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 /* Data-on-MDT specific value to set larger reply buffer for possible
191 * data read along with open/stat requests. By default it tries to use
192 * unused space in reply buffer.
193 * This value is used to ensure that reply buffer has at least as
194 * much free space as value indicates. That free space is gained from
195 * LOV EA buffer which is small for DoM files and on big systems can
196 * provide up to 32KB of extra space in reply buffer.
197 * Default value is 8K now.
199 __u32 cl_dom_min_inline_repsize;
201 enum lustre_sec_part cl_sp_me;
202 enum lustre_sec_part cl_sp_to;
203 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
205 /* the grant values are protected by loi_list_lock below */
206 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
207 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
208 unsigned long cl_dirty_transit; /* dirty synchronous */
209 unsigned long cl_avail_grant; /* bytes of credit for ost */
210 unsigned long cl_lost_grant; /* lost credits (trunc) */
211 /* grant consumed for dirty pages */
212 unsigned long cl_dirty_grant;
214 /* since we allocate grant by blocks, we don't know how many grant will
215 * be used to add a page into cache. As a solution, we reserve maximum
216 * grant before trying to dirty a page and unreserve the rest.
217 * See osc_{reserve|unreserve}_grant for details. */
218 long cl_reserved_grant;
219 struct list_head cl_cache_waiters; /* waiting for cache/grant */
220 time64_t cl_next_shrink_grant; /* seconds */
221 struct list_head cl_grant_chain;
222 time64_t cl_grant_shrink_interval; /* seconds */
224 /* A chunk is an optimal size used by osc_extent to determine
225 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
227 /* extent insertion metadata overhead to be accounted in grant,
229 unsigned int cl_grant_extent_tax;
230 /* maximum extent size, in number of pages */
231 unsigned int cl_max_extent_pages;
233 /* keep track of objects that have lois that contain pages which
234 * have been queued for async brw. this lock also protects the
235 * lists of osc_client_pages that hang off of the loi */
237 * ->cl_loi_list_lock protects consistency of
238 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
239 * ->ap_completion() call-backs are executed under this lock. As we
240 * cannot guarantee that these call-backs never block on all platforms
241 * (as a matter of fact they do block on Mac OS X), type of
242 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
243 * and blocking mutex on Mac OS X. (Alternative is to make this lock
244 * blocking everywhere, but we don't want to slow down fast-path of
245 * our main platform.)
247 * NB by Jinshan: though field names are still _loi_, but actually
248 * osc_object{}s are in the list.
250 spinlock_t cl_loi_list_lock;
251 struct list_head cl_loi_ready_list;
252 struct list_head cl_loi_hp_ready_list;
253 struct list_head cl_loi_write_list;
254 struct list_head cl_loi_read_list;
255 __u32 cl_r_in_flight;
256 __u32 cl_w_in_flight;
257 /* just a sum of the loi/lop pending numbers to be exported by /proc */
258 atomic_t cl_pending_w_pages;
259 atomic_t cl_pending_r_pages;
260 __u32 cl_max_pages_per_rpc;
261 __u32 cl_max_rpcs_in_flight;
262 __u32 cl_short_io_bytes;
263 struct obd_histogram cl_read_rpc_hist;
264 struct obd_histogram cl_write_rpc_hist;
265 struct obd_histogram cl_read_page_hist;
266 struct obd_histogram cl_write_page_hist;
267 struct obd_histogram cl_read_offset_hist;
268 struct obd_histogram cl_write_offset_hist;
270 /** LRU for osc caching pages */
271 struct cl_client_cache *cl_cache;
272 /** member of cl_cache->ccc_lru */
273 struct list_head cl_lru_osc;
274 /** # of available LRU slots left in the per-OSC cache.
275 * Available LRU slots are shared by all OSCs of the same file system,
276 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
277 atomic_long_t *cl_lru_left;
278 /** # of busy LRU pages. A page is considered busy if it's in writeback
279 * queue, or in transfer. Busy pages can't be discarded so they are not
281 atomic_long_t cl_lru_busy;
282 /** # of LRU pages in the cache for this client_obd */
283 atomic_long_t cl_lru_in_list;
284 /** # of threads are shrinking LRU cache. To avoid contention, it's not
285 * allowed to have multiple threads shrinking LRU cache. */
286 atomic_t cl_lru_shrinkers;
287 /** The time when this LRU cache was last used. */
288 time64_t cl_lru_last_used;
289 /** stats: how many reclaims have happened for this client_obd.
290 * reclaim and shrink - shrink is async, voluntarily rebalancing;
291 * reclaim is sync, initiated by IO thread when the LRU slots are
293 __u64 cl_lru_reclaim;
294 /** List of LRU pages for this client_obd */
295 struct list_head cl_lru_list;
296 /** Lock for LRU page list */
297 spinlock_t cl_lru_list_lock;
298 /** # of unstable pages in this client_obd.
299 * An unstable page is a page state that WRITE RPC has finished but
300 * the transaction has NOT yet committed. */
301 atomic_long_t cl_unstable_count;
302 /** Link to osc_shrinker_list */
303 struct list_head cl_shrink_list;
305 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
306 atomic_t cl_destroy_in_flight;
307 wait_queue_head_t cl_destroy_waitq;
309 struct mdc_rpc_lock *cl_rpc_lock;
311 /* modify rpcs in flight
312 * currently used for metadata only */
313 spinlock_t cl_mod_rpcs_lock;
314 __u16 cl_max_mod_rpcs_in_flight;
315 __u16 cl_mod_rpcs_in_flight;
316 __u16 cl_close_rpcs_in_flight;
317 wait_queue_head_t cl_mod_rpcs_waitq;
318 unsigned long *cl_mod_tag_bitmap;
319 struct obd_histogram cl_mod_rpcs_hist;
322 struct mutex cl_mgc_mutex;
323 struct local_oid_storage *cl_mgc_los;
324 struct dt_object *cl_mgc_configs_dir;
325 struct obd_export *cl_mgc_mgsexp;
326 atomic_t cl_mgc_refcount;
327 /* in-flight control list and total RPCs counter */
328 struct list_head cl_flight_waiters;
329 __u32 cl_rpcs_in_flight;
331 /* checksumming for data sent over the network */
332 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
333 cl_checksum_dump:1; /* same */
334 /* supported checksum types that are worked out at connect time */
335 __u32 cl_supp_cksum_types;
336 /* checksum algorithm to be used */
337 enum cksum_types cl_cksum_type;
339 /* also protected by the poorly named _loi_list_lock lock above */
340 struct osc_async_rc cl_ar;
342 /* sequence manager */
343 struct lu_client_seq *cl_seq;
344 struct rw_semaphore cl_seq_rwsem;
346 atomic_t cl_resends; /* resend count */
348 /* ptlrpc work for writeback in ptlrpcd context */
349 void *cl_writeback_work;
351 /* hash tables for osc_quota_info */
352 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
353 /* Links to the global list of registered changelog devices */
354 struct list_head cl_chg_dev_linkage;
356 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
363 struct echo_client_obd {
364 struct obd_export *ec_exp; /* the local connection to osc/lov */
366 struct list_head ec_objects;
367 struct list_head ec_locks;
371 /* Generic subset of OSTs */
373 __u32 *op_array; /* array of index of
375 unsigned int op_count; /* number of OSTs in the array */
376 unsigned int op_size; /* allocated size of lp_array */
377 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
380 /* allow statfs data caching for 1 second */
381 #define OBD_STATFS_CACHE_SECONDS 1
383 struct lov_tgt_desc {
384 struct list_head ltd_kill;
385 struct obd_uuid ltd_uuid;
386 struct obd_device *ltd_obd;
387 struct obd_export *ltd_exp;
389 __u32 ltd_index; /* index in lov_obd->tgts */
390 unsigned long ltd_active:1,/* is this target up for requests */
391 ltd_activate:1,/* should target be activated */
392 ltd_reap:1; /* should this target be deleted */
395 struct lov_md_tgt_desc {
396 struct obd_device *lmtd_mdc;
401 struct lov_desc desc;
402 struct lov_tgt_desc **lov_tgts; /* sparse array */
403 struct ost_pool lov_packed; /* all OSTs in a packed
405 struct mutex lov_lock;
406 struct obd_connect_data lov_ocd;
407 struct proc_dir_entry *targets_proc_entry;
408 atomic_t lov_refcount;
409 __u32 lov_death_row; /* tgts scheduled to be deleted */
410 __u32 lov_tgt_size; /* size of tgts array */
413 struct cfs_hash *lov_pools_hash_body; /* used for key access */
414 struct list_head lov_pool_list; /* used for sequential access */
415 struct proc_dir_entry *lov_pool_proc_entry;
416 enum lustre_sec_part lov_sp_me;
418 /* Cached LRU and unstable data from upper layer */
419 struct cl_client_cache *lov_cache;
421 struct rw_semaphore lov_notify_lock;
422 /* Data-on-MDT: MDC array */
423 struct lov_md_tgt_desc *lov_mdc_tgts;
425 struct kobject *lov_tgts_kobj;
428 struct lmv_tgt_desc {
429 struct obd_uuid ltd_uuid;
430 struct obd_device *ltd_obd;
431 struct obd_export *ltd_exp;
433 struct mutex ltd_fid_mutex;
434 unsigned long ltd_active:1; /* target up for requests */
438 struct lu_client_fld lmv_fld;
440 struct lmv_desc desc;
442 struct mutex lmv_init_mutex;
447 __u32 tgts_size; /* size of tgts array */
448 struct lmv_tgt_desc **tgts;
450 struct obd_connect_data conn_data;
451 struct kobject *lmv_tgts_kobj;
454 struct niobuf_local {
455 __u64 lnb_file_offset;
456 __u32 lnb_page_offset;
460 struct page *lnb_page;
464 struct tgt_thread_big_cache {
465 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
468 #define LUSTRE_FLD_NAME "fld"
469 #define LUSTRE_SEQ_NAME "seq"
471 #define LUSTRE_MDD_NAME "mdd"
472 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
473 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
474 #define LUSTRE_VVP_NAME "vvp"
475 #define LUSTRE_LMV_NAME "lmv"
476 #define LUSTRE_SLP_NAME "slp"
477 #define LUSTRE_LOD_NAME "lod"
478 #define LUSTRE_OSP_NAME "osp"
479 #define LUSTRE_LWP_NAME "lwp"
481 /* obd device type names */
482 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
483 #define LUSTRE_MDS_NAME "mds"
484 #define LUSTRE_MDT_NAME "mdt"
485 #define LUSTRE_MDC_NAME "mdc"
486 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
487 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
488 #define LUSTRE_OSC_NAME "osc"
489 #define LUSTRE_LOV_NAME "lov"
490 #define LUSTRE_MGS_NAME "mgs"
491 #define LUSTRE_MGC_NAME "mgc"
493 #define LUSTRE_ECHO_NAME "obdecho"
494 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
495 #define LUSTRE_QMT_NAME "qmt"
497 /* Constant obd names (post-rename) */
498 #define LUSTRE_MDS_OBDNAME "MDS"
499 #define LUSTRE_OSS_OBDNAME "OSS"
500 #define LUSTRE_MGS_OBDNAME "MGS"
501 #define LUSTRE_MGC_OBDNAME "MGC"
503 static inline int is_lwp_on_mdt(char *name)
507 ptr = strrchr(name, '-');
509 CERROR("%s is not a obdname\n", name);
513 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
515 if (strncmp(ptr + 1, "MDT", 3) != 0)
518 while (*(--ptr) != '-' && ptr != name);
523 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
529 static inline int is_lwp_on_ost(char *name)
533 ptr = strrchr(name, '-');
535 CERROR("%s is not a obdname\n", name);
539 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
541 if (strncmp(ptr + 1, "OST", 3) != 0)
544 while (*(--ptr) != '-' && ptr != name);
549 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
556 * Events signalled through obd_notify() upcall-chain.
558 enum obd_notify_event {
559 /* Device connect start */
561 /* Device activated */
563 /* Device deactivated */
565 /* Connect data for import were changed */
567 /* Administratively deactivate/activate event */
568 OBD_NOTIFY_DEACTIVATE,
573 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
574 * being main example).
576 struct obd_notify_upcall {
577 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
578 enum obd_notify_event ev, void *owner);
579 /* Opaque datum supplied by upper layer listener */
583 struct target_recovery_data {
584 svc_handler_t trd_recovery_handler;
585 pid_t trd_processing_task;
586 struct completion trd_starting;
587 struct completion trd_finishing;
590 struct obd_llog_group {
591 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
592 wait_queue_head_t olg_waitq;
594 struct mutex olg_cat_processing;
597 /* corresponds to one of the obd's */
598 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
601 struct obd_type *obd_type;
602 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
603 int obd_minor; /* device number: lctl dl */
604 struct lu_device *obd_lu_dev;
606 /* common and UUID name of this device */
607 struct obd_uuid obd_uuid;
608 char obd_name[MAX_OBD_NAME];
610 /* bitfield modification is protected by obd_dev_lock */
612 obd_attached:1, /* finished attach */
613 obd_set_up:1, /* finished setup */
614 obd_recovering:1, /* there are recoverable clients */
615 obd_abort_recovery:1, /* recovery expired */
616 obd_version_recov:1, /* obd uses version checking */
617 obd_replayable:1, /* recovery enabled; inform clients */
618 obd_no_transno:1, /* no committed-transno notification */
619 obd_no_recov:1, /* fail instead of retry messages */
620 obd_stopping:1, /* started cleanup */
621 obd_starting:1, /* started setup */
622 obd_force:1, /* cleanup with > 0 obd refcount */
623 obd_fail:1, /* cleanup with failover */
624 obd_no_conn:1, /* deny new connections */
625 obd_inactive:1, /* device active/inactive
626 * (for /proc/status only!!) */
627 obd_no_ir:1, /* no imperative recovery. */
628 obd_process_conf:1, /* device is processing mgs config */
629 obd_checksum_dump:1; /* dump pages upon cksum error */
631 /* use separate field as it is set in interrupt to don't mess with
632 * protection of other bits using _bh lock */
633 unsigned long obd_recovery_expired:1;
634 /* uuid-export hash body */
635 struct cfs_hash *obd_uuid_hash;
636 /* nid-export hash body */
637 struct cfs_hash *obd_nid_hash;
639 struct cfs_hash *obd_nid_stats_hash;
640 /* client_generation-export hash body */
641 struct cfs_hash *obd_gen_hash;
642 struct list_head obd_nid_stats;
643 struct list_head obd_exports;
644 struct list_head obd_unlinked_exports;
645 struct list_head obd_delayed_exports;
646 struct list_head obd_lwp_list;
647 atomic_t obd_refcount;
649 spinlock_t obd_nid_lock;
650 struct ldlm_namespace *obd_namespace;
651 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
652 /* a spinlock is OK for what we do now, may need a semaphore later */
653 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
654 spinlock_t obd_osfs_lock;
655 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
656 time64_t obd_osfs_age;
657 __u64 obd_last_committed;
658 struct mutex obd_dev_mutex;
659 struct lvfs_run_ctxt obd_lvfs_ctxt;
660 struct obd_llog_group obd_olg; /* default llog group */
661 struct obd_device *obd_observer;
662 struct rw_semaphore obd_observer_link_sem;
663 struct obd_notify_upcall obd_upcall;
664 struct obd_export *obd_self_export;
665 struct obd_export *obd_lwp_export;
666 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
667 struct list_head obd_exports_timed;
668 time64_t obd_eviction_timer; /* for ping evictor */
670 int obd_max_recoverable_clients;
671 atomic_t obd_connected_clients;
672 int obd_stale_clients;
673 /* this lock protects all recovery list_heads, timer and
674 * obd_next_recovery_transno value */
675 spinlock_t obd_recovery_task_lock;
676 __u64 obd_next_recovery_transno;
677 int obd_replayed_requests;
678 int obd_requests_queued_for_recovery;
679 wait_queue_head_t obd_next_transno_waitq;
680 /* protected by obd_recovery_task_lock */
681 struct timer_list obd_recovery_timer;
683 time64_t obd_recovery_start;
684 /* seconds, for lprocfs_status */
685 time64_t obd_recovery_end;
686 time64_t obd_recovery_time_hard;
687 time64_t obd_recovery_timeout;
688 int obd_recovery_ir_factor;
690 /* new recovery stuff from CMD2 */
691 int obd_replayed_locks;
692 atomic_t obd_req_replay_clients;
693 atomic_t obd_lock_replay_clients;
694 struct target_recovery_data obd_recovery_data;
696 /* all lists are protected by obd_recovery_task_lock */
697 struct list_head obd_req_replay_queue;
698 struct list_head obd_lock_replay_queue;
699 struct list_head obd_final_req_queue;
702 #ifdef HAVE_SERVER_SUPPORT
703 struct obd_device_target obt;
704 struct filter_obd filter;
706 struct echo_obd echo;
708 struct client_obd cli;
709 struct echo_client_obd echo_client;
714 /* Fields used by LProcFS */
715 struct lprocfs_stats *obd_stats;
717 unsigned int obd_md_cntr_base;
718 struct lprocfs_stats *obd_md_stats;
720 struct dentry *obd_debugfs_entry;
721 struct proc_dir_entry *obd_proc_entry;
722 struct proc_dir_entry *obd_proc_exports_entry;
723 struct dentry *obd_svc_debugfs_entry;
724 struct lprocfs_stats *obd_svc_stats;
725 const struct attribute **obd_attrs;
726 struct lprocfs_vars *obd_vars;
727 atomic_t obd_evict_inprogress;
728 wait_queue_head_t obd_evict_inprogress_waitq;
729 struct list_head obd_evict_list; /* protected with pet_lock */
732 * LDLM pool part. Save last calculated SLV and Limit.
734 rwlock_t obd_pool_lock;
738 int obd_conn_inprogress;
741 * List of outstanding class_incref()'s fo this OBD. For debugging. */
742 struct lu_ref obd_reference;
744 struct kset obd_kset; /* sysfs object collection */
745 struct kobj_type obd_ktype;
746 struct completion obd_kobj_unregister;
749 /* get/set_info keys */
750 #define KEY_ASYNC "async"
751 #define KEY_CHANGELOG_CLEAR "changelog_clear"
752 #define KEY_FID2PATH "fid2path"
753 #define KEY_CHECKSUM "checksum"
754 #define KEY_CLEAR_FS "clear_fs"
755 #define KEY_CONN_DATA "conn_data"
756 #define KEY_EVICT_BY_NID "evict_by_nid"
757 #define KEY_FIEMAP "fiemap"
758 #define KEY_FLUSH_CTX "flush_ctx"
759 #define KEY_GRANT_SHRINK "grant_shrink"
760 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
761 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
762 #define KEY_INTERMDS "inter_mds"
763 #define KEY_LAST_ID "last_id"
764 #define KEY_LAST_FID "last_fid"
765 #define KEY_MAX_EASIZE "max_easize"
766 #define KEY_DEFAULT_EASIZE "default_easize"
767 #define KEY_MGSSEC "mgssec"
768 #define KEY_READ_ONLY "read-only"
769 #define KEY_REGISTER_TARGET "register_target"
770 #define KEY_SET_FS "set_fs"
771 #define KEY_TGT_COUNT "tgt_count"
772 /* KEY_SET_INFO in lustre_idl.h */
773 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
775 #define KEY_CACHE_SET "cache_set"
776 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
777 #define KEY_OSP_CONNECTED "osp_connected"
781 static inline int it_to_lock_mode(struct lookup_intent *it)
783 /* CREAT needs to be tested before open (both could be set) */
784 if (it->it_op & IT_CREAT)
786 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
788 else if (it->it_op & IT_LAYOUT)
789 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
790 else if (it->it_op & IT_READDIR)
792 else if (it->it_op & IT_GETXATTR)
795 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
800 MF_MDC_CANCEL_FID1 = 1 << 0,
801 MF_MDC_CANCEL_FID2 = 1 << 1,
802 MF_MDC_CANCEL_FID3 = 1 << 2,
803 MF_MDC_CANCEL_FID4 = 1 << 3,
804 MF_GET_MDT_IDX = 1 << 4,
808 CLI_SET_MEA = 1 << 0,
809 CLI_RM_ENTRY = 1 << 1,
812 CLI_MIGRATE = 1 << 4,
816 * GETXATTR is not included as only a couple of fields in the reply body
817 * is filled, but not FID which is needed for common intent handling in
818 * mdc_finish_intent_lock()
820 static inline bool it_has_reply_body(const struct lookup_intent *it)
822 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
826 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
827 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
828 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
829 struct lu_fid op_fid4; /* to the operation locks. */
830 u32 op_mds; /* what mds server open will go to */
832 struct lustre_handle op_handle;
836 struct lmv_stripe_md *op_mea1;
837 struct lmv_stripe_md *op_mea2;
838 __u32 op_suppgids[2];
845 /* iattr fields and blocks. */
846 struct iattr op_attr;
847 loff_t op_attr_blocks;
848 __u64 op_valid; /* OBD_MD_* */
849 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
851 enum md_op_flags op_flags;
853 /* Various operation flags. */
854 enum mds_op_bias op_bias;
856 /* used to transfer info between the stacks of MD client
857 * see enum op_cli_flags */
858 enum md_cli_flags op_cli_flags;
860 /* File object data version for HSM release, on client */
861 __u64 op_data_version;
862 struct lustre_handle op_lease_handle;
864 /* File security context, for creates. */
865 const char *op_file_secctx_name;
866 void *op_file_secctx;
867 __u32 op_file_secctx_size;
869 /* default stripe offset */
870 __u32 op_default_stripe_offset;
874 /* Used by readdir */
875 unsigned int op_max_pages;
881 int (*md_blocking_ast)(struct ldlm_lock *lock,
882 struct ldlm_lock_desc *desc,
883 void *data, int flag);
886 struct md_enqueue_info;
887 /* metadata stat-ahead */
888 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
889 struct md_enqueue_info *minfo,
892 struct md_enqueue_info {
893 struct md_op_data mi_data;
894 struct lookup_intent mi_it;
895 struct lustre_handle mi_lockh;
896 struct inode *mi_dir;
897 struct ldlm_enqueue_info mi_einfo;
898 md_enqueue_cb_t mi_cb;
903 struct module *o_owner;
904 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
905 void *karg, void __user *uarg);
906 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
907 __u32 keylen, void *key, __u32 *vallen, void *val);
908 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
909 __u32 keylen, void *key,
910 __u32 vallen, void *val,
911 struct ptlrpc_request_set *set);
912 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
913 int (*o_precleanup)(struct obd_device *dev);
914 int (*o_cleanup)(struct obd_device *dev);
915 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
916 int (*o_postrecov)(struct obd_device *dev);
917 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
919 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
920 /* connect to the target device with given connection
921 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
922 * granted by the target, which are guaranteed to be a subset of flags
923 * asked for. If @ocd == NULL, use default parameters. */
924 int (*o_connect)(const struct lu_env *env,
925 struct obd_export **exp, struct obd_device *src,
926 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
928 int (*o_reconnect)(const struct lu_env *env,
929 struct obd_export *exp, struct obd_device *src,
930 struct obd_uuid *cluuid,
931 struct obd_connect_data *ocd,
933 int (*o_disconnect)(struct obd_export *exp);
935 /* Initialize/finalize fids infrastructure. */
936 int (*o_fid_init)(struct obd_device *obd,
937 struct obd_export *exp, enum lu_cli_type type);
938 int (*o_fid_fini)(struct obd_device *obd);
940 /* Allocate new fid according to passed @hint. */
941 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
942 struct lu_fid *fid, struct md_op_data *op_data);
945 * Object with @fid is getting deleted, we may want to do something
948 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
949 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
950 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
951 time64_t max_age, struct ptlrpc_request_set *set);
952 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
954 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
956 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
958 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
960 int (*o_preprw)(const struct lu_env *env, int cmd,
961 struct obd_export *exp, struct obdo *oa, int objcount,
962 struct obd_ioobj *obj, struct niobuf_remote *remote,
963 int *nr_pages, struct niobuf_local *local);
964 int (*o_commitrw)(const struct lu_env *env, int cmd,
965 struct obd_export *exp, struct obdo *oa,
966 int objcount, struct obd_ioobj *obj,
967 struct niobuf_remote *remote, int pages,
968 struct niobuf_local *local, int rc);
969 int (*o_init_export)(struct obd_export *exp);
970 int (*o_destroy_export)(struct obd_export *exp);
972 int (*o_import_event)(struct obd_device *, struct obd_import *,
973 enum obd_import_event);
975 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
976 enum obd_notify_event ev);
978 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
979 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
982 int (*o_quotactl)(struct obd_device *, struct obd_export *,
983 struct obd_quotactl *);
985 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
988 int (*o_pool_new)(struct obd_device *obd, char *poolname);
989 int (*o_pool_del)(struct obd_device *obd, char *poolname);
990 int (*o_pool_add)(struct obd_device *obd, char *poolname,
992 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
998 struct mdt_body *body;
999 struct lu_buf layout;
1000 struct lmv_stripe_md *lmv;
1001 #ifdef CONFIG_FS_POSIX_ACL
1002 struct posix_acl *posix_acl;
1006 struct md_open_data {
1007 struct obd_client_handle *mod_och;
1008 struct ptlrpc_request *mod_open_req;
1009 struct ptlrpc_request *mod_close_req;
1010 atomic_t mod_refcount;
1014 struct obd_client_handle {
1015 struct lustre_handle och_fh;
1016 struct lu_fid och_fid;
1017 struct md_open_data *och_mod;
1018 struct lustre_handle och_lease_handle; /* open lock for lease */
1023 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1025 struct lookup_intent;
1029 /* Every operation from MD_STATS_FIRST_OP up to and including
1030 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1031 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1032 * Operations after MD_STATS_LAST_OP are excluded from stats.
1033 * There are a few reasons for doing this: we prune the 17
1034 * counters which will be of minimal use in understanding
1035 * metadata utilization, we save memory by allocating 15
1036 * instead of 32 counters, we save cycles by not counting.
1038 * MD_STATS_FIRST_OP must be the first member of md_ops.
1040 #define MD_STATS_FIRST_OP m_close
1041 int (*m_close)(struct obd_export *, struct md_op_data *,
1042 struct md_open_data *, struct ptlrpc_request **);
1044 int (*m_create)(struct obd_export *, struct md_op_data *,
1045 const void *, size_t, umode_t, uid_t, gid_t,
1046 cfs_cap_t, __u64, struct ptlrpc_request **);
1048 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1049 const union ldlm_policy_data *, struct md_op_data *,
1050 struct lustre_handle *, __u64);
1052 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1053 struct ptlrpc_request **);
1055 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1056 struct lookup_intent *,
1057 struct ptlrpc_request **,
1058 ldlm_blocking_callback, __u64);
1060 int (*m_link)(struct obd_export *, struct md_op_data *,
1061 struct ptlrpc_request **);
1063 int (*m_rename)(struct obd_export *, struct md_op_data *,
1064 const char *, size_t, const char *, size_t,
1065 struct ptlrpc_request **);
1067 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1068 size_t , struct ptlrpc_request **);
1070 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1071 struct ptlrpc_request **);
1073 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1074 struct md_callback *cb_op, __u64 hash_offset,
1075 struct page **ppage);
1077 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1078 struct ptlrpc_request **);
1080 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1081 u64, const char *, const void *, size_t, unsigned int,
1082 u32, struct ptlrpc_request **);
1084 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1085 u64, const char *, size_t, struct ptlrpc_request **);
1087 int (*m_intent_getattr_async)(struct obd_export *,
1088 struct md_enqueue_info *);
1090 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1091 struct lu_fid *, __u64 *bits);
1093 #define MD_STATS_LAST_OP m_revalidate_lock
1095 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1097 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1098 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1100 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1101 struct ptlrpc_request **);
1103 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1105 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1106 struct obd_export *, struct obd_export *,
1107 struct lustre_md *);
1109 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1111 int (*m_merge_attr)(struct obd_export *,
1112 const struct lmv_stripe_md *lsm,
1113 struct cl_attr *attr, ldlm_blocking_callback);
1115 int (*m_set_open_replay_data)(struct obd_export *,
1116 struct obd_client_handle *,
1117 struct lookup_intent *);
1119 int (*m_clear_open_replay_data)(struct obd_export *,
1120 struct obd_client_handle *);
1122 int (*m_set_lock_data)(struct obd_export *,
1123 const struct lustre_handle *, void *, __u64 *);
1125 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1126 const struct lu_fid *, enum ldlm_type,
1127 union ldlm_policy_data *, enum ldlm_mode,
1128 struct lustre_handle *);
1130 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1131 union ldlm_policy_data *, enum ldlm_mode,
1132 enum ldlm_cancel_flags flags, void *opaque);
1134 int (*m_get_fid_from_lsm)(struct obd_export *,
1135 const struct lmv_stripe_md *,
1136 const char *name, int namelen,
1137 struct lu_fid *fid);
1138 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1139 const union lmv_mds_md *lmv, size_t lmv_size);
1142 static inline struct md_open_data *obd_mod_alloc(void)
1144 struct md_open_data *mod;
1148 atomic_set(&mod->mod_refcount, 1);
1152 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1153 #define obd_mod_put(mod) \
1155 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1156 if ((mod)->mod_open_req) \
1157 ptlrpc_req_finished((mod)->mod_open_req); \
1158 OBD_FREE_PTR(mod); \
1162 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1163 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1164 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1166 /* return 1 if client should be resend request */
1167 static inline int client_should_resend(int resend, struct client_obd *cli)
1169 return atomic_read(&cli->cl_resends) ?
1170 atomic_read(&cli->cl_resends) > resend : 1;
1174 * Return device name for this device
1176 * XXX: lu_device is declared before obd_device, while a pointer pointing
1177 * back to obd_device in lu_device, so this helper function defines here
1178 * instead of in lu_object.h
1180 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1182 return lu_dev->ld_obd->obd_name;
1185 static inline bool filename_is_volatile(const char *name, size_t namelen,
1191 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1194 /* caller does not care of idx */
1198 /* volatile file, the MDT can be set from name */
1199 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1200 /* if no MDT is specified, use std way */
1201 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1203 /* test for no MDT idx case */
1204 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1205 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1209 /* we have an idx, read it */
1210 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1211 *idx = simple_strtoul(start, &end, 16);
1213 * no digit, no trailing :, negative value
1215 if (((*idx == 0) && (end == start)) ||
1216 (*end != ':') || (*idx < 0))
1221 /* bad format of mdt idx, we cannot return an error
1222 * to caller so we use hash algo */
1223 CERROR("Bad volatile file name format: %s\n",
1224 name + LUSTRE_VOLATILE_HDR_LEN);
1228 static inline int cli_brw_size(struct obd_device *obd)
1230 LASSERT(obd != NULL);
1231 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1235 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1236 * of the client should be increased accordingly to avoid sending fragmented
1237 * RPCs over the network when the client runs out of the maximum dirty space
1238 * when so many RPCs are being generated.
1240 static inline void client_adjust_max_dirty(struct client_obd *cli)
1243 if (cli->cl_dirty_max_pages <= 0) {
1244 cli->cl_dirty_max_pages =
1245 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1247 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1248 cli->cl_max_pages_per_rpc;
1250 if (dirty_max > cli->cl_dirty_max_pages)
1251 cli->cl_dirty_max_pages = dirty_max;
1254 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1255 cli->cl_dirty_max_pages = totalram_pages / 8;
1257 /* This value is exported to userspace through the max_dirty_mb
1258 * parameter. So we round up the number of pages to make it a round
1260 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1261 1 << (20 - PAGE_SHIFT));
1264 #endif /* __OBD_H */