4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
36 #include <linux/spinlock.h>
38 #define IOC_OSC_TYPE 'h'
39 #define IOC_OSC_MIN_NR 20
40 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
41 #define IOC_OSC_MAX_NR 50
43 #define IOC_MDC_TYPE 'i'
44 #define IOC_MDC_MIN_NR 20
45 #define IOC_MDC_MAX_NR 50
47 #include "lustre/lustre_idl.h"
48 #include "lustre_lib.h"
50 #include "lustre_export.h"
51 #include "lustre_fid.h"
52 #include "lustre_fld.h"
53 #include "lustre_intent.h"
55 #define MAX_OBD_DEVICES 8192
63 struct lov_oinfo { /* per-stripe data structure */
64 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
65 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
66 int loi_ost_gen; /* generation of this loi_ost_idx */
68 unsigned long loi_kms_valid:1;
69 __u64 loi_kms; /* known minimum size */
70 struct ost_lvb loi_lvb;
71 struct osc_async_rc loi_ar;
74 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
77 oinfo->loi_kms_valid = 1;
80 static inline void loi_init(struct lov_oinfo *loi)
84 struct lov_stripe_md {
87 pid_t lsm_lock_owner; /* debugging */
89 /* maximum possible file size, might change as OSTs status changes,
90 * e.g. disconnected, deactivated
95 struct ost_id lw_object_oi; /* lov object id/seq */
97 /* LOV-private members start here -- only for use in lov/. */
99 __u32 lw_stripe_size; /* size of the stripe */
100 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
101 __u16 lw_stripe_count; /* number of objects being striped over */
102 __u16 lw_layout_gen; /* generation of the layout */
103 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
106 struct lov_oinfo *lsm_oinfo[0];
109 #define lsm_oi lsm_wire.lw_object_oi
110 #define lsm_magic lsm_wire.lw_magic
111 #define lsm_layout_gen lsm_wire.lw_layout_gen
112 #define lsm_stripe_size lsm_wire.lw_stripe_size
113 #define lsm_pattern lsm_wire.lw_pattern
114 #define lsm_stripe_count lsm_wire.lw_stripe_count
115 #define lsm_pool_name lsm_wire.lw_pool_name
117 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
119 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
122 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
126 if (lsm_is_released(lsm))
131 static inline int lov_stripe_md_size(unsigned int stripe_count)
133 struct lov_stripe_md lsm;
135 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
140 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
142 /* obd info for a particular level (lov, osc). */
144 /* Flags used for set request specific flags:
145 - while lock handling, the flags obtained on the enqueue
146 request are set here.
147 - while stats, the flags used for control delay/resend.
148 - while setattr, the flags used for distinguish punch operation
151 /* lsm data specific for every OSC. */
152 struct lov_stripe_md *oi_md;
153 /* obdo data specific for every OSC, if needed at all. */
155 /* statfs data specific for every OSC, if needed at all. */
156 struct obd_statfs *oi_osfs;
157 /* An update callback which is called to update some data on upper
158 * level. E.g. it is used for update lsm->lsm_oinfo at every received
159 * request in osc level for enqueue requests. It is also possible to
160 * update some caller data from LOV layer if needed.
162 obd_enqueue_update_f oi_cb_up;
166 struct list_head typ_chain;
167 struct obd_ops *typ_dt_ops;
168 struct md_ops *typ_md_ops;
169 struct dentry *typ_debugfs_entry;
172 struct lu_device_type *typ_lu;
173 spinlock_t obd_type_lock;
174 struct kobject *typ_kobj;
186 LLOG_CONFIG_ORIG_CTXT = 0,
187 LLOG_CONFIG_REPL_CTXT,
188 LLOG_MDS_OST_ORIG_CTXT,
189 LLOG_MDS_OST_REPL_CTXT,
196 LLOG_LOVEA_ORIG_CTXT,
197 LLOG_LOVEA_REPL_CTXT,
198 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
199 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
200 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
201 LLOG_AGENT_ORIG_CTXT, /**< agent requests generation on cdt */
205 struct timeout_item {
206 enum timeout_event ti_event;
207 unsigned long ti_timeout;
210 struct list_head ti_obd_list;
211 struct list_head ti_chain;
214 #define OSC_MAX_RIF_DEFAULT 8
215 #define OSC_MAX_RIF_MAX 256
216 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
217 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
218 #define OSC_DEFAULT_RESENDS 10
220 /* possible values for fo_sync_lock_cancel */
222 NEVER_SYNC_ON_CANCEL = 0,
223 BLOCKING_SYNC_ON_CANCEL = 1,
224 ALWAYS_SYNC_ON_CANCEL = 2,
225 NUM_SYNC_ON_CANCEL_STATES
228 #define MDC_MAX_RIF_DEFAULT 8
229 #define MDC_MAX_RIF_MAX 512
231 enum obd_cl_sem_lock_class {
240 struct rw_semaphore cl_sem;
241 struct obd_uuid cl_target_uuid;
242 struct obd_import *cl_import; /* ptlrpc connection state */
244 /* max_mds_easize is purely a performance thing so we don't have to
245 * call obd_size_diskmd() all the time.
247 int cl_default_mds_easize;
248 int cl_max_mds_easize;
249 int cl_default_mds_cookiesize;
250 int cl_max_mds_cookiesize;
252 enum lustre_sec_part cl_sp_me;
253 enum lustre_sec_part cl_sp_to;
254 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
256 /* the grant values are protected by loi_list_lock below */
257 long cl_dirty; /* all _dirty_ in bytes */
258 long cl_dirty_max; /* allowed w/o rpc */
259 long cl_dirty_transit; /* dirty synchronous */
260 long cl_avail_grant; /* bytes of credit for ost */
261 long cl_lost_grant; /* lost credits (trunc) */
263 /* since we allocate grant by blocks, we don't know how many grant will
264 * be used to add a page into cache. As a solution, we reserve maximum
265 * grant before trying to dirty a page and unreserve the rest.
266 * See osc_{reserve|unreserve}_grant for details.
268 long cl_reserved_grant;
269 struct list_head cl_cache_waiters; /* waiting for cache/grant */
270 unsigned long cl_next_shrink_grant; /* jiffies */
271 struct list_head cl_grant_shrink_list; /* Timeout event list */
272 int cl_grant_shrink_interval; /* seconds */
274 /* A chunk is an optimal size used by osc_extent to determine
275 * the extent size. A chunk is max(PAGE_SIZE, OST block size)
279 int cl_extent_tax; /* extent overhead, by bytes */
281 /* keep track of objects that have lois that contain pages which
282 * have been queued for async brw. this lock also protects the
283 * lists of osc_client_pages that hang off of the loi
286 * ->cl_loi_list_lock protects consistency of
287 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
288 * ->ap_completion() call-backs are executed under this lock. As we
289 * cannot guarantee that these call-backs never block on all platforms
290 * (as a matter of fact they do block on Mac OS X), type of
291 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
292 * and blocking mutex on Mac OS X. (Alternative is to make this lock
293 * blocking everywhere, but we don't want to slow down fast-path of
294 * our main platform.)
296 * NB by Jinshan: though field names are still _loi_, but actually
297 * osc_object{}s are in the list.
299 spinlock_t cl_loi_list_lock;
300 struct list_head cl_loi_ready_list;
301 struct list_head cl_loi_hp_ready_list;
302 struct list_head cl_loi_write_list;
303 struct list_head cl_loi_read_list;
306 /* just a sum of the loi/lop pending numbers to be exported by sysfs */
307 atomic_t cl_pending_w_pages;
308 atomic_t cl_pending_r_pages;
309 __u32 cl_max_pages_per_rpc;
310 int cl_max_rpcs_in_flight;
311 struct obd_histogram cl_read_rpc_hist;
312 struct obd_histogram cl_write_rpc_hist;
313 struct obd_histogram cl_read_page_hist;
314 struct obd_histogram cl_write_page_hist;
315 struct obd_histogram cl_read_offset_hist;
316 struct obd_histogram cl_write_offset_hist;
318 /* lru for osc caching pages */
319 struct cl_client_cache *cl_cache;
320 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
321 atomic_t *cl_lru_left;
322 atomic_t cl_lru_busy;
323 atomic_t cl_lru_shrinkers;
324 atomic_t cl_lru_in_list;
325 struct list_head cl_lru_list; /* lru page list */
326 spinlock_t cl_lru_list_lock; /* page list protector */
327 atomic_t cl_unstable_count;
329 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
330 atomic_t cl_destroy_in_flight;
331 wait_queue_head_t cl_destroy_waitq;
333 struct mdc_rpc_lock *cl_rpc_lock;
334 struct mdc_rpc_lock *cl_close_lock;
337 atomic_t cl_mgc_refcount;
338 struct obd_export *cl_mgc_mgsexp;
340 /* checksumming for data sent over the network */
341 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
342 /* supported checksum types that are worked out at connect time */
343 __u32 cl_supp_cksum_types;
344 /* checksum algorithm to be used */
345 enum cksum_type cl_cksum_type;
347 /* also protected by the poorly named _loi_list_lock lock above */
348 struct osc_async_rc cl_ar;
350 /* used by quotacheck when the servers are older than 2.4 */
351 int cl_qchk_stat; /* quotacheck stat of the peer */
352 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
353 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
354 #warning "please consider removing quotacheck compatibility code"
357 /* sequence manager */
358 struct lu_client_seq *cl_seq;
360 atomic_t cl_resends; /* resend count */
362 /* ptlrpc work for writeback in ptlrpcd context */
363 void *cl_writeback_work;
365 /* hash tables for osc_quota_info */
366 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
369 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
376 struct echo_client_obd {
377 struct obd_export *ec_exp; /* the local connection to osc/lov */
379 struct list_head ec_objects;
380 struct list_head ec_locks;
384 /* Generic subset of OSTs */
386 __u32 *op_array; /* array of index of lov_obd->lov_tgts */
387 unsigned int op_count; /* number of OSTs in the array */
388 unsigned int op_size; /* allocated size of lp_array */
389 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
392 /* allow statfs data caching for 1 second */
393 #define OBD_STATFS_CACHE_SECONDS 1
395 struct lov_tgt_desc {
396 struct list_head ltd_kill;
397 struct obd_uuid ltd_uuid;
398 struct obd_device *ltd_obd;
399 struct obd_export *ltd_exp;
401 __u32 ltd_index; /* index in lov_obd->tgts */
402 unsigned long ltd_active:1,/* is this target up for requests */
403 ltd_activate:1,/* should target be activated */
404 ltd_reap:1; /* should this target be deleted */
408 struct lov_desc desc;
409 struct lov_tgt_desc **lov_tgts; /* sparse array */
410 struct ost_pool lov_packed; /* all OSTs in a packed array */
411 struct mutex lov_lock;
412 struct obd_connect_data lov_ocd;
413 atomic_t lov_refcount;
414 __u32 lov_death_row;/* tgts scheduled to be deleted */
415 __u32 lov_tgt_size; /* size of tgts array */
418 struct cfs_hash *lov_pools_hash_body; /* used for key access */
419 struct list_head lov_pool_list; /* used for sequential access */
420 struct dentry *lov_pool_debugfs_entry;
421 enum lustre_sec_part lov_sp_me;
423 /* Cached LRU and unstable data from upper layer */
424 struct cl_client_cache *lov_cache;
426 struct rw_semaphore lov_notify_lock;
428 struct kobject *lov_tgts_kobj;
431 struct lmv_tgt_desc {
432 struct obd_uuid ltd_uuid;
433 struct obd_export *ltd_exp;
435 struct mutex ltd_fid_mutex;
436 unsigned long ltd_active:1; /* target up for requests */
439 enum placement_policy {
440 PLACEMENT_CHAR_POLICY = 0,
441 PLACEMENT_NID_POLICY = 1,
442 PLACEMENT_INVAL_POLICY = 2,
448 struct lu_client_fld lmv_fld;
450 enum placement_policy lmv_placement;
451 struct lmv_desc desc;
452 struct obd_uuid cluuid;
453 struct obd_export *exp;
455 struct mutex lmv_init_mutex;
460 int max_def_cookiesize;
463 int tgts_size; /* size of tgts array */
464 struct lmv_tgt_desc **tgts;
466 struct obd_connect_data conn_data;
467 struct kobject *lmv_tgts_kobj;
470 struct niobuf_local {
471 __u64 lnb_file_offset;
472 __u32 lnb_page_offset;
476 struct dentry *dentry;
481 #define LUSTRE_FLD_NAME "fld"
482 #define LUSTRE_SEQ_NAME "seq"
484 #define LUSTRE_MDD_NAME "mdd"
485 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
486 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
487 #define LUSTRE_VVP_NAME "vvp"
488 #define LUSTRE_LMV_NAME "lmv"
489 #define LUSTRE_SLP_NAME "slp"
490 #define LUSTRE_LOD_NAME "lod"
491 #define LUSTRE_OSP_NAME "osp"
492 #define LUSTRE_LWP_NAME "lwp"
494 /* obd device type names */
495 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
496 #define LUSTRE_MDS_NAME "mds"
497 #define LUSTRE_MDT_NAME "mdt"
498 #define LUSTRE_MDC_NAME "mdc"
499 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
500 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
501 #define LUSTRE_OSC_NAME "osc"
502 #define LUSTRE_LOV_NAME "lov"
503 #define LUSTRE_MGS_NAME "mgs"
504 #define LUSTRE_MGC_NAME "mgc"
506 #define LUSTRE_ECHO_NAME "obdecho"
507 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
508 #define LUSTRE_QMT_NAME "qmt"
510 /* Constant obd names (post-rename) */
511 #define LUSTRE_MDS_OBDNAME "MDS"
512 #define LUSTRE_OSS_OBDNAME "OSS"
513 #define LUSTRE_MGS_OBDNAME "MGS"
514 #define LUSTRE_MGC_OBDNAME "MGC"
516 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
517 #define N_LOCAL_TEMP_PAGE 0x10000000
519 struct obd_trans_info {
522 /* Only used on the server side for tracking acks. */
523 struct oti_req_ack_lock {
524 struct lustre_handle lock;
528 struct llog_cookie oti_onecookie;
529 struct llog_cookie *oti_logcookies;
531 /** synchronous write is needed */
532 unsigned long oti_sync_write:1;
534 /* initial thread handling transaction */
535 struct ptlrpc_thread *oti_thread;
538 __u64 oti_pre_version;
542 struct obd_uuid *oti_ost_uuid;
545 static inline void oti_alloc_cookies(struct obd_trans_info *oti,
551 if (num_cookies == 1)
552 oti->oti_logcookies = &oti->oti_onecookie;
554 oti->oti_logcookies = libcfs_kvzalloc(num_cookies * sizeof(oti->oti_onecookie),
557 oti->oti_numcookies = num_cookies;
560 static inline void oti_free_cookies(struct obd_trans_info *oti)
562 if (!oti || !oti->oti_logcookies)
565 if (oti->oti_logcookies == &oti->oti_onecookie)
566 LASSERT(oti->oti_numcookies == 1);
568 kvfree(oti->oti_logcookies);
570 oti->oti_logcookies = NULL;
571 oti->oti_numcookies = 0;
575 * Events signalled through obd_notify() upcall-chain.
577 enum obd_notify_event {
580 /* Device connect start */
582 /* Device activated */
584 /* Device deactivated */
586 /* Device disconnected */
588 /* Connect data for import were changed */
591 OBD_NOTIFY_SYNC_NONBLOCK,
593 /* Configuration event */
595 /* Administratively deactivate/activate event */
596 OBD_NOTIFY_DEACTIVATE,
601 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
602 * being main example).
604 struct obd_notify_upcall {
605 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
606 enum obd_notify_event ev, void *owner, void *data);
607 /* Opaque datum supplied by upper layer listener */
611 struct target_recovery_data {
612 svc_handler_t trd_recovery_handler;
613 pid_t trd_processing_task;
614 struct completion trd_starting;
615 struct completion trd_finishing;
618 struct obd_llog_group {
620 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
621 wait_queue_head_t olg_waitq;
623 struct mutex olg_cat_processing;
626 /* corresponds to one of the obd's */
627 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
628 #define OBD_DEV_BY_DEVNAME 0xffffd0de
630 struct lvfs_run_ctxt {
631 struct dt_device *dt;
635 struct obd_type *obd_type;
638 /* common and UUID name of this device */
639 char obd_name[MAX_OBD_NAME];
640 struct obd_uuid obd_uuid;
642 struct lu_device *obd_lu_dev;
645 /* bitfield modification is protected by obd_dev_lock */
646 unsigned long obd_attached:1, /* finished attach */
647 obd_set_up:1, /* finished setup */
648 obd_version_recov:1, /* obd uses version checking */
649 obd_replayable:1,/* recovery is enabled; inform clients */
650 obd_no_transno:1, /* no committed-transno notification */
651 obd_no_recov:1, /* fail instead of retry messages */
652 obd_stopping:1, /* started cleanup */
653 obd_starting:1, /* started setup */
654 obd_force:1, /* cleanup with > 0 obd refcount */
655 obd_fail:1, /* cleanup with failover */
656 obd_async_recov:1, /* allow asynchronous orphan cleanup */
657 obd_no_conn:1, /* deny new connections */
658 obd_inactive:1, /* device active/inactive
659 * (for sysfs status only!!)
661 obd_no_ir:1, /* no imperative recovery. */
662 obd_process_conf:1; /* device is processing mgs config */
663 /* use separate field as it is set in interrupt to don't mess with
664 * protection of other bits using _bh lock
666 unsigned long obd_recovery_expired:1;
667 /* uuid-export hash body */
668 struct cfs_hash *obd_uuid_hash;
669 atomic_t obd_refcount;
670 wait_queue_head_t obd_refcount_waitq;
671 struct list_head obd_exports;
672 struct list_head obd_unlinked_exports;
673 struct list_head obd_delayed_exports;
675 spinlock_t obd_nid_lock;
676 struct ldlm_namespace *obd_namespace;
677 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
678 /* a spinlock is OK for what we do now, may need a semaphore later */
679 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
680 struct mutex obd_dev_mutex;
681 __u64 obd_last_committed;
682 spinlock_t obd_osfs_lock;
683 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
685 struct lvfs_run_ctxt obd_lvfs_ctxt;
686 struct obd_llog_group obd_olg; /* default llog group */
687 struct obd_device *obd_observer;
688 struct rw_semaphore obd_observer_link_sem;
689 struct obd_notify_upcall obd_upcall;
690 struct obd_export *obd_self_export;
693 struct client_obd cli;
694 struct echo_client_obd echo_client;
698 /* Fields used by LProcFS */
699 unsigned int obd_cntr_base;
700 struct lprocfs_stats *obd_stats;
702 unsigned int md_cntr_base;
703 struct lprocfs_stats *md_stats;
705 struct dentry *obd_debugfs_entry;
706 struct dentry *obd_svc_debugfs_entry;
707 struct lprocfs_stats *obd_svc_stats;
708 atomic_t obd_evict_inprogress;
709 wait_queue_head_t obd_evict_inprogress_waitq;
710 struct list_head obd_evict_list; /* protected with pet_lock */
713 * Ldlm pool part. Save last calculated SLV and Limit.
715 rwlock_t obd_pool_lock;
720 * A list of outstanding class_incref()'s against this obd. For
723 struct lu_ref obd_reference;
725 int obd_conn_inprogress;
727 struct kobject obd_kobj; /* sysfs object */
728 struct completion obd_kobj_unregister;
731 #define OBD_LLOG_FL_SENDNOW 0x0001
732 #define OBD_LLOG_FL_EXIT 0x0002
734 enum obd_cleanup_stage {
735 /* Special case hack for MDS LOVs */
737 /* can be directly mapped to .ldto_device_fini() */
741 /* get/set_info keys */
742 #define KEY_ASYNC "async"
743 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
744 #define KEY_BLOCKSIZE "blocksize"
745 #define KEY_CHANGELOG_CLEAR "changelog_clear"
746 #define KEY_FID2PATH "fid2path"
747 #define KEY_CHECKSUM "checksum"
748 #define KEY_CLEAR_FS "clear_fs"
749 #define KEY_CONN_DATA "conn_data"
750 #define KEY_EVICT_BY_NID "evict_by_nid"
751 #define KEY_FIEMAP "fiemap"
752 #define KEY_FLUSH_CTX "flush_ctx"
753 #define KEY_GRANT_SHRINK "grant_shrink"
754 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
755 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
756 #define KEY_INIT_RECOV "initial_recov"
757 #define KEY_INTERMDS "inter_mds"
758 #define KEY_LAST_ID "last_id"
759 #define KEY_LAST_FID "last_fid"
760 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
761 #define KEY_LOVDESC "lovdesc"
762 #define KEY_LOV_IDX "lov_idx"
763 #define KEY_MAX_EASIZE "max_easize"
764 #define KEY_DEFAULT_EASIZE "default_easize"
765 #define KEY_MDS_CONN "mds_conn"
766 #define KEY_MGSSEC "mgssec"
767 #define KEY_NEXT_ID "next_id"
768 #define KEY_READ_ONLY "read-only"
769 #define KEY_REGISTER_TARGET "register_target"
770 #define KEY_SET_FS "set_fs"
771 #define KEY_TGT_COUNT "tgt_count"
772 /* KEY_SET_INFO in lustre_idl.h */
773 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
774 #define KEY_CONNECT_FLAG "connect_flags"
775 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
777 #define KEY_CACHE_SET "cache_set"
778 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
779 #define KEY_CHANGELOG_INDEX "changelog_index"
783 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
784 #define IT_OPEN (1 << 0)
785 #define IT_CREAT (1 << 1)
786 #define IT_READDIR (1 << 2)
787 #define IT_GETATTR (1 << 3)
788 #define IT_LOOKUP (1 << 4)
789 #define IT_UNLINK (1 << 5)
790 #define IT_TRUNC (1 << 6)
791 #define IT_GETXATTR (1 << 7)
792 #define IT_EXEC (1 << 8)
793 #define IT_PIN (1 << 9)
794 #define IT_LAYOUT (1 << 10)
795 #define IT_QUOTA_DQACQ (1 << 11)
796 #define IT_QUOTA_CONN (1 << 12)
797 #define IT_SETXATTR (1 << 13)
799 static inline int it_to_lock_mode(struct lookup_intent *it)
801 /* CREAT needs to be tested before open (both could be set) */
802 if (it->it_op & IT_CREAT)
804 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
807 else if (it->it_op & IT_GETXATTR)
809 else if (it->it_op & IT_SETXATTR)
812 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
817 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
818 struct lu_fid op_fid2; /* operation fid2 (usually child) */
819 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
820 struct lu_fid op_fid4; /* to the operation locks. */
821 u32 op_mds; /* what mds server open will go to */
822 struct lustre_handle op_handle;
827 struct lmv_stripe_md *op_mea1;
828 struct lmv_stripe_md *op_mea2;
829 __u32 op_suppgids[2];
835 /* iattr fields and blocks. */
836 struct iattr op_attr;
837 unsigned int op_attr_flags;
839 loff_t op_attr_blocks;
841 /* Size-on-MDS epoch and flags. */
845 /* Various operation flags. */
846 enum mds_op_bias op_bias;
851 /* Used by readdir */
854 /* Used by readdir */
857 /* used to transfer info between the stacks of MD client
858 * see enum op_cli_flags
862 /* File object data version for HSM release, on client */
863 __u64 op_data_version;
864 struct lustre_handle op_lease_handle;
868 CLI_SET_MEA = 1 << 0,
869 CLI_RM_ENTRY = 1 << 1,
872 struct md_enqueue_info;
873 /* metadata stat-ahead */
875 struct md_enqueue_info {
876 struct md_op_data mi_data;
877 struct lookup_intent mi_it;
878 struct lustre_handle mi_lockh;
879 struct inode *mi_dir;
880 int (*mi_cb)(struct ptlrpc_request *req,
881 struct md_enqueue_info *minfo, int rc);
883 unsigned int mi_generation;
887 struct module *owner;
888 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
889 void *karg, void __user *uarg);
890 int (*get_info)(const struct lu_env *env, struct obd_export *,
891 __u32 keylen, void *key, __u32 *vallen, void *val,
892 struct lov_stripe_md *lsm);
893 int (*set_info_async)(const struct lu_env *, struct obd_export *,
894 __u32 keylen, void *key,
895 __u32 vallen, void *val,
896 struct ptlrpc_request_set *set);
897 int (*attach)(struct obd_device *dev, u32 len, void *data);
898 int (*detach)(struct obd_device *dev);
899 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
900 int (*precleanup)(struct obd_device *dev,
901 enum obd_cleanup_stage cleanup_stage);
902 int (*cleanup)(struct obd_device *dev);
903 int (*process_config)(struct obd_device *dev, u32 len, void *data);
904 int (*postrecov)(struct obd_device *dev);
905 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
907 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
908 /* connect to the target device with given connection
909 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
910 * granted by the target, which are guaranteed to be a subset of flags
911 * asked for. If @ocd == NULL, use default parameters.
913 int (*connect)(const struct lu_env *env,
914 struct obd_export **exp, struct obd_device *src,
915 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
917 int (*reconnect)(const struct lu_env *env,
918 struct obd_export *exp, struct obd_device *src,
919 struct obd_uuid *cluuid,
920 struct obd_connect_data *ocd,
922 int (*disconnect)(struct obd_export *exp);
924 /* Initialize/finalize fids infrastructure. */
925 int (*fid_init)(struct obd_device *obd,
926 struct obd_export *exp, enum lu_cli_type type);
927 int (*fid_fini)(struct obd_device *obd);
929 /* Allocate new fid according to passed @hint. */
930 int (*fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
931 struct md_op_data *op_data);
934 * Object with @fid is getting deleted, we may want to do something
937 int (*statfs)(const struct lu_env *, struct obd_export *exp,
938 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
939 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
940 __u64 max_age, struct ptlrpc_request_set *set);
941 int (*packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
942 struct lov_stripe_md *mem_src);
943 int (*unpackmd)(struct obd_export *exp,
944 struct lov_stripe_md **mem_tgt,
945 struct lov_mds_md *disk_src, int disk_len);
946 int (*preallocate)(struct lustre_handle *, u32 *req, u64 *ids);
947 int (*create)(const struct lu_env *env, struct obd_export *exp,
948 struct obdo *oa, struct lov_stripe_md **ea,
949 struct obd_trans_info *oti);
950 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
951 struct obdo *oa, struct lov_stripe_md *ea,
952 struct obd_trans_info *oti, struct obd_export *md_exp);
953 int (*setattr)(const struct lu_env *, struct obd_export *exp,
954 struct obd_info *oinfo, struct obd_trans_info *oti);
955 int (*setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
956 struct obd_trans_info *oti,
957 struct ptlrpc_request_set *rqset);
958 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
959 struct obd_info *oinfo);
960 int (*getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
961 struct ptlrpc_request_set *set);
962 int (*adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
963 u64 size, int shrink);
964 int (*preprw)(const struct lu_env *env, int cmd,
965 struct obd_export *exp, struct obdo *oa, int objcount,
966 struct obd_ioobj *obj, struct niobuf_remote *remote,
967 int *nr_pages, struct niobuf_local *local,
968 struct obd_trans_info *oti);
969 int (*commitrw)(const struct lu_env *env, int cmd,
970 struct obd_export *exp, struct obdo *oa,
971 int objcount, struct obd_ioobj *obj,
972 struct niobuf_remote *remote, int pages,
973 struct niobuf_local *local,
974 struct obd_trans_info *oti, int rc);
975 int (*find_cbdata)(struct obd_export *, struct lov_stripe_md *,
976 ldlm_iterator_t it, void *data);
977 int (*init_export)(struct obd_export *exp);
978 int (*destroy_export)(struct obd_export *exp);
980 /* metadata-only methods */
981 int (*import_event)(struct obd_device *, struct obd_import *,
982 enum obd_import_event);
984 int (*notify)(struct obd_device *obd, struct obd_device *watched,
985 enum obd_notify_event ev, void *data);
987 int (*health_check)(const struct lu_env *env, struct obd_device *);
988 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
991 int (*quotacheck)(struct obd_device *, struct obd_export *,
992 struct obd_quotactl *);
993 int (*quotactl)(struct obd_device *, struct obd_export *,
994 struct obd_quotactl *);
997 int (*pool_new)(struct obd_device *obd, char *poolname);
998 int (*pool_del)(struct obd_device *obd, char *poolname);
999 int (*pool_add)(struct obd_device *obd, char *poolname,
1001 int (*pool_rem)(struct obd_device *obd, char *poolname,
1003 void (*getref)(struct obd_device *obd);
1004 void (*putref)(struct obd_device *obd);
1006 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1007 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1008 * Also, add a wrapper function in include/linux/obd_class.h.
1013 LUSTRE_OPC_MKDIR = (1 << 0),
1014 LUSTRE_OPC_SYMLINK = (1 << 1),
1015 LUSTRE_OPC_MKNOD = (1 << 2),
1016 LUSTRE_OPC_CREATE = (1 << 3),
1017 LUSTRE_OPC_ANY = (1 << 4)
1020 /* lmv structures */
1021 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1022 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1023 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1025 #define MAX_HASH_SIZE_32 0x7fffffffUL
1026 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1027 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1030 struct mdt_body *body;
1031 struct lov_stripe_md *lsm;
1032 struct lmv_stripe_md *mea;
1033 #ifdef CONFIG_FS_POSIX_ACL
1034 struct posix_acl *posix_acl;
1036 struct mdt_remote_perm *remote_perm;
1039 struct md_open_data {
1040 struct obd_client_handle *mod_och;
1041 struct ptlrpc_request *mod_open_req;
1042 struct ptlrpc_request *mod_close_req;
1043 atomic_t mod_refcount;
1047 struct lookup_intent;
1050 int (*getstatus)(struct obd_export *, struct lu_fid *);
1051 int (*null_inode)(struct obd_export *, const struct lu_fid *);
1052 int (*find_cbdata)(struct obd_export *, const struct lu_fid *,
1053 ldlm_iterator_t, void *);
1054 int (*close)(struct obd_export *, struct md_op_data *,
1055 struct md_open_data *, struct ptlrpc_request **);
1056 int (*create)(struct obd_export *, struct md_op_data *,
1057 const void *, int, int, __u32, __u32, cfs_cap_t,
1058 __u64, struct ptlrpc_request **);
1059 int (*done_writing)(struct obd_export *, struct md_op_data *,
1060 struct md_open_data *);
1061 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1062 struct lookup_intent *, struct md_op_data *,
1063 struct lustre_handle *, void *, int,
1064 struct ptlrpc_request **, __u64);
1065 int (*getattr)(struct obd_export *, struct md_op_data *,
1066 struct ptlrpc_request **);
1067 int (*getattr_name)(struct obd_export *, struct md_op_data *,
1068 struct ptlrpc_request **);
1069 int (*intent_lock)(struct obd_export *, struct md_op_data *,
1070 void *, int, struct lookup_intent *, int,
1071 struct ptlrpc_request **,
1072 ldlm_blocking_callback, __u64);
1073 int (*link)(struct obd_export *, struct md_op_data *,
1074 struct ptlrpc_request **);
1075 int (*rename)(struct obd_export *, struct md_op_data *,
1076 const char *, int, const char *, int,
1077 struct ptlrpc_request **);
1078 int (*is_subdir)(struct obd_export *, const struct lu_fid *,
1079 const struct lu_fid *,
1080 struct ptlrpc_request **);
1081 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
1082 int, void *, int, struct ptlrpc_request **,
1083 struct md_open_data **mod);
1084 int (*sync)(struct obd_export *, const struct lu_fid *,
1085 struct ptlrpc_request **);
1086 int (*readpage)(struct obd_export *, struct md_op_data *,
1087 struct page **, struct ptlrpc_request **);
1089 int (*unlink)(struct obd_export *, struct md_op_data *,
1090 struct ptlrpc_request **);
1092 int (*setxattr)(struct obd_export *, const struct lu_fid *,
1093 u64, const char *, const char *, int, int, int, __u32,
1094 struct ptlrpc_request **);
1096 int (*getxattr)(struct obd_export *, const struct lu_fid *,
1097 u64, const char *, const char *, int, int, int,
1098 struct ptlrpc_request **);
1100 int (*init_ea_size)(struct obd_export *, int, int, int, int);
1102 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1103 struct obd_export *, struct obd_export *,
1104 struct lustre_md *);
1106 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
1108 int (*set_open_replay_data)(struct obd_export *,
1109 struct obd_client_handle *,
1110 struct lookup_intent *);
1111 int (*clear_open_replay_data)(struct obd_export *,
1112 struct obd_client_handle *);
1113 int (*set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1115 enum ldlm_mode (*lock_match)(struct obd_export *, __u64,
1116 const struct lu_fid *, enum ldlm_type,
1117 ldlm_policy_data_t *, enum ldlm_mode,
1118 struct lustre_handle *);
1120 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
1121 ldlm_policy_data_t *, enum ldlm_mode,
1122 enum ldlm_cancel_flags flags, void *opaque);
1124 int (*intent_getattr_async)(struct obd_export *,
1125 struct md_enqueue_info *,
1126 struct ldlm_enqueue_info *);
1128 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
1129 struct lu_fid *, __u64 *bits);
1132 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1133 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1134 * wrapper function in include/linux/obd_class.h.
1138 struct lsm_operations {
1139 void (*lsm_free)(struct lov_stripe_md *);
1140 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1141 struct obd_export *md_exp);
1142 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, u64 *,
1144 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, u64 *,
1146 int (*lsm_lmm_verify)(struct lov_mds_md *lmm, int lmm_bytes,
1147 __u16 *stripe_count);
1148 int (*lsm_unpackmd)(struct lov_obd *lov, struct lov_stripe_md *lsm,
1149 struct lov_mds_md *lmm);
1152 extern const struct lsm_operations lsm_v1_ops;
1153 extern const struct lsm_operations lsm_v3_ops;
1154 static inline const struct lsm_operations *lsm_op_find(int magic)
1162 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1167 /* Requests for obd_extent_calc() */
1168 #define OBD_CALC_STRIPE_START 1
1169 #define OBD_CALC_STRIPE_END 2
1171 static inline struct md_open_data *obd_mod_alloc(void)
1173 struct md_open_data *mod;
1175 mod = kzalloc(sizeof(*mod), GFP_NOFS);
1178 atomic_set(&mod->mod_refcount, 1);
1182 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1183 #define obd_mod_put(mod) \
1185 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1186 if ((mod)->mod_open_req) \
1187 ptlrpc_req_finished((mod)->mod_open_req); \
1192 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1193 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1195 /* return 1 if client should be resend request */
1196 static inline int client_should_resend(int resend, struct client_obd *cli)
1198 return atomic_read(&cli->cl_resends) ?
1199 atomic_read(&cli->cl_resends) > resend : 1;
1203 * Return device name for this device
1205 * XXX: lu_device is declared before obd_device, while a pointer pointing
1206 * back to obd_device in lu_device, so this helper function defines here
1207 * instead of in lu_object.h
1209 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1211 return lu_dev->ld_obd->obd_name;
1214 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1219 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1222 /* caller does not care of idx */
1226 /* volatile file, the MDT can be set from name */
1227 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1228 /* if no MDT is specified, use std way */
1229 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1231 /* test for no MDT idx case */
1232 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1233 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1237 /* we have an idx, read it */
1238 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1239 *idx = simple_strtoul(start, &end, 0);
1241 * no digit, no trailing :, negative value
1243 if (((*idx == 0) && (end == start)) ||
1244 (*end != ':') || (*idx < 0))
1249 /* bad format of mdt idx, we cannot return an error
1250 * to caller so we use hash algo
1252 CERROR("Bad volatile file name format: %s\n",
1253 name + LUSTRE_VOLATILE_HDR_LEN);
1257 static inline int cli_brw_size(struct obd_device *obd)
1259 return obd->u.cli.cl_max_pages_per_rpc << PAGE_SHIFT;
1262 #endif /* __OBD_H */