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, 2015, 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/obd.h"
42 #define IOC_OSC_TYPE 'h'
43 #define IOC_OSC_MIN_NR 20
44 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
45 #define IOC_OSC_MAX_NR 50
47 #define IOC_MDC_TYPE 'i'
48 #define IOC_MDC_MIN_NR 20
49 #define IOC_MDC_MAX_NR 50
51 #include "lustre/lustre_idl.h"
52 #include "lustre_lib.h"
54 #include "lustre_export.h"
55 #include "lustre_fid.h"
56 #include "lustre_fld.h"
58 #define MAX_OBD_DEVICES 8192
66 struct lov_oinfo { /* per-stripe data structure */
67 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
68 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
69 int loi_ost_gen; /* generation of this loi_ost_idx */
71 unsigned long loi_kms_valid:1;
72 __u64 loi_kms; /* known minimum size */
73 struct ost_lvb loi_lvb;
74 struct osc_async_rc loi_ar;
77 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
80 oinfo->loi_kms_valid = 1;
83 static inline void loi_init(struct lov_oinfo *loi)
87 struct lov_stripe_md {
90 pid_t lsm_lock_owner; /* debugging */
92 /* maximum possible file size, might change as OSTs status changes,
93 * e.g. disconnected, deactivated */
97 struct ost_id lw_object_oi; /* lov object id/seq */
99 /* LOV-private members start here -- only for use in lov/. */
101 __u32 lw_stripe_size; /* size of the stripe */
102 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
103 __u16 lw_stripe_count; /* number of objects being striped over */
104 __u16 lw_layout_gen; /* generation of the layout */
105 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
108 struct lov_oinfo *lsm_oinfo[0];
111 #define lsm_oi lsm_wire.lw_object_oi
112 #define lsm_magic lsm_wire.lw_magic
113 #define lsm_layout_gen lsm_wire.lw_layout_gen
114 #define lsm_stripe_size lsm_wire.lw_stripe_size
115 #define lsm_pattern lsm_wire.lw_pattern
116 #define lsm_stripe_count lsm_wire.lw_stripe_count
117 #define lsm_pool_name lsm_wire.lw_pool_name
119 static inline bool lsm_is_released(struct lov_stripe_md *lsm)
121 return !!(lsm->lsm_pattern & LOV_PATTERN_F_RELEASED);
124 static inline bool lsm_has_objects(struct lov_stripe_md *lsm)
128 if (lsm_is_released(lsm))
133 static inline int lov_stripe_md_size(unsigned int stripe_count)
135 struct lov_stripe_md lsm;
137 return sizeof(lsm) + stripe_count * sizeof(lsm.lsm_oinfo[0]);
142 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
144 /* obd info for a particular level (lov, osc). */
146 /* Flags used for set request specific flags:
147 - while lock handling, the flags obtained on the enqueue
148 request are set here.
149 - while stats, the flags used for control delay/resend.
150 - while setattr, the flags used for distinguish punch operation
153 /* lsm data specific for every OSC. */
154 struct lov_stripe_md *oi_md;
155 /* obdo data specific for every OSC, if needed at all. */
157 /* statfs data specific for every OSC, if needed at all. */
158 struct obd_statfs *oi_osfs;
159 /* An update callback which is called to update some data on upper
160 * level. E.g. it is used for update lsm->lsm_oinfo at every received
161 * request in osc level for enqueue requests. It is also possible to
162 * update some caller data from LOV layer if needed. */
163 obd_enqueue_update_f oi_cb_up;
166 void lov_stripe_lock(struct lov_stripe_md *md);
167 void lov_stripe_unlock(struct lov_stripe_md *md);
170 struct list_head typ_chain;
171 struct obd_ops *typ_dt_ops;
172 struct md_ops *typ_md_ops;
173 struct dentry *typ_debugfs_entry;
176 struct lu_device_type *typ_lu;
177 spinlock_t obd_type_lock;
178 struct kobject *typ_kobj;
190 LLOG_CONFIG_ORIG_CTXT = 0,
191 LLOG_CONFIG_REPL_CTXT,
192 LLOG_MDS_OST_ORIG_CTXT,
193 LLOG_MDS_OST_REPL_CTXT,
200 LLOG_LOVEA_ORIG_CTXT,
201 LLOG_LOVEA_REPL_CTXT,
202 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
203 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
204 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
205 LLOG_AGENT_ORIG_CTXT, /**< agent requests generation on cdt */
209 struct timeout_item {
210 enum timeout_event ti_event;
211 unsigned long ti_timeout;
214 struct list_head ti_obd_list;
215 struct list_head ti_chain;
218 #define OSC_MAX_RIF_DEFAULT 8
219 #define MDS_OSC_MAX_RIF_DEFAULT 50
220 #define OSC_MAX_RIF_MAX 256
221 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
222 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
223 #define OSC_DEFAULT_RESENDS 10
225 /* possible values for fo_sync_lock_cancel */
227 NEVER_SYNC_ON_CANCEL = 0,
228 BLOCKING_SYNC_ON_CANCEL = 1,
229 ALWAYS_SYNC_ON_CANCEL = 2,
230 NUM_SYNC_ON_CANCEL_STATES
233 #define MDC_MAX_RIF_DEFAULT 8
234 #define MDC_MAX_RIF_MAX 512
239 struct rw_semaphore cl_sem;
240 struct obd_uuid cl_target_uuid;
241 struct obd_import *cl_import; /* ptlrpc connection state */
243 /* max_mds_easize is purely a performance thing so we don't have to
244 * call obd_size_diskmd() all the time. */
245 int cl_default_mds_easize;
246 int cl_max_mds_easize;
247 int cl_default_mds_cookiesize;
248 int cl_max_mds_cookiesize;
250 enum lustre_sec_part cl_sp_me;
251 enum lustre_sec_part cl_sp_to;
252 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
254 /* the grant values are protected by loi_list_lock below */
255 long cl_dirty; /* all _dirty_ in bytes */
256 long cl_dirty_max; /* allowed w/o rpc */
257 long cl_dirty_transit; /* dirty synchronous */
258 long cl_avail_grant; /* bytes of credit for ost */
259 long cl_lost_grant; /* lost credits (trunc) */
261 /* since we allocate grant by blocks, we don't know how many grant will
262 * be used to add a page into cache. As a solution, we reserve maximum
263 * grant before trying to dirty a page and unreserve the rest.
264 * See osc_{reserve|unreserve}_grant for details. */
265 long cl_reserved_grant;
266 struct list_head cl_cache_waiters; /* waiting for cache/grant */
267 unsigned long cl_next_shrink_grant; /* jiffies */
268 struct list_head cl_grant_shrink_list; /* Timeout event list */
269 int cl_grant_shrink_interval; /* seconds */
271 /* A chunk is an optimal size used by osc_extent to determine
272 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
275 int cl_extent_tax; /* extent overhead, by bytes */
277 /* keep track of objects that have lois that contain pages which
278 * have been queued for async brw. this lock also protects the
279 * lists of osc_client_pages that hang off of the loi */
281 * ->cl_loi_list_lock protects consistency of
282 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
283 * ->ap_completion() call-backs are executed under this lock. As we
284 * cannot guarantee that these call-backs never block on all platforms
285 * (as a matter of fact they do block on Mac OS X), type of
286 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
287 * and blocking mutex on Mac OS X. (Alternative is to make this lock
288 * blocking everywhere, but we don't want to slow down fast-path of
289 * our main platform.)
291 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
292 * with client_obd_list_{un,}lock() and
293 * client_obd_list_lock_{init,done}() functions.
295 * NB by Jinshan: though field names are still _loi_, but actually
296 * osc_object{}s are in the list.
298 client_obd_lock_t cl_loi_list_lock;
299 struct list_head cl_loi_ready_list;
300 struct list_head cl_loi_hp_ready_list;
301 struct list_head cl_loi_write_list;
302 struct list_head cl_loi_read_list;
305 /* just a sum of the loi/lop pending numbers to be exported by /proc */
306 atomic_t cl_pending_w_pages;
307 atomic_t cl_pending_r_pages;
308 __u32 cl_max_pages_per_rpc;
309 int cl_max_rpcs_in_flight;
310 struct obd_histogram cl_read_rpc_hist;
311 struct obd_histogram cl_write_rpc_hist;
312 struct obd_histogram cl_read_page_hist;
313 struct obd_histogram cl_write_page_hist;
314 struct obd_histogram cl_read_offset_hist;
315 struct obd_histogram cl_write_offset_hist;
317 /* lru for osc caching pages */
318 struct cl_client_cache *cl_cache;
319 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
320 atomic_t *cl_lru_left;
321 atomic_t cl_lru_busy;
322 atomic_t cl_lru_shrinkers;
323 atomic_t cl_lru_in_list;
324 struct list_head cl_lru_list; /* lru page list */
325 client_obd_lock_t cl_lru_list_lock; /* page list protector */
327 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
328 atomic_t cl_destroy_in_flight;
329 wait_queue_head_t cl_destroy_waitq;
331 struct mdc_rpc_lock *cl_rpc_lock;
332 struct mdc_rpc_lock *cl_close_lock;
335 atomic_t cl_mgc_refcount;
336 struct obd_export *cl_mgc_mgsexp;
338 /* checksumming for data sent over the network */
339 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
340 /* supported checksum types that are worked out at connect time */
341 __u32 cl_supp_cksum_types;
342 /* checksum algorithm to be used */
343 cksum_type_t cl_cksum_type;
345 /* also protected by the poorly named _loi_list_lock lock above */
346 struct osc_async_rc cl_ar;
348 /* used by quotacheck when the servers are older than 2.4 */
349 int cl_qchk_stat; /* quotacheck stat of the peer */
350 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
351 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
352 #warning "please consider removing quotacheck compatibility code"
355 /* sequence manager */
356 struct lu_client_seq *cl_seq;
358 atomic_t cl_resends; /* resend count */
360 /* ptlrpc work for writeback in ptlrpcd context */
361 void *cl_writeback_work;
362 /* hash tables for osc_quota_info */
363 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
366 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
373 struct echo_client_obd {
374 struct obd_export *ec_exp; /* the local connection to osc/lov */
376 struct list_head ec_objects;
377 struct list_head ec_locks;
381 /* Generic subset of OSTs */
383 __u32 *op_array; /* array of index of
385 unsigned int op_count; /* number of OSTs in the array */
386 unsigned int op_size; /* allocated size of lp_array */
387 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
390 /* Round-robin allocator data */
392 __u32 lqr_start_idx; /* start index of new inode */
393 __u32 lqr_offset_idx; /* aliasing for start_idx */
394 int lqr_start_count; /* reseed counter */
395 struct ost_pool lqr_pool; /* round-robin optimized list */
396 unsigned long lqr_dirty:1; /* recalc round-robin list */
399 /* allow statfs data caching for 1 second */
400 #define OBD_STATFS_CACHE_SECONDS 1
402 struct lov_statfs_data {
403 struct obd_info lsd_oi;
404 struct obd_statfs lsd_statfs;
407 /* Stripe placement optimization */
409 struct list_head lq_oss_list; /* list of OSSs that targets use */
410 struct rw_semaphore lq_rw_sem;
411 __u32 lq_active_oss_count;
412 unsigned int lq_prio_free; /* priority for free space */
413 unsigned int lq_threshold_rr;/* priority for rr */
414 struct lov_qos_rr lq_rr; /* round robin qos data */
415 unsigned long lq_dirty:1, /* recalc qos data */
416 lq_same_space:1,/* the ost's all have approx.
417 the same space avail */
418 lq_reset:1, /* zero current penalties */
419 lq_statfs_in_progress:1; /* statfs op in
421 /* qos statfs data */
422 struct lov_statfs_data *lq_statfs_data;
423 wait_queue_head_t lq_statfs_waitq; /* waitqueue to notify statfs
424 * requests completion */
427 struct lov_tgt_desc {
428 struct list_head ltd_kill;
429 struct obd_uuid ltd_uuid;
430 struct obd_device *ltd_obd;
431 struct obd_export *ltd_exp;
433 __u32 ltd_index; /* index in lov_obd->tgts */
434 unsigned long ltd_active:1,/* is this target up for requests */
435 ltd_activate:1,/* should target be activated */
436 ltd_reap:1; /* should this target be deleted */
440 #define pool_tgt_size(_p) _p->pool_obds.op_size
441 #define pool_tgt_count(_p) _p->pool_obds.op_count
442 #define pool_tgt_array(_p) _p->pool_obds.op_array
443 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
446 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
447 struct ost_pool pool_obds; /* pool members */
448 atomic_t pool_refcount; /* pool ref. counter */
449 struct lov_qos_rr pool_rr; /* round robin qos */
450 struct hlist_node pool_hash; /* access by poolname */
451 struct list_head pool_list; /* serial access */
452 struct dentry *pool_debugfs_entry; /* file in /proc */
453 struct obd_device *pool_lobd; /* obd of the lov/lod to which
454 * this pool belongs */
458 struct lov_desc desc;
459 struct lov_tgt_desc **lov_tgts; /* sparse array */
460 struct ost_pool lov_packed; /* all OSTs in a packed
462 struct mutex lov_lock;
463 struct obd_connect_data lov_ocd;
464 atomic_t lov_refcount;
465 __u32 lov_tgt_count; /* how many OBD's */
466 __u32 lov_active_tgt_count; /* how many active */
467 __u32 lov_death_row;/* tgts scheduled to be deleted */
468 __u32 lov_tgt_size; /* size of tgts array */
471 struct cfs_hash *lov_pools_hash_body; /* used for key access */
472 struct list_head lov_pool_list; /* used for sequential access */
473 struct dentry *lov_pool_debugfs_entry;
474 enum lustre_sec_part lov_sp_me;
476 /* Cached LRU pages from upper layer */
479 struct rw_semaphore lov_notify_lock;
481 struct kobject *lov_tgts_kobj;
484 struct lmv_tgt_desc {
485 struct obd_uuid ltd_uuid;
486 struct obd_export *ltd_exp;
488 struct mutex ltd_fid_mutex;
489 unsigned long ltd_active:1; /* target up for requests */
492 enum placement_policy {
493 PLACEMENT_CHAR_POLICY = 0,
494 PLACEMENT_NID_POLICY = 1,
495 PLACEMENT_INVAL_POLICY = 2,
501 struct lu_client_fld lmv_fld;
503 enum placement_policy lmv_placement;
504 struct lmv_desc desc;
505 struct obd_uuid cluuid;
506 struct obd_export *exp;
508 struct mutex init_mutex;
513 int max_def_cookiesize;
516 int tgts_size; /* size of tgts array */
517 struct lmv_tgt_desc **tgts;
519 struct obd_connect_data conn_data;
520 struct kobject *lmv_tgts_kobj;
523 struct niobuf_local {
524 __u64 lnb_file_offset;
525 __u32 lnb_page_offset;
529 struct dentry *dentry;
534 #define LUSTRE_FLD_NAME "fld"
535 #define LUSTRE_SEQ_NAME "seq"
537 #define LUSTRE_MDD_NAME "mdd"
538 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
539 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
540 #define LUSTRE_VVP_NAME "vvp"
541 #define LUSTRE_LMV_NAME "lmv"
542 #define LUSTRE_SLP_NAME "slp"
543 #define LUSTRE_LOD_NAME "lod"
544 #define LUSTRE_OSP_NAME "osp"
545 #define LUSTRE_LWP_NAME "lwp"
547 /* obd device type names */
548 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
549 #define LUSTRE_MDS_NAME "mds"
550 #define LUSTRE_MDT_NAME "mdt"
551 #define LUSTRE_MDC_NAME "mdc"
552 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
553 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
554 #define LUSTRE_OSC_NAME "osc"
555 #define LUSTRE_LOV_NAME "lov"
556 #define LUSTRE_MGS_NAME "mgs"
557 #define LUSTRE_MGC_NAME "mgc"
559 #define LUSTRE_ECHO_NAME "obdecho"
560 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
561 #define LUSTRE_QMT_NAME "qmt"
563 /* Constant obd names (post-rename) */
564 #define LUSTRE_MDS_OBDNAME "MDS"
565 #define LUSTRE_OSS_OBDNAME "OSS"
566 #define LUSTRE_MGS_OBDNAME "MGS"
567 #define LUSTRE_MGC_OBDNAME "MGC"
569 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
570 #define N_LOCAL_TEMP_PAGE 0x10000000
572 struct obd_trans_info {
575 /* Only used on the server side for tracking acks. */
576 struct oti_req_ack_lock {
577 struct lustre_handle lock;
581 struct llog_cookie oti_onecookie;
582 struct llog_cookie *oti_logcookies;
584 /** synchronous write is needed */
585 unsigned long oti_sync_write:1;
587 /* initial thread handling transaction */
588 struct ptlrpc_thread *oti_thread;
591 __u64 oti_pre_version;
595 struct obd_uuid *oti_ost_uuid;
598 static inline void oti_init(struct obd_trans_info *oti,
599 struct ptlrpc_request *req)
603 memset(oti, 0, sizeof(*oti));
608 oti->oti_xid = req->rq_xid;
609 /** VBR: take versions from request */
610 if (req->rq_reqmsg != NULL &&
611 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
612 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
614 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
615 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
618 /** called from mds_create_objects */
619 if (req->rq_repmsg != NULL)
620 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
621 oti->oti_thread = req->rq_svc_thread;
622 if (req->rq_reqmsg != NULL)
623 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
626 static inline void oti_alloc_cookies(struct obd_trans_info *oti,
632 if (num_cookies == 1)
633 oti->oti_logcookies = &oti->oti_onecookie;
635 oti->oti_logcookies = libcfs_kvzalloc(num_cookies * sizeof(oti->oti_onecookie),
638 oti->oti_numcookies = num_cookies;
641 static inline void oti_free_cookies(struct obd_trans_info *oti)
643 if (!oti || !oti->oti_logcookies)
646 if (oti->oti_logcookies == &oti->oti_onecookie)
647 LASSERT(oti->oti_numcookies == 1);
649 kvfree(oti->oti_logcookies);
651 oti->oti_logcookies = NULL;
652 oti->oti_numcookies = 0;
656 * Events signalled through obd_notify() upcall-chain.
658 enum obd_notify_event {
661 /* Device connect start */
663 /* Device activated */
665 /* Device deactivated */
667 /* Device disconnected */
669 /* Connect data for import were changed */
672 OBD_NOTIFY_SYNC_NONBLOCK,
674 /* Configuration event */
676 /* Administratively deactivate/activate event */
677 OBD_NOTIFY_DEACTIVATE,
682 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
683 * and liblustre being main examples).
685 struct obd_notify_upcall {
686 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
687 enum obd_notify_event ev, void *owner, void *data);
688 /* Opaque datum supplied by upper layer listener */
692 struct target_recovery_data {
693 svc_handler_t trd_recovery_handler;
694 pid_t trd_processing_task;
695 struct completion trd_starting;
696 struct completion trd_finishing;
699 struct obd_llog_group {
701 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
702 wait_queue_head_t olg_waitq;
704 struct mutex olg_cat_processing;
707 /* corresponds to one of the obd's */
708 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
709 #define OBD_DEV_BY_DEVNAME 0xffffd0de
711 struct lvfs_run_ctxt {
712 struct dt_device *dt;
716 struct obd_type *obd_type;
719 /* common and UUID name of this device */
720 char obd_name[MAX_OBD_NAME];
721 struct obd_uuid obd_uuid;
723 struct lu_device *obd_lu_dev;
726 /* bitfield modification is protected by obd_dev_lock */
727 unsigned long obd_attached:1, /* finished attach */
728 obd_set_up:1, /* finished setup */
729 obd_version_recov:1, /* obd uses version checking */
730 obd_replayable:1, /* recovery is enabled; inform clients */
731 obd_no_transno:1, /* no committed-transno notification */
732 obd_no_recov:1, /* fail instead of retry messages */
733 obd_stopping:1, /* started cleanup */
734 obd_starting:1, /* started setup */
735 obd_force:1, /* cleanup with > 0 obd refcount */
736 obd_fail:1, /* cleanup with failover */
737 obd_async_recov:1, /* allow asynchronous orphan cleanup */
738 obd_no_conn:1, /* deny new connections */
739 obd_inactive:1, /* device active/inactive
740 * (for /proc/status only!!) */
741 obd_no_ir:1, /* no imperative recovery. */
742 obd_process_conf:1; /* device is processing mgs config */
743 /* use separate field as it is set in interrupt to don't mess with
744 * protection of other bits using _bh lock */
745 unsigned long obd_recovery_expired:1;
746 /* uuid-export hash body */
747 struct cfs_hash *obd_uuid_hash;
748 atomic_t obd_refcount;
749 wait_queue_head_t obd_refcount_waitq;
750 struct list_head obd_exports;
751 struct list_head obd_unlinked_exports;
752 struct list_head obd_delayed_exports;
754 spinlock_t obd_nid_lock;
755 struct ldlm_namespace *obd_namespace;
756 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
757 /* a spinlock is OK for what we do now, may need a semaphore later */
758 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
759 struct mutex obd_dev_mutex;
760 __u64 obd_last_committed;
761 spinlock_t obd_osfs_lock;
762 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
764 struct lvfs_run_ctxt obd_lvfs_ctxt;
765 struct obd_llog_group obd_olg; /* default llog group */
766 struct obd_device *obd_observer;
767 struct rw_semaphore obd_observer_link_sem;
768 struct obd_notify_upcall obd_upcall;
769 struct obd_export *obd_self_export;
772 struct client_obd cli;
773 struct echo_client_obd echo_client;
777 /* Fields used by LProcFS */
778 unsigned int obd_cntr_base;
779 struct lprocfs_stats *obd_stats;
781 unsigned int md_cntr_base;
782 struct lprocfs_stats *md_stats;
784 struct dentry *obd_debugfs_entry;
785 struct dentry *obd_svc_debugfs_entry;
786 struct lprocfs_stats *obd_svc_stats;
787 atomic_t obd_evict_inprogress;
788 wait_queue_head_t obd_evict_inprogress_waitq;
789 struct list_head obd_evict_list; /* protected with pet_lock */
792 * Ldlm pool part. Save last calculated SLV and Limit.
794 rwlock_t obd_pool_lock;
799 * A list of outstanding class_incref()'s against this obd. For
802 struct lu_ref obd_reference;
804 int obd_conn_inprogress;
806 struct kobject obd_kobj; /* sysfs object */
807 struct completion obd_kobj_unregister;
810 #define OBD_LLOG_FL_SENDNOW 0x0001
811 #define OBD_LLOG_FL_EXIT 0x0002
813 enum obd_cleanup_stage {
814 /* Special case hack for MDS LOVs */
816 /* can be directly mapped to .ldto_device_fini() */
820 /* get/set_info keys */
821 #define KEY_ASYNC "async"
822 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
823 #define KEY_BLOCKSIZE "blocksize"
824 #define KEY_CHANGELOG_CLEAR "changelog_clear"
825 #define KEY_FID2PATH "fid2path"
826 #define KEY_CHECKSUM "checksum"
827 #define KEY_CLEAR_FS "clear_fs"
828 #define KEY_CONN_DATA "conn_data"
829 #define KEY_EVICT_BY_NID "evict_by_nid"
830 #define KEY_FIEMAP "fiemap"
831 #define KEY_FLUSH_CTX "flush_ctx"
832 #define KEY_GRANT_SHRINK "grant_shrink"
833 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
834 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
835 #define KEY_INIT_RECOV "initial_recov"
836 #define KEY_INTERMDS "inter_mds"
837 #define KEY_LAST_ID "last_id"
838 #define KEY_LAST_FID "last_fid"
839 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
840 #define KEY_LOVDESC "lovdesc"
841 #define KEY_LOV_IDX "lov_idx"
842 #define KEY_MAX_EASIZE "max_easize"
843 #define KEY_DEFAULT_EASIZE "default_easize"
844 #define KEY_MDS_CONN "mds_conn"
845 #define KEY_MGSSEC "mgssec"
846 #define KEY_NEXT_ID "next_id"
847 #define KEY_READ_ONLY "read-only"
848 #define KEY_REGISTER_TARGET "register_target"
849 #define KEY_SET_FS "set_fs"
850 #define KEY_TGT_COUNT "tgt_count"
851 /* KEY_SET_INFO in lustre_idl.h */
852 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
853 #define KEY_CONNECT_FLAG "connect_flags"
854 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
856 #define KEY_CACHE_SET "cache_set"
857 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
858 #define KEY_CHANGELOG_INDEX "changelog_index"
862 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
863 #define IT_OPEN (1 << 0)
864 #define IT_CREAT (1 << 1)
865 #define IT_READDIR (1 << 2)
866 #define IT_GETATTR (1 << 3)
867 #define IT_LOOKUP (1 << 4)
868 #define IT_UNLINK (1 << 5)
869 #define IT_TRUNC (1 << 6)
870 #define IT_GETXATTR (1 << 7)
871 #define IT_EXEC (1 << 8)
872 #define IT_PIN (1 << 9)
873 #define IT_LAYOUT (1 << 10)
874 #define IT_QUOTA_DQACQ (1 << 11)
875 #define IT_QUOTA_CONN (1 << 12)
876 #define IT_SETXATTR (1 << 13)
878 static inline int it_to_lock_mode(struct lookup_intent *it)
880 /* CREAT needs to be tested before open (both could be set) */
881 if (it->it_op & IT_CREAT)
883 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
886 else if (it->it_op & IT_GETXATTR)
888 else if (it->it_op & IT_SETXATTR)
891 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
896 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
897 struct lu_fid op_fid2; /* operation fid2 (usually child) */
898 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
899 struct lu_fid op_fid4; /* to the operation locks. */
900 u32 op_mds; /* what mds server open will go to */
901 struct lustre_handle op_handle;
906 struct lmv_stripe_md *op_mea1;
907 struct lmv_stripe_md *op_mea2;
908 __u32 op_suppgids[2];
914 /* iattr fields and blocks. */
915 struct iattr op_attr;
916 unsigned int op_attr_flags;
918 loff_t op_attr_blocks;
920 /* Size-on-MDS epoch and flags. */
924 /* Various operation flags. */
925 enum mds_op_bias op_bias;
930 /* Used by readdir */
933 /* Used by readdir */
936 /* used to transfer info between the stacks of MD client
937 * see enum op_cli_flags */
940 /* File object data version for HSM release, on client */
941 __u64 op_data_version;
942 struct lustre_handle op_lease_handle;
946 CLI_SET_MEA = 1 << 0,
947 CLI_RM_ENTRY = 1 << 1,
950 struct md_enqueue_info;
951 /* metadata stat-ahead */
953 struct md_enqueue_info {
954 struct md_op_data mi_data;
955 struct lookup_intent mi_it;
956 struct lustre_handle mi_lockh;
957 struct inode *mi_dir;
958 int (*mi_cb)(struct ptlrpc_request *req,
959 struct md_enqueue_info *minfo, int rc);
961 unsigned int mi_generation;
965 struct module *owner;
966 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
967 void *karg, void __user *uarg);
968 int (*get_info)(const struct lu_env *env, struct obd_export *,
969 __u32 keylen, void *key, __u32 *vallen, void *val,
970 struct lov_stripe_md *lsm);
971 int (*set_info_async)(const struct lu_env *, struct obd_export *,
972 __u32 keylen, void *key,
973 __u32 vallen, void *val,
974 struct ptlrpc_request_set *set);
975 int (*attach)(struct obd_device *dev, u32 len, void *data);
976 int (*detach)(struct obd_device *dev);
977 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
978 int (*precleanup)(struct obd_device *dev,
979 enum obd_cleanup_stage cleanup_stage);
980 int (*cleanup)(struct obd_device *dev);
981 int (*process_config)(struct obd_device *dev, u32 len, void *data);
982 int (*postrecov)(struct obd_device *dev);
983 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
985 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
986 /* connect to the target device with given connection
987 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
988 * granted by the target, which are guaranteed to be a subset of flags
989 * asked for. If @ocd == NULL, use default parameters. */
990 int (*connect)(const struct lu_env *env,
991 struct obd_export **exp, struct obd_device *src,
992 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
994 int (*reconnect)(const struct lu_env *env,
995 struct obd_export *exp, struct obd_device *src,
996 struct obd_uuid *cluuid,
997 struct obd_connect_data *ocd,
999 int (*disconnect)(struct obd_export *exp);
1001 /* Initialize/finalize fids infrastructure. */
1002 int (*fid_init)(struct obd_device *obd,
1003 struct obd_export *exp, enum lu_cli_type type);
1004 int (*fid_fini)(struct obd_device *obd);
1006 /* Allocate new fid according to passed @hint. */
1007 int (*fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1008 struct md_op_data *op_data);
1011 * Object with @fid is getting deleted, we may want to do something
1014 int (*statfs)(const struct lu_env *, struct obd_export *exp,
1015 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1016 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1017 __u64 max_age, struct ptlrpc_request_set *set);
1018 int (*packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1019 struct lov_stripe_md *mem_src);
1020 int (*unpackmd)(struct obd_export *exp,
1021 struct lov_stripe_md **mem_tgt,
1022 struct lov_mds_md *disk_src, int disk_len);
1023 int (*preallocate)(struct lustre_handle *, u32 *req, u64 *ids);
1024 int (*create)(const struct lu_env *env, struct obd_export *exp,
1025 struct obdo *oa, struct lov_stripe_md **ea,
1026 struct obd_trans_info *oti);
1027 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
1028 struct obdo *oa, struct lov_stripe_md *ea,
1029 struct obd_trans_info *oti, struct obd_export *md_exp);
1030 int (*setattr)(const struct lu_env *, struct obd_export *exp,
1031 struct obd_info *oinfo, struct obd_trans_info *oti);
1032 int (*setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1033 struct obd_trans_info *oti,
1034 struct ptlrpc_request_set *rqset);
1035 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
1036 struct obd_info *oinfo);
1037 int (*getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1038 struct ptlrpc_request_set *set);
1039 int (*adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1040 u64 size, int shrink);
1041 int (*preprw)(const struct lu_env *env, int cmd,
1042 struct obd_export *exp, struct obdo *oa, int objcount,
1043 struct obd_ioobj *obj, struct niobuf_remote *remote,
1044 int *nr_pages, struct niobuf_local *local,
1045 struct obd_trans_info *oti);
1046 int (*commitrw)(const struct lu_env *env, int cmd,
1047 struct obd_export *exp, struct obdo *oa,
1048 int objcount, struct obd_ioobj *obj,
1049 struct niobuf_remote *remote, int pages,
1050 struct niobuf_local *local,
1051 struct obd_trans_info *oti, int rc);
1052 int (*find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1053 ldlm_iterator_t it, void *data);
1054 int (*init_export)(struct obd_export *exp);
1055 int (*destroy_export)(struct obd_export *exp);
1057 /* metadata-only methods */
1058 int (*import_event)(struct obd_device *, struct obd_import *,
1059 enum obd_import_event);
1061 int (*notify)(struct obd_device *obd, struct obd_device *watched,
1062 enum obd_notify_event ev, void *data);
1064 int (*health_check)(const struct lu_env *env, struct obd_device *);
1065 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
1068 int (*quotacheck)(struct obd_device *, struct obd_export *,
1069 struct obd_quotactl *);
1070 int (*quotactl)(struct obd_device *, struct obd_export *,
1071 struct obd_quotactl *);
1074 int (*pool_new)(struct obd_device *obd, char *poolname);
1075 int (*pool_del)(struct obd_device *obd, char *poolname);
1076 int (*pool_add)(struct obd_device *obd, char *poolname,
1078 int (*pool_rem)(struct obd_device *obd, char *poolname,
1080 void (*getref)(struct obd_device *obd);
1081 void (*putref)(struct obd_device *obd);
1083 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1084 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1085 * Also, add a wrapper function in include/linux/obd_class.h. */
1089 LUSTRE_OPC_MKDIR = (1 << 0),
1090 LUSTRE_OPC_SYMLINK = (1 << 1),
1091 LUSTRE_OPC_MKNOD = (1 << 2),
1092 LUSTRE_OPC_CREATE = (1 << 3),
1093 LUSTRE_OPC_ANY = (1 << 4)
1096 /* lmv structures */
1097 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1098 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1099 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1101 #define MAX_HASH_SIZE_32 0x7fffffffUL
1102 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1103 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1106 struct mdt_body *body;
1107 struct lov_stripe_md *lsm;
1108 struct lmv_stripe_md *mea;
1109 #ifdef CONFIG_FS_POSIX_ACL
1110 struct posix_acl *posix_acl;
1112 struct mdt_remote_perm *remote_perm;
1115 struct md_open_data {
1116 struct obd_client_handle *mod_och;
1117 struct ptlrpc_request *mod_open_req;
1118 struct ptlrpc_request *mod_close_req;
1119 atomic_t mod_refcount;
1123 struct lookup_intent;
1126 int (*getstatus)(struct obd_export *, struct lu_fid *);
1127 int (*null_inode)(struct obd_export *, const struct lu_fid *);
1128 int (*find_cbdata)(struct obd_export *, const struct lu_fid *,
1129 ldlm_iterator_t, void *);
1130 int (*close)(struct obd_export *, struct md_op_data *,
1131 struct md_open_data *, struct ptlrpc_request **);
1132 int (*create)(struct obd_export *, struct md_op_data *,
1133 const void *, int, int, __u32, __u32, cfs_cap_t,
1134 __u64, struct ptlrpc_request **);
1135 int (*done_writing)(struct obd_export *, struct md_op_data *,
1136 struct md_open_data *);
1137 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1138 struct lookup_intent *, struct md_op_data *,
1139 struct lustre_handle *, void *, int,
1140 struct ptlrpc_request **, __u64);
1141 int (*getattr)(struct obd_export *, struct md_op_data *,
1142 struct ptlrpc_request **);
1143 int (*getattr_name)(struct obd_export *, struct md_op_data *,
1144 struct ptlrpc_request **);
1145 int (*intent_lock)(struct obd_export *, struct md_op_data *,
1146 void *, int, struct lookup_intent *, int,
1147 struct ptlrpc_request **,
1148 ldlm_blocking_callback, __u64);
1149 int (*link)(struct obd_export *, struct md_op_data *,
1150 struct ptlrpc_request **);
1151 int (*rename)(struct obd_export *, struct md_op_data *,
1152 const char *, int, const char *, int,
1153 struct ptlrpc_request **);
1154 int (*is_subdir)(struct obd_export *, const struct lu_fid *,
1155 const struct lu_fid *,
1156 struct ptlrpc_request **);
1157 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
1158 int, void *, int, struct ptlrpc_request **,
1159 struct md_open_data **mod);
1160 int (*sync)(struct obd_export *, const struct lu_fid *,
1161 struct ptlrpc_request **);
1162 int (*readpage)(struct obd_export *, struct md_op_data *,
1163 struct page **, struct ptlrpc_request **);
1165 int (*unlink)(struct obd_export *, struct md_op_data *,
1166 struct ptlrpc_request **);
1168 int (*setxattr)(struct obd_export *, const struct lu_fid *,
1169 u64, const char *, const char *, int, int, int, __u32,
1170 struct ptlrpc_request **);
1172 int (*getxattr)(struct obd_export *, const struct lu_fid *,
1173 u64, const char *, const char *, int, int, int,
1174 struct ptlrpc_request **);
1176 int (*init_ea_size)(struct obd_export *, int, int, int, int);
1178 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1179 struct obd_export *, struct obd_export *,
1180 struct lustre_md *);
1182 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
1184 int (*set_open_replay_data)(struct obd_export *,
1185 struct obd_client_handle *,
1186 struct lookup_intent *);
1187 int (*clear_open_replay_data)(struct obd_export *,
1188 struct obd_client_handle *);
1189 int (*set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1191 ldlm_mode_t (*lock_match)(struct obd_export *, __u64,
1192 const struct lu_fid *, ldlm_type_t,
1193 ldlm_policy_data_t *, ldlm_mode_t,
1194 struct lustre_handle *);
1196 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
1197 ldlm_policy_data_t *, ldlm_mode_t,
1198 ldlm_cancel_flags_t flags, void *opaque);
1200 int (*get_remote_perm)(struct obd_export *, const struct lu_fid *,
1201 __u32, struct ptlrpc_request **);
1203 int (*intent_getattr_async)(struct obd_export *,
1204 struct md_enqueue_info *,
1205 struct ldlm_enqueue_info *);
1207 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
1208 struct lu_fid *, __u64 *bits);
1211 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1212 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1213 * wrapper function in include/linux/obd_class.h.
1217 struct lsm_operations {
1218 void (*lsm_free)(struct lov_stripe_md *);
1219 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1220 struct obd_export *md_exp);
1221 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, u64 *,
1223 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, u64 *,
1225 int (*lsm_lmm_verify)(struct lov_mds_md *lmm, int lmm_bytes,
1226 __u16 *stripe_count);
1227 int (*lsm_unpackmd)(struct lov_obd *lov, struct lov_stripe_md *lsm,
1228 struct lov_mds_md *lmm);
1231 extern const struct lsm_operations lsm_v1_ops;
1232 extern const struct lsm_operations lsm_v3_ops;
1233 static inline const struct lsm_operations *lsm_op_find(int magic)
1241 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1246 /* Requests for obd_extent_calc() */
1247 #define OBD_CALC_STRIPE_START 1
1248 #define OBD_CALC_STRIPE_END 2
1250 static inline struct md_open_data *obd_mod_alloc(void)
1252 struct md_open_data *mod;
1254 mod = kzalloc(sizeof(*mod), GFP_NOFS);
1257 atomic_set(&mod->mod_refcount, 1);
1261 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1262 #define obd_mod_put(mod) \
1264 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1265 if ((mod)->mod_open_req) \
1266 ptlrpc_req_finished((mod)->mod_open_req); \
1271 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1272 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1274 /* return 1 if client should be resend request */
1275 static inline int client_should_resend(int resend, struct client_obd *cli)
1277 return atomic_read(&cli->cl_resends) ?
1278 atomic_read(&cli->cl_resends) > resend : 1;
1282 * Return device name for this device
1284 * XXX: lu_device is declared before obd_device, while a pointer pointing
1285 * back to obd_device in lu_device, so this helper function defines here
1286 * instead of in lu_object.h
1288 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1290 return lu_dev->ld_obd->obd_name;
1293 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1298 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1301 /* caller does not care of idx */
1305 /* volatile file, the MDT can be set from name */
1306 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1307 /* if no MDT is specified, use std way */
1308 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1310 /* test for no MDT idx case */
1311 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1312 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1316 /* we have an idx, read it */
1317 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1318 *idx = simple_strtoul(start, &end, 0);
1320 * no digit, no trailing :, negative value
1322 if (((*idx == 0) && (end == start)) ||
1323 (*end != ':') || (*idx < 0))
1328 /* bad format of mdt idx, we cannot return an error
1329 * to caller so we use hash algo */
1330 CERROR("Bad volatile file name format: %s\n",
1331 name + LUSTRE_VOLATILE_HDR_LEN);
1335 static inline int cli_brw_size(struct obd_device *obd)
1337 LASSERT(obd != NULL);
1338 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1341 #endif /* __OBD_H */