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/
36 #include <linux/posix_acl.h>
37 #include <linux/kobject.h>
38 #include <linux/spinlock.h>
39 #include <linux/sysfs.h>
40 #include <linux/xarray.h>
42 #include <uapi/linux/lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
48 # include <lustre_quota.h>
51 #include <lustre_export.h>
52 #include <lustre_fid.h>
53 #include <lustre_fld.h>
54 #include <lustre_handles.h>
55 #include <lustre_intent.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 void lov_fix_ea_for_replay(void *lovea);
79 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
82 oinfo->loi_kms_valid = 1;
88 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
90 /* obd info for a particular level (lov, osc). */
92 /* OBD_STATFS_* flags */
94 struct obd_device *oi_obd;
95 struct lu_tgt_desc *oi_tgt;
96 /* statfs data specific for every OSC, if needed at all. */
97 struct obd_statfs *oi_osfs;
98 /* An update callback which is called to update some data on upper
99 * level. E.g. it is used for update lsm->lsm_oinfo at every received
100 * request in osc level for enqueue requests. It is also possible to
101 * update some caller data from LOV layer if needed. */
102 obd_enqueue_update_f oi_cb_up;
106 const struct obd_ops *typ_dt_ops;
107 const struct md_ops *typ_md_ops;
108 struct proc_dir_entry *typ_procroot;
109 struct dentry *typ_debugfs_entry;
110 #ifdef HAVE_SERVER_SUPPORT
114 struct lu_device_type *typ_lu;
115 struct kobject typ_kobj;
117 #define typ_name typ_kobj.name
118 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
125 /* used for encryption: difference with offset in clear text page */
127 /* used for encryption: difference with count in clear text page */
132 struct timeout_item {
133 enum timeout_event ti_event;
137 struct list_head ti_obd_list;
138 struct list_head ti_chain;
141 #define OBD_MAX_RIF_DEFAULT 8
142 #define OBD_MAX_RIF_MAX 512
143 #define OSC_MAX_RIF_MAX 256
144 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
145 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
146 #define OSC_DEFAULT_RESENDS 10
148 /* possible values for lut_sync_lock_cancel */
149 enum tgt_sync_lock_cancel {
150 SYNC_LOCK_CANCEL_NEVER = 0,
151 SYNC_LOCK_CANCEL_BLOCKING = 1,
152 SYNC_LOCK_CANCEL_ALWAYS = 2,
156 * Limit reply buffer size for striping data to one x86_64 page. This
157 * value is chosen to fit the striping data for common use cases while
158 * staying well below the limit at which the buffer must be backed by
159 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
162 #define OBD_MAX_DEFAULT_EA_SIZE 4096
165 * Lustre can handle larger xattrs internally, but we must respect the Linux
166 * VFS limitation or tools like tar cannot interact with Lustre volumes
169 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
172 enum obd_cl_sem_lock_class {
181 struct rw_semaphore cl_sem;
182 struct obd_uuid cl_target_uuid;
183 struct obd_import *cl_import; /* ptlrpc connection state */
184 size_t cl_conn_count;
186 /* Cache maximum and default values for easize. This is
187 * strictly a performance optimization to minimize calls to
188 * obd_size_diskmd(). The default values are used to calculate the
189 * initial size of a request buffer. The ptlrpc layer will resize the
190 * buffer as needed to accommodate a larger reply from the
191 * server. The default values should be small enough to avoid wasted
192 * memory and excessive use of vmalloc(), yet large enough to avoid
193 * reallocating the buffer in the common use case. */
195 /* Default EA size for striping attributes. It is initialized at
196 * mount-time based on the default stripe width of the filesystem,
197 * then it tracks the largest observed EA size advertised by
198 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
199 __u32 cl_default_mds_easize;
201 /* Maximum possible EA size computed at mount-time based on
202 * the number of OSTs in the filesystem. May be increased at
203 * run-time if a larger observed size is advertised by the MDT. */
204 __u32 cl_max_mds_easize;
206 /* Data-on-MDT specific value to set larger reply buffer for possible
207 * data read along with open/stat requests. By default it tries to use
208 * unused space in reply buffer.
209 * This value is used to ensure that reply buffer has at least as
210 * much free space as value indicates. That free space is gained from
211 * LOV EA buffer which is small for DoM files and on big systems can
212 * provide up to 32KB of extra space in reply buffer.
213 * Default value is 8K now.
215 __u32 cl_dom_min_inline_repsize;
217 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
218 cl_checksum_dump:1, /* same */
219 cl_ocd_grant_param:1,
220 cl_lsom_update:1; /* send LSOM updates */
221 enum lustre_sec_part cl_sp_me;
222 enum lustre_sec_part cl_sp_to;
223 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
225 /* the grant values are protected by loi_list_lock below */
226 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
227 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
228 unsigned long cl_avail_grant; /* bytes of credit for ost */
229 unsigned long cl_lost_grant; /* lost credits (trunc) */
230 /* grant consumed for dirty pages */
231 unsigned long cl_dirty_grant;
233 /* since we allocate grant by blocks, we don't know how many grant will
234 * be used to add a page into cache. As a solution, we reserve maximum
235 * grant before trying to dirty a page and unreserve the rest.
236 * See osc_{reserve|unreserve}_grant for details. */
237 long cl_reserved_grant;
238 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
239 time64_t cl_next_shrink_grant; /* seconds */
240 struct list_head cl_grant_chain;
241 time64_t cl_grant_shrink_interval; /* seconds */
243 int cl_root_squash; /* if root squash enabled*/
245 /* A chunk is an optimal size used by osc_extent to determine
246 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
248 /* extent insertion metadata overhead to be accounted in grant,
250 unsigned int cl_grant_extent_tax;
251 /* maximum extent size, in number of pages */
252 unsigned int cl_max_extent_pages;
254 /* keep track of objects that have lois that contain pages which
255 * have been queued for async brw. this lock also protects the
256 * lists of osc_client_pages that hang off of the loi */
258 * ->cl_loi_list_lock protects consistency of
259 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
260 * ->ap_completion() call-backs are executed under this lock. As we
261 * cannot guarantee that these call-backs never block on all platforms
262 * (as a matter of fact they do block on Mac OS X), type of
263 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
264 * and blocking mutex on Mac OS X. (Alternative is to make this lock
265 * blocking everywhere, but we don't want to slow down fast-path of
266 * our main platform.)
268 * NB by Jinshan: though field names are still _loi_, but actually
269 * osc_object{}s are in the list.
271 spinlock_t cl_loi_list_lock;
272 struct list_head cl_loi_ready_list;
273 struct list_head cl_loi_hp_ready_list;
274 struct list_head cl_loi_write_list;
275 struct list_head cl_loi_read_list;
276 __u32 cl_r_in_flight;
277 __u32 cl_w_in_flight;
278 /* just a sum of the loi/lop pending numbers to be exported by /proc */
279 atomic_t cl_pending_w_pages;
280 atomic_t cl_pending_r_pages;
281 u32 cl_max_pages_per_rpc;
282 u32 cl_max_rpcs_in_flight;
283 u32 cl_max_short_io_bytes;
284 ktime_t cl_stats_init;
285 struct obd_histogram cl_read_rpc_hist;
286 struct obd_histogram cl_write_rpc_hist;
287 struct obd_histogram cl_read_page_hist;
288 struct obd_histogram cl_write_page_hist;
289 struct obd_histogram cl_read_offset_hist;
290 struct obd_histogram cl_write_offset_hist;
292 /** LRU for osc caching pages */
293 struct cl_client_cache *cl_cache;
294 /** member of cl_cache->ccc_lru */
295 struct list_head cl_lru_osc;
296 /** # of available LRU slots left in the per-OSC cache.
297 * Available LRU slots are shared by all OSCs of the same file system,
298 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
299 atomic_long_t *cl_lru_left;
300 /** # of busy LRU pages. A page is considered busy if it's in writeback
301 * queue, or in transfer. Busy pages can't be discarded so they are not
303 atomic_long_t cl_lru_busy;
304 /** # of LRU pages in the cache for this client_obd */
305 atomic_long_t cl_lru_in_list;
306 /** # of threads are shrinking LRU cache. To avoid contention, it's not
307 * allowed to have multiple threads shrinking LRU cache. */
308 atomic_t cl_lru_shrinkers;
309 /** The time when this LRU cache was last used. */
310 time64_t cl_lru_last_used;
311 /** stats: how many reclaims have happened for this client_obd.
312 * reclaim and shrink - shrink is async, voluntarily rebalancing;
313 * reclaim is sync, initiated by IO thread when the LRU slots are
315 __u64 cl_lru_reclaim;
316 /** List of LRU pages for this client_obd */
317 struct list_head cl_lru_list;
318 /** Lock for LRU page list */
319 spinlock_t cl_lru_list_lock;
320 /** # of unstable pages in this client_obd.
321 * An unstable page is a page state that WRITE RPC has finished but
322 * the transaction has NOT yet committed. */
323 atomic_long_t cl_unstable_count;
324 /** Link to osc_shrinker_list */
325 struct list_head cl_shrink_list;
327 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
328 atomic_t cl_destroy_in_flight;
329 wait_queue_head_t cl_destroy_waitq;
331 /* modify rpcs in flight
332 * currently used for metadata only */
333 spinlock_t cl_mod_rpcs_lock;
334 __u16 cl_max_mod_rpcs_in_flight;
335 __u16 cl_mod_rpcs_in_flight;
336 __u16 cl_close_rpcs_in_flight;
337 wait_queue_head_t cl_mod_rpcs_waitq;
338 unsigned long *cl_mod_tag_bitmap;
339 ktime_t cl_mod_rpcs_init;
340 struct obd_histogram cl_mod_rpcs_hist;
343 struct mutex cl_mgc_mutex;
344 struct local_oid_storage *cl_mgc_los;
345 struct dt_object *cl_mgc_configs_dir;
346 struct obd_export *cl_mgc_mgsexp;
347 atomic_t cl_mgc_refcount;
348 /* in-flight control list and total RPCs counter */
349 struct list_head cl_flight_waiters;
350 __u32 cl_rpcs_in_flight;
352 /* supported checksum types that are worked out at connect time */
353 __u32 cl_supp_cksum_types;
354 /* checksum algorithm to be used */
355 enum cksum_types cl_cksum_type;
356 /* preferred checksum algorithm to be used */
357 enum cksum_types cl_preferred_cksum_type;
359 /* also protected by the poorly named _loi_list_lock lock above */
360 struct osc_async_rc cl_ar;
362 /* sequence manager */
363 struct lu_client_seq *cl_seq;
364 struct rw_semaphore cl_seq_rwsem;
366 atomic_t cl_resends; /* resend count */
368 /* ptlrpc work for writeback in ptlrpcd context */
369 void *cl_writeback_work;
371 struct mutex cl_quota_mutex;
372 /* hash tables for osc_quota_info */
373 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
374 /* the xid of the request updating the hash tables */
375 __u64 cl_quota_last_xid;
376 /* Links to the global list of registered changelog devices */
377 struct list_head cl_chg_dev_linkage;
379 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
386 struct echo_client_obd {
387 struct obd_export *ec_exp; /* the local connection to osc/lov */
389 struct list_head ec_objects;
390 struct list_head ec_locks;
394 /* allow statfs data caching for 1 second */
395 #define OBD_STATFS_CACHE_SECONDS 1
396 /* arbitrary maximum. larger would be useless, allows catching bogus input */
397 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
399 #define lov_tgt_desc lu_tgt_desc
401 struct lov_md_tgt_desc {
402 struct obd_device *lmtd_mdc;
407 struct lov_desc desc;
408 struct lov_tgt_desc **lov_tgts; /* sparse array */
409 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
411 struct mutex lov_lock;
412 struct obd_connect_data lov_ocd;
413 atomic_t lov_refcount;
414 __u32 lov_death_row; /* tgts scheduled to be deleted */
415 __u32 lov_tgt_size; /* size of tgts array */
418 struct rhashtable lov_pools_hash_body; /* used for key access */
419 struct list_head lov_pool_list; /* used for sequential access */
420 struct proc_dir_entry *lov_pool_proc_entry;
421 enum lustre_sec_part lov_sp_me;
423 /* Cached LRU and unstable data from upper layer */
424 struct cl_client_cache *lov_cache;
426 struct rw_semaphore lov_notify_lock;
427 /* Data-on-MDT: MDC array */
428 struct lov_md_tgt_desc *lov_mdc_tgts;
430 struct kobject *lov_tgts_kobj;
433 #define lmv_tgt_desc lu_tgt_desc
436 struct lu_client_fld lmv_fld;
442 u32 lmv_statfs_start;
444 struct lu_tgt_descs lmv_mdt_descs;
446 struct obd_connect_data conn_data;
447 struct kobject *lmv_tgts_kobj;
450 __u32 lmv_qos_rr_index;
453 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
454 #define lmv_qos lmv_mdt_descs.ltd_qos
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;
470 /* separate unlock for read path to allow shared access */
474 struct tgt_thread_big_cache {
475 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
478 #define LUSTRE_FLD_NAME "fld"
479 #define LUSTRE_SEQ_NAME "seq"
481 #define LUSTRE_MDD_NAME "mdd"
482 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
483 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
484 #define LUSTRE_VVP_NAME "vvp"
485 #define LUSTRE_LMV_NAME "lmv"
486 #define LUSTRE_SLP_NAME "slp"
487 #define LUSTRE_LOD_NAME "lod"
488 #define LUSTRE_OSP_NAME "osp"
489 #define LUSTRE_LWP_NAME "lwp"
491 /* obd device type names */
492 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
493 #define LUSTRE_MDS_NAME "mds"
494 #define LUSTRE_MDT_NAME "mdt"
495 #define LUSTRE_MDC_NAME "mdc"
496 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
497 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
498 #define LUSTRE_OSC_NAME "osc"
499 #define LUSTRE_LOV_NAME "lov"
500 #define LUSTRE_MGS_NAME "mgs"
501 #define LUSTRE_MGC_NAME "mgc"
503 #define LUSTRE_ECHO_NAME "obdecho"
504 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
505 #define LUSTRE_QMT_NAME "qmt"
507 /* Constant obd names (post-rename) */
508 #define LUSTRE_MDS_OBDNAME "MDS"
509 #define LUSTRE_OSS_OBDNAME "OSS"
510 #define LUSTRE_MGS_OBDNAME "MGS"
511 #define LUSTRE_MGC_OBDNAME "MGC"
513 static inline int is_lwp_on_mdt(char *name)
517 ptr = strrchr(name, '-');
519 CERROR("%s is not a obdname\n", name);
523 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
525 if (strncmp(ptr + 1, "MDT", 3) != 0)
528 while (*(--ptr) != '-' && ptr != name);
533 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
539 static inline int is_lwp_on_ost(char *name)
543 ptr = strrchr(name, '-');
545 CERROR("%s is not a obdname\n", name);
549 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
551 if (strncmp(ptr + 1, "OST", 3) != 0)
554 while (*(--ptr) != '-' && ptr != name);
559 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
566 * Events signalled through obd_notify() upcall-chain.
568 enum obd_notify_event {
569 /* Device connect start */
571 /* Device activated */
573 /* Device deactivated */
575 /* Connect data for import were changed */
577 /* Administratively deactivate/activate event */
578 OBD_NOTIFY_DEACTIVATE,
583 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
584 * being main example).
586 struct obd_notify_upcall {
587 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
588 enum obd_notify_event ev, void *owner);
589 /* Opaque datum supplied by upper layer listener */
593 struct target_recovery_data {
594 svc_handler_t trd_recovery_handler;
595 pid_t trd_processing_task;
596 struct completion trd_starting;
597 struct completion trd_finishing;
600 struct obd_llog_group {
601 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
602 wait_queue_head_t olg_waitq;
606 /* corresponds to one of the obd's */
607 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
610 struct obd_type *obd_type;
611 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
612 int obd_minor; /* device number: lctl dl */
613 struct lu_device *obd_lu_dev;
615 /* common and UUID name of this device */
616 struct obd_uuid obd_uuid;
617 char obd_name[MAX_OBD_NAME];
619 /* bitfield modification is protected by obd_dev_lock */
621 obd_attached:1, /* finished attach */
622 obd_set_up:1, /* finished setup */
623 obd_recovering:1, /* there are recoverable clients */
624 obd_abort_recovery:1, /* recovery expired */
625 obd_abort_recov_mdt:1, /* only abort recovery between MDTs */
626 obd_version_recov:1, /* obd uses version checking */
627 obd_replayable:1, /* recovery enabled; inform clients */
628 obd_no_recov:1, /* fail instead of retry messages */
629 obd_stopping:1, /* started cleanup */
630 obd_starting:1, /* started setup */
631 obd_force:1, /* cleanup with > 0 obd refcount */
632 obd_fail:1, /* cleanup with failover */
633 obd_no_conn:1, /* deny new connections */
634 obd_inactive:1, /* device active/inactive
635 * (for /proc/status only!!) */
636 obd_no_ir:1, /* no imperative recovery. */
637 obd_process_conf:1, /* device is processing mgs config */
638 obd_checksum_dump:1, /* dump pages upon cksum error */
639 obd_dynamic_nids:1; /* Allow dynamic NIDs on device */
640 #ifdef HAVE_SERVER_SUPPORT
641 /* no committed-transno notification */
642 unsigned long obd_no_transno:1;
645 /* use separate field as it is set in interrupt to don't mess with
646 * protection of other bits using _bh lock */
647 unsigned long obd_recovery_expired:1;
648 /* uuid-export hash body */
649 struct rhashtable obd_uuid_hash;
650 /* nid-export hash body */
651 struct rhltable obd_nid_hash;
653 struct cfs_hash *obd_nid_stats_hash;
654 /* client_generation-export hash body */
655 struct cfs_hash *obd_gen_hash;
656 struct list_head obd_nid_stats;
657 struct list_head obd_exports;
658 struct list_head obd_unlinked_exports;
659 struct list_head obd_delayed_exports;
660 struct list_head obd_lwp_list;
661 atomic_t obd_refcount;
663 int obd_grant_check_threshold;
664 spinlock_t obd_nid_lock;
665 struct ldlm_namespace *obd_namespace;
666 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
667 /* a spinlock is OK for what we do now, may need a semaphore later */
668 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
669 spinlock_t obd_osfs_lock;
670 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
671 time64_t obd_osfs_age;
672 __u64 obd_last_committed;
673 struct mutex obd_dev_mutex;
674 struct lvfs_run_ctxt obd_lvfs_ctxt;
675 struct obd_llog_group obd_olg; /* default llog group */
676 struct obd_device *obd_observer;
677 struct rw_semaphore obd_observer_link_sem;
678 struct obd_notify_upcall obd_upcall;
679 struct obd_export *obd_self_export;
680 struct obd_export *obd_lwp_export;
681 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
682 struct list_head obd_exports_timed;
683 time64_t obd_eviction_timer; /* for ping evictor */
685 atomic_t obd_max_recoverable_clients;
686 atomic_t obd_connected_clients;
687 int obd_stale_clients;
688 /* this lock protects all recovery list_heads, timer and
689 * obd_next_recovery_transno value */
690 spinlock_t obd_recovery_task_lock;
691 __u64 obd_next_recovery_transno;
692 int obd_replayed_requests;
693 int obd_requests_queued_for_recovery;
694 wait_queue_head_t obd_next_transno_waitq;
695 /* protected by obd_recovery_task_lock */
696 struct hrtimer obd_recovery_timer;
698 time64_t obd_recovery_start;
699 /* seconds, for lprocfs_status */
700 time64_t obd_recovery_end;
701 /* To tell timeouts from time stamps Lustre uses timeout_t
702 * instead of time64_t.
704 timeout_t obd_recovery_time_hard;
705 timeout_t obd_recovery_timeout;
706 int obd_recovery_ir_factor;
708 /* new recovery stuff from CMD2 */
709 int obd_replayed_locks;
710 atomic_t obd_req_replay_clients;
711 atomic_t obd_lock_replay_clients;
712 struct target_recovery_data obd_recovery_data;
714 /* all lists are protected by obd_recovery_task_lock */
715 struct list_head obd_req_replay_queue;
716 struct list_head obd_lock_replay_queue;
717 struct list_head obd_final_req_queue;
720 #ifdef HAVE_SERVER_SUPPORT
721 struct obd_device_target obt;
722 struct filter_obd filter;
724 struct echo_obd echo;
726 struct client_obd cli;
727 struct echo_client_obd echo_client;
732 /* Fields used by LProcFS */
733 struct lprocfs_stats *obd_stats;
735 struct lprocfs_stats *obd_md_stats;
737 struct dentry *obd_debugfs_entry;
738 struct proc_dir_entry *obd_proc_entry;
739 struct proc_dir_entry *obd_proc_exports_entry;
740 struct dentry *obd_svc_debugfs_entry;
741 struct lprocfs_stats *obd_svc_stats;
742 const struct attribute **obd_attrs;
743 struct lprocfs_vars *obd_vars;
744 struct ldebugfs_vars *obd_debugfs_vars;
745 atomic_t obd_evict_inprogress;
746 wait_queue_head_t obd_evict_inprogress_waitq;
747 struct list_head obd_evict_list; /* protected with pet_lock */
750 * LDLM pool part. Save last calculated SLV and Limit.
752 rwlock_t obd_pool_lock;
756 int obd_conn_inprogress;
759 * List of outstanding class_incref()'s fo this OBD. For debugging. */
760 struct lu_ref obd_reference;
762 struct kset obd_kset; /* sysfs object collection */
763 struct kobj_type obd_ktype;
764 struct completion obd_kobj_unregister;
767 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
768 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
769 #ifdef HAVE_SERVER_SUPPORT
770 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
771 struct obd_uuid *uuid);
773 int obd_nid_export_for_each(struct obd_device *obd, struct lnet_nid *nid,
774 int cb(struct obd_export *exp, void *data),
776 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
777 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
780 /* get/set_info keys */
781 #define KEY_ASYNC "async"
782 #define KEY_CHANGELOG_CLEAR "changelog_clear"
783 #define KEY_FID2PATH "fid2path"
784 #define KEY_CHECKSUM "checksum"
785 #define KEY_CLEAR_FS "clear_fs"
786 #define KEY_CONN_DATA "conn_data"
787 #define KEY_EVICT_BY_NID "evict_by_nid"
788 #define KEY_FIEMAP "fiemap"
789 #define KEY_FLUSH_CTX "flush_ctx"
790 #define KEY_GRANT_SHRINK "grant_shrink"
791 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
792 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
793 #define KEY_INTERMDS "inter_mds"
794 #define KEY_LAST_ID "last_id"
795 #define KEY_LAST_FID "last_fid"
796 #define KEY_MAX_EASIZE "max_easize"
797 #define KEY_DEFAULT_EASIZE "default_easize"
798 #define KEY_MGSSEC "mgssec"
799 #define KEY_READ_ONLY "read-only"
800 #define KEY_REGISTER_TARGET "register_target"
801 #define KEY_SET_FS "set_fs"
802 #define KEY_TGT_COUNT "tgt_count"
803 /* KEY_SET_INFO in lustre_idl.h */
804 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
806 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
807 #define KEY_OSP_CONNECTED "osp_connected"
809 /* Flags for op_xvalid */
811 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
812 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
813 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
814 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
815 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
816 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
817 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
822 static inline int it_to_lock_mode(struct lookup_intent *it)
824 /* CREAT needs to be tested before open (both could be set) */
825 if (it->it_op & IT_CREAT)
827 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
829 else if (it->it_op & IT_LAYOUT)
830 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
831 else if (it->it_op & IT_READDIR)
833 else if (it->it_op & IT_GETXATTR)
836 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
841 MF_MDC_CANCEL_FID1 = BIT(0),
842 MF_MDC_CANCEL_FID2 = BIT(1),
843 MF_MDC_CANCEL_FID3 = BIT(2),
844 MF_MDC_CANCEL_FID4 = BIT(3),
845 MF_GET_MDT_IDX = BIT(4),
846 MF_GETATTR_BY_FID = BIT(5),
847 MF_QOS_MKDIR = BIT(6),
848 MF_RR_MKDIR = BIT(7),
849 MF_OPNAME_KMALLOCED = BIT(8),
853 CLI_SET_MEA = BIT(0),
854 CLI_RM_ENTRY = BIT(1),
857 CLI_MIGRATE = BIT(4),
858 CLI_DIRTY_DATA = BIT(5),
859 CLI_NO_SLOT = BIT(6),
863 LUSTRE_OPC_MKDIR = 1,
873 * GETXATTR is not included as only a couple of fields in the reply body
874 * is filled, but not FID which is needed for common intent handling in
875 * mdc_finish_intent_lock()
877 static inline bool it_has_reply_body(const struct lookup_intent *it)
879 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
883 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
884 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
885 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
886 struct lu_fid op_fid4; /* to the operation locks. */
887 u32 op_mds; /* what mds server open will go to */
889 enum md_op_code op_code;
890 struct lustre_handle op_open_handle;
894 struct rw_semaphore *op_mea1_sem;
895 struct rw_semaphore *op_mea2_sem;
896 struct lmv_stripe_md *op_mea1;
897 struct lmv_stripe_md *op_mea2;
898 struct lmv_stripe_md *op_default_mea1; /* default LMV */
899 __u32 op_suppgids[2];
906 /* iattr fields and blocks. */
907 struct iattr op_attr;
908 enum op_xvalid op_xvalid; /* eXtra validity flags */
909 loff_t op_attr_blocks;
910 u64 op_valid; /* OBD_MD_* */
911 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
913 enum md_op_flags op_flags;
915 /* Various operation flags. */
916 enum mds_op_bias op_bias;
918 /* used to transfer info between the stacks of MD client
919 * see enum op_cli_flags */
920 enum md_cli_flags op_cli_flags;
922 /* File object data version for HSM release, on client */
923 __u64 op_data_version;
924 struct lustre_handle op_lease_handle;
926 /* File security context, for creates/metadata ops */
927 const char *op_file_secctx_name;
928 __u32 op_file_secctx_name_size;
929 void *op_file_secctx;
930 __u32 op_file_secctx_size;
932 /* File encryption context, for creates/metadata ops */
933 void *op_file_encctx;
934 __u32 op_file_encctx_size;
939 /* Used by readdir */
940 unsigned int op_max_pages;
942 unsigned short op_dir_depth;
948 * used to access dir that is changing layout: if it's set, access
949 * dir by new layout, otherwise old layout.
950 * By default it's not set, because new files are created under new
951 * layout, if we can't find file with name under both old and new
952 * layout, we are sure file with name doesn't exist, but in reverse
953 * order there may be a race with creation by others.
956 /* used to access dir with bash hash */
957 __u32 op_stripe_index;
958 /* Archive ID for PCC attach */
962 struct md_readdir_info {
963 int (*mr_blocking_ast)(struct ldlm_lock *lock,
964 struct ldlm_lock_desc *desc,
965 void *data, int flag);
966 /* if striped directory is partially read, the result is stored here */
967 int mr_partial_readdir_rc;
970 struct md_enqueue_info;
971 /* metadata stat-ahead */
972 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
973 struct md_enqueue_info *minfo,
976 struct md_enqueue_info {
977 struct md_op_data mi_data;
978 struct lookup_intent mi_it;
979 struct lustre_handle mi_lockh;
980 struct inode *mi_dir;
981 struct ldlm_enqueue_info mi_einfo;
982 md_enqueue_cb_t mi_cb;
987 struct module *o_owner;
988 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
989 void *karg, void __user *uarg);
990 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
991 __u32 keylen, void *key, __u32 *vallen, void *val);
992 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
993 __u32 keylen, void *key,
994 __u32 vallen, void *val,
995 struct ptlrpc_request_set *set);
996 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
997 int (*o_precleanup)(struct obd_device *obd);
998 int (*o_cleanup)(struct obd_device *obd);
999 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
1000 int (*o_postrecov)(struct obd_device *obd);
1001 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1003 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1004 /* connect to the target device with given connection
1005 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1006 * granted by the target, which are guaranteed to be a subset of flags
1007 * asked for. If @ocd == NULL, use default parameters. */
1008 int (*o_connect)(const struct lu_env *env,
1009 struct obd_export **exp, struct obd_device *src,
1010 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1012 int (*o_reconnect)(const struct lu_env *env,
1013 struct obd_export *exp, struct obd_device *src,
1014 struct obd_uuid *cluuid,
1015 struct obd_connect_data *ocd,
1017 int (*o_disconnect)(struct obd_export *exp);
1019 /* Initialize/finalize fids infrastructure. */
1020 int (*o_fid_init)(struct obd_device *obd,
1021 struct obd_export *exp, enum lu_cli_type type);
1022 int (*o_fid_fini)(struct obd_device *obd);
1024 /* Allocate new fid according to passed @hint. */
1025 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1026 struct lu_fid *fid, struct md_op_data *op_data);
1029 * Object with @fid is getting deleted, we may want to do something
1032 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1033 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1034 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1035 time64_t max_age, struct ptlrpc_request_set *set);
1036 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1038 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1040 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1042 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1044 int (*o_preprw)(const struct lu_env *env, int cmd,
1045 struct obd_export *exp, struct obdo *oa, int objcount,
1046 struct obd_ioobj *obj, struct niobuf_remote *remote,
1047 int *nr_pages, struct niobuf_local *local);
1048 int (*o_commitrw)(const struct lu_env *env, int cmd,
1049 struct obd_export *exp, struct obdo *oa,
1050 int objcount, struct obd_ioobj *obj,
1051 struct niobuf_remote *remote, int pages,
1052 struct niobuf_local *local, int rc, int nob,
1054 int (*o_init_export)(struct obd_export *exp);
1055 int (*o_destroy_export)(struct obd_export *exp);
1057 int (*o_import_event)(struct obd_device *, struct obd_import *,
1058 enum obd_import_event);
1060 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1061 enum obd_notify_event ev);
1063 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1064 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1067 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1068 struct obd_quotactl *);
1071 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1072 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1073 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1075 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1079 /* lmv structures */
1081 struct mdt_body *body;
1082 struct lu_buf layout;
1084 struct lmv_stripe_md *lmv;
1085 struct lmv_foreign_md *lfm;
1087 struct lmv_stripe_md *default_lmv;
1088 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1089 struct posix_acl *posix_acl;
1093 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1094 static inline void lmd_clear_acl(struct lustre_md *md)
1096 if (md->posix_acl) {
1097 posix_acl_release(md->posix_acl);
1098 md->posix_acl = NULL;
1102 #define OBD_CONNECT_ACL_FLAGS \
1103 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1105 static inline void lmd_clear_acl(struct lustre_md *md)
1109 #define OBD_CONNECT_ACL_FLAGS (0)
1112 struct md_open_data {
1113 struct obd_client_handle *mod_och;
1114 struct ptlrpc_request *mod_open_req;
1115 struct ptlrpc_request *mod_close_req;
1116 atomic_t mod_refcount;
1120 struct obd_client_handle {
1121 struct lustre_handle och_open_handle;
1122 struct lu_fid och_fid;
1123 struct md_open_data *och_mod;
1124 struct lustre_handle och_lease_handle; /* open lock for lease */
1129 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1131 struct lookup_intent;
1135 int (*m_close)(struct obd_export *, struct md_op_data *,
1136 struct md_open_data *, struct ptlrpc_request **);
1138 int (*m_create)(struct obd_export *, struct md_op_data *,
1139 const void *, size_t, umode_t, uid_t, gid_t,
1140 kernel_cap_t, __u64, struct ptlrpc_request **);
1142 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1143 const union ldlm_policy_data *, struct md_op_data *,
1144 struct lustre_handle *, __u64);
1146 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1147 struct ptlrpc_request **);
1149 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1150 struct lookup_intent *,
1151 struct ptlrpc_request **,
1152 ldlm_blocking_callback, __u64);
1154 int (*m_link)(struct obd_export *, struct md_op_data *,
1155 struct ptlrpc_request **);
1157 int (*m_rename)(struct obd_export *, struct md_op_data *,
1158 const char *, size_t, const char *, size_t,
1159 struct ptlrpc_request **);
1161 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1162 size_t , struct ptlrpc_request **);
1164 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1165 struct ptlrpc_request **);
1167 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1168 struct md_readdir_info *mrinfo, __u64 hash_offset,
1169 struct page **ppage);
1171 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1172 struct ptlrpc_request **);
1174 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1175 u64, const char *, const void *, size_t, unsigned int,
1176 u32, struct ptlrpc_request **);
1178 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1179 u64, const char *, size_t, struct ptlrpc_request **);
1181 int (*m_intent_getattr_async)(struct obd_export *,
1182 struct md_enqueue_info *);
1184 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1185 struct lu_fid *, __u64 *bits);
1187 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1189 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1190 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1192 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1193 struct ptlrpc_request **);
1195 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1197 int (*m_get_lustre_md)(struct obd_export *, struct req_capsule *,
1198 struct obd_export *, struct obd_export *,
1199 struct lustre_md *);
1201 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1203 int (*m_merge_attr)(struct obd_export *,
1204 const struct lmv_stripe_md *lsm,
1205 struct cl_attr *attr, ldlm_blocking_callback);
1207 int (*m_set_open_replay_data)(struct obd_export *,
1208 struct obd_client_handle *,
1209 struct lookup_intent *);
1211 int (*m_clear_open_replay_data)(struct obd_export *,
1212 struct obd_client_handle *);
1214 int (*m_set_lock_data)(struct obd_export *,
1215 const struct lustre_handle *, void *, __u64 *);
1217 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1218 const struct lu_fid *, enum ldlm_type,
1219 union ldlm_policy_data *, enum ldlm_mode,
1220 struct lustre_handle *);
1222 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1223 union ldlm_policy_data *, enum ldlm_mode,
1224 enum ldlm_cancel_flags flags, void *opaque);
1226 int (*m_get_fid_from_lsm)(struct obd_export *,
1227 const struct lmv_stripe_md *,
1228 const char *name, int namelen,
1229 struct lu_fid *fid);
1230 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1231 const union lmv_mds_md *lmv, size_t lmv_size);
1232 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1233 struct ptlrpc_request_set *set);
1236 static inline struct md_open_data *obd_mod_alloc(void)
1238 struct md_open_data *mod;
1242 atomic_set(&mod->mod_refcount, 1);
1246 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1247 #define obd_mod_put(mod) \
1249 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1250 if ((mod)->mod_open_req) \
1251 ptlrpc_req_finished((mod)->mod_open_req); \
1252 OBD_FREE_PTR(mod); \
1256 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1257 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1258 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1260 /* return 1 if client should be resend request */
1261 static inline int client_should_resend(int resend, struct client_obd *cli)
1263 return atomic_read(&cli->cl_resends) ?
1264 atomic_read(&cli->cl_resends) > resend : 1;
1268 * Return device name for this device
1270 * XXX: lu_device is declared before obd_device, while a pointer pointing
1271 * back to obd_device in lu_device, so this helper function defines here
1272 * instead of in lu_object.h
1274 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1276 return lu_dev->ld_obd->obd_name;
1279 static inline bool filename_is_volatile(const char *name, size_t namelen,
1285 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1288 /* caller does not care of idx */
1292 /* volatile file, the MDT can be set from name */
1293 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1294 /* if no MDT is specified, use std way */
1295 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1297 /* test for no MDT idx case */
1298 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1299 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1303 /* we have an idx, read it */
1304 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1305 *idx = simple_strtoul(start, &end, 16);
1307 * no digit, no trailing :, negative value
1309 if (((*idx == 0) && (end == start)) ||
1310 (*end != ':') || (*idx < 0))
1315 /* bad format of mdt idx, we cannot return an error
1316 * to caller so we use hash algo */
1317 CERROR("Bad volatile file name format: %s\n",
1318 name + LUSTRE_VOLATILE_HDR_LEN);
1322 static inline int cli_brw_size(struct obd_device *obd)
1324 LASSERT(obd != NULL);
1325 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1329 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1330 * of the client should be increased accordingly to avoid sending fragmented
1331 * RPCs over the network when the client runs out of the maximum dirty space
1332 * when so many RPCs are being generated.
1334 static inline void client_adjust_max_dirty(struct client_obd *cli)
1337 if (cli->cl_dirty_max_pages <= 0) {
1338 cli->cl_dirty_max_pages =
1339 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1341 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1342 cli->cl_max_pages_per_rpc;
1344 if (dirty_max > cli->cl_dirty_max_pages)
1345 cli->cl_dirty_max_pages = dirty_max;
1348 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1349 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1351 /* This value is exported to userspace through the max_dirty_mb
1352 * parameter. So we round up the number of pages to make it a round
1354 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1355 1 << (20 - PAGE_SHIFT));
1358 /* Must be used for page cache pages only,
1359 * not safe otherwise (e.g. direct IO pages)
1361 static inline struct inode *page2inode(struct page *page)
1363 if (page->mapping) {
1367 return page->mapping->host;
1373 #endif /* __OBD_H */