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 struct obd_device *oi_obd;
91 struct lmv_tgt_desc *oi_tgt;
92 /* statfs data specific for every OSC, if needed at all. */
93 struct obd_statfs *oi_osfs;
94 /* An update callback which is called to update some data on upper
95 * level. E.g. it is used for update lsm->lsm_oinfo at every received
96 * request in osc level for enqueue requests. It is also possible to
97 * update some caller data from LOV layer if needed. */
98 obd_enqueue_update_f oi_cb_up;
102 struct list_head typ_chain;
103 struct obd_ops *typ_dt_ops;
104 struct md_ops *typ_md_ops;
105 struct proc_dir_entry *typ_procroot;
106 struct proc_dir_entry *typ_procsym;
107 struct dentry *typ_debugfs_entry;
108 #ifdef HAVE_SERVER_SUPPORT
113 struct lu_device_type *typ_lu;
114 spinlock_t obd_type_lock;
115 struct kobject typ_kobj;
125 struct timeout_item {
126 enum timeout_event ti_event;
130 struct list_head ti_obd_list;
131 struct list_head ti_chain;
134 #define OBD_MAX_RIF_DEFAULT 8
135 #define OBD_MAX_RIF_MAX 512
136 #define OSC_MAX_RIF_MAX 256
137 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
138 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
139 #define OSC_DEFAULT_RESENDS 10
141 /* possible values for fo_sync_lock_cancel */
143 NEVER_SYNC_ON_CANCEL = 0,
144 BLOCKING_SYNC_ON_CANCEL = 1,
145 ALWAYS_SYNC_ON_CANCEL = 2,
146 NUM_SYNC_ON_CANCEL_STATES
150 * Limit reply buffer size for striping data to one x86_64 page. This
151 * value is chosen to fit the striping data for common use cases while
152 * staying well below the limit at which the buffer must be backed by
153 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
156 #define OBD_MAX_DEFAULT_EA_SIZE 4096
159 * Lustre can handle larger xattrs internally, but we must respect the Linux
160 * VFS limitation or tools like tar cannot interact with Lustre volumes
163 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
166 enum obd_cl_sem_lock_class {
175 struct rw_semaphore cl_sem;
176 struct obd_uuid cl_target_uuid;
177 struct obd_import *cl_import; /* ptlrpc connection state */
178 size_t cl_conn_count;
180 /* Cache maximum and default values for easize. This is
181 * strictly a performance optimization to minimize calls to
182 * obd_size_diskmd(). The default values are used to calculate the
183 * initial size of a request buffer. The ptlrpc layer will resize the
184 * buffer as needed to accommodate a larger reply from the
185 * server. The default values should be small enough to avoid wasted
186 * memory and excessive use of vmalloc(), yet large enough to avoid
187 * reallocating the buffer in the common use case. */
189 /* Default EA size for striping attributes. It is initialized at
190 * mount-time based on the default stripe width of the filesystem,
191 * then it tracks the largest observed EA size advertised by
192 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
193 __u32 cl_default_mds_easize;
195 /* Maximum possible EA size computed at mount-time based on
196 * the number of OSTs in the filesystem. May be increased at
197 * run-time if a larger observed size is advertised by the MDT. */
198 __u32 cl_max_mds_easize;
200 /* Data-on-MDT specific value to set larger reply buffer for possible
201 * data read along with open/stat requests. By default it tries to use
202 * unused space in reply buffer.
203 * This value is used to ensure that reply buffer has at least as
204 * much free space as value indicates. That free space is gained from
205 * LOV EA buffer which is small for DoM files and on big systems can
206 * provide up to 32KB of extra space in reply buffer.
207 * Default value is 8K now.
209 __u32 cl_dom_min_inline_repsize;
211 enum lustre_sec_part cl_sp_me;
212 enum lustre_sec_part cl_sp_to;
213 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
215 /* the grant values are protected by loi_list_lock below */
216 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
217 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
218 unsigned long cl_dirty_transit; /* dirty synchronous */
219 unsigned long cl_avail_grant; /* bytes of credit for ost */
220 unsigned long cl_lost_grant; /* lost credits (trunc) */
221 /* grant consumed for dirty pages */
222 unsigned long cl_dirty_grant;
224 /* since we allocate grant by blocks, we don't know how many grant will
225 * be used to add a page into cache. As a solution, we reserve maximum
226 * grant before trying to dirty a page and unreserve the rest.
227 * See osc_{reserve|unreserve}_grant for details. */
228 long cl_reserved_grant;
229 struct list_head cl_cache_waiters; /* waiting for cache/grant */
230 time64_t cl_next_shrink_grant; /* seconds */
231 struct list_head cl_grant_chain;
232 time64_t cl_grant_shrink_interval; /* seconds */
234 /* A chunk is an optimal size used by osc_extent to determine
235 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
237 /* extent insertion metadata overhead to be accounted in grant,
239 unsigned int cl_grant_extent_tax;
240 /* maximum extent size, in number of pages */
241 unsigned int cl_max_extent_pages;
243 /* keep track of objects that have lois that contain pages which
244 * have been queued for async brw. this lock also protects the
245 * lists of osc_client_pages that hang off of the loi */
247 * ->cl_loi_list_lock protects consistency of
248 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
249 * ->ap_completion() call-backs are executed under this lock. As we
250 * cannot guarantee that these call-backs never block on all platforms
251 * (as a matter of fact they do block on Mac OS X), type of
252 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
253 * and blocking mutex on Mac OS X. (Alternative is to make this lock
254 * blocking everywhere, but we don't want to slow down fast-path of
255 * our main platform.)
257 * NB by Jinshan: though field names are still _loi_, but actually
258 * osc_object{}s are in the list.
260 spinlock_t cl_loi_list_lock;
261 struct list_head cl_loi_ready_list;
262 struct list_head cl_loi_hp_ready_list;
263 struct list_head cl_loi_write_list;
264 struct list_head cl_loi_read_list;
265 __u32 cl_r_in_flight;
266 __u32 cl_w_in_flight;
267 /* just a sum of the loi/lop pending numbers to be exported by /proc */
268 atomic_t cl_pending_w_pages;
269 atomic_t cl_pending_r_pages;
270 u32 cl_max_pages_per_rpc;
271 u32 cl_max_rpcs_in_flight;
272 u32 cl_max_short_io_bytes;
273 struct obd_histogram cl_read_rpc_hist;
274 struct obd_histogram cl_write_rpc_hist;
275 struct obd_histogram cl_read_page_hist;
276 struct obd_histogram cl_write_page_hist;
277 struct obd_histogram cl_read_offset_hist;
278 struct obd_histogram cl_write_offset_hist;
280 /** LRU for osc caching pages */
281 struct cl_client_cache *cl_cache;
282 /** member of cl_cache->ccc_lru */
283 struct list_head cl_lru_osc;
284 /** # of available LRU slots left in the per-OSC cache.
285 * Available LRU slots are shared by all OSCs of the same file system,
286 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
287 atomic_long_t *cl_lru_left;
288 /** # of busy LRU pages. A page is considered busy if it's in writeback
289 * queue, or in transfer. Busy pages can't be discarded so they are not
291 atomic_long_t cl_lru_busy;
292 /** # of LRU pages in the cache for this client_obd */
293 atomic_long_t cl_lru_in_list;
294 /** # of threads are shrinking LRU cache. To avoid contention, it's not
295 * allowed to have multiple threads shrinking LRU cache. */
296 atomic_t cl_lru_shrinkers;
297 /** The time when this LRU cache was last used. */
298 time64_t cl_lru_last_used;
299 /** stats: how many reclaims have happened for this client_obd.
300 * reclaim and shrink - shrink is async, voluntarily rebalancing;
301 * reclaim is sync, initiated by IO thread when the LRU slots are
303 __u64 cl_lru_reclaim;
304 /** List of LRU pages for this client_obd */
305 struct list_head cl_lru_list;
306 /** Lock for LRU page list */
307 spinlock_t cl_lru_list_lock;
308 /** # of unstable pages in this client_obd.
309 * An unstable page is a page state that WRITE RPC has finished but
310 * the transaction has NOT yet committed. */
311 atomic_long_t cl_unstable_count;
312 /** Link to osc_shrinker_list */
313 struct list_head cl_shrink_list;
315 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
316 atomic_t cl_destroy_in_flight;
317 wait_queue_head_t cl_destroy_waitq;
319 /* modify rpcs in flight
320 * currently used for metadata only */
321 spinlock_t cl_mod_rpcs_lock;
322 __u16 cl_max_mod_rpcs_in_flight;
323 __u16 cl_mod_rpcs_in_flight;
324 __u16 cl_close_rpcs_in_flight;
325 wait_queue_head_t cl_mod_rpcs_waitq;
326 unsigned long *cl_mod_tag_bitmap;
327 struct obd_histogram cl_mod_rpcs_hist;
330 struct mutex cl_mgc_mutex;
331 struct local_oid_storage *cl_mgc_los;
332 struct dt_object *cl_mgc_configs_dir;
333 struct obd_export *cl_mgc_mgsexp;
334 atomic_t cl_mgc_refcount;
335 /* in-flight control list and total RPCs counter */
336 struct list_head cl_flight_waiters;
337 __u32 cl_rpcs_in_flight;
339 /* checksumming for data sent over the network */
340 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
341 cl_checksum_dump:1; /* same */
342 /* supported checksum types that are worked out at connect time */
343 __u32 cl_supp_cksum_types;
344 /* checksum algorithm to be used */
345 enum cksum_types cl_cksum_type;
347 /* also protected by the poorly named _loi_list_lock lock above */
348 struct osc_async_rc cl_ar;
350 /* sequence manager */
351 struct lu_client_seq *cl_seq;
352 struct rw_semaphore cl_seq_rwsem;
354 atomic_t cl_resends; /* resend count */
356 /* ptlrpc work for writeback in ptlrpcd context */
357 void *cl_writeback_work;
359 struct mutex cl_quota_mutex;
360 /* hash tables for osc_quota_info */
361 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
362 /* the xid of the request updating the hash tables */
363 __u64 cl_quota_last_xid;
364 /* Links to the global list of registered changelog devices */
365 struct list_head cl_chg_dev_linkage;
367 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
374 struct echo_client_obd {
375 struct obd_export *ec_exp; /* the local connection to osc/lov */
377 struct list_head ec_objects;
378 struct list_head ec_locks;
382 /* Generic subset of OSTs */
384 __u32 *op_array; /* array of index of
386 unsigned int op_count; /* number of OSTs in the array */
387 unsigned int op_size; /* allocated size of lp_array */
388 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
391 /* allow statfs data caching for 1 second */
392 #define OBD_STATFS_CACHE_SECONDS 1
394 struct lov_tgt_desc {
395 struct list_head ltd_kill;
396 struct obd_uuid ltd_uuid;
397 struct obd_device *ltd_obd;
398 struct obd_export *ltd_exp;
400 __u32 ltd_index; /* index in lov_obd->tgts */
401 unsigned long ltd_active:1,/* is this target up for requests */
402 ltd_activate:1,/* should target be activated */
403 ltd_reap:1; /* should this target be deleted */
406 struct lov_md_tgt_desc {
407 struct obd_device *lmtd_mdc;
412 struct lov_desc desc;
413 struct lov_tgt_desc **lov_tgts; /* sparse array */
414 struct ost_pool lov_packed; /* all OSTs in a packed
416 struct mutex lov_lock;
417 struct obd_connect_data lov_ocd;
418 struct proc_dir_entry *targets_proc_entry;
419 atomic_t lov_refcount;
420 __u32 lov_death_row; /* tgts scheduled to be deleted */
421 __u32 lov_tgt_size; /* size of tgts array */
424 struct cfs_hash *lov_pools_hash_body; /* used for key access */
425 struct list_head lov_pool_list; /* used for sequential access */
426 struct proc_dir_entry *lov_pool_proc_entry;
427 enum lustre_sec_part lov_sp_me;
429 /* Cached LRU and unstable data from upper layer */
430 struct cl_client_cache *lov_cache;
432 struct rw_semaphore lov_notify_lock;
433 /* Data-on-MDT: MDC array */
434 struct lov_md_tgt_desc *lov_mdc_tgts;
436 struct kobject *lov_tgts_kobj;
439 struct lmv_tgt_desc {
440 struct obd_uuid ltd_uuid;
441 struct obd_device *ltd_obd;
442 struct obd_export *ltd_exp;
444 struct mutex ltd_fid_mutex;
445 struct obd_statfs ltd_statfs;
446 time64_t ltd_statfs_age;
447 unsigned long ltd_active:1; /* target up for requests */
451 struct lu_client_fld lmv_fld;
453 struct lmv_desc desc;
455 struct mutex lmv_init_mutex;
459 u32 lmv_statfs_start;
461 u32 tgts_size; /* size of tgts array */
462 struct lmv_tgt_desc **tgts;
464 struct obd_connect_data conn_data;
465 struct kobject *lmv_tgts_kobj;
469 /* Minimum sector size is 512 */
470 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
472 struct niobuf_local {
473 __u64 lnb_file_offset;
474 __u32 lnb_page_offset;
478 struct page *lnb_page;
480 __u16 lnb_guards[MAX_GUARD_NUMBER];
481 __u16 lnb_guard_rpc:1;
482 __u16 lnb_guard_disk:1;
485 struct tgt_thread_big_cache {
486 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
489 #define LUSTRE_FLD_NAME "fld"
490 #define LUSTRE_SEQ_NAME "seq"
492 #define LUSTRE_MDD_NAME "mdd"
493 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
494 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
495 #define LUSTRE_VVP_NAME "vvp"
496 #define LUSTRE_LMV_NAME "lmv"
497 #define LUSTRE_SLP_NAME "slp"
498 #define LUSTRE_LOD_NAME "lod"
499 #define LUSTRE_OSP_NAME "osp"
500 #define LUSTRE_LWP_NAME "lwp"
502 /* obd device type names */
503 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
504 #define LUSTRE_MDS_NAME "mds"
505 #define LUSTRE_MDT_NAME "mdt"
506 #define LUSTRE_MDC_NAME "mdc"
507 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
508 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
509 #define LUSTRE_OSC_NAME "osc"
510 #define LUSTRE_LOV_NAME "lov"
511 #define LUSTRE_MGS_NAME "mgs"
512 #define LUSTRE_MGC_NAME "mgc"
514 #define LUSTRE_ECHO_NAME "obdecho"
515 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
516 #define LUSTRE_QMT_NAME "qmt"
518 /* Constant obd names (post-rename) */
519 #define LUSTRE_MDS_OBDNAME "MDS"
520 #define LUSTRE_OSS_OBDNAME "OSS"
521 #define LUSTRE_MGS_OBDNAME "MGS"
522 #define LUSTRE_MGC_OBDNAME "MGC"
524 static inline int is_lwp_on_mdt(char *name)
528 ptr = strrchr(name, '-');
530 CERROR("%s is not a obdname\n", name);
534 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
536 if (strncmp(ptr + 1, "MDT", 3) != 0)
539 while (*(--ptr) != '-' && ptr != name);
544 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
550 static inline int is_lwp_on_ost(char *name)
554 ptr = strrchr(name, '-');
556 CERROR("%s is not a obdname\n", name);
560 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
562 if (strncmp(ptr + 1, "OST", 3) != 0)
565 while (*(--ptr) != '-' && ptr != name);
570 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
577 * Events signalled through obd_notify() upcall-chain.
579 enum obd_notify_event {
580 /* Device connect start */
582 /* Device activated */
584 /* Device deactivated */
586 /* Connect data for import were changed */
588 /* Administratively deactivate/activate event */
589 OBD_NOTIFY_DEACTIVATE,
594 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
595 * being main example).
597 struct obd_notify_upcall {
598 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
599 enum obd_notify_event ev, void *owner);
600 /* Opaque datum supplied by upper layer listener */
604 struct target_recovery_data {
605 svc_handler_t trd_recovery_handler;
606 pid_t trd_processing_task;
607 struct completion trd_starting;
608 struct completion trd_finishing;
611 struct obd_llog_group {
612 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
613 wait_queue_head_t olg_waitq;
615 struct mutex olg_cat_processing;
618 /* corresponds to one of the obd's */
619 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
622 struct obd_type *obd_type;
623 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
624 int obd_minor; /* device number: lctl dl */
625 struct lu_device *obd_lu_dev;
627 /* common and UUID name of this device */
628 struct obd_uuid obd_uuid;
629 char obd_name[MAX_OBD_NAME];
631 /* bitfield modification is protected by obd_dev_lock */
633 obd_attached:1, /* finished attach */
634 obd_set_up:1, /* finished setup */
635 obd_recovering:1, /* there are recoverable clients */
636 obd_abort_recovery:1, /* recovery expired */
637 obd_version_recov:1, /* obd uses version checking */
638 obd_replayable:1, /* recovery enabled; inform clients */
639 obd_no_transno:1, /* no committed-transno notification */
640 obd_no_recov:1, /* fail instead of retry messages */
641 obd_stopping:1, /* started cleanup */
642 obd_starting:1, /* started setup */
643 obd_force:1, /* cleanup with > 0 obd refcount */
644 obd_fail:1, /* cleanup with failover */
645 obd_no_conn:1, /* deny new connections */
646 obd_inactive:1, /* device active/inactive
647 * (for /proc/status only!!) */
648 obd_no_ir:1, /* no imperative recovery. */
649 obd_process_conf:1, /* device is processing mgs config */
650 obd_checksum_dump:1; /* dump pages upon cksum error */
652 /* use separate field as it is set in interrupt to don't mess with
653 * protection of other bits using _bh lock */
654 unsigned long obd_recovery_expired:1;
655 /* uuid-export hash body */
656 struct cfs_hash *obd_uuid_hash;
657 /* nid-export hash body */
658 struct cfs_hash *obd_nid_hash;
660 struct cfs_hash *obd_nid_stats_hash;
661 /* client_generation-export hash body */
662 struct cfs_hash *obd_gen_hash;
663 struct list_head obd_nid_stats;
664 struct list_head obd_exports;
665 struct list_head obd_unlinked_exports;
666 struct list_head obd_delayed_exports;
667 struct list_head obd_lwp_list;
668 atomic_t obd_refcount;
670 spinlock_t obd_nid_lock;
671 struct ldlm_namespace *obd_namespace;
672 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
673 /* a spinlock is OK for what we do now, may need a semaphore later */
674 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
675 spinlock_t obd_osfs_lock;
676 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
677 time64_t obd_osfs_age;
678 __u64 obd_last_committed;
679 struct mutex obd_dev_mutex;
680 struct lvfs_run_ctxt obd_lvfs_ctxt;
681 struct obd_llog_group obd_olg; /* default llog group */
682 struct obd_device *obd_observer;
683 struct rw_semaphore obd_observer_link_sem;
684 struct obd_notify_upcall obd_upcall;
685 struct obd_export *obd_self_export;
686 struct obd_export *obd_lwp_export;
687 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
688 struct list_head obd_exports_timed;
689 time64_t obd_eviction_timer; /* for ping evictor */
691 int obd_max_recoverable_clients;
692 atomic_t obd_connected_clients;
693 int obd_stale_clients;
694 /* this lock protects all recovery list_heads, timer and
695 * obd_next_recovery_transno value */
696 spinlock_t obd_recovery_task_lock;
697 __u64 obd_next_recovery_transno;
698 int obd_replayed_requests;
699 int obd_requests_queued_for_recovery;
700 wait_queue_head_t obd_next_transno_waitq;
701 /* protected by obd_recovery_task_lock */
702 struct hrtimer obd_recovery_timer;
704 time64_t obd_recovery_start;
705 /* seconds, for lprocfs_status */
706 time64_t obd_recovery_end;
707 time64_t obd_recovery_time_hard;
708 time64_t obd_recovery_timeout;
709 int obd_recovery_ir_factor;
711 /* new recovery stuff from CMD2 */
712 int obd_replayed_locks;
713 atomic_t obd_req_replay_clients;
714 atomic_t obd_lock_replay_clients;
715 struct target_recovery_data obd_recovery_data;
717 /* all lists are protected by obd_recovery_task_lock */
718 struct list_head obd_req_replay_queue;
719 struct list_head obd_lock_replay_queue;
720 struct list_head obd_final_req_queue;
723 #ifdef HAVE_SERVER_SUPPORT
724 struct obd_device_target obt;
725 struct filter_obd filter;
727 struct echo_obd echo;
729 struct client_obd cli;
730 struct echo_client_obd echo_client;
735 /* Fields used by LProcFS */
736 struct lprocfs_stats *obd_stats;
738 struct lprocfs_stats *obd_md_stats;
740 struct dentry *obd_debugfs_entry;
741 struct proc_dir_entry *obd_proc_entry;
742 struct proc_dir_entry *obd_proc_exports_entry;
743 struct dentry *obd_svc_debugfs_entry;
744 struct lprocfs_stats *obd_svc_stats;
745 const struct attribute **obd_attrs;
746 struct lprocfs_vars *obd_vars;
747 atomic_t obd_evict_inprogress;
748 wait_queue_head_t obd_evict_inprogress_waitq;
749 struct list_head obd_evict_list; /* protected with pet_lock */
752 * LDLM pool part. Save last calculated SLV and Limit.
754 rwlock_t obd_pool_lock;
758 int obd_conn_inprogress;
761 * List of outstanding class_incref()'s fo this OBD. For debugging. */
762 struct lu_ref obd_reference;
764 struct kset obd_kset; /* sysfs object collection */
765 struct kobj_type obd_ktype;
766 struct completion obd_kobj_unregister;
769 /* get/set_info keys */
770 #define KEY_ASYNC "async"
771 #define KEY_CHANGELOG_CLEAR "changelog_clear"
772 #define KEY_FID2PATH "fid2path"
773 #define KEY_CHECKSUM "checksum"
774 #define KEY_CLEAR_FS "clear_fs"
775 #define KEY_CONN_DATA "conn_data"
776 #define KEY_EVICT_BY_NID "evict_by_nid"
777 #define KEY_FIEMAP "fiemap"
778 #define KEY_FLUSH_CTX "flush_ctx"
779 #define KEY_GRANT_SHRINK "grant_shrink"
780 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
781 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
782 #define KEY_INTERMDS "inter_mds"
783 #define KEY_LAST_ID "last_id"
784 #define KEY_LAST_FID "last_fid"
785 #define KEY_MAX_EASIZE "max_easize"
786 #define KEY_DEFAULT_EASIZE "default_easize"
787 #define KEY_MGSSEC "mgssec"
788 #define KEY_READ_ONLY "read-only"
789 #define KEY_REGISTER_TARGET "register_target"
790 #define KEY_SET_FS "set_fs"
791 #define KEY_TGT_COUNT "tgt_count"
792 /* KEY_SET_INFO in lustre_idl.h */
793 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
795 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
796 #define KEY_OSP_CONNECTED "osp_connected"
798 /* Flags for op_xvalid */
800 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
801 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
802 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
803 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
804 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
805 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
806 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
811 static inline int it_to_lock_mode(struct lookup_intent *it)
813 /* CREAT needs to be tested before open (both could be set) */
814 if (it->it_op & IT_CREAT)
816 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
818 else if (it->it_op & IT_LAYOUT)
819 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
820 else if (it->it_op & IT_READDIR)
822 else if (it->it_op & IT_GETXATTR)
825 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
830 MF_MDC_CANCEL_FID1 = 1 << 0,
831 MF_MDC_CANCEL_FID2 = 1 << 1,
832 MF_MDC_CANCEL_FID3 = 1 << 2,
833 MF_MDC_CANCEL_FID4 = 1 << 3,
834 MF_GET_MDT_IDX = 1 << 4,
838 CLI_SET_MEA = 1 << 0,
839 CLI_RM_ENTRY = 1 << 1,
842 CLI_MIGRATE = 1 << 4,
846 LUSTRE_OPC_MKDIR = 0,
847 LUSTRE_OPC_SYMLINK = 1,
848 LUSTRE_OPC_MKNOD = 2,
849 LUSTRE_OPC_CREATE = 3,
854 * GETXATTR is not included as only a couple of fields in the reply body
855 * is filled, but not FID which is needed for common intent handling in
856 * mdc_finish_intent_lock()
858 static inline bool it_has_reply_body(const struct lookup_intent *it)
860 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
864 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
865 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
866 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
867 struct lu_fid op_fid4; /* to the operation locks. */
868 u32 op_mds; /* what mds server open will go to */
870 enum md_op_code op_code;
871 struct lustre_handle op_open_handle;
875 struct rw_semaphore *op_mea1_sem;
876 struct rw_semaphore *op_mea2_sem;
877 struct lmv_stripe_md *op_mea1;
878 struct lmv_stripe_md *op_mea2;
879 struct lmv_stripe_md *op_default_mea1; /* default LMV */
880 __u32 op_suppgids[2];
887 /* iattr fields and blocks. */
888 struct iattr op_attr;
889 enum op_xvalid op_xvalid; /* eXtra validity flags */
890 loff_t op_attr_blocks;
891 u64 op_valid; /* OBD_MD_* */
892 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
894 enum md_op_flags op_flags;
896 /* Various operation flags. */
897 enum mds_op_bias op_bias;
899 /* used to transfer info between the stacks of MD client
900 * see enum op_cli_flags */
901 enum md_cli_flags op_cli_flags;
903 /* File object data version for HSM release, on client */
904 __u64 op_data_version;
905 struct lustre_handle op_lease_handle;
907 /* File security context, for creates/metadata ops */
908 const char *op_file_secctx_name;
909 __u32 op_file_secctx_name_size;
910 void *op_file_secctx;
911 __u32 op_file_secctx_size;
915 /* Used by readdir */
916 unsigned int op_max_pages;
921 * used to access migrating dir: if it's set, assume migration is
922 * finished, use the new layout to access dir, otherwise use old layout.
923 * By default it's not set, because new files are created under new
924 * layout, if we can't find file with name under both old and new
925 * layout, we are sure file with name doesn't exist, but in reverse
926 * order there may be a race with creation by others.
928 bool op_post_migrate;
929 /* used to access dir with bash hash */
930 __u32 op_stripe_index;
934 int (*md_blocking_ast)(struct ldlm_lock *lock,
935 struct ldlm_lock_desc *desc,
936 void *data, int flag);
939 struct md_enqueue_info;
940 /* metadata stat-ahead */
941 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
942 struct md_enqueue_info *minfo,
945 struct md_enqueue_info {
946 struct md_op_data mi_data;
947 struct lookup_intent mi_it;
948 struct lustre_handle mi_lockh;
949 struct inode *mi_dir;
950 struct ldlm_enqueue_info mi_einfo;
951 md_enqueue_cb_t mi_cb;
956 struct module *o_owner;
957 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
958 void *karg, void __user *uarg);
959 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
960 __u32 keylen, void *key, __u32 *vallen, void *val);
961 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
962 __u32 keylen, void *key,
963 __u32 vallen, void *val,
964 struct ptlrpc_request_set *set);
965 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
966 int (*o_precleanup)(struct obd_device *dev);
967 int (*o_cleanup)(struct obd_device *dev);
968 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
969 int (*o_postrecov)(struct obd_device *dev);
970 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
972 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
973 /* connect to the target device with given connection
974 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
975 * granted by the target, which are guaranteed to be a subset of flags
976 * asked for. If @ocd == NULL, use default parameters. */
977 int (*o_connect)(const struct lu_env *env,
978 struct obd_export **exp, struct obd_device *src,
979 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
981 int (*o_reconnect)(const struct lu_env *env,
982 struct obd_export *exp, struct obd_device *src,
983 struct obd_uuid *cluuid,
984 struct obd_connect_data *ocd,
986 int (*o_disconnect)(struct obd_export *exp);
988 /* Initialize/finalize fids infrastructure. */
989 int (*o_fid_init)(struct obd_device *obd,
990 struct obd_export *exp, enum lu_cli_type type);
991 int (*o_fid_fini)(struct obd_device *obd);
993 /* Allocate new fid according to passed @hint. */
994 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
995 struct lu_fid *fid, struct md_op_data *op_data);
998 * Object with @fid is getting deleted, we may want to do something
1001 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1002 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1003 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1004 time64_t max_age, struct ptlrpc_request_set *set);
1005 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1007 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1009 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1011 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1013 int (*o_preprw)(const struct lu_env *env, int cmd,
1014 struct obd_export *exp, struct obdo *oa, int objcount,
1015 struct obd_ioobj *obj, struct niobuf_remote *remote,
1016 int *nr_pages, struct niobuf_local *local);
1017 int (*o_commitrw)(const struct lu_env *env, int cmd,
1018 struct obd_export *exp, struct obdo *oa,
1019 int objcount, struct obd_ioobj *obj,
1020 struct niobuf_remote *remote, int pages,
1021 struct niobuf_local *local, int rc);
1022 int (*o_init_export)(struct obd_export *exp);
1023 int (*o_destroy_export)(struct obd_export *exp);
1025 int (*o_import_event)(struct obd_device *, struct obd_import *,
1026 enum obd_import_event);
1028 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1029 enum obd_notify_event ev);
1031 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1032 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1035 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1036 struct obd_quotactl *);
1039 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1040 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1041 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1043 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1047 /* lmv structures */
1049 struct mdt_body *body;
1050 struct lu_buf layout;
1052 struct lmv_stripe_md *lmv;
1053 struct lmv_foreign_md *lfm;
1055 struct lmv_stripe_md *default_lmv;
1056 #ifdef CONFIG_FS_POSIX_ACL
1057 struct posix_acl *posix_acl;
1061 struct md_open_data {
1062 struct obd_client_handle *mod_och;
1063 struct ptlrpc_request *mod_open_req;
1064 struct ptlrpc_request *mod_close_req;
1065 atomic_t mod_refcount;
1069 struct obd_client_handle {
1070 struct lustre_handle och_open_handle;
1071 struct lu_fid och_fid;
1072 struct md_open_data *och_mod;
1073 struct lustre_handle och_lease_handle; /* open lock for lease */
1078 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1080 struct lookup_intent;
1084 int (*m_close)(struct obd_export *, struct md_op_data *,
1085 struct md_open_data *, struct ptlrpc_request **);
1087 int (*m_create)(struct obd_export *, struct md_op_data *,
1088 const void *, size_t, umode_t, uid_t, gid_t,
1089 cfs_cap_t, __u64, struct ptlrpc_request **);
1091 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1092 const union ldlm_policy_data *, struct md_op_data *,
1093 struct lustre_handle *, __u64);
1095 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1096 struct ptlrpc_request **);
1098 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1099 struct lookup_intent *,
1100 struct ptlrpc_request **,
1101 ldlm_blocking_callback, __u64);
1103 int (*m_link)(struct obd_export *, struct md_op_data *,
1104 struct ptlrpc_request **);
1106 int (*m_rename)(struct obd_export *, struct md_op_data *,
1107 const char *, size_t, const char *, size_t,
1108 struct ptlrpc_request **);
1110 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1111 size_t , struct ptlrpc_request **);
1113 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1114 struct ptlrpc_request **);
1116 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1117 struct md_callback *cb_op, __u64 hash_offset,
1118 struct page **ppage);
1120 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1121 struct ptlrpc_request **);
1123 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1124 u64, const char *, const void *, size_t, unsigned int,
1125 u32, struct ptlrpc_request **);
1127 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1128 u64, const char *, size_t, struct ptlrpc_request **);
1130 int (*m_intent_getattr_async)(struct obd_export *,
1131 struct md_enqueue_info *);
1133 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1134 struct lu_fid *, __u64 *bits);
1136 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1138 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1139 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1141 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1142 struct ptlrpc_request **);
1144 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1146 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1147 struct obd_export *, struct obd_export *,
1148 struct lustre_md *);
1150 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1152 int (*m_merge_attr)(struct obd_export *,
1153 const struct lmv_stripe_md *lsm,
1154 struct cl_attr *attr, ldlm_blocking_callback);
1156 int (*m_set_open_replay_data)(struct obd_export *,
1157 struct obd_client_handle *,
1158 struct lookup_intent *);
1160 int (*m_clear_open_replay_data)(struct obd_export *,
1161 struct obd_client_handle *);
1163 int (*m_set_lock_data)(struct obd_export *,
1164 const struct lustre_handle *, void *, __u64 *);
1166 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1167 const struct lu_fid *, enum ldlm_type,
1168 union ldlm_policy_data *, enum ldlm_mode,
1169 struct lustre_handle *);
1171 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1172 union ldlm_policy_data *, enum ldlm_mode,
1173 enum ldlm_cancel_flags flags, void *opaque);
1175 int (*m_get_fid_from_lsm)(struct obd_export *,
1176 const struct lmv_stripe_md *,
1177 const char *name, int namelen,
1178 struct lu_fid *fid);
1179 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1180 const union lmv_mds_md *lmv, size_t lmv_size);
1183 static inline struct md_open_data *obd_mod_alloc(void)
1185 struct md_open_data *mod;
1189 atomic_set(&mod->mod_refcount, 1);
1193 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1194 #define obd_mod_put(mod) \
1196 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1197 if ((mod)->mod_open_req) \
1198 ptlrpc_req_finished((mod)->mod_open_req); \
1199 OBD_FREE_PTR(mod); \
1203 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1204 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1205 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1207 /* return 1 if client should be resend request */
1208 static inline int client_should_resend(int resend, struct client_obd *cli)
1210 return atomic_read(&cli->cl_resends) ?
1211 atomic_read(&cli->cl_resends) > resend : 1;
1215 * Return device name for this device
1217 * XXX: lu_device is declared before obd_device, while a pointer pointing
1218 * back to obd_device in lu_device, so this helper function defines here
1219 * instead of in lu_object.h
1221 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1223 return lu_dev->ld_obd->obd_name;
1226 static inline bool filename_is_volatile(const char *name, size_t namelen,
1232 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1235 /* caller does not care of idx */
1239 /* volatile file, the MDT can be set from name */
1240 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1241 /* if no MDT is specified, use std way */
1242 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1244 /* test for no MDT idx case */
1245 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1246 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1250 /* we have an idx, read it */
1251 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1252 *idx = simple_strtoul(start, &end, 16);
1254 * no digit, no trailing :, negative value
1256 if (((*idx == 0) && (end == start)) ||
1257 (*end != ':') || (*idx < 0))
1262 /* bad format of mdt idx, we cannot return an error
1263 * to caller so we use hash algo */
1264 CERROR("Bad volatile file name format: %s\n",
1265 name + LUSTRE_VOLATILE_HDR_LEN);
1269 static inline int cli_brw_size(struct obd_device *obd)
1271 LASSERT(obd != NULL);
1272 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1276 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1277 * of the client should be increased accordingly to avoid sending fragmented
1278 * RPCs over the network when the client runs out of the maximum dirty space
1279 * when so many RPCs are being generated.
1281 static inline void client_adjust_max_dirty(struct client_obd *cli)
1284 if (cli->cl_dirty_max_pages <= 0) {
1285 cli->cl_dirty_max_pages =
1286 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1288 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1289 cli->cl_max_pages_per_rpc;
1291 if (dirty_max > cli->cl_dirty_max_pages)
1292 cli->cl_dirty_max_pages = dirty_max;
1295 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1296 cli->cl_dirty_max_pages = totalram_pages / 8;
1298 /* This value is exported to userspace through the max_dirty_mb
1299 * parameter. So we round up the number of pages to make it a round
1301 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1302 1 << (20 - PAGE_SHIFT));
1305 #endif /* __OBD_H */