1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 /* Moved to lustre_user.h
58 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_ioctl_data *)
59 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
60 #define IOC_MDC_MAX_NR 50
62 #include <lustre/lustre_idl.h>
63 #include <lu_target.h>
65 #include <lustre_lib.h>
66 #include <lustre_export.h>
67 #include <lustre_quota.h>
68 #include <lustre_fld.h>
69 #include <lustre_capa.h>
71 #include <libcfs/bitmap.h>
74 #define MAX_OBD_DEVICES 8192
76 /* this is really local to the OSC */
77 struct loi_oap_pages {
78 cfs_list_t lop_pending;
79 cfs_list_t lop_urgent;
80 cfs_list_t lop_pending_group;
90 struct lov_oinfo { /* per-stripe data structure */
91 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
92 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
93 int loi_ost_gen; /* generation of this loi_ost_idx */
95 /* used by the osc to keep track of what objects to build into rpcs */
96 struct loi_oap_pages loi_read_lop;
97 struct loi_oap_pages loi_write_lop;
98 cfs_list_t loi_ready_item;
99 cfs_list_t loi_hp_ready_item;
100 cfs_list_t loi_write_item;
101 cfs_list_t loi_read_item;
103 unsigned long loi_kms_valid:1;
104 __u64 loi_kms; /* known minimum size */
105 struct ost_lvb loi_lvb;
106 struct osc_async_rc loi_ar;
108 #define loi_id loi_oi.oi_id
109 #define loi_seq loi_oi.oi_seq
111 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
113 oinfo->loi_kms = kms;
114 oinfo->loi_kms_valid = 1;
117 static inline void loi_init(struct lov_oinfo *loi)
119 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
120 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
121 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
122 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
123 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
124 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
125 CFS_INIT_LIST_HEAD(&loi->loi_ready_item);
126 CFS_INIT_LIST_HEAD(&loi->loi_hp_ready_item);
127 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
128 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
131 struct lov_stripe_md {
132 cfs_spinlock_t lsm_lock;
133 pid_t lsm_lock_owner; /* debugging */
136 /* Public members. */
137 __u64 lw_object_id; /* lov object id */
138 __u64 lw_object_seq; /* lov object seq */
139 __u64 lw_maxbytes; /* maximum possible file size */
141 /* LOV-private members start here -- only for use in lov/. */
143 __u32 lw_stripe_size; /* size of the stripe */
144 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
145 unsigned lw_stripe_count; /* number of objects being striped over */
146 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
149 struct lov_oinfo *lsm_oinfo[0];
152 #define lsm_object_id lsm_wire.lw_object_id
153 #define lsm_object_seq lsm_wire.lw_object_seq
154 #define lsm_maxbytes lsm_wire.lw_maxbytes
155 #define lsm_magic lsm_wire.lw_magic
156 #define lsm_stripe_size lsm_wire.lw_stripe_size
157 #define lsm_pattern lsm_wire.lw_pattern
158 #define lsm_stripe_count lsm_wire.lw_stripe_count
159 #define lsm_pool_name lsm_wire.lw_pool_name
163 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
165 /* obd info for a particular level (lov, osc). */
167 /* Lock policy. It keeps an extent which is specific for a particular
168 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
169 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
170 ldlm_policy_data_t oi_policy;
171 /* Flags used for set request specific flags:
172 - while lock handling, the flags obtained on the enqueue
173 request are set here.
174 - while stats, the flags used for control delay/resend.
177 /* Lock handle specific for every OSC lock. */
178 struct lustre_handle *oi_lockh;
179 /* lsm data specific for every OSC. */
180 struct lov_stripe_md *oi_md;
181 /* obdo data specific for every OSC, if needed at all. */
183 /* statfs data specific for every OSC, if needed at all. */
184 struct obd_statfs *oi_osfs;
185 /* An update callback which is called to update some data on upper
186 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
187 * request in osc level for enqueue requests. It is also possible to
188 * update some caller data from LOV layer if needed. */
189 obd_enqueue_update_f oi_cb_up;
190 /* oss capability, its type is obd_capa in client to avoid copy.
191 * in contrary its type is lustre_capa in OSS. */
195 /* compare all relevant fields. */
196 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
197 struct lov_stripe_md *m2)
200 * ->lsm_wire contains padding, but it should be zeroed out during
203 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
206 void lov_stripe_lock(struct lov_stripe_md *md);
207 void lov_stripe_unlock(struct lov_stripe_md *md);
210 cfs_list_t typ_chain;
211 struct obd_ops *typ_dt_ops;
212 struct md_ops *typ_md_ops;
213 cfs_proc_dir_entry_t *typ_procroot;
216 struct lu_device_type *typ_lu;
217 cfs_spinlock_t obd_type_lock;
227 /* Individual type definitions */
229 struct ost_server_data;
231 #define OBT_MAGIC 0xBDDECEAE
232 /* hold common fields for "target" device */
233 struct obd_device_target {
235 struct super_block *obt_sb;
236 /** last_rcvd file */
237 struct file *obt_rcvd_filp;
238 struct lu_target *obt_lut;
239 __u64 obt_mount_count;
240 cfs_semaphore_t obt_quotachecking;
241 struct lustre_quota_ctxt obt_qctxt;
242 lustre_quota_version_t obt_qfmt;
243 cfs_rw_semaphore_t obt_rwsem;
244 struct vfsmount *obt_vfsmnt;
245 struct file *obt_health_check_filp;
250 LLOG_CONFIG_ORIG_CTXT = 0,
251 LLOG_CONFIG_REPL_CTXT,
252 LLOG_MDS_OST_ORIG_CTXT,
253 LLOG_MDS_OST_REPL_CTXT,
260 LLOG_LOVEA_ORIG_CTXT,
261 LLOG_LOVEA_REPL_CTXT,
262 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
263 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
264 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
268 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
270 struct filter_subdirs {
271 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
281 /* NB this field MUST be first */
282 struct obd_device_target fo_obt;
283 const char *fo_fstype;
286 cfs_dentry_t *fo_dentry_O;
287 cfs_dentry_t **fo_dentry_O_groups;
288 struct filter_subdirs *fo_dentry_O_sub;
289 cfs_semaphore_t fo_init_lock; /* group initialization lock */
290 int fo_committed_group;
292 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
294 unsigned long fo_destroys_in_progress;
295 cfs_semaphore_t fo_create_locks[FILTER_SUBDIR_COUNT];
297 cfs_list_t fo_export_list;
300 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
301 obd_size fo_tot_granted; /* all values in bytes */
302 obd_size fo_tot_pending;
303 int fo_tot_granted_clients;
305 obd_size fo_readcache_max_filesize;
306 int fo_read_cache:1, /**< enable read-only cache */
307 fo_writethrough_cache:1,/**< read cache writes */
308 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
309 fo_raid_degraded:1;/**< RAID device degraded */
311 struct obd_import *fo_mdc_imp;
312 struct obd_uuid fo_mdc_uuid;
313 struct lustre_handle fo_mdc_conn;
314 struct file **fo_last_objid_files;
315 __u64 *fo_last_objids; /* last created objid for groups,
316 * protected by fo_objidlock */
318 cfs_semaphore_t fo_alloc_lock;
320 cfs_atomic_t fo_r_in_flight;
321 cfs_atomic_t fo_w_in_flight;
324 * per-filter pool of kiobuf's allocated by filter_common_setup() and
325 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
326 * which ->fo_iobuf_count were allocated.
328 * This pool contains kiobuf used by
329 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
332 * Locking: none, each OST thread uses only one element, determined by
333 * its "ordinal number", ->t_id.
335 struct filter_iobuf **fo_iobuf_pool;
338 cfs_list_t fo_llog_list;
339 cfs_spinlock_t fo_llog_list_lock;
341 struct brw_stats fo_filter_stats;
342 struct lustre_quota_ctxt fo_quota_ctxt;
343 cfs_spinlock_t fo_quotacheck_lock;
344 cfs_atomic_t fo_quotachecking;
346 int fo_fmd_max_num; /* per exp filter_mod_data */
347 int fo_fmd_max_age; /* jiffies to fmd expiry */
348 unsigned long fo_syncjournal:1, /* sync journal on writes */
349 fo_sync_lock_cancel:2;/* sync on lock cancel */
353 cfs_rwlock_t fo_sptlrpc_lock;
354 struct sptlrpc_rule_set fo_sptlrpc_rset;
356 /* capability related */
357 unsigned int fo_fl_oss_capa;
358 cfs_list_t fo_capa_keys;
359 cfs_hlist_head_t *fo_capa_hash;
360 struct llog_commit_master *fo_lcm;
364 struct timeout_item {
365 enum timeout_event ti_event;
366 cfs_time_t ti_timeout;
369 cfs_list_t ti_obd_list;
373 #define OSC_MAX_RIF_DEFAULT 8
374 #define OSC_MAX_RIF_MAX 256
375 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
376 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
377 #define OSC_DEFAULT_RESENDS 10
379 /* possible values for fo_sync_lock_cancel */
381 NEVER_SYNC_ON_CANCEL = 0,
382 BLOCKING_SYNC_ON_CANCEL = 1,
383 ALWAYS_SYNC_ON_CANCEL = 2,
384 NUM_SYNC_ON_CANCEL_STATES
387 #define MDC_MAX_RIF_DEFAULT 8
388 #define MDC_MAX_RIF_MAX 512
393 cfs_rw_semaphore_t cl_sem;
394 struct obd_uuid cl_target_uuid;
395 struct obd_import *cl_import; /* ptlrpc connection state */
397 /* max_mds_easize is purely a performance thing so we don't have to
398 * call obd_size_diskmd() all the time. */
399 int cl_default_mds_easize;
400 int cl_max_mds_easize;
401 int cl_max_mds_cookiesize;
403 enum lustre_sec_part cl_sp_me;
404 enum lustre_sec_part cl_sp_to;
405 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
407 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
408 void *cl_llcd_offset;
410 /* the grant values are protected by loi_list_lock below */
411 long cl_dirty; /* all _dirty_ in bytes */
412 long cl_dirty_max; /* allowed w/o rpc */
413 long cl_dirty_transit; /* dirty synchronous */
414 long cl_avail_grant; /* bytes of credit for ost */
415 long cl_lost_grant; /* lost credits (trunc) */
416 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
417 cfs_time_t cl_next_shrink_grant; /* jiffies */
418 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
419 cfs_semaphore_t cl_grant_sem; /*grant shrink list cfs_semaphore*/
420 int cl_grant_shrink_interval; /* seconds */
422 /* keep track of objects that have lois that contain pages which
423 * have been queued for async brw. this lock also protects the
424 * lists of osc_client_pages that hang off of the loi */
426 * ->cl_loi_list_lock protects consistency of
427 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
428 * ->ap_completion() call-backs are executed under this lock. As we
429 * cannot guarantee that these call-backs never block on all platforms
430 * (as a matter of fact they do block on Mac OS X), type of
431 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
432 * and blocking mutex on Mac OS X. (Alternative is to make this lock
433 * blocking everywhere, but we don't want to slow down fast-path of
434 * our main platform.)
436 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
437 * with client_obd_list_{un,}lock() and
438 * client_obd_list_lock_{init,done}() functions.
440 client_obd_lock_t cl_loi_list_lock;
441 cfs_list_t cl_loi_ready_list;
442 cfs_list_t cl_loi_hp_ready_list;
443 cfs_list_t cl_loi_write_list;
444 cfs_list_t cl_loi_read_list;
447 /* just a sum of the loi/lop pending numbers to be exported by /proc */
448 int cl_pending_w_pages;
449 int cl_pending_r_pages;
450 int cl_max_pages_per_rpc;
451 int cl_max_rpcs_in_flight;
452 struct obd_histogram cl_read_rpc_hist;
453 struct obd_histogram cl_write_rpc_hist;
454 struct obd_histogram cl_read_page_hist;
455 struct obd_histogram cl_write_page_hist;
456 struct obd_histogram cl_read_offset_hist;
457 struct obd_histogram cl_write_offset_hist;
459 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
460 cfs_atomic_t cl_destroy_in_flight;
461 cfs_waitq_t cl_destroy_waitq;
463 struct mdc_rpc_lock *cl_rpc_lock;
464 struct mdc_rpc_lock *cl_setattr_lock;
465 struct mdc_rpc_lock *cl_close_lock;
466 struct osc_creator cl_oscc;
469 cfs_semaphore_t cl_mgc_sem;
470 struct vfsmount *cl_mgc_vfsmnt;
471 struct dentry *cl_mgc_configs_dir;
472 cfs_atomic_t cl_mgc_refcount;
473 struct obd_export *cl_mgc_mgsexp;
475 /* checksumming for data sent over the network */
476 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
477 /* supported checksum types that are worked out at connect time */
478 __u32 cl_supp_cksum_types;
479 /* checksum algorithm to be used */
480 cksum_type_t cl_cksum_type;
482 /* also protected by the poorly named _loi_list_lock lock above */
483 struct osc_async_rc cl_ar;
485 /* used by quotacheck */
486 int cl_qchk_stat; /* quotacheck stat of the peer */
488 /* sequence manager */
489 struct lu_client_seq *cl_seq;
491 cfs_atomic_t cl_resends; /* resend count */
493 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
495 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
498 struct ptlrpc_service *mgs_service;
499 struct vfsmount *mgs_vfsmnt;
500 struct super_block *mgs_sb;
501 struct dentry *mgs_configs_dir;
502 struct dentry *mgs_fid_de;
503 cfs_list_t mgs_fs_db_list;
504 cfs_semaphore_t mgs_sem;
505 cfs_proc_dir_entry_t *mgs_proc_live;
509 /* NB this field MUST be first */
510 struct obd_device_target mds_obt;
511 struct ptlrpc_service *mds_service;
512 struct ptlrpc_service *mds_setattr_service;
513 struct ptlrpc_service *mds_readpage_service;
514 cfs_dentry_t *mds_fid_de;
516 int mds_max_cookiesize;
518 unsigned long mds_atime_diff;
519 cfs_semaphore_t mds_epoch_sem;
520 struct ll_fid mds_rootfid;
521 cfs_dentry_t *mds_pending_dir;
522 cfs_dentry_t *mds_logs_dir;
523 cfs_dentry_t *mds_objects_dir;
524 struct llog_handle *mds_cfg_llh;
525 struct obd_device *mds_lov_obd;
526 struct obd_uuid mds_lov_uuid;
528 struct obd_export *mds_lov_exp;
529 struct lov_desc mds_lov_desc;
532 /* mark pages dirty for write. */
533 cfs_bitmap_t *mds_lov_page_dirty;
534 /* array for store pages with obd_id */
535 void **mds_lov_page_array;
536 /* file for store objid */
537 struct file *mds_lov_objid_filp;
538 __u32 mds_lov_objid_count;
539 __u32 mds_lov_objid_max_index;
540 __u32 mds_lov_objid_lastpage;
541 __u32 mds_lov_objid_lastidx;
544 struct lustre_quota_info mds_quota_info;
545 cfs_rw_semaphore_t mds_qonoff_sem;
546 cfs_semaphore_t mds_health_sem;
547 unsigned long mds_fl_user_xattr:1,
549 mds_evict_ost_nids:1,
553 mds_fl_target:1; /* mds have one or
556 struct upcall_cache *mds_identity_cache;
558 /* for capability keys update */
559 struct lustre_capa_key *mds_capa_keys;
560 cfs_rw_semaphore_t mds_notify_lock;
564 extern __u32 mds_max_ost_index;
566 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
568 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
570 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
572 extern int mds_lov_init_objids(struct obd_device *obd);
573 extern void mds_lov_destroy_objids(struct obd_device *obd);
584 cfs_spinlock_t eo_lock;
586 struct lustre_handle eo_nl_lock;
587 cfs_atomic_t eo_prep;
591 struct ptlrpc_service *ost_service;
592 struct ptlrpc_service *ost_create_service;
593 struct ptlrpc_service *ost_io_service;
594 cfs_semaphore_t ost_health_sem;
597 struct echo_client_obd {
598 struct obd_export *ec_exp; /* the local connection to osc/lov */
599 cfs_spinlock_t ec_lock;
600 cfs_list_t ec_objects;
607 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
608 cfs_list_t lqo_oss_list; /* link to lov_qos */
609 __u64 lqo_bavail; /* total bytes avail on OSS */
610 __u64 lqo_penalty; /* current penalty */
611 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
612 time_t lqo_used; /* last used time, seconds */
613 __u32 lqo_ost_count; /* number of osts on this oss */
617 struct lov_qos_oss *ltq_oss; /* oss info */
618 __u64 ltq_penalty; /* current penalty */
619 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
620 __u64 ltq_weight; /* net weighting */
621 time_t ltq_used; /* last used time, seconds */
622 unsigned int ltq_usable:1; /* usable for striping */
625 /* Generic subset of OSTs */
627 __u32 *op_array; /* array of index of
629 unsigned int op_count; /* number of OSTs in the array */
630 unsigned int op_size; /* allocated size of lp_array */
631 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
634 /* Round-robin allocator data */
636 __u32 lqr_start_idx; /* start index of new inode */
637 __u32 lqr_offset_idx; /* aliasing for start_idx */
638 int lqr_start_count; /* reseed counter */
639 struct ost_pool lqr_pool; /* round-robin optimized list */
640 unsigned long lqr_dirty:1; /* recalc round-robin list */
643 /* allow statfs data caching for 1 second */
644 #define OBD_STATFS_CACHE_SECONDS 1
646 struct lov_statfs_data {
647 struct obd_info lsd_oi;
648 struct obd_statfs lsd_statfs;
650 /* Stripe placement optimization */
652 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
653 cfs_rw_semaphore_t lq_rw_sem;
654 __u32 lq_active_oss_count;
655 unsigned int lq_prio_free; /* priority for free space */
656 unsigned int lq_threshold_rr;/* priority for rr */
657 struct lov_qos_rr lq_rr; /* round robin qos data */
658 unsigned long lq_dirty:1, /* recalc qos data */
659 lq_same_space:1,/* the ost's all have approx.
660 the same space avail */
661 lq_reset:1, /* zero current penalties */
662 lq_statfs_in_progress:1; /* statfs op in
664 /* qos statfs data */
665 struct lov_statfs_data *lq_statfs_data;
666 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
667 * requests completion */
670 struct lov_tgt_desc {
672 struct obd_uuid ltd_uuid;
673 struct obd_device *ltd_obd;
674 struct obd_export *ltd_exp;
675 struct ltd_qos ltd_qos; /* qos info per target */
677 __u32 ltd_index; /* index in lov_obd->tgts */
678 unsigned long ltd_active:1,/* is this target up for requests */
679 ltd_activate:1,/* should target be activated */
680 ltd_reap:1; /* should this target be deleted */
684 #define pool_tgt_size(_p) _p->pool_obds.op_size
685 #define pool_tgt_count(_p) _p->pool_obds.op_count
686 #define pool_tgt_array(_p) _p->pool_obds.op_array
687 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
688 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
691 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
692 struct ost_pool pool_obds; /* pool members */
693 cfs_atomic_t pool_refcount; /* pool ref. counter */
694 struct lov_qos_rr pool_rr; /* round robin qos */
695 cfs_hlist_node_t pool_hash; /* access by poolname */
696 cfs_list_t pool_list; /* serial access */
697 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
698 struct lov_obd *pool_lov; /* lov obd to which this
703 struct lov_desc desc;
704 struct lov_tgt_desc **lov_tgts; /* sparse array */
705 struct ost_pool lov_packed; /* all OSTs in a packed
707 cfs_semaphore_t lov_lock;
708 struct obd_connect_data lov_ocd;
709 struct lov_qos lov_qos; /* qos info per lov */
710 cfs_atomic_t lov_refcount;
711 __u32 lov_tgt_count; /* how many OBD's */
712 __u32 lov_active_tgt_count; /* how many active */
713 __u32 lov_death_row;/* tgts scheduled to be deleted */
714 __u32 lov_tgt_size; /* size of tgts array */
717 cfs_hash_t *lov_pools_hash_body; /* used for key access */
718 cfs_list_t lov_pool_list; /* used for sequential access */
719 cfs_proc_dir_entry_t *lov_pool_proc_entry;
720 enum lustre_sec_part lov_sp_me;
723 struct lmv_tgt_desc {
724 struct obd_uuid ltd_uuid;
725 struct obd_export *ltd_exp;
726 int ltd_active; /* is this target up for requests */
728 cfs_semaphore_t ltd_fid_sem;
731 enum placement_policy {
732 PLACEMENT_CHAR_POLICY = 0,
733 PLACEMENT_NID_POLICY = 1,
734 PLACEMENT_INVAL_POLICY = 2,
738 typedef enum placement_policy placement_policy_t;
742 struct lu_client_fld lmv_fld;
743 cfs_spinlock_t lmv_lock;
744 placement_policy_t lmv_placement;
745 struct lmv_desc desc;
746 struct obd_uuid cluuid;
747 struct obd_export *exp;
754 cfs_semaphore_t init_sem;
756 struct lmv_tgt_desc *tgts;
759 struct obd_connect_data *datas;
762 struct obd_connect_data conn_data;
765 struct niobuf_local {
770 cfs_dentry_t *dentry;
775 #define LUSTRE_FLD_NAME "fld"
776 #define LUSTRE_SEQ_NAME "seq"
778 #define LUSTRE_CMM_NAME "cmm"
779 #define LUSTRE_MDD_NAME "mdd"
780 #define LUSTRE_OSD_NAME "osd-ldiskfs"
781 #define LUSTRE_VVP_NAME "vvp"
782 #define LUSTRE_LMV_NAME "lmv"
783 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
784 #define LUSTRE_SLP_NAME "slp"
786 /* obd device type names */
787 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
788 #define LUSTRE_MDS_NAME "mds"
789 #define LUSTRE_MDT_NAME "mdt"
790 #define LUSTRE_MDC_NAME "mdc"
791 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
792 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
793 #define LUSTRE_OSC_NAME "osc"
794 #define LUSTRE_LOV_NAME "lov"
795 #define LUSTRE_MGS_NAME "mgs"
796 #define LUSTRE_MGC_NAME "mgc"
798 #define LUSTRE_CACHEOBD_NAME "cobd"
799 #define LUSTRE_ECHO_NAME "obdecho"
800 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
802 /* Constant obd names (post-rename) */
803 #define LUSTRE_MDS_OBDNAME "MDS"
804 #define LUSTRE_OSS_OBDNAME "OSS"
805 #define LUSTRE_MGS_OBDNAME "MGS"
806 #define LUSTRE_MGC_OBDNAME "MGC"
808 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
809 #define N_LOCAL_TEMP_PAGE 0x10000000
811 struct obd_trans_info {
814 /* Only used on the server side for tracking acks. */
815 struct oti_req_ack_lock {
816 struct lustre_handle lock;
820 struct llog_cookie oti_onecookie;
821 struct llog_cookie *oti_logcookies;
824 /* initial thread handling transaction */
825 struct ptlrpc_thread * oti_thread;
828 __u64 oti_pre_version;
830 struct obd_uuid *oti_ost_uuid;
833 static inline void oti_init(struct obd_trans_info *oti,
834 struct ptlrpc_request *req)
838 memset(oti, 0, sizeof(*oti));
843 oti->oti_xid = req->rq_xid;
844 /** VBR: take versions from request */
845 if (req->rq_reqmsg != NULL &&
846 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
847 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
848 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
849 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
852 /** called from mds_create_objects */
853 if (req->rq_repmsg != NULL)
854 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
855 oti->oti_thread = req->rq_svc_thread;
856 if (req->rq_reqmsg != NULL)
857 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
860 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
865 if (num_cookies == 1)
866 oti->oti_logcookies = &oti->oti_onecookie;
868 OBD_ALLOC(oti->oti_logcookies,
869 num_cookies * sizeof(oti->oti_onecookie));
871 oti->oti_numcookies = num_cookies;
874 static inline void oti_free_cookies(struct obd_trans_info *oti)
876 if (!oti || !oti->oti_logcookies)
879 if (oti->oti_logcookies == &oti->oti_onecookie)
880 LASSERT(oti->oti_numcookies == 1);
882 OBD_FREE(oti->oti_logcookies,
883 oti->oti_numcookies * sizeof(oti->oti_onecookie));
884 oti->oti_logcookies = NULL;
885 oti->oti_numcookies = 0;
889 * Events signalled through obd_notify() upcall-chain.
891 enum obd_notify_event {
892 /* Device connect start */
894 /* Device activated */
896 /* Device deactivated */
898 /* Device disconnected */
900 /* Connect data for import were changed */
903 OBD_NOTIFY_SYNC_NONBLOCK,
905 /* Configuration event */
907 /* Trigger quota recovery */
911 /* bit-mask flags for config events */
913 CONFIG_LOG = 0x1, /* finished processing config log */
914 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
915 CONFIG_TARGET = 0x4 /* one target is added */
919 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
920 * and liblustre being main examples).
922 struct obd_notify_upcall {
923 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
924 enum obd_notify_event ev, void *owner, void *data);
925 /* Opaque datum supplied by upper layer listener */
929 struct target_recovery_data {
930 svc_handler_t trd_recovery_handler;
931 pid_t trd_processing_task;
932 cfs_completion_t trd_starting;
933 cfs_completion_t trd_finishing;
937 * In HEAD for CMD, the object is created in group number which is 3>=
938 * or indexing starts from 3. To test this assertions are added to disallow
939 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
940 * object in group 0 needs to be allowed.
941 * So for interop mode following changes needs to be done:
942 * 1. No need to assert on group 0 or allow group 0
943 * 2. The group number indexing starts from 0 instead of 3
946 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
948 static inline __u64 objseq_to_mdsno(obd_seq seq)
950 LASSERT_SEQ_IS_MDT(seq);
951 if (seq == FID_SEQ_OST_MDT0)
953 return seq - FID_SEQ_OST_MDT1 + 1;
956 static inline int mdt_to_obd_objseq(int mdtid)
959 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
963 return FID_SEQ_OST_MDT1 + mdtid - 1;
967 struct obd_llog_group {
970 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
971 cfs_waitq_t olg_waitq;
972 cfs_spinlock_t olg_lock;
973 struct obd_export *olg_exp;
974 int olg_initializing;
975 cfs_semaphore_t olg_cat_processing;
978 /* corresponds to one of the obd's */
979 #define MAX_OBD_NAME 128
980 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
981 #define OBD_DEV_BY_DEVNAME 0xffffd0de
983 struct obd_type *obd_type;
986 /* common and UUID name of this device */
987 char obd_name[MAX_OBD_NAME];
988 struct obd_uuid obd_uuid;
990 struct lu_device *obd_lu_dev;
993 unsigned long obd_attached:1, /* finished attach */
994 obd_set_up:1, /* finished setup */
995 obd_recovering:1, /* there are recoverable clients */
996 obd_abort_recovery:1,/* recovery expired */
997 obd_version_recov:1, /* obd uses version checking */
998 obd_recovery_expired:1,
999 obd_replayable:1, /* recovery is enabled; inform clients */
1000 obd_no_transno:1, /* no committed-transno notification */
1001 obd_no_recov:1, /* fail instead of retry messages */
1002 obd_req_replaying:1, /* replaying requests */
1003 obd_stopping:1, /* started cleanup */
1004 obd_starting:1, /* started setup */
1005 obd_force:1, /* cleanup with > 0 obd refcount */
1006 obd_fail:1, /* cleanup with failover */
1007 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1008 obd_no_conn:1, /* deny new connections */
1009 obd_inactive:1, /* device active/inactive
1010 * (for /proc/status only!!) */
1011 obd_process_conf:1; /* device is processing mgs config */
1012 /* uuid-export hash body */
1013 cfs_hash_t *obd_uuid_hash;
1014 /* nid-export hash body */
1015 cfs_hash_t *obd_nid_hash;
1016 /* nid stats body */
1017 cfs_hash_t *obd_nid_stats_hash;
1018 cfs_list_t obd_nid_stats;
1019 cfs_atomic_t obd_refcount;
1020 cfs_waitq_t obd_refcount_waitq;
1021 cfs_list_t obd_exports;
1022 cfs_list_t obd_unlinked_exports;
1023 cfs_list_t obd_delayed_exports;
1024 int obd_num_exports;
1025 cfs_spinlock_t obd_nid_lock;
1026 struct ldlm_namespace *obd_namespace;
1027 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1028 /* a spinlock is OK for what we do now, may need a semaphore later */
1029 cfs_spinlock_t obd_dev_lock;
1030 cfs_semaphore_t obd_dev_sem;
1031 __u64 obd_last_committed;
1032 struct fsfilt_operations *obd_fsops;
1033 cfs_spinlock_t obd_osfs_lock;
1034 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1036 struct lvfs_run_ctxt obd_lvfs_ctxt;
1037 struct obd_llog_group obd_olg; /* default llog group */
1038 struct obd_device *obd_observer;
1039 cfs_rw_semaphore_t obd_observer_link_sem;
1040 struct obd_notify_upcall obd_upcall;
1041 struct obd_export *obd_self_export;
1042 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1043 cfs_list_t obd_exports_timed;
1044 time_t obd_eviction_timer; /* for ping evictor */
1046 int obd_max_recoverable_clients;
1047 int obd_connected_clients;
1048 int obd_stale_clients;
1049 int obd_delayed_clients;
1050 cfs_spinlock_t obd_processing_task_lock; /* BH lock (timer) */
1051 __u64 obd_next_recovery_transno;
1052 int obd_replayed_requests;
1053 int obd_requests_queued_for_recovery;
1054 cfs_waitq_t obd_next_transno_waitq;
1055 cfs_timer_t obd_recovery_timer;
1056 time_t obd_recovery_start; /* seconds */
1057 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1058 time_t obd_recovery_time_hard;
1059 int obd_recovery_timeout;
1061 /* new recovery stuff from CMD2 */
1062 struct target_recovery_data obd_recovery_data;
1063 int obd_replayed_locks;
1064 cfs_atomic_t obd_req_replay_clients;
1065 cfs_atomic_t obd_lock_replay_clients;
1066 cfs_list_t obd_req_replay_queue;
1067 cfs_list_t obd_lock_replay_queue;
1068 cfs_list_t obd_final_req_queue;
1069 int obd_recovery_stage;
1072 struct obd_device_target obt;
1073 struct filter_obd filter;
1075 struct client_obd cli;
1077 struct echo_client_obd echo_client;
1078 struct echo_obd echo;
1083 /* Fields used by LProcFS */
1084 unsigned int obd_cntr_base;
1085 struct lprocfs_stats *obd_stats;
1087 unsigned int md_cntr_base;
1088 struct lprocfs_stats *md_stats;
1090 cfs_proc_dir_entry_t *obd_proc_entry;
1091 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1092 cfs_proc_dir_entry_t *obd_svc_procroot;
1093 struct lprocfs_stats *obd_svc_stats;
1094 cfs_atomic_t obd_evict_inprogress;
1095 cfs_waitq_t obd_evict_inprogress_waitq;
1096 cfs_list_t obd_evict_list; /* protected with pet_lock */
1099 * Ldlm pool part. Save last calculated SLV and Limit.
1101 cfs_rwlock_t obd_pool_lock;
1106 * A list of outstanding class_incref()'s against this obd. For
1109 struct lu_ref obd_reference;
1112 #define OBD_LLOG_FL_SENDNOW 0x0001
1114 enum obd_cleanup_stage {
1115 /* Special case hack for MDS LOVs */
1117 /* can be directly mapped to .ldto_device_fini() */
1118 OBD_CLEANUP_EXPORTS,
1121 /* get/set_info keys */
1122 #define KEY_ASYNC "async"
1123 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1124 #define KEY_BLOCKSIZE "blocksize"
1125 #define KEY_CAPA_KEY "capa_key"
1126 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1127 #define KEY_FID2PATH "fid2path"
1128 #define KEY_CHECKSUM "checksum"
1129 #define KEY_CLEAR_FS "clear_fs"
1130 #define KEY_CONN_DATA "conn_data"
1131 #define KEY_EVICT_BY_NID "evict_by_nid"
1132 #define KEY_FIEMAP "fiemap"
1133 #define KEY_FLUSH_CTX "flush_ctx"
1134 #define KEY_GRANT_SHRINK "grant_shrink"
1135 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1136 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1137 #define KEY_INIT_RECOV "initial_recov"
1138 #define KEY_INTERMDS "inter_mds"
1139 #define KEY_LAST_ID "last_id"
1140 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1141 #define KEY_LOVDESC "lovdesc"
1142 #define KEY_LOV_IDX "lov_idx"
1143 #define KEY_MAX_EASIZE "max_easize"
1144 #define KEY_MDS_CONN "mds_conn"
1145 #define KEY_MGSSEC "mgssec"
1146 #define KEY_NEXT_ID "next_id"
1147 #define KEY_READ_ONLY "read-only"
1148 #define KEY_REGISTER_TARGET "register_target"
1149 #define KEY_REVIMP_UPD "revimp_update"
1150 #define KEY_SET_FS "set_fs"
1151 /* KEY_SET_INFO in lustre_idl.h */
1152 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1153 #define KEY_CONNECT_FLAG "connect_flags"
1154 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1159 static inline int it_to_lock_mode(struct lookup_intent *it)
1161 /* CREAT needs to be tested before open (both could be set) */
1162 if (it->it_op & IT_CREAT)
1164 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
1167 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1172 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1173 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1174 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1175 struct lu_fid op_fid4; /* to the operation locks. */
1176 mdsno_t op_mds; /* what mds server open will go to */
1177 struct lustre_handle op_handle;
1179 const char *op_name;
1182 struct lmv_stripe_md *op_mea1;
1183 struct lmv_stripe_md *op_mea2;
1184 __u32 op_suppgids[2];
1190 /* iattr fields and blocks. */
1191 struct iattr op_attr;
1193 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
1194 unsigned int op_attr_flags;
1198 loff_t op_attr_blocks;
1200 /* Size-on-MDS epoch and flags. */
1205 struct obd_capa *op_capa1;
1206 struct obd_capa *op_capa2;
1208 /* Various operation flags. */
1211 /* Operation type */
1215 struct md_enqueue_info;
1216 /* metadata stat-ahead */
1217 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1218 struct md_enqueue_info *minfo,
1221 struct md_enqueue_info {
1222 struct md_op_data mi_data;
1223 struct lookup_intent mi_it;
1224 struct lustre_handle mi_lockh;
1225 struct dentry *mi_dentry;
1226 struct inode *mi_dir;
1227 md_enqueue_cb_t mi_cb;
1228 unsigned int mi_generation;
1233 cfs_module_t *o_owner;
1234 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1235 void *karg, void *uarg);
1236 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
1237 __u32 *vallen, void *val, struct lov_stripe_md *lsm);
1238 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
1239 __u32 vallen, void *val,
1240 struct ptlrpc_request_set *set);
1241 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1242 int (*o_detach)(struct obd_device *dev);
1243 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1244 int (*o_precleanup)(struct obd_device *dev,
1245 enum obd_cleanup_stage cleanup_stage);
1246 int (*o_cleanup)(struct obd_device *dev);
1247 int (*o_process_config)(struct obd_device *dev, obd_count len,
1249 int (*o_postrecov)(struct obd_device *dev);
1250 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1252 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1253 /* connect to the target device with given connection
1254 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1255 * granted by the target, which are guaranteed to be a subset of flags
1256 * asked for. If @ocd == NULL, use default parameters. */
1257 int (*o_connect)(const struct lu_env *env,
1258 struct obd_export **exp, struct obd_device *src,
1259 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1261 int (*o_reconnect)(const struct lu_env *env,
1262 struct obd_export *exp, struct obd_device *src,
1263 struct obd_uuid *cluuid,
1264 struct obd_connect_data *ocd,
1266 int (*o_disconnect)(struct obd_export *exp);
1268 /* Initialize/finalize fids infrastructure. */
1269 int (*o_fid_init)(struct obd_export *exp);
1270 int (*o_fid_fini)(struct obd_export *exp);
1272 /* Allocate new fid according to passed @hint. */
1273 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1274 struct md_op_data *op_data);
1277 * Object with @fid is getting deleted, we may want to do something
1280 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1282 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
1283 __u64 max_age, __u32 flags);
1284 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
1285 __u64 max_age, struct ptlrpc_request_set *set);
1286 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1287 struct lov_stripe_md *mem_src);
1288 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1289 struct lov_mds_md *disk_src, int disk_len);
1290 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1292 /* FIXME: add fid capability support for create & destroy! */
1293 int (*o_precreate)(struct obd_export *exp);
1294 int (*o_create)(struct obd_export *exp, struct obdo *oa,
1295 struct lov_stripe_md **ea, struct obd_trans_info *oti);
1296 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1297 struct lov_stripe_md **ea,
1298 struct obd_trans_info *oti);
1299 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
1300 struct lov_stripe_md *ea, struct obd_trans_info *oti,
1301 struct obd_export *md_exp, void *capa);
1302 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
1303 struct obd_trans_info *oti);
1304 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1305 struct obd_trans_info *oti,
1306 struct ptlrpc_request_set *rqset);
1307 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
1308 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1309 struct ptlrpc_request_set *set);
1310 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1311 obd_count oa_bufs, struct brw_page *pgarr,
1312 struct obd_trans_info *oti);
1313 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1314 struct ost_lvb *lvb, int kms_only);
1315 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1316 obd_off size, int shrink);
1317 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
1318 struct obd_trans_info *oti,
1319 struct ptlrpc_request_set *rqset);
1320 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
1321 struct lov_stripe_md *ea, obd_size start, obd_size end,
1323 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1324 struct lov_stripe_md *src, obd_size start,
1325 obd_size end, struct obd_trans_info *oti);
1326 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1327 struct lustre_handle *srconn, struct lov_stripe_md *src,
1328 obd_size start, obd_size end, struct obd_trans_info *);
1329 int (*o_iterate)(struct lustre_handle *conn,
1330 int (*)(obd_id, obd_seq, void *),
1331 obd_id *startid, obd_seq seq, void *data);
1332 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
1333 int objcount, struct obd_ioobj *obj,
1334 struct niobuf_remote *remote, int *nr_pages,
1335 struct niobuf_local *local,
1336 struct obd_trans_info *oti,
1337 struct lustre_capa *capa);
1338 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
1339 int objcount, struct obd_ioobj *obj,
1340 struct niobuf_remote *remote, int pages,
1341 struct niobuf_local *local,
1342 struct obd_trans_info *oti, int rc);
1343 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1344 struct ldlm_enqueue_info *einfo,
1345 struct ptlrpc_request_set *rqset);
1346 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1347 ldlm_iterator_t it, void *data);
1348 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1349 ldlm_iterator_t it, void *data);
1350 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1351 __u32 mode, struct lustre_handle *);
1352 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1353 ldlm_cancel_flags_t flags, void *opaque);
1354 int (*o_init_export)(struct obd_export *exp);
1355 int (*o_destroy_export)(struct obd_export *exp);
1356 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1357 int cmd, obd_off *);
1359 /* llog related obd_methods */
1360 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1361 struct obd_device *disk_obd, int *idx);
1362 int (*o_llog_finish)(struct obd_device *obd, int count);
1363 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1365 /* metadata-only methods */
1366 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1367 struct obd_capa *, struct obd_client_handle *, int flag);
1368 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1370 int (*o_import_event)(struct obd_device *, struct obd_import *,
1371 enum obd_import_event);
1373 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1374 enum obd_notify_event ev, void *data);
1376 int (*o_health_check)(struct obd_device *);
1377 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1380 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1381 struct obd_quotactl *);
1382 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1383 struct obd_quotactl *);
1384 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1385 struct quota_adjust_qunit *oqaq,
1386 struct lustre_quota_ctxt *qctxt);
1389 int (*o_ping)(struct obd_export *exp);
1392 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1393 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1394 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1396 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1398 void (*o_getref)(struct obd_device *obd);
1399 void (*o_putref)(struct obd_device *obd);
1401 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1402 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1403 * Also, add a wrapper function in include/linux/obd_class.h. */
1407 LUSTRE_OPC_MKDIR = (1 << 0),
1408 LUSTRE_OPC_SYMLINK = (1 << 1),
1409 LUSTRE_OPC_MKNOD = (1 << 2),
1410 LUSTRE_OPC_CREATE = (1 << 3),
1411 LUSTRE_OPC_ANY = (1 << 4)
1414 /* lmv structures */
1415 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1416 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1417 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1419 #define MAX_HASH_SIZE_32 0x7fffffffUL
1420 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1421 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1424 struct mdt_body *body;
1425 struct lov_stripe_md *lsm;
1426 struct lmv_stripe_md *mea;
1427 #ifdef CONFIG_FS_POSIX_ACL
1428 struct posix_acl *posix_acl;
1430 struct mdt_remote_perm *remote_perm;
1431 struct obd_capa *mds_capa;
1432 struct obd_capa *oss_capa;
1435 struct md_open_data {
1436 struct obd_client_handle *mod_och;
1437 struct ptlrpc_request *mod_open_req;
1438 struct ptlrpc_request *mod_close_req;
1439 cfs_atomic_t mod_refcount;
1442 struct lookup_intent;
1445 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1446 struct obd_capa **);
1447 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1448 ldlm_iterator_t, void *);
1449 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1450 ldlm_iterator_t, void *);
1451 int (*m_close)(struct obd_export *, struct md_op_data *,
1452 struct md_open_data *, struct ptlrpc_request **);
1453 int (*m_create)(struct obd_export *, struct md_op_data *,
1454 const void *, int, int, __u32, __u32, cfs_cap_t,
1455 __u64, struct ptlrpc_request **);
1456 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1457 struct md_open_data *);
1458 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1459 struct lookup_intent *, struct md_op_data *,
1460 struct lustre_handle *, void *, int,
1461 struct ptlrpc_request **, int);
1462 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1463 struct ptlrpc_request **);
1464 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1465 struct ptlrpc_request **);
1466 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1467 void *, int, struct lookup_intent *, int,
1468 struct ptlrpc_request **,
1469 ldlm_blocking_callback, int);
1470 int (*m_link)(struct obd_export *, struct md_op_data *,
1471 struct ptlrpc_request **);
1472 int (*m_rename)(struct obd_export *, struct md_op_data *,
1473 const char *, int, const char *, int,
1474 struct ptlrpc_request **);
1475 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1476 const struct lu_fid *,
1477 struct ptlrpc_request **);
1478 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1479 int , void *, int, struct ptlrpc_request **,
1480 struct md_open_data **mod);
1481 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1482 struct obd_capa *, struct ptlrpc_request **);
1483 int (*m_readpage)(struct obd_export *, const struct lu_fid *,
1484 struct obd_capa *, __u64, struct page *,
1485 struct ptlrpc_request **);
1487 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1488 struct ptlrpc_request **);
1490 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1491 struct obd_capa *, obd_valid, const char *,
1492 const char *, int, int, int, __u32,
1493 struct ptlrpc_request **);
1495 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1496 struct obd_capa *, obd_valid, const char *,
1497 const char *, int, int, int,
1498 struct ptlrpc_request **);
1500 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1502 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1503 struct obd_export *, struct obd_export *,
1504 struct lustre_md *);
1506 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1508 int (*m_set_open_replay_data)(struct obd_export *,
1509 struct obd_client_handle *,
1510 struct ptlrpc_request *);
1511 int (*m_clear_open_replay_data)(struct obd_export *,
1512 struct obd_client_handle *);
1513 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u32 *);
1515 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1516 const struct lu_fid *, ldlm_type_t,
1517 ldlm_policy_data_t *, ldlm_mode_t,
1518 struct lustre_handle *);
1520 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1521 ldlm_policy_data_t *, ldlm_mode_t,
1522 ldlm_cancel_flags_t flags, void *opaque);
1523 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1524 renew_capa_cb_t cb);
1525 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1526 const struct req_msg_field *, struct obd_capa **);
1528 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1529 struct obd_capa *, __u32,
1530 struct ptlrpc_request **);
1532 int (*m_intent_getattr_async)(struct obd_export *,
1533 struct md_enqueue_info *,
1534 struct ldlm_enqueue_info *);
1536 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1537 struct lu_fid *, __u32 *);
1540 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1541 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1542 * wrapper function in include/linux/obd_class.h.
1546 struct lsm_operations {
1547 void (*lsm_free)(struct lov_stripe_md *);
1548 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1549 struct obd_export *md_exp);
1550 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1552 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1554 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1556 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1557 struct lov_mds_md *lmm);
1560 extern const struct lsm_operations lsm_v1_ops;
1561 extern const struct lsm_operations lsm_v3_ops;
1562 static inline const struct lsm_operations *lsm_op_find(int magic)
1570 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1575 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1577 /* Requests for obd_extent_calc() */
1578 #define OBD_CALC_STRIPE_START 1
1579 #define OBD_CALC_STRIPE_END 2
1581 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1582 struct obd_export *exp, int error)
1585 CERROR("%s: transno "LPU64" commit error: %d\n",
1586 obd->obd_name, transno, error);
1589 if (exp && transno > exp->exp_last_committed) {
1590 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1591 obd->obd_name, transno);
1592 exp->exp_last_committed = transno;
1593 ptlrpc_commit_replies(exp);
1595 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1596 obd->obd_name, transno);
1598 if (transno > obd->obd_last_committed)
1599 obd->obd_last_committed = transno;
1602 static inline void init_obd_quota_ops(quota_interface_t *interface,
1603 struct obd_ops *obd_ops)
1609 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1610 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1611 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1614 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1616 return oinfo->oi_capa;
1619 static inline struct md_open_data *obd_mod_alloc(void)
1621 struct md_open_data *mod;
1625 cfs_atomic_set(&mod->mod_refcount, 1);
1629 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1630 #define obd_mod_put(mod) \
1632 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1633 if ((mod)->mod_open_req) \
1634 ptlrpc_req_finished((mod)->mod_open_req); \
1635 OBD_FREE_PTR(mod); \
1639 extern void obdo_from_inode(struct obdo *dst, struct inode *src,
1640 struct lu_fid *parent, obd_flag valid);
1642 #endif /* __OBD_H */