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;
430 struct mutex lmv_init_mutex;
434 u32 lmv_statfs_start;
436 u32 tgts_size; /* size of tgts array */
437 struct lmv_tgt_desc **tgts;
439 struct obd_connect_data conn_data;
440 struct kobject *lmv_tgts_kobj;
443 struct lu_qos lmv_qos;
444 __u32 lmv_qos_rr_index;
447 /* Minimum sector size is 512 */
448 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
450 struct niobuf_local {
451 __u64 lnb_file_offset;
452 __u32 lnb_page_offset;
456 struct page *lnb_page;
458 __u16 lnb_guards[MAX_GUARD_NUMBER];
459 __u16 lnb_guard_rpc:1;
460 __u16 lnb_guard_disk:1;
461 /* separate unlock for read path to allow shared access */
465 struct tgt_thread_big_cache {
466 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
469 #define LUSTRE_FLD_NAME "fld"
470 #define LUSTRE_SEQ_NAME "seq"
472 #define LUSTRE_MDD_NAME "mdd"
473 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
474 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
475 #define LUSTRE_VVP_NAME "vvp"
476 #define LUSTRE_LMV_NAME "lmv"
477 #define LUSTRE_SLP_NAME "slp"
478 #define LUSTRE_LOD_NAME "lod"
479 #define LUSTRE_OSP_NAME "osp"
480 #define LUSTRE_LWP_NAME "lwp"
482 /* obd device type names */
483 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
484 #define LUSTRE_MDS_NAME "mds"
485 #define LUSTRE_MDT_NAME "mdt"
486 #define LUSTRE_MDC_NAME "mdc"
487 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
488 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
489 #define LUSTRE_OSC_NAME "osc"
490 #define LUSTRE_LOV_NAME "lov"
491 #define LUSTRE_MGS_NAME "mgs"
492 #define LUSTRE_MGC_NAME "mgc"
494 #define LUSTRE_ECHO_NAME "obdecho"
495 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
496 #define LUSTRE_QMT_NAME "qmt"
498 /* Constant obd names (post-rename) */
499 #define LUSTRE_MDS_OBDNAME "MDS"
500 #define LUSTRE_OSS_OBDNAME "OSS"
501 #define LUSTRE_MGS_OBDNAME "MGS"
502 #define LUSTRE_MGC_OBDNAME "MGC"
504 static inline int is_lwp_on_mdt(char *name)
508 ptr = strrchr(name, '-');
510 CERROR("%s is not a obdname\n", name);
514 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
516 if (strncmp(ptr + 1, "MDT", 3) != 0)
519 while (*(--ptr) != '-' && ptr != name);
524 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
530 static inline int is_lwp_on_ost(char *name)
534 ptr = strrchr(name, '-');
536 CERROR("%s is not a obdname\n", name);
540 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
542 if (strncmp(ptr + 1, "OST", 3) != 0)
545 while (*(--ptr) != '-' && ptr != name);
550 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
557 * Events signalled through obd_notify() upcall-chain.
559 enum obd_notify_event {
560 /* Device connect start */
562 /* Device activated */
564 /* Device deactivated */
566 /* Connect data for import were changed */
568 /* Administratively deactivate/activate event */
569 OBD_NOTIFY_DEACTIVATE,
574 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
575 * being main example).
577 struct obd_notify_upcall {
578 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
579 enum obd_notify_event ev, void *owner);
580 /* Opaque datum supplied by upper layer listener */
584 struct target_recovery_data {
585 svc_handler_t trd_recovery_handler;
586 pid_t trd_processing_task;
587 struct completion trd_starting;
588 struct completion trd_finishing;
591 struct obd_llog_group {
592 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
593 wait_queue_head_t olg_waitq;
595 struct mutex olg_cat_processing;
598 /* corresponds to one of the obd's */
599 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
602 struct obd_type *obd_type;
603 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
604 int obd_minor; /* device number: lctl dl */
605 struct lu_device *obd_lu_dev;
607 /* common and UUID name of this device */
608 struct obd_uuid obd_uuid;
609 char obd_name[MAX_OBD_NAME];
611 /* bitfield modification is protected by obd_dev_lock */
613 obd_attached:1, /* finished attach */
614 obd_set_up:1, /* finished setup */
615 obd_recovering:1, /* there are recoverable clients */
616 obd_abort_recovery:1, /* recovery expired */
617 obd_version_recov:1, /* obd uses version checking */
618 obd_replayable:1, /* recovery enabled; inform clients */
619 obd_no_transno:1, /* no committed-transno notification */
620 obd_no_recov:1, /* fail instead of retry messages */
621 obd_stopping:1, /* started cleanup */
622 obd_starting:1, /* started setup */
623 obd_force:1, /* cleanup with > 0 obd refcount */
624 obd_fail:1, /* cleanup with failover */
625 obd_no_conn:1, /* deny new connections */
626 obd_inactive:1, /* device active/inactive
627 * (for /proc/status only!!) */
628 obd_no_ir:1, /* no imperative recovery. */
629 obd_process_conf:1, /* device is processing mgs config */
630 obd_checksum_dump:1; /* dump pages upon cksum error */
632 /* use separate field as it is set in interrupt to don't mess with
633 * protection of other bits using _bh lock */
634 unsigned long obd_recovery_expired:1;
635 /* uuid-export hash body */
636 struct cfs_hash *obd_uuid_hash;
637 /* nid-export hash body */
638 struct cfs_hash *obd_nid_hash;
640 struct cfs_hash *obd_nid_stats_hash;
641 /* client_generation-export hash body */
642 struct cfs_hash *obd_gen_hash;
643 struct list_head obd_nid_stats;
644 struct list_head obd_exports;
645 struct list_head obd_unlinked_exports;
646 struct list_head obd_delayed_exports;
647 struct list_head obd_lwp_list;
648 atomic_t obd_refcount;
650 spinlock_t obd_nid_lock;
651 struct ldlm_namespace *obd_namespace;
652 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
653 /* a spinlock is OK for what we do now, may need a semaphore later */
654 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
655 spinlock_t obd_osfs_lock;
656 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
657 time64_t obd_osfs_age;
658 __u64 obd_last_committed;
659 struct mutex obd_dev_mutex;
660 struct lvfs_run_ctxt obd_lvfs_ctxt;
661 struct obd_llog_group obd_olg; /* default llog group */
662 struct obd_device *obd_observer;
663 struct rw_semaphore obd_observer_link_sem;
664 struct obd_notify_upcall obd_upcall;
665 struct obd_export *obd_self_export;
666 struct obd_export *obd_lwp_export;
667 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
668 struct list_head obd_exports_timed;
669 time64_t obd_eviction_timer; /* for ping evictor */
671 int obd_max_recoverable_clients;
672 atomic_t obd_connected_clients;
673 int obd_stale_clients;
674 /* this lock protects all recovery list_heads, timer and
675 * obd_next_recovery_transno value */
676 spinlock_t obd_recovery_task_lock;
677 __u64 obd_next_recovery_transno;
678 int obd_replayed_requests;
679 int obd_requests_queued_for_recovery;
680 wait_queue_head_t obd_next_transno_waitq;
681 /* protected by obd_recovery_task_lock */
682 struct hrtimer obd_recovery_timer;
684 time64_t obd_recovery_start;
685 /* seconds, for lprocfs_status */
686 time64_t obd_recovery_end;
687 /* To tell timeouts from time stamps Lustre uses time_t
688 * instead of time64_t.
690 time_t obd_recovery_time_hard;
691 time_t obd_recovery_timeout;
692 int obd_recovery_ir_factor;
694 /* new recovery stuff from CMD2 */
695 int obd_replayed_locks;
696 atomic_t obd_req_replay_clients;
697 atomic_t obd_lock_replay_clients;
698 struct target_recovery_data obd_recovery_data;
700 /* all lists are protected by obd_recovery_task_lock */
701 struct list_head obd_req_replay_queue;
702 struct list_head obd_lock_replay_queue;
703 struct list_head obd_final_req_queue;
706 #ifdef HAVE_SERVER_SUPPORT
707 struct obd_device_target obt;
708 struct filter_obd filter;
710 struct echo_obd echo;
712 struct client_obd cli;
713 struct echo_client_obd echo_client;
718 /* Fields used by LProcFS */
719 struct lprocfs_stats *obd_stats;
721 struct lprocfs_stats *obd_md_stats;
723 struct dentry *obd_debugfs_entry;
724 struct proc_dir_entry *obd_proc_entry;
725 struct proc_dir_entry *obd_proc_exports_entry;
726 struct dentry *obd_svc_debugfs_entry;
727 struct lprocfs_stats *obd_svc_stats;
728 const struct attribute **obd_attrs;
729 struct lprocfs_vars *obd_vars;
730 atomic_t obd_evict_inprogress;
731 wait_queue_head_t obd_evict_inprogress_waitq;
732 struct list_head obd_evict_list; /* protected with pet_lock */
735 * LDLM pool part. Save last calculated SLV and Limit.
737 rwlock_t obd_pool_lock;
741 int obd_conn_inprogress;
744 * List of outstanding class_incref()'s fo this OBD. For debugging. */
745 struct lu_ref obd_reference;
747 struct kset obd_kset; /* sysfs object collection */
748 struct kobj_type obd_ktype;
749 struct completion obd_kobj_unregister;
752 /* get/set_info keys */
753 #define KEY_ASYNC "async"
754 #define KEY_CHANGELOG_CLEAR "changelog_clear"
755 #define KEY_FID2PATH "fid2path"
756 #define KEY_CHECKSUM "checksum"
757 #define KEY_CLEAR_FS "clear_fs"
758 #define KEY_CONN_DATA "conn_data"
759 #define KEY_EVICT_BY_NID "evict_by_nid"
760 #define KEY_FIEMAP "fiemap"
761 #define KEY_FLUSH_CTX "flush_ctx"
762 #define KEY_GRANT_SHRINK "grant_shrink"
763 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
764 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
765 #define KEY_INTERMDS "inter_mds"
766 #define KEY_LAST_ID "last_id"
767 #define KEY_LAST_FID "last_fid"
768 #define KEY_MAX_EASIZE "max_easize"
769 #define KEY_DEFAULT_EASIZE "default_easize"
770 #define KEY_MGSSEC "mgssec"
771 #define KEY_READ_ONLY "read-only"
772 #define KEY_REGISTER_TARGET "register_target"
773 #define KEY_SET_FS "set_fs"
774 #define KEY_TGT_COUNT "tgt_count"
775 /* KEY_SET_INFO in lustre_idl.h */
776 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
778 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
779 #define KEY_OSP_CONNECTED "osp_connected"
781 /* Flags for op_xvalid */
783 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
784 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
785 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
786 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
787 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
788 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
789 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
794 static inline int it_to_lock_mode(struct lookup_intent *it)
796 /* CREAT needs to be tested before open (both could be set) */
797 if (it->it_op & IT_CREAT)
799 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
801 else if (it->it_op & IT_LAYOUT)
802 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
803 else if (it->it_op & IT_READDIR)
805 else if (it->it_op & IT_GETXATTR)
808 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
813 MF_MDC_CANCEL_FID1 = 1 << 0,
814 MF_MDC_CANCEL_FID2 = 1 << 1,
815 MF_MDC_CANCEL_FID3 = 1 << 2,
816 MF_MDC_CANCEL_FID4 = 1 << 3,
817 MF_GET_MDT_IDX = 1 << 4,
821 CLI_SET_MEA = 1 << 0,
822 CLI_RM_ENTRY = 1 << 1,
825 CLI_MIGRATE = 1 << 4,
829 LUSTRE_OPC_MKDIR = 0,
830 LUSTRE_OPC_SYMLINK = 1,
831 LUSTRE_OPC_MKNOD = 2,
832 LUSTRE_OPC_CREATE = 3,
837 * GETXATTR is not included as only a couple of fields in the reply body
838 * is filled, but not FID which is needed for common intent handling in
839 * mdc_finish_intent_lock()
841 static inline bool it_has_reply_body(const struct lookup_intent *it)
843 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
847 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
848 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
849 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
850 struct lu_fid op_fid4; /* to the operation locks. */
851 u32 op_mds; /* what mds server open will go to */
853 enum md_op_code op_code;
854 struct lustre_handle op_open_handle;
858 struct rw_semaphore *op_mea1_sem;
859 struct rw_semaphore *op_mea2_sem;
860 struct lmv_stripe_md *op_mea1;
861 struct lmv_stripe_md *op_mea2;
862 struct lmv_stripe_md *op_default_mea1; /* default LMV */
863 __u32 op_suppgids[2];
870 /* iattr fields and blocks. */
871 struct iattr op_attr;
872 enum op_xvalid op_xvalid; /* eXtra validity flags */
873 loff_t op_attr_blocks;
874 u64 op_valid; /* OBD_MD_* */
875 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
877 enum md_op_flags op_flags;
879 /* Various operation flags. */
880 enum mds_op_bias op_bias;
882 /* used to transfer info between the stacks of MD client
883 * see enum op_cli_flags */
884 enum md_cli_flags op_cli_flags;
886 /* File object data version for HSM release, on client */
887 __u64 op_data_version;
888 struct lustre_handle op_lease_handle;
890 /* File security context, for creates/metadata ops */
891 const char *op_file_secctx_name;
892 __u32 op_file_secctx_name_size;
893 void *op_file_secctx;
894 __u32 op_file_secctx_size;
898 /* Used by readdir */
899 unsigned int op_max_pages;
904 * used to access migrating dir: if it's set, assume migration is
905 * finished, use the new layout to access dir, otherwise use old layout.
906 * By default it's not set, because new files are created under new
907 * layout, if we can't find file with name under both old and new
908 * layout, we are sure file with name doesn't exist, but in reverse
909 * order there may be a race with creation by others.
911 bool op_post_migrate;
912 /* used to access dir with bash hash */
913 __u32 op_stripe_index;
914 /* Archive ID for PCC attach */
919 int (*md_blocking_ast)(struct ldlm_lock *lock,
920 struct ldlm_lock_desc *desc,
921 void *data, int flag);
924 struct md_enqueue_info;
925 /* metadata stat-ahead */
926 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
927 struct md_enqueue_info *minfo,
930 struct md_enqueue_info {
931 struct md_op_data mi_data;
932 struct lookup_intent mi_it;
933 struct lustre_handle mi_lockh;
934 struct inode *mi_dir;
935 struct ldlm_enqueue_info mi_einfo;
936 md_enqueue_cb_t mi_cb;
941 struct module *o_owner;
942 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
943 void *karg, void __user *uarg);
944 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
945 __u32 keylen, void *key, __u32 *vallen, void *val);
946 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
947 __u32 keylen, void *key,
948 __u32 vallen, void *val,
949 struct ptlrpc_request_set *set);
950 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
951 int (*o_precleanup)(struct obd_device *dev);
952 int (*o_cleanup)(struct obd_device *dev);
953 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
954 int (*o_postrecov)(struct obd_device *dev);
955 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
957 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
958 /* connect to the target device with given connection
959 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
960 * granted by the target, which are guaranteed to be a subset of flags
961 * asked for. If @ocd == NULL, use default parameters. */
962 int (*o_connect)(const struct lu_env *env,
963 struct obd_export **exp, struct obd_device *src,
964 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
966 int (*o_reconnect)(const struct lu_env *env,
967 struct obd_export *exp, struct obd_device *src,
968 struct obd_uuid *cluuid,
969 struct obd_connect_data *ocd,
971 int (*o_disconnect)(struct obd_export *exp);
973 /* Initialize/finalize fids infrastructure. */
974 int (*o_fid_init)(struct obd_device *obd,
975 struct obd_export *exp, enum lu_cli_type type);
976 int (*o_fid_fini)(struct obd_device *obd);
978 /* Allocate new fid according to passed @hint. */
979 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
980 struct lu_fid *fid, struct md_op_data *op_data);
983 * Object with @fid is getting deleted, we may want to do something
986 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
987 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
988 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
989 time64_t max_age, struct ptlrpc_request_set *set);
990 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
992 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
994 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
996 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
998 int (*o_preprw)(const struct lu_env *env, int cmd,
999 struct obd_export *exp, struct obdo *oa, int objcount,
1000 struct obd_ioobj *obj, struct niobuf_remote *remote,
1001 int *nr_pages, struct niobuf_local *local);
1002 int (*o_commitrw)(const struct lu_env *env, int cmd,
1003 struct obd_export *exp, struct obdo *oa,
1004 int objcount, struct obd_ioobj *obj,
1005 struct niobuf_remote *remote, int pages,
1006 struct niobuf_local *local, int rc);
1007 int (*o_init_export)(struct obd_export *exp);
1008 int (*o_destroy_export)(struct obd_export *exp);
1010 int (*o_import_event)(struct obd_device *, struct obd_import *,
1011 enum obd_import_event);
1013 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1014 enum obd_notify_event ev);
1016 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1017 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1020 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1021 struct obd_quotactl *);
1024 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1025 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1026 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1028 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1032 /* lmv structures */
1034 struct mdt_body *body;
1035 struct lu_buf layout;
1037 struct lmv_stripe_md *lmv;
1038 struct lmv_foreign_md *lfm;
1040 struct lmv_stripe_md *default_lmv;
1041 #ifdef CONFIG_FS_POSIX_ACL
1042 struct posix_acl *posix_acl;
1046 struct md_open_data {
1047 struct obd_client_handle *mod_och;
1048 struct ptlrpc_request *mod_open_req;
1049 struct ptlrpc_request *mod_close_req;
1050 atomic_t mod_refcount;
1054 struct obd_client_handle {
1055 struct lustre_handle och_open_handle;
1056 struct lu_fid och_fid;
1057 struct md_open_data *och_mod;
1058 struct lustre_handle och_lease_handle; /* open lock for lease */
1063 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1065 struct lookup_intent;
1069 int (*m_close)(struct obd_export *, struct md_op_data *,
1070 struct md_open_data *, struct ptlrpc_request **);
1072 int (*m_create)(struct obd_export *, struct md_op_data *,
1073 const void *, size_t, umode_t, uid_t, gid_t,
1074 cfs_cap_t, __u64, struct ptlrpc_request **);
1076 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1077 const union ldlm_policy_data *, struct md_op_data *,
1078 struct lustre_handle *, __u64);
1080 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1081 struct ptlrpc_request **);
1083 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1084 struct lookup_intent *,
1085 struct ptlrpc_request **,
1086 ldlm_blocking_callback, __u64);
1088 int (*m_link)(struct obd_export *, struct md_op_data *,
1089 struct ptlrpc_request **);
1091 int (*m_rename)(struct obd_export *, struct md_op_data *,
1092 const char *, size_t, const char *, size_t,
1093 struct ptlrpc_request **);
1095 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1096 size_t , struct ptlrpc_request **);
1098 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1099 struct ptlrpc_request **);
1101 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1102 struct md_callback *cb_op, __u64 hash_offset,
1103 struct page **ppage);
1105 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1106 struct ptlrpc_request **);
1108 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1109 u64, const char *, const void *, size_t, unsigned int,
1110 u32, struct ptlrpc_request **);
1112 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1113 u64, const char *, size_t, struct ptlrpc_request **);
1115 int (*m_intent_getattr_async)(struct obd_export *,
1116 struct md_enqueue_info *);
1118 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1119 struct lu_fid *, __u64 *bits);
1121 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1123 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1124 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1126 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1127 struct ptlrpc_request **);
1129 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1131 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1132 struct obd_export *, struct obd_export *,
1133 struct lustre_md *);
1135 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1137 int (*m_merge_attr)(struct obd_export *,
1138 const struct lmv_stripe_md *lsm,
1139 struct cl_attr *attr, ldlm_blocking_callback);
1141 int (*m_set_open_replay_data)(struct obd_export *,
1142 struct obd_client_handle *,
1143 struct lookup_intent *);
1145 int (*m_clear_open_replay_data)(struct obd_export *,
1146 struct obd_client_handle *);
1148 int (*m_set_lock_data)(struct obd_export *,
1149 const struct lustre_handle *, void *, __u64 *);
1151 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1152 const struct lu_fid *, enum ldlm_type,
1153 union ldlm_policy_data *, enum ldlm_mode,
1154 struct lustre_handle *);
1156 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1157 union ldlm_policy_data *, enum ldlm_mode,
1158 enum ldlm_cancel_flags flags, void *opaque);
1160 int (*m_get_fid_from_lsm)(struct obd_export *,
1161 const struct lmv_stripe_md *,
1162 const char *name, int namelen,
1163 struct lu_fid *fid);
1164 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1165 const union lmv_mds_md *lmv, size_t lmv_size);
1166 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1167 struct ptlrpc_request_set *set);
1170 static inline struct md_open_data *obd_mod_alloc(void)
1172 struct md_open_data *mod;
1176 atomic_set(&mod->mod_refcount, 1);
1180 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1181 #define obd_mod_put(mod) \
1183 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1184 if ((mod)->mod_open_req) \
1185 ptlrpc_req_finished((mod)->mod_open_req); \
1186 OBD_FREE_PTR(mod); \
1190 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1191 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1192 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1194 /* return 1 if client should be resend request */
1195 static inline int client_should_resend(int resend, struct client_obd *cli)
1197 return atomic_read(&cli->cl_resends) ?
1198 atomic_read(&cli->cl_resends) > resend : 1;
1202 * Return device name for this device
1204 * XXX: lu_device is declared before obd_device, while a pointer pointing
1205 * back to obd_device in lu_device, so this helper function defines here
1206 * instead of in lu_object.h
1208 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1210 return lu_dev->ld_obd->obd_name;
1213 static inline bool filename_is_volatile(const char *name, size_t namelen,
1219 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1222 /* caller does not care of idx */
1226 /* volatile file, the MDT can be set from name */
1227 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1228 /* if no MDT is specified, use std way */
1229 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1231 /* test for no MDT idx case */
1232 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1233 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1237 /* we have an idx, read it */
1238 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1239 *idx = simple_strtoul(start, &end, 16);
1241 * no digit, no trailing :, negative value
1243 if (((*idx == 0) && (end == start)) ||
1244 (*end != ':') || (*idx < 0))
1249 /* bad format of mdt idx, we cannot return an error
1250 * to caller so we use hash algo */
1251 CERROR("Bad volatile file name format: %s\n",
1252 name + LUSTRE_VOLATILE_HDR_LEN);
1256 static inline int cli_brw_size(struct obd_device *obd)
1258 LASSERT(obd != NULL);
1259 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1263 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1264 * of the client should be increased accordingly to avoid sending fragmented
1265 * RPCs over the network when the client runs out of the maximum dirty space
1266 * when so many RPCs are being generated.
1268 static inline void client_adjust_max_dirty(struct client_obd *cli)
1271 if (cli->cl_dirty_max_pages <= 0) {
1272 cli->cl_dirty_max_pages =
1273 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1275 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1276 cli->cl_max_pages_per_rpc;
1278 if (dirty_max > cli->cl_dirty_max_pages)
1279 cli->cl_dirty_max_pages = dirty_max;
1282 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1283 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1285 /* This value is exported to userspace through the max_dirty_mb
1286 * parameter. So we round up the number of pages to make it a round
1288 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1289 1 << (20 - PAGE_SHIFT));
1292 #endif /* __OBD_H */