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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #include <linux/spinlock.h>
42 #include <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>
50 #include <lustre_export.h>
51 #include <lustre_fid.h>
52 #include <lustre_fld.h>
53 #include <lustre_handles.h>
54 #include <lustre_intent.h>
55 #include <lustre_capa.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 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
87 /* If we are unable to get the maximum object size from the OST in
88 * ocd_maxbytes using OBD_CONNECT_MAXBYTES, then we fall back to using
89 * the old maximum object size from ext3. */
90 #define LUSTRE_EXT3_STRIPE_MAXBYTES 0x1fffffff000ULL
92 struct lov_stripe_md {
95 pid_t lsm_lock_owner; /* debugging */
97 /* maximum possible file size, might change as OSTs status changes,
98 * e.g. disconnected, deactivated */
100 struct ost_id lsm_oi;
102 __u32 lsm_stripe_size;
103 __u32 lsm_pattern; /* RAID0, RAID1, released, ... */
104 __u16 lsm_stripe_count;
105 __u16 lsm_layout_gen;
106 char lsm_pool_name[LOV_MAXPOOLNAME + 1];
107 struct lov_oinfo *lsm_oinfo[0];
110 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
112 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
115 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
119 if (lsm_is_released(lsm))
124 static inline int lov_stripe_md_size(unsigned int stripe_count)
126 struct lov_stripe_md lsm;
128 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
133 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
135 /* obd info for a particular level (lov, osc). */
137 /* Lock policy. It keeps an extent which is specific for a particular
138 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
139 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
140 ldlm_policy_data_t oi_policy;
141 /* Flags used for set request specific flags:
142 - while lock handling, the flags obtained on the enqueue
143 request are set here.
144 - while stats, the flags used for control delay/resend.
145 - while setattr, the flags used for distinguish punch operation
148 /* obdo data specific for every OSC, if needed at all. */
150 /* statfs data specific for every OSC, if needed at all. */
151 struct obd_statfs *oi_osfs;
152 /* An update callback which is called to update some data on upper
153 * level. E.g. it is used for update lsm->lsm_oinfo at every received
154 * request in osc level for enqueue requests. It is also possible to
155 * update some caller data from LOV layer if needed. */
156 obd_enqueue_update_f oi_cb_up;
157 /* oss capability, its type is obd_capa in client to avoid copy.
158 * in contrary its type is lustre_capa in OSS. */
163 struct list_head typ_chain;
164 struct obd_ops *typ_dt_ops;
165 struct md_ops *typ_md_ops;
166 struct proc_dir_entry *typ_procroot;
167 struct proc_dir_entry *typ_procsym;
168 __u32 typ_sym_filter;
171 struct lu_device_type *typ_lu;
172 spinlock_t obd_type_lock;
182 struct timeout_item {
183 enum timeout_event ti_event;
184 cfs_time_t ti_timeout;
187 struct list_head ti_obd_list;
188 struct list_head ti_chain;
191 #define OBD_MAX_RIF_DEFAULT 8
192 #define OBD_MAX_RIF_MAX 512
193 #define OSC_MAX_RIF_MAX 256
194 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
195 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
196 #define OSC_DEFAULT_RESENDS 10
198 /* possible values for fo_sync_lock_cancel */
200 NEVER_SYNC_ON_CANCEL = 0,
201 BLOCKING_SYNC_ON_CANCEL = 1,
202 ALWAYS_SYNC_ON_CANCEL = 2,
203 NUM_SYNC_ON_CANCEL_STATES
207 * Limit reply buffer size for striping data to one x86_64 page. This
208 * value is chosen to fit the striping data for common use cases while
209 * staying well below the limit at which the buffer must be backed by
210 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
213 #define OBD_MAX_DEFAULT_EA_SIZE 4096
218 struct rw_semaphore cl_sem;
219 struct obd_uuid cl_target_uuid;
220 struct obd_import *cl_import; /* ptlrpc connection state */
221 size_t cl_conn_count;
223 /* Cache maximum and default values for easize. This is
224 * strictly a performance optimization to minimize calls to
225 * obd_size_diskmd(). The default values are used to calculate the
226 * initial size of a request buffer. The ptlrpc layer will resize the
227 * buffer as needed to accommodate a larger reply from the
228 * server. The default values should be small enough to avoid wasted
229 * memory and excessive use of vmalloc(), yet large enough to avoid
230 * reallocating the buffer in the common use case. */
232 /* Default EA size for striping attributes. It is initialized at
233 * mount-time based on the default stripe width of the filesystem,
234 * then it tracks the largest observed EA size advertised by
235 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
236 __u32 cl_default_mds_easize;
238 /* Maximum possible EA size computed at mount-time based on
239 * the number of OSTs in the filesystem. May be increased at
240 * run-time if a larger observed size is advertised by the MDT. */
241 __u32 cl_max_mds_easize;
243 enum lustre_sec_part cl_sp_me;
244 enum lustre_sec_part cl_sp_to;
245 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
247 /* the grant values are protected by loi_list_lock below */
248 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
249 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
250 unsigned long cl_dirty_transit; /* dirty synchronous */
251 unsigned long cl_avail_grant; /* bytes of credit for ost */
252 unsigned long cl_lost_grant; /* lost credits (trunc) */
254 /* since we allocate grant by blocks, we don't know how many grant will
255 * be used to add a page into cache. As a solution, we reserve maximum
256 * grant before trying to dirty a page and unreserve the rest.
257 * See osc_{reserve|unreserve}_grant for details. */
258 long cl_reserved_grant;
259 struct list_head cl_cache_waiters; /* waiting for cache/grant */
260 cfs_time_t cl_next_shrink_grant; /* jiffies */
261 struct list_head cl_grant_shrink_list; /* Timeout event list */
262 int cl_grant_shrink_interval; /* seconds */
264 /* A chunk is an optimal size used by osc_extent to determine
265 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
267 unsigned int cl_extent_tax; /* extent overhead, by bytes */
269 /* keep track of objects that have lois that contain pages which
270 * have been queued for async brw. this lock also protects the
271 * lists of osc_client_pages that hang off of the loi */
273 * ->cl_loi_list_lock protects consistency of
274 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
275 * ->ap_completion() call-backs are executed under this lock. As we
276 * cannot guarantee that these call-backs never block on all platforms
277 * (as a matter of fact they do block on Mac OS X), type of
278 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
279 * and blocking mutex on Mac OS X. (Alternative is to make this lock
280 * blocking everywhere, but we don't want to slow down fast-path of
281 * our main platform.)
283 * NB by Jinshan: though field names are still _loi_, but actually
284 * osc_object{}s are in the list.
286 spinlock_t cl_loi_list_lock;
287 struct list_head cl_loi_ready_list;
288 struct list_head cl_loi_hp_ready_list;
289 struct list_head cl_loi_write_list;
290 struct list_head cl_loi_read_list;
291 __u32 cl_r_in_flight;
292 __u32 cl_w_in_flight;
293 /* just a sum of the loi/lop pending numbers to be exported by /proc */
294 atomic_t cl_pending_w_pages;
295 atomic_t cl_pending_r_pages;
296 __u32 cl_max_pages_per_rpc;
297 __u32 cl_max_rpcs_in_flight;
298 struct obd_histogram cl_read_rpc_hist;
299 struct obd_histogram cl_write_rpc_hist;
300 struct obd_histogram cl_read_page_hist;
301 struct obd_histogram cl_write_page_hist;
302 struct obd_histogram cl_read_offset_hist;
303 struct obd_histogram cl_write_offset_hist;
305 /* lru for osc caching pages */
306 struct cl_client_cache *cl_cache;
307 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
308 atomic_long_t *cl_lru_left;
309 atomic_long_t cl_lru_busy;
310 atomic_long_t cl_lru_in_list;
311 atomic_long_t cl_unstable_count;
312 struct list_head cl_lru_list; /* lru page list */
313 spinlock_t cl_lru_list_lock; /* page list protector */
314 atomic_t cl_lru_shrinkers;
316 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
317 atomic_t cl_destroy_in_flight;
318 wait_queue_head_t cl_destroy_waitq;
320 struct mdc_rpc_lock *cl_rpc_lock;
321 struct mdc_rpc_lock *cl_close_lock;
323 /* modify rpcs in flight
324 * currently used for metadata only */
325 spinlock_t cl_mod_rpcs_lock;
326 __u16 cl_max_mod_rpcs_in_flight;
329 struct mutex cl_mgc_mutex;
330 struct local_oid_storage *cl_mgc_los;
331 struct dt_object *cl_mgc_configs_dir;
332 atomic_t cl_mgc_refcount;
333 struct obd_export *cl_mgc_mgsexp;
335 /* checksumming for data sent over the network */
336 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
337 /* supported checksum types that are worked out at connect time */
338 __u32 cl_supp_cksum_types;
339 /* checksum algorithm to be used */
340 cksum_type_t cl_cksum_type;
342 /* also protected by the poorly named _loi_list_lock lock above */
343 struct osc_async_rc cl_ar;
345 /* sequence manager */
346 struct lu_client_seq *cl_seq;
348 atomic_t cl_resends; /* resend count */
350 /* ptlrpc work for writeback in ptlrpcd context */
351 void *cl_writeback_work;
353 /* hash tables for osc_quota_info */
354 cfs_hash_t *cl_quota_hash[MAXQUOTAS];
356 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
363 struct echo_client_obd {
364 struct obd_export *ec_exp; /* the local connection to osc/lov */
366 struct list_head ec_objects;
367 struct list_head ec_locks;
371 /* Generic subset of OSTs */
373 __u32 *op_array; /* array of index of
375 unsigned int op_count; /* number of OSTs in the array */
376 unsigned int op_size; /* allocated size of lp_array */
377 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
380 /* allow statfs data caching for 1 second */
381 #define OBD_STATFS_CACHE_SECONDS 1
383 struct lov_tgt_desc {
384 struct list_head ltd_kill;
385 struct obd_uuid ltd_uuid;
386 struct obd_device *ltd_obd;
387 struct obd_export *ltd_exp;
389 __u32 ltd_index; /* index in lov_obd->tgts */
390 unsigned long ltd_active:1,/* is this target up for requests */
391 ltd_activate:1,/* should target be activated */
392 ltd_reap:1; /* should this target be deleted */
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 cfs_hash_t *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;
419 struct lmv_tgt_desc {
420 struct obd_uuid ltd_uuid;
421 struct obd_export *ltd_exp;
423 struct mutex ltd_fid_mutex;
424 unsigned long ltd_active:1; /* target up for requests */
427 enum placement_policy {
428 PLACEMENT_CHAR_POLICY = 0,
429 PLACEMENT_NID_POLICY = 1,
430 PLACEMENT_INVAL_POLICY = 2,
434 typedef enum placement_policy placement_policy_t;
438 struct lu_client_fld lmv_fld;
440 placement_policy_t lmv_placement;
441 struct lmv_desc desc;
442 struct obd_uuid cluuid;
443 struct obd_export *exp;
444 struct proc_dir_entry *targets_proc_entry;
446 struct mutex lmv_init_mutex;
451 __u32 tgts_size; /* size of tgts array */
452 struct lmv_tgt_desc **tgts;
454 struct obd_connect_data conn_data;
457 struct niobuf_local {
458 __u64 lnb_file_offset;
459 __u32 lnb_page_offset;
462 struct page *lnb_page;
467 #define LUSTRE_FLD_NAME "fld"
468 #define LUSTRE_SEQ_NAME "seq"
470 #define LUSTRE_MDD_NAME "mdd"
471 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
472 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
473 #define LUSTRE_VVP_NAME "vvp"
474 #define LUSTRE_LMV_NAME "lmv"
475 #define LUSTRE_SLP_NAME "slp"
476 #define LUSTRE_LOD_NAME "lod"
477 #define LUSTRE_OSP_NAME "osp"
478 #define LUSTRE_LWP_NAME "lwp"
480 /* obd device type names */
481 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
482 #define LUSTRE_MDS_NAME "mds"
483 #define LUSTRE_MDT_NAME "mdt"
484 #define LUSTRE_MDC_NAME "mdc"
485 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
486 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
487 #define LUSTRE_OSC_NAME "osc"
488 #define LUSTRE_LOV_NAME "lov"
489 #define LUSTRE_MGS_NAME "mgs"
490 #define LUSTRE_MGC_NAME "mgc"
492 #define LUSTRE_ECHO_NAME "obdecho"
493 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
494 #define LUSTRE_QMT_NAME "qmt"
496 /* Constant obd names (post-rename) */
497 #define LUSTRE_MDS_OBDNAME "MDS"
498 #define LUSTRE_OSS_OBDNAME "OSS"
499 #define LUSTRE_MGS_OBDNAME "MGS"
500 #define LUSTRE_MGC_OBDNAME "MGC"
503 * Events signalled through obd_notify() upcall-chain.
505 enum obd_notify_event {
508 /* Device connect start */
510 /* Device activated */
512 /* Device deactivated */
514 /* Device disconnected */
516 /* Connect data for import were changed */
519 OBD_NOTIFY_SYNC_NONBLOCK,
521 /* Configuration event */
523 /* Administratively deactivate/activate event */
524 OBD_NOTIFY_DEACTIVATE,
528 /* bit-mask flags for config events */
530 CONFIG_LOG = 0x1, /* finished processing config log */
531 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
532 CONFIG_TARGET = 0x4 /* one target is added */
536 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
537 * and liblustre being main examples).
539 struct obd_notify_upcall {
540 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
541 enum obd_notify_event ev, void *owner, void *data);
542 /* Opaque datum supplied by upper layer listener */
546 struct target_recovery_data {
547 svc_handler_t trd_recovery_handler;
548 pid_t trd_processing_task;
549 struct completion trd_starting;
550 struct completion trd_finishing;
553 struct obd_llog_group {
554 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
555 wait_queue_head_t olg_waitq;
557 struct mutex olg_cat_processing;
560 /* corresponds to one of the obd's */
561 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
564 struct obd_type *obd_type;
567 /* common and UUID name of this device */
568 char obd_name[MAX_OBD_NAME];
569 struct obd_uuid obd_uuid;
571 struct lu_device *obd_lu_dev;
573 /* bitfield modification is protected by obd_dev_lock */
575 obd_attached:1, /* finished attach */
576 obd_set_up:1, /* finished setup */
577 obd_recovering:1, /* there are recoverable clients */
578 obd_abort_recovery:1, /* recovery expired */
579 obd_version_recov:1, /* obd uses version checking */
580 obd_replayable:1, /* recovery is enabled;
582 obd_no_transno:1, /* no committed-transno notification */
583 obd_no_recov:1, /* fail instead of retry messages */
584 obd_stopping:1, /* started cleanup */
585 obd_starting:1, /* started setup */
586 obd_force:1, /* cleanup with > 0 obd refcount */
587 obd_fail:1, /* cleanup with failover */
588 obd_no_conn:1, /* deny new connections */
589 obd_inactive:1, /* device active/inactive
590 * (for /proc/status only!!) */
591 obd_no_ir:1, /* no imperative recovery. */
592 obd_process_conf:1, /* device is processing mgs config */
593 obd_uses_nid_stats:1; /* maintain per-client OBD stats */
595 /* use separate field as it is set in interrupt to don't mess with
596 * protection of other bits using _bh lock */
597 unsigned long obd_recovery_expired:1;
598 /* uuid-export hash body */
599 cfs_hash_t *obd_uuid_hash;
600 /* nid-export hash body */
601 cfs_hash_t *obd_nid_hash;
603 cfs_hash_t *obd_nid_stats_hash;
604 struct list_head obd_nid_stats;
605 atomic_t obd_refcount;
606 struct list_head obd_exports;
607 struct list_head obd_unlinked_exports;
608 struct list_head obd_delayed_exports;
609 struct list_head obd_lwp_list;
611 spinlock_t obd_nid_lock;
612 struct ldlm_namespace *obd_namespace;
613 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
614 /* a spinlock is OK for what we do now, may need a semaphore later */
615 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
616 struct mutex obd_dev_mutex;
617 __u64 obd_last_committed;
618 spinlock_t obd_osfs_lock;
619 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
621 struct lvfs_run_ctxt obd_lvfs_ctxt;
622 struct obd_llog_group obd_olg; /* default llog group */
623 struct obd_device *obd_observer;
624 struct rw_semaphore obd_observer_link_sem;
625 struct obd_notify_upcall obd_upcall;
626 struct obd_export *obd_self_export;
627 struct obd_export *obd_lwp_export;
628 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
629 struct list_head obd_exports_timed;
630 time_t obd_eviction_timer; /* for ping evictor */
632 int obd_max_recoverable_clients;
633 atomic_t obd_connected_clients;
634 int obd_stale_clients;
635 /* this lock protects all recovery list_heads, timer and
636 * obd_next_recovery_transno value */
637 spinlock_t obd_recovery_task_lock;
638 __u64 obd_next_recovery_transno;
639 int obd_replayed_requests;
640 int obd_requests_queued_for_recovery;
641 wait_queue_head_t obd_next_transno_waitq;
642 /* protected by obd_recovery_task_lock */
643 struct timer_list obd_recovery_timer;
645 time_t obd_recovery_start;
646 /* seconds, for lprocfs_status */
647 time_t obd_recovery_end;
648 int obd_recovery_time_hard;
649 int obd_recovery_timeout;
650 int obd_recovery_ir_factor;
652 /* new recovery stuff from CMD2 */
653 struct target_recovery_data obd_recovery_data;
654 int obd_replayed_locks;
655 atomic_t obd_req_replay_clients;
656 atomic_t obd_lock_replay_clients;
657 /* all lists are protected by obd_recovery_task_lock */
658 struct list_head obd_req_replay_queue;
659 struct list_head obd_lock_replay_queue;
660 struct list_head obd_final_req_queue;
663 #ifdef HAVE_SERVER_SUPPORT
664 struct obd_device_target obt;
665 struct filter_obd filter;
667 struct echo_obd echo;
669 struct client_obd cli;
670 struct echo_client_obd echo_client;
674 /* Fields used by LProcFS */
675 unsigned int obd_cntr_base;
676 struct lprocfs_stats *obd_stats;
678 unsigned int obd_md_cntr_base;
679 struct lprocfs_stats *obd_md_stats;
681 struct proc_dir_entry *obd_proc_entry;
682 struct proc_dir_entry *obd_proc_exports_entry;
683 struct proc_dir_entry *obd_svc_procroot;
684 struct lprocfs_stats *obd_svc_stats;
685 struct lprocfs_vars *obd_vars;
686 atomic_t obd_evict_inprogress;
687 wait_queue_head_t obd_evict_inprogress_waitq;
688 struct list_head obd_evict_list; /* protected with pet_lock */
691 * Ldlm pool part. Save last calculated SLV and Limit.
693 rwlock_t obd_pool_lock;
698 * A list of outstanding class_incref()'s against this obd. For
701 struct lu_ref obd_reference;
703 int obd_conn_inprogress;
706 enum obd_cleanup_stage {
707 /* Special case hack for MDS LOVs */
709 /* can be directly mapped to .ldto_device_fini() */
713 /* get/set_info keys */
714 #define KEY_ASYNC "async"
715 #define KEY_CAPA_KEY "capa_key"
716 #define KEY_CHANGELOG_CLEAR "changelog_clear"
717 #define KEY_FID2PATH "fid2path"
718 #define KEY_CHECKSUM "checksum"
719 #define KEY_CLEAR_FS "clear_fs"
720 #define KEY_CONN_DATA "conn_data"
721 #define KEY_EVICT_BY_NID "evict_by_nid"
722 #define KEY_FIEMAP "fiemap"
723 #define KEY_FLUSH_CTX "flush_ctx"
724 #define KEY_GRANT_SHRINK "grant_shrink"
725 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
726 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
727 #define KEY_INTERMDS "inter_mds"
728 #define KEY_LAST_ID "last_id"
729 #define KEY_LAST_FID "last_fid"
730 #define KEY_MAX_EASIZE "max_easize"
731 #define KEY_DEFAULT_EASIZE "default_easize"
732 #define KEY_MGSSEC "mgssec"
733 #define KEY_READ_ONLY "read-only"
734 #define KEY_REGISTER_TARGET "register_target"
735 #define KEY_SET_FS "set_fs"
736 #define KEY_TGT_COUNT "tgt_count"
737 /* KEY_SET_INFO in lustre_idl.h */
738 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
740 #define KEY_CACHE_SET "cache_set"
741 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
742 #define KEY_OSP_CONNECTED "osp_connected"
746 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
747 #define IT_OPEN (1 << 0)
748 #define IT_CREAT (1 << 1)
749 #define IT_READDIR (1 << 2)
750 #define IT_GETATTR (1 << 3)
751 #define IT_LOOKUP (1 << 4)
752 #define IT_UNLINK (1 << 5)
753 #define IT_TRUNC (1 << 6)
754 #define IT_GETXATTR (1 << 7)
755 #define IT_EXEC (1 << 8)
756 #define IT_PIN (1 << 9)
757 #define IT_LAYOUT (1 << 10)
758 #define IT_QUOTA_DQACQ (1 << 11)
759 #define IT_QUOTA_CONN (1 << 12)
760 #define IT_SETXATTR (1 << 13)
762 static inline int it_to_lock_mode(struct lookup_intent *it)
764 /* CREAT needs to be tested before open (both could be set) */
765 if (it->it_op & IT_CREAT)
767 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
770 else if (it->it_op & IT_READDIR)
772 else if (it->it_op & IT_GETXATTR)
774 else if (it->it_op & IT_SETXATTR)
777 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
782 MF_MDC_CANCEL_FID1 = 1 << 0,
783 MF_MDC_CANCEL_FID2 = 1 << 1,
784 MF_MDC_CANCEL_FID3 = 1 << 2,
785 MF_MDC_CANCEL_FID4 = 1 << 3,
786 MF_GET_MDT_IDX = 1 << 4,
790 CLI_SET_MEA = 1 << 0,
791 CLI_RM_ENTRY = 1 << 1,
794 CLI_MIGRATE = 1 << 4,
798 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
799 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
800 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
801 struct lu_fid op_fid4; /* to the operation locks. */
802 u32 op_mds; /* what mds server open will go to */
803 struct lustre_handle op_handle;
808 struct lmv_stripe_md *op_mea1;
809 struct lmv_stripe_md *op_mea2;
810 __u32 op_suppgids[2];
817 /* iattr fields and blocks. */
818 struct iattr op_attr;
819 loff_t op_attr_blocks;
820 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
821 __u64 op_valid; /* OBD_MD_* */
823 enum md_op_flags op_flags;
826 struct obd_capa *op_capa1;
827 struct obd_capa *op_capa2;
829 /* Various operation flags. */
830 enum mds_op_bias op_bias;
832 /* Used by readdir */
833 unsigned int op_max_pages;
835 /* used to transfer info between the stacks of MD client
836 * see enum op_cli_flags */
837 enum md_cli_flags op_cli_flags;
839 /* File object data version for HSM release, on client */
840 __u64 op_data_version;
841 struct lustre_handle op_lease_handle;
843 /* default stripe offset */
844 __u32 op_default_stripe_offset;
848 int (*md_blocking_ast)(struct ldlm_lock *lock,
849 struct ldlm_lock_desc *desc,
850 void *data, int flag);
853 struct md_enqueue_info;
854 /* metadata stat-ahead */
855 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
856 struct md_enqueue_info *minfo,
859 struct md_enqueue_info {
860 struct md_op_data mi_data;
861 struct lookup_intent mi_it;
862 struct lustre_handle mi_lockh;
863 struct inode *mi_dir;
864 md_enqueue_cb_t mi_cb;
869 struct module *o_owner;
870 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
871 void *karg, void __user *uarg);
872 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
873 __u32 keylen, void *key, __u32 *vallen, void *val);
874 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
875 __u32 keylen, void *key,
876 __u32 vallen, void *val,
877 struct ptlrpc_request_set *set);
878 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
879 int (*o_precleanup)(struct obd_device *dev,
880 enum obd_cleanup_stage cleanup_stage);
881 int (*o_cleanup)(struct obd_device *dev);
882 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
883 int (*o_postrecov)(struct obd_device *dev);
884 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
886 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
887 /* connect to the target device with given connection
888 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
889 * granted by the target, which are guaranteed to be a subset of flags
890 * asked for. If @ocd == NULL, use default parameters. */
891 int (*o_connect)(const struct lu_env *env,
892 struct obd_export **exp, struct obd_device *src,
893 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
895 int (*o_reconnect)(const struct lu_env *env,
896 struct obd_export *exp, struct obd_device *src,
897 struct obd_uuid *cluuid,
898 struct obd_connect_data *ocd,
900 int (*o_disconnect)(struct obd_export *exp);
902 /* Initialize/finalize fids infrastructure. */
903 int (*o_fid_init)(struct obd_device *obd,
904 struct obd_export *exp, enum lu_cli_type type);
905 int (*o_fid_fini)(struct obd_device *obd);
907 /* Allocate new fid according to passed @hint. */
908 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
909 struct lu_fid *fid, struct md_op_data *op_data);
912 * Object with @fid is getting deleted, we may want to do something
915 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
916 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
917 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
918 __u64 max_age, struct ptlrpc_request_set *set);
919 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
920 struct lov_stripe_md *mem_src);
921 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
922 struct lov_mds_md *disk_src, int disk_len);
923 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
925 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
927 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
928 struct obd_info *oinfo);
929 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
930 struct obd_info *oinfo);
931 int (*o_preprw)(const struct lu_env *env, int cmd,
932 struct obd_export *exp, struct obdo *oa, int objcount,
933 struct obd_ioobj *obj, struct niobuf_remote *remote,
934 int *nr_pages, struct niobuf_local *local);
935 int (*o_commitrw)(const struct lu_env *env, int cmd,
936 struct obd_export *exp, struct obdo *oa,
937 int objcount, struct obd_ioobj *obj,
938 struct niobuf_remote *remote, int pages,
939 struct niobuf_local *local, int rc);
940 int (*o_init_export)(struct obd_export *exp);
941 int (*o_destroy_export)(struct obd_export *exp);
943 int (*o_import_event)(struct obd_device *, struct obd_import *,
944 enum obd_import_event);
946 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
947 enum obd_notify_event ev, void *data);
949 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
950 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
953 int (*o_quotactl)(struct obd_device *, struct obd_export *,
954 struct obd_quotactl *);
956 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
959 int (*o_pool_new)(struct obd_device *obd, char *poolname);
960 int (*o_pool_del)(struct obd_device *obd, char *poolname);
961 int (*o_pool_add)(struct obd_device *obd, char *poolname,
963 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
965 void (*o_getref)(struct obd_device *obd);
966 void (*o_putref)(struct obd_device *obd);
968 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
969 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
970 * Also, add a wrapper function in include/linux/obd_class.h. */
975 struct mdt_body *body;
976 struct lov_stripe_md *lsm;
977 struct lmv_stripe_md *lmv;
978 #ifdef CONFIG_FS_POSIX_ACL
979 struct posix_acl *posix_acl;
981 struct mdt_remote_perm *remote_perm;
982 struct obd_capa *mds_capa;
983 struct obd_capa *oss_capa;
986 struct md_open_data {
987 struct obd_client_handle *mod_och;
988 struct ptlrpc_request *mod_open_req;
989 struct ptlrpc_request *mod_close_req;
990 atomic_t mod_refcount;
994 struct obd_client_handle {
995 struct lustre_handle och_fh;
996 struct lu_fid och_fid;
997 struct md_open_data *och_mod;
998 struct lustre_handle och_lease_handle; /* open lock for lease */
1003 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1005 struct lookup_intent;
1009 /* Every operation from MD_STATS_FIRST_OP up to and including
1010 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1011 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1012 * Operations after MD_STATS_LAST_OP are excluded from stats.
1013 * There are a few reasons for doing this: we prune the 17
1014 * counters which will be of minimal use in understanding
1015 * metadata utilization, we save memory by allocating 15
1016 * instead of 32 counters, we save cycles by not counting.
1018 * MD_STATS_FIRST_OP must be the first member of md_ops.
1020 #define MD_STATS_FIRST_OP m_close
1021 int (*m_close)(struct obd_export *, struct md_op_data *,
1022 struct md_open_data *, struct ptlrpc_request **);
1024 int (*m_create)(struct obd_export *, struct md_op_data *,
1025 const void *, size_t, umode_t, uid_t, gid_t,
1026 cfs_cap_t, __u64, struct ptlrpc_request **);
1028 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1029 const union ldlm_policy_data *,
1030 struct lookup_intent *, struct md_op_data *,
1031 struct lustre_handle *, __u64);
1033 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1034 struct ptlrpc_request **);
1036 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1037 struct lookup_intent *,
1038 struct ptlrpc_request **,
1039 ldlm_blocking_callback, __u64);
1041 int (*m_link)(struct obd_export *, struct md_op_data *,
1042 struct ptlrpc_request **);
1044 int (*m_rename)(struct obd_export *, struct md_op_data *,
1045 const char *, size_t, const char *, size_t,
1046 struct ptlrpc_request **);
1048 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1049 size_t , struct ptlrpc_request **);
1051 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1052 struct obd_capa *, struct ptlrpc_request **);
1054 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1055 struct md_callback *cb_op, __u64 hash_offset,
1056 struct page **ppage);
1058 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1059 struct ptlrpc_request **);
1061 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1062 struct obd_capa *, u64, const char *,
1063 const char *, int, int, int, __u32,
1064 struct ptlrpc_request **);
1066 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1067 struct obd_capa *, u64, const char *,
1068 const char *, int, int, int,
1069 struct ptlrpc_request **);
1071 int (*m_intent_getattr_async)(struct obd_export *,
1072 struct md_enqueue_info *,
1073 struct ldlm_enqueue_info *);
1075 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1076 struct lu_fid *, __u64 *bits);
1078 #define MD_STATS_LAST_OP m_revalidate_lock
1080 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1081 struct obd_capa **);
1083 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1085 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1086 ldlm_iterator_t, void *);
1088 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1089 struct ptlrpc_request **);
1091 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1093 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1094 struct obd_export *, struct obd_export *,
1095 struct lustre_md *);
1097 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1099 int (*m_merge_attr)(struct obd_export *,
1100 const struct lmv_stripe_md *lsm,
1101 struct cl_attr *attr, ldlm_blocking_callback);
1103 int (*m_set_open_replay_data)(struct obd_export *,
1104 struct obd_client_handle *,
1105 struct lookup_intent *);
1107 int (*m_clear_open_replay_data)(struct obd_export *,
1108 struct obd_client_handle *);
1110 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1112 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1113 const struct lu_fid *, ldlm_type_t,
1114 ldlm_policy_data_t *, ldlm_mode_t,
1115 struct lustre_handle *);
1117 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1118 ldlm_policy_data_t *, ldlm_mode_t,
1119 ldlm_cancel_flags_t flags, void *opaque);
1121 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1122 renew_capa_cb_t cb);
1124 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1125 const struct req_msg_field *, struct obd_capa **);
1127 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1128 struct obd_capa *, __u32,
1129 struct ptlrpc_request **);
1131 int (*m_get_fid_from_lsm)(struct obd_export *,
1132 const struct lmv_stripe_md *,
1133 const char *name, int namelen,
1134 struct lu_fid *fid);
1137 struct lsm_operations {
1138 void (*lsm_free)(struct lov_stripe_md *);
1139 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *,
1140 loff_t *, loff_t *);
1141 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *,
1142 loff_t *, loff_t *);
1143 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1144 __u16 *stripe_count);
1145 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1146 struct lov_mds_md *lmm);
1149 extern const struct lsm_operations lsm_v1_ops;
1150 extern const struct lsm_operations lsm_v3_ops;
1151 static inline const struct lsm_operations *lsm_op_find(u32 magic)
1159 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1164 static inline struct md_open_data *obd_mod_alloc(void)
1166 struct md_open_data *mod;
1170 atomic_set(&mod->mod_refcount, 1);
1174 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1175 #define obd_mod_put(mod) \
1177 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1178 if ((mod)->mod_open_req) \
1179 ptlrpc_req_finished((mod)->mod_open_req); \
1180 OBD_FREE_PTR(mod); \
1184 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1185 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1187 /* return 1 if client should be resend request */
1188 static inline int client_should_resend(int resend, struct client_obd *cli)
1190 return atomic_read(&cli->cl_resends) ?
1191 atomic_read(&cli->cl_resends) > resend : 1;
1195 * Return device name for this device
1197 * XXX: lu_device is declared before obd_device, while a pointer pointing
1198 * back to obd_device in lu_device, so this helper function defines here
1199 * instead of in lu_object.h
1201 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1203 return lu_dev->ld_obd->obd_name;
1206 static inline bool filename_is_volatile(const char *name, size_t namelen,
1212 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1215 /* caller does not care of idx */
1219 /* volatile file, the MDT can be set from name */
1220 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1221 /* if no MDT is specified, use std way */
1222 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1224 /* test for no MDT idx case */
1225 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1226 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1230 /* we have an idx, read it */
1231 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1232 *idx = simple_strtoul(start, &end, 16);
1234 * no digit, no trailing :, negative value
1236 if (((*idx == 0) && (end == start)) ||
1237 (*end != ':') || (*idx < 0))
1242 /* bad format of mdt idx, we cannot return an error
1243 * to caller so we use hash algo */
1244 CERROR("Bad volatile file name format: %s\n",
1245 name + LUSTRE_VOLATILE_HDR_LEN);
1249 static inline int cli_brw_size(struct obd_device *obd)
1251 LASSERT(obd != NULL);
1252 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1255 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1256 * of the client should be increased accordingly to avoid sending fragmented
1257 * RPCs over the network when the client runs out of the maximum dirty space
1258 * when so many RPCs are being generated.
1260 static inline void client_adjust_max_dirty(struct client_obd *cli)
1263 if (cli->cl_dirty_max_pages <= 0)
1264 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1265 >> PAGE_CACHE_SHIFT;
1267 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1268 cli->cl_max_pages_per_rpc;
1270 if (dirty_max > cli->cl_dirty_max_pages)
1271 cli->cl_dirty_max_pages = dirty_max;
1274 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1275 cli->cl_dirty_max_pages = totalram_pages / 8;
1278 #endif /* __OBD_H */