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 received
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
209 * Limit reply buffer size for striping data to one x86_64 page. This
210 * value is chosen to fit the striping data for common use cases while
211 * staying well below the limit at which the buffer must be backed by
212 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
215 #define OBD_MAX_DEFAULT_EA_SIZE 4096
216 #define OBD_MAX_DEFAULT_COOKIE_SIZE 4096
221 struct rw_semaphore cl_sem;
222 struct obd_uuid cl_target_uuid;
223 struct obd_import *cl_import; /* ptlrpc connection state */
224 size_t cl_conn_count;
226 /* Cache maximum and default values for easize and cookiesize. This is
227 * strictly a performance optimization to minimize calls to
228 * obd_size_diskmd(). The default values are used to calculate the
229 * initial size of a request buffer. The ptlrpc layer will resize the
230 * buffer as needed to accommodate a larger reply from the
231 * server. The default values should be small enough to avoid wasted
232 * memory and excessive use of vmalloc(), yet large enough to avoid
233 * reallocating the buffer in the common use case. */
235 /* Default EA size for striping attributes. It is initialized at
236 * mount-time based on the default stripe width of the filesystem,
237 * then it tracks the largest observed EA size advertised by
238 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
239 __u32 cl_default_mds_easize;
241 /* Maximum possible EA size computed at mount-time based on
242 * the number of OSTs in the filesystem. May be increased at
243 * run-time if a larger observed size is advertised by the MDT. */
244 __u32 cl_max_mds_easize;
246 /* Default cookie size for llog cookies (see struct llog_cookie). It is
247 * initialized to zero at mount-time, then it tracks the largest
248 * observed cookie size advertised by the MDT, up to a maximum value of
249 * OBD_MAX_DEFAULT_COOKIE_SIZE. Note that llog_cookies are not
250 * used by clients communicating with MDS versions 2.4.0 and later.*/
251 __u32 cl_default_mds_cookiesize;
253 /* Maximum possible cookie size computed at mount-time based on
254 * the number of OSTs in the filesystem. May be increased at
255 * run-time if a larger observed size is advertised by the MDT. */
256 __u32 cl_max_mds_cookiesize;
258 enum lustre_sec_part cl_sp_me;
259 enum lustre_sec_part cl_sp_to;
260 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
262 /* the grant values are protected by loi_list_lock below */
263 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
264 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
265 unsigned long cl_dirty_transit; /* dirty synchronous */
266 unsigned long cl_avail_grant; /* bytes of credit for ost */
267 unsigned long cl_lost_grant; /* lost credits (trunc) */
269 /* since we allocate grant by blocks, we don't know how many grant will
270 * be used to add a page into cache. As a solution, we reserve maximum
271 * grant before trying to dirty a page and unreserve the rest.
272 * See osc_{reserve|unreserve}_grant for details. */
273 long cl_reserved_grant;
274 struct list_head cl_cache_waiters; /* waiting for cache/grant */
275 cfs_time_t cl_next_shrink_grant; /* jiffies */
276 struct list_head cl_grant_shrink_list; /* Timeout event list */
277 int cl_grant_shrink_interval; /* seconds */
279 /* A chunk is an optimal size used by osc_extent to determine
280 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
282 unsigned int cl_extent_tax; /* extent overhead, by bytes */
284 /* keep track of objects that have lois that contain pages which
285 * have been queued for async brw. this lock also protects the
286 * lists of osc_client_pages that hang off of the loi */
288 * ->cl_loi_list_lock protects consistency of
289 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
290 * ->ap_completion() call-backs are executed under this lock. As we
291 * cannot guarantee that these call-backs never block on all platforms
292 * (as a matter of fact they do block on Mac OS X), type of
293 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
294 * and blocking mutex on Mac OS X. (Alternative is to make this lock
295 * blocking everywhere, but we don't want to slow down fast-path of
296 * our main platform.)
298 * NB by Jinshan: though field names are still _loi_, but actually
299 * osc_object{}s are in the list.
301 spinlock_t cl_loi_list_lock;
302 struct list_head cl_loi_ready_list;
303 struct list_head cl_loi_hp_ready_list;
304 struct list_head cl_loi_write_list;
305 struct list_head cl_loi_read_list;
306 __u32 cl_r_in_flight;
307 __u32 cl_w_in_flight;
308 /* just a sum of the loi/lop pending numbers to be exported by /proc */
309 atomic_t cl_pending_w_pages;
310 atomic_t cl_pending_r_pages;
311 __u32 cl_max_pages_per_rpc;
312 __u32 cl_max_rpcs_in_flight;
313 struct obd_histogram cl_read_rpc_hist;
314 struct obd_histogram cl_write_rpc_hist;
315 struct obd_histogram cl_read_page_hist;
316 struct obd_histogram cl_write_page_hist;
317 struct obd_histogram cl_read_offset_hist;
318 struct obd_histogram cl_write_offset_hist;
320 /* lru for osc caching pages */
321 struct cl_client_cache *cl_cache;
322 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
323 atomic_long_t *cl_lru_left;
324 atomic_long_t cl_lru_busy;
325 atomic_long_t cl_lru_in_list;
326 atomic_long_t cl_unstable_count;
327 struct list_head cl_lru_list; /* lru page list */
328 spinlock_t cl_lru_list_lock; /* page list protector */
329 atomic_t cl_lru_shrinkers;
331 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
332 atomic_t cl_destroy_in_flight;
333 wait_queue_head_t cl_destroy_waitq;
335 struct mdc_rpc_lock *cl_rpc_lock;
336 struct mdc_rpc_lock *cl_close_lock;
339 struct mutex cl_mgc_mutex;
340 struct local_oid_storage *cl_mgc_los;
341 struct dt_object *cl_mgc_configs_dir;
342 atomic_t cl_mgc_refcount;
343 struct obd_export *cl_mgc_mgsexp;
345 /* checksumming for data sent over the network */
346 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
347 /* supported checksum types that are worked out at connect time */
348 __u32 cl_supp_cksum_types;
349 /* checksum algorithm to be used */
350 cksum_type_t cl_cksum_type;
352 /* also protected by the poorly named _loi_list_lock lock above */
353 struct osc_async_rc cl_ar;
355 /* used by quotacheck when the servers are older than 2.4 */
356 int cl_qchk_stat; /* quotacheck stat of the peer */
357 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
358 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
359 #warning "please consider removing quotacheck compatibility code"
362 /* sequence manager */
363 struct lu_client_seq *cl_seq;
365 atomic_t cl_resends; /* resend count */
367 /* ptlrpc work for writeback in ptlrpcd context */
368 void *cl_writeback_work;
370 /* hash tables for osc_quota_info */
371 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
373 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
380 struct echo_client_obd {
381 struct obd_export *ec_exp; /* the local connection to osc/lov */
383 struct list_head ec_objects;
384 struct list_head ec_locks;
388 /* Generic subset of OSTs */
390 __u32 *op_array; /* array of index of
392 unsigned int op_count; /* number of OSTs in the array */
393 unsigned int op_size; /* allocated size of lp_array */
394 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
397 /* allow statfs data caching for 1 second */
398 #define OBD_STATFS_CACHE_SECONDS 1
400 struct lov_tgt_desc {
401 struct list_head ltd_kill;
402 struct obd_uuid ltd_uuid;
403 struct obd_device *ltd_obd;
404 struct obd_export *ltd_exp;
406 __u32 ltd_index; /* index in lov_obd->tgts */
407 unsigned long ltd_active:1,/* is this target up for requests */
408 ltd_activate:1,/* should target be activated */
409 ltd_reap:1; /* should this target be deleted */
413 struct lov_desc desc;
414 struct lov_tgt_desc **lov_tgts; /* sparse array */
415 struct ost_pool lov_packed; /* all OSTs in a packed
417 struct mutex lov_lock;
418 struct obd_connect_data lov_ocd;
419 struct proc_dir_entry *targets_proc_entry;
420 atomic_t lov_refcount;
421 __u32 lov_death_row; /* tgts scheduled to be deleted */
422 __u32 lov_tgt_size; /* size of tgts array */
425 cfs_hash_t *lov_pools_hash_body; /* used for key access */
426 struct list_head lov_pool_list; /* used for sequential access */
427 struct proc_dir_entry *lov_pool_proc_entry;
428 enum lustre_sec_part lov_sp_me;
430 /* Cached LRU and unstable data from upper layer */
431 struct cl_client_cache *lov_cache;
433 struct rw_semaphore lov_notify_lock;
436 struct lmv_tgt_desc {
437 struct obd_uuid ltd_uuid;
438 struct obd_export *ltd_exp;
440 struct mutex ltd_fid_mutex;
441 unsigned long ltd_active:1; /* target up for requests */
444 enum placement_policy {
445 PLACEMENT_CHAR_POLICY = 0,
446 PLACEMENT_NID_POLICY = 1,
447 PLACEMENT_INVAL_POLICY = 2,
451 typedef enum placement_policy placement_policy_t;
455 struct lu_client_fld lmv_fld;
457 placement_policy_t lmv_placement;
458 struct lmv_desc desc;
459 struct obd_uuid cluuid;
460 struct obd_export *exp;
461 struct proc_dir_entry *targets_proc_entry;
463 struct mutex lmv_init_mutex;
468 int max_def_cookiesize;
470 __u32 tgts_size; /* size of tgts array */
471 struct lmv_tgt_desc **tgts;
473 struct obd_connect_data conn_data;
476 struct niobuf_local {
477 __u64 lnb_file_offset;
478 __u32 lnb_page_offset;
481 struct page *lnb_page;
486 #define LUSTRE_FLD_NAME "fld"
487 #define LUSTRE_SEQ_NAME "seq"
489 #define LUSTRE_MDD_NAME "mdd"
490 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
491 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
492 #define LUSTRE_VVP_NAME "vvp"
493 #define LUSTRE_LMV_NAME "lmv"
494 #define LUSTRE_SLP_NAME "slp"
495 #define LUSTRE_LOD_NAME "lod"
496 #define LUSTRE_OSP_NAME "osp"
497 #define LUSTRE_LWP_NAME "lwp"
499 /* obd device type names */
500 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
501 #define LUSTRE_MDS_NAME "mds"
502 #define LUSTRE_MDT_NAME "mdt"
503 #define LUSTRE_MDC_NAME "mdc"
504 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
505 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
506 #define LUSTRE_OSC_NAME "osc"
507 #define LUSTRE_LOV_NAME "lov"
508 #define LUSTRE_MGS_NAME "mgs"
509 #define LUSTRE_MGC_NAME "mgc"
511 #define LUSTRE_ECHO_NAME "obdecho"
512 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
513 #define LUSTRE_QMT_NAME "qmt"
515 /* Constant obd names (post-rename) */
516 #define LUSTRE_MDS_OBDNAME "MDS"
517 #define LUSTRE_OSS_OBDNAME "OSS"
518 #define LUSTRE_MGS_OBDNAME "MGS"
519 #define LUSTRE_MGC_OBDNAME "MGC"
521 struct obd_trans_info {
523 /* Only used on the server side for tracking acks. */
524 struct oti_req_ack_lock {
525 struct lustre_handle lock;
529 struct llog_cookie oti_onecookie;
530 struct llog_cookie *oti_logcookies;
533 __u64 oti_pre_version;
537 * Events signalled through obd_notify() upcall-chain.
539 enum obd_notify_event {
542 /* Device connect start */
544 /* Device activated */
546 /* Device deactivated */
548 /* Device disconnected */
550 /* Connect data for import were changed */
553 OBD_NOTIFY_SYNC_NONBLOCK,
555 /* Configuration event */
557 /* Administratively deactivate/activate event */
558 OBD_NOTIFY_DEACTIVATE,
562 /* bit-mask flags for config events */
564 CONFIG_LOG = 0x1, /* finished processing config log */
565 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
566 CONFIG_TARGET = 0x4 /* one target is added */
570 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
571 * and liblustre being main examples).
573 struct obd_notify_upcall {
574 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
575 enum obd_notify_event ev, void *owner, void *data);
576 /* Opaque datum supplied by upper layer listener */
580 struct target_recovery_data {
581 svc_handler_t trd_recovery_handler;
582 pid_t trd_processing_task;
583 struct completion trd_starting;
584 struct completion trd_finishing;
587 struct obd_llog_group {
588 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
589 wait_queue_head_t olg_waitq;
591 struct mutex olg_cat_processing;
594 /* corresponds to one of the obd's */
595 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
598 struct obd_type *obd_type;
601 /* common and UUID name of this device */
602 char obd_name[MAX_OBD_NAME];
603 struct obd_uuid obd_uuid;
605 struct lu_device *obd_lu_dev;
607 /* bitfield modification is protected by obd_dev_lock */
609 obd_attached:1, /* finished attach */
610 obd_set_up:1, /* finished setup */
611 obd_recovering:1, /* there are recoverable clients */
612 obd_abort_recovery:1, /* recovery expired */
613 obd_version_recov:1, /* obd uses version checking */
614 obd_replayable:1, /* recovery is enabled;
616 obd_no_transno:1, /* no committed-transno notification */
617 obd_no_recov:1, /* fail instead of retry messages */
618 obd_stopping:1, /* started cleanup */
619 obd_starting:1, /* started setup */
620 obd_force:1, /* cleanup with > 0 obd refcount */
621 obd_fail:1, /* cleanup with failover */
622 obd_no_conn:1, /* deny new connections */
623 obd_inactive:1, /* device active/inactive
624 * (for /proc/status only!!) */
625 obd_no_ir:1, /* no imperative recovery. */
626 obd_process_conf:1, /* device is processing mgs config */
627 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
629 /* use separate field as it is set in interrupt to don't mess with
630 * protection of other bits using _bh lock */
631 unsigned long obd_recovery_expired:1;
632 /* uuid-export hash body */
633 cfs_hash_t *obd_uuid_hash;
634 /* nid-export hash body */
635 cfs_hash_t *obd_nid_hash;
637 cfs_hash_t *obd_nid_stats_hash;
638 struct list_head obd_nid_stats;
639 atomic_t obd_refcount;
640 struct list_head obd_exports;
641 struct list_head obd_unlinked_exports;
642 struct list_head obd_delayed_exports;
643 struct list_head obd_lwp_list;
645 spinlock_t obd_nid_lock;
646 struct ldlm_namespace *obd_namespace;
647 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
648 /* a spinlock is OK for what we do now, may need a semaphore later */
649 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
650 struct mutex obd_dev_mutex;
651 __u64 obd_last_committed;
652 spinlock_t obd_osfs_lock;
653 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
655 struct lvfs_run_ctxt obd_lvfs_ctxt;
656 struct obd_llog_group obd_olg; /* default llog group */
657 struct obd_device *obd_observer;
658 struct rw_semaphore obd_observer_link_sem;
659 struct obd_notify_upcall obd_upcall;
660 struct obd_export *obd_self_export;
661 struct obd_export *obd_lwp_export;
662 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
663 struct list_head obd_exports_timed;
664 time_t obd_eviction_timer; /* for ping evictor */
666 int obd_max_recoverable_clients;
667 atomic_t obd_connected_clients;
668 int obd_stale_clients;
669 /* this lock protects all recovery list_heads, timer and
670 * obd_next_recovery_transno value */
671 spinlock_t obd_recovery_task_lock;
672 __u64 obd_next_recovery_transno;
673 int obd_replayed_requests;
674 int obd_requests_queued_for_recovery;
675 wait_queue_head_t obd_next_transno_waitq;
676 /* protected by obd_recovery_task_lock */
677 struct timer_list obd_recovery_timer;
679 time_t obd_recovery_start;
680 /* seconds, for lprocfs_status */
681 time_t obd_recovery_end;
682 int obd_recovery_time_hard;
683 int obd_recovery_timeout;
684 int obd_recovery_ir_factor;
686 /* new recovery stuff from CMD2 */
687 struct target_recovery_data obd_recovery_data;
688 int obd_replayed_locks;
689 atomic_t obd_req_replay_clients;
690 atomic_t obd_lock_replay_clients;
691 /* all lists are protected by obd_recovery_task_lock */
692 struct list_head obd_req_replay_queue;
693 struct list_head obd_lock_replay_queue;
694 struct list_head obd_final_req_queue;
697 #ifdef HAVE_SERVER_SUPPORT
698 struct obd_device_target obt;
699 struct filter_obd filter;
701 struct echo_obd echo;
703 struct client_obd cli;
704 struct echo_client_obd echo_client;
708 /* Fields used by LProcFS */
709 unsigned int obd_cntr_base;
710 struct lprocfs_stats *obd_stats;
712 unsigned int obd_md_cntr_base;
713 struct lprocfs_stats *obd_md_stats;
715 struct proc_dir_entry *obd_proc_entry;
716 struct proc_dir_entry *obd_proc_exports_entry;
717 struct proc_dir_entry *obd_svc_procroot;
718 struct lprocfs_stats *obd_svc_stats;
719 struct lprocfs_vars *obd_vars;
720 atomic_t obd_evict_inprogress;
721 wait_queue_head_t obd_evict_inprogress_waitq;
722 struct list_head obd_evict_list; /* protected with pet_lock */
725 * Ldlm pool part. Save last calculated SLV and Limit.
727 rwlock_t obd_pool_lock;
732 * A list of outstanding class_incref()'s against this obd. For
735 struct lu_ref obd_reference;
737 int obd_conn_inprogress;
740 enum obd_cleanup_stage {
741 /* Special case hack for MDS LOVs */
743 /* can be directly mapped to .ldto_device_fini() */
747 /* get/set_info keys */
748 #define KEY_ASYNC "async"
749 #define KEY_CAPA_KEY "capa_key"
750 #define KEY_CHANGELOG_CLEAR "changelog_clear"
751 #define KEY_FID2PATH "fid2path"
752 #define KEY_CHECKSUM "checksum"
753 #define KEY_CLEAR_FS "clear_fs"
754 #define KEY_CONN_DATA "conn_data"
755 #define KEY_EVICT_BY_NID "evict_by_nid"
756 #define KEY_FIEMAP "fiemap"
757 #define KEY_FLUSH_CTX "flush_ctx"
758 #define KEY_GRANT_SHRINK "grant_shrink"
759 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
760 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
761 #define KEY_INTERMDS "inter_mds"
762 #define KEY_LAST_ID "last_id"
763 #define KEY_LAST_FID "last_fid"
764 #define KEY_LOVDESC "lovdesc"
765 #define KEY_MAX_EASIZE "max_easize"
766 #define KEY_DEFAULT_EASIZE "default_easize"
767 #define KEY_MAX_COOKIESIZE "max_cookiesize"
768 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
769 #define KEY_MGSSEC "mgssec"
770 #define KEY_READ_ONLY "read-only"
771 #define KEY_REGISTER_TARGET "register_target"
772 #define KEY_SET_FS "set_fs"
773 #define KEY_TGT_COUNT "tgt_count"
774 /* KEY_SET_INFO in lustre_idl.h */
775 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
777 #define KEY_CACHE_SET "cache_set"
778 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
779 #define KEY_OSP_CONNECTED "osp_connected"
783 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
784 #define IT_OPEN (1 << 0)
785 #define IT_CREAT (1 << 1)
786 #define IT_READDIR (1 << 2)
787 #define IT_GETATTR (1 << 3)
788 #define IT_LOOKUP (1 << 4)
789 #define IT_UNLINK (1 << 5)
790 #define IT_TRUNC (1 << 6)
791 #define IT_GETXATTR (1 << 7)
792 #define IT_EXEC (1 << 8)
793 #define IT_PIN (1 << 9)
794 #define IT_LAYOUT (1 << 10)
795 #define IT_QUOTA_DQACQ (1 << 11)
796 #define IT_QUOTA_CONN (1 << 12)
797 #define IT_SETXATTR (1 << 13)
799 static inline int it_to_lock_mode(struct lookup_intent *it)
801 /* CREAT needs to be tested before open (both could be set) */
802 if (it->it_op & IT_CREAT)
804 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
807 else if (it->it_op & IT_READDIR)
809 else if (it->it_op & IT_GETXATTR)
811 else if (it->it_op & IT_SETXATTR)
814 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
819 MF_MDC_CANCEL_FID1 = 1 << 0,
820 MF_MDC_CANCEL_FID2 = 1 << 1,
821 MF_MDC_CANCEL_FID3 = 1 << 2,
822 MF_MDC_CANCEL_FID4 = 1 << 3,
823 MF_GET_MDT_IDX = 1 << 4,
827 CLI_SET_MEA = 1 << 0,
828 CLI_RM_ENTRY = 1 << 1,
831 CLI_MIGRATE = 1 << 4,
835 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
836 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
837 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
838 struct lu_fid op_fid4; /* to the operation locks. */
839 u32 op_mds; /* what mds server open will go to */
840 struct lustre_handle op_handle;
845 struct lmv_stripe_md *op_mea1;
846 struct lmv_stripe_md *op_mea2;
847 __u32 op_suppgids[2];
854 /* iattr fields and blocks. */
855 struct iattr op_attr;
856 loff_t op_attr_blocks;
857 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
858 __u64 op_valid; /* OBD_MD_* */
860 enum md_op_flags op_flags;
863 struct obd_capa *op_capa1;
864 struct obd_capa *op_capa2;
866 /* Various operation flags. */
867 enum mds_op_bias op_bias;
869 /* Used by readdir */
870 unsigned int op_max_pages;
872 /* used to transfer info between the stacks of MD client
873 * see enum op_cli_flags */
874 enum md_cli_flags op_cli_flags;
876 /* File object data version for HSM release, on client */
877 __u64 op_data_version;
878 struct lustre_handle op_lease_handle;
880 /* default stripe offset */
881 __u32 op_default_stripe_offset;
885 int (*md_blocking_ast)(struct ldlm_lock *lock,
886 struct ldlm_lock_desc *desc,
887 void *data, int flag);
890 struct md_enqueue_info;
891 /* metadata stat-ahead */
892 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
893 struct md_enqueue_info *minfo,
896 struct md_enqueue_info {
897 struct md_op_data mi_data;
898 struct lookup_intent mi_it;
899 struct lustre_handle mi_lockh;
900 struct inode *mi_dir;
901 md_enqueue_cb_t mi_cb;
906 struct module *o_owner;
907 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
908 void *karg, void __user *uarg);
909 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
910 __u32 keylen, void *key, __u32 *vallen, void *val);
911 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
912 __u32 keylen, void *key,
913 __u32 vallen, void *val,
914 struct ptlrpc_request_set *set);
915 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
916 int (*o_precleanup)(struct obd_device *dev,
917 enum obd_cleanup_stage cleanup_stage);
918 int (*o_cleanup)(struct obd_device *dev);
919 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
920 int (*o_postrecov)(struct obd_device *dev);
921 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
923 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
924 /* connect to the target device with given connection
925 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
926 * granted by the target, which are guaranteed to be a subset of flags
927 * asked for. If @ocd == NULL, use default parameters. */
928 int (*o_connect)(const struct lu_env *env,
929 struct obd_export **exp, struct obd_device *src,
930 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
932 int (*o_reconnect)(const struct lu_env *env,
933 struct obd_export *exp, struct obd_device *src,
934 struct obd_uuid *cluuid,
935 struct obd_connect_data *ocd,
937 int (*o_disconnect)(struct obd_export *exp);
939 /* Initialize/finalize fids infrastructure. */
940 int (*o_fid_init)(struct obd_device *obd,
941 struct obd_export *exp, enum lu_cli_type type);
942 int (*o_fid_fini)(struct obd_device *obd);
944 /* Allocate new fid according to passed @hint. */
945 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
946 struct lu_fid *fid, struct md_op_data *op_data);
949 * Object with @fid is getting deleted, we may want to do something
952 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
953 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
954 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
955 __u64 max_age, struct ptlrpc_request_set *set);
956 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
957 struct lov_stripe_md *mem_src);
958 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
959 struct lov_mds_md *disk_src, int disk_len);
960 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
961 struct obdo *oa, struct obd_trans_info *oti);
962 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
963 struct obdo *oa, struct obd_trans_info *oti);
964 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
965 struct obd_info *oinfo, struct obd_trans_info *oti);
966 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
967 struct obd_info *oinfo);
968 int (*o_preprw)(const struct lu_env *env, int cmd,
969 struct obd_export *exp, struct obdo *oa, int objcount,
970 struct obd_ioobj *obj, struct niobuf_remote *remote,
971 int *nr_pages, struct niobuf_local *local,
972 struct obd_trans_info *oti);
973 int (*o_commitrw)(const struct lu_env *env, int cmd,
974 struct obd_export *exp, struct obdo *oa,
975 int objcount, struct obd_ioobj *obj,
976 struct niobuf_remote *remote, int pages,
977 struct niobuf_local *local,
978 struct obd_trans_info *oti, int rc);
979 int (*o_init_export)(struct obd_export *exp);
980 int (*o_destroy_export)(struct obd_export *exp);
982 int (*o_import_event)(struct obd_device *, struct obd_import *,
983 enum obd_import_event);
985 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
986 enum obd_notify_event ev, void *data);
988 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
989 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
992 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
993 struct obd_quotactl *);
994 int (*o_quotactl)(struct obd_device *, struct obd_export *,
995 struct obd_quotactl *);
997 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1000 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1001 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1002 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1004 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1006 void (*o_getref)(struct obd_device *obd);
1007 void (*o_putref)(struct obd_device *obd);
1009 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1010 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1011 * Also, add a wrapper function in include/linux/obd_class.h. */
1014 /* lmv structures */
1016 struct mdt_body *body;
1017 struct lov_stripe_md *lsm;
1018 struct lmv_stripe_md *lmv;
1019 #ifdef CONFIG_FS_POSIX_ACL
1020 struct posix_acl *posix_acl;
1022 struct mdt_remote_perm *remote_perm;
1023 struct obd_capa *mds_capa;
1024 struct obd_capa *oss_capa;
1027 struct md_open_data {
1028 struct obd_client_handle *mod_och;
1029 struct ptlrpc_request *mod_open_req;
1030 struct ptlrpc_request *mod_close_req;
1031 atomic_t mod_refcount;
1035 struct obd_client_handle {
1036 struct lustre_handle och_fh;
1037 struct lu_fid och_fid;
1038 struct md_open_data *och_mod;
1039 struct lustre_handle och_lease_handle; /* open lock for lease */
1044 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1046 struct lookup_intent;
1050 /* Every operation from MD_STATS_FIRST_OP up to and including
1051 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1052 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1053 * Operations after MD_STATS_LAST_OP are excluded from stats.
1054 * There are a few reasons for doing this: we prune the 17
1055 * counters which will be of minimal use in understanding
1056 * metadata utilization, we save memory by allocating 15
1057 * instead of 32 counters, we save cycles by not counting.
1059 * MD_STATS_FIRST_OP must be the first member of md_ops.
1061 #define MD_STATS_FIRST_OP m_close
1062 int (*m_close)(struct obd_export *, struct md_op_data *,
1063 struct md_open_data *, struct ptlrpc_request **);
1065 int (*m_create)(struct obd_export *, struct md_op_data *,
1066 const void *, size_t, umode_t, uid_t, gid_t,
1067 cfs_cap_t, __u64, struct ptlrpc_request **);
1069 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1070 const union ldlm_policy_data *,
1071 struct lookup_intent *, struct md_op_data *,
1072 struct lustre_handle *, __u64);
1074 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1075 struct ptlrpc_request **);
1077 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1078 struct lookup_intent *,
1079 struct ptlrpc_request **,
1080 ldlm_blocking_callback, __u64);
1082 int (*m_link)(struct obd_export *, struct md_op_data *,
1083 struct ptlrpc_request **);
1085 int (*m_rename)(struct obd_export *, struct md_op_data *,
1086 const char *, size_t, const char *, size_t,
1087 struct ptlrpc_request **);
1089 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1090 size_t , struct ptlrpc_request **);
1092 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1093 struct obd_capa *, struct ptlrpc_request **);
1095 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1096 struct md_callback *cb_op, __u64 hash_offset,
1097 struct page **ppage);
1099 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1100 struct ptlrpc_request **);
1102 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1103 struct obd_capa *, u64, const char *,
1104 const char *, int, int, int, __u32,
1105 struct ptlrpc_request **);
1107 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1108 struct obd_capa *, u64, const char *,
1109 const char *, int, int, int,
1110 struct ptlrpc_request **);
1112 int (*m_intent_getattr_async)(struct obd_export *,
1113 struct md_enqueue_info *,
1114 struct ldlm_enqueue_info *);
1116 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1117 struct lu_fid *, __u64 *bits);
1119 #define MD_STATS_LAST_OP m_revalidate_lock
1121 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1122 struct obd_capa **);
1124 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1126 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1127 ldlm_iterator_t, void *);
1129 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1130 struct ptlrpc_request **);
1132 int (*m_init_ea_size)(struct obd_export *, __u32, __u32, __u32, __u32);
1134 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1135 struct obd_export *, struct obd_export *,
1136 struct lustre_md *);
1138 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1140 int (*m_merge_attr)(struct obd_export *,
1141 const struct lmv_stripe_md *lsm,
1142 struct cl_attr *attr, ldlm_blocking_callback);
1144 int (*m_set_open_replay_data)(struct obd_export *,
1145 struct obd_client_handle *,
1146 struct lookup_intent *);
1148 int (*m_clear_open_replay_data)(struct obd_export *,
1149 struct obd_client_handle *);
1151 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1153 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1154 const struct lu_fid *, ldlm_type_t,
1155 ldlm_policy_data_t *, ldlm_mode_t,
1156 struct lustre_handle *);
1158 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1159 ldlm_policy_data_t *, ldlm_mode_t,
1160 ldlm_cancel_flags_t flags, void *opaque);
1162 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1163 renew_capa_cb_t cb);
1165 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1166 const struct req_msg_field *, struct obd_capa **);
1168 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1169 struct obd_capa *, __u32,
1170 struct ptlrpc_request **);
1172 int (*m_get_fid_from_lsm)(struct obd_export *,
1173 const struct lmv_stripe_md *,
1174 const char *name, int namelen,
1175 struct lu_fid *fid);
1178 struct lsm_operations {
1179 void (*lsm_free)(struct lov_stripe_md *);
1180 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *,
1181 loff_t *, loff_t *);
1182 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *,
1183 loff_t *, loff_t *);
1184 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1185 __u16 *stripe_count);
1186 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1187 struct lov_mds_md *lmm);
1190 extern const struct lsm_operations lsm_v1_ops;
1191 extern const struct lsm_operations lsm_v3_ops;
1192 static inline const struct lsm_operations *lsm_op_find(u32 magic)
1200 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1205 static inline struct md_open_data *obd_mod_alloc(void)
1207 struct md_open_data *mod;
1211 atomic_set(&mod->mod_refcount, 1);
1215 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1216 #define obd_mod_put(mod) \
1218 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1219 if ((mod)->mod_open_req) \
1220 ptlrpc_req_finished((mod)->mod_open_req); \
1221 OBD_FREE_PTR(mod); \
1225 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1226 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1228 /* return 1 if client should be resend request */
1229 static inline int client_should_resend(int resend, struct client_obd *cli)
1231 return atomic_read(&cli->cl_resends) ?
1232 atomic_read(&cli->cl_resends) > resend : 1;
1236 * Return device name for this device
1238 * XXX: lu_device is declared before obd_device, while a pointer pointing
1239 * back to obd_device in lu_device, so this helper function defines here
1240 * instead of in lu_object.h
1242 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1244 return lu_dev->ld_obd->obd_name;
1247 static inline bool filename_is_volatile(const char *name, size_t namelen,
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 = simple_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 */