */
586 char obd_name[MAX_OBD_NAME];
587 struct obd_uuid obd_uuid;
589 struct lu_device *obd_lu_dev;
591 /* bitfield modification is protected by obd_dev_lock */
593 obd_attached:1, /* finished attach */
594 obd_set_up:1, /* finished setup */
595 obd_recovering:1, /* there are recoverable clients */
596 obd_abort_recovery:1, /* recovery expired */
597 obd_version_recov:1, /* obd uses version checking */
598 obd_replayable:1, /* recovery is enabled;
600 obd_no_transno:1, /* no committed-transno notification */
601 obd_no_recov:1, /* fail instead of retry messages */
602 obd_stopping:1, /* started cleanup */
603 obd_starting:1, /* started setup */
604 obd_force:1, /* cleanup with > 0 obd refcount */
605 obd_fail:1, /* cleanup with failover */
606 obd_no_conn:1, /* deny new connections */
607 obd_inactive:1, /* device active/inactive
608 * (for /proc/status only!!) */
609 obd_no_ir:1, /* no imperative recovery. */
610 obd_process_conf:1, /* device is processing mgs config */
611 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
613 /* use separate field as it is set in interrupt to don't mess with
614 * protection of other bits using _bh lock */
615 unsigned long obd_recovery_expired:1;
616 /* uuid-export hash body */
617 cfs_hash_t *obd_uuid_hash;
618 /* nid-export hash body */
619 cfs_hash_t *obd_nid_hash;
621 cfs_hash_t *obd_nid_stats_hash;
622 struct list_head obd_nid_stats;
623 atomic_t obd_refcount;
624 struct list_head obd_exports;
625 struct list_head obd_unlinked_exports;
626 struct list_head obd_delayed_exports;
627 struct list_head obd_lwp_list;
629 spinlock_t obd_nid_lock;
630 struct ldlm_namespace *obd_namespace;
631 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
632 /* a spinlock is OK for what we do now, may need a semaphore later */
633 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
634 struct mutex obd_dev_mutex;
635 __u64 obd_last_committed;
636 spinlock_t obd_osfs_lock;
637 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
639 struct lvfs_run_ctxt obd_lvfs_ctxt;
640 struct obd_llog_group obd_olg; /* default llog group */
641 struct obd_device *obd_observer;
642 struct rw_semaphore obd_observer_link_sem;
643 struct obd_notify_upcall obd_upcall;
644 struct obd_export *obd_self_export;
645 struct obd_export *obd_lwp_export;
646 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
647 struct list_head obd_exports_timed;
648 time_t obd_eviction_timer; /* for ping evictor */
650 int obd_max_recoverable_clients;
651 atomic_t obd_connected_clients;
652 int obd_stale_clients;
653 /* this lock protects all recovery list_heads, timer and
654 * obd_next_recovery_transno value */
655 spinlock_t obd_recovery_task_lock;
656 __u64 obd_next_recovery_transno;
657 int obd_replayed_requests;
658 int obd_requests_queued_for_recovery;
659 wait_queue_head_t obd_next_transno_waitq;
660 /* protected by obd_recovery_task_lock */
661 struct timer_list obd_recovery_timer;
663 time_t obd_recovery_start;
664 /* seconds, for lprocfs_status */
665 time_t obd_recovery_end;
666 int obd_recovery_time_hard;
667 int obd_recovery_timeout;
668 int obd_recovery_ir_factor;
670 /* new recovery stuff from CMD2 */
671 struct target_recovery_data obd_recovery_data;
672 int obd_replayed_locks;
673 atomic_t obd_req_replay_clients;
674 atomic_t obd_lock_replay_clients;
675 /* all lists are protected by obd_recovery_task_lock */
676 struct list_head obd_req_replay_queue;
677 struct list_head obd_lock_replay_queue;
678 struct list_head obd_final_req_queue;
681 #ifdef HAVE_SERVER_SUPPORT
682 struct obd_device_target obt;
683 struct filter_obd filter;
685 struct echo_obd echo;
687 struct client_obd cli;
688 struct echo_client_obd echo_client;
692 /* Fields used by LProcFS */
693 unsigned int obd_cntr_base;
694 struct lprocfs_stats *obd_stats;
696 unsigned int obd_md_cntr_base;
697 struct lprocfs_stats *obd_md_stats;
699 struct proc_dir_entry *obd_proc_entry;
700 struct proc_dir_entry *obd_proc_exports_entry;
701 struct proc_dir_entry *obd_svc_procroot;
702 struct lprocfs_stats *obd_svc_stats;
703 struct lprocfs_seq_vars *obd_vars;
704 atomic_t obd_evict_inprogress;
705 wait_queue_head_t obd_evict_inprogress_waitq;
706 struct list_head obd_evict_list; /* protected with pet_lock */
709 * Ldlm pool part. Save last calculated SLV and Limit.
711 rwlock_t obd_pool_lock;
716 * A list of outstanding class_incref()'s against this obd. For
719 struct lu_ref obd_reference;
721 int obd_conn_inprogress;
724 enum obd_cleanup_stage {
725 /* Special case hack for MDS LOVs */
727 /* can be directly mapped to .ldto_device_fini() */
731 /* get/set_info keys */
732 #define KEY_ASYNC "async"
733 #define KEY_CAPA_KEY "capa_key"
734 #define KEY_CHANGELOG_CLEAR "changelog_clear"
735 #define KEY_FID2PATH "fid2path"
736 #define KEY_CHECKSUM "checksum"
737 #define KEY_CLEAR_FS "clear_fs"
738 #define KEY_CONN_DATA "conn_data"
739 #define KEY_EVICT_BY_NID "evict_by_nid"
740 #define KEY_FIEMAP "fiemap"
741 #define KEY_FLUSH_CTX "flush_ctx"
742 #define KEY_GRANT_SHRINK "grant_shrink"
743 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
744 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
745 #define KEY_INTERMDS "inter_mds"
746 #define KEY_LAST_ID "last_id"
747 #define KEY_LAST_FID "last_fid"
748 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
749 #define KEY_LOVDESC "lovdesc"
750 #define KEY_MAX_EASIZE "max_easize"
751 #define KEY_DEFAULT_EASIZE "default_easize"
752 #define KEY_MAX_COOKIESIZE "max_cookiesize"
753 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
754 #define KEY_MDS_CONN "mds_conn"
755 #define KEY_MGSSEC "mgssec"
756 #define KEY_NEXT_ID "next_id"
757 #define KEY_READ_ONLY "read-only"
758 #define KEY_REGISTER_TARGET "register_target"
759 #define KEY_SET_FS "set_fs"
760 #define KEY_TGT_COUNT "tgt_count"
761 /* KEY_SET_INFO in lustre_idl.h */
762 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
763 #define KEY_CONNECT_FLAG "connect_flags"
765 #define KEY_CACHE_SET "cache_set"
766 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
767 #define KEY_OSP_CONNECTED "osp_connected"
771 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
772 #define IT_OPEN (1 << 0)
773 #define IT_CREAT (1 << 1)
774 #define IT_READDIR (1 << 2)
775 #define IT_GETATTR (1 << 3)
776 #define IT_LOOKUP (1 << 4)
777 #define IT_UNLINK (1 << 5)
778 #define IT_TRUNC (1 << 6)
779 #define IT_GETXATTR (1 << 7)
780 #define IT_EXEC (1 << 8)
781 #define IT_PIN (1 << 9)
782 #define IT_LAYOUT (1 << 10)
783 #define IT_QUOTA_DQACQ (1 << 11)
784 #define IT_QUOTA_CONN (1 << 12)
785 #define IT_SETXATTR (1 << 13)
787 static inline int it_to_lock_mode(struct lookup_intent *it)
789 /* CREAT needs to be tested before open (both could be set) */
790 if (it->it_op & IT_CREAT)
792 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
795 else if (it->it_op & IT_READDIR)
797 else if (it->it_op & IT_GETXATTR)
799 else if (it->it_op & IT_SETXATTR)
802 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
807 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
808 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
809 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
810 struct lu_fid op_fid4; /* to the operation locks. */
811 u32 op_mds; /* what mds server open will go to */
812 struct lustre_handle op_handle;
813 obd_time op_mod_time;
817 struct lmv_stripe_md *op_mea1;
818 struct lmv_stripe_md *op_mea2;
819 __u32 op_suppgids[2];
826 /* iattr fields and blocks. */
827 struct iattr op_attr;
829 loff_t op_attr_blocks;
831 /* Size-on-MDS epoch and flags. */
836 struct obd_capa *op_capa1;
837 struct obd_capa *op_capa2;
839 /* Various operation flags. */
840 enum mds_op_bias op_bias;
842 /* Used by readdir */
845 /* used to transfer info between the stacks of MD client
846 * see enum op_cli_flags */
849 /* File object data version for HSM release, on client */
850 __u64 op_data_version;
851 struct lustre_handle op_lease_handle;
854 #define op_stripe_offset op_ioepoch
855 #define op_max_pages op_valid
858 int (*md_blocking_ast)(struct ldlm_lock *lock,
859 struct ldlm_lock_desc *desc,
860 void *data, int flag);
863 struct md_enqueue_info;
864 /* metadata stat-ahead */
865 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
866 struct md_enqueue_info *minfo,
869 struct md_enqueue_info {
870 struct md_op_data mi_data;
871 struct lookup_intent mi_it;
872 struct lustre_handle mi_lockh;
873 struct inode *mi_dir;
874 md_enqueue_cb_t mi_cb;
879 struct module *o_owner;
880 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
881 void *karg, void *uarg);
882 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
883 __u32 keylen, void *key, __u32 *vallen, void *val,
884 struct lov_stripe_md *lsm);
885 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
886 __u32 keylen, void *key,
887 __u32 vallen, void *val,
888 struct ptlrpc_request_set *set);
889 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
890 int (*o_precleanup)(struct obd_device *dev,
891 enum obd_cleanup_stage cleanup_stage);
892 int (*o_cleanup)(struct obd_device *dev);
893 int (*o_process_config)(struct obd_device *dev, obd_count len,
895 int (*o_postrecov)(struct obd_device *dev);
896 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
898 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
899 /* connect to the target device with given connection
900 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
901 * granted by the target, which are guaranteed to be a subset of flags
902 * asked for. If @ocd == NULL, use default parameters. */
903 int (*o_connect)(const struct lu_env *env,
904 struct obd_export **exp, struct obd_device *src,
905 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
907 int (*o_reconnect)(const struct lu_env *env,
908 struct obd_export *exp, struct obd_device *src,
909 struct obd_uuid *cluuid,
910 struct obd_connect_data *ocd,
912 int (*o_disconnect)(struct obd_export *exp);
914 /* Initialize/finalize fids infrastructure. */
915 int (*o_fid_init)(struct obd_device *obd,
916 struct obd_export *exp, enum lu_cli_type type);
917 int (*o_fid_fini)(struct obd_device *obd);
919 /* Allocate new fid according to passed @hint. */
920 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
921 struct lu_fid *fid, struct md_op_data *op_data);
924 * Object with @fid is getting deleted, we may want to do something
927 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
928 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
929 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
930 __u64 max_age, struct ptlrpc_request_set *set);
931 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
932 struct lov_stripe_md *mem_src);
933 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
934 struct lov_mds_md *disk_src, int disk_len);
935 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
936 struct obdo *oa, struct lov_stripe_md **ea,
937 struct obd_trans_info *oti);
938 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
939 struct obdo *oa, struct lov_stripe_md *ea,
940 struct obd_trans_info *oti, struct obd_export *md_exp,
942 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
943 struct obd_info *oinfo, struct obd_trans_info *oti);
944 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
945 struct obd_trans_info *oti,
946 struct ptlrpc_request_set *rqset);
947 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
948 struct obd_info *oinfo);
949 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
950 struct ptlrpc_request_set *set);
951 int (*o_preprw)(const struct lu_env *env, int cmd,
952 struct obd_export *exp, struct obdo *oa, int objcount,
953 struct obd_ioobj *obj, struct niobuf_remote *remote,
954 int *nr_pages, struct niobuf_local *local,
955 struct obd_trans_info *oti, struct lustre_capa *capa);
956 int (*o_commitrw)(const struct lu_env *env, int cmd,
957 struct obd_export *exp, struct obdo *oa,
958 int objcount, struct obd_ioobj *obj,
959 struct niobuf_remote *remote, int pages,
960 struct niobuf_local *local,
961 struct obd_trans_info *oti, int rc);
962 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
963 ldlm_iterator_t it, void *data);
964 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
965 ldlm_iterator_t it, void *data);
966 int (*o_init_export)(struct obd_export *exp);
967 int (*o_destroy_export)(struct obd_export *exp);
969 /* llog related obd_methods */
970 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
971 struct obd_device *disk_obd, int *idx);
972 int (*o_llog_finish)(struct obd_device *obd, int count);
974 int (*o_import_event)(struct obd_device *, struct obd_import *,
975 enum obd_import_event);
977 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
978 enum obd_notify_event ev, void *data);
980 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
981 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
984 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
985 struct obd_quotactl *);
986 int (*o_quotactl)(struct obd_device *, struct obd_export *,
987 struct obd_quotactl *);
989 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
992 int (*o_pool_new)(struct obd_device *obd, char *poolname);
993 int (*o_pool_del)(struct obd_device *obd, char *poolname);
994 int (*o_pool_add)(struct obd_device *obd, char *poolname,
996 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
998 void (*o_getref)(struct obd_device *obd);
999 void (*o_putref)(struct obd_device *obd);
1001 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1002 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1003 * Also, add a wrapper function in include/linux/obd_class.h. */
1006 /* lmv structures */
1008 struct mdt_body *body;
1009 struct lov_stripe_md *lsm;
1010 struct lmv_stripe_md *lmv;
1011 #ifdef CONFIG_FS_POSIX_ACL
1012 struct posix_acl *posix_acl;
1014 struct mdt_remote_perm *remote_perm;
1015 struct obd_capa *mds_capa;
1016 struct obd_capa *oss_capa;
1019 struct md_open_data {
1020 struct obd_client_handle *mod_och;
1021 struct ptlrpc_request *mod_open_req;
1022 struct ptlrpc_request *mod_close_req;
1023 atomic_t mod_refcount;
1027 struct obd_client_handle {
1028 struct lustre_handle och_fh;
1029 struct lu_fid och_fid;
1030 struct md_open_data *och_mod;
1031 struct lustre_handle och_lease_handle; /* open lock for lease */
1036 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1038 struct lookup_intent;
1042 /* Every operation from MD_STATS_FIRST_OP up to and including
1043 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1044 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1045 * Operations after MD_STATS_LAST_OP are excluded from stats.
1046 * There are a few reasons for doing this: we prune the 17
1047 * counters which will be of minimal use in understanding
1048 * metadata utilization, we save memory by allocating 15
1049 * instead of 32 counters, we save cycles by not counting.
1051 * MD_STATS_FIRST_OP must be the first member of md_ops.
1053 #define MD_STATS_FIRST_OP m_close
1054 int (*m_close)(struct obd_export *, struct md_op_data *,
1055 struct md_open_data *, struct ptlrpc_request **);
1057 int (*m_create)(struct obd_export *, struct md_op_data *,
1058 const void *, int, int, __u32, __u32, cfs_cap_t,
1059 __u64, struct ptlrpc_request **);
1061 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1062 const union ldlm_policy_data *,
1063 struct lookup_intent *, struct md_op_data *,
1064 struct lustre_handle *, __u64);
1066 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1067 struct ptlrpc_request **);
1069 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1070 struct lookup_intent *,
1071 struct ptlrpc_request **,
1072 ldlm_blocking_callback, __u64);
1074 int (*m_link)(struct obd_export *, struct md_op_data *,
1075 struct ptlrpc_request **);
1077 int (*m_rename)(struct obd_export *, struct md_op_data *,
1078 const char *, int, const char *, int,
1079 struct ptlrpc_request **);
1081 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1082 int , void *, int, struct ptlrpc_request **,
1083 struct md_open_data **mod);
1085 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1086 struct obd_capa *, struct ptlrpc_request **);
1088 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1089 struct md_callback *cb_op, __u64 hash_offset,
1090 struct page **ppage);
1092 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1093 struct ptlrpc_request **);
1095 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1096 struct obd_capa *, obd_valid, const char *,
1097 const char *, int, int, int, __u32,
1098 struct ptlrpc_request **);
1100 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1101 struct obd_capa *, obd_valid, const char *,
1102 const char *, int, int, int,
1103 struct ptlrpc_request **);
1105 int (*m_intent_getattr_async)(struct obd_export *,
1106 struct md_enqueue_info *,
1107 struct ldlm_enqueue_info *);
1109 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1110 struct lu_fid *, __u64 *bits);
1112 #define MD_STATS_LAST_OP m_revalidate_lock
1114 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1115 struct obd_capa **);
1117 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1119 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1120 ldlm_iterator_t, void *);
1122 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1123 struct md_open_data *);
1125 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1126 struct ptlrpc_request **);
1128 int (*m_init_ea_size)(struct obd_export *, int, int, int, int);
1130 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1131 struct obd_export *, struct obd_export *,
1132 struct lustre_md *);
1134 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1136 int (*m_merge_attr)(struct obd_export *,
1137 const struct lmv_stripe_md *lsm,
1138 struct cl_attr *attr);
1140 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1141 struct mdt_body *, ldlm_blocking_callback);
1143 int (*m_set_open_replay_data)(struct obd_export *,
1144 struct obd_client_handle *,
1145 struct lookup_intent *);
1147 int (*m_clear_open_replay_data)(struct obd_export *,
1148 struct obd_client_handle *);
1150 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1152 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1153 const struct lu_fid *, ldlm_type_t,
1154 ldlm_policy_data_t *, ldlm_mode_t,
1155 struct lustre_handle *);
1157 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1158 ldlm_policy_data_t *, ldlm_mode_t,
1159 ldlm_cancel_flags_t flags, void *opaque);
1161 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1162 renew_capa_cb_t cb);
1164 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1165 const struct req_msg_field *, struct obd_capa **);
1167 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1168 struct obd_capa *, __u32,
1169 struct ptlrpc_request **);
1171 int (*m_get_fid_from_lsm)(struct obd_export *,
1172 const struct lmv_stripe_md *,
1173 const char *name, int namelen,
1174 struct lu_fid *fid);
1177 struct lsm_operations {
1178 void (*lsm_free)(struct lov_stripe_md *);
1179 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1180 struct obd_export *md_exp);
1181 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1183 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1185 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1186 __u16 *stripe_count);
1187 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1188 struct lov_mds_md *lmm);
1191 extern const struct lsm_operations lsm_v1_ops;
1192 extern const struct lsm_operations lsm_v3_ops;
1193 static inline const struct lsm_operations *lsm_op_find(int magic)
1201 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1206 static inline struct md_open_data *obd_mod_alloc(void)
1208 struct md_open_data *mod;
1212 atomic_set(&mod->mod_refcount, 1);
1216 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1217 #define obd_mod_put(mod) \
1219 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1220 if ((mod)->mod_open_req) \
1221 ptlrpc_req_finished((mod)->mod_open_req); \
1222 OBD_FREE_PTR(mod); \
1226 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1227 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1229 /* return 1 if client should be resend request */
1230 static inline int client_should_resend(int resend, struct client_obd *cli)
1232 return atomic_read(&cli->cl_resends) ?
1233 atomic_read(&cli->cl_resends) > resend : 1;
1237 * Return device name for this device
1239 * XXX: lu_device is declared before obd_device, while a pointer pointing
1240 * back to obd_device in lu_device, so this helper function defines here
1241 * instead of in lu_object.h
1243 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1245 return lu_dev->ld_obd->obd_name;
1248 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1253 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1256 /* caller does not care of idx */
1260 /* volatile file, the MDT can be set from name */
1261 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1262 /* if no MDT is specified, use std way */
1263 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1265 /* test for no MDT idx case */
1266 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1267 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1271 /* we have an idx, read it */
1272 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1273 *idx = strtoul(start, &end, 16);
1275 * no digit, no trailing :, negative value
1277 if (((*idx == 0) && (end == start)) ||
1278 (*end != ':') || (*idx < 0))
1283 /* bad format of mdt idx, we cannot return an error
1284 * to caller so we use hash algo */
1285 CERROR("Bad volatile file name format: %s\n",
1286 name + LUSTRE_VOLATILE_HDR_LEN);
1290 static inline int cli_brw_size(struct obd_device *obd)
1292 LASSERT(obd != NULL);
1293 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1296 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1297 * of the client should be increased accordingly to avoid sending fragmented
1298 * RPCs over the network when the client runs out of the maximum dirty space
1299 * when so many RPCs are being generated.
1301 static inline void client_adjust_max_dirty(struct client_obd *cli)
1304 if (cli->cl_dirty_max_pages <= 0)
1305 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1306 >> PAGE_CACHE_SHIFT;
1308 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1309 cli->cl_max_pages_per_rpc;
1311 if (dirty_max > cli->cl_dirty_max_pages)
1312 cli->cl_dirty_max_pages = dirty_max;
1315 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1316 cli->cl_dirty_max_pages = totalram_pages / 8;
1319 #endif /* __OBD_H */