4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/kobject.h>
37 #include <linux/spinlock.h>
38 #include <linux/sysfs.h>
40 #include <uapi/linux/lustre/lustre_idl.h>
41 #include <lustre_lib.h>
42 #include <libcfs/bitmap.h>
43 #ifdef HAVE_SERVER_SUPPORT
44 # include <lu_target.h>
45 # include <obd_target.h>
48 #include <lustre_export.h>
49 #include <lustre_fid.h>
50 #include <lustre_fld.h>
51 #include <lustre_handles.h>
52 #include <lustre_intent.h>
54 #include <lustre_quota.h>
56 #define MAX_OBD_DEVICES 8192
64 struct lov_oinfo { /* per-stripe data structure */
65 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
66 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
67 int loi_ost_gen; /* generation of this loi_ost_idx */
69 unsigned long loi_kms_valid:1;
70 __u64 loi_kms; /* known minimum size */
71 struct ost_lvb loi_lvb;
72 struct osc_async_rc loi_ar;
75 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
78 oinfo->loi_kms_valid = 1;
84 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
86 /* obd info for a particular level (lov, osc). */
88 /* OBD_STATFS_* flags */
90 /* statfs data specific for every OSC, if needed at all. */
91 struct obd_statfs *oi_osfs;
92 /* An update callback which is called to update some data on upper
93 * level. E.g. it is used for update lsm->lsm_oinfo at every received
94 * request in osc level for enqueue requests. It is also possible to
95 * update some caller data from LOV layer if needed. */
96 obd_enqueue_update_f oi_cb_up;
100 struct list_head typ_chain;
101 struct obd_ops *typ_dt_ops;
102 struct md_ops *typ_md_ops;
103 struct proc_dir_entry *typ_procroot;
104 struct proc_dir_entry *typ_procsym;
105 struct dentry *typ_debugfs_entry;
106 #ifdef HAVE_SERVER_SUPPORT
111 struct lu_device_type *typ_lu;
112 spinlock_t obd_type_lock;
113 struct kobject *typ_kobj;
123 struct timeout_item {
124 enum timeout_event ti_event;
128 struct list_head ti_obd_list;
129 struct list_head ti_chain;
132 #define OBD_MAX_RIF_DEFAULT 8
133 #define OBD_MAX_RIF_MAX 512
134 #define OSC_MAX_RIF_MAX 256
135 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
136 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
137 #define OSC_DEFAULT_RESENDS 10
139 /* possible values for fo_sync_lock_cancel */
141 NEVER_SYNC_ON_CANCEL = 0,
142 BLOCKING_SYNC_ON_CANCEL = 1,
143 ALWAYS_SYNC_ON_CANCEL = 2,
144 NUM_SYNC_ON_CANCEL_STATES
148 * Limit reply buffer size for striping data to one x86_64 page. This
149 * value is chosen to fit the striping data for common use cases while
150 * staying well below the limit at which the buffer must be backed by
151 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
154 #define OBD_MAX_DEFAULT_EA_SIZE 4096
156 enum obd_cl_sem_lock_class {
165 struct rw_semaphore cl_sem;
166 struct obd_uuid cl_target_uuid;
167 struct obd_import *cl_import; /* ptlrpc connection state */
168 size_t cl_conn_count;
170 /* Cache maximum and default values for easize. This is
171 * strictly a performance optimization to minimize calls to
172 * obd_size_diskmd(). The default values are used to calculate the
173 * initial size of a request buffer. The ptlrpc layer will resize the
174 * buffer as needed to accommodate a larger reply from the
175 * server. The default values should be small enough to avoid wasted
176 * memory and excessive use of vmalloc(), yet large enough to avoid
177 * reallocating the buffer in the common use case. */
179 /* Default EA size for striping attributes. It is initialized at
180 * mount-time based on the default stripe width of the filesystem,
181 * then it tracks the largest observed EA size advertised by
182 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
183 __u32 cl_default_mds_easize;
185 /* Maximum possible EA size computed at mount-time based on
186 * the number of OSTs in the filesystem. May be increased at
187 * run-time if a larger observed size is advertised by the MDT. */
188 __u32 cl_max_mds_easize;
190 /* Data-on-MDT specific value to set larger reply buffer for possible
191 * data read along with open/stat requests. By default it tries to use
192 * unused space in reply buffer.
193 * This value is used to ensure that reply buffer has at least as
194 * much free space as value indicates. That free space is gained from
195 * LOV EA buffer which is small for DoM files and on big systems can
196 * provide up to 32KB of extra space in reply buffer.
197 * Default value is 8K now.
199 __u32 cl_dom_min_inline_repsize;
201 enum lustre_sec_part cl_sp_me;
202 enum lustre_sec_part cl_sp_to;
203 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
205 /* the grant values are protected by loi_list_lock below */
206 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
207 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
208 unsigned long cl_dirty_transit; /* dirty synchronous */
209 unsigned long cl_avail_grant; /* bytes of credit for ost */
210 unsigned long cl_lost_grant; /* lost credits (trunc) */
211 /* grant consumed for dirty pages */
212 unsigned long cl_dirty_grant;
214 /* since we allocate grant by blocks, we don't know how many grant will
215 * be used to add a page into cache. As a solution, we reserve maximum
216 * grant before trying to dirty a page and unreserve the rest.
217 * See osc_{reserve|unreserve}_grant for details. */
218 long cl_reserved_grant;
219 struct list_head cl_cache_waiters; /* waiting for cache/grant */
220 time64_t cl_next_shrink_grant; /* seconds */
221 struct list_head cl_grant_chain;
222 time64_t cl_grant_shrink_interval; /* seconds */
224 /* A chunk is an optimal size used by osc_extent to determine
225 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
227 /* extent insertion metadata overhead to be accounted in grant,
229 unsigned int cl_grant_extent_tax;
230 /* maximum extent size, in number of pages */
231 unsigned int cl_max_extent_pages;
233 /* keep track of objects that have lois that contain pages which
234 * have been queued for async brw. this lock also protects the
235 * lists of osc_client_pages that hang off of the loi */
237 * ->cl_loi_list_lock protects consistency of
238 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
239 * ->ap_completion() call-backs are executed under this lock. As we
240 * cannot guarantee that these call-backs never block on all platforms
241 * (as a matter of fact they do block on Mac OS X), type of
242 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
243 * and blocking mutex on Mac OS X. (Alternative is to make this lock
244 * blocking everywhere, but we don't want to slow down fast-path of
245 * our main platform.)
247 * NB by Jinshan: though field names are still _loi_, but actually
248 * osc_object{}s are in the list.
250 spinlock_t cl_loi_list_lock;
251 struct list_head cl_loi_ready_list;
252 struct list_head cl_loi_hp_ready_list;
253 struct list_head cl_loi_write_list;
254 struct list_head cl_loi_read_list;
255 __u32 cl_r_in_flight;
256 __u32 cl_w_in_flight;
257 /* just a sum of the loi/lop pending numbers to be exported by /proc */
258 atomic_t cl_pending_w_pages;
259 atomic_t cl_pending_r_pages;
260 u32 cl_max_pages_per_rpc;
261 u32 cl_max_rpcs_in_flight;
262 u32 cl_max_short_io_bytes;
263 struct obd_histogram cl_read_rpc_hist;
264 struct obd_histogram cl_write_rpc_hist;
265 struct obd_histogram cl_read_page_hist;
266 struct obd_histogram cl_write_page_hist;
267 struct obd_histogram cl_read_offset_hist;
268 struct obd_histogram cl_write_offset_hist;
270 /** LRU for osc caching pages */
271 struct cl_client_cache *cl_cache;
272 /** member of cl_cache->ccc_lru */
273 struct list_head cl_lru_osc;
274 /** # of available LRU slots left in the per-OSC cache.
275 * Available LRU slots are shared by all OSCs of the same file system,
276 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
277 atomic_long_t *cl_lru_left;
278 /** # of busy LRU pages. A page is considered busy if it's in writeback
279 * queue, or in transfer. Busy pages can't be discarded so they are not
281 atomic_long_t cl_lru_busy;
282 /** # of LRU pages in the cache for this client_obd */
283 atomic_long_t cl_lru_in_list;
284 /** # of threads are shrinking LRU cache. To avoid contention, it's not
285 * allowed to have multiple threads shrinking LRU cache. */
286 atomic_t cl_lru_shrinkers;
287 /** The time when this LRU cache was last used. */
288 time64_t cl_lru_last_used;
289 /** stats: how many reclaims have happened for this client_obd.
290 * reclaim and shrink - shrink is async, voluntarily rebalancing;
291 * reclaim is sync, initiated by IO thread when the LRU slots are
293 __u64 cl_lru_reclaim;
294 /** List of LRU pages for this client_obd */
295 struct list_head cl_lru_list;
296 /** Lock for LRU page list */
297 spinlock_t cl_lru_list_lock;
298 /** # of unstable pages in this client_obd.
299 * An unstable page is a page state that WRITE RPC has finished but
300 * the transaction has NOT yet committed. */
301 atomic_long_t cl_unstable_count;
302 /** Link to osc_shrinker_list */
303 struct list_head cl_shrink_list;
305 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
306 atomic_t cl_destroy_in_flight;
307 wait_queue_head_t cl_destroy_waitq;
309 /* modify rpcs in flight
310 * currently used for metadata only */
311 spinlock_t cl_mod_rpcs_lock;
312 __u16 cl_max_mod_rpcs_in_flight;
313 __u16 cl_mod_rpcs_in_flight;
314 __u16 cl_close_rpcs_in_flight;
315 wait_queue_head_t cl_mod_rpcs_waitq;
316 unsigned long *cl_mod_tag_bitmap;
317 struct obd_histogram cl_mod_rpcs_hist;
320 struct mutex cl_mgc_mutex;
321 struct local_oid_storage *cl_mgc_los;
322 struct dt_object *cl_mgc_configs_dir;
323 struct obd_export *cl_mgc_mgsexp;
324 atomic_t cl_mgc_refcount;
325 /* in-flight control list and total RPCs counter */
326 struct list_head cl_flight_waiters;
327 __u32 cl_rpcs_in_flight;
329 /* checksumming for data sent over the network */
330 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
331 cl_checksum_dump:1; /* same */
332 /* supported checksum types that are worked out at connect time */
333 __u32 cl_supp_cksum_types;
334 /* checksum algorithm to be used */
335 enum cksum_types cl_cksum_type;
337 /* also protected by the poorly named _loi_list_lock lock above */
338 struct osc_async_rc cl_ar;
340 /* sequence manager */
341 struct lu_client_seq *cl_seq;
342 struct rw_semaphore cl_seq_rwsem;
344 atomic_t cl_resends; /* resend count */
346 /* ptlrpc work for writeback in ptlrpcd context */
347 void *cl_writeback_work;
349 struct mutex cl_quota_mutex;
350 /* hash tables for osc_quota_info */
351 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
352 /* the xid of the request updating the hash tables */
353 __u64 cl_quota_last_xid;
354 /* Links to the global list of registered changelog devices */
355 struct list_head cl_chg_dev_linkage;
357 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
364 struct echo_client_obd {
365 struct obd_export *ec_exp; /* the local connection to osc/lov */
367 struct list_head ec_objects;
368 struct list_head ec_locks;
372 /* Generic subset of OSTs */
374 __u32 *op_array; /* array of index of
376 unsigned int op_count; /* number of OSTs in the array */
377 unsigned int op_size; /* allocated size of lp_array */
378 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
381 /* allow statfs data caching for 1 second */
382 #define OBD_STATFS_CACHE_SECONDS 1
384 struct lov_tgt_desc {
385 struct list_head ltd_kill;
386 struct obd_uuid ltd_uuid;
387 struct obd_device *ltd_obd;
388 struct obd_export *ltd_exp;
390 __u32 ltd_index; /* index in lov_obd->tgts */
391 unsigned long ltd_active:1,/* is this target up for requests */
392 ltd_activate:1,/* should target be activated */
393 ltd_reap:1; /* should this target be deleted */
396 struct lov_md_tgt_desc {
397 struct obd_device *lmtd_mdc;
402 struct lov_desc desc;
403 struct lov_tgt_desc **lov_tgts; /* sparse array */
404 struct ost_pool lov_packed; /* all OSTs in a packed
406 struct mutex lov_lock;
407 struct obd_connect_data lov_ocd;
408 struct proc_dir_entry *targets_proc_entry;
409 atomic_t lov_refcount;
410 __u32 lov_death_row; /* tgts scheduled to be deleted */
411 __u32 lov_tgt_size; /* size of tgts array */
414 struct cfs_hash *lov_pools_hash_body; /* used for key access */
415 struct list_head lov_pool_list; /* used for sequential access */
416 struct proc_dir_entry *lov_pool_proc_entry;
417 enum lustre_sec_part lov_sp_me;
419 /* Cached LRU and unstable data from upper layer */
420 struct cl_client_cache *lov_cache;
422 struct rw_semaphore lov_notify_lock;
423 /* Data-on-MDT: MDC array */
424 struct lov_md_tgt_desc *lov_mdc_tgts;
426 struct kobject *lov_tgts_kobj;
429 struct lmv_tgt_desc {
430 struct obd_uuid ltd_uuid;
431 struct obd_device *ltd_obd;
432 struct obd_export *ltd_exp;
434 struct mutex ltd_fid_mutex;
435 unsigned long ltd_active:1; /* target up for requests */
439 struct lu_client_fld lmv_fld;
441 struct lmv_desc desc;
443 struct mutex lmv_init_mutex;
447 u32 lmv_statfs_start;
449 u32 tgts_size; /* size of tgts array */
450 struct lmv_tgt_desc **tgts;
452 struct obd_connect_data conn_data;
453 struct kobject *lmv_tgts_kobj;
456 /* Minimum sector size is 512 */
457 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
459 struct niobuf_local {
460 __u64 lnb_file_offset;
461 __u32 lnb_page_offset;
465 struct page *lnb_page;
467 __u16 lnb_guards[MAX_GUARD_NUMBER];
468 __u16 lnb_guard_rpc:1;
469 __u16 lnb_guard_disk:1;
472 struct tgt_thread_big_cache {
473 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
476 #define LUSTRE_FLD_NAME "fld"
477 #define LUSTRE_SEQ_NAME "seq"
479 #define LUSTRE_MDD_NAME "mdd"
480 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
481 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
482 #define LUSTRE_VVP_NAME "vvp"
483 #define LUSTRE_LMV_NAME "lmv"
484 #define LUSTRE_SLP_NAME "slp"
485 #define LUSTRE_LOD_NAME "lod"
486 #define LUSTRE_OSP_NAME "osp"
487 #define LUSTRE_LWP_NAME "lwp"
489 /* obd device type names */
490 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
491 #define LUSTRE_MDS_NAME "mds"
492 #define LUSTRE_MDT_NAME "mdt"
493 #define LUSTRE_MDC_NAME "mdc"
494 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
495 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
496 #define LUSTRE_OSC_NAME "osc"
497 #define LUSTRE_LOV_NAME "lov"
498 #define LUSTRE_MGS_NAME "mgs"
499 #define LUSTRE_MGC_NAME "mgc"
501 #define LUSTRE_ECHO_NAME "obdecho"
502 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
503 #define LUSTRE_QMT_NAME "qmt"
505 /* Constant obd names (post-rename) */
506 #define LUSTRE_MDS_OBDNAME "MDS"
507 #define LUSTRE_OSS_OBDNAME "OSS"
508 #define LUSTRE_MGS_OBDNAME "MGS"
509 #define LUSTRE_MGC_OBDNAME "MGC"
511 static inline int is_lwp_on_mdt(char *name)
515 ptr = strrchr(name, '-');
517 CERROR("%s is not a obdname\n", name);
521 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
523 if (strncmp(ptr + 1, "MDT", 3) != 0)
526 while (*(--ptr) != '-' && ptr != name);
531 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
537 static inline int is_lwp_on_ost(char *name)
541 ptr = strrchr(name, '-');
543 CERROR("%s is not a obdname\n", name);
547 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
549 if (strncmp(ptr + 1, "OST", 3) != 0)
552 while (*(--ptr) != '-' && ptr != name);
557 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
564 * Events signalled through obd_notify() upcall-chain.
566 enum obd_notify_event {
567 /* Device connect start */
569 /* Device activated */
571 /* Device deactivated */
573 /* Connect data for import were changed */
575 /* Administratively deactivate/activate event */
576 OBD_NOTIFY_DEACTIVATE,
581 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
582 * being main example).
584 struct obd_notify_upcall {
585 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
586 enum obd_notify_event ev, void *owner);
587 /* Opaque datum supplied by upper layer listener */
591 struct target_recovery_data {
592 svc_handler_t trd_recovery_handler;
593 pid_t trd_processing_task;
594 struct completion trd_starting;
595 struct completion trd_finishing;
598 struct obd_llog_group {
599 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
600 wait_queue_head_t olg_waitq;
602 struct mutex olg_cat_processing;
605 /* corresponds to one of the obd's */
606 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
609 struct obd_type *obd_type;
610 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
611 int obd_minor; /* device number: lctl dl */
612 struct lu_device *obd_lu_dev;
614 /* common and UUID name of this device */
615 struct obd_uuid obd_uuid;
616 char obd_name[MAX_OBD_NAME];
618 /* bitfield modification is protected by obd_dev_lock */
620 obd_attached:1, /* finished attach */
621 obd_set_up:1, /* finished setup */
622 obd_recovering:1, /* there are recoverable clients */
623 obd_abort_recovery:1, /* recovery expired */
624 obd_version_recov:1, /* obd uses version checking */
625 obd_replayable:1, /* recovery enabled; inform clients */
626 obd_no_transno:1, /* no committed-transno notification */
627 obd_no_recov:1, /* fail instead of retry messages */
628 obd_stopping:1, /* started cleanup */
629 obd_starting:1, /* started setup */
630 obd_force:1, /* cleanup with > 0 obd refcount */
631 obd_fail:1, /* cleanup with failover */
632 obd_no_conn:1, /* deny new connections */
633 obd_inactive:1, /* device active/inactive
634 * (for /proc/status only!!) */
635 obd_no_ir:1, /* no imperative recovery. */
636 obd_process_conf:1, /* device is processing mgs config */
637 obd_checksum_dump:1; /* dump pages upon cksum error */
639 /* use separate field as it is set in interrupt to don't mess with
640 * protection of other bits using _bh lock */
641 unsigned long obd_recovery_expired:1;
642 /* uuid-export hash body */
643 struct cfs_hash *obd_uuid_hash;
644 /* nid-export hash body */
645 struct cfs_hash *obd_nid_hash;
647 struct cfs_hash *obd_nid_stats_hash;
648 /* client_generation-export hash body */
649 struct cfs_hash *obd_gen_hash;
650 struct list_head obd_nid_stats;
651 struct list_head obd_exports;
652 struct list_head obd_unlinked_exports;
653 struct list_head obd_delayed_exports;
654 struct list_head obd_lwp_list;
655 atomic_t obd_refcount;
657 spinlock_t obd_nid_lock;
658 struct ldlm_namespace *obd_namespace;
659 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
660 /* a spinlock is OK for what we do now, may need a semaphore later */
661 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
662 spinlock_t obd_osfs_lock;
663 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
664 time64_t obd_osfs_age;
665 __u64 obd_last_committed;
666 struct mutex obd_dev_mutex;
667 struct lvfs_run_ctxt obd_lvfs_ctxt;
668 struct obd_llog_group obd_olg; /* default llog group */
669 struct obd_device *obd_observer;
670 struct rw_semaphore obd_observer_link_sem;
671 struct obd_notify_upcall obd_upcall;
672 struct obd_export *obd_self_export;
673 struct obd_export *obd_lwp_export;
674 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
675 struct list_head obd_exports_timed;
676 time64_t obd_eviction_timer; /* for ping evictor */
678 int obd_max_recoverable_clients;
679 atomic_t obd_connected_clients;
680 int obd_stale_clients;
681 /* this lock protects all recovery list_heads, timer and
682 * obd_next_recovery_transno value */
683 spinlock_t obd_recovery_task_lock;
684 __u64 obd_next_recovery_transno;
685 int obd_replayed_requests;
686 int obd_requests_queued_for_recovery;
687 wait_queue_head_t obd_next_transno_waitq;
688 /* protected by obd_recovery_task_lock */
689 struct timer_list obd_recovery_timer;
691 time64_t obd_recovery_start;
692 /* seconds, for lprocfs_status */
693 time64_t obd_recovery_end;
694 time64_t obd_recovery_time_hard;
695 time64_t obd_recovery_timeout;
696 int obd_recovery_ir_factor;
698 /* new recovery stuff from CMD2 */
699 int obd_replayed_locks;
700 atomic_t obd_req_replay_clients;
701 atomic_t obd_lock_replay_clients;
702 struct target_recovery_data obd_recovery_data;
704 /* all lists are protected by obd_recovery_task_lock */
705 struct list_head obd_req_replay_queue;
706 struct list_head obd_lock_replay_queue;
707 struct list_head obd_final_req_queue;
710 #ifdef HAVE_SERVER_SUPPORT
711 struct obd_device_target obt;
712 struct filter_obd filter;
714 struct echo_obd echo;
716 struct client_obd cli;
717 struct echo_client_obd echo_client;
722 /* Fields used by LProcFS */
723 struct lprocfs_stats *obd_stats;
725 struct lprocfs_stats *obd_md_stats;
727 struct dentry *obd_debugfs_entry;
728 struct proc_dir_entry *obd_proc_entry;
729 struct proc_dir_entry *obd_proc_exports_entry;
730 struct dentry *obd_svc_debugfs_entry;
731 struct lprocfs_stats *obd_svc_stats;
732 const struct attribute **obd_attrs;
733 struct lprocfs_vars *obd_vars;
734 atomic_t obd_evict_inprogress;
735 wait_queue_head_t obd_evict_inprogress_waitq;
736 struct list_head obd_evict_list; /* protected with pet_lock */
739 * LDLM pool part. Save last calculated SLV and Limit.
741 rwlock_t obd_pool_lock;
745 int obd_conn_inprogress;
748 * List of outstanding class_incref()'s fo this OBD. For debugging. */
749 struct lu_ref obd_reference;
751 struct kset obd_kset; /* sysfs object collection */
752 struct kobj_type obd_ktype;
753 struct completion obd_kobj_unregister;
756 /* get/set_info keys */
757 #define KEY_ASYNC "async"
758 #define KEY_CHANGELOG_CLEAR "changelog_clear"
759 #define KEY_FID2PATH "fid2path"
760 #define KEY_CHECKSUM "checksum"
761 #define KEY_CLEAR_FS "clear_fs"
762 #define KEY_CONN_DATA "conn_data"
763 #define KEY_EVICT_BY_NID "evict_by_nid"
764 #define KEY_FIEMAP "fiemap"
765 #define KEY_FLUSH_CTX "flush_ctx"
766 #define KEY_GRANT_SHRINK "grant_shrink"
767 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
768 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
769 #define KEY_INTERMDS "inter_mds"
770 #define KEY_LAST_ID "last_id"
771 #define KEY_LAST_FID "last_fid"
772 #define KEY_MAX_EASIZE "max_easize"
773 #define KEY_DEFAULT_EASIZE "default_easize"
774 #define KEY_MGSSEC "mgssec"
775 #define KEY_READ_ONLY "read-only"
776 #define KEY_REGISTER_TARGET "register_target"
777 #define KEY_SET_FS "set_fs"
778 #define KEY_TGT_COUNT "tgt_count"
779 /* KEY_SET_INFO in lustre_idl.h */
780 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
782 #define KEY_CACHE_SET "cache_set"
783 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
784 #define KEY_OSP_CONNECTED "osp_connected"
786 /* Flags for op_xvalid */
788 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
789 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
790 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
791 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
792 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
793 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
794 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
799 static inline int it_to_lock_mode(struct lookup_intent *it)
801 /* CREAT needs to be tested before open (both could be set) */
802 if (it->it_op & IT_CREAT)
804 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
806 else if (it->it_op & IT_LAYOUT)
807 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
808 else if (it->it_op & IT_READDIR)
810 else if (it->it_op & IT_GETXATTR)
813 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
818 MF_MDC_CANCEL_FID1 = 1 << 0,
819 MF_MDC_CANCEL_FID2 = 1 << 1,
820 MF_MDC_CANCEL_FID3 = 1 << 2,
821 MF_MDC_CANCEL_FID4 = 1 << 3,
822 MF_GET_MDT_IDX = 1 << 4,
826 CLI_SET_MEA = 1 << 0,
827 CLI_RM_ENTRY = 1 << 1,
830 CLI_MIGRATE = 1 << 4,
834 * GETXATTR is not included as only a couple of fields in the reply body
835 * is filled, but not FID which is needed for common intent handling in
836 * mdc_finish_intent_lock()
838 static inline bool it_has_reply_body(const struct lookup_intent *it)
840 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
844 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
845 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
846 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
847 struct lu_fid op_fid4; /* to the operation locks. */
848 u32 op_mds; /* what mds server open will go to */
850 struct lustre_handle op_open_handle;
854 struct rw_semaphore *op_mea1_sem;
855 struct rw_semaphore *op_mea2_sem;
856 struct lmv_stripe_md *op_mea1;
857 struct lmv_stripe_md *op_mea2;
858 __u32 op_suppgids[2];
865 /* iattr fields and blocks. */
866 struct iattr op_attr;
867 enum op_xvalid op_xvalid; /* eXtra validity flags */
868 loff_t op_attr_blocks;
869 u64 op_valid; /* OBD_MD_* */
870 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
872 enum md_op_flags op_flags;
874 /* Various operation flags. */
875 enum mds_op_bias op_bias;
877 /* used to transfer info between the stacks of MD client
878 * see enum op_cli_flags */
879 enum md_cli_flags op_cli_flags;
881 /* File object data version for HSM release, on client */
882 __u64 op_data_version;
883 struct lustre_handle op_lease_handle;
885 /* File security context, for creates. */
886 const char *op_file_secctx_name;
887 void *op_file_secctx;
888 __u32 op_file_secctx_size;
890 /* default stripe offset */
891 __u32 op_default_stripe_offset;
895 /* Used by readdir */
896 unsigned int op_max_pages;
901 * used to access migrating dir: if it's set, assume migration is
902 * finished, use the new layout to access dir, otherwise use old layout.
903 * By default it's not set, because new files are created under new
904 * layout, if we can't find file with name under both old and new
905 * layout, we are sure file with name doesn't exist, but in reverse
906 * order there may be a race with creation by others.
908 bool op_post_migrate;
909 /* used to access dir with bash hash */
910 __u32 op_stripe_index;
914 int (*md_blocking_ast)(struct ldlm_lock *lock,
915 struct ldlm_lock_desc *desc,
916 void *data, int flag);
919 struct md_enqueue_info;
920 /* metadata stat-ahead */
921 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
922 struct md_enqueue_info *minfo,
925 struct md_enqueue_info {
926 struct md_op_data mi_data;
927 struct lookup_intent mi_it;
928 struct lustre_handle mi_lockh;
929 struct inode *mi_dir;
930 struct ldlm_enqueue_info mi_einfo;
931 md_enqueue_cb_t mi_cb;
936 struct module *o_owner;
937 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
938 void *karg, void __user *uarg);
939 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
940 __u32 keylen, void *key, __u32 *vallen, void *val);
941 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
942 __u32 keylen, void *key,
943 __u32 vallen, void *val,
944 struct ptlrpc_request_set *set);
945 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
946 int (*o_precleanup)(struct obd_device *dev);
947 int (*o_cleanup)(struct obd_device *dev);
948 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
949 int (*o_postrecov)(struct obd_device *dev);
950 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
952 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
953 /* connect to the target device with given connection
954 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
955 * granted by the target, which are guaranteed to be a subset of flags
956 * asked for. If @ocd == NULL, use default parameters. */
957 int (*o_connect)(const struct lu_env *env,
958 struct obd_export **exp, struct obd_device *src,
959 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
961 int (*o_reconnect)(const struct lu_env *env,
962 struct obd_export *exp, struct obd_device *src,
963 struct obd_uuid *cluuid,
964 struct obd_connect_data *ocd,
966 int (*o_disconnect)(struct obd_export *exp);
968 /* Initialize/finalize fids infrastructure. */
969 int (*o_fid_init)(struct obd_device *obd,
970 struct obd_export *exp, enum lu_cli_type type);
971 int (*o_fid_fini)(struct obd_device *obd);
973 /* Allocate new fid according to passed @hint. */
974 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
975 struct lu_fid *fid, struct md_op_data *op_data);
978 * Object with @fid is getting deleted, we may want to do something
981 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
982 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
983 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
984 time64_t max_age, struct ptlrpc_request_set *set);
985 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
987 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
989 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
991 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
993 int (*o_preprw)(const struct lu_env *env, int cmd,
994 struct obd_export *exp, struct obdo *oa, int objcount,
995 struct obd_ioobj *obj, struct niobuf_remote *remote,
996 int *nr_pages, struct niobuf_local *local);
997 int (*o_commitrw)(const struct lu_env *env, int cmd,
998 struct obd_export *exp, struct obdo *oa,
999 int objcount, struct obd_ioobj *obj,
1000 struct niobuf_remote *remote, int pages,
1001 struct niobuf_local *local, int rc);
1002 int (*o_init_export)(struct obd_export *exp);
1003 int (*o_destroy_export)(struct obd_export *exp);
1005 int (*o_import_event)(struct obd_device *, struct obd_import *,
1006 enum obd_import_event);
1008 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1009 enum obd_notify_event ev);
1011 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1012 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1015 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1016 struct obd_quotactl *);
1019 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1020 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1021 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1023 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1027 /* lmv structures */
1029 struct mdt_body *body;
1030 struct lu_buf layout;
1031 struct lmv_stripe_md *lmv;
1032 #ifdef CONFIG_FS_POSIX_ACL
1033 struct posix_acl *posix_acl;
1037 struct md_open_data {
1038 struct obd_client_handle *mod_och;
1039 struct ptlrpc_request *mod_open_req;
1040 struct ptlrpc_request *mod_close_req;
1041 atomic_t mod_refcount;
1045 struct obd_client_handle {
1046 struct lustre_handle och_open_handle;
1047 struct lu_fid och_fid;
1048 struct md_open_data *och_mod;
1049 struct lustre_handle och_lease_handle; /* open lock for lease */
1054 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1056 struct lookup_intent;
1060 int (*m_close)(struct obd_export *, struct md_op_data *,
1061 struct md_open_data *, struct ptlrpc_request **);
1063 int (*m_create)(struct obd_export *, struct md_op_data *,
1064 const void *, size_t, umode_t, uid_t, gid_t,
1065 cfs_cap_t, __u64, struct ptlrpc_request **);
1067 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1068 const union ldlm_policy_data *, struct md_op_data *,
1069 struct lustre_handle *, __u64);
1071 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1072 struct ptlrpc_request **);
1074 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1075 struct lookup_intent *,
1076 struct ptlrpc_request **,
1077 ldlm_blocking_callback, __u64);
1079 int (*m_link)(struct obd_export *, struct md_op_data *,
1080 struct ptlrpc_request **);
1082 int (*m_rename)(struct obd_export *, struct md_op_data *,
1083 const char *, size_t, const char *, size_t,
1084 struct ptlrpc_request **);
1086 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1087 size_t , struct ptlrpc_request **);
1089 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1090 struct ptlrpc_request **);
1092 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1093 struct md_callback *cb_op, __u64 hash_offset,
1094 struct page **ppage);
1096 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1097 struct ptlrpc_request **);
1099 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1100 u64, const char *, const void *, size_t, unsigned int,
1101 u32, struct ptlrpc_request **);
1103 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1104 u64, const char *, size_t, struct ptlrpc_request **);
1106 int (*m_intent_getattr_async)(struct obd_export *,
1107 struct md_enqueue_info *);
1109 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1110 struct lu_fid *, __u64 *bits);
1112 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1114 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1115 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1117 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1118 struct ptlrpc_request **);
1120 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1122 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1123 struct obd_export *, struct obd_export *,
1124 struct lustre_md *);
1126 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1128 int (*m_merge_attr)(struct obd_export *,
1129 const struct lmv_stripe_md *lsm,
1130 struct cl_attr *attr, ldlm_blocking_callback);
1132 int (*m_set_open_replay_data)(struct obd_export *,
1133 struct obd_client_handle *,
1134 struct lookup_intent *);
1136 int (*m_clear_open_replay_data)(struct obd_export *,
1137 struct obd_client_handle *);
1139 int (*m_set_lock_data)(struct obd_export *,
1140 const struct lustre_handle *, void *, __u64 *);
1142 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1143 const struct lu_fid *, enum ldlm_type,
1144 union ldlm_policy_data *, enum ldlm_mode,
1145 struct lustre_handle *);
1147 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1148 union ldlm_policy_data *, enum ldlm_mode,
1149 enum ldlm_cancel_flags flags, void *opaque);
1151 int (*m_get_fid_from_lsm)(struct obd_export *,
1152 const struct lmv_stripe_md *,
1153 const char *name, int namelen,
1154 struct lu_fid *fid);
1155 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1156 const union lmv_mds_md *lmv, size_t lmv_size);
1159 static inline struct md_open_data *obd_mod_alloc(void)
1161 struct md_open_data *mod;
1165 atomic_set(&mod->mod_refcount, 1);
1169 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1170 #define obd_mod_put(mod) \
1172 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1173 if ((mod)->mod_open_req) \
1174 ptlrpc_req_finished((mod)->mod_open_req); \
1175 OBD_FREE_PTR(mod); \
1179 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1180 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1181 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1183 /* return 1 if client should be resend request */
1184 static inline int client_should_resend(int resend, struct client_obd *cli)
1186 return atomic_read(&cli->cl_resends) ?
1187 atomic_read(&cli->cl_resends) > resend : 1;
1191 * Return device name for this device
1193 * XXX: lu_device is declared before obd_device, while a pointer pointing
1194 * back to obd_device in lu_device, so this helper function defines here
1195 * instead of in lu_object.h
1197 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1199 return lu_dev->ld_obd->obd_name;
1202 static inline bool filename_is_volatile(const char *name, size_t namelen,
1208 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1211 /* caller does not care of idx */
1215 /* volatile file, the MDT can be set from name */
1216 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1217 /* if no MDT is specified, use std way */
1218 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1220 /* test for no MDT idx case */
1221 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1222 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1226 /* we have an idx, read it */
1227 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1228 *idx = simple_strtoul(start, &end, 16);
1230 * no digit, no trailing :, negative value
1232 if (((*idx == 0) && (end == start)) ||
1233 (*end != ':') || (*idx < 0))
1238 /* bad format of mdt idx, we cannot return an error
1239 * to caller so we use hash algo */
1240 CERROR("Bad volatile file name format: %s\n",
1241 name + LUSTRE_VOLATILE_HDR_LEN);
1245 static inline int cli_brw_size(struct obd_device *obd)
1247 LASSERT(obd != NULL);
1248 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1252 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1253 * of the client should be increased accordingly to avoid sending fragmented
1254 * RPCs over the network when the client runs out of the maximum dirty space
1255 * when so many RPCs are being generated.
1257 static inline void client_adjust_max_dirty(struct client_obd *cli)
1260 if (cli->cl_dirty_max_pages <= 0) {
1261 cli->cl_dirty_max_pages =
1262 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1264 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1265 cli->cl_max_pages_per_rpc;
1267 if (dirty_max > cli->cl_dirty_max_pages)
1268 cli->cl_dirty_max_pages = dirty_max;
1271 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1272 cli->cl_dirty_max_pages = totalram_pages / 8;
1274 /* This value is exported to userspace through the max_dirty_mb
1275 * parameter. So we round up the number of pages to make it a round
1277 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1278 1 << (20 - PAGE_SHIFT));
1281 #endif /* __OBD_H */