1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
6 * This code is issued under the GNU General Public License.
7 * See the file COPYING in this distribution
13 #define IOC_OSC_TYPE 'h'
14 #define IOC_OSC_MIN_NR 20
15 #define IOC_OSC_SET_ACTIVE _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
16 #define IOC_OSC_MAX_NR 50
18 #define IOC_MDC_TYPE 'i'
19 #define IOC_MDC_MIN_NR 20
20 #define IOC_MDC_LOOKUP _IOWR(IOC_MDC_TYPE, 20, struct obd_device *)
21 #define IOC_MDC_GETSTRIPE _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *)
22 #define IOC_MDC_MAX_NR 50
25 # include <linux/fs.h>
26 # include <linux/list.h>
27 # include <linux/sched.h> /* for struct task_struct, for current.h */
28 # include <asm/current.h> /* for smp_lock.h */
29 # include <linux/smp_lock.h>
30 # include <linux/proc_fs.h>
31 # include <linux/mount.h>
34 #include <linux/lustre_lib.h>
35 #include <linux/lustre_idl.h>
36 #include <linux/lustre_export.h>
38 /* this is really local to the OSC */
39 struct loi_oap_pages {
40 struct list_head lop_pending;
42 struct list_head lop_urgent;
43 struct list_head lop_pending_sync;
46 struct lov_oinfo { /* per-stripe data structure */
47 __u64 loi_id; /* object ID on the target OST */
48 __u64 loi_gr; /* object group on the target OST */
49 struct lustre_handle *loi_handle; /* open file handle for obj on OST */
50 int loi_ost_idx; /* OST stripe index in lov_tgt_desc->tgts */
51 int loi_ost_gen; /* generation of this loi_ost_idx */
53 /* tracking offsets per file, per stripe.. */
54 struct loi_oap_pages loi_read_lop;
55 struct loi_oap_pages loi_write_lop;
56 struct list_head loi_cli_item;
59 static inline void loi_init(struct lov_oinfo *loi)
61 INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
62 INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
63 INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_sync);
64 INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
65 INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
66 INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_sync);
67 INIT_LIST_HEAD(&loi->loi_cli_item);
70 struct lov_stripe_md {
72 __u64 lsm_object_id; /* lov object id */
73 __u64 lsm_object_gr; /* lov object id */
76 /* LOV-private members start here -- only for use in lov/. */
78 __u32 lsm_stripe_size; /* size of the stripe */
79 __u32 lsm_pattern; /* striping pattern (RAID0, RAID1) */
80 unsigned lsm_stripe_count; /* number of objects being striped over */
81 struct lov_oinfo lsm_oinfo[0];
85 struct list_head typ_chain;
86 struct obd_ops *typ_ops;
87 struct proc_dir_entry *typ_procroot;
102 ASYNC_COUNT_STABLE = 0x4,
105 struct obd_async_page_ops {
106 int (*ap_make_ready)(void *data, int cmd);
107 int (*ap_refresh_count)(void *data, int cmd);
108 void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
109 void (*ap_completion)(void *data, int cmd, int rc);
112 struct obd_sync_io_container {
113 spinlock_t osic_lock;
114 atomic_t osic_refcount;
117 wait_queue_head_t osic_waitq;
120 /* if we find more consumers this could be generalized */
121 #define OBD_HIST_MAX 32
122 struct obd_histogram {
124 unsigned long oh_buckets[OBD_HIST_MAX];
127 /* Individual type definitions */
129 struct ost_server_data;
132 const char *fo_fstype;
133 struct super_block *fo_sb;
134 struct vfsmount *fo_vfsmnt;
135 struct dentry *fo_dentry_O;
136 struct dentry **fo_dentry_O_groups;
137 struct dentry **fo_dentry_O_sub;
138 spinlock_t fo_objidlock; /* protect fo_lastobjid increment */
139 spinlock_t fo_translock; /* protect fsd_last_rcvd increment */
140 struct file *fo_rcvd_filp;
141 struct filter_server_data *fo_fsd;
142 unsigned long *fo_last_rcvd_slots;
143 __u64 fo_mount_count;
145 struct file_operations *fo_fop;
146 struct inode_operations *fo_iop;
147 struct address_space_operations *fo_aops;
149 struct list_head fo_export_list;
151 spinlock_t fo_grant_lock; /* protects tot_granted */
152 obd_size fo_tot_granted;
153 obd_size fo_tot_cached;
155 struct obd_import *fo_mdc_imp;
156 struct obd_uuid fo_mdc_uuid;
157 struct lustre_handle fo_mdc_conn;
159 struct ptlrpc_client fo_mdc_client;
161 struct file **fo_last_objid_files;
162 __u64 *fo_last_objids; //last created object ID for groups
164 struct semaphore fo_alloc_lock;
166 struct obd_histogram fo_r_pages;
167 struct obd_histogram fo_w_pages;
168 struct obd_histogram fo_r_discont_pages;
169 struct obd_histogram fo_w_discont_pages;
170 struct obd_histogram fo_r_discont_blocks;
171 struct obd_histogram fo_w_discont_blocks;
174 struct mds_server_data;
178 struct obd_import *cl_import;
179 struct semaphore cl_sem;
181 /* max_mds_easize is purely a performance thing so we don't have to
182 * call obd_size_wiremd() all the time. */
183 int cl_max_mds_easize;
184 int cl_max_mds_cookiesize;
187 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
188 void *cl_llcd_offset;
190 struct semaphore cl_dirty_sem;
191 obd_size cl_dirty; /* all _dirty_ in bytes */
192 obd_size cl_dirty_granted; /* from ost */
193 obd_size cl_dirty_max; /* allowed w/o rpc */
194 struct list_head cl_cache_waiters;
196 struct obd_device *cl_mgmtcli_obd;
198 /* this is just to keep existing infinitely caching behaviour between
199 * clients and OSTs that don't have the grant code in yet.. it can
200 * be yanked once everything speaks grants */
201 char cl_ost_can_grant;
203 /* keep track of objects that have lois that contain pages which
204 * have been queued for async brw. this lock also protects the
205 * lists of obd_client_pages that hang off of the loi */
206 spinlock_t cl_loi_list_lock;
207 struct list_head cl_loi_ready_list;
208 int cl_brw_in_flight;
209 /* just a sum of the loi/lop pending numbers to be exported by /proc */
210 int cl_pending_w_pages;
211 int cl_pending_r_pages;
212 int cl_max_pages_per_rpc;
213 int cl_max_rpcs_in_flight;
214 struct obd_histogram cl_read_rpc_hist;
215 struct obd_histogram cl_write_rpc_hist;
216 struct obd_histogram cl_read_page_hist;
217 struct obd_histogram cl_write_page_hist;
219 struct mdc_rpc_lock *cl_rpc_lock;
220 struct mdc_rpc_lock *cl_setattr_lock;
223 /* Like a client, with some hangers-on. Keep mc_client_obd first so that we
224 * can reuse the various client setup/connect functions. */
226 struct client_obd mc_client_obd; /* nested */
227 struct ptlrpc_thread *mc_ping_thread;
228 struct obd_export *mc_ping_exp; /* XXX single-target */
229 struct list_head mc_registered;
233 #define mc_import mc_client_obd.cl_import
236 struct ptlrpc_service *mds_service;
237 struct ptlrpc_service *mds_setattr_service;
238 struct ptlrpc_service *mds_readpage_service;
239 struct super_block *mds_sb;
240 struct vfsmount *mds_vfsmnt;
241 struct dentry *mds_fid_de;
243 int mds_max_cookiesize;
244 struct file *mds_rcvd_filp;
245 spinlock_t mds_transno_lock;
246 __u64 mds_last_transno;
247 __u64 mds_mount_count;
249 struct semaphore mds_epoch_sem;
250 struct ll_fid mds_rootfid;
251 struct mds_server_data *mds_server_data;
252 struct dentry *mds_pending_dir;
253 struct dentry *mds_logs_dir;
254 struct dentry *mds_objects_dir;
255 struct llog_handle *mds_cfg_llh;
256 // struct llog_handle *mds_catalog;
257 struct obd_device *mds_osc_obd; /* XXX lov_obd */
258 struct obd_uuid mds_lov_uuid;
260 struct obd_export *mds_osc_exp; /* XXX lov_exp */
261 int mds_has_lov_desc;
262 struct lov_desc mds_lov_desc;
263 obd_id *mds_lov_objids;
264 int mds_lov_objids_valid;
265 int mds_lov_nextid_set;
266 struct file *mds_lov_objid_filp;
267 unsigned long *mds_client_bitmap;
268 struct semaphore mds_orphan_recovery_sem;
270 atomic_t mds_open_count;
287 * this struct does double-duty acting as either a client or
288 * server instance .. maybe not wise.
292 struct ptlrpc_service *ptlbd_service;
295 struct ptlrpc_client bd_client;
296 struct obd_import *bd_import;
297 struct obd_uuid bd_server_uuid;
298 struct obd_export *bd_exp;
299 int refcount; /* XXX sigh */
303 spinlock_t recovd_lock;
304 struct list_head recovd_managed_items; /* items managed */
305 struct list_head recovd_troubled_items; /* items in recovery */
307 wait_queue_head_t recovd_recovery_waitq;
308 wait_queue_head_t recovd_ctl_waitq;
309 wait_queue_head_t recovd_waitq;
310 struct task_struct *recovd_thread;
315 struct ptlrpc_service *ost_service;
316 struct ptlrpc_service *ost_create_service;
319 struct echo_client_obd {
320 struct obd_export *ec_exp; /* the local connection to osc/lov */
322 struct list_head ec_objects;
328 struct obd_export *cobd_target_exp;/* local connection to target obd */
329 struct obd_export *cobd_cache_exp; /* local connection to cache obd */
332 struct lov_tgt_desc {
333 struct obd_uuid uuid;
334 struct obd_export *ltd_exp;
335 int active; /* is this target up for requests */
340 struct lov_desc desc;
343 int lo_catalog_loaded:1;
344 struct lov_tgt_desc *tgts;
347 struct niobuf_local {
353 struct dentry *dentry;
358 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
359 #define N_LOCAL_TEMP_PAGE 0x10000000
361 struct obd_trans_info {
364 /* Only used on the server side for tracking acks. */
365 struct oti_req_ack_lock {
366 struct lustre_handle lock;
370 struct llog_cookie oti_onecookie;
371 struct llog_cookie *oti_logcookies;
375 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
380 if (num_cookies == 1)
381 oti->oti_logcookies = &oti->oti_onecookie;
383 OBD_ALLOC(oti->oti_logcookies,
384 num_cookies * sizeof(oti->oti_onecookie));
386 oti->oti_numcookies = num_cookies;
389 static inline void oti_free_cookies(struct obd_trans_info *oti)
391 if (!oti || !oti->oti_logcookies)
394 if (oti->oti_logcookies == &oti->oti_onecookie)
395 LASSERT(oti->oti_numcookies == 1);
397 OBD_FREE(oti->oti_logcookies,
398 oti->oti_numcookies * sizeof(oti->oti_onecookie));
399 oti->oti_logcookies = NULL;
400 oti->oti_numcookies = 0;
405 LLOG_CONFIG_ORIG_CTXT = 0,
406 LLOG_CONFIG_REPL_CTXT = 1,
407 LLOG_UNLINK_ORIG_CTXT = 2,
408 LLOG_UNLINK_REPL_CTXT = 3,
409 LLOG_SIZE_ORIG_CTXT = 4,
410 LLOG_SIZE_REPL_CTXT = 5,
411 LLOG_MD_ORIG_CTXT = 6,
412 LLOG_MD_REPL_CTXT = 7,
413 LLOG_RD1_ORIG_CTXT = 8,
414 LLOG_RD1_REPL_CTXT = 9,
415 LLOG_TEST_ORIG_CTXT = 10,
416 LLOG_TEST_REPL_CTXT = 11,
421 /* corresponds to one of the obd's */
423 struct obd_type *obd_type;
425 /* common and UUID name of this device */
427 struct obd_uuid obd_uuid;
430 int obd_attached:1, obd_set_up:1, obd_recovering:1,
431 obd_abort_recovery:1, obd_replayable:1, obd_no_transno:1,
432 obd_no_recov:1, obd_stopping:1;
433 atomic_t obd_refcount;
434 wait_queue_head_t obd_refcount_waitq;
435 struct proc_dir_entry *obd_proc_entry;
436 struct list_head obd_exports;
438 struct ldlm_namespace *obd_namespace;
439 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
440 /* a spinlock is OK for what we do now, may need a semaphore later */
441 spinlock_t obd_dev_lock;
442 __u64 obd_last_committed;
443 struct fsfilt_operations *obd_fsops;
444 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
445 struct obd_statfs obd_osfs;
446 unsigned long obd_osfs_age;
447 struct obd_run_ctxt obd_ctxt;
448 struct obd_device *obd_observer;
449 struct obd_export *obd_self_export;
451 /* XXX encapsulate all this recovery data into one struct */
452 svc_handler_t obd_recovery_handler;
453 int obd_max_recoverable_clients;
454 int obd_connected_clients;
455 int obd_recoverable_clients;
456 spinlock_t obd_processing_task_lock;
457 pid_t obd_processing_task;
458 __u64 obd_next_recovery_transno;
459 int obd_replayed_requests;
460 int obd_requests_queued_for_recovery;
461 wait_queue_head_t obd_next_transno_waitq;
462 wait_queue_head_t obd_commit_waitq;
463 struct timer_list obd_recovery_timer;
464 struct list_head obd_recovery_queue;
465 struct list_head obd_delayed_reply_queue;
468 struct filter_obd filter;
470 struct client_obd cli;
472 struct echo_client_obd echo_client;
473 struct echo_obd echo;
474 struct recovd_obd recovd;
476 struct cache_obd cobd;
477 struct ptlbd_obd ptlbd;
478 struct mgmtcli_obd mgmtcli;
480 /* Fields used by LProcFS */
481 unsigned int obd_cntr_base;
482 struct lprocfs_stats *obd_stats;
483 struct proc_dir_entry *obd_svc_procroot;
484 struct lprocfs_stats *obd_svc_stats;
487 #define OBD_OPT_FORCE 0x0001
488 #define OBD_OPT_FAILOVER 0x0002
490 #define OBD_LLOG_FL_SENDNOW 0x0001
493 struct module *o_owner;
494 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
495 void *karg, void *uarg);
496 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
497 __u32 *vallen, void *val);
498 int (*o_set_info)(struct obd_export *, __u32 keylen, void *key,
499 __u32 vallen, void *val);
500 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
501 int (*o_detach)(struct obd_device *dev);
502 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
503 int (*o_postsetup) (struct obd_device *dev);
504 int (*o_precleanup)(struct obd_device *dev, int flags);
505 int (*o_cleanup)(struct obd_device *dev, int flags);
506 int (*o_postrecov)(struct obd_device *dev);
507 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
508 struct obd_uuid *cluuid);
509 int (*o_disconnect)(struct obd_export *exp, int flags);
511 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
512 unsigned long max_age);
513 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
514 struct lov_stripe_md *mem_src);
515 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
516 struct lov_mds_md *disk_src, int disk_len);
517 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
519 int (*o_create)(struct obd_export *exp, struct obdo *oa,
520 struct lov_stripe_md **ea, struct obd_trans_info *oti);
521 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
522 struct lov_stripe_md *ea, struct obd_trans_info *oti);
523 int (*o_setattr)(struct obd_export *exp, struct obdo *oa,
524 struct lov_stripe_md *ea, struct obd_trans_info *oti);
525 int (*o_getattr)(struct obd_export *exp, struct obdo *oa,
526 struct lov_stripe_md *ea);
527 int (*o_getattr_async)(struct obd_export *exp, struct obdo *oa,
528 struct lov_stripe_md *ea,
529 struct ptlrpc_request_set *set);
530 int (*o_brw)(int rw, struct obd_export *exp, struct obdo *oa,
531 struct lov_stripe_md *ea, obd_count oa_bufs,
532 struct brw_page *pgarr, struct obd_trans_info *oti);
533 int (*o_brw_async)(int rw, struct obd_export *exp, struct obdo *oa,
534 struct lov_stripe_md *ea, obd_count oa_bufs,
535 struct brw_page *pgarr, struct ptlrpc_request_set *,
536 struct obd_trans_info *oti);
537 int (*o_prep_async_page)(struct obd_export *exp,
538 struct lov_stripe_md *lsm,
539 struct lov_oinfo *loi,
540 struct page *page, obd_off offset,
541 struct obd_async_page_ops *ops, void *data,
543 int (*o_queue_async_io)(struct obd_export *exp,
544 struct lov_stripe_md *lsm,
545 struct lov_oinfo *loi, void *cookie,
546 int cmd, obd_off off, int count,
547 obd_flag brw_flags, obd_flag async_flags);
548 int (*o_queue_sync_io)(struct obd_export *exp,
549 struct lov_stripe_md *lsm,
550 struct lov_oinfo *loi,
551 struct obd_sync_io_container *osic,
552 void *cookie, int cmd, obd_off off, int count,
554 int (*o_trigger_sync_io)(struct obd_export *exp,
555 struct lov_stripe_md *lsm,
556 struct lov_oinfo *loi,
557 struct obd_sync_io_container *osic);
558 int (*o_set_async_flags)(struct obd_export *exp,
559 struct lov_stripe_md *lsm,
560 struct lov_oinfo *loi, void *cookie,
561 obd_flag async_flags);
562 int (*o_teardown_async_page)(struct obd_export *exp,
563 struct lov_stripe_md *lsm,
564 struct lov_oinfo *loi, void *cookie);
565 int (*o_punch)(struct obd_export *exp, struct obdo *oa,
566 struct lov_stripe_md *ea, obd_size start,
567 obd_size end, struct obd_trans_info *oti);
568 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
569 struct lov_stripe_md *ea, obd_size start, obd_size end);
570 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
571 struct lov_stripe_md *src, obd_size start,
572 obd_size end, struct obd_trans_info *oti);
573 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
574 struct lustre_handle *srconn, struct lov_stripe_md *src,
575 obd_size start, obd_size end, struct obd_trans_info *);
576 int (*o_iterate)(struct lustre_handle *conn,
577 int (*)(obd_id, obd_gr, void *),
578 obd_id *startid, obd_gr group, void *data);
579 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
580 int objcount, struct obd_ioobj *obj,
581 int niocount, struct niobuf_remote *remote,
582 struct niobuf_local *local, struct obd_trans_info *oti);
583 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
584 int objcount, struct obd_ioobj *obj,
585 int niocount, struct niobuf_local *local,
586 struct obd_trans_info *oti);
587 int (*o_enqueue)(struct obd_export *exp, struct lov_stripe_md *md,
588 struct lustre_handle *parent_lock,
589 __u32 type, void *cookie, int cookielen, __u32 mode,
590 int *flags, void *cb, void *data,
591 struct lustre_handle *lockh);
592 int (*o_match)(struct obd_export *exp, struct lov_stripe_md *md,
593 __u32 type, void *cookie, int cookielen, __u32 mode,
594 int *flags, void *data, struct lustre_handle *lockh);
595 int (*o_change_cbdata)(struct obd_export *exp,
596 struct lov_stripe_md *lsm, ldlm_iterator_t it,
598 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
599 __u32 mode, struct lustre_handle *);
600 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
601 int flags, void *opaque);
602 int (*o_san_preprw)(int cmd, struct obd_export *exp,
603 struct obdo *oa, int objcount,
604 struct obd_ioobj *obj, int niocount,
605 struct niobuf_remote *remote);
606 int (*o_init_export)(struct obd_export *exp);
607 int (*o_destroy_export)(struct obd_export *exp);
609 /* llog related obd_methods */
610 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
611 int count, struct llog_logid *logid);
612 int (*o_llog_finish)(struct obd_device *obd, int count);
614 /* only until proper file size mechanics arrive */
615 int (*o_lock_contains)(struct obd_export *exp,
616 struct lov_stripe_md *lsm,
617 struct ldlm_lock *lock, obd_off offset);
619 /* metadata-only methods */
620 int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
621 struct obd_client_handle *, int flag);
622 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
624 int (*o_invalidate_import)(struct obd_device *, struct obd_import *);
626 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
629 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
630 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
631 * Also, add a wrapper function in include/linux/obd_class.h.
637 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
641 CERROR("%s: transno "LPD64" commit error: %d\n",
642 obd->obd_name, transno, error);
645 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
646 obd->obd_name, transno);
647 if (transno > obd->obd_last_committed) {
648 obd->obd_last_committed = transno;
649 wake_up(&obd->obd_commit_waitq);