4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/spinlock.h>
38 #include <lustre/lustre_idl.h>
39 #include <lustre_lib.h>
40 #include <libcfs/bitmap.h>
41 #ifdef HAVE_SERVER_SUPPORT
42 # include <lu_target.h>
43 # include <obd_target.h>
46 #include <lustre_export.h>
47 #include <lustre_fid.h>
48 #include <lustre_fld.h>
49 #include <lustre_handles.h>
50 #include <lustre_intent.h>
52 #include <lustre_quota.h>
54 #define MAX_OBD_DEVICES 8192
62 struct lov_oinfo { /* per-stripe data structure */
63 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
64 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
65 int loi_ost_gen; /* generation of this loi_ost_idx */
67 unsigned long loi_kms_valid:1;
68 __u64 loi_kms; /* known minimum size */
69 struct ost_lvb loi_lvb;
70 struct osc_async_rc loi_ar;
73 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
76 oinfo->loi_kms_valid = 1;
82 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
84 /* obd info for a particular level (lov, osc). */
86 /* OBD_STATFS_* flags */
88 /* statfs data specific for every OSC, if needed at all. */
89 struct obd_statfs *oi_osfs;
90 /* An update callback which is called to update some data on upper
91 * level. E.g. it is used for update lsm->lsm_oinfo at every received
92 * request in osc level for enqueue requests. It is also possible to
93 * update some caller data from LOV layer if needed. */
94 obd_enqueue_update_f oi_cb_up;
98 struct list_head typ_chain;
99 struct obd_ops *typ_dt_ops;
100 struct md_ops *typ_md_ops;
101 struct proc_dir_entry *typ_procroot;
102 struct proc_dir_entry *typ_procsym;
103 __u32 typ_sym_filter;
106 struct lu_device_type *typ_lu;
107 spinlock_t obd_type_lock;
108 struct kobject *typ_kobj;
118 struct timeout_item {
119 enum timeout_event ti_event;
120 cfs_time_t ti_timeout;
123 struct list_head ti_obd_list;
124 struct list_head ti_chain;
127 #define OBD_MAX_RIF_DEFAULT 8
128 #define OBD_MAX_RIF_MAX 512
129 #define OSC_MAX_RIF_MAX 256
130 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
131 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
132 #define OSC_DEFAULT_RESENDS 10
134 /* possible values for fo_sync_lock_cancel */
136 NEVER_SYNC_ON_CANCEL = 0,
137 BLOCKING_SYNC_ON_CANCEL = 1,
138 ALWAYS_SYNC_ON_CANCEL = 2,
139 NUM_SYNC_ON_CANCEL_STATES
143 * Limit reply buffer size for striping data to one x86_64 page. This
144 * value is chosen to fit the striping data for common use cases while
145 * staying well below the limit at which the buffer must be backed by
146 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
149 #define OBD_MAX_DEFAULT_EA_SIZE 4096
151 enum obd_cl_sem_lock_class {
160 struct rw_semaphore cl_sem;
161 struct obd_uuid cl_target_uuid;
162 struct obd_import *cl_import; /* ptlrpc connection state */
163 size_t cl_conn_count;
165 /* Cache maximum and default values for easize. This is
166 * strictly a performance optimization to minimize calls to
167 * obd_size_diskmd(). The default values are used to calculate the
168 * initial size of a request buffer. The ptlrpc layer will resize the
169 * buffer as needed to accommodate a larger reply from the
170 * server. The default values should be small enough to avoid wasted
171 * memory and excessive use of vmalloc(), yet large enough to avoid
172 * reallocating the buffer in the common use case. */
174 /* Default EA size for striping attributes. It is initialized at
175 * mount-time based on the default stripe width of the filesystem,
176 * then it tracks the largest observed EA size advertised by
177 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
178 __u32 cl_default_mds_easize;
180 /* Maximum possible EA size computed at mount-time based on
181 * the number of OSTs in the filesystem. May be increased at
182 * run-time if a larger observed size is advertised by the MDT. */
183 __u32 cl_max_mds_easize;
185 enum lustre_sec_part cl_sp_me;
186 enum lustre_sec_part cl_sp_to;
187 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
189 /* the grant values are protected by loi_list_lock below */
190 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
191 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
192 unsigned long cl_dirty_transit; /* dirty synchronous */
193 unsigned long cl_avail_grant; /* bytes of credit for ost */
194 unsigned long cl_lost_grant; /* lost credits (trunc) */
195 /* grant consumed for dirty pages */
196 unsigned long cl_dirty_grant;
198 /* since we allocate grant by blocks, we don't know how many grant will
199 * be used to add a page into cache. As a solution, we reserve maximum
200 * grant before trying to dirty a page and unreserve the rest.
201 * See osc_{reserve|unreserve}_grant for details. */
202 long cl_reserved_grant;
203 struct list_head cl_cache_waiters; /* waiting for cache/grant */
204 cfs_time_t cl_next_shrink_grant; /* jiffies */
205 struct list_head cl_grant_shrink_list; /* Timeout event list */
206 int cl_grant_shrink_interval; /* seconds */
208 /* A chunk is an optimal size used by osc_extent to determine
209 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
211 /* extent insertion metadata overhead to be accounted in grant,
213 unsigned int cl_grant_extent_tax;
214 /* maximum extent size, in number of pages */
215 unsigned int cl_max_extent_pages;
217 /* keep track of objects that have lois that contain pages which
218 * have been queued for async brw. this lock also protects the
219 * lists of osc_client_pages that hang off of the loi */
221 * ->cl_loi_list_lock protects consistency of
222 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
223 * ->ap_completion() call-backs are executed under this lock. As we
224 * cannot guarantee that these call-backs never block on all platforms
225 * (as a matter of fact they do block on Mac OS X), type of
226 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
227 * and blocking mutex on Mac OS X. (Alternative is to make this lock
228 * blocking everywhere, but we don't want to slow down fast-path of
229 * our main platform.)
231 * NB by Jinshan: though field names are still _loi_, but actually
232 * osc_object{}s are in the list.
234 spinlock_t cl_loi_list_lock;
235 struct list_head cl_loi_ready_list;
236 struct list_head cl_loi_hp_ready_list;
237 struct list_head cl_loi_write_list;
238 struct list_head cl_loi_read_list;
239 __u32 cl_r_in_flight;
240 __u32 cl_w_in_flight;
241 /* just a sum of the loi/lop pending numbers to be exported by /proc */
242 atomic_t cl_pending_w_pages;
243 atomic_t cl_pending_r_pages;
244 __u32 cl_max_pages_per_rpc;
245 __u32 cl_max_rpcs_in_flight;
246 struct obd_histogram cl_read_rpc_hist;
247 struct obd_histogram cl_write_rpc_hist;
248 struct obd_histogram cl_read_page_hist;
249 struct obd_histogram cl_write_page_hist;
250 struct obd_histogram cl_read_offset_hist;
251 struct obd_histogram cl_write_offset_hist;
253 /** LRU for osc caching pages */
254 struct cl_client_cache *cl_cache;
255 /** member of cl_cache->ccc_lru */
256 struct list_head cl_lru_osc;
257 /** # of available LRU slots left in the per-OSC cache.
258 * Available LRU slots are shared by all OSCs of the same file system,
259 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
260 atomic_long_t *cl_lru_left;
261 /** # of busy LRU pages. A page is considered busy if it's in writeback
262 * queue, or in transfer. Busy pages can't be discarded so they are not
264 atomic_long_t cl_lru_busy;
265 /** # of LRU pages in the cache for this client_obd */
266 atomic_long_t cl_lru_in_list;
267 /** # of threads are shrinking LRU cache. To avoid contention, it's not
268 * allowed to have multiple threads shrinking LRU cache. */
269 atomic_t cl_lru_shrinkers;
270 /** The time when this LRU cache was last used. */
271 time64_t cl_lru_last_used;
272 /** stats: how many reclaims have happened for this client_obd.
273 * reclaim and shrink - shrink is async, voluntarily rebalancing;
274 * reclaim is sync, initiated by IO thread when the LRU slots are
276 __u64 cl_lru_reclaim;
277 /** List of LRU pages for this client_obd */
278 struct list_head cl_lru_list;
279 /** Lock for LRU page list */
280 spinlock_t cl_lru_list_lock;
281 /** # of unstable pages in this client_obd.
282 * An unstable page is a page state that WRITE RPC has finished but
283 * the transaction has NOT yet committed. */
284 atomic_long_t cl_unstable_count;
285 /** Link to osc_shrinker_list */
286 struct list_head cl_shrink_list;
288 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
289 atomic_t cl_destroy_in_flight;
290 wait_queue_head_t cl_destroy_waitq;
292 struct mdc_rpc_lock *cl_rpc_lock;
294 /* modify rpcs in flight
295 * currently used for metadata only */
296 spinlock_t cl_mod_rpcs_lock;
297 __u16 cl_max_mod_rpcs_in_flight;
298 __u16 cl_mod_rpcs_in_flight;
299 __u16 cl_close_rpcs_in_flight;
300 wait_queue_head_t cl_mod_rpcs_waitq;
301 unsigned long *cl_mod_tag_bitmap;
302 struct obd_histogram cl_mod_rpcs_hist;
305 struct mutex cl_mgc_mutex;
306 struct local_oid_storage *cl_mgc_los;
307 struct dt_object *cl_mgc_configs_dir;
308 atomic_t cl_mgc_refcount;
309 struct obd_export *cl_mgc_mgsexp;
311 /* checksumming for data sent over the network */
312 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
313 cl_checksum_dump:1; /* same */
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;
324 struct rw_semaphore cl_seq_rwsem;
326 atomic_t cl_resends; /* resend count */
328 /* ptlrpc work for writeback in ptlrpcd context */
329 void *cl_writeback_work;
331 /* hash tables for osc_quota_info */
332 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
333 /* Links to the global list of registered changelog devices */
334 struct list_head cl_chg_dev_linkage;
336 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
343 struct echo_client_obd {
344 struct obd_export *ec_exp; /* the local connection to osc/lov */
346 struct list_head ec_objects;
347 struct list_head ec_locks;
351 /* Generic subset of OSTs */
353 __u32 *op_array; /* array of index of
355 unsigned int op_count; /* number of OSTs in the array */
356 unsigned int op_size; /* allocated size of lp_array */
357 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
360 /* allow statfs data caching for 1 second */
361 #define OBD_STATFS_CACHE_SECONDS 1
363 struct lov_tgt_desc {
364 struct list_head ltd_kill;
365 struct obd_uuid ltd_uuid;
366 struct obd_device *ltd_obd;
367 struct obd_export *ltd_exp;
369 __u32 ltd_index; /* index in lov_obd->tgts */
370 unsigned long ltd_active:1,/* is this target up for requests */
371 ltd_activate:1,/* should target be activated */
372 ltd_reap:1; /* should this target be deleted */
376 struct lov_desc desc;
377 struct lov_tgt_desc **lov_tgts; /* sparse array */
378 struct ost_pool lov_packed; /* all OSTs in a packed
380 struct mutex lov_lock;
381 struct obd_connect_data lov_ocd;
382 struct proc_dir_entry *targets_proc_entry;
383 atomic_t lov_refcount;
384 __u32 lov_death_row; /* tgts scheduled to be deleted */
385 __u32 lov_tgt_size; /* size of tgts array */
388 struct cfs_hash *lov_pools_hash_body; /* used for key access */
389 struct list_head lov_pool_list; /* used for sequential access */
390 struct proc_dir_entry *lov_pool_proc_entry;
391 enum lustre_sec_part lov_sp_me;
393 /* Cached LRU and unstable data from upper layer */
394 struct cl_client_cache *lov_cache;
396 struct rw_semaphore lov_notify_lock;
399 struct lmv_tgt_desc {
400 struct obd_uuid ltd_uuid;
401 struct obd_export *ltd_exp;
403 struct mutex ltd_fid_mutex;
404 unsigned long ltd_active:1; /* target up for requests */
408 struct lu_client_fld lmv_fld;
410 struct lmv_desc desc;
411 struct obd_uuid cluuid;
412 struct proc_dir_entry *targets_proc_entry;
414 struct mutex lmv_init_mutex;
419 __u32 tgts_size; /* size of tgts array */
420 struct lmv_tgt_desc **tgts;
422 struct obd_connect_data conn_data;
425 struct niobuf_local {
426 __u64 lnb_file_offset;
427 __u32 lnb_page_offset;
431 struct page *lnb_page;
435 struct tgt_thread_big_cache {
436 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
439 #define LUSTRE_FLD_NAME "fld"
440 #define LUSTRE_SEQ_NAME "seq"
442 #define LUSTRE_MDD_NAME "mdd"
443 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
444 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
445 #define LUSTRE_VVP_NAME "vvp"
446 #define LUSTRE_LMV_NAME "lmv"
447 #define LUSTRE_SLP_NAME "slp"
448 #define LUSTRE_LOD_NAME "lod"
449 #define LUSTRE_OSP_NAME "osp"
450 #define LUSTRE_LWP_NAME "lwp"
452 /* obd device type names */
453 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
454 #define LUSTRE_MDS_NAME "mds"
455 #define LUSTRE_MDT_NAME "mdt"
456 #define LUSTRE_MDC_NAME "mdc"
457 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
458 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
459 #define LUSTRE_OSC_NAME "osc"
460 #define LUSTRE_LOV_NAME "lov"
461 #define LUSTRE_MGS_NAME "mgs"
462 #define LUSTRE_MGC_NAME "mgc"
464 #define LUSTRE_ECHO_NAME "obdecho"
465 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
466 #define LUSTRE_QMT_NAME "qmt"
468 /* Constant obd names (post-rename) */
469 #define LUSTRE_MDS_OBDNAME "MDS"
470 #define LUSTRE_OSS_OBDNAME "OSS"
471 #define LUSTRE_MGS_OBDNAME "MGS"
472 #define LUSTRE_MGC_OBDNAME "MGC"
474 static inline int is_lwp_on_mdt(char *name)
478 ptr = strrchr(name, '-');
480 CERROR("%s is not a obdname\n", name);
484 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
486 if (strncmp(ptr + 1, "MDT", 3) != 0)
489 while (*(--ptr) != '-' && ptr != name);
494 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
500 static inline int is_lwp_on_ost(char *name)
504 ptr = strrchr(name, '-');
506 CERROR("%s is not a obdname\n", name);
510 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
512 if (strncmp(ptr + 1, "OST", 3) != 0)
515 while (*(--ptr) != '-' && ptr != name);
520 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
527 * Events signalled through obd_notify() upcall-chain.
529 enum obd_notify_event {
530 /* Device connect start */
532 /* Device activated */
534 /* Device deactivated */
536 /* Connect data for import were changed */
538 /* Administratively deactivate/activate event */
539 OBD_NOTIFY_DEACTIVATE,
544 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
545 * and liblustre being main examples).
547 struct obd_notify_upcall {
548 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
549 enum obd_notify_event ev, void *owner);
550 /* Opaque datum supplied by upper layer listener */
554 struct target_recovery_data {
555 svc_handler_t trd_recovery_handler;
556 pid_t trd_processing_task;
557 struct completion trd_starting;
558 struct completion trd_finishing;
561 struct obd_llog_group {
562 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
563 wait_queue_head_t olg_waitq;
565 struct mutex olg_cat_processing;
568 /* corresponds to one of the obd's */
569 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
572 struct obd_type *obd_type;
573 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
574 int obd_minor; /* device number: lctl dl */
575 struct lu_device *obd_lu_dev;
577 /* common and UUID name of this device */
578 struct obd_uuid obd_uuid;
579 char obd_name[MAX_OBD_NAME];
581 /* bitfield modification is protected by obd_dev_lock */
583 obd_attached:1, /* finished attach */
584 obd_set_up:1, /* finished setup */
585 obd_recovering:1, /* there are recoverable clients */
586 obd_abort_recovery:1, /* recovery expired */
587 obd_version_recov:1, /* obd uses version checking */
588 obd_replayable:1, /* recovery enabled; inform clients */
589 obd_no_transno:1, /* no committed-transno notification */
590 obd_no_recov:1, /* fail instead of retry messages */
591 obd_stopping:1, /* started cleanup */
592 obd_starting:1, /* started setup */
593 obd_force:1, /* cleanup with > 0 obd refcount */
594 obd_fail:1, /* cleanup with failover */
595 obd_no_conn:1, /* deny new connections */
596 obd_inactive:1, /* device active/inactive
597 * (for /proc/status only!!) */
598 obd_no_ir:1, /* no imperative recovery. */
599 obd_process_conf:1, /* device is processing mgs config */
600 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
601 obd_checksum_dump:1; /* dump pages upon cksum error */
603 /* use separate field as it is set in interrupt to don't mess with
604 * protection of other bits using _bh lock */
605 unsigned long obd_recovery_expired:1;
606 /* uuid-export hash body */
607 struct cfs_hash *obd_uuid_hash;
608 /* nid-export hash body */
609 struct cfs_hash *obd_nid_hash;
611 struct cfs_hash *obd_nid_stats_hash;
612 /* client_generation-export hash body */
613 struct cfs_hash *obd_gen_hash;
614 struct list_head obd_nid_stats;
615 struct list_head obd_exports;
616 struct list_head obd_unlinked_exports;
617 struct list_head obd_delayed_exports;
618 struct list_head obd_lwp_list;
619 atomic_t obd_refcount;
621 spinlock_t obd_nid_lock;
622 struct ldlm_namespace *obd_namespace;
623 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
624 /* a spinlock is OK for what we do now, may need a semaphore later */
625 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
626 spinlock_t obd_osfs_lock;
627 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
629 __u64 obd_last_committed;
630 struct mutex obd_dev_mutex;
631 struct lvfs_run_ctxt obd_lvfs_ctxt;
632 struct obd_llog_group obd_olg; /* default llog group */
633 struct obd_device *obd_observer;
634 struct rw_semaphore obd_observer_link_sem;
635 struct obd_notify_upcall obd_upcall;
636 struct obd_export *obd_self_export;
637 struct obd_export *obd_lwp_export;
638 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
639 struct list_head obd_exports_timed;
640 time_t obd_eviction_timer; /* for ping evictor */
642 int obd_max_recoverable_clients;
643 atomic_t obd_connected_clients;
644 int obd_stale_clients;
645 /* this lock protects all recovery list_heads, timer and
646 * obd_next_recovery_transno value */
647 spinlock_t obd_recovery_task_lock;
648 __u64 obd_next_recovery_transno;
649 int obd_replayed_requests;
650 int obd_requests_queued_for_recovery;
651 wait_queue_head_t obd_next_transno_waitq;
652 /* protected by obd_recovery_task_lock */
653 struct timer_list obd_recovery_timer;
655 time64_t obd_recovery_start;
656 /* seconds, for lprocfs_status */
657 time64_t obd_recovery_end;
658 time64_t obd_recovery_time_hard;
659 time64_t obd_recovery_timeout;
660 int obd_recovery_ir_factor;
662 /* new recovery stuff from CMD2 */
663 int obd_replayed_locks;
664 atomic_t obd_req_replay_clients;
665 atomic_t obd_lock_replay_clients;
666 struct target_recovery_data obd_recovery_data;
668 /* all lists are protected by obd_recovery_task_lock */
669 struct list_head obd_req_replay_queue;
670 struct list_head obd_lock_replay_queue;
671 struct list_head obd_final_req_queue;
674 #ifdef HAVE_SERVER_SUPPORT
675 struct obd_device_target obt;
676 struct filter_obd filter;
678 struct echo_obd echo;
680 struct client_obd cli;
681 struct echo_client_obd echo_client;
686 /* Fields used by LProcFS */
687 struct lprocfs_stats *obd_stats;
688 unsigned int obd_cntr_base;
690 unsigned int obd_md_cntr_base;
691 struct lprocfs_stats *obd_md_stats;
693 struct proc_dir_entry *obd_proc_entry;
694 struct proc_dir_entry *obd_proc_exports_entry;
695 struct proc_dir_entry *obd_svc_procroot;
696 struct lprocfs_stats *obd_svc_stats;
697 struct attribute_group *obd_attrs;
698 struct lprocfs_vars *obd_vars;
699 atomic_t obd_evict_inprogress;
700 wait_queue_head_t obd_evict_inprogress_waitq;
701 struct list_head obd_evict_list; /* protected with pet_lock */
704 * LDLM pool part. Save last calculated SLV and Limit.
706 rwlock_t obd_pool_lock;
710 int obd_conn_inprogress;
713 * List of outstanding class_incref()'s fo this OBD. For debugging. */
714 struct lu_ref obd_reference;
716 struct kobject obd_kobj; /* sysfs object */
717 struct completion obd_kobj_unregister;
720 /* get/set_info keys */
721 #define KEY_ASYNC "async"
722 #define KEY_CHANGELOG_CLEAR "changelog_clear"
723 #define KEY_FID2PATH "fid2path"
724 #define KEY_CHECKSUM "checksum"
725 #define KEY_CLEAR_FS "clear_fs"
726 #define KEY_CONN_DATA "conn_data"
727 #define KEY_EVICT_BY_NID "evict_by_nid"
728 #define KEY_FIEMAP "fiemap"
729 #define KEY_FLUSH_CTX "flush_ctx"
730 #define KEY_GRANT_SHRINK "grant_shrink"
731 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
732 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
733 #define KEY_INTERMDS "inter_mds"
734 #define KEY_LAST_ID "last_id"
735 #define KEY_LAST_FID "last_fid"
736 #define KEY_MAX_EASIZE "max_easize"
737 #define KEY_DEFAULT_EASIZE "default_easize"
738 #define KEY_MGSSEC "mgssec"
739 #define KEY_READ_ONLY "read-only"
740 #define KEY_REGISTER_TARGET "register_target"
741 #define KEY_SET_FS "set_fs"
742 #define KEY_TGT_COUNT "tgt_count"
743 /* KEY_SET_INFO in lustre_idl.h */
744 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
746 #define KEY_CACHE_SET "cache_set"
747 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
748 #define KEY_OSP_CONNECTED "osp_connected"
752 static inline int it_to_lock_mode(struct lookup_intent *it)
754 /* CREAT needs to be tested before open (both could be set) */
755 if (it->it_op & IT_CREAT)
757 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
760 else if (it->it_op & IT_READDIR)
762 else if (it->it_op & IT_GETXATTR)
764 else if (it->it_op & IT_SETXATTR)
767 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
772 MF_MDC_CANCEL_FID1 = 1 << 0,
773 MF_MDC_CANCEL_FID2 = 1 << 1,
774 MF_MDC_CANCEL_FID3 = 1 << 2,
775 MF_MDC_CANCEL_FID4 = 1 << 3,
776 MF_GET_MDT_IDX = 1 << 4,
780 CLI_SET_MEA = 1 << 0,
781 CLI_RM_ENTRY = 1 << 1,
784 CLI_MIGRATE = 1 << 4,
788 * GETXATTR is not included as only a couple of fields in the reply body
789 * is filled, but not FID which is needed for common intent handling in
790 * mdc_finish_intent_lock()
792 static inline bool it_has_reply_body(const struct lookup_intent *it)
794 return it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR);
798 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
799 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
800 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
801 struct lu_fid op_fid4; /* to the operation locks. */
802 u32 op_mds; /* what mds server open will go to */
804 struct lustre_handle op_handle;
808 struct lmv_stripe_md *op_mea1;
809 struct lmv_stripe_md *op_mea2;
810 __u32 op_suppgids[2];
817 /* iattr fields and blocks. */
818 struct iattr op_attr;
819 loff_t op_attr_blocks;
820 __u64 op_valid; /* OBD_MD_* */
821 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
823 enum md_op_flags op_flags;
825 /* Various operation flags. */
826 enum mds_op_bias op_bias;
828 /* used to transfer info between the stacks of MD client
829 * see enum op_cli_flags */
830 enum md_cli_flags op_cli_flags;
832 /* File object data version for HSM release, on client */
833 __u64 op_data_version;
834 struct lustre_handle op_lease_handle;
836 /* File security context, for creates. */
837 const char *op_file_secctx_name;
838 void *op_file_secctx;
839 __u32 op_file_secctx_size;
841 /* default stripe offset */
842 __u32 op_default_stripe_offset;
846 /* Used by readdir */
847 unsigned int op_max_pages;
852 int (*md_blocking_ast)(struct ldlm_lock *lock,
853 struct ldlm_lock_desc *desc,
854 void *data, int flag);
857 struct md_enqueue_info;
858 /* metadata stat-ahead */
859 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
860 struct md_enqueue_info *minfo,
863 struct md_enqueue_info {
864 struct md_op_data mi_data;
865 struct lookup_intent mi_it;
866 struct lustre_handle mi_lockh;
867 struct inode *mi_dir;
868 struct ldlm_enqueue_info mi_einfo;
869 md_enqueue_cb_t mi_cb;
874 struct module *o_owner;
875 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
876 void *karg, void __user *uarg);
877 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
878 __u32 keylen, void *key, __u32 *vallen, void *val);
879 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
880 __u32 keylen, void *key,
881 __u32 vallen, void *val,
882 struct ptlrpc_request_set *set);
883 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
884 int (*o_precleanup)(struct obd_device *dev);
885 int (*o_cleanup)(struct obd_device *dev);
886 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
887 int (*o_postrecov)(struct obd_device *dev);
888 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
890 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
891 /* connect to the target device with given connection
892 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
893 * granted by the target, which are guaranteed to be a subset of flags
894 * asked for. If @ocd == NULL, use default parameters. */
895 int (*o_connect)(const struct lu_env *env,
896 struct obd_export **exp, struct obd_device *src,
897 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
899 int (*o_reconnect)(const struct lu_env *env,
900 struct obd_export *exp, struct obd_device *src,
901 struct obd_uuid *cluuid,
902 struct obd_connect_data *ocd,
904 int (*o_disconnect)(struct obd_export *exp);
906 /* Initialize/finalize fids infrastructure. */
907 int (*o_fid_init)(struct obd_device *obd,
908 struct obd_export *exp, enum lu_cli_type type);
909 int (*o_fid_fini)(struct obd_device *obd);
911 /* Allocate new fid according to passed @hint. */
912 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
913 struct lu_fid *fid, struct md_op_data *op_data);
916 * Object with @fid is getting deleted, we may want to do something
919 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
920 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
921 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
922 __u64 max_age, struct ptlrpc_request_set *set);
923 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
925 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
927 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
929 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
931 int (*o_preprw)(const struct lu_env *env, int cmd,
932 struct obd_export *exp, struct obdo *oa, int objcount,
933 struct obd_ioobj *obj, struct niobuf_remote *remote,
934 int *nr_pages, struct niobuf_local *local);
935 int (*o_commitrw)(const struct lu_env *env, int cmd,
936 struct obd_export *exp, struct obdo *oa,
937 int objcount, struct obd_ioobj *obj,
938 struct niobuf_remote *remote, int pages,
939 struct niobuf_local *local, int rc);
940 int (*o_init_export)(struct obd_export *exp);
941 int (*o_destroy_export)(struct obd_export *exp);
943 int (*o_import_event)(struct obd_device *, struct obd_import *,
944 enum obd_import_event);
946 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
947 enum obd_notify_event ev);
949 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
950 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
953 int (*o_quotactl)(struct obd_device *, struct obd_export *,
954 struct obd_quotactl *);
956 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
959 int (*o_pool_new)(struct obd_device *obd, char *poolname);
960 int (*o_pool_del)(struct obd_device *obd, char *poolname);
961 int (*o_pool_add)(struct obd_device *obd, char *poolname,
963 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
965 void (*o_getref)(struct obd_device *obd);
966 void (*o_putref)(struct obd_device *obd);
968 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
969 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
970 * Also, add a wrapper function in include/linux/obd_class.h. */
975 struct mdt_body *body;
976 struct lu_buf layout;
977 struct lmv_stripe_md *lmv;
978 #ifdef CONFIG_FS_POSIX_ACL
979 struct posix_acl *posix_acl;
983 struct md_open_data {
984 struct obd_client_handle *mod_och;
985 struct ptlrpc_request *mod_open_req;
986 struct ptlrpc_request *mod_close_req;
987 atomic_t mod_refcount;
991 struct obd_client_handle {
992 struct lustre_handle och_fh;
993 struct lu_fid och_fid;
994 struct md_open_data *och_mod;
995 struct lustre_handle och_lease_handle; /* open lock for lease */
1000 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1002 struct lookup_intent;
1006 /* Every operation from MD_STATS_FIRST_OP up to and including
1007 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1008 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1009 * Operations after MD_STATS_LAST_OP are excluded from stats.
1010 * There are a few reasons for doing this: we prune the 17
1011 * counters which will be of minimal use in understanding
1012 * metadata utilization, we save memory by allocating 15
1013 * instead of 32 counters, we save cycles by not counting.
1015 * MD_STATS_FIRST_OP must be the first member of md_ops.
1017 #define MD_STATS_FIRST_OP m_close
1018 int (*m_close)(struct obd_export *, struct md_op_data *,
1019 struct md_open_data *, struct ptlrpc_request **);
1021 int (*m_create)(struct obd_export *, struct md_op_data *,
1022 const void *, size_t, umode_t, uid_t, gid_t,
1023 cfs_cap_t, __u64, struct ptlrpc_request **);
1025 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1026 const union ldlm_policy_data *, 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 *,
1099 const struct lustre_handle *, void *, __u64 *);
1101 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1102 const struct lu_fid *, enum ldlm_type,
1103 union ldlm_policy_data *, enum ldlm_mode,
1104 struct lustre_handle *);
1106 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1107 union ldlm_policy_data *, enum ldlm_mode,
1108 enum ldlm_cancel_flags flags, void *opaque);
1110 int (*m_get_fid_from_lsm)(struct obd_export *,
1111 const struct lmv_stripe_md *,
1112 const char *name, int namelen,
1113 struct lu_fid *fid);
1114 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1115 const union lmv_mds_md *lmv, size_t lmv_size);
1118 static inline struct md_open_data *obd_mod_alloc(void)
1120 struct md_open_data *mod;
1124 atomic_set(&mod->mod_refcount, 1);
1128 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1129 #define obd_mod_put(mod) \
1131 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1132 if ((mod)->mod_open_req) \
1133 ptlrpc_req_finished((mod)->mod_open_req); \
1134 OBD_FREE_PTR(mod); \
1138 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1139 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1140 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1142 /* return 1 if client should be resend request */
1143 static inline int client_should_resend(int resend, struct client_obd *cli)
1145 return atomic_read(&cli->cl_resends) ?
1146 atomic_read(&cli->cl_resends) > resend : 1;
1150 * Return device name for this device
1152 * XXX: lu_device is declared before obd_device, while a pointer pointing
1153 * back to obd_device in lu_device, so this helper function defines here
1154 * instead of in lu_object.h
1156 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1158 return lu_dev->ld_obd->obd_name;
1161 static inline bool filename_is_volatile(const char *name, size_t namelen,
1167 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1170 /* caller does not care of idx */
1174 /* volatile file, the MDT can be set from name */
1175 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1176 /* if no MDT is specified, use std way */
1177 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1179 /* test for no MDT idx case */
1180 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1181 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1185 /* we have an idx, read it */
1186 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1187 *idx = simple_strtoul(start, &end, 16);
1189 * no digit, no trailing :, negative value
1191 if (((*idx == 0) && (end == start)) ||
1192 (*end != ':') || (*idx < 0))
1197 /* bad format of mdt idx, we cannot return an error
1198 * to caller so we use hash algo */
1199 CERROR("Bad volatile file name format: %s\n",
1200 name + LUSTRE_VOLATILE_HDR_LEN);
1204 static inline int cli_brw_size(struct obd_device *obd)
1206 LASSERT(obd != NULL);
1207 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1210 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1211 * of the client should be increased accordingly to avoid sending fragmented
1212 * RPCs over the network when the client runs out of the maximum dirty space
1213 * when so many RPCs are being generated.
1215 static inline void client_adjust_max_dirty(struct client_obd *cli)
1218 if (cli->cl_dirty_max_pages <= 0)
1219 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1222 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1223 cli->cl_max_pages_per_rpc;
1225 if (dirty_max > cli->cl_dirty_max_pages)
1226 cli->cl_dirty_max_pages = dirty_max;
1229 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1230 cli->cl_dirty_max_pages = totalram_pages / 8;
1233 #endif /* __OBD_H */