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_spinlock_t lsm_lock;
106 pid_t lsm_lock_owner; /* debugging */
108 /* maximum possible file size, might change as OSTs status changes,
109 * e.g. disconnected, deactivated */
112 /* Public members. */
113 __u64 lw_object_id; /* lov object id */
114 __u64 lw_object_seq; /* lov object seq */
116 /* LOV-private members start here -- only for use in lov/. */
118 __u32 lw_stripe_size; /* size of the stripe */
119 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
120 __u16 lw_stripe_count; /* number of objects being striped over */
121 __u16 lw_layout_gen; /* generation of the layout */
122 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
125 struct lov_oinfo *lsm_oinfo[0];
128 #define lsm_object_id lsm_wire.lw_object_id
129 #define lsm_object_seq lsm_wire.lw_object_seq
130 #define lsm_magic lsm_wire.lw_magic
131 #define lsm_layout_gen lsm_wire.lw_layout_gen
132 #define lsm_stripe_size lsm_wire.lw_stripe_size
133 #define lsm_pattern lsm_wire.lw_pattern
134 #define lsm_stripe_count lsm_wire.lw_stripe_count
135 #define lsm_pool_name lsm_wire.lw_pool_name
139 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
141 /* obd info for a particular level (lov, osc). */
143 /* Lock policy. It keeps an extent which is specific for a particular
144 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
145 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
146 ldlm_policy_data_t oi_policy;
147 /* Flags used for set request specific flags:
148 - while lock handling, the flags obtained on the enqueue
149 request are set here.
150 - while stats, the flags used for control delay/resend.
151 - while setattr, the flags used for distinguish punch operation
154 /* Lock handle specific for every OSC lock. */
155 struct lustre_handle *oi_lockh;
156 /* lsm data specific for every OSC. */
157 struct lov_stripe_md *oi_md;
158 /* obdo data specific for every OSC, if needed at all. */
160 /* statfs data specific for every OSC, if needed at all. */
161 struct obd_statfs *oi_osfs;
162 /* An update callback which is called to update some data on upper
163 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
164 * request in osc level for enqueue requests. It is also possible to
165 * update some caller data from LOV layer if needed. */
166 obd_enqueue_update_f oi_cb_up;
167 /* oss capability, its type is obd_capa in client to avoid copy.
168 * in contrary its type is lustre_capa in OSS. */
170 /* transfer jobid from ost_sync() to filter_sync()... */
174 /* compare all relevant fields. */
175 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
176 struct lov_stripe_md *m2)
179 * ->lsm_wire contains padding, but it should be zeroed out during
182 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
185 static inline int lov_lum_lsm_cmp(struct lov_user_md *lum,
186 struct lov_stripe_md *lsm)
188 if (lsm->lsm_magic != lum->lmm_magic)
190 if ((lsm->lsm_stripe_count != 0) && (lum->lmm_stripe_count != 0) &&
191 (lsm->lsm_stripe_count != lum->lmm_stripe_count))
193 if ((lsm->lsm_stripe_size != 0) && (lum->lmm_stripe_size != 0) &&
194 (lsm->lsm_stripe_size != lum->lmm_stripe_size))
196 if ((lsm->lsm_pattern != 0) && (lum->lmm_pattern != 0) &&
197 (lsm->lsm_pattern != lum->lmm_pattern))
199 if ((lsm->lsm_magic == LOV_MAGIC_V3) &&
200 (strncmp(lsm->lsm_pool_name,
201 ((struct lov_user_md_v3 *)lum)->lmm_pool_name,
202 LOV_MAXPOOLNAME) != 0))
207 static inline int lov_lum_swab_if_needed(struct lov_user_md_v3 *lumv3,
209 struct lov_user_md *lum)
211 if (lum && cfs_copy_from_user(lumv3, lum,sizeof(struct lov_user_md_v1)))
214 *lmm_magic = lumv3->lmm_magic;
216 if (*lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
217 lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lumv3);
218 *lmm_magic = LOV_USER_MAGIC_V1;
219 } else if (*lmm_magic == LOV_USER_MAGIC_V3) {
220 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
222 } else if (*lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
223 if (lum && cfs_copy_from_user(lumv3, lum, sizeof(*lumv3)))
225 lustre_swab_lov_user_md_v3(lumv3);
226 *lmm_magic = LOV_USER_MAGIC_V3;
227 } else if (*lmm_magic != LOV_USER_MAGIC_V1) {
229 "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
230 *lmm_magic, LOV_USER_MAGIC_V1, LOV_USER_MAGIC_V3);
236 void lov_stripe_lock(struct lov_stripe_md *md);
237 void lov_stripe_unlock(struct lov_stripe_md *md);
240 cfs_list_t typ_chain;
241 struct obd_ops *typ_dt_ops;
242 struct md_ops *typ_md_ops;
243 cfs_proc_dir_entry_t *typ_procroot;
246 struct lu_device_type *typ_lu;
247 cfs_spinlock_t obd_type_lock;
257 /* Individual type definitions */
259 struct ost_server_data;
261 struct osd_properties {
262 size_t osd_max_ea_size;
265 #define OBT_MAGIC 0xBDDECEAE
266 /* hold common fields for "target" device */
267 struct obd_device_target {
270 struct super_block *obt_sb;
271 /** last_rcvd file */
272 struct file *obt_rcvd_filp;
273 #ifdef HAVE_SERVER_SUPPORT
274 struct lu_target *obt_lut;
276 __u64 obt_mount_count;
277 cfs_semaphore_t obt_quotachecking;
278 struct lustre_quota_ctxt obt_qctxt;
279 lustre_quota_version_t obt_qfmt;
280 cfs_rw_semaphore_t obt_rwsem;
281 struct vfsmount *obt_vfsmnt;
282 struct file *obt_health_check_filp;
283 struct osd_properties obt_osd_properties;
284 struct obd_job_stats obt_jobstats;
289 LLOG_CONFIG_ORIG_CTXT = 0,
290 LLOG_CONFIG_REPL_CTXT,
291 LLOG_MDS_OST_ORIG_CTXT,
292 LLOG_MDS_OST_REPL_CTXT,
299 LLOG_LOVEA_ORIG_CTXT,
300 LLOG_LOVEA_REPL_CTXT,
301 LLOG_CHANGELOG_ORIG_CTXT, /**< changelog generation on mdd */
302 LLOG_CHANGELOG_REPL_CTXT, /**< changelog access on clients */
303 LLOG_CHANGELOG_USER_ORIG_CTXT, /**< for multiple changelog consumers */
307 #define FILTER_SUBDIR_COUNT 32 /* set to zero for no subdirs */
309 struct filter_subdirs {
310 cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
320 /* NB this field MUST be first */
321 struct obd_device_target fo_obt;
322 const char *fo_fstype;
325 cfs_dentry_t *fo_dentry_O;
326 cfs_dentry_t **fo_dentry_O_groups;
327 struct filter_subdirs *fo_dentry_O_sub;
328 cfs_mutex_t fo_init_lock; /* group initialization lock */
329 int fo_committed_group;
331 cfs_spinlock_t fo_objidlock; /* protect fo_lastobjid */
333 unsigned long fo_destroys_in_progress;
334 cfs_mutex_t fo_create_locks[FILTER_SUBDIR_COUNT];
336 cfs_list_t fo_export_list;
339 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
340 obd_size fo_tot_granted; /* all values in bytes */
341 obd_size fo_tot_pending;
342 int fo_tot_granted_clients;
344 obd_size fo_readcache_max_filesize;
345 cfs_spinlock_t fo_flags_lock;
346 unsigned int fo_read_cache:1, /**< enable read-only cache */
347 fo_writethrough_cache:1,/**< read cache writes */
348 fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
349 fo_raid_degraded:1;/**< RAID device degraded */
351 struct obd_import *fo_mdc_imp;
352 struct obd_uuid fo_mdc_uuid;
353 struct lustre_handle fo_mdc_conn;
354 struct file **fo_last_objid_files;
355 __u64 *fo_last_objids; /* last created objid for groups,
356 * protected by fo_objidlock */
358 cfs_mutex_t fo_alloc_lock;
360 cfs_atomic_t fo_r_in_flight;
361 cfs_atomic_t fo_w_in_flight;
364 * per-filter pool of kiobuf's allocated by filter_common_setup() and
365 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
366 * which ->fo_iobuf_count were allocated.
368 * This pool contains kiobuf used by
369 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
372 * Locking: none, each OST thread uses only one element, determined by
373 * its "ordinal number", ->t_id.
375 struct filter_iobuf **fo_iobuf_pool;
378 cfs_list_t fo_llog_list;
379 cfs_spinlock_t fo_llog_list_lock;
381 struct brw_stats fo_filter_stats;
382 struct lustre_quota_ctxt fo_quota_ctxt;
383 cfs_spinlock_t fo_quotacheck_lock;
384 cfs_atomic_t fo_quotachecking;
386 int fo_fmd_max_num; /* per exp filter_mod_data */
387 int fo_fmd_max_age; /* jiffies to fmd expiry */
388 unsigned long fo_syncjournal:1, /* sync journal on writes */
389 fo_sync_lock_cancel:2;/* sync on lock cancel */
393 cfs_rwlock_t fo_sptlrpc_lock;
394 struct sptlrpc_rule_set fo_sptlrpc_rset;
396 /* capability related */
397 unsigned int fo_fl_oss_capa;
398 cfs_list_t fo_capa_keys;
399 cfs_hlist_head_t *fo_capa_hash;
400 struct llog_commit_master *fo_lcm;
404 struct timeout_item {
405 enum timeout_event ti_event;
406 cfs_time_t ti_timeout;
409 cfs_list_t ti_obd_list;
413 #define OSC_MAX_RIF_DEFAULT 8
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) */
456 cfs_list_t cl_cache_waiters; /* waiting for cache/grant */
457 cfs_time_t cl_next_shrink_grant; /* jiffies */
458 cfs_list_t cl_grant_shrink_list; /* Timeout event list */
459 int cl_grant_shrink_interval; /* seconds */
461 /* keep track of objects that have lois that contain pages which
462 * have been queued for async brw. this lock also protects the
463 * lists of osc_client_pages that hang off of the loi */
465 * ->cl_loi_list_lock protects consistency of
466 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
467 * ->ap_completion() call-backs are executed under this lock. As we
468 * cannot guarantee that these call-backs never block on all platforms
469 * (as a matter of fact they do block on Mac OS X), type of
470 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
471 * and blocking mutex on Mac OS X. (Alternative is to make this lock
472 * blocking everywhere, but we don't want to slow down fast-path of
473 * our main platform.)
475 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
476 * with client_obd_list_{un,}lock() and
477 * client_obd_list_lock_{init,done}() functions.
479 * NB by Jinshan: though field names are still _loi_, but actually
480 * osc_object{}s are in the list.
482 client_obd_lock_t cl_loi_list_lock;
483 cfs_list_t cl_loi_ready_list;
484 cfs_list_t cl_loi_hp_ready_list;
485 cfs_list_t cl_loi_write_list;
486 cfs_list_t cl_loi_read_list;
489 /* just a sum of the loi/lop pending numbers to be exported by /proc */
490 int cl_pending_w_pages;
491 int cl_pending_r_pages;
492 int cl_max_pages_per_rpc;
493 int cl_max_rpcs_in_flight;
494 struct obd_histogram cl_read_rpc_hist;
495 struct obd_histogram cl_write_rpc_hist;
496 struct obd_histogram cl_read_page_hist;
497 struct obd_histogram cl_write_page_hist;
498 struct obd_histogram cl_read_offset_hist;
499 struct obd_histogram cl_write_offset_hist;
501 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
502 cfs_atomic_t cl_destroy_in_flight;
503 cfs_waitq_t cl_destroy_waitq;
505 struct mdc_rpc_lock *cl_rpc_lock;
506 struct mdc_rpc_lock *cl_close_lock;
507 struct osc_creator cl_oscc;
510 cfs_semaphore_t cl_mgc_sem;
511 struct vfsmount *cl_mgc_vfsmnt;
512 struct dentry *cl_mgc_configs_dir;
513 cfs_atomic_t cl_mgc_refcount;
514 struct obd_export *cl_mgc_mgsexp;
516 /* checksumming for data sent over the network */
517 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
518 /* supported checksum types that are worked out at connect time */
519 __u32 cl_supp_cksum_types;
520 /* checksum algorithm to be used */
521 cksum_type_t cl_cksum_type;
523 /* also protected by the poorly named _loi_list_lock lock above */
524 struct osc_async_rc cl_ar;
526 /* used by quotacheck */
527 int cl_qchk_stat; /* quotacheck stat of the peer */
529 /* sequence manager */
530 struct lu_client_seq *cl_seq;
532 cfs_atomic_t cl_resends; /* resend count */
534 /* ptlrpc work for writeback in ptlrpcd context */
535 void *cl_writeback_work;
537 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
539 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
542 struct obd_device_target mgs_obt;
543 struct ptlrpc_service *mgs_service;
544 struct vfsmount *mgs_vfsmnt;
545 struct super_block *mgs_sb;
546 struct dentry *mgs_configs_dir;
547 cfs_list_t mgs_fs_db_list;
548 cfs_mutex_t mgs_mutex;
549 cfs_proc_dir_entry_t *mgs_proc_live;
550 cfs_time_t mgs_start_time;
554 /* NB this field MUST be first */
555 struct obd_device_target mds_obt;
556 struct ptlrpc_service *mds_service;
557 struct ptlrpc_service *mds_setattr_service;
558 struct ptlrpc_service *mds_readpage_service;
560 int mds_max_cookiesize;
562 unsigned long mds_atime_diff;
563 struct ll_fid mds_rootfid;
564 cfs_dentry_t *mds_pending_dir;
565 cfs_dentry_t *mds_logs_dir;
566 cfs_dentry_t *mds_objects_dir;
567 struct llog_handle *mds_cfg_llh;
568 struct obd_device *mds_lov_obd;
569 struct obd_uuid mds_lov_uuid;
571 struct obd_export *mds_lov_exp;
572 struct lov_desc mds_lov_desc;
575 /* mark pages dirty for write. */
576 cfs_bitmap_t *mds_lov_page_dirty;
577 /* array for store pages with obd_id */
578 void **mds_lov_page_array;
579 /* file for store objid */
580 struct file *mds_lov_objid_filp;
581 __u32 mds_lov_objid_count;
582 __u32 mds_lov_objid_max_index;
583 __u32 mds_lov_objid_lastpage;
584 __u32 mds_lov_objid_lastidx;
587 struct lustre_quota_info mds_quota_info;
588 cfs_rw_semaphore_t mds_qonoff_sem;
589 unsigned long mds_fl_user_xattr:1,
591 mds_evict_ost_nids:1,
595 mds_fl_target:1; /* mds have one or
598 struct upcall_cache *mds_identity_cache;
600 /* for capability keys update */
601 struct lustre_capa_key *mds_capa_keys;
602 cfs_rw_semaphore_t mds_notify_lock;
606 extern __u32 mds_max_ost_index;
608 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
610 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
612 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
614 extern int mds_lov_init_objids(struct obd_device *obd);
615 extern void mds_lov_destroy_objids(struct obd_device *obd);
625 struct obd_device_target eo_obt;
627 cfs_spinlock_t eo_lock;
629 struct lustre_handle eo_nl_lock;
630 cfs_atomic_t eo_prep;
634 struct ptlrpc_service *ost_service;
635 struct ptlrpc_service *ost_create_service;
636 struct ptlrpc_service *ost_io_service;
637 cfs_mutex_t ost_health_mutex;
640 struct echo_client_obd {
641 struct obd_export *ec_exp; /* the local connection to osc/lov */
642 cfs_spinlock_t ec_lock;
643 cfs_list_t ec_objects;
650 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
651 cfs_list_t lqo_oss_list; /* link to lov_qos */
652 __u64 lqo_bavail; /* total bytes avail on OSS */
653 __u64 lqo_penalty; /* current penalty */
654 __u64 lqo_penalty_per_obj;/* penalty decrease every obj*/
655 time_t lqo_used; /* last used time, seconds */
656 __u32 lqo_ost_count; /* number of osts on this oss */
660 struct lov_qos_oss *ltq_oss; /* oss info */
661 __u64 ltq_penalty; /* current penalty */
662 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
663 __u64 ltq_weight; /* net weighting */
664 time_t ltq_used; /* last used time, seconds */
665 unsigned int ltq_usable:1; /* usable for striping */
668 /* Generic subset of OSTs */
670 __u32 *op_array; /* array of index of
672 unsigned int op_count; /* number of OSTs in the array */
673 unsigned int op_size; /* allocated size of lp_array */
674 cfs_rw_semaphore_t op_rw_sem; /* to protect ost_pool use */
677 /* Round-robin allocator data */
679 __u32 lqr_start_idx; /* start index of new inode */
680 __u32 lqr_offset_idx; /* aliasing for start_idx */
681 int lqr_start_count; /* reseed counter */
682 struct ost_pool lqr_pool; /* round-robin optimized list */
683 unsigned long lqr_dirty:1; /* recalc round-robin list */
686 /* allow statfs data caching for 1 second */
687 #define OBD_STATFS_CACHE_SECONDS 1
689 struct lov_statfs_data {
690 struct obd_info lsd_oi;
691 struct obd_statfs lsd_statfs;
693 /* Stripe placement optimization */
695 cfs_list_t lq_oss_list; /* list of OSSs that targets use */
696 cfs_rw_semaphore_t lq_rw_sem;
697 __u32 lq_active_oss_count;
698 unsigned int lq_prio_free; /* priority for free space */
699 unsigned int lq_threshold_rr;/* priority for rr */
700 struct lov_qos_rr lq_rr; /* round robin qos data */
701 unsigned long lq_dirty:1, /* recalc qos data */
702 lq_same_space:1,/* the ost's all have approx.
703 the same space avail */
704 lq_reset:1, /* zero current penalties */
705 lq_statfs_in_progress:1; /* statfs op in
707 /* qos statfs data */
708 struct lov_statfs_data *lq_statfs_data;
709 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
710 * requests completion */
713 struct lov_tgt_desc {
715 struct obd_uuid ltd_uuid;
716 struct obd_device *ltd_obd;
717 struct obd_export *ltd_exp;
718 struct ltd_qos ltd_qos; /* qos info per target */
720 __u32 ltd_index; /* index in lov_obd->tgts */
721 unsigned long ltd_active:1,/* is this target up for requests */
722 ltd_activate:1,/* should target be activated */
723 ltd_reap:1; /* should this target be deleted */
727 #define pool_tgt_size(_p) _p->pool_obds.op_size
728 #define pool_tgt_count(_p) _p->pool_obds.op_count
729 #define pool_tgt_array(_p) _p->pool_obds.op_array
730 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
731 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
734 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
735 struct ost_pool pool_obds; /* pool members */
736 cfs_atomic_t pool_refcount; /* pool ref. counter */
737 struct lov_qos_rr pool_rr; /* round robin qos */
738 cfs_hlist_node_t pool_hash; /* access by poolname */
739 cfs_list_t pool_list; /* serial access */
740 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
741 struct lov_obd *pool_lov; /* lov obd to which this
746 struct lov_desc desc;
747 struct lov_tgt_desc **lov_tgts; /* sparse array */
748 struct ost_pool lov_packed; /* all OSTs in a packed
750 cfs_mutex_t lov_lock;
751 struct obd_connect_data lov_ocd;
752 struct lov_qos lov_qos; /* qos info per lov */
753 cfs_atomic_t lov_refcount;
754 __u32 lov_tgt_count; /* how many OBD's */
755 __u32 lov_active_tgt_count; /* how many active */
756 __u32 lov_death_row;/* tgts scheduled to be deleted */
757 __u32 lov_tgt_size; /* size of tgts array */
760 cfs_hash_t *lov_pools_hash_body; /* used for key access */
761 cfs_list_t lov_pool_list; /* used for sequential access */
762 cfs_proc_dir_entry_t *lov_pool_proc_entry;
763 enum lustre_sec_part lov_sp_me;
766 struct lmv_tgt_desc {
767 struct obd_uuid ltd_uuid;
768 struct obd_export *ltd_exp;
769 int ltd_active; /* is this target up for requests */
771 cfs_mutex_t ltd_fid_mutex;
774 enum placement_policy {
775 PLACEMENT_CHAR_POLICY = 0,
776 PLACEMENT_NID_POLICY = 1,
777 PLACEMENT_INVAL_POLICY = 2,
781 typedef enum placement_policy placement_policy_t;
785 struct lu_client_fld lmv_fld;
786 cfs_spinlock_t lmv_lock;
787 placement_policy_t lmv_placement;
788 struct lmv_desc desc;
789 struct obd_uuid cluuid;
790 struct obd_export *exp;
797 cfs_mutex_t init_mutex;
799 struct lmv_tgt_desc *tgts;
802 struct obd_connect_data *datas;
805 struct obd_connect_data conn_data;
808 struct niobuf_local {
813 cfs_dentry_t *dentry;
818 #define LUSTRE_FLD_NAME "fld"
819 #define LUSTRE_SEQ_NAME "seq"
821 #define LUSTRE_CMM_NAME "cmm"
822 #define LUSTRE_MDD_NAME "mdd"
823 #define LUSTRE_OSD_NAME "osd-ldiskfs"
824 #define LUSTRE_VVP_NAME "vvp"
825 #define LUSTRE_LMV_NAME "lmv"
826 #define LUSTRE_CMM_MDC_NAME "cmm-mdc"
827 #define LUSTRE_SLP_NAME "slp"
829 /* obd device type names */
830 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
831 #define LUSTRE_MDS_NAME "mds"
832 #define LUSTRE_MDT_NAME "mdt"
833 #define LUSTRE_MDC_NAME "mdc"
834 #define LUSTRE_OSS_NAME "ost" /* FIXME change name to oss */
835 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost */
836 #define LUSTRE_OSC_NAME "osc"
837 #define LUSTRE_LOV_NAME "lov"
838 #define LUSTRE_MGS_NAME "mgs"
839 #define LUSTRE_MGC_NAME "mgc"
841 #define LUSTRE_CACHEOBD_NAME "cobd"
842 #define LUSTRE_ECHO_NAME "obdecho"
843 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
845 /* Constant obd names (post-rename) */
846 #define LUSTRE_MDS_OBDNAME "MDS"
847 #define LUSTRE_OSS_OBDNAME "OSS"
848 #define LUSTRE_MGS_OBDNAME "MGS"
849 #define LUSTRE_MGC_OBDNAME "MGC"
851 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
852 #define N_LOCAL_TEMP_PAGE 0x10000000
854 struct obd_trans_info {
857 /* Only used on the server side for tracking acks. */
858 struct oti_req_ack_lock {
859 struct lustre_handle lock;
863 struct llog_cookie oti_onecookie;
864 struct llog_cookie *oti_logcookies;
867 /* initial thread handling transaction */
868 struct ptlrpc_thread * oti_thread;
871 __u64 oti_pre_version;
875 struct obd_uuid *oti_ost_uuid;
878 static inline void oti_init(struct obd_trans_info *oti,
879 struct ptlrpc_request *req)
883 memset(oti, 0, sizeof(*oti));
888 oti->oti_xid = req->rq_xid;
889 /** VBR: take versions from request */
890 if (req->rq_reqmsg != NULL &&
891 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
892 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
893 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
894 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
897 /** called from mds_create_objects */
898 if (req->rq_repmsg != NULL)
899 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
900 oti->oti_thread = req->rq_svc_thread;
901 if (req->rq_reqmsg != NULL)
902 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
905 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
910 if (num_cookies == 1)
911 oti->oti_logcookies = &oti->oti_onecookie;
913 OBD_ALLOC_LARGE(oti->oti_logcookies,
914 num_cookies * sizeof(oti->oti_onecookie));
916 oti->oti_numcookies = num_cookies;
919 static inline void oti_free_cookies(struct obd_trans_info *oti)
921 if (!oti || !oti->oti_logcookies)
924 if (oti->oti_logcookies == &oti->oti_onecookie)
925 LASSERT(oti->oti_numcookies == 1);
927 OBD_FREE_LARGE(oti->oti_logcookies,
928 oti->oti_numcookies*sizeof(oti->oti_onecookie));
929 oti->oti_logcookies = NULL;
930 oti->oti_numcookies = 0;
934 * Events signalled through obd_notify() upcall-chain.
936 enum obd_notify_event {
939 /* Device connect start */
941 /* Device activated */
943 /* Device deactivated */
945 /* Device disconnected */
947 /* Connect data for import were changed */
950 OBD_NOTIFY_SYNC_NONBLOCK,
952 /* Configuration event */
954 /* Trigger quota recovery */
956 /* Administratively deactivate/activate event */
957 OBD_NOTIFY_DEACTIVATE,
961 /* bit-mask flags for config events */
963 CONFIG_LOG = 0x1, /* finished processing config log */
964 CONFIG_SYNC = 0x2, /* mdt synced 1 ost */
965 CONFIG_TARGET = 0x4 /* one target is added */
969 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
970 * and liblustre being main examples).
972 struct obd_notify_upcall {
973 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
974 enum obd_notify_event ev, void *owner, void *data);
975 /* Opaque datum supplied by upper layer listener */
979 struct target_recovery_data {
980 svc_handler_t trd_recovery_handler;
981 pid_t trd_processing_task;
982 cfs_completion_t trd_starting;
983 cfs_completion_t trd_finishing;
987 * In HEAD for CMD, the object is created in group number which is 3>=
988 * or indexing starts from 3. To test this assertions are added to disallow
989 * group 0. But to run 2.0 mds server on 1.8.x disk format (i.e. interop_mode)
990 * object in group 0 needs to be allowed.
991 * So for interop mode following changes needs to be done:
992 * 1. No need to assert on group 0 or allow group 0
993 * 2. The group number indexing starts from 0 instead of 3
996 #define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
998 static inline __u64 objseq_to_mdsno(obd_seq seq)
1000 LASSERT_SEQ_IS_MDT(seq);
1001 if (seq == FID_SEQ_OST_MDT0)
1003 return seq - FID_SEQ_OST_MDT1 + 1;
1006 static inline int mdt_to_obd_objseq(int mdtid)
1009 * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
1013 return FID_SEQ_OST_MDT1 + mdtid - 1;
1017 struct obd_llog_group {
1018 cfs_list_t olg_list;
1020 struct llog_ctxt *olg_ctxts[LLOG_MAX_CTXTS];
1021 cfs_waitq_t olg_waitq;
1022 cfs_spinlock_t olg_lock;
1023 struct obd_export *olg_exp;
1024 int olg_initializing;
1025 cfs_mutex_t olg_cat_processing;
1028 /* corresponds to one of the obd's */
1029 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
1030 #define OBD_DEV_BY_DEVNAME 0xffffd0de
1033 struct obd_type *obd_type;
1036 /* common and UUID name of this device */
1037 char obd_name[MAX_OBD_NAME];
1038 struct obd_uuid obd_uuid;
1040 struct lu_device *obd_lu_dev;
1043 /* bitfield modification is protected by obd_dev_lock */
1044 unsigned long obd_attached:1, /* finished attach */
1045 obd_set_up:1, /* finished setup */
1046 obd_recovering:1, /* there are recoverable clients */
1047 obd_abort_recovery:1,/* recovery expired */
1048 obd_version_recov:1, /* obd uses version checking */
1049 obd_replayable:1, /* recovery is enabled; inform clients */
1050 obd_no_transno:1, /* no committed-transno notification */
1051 obd_no_recov:1, /* fail instead of retry messages */
1052 obd_stopping:1, /* started cleanup */
1053 obd_starting:1, /* started setup */
1054 obd_force:1, /* cleanup with > 0 obd refcount */
1055 obd_fail:1, /* cleanup with failover */
1056 obd_async_recov:1, /* allow asynchronous orphan cleanup */
1057 obd_no_conn:1, /* deny new connections */
1058 obd_inactive:1, /* device active/inactive
1059 * (for /proc/status only!!) */
1060 obd_no_ir:1, /* no imperative recovery. */
1061 obd_process_conf:1; /* device is processing mgs config */
1062 /* use separate field as it is set in interrupt to don't mess with
1063 * protection of other bits using _bh lock */
1064 unsigned long obd_recovery_expired:1;
1065 /* uuid-export hash body */
1066 cfs_hash_t *obd_uuid_hash;
1067 /* nid-export hash body */
1068 cfs_hash_t *obd_nid_hash;
1069 /* nid stats body */
1070 cfs_hash_t *obd_nid_stats_hash;
1071 cfs_list_t obd_nid_stats;
1072 cfs_atomic_t obd_refcount;
1073 cfs_waitq_t obd_refcount_waitq;
1074 cfs_list_t obd_exports;
1075 cfs_list_t obd_unlinked_exports;
1076 cfs_list_t obd_delayed_exports;
1077 int obd_num_exports;
1078 cfs_spinlock_t obd_nid_lock;
1079 struct ldlm_namespace *obd_namespace;
1080 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1081 /* a spinlock is OK for what we do now, may need a semaphore later */
1082 cfs_spinlock_t obd_dev_lock; /* protects obd bitfield above */
1083 cfs_mutex_t obd_dev_mutex;
1084 __u64 obd_last_committed;
1085 struct fsfilt_operations *obd_fsops;
1086 cfs_spinlock_t obd_osfs_lock;
1087 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1089 struct lvfs_run_ctxt obd_lvfs_ctxt;
1090 struct obd_llog_group obd_olg; /* default llog group */
1091 struct obd_device *obd_observer;
1092 cfs_rw_semaphore_t obd_observer_link_sem;
1093 struct obd_notify_upcall obd_upcall;
1094 struct obd_export *obd_self_export;
1095 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1096 cfs_list_t obd_exports_timed;
1097 time_t obd_eviction_timer; /* for ping evictor */
1099 int obd_max_recoverable_clients;
1100 cfs_atomic_t obd_connected_clients;
1101 int obd_stale_clients;
1102 int obd_delayed_clients;
1103 /* this lock protects all recovery list_heads, timer and
1104 * obd_next_recovery_transno value */
1105 cfs_spinlock_t obd_recovery_task_lock;
1106 __u64 obd_next_recovery_transno;
1107 int obd_replayed_requests;
1108 int obd_requests_queued_for_recovery;
1109 cfs_waitq_t obd_next_transno_waitq;
1110 /* protected by obd_recovery_task_lock */
1111 cfs_timer_t obd_recovery_timer;
1112 time_t obd_recovery_start; /* seconds */
1113 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1114 int obd_recovery_time_hard;
1115 int obd_recovery_timeout;
1116 int obd_recovery_ir_factor;
1118 /* new recovery stuff from CMD2 */
1119 struct target_recovery_data obd_recovery_data;
1120 int obd_replayed_locks;
1121 cfs_atomic_t obd_req_replay_clients;
1122 cfs_atomic_t obd_lock_replay_clients;
1123 /* all lists are protected by obd_recovery_task_lock */
1124 cfs_list_t obd_req_replay_queue;
1125 cfs_list_t obd_lock_replay_queue;
1126 cfs_list_t obd_final_req_queue;
1127 int obd_recovery_stage;
1130 struct obd_device_target obt;
1131 struct filter_obd filter;
1133 struct client_obd cli;
1135 struct echo_client_obd echo_client;
1136 struct echo_obd echo;
1141 /* Fields used by LProcFS */
1142 unsigned int obd_cntr_base;
1143 struct lprocfs_stats *obd_stats;
1145 unsigned int md_cntr_base;
1146 struct lprocfs_stats *md_stats;
1148 cfs_proc_dir_entry_t *obd_proc_entry;
1149 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1150 cfs_proc_dir_entry_t *obd_svc_procroot;
1151 struct lprocfs_stats *obd_svc_stats;
1152 cfs_atomic_t obd_evict_inprogress;
1153 cfs_waitq_t obd_evict_inprogress_waitq;
1154 cfs_list_t obd_evict_list; /* protected with pet_lock */
1157 * Ldlm pool part. Save last calculated SLV and Limit.
1159 cfs_rwlock_t obd_pool_lock;
1164 * A list of outstanding class_incref()'s against this obd. For
1167 struct lu_ref obd_reference;
1170 #define OBD_LLOG_FL_SENDNOW 0x0001
1172 enum obd_cleanup_stage {
1173 /* Special case hack for MDS LOVs */
1175 /* can be directly mapped to .ldto_device_fini() */
1176 OBD_CLEANUP_EXPORTS,
1179 /* get/set_info keys */
1180 #define KEY_ASYNC "async"
1181 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1182 #define KEY_BLOCKSIZE "blocksize"
1183 #define KEY_CAPA_KEY "capa_key"
1184 #define KEY_CHANGELOG_CLEAR "changelog_clear"
1185 #define KEY_FID2PATH "fid2path"
1186 #define KEY_CHECKSUM "checksum"
1187 #define KEY_CLEAR_FS "clear_fs"
1188 #define KEY_CONN_DATA "conn_data"
1189 #define KEY_EVICT_BY_NID "evict_by_nid"
1190 #define KEY_FIEMAP "fiemap"
1191 #define KEY_FLUSH_CTX "flush_ctx"
1192 #define KEY_GRANT_SHRINK "grant_shrink"
1193 #define KEY_HSM_COPYTOOL_SEND "hsm_send"
1194 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1195 #define KEY_INIT_RECOV "initial_recov"
1196 #define KEY_INTERMDS "inter_mds"
1197 #define KEY_LAST_ID "last_id"
1198 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1199 #define KEY_LOVDESC "lovdesc"
1200 #define KEY_LOV_IDX "lov_idx"
1201 #define KEY_MAX_EASIZE "max_easize"
1202 #define KEY_MDS_CONN "mds_conn"
1203 #define KEY_MGSSEC "mgssec"
1204 #define KEY_NEXT_ID "next_id"
1205 #define KEY_READ_ONLY "read-only"
1206 #define KEY_REGISTER_TARGET "register_target"
1207 #define KEY_REVIMP_UPD "revimp_update"
1208 #define KEY_SET_FS "set_fs"
1209 #define KEY_TGT_COUNT "tgt_count"
1210 /* KEY_SET_INFO in lustre_idl.h */
1211 #define KEY_SPTLRPC_CONF "sptlrpc_conf"
1212 #define KEY_CONNECT_FLAG "connect_flags"
1213 #define KEY_SYNC_LOCK_CANCEL "sync_lock_cancel"
1217 /* /!\ must be coherent with include/linux/namei.h on patched kernel */
1218 #define IT_OPEN (1 << 0)
1219 #define IT_CREAT (1 << 1)
1220 #define IT_READDIR (1 << 2)
1221 #define IT_GETATTR (1 << 3)
1222 #define IT_LOOKUP (1 << 4)
1223 #define IT_UNLINK (1 << 5)
1224 #define IT_TRUNC (1 << 6)
1225 #define IT_GETXATTR (1 << 7)
1226 #define IT_EXEC (1 << 8)
1227 #define IT_PIN (1 << 9)
1228 #define IT_LAYOUT (1 << 10)
1230 static inline int it_to_lock_mode(struct lookup_intent *it)
1232 /* CREAT needs to be tested before open (both could be set) */
1233 if (it->it_op & IT_CREAT)
1235 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP |
1239 LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
1244 struct lu_fid op_fid1; /* operation fid1 (usualy parent) */
1245 struct lu_fid op_fid2; /* operation fid2 (usualy child) */
1246 struct lu_fid op_fid3; /* 2 extra fids to find conflicting */
1247 struct lu_fid op_fid4; /* to the operation locks. */
1248 mdsno_t op_mds; /* what mds server open will go to */
1249 struct lustre_handle op_handle;
1250 obd_time op_mod_time;
1251 const char *op_name;
1254 struct lmv_stripe_md *op_mea1;
1255 struct lmv_stripe_md *op_mea2;
1256 __u32 op_suppgids[2];
1262 /* iattr fields and blocks. */
1263 struct iattr op_attr;
1265 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
1266 unsigned int op_attr_flags;
1270 loff_t op_attr_blocks;
1272 /* Size-on-MDS epoch and flags. */
1277 struct obd_capa *op_capa1;
1278 struct obd_capa *op_capa2;
1280 /* Various operation flags. */
1283 /* Operation type */
1286 /* Used by readdir */
1291 struct md_enqueue_info;
1292 /* metadata stat-ahead */
1293 typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
1294 struct md_enqueue_info *minfo,
1297 struct md_enqueue_info {
1298 struct md_op_data mi_data;
1299 struct lookup_intent mi_it;
1300 struct lustre_handle mi_lockh;
1301 struct inode *mi_dir;
1302 md_enqueue_cb_t mi_cb;
1304 unsigned int mi_generation;
1308 cfs_module_t *o_owner;
1309 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1310 void *karg, void *uarg);
1311 int (*o_get_info)(const struct lu_env *env, struct obd_export *,
1312 __u32 keylen, void *key, __u32 *vallen, void *val,
1313 struct lov_stripe_md *lsm);
1314 int (*o_set_info_async)(const struct lu_env *, struct obd_export *,
1315 __u32 keylen, void *key,
1316 __u32 vallen, void *val,
1317 struct ptlrpc_request_set *set);
1318 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1319 int (*o_detach)(struct obd_device *dev);
1320 int (*o_setup) (struct obd_device *dev, struct lustre_cfg *cfg);
1321 int (*o_precleanup)(struct obd_device *dev,
1322 enum obd_cleanup_stage cleanup_stage);
1323 int (*o_cleanup)(struct obd_device *dev);
1324 int (*o_process_config)(struct obd_device *dev, obd_count len,
1326 int (*o_postrecov)(struct obd_device *dev);
1327 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1329 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1330 /* connect to the target device with given connection
1331 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1332 * granted by the target, which are guaranteed to be a subset of flags
1333 * asked for. If @ocd == NULL, use default parameters. */
1334 int (*o_connect)(const struct lu_env *env,
1335 struct obd_export **exp, struct obd_device *src,
1336 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1338 int (*o_reconnect)(const struct lu_env *env,
1339 struct obd_export *exp, struct obd_device *src,
1340 struct obd_uuid *cluuid,
1341 struct obd_connect_data *ocd,
1343 int (*o_disconnect)(struct obd_export *exp);
1345 /* Initialize/finalize fids infrastructure. */
1346 int (*o_fid_init)(struct obd_export *exp);
1347 int (*o_fid_fini)(struct obd_export *exp);
1349 /* Allocate new fid according to passed @hint. */
1350 int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
1351 struct md_op_data *op_data);
1354 * Object with @fid is getting deleted, we may want to do something
1357 int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
1359 int (*o_statfs)(const struct lu_env *, struct obd_export *exp,
1360 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
1361 int (*o_statfs_async)(struct obd_export *exp, struct obd_info *oinfo,
1362 __u64 max_age, struct ptlrpc_request_set *set);
1363 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1364 struct lov_stripe_md *mem_src);
1365 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1366 struct lov_mds_md *disk_src, int disk_len);
1367 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1369 /* FIXME: add fid capability support for create & destroy! */
1370 int (*o_precreate)(struct obd_export *exp);
1371 int (*o_create)(const struct lu_env *env, struct obd_export *exp,
1372 struct obdo *oa, struct lov_stripe_md **ea,
1373 struct obd_trans_info *oti);
1374 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1375 struct lov_stripe_md **ea,
1376 struct obd_trans_info *oti);
1377 int (*o_destroy)(const struct lu_env *env, struct obd_export *exp,
1378 struct obdo *oa, struct lov_stripe_md *ea,
1379 struct obd_trans_info *oti, struct obd_export *md_exp,
1381 int (*o_setattr)(const struct lu_env *, struct obd_export *exp,
1382 struct obd_info *oinfo, struct obd_trans_info *oti);
1383 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1384 struct obd_trans_info *oti,
1385 struct ptlrpc_request_set *rqset);
1386 int (*o_getattr)(const struct lu_env *env, struct obd_export *exp,
1387 struct obd_info *oinfo);
1388 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1389 struct ptlrpc_request_set *set);
1390 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1391 obd_count oa_bufs, struct brw_page *pgarr,
1392 struct obd_trans_info *oti);
1393 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1394 struct ost_lvb *lvb, int kms_only);
1395 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1396 obd_off size, int shrink);
1397 int (*o_punch)(const struct lu_env *, struct obd_export *exp,
1398 struct obd_info *oinfo, struct obd_trans_info *oti,
1399 struct ptlrpc_request_set *rqset);
1400 int (*o_sync)(const struct lu_env *env, struct obd_export *exp,
1401 struct obd_info *oinfo, obd_size start, obd_size end,
1402 struct ptlrpc_request_set *set);
1403 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1404 struct lov_stripe_md *src, obd_size start,
1405 obd_size end, struct obd_trans_info *oti);
1406 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1407 struct lustre_handle *srconn, struct lov_stripe_md *src,
1408 obd_size start, obd_size end, struct obd_trans_info *);
1409 int (*o_iterate)(struct lustre_handle *conn,
1410 int (*)(obd_id, obd_seq, void *),
1411 obd_id *startid, obd_seq seq, void *data);
1412 int (*o_preprw)(const struct lu_env *env, int cmd,
1413 struct obd_export *exp, struct obdo *oa, int objcount,
1414 struct obd_ioobj *obj, struct niobuf_remote *remote,
1415 int *nr_pages, struct niobuf_local *local,
1416 struct obd_trans_info *oti, struct lustre_capa *capa);
1417 int (*o_commitrw)(const struct lu_env *env, int cmd,
1418 struct obd_export *exp, struct obdo *oa,
1419 int objcount, struct obd_ioobj *obj,
1420 struct niobuf_remote *remote, int pages,
1421 struct niobuf_local *local,
1422 struct obd_trans_info *oti, int rc);
1423 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1424 struct ldlm_enqueue_info *einfo,
1425 struct ptlrpc_request_set *rqset);
1426 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1427 ldlm_iterator_t it, void *data);
1428 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1429 ldlm_iterator_t it, void *data);
1430 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1431 __u32 mode, struct lustre_handle *);
1432 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1433 ldlm_cancel_flags_t flags, void *opaque);
1434 int (*o_init_export)(struct obd_export *exp);
1435 int (*o_destroy_export)(struct obd_export *exp);
1436 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1437 int cmd, obd_off *);
1439 /* llog related obd_methods */
1440 int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
1441 struct obd_device *disk_obd, int *idx);
1442 int (*o_llog_finish)(struct obd_device *obd, int count);
1443 int (*o_llog_connect)(struct obd_export *, struct llogd_conn_body *);
1445 /* metadata-only methods */
1446 int (*o_pin)(struct obd_export *, const struct lu_fid *fid,
1447 struct obd_capa *, struct obd_client_handle *, int flag);
1448 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1450 int (*o_import_event)(struct obd_device *, struct obd_import *,
1451 enum obd_import_event);
1453 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1454 enum obd_notify_event ev, void *data);
1456 int (*o_health_check)(const struct lu_env *env, struct obd_device *);
1457 struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
1460 int (*o_quotacheck)(struct obd_device *, struct obd_export *,
1461 struct obd_quotactl *);
1462 int (*o_quotactl)(struct obd_device *, struct obd_export *,
1463 struct obd_quotactl *);
1464 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1465 struct quota_adjust_qunit *oqaq,
1466 struct lustre_quota_ctxt *qctxt,
1467 struct ptlrpc_request_set *rqset);
1470 int (*o_ping)(const struct lu_env *, struct obd_export *exp);
1473 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1474 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1475 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1477 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1479 void (*o_getref)(struct obd_device *obd);
1480 void (*o_putref)(struct obd_device *obd);
1482 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1483 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1484 * Also, add a wrapper function in include/linux/obd_class.h. */
1488 LUSTRE_OPC_MKDIR = (1 << 0),
1489 LUSTRE_OPC_SYMLINK = (1 << 1),
1490 LUSTRE_OPC_MKNOD = (1 << 2),
1491 LUSTRE_OPC_CREATE = (1 << 3),
1492 LUSTRE_OPC_ANY = (1 << 4)
1495 /* lmv structures */
1496 #define MEA_MAGIC_LAST_CHAR 0xb2221ca1
1497 #define MEA_MAGIC_ALL_CHARS 0xb222a11c
1498 #define MEA_MAGIC_HASH_SEGMENT 0xb222a11b
1500 #define MAX_HASH_SIZE_32 0x7fffffffUL
1501 #define MAX_HASH_SIZE 0x7fffffffffffffffULL
1502 #define MAX_HASH_HIGHEST_BIT 0x1000000000000000ULL
1505 struct mdt_body *body;
1506 struct lov_stripe_md *lsm;
1507 struct lmv_stripe_md *mea;
1508 #ifdef CONFIG_FS_POSIX_ACL
1509 struct posix_acl *posix_acl;
1511 struct mdt_remote_perm *remote_perm;
1512 struct obd_capa *mds_capa;
1513 struct obd_capa *oss_capa;
1516 struct md_open_data {
1517 struct obd_client_handle *mod_och;
1518 struct ptlrpc_request *mod_open_req;
1519 struct ptlrpc_request *mod_close_req;
1520 cfs_atomic_t mod_refcount;
1523 struct lookup_intent;
1526 int (*m_getstatus)(struct obd_export *, struct lu_fid *,
1527 struct obd_capa **);
1528 int (*m_change_cbdata)(struct obd_export *, const struct lu_fid *,
1529 ldlm_iterator_t, void *);
1530 int (*m_find_cbdata)(struct obd_export *, const struct lu_fid *,
1531 ldlm_iterator_t, void *);
1532 int (*m_close)(struct obd_export *, struct md_op_data *,
1533 struct md_open_data *, struct ptlrpc_request **);
1534 int (*m_create)(struct obd_export *, struct md_op_data *,
1535 const void *, int, int, __u32, __u32, cfs_cap_t,
1536 __u64, struct ptlrpc_request **);
1537 int (*m_done_writing)(struct obd_export *, struct md_op_data *,
1538 struct md_open_data *);
1539 int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
1540 struct lookup_intent *, struct md_op_data *,
1541 struct lustre_handle *, void *, int,
1542 struct ptlrpc_request **, int);
1543 int (*m_getattr)(struct obd_export *, struct md_op_data *,
1544 struct ptlrpc_request **);
1545 int (*m_getattr_name)(struct obd_export *, struct md_op_data *,
1546 struct ptlrpc_request **);
1547 int (*m_intent_lock)(struct obd_export *, struct md_op_data *,
1548 void *, int, struct lookup_intent *, int,
1549 struct ptlrpc_request **,
1550 ldlm_blocking_callback, int);
1551 int (*m_link)(struct obd_export *, struct md_op_data *,
1552 struct ptlrpc_request **);
1553 int (*m_rename)(struct obd_export *, struct md_op_data *,
1554 const char *, int, const char *, int,
1555 struct ptlrpc_request **);
1556 int (*m_is_subdir)(struct obd_export *, const struct lu_fid *,
1557 const struct lu_fid *,
1558 struct ptlrpc_request **);
1559 int (*m_setattr)(struct obd_export *, struct md_op_data *, void *,
1560 int , void *, int, struct ptlrpc_request **,
1561 struct md_open_data **mod);
1562 int (*m_sync)(struct obd_export *, const struct lu_fid *,
1563 struct obd_capa *, struct ptlrpc_request **);
1564 int (*m_readpage)(struct obd_export *, struct md_op_data *,
1565 struct page **, struct ptlrpc_request **);
1567 int (*m_unlink)(struct obd_export *, struct md_op_data *,
1568 struct ptlrpc_request **);
1570 int (*m_setxattr)(struct obd_export *, const struct lu_fid *,
1571 struct obd_capa *, obd_valid, const char *,
1572 const char *, int, int, int, __u32,
1573 struct ptlrpc_request **);
1575 int (*m_getxattr)(struct obd_export *, const struct lu_fid *,
1576 struct obd_capa *, obd_valid, const char *,
1577 const char *, int, int, int,
1578 struct ptlrpc_request **);
1580 int (*m_init_ea_size)(struct obd_export *, int, int, int);
1582 int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
1583 struct obd_export *, struct obd_export *,
1584 struct lustre_md *);
1586 int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
1588 int (*m_set_open_replay_data)(struct obd_export *,
1589 struct obd_client_handle *,
1590 struct ptlrpc_request *);
1591 int (*m_clear_open_replay_data)(struct obd_export *,
1592 struct obd_client_handle *);
1593 int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
1595 ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
1596 const struct lu_fid *, ldlm_type_t,
1597 ldlm_policy_data_t *, ldlm_mode_t,
1598 struct lustre_handle *);
1600 int (*m_cancel_unused)(struct obd_export *, const struct lu_fid *,
1601 ldlm_policy_data_t *, ldlm_mode_t,
1602 ldlm_cancel_flags_t flags, void *opaque);
1603 int (*m_renew_capa)(struct obd_export *, struct obd_capa *oc,
1604 renew_capa_cb_t cb);
1605 int (*m_unpack_capa)(struct obd_export *, struct ptlrpc_request *,
1606 const struct req_msg_field *, struct obd_capa **);
1608 int (*m_get_remote_perm)(struct obd_export *, const struct lu_fid *,
1609 struct obd_capa *, __u32,
1610 struct ptlrpc_request **);
1612 int (*m_intent_getattr_async)(struct obd_export *,
1613 struct md_enqueue_info *,
1614 struct ldlm_enqueue_info *);
1616 int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
1617 struct lu_fid *, __u64 *bits);
1620 * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
1621 * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
1622 * wrapper function in include/linux/obd_class.h.
1626 struct lsm_operations {
1627 void (*lsm_free)(struct lov_stripe_md *);
1628 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1629 struct obd_export *md_exp);
1630 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1632 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1634 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1635 __u16 *stripe_count);
1636 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1637 struct lov_mds_md *lmm);
1640 extern const struct lsm_operations lsm_v1_ops;
1641 extern const struct lsm_operations lsm_v3_ops;
1642 static inline const struct lsm_operations *lsm_op_find(int magic)
1650 CERROR("Cannot recognize lsm_magic %08x\n", magic);
1655 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1657 /* Requests for obd_extent_calc() */
1658 #define OBD_CALC_STRIPE_START 1
1659 #define OBD_CALC_STRIPE_END 2
1661 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1662 struct obd_export *exp, int error)
1665 CERROR("%s: transno "LPU64" commit error: %d\n",
1666 obd->obd_name, transno, error);
1669 if (exp && transno > exp->exp_last_committed) {
1670 CDEBUG(D_HA, "%s: transno "LPU64" committed\n",
1671 obd->obd_name, transno);
1672 exp->exp_last_committed = transno;
1673 ptlrpc_commit_replies(exp);
1675 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1676 obd->obd_name, transno);
1678 if (transno > obd->obd_last_committed)
1679 obd->obd_last_committed = transno;
1682 static inline void init_obd_quota_ops(quota_interface_t *interface,
1683 struct obd_ops *obd_ops)
1689 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1690 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1691 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1694 static inline struct lustre_capa *oinfo_capa(struct obd_info *oinfo)
1696 return oinfo->oi_capa;
1699 static inline struct md_open_data *obd_mod_alloc(void)
1701 struct md_open_data *mod;
1705 cfs_atomic_set(&mod->mod_refcount, 1);
1709 #define obd_mod_get(mod) cfs_atomic_inc(&(mod)->mod_refcount)
1710 #define obd_mod_put(mod) \
1712 if (cfs_atomic_dec_and_test(&(mod)->mod_refcount)) { \
1713 if ((mod)->mod_open_req) \
1714 ptlrpc_req_finished((mod)->mod_open_req); \
1715 OBD_FREE_PTR(mod); \
1719 void obdo_from_inode(struct obdo *dst, struct inode *src, obd_flag valid);
1720 void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent);
1722 /* return 1 if client should be resend request */
1723 static inline int client_should_resend(int resend, struct client_obd *cli)
1725 return cfs_atomic_read(&cli->cl_resends) ?
1726 cfs_atomic_read(&cli->cl_resends) > resend : 1;
1730 * Return device name for this device
1732 * XXX: lu_device is declared before obd_device, while a pointer pointing
1733 * back to obd_device in lu_device, so this helper function defines here
1734 * instead of in lu_object.h
1736 static inline const char *lu_dev_name(const struct lu_device *lu_dev)
1738 return lu_dev->ld_obd->obd_name;
1741 #endif /* __OBD_H */