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/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 __u32 typ_sym_filter;
108 struct lu_device_type *typ_lu;
109 spinlock_t obd_type_lock;
110 struct kobject typ_kobj;
111 struct completion typ_kobj_unregister;
121 struct timeout_item {
122 enum timeout_event ti_event;
126 struct list_head ti_obd_list;
127 struct list_head ti_chain;
130 #define OBD_MAX_RIF_DEFAULT 8
131 #define OBD_MAX_RIF_MAX 512
132 #define OSC_MAX_RIF_MAX 256
133 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
134 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
135 #define OSC_DEFAULT_RESENDS 10
137 /* possible values for fo_sync_lock_cancel */
139 NEVER_SYNC_ON_CANCEL = 0,
140 BLOCKING_SYNC_ON_CANCEL = 1,
141 ALWAYS_SYNC_ON_CANCEL = 2,
142 NUM_SYNC_ON_CANCEL_STATES
146 * Limit reply buffer size for striping data to one x86_64 page. This
147 * value is chosen to fit the striping data for common use cases while
148 * staying well below the limit at which the buffer must be backed by
149 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
152 #define OBD_MAX_DEFAULT_EA_SIZE 4096
154 enum obd_cl_sem_lock_class {
163 struct rw_semaphore cl_sem;
164 struct obd_uuid cl_target_uuid;
165 struct obd_import *cl_import; /* ptlrpc connection state */
166 size_t cl_conn_count;
168 /* Cache maximum and default values for easize. This is
169 * strictly a performance optimization to minimize calls to
170 * obd_size_diskmd(). The default values are used to calculate the
171 * initial size of a request buffer. The ptlrpc layer will resize the
172 * buffer as needed to accommodate a larger reply from the
173 * server. The default values should be small enough to avoid wasted
174 * memory and excessive use of vmalloc(), yet large enough to avoid
175 * reallocating the buffer in the common use case. */
177 /* Default EA size for striping attributes. It is initialized at
178 * mount-time based on the default stripe width of the filesystem,
179 * then it tracks the largest observed EA size advertised by
180 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
181 __u32 cl_default_mds_easize;
183 /* Maximum possible EA size computed at mount-time based on
184 * the number of OSTs in the filesystem. May be increased at
185 * run-time if a larger observed size is advertised by the MDT. */
186 __u32 cl_max_mds_easize;
188 enum lustre_sec_part cl_sp_me;
189 enum lustre_sec_part cl_sp_to;
190 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
192 /* the grant values are protected by loi_list_lock below */
193 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
194 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
195 unsigned long cl_dirty_transit; /* dirty synchronous */
196 unsigned long cl_avail_grant; /* bytes of credit for ost */
197 unsigned long cl_lost_grant; /* lost credits (trunc) */
198 /* grant consumed for dirty pages */
199 unsigned long cl_dirty_grant;
201 /* since we allocate grant by blocks, we don't know how many grant will
202 * be used to add a page into cache. As a solution, we reserve maximum
203 * grant before trying to dirty a page and unreserve the rest.
204 * See osc_{reserve|unreserve}_grant for details. */
205 long cl_reserved_grant;
206 struct list_head cl_cache_waiters; /* waiting for cache/grant */
207 time64_t cl_next_shrink_grant; /* seconds */
208 struct list_head cl_grant_shrink_list; /* Timeout event list */
209 time64_t cl_grant_shrink_interval; /* seconds */
211 /* A chunk is an optimal size used by osc_extent to determine
212 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
214 /* extent insertion metadata overhead to be accounted in grant,
216 unsigned int cl_grant_extent_tax;
217 /* maximum extent size, in number of pages */
218 unsigned int cl_max_extent_pages;
220 /* keep track of objects that have lois that contain pages which
221 * have been queued for async brw. this lock also protects the
222 * lists of osc_client_pages that hang off of the loi */
224 * ->cl_loi_list_lock protects consistency of
225 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
226 * ->ap_completion() call-backs are executed under this lock. As we
227 * cannot guarantee that these call-backs never block on all platforms
228 * (as a matter of fact they do block on Mac OS X), type of
229 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
230 * and blocking mutex on Mac OS X. (Alternative is to make this lock
231 * blocking everywhere, but we don't want to slow down fast-path of
232 * our main platform.)
234 * NB by Jinshan: though field names are still _loi_, but actually
235 * osc_object{}s are in the list.
237 spinlock_t cl_loi_list_lock;
238 struct list_head cl_loi_ready_list;
239 struct list_head cl_loi_hp_ready_list;
240 struct list_head cl_loi_write_list;
241 struct list_head cl_loi_read_list;
242 __u32 cl_r_in_flight;
243 __u32 cl_w_in_flight;
244 /* just a sum of the loi/lop pending numbers to be exported by /proc */
245 atomic_t cl_pending_w_pages;
246 atomic_t cl_pending_r_pages;
247 __u32 cl_max_pages_per_rpc;
248 __u32 cl_max_rpcs_in_flight;
249 __u32 cl_short_io_bytes;
250 struct obd_histogram cl_read_rpc_hist;
251 struct obd_histogram cl_write_rpc_hist;
252 struct obd_histogram cl_read_page_hist;
253 struct obd_histogram cl_write_page_hist;
254 struct obd_histogram cl_read_offset_hist;
255 struct obd_histogram cl_write_offset_hist;
257 /** LRU for osc caching pages */
258 struct cl_client_cache *cl_cache;
259 /** member of cl_cache->ccc_lru */
260 struct list_head cl_lru_osc;
261 /** # of available LRU slots left in the per-OSC cache.
262 * Available LRU slots are shared by all OSCs of the same file system,
263 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
264 atomic_long_t *cl_lru_left;
265 /** # of busy LRU pages. A page is considered busy if it's in writeback
266 * queue, or in transfer. Busy pages can't be discarded so they are not
268 atomic_long_t cl_lru_busy;
269 /** # of LRU pages in the cache for this client_obd */
270 atomic_long_t cl_lru_in_list;
271 /** # of threads are shrinking LRU cache. To avoid contention, it's not
272 * allowed to have multiple threads shrinking LRU cache. */
273 atomic_t cl_lru_shrinkers;
274 /** The time when this LRU cache was last used. */
275 time64_t cl_lru_last_used;
276 /** stats: how many reclaims have happened for this client_obd.
277 * reclaim and shrink - shrink is async, voluntarily rebalancing;
278 * reclaim is sync, initiated by IO thread when the LRU slots are
280 __u64 cl_lru_reclaim;
281 /** List of LRU pages for this client_obd */
282 struct list_head cl_lru_list;
283 /** Lock for LRU page list */
284 spinlock_t cl_lru_list_lock;
285 /** # of unstable pages in this client_obd.
286 * An unstable page is a page state that WRITE RPC has finished but
287 * the transaction has NOT yet committed. */
288 atomic_long_t cl_unstable_count;
289 /** Link to osc_shrinker_list */
290 struct list_head cl_shrink_list;
292 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
293 atomic_t cl_destroy_in_flight;
294 wait_queue_head_t cl_destroy_waitq;
296 struct mdc_rpc_lock *cl_rpc_lock;
298 /* modify rpcs in flight
299 * currently used for metadata only */
300 spinlock_t cl_mod_rpcs_lock;
301 __u16 cl_max_mod_rpcs_in_flight;
302 __u16 cl_mod_rpcs_in_flight;
303 __u16 cl_close_rpcs_in_flight;
304 wait_queue_head_t cl_mod_rpcs_waitq;
305 unsigned long *cl_mod_tag_bitmap;
306 struct obd_histogram cl_mod_rpcs_hist;
309 struct mutex cl_mgc_mutex;
310 struct local_oid_storage *cl_mgc_los;
311 struct dt_object *cl_mgc_configs_dir;
312 atomic_t cl_mgc_refcount;
313 struct obd_export *cl_mgc_mgsexp;
315 /* checksumming for data sent over the network */
316 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
317 cl_checksum_dump:1; /* same */
318 /* supported checksum types that are worked out at connect time */
319 __u32 cl_supp_cksum_types;
320 /* checksum algorithm to be used */
321 enum cksum_types cl_cksum_type;
323 /* also protected by the poorly named _loi_list_lock lock above */
324 struct osc_async_rc cl_ar;
326 /* sequence manager */
327 struct lu_client_seq *cl_seq;
328 struct rw_semaphore cl_seq_rwsem;
330 atomic_t cl_resends; /* resend count */
332 /* ptlrpc work for writeback in ptlrpcd context */
333 void *cl_writeback_work;
335 /* hash tables for osc_quota_info */
336 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
337 /* Links to the global list of registered changelog devices */
338 struct list_head cl_chg_dev_linkage;
340 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
347 struct echo_client_obd {
348 struct obd_export *ec_exp; /* the local connection to osc/lov */
350 struct list_head ec_objects;
351 struct list_head ec_locks;
355 /* Generic subset of OSTs */
357 __u32 *op_array; /* array of index of
359 unsigned int op_count; /* number of OSTs in the array */
360 unsigned int op_size; /* allocated size of lp_array */
361 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
364 /* allow statfs data caching for 1 second */
365 #define OBD_STATFS_CACHE_SECONDS 1
367 struct lov_tgt_desc {
368 struct list_head ltd_kill;
369 struct obd_uuid ltd_uuid;
370 struct obd_device *ltd_obd;
371 struct obd_export *ltd_exp;
373 __u32 ltd_index; /* index in lov_obd->tgts */
374 unsigned long ltd_active:1,/* is this target up for requests */
375 ltd_activate:1,/* should target be activated */
376 ltd_reap:1; /* should this target be deleted */
380 struct lov_desc desc;
381 struct lov_tgt_desc **lov_tgts; /* sparse array */
382 struct ost_pool lov_packed; /* all OSTs in a packed
384 struct mutex lov_lock;
385 struct obd_connect_data lov_ocd;
386 struct proc_dir_entry *targets_proc_entry;
387 atomic_t lov_refcount;
388 __u32 lov_death_row; /* tgts scheduled to be deleted */
389 __u32 lov_tgt_size; /* size of tgts array */
392 struct cfs_hash *lov_pools_hash_body; /* used for key access */
393 struct list_head lov_pool_list; /* used for sequential access */
394 struct proc_dir_entry *lov_pool_proc_entry;
395 enum lustre_sec_part lov_sp_me;
397 /* Cached LRU and unstable data from upper layer */
398 struct cl_client_cache *lov_cache;
400 struct rw_semaphore lov_notify_lock;
403 struct lmv_tgt_desc {
404 struct obd_uuid ltd_uuid;
405 struct obd_export *ltd_exp;
407 struct mutex ltd_fid_mutex;
408 unsigned long ltd_active:1; /* target up for requests */
412 struct lu_client_fld lmv_fld;
414 struct lmv_desc desc;
415 struct obd_uuid cluuid;
416 struct proc_dir_entry *targets_proc_entry;
418 struct mutex lmv_init_mutex;
423 __u32 tgts_size; /* size of tgts array */
424 struct lmv_tgt_desc **tgts;
426 struct obd_connect_data conn_data;
429 struct niobuf_local {
430 __u64 lnb_file_offset;
431 __u32 lnb_page_offset;
435 struct page *lnb_page;
439 struct tgt_thread_big_cache {
440 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
443 #define LUSTRE_FLD_NAME "fld"
444 #define LUSTRE_SEQ_NAME "seq"
446 #define LUSTRE_MDD_NAME "mdd"
447 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
448 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
449 #define LUSTRE_VVP_NAME "vvp"
450 #define LUSTRE_LMV_NAME "lmv"
451 #define LUSTRE_SLP_NAME "slp"
452 #define LUSTRE_LOD_NAME "lod"
453 #define LUSTRE_OSP_NAME "osp"
454 #define LUSTRE_LWP_NAME "lwp"
456 /* obd device type names */
457 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
458 #define LUSTRE_MDS_NAME "mds"
459 #define LUSTRE_MDT_NAME "mdt"
460 #define LUSTRE_MDC_NAME "mdc"
461 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
462 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
463 #define LUSTRE_OSC_NAME "osc"
464 #define LUSTRE_LOV_NAME "lov"
465 #define LUSTRE_MGS_NAME "mgs"
466 #define LUSTRE_MGC_NAME "mgc"
468 #define LUSTRE_ECHO_NAME "obdecho"
469 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
470 #define LUSTRE_QMT_NAME "qmt"
472 /* Constant obd names (post-rename) */
473 #define LUSTRE_MDS_OBDNAME "MDS"
474 #define LUSTRE_OSS_OBDNAME "OSS"
475 #define LUSTRE_MGS_OBDNAME "MGS"
476 #define LUSTRE_MGC_OBDNAME "MGC"
478 static inline int is_lwp_on_mdt(char *name)
482 ptr = strrchr(name, '-');
484 CERROR("%s is not a obdname\n", name);
488 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
490 if (strncmp(ptr + 1, "MDT", 3) != 0)
493 while (*(--ptr) != '-' && ptr != name);
498 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
504 static inline int is_lwp_on_ost(char *name)
508 ptr = strrchr(name, '-');
510 CERROR("%s is not a obdname\n", name);
514 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
516 if (strncmp(ptr + 1, "OST", 3) != 0)
519 while (*(--ptr) != '-' && ptr != name);
524 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
531 * Events signalled through obd_notify() upcall-chain.
533 enum obd_notify_event {
534 /* Device connect start */
536 /* Device activated */
538 /* Device deactivated */
540 /* Connect data for import were changed */
542 /* Administratively deactivate/activate event */
543 OBD_NOTIFY_DEACTIVATE,
548 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
549 * being main example).
551 struct obd_notify_upcall {
552 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
553 enum obd_notify_event ev, void *owner);
554 /* Opaque datum supplied by upper layer listener */
558 struct target_recovery_data {
559 svc_handler_t trd_recovery_handler;
560 pid_t trd_processing_task;
561 struct completion trd_starting;
562 struct completion trd_finishing;
565 struct obd_llog_group {
566 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
567 wait_queue_head_t olg_waitq;
569 struct mutex olg_cat_processing;
572 /* corresponds to one of the obd's */
573 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
576 struct obd_type *obd_type;
577 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
578 int obd_minor; /* device number: lctl dl */
579 struct lu_device *obd_lu_dev;
581 /* common and UUID name of this device */
582 struct obd_uuid obd_uuid;
583 char obd_name[MAX_OBD_NAME];
585 /* bitfield modification is protected by obd_dev_lock */
587 obd_attached:1, /* finished attach */
588 obd_set_up:1, /* finished setup */
589 obd_recovering:1, /* there are recoverable clients */
590 obd_abort_recovery:1, /* recovery expired */
591 obd_version_recov:1, /* obd uses version checking */
592 obd_replayable:1, /* recovery enabled; inform clients */
593 obd_no_transno:1, /* no committed-transno notification */
594 obd_no_recov:1, /* fail instead of retry messages */
595 obd_stopping:1, /* started cleanup */
596 obd_starting:1, /* started setup */
597 obd_force:1, /* cleanup with > 0 obd refcount */
598 obd_fail:1, /* cleanup with failover */
599 obd_no_conn:1, /* deny new connections */
600 obd_inactive:1, /* device active/inactive
601 * (for /proc/status only!!) */
602 obd_no_ir:1, /* no imperative recovery. */
603 obd_process_conf:1, /* device is processing mgs config */
604 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
605 obd_checksum_dump:1; /* dump pages upon cksum error */
607 /* use separate field as it is set in interrupt to don't mess with
608 * protection of other bits using _bh lock */
609 unsigned long obd_recovery_expired:1;
610 /* uuid-export hash body */
611 struct cfs_hash *obd_uuid_hash;
612 /* nid-export hash body */
613 struct cfs_hash *obd_nid_hash;
615 struct cfs_hash *obd_nid_stats_hash;
616 /* client_generation-export hash body */
617 struct cfs_hash *obd_gen_hash;
618 struct list_head obd_nid_stats;
619 struct list_head obd_exports;
620 struct list_head obd_unlinked_exports;
621 struct list_head obd_delayed_exports;
622 struct list_head obd_lwp_list;
623 atomic_t obd_refcount;
625 spinlock_t obd_nid_lock;
626 struct ldlm_namespace *obd_namespace;
627 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
628 /* a spinlock is OK for what we do now, may need a semaphore later */
629 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
630 spinlock_t obd_osfs_lock;
631 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
633 __u64 obd_last_committed;
634 struct mutex obd_dev_mutex;
635 struct lvfs_run_ctxt obd_lvfs_ctxt;
636 struct obd_llog_group obd_olg; /* default llog group */
637 struct obd_device *obd_observer;
638 struct rw_semaphore obd_observer_link_sem;
639 struct obd_notify_upcall obd_upcall;
640 struct obd_export *obd_self_export;
641 struct obd_export *obd_lwp_export;
642 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
643 struct list_head obd_exports_timed;
644 time64_t obd_eviction_timer; /* for ping evictor */
646 int obd_max_recoverable_clients;
647 atomic_t obd_connected_clients;
648 int obd_stale_clients;
649 /* this lock protects all recovery list_heads, timer and
650 * obd_next_recovery_transno value */
651 spinlock_t obd_recovery_task_lock;
652 __u64 obd_next_recovery_transno;
653 int obd_replayed_requests;
654 int obd_requests_queued_for_recovery;
655 wait_queue_head_t obd_next_transno_waitq;
656 /* protected by obd_recovery_task_lock */
657 struct timer_list obd_recovery_timer;
659 time64_t obd_recovery_start;
660 /* seconds, for lprocfs_status */
661 time64_t obd_recovery_end;
662 time64_t obd_recovery_time_hard;
663 time64_t obd_recovery_timeout;
664 int obd_recovery_ir_factor;
666 /* new recovery stuff from CMD2 */
667 int obd_replayed_locks;
668 atomic_t obd_req_replay_clients;
669 atomic_t obd_lock_replay_clients;
670 struct target_recovery_data obd_recovery_data;
672 /* all lists are protected by obd_recovery_task_lock */
673 struct list_head obd_req_replay_queue;
674 struct list_head obd_lock_replay_queue;
675 struct list_head obd_final_req_queue;
678 #ifdef HAVE_SERVER_SUPPORT
679 struct obd_device_target obt;
680 struct filter_obd filter;
682 struct echo_obd echo;
684 struct client_obd cli;
685 struct echo_client_obd echo_client;
690 /* Fields used by LProcFS */
691 struct lprocfs_stats *obd_stats;
692 unsigned int obd_cntr_base;
694 unsigned int obd_md_cntr_base;
695 struct lprocfs_stats *obd_md_stats;
697 struct proc_dir_entry *obd_proc_entry;
698 struct proc_dir_entry *obd_proc_exports_entry;
699 struct proc_dir_entry *obd_svc_procroot;
700 struct lprocfs_stats *obd_svc_stats;
701 struct attribute_group obd_attrs_group;
702 struct attribute **obd_attrs;
703 struct lprocfs_vars *obd_vars;
704 atomic_t obd_evict_inprogress;
705 wait_queue_head_t obd_evict_inprogress_waitq;
706 struct list_head obd_evict_list; /* protected with pet_lock */
709 * LDLM pool part. Save last calculated SLV and Limit.
711 rwlock_t obd_pool_lock;
715 int obd_conn_inprogress;
718 * List of outstanding class_incref()'s fo this OBD. For debugging. */
719 struct lu_ref obd_reference;
721 struct kset obd_kset; /* sysfs object collection */
722 struct kobj_type obd_ktype;
723 struct completion obd_kobj_unregister;
726 /* get/set_info keys */
727 #define KEY_ASYNC "async"
728 #define KEY_CHANGELOG_CLEAR "changelog_clear"
729 #define KEY_FID2PATH "fid2path"
730 #define KEY_CHECKSUM "checksum"
731 #define KEY_CLEAR_FS "clear_fs"
732 #define KEY_CONN_DATA "conn_data"
733 #define KEY_EVICT_BY_NID "evict_by_nid"
734 #define KEY_FIEMAP "fiemap"
735 #define KEY_FLUSH_CTX "flush_ctx"
736 #define KEY_GRANT_SHRINK "grant_shrink"
737 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
738 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
739 #define KEY_INTERMDS "inter_mds"
740 #define KEY_LAST_ID "last_id"
741 #define KEY_LAST_FID "last_fid"
742 #define KEY_MAX_EASIZE "max_easize"
743 #define KEY_DEFAULT_EASIZE "default_easize"
744 #define KEY_MGSSEC "mgssec"
745 #define KEY_READ_ONLY "read-only"
746 #define KEY_REGISTER_TARGET "register_target"
747 #define KEY_SET_FS "set_fs"
748 #define KEY_TGT_COUNT "tgt_count"
749 /* KEY_SET_INFO in lustre_idl.h */
750 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
752 #define KEY_CACHE_SET "cache_set"
753 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
754 #define KEY_OSP_CONNECTED "osp_connected"
758 static inline int it_to_lock_mode(struct lookup_intent *it)
760 /* CREAT needs to be tested before open (both could be set) */
761 if (it->it_op & IT_CREAT)
763 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
766 else if (it->it_op & IT_READDIR)
768 else if (it->it_op & IT_GETXATTR)
770 else if (it->it_op & IT_SETXATTR)
773 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
778 MF_MDC_CANCEL_FID1 = 1 << 0,
779 MF_MDC_CANCEL_FID2 = 1 << 1,
780 MF_MDC_CANCEL_FID3 = 1 << 2,
781 MF_MDC_CANCEL_FID4 = 1 << 3,
782 MF_GET_MDT_IDX = 1 << 4,
786 CLI_SET_MEA = 1 << 0,
787 CLI_RM_ENTRY = 1 << 1,
790 CLI_MIGRATE = 1 << 4,
794 * GETXATTR is not included as only a couple of fields in the reply body
795 * is filled, but not FID which is needed for common intent handling in
796 * mdc_finish_intent_lock()
798 static inline bool it_has_reply_body(const struct lookup_intent *it)
800 return it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR);
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 */
810 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 __u64 op_valid; /* OBD_MD_* */
827 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
829 enum md_op_flags op_flags;
831 /* Various operation flags. */
832 enum mds_op_bias op_bias;
834 /* used to transfer info between the stacks of MD client
835 * see enum op_cli_flags */
836 enum md_cli_flags op_cli_flags;
838 /* File object data version for HSM release, on client */
839 __u64 op_data_version;
840 struct lustre_handle op_lease_handle;
842 /* File security context, for creates. */
843 const char *op_file_secctx_name;
844 void *op_file_secctx;
845 __u32 op_file_secctx_size;
847 /* default stripe offset */
848 __u32 op_default_stripe_offset;
852 /* Used by readdir */
853 unsigned int op_max_pages;
858 int (*md_blocking_ast)(struct ldlm_lock *lock,
859 struct ldlm_lock_desc *desc,
860 void *data, int flag);
863 struct md_enqueue_info;
864 /* metadata stat-ahead */
865 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
866 struct md_enqueue_info *minfo,
869 struct md_enqueue_info {
870 struct md_op_data mi_data;
871 struct lookup_intent mi_it;
872 struct lustre_handle mi_lockh;
873 struct inode *mi_dir;
874 struct ldlm_enqueue_info mi_einfo;
875 md_enqueue_cb_t mi_cb;
880 struct module *o_owner;
881 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
882 void *karg, void __user *uarg);
883 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
884 __u32 keylen, void *key, __u32 *vallen, void *val);
885 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
886 __u32 keylen, void *key,
887 __u32 vallen, void *val,
888 struct ptlrpc_request_set *set);
889 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
890 int (*o_precleanup)(struct obd_device *dev);
891 int (*o_cleanup)(struct obd_device *dev);
892 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
893 int (*o_postrecov)(struct obd_device *dev);
894 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
896 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
897 /* connect to the target device with given connection
898 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
899 * granted by the target, which are guaranteed to be a subset of flags
900 * asked for. If @ocd == NULL, use default parameters. */
901 int (*o_connect)(const struct lu_env *env,
902 struct obd_export **exp, struct obd_device *src,
903 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
905 int (*o_reconnect)(const struct lu_env *env,
906 struct obd_export *exp, struct obd_device *src,
907 struct obd_uuid *cluuid,
908 struct obd_connect_data *ocd,
910 int (*o_disconnect)(struct obd_export *exp);
912 /* Initialize/finalize fids infrastructure. */
913 int (*o_fid_init)(struct obd_device *obd,
914 struct obd_export *exp, enum lu_cli_type type);
915 int (*o_fid_fini)(struct obd_device *obd);
917 /* Allocate new fid according to passed @hint. */
918 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
919 struct lu_fid *fid, struct md_op_data *op_data);
922 * Object with @fid is getting deleted, we may want to do something
925 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
926 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
927 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
928 __u64 max_age, struct ptlrpc_request_set *set);
929 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
931 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
933 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
935 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
937 int (*o_preprw)(const struct lu_env *env, int cmd,
938 struct obd_export *exp, struct obdo *oa, int objcount,
939 struct obd_ioobj *obj, struct niobuf_remote *remote,
940 int *nr_pages, struct niobuf_local *local);
941 int (*o_commitrw)(const struct lu_env *env, int cmd,
942 struct obd_export *exp, struct obdo *oa,
943 int objcount, struct obd_ioobj *obj,
944 struct niobuf_remote *remote, int pages,
945 struct niobuf_local *local, int rc);
946 int (*o_init_export)(struct obd_export *exp);
947 int (*o_destroy_export)(struct obd_export *exp);
949 int (*o_import_event)(struct obd_device *, struct obd_import *,
950 enum obd_import_event);
952 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
953 enum obd_notify_event ev);
955 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
956 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
959 int (*o_quotactl)(struct obd_device *, struct obd_export *,
960 struct obd_quotactl *);
962 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
965 int (*o_pool_new)(struct obd_device *obd, char *poolname);
966 int (*o_pool_del)(struct obd_device *obd, char *poolname);
967 int (*o_pool_add)(struct obd_device *obd, char *poolname,
969 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
971 void (*o_getref)(struct obd_device *obd);
972 void (*o_putref)(struct obd_device *obd);
974 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
975 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
976 * Also, add a wrapper function in include/linux/obd_class.h. */
981 struct mdt_body *body;
982 struct lu_buf layout;
983 struct lmv_stripe_md *lmv;
984 #ifdef CONFIG_FS_POSIX_ACL
985 struct posix_acl *posix_acl;
989 struct md_open_data {
990 struct obd_client_handle *mod_och;
991 struct ptlrpc_request *mod_open_req;
992 struct ptlrpc_request *mod_close_req;
993 atomic_t mod_refcount;
997 struct obd_client_handle {
998 struct lustre_handle och_fh;
999 struct lu_fid och_fid;
1000 struct md_open_data *och_mod;
1001 struct lustre_handle och_lease_handle; /* open lock for lease */
1006 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1008 struct lookup_intent;
1012 /* Every operation from MD_STATS_FIRST_OP up to and including
1013 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1014 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1015 * Operations after MD_STATS_LAST_OP are excluded from stats.
1016 * There are a few reasons for doing this: we prune the 17
1017 * counters which will be of minimal use in understanding
1018 * metadata utilization, we save memory by allocating 15
1019 * instead of 32 counters, we save cycles by not counting.
1021 * MD_STATS_FIRST_OP must be the first member of md_ops.
1023 #define MD_STATS_FIRST_OP m_close
1024 int (*m_close)(struct obd_export *, struct md_op_data *,
1025 struct md_open_data *, struct ptlrpc_request **);
1027 int (*m_create)(struct obd_export *, struct md_op_data *,
1028 const void *, size_t, umode_t, uid_t, gid_t,
1029 cfs_cap_t, __u64, struct ptlrpc_request **);
1031 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1032 const union ldlm_policy_data *, struct md_op_data *,
1033 struct lustre_handle *, __u64);
1035 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1036 struct ptlrpc_request **);
1038 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1039 struct lookup_intent *,
1040 struct ptlrpc_request **,
1041 ldlm_blocking_callback, __u64);
1043 int (*m_link)(struct obd_export *, struct md_op_data *,
1044 struct ptlrpc_request **);
1046 int (*m_rename)(struct obd_export *, struct md_op_data *,
1047 const char *, size_t, const char *, size_t,
1048 struct ptlrpc_request **);
1050 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1051 size_t , struct ptlrpc_request **);
1053 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1054 struct ptlrpc_request **);
1056 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1057 struct md_callback *cb_op, __u64 hash_offset,
1058 struct page **ppage);
1060 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1061 struct ptlrpc_request **);
1063 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1064 u64, const char *, const char *, int, int, int, u32,
1065 struct ptlrpc_request **);
1067 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1068 u64, const char *, const char *, int, int, int,
1069 struct ptlrpc_request **);
1071 int (*m_intent_getattr_async)(struct obd_export *,
1072 struct md_enqueue_info *);
1074 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1075 struct lu_fid *, __u64 *bits);
1077 #define MD_STATS_LAST_OP m_revalidate_lock
1079 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1080 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1082 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1083 struct ptlrpc_request **);
1085 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1087 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1088 struct obd_export *, struct obd_export *,
1089 struct lustre_md *);
1091 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1093 int (*m_merge_attr)(struct obd_export *,
1094 const struct lmv_stripe_md *lsm,
1095 struct cl_attr *attr, ldlm_blocking_callback);
1097 int (*m_set_open_replay_data)(struct obd_export *,
1098 struct obd_client_handle *,
1099 struct lookup_intent *);
1101 int (*m_clear_open_replay_data)(struct obd_export *,
1102 struct obd_client_handle *);
1104 int (*m_set_lock_data)(struct obd_export *,
1105 const struct lustre_handle *, void *, __u64 *);
1107 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1108 const struct lu_fid *, enum ldlm_type,
1109 union ldlm_policy_data *, enum ldlm_mode,
1110 struct lustre_handle *);
1112 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1113 union ldlm_policy_data *, enum ldlm_mode,
1114 enum ldlm_cancel_flags flags, void *opaque);
1116 int (*m_get_fid_from_lsm)(struct obd_export *,
1117 const struct lmv_stripe_md *,
1118 const char *name, int namelen,
1119 struct lu_fid *fid);
1120 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1121 const union lmv_mds_md *lmv, size_t lmv_size);
1124 static inline struct md_open_data *obd_mod_alloc(void)
1126 struct md_open_data *mod;
1130 atomic_set(&mod->mod_refcount, 1);
1134 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1135 #define obd_mod_put(mod) \
1137 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1138 if ((mod)->mod_open_req) \
1139 ptlrpc_req_finished((mod)->mod_open_req); \
1140 OBD_FREE_PTR(mod); \
1144 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1145 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1146 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1148 /* return 1 if client should be resend request */
1149 static inline int client_should_resend(int resend, struct client_obd *cli)
1151 return atomic_read(&cli->cl_resends) ?
1152 atomic_read(&cli->cl_resends) > resend : 1;
1156 * Return device name for this device
1158 * XXX: lu_device is declared before obd_device, while a pointer pointing
1159 * back to obd_device in lu_device, so this helper function defines here
1160 * instead of in lu_object.h
1162 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1164 return lu_dev->ld_obd->obd_name;
1167 static inline bool filename_is_volatile(const char *name, size_t namelen,
1173 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1176 /* caller does not care of idx */
1180 /* volatile file, the MDT can be set from name */
1181 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1182 /* if no MDT is specified, use std way */
1183 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1185 /* test for no MDT idx case */
1186 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1187 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1191 /* we have an idx, read it */
1192 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1193 *idx = simple_strtoul(start, &end, 16);
1195 * no digit, no trailing :, negative value
1197 if (((*idx == 0) && (end == start)) ||
1198 (*end != ':') || (*idx < 0))
1203 /* bad format of mdt idx, we cannot return an error
1204 * to caller so we use hash algo */
1205 CERROR("Bad volatile file name format: %s\n",
1206 name + LUSTRE_VOLATILE_HDR_LEN);
1210 static inline int cli_brw_size(struct obd_device *obd)
1212 LASSERT(obd != NULL);
1213 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1216 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1217 * of the client should be increased accordingly to avoid sending fragmented
1218 * RPCs over the network when the client runs out of the maximum dirty space
1219 * when so many RPCs are being generated.
1221 static inline void client_adjust_max_dirty(struct client_obd *cli)
1224 if (cli->cl_dirty_max_pages <= 0)
1225 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1228 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1229 cli->cl_max_pages_per_rpc;
1231 if (dirty_max > cli->cl_dirty_max_pages)
1232 cli->cl_dirty_max_pages = dirty_max;
1235 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1236 cli->cl_dirty_max_pages = totalram_pages / 8;
1239 #endif /* __OBD_H */