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 obd_size fo_tot_granted; /* protected by obd_osfs_lock */
152 obd_size fo_tot_cached;
154 struct obd_import *fo_mdc_imp;
155 struct obd_uuid fo_mdc_uuid;
156 struct lustre_handle fo_mdc_conn;
158 struct ptlrpc_client fo_mdc_client;
160 struct file **fo_last_objid_files;
161 __u64 *fo_last_objids; //last created object ID for groups
163 struct semaphore fo_alloc_lock;
165 struct obd_histogram fo_r_pages;
166 struct obd_histogram fo_w_pages;
167 struct obd_histogram fo_r_discont_pages;
168 struct obd_histogram fo_w_discont_pages;
169 struct obd_histogram fo_r_discont_blocks;
170 struct obd_histogram fo_w_discont_blocks;
173 struct mds_server_data;
177 struct obd_import *cl_import;
178 struct semaphore cl_sem;
180 /* max_mds_easize is purely a performance thing so we don't have to
181 * call obd_size_wiremd() all the time. */
182 int cl_max_mds_easize;
183 int cl_max_mds_cookiesize;
186 //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
187 void *cl_llcd_offset;
189 obd_size cl_dirty; /* all _dirty_ in bytes */
190 obd_size cl_dirty_granted; /* from ost */
191 obd_size cl_dirty_max; /* allowed w/o rpc */
192 struct list_head cl_cache_waiters;
194 struct obd_device *cl_mgmtcli_obd;
196 /* this is just to keep existing infinitely caching behaviour between
197 * clients and OSTs that don't have the grant code in yet.. it can
198 * be yanked once everything speaks grants */
199 char cl_ost_can_grant;
201 /* keep track of objects that have lois that contain pages which
202 * have been queued for async brw. this lock also protects the
203 * lists of obd_client_pages that hang off of the loi */
204 spinlock_t cl_loi_list_lock;
205 struct list_head cl_loi_ready_list;
206 int cl_brw_in_flight;
207 /* just a sum of the loi/lop pending numbers to be exported by /proc */
208 int cl_pending_w_pages;
209 int cl_pending_r_pages;
210 int cl_max_pages_per_rpc;
211 int cl_max_rpcs_in_flight;
212 struct obd_histogram cl_read_rpc_hist;
213 struct obd_histogram cl_write_rpc_hist;
214 struct obd_histogram cl_read_page_hist;
215 struct obd_histogram cl_write_page_hist;
217 struct mdc_rpc_lock *cl_rpc_lock;
218 struct mdc_rpc_lock *cl_setattr_lock;
221 /* Like a client, with some hangers-on. Keep mc_client_obd first so that we
222 * can reuse the various client setup/connect functions. */
224 struct client_obd mc_client_obd; /* nested */
225 struct ptlrpc_thread *mc_ping_thread;
226 struct obd_export *mc_ping_exp; /* XXX single-target */
227 struct list_head mc_registered;
231 #define mc_import mc_client_obd.cl_import
234 struct ptlrpc_service *mds_service;
235 struct ptlrpc_service *mds_setattr_service;
236 struct ptlrpc_service *mds_readpage_service;
237 struct super_block *mds_sb;
238 struct vfsmount *mds_vfsmnt;
239 struct dentry *mds_fid_de;
241 int mds_max_cookiesize;
242 struct file *mds_rcvd_filp;
243 spinlock_t mds_transno_lock;
244 __u64 mds_last_transno;
245 __u64 mds_mount_count;
247 struct semaphore mds_epoch_sem;
248 struct ll_fid mds_rootfid;
249 struct mds_server_data *mds_server_data;
250 struct dentry *mds_pending_dir;
251 struct dentry *mds_logs_dir;
252 struct dentry *mds_objects_dir;
253 struct llog_handle *mds_cfg_llh;
254 // struct llog_handle *mds_catalog;
255 struct obd_device *mds_osc_obd; /* XXX lov_obd */
256 struct obd_uuid mds_lov_uuid;
258 struct obd_export *mds_osc_exp; /* XXX lov_exp */
259 int mds_has_lov_desc;
260 struct lov_desc mds_lov_desc;
261 obd_id *mds_lov_objids;
262 int mds_lov_objids_valid;
263 int mds_lov_nextid_set;
264 struct file *mds_lov_objid_filp;
265 unsigned long *mds_client_bitmap;
266 struct semaphore mds_orphan_recovery_sem;
268 atomic_t mds_open_count;
285 * this struct does double-duty acting as either a client or
286 * server instance .. maybe not wise.
290 struct ptlrpc_service *ptlbd_service;
293 struct ptlrpc_client bd_client;
294 struct obd_import *bd_import;
295 struct obd_uuid bd_server_uuid;
296 struct obd_export *bd_exp;
297 int refcount; /* XXX sigh */
301 spinlock_t recovd_lock;
302 struct list_head recovd_managed_items; /* items managed */
303 struct list_head recovd_troubled_items; /* items in recovery */
305 wait_queue_head_t recovd_recovery_waitq;
306 wait_queue_head_t recovd_ctl_waitq;
307 wait_queue_head_t recovd_waitq;
308 struct task_struct *recovd_thread;
313 struct ptlrpc_service *ost_service;
314 struct ptlrpc_service *ost_create_service;
317 struct echo_client_obd {
318 struct obd_export *ec_exp; /* the local connection to osc/lov */
320 struct list_head ec_objects;
326 struct obd_export *cobd_target_exp;/* local connection to target obd */
327 struct obd_export *cobd_cache_exp; /* local connection to cache obd */
330 struct lov_tgt_desc {
331 struct obd_uuid uuid;
332 struct obd_export *ltd_exp;
333 int active; /* is this target up for requests */
338 struct lov_desc desc;
341 int lo_catalog_loaded:1;
342 struct lov_tgt_desc *tgts;
345 struct niobuf_local {
351 struct dentry *dentry;
356 /* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
357 #define N_LOCAL_TEMP_PAGE 0x10000000
359 struct obd_trans_info {
362 /* Only used on the server side for tracking acks. */
363 struct oti_req_ack_lock {
364 struct lustre_handle lock;
368 struct llog_cookie oti_onecookie;
369 struct llog_cookie *oti_logcookies;
373 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
378 if (num_cookies == 1)
379 oti->oti_logcookies = &oti->oti_onecookie;
381 OBD_ALLOC(oti->oti_logcookies,
382 num_cookies * sizeof(oti->oti_onecookie));
384 oti->oti_numcookies = num_cookies;
387 static inline void oti_free_cookies(struct obd_trans_info *oti)
389 if (!oti || !oti->oti_logcookies)
392 if (oti->oti_logcookies == &oti->oti_onecookie)
393 LASSERT(oti->oti_numcookies == 1);
395 OBD_FREE(oti->oti_logcookies,
396 oti->oti_numcookies * sizeof(oti->oti_onecookie));
397 oti->oti_logcookies = NULL;
398 oti->oti_numcookies = 0;
403 LLOG_CONFIG_ORIG_CTXT = 0,
404 LLOG_CONFIG_REPL_CTXT = 1,
405 LLOG_UNLINK_ORIG_CTXT = 2,
406 LLOG_UNLINK_REPL_CTXT = 3,
407 LLOG_SIZE_ORIG_CTXT = 4,
408 LLOG_SIZE_REPL_CTXT = 5,
409 LLOG_MD_ORIG_CTXT = 6,
410 LLOG_MD_REPL_CTXT = 7,
411 LLOG_RD1_ORIG_CTXT = 8,
412 LLOG_RD1_REPL_CTXT = 9,
413 LLOG_TEST_ORIG_CTXT = 10,
414 LLOG_TEST_REPL_CTXT = 11,
419 /* corresponds to one of the obd's */
421 struct obd_type *obd_type;
423 /* common and UUID name of this device */
425 struct obd_uuid obd_uuid;
428 int obd_attached:1, obd_set_up:1, obd_recovering:1,
429 obd_abort_recovery:1, obd_replayable:1, obd_no_transno:1,
430 obd_no_recov:1, obd_stopping:1;
431 atomic_t obd_refcount;
432 wait_queue_head_t obd_refcount_waitq;
433 struct proc_dir_entry *obd_proc_entry;
434 struct list_head obd_exports;
436 struct ldlm_namespace *obd_namespace;
437 struct ptlrpc_client obd_ldlm_client; /* XXX OST/MDS only */
438 /* a spinlock is OK for what we do now, may need a semaphore later */
439 spinlock_t obd_dev_lock;
440 __u64 obd_last_committed;
441 struct fsfilt_operations *obd_fsops;
442 spinlock_t obd_osfs_lock;
443 struct llog_ctxt *obd_llog_ctxt[LLOG_MAX_CTXTS];
444 struct obd_statfs obd_osfs;
445 unsigned long obd_osfs_age;
446 struct obd_run_ctxt obd_ctxt;
447 struct obd_device *obd_observer;
448 struct obd_export *obd_self_export;
450 /* XXX encapsulate all this recovery data into one struct */
451 svc_handler_t obd_recovery_handler;
452 int obd_max_recoverable_clients;
453 int obd_connected_clients;
454 int obd_recoverable_clients;
455 spinlock_t obd_processing_task_lock;
456 pid_t obd_processing_task;
457 __u64 obd_next_recovery_transno;
458 int obd_replayed_requests;
459 int obd_requests_queued_for_recovery;
460 wait_queue_head_t obd_next_transno_waitq;
461 wait_queue_head_t obd_commit_waitq;
462 struct timer_list obd_recovery_timer;
463 struct list_head obd_recovery_queue;
464 struct list_head obd_delayed_reply_queue;
467 struct filter_obd filter;
469 struct client_obd cli;
471 struct echo_client_obd echo_client;
472 struct echo_obd echo;
473 struct recovd_obd recovd;
475 struct cache_obd cobd;
476 struct ptlbd_obd ptlbd;
477 struct mgmtcli_obd mgmtcli;
479 /* Fields used by LProcFS */
480 unsigned int obd_cntr_base;
481 struct lprocfs_stats *obd_stats;
482 struct proc_dir_entry *obd_svc_procroot;
483 struct lprocfs_stats *obd_svc_stats;
486 #define OBD_OPT_FORCE 0x0001
487 #define OBD_OPT_FAILOVER 0x0002
489 #define OBD_LLOG_FL_SENDNOW 0x0001
492 struct module *o_owner;
493 int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
494 void *karg, void *uarg);
495 int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
496 __u32 *vallen, void *val);
497 int (*o_set_info)(struct obd_export *, __u32 keylen, void *key,
498 __u32 vallen, void *val);
499 int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
500 int (*o_detach)(struct obd_device *dev);
501 int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
502 int (*o_postsetup) (struct obd_device *dev);
503 int (*o_precleanup)(struct obd_device *dev, int flags);
504 int (*o_cleanup)(struct obd_device *dev, int flags);
505 int (*o_postrecov)(struct obd_device *dev);
506 int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
507 struct obd_uuid *cluuid);
508 int (*o_disconnect)(struct obd_export *exp, int flags);
510 int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
511 unsigned long max_age);
512 int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
513 struct lov_stripe_md *mem_src);
514 int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
515 struct lov_mds_md *disk_src, int disk_len);
516 int (*o_preallocate)(struct lustre_handle *, obd_count *req,
518 int (*o_create)(struct obd_export *exp, struct obdo *oa,
519 struct lov_stripe_md **ea, struct obd_trans_info *oti);
520 int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
521 struct lov_stripe_md *ea, struct obd_trans_info *oti);
522 int (*o_setattr)(struct obd_export *exp, struct obdo *oa,
523 struct lov_stripe_md *ea, struct obd_trans_info *oti);
524 int (*o_getattr)(struct obd_export *exp, struct obdo *oa,
525 struct lov_stripe_md *ea);
526 int (*o_getattr_async)(struct obd_export *exp, struct obdo *oa,
527 struct lov_stripe_md *ea,
528 struct ptlrpc_request_set *set);
529 int (*o_brw)(int rw, struct obd_export *exp, struct obdo *oa,
530 struct lov_stripe_md *ea, obd_count oa_bufs,
531 struct brw_page *pgarr, struct obd_trans_info *oti);
532 int (*o_brw_async)(int rw, struct obd_export *exp, struct obdo *oa,
533 struct lov_stripe_md *ea, obd_count oa_bufs,
534 struct brw_page *pgarr, struct ptlrpc_request_set *,
535 struct obd_trans_info *oti);
536 int (*o_prep_async_page)(struct obd_export *exp,
537 struct lov_stripe_md *lsm,
538 struct lov_oinfo *loi,
539 struct page *page, obd_off offset,
540 struct obd_async_page_ops *ops, void *data,
542 int (*o_queue_async_io)(struct obd_export *exp,
543 struct lov_stripe_md *lsm,
544 struct lov_oinfo *loi, void *cookie,
545 int cmd, obd_off off, int count,
546 obd_flag brw_flags, obd_flag async_flags);
547 int (*o_queue_sync_io)(struct obd_export *exp,
548 struct lov_stripe_md *lsm,
549 struct lov_oinfo *loi,
550 struct obd_sync_io_container *osic,
551 void *cookie, int cmd, obd_off off, int count,
553 int (*o_trigger_sync_io)(struct obd_export *exp,
554 struct lov_stripe_md *lsm,
555 struct lov_oinfo *loi,
556 struct obd_sync_io_container *osic);
557 int (*o_set_async_flags)(struct obd_export *exp,
558 struct lov_stripe_md *lsm,
559 struct lov_oinfo *loi, void *cookie,
560 obd_flag async_flags);
561 int (*o_teardown_async_page)(struct obd_export *exp,
562 struct lov_stripe_md *lsm,
563 struct lov_oinfo *loi, void *cookie);
564 int (*o_punch)(struct obd_export *exp, struct obdo *oa,
565 struct lov_stripe_md *ea, obd_size start,
566 obd_size end, struct obd_trans_info *oti);
567 int (*o_sync)(struct obd_export *exp, struct obdo *oa,
568 struct lov_stripe_md *ea, obd_size start, obd_size end);
569 int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
570 struct lov_stripe_md *src, obd_size start,
571 obd_size end, struct obd_trans_info *oti);
572 int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
573 struct lustre_handle *srconn, struct lov_stripe_md *src,
574 obd_size start, obd_size end, struct obd_trans_info *);
575 int (*o_iterate)(struct lustre_handle *conn,
576 int (*)(obd_id, obd_gr, void *),
577 obd_id *startid, obd_gr group, void *data);
578 int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
579 int objcount, struct obd_ioobj *obj,
580 int niocount, struct niobuf_remote *remote,
581 struct niobuf_local *local, struct obd_trans_info *oti);
582 int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
583 int objcount, struct obd_ioobj *obj,
584 int niocount, struct niobuf_local *local,
585 struct obd_trans_info *oti);
586 int (*o_enqueue)(struct obd_export *exp, struct lov_stripe_md *md,
587 struct lustre_handle *parent_lock,
588 __u32 type, void *cookie, int cookielen, __u32 mode,
589 int *flags, void *cb, void *data,
590 struct lustre_handle *lockh);
591 int (*o_match)(struct obd_export *exp, struct lov_stripe_md *md,
592 __u32 type, void *cookie, int cookielen, __u32 mode,
593 int *flags, void *data, struct lustre_handle *lockh);
594 int (*o_change_cbdata)(struct obd_export *exp,
595 struct lov_stripe_md *lsm, ldlm_iterator_t it,
597 int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
598 __u32 mode, struct lustre_handle *);
599 int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
600 int flags, void *opaque);
601 int (*o_san_preprw)(int cmd, struct obd_export *exp,
602 struct obdo *oa, int objcount,
603 struct obd_ioobj *obj, int niocount,
604 struct niobuf_remote *remote);
605 int (*o_init_export)(struct obd_export *exp);
606 int (*o_destroy_export)(struct obd_export *exp);
608 /* llog related obd_methods */
609 int (*o_llog_init)(struct obd_device *obd, struct obd_device *disk_obd,
610 int count, struct llog_logid *logid);
611 int (*o_llog_finish)(struct obd_device *obd, int count);
613 /* only until proper file size mechanics arrive */
614 int (*o_lock_contains)(struct obd_export *exp,
615 struct lov_stripe_md *lsm,
616 struct ldlm_lock *lock, obd_off offset);
618 /* metadata-only methods */
619 int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
620 struct obd_client_handle *, int flag);
621 int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
623 int (*o_invalidate_import)(struct obd_device *, struct obd_import *);
625 int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
628 * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
629 * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
630 * Also, add a wrapper function in include/linux/obd_class.h.
636 static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
640 CERROR("%s: transno "LPD64" commit error: %d\n",
641 obd->obd_name, transno, error);
644 CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
645 obd->obd_name, transno);
646 if (transno > obd->obd_last_committed) {
647 obd->obd_last_committed = transno;
648 wake_up(&obd->obd_commit_waitq);