1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
40 #if defined(__linux__)
41 #include <linux/obd.h>
42 #elif defined(__APPLE__)
43 #include <darwin/obd.h>
44 #elif defined(__WINNT__)
45 #include <winnt/obd.h>
47 #error Unsupported operating system.
50 #define IOC_OSC_TYPE 'h'
51 #define IOC_OSC_MIN_NR 20
52 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
53 #define IOC_OSC_MAX_NR 50
55 #define IOC_MDC_TYPE 'i'
56 #define IOC_MDC_MIN_NR 20
57 /* Moved to lustre_user.h
58 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_ioctl_data *)
59 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
60 #define IOC_MDC_MAX_NR 50
62 #include <lustre_lib.h>
63 #include <lustre/lustre_idl.h>
64 #include <lustre_export.h>
65 #include <lustre_quota.h>
66 #include <class_hash.h>
68 #include <libcfs/bitmap.h>
71 #define MAX_OBD_DEVICES 8192
73 /* this is really local to the OSC */
74 struct loi_oap_pages {
75 struct list_head lop_pending;
76 struct list_head lop_urgent;
77 struct list_head lop_pending_group;
87 struct lov_oinfo { /* per-stripe data structure */
88 __u64 loi_id; /* object ID on the target OST */
89 __u64 loi_gr; /* object group on the target OST */
90 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
91 int loi_ost_gen; /* generation of this loi_ost_idx */
93 /* used by the osc to keep track of what objects to build into rpcs */
94 struct loi_oap_pages loi_read_lop;
95 struct loi_oap_pages loi_write_lop;
96 struct list_head loi_ready_item;
97 struct list_head loi_hp_ready_item;
98 struct list_head loi_write_item;
99 struct list_head loi_read_item;
101 unsigned long loi_kms_valid:1;
102 __u64 loi_kms; /* known minimum size */
103 struct ost_lvb loi_lvb;
104 struct osc_async_rc loi_ar;
107 static inline void loi_init(struct lov_oinfo *loi)
109 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
110 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
111 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
112 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
113 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
114 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
115 CFS_INIT_LIST_HEAD(&loi->loi_ready_item);
116 CFS_INIT_LIST_HEAD(&loi->loi_hp_ready_item);
117 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
118 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
121 /*extent array item for describing the joined file extent info*/
123 __u64 le_start; /* extent start */
124 __u64 le_len; /* extent length */
125 int le_loi_idx; /* extent #1 loi's index in lsm loi array */
126 int le_stripe_count; /* extent stripe count*/
129 /*Lov array info for describing joined file array EA info*/
130 struct lov_array_info {
131 struct llog_logid lai_array_id; /* MDS med llog object id */
132 unsigned lai_ext_count; /* number of extent count */
133 struct lov_extent *lai_ext_array; /* extent desc array */
136 struct lov_stripe_md {
138 void *lsm_lock_owner; /* debugging */
141 /* Public members. */
142 __u64 lw_object_id; /* lov object id */
143 __u64 lw_object_gr; /* lov object group */
144 __u64 lw_maxbytes; /* maximum possible file size */
146 /* LOV-private members start here -- only for use in lov/. */
148 __u32 lw_stripe_size; /* size of the stripe */
149 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
150 unsigned lw_stripe_count; /* number of objects being striped over */
151 char lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
154 struct lov_array_info *lsm_array; /*Only for joined file array info*/
155 struct lov_oinfo *lsm_oinfo[0];
158 #define lsm_object_id lsm_wire.lw_object_id
159 #define lsm_object_gr lsm_wire.lw_object_gr
160 #define lsm_maxbytes lsm_wire.lw_maxbytes
161 #define lsm_magic lsm_wire.lw_magic
162 #define lsm_stripe_size lsm_wire.lw_stripe_size
163 #define lsm_pattern lsm_wire.lw_pattern
164 #define lsm_stripe_count lsm_wire.lw_stripe_count
165 #define lsm_pool_name lsm_wire.lw_pool_name
169 typedef int (*obd_enqueue_update_f)(struct obd_info *oinfo, int rc);
171 /* obd info for a particular level (lov, osc). */
173 /* Lock policy. It keeps an extent which is specific for a particular
174 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
175 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
176 ldlm_policy_data_t oi_policy;
177 /* Flags used for set request specific flags:
178 - while lock handling, the flags obtained on the enqueue
179 request are set here.
180 - while stats, the flags used for control delay/resend.
183 /* Lock handle specific for every OSC lock. */
184 struct lustre_handle *oi_lockh;
185 /* lsm data specific for every OSC. */
186 struct lov_stripe_md *oi_md;
187 /* obdo data specific for every OSC, if needed at all. */
189 /* statfs data specific for every OSC, if needed at all. */
190 struct obd_statfs *oi_osfs;
191 /* An update callback which is called to update some data on upper
192 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
193 * request in osc level for enqueue requests. It is also possible to
194 * update some caller data from LOV layer if needed. */
195 obd_enqueue_update_f oi_cb_up;
198 /* compare all relevant fields. */
199 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
200 struct lov_stripe_md *m2)
202 if ((m1->lsm_magic == LOV_MAGIC_V1 && m2->lsm_magic == LOV_MAGIC_V3) ||
203 (m1->lsm_magic == LOV_MAGIC_V3 && m2->lsm_magic == LOV_MAGIC_V1))
205 * Dowgrade/upgrade case (b=20318) to a version that
206 * doesn't/does OST pools. Compare only objid and stripe
207 * size/count in this case
209 return m1->lsm_object_id != m2->lsm_object_id ||
210 m1->lsm_stripe_size != m2->lsm_stripe_size ||
211 m1->lsm_stripe_count != m2->lsm_stripe_count;
214 * ->lsm_wire contains padding, but it should be zeroed out during
217 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof(m1->lsm_wire));
220 void lov_stripe_lock(struct lov_stripe_md *md);
221 void lov_stripe_unlock(struct lov_stripe_md *md);
224 struct list_head typ_chain;
225 struct obd_ops *typ_ops;
226 cfs_proc_dir_entry_t *typ_procroot;
229 spinlock_t obd_type_lock;
240 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
241 page is added to an rpc */
242 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
243 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
244 to give the caller a chance to update
245 or cancel the size of the io */
246 ASYNC_GROUP_SYNC = 0x8, /* ap_completion will not be called, instead
247 the page is accounted for in the
248 obd_io_group given to
249 obd_queue_group_io */
253 struct obd_async_page_ops {
254 int (*ap_make_ready)(void *data, int cmd);
255 int (*ap_refresh_count)(void *data, int cmd);
256 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
257 void (*ap_update_obdo)(void *data, int cmd, struct obdo *oa,
259 int (*ap_completion)(void *data, int cmd, struct obdo *oa, int rc);
262 /* the `oig' is passed down from a caller of obd rw methods. the callee
263 * records enough state such that the caller can sleep on the oig and
264 * be woken when all the callees have finished their work */
265 struct obd_io_group {
267 atomic_t oig_refcount;
270 struct list_head oig_occ_list;
271 cfs_waitq_t oig_waitq;
274 /* the oig callback context lets the callee of obd rw methods register
275 * for callbacks from the caller. */
276 struct oig_callback_context {
277 struct list_head occ_oig_item;
278 /* called when the caller has received a signal while sleeping.
279 * callees of this method are encouraged to abort their state
280 * in the oig. This may be called multiple times. */
281 void (*occ_interrupted)(struct oig_callback_context *occ);
282 unsigned long interrupted:1;
285 /* Individual type definitions */
287 struct ost_server_data;
289 /* hold common fields for "target" device */
290 struct obd_device_target {
291 struct super_block *obt_sb;
292 /** last_rcvd file */
293 struct file *obt_rcvd_filp;
294 /** server data in last_rcvd file */
295 struct lr_server_data *obt_lsd;
296 /** Lock protecting client bitmap */
297 spinlock_t obt_client_bitmap_lock;
298 /** Bitmap of known clients */
299 unsigned long *obt_client_bitmap;
300 /** Server last transaction number */
301 __u64 obt_last_transno;
302 /** Lock protecting last transaction number */
303 spinlock_t obt_translock;
304 /** Number of mounts */
305 __u64 obt_mount_count;
306 atomic_t obt_quotachecking;
307 struct lustre_quota_ctxt obt_qctxt;
308 lustre_quota_version_t obt_qfmt;
309 __u32 obt_stale_export_age;
310 spinlock_t obt_trans_table_lock;
311 struct vfsmount *obt_vfsmnt;
312 struct file *obt_health_check_filp;
315 typedef void (*obd_pin_extent_cb)(void *data);
316 typedef int (*obd_page_removal_cb_t)(void *data, int discard);
317 typedef int (*obd_lock_cancel_cb)(struct ldlm_lock *,struct ldlm_lock_desc *,
320 #define FILTER_GROUP_LLOG 1
321 #define FILTER_GROUP_ECHO 2
329 /* NB this field MUST be first */
330 struct obd_device_target fo_obt;
331 const char *fo_fstype;
332 cfs_dentry_t *fo_dentry_O;
333 cfs_dentry_t **fo_dentry_O_groups;
334 cfs_dentry_t **fo_dentry_O_sub;
335 spinlock_t fo_objidlock; /* protect fo_lastobjid */
337 int fo_destroy_in_progress;
338 struct semaphore fo_create_lock;
340 struct list_head fo_export_list;
343 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
344 obd_size fo_tot_granted; /* all values in bytes */
345 obd_size fo_tot_pending;
346 int fo_tot_granted_clients;
348 obd_size fo_readcache_max_filesize;
349 unsigned long fo_read_cache:1, /* read-only cache */
350 fo_writethrough_cache:1, /* writetrhough cache */
351 fo_syncjournal:1, /* sync journal on writes */
352 fo_raid_degraded:1; /* RAID device degraded */
354 struct obd_import *fo_mdc_imp;
355 struct obd_uuid fo_mdc_uuid;
356 struct lustre_handle fo_mdc_conn;
357 struct file **fo_last_objid_files;
358 __u64 *fo_last_objids; /* last created objid for groups,
359 * protected by fo_objidlock */
361 struct semaphore fo_alloc_lock;
363 atomic_t fo_r_in_flight;
364 atomic_t fo_w_in_flight;
367 * per-filter pool of kiobuf's allocated by filter_common_setup() and
368 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
369 * which ->fo_iobuf_count were allocated.
371 * This pool contains kiobuf used by
372 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
375 * Locking: none, each OST thread uses only one element, determined by
376 * its "ordinal number", ->t_id.
378 struct filter_iobuf **fo_iobuf_pool;
381 struct brw_stats fo_filter_stats;
382 struct lustre_quota_ctxt fo_quota_ctxt;
383 spinlock_t fo_quotacheck_lock;
384 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 struct llog_commit_master *fo_lcm;
391 #define fo_translock fo_obt.obt_translock
392 #define fo_rcvd_filp fo_obt.obt_rcvd_filp
393 #define fo_fsd fo_obt.obt_lsd
394 #define fo_last_rcvd_slots fo_obt.obt_client_bitmap
395 #define fo_mount_count fo_obt.obt_mount_count
396 #define fo_vfsmnt fo_obt.obt_vfsmnt
398 #define OSC_MAX_RIF_DEFAULT 8
399 #define OSC_MAX_RIF_MAX 256
400 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
401 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
402 #define OSC_DEFAULT_RESENDS 10
404 #define MDC_MAX_RIF_DEFAULT 8
405 #define MDC_MAX_RIF_MAX 512
412 struct timeout_item {
413 enum timeout_event ti_event;
414 cfs_time_t ti_timeout;
417 struct list_head ti_obd_list;
418 struct list_head ti_chain;
421 struct lu_client_seq {
422 /* Sequence-controller export. */
423 struct obd_export *lcs_exp;
424 struct semaphore lcs_sem;
427 * Range of allowed for allocation sequences. When using lu_client_seq
428 * on clients, this contains meta-sequence range. And for servers this
429 * contains super-sequence range.
431 struct lu_seq_range lcs_space;
433 /* This holds last allocated fid in last obtained seq */
434 struct lu_fid lcs_fid;
436 /* LUSTRE_SEQ_METADATA or LUSTRE_SEQ_DATA */
437 enum lu_cli_type lcs_type;
439 * Service uuid, passed from MDT + seq name to form unique seq name to
440 * use it with procfs.
445 * Sequence width, that is how many objects may be allocated in one
446 * sequence. Default value for it is LUSTRE_SEQ_MAX_WIDTH.
453 struct rw_semaphore cl_sem;
454 struct obd_uuid cl_target_uuid;
455 struct obd_import *cl_import; /* ptlrpc connection state */
457 /* max_mds_easize is purely a performance thing so we don't have to
458 * call obd_size_diskmd() all the time. */
459 unsigned cl_default_mds_easize;
460 unsigned cl_max_mds_easize;
461 unsigned cl_max_mds_cookiesize;
463 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
464 void *cl_llcd_offset;
466 /* the grant values are protected by loi_list_lock below */
467 long cl_dirty; /* all _dirty_ in bytes */
468 long cl_dirty_max; /* allowed w/o rpc */
469 long cl_avail_grant; /* bytes of credit for ost */
470 long cl_lost_grant; /* lost credits (trunc) */
471 struct list_head cl_cache_waiters; /* waiting for cache/grant */
472 cfs_time_t cl_next_shrink_grant; /* jiffies */
473 struct list_head cl_grant_shrink_list; /* Timeout event list */
474 struct semaphore cl_grant_sem; /*grant shrink list semaphore*/
475 int cl_grant_shrink_interval; /* seconds */
477 /* keep track of objects that have lois that contain pages which
478 * have been queued for async brw. this lock also protects the
479 * lists of osc_client_pages that hang off of the loi */
481 * ->cl_loi_list_lock protects consistency of
482 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
483 * ->ap_completion() call-backs are executed under this lock. As we
484 * cannot guarantee that these call-backs never block on all platforms
485 * (as a matter of fact they do block on Mac OS X), type of
486 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
487 * and blocking mutex on Mac OS X. (Alternative is to make this lock
488 * blocking everywhere, but we don't want to slow down fast-path of
489 * our main platform.)
491 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
492 * with client_obd_list_{un,}lock() and
493 * client_obd_list_lock_{init,done}() functions.
495 client_obd_lock_t cl_loi_list_lock;
496 struct list_head cl_loi_ready_list;
497 struct list_head cl_loi_hp_ready_list;
498 struct list_head cl_loi_write_list;
499 struct list_head cl_loi_read_list;
502 /* just a sum of the loi/lop pending numbers to be exported by /proc */
503 int cl_pending_w_pages;
504 int cl_pending_r_pages;
505 int cl_max_pages_per_rpc;
506 int cl_max_rpcs_in_flight;
507 struct obd_histogram cl_read_rpc_hist;
508 struct obd_histogram cl_write_rpc_hist;
509 struct obd_histogram cl_read_page_hist;
510 struct obd_histogram cl_write_page_hist;
511 struct obd_histogram cl_read_offset_hist;
512 struct obd_histogram cl_write_offset_hist;
514 /* number of in flight destroy rpcs is limited to max_rpcs_in_flight */
515 atomic_t cl_destroy_in_flight;
516 cfs_waitq_t cl_destroy_waitq;
518 struct mdc_rpc_lock *cl_rpc_lock;
519 struct mdc_rpc_lock *cl_setattr_lock;
520 struct mdc_rpc_lock *cl_close_lock;
521 struct osc_creator cl_oscc;
524 struct semaphore cl_mgc_sem;
525 struct vfsmount *cl_mgc_vfsmnt;
526 struct dentry *cl_mgc_configs_dir;
527 atomic_t cl_mgc_refcount;
528 struct obd_export *cl_mgc_mgsexp;
530 /* checksumming for data sent over the network */
531 unsigned int cl_checksum:1; /* 0 = disabled, 1 = enabled */
532 /* supported checksum types that are worked out at connect time */
533 __u32 cl_supp_cksum_types;
534 /* checksum algorithm to be used */
535 cksum_type_t cl_cksum_type;
537 /* also protected by the poorly named _loi_list_lock lock above */
538 struct osc_async_rc cl_ar;
540 /* used by quotacheck */
541 int cl_qchk_stat; /* quotacheck stat of the peer */
543 /* sequence manager */
544 struct lu_client_seq *cl_seq;
546 atomic_t cl_resends; /* resend count */
547 /* Cache of triples */
548 struct lustre_cache *cl_cache;
549 obd_lock_cancel_cb cl_ext_lock_cancel_cb;
551 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
553 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
556 struct ptlrpc_service *mgs_service;
557 struct vfsmount *mgs_vfsmnt;
558 struct super_block *mgs_sb;
559 struct dentry *mgs_configs_dir;
560 struct dentry *mgs_fid_de;
561 struct list_head mgs_fs_db_list;
562 struct semaphore mgs_sem;
563 cfs_proc_dir_entry_t *mgs_proc_live;
567 /* NB this field MUST be first */
568 struct obd_device_target mds_obt;
569 struct ptlrpc_service *mds_service;
570 struct ptlrpc_service *mds_setattr_service;
571 struct ptlrpc_service *mds_readpage_service;
572 cfs_dentry_t *mds_fid_de;
574 int mds_max_cookiesize;
576 unsigned long mds_atime_diff;
577 struct semaphore mds_epoch_sem;
578 struct ll_fid mds_rootfid;
579 cfs_dentry_t *mds_pending_dir;
580 cfs_dentry_t *mds_logs_dir;
581 cfs_dentry_t *mds_objects_dir;
582 struct llog_handle *mds_cfg_llh;
583 // struct llog_handle *mds_catalog;
584 struct obd_device *mds_lov_obd;
585 struct obd_uuid mds_lov_uuid;
587 struct obd_export *mds_lov_exp;
588 struct lov_desc mds_lov_desc;
590 /* mark pages dirty for write. */
591 bitmap_t *mds_lov_page_dirty;
592 /* array for store pages with obd_id */
593 void **mds_lov_page_array;
594 /* file for store objid */
595 struct file *mds_lov_objid_filp;
596 __u32 mds_lov_objid_count;
597 __u32 mds_lov_objid_max_index;
598 __u32 mds_lov_objid_lastpage;
599 __u32 mds_lov_objid_lastidx;
601 struct upcall_cache *mds_group_hash;
603 struct lustre_quota_info mds_quota_info;
604 struct semaphore mds_qonoff_sem;
605 struct semaphore mds_health_sem;
606 unsigned long mds_fl_user_xattr:1,
610 mds_fl_target:1, /* mds have one or
612 mds_evict_ost_nids:1;
614 uid_t mds_squash_uid;
615 gid_t mds_squash_gid;
616 lnet_nid_t mds_nosquash_nid;
617 /* do we need permission sync */
618 unsigned int mds_sync_permission;
621 #define mds_transno_lock mds_obt.obt_translock
622 #define mds_rcvd_filp mds_obt.obt_rcvd_filp
623 #define mds_server_data mds_obt.obt_lsd
624 #define mds_client_bitmap mds_obt.obt_client_bitmap
625 #define mds_mount_count mds_obt.obt_mount_count
626 #define mds_last_transno mds_obt.obt_last_transno
627 #define mds_vfsmnt mds_obt.obt_vfsmnt
630 #define mds_max_ost_index (0xFFFF)
631 #define MDS_LOV_ALLOC_SIZE (CFS_PAGE_SIZE)
633 #define OBJID_PER_PAGE() (MDS_LOV_ALLOC_SIZE / sizeof(obd_id))
635 #define MDS_LOV_OBJID_PAGES_COUNT (mds_max_ost_index/OBJID_PER_PAGE())
637 extern int mds_lov_init_objids(struct obd_device *obd);
638 extern void mds_lov_destroy_objids(struct obd_device *obd);
651 struct lustre_handle eo_nl_lock;
656 struct ptlrpc_service *ost_service;
657 struct ptlrpc_service *ost_create_service;
658 struct ptlrpc_service *ost_io_service;
659 struct semaphore ost_health_sem;
660 int ost_sync_on_lock_cancel;
663 struct echo_client_obd {
664 struct obd_export *ec_exp; /* the local connection to osc/lov */
666 struct list_head ec_objects;
672 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
673 struct list_head lqo_oss_list; /* link to lov_qos */
674 __u64 lqo_bavail; /* total bytes avail on OSS */
675 __u64 lqo_penalty; /* current penalty */
676 __u64 lqo_penalty_per_obj; /* penalty decrease every obj*/
677 time_t lqo_used; /* last used time, seconds */
678 __u32 lqo_ost_count; /* number of osts on this oss */
682 struct lov_qos_oss *ltq_oss; /* oss info */
683 __u64 ltq_penalty; /* current penalty */
684 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
685 __u64 ltq_weight; /* net weighting */
686 time_t ltq_used; /* last used time, seconds */
687 unsigned int ltq_usable:1; /* usable for striping */
690 /* Generic subset of OSTs */
692 __u32 *op_array; /* array of index of
694 unsigned int op_count; /* number of OSTs in the array */
695 unsigned int op_size; /* allocated size of lp_array */
696 struct rw_semaphore op_rw_sem; /* to protect ost_pool use */
699 /* Round-robin allocator data */
701 __u32 lqr_start_idx; /* start index of new inode */
702 __u32 lqr_offset_idx; /* aliasing for start_idx */
703 int lqr_start_count; /* reseed counter */
704 struct ost_pool lqr_pool; /* round-robin optimized list */
705 unsigned long lqr_dirty:1; /* recalc round-robin list */
708 /* allow statfs data caching for 1 second */
709 #define OBD_STATFS_CACHE_SECONDS 1
711 struct lov_statfs_data {
712 struct obd_info lsd_oi;
713 struct obd_statfs lsd_statfs;
715 /* Stripe placement optimization */
717 struct list_head lq_oss_list; /* list of OSSs that targets use */
718 struct rw_semaphore lq_rw_sem;
719 __u32 lq_active_oss_count;
720 unsigned int lq_prio_free; /* priority for free space */
721 unsigned int lq_threshold_rr;/* priority for rr */
722 struct lov_qos_rr lq_rr; /* round robin qos data */
723 unsigned long lq_dirty:1, /* recalc qos data */
724 lq_same_space:1,/* the ost's all have approx.
725 the same space avail */
726 lq_reset:1, /* zero current penalties */
727 lq_statfs_in_progress:1; /* statfs op in progress */
728 /* qos statfs data */
729 struct lov_statfs_data *lq_statfs_data;
730 cfs_waitq_t lq_statfs_waitq; /* waitqueue to notify statfs
731 * requests completion */
734 struct lov_tgt_desc {
735 struct list_head ltd_kill;
736 struct obd_uuid ltd_uuid;
737 struct obd_device *ltd_obd;
738 struct obd_export *ltd_exp;
739 struct ltd_qos ltd_qos; /* qos info per target */
741 __u32 ltd_index; /* index in lov_obd->tgts */
742 unsigned long ltd_active:1,/* is this target up for requests */
743 ltd_activate:1,/* should this target be activated */
744 ltd_reap:1; /* should this target be deleted */
748 #define pool_tgt_size(_p) _p->pool_obds.op_size
749 #define pool_tgt_count(_p) _p->pool_obds.op_count
750 #define pool_tgt_array(_p) _p->pool_obds.op_array
751 #define pool_tgt_rw_sem(_p) _p->pool_obds.op_rw_sem
752 #define pool_tgt(_p, _i) _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
755 char pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
756 struct ost_pool pool_obds; /* pool members */
757 atomic_t pool_refcount; /* pool ref. counter */
758 struct lov_qos_rr pool_rr; /* round robin qos */
759 struct hlist_node pool_hash; /* access by poolname */
760 struct list_head pool_list; /* serial access */
761 cfs_proc_dir_entry_t *pool_proc_entry; /* file in /proc */
762 struct lov_obd *pool_lov; /* lov obd to which this
767 struct lov_desc desc;
768 struct lov_tgt_desc **lov_tgts; /* sparse array */
769 struct ost_pool lov_packed; /* all OSTs in a packed
771 struct semaphore lov_lock;
772 struct obd_connect_data lov_ocd;
773 struct lov_qos lov_qos; /* qos info per lov */
774 atomic_t lov_refcount;
775 __u32 lov_tgt_count; /* how many OBD's */
776 __u32 lov_active_tgt_count; /* how many active */
777 __u32 lov_death_row;/* tgts scheduled to be deleted */
778 __u32 lov_tgt_size; /* size of tgts array */
780 obd_page_removal_cb_t lov_page_removal_cb;
781 obd_pin_extent_cb lov_page_pin_cb;
782 obd_lock_cancel_cb lov_lock_cancel_cb;
784 lustre_hash_t *lov_pools_hash_body; /* used for key access */
785 struct list_head lov_pool_list; /* used for sequential access */
786 cfs_proc_dir_entry_t *lov_pool_proc_entry;
789 struct niobuf_local {
794 cfs_dentry_t *dentry;
799 /* obd device type names */
800 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
801 #define LUSTRE_MDS_NAME "mds"
802 #define LUSTRE_MDT_NAME "mdt"
803 #define LUSTRE_MDC_NAME "mdc"
804 #define LUSTRE_OSS_NAME "ost" /*FIXME change name to oss*/
805 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost*/
806 #define LUSTRE_OSC_NAME "osc"
807 #define LUSTRE_LOV_NAME "lov"
808 #define LUSTRE_MGS_NAME "mgs"
809 #define LUSTRE_MGC_NAME "mgc"
811 #define LUSTRE_CACHEOBD_NAME "cobd"
812 #define LUSTRE_ECHO_NAME "obdecho"
813 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
815 /* Constant obd names (post-rename) */
816 #define LUSTRE_MDS_OBDNAME "MDS"
817 #define LUSTRE_OSS_OBDNAME "OSS"
818 #define LUSTRE_MGS_OBDNAME "MGS"
819 #define LUSTRE_MGC_OBDNAME "MGC"
821 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
822 #define N_LOCAL_TEMP_PAGE 0x10000000
824 struct obd_trans_info {
827 /* Only used on the server side for tracking acks. */
828 struct oti_req_ack_lock {
829 struct lustre_handle lock;
833 struct llog_cookie oti_onecookie;
834 struct llog_cookie *oti_logcookies;
837 /* initial thread handling transaction */
838 struct ptlrpc_thread * oti_thread;
841 __u64 oti_pre_version;
843 struct obd_uuid *oti_ost_uuid;
846 static inline void oti_init(struct obd_trans_info *oti,
847 struct ptlrpc_request *req)
851 memset(oti, 0, sizeof(*oti));
856 oti->oti_xid = req->rq_xid;
857 /* VBR: take versions from request */
858 if (req->rq_reqmsg != NULL &&
859 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
860 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
861 /* b1.6 interoperability check. pre_versions may be NULL */
862 oti->oti_pre_version = pre_version ? pre_version[0] : 0;
863 oti->oti_transno = lustre_msg_get_transno(req->rq_reqmsg);
866 /* called from mds_create_objects */
867 if (req->rq_repmsg != NULL)
868 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
869 oti->oti_thread = req->rq_svc_thread;
870 if (req->rq_reqmsg != NULL)
871 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
874 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
879 if (num_cookies == 1)
880 oti->oti_logcookies = &oti->oti_onecookie;
882 OBD_ALLOC(oti->oti_logcookies,
883 num_cookies * sizeof(oti->oti_onecookie));
885 oti->oti_numcookies = num_cookies;
888 static inline void oti_free_cookies(struct obd_trans_info *oti)
890 if (!oti || !oti->oti_logcookies)
893 if (oti->oti_logcookies == &oti->oti_onecookie)
894 LASSERT(oti->oti_numcookies == 1);
896 OBD_FREE(oti->oti_logcookies,
897 oti->oti_numcookies * sizeof(oti->oti_onecookie));
898 oti->oti_logcookies = NULL;
899 oti->oti_numcookies = 0;
904 LLOG_CONFIG_ORIG_CTXT = 0,
905 LLOG_CONFIG_REPL_CTXT = 1,
906 LLOG_MDS_OST_ORIG_CTXT = 2,
907 LLOG_MDS_OST_REPL_CTXT = 3,
908 LLOG_SIZE_ORIG_CTXT = 4,
909 LLOG_SIZE_REPL_CTXT = 5,
910 LLOG_MD_ORIG_CTXT = 6,
911 LLOG_MD_REPL_CTXT = 7,
912 LLOG_RD1_ORIG_CTXT = 8,
913 LLOG_RD1_REPL_CTXT = 9,
914 LLOG_TEST_ORIG_CTXT = 10,
915 LLOG_TEST_REPL_CTXT = 11,
916 LLOG_LOVEA_ORIG_CTXT = 12,
917 LLOG_LOVEA_REPL_CTXT = 13,
922 * Events signalled through obd_notify() upcall-chain.
924 enum obd_notify_event {
925 /* device target is created */
927 /* DEVICE connect start */
929 /* Device activated */
931 /* Device deactivated */
933 /* Connect data for import were changed */
936 OBD_NOTIFY_SYNC_NONBLOCK,
938 /* Configuration event */
942 #define CONFIG_LOG 0x1 /* finished processing config log */
943 #define CONFIG_SYNC 0x2 /* mdt synced 1 ost */
944 #define CONFIG_TARGET 0x4 /* one target is added */
947 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
948 * and liblustre being main examples).
950 struct obd_notify_upcall {
951 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
952 enum obd_notify_event ev, void *owner);
953 /* Opaque datum supplied by upper layer listener */
957 /* corresponds to one of the obd's */
958 #define MAX_OBD_NAME 128
959 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
960 #define OBD_DEV_BY_DEVNAME 0xffffd0de
962 struct obd_type *obd_type;
965 /* common and UUID name of this device */
966 char obd_name[MAX_OBD_NAME];
967 struct obd_uuid obd_uuid;
970 unsigned long obd_attached:1, /* finished attach */
971 obd_set_up:1, /* finished setup */
972 obd_recovering:1, /* there are recoverable clients */
973 obd_abort_recovery:1,/* recovery expired */
974 obd_version_recov:1, /* obd uses version checking */
975 obd_replayable:1, /* recovery is enabled; inform clients */
976 obd_no_transno:1, /* no committed-transno notification */
977 obd_no_recov:1, /* fail instead of retry messages */
978 obd_stopping:1, /* started cleanup */
979 obd_starting:1, /* started setup */
980 obd_force:1, /* cleanup with > 0 obd refcount */
981 obd_fail:1, /* cleanup with failover */
982 obd_async_recov:1, /* allow asyncronous orphan cleanup */
983 obd_no_conn:1, /* deny new connections */
984 obd_inactive:1; /* device active/inactive
985 * (for /proc/status only!!) */
986 /* uuid-export hash body */
987 struct lustre_hash *obd_uuid_hash;
988 /* nid-export hash body */
989 struct lustre_hash *obd_nid_hash;
991 struct lustre_hash *obd_nid_stats_hash;
992 struct list_head obd_nid_stats;
993 atomic_t obd_refcount;
994 cfs_waitq_t obd_refcount_waitq;
995 struct list_head obd_exports;
996 struct list_head obd_delayed_exports;
998 spinlock_t obd_nid_lock;
999 struct ldlm_namespace *obd_namespace;
1000 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
1001 /* a spinlock is OK for what we do now, may need a semaphore later */
1002 spinlock_t obd_dev_lock;
1003 struct semaphore obd_dev_sem;
1004 __u64 obd_last_committed;
1005 struct fsfilt_operations *obd_fsops;
1006 spinlock_t obd_osfs_lock;
1007 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
1009 struct lvfs_run_ctxt obd_lvfs_ctxt;
1011 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
1012 struct semaphore obd_llog_cat_process;
1013 cfs_waitq_t obd_llog_waitq;
1015 struct obd_device *obd_observer;
1016 struct rw_semaphore obd_observer_link_sem;
1017 struct obd_notify_upcall obd_upcall;
1018 struct obd_export *obd_self_export;
1019 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
1020 struct list_head obd_exports_timed;
1021 time_t obd_eviction_timer; /* for ping evictor */
1023 /* XXX encapsulate all this recovery data into one struct */
1024 svc_handler_t obd_recovery_handler;
1025 int obd_max_recoverable_clients;
1026 int obd_connected_clients;
1027 int obd_recoverable_clients;
1028 int obd_stale_clients;
1029 int obd_delayed_clients;
1030 spinlock_t obd_processing_task_lock; /* BH lock (timer) */
1031 pid_t obd_processing_task;
1032 /* thread to handle recovery queue */
1033 struct ptlrpc_thread *obd_recovery_thread;
1034 __u64 obd_next_recovery_transno;
1035 int obd_replayed_requests;
1036 int obd_requests_queued_for_recovery;
1037 cfs_waitq_t obd_next_transno_waitq;
1038 cfs_timer_t obd_recovery_timer;
1039 struct list_head obd_recovery_queue;
1040 struct list_head obd_delayed_reply_queue;
1041 time_t obd_recovery_start; /* seconds */
1042 time_t obd_recovery_end; /* seconds, for lprocfs_status */
1043 time_t obd_recovery_time_hard;
1044 int obd_recovery_timeout;
1047 struct obd_device_target obt;
1048 struct filter_obd filter;
1050 struct client_obd cli;
1052 struct echo_client_obd echo_client;
1053 struct echo_obd echo;
1057 /* Fields used by LProcFS */
1058 cfs_proc_dir_entry_t *obd_proc_entry;
1059 cfs_proc_dir_entry_t *obd_proc_exports_entry;
1060 cfs_proc_dir_entry_t *obd_svc_procroot;
1061 struct lprocfs_stats *obd_stats;
1062 struct lprocfs_stats *obd_svc_stats;
1063 unsigned int obd_cntr_base;
1064 atomic_t obd_evict_inprogress;
1065 cfs_waitq_t obd_evict_inprogress_waitq;
1066 struct list_head obd_evict_list; /* protected with pet_lock */
1068 /* Ldlm pool part. Save last calculated SLV and Limit. */
1069 rwlock_t obd_pool_lock;
1074 #define OBD_LLOG_FL_SENDNOW 0x0001
1076 enum obd_cleanup_stage {
1077 /* Special case hack for MDS LOVs */
1079 /* Precleanup stage 1, we must make sure all exports (other than the
1080 self-export) get destroyed. */
1081 OBD_CLEANUP_EXPORTS,
1082 /* Precleanup stage 2, do other type-specific cleanup requiring the
1084 OBD_CLEANUP_SELF_EXP,
1085 /* FIXME we should eliminate the "precleanup" function and make them stages
1086 of the "cleanup" function. */
1090 /* get/set_info keys */
1091 #define KEY_MDS_CONN "mds_conn"
1092 #define KEY_NEXT_ID "next_id"
1093 #define KEY_LOVDESC "lovdesc"
1094 #define KEY_INIT_RECOV "initial_recov"
1095 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
1096 #define KEY_LAST_ID "last_id"
1097 #define KEY_LOCK_TO_STRIPE "lock_to_stripe"
1098 #define KEY_CHECKSUM "checksum"
1099 #define KEY_READONLY "read-only"
1100 #define KEY_READONLY_166COMPAT "readonly"
1101 #define KEY_EVICT_BY_NID "evict_by_nid"
1102 #define KEY_REGISTER_TARGET "register_target"
1103 #define KEY_SET_FS "set_fs"
1104 #define KEY_CLEAR_FS "clear_fs"
1105 #define KEY_SET_INFO "set_info"
1106 #define KEY_BLOCKSIZE "blocksize"
1107 #define KEY_BLOCKSIZE_BITS "blocksize_bits"
1108 #define KEY_MAX_EASIZE "max_ea_size"
1109 #define KEY_FIEMAP "fiemap"
1110 #define KEY_CONNECT_FLAG "connect_flags"
1112 #define KEY_ASYNC "async"
1113 #define KEY_CAPA_KEY "capa_key"
1114 #define KEY_GRANT_SHRINK "grant_shrink"
1115 #define KEY_OFF_RPCSIZE "off_rpcsize"
1118 struct module *o_owner;
1119 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
1120 void *karg, void *uarg);
1121 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
1122 __u32 *vallen, void *val, struct lov_stripe_md *lsm);
1123 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
1124 __u32 vallen, void *val,
1125 struct ptlrpc_request_set *set);
1126 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
1127 int (*o_detach)(struct obd_device *dev);
1128 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
1129 int (*o_precleanup)(struct obd_device *dev,
1130 enum obd_cleanup_stage cleanup_stage);
1131 int (*o_cleanup)(struct obd_device *dev);
1132 int (*o_process_config)(struct obd_device *dev, obd_count len,
1134 int (*o_postrecov)(struct obd_device *dev);
1135 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
1137 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
1138 /* connect to the target device with given connection
1139 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
1140 * granted by the target, which are guaranteed to be a subset of flags
1141 * asked for. If @ocd == NULL, use default parameters. */
1142 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
1143 struct obd_uuid *cluuid, struct obd_connect_data *ocd,
1145 int (*o_reconnect)(struct obd_export *exp, struct obd_device *src,
1146 struct obd_uuid *cluuid,
1147 struct obd_connect_data *ocd,
1149 int (*o_disconnect)(struct obd_export *exp);
1151 /* Initialize/finalize fids infrastructure. */
1152 int (*o_fid_init)(struct obd_export *exp);
1153 int (*o_fid_fini)(struct obd_export *exp);
1155 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
1156 __u64 max_age, __u32 flags);
1157 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
1158 __u64 max_age, struct ptlrpc_request_set *set);
1159 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
1160 struct lov_stripe_md *mem_src);
1161 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
1162 struct lov_mds_md *disk_src, int disk_len);
1163 int (*o_checkmd)(struct obd_export *exp, struct obd_export *md_exp,
1164 struct lov_stripe_md *mem_tgt);
1165 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
1167 int (*o_precreate)(struct obd_export *exp);
1168 int (*o_create)(struct obd_export *exp, struct obdo *oa,
1169 struct lov_stripe_md **ea, struct obd_trans_info *oti);
1170 int (*o_create_async)(struct obd_export *exp, struct obd_info *oinfo,
1171 struct lov_stripe_md **ea,
1172 struct obd_trans_info *oti);
1173 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
1174 struct lov_stripe_md *ea, struct obd_trans_info *oti,
1175 struct obd_export *md_exp);
1176 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
1177 struct obd_trans_info *oti);
1178 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1179 struct obd_trans_info *oti,
1180 struct ptlrpc_request_set *rqset);
1181 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
1182 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
1183 struct ptlrpc_request_set *set);
1184 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
1185 obd_count oa_bufs, struct brw_page *pgarr,
1186 struct obd_trans_info *oti);
1187 int (*o_brw_async)(int rw, struct obd_export *exp,
1188 struct obd_info *oinfo, obd_count oa_bufs,
1189 struct brw_page *pgarr, struct obd_trans_info *oti,
1190 struct ptlrpc_request_set *, int pshift);
1191 int (*o_prep_async_page)(struct obd_export *exp,
1192 struct lov_stripe_md *lsm,
1193 struct lov_oinfo *loi,
1194 cfs_page_t *page, obd_off offset,
1195 struct obd_async_page_ops *ops, void *data,
1196 void **res, int flags,
1197 struct lustre_handle *lockh);
1198 int (*o_get_lock)(struct obd_export *exp, struct lov_stripe_md *lsm,
1199 void **res, int rw, obd_off start, obd_off end,
1200 struct lustre_handle *lockh, int flags);
1201 int (*o_queue_async_io)(struct obd_export *exp,
1202 struct lov_stripe_md *lsm,
1203 struct lov_oinfo *loi, void *cookie,
1204 int cmd, obd_off off, int count,
1205 obd_flag brw_flags, obd_flag async_flags);
1206 int (*o_queue_group_io)(struct obd_export *exp,
1207 struct lov_stripe_md *lsm,
1208 struct lov_oinfo *loi,
1209 struct obd_io_group *oig,
1210 void *cookie, int cmd, obd_off off, int count,
1211 obd_flag brw_flags, obd_flag async_flags);
1212 int (*o_trigger_group_io)(struct obd_export *exp,
1213 struct lov_stripe_md *lsm,
1214 struct lov_oinfo *loi,
1215 struct obd_io_group *oig);
1216 int (*o_set_async_flags)(struct obd_export *exp,
1217 struct lov_stripe_md *lsm,
1218 struct lov_oinfo *loi, void *cookie,
1219 obd_flag async_flags);
1220 int (*o_teardown_async_page)(struct obd_export *exp,
1221 struct lov_stripe_md *lsm,
1222 struct lov_oinfo *loi, void *cookie);
1223 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1224 struct ost_lvb *lvb, int kms_only);
1225 int (*o_update_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
1226 struct ost_lvb *lvb, obd_flag valid);
1227 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
1228 obd_off size, int shrink);
1229 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
1230 struct obd_trans_info *oti,
1231 struct ptlrpc_request_set *rqset);
1232 int (*o_sync)(struct obd_export *exp, struct obd_info *oinfo,
1233 obd_size start, obd_size end,
1234 struct ptlrpc_request_set *rqset);
1235 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
1236 struct lov_stripe_md *src, obd_size start,
1237 obd_size end, struct obd_trans_info *oti);
1238 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
1239 struct lustre_handle *srconn, struct lov_stripe_md *src,
1240 obd_size start, obd_size end, struct obd_trans_info *);
1241 int (*o_iterate)(struct lustre_handle *conn,
1242 int (*)(obd_id, obd_gr, void *),
1243 obd_id *startid, obd_gr group, void *data);
1244 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
1245 int objcount, struct obd_ioobj *obj,
1246 struct niobuf_remote *remote, int *nr_pages,
1247 struct niobuf_local *local,
1248 struct obd_trans_info *oti);
1249 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
1250 int objcount, struct obd_ioobj *obj,
1251 struct niobuf_remote *remote, int pages,
1252 struct niobuf_local *local,
1253 struct obd_trans_info *oti, int rc);
1254 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
1255 struct ldlm_enqueue_info *einfo,
1256 struct ptlrpc_request_set *rqset);
1257 int (*o_match)(struct obd_export *, struct lov_stripe_md *, __u32 type,
1258 ldlm_policy_data_t *, __u32 mode, int *flags, void *data,
1259 struct lustre_handle *lockh, int *n_matches);
1260 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
1261 ldlm_iterator_t it, void *data);
1262 int (*o_find_cbdata)(struct obd_export *, struct lov_stripe_md *,
1263 ldlm_iterator_t it, void *data);
1264 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
1265 __u32 mode, struct lustre_handle *, int flags,
1267 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
1268 int flags, void *opaque);
1269 int (*o_join_lru)(struct obd_export *, struct lov_stripe_md *,
1271 int (*o_init_export)(struct obd_export *exp);
1272 int (*o_destroy_export)(struct obd_export *exp);
1273 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
1274 int cmd, obd_off *);
1276 /* llog related obd_methods */
1277 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
1279 int (*o_llog_finish)(struct obd_device *obd, int count);
1281 /* metadata-only methods */
1282 int (*o_pin)(struct obd_export *, struct ll_fid *,
1283 struct obd_client_handle *, int flag);
1284 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
1286 int (*o_import_event)(struct obd_device *, struct obd_import *,
1287 enum obd_import_event);
1289 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
1290 enum obd_notify_event ev, void *data);
1292 int (*o_health_check)(struct obd_device *);
1295 int (*o_quotacheck)(struct obd_export *, struct obd_quotactl *);
1296 int (*o_quotactl)(struct obd_export *, struct obd_quotactl *);
1297 int (*o_quota_adjust_qunit)(struct obd_export *exp,
1298 struct quota_adjust_qunit *oqaq,
1299 struct lustre_quota_ctxt *qctxt);
1302 int (*o_ping)(struct obd_export *exp);
1304 int (*o_register_page_removal_cb)(struct obd_device *obd,
1305 obd_page_removal_cb_t cb,
1306 obd_pin_extent_cb pin_cb);
1307 int (*o_unregister_page_removal_cb)(struct obd_device *obd,
1308 obd_page_removal_cb_t cb);
1309 int (*o_register_lock_cancel_cb)(struct obd_device *obd,
1310 obd_lock_cancel_cb cb);
1311 int (*o_unregister_lock_cancel_cb)(struct obd_device *obd,
1312 obd_lock_cancel_cb cb);
1314 int (*o_pool_new)(struct obd_device *obd, char *poolname);
1315 int (*o_pool_del)(struct obd_device *obd, char *poolname);
1316 int (*o_pool_add)(struct obd_device *obd, char *poolname,
1318 int (*o_pool_rem)(struct obd_device *obd, char *poolname,
1320 void (*o_getref)(struct obd_device *obd);
1321 void (*o_putref)(struct obd_device *obd);
1324 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
1325 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
1326 * Also, add a wrapper function in include/linux/obd_class.h. */
1329 struct lsm_operations {
1330 void (*lsm_free)(struct lov_stripe_md *);
1331 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
1332 struct obd_export *md_exp);
1333 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1335 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1337 obd_off (*lsm_stripe_offset_by_index)(struct lov_stripe_md *, int);
1338 obd_off (*lsm_stripe_offset_by_offset)(struct lov_stripe_md *, obd_off);
1339 int (*lsm_stripe_index_by_offset)(struct lov_stripe_md *, obd_off);
1340 int (*lsm_revalidate) (struct lov_stripe_md *, struct obd_device *obd);
1341 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1343 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1344 struct lov_mds_md *lmm);
1347 extern struct lsm_operations lsm_v1_ops;
1348 extern struct lsm_operations lsm_join_ops;
1349 extern struct lsm_operations lsm_v3_ops;
1350 static inline struct lsm_operations *lsm_op_find(int magic)
1355 case LOV_MAGIC_JOIN:
1356 return &lsm_join_ops;
1360 CERROR("Cannot recognize lsm_magic %x\n", magic);
1365 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1367 /* Requests for obd_extent_calc() */
1368 #define OBD_CALC_STRIPE_START 0x0001
1369 #define OBD_CALC_STRIPE_END 0x0010
1370 #define OBD_CALC_STRIPE_RPC_ALIGN 0x0100
1372 #define OBD_CALC_STRIPE_RPC_START_ALIGN (OBD_CALC_STRIPE_START | \
1373 OBD_CALC_STRIPE_RPC_ALIGN)
1374 #define OBD_CALC_STRIPE_RPC_END_ALIGN (OBD_CALC_STRIPE_END | \
1375 OBD_CALC_STRIPE_RPC_ALIGN)
1377 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1378 struct obd_export *exp, int error)
1381 CERROR("%s: transno "LPU64" commit error: %d\n",
1382 obd->obd_name, transno, error);
1385 if (exp && transno > exp->exp_last_committed) {
1386 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1387 obd->obd_name, transno);
1388 exp->exp_last_committed = transno;
1389 ptlrpc_commit_replies(exp);
1391 CDEBUG(D_INFO, "%s: transno "LPU64" committed\n",
1392 obd->obd_name, transno);
1394 if (transno > obd->obd_last_committed)
1395 obd->obd_last_committed = transno;
1398 static inline void init_obd_quota_ops(quota_interface_t *interface,
1399 struct obd_ops *obd_ops)
1405 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1406 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1407 obd_ops->o_quota_adjust_qunit = QUOTA_OP(interface, adjust_qunit);
1415 /* Default preferred checksum algorithm to use (if supported by the server) */
1416 #define OSC_DEFAULT_CKSUM OBD_CKSUM_ADLER
1417 /* Adler-32 is supported */
1418 #define CHECKSUM_ADLER OBD_CKSUM_ADLER
1420 #define OSC_DEFAULT_CKSUM OBD_CKSUM_CRC32
1421 #define CHECKSUM_ADLER 0
1424 #define OBD_CKSUM_ALL (OBD_CKSUM_CRC32 | CHECKSUM_ADLER)
1426 /* Checksum algorithm names. Must be defined in the same order as the
1427 * OBD_CKSUM_* flags. */
1428 #define DECLARE_CKSUM_NAME char *cksum_name[] = {"crc32", "adler"}
1430 #endif /* __OBD_H */