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 enum lustre_sec_part cl_sp_me;
221 enum lustre_sec_part cl_sp_to;
222 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
224 /* the grant values are protected by loi_list_lock below */
225 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
226 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
227 unsigned long cl_avail_grant; /* bytes of credit for ost */
228 unsigned long cl_lost_grant; /* lost credits (trunc) */
229 /* grant consumed for dirty pages */
230 unsigned long cl_dirty_grant;
232 /* since we allocate grant by blocks, we don't know how many grant will
233 * be used to add a page into cache. As a solution, we reserve maximum
234 * grant before trying to dirty a page and unreserve the rest.
235 * See osc_{reserve|unreserve}_grant for details. */
236 long cl_reserved_grant;
237 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
238 time64_t cl_next_shrink_grant; /* seconds */
239 struct list_head cl_grant_chain;
240 time64_t cl_grant_shrink_interval; /* seconds */
242 int cl_root_squash; /* if root squash enabled*/
244 /* A chunk is an optimal size used by osc_extent to determine
245 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
247 /* extent insertion metadata overhead to be accounted in grant,
249 unsigned int cl_grant_extent_tax;
250 /* maximum extent size, in number of pages */
251 unsigned int cl_max_extent_pages;
253 /* keep track of objects that have lois that contain pages which
254 * have been queued for async brw. this lock also protects the
255 * lists of osc_client_pages that hang off of the loi */
257 * ->cl_loi_list_lock protects consistency of
258 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
259 * ->ap_completion() call-backs are executed under this lock. As we
260 * cannot guarantee that these call-backs never block on all platforms
261 * (as a matter of fact they do block on Mac OS X), type of
262 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
263 * and blocking mutex on Mac OS X. (Alternative is to make this lock
264 * blocking everywhere, but we don't want to slow down fast-path of
265 * our main platform.)
267 * NB by Jinshan: though field names are still _loi_, but actually
268 * osc_object{}s are in the list.
270 spinlock_t cl_loi_list_lock;
271 struct list_head cl_loi_ready_list;
272 struct list_head cl_loi_hp_ready_list;
273 struct list_head cl_loi_write_list;
274 struct list_head cl_loi_read_list;
275 __u32 cl_r_in_flight;
276 __u32 cl_w_in_flight;
277 /* just a sum of the loi/lop pending numbers to be exported by /proc */
278 atomic_t cl_pending_w_pages;
279 atomic_t cl_pending_r_pages;
280 u32 cl_max_pages_per_rpc;
281 u32 cl_max_rpcs_in_flight;
282 u32 cl_max_short_io_bytes;
283 ktime_t cl_stats_init;
284 struct obd_histogram cl_read_rpc_hist;
285 struct obd_histogram cl_write_rpc_hist;
286 struct obd_histogram cl_read_page_hist;
287 struct obd_histogram cl_write_page_hist;
288 struct obd_histogram cl_read_offset_hist;
289 struct obd_histogram cl_write_offset_hist;
291 /** LRU for osc caching pages */
292 struct cl_client_cache *cl_cache;
293 /** member of cl_cache->ccc_lru */
294 struct list_head cl_lru_osc;
295 /** # of available LRU slots left in the per-OSC cache.
296 * Available LRU slots are shared by all OSCs of the same file system,
297 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
298 atomic_long_t *cl_lru_left;
299 /** # of busy LRU pages. A page is considered busy if it's in writeback
300 * queue, or in transfer. Busy pages can't be discarded so they are not
302 atomic_long_t cl_lru_busy;
303 /** # of LRU pages in the cache for this client_obd */
304 atomic_long_t cl_lru_in_list;
305 /** # of threads are shrinking LRU cache. To avoid contention, it's not
306 * allowed to have multiple threads shrinking LRU cache. */
307 atomic_t cl_lru_shrinkers;
308 /** The time when this LRU cache was last used. */
309 time64_t cl_lru_last_used;
310 /** stats: how many reclaims have happened for this client_obd.
311 * reclaim and shrink - shrink is async, voluntarily rebalancing;
312 * reclaim is sync, initiated by IO thread when the LRU slots are
314 __u64 cl_lru_reclaim;
315 /** List of LRU pages for this client_obd */
316 struct list_head cl_lru_list;
317 /** Lock for LRU page list */
318 spinlock_t cl_lru_list_lock;
319 /** # of unstable pages in this client_obd.
320 * An unstable page is a page state that WRITE RPC has finished but
321 * the transaction has NOT yet committed. */
322 atomic_long_t cl_unstable_count;
323 /** Link to osc_shrinker_list */
324 struct list_head cl_shrink_list;
326 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
327 atomic_t cl_destroy_in_flight;
328 wait_queue_head_t cl_destroy_waitq;
330 /* modify rpcs in flight
331 * currently used for metadata only */
332 spinlock_t cl_mod_rpcs_lock;
333 __u16 cl_max_mod_rpcs_in_flight;
334 __u16 cl_mod_rpcs_in_flight;
335 __u16 cl_close_rpcs_in_flight;
336 wait_queue_head_t cl_mod_rpcs_waitq;
337 unsigned long *cl_mod_tag_bitmap;
338 ktime_t cl_mod_rpcs_init;
339 struct obd_histogram cl_mod_rpcs_hist;
342 struct mutex cl_mgc_mutex;
343 struct local_oid_storage *cl_mgc_los;
344 struct dt_object *cl_mgc_configs_dir;
345 struct obd_export *cl_mgc_mgsexp;
346 atomic_t cl_mgc_refcount;
347 /* in-flight control list and total RPCs counter */
348 struct list_head cl_flight_waiters;
349 __u32 cl_rpcs_in_flight;
351 /* supported checksum types that are worked out at connect time */
352 __u32 cl_supp_cksum_types;
353 /* checksum algorithm to be used */
354 enum cksum_types cl_cksum_type;
355 /* preferred checksum algorithm to be used */
356 enum cksum_types cl_preferred_cksum_type;
358 /* also protected by the poorly named _loi_list_lock lock above */
359 struct osc_async_rc cl_ar;
361 /* sequence manager */
362 struct lu_client_seq *cl_seq;
363 struct rw_semaphore cl_seq_rwsem;
365 atomic_t cl_resends; /* resend count */
367 /* ptlrpc work for writeback in ptlrpcd context */
368 void *cl_writeback_work;
370 struct mutex cl_quota_mutex;
371 /* hash tables for osc_quota_info */
372 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
373 /* the xid of the request updating the hash tables */
374 __u64 cl_quota_last_xid;
375 /* Links to the global list of registered changelog devices */
376 struct list_head cl_chg_dev_linkage;
378 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
385 struct echo_client_obd {
386 struct obd_export *ec_exp; /* the local connection to osc/lov */
388 struct list_head ec_objects;
389 struct list_head ec_locks;
393 /* allow statfs data caching for 1 second */
394 #define OBD_STATFS_CACHE_SECONDS 1
395 /* arbitrary maximum. larger would be useless, allows catching bogus input */
396 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
398 #define lov_tgt_desc lu_tgt_desc
400 struct lov_md_tgt_desc {
401 struct obd_device *lmtd_mdc;
406 struct lov_desc desc;
407 struct lov_tgt_desc **lov_tgts; /* sparse array */
408 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
410 struct mutex lov_lock;
411 struct obd_connect_data lov_ocd;
412 atomic_t lov_refcount;
413 __u32 lov_death_row; /* tgts scheduled to be deleted */
414 __u32 lov_tgt_size; /* size of tgts array */
417 struct rhashtable lov_pools_hash_body; /* used for key access */
418 struct list_head lov_pool_list; /* used for sequential access */
419 struct proc_dir_entry *lov_pool_proc_entry;
420 enum lustre_sec_part lov_sp_me;
422 /* Cached LRU and unstable data from upper layer */
423 struct cl_client_cache *lov_cache;
425 struct rw_semaphore lov_notify_lock;
426 /* Data-on-MDT: MDC array */
427 struct lov_md_tgt_desc *lov_mdc_tgts;
429 struct kobject *lov_tgts_kobj;
432 #define lmv_tgt_desc lu_tgt_desc
435 struct lu_client_fld lmv_fld;
441 u32 lmv_statfs_start;
443 struct lu_tgt_descs lmv_mdt_descs;
445 struct obd_connect_data conn_data;
446 struct kobject *lmv_tgts_kobj;
449 __u32 lmv_qos_rr_index;
452 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
453 #define lmv_qos lmv_mdt_descs.ltd_qos
455 /* Minimum sector size is 512 */
456 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
458 struct niobuf_local {
459 __u64 lnb_file_offset;
460 __u32 lnb_page_offset;
464 struct page *lnb_page;
466 __u16 lnb_guards[MAX_GUARD_NUMBER];
467 __u16 lnb_guard_rpc:1;
468 __u16 lnb_guard_disk:1;
469 /* separate unlock for read path to allow shared access */
473 struct tgt_thread_big_cache {
474 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
477 #define LUSTRE_FLD_NAME "fld"
478 #define LUSTRE_SEQ_NAME "seq"
480 #define LUSTRE_MDD_NAME "mdd"
481 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
482 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
483 #define LUSTRE_VVP_NAME "vvp"
484 #define LUSTRE_LMV_NAME "lmv"
485 #define LUSTRE_SLP_NAME "slp"
486 #define LUSTRE_LOD_NAME "lod"
487 #define LUSTRE_OSP_NAME "osp"
488 #define LUSTRE_LWP_NAME "lwp"
490 /* obd device type names */
491 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
492 #define LUSTRE_MDS_NAME "mds"
493 #define LUSTRE_MDT_NAME "mdt"
494 #define LUSTRE_MDC_NAME "mdc"
495 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
496 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
497 #define LUSTRE_OSC_NAME "osc"
498 #define LUSTRE_LOV_NAME "lov"
499 #define LUSTRE_MGS_NAME "mgs"
500 #define LUSTRE_MGC_NAME "mgc"
502 #define LUSTRE_ECHO_NAME "obdecho"
503 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
504 #define LUSTRE_QMT_NAME "qmt"
506 /* Constant obd names (post-rename) */
507 #define LUSTRE_MDS_OBDNAME "MDS"
508 #define LUSTRE_OSS_OBDNAME "OSS"
509 #define LUSTRE_MGS_OBDNAME "MGS"
510 #define LUSTRE_MGC_OBDNAME "MGC"
512 static inline int is_lwp_on_mdt(char *name)
516 ptr = strrchr(name, '-');
518 CERROR("%s is not a obdname\n", name);
522 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
524 if (strncmp(ptr + 1, "MDT", 3) != 0)
527 while (*(--ptr) != '-' && ptr != name);
532 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
538 static inline int is_lwp_on_ost(char *name)
542 ptr = strrchr(name, '-');
544 CERROR("%s is not a obdname\n", name);
548 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
550 if (strncmp(ptr + 1, "OST", 3) != 0)
553 while (*(--ptr) != '-' && ptr != name);
558 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
565 * Events signalled through obd_notify() upcall-chain.
567 enum obd_notify_event {
568 /* Device connect start */
570 /* Device activated */
572 /* Device deactivated */
574 /* Connect data for import were changed */
576 /* Administratively deactivate/activate event */
577 OBD_NOTIFY_DEACTIVATE,
582 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
583 * being main example).
585 struct obd_notify_upcall {
586 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
587 enum obd_notify_event ev, void *owner);
588 /* Opaque datum supplied by upper layer listener */
592 struct target_recovery_data {
593 svc_handler_t trd_recovery_handler;
594 pid_t trd_processing_task;
595 struct completion trd_starting;
596 struct completion trd_finishing;
599 struct obd_llog_group {
600 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
601 wait_queue_head_t olg_waitq;
605 /* corresponds to one of the obd's */
606 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
609 struct obd_type *obd_type;
610 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
611 int obd_minor; /* device number: lctl dl */
612 struct lu_device *obd_lu_dev;
614 /* common and UUID name of this device */
615 struct obd_uuid obd_uuid;
616 char obd_name[MAX_OBD_NAME];
618 /* bitfield modification is protected by obd_dev_lock */
620 obd_attached:1, /* finished attach */
621 obd_set_up:1, /* finished setup */
622 obd_recovering:1, /* there are recoverable clients */
623 obd_abort_recovery:1, /* recovery expired */
624 obd_abort_recov_mdt:1, /* only abort recovery between MDTs */
625 obd_version_recov:1, /* obd uses version checking */
626 obd_replayable:1, /* recovery enabled; inform clients */
627 obd_no_recov:1, /* fail instead of retry messages */
628 obd_stopping:1, /* started cleanup */
629 obd_starting:1, /* started setup */
630 obd_force:1, /* cleanup with > 0 obd refcount */
631 obd_fail:1, /* cleanup with failover */
632 obd_no_conn:1, /* deny new connections */
633 obd_inactive:1, /* device active/inactive
634 * (for /proc/status only!!) */
635 obd_no_ir:1, /* no imperative recovery. */
636 obd_process_conf:1, /* device is processing mgs config */
637 obd_checksum_dump:1, /* dump pages upon cksum error */
638 obd_dynamic_nids:1; /* Allow dynamic NIDs on device */
639 #ifdef HAVE_SERVER_SUPPORT
640 /* no committed-transno notification */
641 unsigned long obd_no_transno:1;
644 /* use separate field as it is set in interrupt to don't mess with
645 * protection of other bits using _bh lock */
646 unsigned long obd_recovery_expired:1;
647 /* uuid-export hash body */
648 struct rhashtable obd_uuid_hash;
649 /* nid-export hash body */
650 struct rhltable obd_nid_hash;
652 struct cfs_hash *obd_nid_stats_hash;
653 /* client_generation-export hash body */
654 struct cfs_hash *obd_gen_hash;
655 struct list_head obd_nid_stats;
656 struct list_head obd_exports;
657 struct list_head obd_unlinked_exports;
658 struct list_head obd_delayed_exports;
659 struct list_head obd_lwp_list;
660 atomic_t obd_refcount;
662 spinlock_t obd_nid_lock;
663 struct ldlm_namespace *obd_namespace;
664 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
665 /* a spinlock is OK for what we do now, may need a semaphore later */
666 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
667 spinlock_t obd_osfs_lock;
668 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
669 time64_t obd_osfs_age;
670 __u64 obd_last_committed;
671 struct mutex obd_dev_mutex;
672 struct lvfs_run_ctxt obd_lvfs_ctxt;
673 struct obd_llog_group obd_olg; /* default llog group */
674 struct obd_device *obd_observer;
675 struct rw_semaphore obd_observer_link_sem;
676 struct obd_notify_upcall obd_upcall;
677 struct obd_export *obd_self_export;
678 struct obd_export *obd_lwp_export;
679 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
680 struct list_head obd_exports_timed;
681 time64_t obd_eviction_timer; /* for ping evictor */
683 atomic_t obd_max_recoverable_clients;
684 atomic_t obd_connected_clients;
685 int obd_stale_clients;
686 /* this lock protects all recovery list_heads, timer and
687 * obd_next_recovery_transno value */
688 spinlock_t obd_recovery_task_lock;
689 __u64 obd_next_recovery_transno;
690 int obd_replayed_requests;
691 int obd_requests_queued_for_recovery;
692 wait_queue_head_t obd_next_transno_waitq;
693 /* protected by obd_recovery_task_lock */
694 struct hrtimer obd_recovery_timer;
696 time64_t obd_recovery_start;
697 /* seconds, for lprocfs_status */
698 time64_t obd_recovery_end;
699 /* To tell timeouts from time stamps Lustre uses timeout_t
700 * instead of time64_t.
702 timeout_t obd_recovery_time_hard;
703 timeout_t obd_recovery_timeout;
704 int obd_recovery_ir_factor;
706 /* new recovery stuff from CMD2 */
707 int obd_replayed_locks;
708 atomic_t obd_req_replay_clients;
709 atomic_t obd_lock_replay_clients;
710 struct target_recovery_data obd_recovery_data;
712 /* all lists are protected by obd_recovery_task_lock */
713 struct list_head obd_req_replay_queue;
714 struct list_head obd_lock_replay_queue;
715 struct list_head obd_final_req_queue;
718 #ifdef HAVE_SERVER_SUPPORT
719 struct obd_device_target obt;
720 struct filter_obd filter;
722 struct echo_obd echo;
724 struct client_obd cli;
725 struct echo_client_obd echo_client;
730 /* Fields used by LProcFS */
731 struct lprocfs_stats *obd_stats;
733 struct lprocfs_stats *obd_md_stats;
735 struct dentry *obd_debugfs_entry;
736 struct proc_dir_entry *obd_proc_entry;
737 struct proc_dir_entry *obd_proc_exports_entry;
738 struct dentry *obd_svc_debugfs_entry;
739 struct lprocfs_stats *obd_svc_stats;
740 const struct attribute **obd_attrs;
741 struct lprocfs_vars *obd_vars;
742 struct ldebugfs_vars *obd_debugfs_vars;
743 atomic_t obd_evict_inprogress;
744 wait_queue_head_t obd_evict_inprogress_waitq;
745 struct list_head obd_evict_list; /* protected with pet_lock */
748 * LDLM pool part. Save last calculated SLV and Limit.
750 rwlock_t obd_pool_lock;
754 int obd_conn_inprogress;
757 * List of outstanding class_incref()'s fo this OBD. For debugging. */
758 struct lu_ref obd_reference;
760 struct kset obd_kset; /* sysfs object collection */
761 struct kobj_type obd_ktype;
762 struct completion obd_kobj_unregister;
765 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
766 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
767 #ifdef HAVE_SERVER_SUPPORT
768 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
769 struct obd_uuid *uuid);
771 int obd_nid_export_for_each(struct obd_device *obd, lnet_nid_t nid,
772 int cb(struct obd_export *exp, void *data),
774 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
775 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
778 /* get/set_info keys */
779 #define KEY_ASYNC "async"
780 #define KEY_CHANGELOG_CLEAR "changelog_clear"
781 #define KEY_FID2PATH "fid2path"
782 #define KEY_CHECKSUM "checksum"
783 #define KEY_CLEAR_FS "clear_fs"
784 #define KEY_CONN_DATA "conn_data"
785 #define KEY_EVICT_BY_NID "evict_by_nid"
786 #define KEY_FIEMAP "fiemap"
787 #define KEY_FLUSH_CTX "flush_ctx"
788 #define KEY_GRANT_SHRINK "grant_shrink"
789 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
790 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
791 #define KEY_INTERMDS "inter_mds"
792 #define KEY_LAST_ID "last_id"
793 #define KEY_LAST_FID "last_fid"
794 #define KEY_MAX_EASIZE "max_easize"
795 #define KEY_DEFAULT_EASIZE "default_easize"
796 #define KEY_MGSSEC "mgssec"
797 #define KEY_READ_ONLY "read-only"
798 #define KEY_REGISTER_TARGET "register_target"
799 #define KEY_SET_FS "set_fs"
800 #define KEY_TGT_COUNT "tgt_count"
801 /* KEY_SET_INFO in lustre_idl.h */
802 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
804 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
805 #define KEY_OSP_CONNECTED "osp_connected"
807 /* Flags for op_xvalid */
809 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
810 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
811 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
812 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
813 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
814 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
815 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
820 static inline int it_to_lock_mode(struct lookup_intent *it)
822 /* CREAT needs to be tested before open (both could be set) */
823 if (it->it_op & IT_CREAT)
825 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
827 else if (it->it_op & IT_LAYOUT)
828 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
829 else if (it->it_op & IT_READDIR)
831 else if (it->it_op & IT_GETXATTR)
834 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
839 MF_MDC_CANCEL_FID1 = BIT(0),
840 MF_MDC_CANCEL_FID2 = BIT(1),
841 MF_MDC_CANCEL_FID3 = BIT(2),
842 MF_MDC_CANCEL_FID4 = BIT(3),
843 MF_GET_MDT_IDX = BIT(4),
844 MF_GETATTR_BY_FID = BIT(5),
845 MF_QOS_MKDIR = BIT(6),
846 MF_RR_MKDIR = BIT(7),
847 MF_OPNAME_KMALLOCED = BIT(8),
851 CLI_SET_MEA = BIT(0),
852 CLI_RM_ENTRY = BIT(1),
855 CLI_MIGRATE = BIT(4),
856 CLI_DIRTY_DATA = BIT(5),
860 LUSTRE_OPC_MKDIR = 1,
871 * GETXATTR is not included as only a couple of fields in the reply body
872 * is filled, but not FID which is needed for common intent handling in
873 * mdc_finish_intent_lock()
875 static inline bool it_has_reply_body(const struct lookup_intent *it)
877 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
881 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
882 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
883 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
884 struct lu_fid op_fid4; /* to the operation locks. */
885 u32 op_mds; /* what mds server open will go to */
887 enum md_op_code op_code;
888 struct lustre_handle op_open_handle;
892 struct rw_semaphore *op_mea1_sem;
893 struct rw_semaphore *op_mea2_sem;
894 struct lmv_stripe_md *op_mea1;
895 struct lmv_stripe_md *op_mea2;
896 struct lmv_stripe_md *op_default_mea1; /* default LMV */
897 __u32 op_suppgids[2];
904 /* iattr fields and blocks. */
905 struct iattr op_attr;
906 enum op_xvalid op_xvalid; /* eXtra validity flags */
907 loff_t op_attr_blocks;
908 u64 op_valid; /* OBD_MD_* */
909 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
911 enum md_op_flags op_flags;
913 /* Various operation flags. */
914 enum mds_op_bias op_bias;
916 /* used to transfer info between the stacks of MD client
917 * see enum op_cli_flags */
918 enum md_cli_flags op_cli_flags;
920 /* File object data version for HSM release, on client */
921 __u64 op_data_version;
922 struct lustre_handle op_lease_handle;
924 /* File security context, for creates/metadata ops */
925 const char *op_file_secctx_name;
926 __u32 op_file_secctx_name_size;
927 void *op_file_secctx;
928 __u32 op_file_secctx_size;
930 /* File encryption context, for creates/metadata ops */
931 void *op_file_encctx;
932 __u32 op_file_encctx_size;
937 /* Used by readdir */
938 unsigned int op_max_pages;
940 unsigned short op_dir_depth;
946 * used to access dir that is changing layout: if it's set, access
947 * dir by new layout, otherwise old layout.
948 * By default it's not set, because new files are created under new
949 * layout, if we can't find file with name under both old and new
950 * layout, we are sure file with name doesn't exist, but in reverse
951 * order there may be a race with creation by others.
954 /* used to access dir with bash hash */
955 __u32 op_stripe_index;
956 /* Archive ID for PCC attach */
961 int (*md_blocking_ast)(struct ldlm_lock *lock,
962 struct ldlm_lock_desc *desc,
963 void *data, int flag);
966 struct md_enqueue_info;
967 /* metadata stat-ahead */
968 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
969 struct md_enqueue_info *minfo,
972 struct md_enqueue_info {
973 struct md_op_data mi_data;
974 struct lookup_intent mi_it;
975 struct lustre_handle mi_lockh;
976 struct inode *mi_dir;
977 struct ldlm_enqueue_info mi_einfo;
978 md_enqueue_cb_t mi_cb;
983 struct module *o_owner;
984 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
985 void *karg, void __user *uarg);
986 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
987 __u32 keylen, void *key, __u32 *vallen, void *val);
988 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
989 __u32 keylen, void *key,
990 __u32 vallen, void *val,
991 struct ptlrpc_request_set *set);
992 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
993 int (*o_precleanup)(struct obd_device *obd);
994 int (*o_cleanup)(struct obd_device *obd);
995 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
996 int (*o_postrecov)(struct obd_device *obd);
997 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
999 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1000 /* connect to the target device with given connection
1001 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1002 * granted by the target, which are guaranteed to be a subset of flags
1003 * asked for. If @ocd == NULL, use default parameters. */
1004 int (*o_connect)(const struct lu_env *env,
1005 struct obd_export **exp, struct obd_device *src,
1006 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1008 int (*o_reconnect)(const struct lu_env *env,
1009 struct obd_export *exp, struct obd_device *src,
1010 struct obd_uuid *cluuid,
1011 struct obd_connect_data *ocd,
1013 int (*o_disconnect)(struct obd_export *exp);
1015 /* Initialize/finalize fids infrastructure. */
1016 int (*o_fid_init)(struct obd_device *obd,
1017 struct obd_export *exp, enum lu_cli_type type);
1018 int (*o_fid_fini)(struct obd_device *obd);
1020 /* Allocate new fid according to passed @hint. */
1021 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1022 struct lu_fid *fid, struct md_op_data *op_data);
1025 * Object with @fid is getting deleted, we may want to do something
1028 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1029 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1030 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1031 time64_t max_age, struct ptlrpc_request_set *set);
1032 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1034 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1036 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1038 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1040 int (*o_preprw)(const struct lu_env *env, int cmd,
1041 struct obd_export *exp, struct obdo *oa, int objcount,
1042 struct obd_ioobj *obj, struct niobuf_remote *remote,
1043 int *nr_pages, struct niobuf_local *local);
1044 int (*o_commitrw)(const struct lu_env *env, int cmd,
1045 struct obd_export *exp, struct obdo *oa,
1046 int objcount, struct obd_ioobj *obj,
1047 struct niobuf_remote *remote, int pages,
1048 struct niobuf_local *local, int rc);
1049 int (*o_init_export)(struct obd_export *exp);
1050 int (*o_destroy_export)(struct obd_export *exp);
1052 int (*o_import_event)(struct obd_device *, struct obd_import *,
1053 enum obd_import_event);
1055 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1056 enum obd_notify_event ev);
1058 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1059 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1062 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1063 struct obd_quotactl *);
1066 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1067 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1068 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1070 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1074 /* lmv structures */
1076 struct mdt_body *body;
1077 struct lu_buf layout;
1079 struct lmv_stripe_md *lmv;
1080 struct lmv_foreign_md *lfm;
1082 struct lmv_stripe_md *default_lmv;
1083 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1084 struct posix_acl *posix_acl;
1088 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1089 static inline void lmd_clear_acl(struct lustre_md *md)
1091 if (md->posix_acl) {
1092 posix_acl_release(md->posix_acl);
1093 md->posix_acl = NULL;
1097 #define OBD_CONNECT_ACL_FLAGS \
1098 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1100 static inline void lmd_clear_acl(struct lustre_md *md)
1104 #define OBD_CONNECT_ACL_FLAGS (0)
1107 struct md_open_data {
1108 struct obd_client_handle *mod_och;
1109 struct ptlrpc_request *mod_open_req;
1110 struct ptlrpc_request *mod_close_req;
1111 atomic_t mod_refcount;
1115 struct obd_client_handle {
1116 struct lustre_handle och_open_handle;
1117 struct lu_fid och_fid;
1118 struct md_open_data *och_mod;
1119 struct lustre_handle och_lease_handle; /* open lock for lease */
1124 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1126 struct lookup_intent;
1130 int (*m_close)(struct obd_export *, struct md_op_data *,
1131 struct md_open_data *, struct ptlrpc_request **);
1133 int (*m_create)(struct obd_export *, struct md_op_data *,
1134 const void *, size_t, umode_t, uid_t, gid_t,
1135 kernel_cap_t, __u64, struct ptlrpc_request **);
1137 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1138 const union ldlm_policy_data *, struct md_op_data *,
1139 struct lustre_handle *, __u64);
1141 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1142 struct ptlrpc_request **);
1144 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1145 struct lookup_intent *,
1146 struct ptlrpc_request **,
1147 ldlm_blocking_callback, __u64);
1149 int (*m_link)(struct obd_export *, struct md_op_data *,
1150 struct ptlrpc_request **);
1152 int (*m_rename)(struct obd_export *, struct md_op_data *,
1153 const char *, size_t, const char *, size_t,
1154 struct ptlrpc_request **);
1156 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1157 size_t , struct ptlrpc_request **);
1159 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1160 struct ptlrpc_request **);
1162 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1163 struct md_callback *cb_op, __u64 hash_offset,
1164 struct page **ppage);
1166 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1167 struct ptlrpc_request **);
1169 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1170 u64, const char *, const void *, size_t, unsigned int,
1171 u32, struct ptlrpc_request **);
1173 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1174 u64, const char *, size_t, struct ptlrpc_request **);
1176 int (*m_intent_getattr_async)(struct obd_export *,
1177 struct md_enqueue_info *);
1179 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1180 struct lu_fid *, __u64 *bits);
1182 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1184 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1185 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1187 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1188 struct ptlrpc_request **);
1190 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1192 int (*m_get_lustre_md)(struct obd_export *, struct req_capsule *,
1193 struct obd_export *, struct obd_export *,
1194 struct lustre_md *);
1196 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1198 int (*m_merge_attr)(struct obd_export *,
1199 const struct lmv_stripe_md *lsm,
1200 struct cl_attr *attr, ldlm_blocking_callback);
1202 int (*m_set_open_replay_data)(struct obd_export *,
1203 struct obd_client_handle *,
1204 struct lookup_intent *);
1206 int (*m_clear_open_replay_data)(struct obd_export *,
1207 struct obd_client_handle *);
1209 int (*m_set_lock_data)(struct obd_export *,
1210 const struct lustre_handle *, void *, __u64 *);
1212 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1213 const struct lu_fid *, enum ldlm_type,
1214 union ldlm_policy_data *, enum ldlm_mode,
1215 struct lustre_handle *);
1217 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1218 union ldlm_policy_data *, enum ldlm_mode,
1219 enum ldlm_cancel_flags flags, void *opaque);
1221 int (*m_get_fid_from_lsm)(struct obd_export *,
1222 const struct lmv_stripe_md *,
1223 const char *name, int namelen,
1224 struct lu_fid *fid);
1225 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1226 const union lmv_mds_md *lmv, size_t lmv_size);
1227 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1228 struct ptlrpc_request_set *set);
1231 static inline struct md_open_data *obd_mod_alloc(void)
1233 struct md_open_data *mod;
1237 atomic_set(&mod->mod_refcount, 1);
1241 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1242 #define obd_mod_put(mod) \
1244 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1245 if ((mod)->mod_open_req) \
1246 ptlrpc_req_finished((mod)->mod_open_req); \
1247 OBD_FREE_PTR(mod); \
1251 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1252 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1253 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1255 /* return 1 if client should be resend request */
1256 static inline int client_should_resend(int resend, struct client_obd *cli)
1258 return atomic_read(&cli->cl_resends) ?
1259 atomic_read(&cli->cl_resends) > resend : 1;
1263 * Return device name for this device
1265 * XXX: lu_device is declared before obd_device, while a pointer pointing
1266 * back to obd_device in lu_device, so this helper function defines here
1267 * instead of in lu_object.h
1269 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1271 return lu_dev->ld_obd->obd_name;
1274 static inline bool filename_is_volatile(const char *name, size_t namelen,
1280 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1283 /* caller does not care of idx */
1287 /* volatile file, the MDT can be set from name */
1288 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1289 /* if no MDT is specified, use std way */
1290 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1292 /* test for no MDT idx case */
1293 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1294 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1298 /* we have an idx, read it */
1299 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1300 *idx = simple_strtoul(start, &end, 16);
1302 * no digit, no trailing :, negative value
1304 if (((*idx == 0) && (end == start)) ||
1305 (*end != ':') || (*idx < 0))
1310 /* bad format of mdt idx, we cannot return an error
1311 * to caller so we use hash algo */
1312 CERROR("Bad volatile file name format: %s\n",
1313 name + LUSTRE_VOLATILE_HDR_LEN);
1317 static inline int cli_brw_size(struct obd_device *obd)
1319 LASSERT(obd != NULL);
1320 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1324 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1325 * of the client should be increased accordingly to avoid sending fragmented
1326 * RPCs over the network when the client runs out of the maximum dirty space
1327 * when so many RPCs are being generated.
1329 static inline void client_adjust_max_dirty(struct client_obd *cli)
1332 if (cli->cl_dirty_max_pages <= 0) {
1333 cli->cl_dirty_max_pages =
1334 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1336 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1337 cli->cl_max_pages_per_rpc;
1339 if (dirty_max > cli->cl_dirty_max_pages)
1340 cli->cl_dirty_max_pages = dirty_max;
1343 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1344 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1346 /* This value is exported to userspace through the max_dirty_mb
1347 * parameter. So we round up the number of pages to make it a round
1349 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1350 1 << (20 - PAGE_SHIFT));
1353 static inline struct inode *page2inode(struct page *page)
1355 if (page->mapping) {
1359 return page->mapping->host;
1365 #endif /* __OBD_H */