4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
36 #include <linux/posix_acl.h>
37 #include <linux/kobject.h>
38 #include <linux/spinlock.h>
39 #include <linux/sysfs.h>
40 #include <linux/xarray.h>
42 #include <uapi/linux/lustre/lustre_idl.h>
43 #include <lustre_lib.h>
44 #include <libcfs/bitmap.h>
45 #ifdef HAVE_SERVER_SUPPORT
46 # include <lu_target.h>
47 # include <obd_target.h>
48 # include <lustre_quota.h>
51 #include <lustre_export.h>
52 #include <lustre_fid.h>
53 #include <lustre_fld.h>
54 #include <lustre_handles.h>
55 #include <lustre_intent.h>
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 void lov_fix_ea_for_replay(void *lovea);
79 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
82 oinfo->loi_kms_valid = 1;
88 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
90 /* obd info for a particular level (lov, osc). */
92 /* OBD_STATFS_* flags */
94 struct obd_device *oi_obd;
95 struct lu_tgt_desc *oi_tgt;
96 /* statfs data specific for every OSC, if needed at all. */
97 struct obd_statfs *oi_osfs;
98 /* An update callback which is called to update some data on upper
99 * level. E.g. it is used for update lsm->lsm_oinfo at every received
100 * request in osc level for enqueue requests. It is also possible to
101 * update some caller data from LOV layer if needed. */
102 obd_enqueue_update_f oi_cb_up;
106 const struct obd_ops *typ_dt_ops;
107 const struct md_ops *typ_md_ops;
108 struct proc_dir_entry *typ_procroot;
109 struct dentry *typ_debugfs_entry;
110 #ifdef HAVE_SERVER_SUPPORT
114 struct lu_device_type *typ_lu;
115 struct kobject typ_kobj;
117 #define typ_name typ_kobj.name
118 #define OBD_LU_TYPE_SETUP ((void *)0x01UL)
125 /* used for encryption: difference with offset in clear text page */
127 /* used for encryption: difference with count in clear text page */
132 struct timeout_item {
133 enum timeout_event ti_event;
137 struct list_head ti_obd_list;
138 struct list_head ti_chain;
141 #define OBD_MAX_RIF_DEFAULT 8
142 #define OBD_MAX_RIF_MAX 512
143 #define OSC_MAX_RIF_MAX 256
144 #define OSC_MAX_DIRTY_DEFAULT 2000 /* Arbitrary large value */
145 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
146 #define OSC_DEFAULT_RESENDS 10
148 /* possible values for lut_sync_lock_cancel */
149 enum tgt_sync_lock_cancel {
150 SYNC_LOCK_CANCEL_NEVER = 0,
151 SYNC_LOCK_CANCEL_BLOCKING = 1,
152 SYNC_LOCK_CANCEL_ALWAYS = 2,
156 * Limit reply buffer size for striping data to one x86_64 page. This
157 * value is chosen to fit the striping data for common use cases while
158 * staying well below the limit at which the buffer must be backed by
159 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
162 #define OBD_MAX_DEFAULT_EA_SIZE 4096
165 * Lustre can handle larger xattrs internally, but we must respect the Linux
166 * VFS limitation or tools like tar cannot interact with Lustre volumes
169 #define OBD_MAX_EA_SIZE XATTR_SIZE_MAX
172 enum obd_cl_sem_lock_class {
181 struct rw_semaphore cl_sem;
182 struct obd_uuid cl_target_uuid;
183 struct obd_import *cl_import; /* ptlrpc connection state */
184 size_t cl_conn_count;
186 /* Cache maximum and default values for easize. This is
187 * strictly a performance optimization to minimize calls to
188 * obd_size_diskmd(). The default values are used to calculate the
189 * initial size of a request buffer. The ptlrpc layer will resize the
190 * buffer as needed to accommodate a larger reply from the
191 * server. The default values should be small enough to avoid wasted
192 * memory and excessive use of vmalloc(), yet large enough to avoid
193 * reallocating the buffer in the common use case. */
195 /* Default EA size for striping attributes. It is initialized at
196 * mount-time based on the default stripe width of the filesystem,
197 * then it tracks the largest observed EA size advertised by
198 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
199 __u32 cl_default_mds_easize;
201 /* Maximum possible EA size computed at mount-time based on
202 * the number of OSTs in the filesystem. May be increased at
203 * run-time if a larger observed size is advertised by the MDT. */
204 __u32 cl_max_mds_easize;
206 /* Data-on-MDT specific value to set larger reply buffer for possible
207 * data read along with open/stat requests. By default it tries to use
208 * unused space in reply buffer.
209 * This value is used to ensure that reply buffer has at least as
210 * much free space as value indicates. That free space is gained from
211 * LOV EA buffer which is small for DoM files and on big systems can
212 * provide up to 32KB of extra space in reply buffer.
213 * Default value is 8K now.
215 __u32 cl_dom_min_inline_repsize;
217 unsigned int cl_checksum:1, /* 0 = disabled, 1 = enabled */
218 cl_checksum_dump:1, /* same */
219 cl_ocd_grant_param:1;
220 enum lustre_sec_part cl_sp_me;
221 enum lustre_sec_part cl_sp_to;
222 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
224 /* the grant values are protected by loi_list_lock below */
225 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
226 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
227 unsigned long cl_avail_grant; /* bytes of credit for ost */
228 unsigned long cl_lost_grant; /* lost credits (trunc) */
229 /* grant consumed for dirty pages */
230 unsigned long cl_dirty_grant;
232 /* since we allocate grant by blocks, we don't know how many grant will
233 * be used to add a page into cache. As a solution, we reserve maximum
234 * grant before trying to dirty a page and unreserve the rest.
235 * See osc_{reserve|unreserve}_grant for details. */
236 long cl_reserved_grant;
237 wait_queue_head_t cl_cache_waiters; /* waiting for cache/grant */
238 time64_t cl_next_shrink_grant; /* seconds */
239 struct list_head cl_grant_chain;
240 time64_t cl_grant_shrink_interval; /* seconds */
242 int cl_root_squash; /* if root squash enabled*/
244 /* A chunk is an optimal size used by osc_extent to determine
245 * the extent size. A chunk is max(PAGE_SIZE, OST block size) */
247 /* extent insertion metadata overhead to be accounted in grant,
249 unsigned int cl_grant_extent_tax;
250 /* maximum extent size, in number of pages */
251 unsigned int cl_max_extent_pages;
253 /* keep track of objects that have lois that contain pages which
254 * have been queued for async brw. this lock also protects the
255 * lists of osc_client_pages that hang off of the loi */
257 * ->cl_loi_list_lock protects consistency of
258 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
259 * ->ap_completion() call-backs are executed under this lock. As we
260 * cannot guarantee that these call-backs never block on all platforms
261 * (as a matter of fact they do block on Mac OS X), type of
262 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
263 * and blocking mutex on Mac OS X. (Alternative is to make this lock
264 * blocking everywhere, but we don't want to slow down fast-path of
265 * our main platform.)
267 * NB by Jinshan: though field names are still _loi_, but actually
268 * osc_object{}s are in the list.
270 spinlock_t cl_loi_list_lock;
271 struct list_head cl_loi_ready_list;
272 struct list_head cl_loi_hp_ready_list;
273 struct list_head cl_loi_write_list;
274 struct list_head cl_loi_read_list;
275 __u32 cl_r_in_flight;
276 __u32 cl_w_in_flight;
277 /* just a sum of the loi/lop pending numbers to be exported by /proc */
278 atomic_t cl_pending_w_pages;
279 atomic_t cl_pending_r_pages;
280 u32 cl_max_pages_per_rpc;
281 u32 cl_max_rpcs_in_flight;
282 u32 cl_max_short_io_bytes;
283 struct obd_histogram cl_read_rpc_hist;
284 struct obd_histogram cl_write_rpc_hist;
285 struct obd_histogram cl_read_page_hist;
286 struct obd_histogram cl_write_page_hist;
287 struct obd_histogram cl_read_offset_hist;
288 struct obd_histogram cl_write_offset_hist;
290 /** LRU for osc caching pages */
291 struct cl_client_cache *cl_cache;
292 /** member of cl_cache->ccc_lru */
293 struct list_head cl_lru_osc;
294 /** # of available LRU slots left in the per-OSC cache.
295 * Available LRU slots are shared by all OSCs of the same file system,
296 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
297 atomic_long_t *cl_lru_left;
298 /** # of busy LRU pages. A page is considered busy if it's in writeback
299 * queue, or in transfer. Busy pages can't be discarded so they are not
301 atomic_long_t cl_lru_busy;
302 /** # of LRU pages in the cache for this client_obd */
303 atomic_long_t cl_lru_in_list;
304 /** # of threads are shrinking LRU cache. To avoid contention, it's not
305 * allowed to have multiple threads shrinking LRU cache. */
306 atomic_t cl_lru_shrinkers;
307 /** The time when this LRU cache was last used. */
308 time64_t cl_lru_last_used;
309 /** stats: how many reclaims have happened for this client_obd.
310 * reclaim and shrink - shrink is async, voluntarily rebalancing;
311 * reclaim is sync, initiated by IO thread when the LRU slots are
313 __u64 cl_lru_reclaim;
314 /** List of LRU pages for this client_obd */
315 struct list_head cl_lru_list;
316 /** Lock for LRU page list */
317 spinlock_t cl_lru_list_lock;
318 /** # of unstable pages in this client_obd.
319 * An unstable page is a page state that WRITE RPC has finished but
320 * the transaction has NOT yet committed. */
321 atomic_long_t cl_unstable_count;
322 /** Link to osc_shrinker_list */
323 struct list_head cl_shrink_list;
325 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
326 atomic_t cl_destroy_in_flight;
327 wait_queue_head_t cl_destroy_waitq;
329 /* modify rpcs in flight
330 * currently used for metadata only */
331 spinlock_t cl_mod_rpcs_lock;
332 __u16 cl_max_mod_rpcs_in_flight;
333 __u16 cl_mod_rpcs_in_flight;
334 __u16 cl_close_rpcs_in_flight;
335 wait_queue_head_t cl_mod_rpcs_waitq;
336 unsigned long *cl_mod_tag_bitmap;
337 struct obd_histogram cl_mod_rpcs_hist;
340 struct mutex cl_mgc_mutex;
341 struct local_oid_storage *cl_mgc_los;
342 struct dt_object *cl_mgc_configs_dir;
343 struct obd_export *cl_mgc_mgsexp;
344 atomic_t cl_mgc_refcount;
345 /* in-flight control list and total RPCs counter */
346 struct list_head cl_flight_waiters;
347 __u32 cl_rpcs_in_flight;
349 /* supported checksum types that are worked out at connect time */
350 __u32 cl_supp_cksum_types;
351 /* checksum algorithm to be used */
352 enum cksum_types cl_cksum_type;
353 /* preferred checksum algorithm to be used */
354 enum cksum_types cl_preferred_cksum_type;
356 /* also protected by the poorly named _loi_list_lock lock above */
357 struct osc_async_rc cl_ar;
359 /* sequence manager */
360 struct lu_client_seq *cl_seq;
361 struct rw_semaphore cl_seq_rwsem;
363 atomic_t cl_resends; /* resend count */
365 /* ptlrpc work for writeback in ptlrpcd context */
366 void *cl_writeback_work;
368 struct mutex cl_quota_mutex;
369 /* hash tables for osc_quota_info */
370 struct cfs_hash *cl_quota_hash[LL_MAXQUOTAS];
371 /* the xid of the request updating the hash tables */
372 __u64 cl_quota_last_xid;
373 /* Links to the global list of registered changelog devices */
374 struct list_head cl_chg_dev_linkage;
376 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
383 struct echo_client_obd {
384 struct obd_export *ec_exp; /* the local connection to osc/lov */
386 struct list_head ec_objects;
387 struct list_head ec_locks;
391 /* allow statfs data caching for 1 second */
392 #define OBD_STATFS_CACHE_SECONDS 1
393 /* arbitrary maximum. larger would be useless, allows catching bogus input */
394 #define OBD_STATFS_CACHE_MAX_AGE 3600 /* seconds */
396 #define lov_tgt_desc lu_tgt_desc
398 struct lov_md_tgt_desc {
399 struct obd_device *lmtd_mdc;
404 struct lov_desc desc;
405 struct lov_tgt_desc **lov_tgts; /* sparse array */
406 struct lu_tgt_pool lov_packed; /* all OSTs in a packed
408 struct mutex lov_lock;
409 struct obd_connect_data lov_ocd;
410 atomic_t lov_refcount;
411 __u32 lov_death_row; /* tgts scheduled to be deleted */
412 __u32 lov_tgt_size; /* size of tgts array */
415 struct rhashtable lov_pools_hash_body; /* used for key access */
416 struct list_head lov_pool_list; /* used for sequential access */
417 struct proc_dir_entry *lov_pool_proc_entry;
418 enum lustre_sec_part lov_sp_me;
420 /* Cached LRU and unstable data from upper layer */
421 struct cl_client_cache *lov_cache;
423 struct rw_semaphore lov_notify_lock;
424 /* Data-on-MDT: MDC array */
425 struct lov_md_tgt_desc *lov_mdc_tgts;
427 struct kobject *lov_tgts_kobj;
430 #define lmv_tgt_desc lu_tgt_desc
433 struct lu_client_fld lmv_fld;
439 u32 lmv_statfs_start;
441 struct lu_tgt_descs lmv_mdt_descs;
443 struct obd_connect_data conn_data;
444 struct kobject *lmv_tgts_kobj;
447 __u32 lmv_qos_rr_index;
450 #define lmv_mdt_count lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count
451 #define lmv_qos lmv_mdt_descs.ltd_qos
453 /* Minimum sector size is 512 */
454 #define MAX_GUARD_NUMBER (PAGE_SIZE / 512)
456 struct niobuf_local {
457 __u64 lnb_file_offset;
458 __u32 lnb_page_offset;
462 struct page *lnb_page;
464 __u16 lnb_guards[MAX_GUARD_NUMBER];
465 __u16 lnb_guard_rpc:1;
466 __u16 lnb_guard_disk:1;
467 /* separate unlock for read path to allow shared access */
471 struct tgt_thread_big_cache {
472 struct niobuf_local local[PTLRPC_MAX_BRW_PAGES];
475 #define LUSTRE_FLD_NAME "fld"
476 #define LUSTRE_SEQ_NAME "seq"
478 #define LUSTRE_MDD_NAME "mdd"
479 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
480 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
481 #define LUSTRE_VVP_NAME "vvp"
482 #define LUSTRE_LMV_NAME "lmv"
483 #define LUSTRE_SLP_NAME "slp"
484 #define LUSTRE_LOD_NAME "lod"
485 #define LUSTRE_OSP_NAME "osp"
486 #define LUSTRE_LWP_NAME "lwp"
488 /* obd device type names */
489 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
490 #define LUSTRE_MDS_NAME "mds"
491 #define LUSTRE_MDT_NAME "mdt"
492 #define LUSTRE_MDC_NAME "mdc"
493 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
494 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
495 #define LUSTRE_OSC_NAME "osc"
496 #define LUSTRE_LOV_NAME "lov"
497 #define LUSTRE_MGS_NAME "mgs"
498 #define LUSTRE_MGC_NAME "mgc"
500 #define LUSTRE_ECHO_NAME "obdecho"
501 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
502 #define LUSTRE_QMT_NAME "qmt"
504 /* Constant obd names (post-rename) */
505 #define LUSTRE_MDS_OBDNAME "MDS"
506 #define LUSTRE_OSS_OBDNAME "OSS"
507 #define LUSTRE_MGS_OBDNAME "MGS"
508 #define LUSTRE_MGC_OBDNAME "MGC"
510 static inline int is_lwp_on_mdt(char *name)
514 ptr = strrchr(name, '-');
516 CERROR("%s is not a obdname\n", name);
520 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
522 if (strncmp(ptr + 1, "MDT", 3) != 0)
525 while (*(--ptr) != '-' && ptr != name);
530 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
536 static inline int is_lwp_on_ost(char *name)
540 ptr = strrchr(name, '-');
542 CERROR("%s is not a obdname\n", name);
546 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
548 if (strncmp(ptr + 1, "OST", 3) != 0)
551 while (*(--ptr) != '-' && ptr != name);
556 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
563 * Events signalled through obd_notify() upcall-chain.
565 enum obd_notify_event {
566 /* Device connect start */
568 /* Device activated */
570 /* Device deactivated */
572 /* Connect data for import were changed */
574 /* Administratively deactivate/activate event */
575 OBD_NOTIFY_DEACTIVATE,
580 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
581 * being main example).
583 struct obd_notify_upcall {
584 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
585 enum obd_notify_event ev, void *owner);
586 /* Opaque datum supplied by upper layer listener */
590 struct target_recovery_data {
591 svc_handler_t trd_recovery_handler;
592 pid_t trd_processing_task;
593 struct completion trd_starting;
594 struct completion trd_finishing;
597 struct obd_llog_group {
598 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
599 wait_queue_head_t olg_waitq;
603 /* corresponds to one of the obd's */
604 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
607 struct obd_type *obd_type;
608 __u32 obd_magic; /* OBD_DEVICE_MAGIC */
609 int obd_minor; /* device number: lctl dl */
610 struct lu_device *obd_lu_dev;
612 /* common and UUID name of this device */
613 struct obd_uuid obd_uuid;
614 char obd_name[MAX_OBD_NAME];
616 /* bitfield modification is protected by obd_dev_lock */
618 obd_attached:1, /* finished attach */
619 obd_set_up:1, /* finished setup */
620 obd_recovering:1, /* there are recoverable clients */
621 obd_abort_recovery:1, /* recovery expired */
622 obd_abort_recov_mdt:1, /* only abort recovery between MDTs */
623 obd_version_recov:1, /* obd uses version checking */
624 obd_replayable:1, /* recovery enabled; inform clients */
625 obd_no_recov:1, /* fail instead of retry messages */
626 obd_stopping:1, /* started cleanup */
627 obd_starting:1, /* started setup */
628 obd_force:1, /* cleanup with > 0 obd refcount */
629 obd_fail:1, /* cleanup with failover */
630 obd_no_conn:1, /* deny new connections */
631 obd_inactive:1, /* device active/inactive
632 * (for /proc/status only!!) */
633 obd_no_ir:1, /* no imperative recovery. */
634 obd_process_conf:1, /* device is processing mgs config */
635 obd_checksum_dump:1, /* dump pages upon cksum error */
636 obd_dynamic_nids:1; /* Allow dynamic NIDs on device */
637 #ifdef HAVE_SERVER_SUPPORT
638 /* no committed-transno notification */
639 unsigned long obd_no_transno:1;
642 /* use separate field as it is set in interrupt to don't mess with
643 * protection of other bits using _bh lock */
644 unsigned long obd_recovery_expired:1;
645 /* uuid-export hash body */
646 struct rhashtable obd_uuid_hash;
647 /* nid-export hash body */
648 struct rhltable obd_nid_hash;
650 struct cfs_hash *obd_nid_stats_hash;
651 /* client_generation-export hash body */
652 struct cfs_hash *obd_gen_hash;
653 struct list_head obd_nid_stats;
654 struct list_head obd_exports;
655 struct list_head obd_unlinked_exports;
656 struct list_head obd_delayed_exports;
657 struct list_head obd_lwp_list;
658 atomic_t obd_refcount;
660 spinlock_t obd_nid_lock;
661 struct ldlm_namespace *obd_namespace;
662 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
663 /* a spinlock is OK for what we do now, may need a semaphore later */
664 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
665 spinlock_t obd_osfs_lock;
666 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
667 time64_t obd_osfs_age;
668 __u64 obd_last_committed;
669 struct mutex obd_dev_mutex;
670 struct lvfs_run_ctxt obd_lvfs_ctxt;
671 struct obd_llog_group obd_olg; /* default llog group */
672 struct obd_device *obd_observer;
673 struct rw_semaphore obd_observer_link_sem;
674 struct obd_notify_upcall obd_upcall;
675 struct obd_export *obd_self_export;
676 struct obd_export *obd_lwp_export;
677 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
678 struct list_head obd_exports_timed;
679 time64_t obd_eviction_timer; /* for ping evictor */
681 atomic_t obd_max_recoverable_clients;
682 atomic_t obd_connected_clients;
683 int obd_stale_clients;
684 /* this lock protects all recovery list_heads, timer and
685 * obd_next_recovery_transno value */
686 spinlock_t obd_recovery_task_lock;
687 __u64 obd_next_recovery_transno;
688 int obd_replayed_requests;
689 int obd_requests_queued_for_recovery;
690 wait_queue_head_t obd_next_transno_waitq;
691 /* protected by obd_recovery_task_lock */
692 struct hrtimer obd_recovery_timer;
694 time64_t obd_recovery_start;
695 /* seconds, for lprocfs_status */
696 time64_t obd_recovery_end;
697 /* To tell timeouts from time stamps Lustre uses timeout_t
698 * instead of time64_t.
700 timeout_t obd_recovery_time_hard;
701 timeout_t obd_recovery_timeout;
702 int obd_recovery_ir_factor;
704 /* new recovery stuff from CMD2 */
705 int obd_replayed_locks;
706 atomic_t obd_req_replay_clients;
707 atomic_t obd_lock_replay_clients;
708 struct target_recovery_data obd_recovery_data;
710 /* all lists are protected by obd_recovery_task_lock */
711 struct list_head obd_req_replay_queue;
712 struct list_head obd_lock_replay_queue;
713 struct list_head obd_final_req_queue;
716 #ifdef HAVE_SERVER_SUPPORT
717 struct obd_device_target obt;
718 struct filter_obd filter;
720 struct echo_obd echo;
722 struct client_obd cli;
723 struct echo_client_obd echo_client;
728 /* Fields used by LProcFS */
729 struct lprocfs_stats *obd_stats;
731 struct lprocfs_stats *obd_md_stats;
733 struct dentry *obd_debugfs_entry;
734 struct proc_dir_entry *obd_proc_entry;
735 struct proc_dir_entry *obd_proc_exports_entry;
736 struct dentry *obd_svc_debugfs_entry;
737 struct lprocfs_stats *obd_svc_stats;
738 const struct attribute **obd_attrs;
739 struct lprocfs_vars *obd_vars;
740 struct ldebugfs_vars *obd_debugfs_vars;
741 atomic_t obd_evict_inprogress;
742 wait_queue_head_t obd_evict_inprogress_waitq;
743 struct list_head obd_evict_list; /* protected with pet_lock */
746 * LDLM pool part. Save last calculated SLV and Limit.
748 rwlock_t obd_pool_lock;
752 int obd_conn_inprogress;
755 * List of outstanding class_incref()'s fo this OBD. For debugging. */
756 struct lu_ref obd_reference;
758 struct kset obd_kset; /* sysfs object collection */
759 struct kobj_type obd_ktype;
760 struct completion obd_kobj_unregister;
763 int obd_uuid_add(struct obd_device *obd, struct obd_export *export);
764 void obd_uuid_del(struct obd_device *obd, struct obd_export *export);
765 #ifdef HAVE_SERVER_SUPPORT
766 struct obd_export *obd_uuid_lookup(struct obd_device *obd,
767 struct obd_uuid *uuid);
769 int obd_nid_export_for_each(struct obd_device *obd, lnet_nid_t nid,
770 int cb(struct obd_export *exp, void *data),
772 int obd_nid_add(struct obd_device *obd, struct obd_export *exp);
773 void obd_nid_del(struct obd_device *obd, struct obd_export *exp);
776 /* get/set_info keys */
777 #define KEY_ASYNC "async"
778 #define KEY_CHANGELOG_CLEAR "changelog_clear"
779 #define KEY_FID2PATH "fid2path"
780 #define KEY_CHECKSUM "checksum"
781 #define KEY_CLEAR_FS "clear_fs"
782 #define KEY_CONN_DATA "conn_data"
783 #define KEY_EVICT_BY_NID "evict_by_nid"
784 #define KEY_FIEMAP "fiemap"
785 #define KEY_FLUSH_CTX "flush_ctx"
786 #define KEY_GRANT_SHRINK "grant_shrink"
787 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
788 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
789 #define KEY_INTERMDS "inter_mds"
790 #define KEY_LAST_ID "last_id"
791 #define KEY_LAST_FID "last_fid"
792 #define KEY_MAX_EASIZE "max_easize"
793 #define KEY_DEFAULT_EASIZE "default_easize"
794 #define KEY_MGSSEC "mgssec"
795 #define KEY_READ_ONLY "read-only"
796 #define KEY_REGISTER_TARGET "register_target"
797 #define KEY_SET_FS "set_fs"
798 #define KEY_TGT_COUNT "tgt_count"
799 /* KEY_SET_INFO in lustre_idl.h */
800 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
802 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
803 #define KEY_OSP_CONNECTED "osp_connected"
805 /* Flags for op_xvalid */
807 OP_XVALID_CTIME_SET = BIT(0), /* 0x0001 */
808 OP_XVALID_BLOCKS = BIT(1), /* 0x0002 */
809 OP_XVALID_OWNEROVERRIDE = BIT(2), /* 0x0004 */
810 OP_XVALID_FLAGS = BIT(3), /* 0x0008 */
811 OP_XVALID_PROJID = BIT(4), /* 0x0010 */
812 OP_XVALID_LAZYSIZE = BIT(5), /* 0x0020 */
813 OP_XVALID_LAZYBLOCKS = BIT(6), /* 0x0040 */
818 static inline int it_to_lock_mode(struct lookup_intent *it)
820 /* CREAT needs to be tested before open (both could be set) */
821 if (it->it_op & IT_CREAT)
823 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP))
825 else if (it->it_op & IT_LAYOUT)
826 return (it->it_flags & FMODE_WRITE) ? LCK_EX : LCK_CR;
827 else if (it->it_op & IT_READDIR)
829 else if (it->it_op & IT_GETXATTR)
832 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
837 MF_MDC_CANCEL_FID1 = BIT(0),
838 MF_MDC_CANCEL_FID2 = BIT(1),
839 MF_MDC_CANCEL_FID3 = BIT(2),
840 MF_MDC_CANCEL_FID4 = BIT(3),
841 MF_GET_MDT_IDX = BIT(4),
842 MF_GETATTR_BY_FID = BIT(5),
843 MF_QOS_MKDIR = BIT(6),
844 MF_RR_MKDIR = BIT(7),
845 MF_OPNAME_KMALLOCED = BIT(8),
849 CLI_SET_MEA = BIT(0),
850 CLI_RM_ENTRY = BIT(1),
853 CLI_MIGRATE = BIT(4),
854 CLI_DIRTY_DATA = BIT(5),
858 LUSTRE_OPC_MKDIR = 1,
869 * GETXATTR is not included as only a couple of fields in the reply body
870 * is filled, but not FID which is needed for common intent handling in
871 * mdc_finish_intent_lock()
873 static inline bool it_has_reply_body(const struct lookup_intent *it)
875 return it->it_op & (IT_OPEN | IT_LOOKUP | IT_GETATTR);
879 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
880 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
881 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
882 struct lu_fid op_fid4; /* to the operation locks. */
883 u32 op_mds; /* what mds server open will go to */
885 enum md_op_code op_code;
886 struct lustre_handle op_open_handle;
890 struct rw_semaphore *op_mea1_sem;
891 struct rw_semaphore *op_mea2_sem;
892 struct lmv_stripe_md *op_mea1;
893 struct lmv_stripe_md *op_mea2;
894 struct lmv_stripe_md *op_default_mea1; /* default LMV */
895 __u32 op_suppgids[2];
902 /* iattr fields and blocks. */
903 struct iattr op_attr;
904 enum op_xvalid op_xvalid; /* eXtra validity flags */
905 loff_t op_attr_blocks;
906 u64 op_valid; /* OBD_MD_* */
907 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
909 enum md_op_flags op_flags;
911 /* Various operation flags. */
912 enum mds_op_bias op_bias;
914 /* used to transfer info between the stacks of MD client
915 * see enum op_cli_flags */
916 enum md_cli_flags op_cli_flags;
918 /* File object data version for HSM release, on client */
919 __u64 op_data_version;
920 struct lustre_handle op_lease_handle;
922 /* File security context, for creates/metadata ops */
923 const char *op_file_secctx_name;
924 __u32 op_file_secctx_name_size;
925 void *op_file_secctx;
926 __u32 op_file_secctx_size;
928 /* File encryption context, for creates/metadata ops */
929 void *op_file_encctx;
930 __u32 op_file_encctx_size;
935 /* Used by readdir */
936 unsigned int op_max_pages;
938 unsigned short op_dir_depth;
944 * used to access dir that is changing layout: if it's set, access
945 * dir by new layout, otherwise old layout.
946 * By default it's not set, because new files are created under new
947 * layout, if we can't find file with name under both old and new
948 * layout, we are sure file with name doesn't exist, but in reverse
949 * order there may be a race with creation by others.
952 /* used to access dir with bash hash */
953 __u32 op_stripe_index;
954 /* Archive ID for PCC attach */
959 int (*md_blocking_ast)(struct ldlm_lock *lock,
960 struct ldlm_lock_desc *desc,
961 void *data, int flag);
964 struct md_enqueue_info;
965 /* metadata stat-ahead */
966 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
967 struct md_enqueue_info *minfo,
970 struct md_enqueue_info {
971 struct md_op_data mi_data;
972 struct lookup_intent mi_it;
973 struct lustre_handle mi_lockh;
974 struct inode *mi_dir;
975 struct ldlm_enqueue_info mi_einfo;
976 md_enqueue_cb_t mi_cb;
981 struct module *o_owner;
982 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
983 void *karg, void __user *uarg);
984 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
985 __u32 keylen, void *key, __u32 *vallen, void *val);
986 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
987 __u32 keylen, void *key,
988 __u32 vallen, void *val,
989 struct ptlrpc_request_set *set);
990 int (*o_setup) (struct obd_device *obd, struct lustre_cfg *cfg);
991 int (*o_precleanup)(struct obd_device *obd);
992 int (*o_cleanup)(struct obd_device *obd);
993 int (*o_process_config)(struct obd_device *obd, size_t len, void *data);
994 int (*o_postrecov)(struct obd_device *obd);
995 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
997 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
998 /* connect to the target device with given connection
999 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1000 * granted by the target, which are guaranteed to be a subset of flags
1001 * asked for. If @ocd == NULL, use default parameters. */
1002 int (*o_connect)(const struct lu_env *env,
1003 struct obd_export **exp, struct obd_device *src,
1004 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1006 int (*o_reconnect)(const struct lu_env *env,
1007 struct obd_export *exp, struct obd_device *src,
1008 struct obd_uuid *cluuid,
1009 struct obd_connect_data *ocd,
1011 int (*o_disconnect)(struct obd_export *exp);
1013 /* Initialize/finalize fids infrastructure. */
1014 int (*o_fid_init)(struct obd_device *obd,
1015 struct obd_export *exp, enum lu_cli_type type);
1016 int (*o_fid_fini)(struct obd_device *obd);
1018 /* Allocate new fid according to passed @hint. */
1019 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
1020 struct lu_fid *fid, struct md_op_data *op_data);
1023 * Object with @fid is getting deleted, we may want to do something
1026 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1027 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
1028 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1029 time64_t max_age, struct ptlrpc_request_set *set);
1030 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1032 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1034 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1036 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1038 int (*o_preprw)(const struct lu_env *env, int cmd,
1039 struct obd_export *exp, struct obdo *oa, int objcount,
1040 struct obd_ioobj *obj, struct niobuf_remote *remote,
1041 int *nr_pages, struct niobuf_local *local);
1042 int (*o_commitrw)(const struct lu_env *env, int cmd,
1043 struct obd_export *exp, struct obdo *oa,
1044 int objcount, struct obd_ioobj *obj,
1045 struct niobuf_remote *remote, int pages,
1046 struct niobuf_local *local, int rc);
1047 int (*o_init_export)(struct obd_export *exp);
1048 int (*o_destroy_export)(struct obd_export *exp);
1050 int (*o_import_event)(struct obd_device *, struct obd_import *,
1051 enum obd_import_event);
1053 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1054 enum obd_notify_event ev);
1056 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1057 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1060 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1061 struct obd_quotactl *);
1064 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1065 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1066 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1068 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1072 /* lmv structures */
1074 struct mdt_body *body;
1075 struct lu_buf layout;
1077 struct lmv_stripe_md *lmv;
1078 struct lmv_foreign_md *lfm;
1080 struct lmv_stripe_md *default_lmv;
1081 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1082 struct posix_acl *posix_acl;
1086 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1087 static inline void lmd_clear_acl(struct lustre_md *md)
1089 if (md->posix_acl) {
1090 posix_acl_release(md->posix_acl);
1091 md->posix_acl = NULL;
1095 #define OBD_CONNECT_ACL_FLAGS \
1096 (OBD_CONNECT_ACL | OBD_CONNECT_UMASK | OBD_CONNECT_LARGE_ACL)
1098 static inline void lmd_clear_acl(struct lustre_md *md)
1102 #define OBD_CONNECT_ACL_FLAGS (0)
1105 struct md_open_data {
1106 struct obd_client_handle *mod_och;
1107 struct ptlrpc_request *mod_open_req;
1108 struct ptlrpc_request *mod_close_req;
1109 atomic_t mod_refcount;
1113 struct obd_client_handle {
1114 struct lustre_handle och_open_handle;
1115 struct lu_fid och_fid;
1116 struct md_open_data *och_mod;
1117 struct lustre_handle och_lease_handle; /* open lock for lease */
1122 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1124 struct lookup_intent;
1128 int (*m_close)(struct obd_export *, struct md_op_data *,
1129 struct md_open_data *, struct ptlrpc_request **);
1131 int (*m_create)(struct obd_export *, struct md_op_data *,
1132 const void *, size_t, umode_t, uid_t, gid_t,
1133 kernel_cap_t, __u64, struct ptlrpc_request **);
1135 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1136 const union ldlm_policy_data *, struct md_op_data *,
1137 struct lustre_handle *, __u64);
1139 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1140 struct ptlrpc_request **);
1142 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1143 struct lookup_intent *,
1144 struct ptlrpc_request **,
1145 ldlm_blocking_callback, __u64);
1147 int (*m_link)(struct obd_export *, struct md_op_data *,
1148 struct ptlrpc_request **);
1150 int (*m_rename)(struct obd_export *, struct md_op_data *,
1151 const char *, size_t, const char *, size_t,
1152 struct ptlrpc_request **);
1154 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1155 size_t , struct ptlrpc_request **);
1157 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1158 struct ptlrpc_request **);
1160 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1161 struct md_callback *cb_op, __u64 hash_offset,
1162 struct page **ppage);
1164 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1165 struct ptlrpc_request **);
1167 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1168 u64, const char *, const void *, size_t, unsigned int,
1169 u32, struct ptlrpc_request **);
1171 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1172 u64, const char *, size_t, struct ptlrpc_request **);
1174 int (*m_intent_getattr_async)(struct obd_export *,
1175 struct md_enqueue_info *);
1177 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1178 struct lu_fid *, __u64 *bits);
1180 int (*m_file_resync)(struct obd_export *, struct md_op_data *);
1182 int (*m_get_root)(struct obd_export *, const char *, struct lu_fid *);
1183 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1185 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1186 struct ptlrpc_request **);
1188 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1190 int (*m_get_lustre_md)(struct obd_export *, struct req_capsule *,
1191 struct obd_export *, struct obd_export *,
1192 struct lustre_md *);
1194 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1196 int (*m_merge_attr)(struct obd_export *,
1197 const struct lmv_stripe_md *lsm,
1198 struct cl_attr *attr, ldlm_blocking_callback);
1200 int (*m_set_open_replay_data)(struct obd_export *,
1201 struct obd_client_handle *,
1202 struct lookup_intent *);
1204 int (*m_clear_open_replay_data)(struct obd_export *,
1205 struct obd_client_handle *);
1207 int (*m_set_lock_data)(struct obd_export *,
1208 const struct lustre_handle *, void *, __u64 *);
1210 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1211 const struct lu_fid *, enum ldlm_type,
1212 union ldlm_policy_data *, enum ldlm_mode,
1213 struct lustre_handle *);
1215 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1216 union ldlm_policy_data *, enum ldlm_mode,
1217 enum ldlm_cancel_flags flags, void *opaque);
1219 int (*m_get_fid_from_lsm)(struct obd_export *,
1220 const struct lmv_stripe_md *,
1221 const char *name, int namelen,
1222 struct lu_fid *fid);
1223 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1224 const union lmv_mds_md *lmv, size_t lmv_size);
1225 int (*m_rmfid)(struct obd_export *exp, struct fid_array *fa, int *rcs,
1226 struct ptlrpc_request_set *set);
1229 static inline struct md_open_data *obd_mod_alloc(void)
1231 struct md_open_data *mod;
1235 atomic_set(&mod->mod_refcount, 1);
1239 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1240 #define obd_mod_put(mod) \
1242 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1243 if ((mod)->mod_open_req) \
1244 ptlrpc_req_finished((mod)->mod_open_req); \
1245 OBD_FREE_PTR(mod); \
1249 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1250 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1251 void obdo_set_o_projid(struct obdo *dst, u32 projid);
1253 /* return 1 if client should be resend request */
1254 static inline int client_should_resend(int resend, struct client_obd *cli)
1256 return atomic_read(&cli->cl_resends) ?
1257 atomic_read(&cli->cl_resends) > resend : 1;
1261 * Return device name for this device
1263 * XXX: lu_device is declared before obd_device, while a pointer pointing
1264 * back to obd_device in lu_device, so this helper function defines here
1265 * instead of in lu_object.h
1267 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1269 return lu_dev->ld_obd->obd_name;
1272 static inline bool filename_is_volatile(const char *name, size_t namelen,
1278 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1281 /* caller does not care of idx */
1285 /* volatile file, the MDT can be set from name */
1286 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1287 /* if no MDT is specified, use std way */
1288 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1290 /* test for no MDT idx case */
1291 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1292 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1296 /* we have an idx, read it */
1297 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1298 *idx = simple_strtoul(start, &end, 16);
1300 * no digit, no trailing :, negative value
1302 if (((*idx == 0) && (end == start)) ||
1303 (*end != ':') || (*idx < 0))
1308 /* bad format of mdt idx, we cannot return an error
1309 * to caller so we use hash algo */
1310 CERROR("Bad volatile file name format: %s\n",
1311 name + LUSTRE_VOLATILE_HDR_LEN);
1315 static inline int cli_brw_size(struct obd_device *obd)
1317 LASSERT(obd != NULL);
1318 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1322 * When RPC size or the max RPCs in flight is increased, the max dirty pages
1323 * of the client should be increased accordingly to avoid sending fragmented
1324 * RPCs over the network when the client runs out of the maximum dirty space
1325 * when so many RPCs are being generated.
1327 static inline void client_adjust_max_dirty(struct client_obd *cli)
1330 if (cli->cl_dirty_max_pages <= 0) {
1331 cli->cl_dirty_max_pages =
1332 (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024) >> PAGE_SHIFT;
1334 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1335 cli->cl_max_pages_per_rpc;
1337 if (dirty_max > cli->cl_dirty_max_pages)
1338 cli->cl_dirty_max_pages = dirty_max;
1341 if (cli->cl_dirty_max_pages > cfs_totalram_pages() / 8)
1342 cli->cl_dirty_max_pages = cfs_totalram_pages() / 8;
1344 /* This value is exported to userspace through the max_dirty_mb
1345 * parameter. So we round up the number of pages to make it a round
1347 cli->cl_dirty_max_pages = round_up(cli->cl_dirty_max_pages,
1348 1 << (20 - PAGE_SHIFT));
1351 static inline struct inode *page2inode(struct page *page)
1353 if (page->mapping) {
1357 return page->mapping->host;
1363 #endif /* __OBD_H */