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/spinlock.h>
41 #include <linux/obd.h>
43 #include <lustre/lustre_idl.h>
44 #include <lustre_lib.h>
45 #include <libcfs/bitmap.h>
46 #ifdef HAVE_SERVER_SUPPORT
47 # include <lu_target.h>
48 # include <obd_target.h>
51 #include <lustre_export.h>
52 #include <lustre_fid.h>
53 #include <lustre_fld.h>
54 #include <lustre_handles.h>
55 #include <lustre_intent.h>
56 #include <lustre_capa.h>
59 #define MAX_OBD_DEVICES 8192
67 struct lov_oinfo { /* per-stripe data structure */
68 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
69 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
70 int loi_ost_gen; /* generation of this loi_ost_idx */
72 unsigned long loi_kms_valid:1;
73 __u64 loi_kms; /* known minimum size */
74 struct ost_lvb loi_lvb;
75 struct osc_async_rc loi_ar;
78 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
81 oinfo->loi_kms_valid = 1;
84 static inline void loi_init(struct lov_oinfo *loi)
88 /* If we are unable to get the maximum object size from the OST in
89 * ocd_maxbytes using OBD_CONNECT_MAXBYTES, then we fall back to using
90 * the old maximum object size from ext3. */
91 #define LUSTRE_EXT3_STRIPE_MAXBYTES 0x1fffffff000ULL
93 struct lov_stripe_md {
96 pid_t lsm_lock_owner; /* debugging */
98 /* maximum possible file size, might change as OSTs status changes,
99 * e.g. disconnected, deactivated */
102 /* Public members. */
103 struct ost_id lw_object_oi; /* lov object id/seq */
105 /* LOV-private members start here -- only for use in lov/. */
107 __u32 lw_stripe_size; /* size of the stripe */
108 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
109 __u16 lw_stripe_count; /* number of objects being striped over */
110 __u16 lw_layout_gen; /* generation of the layout */
111 char lw_pool_name[LOV_MAXPOOLNAME + 1]; /* pool name */
114 struct lov_oinfo *lsm_oinfo[0];
117 #define lsm_oi lsm_wire.lw_object_oi
118 #define lsm_magic lsm_wire.lw_magic
119 #define lsm_layout_gen lsm_wire.lw_layout_gen
120 #define lsm_stripe_size lsm_wire.lw_stripe_size
121 #define lsm_pattern lsm_wire.lw_pattern
122 #define lsm_stripe_count lsm_wire.lw_stripe_count
123 #define lsm_pool_name lsm_wire.lw_pool_name
125 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
127 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
130 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
134 if (lsm_is_released(lsm))
139 static inline int lov_stripe_md_size(unsigned int stripe_count)
141 struct lov_stripe_md lsm;
143 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
148 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
150 /* obd info for a particular level (lov, osc). */
152 /* Lock policy. It keeps an extent which is specific for a particular
153 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
154 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
155 ldlm_policy_data_t oi_policy;
156 /* Flags used for set request specific flags:
157 - while lock handling, the flags obtained on the enqueue
158 request are set here.
159 - while stats, the flags used for control delay/resend.
160 - while setattr, the flags used for distinguish punch operation
163 /* lsm data specific for every OSC. */
164 struct lov_stripe_md *oi_md;
165 /* obdo data specific for every OSC, if needed at all. */
167 /* statfs data specific for every OSC, if needed at all. */
168 struct obd_statfs *oi_osfs;
169 /* An update callback which is called to update some data on upper
170 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
171 * request in osc level for enqueue requests. It is also possible to
172 * update some caller data from LOV layer if needed. */
173 obd_enqueue_update_f oi_cb_up;
174 /* oss capability, its type is obd_capa in client to avoid copy.
175 * in contrary its type is lustre_capa in OSS. */
180 struct list_head typ_chain;
181 struct obd_ops *typ_dt_ops;
182 struct md_ops *typ_md_ops;
183 struct proc_dir_entry *typ_procroot;
184 struct proc_dir_entry *typ_procsym;
185 __u32 typ_sym_filter;
188 struct lu_device_type *typ_lu;
189 spinlock_t obd_type_lock;
199 struct timeout_item {
200 enum timeout_event ti_event;
201 cfs_time_t ti_timeout;
204 struct list_head ti_obd_list;
205 struct list_head ti_chain;
208 #define OBD_MAX_RIF_DEFAULT 8
209 #define OBD_MAX_RIF_MAX 512
210 #define OSC_MAX_RIF_MAX 256
211 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
212 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
213 #define OSC_DEFAULT_RESENDS 10
215 /* possible values for fo_sync_lock_cancel */
217 NEVER_SYNC_ON_CANCEL = 0,
218 BLOCKING_SYNC_ON_CANCEL = 1,
219 ALWAYS_SYNC_ON_CANCEL = 2,
220 NUM_SYNC_ON_CANCEL_STATES
226 struct rw_semaphore cl_sem;
227 struct obd_uuid cl_target_uuid;
228 struct obd_import *cl_import; /* ptlrpc connection state */
230 /* max_mds_easize is purely a performance thing so we don't have to
231 * call obd_size_diskmd() all the time. */
232 __u32 cl_default_mds_easize;
233 __u32 cl_max_mds_easize;
234 __u32 cl_default_mds_cookiesize;
235 __u32 cl_max_mds_cookiesize;
237 enum lustre_sec_part cl_sp_me;
238 enum lustre_sec_part cl_sp_to;
239 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
241 /* the grant values are protected by loi_list_lock below */
242 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
243 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
244 unsigned long cl_dirty_transit; /* dirty synchronous */
245 unsigned long cl_avail_grant; /* bytes of credit for ost */
246 unsigned long cl_lost_grant; /* lost credits (trunc) */
248 /* since we allocate grant by blocks, we don't know how many grant will
249 * be used to add a page into cache. As a solution, we reserve maximum
250 * grant before trying to dirty a page and unreserve the rest.
251 * See osc_{reserve|unreserve}_grant for details. */
252 long cl_reserved_grant;
253 struct list_head cl_cache_waiters; /* waiting for cache/grant */
254 cfs_time_t cl_next_shrink_grant; /* jiffies */
255 struct list_head cl_grant_shrink_list; /* Timeout event list */
256 int cl_grant_shrink_interval; /* seconds */
258 /* A chunk is an optimal size used by osc_extent to determine
259 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
261 int cl_extent_tax; /* extent overhead, by bytes */
263 /* keep track of objects that have lois that contain pages which
264 * have been queued for async brw. this lock also protects the
265 * lists of osc_client_pages that hang off of the loi */
267 * ->cl_loi_list_lock protects consistency of
268 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
269 * ->ap_completion() call-backs are executed under this lock. As we
270 * cannot guarantee that these call-backs never block on all platforms
271 * (as a matter of fact they do block on Mac OS X), type of
272 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
273 * and blocking mutex on Mac OS X. (Alternative is to make this lock
274 * blocking everywhere, but we don't want to slow down fast-path of
275 * our main platform.)
277 * NB by Jinshan: though field names are still _loi_, but actually
278 * osc_object{}s are in the list.
280 spinlock_t cl_loi_list_lock;
281 struct list_head cl_loi_ready_list;
282 struct list_head cl_loi_hp_ready_list;
283 struct list_head cl_loi_write_list;
284 struct list_head cl_loi_read_list;
285 __u32 cl_r_in_flight;
286 __u32 cl_w_in_flight;
287 /* just a sum of the loi/lop pending numbers to be exported by /proc */
288 atomic_t cl_pending_w_pages;
289 atomic_t cl_pending_r_pages;
290 __u32 cl_max_pages_per_rpc;
291 __u32 cl_max_rpcs_in_flight;
292 struct obd_histogram cl_read_rpc_hist;
293 struct obd_histogram cl_write_rpc_hist;
294 struct obd_histogram cl_read_page_hist;
295 struct obd_histogram cl_write_page_hist;
296 struct obd_histogram cl_read_offset_hist;
297 struct obd_histogram cl_write_offset_hist;
299 /* lru for osc caching pages */
300 struct cl_client_cache *cl_cache;
301 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
302 atomic_long_t *cl_lru_left;
303 atomic_long_t cl_lru_busy;
304 atomic_long_t cl_lru_in_list;
305 atomic_long_t cl_unstable_count;
306 struct list_head cl_lru_list; /* lru page list */
307 spinlock_t cl_lru_list_lock; /* page list protector */
308 atomic_t cl_lru_shrinkers;
310 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
311 atomic_t cl_destroy_in_flight;
312 wait_queue_head_t cl_destroy_waitq;
314 struct mdc_rpc_lock *cl_rpc_lock;
315 struct mdc_rpc_lock *cl_close_lock;
318 struct mutex cl_mgc_mutex;
319 struct local_oid_storage *cl_mgc_los;
320 struct dt_object *cl_mgc_configs_dir;
321 atomic_t cl_mgc_refcount;
322 struct obd_export *cl_mgc_mgsexp;
324 /* checksumming for data sent over the network */
325 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
326 /* supported checksum types that are worked out at connect time */
327 __u32 cl_supp_cksum_types;
328 /* checksum algorithm to be used */
329 cksum_type_t cl_cksum_type;
331 /* also protected by the poorly named _loi_list_lock lock above */
332 struct osc_async_rc cl_ar;
334 /* used by quotacheck when the servers are older than 2.4 */
335 int cl_qchk_stat; /* quotacheck stat of the peer */
336 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
337 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 53, 0)
338 #warning "please consider removing quotacheck compatibility code"
341 /* sequence manager */
342 struct lu_client_seq *cl_seq;
344 atomic_t cl_resends; /* resend count */
346 /* ptlrpc work for writeback in ptlrpcd context */
347 void *cl_writeback_work;
349 /* hash tables for osc_quota_info */
350 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
352 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
359 struct echo_client_obd {
360 struct obd_export *ec_exp; /* the local connection to osc/lov */
362 struct list_head ec_objects;
363 struct list_head ec_locks;
367 /* Generic subset of OSTs */
369 __u32 *op_array; /* array of index of
371 unsigned int op_count; /* number of OSTs in the array */
372 unsigned int op_size; /* allocated size of lp_array */
373 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
376 /* allow statfs data caching for 1 second */
377 #define OBD_STATFS_CACHE_SECONDS 1
379 struct lov_tgt_desc {
380 struct list_head ltd_kill;
381 struct obd_uuid ltd_uuid;
382 struct obd_device *ltd_obd;
383 struct obd_export *ltd_exp;
385 __u32 ltd_index; /* index in lov_obd->tgts */
386 unsigned long ltd_active:1,/* is this target up for requests */
387 ltd_activate:1,/* should target be activated */
388 ltd_reap:1; /* should this target be deleted */
392 struct lov_desc desc;
393 struct lov_tgt_desc **lov_tgts; /* sparse array */
394 struct ost_pool lov_packed; /* all OSTs in a packed
396 struct mutex lov_lock;
397 struct obd_connect_data lov_ocd;
398 struct proc_dir_entry *targets_proc_entry;
399 atomic_t lov_refcount;
400 __u32 lov_death_row; /* tgts scheduled to be deleted */
401 __u32 lov_tgt_size; /* size of tgts array */
404 cfs_hash_t *lov_pools_hash_body; /* used for key access */
405 struct list_head lov_pool_list; /* used for sequential access */
406 struct proc_dir_entry *lov_pool_proc_entry;
407 enum lustre_sec_part lov_sp_me;
409 /* Cached LRU and unstable data from upper layer */
412 struct rw_semaphore lov_notify_lock;
415 struct lmv_tgt_desc {
416 struct obd_uuid ltd_uuid;
417 struct obd_export *ltd_exp;
419 struct mutex ltd_fid_mutex;
420 unsigned long ltd_active:1; /* target up for requests */
423 enum placement_policy {
424 PLACEMENT_CHAR_POLICY = 0,
425 PLACEMENT_NID_POLICY = 1,
426 PLACEMENT_INVAL_POLICY = 2,
430 typedef enum placement_policy placement_policy_t;
434 struct lu_client_fld lmv_fld;
436 placement_policy_t lmv_placement;
437 struct lmv_desc desc;
438 struct obd_uuid cluuid;
439 struct obd_export *exp;
440 struct proc_dir_entry *targets_proc_entry;
442 struct mutex lmv_init_mutex;
447 int max_def_cookiesize;
449 __u32 tgts_size; /* size of tgts array */
450 struct lmv_tgt_desc **tgts;
452 struct obd_connect_data conn_data;
455 struct niobuf_local {
456 __u64 lnb_file_offset;
457 __u32 lnb_page_offset;
460 struct page *lnb_page;
465 #define LUSTRE_FLD_NAME "fld"
466 #define LUSTRE_SEQ_NAME "seq"
468 #define LUSTRE_MDD_NAME "mdd"
469 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
470 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
471 #define LUSTRE_VVP_NAME "vvp"
472 #define LUSTRE_LMV_NAME "lmv"
473 #define LUSTRE_SLP_NAME "slp"
474 #define LUSTRE_LOD_NAME "lod"
475 #define LUSTRE_OSP_NAME "osp"
476 #define LUSTRE_LWP_NAME "lwp"
478 /* obd device type names */
479 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
480 #define LUSTRE_MDS_NAME "mds"
481 #define LUSTRE_MDT_NAME "mdt"
482 #define LUSTRE_MDC_NAME "mdc"
483 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
484 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
485 #define LUSTRE_OSC_NAME "osc"
486 #define LUSTRE_LOV_NAME "lov"
487 #define LUSTRE_MGS_NAME "mgs"
488 #define LUSTRE_MGC_NAME "mgc"
490 #define LUSTRE_ECHO_NAME "obdecho"
491 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
492 #define LUSTRE_QMT_NAME "qmt"
494 /* Constant obd names (post-rename) */
495 #define LUSTRE_MDS_OBDNAME "MDS"
496 #define LUSTRE_OSS_OBDNAME "OSS"
497 #define LUSTRE_MGS_OBDNAME "MGS"
498 #define LUSTRE_MGC_OBDNAME "MGC"
500 struct obd_trans_info {
502 /* Only used on the server side for tracking acks. */
503 struct oti_req_ack_lock {
504 struct lustre_handle lock;
508 struct llog_cookie oti_onecookie;
509 struct llog_cookie *oti_logcookies;
512 __u64 oti_pre_version;
516 * Events signalled through obd_notify() upcall-chain.
518 enum obd_notify_event {
521 /* Device connect start */
523 /* Device activated */
525 /* Device deactivated */
527 /* Device disconnected */
529 /* Connect data for import were changed */
532 OBD_NOTIFY_SYNC_NONBLOCK,
534 /* Configuration event */
536 /* Administratively deactivate/activate event */
537 OBD_NOTIFY_DEACTIVATE,
541 /* bit-mask flags for config events */
543 CONFIG_LOG = 0x1, /* finished processing config log */
544 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
545 CONFIG_TARGET = 0x4 /* one target is added */
549 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
550 * and liblustre being main examples).
552 struct obd_notify_upcall {
553 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
554 enum obd_notify_event ev, void *owner, void *data);
555 /* Opaque datum supplied by upper layer listener */
559 struct target_recovery_data {
560 svc_handler_t trd_recovery_handler;
561 pid_t trd_processing_task;
562 struct completion trd_starting;
563 struct completion trd_finishing;
566 struct obd_llog_group {
567 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
568 wait_queue_head_t olg_waitq;
570 struct mutex olg_cat_processing;
573 /* corresponds to one of the obd's */
574 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
577 struct obd_type *obd_type;
580 /* common and UUID name of this device */
581 char obd_name[MAX_OBD_NAME];
582 struct obd_uuid obd_uuid;
584 struct lu_device *obd_lu_dev;
586 /* bitfield modification is protected by obd_dev_lock */
588 obd_attached:1, /* finished attach */
589 obd_set_up:1, /* finished setup */
590 obd_recovering:1, /* there are recoverable clients */
591 obd_abort_recovery:1, /* recovery expired */
592 obd_version_recov:1, /* obd uses version checking */
593 obd_replayable:1, /* recovery is enabled;
595 obd_no_transno:1, /* no committed-transno notification */
596 obd_no_recov:1, /* fail instead of retry messages */
597 obd_stopping:1, /* started cleanup */
598 obd_starting:1, /* started setup */
599 obd_force:1, /* cleanup with > 0 obd refcount */
600 obd_fail:1, /* cleanup with failover */
601 obd_no_conn:1, /* deny new connections */
602 obd_inactive:1, /* device active/inactive
603 * (for /proc/status only!!) */
604 obd_no_ir:1, /* no imperative recovery. */
605 obd_process_conf:1, /* device is processing mgs config */
606 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
608 /* use separate field as it is set in interrupt to don't mess with
609 * protection of other bits using _bh lock */
610 unsigned long obd_recovery_expired:1;
611 /* uuid-export hash body */
612 cfs_hash_t *obd_uuid_hash;
613 /* nid-export hash body */
614 cfs_hash_t *obd_nid_hash;
616 cfs_hash_t *obd_nid_stats_hash;
617 struct list_head obd_nid_stats;
618 atomic_t obd_refcount;
619 struct list_head obd_exports;
620 struct list_head obd_unlinked_exports;
621 struct list_head obd_delayed_exports;
622 struct list_head obd_lwp_list;
624 spinlock_t obd_nid_lock;
625 struct ldlm_namespace *obd_namespace;
626 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
627 /* a spinlock is OK for what we do now, may need a semaphore later */
628 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
629 struct mutex obd_dev_mutex;
630 __u64 obd_last_committed;
631 spinlock_t obd_osfs_lock;
632 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
634 struct lvfs_run_ctxt obd_lvfs_ctxt;
635 struct obd_llog_group obd_olg; /* default llog group */
636 struct obd_device *obd_observer;
637 struct rw_semaphore obd_observer_link_sem;
638 struct obd_notify_upcall obd_upcall;
639 struct obd_export *obd_self_export;
640 struct obd_export *obd_lwp_export;
641 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
642 struct list_head obd_exports_timed;
643 time_t obd_eviction_timer; /* for ping evictor */
645 int obd_max_recoverable_clients;
646 atomic_t obd_connected_clients;
647 int obd_stale_clients;
648 /* this lock protects all recovery list_heads, timer and
649 * obd_next_recovery_transno value */
650 spinlock_t obd_recovery_task_lock;
651 __u64 obd_next_recovery_transno;
652 int obd_replayed_requests;
653 int obd_requests_queued_for_recovery;
654 wait_queue_head_t obd_next_transno_waitq;
655 /* protected by obd_recovery_task_lock */
656 struct timer_list obd_recovery_timer;
658 time_t obd_recovery_start;
659 /* seconds, for lprocfs_status */
660 time_t obd_recovery_end;
661 int obd_recovery_time_hard;
662 int obd_recovery_timeout;
663 int obd_recovery_ir_factor;
665 /* new recovery stuff from CMD2 */
666 struct target_recovery_data obd_recovery_data;
667 int obd_replayed_locks;
668 atomic_t obd_req_replay_clients;
669 atomic_t obd_lock_replay_clients;
670 /* all lists are protected by obd_recovery_task_lock */
671 struct list_head obd_req_replay_queue;
672 struct list_head obd_lock_replay_queue;
673 struct list_head obd_final_req_queue;
676 #ifdef HAVE_SERVER_SUPPORT
677 struct obd_device_target obt;
678 struct filter_obd filter;
680 struct echo_obd echo;
682 struct client_obd cli;
683 struct echo_client_obd echo_client;
687 /* Fields used by LProcFS */
688 unsigned int obd_cntr_base;
689 struct lprocfs_stats *obd_stats;
691 unsigned int obd_md_cntr_base;
692 struct lprocfs_stats *obd_md_stats;
694 struct proc_dir_entry *obd_proc_entry;
695 struct proc_dir_entry *obd_proc_exports_entry;
696 struct proc_dir_entry *obd_svc_procroot;
697 struct lprocfs_stats *obd_svc_stats;
698 struct lprocfs_seq_vars *obd_vars;
699 atomic_t obd_evict_inprogress;
700 wait_queue_head_t obd_evict_inprogress_waitq;
701 struct list_head obd_evict_list; /* protected with pet_lock */
704 * Ldlm pool part. Save last calculated SLV and Limit.
706 rwlock_t obd_pool_lock;
711 * A list of outstanding class_incref()'s against this obd. For
714 struct lu_ref obd_reference;
716 int obd_conn_inprogress;
719 enum obd_cleanup_stage {
720 /* Special case hack for MDS LOVs */
722 /* can be directly mapped to .ldto_device_fini() */
726 /* get/set_info keys */
727 #define KEY_ASYNC "async"
728 #define KEY_CAPA_KEY "capa_key"
729 #define KEY_CHANGELOG_CLEAR "changelog_clear"
730 #define KEY_FID2PATH "fid2path"
731 #define KEY_CHECKSUM "checksum"
732 #define KEY_CLEAR_FS "clear_fs"
733 #define KEY_CONN_DATA "conn_data"
734 #define KEY_EVICT_BY_NID "evict_by_nid"
735 #define KEY_FIEMAP "fiemap"
736 #define KEY_FLUSH_CTX "flush_ctx"
737 #define KEY_GRANT_SHRINK "grant_shrink"
738 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
739 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
740 #define KEY_INTERMDS "inter_mds"
741 #define KEY_LAST_ID "last_id"
742 #define KEY_LAST_FID "last_fid"
743 #define KEY_LOVDESC "lovdesc"
744 #define KEY_MAX_EASIZE "max_easize"
745 #define KEY_DEFAULT_EASIZE "default_easize"
746 #define KEY_MAX_COOKIESIZE "max_cookiesize"
747 #define KEY_DEFAULT_COOKIESIZE "default_cookiesize"
748 #define KEY_MGSSEC "mgssec"
749 #define KEY_READ_ONLY "read-only"
750 #define KEY_REGISTER_TARGET "register_target"
751 #define KEY_SET_FS "set_fs"
752 #define KEY_TGT_COUNT "tgt_count"
753 /* KEY_SET_INFO in lustre_idl.h */
754 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
756 #define KEY_CACHE_SET "cache_set"
757 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
758 #define KEY_OSP_CONNECTED "osp_connected"
762 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
763 #define IT_OPEN (1 << 0)
764 #define IT_CREAT (1 << 1)
765 #define IT_READDIR (1 << 2)
766 #define IT_GETATTR (1 << 3)
767 #define IT_LOOKUP (1 << 4)
768 #define IT_UNLINK (1 << 5)
769 #define IT_TRUNC (1 << 6)
770 #define IT_GETXATTR (1 << 7)
771 #define IT_EXEC (1 << 8)
772 #define IT_PIN (1 << 9)
773 #define IT_LAYOUT (1 << 10)
774 #define IT_QUOTA_DQACQ (1 << 11)
775 #define IT_QUOTA_CONN (1 << 12)
776 #define IT_SETXATTR (1 << 13)
778 static inline int it_to_lock_mode(struct lookup_intent *it)
780 /* CREAT needs to be tested before open (both could be set) */
781 if (it->it_op & IT_CREAT)
783 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
786 else if (it->it_op & IT_READDIR)
788 else if (it->it_op & IT_GETXATTR)
790 else if (it->it_op & IT_SETXATTR)
793 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
798 CLI_SET_MEA = 1 << 0,
799 CLI_RM_ENTRY = 1 << 1,
802 CLI_MIGRATE = 1 << 4,
806 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
807 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
808 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
809 struct lu_fid op_fid4; /* to the operation locks. */
810 u32 op_mds; /* what mds server open will go to */
811 struct lustre_handle op_handle;
812 obd_time op_mod_time;
816 struct lmv_stripe_md *op_mea1;
817 struct lmv_stripe_md *op_mea2;
818 __u32 op_suppgids[2];
825 /* iattr fields and blocks. */
826 struct iattr op_attr;
827 loff_t op_attr_blocks;
828 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
829 __u64 op_valid; /* OBD_MD_* */
831 /* Size-on-MDS epoch and flags. */
833 enum md_op_flags op_flags;
836 struct obd_capa *op_capa1;
837 struct obd_capa *op_capa2;
839 /* Various operation flags. */
840 enum mds_op_bias op_bias;
842 /* Used by readdir */
843 unsigned int op_max_pages;
845 /* used to transfer info between the stacks of MD client
846 * see enum op_cli_flags */
847 enum md_cli_flags op_cli_flags;
849 /* File object data version for HSM release, on client */
850 __u64 op_data_version;
851 struct lustre_handle op_lease_handle;
855 int (*md_blocking_ast)(struct ldlm_lock *lock,
856 struct ldlm_lock_desc *desc,
857 void *data, int flag);
860 struct md_enqueue_info;
861 /* metadata stat-ahead */
862 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
863 struct md_enqueue_info *minfo,
866 struct md_enqueue_info {
867 struct md_op_data mi_data;
868 struct lookup_intent mi_it;
869 struct lustre_handle mi_lockh;
870 struct inode *mi_dir;
871 md_enqueue_cb_t mi_cb;
876 struct module *o_owner;
877 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
878 void *karg, void __user *uarg);
879 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
880 __u32 keylen, void *key, __u32 *vallen, void *val,
881 struct lov_stripe_md *lsm);
882 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
883 __u32 keylen, void *key,
884 __u32 vallen, void *val,
885 struct ptlrpc_request_set *set);
886 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
887 int (*o_precleanup)(struct obd_device *dev,
888 enum obd_cleanup_stage cleanup_stage);
889 int (*o_cleanup)(struct obd_device *dev);
890 int (*o_process_config)(struct obd_device *dev, obd_count len,
892 int (*o_postrecov)(struct obd_device *dev);
893 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
895 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
896 /* connect to the target device with given connection
897 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
898 * granted by the target, which are guaranteed to be a subset of flags
899 * asked for. If @ocd == NULL, use default parameters. */
900 int (*o_connect)(const struct lu_env *env,
901 struct obd_export **exp, struct obd_device *src,
902 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
904 int (*o_reconnect)(const struct lu_env *env,
905 struct obd_export *exp, struct obd_device *src,
906 struct obd_uuid *cluuid,
907 struct obd_connect_data *ocd,
909 int (*o_disconnect)(struct obd_export *exp);
911 /* Initialize/finalize fids infrastructure. */
912 int (*o_fid_init)(struct obd_device *obd,
913 struct obd_export *exp, enum lu_cli_type type);
914 int (*o_fid_fini)(struct obd_device *obd);
916 /* Allocate new fid according to passed @hint. */
917 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
918 struct lu_fid *fid, struct md_op_data *op_data);
921 * Object with @fid is getting deleted, we may want to do something
924 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
925 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
926 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
927 __u64 max_age, struct ptlrpc_request_set *set);
928 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
929 struct lov_stripe_md *mem_src);
930 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
931 struct lov_mds_md *disk_src, int disk_len);
932 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
933 struct obdo *oa, struct obd_trans_info *oti);
934 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
935 struct obdo *oa, struct obd_trans_info *oti);
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 int (*o_import_event)(struct obd_device *, struct obd_import *,
964 enum obd_import_event);
966 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
967 enum obd_notify_event ev, void *data);
969 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
970 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
973 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
974 struct obd_quotactl *);
975 int (*o_quotactl)(struct obd_device *, struct obd_export *,
976 struct obd_quotactl *);
978 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
981 int (*o_pool_new)(struct obd_device *obd, char *poolname);
982 int (*o_pool_del)(struct obd_device *obd, char *poolname);
983 int (*o_pool_add)(struct obd_device *obd, char *poolname,
985 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
987 void (*o_getref)(struct obd_device *obd);
988 void (*o_putref)(struct obd_device *obd);
990 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
991 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
992 * Also, add a wrapper function in include/linux/obd_class.h. */
997 struct mdt_body *body;
998 struct lov_stripe_md *lsm;
999 struct lmv_stripe_md *lmv;
1000 #ifdef CONFIG_FS_POSIX_ACL
1001 struct posix_acl *posix_acl;
1003 struct mdt_remote_perm *remote_perm;
1004 struct obd_capa *mds_capa;
1005 struct obd_capa *oss_capa;
1008 struct md_open_data {
1009 struct obd_client_handle *mod_och;
1010 struct ptlrpc_request *mod_open_req;
1011 struct ptlrpc_request *mod_close_req;
1012 atomic_t mod_refcount;
1016 struct obd_client_handle {
1017 struct lustre_handle och_fh;
1018 struct lu_fid och_fid;
1019 struct md_open_data *och_mod;
1020 struct lustre_handle och_lease_handle; /* open lock for lease */
1025 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1027 struct lookup_intent;
1031 /* Every operation from MD_STATS_FIRST_OP up to and including
1032 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1033 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1034 * Operations after MD_STATS_LAST_OP are excluded from stats.
1035 * There are a few reasons for doing this: we prune the 17
1036 * counters which will be of minimal use in understanding
1037 * metadata utilization, we save memory by allocating 15
1038 * instead of 32 counters, we save cycles by not counting.
1040 * MD_STATS_FIRST_OP must be the first member of md_ops.
1042 #define MD_STATS_FIRST_OP m_close
1043 int (*m_close)(struct obd_export *, struct md_op_data *,
1044 struct md_open_data *, struct ptlrpc_request **);
1046 int (*m_create)(struct obd_export *, struct md_op_data *,
1047 const void *, size_t, umode_t, uid_t, gid_t,
1048 cfs_cap_t, __u64, struct ptlrpc_request **);
1050 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1051 const union ldlm_policy_data *,
1052 struct lookup_intent *, struct md_op_data *,
1053 struct lustre_handle *, __u64);
1055 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1056 struct ptlrpc_request **);
1058 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1059 struct lookup_intent *,
1060 struct ptlrpc_request **,
1061 ldlm_blocking_callback, __u64);
1063 int (*m_link)(struct obd_export *, struct md_op_data *,
1064 struct ptlrpc_request **);
1066 int (*m_rename)(struct obd_export *, struct md_op_data *,
1067 const char *, size_t, const char *, size_t,
1068 struct ptlrpc_request **);
1070 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1071 size_t , void *, size_t, struct ptlrpc_request **,
1072 struct md_open_data **mod);
1074 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1075 struct obd_capa *, struct ptlrpc_request **);
1077 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1078 struct md_callback *cb_op, __u64 hash_offset,
1079 struct page **ppage);
1081 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1082 struct ptlrpc_request **);
1084 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1085 struct obd_capa *, obd_valid, const char *,
1086 const char *, int, int, int, __u32,
1087 struct ptlrpc_request **);
1089 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1090 struct obd_capa *, obd_valid, const char *,
1091 const char *, int, int, int,
1092 struct ptlrpc_request **);
1094 int (*m_intent_getattr_async)(struct obd_export *,
1095 struct md_enqueue_info *,
1096 struct ldlm_enqueue_info *);
1098 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1099 struct lu_fid *, __u64 *bits);
1101 #define MD_STATS_LAST_OP m_revalidate_lock
1103 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1104 struct obd_capa **);
1106 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1108 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1109 ldlm_iterator_t, void *);
1111 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1112 struct md_open_data *);
1114 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1115 struct ptlrpc_request **);
1117 int (*m_init_ea_size)(struct obd_export *, __u32, __u32, __u32, __u32);
1119 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1120 struct obd_export *, struct obd_export *,
1121 struct lustre_md *);
1123 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1125 int (*m_merge_attr)(struct obd_export *,
1126 const struct lmv_stripe_md *lsm,
1127 struct cl_attr *attr);
1129 int (*m_update_lsm_md)(struct obd_export *, struct lmv_stripe_md *lsm,
1130 struct mdt_body *, ldlm_blocking_callback);
1132 int (*m_set_open_replay_data)(struct obd_export *,
1133 struct obd_client_handle *,
1134 struct lookup_intent *);
1136 int (*m_clear_open_replay_data)(struct obd_export *,
1137 struct obd_client_handle *);
1139 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1141 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1142 const struct lu_fid *, ldlm_type_t,
1143 ldlm_policy_data_t *, ldlm_mode_t,
1144 struct lustre_handle *);
1146 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1147 ldlm_policy_data_t *, ldlm_mode_t,
1148 ldlm_cancel_flags_t flags, void *opaque);
1150 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1151 renew_capa_cb_t cb);
1153 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1154 const struct req_msg_field *, struct obd_capa **);
1156 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1157 struct obd_capa *, __u32,
1158 struct ptlrpc_request **);
1160 int (*m_get_fid_from_lsm)(struct obd_export *,
1161 const struct lmv_stripe_md *,
1162 const char *name, int namelen,
1163 struct lu_fid *fid);
1166 struct lsm_operations {
1167 void (*lsm_free)(struct lov_stripe_md *);
1168 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1170 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1172 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1173 __u16 *stripe_count);
1174 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1175 struct lov_mds_md *lmm);
1178 extern const struct lsm_operations lsm_v1_ops;
1179 extern const struct lsm_operations lsm_v3_ops;
1180 static inline const struct lsm_operations *lsm_op_find(int magic)
1188 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1193 static inline struct md_open_data *obd_mod_alloc(void)
1195 struct md_open_data *mod;
1199 atomic_set(&mod->mod_refcount, 1);
1203 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1204 #define obd_mod_put(mod) \
1206 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1207 if ((mod)->mod_open_req) \
1208 ptlrpc_req_finished((mod)->mod_open_req); \
1209 OBD_FREE_PTR(mod); \
1213 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1214 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1216 /* return 1 if client should be resend request */
1217 static inline int client_should_resend(int resend, struct client_obd *cli)
1219 return atomic_read(&cli->cl_resends) ?
1220 atomic_read(&cli->cl_resends) > resend : 1;
1224 * Return device name for this device
1226 * XXX: lu_device is declared before obd_device, while a pointer pointing
1227 * back to obd_device in lu_device, so this helper function defines here
1228 * instead of in lu_object.h
1230 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1232 return lu_dev->ld_obd->obd_name;
1235 static inline bool filename_is_volatile(const char *name, size_t namelen,
1241 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1244 /* caller does not care of idx */
1248 /* volatile file, the MDT can be set from name */
1249 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1250 /* if no MDT is specified, use std way */
1251 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1253 /* test for no MDT idx case */
1254 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1255 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1259 /* we have an idx, read it */
1260 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1261 *idx = strtoul(start, &end, 16);
1263 * no digit, no trailing :, negative value
1265 if (((*idx == 0) && (end == start)) ||
1266 (*end != ':') || (*idx < 0))
1271 /* bad format of mdt idx, we cannot return an error
1272 * to caller so we use hash algo */
1273 CERROR("Bad volatile file name format: %s\n",
1274 name + LUSTRE_VOLATILE_HDR_LEN);
1278 static inline int cli_brw_size(struct obd_device *obd)
1280 LASSERT(obd != NULL);
1281 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1284 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1285 * of the client should be increased accordingly to avoid sending fragmented
1286 * RPCs over the network when the client runs out of the maximum dirty space
1287 * when so many RPCs are being generated.
1289 static inline void client_adjust_max_dirty(struct client_obd *cli)
1292 if (cli->cl_dirty_max_pages <= 0)
1293 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1294 >> PAGE_CACHE_SHIFT;
1296 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1297 cli->cl_max_pages_per_rpc;
1299 if (dirty_max > cli->cl_dirty_max_pages)
1300 cli->cl_dirty_max_pages = dirty_max;
1303 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1304 cli->cl_dirty_max_pages = totalram_pages / 8;
1307 #endif /* __OBD_H */