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, 2017, 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/kobject.h>
37 #include <linux/spinlock.h>
38 #include <linux/sysfs.h>
40 #include <uapi/linux/lustre/lustre_idl.h>
41 #include <lustre_lib.h>
42 #include <libcfs/bitmap.h>
43 #ifdef HAVE_SERVER_SUPPORT
44 # include <lu_target.h>
45 # include <obd_target.h>
48 #include <lustre_export.h>
49 #include <lustre_fid.h>
50 #include <lustre_fld.h>
51 #include <lustre_handles.h>
52 #include <lustre_intent.h>
54 #include <lustre_quota.h>
56 #define MAX_OBD_DEVICES 8192
64 struct lov_oinfo { /* per-stripe data structure */
65 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
66 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
67 int loi_ost_gen; /* generation of this loi_ost_idx */
69 unsigned long loi_kms_valid:1;
70 __u64 loi_kms; /* known minimum size */
71 struct ost_lvb loi_lvb;
72 struct osc_async_rc loi_ar;
75 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
78 oinfo->loi_kms_valid = 1;
84 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
86 /* obd info for a particular level (lov, osc). */
88 /* OBD_STATFS_* flags */
90 /* statfs data specific for every OSC, if needed at all. */
91 struct obd_statfs *oi_osfs;
92 /* An update callback which is called to update some data on upper
93 * level. E.g. it is used for update lsm->lsm_oinfo at every received
94 * request in osc level for enqueue requests. It is also possible to
95 * update some caller data from LOV layer if needed. */
96 obd_enqueue_update_f oi_cb_up;
100 struct list_head typ_chain;
101 struct obd_ops *typ_dt_ops;
102 struct md_ops *typ_md_ops;
103 struct proc_dir_entry *typ_procroot;
104 struct proc_dir_entry *typ_procsym;
105 struct dentry *typ_debugfs_entry;
106 #ifdef HAVE_SERVER_SUPPORT
111 struct lu_device_type *typ_lu;
112 spinlock_t obd_type_lock;
113 struct kobject *typ_kobj;
123 struct timeout_item {
124 enum timeout_event ti_event;
128 struct list_head ti_obd_list;
129 struct list_head ti_chain;
132 #define OBD_MAX_RIF_DEFAULT 8
133 #define OBD_MAX_RIF_MAX 512
134 #define OSC_MAX_RIF_MAX 256
135 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
136 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
137 #define OSC_DEFAULT_RESENDS 10
139 /* possible values for fo_sync_lock_cancel */
141 NEVER_SYNC_ON_CANCEL = 0,
142 BLOCKING_SYNC_ON_CANCEL = 1,
143 ALWAYS_SYNC_ON_CANCEL = 2,
144 NUM_SYNC_ON_CANCEL_STATES
148 * Limit reply buffer size for striping data to one x86_64 page. This
149 * value is chosen to fit the striping data for common use cases while
150 * staying well below the limit at which the buffer must be backed by
151 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
154 #define OBD_MAX_DEFAULT_EA_SIZE 4096
156 enum obd_cl_sem_lock_class {
165 struct rw_semaphore cl_sem;
166 struct obd_uuid cl_target_uuid;
167 struct obd_import *cl_import; /* ptlrpc connection state */
168 size_t cl_conn_count;
170 /* Cache maximum and default values for easize. This is
171 * strictly a performance optimization to minimize calls to
172 * obd_size_diskmd(). The default values are used to calculate the
173 * initial size of a request buffer. The ptlrpc layer will resize the
174 * buffer as needed to accommodate a larger reply from the
175 * server. The default values should be small enough to avoid wasted
176 * memory and excessive use of vmalloc(), yet large enough to avoid
177 * reallocating the buffer in the common use case. */
179 /* Default EA size for striping attributes. It is initialized at
180 * mount-time based on the default stripe width of the filesystem,
181 * then it tracks the largest observed EA size advertised by
182 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
183 __u32 cl_default_mds_easize;
185 /* Maximum possible EA size computed at mount-time based on
186 * the number of OSTs in the filesystem. May be increased at
187 * run-time if a larger observed size is advertised by the MDT. */
188 __u32 cl_max_mds_easize;
190 enum lustre_sec_part cl_sp_me;
191 enum lustre_sec_part cl_sp_to;
192 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
194 /* the grant values are protected by loi_list_lock below */
195 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
196 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
197 unsigned long cl_dirty_transit; /* dirty synchronous */
198 unsigned long cl_avail_grant; /* bytes of credit for ost */
199 unsigned long cl_lost_grant; /* lost credits (trunc) */
200 /* grant consumed for dirty pages */
201 unsigned long cl_dirty_grant;
203 /* since we allocate grant by blocks, we don't know how many grant will
204 * be used to add a page into cache. As a solution, we reserve maximum
205 * grant before trying to dirty a page and unreserve the rest.
206 * See osc_{reserve|unreserve}_grant for details. */
207 long cl_reserved_grant;
208 struct list_head cl_cache_waiters; /* waiting for cache/grant */
209 time64_t cl_next_shrink_grant; /* seconds */
210 struct list_head cl_grant_shrink_list; /* Timeout event list */
211 time64_t cl_grant_shrink_interval; /* seconds */
213 /* A chunk is an optimal size used by osc_extent to determine
214 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
216 /* extent insertion metadata overhead to be accounted in grant,
218 unsigned int cl_grant_extent_tax;
219 /* maximum extent size, in number of pages */
220 unsigned int cl_max_extent_pages;
222 /* keep track of objects that have lois that contain pages which
223 * have been queued for async brw. this lock also protects the
224 * lists of osc_client_pages that hang off of the loi */
226 * ->cl_loi_list_lock protects consistency of
227 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
228 * ->ap_completion() call-backs are executed under this lock. As we
229 * cannot guarantee that these call-backs never block on all platforms
230 * (as a matter of fact they do block on Mac OS X), type of
231 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
232 * and blocking mutex on Mac OS X. (Alternative is to make this lock
233 * blocking everywhere, but we don't want to slow down fast-path of
234 * our main platform.)
236 * NB by Jinshan: though field names are still _loi_, but actually
237 * osc_object{}s are in the list.
239 spinlock_t cl_loi_list_lock;
240 struct list_head cl_loi_ready_list;
241 struct list_head cl_loi_hp_ready_list;
242 struct list_head cl_loi_write_list;
243 struct list_head cl_loi_read_list;
244 __u32 cl_r_in_flight;
245 __u32 cl_w_in_flight;
246 /* just a sum of the loi/lop pending numbers to be exported by /proc */
247 atomic_t cl_pending_w_pages;
248 atomic_t cl_pending_r_pages;
249 __u32 cl_max_pages_per_rpc;
250 __u32 cl_max_rpcs_in_flight;
251 __u32 cl_short_io_bytes;
252 struct obd_histogram cl_read_rpc_hist;
253 struct obd_histogram cl_write_rpc_hist;
254 struct obd_histogram cl_read_page_hist;
255 struct obd_histogram cl_write_page_hist;
256 struct obd_histogram cl_read_offset_hist;
257 struct obd_histogram cl_write_offset_hist;
259 /** LRU for osc caching pages */
260 struct cl_client_cache *cl_cache;
261 /** member of cl_cache->ccc_lru */
262 struct list_head cl_lru_osc;
263 /** # of available LRU slots left in the per-OSC cache.
264 * Available LRU slots are shared by all OSCs of the same file system,
265 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
266 atomic_long_t *cl_lru_left;
267 /** # of busy LRU pages. A page is considered busy if it's in writeback
268 * queue, or in transfer. Busy pages can't be discarded so they are not
270 atomic_long_t cl_lru_busy;
271 /** # of LRU pages in the cache for this client_obd */
272 atomic_long_t cl_lru_in_list;
273 /** # of threads are shrinking LRU cache. To avoid contention, it's not
274 * allowed to have multiple threads shrinking LRU cache. */
275 atomic_t cl_lru_shrinkers;
276 /** The time when this LRU cache was last used. */
277 time64_t cl_lru_last_used;
278 /** stats: how many reclaims have happened for this client_obd.
279 * reclaim and shrink - shrink is async, voluntarily rebalancing;
280 * reclaim is sync, initiated by IO thread when the LRU slots are
282 __u64 cl_lru_reclaim;
283 /** List of LRU pages for this client_obd */
284 struct list_head cl_lru_list;
285 /** Lock for LRU page list */
286 spinlock_t cl_lru_list_lock;
287 /** # of unstable pages in this client_obd.
288 * An unstable page is a page state that WRITE RPC has finished but
289 * the transaction has NOT yet committed. */
290 atomic_long_t cl_unstable_count;
291 /** Link to osc_shrinker_list */
292 struct list_head cl_shrink_list;
294 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
295 atomic_t cl_destroy_in_flight;
296 wait_queue_head_t cl_destroy_waitq;
298 struct mdc_rpc_lock *cl_rpc_lock;
300 /* modify rpcs in flight
301 * currently used for metadata only */
302 spinlock_t cl_mod_rpcs_lock;
303 __u16 cl_max_mod_rpcs_in_flight;
304 __u16 cl_mod_rpcs_in_flight;
305 __u16 cl_close_rpcs_in_flight;
306 wait_queue_head_t cl_mod_rpcs_waitq;
307 unsigned long *cl_mod_tag_bitmap;
308 struct obd_histogram cl_mod_rpcs_hist;
311 struct mutex cl_mgc_mutex;
312 struct local_oid_storage *cl_mgc_los;
313 struct dt_object *cl_mgc_configs_dir;
314 struct obd_export *cl_mgc_mgsexp;
315 atomic_t cl_mgc_refcount;
316 /* in-flight control list and total RPCs counter */
317 struct list_head cl_flight_waiters;
318 __u32 cl_rpcs_in_flight;
320 /* checksumming for data sent over the network */
321 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
322 cl_checksum_dump:1; /* same */
323 /* supported checksum types that are worked out at connect time */
324 __u32 cl_supp_cksum_types;
325 /* checksum algorithm to be used */
326 enum cksum_types cl_cksum_type;
328 /* also protected by the poorly named _loi_list_lock lock above */
329 struct osc_async_rc cl_ar;
331 /* sequence manager */
332 struct lu_client_seq *cl_seq;
333 struct rw_semaphore cl_seq_rwsem;
335 atomic_t cl_resends; /* resend count */
337 /* ptlrpc work for writeback in ptlrpcd context */
338 void *cl_writeback_work;
340 /* hash tables for osc_quota_info */
341 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
342 /* Links to the global list of registered changelog devices */
343 struct list_head cl_chg_dev_linkage;
345 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
352 struct echo_client_obd {
353 struct obd_export *ec_exp; /* the local connection to osc/lov */
355 struct list_head ec_objects;
356 struct list_head ec_locks;
360 /* Generic subset of OSTs */
362 __u32 *op_array; /* array of index of
364 unsigned int op_count; /* number of OSTs in the array */
365 unsigned int op_size; /* allocated size of lp_array */
366 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
369 /* allow statfs data caching for 1 second */
370 #define OBD_STATFS_CACHE_SECONDS 1
372 struct lov_tgt_desc {
373 struct list_head ltd_kill;
374 struct obd_uuid ltd_uuid;
375 struct obd_device *ltd_obd;
376 struct obd_export *ltd_exp;
378 __u32 ltd_index; /* index in lov_obd->tgts */
379 unsigned long ltd_active:1,/* is this target up for requests */
380 ltd_activate:1,/* should target be activated */
381 ltd_reap:1; /* should this target be deleted */
384 struct lov_md_tgt_desc {
385 struct obd_device *lmtd_mdc;
390 struct lov_desc desc;
391 struct lov_tgt_desc **lov_tgts; /* sparse array */
392 struct ost_pool lov_packed; /* all OSTs in a packed
394 struct mutex lov_lock;
395 struct obd_connect_data lov_ocd;
396 struct proc_dir_entry *targets_proc_entry;
397 atomic_t lov_refcount;
398 __u32 lov_death_row; /* tgts scheduled to be deleted */
399 __u32 lov_tgt_size; /* size of tgts array */
402 struct cfs_hash *lov_pools_hash_body; /* used for key access */
403 struct list_head lov_pool_list; /* used for sequential access */
404 struct proc_dir_entry *lov_pool_proc_entry;
405 enum lustre_sec_part lov_sp_me;
407 /* Cached LRU and unstable data from upper layer */
408 struct cl_client_cache *lov_cache;
410 struct rw_semaphore lov_notify_lock;
411 /* Data-on-MDT: MDC array */
412 struct lov_md_tgt_desc *lov_mdc_tgts;
414 struct kobject *lov_tgts_kobj;
417 struct lmv_tgt_desc {
418 struct obd_uuid ltd_uuid;
419 struct obd_device *ltd_obd;
420 struct obd_export *ltd_exp;
422 struct mutex ltd_fid_mutex;
423 unsigned long ltd_active:1; /* target up for requests */
427 struct lu_client_fld lmv_fld;
429 struct lmv_desc desc;
431 struct mutex lmv_init_mutex;
436 __u32 tgts_size; /* size of tgts array */
437 struct lmv_tgt_desc **tgts;
439 struct obd_connect_data conn_data;
440 struct kobject *lmv_tgts_kobj;
443 struct niobuf_local {
444 __u64 lnb_file_offset;
445 __u32 lnb_page_offset;
449 struct page *lnb_page;
453 struct tgt_thread_big_cache {
454 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
457 #define LUSTRE_FLD_NAME "fld"
458 #define LUSTRE_SEQ_NAME "seq"
460 #define LUSTRE_MDD_NAME "mdd"
461 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
462 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
463 #define LUSTRE_VVP_NAME "vvp"
464 #define LUSTRE_LMV_NAME "lmv"
465 #define LUSTRE_SLP_NAME "slp"
466 #define LUSTRE_LOD_NAME "lod"
467 #define LUSTRE_OSP_NAME "osp"
468 #define LUSTRE_LWP_NAME "lwp"
470 /* obd device type names */
471 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
472 #define LUSTRE_MDS_NAME "mds"
473 #define LUSTRE_MDT_NAME "mdt"
474 #define LUSTRE_MDC_NAME "mdc"
475 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
476 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
477 #define LUSTRE_OSC_NAME "osc"
478 #define LUSTRE_LOV_NAME "lov"
479 #define LUSTRE_MGS_NAME "mgs"
480 #define LUSTRE_MGC_NAME "mgc"
482 #define LUSTRE_ECHO_NAME "obdecho"
483 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
484 #define LUSTRE_QMT_NAME "qmt"
486 /* Constant obd names (post-rename) */
487 #define LUSTRE_MDS_OBDNAME "MDS"
488 #define LUSTRE_OSS_OBDNAME "OSS"
489 #define LUSTRE_MGS_OBDNAME "MGS"
490 #define LUSTRE_MGC_OBDNAME "MGC"
492 static inline int is_lwp_on_mdt(char *name)
496 ptr = strrchr(name, '-');
498 CERROR("%s is not a obdname\n", name);
502 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
504 if (strncmp(ptr + 1, "MDT", 3) != 0)
507 while (*(--ptr) != '-' && ptr != name);
512 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
518 static inline int is_lwp_on_ost(char *name)
522 ptr = strrchr(name, '-');
524 CERROR("%s is not a obdname\n", name);
528 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
530 if (strncmp(ptr + 1, "OST", 3) != 0)
533 while (*(--ptr) != '-' && ptr != name);
538 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
545 * Events signalled through obd_notify() upcall-chain.
547 enum obd_notify_event {
548 /* Device connect start */
550 /* Device activated */
552 /* Device deactivated */
554 /* Connect data for import were changed */
556 /* Administratively deactivate/activate event */
557 OBD_NOTIFY_DEACTIVATE,
562 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
563 * being main example).
565 struct obd_notify_upcall {
566 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
567 enum obd_notify_event ev, void *owner);
568 /* Opaque datum supplied by upper layer listener */
572 struct target_recovery_data {
573 svc_handler_t trd_recovery_handler;
574 pid_t trd_processing_task;
575 struct completion trd_starting;
576 struct completion trd_finishing;
579 struct obd_llog_group {
580 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
581 wait_queue_head_t olg_waitq;
583 struct mutex olg_cat_processing;
586 /* corresponds to one of the obd's */
587 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
590 struct obd_type *obd_type;
591 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
592 int obd_minor; /* device number: lctl dl */
593 struct lu_device *obd_lu_dev;
595 /* common and UUID name of this device */
596 struct obd_uuid obd_uuid;
597 char obd_name[MAX_OBD_NAME];
599 /* bitfield modification is protected by obd_dev_lock */
601 obd_attached:1, /* finished attach */
602 obd_set_up:1, /* finished setup */
603 obd_recovering:1, /* there are recoverable clients */
604 obd_abort_recovery:1, /* recovery expired */
605 obd_version_recov:1, /* obd uses version checking */
606 obd_replayable:1, /* recovery enabled; inform clients */
607 obd_no_transno:1, /* no committed-transno notification */
608 obd_no_recov:1, /* fail instead of retry messages */
609 obd_stopping:1, /* started cleanup */
610 obd_starting:1, /* started setup */
611 obd_force:1, /* cleanup with > 0 obd refcount */
612 obd_fail:1, /* cleanup with failover */
613 obd_no_conn:1, /* deny new connections */
614 obd_inactive:1, /* device active/inactive
615 * (for /proc/status only!!) */
616 obd_no_ir:1, /* no imperative recovery. */
617 obd_process_conf:1, /* device is processing mgs config */
618 obd_checksum_dump:1; /* dump pages upon cksum error */
620 /* use separate field as it is set in interrupt to don't mess with
621 * protection of other bits using _bh lock */
622 unsigned long obd_recovery_expired:1;
623 /* uuid-export hash body */
624 struct cfs_hash *obd_uuid_hash;
625 /* nid-export hash body */
626 struct cfs_hash *obd_nid_hash;
628 struct cfs_hash *obd_nid_stats_hash;
629 /* client_generation-export hash body */
630 struct cfs_hash *obd_gen_hash;
631 struct list_head obd_nid_stats;
632 struct list_head obd_exports;
633 struct list_head obd_unlinked_exports;
634 struct list_head obd_delayed_exports;
635 struct list_head obd_lwp_list;
636 atomic_t obd_refcount;
638 spinlock_t obd_nid_lock;
639 struct ldlm_namespace *obd_namespace;
640 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
641 /* a spinlock is OK for what we do now, may need a semaphore later */
642 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
643 spinlock_t obd_osfs_lock;
644 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
645 time64_t obd_osfs_age;
646 __u64 obd_last_committed;
647 struct mutex obd_dev_mutex;
648 struct lvfs_run_ctxt obd_lvfs_ctxt;
649 struct obd_llog_group obd_olg; /* default llog group */
650 struct obd_device *obd_observer;
651 struct rw_semaphore obd_observer_link_sem;
652 struct obd_notify_upcall obd_upcall;
653 struct obd_export *obd_self_export;
654 struct obd_export *obd_lwp_export;
655 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
656 struct list_head obd_exports_timed;
657 time64_t obd_eviction_timer; /* for ping evictor */
659 int obd_max_recoverable_clients;
660 atomic_t obd_connected_clients;
661 int obd_stale_clients;
662 /* this lock protects all recovery list_heads, timer and
663 * obd_next_recovery_transno value */
664 spinlock_t obd_recovery_task_lock;
665 __u64 obd_next_recovery_transno;
666 int obd_replayed_requests;
667 int obd_requests_queued_for_recovery;
668 wait_queue_head_t obd_next_transno_waitq;
669 /* protected by obd_recovery_task_lock */
670 struct timer_list obd_recovery_timer;
672 time64_t obd_recovery_start;
673 /* seconds, for lprocfs_status */
674 time64_t obd_recovery_end;
675 time64_t obd_recovery_time_hard;
676 time64_t obd_recovery_timeout;
677 int obd_recovery_ir_factor;
679 /* new recovery stuff from CMD2 */
680 int obd_replayed_locks;
681 atomic_t obd_req_replay_clients;
682 atomic_t obd_lock_replay_clients;
683 struct target_recovery_data obd_recovery_data;
685 /* all lists are protected by obd_recovery_task_lock */
686 struct list_head obd_req_replay_queue;
687 struct list_head obd_lock_replay_queue;
688 struct list_head obd_final_req_queue;
691 #ifdef HAVE_SERVER_SUPPORT
692 struct obd_device_target obt;
693 struct filter_obd filter;
695 struct echo_obd echo;
697 struct client_obd cli;
698 struct echo_client_obd echo_client;
703 /* Fields used by LProcFS */
704 struct lprocfs_stats *obd_stats;
706 unsigned int obd_md_cntr_base;
707 struct lprocfs_stats *obd_md_stats;
709 struct dentry *obd_debugfs_entry;
710 struct proc_dir_entry *obd_proc_entry;
711 struct proc_dir_entry *obd_proc_exports_entry;
712 struct dentry *obd_svc_debugfs_entry;
713 struct lprocfs_stats *obd_svc_stats;
714 struct attribute_group obd_attrs_group;
715 struct attribute **obd_attrs;
716 struct lprocfs_vars *obd_vars;
717 atomic_t obd_evict_inprogress;
718 wait_queue_head_t obd_evict_inprogress_waitq;
719 struct list_head obd_evict_list; /* protected with pet_lock */
722 * LDLM pool part. Save last calculated SLV and Limit.
724 rwlock_t obd_pool_lock;
728 int obd_conn_inprogress;
731 * List of outstanding class_incref()'s fo this OBD. For debugging. */
732 struct lu_ref obd_reference;
734 struct kset obd_kset; /* sysfs object collection */
735 struct kobj_type obd_ktype;
736 struct completion obd_kobj_unregister;
739 /* get/set_info keys */
740 #define KEY_ASYNC "async"
741 #define KEY_CHANGELOG_CLEAR "changelog_clear"
742 #define KEY_FID2PATH "fid2path"
743 #define KEY_CHECKSUM "checksum"
744 #define KEY_CLEAR_FS "clear_fs"
745 #define KEY_CONN_DATA "conn_data"
746 #define KEY_EVICT_BY_NID "evict_by_nid"
747 #define KEY_FIEMAP "fiemap"
748 #define KEY_FLUSH_CTX "flush_ctx"
749 #define KEY_GRANT_SHRINK "grant_shrink"
750 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
751 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
752 #define KEY_INTERMDS "inter_mds"
753 #define KEY_LAST_ID "last_id"
754 #define KEY_LAST_FID "last_fid"
755 #define KEY_MAX_EASIZE "max_easize"
756 #define KEY_DEFAULT_EASIZE "default_easize"
757 #define KEY_MGSSEC "mgssec"
758 #define KEY_READ_ONLY "read-only"
759 #define KEY_REGISTER_TARGET "register_target"
760 #define KEY_SET_FS "set_fs"
761 #define KEY_TGT_COUNT "tgt_count"
762 /* KEY_SET_INFO in lustre_idl.h */
763 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
765 #define KEY_CACHE_SET "cache_set"
766 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
767 #define KEY_OSP_CONNECTED "osp_connected"
771 static inline int it_to_lock_mode(struct lookup_intent *it)
773 /* CREAT needs to be tested before open (both could be set) */
774 if (it->it_op & IT_CREAT)
776 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
778 else if (it->it_op & IT_LAYOUT)
779 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
780 else if (it->it_op & IT_READDIR)
782 else if (it->it_op & IT_GETXATTR)
784 else if (it->it_op & IT_SETXATTR)
787 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
792 MF_MDC_CANCEL_FID1 = 1 << 0,
793 MF_MDC_CANCEL_FID2 = 1 << 1,
794 MF_MDC_CANCEL_FID3 = 1 << 2,
795 MF_MDC_CANCEL_FID4 = 1 << 3,
796 MF_GET_MDT_IDX = 1 << 4,
800 CLI_SET_MEA = 1 << 0,
801 CLI_RM_ENTRY = 1 << 1,
804 CLI_MIGRATE = 1 << 4,
808 * GETXATTR is not included as only a couple of fields in the reply body
809 * is filled, but not FID which is needed for common intent handling in
810 * mdc_finish_intent_lock()
812 static inline bool it_has_reply_body(const struct lookup_intent *it)
814 return it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR);
818 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
819 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
820 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
821 struct lu_fid op_fid4; /* to the operation locks. */
822 u32 op_mds; /* what mds server open will go to */
824 struct lustre_handle op_handle;
828 struct lmv_stripe_md *op_mea1;
829 struct lmv_stripe_md *op_mea2;
830 __u32 op_suppgids[2];
837 /* iattr fields and blocks. */
838 struct iattr op_attr;
839 loff_t op_attr_blocks;
840 __u64 op_valid; /* OBD_MD_* */
841 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
843 enum md_op_flags op_flags;
845 /* Various operation flags. */
846 enum mds_op_bias op_bias;
848 /* used to transfer info between the stacks of MD client
849 * see enum op_cli_flags */
850 enum md_cli_flags op_cli_flags;
852 /* File object data version for HSM release, on client */
853 __u64 op_data_version;
854 struct lustre_handle op_lease_handle;
856 /* File security context, for creates. */
857 const char *op_file_secctx_name;
858 void *op_file_secctx;
859 __u32 op_file_secctx_size;
861 /* default stripe offset */
862 __u32 op_default_stripe_offset;
866 /* Used by readdir */
867 unsigned int op_max_pages;
873 int (*md_blocking_ast)(struct ldlm_lock *lock,
874 struct ldlm_lock_desc *desc,
875 void *data, int flag);
878 struct md_enqueue_info;
879 /* metadata stat-ahead */
880 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
881 struct md_enqueue_info *minfo,
884 struct md_enqueue_info {
885 struct md_op_data mi_data;
886 struct lookup_intent mi_it;
887 struct lustre_handle mi_lockh;
888 struct inode *mi_dir;
889 struct ldlm_enqueue_info mi_einfo;
890 md_enqueue_cb_t mi_cb;
895 struct module *o_owner;
896 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
897 void *karg, void __user *uarg);
898 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
899 __u32 keylen, void *key, __u32 *vallen, void *val);
900 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
901 __u32 keylen, void *key,
902 __u32 vallen, void *val,
903 struct ptlrpc_request_set *set);
904 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
905 int (*o_precleanup)(struct obd_device *dev);
906 int (*o_cleanup)(struct obd_device *dev);
907 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
908 int (*o_postrecov)(struct obd_device *dev);
909 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
911 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
912 /* connect to the target device with given connection
913 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
914 * granted by the target, which are guaranteed to be a subset of flags
915 * asked for. If @ocd == NULL, use default parameters. */
916 int (*o_connect)(const struct lu_env *env,
917 struct obd_export **exp, struct obd_device *src,
918 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
920 int (*o_reconnect)(const struct lu_env *env,
921 struct obd_export *exp, struct obd_device *src,
922 struct obd_uuid *cluuid,
923 struct obd_connect_data *ocd,
925 int (*o_disconnect)(struct obd_export *exp);
927 /* Initialize/finalize fids infrastructure. */
928 int (*o_fid_init)(struct obd_device *obd,
929 struct obd_export *exp, enum lu_cli_type type);
930 int (*o_fid_fini)(struct obd_device *obd);
932 /* Allocate new fid according to passed @hint. */
933 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
934 struct lu_fid *fid, struct md_op_data *op_data);
937 * Object with @fid is getting deleted, we may want to do something
940 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
941 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
942 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
943 time64_t max_age, struct ptlrpc_request_set *set);
944 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
946 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
948 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
950 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
952 int (*o_preprw)(const struct lu_env *env, int cmd,
953 struct obd_export *exp, struct obdo *oa, int objcount,
954 struct obd_ioobj *obj, struct niobuf_remote *remote,
955 int *nr_pages, struct niobuf_local *local);
956 int (*o_commitrw)(const struct lu_env *env, int cmd,
957 struct obd_export *exp, struct obdo *oa,
958 int objcount, struct obd_ioobj *obj,
959 struct niobuf_remote *remote, int pages,
960 struct niobuf_local *local, int rc);
961 int (*o_init_export)(struct obd_export *exp);
962 int (*o_destroy_export)(struct obd_export *exp);
964 int (*o_import_event)(struct obd_device *, struct obd_import *,
965 enum obd_import_event);
967 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
968 enum obd_notify_event ev);
970 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
971 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
974 int (*o_quotactl)(struct obd_device *, struct obd_export *,
975 struct obd_quotactl *);
977 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
980 int (*o_pool_new)(struct obd_device *obd, char *poolname);
981 int (*o_pool_del)(struct obd_device *obd, char *poolname);
982 int (*o_pool_add)(struct obd_device *obd, char *poolname,
984 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
990 struct mdt_body *body;
991 struct lu_buf layout;
992 struct lmv_stripe_md *lmv;
993 #ifdef CONFIG_FS_POSIX_ACL
994 struct posix_acl *posix_acl;
998 struct md_open_data {
999 struct obd_client_handle *mod_och;
1000 struct ptlrpc_request *mod_open_req;
1001 struct ptlrpc_request *mod_close_req;
1002 atomic_t mod_refcount;
1006 struct obd_client_handle {
1007 struct lustre_handle och_fh;
1008 struct lu_fid och_fid;
1009 struct md_open_data *och_mod;
1010 struct lustre_handle och_lease_handle; /* open lock for lease */
1015 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1017 struct lookup_intent;
1021 /* Every operation from MD_STATS_FIRST_OP up to and including
1022 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1023 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1024 * Operations after MD_STATS_LAST_OP are excluded from stats.
1025 * There are a few reasons for doing this: we prune the 17
1026 * counters which will be of minimal use in understanding
1027 * metadata utilization, we save memory by allocating 15
1028 * instead of 32 counters, we save cycles by not counting.
1030 * MD_STATS_FIRST_OP must be the first member of md_ops.
1032 #define MD_STATS_FIRST_OP m_close
1033 int (*m_close)(struct obd_export *, struct md_op_data *,
1034 struct md_open_data *, struct ptlrpc_request **);
1036 int (*m_create)(struct obd_export *, struct md_op_data *,
1037 const void *, size_t, umode_t, uid_t, gid_t,
1038 cfs_cap_t, __u64, struct ptlrpc_request **);
1040 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1041 const union ldlm_policy_data *, struct md_op_data *,
1042 struct lustre_handle *, __u64);
1044 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1045 struct ptlrpc_request **);
1047 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1048 struct lookup_intent *,
1049 struct ptlrpc_request **,
1050 ldlm_blocking_callback, __u64);
1052 int (*m_link)(struct obd_export *, struct md_op_data *,
1053 struct ptlrpc_request **);
1055 int (*m_rename)(struct obd_export *, struct md_op_data *,
1056 const char *, size_t, const char *, size_t,
1057 struct ptlrpc_request **);
1059 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1060 size_t , struct ptlrpc_request **);
1062 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1063 struct ptlrpc_request **);
1065 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1066 struct md_callback *cb_op, __u64 hash_offset,
1067 struct page **ppage);
1069 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1070 struct ptlrpc_request **);
1072 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1073 u64, const char *, const void *, size_t, unsigned int,
1074 u32, struct ptlrpc_request **);
1076 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1077 u64, const char *, size_t, struct ptlrpc_request **);
1079 int (*m_intent_getattr_async)(struct obd_export *,
1080 struct md_enqueue_info *);
1082 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1083 struct lu_fid *, __u64 *bits);
1085 #define MD_STATS_LAST_OP m_revalidate_lock
1087 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1089 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1090 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1092 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1093 struct ptlrpc_request **);
1095 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1097 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1098 struct obd_export *, struct obd_export *,
1099 struct lustre_md *);
1101 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1103 int (*m_merge_attr)(struct obd_export *,
1104 const struct lmv_stripe_md *lsm,
1105 struct cl_attr *attr, ldlm_blocking_callback);
1107 int (*m_set_open_replay_data)(struct obd_export *,
1108 struct obd_client_handle *,
1109 struct lookup_intent *);
1111 int (*m_clear_open_replay_data)(struct obd_export *,
1112 struct obd_client_handle *);
1114 int (*m_set_lock_data)(struct obd_export *,
1115 const struct lustre_handle *, void *, __u64 *);
1117 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1118 const struct lu_fid *, enum ldlm_type,
1119 union ldlm_policy_data *, enum ldlm_mode,
1120 struct lustre_handle *);
1122 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1123 union ldlm_policy_data *, enum ldlm_mode,
1124 enum ldlm_cancel_flags flags, void *opaque);
1126 int (*m_get_fid_from_lsm)(struct obd_export *,
1127 const struct lmv_stripe_md *,
1128 const char *name, int namelen,
1129 struct lu_fid *fid);
1130 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1131 const union lmv_mds_md *lmv, size_t lmv_size);
1134 static inline struct md_open_data *obd_mod_alloc(void)
1136 struct md_open_data *mod;
1140 atomic_set(&mod->mod_refcount, 1);
1144 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1145 #define obd_mod_put(mod) \
1147 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1148 if ((mod)->mod_open_req) \
1149 ptlrpc_req_finished((mod)->mod_open_req); \
1150 OBD_FREE_PTR(mod); \
1154 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1155 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1156 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1158 /* return 1 if client should be resend request */
1159 static inline int client_should_resend(int resend, struct client_obd *cli)
1161 return atomic_read(&cli->cl_resends) ?
1162 atomic_read(&cli->cl_resends) > resend : 1;
1166 * Return device name for this device
1168 * XXX: lu_device is declared before obd_device, while a pointer pointing
1169 * back to obd_device in lu_device, so this helper function defines here
1170 * instead of in lu_object.h
1172 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1174 return lu_dev->ld_obd->obd_name;
1177 static inline bool filename_is_volatile(const char *name, size_t namelen,
1183 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1186 /* caller does not care of idx */
1190 /* volatile file, the MDT can be set from name */
1191 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1192 /* if no MDT is specified, use std way */
1193 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1195 /* test for no MDT idx case */
1196 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1197 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1201 /* we have an idx, read it */
1202 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1203 *idx = simple_strtoul(start, &end, 16);
1205 * no digit, no trailing :, negative value
1207 if (((*idx == 0) && (end == start)) ||
1208 (*end != ':') || (*idx < 0))
1213 /* bad format of mdt idx, we cannot return an error
1214 * to caller so we use hash algo */
1215 CERROR("Bad volatile file name format: %s\n",
1216 name + LUSTRE_VOLATILE_HDR_LEN);
1220 static inline int cli_brw_size(struct obd_device *obd)
1222 LASSERT(obd != NULL);
1223 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1227 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1228 * of the client should be increased accordingly to avoid sending fragmented
1229 * RPCs over the network when the client runs out of the maximum dirty space
1230 * when so many RPCs are being generated.
1232 static inline void client_adjust_max_dirty(struct client_obd *cli)
1235 if (cli->cl_dirty_max_pages <= 0) {
1236 cli->cl_dirty_max_pages =
1237 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1239 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1240 cli->cl_max_pages_per_rpc;
1242 if (dirty_max > cli->cl_dirty_max_pages)
1243 cli->cl_dirty_max_pages = dirty_max;
1246 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1247 cli->cl_dirty_max_pages = totalram_pages / 8;
1249 /* This value is exported to userspace through the max_dirty_mb
1250 * parameter. So we round up the number of pages to make it a round
1252 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1253 1 << (20 - PAGE_SHIFT));
1256 #endif /* __OBD_H */