4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/kobject.h>
37 #include <linux/spinlock.h>
38 #include <linux/sysfs.h>
40 #include <uapi/linux/lustre/lustre_idl.h>
41 #include <lustre_lib.h>
42 #include <libcfs/bitmap.h>
43 #ifdef HAVE_SERVER_SUPPORT
44 # include <lu_target.h>
45 # include <obd_target.h>
48 #include <lustre_export.h>
49 #include <lustre_fid.h>
50 #include <lustre_fld.h>
51 #include <lustre_handles.h>
52 #include <lustre_intent.h>
54 #include <lustre_quota.h>
56 #define MAX_OBD_DEVICES 8192
64 struct lov_oinfo { /* per-stripe data structure */
65 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
66 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
67 int loi_ost_gen; /* generation of this loi_ost_idx */
69 unsigned long loi_kms_valid:1;
70 __u64 loi_kms; /* known minimum size */
71 struct ost_lvb loi_lvb;
72 struct osc_async_rc loi_ar;
75 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
78 oinfo->loi_kms_valid = 1;
84 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
86 /* obd info for a particular level (lov, osc). */
88 /* OBD_STATFS_* flags */
90 struct obd_device *oi_obd;
91 struct lu_tgt_desc *oi_tgt;
92 /* statfs data specific for every OSC, if needed at all. */
93 struct obd_statfs *oi_osfs;
94 /* An update callback which is called to update some data on upper
95 * level. E.g. it is used for update lsm->lsm_oinfo at every received
96 * request in osc level for enqueue requests. It is also possible to
97 * update some caller data from LOV layer if needed. */
98 obd_enqueue_update_f oi_cb_up;
102 struct list_head typ_chain;
103 struct obd_ops *typ_dt_ops;
104 struct md_ops *typ_md_ops;
105 struct proc_dir_entry *typ_procroot;
106 struct dentry *typ_debugfs_entry;
107 #ifdef HAVE_SERVER_SUPPORT
111 struct lu_device_type *typ_lu;
112 spinlock_t obd_type_lock;
113 struct kobject typ_kobj;
115 #define typ_name typ_kobj.name
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 fo_sync_lock_cancel */
142 NEVER_SYNC_ON_CANCEL = 0,
143 BLOCKING_SYNC_ON_CANCEL = 1,
144 ALWAYS_SYNC_ON_CANCEL = 2,
145 NUM_SYNC_ON_CANCEL_STATES
149 * Limit reply buffer size for striping data to one x86_64 page. This
150 * value is chosen to fit the striping data for common use cases while
151 * staying well below the limit at which the buffer must be backed by
152 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
155 #define OBD_MAX_DEFAULT_EA_SIZE 4096
158 * Lustre can handle larger xattrs internally, but we must respect the Linux
159 * VFS limitation or tools like tar cannot interact with Lustre volumes
162 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
165 enum obd_cl_sem_lock_class {
174 struct rw_semaphore cl_sem;
175 struct obd_uuid cl_target_uuid;
176 struct obd_import *cl_import; /* ptlrpc connection state */
177 size_t cl_conn_count;
179 /* Cache maximum and default values for easize. This is
180 * strictly a performance optimization to minimize calls to
181 * obd_size_diskmd(). The default values are used to calculate the
182 * initial size of a request buffer. The ptlrpc layer will resize the
183 * buffer as needed to accommodate a larger reply from the
184 * server. The default values should be small enough to avoid wasted
185 * memory and excessive use of vmalloc(), yet large enough to avoid
186 * reallocating the buffer in the common use case. */
188 /* Default EA size for striping attributes. It is initialized at
189 * mount-time based on the default stripe width of the filesystem,
190 * then it tracks the largest observed EA size advertised by
191 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
192 __u32 cl_default_mds_easize;
194 /* Maximum possible EA size computed at mount-time based on
195 * the number of OSTs in the filesystem. May be increased at
196 * run-time if a larger observed size is advertised by the MDT. */
197 __u32 cl_max_mds_easize;
199 /* Data-on-MDT specific value to set larger reply buffer for possible
200 * data read along with open/stat requests. By default it tries to use
201 * unused space in reply buffer.
202 * This value is used to ensure that reply buffer has at least as
203 * much free space as value indicates. That free space is gained from
204 * LOV EA buffer which is small for DoM files and on big systems can
205 * provide up to 32KB of extra space in reply buffer.
206 * Default value is 8K now.
208 __u32 cl_dom_min_inline_repsize;
210 enum lustre_sec_part cl_sp_me;
211 enum lustre_sec_part cl_sp_to;
212 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
214 /* the grant values are protected by loi_list_lock below */
215 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
216 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
217 unsigned long cl_dirty_transit; /* dirty synchronous */
218 unsigned long cl_avail_grant; /* bytes of credit for ost */
219 unsigned long cl_lost_grant; /* lost credits (trunc) */
220 /* grant consumed for dirty pages */
221 unsigned long cl_dirty_grant;
223 /* since we allocate grant by blocks, we don't know how many grant will
224 * be used to add a page into cache. As a solution, we reserve maximum
225 * grant before trying to dirty a page and unreserve the rest.
226 * See osc_{reserve|unreserve}_grant for details. */
227 long cl_reserved_grant;
228 struct list_head cl_cache_waiters; /* waiting for cache/grant */
229 time64_t cl_next_shrink_grant; /* seconds */
230 struct list_head cl_grant_chain;
231 time64_t cl_grant_shrink_interval; /* seconds */
233 /* A chunk is an optimal size used by osc_extent to determine
234 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
236 /* extent insertion metadata overhead to be accounted in grant,
238 unsigned int cl_grant_extent_tax;
239 /* maximum extent size, in number of pages */
240 unsigned int cl_max_extent_pages;
242 /* keep track of objects that have lois that contain pages which
243 * have been queued for async brw. this lock also protects the
244 * lists of osc_client_pages that hang off of the loi */
246 * ->cl_loi_list_lock protects consistency of
247 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
248 * ->ap_completion() call-backs are executed under this lock. As we
249 * cannot guarantee that these call-backs never block on all platforms
250 * (as a matter of fact they do block on Mac OS X), type of
251 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
252 * and blocking mutex on Mac OS X. (Alternative is to make this lock
253 * blocking everywhere, but we don't want to slow down fast-path of
254 * our main platform.)
256 * NB by Jinshan: though field names are still _loi_, but actually
257 * osc_object{}s are in the list.
259 spinlock_t cl_loi_list_lock;
260 struct list_head cl_loi_ready_list;
261 struct list_head cl_loi_hp_ready_list;
262 struct list_head cl_loi_write_list;
263 struct list_head cl_loi_read_list;
264 __u32 cl_r_in_flight;
265 __u32 cl_w_in_flight;
266 /* just a sum of the loi/lop pending numbers to be exported by /proc */
267 atomic_t cl_pending_w_pages;
268 atomic_t cl_pending_r_pages;
269 u32 cl_max_pages_per_rpc;
270 u32 cl_max_rpcs_in_flight;
271 u32 cl_max_short_io_bytes;
272 struct obd_histogram cl_read_rpc_hist;
273 struct obd_histogram cl_write_rpc_hist;
274 struct obd_histogram cl_read_page_hist;
275 struct obd_histogram cl_write_page_hist;
276 struct obd_histogram cl_read_offset_hist;
277 struct obd_histogram cl_write_offset_hist;
279 /** LRU for osc caching pages */
280 struct cl_client_cache *cl_cache;
281 /** member of cl_cache->ccc_lru */
282 struct list_head cl_lru_osc;
283 /** # of available LRU slots left in the per-OSC cache.
284 * Available LRU slots are shared by all OSCs of the same file system,
285 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
286 atomic_long_t *cl_lru_left;
287 /** # of busy LRU pages. A page is considered busy if it's in writeback
288 * queue, or in transfer. Busy pages can't be discarded so they are not
290 atomic_long_t cl_lru_busy;
291 /** # of LRU pages in the cache for this client_obd */
292 atomic_long_t cl_lru_in_list;
293 /** # of threads are shrinking LRU cache. To avoid contention, it's not
294 * allowed to have multiple threads shrinking LRU cache. */
295 atomic_t cl_lru_shrinkers;
296 /** The time when this LRU cache was last used. */
297 time64_t cl_lru_last_used;
298 /** stats: how many reclaims have happened for this client_obd.
299 * reclaim and shrink - shrink is async, voluntarily rebalancing;
300 * reclaim is sync, initiated by IO thread when the LRU slots are
302 __u64 cl_lru_reclaim;
303 /** List of LRU pages for this client_obd */
304 struct list_head cl_lru_list;
305 /** Lock for LRU page list */
306 spinlock_t cl_lru_list_lock;
307 /** # of unstable pages in this client_obd.
308 * An unstable page is a page state that WRITE RPC has finished but
309 * the transaction has NOT yet committed. */
310 atomic_long_t cl_unstable_count;
311 /** Link to osc_shrinker_list */
312 struct list_head cl_shrink_list;
314 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
315 atomic_t cl_destroy_in_flight;
316 wait_queue_head_t cl_destroy_waitq;
318 /* modify rpcs in flight
319 * currently used for metadata only */
320 spinlock_t cl_mod_rpcs_lock;
321 __u16 cl_max_mod_rpcs_in_flight;
322 __u16 cl_mod_rpcs_in_flight;
323 __u16 cl_close_rpcs_in_flight;
324 wait_queue_head_t cl_mod_rpcs_waitq;
325 unsigned long *cl_mod_tag_bitmap;
326 struct obd_histogram cl_mod_rpcs_hist;
329 struct mutex cl_mgc_mutex;
330 struct local_oid_storage *cl_mgc_los;
331 struct dt_object *cl_mgc_configs_dir;
332 struct obd_export *cl_mgc_mgsexp;
333 atomic_t cl_mgc_refcount;
334 /* in-flight control list and total RPCs counter */
335 struct list_head cl_flight_waiters;
336 __u32 cl_rpcs_in_flight;
338 /* checksumming for data sent over the network */
339 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
340 cl_checksum_dump:1; /* same */
341 /* supported checksum types that are worked out at connect time */
342 __u32 cl_supp_cksum_types;
343 /* checksum algorithm to be used */
344 enum cksum_types cl_cksum_type;
345 /* preferred checksum algorithm to be used */
346 enum cksum_types cl_preferred_cksum_type;
348 /* also protected by the poorly named _loi_list_lock lock above */
349 struct osc_async_rc cl_ar;
351 /* sequence manager */
352 struct lu_client_seq *cl_seq;
353 struct rw_semaphore cl_seq_rwsem;
355 atomic_t cl_resends; /* resend count */
357 /* ptlrpc work for writeback in ptlrpcd context */
358 void *cl_writeback_work;
360 struct mutex cl_quota_mutex;
361 /* hash tables for osc_quota_info */
362 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
363 /* the xid of the request updating the hash tables */
364 __u64 cl_quota_last_xid;
365 /* Links to the global list of registered changelog devices */
366 struct list_head cl_chg_dev_linkage;
368 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
375 struct echo_client_obd {
376 struct obd_export *ec_exp; /* the local connection to osc/lov */
378 struct list_head ec_objects;
379 struct list_head ec_locks;
383 /* allow statfs data caching for 1 second */
384 #define OBD_STATFS_CACHE_SECONDS 1
385 /* arbitrary maximum. larger would be useless, allows catching bogus input */
386 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
388 #define lov_tgt_desc lu_tgt_desc
390 struct lov_md_tgt_desc {
391 struct obd_device *lmtd_mdc;
396 struct lov_desc desc;
397 struct lov_tgt_desc **lov_tgts; /* sparse array */
398 struct ost_pool lov_packed; /* all OSTs in a packed
400 struct mutex lov_lock;
401 struct obd_connect_data lov_ocd;
402 struct proc_dir_entry *targets_proc_entry;
403 atomic_t lov_refcount;
404 __u32 lov_death_row; /* tgts scheduled to be deleted */
405 __u32 lov_tgt_size; /* size of tgts array */
408 struct cfs_hash *lov_pools_hash_body; /* used for key access */
409 struct list_head lov_pool_list; /* used for sequential access */
410 struct proc_dir_entry *lov_pool_proc_entry;
411 enum lustre_sec_part lov_sp_me;
413 /* Cached LRU and unstable data from upper layer */
414 struct cl_client_cache *lov_cache;
416 struct rw_semaphore lov_notify_lock;
417 /* Data-on-MDT: MDC array */
418 struct lov_md_tgt_desc *lov_mdc_tgts;
420 struct kobject *lov_tgts_kobj;
423 #define lmv_tgt_desc lu_tgt_desc
426 struct lu_client_fld lmv_fld;
428 struct lmv_desc desc;
433 u32 lmv_statfs_start;
435 struct lu_tgt_descs lmv_mdt_descs;
437 struct obd_connect_data conn_data;
438 struct kobject *lmv_tgts_kobj;
441 struct lu_qos lmv_qos;
442 __u32 lmv_qos_rr_index;
445 /* Minimum sector size is 512 */
446 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
448 struct niobuf_local {
449 __u64 lnb_file_offset;
450 __u32 lnb_page_offset;
454 struct page *lnb_page;
456 __u16 lnb_guards[MAX_GUARD_NUMBER];
457 __u16 lnb_guard_rpc:1;
458 __u16 lnb_guard_disk:1;
459 /* separate unlock for read path to allow shared access */
463 struct tgt_thread_big_cache {
464 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
467 #define LUSTRE_FLD_NAME "fld"
468 #define LUSTRE_SEQ_NAME "seq"
470 #define LUSTRE_MDD_NAME "mdd"
471 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
472 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
473 #define LUSTRE_VVP_NAME "vvp"
474 #define LUSTRE_LMV_NAME "lmv"
475 #define LUSTRE_SLP_NAME "slp"
476 #define LUSTRE_LOD_NAME "lod"
477 #define LUSTRE_OSP_NAME "osp"
478 #define LUSTRE_LWP_NAME "lwp"
480 /* obd device type names */
481 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
482 #define LUSTRE_MDS_NAME "mds"
483 #define LUSTRE_MDT_NAME "mdt"
484 #define LUSTRE_MDC_NAME "mdc"
485 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
486 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
487 #define LUSTRE_OSC_NAME "osc"
488 #define LUSTRE_LOV_NAME "lov"
489 #define LUSTRE_MGS_NAME "mgs"
490 #define LUSTRE_MGC_NAME "mgc"
492 #define LUSTRE_ECHO_NAME "obdecho"
493 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
494 #define LUSTRE_QMT_NAME "qmt"
496 /* Constant obd names (post-rename) */
497 #define LUSTRE_MDS_OBDNAME "MDS"
498 #define LUSTRE_OSS_OBDNAME "OSS"
499 #define LUSTRE_MGS_OBDNAME "MGS"
500 #define LUSTRE_MGC_OBDNAME "MGC"
502 static inline int is_lwp_on_mdt(char *name)
506 ptr = strrchr(name, '-');
508 CERROR("%s is not a obdname\n", name);
512 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
514 if (strncmp(ptr + 1, "MDT", 3) != 0)
517 while (*(--ptr) != '-' && ptr != name);
522 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
528 static inline int is_lwp_on_ost(char *name)
532 ptr = strrchr(name, '-');
534 CERROR("%s is not a obdname\n", name);
538 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
540 if (strncmp(ptr + 1, "OST", 3) != 0)
543 while (*(--ptr) != '-' && ptr != name);
548 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
555 * Events signalled through obd_notify() upcall-chain.
557 enum obd_notify_event {
558 /* Device connect start */
560 /* Device activated */
562 /* Device deactivated */
564 /* Connect data for import were changed */
566 /* Administratively deactivate/activate event */
567 OBD_NOTIFY_DEACTIVATE,
572 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
573 * being main example).
575 struct obd_notify_upcall {
576 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
577 enum obd_notify_event ev, void *owner);
578 /* Opaque datum supplied by upper layer listener */
582 struct target_recovery_data {
583 svc_handler_t trd_recovery_handler;
584 pid_t trd_processing_task;
585 struct completion trd_starting;
586 struct completion trd_finishing;
589 struct obd_llog_group {
590 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
591 wait_queue_head_t olg_waitq;
595 /* corresponds to one of the obd's */
596 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
599 struct obd_type *obd_type;
600 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
601 int obd_minor; /* device number: lctl dl */
602 struct lu_device *obd_lu_dev;
604 /* common and UUID name of this device */
605 struct obd_uuid obd_uuid;
606 char obd_name[MAX_OBD_NAME];
608 /* bitfield modification is protected by obd_dev_lock */
610 obd_attached:1, /* finished attach */
611 obd_set_up:1, /* finished setup */
612 obd_recovering:1, /* there are recoverable clients */
613 obd_abort_recovery:1, /* recovery expired */
614 obd_version_recov:1, /* obd uses version checking */
615 obd_replayable:1, /* recovery enabled; inform clients */
616 obd_no_transno:1, /* no committed-transno notification */
617 obd_no_recov:1, /* fail instead of retry messages */
618 obd_stopping:1, /* started cleanup */
619 obd_starting:1, /* started setup */
620 obd_force:1, /* cleanup with > 0 obd refcount */
621 obd_fail:1, /* cleanup with failover */
622 obd_no_conn:1, /* deny new connections */
623 obd_inactive:1, /* device active/inactive
624 * (for /proc/status only!!) */
625 obd_no_ir:1, /* no imperative recovery. */
626 obd_process_conf:1, /* device is processing mgs config */
627 obd_checksum_dump:1; /* dump pages upon cksum error */
629 /* use separate field as it is set in interrupt to don't mess with
630 * protection of other bits using _bh lock */
631 unsigned long obd_recovery_expired:1;
632 /* uuid-export hash body */
633 struct cfs_hash *obd_uuid_hash;
634 /* nid-export hash body */
635 struct cfs_hash *obd_nid_hash;
637 struct cfs_hash *obd_nid_stats_hash;
638 /* client_generation-export hash body */
639 struct cfs_hash *obd_gen_hash;
640 struct list_head obd_nid_stats;
641 struct list_head obd_exports;
642 struct list_head obd_unlinked_exports;
643 struct list_head obd_delayed_exports;
644 struct list_head obd_lwp_list;
645 atomic_t obd_refcount;
647 spinlock_t obd_nid_lock;
648 struct ldlm_namespace *obd_namespace;
649 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
650 /* a spinlock is OK for what we do now, may need a semaphore later */
651 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
652 spinlock_t obd_osfs_lock;
653 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
654 time64_t obd_osfs_age;
655 __u64 obd_last_committed;
656 struct mutex obd_dev_mutex;
657 struct lvfs_run_ctxt obd_lvfs_ctxt;
658 struct obd_llog_group obd_olg; /* default llog group */
659 struct obd_device *obd_observer;
660 struct rw_semaphore obd_observer_link_sem;
661 struct obd_notify_upcall obd_upcall;
662 struct obd_export *obd_self_export;
663 struct obd_export *obd_lwp_export;
664 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
665 struct list_head obd_exports_timed;
666 time64_t obd_eviction_timer; /* for ping evictor */
668 int obd_max_recoverable_clients;
669 atomic_t obd_connected_clients;
670 int obd_stale_clients;
671 /* this lock protects all recovery list_heads, timer and
672 * obd_next_recovery_transno value */
673 spinlock_t obd_recovery_task_lock;
674 __u64 obd_next_recovery_transno;
675 int obd_replayed_requests;
676 int obd_requests_queued_for_recovery;
677 wait_queue_head_t obd_next_transno_waitq;
678 /* protected by obd_recovery_task_lock */
679 struct hrtimer obd_recovery_timer;
681 time64_t obd_recovery_start;
682 /* seconds, for lprocfs_status */
683 time64_t obd_recovery_end;
684 /* To tell timeouts from time stamps Lustre uses time_t
685 * instead of time64_t.
687 time_t obd_recovery_time_hard;
688 time_t obd_recovery_timeout;
689 int obd_recovery_ir_factor;
691 /* new recovery stuff from CMD2 */
692 int obd_replayed_locks;
693 atomic_t obd_req_replay_clients;
694 atomic_t obd_lock_replay_clients;
695 struct target_recovery_data obd_recovery_data;
697 /* all lists are protected by obd_recovery_task_lock */
698 struct list_head obd_req_replay_queue;
699 struct list_head obd_lock_replay_queue;
700 struct list_head obd_final_req_queue;
703 #ifdef HAVE_SERVER_SUPPORT
704 struct obd_device_target obt;
705 struct filter_obd filter;
707 struct echo_obd echo;
709 struct client_obd cli;
710 struct echo_client_obd echo_client;
715 /* Fields used by LProcFS */
716 struct lprocfs_stats *obd_stats;
718 struct lprocfs_stats *obd_md_stats;
720 struct dentry *obd_debugfs_entry;
721 struct proc_dir_entry *obd_proc_entry;
722 struct proc_dir_entry *obd_proc_exports_entry;
723 struct dentry *obd_svc_debugfs_entry;
724 struct lprocfs_stats *obd_svc_stats;
725 const struct attribute **obd_attrs;
726 struct lprocfs_vars *obd_vars;
727 atomic_t obd_evict_inprogress;
728 wait_queue_head_t obd_evict_inprogress_waitq;
729 struct list_head obd_evict_list; /* protected with pet_lock */
732 * LDLM pool part. Save last calculated SLV and Limit.
734 rwlock_t obd_pool_lock;
738 int obd_conn_inprogress;
741 * List of outstanding class_incref()'s fo this OBD. For debugging. */
742 struct lu_ref obd_reference;
744 struct kset obd_kset; /* sysfs object collection */
745 struct kobj_type obd_ktype;
746 struct completion obd_kobj_unregister;
749 /* get/set_info keys */
750 #define KEY_ASYNC "async"
751 #define KEY_CHANGELOG_CLEAR "changelog_clear"
752 #define KEY_FID2PATH "fid2path"
753 #define KEY_CHECKSUM "checksum"
754 #define KEY_CLEAR_FS "clear_fs"
755 #define KEY_CONN_DATA "conn_data"
756 #define KEY_EVICT_BY_NID "evict_by_nid"
757 #define KEY_FIEMAP "fiemap"
758 #define KEY_FLUSH_CTX "flush_ctx"
759 #define KEY_GRANT_SHRINK "grant_shrink"
760 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
761 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
762 #define KEY_INTERMDS "inter_mds"
763 #define KEY_LAST_ID "last_id"
764 #define KEY_LAST_FID "last_fid"
765 #define KEY_MAX_EASIZE "max_easize"
766 #define KEY_DEFAULT_EASIZE "default_easize"
767 #define KEY_MGSSEC "mgssec"
768 #define KEY_READ_ONLY "read-only"
769 #define KEY_REGISTER_TARGET "register_target"
770 #define KEY_SET_FS "set_fs"
771 #define KEY_TGT_COUNT "tgt_count"
772 /* KEY_SET_INFO in lustre_idl.h */
773 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
775 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
776 #define KEY_OSP_CONNECTED "osp_connected"
778 /* Flags for op_xvalid */
780 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
781 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
782 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
783 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
784 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
785 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
786 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
791 static inline int it_to_lock_mode(struct lookup_intent *it)
793 /* CREAT needs to be tested before open (both could be set) */
794 if (it->it_op & IT_CREAT)
796 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
798 else if (it->it_op & IT_LAYOUT)
799 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
800 else if (it->it_op & IT_READDIR)
802 else if (it->it_op & IT_GETXATTR)
805 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
810 MF_MDC_CANCEL_FID1 = 1 << 0,
811 MF_MDC_CANCEL_FID2 = 1 << 1,
812 MF_MDC_CANCEL_FID3 = 1 << 2,
813 MF_MDC_CANCEL_FID4 = 1 << 3,
814 MF_GET_MDT_IDX = 1 << 4,
818 CLI_SET_MEA = 1 << 0,
819 CLI_RM_ENTRY = 1 << 1,
822 CLI_MIGRATE = 1 << 4,
826 LUSTRE_OPC_MKDIR = 0,
827 LUSTRE_OPC_SYMLINK = 1,
828 LUSTRE_OPC_MKNOD = 2,
829 LUSTRE_OPC_CREATE = 3,
834 * GETXATTR is not included as only a couple of fields in the reply body
835 * is filled, but not FID which is needed for common intent handling in
836 * mdc_finish_intent_lock()
838 static inline bool it_has_reply_body(const struct lookup_intent *it)
840 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
844 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
845 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
846 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
847 struct lu_fid op_fid4; /* to the operation locks. */
848 u32 op_mds; /* what mds server open will go to */
850 enum md_op_code op_code;
851 struct lustre_handle op_open_handle;
855 struct rw_semaphore *op_mea1_sem;
856 struct rw_semaphore *op_mea2_sem;
857 struct lmv_stripe_md *op_mea1;
858 struct lmv_stripe_md *op_mea2;
859 struct lmv_stripe_md *op_default_mea1; /* default LMV */
860 __u32 op_suppgids[2];
867 /* iattr fields and blocks. */
868 struct iattr op_attr;
869 enum op_xvalid op_xvalid; /* eXtra validity flags */
870 loff_t op_attr_blocks;
871 u64 op_valid; /* OBD_MD_* */
872 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
874 enum md_op_flags op_flags;
876 /* Various operation flags. */
877 enum mds_op_bias op_bias;
879 /* used to transfer info between the stacks of MD client
880 * see enum op_cli_flags */
881 enum md_cli_flags op_cli_flags;
883 /* File object data version for HSM release, on client */
884 __u64 op_data_version;
885 struct lustre_handle op_lease_handle;
887 /* File security context, for creates/metadata ops */
888 const char *op_file_secctx_name;
889 __u32 op_file_secctx_name_size;
890 void *op_file_secctx;
891 __u32 op_file_secctx_size;
895 /* Used by readdir */
896 unsigned int op_max_pages;
901 * used to access migrating dir: if it's set, assume migration is
902 * finished, use the new layout to access dir, otherwise use old layout.
903 * By default it's not set, because new files are created under new
904 * layout, if we can't find file with name under both old and new
905 * layout, we are sure file with name doesn't exist, but in reverse
906 * order there may be a race with creation by others.
908 bool op_post_migrate;
909 /* used to access dir with bash hash */
910 __u32 op_stripe_index;
911 /* Archive ID for PCC attach */
916 int (*md_blocking_ast)(struct ldlm_lock *lock,
917 struct ldlm_lock_desc *desc,
918 void *data, int flag);
921 struct md_enqueue_info;
922 /* metadata stat-ahead */
923 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
924 struct md_enqueue_info *minfo,
927 struct md_enqueue_info {
928 struct md_op_data mi_data;
929 struct lookup_intent mi_it;
930 struct lustre_handle mi_lockh;
931 struct inode *mi_dir;
932 struct ldlm_enqueue_info mi_einfo;
933 md_enqueue_cb_t mi_cb;
938 struct module *o_owner;
939 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
940 void *karg, void __user *uarg);
941 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
942 __u32 keylen, void *key, __u32 *vallen, void *val);
943 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
944 __u32 keylen, void *key,
945 __u32 vallen, void *val,
946 struct ptlrpc_request_set *set);
947 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
948 int (*o_precleanup)(struct obd_device *dev);
949 int (*o_cleanup)(struct obd_device *dev);
950 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
951 int (*o_postrecov)(struct obd_device *dev);
952 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
954 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
955 /* connect to the target device with given connection
956 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
957 * granted by the target, which are guaranteed to be a subset of flags
958 * asked for. If @ocd == NULL, use default parameters. */
959 int (*o_connect)(const struct lu_env *env,
960 struct obd_export **exp, struct obd_device *src,
961 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
963 int (*o_reconnect)(const struct lu_env *env,
964 struct obd_export *exp, struct obd_device *src,
965 struct obd_uuid *cluuid,
966 struct obd_connect_data *ocd,
968 int (*o_disconnect)(struct obd_export *exp);
970 /* Initialize/finalize fids infrastructure. */
971 int (*o_fid_init)(struct obd_device *obd,
972 struct obd_export *exp, enum lu_cli_type type);
973 int (*o_fid_fini)(struct obd_device *obd);
975 /* Allocate new fid according to passed @hint. */
976 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
977 struct lu_fid *fid, struct md_op_data *op_data);
980 * Object with @fid is getting deleted, we may want to do something
983 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
984 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
985 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
986 time64_t max_age, struct ptlrpc_request_set *set);
987 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
989 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
991 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
993 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
995 int (*o_preprw)(const struct lu_env *env, int cmd,
996 struct obd_export *exp, struct obdo *oa, int objcount,
997 struct obd_ioobj *obj, struct niobuf_remote *remote,
998 int *nr_pages, struct niobuf_local *local);
999 int (*o_commitrw)(const struct lu_env *env, int cmd,
1000 struct obd_export *exp, struct obdo *oa,
1001 int objcount, struct obd_ioobj *obj,
1002 struct niobuf_remote *remote, int pages,
1003 struct niobuf_local *local, int rc);
1004 int (*o_init_export)(struct obd_export *exp);
1005 int (*o_destroy_export)(struct obd_export *exp);
1007 int (*o_import_event)(struct obd_device *, struct obd_import *,
1008 enum obd_import_event);
1010 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1011 enum obd_notify_event ev);
1013 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1014 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1017 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1018 struct obd_quotactl *);
1021 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1022 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1023 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1025 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1029 /* lmv structures */
1031 struct mdt_body *body;
1032 struct lu_buf layout;
1034 struct lmv_stripe_md *lmv;
1035 struct lmv_foreign_md *lfm;
1037 struct lmv_stripe_md *default_lmv;
1038 #ifdef CONFIG_FS_POSIX_ACL
1039 struct posix_acl *posix_acl;
1043 struct md_open_data {
1044 struct obd_client_handle *mod_och;
1045 struct ptlrpc_request *mod_open_req;
1046 struct ptlrpc_request *mod_close_req;
1047 atomic_t mod_refcount;
1051 struct obd_client_handle {
1052 struct lustre_handle och_open_handle;
1053 struct lu_fid och_fid;
1054 struct md_open_data *och_mod;
1055 struct lustre_handle och_lease_handle; /* open lock for lease */
1060 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1062 struct lookup_intent;
1066 int (*m_close)(struct obd_export *, struct md_op_data *,
1067 struct md_open_data *, struct ptlrpc_request **);
1069 int (*m_create)(struct obd_export *, struct md_op_data *,
1070 const void *, size_t, umode_t, uid_t, gid_t,
1071 cfs_cap_t, __u64, struct ptlrpc_request **);
1073 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1074 const union ldlm_policy_data *, struct md_op_data *,
1075 struct lustre_handle *, __u64);
1077 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1078 struct ptlrpc_request **);
1080 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1081 struct lookup_intent *,
1082 struct ptlrpc_request **,
1083 ldlm_blocking_callback, __u64);
1085 int (*m_link)(struct obd_export *, struct md_op_data *,
1086 struct ptlrpc_request **);
1088 int (*m_rename)(struct obd_export *, struct md_op_data *,
1089 const char *, size_t, const char *, size_t,
1090 struct ptlrpc_request **);
1092 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1093 size_t , struct ptlrpc_request **);
1095 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1096 struct ptlrpc_request **);
1098 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1099 struct md_callback *cb_op, __u64 hash_offset,
1100 struct page **ppage);
1102 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1103 struct ptlrpc_request **);
1105 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1106 u64, const char *, const void *, size_t, unsigned int,
1107 u32, struct ptlrpc_request **);
1109 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1110 u64, const char *, size_t, struct ptlrpc_request **);
1112 int (*m_intent_getattr_async)(struct obd_export *,
1113 struct md_enqueue_info *);
1115 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1116 struct lu_fid *, __u64 *bits);
1118 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1120 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1121 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1123 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1124 struct ptlrpc_request **);
1126 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1128 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1129 struct obd_export *, struct obd_export *,
1130 struct lustre_md *);
1132 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1134 int (*m_merge_attr)(struct obd_export *,
1135 const struct lmv_stripe_md *lsm,
1136 struct cl_attr *attr, ldlm_blocking_callback);
1138 int (*m_set_open_replay_data)(struct obd_export *,
1139 struct obd_client_handle *,
1140 struct lookup_intent *);
1142 int (*m_clear_open_replay_data)(struct obd_export *,
1143 struct obd_client_handle *);
1145 int (*m_set_lock_data)(struct obd_export *,
1146 const struct lustre_handle *, void *, __u64 *);
1148 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1149 const struct lu_fid *, enum ldlm_type,
1150 union ldlm_policy_data *, enum ldlm_mode,
1151 struct lustre_handle *);
1153 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1154 union ldlm_policy_data *, enum ldlm_mode,
1155 enum ldlm_cancel_flags flags, void *opaque);
1157 int (*m_get_fid_from_lsm)(struct obd_export *,
1158 const struct lmv_stripe_md *,
1159 const char *name, int namelen,
1160 struct lu_fid *fid);
1161 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1162 const union lmv_mds_md *lmv, size_t lmv_size);
1163 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1164 struct ptlrpc_request_set *set);
1167 static inline struct md_open_data *obd_mod_alloc(void)
1169 struct md_open_data *mod;
1173 atomic_set(&mod->mod_refcount, 1);
1177 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1178 #define obd_mod_put(mod) \
1180 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1181 if ((mod)->mod_open_req) \
1182 ptlrpc_req_finished((mod)->mod_open_req); \
1183 OBD_FREE_PTR(mod); \
1187 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1188 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1189 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1191 /* return 1 if client should be resend request */
1192 static inline int client_should_resend(int resend, struct client_obd *cli)
1194 return atomic_read(&cli->cl_resends) ?
1195 atomic_read(&cli->cl_resends) > resend : 1;
1199 * Return device name for this device
1201 * XXX: lu_device is declared before obd_device, while a pointer pointing
1202 * back to obd_device in lu_device, so this helper function defines here
1203 * instead of in lu_object.h
1205 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1207 return lu_dev->ld_obd->obd_name;
1210 static inline bool filename_is_volatile(const char *name, size_t namelen,
1216 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1219 /* caller does not care of idx */
1223 /* volatile file, the MDT can be set from name */
1224 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1225 /* if no MDT is specified, use std way */
1226 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1228 /* test for no MDT idx case */
1229 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1230 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1234 /* we have an idx, read it */
1235 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1236 *idx = simple_strtoul(start, &end, 16);
1238 * no digit, no trailing :, negative value
1240 if (((*idx == 0) && (end == start)) ||
1241 (*end != ':') || (*idx < 0))
1246 /* bad format of mdt idx, we cannot return an error
1247 * to caller so we use hash algo */
1248 CERROR("Bad volatile file name format: %s\n",
1249 name + LUSTRE_VOLATILE_HDR_LEN);
1253 static inline int cli_brw_size(struct obd_device *obd)
1255 LASSERT(obd != NULL);
1256 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1260 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1261 * of the client should be increased accordingly to avoid sending fragmented
1262 * RPCs over the network when the client runs out of the maximum dirty space
1263 * when so many RPCs are being generated.
1265 static inline void client_adjust_max_dirty(struct client_obd *cli)
1268 if (cli->cl_dirty_max_pages <= 0) {
1269 cli->cl_dirty_max_pages =
1270 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1272 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1273 cli->cl_max_pages_per_rpc;
1275 if (dirty_max > cli->cl_dirty_max_pages)
1276 cli->cl_dirty_max_pages = dirty_max;
1279 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1280 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1282 /* This value is exported to userspace through the max_dirty_mb
1283 * parameter. So we round up the number of pages to make it a round
1285 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1286 1 << (20 - PAGE_SHIFT));
1289 #endif /* __OBD_H */