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 lut_sync_lock_cancel */
140 enum tgt_sync_lock_cancel {
141 SYNC_LOCK_CANCEL_NEVER = 0,
142 SYNC_LOCK_CANCEL_BLOCKING = 1,
143 SYNC_LOCK_CANCEL_ALWAYS = 2,
147 * Limit reply buffer size for striping data to one x86_64 page. This
148 * value is chosen to fit the striping data for common use cases while
149 * staying well below the limit at which the buffer must be backed by
150 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
153 #define OBD_MAX_DEFAULT_EA_SIZE 4096
155 enum obd_cl_sem_lock_class {
164 struct rw_semaphore cl_sem;
165 struct obd_uuid cl_target_uuid;
166 struct obd_import *cl_import; /* ptlrpc connection state */
167 size_t cl_conn_count;
169 /* Cache maximum and default values for easize. This is
170 * strictly a performance optimization to minimize calls to
171 * obd_size_diskmd(). The default values are used to calculate the
172 * initial size of a request buffer. The ptlrpc layer will resize the
173 * buffer as needed to accommodate a larger reply from the
174 * server. The default values should be small enough to avoid wasted
175 * memory and excessive use of vmalloc(), yet large enough to avoid
176 * reallocating the buffer in the common use case. */
178 /* Default EA size for striping attributes. It is initialized at
179 * mount-time based on the default stripe width of the filesystem,
180 * then it tracks the largest observed EA size advertised by
181 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
182 __u32 cl_default_mds_easize;
184 /* Maximum possible EA size computed at mount-time based on
185 * the number of OSTs in the filesystem. May be increased at
186 * run-time if a larger observed size is advertised by the MDT. */
187 __u32 cl_max_mds_easize;
189 /* Data-on-MDT specific value to set larger reply buffer for possible
190 * data read along with open/stat requests. By default it tries to use
191 * unused space in reply buffer.
192 * This value is used to ensure that reply buffer has at least as
193 * much free space as value indicates. That free space is gained from
194 * LOV EA buffer which is small for DoM files and on big systems can
195 * provide up to 32KB of extra space in reply buffer.
196 * Default value is 8K now.
198 __u32 cl_dom_min_inline_repsize;
200 enum lustre_sec_part cl_sp_me;
201 enum lustre_sec_part cl_sp_to;
202 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
204 /* the grant values are protected by loi_list_lock below */
205 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
206 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
207 unsigned long cl_avail_grant; /* bytes of credit for ost */
208 unsigned long cl_lost_grant; /* lost credits (trunc) */
209 /* grant consumed for dirty pages */
210 unsigned long cl_dirty_grant;
212 /* since we allocate grant by blocks, we don't know how many grant will
213 * be used to add a page into cache. As a solution, we reserve maximum
214 * grant before trying to dirty a page and unreserve the rest.
215 * See osc_{reserve|unreserve}_grant for details. */
216 long cl_reserved_grant;
217 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
218 time64_t cl_next_shrink_grant; /* seconds */
219 struct list_head cl_grant_chain;
220 time64_t cl_grant_shrink_interval; /* seconds */
222 /* A chunk is an optimal size used by osc_extent to determine
223 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
225 /* extent insertion metadata overhead to be accounted in grant,
227 unsigned int cl_grant_extent_tax;
228 /* maximum extent size, in number of pages */
229 unsigned int cl_max_extent_pages;
231 /* keep track of objects that have lois that contain pages which
232 * have been queued for async brw. this lock also protects the
233 * lists of osc_client_pages that hang off of the loi */
235 * ->cl_loi_list_lock protects consistency of
236 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
237 * ->ap_completion() call-backs are executed under this lock. As we
238 * cannot guarantee that these call-backs never block on all platforms
239 * (as a matter of fact they do block on Mac OS X), type of
240 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
241 * and blocking mutex on Mac OS X. (Alternative is to make this lock
242 * blocking everywhere, but we don't want to slow down fast-path of
243 * our main platform.)
245 * NB by Jinshan: though field names are still _loi_, but actually
246 * osc_object{}s are in the list.
248 spinlock_t cl_loi_list_lock;
249 struct list_head cl_loi_ready_list;
250 struct list_head cl_loi_hp_ready_list;
251 struct list_head cl_loi_write_list;
252 struct list_head cl_loi_read_list;
253 __u32 cl_r_in_flight;
254 __u32 cl_w_in_flight;
255 /* just a sum of the loi/lop pending numbers to be exported by /proc */
256 atomic_t cl_pending_w_pages;
257 atomic_t cl_pending_r_pages;
258 u32 cl_max_pages_per_rpc;
259 u32 cl_max_rpcs_in_flight;
260 u32 cl_max_short_io_bytes;
261 struct obd_histogram cl_read_rpc_hist;
262 struct obd_histogram cl_write_rpc_hist;
263 struct obd_histogram cl_read_page_hist;
264 struct obd_histogram cl_write_page_hist;
265 struct obd_histogram cl_read_offset_hist;
266 struct obd_histogram cl_write_offset_hist;
268 /** LRU for osc caching pages */
269 struct cl_client_cache *cl_cache;
270 /** member of cl_cache->ccc_lru */
271 struct list_head cl_lru_osc;
272 /** # of available LRU slots left in the per-OSC cache.
273 * Available LRU slots are shared by all OSCs of the same file system,
274 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
275 atomic_long_t *cl_lru_left;
276 /** # of busy LRU pages. A page is considered busy if it's in writeback
277 * queue, or in transfer. Busy pages can't be discarded so they are not
279 atomic_long_t cl_lru_busy;
280 /** # of LRU pages in the cache for this client_obd */
281 atomic_long_t cl_lru_in_list;
282 /** # of threads are shrinking LRU cache. To avoid contention, it's not
283 * allowed to have multiple threads shrinking LRU cache. */
284 atomic_t cl_lru_shrinkers;
285 /** The time when this LRU cache was last used. */
286 time64_t cl_lru_last_used;
287 /** stats: how many reclaims have happened for this client_obd.
288 * reclaim and shrink - shrink is async, voluntarily rebalancing;
289 * reclaim is sync, initiated by IO thread when the LRU slots are
291 __u64 cl_lru_reclaim;
292 /** List of LRU pages for this client_obd */
293 struct list_head cl_lru_list;
294 /** Lock for LRU page list */
295 spinlock_t cl_lru_list_lock;
296 /** # of unstable pages in this client_obd.
297 * An unstable page is a page state that WRITE RPC has finished but
298 * the transaction has NOT yet committed. */
299 atomic_long_t cl_unstable_count;
300 /** Link to osc_shrinker_list */
301 struct list_head cl_shrink_list;
303 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
304 atomic_t cl_destroy_in_flight;
305 wait_queue_head_t cl_destroy_waitq;
307 /* modify rpcs in flight
308 * currently used for metadata only */
309 spinlock_t cl_mod_rpcs_lock;
310 __u16 cl_max_mod_rpcs_in_flight;
311 __u16 cl_mod_rpcs_in_flight;
312 __u16 cl_close_rpcs_in_flight;
313 wait_queue_head_t cl_mod_rpcs_waitq;
314 unsigned long *cl_mod_tag_bitmap;
315 struct obd_histogram cl_mod_rpcs_hist;
318 struct mutex cl_mgc_mutex;
319 struct local_oid_storage *cl_mgc_los;
320 struct dt_object *cl_mgc_configs_dir;
321 struct obd_export *cl_mgc_mgsexp;
322 atomic_t cl_mgc_refcount;
323 /* in-flight control list and total RPCs counter */
324 struct list_head cl_flight_waiters;
325 __u32 cl_rpcs_in_flight;
327 /* checksumming for data sent over the network */
328 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
329 cl_checksum_dump:1; /* same */
330 /* supported checksum types that are worked out at connect time */
331 __u32 cl_supp_cksum_types;
332 /* checksum algorithm to be used */
333 enum cksum_types cl_cksum_type;
335 /* also protected by the poorly named _loi_list_lock lock above */
336 struct osc_async_rc cl_ar;
338 /* sequence manager */
339 struct lu_client_seq *cl_seq;
340 struct rw_semaphore cl_seq_rwsem;
342 atomic_t cl_resends; /* resend count */
344 /* ptlrpc work for writeback in ptlrpcd context */
345 void *cl_writeback_work;
347 struct mutex cl_quota_mutex;
348 /* hash tables for osc_quota_info */
349 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
350 /* the xid of the request updating the hash tables */
351 __u64 cl_quota_last_xid;
352 /* Links to the global list of registered changelog devices */
353 struct list_head cl_chg_dev_linkage;
355 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
362 struct echo_client_obd {
363 struct obd_export *ec_exp; /* the local connection to osc/lov */
365 struct list_head ec_objects;
366 struct list_head ec_locks;
370 /* Generic subset of OSTs */
372 __u32 *op_array; /* array of index of
374 unsigned int op_count; /* number of OSTs in the array */
375 unsigned int op_size; /* allocated size of lp_array */
376 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
379 /* allow statfs data caching for 1 second */
380 #define OBD_STATFS_CACHE_SECONDS 1
381 /* arbitrary maximum. larger would be useless, allows catching bogus input */
382 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
384 struct lov_tgt_desc {
385 struct list_head ltd_kill;
386 struct obd_uuid ltd_uuid;
387 struct obd_device *ltd_obd;
388 struct obd_export *ltd_exp;
390 __u32 ltd_index; /* index in lov_obd->tgts */
391 unsigned long ltd_active:1,/* is this target up for requests */
392 ltd_activate:1,/* should target be activated */
393 ltd_reap:1; /* should this target be deleted */
396 struct lov_md_tgt_desc {
397 struct obd_device *lmtd_mdc;
402 struct lov_desc desc;
403 struct lov_tgt_desc **lov_tgts; /* sparse array */
404 struct ost_pool lov_packed; /* all OSTs in a packed
406 struct mutex lov_lock;
407 struct obd_connect_data lov_ocd;
408 struct proc_dir_entry *targets_proc_entry;
409 atomic_t lov_refcount;
410 __u32 lov_death_row; /* tgts scheduled to be deleted */
411 __u32 lov_tgt_size; /* size of tgts array */
414 struct cfs_hash *lov_pools_hash_body; /* used for key access */
415 struct list_head lov_pool_list; /* used for sequential access */
416 struct proc_dir_entry *lov_pool_proc_entry;
417 enum lustre_sec_part lov_sp_me;
419 /* Cached LRU and unstable data from upper layer */
420 struct cl_client_cache *lov_cache;
422 struct rw_semaphore lov_notify_lock;
423 /* Data-on-MDT: MDC array */
424 struct lov_md_tgt_desc *lov_mdc_tgts;
426 struct kobject *lov_tgts_kobj;
429 struct lmv_tgt_desc {
430 struct obd_uuid ltd_uuid;
431 struct obd_device *ltd_obd;
432 struct obd_export *ltd_exp;
434 struct mutex ltd_fid_mutex;
435 unsigned long ltd_active:1; /* target up for requests */
439 struct lu_client_fld lmv_fld;
441 struct lmv_desc desc;
443 struct mutex lmv_init_mutex;
447 u32 lmv_statfs_start;
449 u32 tgts_size; /* size of tgts array */
450 struct lmv_tgt_desc **tgts;
452 struct obd_connect_data conn_data;
453 struct kobject *lmv_tgts_kobj;
456 /* Minimum sector size is 512 */
457 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
459 struct niobuf_local {
460 __u64 lnb_file_offset;
461 __u32 lnb_page_offset;
465 struct page *lnb_page;
467 __u16 lnb_guards[MAX_GUARD_NUMBER];
468 __u16 lnb_guard_rpc:1;
469 __u16 lnb_guard_disk:1;
470 /* separate unlock for read path to allow shared access */
474 struct tgt_thread_big_cache {
475 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
478 #define LUSTRE_FLD_NAME "fld"
479 #define LUSTRE_SEQ_NAME "seq"
481 #define LUSTRE_MDD_NAME "mdd"
482 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
483 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
484 #define LUSTRE_VVP_NAME "vvp"
485 #define LUSTRE_LMV_NAME "lmv"
486 #define LUSTRE_SLP_NAME "slp"
487 #define LUSTRE_LOD_NAME "lod"
488 #define LUSTRE_OSP_NAME "osp"
489 #define LUSTRE_LWP_NAME "lwp"
491 /* obd device type names */
492 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
493 #define LUSTRE_MDS_NAME "mds"
494 #define LUSTRE_MDT_NAME "mdt"
495 #define LUSTRE_MDC_NAME "mdc"
496 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
497 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
498 #define LUSTRE_OSC_NAME "osc"
499 #define LUSTRE_LOV_NAME "lov"
500 #define LUSTRE_MGS_NAME "mgs"
501 #define LUSTRE_MGC_NAME "mgc"
503 #define LUSTRE_ECHO_NAME "obdecho"
504 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
505 #define LUSTRE_QMT_NAME "qmt"
507 /* Constant obd names (post-rename) */
508 #define LUSTRE_MDS_OBDNAME "MDS"
509 #define LUSTRE_OSS_OBDNAME "OSS"
510 #define LUSTRE_MGS_OBDNAME "MGS"
511 #define LUSTRE_MGC_OBDNAME "MGC"
513 static inline int is_lwp_on_mdt(char *name)
517 ptr = strrchr(name, '-');
519 CERROR("%s is not a obdname\n", name);
523 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
525 if (strncmp(ptr + 1, "MDT", 3) != 0)
528 while (*(--ptr) != '-' && ptr != name);
533 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
539 static inline int is_lwp_on_ost(char *name)
543 ptr = strrchr(name, '-');
545 CERROR("%s is not a obdname\n", name);
549 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
551 if (strncmp(ptr + 1, "OST", 3) != 0)
554 while (*(--ptr) != '-' && ptr != name);
559 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
566 * Events signalled through obd_notify() upcall-chain.
568 enum obd_notify_event {
569 /* Device connect start */
571 /* Device activated */
573 /* Device deactivated */
575 /* Connect data for import were changed */
577 /* Administratively deactivate/activate event */
578 OBD_NOTIFY_DEACTIVATE,
583 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
584 * being main example).
586 struct obd_notify_upcall {
587 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
588 enum obd_notify_event ev, void *owner);
589 /* Opaque datum supplied by upper layer listener */
593 struct target_recovery_data {
594 svc_handler_t trd_recovery_handler;
595 pid_t trd_processing_task;
596 struct completion trd_starting;
597 struct completion trd_finishing;
600 struct obd_llog_group {
601 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
602 wait_queue_head_t olg_waitq;
604 struct mutex olg_cat_processing;
607 /* corresponds to one of the obd's */
608 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
611 struct obd_type *obd_type;
612 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
613 int obd_minor; /* device number: lctl dl */
614 struct lu_device *obd_lu_dev;
616 /* common and UUID name of this device */
617 struct obd_uuid obd_uuid;
618 char obd_name[MAX_OBD_NAME];
620 /* bitfield modification is protected by obd_dev_lock */
622 obd_attached:1, /* finished attach */
623 obd_set_up:1, /* finished setup */
624 obd_recovering:1, /* there are recoverable clients */
625 obd_abort_recovery:1, /* recovery expired */
626 obd_version_recov:1, /* obd uses version checking */
627 obd_replayable:1, /* recovery enabled; inform clients */
628 obd_no_transno:1, /* no committed-transno notification */
629 obd_no_recov:1, /* fail instead of retry messages */
630 obd_stopping:1, /* started cleanup */
631 obd_starting:1, /* started setup */
632 obd_force:1, /* cleanup with > 0 obd refcount */
633 obd_fail:1, /* cleanup with failover */
634 obd_no_conn:1, /* deny new connections */
635 obd_inactive:1, /* device active/inactive
636 * (for /proc/status only!!) */
637 obd_no_ir:1, /* no imperative recovery. */
638 obd_process_conf:1, /* device is processing mgs config */
639 obd_checksum_dump:1; /* dump pages upon cksum error */
641 /* use separate field as it is set in interrupt to don't mess with
642 * protection of other bits using _bh lock */
643 unsigned long obd_recovery_expired:1;
644 /* uuid-export hash body */
645 struct cfs_hash *obd_uuid_hash;
646 /* nid-export hash body */
647 struct cfs_hash *obd_nid_hash;
649 struct cfs_hash *obd_nid_stats_hash;
650 /* client_generation-export hash body */
651 struct cfs_hash *obd_gen_hash;
652 struct list_head obd_nid_stats;
653 struct list_head obd_exports;
654 struct list_head obd_unlinked_exports;
655 struct list_head obd_delayed_exports;
656 struct list_head obd_lwp_list;
657 atomic_t obd_refcount;
659 spinlock_t obd_nid_lock;
660 struct ldlm_namespace *obd_namespace;
661 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
662 /* a spinlock is OK for what we do now, may need a semaphore later */
663 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
664 spinlock_t obd_osfs_lock;
665 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
666 time64_t obd_osfs_age;
667 __u64 obd_last_committed;
668 struct mutex obd_dev_mutex;
669 struct lvfs_run_ctxt obd_lvfs_ctxt;
670 struct obd_llog_group obd_olg; /* default llog group */
671 struct obd_device *obd_observer;
672 struct rw_semaphore obd_observer_link_sem;
673 struct obd_notify_upcall obd_upcall;
674 struct obd_export *obd_self_export;
675 struct obd_export *obd_lwp_export;
676 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
677 struct list_head obd_exports_timed;
678 time64_t obd_eviction_timer; /* for ping evictor */
680 atomic_t obd_max_recoverable_clients;
681 atomic_t obd_connected_clients;
682 int obd_stale_clients;
683 /* this lock protects all recovery list_heads, timer and
684 * obd_next_recovery_transno value */
685 spinlock_t obd_recovery_task_lock;
686 __u64 obd_next_recovery_transno;
687 int obd_replayed_requests;
688 int obd_requests_queued_for_recovery;
689 wait_queue_head_t obd_next_transno_waitq;
690 /* protected by obd_recovery_task_lock */
691 struct hrtimer obd_recovery_timer;
693 time64_t obd_recovery_start;
694 /* seconds, for lprocfs_status */
695 time64_t obd_recovery_end;
696 time64_t obd_recovery_time_hard;
697 time64_t obd_recovery_timeout;
698 int obd_recovery_ir_factor;
700 /* new recovery stuff from CMD2 */
701 int obd_replayed_locks;
702 atomic_t obd_req_replay_clients;
703 atomic_t obd_lock_replay_clients;
704 struct target_recovery_data obd_recovery_data;
706 /* all lists are protected by obd_recovery_task_lock */
707 struct list_head obd_req_replay_queue;
708 struct list_head obd_lock_replay_queue;
709 struct list_head obd_final_req_queue;
712 #ifdef HAVE_SERVER_SUPPORT
713 struct obd_device_target obt;
714 struct filter_obd filter;
716 struct echo_obd echo;
718 struct client_obd cli;
719 struct echo_client_obd echo_client;
724 /* Fields used by LProcFS */
725 struct lprocfs_stats *obd_stats;
727 struct lprocfs_stats *obd_md_stats;
729 struct dentry *obd_debugfs_entry;
730 struct proc_dir_entry *obd_proc_entry;
731 struct proc_dir_entry *obd_proc_exports_entry;
732 struct dentry *obd_svc_debugfs_entry;
733 struct lprocfs_stats *obd_svc_stats;
734 const struct attribute **obd_attrs;
735 struct lprocfs_vars *obd_vars;
736 atomic_t obd_evict_inprogress;
737 wait_queue_head_t obd_evict_inprogress_waitq;
738 struct list_head obd_evict_list; /* protected with pet_lock */
741 * LDLM pool part. Save last calculated SLV and Limit.
743 rwlock_t obd_pool_lock;
747 int obd_conn_inprogress;
750 * List of outstanding class_incref()'s fo this OBD. For debugging. */
751 struct lu_ref obd_reference;
753 struct kset obd_kset; /* sysfs object collection */
754 struct kobj_type obd_ktype;
755 struct completion obd_kobj_unregister;
758 /* get/set_info keys */
759 #define KEY_ASYNC "async"
760 #define KEY_CHANGELOG_CLEAR "changelog_clear"
761 #define KEY_FID2PATH "fid2path"
762 #define KEY_CHECKSUM "checksum"
763 #define KEY_CLEAR_FS "clear_fs"
764 #define KEY_CONN_DATA "conn_data"
765 #define KEY_EVICT_BY_NID "evict_by_nid"
766 #define KEY_FIEMAP "fiemap"
767 #define KEY_FLUSH_CTX "flush_ctx"
768 #define KEY_GRANT_SHRINK "grant_shrink"
769 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
770 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
771 #define KEY_INTERMDS "inter_mds"
772 #define KEY_LAST_ID "last_id"
773 #define KEY_LAST_FID "last_fid"
774 #define KEY_MAX_EASIZE "max_easize"
775 #define KEY_DEFAULT_EASIZE "default_easize"
776 #define KEY_MGSSEC "mgssec"
777 #define KEY_READ_ONLY "read-only"
778 #define KEY_REGISTER_TARGET "register_target"
779 #define KEY_SET_FS "set_fs"
780 #define KEY_TGT_COUNT "tgt_count"
781 /* KEY_SET_INFO in lustre_idl.h */
782 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
784 #define KEY_CACHE_SET "cache_set"
785 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
786 #define KEY_OSP_CONNECTED "osp_connected"
788 /* Flags for op_xvalid */
790 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
791 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
792 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
793 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
794 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
795 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
796 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
801 static inline int it_to_lock_mode(struct lookup_intent *it)
803 /* CREAT needs to be tested before open (both could be set) */
804 if (it->it_op & IT_CREAT)
806 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
808 else if (it->it_op & IT_LAYOUT)
809 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
810 else if (it->it_op & IT_READDIR)
812 else if (it->it_op & IT_GETXATTR)
815 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
820 MF_MDC_CANCEL_FID1 = 1 << 0,
821 MF_MDC_CANCEL_FID2 = 1 << 1,
822 MF_MDC_CANCEL_FID3 = 1 << 2,
823 MF_MDC_CANCEL_FID4 = 1 << 3,
824 MF_GET_MDT_IDX = 1 << 4,
825 MF_GETATTR_BY_FID = 1 << 5,
829 CLI_SET_MEA = 1 << 0,
830 CLI_RM_ENTRY = 1 << 1,
833 CLI_MIGRATE = 1 << 4,
837 * GETXATTR is not included as only a couple of fields in the reply body
838 * is filled, but not FID which is needed for common intent handling in
839 * mdc_finish_intent_lock()
841 static inline bool it_has_reply_body(const struct lookup_intent *it)
843 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
847 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
848 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
849 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
850 struct lu_fid op_fid4; /* to the operation locks. */
851 u32 op_mds; /* what mds server open will go to */
853 struct lustre_handle op_open_handle;
857 struct rw_semaphore *op_mea1_sem;
858 struct rw_semaphore *op_mea2_sem;
859 struct lmv_stripe_md *op_mea1;
860 struct lmv_stripe_md *op_mea2;
861 __u32 op_suppgids[2];
868 /* iattr fields and blocks. */
869 struct iattr op_attr;
870 enum op_xvalid op_xvalid; /* eXtra validity flags */
871 loff_t op_attr_blocks;
872 u64 op_valid; /* OBD_MD_* */
873 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
875 enum md_op_flags op_flags;
877 /* Various operation flags. */
878 enum mds_op_bias op_bias;
880 /* used to transfer info between the stacks of MD client
881 * see enum op_cli_flags */
882 enum md_cli_flags op_cli_flags;
884 /* File object data version for HSM release, on client */
885 __u64 op_data_version;
886 struct lustre_handle op_lease_handle;
888 /* File security context, for creates/metadata ops */
889 const char *op_file_secctx_name;
890 __u32 op_file_secctx_name_size;
891 void *op_file_secctx;
892 __u32 op_file_secctx_size;
894 /* default stripe offset */
895 __u32 op_default_stripe_offset;
899 /* Used by readdir */
900 unsigned int op_max_pages;
905 * used to access migrating dir: if it's set, assume migration is
906 * finished, use the new layout to access dir, otherwise use old layout.
907 * By default it's not set, because new files are created under new
908 * layout, if we can't find file with name under both old and new
909 * layout, we are sure file with name doesn't exist, but in reverse
910 * order there may be a race with creation by others.
912 bool op_post_migrate;
913 /* used to access dir with bash hash */
914 __u32 op_stripe_index;
918 int (*md_blocking_ast)(struct ldlm_lock *lock,
919 struct ldlm_lock_desc *desc,
920 void *data, int flag);
923 struct md_enqueue_info;
924 /* metadata stat-ahead */
925 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
926 struct md_enqueue_info *minfo,
929 struct md_enqueue_info {
930 struct md_op_data mi_data;
931 struct lookup_intent mi_it;
932 struct lustre_handle mi_lockh;
933 struct inode *mi_dir;
934 struct ldlm_enqueue_info mi_einfo;
935 md_enqueue_cb_t mi_cb;
940 struct module *o_owner;
941 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
942 void *karg, void __user *uarg);
943 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
944 __u32 keylen, void *key, __u32 *vallen, void *val);
945 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
946 __u32 keylen, void *key,
947 __u32 vallen, void *val,
948 struct ptlrpc_request_set *set);
949 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
950 int (*o_precleanup)(struct obd_device *dev);
951 int (*o_cleanup)(struct obd_device *dev);
952 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
953 int (*o_postrecov)(struct obd_device *dev);
954 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
956 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
957 /* connect to the target device with given connection
958 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
959 * granted by the target, which are guaranteed to be a subset of flags
960 * asked for. If @ocd == NULL, use default parameters. */
961 int (*o_connect)(const struct lu_env *env,
962 struct obd_export **exp, struct obd_device *src,
963 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
965 int (*o_reconnect)(const struct lu_env *env,
966 struct obd_export *exp, struct obd_device *src,
967 struct obd_uuid *cluuid,
968 struct obd_connect_data *ocd,
970 int (*o_disconnect)(struct obd_export *exp);
972 /* Initialize/finalize fids infrastructure. */
973 int (*o_fid_init)(struct obd_device *obd,
974 struct obd_export *exp, enum lu_cli_type type);
975 int (*o_fid_fini)(struct obd_device *obd);
977 /* Allocate new fid according to passed @hint. */
978 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
979 struct lu_fid *fid, struct md_op_data *op_data);
982 * Object with @fid is getting deleted, we may want to do something
985 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
986 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
987 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
988 time64_t max_age, struct ptlrpc_request_set *set);
989 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
991 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
993 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
995 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
997 int (*o_preprw)(const struct lu_env *env, int cmd,
998 struct obd_export *exp, struct obdo *oa, int objcount,
999 struct obd_ioobj *obj, struct niobuf_remote *remote,
1000 int *nr_pages, struct niobuf_local *local);
1001 int (*o_commitrw)(const struct lu_env *env, int cmd,
1002 struct obd_export *exp, struct obdo *oa,
1003 int objcount, struct obd_ioobj *obj,
1004 struct niobuf_remote *remote, int pages,
1005 struct niobuf_local *local, int rc);
1006 int (*o_init_export)(struct obd_export *exp);
1007 int (*o_destroy_export)(struct obd_export *exp);
1009 int (*o_import_event)(struct obd_device *, struct obd_import *,
1010 enum obd_import_event);
1012 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1013 enum obd_notify_event ev);
1015 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1016 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1019 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1020 struct obd_quotactl *);
1023 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1024 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1025 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1027 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1031 /* lmv structures */
1033 struct mdt_body *body;
1034 struct lu_buf layout;
1035 struct lmv_stripe_md *lmv;
1036 #ifdef CONFIG_FS_POSIX_ACL
1037 struct posix_acl *posix_acl;
1041 struct md_open_data {
1042 struct obd_client_handle *mod_och;
1043 struct ptlrpc_request *mod_open_req;
1044 struct ptlrpc_request *mod_close_req;
1045 atomic_t mod_refcount;
1049 struct obd_client_handle {
1050 struct lustre_handle och_open_handle;
1051 struct lu_fid och_fid;
1052 struct md_open_data *och_mod;
1053 struct lustre_handle och_lease_handle; /* open lock for lease */
1058 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1060 struct lookup_intent;
1064 int (*m_close)(struct obd_export *, struct md_op_data *,
1065 struct md_open_data *, struct ptlrpc_request **);
1067 int (*m_create)(struct obd_export *, struct md_op_data *,
1068 const void *, size_t, umode_t, uid_t, gid_t,
1069 cfs_cap_t, __u64, struct ptlrpc_request **);
1071 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1072 const union ldlm_policy_data *, struct md_op_data *,
1073 struct lustre_handle *, __u64);
1075 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1076 struct ptlrpc_request **);
1078 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1079 struct lookup_intent *,
1080 struct ptlrpc_request **,
1081 ldlm_blocking_callback, __u64);
1083 int (*m_link)(struct obd_export *, struct md_op_data *,
1084 struct ptlrpc_request **);
1086 int (*m_rename)(struct obd_export *, struct md_op_data *,
1087 const char *, size_t, const char *, size_t,
1088 struct ptlrpc_request **);
1090 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1091 size_t , struct ptlrpc_request **);
1093 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1094 struct ptlrpc_request **);
1096 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1097 struct md_callback *cb_op, __u64 hash_offset,
1098 struct page **ppage);
1100 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1101 struct ptlrpc_request **);
1103 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1104 u64, const char *, const void *, size_t, unsigned int,
1105 u32, struct ptlrpc_request **);
1107 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1108 u64, const char *, size_t, struct ptlrpc_request **);
1110 int (*m_intent_getattr_async)(struct obd_export *,
1111 struct md_enqueue_info *);
1113 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1114 struct lu_fid *, __u64 *bits);
1116 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1118 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1119 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1121 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1122 struct ptlrpc_request **);
1124 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1126 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1127 struct obd_export *, struct obd_export *,
1128 struct lustre_md *);
1130 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1132 int (*m_merge_attr)(struct obd_export *,
1133 const struct lmv_stripe_md *lsm,
1134 struct cl_attr *attr, ldlm_blocking_callback);
1136 int (*m_set_open_replay_data)(struct obd_export *,
1137 struct obd_client_handle *,
1138 struct lookup_intent *);
1140 int (*m_clear_open_replay_data)(struct obd_export *,
1141 struct obd_client_handle *);
1143 int (*m_set_lock_data)(struct obd_export *,
1144 const struct lustre_handle *, void *, __u64 *);
1146 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1147 const struct lu_fid *, enum ldlm_type,
1148 union ldlm_policy_data *, enum ldlm_mode,
1149 struct lustre_handle *);
1151 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1152 union ldlm_policy_data *, enum ldlm_mode,
1153 enum ldlm_cancel_flags flags, void *opaque);
1155 int (*m_get_fid_from_lsm)(struct obd_export *,
1156 const struct lmv_stripe_md *,
1157 const char *name, int namelen,
1158 struct lu_fid *fid);
1159 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1160 const union lmv_mds_md *lmv, size_t lmv_size);
1161 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1162 struct ptlrpc_request_set *set);
1165 static inline struct md_open_data *obd_mod_alloc(void)
1167 struct md_open_data *mod;
1171 atomic_set(&mod->mod_refcount, 1);
1175 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1176 #define obd_mod_put(mod) \
1178 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1179 if ((mod)->mod_open_req) \
1180 ptlrpc_req_finished((mod)->mod_open_req); \
1181 OBD_FREE_PTR(mod); \
1185 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1186 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1187 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1189 /* return 1 if client should be resend request */
1190 static inline int client_should_resend(int resend, struct client_obd *cli)
1192 return atomic_read(&cli->cl_resends) ?
1193 atomic_read(&cli->cl_resends) > resend : 1;
1197 * Return device name for this device
1199 * XXX: lu_device is declared before obd_device, while a pointer pointing
1200 * back to obd_device in lu_device, so this helper function defines here
1201 * instead of in lu_object.h
1203 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1205 return lu_dev->ld_obd->obd_name;
1208 static inline bool filename_is_volatile(const char *name, size_t namelen,
1214 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1217 /* caller does not care of idx */
1221 /* volatile file, the MDT can be set from name */
1222 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1223 /* if no MDT is specified, use std way */
1224 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1226 /* test for no MDT idx case */
1227 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1228 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1232 /* we have an idx, read it */
1233 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1234 *idx = simple_strtoul(start, &end, 16);
1236 * no digit, no trailing :, negative value
1238 if (((*idx == 0) && (end == start)) ||
1239 (*end != ':') || (*idx < 0))
1244 /* bad format of mdt idx, we cannot return an error
1245 * to caller so we use hash algo */
1246 CERROR("Bad volatile file name format: %s\n",
1247 name + LUSTRE_VOLATILE_HDR_LEN);
1251 static inline int cli_brw_size(struct obd_device *obd)
1253 LASSERT(obd != NULL);
1254 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1258 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1259 * of the client should be increased accordingly to avoid sending fragmented
1260 * RPCs over the network when the client runs out of the maximum dirty space
1261 * when so many RPCs are being generated.
1263 static inline void client_adjust_max_dirty(struct client_obd *cli)
1266 if (cli->cl_dirty_max_pages <= 0) {
1267 cli->cl_dirty_max_pages =
1268 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1270 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1271 cli->cl_max_pages_per_rpc;
1273 if (dirty_max > cli->cl_dirty_max_pages)
1274 cli->cl_dirty_max_pages = dirty_max;
1277 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1278 cli->cl_dirty_max_pages = totalram_pages / 8;
1280 /* This value is exported to userspace through the max_dirty_mb
1281 * parameter. So we round up the number of pages to make it a round
1283 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1284 1 << (20 - PAGE_SHIFT));
1287 #endif /* __OBD_H */