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, 2014, 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/spinlock.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 */
100 struct ost_id lsm_oi;
102 __u32 lsm_stripe_size;
103 __u32 lsm_pattern; /* RAID0, RAID1, released, ... */
104 __u16 lsm_stripe_count;
105 __u16 lsm_layout_gen;
106 char lsm_pool_name[LOV_MAXPOOLNAME + 1];
107 struct lov_oinfo *lsm_oinfo[0];
110 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
112 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
115 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
119 if (lsm_is_released(lsm))
124 static inline int lov_stripe_md_size(unsigned int stripe_count)
126 struct lov_stripe_md lsm;
128 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
133 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
135 /* obd info for a particular level (lov, osc). */
137 /* Lock policy. It keeps an extent which is specific for a particular
138 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
139 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
140 ldlm_policy_data_t oi_policy;
141 /* Flags used for set request specific flags:
142 - while lock handling, the flags obtained on the enqueue
143 request are set here.
144 - while stats, the flags used for control delay/resend.
145 - while setattr, the flags used for distinguish punch operation
148 /* lsm data specific for every OSC. */
149 struct lov_stripe_md *oi_md;
150 /* obdo data specific for every OSC, if needed at all. */
152 /* statfs data specific for every OSC, if needed at all. */
153 struct obd_statfs *oi_osfs;
154 /* An update callback which is called to update some data on upper
155 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
156 * request in osc level for enqueue requests. It is also possible to
157 * update some caller data from LOV layer if needed. */
158 obd_enqueue_update_f oi_cb_up;
159 /* oss capability, its type is obd_capa in client to avoid copy.
160 * in contrary its type is lustre_capa in OSS. */
165 struct list_head typ_chain;
166 struct obd_ops *typ_dt_ops;
167 struct md_ops *typ_md_ops;
168 struct proc_dir_entry *typ_procroot;
169 struct proc_dir_entry *typ_procsym;
170 __u32 typ_sym_filter;
173 struct lu_device_type *typ_lu;
174 spinlock_t obd_type_lock;
184 struct timeout_item {
185 enum timeout_event ti_event;
186 cfs_time_t ti_timeout;
189 struct list_head ti_obd_list;
190 struct list_head ti_chain;
193 #define OBD_MAX_RIF_DEFAULT 8
194 #define OBD_MAX_RIF_MAX 512
195 #define OSC_MAX_RIF_MAX 256
196 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
197 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
198 #define OSC_DEFAULT_RESENDS 10
200 /* possible values for fo_sync_lock_cancel */
202 NEVER_SYNC_ON_CANCEL = 0,
203 BLOCKING_SYNC_ON_CANCEL = 1,
204 ALWAYS_SYNC_ON_CANCEL = 2,
205 NUM_SYNC_ON_CANCEL_STATES
211 struct rw_semaphore cl_sem;
212 struct obd_uuid cl_target_uuid;
213 struct obd_import *cl_import; /* ptlrpc connection state */
214 size_t cl_conn_count;
215 /* max_mds_easize is purely a performance thing so we don't have to
216 * call obd_size_diskmd() all the time. */
217 __u32 cl_default_mds_easize;
218 __u32 cl_max_mds_easize;
219 __u32 cl_default_mds_cookiesize;
220 __u32 cl_max_mds_cookiesize;
222 enum lustre_sec_part cl_sp_me;
223 enum lustre_sec_part cl_sp_to;
224 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
226 /* the grant values are protected by loi_list_lock below */
227 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
228 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
229 unsigned long cl_dirty_transit; /* dirty synchronous */
230 unsigned long cl_avail_grant; /* bytes of credit for ost */
231 unsigned long cl_lost_grant; /* lost credits (trunc) */
233 /* since we allocate grant by blocks, we don't know how many grant will
234 * be used to add a page into cache. As a solution, we reserve maximum
235 * grant before trying to dirty a page and unreserve the rest.
236 * See osc_{reserve|unreserve}_grant for details. */
237 long cl_reserved_grant;
238 struct list_head cl_cache_waiters; /* waiting for cache/grant */
239 cfs_time_t cl_next_shrink_grant; /* jiffies */
240 struct list_head cl_grant_shrink_list; /* Timeout event list */
241 int cl_grant_shrink_interval; /* seconds */
243 /* A chunk is an optimal size used by osc_extent to determine
244 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
246 unsigned int cl_extent_tax; /* extent overhead, by bytes */
248 /* keep track of objects that have lois that contain pages which
249 * have been queued for async brw. this lock also protects the
250 * lists of osc_client_pages that hang off of the loi */
252 * ->cl_loi_list_lock protects consistency of
253 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
254 * ->ap_completion() call-backs are executed under this lock. As we
255 * cannot guarantee that these call-backs never block on all platforms
256 * (as a matter of fact they do block on Mac OS X), type of
257 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
258 * and blocking mutex on Mac OS X. (Alternative is to make this lock
259 * blocking everywhere, but we don't want to slow down fast-path of
260 * our main platform.)
262 * NB by Jinshan: though field names are still _loi_, but actually
263 * osc_object{}s are in the list.
265 spinlock_t cl_loi_list_lock;
266 struct list_head cl_loi_ready_list;
267 struct list_head cl_loi_hp_ready_list;
268 struct list_head cl_loi_write_list;
269 struct list_head cl_loi_read_list;
270 __u32 cl_r_in_flight;
271 __u32 cl_w_in_flight;
272 /* just a sum of the loi/lop pending numbers to be exported by /proc */
273 atomic_t cl_pending_w_pages;
274 atomic_t cl_pending_r_pages;
275 __u32 cl_max_pages_per_rpc;
276 __u32 cl_max_rpcs_in_flight;
277 struct obd_histogram cl_read_rpc_hist;
278 struct obd_histogram cl_write_rpc_hist;
279 struct obd_histogram cl_read_page_hist;
280 struct obd_histogram cl_write_page_hist;
281 struct obd_histogram cl_read_offset_hist;
282 struct obd_histogram cl_write_offset_hist;
284 /* lru for osc caching pages */
285 struct cl_client_cache *cl_cache;
286 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
287 atomic_long_t *cl_lru_left;
288 atomic_long_t cl_lru_busy;
289 atomic_long_t cl_lru_in_list;
290 atomic_long_t cl_unstable_count;
291 struct list_head cl_lru_list; /* lru page list */
292 spinlock_t cl_lru_list_lock; /* page list protector */
293 atomic_t cl_lru_shrinkers;
295 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
296 atomic_t cl_destroy_in_flight;
297 wait_queue_head_t cl_destroy_waitq;
299 struct mdc_rpc_lock *cl_rpc_lock;
300 struct mdc_rpc_lock *cl_close_lock;
303 struct mutex cl_mgc_mutex;
304 struct local_oid_storage *cl_mgc_los;
305 struct dt_object *cl_mgc_configs_dir;
306 atomic_t cl_mgc_refcount;
307 struct obd_export *cl_mgc_mgsexp;
309 /* checksumming for data sent over the network */
310 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
311 /* supported checksum types that are worked out at connect time */
312 __u32 cl_supp_cksum_types;
313 /* checksum algorithm to be used */
314 cksum_type_t cl_cksum_type;
316 /* also protected by the poorly named _loi_list_lock lock above */
317 struct osc_async_rc cl_ar;
319 /* used by quotacheck when the servers are older than 2.4 */
320 int cl_qchk_stat; /* quotacheck stat of the peer */
321 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
322 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
323 #warning "please consider removing quotacheck compatibility code"
326 /* sequence manager */
327 struct lu_client_seq *cl_seq;
329 atomic_t cl_resends; /* resend count */
331 /* ptlrpc work for writeback in ptlrpcd context */
332 void *cl_writeback_work;
334 /* hash tables for osc_quota_info */
335 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
337 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
344 struct echo_client_obd {
345 struct obd_export *ec_exp; /* the local connection to osc/lov */
347 struct list_head ec_objects;
348 struct list_head ec_locks;
352 /* Generic subset of OSTs */
354 __u32 *op_array; /* array of index of
356 unsigned int op_count; /* number of OSTs in the array */
357 unsigned int op_size; /* allocated size of lp_array */
358 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
361 /* allow statfs data caching for 1 second */
362 #define OBD_STATFS_CACHE_SECONDS 1
364 struct lov_tgt_desc {
365 struct list_head ltd_kill;
366 struct obd_uuid ltd_uuid;
367 struct obd_device *ltd_obd;
368 struct obd_export *ltd_exp;
370 __u32 ltd_index; /* index in lov_obd->tgts */
371 unsigned long ltd_active:1,/* is this target up for requests */
372 ltd_activate:1,/* should target be activated */
373 ltd_reap:1; /* should this target be deleted */
377 struct lov_desc desc;
378 struct lov_tgt_desc **lov_tgts; /* sparse array */
379 struct ost_pool lov_packed; /* all OSTs in a packed
381 struct mutex lov_lock;
382 struct obd_connect_data lov_ocd;
383 struct proc_dir_entry *targets_proc_entry;
384 atomic_t lov_refcount;
385 __u32 lov_death_row; /* tgts scheduled to be deleted */
386 __u32 lov_tgt_size; /* size of tgts array */
389 cfs_hash_t *lov_pools_hash_body; /* used for key access */
390 struct list_head lov_pool_list; /* used for sequential access */
391 struct proc_dir_entry *lov_pool_proc_entry;
392 enum lustre_sec_part lov_sp_me;
394 /* Cached LRU and unstable data from upper layer */
397 struct rw_semaphore lov_notify_lock;
400 struct lmv_tgt_desc {
401 struct obd_uuid ltd_uuid;
402 struct obd_export *ltd_exp;
404 struct mutex ltd_fid_mutex;
405 unsigned long ltd_active:1; /* target up for requests */
408 enum placement_policy {
409 PLACEMENT_CHAR_POLICY = 0,
410 PLACEMENT_NID_POLICY = 1,
411 PLACEMENT_INVAL_POLICY = 2,
415 typedef enum placement_policy placement_policy_t;
419 struct lu_client_fld lmv_fld;
421 placement_policy_t lmv_placement;
422 struct lmv_desc desc;
423 struct obd_uuid cluuid;
424 struct obd_export *exp;
425 struct proc_dir_entry *targets_proc_entry;
427 struct mutex lmv_init_mutex;
432 int max_def_cookiesize;
434 __u32 tgts_size; /* size of tgts array */
435 struct lmv_tgt_desc **tgts;
437 struct obd_connect_data conn_data;
440 struct niobuf_local {
441 __u64 lnb_file_offset;
442 __u32 lnb_page_offset;
445 struct page *lnb_page;
450 #define LUSTRE_FLD_NAME "fld"
451 #define LUSTRE_SEQ_NAME "seq"
453 #define LUSTRE_MDD_NAME "mdd"
454 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
455 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
456 #define LUSTRE_VVP_NAME "vvp"
457 #define LUSTRE_LMV_NAME "lmv"
458 #define LUSTRE_SLP_NAME "slp"
459 #define LUSTRE_LOD_NAME "lod"
460 #define LUSTRE_OSP_NAME "osp"
461 #define LUSTRE_LWP_NAME "lwp"
463 /* obd device type names */
464 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
465 #define LUSTRE_MDS_NAME "mds"
466 #define LUSTRE_MDT_NAME "mdt"
467 #define LUSTRE_MDC_NAME "mdc"
468 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
469 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
470 #define LUSTRE_OSC_NAME "osc"
471 #define LUSTRE_LOV_NAME "lov"
472 #define LUSTRE_MGS_NAME "mgs"
473 #define LUSTRE_MGC_NAME "mgc"
475 #define LUSTRE_ECHO_NAME "obdecho"
476 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
477 #define LUSTRE_QMT_NAME "qmt"
479 /* Constant obd names (post-rename) */
480 #define LUSTRE_MDS_OBDNAME "MDS"
481 #define LUSTRE_OSS_OBDNAME "OSS"
482 #define LUSTRE_MGS_OBDNAME "MGS"
483 #define LUSTRE_MGC_OBDNAME "MGC"
485 struct obd_trans_info {
487 /* Only used on the server side for tracking acks. */
488 struct oti_req_ack_lock {
489 struct lustre_handle lock;
493 struct llog_cookie oti_onecookie;
494 struct llog_cookie *oti_logcookies;
497 __u64 oti_pre_version;
501 * Events signalled through obd_notify() upcall-chain.
503 enum obd_notify_event {
506 /* Device connect start */
508 /* Device activated */
510 /* Device deactivated */
512 /* Device disconnected */
514 /* Connect data for import were changed */
517 OBD_NOTIFY_SYNC_NONBLOCK,
519 /* Configuration event */
521 /* Administratively deactivate/activate event */
522 OBD_NOTIFY_DEACTIVATE,
526 /* bit-mask flags for config events */
528 CONFIG_LOG = 0x1, /* finished processing config log */
529 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
530 CONFIG_TARGET = 0x4 /* one target is added */
534 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
535 * and liblustre being main examples).
537 struct obd_notify_upcall {
538 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
539 enum obd_notify_event ev, void *owner, void *data);
540 /* Opaque datum supplied by upper layer listener */
544 struct target_recovery_data {
545 svc_handler_t trd_recovery_handler;
546 pid_t trd_processing_task;
547 struct completion trd_starting;
548 struct completion trd_finishing;
551 struct obd_llog_group {
552 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
553 wait_queue_head_t olg_waitq;
555 struct mutex olg_cat_processing;
558 /* corresponds to one of the obd's */
559 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
562 struct obd_type *obd_type;
565 /* common and UUID name of this device */
566 char obd_name[MAX_OBD_NAME];
567 struct obd_uuid obd_uuid;
569 struct lu_device *obd_lu_dev;
571 /* bitfield modification is protected by obd_dev_lock */
573 obd_attached:1, /* finished attach */
574 obd_set_up:1, /* finished setup */
575 obd_recovering:1, /* there are recoverable clients */
576 obd_abort_recovery:1, /* recovery expired */
577 obd_version_recov:1, /* obd uses version checking */
578 obd_replayable:1, /* recovery is enabled;
580 obd_no_transno:1, /* no committed-transno notification */
581 obd_no_recov:1, /* fail instead of retry messages */
582 obd_stopping:1, /* started cleanup */
583 obd_starting:1, /* started setup */
584 obd_force:1, /* cleanup with > 0 obd refcount */
585 obd_fail:1, /* cleanup with failover */
586 obd_no_conn:1, /* deny new connections */
587 obd_inactive:1, /* device active/inactive
588 * (for /proc/status only!!) */
589 obd_no_ir:1, /* no imperative recovery. */
590 obd_process_conf:1, /* device is processing mgs config */
591 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
593 /* use separate field as it is set in interrupt to don't mess with
594 * protection of other bits using _bh lock */
595 unsigned long obd_recovery_expired:1;
596 /* uuid-export hash body */
597 cfs_hash_t *obd_uuid_hash;
598 /* nid-export hash body */
599 cfs_hash_t *obd_nid_hash;
601 cfs_hash_t *obd_nid_stats_hash;
602 struct list_head obd_nid_stats;
603 atomic_t obd_refcount;
604 struct list_head obd_exports;
605 struct list_head obd_unlinked_exports;
606 struct list_head obd_delayed_exports;
607 struct list_head obd_lwp_list;
609 spinlock_t obd_nid_lock;
610 struct ldlm_namespace *obd_namespace;
611 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
612 /* a spinlock is OK for what we do now, may need a semaphore later */
613 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
614 struct mutex obd_dev_mutex;
615 __u64 obd_last_committed;
616 spinlock_t obd_osfs_lock;
617 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
619 struct lvfs_run_ctxt obd_lvfs_ctxt;
620 struct obd_llog_group obd_olg; /* default llog group */
621 struct obd_device *obd_observer;
622 struct rw_semaphore obd_observer_link_sem;
623 struct obd_notify_upcall obd_upcall;
624 struct obd_export *obd_self_export;
625 struct obd_export *obd_lwp_export;
626 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
627 struct list_head obd_exports_timed;
628 time_t obd_eviction_timer; /* for ping evictor */
630 int obd_max_recoverable_clients;
631 atomic_t obd_connected_clients;
632 int obd_stale_clients;
633 /* this lock protects all recovery list_heads, timer and
634 * obd_next_recovery_transno value */
635 spinlock_t obd_recovery_task_lock;
636 __u64 obd_next_recovery_transno;
637 int obd_replayed_requests;
638 int obd_requests_queued_for_recovery;
639 wait_queue_head_t obd_next_transno_waitq;
640 /* protected by obd_recovery_task_lock */
641 struct timer_list obd_recovery_timer;
643 time_t obd_recovery_start;
644 /* seconds, for lprocfs_status */
645 time_t obd_recovery_end;
646 int obd_recovery_time_hard;
647 int obd_recovery_timeout;
648 int obd_recovery_ir_factor;
650 /* new recovery stuff from CMD2 */
651 struct target_recovery_data obd_recovery_data;
652 int obd_replayed_locks;
653 atomic_t obd_req_replay_clients;
654 atomic_t obd_lock_replay_clients;
655 /* all lists are protected by obd_recovery_task_lock */
656 struct list_head obd_req_replay_queue;
657 struct list_head obd_lock_replay_queue;
658 struct list_head obd_final_req_queue;
661 #ifdef HAVE_SERVER_SUPPORT
662 struct obd_device_target obt;
663 struct filter_obd filter;
665 struct echo_obd echo;
667 struct client_obd cli;
668 struct echo_client_obd echo_client;
672 /* Fields used by LProcFS */
673 unsigned int obd_cntr_base;
674 struct lprocfs_stats *obd_stats;
676 unsigned int obd_md_cntr_base;
677 struct lprocfs_stats *obd_md_stats;
679 struct proc_dir_entry *obd_proc_entry;
680 struct proc_dir_entry *obd_proc_exports_entry;
681 struct proc_dir_entry *obd_svc_procroot;
682 struct lprocfs_stats *obd_svc_stats;
683 struct lprocfs_vars *obd_vars;
684 atomic_t obd_evict_inprogress;
685 wait_queue_head_t obd_evict_inprogress_waitq;
686 struct list_head obd_evict_list; /* protected with pet_lock */
689 * Ldlm pool part. Save last calculated SLV and Limit.
691 rwlock_t obd_pool_lock;
696 * A list of outstanding class_incref()'s against this obd. For
699 struct lu_ref obd_reference;
701 int obd_conn_inprogress;
704 enum obd_cleanup_stage {
705 /* Special case hack for MDS LOVs */
707 /* can be directly mapped to .ldto_device_fini() */
711 /* get/set_info keys */
712 #define KEY_ASYNC "async"
713 #define KEY_CAPA_KEY "capa_key"
714 #define KEY_CHANGELOG_CLEAR "changelog_clear"
715 #define KEY_FID2PATH "fid2path"
716 #define KEY_CHECKSUM "checksum"
717 #define KEY_CLEAR_FS "clear_fs"
718 #define KEY_CONN_DATA "conn_data"
719 #define KEY_EVICT_BY_NID "evict_by_nid"
720 #define KEY_FIEMAP "fiemap"
721 #define KEY_FLUSH_CTX "flush_ctx"
722 #define KEY_GRANT_SHRINK "grant_shrink"
723 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
724 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
725 #define KEY_INTERMDS "inter_mds"
726 #define KEY_LAST_ID "last_id"
727 #define KEY_LAST_FID "last_fid"
728 #define KEY_LOVDESC "lovdesc"
729 #define KEY_MAX_EASIZE "max_easize"
730 #define KEY_DEFAULT_EASIZE "default_easize"
731 #define KEY_MAX_COOKIESIZE "max_cookiesize"
732 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
733 #define KEY_MGSSEC "mgssec"
734 #define KEY_READ_ONLY "read-only"
735 #define KEY_REGISTER_TARGET "register_target"
736 #define KEY_SET_FS "set_fs"
737 #define KEY_TGT_COUNT "tgt_count"
738 /* KEY_SET_INFO in lustre_idl.h */
739 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
741 #define KEY_CACHE_SET "cache_set"
742 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
743 #define KEY_OSP_CONNECTED "osp_connected"
747 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
748 #define IT_OPEN (1 << 0)
749 #define IT_CREAT (1 << 1)
750 #define IT_READDIR (1 << 2)
751 #define IT_GETATTR (1 << 3)
752 #define IT_LOOKUP (1 << 4)
753 #define IT_UNLINK (1 << 5)
754 #define IT_TRUNC (1 << 6)
755 #define IT_GETXATTR (1 << 7)
756 #define IT_EXEC (1 << 8)
757 #define IT_PIN (1 << 9)
758 #define IT_LAYOUT (1 << 10)
759 #define IT_QUOTA_DQACQ (1 << 11)
760 #define IT_QUOTA_CONN (1 << 12)
761 #define IT_SETXATTR (1 << 13)
763 static inline int it_to_lock_mode(struct lookup_intent *it)
765 /* CREAT needs to be tested before open (both could be set) */
766 if (it->it_op & IT_CREAT)
768 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
771 else if (it->it_op & IT_READDIR)
773 else if (it->it_op & IT_GETXATTR)
775 else if (it->it_op & IT_SETXATTR)
778 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
783 CLI_SET_MEA = 1 << 0,
784 CLI_RM_ENTRY = 1 << 1,
787 CLI_MIGRATE = 1 << 4,
791 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
792 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
793 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
794 struct lu_fid op_fid4; /* to the operation locks. */
795 u32 op_mds; /* what mds server open will go to */
796 struct lustre_handle op_handle;
801 struct lmv_stripe_md *op_mea1;
802 struct lmv_stripe_md *op_mea2;
803 __u32 op_suppgids[2];
810 /* iattr fields and blocks. */
811 struct iattr op_attr;
812 loff_t op_attr_blocks;
813 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
814 __u64 op_valid; /* OBD_MD_* */
816 /* Size-on-MDS epoch and flags. */
818 enum md_op_flags op_flags;
821 struct obd_capa *op_capa1;
822 struct obd_capa *op_capa2;
824 /* Various operation flags. */
825 enum mds_op_bias op_bias;
827 /* Used by readdir */
828 unsigned int op_max_pages;
830 /* used to transfer info between the stacks of MD client
831 * see enum op_cli_flags */
832 enum md_cli_flags op_cli_flags;
834 /* File object data version for HSM release, on client */
835 __u64 op_data_version;
836 struct lustre_handle op_lease_handle;
838 /* default stripe offset */
839 __u32 op_default_stripe_offset;
843 int (*md_blocking_ast)(struct ldlm_lock *lock,
844 struct ldlm_lock_desc *desc,
845 void *data, int flag);
848 struct md_enqueue_info;
849 /* metadata stat-ahead */
850 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
851 struct md_enqueue_info *minfo,
854 struct md_enqueue_info {
855 struct md_op_data mi_data;
856 struct lookup_intent mi_it;
857 struct lustre_handle mi_lockh;
858 struct inode *mi_dir;
859 md_enqueue_cb_t mi_cb;
864 struct module *o_owner;
865 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
866 void *karg, void __user *uarg);
867 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
868 __u32 keylen, void *key,
869 __u32 *vallen, void *val,
870 struct lov_stripe_md *lsm);
871 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
872 __u32 keylen, void *key,
873 __u32 vallen, void *val,
874 struct ptlrpc_request_set *set);
875 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
876 int (*o_precleanup)(struct obd_device *dev,
877 enum obd_cleanup_stage cleanup_stage);
878 int (*o_cleanup)(struct obd_device *dev);
879 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
880 int (*o_postrecov)(struct obd_device *dev);
881 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
883 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
884 /* connect to the target device with given connection
885 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
886 * granted by the target, which are guaranteed to be a subset of flags
887 * asked for. If @ocd == NULL, use default parameters. */
888 int (*o_connect)(const struct lu_env *env,
889 struct obd_export **exp, struct obd_device *src,
890 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
892 int (*o_reconnect)(const struct lu_env *env,
893 struct obd_export *exp, struct obd_device *src,
894 struct obd_uuid *cluuid,
895 struct obd_connect_data *ocd,
897 int (*o_disconnect)(struct obd_export *exp);
899 /* Initialize/finalize fids infrastructure. */
900 int (*o_fid_init)(struct obd_device *obd,
901 struct obd_export *exp, enum lu_cli_type type);
902 int (*o_fid_fini)(struct obd_device *obd);
904 /* Allocate new fid according to passed @hint. */
905 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
906 struct lu_fid *fid, struct md_op_data *op_data);
909 * Object with @fid is getting deleted, we may want to do something
912 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
913 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
914 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
915 __u64 max_age, struct ptlrpc_request_set *set);
916 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
917 struct lov_stripe_md *mem_src);
918 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
919 struct lov_mds_md *disk_src, int disk_len);
920 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
921 struct obdo *oa, struct obd_trans_info *oti);
922 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
923 struct obdo *oa, struct obd_trans_info *oti);
924 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
925 struct obd_info *oinfo, struct obd_trans_info *oti);
926 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
927 struct obd_trans_info *oti,
928 struct ptlrpc_request_set *rqset);
929 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
930 struct obd_info *oinfo);
931 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
932 struct ptlrpc_request_set *set);
933 int (*o_preprw)(const struct lu_env *env, int cmd,
934 struct obd_export *exp, struct obdo *oa, int objcount,
935 struct obd_ioobj *obj, struct niobuf_remote *remote,
936 int *nr_pages, struct niobuf_local *local,
937 struct obd_trans_info *oti, struct lustre_capa *capa);
938 int (*o_commitrw)(const struct lu_env *env, int cmd,
939 struct obd_export *exp, struct obdo *oa,
940 int objcount, struct obd_ioobj *obj,
941 struct niobuf_remote *remote, int pages,
942 struct niobuf_local *local,
943 struct obd_trans_info *oti, int rc);
944 int (*o_init_export)(struct obd_export *exp);
945 int (*o_destroy_export)(struct obd_export *exp);
947 int (*o_import_event)(struct obd_device *, struct obd_import *,
948 enum obd_import_event);
950 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
951 enum obd_notify_event ev, void *data);
953 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
954 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
957 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
958 struct obd_quotactl *);
959 int (*o_quotactl)(struct obd_device *, struct obd_export *,
960 struct obd_quotactl *);
962 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
965 int (*o_pool_new)(struct obd_device *obd, char *poolname);
966 int (*o_pool_del)(struct obd_device *obd, char *poolname);
967 int (*o_pool_add)(struct obd_device *obd, char *poolname,
969 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
971 void (*o_getref)(struct obd_device *obd);
972 void (*o_putref)(struct obd_device *obd);
974 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
975 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
976 * Also, add a wrapper function in include/linux/obd_class.h. */
981 struct mdt_body *body;
982 struct lov_stripe_md *lsm;
983 struct lmv_stripe_md *lmv;
984 #ifdef CONFIG_FS_POSIX_ACL
985 struct posix_acl *posix_acl;
987 struct mdt_remote_perm *remote_perm;
988 struct obd_capa *mds_capa;
989 struct obd_capa *oss_capa;
992 struct md_open_data {
993 struct obd_client_handle *mod_och;
994 struct ptlrpc_request *mod_open_req;
995 struct ptlrpc_request *mod_close_req;
996 atomic_t mod_refcount;
1000 struct obd_client_handle {
1001 struct lustre_handle och_fh;
1002 struct lu_fid och_fid;
1003 struct md_open_data *och_mod;
1004 struct lustre_handle och_lease_handle; /* open lock for lease */
1009 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1011 struct lookup_intent;
1015 /* Every operation from MD_STATS_FIRST_OP up to and including
1016 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1017 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1018 * Operations after MD_STATS_LAST_OP are excluded from stats.
1019 * There are a few reasons for doing this: we prune the 17
1020 * counters which will be of minimal use in understanding
1021 * metadata utilization, we save memory by allocating 15
1022 * instead of 32 counters, we save cycles by not counting.
1024 * MD_STATS_FIRST_OP must be the first member of md_ops.
1026 #define MD_STATS_FIRST_OP m_close
1027 int (*m_close)(struct obd_export *, struct md_op_data *,
1028 struct md_open_data *, struct ptlrpc_request **);
1030 int (*m_create)(struct obd_export *, struct md_op_data *,
1031 const void *, size_t, umode_t, uid_t, gid_t,
1032 cfs_cap_t, __u64, struct ptlrpc_request **);
1034 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1035 const union ldlm_policy_data *,
1036 struct lookup_intent *, struct md_op_data *,
1037 struct lustre_handle *, __u64);
1039 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1040 struct ptlrpc_request **);
1042 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1043 struct lookup_intent *,
1044 struct ptlrpc_request **,
1045 ldlm_blocking_callback, __u64);
1047 int (*m_link)(struct obd_export *, struct md_op_data *,
1048 struct ptlrpc_request **);
1050 int (*m_rename)(struct obd_export *, struct md_op_data *,
1051 const char *, size_t, const char *, size_t,
1052 struct ptlrpc_request **);
1054 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1055 size_t , void *, size_t, struct ptlrpc_request **,
1056 struct md_open_data **mod);
1058 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1059 struct obd_capa *, struct ptlrpc_request **);
1061 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1062 struct md_callback *cb_op, __u64 hash_offset,
1063 struct page **ppage);
1065 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1066 struct ptlrpc_request **);
1068 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1069 struct obd_capa *, u64, const char *,
1070 const char *, int, int, int, __u32,
1071 struct ptlrpc_request **);
1073 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1074 struct obd_capa *, u64, const char *,
1075 const char *, int, int, int,
1076 struct ptlrpc_request **);
1078 int (*m_intent_getattr_async)(struct obd_export *,
1079 struct md_enqueue_info *,
1080 struct ldlm_enqueue_info *);
1082 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1083 struct lu_fid *, __u64 *bits);
1085 #define MD_STATS_LAST_OP m_revalidate_lock
1087 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1088 struct obd_capa **);
1090 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1092 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1093 ldlm_iterator_t, void *);
1095 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1096 struct md_open_data *);
1098 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1099 struct ptlrpc_request **);
1101 int (*m_init_ea_size)(struct obd_export *, __u32, __u32, __u32, __u32);
1103 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1104 struct obd_export *, struct obd_export *,
1105 struct lustre_md *);
1107 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1109 int (*m_merge_attr)(struct obd_export *,
1110 const struct lmv_stripe_md *lsm,
1111 struct cl_attr *attr, ldlm_blocking_callback);
1113 int (*m_set_open_replay_data)(struct obd_export *,
1114 struct obd_client_handle *,
1115 struct lookup_intent *);
1117 int (*m_clear_open_replay_data)(struct obd_export *,
1118 struct obd_client_handle *);
1120 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1122 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1123 const struct lu_fid *, ldlm_type_t,
1124 ldlm_policy_data_t *, ldlm_mode_t,
1125 struct lustre_handle *);
1127 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1128 ldlm_policy_data_t *, ldlm_mode_t,
1129 ldlm_cancel_flags_t flags, void *opaque);
1131 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1132 renew_capa_cb_t cb);
1134 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1135 const struct req_msg_field *, struct obd_capa **);
1137 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1138 struct obd_capa *, __u32,
1139 struct ptlrpc_request **);
1141 int (*m_get_fid_from_lsm)(struct obd_export *,
1142 const struct lmv_stripe_md *,
1143 const char *name, int namelen,
1144 struct lu_fid *fid);
1147 struct lsm_operations {
1148 void (*lsm_free)(struct lov_stripe_md *);
1149 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *,
1150 loff_t *, loff_t *);
1151 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *,
1152 loff_t *, loff_t *);
1153 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1154 __u16 *stripe_count);
1155 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1156 struct lov_mds_md *lmm);
1159 extern const struct lsm_operations lsm_v1_ops;
1160 extern const struct lsm_operations lsm_v3_ops;
1161 static inline const struct lsm_operations *lsm_op_find(u32 magic)
1169 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1174 static inline struct md_open_data *obd_mod_alloc(void)
1176 struct md_open_data *mod;
1180 atomic_set(&mod->mod_refcount, 1);
1184 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1185 #define obd_mod_put(mod) \
1187 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1188 if ((mod)->mod_open_req) \
1189 ptlrpc_req_finished((mod)->mod_open_req); \
1190 OBD_FREE_PTR(mod); \
1194 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1195 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1197 /* return 1 if client should be resend request */
1198 static inline int client_should_resend(int resend, struct client_obd *cli)
1200 return atomic_read(&cli->cl_resends) ?
1201 atomic_read(&cli->cl_resends) > resend : 1;
1205 * Return device name for this device
1207 * XXX: lu_device is declared before obd_device, while a pointer pointing
1208 * back to obd_device in lu_device, so this helper function defines here
1209 * instead of in lu_object.h
1211 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1213 return lu_dev->ld_obd->obd_name;
1216 static inline bool filename_is_volatile(const char *name, size_t namelen,
1222 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1225 /* caller does not care of idx */
1229 /* volatile file, the MDT can be set from name */
1230 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1231 /* if no MDT is specified, use std way */
1232 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1234 /* test for no MDT idx case */
1235 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1236 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1240 /* we have an idx, read it */
1241 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1242 *idx = strtoul(start, &end, 16);
1244 * no digit, no trailing :, negative value
1246 if (((*idx == 0) && (end == start)) ||
1247 (*end != ':') || (*idx < 0))
1252 /* bad format of mdt idx, we cannot return an error
1253 * to caller so we use hash algo */
1254 CERROR("Bad volatile file name format: %s\n",
1255 name + LUSTRE_VOLATILE_HDR_LEN);
1259 static inline int cli_brw_size(struct obd_device *obd)
1261 LASSERT(obd != NULL);
1262 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1265 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1266 * of the client should be increased accordingly to avoid sending fragmented
1267 * RPCs over the network when the client runs out of the maximum dirty space
1268 * when so many RPCs are being generated.
1270 static inline void client_adjust_max_dirty(struct client_obd *cli)
1273 if (cli->cl_dirty_max_pages <= 0)
1274 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1275 >> PAGE_CACHE_SHIFT;
1277 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1278 cli->cl_max_pages_per_rpc;
1280 if (dirty_max > cli->cl_dirty_max_pages)
1281 cli->cl_dirty_max_pages = dirty_max;
1284 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1285 cli->cl_dirty_max_pages = totalram_pages / 8;
1288 #endif /* __OBD_H */