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.
37 #include <linux/posix_acl.h>
38 #include <linux/kobject.h>
39 #include <linux/spinlock.h>
40 #include <linux/sysfs.h>
41 #include <linux/xarray.h>
43 #include <uapi/linux/lustre/lustre_idl.h>
44 #include <lustre_lib.h>
45 #include <libcfs/bitmap.h>
46 #ifdef HAVE_SERVER_SUPPORT
47 # include <lu_target.h>
48 # include <obd_target.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>
57 #include <lustre_quota.h>
59 #define MAX_OBD_DEVICES 8192
67 struct lov_oinfo { /* per-stripe data structure */
68 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
69 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
70 int loi_ost_gen; /* generation of this loi_ost_idx */
72 unsigned long loi_kms_valid:1;
73 __u64 loi_kms; /* known minimum size */
74 struct ost_lvb loi_lvb;
75 struct osc_async_rc loi_ar;
78 void lov_fix_ea_for_replay(void *lovea);
80 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
83 oinfo->loi_kms_valid = 1;
89 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
91 /* obd info for a particular level (lov, osc). */
93 /* OBD_STATFS_* flags */
95 struct obd_device *oi_obd;
96 struct lu_tgt_desc *oi_tgt;
97 /* statfs data specific for every OSC, if needed at all. */
98 struct obd_statfs *oi_osfs;
99 /* An update callback which is called to update some data on upper
100 * level. E.g. it is used for update lsm->lsm_oinfo at every received
101 * request in osc level for enqueue requests. It is also possible to
102 * update some caller data from LOV layer if needed. */
103 obd_enqueue_update_f oi_cb_up;
107 const struct obd_ops *typ_dt_ops;
108 const struct md_ops *typ_md_ops;
109 struct proc_dir_entry *typ_procroot;
110 struct dentry *typ_debugfs_entry;
111 #ifdef HAVE_SERVER_SUPPORT
115 struct lu_device_type *typ_lu;
116 struct kobject typ_kobj;
118 #define typ_name typ_kobj.name
119 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
126 /* used for encryption: difference with offset in clear text page */
128 /* used for encryption: difference with count in clear text page */
133 struct timeout_item {
134 enum timeout_event ti_event;
138 struct list_head ti_obd_list;
139 struct list_head ti_chain;
142 #define OBD_MAX_RIF_DEFAULT 8
143 #define OBD_MAX_RIF_MAX 512
144 #define OSC_MAX_RIF_MAX 256
145 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
146 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
147 #define OSC_DEFAULT_RESENDS 10
149 /* possible values for lut_sync_lock_cancel */
150 enum tgt_sync_lock_cancel {
151 SYNC_LOCK_CANCEL_NEVER = 0,
152 SYNC_LOCK_CANCEL_BLOCKING = 1,
153 SYNC_LOCK_CANCEL_ALWAYS = 2,
157 * Limit reply buffer size for striping data to one x86_64 page. This
158 * value is chosen to fit the striping data for common use cases while
159 * staying well below the limit at which the buffer must be backed by
160 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
163 #define OBD_MAX_DEFAULT_EA_SIZE 4096
166 * Lustre can handle larger xattrs internally, but we must respect the Linux
167 * VFS limitation or tools like tar cannot interact with Lustre volumes
170 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
173 enum obd_cl_sem_lock_class {
182 struct rw_semaphore cl_sem;
183 struct obd_uuid cl_target_uuid;
184 struct obd_import *cl_import; /* ptlrpc connection state */
185 size_t cl_conn_count;
187 /* Cache maximum and default values for easize. This is
188 * strictly a performance optimization to minimize calls to
189 * obd_size_diskmd(). The default values are used to calculate the
190 * initial size of a request buffer. The ptlrpc layer will resize the
191 * buffer as needed to accommodate a larger reply from the
192 * server. The default values should be small enough to avoid wasted
193 * memory and excessive use of vmalloc(), yet large enough to avoid
194 * reallocating the buffer in the common use case. */
196 /* Default EA size for striping attributes. It is initialized at
197 * mount-time based on the default stripe width of the filesystem,
198 * then it tracks the largest observed EA size advertised by
199 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
200 __u32 cl_default_mds_easize;
202 /* Maximum possible EA size computed at mount-time based on
203 * the number of OSTs in the filesystem. May be increased at
204 * run-time if a larger observed size is advertised by the MDT. */
205 __u32 cl_max_mds_easize;
207 /* Data-on-MDT specific value to set larger reply buffer for possible
208 * data read along with open/stat requests. By default it tries to use
209 * unused space in reply buffer.
210 * This value is used to ensure that reply buffer has at least as
211 * much free space as value indicates. That free space is gained from
212 * LOV EA buffer which is small for DoM files and on big systems can
213 * provide up to 32KB of extra space in reply buffer.
214 * Default value is 8K now.
216 __u32 cl_dom_min_inline_repsize;
218 enum lustre_sec_part cl_sp_me;
219 enum lustre_sec_part cl_sp_to;
220 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
222 /* the grant values are protected by loi_list_lock below */
223 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
224 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
225 unsigned long cl_avail_grant; /* bytes of credit for ost */
226 unsigned long cl_lost_grant; /* lost credits (trunc) */
227 /* grant consumed for dirty pages */
228 unsigned long cl_dirty_grant;
230 /* since we allocate grant by blocks, we don't know how many grant will
231 * be used to add a page into cache. As a solution, we reserve maximum
232 * grant before trying to dirty a page and unreserve the rest.
233 * See osc_{reserve|unreserve}_grant for details. */
234 long cl_reserved_grant;
235 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
236 time64_t cl_next_shrink_grant; /* seconds */
237 struct list_head cl_grant_chain;
238 time64_t cl_grant_shrink_interval; /* seconds */
240 /* A chunk is an optimal size used by osc_extent to determine
241 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
243 /* extent insertion metadata overhead to be accounted in grant,
245 unsigned int cl_grant_extent_tax;
246 /* maximum extent size, in number of pages */
247 unsigned int cl_max_extent_pages;
249 /* keep track of objects that have lois that contain pages which
250 * have been queued for async brw. this lock also protects the
251 * lists of osc_client_pages that hang off of the loi */
253 * ->cl_loi_list_lock protects consistency of
254 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
255 * ->ap_completion() call-backs are executed under this lock. As we
256 * cannot guarantee that these call-backs never block on all platforms
257 * (as a matter of fact they do block on Mac OS X), type of
258 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
259 * and blocking mutex on Mac OS X. (Alternative is to make this lock
260 * blocking everywhere, but we don't want to slow down fast-path of
261 * our main platform.)
263 * NB by Jinshan: though field names are still _loi_, but actually
264 * osc_object{}s are in the list.
266 spinlock_t cl_loi_list_lock;
267 struct list_head cl_loi_ready_list;
268 struct list_head cl_loi_hp_ready_list;
269 struct list_head cl_loi_write_list;
270 struct list_head cl_loi_read_list;
271 __u32 cl_r_in_flight;
272 __u32 cl_w_in_flight;
273 /* just a sum of the loi/lop pending numbers to be exported by /proc */
274 atomic_t cl_pending_w_pages;
275 atomic_t cl_pending_r_pages;
276 u32 cl_max_pages_per_rpc;
277 u32 cl_max_rpcs_in_flight;
278 u32 cl_max_short_io_bytes;
279 struct obd_histogram cl_read_rpc_hist;
280 struct obd_histogram cl_write_rpc_hist;
281 struct obd_histogram cl_read_page_hist;
282 struct obd_histogram cl_write_page_hist;
283 struct obd_histogram cl_read_offset_hist;
284 struct obd_histogram cl_write_offset_hist;
286 /** LRU for osc caching pages */
287 struct cl_client_cache *cl_cache;
288 /** member of cl_cache->ccc_lru */
289 struct list_head cl_lru_osc;
290 /** # of available LRU slots left in the per-OSC cache.
291 * Available LRU slots are shared by all OSCs of the same file system,
292 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
293 atomic_long_t *cl_lru_left;
294 /** # of busy LRU pages. A page is considered busy if it's in writeback
295 * queue, or in transfer. Busy pages can't be discarded so they are not
297 atomic_long_t cl_lru_busy;
298 /** # of LRU pages in the cache for this client_obd */
299 atomic_long_t cl_lru_in_list;
300 /** # of threads are shrinking LRU cache. To avoid contention, it's not
301 * allowed to have multiple threads shrinking LRU cache. */
302 atomic_t cl_lru_shrinkers;
303 /** The time when this LRU cache was last used. */
304 time64_t cl_lru_last_used;
305 /** stats: how many reclaims have happened for this client_obd.
306 * reclaim and shrink - shrink is async, voluntarily rebalancing;
307 * reclaim is sync, initiated by IO thread when the LRU slots are
309 __u64 cl_lru_reclaim;
310 /** List of LRU pages for this client_obd */
311 struct list_head cl_lru_list;
312 /** Lock for LRU page list */
313 spinlock_t cl_lru_list_lock;
314 /** # of unstable pages in this client_obd.
315 * An unstable page is a page state that WRITE RPC has finished but
316 * the transaction has NOT yet committed. */
317 atomic_long_t cl_unstable_count;
318 /** Link to osc_shrinker_list */
319 struct list_head cl_shrink_list;
321 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
322 atomic_t cl_destroy_in_flight;
323 wait_queue_head_t cl_destroy_waitq;
325 /* modify rpcs in flight
326 * currently used for metadata only */
327 spinlock_t cl_mod_rpcs_lock;
328 __u16 cl_max_mod_rpcs_in_flight;
329 __u16 cl_mod_rpcs_in_flight;
330 __u16 cl_close_rpcs_in_flight;
331 wait_queue_head_t cl_mod_rpcs_waitq;
332 unsigned long *cl_mod_tag_bitmap;
333 struct obd_histogram cl_mod_rpcs_hist;
336 struct mutex cl_mgc_mutex;
337 struct local_oid_storage *cl_mgc_los;
338 struct dt_object *cl_mgc_configs_dir;
339 struct obd_export *cl_mgc_mgsexp;
340 atomic_t cl_mgc_refcount;
341 /* in-flight control list and total RPCs counter */
342 struct list_head cl_flight_waiters;
343 __u32 cl_rpcs_in_flight;
345 /* checksumming for data sent over the network */
346 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
347 cl_checksum_dump:1; /* same */
348 /* supported checksum types that are worked out at connect time */
349 __u32 cl_supp_cksum_types;
350 /* checksum algorithm to be used */
351 enum cksum_types cl_cksum_type;
352 /* preferred checksum algorithm to be used */
353 enum cksum_types cl_preferred_cksum_type;
355 /* also protected by the poorly named _loi_list_lock lock above */
356 struct osc_async_rc cl_ar;
358 /* sequence manager */
359 struct lu_client_seq *cl_seq;
360 struct rw_semaphore cl_seq_rwsem;
362 atomic_t cl_resends; /* resend count */
364 /* ptlrpc work for writeback in ptlrpcd context */
365 void *cl_writeback_work;
367 struct mutex cl_quota_mutex;
368 /* hash tables for osc_quota_info */
369 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
370 /* the xid of the request updating the hash tables */
371 __u64 cl_quota_last_xid;
372 /* Links to the global list of registered changelog devices */
373 struct list_head cl_chg_dev_linkage;
375 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
382 struct echo_client_obd {
383 struct obd_export *ec_exp; /* the local connection to osc/lov */
385 struct list_head ec_objects;
386 struct list_head ec_locks;
390 /* allow statfs data caching for 1 second */
391 #define OBD_STATFS_CACHE_SECONDS 1
392 /* arbitrary maximum. larger would be useless, allows catching bogus input */
393 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
395 #define lov_tgt_desc lu_tgt_desc
397 struct lov_md_tgt_desc {
398 struct obd_device *lmtd_mdc;
403 struct lov_desc desc;
404 struct lov_tgt_desc **lov_tgts; /* sparse array */
405 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
407 struct mutex lov_lock;
408 struct obd_connect_data lov_ocd;
409 struct proc_dir_entry *targets_proc_entry;
410 atomic_t lov_refcount;
411 __u32 lov_death_row; /* tgts scheduled to be deleted */
412 __u32 lov_tgt_size; /* size of tgts array */
415 struct rhashtable lov_pools_hash_body; /* used for key access */
416 struct list_head lov_pool_list; /* used for sequential access */
417 struct proc_dir_entry *lov_pool_proc_entry;
418 enum lustre_sec_part lov_sp_me;
420 /* Cached LRU and unstable data from upper layer */
421 struct cl_client_cache *lov_cache;
423 struct rw_semaphore lov_notify_lock;
424 /* Data-on-MDT: MDC array */
425 struct lov_md_tgt_desc *lov_mdc_tgts;
427 struct kobject *lov_tgts_kobj;
430 #define lmv_tgt_desc lu_tgt_desc
433 struct lu_client_fld lmv_fld;
439 u32 lmv_statfs_start;
441 struct lu_tgt_descs lmv_mdt_descs;
443 struct obd_connect_data conn_data;
444 struct kobject *lmv_tgts_kobj;
447 __u32 lmv_qos_rr_index;
450 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
451 #define lmv_qos lmv_mdt_descs.ltd_qos
453 /* Minimum sector size is 512 */
454 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
456 struct niobuf_local {
457 __u64 lnb_file_offset;
458 __u32 lnb_page_offset;
462 struct page *lnb_page;
464 __u16 lnb_guards[MAX_GUARD_NUMBER];
465 __u16 lnb_guard_rpc:1;
466 __u16 lnb_guard_disk:1;
467 /* separate unlock for read path to allow shared access */
471 struct tgt_thread_big_cache {
472 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
475 #define LUSTRE_FLD_NAME "fld"
476 #define LUSTRE_SEQ_NAME "seq"
478 #define LUSTRE_MDD_NAME "mdd"
479 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
480 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
481 #define LUSTRE_VVP_NAME "vvp"
482 #define LUSTRE_LMV_NAME "lmv"
483 #define LUSTRE_SLP_NAME "slp"
484 #define LUSTRE_LOD_NAME "lod"
485 #define LUSTRE_OSP_NAME "osp"
486 #define LUSTRE_LWP_NAME "lwp"
488 /* obd device type names */
489 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
490 #define LUSTRE_MDS_NAME "mds"
491 #define LUSTRE_MDT_NAME "mdt"
492 #define LUSTRE_MDC_NAME "mdc"
493 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
494 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
495 #define LUSTRE_OSC_NAME "osc"
496 #define LUSTRE_LOV_NAME "lov"
497 #define LUSTRE_MGS_NAME "mgs"
498 #define LUSTRE_MGC_NAME "mgc"
500 #define LUSTRE_ECHO_NAME "obdecho"
501 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
502 #define LUSTRE_QMT_NAME "qmt"
504 /* Constant obd names (post-rename) */
505 #define LUSTRE_MDS_OBDNAME "MDS"
506 #define LUSTRE_OSS_OBDNAME "OSS"
507 #define LUSTRE_MGS_OBDNAME "MGS"
508 #define LUSTRE_MGC_OBDNAME "MGC"
510 static inline int is_lwp_on_mdt(char *name)
514 ptr = strrchr(name, '-');
516 CERROR("%s is not a obdname\n", name);
520 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
522 if (strncmp(ptr + 1, "MDT", 3) != 0)
525 while (*(--ptr) != '-' && ptr != name);
530 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
536 static inline int is_lwp_on_ost(char *name)
540 ptr = strrchr(name, '-');
542 CERROR("%s is not a obdname\n", name);
546 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
548 if (strncmp(ptr + 1, "OST", 3) != 0)
551 while (*(--ptr) != '-' && ptr != name);
556 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
563 * Events signalled through obd_notify() upcall-chain.
565 enum obd_notify_event {
566 /* Device connect start */
568 /* Device activated */
570 /* Device deactivated */
572 /* Connect data for import were changed */
574 /* Administratively deactivate/activate event */
575 OBD_NOTIFY_DEACTIVATE,
580 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
581 * being main example).
583 struct obd_notify_upcall {
584 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
585 enum obd_notify_event ev, void *owner);
586 /* Opaque datum supplied by upper layer listener */
590 struct target_recovery_data {
591 svc_handler_t trd_recovery_handler;
592 pid_t trd_processing_task;
593 struct completion trd_starting;
594 struct completion trd_finishing;
597 struct obd_llog_group {
598 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
599 wait_queue_head_t olg_waitq;
603 /* corresponds to one of the obd's */
604 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
607 struct obd_type *obd_type;
608 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
609 int obd_minor; /* device number: lctl dl */
610 struct lu_device *obd_lu_dev;
612 /* common and UUID name of this device */
613 struct obd_uuid obd_uuid;
614 char obd_name[MAX_OBD_NAME];
616 /* bitfield modification is protected by obd_dev_lock */
618 obd_attached:1, /* finished attach */
619 obd_set_up:1, /* finished setup */
620 obd_recovering:1, /* there are recoverable clients */
621 obd_abort_recovery:1, /* recovery expired */
622 obd_version_recov:1, /* obd uses version checking */
623 obd_replayable:1, /* recovery enabled; inform clients */
624 obd_no_recov:1, /* fail instead of retry messages */
625 obd_stopping:1, /* started cleanup */
626 obd_starting:1, /* started setup */
627 obd_force:1, /* cleanup with > 0 obd refcount */
628 obd_fail:1, /* cleanup with failover */
629 obd_no_conn:1, /* deny new connections */
630 obd_inactive:1, /* device active/inactive
631 * (for /proc/status only!!) */
632 obd_no_ir:1, /* no imperative recovery. */
633 obd_process_conf:1, /* device is processing mgs config */
634 obd_checksum_dump:1; /* dump pages upon cksum error */
635 #ifdef HAVE_SERVER_SUPPORT
636 /* no committed-transno notification */
637 unsigned long obd_no_transno:1;
640 /* use separate field as it is set in interrupt to don't mess with
641 * protection of other bits using _bh lock */
642 unsigned long obd_recovery_expired:1;
643 /* uuid-export hash body */
644 struct rhashtable obd_uuid_hash;
645 /* nid-export hash body */
646 struct rhltable obd_nid_hash;
648 struct cfs_hash *obd_nid_stats_hash;
649 /* client_generation-export hash body */
650 struct cfs_hash *obd_gen_hash;
651 struct list_head obd_nid_stats;
652 struct list_head obd_exports;
653 struct list_head obd_unlinked_exports;
654 struct list_head obd_delayed_exports;
655 struct list_head obd_lwp_list;
656 atomic_t obd_refcount;
658 spinlock_t obd_nid_lock;
659 struct ldlm_namespace *obd_namespace;
660 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
661 /* a spinlock is OK for what we do now, may need a semaphore later */
662 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
663 spinlock_t obd_osfs_lock;
664 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
665 time64_t obd_osfs_age;
666 __u64 obd_last_committed;
667 struct mutex obd_dev_mutex;
668 struct lvfs_run_ctxt obd_lvfs_ctxt;
669 struct obd_llog_group obd_olg; /* default llog group */
670 struct obd_device *obd_observer;
671 struct rw_semaphore obd_observer_link_sem;
672 struct obd_notify_upcall obd_upcall;
673 struct obd_export *obd_self_export;
674 struct obd_export *obd_lwp_export;
675 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
676 struct list_head obd_exports_timed;
677 time64_t obd_eviction_timer; /* for ping evictor */
679 atomic_t obd_max_recoverable_clients;
680 atomic_t obd_connected_clients;
681 int obd_stale_clients;
682 /* this lock protects all recovery list_heads, timer and
683 * obd_next_recovery_transno value */
684 spinlock_t obd_recovery_task_lock;
685 __u64 obd_next_recovery_transno;
686 int obd_replayed_requests;
687 int obd_requests_queued_for_recovery;
688 wait_queue_head_t obd_next_transno_waitq;
689 /* protected by obd_recovery_task_lock */
690 struct hrtimer obd_recovery_timer;
692 time64_t obd_recovery_start;
693 /* seconds, for lprocfs_status */
694 time64_t obd_recovery_end;
695 /* To tell timeouts from time stamps Lustre uses timeout_t
696 * instead of time64_t.
698 timeout_t obd_recovery_time_hard;
699 timeout_t obd_recovery_timeout;
700 int obd_recovery_ir_factor;
702 /* new recovery stuff from CMD2 */
703 int obd_replayed_locks;
704 atomic_t obd_req_replay_clients;
705 atomic_t obd_lock_replay_clients;
706 struct target_recovery_data obd_recovery_data;
708 /* all lists are protected by obd_recovery_task_lock */
709 struct list_head obd_req_replay_queue;
710 struct list_head obd_lock_replay_queue;
711 struct list_head obd_final_req_queue;
714 #ifdef HAVE_SERVER_SUPPORT
715 struct obd_device_target obt;
716 struct filter_obd filter;
718 struct echo_obd echo;
720 struct client_obd cli;
721 struct echo_client_obd echo_client;
726 /* Fields used by LProcFS */
727 struct lprocfs_stats *obd_stats;
729 struct lprocfs_stats *obd_md_stats;
731 struct dentry *obd_debugfs_entry;
732 struct proc_dir_entry *obd_proc_entry;
733 struct proc_dir_entry *obd_proc_exports_entry;
734 struct dentry *obd_svc_debugfs_entry;
735 struct lprocfs_stats *obd_svc_stats;
736 const struct attribute **obd_attrs;
737 struct lprocfs_vars *obd_vars;
738 struct ldebugfs_vars *obd_debugfs_vars;
739 atomic_t obd_evict_inprogress;
740 wait_queue_head_t obd_evict_inprogress_waitq;
741 struct list_head obd_evict_list; /* protected with pet_lock */
744 * LDLM pool part. Save last calculated SLV and Limit.
746 rwlock_t obd_pool_lock;
750 int obd_conn_inprogress;
753 * List of outstanding class_incref()'s fo this OBD. For debugging. */
754 struct lu_ref obd_reference;
756 struct kset obd_kset; /* sysfs object collection */
757 struct kobj_type obd_ktype;
758 struct completion obd_kobj_unregister;
761 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
762 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
763 #ifdef HAVE_SERVER_SUPPORT
764 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
765 struct obd_uuid *uuid);
767 int obd_nid_export_for_each(struct obd_device *obd, lnet_nid_t nid,
768 int cb(struct obd_export *exp, void *data),
770 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
771 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
774 /* get/set_info keys */
775 #define KEY_ASYNC "async"
776 #define KEY_CHANGELOG_CLEAR "changelog_clear"
777 #define KEY_FID2PATH "fid2path"
778 #define KEY_CHECKSUM "checksum"
779 #define KEY_CLEAR_FS "clear_fs"
780 #define KEY_CONN_DATA "conn_data"
781 #define KEY_EVICT_BY_NID "evict_by_nid"
782 #define KEY_FIEMAP "fiemap"
783 #define KEY_FLUSH_CTX "flush_ctx"
784 #define KEY_GRANT_SHRINK "grant_shrink"
785 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
786 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
787 #define KEY_INTERMDS "inter_mds"
788 #define KEY_LAST_ID "last_id"
789 #define KEY_LAST_FID "last_fid"
790 #define KEY_MAX_EASIZE "max_easize"
791 #define KEY_DEFAULT_EASIZE "default_easize"
792 #define KEY_MGSSEC "mgssec"
793 #define KEY_READ_ONLY "read-only"
794 #define KEY_REGISTER_TARGET "register_target"
795 #define KEY_SET_FS "set_fs"
796 #define KEY_TGT_COUNT "tgt_count"
797 /* KEY_SET_INFO in lustre_idl.h */
798 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
800 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
801 #define KEY_OSP_CONNECTED "osp_connected"
803 /* Flags for op_xvalid */
805 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
806 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
807 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
808 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
809 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
810 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
811 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
816 static inline int it_to_lock_mode(struct lookup_intent *it)
818 /* CREAT needs to be tested before open (both could be set) */
819 if (it->it_op & IT_CREAT)
821 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
823 else if (it->it_op & IT_LAYOUT)
824 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
825 else if (it->it_op & IT_READDIR)
827 else if (it->it_op & IT_GETXATTR)
830 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
835 MF_MDC_CANCEL_FID1 = BIT(0),
836 MF_MDC_CANCEL_FID2 = BIT(1),
837 MF_MDC_CANCEL_FID3 = BIT(2),
838 MF_MDC_CANCEL_FID4 = BIT(3),
839 MF_GET_MDT_IDX = BIT(4),
843 CLI_SET_MEA = BIT(0),
844 CLI_RM_ENTRY = BIT(1),
847 CLI_MIGRATE = BIT(4),
848 CLI_DIRTY_DATA = BIT(5),
852 LUSTRE_OPC_MKDIR = 0,
853 LUSTRE_OPC_SYMLINK = 1,
854 LUSTRE_OPC_MKNOD = 2,
855 LUSTRE_OPC_CREATE = 3,
860 * GETXATTR is not included as only a couple of fields in the reply body
861 * is filled, but not FID which is needed for common intent handling in
862 * mdc_finish_intent_lock()
864 static inline bool it_has_reply_body(const struct lookup_intent *it)
866 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
870 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
871 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
872 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
873 struct lu_fid op_fid4; /* to the operation locks. */
874 u32 op_mds; /* what mds server open will go to */
876 enum md_op_code op_code;
877 struct lustre_handle op_open_handle;
881 struct rw_semaphore *op_mea1_sem;
882 struct rw_semaphore *op_mea2_sem;
883 struct lmv_stripe_md *op_mea1;
884 struct lmv_stripe_md *op_mea2;
885 struct lmv_stripe_md *op_default_mea1; /* default LMV */
886 __u32 op_suppgids[2];
893 /* iattr fields and blocks. */
894 struct iattr op_attr;
895 enum op_xvalid op_xvalid; /* eXtra validity flags */
896 loff_t op_attr_blocks;
897 u64 op_valid; /* OBD_MD_* */
898 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
900 enum md_op_flags op_flags;
902 /* Various operation flags. */
903 enum mds_op_bias op_bias;
905 /* used to transfer info between the stacks of MD client
906 * see enum op_cli_flags */
907 enum md_cli_flags op_cli_flags;
909 /* File object data version for HSM release, on client */
910 __u64 op_data_version;
911 struct lustre_handle op_lease_handle;
913 /* File security context, for creates/metadata ops */
914 const char *op_file_secctx_name;
915 __u32 op_file_secctx_name_size;
916 void *op_file_secctx;
917 __u32 op_file_secctx_size;
919 /* File encryption context, for creates/metadata ops */
920 void *op_file_encctx;
921 __u32 op_file_encctx_size;
925 /* Used by readdir */
926 unsigned int op_max_pages;
931 * used to access dir that is changing layout: if it's set, access
932 * dir by new layout, otherwise old layout.
933 * By default it's not set, because new files are created under new
934 * layout, if we can't find file with name under both old and new
935 * layout, we are sure file with name doesn't exist, but in reverse
936 * order there may be a race with creation by others.
939 /* used to access dir with bash hash */
940 __u32 op_stripe_index;
941 /* Archive ID for PCC attach */
946 int (*md_blocking_ast)(struct ldlm_lock *lock,
947 struct ldlm_lock_desc *desc,
948 void *data, int flag);
951 struct md_enqueue_info;
952 /* metadata stat-ahead */
953 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
954 struct md_enqueue_info *minfo,
957 struct md_enqueue_info {
958 struct md_op_data mi_data;
959 struct lookup_intent mi_it;
960 struct lustre_handle mi_lockh;
961 struct inode *mi_dir;
962 struct ldlm_enqueue_info mi_einfo;
963 md_enqueue_cb_t mi_cb;
968 struct module *o_owner;
969 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
970 void *karg, void __user *uarg);
971 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
972 __u32 keylen, void *key, __u32 *vallen, void *val);
973 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
974 __u32 keylen, void *key,
975 __u32 vallen, void *val,
976 struct ptlrpc_request_set *set);
977 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
978 int (*o_precleanup)(struct obd_device *obd);
979 int (*o_cleanup)(struct obd_device *obd);
980 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
981 int (*o_postrecov)(struct obd_device *obd);
982 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
984 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
985 /* connect to the target device with given connection
986 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
987 * granted by the target, which are guaranteed to be a subset of flags
988 * asked for. If @ocd == NULL, use default parameters. */
989 int (*o_connect)(const struct lu_env *env,
990 struct obd_export **exp, struct obd_device *src,
991 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
993 int (*o_reconnect)(const struct lu_env *env,
994 struct obd_export *exp, struct obd_device *src,
995 struct obd_uuid *cluuid,
996 struct obd_connect_data *ocd,
998 int (*o_disconnect)(struct obd_export *exp);
1000 /* Initialize/finalize fids infrastructure. */
1001 int (*o_fid_init)(struct obd_device *obd,
1002 struct obd_export *exp, enum lu_cli_type type);
1003 int (*o_fid_fini)(struct obd_device *obd);
1005 /* Allocate new fid according to passed @hint. */
1006 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1007 struct lu_fid *fid, struct md_op_data *op_data);
1010 * Object with @fid is getting deleted, we may want to do something
1013 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1014 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1015 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1016 time64_t max_age, struct ptlrpc_request_set *set);
1017 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1019 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1021 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1023 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1025 int (*o_preprw)(const struct lu_env *env, int cmd,
1026 struct obd_export *exp, struct obdo *oa, int objcount,
1027 struct obd_ioobj *obj, struct niobuf_remote *remote,
1028 int *nr_pages, struct niobuf_local *local);
1029 int (*o_commitrw)(const struct lu_env *env, int cmd,
1030 struct obd_export *exp, struct obdo *oa,
1031 int objcount, struct obd_ioobj *obj,
1032 struct niobuf_remote *remote, int pages,
1033 struct niobuf_local *local, int rc);
1034 int (*o_init_export)(struct obd_export *exp);
1035 int (*o_destroy_export)(struct obd_export *exp);
1037 int (*o_import_event)(struct obd_device *, struct obd_import *,
1038 enum obd_import_event);
1040 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1041 enum obd_notify_event ev);
1043 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1044 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1047 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1048 struct obd_quotactl *);
1051 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1052 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1053 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1055 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1059 /* lmv structures */
1061 struct mdt_body *body;
1062 struct lu_buf layout;
1064 struct lmv_stripe_md *lmv;
1065 struct lmv_foreign_md *lfm;
1067 struct lmv_stripe_md *default_lmv;
1068 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1069 struct posix_acl *posix_acl;
1073 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1074 static inline void lmd_clear_acl(struct lustre_md *md)
1076 if (md->posix_acl) {
1077 posix_acl_release(md->posix_acl);
1078 md->posix_acl = NULL;
1082 #define OBD_CONNECT_ACL_FLAGS \
1083 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1085 static inline void lmd_clear_acl(struct lustre_md *md)
1089 #define OBD_CONNECT_ACL_FLAGS (0)
1092 struct md_open_data {
1093 struct obd_client_handle *mod_och;
1094 struct ptlrpc_request *mod_open_req;
1095 struct ptlrpc_request *mod_close_req;
1096 atomic_t mod_refcount;
1100 struct obd_client_handle {
1101 struct lustre_handle och_open_handle;
1102 struct lu_fid och_fid;
1103 struct md_open_data *och_mod;
1104 struct lustre_handle och_lease_handle; /* open lock for lease */
1109 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1111 struct lookup_intent;
1115 int (*m_close)(struct obd_export *, struct md_op_data *,
1116 struct md_open_data *, struct ptlrpc_request **);
1118 int (*m_create)(struct obd_export *, struct md_op_data *,
1119 const void *, size_t, umode_t, uid_t, gid_t,
1120 cfs_cap_t, __u64, struct ptlrpc_request **);
1122 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1123 const union ldlm_policy_data *, struct md_op_data *,
1124 struct lustre_handle *, __u64);
1126 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1127 struct ptlrpc_request **);
1129 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1130 struct lookup_intent *,
1131 struct ptlrpc_request **,
1132 ldlm_blocking_callback, __u64);
1134 int (*m_link)(struct obd_export *, struct md_op_data *,
1135 struct ptlrpc_request **);
1137 int (*m_rename)(struct obd_export *, struct md_op_data *,
1138 const char *, size_t, const char *, size_t,
1139 struct ptlrpc_request **);
1141 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1142 size_t , struct ptlrpc_request **);
1144 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1145 struct ptlrpc_request **);
1147 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1148 struct md_callback *cb_op, __u64 hash_offset,
1149 struct page **ppage);
1151 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1152 struct ptlrpc_request **);
1154 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1155 u64, const char *, const void *, size_t, unsigned int,
1156 u32, struct ptlrpc_request **);
1158 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1159 u64, const char *, size_t, struct ptlrpc_request **);
1161 int (*m_intent_getattr_async)(struct obd_export *,
1162 struct md_enqueue_info *);
1164 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1165 struct lu_fid *, __u64 *bits);
1167 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1169 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1170 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1172 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1173 struct ptlrpc_request **);
1175 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1177 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1178 struct obd_export *, struct obd_export *,
1179 struct lustre_md *);
1181 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1183 int (*m_merge_attr)(struct obd_export *, const struct lu_fid *fid,
1184 const struct lmv_stripe_md *lsm,
1185 struct cl_attr *attr, ldlm_blocking_callback);
1187 int (*m_set_open_replay_data)(struct obd_export *,
1188 struct obd_client_handle *,
1189 struct lookup_intent *);
1191 int (*m_clear_open_replay_data)(struct obd_export *,
1192 struct obd_client_handle *);
1194 int (*m_set_lock_data)(struct obd_export *,
1195 const struct lustre_handle *, void *, __u64 *);
1197 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1198 const struct lu_fid *, enum ldlm_type,
1199 union ldlm_policy_data *, enum ldlm_mode,
1200 struct lustre_handle *);
1202 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1203 union ldlm_policy_data *, enum ldlm_mode,
1204 enum ldlm_cancel_flags flags, void *opaque);
1206 int (*m_get_fid_from_lsm)(struct obd_export *,
1207 const struct lmv_stripe_md *,
1208 const char *name, int namelen,
1209 struct lu_fid *fid);
1210 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1211 const union lmv_mds_md *lmv, size_t lmv_size);
1212 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1213 struct ptlrpc_request_set *set);
1216 static inline struct md_open_data *obd_mod_alloc(void)
1218 struct md_open_data *mod;
1222 atomic_set(&mod->mod_refcount, 1);
1226 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1227 #define obd_mod_put(mod) \
1229 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1230 if ((mod)->mod_open_req) \
1231 ptlrpc_req_finished((mod)->mod_open_req); \
1232 OBD_FREE_PTR(mod); \
1236 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1237 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1238 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1240 /* return 1 if client should be resend request */
1241 static inline int client_should_resend(int resend, struct client_obd *cli)
1243 return atomic_read(&cli->cl_resends) ?
1244 atomic_read(&cli->cl_resends) > resend : 1;
1248 * Return device name for this device
1250 * XXX: lu_device is declared before obd_device, while a pointer pointing
1251 * back to obd_device in lu_device, so this helper function defines here
1252 * instead of in lu_object.h
1254 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1256 return lu_dev->ld_obd->obd_name;
1259 static inline bool filename_is_volatile(const char *name, size_t namelen,
1265 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1268 /* caller does not care of idx */
1272 /* volatile file, the MDT can be set from name */
1273 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1274 /* if no MDT is specified, use std way */
1275 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1277 /* test for no MDT idx case */
1278 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1279 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1283 /* we have an idx, read it */
1284 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1285 *idx = simple_strtoul(start, &end, 16);
1287 * no digit, no trailing :, negative value
1289 if (((*idx == 0) && (end == start)) ||
1290 (*end != ':') || (*idx < 0))
1295 /* bad format of mdt idx, we cannot return an error
1296 * to caller so we use hash algo */
1297 CERROR("Bad volatile file name format: %s\n",
1298 name + LUSTRE_VOLATILE_HDR_LEN);
1302 static inline int cli_brw_size(struct obd_device *obd)
1304 LASSERT(obd != NULL);
1305 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1309 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1310 * of the client should be increased accordingly to avoid sending fragmented
1311 * RPCs over the network when the client runs out of the maximum dirty space
1312 * when so many RPCs are being generated.
1314 static inline void client_adjust_max_dirty(struct client_obd *cli)
1317 if (cli->cl_dirty_max_pages <= 0) {
1318 cli->cl_dirty_max_pages =
1319 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1321 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1322 cli->cl_max_pages_per_rpc;
1324 if (dirty_max > cli->cl_dirty_max_pages)
1325 cli->cl_dirty_max_pages = dirty_max;
1328 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1329 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1331 /* This value is exported to userspace through the max_dirty_mb
1332 * parameter. So we round up the number of pages to make it a round
1334 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1335 1 << (20 - PAGE_SHIFT));
1338 static inline struct inode *page2inode(struct page *page)
1340 if (page->mapping) {
1344 return page->mapping->host;
1350 #endif /* __OBD_H */