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>
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;
82 static inline void loi_init(struct lov_oinfo *loi)
89 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
91 /* obd info for a particular level (lov, osc). */
93 /* OBD_STATFS_* flags */
95 /* statfs data specific for every OSC, if needed at all. */
96 struct obd_statfs *oi_osfs;
97 /* An update callback which is called to update some data on upper
98 * level. E.g. it is used for update lsm->lsm_oinfo at every received
99 * request in osc level for enqueue requests. It is also possible to
100 * update some caller data from LOV layer if needed. */
101 obd_enqueue_update_f oi_cb_up;
105 struct list_head typ_chain;
106 struct obd_ops *typ_dt_ops;
107 struct md_ops *typ_md_ops;
108 struct proc_dir_entry *typ_procroot;
109 struct proc_dir_entry *typ_procsym;
110 __u32 typ_sym_filter;
113 struct lu_device_type *typ_lu;
114 spinlock_t obd_type_lock;
124 struct timeout_item {
125 enum timeout_event ti_event;
126 cfs_time_t ti_timeout;
129 struct list_head ti_obd_list;
130 struct list_head ti_chain;
133 #define OBD_MAX_RIF_DEFAULT 8
134 #define OBD_MAX_RIF_MAX 512
135 #define OSC_MAX_RIF_MAX 256
136 #define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
137 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
138 #define OSC_DEFAULT_RESENDS 10
140 /* possible values for fo_sync_lock_cancel */
142 NEVER_SYNC_ON_CANCEL = 0,
143 BLOCKING_SYNC_ON_CANCEL = 1,
144 ALWAYS_SYNC_ON_CANCEL = 2,
145 NUM_SYNC_ON_CANCEL_STATES
149 * Limit reply buffer size for striping data to one x86_64 page. This
150 * value is chosen to fit the striping data for common use cases while
151 * staying well below the limit at which the buffer must be backed by
152 * vmalloc(). Excessive use of vmalloc() may cause spinlock contention
155 #define OBD_MAX_DEFAULT_EA_SIZE 4096
160 struct rw_semaphore cl_sem;
161 struct obd_uuid cl_target_uuid;
162 struct obd_import *cl_import; /* ptlrpc connection state */
163 size_t cl_conn_count;
165 /* Cache maximum and default values for easize. This is
166 * strictly a performance optimization to minimize calls to
167 * obd_size_diskmd(). The default values are used to calculate the
168 * initial size of a request buffer. The ptlrpc layer will resize the
169 * buffer as needed to accommodate a larger reply from the
170 * server. The default values should be small enough to avoid wasted
171 * memory and excessive use of vmalloc(), yet large enough to avoid
172 * reallocating the buffer in the common use case. */
174 /* Default EA size for striping attributes. It is initialized at
175 * mount-time based on the default stripe width of the filesystem,
176 * then it tracks the largest observed EA size advertised by
177 * the MDT, up to a maximum value of OBD_MAX_DEFAULT_EA_SIZE. */
178 __u32 cl_default_mds_easize;
180 /* Maximum possible EA size computed at mount-time based on
181 * the number of OSTs in the filesystem. May be increased at
182 * run-time if a larger observed size is advertised by the MDT. */
183 __u32 cl_max_mds_easize;
185 enum lustre_sec_part cl_sp_me;
186 enum lustre_sec_part cl_sp_to;
187 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
189 /* the grant values are protected by loi_list_lock below */
190 unsigned long cl_dirty_pages; /* all _dirty_ in pages */
191 unsigned long cl_dirty_max_pages; /* allowed w/o rpc */
192 unsigned long cl_dirty_transit; /* dirty synchronous */
193 unsigned long cl_avail_grant; /* bytes of credit for ost */
194 unsigned long cl_lost_grant; /* lost credits (trunc) */
196 /* since we allocate grant by blocks, we don't know how many grant will
197 * be used to add a page into cache. As a solution, we reserve maximum
198 * grant before trying to dirty a page and unreserve the rest.
199 * See osc_{reserve|unreserve}_grant for details. */
200 long cl_reserved_grant;
201 struct list_head cl_cache_waiters; /* waiting for cache/grant */
202 cfs_time_t cl_next_shrink_grant; /* jiffies */
203 struct list_head cl_grant_shrink_list; /* Timeout event list */
204 int cl_grant_shrink_interval; /* seconds */
206 /* A chunk is an optimal size used by osc_extent to determine
207 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
209 unsigned int cl_extent_tax; /* extent overhead, by bytes */
211 /* keep track of objects that have lois that contain pages which
212 * have been queued for async brw. this lock also protects the
213 * lists of osc_client_pages that hang off of the loi */
215 * ->cl_loi_list_lock protects consistency of
216 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
217 * ->ap_completion() call-backs are executed under this lock. As we
218 * cannot guarantee that these call-backs never block on all platforms
219 * (as a matter of fact they do block on Mac OS X), type of
220 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
221 * and blocking mutex on Mac OS X. (Alternative is to make this lock
222 * blocking everywhere, but we don't want to slow down fast-path of
223 * our main platform.)
225 * NB by Jinshan: though field names are still _loi_, but actually
226 * osc_object{}s are in the list.
228 spinlock_t cl_loi_list_lock;
229 struct list_head cl_loi_ready_list;
230 struct list_head cl_loi_hp_ready_list;
231 struct list_head cl_loi_write_list;
232 struct list_head cl_loi_read_list;
233 __u32 cl_r_in_flight;
234 __u32 cl_w_in_flight;
235 /* just a sum of the loi/lop pending numbers to be exported by /proc */
236 atomic_t cl_pending_w_pages;
237 atomic_t cl_pending_r_pages;
238 __u32 cl_max_pages_per_rpc;
239 __u32 cl_max_rpcs_in_flight;
240 struct obd_histogram cl_read_rpc_hist;
241 struct obd_histogram cl_write_rpc_hist;
242 struct obd_histogram cl_read_page_hist;
243 struct obd_histogram cl_write_page_hist;
244 struct obd_histogram cl_read_offset_hist;
245 struct obd_histogram cl_write_offset_hist;
247 /* lru for osc caching pages */
248 struct cl_client_cache *cl_cache;
249 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
250 atomic_long_t *cl_lru_left;
251 atomic_long_t cl_lru_busy;
252 atomic_long_t cl_lru_in_list;
253 atomic_long_t cl_unstable_count;
254 struct list_head cl_lru_list; /* lru page list */
255 spinlock_t cl_lru_list_lock; /* page list protector */
256 atomic_t cl_lru_shrinkers;
258 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
259 atomic_t cl_destroy_in_flight;
260 wait_queue_head_t cl_destroy_waitq;
262 struct mdc_rpc_lock *cl_rpc_lock;
264 /* modify rpcs in flight
265 * currently used for metadata only */
266 spinlock_t cl_mod_rpcs_lock;
267 __u16 cl_max_mod_rpcs_in_flight;
268 __u16 cl_mod_rpcs_in_flight;
269 __u16 cl_close_rpcs_in_flight;
270 wait_queue_head_t cl_mod_rpcs_waitq;
271 unsigned long *cl_mod_tag_bitmap;
272 struct obd_histogram cl_mod_rpcs_hist;
275 struct mutex cl_mgc_mutex;
276 struct local_oid_storage *cl_mgc_los;
277 struct dt_object *cl_mgc_configs_dir;
278 atomic_t cl_mgc_refcount;
279 struct obd_export *cl_mgc_mgsexp;
281 /* checksumming for data sent over the network */
282 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
283 /* supported checksum types that are worked out at connect time */
284 __u32 cl_supp_cksum_types;
285 /* checksum algorithm to be used */
286 cksum_type_t cl_cksum_type;
288 /* also protected by the poorly named _loi_list_lock lock above */
289 struct osc_async_rc cl_ar;
291 /* sequence manager */
292 struct lu_client_seq *cl_seq;
294 atomic_t cl_resends; /* resend count */
296 /* ptlrpc work for writeback in ptlrpcd context */
297 void *cl_writeback_work;
299 /* hash tables for osc_quota_info */
300 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
302 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
309 struct echo_client_obd {
310 struct obd_export *ec_exp; /* the local connection to osc/lov */
312 struct list_head ec_objects;
313 struct list_head ec_locks;
317 /* Generic subset of OSTs */
319 __u32 *op_array; /* array of index of
321 unsigned int op_count; /* number of OSTs in the array */
322 unsigned int op_size; /* allocated size of lp_array */
323 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
326 /* allow statfs data caching for 1 second */
327 #define OBD_STATFS_CACHE_SECONDS 1
329 struct lov_tgt_desc {
330 struct list_head ltd_kill;
331 struct obd_uuid ltd_uuid;
332 struct obd_device *ltd_obd;
333 struct obd_export *ltd_exp;
335 __u32 ltd_index; /* index in lov_obd->tgts */
336 unsigned long ltd_active:1,/* is this target up for requests */
337 ltd_activate:1,/* should target be activated */
338 ltd_reap:1; /* should this target be deleted */
342 struct lov_desc desc;
343 struct lov_tgt_desc **lov_tgts; /* sparse array */
344 struct ost_pool lov_packed; /* all OSTs in a packed
346 struct mutex lov_lock;
347 struct obd_connect_data lov_ocd;
348 struct proc_dir_entry *targets_proc_entry;
349 atomic_t lov_refcount;
350 __u32 lov_death_row; /* tgts scheduled to be deleted */
351 __u32 lov_tgt_size; /* size of tgts array */
354 struct cfs_hash *lov_pools_hash_body; /* used for key access */
355 struct list_head lov_pool_list; /* used for sequential access */
356 struct proc_dir_entry *lov_pool_proc_entry;
357 enum lustre_sec_part lov_sp_me;
359 /* Cached LRU and unstable data from upper layer */
360 struct cl_client_cache *lov_cache;
362 struct rw_semaphore lov_notify_lock;
365 struct lmv_tgt_desc {
366 struct obd_uuid ltd_uuid;
367 struct obd_export *ltd_exp;
369 struct mutex ltd_fid_mutex;
370 unsigned long ltd_active:1; /* target up for requests */
373 enum placement_policy {
374 PLACEMENT_CHAR_POLICY = 0,
375 PLACEMENT_NID_POLICY = 1,
376 PLACEMENT_INVAL_POLICY = 2,
380 typedef enum placement_policy placement_policy_t;
384 struct lu_client_fld lmv_fld;
386 placement_policy_t lmv_placement;
387 struct lmv_desc desc;
388 struct obd_uuid cluuid;
389 struct obd_export *exp;
390 struct proc_dir_entry *targets_proc_entry;
392 struct mutex lmv_init_mutex;
397 __u32 tgts_size; /* size of tgts array */
398 struct lmv_tgt_desc **tgts;
400 struct obd_connect_data conn_data;
403 struct niobuf_local {
404 __u64 lnb_file_offset;
405 __u32 lnb_page_offset;
408 struct page *lnb_page;
413 #define LUSTRE_FLD_NAME "fld"
414 #define LUSTRE_SEQ_NAME "seq"
416 #define LUSTRE_MDD_NAME "mdd"
417 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
418 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
419 #define LUSTRE_VVP_NAME "vvp"
420 #define LUSTRE_LMV_NAME "lmv"
421 #define LUSTRE_SLP_NAME "slp"
422 #define LUSTRE_LOD_NAME "lod"
423 #define LUSTRE_OSP_NAME "osp"
424 #define LUSTRE_LWP_NAME "lwp"
426 /* obd device type names */
427 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
428 #define LUSTRE_MDS_NAME "mds"
429 #define LUSTRE_MDT_NAME "mdt"
430 #define LUSTRE_MDC_NAME "mdc"
431 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
432 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
433 #define LUSTRE_OSC_NAME "osc"
434 #define LUSTRE_LOV_NAME "lov"
435 #define LUSTRE_MGS_NAME "mgs"
436 #define LUSTRE_MGC_NAME "mgc"
438 #define LUSTRE_ECHO_NAME "obdecho"
439 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
440 #define LUSTRE_QMT_NAME "qmt"
442 /* Constant obd names (post-rename) */
443 #define LUSTRE_MDS_OBDNAME "MDS"
444 #define LUSTRE_OSS_OBDNAME "OSS"
445 #define LUSTRE_MGS_OBDNAME "MGS"
446 #define LUSTRE_MGC_OBDNAME "MGC"
448 static inline int is_lwp_on_mdt(char *name)
452 ptr = strrchr(name, '-');
454 CERROR("%s is not a obdname\n", name);
458 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
460 if (strncmp(ptr + 1, "MDT", 3) != 0)
463 while (*(--ptr) != '-' && ptr != name);
468 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
474 static inline int is_lwp_on_ost(char *name)
478 ptr = strrchr(name, '-');
480 CERROR("%s is not a obdname\n", name);
484 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
486 if (strncmp(ptr + 1, "OST", 3) != 0)
489 while (*(--ptr) != '-' && ptr != name);
494 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
501 * Events signalled through obd_notify() upcall-chain.
503 enum obd_notify_event {
506 /* Device connect start */
508 /* Device activated */
510 /* Device deactivated */
512 /* Device disconnected */
514 /* Connect data for import were changed */
517 OBD_NOTIFY_SYNC_NONBLOCK,
519 /* Configuration event */
521 /* Administratively deactivate/activate event */
522 OBD_NOTIFY_DEACTIVATE,
526 /* bit-mask flags for config events */
528 CONFIG_LOG = 0x1, /* finished processing config log */
529 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
530 CONFIG_TARGET = 0x4 /* one target is added */
534 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
535 * and liblustre being main examples).
537 struct obd_notify_upcall {
538 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
539 enum obd_notify_event ev, void *owner, void *data);
540 /* Opaque datum supplied by upper layer listener */
544 struct target_recovery_data {
545 svc_handler_t trd_recovery_handler;
546 pid_t trd_processing_task;
547 struct completion trd_starting;
548 struct completion trd_finishing;
551 struct obd_llog_group {
552 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
553 wait_queue_head_t olg_waitq;
555 struct mutex olg_cat_processing;
558 /* corresponds to one of the obd's */
559 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
562 struct obd_type *obd_type;
565 /* common and UUID name of this device */
566 char obd_name[MAX_OBD_NAME];
567 struct obd_uuid obd_uuid;
569 struct lu_device *obd_lu_dev;
571 /* bitfield modification is protected by obd_dev_lock */
573 obd_attached:1, /* finished attach */
574 obd_set_up:1, /* finished setup */
575 obd_recovering:1, /* there are recoverable clients */
576 obd_abort_recovery:1, /* recovery expired */
577 obd_version_recov:1, /* obd uses version checking */
578 obd_replayable:1, /* recovery is enabled;
580 obd_no_transno:1, /* no committed-transno notification */
581 obd_no_recov:1, /* fail instead of retry messages */
582 obd_stopping:1, /* started cleanup */
583 obd_starting:1, /* started setup */
584 obd_force:1, /* cleanup with > 0 obd refcount */
585 obd_fail:1, /* cleanup with failover */
586 obd_no_conn:1, /* deny new connections */
587 obd_inactive:1, /* device active/inactive
588 * (for /proc/status only!!) */
589 obd_no_ir:1, /* no imperative recovery. */
590 obd_process_conf:1, /* device is processing mgs config */
591 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
592 obd_force_abort_recovery:1; /* abort recovery forcely */
594 /* use separate field as it is set in interrupt to don't mess with
595 * protection of other bits using _bh lock */
596 unsigned long obd_recovery_expired:1;
597 /* uuid-export hash body */
598 struct cfs_hash *obd_uuid_hash;
599 /* nid-export hash body */
600 struct cfs_hash *obd_nid_hash;
602 struct cfs_hash *obd_nid_stats_hash;
603 /* client_generation-export hash body */
604 struct cfs_hash *obd_gen_hash;
605 struct list_head obd_nid_stats;
606 atomic_t obd_refcount;
607 struct list_head obd_exports;
608 struct list_head obd_unlinked_exports;
609 struct list_head obd_delayed_exports;
610 struct list_head obd_lwp_list;
612 spinlock_t obd_nid_lock;
613 struct ldlm_namespace *obd_namespace;
614 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
615 /* a spinlock is OK for what we do now, may need a semaphore later */
616 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
617 struct mutex obd_dev_mutex;
618 __u64 obd_last_committed;
619 spinlock_t obd_osfs_lock;
620 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
622 struct lvfs_run_ctxt obd_lvfs_ctxt;
623 struct obd_llog_group obd_olg; /* default llog group */
624 struct obd_device *obd_observer;
625 struct rw_semaphore obd_observer_link_sem;
626 struct obd_notify_upcall obd_upcall;
627 struct obd_export *obd_self_export;
628 struct obd_export *obd_lwp_export;
629 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
630 struct list_head obd_exports_timed;
631 time_t obd_eviction_timer; /* for ping evictor */
633 int obd_max_recoverable_clients;
634 atomic_t obd_connected_clients;
635 int obd_stale_clients;
636 /* this lock protects all recovery list_heads, timer and
637 * obd_next_recovery_transno value */
638 spinlock_t obd_recovery_task_lock;
639 __u64 obd_next_recovery_transno;
640 int obd_replayed_requests;
641 int obd_requests_queued_for_recovery;
642 wait_queue_head_t obd_next_transno_waitq;
643 /* protected by obd_recovery_task_lock */
644 struct timer_list obd_recovery_timer;
646 time_t obd_recovery_start;
647 /* seconds, for lprocfs_status */
648 time_t obd_recovery_end;
649 int obd_recovery_time_hard;
650 int obd_recovery_timeout;
651 int obd_recovery_ir_factor;
653 /* new recovery stuff from CMD2 */
654 struct target_recovery_data obd_recovery_data;
655 int obd_replayed_locks;
656 atomic_t obd_req_replay_clients;
657 atomic_t obd_lock_replay_clients;
658 /* all lists are protected by obd_recovery_task_lock */
659 struct list_head obd_req_replay_queue;
660 struct list_head obd_lock_replay_queue;
661 struct list_head obd_final_req_queue;
664 #ifdef HAVE_SERVER_SUPPORT
665 struct obd_device_target obt;
666 struct filter_obd filter;
668 struct echo_obd echo;
670 struct client_obd cli;
671 struct echo_client_obd echo_client;
675 /* Fields used by LProcFS */
676 unsigned int obd_cntr_base;
677 struct lprocfs_stats *obd_stats;
679 unsigned int obd_md_cntr_base;
680 struct lprocfs_stats *obd_md_stats;
682 struct proc_dir_entry *obd_proc_entry;
683 struct proc_dir_entry *obd_proc_exports_entry;
684 struct proc_dir_entry *obd_svc_procroot;
685 struct lprocfs_stats *obd_svc_stats;
686 struct lprocfs_vars *obd_vars;
687 atomic_t obd_evict_inprogress;
688 wait_queue_head_t obd_evict_inprogress_waitq;
689 struct list_head obd_evict_list; /* protected with pet_lock */
692 * Ldlm pool part. Save last calculated SLV and Limit.
694 rwlock_t obd_pool_lock;
699 * A list of outstanding class_incref()'s against this obd. For
702 struct lu_ref obd_reference;
704 int obd_conn_inprogress;
707 enum obd_cleanup_stage {
708 /* Special case hack for MDS LOVs */
710 /* can be directly mapped to .ldto_device_fini() */
714 /* get/set_info keys */
715 #define KEY_ASYNC "async"
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;
825 /* Various operation flags. */
826 enum mds_op_bias op_bias;
828 /* Used by readdir */
829 unsigned int op_max_pages;
831 /* used to transfer info between the stacks of MD client
832 * see enum op_cli_flags */
833 enum md_cli_flags op_cli_flags;
835 /* File object data version for HSM release, on client */
836 __u64 op_data_version;
837 struct lustre_handle op_lease_handle;
839 /* default stripe offset */
840 __u32 op_default_stripe_offset;
844 int (*md_blocking_ast)(struct ldlm_lock *lock,
845 struct ldlm_lock_desc *desc,
846 void *data, int flag);
849 struct md_enqueue_info;
850 /* metadata stat-ahead */
851 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
852 struct md_enqueue_info *minfo,
855 struct md_enqueue_info {
856 struct md_op_data mi_data;
857 struct lookup_intent mi_it;
858 struct lustre_handle mi_lockh;
859 struct inode *mi_dir;
860 md_enqueue_cb_t mi_cb;
865 struct module *o_owner;
866 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
867 void *karg, void __user *uarg);
868 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
869 __u32 keylen, void *key, __u32 *vallen, void *val);
870 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
871 __u32 keylen, void *key,
872 __u32 vallen, void *val,
873 struct ptlrpc_request_set *set);
874 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
875 int (*o_precleanup)(struct obd_device *dev,
876 enum obd_cleanup_stage cleanup_stage);
877 int (*o_cleanup)(struct obd_device *dev);
878 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
879 int (*o_postrecov)(struct obd_device *dev);
880 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
882 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
883 /* connect to the target device with given connection
884 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
885 * granted by the target, which are guaranteed to be a subset of flags
886 * asked for. If @ocd == NULL, use default parameters. */
887 int (*o_connect)(const struct lu_env *env,
888 struct obd_export **exp, struct obd_device *src,
889 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
891 int (*o_reconnect)(const struct lu_env *env,
892 struct obd_export *exp, struct obd_device *src,
893 struct obd_uuid *cluuid,
894 struct obd_connect_data *ocd,
896 int (*o_disconnect)(struct obd_export *exp);
898 /* Initialize/finalize fids infrastructure. */
899 int (*o_fid_init)(struct obd_device *obd,
900 struct obd_export *exp, enum lu_cli_type type);
901 int (*o_fid_fini)(struct obd_device *obd);
903 /* Allocate new fid according to passed @hint. */
904 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
905 struct lu_fid *fid, struct md_op_data *op_data);
908 * Object with @fid is getting deleted, we may want to do something
911 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
912 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
913 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
914 __u64 max_age, struct ptlrpc_request_set *set);
915 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
917 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
919 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
921 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
923 int (*o_preprw)(const struct lu_env *env, int cmd,
924 struct obd_export *exp, struct obdo *oa, int objcount,
925 struct obd_ioobj *obj, struct niobuf_remote *remote,
926 int *nr_pages, struct niobuf_local *local);
927 int (*o_commitrw)(const struct lu_env *env, int cmd,
928 struct obd_export *exp, struct obdo *oa,
929 int objcount, struct obd_ioobj *obj,
930 struct niobuf_remote *remote, int pages,
931 struct niobuf_local *local, int rc);
932 int (*o_init_export)(struct obd_export *exp);
933 int (*o_destroy_export)(struct obd_export *exp);
935 int (*o_import_event)(struct obd_device *, struct obd_import *,
936 enum obd_import_event);
938 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
939 enum obd_notify_event ev, void *data);
941 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
942 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
945 int (*o_quotactl)(struct obd_device *, struct obd_export *,
946 struct obd_quotactl *);
948 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
951 int (*o_pool_new)(struct obd_device *obd, char *poolname);
952 int (*o_pool_del)(struct obd_device *obd, char *poolname);
953 int (*o_pool_add)(struct obd_device *obd, char *poolname,
955 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
957 void (*o_getref)(struct obd_device *obd);
958 void (*o_putref)(struct obd_device *obd);
960 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
961 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
962 * Also, add a wrapper function in include/linux/obd_class.h. */
967 struct mdt_body *body;
968 struct lu_buf layout;
969 struct lmv_stripe_md *lmv;
970 #ifdef CONFIG_FS_POSIX_ACL
971 struct posix_acl *posix_acl;
973 struct mdt_remote_perm *remote_perm;
976 struct md_open_data {
977 struct obd_client_handle *mod_och;
978 struct ptlrpc_request *mod_open_req;
979 struct ptlrpc_request *mod_close_req;
980 atomic_t mod_refcount;
984 struct obd_client_handle {
985 struct lustre_handle och_fh;
986 struct lu_fid och_fid;
987 struct md_open_data *och_mod;
988 struct lustre_handle och_lease_handle; /* open lock for lease */
993 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
995 struct lookup_intent;
999 /* Every operation from MD_STATS_FIRST_OP up to and including
1000 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1001 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1002 * Operations after MD_STATS_LAST_OP are excluded from stats.
1003 * There are a few reasons for doing this: we prune the 17
1004 * counters which will be of minimal use in understanding
1005 * metadata utilization, we save memory by allocating 15
1006 * instead of 32 counters, we save cycles by not counting.
1008 * MD_STATS_FIRST_OP must be the first member of md_ops.
1010 #define MD_STATS_FIRST_OP m_close
1011 int (*m_close)(struct obd_export *, struct md_op_data *,
1012 struct md_open_data *, struct ptlrpc_request **);
1014 int (*m_create)(struct obd_export *, struct md_op_data *,
1015 const void *, size_t, umode_t, uid_t, gid_t,
1016 cfs_cap_t, __u64, struct ptlrpc_request **);
1018 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1019 const union ldlm_policy_data *,
1020 struct lookup_intent *, struct md_op_data *,
1021 struct lustre_handle *, __u64);
1023 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1024 struct ptlrpc_request **);
1026 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1027 struct lookup_intent *,
1028 struct ptlrpc_request **,
1029 ldlm_blocking_callback, __u64);
1031 int (*m_link)(struct obd_export *, struct md_op_data *,
1032 struct ptlrpc_request **);
1034 int (*m_rename)(struct obd_export *, struct md_op_data *,
1035 const char *, size_t, const char *, size_t,
1036 struct ptlrpc_request **);
1038 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1039 size_t , struct ptlrpc_request **);
1041 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1042 struct ptlrpc_request **);
1044 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1045 struct md_callback *cb_op, __u64 hash_offset,
1046 struct page **ppage);
1048 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1049 struct ptlrpc_request **);
1051 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1052 u64, const char *, const char *, int, int, int, u32,
1053 struct ptlrpc_request **);
1055 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1056 u64, const char *, const char *, int, int, int,
1057 struct ptlrpc_request **);
1059 int (*m_intent_getattr_async)(struct obd_export *,
1060 struct md_enqueue_info *,
1061 struct ldlm_enqueue_info *);
1063 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1064 struct lu_fid *, __u64 *bits);
1066 #define MD_STATS_LAST_OP m_revalidate_lock
1068 int (*m_getstatus)(struct obd_export *, struct lu_fid *);
1069 int (*m_null_inode)(struct obd_export *, const struct lu_fid *);
1071 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1072 ldlm_iterator_t, void *);
1074 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1075 struct ptlrpc_request **);
1077 int (*m_init_ea_size)(struct obd_export *, __u32, __u32);
1079 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1080 struct obd_export *, struct obd_export *,
1081 struct lustre_md *);
1083 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1085 int (*m_merge_attr)(struct obd_export *,
1086 const struct lmv_stripe_md *lsm,
1087 struct cl_attr *attr, ldlm_blocking_callback);
1089 int (*m_set_open_replay_data)(struct obd_export *,
1090 struct obd_client_handle *,
1091 struct lookup_intent *);
1093 int (*m_clear_open_replay_data)(struct obd_export *,
1094 struct obd_client_handle *);
1096 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1098 ldlm_mode_t (*m_lock_match)(struct obd_export *, __u64,
1099 const struct lu_fid *, ldlm_type_t,
1100 ldlm_policy_data_t *, ldlm_mode_t,
1101 struct lustre_handle *);
1103 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1104 ldlm_policy_data_t *, ldlm_mode_t,
1105 ldlm_cancel_flags_t flags, void *opaque);
1107 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1108 u32, struct ptlrpc_request **);
1110 int (*m_get_fid_from_lsm)(struct obd_export *,
1111 const struct lmv_stripe_md *,
1112 const char *name, int namelen,
1113 struct lu_fid *fid);
1114 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1115 const union lmv_mds_md *lmv, size_t lmv_size);
1118 static inline struct md_open_data *obd_mod_alloc(void)
1120 struct md_open_data *mod;
1124 atomic_set(&mod->mod_refcount, 1);
1128 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1129 #define obd_mod_put(mod) \
1131 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1132 if ((mod)->mod_open_req) \
1133 ptlrpc_req_finished((mod)->mod_open_req); \
1134 OBD_FREE_PTR(mod); \
1138 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1139 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1141 /* return 1 if client should be resend request */
1142 static inline int client_should_resend(int resend, struct client_obd *cli)
1144 return atomic_read(&cli->cl_resends) ?
1145 atomic_read(&cli->cl_resends) > resend : 1;
1149 * Return device name for this device
1151 * XXX: lu_device is declared before obd_device, while a pointer pointing
1152 * back to obd_device in lu_device, so this helper function defines here
1153 * instead of in lu_object.h
1155 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1157 return lu_dev->ld_obd->obd_name;
1160 static inline bool filename_is_volatile(const char *name, size_t namelen,
1166 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1169 /* caller does not care of idx */
1173 /* volatile file, the MDT can be set from name */
1174 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1175 /* if no MDT is specified, use std way */
1176 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1178 /* test for no MDT idx case */
1179 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1180 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1184 /* we have an idx, read it */
1185 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1186 *idx = simple_strtoul(start, &end, 16);
1188 * no digit, no trailing :, negative value
1190 if (((*idx == 0) && (end == start)) ||
1191 (*end != ':') || (*idx < 0))
1196 /* bad format of mdt idx, we cannot return an error
1197 * to caller so we use hash algo */
1198 CERROR("Bad volatile file name format: %s\n",
1199 name + LUSTRE_VOLATILE_HDR_LEN);
1203 static inline int cli_brw_size(struct obd_device *obd)
1205 LASSERT(obd != NULL);
1206 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1209 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1210 * of the client should be increased accordingly to avoid sending fragmented
1211 * RPCs over the network when the client runs out of the maximum dirty space
1212 * when so many RPCs are being generated.
1214 static inline void client_adjust_max_dirty(struct client_obd *cli)
1217 if (cli->cl_dirty_max_pages <= 0)
1218 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1219 >> PAGE_CACHE_SHIFT;
1221 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1222 cli->cl_max_pages_per_rpc;
1224 if (dirty_max > cli->cl_dirty_max_pages)
1225 cli->cl_dirty_max_pages = dirty_max;
1228 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1229 cli->cl_dirty_max_pages = totalram_pages / 8;
1232 #endif /* __OBD_H */