1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
10 #elif defined(__APPLE__)
11 #include <darwin/obd.h>
12 #elif defined(__WINNT__)
13 #include <winnt/obd.h>
15 #error Unsupported operating system.
18 #define IOC_OSC_TYPE 'h'
19 #define IOC_OSC_MIN_NR 20
20 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
21 #define IOC_OSC_MAX_NR 50
23 #define IOC_MDC_TYPE 'i'
24 #define IOC_MDC_MIN_NR 20
25 /* Moved to lustre_user.h
26 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_ioctl_data *)
27 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
28 #define IOC_MDC_MAX_NR 50
30 #include <lustre_lib.h>
31 #include <lustre/lustre_idl.h>
32 #include <lustre_export.h>
33 #include <lustre_quota.h>
34 #include <class_hash.h>
36 #define MAX_OBD_DEVICES 8192
38 /* this is really local to the OSC */
39 struct loi_oap_pages {
40 struct list_head lop_pending;
41 struct list_head lop_urgent;
42 struct list_head lop_pending_group;
52 struct lov_oinfo { /* per-stripe data structure */
53 __u64 loi_id; /* object ID on the target OST */
54 __u64 loi_gr; /* object group on the target OST */
55 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
56 int loi_ost_gen; /* generation of this loi_ost_idx */
58 /* used by the osc to keep track of what objects to build into rpcs */
59 struct loi_oap_pages loi_read_lop;
60 struct loi_oap_pages loi_write_lop;
61 /* _cli_ is poorly named, it should be _ready_ */
62 struct list_head loi_cli_item;
63 struct list_head loi_write_item;
64 struct list_head loi_read_item;
66 unsigned loi_kms_valid:1;
67 __u64 loi_kms; /* known minimum size */
68 struct ost_lvb loi_lvb;
69 struct osc_async_rc loi_ar;
72 static inline void loi_init(struct lov_oinfo *loi)
74 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
75 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
76 CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
77 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
78 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
79 CFS_INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
80 CFS_INIT_LIST_HEAD(&loi->loi_cli_item);
81 CFS_INIT_LIST_HEAD(&loi->loi_write_item);
82 CFS_INIT_LIST_HEAD(&loi->loi_read_item);
85 /*extent array item for describing the joined file extent info*/
87 __u64 le_start; /* extent start */
88 __u64 le_len; /* extent length */
89 int le_loi_idx; /* extent #1 loi's index in lsm loi array */
90 int le_stripe_count; /* extent stripe count*/
93 /*Lov array info for describing joined file array EA info*/
94 struct lov_array_info {
95 struct llog_logid lai_array_id; /* MDS med llog object id */
96 unsigned lai_ext_count; /* number of extent count */
97 struct lov_extent *lai_ext_array; /* extent desc array */
100 struct lov_stripe_md {
102 void *lsm_lock_owner; /* debugging */
105 /* Public members. */
106 __u64 lw_object_id; /* lov object id */
107 __u64 lw_object_gr; /* lov object group */
108 __u64 lw_maxbytes; /* maximum possible file size */
110 /* LOV-private members start here -- only for use in lov/. */
112 __u32 lw_stripe_size; /* size of the stripe */
113 __u32 lw_pattern; /* striping pattern (RAID0, RAID1) */
114 unsigned lw_stripe_count; /* number of objects being striped over */
117 struct lov_array_info *lsm_array; /*Only for joined file array info*/
118 struct lov_oinfo *lsm_oinfo[0];
121 #define lsm_object_id lsm_wire.lw_object_id
122 #define lsm_object_gr lsm_wire.lw_object_gr
123 #define lsm_maxbytes lsm_wire.lw_maxbytes
124 #define lsm_magic lsm_wire.lw_magic
125 #define lsm_stripe_size lsm_wire.lw_stripe_size
126 #define lsm_pattern lsm_wire.lw_pattern
127 #define lsm_stripe_count lsm_wire.lw_stripe_count
131 typedef int (*obd_enqueue_update_f)(struct obd_info *oinfo, int rc);
133 /* obd_enqueue parameters common for all levels (lov, osc). */
134 struct obd_enqueue_info {
135 /* Flags used while lock handling. */
137 /* Type of the lock being enqueued. */
139 /* Mode of the lock being enqueued. */
141 /* Different callbacks for lock handling (blocking, completion,
146 /* Data to be passed into callbacks. */
148 /* Request set for OSC async requests. */
149 struct ptlrpc_request_set *ei_rqset;
152 /* obd info for a particular level (lov, osc). */
154 /* Lock policy. It keeps an extent which is specific for a particular
155 * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
156 * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
157 ldlm_policy_data_t oi_policy;
158 /* Lock handle specific for every OSC lock. */
159 struct lustre_handle *oi_lockh;
160 /* lsm data specific for every OSC. */
161 struct lov_stripe_md *oi_md;
162 /* obdo data specific for every OSC, if needed at all. */
164 /* statfs data specific for every OSC, if needed at all. */
165 struct obd_statfs *oi_osfs;
166 /* An update callback which is called to update some data on upper
167 * level. E.g. it is used for update lsm->lsm_oinfo at every recieved
168 * request in osc level for enqueue requests. It is also possible to
169 * update some caller data from LOV layer if needed. */
170 obd_enqueue_update_f oi_cb_up;
173 /* compare all relevant fields. */
174 static inline int lov_stripe_md_cmp(struct lov_stripe_md *m1,
175 struct lov_stripe_md *m2)
178 * ->lsm_wire contains padding, but it should be zeroed out during
181 return memcmp(&m1->lsm_wire, &m2->lsm_wire, sizeof m1->lsm_wire);
184 void lov_stripe_lock(struct lov_stripe_md *md);
185 void lov_stripe_unlock(struct lov_stripe_md *md);
188 struct list_head typ_chain;
189 struct obd_ops *typ_ops;
190 cfs_proc_dir_entry_t *typ_procroot;
193 spinlock_t obd_type_lock;
204 ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
205 page is added to an rpc */
206 ASYNC_URGENT = 0x2, /* page must be put into an RPC before return */
207 ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
208 to give the caller a chance to update
209 or cancel the size of the io */
210 ASYNC_GROUP_SYNC = 0x8, /* ap_completion will not be called, instead
211 the page is accounted for in the
212 obd_io_group given to
213 obd_queue_group_io */
216 struct obd_async_page_ops {
217 int (*ap_make_ready)(void *data, int cmd);
218 int (*ap_refresh_count)(void *data, int cmd);
219 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
220 void (*ap_update_obdo)(void *data, int cmd, struct obdo *oa,
222 int (*ap_completion)(void *data, int cmd, struct obdo *oa, int rc);
225 /* the `oig' is passed down from a caller of obd rw methods. the callee
226 * records enough state such that the caller can sleep on the oig and
227 * be woken when all the callees have finished their work */
228 struct obd_io_group {
230 atomic_t oig_refcount;
233 struct list_head oig_occ_list;
234 cfs_waitq_t oig_waitq;
237 /* the oig callback context lets the callee of obd rw methods register
238 * for callbacks from the caller. */
239 struct oig_callback_context {
240 struct list_head occ_oig_item;
241 /* called when the caller has received a signal while sleeping.
242 * callees of this method are encouraged to abort their state
243 * in the oig. This may be called multiple times. */
244 void (*occ_interrupted)(struct oig_callback_context *occ);
245 unsigned int interrupted:1;
248 /* Individual type definitions */
250 struct ost_server_data;
252 /* hold common fields for "target" device */
253 struct obd_device_target {
254 struct super_block *obt_sb;
255 atomic_t obt_quotachecking;
256 struct lustre_quota_ctxt obt_qctxt;
259 #define FILTER_GROUP_LLOG 1
260 #define FILTER_GROUP_ECHO 2
268 /* NB this field MUST be first */
269 struct obd_device_target fo_obt;
270 const char *fo_fstype;
271 struct vfsmount *fo_vfsmnt;
272 cfs_dentry_t *fo_dentry_O;
273 cfs_dentry_t **fo_dentry_O_groups;
274 cfs_dentry_t **fo_dentry_O_sub;
275 spinlock_t fo_objidlock; /* protect fo_lastobjid */
276 spinlock_t fo_translock; /* protect fsd_last_transno */
277 struct file *fo_rcvd_filp;
278 struct file *fo_health_check_filp;
279 struct lr_server_data *fo_fsd;
280 unsigned long *fo_last_rcvd_slots;
281 __u64 fo_mount_count;
283 int fo_destroy_in_progress;
284 struct semaphore fo_create_lock;
286 struct list_head fo_export_list;
289 obd_size fo_tot_dirty; /* protected by obd_osfs_lock */
290 obd_size fo_tot_granted; /* all values in bytes */
291 obd_size fo_tot_pending;
293 obd_size fo_readcache_max_filesize;
295 struct obd_import *fo_mdc_imp;
296 struct obd_uuid fo_mdc_uuid;
297 struct lustre_handle fo_mdc_conn;
298 struct file **fo_last_objid_files;
299 __u64 *fo_last_objids; /* last created objid for groups,
300 * protected by fo_objidlock */
302 struct semaphore fo_alloc_lock;
304 atomic_t fo_r_in_flight;
305 atomic_t fo_w_in_flight;
308 * per-filter pool of kiobuf's allocated by filter_common_setup() and
309 * torn down by filter_cleanup(). Contains OST_NUM_THREADS elements of
310 * which ->fo_iobuf_count were allocated.
312 * This pool contains kiobuf used by
313 * filter_{prep,commit}rw_{read,write}() and is shared by all OST
316 * Locking: none, each OST thread uses only one element, determined by
317 * its "ordinal number", ->t_id.
319 struct filter_iobuf **fo_iobuf_pool;
322 struct brw_stats fo_filter_stats;
323 struct lustre_quota_ctxt fo_quota_ctxt;
324 spinlock_t fo_quotacheck_lock;
325 atomic_t fo_quotachecking;
327 int fo_fmd_max_num; /* per exp filter_mod_data */
328 int fo_fmd_max_age; /* jiffies to fmd expiry */
331 #define OSC_MAX_RIF_DEFAULT 8
332 #define OSC_MAX_RIF_MAX 256
333 #define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
334 #define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
336 #define MDC_MAX_RIF_DEFAULT 8
337 #define MDC_MAX_RIF_MAX 512
342 struct semaphore cl_sem;
343 struct obd_uuid cl_target_uuid;
344 struct obd_import *cl_import; /* ptlrpc connection state */
346 /* max_mds_easize is purely a performance thing so we don't have to
347 * call obd_size_diskmd() all the time. */
348 int cl_default_mds_easize;
349 int cl_max_mds_easize;
350 int cl_max_mds_cookiesize;
352 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
353 void *cl_llcd_offset;
355 /* the grant values are protected by loi_list_lock below */
356 long cl_dirty; /* all _dirty_ in bytes */
357 long cl_dirty_max; /* allowed w/o rpc */
358 long cl_avail_grant; /* bytes of credit for ost */
359 long cl_lost_grant; /* lost credits (trunc) */
360 struct list_head cl_cache_waiters; /* waiting for cache/grant */
362 /* keep track of objects that have lois that contain pages which
363 * have been queued for async brw. this lock also protects the
364 * lists of osc_client_pages that hang off of the loi */
366 * ->cl_loi_list_lock protects consistency of
367 * ->cl_loi_{ready,read,write}_list. ->ap_make_ready() and
368 * ->ap_completion() call-backs are executed under this lock. As we
369 * cannot guarantee that these call-backs never block on all platforms
370 * (as a matter of fact they do block on Mac OS X), type of
371 * ->cl_loi_list_lock is platform dependent: it's a spin-lock on Linux
372 * and blocking mutex on Mac OS X. (Alternative is to make this lock
373 * blocking everywhere, but we don't want to slow down fast-path of
374 * our main platform.)
376 * Exact type of ->cl_loi_list_lock is defined in arch/obd.h together
377 * with client_obd_list_{un,}lock() and
378 * client_obd_list_lock_{init,done}() functions.
380 client_obd_lock_t cl_loi_list_lock;
381 struct list_head cl_loi_ready_list;
382 struct list_head cl_loi_write_list;
383 struct list_head cl_loi_read_list;
386 /* just a sum of the loi/lop pending numbers to be exported by /proc */
387 int cl_pending_w_pages;
388 int cl_pending_r_pages;
389 int cl_max_pages_per_rpc;
390 int cl_max_rpcs_in_flight;
391 struct obd_histogram cl_read_rpc_hist;
392 struct obd_histogram cl_write_rpc_hist;
393 struct obd_histogram cl_read_page_hist;
394 struct obd_histogram cl_write_page_hist;
395 struct obd_histogram cl_read_offset_hist;
396 struct obd_histogram cl_write_offset_hist;
398 struct mdc_rpc_lock *cl_rpc_lock;
399 struct mdc_rpc_lock *cl_setattr_lock;
400 struct mdc_rpc_lock *cl_close_lock;
401 struct osc_creator cl_oscc;
404 struct semaphore cl_mgc_sem;
405 struct vfsmount *cl_mgc_vfsmnt;
406 struct dentry *cl_mgc_configs_dir;
407 atomic_t cl_mgc_refcount;
408 struct obd_export *cl_mgc_mgsexp;
411 unsigned int cl_checksum:1; /* debug checksums */
413 /* also protected by the poorly named _loi_list_lock lock above */
414 struct osc_async_rc cl_ar;
416 /* used by quotacheck */
417 int cl_qchk_stat; /* quotacheck stat of the peer */
419 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
421 #define CL_NOT_QUOTACHECKED 1 /* client->cl_qchk_stat init value */
424 struct ptlrpc_service *mgs_service;
425 struct vfsmount *mgs_vfsmnt;
426 struct super_block *mgs_sb;
427 struct dentry *mgs_configs_dir;
428 struct dentry *mgs_fid_de;
429 struct list_head mgs_fs_db_list;
430 struct semaphore mgs_sem;
431 cfs_proc_dir_entry_t *mgs_proc_live;
435 /* NB this field MUST be first */
436 struct obd_device_target mds_obt;
437 struct ptlrpc_service *mds_service;
438 struct ptlrpc_service *mds_setattr_service;
439 struct ptlrpc_service *mds_readpage_service;
440 struct vfsmount *mds_vfsmnt;
441 cfs_dentry_t *mds_fid_de;
443 int mds_max_cookiesize;
444 struct file *mds_rcvd_filp;
445 spinlock_t mds_transno_lock;
446 __u64 mds_last_transno;
447 __u64 mds_mount_count;
449 unsigned long mds_atime_diff;
450 struct semaphore mds_epoch_sem;
451 struct ll_fid mds_rootfid;
452 struct lr_server_data *mds_server_data;
453 cfs_dentry_t *mds_pending_dir;
454 cfs_dentry_t *mds_logs_dir;
455 cfs_dentry_t *mds_objects_dir;
456 struct llog_handle *mds_cfg_llh;
457 // struct llog_handle *mds_catalog;
458 struct obd_device *mds_osc_obd; /* XXX lov_obd */
459 struct obd_uuid mds_lov_uuid;
461 struct obd_export *mds_osc_exp; /* XXX lov_exp */
462 struct lov_desc mds_lov_desc;
463 obd_id *mds_lov_objids;
464 int mds_lov_objids_size;
465 __u32 mds_lov_objids_in_file;
466 unsigned int mds_lov_objids_dirty:1;
467 int mds_lov_nextid_set;
468 struct file *mds_lov_objid_filp;
469 struct file *mds_health_check_filp;
470 unsigned long *mds_client_bitmap;
471 struct upcall_cache *mds_group_hash;
473 struct lustre_quota_info mds_quota_info;
474 struct semaphore mds_qonoff_sem;
475 struct semaphore mds_health_sem;
476 unsigned long mds_lov_objids_valid:1,
485 struct lustre_handle eo_nl_lock;
490 struct ptlrpc_service *ost_service;
491 struct ptlrpc_service *ost_create_service;
492 struct ptlrpc_service *ost_io_service;
493 struct semaphore ost_health_sem;
496 struct echo_client_obd {
497 struct obd_export *ec_exp; /* the local connection to osc/lov */
499 struct list_head ec_objects;
505 struct obd_uuid lqo_uuid; /* ptlrpc's c_remote_uuid */
506 struct list_head lqo_oss_list; /* link to lov_qos */
507 __u32 lqo_ost_count; /* number of osts on this oss */
508 __u64 lqo_bavail; /* total bytes avail on OSS */
509 __u64 lqo_penalty; /* current penalty */
510 __u64 lqo_penalty_per_obj; /* penalty decrease every obj*/
514 struct lov_qos_oss *ltq_oss; /* oss info */
515 __u64 ltq_penalty; /* current penalty */
516 __u64 ltq_penalty_per_obj; /* penalty decrease every obj*/
517 __u64 ltq_weight; /* net weighting */
518 unsigned int ltq_usable:1; /* usable for striping */
522 struct list_head lq_oss_list; /* list of OSSs that targets use */
523 struct rw_semaphore lq_rw_sem;
524 __u32 lq_active_oss_count;
525 __u32 *lq_rr_array; /* round-robin optimized list */
526 unsigned int lq_rr_size; /* rr array size */
527 unsigned int lq_prio_free; /* priority for free space */
528 unsigned int lq_dirty:1, /* recalc qos data */
529 lq_dirty_rr:1, /* recalc round-robin list */
530 lq_same_space:1,/* the ost's all have approx.
531 the same space avail */
532 lq_reset:1; /* zero current penalties */
535 struct lov_tgt_desc {
536 struct obd_uuid ltd_uuid;
537 struct obd_export *ltd_exp;
538 struct ltd_qos ltd_qos; /* qos info per target */
540 __u32 ltd_index; /* index in lov_obd->tgts */
541 unsigned int ltd_active:1,/* is this target up for requests */
542 ltd_activate:1,/* should this target be activated */
543 ltd_reap:1; /* should this target be deleted */
547 struct lov_desc desc;
548 struct lov_tgt_desc **lov_tgts;
549 struct semaphore lov_lock;
550 struct obd_connect_data lov_ocd;
551 struct lov_qos lov_qos; /* qos info per lov */
552 atomic_t lov_refcount;
553 __u32 lov_tgt_count; /* how many OBD's */
554 __u32 lov_active_tgt_count; /* how many active */
555 __u32 lov_death_row;/* tgts scheduled to be deleted */
556 __u32 lov_tgt_size; /* size of tgts array */
557 __u32 lov_start_idx; /* start index of new inode */
558 __u32 lov_offset_idx; /* aliasing for start_idx */
559 int lov_start_count;/* reseed counter */
563 struct niobuf_local {
568 cfs_dentry_t *dentry;
573 /* obd device type names */
574 /* FIXME all the references to LUSTRE_MDS_NAME should be swapped with LUSTRE_MDT_NAME */
575 #define LUSTRE_MDS_NAME "mds"
576 #define LUSTRE_MDT_NAME "mdt"
577 #define LUSTRE_MDC_NAME "mdc"
578 #define LUSTRE_OSS_NAME "ost" /*FIXME change name to oss*/
579 #define LUSTRE_OST_NAME "obdfilter" /* FIXME change name to ost*/
580 #define LUSTRE_OSC_NAME "osc"
581 #define LUSTRE_LOV_NAME "lov"
582 #define LUSTRE_MGS_NAME "mgs"
583 #define LUSTRE_MGC_NAME "mgc"
585 #define LUSTRE_CACHEOBD_NAME "cobd"
586 #define LUSTRE_ECHO_NAME "obdecho"
587 #define LUSTRE_ECHO_CLIENT_NAME "echo_client"
589 /* Constant obd names (post-rename) */
590 #define LUSTRE_MDS_OBDNAME "MDS"
591 #define LUSTRE_OSS_OBDNAME "OSS"
592 #define LUSTRE_MGS_OBDNAME "MGS"
593 #define LUSTRE_MGC_OBDNAME "MGC"
595 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
596 #define N_LOCAL_TEMP_PAGE 0x10000000
598 struct obd_trans_info {
602 /* Only used on the server side for tracking acks. */
603 struct oti_req_ack_lock {
604 struct lustre_handle lock;
608 struct llog_cookie oti_onecookie;
609 struct llog_cookie *oti_logcookies;
612 /* initial thread handling transaction */
617 static inline void oti_init(struct obd_trans_info *oti,
618 struct ptlrpc_request *req)
622 memset(oti, 0, sizeof *oti);
627 oti->oti_xid = req->rq_xid;
629 if (req->rq_repmsg && req->rq_reqmsg != 0)
630 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
631 oti->oti_thread_id = req->rq_svc_thread ? req->rq_svc_thread->t_id : -1;
632 oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
635 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
640 if (num_cookies == 1)
641 oti->oti_logcookies = &oti->oti_onecookie;
643 OBD_ALLOC(oti->oti_logcookies,
644 num_cookies * sizeof(oti->oti_onecookie));
646 oti->oti_numcookies = num_cookies;
649 static inline void oti_free_cookies(struct obd_trans_info *oti)
651 if (!oti || !oti->oti_logcookies)
654 if (oti->oti_logcookies == &oti->oti_onecookie)
655 LASSERT(oti->oti_numcookies == 1);
657 OBD_FREE(oti->oti_logcookies,
658 oti->oti_numcookies * sizeof(oti->oti_onecookie));
659 oti->oti_logcookies = NULL;
660 oti->oti_numcookies = 0;
665 LLOG_CONFIG_ORIG_CTXT = 0,
666 LLOG_CONFIG_REPL_CTXT = 1,
667 LLOG_MDS_OST_ORIG_CTXT = 2,
668 LLOG_MDS_OST_REPL_CTXT = 3,
669 LLOG_SIZE_ORIG_CTXT = 4,
670 LLOG_SIZE_REPL_CTXT = 5,
671 LLOG_MD_ORIG_CTXT = 6,
672 LLOG_MD_REPL_CTXT = 7,
673 LLOG_RD1_ORIG_CTXT = 8,
674 LLOG_RD1_REPL_CTXT = 9,
675 LLOG_TEST_ORIG_CTXT = 10,
676 LLOG_TEST_REPL_CTXT = 11,
677 LLOG_LOVEA_ORIG_CTXT = 12,
678 LLOG_LOVEA_REPL_CTXT = 13,
683 * Events signalled through obd_notify() upcall-chain.
685 enum obd_notify_event {
686 /* Device activated */
688 /* Device deactivated */
690 /* Connect data for import were changed */
693 OBD_NOTIFY_SYNC_NONBLOCK,
695 /* Configuration event */
700 * Data structure used to pass obd_notify()-event to non-obd listeners (llite
701 * and liblustre being main examples).
703 struct obd_notify_upcall {
704 int (*onu_upcall)(struct obd_device *host, struct obd_device *watched,
705 enum obd_notify_event ev, void *owner);
706 /* Opaque datum supplied by upper layer listener */
710 /* corresponds to one of the obd's */
711 #define MAX_OBD_NAME 128
712 #define OBD_DEVICE_MAGIC 0XAB5CD6EF
714 struct obd_type *obd_type;
717 /* common and UUID name of this device */
718 char obd_name[MAX_OBD_NAME];
719 struct obd_uuid obd_uuid;
722 unsigned int obd_attached:1, /* finished attach */
723 obd_set_up:1, /* finished setup */
724 obd_recovering:1, /* there are recoverable clients */
725 obd_abort_recovery:1,/* somebody ioctl'ed us to abort */
726 obd_replayable:1, /* recovery is enabled; inform clients */
727 obd_no_transno:1, /* no committed-transno notification */
728 obd_no_recov:1, /* fail instead of retry messages */
729 obd_stopping:1, /* started cleanup */
730 obd_starting:1, /* started setup */
731 obd_force:1, /* cleanup with > 0 obd refcount */
732 obd_fail:1, /* cleanup with failover */
733 obd_async_recov:1, /* allow asyncronous orphan cleanup */
734 obd_no_conn:1; /* deny new connections */
735 atomic_t obd_refcount;
736 /* uuid-export hash body */
737 struct lustre_class_hash_body *obd_uuid_hash_body;
738 /* nid-export hash body */
739 struct lustre_class_hash_body *obd_nid_hash_body;
740 cfs_waitq_t obd_refcount_waitq;
741 struct list_head obd_exports;
743 struct ldlm_namespace *obd_namespace;
744 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
745 /* a spinlock is OK for what we do now, may need a semaphore later */
746 spinlock_t obd_dev_lock;
747 struct semaphore obd_dev_sem;
748 __u64 obd_last_committed;
749 struct fsfilt_operations *obd_fsops;
750 spinlock_t obd_osfs_lock;
751 struct obd_statfs obd_osfs; /* locked by obd_osfs_lock */
753 struct lvfs_run_ctxt obd_lvfs_ctxt;
754 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
755 struct obd_device *obd_observer;
756 struct obd_notify_upcall obd_upcall;
757 struct obd_export *obd_self_export;
758 /* list of exports in LRU order, for ping evictor, with obd_dev_lock */
759 struct list_head obd_exports_timed;
760 time_t obd_eviction_timer; /* for ping evictor */
762 /* XXX encapsulate all this recovery data into one struct */
763 svc_handler_t obd_recovery_handler;
764 int obd_max_recoverable_clients;
765 int obd_connected_clients;
766 int obd_recoverable_clients;
767 spinlock_t obd_processing_task_lock; /* BH lock (timer) */
768 pid_t obd_processing_task;
769 __u64 obd_next_recovery_transno;
770 int obd_replayed_requests;
771 int obd_requests_queued_for_recovery;
772 cfs_waitq_t obd_next_transno_waitq;
773 struct list_head obd_uncommitted_replies;
774 spinlock_t obd_uncommitted_replies_lock;
775 cfs_timer_t obd_recovery_timer;
776 struct list_head obd_recovery_queue;
777 struct list_head obd_delayed_reply_queue;
778 time_t obd_recovery_start;
779 time_t obd_recovery_end;
782 struct obd_device_target obt;
783 struct filter_obd filter;
785 struct client_obd cli;
787 struct echo_client_obd echo_client;
788 struct echo_obd echo;
792 /* Fields used by LProcFS */
793 cfs_proc_dir_entry_t *obd_proc_entry;
794 cfs_proc_dir_entry_t *obd_proc_exports;
795 cfs_proc_dir_entry_t *obd_svc_procroot;
796 struct lprocfs_stats *obd_stats;
797 struct lprocfs_stats *obd_svc_stats;
798 unsigned int obd_cntr_base;
799 struct semaphore obd_proc_exp_sem;
802 #define OBD_OPT_FORCE 0x0001
803 #define OBD_OPT_FAILOVER 0x0002
805 #define OBD_LLOG_FL_SENDNOW 0x0001
807 enum obd_cleanup_stage {
808 /* Special case hack for MDS LOVs */
810 /* Precleanup stage 1, we must make sure all exports (other than the
811 self-export) get destroyed. */
813 /* Precleanup stage 2, do other type-specific cleanup requiring the
815 OBD_CLEANUP_SELF_EXP,
816 /* FIXME we should eliminate the "precleanup" function and make them stages
817 of the "cleanup" function. */
821 /* get/set_info keys */
822 #define KEY_MDS_CONN "mds_conn"
823 #define KEY_NEXT_ID "next_id"
824 #define KEY_LOVDESC "lovdesc"
825 #define KEY_INIT_RECOV "initial_recov"
826 #define KEY_INIT_RECOV_BACKUP "init_recov_bk"
829 struct module *o_owner;
830 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
831 void *karg, void *uarg);
832 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
833 __u32 *vallen, void *val);
834 int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
835 __u32 vallen, void *val,
836 struct ptlrpc_request_set *set);
837 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
838 int (*o_detach)(struct obd_device *dev);
839 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
840 int (*o_precleanup)(struct obd_device *dev,
841 enum obd_cleanup_stage cleanup_stage);
842 int (*o_cleanup)(struct obd_device *dev);
843 int (*o_process_config)(struct obd_device *dev, obd_count len,
845 int (*o_postrecov)(struct obd_device *dev);
846 int (*o_add_conn)(struct obd_import *imp, struct obd_uuid *uuid,
848 int (*o_del_conn)(struct obd_import *imp, struct obd_uuid *uuid);
849 /* connect to the target device with given connection
850 * data. @ocd->ocd_connect_flags is modified to reflect flags actually
851 * granted by the target, which are guaranteed to be a subset of flags
852 * asked for. If @ocd == NULL, use default parameters. */
853 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
854 struct obd_uuid *cluuid, struct obd_connect_data *ocd);
855 int (*o_reconnect)(struct obd_export *exp, struct obd_device *src,
856 struct obd_uuid *cluuid,
857 struct obd_connect_data *ocd);
858 int (*o_disconnect)(struct obd_export *exp);
860 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
862 int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
863 __u64 max_age, struct ptlrpc_request_set *set);
864 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
865 struct lov_stripe_md *mem_src);
866 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
867 struct lov_mds_md *disk_src, int disk_len);
868 int (*o_checkmd)(struct obd_export *exp, struct obd_export *md_exp,
869 struct lov_stripe_md *mem_tgt);
870 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
872 int (*o_create)(struct obd_export *exp, struct obdo *oa,
873 struct lov_stripe_md **ea, struct obd_trans_info *oti);
874 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
875 struct lov_stripe_md *ea, struct obd_trans_info *oti,
876 struct obd_export *md_exp);
877 int (*o_setattr)(struct obd_export *exp, struct obd_info *oinfo,
878 struct obd_trans_info *oti);
879 int (*o_setattr_async)(struct obd_export *exp, struct obd_info *oinfo,
880 struct obd_trans_info *oti,
881 struct ptlrpc_request_set *rqset);
882 int (*o_getattr)(struct obd_export *exp, struct obd_info *oinfo);
883 int (*o_getattr_async)(struct obd_export *exp, struct obd_info *oinfo,
884 struct ptlrpc_request_set *set);
885 int (*o_brw)(int rw, struct obd_export *exp, struct obd_info *oinfo,
886 obd_count oa_bufs, struct brw_page *pgarr,
887 struct obd_trans_info *oti);
888 int (*o_brw_async)(int rw, struct obd_export *exp,
889 struct obd_info *oinfo, obd_count oa_bufs,
890 struct brw_page *pgarr, struct obd_trans_info *oti,
891 struct ptlrpc_request_set *);
892 int (*o_prep_async_page)(struct obd_export *exp,
893 struct lov_stripe_md *lsm,
894 struct lov_oinfo *loi,
895 cfs_page_t *page, obd_off offset,
896 struct obd_async_page_ops *ops, void *data,
898 int (*o_queue_async_io)(struct obd_export *exp,
899 struct lov_stripe_md *lsm,
900 struct lov_oinfo *loi, void *cookie,
901 int cmd, obd_off off, int count,
902 obd_flag brw_flags, obd_flag async_flags);
903 int (*o_queue_group_io)(struct obd_export *exp,
904 struct lov_stripe_md *lsm,
905 struct lov_oinfo *loi,
906 struct obd_io_group *oig,
907 void *cookie, int cmd, obd_off off, int count,
908 obd_flag brw_flags, obd_flag async_flags);
909 int (*o_trigger_group_io)(struct obd_export *exp,
910 struct lov_stripe_md *lsm,
911 struct lov_oinfo *loi,
912 struct obd_io_group *oig);
913 int (*o_set_async_flags)(struct obd_export *exp,
914 struct lov_stripe_md *lsm,
915 struct lov_oinfo *loi, void *cookie,
916 obd_flag async_flags);
917 int (*o_teardown_async_page)(struct obd_export *exp,
918 struct lov_stripe_md *lsm,
919 struct lov_oinfo *loi, void *cookie);
920 int (*o_merge_lvb)(struct obd_export *exp, struct lov_stripe_md *lsm,
921 struct ost_lvb *lvb, int kms_only);
922 int (*o_adjust_kms)(struct obd_export *exp, struct lov_stripe_md *lsm,
923 obd_off size, int shrink);
924 int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
925 struct obd_trans_info *oti,
926 struct ptlrpc_request_set *rqset);
927 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
928 struct lov_stripe_md *ea, obd_size start, obd_size end);
929 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
930 struct lov_stripe_md *src, obd_size start,
931 obd_size end, struct obd_trans_info *oti);
932 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
933 struct lustre_handle *srconn, struct lov_stripe_md *src,
934 obd_size start, obd_size end, struct obd_trans_info *);
935 int (*o_iterate)(struct lustre_handle *conn,
936 int (*)(obd_id, obd_gr, void *),
937 obd_id *startid, obd_gr group, void *data);
938 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
939 int objcount, struct obd_ioobj *obj,
940 int niocount, struct niobuf_remote *remote,
941 struct niobuf_local *local, struct obd_trans_info *oti);
942 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
943 int objcount, struct obd_ioobj *obj,
944 int niocount, struct niobuf_local *local,
945 struct obd_trans_info *oti, int rc);
946 int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
947 struct obd_enqueue_info *einfo);
948 int (*o_match)(struct obd_export *, struct lov_stripe_md *, __u32 type,
949 ldlm_policy_data_t *, __u32 mode, int *flags, void *data,
950 struct lustre_handle *lockh);
951 int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
952 ldlm_iterator_t it, void *data);
953 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
954 __u32 mode, struct lustre_handle *);
955 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
956 int flags, void *opaque);
957 int (*o_join_lru)(struct obd_export *, struct lov_stripe_md *,
959 int (*o_init_export)(struct obd_export *exp);
960 int (*o_destroy_export)(struct obd_export *exp);
961 int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
964 /* llog related obd_methods */
965 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
966 int count, struct llog_catid *logid,
967 struct obd_uuid *uuid);
968 int (*o_llog_finish)(struct obd_device *obd, int count);
970 /* metadata-only methods */
971 int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
972 struct obd_client_handle *, int flag);
973 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
975 int (*o_import_event)(struct obd_device *, struct obd_import *,
976 enum obd_import_event);
978 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
979 enum obd_notify_event ev, void *data);
981 int (*o_health_check)(struct obd_device *);
984 int (*o_quotacheck)(struct obd_export *, struct obd_quotactl *);
985 int (*o_quotactl)(struct obd_export *, struct obd_quotactl *);
987 int (*o_ping)(struct obd_export *exp);
989 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
990 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
991 * Also, add a wrapper function in include/linux/obd_class.h. */
994 struct lsm_operations {
995 void (*lsm_free)(struct lov_stripe_md *);
996 int (*lsm_destroy)(struct lov_stripe_md *, struct obdo *oa,
997 struct obd_export *md_exp);
998 void (*lsm_stripe_by_index)(struct lov_stripe_md *, int *, obd_off *,
1000 void (*lsm_stripe_by_offset)(struct lov_stripe_md *, int *, obd_off *,
1002 obd_off (*lsm_stripe_offset_by_index)(struct lov_stripe_md *, int);
1003 obd_off (*lsm_stripe_offset_by_offset)(struct lov_stripe_md *, obd_off);
1004 int (*lsm_stripe_index_by_offset)(struct lov_stripe_md *, obd_off);
1005 int (*lsm_revalidate) (struct lov_stripe_md *, struct obd_device *obd);
1006 int (*lsm_lmm_verify) (struct lov_mds_md *lmm, int lmm_bytes,
1008 int (*lsm_unpackmd) (struct lov_obd *lov, struct lov_stripe_md *lsm,
1009 struct lov_mds_md *lmm);
1012 extern struct lsm_operations lsm_plain_ops;
1013 extern struct lsm_operations lsm_join_ops;
1014 static inline struct lsm_operations *lsm_op_find(int magic)
1018 return &lsm_plain_ops;
1019 case LOV_MAGIC_JOIN:
1020 return &lsm_join_ops;
1022 CERROR("Cannot recognize lsm_magic %d", magic);
1027 int lvfs_check_io_health(struct obd_device *obd, struct file *file);
1029 /* Requests for obd_extent_calc() */
1030 #define OBD_CALC_STRIPE_START 1
1031 #define OBD_CALC_STRIPE_END 2
1033 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
1037 CERROR("%s: transno "LPD64" commit error: %d\n",
1038 obd->obd_name, transno, error);
1041 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
1042 obd->obd_name, transno);
1043 if (transno > obd->obd_last_committed) {
1044 obd->obd_last_committed = transno;
1045 ptlrpc_commit_replies (obd);
1049 static inline void init_obd_quota_ops(quota_interface_t *interface,
1050 struct obd_ops *obd_ops)
1056 obd_ops->o_quotacheck = QUOTA_OP(interface, check);
1057 obd_ops->o_quotactl = QUOTA_OP(interface, ctl);
1060 #endif /* __OBD_H */