4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #include <linux/spinlock.h>
42 #include <lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
50 #include <lustre_export.h>
51 #include <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <lustre_handles.h>
54 #include <lustre_intent.h>
56 #include <lustre_quota.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
90 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
92 /* obd info for a particular level (lov, osc). */
94 /* OBD_STATFS_* flags */
96 /* statfs data specific for every OSC, if needed at all. */
97 struct obd_statfs *oi_osfs;
98 /* An update callback which is called to update some data on upper
99 * level. E.g. it is used for update lsm->lsm_oinfo at every received
100 * request in osc level for enqueue requests. It is also possible to
101 * update some caller data from LOV layer if needed. */
102 obd_enqueue_update_f oi_cb_up;
106 struct list_head typ_chain;
107 struct obd_ops *typ_dt_ops;
108 struct md_ops *typ_md_ops;
109 struct proc_dir_entry *typ_procroot;
110 struct proc_dir_entry *typ_procsym;
111 __u32 typ_sym_filter;
114 struct lu_device_type *typ_lu;
115 spinlock_t obd_type_lock;
125 struct timeout_item {
126 enum timeout_event ti_event;
127 cfs_time_t ti_timeout;
130 struct list_head ti_obd_list;
131 struct list_head ti_chain;
134 #define OBD_MAX_RIF_DEFAULT 8
135 #define OBD_MAX_RIF_MAX 512
136 #define OSC_MAX_RIF_MAX 256
137 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
138 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
139 #define OSC_DEFAULT_RESENDS 10
141 /* possible values for fo_sync_lock_cancel */
143 NEVER_SYNC_ON_CANCEL = 0,
144 BLOCKING_SYNC_ON_CANCEL = 1,
145 ALWAYS_SYNC_ON_CANCEL = 2,
146 NUM_SYNC_ON_CANCEL_STATES
150 * Limit reply buffer size for striping data to one x86_64 page. This
151 * value is chosen to fit the striping data for common use cases while
152 * staying well below the limit at which the buffer must be backed by
153 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
156 #define OBD_MAX_DEFAULT_EA_SIZE 4096
161 struct rw_semaphore cl_sem;
162 struct obd_uuid cl_target_uuid;
163 struct obd_import *cl_import; /* ptlrpc connection state */
164 size_t cl_conn_count;
166 /* Cache maximum and default values for easize. This is
167 * strictly a performance optimization to minimize calls to
168 * obd_size_diskmd(). The default values are used to calculate the
169 * initial size of a request buffer. The ptlrpc layer will resize the
170 * buffer as needed to accommodate a larger reply from the
171 * server. The default values should be small enough to avoid wasted
172 * memory and excessive use of vmalloc(), yet large enough to avoid
173 * reallocating the buffer in the common use case. */
175 /* Default EA size for striping attributes. It is initialized at
176 * mount-time based on the default stripe width of the filesystem,
177 * then it tracks the largest observed EA size advertised by
178 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
179 __u32 cl_default_mds_easize;
181 /* Maximum possible EA size computed at mount-time based on
182 * the number of OSTs in the filesystem. May be increased at
183 * run-time if a larger observed size is advertised by the MDT. */
184 __u32 cl_max_mds_easize;
186 enum lustre_sec_part cl_sp_me;
187 enum lustre_sec_part cl_sp_to;
188 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
190 /* the grant values are protected by loi_list_lock below */
191 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
192 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
193 unsigned long cl_dirty_transit; /* dirty synchronous */
194 unsigned long cl_avail_grant; /* bytes of credit for ost */
195 unsigned long cl_lost_grant; /* lost credits (trunc) */
196 /* grant consumed for dirty pages */
197 unsigned long cl_dirty_grant;
199 /* since we allocate grant by blocks, we don't know how many grant will
200 * be used to add a page into cache. As a solution, we reserve maximum
201 * grant before trying to dirty a page and unreserve the rest.
202 * See osc_{reserve|unreserve}_grant for details. */
203 long cl_reserved_grant;
204 struct list_head cl_cache_waiters; /* waiting for cache/grant */
205 cfs_time_t cl_next_shrink_grant; /* jiffies */
206 struct list_head cl_grant_shrink_list; /* Timeout event list */
207 int cl_grant_shrink_interval; /* seconds */
209 /* A chunk is an optimal size used by osc_extent to determine
210 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
212 /* extent insertion metadata overhead to be accounted in grant,
214 unsigned int cl_grant_extent_tax;
215 /* maximum extent size, in number of pages */
216 unsigned int cl_max_extent_pages;
218 /* keep track of objects that have lois that contain pages which
219 * have been queued for async brw. this lock also protects the
220 * lists of osc_client_pages that hang off of the loi */
222 * ->cl_loi_list_lock protects consistency of
223 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
224 * ->ap_completion() call-backs are executed under this lock. As we
225 * cannot guarantee that these call-backs never block on all platforms
226 * (as a matter of fact they do block on Mac OS X), type of
227 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
228 * and blocking mutex on Mac OS X. (Alternative is to make this lock
229 * blocking everywhere, but we don't want to slow down fast-path of
230 * our main platform.)
232 * NB by Jinshan: though field names are still _loi_, but actually
233 * osc_object{}s are in the list.
235 spinlock_t cl_loi_list_lock;
236 struct list_head cl_loi_ready_list;
237 struct list_head cl_loi_hp_ready_list;
238 struct list_head cl_loi_write_list;
239 struct list_head cl_loi_read_list;
240 __u32 cl_r_in_flight;
241 __u32 cl_w_in_flight;
242 /* just a sum of the loi/lop pending numbers to be exported by /proc */
243 atomic_t cl_pending_w_pages;
244 atomic_t cl_pending_r_pages;
245 __u32 cl_max_pages_per_rpc;
246 __u32 cl_max_rpcs_in_flight;
247 struct obd_histogram cl_read_rpc_hist;
248 struct obd_histogram cl_write_rpc_hist;
249 struct obd_histogram cl_read_page_hist;
250 struct obd_histogram cl_write_page_hist;
251 struct obd_histogram cl_read_offset_hist;
252 struct obd_histogram cl_write_offset_hist;
254 /** LRU for osc caching pages */
255 struct cl_client_cache *cl_cache;
256 /** member of cl_cache->ccc_lru */
257 struct list_head cl_lru_osc;
258 /** # of available LRU slots left in the per-OSC cache.
259 * Available LRU slots are shared by all OSCs of the same file system,
260 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
261 atomic_long_t *cl_lru_left;
262 /** # of busy LRU pages. A page is considered busy if it's in writeback
263 * queue, or in transfer. Busy pages can't be discarded so they are not
265 atomic_long_t cl_lru_busy;
266 /** # of LRU pages in the cache for this client_obd */
267 atomic_long_t cl_lru_in_list;
268 /** # of threads are shrinking LRU cache. To avoid contention, it's not
269 * allowed to have multiple threads shrinking LRU cache. */
270 atomic_t cl_lru_shrinkers;
271 /** The time when this LRU cache was last used. */
272 time_t cl_lru_last_used;
273 /** stats: how many reclaims have happened for this client_obd.
274 * reclaim and shrink - shrink is async, voluntarily rebalancing;
275 * reclaim is sync, initiated by IO thread when the LRU slots are
277 __u64 cl_lru_reclaim;
278 /** List of LRU pages for this client_obd */
279 struct list_head cl_lru_list;
280 /** Lock for LRU page list */
281 spinlock_t cl_lru_list_lock;
282 /** # of unstable pages in this client_obd.
283 * An unstable page is a page state that WRITE RPC has finished but
284 * the transaction has NOT yet committed. */
285 atomic_long_t cl_unstable_count;
286 /** Link to osc_shrinker_list */
287 struct list_head cl_shrink_list;
289 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
290 atomic_t cl_destroy_in_flight;
291 wait_queue_head_t cl_destroy_waitq;
293 struct mdc_rpc_lock *cl_rpc_lock;
295 /* modify rpcs in flight
296 * currently used for metadata only */
297 spinlock_t cl_mod_rpcs_lock;
298 __u16 cl_max_mod_rpcs_in_flight;
299 __u16 cl_mod_rpcs_in_flight;
300 __u16 cl_close_rpcs_in_flight;
301 wait_queue_head_t cl_mod_rpcs_waitq;
302 unsigned long *cl_mod_tag_bitmap;
303 struct obd_histogram cl_mod_rpcs_hist;
306 struct mutex cl_mgc_mutex;
307 struct local_oid_storage *cl_mgc_los;
308 struct dt_object *cl_mgc_configs_dir;
309 atomic_t cl_mgc_refcount;
310 struct obd_export *cl_mgc_mgsexp;
312 /* checksumming for data sent over the network */
313 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
314 /* supported checksum types that are worked out at connect time */
315 __u32 cl_supp_cksum_types;
316 /* checksum algorithm to be used */
317 cksum_type_t cl_cksum_type;
319 /* also protected by the poorly named _loi_list_lock lock above */
320 struct osc_async_rc cl_ar;
322 /* sequence manager */
323 struct lu_client_seq *cl_seq;
325 atomic_t cl_resends; /* resend count */
327 /* ptlrpc work for writeback in ptlrpcd context */
328 void *cl_writeback_work;
330 /* hash tables for osc_quota_info */
331 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
333 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
340 struct echo_client_obd {
341 struct obd_export *ec_exp; /* the local connection to osc/lov */
343 struct list_head ec_objects;
344 struct list_head ec_locks;
348 /* Generic subset of OSTs */
350 __u32 *op_array; /* array of index of
352 unsigned int op_count; /* number of OSTs in the array */
353 unsigned int op_size; /* allocated size of lp_array */
354 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
357 /* allow statfs data caching for 1 second */
358 #define OBD_STATFS_CACHE_SECONDS 1
360 struct lov_tgt_desc {
361 struct list_head ltd_kill;
362 struct obd_uuid ltd_uuid;
363 struct obd_device *ltd_obd;
364 struct obd_export *ltd_exp;
366 __u32 ltd_index; /* index in lov_obd->tgts */
367 unsigned long ltd_active:1,/* is this target up for requests */
368 ltd_activate:1,/* should target be activated */
369 ltd_reap:1; /* should this target be deleted */
373 struct lov_desc desc;
374 struct lov_tgt_desc **lov_tgts; /* sparse array */
375 struct ost_pool lov_packed; /* all OSTs in a packed
377 struct mutex lov_lock;
378 struct obd_connect_data lov_ocd;
379 struct proc_dir_entry *targets_proc_entry;
380 atomic_t lov_refcount;
381 __u32 lov_death_row; /* tgts scheduled to be deleted */
382 __u32 lov_tgt_size; /* size of tgts array */
385 struct cfs_hash *lov_pools_hash_body; /* used for key access */
386 struct list_head lov_pool_list; /* used for sequential access */
387 struct proc_dir_entry *lov_pool_proc_entry;
388 enum lustre_sec_part lov_sp_me;
390 /* Cached LRU and unstable data from upper layer */
391 struct cl_client_cache *lov_cache;
393 struct rw_semaphore lov_notify_lock;
396 struct lmv_tgt_desc {
397 struct obd_uuid ltd_uuid;
398 struct obd_export *ltd_exp;
400 struct mutex ltd_fid_mutex;
401 unsigned long ltd_active:1; /* target up for requests */
404 enum placement_policy {
405 PLACEMENT_CHAR_POLICY = 0,
406 PLACEMENT_NID_POLICY = 1,
407 PLACEMENT_INVAL_POLICY = 2,
411 typedef enum placement_policy placement_policy_t;
415 struct lu_client_fld lmv_fld;
417 placement_policy_t lmv_placement;
418 struct lmv_desc desc;
419 struct obd_uuid cluuid;
420 struct obd_export *exp;
421 struct proc_dir_entry *targets_proc_entry;
423 struct mutex lmv_init_mutex;
428 __u32 tgts_size; /* size of tgts array */
429 struct lmv_tgt_desc **tgts;
431 struct obd_connect_data conn_data;
434 struct niobuf_local {
435 __u64 lnb_file_offset;
436 __u32 lnb_page_offset;
440 struct page *lnb_page;
444 #define LUSTRE_FLD_NAME "fld"
445 #define LUSTRE_SEQ_NAME "seq"
447 #define LUSTRE_MDD_NAME "mdd"
448 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
449 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
450 #define LUSTRE_VVP_NAME "vvp"
451 #define LUSTRE_LMV_NAME "lmv"
452 #define LUSTRE_SLP_NAME "slp"
453 #define LUSTRE_LOD_NAME "lod"
454 #define LUSTRE_OSP_NAME "osp"
455 #define LUSTRE_LWP_NAME "lwp"
457 /* obd device type names */
458 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
459 #define LUSTRE_MDS_NAME "mds"
460 #define LUSTRE_MDT_NAME "mdt"
461 #define LUSTRE_MDC_NAME "mdc"
462 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
463 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
464 #define LUSTRE_OSC_NAME "osc"
465 #define LUSTRE_LOV_NAME "lov"
466 #define LUSTRE_MGS_NAME "mgs"
467 #define LUSTRE_MGC_NAME "mgc"
469 #define LUSTRE_ECHO_NAME "obdecho"
470 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
471 #define LUSTRE_QMT_NAME "qmt"
473 /* Constant obd names (post-rename) */
474 #define LUSTRE_MDS_OBDNAME "MDS"
475 #define LUSTRE_OSS_OBDNAME "OSS"
476 #define LUSTRE_MGS_OBDNAME "MGS"
477 #define LUSTRE_MGC_OBDNAME "MGC"
479 static inline int is_lwp_on_mdt(char *name)
483 ptr = strrchr(name, '-');
485 CERROR("%s is not a obdname\n", name);
489 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
491 if (strncmp(ptr + 1, "MDT", 3) != 0)
494 while (*(--ptr) != '-' && ptr != name);
499 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
505 static inline int is_lwp_on_ost(char *name)
509 ptr = strrchr(name, '-');
511 CERROR("%s is not a obdname\n", name);
515 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
517 if (strncmp(ptr + 1, "OST", 3) != 0)
520 while (*(--ptr) != '-' && ptr != name);
525 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
532 * Events signalled through obd_notify() upcall-chain.
534 enum obd_notify_event {
537 /* Device connect start */
539 /* Device activated */
541 /* Device deactivated */
543 /* Device disconnected */
545 /* Connect data for import were changed */
548 OBD_NOTIFY_SYNC_NONBLOCK,
550 /* Configuration event */
552 /* Administratively deactivate/activate event */
553 OBD_NOTIFY_DEACTIVATE,
557 /* bit-mask flags for config events */
559 CONFIG_LOG = 0x1, /* finished processing config log */
560 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
561 CONFIG_TARGET = 0x4 /* one target is added */
565 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
566 * and liblustre being main examples).
568 struct obd_notify_upcall {
569 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
570 enum obd_notify_event ev, void *owner, void *data);
571 /* Opaque datum supplied by upper layer listener */
575 struct target_recovery_data {
576 svc_handler_t trd_recovery_handler;
577 pid_t trd_processing_task;
578 struct completion trd_starting;
579 struct completion trd_finishing;
582 struct obd_llog_group {
583 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
584 wait_queue_head_t olg_waitq;
586 struct mutex olg_cat_processing;
589 /* corresponds to one of the obd's */
590 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
593 struct obd_type *obd_type;
594 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
595 int obd_minor; /* device number: lctl dl */
596 struct lu_device *obd_lu_dev;
598 /* common and UUID name of this device */
599 struct obd_uuid obd_uuid;
600 char obd_name[MAX_OBD_NAME];
602 /* bitfield modification is protected by obd_dev_lock */
604 obd_attached:1, /* finished attach */
605 obd_set_up:1, /* finished setup */
606 obd_recovering:1, /* there are recoverable clients */
607 obd_abort_recovery:1, /* recovery expired */
608 obd_version_recov:1, /* obd uses version checking */
609 obd_replayable:1, /* recovery enabled; inform clients */
610 obd_no_transno:1, /* no committed-transno notification */
611 obd_no_recov:1, /* fail instead of retry messages */
612 obd_stopping:1, /* started cleanup */
613 obd_starting:1, /* started setup */
614 obd_force:1, /* cleanup with > 0 obd refcount */
615 obd_fail:1, /* cleanup with failover */
616 obd_no_conn:1, /* deny new connections */
617 obd_inactive:1, /* device active/inactive
618 * (for /proc/status only!!) */
619 obd_no_ir:1, /* no imperative recovery. */
620 obd_process_conf:1, /* device is processing mgs config */
621 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
623 /* use separate field as it is set in interrupt to don't mess with
624 * protection of other bits using _bh lock */
625 unsigned long obd_recovery_expired:1;
626 /* uuid-export hash body */
627 struct cfs_hash *obd_uuid_hash;
628 /* nid-export hash body */
629 struct cfs_hash *obd_nid_hash;
631 struct cfs_hash *obd_nid_stats_hash;
632 /* client_generation-export hash body */
633 struct cfs_hash *obd_gen_hash;
634 struct list_head obd_nid_stats;
635 struct list_head obd_exports;
636 struct list_head obd_unlinked_exports;
637 struct list_head obd_delayed_exports;
638 struct list_head obd_lwp_list;
639 atomic_t obd_refcount;
641 spinlock_t obd_nid_lock;
642 struct ldlm_namespace *obd_namespace;
643 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
644 /* a spinlock is OK for what we do now, may need a semaphore later */
645 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
646 spinlock_t obd_osfs_lock;
647 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
649 __u64 obd_last_committed;
650 struct mutex obd_dev_mutex;
651 struct lvfs_run_ctxt obd_lvfs_ctxt;
652 struct obd_llog_group obd_olg; /* default llog group */
653 struct obd_device *obd_observer;
654 struct rw_semaphore obd_observer_link_sem;
655 struct obd_notify_upcall obd_upcall;
656 struct obd_export *obd_self_export;
657 struct obd_export *obd_lwp_export;
658 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
659 struct list_head obd_exports_timed;
660 time_t obd_eviction_timer; /* for ping evictor */
662 int obd_max_recoverable_clients;
663 atomic_t obd_connected_clients;
664 int obd_stale_clients;
665 /* this lock protects all recovery list_heads, timer and
666 * obd_next_recovery_transno value */
667 spinlock_t obd_recovery_task_lock;
668 __u64 obd_next_recovery_transno;
669 int obd_replayed_requests;
670 int obd_requests_queued_for_recovery;
671 wait_queue_head_t obd_next_transno_waitq;
672 /* protected by obd_recovery_task_lock */
673 struct timer_list obd_recovery_timer;
675 time_t obd_recovery_start;
676 /* seconds, for lprocfs_status */
677 time_t obd_recovery_end;
678 int obd_recovery_time_hard;
679 int obd_recovery_timeout;
680 int obd_recovery_ir_factor;
682 /* new recovery stuff from CMD2 */
683 int obd_replayed_locks;
684 atomic_t obd_req_replay_clients;
685 atomic_t obd_lock_replay_clients;
686 struct target_recovery_data obd_recovery_data;
688 /* all lists are protected by obd_recovery_task_lock */
689 struct list_head obd_req_replay_queue;
690 struct list_head obd_lock_replay_queue;
691 struct list_head obd_final_req_queue;
694 #ifdef HAVE_SERVER_SUPPORT
695 struct obd_device_target obt;
696 struct filter_obd filter;
698 struct echo_obd echo;
700 struct client_obd cli;
701 struct echo_client_obd echo_client;
706 /* Fields used by LProcFS */
707 struct lprocfs_stats *obd_stats;
708 unsigned int obd_cntr_base;
710 unsigned int obd_md_cntr_base;
711 struct lprocfs_stats *obd_md_stats;
713 struct proc_dir_entry *obd_proc_entry;
714 struct proc_dir_entry *obd_proc_exports_entry;
715 struct proc_dir_entry *obd_svc_procroot;
716 struct lprocfs_stats *obd_svc_stats;
717 struct lprocfs_vars *obd_vars;
718 atomic_t obd_evict_inprogress;
719 wait_queue_head_t obd_evict_inprogress_waitq;
720 struct list_head obd_evict_list; /* protected with pet_lock */
723 * LDLM pool part. Save last calculated SLV and Limit.
725 rwlock_t obd_pool_lock;
729 int obd_conn_inprogress;
732 * List of outstanding class_incref()'s fo this OBD. For debugging. */
733 struct lu_ref obd_reference;
736 /* get/set_info keys */
737 #define KEY_ASYNC "async"
738 #define KEY_CHANGELOG_CLEAR "changelog_clear"
739 #define KEY_FID2PATH "fid2path"
740 #define KEY_CHECKSUM "checksum"
741 #define KEY_CLEAR_FS "clear_fs"
742 #define KEY_CONN_DATA "conn_data"
743 #define KEY_EVICT_BY_NID "evict_by_nid"
744 #define KEY_FIEMAP "fiemap"
745 #define KEY_FLUSH_CTX "flush_ctx"
746 #define KEY_GRANT_SHRINK "grant_shrink"
747 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
748 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
749 #define KEY_INTERMDS "inter_mds"
750 #define KEY_LAST_ID "last_id"
751 #define KEY_LAST_FID "last_fid"
752 #define KEY_MAX_EASIZE "max_easize"
753 #define KEY_DEFAULT_EASIZE "default_easize"
754 #define KEY_MGSSEC "mgssec"
755 #define KEY_READ_ONLY "read-only"
756 #define KEY_REGISTER_TARGET "register_target"
757 #define KEY_SET_FS "set_fs"
758 #define KEY_TGT_COUNT "tgt_count"
759 /* KEY_SET_INFO in lustre_idl.h */
760 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
762 #define KEY_CACHE_SET "cache_set"
763 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
764 #define KEY_OSP_CONNECTED "osp_connected"
768 static inline int it_to_lock_mode(struct lookup_intent *it)
770 /* CREAT needs to be tested before open (both could be set) */
771 if (it->it_op & IT_CREAT)
773 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
776 else if (it->it_op & IT_READDIR)
778 else if (it->it_op & IT_GETXATTR)
780 else if (it->it_op & IT_SETXATTR)
783 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
788 MF_MDC_CANCEL_FID1 = 1 << 0,
789 MF_MDC_CANCEL_FID2 = 1 << 1,
790 MF_MDC_CANCEL_FID3 = 1 << 2,
791 MF_MDC_CANCEL_FID4 = 1 << 3,
792 MF_GET_MDT_IDX = 1 << 4,
796 CLI_SET_MEA = 1 << 0,
797 CLI_RM_ENTRY = 1 << 1,
800 CLI_MIGRATE = 1 << 4,
804 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
805 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
806 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
807 struct lu_fid op_fid4; /* to the operation locks. */
808 u32 op_mds; /* what mds server open will go to */
809 struct lustre_handle op_handle;
814 struct lmv_stripe_md *op_mea1;
815 struct lmv_stripe_md *op_mea2;
816 __u32 op_suppgids[2];
823 /* iattr fields and blocks. */
824 struct iattr op_attr;
825 loff_t op_attr_blocks;
826 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
827 __u64 op_valid; /* OBD_MD_* */
829 enum md_op_flags op_flags;
831 /* Various operation flags. */
832 enum mds_op_bias op_bias;
834 /* Used by readdir */
835 unsigned int op_max_pages;
837 /* used to transfer info between the stacks of MD client
838 * see enum op_cli_flags */
839 enum md_cli_flags op_cli_flags;
841 /* File object data version for HSM release, on client */
842 __u64 op_data_version;
843 struct lustre_handle op_lease_handle;
845 /* default stripe offset */
846 __u32 op_default_stripe_offset;
850 int (*md_blocking_ast)(struct ldlm_lock *lock,
851 struct ldlm_lock_desc *desc,
852 void *data, int flag);
855 struct md_enqueue_info;
856 /* metadata stat-ahead */
857 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
858 struct md_enqueue_info *minfo,
861 struct md_enqueue_info {
862 struct md_op_data mi_data;
863 struct lookup_intent mi_it;
864 struct lustre_handle mi_lockh;
865 struct inode *mi_dir;
866 struct ldlm_enqueue_info mi_einfo;
867 md_enqueue_cb_t mi_cb;
872 struct module *o_owner;
873 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
874 void *karg, void __user *uarg);
875 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
876 __u32 keylen, void *key, __u32 *vallen, void *val);
877 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
878 __u32 keylen, void *key,
879 __u32 vallen, void *val,
880 struct ptlrpc_request_set *set);
881 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
882 int (*o_precleanup)(struct obd_device *dev);
883 int (*o_cleanup)(struct obd_device *dev);
884 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
885 int (*o_postrecov)(struct obd_device *dev);
886 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
888 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
889 /* connect to the target device with given connection
890 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
891 * granted by the target, which are guaranteed to be a subset of flags
892 * asked for. If @ocd == NULL, use default parameters. */
893 int (*o_connect)(const struct lu_env *env,
894 struct obd_export **exp, struct obd_device *src,
895 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
897 int (*o_reconnect)(const struct lu_env *env,
898 struct obd_export *exp, struct obd_device *src,
899 struct obd_uuid *cluuid,
900 struct obd_connect_data *ocd,
902 int (*o_disconnect)(struct obd_export *exp);
904 /* Initialize/finalize fids infrastructure. */
905 int (*o_fid_init)(struct obd_device *obd,
906 struct obd_export *exp, enum lu_cli_type type);
907 int (*o_fid_fini)(struct obd_device *obd);
909 /* Allocate new fid according to passed @hint. */
910 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
911 struct lu_fid *fid, struct md_op_data *op_data);
914 * Object with @fid is getting deleted, we may want to do something
917 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
918 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
919 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
920 __u64 max_age, struct ptlrpc_request_set *set);
921 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
923 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
925 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
927 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
929 int (*o_preprw)(const struct lu_env *env, int cmd,
930 struct obd_export *exp, struct obdo *oa, int objcount,
931 struct obd_ioobj *obj, struct niobuf_remote *remote,
932 int *nr_pages, struct niobuf_local *local);
933 int (*o_commitrw)(const struct lu_env *env, int cmd,
934 struct obd_export *exp, struct obdo *oa,
935 int objcount, struct obd_ioobj *obj,
936 struct niobuf_remote *remote, int pages,
937 struct niobuf_local *local, int rc);
938 int (*o_init_export)(struct obd_export *exp);
939 int (*o_destroy_export)(struct obd_export *exp);
941 int (*o_import_event)(struct obd_device *, struct obd_import *,
942 enum obd_import_event);
944 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
945 enum obd_notify_event ev, void *data);
947 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
948 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
951 int (*o_quotactl)(struct obd_device *, struct obd_export *,
952 struct obd_quotactl *);
954 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
957 int (*o_pool_new)(struct obd_device *obd, char *poolname);
958 int (*o_pool_del)(struct obd_device *obd, char *poolname);
959 int (*o_pool_add)(struct obd_device *obd, char *poolname,
961 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
963 void (*o_getref)(struct obd_device *obd);
964 void (*o_putref)(struct obd_device *obd);
966 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
967 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
968 * Also, add a wrapper function in include/linux/obd_class.h. */
973 struct mdt_body *body;
974 struct lu_buf layout;
975 struct lmv_stripe_md *lmv;
976 #ifdef CONFIG_FS_POSIX_ACL
977 struct posix_acl *posix_acl;
979 struct mdt_remote_perm *remote_perm;
982 struct md_open_data {
983 struct obd_client_handle *mod_och;
984 struct ptlrpc_request *mod_open_req;
985 struct ptlrpc_request *mod_close_req;
986 atomic_t mod_refcount;
990 struct obd_client_handle {
991 struct lustre_handle och_fh;
992 struct lu_fid och_fid;
993 struct md_open_data *och_mod;
994 struct lustre_handle och_lease_handle; /* open lock for lease */
999 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1001 struct lookup_intent;
1005 /* Every operation from MD_STATS_FIRST_OP up to and including
1006 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1007 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1008 * Operations after MD_STATS_LAST_OP are excluded from stats.
1009 * There are a few reasons for doing this: we prune the 17
1010 * counters which will be of minimal use in understanding
1011 * metadata utilization, we save memory by allocating 15
1012 * instead of 32 counters, we save cycles by not counting.
1014 * MD_STATS_FIRST_OP must be the first member of md_ops.
1016 #define MD_STATS_FIRST_OP m_close
1017 int (*m_close)(struct obd_export *, struct md_op_data *,
1018 struct md_open_data *, struct ptlrpc_request **);
1020 int (*m_create)(struct obd_export *, struct md_op_data *,
1021 const void *, size_t, umode_t, uid_t, gid_t,
1022 cfs_cap_t, __u64, struct ptlrpc_request **);
1024 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1025 const union ldlm_policy_data *,
1026 struct lookup_intent *, struct md_op_data *,
1027 struct lustre_handle *, __u64);
1029 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1030 struct ptlrpc_request **);
1032 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1033 struct lookup_intent *,
1034 struct ptlrpc_request **,
1035 ldlm_blocking_callback, __u64);
1037 int (*m_link)(struct obd_export *, struct md_op_data *,
1038 struct ptlrpc_request **);
1040 int (*m_rename)(struct obd_export *, struct md_op_data *,
1041 const char *, size_t, const char *, size_t,
1042 struct ptlrpc_request **);
1044 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1045 size_t , struct ptlrpc_request **);
1047 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1048 struct ptlrpc_request **);
1050 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1051 struct md_callback *cb_op, __u64 hash_offset,
1052 struct page **ppage);
1054 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1055 struct ptlrpc_request **);
1057 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1058 u64, const char *, const char *, int, int, int, u32,
1059 struct ptlrpc_request **);
1061 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1062 u64, const char *, const char *, int, int, int,
1063 struct ptlrpc_request **);
1065 int (*m_intent_getattr_async)(struct obd_export *,
1066 struct md_enqueue_info *);
1068 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1069 struct lu_fid *, __u64 *bits);
1071 #define MD_STATS_LAST_OP m_revalidate_lock
1073 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1074 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1076 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1077 struct ptlrpc_request **);
1079 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1081 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1082 struct obd_export *, struct obd_export *,
1083 struct lustre_md *);
1085 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1087 int (*m_merge_attr)(struct obd_export *,
1088 const struct lmv_stripe_md *lsm,
1089 struct cl_attr *attr, ldlm_blocking_callback);
1091 int (*m_set_open_replay_data)(struct obd_export *,
1092 struct obd_client_handle *,
1093 struct lookup_intent *);
1095 int (*m_clear_open_replay_data)(struct obd_export *,
1096 struct obd_client_handle *);
1098 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1100 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1101 const struct lu_fid *, enum ldlm_type,
1102 union ldlm_policy_data *, enum ldlm_mode,
1103 struct lustre_handle *);
1105 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1106 union ldlm_policy_data *, enum ldlm_mode,
1107 enum ldlm_cancel_flags flags, void *opaque);
1109 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1110 u32, struct ptlrpc_request **);
1112 int (*m_get_fid_from_lsm)(struct obd_export *,
1113 const struct lmv_stripe_md *,
1114 const char *name, int namelen,
1115 struct lu_fid *fid);
1116 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1117 const union lmv_mds_md *lmv, size_t lmv_size);
1120 static inline struct md_open_data *obd_mod_alloc(void)
1122 struct md_open_data *mod;
1126 atomic_set(&mod->mod_refcount, 1);
1130 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1131 #define obd_mod_put(mod) \
1133 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1134 if ((mod)->mod_open_req) \
1135 ptlrpc_req_finished((mod)->mod_open_req); \
1136 OBD_FREE_PTR(mod); \
1140 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1141 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1143 /* return 1 if client should be resend request */
1144 static inline int client_should_resend(int resend, struct client_obd *cli)
1146 return atomic_read(&cli->cl_resends) ?
1147 atomic_read(&cli->cl_resends) > resend : 1;
1151 * Return device name for this device
1153 * XXX: lu_device is declared before obd_device, while a pointer pointing
1154 * back to obd_device in lu_device, so this helper function defines here
1155 * instead of in lu_object.h
1157 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1159 return lu_dev->ld_obd->obd_name;
1162 static inline bool filename_is_volatile(const char *name, size_t namelen,
1168 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1171 /* caller does not care of idx */
1175 /* volatile file, the MDT can be set from name */
1176 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1177 /* if no MDT is specified, use std way */
1178 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1180 /* test for no MDT idx case */
1181 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1182 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1186 /* we have an idx, read it */
1187 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1188 *idx = simple_strtoul(start, &end, 16);
1190 * no digit, no trailing :, negative value
1192 if (((*idx == 0) && (end == start)) ||
1193 (*end != ':') || (*idx < 0))
1198 /* bad format of mdt idx, we cannot return an error
1199 * to caller so we use hash algo */
1200 CERROR("Bad volatile file name format: %s\n",
1201 name + LUSTRE_VOLATILE_HDR_LEN);
1205 static inline int cli_brw_size(struct obd_device *obd)
1207 LASSERT(obd != NULL);
1208 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1211 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1212 * of the client should be increased accordingly to avoid sending fragmented
1213 * RPCs over the network when the client runs out of the maximum dirty space
1214 * when so many RPCs are being generated.
1216 static inline void client_adjust_max_dirty(struct client_obd *cli)
1219 if (cli->cl_dirty_max_pages <= 0)
1220 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1221 >> PAGE_CACHE_SHIFT;
1223 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1224 cli->cl_max_pages_per_rpc;
1226 if (dirty_max > cli->cl_dirty_max_pages)
1227 cli->cl_dirty_max_pages = dirty_max;
1230 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1231 cli->cl_dirty_max_pages = totalram_pages / 8;
1234 #endif /* __OBD_H */