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_capa.h>
56 #define MAX_OBD_DEVICES 8192
64 struct lov_oinfo { /* per-stripe data structure */
65 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
66 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
67 int loi_ost_gen; /* generation of this loi_ost_idx */
69 unsigned long loi_kms_valid:1;
70 __u64 loi_kms; /* known minimum size */
71 struct ost_lvb loi_lvb;
72 struct osc_async_rc loi_ar;
75 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
78 oinfo->loi_kms_valid = 1;
81 static inline void loi_init(struct lov_oinfo *loi)
85 struct lov_stripe_md {
88 pid_t lsm_lock_owner; /* debugging */
90 /* maximum possible file size, might change as OSTs status changes,
91 * e.g. disconnected, deactivated */
95 struct ost_id lw_object_oi; /* lov object id/seq */
97 /* LOV-private members start here -- only for use in lov/. */
99 __u32 lw_stripe_size; /* size of the stripe */
100 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
101 __u16 lw_stripe_count; /* number of objects being striped over */
102 __u16 lw_layout_gen; /* generation of the layout */
103 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
106 struct lov_oinfo *lsm_oinfo[0];
109 #define lsm_oi lsm_wire.lw_object_oi
110 #define lsm_magic lsm_wire.lw_magic
111 #define lsm_layout_gen lsm_wire.lw_layout_gen
112 #define lsm_stripe_size lsm_wire.lw_stripe_size
113 #define lsm_pattern lsm_wire.lw_pattern
114 #define lsm_stripe_count lsm_wire.lw_stripe_count
115 #define lsm_pool_name lsm_wire.lw_pool_name
117 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
119 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
122 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
126 if (lsm_is_released(lsm))
131 static inline int lov_stripe_md_size(unsigned int stripe_count)
133 struct lov_stripe_md lsm;
135 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
140 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
142 /* obd info for a particular level (lov, osc). */
144 /* Lock policy. It keeps an extent which is specific for a particular
145 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
146 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
147 ldlm_policy_data_t oi_policy;
148 /* Flags used for set request specific flags:
149 - while lock handling, the flags obtained on the enqueue
150 request are set here.
151 - while stats, the flags used for control delay/resend.
152 - while setattr, the flags used for distinguish punch operation
155 /* lsm data specific for every OSC. */
156 struct lov_stripe_md *oi_md;
157 /* obdo data specific for every OSC, if needed at all. */
159 /* statfs data specific for every OSC, if needed at all. */
160 struct obd_statfs *oi_osfs;
161 /* An update callback which is called to update some data on upper
162 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
163 * request in osc level for enqueue requests. It is also possible to
164 * update some caller data from LOV layer if needed. */
165 obd_enqueue_update_f oi_cb_up;
166 /* oss capability, its type is obd_capa in client to avoid copy.
167 * in contrary its type is lustre_capa in OSS. */
172 struct list_head typ_chain;
173 struct obd_ops *typ_dt_ops;
174 struct md_ops *typ_md_ops;
175 struct proc_dir_entry *typ_procroot;
176 struct proc_dir_entry *typ_procsym;
177 __u32 typ_sym_filter;
180 struct lu_device_type *typ_lu;
181 spinlock_t obd_type_lock;
191 struct timeout_item {
192 enum timeout_event ti_event;
193 cfs_time_t ti_timeout;
196 struct list_head ti_obd_list;
197 struct list_head ti_chain;
200 #define OBD_MAX_RIF_DEFAULT 8
201 #define OBD_MAX_RIF_MAX 512
202 #define OSC_MAX_RIF_MAX 256
203 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
204 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
205 #define OSC_DEFAULT_RESENDS 10
207 /* possible values for fo_sync_lock_cancel */
209 NEVER_SYNC_ON_CANCEL = 0,
210 BLOCKING_SYNC_ON_CANCEL = 1,
211 ALWAYS_SYNC_ON_CANCEL = 2,
212 NUM_SYNC_ON_CANCEL_STATES
218 struct rw_semaphore cl_sem;
219 struct obd_uuid cl_target_uuid;
220 struct obd_import *cl_import; /* ptlrpc connection state */
222 /* max_mds_easize is purely a performance thing so we don't have to
223 * call obd_size_diskmd() all the time. */
224 int cl_default_mds_easize;
225 int cl_max_mds_easize;
226 int cl_default_mds_cookiesize;
227 int cl_max_mds_cookiesize;
229 enum lustre_sec_part cl_sp_me;
230 enum lustre_sec_part cl_sp_to;
231 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
233 /* the grant values are protected by loi_list_lock below */
234 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
235 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
236 unsigned long cl_dirty_transit; /* dirty synchronous */
237 unsigned long cl_avail_grant; /* bytes of credit for ost */
238 unsigned long cl_lost_grant; /* lost credits (trunc) */
240 /* since we allocate grant by blocks, we don't know how many grant will
241 * be used to add a page into cache. As a solution, we reserve maximum
242 * grant before trying to dirty a page and unreserve the rest.
243 * See osc_{reserve|unreserve}_grant for details. */
244 long cl_reserved_grant;
245 struct list_head cl_cache_waiters; /* waiting for cache/grant */
246 cfs_time_t cl_next_shrink_grant; /* jiffies */
247 struct list_head cl_grant_shrink_list; /* Timeout event list */
248 int cl_grant_shrink_interval; /* seconds */
250 /* A chunk is an optimal size used by osc_extent to determine
251 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
253 int cl_extent_tax; /* extent overhead, by bytes */
255 /* keep track of objects that have lois that contain pages which
256 * have been queued for async brw. this lock also protects the
257 * lists of osc_client_pages that hang off of the loi */
259 * ->cl_loi_list_lock protects consistency of
260 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
261 * ->ap_completion() call-backs are executed under this lock. As we
262 * cannot guarantee that these call-backs never block on all platforms
263 * (as a matter of fact they do block on Mac OS X), type of
264 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
265 * and blocking mutex on Mac OS X. (Alternative is to make this lock
266 * blocking everywhere, but we don't want to slow down fast-path of
267 * our main platform.)
269 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
270 * with client_obd_list_{un,}lock() and
271 * client_obd_list_lock_{init,done}() functions.
273 * NB by Jinshan: though field names are still _loi_, but actually
274 * osc_object{}s are in the list.
276 client_obd_lock_t cl_loi_list_lock;
277 struct list_head cl_loi_ready_list;
278 struct list_head cl_loi_hp_ready_list;
279 struct list_head cl_loi_write_list;
280 struct list_head cl_loi_read_list;
281 __u32 cl_r_in_flight;
282 __u32 cl_w_in_flight;
283 /* just a sum of the loi/lop pending numbers to be exported by /proc */
284 atomic_t cl_pending_w_pages;
285 atomic_t cl_pending_r_pages;
286 __u32 cl_max_pages_per_rpc;
287 __u32 cl_max_rpcs_in_flight;
288 struct obd_histogram cl_read_rpc_hist;
289 struct obd_histogram cl_write_rpc_hist;
290 struct obd_histogram cl_read_page_hist;
291 struct obd_histogram cl_write_page_hist;
292 struct obd_histogram cl_read_offset_hist;
293 struct obd_histogram cl_write_offset_hist;
295 /* lru for osc caching pages */
296 struct cl_client_cache *cl_cache;
297 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
298 atomic_t *cl_lru_left;
299 atomic_t cl_lru_busy;
300 atomic_t cl_lru_shrinkers;
301 atomic_t cl_lru_in_list;
302 struct list_head cl_lru_list; /* lru page list */
303 client_obd_lock_t cl_lru_list_lock; /* page list protector */
304 atomic_t cl_unstable_count;
306 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
307 atomic_t cl_destroy_in_flight;
308 wait_queue_head_t cl_destroy_waitq;
310 struct mdc_rpc_lock *cl_rpc_lock;
311 struct mdc_rpc_lock *cl_close_lock;
314 struct mutex cl_mgc_mutex;
315 struct local_oid_storage *cl_mgc_los;
316 struct dt_object *cl_mgc_configs_dir;
317 atomic_t cl_mgc_refcount;
318 struct obd_export *cl_mgc_mgsexp;
320 /* checksumming for data sent over the network */
321 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
322 /* supported checksum types that are worked out at connect time */
323 __u32 cl_supp_cksum_types;
324 /* checksum algorithm to be used */
325 cksum_type_t cl_cksum_type;
327 /* also protected by the poorly named _loi_list_lock lock above */
328 struct osc_async_rc cl_ar;
330 /* used by quotacheck when the servers are older than 2.4 */
331 int cl_qchk_stat; /* quotacheck stat of the peer */
332 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
333 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
334 #warning "please consider removing quotacheck compatibility code"
337 /* sequence manager */
338 struct lu_client_seq *cl_seq;
340 atomic_t cl_resends; /* resend count */
342 /* ptlrpc work for writeback in ptlrpcd context */
343 void *cl_writeback_work;
345 /* hash tables for osc_quota_info */
346 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
348 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
355 struct echo_client_obd {
356 struct obd_export *ec_exp; /* the local connection to osc/lov */
358 struct list_head ec_objects;
359 struct list_head ec_locks;
364 /* Generic subset of OSTs */
366 __u32 *op_array; /* array of index of
368 unsigned int op_count; /* number of OSTs in the array */
369 unsigned int op_size; /* allocated size of lp_array */
370 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
373 /* allow statfs data caching for 1 second */
374 #define OBD_STATFS_CACHE_SECONDS 1
376 struct lov_tgt_desc {
377 struct list_head ltd_kill;
378 struct obd_uuid ltd_uuid;
379 struct obd_device *ltd_obd;
380 struct obd_export *ltd_exp;
382 __u32 ltd_index; /* index in lov_obd->tgts */
383 unsigned long ltd_active:1,/* is this target up for requests */
384 ltd_activate:1,/* should target be activated */
385 ltd_reap:1; /* should this target be deleted */
389 struct lov_desc desc;
390 struct lov_tgt_desc **lov_tgts; /* sparse array */
391 struct ost_pool lov_packed; /* all OSTs in a packed
393 struct mutex lov_lock;
394 struct obd_connect_data lov_ocd;
395 struct proc_dir_entry *targets_proc_entry;
396 atomic_t lov_refcount;
397 __u32 lov_death_row; /* tgts scheduled to be deleted */
398 __u32 lov_tgt_size; /* size of tgts array */
401 cfs_hash_t *lov_pools_hash_body; /* used for key access */
402 struct list_head lov_pool_list; /* used for sequential access */
403 struct proc_dir_entry *lov_pool_proc_entry;
404 enum lustre_sec_part lov_sp_me;
406 /* Cached LRU and unstable data from upper layer */
409 struct rw_semaphore lov_notify_lock;
412 struct lmv_tgt_desc {
413 struct obd_uuid ltd_uuid;
414 struct obd_export *ltd_exp;
416 struct mutex ltd_fid_mutex;
417 unsigned long ltd_active:1; /* target up for requests */
420 enum placement_policy {
421 PLACEMENT_CHAR_POLICY = 0,
422 PLACEMENT_NID_POLICY = 1,
423 PLACEMENT_INVAL_POLICY = 2,
427 typedef enum placement_policy placement_policy_t;
431 struct lu_client_fld lmv_fld;
433 placement_policy_t lmv_placement;
434 struct lmv_desc desc;
435 struct obd_uuid cluuid;
436 struct obd_export *exp;
437 struct proc_dir_entry *targets_proc_entry;
439 struct mutex init_mutex;
444 int max_def_cookiesize;
447 int tgts_size; /* size of tgts array */
448 struct lmv_tgt_desc **tgts;
450 struct obd_connect_data conn_data;
453 struct niobuf_local {
454 __u64 lnb_file_offset;
455 __u32 lnb_page_offset;
458 struct page *lnb_page;
463 #define LUSTRE_FLD_NAME "fld"
464 #define LUSTRE_SEQ_NAME "seq"
466 #define LUSTRE_MDD_NAME "mdd"
467 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
468 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
469 #define LUSTRE_VVP_NAME "vvp"
470 #define LUSTRE_LMV_NAME "lmv"
471 #define LUSTRE_SLP_NAME "slp"
472 #define LUSTRE_LOD_NAME "lod"
473 #define LUSTRE_OSP_NAME "osp"
474 #define LUSTRE_LWP_NAME "lwp"
476 /* obd device type names */
477 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
478 #define LUSTRE_MDS_NAME "mds"
479 #define LUSTRE_MDT_NAME "mdt"
480 #define LUSTRE_MDC_NAME "mdc"
481 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
482 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
483 #define LUSTRE_OSC_NAME "osc"
484 #define LUSTRE_LOV_NAME "lov"
485 #define LUSTRE_MGS_NAME "mgs"
486 #define LUSTRE_MGC_NAME "mgc"
488 #define LUSTRE_ECHO_NAME "obdecho"
489 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
490 #define LUSTRE_QMT_NAME "qmt"
492 /* Constant obd names (post-rename) */
493 #define LUSTRE_MDS_OBDNAME "MDS"
494 #define LUSTRE_OSS_OBDNAME "OSS"
495 #define LUSTRE_MGS_OBDNAME "MGS"
496 #define LUSTRE_MGC_OBDNAME "MGC"
498 struct obd_trans_info {
500 /* Only used on the server side for tracking acks. */
501 struct oti_req_ack_lock {
502 struct lustre_handle lock;
506 struct llog_cookie oti_onecookie;
507 struct llog_cookie *oti_logcookies;
510 __u64 oti_pre_version;
514 * Events signalled through obd_notify() upcall-chain.
516 enum obd_notify_event {
519 /* Device connect start */
521 /* Device activated */
523 /* Device deactivated */
525 /* Device disconnected */
527 /* Connect data for import were changed */
530 OBD_NOTIFY_SYNC_NONBLOCK,
532 /* Configuration event */
534 /* Administratively deactivate/activate event */
535 OBD_NOTIFY_DEACTIVATE,
539 /* bit-mask flags for config events */
541 CONFIG_LOG = 0x1, /* finished processing config log */
542 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
543 CONFIG_TARGET = 0x4 /* one target is added */
547 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
548 * and liblustre being main examples).
550 struct obd_notify_upcall {
551 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
552 enum obd_notify_event ev, void *owner, void *data);
553 /* Opaque datum supplied by upper layer listener */
557 struct target_recovery_data {
558 svc_handler_t trd_recovery_handler;
559 pid_t trd_processing_task;
560 struct completion trd_starting;
561 struct completion trd_finishing;
564 struct obd_llog_group {
565 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
566 wait_queue_head_t olg_waitq;
568 struct mutex olg_cat_processing;
571 /* corresponds to one of the obd's */
572 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
575 struct obd_type *obd_type;
578 /* common and UUID name of this device */
579 char obd_name[MAX_OBD_NAME];
580 struct obd_uuid obd_uuid;
582 struct lu_device *obd_lu_dev;
584 /* bitfield modification is protected by obd_dev_lock */
586 obd_attached:1, /* finished attach */
587 obd_set_up:1, /* finished setup */
588 obd_recovering:1, /* there are recoverable clients */
589 obd_abort_recovery:1, /* recovery expired */
590 obd_version_recov:1, /* obd uses version checking */
591 obd_replayable:1, /* recovery is enabled;
593 obd_no_transno:1, /* no committed-transno notification */
594 obd_no_recov:1, /* fail instead of retry messages */
595 obd_stopping:1, /* started cleanup */
596 obd_starting:1, /* started setup */
597 obd_force:1, /* cleanup with > 0 obd refcount */
598 obd_fail:1, /* cleanup with failover */
599 obd_no_conn:1, /* deny new connections */
600 obd_inactive:1, /* device active/inactive
601 * (for /proc/status only!!) */
602 obd_no_ir:1, /* no imperative recovery. */
603 obd_process_conf:1, /* device is processing mgs config */
604 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
606 /* use separate field as it is set in interrupt to don't mess with
607 * protection of other bits using _bh lock */
608 unsigned long obd_recovery_expired:1;
609 /* uuid-export hash body */
610 cfs_hash_t *obd_uuid_hash;
611 /* nid-export hash body */
612 cfs_hash_t *obd_nid_hash;
614 cfs_hash_t *obd_nid_stats_hash;
615 struct list_head obd_nid_stats;
616 atomic_t obd_refcount;
617 struct list_head obd_exports;
618 struct list_head obd_unlinked_exports;
619 struct list_head obd_delayed_exports;
620 struct list_head obd_lwp_list;
622 spinlock_t obd_nid_lock;
623 struct ldlm_namespace *obd_namespace;
624 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
625 /* a spinlock is OK for what we do now, may need a semaphore later */
626 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
627 struct mutex obd_dev_mutex;
628 __u64 obd_last_committed;
629 spinlock_t obd_osfs_lock;
630 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
632 struct lvfs_run_ctxt obd_lvfs_ctxt;
633 struct obd_llog_group obd_olg; /* default llog group */
634 struct obd_device *obd_observer;
635 struct rw_semaphore obd_observer_link_sem;
636 struct obd_notify_upcall obd_upcall;
637 struct obd_export *obd_self_export;
638 struct obd_export *obd_lwp_export;
639 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
640 struct list_head obd_exports_timed;
641 time_t obd_eviction_timer; /* for ping evictor */
643 int obd_max_recoverable_clients;
644 atomic_t obd_connected_clients;
645 int obd_stale_clients;
646 /* this lock protects all recovery list_heads, timer and
647 * obd_next_recovery_transno value */
648 spinlock_t obd_recovery_task_lock;
649 __u64 obd_next_recovery_transno;
650 int obd_replayed_requests;
651 int obd_requests_queued_for_recovery;
652 wait_queue_head_t obd_next_transno_waitq;
653 /* protected by obd_recovery_task_lock */
654 struct timer_list obd_recovery_timer;
656 time_t obd_recovery_start;
657 /* seconds, for lprocfs_status */
658 time_t obd_recovery_end;
659 int obd_recovery_time_hard;
660 int obd_recovery_timeout;
661 int obd_recovery_ir_factor;
663 /* new recovery stuff from CMD2 */
664 struct target_recovery_data obd_recovery_data;
665 int obd_replayed_locks;
666 atomic_t obd_req_replay_clients;
667 atomic_t obd_lock_replay_clients;
668 /* all lists are protected by obd_recovery_task_lock */
669 struct list_head obd_req_replay_queue;
670 struct list_head obd_lock_replay_queue;
671 struct list_head obd_final_req_queue;
674 #ifdef HAVE_SERVER_SUPPORT
675 struct obd_device_target obt;
676 struct filter_obd filter;
678 struct echo_obd echo;
680 struct client_obd cli;
681 struct echo_client_obd echo_client;
685 /* Fields used by LProcFS */
686 unsigned int obd_cntr_base;
687 struct lprocfs_stats *obd_stats;
689 unsigned int obd_md_cntr_base;
690 struct lprocfs_stats *obd_md_stats;
692 struct proc_dir_entry *obd_proc_entry;
693 struct proc_dir_entry *obd_proc_exports_entry;
694 struct proc_dir_entry *obd_svc_procroot;
695 struct lprocfs_stats *obd_svc_stats;
696 struct lprocfs_seq_vars *obd_vars;
697 atomic_t obd_evict_inprogress;
698 wait_queue_head_t obd_evict_inprogress_waitq;
699 struct list_head obd_evict_list; /* protected with pet_lock */
702 * Ldlm pool part. Save last calculated SLV and Limit.
704 rwlock_t obd_pool_lock;
709 * A list of outstanding class_incref()'s against this obd. For
712 struct lu_ref obd_reference;
714 int obd_conn_inprogress;
717 enum obd_cleanup_stage {
718 /* Special case hack for MDS LOVs */
720 /* can be directly mapped to .ldto_device_fini() */
724 /* get/set_info keys */
725 #define KEY_ASYNC "async"
726 #define KEY_CAPA_KEY "capa_key"
727 #define KEY_CHANGELOG_CLEAR "changelog_clear"
728 #define KEY_FID2PATH "fid2path"
729 #define KEY_CHECKSUM "checksum"
730 #define KEY_CLEAR_FS "clear_fs"
731 #define KEY_CONN_DATA "conn_data"
732 #define KEY_EVICT_BY_NID "evict_by_nid"
733 #define KEY_FIEMAP "fiemap"
734 #define KEY_FLUSH_CTX "flush_ctx"
735 #define KEY_GRANT_SHRINK "grant_shrink"
736 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
737 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
738 #define KEY_INTERMDS "inter_mds"
739 #define KEY_LAST_ID "last_id"
740 #define KEY_LAST_FID "last_fid"
741 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
742 #define KEY_LOVDESC "lovdesc"
743 #define KEY_MAX_EASIZE "max_easize"
744 #define KEY_DEFAULT_EASIZE "default_easize"
745 #define KEY_MAX_COOKIESIZE "max_cookiesize"
746 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
747 #define KEY_MDS_CONN "mds_conn"
748 #define KEY_MGSSEC "mgssec"
749 #define KEY_NEXT_ID "next_id"
750 #define KEY_READ_ONLY "read-only"
751 #define KEY_REGISTER_TARGET "register_target"
752 #define KEY_SET_FS "set_fs"
753 #define KEY_TGT_COUNT "tgt_count"
754 /* KEY_SET_INFO in lustre_idl.h */
755 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
756 #define KEY_CONNECT_FLAG "connect_flags"
758 #define KEY_CACHE_SET "cache_set"
759 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
760 #define KEY_OSP_CONNECTED "osp_connected"
764 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
765 #define IT_OPEN (1 << 0)
766 #define IT_CREAT (1 << 1)
767 #define IT_READDIR (1 << 2)
768 #define IT_GETATTR (1 << 3)
769 #define IT_LOOKUP (1 << 4)
770 #define IT_UNLINK (1 << 5)
771 #define IT_TRUNC (1 << 6)
772 #define IT_GETXATTR (1 << 7)
773 #define IT_EXEC (1 << 8)
774 #define IT_PIN (1 << 9)
775 #define IT_LAYOUT (1 << 10)
776 #define IT_QUOTA_DQACQ (1 << 11)
777 #define IT_QUOTA_CONN (1 << 12)
778 #define IT_SETXATTR (1 << 13)
780 static inline int it_to_lock_mode(struct lookup_intent *it)
782 /* CREAT needs to be tested before open (both could be set) */
783 if (it->it_op & IT_CREAT)
785 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
788 else if (it->it_op & IT_READDIR)
790 else if (it->it_op & IT_GETXATTR)
792 else if (it->it_op & IT_SETXATTR)
795 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
800 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
801 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
802 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
803 struct lu_fid op_fid4; /* to the operation locks. */
804 u32 op_mds; /* what mds server open will go to */
805 struct lustre_handle op_handle;
806 obd_time op_mod_time;
810 struct lmv_stripe_md *op_mea1;
811 struct lmv_stripe_md *op_mea2;
812 __u32 op_suppgids[2];
819 /* iattr fields and blocks. */
820 struct iattr op_attr;
822 loff_t op_attr_blocks;
824 /* Size-on-MDS epoch and flags. */
829 struct obd_capa *op_capa1;
830 struct obd_capa *op_capa2;
832 /* Various operation flags. */
833 enum mds_op_bias op_bias;
835 /* Used by readdir */
838 /* used to transfer info between the stacks of MD client
839 * see enum op_cli_flags */
842 /* File object data version for HSM release, on client */
843 __u64 op_data_version;
844 struct lustre_handle op_lease_handle;
847 #define op_stripe_offset op_ioepoch
848 #define op_max_pages op_valid
851 int (*md_blocking_ast)(struct ldlm_lock *lock,
852 struct ldlm_lock_desc *desc,
853 void *data, int flag);
856 struct md_enqueue_info;
857 /* metadata stat-ahead */
858 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
859 struct md_enqueue_info *minfo,
862 struct md_enqueue_info {
863 struct md_op_data mi_data;
864 struct lookup_intent mi_it;
865 struct lustre_handle mi_lockh;
866 struct inode *mi_dir;
867 md_enqueue_cb_t mi_cb;
869 unsigned int mi_generation;
873 struct module *o_owner;
874 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
875 void *karg, void *uarg);
876 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
877 __u32 keylen, void *key, __u32 *vallen, void *val,
878 struct lov_stripe_md *lsm);
879 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
880 __u32 keylen, void *key,
881 __u32 vallen, void *val,
882 struct ptlrpc_request_set *set);
883 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
884 int (*o_precleanup)(struct obd_device *dev,
885 enum obd_cleanup_stage cleanup_stage);
886 int (*o_cleanup)(struct obd_device *dev);
887 int (*o_process_config)(struct obd_device *dev, obd_count len,
889 int (*o_postrecov)(struct obd_device *dev);
890 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
892 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
893 /* connect to the target device with given connection
894 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
895 * granted by the target, which are guaranteed to be a subset of flags
896 * asked for. If @ocd == NULL, use default parameters. */
897 int (*o_connect)(const struct lu_env *env,
898 struct obd_export **exp, struct obd_device *src,
899 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
901 int (*o_reconnect)(const struct lu_env *env,
902 struct obd_export *exp, struct obd_device *src,
903 struct obd_uuid *cluuid,
904 struct obd_connect_data *ocd,
906 int (*o_disconnect)(struct obd_export *exp);
908 /* Initialize/finalize fids infrastructure. */
909 int (*o_fid_init)(struct obd_device *obd,
910 struct obd_export *exp, enum lu_cli_type type);
911 int (*o_fid_fini)(struct obd_device *obd);
913 /* Allocate new fid according to passed @hint. */
914 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
915 struct lu_fid *fid, struct md_op_data *op_data);
918 * Object with @fid is getting deleted, we may want to do something
921 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
922 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
923 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
924 __u64 max_age, struct ptlrpc_request_set *set);
925 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
926 struct lov_stripe_md *mem_src);
927 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
928 struct lov_mds_md *disk_src, int disk_len);
929 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
930 struct obdo *oa, struct lov_stripe_md **ea,
931 struct obd_trans_info *oti);
932 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
933 struct obdo *oa, struct lov_stripe_md *ea,
934 struct obd_trans_info *oti, struct obd_export *md_exp,
936 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
937 struct obd_info *oinfo, struct obd_trans_info *oti);
938 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
939 struct obd_trans_info *oti,
940 struct ptlrpc_request_set *rqset);
941 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
942 struct obd_info *oinfo);
943 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
944 struct ptlrpc_request_set *set);
945 int (*o_preprw)(const struct lu_env *env, int cmd,
946 struct obd_export *exp, struct obdo *oa, int objcount,
947 struct obd_ioobj *obj, struct niobuf_remote *remote,
948 int *nr_pages, struct niobuf_local *local,
949 struct obd_trans_info *oti, struct lustre_capa *capa);
950 int (*o_commitrw)(const struct lu_env *env, int cmd,
951 struct obd_export *exp, struct obdo *oa,
952 int objcount, struct obd_ioobj *obj,
953 struct niobuf_remote *remote, int pages,
954 struct niobuf_local *local,
955 struct obd_trans_info *oti, int rc);
956 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
957 ldlm_iterator_t it, void *data);
958 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
959 ldlm_iterator_t it, void *data);
960 int (*o_init_export)(struct obd_export *exp);
961 int (*o_destroy_export)(struct obd_export *exp);
963 /* llog related obd_methods */
964 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
965 struct obd_device *disk_obd, int *idx);
966 int (*o_llog_finish)(struct obd_device *obd, int count);
968 int (*o_import_event)(struct obd_device *, struct obd_import *,
969 enum obd_import_event);
971 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
972 enum obd_notify_event ev, void *data);
974 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
975 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
978 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
979 struct obd_quotactl *);
980 int (*o_quotactl)(struct obd_device *, struct obd_export *,
981 struct obd_quotactl *);
983 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
986 int (*o_pool_new)(struct obd_device *obd, char *poolname);
987 int (*o_pool_del)(struct obd_device *obd, char *poolname);
988 int (*o_pool_add)(struct obd_device *obd, char *poolname,
990 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
992 void (*o_getref)(struct obd_device *obd);
993 void (*o_putref)(struct obd_device *obd);
995 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
996 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
997 * Also, add a wrapper function in include/linux/obd_class.h. */
1000 /* lmv structures */
1002 struct mdt_body *body;
1003 struct lov_stripe_md *lsm;
1004 struct lmv_stripe_md *lmv;
1005 #ifdef CONFIG_FS_POSIX_ACL
1006 struct posix_acl *posix_acl;
1008 struct mdt_remote_perm *remote_perm;
1009 struct obd_capa *mds_capa;
1010 struct obd_capa *oss_capa;
1013 struct md_open_data {
1014 struct obd_client_handle *mod_och;
1015 struct ptlrpc_request *mod_open_req;
1016 struct ptlrpc_request *mod_close_req;
1017 atomic_t mod_refcount;
1021 struct lookup_intent;
1025 /* Every operation from MD_STATS_FIRST_OP up to and including
1026 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1027 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1028 * Operations after MD_STATS_LAST_OP are excluded from stats.
1029 * There are a few reasons for doing this: we prune the 17
1030 * counters which will be of minimal use in understanding
1031 * metadata utilization, we save memory by allocating 15
1032 * instead of 32 counters, we save cycles by not counting.
1034 * MD_STATS_FIRST_OP must be the first member of md_ops.
1036 #define MD_STATS_FIRST_OP m_close
1037 int (*m_close)(struct obd_export *, struct md_op_data *,
1038 struct md_open_data *, struct ptlrpc_request **);
1040 int (*m_create)(struct obd_export *, struct md_op_data *,
1041 const void *, int, int, __u32, __u32, cfs_cap_t,
1042 __u64, struct ptlrpc_request **);
1044 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1045 const union ldlm_policy_data *,
1046 struct lookup_intent *, struct md_op_data *,
1047 struct lustre_handle *, __u64);
1049 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1050 struct ptlrpc_request **);
1052 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1053 struct lookup_intent *,
1054 struct ptlrpc_request **,
1055 ldlm_blocking_callback, __u64);
1057 int (*m_link)(struct obd_export *, struct md_op_data *,
1058 struct ptlrpc_request **);
1060 int (*m_rename)(struct obd_export *, struct md_op_data *,
1061 const char *, int, const char *, int,
1062 struct ptlrpc_request **);
1064 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1065 int , void *, int, struct ptlrpc_request **,
1066 struct md_open_data **mod);
1068 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1069 struct obd_capa *, struct ptlrpc_request **);
1071 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1072 struct md_callback *cb_op, __u64 hash_offset,
1073 struct page **ppage);
1075 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1076 struct ptlrpc_request **);
1078 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1079 struct obd_capa *, obd_valid, const char *,
1080 const char *, int, int, int, __u32,
1081 struct ptlrpc_request **);
1083 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1084 struct obd_capa *, obd_valid, const char *,
1085 const char *, int, int, int,
1086 struct ptlrpc_request **);
1088 int (*m_intent_getattr_async)(struct obd_export *,
1089 struct md_enqueue_info *,
1090 struct ldlm_enqueue_info *);
1092 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1093 struct lu_fid *, __u64 *bits);
1095 #define MD_STATS_LAST_OP m_revalidate_lock
1097 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1098 struct obd_capa **);
1100 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1102 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1103 ldlm_iterator_t, void *);
1105 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1106 struct md_open_data *);
1108 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1109 struct ptlrpc_request **);
1111 int (*m_init_ea_size)(struct obd_export *, int, int, int, int);
1113 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1114 struct obd_export *, struct obd_export *,
1115 struct lustre_md *);
1117 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1119 int (*m_merge_attr)(struct obd_export *,
1120 const struct lmv_stripe_md *lsm,
1121 struct cl_attr *attr);
1123 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1124 struct mdt_body *, ldlm_blocking_callback);
1126 int (*m_set_open_replay_data)(struct obd_export *,
1127 struct obd_client_handle *,
1128 struct lookup_intent *);
1130 int (*m_clear_open_replay_data)(struct obd_export *,
1131 struct obd_client_handle *);
1133 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1135 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1136 const struct lu_fid *, ldlm_type_t,
1137 ldlm_policy_data_t *, ldlm_mode_t,
1138 struct lustre_handle *);
1140 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1141 ldlm_policy_data_t *, ldlm_mode_t,
1142 ldlm_cancel_flags_t flags, void *opaque);
1144 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1145 renew_capa_cb_t cb);
1147 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1148 const struct req_msg_field *, struct obd_capa **);
1150 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1151 struct obd_capa *, __u32,
1152 struct ptlrpc_request **);
1154 int (*m_get_fid_from_lsm)(struct obd_export *,
1155 const struct lmv_stripe_md *,
1156 const char *name, int namelen,
1157 struct lu_fid *fid);
1160 struct lsm_operations {
1161 void (*lsm_free)(struct lov_stripe_md *);
1162 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1163 struct obd_export *md_exp);
1164 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1166 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1168 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1169 __u16 *stripe_count);
1170 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1171 struct lov_mds_md *lmm);
1174 extern const struct lsm_operations lsm_v1_ops;
1175 extern const struct lsm_operations lsm_v3_ops;
1176 static inline const struct lsm_operations *lsm_op_find(int magic)
1184 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1189 static inline struct md_open_data *obd_mod_alloc(void)
1191 struct md_open_data *mod;
1195 atomic_set(&mod->mod_refcount, 1);
1199 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1200 #define obd_mod_put(mod) \
1202 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1203 if ((mod)->mod_open_req) \
1204 ptlrpc_req_finished((mod)->mod_open_req); \
1205 OBD_FREE_PTR(mod); \
1209 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1210 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1212 /* return 1 if client should be resend request */
1213 static inline int client_should_resend(int resend, struct client_obd *cli)
1215 return atomic_read(&cli->cl_resends) ?
1216 atomic_read(&cli->cl_resends) > resend : 1;
1220 * Return device name for this device
1222 * XXX: lu_device is declared before obd_device, while a pointer pointing
1223 * back to obd_device in lu_device, so this helper function defines here
1224 * instead of in lu_object.h
1226 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1228 return lu_dev->ld_obd->obd_name;
1231 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1236 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1239 /* caller does not care of idx */
1243 /* volatile file, the MDT can be set from name */
1244 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1245 /* if no MDT is specified, use std way */
1246 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1248 /* test for no MDT idx case */
1249 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1250 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1254 /* we have an idx, read it */
1255 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1256 *idx = strtoul(start, &end, 16);
1258 * no digit, no trailing :, negative value
1260 if (((*idx == 0) && (end == start)) ||
1261 (*end != ':') || (*idx < 0))
1266 /* bad format of mdt idx, we cannot return an error
1267 * to caller so we use hash algo */
1268 CERROR("Bad volatile file name format: %s\n",
1269 name + LUSTRE_VOLATILE_HDR_LEN);
1273 static inline int cli_brw_size(struct obd_device *obd)
1275 LASSERT(obd != NULL);
1276 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1279 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1280 * of the client should be increased accordingly to avoid sending fragmented
1281 * RPCs over the network when the client runs out of the maximum dirty space
1282 * when so many RPCs are being generated.
1284 static inline void client_adjust_max_dirty(struct client_obd *cli)
1287 if (cli->cl_dirty_max_pages <= 0)
1288 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1289 >> PAGE_CACHE_SHIFT;
1291 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1292 cli->cl_max_pages_per_rpc;
1294 if (dirty_max > cli->cl_dirty_max_pages)
1295 cli->cl_dirty_max_pages = dirty_max;
1298 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1299 cli->cl_dirty_max_pages = totalram_pages / 8;
1302 #endif /* __OBD_H */