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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/spinlock.h>
38 #include <lustre/lustre_idl.h>
39 #include <lustre_lib.h>
40 #include <libcfs/bitmap.h>
41 #ifdef HAVE_SERVER_SUPPORT
42 # include <lu_target.h>
43 # include <obd_target.h>
46 #include <lustre_export.h>
47 #include <lustre_fid.h>
48 #include <lustre_fld.h>
49 #include <lustre_handles.h>
50 #include <lustre_intent.h>
52 #include <lustre_quota.h>
54 #define MAX_OBD_DEVICES 8192
62 struct lov_oinfo { /* per-stripe data structure */
63 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
64 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
65 int loi_ost_gen; /* generation of this loi_ost_idx */
67 unsigned long loi_kms_valid:1;
68 __u64 loi_kms; /* known minimum size */
69 struct ost_lvb loi_lvb;
70 struct osc_async_rc loi_ar;
73 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
76 oinfo->loi_kms_valid = 1;
82 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
84 /* obd info for a particular level (lov, osc). */
86 /* OBD_STATFS_* flags */
88 /* statfs data specific for every OSC, if needed at all. */
89 struct obd_statfs *oi_osfs;
90 /* An update callback which is called to update some data on upper
91 * level. E.g. it is used for update lsm->lsm_oinfo at every received
92 * request in osc level for enqueue requests. It is also possible to
93 * update some caller data from LOV layer if needed. */
94 obd_enqueue_update_f oi_cb_up;
98 struct list_head typ_chain;
99 struct obd_ops *typ_dt_ops;
100 struct md_ops *typ_md_ops;
101 struct proc_dir_entry *typ_procroot;
102 struct proc_dir_entry *typ_procsym;
103 __u32 typ_sym_filter;
106 struct lu_device_type *typ_lu;
107 spinlock_t obd_type_lock;
117 struct timeout_item {
118 enum timeout_event ti_event;
119 cfs_time_t ti_timeout;
122 struct list_head ti_obd_list;
123 struct list_head ti_chain;
126 #define OBD_MAX_RIF_DEFAULT 8
127 #define OBD_MAX_RIF_MAX 512
128 #define OSC_MAX_RIF_MAX 256
129 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
130 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
131 #define OSC_DEFAULT_RESENDS 10
133 /* possible values for fo_sync_lock_cancel */
135 NEVER_SYNC_ON_CANCEL = 0,
136 BLOCKING_SYNC_ON_CANCEL = 1,
137 ALWAYS_SYNC_ON_CANCEL = 2,
138 NUM_SYNC_ON_CANCEL_STATES
142 * Limit reply buffer size for striping data to one x86_64 page. This
143 * value is chosen to fit the striping data for common use cases while
144 * staying well below the limit at which the buffer must be backed by
145 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
148 #define OBD_MAX_DEFAULT_EA_SIZE 4096
150 enum obd_cl_sem_lock_class {
159 struct rw_semaphore cl_sem;
160 struct obd_uuid cl_target_uuid;
161 struct obd_import *cl_import; /* ptlrpc connection state */
162 size_t cl_conn_count;
164 /* Cache maximum and default values for easize. This is
165 * strictly a performance optimization to minimize calls to
166 * obd_size_diskmd(). The default values are used to calculate the
167 * initial size of a request buffer. The ptlrpc layer will resize the
168 * buffer as needed to accommodate a larger reply from the
169 * server. The default values should be small enough to avoid wasted
170 * memory and excessive use of vmalloc(), yet large enough to avoid
171 * reallocating the buffer in the common use case. */
173 /* Default EA size for striping attributes. It is initialized at
174 * mount-time based on the default stripe width of the filesystem,
175 * then it tracks the largest observed EA size advertised by
176 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
177 __u32 cl_default_mds_easize;
179 /* Maximum possible EA size computed at mount-time based on
180 * the number of OSTs in the filesystem. May be increased at
181 * run-time if a larger observed size is advertised by the MDT. */
182 __u32 cl_max_mds_easize;
184 enum lustre_sec_part cl_sp_me;
185 enum lustre_sec_part cl_sp_to;
186 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
188 /* the grant values are protected by loi_list_lock below */
189 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
190 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
191 unsigned long cl_dirty_transit; /* dirty synchronous */
192 unsigned long cl_avail_grant; /* bytes of credit for ost */
193 unsigned long cl_lost_grant; /* lost credits (trunc) */
194 /* grant consumed for dirty pages */
195 unsigned long cl_dirty_grant;
197 /* since we allocate grant by blocks, we don't know how many grant will
198 * be used to add a page into cache. As a solution, we reserve maximum
199 * grant before trying to dirty a page and unreserve the rest.
200 * See osc_{reserve|unreserve}_grant for details. */
201 long cl_reserved_grant;
202 struct list_head cl_cache_waiters; /* waiting for cache/grant */
203 cfs_time_t cl_next_shrink_grant; /* jiffies */
204 struct list_head cl_grant_shrink_list; /* Timeout event list */
205 int cl_grant_shrink_interval; /* seconds */
207 /* A chunk is an optimal size used by osc_extent to determine
208 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
210 /* extent insertion metadata overhead to be accounted in grant,
212 unsigned int cl_grant_extent_tax;
213 /* maximum extent size, in number of pages */
214 unsigned int cl_max_extent_pages;
216 /* keep track of objects that have lois that contain pages which
217 * have been queued for async brw. this lock also protects the
218 * lists of osc_client_pages that hang off of the loi */
220 * ->cl_loi_list_lock protects consistency of
221 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
222 * ->ap_completion() call-backs are executed under this lock. As we
223 * cannot guarantee that these call-backs never block on all platforms
224 * (as a matter of fact they do block on Mac OS X), type of
225 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
226 * and blocking mutex on Mac OS X. (Alternative is to make this lock
227 * blocking everywhere, but we don't want to slow down fast-path of
228 * our main platform.)
230 * NB by Jinshan: though field names are still _loi_, but actually
231 * osc_object{}s are in the list.
233 spinlock_t cl_loi_list_lock;
234 struct list_head cl_loi_ready_list;
235 struct list_head cl_loi_hp_ready_list;
236 struct list_head cl_loi_write_list;
237 struct list_head cl_loi_read_list;
238 __u32 cl_r_in_flight;
239 __u32 cl_w_in_flight;
240 /* just a sum of the loi/lop pending numbers to be exported by /proc */
241 atomic_t cl_pending_w_pages;
242 atomic_t cl_pending_r_pages;
243 __u32 cl_max_pages_per_rpc;
244 __u32 cl_max_rpcs_in_flight;
245 struct obd_histogram cl_read_rpc_hist;
246 struct obd_histogram cl_write_rpc_hist;
247 struct obd_histogram cl_read_page_hist;
248 struct obd_histogram cl_write_page_hist;
249 struct obd_histogram cl_read_offset_hist;
250 struct obd_histogram cl_write_offset_hist;
252 /** LRU for osc caching pages */
253 struct cl_client_cache *cl_cache;
254 /** member of cl_cache->ccc_lru */
255 struct list_head cl_lru_osc;
256 /** # of available LRU slots left in the per-OSC cache.
257 * Available LRU slots are shared by all OSCs of the same file system,
258 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
259 atomic_long_t *cl_lru_left;
260 /** # of busy LRU pages. A page is considered busy if it's in writeback
261 * queue, or in transfer. Busy pages can't be discarded so they are not
263 atomic_long_t cl_lru_busy;
264 /** # of LRU pages in the cache for this client_obd */
265 atomic_long_t cl_lru_in_list;
266 /** # of threads are shrinking LRU cache. To avoid contention, it's not
267 * allowed to have multiple threads shrinking LRU cache. */
268 atomic_t cl_lru_shrinkers;
269 /** The time when this LRU cache was last used. */
270 time64_t cl_lru_last_used;
271 /** stats: how many reclaims have happened for this client_obd.
272 * reclaim and shrink - shrink is async, voluntarily rebalancing;
273 * reclaim is sync, initiated by IO thread when the LRU slots are
275 __u64 cl_lru_reclaim;
276 /** List of LRU pages for this client_obd */
277 struct list_head cl_lru_list;
278 /** Lock for LRU page list */
279 spinlock_t cl_lru_list_lock;
280 /** # of unstable pages in this client_obd.
281 * An unstable page is a page state that WRITE RPC has finished but
282 * the transaction has NOT yet committed. */
283 atomic_long_t cl_unstable_count;
284 /** Link to osc_shrinker_list */
285 struct list_head cl_shrink_list;
287 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
288 atomic_t cl_destroy_in_flight;
289 wait_queue_head_t cl_destroy_waitq;
291 struct mdc_rpc_lock *cl_rpc_lock;
293 /* modify rpcs in flight
294 * currently used for metadata only */
295 spinlock_t cl_mod_rpcs_lock;
296 __u16 cl_max_mod_rpcs_in_flight;
297 __u16 cl_mod_rpcs_in_flight;
298 __u16 cl_close_rpcs_in_flight;
299 wait_queue_head_t cl_mod_rpcs_waitq;
300 unsigned long *cl_mod_tag_bitmap;
301 struct obd_histogram cl_mod_rpcs_hist;
304 struct mutex cl_mgc_mutex;
305 struct local_oid_storage *cl_mgc_los;
306 struct dt_object *cl_mgc_configs_dir;
307 atomic_t cl_mgc_refcount;
308 struct obd_export *cl_mgc_mgsexp;
310 /* checksumming for data sent over the network */
311 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
312 /* supported checksum types that are worked out at connect time */
313 __u32 cl_supp_cksum_types;
314 /* checksum algorithm to be used */
315 cksum_type_t cl_cksum_type;
317 /* also protected by the poorly named _loi_list_lock lock above */
318 struct osc_async_rc cl_ar;
320 /* sequence manager */
321 struct lu_client_seq *cl_seq;
322 struct rw_semaphore cl_seq_rwsem;
324 atomic_t cl_resends; /* resend count */
326 /* ptlrpc work for writeback in ptlrpcd context */
327 void *cl_writeback_work;
329 /* hash tables for osc_quota_info */
330 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
332 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
339 struct echo_client_obd {
340 struct obd_export *ec_exp; /* the local connection to osc/lov */
342 struct list_head ec_objects;
343 struct list_head ec_locks;
347 /* Generic subset of OSTs */
349 __u32 *op_array; /* array of index of
351 unsigned int op_count; /* number of OSTs in the array */
352 unsigned int op_size; /* allocated size of lp_array */
353 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
356 /* allow statfs data caching for 1 second */
357 #define OBD_STATFS_CACHE_SECONDS 1
359 struct lov_tgt_desc {
360 struct list_head ltd_kill;
361 struct obd_uuid ltd_uuid;
362 struct obd_device *ltd_obd;
363 struct obd_export *ltd_exp;
365 __u32 ltd_index; /* index in lov_obd->tgts */
366 unsigned long ltd_active:1,/* is this target up for requests */
367 ltd_activate:1,/* should target be activated */
368 ltd_reap:1; /* should this target be deleted */
372 struct lov_desc desc;
373 struct lov_tgt_desc **lov_tgts; /* sparse array */
374 struct ost_pool lov_packed; /* all OSTs in a packed
376 struct mutex lov_lock;
377 struct obd_connect_data lov_ocd;
378 struct proc_dir_entry *targets_proc_entry;
379 atomic_t lov_refcount;
380 __u32 lov_death_row; /* tgts scheduled to be deleted */
381 __u32 lov_tgt_size; /* size of tgts array */
384 struct cfs_hash *lov_pools_hash_body; /* used for key access */
385 struct list_head lov_pool_list; /* used for sequential access */
386 struct proc_dir_entry *lov_pool_proc_entry;
387 enum lustre_sec_part lov_sp_me;
389 /* Cached LRU and unstable data from upper layer */
390 struct cl_client_cache *lov_cache;
392 struct rw_semaphore lov_notify_lock;
395 struct lmv_tgt_desc {
396 struct obd_uuid ltd_uuid;
397 struct obd_export *ltd_exp;
399 struct mutex ltd_fid_mutex;
400 unsigned long ltd_active:1; /* target up for requests */
404 struct lu_client_fld lmv_fld;
406 struct lmv_desc desc;
407 struct obd_uuid cluuid;
408 struct proc_dir_entry *targets_proc_entry;
410 struct mutex lmv_init_mutex;
415 __u32 tgts_size; /* size of tgts array */
416 struct lmv_tgt_desc **tgts;
418 struct obd_connect_data conn_data;
421 struct niobuf_local {
422 __u64 lnb_file_offset;
423 __u32 lnb_page_offset;
427 struct page *lnb_page;
431 struct tgt_thread_big_cache {
432 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
435 #define LUSTRE_FLD_NAME "fld"
436 #define LUSTRE_SEQ_NAME "seq"
438 #define LUSTRE_MDD_NAME "mdd"
439 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
440 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
441 #define LUSTRE_VVP_NAME "vvp"
442 #define LUSTRE_LMV_NAME "lmv"
443 #define LUSTRE_SLP_NAME "slp"
444 #define LUSTRE_LOD_NAME "lod"
445 #define LUSTRE_OSP_NAME "osp"
446 #define LUSTRE_LWP_NAME "lwp"
448 /* obd device type names */
449 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
450 #define LUSTRE_MDS_NAME "mds"
451 #define LUSTRE_MDT_NAME "mdt"
452 #define LUSTRE_MDC_NAME "mdc"
453 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
454 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
455 #define LUSTRE_OSC_NAME "osc"
456 #define LUSTRE_LOV_NAME "lov"
457 #define LUSTRE_MGS_NAME "mgs"
458 #define LUSTRE_MGC_NAME "mgc"
460 #define LUSTRE_ECHO_NAME "obdecho"
461 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
462 #define LUSTRE_QMT_NAME "qmt"
464 /* Constant obd names (post-rename) */
465 #define LUSTRE_MDS_OBDNAME "MDS"
466 #define LUSTRE_OSS_OBDNAME "OSS"
467 #define LUSTRE_MGS_OBDNAME "MGS"
468 #define LUSTRE_MGC_OBDNAME "MGC"
470 static inline int is_lwp_on_mdt(char *name)
474 ptr = strrchr(name, '-');
476 CERROR("%s is not a obdname\n", name);
480 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
482 if (strncmp(ptr + 1, "MDT", 3) != 0)
485 while (*(--ptr) != '-' && ptr != name);
490 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
496 static inline int is_lwp_on_ost(char *name)
500 ptr = strrchr(name, '-');
502 CERROR("%s is not a obdname\n", name);
506 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
508 if (strncmp(ptr + 1, "OST", 3) != 0)
511 while (*(--ptr) != '-' && ptr != name);
516 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
523 * Events signalled through obd_notify() upcall-chain.
525 enum obd_notify_event {
526 /* Device connect start */
528 /* Device activated */
530 /* Device deactivated */
532 /* Connect data for import were changed */
534 /* Administratively deactivate/activate event */
535 OBD_NOTIFY_DEACTIVATE,
540 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
541 * and liblustre being main examples).
543 struct obd_notify_upcall {
544 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
545 enum obd_notify_event ev, void *owner);
546 /* Opaque datum supplied by upper layer listener */
550 struct target_recovery_data {
551 svc_handler_t trd_recovery_handler;
552 pid_t trd_processing_task;
553 struct completion trd_starting;
554 struct completion trd_finishing;
557 struct obd_llog_group {
558 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
559 wait_queue_head_t olg_waitq;
561 struct mutex olg_cat_processing;
564 /* corresponds to one of the obd's */
565 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
568 struct obd_type *obd_type;
569 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
570 int obd_minor; /* device number: lctl dl */
571 struct lu_device *obd_lu_dev;
573 /* common and UUID name of this device */
574 struct obd_uuid obd_uuid;
575 char obd_name[MAX_OBD_NAME];
577 /* bitfield modification is protected by obd_dev_lock */
579 obd_attached:1, /* finished attach */
580 obd_set_up:1, /* finished setup */
581 obd_recovering:1, /* there are recoverable clients */
582 obd_abort_recovery:1, /* recovery expired */
583 obd_version_recov:1, /* obd uses version checking */
584 obd_replayable:1, /* recovery enabled; inform clients */
585 obd_no_transno:1, /* no committed-transno notification */
586 obd_no_recov:1, /* fail instead of retry messages */
587 obd_stopping:1, /* started cleanup */
588 obd_starting:1, /* started setup */
589 obd_force:1, /* cleanup with > 0 obd refcount */
590 obd_fail:1, /* cleanup with failover */
591 obd_no_conn:1, /* deny new connections */
592 obd_inactive:1, /* device active/inactive
593 * (for /proc/status only!!) */
594 obd_no_ir:1, /* no imperative recovery. */
595 obd_process_conf:1, /* device is processing mgs config */
596 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
598 /* use separate field as it is set in interrupt to don't mess with
599 * protection of other bits using _bh lock */
600 unsigned long obd_recovery_expired:1;
601 /* uuid-export hash body */
602 struct cfs_hash *obd_uuid_hash;
603 /* nid-export hash body */
604 struct cfs_hash *obd_nid_hash;
606 struct cfs_hash *obd_nid_stats_hash;
607 /* client_generation-export hash body */
608 struct cfs_hash *obd_gen_hash;
609 struct list_head obd_nid_stats;
610 struct list_head obd_exports;
611 struct list_head obd_unlinked_exports;
612 struct list_head obd_delayed_exports;
613 struct list_head obd_lwp_list;
614 atomic_t obd_refcount;
616 spinlock_t obd_nid_lock;
617 struct ldlm_namespace *obd_namespace;
618 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
619 /* a spinlock is OK for what we do now, may need a semaphore later */
620 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
621 spinlock_t obd_osfs_lock;
622 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
624 __u64 obd_last_committed;
625 struct mutex obd_dev_mutex;
626 struct lvfs_run_ctxt obd_lvfs_ctxt;
627 struct obd_llog_group obd_olg; /* default llog group */
628 struct obd_device *obd_observer;
629 struct rw_semaphore obd_observer_link_sem;
630 struct obd_notify_upcall obd_upcall;
631 struct obd_export *obd_self_export;
632 struct obd_export *obd_lwp_export;
633 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
634 struct list_head obd_exports_timed;
635 time_t obd_eviction_timer; /* for ping evictor */
637 int obd_max_recoverable_clients;
638 atomic_t obd_connected_clients;
639 int obd_stale_clients;
640 /* this lock protects all recovery list_heads, timer and
641 * obd_next_recovery_transno value */
642 spinlock_t obd_recovery_task_lock;
643 __u64 obd_next_recovery_transno;
644 int obd_replayed_requests;
645 int obd_requests_queued_for_recovery;
646 wait_queue_head_t obd_next_transno_waitq;
647 /* protected by obd_recovery_task_lock */
648 struct timer_list obd_recovery_timer;
650 time64_t obd_recovery_start;
651 /* seconds, for lprocfs_status */
652 time64_t obd_recovery_end;
653 time64_t obd_recovery_time_hard;
654 time64_t obd_recovery_timeout;
655 int obd_recovery_ir_factor;
657 /* new recovery stuff from CMD2 */
658 int obd_replayed_locks;
659 atomic_t obd_req_replay_clients;
660 atomic_t obd_lock_replay_clients;
661 struct target_recovery_data obd_recovery_data;
663 /* all lists are protected by obd_recovery_task_lock */
664 struct list_head obd_req_replay_queue;
665 struct list_head obd_lock_replay_queue;
666 struct list_head obd_final_req_queue;
669 #ifdef HAVE_SERVER_SUPPORT
670 struct obd_device_target obt;
671 struct filter_obd filter;
673 struct echo_obd echo;
675 struct client_obd cli;
676 struct echo_client_obd echo_client;
681 /* Fields used by LProcFS */
682 struct lprocfs_stats *obd_stats;
683 unsigned int obd_cntr_base;
685 unsigned int obd_md_cntr_base;
686 struct lprocfs_stats *obd_md_stats;
688 struct proc_dir_entry *obd_proc_entry;
689 struct proc_dir_entry *obd_proc_exports_entry;
690 struct proc_dir_entry *obd_svc_procroot;
691 struct lprocfs_stats *obd_svc_stats;
692 struct lprocfs_vars *obd_vars;
693 atomic_t obd_evict_inprogress;
694 wait_queue_head_t obd_evict_inprogress_waitq;
695 struct list_head obd_evict_list; /* protected with pet_lock */
698 * LDLM pool part. Save last calculated SLV and Limit.
700 rwlock_t obd_pool_lock;
704 int obd_conn_inprogress;
707 * List of outstanding class_incref()'s fo this OBD. For debugging. */
708 struct lu_ref obd_reference;
711 /* get/set_info keys */
712 #define KEY_ASYNC "async"
713 #define KEY_CHANGELOG_CLEAR "changelog_clear"
714 #define KEY_FID2PATH "fid2path"
715 #define KEY_CHECKSUM "checksum"
716 #define KEY_CLEAR_FS "clear_fs"
717 #define KEY_CONN_DATA "conn_data"
718 #define KEY_EVICT_BY_NID "evict_by_nid"
719 #define KEY_FIEMAP "fiemap"
720 #define KEY_FLUSH_CTX "flush_ctx"
721 #define KEY_GRANT_SHRINK "grant_shrink"
722 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
723 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
724 #define KEY_INTERMDS "inter_mds"
725 #define KEY_LAST_ID "last_id"
726 #define KEY_LAST_FID "last_fid"
727 #define KEY_MAX_EASIZE "max_easize"
728 #define KEY_DEFAULT_EASIZE "default_easize"
729 #define KEY_MGSSEC "mgssec"
730 #define KEY_READ_ONLY "read-only"
731 #define KEY_REGISTER_TARGET "register_target"
732 #define KEY_SET_FS "set_fs"
733 #define KEY_TGT_COUNT "tgt_count"
734 /* KEY_SET_INFO in lustre_idl.h */
735 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
737 #define KEY_CACHE_SET "cache_set"
738 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
739 #define KEY_OSP_CONNECTED "osp_connected"
743 static inline int it_to_lock_mode(struct lookup_intent *it)
745 /* CREAT needs to be tested before open (both could be set) */
746 if (it->it_op & IT_CREAT)
748 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
751 else if (it->it_op & IT_READDIR)
753 else if (it->it_op & IT_GETXATTR)
755 else if (it->it_op & IT_SETXATTR)
758 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
763 MF_MDC_CANCEL_FID1 = 1 << 0,
764 MF_MDC_CANCEL_FID2 = 1 << 1,
765 MF_MDC_CANCEL_FID3 = 1 << 2,
766 MF_MDC_CANCEL_FID4 = 1 << 3,
767 MF_GET_MDT_IDX = 1 << 4,
771 CLI_SET_MEA = 1 << 0,
772 CLI_RM_ENTRY = 1 << 1,
775 CLI_MIGRATE = 1 << 4,
779 * GETXATTR is not included as only a couple of fields in the reply body
780 * is filled, but not FID which is needed for common intent handling in
781 * mdc_finish_intent_lock()
783 static inline bool it_has_reply_body(const struct lookup_intent *it)
785 return it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR);
789 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
790 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
791 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
792 struct lu_fid op_fid4; /* to the operation locks. */
793 u32 op_mds; /* what mds server open will go to */
794 struct lustre_handle op_handle;
799 struct lmv_stripe_md *op_mea1;
800 struct lmv_stripe_md *op_mea2;
801 __u32 op_suppgids[2];
808 /* iattr fields and blocks. */
809 struct iattr op_attr;
810 loff_t op_attr_blocks;
811 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
812 __u64 op_valid; /* OBD_MD_* */
814 enum md_op_flags op_flags;
816 /* Various operation flags. */
817 enum mds_op_bias op_bias;
819 /* Used by readdir */
820 unsigned int op_max_pages;
822 /* used to transfer info between the stacks of MD client
823 * see enum op_cli_flags */
824 enum md_cli_flags op_cli_flags;
826 /* File object data version for HSM release, on client */
827 __u64 op_data_version;
828 struct lustre_handle op_lease_handle;
830 /* File security context, for creates. */
831 const char *op_file_secctx_name;
832 void *op_file_secctx;
833 __u32 op_file_secctx_size;
835 /* default stripe offset */
836 __u32 op_default_stripe_offset;
840 int (*md_blocking_ast)(struct ldlm_lock *lock,
841 struct ldlm_lock_desc *desc,
842 void *data, int flag);
845 struct md_enqueue_info;
846 /* metadata stat-ahead */
847 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
848 struct md_enqueue_info *minfo,
851 struct md_enqueue_info {
852 struct md_op_data mi_data;
853 struct lookup_intent mi_it;
854 struct lustre_handle mi_lockh;
855 struct inode *mi_dir;
856 struct ldlm_enqueue_info mi_einfo;
857 md_enqueue_cb_t mi_cb;
862 struct module *o_owner;
863 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
864 void *karg, void __user *uarg);
865 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
866 __u32 keylen, void *key, __u32 *vallen, void *val);
867 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
868 __u32 keylen, void *key,
869 __u32 vallen, void *val,
870 struct ptlrpc_request_set *set);
871 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
872 int (*o_precleanup)(struct obd_device *dev);
873 int (*o_cleanup)(struct obd_device *dev);
874 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
875 int (*o_postrecov)(struct obd_device *dev);
876 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
878 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
879 /* connect to the target device with given connection
880 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
881 * granted by the target, which are guaranteed to be a subset of flags
882 * asked for. If @ocd == NULL, use default parameters. */
883 int (*o_connect)(const struct lu_env *env,
884 struct obd_export **exp, struct obd_device *src,
885 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
887 int (*o_reconnect)(const struct lu_env *env,
888 struct obd_export *exp, struct obd_device *src,
889 struct obd_uuid *cluuid,
890 struct obd_connect_data *ocd,
892 int (*o_disconnect)(struct obd_export *exp);
894 /* Initialize/finalize fids infrastructure. */
895 int (*o_fid_init)(struct obd_device *obd,
896 struct obd_export *exp, enum lu_cli_type type);
897 int (*o_fid_fini)(struct obd_device *obd);
899 /* Allocate new fid according to passed @hint. */
900 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
901 struct lu_fid *fid, struct md_op_data *op_data);
904 * Object with @fid is getting deleted, we may want to do something
907 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
908 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
909 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
910 __u64 max_age, struct ptlrpc_request_set *set);
911 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
913 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
915 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
917 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
919 int (*o_preprw)(const struct lu_env *env, int cmd,
920 struct obd_export *exp, struct obdo *oa, int objcount,
921 struct obd_ioobj *obj, struct niobuf_remote *remote,
922 int *nr_pages, struct niobuf_local *local);
923 int (*o_commitrw)(const struct lu_env *env, int cmd,
924 struct obd_export *exp, struct obdo *oa,
925 int objcount, struct obd_ioobj *obj,
926 struct niobuf_remote *remote, int pages,
927 struct niobuf_local *local, int rc);
928 int (*o_init_export)(struct obd_export *exp);
929 int (*o_destroy_export)(struct obd_export *exp);
931 int (*o_import_event)(struct obd_device *, struct obd_import *,
932 enum obd_import_event);
934 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
935 enum obd_notify_event ev);
937 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
938 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
941 int (*o_quotactl)(struct obd_device *, struct obd_export *,
942 struct obd_quotactl *);
944 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
947 int (*o_pool_new)(struct obd_device *obd, char *poolname);
948 int (*o_pool_del)(struct obd_device *obd, char *poolname);
949 int (*o_pool_add)(struct obd_device *obd, char *poolname,
951 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
953 void (*o_getref)(struct obd_device *obd);
954 void (*o_putref)(struct obd_device *obd);
956 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
957 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
958 * Also, add a wrapper function in include/linux/obd_class.h. */
963 struct mdt_body *body;
964 struct lu_buf layout;
965 struct lmv_stripe_md *lmv;
966 #ifdef CONFIG_FS_POSIX_ACL
967 struct posix_acl *posix_acl;
971 struct md_open_data {
972 struct obd_client_handle *mod_och;
973 struct ptlrpc_request *mod_open_req;
974 struct ptlrpc_request *mod_close_req;
975 atomic_t mod_refcount;
979 struct obd_client_handle {
980 struct lustre_handle och_fh;
981 struct lu_fid och_fid;
982 struct md_open_data *och_mod;
983 struct lustre_handle och_lease_handle; /* open lock for lease */
988 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
990 struct lookup_intent;
994 /* Every operation from MD_STATS_FIRST_OP up to and including
995 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
996 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
997 * Operations after MD_STATS_LAST_OP are excluded from stats.
998 * There are a few reasons for doing this: we prune the 17
999 * counters which will be of minimal use in understanding
1000 * metadata utilization, we save memory by allocating 15
1001 * instead of 32 counters, we save cycles by not counting.
1003 * MD_STATS_FIRST_OP must be the first member of md_ops.
1005 #define MD_STATS_FIRST_OP m_close
1006 int (*m_close)(struct obd_export *, struct md_op_data *,
1007 struct md_open_data *, struct ptlrpc_request **);
1009 int (*m_create)(struct obd_export *, struct md_op_data *,
1010 const void *, size_t, umode_t, uid_t, gid_t,
1011 cfs_cap_t, __u64, struct ptlrpc_request **);
1013 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1014 const union ldlm_policy_data *, struct md_op_data *,
1015 struct lustre_handle *, __u64);
1017 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1018 struct ptlrpc_request **);
1020 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1021 struct lookup_intent *,
1022 struct ptlrpc_request **,
1023 ldlm_blocking_callback, __u64);
1025 int (*m_link)(struct obd_export *, struct md_op_data *,
1026 struct ptlrpc_request **);
1028 int (*m_rename)(struct obd_export *, struct md_op_data *,
1029 const char *, size_t, const char *, size_t,
1030 struct ptlrpc_request **);
1032 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1033 size_t , struct ptlrpc_request **);
1035 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1036 struct ptlrpc_request **);
1038 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1039 struct md_callback *cb_op, __u64 hash_offset,
1040 struct page **ppage);
1042 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1043 struct ptlrpc_request **);
1045 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1046 u64, const char *, const char *, int, int, int, u32,
1047 struct ptlrpc_request **);
1049 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1050 u64, const char *, const char *, int, int, int,
1051 struct ptlrpc_request **);
1053 int (*m_intent_getattr_async)(struct obd_export *,
1054 struct md_enqueue_info *);
1056 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1057 struct lu_fid *, __u64 *bits);
1059 #define MD_STATS_LAST_OP m_revalidate_lock
1061 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1062 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1064 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1065 struct ptlrpc_request **);
1067 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1069 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1070 struct obd_export *, struct obd_export *,
1071 struct lustre_md *);
1073 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1075 int (*m_merge_attr)(struct obd_export *,
1076 const struct lmv_stripe_md *lsm,
1077 struct cl_attr *attr, ldlm_blocking_callback);
1079 int (*m_set_open_replay_data)(struct obd_export *,
1080 struct obd_client_handle *,
1081 struct lookup_intent *);
1083 int (*m_clear_open_replay_data)(struct obd_export *,
1084 struct obd_client_handle *);
1086 int (*m_set_lock_data)(struct obd_export *,
1087 const struct lustre_handle *, void *, __u64 *);
1089 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1090 const struct lu_fid *, enum ldlm_type,
1091 union ldlm_policy_data *, enum ldlm_mode,
1092 struct lustre_handle *);
1094 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1095 union ldlm_policy_data *, enum ldlm_mode,
1096 enum ldlm_cancel_flags flags, void *opaque);
1098 int (*m_get_fid_from_lsm)(struct obd_export *,
1099 const struct lmv_stripe_md *,
1100 const char *name, int namelen,
1101 struct lu_fid *fid);
1102 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1103 const union lmv_mds_md *lmv, size_t lmv_size);
1106 static inline struct md_open_data *obd_mod_alloc(void)
1108 struct md_open_data *mod;
1112 atomic_set(&mod->mod_refcount, 1);
1116 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1117 #define obd_mod_put(mod) \
1119 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1120 if ((mod)->mod_open_req) \
1121 ptlrpc_req_finished((mod)->mod_open_req); \
1122 OBD_FREE_PTR(mod); \
1126 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1127 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1129 /* return 1 if client should be resend request */
1130 static inline int client_should_resend(int resend, struct client_obd *cli)
1132 return atomic_read(&cli->cl_resends) ?
1133 atomic_read(&cli->cl_resends) > resend : 1;
1137 * Return device name for this device
1139 * XXX: lu_device is declared before obd_device, while a pointer pointing
1140 * back to obd_device in lu_device, so this helper function defines here
1141 * instead of in lu_object.h
1143 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1145 return lu_dev->ld_obd->obd_name;
1148 static inline bool filename_is_volatile(const char *name, size_t namelen,
1154 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1157 /* caller does not care of idx */
1161 /* volatile file, the MDT can be set from name */
1162 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1163 /* if no MDT is specified, use std way */
1164 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1166 /* test for no MDT idx case */
1167 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1168 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1172 /* we have an idx, read it */
1173 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1174 *idx = simple_strtoul(start, &end, 16);
1176 * no digit, no trailing :, negative value
1178 if (((*idx == 0) && (end == start)) ||
1179 (*end != ':') || (*idx < 0))
1184 /* bad format of mdt idx, we cannot return an error
1185 * to caller so we use hash algo */
1186 CERROR("Bad volatile file name format: %s\n",
1187 name + LUSTRE_VOLATILE_HDR_LEN);
1191 static inline int cli_brw_size(struct obd_device *obd)
1193 LASSERT(obd != NULL);
1194 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1197 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1198 * of the client should be increased accordingly to avoid sending fragmented
1199 * RPCs over the network when the client runs out of the maximum dirty space
1200 * when so many RPCs are being generated.
1202 static inline void client_adjust_max_dirty(struct client_obd *cli)
1205 if (cli->cl_dirty_max_pages <= 0)
1206 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1209 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1210 cli->cl_max_pages_per_rpc;
1212 if (dirty_max > cli->cl_dirty_max_pages)
1213 cli->cl_dirty_max_pages = dirty_max;
1216 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1217 cli->cl_dirty_max_pages = totalram_pages / 8;
1220 #endif /* __OBD_H */