4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/kobject.h>
37 #include <linux/spinlock.h>
38 #include <linux/sysfs.h>
39 #include <linux/xarray.h>
41 #include <uapi/linux/lustre/lustre_idl.h>
42 #include <lustre_lib.h>
43 #include <libcfs/bitmap.h>
44 #ifdef HAVE_SERVER_SUPPORT
45 # include <lu_target.h>
46 # include <obd_target.h>
49 #include <lustre_export.h>
50 #include <lustre_fid.h>
51 #include <lustre_fld.h>
52 #include <lustre_handles.h>
53 #include <lustre_intent.h>
55 #include <lustre_quota.h>
57 #define MAX_OBD_DEVICES 8192
65 struct lov_oinfo { /* per-stripe data structure */
66 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
67 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
68 int loi_ost_gen; /* generation of this loi_ost_idx */
70 unsigned long loi_kms_valid:1;
71 __u64 loi_kms; /* known minimum size */
72 struct ost_lvb loi_lvb;
73 struct osc_async_rc loi_ar;
76 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
79 oinfo->loi_kms_valid = 1;
85 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
87 /* obd info for a particular level (lov, osc). */
89 /* OBD_STATFS_* flags */
91 struct obd_device *oi_obd;
92 struct lu_tgt_desc *oi_tgt;
93 /* statfs data specific for every OSC, if needed at all. */
94 struct obd_statfs *oi_osfs;
95 /* An update callback which is called to update some data on upper
96 * level. E.g. it is used for update lsm->lsm_oinfo at every received
97 * request in osc level for enqueue requests. It is also possible to
98 * update some caller data from LOV layer if needed. */
99 obd_enqueue_update_f oi_cb_up;
103 const struct obd_ops *typ_dt_ops;
104 const struct md_ops *typ_md_ops;
105 struct proc_dir_entry *typ_procroot;
106 struct dentry *typ_debugfs_entry;
107 #ifdef HAVE_SERVER_SUPPORT
111 struct lu_device_type *typ_lu;
112 struct kobject typ_kobj;
114 #define typ_name typ_kobj.name
115 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
124 struct timeout_item {
125 enum timeout_event ti_event;
129 struct list_head ti_obd_list;
130 struct list_head ti_chain;
133 #define OBD_MAX_RIF_DEFAULT 8
134 #define OBD_MAX_RIF_MAX 512
135 #define OSC_MAX_RIF_MAX 256
136 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
137 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
138 #define OSC_DEFAULT_RESENDS 10
140 /* possible values for lut_sync_lock_cancel */
141 enum tgt_sync_lock_cancel {
142 SYNC_LOCK_CANCEL_NEVER = 0,
143 SYNC_LOCK_CANCEL_BLOCKING = 1,
144 SYNC_LOCK_CANCEL_ALWAYS = 2,
148 * Limit reply buffer size for striping data to one x86_64 page. This
149 * value is chosen to fit the striping data for common use cases while
150 * staying well below the limit at which the buffer must be backed by
151 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
154 #define OBD_MAX_DEFAULT_EA_SIZE 4096
157 * Lustre can handle larger xattrs internally, but we must respect the Linux
158 * VFS limitation or tools like tar cannot interact with Lustre volumes
161 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
164 enum obd_cl_sem_lock_class {
173 struct rw_semaphore cl_sem;
174 struct obd_uuid cl_target_uuid;
175 struct obd_import *cl_import; /* ptlrpc connection state */
176 size_t cl_conn_count;
178 /* Cache maximum and default values for easize. This is
179 * strictly a performance optimization to minimize calls to
180 * obd_size_diskmd(). The default values are used to calculate the
181 * initial size of a request buffer. The ptlrpc layer will resize the
182 * buffer as needed to accommodate a larger reply from the
183 * server. The default values should be small enough to avoid wasted
184 * memory and excessive use of vmalloc(), yet large enough to avoid
185 * reallocating the buffer in the common use case. */
187 /* Default EA size for striping attributes. It is initialized at
188 * mount-time based on the default stripe width of the filesystem,
189 * then it tracks the largest observed EA size advertised by
190 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
191 __u32 cl_default_mds_easize;
193 /* Maximum possible EA size computed at mount-time based on
194 * the number of OSTs in the filesystem. May be increased at
195 * run-time if a larger observed size is advertised by the MDT. */
196 __u32 cl_max_mds_easize;
198 /* Data-on-MDT specific value to set larger reply buffer for possible
199 * data read along with open/stat requests. By default it tries to use
200 * unused space in reply buffer.
201 * This value is used to ensure that reply buffer has at least as
202 * much free space as value indicates. That free space is gained from
203 * LOV EA buffer which is small for DoM files and on big systems can
204 * provide up to 32KB of extra space in reply buffer.
205 * Default value is 8K now.
207 __u32 cl_dom_min_inline_repsize;
209 enum lustre_sec_part cl_sp_me;
210 enum lustre_sec_part cl_sp_to;
211 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
213 /* the grant values are protected by loi_list_lock below */
214 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
215 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
216 unsigned long cl_avail_grant; /* bytes of credit for ost */
217 unsigned long cl_lost_grant; /* lost credits (trunc) */
218 /* grant consumed for dirty pages */
219 unsigned long cl_dirty_grant;
221 /* since we allocate grant by blocks, we don't know how many grant will
222 * be used to add a page into cache. As a solution, we reserve maximum
223 * grant before trying to dirty a page and unreserve the rest.
224 * See osc_{reserve|unreserve}_grant for details. */
225 long cl_reserved_grant;
226 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
227 time64_t cl_next_shrink_grant; /* seconds */
228 struct list_head cl_grant_chain;
229 time64_t cl_grant_shrink_interval; /* seconds */
231 /* A chunk is an optimal size used by osc_extent to determine
232 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
234 /* extent insertion metadata overhead to be accounted in grant,
236 unsigned int cl_grant_extent_tax;
237 /* maximum extent size, in number of pages */
238 unsigned int cl_max_extent_pages;
240 /* keep track of objects that have lois that contain pages which
241 * have been queued for async brw. this lock also protects the
242 * lists of osc_client_pages that hang off of the loi */
244 * ->cl_loi_list_lock protects consistency of
245 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
246 * ->ap_completion() call-backs are executed under this lock. As we
247 * cannot guarantee that these call-backs never block on all platforms
248 * (as a matter of fact they do block on Mac OS X), type of
249 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
250 * and blocking mutex on Mac OS X. (Alternative is to make this lock
251 * blocking everywhere, but we don't want to slow down fast-path of
252 * our main platform.)
254 * NB by Jinshan: though field names are still _loi_, but actually
255 * osc_object{}s are in the list.
257 spinlock_t cl_loi_list_lock;
258 struct list_head cl_loi_ready_list;
259 struct list_head cl_loi_hp_ready_list;
260 struct list_head cl_loi_write_list;
261 struct list_head cl_loi_read_list;
262 __u32 cl_r_in_flight;
263 __u32 cl_w_in_flight;
264 /* just a sum of the loi/lop pending numbers to be exported by /proc */
265 atomic_t cl_pending_w_pages;
266 atomic_t cl_pending_r_pages;
267 u32 cl_max_pages_per_rpc;
268 u32 cl_max_rpcs_in_flight;
269 u32 cl_max_short_io_bytes;
270 struct obd_histogram cl_read_rpc_hist;
271 struct obd_histogram cl_write_rpc_hist;
272 struct obd_histogram cl_read_page_hist;
273 struct obd_histogram cl_write_page_hist;
274 struct obd_histogram cl_read_offset_hist;
275 struct obd_histogram cl_write_offset_hist;
277 /** LRU for osc caching pages */
278 struct cl_client_cache *cl_cache;
279 /** member of cl_cache->ccc_lru */
280 struct list_head cl_lru_osc;
281 /** # of available LRU slots left in the per-OSC cache.
282 * Available LRU slots are shared by all OSCs of the same file system,
283 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
284 atomic_long_t *cl_lru_left;
285 /** # of busy LRU pages. A page is considered busy if it's in writeback
286 * queue, or in transfer. Busy pages can't be discarded so they are not
288 atomic_long_t cl_lru_busy;
289 /** # of LRU pages in the cache for this client_obd */
290 atomic_long_t cl_lru_in_list;
291 /** # of threads are shrinking LRU cache. To avoid contention, it's not
292 * allowed to have multiple threads shrinking LRU cache. */
293 atomic_t cl_lru_shrinkers;
294 /** The time when this LRU cache was last used. */
295 time64_t cl_lru_last_used;
296 /** stats: how many reclaims have happened for this client_obd.
297 * reclaim and shrink - shrink is async, voluntarily rebalancing;
298 * reclaim is sync, initiated by IO thread when the LRU slots are
300 __u64 cl_lru_reclaim;
301 /** List of LRU pages for this client_obd */
302 struct list_head cl_lru_list;
303 /** Lock for LRU page list */
304 spinlock_t cl_lru_list_lock;
305 /** # of unstable pages in this client_obd.
306 * An unstable page is a page state that WRITE RPC has finished but
307 * the transaction has NOT yet committed. */
308 atomic_long_t cl_unstable_count;
309 /** Link to osc_shrinker_list */
310 struct list_head cl_shrink_list;
312 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
313 atomic_t cl_destroy_in_flight;
314 wait_queue_head_t cl_destroy_waitq;
316 /* modify rpcs in flight
317 * currently used for metadata only */
318 spinlock_t cl_mod_rpcs_lock;
319 __u16 cl_max_mod_rpcs_in_flight;
320 __u16 cl_mod_rpcs_in_flight;
321 __u16 cl_close_rpcs_in_flight;
322 wait_queue_head_t cl_mod_rpcs_waitq;
323 unsigned long *cl_mod_tag_bitmap;
324 struct obd_histogram cl_mod_rpcs_hist;
327 struct mutex cl_mgc_mutex;
328 struct local_oid_storage *cl_mgc_los;
329 struct dt_object *cl_mgc_configs_dir;
330 struct obd_export *cl_mgc_mgsexp;
331 atomic_t cl_mgc_refcount;
332 /* in-flight control list and total RPCs counter */
333 struct list_head cl_flight_waiters;
334 __u32 cl_rpcs_in_flight;
336 /* checksumming for data sent over the network */
337 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
338 cl_checksum_dump:1; /* same */
339 /* supported checksum types that are worked out at connect time */
340 __u32 cl_supp_cksum_types;
341 /* checksum algorithm to be used */
342 enum cksum_types cl_cksum_type;
343 /* preferred checksum algorithm to be used */
344 enum cksum_types cl_preferred_cksum_type;
346 /* also protected by the poorly named _loi_list_lock lock above */
347 struct osc_async_rc cl_ar;
349 /* sequence manager */
350 struct lu_client_seq *cl_seq;
351 struct rw_semaphore cl_seq_rwsem;
353 atomic_t cl_resends; /* resend count */
355 /* ptlrpc work for writeback in ptlrpcd context */
356 void *cl_writeback_work;
358 struct mutex cl_quota_mutex;
359 /* hash tables for osc_quota_info */
360 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
361 /* the xid of the request updating the hash tables */
362 __u64 cl_quota_last_xid;
363 /* Links to the global list of registered changelog devices */
364 struct list_head cl_chg_dev_linkage;
366 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
373 struct echo_client_obd {
374 struct obd_export *ec_exp; /* the local connection to osc/lov */
376 struct list_head ec_objects;
377 struct list_head ec_locks;
381 /* allow statfs data caching for 1 second */
382 #define OBD_STATFS_CACHE_SECONDS 1
383 /* arbitrary maximum. larger would be useless, allows catching bogus input */
384 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
386 #define lov_tgt_desc lu_tgt_desc
388 struct lov_md_tgt_desc {
389 struct obd_device *lmtd_mdc;
394 struct lov_desc desc;
395 struct lov_tgt_desc **lov_tgts; /* sparse array */
396 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
398 struct mutex lov_lock;
399 struct obd_connect_data lov_ocd;
400 struct proc_dir_entry *targets_proc_entry;
401 atomic_t lov_refcount;
402 __u32 lov_death_row; /* tgts scheduled to be deleted */
403 __u32 lov_tgt_size; /* size of tgts array */
406 struct rhashtable lov_pools_hash_body; /* used for key access */
407 struct list_head lov_pool_list; /* used for sequential access */
408 struct proc_dir_entry *lov_pool_proc_entry;
409 enum lustre_sec_part lov_sp_me;
411 /* Cached LRU and unstable data from upper layer */
412 struct cl_client_cache *lov_cache;
414 struct rw_semaphore lov_notify_lock;
415 /* Data-on-MDT: MDC array */
416 struct lov_md_tgt_desc *lov_mdc_tgts;
418 struct kobject *lov_tgts_kobj;
421 #define lmv_tgt_desc lu_tgt_desc
424 struct lu_client_fld lmv_fld;
430 u32 lmv_statfs_start;
432 struct lu_tgt_descs lmv_mdt_descs;
434 struct obd_connect_data conn_data;
435 struct kobject *lmv_tgts_kobj;
438 __u32 lmv_qos_rr_index;
441 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
442 #define lmv_qos lmv_mdt_descs.ltd_qos
444 /* Minimum sector size is 512 */
445 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
447 struct niobuf_local {
448 __u64 lnb_file_offset;
449 __u32 lnb_page_offset;
453 struct page *lnb_page;
455 __u16 lnb_guards[MAX_GUARD_NUMBER];
456 __u16 lnb_guard_rpc:1;
457 __u16 lnb_guard_disk:1;
458 /* separate unlock for read path to allow shared access */
462 struct tgt_thread_big_cache {
463 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
466 #define LUSTRE_FLD_NAME "fld"
467 #define LUSTRE_SEQ_NAME "seq"
469 #define LUSTRE_MDD_NAME "mdd"
470 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
471 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
472 #define LUSTRE_VVP_NAME "vvp"
473 #define LUSTRE_LMV_NAME "lmv"
474 #define LUSTRE_SLP_NAME "slp"
475 #define LUSTRE_LOD_NAME "lod"
476 #define LUSTRE_OSP_NAME "osp"
477 #define LUSTRE_LWP_NAME "lwp"
479 /* obd device type names */
480 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
481 #define LUSTRE_MDS_NAME "mds"
482 #define LUSTRE_MDT_NAME "mdt"
483 #define LUSTRE_MDC_NAME "mdc"
484 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
485 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
486 #define LUSTRE_OSC_NAME "osc"
487 #define LUSTRE_LOV_NAME "lov"
488 #define LUSTRE_MGS_NAME "mgs"
489 #define LUSTRE_MGC_NAME "mgc"
491 #define LUSTRE_ECHO_NAME "obdecho"
492 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
493 #define LUSTRE_QMT_NAME "qmt"
495 /* Constant obd names (post-rename) */
496 #define LUSTRE_MDS_OBDNAME "MDS"
497 #define LUSTRE_OSS_OBDNAME "OSS"
498 #define LUSTRE_MGS_OBDNAME "MGS"
499 #define LUSTRE_MGC_OBDNAME "MGC"
501 static inline int is_lwp_on_mdt(char *name)
505 ptr = strrchr(name, '-');
507 CERROR("%s is not a obdname\n", name);
511 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
513 if (strncmp(ptr + 1, "MDT", 3) != 0)
516 while (*(--ptr) != '-' && ptr != name);
521 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
527 static inline int is_lwp_on_ost(char *name)
531 ptr = strrchr(name, '-');
533 CERROR("%s is not a obdname\n", name);
537 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
539 if (strncmp(ptr + 1, "OST", 3) != 0)
542 while (*(--ptr) != '-' && ptr != name);
547 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
554 * Events signalled through obd_notify() upcall-chain.
556 enum obd_notify_event {
557 /* Device connect start */
559 /* Device activated */
561 /* Device deactivated */
563 /* Connect data for import were changed */
565 /* Administratively deactivate/activate event */
566 OBD_NOTIFY_DEACTIVATE,
571 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
572 * being main example).
574 struct obd_notify_upcall {
575 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
576 enum obd_notify_event ev, void *owner);
577 /* Opaque datum supplied by upper layer listener */
581 struct target_recovery_data {
582 svc_handler_t trd_recovery_handler;
583 pid_t trd_processing_task;
584 struct completion trd_starting;
585 struct completion trd_finishing;
588 struct obd_llog_group {
589 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
590 wait_queue_head_t olg_waitq;
594 /* corresponds to one of the obd's */
595 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
598 struct obd_type *obd_type;
599 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
600 int obd_minor; /* device number: lctl dl */
601 struct lu_device *obd_lu_dev;
603 /* common and UUID name of this device */
604 struct obd_uuid obd_uuid;
605 char obd_name[MAX_OBD_NAME];
607 /* bitfield modification is protected by obd_dev_lock */
609 obd_attached:1, /* finished attach */
610 obd_set_up:1, /* finished setup */
611 obd_recovering:1, /* there are recoverable clients */
612 obd_abort_recovery:1, /* recovery expired */
613 obd_version_recov:1, /* obd uses version checking */
614 obd_replayable:1, /* recovery enabled; inform clients */
615 obd_no_recov:1, /* fail instead of retry messages */
616 obd_stopping:1, /* started cleanup */
617 obd_starting:1, /* started setup */
618 obd_force:1, /* cleanup with > 0 obd refcount */
619 obd_fail:1, /* cleanup with failover */
620 obd_no_conn:1, /* deny new connections */
621 obd_inactive:1, /* device active/inactive
622 * (for /proc/status only!!) */
623 obd_no_ir:1, /* no imperative recovery. */
624 obd_process_conf:1, /* device is processing mgs config */
625 obd_checksum_dump:1; /* dump pages upon cksum error */
626 #ifdef HAVE_SERVER_SUPPORT
627 /* no committed-transno notification */
628 unsigned long obd_no_transno:1;
631 /* use separate field as it is set in interrupt to don't mess with
632 * protection of other bits using _bh lock */
633 unsigned long obd_recovery_expired:1;
634 /* uuid-export hash body */
635 struct rhashtable obd_uuid_hash;
636 /* nid-export hash body */
637 struct cfs_hash *obd_nid_hash;
639 struct cfs_hash *obd_nid_stats_hash;
640 /* client_generation-export hash body */
641 struct cfs_hash *obd_gen_hash;
642 struct list_head obd_nid_stats;
643 struct list_head obd_exports;
644 struct list_head obd_unlinked_exports;
645 struct list_head obd_delayed_exports;
646 struct list_head obd_lwp_list;
647 atomic_t obd_refcount;
649 spinlock_t obd_nid_lock;
650 struct ldlm_namespace *obd_namespace;
651 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
652 /* a spinlock is OK for what we do now, may need a semaphore later */
653 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
654 spinlock_t obd_osfs_lock;
655 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
656 time64_t obd_osfs_age;
657 __u64 obd_last_committed;
658 struct mutex obd_dev_mutex;
659 struct lvfs_run_ctxt obd_lvfs_ctxt;
660 struct obd_llog_group obd_olg; /* default llog group */
661 struct obd_device *obd_observer;
662 struct rw_semaphore obd_observer_link_sem;
663 struct obd_notify_upcall obd_upcall;
664 struct obd_export *obd_self_export;
665 struct obd_export *obd_lwp_export;
666 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
667 struct list_head obd_exports_timed;
668 time64_t obd_eviction_timer; /* for ping evictor */
670 atomic_t obd_max_recoverable_clients;
671 atomic_t obd_connected_clients;
672 int obd_stale_clients;
673 /* this lock protects all recovery list_heads, timer and
674 * obd_next_recovery_transno value */
675 spinlock_t obd_recovery_task_lock;
676 __u64 obd_next_recovery_transno;
677 int obd_replayed_requests;
678 int obd_requests_queued_for_recovery;
679 wait_queue_head_t obd_next_transno_waitq;
680 /* protected by obd_recovery_task_lock */
681 struct hrtimer obd_recovery_timer;
683 time64_t obd_recovery_start;
684 /* seconds, for lprocfs_status */
685 time64_t obd_recovery_end;
686 /* To tell timeouts from time stamps Lustre uses timeout_t
687 * instead of time64_t.
689 timeout_t obd_recovery_time_hard;
690 timeout_t obd_recovery_timeout;
691 int obd_recovery_ir_factor;
693 /* new recovery stuff from CMD2 */
694 int obd_replayed_locks;
695 atomic_t obd_req_replay_clients;
696 atomic_t obd_lock_replay_clients;
697 struct target_recovery_data obd_recovery_data;
699 /* all lists are protected by obd_recovery_task_lock */
700 struct list_head obd_req_replay_queue;
701 struct list_head obd_lock_replay_queue;
702 struct list_head obd_final_req_queue;
705 #ifdef HAVE_SERVER_SUPPORT
706 struct obd_device_target obt;
707 struct filter_obd filter;
709 struct echo_obd echo;
711 struct client_obd cli;
712 struct echo_client_obd echo_client;
717 /* Fields used by LProcFS */
718 struct lprocfs_stats *obd_stats;
720 struct lprocfs_stats *obd_md_stats;
722 struct dentry *obd_debugfs_entry;
723 struct proc_dir_entry *obd_proc_entry;
724 struct proc_dir_entry *obd_proc_exports_entry;
725 struct dentry *obd_svc_debugfs_entry;
726 struct lprocfs_stats *obd_svc_stats;
727 const struct attribute **obd_attrs;
728 struct lprocfs_vars *obd_vars;
729 atomic_t obd_evict_inprogress;
730 wait_queue_head_t obd_evict_inprogress_waitq;
731 struct list_head obd_evict_list; /* protected with pet_lock */
734 * LDLM pool part. Save last calculated SLV and Limit.
736 rwlock_t obd_pool_lock;
740 int obd_conn_inprogress;
743 * List of outstanding class_incref()'s fo this OBD. For debugging. */
744 struct lu_ref obd_reference;
746 struct kset obd_kset; /* sysfs object collection */
747 struct kobj_type obd_ktype;
748 struct completion obd_kobj_unregister;
751 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
752 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
753 #ifdef HAVE_SERVER_SUPPORT
754 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
755 struct obd_uuid *uuid);
758 /* get/set_info keys */
759 #define KEY_ASYNC "async"
760 #define KEY_CHANGELOG_CLEAR "changelog_clear"
761 #define KEY_FID2PATH "fid2path"
762 #define KEY_CHECKSUM "checksum"
763 #define KEY_CLEAR_FS "clear_fs"
764 #define KEY_CONN_DATA "conn_data"
765 #define KEY_EVICT_BY_NID "evict_by_nid"
766 #define KEY_FIEMAP "fiemap"
767 #define KEY_FLUSH_CTX "flush_ctx"
768 #define KEY_GRANT_SHRINK "grant_shrink"
769 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
770 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
771 #define KEY_INTERMDS "inter_mds"
772 #define KEY_LAST_ID "last_id"
773 #define KEY_LAST_FID "last_fid"
774 #define KEY_MAX_EASIZE "max_easize"
775 #define KEY_DEFAULT_EASIZE "default_easize"
776 #define KEY_MGSSEC "mgssec"
777 #define KEY_READ_ONLY "read-only"
778 #define KEY_REGISTER_TARGET "register_target"
779 #define KEY_SET_FS "set_fs"
780 #define KEY_TGT_COUNT "tgt_count"
781 /* KEY_SET_INFO in lustre_idl.h */
782 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
784 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
785 #define KEY_OSP_CONNECTED "osp_connected"
787 /* Flags for op_xvalid */
789 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
790 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
791 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
792 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
793 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
794 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
795 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
800 static inline int it_to_lock_mode(struct lookup_intent *it)
802 /* CREAT needs to be tested before open (both could be set) */
803 if (it->it_op & IT_CREAT)
805 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
807 else if (it->it_op & IT_LAYOUT)
808 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
809 else if (it->it_op & IT_READDIR)
811 else if (it->it_op & IT_GETXATTR)
814 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
819 MF_MDC_CANCEL_FID1 = BIT(0),
820 MF_MDC_CANCEL_FID2 = BIT(1),
821 MF_MDC_CANCEL_FID3 = BIT(2),
822 MF_MDC_CANCEL_FID4 = BIT(3),
823 MF_GET_MDT_IDX = BIT(4),
827 CLI_SET_MEA = BIT(0),
828 CLI_RM_ENTRY = BIT(1),
831 CLI_MIGRATE = BIT(4),
832 CLI_DIRTY_DATA = BIT(5),
836 LUSTRE_OPC_MKDIR = 0,
837 LUSTRE_OPC_SYMLINK = 1,
838 LUSTRE_OPC_MKNOD = 2,
839 LUSTRE_OPC_CREATE = 3,
844 * GETXATTR is not included as only a couple of fields in the reply body
845 * is filled, but not FID which is needed for common intent handling in
846 * mdc_finish_intent_lock()
848 static inline bool it_has_reply_body(const struct lookup_intent *it)
850 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
854 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
855 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
856 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
857 struct lu_fid op_fid4; /* to the operation locks. */
858 u32 op_mds; /* what mds server open will go to */
860 enum md_op_code op_code;
861 struct lustre_handle op_open_handle;
865 struct rw_semaphore *op_mea1_sem;
866 struct rw_semaphore *op_mea2_sem;
867 struct lmv_stripe_md *op_mea1;
868 struct lmv_stripe_md *op_mea2;
869 struct lmv_stripe_md *op_default_mea1; /* default LMV */
870 __u32 op_suppgids[2];
877 /* iattr fields and blocks. */
878 struct iattr op_attr;
879 enum op_xvalid op_xvalid; /* eXtra validity flags */
880 loff_t op_attr_blocks;
881 u64 op_valid; /* OBD_MD_* */
882 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
884 enum md_op_flags op_flags;
886 /* Various operation flags. */
887 enum mds_op_bias op_bias;
889 /* used to transfer info between the stacks of MD client
890 * see enum op_cli_flags */
891 enum md_cli_flags op_cli_flags;
893 /* File object data version for HSM release, on client */
894 __u64 op_data_version;
895 struct lustre_handle op_lease_handle;
897 /* File security context, for creates/metadata ops */
898 const char *op_file_secctx_name;
899 __u32 op_file_secctx_name_size;
900 void *op_file_secctx;
901 __u32 op_file_secctx_size;
905 /* Used by readdir */
906 unsigned int op_max_pages;
911 * used to access dir that is changing layout: if it's set, access
912 * dir by new layout, otherwise old layout.
913 * By default it's not set, because new files are created under new
914 * layout, if we can't find file with name under both old and new
915 * layout, we are sure file with name doesn't exist, but in reverse
916 * order there may be a race with creation by others.
919 /* used to access dir with bash hash */
920 __u32 op_stripe_index;
921 /* Archive ID for PCC attach */
926 int (*md_blocking_ast)(struct ldlm_lock *lock,
927 struct ldlm_lock_desc *desc,
928 void *data, int flag);
931 struct md_enqueue_info;
932 /* metadata stat-ahead */
933 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
934 struct md_enqueue_info *minfo,
937 struct md_enqueue_info {
938 struct md_op_data mi_data;
939 struct lookup_intent mi_it;
940 struct lustre_handle mi_lockh;
941 struct inode *mi_dir;
942 struct ldlm_enqueue_info mi_einfo;
943 md_enqueue_cb_t mi_cb;
948 struct module *o_owner;
949 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
950 void *karg, void __user *uarg);
951 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
952 __u32 keylen, void *key, __u32 *vallen, void *val);
953 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
954 __u32 keylen, void *key,
955 __u32 vallen, void *val,
956 struct ptlrpc_request_set *set);
957 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
958 int (*o_precleanup)(struct obd_device *obd);
959 int (*o_cleanup)(struct obd_device *obd);
960 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
961 int (*o_postrecov)(struct obd_device *obd);
962 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
964 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
965 /* connect to the target device with given connection
966 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
967 * granted by the target, which are guaranteed to be a subset of flags
968 * asked for. If @ocd == NULL, use default parameters. */
969 int (*o_connect)(const struct lu_env *env,
970 struct obd_export **exp, struct obd_device *src,
971 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
973 int (*o_reconnect)(const struct lu_env *env,
974 struct obd_export *exp, struct obd_device *src,
975 struct obd_uuid *cluuid,
976 struct obd_connect_data *ocd,
978 int (*o_disconnect)(struct obd_export *exp);
980 /* Initialize/finalize fids infrastructure. */
981 int (*o_fid_init)(struct obd_device *obd,
982 struct obd_export *exp, enum lu_cli_type type);
983 int (*o_fid_fini)(struct obd_device *obd);
985 /* Allocate new fid according to passed @hint. */
986 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
987 struct lu_fid *fid, struct md_op_data *op_data);
990 * Object with @fid is getting deleted, we may want to do something
993 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
994 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
995 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
996 time64_t max_age, struct ptlrpc_request_set *set);
997 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
999 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1001 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1003 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1005 int (*o_preprw)(const struct lu_env *env, int cmd,
1006 struct obd_export *exp, struct obdo *oa, int objcount,
1007 struct obd_ioobj *obj, struct niobuf_remote *remote,
1008 int *nr_pages, struct niobuf_local *local);
1009 int (*o_commitrw)(const struct lu_env *env, int cmd,
1010 struct obd_export *exp, struct obdo *oa,
1011 int objcount, struct obd_ioobj *obj,
1012 struct niobuf_remote *remote, int pages,
1013 struct niobuf_local *local, int rc);
1014 int (*o_init_export)(struct obd_export *exp);
1015 int (*o_destroy_export)(struct obd_export *exp);
1017 int (*o_import_event)(struct obd_device *, struct obd_import *,
1018 enum obd_import_event);
1020 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1021 enum obd_notify_event ev);
1023 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1024 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1027 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1028 struct obd_quotactl *);
1031 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1032 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1033 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1035 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1039 /* lmv structures */
1041 struct mdt_body *body;
1042 struct lu_buf layout;
1044 struct lmv_stripe_md *lmv;
1045 struct lmv_foreign_md *lfm;
1047 struct lmv_stripe_md *default_lmv;
1048 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1049 struct posix_acl *posix_acl;
1053 struct md_open_data {
1054 struct obd_client_handle *mod_och;
1055 struct ptlrpc_request *mod_open_req;
1056 struct ptlrpc_request *mod_close_req;
1057 atomic_t mod_refcount;
1061 struct obd_client_handle {
1062 struct lustre_handle och_open_handle;
1063 struct lu_fid och_fid;
1064 struct md_open_data *och_mod;
1065 struct lustre_handle och_lease_handle; /* open lock for lease */
1070 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1072 struct lookup_intent;
1076 int (*m_close)(struct obd_export *, struct md_op_data *,
1077 struct md_open_data *, struct ptlrpc_request **);
1079 int (*m_create)(struct obd_export *, struct md_op_data *,
1080 const void *, size_t, umode_t, uid_t, gid_t,
1081 cfs_cap_t, __u64, struct ptlrpc_request **);
1083 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1084 const union ldlm_policy_data *, struct md_op_data *,
1085 struct lustre_handle *, __u64);
1087 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1088 struct ptlrpc_request **);
1090 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1091 struct lookup_intent *,
1092 struct ptlrpc_request **,
1093 ldlm_blocking_callback, __u64);
1095 int (*m_link)(struct obd_export *, struct md_op_data *,
1096 struct ptlrpc_request **);
1098 int (*m_rename)(struct obd_export *, struct md_op_data *,
1099 const char *, size_t, const char *, size_t,
1100 struct ptlrpc_request **);
1102 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1103 size_t , struct ptlrpc_request **);
1105 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1106 struct ptlrpc_request **);
1108 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1109 struct md_callback *cb_op, __u64 hash_offset,
1110 struct page **ppage);
1112 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1113 struct ptlrpc_request **);
1115 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1116 u64, const char *, const void *, size_t, unsigned int,
1117 u32, struct ptlrpc_request **);
1119 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1120 u64, const char *, size_t, struct ptlrpc_request **);
1122 int (*m_intent_getattr_async)(struct obd_export *,
1123 struct md_enqueue_info *);
1125 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1126 struct lu_fid *, __u64 *bits);
1128 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1130 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1131 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1133 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1134 struct ptlrpc_request **);
1136 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1138 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1139 struct obd_export *, struct obd_export *,
1140 struct lustre_md *);
1142 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1144 int (*m_merge_attr)(struct obd_export *,
1145 const struct lmv_stripe_md *lsm,
1146 struct cl_attr *attr, ldlm_blocking_callback);
1148 int (*m_set_open_replay_data)(struct obd_export *,
1149 struct obd_client_handle *,
1150 struct lookup_intent *);
1152 int (*m_clear_open_replay_data)(struct obd_export *,
1153 struct obd_client_handle *);
1155 int (*m_set_lock_data)(struct obd_export *,
1156 const struct lustre_handle *, void *, __u64 *);
1158 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1159 const struct lu_fid *, enum ldlm_type,
1160 union ldlm_policy_data *, enum ldlm_mode,
1161 struct lustre_handle *);
1163 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1164 union ldlm_policy_data *, enum ldlm_mode,
1165 enum ldlm_cancel_flags flags, void *opaque);
1167 int (*m_get_fid_from_lsm)(struct obd_export *,
1168 const struct lmv_stripe_md *,
1169 const char *name, int namelen,
1170 struct lu_fid *fid);
1171 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1172 const union lmv_mds_md *lmv, size_t lmv_size);
1173 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1174 struct ptlrpc_request_set *set);
1177 static inline struct md_open_data *obd_mod_alloc(void)
1179 struct md_open_data *mod;
1183 atomic_set(&mod->mod_refcount, 1);
1187 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1188 #define obd_mod_put(mod) \
1190 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1191 if ((mod)->mod_open_req) \
1192 ptlrpc_req_finished((mod)->mod_open_req); \
1193 OBD_FREE_PTR(mod); \
1197 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1198 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1199 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1201 /* return 1 if client should be resend request */
1202 static inline int client_should_resend(int resend, struct client_obd *cli)
1204 return atomic_read(&cli->cl_resends) ?
1205 atomic_read(&cli->cl_resends) > resend : 1;
1209 * Return device name for this device
1211 * XXX: lu_device is declared before obd_device, while a pointer pointing
1212 * back to obd_device in lu_device, so this helper function defines here
1213 * instead of in lu_object.h
1215 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1217 return lu_dev->ld_obd->obd_name;
1220 static inline bool filename_is_volatile(const char *name, size_t namelen,
1226 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1229 /* caller does not care of idx */
1233 /* volatile file, the MDT can be set from name */
1234 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1235 /* if no MDT is specified, use std way */
1236 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1238 /* test for no MDT idx case */
1239 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1240 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1244 /* we have an idx, read it */
1245 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1246 *idx = simple_strtoul(start, &end, 16);
1248 * no digit, no trailing :, negative value
1250 if (((*idx == 0) && (end == start)) ||
1251 (*end != ':') || (*idx < 0))
1256 /* bad format of mdt idx, we cannot return an error
1257 * to caller so we use hash algo */
1258 CERROR("Bad volatile file name format: %s\n",
1259 name + LUSTRE_VOLATILE_HDR_LEN);
1263 static inline int cli_brw_size(struct obd_device *obd)
1265 LASSERT(obd != NULL);
1266 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1270 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1271 * of the client should be increased accordingly to avoid sending fragmented
1272 * RPCs over the network when the client runs out of the maximum dirty space
1273 * when so many RPCs are being generated.
1275 static inline void client_adjust_max_dirty(struct client_obd *cli)
1278 if (cli->cl_dirty_max_pages <= 0) {
1279 cli->cl_dirty_max_pages =
1280 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1282 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1283 cli->cl_max_pages_per_rpc;
1285 if (dirty_max > cli->cl_dirty_max_pages)
1286 cli->cl_dirty_max_pages = dirty_max;
1289 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1290 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1292 /* This value is exported to userspace through the max_dirty_mb
1293 * parameter. So we round up the number of pages to make it a round
1295 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1296 1 << (20 - PAGE_SHIFT));
1299 #endif /* __OBD_H */