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 OSC_MAX_RIF_MAX 256
220 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
221 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
222 #define OSC_DEFAULT_RESENDS 10
224 /* possible values for fo_sync_lock_cancel */
226 NEVER_SYNC_ON_CANCEL = 0,
227 BLOCKING_SYNC_ON_CANCEL = 1,
228 ALWAYS_SYNC_ON_CANCEL = 2,
229 NUM_SYNC_ON_CANCEL_STATES
232 #define MDC_MAX_RIF_DEFAULT 8
233 #define MDC_MAX_RIF_MAX 512
238 struct rw_semaphore cl_sem;
239 struct obd_uuid cl_target_uuid;
240 struct obd_import *cl_import; /* ptlrpc connection state */
242 /* max_mds_easize is purely a performance thing so we don't have to
243 * call obd_size_diskmd() all the time. */
244 int cl_default_mds_easize;
245 int cl_max_mds_easize;
246 int cl_default_mds_cookiesize;
247 int cl_max_mds_cookiesize;
249 enum lustre_sec_part cl_sp_me;
250 enum lustre_sec_part cl_sp_to;
251 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
253 /* the grant values are protected by loi_list_lock below */
254 long cl_dirty; /* all _dirty_ in bytes */
255 long cl_dirty_max; /* allowed w/o rpc */
256 long cl_dirty_transit; /* dirty synchronous */
257 long cl_avail_grant; /* bytes of credit for ost */
258 long cl_lost_grant; /* lost credits (trunc) */
260 /* since we allocate grant by blocks, we don't know how many grant will
261 * be used to add a page into cache. As a solution, we reserve maximum
262 * grant before trying to dirty a page and unreserve the rest.
263 * See osc_{reserve|unreserve}_grant for details. */
264 long cl_reserved_grant;
265 struct list_head cl_cache_waiters; /* waiting for cache/grant */
266 unsigned long cl_next_shrink_grant; /* jiffies */
267 struct list_head cl_grant_shrink_list; /* Timeout event list */
268 int cl_grant_shrink_interval; /* seconds */
270 /* A chunk is an optimal size used by osc_extent to determine
271 * the extent size. A chunk is max(PAGE_CACHE_SIZE, OST block size) */
274 int cl_extent_tax; /* extent overhead, by bytes */
276 /* keep track of objects that have lois that contain pages which
277 * have been queued for async brw. this lock also protects the
278 * lists of osc_client_pages that hang off of the loi */
280 * ->cl_loi_list_lock protects consistency of
281 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
282 * ->ap_completion() call-backs are executed under this lock. As we
283 * cannot guarantee that these call-backs never block on all platforms
284 * (as a matter of fact they do block on Mac OS X), type of
285 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
286 * and blocking mutex on Mac OS X. (Alternative is to make this lock
287 * blocking everywhere, but we don't want to slow down fast-path of
288 * our main platform.)
290 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
291 * with client_obd_list_{un,}lock() and
292 * client_obd_list_lock_{init,done}() functions.
294 * NB by Jinshan: though field names are still _loi_, but actually
295 * osc_object{}s are in the list.
297 struct client_obd_lock cl_loi_list_lock;
298 struct list_head cl_loi_ready_list;
299 struct list_head cl_loi_hp_ready_list;
300 struct list_head cl_loi_write_list;
301 struct list_head cl_loi_read_list;
304 /* just a sum of the loi/lop pending numbers to be exported by sysfs */
305 atomic_t cl_pending_w_pages;
306 atomic_t cl_pending_r_pages;
307 __u32 cl_max_pages_per_rpc;
308 int cl_max_rpcs_in_flight;
309 struct obd_histogram cl_read_rpc_hist;
310 struct obd_histogram cl_write_rpc_hist;
311 struct obd_histogram cl_read_page_hist;
312 struct obd_histogram cl_write_page_hist;
313 struct obd_histogram cl_read_offset_hist;
314 struct obd_histogram cl_write_offset_hist;
316 /* lru for osc caching pages */
317 struct cl_client_cache *cl_cache;
318 struct list_head cl_lru_osc; /* member of cl_cache->ccc_lru */
319 atomic_t *cl_lru_left;
320 atomic_t cl_lru_busy;
321 atomic_t cl_lru_shrinkers;
322 atomic_t cl_lru_in_list;
323 struct list_head cl_lru_list; /* lru page list */
324 struct client_obd_lock cl_lru_list_lock; /* page list protector */
326 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
327 atomic_t cl_destroy_in_flight;
328 wait_queue_head_t cl_destroy_waitq;
330 struct mdc_rpc_lock *cl_rpc_lock;
331 struct mdc_rpc_lock *cl_close_lock;
334 atomic_t cl_mgc_refcount;
335 struct obd_export *cl_mgc_mgsexp;
337 /* checksumming for data sent over the network */
338 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
339 /* supported checksum types that are worked out at connect time */
340 __u32 cl_supp_cksum_types;
341 /* checksum algorithm to be used */
342 cksum_type_t cl_cksum_type;
344 /* also protected by the poorly named _loi_list_lock lock above */
345 struct osc_async_rc cl_ar;
347 /* used by quotacheck when the servers are older than 2.4 */
348 int cl_qchk_stat; /* quotacheck stat of the peer */
349 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
350 #if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 7, 50, 0)
351 #warning "please consider removing quotacheck compatibility code"
354 /* sequence manager */
355 struct lu_client_seq *cl_seq;
357 atomic_t cl_resends; /* resend count */
359 /* ptlrpc work for writeback in ptlrpcd context */
360 void *cl_writeback_work;
361 /* hash tables for osc_quota_info */
362 struct cfs_hash *cl_quota_hash[MAXQUOTAS];
365 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
372 struct echo_client_obd {
373 struct obd_export *ec_exp; /* the local connection to osc/lov */
375 struct list_head ec_objects;
376 struct list_head ec_locks;
380 /* Generic subset of OSTs */
382 __u32 *op_array; /* array of index of
384 unsigned int op_count; /* number of OSTs in the array */
385 unsigned int op_size; /* allocated size of lp_array */
386 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
389 /* Round-robin allocator data */
391 __u32 lqr_start_idx; /* start index of new inode */
392 __u32 lqr_offset_idx; /* aliasing for start_idx */
393 int lqr_start_count; /* reseed counter */
394 struct ost_pool lqr_pool; /* round-robin optimized list */
395 unsigned long lqr_dirty:1; /* recalc round-robin list */
398 /* allow statfs data caching for 1 second */
399 #define OBD_STATFS_CACHE_SECONDS 1
401 struct lov_statfs_data {
402 struct obd_info lsd_oi;
403 struct obd_statfs lsd_statfs;
406 /* Stripe placement optimization */
408 struct list_head lq_oss_list; /* list of OSSs that targets use */
409 struct rw_semaphore lq_rw_sem;
410 __u32 lq_active_oss_count;
411 unsigned int lq_prio_free; /* priority for free space */
412 unsigned int lq_threshold_rr;/* priority for rr */
413 struct lov_qos_rr lq_rr; /* round robin qos data */
414 unsigned long lq_dirty:1, /* recalc qos data */
415 lq_same_space:1,/* the ost's all have approx.
416 the same space avail */
417 lq_reset:1, /* zero current penalties */
418 lq_statfs_in_progress:1; /* statfs op in
420 /* qos statfs data */
421 struct lov_statfs_data *lq_statfs_data;
422 wait_queue_head_t lq_statfs_waitq; /* waitqueue to notify statfs
423 * requests completion */
426 struct lov_tgt_desc {
427 struct list_head ltd_kill;
428 struct obd_uuid ltd_uuid;
429 struct obd_device *ltd_obd;
430 struct obd_export *ltd_exp;
432 __u32 ltd_index; /* index in lov_obd->tgts */
433 unsigned long ltd_active:1,/* is this target up for requests */
434 ltd_activate:1,/* should target be activated */
435 ltd_reap:1; /* should this target be deleted */
439 #define pool_tgt_size(_p) _p->pool_obds.op_size
440 #define pool_tgt_count(_p) _p->pool_obds.op_count
441 #define pool_tgt_array(_p) _p->pool_obds.op_array
442 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
445 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
446 struct ost_pool pool_obds; /* pool members */
447 atomic_t pool_refcount; /* pool ref. counter */
448 struct lov_qos_rr pool_rr; /* round robin qos */
449 struct hlist_node pool_hash; /* access by poolname */
450 struct list_head pool_list; /* serial access */
451 struct dentry *pool_debugfs_entry; /* file in debugfs */
452 struct obd_device *pool_lobd; /* obd of the lov/lod to which
453 * this pool belongs */
457 struct lov_desc desc;
458 struct lov_tgt_desc **lov_tgts; /* sparse array */
459 struct ost_pool lov_packed; /* all OSTs in a packed
461 struct mutex lov_lock;
462 struct obd_connect_data lov_ocd;
463 atomic_t lov_refcount;
464 __u32 lov_tgt_count; /* how many OBD's */
465 __u32 lov_active_tgt_count; /* how many active */
466 __u32 lov_death_row;/* tgts scheduled to be deleted */
467 __u32 lov_tgt_size; /* size of tgts array */
470 struct cfs_hash *lov_pools_hash_body; /* used for key access */
471 struct list_head lov_pool_list; /* used for sequential access */
472 struct dentry *lov_pool_debugfs_entry;
473 enum lustre_sec_part lov_sp_me;
475 /* Cached LRU pages from upper layer */
478 struct rw_semaphore lov_notify_lock;
480 struct kobject *lov_tgts_kobj;
483 struct lmv_tgt_desc {
484 struct obd_uuid ltd_uuid;
485 struct obd_export *ltd_exp;
487 struct mutex ltd_fid_mutex;
488 unsigned long ltd_active:1; /* target up for requests */
491 enum placement_policy {
492 PLACEMENT_CHAR_POLICY = 0,
493 PLACEMENT_NID_POLICY = 1,
494 PLACEMENT_INVAL_POLICY = 2,
500 struct lu_client_fld lmv_fld;
502 enum placement_policy lmv_placement;
503 struct lmv_desc desc;
504 struct obd_uuid cluuid;
505 struct obd_export *exp;
507 struct mutex init_mutex;
512 int max_def_cookiesize;
515 int tgts_size; /* size of tgts array */
516 struct lmv_tgt_desc **tgts;
518 struct obd_connect_data conn_data;
519 struct kobject *lmv_tgts_kobj;
522 struct niobuf_local {
523 __u64 lnb_file_offset;
524 __u32 lnb_page_offset;
528 struct dentry *dentry;
533 #define LUSTRE_FLD_NAME "fld"
534 #define LUSTRE_SEQ_NAME "seq"
536 #define LUSTRE_MDD_NAME "mdd"
537 #define LUSTRE_OSD_LDISKFS_NAME "osd-ldiskfs"
538 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
539 #define LUSTRE_VVP_NAME "vvp"
540 #define LUSTRE_LMV_NAME "lmv"
541 #define LUSTRE_SLP_NAME "slp"
542 #define LUSTRE_LOD_NAME "lod"
543 #define LUSTRE_OSP_NAME "osp"
544 #define LUSTRE_LWP_NAME "lwp"
546 /* obd device type names */
547 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
548 #define LUSTRE_MDS_NAME "mds"
549 #define LUSTRE_MDT_NAME "mdt"
550 #define LUSTRE_MDC_NAME "mdc"
551 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
552 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
553 #define LUSTRE_OSC_NAME "osc"
554 #define LUSTRE_LOV_NAME "lov"
555 #define LUSTRE_MGS_NAME "mgs"
556 #define LUSTRE_MGC_NAME "mgc"
558 #define LUSTRE_ECHO_NAME "obdecho"
559 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
560 #define LUSTRE_QMT_NAME "qmt"
562 /* Constant obd names (post-rename) */
563 #define LUSTRE_MDS_OBDNAME "MDS"
564 #define LUSTRE_OSS_OBDNAME "OSS"
565 #define LUSTRE_MGS_OBDNAME "MGS"
566 #define LUSTRE_MGC_OBDNAME "MGC"
568 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
569 #define N_LOCAL_TEMP_PAGE 0x10000000
571 struct obd_trans_info {
574 /* Only used on the server side for tracking acks. */
575 struct oti_req_ack_lock {
576 struct lustre_handle lock;
580 struct llog_cookie oti_onecookie;
581 struct llog_cookie *oti_logcookies;
583 /** synchronous write is needed */
584 unsigned long oti_sync_write:1;
586 /* initial thread handling transaction */
587 struct ptlrpc_thread *oti_thread;
590 __u64 oti_pre_version;
594 struct obd_uuid *oti_ost_uuid;
597 static inline void oti_alloc_cookies(struct obd_trans_info *oti,
603 if (num_cookies == 1)
604 oti->oti_logcookies = &oti->oti_onecookie;
606 oti->oti_logcookies = libcfs_kvzalloc(num_cookies * sizeof(oti->oti_onecookie),
609 oti->oti_numcookies = num_cookies;
612 static inline void oti_free_cookies(struct obd_trans_info *oti)
614 if (!oti || !oti->oti_logcookies)
617 if (oti->oti_logcookies == &oti->oti_onecookie)
618 LASSERT(oti->oti_numcookies == 1);
620 kvfree(oti->oti_logcookies);
622 oti->oti_logcookies = NULL;
623 oti->oti_numcookies = 0;
627 * Events signalled through obd_notify() upcall-chain.
629 enum obd_notify_event {
632 /* Device connect start */
634 /* Device activated */
636 /* Device deactivated */
638 /* Device disconnected */
640 /* Connect data for import were changed */
643 OBD_NOTIFY_SYNC_NONBLOCK,
645 /* Configuration event */
647 /* Administratively deactivate/activate event */
648 OBD_NOTIFY_DEACTIVATE,
653 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
654 * and liblustre being main examples).
656 struct obd_notify_upcall {
657 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
658 enum obd_notify_event ev, void *owner, void *data);
659 /* Opaque datum supplied by upper layer listener */
663 struct target_recovery_data {
664 svc_handler_t trd_recovery_handler;
665 pid_t trd_processing_task;
666 struct completion trd_starting;
667 struct completion trd_finishing;
670 struct obd_llog_group {
672 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
673 wait_queue_head_t olg_waitq;
675 struct mutex olg_cat_processing;
678 /* corresponds to one of the obd's */
679 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
680 #define OBD_DEV_BY_DEVNAME 0xffffd0de
682 struct lvfs_run_ctxt {
683 struct dt_device *dt;
687 struct obd_type *obd_type;
690 /* common and UUID name of this device */
691 char obd_name[MAX_OBD_NAME];
692 struct obd_uuid obd_uuid;
694 struct lu_device *obd_lu_dev;
697 /* bitfield modification is protected by obd_dev_lock */
698 unsigned long obd_attached:1, /* finished attach */
699 obd_set_up:1, /* finished setup */
700 obd_version_recov:1, /* obd uses version checking */
701 obd_replayable:1, /* recovery is enabled; inform clients */
702 obd_no_transno:1, /* no committed-transno notification */
703 obd_no_recov:1, /* fail instead of retry messages */
704 obd_stopping:1, /* started cleanup */
705 obd_starting:1, /* started setup */
706 obd_force:1, /* cleanup with > 0 obd refcount */
707 obd_fail:1, /* cleanup with failover */
708 obd_async_recov:1, /* allow asynchronous orphan cleanup */
709 obd_no_conn:1, /* deny new connections */
710 obd_inactive:1, /* device active/inactive
711 * (for sysfs status only!!)
713 obd_no_ir:1, /* no imperative recovery. */
714 obd_process_conf:1; /* device is processing mgs config */
715 /* use separate field as it is set in interrupt to don't mess with
716 * protection of other bits using _bh lock */
717 unsigned long obd_recovery_expired:1;
718 /* uuid-export hash body */
719 struct cfs_hash *obd_uuid_hash;
720 atomic_t obd_refcount;
721 wait_queue_head_t obd_refcount_waitq;
722 struct list_head obd_exports;
723 struct list_head obd_unlinked_exports;
724 struct list_head obd_delayed_exports;
726 spinlock_t obd_nid_lock;
727 struct ldlm_namespace *obd_namespace;
728 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
729 /* a spinlock is OK for what we do now, may need a semaphore later */
730 spinlock_t obd_dev_lock; /* protect OBD bitfield above */
731 struct mutex obd_dev_mutex;
732 __u64 obd_last_committed;
733 spinlock_t obd_osfs_lock;
734 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
736 struct lvfs_run_ctxt obd_lvfs_ctxt;
737 struct obd_llog_group obd_olg; /* default llog group */
738 struct obd_device *obd_observer;
739 struct rw_semaphore obd_observer_link_sem;
740 struct obd_notify_upcall obd_upcall;
741 struct obd_export *obd_self_export;
744 struct client_obd cli;
745 struct echo_client_obd echo_client;
749 /* Fields used by LProcFS */
750 unsigned int obd_cntr_base;
751 struct lprocfs_stats *obd_stats;
753 unsigned int md_cntr_base;
754 struct lprocfs_stats *md_stats;
756 struct dentry *obd_debugfs_entry;
757 struct dentry *obd_svc_debugfs_entry;
758 struct lprocfs_stats *obd_svc_stats;
759 atomic_t obd_evict_inprogress;
760 wait_queue_head_t obd_evict_inprogress_waitq;
761 struct list_head obd_evict_list; /* protected with pet_lock */
764 * Ldlm pool part. Save last calculated SLV and Limit.
766 rwlock_t obd_pool_lock;
771 * A list of outstanding class_incref()'s against this obd. For
774 struct lu_ref obd_reference;
776 int obd_conn_inprogress;
778 struct kobject obd_kobj; /* sysfs object */
779 struct completion obd_kobj_unregister;
782 #define OBD_LLOG_FL_SENDNOW 0x0001
783 #define OBD_LLOG_FL_EXIT 0x0002
785 enum obd_cleanup_stage {
786 /* Special case hack for MDS LOVs */
788 /* can be directly mapped to .ldto_device_fini() */
792 /* get/set_info keys */
793 #define KEY_ASYNC "async"
794 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
795 #define KEY_BLOCKSIZE "blocksize"
796 #define KEY_CHANGELOG_CLEAR "changelog_clear"
797 #define KEY_FID2PATH "fid2path"
798 #define KEY_CHECKSUM "checksum"
799 #define KEY_CLEAR_FS "clear_fs"
800 #define KEY_CONN_DATA "conn_data"
801 #define KEY_EVICT_BY_NID "evict_by_nid"
802 #define KEY_FIEMAP "fiemap"
803 #define KEY_FLUSH_CTX "flush_ctx"
804 #define KEY_GRANT_SHRINK "grant_shrink"
805 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
806 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
807 #define KEY_INIT_RECOV "initial_recov"
808 #define KEY_INTERMDS "inter_mds"
809 #define KEY_LAST_ID "last_id"
810 #define KEY_LAST_FID "last_fid"
811 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
812 #define KEY_LOVDESC "lovdesc"
813 #define KEY_LOV_IDX "lov_idx"
814 #define KEY_MAX_EASIZE "max_easize"
815 #define KEY_DEFAULT_EASIZE "default_easize"
816 #define KEY_MDS_CONN "mds_conn"
817 #define KEY_MGSSEC "mgssec"
818 #define KEY_NEXT_ID "next_id"
819 #define KEY_READ_ONLY "read-only"
820 #define KEY_REGISTER_TARGET "register_target"
821 #define KEY_SET_FS "set_fs"
822 #define KEY_TGT_COUNT "tgt_count"
823 /* KEY_SET_INFO in lustre_idl.h */
824 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
825 #define KEY_CONNECT_FLAG "connect_flags"
826 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
828 #define KEY_CACHE_SET "cache_set"
829 #define KEY_CACHE_LRU_SHRINK "cache_lru_shrink"
830 #define KEY_CHANGELOG_INDEX "changelog_index"
834 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
835 #define IT_OPEN (1 << 0)
836 #define IT_CREAT (1 << 1)
837 #define IT_READDIR (1 << 2)
838 #define IT_GETATTR (1 << 3)
839 #define IT_LOOKUP (1 << 4)
840 #define IT_UNLINK (1 << 5)
841 #define IT_TRUNC (1 << 6)
842 #define IT_GETXATTR (1 << 7)
843 #define IT_EXEC (1 << 8)
844 #define IT_PIN (1 << 9)
845 #define IT_LAYOUT (1 << 10)
846 #define IT_QUOTA_DQACQ (1 << 11)
847 #define IT_QUOTA_CONN (1 << 12)
848 #define IT_SETXATTR (1 << 13)
850 static inline int it_to_lock_mode(struct lookup_intent *it)
852 /* CREAT needs to be tested before open (both could be set) */
853 if (it->it_op & IT_CREAT)
855 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
858 else if (it->it_op & IT_GETXATTR)
860 else if (it->it_op & IT_SETXATTR)
863 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
868 struct lu_fid op_fid1; /* operation fid1 (usually parent) */
869 struct lu_fid op_fid2; /* operation fid2 (usually child) */
870 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
871 struct lu_fid op_fid4; /* to the operation locks. */
872 u32 op_mds; /* what mds server open will go to */
873 struct lustre_handle op_handle;
878 struct lmv_stripe_md *op_mea1;
879 struct lmv_stripe_md *op_mea2;
880 __u32 op_suppgids[2];
886 /* iattr fields and blocks. */
887 struct iattr op_attr;
888 unsigned int op_attr_flags;
890 loff_t op_attr_blocks;
892 /* Size-on-MDS epoch and flags. */
896 /* Various operation flags. */
897 enum mds_op_bias op_bias;
902 /* Used by readdir */
905 /* Used by readdir */
908 /* used to transfer info between the stacks of MD client
909 * see enum op_cli_flags */
912 /* File object data version for HSM release, on client */
913 __u64 op_data_version;
914 struct lustre_handle op_lease_handle;
918 CLI_SET_MEA = 1 << 0,
919 CLI_RM_ENTRY = 1 << 1,
922 struct md_enqueue_info;
923 /* metadata stat-ahead */
925 struct md_enqueue_info {
926 struct md_op_data mi_data;
927 struct lookup_intent mi_it;
928 struct lustre_handle mi_lockh;
929 struct inode *mi_dir;
930 int (*mi_cb)(struct ptlrpc_request *req,
931 struct md_enqueue_info *minfo, int rc);
933 unsigned int mi_generation;
937 struct module *owner;
938 int (*iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
939 void *karg, void __user *uarg);
940 int (*get_info)(const struct lu_env *env, struct obd_export *,
941 __u32 keylen, void *key, __u32 *vallen, void *val,
942 struct lov_stripe_md *lsm);
943 int (*set_info_async)(const struct lu_env *, struct obd_export *,
944 __u32 keylen, void *key,
945 __u32 vallen, void *val,
946 struct ptlrpc_request_set *set);
947 int (*attach)(struct obd_device *dev, u32 len, void *data);
948 int (*detach)(struct obd_device *dev);
949 int (*setup)(struct obd_device *dev, struct lustre_cfg *cfg);
950 int (*precleanup)(struct obd_device *dev,
951 enum obd_cleanup_stage cleanup_stage);
952 int (*cleanup)(struct obd_device *dev);
953 int (*process_config)(struct obd_device *dev, u32 len, void *data);
954 int (*postrecov)(struct obd_device *dev);
955 int (*add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
957 int (*del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
958 /* connect to the target device with given connection
959 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
960 * granted by the target, which are guaranteed to be a subset of flags
961 * asked for. If @ocd == NULL, use default parameters. */
962 int (*connect)(const struct lu_env *env,
963 struct obd_export **exp, struct obd_device *src,
964 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
966 int (*reconnect)(const struct lu_env *env,
967 struct obd_export *exp, struct obd_device *src,
968 struct obd_uuid *cluuid,
969 struct obd_connect_data *ocd,
971 int (*disconnect)(struct obd_export *exp);
973 /* Initialize/finalize fids infrastructure. */
974 int (*fid_init)(struct obd_device *obd,
975 struct obd_export *exp, enum lu_cli_type type);
976 int (*fid_fini)(struct obd_device *obd);
978 /* Allocate new fid according to passed @hint. */
979 int (*fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
980 struct md_op_data *op_data);
983 * Object with @fid is getting deleted, we may want to do something
986 int (*statfs)(const struct lu_env *, struct obd_export *exp,
987 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
988 int (*statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
989 __u64 max_age, struct ptlrpc_request_set *set);
990 int (*packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
991 struct lov_stripe_md *mem_src);
992 int (*unpackmd)(struct obd_export *exp,
993 struct lov_stripe_md **mem_tgt,
994 struct lov_mds_md *disk_src, int disk_len);
995 int (*preallocate)(struct lustre_handle *, u32 *req, u64 *ids);
996 int (*create)(const struct lu_env *env, struct obd_export *exp,
997 struct obdo *oa, struct lov_stripe_md **ea,
998 struct obd_trans_info *oti);
999 int (*destroy)(const struct lu_env *env, struct obd_export *exp,
1000 struct obdo *oa, struct lov_stripe_md *ea,
1001 struct obd_trans_info *oti, struct obd_export *md_exp);
1002 int (*setattr)(const struct lu_env *, struct obd_export *exp,
1003 struct obd_info *oinfo, struct obd_trans_info *oti);
1004 int (*setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1005 struct obd_trans_info *oti,
1006 struct ptlrpc_request_set *rqset);
1007 int (*getattr)(const struct lu_env *env, struct obd_export *exp,
1008 struct obd_info *oinfo);
1009 int (*getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1010 struct ptlrpc_request_set *set);
1011 int (*adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1012 u64 size, int shrink);
1013 int (*preprw)(const struct lu_env *env, int cmd,
1014 struct obd_export *exp, struct obdo *oa, int objcount,
1015 struct obd_ioobj *obj, struct niobuf_remote *remote,
1016 int *nr_pages, struct niobuf_local *local,
1017 struct obd_trans_info *oti);
1018 int (*commitrw)(const struct lu_env *env, int cmd,
1019 struct obd_export *exp, struct obdo *oa,
1020 int objcount, struct obd_ioobj *obj,
1021 struct niobuf_remote *remote, int pages,
1022 struct niobuf_local *local,
1023 struct obd_trans_info *oti, int rc);
1024 int (*find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1025 ldlm_iterator_t it, void *data);
1026 int (*init_export)(struct obd_export *exp);
1027 int (*destroy_export)(struct obd_export *exp);
1029 /* metadata-only methods */
1030 int (*import_event)(struct obd_device *, struct obd_import *,
1031 enum obd_import_event);
1033 int (*notify)(struct obd_device *obd, struct obd_device *watched,
1034 enum obd_notify_event ev, void *data);
1036 int (*health_check)(const struct lu_env *env, struct obd_device *);
1037 struct obd_uuid *(*get_uuid)(struct obd_export *exp);
1040 int (*quotacheck)(struct obd_device *, struct obd_export *,
1041 struct obd_quotactl *);
1042 int (*quotactl)(struct obd_device *, struct obd_export *,
1043 struct obd_quotactl *);
1046 int (*pool_new)(struct obd_device *obd, char *poolname);
1047 int (*pool_del)(struct obd_device *obd, char *poolname);
1048 int (*pool_add)(struct obd_device *obd, char *poolname,
1050 int (*pool_rem)(struct obd_device *obd, char *poolname,
1052 void (*getref)(struct obd_device *obd);
1053 void (*putref)(struct obd_device *obd);
1055 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1056 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1057 * Also, add a wrapper function in include/linux/obd_class.h. */
1061 LUSTRE_OPC_MKDIR = (1 << 0),
1062 LUSTRE_OPC_SYMLINK = (1 << 1),
1063 LUSTRE_OPC_MKNOD = (1 << 2),
1064 LUSTRE_OPC_CREATE = (1 << 3),
1065 LUSTRE_OPC_ANY = (1 << 4)
1068 /* lmv structures */
1069 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1070 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1071 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1073 #define MAX_HASH_SIZE_32 0x7fffffffUL
1074 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1075 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1078 struct mdt_body *body;
1079 struct lov_stripe_md *lsm;
1080 struct lmv_stripe_md *mea;
1081 #ifdef CONFIG_FS_POSIX_ACL
1082 struct posix_acl *posix_acl;
1084 struct mdt_remote_perm *remote_perm;
1087 struct md_open_data {
1088 struct obd_client_handle *mod_och;
1089 struct ptlrpc_request *mod_open_req;
1090 struct ptlrpc_request *mod_close_req;
1091 atomic_t mod_refcount;
1095 struct lookup_intent;
1098 int (*getstatus)(struct obd_export *, struct lu_fid *);
1099 int (*null_inode)(struct obd_export *, const struct lu_fid *);
1100 int (*find_cbdata)(struct obd_export *, const struct lu_fid *,
1101 ldlm_iterator_t, void *);
1102 int (*close)(struct obd_export *, struct md_op_data *,
1103 struct md_open_data *, struct ptlrpc_request **);
1104 int (*create)(struct obd_export *, struct md_op_data *,
1105 const void *, int, int, __u32, __u32, cfs_cap_t,
1106 __u64, struct ptlrpc_request **);
1107 int (*done_writing)(struct obd_export *, struct md_op_data *,
1108 struct md_open_data *);
1109 int (*enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1110 struct lookup_intent *, struct md_op_data *,
1111 struct lustre_handle *, void *, int,
1112 struct ptlrpc_request **, __u64);
1113 int (*getattr)(struct obd_export *, struct md_op_data *,
1114 struct ptlrpc_request **);
1115 int (*getattr_name)(struct obd_export *, struct md_op_data *,
1116 struct ptlrpc_request **);
1117 int (*intent_lock)(struct obd_export *, struct md_op_data *,
1118 void *, int, struct lookup_intent *, int,
1119 struct ptlrpc_request **,
1120 ldlm_blocking_callback, __u64);
1121 int (*link)(struct obd_export *, struct md_op_data *,
1122 struct ptlrpc_request **);
1123 int (*rename)(struct obd_export *, struct md_op_data *,
1124 const char *, int, const char *, int,
1125 struct ptlrpc_request **);
1126 int (*is_subdir)(struct obd_export *, const struct lu_fid *,
1127 const struct lu_fid *,
1128 struct ptlrpc_request **);
1129 int (*setattr)(struct obd_export *, struct md_op_data *, void *,
1130 int, void *, int, struct ptlrpc_request **,
1131 struct md_open_data **mod);
1132 int (*sync)(struct obd_export *, const struct lu_fid *,
1133 struct ptlrpc_request **);
1134 int (*readpage)(struct obd_export *, struct md_op_data *,
1135 struct page **, struct ptlrpc_request **);
1137 int (*unlink)(struct obd_export *, struct md_op_data *,
1138 struct ptlrpc_request **);
1140 int (*setxattr)(struct obd_export *, const struct lu_fid *,
1141 u64, const char *, const char *, int, int, int, __u32,
1142 struct ptlrpc_request **);
1144 int (*getxattr)(struct obd_export *, const struct lu_fid *,
1145 u64, const char *, const char *, int, int, int,
1146 struct ptlrpc_request **);
1148 int (*init_ea_size)(struct obd_export *, int, int, int, int);
1150 int (*get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1151 struct obd_export *, struct obd_export *,
1152 struct lustre_md *);
1154 int (*free_lustre_md)(struct obd_export *, struct lustre_md *);
1156 int (*set_open_replay_data)(struct obd_export *,
1157 struct obd_client_handle *,
1158 struct lookup_intent *);
1159 int (*clear_open_replay_data)(struct obd_export *,
1160 struct obd_client_handle *);
1161 int (*set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1163 ldlm_mode_t (*lock_match)(struct obd_export *, __u64,
1164 const struct lu_fid *, ldlm_type_t,
1165 ldlm_policy_data_t *, ldlm_mode_t,
1166 struct lustre_handle *);
1168 int (*cancel_unused)(struct obd_export *, const struct lu_fid *,
1169 ldlm_policy_data_t *, ldlm_mode_t,
1170 ldlm_cancel_flags_t flags, void *opaque);
1172 int (*get_remote_perm)(struct obd_export *, const struct lu_fid *,
1173 __u32, struct ptlrpc_request **);
1175 int (*intent_getattr_async)(struct obd_export *,
1176 struct md_enqueue_info *,
1177 struct ldlm_enqueue_info *);
1179 int (*revalidate_lock)(struct obd_export *, struct lookup_intent *,
1180 struct lu_fid *, __u64 *bits);
1183 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1184 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1185 * wrapper function in include/linux/obd_class.h.
1189 struct lsm_operations {
1190 void (*lsm_free)(struct lov_stripe_md *);
1191 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1192 struct obd_export *md_exp);
1193 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, u64 *,
1195 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, u64 *,
1197 int (*lsm_lmm_verify)(struct lov_mds_md *lmm, int lmm_bytes,
1198 __u16 *stripe_count);
1199 int (*lsm_unpackmd)(struct lov_obd *lov, struct lov_stripe_md *lsm,
1200 struct lov_mds_md *lmm);
1203 extern const struct lsm_operations lsm_v1_ops;
1204 extern const struct lsm_operations lsm_v3_ops;
1205 static inline const struct lsm_operations *lsm_op_find(int magic)
1213 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1218 /* Requests for obd_extent_calc() */
1219 #define OBD_CALC_STRIPE_START 1
1220 #define OBD_CALC_STRIPE_END 2
1222 static inline struct md_open_data *obd_mod_alloc(void)
1224 struct md_open_data *mod;
1226 mod = kzalloc(sizeof(*mod), GFP_NOFS);
1229 atomic_set(&mod->mod_refcount, 1);
1233 #define obd_mod_get(mod) atomic_inc(&(mod)->mod_refcount)
1234 #define obd_mod_put(mod) \
1236 if (atomic_dec_and_test(&(mod)->mod_refcount)) { \
1237 if ((mod)->mod_open_req) \
1238 ptlrpc_req_finished((mod)->mod_open_req); \
1243 void obdo_from_inode(struct obdo *dst, struct inode *src, u32 valid);
1244 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1246 /* return 1 if client should be resend request */
1247 static inline int client_should_resend(int resend, struct client_obd *cli)
1249 return atomic_read(&cli->cl_resends) ?
1250 atomic_read(&cli->cl_resends) > resend : 1;
1254 * Return device name for this device
1256 * XXX: lu_device is declared before obd_device, while a pointer pointing
1257 * back to obd_device in lu_device, so this helper function defines here
1258 * instead of in lu_object.h
1260 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1262 return lu_dev->ld_obd->obd_name;
1265 static inline bool filename_is_volatile(const char *name, int namelen, int *idx)
1270 if (strncmp(name, LUSTRE_VOLATILE_HDR, LUSTRE_VOLATILE_HDR_LEN) != 0)
1273 /* caller does not care of idx */
1277 /* volatile file, the MDT can be set from name */
1278 /* name format is LUSTRE_VOLATILE_HDR:[idx]: */
1279 /* if no MDT is specified, use std way */
1280 if (namelen < LUSTRE_VOLATILE_HDR_LEN + 2)
1282 /* test for no MDT idx case */
1283 if ((*(name + LUSTRE_VOLATILE_HDR_LEN) == ':') &&
1284 (*(name + LUSTRE_VOLATILE_HDR_LEN + 1) == ':')) {
1288 /* we have an idx, read it */
1289 start = name + LUSTRE_VOLATILE_HDR_LEN + 1;
1290 *idx = simple_strtoul(start, &end, 0);
1292 * no digit, no trailing :, negative value
1294 if (((*idx == 0) && (end == start)) ||
1295 (*end != ':') || (*idx < 0))
1300 /* bad format of mdt idx, we cannot return an error
1301 * to caller so we use hash algo */
1302 CERROR("Bad volatile file name format: %s\n",
1303 name + LUSTRE_VOLATILE_HDR_LEN);
1307 static inline int cli_brw_size(struct obd_device *obd)
1309 return obd->u.cli.cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
1312 #endif /* __OBD_H */