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, 2012, Whamcloud, Inc.
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 #define IOC_MDC_MAX_NR 50
59 #include <lustre/lustre_idl.h>
60 #ifdef HAVE_SERVER_SUPPORT
61 # include <lu_target.h>
64 #include <lustre_lib.h>
65 #include <lustre_export.h>
66 #include <lustre_quota.h>
67 #include <lustre_fld.h>
68 #include <lustre_capa.h>
70 #include <libcfs/bitmap.h>
73 #define MAX_OBD_DEVICES 8192
81 struct lov_oinfo { /* per-stripe data structure */
82 struct ost_id loi_oi; /* object ID/Sequence on the target OST */
83 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
84 int loi_ost_gen; /* generation of this loi_ost_idx */
86 unsigned long loi_kms_valid:1;
87 __u64 loi_kms; /* known minimum size */
88 struct ost_lvb loi_lvb;
89 struct osc_async_rc loi_ar;
91 #define loi_id loi_oi.oi_id
92 #define loi_seq loi_oi.oi_seq
94 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
97 oinfo->loi_kms_valid = 1;
100 static inline void loi_init(struct lov_oinfo *loi)
104 struct lov_stripe_md {
105 cfs_atomic_t lsm_refc;
106 cfs_spinlock_t lsm_lock;
107 pid_t lsm_lock_owner; /* debugging */
109 /* maximum possible file size, might change as OSTs status changes,
110 * e.g. disconnected, deactivated */
113 /* Public members. */
114 __u64 lw_object_id; /* lov object id */
115 __u64 lw_object_seq; /* lov object seq */
117 /* LOV-private members start here -- only for use in lov/. */
119 __u32 lw_stripe_size; /* size of the stripe */
120 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
121 __u16 lw_stripe_count; /* number of objects being striped over */
122 __u16 lw_layout_gen; /* generation of the layout */
123 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
126 struct lov_oinfo *lsm_oinfo[0];
129 #define lsm_object_id lsm_wire.lw_object_id
130 #define lsm_object_seq lsm_wire.lw_object_seq
131 #define lsm_magic lsm_wire.lw_magic
132 #define lsm_layout_gen lsm_wire.lw_layout_gen
133 #define lsm_stripe_size lsm_wire.lw_stripe_size
134 #define lsm_pattern lsm_wire.lw_pattern
135 #define lsm_stripe_count lsm_wire.lw_stripe_count
136 #define lsm_pool_name lsm_wire.lw_pool_name
140 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
142 /* obd info for a particular level (lov, osc). */
144 /* Lock policy. It keeps an extent which is specific for a particular
145 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
146 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
147 ldlm_policy_data_t oi_policy;
148 /* Flags used for set request specific flags:
149 - while lock handling, the flags obtained on the enqueue
150 request are set here.
151 - while stats, the flags used for control delay/resend.
152 - while setattr, the flags used for distinguish punch operation
155 /* Lock handle specific for every OSC lock. */
156 struct lustre_handle *oi_lockh;
157 /* lsm data specific for every OSC. */
158 struct lov_stripe_md *oi_md;
159 /* obdo data specific for every OSC, if needed at all. */
161 /* statfs data specific for every OSC, if needed at all. */
162 struct obd_statfs *oi_osfs;
163 /* An update callback which is called to update some data on upper
164 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
165 * request in osc level for enqueue requests. It is also possible to
166 * update some caller data from LOV layer if needed. */
167 obd_enqueue_update_f oi_cb_up;
168 /* oss capability, its type is obd_capa in client to avoid copy.
169 * in contrary its type is lustre_capa in OSS. */
171 /* transfer jobid from ost_sync() to filter_sync()... */
175 /* compare all relevant fields. */
176 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
177 struct lov_stripe_md *m2)
180 * ->lsm_wire contains padding, but it should be zeroed out during
183 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
186 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
187 struct lov_stripe_md *lsm)
189 if (lsm->lsm_magic != lum->lmm_magic)
191 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
192 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
194 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
195 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
197 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
198 (lsm->lsm_pattern != lum->lmm_pattern))
200 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
201 (strncmp(lsm->lsm_pool_name,
202 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
203 LOV_MAXPOOLNAME) != 0))
208 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
210 struct lov_user_md *lum)
212 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
215 *lmm_magic = lumv3->lmm_magic;
217 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
218 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
219 *lmm_magic = LOV_USER_MAGIC_V1;
220 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
221 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
223 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
224 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
226 lustre_swab_lov_user_md_v3(lumv3);
227 *lmm_magic = LOV_USER_MAGIC_V3;
228 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
230 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
231 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
237 void lov_stripe_lock(struct lov_stripe_md *md);
238 void lov_stripe_unlock(struct lov_stripe_md *md);
241 cfs_list_t typ_chain;
242 struct obd_ops *typ_dt_ops;
243 struct md_ops *typ_md_ops;
244 cfs_proc_dir_entry_t *typ_procroot;
247 struct lu_device_type *typ_lu;
248 cfs_spinlock_t obd_type_lock;
258 /* Individual type definitions */
260 struct ost_server_data;
262 struct osd_properties {
263 size_t osd_max_ea_size;
266 #define OBT_MAGIC 0xBDDECEAE
267 /* hold common fields for "target" device */
268 struct obd_device_target {
271 struct super_block *obt_sb;
272 /** last_rcvd file */
273 struct file *obt_rcvd_filp;
274 #ifdef HAVE_SERVER_SUPPORT
275 struct lu_target *obt_lut;
277 __u64 obt_mount_count;
278 cfs_semaphore_t obt_quotachecking;
279 struct lustre_quota_ctxt obt_qctxt;
280 lustre_quota_version_t obt_qfmt;
281 cfs_rw_semaphore_t obt_rwsem;
282 struct vfsmount *obt_vfsmnt;
283 struct file *obt_health_check_filp;
284 struct osd_properties obt_osd_properties;
285 struct obd_job_stats obt_jobstats;
290 LLOG_CONFIG_ORIG_CTXT = 0,
291 LLOG_CONFIG_REPL_CTXT,
292 LLOG_MDS_OST_ORIG_CTXT,
293 LLOG_MDS_OST_REPL_CTXT,
300 LLOG_LOVEA_ORIG_CTXT,
301 LLOG_LOVEA_REPL_CTXT,
302 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
303 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
304 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
308 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
310 struct filter_subdirs {
311 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
321 /* NB this field MUST be first */
322 struct obd_device_target fo_obt;
323 const char *fo_fstype;
326 cfs_dentry_t *fo_dentry_O;
327 cfs_dentry_t **fo_dentry_O_groups;
328 struct filter_subdirs *fo_dentry_O_sub;
329 cfs_mutex_t fo_init_lock; /* group initialization lock */
330 int fo_committed_group;
332 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
334 unsigned long fo_destroys_in_progress;
335 cfs_mutex_t fo_create_locks[FILTER_SUBDIR_COUNT];
337 cfs_list_t fo_export_list;
340 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
341 obd_size fo_tot_granted; /* all values in bytes */
342 obd_size fo_tot_pending;
343 int fo_tot_granted_clients;
345 obd_size fo_readcache_max_filesize;
346 cfs_spinlock_t fo_flags_lock;
347 unsigned int fo_read_cache:1, /**< enable read-only cache */
348 fo_writethrough_cache:1,/**< read cache writes */
349 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
350 fo_raid_degraded:1;/**< RAID device degraded */
352 struct obd_import *fo_mdc_imp;
353 struct obd_uuid fo_mdc_uuid;
354 struct lustre_handle fo_mdc_conn;
355 struct file **fo_last_objid_files;
356 __u64 *fo_last_objids; /* last created objid for groups,
357 * protected by fo_objidlock */
359 cfs_mutex_t fo_alloc_lock;
361 cfs_atomic_t fo_r_in_flight;
362 cfs_atomic_t fo_w_in_flight;
365 * per-filter pool of kiobuf's allocated by filter_common_setup() and
366 * torn down by filter_cleanup().
368 * This pool contains kiobuf used by
369 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
372 * Locking: protected by internal lock of cfs_hash, pool can be
373 * found from this hash table by t_id of ptlrpc_thread.
375 struct cfs_hash *fo_iobuf_hash;
377 cfs_list_t fo_llog_list;
378 cfs_spinlock_t fo_llog_list_lock;
380 struct brw_stats fo_filter_stats;
381 struct lustre_quota_ctxt fo_quota_ctxt;
382 cfs_spinlock_t fo_quotacheck_lock;
383 cfs_atomic_t fo_quotachecking;
385 int fo_fmd_max_num; /* per exp filter_mod_data */
386 int fo_fmd_max_age; /* jiffies to fmd expiry */
387 unsigned long fo_syncjournal:1, /* sync journal on writes */
388 fo_sync_lock_cancel:2;/* sync on lock cancel */
392 cfs_rwlock_t fo_sptlrpc_lock;
393 struct sptlrpc_rule_set fo_sptlrpc_rset;
395 /* capability related */
396 unsigned int fo_fl_oss_capa;
397 cfs_list_t fo_capa_keys;
398 cfs_hlist_head_t *fo_capa_hash;
399 struct llog_commit_master *fo_lcm;
403 struct timeout_item {
404 enum timeout_event ti_event;
405 cfs_time_t ti_timeout;
408 cfs_list_t ti_obd_list;
412 #define OSC_MAX_RIF_DEFAULT 8
413 #define MDS_OSC_MAX_RIF_DEFAULT 50
414 #define OSC_MAX_RIF_MAX 256
415 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
416 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
417 #define OSC_DEFAULT_RESENDS 10
419 /* possible values for fo_sync_lock_cancel */
421 NEVER_SYNC_ON_CANCEL = 0,
422 BLOCKING_SYNC_ON_CANCEL = 1,
423 ALWAYS_SYNC_ON_CANCEL = 2,
424 NUM_SYNC_ON_CANCEL_STATES
427 #define MDC_MAX_RIF_DEFAULT 8
428 #define MDC_MAX_RIF_MAX 512
433 cfs_rw_semaphore_t cl_sem;
434 struct obd_uuid cl_target_uuid;
435 struct obd_import *cl_import; /* ptlrpc connection state */
437 /* max_mds_easize is purely a performance thing so we don't have to
438 * call obd_size_diskmd() all the time. */
439 int cl_default_mds_easize;
440 int cl_max_mds_easize;
441 int cl_max_mds_cookiesize;
443 enum lustre_sec_part cl_sp_me;
444 enum lustre_sec_part cl_sp_to;
445 struct sptlrpc_flavor cl_flvr_mgc; /* fixed flavor of mgc->mgs */
447 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
448 void *cl_llcd_offset;
450 /* the grant values are protected by loi_list_lock below */
451 long cl_dirty; /* all _dirty_ in bytes */
452 long cl_dirty_max; /* allowed w/o rpc */
453 long cl_dirty_transit; /* dirty synchronous */
454 long cl_avail_grant; /* bytes of credit for ost */
455 long cl_lost_grant; /* lost credits (trunc) */
457 /* since we allocate grant by blocks, we don't know how many grant will
458 * be used to add a page into cache. As a solution, we reserve maximum
459 * grant before trying to dirty a page and unreserve the rest.
460 * See osc_{reserve|unreserve}_grant for details. */
461 long cl_reserved_grant;
462 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
463 cfs_time_t cl_next_shrink_grant; /* jiffies */
464 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
465 int cl_grant_shrink_interval; /* seconds */
467 /* A chunk is an optimal size used by osc_extent to determine
468 * the extent size. A chunk is max(CFS_PAGE_SIZE, OST block size) */
471 int cl_extent_tax; /* extent overhead, by bytes */
473 /* keep track of objects that have lois that contain pages which
474 * have been queued for async brw. this lock also protects the
475 * lists of osc_client_pages that hang off of the loi */
477 * ->cl_loi_list_lock protects consistency of
478 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
479 * ->ap_completion() call-backs are executed under this lock. As we
480 * cannot guarantee that these call-backs never block on all platforms
481 * (as a matter of fact they do block on Mac OS X), type of
482 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
483 * and blocking mutex on Mac OS X. (Alternative is to make this lock
484 * blocking everywhere, but we don't want to slow down fast-path of
485 * our main platform.)
487 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
488 * with client_obd_list_{un,}lock() and
489 * client_obd_list_lock_{init,done}() functions.
491 * NB by Jinshan: though field names are still _loi_, but actually
492 * osc_object{}s are in the list.
494 client_obd_lock_t cl_loi_list_lock;
495 cfs_list_t cl_loi_ready_list;
496 cfs_list_t cl_loi_hp_ready_list;
497 cfs_list_t cl_loi_write_list;
498 cfs_list_t cl_loi_read_list;
501 /* just a sum of the loi/lop pending numbers to be exported by /proc */
502 cfs_atomic_t cl_pending_w_pages;
503 cfs_atomic_t cl_pending_r_pages;
504 int cl_max_pages_per_rpc;
505 int cl_max_rpcs_in_flight;
506 struct obd_histogram cl_read_rpc_hist;
507 struct obd_histogram cl_write_rpc_hist;
508 struct obd_histogram cl_read_page_hist;
509 struct obd_histogram cl_write_page_hist;
510 struct obd_histogram cl_read_offset_hist;
511 struct obd_histogram cl_write_offset_hist;
513 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
514 cfs_atomic_t cl_destroy_in_flight;
515 cfs_waitq_t cl_destroy_waitq;
517 struct mdc_rpc_lock *cl_rpc_lock;
518 struct mdc_rpc_lock *cl_close_lock;
519 struct osc_creator cl_oscc;
522 cfs_semaphore_t cl_mgc_sem;
523 struct vfsmount *cl_mgc_vfsmnt;
524 struct dentry *cl_mgc_configs_dir;
525 cfs_atomic_t cl_mgc_refcount;
526 struct obd_export *cl_mgc_mgsexp;
528 /* checksumming for data sent over the network */
529 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
530 /* supported checksum types that are worked out at connect time */
531 __u32 cl_supp_cksum_types;
532 /* checksum algorithm to be used */
533 cksum_type_t cl_cksum_type;
535 /* also protected by the poorly named _loi_list_lock lock above */
536 struct osc_async_rc cl_ar;
538 /* used by quotacheck */
539 int cl_qchk_stat; /* quotacheck stat of the peer */
541 /* sequence manager */
542 struct lu_client_seq *cl_seq;
544 cfs_atomic_t cl_resends; /* resend count */
546 /* ptlrpc work for writeback in ptlrpcd context */
547 void *cl_writeback_work;
549 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
551 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
554 struct obd_device_target mgs_obt;
555 struct ptlrpc_service *mgs_service;
556 struct vfsmount *mgs_vfsmnt;
557 struct super_block *mgs_sb;
558 struct dentry *mgs_configs_dir;
559 cfs_list_t mgs_fs_db_list;
560 cfs_mutex_t mgs_mutex;
561 cfs_proc_dir_entry_t *mgs_proc_live;
562 cfs_time_t mgs_start_time;
566 /* NB this field MUST be first */
567 struct obd_device_target mds_obt;
568 struct ptlrpc_service *mds_service;
569 struct ptlrpc_service *mds_setattr_service;
570 struct ptlrpc_service *mds_readpage_service;
572 int mds_max_cookiesize;
574 unsigned long mds_atime_diff;
575 struct ll_fid mds_rootfid;
576 cfs_dentry_t *mds_pending_dir;
577 cfs_dentry_t *mds_logs_dir;
578 cfs_dentry_t *mds_objects_dir;
579 struct llog_handle *mds_cfg_llh;
580 struct obd_device *mds_lov_obd;
581 struct obd_uuid mds_lov_uuid;
583 struct obd_export *mds_lov_exp;
584 struct lov_desc mds_lov_desc;
587 /* mark pages dirty for write. */
588 cfs_bitmap_t *mds_lov_page_dirty;
589 /* array for store pages with obd_id */
590 void **mds_lov_page_array;
591 /* file for store objid */
592 struct file *mds_lov_objid_filp;
593 __u32 mds_lov_objid_count;
594 __u32 mds_lov_objid_max_index;
595 __u32 mds_lov_objid_lastpage;
596 __u32 mds_lov_objid_lastidx;
599 struct lustre_quota_info mds_quota_info;
600 cfs_rw_semaphore_t mds_qonoff_sem;
601 unsigned long mds_fl_user_xattr:1,
603 mds_evict_ost_nids:1,
607 mds_fl_target:1; /* mds have one or
610 struct upcall_cache *mds_identity_cache;
612 /* for capability keys update */
613 struct lustre_capa_key *mds_capa_keys;
614 cfs_rw_semaphore_t mds_notify_lock;
618 extern __u32 mds_max_ost_index;
620 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
622 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
624 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
626 extern int mds_lov_init_objids(struct obd_device *obd);
627 extern void mds_lov_destroy_objids(struct obd_device *obd);
637 struct obd_device_target eo_obt;
639 cfs_spinlock_t eo_lock;
641 struct lustre_handle eo_nl_lock;
642 cfs_atomic_t eo_prep;
646 struct ptlrpc_service *ost_service;
647 struct ptlrpc_service *ost_create_service;
648 struct ptlrpc_service *ost_io_service;
649 cfs_mutex_t ost_health_mutex;
652 struct echo_client_obd {
653 struct obd_export *ec_exp; /* the local connection to osc/lov */
654 cfs_spinlock_t ec_lock;
655 cfs_list_t ec_objects;
662 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
663 cfs_list_t lqo_oss_list; /* link to lov_qos */
664 __u64 lqo_bavail; /* total bytes avail on OSS */
665 __u64 lqo_penalty; /* current penalty */
666 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
667 time_t lqo_used; /* last used time, seconds */
668 __u32 lqo_ost_count; /* number of osts on this oss */
672 struct lov_qos_oss *ltq_oss; /* oss info */
673 __u64 ltq_penalty; /* current penalty */
674 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
675 __u64 ltq_weight; /* net weighting */
676 time_t ltq_used; /* last used time, seconds */
677 unsigned int ltq_usable:1; /* usable for striping */
680 /* Generic subset of OSTs */
682 __u32 *op_array; /* array of index of
684 unsigned int op_count; /* number of OSTs in the array */
685 unsigned int op_size; /* allocated size of lp_array */
686 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
689 /* Round-robin allocator data */
691 __u32 lqr_start_idx; /* start index of new inode */
692 __u32 lqr_offset_idx; /* aliasing for start_idx */
693 int lqr_start_count; /* reseed counter */
694 struct ost_pool lqr_pool; /* round-robin optimized list */
695 unsigned long lqr_dirty:1; /* recalc round-robin list */
698 /* allow statfs data caching for 1 second */
699 #define OBD_STATFS_CACHE_SECONDS 1
701 struct lov_statfs_data {
702 struct obd_info lsd_oi;
703 struct obd_statfs lsd_statfs;
705 /* Stripe placement optimization */
707 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
708 cfs_rw_semaphore_t lq_rw_sem;
709 __u32 lq_active_oss_count;
710 unsigned int lq_prio_free; /* priority for free space */
711 unsigned int lq_threshold_rr;/* priority for rr */
712 struct lov_qos_rr lq_rr; /* round robin qos data */
713 unsigned long lq_dirty:1, /* recalc qos data */
714 lq_same_space:1,/* the ost's all have approx.
715 the same space avail */
716 lq_reset:1, /* zero current penalties */
717 lq_statfs_in_progress:1; /* statfs op in
719 /* qos statfs data */
720 struct lov_statfs_data *lq_statfs_data;
721 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
722 * requests completion */
725 struct lov_tgt_desc {
727 struct obd_uuid ltd_uuid;
728 struct obd_device *ltd_obd;
729 struct obd_export *ltd_exp;
730 struct ltd_qos ltd_qos; /* qos info per target */
732 __u32 ltd_index; /* index in lov_obd->tgts */
733 unsigned long ltd_active:1,/* is this target up for requests */
734 ltd_activate:1,/* should target be activated */
735 ltd_reap:1; /* should this target be deleted */
739 #define pool_tgt_size(_p) _p->pool_obds.op_size
740 #define pool_tgt_count(_p) _p->pool_obds.op_count
741 #define pool_tgt_array(_p) _p->pool_obds.op_array
742 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
743 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
746 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
747 struct ost_pool pool_obds; /* pool members */
748 cfs_atomic_t pool_refcount; /* pool ref. counter */
749 struct lov_qos_rr pool_rr; /* round robin qos */
750 cfs_hlist_node_t pool_hash; /* access by poolname */
751 cfs_list_t pool_list; /* serial access */
752 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
753 struct lov_obd *pool_lov; /* lov obd to which this
758 struct lov_desc desc;
759 struct lov_tgt_desc **lov_tgts; /* sparse array */
760 struct ost_pool lov_packed; /* all OSTs in a packed
762 cfs_mutex_t lov_lock;
763 struct obd_connect_data lov_ocd;
764 struct lov_qos lov_qos; /* qos info per lov */
765 cfs_atomic_t lov_refcount;
766 __u32 lov_tgt_count; /* how many OBD's */
767 __u32 lov_active_tgt_count; /* how many active */
768 __u32 lov_death_row;/* tgts scheduled to be deleted */
769 __u32 lov_tgt_size; /* size of tgts array */
772 cfs_hash_t *lov_pools_hash_body; /* used for key access */
773 cfs_list_t lov_pool_list; /* used for sequential access */
774 cfs_proc_dir_entry_t *lov_pool_proc_entry;
775 enum lustre_sec_part lov_sp_me;
778 struct lmv_tgt_desc {
779 struct obd_uuid ltd_uuid;
780 struct obd_export *ltd_exp;
781 int ltd_active; /* is this target up for requests */
783 cfs_mutex_t ltd_fid_mutex;
786 enum placement_policy {
787 PLACEMENT_CHAR_POLICY = 0,
788 PLACEMENT_NID_POLICY = 1,
789 PLACEMENT_INVAL_POLICY = 2,
793 typedef enum placement_policy placement_policy_t;
797 struct lu_client_fld lmv_fld;
798 cfs_spinlock_t lmv_lock;
799 placement_policy_t lmv_placement;
800 struct lmv_desc desc;
801 struct obd_uuid cluuid;
802 struct obd_export *exp;
809 cfs_mutex_t init_mutex;
811 struct lmv_tgt_desc *tgts;
814 struct obd_connect_data *datas;
817 struct obd_connect_data conn_data;
820 struct niobuf_local {
821 __u64 lnb_file_offset;
826 cfs_dentry_t *dentry;
831 #define LUSTRE_FLD_NAME "fld"
832 #define LUSTRE_SEQ_NAME "seq"
834 #define LUSTRE_CMM_NAME "cmm"
835 #define LUSTRE_MDD_NAME "mdd"
836 #define LUSTRE_OSD_NAME "osd-ldiskfs"
837 #define LUSTRE_OSD_ZFS_NAME "osd-zfs"
838 #define LUSTRE_VVP_NAME "vvp"
839 #define LUSTRE_LMV_NAME "lmv"
840 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
841 #define LUSTRE_SLP_NAME "slp"
843 /* obd device type names */
844 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
845 #define LUSTRE_MDS_NAME "mds"
846 #define LUSTRE_MDT_NAME "mdt"
847 #define LUSTRE_MDC_NAME "mdc"
848 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
849 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
850 #define LUSTRE_OSC_NAME "osc"
851 #define LUSTRE_LOV_NAME "lov"
852 #define LUSTRE_MGS_NAME "mgs"
853 #define LUSTRE_MGC_NAME "mgc"
855 #define LUSTRE_CACHEOBD_NAME "cobd"
856 #define LUSTRE_ECHO_NAME "obdecho"
857 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
859 /* Constant obd names (post-rename) */
860 #define LUSTRE_MDS_OBDNAME "MDS"
861 #define LUSTRE_OSS_OBDNAME "OSS"
862 #define LUSTRE_MGS_OBDNAME "MGS"
863 #define LUSTRE_MGC_OBDNAME "MGC"
865 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
866 #define N_LOCAL_TEMP_PAGE 0x10000000
868 struct obd_trans_info {
871 /* Only used on the server side for tracking acks. */
872 struct oti_req_ack_lock {
873 struct lustre_handle lock;
877 struct llog_cookie oti_onecookie;
878 struct llog_cookie *oti_logcookies;
880 /** synchronous write is needed */
881 long oti_sync_write:1;
883 /* initial thread handling transaction */
884 struct ptlrpc_thread * oti_thread;
887 __u64 oti_pre_version;
891 struct obd_uuid *oti_ost_uuid;
894 static inline void oti_init(struct obd_trans_info *oti,
895 struct ptlrpc_request *req)
899 memset(oti, 0, sizeof(*oti));
904 oti->oti_xid = req->rq_xid;
905 /** VBR: take versions from request */
906 if (req->rq_reqmsg != NULL &&
907 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
908 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
909 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
910 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
913 /** called from mds_create_objects */
914 if (req->rq_repmsg != NULL)
915 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
916 oti->oti_thread = req->rq_svc_thread;
917 if (req->rq_reqmsg != NULL)
918 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
921 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
926 if (num_cookies == 1)
927 oti->oti_logcookies = &oti->oti_onecookie;
929 OBD_ALLOC_LARGE(oti->oti_logcookies,
930 num_cookies * sizeof(oti->oti_onecookie));
932 oti->oti_numcookies = num_cookies;
935 static inline void oti_free_cookies(struct obd_trans_info *oti)
937 if (!oti || !oti->oti_logcookies)
940 if (oti->oti_logcookies == &oti->oti_onecookie)
941 LASSERT(oti->oti_numcookies == 1);
943 OBD_FREE_LARGE(oti->oti_logcookies,
944 oti->oti_numcookies*sizeof(oti->oti_onecookie));
945 oti->oti_logcookies = NULL;
946 oti->oti_numcookies = 0;
950 * Events signalled through obd_notify() upcall-chain.
952 enum obd_notify_event {
955 /* Device connect start */
957 /* Device activated */
959 /* Device deactivated */
961 /* Device disconnected */
963 /* Connect data for import were changed */
966 OBD_NOTIFY_SYNC_NONBLOCK,
968 /* Configuration event */
970 /* Trigger quota recovery */
972 /* Administratively deactivate/activate event */
973 OBD_NOTIFY_DEACTIVATE,
977 /* bit-mask flags for config events */
979 CONFIG_LOG = 0x1, /* finished processing config log */
980 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
981 CONFIG_TARGET = 0x4 /* one target is added */
985 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
986 * and liblustre being main examples).
988 struct obd_notify_upcall {
989 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
990 enum obd_notify_event ev, void *owner, void *data);
991 /* Opaque datum supplied by upper layer listener */
995 struct target_recovery_data {
996 svc_handler_t trd_recovery_handler;
997 pid_t trd_processing_task;
998 cfs_completion_t trd_starting;
999 cfs_completion_t trd_finishing;
1003 * In HEAD for CMD, the object is created in group number which is 3>=
1004 * or indexing starts from 3. To test this assertions are added to disallow
1005 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
1006 * object in group 0 needs to be allowed.
1007 * So for interop mode following changes needs to be done:
1008 * 1. No need to assert on group 0 or allow group 0
1009 * 2. The group number indexing starts from 0 instead of 3
1012 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
1014 static inline __u64 objseq_to_mdsno(obd_seq seq)
1016 LASSERT_SEQ_IS_MDT(seq);
1017 if (seq == FID_SEQ_OST_MDT0)
1019 return seq - FID_SEQ_OST_MDT1 + 1;
1022 static inline int mdt_to_obd_objseq(int mdtid)
1025 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
1029 return FID_SEQ_OST_MDT1 + mdtid - 1;
1033 struct obd_llog_group {
1034 cfs_list_t olg_list;
1036 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
1037 cfs_waitq_t olg_waitq;
1038 cfs_spinlock_t olg_lock;
1039 struct obd_export *olg_exp;
1040 int olg_initializing;
1041 cfs_mutex_t olg_cat_processing;
1044 /* corresponds to one of the obd's */
1045 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1046 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1049 struct obd_type *obd_type;
1052 /* common and UUID name of this device */
1053 char obd_name[MAX_OBD_NAME];
1054 struct obd_uuid obd_uuid;
1056 struct lu_device *obd_lu_dev;
1059 /* bitfield modification is protected by obd_dev_lock */
1060 unsigned long obd_attached:1, /* finished attach */
1061 obd_set_up:1, /* finished setup */
1062 obd_recovering:1, /* there are recoverable clients */
1063 obd_abort_recovery:1,/* recovery expired */
1064 obd_version_recov:1, /* obd uses version checking */
1065 obd_replayable:1, /* recovery is enabled; inform clients */
1066 obd_no_transno:1, /* no committed-transno notification */
1067 obd_no_recov:1, /* fail instead of retry messages */
1068 obd_stopping:1, /* started cleanup */
1069 obd_starting:1, /* started setup */
1070 obd_force:1, /* cleanup with > 0 obd refcount */
1071 obd_fail:1, /* cleanup with failover */
1072 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1073 obd_no_conn:1, /* deny new connections */
1074 obd_inactive:1, /* device active/inactive
1075 * (for /proc/status only!!) */
1076 obd_no_ir:1, /* no imperative recovery. */
1077 obd_process_conf:1; /* device is processing mgs config */
1078 /* use separate field as it is set in interrupt to don't mess with
1079 * protection of other bits using _bh lock */
1080 unsigned long obd_recovery_expired:1;
1081 /* uuid-export hash body */
1082 cfs_hash_t *obd_uuid_hash;
1083 /* nid-export hash body */
1084 cfs_hash_t *obd_nid_hash;
1085 /* nid stats body */
1086 cfs_hash_t *obd_nid_stats_hash;
1087 cfs_list_t obd_nid_stats;
1088 cfs_atomic_t obd_refcount;
1089 cfs_waitq_t obd_refcount_waitq;
1090 cfs_list_t obd_exports;
1091 cfs_list_t obd_unlinked_exports;
1092 cfs_list_t obd_delayed_exports;
1093 int obd_num_exports;
1094 cfs_spinlock_t obd_nid_lock;
1095 struct ldlm_namespace *obd_namespace;
1096 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1097 /* a spinlock is OK for what we do now, may need a semaphore later */
1098 cfs_spinlock_t obd_dev_lock; /* protects obd bitfield above */
1099 cfs_mutex_t obd_dev_mutex;
1100 __u64 obd_last_committed;
1101 struct fsfilt_operations *obd_fsops;
1102 cfs_spinlock_t obd_osfs_lock;
1103 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1105 struct lvfs_run_ctxt obd_lvfs_ctxt;
1106 struct obd_llog_group obd_olg; /* default llog group */
1107 struct obd_device *obd_observer;
1108 cfs_rw_semaphore_t obd_observer_link_sem;
1109 struct obd_notify_upcall obd_upcall;
1110 struct obd_export *obd_self_export;
1111 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1112 cfs_list_t obd_exports_timed;
1113 time_t obd_eviction_timer; /* for ping evictor */
1115 int obd_max_recoverable_clients;
1116 cfs_atomic_t obd_connected_clients;
1117 int obd_stale_clients;
1118 int obd_delayed_clients;
1119 /* this lock protects all recovery list_heads, timer and
1120 * obd_next_recovery_transno value */
1121 cfs_spinlock_t obd_recovery_task_lock;
1122 __u64 obd_next_recovery_transno;
1123 int obd_replayed_requests;
1124 int obd_requests_queued_for_recovery;
1125 cfs_waitq_t obd_next_transno_waitq;
1126 /* protected by obd_recovery_task_lock */
1127 cfs_timer_t obd_recovery_timer;
1128 time_t obd_recovery_start; /* seconds */
1129 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1130 int obd_recovery_time_hard;
1131 int obd_recovery_timeout;
1132 int obd_recovery_ir_factor;
1134 /* new recovery stuff from CMD2 */
1135 struct target_recovery_data obd_recovery_data;
1136 int obd_replayed_locks;
1137 cfs_atomic_t obd_req_replay_clients;
1138 cfs_atomic_t obd_lock_replay_clients;
1139 /* all lists are protected by obd_recovery_task_lock */
1140 cfs_list_t obd_req_replay_queue;
1141 cfs_list_t obd_lock_replay_queue;
1142 cfs_list_t obd_final_req_queue;
1143 int obd_recovery_stage;
1146 struct obd_device_target obt;
1147 struct filter_obd filter;
1149 struct client_obd cli;
1151 struct echo_client_obd echo_client;
1152 struct echo_obd echo;
1157 /* Fields used by LProcFS */
1158 unsigned int obd_cntr_base;
1159 struct lprocfs_stats *obd_stats;
1161 unsigned int md_cntr_base;
1162 struct lprocfs_stats *md_stats;
1164 cfs_proc_dir_entry_t *obd_proc_entry;
1165 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1166 cfs_proc_dir_entry_t *obd_svc_procroot;
1167 struct lprocfs_stats *obd_svc_stats;
1168 cfs_atomic_t obd_evict_inprogress;
1169 cfs_waitq_t obd_evict_inprogress_waitq;
1170 cfs_list_t obd_evict_list; /* protected with pet_lock */
1173 * Ldlm pool part. Save last calculated SLV and Limit.
1175 cfs_rwlock_t obd_pool_lock;
1180 * A list of outstanding class_incref()'s against this obd. For
1183 struct lu_ref obd_reference;
1185 int obd_conn_inprogress;
1188 #define OBD_LLOG_FL_SENDNOW 0x0001
1189 #define OBD_LLOG_FL_EXIT 0x0002
1191 enum obd_cleanup_stage {
1192 /* Special case hack for MDS LOVs */
1194 /* can be directly mapped to .ldto_device_fini() */
1195 OBD_CLEANUP_EXPORTS,
1198 /* get/set_info keys */
1199 #define KEY_ASYNC "async"
1200 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1201 #define KEY_BLOCKSIZE "blocksize"
1202 #define KEY_CAPA_KEY "capa_key"
1203 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1204 #define KEY_FID2PATH "fid2path"
1205 #define KEY_CHECKSUM "checksum"
1206 #define KEY_CLEAR_FS "clear_fs"
1207 #define KEY_CONN_DATA "conn_data"
1208 #define KEY_EVICT_BY_NID "evict_by_nid"
1209 #define KEY_FIEMAP "fiemap"
1210 #define KEY_FLUSH_CTX "flush_ctx"
1211 #define KEY_GRANT_SHRINK "grant_shrink"
1212 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1213 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1214 #define KEY_INIT_RECOV "initial_recov"
1215 #define KEY_INTERMDS "inter_mds"
1216 #define KEY_LAST_ID "last_id"
1217 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1218 #define KEY_LOVDESC "lovdesc"
1219 #define KEY_LOV_IDX "lov_idx"
1220 #define KEY_MAX_EASIZE "max_easize"
1221 #define KEY_MDS_CONN "mds_conn"
1222 #define KEY_MGSSEC "mgssec"
1223 #define KEY_NEXT_ID "next_id"
1224 #define KEY_READ_ONLY "read-only"
1225 #define KEY_REGISTER_TARGET "register_target"
1226 #define KEY_REVIMP_UPD "revimp_update"
1227 #define KEY_SET_FS "set_fs"
1228 #define KEY_TGT_COUNT "tgt_count"
1229 /* KEY_SET_INFO in lustre_idl.h */
1230 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1231 #define KEY_CONNECT_FLAG "connect_flags"
1232 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1236 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1237 #define IT_OPEN (1 << 0)
1238 #define IT_CREAT (1 << 1)
1239 #define IT_READDIR (1 << 2)
1240 #define IT_GETATTR (1 << 3)
1241 #define IT_LOOKUP (1 << 4)
1242 #define IT_UNLINK (1 << 5)
1243 #define IT_TRUNC (1 << 6)
1244 #define IT_GETXATTR (1 << 7)
1245 #define IT_EXEC (1 << 8)
1246 #define IT_PIN (1 << 9)
1247 #define IT_LAYOUT (1 << 10)
1249 static inline int it_to_lock_mode(struct lookup_intent *it)
1251 /* CREAT needs to be tested before open (both could be set) */
1252 if (it->it_op & IT_CREAT)
1254 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1258 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1263 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1264 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1265 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1266 struct lu_fid op_fid4; /* to the operation locks. */
1267 mdsno_t op_mds; /* what mds server open will go to */
1268 struct lustre_handle op_handle;
1269 obd_time op_mod_time;
1270 const char *op_name;
1273 struct lmv_stripe_md *op_mea1;
1274 struct lmv_stripe_md *op_mea2;
1275 __u32 op_suppgids[2];
1281 /* iattr fields and blocks. */
1282 struct iattr op_attr;
1284 unsigned int op_attr_flags;
1287 loff_t op_attr_blocks;
1289 /* Size-on-MDS epoch and flags. */
1294 struct obd_capa *op_capa1;
1295 struct obd_capa *op_capa2;
1297 /* Various operation flags. */
1300 /* Operation type */
1303 /* Used by readdir */
1308 struct md_enqueue_info;
1309 /* metadata stat-ahead */
1310 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1311 struct md_enqueue_info *minfo,
1314 struct md_enqueue_info {
1315 struct md_op_data mi_data;
1316 struct lookup_intent mi_it;
1317 struct lustre_handle mi_lockh;
1318 struct inode *mi_dir;
1319 md_enqueue_cb_t mi_cb;
1321 unsigned int mi_generation;
1325 cfs_module_t *o_owner;
1326 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1327 void *karg, void *uarg);
1328 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1329 __u32 keylen, void *key, __u32 *vallen, void *val,
1330 struct lov_stripe_md *lsm);
1331 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1332 __u32 keylen, void *key,
1333 __u32 vallen, void *val,
1334 struct ptlrpc_request_set *set);
1335 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1336 int (*o_detach)(struct obd_device *dev);
1337 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1338 int (*o_precleanup)(struct obd_device *dev,
1339 enum obd_cleanup_stage cleanup_stage);
1340 int (*o_cleanup)(struct obd_device *dev);
1341 int (*o_process_config)(struct obd_device *dev, obd_count len,
1343 int (*o_postrecov)(struct obd_device *dev);
1344 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1346 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1347 /* connect to the target device with given connection
1348 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1349 * granted by the target, which are guaranteed to be a subset of flags
1350 * asked for. If @ocd == NULL, use default parameters. */
1351 int (*o_connect)(const struct lu_env *env,
1352 struct obd_export **exp, struct obd_device *src,
1353 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1355 int (*o_reconnect)(const struct lu_env *env,
1356 struct obd_export *exp, struct obd_device *src,
1357 struct obd_uuid *cluuid,
1358 struct obd_connect_data *ocd,
1360 int (*o_disconnect)(struct obd_export *exp);
1362 /* Initialize/finalize fids infrastructure. */
1363 int (*o_fid_init)(struct obd_export *exp);
1364 int (*o_fid_fini)(struct obd_export *exp);
1366 /* Allocate new fid according to passed @hint. */
1367 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1368 struct md_op_data *op_data);
1371 * Object with @fid is getting deleted, we may want to do something
1374 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1376 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1377 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1378 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1379 __u64 max_age, struct ptlrpc_request_set *set);
1380 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1381 struct lov_stripe_md *mem_src);
1382 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1383 struct lov_mds_md *disk_src, int disk_len);
1384 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1386 /* FIXME: add fid capability support for create & destroy! */
1387 int (*o_precreate)(struct obd_export *exp);
1388 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1389 struct obdo *oa, struct lov_stripe_md **ea,
1390 struct obd_trans_info *oti);
1391 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1392 struct lov_stripe_md **ea,
1393 struct obd_trans_info *oti);
1394 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1395 struct obdo *oa, struct lov_stripe_md *ea,
1396 struct obd_trans_info *oti, struct obd_export *md_exp,
1398 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1399 struct obd_info *oinfo, struct obd_trans_info *oti);
1400 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1401 struct obd_trans_info *oti,
1402 struct ptlrpc_request_set *rqset);
1403 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1404 struct obd_info *oinfo);
1405 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1406 struct ptlrpc_request_set *set);
1407 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1408 obd_count oa_bufs, struct brw_page *pgarr,
1409 struct obd_trans_info *oti);
1410 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1411 struct ost_lvb *lvb, int kms_only);
1412 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1413 obd_off size, int shrink);
1414 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1415 struct obd_info *oinfo, struct obd_trans_info *oti,
1416 struct ptlrpc_request_set *rqset);
1417 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1418 struct obd_info *oinfo, obd_size start, obd_size end,
1419 struct ptlrpc_request_set *set);
1420 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1421 struct lov_stripe_md *src, obd_size start,
1422 obd_size end, struct obd_trans_info *oti);
1423 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1424 struct lustre_handle *srconn, struct lov_stripe_md *src,
1425 obd_size start, obd_size end, struct obd_trans_info *);
1426 int (*o_iterate)(struct lustre_handle *conn,
1427 int (*)(obd_id, obd_seq, void *),
1428 obd_id *startid, obd_seq seq, void *data);
1429 int (*o_preprw)(const struct lu_env *env, int cmd,
1430 struct obd_export *exp, struct obdo *oa, int objcount,
1431 struct obd_ioobj *obj, struct niobuf_remote *remote,
1432 int *nr_pages, struct niobuf_local *local,
1433 struct obd_trans_info *oti, struct lustre_capa *capa);
1434 int (*o_commitrw)(const struct lu_env *env, int cmd,
1435 struct obd_export *exp, struct obdo *oa,
1436 int objcount, struct obd_ioobj *obj,
1437 struct niobuf_remote *remote, int pages,
1438 struct niobuf_local *local,
1439 struct obd_trans_info *oti, int rc);
1440 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1441 struct ldlm_enqueue_info *einfo,
1442 struct ptlrpc_request_set *rqset);
1443 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1444 ldlm_iterator_t it, void *data);
1445 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1446 ldlm_iterator_t it, void *data);
1447 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1448 __u32 mode, struct lustre_handle *);
1449 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1450 ldlm_cancel_flags_t flags, void *opaque);
1451 int (*o_init_export)(struct obd_export *exp);
1452 int (*o_destroy_export)(struct obd_export *exp);
1453 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1454 int cmd, obd_off *);
1456 /* llog related obd_methods */
1457 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1458 struct obd_device *disk_obd, int *idx);
1459 int (*o_llog_finish)(struct obd_device *obd, int count);
1460 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1462 /* metadata-only methods */
1463 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1464 struct obd_capa *, struct obd_client_handle *, int flag);
1465 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1467 int (*o_import_event)(struct obd_device *, struct obd_import *,
1468 enum obd_import_event);
1470 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1471 enum obd_notify_event ev, void *data);
1473 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1474 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1477 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1478 struct obd_quotactl *);
1479 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1480 struct obd_quotactl *);
1481 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1482 struct quota_adjust_qunit *oqaq,
1483 struct lustre_quota_ctxt *qctxt,
1484 struct ptlrpc_request_set *rqset);
1487 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1490 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1491 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1492 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1494 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1496 void (*o_getref)(struct obd_device *obd);
1497 void (*o_putref)(struct obd_device *obd);
1499 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1500 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1501 * Also, add a wrapper function in include/linux/obd_class.h. */
1505 LUSTRE_OPC_MKDIR = (1 << 0),
1506 LUSTRE_OPC_SYMLINK = (1 << 1),
1507 LUSTRE_OPC_MKNOD = (1 << 2),
1508 LUSTRE_OPC_CREATE = (1 << 3),
1509 LUSTRE_OPC_ANY = (1 << 4)
1512 /* lmv structures */
1513 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1514 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1515 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1517 #define MAX_HASH_SIZE_32 0x7fffffffUL
1518 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1519 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1522 struct mdt_body *body;
1523 struct lov_stripe_md *lsm;
1524 struct lmv_stripe_md *mea;
1525 #ifdef CONFIG_FS_POSIX_ACL
1526 struct posix_acl *posix_acl;
1528 struct mdt_remote_perm *remote_perm;
1529 struct obd_capa *mds_capa;
1530 struct obd_capa *oss_capa;
1533 struct md_open_data {
1534 struct obd_client_handle *mod_och;
1535 struct ptlrpc_request *mod_open_req;
1536 struct ptlrpc_request *mod_close_req;
1537 cfs_atomic_t mod_refcount;
1540 struct lookup_intent;
1543 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1544 struct obd_capa **);
1545 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1546 ldlm_iterator_t, void *);
1547 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1548 ldlm_iterator_t, void *);
1549 int (*m_close)(struct obd_export *, struct md_op_data *,
1550 struct md_open_data *, struct ptlrpc_request **);
1551 int (*m_create)(struct obd_export *, struct md_op_data *,
1552 const void *, int, int, __u32, __u32, cfs_cap_t,
1553 __u64, struct ptlrpc_request **);
1554 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1555 struct md_open_data *);
1556 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1557 struct lookup_intent *, struct md_op_data *,
1558 struct lustre_handle *, void *, int,
1559 struct ptlrpc_request **, int);
1560 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1561 struct ptlrpc_request **);
1562 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1563 struct ptlrpc_request **);
1564 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1565 void *, int, struct lookup_intent *, int,
1566 struct ptlrpc_request **,
1567 ldlm_blocking_callback, int);
1568 int (*m_link)(struct obd_export *, struct md_op_data *,
1569 struct ptlrpc_request **);
1570 int (*m_rename)(struct obd_export *, struct md_op_data *,
1571 const char *, int, const char *, int,
1572 struct ptlrpc_request **);
1573 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1574 const struct lu_fid *,
1575 struct ptlrpc_request **);
1576 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1577 int , void *, int, struct ptlrpc_request **,
1578 struct md_open_data **mod);
1579 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1580 struct obd_capa *, struct ptlrpc_request **);
1581 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1582 struct page **, struct ptlrpc_request **);
1584 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1585 struct ptlrpc_request **);
1587 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1588 struct obd_capa *, obd_valid, const char *,
1589 const char *, int, int, int, __u32,
1590 struct ptlrpc_request **);
1592 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1593 struct obd_capa *, obd_valid, const char *,
1594 const char *, int, int, int,
1595 struct ptlrpc_request **);
1597 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1599 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1600 struct obd_export *, struct obd_export *,
1601 struct lustre_md *);
1603 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1605 int (*m_set_open_replay_data)(struct obd_export *,
1606 struct obd_client_handle *,
1607 struct ptlrpc_request *);
1608 int (*m_clear_open_replay_data)(struct obd_export *,
1609 struct obd_client_handle *);
1610 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1612 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1613 const struct lu_fid *, ldlm_type_t,
1614 ldlm_policy_data_t *, ldlm_mode_t,
1615 struct lustre_handle *);
1617 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1618 ldlm_policy_data_t *, ldlm_mode_t,
1619 ldlm_cancel_flags_t flags, void *opaque);
1620 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1621 renew_capa_cb_t cb);
1622 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1623 const struct req_msg_field *, struct obd_capa **);
1625 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1626 struct obd_capa *, __u32,
1627 struct ptlrpc_request **);
1629 int (*m_intent_getattr_async)(struct obd_export *,
1630 struct md_enqueue_info *,
1631 struct ldlm_enqueue_info *);
1633 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1634 struct lu_fid *, __u64 *bits);
1637 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1638 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1639 * wrapper function in include/linux/obd_class.h.
1643 struct lsm_operations {
1644 void (*lsm_free)(struct lov_stripe_md *);
1645 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1646 struct obd_export *md_exp);
1647 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1649 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1651 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1652 __u16 *stripe_count);
1653 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1654 struct lov_mds_md *lmm);
1657 extern const struct lsm_operations lsm_v1_ops;
1658 extern const struct lsm_operations lsm_v3_ops;
1659 static inline const struct lsm_operations *lsm_op_find(int magic)
1667 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1672 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1674 /* Requests for obd_extent_calc() */
1675 #define OBD_CALC_STRIPE_START 1
1676 #define OBD_CALC_STRIPE_END 2
1678 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1679 struct obd_export *exp, int error)
1682 CERROR("%s: transno "LPU64" commit error: %d\n",
1683 obd->obd_name, transno, error);
1686 if (exp && transno > exp->exp_last_committed) {
1687 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1688 obd->obd_name, transno);
1689 exp->exp_last_committed = transno;
1690 ptlrpc_commit_replies(exp);
1692 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1693 obd->obd_name, transno);
1695 if (transno > obd->obd_last_committed)
1696 obd->obd_last_committed = transno;
1699 static inline void init_obd_quota_ops(quota_interface_t *interface,
1700 struct obd_ops *obd_ops)
1706 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1707 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1708 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1711 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1713 return oinfo->oi_capa;
1716 static inline struct md_open_data *obd_mod_alloc(void)
1718 struct md_open_data *mod;
1722 cfs_atomic_set(&mod->mod_refcount, 1);
1726 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1727 #define obd_mod_put(mod) \
1729 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1730 if ((mod)->mod_open_req) \
1731 ptlrpc_req_finished((mod)->mod_open_req); \
1732 OBD_FREE_PTR(mod); \
1736 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1737 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1739 /* return 1 if client should be resend request */
1740 static inline int client_should_resend(int resend, struct client_obd *cli)
1742 return cfs_atomic_read(&cli->cl_resends) ?
1743 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1747 * Return device name for this device
1749 * XXX: lu_device is declared before obd_device, while a pointer pointing
1750 * back to obd_device in lu_device, so this helper function defines here
1751 * instead of in lu_object.h
1753 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1755 return lu_dev->ld_obd->obd_name;
1758 #endif /* __OBD_H */