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 /** member of cl_cache->ccc_lru */
250 struct list_head cl_lru_osc;
251 /** # of available LRU slots left in the per-OSC cache.
252 * Available LRU slots are shared by all OSCs of the same file system,
253 * therefore this is a pointer to cl_client_cache::ccc_lru_left. */
254 atomic_long_t *cl_lru_left;
255 /** # of busy LRU pages. A page is considered busy if it's in writeback
256 * queue, or in transfer. Busy pages can't be discarded so they are not
258 atomic_long_t cl_lru_busy;
259 /** # of LRU pages in the cache for this client_obd */
260 atomic_long_t cl_lru_in_list;
261 /** # of threads are shrinking LRU cache. To avoid contention, it's not
262 * allowed to have multiple threads shrinking LRU cache. */
263 atomic_t cl_lru_shrinkers;
264 /** The time when this LRU cache was last used. */
265 time_t cl_lru_last_used;
266 /** stats: how many reclaims have happened for this client_obd.
267 * reclaim and shrink - shrink is async, voluntarily rebalancing;
268 * reclaim is sync, initiated by IO thread when the LRU slots are
270 __u64 cl_lru_reclaim;
271 /** List of LRU pages for this client_obd */
272 struct list_head cl_lru_list;
273 /** Lock for LRU page list */
274 spinlock_t cl_lru_list_lock;
275 /** # of unstable pages in this client_obd.
276 * An unstable page is a page state that WRITE RPC has finished but
277 * the transaction has NOT yet committed. */
278 atomic_long_t cl_unstable_count;
280 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
281 atomic_t cl_destroy_in_flight;
282 wait_queue_head_t cl_destroy_waitq;
284 struct mdc_rpc_lock *cl_rpc_lock;
286 /* modify rpcs in flight
287 * currently used for metadata only */
288 spinlock_t cl_mod_rpcs_lock;
289 __u16 cl_max_mod_rpcs_in_flight;
290 __u16 cl_mod_rpcs_in_flight;
291 __u16 cl_close_rpcs_in_flight;
292 wait_queue_head_t cl_mod_rpcs_waitq;
293 unsigned long *cl_mod_tag_bitmap;
294 struct obd_histogram cl_mod_rpcs_hist;
297 struct mutex cl_mgc_mutex;
298 struct local_oid_storage *cl_mgc_los;
299 struct dt_object *cl_mgc_configs_dir;
300 atomic_t cl_mgc_refcount;
301 struct obd_export *cl_mgc_mgsexp;
303 /* checksumming for data sent over the network */
304 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
305 /* supported checksum types that are worked out at connect time */
306 __u32 cl_supp_cksum_types;
307 /* checksum algorithm to be used */
308 cksum_type_t cl_cksum_type;
310 /* also protected by the poorly named _loi_list_lock lock above */
311 struct osc_async_rc cl_ar;
313 /* sequence manager */
314 struct lu_client_seq *cl_seq;
316 atomic_t cl_resends; /* resend count */
318 /* ptlrpc work for writeback in ptlrpcd context */
319 void *cl_writeback_work;
321 /* hash tables for osc_quota_info */
322 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
324 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
331 struct echo_client_obd {
332 struct obd_export *ec_exp; /* the local connection to osc/lov */
334 struct list_head ec_objects;
335 struct list_head ec_locks;
339 /* Generic subset of OSTs */
341 __u32 *op_array; /* array of index of
343 unsigned int op_count; /* number of OSTs in the array */
344 unsigned int op_size; /* allocated size of lp_array */
345 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
348 /* allow statfs data caching for 1 second */
349 #define OBD_STATFS_CACHE_SECONDS 1
351 struct lov_tgt_desc {
352 struct list_head ltd_kill;
353 struct obd_uuid ltd_uuid;
354 struct obd_device *ltd_obd;
355 struct obd_export *ltd_exp;
357 __u32 ltd_index; /* index in lov_obd->tgts */
358 unsigned long ltd_active:1,/* is this target up for requests */
359 ltd_activate:1,/* should target be activated */
360 ltd_reap:1; /* should this target be deleted */
364 struct lov_desc desc;
365 struct lov_tgt_desc **lov_tgts; /* sparse array */
366 struct ost_pool lov_packed; /* all OSTs in a packed
368 struct mutex lov_lock;
369 struct obd_connect_data lov_ocd;
370 struct proc_dir_entry *targets_proc_entry;
371 atomic_t lov_refcount;
372 __u32 lov_death_row; /* tgts scheduled to be deleted */
373 __u32 lov_tgt_size; /* size of tgts array */
376 struct cfs_hash *lov_pools_hash_body; /* used for key access */
377 struct list_head lov_pool_list; /* used for sequential access */
378 struct proc_dir_entry *lov_pool_proc_entry;
379 enum lustre_sec_part lov_sp_me;
381 /* Cached LRU and unstable data from upper layer */
382 struct cl_client_cache *lov_cache;
384 struct rw_semaphore lov_notify_lock;
387 struct lmv_tgt_desc {
388 struct obd_uuid ltd_uuid;
389 struct obd_export *ltd_exp;
391 struct mutex ltd_fid_mutex;
392 unsigned long ltd_active:1; /* target up for requests */
395 enum placement_policy {
396 PLACEMENT_CHAR_POLICY = 0,
397 PLACEMENT_NID_POLICY = 1,
398 PLACEMENT_INVAL_POLICY = 2,
402 typedef enum placement_policy placement_policy_t;
406 struct lu_client_fld lmv_fld;
408 placement_policy_t lmv_placement;
409 struct lmv_desc desc;
410 struct obd_uuid cluuid;
411 struct obd_export *exp;
412 struct proc_dir_entry *targets_proc_entry;
414 struct mutex lmv_init_mutex;
419 __u32 tgts_size; /* size of tgts array */
420 struct lmv_tgt_desc **tgts;
422 struct obd_connect_data conn_data;
425 struct niobuf_local {
426 __u64 lnb_file_offset;
427 __u32 lnb_page_offset;
430 struct page *lnb_page;
435 #define LUSTRE_FLD_NAME "fld"
436 #define LUSTRE_SEQ_NAME "seq"
438 #define LUSTRE_MDD_NAME "mdd"
439 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
440 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
441 #define LUSTRE_VVP_NAME "vvp"
442 #define LUSTRE_LMV_NAME "lmv"
443 #define LUSTRE_SLP_NAME "slp"
444 #define LUSTRE_LOD_NAME "lod"
445 #define LUSTRE_OSP_NAME "osp"
446 #define LUSTRE_LWP_NAME "lwp"
448 /* obd device type names */
449 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
450 #define LUSTRE_MDS_NAME "mds"
451 #define LUSTRE_MDT_NAME "mdt"
452 #define LUSTRE_MDC_NAME "mdc"
453 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
454 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
455 #define LUSTRE_OSC_NAME "osc"
456 #define LUSTRE_LOV_NAME "lov"
457 #define LUSTRE_MGS_NAME "mgs"
458 #define LUSTRE_MGC_NAME "mgc"
460 #define LUSTRE_ECHO_NAME "obdecho"
461 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
462 #define LUSTRE_QMT_NAME "qmt"
464 /* Constant obd names (post-rename) */
465 #define LUSTRE_MDS_OBDNAME "MDS"
466 #define LUSTRE_OSS_OBDNAME "OSS"
467 #define LUSTRE_MGS_OBDNAME "MGS"
468 #define LUSTRE_MGC_OBDNAME "MGC"
470 static inline int is_lwp_on_mdt(char *name)
474 ptr = strrchr(name, '-');
476 CERROR("%s is not a obdname\n", name);
480 /* LWP name on MDT is fsname-MDTxxxx-lwp-MDTxxxx */
482 if (strncmp(ptr + 1, "MDT", 3) != 0)
485 while (*(--ptr) != '-' && ptr != name);
490 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
496 static inline int is_lwp_on_ost(char *name)
500 ptr = strrchr(name, '-');
502 CERROR("%s is not a obdname\n", name);
506 /* LWP name on OST is fsname-MDTxxxx-lwp-OSTxxxx */
508 if (strncmp(ptr + 1, "OST", 3) != 0)
511 while (*(--ptr) != '-' && ptr != name);
516 if (strncmp(ptr + 1, LUSTRE_LWP_NAME, strlen(LUSTRE_LWP_NAME)) != 0)
523 * Events signalled through obd_notify() upcall-chain.
525 enum obd_notify_event {
528 /* Device connect start */
530 /* Device activated */
532 /* Device deactivated */
534 /* Device disconnected */
536 /* Connect data for import were changed */
539 OBD_NOTIFY_SYNC_NONBLOCK,
541 /* Configuration event */
543 /* Administratively deactivate/activate event */
544 OBD_NOTIFY_DEACTIVATE,
548 /* bit-mask flags for config events */
550 CONFIG_LOG = 0x1, /* finished processing config log */
551 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
552 CONFIG_TARGET = 0x4 /* one target is added */
556 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
557 * and liblustre being main examples).
559 struct obd_notify_upcall {
560 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
561 enum obd_notify_event ev, void *owner, void *data);
562 /* Opaque datum supplied by upper layer listener */
566 struct target_recovery_data {
567 svc_handler_t trd_recovery_handler;
568 pid_t trd_processing_task;
569 struct completion trd_starting;
570 struct completion trd_finishing;
573 struct obd_llog_group {
574 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
575 wait_queue_head_t olg_waitq;
577 struct mutex olg_cat_processing;
580 /* corresponds to one of the obd's */
581 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
584 struct obd_type *obd_type;
587 /* common and UUID name of this device */
588 char obd_name[MAX_OBD_NAME];
589 struct obd_uuid obd_uuid;
591 struct lu_device *obd_lu_dev;
593 /* bitfield modification is protected by obd_dev_lock */
595 obd_attached:1, /* finished attach */
596 obd_set_up:1, /* finished setup */
597 obd_recovering:1, /* there are recoverable clients */
598 obd_abort_recovery:1, /* recovery expired */
599 obd_version_recov:1, /* obd uses version checking */
600 obd_replayable:1, /* recovery is enabled;
602 obd_no_transno:1, /* no committed-transno notification */
603 obd_no_recov:1, /* fail instead of retry messages */
604 obd_stopping:1, /* started cleanup */
605 obd_starting:1, /* started setup */
606 obd_force:1, /* cleanup with > 0 obd refcount */
607 obd_fail:1, /* cleanup with failover */
608 obd_no_conn:1, /* deny new connections */
609 obd_inactive:1, /* device active/inactive
610 * (for /proc/status only!!) */
611 obd_no_ir:1, /* no imperative recovery. */
612 obd_process_conf:1, /* device is processing mgs config */
613 obd_uses_nid_stats:1, /* maintain per-client OBD stats */
614 obd_force_abort_recovery:1; /* abort recovery forcely */
616 /* use separate field as it is set in interrupt to don't mess with
617 * protection of other bits using _bh lock */
618 unsigned long obd_recovery_expired:1;
619 /* uuid-export hash body */
620 struct cfs_hash *obd_uuid_hash;
621 /* nid-export hash body */
622 struct cfs_hash *obd_nid_hash;
624 struct cfs_hash *obd_nid_stats_hash;
625 /* client_generation-export hash body */
626 struct cfs_hash *obd_gen_hash;
627 struct list_head obd_nid_stats;
628 atomic_t obd_refcount;
629 struct list_head obd_exports;
630 struct list_head obd_unlinked_exports;
631 struct list_head obd_delayed_exports;
632 struct list_head obd_lwp_list;
634 spinlock_t obd_nid_lock;
635 struct ldlm_namespace *obd_namespace;
636 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
637 /* a spinlock is OK for what we do now, may need a semaphore later */
638 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
639 struct mutex obd_dev_mutex;
640 __u64 obd_last_committed;
641 spinlock_t obd_osfs_lock;
642 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
644 struct lvfs_run_ctxt obd_lvfs_ctxt;
645 struct obd_llog_group obd_olg; /* default llog group */
646 struct obd_device *obd_observer;
647 struct rw_semaphore obd_observer_link_sem;
648 struct obd_notify_upcall obd_upcall;
649 struct obd_export *obd_self_export;
650 struct obd_export *obd_lwp_export;
651 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
652 struct list_head obd_exports_timed;
653 time_t obd_eviction_timer; /* for ping evictor */
655 int obd_max_recoverable_clients;
656 atomic_t obd_connected_clients;
657 int obd_stale_clients;
658 /* this lock protects all recovery list_heads, timer and
659 * obd_next_recovery_transno value */
660 spinlock_t obd_recovery_task_lock;
661 __u64 obd_next_recovery_transno;
662 int obd_replayed_requests;
663 int obd_requests_queued_for_recovery;
664 wait_queue_head_t obd_next_transno_waitq;
665 /* protected by obd_recovery_task_lock */
666 struct timer_list obd_recovery_timer;
668 time_t obd_recovery_start;
669 /* seconds, for lprocfs_status */
670 time_t obd_recovery_end;
671 int obd_recovery_time_hard;
672 int obd_recovery_timeout;
673 int obd_recovery_ir_factor;
675 /* new recovery stuff from CMD2 */
676 struct target_recovery_data obd_recovery_data;
677 int obd_replayed_locks;
678 atomic_t obd_req_replay_clients;
679 atomic_t obd_lock_replay_clients;
680 /* all lists are protected by obd_recovery_task_lock */
681 struct list_head obd_req_replay_queue;
682 struct list_head obd_lock_replay_queue;
683 struct list_head obd_final_req_queue;
686 #ifdef HAVE_SERVER_SUPPORT
687 struct obd_device_target obt;
688 struct filter_obd filter;
690 struct echo_obd echo;
692 struct client_obd cli;
693 struct echo_client_obd echo_client;
697 /* Fields used by LProcFS */
698 unsigned int obd_cntr_base;
699 struct lprocfs_stats *obd_stats;
701 unsigned int obd_md_cntr_base;
702 struct lprocfs_stats *obd_md_stats;
704 struct proc_dir_entry *obd_proc_entry;
705 struct proc_dir_entry *obd_proc_exports_entry;
706 struct proc_dir_entry *obd_svc_procroot;
707 struct lprocfs_stats *obd_svc_stats;
708 struct lprocfs_vars *obd_vars;
709 atomic_t obd_evict_inprogress;
710 wait_queue_head_t obd_evict_inprogress_waitq;
711 struct list_head obd_evict_list; /* protected with pet_lock */
714 * Ldlm pool part. Save last calculated SLV and Limit.
716 rwlock_t obd_pool_lock;
721 * A list of outstanding class_incref()'s against this obd. For
724 struct lu_ref obd_reference;
726 int obd_conn_inprogress;
729 /* get/set_info keys */
730 #define KEY_ASYNC "async"
731 #define KEY_CHANGELOG_CLEAR "changelog_clear"
732 #define KEY_FID2PATH "fid2path"
733 #define KEY_CHECKSUM "checksum"
734 #define KEY_CLEAR_FS "clear_fs"
735 #define KEY_CONN_DATA "conn_data"
736 #define KEY_EVICT_BY_NID "evict_by_nid"
737 #define KEY_FIEMAP "fiemap"
738 #define KEY_FLUSH_CTX "flush_ctx"
739 #define KEY_GRANT_SHRINK "grant_shrink"
740 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
741 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
742 #define KEY_INTERMDS "inter_mds"
743 #define KEY_LAST_ID "last_id"
744 #define KEY_LAST_FID "last_fid"
745 #define KEY_MAX_EASIZE "max_easize"
746 #define KEY_DEFAULT_EASIZE "default_easize"
747 #define KEY_MGSSEC "mgssec"
748 #define KEY_READ_ONLY "read-only"
749 #define KEY_REGISTER_TARGET "register_target"
750 #define KEY_SET_FS "set_fs"
751 #define KEY_TGT_COUNT "tgt_count"
752 /* KEY_SET_INFO in lustre_idl.h */
753 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
755 #define KEY_CACHE_SET "cache_set"
756 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
757 #define KEY_OSP_CONNECTED "osp_connected"
761 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
762 #define IT_OPEN (1 << 0)
763 #define IT_CREAT (1 << 1)
764 #define IT_READDIR (1 << 2)
765 #define IT_GETATTR (1 << 3)
766 #define IT_LOOKUP (1 << 4)
767 #define IT_UNLINK (1 << 5)
768 #define IT_TRUNC (1 << 6)
769 #define IT_GETXATTR (1 << 7)
770 #define IT_EXEC (1 << 8)
771 #define IT_PIN (1 << 9)
772 #define IT_LAYOUT (1 << 10)
773 #define IT_QUOTA_DQACQ (1 << 11)
774 #define IT_QUOTA_CONN (1 << 12)
775 #define IT_SETXATTR (1 << 13)
777 static inline int it_to_lock_mode(struct lookup_intent *it)
779 /* CREAT needs to be tested before open (both could be set) */
780 if (it->it_op & IT_CREAT)
782 else if (it->it_op & (IT_GETATTR | IT_OPEN | IT_LOOKUP |
785 else if (it->it_op & IT_READDIR)
787 else if (it->it_op & IT_GETXATTR)
789 else if (it->it_op & IT_SETXATTR)
792 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
797 MF_MDC_CANCEL_FID1 = 1 << 0,
798 MF_MDC_CANCEL_FID2 = 1 << 1,
799 MF_MDC_CANCEL_FID3 = 1 << 2,
800 MF_MDC_CANCEL_FID4 = 1 << 3,
801 MF_GET_MDT_IDX = 1 << 4,
805 CLI_SET_MEA = 1 << 0,
806 CLI_RM_ENTRY = 1 << 1,
809 CLI_MIGRATE = 1 << 4,
813 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
814 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
815 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
816 struct lu_fid op_fid4; /* to the operation locks. */
817 u32 op_mds; /* what mds server open will go to */
818 struct lustre_handle op_handle;
823 struct lmv_stripe_md *op_mea1;
824 struct lmv_stripe_md *op_mea2;
825 __u32 op_suppgids[2];
832 /* iattr fields and blocks. */
833 struct iattr op_attr;
834 loff_t op_attr_blocks;
835 unsigned int op_attr_flags; /* LUSTRE_{SYNC,..}_FL */
836 __u64 op_valid; /* OBD_MD_* */
838 enum md_op_flags op_flags;
840 /* Various operation flags. */
841 enum mds_op_bias op_bias;
843 /* Used by readdir */
844 unsigned int op_max_pages;
846 /* used to transfer info between the stacks of MD client
847 * see enum op_cli_flags */
848 enum md_cli_flags op_cli_flags;
850 /* File object data version for HSM release, on client */
851 __u64 op_data_version;
852 struct lustre_handle op_lease_handle;
854 /* default stripe offset */
855 __u32 op_default_stripe_offset;
859 int (*md_blocking_ast)(struct ldlm_lock *lock,
860 struct ldlm_lock_desc *desc,
861 void *data, int flag);
864 struct md_enqueue_info;
865 /* metadata stat-ahead */
866 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
867 struct md_enqueue_info *minfo,
870 struct md_enqueue_info {
871 struct md_op_data mi_data;
872 struct lookup_intent mi_it;
873 struct lustre_handle mi_lockh;
874 struct inode *mi_dir;
875 struct ldlm_enqueue_info mi_einfo;
876 md_enqueue_cb_t mi_cb;
881 struct module *o_owner;
882 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
883 void *karg, void __user *uarg);
884 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
885 __u32 keylen, void *key, __u32 *vallen, void *val);
886 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
887 __u32 keylen, void *key,
888 __u32 vallen, void *val,
889 struct ptlrpc_request_set *set);
890 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
891 int (*o_precleanup)(struct obd_device *dev);
892 int (*o_cleanup)(struct obd_device *dev);
893 int (*o_process_config)(struct obd_device *dev, size_t len, void *data);
894 int (*o_postrecov)(struct obd_device *dev);
895 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
897 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
898 /* connect to the target device with given connection
899 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
900 * granted by the target, which are guaranteed to be a subset of flags
901 * asked for. If @ocd == NULL, use default parameters. */
902 int (*o_connect)(const struct lu_env *env,
903 struct obd_export **exp, struct obd_device *src,
904 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
906 int (*o_reconnect)(const struct lu_env *env,
907 struct obd_export *exp, struct obd_device *src,
908 struct obd_uuid *cluuid,
909 struct obd_connect_data *ocd,
911 int (*o_disconnect)(struct obd_export *exp);
913 /* Initialize/finalize fids infrastructure. */
914 int (*o_fid_init)(struct obd_device *obd,
915 struct obd_export *exp, enum lu_cli_type type);
916 int (*o_fid_fini)(struct obd_device *obd);
918 /* Allocate new fid according to passed @hint. */
919 int (*o_fid_alloc)(const struct lu_env *env, struct obd_export *exp,
920 struct lu_fid *fid, struct md_op_data *op_data);
923 * Object with @fid is getting deleted, we may want to do something
926 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
927 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
928 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
929 __u64 max_age, struct ptlrpc_request_set *set);
930 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
932 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
934 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
936 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
938 int (*o_preprw)(const struct lu_env *env, int cmd,
939 struct obd_export *exp, struct obdo *oa, int objcount,
940 struct obd_ioobj *obj, struct niobuf_remote *remote,
941 int *nr_pages, struct niobuf_local *local);
942 int (*o_commitrw)(const struct lu_env *env, int cmd,
943 struct obd_export *exp, struct obdo *oa,
944 int objcount, struct obd_ioobj *obj,
945 struct niobuf_remote *remote, int pages,
946 struct niobuf_local *local, int rc);
947 int (*o_init_export)(struct obd_export *exp);
948 int (*o_destroy_export)(struct obd_export *exp);
950 int (*o_import_event)(struct obd_device *, struct obd_import *,
951 enum obd_import_event);
953 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
954 enum obd_notify_event ev, void *data);
956 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
957 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
960 int (*o_quotactl)(struct obd_device *, struct obd_export *,
961 struct obd_quotactl *);
963 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
966 int (*o_pool_new)(struct obd_device *obd, char *poolname);
967 int (*o_pool_del)(struct obd_device *obd, char *poolname);
968 int (*o_pool_add)(struct obd_device *obd, char *poolname,
970 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
972 void (*o_getref)(struct obd_device *obd);
973 void (*o_putref)(struct obd_device *obd);
975 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
976 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
977 * Also, add a wrapper function in include/linux/obd_class.h. */
982 struct mdt_body *body;
983 struct lu_buf layout;
984 struct lmv_stripe_md *lmv;
985 #ifdef CONFIG_FS_POSIX_ACL
986 struct posix_acl *posix_acl;
988 struct mdt_remote_perm *remote_perm;
991 struct md_open_data {
992 struct obd_client_handle *mod_och;
993 struct ptlrpc_request *mod_open_req;
994 struct ptlrpc_request *mod_close_req;
995 atomic_t mod_refcount;
999 struct obd_client_handle {
1000 struct lustre_handle och_fh;
1001 struct lu_fid och_fid;
1002 struct md_open_data *och_mod;
1003 struct lustre_handle och_lease_handle; /* open lock for lease */
1008 #define OBD_CLIENT_HANDLE_MAGIC 0xd15ea5ed
1010 struct lookup_intent;
1014 /* Every operation from MD_STATS_FIRST_OP up to and including
1015 * MD_STATS_LAST_OP will be counted by EXP_MD_OP_INCREMENT()
1016 * and will appear in /proc/fs/lustre/{lmv,mdc}/.../md_stats.
1017 * Operations after MD_STATS_LAST_OP are excluded from stats.
1018 * There are a few reasons for doing this: we prune the 17
1019 * counters which will be of minimal use in understanding
1020 * metadata utilization, we save memory by allocating 15
1021 * instead of 32 counters, we save cycles by not counting.
1023 * MD_STATS_FIRST_OP must be the first member of md_ops.
1025 #define MD_STATS_FIRST_OP m_close
1026 int (*m_close)(struct obd_export *, struct md_op_data *,
1027 struct md_open_data *, struct ptlrpc_request **);
1029 int (*m_create)(struct obd_export *, struct md_op_data *,
1030 const void *, size_t, umode_t, uid_t, gid_t,
1031 cfs_cap_t, __u64, struct ptlrpc_request **);
1033 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1034 const union ldlm_policy_data *,
1035 struct lookup_intent *, struct md_op_data *,
1036 struct lustre_handle *, __u64);
1038 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1039 struct ptlrpc_request **);
1041 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1042 struct lookup_intent *,
1043 struct ptlrpc_request **,
1044 ldlm_blocking_callback, __u64);
1046 int (*m_link)(struct obd_export *, struct md_op_data *,
1047 struct ptlrpc_request **);
1049 int (*m_rename)(struct obd_export *, struct md_op_data *,
1050 const char *, size_t, const char *, size_t,
1051 struct ptlrpc_request **);
1053 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1054 size_t , struct ptlrpc_request **);
1056 int (*m_fsync)(struct obd_export *, const struct lu_fid *,
1057 struct ptlrpc_request **);
1059 int (*m_read_page)(struct obd_export *, struct md_op_data *,
1060 struct md_callback *cb_op, __u64 hash_offset,
1061 struct page **ppage);
1063 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1064 struct ptlrpc_request **);
1066 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1067 u64, const char *, const char *, int, int, int, u32,
1068 struct ptlrpc_request **);
1070 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1071 u64, const char *, const char *, int, int, int,
1072 struct ptlrpc_request **);
1074 int (*m_intent_getattr_async)(struct obd_export *,
1075 struct md_enqueue_info *);
1077 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1078 struct lu_fid *, __u64 *bits);
1080 #define MD_STATS_LAST_OP m_revalidate_lock
1082 int (*m_getstatus)(struct obd_export *, struct lu_fid *);
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 enum ldlm_mode (*m_lock_match)(struct obd_export *, __u64,
1113 const struct lu_fid *, enum ldlm_type,
1114 union ldlm_policy_data *, enum ldlm_mode,
1115 struct lustre_handle *);
1117 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1118 union ldlm_policy_data *, enum ldlm_mode,
1119 enum ldlm_cancel_flags flags, void *opaque);
1121 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1122 u32, struct ptlrpc_request **);
1124 int (*m_get_fid_from_lsm)(struct obd_export *,
1125 const struct lmv_stripe_md *,
1126 const char *name, int namelen,
1127 struct lu_fid *fid);
1128 int (*m_unpackmd)(struct obd_export *exp, struct lmv_stripe_md **plsm,
1129 const union lmv_mds_md *lmv, size_t lmv_size);
1132 static inline struct md_open_data *obd_mod_alloc(void)
1134 struct md_open_data *mod;
1138 atomic_set(&mod->mod_refcount, 1);
1142 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1143 #define obd_mod_put(mod) \
1145 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1146 if ((mod)->mod_open_req) \
1147 ptlrpc_req_finished((mod)->mod_open_req); \
1148 OBD_FREE_PTR(mod); \
1152 void obdo_from_inode(struct obdo *dst, struct inode *src, u64 valid);
1153 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1155 /* return 1 if client should be resend request */
1156 static inline int client_should_resend(int resend, struct client_obd *cli)
1158 return atomic_read(&cli->cl_resends) ?
1159 atomic_read(&cli->cl_resends) > resend : 1;
1163 * Return device name for this device
1165 * XXX: lu_device is declared before obd_device, while a pointer pointing
1166 * back to obd_device in lu_device, so this helper function defines here
1167 * instead of in lu_object.h
1169 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1171 return lu_dev->ld_obd->obd_name;
1174 static inline bool filename_is_volatile(const char *name, size_t namelen,
1180 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1183 /* caller does not care of idx */
1187 /* volatile file, the MDT can be set from name */
1188 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1189 /* if no MDT is specified, use std way */
1190 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1192 /* test for no MDT idx case */
1193 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1194 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1198 /* we have an idx, read it */
1199 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1200 *idx = simple_strtoul(start, &end, 16);
1202 * no digit, no trailing :, negative value
1204 if (((*idx == 0) && (end == start)) ||
1205 (*end != ':') || (*idx < 0))
1210 /* bad format of mdt idx, we cannot return an error
1211 * to caller so we use hash algo */
1212 CERROR("Bad volatile file name format: %s\n",
1213 name + LUSTRE_VOLATILE_HDR_LEN);
1217 static inline int cli_brw_size(struct obd_device *obd)
1219 LASSERT(obd != NULL);
1220 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1223 /* when RPC size or the max RPCs in flight is increased, the max dirty pages
1224 * of the client should be increased accordingly to avoid sending fragmented
1225 * RPCs over the network when the client runs out of the maximum dirty space
1226 * when so many RPCs are being generated.
1228 static inline void client_adjust_max_dirty(struct client_obd *cli)
1231 if (cli->cl_dirty_max_pages <= 0)
1232 cli->cl_dirty_max_pages = (OSC_MAX_DIRTY_DEFAULT * 1024 * 1024)
1233 >> PAGE_CACHE_SHIFT;
1235 unsigned long dirty_max = cli->cl_max_rpcs_in_flight *
1236 cli->cl_max_pages_per_rpc;
1238 if (dirty_max > cli->cl_dirty_max_pages)
1239 cli->cl_dirty_max_pages = dirty_max;
1242 if (cli->cl_dirty_max_pages > totalram_pages / 8)
1243 cli->cl_dirty_max_pages = totalram_pages / 8;
1246 #endif /* __OBD_H */