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)
122 /* used for encryption: difference with offset in clear text page */
124 /* used for encryption: difference with count in clear text page */
129 struct timeout_item {
130 enum timeout_event ti_event;
134 struct list_head ti_obd_list;
135 struct list_head ti_chain;
138 #define OBD_MAX_RIF_DEFAULT 8
139 #define OBD_MAX_RIF_MAX 512
140 #define OSC_MAX_RIF_MAX 256
141 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
142 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
143 #define OSC_DEFAULT_RESENDS 10
145 /* possible values for lut_sync_lock_cancel */
146 enum tgt_sync_lock_cancel {
147 SYNC_LOCK_CANCEL_NEVER = 0,
148 SYNC_LOCK_CANCEL_BLOCKING = 1,
149 SYNC_LOCK_CANCEL_ALWAYS = 2,
153 * Limit reply buffer size for striping data to one x86_64 page. This
154 * value is chosen to fit the striping data for common use cases while
155 * staying well below the limit at which the buffer must be backed by
156 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
159 #define OBD_MAX_DEFAULT_EA_SIZE 4096
162 * Lustre can handle larger xattrs internally, but we must respect the Linux
163 * VFS limitation or tools like tar cannot interact with Lustre volumes
166 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
169 enum obd_cl_sem_lock_class {
178 struct rw_semaphore cl_sem;
179 struct obd_uuid cl_target_uuid;
180 struct obd_import *cl_import; /* ptlrpc connection state */
181 size_t cl_conn_count;
183 /* Cache maximum and default values for easize. This is
184 * strictly a performance optimization to minimize calls to
185 * obd_size_diskmd(). The default values are used to calculate the
186 * initial size of a request buffer. The ptlrpc layer will resize the
187 * buffer as needed to accommodate a larger reply from the
188 * server. The default values should be small enough to avoid wasted
189 * memory and excessive use of vmalloc(), yet large enough to avoid
190 * reallocating the buffer in the common use case. */
192 /* Default EA size for striping attributes. It is initialized at
193 * mount-time based on the default stripe width of the filesystem,
194 * then it tracks the largest observed EA size advertised by
195 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
196 __u32 cl_default_mds_easize;
198 /* Maximum possible EA size computed at mount-time based on
199 * the number of OSTs in the filesystem. May be increased at
200 * run-time if a larger observed size is advertised by the MDT. */
201 __u32 cl_max_mds_easize;
203 /* Data-on-MDT specific value to set larger reply buffer for possible
204 * data read along with open/stat requests. By default it tries to use
205 * unused space in reply buffer.
206 * This value is used to ensure that reply buffer has at least as
207 * much free space as value indicates. That free space is gained from
208 * LOV EA buffer which is small for DoM files and on big systems can
209 * provide up to 32KB of extra space in reply buffer.
210 * Default value is 8K now.
212 __u32 cl_dom_min_inline_repsize;
214 enum lustre_sec_part cl_sp_me;
215 enum lustre_sec_part cl_sp_to;
216 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
218 /* the grant values are protected by loi_list_lock below */
219 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
220 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
221 unsigned long cl_avail_grant; /* bytes of credit for ost */
222 unsigned long cl_lost_grant; /* lost credits (trunc) */
223 /* grant consumed for dirty pages */
224 unsigned long cl_dirty_grant;
226 /* since we allocate grant by blocks, we don't know how many grant will
227 * be used to add a page into cache. As a solution, we reserve maximum
228 * grant before trying to dirty a page and unreserve the rest.
229 * See osc_{reserve|unreserve}_grant for details. */
230 long cl_reserved_grant;
231 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
232 time64_t cl_next_shrink_grant; /* seconds */
233 struct list_head cl_grant_chain;
234 time64_t cl_grant_shrink_interval; /* seconds */
236 /* A chunk is an optimal size used by osc_extent to determine
237 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
239 /* extent insertion metadata overhead to be accounted in grant,
241 unsigned int cl_grant_extent_tax;
242 /* maximum extent size, in number of pages */
243 unsigned int cl_max_extent_pages;
245 /* keep track of objects that have lois that contain pages which
246 * have been queued for async brw. this lock also protects the
247 * lists of osc_client_pages that hang off of the loi */
249 * ->cl_loi_list_lock protects consistency of
250 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
251 * ->ap_completion() call-backs are executed under this lock. As we
252 * cannot guarantee that these call-backs never block on all platforms
253 * (as a matter of fact they do block on Mac OS X), type of
254 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
255 * and blocking mutex on Mac OS X. (Alternative is to make this lock
256 * blocking everywhere, but we don't want to slow down fast-path of
257 * our main platform.)
259 * NB by Jinshan: though field names are still _loi_, but actually
260 * osc_object{}s are in the list.
262 spinlock_t cl_loi_list_lock;
263 struct list_head cl_loi_ready_list;
264 struct list_head cl_loi_hp_ready_list;
265 struct list_head cl_loi_write_list;
266 struct list_head cl_loi_read_list;
267 __u32 cl_r_in_flight;
268 __u32 cl_w_in_flight;
269 /* just a sum of the loi/lop pending numbers to be exported by /proc */
270 atomic_t cl_pending_w_pages;
271 atomic_t cl_pending_r_pages;
272 u32 cl_max_pages_per_rpc;
273 u32 cl_max_rpcs_in_flight;
274 u32 cl_max_short_io_bytes;
275 struct obd_histogram cl_read_rpc_hist;
276 struct obd_histogram cl_write_rpc_hist;
277 struct obd_histogram cl_read_page_hist;
278 struct obd_histogram cl_write_page_hist;
279 struct obd_histogram cl_read_offset_hist;
280 struct obd_histogram cl_write_offset_hist;
282 /** LRU for osc caching pages */
283 struct cl_client_cache *cl_cache;
284 /** member of cl_cache->ccc_lru */
285 struct list_head cl_lru_osc;
286 /** # of available LRU slots left in the per-OSC cache.
287 * Available LRU slots are shared by all OSCs of the same file system,
288 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
289 atomic_long_t *cl_lru_left;
290 /** # of busy LRU pages. A page is considered busy if it's in writeback
291 * queue, or in transfer. Busy pages can't be discarded so they are not
293 atomic_long_t cl_lru_busy;
294 /** # of LRU pages in the cache for this client_obd */
295 atomic_long_t cl_lru_in_list;
296 /** # of threads are shrinking LRU cache. To avoid contention, it's not
297 * allowed to have multiple threads shrinking LRU cache. */
298 atomic_t cl_lru_shrinkers;
299 /** The time when this LRU cache was last used. */
300 time64_t cl_lru_last_used;
301 /** stats: how many reclaims have happened for this client_obd.
302 * reclaim and shrink - shrink is async, voluntarily rebalancing;
303 * reclaim is sync, initiated by IO thread when the LRU slots are
305 __u64 cl_lru_reclaim;
306 /** List of LRU pages for this client_obd */
307 struct list_head cl_lru_list;
308 /** Lock for LRU page list */
309 spinlock_t cl_lru_list_lock;
310 /** # of unstable pages in this client_obd.
311 * An unstable page is a page state that WRITE RPC has finished but
312 * the transaction has NOT yet committed. */
313 atomic_long_t cl_unstable_count;
314 /** Link to osc_shrinker_list */
315 struct list_head cl_shrink_list;
317 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
318 atomic_t cl_destroy_in_flight;
319 wait_queue_head_t cl_destroy_waitq;
321 /* modify rpcs in flight
322 * currently used for metadata only */
323 spinlock_t cl_mod_rpcs_lock;
324 __u16 cl_max_mod_rpcs_in_flight;
325 __u16 cl_mod_rpcs_in_flight;
326 __u16 cl_close_rpcs_in_flight;
327 wait_queue_head_t cl_mod_rpcs_waitq;
328 unsigned long *cl_mod_tag_bitmap;
329 struct obd_histogram cl_mod_rpcs_hist;
332 struct mutex cl_mgc_mutex;
333 struct local_oid_storage *cl_mgc_los;
334 struct dt_object *cl_mgc_configs_dir;
335 struct obd_export *cl_mgc_mgsexp;
336 atomic_t cl_mgc_refcount;
337 /* in-flight control list and total RPCs counter */
338 struct list_head cl_flight_waiters;
339 __u32 cl_rpcs_in_flight;
341 /* checksumming for data sent over the network */
342 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
343 cl_checksum_dump:1; /* same */
344 /* supported checksum types that are worked out at connect time */
345 __u32 cl_supp_cksum_types;
346 /* checksum algorithm to be used */
347 enum cksum_types cl_cksum_type;
348 /* preferred checksum algorithm to be used */
349 enum cksum_types cl_preferred_cksum_type;
351 /* also protected by the poorly named _loi_list_lock lock above */
352 struct osc_async_rc cl_ar;
354 /* sequence manager */
355 struct lu_client_seq *cl_seq;
356 struct rw_semaphore cl_seq_rwsem;
358 atomic_t cl_resends; /* resend count */
360 /* ptlrpc work for writeback in ptlrpcd context */
361 void *cl_writeback_work;
363 struct mutex cl_quota_mutex;
364 /* hash tables for osc_quota_info */
365 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
366 /* the xid of the request updating the hash tables */
367 __u64 cl_quota_last_xid;
368 /* Links to the global list of registered changelog devices */
369 struct list_head cl_chg_dev_linkage;
371 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
378 struct echo_client_obd {
379 struct obd_export *ec_exp; /* the local connection to osc/lov */
381 struct list_head ec_objects;
382 struct list_head ec_locks;
386 /* allow statfs data caching for 1 second */
387 #define OBD_STATFS_CACHE_SECONDS 1
388 /* arbitrary maximum. larger would be useless, allows catching bogus input */
389 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
391 #define lov_tgt_desc lu_tgt_desc
393 struct lov_md_tgt_desc {
394 struct obd_device *lmtd_mdc;
399 struct lov_desc desc;
400 struct lov_tgt_desc **lov_tgts; /* sparse array */
401 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
403 struct mutex lov_lock;
404 struct obd_connect_data lov_ocd;
405 struct proc_dir_entry *targets_proc_entry;
406 atomic_t lov_refcount;
407 __u32 lov_death_row; /* tgts scheduled to be deleted */
408 __u32 lov_tgt_size; /* size of tgts array */
411 struct rhashtable lov_pools_hash_body; /* used for key access */
412 struct list_head lov_pool_list; /* used for sequential access */
413 struct proc_dir_entry *lov_pool_proc_entry;
414 enum lustre_sec_part lov_sp_me;
416 /* Cached LRU and unstable data from upper layer */
417 struct cl_client_cache *lov_cache;
419 struct rw_semaphore lov_notify_lock;
420 /* Data-on-MDT: MDC array */
421 struct lov_md_tgt_desc *lov_mdc_tgts;
423 struct kobject *lov_tgts_kobj;
426 #define lmv_tgt_desc lu_tgt_desc
429 struct lu_client_fld lmv_fld;
435 u32 lmv_statfs_start;
437 struct lu_tgt_descs lmv_mdt_descs;
439 struct obd_connect_data conn_data;
440 struct kobject *lmv_tgts_kobj;
443 __u32 lmv_qos_rr_index;
446 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
447 #define lmv_qos lmv_mdt_descs.ltd_qos
449 /* Minimum sector size is 512 */
450 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
452 struct niobuf_local {
453 __u64 lnb_file_offset;
454 __u32 lnb_page_offset;
458 struct page *lnb_page;
460 __u16 lnb_guards[MAX_GUARD_NUMBER];
461 __u16 lnb_guard_rpc:1;
462 __u16 lnb_guard_disk:1;
463 /* separate unlock for read path to allow shared access */
467 struct tgt_thread_big_cache {
468 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
471 #define LUSTRE_FLD_NAME "fld"
472 #define LUSTRE_SEQ_NAME "seq"
474 #define LUSTRE_MDD_NAME "mdd"
475 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
476 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
477 #define LUSTRE_VVP_NAME "vvp"
478 #define LUSTRE_LMV_NAME "lmv"
479 #define LUSTRE_SLP_NAME "slp"
480 #define LUSTRE_LOD_NAME "lod"
481 #define LUSTRE_OSP_NAME "osp"
482 #define LUSTRE_LWP_NAME "lwp"
484 /* obd device type names */
485 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
486 #define LUSTRE_MDS_NAME "mds"
487 #define LUSTRE_MDT_NAME "mdt"
488 #define LUSTRE_MDC_NAME "mdc"
489 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
490 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
491 #define LUSTRE_OSC_NAME "osc"
492 #define LUSTRE_LOV_NAME "lov"
493 #define LUSTRE_MGS_NAME "mgs"
494 #define LUSTRE_MGC_NAME "mgc"
496 #define LUSTRE_ECHO_NAME "obdecho"
497 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
498 #define LUSTRE_QMT_NAME "qmt"
500 /* Constant obd names (post-rename) */
501 #define LUSTRE_MDS_OBDNAME "MDS"
502 #define LUSTRE_OSS_OBDNAME "OSS"
503 #define LUSTRE_MGS_OBDNAME "MGS"
504 #define LUSTRE_MGC_OBDNAME "MGC"
506 static inline int is_lwp_on_mdt(char *name)
510 ptr = strrchr(name, '-');
512 CERROR("%s is not a obdname\n", name);
516 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
518 if (strncmp(ptr + 1, "MDT", 3) != 0)
521 while (*(--ptr) != '-' && ptr != name);
526 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
532 static inline int is_lwp_on_ost(char *name)
536 ptr = strrchr(name, '-');
538 CERROR("%s is not a obdname\n", name);
542 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
544 if (strncmp(ptr + 1, "OST", 3) != 0)
547 while (*(--ptr) != '-' && ptr != name);
552 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
559 * Events signalled through obd_notify() upcall-chain.
561 enum obd_notify_event {
562 /* Device connect start */
564 /* Device activated */
566 /* Device deactivated */
568 /* Connect data for import were changed */
570 /* Administratively deactivate/activate event */
571 OBD_NOTIFY_DEACTIVATE,
576 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
577 * being main example).
579 struct obd_notify_upcall {
580 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
581 enum obd_notify_event ev, void *owner);
582 /* Opaque datum supplied by upper layer listener */
586 struct target_recovery_data {
587 svc_handler_t trd_recovery_handler;
588 pid_t trd_processing_task;
589 struct completion trd_starting;
590 struct completion trd_finishing;
593 struct obd_llog_group {
594 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
595 wait_queue_head_t olg_waitq;
599 /* corresponds to one of the obd's */
600 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
603 struct obd_type *obd_type;
604 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
605 int obd_minor; /* device number: lctl dl */
606 struct lu_device *obd_lu_dev;
608 /* common and UUID name of this device */
609 struct obd_uuid obd_uuid;
610 char obd_name[MAX_OBD_NAME];
612 /* bitfield modification is protected by obd_dev_lock */
614 obd_attached:1, /* finished attach */
615 obd_set_up:1, /* finished setup */
616 obd_recovering:1, /* there are recoverable clients */
617 obd_abort_recovery:1, /* recovery expired */
618 obd_version_recov:1, /* obd uses version checking */
619 obd_replayable:1, /* recovery enabled; inform clients */
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 */
631 #ifdef HAVE_SERVER_SUPPORT
632 /* no committed-transno notification */
633 unsigned long obd_no_transno:1;
636 /* use separate field as it is set in interrupt to don't mess with
637 * protection of other bits using _bh lock */
638 unsigned long obd_recovery_expired:1;
639 /* uuid-export hash body */
640 struct rhashtable obd_uuid_hash;
641 /* nid-export hash body */
642 struct rhltable obd_nid_hash;
644 struct cfs_hash *obd_nid_stats_hash;
645 /* client_generation-export hash body */
646 struct cfs_hash *obd_gen_hash;
647 struct list_head obd_nid_stats;
648 struct list_head obd_exports;
649 struct list_head obd_unlinked_exports;
650 struct list_head obd_delayed_exports;
651 struct list_head obd_lwp_list;
652 atomic_t obd_refcount;
654 spinlock_t obd_nid_lock;
655 struct ldlm_namespace *obd_namespace;
656 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
657 /* a spinlock is OK for what we do now, may need a semaphore later */
658 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
659 spinlock_t obd_osfs_lock;
660 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
661 time64_t obd_osfs_age;
662 __u64 obd_last_committed;
663 struct mutex obd_dev_mutex;
664 struct lvfs_run_ctxt obd_lvfs_ctxt;
665 struct obd_llog_group obd_olg; /* default llog group */
666 struct obd_device *obd_observer;
667 struct rw_semaphore obd_observer_link_sem;
668 struct obd_notify_upcall obd_upcall;
669 struct obd_export *obd_self_export;
670 struct obd_export *obd_lwp_export;
671 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
672 struct list_head obd_exports_timed;
673 time64_t obd_eviction_timer; /* for ping evictor */
675 atomic_t obd_max_recoverable_clients;
676 atomic_t obd_connected_clients;
677 int obd_stale_clients;
678 /* this lock protects all recovery list_heads, timer and
679 * obd_next_recovery_transno value */
680 spinlock_t obd_recovery_task_lock;
681 __u64 obd_next_recovery_transno;
682 int obd_replayed_requests;
683 int obd_requests_queued_for_recovery;
684 wait_queue_head_t obd_next_transno_waitq;
685 /* protected by obd_recovery_task_lock */
686 struct hrtimer obd_recovery_timer;
688 time64_t obd_recovery_start;
689 /* seconds, for lprocfs_status */
690 time64_t obd_recovery_end;
691 /* To tell timeouts from time stamps Lustre uses timeout_t
692 * instead of time64_t.
694 timeout_t obd_recovery_time_hard;
695 timeout_t obd_recovery_timeout;
696 int obd_recovery_ir_factor;
698 /* new recovery stuff from CMD2 */
699 int obd_replayed_locks;
700 atomic_t obd_req_replay_clients;
701 atomic_t obd_lock_replay_clients;
702 struct target_recovery_data obd_recovery_data;
704 /* all lists are protected by obd_recovery_task_lock */
705 struct list_head obd_req_replay_queue;
706 struct list_head obd_lock_replay_queue;
707 struct list_head obd_final_req_queue;
710 #ifdef HAVE_SERVER_SUPPORT
711 struct obd_device_target obt;
712 struct filter_obd filter;
714 struct echo_obd echo;
716 struct client_obd cli;
717 struct echo_client_obd echo_client;
722 /* Fields used by LProcFS */
723 struct lprocfs_stats *obd_stats;
725 struct lprocfs_stats *obd_md_stats;
727 struct dentry *obd_debugfs_entry;
728 struct proc_dir_entry *obd_proc_entry;
729 struct proc_dir_entry *obd_proc_exports_entry;
730 struct dentry *obd_svc_debugfs_entry;
731 struct lprocfs_stats *obd_svc_stats;
732 const struct attribute **obd_attrs;
733 struct lprocfs_vars *obd_vars;
734 atomic_t obd_evict_inprogress;
735 wait_queue_head_t obd_evict_inprogress_waitq;
736 struct list_head obd_evict_list; /* protected with pet_lock */
739 * LDLM pool part. Save last calculated SLV and Limit.
741 rwlock_t obd_pool_lock;
745 int obd_conn_inprogress;
748 * List of outstanding class_incref()'s fo this OBD. For debugging. */
749 struct lu_ref obd_reference;
751 struct kset obd_kset; /* sysfs object collection */
752 struct kobj_type obd_ktype;
753 struct completion obd_kobj_unregister;
756 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
757 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
758 #ifdef HAVE_SERVER_SUPPORT
759 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
760 struct obd_uuid *uuid);
762 int obd_nid_export_for_each(struct obd_device *obd, lnet_nid_t nid,
763 int cb(struct obd_export *exp, void *data),
765 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
766 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
769 /* get/set_info keys */
770 #define KEY_ASYNC "async"
771 #define KEY_CHANGELOG_CLEAR "changelog_clear"
772 #define KEY_FID2PATH "fid2path"
773 #define KEY_CHECKSUM "checksum"
774 #define KEY_CLEAR_FS "clear_fs"
775 #define KEY_CONN_DATA "conn_data"
776 #define KEY_EVICT_BY_NID "evict_by_nid"
777 #define KEY_FIEMAP "fiemap"
778 #define KEY_FLUSH_CTX "flush_ctx"
779 #define KEY_GRANT_SHRINK "grant_shrink"
780 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
781 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
782 #define KEY_INTERMDS "inter_mds"
783 #define KEY_LAST_ID "last_id"
784 #define KEY_LAST_FID "last_fid"
785 #define KEY_MAX_EASIZE "max_easize"
786 #define KEY_DEFAULT_EASIZE "default_easize"
787 #define KEY_MGSSEC "mgssec"
788 #define KEY_READ_ONLY "read-only"
789 #define KEY_REGISTER_TARGET "register_target"
790 #define KEY_SET_FS "set_fs"
791 #define KEY_TGT_COUNT "tgt_count"
792 /* KEY_SET_INFO in lustre_idl.h */
793 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
795 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
796 #define KEY_OSP_CONNECTED "osp_connected"
798 /* Flags for op_xvalid */
800 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
801 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
802 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
803 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
804 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
805 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
806 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
811 static inline int it_to_lock_mode(struct lookup_intent *it)
813 /* CREAT needs to be tested before open (both could be set) */
814 if (it->it_op & IT_CREAT)
816 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
818 else if (it->it_op & IT_LAYOUT)
819 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
820 else if (it->it_op & IT_READDIR)
822 else if (it->it_op & IT_GETXATTR)
825 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
830 MF_MDC_CANCEL_FID1 = BIT(0),
831 MF_MDC_CANCEL_FID2 = BIT(1),
832 MF_MDC_CANCEL_FID3 = BIT(2),
833 MF_MDC_CANCEL_FID4 = BIT(3),
834 MF_GET_MDT_IDX = BIT(4),
838 CLI_SET_MEA = BIT(0),
839 CLI_RM_ENTRY = BIT(1),
842 CLI_MIGRATE = BIT(4),
843 CLI_DIRTY_DATA = BIT(5),
847 LUSTRE_OPC_MKDIR = 0,
848 LUSTRE_OPC_SYMLINK = 1,
849 LUSTRE_OPC_MKNOD = 2,
850 LUSTRE_OPC_CREATE = 3,
855 * GETXATTR is not included as only a couple of fields in the reply body
856 * is filled, but not FID which is needed for common intent handling in
857 * mdc_finish_intent_lock()
859 static inline bool it_has_reply_body(const struct lookup_intent *it)
861 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
865 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
866 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
867 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
868 struct lu_fid op_fid4; /* to the operation locks. */
869 u32 op_mds; /* what mds server open will go to */
871 enum md_op_code op_code;
872 struct lustre_handle op_open_handle;
876 struct rw_semaphore *op_mea1_sem;
877 struct rw_semaphore *op_mea2_sem;
878 struct lmv_stripe_md *op_mea1;
879 struct lmv_stripe_md *op_mea2;
880 struct lmv_stripe_md *op_default_mea1; /* default LMV */
881 __u32 op_suppgids[2];
888 /* iattr fields and blocks. */
889 struct iattr op_attr;
890 enum op_xvalid op_xvalid; /* eXtra validity flags */
891 loff_t op_attr_blocks;
892 u64 op_valid; /* OBD_MD_* */
893 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
895 enum md_op_flags op_flags;
897 /* Various operation flags. */
898 enum mds_op_bias op_bias;
900 /* used to transfer info between the stacks of MD client
901 * see enum op_cli_flags */
902 enum md_cli_flags op_cli_flags;
904 /* File object data version for HSM release, on client */
905 __u64 op_data_version;
906 struct lustre_handle op_lease_handle;
908 /* File security context, for creates/metadata ops */
909 const char *op_file_secctx_name;
910 __u32 op_file_secctx_name_size;
911 void *op_file_secctx;
912 __u32 op_file_secctx_size;
916 /* Used by readdir */
917 unsigned int op_max_pages;
922 * used to access dir that is changing layout: if it's set, access
923 * dir by new layout, otherwise old layout.
924 * By default it's not set, because new files are created under new
925 * layout, if we can't find file with name under both old and new
926 * layout, we are sure file with name doesn't exist, but in reverse
927 * order there may be a race with creation by others.
930 /* used to access dir with bash hash */
931 __u32 op_stripe_index;
932 /* Archive ID for PCC attach */
937 int (*md_blocking_ast)(struct ldlm_lock *lock,
938 struct ldlm_lock_desc *desc,
939 void *data, int flag);
942 struct md_enqueue_info;
943 /* metadata stat-ahead */
944 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
945 struct md_enqueue_info *minfo,
948 struct md_enqueue_info {
949 struct md_op_data mi_data;
950 struct lookup_intent mi_it;
951 struct lustre_handle mi_lockh;
952 struct inode *mi_dir;
953 struct ldlm_enqueue_info mi_einfo;
954 md_enqueue_cb_t mi_cb;
959 struct module *o_owner;
960 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
961 void *karg, void __user *uarg);
962 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
963 __u32 keylen, void *key, __u32 *vallen, void *val);
964 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
965 __u32 keylen, void *key,
966 __u32 vallen, void *val,
967 struct ptlrpc_request_set *set);
968 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
969 int (*o_precleanup)(struct obd_device *obd);
970 int (*o_cleanup)(struct obd_device *obd);
971 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
972 int (*o_postrecov)(struct obd_device *obd);
973 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
975 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
976 /* connect to the target device with given connection
977 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
978 * granted by the target, which are guaranteed to be a subset of flags
979 * asked for. If @ocd == NULL, use default parameters. */
980 int (*o_connect)(const struct lu_env *env,
981 struct obd_export **exp, struct obd_device *src,
982 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
984 int (*o_reconnect)(const struct lu_env *env,
985 struct obd_export *exp, struct obd_device *src,
986 struct obd_uuid *cluuid,
987 struct obd_connect_data *ocd,
989 int (*o_disconnect)(struct obd_export *exp);
991 /* Initialize/finalize fids infrastructure. */
992 int (*o_fid_init)(struct obd_device *obd,
993 struct obd_export *exp, enum lu_cli_type type);
994 int (*o_fid_fini)(struct obd_device *obd);
996 /* Allocate new fid according to passed @hint. */
997 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
998 struct lu_fid *fid, struct md_op_data *op_data);
1001 * Object with @fid is getting deleted, we may want to do something
1004 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1005 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1006 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1007 time64_t max_age, struct ptlrpc_request_set *set);
1008 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1010 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1012 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1014 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1016 int (*o_preprw)(const struct lu_env *env, int cmd,
1017 struct obd_export *exp, struct obdo *oa, int objcount,
1018 struct obd_ioobj *obj, struct niobuf_remote *remote,
1019 int *nr_pages, struct niobuf_local *local);
1020 int (*o_commitrw)(const struct lu_env *env, int cmd,
1021 struct obd_export *exp, struct obdo *oa,
1022 int objcount, struct obd_ioobj *obj,
1023 struct niobuf_remote *remote, int pages,
1024 struct niobuf_local *local, int rc);
1025 int (*o_init_export)(struct obd_export *exp);
1026 int (*o_destroy_export)(struct obd_export *exp);
1028 int (*o_import_event)(struct obd_device *, struct obd_import *,
1029 enum obd_import_event);
1031 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1032 enum obd_notify_event ev);
1034 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1035 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1038 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1039 struct obd_quotactl *);
1042 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1043 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1044 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1046 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1050 /* lmv structures */
1052 struct mdt_body *body;
1053 struct lu_buf layout;
1055 struct lmv_stripe_md *lmv;
1056 struct lmv_foreign_md *lfm;
1058 struct lmv_stripe_md *default_lmv;
1059 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1060 struct posix_acl *posix_acl;
1064 struct md_open_data {
1065 struct obd_client_handle *mod_och;
1066 struct ptlrpc_request *mod_open_req;
1067 struct ptlrpc_request *mod_close_req;
1068 atomic_t mod_refcount;
1072 struct obd_client_handle {
1073 struct lustre_handle och_open_handle;
1074 struct lu_fid och_fid;
1075 struct md_open_data *och_mod;
1076 struct lustre_handle och_lease_handle; /* open lock for lease */
1081 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1083 struct lookup_intent;
1087 int (*m_close)(struct obd_export *, struct md_op_data *,
1088 struct md_open_data *, struct ptlrpc_request **);
1090 int (*m_create)(struct obd_export *, struct md_op_data *,
1091 const void *, size_t, umode_t, uid_t, gid_t,
1092 cfs_cap_t, __u64, struct ptlrpc_request **);
1094 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1095 const union ldlm_policy_data *, struct md_op_data *,
1096 struct lustre_handle *, __u64);
1098 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1099 struct ptlrpc_request **);
1101 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1102 struct lookup_intent *,
1103 struct ptlrpc_request **,
1104 ldlm_blocking_callback, __u64);
1106 int (*m_link)(struct obd_export *, struct md_op_data *,
1107 struct ptlrpc_request **);
1109 int (*m_rename)(struct obd_export *, struct md_op_data *,
1110 const char *, size_t, const char *, size_t,
1111 struct ptlrpc_request **);
1113 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1114 size_t , struct ptlrpc_request **);
1116 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1117 struct ptlrpc_request **);
1119 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1120 struct md_callback *cb_op, __u64 hash_offset,
1121 struct page **ppage);
1123 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1124 struct ptlrpc_request **);
1126 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1127 u64, const char *, const void *, size_t, unsigned int,
1128 u32, struct ptlrpc_request **);
1130 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1131 u64, const char *, size_t, struct ptlrpc_request **);
1133 int (*m_intent_getattr_async)(struct obd_export *,
1134 struct md_enqueue_info *);
1136 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1137 struct lu_fid *, __u64 *bits);
1139 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1141 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1142 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1144 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1145 struct ptlrpc_request **);
1147 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1149 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1150 struct obd_export *, struct obd_export *,
1151 struct lustre_md *);
1153 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1155 int (*m_merge_attr)(struct obd_export *,
1156 const struct lmv_stripe_md *lsm,
1157 struct cl_attr *attr, ldlm_blocking_callback);
1159 int (*m_set_open_replay_data)(struct obd_export *,
1160 struct obd_client_handle *,
1161 struct lookup_intent *);
1163 int (*m_clear_open_replay_data)(struct obd_export *,
1164 struct obd_client_handle *);
1166 int (*m_set_lock_data)(struct obd_export *,
1167 const struct lustre_handle *, void *, __u64 *);
1169 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1170 const struct lu_fid *, enum ldlm_type,
1171 union ldlm_policy_data *, enum ldlm_mode,
1172 struct lustre_handle *);
1174 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1175 union ldlm_policy_data *, enum ldlm_mode,
1176 enum ldlm_cancel_flags flags, void *opaque);
1178 int (*m_get_fid_from_lsm)(struct obd_export *,
1179 const struct lmv_stripe_md *,
1180 const char *name, int namelen,
1181 struct lu_fid *fid);
1182 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1183 const union lmv_mds_md *lmv, size_t lmv_size);
1184 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1185 struct ptlrpc_request_set *set);
1188 static inline struct md_open_data *obd_mod_alloc(void)
1190 struct md_open_data *mod;
1194 atomic_set(&mod->mod_refcount, 1);
1198 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1199 #define obd_mod_put(mod) \
1201 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1202 if ((mod)->mod_open_req) \
1203 ptlrpc_req_finished((mod)->mod_open_req); \
1204 OBD_FREE_PTR(mod); \
1208 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1209 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1210 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1212 /* return 1 if client should be resend request */
1213 static inline int client_should_resend(int resend, struct client_obd *cli)
1215 return atomic_read(&cli->cl_resends) ?
1216 atomic_read(&cli->cl_resends) > resend : 1;
1220 * Return device name for this device
1222 * XXX: lu_device is declared before obd_device, while a pointer pointing
1223 * back to obd_device in lu_device, so this helper function defines here
1224 * instead of in lu_object.h
1226 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1228 return lu_dev->ld_obd->obd_name;
1231 static inline bool filename_is_volatile(const char *name, size_t namelen,
1237 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1240 /* caller does not care of idx */
1244 /* volatile file, the MDT can be set from name */
1245 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1246 /* if no MDT is specified, use std way */
1247 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1249 /* test for no MDT idx case */
1250 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1251 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1255 /* we have an idx, read it */
1256 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1257 *idx = simple_strtoul(start, &end, 16);
1259 * no digit, no trailing :, negative value
1261 if (((*idx == 0) && (end == start)) ||
1262 (*end != ':') || (*idx < 0))
1267 /* bad format of mdt idx, we cannot return an error
1268 * to caller so we use hash algo */
1269 CERROR("Bad volatile file name format: %s\n",
1270 name + LUSTRE_VOLATILE_HDR_LEN);
1274 static inline int cli_brw_size(struct obd_device *obd)
1276 LASSERT(obd != NULL);
1277 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1281 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1282 * of the client should be increased accordingly to avoid sending fragmented
1283 * RPCs over the network when the client runs out of the maximum dirty space
1284 * when so many RPCs are being generated.
1286 static inline void client_adjust_max_dirty(struct client_obd *cli)
1289 if (cli->cl_dirty_max_pages <= 0) {
1290 cli->cl_dirty_max_pages =
1291 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1293 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1294 cli->cl_max_pages_per_rpc;
1296 if (dirty_max > cli->cl_dirty_max_pages)
1297 cli->cl_dirty_max_pages = dirty_max;
1300 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1301 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1303 /* This value is exported to userspace through the max_dirty_mb
1304 * parameter. So we round up the number of pages to make it a round
1306 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1307 1 << (20 - PAGE_SHIFT));
1310 static inline struct inode *page2inode(struct page *page)
1312 if (page->mapping) {
1316 return page->mapping->host;
1322 #endif /* __OBD_H */